root / fpu / softfloat.c @ 7fee199c
History | View | Annotate | Download (228.5 kB)
1 | 8d725fac | Andreas Färber | /*
|
---|---|---|---|
2 | 8d725fac | Andreas Färber | * QEMU float support
|
3 | 8d725fac | Andreas Färber | *
|
4 | 8d725fac | Andreas Färber | * Derived from SoftFloat.
|
5 | 8d725fac | Andreas Färber | */
|
6 | 158142c2 | bellard | |
7 | 158142c2 | bellard | /*============================================================================
|
8 | 158142c2 | bellard | |
9 | 158142c2 | bellard | This C source file is part of the SoftFloat IEC/IEEE Floating-point Arithmetic
|
10 | 158142c2 | bellard | Package, Release 2b.
|
11 | 158142c2 | bellard | |
12 | 158142c2 | bellard | Written by John R. Hauser. This work was made possible in part by the
|
13 | 158142c2 | bellard | International Computer Science Institute, located at Suite 600, 1947 Center
|
14 | 158142c2 | bellard | Street, Berkeley, California 94704. Funding was partially provided by the
|
15 | 158142c2 | bellard | National Science Foundation under grant MIP-9311980. The original version
|
16 | 158142c2 | bellard | of this code was written as part of a project to build a fixed-point vector
|
17 | 158142c2 | bellard | processor in collaboration with the University of California at Berkeley,
|
18 | 158142c2 | bellard | overseen by Profs. Nelson Morgan and John Wawrzynek. More information
|
19 | 158142c2 | bellard | is available through the Web page `http://www.cs.berkeley.edu/~jhauser/
|
20 | 158142c2 | bellard | arithmetic/SoftFloat.html'.
|
21 | 158142c2 | bellard | |
22 | 158142c2 | bellard | THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE. Although reasonable effort has
|
23 | 158142c2 | bellard | been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT TIMES
|
24 | 158142c2 | bellard | RESULT IN INCORRECT BEHAVIOR. USE OF THIS SOFTWARE IS RESTRICTED TO PERSONS
|
25 | 158142c2 | bellard | AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ALL LOSSES,
|
26 | 158142c2 | bellard | COSTS, OR OTHER PROBLEMS THEY INCUR DUE TO THE SOFTWARE, AND WHO FURTHERMORE
|
27 | 158142c2 | bellard | EFFECTIVELY INDEMNIFY JOHN HAUSER AND THE INTERNATIONAL COMPUTER SCIENCE
|
28 | 158142c2 | bellard | INSTITUTE (possibly via similar legal warning) AGAINST ALL LOSSES, COSTS, OR
|
29 | 158142c2 | bellard | OTHER PROBLEMS INCURRED BY THEIR CUSTOMERS AND CLIENTS DUE TO THE SOFTWARE.
|
30 | 158142c2 | bellard | |
31 | 158142c2 | bellard | Derivative works are acceptable, even for commercial purposes, so long as
|
32 | 158142c2 | bellard | (1) the source code for the derivative work includes prominent notice that
|
33 | 158142c2 | bellard | the work is derivative, and (2) the source code includes prominent notice with
|
34 | 158142c2 | bellard | these four paragraphs for those parts of this code that are retained.
|
35 | 158142c2 | bellard | |
36 | 158142c2 | bellard | =============================================================================*/
|
37 | 158142c2 | bellard | |
38 | 158142c2 | bellard | #include "softfloat.h" |
39 | 158142c2 | bellard | |
40 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
41 | 158142c2 | bellard | | Primitive arithmetic functions, including multi-word arithmetic, and
|
42 | 158142c2 | bellard | | division and square root approximations. (Can be specialized to target if
|
43 | 158142c2 | bellard | | desired.)
|
44 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
45 | 158142c2 | bellard | #include "softfloat-macros.h" |
46 | 158142c2 | bellard | |
47 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
48 | 158142c2 | bellard | | Functions and definitions to determine: (1) whether tininess for underflow
|
49 | 158142c2 | bellard | | is detected before or after rounding by default, (2) what (if anything)
|
50 | 158142c2 | bellard | | happens when exceptions are raised, (3) how signaling NaNs are distinguished
|
51 | 158142c2 | bellard | | from quiet NaNs, (4) the default generated quiet NaNs, and (5) how NaNs
|
52 | 158142c2 | bellard | | are propagated from function inputs to output. These details are target-
|
53 | 158142c2 | bellard | | specific.
|
54 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
55 | 158142c2 | bellard | #include "softfloat-specialize.h" |
56 | 158142c2 | bellard | |
57 | 158142c2 | bellard | void set_float_rounding_mode(int val STATUS_PARAM) |
58 | 158142c2 | bellard | { |
59 | 158142c2 | bellard | STATUS(float_rounding_mode) = val; |
60 | 158142c2 | bellard | } |
61 | 158142c2 | bellard | |
62 | 1d6bda35 | bellard | void set_float_exception_flags(int val STATUS_PARAM) |
63 | 1d6bda35 | bellard | { |
64 | 1d6bda35 | bellard | STATUS(float_exception_flags) = val; |
65 | 1d6bda35 | bellard | } |
66 | 1d6bda35 | bellard | |
67 | 158142c2 | bellard | void set_floatx80_rounding_precision(int val STATUS_PARAM) |
68 | 158142c2 | bellard | { |
69 | 158142c2 | bellard | STATUS(floatx80_rounding_precision) = val; |
70 | 158142c2 | bellard | } |
71 | 158142c2 | bellard | |
72 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
73 | bb4d4bb3 | Peter Maydell | | Returns the fraction bits of the half-precision floating-point value `a'.
|
74 | bb4d4bb3 | Peter Maydell | *----------------------------------------------------------------------------*/
|
75 | bb4d4bb3 | Peter Maydell | |
76 | bb4d4bb3 | Peter Maydell | INLINE uint32_t extractFloat16Frac(float16 a) |
77 | bb4d4bb3 | Peter Maydell | { |
78 | bb4d4bb3 | Peter Maydell | return float16_val(a) & 0x3ff; |
79 | bb4d4bb3 | Peter Maydell | } |
80 | bb4d4bb3 | Peter Maydell | |
81 | bb4d4bb3 | Peter Maydell | /*----------------------------------------------------------------------------
|
82 | bb4d4bb3 | Peter Maydell | | Returns the exponent bits of the half-precision floating-point value `a'.
|
83 | bb4d4bb3 | Peter Maydell | *----------------------------------------------------------------------------*/
|
84 | bb4d4bb3 | Peter Maydell | |
85 | bb4d4bb3 | Peter Maydell | INLINE int16 extractFloat16Exp(float16 a) |
86 | bb4d4bb3 | Peter Maydell | { |
87 | bb4d4bb3 | Peter Maydell | return (float16_val(a) >> 10) & 0x1f; |
88 | bb4d4bb3 | Peter Maydell | } |
89 | bb4d4bb3 | Peter Maydell | |
90 | bb4d4bb3 | Peter Maydell | /*----------------------------------------------------------------------------
|
91 | bb4d4bb3 | Peter Maydell | | Returns the sign bit of the single-precision floating-point value `a'.
|
92 | bb4d4bb3 | Peter Maydell | *----------------------------------------------------------------------------*/
|
93 | bb4d4bb3 | Peter Maydell | |
94 | bb4d4bb3 | Peter Maydell | INLINE flag extractFloat16Sign(float16 a) |
95 | bb4d4bb3 | Peter Maydell | { |
96 | bb4d4bb3 | Peter Maydell | return float16_val(a)>>15; |
97 | bb4d4bb3 | Peter Maydell | } |
98 | bb4d4bb3 | Peter Maydell | |
99 | bb4d4bb3 | Peter Maydell | /*----------------------------------------------------------------------------
|
100 | 158142c2 | bellard | | Takes a 64-bit fixed-point value `absZ' with binary point between bits 6
|
101 | 158142c2 | bellard | | and 7, and returns the properly rounded 32-bit integer corresponding to the
|
102 | 158142c2 | bellard | | input. If `zSign' is 1, the input is negated before being converted to an
|
103 | 158142c2 | bellard | | integer. Bit 63 of `absZ' must be zero. Ordinarily, the fixed-point input
|
104 | 158142c2 | bellard | | is simply rounded to an integer, with the inexact exception raised if the
|
105 | 158142c2 | bellard | | input cannot be represented exactly as an integer. However, if the fixed-
|
106 | 158142c2 | bellard | | point input is too large, the invalid exception is raised and the largest
|
107 | 158142c2 | bellard | | positive or negative integer is returned.
|
108 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
109 | 158142c2 | bellard | |
110 | bb98fe42 | Andreas Färber | static int32 roundAndPackInt32( flag zSign, uint64_t absZ STATUS_PARAM)
|
111 | 158142c2 | bellard | { |
112 | 158142c2 | bellard | int8 roundingMode; |
113 | 158142c2 | bellard | flag roundNearestEven; |
114 | 158142c2 | bellard | int8 roundIncrement, roundBits; |
115 | 158142c2 | bellard | int32 z; |
116 | 158142c2 | bellard | |
117 | 158142c2 | bellard | roundingMode = STATUS(float_rounding_mode); |
118 | 158142c2 | bellard | roundNearestEven = ( roundingMode == float_round_nearest_even ); |
119 | 158142c2 | bellard | roundIncrement = 0x40;
|
120 | 158142c2 | bellard | if ( ! roundNearestEven ) {
|
121 | 158142c2 | bellard | if ( roundingMode == float_round_to_zero ) {
|
122 | 158142c2 | bellard | roundIncrement = 0;
|
123 | 158142c2 | bellard | } |
124 | 158142c2 | bellard | else {
|
125 | 158142c2 | bellard | roundIncrement = 0x7F;
|
126 | 158142c2 | bellard | if ( zSign ) {
|
127 | 158142c2 | bellard | if ( roundingMode == float_round_up ) roundIncrement = 0; |
128 | 158142c2 | bellard | } |
129 | 158142c2 | bellard | else {
|
130 | 158142c2 | bellard | if ( roundingMode == float_round_down ) roundIncrement = 0; |
131 | 158142c2 | bellard | } |
132 | 158142c2 | bellard | } |
133 | 158142c2 | bellard | } |
134 | 158142c2 | bellard | roundBits = absZ & 0x7F;
|
135 | 158142c2 | bellard | absZ = ( absZ + roundIncrement )>>7;
|
136 | 158142c2 | bellard | absZ &= ~ ( ( ( roundBits ^ 0x40 ) == 0 ) & roundNearestEven ); |
137 | 158142c2 | bellard | z = absZ; |
138 | 158142c2 | bellard | if ( zSign ) z = - z;
|
139 | 158142c2 | bellard | if ( ( absZ>>32 ) || ( z && ( ( z < 0 ) ^ zSign ) ) ) { |
140 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
141 | bb98fe42 | Andreas Färber | return zSign ? (int32_t) 0x80000000 : 0x7FFFFFFF; |
142 | 158142c2 | bellard | } |
143 | 158142c2 | bellard | if ( roundBits ) STATUS(float_exception_flags) |= float_flag_inexact;
|
144 | 158142c2 | bellard | return z;
|
145 | 158142c2 | bellard | |
146 | 158142c2 | bellard | } |
147 | 158142c2 | bellard | |
148 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
149 | 158142c2 | bellard | | Takes the 128-bit fixed-point value formed by concatenating `absZ0' and
|
150 | 158142c2 | bellard | | `absZ1', with binary point between bits 63 and 64 (between the input words),
|
151 | 158142c2 | bellard | | and returns the properly rounded 64-bit integer corresponding to the input.
|
152 | 158142c2 | bellard | | If `zSign' is 1, the input is negated before being converted to an integer.
|
153 | 158142c2 | bellard | | Ordinarily, the fixed-point input is simply rounded to an integer, with
|
154 | 158142c2 | bellard | | the inexact exception raised if the input cannot be represented exactly as
|
155 | 158142c2 | bellard | | an integer. However, if the fixed-point input is too large, the invalid
|
156 | 158142c2 | bellard | | exception is raised and the largest positive or negative integer is
|
157 | 158142c2 | bellard | | returned.
|
158 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
159 | 158142c2 | bellard | |
160 | bb98fe42 | Andreas Färber | static int64 roundAndPackInt64( flag zSign, uint64_t absZ0, uint64_t absZ1 STATUS_PARAM)
|
161 | 158142c2 | bellard | { |
162 | 158142c2 | bellard | int8 roundingMode; |
163 | 158142c2 | bellard | flag roundNearestEven, increment; |
164 | 158142c2 | bellard | int64 z; |
165 | 158142c2 | bellard | |
166 | 158142c2 | bellard | roundingMode = STATUS(float_rounding_mode); |
167 | 158142c2 | bellard | roundNearestEven = ( roundingMode == float_round_nearest_even ); |
168 | bb98fe42 | Andreas Färber | increment = ( (int64_t) absZ1 < 0 );
|
169 | 158142c2 | bellard | if ( ! roundNearestEven ) {
|
170 | 158142c2 | bellard | if ( roundingMode == float_round_to_zero ) {
|
171 | 158142c2 | bellard | increment = 0;
|
172 | 158142c2 | bellard | } |
173 | 158142c2 | bellard | else {
|
174 | 158142c2 | bellard | if ( zSign ) {
|
175 | 158142c2 | bellard | increment = ( roundingMode == float_round_down ) && absZ1; |
176 | 158142c2 | bellard | } |
177 | 158142c2 | bellard | else {
|
178 | 158142c2 | bellard | increment = ( roundingMode == float_round_up ) && absZ1; |
179 | 158142c2 | bellard | } |
180 | 158142c2 | bellard | } |
181 | 158142c2 | bellard | } |
182 | 158142c2 | bellard | if ( increment ) {
|
183 | 158142c2 | bellard | ++absZ0; |
184 | 158142c2 | bellard | if ( absZ0 == 0 ) goto overflow; |
185 | bb98fe42 | Andreas Färber | absZ0 &= ~ ( ( (uint64_t) ( absZ1<<1 ) == 0 ) & roundNearestEven ); |
186 | 158142c2 | bellard | } |
187 | 158142c2 | bellard | z = absZ0; |
188 | 158142c2 | bellard | if ( zSign ) z = - z;
|
189 | 158142c2 | bellard | if ( z && ( ( z < 0 ) ^ zSign ) ) { |
190 | 158142c2 | bellard | overflow:
|
191 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
192 | 158142c2 | bellard | return
|
193 | bb98fe42 | Andreas Färber | zSign ? (int64_t) LIT64( 0x8000000000000000 )
|
194 | 158142c2 | bellard | : LIT64( 0x7FFFFFFFFFFFFFFF );
|
195 | 158142c2 | bellard | } |
196 | 158142c2 | bellard | if ( absZ1 ) STATUS(float_exception_flags) |= float_flag_inexact;
|
197 | 158142c2 | bellard | return z;
|
198 | 158142c2 | bellard | |
199 | 158142c2 | bellard | } |
200 | 158142c2 | bellard | |
201 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
202 | 158142c2 | bellard | | Returns the fraction bits of the single-precision floating-point value `a'.
|
203 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
204 | 158142c2 | bellard | |
205 | bb98fe42 | Andreas Färber | INLINE uint32_t extractFloat32Frac( float32 a ) |
206 | 158142c2 | bellard | { |
207 | 158142c2 | bellard | |
208 | f090c9d4 | pbrook | return float32_val(a) & 0x007FFFFF; |
209 | 158142c2 | bellard | |
210 | 158142c2 | bellard | } |
211 | 158142c2 | bellard | |
212 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
213 | 158142c2 | bellard | | Returns the exponent bits of the single-precision floating-point value `a'.
|
214 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
215 | 158142c2 | bellard | |
216 | 158142c2 | bellard | INLINE int16 extractFloat32Exp( float32 a ) |
217 | 158142c2 | bellard | { |
218 | 158142c2 | bellard | |
219 | f090c9d4 | pbrook | return ( float32_val(a)>>23 ) & 0xFF; |
220 | 158142c2 | bellard | |
221 | 158142c2 | bellard | } |
222 | 158142c2 | bellard | |
223 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
224 | 158142c2 | bellard | | Returns the sign bit of the single-precision floating-point value `a'.
|
225 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
226 | 158142c2 | bellard | |
227 | 158142c2 | bellard | INLINE flag extractFloat32Sign( float32 a ) |
228 | 158142c2 | bellard | { |
229 | 158142c2 | bellard | |
230 | f090c9d4 | pbrook | return float32_val(a)>>31; |
231 | 158142c2 | bellard | |
232 | 158142c2 | bellard | } |
233 | 158142c2 | bellard | |
234 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
235 | 37d18660 | Peter Maydell | | If `a' is denormal and we are in flush-to-zero mode then set the
|
236 | 37d18660 | Peter Maydell | | input-denormal exception and return zero. Otherwise just return the value.
|
237 | 37d18660 | Peter Maydell | *----------------------------------------------------------------------------*/
|
238 | 37d18660 | Peter Maydell | static float32 float32_squash_input_denormal(float32 a STATUS_PARAM)
|
239 | 37d18660 | Peter Maydell | { |
240 | 37d18660 | Peter Maydell | if (STATUS(flush_inputs_to_zero)) {
|
241 | 37d18660 | Peter Maydell | if (extractFloat32Exp(a) == 0 && extractFloat32Frac(a) != 0) { |
242 | 37d18660 | Peter Maydell | float_raise(float_flag_input_denormal STATUS_VAR); |
243 | 37d18660 | Peter Maydell | return make_float32(float32_val(a) & 0x80000000); |
244 | 37d18660 | Peter Maydell | } |
245 | 37d18660 | Peter Maydell | } |
246 | 37d18660 | Peter Maydell | return a;
|
247 | 37d18660 | Peter Maydell | } |
248 | 37d18660 | Peter Maydell | |
249 | 37d18660 | Peter Maydell | /*----------------------------------------------------------------------------
|
250 | 158142c2 | bellard | | Normalizes the subnormal single-precision floating-point value represented
|
251 | 158142c2 | bellard | | by the denormalized significand `aSig'. The normalized exponent and
|
252 | 158142c2 | bellard | | significand are stored at the locations pointed to by `zExpPtr' and
|
253 | 158142c2 | bellard | | `zSigPtr', respectively.
|
254 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
255 | 158142c2 | bellard | |
256 | 158142c2 | bellard | static void |
257 | bb98fe42 | Andreas Färber | normalizeFloat32Subnormal( uint32_t aSig, int16 *zExpPtr, uint32_t *zSigPtr ) |
258 | 158142c2 | bellard | { |
259 | 158142c2 | bellard | int8 shiftCount; |
260 | 158142c2 | bellard | |
261 | 158142c2 | bellard | shiftCount = countLeadingZeros32( aSig ) - 8;
|
262 | 158142c2 | bellard | *zSigPtr = aSig<<shiftCount; |
263 | 158142c2 | bellard | *zExpPtr = 1 - shiftCount;
|
264 | 158142c2 | bellard | |
265 | 158142c2 | bellard | } |
266 | 158142c2 | bellard | |
267 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
268 | 158142c2 | bellard | | Packs the sign `zSign', exponent `zExp', and significand `zSig' into a
|
269 | 158142c2 | bellard | | single-precision floating-point value, returning the result. After being
|
270 | 158142c2 | bellard | | shifted into the proper positions, the three fields are simply added
|
271 | 158142c2 | bellard | | together to form the result. This means that any integer portion of `zSig'
|
272 | 158142c2 | bellard | | will be added into the exponent. Since a properly normalized significand
|
273 | 158142c2 | bellard | | will have an integer portion equal to 1, the `zExp' input should be 1 less
|
274 | 158142c2 | bellard | | than the desired result exponent whenever `zSig' is a complete, normalized
|
275 | 158142c2 | bellard | | significand.
|
276 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
277 | 158142c2 | bellard | |
278 | bb98fe42 | Andreas Färber | INLINE float32 packFloat32( flag zSign, int16 zExp, uint32_t zSig ) |
279 | 158142c2 | bellard | { |
280 | 158142c2 | bellard | |
281 | f090c9d4 | pbrook | return make_float32(
|
282 | bb98fe42 | Andreas Färber | ( ( (uint32_t) zSign )<<31 ) + ( ( (uint32_t) zExp )<<23 ) + zSig); |
283 | 158142c2 | bellard | |
284 | 158142c2 | bellard | } |
285 | 158142c2 | bellard | |
286 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
287 | 158142c2 | bellard | | Takes an abstract floating-point value having sign `zSign', exponent `zExp',
|
288 | 158142c2 | bellard | | and significand `zSig', and returns the proper single-precision floating-
|
289 | 158142c2 | bellard | | point value corresponding to the abstract input. Ordinarily, the abstract
|
290 | 158142c2 | bellard | | value is simply rounded and packed into the single-precision format, with
|
291 | 158142c2 | bellard | | the inexact exception raised if the abstract input cannot be represented
|
292 | 158142c2 | bellard | | exactly. However, if the abstract value is too large, the overflow and
|
293 | 158142c2 | bellard | | inexact exceptions are raised and an infinity or maximal finite value is
|
294 | 158142c2 | bellard | | returned. If the abstract value is too small, the input value is rounded to
|
295 | 158142c2 | bellard | | a subnormal number, and the underflow and inexact exceptions are raised if
|
296 | 158142c2 | bellard | | the abstract input cannot be represented exactly as a subnormal single-
|
297 | 158142c2 | bellard | | precision floating-point number.
|
298 | 158142c2 | bellard | | The input significand `zSig' has its binary point between bits 30
|
299 | 158142c2 | bellard | | and 29, which is 7 bits to the left of the usual location. This shifted
|
300 | 158142c2 | bellard | | significand must be normalized or smaller. If `zSig' is not normalized,
|
301 | 158142c2 | bellard | | `zExp' must be 0; in that case, the result returned is a subnormal number,
|
302 | 158142c2 | bellard | | and it must not require rounding. In the usual case that `zSig' is
|
303 | 158142c2 | bellard | | normalized, `zExp' must be 1 less than the ``true'' floating-point exponent.
|
304 | 158142c2 | bellard | | The handling of underflow and overflow follows the IEC/IEEE Standard for
|
305 | 158142c2 | bellard | | Binary Floating-Point Arithmetic.
|
306 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
307 | 158142c2 | bellard | |
308 | bb98fe42 | Andreas Färber | static float32 roundAndPackFloat32( flag zSign, int16 zExp, uint32_t zSig STATUS_PARAM)
|
309 | 158142c2 | bellard | { |
310 | 158142c2 | bellard | int8 roundingMode; |
311 | 158142c2 | bellard | flag roundNearestEven; |
312 | 158142c2 | bellard | int8 roundIncrement, roundBits; |
313 | 158142c2 | bellard | flag isTiny; |
314 | 158142c2 | bellard | |
315 | 158142c2 | bellard | roundingMode = STATUS(float_rounding_mode); |
316 | 158142c2 | bellard | roundNearestEven = ( roundingMode == float_round_nearest_even ); |
317 | 158142c2 | bellard | roundIncrement = 0x40;
|
318 | 158142c2 | bellard | if ( ! roundNearestEven ) {
|
319 | 158142c2 | bellard | if ( roundingMode == float_round_to_zero ) {
|
320 | 158142c2 | bellard | roundIncrement = 0;
|
321 | 158142c2 | bellard | } |
322 | 158142c2 | bellard | else {
|
323 | 158142c2 | bellard | roundIncrement = 0x7F;
|
324 | 158142c2 | bellard | if ( zSign ) {
|
325 | 158142c2 | bellard | if ( roundingMode == float_round_up ) roundIncrement = 0; |
326 | 158142c2 | bellard | } |
327 | 158142c2 | bellard | else {
|
328 | 158142c2 | bellard | if ( roundingMode == float_round_down ) roundIncrement = 0; |
329 | 158142c2 | bellard | } |
330 | 158142c2 | bellard | } |
331 | 158142c2 | bellard | } |
332 | 158142c2 | bellard | roundBits = zSig & 0x7F;
|
333 | bb98fe42 | Andreas Färber | if ( 0xFD <= (uint16_t) zExp ) { |
334 | 158142c2 | bellard | if ( ( 0xFD < zExp ) |
335 | 158142c2 | bellard | || ( ( zExp == 0xFD )
|
336 | bb98fe42 | Andreas Färber | && ( (int32_t) ( zSig + roundIncrement ) < 0 ) )
|
337 | 158142c2 | bellard | ) { |
338 | 158142c2 | bellard | float_raise( float_flag_overflow | float_flag_inexact STATUS_VAR); |
339 | f090c9d4 | pbrook | return packFloat32( zSign, 0xFF, - ( roundIncrement == 0 )); |
340 | 158142c2 | bellard | } |
341 | 158142c2 | bellard | if ( zExp < 0 ) { |
342 | e6afc87f | Peter Maydell | if (STATUS(flush_to_zero)) {
|
343 | e6afc87f | Peter Maydell | float_raise(float_flag_output_denormal STATUS_VAR); |
344 | e6afc87f | Peter Maydell | return packFloat32(zSign, 0, 0); |
345 | e6afc87f | Peter Maydell | } |
346 | 158142c2 | bellard | isTiny = |
347 | 158142c2 | bellard | ( STATUS(float_detect_tininess) == float_tininess_before_rounding ) |
348 | 158142c2 | bellard | || ( zExp < -1 )
|
349 | 158142c2 | bellard | || ( zSig + roundIncrement < 0x80000000 );
|
350 | 158142c2 | bellard | shift32RightJamming( zSig, - zExp, &zSig ); |
351 | 158142c2 | bellard | zExp = 0;
|
352 | 158142c2 | bellard | roundBits = zSig & 0x7F;
|
353 | 158142c2 | bellard | if ( isTiny && roundBits ) float_raise( float_flag_underflow STATUS_VAR);
|
354 | 158142c2 | bellard | } |
355 | 158142c2 | bellard | } |
356 | 158142c2 | bellard | if ( roundBits ) STATUS(float_exception_flags) |= float_flag_inexact;
|
357 | 158142c2 | bellard | zSig = ( zSig + roundIncrement )>>7;
|
358 | 158142c2 | bellard | zSig &= ~ ( ( ( roundBits ^ 0x40 ) == 0 ) & roundNearestEven ); |
359 | 158142c2 | bellard | if ( zSig == 0 ) zExp = 0; |
360 | 158142c2 | bellard | return packFloat32( zSign, zExp, zSig );
|
361 | 158142c2 | bellard | |
362 | 158142c2 | bellard | } |
363 | 158142c2 | bellard | |
364 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
365 | 158142c2 | bellard | | Takes an abstract floating-point value having sign `zSign', exponent `zExp',
|
366 | 158142c2 | bellard | | and significand `zSig', and returns the proper single-precision floating-
|
367 | 158142c2 | bellard | | point value corresponding to the abstract input. This routine is just like
|
368 | 158142c2 | bellard | | `roundAndPackFloat32' except that `zSig' does not have to be normalized.
|
369 | 158142c2 | bellard | | Bit 31 of `zSig' must be zero, and `zExp' must be 1 less than the ``true''
|
370 | 158142c2 | bellard | | floating-point exponent.
|
371 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
372 | 158142c2 | bellard | |
373 | 158142c2 | bellard | static float32
|
374 | bb98fe42 | Andreas Färber | normalizeRoundAndPackFloat32( flag zSign, int16 zExp, uint32_t zSig STATUS_PARAM) |
375 | 158142c2 | bellard | { |
376 | 158142c2 | bellard | int8 shiftCount; |
377 | 158142c2 | bellard | |
378 | 158142c2 | bellard | shiftCount = countLeadingZeros32( zSig ) - 1;
|
379 | 158142c2 | bellard | return roundAndPackFloat32( zSign, zExp - shiftCount, zSig<<shiftCount STATUS_VAR);
|
380 | 158142c2 | bellard | |
381 | 158142c2 | bellard | } |
382 | 158142c2 | bellard | |
383 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
384 | 158142c2 | bellard | | Returns the fraction bits of the double-precision floating-point value `a'.
|
385 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
386 | 158142c2 | bellard | |
387 | bb98fe42 | Andreas Färber | INLINE uint64_t extractFloat64Frac( float64 a ) |
388 | 158142c2 | bellard | { |
389 | 158142c2 | bellard | |
390 | f090c9d4 | pbrook | return float64_val(a) & LIT64( 0x000FFFFFFFFFFFFF ); |
391 | 158142c2 | bellard | |
392 | 158142c2 | bellard | } |
393 | 158142c2 | bellard | |
394 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
395 | 158142c2 | bellard | | Returns the exponent bits of the double-precision floating-point value `a'.
|
396 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
397 | 158142c2 | bellard | |
398 | 158142c2 | bellard | INLINE int16 extractFloat64Exp( float64 a ) |
399 | 158142c2 | bellard | { |
400 | 158142c2 | bellard | |
401 | f090c9d4 | pbrook | return ( float64_val(a)>>52 ) & 0x7FF; |
402 | 158142c2 | bellard | |
403 | 158142c2 | bellard | } |
404 | 158142c2 | bellard | |
405 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
406 | 158142c2 | bellard | | Returns the sign bit of the double-precision floating-point value `a'.
|
407 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
408 | 158142c2 | bellard | |
409 | 158142c2 | bellard | INLINE flag extractFloat64Sign( float64 a ) |
410 | 158142c2 | bellard | { |
411 | 158142c2 | bellard | |
412 | f090c9d4 | pbrook | return float64_val(a)>>63; |
413 | 158142c2 | bellard | |
414 | 158142c2 | bellard | } |
415 | 158142c2 | bellard | |
416 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
417 | 37d18660 | Peter Maydell | | If `a' is denormal and we are in flush-to-zero mode then set the
|
418 | 37d18660 | Peter Maydell | | input-denormal exception and return zero. Otherwise just return the value.
|
419 | 37d18660 | Peter Maydell | *----------------------------------------------------------------------------*/
|
420 | 37d18660 | Peter Maydell | static float64 float64_squash_input_denormal(float64 a STATUS_PARAM)
|
421 | 37d18660 | Peter Maydell | { |
422 | 37d18660 | Peter Maydell | if (STATUS(flush_inputs_to_zero)) {
|
423 | 37d18660 | Peter Maydell | if (extractFloat64Exp(a) == 0 && extractFloat64Frac(a) != 0) { |
424 | 37d18660 | Peter Maydell | float_raise(float_flag_input_denormal STATUS_VAR); |
425 | 37d18660 | Peter Maydell | return make_float64(float64_val(a) & (1ULL << 63)); |
426 | 37d18660 | Peter Maydell | } |
427 | 37d18660 | Peter Maydell | } |
428 | 37d18660 | Peter Maydell | return a;
|
429 | 37d18660 | Peter Maydell | } |
430 | 37d18660 | Peter Maydell | |
431 | 37d18660 | Peter Maydell | /*----------------------------------------------------------------------------
|
432 | 158142c2 | bellard | | Normalizes the subnormal double-precision floating-point value represented
|
433 | 158142c2 | bellard | | by the denormalized significand `aSig'. The normalized exponent and
|
434 | 158142c2 | bellard | | significand are stored at the locations pointed to by `zExpPtr' and
|
435 | 158142c2 | bellard | | `zSigPtr', respectively.
|
436 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
437 | 158142c2 | bellard | |
438 | 158142c2 | bellard | static void |
439 | bb98fe42 | Andreas Färber | normalizeFloat64Subnormal( uint64_t aSig, int16 *zExpPtr, uint64_t *zSigPtr ) |
440 | 158142c2 | bellard | { |
441 | 158142c2 | bellard | int8 shiftCount; |
442 | 158142c2 | bellard | |
443 | 158142c2 | bellard | shiftCount = countLeadingZeros64( aSig ) - 11;
|
444 | 158142c2 | bellard | *zSigPtr = aSig<<shiftCount; |
445 | 158142c2 | bellard | *zExpPtr = 1 - shiftCount;
|
446 | 158142c2 | bellard | |
447 | 158142c2 | bellard | } |
448 | 158142c2 | bellard | |
449 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
450 | 158142c2 | bellard | | Packs the sign `zSign', exponent `zExp', and significand `zSig' into a
|
451 | 158142c2 | bellard | | double-precision floating-point value, returning the result. After being
|
452 | 158142c2 | bellard | | shifted into the proper positions, the three fields are simply added
|
453 | 158142c2 | bellard | | together to form the result. This means that any integer portion of `zSig'
|
454 | 158142c2 | bellard | | will be added into the exponent. Since a properly normalized significand
|
455 | 158142c2 | bellard | | will have an integer portion equal to 1, the `zExp' input should be 1 less
|
456 | 158142c2 | bellard | | than the desired result exponent whenever `zSig' is a complete, normalized
|
457 | 158142c2 | bellard | | significand.
|
458 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
459 | 158142c2 | bellard | |
460 | bb98fe42 | Andreas Färber | INLINE float64 packFloat64( flag zSign, int16 zExp, uint64_t zSig ) |
461 | 158142c2 | bellard | { |
462 | 158142c2 | bellard | |
463 | f090c9d4 | pbrook | return make_float64(
|
464 | bb98fe42 | Andreas Färber | ( ( (uint64_t) zSign )<<63 ) + ( ( (uint64_t) zExp )<<52 ) + zSig); |
465 | 158142c2 | bellard | |
466 | 158142c2 | bellard | } |
467 | 158142c2 | bellard | |
468 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
469 | 158142c2 | bellard | | Takes an abstract floating-point value having sign `zSign', exponent `zExp',
|
470 | 158142c2 | bellard | | and significand `zSig', and returns the proper double-precision floating-
|
471 | 158142c2 | bellard | | point value corresponding to the abstract input. Ordinarily, the abstract
|
472 | 158142c2 | bellard | | value is simply rounded and packed into the double-precision format, with
|
473 | 158142c2 | bellard | | the inexact exception raised if the abstract input cannot be represented
|
474 | 158142c2 | bellard | | exactly. However, if the abstract value is too large, the overflow and
|
475 | 158142c2 | bellard | | inexact exceptions are raised and an infinity or maximal finite value is
|
476 | 158142c2 | bellard | | returned. If the abstract value is too small, the input value is rounded
|
477 | 158142c2 | bellard | | to a subnormal number, and the underflow and inexact exceptions are raised
|
478 | 158142c2 | bellard | | if the abstract input cannot be represented exactly as a subnormal double-
|
479 | 158142c2 | bellard | | precision floating-point number.
|
480 | 158142c2 | bellard | | The input significand `zSig' has its binary point between bits 62
|
481 | 158142c2 | bellard | | and 61, which is 10 bits to the left of the usual location. This shifted
|
482 | 158142c2 | bellard | | significand must be normalized or smaller. If `zSig' is not normalized,
|
483 | 158142c2 | bellard | | `zExp' must be 0; in that case, the result returned is a subnormal number,
|
484 | 158142c2 | bellard | | and it must not require rounding. In the usual case that `zSig' is
|
485 | 158142c2 | bellard | | normalized, `zExp' must be 1 less than the ``true'' floating-point exponent.
|
486 | 158142c2 | bellard | | The handling of underflow and overflow follows the IEC/IEEE Standard for
|
487 | 158142c2 | bellard | | Binary Floating-Point Arithmetic.
|
488 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
489 | 158142c2 | bellard | |
490 | bb98fe42 | Andreas Färber | static float64 roundAndPackFloat64( flag zSign, int16 zExp, uint64_t zSig STATUS_PARAM)
|
491 | 158142c2 | bellard | { |
492 | 158142c2 | bellard | int8 roundingMode; |
493 | 158142c2 | bellard | flag roundNearestEven; |
494 | 158142c2 | bellard | int16 roundIncrement, roundBits; |
495 | 158142c2 | bellard | flag isTiny; |
496 | 158142c2 | bellard | |
497 | 158142c2 | bellard | roundingMode = STATUS(float_rounding_mode); |
498 | 158142c2 | bellard | roundNearestEven = ( roundingMode == float_round_nearest_even ); |
499 | 158142c2 | bellard | roundIncrement = 0x200;
|
500 | 158142c2 | bellard | if ( ! roundNearestEven ) {
|
501 | 158142c2 | bellard | if ( roundingMode == float_round_to_zero ) {
|
502 | 158142c2 | bellard | roundIncrement = 0;
|
503 | 158142c2 | bellard | } |
504 | 158142c2 | bellard | else {
|
505 | 158142c2 | bellard | roundIncrement = 0x3FF;
|
506 | 158142c2 | bellard | if ( zSign ) {
|
507 | 158142c2 | bellard | if ( roundingMode == float_round_up ) roundIncrement = 0; |
508 | 158142c2 | bellard | } |
509 | 158142c2 | bellard | else {
|
510 | 158142c2 | bellard | if ( roundingMode == float_round_down ) roundIncrement = 0; |
511 | 158142c2 | bellard | } |
512 | 158142c2 | bellard | } |
513 | 158142c2 | bellard | } |
514 | 158142c2 | bellard | roundBits = zSig & 0x3FF;
|
515 | bb98fe42 | Andreas Färber | if ( 0x7FD <= (uint16_t) zExp ) { |
516 | 158142c2 | bellard | if ( ( 0x7FD < zExp ) |
517 | 158142c2 | bellard | || ( ( zExp == 0x7FD )
|
518 | bb98fe42 | Andreas Färber | && ( (int64_t) ( zSig + roundIncrement ) < 0 ) )
|
519 | 158142c2 | bellard | ) { |
520 | 158142c2 | bellard | float_raise( float_flag_overflow | float_flag_inexact STATUS_VAR); |
521 | f090c9d4 | pbrook | return packFloat64( zSign, 0x7FF, - ( roundIncrement == 0 )); |
522 | 158142c2 | bellard | } |
523 | 158142c2 | bellard | if ( zExp < 0 ) { |
524 | e6afc87f | Peter Maydell | if (STATUS(flush_to_zero)) {
|
525 | e6afc87f | Peter Maydell | float_raise(float_flag_output_denormal STATUS_VAR); |
526 | e6afc87f | Peter Maydell | return packFloat64(zSign, 0, 0); |
527 | e6afc87f | Peter Maydell | } |
528 | 158142c2 | bellard | isTiny = |
529 | 158142c2 | bellard | ( STATUS(float_detect_tininess) == float_tininess_before_rounding ) |
530 | 158142c2 | bellard | || ( zExp < -1 )
|
531 | 158142c2 | bellard | || ( zSig + roundIncrement < LIT64( 0x8000000000000000 ) );
|
532 | 158142c2 | bellard | shift64RightJamming( zSig, - zExp, &zSig ); |
533 | 158142c2 | bellard | zExp = 0;
|
534 | 158142c2 | bellard | roundBits = zSig & 0x3FF;
|
535 | 158142c2 | bellard | if ( isTiny && roundBits ) float_raise( float_flag_underflow STATUS_VAR);
|
536 | 158142c2 | bellard | } |
537 | 158142c2 | bellard | } |
538 | 158142c2 | bellard | if ( roundBits ) STATUS(float_exception_flags) |= float_flag_inexact;
|
539 | 158142c2 | bellard | zSig = ( zSig + roundIncrement )>>10;
|
540 | 158142c2 | bellard | zSig &= ~ ( ( ( roundBits ^ 0x200 ) == 0 ) & roundNearestEven ); |
541 | 158142c2 | bellard | if ( zSig == 0 ) zExp = 0; |
542 | 158142c2 | bellard | return packFloat64( zSign, zExp, zSig );
|
543 | 158142c2 | bellard | |
544 | 158142c2 | bellard | } |
545 | 158142c2 | bellard | |
546 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
547 | 158142c2 | bellard | | Takes an abstract floating-point value having sign `zSign', exponent `zExp',
|
548 | 158142c2 | bellard | | and significand `zSig', and returns the proper double-precision floating-
|
549 | 158142c2 | bellard | | point value corresponding to the abstract input. This routine is just like
|
550 | 158142c2 | bellard | | `roundAndPackFloat64' except that `zSig' does not have to be normalized.
|
551 | 158142c2 | bellard | | Bit 63 of `zSig' must be zero, and `zExp' must be 1 less than the ``true''
|
552 | 158142c2 | bellard | | floating-point exponent.
|
553 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
554 | 158142c2 | bellard | |
555 | 158142c2 | bellard | static float64
|
556 | bb98fe42 | Andreas Färber | normalizeRoundAndPackFloat64( flag zSign, int16 zExp, uint64_t zSig STATUS_PARAM) |
557 | 158142c2 | bellard | { |
558 | 158142c2 | bellard | int8 shiftCount; |
559 | 158142c2 | bellard | |
560 | 158142c2 | bellard | shiftCount = countLeadingZeros64( zSig ) - 1;
|
561 | 158142c2 | bellard | return roundAndPackFloat64( zSign, zExp - shiftCount, zSig<<shiftCount STATUS_VAR);
|
562 | 158142c2 | bellard | |
563 | 158142c2 | bellard | } |
564 | 158142c2 | bellard | |
565 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
566 | 158142c2 | bellard | | Returns the fraction bits of the extended double-precision floating-point
|
567 | 158142c2 | bellard | | value `a'.
|
568 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
569 | 158142c2 | bellard | |
570 | bb98fe42 | Andreas Färber | INLINE uint64_t extractFloatx80Frac( floatx80 a ) |
571 | 158142c2 | bellard | { |
572 | 158142c2 | bellard | |
573 | 158142c2 | bellard | return a.low;
|
574 | 158142c2 | bellard | |
575 | 158142c2 | bellard | } |
576 | 158142c2 | bellard | |
577 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
578 | 158142c2 | bellard | | Returns the exponent bits of the extended double-precision floating-point
|
579 | 158142c2 | bellard | | value `a'.
|
580 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
581 | 158142c2 | bellard | |
582 | 158142c2 | bellard | INLINE int32 extractFloatx80Exp( floatx80 a ) |
583 | 158142c2 | bellard | { |
584 | 158142c2 | bellard | |
585 | 158142c2 | bellard | return a.high & 0x7FFF; |
586 | 158142c2 | bellard | |
587 | 158142c2 | bellard | } |
588 | 158142c2 | bellard | |
589 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
590 | 158142c2 | bellard | | Returns the sign bit of the extended double-precision floating-point value
|
591 | 158142c2 | bellard | | `a'.
|
592 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
593 | 158142c2 | bellard | |
594 | 158142c2 | bellard | INLINE flag extractFloatx80Sign( floatx80 a ) |
595 | 158142c2 | bellard | { |
596 | 158142c2 | bellard | |
597 | 158142c2 | bellard | return a.high>>15; |
598 | 158142c2 | bellard | |
599 | 158142c2 | bellard | } |
600 | 158142c2 | bellard | |
601 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
602 | 158142c2 | bellard | | Normalizes the subnormal extended double-precision floating-point value
|
603 | 158142c2 | bellard | | represented by the denormalized significand `aSig'. The normalized exponent
|
604 | 158142c2 | bellard | | and significand are stored at the locations pointed to by `zExpPtr' and
|
605 | 158142c2 | bellard | | `zSigPtr', respectively.
|
606 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
607 | 158142c2 | bellard | |
608 | 158142c2 | bellard | static void |
609 | bb98fe42 | Andreas Färber | normalizeFloatx80Subnormal( uint64_t aSig, int32 *zExpPtr, uint64_t *zSigPtr ) |
610 | 158142c2 | bellard | { |
611 | 158142c2 | bellard | int8 shiftCount; |
612 | 158142c2 | bellard | |
613 | 158142c2 | bellard | shiftCount = countLeadingZeros64( aSig ); |
614 | 158142c2 | bellard | *zSigPtr = aSig<<shiftCount; |
615 | 158142c2 | bellard | *zExpPtr = 1 - shiftCount;
|
616 | 158142c2 | bellard | |
617 | 158142c2 | bellard | } |
618 | 158142c2 | bellard | |
619 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
620 | 158142c2 | bellard | | Packs the sign `zSign', exponent `zExp', and significand `zSig' into an
|
621 | 158142c2 | bellard | | extended double-precision floating-point value, returning the result.
|
622 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
623 | 158142c2 | bellard | |
624 | bb98fe42 | Andreas Färber | INLINE floatx80 packFloatx80( flag zSign, int32 zExp, uint64_t zSig ) |
625 | 158142c2 | bellard | { |
626 | 158142c2 | bellard | floatx80 z; |
627 | 158142c2 | bellard | |
628 | 158142c2 | bellard | z.low = zSig; |
629 | bb98fe42 | Andreas Färber | z.high = ( ( (uint16_t) zSign )<<15 ) + zExp;
|
630 | 158142c2 | bellard | return z;
|
631 | 158142c2 | bellard | |
632 | 158142c2 | bellard | } |
633 | 158142c2 | bellard | |
634 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
635 | 158142c2 | bellard | | Takes an abstract floating-point value having sign `zSign', exponent `zExp',
|
636 | 158142c2 | bellard | | and extended significand formed by the concatenation of `zSig0' and `zSig1',
|
637 | 158142c2 | bellard | | and returns the proper extended double-precision floating-point value
|
638 | 158142c2 | bellard | | corresponding to the abstract input. Ordinarily, the abstract value is
|
639 | 158142c2 | bellard | | rounded and packed into the extended double-precision format, with the
|
640 | 158142c2 | bellard | | inexact exception raised if the abstract input cannot be represented
|
641 | 158142c2 | bellard | | exactly. However, if the abstract value is too large, the overflow and
|
642 | 158142c2 | bellard | | inexact exceptions are raised and an infinity or maximal finite value is
|
643 | 158142c2 | bellard | | returned. If the abstract value is too small, the input value is rounded to
|
644 | 158142c2 | bellard | | a subnormal number, and the underflow and inexact exceptions are raised if
|
645 | 158142c2 | bellard | | the abstract input cannot be represented exactly as a subnormal extended
|
646 | 158142c2 | bellard | | double-precision floating-point number.
|
647 | 158142c2 | bellard | | If `roundingPrecision' is 32 or 64, the result is rounded to the same
|
648 | 158142c2 | bellard | | number of bits as single or double precision, respectively. Otherwise, the
|
649 | 158142c2 | bellard | | result is rounded to the full precision of the extended double-precision
|
650 | 158142c2 | bellard | | format.
|
651 | 158142c2 | bellard | | The input significand must be normalized or smaller. If the input
|
652 | 158142c2 | bellard | | significand is not normalized, `zExp' must be 0; in that case, the result
|
653 | 158142c2 | bellard | | returned is a subnormal number, and it must not require rounding. The
|
654 | 158142c2 | bellard | | handling of underflow and overflow follows the IEC/IEEE Standard for Binary
|
655 | 158142c2 | bellard | | Floating-Point Arithmetic.
|
656 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
657 | 158142c2 | bellard | |
658 | 158142c2 | bellard | static floatx80
|
659 | 158142c2 | bellard | roundAndPackFloatx80( |
660 | bb98fe42 | Andreas Färber | int8 roundingPrecision, flag zSign, int32 zExp, uint64_t zSig0, uint64_t zSig1 |
661 | 158142c2 | bellard | STATUS_PARAM) |
662 | 158142c2 | bellard | { |
663 | 158142c2 | bellard | int8 roundingMode; |
664 | 158142c2 | bellard | flag roundNearestEven, increment, isTiny; |
665 | 158142c2 | bellard | int64 roundIncrement, roundMask, roundBits; |
666 | 158142c2 | bellard | |
667 | 158142c2 | bellard | roundingMode = STATUS(float_rounding_mode); |
668 | 158142c2 | bellard | roundNearestEven = ( roundingMode == float_round_nearest_even ); |
669 | 158142c2 | bellard | if ( roundingPrecision == 80 ) goto precision80; |
670 | 158142c2 | bellard | if ( roundingPrecision == 64 ) { |
671 | 158142c2 | bellard | roundIncrement = LIT64( 0x0000000000000400 );
|
672 | 158142c2 | bellard | roundMask = LIT64( 0x00000000000007FF );
|
673 | 158142c2 | bellard | } |
674 | 158142c2 | bellard | else if ( roundingPrecision == 32 ) { |
675 | 158142c2 | bellard | roundIncrement = LIT64( 0x0000008000000000 );
|
676 | 158142c2 | bellard | roundMask = LIT64( 0x000000FFFFFFFFFF );
|
677 | 158142c2 | bellard | } |
678 | 158142c2 | bellard | else {
|
679 | 158142c2 | bellard | goto precision80;
|
680 | 158142c2 | bellard | } |
681 | 158142c2 | bellard | zSig0 |= ( zSig1 != 0 );
|
682 | 158142c2 | bellard | if ( ! roundNearestEven ) {
|
683 | 158142c2 | bellard | if ( roundingMode == float_round_to_zero ) {
|
684 | 158142c2 | bellard | roundIncrement = 0;
|
685 | 158142c2 | bellard | } |
686 | 158142c2 | bellard | else {
|
687 | 158142c2 | bellard | roundIncrement = roundMask; |
688 | 158142c2 | bellard | if ( zSign ) {
|
689 | 158142c2 | bellard | if ( roundingMode == float_round_up ) roundIncrement = 0; |
690 | 158142c2 | bellard | } |
691 | 158142c2 | bellard | else {
|
692 | 158142c2 | bellard | if ( roundingMode == float_round_down ) roundIncrement = 0; |
693 | 158142c2 | bellard | } |
694 | 158142c2 | bellard | } |
695 | 158142c2 | bellard | } |
696 | 158142c2 | bellard | roundBits = zSig0 & roundMask; |
697 | bb98fe42 | Andreas Färber | if ( 0x7FFD <= (uint32_t) ( zExp - 1 ) ) { |
698 | 158142c2 | bellard | if ( ( 0x7FFE < zExp ) |
699 | 158142c2 | bellard | || ( ( zExp == 0x7FFE ) && ( zSig0 + roundIncrement < zSig0 ) )
|
700 | 158142c2 | bellard | ) { |
701 | 158142c2 | bellard | goto overflow;
|
702 | 158142c2 | bellard | } |
703 | 158142c2 | bellard | if ( zExp <= 0 ) { |
704 | e6afc87f | Peter Maydell | if (STATUS(flush_to_zero)) {
|
705 | e6afc87f | Peter Maydell | float_raise(float_flag_output_denormal STATUS_VAR); |
706 | e6afc87f | Peter Maydell | return packFloatx80(zSign, 0, 0); |
707 | e6afc87f | Peter Maydell | } |
708 | 158142c2 | bellard | isTiny = |
709 | 158142c2 | bellard | ( STATUS(float_detect_tininess) == float_tininess_before_rounding ) |
710 | 158142c2 | bellard | || ( zExp < 0 )
|
711 | 158142c2 | bellard | || ( zSig0 <= zSig0 + roundIncrement ); |
712 | 158142c2 | bellard | shift64RightJamming( zSig0, 1 - zExp, &zSig0 );
|
713 | 158142c2 | bellard | zExp = 0;
|
714 | 158142c2 | bellard | roundBits = zSig0 & roundMask; |
715 | 158142c2 | bellard | if ( isTiny && roundBits ) float_raise( float_flag_underflow STATUS_VAR);
|
716 | 158142c2 | bellard | if ( roundBits ) STATUS(float_exception_flags) |= float_flag_inexact;
|
717 | 158142c2 | bellard | zSig0 += roundIncrement; |
718 | bb98fe42 | Andreas Färber | if ( (int64_t) zSig0 < 0 ) zExp = 1; |
719 | 158142c2 | bellard | roundIncrement = roundMask + 1;
|
720 | 158142c2 | bellard | if ( roundNearestEven && ( roundBits<<1 == roundIncrement ) ) { |
721 | 158142c2 | bellard | roundMask |= roundIncrement; |
722 | 158142c2 | bellard | } |
723 | 158142c2 | bellard | zSig0 &= ~ roundMask; |
724 | 158142c2 | bellard | return packFloatx80( zSign, zExp, zSig0 );
|
725 | 158142c2 | bellard | } |
726 | 158142c2 | bellard | } |
727 | 158142c2 | bellard | if ( roundBits ) STATUS(float_exception_flags) |= float_flag_inexact;
|
728 | 158142c2 | bellard | zSig0 += roundIncrement; |
729 | 158142c2 | bellard | if ( zSig0 < roundIncrement ) {
|
730 | 158142c2 | bellard | ++zExp; |
731 | 158142c2 | bellard | zSig0 = LIT64( 0x8000000000000000 );
|
732 | 158142c2 | bellard | } |
733 | 158142c2 | bellard | roundIncrement = roundMask + 1;
|
734 | 158142c2 | bellard | if ( roundNearestEven && ( roundBits<<1 == roundIncrement ) ) { |
735 | 158142c2 | bellard | roundMask |= roundIncrement; |
736 | 158142c2 | bellard | } |
737 | 158142c2 | bellard | zSig0 &= ~ roundMask; |
738 | 158142c2 | bellard | if ( zSig0 == 0 ) zExp = 0; |
739 | 158142c2 | bellard | return packFloatx80( zSign, zExp, zSig0 );
|
740 | 158142c2 | bellard | precision80:
|
741 | bb98fe42 | Andreas Färber | increment = ( (int64_t) zSig1 < 0 );
|
742 | 158142c2 | bellard | if ( ! roundNearestEven ) {
|
743 | 158142c2 | bellard | if ( roundingMode == float_round_to_zero ) {
|
744 | 158142c2 | bellard | increment = 0;
|
745 | 158142c2 | bellard | } |
746 | 158142c2 | bellard | else {
|
747 | 158142c2 | bellard | if ( zSign ) {
|
748 | 158142c2 | bellard | increment = ( roundingMode == float_round_down ) && zSig1; |
749 | 158142c2 | bellard | } |
750 | 158142c2 | bellard | else {
|
751 | 158142c2 | bellard | increment = ( roundingMode == float_round_up ) && zSig1; |
752 | 158142c2 | bellard | } |
753 | 158142c2 | bellard | } |
754 | 158142c2 | bellard | } |
755 | bb98fe42 | Andreas Färber | if ( 0x7FFD <= (uint32_t) ( zExp - 1 ) ) { |
756 | 158142c2 | bellard | if ( ( 0x7FFE < zExp ) |
757 | 158142c2 | bellard | || ( ( zExp == 0x7FFE )
|
758 | 158142c2 | bellard | && ( zSig0 == LIT64( 0xFFFFFFFFFFFFFFFF ) )
|
759 | 158142c2 | bellard | && increment |
760 | 158142c2 | bellard | ) |
761 | 158142c2 | bellard | ) { |
762 | 158142c2 | bellard | roundMask = 0;
|
763 | 158142c2 | bellard | overflow:
|
764 | 158142c2 | bellard | float_raise( float_flag_overflow | float_flag_inexact STATUS_VAR); |
765 | 158142c2 | bellard | if ( ( roundingMode == float_round_to_zero )
|
766 | 158142c2 | bellard | || ( zSign && ( roundingMode == float_round_up ) ) |
767 | 158142c2 | bellard | || ( ! zSign && ( roundingMode == float_round_down ) ) |
768 | 158142c2 | bellard | ) { |
769 | 158142c2 | bellard | return packFloatx80( zSign, 0x7FFE, ~ roundMask ); |
770 | 158142c2 | bellard | } |
771 | 158142c2 | bellard | return packFloatx80( zSign, 0x7FFF, LIT64( 0x8000000000000000 ) ); |
772 | 158142c2 | bellard | } |
773 | 158142c2 | bellard | if ( zExp <= 0 ) { |
774 | 158142c2 | bellard | isTiny = |
775 | 158142c2 | bellard | ( STATUS(float_detect_tininess) == float_tininess_before_rounding ) |
776 | 158142c2 | bellard | || ( zExp < 0 )
|
777 | 158142c2 | bellard | || ! increment |
778 | 158142c2 | bellard | || ( zSig0 < LIT64( 0xFFFFFFFFFFFFFFFF ) );
|
779 | 158142c2 | bellard | shift64ExtraRightJamming( zSig0, zSig1, 1 - zExp, &zSig0, &zSig1 );
|
780 | 158142c2 | bellard | zExp = 0;
|
781 | 158142c2 | bellard | if ( isTiny && zSig1 ) float_raise( float_flag_underflow STATUS_VAR);
|
782 | 158142c2 | bellard | if ( zSig1 ) STATUS(float_exception_flags) |= float_flag_inexact;
|
783 | 158142c2 | bellard | if ( roundNearestEven ) {
|
784 | bb98fe42 | Andreas Färber | increment = ( (int64_t) zSig1 < 0 );
|
785 | 158142c2 | bellard | } |
786 | 158142c2 | bellard | else {
|
787 | 158142c2 | bellard | if ( zSign ) {
|
788 | 158142c2 | bellard | increment = ( roundingMode == float_round_down ) && zSig1; |
789 | 158142c2 | bellard | } |
790 | 158142c2 | bellard | else {
|
791 | 158142c2 | bellard | increment = ( roundingMode == float_round_up ) && zSig1; |
792 | 158142c2 | bellard | } |
793 | 158142c2 | bellard | } |
794 | 158142c2 | bellard | if ( increment ) {
|
795 | 158142c2 | bellard | ++zSig0; |
796 | 158142c2 | bellard | zSig0 &= |
797 | bb98fe42 | Andreas Färber | ~ ( ( (uint64_t) ( zSig1<<1 ) == 0 ) & roundNearestEven ); |
798 | bb98fe42 | Andreas Färber | if ( (int64_t) zSig0 < 0 ) zExp = 1; |
799 | 158142c2 | bellard | } |
800 | 158142c2 | bellard | return packFloatx80( zSign, zExp, zSig0 );
|
801 | 158142c2 | bellard | } |
802 | 158142c2 | bellard | } |
803 | 158142c2 | bellard | if ( zSig1 ) STATUS(float_exception_flags) |= float_flag_inexact;
|
804 | 158142c2 | bellard | if ( increment ) {
|
805 | 158142c2 | bellard | ++zSig0; |
806 | 158142c2 | bellard | if ( zSig0 == 0 ) { |
807 | 158142c2 | bellard | ++zExp; |
808 | 158142c2 | bellard | zSig0 = LIT64( 0x8000000000000000 );
|
809 | 158142c2 | bellard | } |
810 | 158142c2 | bellard | else {
|
811 | bb98fe42 | Andreas Färber | zSig0 &= ~ ( ( (uint64_t) ( zSig1<<1 ) == 0 ) & roundNearestEven ); |
812 | 158142c2 | bellard | } |
813 | 158142c2 | bellard | } |
814 | 158142c2 | bellard | else {
|
815 | 158142c2 | bellard | if ( zSig0 == 0 ) zExp = 0; |
816 | 158142c2 | bellard | } |
817 | 158142c2 | bellard | return packFloatx80( zSign, zExp, zSig0 );
|
818 | 158142c2 | bellard | |
819 | 158142c2 | bellard | } |
820 | 158142c2 | bellard | |
821 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
822 | 158142c2 | bellard | | Takes an abstract floating-point value having sign `zSign', exponent
|
823 | 158142c2 | bellard | | `zExp', and significand formed by the concatenation of `zSig0' and `zSig1',
|
824 | 158142c2 | bellard | | and returns the proper extended double-precision floating-point value
|
825 | 158142c2 | bellard | | corresponding to the abstract input. This routine is just like
|
826 | 158142c2 | bellard | | `roundAndPackFloatx80' except that the input significand does not have to be
|
827 | 158142c2 | bellard | | normalized.
|
828 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
829 | 158142c2 | bellard | |
830 | 158142c2 | bellard | static floatx80
|
831 | 158142c2 | bellard | normalizeRoundAndPackFloatx80( |
832 | bb98fe42 | Andreas Färber | int8 roundingPrecision, flag zSign, int32 zExp, uint64_t zSig0, uint64_t zSig1 |
833 | 158142c2 | bellard | STATUS_PARAM) |
834 | 158142c2 | bellard | { |
835 | 158142c2 | bellard | int8 shiftCount; |
836 | 158142c2 | bellard | |
837 | 158142c2 | bellard | if ( zSig0 == 0 ) { |
838 | 158142c2 | bellard | zSig0 = zSig1; |
839 | 158142c2 | bellard | zSig1 = 0;
|
840 | 158142c2 | bellard | zExp -= 64;
|
841 | 158142c2 | bellard | } |
842 | 158142c2 | bellard | shiftCount = countLeadingZeros64( zSig0 ); |
843 | 158142c2 | bellard | shortShift128Left( zSig0, zSig1, shiftCount, &zSig0, &zSig1 ); |
844 | 158142c2 | bellard | zExp -= shiftCount; |
845 | 158142c2 | bellard | return
|
846 | 158142c2 | bellard | roundAndPackFloatx80( roundingPrecision, zSign, zExp, zSig0, zSig1 STATUS_VAR); |
847 | 158142c2 | bellard | |
848 | 158142c2 | bellard | } |
849 | 158142c2 | bellard | |
850 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
851 | 158142c2 | bellard | | Returns the least-significant 64 fraction bits of the quadruple-precision
|
852 | 158142c2 | bellard | | floating-point value `a'.
|
853 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
854 | 158142c2 | bellard | |
855 | bb98fe42 | Andreas Färber | INLINE uint64_t extractFloat128Frac1( float128 a ) |
856 | 158142c2 | bellard | { |
857 | 158142c2 | bellard | |
858 | 158142c2 | bellard | return a.low;
|
859 | 158142c2 | bellard | |
860 | 158142c2 | bellard | } |
861 | 158142c2 | bellard | |
862 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
863 | 158142c2 | bellard | | Returns the most-significant 48 fraction bits of the quadruple-precision
|
864 | 158142c2 | bellard | | floating-point value `a'.
|
865 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
866 | 158142c2 | bellard | |
867 | bb98fe42 | Andreas Färber | INLINE uint64_t extractFloat128Frac0( float128 a ) |
868 | 158142c2 | bellard | { |
869 | 158142c2 | bellard | |
870 | 158142c2 | bellard | return a.high & LIT64( 0x0000FFFFFFFFFFFF ); |
871 | 158142c2 | bellard | |
872 | 158142c2 | bellard | } |
873 | 158142c2 | bellard | |
874 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
875 | 158142c2 | bellard | | Returns the exponent bits of the quadruple-precision floating-point value
|
876 | 158142c2 | bellard | | `a'.
|
877 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
878 | 158142c2 | bellard | |
879 | 158142c2 | bellard | INLINE int32 extractFloat128Exp( float128 a ) |
880 | 158142c2 | bellard | { |
881 | 158142c2 | bellard | |
882 | 158142c2 | bellard | return ( a.high>>48 ) & 0x7FFF; |
883 | 158142c2 | bellard | |
884 | 158142c2 | bellard | } |
885 | 158142c2 | bellard | |
886 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
887 | 158142c2 | bellard | | Returns the sign bit of the quadruple-precision floating-point value `a'.
|
888 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
889 | 158142c2 | bellard | |
890 | 158142c2 | bellard | INLINE flag extractFloat128Sign( float128 a ) |
891 | 158142c2 | bellard | { |
892 | 158142c2 | bellard | |
893 | 158142c2 | bellard | return a.high>>63; |
894 | 158142c2 | bellard | |
895 | 158142c2 | bellard | } |
896 | 158142c2 | bellard | |
897 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
898 | 158142c2 | bellard | | Normalizes the subnormal quadruple-precision floating-point value
|
899 | 158142c2 | bellard | | represented by the denormalized significand formed by the concatenation of
|
900 | 158142c2 | bellard | | `aSig0' and `aSig1'. The normalized exponent is stored at the location
|
901 | 158142c2 | bellard | | pointed to by `zExpPtr'. The most significant 49 bits of the normalized
|
902 | 158142c2 | bellard | | significand are stored at the location pointed to by `zSig0Ptr', and the
|
903 | 158142c2 | bellard | | least significant 64 bits of the normalized significand are stored at the
|
904 | 158142c2 | bellard | | location pointed to by `zSig1Ptr'.
|
905 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
906 | 158142c2 | bellard | |
907 | 158142c2 | bellard | static void |
908 | 158142c2 | bellard | normalizeFloat128Subnormal( |
909 | bb98fe42 | Andreas Färber | uint64_t aSig0, |
910 | bb98fe42 | Andreas Färber | uint64_t aSig1, |
911 | 158142c2 | bellard | int32 *zExpPtr, |
912 | bb98fe42 | Andreas Färber | uint64_t *zSig0Ptr, |
913 | bb98fe42 | Andreas Färber | uint64_t *zSig1Ptr |
914 | 158142c2 | bellard | ) |
915 | 158142c2 | bellard | { |
916 | 158142c2 | bellard | int8 shiftCount; |
917 | 158142c2 | bellard | |
918 | 158142c2 | bellard | if ( aSig0 == 0 ) { |
919 | 158142c2 | bellard | shiftCount = countLeadingZeros64( aSig1 ) - 15;
|
920 | 158142c2 | bellard | if ( shiftCount < 0 ) { |
921 | 158142c2 | bellard | *zSig0Ptr = aSig1>>( - shiftCount ); |
922 | 158142c2 | bellard | *zSig1Ptr = aSig1<<( shiftCount & 63 );
|
923 | 158142c2 | bellard | } |
924 | 158142c2 | bellard | else {
|
925 | 158142c2 | bellard | *zSig0Ptr = aSig1<<shiftCount; |
926 | 158142c2 | bellard | *zSig1Ptr = 0;
|
927 | 158142c2 | bellard | } |
928 | 158142c2 | bellard | *zExpPtr = - shiftCount - 63;
|
929 | 158142c2 | bellard | } |
930 | 158142c2 | bellard | else {
|
931 | 158142c2 | bellard | shiftCount = countLeadingZeros64( aSig0 ) - 15;
|
932 | 158142c2 | bellard | shortShift128Left( aSig0, aSig1, shiftCount, zSig0Ptr, zSig1Ptr ); |
933 | 158142c2 | bellard | *zExpPtr = 1 - shiftCount;
|
934 | 158142c2 | bellard | } |
935 | 158142c2 | bellard | |
936 | 158142c2 | bellard | } |
937 | 158142c2 | bellard | |
938 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
939 | 158142c2 | bellard | | Packs the sign `zSign', the exponent `zExp', and the significand formed
|
940 | 158142c2 | bellard | | by the concatenation of `zSig0' and `zSig1' into a quadruple-precision
|
941 | 158142c2 | bellard | | floating-point value, returning the result. After being shifted into the
|
942 | 158142c2 | bellard | | proper positions, the three fields `zSign', `zExp', and `zSig0' are simply
|
943 | 158142c2 | bellard | | added together to form the most significant 32 bits of the result. This
|
944 | 158142c2 | bellard | | means that any integer portion of `zSig0' will be added into the exponent.
|
945 | 158142c2 | bellard | | Since a properly normalized significand will have an integer portion equal
|
946 | 158142c2 | bellard | | to 1, the `zExp' input should be 1 less than the desired result exponent
|
947 | 158142c2 | bellard | | whenever `zSig0' and `zSig1' concatenated form a complete, normalized
|
948 | 158142c2 | bellard | | significand.
|
949 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
950 | 158142c2 | bellard | |
951 | 158142c2 | bellard | INLINE float128 |
952 | bb98fe42 | Andreas Färber | packFloat128( flag zSign, int32 zExp, uint64_t zSig0, uint64_t zSig1 ) |
953 | 158142c2 | bellard | { |
954 | 158142c2 | bellard | float128 z; |
955 | 158142c2 | bellard | |
956 | 158142c2 | bellard | z.low = zSig1; |
957 | bb98fe42 | Andreas Färber | z.high = ( ( (uint64_t) zSign )<<63 ) + ( ( (uint64_t) zExp )<<48 ) + zSig0; |
958 | 158142c2 | bellard | return z;
|
959 | 158142c2 | bellard | |
960 | 158142c2 | bellard | } |
961 | 158142c2 | bellard | |
962 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
963 | 158142c2 | bellard | | Takes an abstract floating-point value having sign `zSign', exponent `zExp',
|
964 | 158142c2 | bellard | | and extended significand formed by the concatenation of `zSig0', `zSig1',
|
965 | 158142c2 | bellard | | and `zSig2', and returns the proper quadruple-precision floating-point value
|
966 | 158142c2 | bellard | | corresponding to the abstract input. Ordinarily, the abstract value is
|
967 | 158142c2 | bellard | | simply rounded and packed into the quadruple-precision format, with the
|
968 | 158142c2 | bellard | | inexact exception raised if the abstract input cannot be represented
|
969 | 158142c2 | bellard | | exactly. However, if the abstract value is too large, the overflow and
|
970 | 158142c2 | bellard | | inexact exceptions are raised and an infinity or maximal finite value is
|
971 | 158142c2 | bellard | | returned. If the abstract value is too small, the input value is rounded to
|
972 | 158142c2 | bellard | | a subnormal number, and the underflow and inexact exceptions are raised if
|
973 | 158142c2 | bellard | | the abstract input cannot be represented exactly as a subnormal quadruple-
|
974 | 158142c2 | bellard | | precision floating-point number.
|
975 | 158142c2 | bellard | | The input significand must be normalized or smaller. If the input
|
976 | 158142c2 | bellard | | significand is not normalized, `zExp' must be 0; in that case, the result
|
977 | 158142c2 | bellard | | returned is a subnormal number, and it must not require rounding. In the
|
978 | 158142c2 | bellard | | usual case that the input significand is normalized, `zExp' must be 1 less
|
979 | 158142c2 | bellard | | than the ``true'' floating-point exponent. The handling of underflow and
|
980 | 158142c2 | bellard | | overflow follows the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
981 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
982 | 158142c2 | bellard | |
983 | 158142c2 | bellard | static float128
|
984 | 158142c2 | bellard | roundAndPackFloat128( |
985 | bb98fe42 | Andreas Färber | flag zSign, int32 zExp, uint64_t zSig0, uint64_t zSig1, uint64_t zSig2 STATUS_PARAM) |
986 | 158142c2 | bellard | { |
987 | 158142c2 | bellard | int8 roundingMode; |
988 | 158142c2 | bellard | flag roundNearestEven, increment, isTiny; |
989 | 158142c2 | bellard | |
990 | 158142c2 | bellard | roundingMode = STATUS(float_rounding_mode); |
991 | 158142c2 | bellard | roundNearestEven = ( roundingMode == float_round_nearest_even ); |
992 | bb98fe42 | Andreas Färber | increment = ( (int64_t) zSig2 < 0 );
|
993 | 158142c2 | bellard | if ( ! roundNearestEven ) {
|
994 | 158142c2 | bellard | if ( roundingMode == float_round_to_zero ) {
|
995 | 158142c2 | bellard | increment = 0;
|
996 | 158142c2 | bellard | } |
997 | 158142c2 | bellard | else {
|
998 | 158142c2 | bellard | if ( zSign ) {
|
999 | 158142c2 | bellard | increment = ( roundingMode == float_round_down ) && zSig2; |
1000 | 158142c2 | bellard | } |
1001 | 158142c2 | bellard | else {
|
1002 | 158142c2 | bellard | increment = ( roundingMode == float_round_up ) && zSig2; |
1003 | 158142c2 | bellard | } |
1004 | 158142c2 | bellard | } |
1005 | 158142c2 | bellard | } |
1006 | bb98fe42 | Andreas Färber | if ( 0x7FFD <= (uint32_t) zExp ) { |
1007 | 158142c2 | bellard | if ( ( 0x7FFD < zExp ) |
1008 | 158142c2 | bellard | || ( ( zExp == 0x7FFD )
|
1009 | 158142c2 | bellard | && eq128( |
1010 | 158142c2 | bellard | LIT64( 0x0001FFFFFFFFFFFF ),
|
1011 | 158142c2 | bellard | LIT64( 0xFFFFFFFFFFFFFFFF ),
|
1012 | 158142c2 | bellard | zSig0, |
1013 | 158142c2 | bellard | zSig1 |
1014 | 158142c2 | bellard | ) |
1015 | 158142c2 | bellard | && increment |
1016 | 158142c2 | bellard | ) |
1017 | 158142c2 | bellard | ) { |
1018 | 158142c2 | bellard | float_raise( float_flag_overflow | float_flag_inexact STATUS_VAR); |
1019 | 158142c2 | bellard | if ( ( roundingMode == float_round_to_zero )
|
1020 | 158142c2 | bellard | || ( zSign && ( roundingMode == float_round_up ) ) |
1021 | 158142c2 | bellard | || ( ! zSign && ( roundingMode == float_round_down ) ) |
1022 | 158142c2 | bellard | ) { |
1023 | 158142c2 | bellard | return
|
1024 | 158142c2 | bellard | packFloat128( |
1025 | 158142c2 | bellard | zSign, |
1026 | 158142c2 | bellard | 0x7FFE,
|
1027 | 158142c2 | bellard | LIT64( 0x0000FFFFFFFFFFFF ),
|
1028 | 158142c2 | bellard | LIT64( 0xFFFFFFFFFFFFFFFF )
|
1029 | 158142c2 | bellard | ); |
1030 | 158142c2 | bellard | } |
1031 | 158142c2 | bellard | return packFloat128( zSign, 0x7FFF, 0, 0 ); |
1032 | 158142c2 | bellard | } |
1033 | 158142c2 | bellard | if ( zExp < 0 ) { |
1034 | e6afc87f | Peter Maydell | if (STATUS(flush_to_zero)) {
|
1035 | e6afc87f | Peter Maydell | float_raise(float_flag_output_denormal STATUS_VAR); |
1036 | e6afc87f | Peter Maydell | return packFloat128(zSign, 0, 0, 0); |
1037 | e6afc87f | Peter Maydell | } |
1038 | 158142c2 | bellard | isTiny = |
1039 | 158142c2 | bellard | ( STATUS(float_detect_tininess) == float_tininess_before_rounding ) |
1040 | 158142c2 | bellard | || ( zExp < -1 )
|
1041 | 158142c2 | bellard | || ! increment |
1042 | 158142c2 | bellard | || lt128( |
1043 | 158142c2 | bellard | zSig0, |
1044 | 158142c2 | bellard | zSig1, |
1045 | 158142c2 | bellard | LIT64( 0x0001FFFFFFFFFFFF ),
|
1046 | 158142c2 | bellard | LIT64( 0xFFFFFFFFFFFFFFFF )
|
1047 | 158142c2 | bellard | ); |
1048 | 158142c2 | bellard | shift128ExtraRightJamming( |
1049 | 158142c2 | bellard | zSig0, zSig1, zSig2, - zExp, &zSig0, &zSig1, &zSig2 ); |
1050 | 158142c2 | bellard | zExp = 0;
|
1051 | 158142c2 | bellard | if ( isTiny && zSig2 ) float_raise( float_flag_underflow STATUS_VAR);
|
1052 | 158142c2 | bellard | if ( roundNearestEven ) {
|
1053 | bb98fe42 | Andreas Färber | increment = ( (int64_t) zSig2 < 0 );
|
1054 | 158142c2 | bellard | } |
1055 | 158142c2 | bellard | else {
|
1056 | 158142c2 | bellard | if ( zSign ) {
|
1057 | 158142c2 | bellard | increment = ( roundingMode == float_round_down ) && zSig2; |
1058 | 158142c2 | bellard | } |
1059 | 158142c2 | bellard | else {
|
1060 | 158142c2 | bellard | increment = ( roundingMode == float_round_up ) && zSig2; |
1061 | 158142c2 | bellard | } |
1062 | 158142c2 | bellard | } |
1063 | 158142c2 | bellard | } |
1064 | 158142c2 | bellard | } |
1065 | 158142c2 | bellard | if ( zSig2 ) STATUS(float_exception_flags) |= float_flag_inexact;
|
1066 | 158142c2 | bellard | if ( increment ) {
|
1067 | 158142c2 | bellard | add128( zSig0, zSig1, 0, 1, &zSig0, &zSig1 ); |
1068 | 158142c2 | bellard | zSig1 &= ~ ( ( zSig2 + zSig2 == 0 ) & roundNearestEven );
|
1069 | 158142c2 | bellard | } |
1070 | 158142c2 | bellard | else {
|
1071 | 158142c2 | bellard | if ( ( zSig0 | zSig1 ) == 0 ) zExp = 0; |
1072 | 158142c2 | bellard | } |
1073 | 158142c2 | bellard | return packFloat128( zSign, zExp, zSig0, zSig1 );
|
1074 | 158142c2 | bellard | |
1075 | 158142c2 | bellard | } |
1076 | 158142c2 | bellard | |
1077 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1078 | 158142c2 | bellard | | Takes an abstract floating-point value having sign `zSign', exponent `zExp',
|
1079 | 158142c2 | bellard | | and significand formed by the concatenation of `zSig0' and `zSig1', and
|
1080 | 158142c2 | bellard | | returns the proper quadruple-precision floating-point value corresponding
|
1081 | 158142c2 | bellard | | to the abstract input. This routine is just like `roundAndPackFloat128'
|
1082 | 158142c2 | bellard | | except that the input significand has fewer bits and does not have to be
|
1083 | 158142c2 | bellard | | normalized. In all cases, `zExp' must be 1 less than the ``true'' floating-
|
1084 | 158142c2 | bellard | | point exponent.
|
1085 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1086 | 158142c2 | bellard | |
1087 | 158142c2 | bellard | static float128
|
1088 | 158142c2 | bellard | normalizeRoundAndPackFloat128( |
1089 | bb98fe42 | Andreas Färber | flag zSign, int32 zExp, uint64_t zSig0, uint64_t zSig1 STATUS_PARAM) |
1090 | 158142c2 | bellard | { |
1091 | 158142c2 | bellard | int8 shiftCount; |
1092 | bb98fe42 | Andreas Färber | uint64_t zSig2; |
1093 | 158142c2 | bellard | |
1094 | 158142c2 | bellard | if ( zSig0 == 0 ) { |
1095 | 158142c2 | bellard | zSig0 = zSig1; |
1096 | 158142c2 | bellard | zSig1 = 0;
|
1097 | 158142c2 | bellard | zExp -= 64;
|
1098 | 158142c2 | bellard | } |
1099 | 158142c2 | bellard | shiftCount = countLeadingZeros64( zSig0 ) - 15;
|
1100 | 158142c2 | bellard | if ( 0 <= shiftCount ) { |
1101 | 158142c2 | bellard | zSig2 = 0;
|
1102 | 158142c2 | bellard | shortShift128Left( zSig0, zSig1, shiftCount, &zSig0, &zSig1 ); |
1103 | 158142c2 | bellard | } |
1104 | 158142c2 | bellard | else {
|
1105 | 158142c2 | bellard | shift128ExtraRightJamming( |
1106 | 158142c2 | bellard | zSig0, zSig1, 0, - shiftCount, &zSig0, &zSig1, &zSig2 );
|
1107 | 158142c2 | bellard | } |
1108 | 158142c2 | bellard | zExp -= shiftCount; |
1109 | 158142c2 | bellard | return roundAndPackFloat128( zSign, zExp, zSig0, zSig1, zSig2 STATUS_VAR);
|
1110 | 158142c2 | bellard | |
1111 | 158142c2 | bellard | } |
1112 | 158142c2 | bellard | |
1113 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1114 | 158142c2 | bellard | | Returns the result of converting the 32-bit two's complement integer `a'
|
1115 | 158142c2 | bellard | | to the single-precision floating-point format. The conversion is performed
|
1116 | 158142c2 | bellard | | according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
1117 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1118 | 158142c2 | bellard | |
1119 | 158142c2 | bellard | float32 int32_to_float32( int32 a STATUS_PARAM ) |
1120 | 158142c2 | bellard | { |
1121 | 158142c2 | bellard | flag zSign; |
1122 | 158142c2 | bellard | |
1123 | f090c9d4 | pbrook | if ( a == 0 ) return float32_zero; |
1124 | bb98fe42 | Andreas Färber | if ( a == (int32_t) 0x80000000 ) return packFloat32( 1, 0x9E, 0 ); |
1125 | 158142c2 | bellard | zSign = ( a < 0 );
|
1126 | 158142c2 | bellard | return normalizeRoundAndPackFloat32( zSign, 0x9C, zSign ? - a : a STATUS_VAR ); |
1127 | 158142c2 | bellard | |
1128 | 158142c2 | bellard | } |
1129 | 158142c2 | bellard | |
1130 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1131 | 158142c2 | bellard | | Returns the result of converting the 32-bit two's complement integer `a'
|
1132 | 158142c2 | bellard | | to the double-precision floating-point format. The conversion is performed
|
1133 | 158142c2 | bellard | | according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
1134 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1135 | 158142c2 | bellard | |
1136 | 158142c2 | bellard | float64 int32_to_float64( int32 a STATUS_PARAM ) |
1137 | 158142c2 | bellard | { |
1138 | 158142c2 | bellard | flag zSign; |
1139 | 158142c2 | bellard | uint32 absA; |
1140 | 158142c2 | bellard | int8 shiftCount; |
1141 | bb98fe42 | Andreas Färber | uint64_t zSig; |
1142 | 158142c2 | bellard | |
1143 | f090c9d4 | pbrook | if ( a == 0 ) return float64_zero; |
1144 | 158142c2 | bellard | zSign = ( a < 0 );
|
1145 | 158142c2 | bellard | absA = zSign ? - a : a; |
1146 | 158142c2 | bellard | shiftCount = countLeadingZeros32( absA ) + 21;
|
1147 | 158142c2 | bellard | zSig = absA; |
1148 | 158142c2 | bellard | return packFloat64( zSign, 0x432 - shiftCount, zSig<<shiftCount ); |
1149 | 158142c2 | bellard | |
1150 | 158142c2 | bellard | } |
1151 | 158142c2 | bellard | |
1152 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1153 | 158142c2 | bellard | | Returns the result of converting the 32-bit two's complement integer `a'
|
1154 | 158142c2 | bellard | | to the extended double-precision floating-point format. The conversion
|
1155 | 158142c2 | bellard | | is performed according to the IEC/IEEE Standard for Binary Floating-Point
|
1156 | 158142c2 | bellard | | Arithmetic.
|
1157 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1158 | 158142c2 | bellard | |
1159 | 158142c2 | bellard | floatx80 int32_to_floatx80( int32 a STATUS_PARAM ) |
1160 | 158142c2 | bellard | { |
1161 | 158142c2 | bellard | flag zSign; |
1162 | 158142c2 | bellard | uint32 absA; |
1163 | 158142c2 | bellard | int8 shiftCount; |
1164 | bb98fe42 | Andreas Färber | uint64_t zSig; |
1165 | 158142c2 | bellard | |
1166 | 158142c2 | bellard | if ( a == 0 ) return packFloatx80( 0, 0, 0 ); |
1167 | 158142c2 | bellard | zSign = ( a < 0 );
|
1168 | 158142c2 | bellard | absA = zSign ? - a : a; |
1169 | 158142c2 | bellard | shiftCount = countLeadingZeros32( absA ) + 32;
|
1170 | 158142c2 | bellard | zSig = absA; |
1171 | 158142c2 | bellard | return packFloatx80( zSign, 0x403E - shiftCount, zSig<<shiftCount ); |
1172 | 158142c2 | bellard | |
1173 | 158142c2 | bellard | } |
1174 | 158142c2 | bellard | |
1175 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1176 | 158142c2 | bellard | | Returns the result of converting the 32-bit two's complement integer `a' to
|
1177 | 158142c2 | bellard | | the quadruple-precision floating-point format. The conversion is performed
|
1178 | 158142c2 | bellard | | according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
1179 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1180 | 158142c2 | bellard | |
1181 | 158142c2 | bellard | float128 int32_to_float128( int32 a STATUS_PARAM ) |
1182 | 158142c2 | bellard | { |
1183 | 158142c2 | bellard | flag zSign; |
1184 | 158142c2 | bellard | uint32 absA; |
1185 | 158142c2 | bellard | int8 shiftCount; |
1186 | bb98fe42 | Andreas Färber | uint64_t zSig0; |
1187 | 158142c2 | bellard | |
1188 | 158142c2 | bellard | if ( a == 0 ) return packFloat128( 0, 0, 0, 0 ); |
1189 | 158142c2 | bellard | zSign = ( a < 0 );
|
1190 | 158142c2 | bellard | absA = zSign ? - a : a; |
1191 | 158142c2 | bellard | shiftCount = countLeadingZeros32( absA ) + 17;
|
1192 | 158142c2 | bellard | zSig0 = absA; |
1193 | 158142c2 | bellard | return packFloat128( zSign, 0x402E - shiftCount, zSig0<<shiftCount, 0 ); |
1194 | 158142c2 | bellard | |
1195 | 158142c2 | bellard | } |
1196 | 158142c2 | bellard | |
1197 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1198 | 158142c2 | bellard | | Returns the result of converting the 64-bit two's complement integer `a'
|
1199 | 158142c2 | bellard | | to the single-precision floating-point format. The conversion is performed
|
1200 | 158142c2 | bellard | | according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
1201 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1202 | 158142c2 | bellard | |
1203 | 158142c2 | bellard | float32 int64_to_float32( int64 a STATUS_PARAM ) |
1204 | 158142c2 | bellard | { |
1205 | 158142c2 | bellard | flag zSign; |
1206 | 158142c2 | bellard | uint64 absA; |
1207 | 158142c2 | bellard | int8 shiftCount; |
1208 | 158142c2 | bellard | |
1209 | f090c9d4 | pbrook | if ( a == 0 ) return float32_zero; |
1210 | 158142c2 | bellard | zSign = ( a < 0 );
|
1211 | 158142c2 | bellard | absA = zSign ? - a : a; |
1212 | 158142c2 | bellard | shiftCount = countLeadingZeros64( absA ) - 40;
|
1213 | 158142c2 | bellard | if ( 0 <= shiftCount ) { |
1214 | 158142c2 | bellard | return packFloat32( zSign, 0x95 - shiftCount, absA<<shiftCount ); |
1215 | 158142c2 | bellard | } |
1216 | 158142c2 | bellard | else {
|
1217 | 158142c2 | bellard | shiftCount += 7;
|
1218 | 158142c2 | bellard | if ( shiftCount < 0 ) { |
1219 | 158142c2 | bellard | shift64RightJamming( absA, - shiftCount, &absA ); |
1220 | 158142c2 | bellard | } |
1221 | 158142c2 | bellard | else {
|
1222 | 158142c2 | bellard | absA <<= shiftCount; |
1223 | 158142c2 | bellard | } |
1224 | 158142c2 | bellard | return roundAndPackFloat32( zSign, 0x9C - shiftCount, absA STATUS_VAR ); |
1225 | 158142c2 | bellard | } |
1226 | 158142c2 | bellard | |
1227 | 158142c2 | bellard | } |
1228 | 158142c2 | bellard | |
1229 | 3430b0be | j_mayer | float32 uint64_to_float32( uint64 a STATUS_PARAM ) |
1230 | 75d62a58 | j_mayer | { |
1231 | 75d62a58 | j_mayer | int8 shiftCount; |
1232 | 75d62a58 | j_mayer | |
1233 | f090c9d4 | pbrook | if ( a == 0 ) return float32_zero; |
1234 | 75d62a58 | j_mayer | shiftCount = countLeadingZeros64( a ) - 40;
|
1235 | 75d62a58 | j_mayer | if ( 0 <= shiftCount ) { |
1236 | 75d62a58 | j_mayer | return packFloat32( 1 > 0, 0x95 - shiftCount, a<<shiftCount ); |
1237 | 75d62a58 | j_mayer | } |
1238 | 75d62a58 | j_mayer | else {
|
1239 | 75d62a58 | j_mayer | shiftCount += 7;
|
1240 | 75d62a58 | j_mayer | if ( shiftCount < 0 ) { |
1241 | 75d62a58 | j_mayer | shift64RightJamming( a, - shiftCount, &a ); |
1242 | 75d62a58 | j_mayer | } |
1243 | 75d62a58 | j_mayer | else {
|
1244 | 75d62a58 | j_mayer | a <<= shiftCount; |
1245 | 75d62a58 | j_mayer | } |
1246 | 75d62a58 | j_mayer | return roundAndPackFloat32( 1 > 0, 0x9C - shiftCount, a STATUS_VAR ); |
1247 | 75d62a58 | j_mayer | } |
1248 | 75d62a58 | j_mayer | } |
1249 | 75d62a58 | j_mayer | |
1250 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1251 | 158142c2 | bellard | | Returns the result of converting the 64-bit two's complement integer `a'
|
1252 | 158142c2 | bellard | | to the double-precision floating-point format. The conversion is performed
|
1253 | 158142c2 | bellard | | according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
1254 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1255 | 158142c2 | bellard | |
1256 | 158142c2 | bellard | float64 int64_to_float64( int64 a STATUS_PARAM ) |
1257 | 158142c2 | bellard | { |
1258 | 158142c2 | bellard | flag zSign; |
1259 | 158142c2 | bellard | |
1260 | f090c9d4 | pbrook | if ( a == 0 ) return float64_zero; |
1261 | bb98fe42 | Andreas Färber | if ( a == (int64_t) LIT64( 0x8000000000000000 ) ) { |
1262 | 158142c2 | bellard | return packFloat64( 1, 0x43E, 0 ); |
1263 | 158142c2 | bellard | } |
1264 | 158142c2 | bellard | zSign = ( a < 0 );
|
1265 | 158142c2 | bellard | return normalizeRoundAndPackFloat64( zSign, 0x43C, zSign ? - a : a STATUS_VAR ); |
1266 | 158142c2 | bellard | |
1267 | 158142c2 | bellard | } |
1268 | 158142c2 | bellard | |
1269 | 75d62a58 | j_mayer | float64 uint64_to_float64( uint64 a STATUS_PARAM ) |
1270 | 75d62a58 | j_mayer | { |
1271 | f090c9d4 | pbrook | if ( a == 0 ) return float64_zero; |
1272 | 75d62a58 | j_mayer | return normalizeRoundAndPackFloat64( 0, 0x43C, a STATUS_VAR ); |
1273 | 75d62a58 | j_mayer | |
1274 | 75d62a58 | j_mayer | } |
1275 | 75d62a58 | j_mayer | |
1276 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1277 | 158142c2 | bellard | | Returns the result of converting the 64-bit two's complement integer `a'
|
1278 | 158142c2 | bellard | | to the extended double-precision floating-point format. The conversion
|
1279 | 158142c2 | bellard | | is performed according to the IEC/IEEE Standard for Binary Floating-Point
|
1280 | 158142c2 | bellard | | Arithmetic.
|
1281 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1282 | 158142c2 | bellard | |
1283 | 158142c2 | bellard | floatx80 int64_to_floatx80( int64 a STATUS_PARAM ) |
1284 | 158142c2 | bellard | { |
1285 | 158142c2 | bellard | flag zSign; |
1286 | 158142c2 | bellard | uint64 absA; |
1287 | 158142c2 | bellard | int8 shiftCount; |
1288 | 158142c2 | bellard | |
1289 | 158142c2 | bellard | if ( a == 0 ) return packFloatx80( 0, 0, 0 ); |
1290 | 158142c2 | bellard | zSign = ( a < 0 );
|
1291 | 158142c2 | bellard | absA = zSign ? - a : a; |
1292 | 158142c2 | bellard | shiftCount = countLeadingZeros64( absA ); |
1293 | 158142c2 | bellard | return packFloatx80( zSign, 0x403E - shiftCount, absA<<shiftCount ); |
1294 | 158142c2 | bellard | |
1295 | 158142c2 | bellard | } |
1296 | 158142c2 | bellard | |
1297 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1298 | 158142c2 | bellard | | Returns the result of converting the 64-bit two's complement integer `a' to
|
1299 | 158142c2 | bellard | | the quadruple-precision floating-point format. The conversion is performed
|
1300 | 158142c2 | bellard | | according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
1301 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1302 | 158142c2 | bellard | |
1303 | 158142c2 | bellard | float128 int64_to_float128( int64 a STATUS_PARAM ) |
1304 | 158142c2 | bellard | { |
1305 | 158142c2 | bellard | flag zSign; |
1306 | 158142c2 | bellard | uint64 absA; |
1307 | 158142c2 | bellard | int8 shiftCount; |
1308 | 158142c2 | bellard | int32 zExp; |
1309 | bb98fe42 | Andreas Färber | uint64_t zSig0, zSig1; |
1310 | 158142c2 | bellard | |
1311 | 158142c2 | bellard | if ( a == 0 ) return packFloat128( 0, 0, 0, 0 ); |
1312 | 158142c2 | bellard | zSign = ( a < 0 );
|
1313 | 158142c2 | bellard | absA = zSign ? - a : a; |
1314 | 158142c2 | bellard | shiftCount = countLeadingZeros64( absA ) + 49;
|
1315 | 158142c2 | bellard | zExp = 0x406E - shiftCount;
|
1316 | 158142c2 | bellard | if ( 64 <= shiftCount ) { |
1317 | 158142c2 | bellard | zSig1 = 0;
|
1318 | 158142c2 | bellard | zSig0 = absA; |
1319 | 158142c2 | bellard | shiftCount -= 64;
|
1320 | 158142c2 | bellard | } |
1321 | 158142c2 | bellard | else {
|
1322 | 158142c2 | bellard | zSig1 = absA; |
1323 | 158142c2 | bellard | zSig0 = 0;
|
1324 | 158142c2 | bellard | } |
1325 | 158142c2 | bellard | shortShift128Left( zSig0, zSig1, shiftCount, &zSig0, &zSig1 ); |
1326 | 158142c2 | bellard | return packFloat128( zSign, zExp, zSig0, zSig1 );
|
1327 | 158142c2 | bellard | |
1328 | 158142c2 | bellard | } |
1329 | 158142c2 | bellard | |
1330 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1331 | 158142c2 | bellard | | Returns the result of converting the single-precision floating-point value
|
1332 | 158142c2 | bellard | | `a' to the 32-bit two's complement integer format. The conversion is
|
1333 | 158142c2 | bellard | | performed according to the IEC/IEEE Standard for Binary Floating-Point
|
1334 | 158142c2 | bellard | | Arithmetic---which means in particular that the conversion is rounded
|
1335 | 158142c2 | bellard | | according to the current rounding mode. If `a' is a NaN, the largest
|
1336 | 158142c2 | bellard | | positive integer is returned. Otherwise, if the conversion overflows, the
|
1337 | 158142c2 | bellard | | largest integer with the same sign as `a' is returned.
|
1338 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1339 | 158142c2 | bellard | |
1340 | 158142c2 | bellard | int32 float32_to_int32( float32 a STATUS_PARAM ) |
1341 | 158142c2 | bellard | { |
1342 | 158142c2 | bellard | flag aSign; |
1343 | 158142c2 | bellard | int16 aExp, shiftCount; |
1344 | bb98fe42 | Andreas Färber | uint32_t aSig; |
1345 | bb98fe42 | Andreas Färber | uint64_t aSig64; |
1346 | 158142c2 | bellard | |
1347 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
1348 | 158142c2 | bellard | aSig = extractFloat32Frac( a ); |
1349 | 158142c2 | bellard | aExp = extractFloat32Exp( a ); |
1350 | 158142c2 | bellard | aSign = extractFloat32Sign( a ); |
1351 | 158142c2 | bellard | if ( ( aExp == 0xFF ) && aSig ) aSign = 0; |
1352 | 158142c2 | bellard | if ( aExp ) aSig |= 0x00800000; |
1353 | 158142c2 | bellard | shiftCount = 0xAF - aExp;
|
1354 | 158142c2 | bellard | aSig64 = aSig; |
1355 | 158142c2 | bellard | aSig64 <<= 32;
|
1356 | 158142c2 | bellard | if ( 0 < shiftCount ) shift64RightJamming( aSig64, shiftCount, &aSig64 ); |
1357 | 158142c2 | bellard | return roundAndPackInt32( aSign, aSig64 STATUS_VAR );
|
1358 | 158142c2 | bellard | |
1359 | 158142c2 | bellard | } |
1360 | 158142c2 | bellard | |
1361 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1362 | 158142c2 | bellard | | Returns the result of converting the single-precision floating-point value
|
1363 | 158142c2 | bellard | | `a' to the 32-bit two's complement integer format. The conversion is
|
1364 | 158142c2 | bellard | | performed according to the IEC/IEEE Standard for Binary Floating-Point
|
1365 | 158142c2 | bellard | | Arithmetic, except that the conversion is always rounded toward zero.
|
1366 | 158142c2 | bellard | | If `a' is a NaN, the largest positive integer is returned. Otherwise, if
|
1367 | 158142c2 | bellard | | the conversion overflows, the largest integer with the same sign as `a' is
|
1368 | 158142c2 | bellard | | returned.
|
1369 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1370 | 158142c2 | bellard | |
1371 | 158142c2 | bellard | int32 float32_to_int32_round_to_zero( float32 a STATUS_PARAM ) |
1372 | 158142c2 | bellard | { |
1373 | 158142c2 | bellard | flag aSign; |
1374 | 158142c2 | bellard | int16 aExp, shiftCount; |
1375 | bb98fe42 | Andreas Färber | uint32_t aSig; |
1376 | 158142c2 | bellard | int32 z; |
1377 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
1378 | 158142c2 | bellard | |
1379 | 158142c2 | bellard | aSig = extractFloat32Frac( a ); |
1380 | 158142c2 | bellard | aExp = extractFloat32Exp( a ); |
1381 | 158142c2 | bellard | aSign = extractFloat32Sign( a ); |
1382 | 158142c2 | bellard | shiftCount = aExp - 0x9E;
|
1383 | 158142c2 | bellard | if ( 0 <= shiftCount ) { |
1384 | f090c9d4 | pbrook | if ( float32_val(a) != 0xCF000000 ) { |
1385 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
1386 | 158142c2 | bellard | if ( ! aSign || ( ( aExp == 0xFF ) && aSig ) ) return 0x7FFFFFFF; |
1387 | 158142c2 | bellard | } |
1388 | bb98fe42 | Andreas Färber | return (int32_t) 0x80000000; |
1389 | 158142c2 | bellard | } |
1390 | 158142c2 | bellard | else if ( aExp <= 0x7E ) { |
1391 | 158142c2 | bellard | if ( aExp | aSig ) STATUS(float_exception_flags) |= float_flag_inexact;
|
1392 | 158142c2 | bellard | return 0; |
1393 | 158142c2 | bellard | } |
1394 | 158142c2 | bellard | aSig = ( aSig | 0x00800000 )<<8; |
1395 | 158142c2 | bellard | z = aSig>>( - shiftCount ); |
1396 | bb98fe42 | Andreas Färber | if ( (uint32_t) ( aSig<<( shiftCount & 31 ) ) ) { |
1397 | 158142c2 | bellard | STATUS(float_exception_flags) |= float_flag_inexact; |
1398 | 158142c2 | bellard | } |
1399 | 158142c2 | bellard | if ( aSign ) z = - z;
|
1400 | 158142c2 | bellard | return z;
|
1401 | 158142c2 | bellard | |
1402 | 158142c2 | bellard | } |
1403 | 158142c2 | bellard | |
1404 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1405 | 158142c2 | bellard | | Returns the result of converting the single-precision floating-point value
|
1406 | cbcef455 | Peter Maydell | | `a' to the 16-bit two's complement integer format. The conversion is
|
1407 | cbcef455 | Peter Maydell | | performed according to the IEC/IEEE Standard for Binary Floating-Point
|
1408 | cbcef455 | Peter Maydell | | Arithmetic, except that the conversion is always rounded toward zero.
|
1409 | cbcef455 | Peter Maydell | | If `a' is a NaN, the largest positive integer is returned. Otherwise, if
|
1410 | cbcef455 | Peter Maydell | | the conversion overflows, the largest integer with the same sign as `a' is
|
1411 | cbcef455 | Peter Maydell | | returned.
|
1412 | cbcef455 | Peter Maydell | *----------------------------------------------------------------------------*/
|
1413 | cbcef455 | Peter Maydell | |
1414 | cbcef455 | Peter Maydell | int16 float32_to_int16_round_to_zero( float32 a STATUS_PARAM ) |
1415 | cbcef455 | Peter Maydell | { |
1416 | cbcef455 | Peter Maydell | flag aSign; |
1417 | cbcef455 | Peter Maydell | int16 aExp, shiftCount; |
1418 | bb98fe42 | Andreas Färber | uint32_t aSig; |
1419 | cbcef455 | Peter Maydell | int32 z; |
1420 | cbcef455 | Peter Maydell | |
1421 | cbcef455 | Peter Maydell | aSig = extractFloat32Frac( a ); |
1422 | cbcef455 | Peter Maydell | aExp = extractFloat32Exp( a ); |
1423 | cbcef455 | Peter Maydell | aSign = extractFloat32Sign( a ); |
1424 | cbcef455 | Peter Maydell | shiftCount = aExp - 0x8E;
|
1425 | cbcef455 | Peter Maydell | if ( 0 <= shiftCount ) { |
1426 | cbcef455 | Peter Maydell | if ( float32_val(a) != 0xC7000000 ) { |
1427 | cbcef455 | Peter Maydell | float_raise( float_flag_invalid STATUS_VAR); |
1428 | cbcef455 | Peter Maydell | if ( ! aSign || ( ( aExp == 0xFF ) && aSig ) ) { |
1429 | cbcef455 | Peter Maydell | return 0x7FFF; |
1430 | cbcef455 | Peter Maydell | } |
1431 | cbcef455 | Peter Maydell | } |
1432 | bb98fe42 | Andreas Färber | return (int32_t) 0xffff8000; |
1433 | cbcef455 | Peter Maydell | } |
1434 | cbcef455 | Peter Maydell | else if ( aExp <= 0x7E ) { |
1435 | cbcef455 | Peter Maydell | if ( aExp | aSig ) {
|
1436 | cbcef455 | Peter Maydell | STATUS(float_exception_flags) |= float_flag_inexact; |
1437 | cbcef455 | Peter Maydell | } |
1438 | cbcef455 | Peter Maydell | return 0; |
1439 | cbcef455 | Peter Maydell | } |
1440 | cbcef455 | Peter Maydell | shiftCount -= 0x10;
|
1441 | cbcef455 | Peter Maydell | aSig = ( aSig | 0x00800000 )<<8; |
1442 | cbcef455 | Peter Maydell | z = aSig>>( - shiftCount ); |
1443 | bb98fe42 | Andreas Färber | if ( (uint32_t) ( aSig<<( shiftCount & 31 ) ) ) { |
1444 | cbcef455 | Peter Maydell | STATUS(float_exception_flags) |= float_flag_inexact; |
1445 | cbcef455 | Peter Maydell | } |
1446 | cbcef455 | Peter Maydell | if ( aSign ) {
|
1447 | cbcef455 | Peter Maydell | z = - z; |
1448 | cbcef455 | Peter Maydell | } |
1449 | cbcef455 | Peter Maydell | return z;
|
1450 | cbcef455 | Peter Maydell | |
1451 | cbcef455 | Peter Maydell | } |
1452 | cbcef455 | Peter Maydell | |
1453 | cbcef455 | Peter Maydell | /*----------------------------------------------------------------------------
|
1454 | cbcef455 | Peter Maydell | | Returns the result of converting the single-precision floating-point value
|
1455 | 158142c2 | bellard | | `a' to the 64-bit two's complement integer format. The conversion is
|
1456 | 158142c2 | bellard | | performed according to the IEC/IEEE Standard for Binary Floating-Point
|
1457 | 158142c2 | bellard | | Arithmetic---which means in particular that the conversion is rounded
|
1458 | 158142c2 | bellard | | according to the current rounding mode. If `a' is a NaN, the largest
|
1459 | 158142c2 | bellard | | positive integer is returned. Otherwise, if the conversion overflows, the
|
1460 | 158142c2 | bellard | | largest integer with the same sign as `a' is returned.
|
1461 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1462 | 158142c2 | bellard | |
1463 | 158142c2 | bellard | int64 float32_to_int64( float32 a STATUS_PARAM ) |
1464 | 158142c2 | bellard | { |
1465 | 158142c2 | bellard | flag aSign; |
1466 | 158142c2 | bellard | int16 aExp, shiftCount; |
1467 | bb98fe42 | Andreas Färber | uint32_t aSig; |
1468 | bb98fe42 | Andreas Färber | uint64_t aSig64, aSigExtra; |
1469 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
1470 | 158142c2 | bellard | |
1471 | 158142c2 | bellard | aSig = extractFloat32Frac( a ); |
1472 | 158142c2 | bellard | aExp = extractFloat32Exp( a ); |
1473 | 158142c2 | bellard | aSign = extractFloat32Sign( a ); |
1474 | 158142c2 | bellard | shiftCount = 0xBE - aExp;
|
1475 | 158142c2 | bellard | if ( shiftCount < 0 ) { |
1476 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
1477 | 158142c2 | bellard | if ( ! aSign || ( ( aExp == 0xFF ) && aSig ) ) { |
1478 | 158142c2 | bellard | return LIT64( 0x7FFFFFFFFFFFFFFF ); |
1479 | 158142c2 | bellard | } |
1480 | bb98fe42 | Andreas Färber | return (int64_t) LIT64( 0x8000000000000000 ); |
1481 | 158142c2 | bellard | } |
1482 | 158142c2 | bellard | if ( aExp ) aSig |= 0x00800000; |
1483 | 158142c2 | bellard | aSig64 = aSig; |
1484 | 158142c2 | bellard | aSig64 <<= 40;
|
1485 | 158142c2 | bellard | shift64ExtraRightJamming( aSig64, 0, shiftCount, &aSig64, &aSigExtra );
|
1486 | 158142c2 | bellard | return roundAndPackInt64( aSign, aSig64, aSigExtra STATUS_VAR );
|
1487 | 158142c2 | bellard | |
1488 | 158142c2 | bellard | } |
1489 | 158142c2 | bellard | |
1490 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1491 | 158142c2 | bellard | | Returns the result of converting the single-precision floating-point value
|
1492 | 158142c2 | bellard | | `a' to the 64-bit two's complement integer format. The conversion is
|
1493 | 158142c2 | bellard | | performed according to the IEC/IEEE Standard for Binary Floating-Point
|
1494 | 158142c2 | bellard | | Arithmetic, except that the conversion is always rounded toward zero. If
|
1495 | 158142c2 | bellard | | `a' is a NaN, the largest positive integer is returned. Otherwise, if the
|
1496 | 158142c2 | bellard | | conversion overflows, the largest integer with the same sign as `a' is
|
1497 | 158142c2 | bellard | | returned.
|
1498 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1499 | 158142c2 | bellard | |
1500 | 158142c2 | bellard | int64 float32_to_int64_round_to_zero( float32 a STATUS_PARAM ) |
1501 | 158142c2 | bellard | { |
1502 | 158142c2 | bellard | flag aSign; |
1503 | 158142c2 | bellard | int16 aExp, shiftCount; |
1504 | bb98fe42 | Andreas Färber | uint32_t aSig; |
1505 | bb98fe42 | Andreas Färber | uint64_t aSig64; |
1506 | 158142c2 | bellard | int64 z; |
1507 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
1508 | 158142c2 | bellard | |
1509 | 158142c2 | bellard | aSig = extractFloat32Frac( a ); |
1510 | 158142c2 | bellard | aExp = extractFloat32Exp( a ); |
1511 | 158142c2 | bellard | aSign = extractFloat32Sign( a ); |
1512 | 158142c2 | bellard | shiftCount = aExp - 0xBE;
|
1513 | 158142c2 | bellard | if ( 0 <= shiftCount ) { |
1514 | f090c9d4 | pbrook | if ( float32_val(a) != 0xDF000000 ) { |
1515 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
1516 | 158142c2 | bellard | if ( ! aSign || ( ( aExp == 0xFF ) && aSig ) ) { |
1517 | 158142c2 | bellard | return LIT64( 0x7FFFFFFFFFFFFFFF ); |
1518 | 158142c2 | bellard | } |
1519 | 158142c2 | bellard | } |
1520 | bb98fe42 | Andreas Färber | return (int64_t) LIT64( 0x8000000000000000 ); |
1521 | 158142c2 | bellard | } |
1522 | 158142c2 | bellard | else if ( aExp <= 0x7E ) { |
1523 | 158142c2 | bellard | if ( aExp | aSig ) STATUS(float_exception_flags) |= float_flag_inexact;
|
1524 | 158142c2 | bellard | return 0; |
1525 | 158142c2 | bellard | } |
1526 | 158142c2 | bellard | aSig64 = aSig | 0x00800000;
|
1527 | 158142c2 | bellard | aSig64 <<= 40;
|
1528 | 158142c2 | bellard | z = aSig64>>( - shiftCount ); |
1529 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( aSig64<<( shiftCount & 63 ) ) ) { |
1530 | 158142c2 | bellard | STATUS(float_exception_flags) |= float_flag_inexact; |
1531 | 158142c2 | bellard | } |
1532 | 158142c2 | bellard | if ( aSign ) z = - z;
|
1533 | 158142c2 | bellard | return z;
|
1534 | 158142c2 | bellard | |
1535 | 158142c2 | bellard | } |
1536 | 158142c2 | bellard | |
1537 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1538 | 158142c2 | bellard | | Returns the result of converting the single-precision floating-point value
|
1539 | 158142c2 | bellard | | `a' to the double-precision floating-point format. The conversion is
|
1540 | 158142c2 | bellard | | performed according to the IEC/IEEE Standard for Binary Floating-Point
|
1541 | 158142c2 | bellard | | Arithmetic.
|
1542 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1543 | 158142c2 | bellard | |
1544 | 158142c2 | bellard | float64 float32_to_float64( float32 a STATUS_PARAM ) |
1545 | 158142c2 | bellard | { |
1546 | 158142c2 | bellard | flag aSign; |
1547 | 158142c2 | bellard | int16 aExp; |
1548 | bb98fe42 | Andreas Färber | uint32_t aSig; |
1549 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
1550 | 158142c2 | bellard | |
1551 | 158142c2 | bellard | aSig = extractFloat32Frac( a ); |
1552 | 158142c2 | bellard | aExp = extractFloat32Exp( a ); |
1553 | 158142c2 | bellard | aSign = extractFloat32Sign( a ); |
1554 | 158142c2 | bellard | if ( aExp == 0xFF ) { |
1555 | bcd4d9af | Christophe Lyon | if ( aSig ) return commonNaNToFloat64( float32ToCommonNaN( a STATUS_VAR ) STATUS_VAR ); |
1556 | 158142c2 | bellard | return packFloat64( aSign, 0x7FF, 0 ); |
1557 | 158142c2 | bellard | } |
1558 | 158142c2 | bellard | if ( aExp == 0 ) { |
1559 | 158142c2 | bellard | if ( aSig == 0 ) return packFloat64( aSign, 0, 0 ); |
1560 | 158142c2 | bellard | normalizeFloat32Subnormal( aSig, &aExp, &aSig ); |
1561 | 158142c2 | bellard | --aExp; |
1562 | 158142c2 | bellard | } |
1563 | bb98fe42 | Andreas Färber | return packFloat64( aSign, aExp + 0x380, ( (uint64_t) aSig )<<29 ); |
1564 | 158142c2 | bellard | |
1565 | 158142c2 | bellard | } |
1566 | 158142c2 | bellard | |
1567 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1568 | 158142c2 | bellard | | Returns the result of converting the single-precision floating-point value
|
1569 | 158142c2 | bellard | | `a' to the extended double-precision floating-point format. The conversion
|
1570 | 158142c2 | bellard | | is performed according to the IEC/IEEE Standard for Binary Floating-Point
|
1571 | 158142c2 | bellard | | Arithmetic.
|
1572 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1573 | 158142c2 | bellard | |
1574 | 158142c2 | bellard | floatx80 float32_to_floatx80( float32 a STATUS_PARAM ) |
1575 | 158142c2 | bellard | { |
1576 | 158142c2 | bellard | flag aSign; |
1577 | 158142c2 | bellard | int16 aExp; |
1578 | bb98fe42 | Andreas Färber | uint32_t aSig; |
1579 | 158142c2 | bellard | |
1580 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
1581 | 158142c2 | bellard | aSig = extractFloat32Frac( a ); |
1582 | 158142c2 | bellard | aExp = extractFloat32Exp( a ); |
1583 | 158142c2 | bellard | aSign = extractFloat32Sign( a ); |
1584 | 158142c2 | bellard | if ( aExp == 0xFF ) { |
1585 | bcd4d9af | Christophe Lyon | if ( aSig ) return commonNaNToFloatx80( float32ToCommonNaN( a STATUS_VAR ) STATUS_VAR ); |
1586 | 158142c2 | bellard | return packFloatx80( aSign, 0x7FFF, LIT64( 0x8000000000000000 ) ); |
1587 | 158142c2 | bellard | } |
1588 | 158142c2 | bellard | if ( aExp == 0 ) { |
1589 | 158142c2 | bellard | if ( aSig == 0 ) return packFloatx80( aSign, 0, 0 ); |
1590 | 158142c2 | bellard | normalizeFloat32Subnormal( aSig, &aExp, &aSig ); |
1591 | 158142c2 | bellard | } |
1592 | 158142c2 | bellard | aSig |= 0x00800000;
|
1593 | bb98fe42 | Andreas Färber | return packFloatx80( aSign, aExp + 0x3F80, ( (uint64_t) aSig )<<40 ); |
1594 | 158142c2 | bellard | |
1595 | 158142c2 | bellard | } |
1596 | 158142c2 | bellard | |
1597 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1598 | 158142c2 | bellard | | Returns the result of converting the single-precision floating-point value
|
1599 | 158142c2 | bellard | | `a' to the double-precision floating-point format. The conversion is
|
1600 | 158142c2 | bellard | | performed according to the IEC/IEEE Standard for Binary Floating-Point
|
1601 | 158142c2 | bellard | | Arithmetic.
|
1602 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1603 | 158142c2 | bellard | |
1604 | 158142c2 | bellard | float128 float32_to_float128( float32 a STATUS_PARAM ) |
1605 | 158142c2 | bellard | { |
1606 | 158142c2 | bellard | flag aSign; |
1607 | 158142c2 | bellard | int16 aExp; |
1608 | bb98fe42 | Andreas Färber | uint32_t aSig; |
1609 | 158142c2 | bellard | |
1610 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
1611 | 158142c2 | bellard | aSig = extractFloat32Frac( a ); |
1612 | 158142c2 | bellard | aExp = extractFloat32Exp( a ); |
1613 | 158142c2 | bellard | aSign = extractFloat32Sign( a ); |
1614 | 158142c2 | bellard | if ( aExp == 0xFF ) { |
1615 | bcd4d9af | Christophe Lyon | if ( aSig ) return commonNaNToFloat128( float32ToCommonNaN( a STATUS_VAR ) STATUS_VAR ); |
1616 | 158142c2 | bellard | return packFloat128( aSign, 0x7FFF, 0, 0 ); |
1617 | 158142c2 | bellard | } |
1618 | 158142c2 | bellard | if ( aExp == 0 ) { |
1619 | 158142c2 | bellard | if ( aSig == 0 ) return packFloat128( aSign, 0, 0, 0 ); |
1620 | 158142c2 | bellard | normalizeFloat32Subnormal( aSig, &aExp, &aSig ); |
1621 | 158142c2 | bellard | --aExp; |
1622 | 158142c2 | bellard | } |
1623 | bb98fe42 | Andreas Färber | return packFloat128( aSign, aExp + 0x3F80, ( (uint64_t) aSig )<<25, 0 ); |
1624 | 158142c2 | bellard | |
1625 | 158142c2 | bellard | } |
1626 | 158142c2 | bellard | |
1627 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1628 | 158142c2 | bellard | | Rounds the single-precision floating-point value `a' to an integer, and
|
1629 | 158142c2 | bellard | | returns the result as a single-precision floating-point value. The
|
1630 | 158142c2 | bellard | | operation is performed according to the IEC/IEEE Standard for Binary
|
1631 | 158142c2 | bellard | | Floating-Point Arithmetic.
|
1632 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1633 | 158142c2 | bellard | |
1634 | 158142c2 | bellard | float32 float32_round_to_int( float32 a STATUS_PARAM) |
1635 | 158142c2 | bellard | { |
1636 | 158142c2 | bellard | flag aSign; |
1637 | 158142c2 | bellard | int16 aExp; |
1638 | bb98fe42 | Andreas Färber | uint32_t lastBitMask, roundBitsMask; |
1639 | 158142c2 | bellard | int8 roundingMode; |
1640 | bb98fe42 | Andreas Färber | uint32_t z; |
1641 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
1642 | 158142c2 | bellard | |
1643 | 158142c2 | bellard | aExp = extractFloat32Exp( a ); |
1644 | 158142c2 | bellard | if ( 0x96 <= aExp ) { |
1645 | 158142c2 | bellard | if ( ( aExp == 0xFF ) && extractFloat32Frac( a ) ) { |
1646 | 158142c2 | bellard | return propagateFloat32NaN( a, a STATUS_VAR );
|
1647 | 158142c2 | bellard | } |
1648 | 158142c2 | bellard | return a;
|
1649 | 158142c2 | bellard | } |
1650 | 158142c2 | bellard | if ( aExp <= 0x7E ) { |
1651 | bb98fe42 | Andreas Färber | if ( (uint32_t) ( float32_val(a)<<1 ) == 0 ) return a; |
1652 | 158142c2 | bellard | STATUS(float_exception_flags) |= float_flag_inexact; |
1653 | 158142c2 | bellard | aSign = extractFloat32Sign( a ); |
1654 | 158142c2 | bellard | switch ( STATUS(float_rounding_mode) ) {
|
1655 | 158142c2 | bellard | case float_round_nearest_even:
|
1656 | 158142c2 | bellard | if ( ( aExp == 0x7E ) && extractFloat32Frac( a ) ) { |
1657 | 158142c2 | bellard | return packFloat32( aSign, 0x7F, 0 ); |
1658 | 158142c2 | bellard | } |
1659 | 158142c2 | bellard | break;
|
1660 | 158142c2 | bellard | case float_round_down:
|
1661 | f090c9d4 | pbrook | return make_float32(aSign ? 0xBF800000 : 0); |
1662 | 158142c2 | bellard | case float_round_up:
|
1663 | f090c9d4 | pbrook | return make_float32(aSign ? 0x80000000 : 0x3F800000); |
1664 | 158142c2 | bellard | } |
1665 | 158142c2 | bellard | return packFloat32( aSign, 0, 0 ); |
1666 | 158142c2 | bellard | } |
1667 | 158142c2 | bellard | lastBitMask = 1;
|
1668 | 158142c2 | bellard | lastBitMask <<= 0x96 - aExp;
|
1669 | 158142c2 | bellard | roundBitsMask = lastBitMask - 1;
|
1670 | f090c9d4 | pbrook | z = float32_val(a); |
1671 | 158142c2 | bellard | roundingMode = STATUS(float_rounding_mode); |
1672 | 158142c2 | bellard | if ( roundingMode == float_round_nearest_even ) {
|
1673 | 158142c2 | bellard | z += lastBitMask>>1;
|
1674 | 158142c2 | bellard | if ( ( z & roundBitsMask ) == 0 ) z &= ~ lastBitMask; |
1675 | 158142c2 | bellard | } |
1676 | 158142c2 | bellard | else if ( roundingMode != float_round_to_zero ) { |
1677 | f090c9d4 | pbrook | if ( extractFloat32Sign( make_float32(z) ) ^ ( roundingMode == float_round_up ) ) {
|
1678 | 158142c2 | bellard | z += roundBitsMask; |
1679 | 158142c2 | bellard | } |
1680 | 158142c2 | bellard | } |
1681 | 158142c2 | bellard | z &= ~ roundBitsMask; |
1682 | f090c9d4 | pbrook | if ( z != float32_val(a) ) STATUS(float_exception_flags) |= float_flag_inexact;
|
1683 | f090c9d4 | pbrook | return make_float32(z);
|
1684 | 158142c2 | bellard | |
1685 | 158142c2 | bellard | } |
1686 | 158142c2 | bellard | |
1687 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1688 | 158142c2 | bellard | | Returns the result of adding the absolute values of the single-precision
|
1689 | 158142c2 | bellard | | floating-point values `a' and `b'. If `zSign' is 1, the sum is negated
|
1690 | 158142c2 | bellard | | before being returned. `zSign' is ignored if the result is a NaN.
|
1691 | 158142c2 | bellard | | The addition is performed according to the IEC/IEEE Standard for Binary
|
1692 | 158142c2 | bellard | | Floating-Point Arithmetic.
|
1693 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1694 | 158142c2 | bellard | |
1695 | 158142c2 | bellard | static float32 addFloat32Sigs( float32 a, float32 b, flag zSign STATUS_PARAM)
|
1696 | 158142c2 | bellard | { |
1697 | 158142c2 | bellard | int16 aExp, bExp, zExp; |
1698 | bb98fe42 | Andreas Färber | uint32_t aSig, bSig, zSig; |
1699 | 158142c2 | bellard | int16 expDiff; |
1700 | 158142c2 | bellard | |
1701 | 158142c2 | bellard | aSig = extractFloat32Frac( a ); |
1702 | 158142c2 | bellard | aExp = extractFloat32Exp( a ); |
1703 | 158142c2 | bellard | bSig = extractFloat32Frac( b ); |
1704 | 158142c2 | bellard | bExp = extractFloat32Exp( b ); |
1705 | 158142c2 | bellard | expDiff = aExp - bExp; |
1706 | 158142c2 | bellard | aSig <<= 6;
|
1707 | 158142c2 | bellard | bSig <<= 6;
|
1708 | 158142c2 | bellard | if ( 0 < expDiff ) { |
1709 | 158142c2 | bellard | if ( aExp == 0xFF ) { |
1710 | 158142c2 | bellard | if ( aSig ) return propagateFloat32NaN( a, b STATUS_VAR ); |
1711 | 158142c2 | bellard | return a;
|
1712 | 158142c2 | bellard | } |
1713 | 158142c2 | bellard | if ( bExp == 0 ) { |
1714 | 158142c2 | bellard | --expDiff; |
1715 | 158142c2 | bellard | } |
1716 | 158142c2 | bellard | else {
|
1717 | 158142c2 | bellard | bSig |= 0x20000000;
|
1718 | 158142c2 | bellard | } |
1719 | 158142c2 | bellard | shift32RightJamming( bSig, expDiff, &bSig ); |
1720 | 158142c2 | bellard | zExp = aExp; |
1721 | 158142c2 | bellard | } |
1722 | 158142c2 | bellard | else if ( expDiff < 0 ) { |
1723 | 158142c2 | bellard | if ( bExp == 0xFF ) { |
1724 | 158142c2 | bellard | if ( bSig ) return propagateFloat32NaN( a, b STATUS_VAR ); |
1725 | 158142c2 | bellard | return packFloat32( zSign, 0xFF, 0 ); |
1726 | 158142c2 | bellard | } |
1727 | 158142c2 | bellard | if ( aExp == 0 ) { |
1728 | 158142c2 | bellard | ++expDiff; |
1729 | 158142c2 | bellard | } |
1730 | 158142c2 | bellard | else {
|
1731 | 158142c2 | bellard | aSig |= 0x20000000;
|
1732 | 158142c2 | bellard | } |
1733 | 158142c2 | bellard | shift32RightJamming( aSig, - expDiff, &aSig ); |
1734 | 158142c2 | bellard | zExp = bExp; |
1735 | 158142c2 | bellard | } |
1736 | 158142c2 | bellard | else {
|
1737 | 158142c2 | bellard | if ( aExp == 0xFF ) { |
1738 | 158142c2 | bellard | if ( aSig | bSig ) return propagateFloat32NaN( a, b STATUS_VAR ); |
1739 | 158142c2 | bellard | return a;
|
1740 | 158142c2 | bellard | } |
1741 | fe76d976 | pbrook | if ( aExp == 0 ) { |
1742 | e6afc87f | Peter Maydell | if (STATUS(flush_to_zero)) {
|
1743 | e6afc87f | Peter Maydell | if (aSig | bSig) {
|
1744 | e6afc87f | Peter Maydell | float_raise(float_flag_output_denormal STATUS_VAR); |
1745 | e6afc87f | Peter Maydell | } |
1746 | e6afc87f | Peter Maydell | return packFloat32(zSign, 0, 0); |
1747 | e6afc87f | Peter Maydell | } |
1748 | fe76d976 | pbrook | return packFloat32( zSign, 0, ( aSig + bSig )>>6 ); |
1749 | fe76d976 | pbrook | } |
1750 | 158142c2 | bellard | zSig = 0x40000000 + aSig + bSig;
|
1751 | 158142c2 | bellard | zExp = aExp; |
1752 | 158142c2 | bellard | goto roundAndPack;
|
1753 | 158142c2 | bellard | } |
1754 | 158142c2 | bellard | aSig |= 0x20000000;
|
1755 | 158142c2 | bellard | zSig = ( aSig + bSig )<<1;
|
1756 | 158142c2 | bellard | --zExp; |
1757 | bb98fe42 | Andreas Färber | if ( (int32_t) zSig < 0 ) { |
1758 | 158142c2 | bellard | zSig = aSig + bSig; |
1759 | 158142c2 | bellard | ++zExp; |
1760 | 158142c2 | bellard | } |
1761 | 158142c2 | bellard | roundAndPack:
|
1762 | 158142c2 | bellard | return roundAndPackFloat32( zSign, zExp, zSig STATUS_VAR );
|
1763 | 158142c2 | bellard | |
1764 | 158142c2 | bellard | } |
1765 | 158142c2 | bellard | |
1766 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1767 | 158142c2 | bellard | | Returns the result of subtracting the absolute values of the single-
|
1768 | 158142c2 | bellard | | precision floating-point values `a' and `b'. If `zSign' is 1, the
|
1769 | 158142c2 | bellard | | difference is negated before being returned. `zSign' is ignored if the
|
1770 | 158142c2 | bellard | | result is a NaN. The subtraction is performed according to the IEC/IEEE
|
1771 | 158142c2 | bellard | | Standard for Binary Floating-Point Arithmetic.
|
1772 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1773 | 158142c2 | bellard | |
1774 | 158142c2 | bellard | static float32 subFloat32Sigs( float32 a, float32 b, flag zSign STATUS_PARAM)
|
1775 | 158142c2 | bellard | { |
1776 | 158142c2 | bellard | int16 aExp, bExp, zExp; |
1777 | bb98fe42 | Andreas Färber | uint32_t aSig, bSig, zSig; |
1778 | 158142c2 | bellard | int16 expDiff; |
1779 | 158142c2 | bellard | |
1780 | 158142c2 | bellard | aSig = extractFloat32Frac( a ); |
1781 | 158142c2 | bellard | aExp = extractFloat32Exp( a ); |
1782 | 158142c2 | bellard | bSig = extractFloat32Frac( b ); |
1783 | 158142c2 | bellard | bExp = extractFloat32Exp( b ); |
1784 | 158142c2 | bellard | expDiff = aExp - bExp; |
1785 | 158142c2 | bellard | aSig <<= 7;
|
1786 | 158142c2 | bellard | bSig <<= 7;
|
1787 | 158142c2 | bellard | if ( 0 < expDiff ) goto aExpBigger; |
1788 | 158142c2 | bellard | if ( expDiff < 0 ) goto bExpBigger; |
1789 | 158142c2 | bellard | if ( aExp == 0xFF ) { |
1790 | 158142c2 | bellard | if ( aSig | bSig ) return propagateFloat32NaN( a, b STATUS_VAR ); |
1791 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
1792 | 158142c2 | bellard | return float32_default_nan;
|
1793 | 158142c2 | bellard | } |
1794 | 158142c2 | bellard | if ( aExp == 0 ) { |
1795 | 158142c2 | bellard | aExp = 1;
|
1796 | 158142c2 | bellard | bExp = 1;
|
1797 | 158142c2 | bellard | } |
1798 | 158142c2 | bellard | if ( bSig < aSig ) goto aBigger; |
1799 | 158142c2 | bellard | if ( aSig < bSig ) goto bBigger; |
1800 | 158142c2 | bellard | return packFloat32( STATUS(float_rounding_mode) == float_round_down, 0, 0 ); |
1801 | 158142c2 | bellard | bExpBigger:
|
1802 | 158142c2 | bellard | if ( bExp == 0xFF ) { |
1803 | 158142c2 | bellard | if ( bSig ) return propagateFloat32NaN( a, b STATUS_VAR ); |
1804 | 158142c2 | bellard | return packFloat32( zSign ^ 1, 0xFF, 0 ); |
1805 | 158142c2 | bellard | } |
1806 | 158142c2 | bellard | if ( aExp == 0 ) { |
1807 | 158142c2 | bellard | ++expDiff; |
1808 | 158142c2 | bellard | } |
1809 | 158142c2 | bellard | else {
|
1810 | 158142c2 | bellard | aSig |= 0x40000000;
|
1811 | 158142c2 | bellard | } |
1812 | 158142c2 | bellard | shift32RightJamming( aSig, - expDiff, &aSig ); |
1813 | 158142c2 | bellard | bSig |= 0x40000000;
|
1814 | 158142c2 | bellard | bBigger:
|
1815 | 158142c2 | bellard | zSig = bSig - aSig; |
1816 | 158142c2 | bellard | zExp = bExp; |
1817 | 158142c2 | bellard | zSign ^= 1;
|
1818 | 158142c2 | bellard | goto normalizeRoundAndPack;
|
1819 | 158142c2 | bellard | aExpBigger:
|
1820 | 158142c2 | bellard | if ( aExp == 0xFF ) { |
1821 | 158142c2 | bellard | if ( aSig ) return propagateFloat32NaN( a, b STATUS_VAR ); |
1822 | 158142c2 | bellard | return a;
|
1823 | 158142c2 | bellard | } |
1824 | 158142c2 | bellard | if ( bExp == 0 ) { |
1825 | 158142c2 | bellard | --expDiff; |
1826 | 158142c2 | bellard | } |
1827 | 158142c2 | bellard | else {
|
1828 | 158142c2 | bellard | bSig |= 0x40000000;
|
1829 | 158142c2 | bellard | } |
1830 | 158142c2 | bellard | shift32RightJamming( bSig, expDiff, &bSig ); |
1831 | 158142c2 | bellard | aSig |= 0x40000000;
|
1832 | 158142c2 | bellard | aBigger:
|
1833 | 158142c2 | bellard | zSig = aSig - bSig; |
1834 | 158142c2 | bellard | zExp = aExp; |
1835 | 158142c2 | bellard | normalizeRoundAndPack:
|
1836 | 158142c2 | bellard | --zExp; |
1837 | 158142c2 | bellard | return normalizeRoundAndPackFloat32( zSign, zExp, zSig STATUS_VAR );
|
1838 | 158142c2 | bellard | |
1839 | 158142c2 | bellard | } |
1840 | 158142c2 | bellard | |
1841 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1842 | 158142c2 | bellard | | Returns the result of adding the single-precision floating-point values `a'
|
1843 | 158142c2 | bellard | | and `b'. The operation is performed according to the IEC/IEEE Standard for
|
1844 | 158142c2 | bellard | | Binary Floating-Point Arithmetic.
|
1845 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1846 | 158142c2 | bellard | |
1847 | 158142c2 | bellard | float32 float32_add( float32 a, float32 b STATUS_PARAM ) |
1848 | 158142c2 | bellard | { |
1849 | 158142c2 | bellard | flag aSign, bSign; |
1850 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
1851 | 37d18660 | Peter Maydell | b = float32_squash_input_denormal(b STATUS_VAR); |
1852 | 158142c2 | bellard | |
1853 | 158142c2 | bellard | aSign = extractFloat32Sign( a ); |
1854 | 158142c2 | bellard | bSign = extractFloat32Sign( b ); |
1855 | 158142c2 | bellard | if ( aSign == bSign ) {
|
1856 | 158142c2 | bellard | return addFloat32Sigs( a, b, aSign STATUS_VAR);
|
1857 | 158142c2 | bellard | } |
1858 | 158142c2 | bellard | else {
|
1859 | 158142c2 | bellard | return subFloat32Sigs( a, b, aSign STATUS_VAR );
|
1860 | 158142c2 | bellard | } |
1861 | 158142c2 | bellard | |
1862 | 158142c2 | bellard | } |
1863 | 158142c2 | bellard | |
1864 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1865 | 158142c2 | bellard | | Returns the result of subtracting the single-precision floating-point values
|
1866 | 158142c2 | bellard | | `a' and `b'. The operation is performed according to the IEC/IEEE Standard
|
1867 | 158142c2 | bellard | | for Binary Floating-Point Arithmetic.
|
1868 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1869 | 158142c2 | bellard | |
1870 | 158142c2 | bellard | float32 float32_sub( float32 a, float32 b STATUS_PARAM ) |
1871 | 158142c2 | bellard | { |
1872 | 158142c2 | bellard | flag aSign, bSign; |
1873 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
1874 | 37d18660 | Peter Maydell | b = float32_squash_input_denormal(b STATUS_VAR); |
1875 | 158142c2 | bellard | |
1876 | 158142c2 | bellard | aSign = extractFloat32Sign( a ); |
1877 | 158142c2 | bellard | bSign = extractFloat32Sign( b ); |
1878 | 158142c2 | bellard | if ( aSign == bSign ) {
|
1879 | 158142c2 | bellard | return subFloat32Sigs( a, b, aSign STATUS_VAR );
|
1880 | 158142c2 | bellard | } |
1881 | 158142c2 | bellard | else {
|
1882 | 158142c2 | bellard | return addFloat32Sigs( a, b, aSign STATUS_VAR );
|
1883 | 158142c2 | bellard | } |
1884 | 158142c2 | bellard | |
1885 | 158142c2 | bellard | } |
1886 | 158142c2 | bellard | |
1887 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1888 | 158142c2 | bellard | | Returns the result of multiplying the single-precision floating-point values
|
1889 | 158142c2 | bellard | | `a' and `b'. The operation is performed according to the IEC/IEEE Standard
|
1890 | 158142c2 | bellard | | for Binary Floating-Point Arithmetic.
|
1891 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1892 | 158142c2 | bellard | |
1893 | 158142c2 | bellard | float32 float32_mul( float32 a, float32 b STATUS_PARAM ) |
1894 | 158142c2 | bellard | { |
1895 | 158142c2 | bellard | flag aSign, bSign, zSign; |
1896 | 158142c2 | bellard | int16 aExp, bExp, zExp; |
1897 | bb98fe42 | Andreas Färber | uint32_t aSig, bSig; |
1898 | bb98fe42 | Andreas Färber | uint64_t zSig64; |
1899 | bb98fe42 | Andreas Färber | uint32_t zSig; |
1900 | 158142c2 | bellard | |
1901 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
1902 | 37d18660 | Peter Maydell | b = float32_squash_input_denormal(b STATUS_VAR); |
1903 | 37d18660 | Peter Maydell | |
1904 | 158142c2 | bellard | aSig = extractFloat32Frac( a ); |
1905 | 158142c2 | bellard | aExp = extractFloat32Exp( a ); |
1906 | 158142c2 | bellard | aSign = extractFloat32Sign( a ); |
1907 | 158142c2 | bellard | bSig = extractFloat32Frac( b ); |
1908 | 158142c2 | bellard | bExp = extractFloat32Exp( b ); |
1909 | 158142c2 | bellard | bSign = extractFloat32Sign( b ); |
1910 | 158142c2 | bellard | zSign = aSign ^ bSign; |
1911 | 158142c2 | bellard | if ( aExp == 0xFF ) { |
1912 | 158142c2 | bellard | if ( aSig || ( ( bExp == 0xFF ) && bSig ) ) { |
1913 | 158142c2 | bellard | return propagateFloat32NaN( a, b STATUS_VAR );
|
1914 | 158142c2 | bellard | } |
1915 | 158142c2 | bellard | if ( ( bExp | bSig ) == 0 ) { |
1916 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
1917 | 158142c2 | bellard | return float32_default_nan;
|
1918 | 158142c2 | bellard | } |
1919 | 158142c2 | bellard | return packFloat32( zSign, 0xFF, 0 ); |
1920 | 158142c2 | bellard | } |
1921 | 158142c2 | bellard | if ( bExp == 0xFF ) { |
1922 | 158142c2 | bellard | if ( bSig ) return propagateFloat32NaN( a, b STATUS_VAR ); |
1923 | 158142c2 | bellard | if ( ( aExp | aSig ) == 0 ) { |
1924 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
1925 | 158142c2 | bellard | return float32_default_nan;
|
1926 | 158142c2 | bellard | } |
1927 | 158142c2 | bellard | return packFloat32( zSign, 0xFF, 0 ); |
1928 | 158142c2 | bellard | } |
1929 | 158142c2 | bellard | if ( aExp == 0 ) { |
1930 | 158142c2 | bellard | if ( aSig == 0 ) return packFloat32( zSign, 0, 0 ); |
1931 | 158142c2 | bellard | normalizeFloat32Subnormal( aSig, &aExp, &aSig ); |
1932 | 158142c2 | bellard | } |
1933 | 158142c2 | bellard | if ( bExp == 0 ) { |
1934 | 158142c2 | bellard | if ( bSig == 0 ) return packFloat32( zSign, 0, 0 ); |
1935 | 158142c2 | bellard | normalizeFloat32Subnormal( bSig, &bExp, &bSig ); |
1936 | 158142c2 | bellard | } |
1937 | 158142c2 | bellard | zExp = aExp + bExp - 0x7F;
|
1938 | 158142c2 | bellard | aSig = ( aSig | 0x00800000 )<<7; |
1939 | 158142c2 | bellard | bSig = ( bSig | 0x00800000 )<<8; |
1940 | bb98fe42 | Andreas Färber | shift64RightJamming( ( (uint64_t) aSig ) * bSig, 32, &zSig64 );
|
1941 | 158142c2 | bellard | zSig = zSig64; |
1942 | bb98fe42 | Andreas Färber | if ( 0 <= (int32_t) ( zSig<<1 ) ) { |
1943 | 158142c2 | bellard | zSig <<= 1;
|
1944 | 158142c2 | bellard | --zExp; |
1945 | 158142c2 | bellard | } |
1946 | 158142c2 | bellard | return roundAndPackFloat32( zSign, zExp, zSig STATUS_VAR );
|
1947 | 158142c2 | bellard | |
1948 | 158142c2 | bellard | } |
1949 | 158142c2 | bellard | |
1950 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
1951 | 158142c2 | bellard | | Returns the result of dividing the single-precision floating-point value `a'
|
1952 | 158142c2 | bellard | | by the corresponding value `b'. The operation is performed according to the
|
1953 | 158142c2 | bellard | | IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
1954 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
1955 | 158142c2 | bellard | |
1956 | 158142c2 | bellard | float32 float32_div( float32 a, float32 b STATUS_PARAM ) |
1957 | 158142c2 | bellard | { |
1958 | 158142c2 | bellard | flag aSign, bSign, zSign; |
1959 | 158142c2 | bellard | int16 aExp, bExp, zExp; |
1960 | bb98fe42 | Andreas Färber | uint32_t aSig, bSig, zSig; |
1961 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
1962 | 37d18660 | Peter Maydell | b = float32_squash_input_denormal(b STATUS_VAR); |
1963 | 158142c2 | bellard | |
1964 | 158142c2 | bellard | aSig = extractFloat32Frac( a ); |
1965 | 158142c2 | bellard | aExp = extractFloat32Exp( a ); |
1966 | 158142c2 | bellard | aSign = extractFloat32Sign( a ); |
1967 | 158142c2 | bellard | bSig = extractFloat32Frac( b ); |
1968 | 158142c2 | bellard | bExp = extractFloat32Exp( b ); |
1969 | 158142c2 | bellard | bSign = extractFloat32Sign( b ); |
1970 | 158142c2 | bellard | zSign = aSign ^ bSign; |
1971 | 158142c2 | bellard | if ( aExp == 0xFF ) { |
1972 | 158142c2 | bellard | if ( aSig ) return propagateFloat32NaN( a, b STATUS_VAR ); |
1973 | 158142c2 | bellard | if ( bExp == 0xFF ) { |
1974 | 158142c2 | bellard | if ( bSig ) return propagateFloat32NaN( a, b STATUS_VAR ); |
1975 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
1976 | 158142c2 | bellard | return float32_default_nan;
|
1977 | 158142c2 | bellard | } |
1978 | 158142c2 | bellard | return packFloat32( zSign, 0xFF, 0 ); |
1979 | 158142c2 | bellard | } |
1980 | 158142c2 | bellard | if ( bExp == 0xFF ) { |
1981 | 158142c2 | bellard | if ( bSig ) return propagateFloat32NaN( a, b STATUS_VAR ); |
1982 | 158142c2 | bellard | return packFloat32( zSign, 0, 0 ); |
1983 | 158142c2 | bellard | } |
1984 | 158142c2 | bellard | if ( bExp == 0 ) { |
1985 | 158142c2 | bellard | if ( bSig == 0 ) { |
1986 | 158142c2 | bellard | if ( ( aExp | aSig ) == 0 ) { |
1987 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
1988 | 158142c2 | bellard | return float32_default_nan;
|
1989 | 158142c2 | bellard | } |
1990 | 158142c2 | bellard | float_raise( float_flag_divbyzero STATUS_VAR); |
1991 | 158142c2 | bellard | return packFloat32( zSign, 0xFF, 0 ); |
1992 | 158142c2 | bellard | } |
1993 | 158142c2 | bellard | normalizeFloat32Subnormal( bSig, &bExp, &bSig ); |
1994 | 158142c2 | bellard | } |
1995 | 158142c2 | bellard | if ( aExp == 0 ) { |
1996 | 158142c2 | bellard | if ( aSig == 0 ) return packFloat32( zSign, 0, 0 ); |
1997 | 158142c2 | bellard | normalizeFloat32Subnormal( aSig, &aExp, &aSig ); |
1998 | 158142c2 | bellard | } |
1999 | 158142c2 | bellard | zExp = aExp - bExp + 0x7D;
|
2000 | 158142c2 | bellard | aSig = ( aSig | 0x00800000 )<<7; |
2001 | 158142c2 | bellard | bSig = ( bSig | 0x00800000 )<<8; |
2002 | 158142c2 | bellard | if ( bSig <= ( aSig + aSig ) ) {
|
2003 | 158142c2 | bellard | aSig >>= 1;
|
2004 | 158142c2 | bellard | ++zExp; |
2005 | 158142c2 | bellard | } |
2006 | bb98fe42 | Andreas Färber | zSig = ( ( (uint64_t) aSig )<<32 ) / bSig;
|
2007 | 158142c2 | bellard | if ( ( zSig & 0x3F ) == 0 ) { |
2008 | bb98fe42 | Andreas Färber | zSig |= ( (uint64_t) bSig * zSig != ( (uint64_t) aSig )<<32 );
|
2009 | 158142c2 | bellard | } |
2010 | 158142c2 | bellard | return roundAndPackFloat32( zSign, zExp, zSig STATUS_VAR );
|
2011 | 158142c2 | bellard | |
2012 | 158142c2 | bellard | } |
2013 | 158142c2 | bellard | |
2014 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
2015 | 158142c2 | bellard | | Returns the remainder of the single-precision floating-point value `a'
|
2016 | 158142c2 | bellard | | with respect to the corresponding value `b'. The operation is performed
|
2017 | 158142c2 | bellard | | according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
2018 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
2019 | 158142c2 | bellard | |
2020 | 158142c2 | bellard | float32 float32_rem( float32 a, float32 b STATUS_PARAM ) |
2021 | 158142c2 | bellard | { |
2022 | ed086f3d | Blue Swirl | flag aSign, zSign; |
2023 | 158142c2 | bellard | int16 aExp, bExp, expDiff; |
2024 | bb98fe42 | Andreas Färber | uint32_t aSig, bSig; |
2025 | bb98fe42 | Andreas Färber | uint32_t q; |
2026 | bb98fe42 | Andreas Färber | uint64_t aSig64, bSig64, q64; |
2027 | bb98fe42 | Andreas Färber | uint32_t alternateASig; |
2028 | bb98fe42 | Andreas Färber | int32_t sigMean; |
2029 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
2030 | 37d18660 | Peter Maydell | b = float32_squash_input_denormal(b STATUS_VAR); |
2031 | 158142c2 | bellard | |
2032 | 158142c2 | bellard | aSig = extractFloat32Frac( a ); |
2033 | 158142c2 | bellard | aExp = extractFloat32Exp( a ); |
2034 | 158142c2 | bellard | aSign = extractFloat32Sign( a ); |
2035 | 158142c2 | bellard | bSig = extractFloat32Frac( b ); |
2036 | 158142c2 | bellard | bExp = extractFloat32Exp( b ); |
2037 | 158142c2 | bellard | if ( aExp == 0xFF ) { |
2038 | 158142c2 | bellard | if ( aSig || ( ( bExp == 0xFF ) && bSig ) ) { |
2039 | 158142c2 | bellard | return propagateFloat32NaN( a, b STATUS_VAR );
|
2040 | 158142c2 | bellard | } |
2041 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
2042 | 158142c2 | bellard | return float32_default_nan;
|
2043 | 158142c2 | bellard | } |
2044 | 158142c2 | bellard | if ( bExp == 0xFF ) { |
2045 | 158142c2 | bellard | if ( bSig ) return propagateFloat32NaN( a, b STATUS_VAR ); |
2046 | 158142c2 | bellard | return a;
|
2047 | 158142c2 | bellard | } |
2048 | 158142c2 | bellard | if ( bExp == 0 ) { |
2049 | 158142c2 | bellard | if ( bSig == 0 ) { |
2050 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
2051 | 158142c2 | bellard | return float32_default_nan;
|
2052 | 158142c2 | bellard | } |
2053 | 158142c2 | bellard | normalizeFloat32Subnormal( bSig, &bExp, &bSig ); |
2054 | 158142c2 | bellard | } |
2055 | 158142c2 | bellard | if ( aExp == 0 ) { |
2056 | 158142c2 | bellard | if ( aSig == 0 ) return a; |
2057 | 158142c2 | bellard | normalizeFloat32Subnormal( aSig, &aExp, &aSig ); |
2058 | 158142c2 | bellard | } |
2059 | 158142c2 | bellard | expDiff = aExp - bExp; |
2060 | 158142c2 | bellard | aSig |= 0x00800000;
|
2061 | 158142c2 | bellard | bSig |= 0x00800000;
|
2062 | 158142c2 | bellard | if ( expDiff < 32 ) { |
2063 | 158142c2 | bellard | aSig <<= 8;
|
2064 | 158142c2 | bellard | bSig <<= 8;
|
2065 | 158142c2 | bellard | if ( expDiff < 0 ) { |
2066 | 158142c2 | bellard | if ( expDiff < -1 ) return a; |
2067 | 158142c2 | bellard | aSig >>= 1;
|
2068 | 158142c2 | bellard | } |
2069 | 158142c2 | bellard | q = ( bSig <= aSig ); |
2070 | 158142c2 | bellard | if ( q ) aSig -= bSig;
|
2071 | 158142c2 | bellard | if ( 0 < expDiff ) { |
2072 | bb98fe42 | Andreas Färber | q = ( ( (uint64_t) aSig )<<32 ) / bSig;
|
2073 | 158142c2 | bellard | q >>= 32 - expDiff;
|
2074 | 158142c2 | bellard | bSig >>= 2;
|
2075 | 158142c2 | bellard | aSig = ( ( aSig>>1 )<<( expDiff - 1 ) ) - bSig * q; |
2076 | 158142c2 | bellard | } |
2077 | 158142c2 | bellard | else {
|
2078 | 158142c2 | bellard | aSig >>= 2;
|
2079 | 158142c2 | bellard | bSig >>= 2;
|
2080 | 158142c2 | bellard | } |
2081 | 158142c2 | bellard | } |
2082 | 158142c2 | bellard | else {
|
2083 | 158142c2 | bellard | if ( bSig <= aSig ) aSig -= bSig;
|
2084 | bb98fe42 | Andreas Färber | aSig64 = ( (uint64_t) aSig )<<40;
|
2085 | bb98fe42 | Andreas Färber | bSig64 = ( (uint64_t) bSig )<<40;
|
2086 | 158142c2 | bellard | expDiff -= 64;
|
2087 | 158142c2 | bellard | while ( 0 < expDiff ) { |
2088 | 158142c2 | bellard | q64 = estimateDiv128To64( aSig64, 0, bSig64 );
|
2089 | 158142c2 | bellard | q64 = ( 2 < q64 ) ? q64 - 2 : 0; |
2090 | 158142c2 | bellard | aSig64 = - ( ( bSig * q64 )<<38 );
|
2091 | 158142c2 | bellard | expDiff -= 62;
|
2092 | 158142c2 | bellard | } |
2093 | 158142c2 | bellard | expDiff += 64;
|
2094 | 158142c2 | bellard | q64 = estimateDiv128To64( aSig64, 0, bSig64 );
|
2095 | 158142c2 | bellard | q64 = ( 2 < q64 ) ? q64 - 2 : 0; |
2096 | 158142c2 | bellard | q = q64>>( 64 - expDiff );
|
2097 | 158142c2 | bellard | bSig <<= 6;
|
2098 | 158142c2 | bellard | aSig = ( ( aSig64>>33 )<<( expDiff - 1 ) ) - bSig * q; |
2099 | 158142c2 | bellard | } |
2100 | 158142c2 | bellard | do {
|
2101 | 158142c2 | bellard | alternateASig = aSig; |
2102 | 158142c2 | bellard | ++q; |
2103 | 158142c2 | bellard | aSig -= bSig; |
2104 | bb98fe42 | Andreas Färber | } while ( 0 <= (int32_t) aSig ); |
2105 | 158142c2 | bellard | sigMean = aSig + alternateASig; |
2106 | 158142c2 | bellard | if ( ( sigMean < 0 ) || ( ( sigMean == 0 ) && ( q & 1 ) ) ) { |
2107 | 158142c2 | bellard | aSig = alternateASig; |
2108 | 158142c2 | bellard | } |
2109 | bb98fe42 | Andreas Färber | zSign = ( (int32_t) aSig < 0 );
|
2110 | 158142c2 | bellard | if ( zSign ) aSig = - aSig;
|
2111 | 158142c2 | bellard | return normalizeRoundAndPackFloat32( aSign ^ zSign, bExp, aSig STATUS_VAR );
|
2112 | 158142c2 | bellard | |
2113 | 158142c2 | bellard | } |
2114 | 158142c2 | bellard | |
2115 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
2116 | 158142c2 | bellard | | Returns the square root of the single-precision floating-point value `a'.
|
2117 | 158142c2 | bellard | | The operation is performed according to the IEC/IEEE Standard for Binary
|
2118 | 158142c2 | bellard | | Floating-Point Arithmetic.
|
2119 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
2120 | 158142c2 | bellard | |
2121 | 158142c2 | bellard | float32 float32_sqrt( float32 a STATUS_PARAM ) |
2122 | 158142c2 | bellard | { |
2123 | 158142c2 | bellard | flag aSign; |
2124 | 158142c2 | bellard | int16 aExp, zExp; |
2125 | bb98fe42 | Andreas Färber | uint32_t aSig, zSig; |
2126 | bb98fe42 | Andreas Färber | uint64_t rem, term; |
2127 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
2128 | 158142c2 | bellard | |
2129 | 158142c2 | bellard | aSig = extractFloat32Frac( a ); |
2130 | 158142c2 | bellard | aExp = extractFloat32Exp( a ); |
2131 | 158142c2 | bellard | aSign = extractFloat32Sign( a ); |
2132 | 158142c2 | bellard | if ( aExp == 0xFF ) { |
2133 | f090c9d4 | pbrook | if ( aSig ) return propagateFloat32NaN( a, float32_zero STATUS_VAR ); |
2134 | 158142c2 | bellard | if ( ! aSign ) return a; |
2135 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
2136 | 158142c2 | bellard | return float32_default_nan;
|
2137 | 158142c2 | bellard | } |
2138 | 158142c2 | bellard | if ( aSign ) {
|
2139 | 158142c2 | bellard | if ( ( aExp | aSig ) == 0 ) return a; |
2140 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
2141 | 158142c2 | bellard | return float32_default_nan;
|
2142 | 158142c2 | bellard | } |
2143 | 158142c2 | bellard | if ( aExp == 0 ) { |
2144 | f090c9d4 | pbrook | if ( aSig == 0 ) return float32_zero; |
2145 | 158142c2 | bellard | normalizeFloat32Subnormal( aSig, &aExp, &aSig ); |
2146 | 158142c2 | bellard | } |
2147 | 158142c2 | bellard | zExp = ( ( aExp - 0x7F )>>1 ) + 0x7E; |
2148 | 158142c2 | bellard | aSig = ( aSig | 0x00800000 )<<8; |
2149 | 158142c2 | bellard | zSig = estimateSqrt32( aExp, aSig ) + 2;
|
2150 | 158142c2 | bellard | if ( ( zSig & 0x7F ) <= 5 ) { |
2151 | 158142c2 | bellard | if ( zSig < 2 ) { |
2152 | 158142c2 | bellard | zSig = 0x7FFFFFFF;
|
2153 | 158142c2 | bellard | goto roundAndPack;
|
2154 | 158142c2 | bellard | } |
2155 | 158142c2 | bellard | aSig >>= aExp & 1;
|
2156 | bb98fe42 | Andreas Färber | term = ( (uint64_t) zSig ) * zSig; |
2157 | bb98fe42 | Andreas Färber | rem = ( ( (uint64_t) aSig )<<32 ) - term;
|
2158 | bb98fe42 | Andreas Färber | while ( (int64_t) rem < 0 ) { |
2159 | 158142c2 | bellard | --zSig; |
2160 | bb98fe42 | Andreas Färber | rem += ( ( (uint64_t) zSig )<<1 ) | 1; |
2161 | 158142c2 | bellard | } |
2162 | 158142c2 | bellard | zSig |= ( rem != 0 );
|
2163 | 158142c2 | bellard | } |
2164 | 158142c2 | bellard | shift32RightJamming( zSig, 1, &zSig );
|
2165 | 158142c2 | bellard | roundAndPack:
|
2166 | 158142c2 | bellard | return roundAndPackFloat32( 0, zExp, zSig STATUS_VAR ); |
2167 | 158142c2 | bellard | |
2168 | 158142c2 | bellard | } |
2169 | 158142c2 | bellard | |
2170 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
2171 | 8229c991 | Aurelien Jarno | | Returns the binary exponential of the single-precision floating-point value
|
2172 | 8229c991 | Aurelien Jarno | | `a'. The operation is performed according to the IEC/IEEE Standard for
|
2173 | 8229c991 | Aurelien Jarno | | Binary Floating-Point Arithmetic.
|
2174 | 8229c991 | Aurelien Jarno | |
|
2175 | 8229c991 | Aurelien Jarno | | Uses the following identities:
|
2176 | 8229c991 | Aurelien Jarno | |
|
2177 | 8229c991 | Aurelien Jarno | | 1. -------------------------------------------------------------------------
|
2178 | 8229c991 | Aurelien Jarno | | x x*ln(2)
|
2179 | 8229c991 | Aurelien Jarno | | 2 = e
|
2180 | 8229c991 | Aurelien Jarno | |
|
2181 | 8229c991 | Aurelien Jarno | | 2. -------------------------------------------------------------------------
|
2182 | 8229c991 | Aurelien Jarno | | 2 3 4 5 n
|
2183 | 8229c991 | Aurelien Jarno | | x x x x x x x
|
2184 | 8229c991 | Aurelien Jarno | | e = 1 + --- + --- + --- + --- + --- + ... + --- + ...
|
2185 | 8229c991 | Aurelien Jarno | | 1! 2! 3! 4! 5! n!
|
2186 | 8229c991 | Aurelien Jarno | *----------------------------------------------------------------------------*/
|
2187 | 8229c991 | Aurelien Jarno | |
2188 | 8229c991 | Aurelien Jarno | static const float64 float32_exp2_coefficients[15] = |
2189 | 8229c991 | Aurelien Jarno | { |
2190 | d5138cf4 | Peter Maydell | const_float64( 0x3ff0000000000000ll ), /* 1 */ |
2191 | d5138cf4 | Peter Maydell | const_float64( 0x3fe0000000000000ll ), /* 2 */ |
2192 | d5138cf4 | Peter Maydell | const_float64( 0x3fc5555555555555ll ), /* 3 */ |
2193 | d5138cf4 | Peter Maydell | const_float64( 0x3fa5555555555555ll ), /* 4 */ |
2194 | d5138cf4 | Peter Maydell | const_float64( 0x3f81111111111111ll ), /* 5 */ |
2195 | d5138cf4 | Peter Maydell | const_float64( 0x3f56c16c16c16c17ll ), /* 6 */ |
2196 | d5138cf4 | Peter Maydell | const_float64( 0x3f2a01a01a01a01all ), /* 7 */ |
2197 | d5138cf4 | Peter Maydell | const_float64( 0x3efa01a01a01a01all ), /* 8 */ |
2198 | d5138cf4 | Peter Maydell | const_float64( 0x3ec71de3a556c734ll ), /* 9 */ |
2199 | d5138cf4 | Peter Maydell | const_float64( 0x3e927e4fb7789f5cll ), /* 10 */ |
2200 | d5138cf4 | Peter Maydell | const_float64( 0x3e5ae64567f544e4ll ), /* 11 */ |
2201 | d5138cf4 | Peter Maydell | const_float64( 0x3e21eed8eff8d898ll ), /* 12 */ |
2202 | d5138cf4 | Peter Maydell | const_float64( 0x3de6124613a86d09ll ), /* 13 */ |
2203 | d5138cf4 | Peter Maydell | const_float64( 0x3da93974a8c07c9dll ), /* 14 */ |
2204 | d5138cf4 | Peter Maydell | const_float64( 0x3d6ae7f3e733b81fll ), /* 15 */ |
2205 | 8229c991 | Aurelien Jarno | }; |
2206 | 8229c991 | Aurelien Jarno | |
2207 | 8229c991 | Aurelien Jarno | float32 float32_exp2( float32 a STATUS_PARAM ) |
2208 | 8229c991 | Aurelien Jarno | { |
2209 | 8229c991 | Aurelien Jarno | flag aSign; |
2210 | 8229c991 | Aurelien Jarno | int16 aExp; |
2211 | bb98fe42 | Andreas Färber | uint32_t aSig; |
2212 | 8229c991 | Aurelien Jarno | float64 r, x, xn; |
2213 | 8229c991 | Aurelien Jarno | int i;
|
2214 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
2215 | 8229c991 | Aurelien Jarno | |
2216 | 8229c991 | Aurelien Jarno | aSig = extractFloat32Frac( a ); |
2217 | 8229c991 | Aurelien Jarno | aExp = extractFloat32Exp( a ); |
2218 | 8229c991 | Aurelien Jarno | aSign = extractFloat32Sign( a ); |
2219 | 8229c991 | Aurelien Jarno | |
2220 | 8229c991 | Aurelien Jarno | if ( aExp == 0xFF) { |
2221 | 8229c991 | Aurelien Jarno | if ( aSig ) return propagateFloat32NaN( a, float32_zero STATUS_VAR ); |
2222 | 8229c991 | Aurelien Jarno | return (aSign) ? float32_zero : a;
|
2223 | 8229c991 | Aurelien Jarno | } |
2224 | 8229c991 | Aurelien Jarno | if (aExp == 0) { |
2225 | 8229c991 | Aurelien Jarno | if (aSig == 0) return float32_one; |
2226 | 8229c991 | Aurelien Jarno | } |
2227 | 8229c991 | Aurelien Jarno | |
2228 | 8229c991 | Aurelien Jarno | float_raise( float_flag_inexact STATUS_VAR); |
2229 | 8229c991 | Aurelien Jarno | |
2230 | 8229c991 | Aurelien Jarno | /* ******************************* */
|
2231 | 8229c991 | Aurelien Jarno | /* using float64 for approximation */
|
2232 | 8229c991 | Aurelien Jarno | /* ******************************* */
|
2233 | 8229c991 | Aurelien Jarno | x = float32_to_float64(a STATUS_VAR); |
2234 | 8229c991 | Aurelien Jarno | x = float64_mul(x, float64_ln2 STATUS_VAR); |
2235 | 8229c991 | Aurelien Jarno | |
2236 | 8229c991 | Aurelien Jarno | xn = x; |
2237 | 8229c991 | Aurelien Jarno | r = float64_one; |
2238 | 8229c991 | Aurelien Jarno | for (i = 0 ; i < 15 ; i++) { |
2239 | 8229c991 | Aurelien Jarno | float64 f; |
2240 | 8229c991 | Aurelien Jarno | |
2241 | 8229c991 | Aurelien Jarno | f = float64_mul(xn, float32_exp2_coefficients[i] STATUS_VAR); |
2242 | 8229c991 | Aurelien Jarno | r = float64_add(r, f STATUS_VAR); |
2243 | 8229c991 | Aurelien Jarno | |
2244 | 8229c991 | Aurelien Jarno | xn = float64_mul(xn, x STATUS_VAR); |
2245 | 8229c991 | Aurelien Jarno | } |
2246 | 8229c991 | Aurelien Jarno | |
2247 | 8229c991 | Aurelien Jarno | return float64_to_float32(r, status);
|
2248 | 8229c991 | Aurelien Jarno | } |
2249 | 8229c991 | Aurelien Jarno | |
2250 | 8229c991 | Aurelien Jarno | /*----------------------------------------------------------------------------
|
2251 | 374dfc33 | aurel32 | | Returns the binary log of the single-precision floating-point value `a'.
|
2252 | 374dfc33 | aurel32 | | The operation is performed according to the IEC/IEEE Standard for Binary
|
2253 | 374dfc33 | aurel32 | | Floating-Point Arithmetic.
|
2254 | 374dfc33 | aurel32 | *----------------------------------------------------------------------------*/
|
2255 | 374dfc33 | aurel32 | float32 float32_log2( float32 a STATUS_PARAM ) |
2256 | 374dfc33 | aurel32 | { |
2257 | 374dfc33 | aurel32 | flag aSign, zSign; |
2258 | 374dfc33 | aurel32 | int16 aExp; |
2259 | bb98fe42 | Andreas Färber | uint32_t aSig, zSig, i; |
2260 | 374dfc33 | aurel32 | |
2261 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
2262 | 374dfc33 | aurel32 | aSig = extractFloat32Frac( a ); |
2263 | 374dfc33 | aurel32 | aExp = extractFloat32Exp( a ); |
2264 | 374dfc33 | aurel32 | aSign = extractFloat32Sign( a ); |
2265 | 374dfc33 | aurel32 | |
2266 | 374dfc33 | aurel32 | if ( aExp == 0 ) { |
2267 | 374dfc33 | aurel32 | if ( aSig == 0 ) return packFloat32( 1, 0xFF, 0 ); |
2268 | 374dfc33 | aurel32 | normalizeFloat32Subnormal( aSig, &aExp, &aSig ); |
2269 | 374dfc33 | aurel32 | } |
2270 | 374dfc33 | aurel32 | if ( aSign ) {
|
2271 | 374dfc33 | aurel32 | float_raise( float_flag_invalid STATUS_VAR); |
2272 | 374dfc33 | aurel32 | return float32_default_nan;
|
2273 | 374dfc33 | aurel32 | } |
2274 | 374dfc33 | aurel32 | if ( aExp == 0xFF ) { |
2275 | 374dfc33 | aurel32 | if ( aSig ) return propagateFloat32NaN( a, float32_zero STATUS_VAR ); |
2276 | 374dfc33 | aurel32 | return a;
|
2277 | 374dfc33 | aurel32 | } |
2278 | 374dfc33 | aurel32 | |
2279 | 374dfc33 | aurel32 | aExp -= 0x7F;
|
2280 | 374dfc33 | aurel32 | aSig |= 0x00800000;
|
2281 | 374dfc33 | aurel32 | zSign = aExp < 0;
|
2282 | 374dfc33 | aurel32 | zSig = aExp << 23;
|
2283 | 374dfc33 | aurel32 | |
2284 | 374dfc33 | aurel32 | for (i = 1 << 22; i > 0; i >>= 1) { |
2285 | bb98fe42 | Andreas Färber | aSig = ( (uint64_t)aSig * aSig ) >> 23;
|
2286 | 374dfc33 | aurel32 | if ( aSig & 0x01000000 ) { |
2287 | 374dfc33 | aurel32 | aSig >>= 1;
|
2288 | 374dfc33 | aurel32 | zSig |= i; |
2289 | 374dfc33 | aurel32 | } |
2290 | 374dfc33 | aurel32 | } |
2291 | 374dfc33 | aurel32 | |
2292 | 374dfc33 | aurel32 | if ( zSign )
|
2293 | 374dfc33 | aurel32 | zSig = -zSig; |
2294 | 374dfc33 | aurel32 | |
2295 | 374dfc33 | aurel32 | return normalizeRoundAndPackFloat32( zSign, 0x85, zSig STATUS_VAR ); |
2296 | 374dfc33 | aurel32 | } |
2297 | 374dfc33 | aurel32 | |
2298 | 374dfc33 | aurel32 | /*----------------------------------------------------------------------------
|
2299 | 158142c2 | bellard | | Returns 1 if the single-precision floating-point value `a' is equal to
|
2300 | b689362d | Aurelien Jarno | | the corresponding value `b', and 0 otherwise. The invalid exception is
|
2301 | b689362d | Aurelien Jarno | | raised if either operand is a NaN. Otherwise, the comparison is performed
|
2302 | 158142c2 | bellard | | according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
2303 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
2304 | 158142c2 | bellard | |
2305 | b689362d | Aurelien Jarno | int float32_eq( float32 a, float32 b STATUS_PARAM )
|
2306 | 158142c2 | bellard | { |
2307 | b689362d | Aurelien Jarno | uint32_t av, bv; |
2308 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
2309 | 37d18660 | Peter Maydell | b = float32_squash_input_denormal(b STATUS_VAR); |
2310 | 158142c2 | bellard | |
2311 | 158142c2 | bellard | if ( ( ( extractFloat32Exp( a ) == 0xFF ) && extractFloat32Frac( a ) ) |
2312 | 158142c2 | bellard | || ( ( extractFloat32Exp( b ) == 0xFF ) && extractFloat32Frac( b ) )
|
2313 | 158142c2 | bellard | ) { |
2314 | b689362d | Aurelien Jarno | float_raise( float_flag_invalid STATUS_VAR); |
2315 | 158142c2 | bellard | return 0; |
2316 | 158142c2 | bellard | } |
2317 | b689362d | Aurelien Jarno | av = float32_val(a); |
2318 | b689362d | Aurelien Jarno | bv = float32_val(b); |
2319 | b689362d | Aurelien Jarno | return ( av == bv ) || ( (uint32_t) ( ( av | bv )<<1 ) == 0 ); |
2320 | 158142c2 | bellard | } |
2321 | 158142c2 | bellard | |
2322 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
2323 | 158142c2 | bellard | | Returns 1 if the single-precision floating-point value `a' is less than
|
2324 | f5a64251 | Aurelien Jarno | | or equal to the corresponding value `b', and 0 otherwise. The invalid
|
2325 | f5a64251 | Aurelien Jarno | | exception is raised if either operand is a NaN. The comparison is performed
|
2326 | f5a64251 | Aurelien Jarno | | according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
2327 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
2328 | 158142c2 | bellard | |
2329 | 750afe93 | bellard | int float32_le( float32 a, float32 b STATUS_PARAM )
|
2330 | 158142c2 | bellard | { |
2331 | 158142c2 | bellard | flag aSign, bSign; |
2332 | bb98fe42 | Andreas Färber | uint32_t av, bv; |
2333 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
2334 | 37d18660 | Peter Maydell | b = float32_squash_input_denormal(b STATUS_VAR); |
2335 | 158142c2 | bellard | |
2336 | 158142c2 | bellard | if ( ( ( extractFloat32Exp( a ) == 0xFF ) && extractFloat32Frac( a ) ) |
2337 | 158142c2 | bellard | || ( ( extractFloat32Exp( b ) == 0xFF ) && extractFloat32Frac( b ) )
|
2338 | 158142c2 | bellard | ) { |
2339 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
2340 | 158142c2 | bellard | return 0; |
2341 | 158142c2 | bellard | } |
2342 | 158142c2 | bellard | aSign = extractFloat32Sign( a ); |
2343 | 158142c2 | bellard | bSign = extractFloat32Sign( b ); |
2344 | f090c9d4 | pbrook | av = float32_val(a); |
2345 | f090c9d4 | pbrook | bv = float32_val(b); |
2346 | bb98fe42 | Andreas Färber | if ( aSign != bSign ) return aSign || ( (uint32_t) ( ( av | bv )<<1 ) == 0 ); |
2347 | f090c9d4 | pbrook | return ( av == bv ) || ( aSign ^ ( av < bv ) );
|
2348 | 158142c2 | bellard | |
2349 | 158142c2 | bellard | } |
2350 | 158142c2 | bellard | |
2351 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
2352 | 158142c2 | bellard | | Returns 1 if the single-precision floating-point value `a' is less than
|
2353 | f5a64251 | Aurelien Jarno | | the corresponding value `b', and 0 otherwise. The invalid exception is
|
2354 | f5a64251 | Aurelien Jarno | | raised if either operand is a NaN. The comparison is performed according
|
2355 | f5a64251 | Aurelien Jarno | | to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
2356 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
2357 | 158142c2 | bellard | |
2358 | 750afe93 | bellard | int float32_lt( float32 a, float32 b STATUS_PARAM )
|
2359 | 158142c2 | bellard | { |
2360 | 158142c2 | bellard | flag aSign, bSign; |
2361 | bb98fe42 | Andreas Färber | uint32_t av, bv; |
2362 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
2363 | 37d18660 | Peter Maydell | b = float32_squash_input_denormal(b STATUS_VAR); |
2364 | 158142c2 | bellard | |
2365 | 158142c2 | bellard | if ( ( ( extractFloat32Exp( a ) == 0xFF ) && extractFloat32Frac( a ) ) |
2366 | 158142c2 | bellard | || ( ( extractFloat32Exp( b ) == 0xFF ) && extractFloat32Frac( b ) )
|
2367 | 158142c2 | bellard | ) { |
2368 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
2369 | 158142c2 | bellard | return 0; |
2370 | 158142c2 | bellard | } |
2371 | 158142c2 | bellard | aSign = extractFloat32Sign( a ); |
2372 | 158142c2 | bellard | bSign = extractFloat32Sign( b ); |
2373 | f090c9d4 | pbrook | av = float32_val(a); |
2374 | f090c9d4 | pbrook | bv = float32_val(b); |
2375 | bb98fe42 | Andreas Färber | if ( aSign != bSign ) return aSign && ( (uint32_t) ( ( av | bv )<<1 ) != 0 ); |
2376 | f090c9d4 | pbrook | return ( av != bv ) && ( aSign ^ ( av < bv ) );
|
2377 | 158142c2 | bellard | |
2378 | 158142c2 | bellard | } |
2379 | 158142c2 | bellard | |
2380 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
2381 | 67b7861d | Aurelien Jarno | | Returns 1 if the single-precision floating-point values `a' and `b' cannot
|
2382 | f5a64251 | Aurelien Jarno | | be compared, and 0 otherwise. The invalid exception is raised if either
|
2383 | f5a64251 | Aurelien Jarno | | operand is a NaN. The comparison is performed according to the IEC/IEEE
|
2384 | f5a64251 | Aurelien Jarno | | Standard for Binary Floating-Point Arithmetic.
|
2385 | 67b7861d | Aurelien Jarno | *----------------------------------------------------------------------------*/
|
2386 | 67b7861d | Aurelien Jarno | |
2387 | 67b7861d | Aurelien Jarno | int float32_unordered( float32 a, float32 b STATUS_PARAM )
|
2388 | 67b7861d | Aurelien Jarno | { |
2389 | 67b7861d | Aurelien Jarno | a = float32_squash_input_denormal(a STATUS_VAR); |
2390 | 67b7861d | Aurelien Jarno | b = float32_squash_input_denormal(b STATUS_VAR); |
2391 | 67b7861d | Aurelien Jarno | |
2392 | 67b7861d | Aurelien Jarno | if ( ( ( extractFloat32Exp( a ) == 0xFF ) && extractFloat32Frac( a ) ) |
2393 | 67b7861d | Aurelien Jarno | || ( ( extractFloat32Exp( b ) == 0xFF ) && extractFloat32Frac( b ) )
|
2394 | 67b7861d | Aurelien Jarno | ) { |
2395 | 67b7861d | Aurelien Jarno | float_raise( float_flag_invalid STATUS_VAR); |
2396 | 67b7861d | Aurelien Jarno | return 1; |
2397 | 67b7861d | Aurelien Jarno | } |
2398 | 67b7861d | Aurelien Jarno | return 0; |
2399 | 67b7861d | Aurelien Jarno | } |
2400 | b689362d | Aurelien Jarno | |
2401 | 67b7861d | Aurelien Jarno | /*----------------------------------------------------------------------------
|
2402 | 158142c2 | bellard | | Returns 1 if the single-precision floating-point value `a' is equal to
|
2403 | f5a64251 | Aurelien Jarno | | the corresponding value `b', and 0 otherwise. Quiet NaNs do not cause an
|
2404 | f5a64251 | Aurelien Jarno | | exception. The comparison is performed according to the IEC/IEEE Standard
|
2405 | f5a64251 | Aurelien Jarno | | for Binary Floating-Point Arithmetic.
|
2406 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
2407 | 158142c2 | bellard | |
2408 | b689362d | Aurelien Jarno | int float32_eq_quiet( float32 a, float32 b STATUS_PARAM )
|
2409 | 158142c2 | bellard | { |
2410 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
2411 | 37d18660 | Peter Maydell | b = float32_squash_input_denormal(b STATUS_VAR); |
2412 | 158142c2 | bellard | |
2413 | 158142c2 | bellard | if ( ( ( extractFloat32Exp( a ) == 0xFF ) && extractFloat32Frac( a ) ) |
2414 | 158142c2 | bellard | || ( ( extractFloat32Exp( b ) == 0xFF ) && extractFloat32Frac( b ) )
|
2415 | 158142c2 | bellard | ) { |
2416 | b689362d | Aurelien Jarno | if ( float32_is_signaling_nan( a ) || float32_is_signaling_nan( b ) ) {
|
2417 | b689362d | Aurelien Jarno | float_raise( float_flag_invalid STATUS_VAR); |
2418 | b689362d | Aurelien Jarno | } |
2419 | 158142c2 | bellard | return 0; |
2420 | 158142c2 | bellard | } |
2421 | b689362d | Aurelien Jarno | return ( float32_val(a) == float32_val(b) ) ||
|
2422 | b689362d | Aurelien Jarno | ( (uint32_t) ( ( float32_val(a) | float32_val(b) )<<1 ) == 0 ); |
2423 | 158142c2 | bellard | } |
2424 | 158142c2 | bellard | |
2425 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
2426 | 158142c2 | bellard | | Returns 1 if the single-precision floating-point value `a' is less than or
|
2427 | 158142c2 | bellard | | equal to the corresponding value `b', and 0 otherwise. Quiet NaNs do not
|
2428 | 158142c2 | bellard | | cause an exception. Otherwise, the comparison is performed according to the
|
2429 | 158142c2 | bellard | | IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
2430 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
2431 | 158142c2 | bellard | |
2432 | 750afe93 | bellard | int float32_le_quiet( float32 a, float32 b STATUS_PARAM )
|
2433 | 158142c2 | bellard | { |
2434 | 158142c2 | bellard | flag aSign, bSign; |
2435 | bb98fe42 | Andreas Färber | uint32_t av, bv; |
2436 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
2437 | 37d18660 | Peter Maydell | b = float32_squash_input_denormal(b STATUS_VAR); |
2438 | 158142c2 | bellard | |
2439 | 158142c2 | bellard | if ( ( ( extractFloat32Exp( a ) == 0xFF ) && extractFloat32Frac( a ) ) |
2440 | 158142c2 | bellard | || ( ( extractFloat32Exp( b ) == 0xFF ) && extractFloat32Frac( b ) )
|
2441 | 158142c2 | bellard | ) { |
2442 | 158142c2 | bellard | if ( float32_is_signaling_nan( a ) || float32_is_signaling_nan( b ) ) {
|
2443 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
2444 | 158142c2 | bellard | } |
2445 | 158142c2 | bellard | return 0; |
2446 | 158142c2 | bellard | } |
2447 | 158142c2 | bellard | aSign = extractFloat32Sign( a ); |
2448 | 158142c2 | bellard | bSign = extractFloat32Sign( b ); |
2449 | f090c9d4 | pbrook | av = float32_val(a); |
2450 | f090c9d4 | pbrook | bv = float32_val(b); |
2451 | bb98fe42 | Andreas Färber | if ( aSign != bSign ) return aSign || ( (uint32_t) ( ( av | bv )<<1 ) == 0 ); |
2452 | f090c9d4 | pbrook | return ( av == bv ) || ( aSign ^ ( av < bv ) );
|
2453 | 158142c2 | bellard | |
2454 | 158142c2 | bellard | } |
2455 | 158142c2 | bellard | |
2456 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
2457 | 158142c2 | bellard | | Returns 1 if the single-precision floating-point value `a' is less than
|
2458 | 158142c2 | bellard | | the corresponding value `b', and 0 otherwise. Quiet NaNs do not cause an
|
2459 | 158142c2 | bellard | | exception. Otherwise, the comparison is performed according to the IEC/IEEE
|
2460 | 158142c2 | bellard | | Standard for Binary Floating-Point Arithmetic.
|
2461 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
2462 | 158142c2 | bellard | |
2463 | 750afe93 | bellard | int float32_lt_quiet( float32 a, float32 b STATUS_PARAM )
|
2464 | 158142c2 | bellard | { |
2465 | 158142c2 | bellard | flag aSign, bSign; |
2466 | bb98fe42 | Andreas Färber | uint32_t av, bv; |
2467 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
2468 | 37d18660 | Peter Maydell | b = float32_squash_input_denormal(b STATUS_VAR); |
2469 | 158142c2 | bellard | |
2470 | 158142c2 | bellard | if ( ( ( extractFloat32Exp( a ) == 0xFF ) && extractFloat32Frac( a ) ) |
2471 | 158142c2 | bellard | || ( ( extractFloat32Exp( b ) == 0xFF ) && extractFloat32Frac( b ) )
|
2472 | 158142c2 | bellard | ) { |
2473 | 158142c2 | bellard | if ( float32_is_signaling_nan( a ) || float32_is_signaling_nan( b ) ) {
|
2474 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
2475 | 158142c2 | bellard | } |
2476 | 158142c2 | bellard | return 0; |
2477 | 158142c2 | bellard | } |
2478 | 158142c2 | bellard | aSign = extractFloat32Sign( a ); |
2479 | 158142c2 | bellard | bSign = extractFloat32Sign( b ); |
2480 | f090c9d4 | pbrook | av = float32_val(a); |
2481 | f090c9d4 | pbrook | bv = float32_val(b); |
2482 | bb98fe42 | Andreas Färber | if ( aSign != bSign ) return aSign && ( (uint32_t) ( ( av | bv )<<1 ) != 0 ); |
2483 | f090c9d4 | pbrook | return ( av != bv ) && ( aSign ^ ( av < bv ) );
|
2484 | 158142c2 | bellard | |
2485 | 158142c2 | bellard | } |
2486 | 158142c2 | bellard | |
2487 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
2488 | 67b7861d | Aurelien Jarno | | Returns 1 if the single-precision floating-point values `a' and `b' cannot
|
2489 | 67b7861d | Aurelien Jarno | | be compared, and 0 otherwise. Quiet NaNs do not cause an exception. The
|
2490 | 67b7861d | Aurelien Jarno | | comparison is performed according to the IEC/IEEE Standard for Binary
|
2491 | 67b7861d | Aurelien Jarno | | Floating-Point Arithmetic.
|
2492 | 67b7861d | Aurelien Jarno | *----------------------------------------------------------------------------*/
|
2493 | 67b7861d | Aurelien Jarno | |
2494 | 67b7861d | Aurelien Jarno | int float32_unordered_quiet( float32 a, float32 b STATUS_PARAM )
|
2495 | 67b7861d | Aurelien Jarno | { |
2496 | 67b7861d | Aurelien Jarno | a = float32_squash_input_denormal(a STATUS_VAR); |
2497 | 67b7861d | Aurelien Jarno | b = float32_squash_input_denormal(b STATUS_VAR); |
2498 | 67b7861d | Aurelien Jarno | |
2499 | 67b7861d | Aurelien Jarno | if ( ( ( extractFloat32Exp( a ) == 0xFF ) && extractFloat32Frac( a ) ) |
2500 | 67b7861d | Aurelien Jarno | || ( ( extractFloat32Exp( b ) == 0xFF ) && extractFloat32Frac( b ) )
|
2501 | 67b7861d | Aurelien Jarno | ) { |
2502 | 67b7861d | Aurelien Jarno | if ( float32_is_signaling_nan( a ) || float32_is_signaling_nan( b ) ) {
|
2503 | 67b7861d | Aurelien Jarno | float_raise( float_flag_invalid STATUS_VAR); |
2504 | 67b7861d | Aurelien Jarno | } |
2505 | 67b7861d | Aurelien Jarno | return 1; |
2506 | 67b7861d | Aurelien Jarno | } |
2507 | 67b7861d | Aurelien Jarno | return 0; |
2508 | 67b7861d | Aurelien Jarno | } |
2509 | 67b7861d | Aurelien Jarno | |
2510 | 67b7861d | Aurelien Jarno | /*----------------------------------------------------------------------------
|
2511 | 158142c2 | bellard | | Returns the result of converting the double-precision floating-point value
|
2512 | 158142c2 | bellard | | `a' to the 32-bit two's complement integer format. The conversion is
|
2513 | 158142c2 | bellard | | performed according to the IEC/IEEE Standard for Binary Floating-Point
|
2514 | 158142c2 | bellard | | Arithmetic---which means in particular that the conversion is rounded
|
2515 | 158142c2 | bellard | | according to the current rounding mode. If `a' is a NaN, the largest
|
2516 | 158142c2 | bellard | | positive integer is returned. Otherwise, if the conversion overflows, the
|
2517 | 158142c2 | bellard | | largest integer with the same sign as `a' is returned.
|
2518 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
2519 | 158142c2 | bellard | |
2520 | 158142c2 | bellard | int32 float64_to_int32( float64 a STATUS_PARAM ) |
2521 | 158142c2 | bellard | { |
2522 | 158142c2 | bellard | flag aSign; |
2523 | 158142c2 | bellard | int16 aExp, shiftCount; |
2524 | bb98fe42 | Andreas Färber | uint64_t aSig; |
2525 | 37d18660 | Peter Maydell | a = float64_squash_input_denormal(a STATUS_VAR); |
2526 | 158142c2 | bellard | |
2527 | 158142c2 | bellard | aSig = extractFloat64Frac( a ); |
2528 | 158142c2 | bellard | aExp = extractFloat64Exp( a ); |
2529 | 158142c2 | bellard | aSign = extractFloat64Sign( a ); |
2530 | 158142c2 | bellard | if ( ( aExp == 0x7FF ) && aSig ) aSign = 0; |
2531 | 158142c2 | bellard | if ( aExp ) aSig |= LIT64( 0x0010000000000000 ); |
2532 | 158142c2 | bellard | shiftCount = 0x42C - aExp;
|
2533 | 158142c2 | bellard | if ( 0 < shiftCount ) shift64RightJamming( aSig, shiftCount, &aSig ); |
2534 | 158142c2 | bellard | return roundAndPackInt32( aSign, aSig STATUS_VAR );
|
2535 | 158142c2 | bellard | |
2536 | 158142c2 | bellard | } |
2537 | 158142c2 | bellard | |
2538 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
2539 | 158142c2 | bellard | | Returns the result of converting the double-precision floating-point value
|
2540 | 158142c2 | bellard | | `a' to the 32-bit two's complement integer format. The conversion is
|
2541 | 158142c2 | bellard | | performed according to the IEC/IEEE Standard for Binary Floating-Point
|
2542 | 158142c2 | bellard | | Arithmetic, except that the conversion is always rounded toward zero.
|
2543 | 158142c2 | bellard | | If `a' is a NaN, the largest positive integer is returned. Otherwise, if
|
2544 | 158142c2 | bellard | | the conversion overflows, the largest integer with the same sign as `a' is
|
2545 | 158142c2 | bellard | | returned.
|
2546 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
2547 | 158142c2 | bellard | |
2548 | 158142c2 | bellard | int32 float64_to_int32_round_to_zero( float64 a STATUS_PARAM ) |
2549 | 158142c2 | bellard | { |
2550 | 158142c2 | bellard | flag aSign; |
2551 | 158142c2 | bellard | int16 aExp, shiftCount; |
2552 | bb98fe42 | Andreas Färber | uint64_t aSig, savedASig; |
2553 | 158142c2 | bellard | int32 z; |
2554 | 37d18660 | Peter Maydell | a = float64_squash_input_denormal(a STATUS_VAR); |
2555 | 158142c2 | bellard | |
2556 | 158142c2 | bellard | aSig = extractFloat64Frac( a ); |
2557 | 158142c2 | bellard | aExp = extractFloat64Exp( a ); |
2558 | 158142c2 | bellard | aSign = extractFloat64Sign( a ); |
2559 | 158142c2 | bellard | if ( 0x41E < aExp ) { |
2560 | 158142c2 | bellard | if ( ( aExp == 0x7FF ) && aSig ) aSign = 0; |
2561 | 158142c2 | bellard | goto invalid;
|
2562 | 158142c2 | bellard | } |
2563 | 158142c2 | bellard | else if ( aExp < 0x3FF ) { |
2564 | 158142c2 | bellard | if ( aExp || aSig ) STATUS(float_exception_flags) |= float_flag_inexact;
|
2565 | 158142c2 | bellard | return 0; |
2566 | 158142c2 | bellard | } |
2567 | 158142c2 | bellard | aSig |= LIT64( 0x0010000000000000 );
|
2568 | 158142c2 | bellard | shiftCount = 0x433 - aExp;
|
2569 | 158142c2 | bellard | savedASig = aSig; |
2570 | 158142c2 | bellard | aSig >>= shiftCount; |
2571 | 158142c2 | bellard | z = aSig; |
2572 | 158142c2 | bellard | if ( aSign ) z = - z;
|
2573 | 158142c2 | bellard | if ( ( z < 0 ) ^ aSign ) { |
2574 | 158142c2 | bellard | invalid:
|
2575 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
2576 | bb98fe42 | Andreas Färber | return aSign ? (int32_t) 0x80000000 : 0x7FFFFFFF; |
2577 | 158142c2 | bellard | } |
2578 | 158142c2 | bellard | if ( ( aSig<<shiftCount ) != savedASig ) {
|
2579 | 158142c2 | bellard | STATUS(float_exception_flags) |= float_flag_inexact; |
2580 | 158142c2 | bellard | } |
2581 | 158142c2 | bellard | return z;
|
2582 | 158142c2 | bellard | |
2583 | 158142c2 | bellard | } |
2584 | 158142c2 | bellard | |
2585 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
2586 | 158142c2 | bellard | | Returns the result of converting the double-precision floating-point value
|
2587 | cbcef455 | Peter Maydell | | `a' to the 16-bit two's complement integer format. The conversion is
|
2588 | cbcef455 | Peter Maydell | | performed according to the IEC/IEEE Standard for Binary Floating-Point
|
2589 | cbcef455 | Peter Maydell | | Arithmetic, except that the conversion is always rounded toward zero.
|
2590 | cbcef455 | Peter Maydell | | If `a' is a NaN, the largest positive integer is returned. Otherwise, if
|
2591 | cbcef455 | Peter Maydell | | the conversion overflows, the largest integer with the same sign as `a' is
|
2592 | cbcef455 | Peter Maydell | | returned.
|
2593 | cbcef455 | Peter Maydell | *----------------------------------------------------------------------------*/
|
2594 | cbcef455 | Peter Maydell | |
2595 | cbcef455 | Peter Maydell | int16 float64_to_int16_round_to_zero( float64 a STATUS_PARAM ) |
2596 | cbcef455 | Peter Maydell | { |
2597 | cbcef455 | Peter Maydell | flag aSign; |
2598 | cbcef455 | Peter Maydell | int16 aExp, shiftCount; |
2599 | bb98fe42 | Andreas Färber | uint64_t aSig, savedASig; |
2600 | cbcef455 | Peter Maydell | int32 z; |
2601 | cbcef455 | Peter Maydell | |
2602 | cbcef455 | Peter Maydell | aSig = extractFloat64Frac( a ); |
2603 | cbcef455 | Peter Maydell | aExp = extractFloat64Exp( a ); |
2604 | cbcef455 | Peter Maydell | aSign = extractFloat64Sign( a ); |
2605 | cbcef455 | Peter Maydell | if ( 0x40E < aExp ) { |
2606 | cbcef455 | Peter Maydell | if ( ( aExp == 0x7FF ) && aSig ) { |
2607 | cbcef455 | Peter Maydell | aSign = 0;
|
2608 | cbcef455 | Peter Maydell | } |
2609 | cbcef455 | Peter Maydell | goto invalid;
|
2610 | cbcef455 | Peter Maydell | } |
2611 | cbcef455 | Peter Maydell | else if ( aExp < 0x3FF ) { |
2612 | cbcef455 | Peter Maydell | if ( aExp || aSig ) {
|
2613 | cbcef455 | Peter Maydell | STATUS(float_exception_flags) |= float_flag_inexact; |
2614 | cbcef455 | Peter Maydell | } |
2615 | cbcef455 | Peter Maydell | return 0; |
2616 | cbcef455 | Peter Maydell | } |
2617 | cbcef455 | Peter Maydell | aSig |= LIT64( 0x0010000000000000 );
|
2618 | cbcef455 | Peter Maydell | shiftCount = 0x433 - aExp;
|
2619 | cbcef455 | Peter Maydell | savedASig = aSig; |
2620 | cbcef455 | Peter Maydell | aSig >>= shiftCount; |
2621 | cbcef455 | Peter Maydell | z = aSig; |
2622 | cbcef455 | Peter Maydell | if ( aSign ) {
|
2623 | cbcef455 | Peter Maydell | z = - z; |
2624 | cbcef455 | Peter Maydell | } |
2625 | cbcef455 | Peter Maydell | if ( ( (int16_t)z < 0 ) ^ aSign ) { |
2626 | cbcef455 | Peter Maydell | invalid:
|
2627 | cbcef455 | Peter Maydell | float_raise( float_flag_invalid STATUS_VAR); |
2628 | bb98fe42 | Andreas Färber | return aSign ? (int32_t) 0xffff8000 : 0x7FFF; |
2629 | cbcef455 | Peter Maydell | } |
2630 | cbcef455 | Peter Maydell | if ( ( aSig<<shiftCount ) != savedASig ) {
|
2631 | cbcef455 | Peter Maydell | STATUS(float_exception_flags) |= float_flag_inexact; |
2632 | cbcef455 | Peter Maydell | } |
2633 | cbcef455 | Peter Maydell | return z;
|
2634 | cbcef455 | Peter Maydell | } |
2635 | cbcef455 | Peter Maydell | |
2636 | cbcef455 | Peter Maydell | /*----------------------------------------------------------------------------
|
2637 | cbcef455 | Peter Maydell | | Returns the result of converting the double-precision floating-point value
|
2638 | 158142c2 | bellard | | `a' to the 64-bit two's complement integer format. The conversion is
|
2639 | 158142c2 | bellard | | performed according to the IEC/IEEE Standard for Binary Floating-Point
|
2640 | 158142c2 | bellard | | Arithmetic---which means in particular that the conversion is rounded
|
2641 | 158142c2 | bellard | | according to the current rounding mode. If `a' is a NaN, the largest
|
2642 | 158142c2 | bellard | | positive integer is returned. Otherwise, if the conversion overflows, the
|
2643 | 158142c2 | bellard | | largest integer with the same sign as `a' is returned.
|
2644 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
2645 | 158142c2 | bellard | |
2646 | 158142c2 | bellard | int64 float64_to_int64( float64 a STATUS_PARAM ) |
2647 | 158142c2 | bellard | { |
2648 | 158142c2 | bellard | flag aSign; |
2649 | 158142c2 | bellard | int16 aExp, shiftCount; |
2650 | bb98fe42 | Andreas Färber | uint64_t aSig, aSigExtra; |
2651 | 37d18660 | Peter Maydell | a = float64_squash_input_denormal(a STATUS_VAR); |
2652 | 158142c2 | bellard | |
2653 | 158142c2 | bellard | aSig = extractFloat64Frac( a ); |
2654 | 158142c2 | bellard | aExp = extractFloat64Exp( a ); |
2655 | 158142c2 | bellard | aSign = extractFloat64Sign( a ); |
2656 | 158142c2 | bellard | if ( aExp ) aSig |= LIT64( 0x0010000000000000 ); |
2657 | 158142c2 | bellard | shiftCount = 0x433 - aExp;
|
2658 | 158142c2 | bellard | if ( shiftCount <= 0 ) { |
2659 | 158142c2 | bellard | if ( 0x43E < aExp ) { |
2660 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
2661 | 158142c2 | bellard | if ( ! aSign
|
2662 | 158142c2 | bellard | || ( ( aExp == 0x7FF )
|
2663 | 158142c2 | bellard | && ( aSig != LIT64( 0x0010000000000000 ) ) )
|
2664 | 158142c2 | bellard | ) { |
2665 | 158142c2 | bellard | return LIT64( 0x7FFFFFFFFFFFFFFF ); |
2666 | 158142c2 | bellard | } |
2667 | bb98fe42 | Andreas Färber | return (int64_t) LIT64( 0x8000000000000000 ); |
2668 | 158142c2 | bellard | } |
2669 | 158142c2 | bellard | aSigExtra = 0;
|
2670 | 158142c2 | bellard | aSig <<= - shiftCount; |
2671 | 158142c2 | bellard | } |
2672 | 158142c2 | bellard | else {
|
2673 | 158142c2 | bellard | shift64ExtraRightJamming( aSig, 0, shiftCount, &aSig, &aSigExtra );
|
2674 | 158142c2 | bellard | } |
2675 | 158142c2 | bellard | return roundAndPackInt64( aSign, aSig, aSigExtra STATUS_VAR );
|
2676 | 158142c2 | bellard | |
2677 | 158142c2 | bellard | } |
2678 | 158142c2 | bellard | |
2679 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
2680 | 158142c2 | bellard | | Returns the result of converting the double-precision floating-point value
|
2681 | 158142c2 | bellard | | `a' to the 64-bit two's complement integer format. The conversion is
|
2682 | 158142c2 | bellard | | performed according to the IEC/IEEE Standard for Binary Floating-Point
|
2683 | 158142c2 | bellard | | Arithmetic, except that the conversion is always rounded toward zero.
|
2684 | 158142c2 | bellard | | If `a' is a NaN, the largest positive integer is returned. Otherwise, if
|
2685 | 158142c2 | bellard | | the conversion overflows, the largest integer with the same sign as `a' is
|
2686 | 158142c2 | bellard | | returned.
|
2687 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
2688 | 158142c2 | bellard | |
2689 | 158142c2 | bellard | int64 float64_to_int64_round_to_zero( float64 a STATUS_PARAM ) |
2690 | 158142c2 | bellard | { |
2691 | 158142c2 | bellard | flag aSign; |
2692 | 158142c2 | bellard | int16 aExp, shiftCount; |
2693 | bb98fe42 | Andreas Färber | uint64_t aSig; |
2694 | 158142c2 | bellard | int64 z; |
2695 | 37d18660 | Peter Maydell | a = float64_squash_input_denormal(a STATUS_VAR); |
2696 | 158142c2 | bellard | |
2697 | 158142c2 | bellard | aSig = extractFloat64Frac( a ); |
2698 | 158142c2 | bellard | aExp = extractFloat64Exp( a ); |
2699 | 158142c2 | bellard | aSign = extractFloat64Sign( a ); |
2700 | 158142c2 | bellard | if ( aExp ) aSig |= LIT64( 0x0010000000000000 ); |
2701 | 158142c2 | bellard | shiftCount = aExp - 0x433;
|
2702 | 158142c2 | bellard | if ( 0 <= shiftCount ) { |
2703 | 158142c2 | bellard | if ( 0x43E <= aExp ) { |
2704 | f090c9d4 | pbrook | if ( float64_val(a) != LIT64( 0xC3E0000000000000 ) ) { |
2705 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
2706 | 158142c2 | bellard | if ( ! aSign
|
2707 | 158142c2 | bellard | || ( ( aExp == 0x7FF )
|
2708 | 158142c2 | bellard | && ( aSig != LIT64( 0x0010000000000000 ) ) )
|
2709 | 158142c2 | bellard | ) { |
2710 | 158142c2 | bellard | return LIT64( 0x7FFFFFFFFFFFFFFF ); |
2711 | 158142c2 | bellard | } |
2712 | 158142c2 | bellard | } |
2713 | bb98fe42 | Andreas Färber | return (int64_t) LIT64( 0x8000000000000000 ); |
2714 | 158142c2 | bellard | } |
2715 | 158142c2 | bellard | z = aSig<<shiftCount; |
2716 | 158142c2 | bellard | } |
2717 | 158142c2 | bellard | else {
|
2718 | 158142c2 | bellard | if ( aExp < 0x3FE ) { |
2719 | 158142c2 | bellard | if ( aExp | aSig ) STATUS(float_exception_flags) |= float_flag_inexact;
|
2720 | 158142c2 | bellard | return 0; |
2721 | 158142c2 | bellard | } |
2722 | 158142c2 | bellard | z = aSig>>( - shiftCount ); |
2723 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( aSig<<( shiftCount & 63 ) ) ) { |
2724 | 158142c2 | bellard | STATUS(float_exception_flags) |= float_flag_inexact; |
2725 | 158142c2 | bellard | } |
2726 | 158142c2 | bellard | } |
2727 | 158142c2 | bellard | if ( aSign ) z = - z;
|
2728 | 158142c2 | bellard | return z;
|
2729 | 158142c2 | bellard | |
2730 | 158142c2 | bellard | } |
2731 | 158142c2 | bellard | |
2732 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
2733 | 158142c2 | bellard | | Returns the result of converting the double-precision floating-point value
|
2734 | 158142c2 | bellard | | `a' to the single-precision floating-point format. The conversion is
|
2735 | 158142c2 | bellard | | performed according to the IEC/IEEE Standard for Binary Floating-Point
|
2736 | 158142c2 | bellard | | Arithmetic.
|
2737 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
2738 | 158142c2 | bellard | |
2739 | 158142c2 | bellard | float32 float64_to_float32( float64 a STATUS_PARAM ) |
2740 | 158142c2 | bellard | { |
2741 | 158142c2 | bellard | flag aSign; |
2742 | 158142c2 | bellard | int16 aExp; |
2743 | bb98fe42 | Andreas Färber | uint64_t aSig; |
2744 | bb98fe42 | Andreas Färber | uint32_t zSig; |
2745 | 37d18660 | Peter Maydell | a = float64_squash_input_denormal(a STATUS_VAR); |
2746 | 158142c2 | bellard | |
2747 | 158142c2 | bellard | aSig = extractFloat64Frac( a ); |
2748 | 158142c2 | bellard | aExp = extractFloat64Exp( a ); |
2749 | 158142c2 | bellard | aSign = extractFloat64Sign( a ); |
2750 | 158142c2 | bellard | if ( aExp == 0x7FF ) { |
2751 | bcd4d9af | Christophe Lyon | if ( aSig ) return commonNaNToFloat32( float64ToCommonNaN( a STATUS_VAR ) STATUS_VAR ); |
2752 | 158142c2 | bellard | return packFloat32( aSign, 0xFF, 0 ); |
2753 | 158142c2 | bellard | } |
2754 | 158142c2 | bellard | shift64RightJamming( aSig, 22, &aSig );
|
2755 | 158142c2 | bellard | zSig = aSig; |
2756 | 158142c2 | bellard | if ( aExp || zSig ) {
|
2757 | 158142c2 | bellard | zSig |= 0x40000000;
|
2758 | 158142c2 | bellard | aExp -= 0x381;
|
2759 | 158142c2 | bellard | } |
2760 | 158142c2 | bellard | return roundAndPackFloat32( aSign, aExp, zSig STATUS_VAR );
|
2761 | 158142c2 | bellard | |
2762 | 158142c2 | bellard | } |
2763 | 158142c2 | bellard | |
2764 | 60011498 | Paul Brook | |
2765 | 60011498 | Paul Brook | /*----------------------------------------------------------------------------
|
2766 | 60011498 | Paul Brook | | Packs the sign `zSign', exponent `zExp', and significand `zSig' into a
|
2767 | 60011498 | Paul Brook | | half-precision floating-point value, returning the result. After being
|
2768 | 60011498 | Paul Brook | | shifted into the proper positions, the three fields are simply added
|
2769 | 60011498 | Paul Brook | | together to form the result. This means that any integer portion of `zSig'
|
2770 | 60011498 | Paul Brook | | will be added into the exponent. Since a properly normalized significand
|
2771 | 60011498 | Paul Brook | | will have an integer portion equal to 1, the `zExp' input should be 1 less
|
2772 | 60011498 | Paul Brook | | than the desired result exponent whenever `zSig' is a complete, normalized
|
2773 | 60011498 | Paul Brook | | significand.
|
2774 | 60011498 | Paul Brook | *----------------------------------------------------------------------------*/
|
2775 | bb98fe42 | Andreas Färber | static float16 packFloat16(flag zSign, int16 zExp, uint16_t zSig)
|
2776 | 60011498 | Paul Brook | { |
2777 | bb4d4bb3 | Peter Maydell | return make_float16(
|
2778 | bb98fe42 | Andreas Färber | (((uint32_t)zSign) << 15) + (((uint32_t)zExp) << 10) + zSig); |
2779 | 60011498 | Paul Brook | } |
2780 | 60011498 | Paul Brook | |
2781 | 60011498 | Paul Brook | /* Half precision floats come in two formats: standard IEEE and "ARM" format.
|
2782 | 60011498 | Paul Brook | The latter gains extra exponent range by omitting the NaN/Inf encodings. */
|
2783 | bb4d4bb3 | Peter Maydell | |
2784 | bb4d4bb3 | Peter Maydell | float32 float16_to_float32(float16 a, flag ieee STATUS_PARAM) |
2785 | 60011498 | Paul Brook | { |
2786 | 60011498 | Paul Brook | flag aSign; |
2787 | 60011498 | Paul Brook | int16 aExp; |
2788 | bb98fe42 | Andreas Färber | uint32_t aSig; |
2789 | 60011498 | Paul Brook | |
2790 | bb4d4bb3 | Peter Maydell | aSign = extractFloat16Sign(a); |
2791 | bb4d4bb3 | Peter Maydell | aExp = extractFloat16Exp(a); |
2792 | bb4d4bb3 | Peter Maydell | aSig = extractFloat16Frac(a); |
2793 | 60011498 | Paul Brook | |
2794 | 60011498 | Paul Brook | if (aExp == 0x1f && ieee) { |
2795 | 60011498 | Paul Brook | if (aSig) {
|
2796 | f591e1be | Peter Maydell | return commonNaNToFloat32(float16ToCommonNaN(a STATUS_VAR) STATUS_VAR);
|
2797 | 60011498 | Paul Brook | } |
2798 | 60011498 | Paul Brook | return packFloat32(aSign, 0xff, aSig << 13); |
2799 | 60011498 | Paul Brook | } |
2800 | 60011498 | Paul Brook | if (aExp == 0) { |
2801 | 60011498 | Paul Brook | int8 shiftCount; |
2802 | 60011498 | Paul Brook | |
2803 | 60011498 | Paul Brook | if (aSig == 0) { |
2804 | 60011498 | Paul Brook | return packFloat32(aSign, 0, 0); |
2805 | 60011498 | Paul Brook | } |
2806 | 60011498 | Paul Brook | |
2807 | 60011498 | Paul Brook | shiftCount = countLeadingZeros32( aSig ) - 21;
|
2808 | 60011498 | Paul Brook | aSig = aSig << shiftCount; |
2809 | 60011498 | Paul Brook | aExp = -shiftCount; |
2810 | 60011498 | Paul Brook | } |
2811 | 60011498 | Paul Brook | return packFloat32( aSign, aExp + 0x70, aSig << 13); |
2812 | 60011498 | Paul Brook | } |
2813 | 60011498 | Paul Brook | |
2814 | bb4d4bb3 | Peter Maydell | float16 float32_to_float16(float32 a, flag ieee STATUS_PARAM) |
2815 | 60011498 | Paul Brook | { |
2816 | 60011498 | Paul Brook | flag aSign; |
2817 | 60011498 | Paul Brook | int16 aExp; |
2818 | bb98fe42 | Andreas Färber | uint32_t aSig; |
2819 | bb98fe42 | Andreas Färber | uint32_t mask; |
2820 | bb98fe42 | Andreas Färber | uint32_t increment; |
2821 | 60011498 | Paul Brook | int8 roundingMode; |
2822 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
2823 | 60011498 | Paul Brook | |
2824 | 60011498 | Paul Brook | aSig = extractFloat32Frac( a ); |
2825 | 60011498 | Paul Brook | aExp = extractFloat32Exp( a ); |
2826 | 60011498 | Paul Brook | aSign = extractFloat32Sign( a ); |
2827 | 60011498 | Paul Brook | if ( aExp == 0xFF ) { |
2828 | 60011498 | Paul Brook | if (aSig) {
|
2829 | 600e30d2 | Peter Maydell | /* Input is a NaN */
|
2830 | 600e30d2 | Peter Maydell | float16 r = commonNaNToFloat16( float32ToCommonNaN( a STATUS_VAR ) STATUS_VAR ); |
2831 | 600e30d2 | Peter Maydell | if (!ieee) {
|
2832 | 600e30d2 | Peter Maydell | return packFloat16(aSign, 0, 0); |
2833 | 600e30d2 | Peter Maydell | } |
2834 | 600e30d2 | Peter Maydell | return r;
|
2835 | 60011498 | Paul Brook | } |
2836 | 600e30d2 | Peter Maydell | /* Infinity */
|
2837 | 600e30d2 | Peter Maydell | if (!ieee) {
|
2838 | 600e30d2 | Peter Maydell | float_raise(float_flag_invalid STATUS_VAR); |
2839 | 600e30d2 | Peter Maydell | return packFloat16(aSign, 0x1f, 0x3ff); |
2840 | 600e30d2 | Peter Maydell | } |
2841 | 600e30d2 | Peter Maydell | return packFloat16(aSign, 0x1f, 0); |
2842 | 60011498 | Paul Brook | } |
2843 | 600e30d2 | Peter Maydell | if (aExp == 0 && aSig == 0) { |
2844 | 60011498 | Paul Brook | return packFloat16(aSign, 0, 0); |
2845 | 60011498 | Paul Brook | } |
2846 | 60011498 | Paul Brook | /* Decimal point between bits 22 and 23. */
|
2847 | 60011498 | Paul Brook | aSig |= 0x00800000;
|
2848 | 60011498 | Paul Brook | aExp -= 0x7f;
|
2849 | 60011498 | Paul Brook | if (aExp < -14) { |
2850 | 600e30d2 | Peter Maydell | mask = 0x00ffffff;
|
2851 | 600e30d2 | Peter Maydell | if (aExp >= -24) { |
2852 | 600e30d2 | Peter Maydell | mask >>= 25 + aExp;
|
2853 | 60011498 | Paul Brook | } |
2854 | 60011498 | Paul Brook | } else {
|
2855 | 60011498 | Paul Brook | mask = 0x00001fff;
|
2856 | 60011498 | Paul Brook | } |
2857 | 60011498 | Paul Brook | if (aSig & mask) {
|
2858 | 60011498 | Paul Brook | float_raise( float_flag_underflow STATUS_VAR ); |
2859 | 60011498 | Paul Brook | roundingMode = STATUS(float_rounding_mode); |
2860 | 60011498 | Paul Brook | switch (roundingMode) {
|
2861 | 60011498 | Paul Brook | case float_round_nearest_even:
|
2862 | 60011498 | Paul Brook | increment = (mask + 1) >> 1; |
2863 | 60011498 | Paul Brook | if ((aSig & mask) == increment) {
|
2864 | 60011498 | Paul Brook | increment = aSig & (increment << 1);
|
2865 | 60011498 | Paul Brook | } |
2866 | 60011498 | Paul Brook | break;
|
2867 | 60011498 | Paul Brook | case float_round_up:
|
2868 | 60011498 | Paul Brook | increment = aSign ? 0 : mask;
|
2869 | 60011498 | Paul Brook | break;
|
2870 | 60011498 | Paul Brook | case float_round_down:
|
2871 | 60011498 | Paul Brook | increment = aSign ? mask : 0;
|
2872 | 60011498 | Paul Brook | break;
|
2873 | 60011498 | Paul Brook | default: /* round_to_zero */ |
2874 | 60011498 | Paul Brook | increment = 0;
|
2875 | 60011498 | Paul Brook | break;
|
2876 | 60011498 | Paul Brook | } |
2877 | 60011498 | Paul Brook | aSig += increment; |
2878 | 60011498 | Paul Brook | if (aSig >= 0x01000000) { |
2879 | 60011498 | Paul Brook | aSig >>= 1;
|
2880 | 60011498 | Paul Brook | aExp++; |
2881 | 60011498 | Paul Brook | } |
2882 | 60011498 | Paul Brook | } else if (aExp < -14 |
2883 | 60011498 | Paul Brook | && STATUS(float_detect_tininess) == float_tininess_before_rounding) { |
2884 | 60011498 | Paul Brook | float_raise( float_flag_underflow STATUS_VAR); |
2885 | 60011498 | Paul Brook | } |
2886 | 60011498 | Paul Brook | |
2887 | 60011498 | Paul Brook | if (ieee) {
|
2888 | 60011498 | Paul Brook | if (aExp > 15) { |
2889 | 60011498 | Paul Brook | float_raise( float_flag_overflow | float_flag_inexact STATUS_VAR); |
2890 | 60011498 | Paul Brook | return packFloat16(aSign, 0x1f, 0); |
2891 | 60011498 | Paul Brook | } |
2892 | 60011498 | Paul Brook | } else {
|
2893 | 60011498 | Paul Brook | if (aExp > 16) { |
2894 | 600e30d2 | Peter Maydell | float_raise(float_flag_invalid | float_flag_inexact STATUS_VAR); |
2895 | 60011498 | Paul Brook | return packFloat16(aSign, 0x1f, 0x3ff); |
2896 | 60011498 | Paul Brook | } |
2897 | 60011498 | Paul Brook | } |
2898 | 60011498 | Paul Brook | if (aExp < -24) { |
2899 | 60011498 | Paul Brook | return packFloat16(aSign, 0, 0); |
2900 | 60011498 | Paul Brook | } |
2901 | 60011498 | Paul Brook | if (aExp < -14) { |
2902 | 60011498 | Paul Brook | aSig >>= -14 - aExp;
|
2903 | 60011498 | Paul Brook | aExp = -14;
|
2904 | 60011498 | Paul Brook | } |
2905 | 60011498 | Paul Brook | return packFloat16(aSign, aExp + 14, aSig >> 13); |
2906 | 60011498 | Paul Brook | } |
2907 | 60011498 | Paul Brook | |
2908 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
2909 | 158142c2 | bellard | | Returns the result of converting the double-precision floating-point value
|
2910 | 158142c2 | bellard | | `a' to the extended double-precision floating-point format. The conversion
|
2911 | 158142c2 | bellard | | is performed according to the IEC/IEEE Standard for Binary Floating-Point
|
2912 | 158142c2 | bellard | | Arithmetic.
|
2913 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
2914 | 158142c2 | bellard | |
2915 | 158142c2 | bellard | floatx80 float64_to_floatx80( float64 a STATUS_PARAM ) |
2916 | 158142c2 | bellard | { |
2917 | 158142c2 | bellard | flag aSign; |
2918 | 158142c2 | bellard | int16 aExp; |
2919 | bb98fe42 | Andreas Färber | uint64_t aSig; |
2920 | 158142c2 | bellard | |
2921 | 37d18660 | Peter Maydell | a = float64_squash_input_denormal(a STATUS_VAR); |
2922 | 158142c2 | bellard | aSig = extractFloat64Frac( a ); |
2923 | 158142c2 | bellard | aExp = extractFloat64Exp( a ); |
2924 | 158142c2 | bellard | aSign = extractFloat64Sign( a ); |
2925 | 158142c2 | bellard | if ( aExp == 0x7FF ) { |
2926 | bcd4d9af | Christophe Lyon | if ( aSig ) return commonNaNToFloatx80( float64ToCommonNaN( a STATUS_VAR ) STATUS_VAR ); |
2927 | 158142c2 | bellard | return packFloatx80( aSign, 0x7FFF, LIT64( 0x8000000000000000 ) ); |
2928 | 158142c2 | bellard | } |
2929 | 158142c2 | bellard | if ( aExp == 0 ) { |
2930 | 158142c2 | bellard | if ( aSig == 0 ) return packFloatx80( aSign, 0, 0 ); |
2931 | 158142c2 | bellard | normalizeFloat64Subnormal( aSig, &aExp, &aSig ); |
2932 | 158142c2 | bellard | } |
2933 | 158142c2 | bellard | return
|
2934 | 158142c2 | bellard | packFloatx80( |
2935 | 158142c2 | bellard | aSign, aExp + 0x3C00, ( aSig | LIT64( 0x0010000000000000 ) )<<11 ); |
2936 | 158142c2 | bellard | |
2937 | 158142c2 | bellard | } |
2938 | 158142c2 | bellard | |
2939 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
2940 | 158142c2 | bellard | | Returns the result of converting the double-precision floating-point value
|
2941 | 158142c2 | bellard | | `a' to the quadruple-precision floating-point format. The conversion is
|
2942 | 158142c2 | bellard | | performed according to the IEC/IEEE Standard for Binary Floating-Point
|
2943 | 158142c2 | bellard | | Arithmetic.
|
2944 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
2945 | 158142c2 | bellard | |
2946 | 158142c2 | bellard | float128 float64_to_float128( float64 a STATUS_PARAM ) |
2947 | 158142c2 | bellard | { |
2948 | 158142c2 | bellard | flag aSign; |
2949 | 158142c2 | bellard | int16 aExp; |
2950 | bb98fe42 | Andreas Färber | uint64_t aSig, zSig0, zSig1; |
2951 | 158142c2 | bellard | |
2952 | 37d18660 | Peter Maydell | a = float64_squash_input_denormal(a STATUS_VAR); |
2953 | 158142c2 | bellard | aSig = extractFloat64Frac( a ); |
2954 | 158142c2 | bellard | aExp = extractFloat64Exp( a ); |
2955 | 158142c2 | bellard | aSign = extractFloat64Sign( a ); |
2956 | 158142c2 | bellard | if ( aExp == 0x7FF ) { |
2957 | bcd4d9af | Christophe Lyon | if ( aSig ) return commonNaNToFloat128( float64ToCommonNaN( a STATUS_VAR ) STATUS_VAR ); |
2958 | 158142c2 | bellard | return packFloat128( aSign, 0x7FFF, 0, 0 ); |
2959 | 158142c2 | bellard | } |
2960 | 158142c2 | bellard | if ( aExp == 0 ) { |
2961 | 158142c2 | bellard | if ( aSig == 0 ) return packFloat128( aSign, 0, 0, 0 ); |
2962 | 158142c2 | bellard | normalizeFloat64Subnormal( aSig, &aExp, &aSig ); |
2963 | 158142c2 | bellard | --aExp; |
2964 | 158142c2 | bellard | } |
2965 | 158142c2 | bellard | shift128Right( aSig, 0, 4, &zSig0, &zSig1 ); |
2966 | 158142c2 | bellard | return packFloat128( aSign, aExp + 0x3C00, zSig0, zSig1 ); |
2967 | 158142c2 | bellard | |
2968 | 158142c2 | bellard | } |
2969 | 158142c2 | bellard | |
2970 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
2971 | 158142c2 | bellard | | Rounds the double-precision floating-point value `a' to an integer, and
|
2972 | 158142c2 | bellard | | returns the result as a double-precision floating-point value. The
|
2973 | 158142c2 | bellard | | operation is performed according to the IEC/IEEE Standard for Binary
|
2974 | 158142c2 | bellard | | Floating-Point Arithmetic.
|
2975 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
2976 | 158142c2 | bellard | |
2977 | 158142c2 | bellard | float64 float64_round_to_int( float64 a STATUS_PARAM ) |
2978 | 158142c2 | bellard | { |
2979 | 158142c2 | bellard | flag aSign; |
2980 | 158142c2 | bellard | int16 aExp; |
2981 | bb98fe42 | Andreas Färber | uint64_t lastBitMask, roundBitsMask; |
2982 | 158142c2 | bellard | int8 roundingMode; |
2983 | bb98fe42 | Andreas Färber | uint64_t z; |
2984 | 37d18660 | Peter Maydell | a = float64_squash_input_denormal(a STATUS_VAR); |
2985 | 158142c2 | bellard | |
2986 | 158142c2 | bellard | aExp = extractFloat64Exp( a ); |
2987 | 158142c2 | bellard | if ( 0x433 <= aExp ) { |
2988 | 158142c2 | bellard | if ( ( aExp == 0x7FF ) && extractFloat64Frac( a ) ) { |
2989 | 158142c2 | bellard | return propagateFloat64NaN( a, a STATUS_VAR );
|
2990 | 158142c2 | bellard | } |
2991 | 158142c2 | bellard | return a;
|
2992 | 158142c2 | bellard | } |
2993 | 158142c2 | bellard | if ( aExp < 0x3FF ) { |
2994 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( float64_val(a)<<1 ) == 0 ) return a; |
2995 | 158142c2 | bellard | STATUS(float_exception_flags) |= float_flag_inexact; |
2996 | 158142c2 | bellard | aSign = extractFloat64Sign( a ); |
2997 | 158142c2 | bellard | switch ( STATUS(float_rounding_mode) ) {
|
2998 | 158142c2 | bellard | case float_round_nearest_even:
|
2999 | 158142c2 | bellard | if ( ( aExp == 0x3FE ) && extractFloat64Frac( a ) ) { |
3000 | 158142c2 | bellard | return packFloat64( aSign, 0x3FF, 0 ); |
3001 | 158142c2 | bellard | } |
3002 | 158142c2 | bellard | break;
|
3003 | 158142c2 | bellard | case float_round_down:
|
3004 | f090c9d4 | pbrook | return make_float64(aSign ? LIT64( 0xBFF0000000000000 ) : 0); |
3005 | 158142c2 | bellard | case float_round_up:
|
3006 | f090c9d4 | pbrook | return make_float64(
|
3007 | f090c9d4 | pbrook | aSign ? LIT64( 0x8000000000000000 ) : LIT64( 0x3FF0000000000000 )); |
3008 | 158142c2 | bellard | } |
3009 | 158142c2 | bellard | return packFloat64( aSign, 0, 0 ); |
3010 | 158142c2 | bellard | } |
3011 | 158142c2 | bellard | lastBitMask = 1;
|
3012 | 158142c2 | bellard | lastBitMask <<= 0x433 - aExp;
|
3013 | 158142c2 | bellard | roundBitsMask = lastBitMask - 1;
|
3014 | f090c9d4 | pbrook | z = float64_val(a); |
3015 | 158142c2 | bellard | roundingMode = STATUS(float_rounding_mode); |
3016 | 158142c2 | bellard | if ( roundingMode == float_round_nearest_even ) {
|
3017 | 158142c2 | bellard | z += lastBitMask>>1;
|
3018 | 158142c2 | bellard | if ( ( z & roundBitsMask ) == 0 ) z &= ~ lastBitMask; |
3019 | 158142c2 | bellard | } |
3020 | 158142c2 | bellard | else if ( roundingMode != float_round_to_zero ) { |
3021 | f090c9d4 | pbrook | if ( extractFloat64Sign( make_float64(z) ) ^ ( roundingMode == float_round_up ) ) {
|
3022 | 158142c2 | bellard | z += roundBitsMask; |
3023 | 158142c2 | bellard | } |
3024 | 158142c2 | bellard | } |
3025 | 158142c2 | bellard | z &= ~ roundBitsMask; |
3026 | f090c9d4 | pbrook | if ( z != float64_val(a) )
|
3027 | f090c9d4 | pbrook | STATUS(float_exception_flags) |= float_flag_inexact; |
3028 | f090c9d4 | pbrook | return make_float64(z);
|
3029 | 158142c2 | bellard | |
3030 | 158142c2 | bellard | } |
3031 | 158142c2 | bellard | |
3032 | e6e5906b | pbrook | float64 float64_trunc_to_int( float64 a STATUS_PARAM) |
3033 | e6e5906b | pbrook | { |
3034 | e6e5906b | pbrook | int oldmode;
|
3035 | e6e5906b | pbrook | float64 res; |
3036 | e6e5906b | pbrook | oldmode = STATUS(float_rounding_mode); |
3037 | e6e5906b | pbrook | STATUS(float_rounding_mode) = float_round_to_zero; |
3038 | e6e5906b | pbrook | res = float64_round_to_int(a STATUS_VAR); |
3039 | e6e5906b | pbrook | STATUS(float_rounding_mode) = oldmode; |
3040 | e6e5906b | pbrook | return res;
|
3041 | e6e5906b | pbrook | } |
3042 | e6e5906b | pbrook | |
3043 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
3044 | 158142c2 | bellard | | Returns the result of adding the absolute values of the double-precision
|
3045 | 158142c2 | bellard | | floating-point values `a' and `b'. If `zSign' is 1, the sum is negated
|
3046 | 158142c2 | bellard | | before being returned. `zSign' is ignored if the result is a NaN.
|
3047 | 158142c2 | bellard | | The addition is performed according to the IEC/IEEE Standard for Binary
|
3048 | 158142c2 | bellard | | Floating-Point Arithmetic.
|
3049 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
3050 | 158142c2 | bellard | |
3051 | 158142c2 | bellard | static float64 addFloat64Sigs( float64 a, float64 b, flag zSign STATUS_PARAM )
|
3052 | 158142c2 | bellard | { |
3053 | 158142c2 | bellard | int16 aExp, bExp, zExp; |
3054 | bb98fe42 | Andreas Färber | uint64_t aSig, bSig, zSig; |
3055 | 158142c2 | bellard | int16 expDiff; |
3056 | 158142c2 | bellard | |
3057 | 158142c2 | bellard | aSig = extractFloat64Frac( a ); |
3058 | 158142c2 | bellard | aExp = extractFloat64Exp( a ); |
3059 | 158142c2 | bellard | bSig = extractFloat64Frac( b ); |
3060 | 158142c2 | bellard | bExp = extractFloat64Exp( b ); |
3061 | 158142c2 | bellard | expDiff = aExp - bExp; |
3062 | 158142c2 | bellard | aSig <<= 9;
|
3063 | 158142c2 | bellard | bSig <<= 9;
|
3064 | 158142c2 | bellard | if ( 0 < expDiff ) { |
3065 | 158142c2 | bellard | if ( aExp == 0x7FF ) { |
3066 | 158142c2 | bellard | if ( aSig ) return propagateFloat64NaN( a, b STATUS_VAR ); |
3067 | 158142c2 | bellard | return a;
|
3068 | 158142c2 | bellard | } |
3069 | 158142c2 | bellard | if ( bExp == 0 ) { |
3070 | 158142c2 | bellard | --expDiff; |
3071 | 158142c2 | bellard | } |
3072 | 158142c2 | bellard | else {
|
3073 | 158142c2 | bellard | bSig |= LIT64( 0x2000000000000000 );
|
3074 | 158142c2 | bellard | } |
3075 | 158142c2 | bellard | shift64RightJamming( bSig, expDiff, &bSig ); |
3076 | 158142c2 | bellard | zExp = aExp; |
3077 | 158142c2 | bellard | } |
3078 | 158142c2 | bellard | else if ( expDiff < 0 ) { |
3079 | 158142c2 | bellard | if ( bExp == 0x7FF ) { |
3080 | 158142c2 | bellard | if ( bSig ) return propagateFloat64NaN( a, b STATUS_VAR ); |
3081 | 158142c2 | bellard | return packFloat64( zSign, 0x7FF, 0 ); |
3082 | 158142c2 | bellard | } |
3083 | 158142c2 | bellard | if ( aExp == 0 ) { |
3084 | 158142c2 | bellard | ++expDiff; |
3085 | 158142c2 | bellard | } |
3086 | 158142c2 | bellard | else {
|
3087 | 158142c2 | bellard | aSig |= LIT64( 0x2000000000000000 );
|
3088 | 158142c2 | bellard | } |
3089 | 158142c2 | bellard | shift64RightJamming( aSig, - expDiff, &aSig ); |
3090 | 158142c2 | bellard | zExp = bExp; |
3091 | 158142c2 | bellard | } |
3092 | 158142c2 | bellard | else {
|
3093 | 158142c2 | bellard | if ( aExp == 0x7FF ) { |
3094 | 158142c2 | bellard | if ( aSig | bSig ) return propagateFloat64NaN( a, b STATUS_VAR ); |
3095 | 158142c2 | bellard | return a;
|
3096 | 158142c2 | bellard | } |
3097 | fe76d976 | pbrook | if ( aExp == 0 ) { |
3098 | e6afc87f | Peter Maydell | if (STATUS(flush_to_zero)) {
|
3099 | e6afc87f | Peter Maydell | if (aSig | bSig) {
|
3100 | e6afc87f | Peter Maydell | float_raise(float_flag_output_denormal STATUS_VAR); |
3101 | e6afc87f | Peter Maydell | } |
3102 | e6afc87f | Peter Maydell | return packFloat64(zSign, 0, 0); |
3103 | e6afc87f | Peter Maydell | } |
3104 | fe76d976 | pbrook | return packFloat64( zSign, 0, ( aSig + bSig )>>9 ); |
3105 | fe76d976 | pbrook | } |
3106 | 158142c2 | bellard | zSig = LIT64( 0x4000000000000000 ) + aSig + bSig;
|
3107 | 158142c2 | bellard | zExp = aExp; |
3108 | 158142c2 | bellard | goto roundAndPack;
|
3109 | 158142c2 | bellard | } |
3110 | 158142c2 | bellard | aSig |= LIT64( 0x2000000000000000 );
|
3111 | 158142c2 | bellard | zSig = ( aSig + bSig )<<1;
|
3112 | 158142c2 | bellard | --zExp; |
3113 | bb98fe42 | Andreas Färber | if ( (int64_t) zSig < 0 ) { |
3114 | 158142c2 | bellard | zSig = aSig + bSig; |
3115 | 158142c2 | bellard | ++zExp; |
3116 | 158142c2 | bellard | } |
3117 | 158142c2 | bellard | roundAndPack:
|
3118 | 158142c2 | bellard | return roundAndPackFloat64( zSign, zExp, zSig STATUS_VAR );
|
3119 | 158142c2 | bellard | |
3120 | 158142c2 | bellard | } |
3121 | 158142c2 | bellard | |
3122 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
3123 | 158142c2 | bellard | | Returns the result of subtracting the absolute values of the double-
|
3124 | 158142c2 | bellard | | precision floating-point values `a' and `b'. If `zSign' is 1, the
|
3125 | 158142c2 | bellard | | difference is negated before being returned. `zSign' is ignored if the
|
3126 | 158142c2 | bellard | | result is a NaN. The subtraction is performed according to the IEC/IEEE
|
3127 | 158142c2 | bellard | | Standard for Binary Floating-Point Arithmetic.
|
3128 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
3129 | 158142c2 | bellard | |
3130 | 158142c2 | bellard | static float64 subFloat64Sigs( float64 a, float64 b, flag zSign STATUS_PARAM )
|
3131 | 158142c2 | bellard | { |
3132 | 158142c2 | bellard | int16 aExp, bExp, zExp; |
3133 | bb98fe42 | Andreas Färber | uint64_t aSig, bSig, zSig; |
3134 | 158142c2 | bellard | int16 expDiff; |
3135 | 158142c2 | bellard | |
3136 | 158142c2 | bellard | aSig = extractFloat64Frac( a ); |
3137 | 158142c2 | bellard | aExp = extractFloat64Exp( a ); |
3138 | 158142c2 | bellard | bSig = extractFloat64Frac( b ); |
3139 | 158142c2 | bellard | bExp = extractFloat64Exp( b ); |
3140 | 158142c2 | bellard | expDiff = aExp - bExp; |
3141 | 158142c2 | bellard | aSig <<= 10;
|
3142 | 158142c2 | bellard | bSig <<= 10;
|
3143 | 158142c2 | bellard | if ( 0 < expDiff ) goto aExpBigger; |
3144 | 158142c2 | bellard | if ( expDiff < 0 ) goto bExpBigger; |
3145 | 158142c2 | bellard | if ( aExp == 0x7FF ) { |
3146 | 158142c2 | bellard | if ( aSig | bSig ) return propagateFloat64NaN( a, b STATUS_VAR ); |
3147 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
3148 | 158142c2 | bellard | return float64_default_nan;
|
3149 | 158142c2 | bellard | } |
3150 | 158142c2 | bellard | if ( aExp == 0 ) { |
3151 | 158142c2 | bellard | aExp = 1;
|
3152 | 158142c2 | bellard | bExp = 1;
|
3153 | 158142c2 | bellard | } |
3154 | 158142c2 | bellard | if ( bSig < aSig ) goto aBigger; |
3155 | 158142c2 | bellard | if ( aSig < bSig ) goto bBigger; |
3156 | 158142c2 | bellard | return packFloat64( STATUS(float_rounding_mode) == float_round_down, 0, 0 ); |
3157 | 158142c2 | bellard | bExpBigger:
|
3158 | 158142c2 | bellard | if ( bExp == 0x7FF ) { |
3159 | 158142c2 | bellard | if ( bSig ) return propagateFloat64NaN( a, b STATUS_VAR ); |
3160 | 158142c2 | bellard | return packFloat64( zSign ^ 1, 0x7FF, 0 ); |
3161 | 158142c2 | bellard | } |
3162 | 158142c2 | bellard | if ( aExp == 0 ) { |
3163 | 158142c2 | bellard | ++expDiff; |
3164 | 158142c2 | bellard | } |
3165 | 158142c2 | bellard | else {
|
3166 | 158142c2 | bellard | aSig |= LIT64( 0x4000000000000000 );
|
3167 | 158142c2 | bellard | } |
3168 | 158142c2 | bellard | shift64RightJamming( aSig, - expDiff, &aSig ); |
3169 | 158142c2 | bellard | bSig |= LIT64( 0x4000000000000000 );
|
3170 | 158142c2 | bellard | bBigger:
|
3171 | 158142c2 | bellard | zSig = bSig - aSig; |
3172 | 158142c2 | bellard | zExp = bExp; |
3173 | 158142c2 | bellard | zSign ^= 1;
|
3174 | 158142c2 | bellard | goto normalizeRoundAndPack;
|
3175 | 158142c2 | bellard | aExpBigger:
|
3176 | 158142c2 | bellard | if ( aExp == 0x7FF ) { |
3177 | 158142c2 | bellard | if ( aSig ) return propagateFloat64NaN( a, b STATUS_VAR ); |
3178 | 158142c2 | bellard | return a;
|
3179 | 158142c2 | bellard | } |
3180 | 158142c2 | bellard | if ( bExp == 0 ) { |
3181 | 158142c2 | bellard | --expDiff; |
3182 | 158142c2 | bellard | } |
3183 | 158142c2 | bellard | else {
|
3184 | 158142c2 | bellard | bSig |= LIT64( 0x4000000000000000 );
|
3185 | 158142c2 | bellard | } |
3186 | 158142c2 | bellard | shift64RightJamming( bSig, expDiff, &bSig ); |
3187 | 158142c2 | bellard | aSig |= LIT64( 0x4000000000000000 );
|
3188 | 158142c2 | bellard | aBigger:
|
3189 | 158142c2 | bellard | zSig = aSig - bSig; |
3190 | 158142c2 | bellard | zExp = aExp; |
3191 | 158142c2 | bellard | normalizeRoundAndPack:
|
3192 | 158142c2 | bellard | --zExp; |
3193 | 158142c2 | bellard | return normalizeRoundAndPackFloat64( zSign, zExp, zSig STATUS_VAR );
|
3194 | 158142c2 | bellard | |
3195 | 158142c2 | bellard | } |
3196 | 158142c2 | bellard | |
3197 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
3198 | 158142c2 | bellard | | Returns the result of adding the double-precision floating-point values `a'
|
3199 | 158142c2 | bellard | | and `b'. The operation is performed according to the IEC/IEEE Standard for
|
3200 | 158142c2 | bellard | | Binary Floating-Point Arithmetic.
|
3201 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
3202 | 158142c2 | bellard | |
3203 | 158142c2 | bellard | float64 float64_add( float64 a, float64 b STATUS_PARAM ) |
3204 | 158142c2 | bellard | { |
3205 | 158142c2 | bellard | flag aSign, bSign; |
3206 | 37d18660 | Peter Maydell | a = float64_squash_input_denormal(a STATUS_VAR); |
3207 | 37d18660 | Peter Maydell | b = float64_squash_input_denormal(b STATUS_VAR); |
3208 | 158142c2 | bellard | |
3209 | 158142c2 | bellard | aSign = extractFloat64Sign( a ); |
3210 | 158142c2 | bellard | bSign = extractFloat64Sign( b ); |
3211 | 158142c2 | bellard | if ( aSign == bSign ) {
|
3212 | 158142c2 | bellard | return addFloat64Sigs( a, b, aSign STATUS_VAR );
|
3213 | 158142c2 | bellard | } |
3214 | 158142c2 | bellard | else {
|
3215 | 158142c2 | bellard | return subFloat64Sigs( a, b, aSign STATUS_VAR );
|
3216 | 158142c2 | bellard | } |
3217 | 158142c2 | bellard | |
3218 | 158142c2 | bellard | } |
3219 | 158142c2 | bellard | |
3220 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
3221 | 158142c2 | bellard | | Returns the result of subtracting the double-precision floating-point values
|
3222 | 158142c2 | bellard | | `a' and `b'. The operation is performed according to the IEC/IEEE Standard
|
3223 | 158142c2 | bellard | | for Binary Floating-Point Arithmetic.
|
3224 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
3225 | 158142c2 | bellard | |
3226 | 158142c2 | bellard | float64 float64_sub( float64 a, float64 b STATUS_PARAM ) |
3227 | 158142c2 | bellard | { |
3228 | 158142c2 | bellard | flag aSign, bSign; |
3229 | 37d18660 | Peter Maydell | a = float64_squash_input_denormal(a STATUS_VAR); |
3230 | 37d18660 | Peter Maydell | b = float64_squash_input_denormal(b STATUS_VAR); |
3231 | 158142c2 | bellard | |
3232 | 158142c2 | bellard | aSign = extractFloat64Sign( a ); |
3233 | 158142c2 | bellard | bSign = extractFloat64Sign( b ); |
3234 | 158142c2 | bellard | if ( aSign == bSign ) {
|
3235 | 158142c2 | bellard | return subFloat64Sigs( a, b, aSign STATUS_VAR );
|
3236 | 158142c2 | bellard | } |
3237 | 158142c2 | bellard | else {
|
3238 | 158142c2 | bellard | return addFloat64Sigs( a, b, aSign STATUS_VAR );
|
3239 | 158142c2 | bellard | } |
3240 | 158142c2 | bellard | |
3241 | 158142c2 | bellard | } |
3242 | 158142c2 | bellard | |
3243 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
3244 | 158142c2 | bellard | | Returns the result of multiplying the double-precision floating-point values
|
3245 | 158142c2 | bellard | | `a' and `b'. The operation is performed according to the IEC/IEEE Standard
|
3246 | 158142c2 | bellard | | for Binary Floating-Point Arithmetic.
|
3247 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
3248 | 158142c2 | bellard | |
3249 | 158142c2 | bellard | float64 float64_mul( float64 a, float64 b STATUS_PARAM ) |
3250 | 158142c2 | bellard | { |
3251 | 158142c2 | bellard | flag aSign, bSign, zSign; |
3252 | 158142c2 | bellard | int16 aExp, bExp, zExp; |
3253 | bb98fe42 | Andreas Färber | uint64_t aSig, bSig, zSig0, zSig1; |
3254 | 158142c2 | bellard | |
3255 | 37d18660 | Peter Maydell | a = float64_squash_input_denormal(a STATUS_VAR); |
3256 | 37d18660 | Peter Maydell | b = float64_squash_input_denormal(b STATUS_VAR); |
3257 | 37d18660 | Peter Maydell | |
3258 | 158142c2 | bellard | aSig = extractFloat64Frac( a ); |
3259 | 158142c2 | bellard | aExp = extractFloat64Exp( a ); |
3260 | 158142c2 | bellard | aSign = extractFloat64Sign( a ); |
3261 | 158142c2 | bellard | bSig = extractFloat64Frac( b ); |
3262 | 158142c2 | bellard | bExp = extractFloat64Exp( b ); |
3263 | 158142c2 | bellard | bSign = extractFloat64Sign( b ); |
3264 | 158142c2 | bellard | zSign = aSign ^ bSign; |
3265 | 158142c2 | bellard | if ( aExp == 0x7FF ) { |
3266 | 158142c2 | bellard | if ( aSig || ( ( bExp == 0x7FF ) && bSig ) ) { |
3267 | 158142c2 | bellard | return propagateFloat64NaN( a, b STATUS_VAR );
|
3268 | 158142c2 | bellard | } |
3269 | 158142c2 | bellard | if ( ( bExp | bSig ) == 0 ) { |
3270 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
3271 | 158142c2 | bellard | return float64_default_nan;
|
3272 | 158142c2 | bellard | } |
3273 | 158142c2 | bellard | return packFloat64( zSign, 0x7FF, 0 ); |
3274 | 158142c2 | bellard | } |
3275 | 158142c2 | bellard | if ( bExp == 0x7FF ) { |
3276 | 158142c2 | bellard | if ( bSig ) return propagateFloat64NaN( a, b STATUS_VAR ); |
3277 | 158142c2 | bellard | if ( ( aExp | aSig ) == 0 ) { |
3278 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
3279 | 158142c2 | bellard | return float64_default_nan;
|
3280 | 158142c2 | bellard | } |
3281 | 158142c2 | bellard | return packFloat64( zSign, 0x7FF, 0 ); |
3282 | 158142c2 | bellard | } |
3283 | 158142c2 | bellard | if ( aExp == 0 ) { |
3284 | 158142c2 | bellard | if ( aSig == 0 ) return packFloat64( zSign, 0, 0 ); |
3285 | 158142c2 | bellard | normalizeFloat64Subnormal( aSig, &aExp, &aSig ); |
3286 | 158142c2 | bellard | } |
3287 | 158142c2 | bellard | if ( bExp == 0 ) { |
3288 | 158142c2 | bellard | if ( bSig == 0 ) return packFloat64( zSign, 0, 0 ); |
3289 | 158142c2 | bellard | normalizeFloat64Subnormal( bSig, &bExp, &bSig ); |
3290 | 158142c2 | bellard | } |
3291 | 158142c2 | bellard | zExp = aExp + bExp - 0x3FF;
|
3292 | 158142c2 | bellard | aSig = ( aSig | LIT64( 0x0010000000000000 ) )<<10; |
3293 | 158142c2 | bellard | bSig = ( bSig | LIT64( 0x0010000000000000 ) )<<11; |
3294 | 158142c2 | bellard | mul64To128( aSig, bSig, &zSig0, &zSig1 ); |
3295 | 158142c2 | bellard | zSig0 |= ( zSig1 != 0 );
|
3296 | bb98fe42 | Andreas Färber | if ( 0 <= (int64_t) ( zSig0<<1 ) ) { |
3297 | 158142c2 | bellard | zSig0 <<= 1;
|
3298 | 158142c2 | bellard | --zExp; |
3299 | 158142c2 | bellard | } |
3300 | 158142c2 | bellard | return roundAndPackFloat64( zSign, zExp, zSig0 STATUS_VAR );
|
3301 | 158142c2 | bellard | |
3302 | 158142c2 | bellard | } |
3303 | 158142c2 | bellard | |
3304 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
3305 | 158142c2 | bellard | | Returns the result of dividing the double-precision floating-point value `a'
|
3306 | 158142c2 | bellard | | by the corresponding value `b'. The operation is performed according to
|
3307 | 158142c2 | bellard | | the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
3308 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
3309 | 158142c2 | bellard | |
3310 | 158142c2 | bellard | float64 float64_div( float64 a, float64 b STATUS_PARAM ) |
3311 | 158142c2 | bellard | { |
3312 | 158142c2 | bellard | flag aSign, bSign, zSign; |
3313 | 158142c2 | bellard | int16 aExp, bExp, zExp; |
3314 | bb98fe42 | Andreas Färber | uint64_t aSig, bSig, zSig; |
3315 | bb98fe42 | Andreas Färber | uint64_t rem0, rem1; |
3316 | bb98fe42 | Andreas Färber | uint64_t term0, term1; |
3317 | 37d18660 | Peter Maydell | a = float64_squash_input_denormal(a STATUS_VAR); |
3318 | 37d18660 | Peter Maydell | b = float64_squash_input_denormal(b STATUS_VAR); |
3319 | 158142c2 | bellard | |
3320 | 158142c2 | bellard | aSig = extractFloat64Frac( a ); |
3321 | 158142c2 | bellard | aExp = extractFloat64Exp( a ); |
3322 | 158142c2 | bellard | aSign = extractFloat64Sign( a ); |
3323 | 158142c2 | bellard | bSig = extractFloat64Frac( b ); |
3324 | 158142c2 | bellard | bExp = extractFloat64Exp( b ); |
3325 | 158142c2 | bellard | bSign = extractFloat64Sign( b ); |
3326 | 158142c2 | bellard | zSign = aSign ^ bSign; |
3327 | 158142c2 | bellard | if ( aExp == 0x7FF ) { |
3328 | 158142c2 | bellard | if ( aSig ) return propagateFloat64NaN( a, b STATUS_VAR ); |
3329 | 158142c2 | bellard | if ( bExp == 0x7FF ) { |
3330 | 158142c2 | bellard | if ( bSig ) return propagateFloat64NaN( a, b STATUS_VAR ); |
3331 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
3332 | 158142c2 | bellard | return float64_default_nan;
|
3333 | 158142c2 | bellard | } |
3334 | 158142c2 | bellard | return packFloat64( zSign, 0x7FF, 0 ); |
3335 | 158142c2 | bellard | } |
3336 | 158142c2 | bellard | if ( bExp == 0x7FF ) { |
3337 | 158142c2 | bellard | if ( bSig ) return propagateFloat64NaN( a, b STATUS_VAR ); |
3338 | 158142c2 | bellard | return packFloat64( zSign, 0, 0 ); |
3339 | 158142c2 | bellard | } |
3340 | 158142c2 | bellard | if ( bExp == 0 ) { |
3341 | 158142c2 | bellard | if ( bSig == 0 ) { |
3342 | 158142c2 | bellard | if ( ( aExp | aSig ) == 0 ) { |
3343 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
3344 | 158142c2 | bellard | return float64_default_nan;
|
3345 | 158142c2 | bellard | } |
3346 | 158142c2 | bellard | float_raise( float_flag_divbyzero STATUS_VAR); |
3347 | 158142c2 | bellard | return packFloat64( zSign, 0x7FF, 0 ); |
3348 | 158142c2 | bellard | } |
3349 | 158142c2 | bellard | normalizeFloat64Subnormal( bSig, &bExp, &bSig ); |
3350 | 158142c2 | bellard | } |
3351 | 158142c2 | bellard | if ( aExp == 0 ) { |
3352 | 158142c2 | bellard | if ( aSig == 0 ) return packFloat64( zSign, 0, 0 ); |
3353 | 158142c2 | bellard | normalizeFloat64Subnormal( aSig, &aExp, &aSig ); |
3354 | 158142c2 | bellard | } |
3355 | 158142c2 | bellard | zExp = aExp - bExp + 0x3FD;
|
3356 | 158142c2 | bellard | aSig = ( aSig | LIT64( 0x0010000000000000 ) )<<10; |
3357 | 158142c2 | bellard | bSig = ( bSig | LIT64( 0x0010000000000000 ) )<<11; |
3358 | 158142c2 | bellard | if ( bSig <= ( aSig + aSig ) ) {
|
3359 | 158142c2 | bellard | aSig >>= 1;
|
3360 | 158142c2 | bellard | ++zExp; |
3361 | 158142c2 | bellard | } |
3362 | 158142c2 | bellard | zSig = estimateDiv128To64( aSig, 0, bSig );
|
3363 | 158142c2 | bellard | if ( ( zSig & 0x1FF ) <= 2 ) { |
3364 | 158142c2 | bellard | mul64To128( bSig, zSig, &term0, &term1 ); |
3365 | 158142c2 | bellard | sub128( aSig, 0, term0, term1, &rem0, &rem1 );
|
3366 | bb98fe42 | Andreas Färber | while ( (int64_t) rem0 < 0 ) { |
3367 | 158142c2 | bellard | --zSig; |
3368 | 158142c2 | bellard | add128( rem0, rem1, 0, bSig, &rem0, &rem1 );
|
3369 | 158142c2 | bellard | } |
3370 | 158142c2 | bellard | zSig |= ( rem1 != 0 );
|
3371 | 158142c2 | bellard | } |
3372 | 158142c2 | bellard | return roundAndPackFloat64( zSign, zExp, zSig STATUS_VAR );
|
3373 | 158142c2 | bellard | |
3374 | 158142c2 | bellard | } |
3375 | 158142c2 | bellard | |
3376 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
3377 | 158142c2 | bellard | | Returns the remainder of the double-precision floating-point value `a'
|
3378 | 158142c2 | bellard | | with respect to the corresponding value `b'. The operation is performed
|
3379 | 158142c2 | bellard | | according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
3380 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
3381 | 158142c2 | bellard | |
3382 | 158142c2 | bellard | float64 float64_rem( float64 a, float64 b STATUS_PARAM ) |
3383 | 158142c2 | bellard | { |
3384 | ed086f3d | Blue Swirl | flag aSign, zSign; |
3385 | 158142c2 | bellard | int16 aExp, bExp, expDiff; |
3386 | bb98fe42 | Andreas Färber | uint64_t aSig, bSig; |
3387 | bb98fe42 | Andreas Färber | uint64_t q, alternateASig; |
3388 | bb98fe42 | Andreas Färber | int64_t sigMean; |
3389 | 158142c2 | bellard | |
3390 | 37d18660 | Peter Maydell | a = float64_squash_input_denormal(a STATUS_VAR); |
3391 | 37d18660 | Peter Maydell | b = float64_squash_input_denormal(b STATUS_VAR); |
3392 | 158142c2 | bellard | aSig = extractFloat64Frac( a ); |
3393 | 158142c2 | bellard | aExp = extractFloat64Exp( a ); |
3394 | 158142c2 | bellard | aSign = extractFloat64Sign( a ); |
3395 | 158142c2 | bellard | bSig = extractFloat64Frac( b ); |
3396 | 158142c2 | bellard | bExp = extractFloat64Exp( b ); |
3397 | 158142c2 | bellard | if ( aExp == 0x7FF ) { |
3398 | 158142c2 | bellard | if ( aSig || ( ( bExp == 0x7FF ) && bSig ) ) { |
3399 | 158142c2 | bellard | return propagateFloat64NaN( a, b STATUS_VAR );
|
3400 | 158142c2 | bellard | } |
3401 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
3402 | 158142c2 | bellard | return float64_default_nan;
|
3403 | 158142c2 | bellard | } |
3404 | 158142c2 | bellard | if ( bExp == 0x7FF ) { |
3405 | 158142c2 | bellard | if ( bSig ) return propagateFloat64NaN( a, b STATUS_VAR ); |
3406 | 158142c2 | bellard | return a;
|
3407 | 158142c2 | bellard | } |
3408 | 158142c2 | bellard | if ( bExp == 0 ) { |
3409 | 158142c2 | bellard | if ( bSig == 0 ) { |
3410 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
3411 | 158142c2 | bellard | return float64_default_nan;
|
3412 | 158142c2 | bellard | } |
3413 | 158142c2 | bellard | normalizeFloat64Subnormal( bSig, &bExp, &bSig ); |
3414 | 158142c2 | bellard | } |
3415 | 158142c2 | bellard | if ( aExp == 0 ) { |
3416 | 158142c2 | bellard | if ( aSig == 0 ) return a; |
3417 | 158142c2 | bellard | normalizeFloat64Subnormal( aSig, &aExp, &aSig ); |
3418 | 158142c2 | bellard | } |
3419 | 158142c2 | bellard | expDiff = aExp - bExp; |
3420 | 158142c2 | bellard | aSig = ( aSig | LIT64( 0x0010000000000000 ) )<<11; |
3421 | 158142c2 | bellard | bSig = ( bSig | LIT64( 0x0010000000000000 ) )<<11; |
3422 | 158142c2 | bellard | if ( expDiff < 0 ) { |
3423 | 158142c2 | bellard | if ( expDiff < -1 ) return a; |
3424 | 158142c2 | bellard | aSig >>= 1;
|
3425 | 158142c2 | bellard | } |
3426 | 158142c2 | bellard | q = ( bSig <= aSig ); |
3427 | 158142c2 | bellard | if ( q ) aSig -= bSig;
|
3428 | 158142c2 | bellard | expDiff -= 64;
|
3429 | 158142c2 | bellard | while ( 0 < expDiff ) { |
3430 | 158142c2 | bellard | q = estimateDiv128To64( aSig, 0, bSig );
|
3431 | 158142c2 | bellard | q = ( 2 < q ) ? q - 2 : 0; |
3432 | 158142c2 | bellard | aSig = - ( ( bSig>>2 ) * q );
|
3433 | 158142c2 | bellard | expDiff -= 62;
|
3434 | 158142c2 | bellard | } |
3435 | 158142c2 | bellard | expDiff += 64;
|
3436 | 158142c2 | bellard | if ( 0 < expDiff ) { |
3437 | 158142c2 | bellard | q = estimateDiv128To64( aSig, 0, bSig );
|
3438 | 158142c2 | bellard | q = ( 2 < q ) ? q - 2 : 0; |
3439 | 158142c2 | bellard | q >>= 64 - expDiff;
|
3440 | 158142c2 | bellard | bSig >>= 2;
|
3441 | 158142c2 | bellard | aSig = ( ( aSig>>1 )<<( expDiff - 1 ) ) - bSig * q; |
3442 | 158142c2 | bellard | } |
3443 | 158142c2 | bellard | else {
|
3444 | 158142c2 | bellard | aSig >>= 2;
|
3445 | 158142c2 | bellard | bSig >>= 2;
|
3446 | 158142c2 | bellard | } |
3447 | 158142c2 | bellard | do {
|
3448 | 158142c2 | bellard | alternateASig = aSig; |
3449 | 158142c2 | bellard | ++q; |
3450 | 158142c2 | bellard | aSig -= bSig; |
3451 | bb98fe42 | Andreas Färber | } while ( 0 <= (int64_t) aSig ); |
3452 | 158142c2 | bellard | sigMean = aSig + alternateASig; |
3453 | 158142c2 | bellard | if ( ( sigMean < 0 ) || ( ( sigMean == 0 ) && ( q & 1 ) ) ) { |
3454 | 158142c2 | bellard | aSig = alternateASig; |
3455 | 158142c2 | bellard | } |
3456 | bb98fe42 | Andreas Färber | zSign = ( (int64_t) aSig < 0 );
|
3457 | 158142c2 | bellard | if ( zSign ) aSig = - aSig;
|
3458 | 158142c2 | bellard | return normalizeRoundAndPackFloat64( aSign ^ zSign, bExp, aSig STATUS_VAR );
|
3459 | 158142c2 | bellard | |
3460 | 158142c2 | bellard | } |
3461 | 158142c2 | bellard | |
3462 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
3463 | 158142c2 | bellard | | Returns the square root of the double-precision floating-point value `a'.
|
3464 | 158142c2 | bellard | | The operation is performed according to the IEC/IEEE Standard for Binary
|
3465 | 158142c2 | bellard | | Floating-Point Arithmetic.
|
3466 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
3467 | 158142c2 | bellard | |
3468 | 158142c2 | bellard | float64 float64_sqrt( float64 a STATUS_PARAM ) |
3469 | 158142c2 | bellard | { |
3470 | 158142c2 | bellard | flag aSign; |
3471 | 158142c2 | bellard | int16 aExp, zExp; |
3472 | bb98fe42 | Andreas Färber | uint64_t aSig, zSig, doubleZSig; |
3473 | bb98fe42 | Andreas Färber | uint64_t rem0, rem1, term0, term1; |
3474 | 37d18660 | Peter Maydell | a = float64_squash_input_denormal(a STATUS_VAR); |
3475 | 158142c2 | bellard | |
3476 | 158142c2 | bellard | aSig = extractFloat64Frac( a ); |
3477 | 158142c2 | bellard | aExp = extractFloat64Exp( a ); |
3478 | 158142c2 | bellard | aSign = extractFloat64Sign( a ); |
3479 | 158142c2 | bellard | if ( aExp == 0x7FF ) { |
3480 | 158142c2 | bellard | if ( aSig ) return propagateFloat64NaN( a, a STATUS_VAR ); |
3481 | 158142c2 | bellard | if ( ! aSign ) return a; |
3482 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
3483 | 158142c2 | bellard | return float64_default_nan;
|
3484 | 158142c2 | bellard | } |
3485 | 158142c2 | bellard | if ( aSign ) {
|
3486 | 158142c2 | bellard | if ( ( aExp | aSig ) == 0 ) return a; |
3487 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
3488 | 158142c2 | bellard | return float64_default_nan;
|
3489 | 158142c2 | bellard | } |
3490 | 158142c2 | bellard | if ( aExp == 0 ) { |
3491 | f090c9d4 | pbrook | if ( aSig == 0 ) return float64_zero; |
3492 | 158142c2 | bellard | normalizeFloat64Subnormal( aSig, &aExp, &aSig ); |
3493 | 158142c2 | bellard | } |
3494 | 158142c2 | bellard | zExp = ( ( aExp - 0x3FF )>>1 ) + 0x3FE; |
3495 | 158142c2 | bellard | aSig |= LIT64( 0x0010000000000000 );
|
3496 | 158142c2 | bellard | zSig = estimateSqrt32( aExp, aSig>>21 );
|
3497 | 158142c2 | bellard | aSig <<= 9 - ( aExp & 1 ); |
3498 | 158142c2 | bellard | zSig = estimateDiv128To64( aSig, 0, zSig<<32 ) + ( zSig<<30 ); |
3499 | 158142c2 | bellard | if ( ( zSig & 0x1FF ) <= 5 ) { |
3500 | 158142c2 | bellard | doubleZSig = zSig<<1;
|
3501 | 158142c2 | bellard | mul64To128( zSig, zSig, &term0, &term1 ); |
3502 | 158142c2 | bellard | sub128( aSig, 0, term0, term1, &rem0, &rem1 );
|
3503 | bb98fe42 | Andreas Färber | while ( (int64_t) rem0 < 0 ) { |
3504 | 158142c2 | bellard | --zSig; |
3505 | 158142c2 | bellard | doubleZSig -= 2;
|
3506 | 158142c2 | bellard | add128( rem0, rem1, zSig>>63, doubleZSig | 1, &rem0, &rem1 ); |
3507 | 158142c2 | bellard | } |
3508 | 158142c2 | bellard | zSig |= ( ( rem0 | rem1 ) != 0 );
|
3509 | 158142c2 | bellard | } |
3510 | 158142c2 | bellard | return roundAndPackFloat64( 0, zExp, zSig STATUS_VAR ); |
3511 | 158142c2 | bellard | |
3512 | 158142c2 | bellard | } |
3513 | 158142c2 | bellard | |
3514 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
3515 | 374dfc33 | aurel32 | | Returns the binary log of the double-precision floating-point value `a'.
|
3516 | 374dfc33 | aurel32 | | The operation is performed according to the IEC/IEEE Standard for Binary
|
3517 | 374dfc33 | aurel32 | | Floating-Point Arithmetic.
|
3518 | 374dfc33 | aurel32 | *----------------------------------------------------------------------------*/
|
3519 | 374dfc33 | aurel32 | float64 float64_log2( float64 a STATUS_PARAM ) |
3520 | 374dfc33 | aurel32 | { |
3521 | 374dfc33 | aurel32 | flag aSign, zSign; |
3522 | 374dfc33 | aurel32 | int16 aExp; |
3523 | bb98fe42 | Andreas Färber | uint64_t aSig, aSig0, aSig1, zSig, i; |
3524 | 37d18660 | Peter Maydell | a = float64_squash_input_denormal(a STATUS_VAR); |
3525 | 374dfc33 | aurel32 | |
3526 | 374dfc33 | aurel32 | aSig = extractFloat64Frac( a ); |
3527 | 374dfc33 | aurel32 | aExp = extractFloat64Exp( a ); |
3528 | 374dfc33 | aurel32 | aSign = extractFloat64Sign( a ); |
3529 | 374dfc33 | aurel32 | |
3530 | 374dfc33 | aurel32 | if ( aExp == 0 ) { |
3531 | 374dfc33 | aurel32 | if ( aSig == 0 ) return packFloat64( 1, 0x7FF, 0 ); |
3532 | 374dfc33 | aurel32 | normalizeFloat64Subnormal( aSig, &aExp, &aSig ); |
3533 | 374dfc33 | aurel32 | } |
3534 | 374dfc33 | aurel32 | if ( aSign ) {
|
3535 | 374dfc33 | aurel32 | float_raise( float_flag_invalid STATUS_VAR); |
3536 | 374dfc33 | aurel32 | return float64_default_nan;
|
3537 | 374dfc33 | aurel32 | } |
3538 | 374dfc33 | aurel32 | if ( aExp == 0x7FF ) { |
3539 | 374dfc33 | aurel32 | if ( aSig ) return propagateFloat64NaN( a, float64_zero STATUS_VAR ); |
3540 | 374dfc33 | aurel32 | return a;
|
3541 | 374dfc33 | aurel32 | } |
3542 | 374dfc33 | aurel32 | |
3543 | 374dfc33 | aurel32 | aExp -= 0x3FF;
|
3544 | 374dfc33 | aurel32 | aSig |= LIT64( 0x0010000000000000 );
|
3545 | 374dfc33 | aurel32 | zSign = aExp < 0;
|
3546 | bb98fe42 | Andreas Färber | zSig = (uint64_t)aExp << 52;
|
3547 | 374dfc33 | aurel32 | for (i = 1LL << 51; i > 0; i >>= 1) { |
3548 | 374dfc33 | aurel32 | mul64To128( aSig, aSig, &aSig0, &aSig1 ); |
3549 | 374dfc33 | aurel32 | aSig = ( aSig0 << 12 ) | ( aSig1 >> 52 ); |
3550 | 374dfc33 | aurel32 | if ( aSig & LIT64( 0x0020000000000000 ) ) { |
3551 | 374dfc33 | aurel32 | aSig >>= 1;
|
3552 | 374dfc33 | aurel32 | zSig |= i; |
3553 | 374dfc33 | aurel32 | } |
3554 | 374dfc33 | aurel32 | } |
3555 | 374dfc33 | aurel32 | |
3556 | 374dfc33 | aurel32 | if ( zSign )
|
3557 | 374dfc33 | aurel32 | zSig = -zSig; |
3558 | 374dfc33 | aurel32 | return normalizeRoundAndPackFloat64( zSign, 0x408, zSig STATUS_VAR ); |
3559 | 374dfc33 | aurel32 | } |
3560 | 374dfc33 | aurel32 | |
3561 | 374dfc33 | aurel32 | /*----------------------------------------------------------------------------
|
3562 | 158142c2 | bellard | | Returns 1 if the double-precision floating-point value `a' is equal to the
|
3563 | b689362d | Aurelien Jarno | | corresponding value `b', and 0 otherwise. The invalid exception is raised
|
3564 | b689362d | Aurelien Jarno | | if either operand is a NaN. Otherwise, the comparison is performed
|
3565 | 158142c2 | bellard | | according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
3566 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
3567 | 158142c2 | bellard | |
3568 | b689362d | Aurelien Jarno | int float64_eq( float64 a, float64 b STATUS_PARAM )
|
3569 | 158142c2 | bellard | { |
3570 | bb98fe42 | Andreas Färber | uint64_t av, bv; |
3571 | 37d18660 | Peter Maydell | a = float64_squash_input_denormal(a STATUS_VAR); |
3572 | 37d18660 | Peter Maydell | b = float64_squash_input_denormal(b STATUS_VAR); |
3573 | 158142c2 | bellard | |
3574 | 158142c2 | bellard | if ( ( ( extractFloat64Exp( a ) == 0x7FF ) && extractFloat64Frac( a ) ) |
3575 | 158142c2 | bellard | || ( ( extractFloat64Exp( b ) == 0x7FF ) && extractFloat64Frac( b ) )
|
3576 | 158142c2 | bellard | ) { |
3577 | b689362d | Aurelien Jarno | float_raise( float_flag_invalid STATUS_VAR); |
3578 | 158142c2 | bellard | return 0; |
3579 | 158142c2 | bellard | } |
3580 | f090c9d4 | pbrook | av = float64_val(a); |
3581 | a1b91bb4 | pbrook | bv = float64_val(b); |
3582 | bb98fe42 | Andreas Färber | return ( av == bv ) || ( (uint64_t) ( ( av | bv )<<1 ) == 0 ); |
3583 | 158142c2 | bellard | |
3584 | 158142c2 | bellard | } |
3585 | 158142c2 | bellard | |
3586 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
3587 | 158142c2 | bellard | | Returns 1 if the double-precision floating-point value `a' is less than or
|
3588 | f5a64251 | Aurelien Jarno | | equal to the corresponding value `b', and 0 otherwise. The invalid
|
3589 | f5a64251 | Aurelien Jarno | | exception is raised if either operand is a NaN. The comparison is performed
|
3590 | f5a64251 | Aurelien Jarno | | according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
3591 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
3592 | 158142c2 | bellard | |
3593 | 750afe93 | bellard | int float64_le( float64 a, float64 b STATUS_PARAM )
|
3594 | 158142c2 | bellard | { |
3595 | 158142c2 | bellard | flag aSign, bSign; |
3596 | bb98fe42 | Andreas Färber | uint64_t av, bv; |
3597 | 37d18660 | Peter Maydell | a = float64_squash_input_denormal(a STATUS_VAR); |
3598 | 37d18660 | Peter Maydell | b = float64_squash_input_denormal(b STATUS_VAR); |
3599 | 158142c2 | bellard | |
3600 | 158142c2 | bellard | if ( ( ( extractFloat64Exp( a ) == 0x7FF ) && extractFloat64Frac( a ) ) |
3601 | 158142c2 | bellard | || ( ( extractFloat64Exp( b ) == 0x7FF ) && extractFloat64Frac( b ) )
|
3602 | 158142c2 | bellard | ) { |
3603 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
3604 | 158142c2 | bellard | return 0; |
3605 | 158142c2 | bellard | } |
3606 | 158142c2 | bellard | aSign = extractFloat64Sign( a ); |
3607 | 158142c2 | bellard | bSign = extractFloat64Sign( b ); |
3608 | f090c9d4 | pbrook | av = float64_val(a); |
3609 | a1b91bb4 | pbrook | bv = float64_val(b); |
3610 | bb98fe42 | Andreas Färber | if ( aSign != bSign ) return aSign || ( (uint64_t) ( ( av | bv )<<1 ) == 0 ); |
3611 | f090c9d4 | pbrook | return ( av == bv ) || ( aSign ^ ( av < bv ) );
|
3612 | 158142c2 | bellard | |
3613 | 158142c2 | bellard | } |
3614 | 158142c2 | bellard | |
3615 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
3616 | 158142c2 | bellard | | Returns 1 if the double-precision floating-point value `a' is less than
|
3617 | f5a64251 | Aurelien Jarno | | the corresponding value `b', and 0 otherwise. The invalid exception is
|
3618 | f5a64251 | Aurelien Jarno | | raised if either operand is a NaN. The comparison is performed according
|
3619 | f5a64251 | Aurelien Jarno | | to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
3620 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
3621 | 158142c2 | bellard | |
3622 | 750afe93 | bellard | int float64_lt( float64 a, float64 b STATUS_PARAM )
|
3623 | 158142c2 | bellard | { |
3624 | 158142c2 | bellard | flag aSign, bSign; |
3625 | bb98fe42 | Andreas Färber | uint64_t av, bv; |
3626 | 158142c2 | bellard | |
3627 | 37d18660 | Peter Maydell | a = float64_squash_input_denormal(a STATUS_VAR); |
3628 | 37d18660 | Peter Maydell | b = float64_squash_input_denormal(b STATUS_VAR); |
3629 | 158142c2 | bellard | if ( ( ( extractFloat64Exp( a ) == 0x7FF ) && extractFloat64Frac( a ) ) |
3630 | 158142c2 | bellard | || ( ( extractFloat64Exp( b ) == 0x7FF ) && extractFloat64Frac( b ) )
|
3631 | 158142c2 | bellard | ) { |
3632 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
3633 | 158142c2 | bellard | return 0; |
3634 | 158142c2 | bellard | } |
3635 | 158142c2 | bellard | aSign = extractFloat64Sign( a ); |
3636 | 158142c2 | bellard | bSign = extractFloat64Sign( b ); |
3637 | f090c9d4 | pbrook | av = float64_val(a); |
3638 | a1b91bb4 | pbrook | bv = float64_val(b); |
3639 | bb98fe42 | Andreas Färber | if ( aSign != bSign ) return aSign && ( (uint64_t) ( ( av | bv )<<1 ) != 0 ); |
3640 | f090c9d4 | pbrook | return ( av != bv ) && ( aSign ^ ( av < bv ) );
|
3641 | 158142c2 | bellard | |
3642 | 158142c2 | bellard | } |
3643 | 158142c2 | bellard | |
3644 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
3645 | 67b7861d | Aurelien Jarno | | Returns 1 if the double-precision floating-point values `a' and `b' cannot
|
3646 | f5a64251 | Aurelien Jarno | | be compared, and 0 otherwise. The invalid exception is raised if either
|
3647 | f5a64251 | Aurelien Jarno | | operand is a NaN. The comparison is performed according to the IEC/IEEE
|
3648 | f5a64251 | Aurelien Jarno | | Standard for Binary Floating-Point Arithmetic.
|
3649 | 67b7861d | Aurelien Jarno | *----------------------------------------------------------------------------*/
|
3650 | 67b7861d | Aurelien Jarno | |
3651 | 67b7861d | Aurelien Jarno | int float64_unordered( float64 a, float64 b STATUS_PARAM )
|
3652 | 67b7861d | Aurelien Jarno | { |
3653 | 67b7861d | Aurelien Jarno | a = float64_squash_input_denormal(a STATUS_VAR); |
3654 | 67b7861d | Aurelien Jarno | b = float64_squash_input_denormal(b STATUS_VAR); |
3655 | 67b7861d | Aurelien Jarno | |
3656 | 67b7861d | Aurelien Jarno | if ( ( ( extractFloat64Exp( a ) == 0x7FF ) && extractFloat64Frac( a ) ) |
3657 | 67b7861d | Aurelien Jarno | || ( ( extractFloat64Exp( b ) == 0x7FF ) && extractFloat64Frac( b ) )
|
3658 | 67b7861d | Aurelien Jarno | ) { |
3659 | 67b7861d | Aurelien Jarno | float_raise( float_flag_invalid STATUS_VAR); |
3660 | 67b7861d | Aurelien Jarno | return 1; |
3661 | 67b7861d | Aurelien Jarno | } |
3662 | 67b7861d | Aurelien Jarno | return 0; |
3663 | 67b7861d | Aurelien Jarno | } |
3664 | 67b7861d | Aurelien Jarno | |
3665 | 67b7861d | Aurelien Jarno | /*----------------------------------------------------------------------------
|
3666 | 158142c2 | bellard | | Returns 1 if the double-precision floating-point value `a' is equal to the
|
3667 | f5a64251 | Aurelien Jarno | | corresponding value `b', and 0 otherwise. Quiet NaNs do not cause an
|
3668 | f5a64251 | Aurelien Jarno | | exception.The comparison is performed according to the IEC/IEEE Standard
|
3669 | f5a64251 | Aurelien Jarno | | for Binary Floating-Point Arithmetic.
|
3670 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
3671 | 158142c2 | bellard | |
3672 | b689362d | Aurelien Jarno | int float64_eq_quiet( float64 a, float64 b STATUS_PARAM )
|
3673 | 158142c2 | bellard | { |
3674 | bb98fe42 | Andreas Färber | uint64_t av, bv; |
3675 | 37d18660 | Peter Maydell | a = float64_squash_input_denormal(a STATUS_VAR); |
3676 | 37d18660 | Peter Maydell | b = float64_squash_input_denormal(b STATUS_VAR); |
3677 | 158142c2 | bellard | |
3678 | 158142c2 | bellard | if ( ( ( extractFloat64Exp( a ) == 0x7FF ) && extractFloat64Frac( a ) ) |
3679 | 158142c2 | bellard | || ( ( extractFloat64Exp( b ) == 0x7FF ) && extractFloat64Frac( b ) )
|
3680 | 158142c2 | bellard | ) { |
3681 | b689362d | Aurelien Jarno | if ( float64_is_signaling_nan( a ) || float64_is_signaling_nan( b ) ) {
|
3682 | b689362d | Aurelien Jarno | float_raise( float_flag_invalid STATUS_VAR); |
3683 | b689362d | Aurelien Jarno | } |
3684 | 158142c2 | bellard | return 0; |
3685 | 158142c2 | bellard | } |
3686 | f090c9d4 | pbrook | av = float64_val(a); |
3687 | a1b91bb4 | pbrook | bv = float64_val(b); |
3688 | bb98fe42 | Andreas Färber | return ( av == bv ) || ( (uint64_t) ( ( av | bv )<<1 ) == 0 ); |
3689 | 158142c2 | bellard | |
3690 | 158142c2 | bellard | } |
3691 | 158142c2 | bellard | |
3692 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
3693 | 158142c2 | bellard | | Returns 1 if the double-precision floating-point value `a' is less than or
|
3694 | 158142c2 | bellard | | equal to the corresponding value `b', and 0 otherwise. Quiet NaNs do not
|
3695 | 158142c2 | bellard | | cause an exception. Otherwise, the comparison is performed according to the
|
3696 | 158142c2 | bellard | | IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
3697 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
3698 | 158142c2 | bellard | |
3699 | 750afe93 | bellard | int float64_le_quiet( float64 a, float64 b STATUS_PARAM )
|
3700 | 158142c2 | bellard | { |
3701 | 158142c2 | bellard | flag aSign, bSign; |
3702 | bb98fe42 | Andreas Färber | uint64_t av, bv; |
3703 | 37d18660 | Peter Maydell | a = float64_squash_input_denormal(a STATUS_VAR); |
3704 | 37d18660 | Peter Maydell | b = float64_squash_input_denormal(b STATUS_VAR); |
3705 | 158142c2 | bellard | |
3706 | 158142c2 | bellard | if ( ( ( extractFloat64Exp( a ) == 0x7FF ) && extractFloat64Frac( a ) ) |
3707 | 158142c2 | bellard | || ( ( extractFloat64Exp( b ) == 0x7FF ) && extractFloat64Frac( b ) )
|
3708 | 158142c2 | bellard | ) { |
3709 | 158142c2 | bellard | if ( float64_is_signaling_nan( a ) || float64_is_signaling_nan( b ) ) {
|
3710 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
3711 | 158142c2 | bellard | } |
3712 | 158142c2 | bellard | return 0; |
3713 | 158142c2 | bellard | } |
3714 | 158142c2 | bellard | aSign = extractFloat64Sign( a ); |
3715 | 158142c2 | bellard | bSign = extractFloat64Sign( b ); |
3716 | f090c9d4 | pbrook | av = float64_val(a); |
3717 | a1b91bb4 | pbrook | bv = float64_val(b); |
3718 | bb98fe42 | Andreas Färber | if ( aSign != bSign ) return aSign || ( (uint64_t) ( ( av | bv )<<1 ) == 0 ); |
3719 | f090c9d4 | pbrook | return ( av == bv ) || ( aSign ^ ( av < bv ) );
|
3720 | 158142c2 | bellard | |
3721 | 158142c2 | bellard | } |
3722 | 158142c2 | bellard | |
3723 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
3724 | 158142c2 | bellard | | Returns 1 if the double-precision floating-point value `a' is less than
|
3725 | 158142c2 | bellard | | the corresponding value `b', and 0 otherwise. Quiet NaNs do not cause an
|
3726 | 158142c2 | bellard | | exception. Otherwise, the comparison is performed according to the IEC/IEEE
|
3727 | 158142c2 | bellard | | Standard for Binary Floating-Point Arithmetic.
|
3728 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
3729 | 158142c2 | bellard | |
3730 | 750afe93 | bellard | int float64_lt_quiet( float64 a, float64 b STATUS_PARAM )
|
3731 | 158142c2 | bellard | { |
3732 | 158142c2 | bellard | flag aSign, bSign; |
3733 | bb98fe42 | Andreas Färber | uint64_t av, bv; |
3734 | 37d18660 | Peter Maydell | a = float64_squash_input_denormal(a STATUS_VAR); |
3735 | 37d18660 | Peter Maydell | b = float64_squash_input_denormal(b STATUS_VAR); |
3736 | 158142c2 | bellard | |
3737 | 158142c2 | bellard | if ( ( ( extractFloat64Exp( a ) == 0x7FF ) && extractFloat64Frac( a ) ) |
3738 | 158142c2 | bellard | || ( ( extractFloat64Exp( b ) == 0x7FF ) && extractFloat64Frac( b ) )
|
3739 | 158142c2 | bellard | ) { |
3740 | 158142c2 | bellard | if ( float64_is_signaling_nan( a ) || float64_is_signaling_nan( b ) ) {
|
3741 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
3742 | 158142c2 | bellard | } |
3743 | 158142c2 | bellard | return 0; |
3744 | 158142c2 | bellard | } |
3745 | 158142c2 | bellard | aSign = extractFloat64Sign( a ); |
3746 | 158142c2 | bellard | bSign = extractFloat64Sign( b ); |
3747 | f090c9d4 | pbrook | av = float64_val(a); |
3748 | a1b91bb4 | pbrook | bv = float64_val(b); |
3749 | bb98fe42 | Andreas Färber | if ( aSign != bSign ) return aSign && ( (uint64_t) ( ( av | bv )<<1 ) != 0 ); |
3750 | f090c9d4 | pbrook | return ( av != bv ) && ( aSign ^ ( av < bv ) );
|
3751 | 158142c2 | bellard | |
3752 | 158142c2 | bellard | } |
3753 | 158142c2 | bellard | |
3754 | 67b7861d | Aurelien Jarno | /*----------------------------------------------------------------------------
|
3755 | 67b7861d | Aurelien Jarno | | Returns 1 if the double-precision floating-point values `a' and `b' cannot
|
3756 | 67b7861d | Aurelien Jarno | | be compared, and 0 otherwise. Quiet NaNs do not cause an exception. The
|
3757 | 67b7861d | Aurelien Jarno | | comparison is performed according to the IEC/IEEE Standard for Binary
|
3758 | 67b7861d | Aurelien Jarno | | Floating-Point Arithmetic.
|
3759 | 67b7861d | Aurelien Jarno | *----------------------------------------------------------------------------*/
|
3760 | 67b7861d | Aurelien Jarno | |
3761 | 67b7861d | Aurelien Jarno | int float64_unordered_quiet( float64 a, float64 b STATUS_PARAM )
|
3762 | 67b7861d | Aurelien Jarno | { |
3763 | 67b7861d | Aurelien Jarno | a = float64_squash_input_denormal(a STATUS_VAR); |
3764 | 67b7861d | Aurelien Jarno | b = float64_squash_input_denormal(b STATUS_VAR); |
3765 | 67b7861d | Aurelien Jarno | |
3766 | 67b7861d | Aurelien Jarno | if ( ( ( extractFloat64Exp( a ) == 0x7FF ) && extractFloat64Frac( a ) ) |
3767 | 67b7861d | Aurelien Jarno | || ( ( extractFloat64Exp( b ) == 0x7FF ) && extractFloat64Frac( b ) )
|
3768 | 67b7861d | Aurelien Jarno | ) { |
3769 | 67b7861d | Aurelien Jarno | if ( float64_is_signaling_nan( a ) || float64_is_signaling_nan( b ) ) {
|
3770 | 67b7861d | Aurelien Jarno | float_raise( float_flag_invalid STATUS_VAR); |
3771 | 67b7861d | Aurelien Jarno | } |
3772 | 67b7861d | Aurelien Jarno | return 1; |
3773 | 67b7861d | Aurelien Jarno | } |
3774 | 67b7861d | Aurelien Jarno | return 0; |
3775 | 67b7861d | Aurelien Jarno | } |
3776 | 67b7861d | Aurelien Jarno | |
3777 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
3778 | 158142c2 | bellard | | Returns the result of converting the extended double-precision floating-
|
3779 | 158142c2 | bellard | | point value `a' to the 32-bit two's complement integer format. The
|
3780 | 158142c2 | bellard | | conversion is performed according to the IEC/IEEE Standard for Binary
|
3781 | 158142c2 | bellard | | Floating-Point Arithmetic---which means in particular that the conversion
|
3782 | 158142c2 | bellard | | is rounded according to the current rounding mode. If `a' is a NaN, the
|
3783 | 158142c2 | bellard | | largest positive integer is returned. Otherwise, if the conversion
|
3784 | 158142c2 | bellard | | overflows, the largest integer with the same sign as `a' is returned.
|
3785 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
3786 | 158142c2 | bellard | |
3787 | 158142c2 | bellard | int32 floatx80_to_int32( floatx80 a STATUS_PARAM ) |
3788 | 158142c2 | bellard | { |
3789 | 158142c2 | bellard | flag aSign; |
3790 | 158142c2 | bellard | int32 aExp, shiftCount; |
3791 | bb98fe42 | Andreas Färber | uint64_t aSig; |
3792 | 158142c2 | bellard | |
3793 | 158142c2 | bellard | aSig = extractFloatx80Frac( a ); |
3794 | 158142c2 | bellard | aExp = extractFloatx80Exp( a ); |
3795 | 158142c2 | bellard | aSign = extractFloatx80Sign( a ); |
3796 | bb98fe42 | Andreas Färber | if ( ( aExp == 0x7FFF ) && (uint64_t) ( aSig<<1 ) ) aSign = 0; |
3797 | 158142c2 | bellard | shiftCount = 0x4037 - aExp;
|
3798 | 158142c2 | bellard | if ( shiftCount <= 0 ) shiftCount = 1; |
3799 | 158142c2 | bellard | shift64RightJamming( aSig, shiftCount, &aSig ); |
3800 | 158142c2 | bellard | return roundAndPackInt32( aSign, aSig STATUS_VAR );
|
3801 | 158142c2 | bellard | |
3802 | 158142c2 | bellard | } |
3803 | 158142c2 | bellard | |
3804 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
3805 | 158142c2 | bellard | | Returns the result of converting the extended double-precision floating-
|
3806 | 158142c2 | bellard | | point value `a' to the 32-bit two's complement integer format. The
|
3807 | 158142c2 | bellard | | conversion is performed according to the IEC/IEEE Standard for Binary
|
3808 | 158142c2 | bellard | | Floating-Point Arithmetic, except that the conversion is always rounded
|
3809 | 158142c2 | bellard | | toward zero. If `a' is a NaN, the largest positive integer is returned.
|
3810 | 158142c2 | bellard | | Otherwise, if the conversion overflows, the largest integer with the same
|
3811 | 158142c2 | bellard | | sign as `a' is returned.
|
3812 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
3813 | 158142c2 | bellard | |
3814 | 158142c2 | bellard | int32 floatx80_to_int32_round_to_zero( floatx80 a STATUS_PARAM ) |
3815 | 158142c2 | bellard | { |
3816 | 158142c2 | bellard | flag aSign; |
3817 | 158142c2 | bellard | int32 aExp, shiftCount; |
3818 | bb98fe42 | Andreas Färber | uint64_t aSig, savedASig; |
3819 | 158142c2 | bellard | int32 z; |
3820 | 158142c2 | bellard | |
3821 | 158142c2 | bellard | aSig = extractFloatx80Frac( a ); |
3822 | 158142c2 | bellard | aExp = extractFloatx80Exp( a ); |
3823 | 158142c2 | bellard | aSign = extractFloatx80Sign( a ); |
3824 | 158142c2 | bellard | if ( 0x401E < aExp ) { |
3825 | bb98fe42 | Andreas Färber | if ( ( aExp == 0x7FFF ) && (uint64_t) ( aSig<<1 ) ) aSign = 0; |
3826 | 158142c2 | bellard | goto invalid;
|
3827 | 158142c2 | bellard | } |
3828 | 158142c2 | bellard | else if ( aExp < 0x3FFF ) { |
3829 | 158142c2 | bellard | if ( aExp || aSig ) STATUS(float_exception_flags) |= float_flag_inexact;
|
3830 | 158142c2 | bellard | return 0; |
3831 | 158142c2 | bellard | } |
3832 | 158142c2 | bellard | shiftCount = 0x403E - aExp;
|
3833 | 158142c2 | bellard | savedASig = aSig; |
3834 | 158142c2 | bellard | aSig >>= shiftCount; |
3835 | 158142c2 | bellard | z = aSig; |
3836 | 158142c2 | bellard | if ( aSign ) z = - z;
|
3837 | 158142c2 | bellard | if ( ( z < 0 ) ^ aSign ) { |
3838 | 158142c2 | bellard | invalid:
|
3839 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
3840 | bb98fe42 | Andreas Färber | return aSign ? (int32_t) 0x80000000 : 0x7FFFFFFF; |
3841 | 158142c2 | bellard | } |
3842 | 158142c2 | bellard | if ( ( aSig<<shiftCount ) != savedASig ) {
|
3843 | 158142c2 | bellard | STATUS(float_exception_flags) |= float_flag_inexact; |
3844 | 158142c2 | bellard | } |
3845 | 158142c2 | bellard | return z;
|
3846 | 158142c2 | bellard | |
3847 | 158142c2 | bellard | } |
3848 | 158142c2 | bellard | |
3849 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
3850 | 158142c2 | bellard | | Returns the result of converting the extended double-precision floating-
|
3851 | 158142c2 | bellard | | point value `a' to the 64-bit two's complement integer format. The
|
3852 | 158142c2 | bellard | | conversion is performed according to the IEC/IEEE Standard for Binary
|
3853 | 158142c2 | bellard | | Floating-Point Arithmetic---which means in particular that the conversion
|
3854 | 158142c2 | bellard | | is rounded according to the current rounding mode. If `a' is a NaN,
|
3855 | 158142c2 | bellard | | the largest positive integer is returned. Otherwise, if the conversion
|
3856 | 158142c2 | bellard | | overflows, the largest integer with the same sign as `a' is returned.
|
3857 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
3858 | 158142c2 | bellard | |
3859 | 158142c2 | bellard | int64 floatx80_to_int64( floatx80 a STATUS_PARAM ) |
3860 | 158142c2 | bellard | { |
3861 | 158142c2 | bellard | flag aSign; |
3862 | 158142c2 | bellard | int32 aExp, shiftCount; |
3863 | bb98fe42 | Andreas Färber | uint64_t aSig, aSigExtra; |
3864 | 158142c2 | bellard | |
3865 | 158142c2 | bellard | aSig = extractFloatx80Frac( a ); |
3866 | 158142c2 | bellard | aExp = extractFloatx80Exp( a ); |
3867 | 158142c2 | bellard | aSign = extractFloatx80Sign( a ); |
3868 | 158142c2 | bellard | shiftCount = 0x403E - aExp;
|
3869 | 158142c2 | bellard | if ( shiftCount <= 0 ) { |
3870 | 158142c2 | bellard | if ( shiftCount ) {
|
3871 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
3872 | 158142c2 | bellard | if ( ! aSign
|
3873 | 158142c2 | bellard | || ( ( aExp == 0x7FFF )
|
3874 | 158142c2 | bellard | && ( aSig != LIT64( 0x8000000000000000 ) ) )
|
3875 | 158142c2 | bellard | ) { |
3876 | 158142c2 | bellard | return LIT64( 0x7FFFFFFFFFFFFFFF ); |
3877 | 158142c2 | bellard | } |
3878 | bb98fe42 | Andreas Färber | return (int64_t) LIT64( 0x8000000000000000 ); |
3879 | 158142c2 | bellard | } |
3880 | 158142c2 | bellard | aSigExtra = 0;
|
3881 | 158142c2 | bellard | } |
3882 | 158142c2 | bellard | else {
|
3883 | 158142c2 | bellard | shift64ExtraRightJamming( aSig, 0, shiftCount, &aSig, &aSigExtra );
|
3884 | 158142c2 | bellard | } |
3885 | 158142c2 | bellard | return roundAndPackInt64( aSign, aSig, aSigExtra STATUS_VAR );
|
3886 | 158142c2 | bellard | |
3887 | 158142c2 | bellard | } |
3888 | 158142c2 | bellard | |
3889 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
3890 | 158142c2 | bellard | | Returns the result of converting the extended double-precision floating-
|
3891 | 158142c2 | bellard | | point value `a' to the 64-bit two's complement integer format. The
|
3892 | 158142c2 | bellard | | conversion is performed according to the IEC/IEEE Standard for Binary
|
3893 | 158142c2 | bellard | | Floating-Point Arithmetic, except that the conversion is always rounded
|
3894 | 158142c2 | bellard | | toward zero. If `a' is a NaN, the largest positive integer is returned.
|
3895 | 158142c2 | bellard | | Otherwise, if the conversion overflows, the largest integer with the same
|
3896 | 158142c2 | bellard | | sign as `a' is returned.
|
3897 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
3898 | 158142c2 | bellard | |
3899 | 158142c2 | bellard | int64 floatx80_to_int64_round_to_zero( floatx80 a STATUS_PARAM ) |
3900 | 158142c2 | bellard | { |
3901 | 158142c2 | bellard | flag aSign; |
3902 | 158142c2 | bellard | int32 aExp, shiftCount; |
3903 | bb98fe42 | Andreas Färber | uint64_t aSig; |
3904 | 158142c2 | bellard | int64 z; |
3905 | 158142c2 | bellard | |
3906 | 158142c2 | bellard | aSig = extractFloatx80Frac( a ); |
3907 | 158142c2 | bellard | aExp = extractFloatx80Exp( a ); |
3908 | 158142c2 | bellard | aSign = extractFloatx80Sign( a ); |
3909 | 158142c2 | bellard | shiftCount = aExp - 0x403E;
|
3910 | 158142c2 | bellard | if ( 0 <= shiftCount ) { |
3911 | 158142c2 | bellard | aSig &= LIT64( 0x7FFFFFFFFFFFFFFF );
|
3912 | 158142c2 | bellard | if ( ( a.high != 0xC03E ) || aSig ) { |
3913 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
3914 | 158142c2 | bellard | if ( ! aSign || ( ( aExp == 0x7FFF ) && aSig ) ) { |
3915 | 158142c2 | bellard | return LIT64( 0x7FFFFFFFFFFFFFFF ); |
3916 | 158142c2 | bellard | } |
3917 | 158142c2 | bellard | } |
3918 | bb98fe42 | Andreas Färber | return (int64_t) LIT64( 0x8000000000000000 ); |
3919 | 158142c2 | bellard | } |
3920 | 158142c2 | bellard | else if ( aExp < 0x3FFF ) { |
3921 | 158142c2 | bellard | if ( aExp | aSig ) STATUS(float_exception_flags) |= float_flag_inexact;
|
3922 | 158142c2 | bellard | return 0; |
3923 | 158142c2 | bellard | } |
3924 | 158142c2 | bellard | z = aSig>>( - shiftCount ); |
3925 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( aSig<<( shiftCount & 63 ) ) ) { |
3926 | 158142c2 | bellard | STATUS(float_exception_flags) |= float_flag_inexact; |
3927 | 158142c2 | bellard | } |
3928 | 158142c2 | bellard | if ( aSign ) z = - z;
|
3929 | 158142c2 | bellard | return z;
|
3930 | 158142c2 | bellard | |
3931 | 158142c2 | bellard | } |
3932 | 158142c2 | bellard | |
3933 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
3934 | 158142c2 | bellard | | Returns the result of converting the extended double-precision floating-
|
3935 | 158142c2 | bellard | | point value `a' to the single-precision floating-point format. The
|
3936 | 158142c2 | bellard | | conversion is performed according to the IEC/IEEE Standard for Binary
|
3937 | 158142c2 | bellard | | Floating-Point Arithmetic.
|
3938 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
3939 | 158142c2 | bellard | |
3940 | 158142c2 | bellard | float32 floatx80_to_float32( floatx80 a STATUS_PARAM ) |
3941 | 158142c2 | bellard | { |
3942 | 158142c2 | bellard | flag aSign; |
3943 | 158142c2 | bellard | int32 aExp; |
3944 | bb98fe42 | Andreas Färber | uint64_t aSig; |
3945 | 158142c2 | bellard | |
3946 | 158142c2 | bellard | aSig = extractFloatx80Frac( a ); |
3947 | 158142c2 | bellard | aExp = extractFloatx80Exp( a ); |
3948 | 158142c2 | bellard | aSign = extractFloatx80Sign( a ); |
3949 | 158142c2 | bellard | if ( aExp == 0x7FFF ) { |
3950 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( aSig<<1 ) ) { |
3951 | bcd4d9af | Christophe Lyon | return commonNaNToFloat32( floatx80ToCommonNaN( a STATUS_VAR ) STATUS_VAR );
|
3952 | 158142c2 | bellard | } |
3953 | 158142c2 | bellard | return packFloat32( aSign, 0xFF, 0 ); |
3954 | 158142c2 | bellard | } |
3955 | 158142c2 | bellard | shift64RightJamming( aSig, 33, &aSig );
|
3956 | 158142c2 | bellard | if ( aExp || aSig ) aExp -= 0x3F81; |
3957 | 158142c2 | bellard | return roundAndPackFloat32( aSign, aExp, aSig STATUS_VAR );
|
3958 | 158142c2 | bellard | |
3959 | 158142c2 | bellard | } |
3960 | 158142c2 | bellard | |
3961 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
3962 | 158142c2 | bellard | | Returns the result of converting the extended double-precision floating-
|
3963 | 158142c2 | bellard | | point value `a' to the double-precision floating-point format. The
|
3964 | 158142c2 | bellard | | conversion is performed according to the IEC/IEEE Standard for Binary
|
3965 | 158142c2 | bellard | | Floating-Point Arithmetic.
|
3966 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
3967 | 158142c2 | bellard | |
3968 | 158142c2 | bellard | float64 floatx80_to_float64( floatx80 a STATUS_PARAM ) |
3969 | 158142c2 | bellard | { |
3970 | 158142c2 | bellard | flag aSign; |
3971 | 158142c2 | bellard | int32 aExp; |
3972 | bb98fe42 | Andreas Färber | uint64_t aSig, zSig; |
3973 | 158142c2 | bellard | |
3974 | 158142c2 | bellard | aSig = extractFloatx80Frac( a ); |
3975 | 158142c2 | bellard | aExp = extractFloatx80Exp( a ); |
3976 | 158142c2 | bellard | aSign = extractFloatx80Sign( a ); |
3977 | 158142c2 | bellard | if ( aExp == 0x7FFF ) { |
3978 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( aSig<<1 ) ) { |
3979 | bcd4d9af | Christophe Lyon | return commonNaNToFloat64( floatx80ToCommonNaN( a STATUS_VAR ) STATUS_VAR );
|
3980 | 158142c2 | bellard | } |
3981 | 158142c2 | bellard | return packFloat64( aSign, 0x7FF, 0 ); |
3982 | 158142c2 | bellard | } |
3983 | 158142c2 | bellard | shift64RightJamming( aSig, 1, &zSig );
|
3984 | 158142c2 | bellard | if ( aExp || aSig ) aExp -= 0x3C01; |
3985 | 158142c2 | bellard | return roundAndPackFloat64( aSign, aExp, zSig STATUS_VAR );
|
3986 | 158142c2 | bellard | |
3987 | 158142c2 | bellard | } |
3988 | 158142c2 | bellard | |
3989 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
3990 | 158142c2 | bellard | | Returns the result of converting the extended double-precision floating-
|
3991 | 158142c2 | bellard | | point value `a' to the quadruple-precision floating-point format. The
|
3992 | 158142c2 | bellard | | conversion is performed according to the IEC/IEEE Standard for Binary
|
3993 | 158142c2 | bellard | | Floating-Point Arithmetic.
|
3994 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
3995 | 158142c2 | bellard | |
3996 | 158142c2 | bellard | float128 floatx80_to_float128( floatx80 a STATUS_PARAM ) |
3997 | 158142c2 | bellard | { |
3998 | 158142c2 | bellard | flag aSign; |
3999 | 158142c2 | bellard | int16 aExp; |
4000 | bb98fe42 | Andreas Färber | uint64_t aSig, zSig0, zSig1; |
4001 | 158142c2 | bellard | |
4002 | 158142c2 | bellard | aSig = extractFloatx80Frac( a ); |
4003 | 158142c2 | bellard | aExp = extractFloatx80Exp( a ); |
4004 | 158142c2 | bellard | aSign = extractFloatx80Sign( a ); |
4005 | bb98fe42 | Andreas Färber | if ( ( aExp == 0x7FFF ) && (uint64_t) ( aSig<<1 ) ) { |
4006 | bcd4d9af | Christophe Lyon | return commonNaNToFloat128( floatx80ToCommonNaN( a STATUS_VAR ) STATUS_VAR );
|
4007 | 158142c2 | bellard | } |
4008 | 158142c2 | bellard | shift128Right( aSig<<1, 0, 16, &zSig0, &zSig1 ); |
4009 | 158142c2 | bellard | return packFloat128( aSign, aExp, zSig0, zSig1 );
|
4010 | 158142c2 | bellard | |
4011 | 158142c2 | bellard | } |
4012 | 158142c2 | bellard | |
4013 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
4014 | 158142c2 | bellard | | Rounds the extended double-precision floating-point value `a' to an integer,
|
4015 | 158142c2 | bellard | | and returns the result as an extended quadruple-precision floating-point
|
4016 | 158142c2 | bellard | | value. The operation is performed according to the IEC/IEEE Standard for
|
4017 | 158142c2 | bellard | | Binary Floating-Point Arithmetic.
|
4018 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
4019 | 158142c2 | bellard | |
4020 | 158142c2 | bellard | floatx80 floatx80_round_to_int( floatx80 a STATUS_PARAM ) |
4021 | 158142c2 | bellard | { |
4022 | 158142c2 | bellard | flag aSign; |
4023 | 158142c2 | bellard | int32 aExp; |
4024 | bb98fe42 | Andreas Färber | uint64_t lastBitMask, roundBitsMask; |
4025 | 158142c2 | bellard | int8 roundingMode; |
4026 | 158142c2 | bellard | floatx80 z; |
4027 | 158142c2 | bellard | |
4028 | 158142c2 | bellard | aExp = extractFloatx80Exp( a ); |
4029 | 158142c2 | bellard | if ( 0x403E <= aExp ) { |
4030 | bb98fe42 | Andreas Färber | if ( ( aExp == 0x7FFF ) && (uint64_t) ( extractFloatx80Frac( a )<<1 ) ) { |
4031 | 158142c2 | bellard | return propagateFloatx80NaN( a, a STATUS_VAR );
|
4032 | 158142c2 | bellard | } |
4033 | 158142c2 | bellard | return a;
|
4034 | 158142c2 | bellard | } |
4035 | 158142c2 | bellard | if ( aExp < 0x3FFF ) { |
4036 | 158142c2 | bellard | if ( ( aExp == 0 ) |
4037 | bb98fe42 | Andreas Färber | && ( (uint64_t) ( extractFloatx80Frac( a )<<1 ) == 0 ) ) { |
4038 | 158142c2 | bellard | return a;
|
4039 | 158142c2 | bellard | } |
4040 | 158142c2 | bellard | STATUS(float_exception_flags) |= float_flag_inexact; |
4041 | 158142c2 | bellard | aSign = extractFloatx80Sign( a ); |
4042 | 158142c2 | bellard | switch ( STATUS(float_rounding_mode) ) {
|
4043 | 158142c2 | bellard | case float_round_nearest_even:
|
4044 | bb98fe42 | Andreas Färber | if ( ( aExp == 0x3FFE ) && (uint64_t) ( extractFloatx80Frac( a )<<1 ) |
4045 | 158142c2 | bellard | ) { |
4046 | 158142c2 | bellard | return
|
4047 | 158142c2 | bellard | packFloatx80( aSign, 0x3FFF, LIT64( 0x8000000000000000 ) ); |
4048 | 158142c2 | bellard | } |
4049 | 158142c2 | bellard | break;
|
4050 | 158142c2 | bellard | case float_round_down:
|
4051 | 158142c2 | bellard | return
|
4052 | 158142c2 | bellard | aSign ? |
4053 | 158142c2 | bellard | packFloatx80( 1, 0x3FFF, LIT64( 0x8000000000000000 ) ) |
4054 | 158142c2 | bellard | : packFloatx80( 0, 0, 0 ); |
4055 | 158142c2 | bellard | case float_round_up:
|
4056 | 158142c2 | bellard | return
|
4057 | 158142c2 | bellard | aSign ? packFloatx80( 1, 0, 0 ) |
4058 | 158142c2 | bellard | : packFloatx80( 0, 0x3FFF, LIT64( 0x8000000000000000 ) ); |
4059 | 158142c2 | bellard | } |
4060 | 158142c2 | bellard | return packFloatx80( aSign, 0, 0 ); |
4061 | 158142c2 | bellard | } |
4062 | 158142c2 | bellard | lastBitMask = 1;
|
4063 | 158142c2 | bellard | lastBitMask <<= 0x403E - aExp;
|
4064 | 158142c2 | bellard | roundBitsMask = lastBitMask - 1;
|
4065 | 158142c2 | bellard | z = a; |
4066 | 158142c2 | bellard | roundingMode = STATUS(float_rounding_mode); |
4067 | 158142c2 | bellard | if ( roundingMode == float_round_nearest_even ) {
|
4068 | 158142c2 | bellard | z.low += lastBitMask>>1;
|
4069 | 158142c2 | bellard | if ( ( z.low & roundBitsMask ) == 0 ) z.low &= ~ lastBitMask; |
4070 | 158142c2 | bellard | } |
4071 | 158142c2 | bellard | else if ( roundingMode != float_round_to_zero ) { |
4072 | 158142c2 | bellard | if ( extractFloatx80Sign( z ) ^ ( roundingMode == float_round_up ) ) {
|
4073 | 158142c2 | bellard | z.low += roundBitsMask; |
4074 | 158142c2 | bellard | } |
4075 | 158142c2 | bellard | } |
4076 | 158142c2 | bellard | z.low &= ~ roundBitsMask; |
4077 | 158142c2 | bellard | if ( z.low == 0 ) { |
4078 | 158142c2 | bellard | ++z.high; |
4079 | 158142c2 | bellard | z.low = LIT64( 0x8000000000000000 );
|
4080 | 158142c2 | bellard | } |
4081 | 158142c2 | bellard | if ( z.low != a.low ) STATUS(float_exception_flags) |= float_flag_inexact;
|
4082 | 158142c2 | bellard | return z;
|
4083 | 158142c2 | bellard | |
4084 | 158142c2 | bellard | } |
4085 | 158142c2 | bellard | |
4086 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
4087 | 158142c2 | bellard | | Returns the result of adding the absolute values of the extended double-
|
4088 | 158142c2 | bellard | | precision floating-point values `a' and `b'. If `zSign' is 1, the sum is
|
4089 | 158142c2 | bellard | | negated before being returned. `zSign' is ignored if the result is a NaN.
|
4090 | 158142c2 | bellard | | The addition is performed according to the IEC/IEEE Standard for Binary
|
4091 | 158142c2 | bellard | | Floating-Point Arithmetic.
|
4092 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
4093 | 158142c2 | bellard | |
4094 | 158142c2 | bellard | static floatx80 addFloatx80Sigs( floatx80 a, floatx80 b, flag zSign STATUS_PARAM)
|
4095 | 158142c2 | bellard | { |
4096 | 158142c2 | bellard | int32 aExp, bExp, zExp; |
4097 | bb98fe42 | Andreas Färber | uint64_t aSig, bSig, zSig0, zSig1; |
4098 | 158142c2 | bellard | int32 expDiff; |
4099 | 158142c2 | bellard | |
4100 | 158142c2 | bellard | aSig = extractFloatx80Frac( a ); |
4101 | 158142c2 | bellard | aExp = extractFloatx80Exp( a ); |
4102 | 158142c2 | bellard | bSig = extractFloatx80Frac( b ); |
4103 | 158142c2 | bellard | bExp = extractFloatx80Exp( b ); |
4104 | 158142c2 | bellard | expDiff = aExp - bExp; |
4105 | 158142c2 | bellard | if ( 0 < expDiff ) { |
4106 | 158142c2 | bellard | if ( aExp == 0x7FFF ) { |
4107 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( aSig<<1 ) ) return propagateFloatx80NaN( a, b STATUS_VAR ); |
4108 | 158142c2 | bellard | return a;
|
4109 | 158142c2 | bellard | } |
4110 | 158142c2 | bellard | if ( bExp == 0 ) --expDiff; |
4111 | 158142c2 | bellard | shift64ExtraRightJamming( bSig, 0, expDiff, &bSig, &zSig1 );
|
4112 | 158142c2 | bellard | zExp = aExp; |
4113 | 158142c2 | bellard | } |
4114 | 158142c2 | bellard | else if ( expDiff < 0 ) { |
4115 | 158142c2 | bellard | if ( bExp == 0x7FFF ) { |
4116 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( bSig<<1 ) ) return propagateFloatx80NaN( a, b STATUS_VAR ); |
4117 | 158142c2 | bellard | return packFloatx80( zSign, 0x7FFF, LIT64( 0x8000000000000000 ) ); |
4118 | 158142c2 | bellard | } |
4119 | 158142c2 | bellard | if ( aExp == 0 ) ++expDiff; |
4120 | 158142c2 | bellard | shift64ExtraRightJamming( aSig, 0, - expDiff, &aSig, &zSig1 );
|
4121 | 158142c2 | bellard | zExp = bExp; |
4122 | 158142c2 | bellard | } |
4123 | 158142c2 | bellard | else {
|
4124 | 158142c2 | bellard | if ( aExp == 0x7FFF ) { |
4125 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( ( aSig | bSig )<<1 ) ) { |
4126 | 158142c2 | bellard | return propagateFloatx80NaN( a, b STATUS_VAR );
|
4127 | 158142c2 | bellard | } |
4128 | 158142c2 | bellard | return a;
|
4129 | 158142c2 | bellard | } |
4130 | 158142c2 | bellard | zSig1 = 0;
|
4131 | 158142c2 | bellard | zSig0 = aSig + bSig; |
4132 | 158142c2 | bellard | if ( aExp == 0 ) { |
4133 | 158142c2 | bellard | normalizeFloatx80Subnormal( zSig0, &zExp, &zSig0 ); |
4134 | 158142c2 | bellard | goto roundAndPack;
|
4135 | 158142c2 | bellard | } |
4136 | 158142c2 | bellard | zExp = aExp; |
4137 | 158142c2 | bellard | goto shiftRight1;
|
4138 | 158142c2 | bellard | } |
4139 | 158142c2 | bellard | zSig0 = aSig + bSig; |
4140 | bb98fe42 | Andreas Färber | if ( (int64_t) zSig0 < 0 ) goto roundAndPack; |
4141 | 158142c2 | bellard | shiftRight1:
|
4142 | 158142c2 | bellard | shift64ExtraRightJamming( zSig0, zSig1, 1, &zSig0, &zSig1 );
|
4143 | 158142c2 | bellard | zSig0 |= LIT64( 0x8000000000000000 );
|
4144 | 158142c2 | bellard | ++zExp; |
4145 | 158142c2 | bellard | roundAndPack:
|
4146 | 158142c2 | bellard | return
|
4147 | 158142c2 | bellard | roundAndPackFloatx80( |
4148 | 158142c2 | bellard | STATUS(floatx80_rounding_precision), zSign, zExp, zSig0, zSig1 STATUS_VAR ); |
4149 | 158142c2 | bellard | |
4150 | 158142c2 | bellard | } |
4151 | 158142c2 | bellard | |
4152 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
4153 | 158142c2 | bellard | | Returns the result of subtracting the absolute values of the extended
|
4154 | 158142c2 | bellard | | double-precision floating-point values `a' and `b'. If `zSign' is 1, the
|
4155 | 158142c2 | bellard | | difference is negated before being returned. `zSign' is ignored if the
|
4156 | 158142c2 | bellard | | result is a NaN. The subtraction is performed according to the IEC/IEEE
|
4157 | 158142c2 | bellard | | Standard for Binary Floating-Point Arithmetic.
|
4158 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
4159 | 158142c2 | bellard | |
4160 | 158142c2 | bellard | static floatx80 subFloatx80Sigs( floatx80 a, floatx80 b, flag zSign STATUS_PARAM )
|
4161 | 158142c2 | bellard | { |
4162 | 158142c2 | bellard | int32 aExp, bExp, zExp; |
4163 | bb98fe42 | Andreas Färber | uint64_t aSig, bSig, zSig0, zSig1; |
4164 | 158142c2 | bellard | int32 expDiff; |
4165 | 158142c2 | bellard | floatx80 z; |
4166 | 158142c2 | bellard | |
4167 | 158142c2 | bellard | aSig = extractFloatx80Frac( a ); |
4168 | 158142c2 | bellard | aExp = extractFloatx80Exp( a ); |
4169 | 158142c2 | bellard | bSig = extractFloatx80Frac( b ); |
4170 | 158142c2 | bellard | bExp = extractFloatx80Exp( b ); |
4171 | 158142c2 | bellard | expDiff = aExp - bExp; |
4172 | 158142c2 | bellard | if ( 0 < expDiff ) goto aExpBigger; |
4173 | 158142c2 | bellard | if ( expDiff < 0 ) goto bExpBigger; |
4174 | 158142c2 | bellard | if ( aExp == 0x7FFF ) { |
4175 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( ( aSig | bSig )<<1 ) ) { |
4176 | 158142c2 | bellard | return propagateFloatx80NaN( a, b STATUS_VAR );
|
4177 | 158142c2 | bellard | } |
4178 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
4179 | 158142c2 | bellard | z.low = floatx80_default_nan_low; |
4180 | 158142c2 | bellard | z.high = floatx80_default_nan_high; |
4181 | 158142c2 | bellard | return z;
|
4182 | 158142c2 | bellard | } |
4183 | 158142c2 | bellard | if ( aExp == 0 ) { |
4184 | 158142c2 | bellard | aExp = 1;
|
4185 | 158142c2 | bellard | bExp = 1;
|
4186 | 158142c2 | bellard | } |
4187 | 158142c2 | bellard | zSig1 = 0;
|
4188 | 158142c2 | bellard | if ( bSig < aSig ) goto aBigger; |
4189 | 158142c2 | bellard | if ( aSig < bSig ) goto bBigger; |
4190 | 158142c2 | bellard | return packFloatx80( STATUS(float_rounding_mode) == float_round_down, 0, 0 ); |
4191 | 158142c2 | bellard | bExpBigger:
|
4192 | 158142c2 | bellard | if ( bExp == 0x7FFF ) { |
4193 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( bSig<<1 ) ) return propagateFloatx80NaN( a, b STATUS_VAR ); |
4194 | 158142c2 | bellard | return packFloatx80( zSign ^ 1, 0x7FFF, LIT64( 0x8000000000000000 ) ); |
4195 | 158142c2 | bellard | } |
4196 | 158142c2 | bellard | if ( aExp == 0 ) ++expDiff; |
4197 | 158142c2 | bellard | shift128RightJamming( aSig, 0, - expDiff, &aSig, &zSig1 );
|
4198 | 158142c2 | bellard | bBigger:
|
4199 | 158142c2 | bellard | sub128( bSig, 0, aSig, zSig1, &zSig0, &zSig1 );
|
4200 | 158142c2 | bellard | zExp = bExp; |
4201 | 158142c2 | bellard | zSign ^= 1;
|
4202 | 158142c2 | bellard | goto normalizeRoundAndPack;
|
4203 | 158142c2 | bellard | aExpBigger:
|
4204 | 158142c2 | bellard | if ( aExp == 0x7FFF ) { |
4205 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( aSig<<1 ) ) return propagateFloatx80NaN( a, b STATUS_VAR ); |
4206 | 158142c2 | bellard | return a;
|
4207 | 158142c2 | bellard | } |
4208 | 158142c2 | bellard | if ( bExp == 0 ) --expDiff; |
4209 | 158142c2 | bellard | shift128RightJamming( bSig, 0, expDiff, &bSig, &zSig1 );
|
4210 | 158142c2 | bellard | aBigger:
|
4211 | 158142c2 | bellard | sub128( aSig, 0, bSig, zSig1, &zSig0, &zSig1 );
|
4212 | 158142c2 | bellard | zExp = aExp; |
4213 | 158142c2 | bellard | normalizeRoundAndPack:
|
4214 | 158142c2 | bellard | return
|
4215 | 158142c2 | bellard | normalizeRoundAndPackFloatx80( |
4216 | 158142c2 | bellard | STATUS(floatx80_rounding_precision), zSign, zExp, zSig0, zSig1 STATUS_VAR ); |
4217 | 158142c2 | bellard | |
4218 | 158142c2 | bellard | } |
4219 | 158142c2 | bellard | |
4220 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
4221 | 158142c2 | bellard | | Returns the result of adding the extended double-precision floating-point
|
4222 | 158142c2 | bellard | | values `a' and `b'. The operation is performed according to the IEC/IEEE
|
4223 | 158142c2 | bellard | | Standard for Binary Floating-Point Arithmetic.
|
4224 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
4225 | 158142c2 | bellard | |
4226 | 158142c2 | bellard | floatx80 floatx80_add( floatx80 a, floatx80 b STATUS_PARAM ) |
4227 | 158142c2 | bellard | { |
4228 | 158142c2 | bellard | flag aSign, bSign; |
4229 | 158142c2 | bellard | |
4230 | 158142c2 | bellard | aSign = extractFloatx80Sign( a ); |
4231 | 158142c2 | bellard | bSign = extractFloatx80Sign( b ); |
4232 | 158142c2 | bellard | if ( aSign == bSign ) {
|
4233 | 158142c2 | bellard | return addFloatx80Sigs( a, b, aSign STATUS_VAR );
|
4234 | 158142c2 | bellard | } |
4235 | 158142c2 | bellard | else {
|
4236 | 158142c2 | bellard | return subFloatx80Sigs( a, b, aSign STATUS_VAR );
|
4237 | 158142c2 | bellard | } |
4238 | 158142c2 | bellard | |
4239 | 158142c2 | bellard | } |
4240 | 158142c2 | bellard | |
4241 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
4242 | 158142c2 | bellard | | Returns the result of subtracting the extended double-precision floating-
|
4243 | 158142c2 | bellard | | point values `a' and `b'. The operation is performed according to the
|
4244 | 158142c2 | bellard | | IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
4245 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
4246 | 158142c2 | bellard | |
4247 | 158142c2 | bellard | floatx80 floatx80_sub( floatx80 a, floatx80 b STATUS_PARAM ) |
4248 | 158142c2 | bellard | { |
4249 | 158142c2 | bellard | flag aSign, bSign; |
4250 | 158142c2 | bellard | |
4251 | 158142c2 | bellard | aSign = extractFloatx80Sign( a ); |
4252 | 158142c2 | bellard | bSign = extractFloatx80Sign( b ); |
4253 | 158142c2 | bellard | if ( aSign == bSign ) {
|
4254 | 158142c2 | bellard | return subFloatx80Sigs( a, b, aSign STATUS_VAR );
|
4255 | 158142c2 | bellard | } |
4256 | 158142c2 | bellard | else {
|
4257 | 158142c2 | bellard | return addFloatx80Sigs( a, b, aSign STATUS_VAR );
|
4258 | 158142c2 | bellard | } |
4259 | 158142c2 | bellard | |
4260 | 158142c2 | bellard | } |
4261 | 158142c2 | bellard | |
4262 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
4263 | 158142c2 | bellard | | Returns the result of multiplying the extended double-precision floating-
|
4264 | 158142c2 | bellard | | point values `a' and `b'. The operation is performed according to the
|
4265 | 158142c2 | bellard | | IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
4266 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
4267 | 158142c2 | bellard | |
4268 | 158142c2 | bellard | floatx80 floatx80_mul( floatx80 a, floatx80 b STATUS_PARAM ) |
4269 | 158142c2 | bellard | { |
4270 | 158142c2 | bellard | flag aSign, bSign, zSign; |
4271 | 158142c2 | bellard | int32 aExp, bExp, zExp; |
4272 | bb98fe42 | Andreas Färber | uint64_t aSig, bSig, zSig0, zSig1; |
4273 | 158142c2 | bellard | floatx80 z; |
4274 | 158142c2 | bellard | |
4275 | 158142c2 | bellard | aSig = extractFloatx80Frac( a ); |
4276 | 158142c2 | bellard | aExp = extractFloatx80Exp( a ); |
4277 | 158142c2 | bellard | aSign = extractFloatx80Sign( a ); |
4278 | 158142c2 | bellard | bSig = extractFloatx80Frac( b ); |
4279 | 158142c2 | bellard | bExp = extractFloatx80Exp( b ); |
4280 | 158142c2 | bellard | bSign = extractFloatx80Sign( b ); |
4281 | 158142c2 | bellard | zSign = aSign ^ bSign; |
4282 | 158142c2 | bellard | if ( aExp == 0x7FFF ) { |
4283 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( aSig<<1 ) |
4284 | bb98fe42 | Andreas Färber | || ( ( bExp == 0x7FFF ) && (uint64_t) ( bSig<<1 ) ) ) { |
4285 | 158142c2 | bellard | return propagateFloatx80NaN( a, b STATUS_VAR );
|
4286 | 158142c2 | bellard | } |
4287 | 158142c2 | bellard | if ( ( bExp | bSig ) == 0 ) goto invalid; |
4288 | 158142c2 | bellard | return packFloatx80( zSign, 0x7FFF, LIT64( 0x8000000000000000 ) ); |
4289 | 158142c2 | bellard | } |
4290 | 158142c2 | bellard | if ( bExp == 0x7FFF ) { |
4291 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( bSig<<1 ) ) return propagateFloatx80NaN( a, b STATUS_VAR ); |
4292 | 158142c2 | bellard | if ( ( aExp | aSig ) == 0 ) { |
4293 | 158142c2 | bellard | invalid:
|
4294 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
4295 | 158142c2 | bellard | z.low = floatx80_default_nan_low; |
4296 | 158142c2 | bellard | z.high = floatx80_default_nan_high; |
4297 | 158142c2 | bellard | return z;
|
4298 | 158142c2 | bellard | } |
4299 | 158142c2 | bellard | return packFloatx80( zSign, 0x7FFF, LIT64( 0x8000000000000000 ) ); |
4300 | 158142c2 | bellard | } |
4301 | 158142c2 | bellard | if ( aExp == 0 ) { |
4302 | 158142c2 | bellard | if ( aSig == 0 ) return packFloatx80( zSign, 0, 0 ); |
4303 | 158142c2 | bellard | normalizeFloatx80Subnormal( aSig, &aExp, &aSig ); |
4304 | 158142c2 | bellard | } |
4305 | 158142c2 | bellard | if ( bExp == 0 ) { |
4306 | 158142c2 | bellard | if ( bSig == 0 ) return packFloatx80( zSign, 0, 0 ); |
4307 | 158142c2 | bellard | normalizeFloatx80Subnormal( bSig, &bExp, &bSig ); |
4308 | 158142c2 | bellard | } |
4309 | 158142c2 | bellard | zExp = aExp + bExp - 0x3FFE;
|
4310 | 158142c2 | bellard | mul64To128( aSig, bSig, &zSig0, &zSig1 ); |
4311 | bb98fe42 | Andreas Färber | if ( 0 < (int64_t) zSig0 ) { |
4312 | 158142c2 | bellard | shortShift128Left( zSig0, zSig1, 1, &zSig0, &zSig1 );
|
4313 | 158142c2 | bellard | --zExp; |
4314 | 158142c2 | bellard | } |
4315 | 158142c2 | bellard | return
|
4316 | 158142c2 | bellard | roundAndPackFloatx80( |
4317 | 158142c2 | bellard | STATUS(floatx80_rounding_precision), zSign, zExp, zSig0, zSig1 STATUS_VAR ); |
4318 | 158142c2 | bellard | |
4319 | 158142c2 | bellard | } |
4320 | 158142c2 | bellard | |
4321 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
4322 | 158142c2 | bellard | | Returns the result of dividing the extended double-precision floating-point
|
4323 | 158142c2 | bellard | | value `a' by the corresponding value `b'. The operation is performed
|
4324 | 158142c2 | bellard | | according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
4325 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
4326 | 158142c2 | bellard | |
4327 | 158142c2 | bellard | floatx80 floatx80_div( floatx80 a, floatx80 b STATUS_PARAM ) |
4328 | 158142c2 | bellard | { |
4329 | 158142c2 | bellard | flag aSign, bSign, zSign; |
4330 | 158142c2 | bellard | int32 aExp, bExp, zExp; |
4331 | bb98fe42 | Andreas Färber | uint64_t aSig, bSig, zSig0, zSig1; |
4332 | bb98fe42 | Andreas Färber | uint64_t rem0, rem1, rem2, term0, term1, term2; |
4333 | 158142c2 | bellard | floatx80 z; |
4334 | 158142c2 | bellard | |
4335 | 158142c2 | bellard | aSig = extractFloatx80Frac( a ); |
4336 | 158142c2 | bellard | aExp = extractFloatx80Exp( a ); |
4337 | 158142c2 | bellard | aSign = extractFloatx80Sign( a ); |
4338 | 158142c2 | bellard | bSig = extractFloatx80Frac( b ); |
4339 | 158142c2 | bellard | bExp = extractFloatx80Exp( b ); |
4340 | 158142c2 | bellard | bSign = extractFloatx80Sign( b ); |
4341 | 158142c2 | bellard | zSign = aSign ^ bSign; |
4342 | 158142c2 | bellard | if ( aExp == 0x7FFF ) { |
4343 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( aSig<<1 ) ) return propagateFloatx80NaN( a, b STATUS_VAR ); |
4344 | 158142c2 | bellard | if ( bExp == 0x7FFF ) { |
4345 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( bSig<<1 ) ) return propagateFloatx80NaN( a, b STATUS_VAR ); |
4346 | 158142c2 | bellard | goto invalid;
|
4347 | 158142c2 | bellard | } |
4348 | 158142c2 | bellard | return packFloatx80( zSign, 0x7FFF, LIT64( 0x8000000000000000 ) ); |
4349 | 158142c2 | bellard | } |
4350 | 158142c2 | bellard | if ( bExp == 0x7FFF ) { |
4351 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( bSig<<1 ) ) return propagateFloatx80NaN( a, b STATUS_VAR ); |
4352 | 158142c2 | bellard | return packFloatx80( zSign, 0, 0 ); |
4353 | 158142c2 | bellard | } |
4354 | 158142c2 | bellard | if ( bExp == 0 ) { |
4355 | 158142c2 | bellard | if ( bSig == 0 ) { |
4356 | 158142c2 | bellard | if ( ( aExp | aSig ) == 0 ) { |
4357 | 158142c2 | bellard | invalid:
|
4358 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
4359 | 158142c2 | bellard | z.low = floatx80_default_nan_low; |
4360 | 158142c2 | bellard | z.high = floatx80_default_nan_high; |
4361 | 158142c2 | bellard | return z;
|
4362 | 158142c2 | bellard | } |
4363 | 158142c2 | bellard | float_raise( float_flag_divbyzero STATUS_VAR); |
4364 | 158142c2 | bellard | return packFloatx80( zSign, 0x7FFF, LIT64( 0x8000000000000000 ) ); |
4365 | 158142c2 | bellard | } |
4366 | 158142c2 | bellard | normalizeFloatx80Subnormal( bSig, &bExp, &bSig ); |
4367 | 158142c2 | bellard | } |
4368 | 158142c2 | bellard | if ( aExp == 0 ) { |
4369 | 158142c2 | bellard | if ( aSig == 0 ) return packFloatx80( zSign, 0, 0 ); |
4370 | 158142c2 | bellard | normalizeFloatx80Subnormal( aSig, &aExp, &aSig ); |
4371 | 158142c2 | bellard | } |
4372 | 158142c2 | bellard | zExp = aExp - bExp + 0x3FFE;
|
4373 | 158142c2 | bellard | rem1 = 0;
|
4374 | 158142c2 | bellard | if ( bSig <= aSig ) {
|
4375 | 158142c2 | bellard | shift128Right( aSig, 0, 1, &aSig, &rem1 ); |
4376 | 158142c2 | bellard | ++zExp; |
4377 | 158142c2 | bellard | } |
4378 | 158142c2 | bellard | zSig0 = estimateDiv128To64( aSig, rem1, bSig ); |
4379 | 158142c2 | bellard | mul64To128( bSig, zSig0, &term0, &term1 ); |
4380 | 158142c2 | bellard | sub128( aSig, rem1, term0, term1, &rem0, &rem1 ); |
4381 | bb98fe42 | Andreas Färber | while ( (int64_t) rem0 < 0 ) { |
4382 | 158142c2 | bellard | --zSig0; |
4383 | 158142c2 | bellard | add128( rem0, rem1, 0, bSig, &rem0, &rem1 );
|
4384 | 158142c2 | bellard | } |
4385 | 158142c2 | bellard | zSig1 = estimateDiv128To64( rem1, 0, bSig );
|
4386 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( zSig1<<1 ) <= 8 ) { |
4387 | 158142c2 | bellard | mul64To128( bSig, zSig1, &term1, &term2 ); |
4388 | 158142c2 | bellard | sub128( rem1, 0, term1, term2, &rem1, &rem2 );
|
4389 | bb98fe42 | Andreas Färber | while ( (int64_t) rem1 < 0 ) { |
4390 | 158142c2 | bellard | --zSig1; |
4391 | 158142c2 | bellard | add128( rem1, rem2, 0, bSig, &rem1, &rem2 );
|
4392 | 158142c2 | bellard | } |
4393 | 158142c2 | bellard | zSig1 |= ( ( rem1 | rem2 ) != 0 );
|
4394 | 158142c2 | bellard | } |
4395 | 158142c2 | bellard | return
|
4396 | 158142c2 | bellard | roundAndPackFloatx80( |
4397 | 158142c2 | bellard | STATUS(floatx80_rounding_precision), zSign, zExp, zSig0, zSig1 STATUS_VAR ); |
4398 | 158142c2 | bellard | |
4399 | 158142c2 | bellard | } |
4400 | 158142c2 | bellard | |
4401 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
4402 | 158142c2 | bellard | | Returns the remainder of the extended double-precision floating-point value
|
4403 | 158142c2 | bellard | | `a' with respect to the corresponding value `b'. The operation is performed
|
4404 | 158142c2 | bellard | | according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
4405 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
4406 | 158142c2 | bellard | |
4407 | 158142c2 | bellard | floatx80 floatx80_rem( floatx80 a, floatx80 b STATUS_PARAM ) |
4408 | 158142c2 | bellard | { |
4409 | ed086f3d | Blue Swirl | flag aSign, zSign; |
4410 | 158142c2 | bellard | int32 aExp, bExp, expDiff; |
4411 | bb98fe42 | Andreas Färber | uint64_t aSig0, aSig1, bSig; |
4412 | bb98fe42 | Andreas Färber | uint64_t q, term0, term1, alternateASig0, alternateASig1; |
4413 | 158142c2 | bellard | floatx80 z; |
4414 | 158142c2 | bellard | |
4415 | 158142c2 | bellard | aSig0 = extractFloatx80Frac( a ); |
4416 | 158142c2 | bellard | aExp = extractFloatx80Exp( a ); |
4417 | 158142c2 | bellard | aSign = extractFloatx80Sign( a ); |
4418 | 158142c2 | bellard | bSig = extractFloatx80Frac( b ); |
4419 | 158142c2 | bellard | bExp = extractFloatx80Exp( b ); |
4420 | 158142c2 | bellard | if ( aExp == 0x7FFF ) { |
4421 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( aSig0<<1 ) |
4422 | bb98fe42 | Andreas Färber | || ( ( bExp == 0x7FFF ) && (uint64_t) ( bSig<<1 ) ) ) { |
4423 | 158142c2 | bellard | return propagateFloatx80NaN( a, b STATUS_VAR );
|
4424 | 158142c2 | bellard | } |
4425 | 158142c2 | bellard | goto invalid;
|
4426 | 158142c2 | bellard | } |
4427 | 158142c2 | bellard | if ( bExp == 0x7FFF ) { |
4428 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( bSig<<1 ) ) return propagateFloatx80NaN( a, b STATUS_VAR ); |
4429 | 158142c2 | bellard | return a;
|
4430 | 158142c2 | bellard | } |
4431 | 158142c2 | bellard | if ( bExp == 0 ) { |
4432 | 158142c2 | bellard | if ( bSig == 0 ) { |
4433 | 158142c2 | bellard | invalid:
|
4434 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
4435 | 158142c2 | bellard | z.low = floatx80_default_nan_low; |
4436 | 158142c2 | bellard | z.high = floatx80_default_nan_high; |
4437 | 158142c2 | bellard | return z;
|
4438 | 158142c2 | bellard | } |
4439 | 158142c2 | bellard | normalizeFloatx80Subnormal( bSig, &bExp, &bSig ); |
4440 | 158142c2 | bellard | } |
4441 | 158142c2 | bellard | if ( aExp == 0 ) { |
4442 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( aSig0<<1 ) == 0 ) return a; |
4443 | 158142c2 | bellard | normalizeFloatx80Subnormal( aSig0, &aExp, &aSig0 ); |
4444 | 158142c2 | bellard | } |
4445 | 158142c2 | bellard | bSig |= LIT64( 0x8000000000000000 );
|
4446 | 158142c2 | bellard | zSign = aSign; |
4447 | 158142c2 | bellard | expDiff = aExp - bExp; |
4448 | 158142c2 | bellard | aSig1 = 0;
|
4449 | 158142c2 | bellard | if ( expDiff < 0 ) { |
4450 | 158142c2 | bellard | if ( expDiff < -1 ) return a; |
4451 | 158142c2 | bellard | shift128Right( aSig0, 0, 1, &aSig0, &aSig1 ); |
4452 | 158142c2 | bellard | expDiff = 0;
|
4453 | 158142c2 | bellard | } |
4454 | 158142c2 | bellard | q = ( bSig <= aSig0 ); |
4455 | 158142c2 | bellard | if ( q ) aSig0 -= bSig;
|
4456 | 158142c2 | bellard | expDiff -= 64;
|
4457 | 158142c2 | bellard | while ( 0 < expDiff ) { |
4458 | 158142c2 | bellard | q = estimateDiv128To64( aSig0, aSig1, bSig ); |
4459 | 158142c2 | bellard | q = ( 2 < q ) ? q - 2 : 0; |
4460 | 158142c2 | bellard | mul64To128( bSig, q, &term0, &term1 ); |
4461 | 158142c2 | bellard | sub128( aSig0, aSig1, term0, term1, &aSig0, &aSig1 ); |
4462 | 158142c2 | bellard | shortShift128Left( aSig0, aSig1, 62, &aSig0, &aSig1 );
|
4463 | 158142c2 | bellard | expDiff -= 62;
|
4464 | 158142c2 | bellard | } |
4465 | 158142c2 | bellard | expDiff += 64;
|
4466 | 158142c2 | bellard | if ( 0 < expDiff ) { |
4467 | 158142c2 | bellard | q = estimateDiv128To64( aSig0, aSig1, bSig ); |
4468 | 158142c2 | bellard | q = ( 2 < q ) ? q - 2 : 0; |
4469 | 158142c2 | bellard | q >>= 64 - expDiff;
|
4470 | 158142c2 | bellard | mul64To128( bSig, q<<( 64 - expDiff ), &term0, &term1 );
|
4471 | 158142c2 | bellard | sub128( aSig0, aSig1, term0, term1, &aSig0, &aSig1 ); |
4472 | 158142c2 | bellard | shortShift128Left( 0, bSig, 64 - expDiff, &term0, &term1 ); |
4473 | 158142c2 | bellard | while ( le128( term0, term1, aSig0, aSig1 ) ) {
|
4474 | 158142c2 | bellard | ++q; |
4475 | 158142c2 | bellard | sub128( aSig0, aSig1, term0, term1, &aSig0, &aSig1 ); |
4476 | 158142c2 | bellard | } |
4477 | 158142c2 | bellard | } |
4478 | 158142c2 | bellard | else {
|
4479 | 158142c2 | bellard | term1 = 0;
|
4480 | 158142c2 | bellard | term0 = bSig; |
4481 | 158142c2 | bellard | } |
4482 | 158142c2 | bellard | sub128( term0, term1, aSig0, aSig1, &alternateASig0, &alternateASig1 ); |
4483 | 158142c2 | bellard | if ( lt128( alternateASig0, alternateASig1, aSig0, aSig1 )
|
4484 | 158142c2 | bellard | || ( eq128( alternateASig0, alternateASig1, aSig0, aSig1 ) |
4485 | 158142c2 | bellard | && ( q & 1 ) )
|
4486 | 158142c2 | bellard | ) { |
4487 | 158142c2 | bellard | aSig0 = alternateASig0; |
4488 | 158142c2 | bellard | aSig1 = alternateASig1; |
4489 | 158142c2 | bellard | zSign = ! zSign; |
4490 | 158142c2 | bellard | } |
4491 | 158142c2 | bellard | return
|
4492 | 158142c2 | bellard | normalizeRoundAndPackFloatx80( |
4493 | 158142c2 | bellard | 80, zSign, bExp + expDiff, aSig0, aSig1 STATUS_VAR );
|
4494 | 158142c2 | bellard | |
4495 | 158142c2 | bellard | } |
4496 | 158142c2 | bellard | |
4497 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
4498 | 158142c2 | bellard | | Returns the square root of the extended double-precision floating-point
|
4499 | 158142c2 | bellard | | value `a'. The operation is performed according to the IEC/IEEE Standard
|
4500 | 158142c2 | bellard | | for Binary Floating-Point Arithmetic.
|
4501 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
4502 | 158142c2 | bellard | |
4503 | 158142c2 | bellard | floatx80 floatx80_sqrt( floatx80 a STATUS_PARAM ) |
4504 | 158142c2 | bellard | { |
4505 | 158142c2 | bellard | flag aSign; |
4506 | 158142c2 | bellard | int32 aExp, zExp; |
4507 | bb98fe42 | Andreas Färber | uint64_t aSig0, aSig1, zSig0, zSig1, doubleZSig0; |
4508 | bb98fe42 | Andreas Färber | uint64_t rem0, rem1, rem2, rem3, term0, term1, term2, term3; |
4509 | 158142c2 | bellard | floatx80 z; |
4510 | 158142c2 | bellard | |
4511 | 158142c2 | bellard | aSig0 = extractFloatx80Frac( a ); |
4512 | 158142c2 | bellard | aExp = extractFloatx80Exp( a ); |
4513 | 158142c2 | bellard | aSign = extractFloatx80Sign( a ); |
4514 | 158142c2 | bellard | if ( aExp == 0x7FFF ) { |
4515 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( aSig0<<1 ) ) return propagateFloatx80NaN( a, a STATUS_VAR ); |
4516 | 158142c2 | bellard | if ( ! aSign ) return a; |
4517 | 158142c2 | bellard | goto invalid;
|
4518 | 158142c2 | bellard | } |
4519 | 158142c2 | bellard | if ( aSign ) {
|
4520 | 158142c2 | bellard | if ( ( aExp | aSig0 ) == 0 ) return a; |
4521 | 158142c2 | bellard | invalid:
|
4522 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
4523 | 158142c2 | bellard | z.low = floatx80_default_nan_low; |
4524 | 158142c2 | bellard | z.high = floatx80_default_nan_high; |
4525 | 158142c2 | bellard | return z;
|
4526 | 158142c2 | bellard | } |
4527 | 158142c2 | bellard | if ( aExp == 0 ) { |
4528 | 158142c2 | bellard | if ( aSig0 == 0 ) return packFloatx80( 0, 0, 0 ); |
4529 | 158142c2 | bellard | normalizeFloatx80Subnormal( aSig0, &aExp, &aSig0 ); |
4530 | 158142c2 | bellard | } |
4531 | 158142c2 | bellard | zExp = ( ( aExp - 0x3FFF )>>1 ) + 0x3FFF; |
4532 | 158142c2 | bellard | zSig0 = estimateSqrt32( aExp, aSig0>>32 );
|
4533 | 158142c2 | bellard | shift128Right( aSig0, 0, 2 + ( aExp & 1 ), &aSig0, &aSig1 ); |
4534 | 158142c2 | bellard | zSig0 = estimateDiv128To64( aSig0, aSig1, zSig0<<32 ) + ( zSig0<<30 ); |
4535 | 158142c2 | bellard | doubleZSig0 = zSig0<<1;
|
4536 | 158142c2 | bellard | mul64To128( zSig0, zSig0, &term0, &term1 ); |
4537 | 158142c2 | bellard | sub128( aSig0, aSig1, term0, term1, &rem0, &rem1 ); |
4538 | bb98fe42 | Andreas Färber | while ( (int64_t) rem0 < 0 ) { |
4539 | 158142c2 | bellard | --zSig0; |
4540 | 158142c2 | bellard | doubleZSig0 -= 2;
|
4541 | 158142c2 | bellard | add128( rem0, rem1, zSig0>>63, doubleZSig0 | 1, &rem0, &rem1 ); |
4542 | 158142c2 | bellard | } |
4543 | 158142c2 | bellard | zSig1 = estimateDiv128To64( rem1, 0, doubleZSig0 );
|
4544 | 158142c2 | bellard | if ( ( zSig1 & LIT64( 0x3FFFFFFFFFFFFFFF ) ) <= 5 ) { |
4545 | 158142c2 | bellard | if ( zSig1 == 0 ) zSig1 = 1; |
4546 | 158142c2 | bellard | mul64To128( doubleZSig0, zSig1, &term1, &term2 ); |
4547 | 158142c2 | bellard | sub128( rem1, 0, term1, term2, &rem1, &rem2 );
|
4548 | 158142c2 | bellard | mul64To128( zSig1, zSig1, &term2, &term3 ); |
4549 | 158142c2 | bellard | sub192( rem1, rem2, 0, 0, term2, term3, &rem1, &rem2, &rem3 ); |
4550 | bb98fe42 | Andreas Färber | while ( (int64_t) rem1 < 0 ) { |
4551 | 158142c2 | bellard | --zSig1; |
4552 | 158142c2 | bellard | shortShift128Left( 0, zSig1, 1, &term2, &term3 ); |
4553 | 158142c2 | bellard | term3 |= 1;
|
4554 | 158142c2 | bellard | term2 |= doubleZSig0; |
4555 | 158142c2 | bellard | add192( rem1, rem2, rem3, 0, term2, term3, &rem1, &rem2, &rem3 );
|
4556 | 158142c2 | bellard | } |
4557 | 158142c2 | bellard | zSig1 |= ( ( rem1 | rem2 | rem3 ) != 0 );
|
4558 | 158142c2 | bellard | } |
4559 | 158142c2 | bellard | shortShift128Left( 0, zSig1, 1, &zSig0, &zSig1 ); |
4560 | 158142c2 | bellard | zSig0 |= doubleZSig0; |
4561 | 158142c2 | bellard | return
|
4562 | 158142c2 | bellard | roundAndPackFloatx80( |
4563 | 158142c2 | bellard | STATUS(floatx80_rounding_precision), 0, zExp, zSig0, zSig1 STATUS_VAR );
|
4564 | 158142c2 | bellard | |
4565 | 158142c2 | bellard | } |
4566 | 158142c2 | bellard | |
4567 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
4568 | b689362d | Aurelien Jarno | | Returns 1 if the extended double-precision floating-point value `a' is equal
|
4569 | b689362d | Aurelien Jarno | | to the corresponding value `b', and 0 otherwise. The invalid exception is
|
4570 | b689362d | Aurelien Jarno | | raised if either operand is a NaN. Otherwise, the comparison is performed
|
4571 | b689362d | Aurelien Jarno | | according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
4572 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
4573 | 158142c2 | bellard | |
4574 | b689362d | Aurelien Jarno | int floatx80_eq( floatx80 a, floatx80 b STATUS_PARAM )
|
4575 | 158142c2 | bellard | { |
4576 | 158142c2 | bellard | |
4577 | 158142c2 | bellard | if ( ( ( extractFloatx80Exp( a ) == 0x7FFF ) |
4578 | bb98fe42 | Andreas Färber | && (uint64_t) ( extractFloatx80Frac( a )<<1 ) )
|
4579 | 158142c2 | bellard | || ( ( extractFloatx80Exp( b ) == 0x7FFF )
|
4580 | bb98fe42 | Andreas Färber | && (uint64_t) ( extractFloatx80Frac( b )<<1 ) )
|
4581 | 158142c2 | bellard | ) { |
4582 | b689362d | Aurelien Jarno | float_raise( float_flag_invalid STATUS_VAR); |
4583 | 158142c2 | bellard | return 0; |
4584 | 158142c2 | bellard | } |
4585 | 158142c2 | bellard | return
|
4586 | 158142c2 | bellard | ( a.low == b.low ) |
4587 | 158142c2 | bellard | && ( ( a.high == b.high ) |
4588 | 158142c2 | bellard | || ( ( a.low == 0 )
|
4589 | bb98fe42 | Andreas Färber | && ( (uint16_t) ( ( a.high | b.high )<<1 ) == 0 ) ) |
4590 | 158142c2 | bellard | ); |
4591 | 158142c2 | bellard | |
4592 | 158142c2 | bellard | } |
4593 | 158142c2 | bellard | |
4594 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
4595 | 158142c2 | bellard | | Returns 1 if the extended double-precision floating-point value `a' is
|
4596 | 158142c2 | bellard | | less than or equal to the corresponding value `b', and 0 otherwise. The
|
4597 | f5a64251 | Aurelien Jarno | | invalid exception is raised if either operand is a NaN. The comparison is
|
4598 | f5a64251 | Aurelien Jarno | | performed according to the IEC/IEEE Standard for Binary Floating-Point
|
4599 | f5a64251 | Aurelien Jarno | | Arithmetic.
|
4600 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
4601 | 158142c2 | bellard | |
4602 | 750afe93 | bellard | int floatx80_le( floatx80 a, floatx80 b STATUS_PARAM )
|
4603 | 158142c2 | bellard | { |
4604 | 158142c2 | bellard | flag aSign, bSign; |
4605 | 158142c2 | bellard | |
4606 | 158142c2 | bellard | if ( ( ( extractFloatx80Exp( a ) == 0x7FFF ) |
4607 | bb98fe42 | Andreas Färber | && (uint64_t) ( extractFloatx80Frac( a )<<1 ) )
|
4608 | 158142c2 | bellard | || ( ( extractFloatx80Exp( b ) == 0x7FFF )
|
4609 | bb98fe42 | Andreas Färber | && (uint64_t) ( extractFloatx80Frac( b )<<1 ) )
|
4610 | 158142c2 | bellard | ) { |
4611 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
4612 | 158142c2 | bellard | return 0; |
4613 | 158142c2 | bellard | } |
4614 | 158142c2 | bellard | aSign = extractFloatx80Sign( a ); |
4615 | 158142c2 | bellard | bSign = extractFloatx80Sign( b ); |
4616 | 158142c2 | bellard | if ( aSign != bSign ) {
|
4617 | 158142c2 | bellard | return
|
4618 | 158142c2 | bellard | aSign |
4619 | bb98fe42 | Andreas Färber | || ( ( ( (uint16_t) ( ( a.high | b.high )<<1 ) ) | a.low | b.low )
|
4620 | 158142c2 | bellard | == 0 );
|
4621 | 158142c2 | bellard | } |
4622 | 158142c2 | bellard | return
|
4623 | 158142c2 | bellard | aSign ? le128( b.high, b.low, a.high, a.low ) |
4624 | 158142c2 | bellard | : le128( a.high, a.low, b.high, b.low ); |
4625 | 158142c2 | bellard | |
4626 | 158142c2 | bellard | } |
4627 | 158142c2 | bellard | |
4628 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
4629 | 158142c2 | bellard | | Returns 1 if the extended double-precision floating-point value `a' is
|
4630 | f5a64251 | Aurelien Jarno | | less than the corresponding value `b', and 0 otherwise. The invalid
|
4631 | f5a64251 | Aurelien Jarno | | exception is raised if either operand is a NaN. The comparison is performed
|
4632 | f5a64251 | Aurelien Jarno | | according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
4633 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
4634 | 158142c2 | bellard | |
4635 | 750afe93 | bellard | int floatx80_lt( floatx80 a, floatx80 b STATUS_PARAM )
|
4636 | 158142c2 | bellard | { |
4637 | 158142c2 | bellard | flag aSign, bSign; |
4638 | 158142c2 | bellard | |
4639 | 158142c2 | bellard | if ( ( ( extractFloatx80Exp( a ) == 0x7FFF ) |
4640 | bb98fe42 | Andreas Färber | && (uint64_t) ( extractFloatx80Frac( a )<<1 ) )
|
4641 | 158142c2 | bellard | || ( ( extractFloatx80Exp( b ) == 0x7FFF )
|
4642 | bb98fe42 | Andreas Färber | && (uint64_t) ( extractFloatx80Frac( b )<<1 ) )
|
4643 | 158142c2 | bellard | ) { |
4644 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
4645 | 158142c2 | bellard | return 0; |
4646 | 158142c2 | bellard | } |
4647 | 158142c2 | bellard | aSign = extractFloatx80Sign( a ); |
4648 | 158142c2 | bellard | bSign = extractFloatx80Sign( b ); |
4649 | 158142c2 | bellard | if ( aSign != bSign ) {
|
4650 | 158142c2 | bellard | return
|
4651 | 158142c2 | bellard | aSign |
4652 | bb98fe42 | Andreas Färber | && ( ( ( (uint16_t) ( ( a.high | b.high )<<1 ) ) | a.low | b.low )
|
4653 | 158142c2 | bellard | != 0 );
|
4654 | 158142c2 | bellard | } |
4655 | 158142c2 | bellard | return
|
4656 | 158142c2 | bellard | aSign ? lt128( b.high, b.low, a.high, a.low ) |
4657 | 158142c2 | bellard | : lt128( a.high, a.low, b.high, b.low ); |
4658 | 158142c2 | bellard | |
4659 | 158142c2 | bellard | } |
4660 | 158142c2 | bellard | |
4661 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
4662 | 67b7861d | Aurelien Jarno | | Returns 1 if the extended double-precision floating-point values `a' and `b'
|
4663 | f5a64251 | Aurelien Jarno | | cannot be compared, and 0 otherwise. The invalid exception is raised if
|
4664 | f5a64251 | Aurelien Jarno | | either operand is a NaN. The comparison is performed according to the
|
4665 | f5a64251 | Aurelien Jarno | | IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
4666 | 67b7861d | Aurelien Jarno | *----------------------------------------------------------------------------*/
|
4667 | 67b7861d | Aurelien Jarno | int floatx80_unordered( floatx80 a, floatx80 b STATUS_PARAM )
|
4668 | 67b7861d | Aurelien Jarno | { |
4669 | 67b7861d | Aurelien Jarno | if ( ( ( extractFloatx80Exp( a ) == 0x7FFF ) |
4670 | 67b7861d | Aurelien Jarno | && (uint64_t) ( extractFloatx80Frac( a )<<1 ) )
|
4671 | 67b7861d | Aurelien Jarno | || ( ( extractFloatx80Exp( b ) == 0x7FFF )
|
4672 | 67b7861d | Aurelien Jarno | && (uint64_t) ( extractFloatx80Frac( b )<<1 ) )
|
4673 | 67b7861d | Aurelien Jarno | ) { |
4674 | 67b7861d | Aurelien Jarno | float_raise( float_flag_invalid STATUS_VAR); |
4675 | 67b7861d | Aurelien Jarno | return 1; |
4676 | 67b7861d | Aurelien Jarno | } |
4677 | 67b7861d | Aurelien Jarno | return 0; |
4678 | 67b7861d | Aurelien Jarno | } |
4679 | 67b7861d | Aurelien Jarno | |
4680 | 67b7861d | Aurelien Jarno | /*----------------------------------------------------------------------------
|
4681 | b689362d | Aurelien Jarno | | Returns 1 if the extended double-precision floating-point value `a' is
|
4682 | f5a64251 | Aurelien Jarno | | equal to the corresponding value `b', and 0 otherwise. Quiet NaNs do not
|
4683 | f5a64251 | Aurelien Jarno | | cause an exception. The comparison is performed according to the IEC/IEEE
|
4684 | f5a64251 | Aurelien Jarno | | Standard for Binary Floating-Point Arithmetic.
|
4685 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
4686 | 158142c2 | bellard | |
4687 | b689362d | Aurelien Jarno | int floatx80_eq_quiet( floatx80 a, floatx80 b STATUS_PARAM )
|
4688 | 158142c2 | bellard | { |
4689 | 158142c2 | bellard | |
4690 | 158142c2 | bellard | if ( ( ( extractFloatx80Exp( a ) == 0x7FFF ) |
4691 | bb98fe42 | Andreas Färber | && (uint64_t) ( extractFloatx80Frac( a )<<1 ) )
|
4692 | 158142c2 | bellard | || ( ( extractFloatx80Exp( b ) == 0x7FFF )
|
4693 | bb98fe42 | Andreas Färber | && (uint64_t) ( extractFloatx80Frac( b )<<1 ) )
|
4694 | 158142c2 | bellard | ) { |
4695 | b689362d | Aurelien Jarno | if ( floatx80_is_signaling_nan( a )
|
4696 | b689362d | Aurelien Jarno | || floatx80_is_signaling_nan( b ) ) { |
4697 | b689362d | Aurelien Jarno | float_raise( float_flag_invalid STATUS_VAR); |
4698 | b689362d | Aurelien Jarno | } |
4699 | 158142c2 | bellard | return 0; |
4700 | 158142c2 | bellard | } |
4701 | 158142c2 | bellard | return
|
4702 | 158142c2 | bellard | ( a.low == b.low ) |
4703 | 158142c2 | bellard | && ( ( a.high == b.high ) |
4704 | 158142c2 | bellard | || ( ( a.low == 0 )
|
4705 | bb98fe42 | Andreas Färber | && ( (uint16_t) ( ( a.high | b.high )<<1 ) == 0 ) ) |
4706 | 158142c2 | bellard | ); |
4707 | 158142c2 | bellard | |
4708 | 158142c2 | bellard | } |
4709 | 158142c2 | bellard | |
4710 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
4711 | 158142c2 | bellard | | Returns 1 if the extended double-precision floating-point value `a' is less
|
4712 | 158142c2 | bellard | | than or equal to the corresponding value `b', and 0 otherwise. Quiet NaNs
|
4713 | 158142c2 | bellard | | do not cause an exception. Otherwise, the comparison is performed according
|
4714 | 158142c2 | bellard | | to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
4715 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
4716 | 158142c2 | bellard | |
4717 | 750afe93 | bellard | int floatx80_le_quiet( floatx80 a, floatx80 b STATUS_PARAM )
|
4718 | 158142c2 | bellard | { |
4719 | 158142c2 | bellard | flag aSign, bSign; |
4720 | 158142c2 | bellard | |
4721 | 158142c2 | bellard | if ( ( ( extractFloatx80Exp( a ) == 0x7FFF ) |
4722 | bb98fe42 | Andreas Färber | && (uint64_t) ( extractFloatx80Frac( a )<<1 ) )
|
4723 | 158142c2 | bellard | || ( ( extractFloatx80Exp( b ) == 0x7FFF )
|
4724 | bb98fe42 | Andreas Färber | && (uint64_t) ( extractFloatx80Frac( b )<<1 ) )
|
4725 | 158142c2 | bellard | ) { |
4726 | 158142c2 | bellard | if ( floatx80_is_signaling_nan( a )
|
4727 | 158142c2 | bellard | || floatx80_is_signaling_nan( b ) ) { |
4728 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
4729 | 158142c2 | bellard | } |
4730 | 158142c2 | bellard | return 0; |
4731 | 158142c2 | bellard | } |
4732 | 158142c2 | bellard | aSign = extractFloatx80Sign( a ); |
4733 | 158142c2 | bellard | bSign = extractFloatx80Sign( b ); |
4734 | 158142c2 | bellard | if ( aSign != bSign ) {
|
4735 | 158142c2 | bellard | return
|
4736 | 158142c2 | bellard | aSign |
4737 | bb98fe42 | Andreas Färber | || ( ( ( (uint16_t) ( ( a.high | b.high )<<1 ) ) | a.low | b.low )
|
4738 | 158142c2 | bellard | == 0 );
|
4739 | 158142c2 | bellard | } |
4740 | 158142c2 | bellard | return
|
4741 | 158142c2 | bellard | aSign ? le128( b.high, b.low, a.high, a.low ) |
4742 | 158142c2 | bellard | : le128( a.high, a.low, b.high, b.low ); |
4743 | 158142c2 | bellard | |
4744 | 158142c2 | bellard | } |
4745 | 158142c2 | bellard | |
4746 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
4747 | 158142c2 | bellard | | Returns 1 if the extended double-precision floating-point value `a' is less
|
4748 | 158142c2 | bellard | | than the corresponding value `b', and 0 otherwise. Quiet NaNs do not cause
|
4749 | 158142c2 | bellard | | an exception. Otherwise, the comparison is performed according to the
|
4750 | 158142c2 | bellard | | IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
4751 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
4752 | 158142c2 | bellard | |
4753 | 750afe93 | bellard | int floatx80_lt_quiet( floatx80 a, floatx80 b STATUS_PARAM )
|
4754 | 158142c2 | bellard | { |
4755 | 158142c2 | bellard | flag aSign, bSign; |
4756 | 158142c2 | bellard | |
4757 | 158142c2 | bellard | if ( ( ( extractFloatx80Exp( a ) == 0x7FFF ) |
4758 | bb98fe42 | Andreas Färber | && (uint64_t) ( extractFloatx80Frac( a )<<1 ) )
|
4759 | 158142c2 | bellard | || ( ( extractFloatx80Exp( b ) == 0x7FFF )
|
4760 | bb98fe42 | Andreas Färber | && (uint64_t) ( extractFloatx80Frac( b )<<1 ) )
|
4761 | 158142c2 | bellard | ) { |
4762 | 158142c2 | bellard | if ( floatx80_is_signaling_nan( a )
|
4763 | 158142c2 | bellard | || floatx80_is_signaling_nan( b ) ) { |
4764 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
4765 | 158142c2 | bellard | } |
4766 | 158142c2 | bellard | return 0; |
4767 | 158142c2 | bellard | } |
4768 | 158142c2 | bellard | aSign = extractFloatx80Sign( a ); |
4769 | 158142c2 | bellard | bSign = extractFloatx80Sign( b ); |
4770 | 158142c2 | bellard | if ( aSign != bSign ) {
|
4771 | 158142c2 | bellard | return
|
4772 | 158142c2 | bellard | aSign |
4773 | bb98fe42 | Andreas Färber | && ( ( ( (uint16_t) ( ( a.high | b.high )<<1 ) ) | a.low | b.low )
|
4774 | 158142c2 | bellard | != 0 );
|
4775 | 158142c2 | bellard | } |
4776 | 158142c2 | bellard | return
|
4777 | 158142c2 | bellard | aSign ? lt128( b.high, b.low, a.high, a.low ) |
4778 | 158142c2 | bellard | : lt128( a.high, a.low, b.high, b.low ); |
4779 | 158142c2 | bellard | |
4780 | 158142c2 | bellard | } |
4781 | 158142c2 | bellard | |
4782 | 67b7861d | Aurelien Jarno | /*----------------------------------------------------------------------------
|
4783 | 67b7861d | Aurelien Jarno | | Returns 1 if the extended double-precision floating-point values `a' and `b'
|
4784 | 67b7861d | Aurelien Jarno | | cannot be compared, and 0 otherwise. Quiet NaNs do not cause an exception.
|
4785 | 67b7861d | Aurelien Jarno | | The comparison is performed according to the IEC/IEEE Standard for Binary
|
4786 | 67b7861d | Aurelien Jarno | | Floating-Point Arithmetic.
|
4787 | 67b7861d | Aurelien Jarno | *----------------------------------------------------------------------------*/
|
4788 | 67b7861d | Aurelien Jarno | int floatx80_unordered_quiet( floatx80 a, floatx80 b STATUS_PARAM )
|
4789 | 67b7861d | Aurelien Jarno | { |
4790 | 67b7861d | Aurelien Jarno | if ( ( ( extractFloatx80Exp( a ) == 0x7FFF ) |
4791 | 67b7861d | Aurelien Jarno | && (uint64_t) ( extractFloatx80Frac( a )<<1 ) )
|
4792 | 67b7861d | Aurelien Jarno | || ( ( extractFloatx80Exp( b ) == 0x7FFF )
|
4793 | 67b7861d | Aurelien Jarno | && (uint64_t) ( extractFloatx80Frac( b )<<1 ) )
|
4794 | 67b7861d | Aurelien Jarno | ) { |
4795 | 67b7861d | Aurelien Jarno | if ( floatx80_is_signaling_nan( a )
|
4796 | 67b7861d | Aurelien Jarno | || floatx80_is_signaling_nan( b ) ) { |
4797 | 67b7861d | Aurelien Jarno | float_raise( float_flag_invalid STATUS_VAR); |
4798 | 67b7861d | Aurelien Jarno | } |
4799 | 67b7861d | Aurelien Jarno | return 1; |
4800 | 67b7861d | Aurelien Jarno | } |
4801 | 67b7861d | Aurelien Jarno | return 0; |
4802 | 67b7861d | Aurelien Jarno | } |
4803 | 67b7861d | Aurelien Jarno | |
4804 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
4805 | 158142c2 | bellard | | Returns the result of converting the quadruple-precision floating-point
|
4806 | 158142c2 | bellard | | value `a' to the 32-bit two's complement integer format. The conversion
|
4807 | 158142c2 | bellard | | is performed according to the IEC/IEEE Standard for Binary Floating-Point
|
4808 | 158142c2 | bellard | | Arithmetic---which means in particular that the conversion is rounded
|
4809 | 158142c2 | bellard | | according to the current rounding mode. If `a' is a NaN, the largest
|
4810 | 158142c2 | bellard | | positive integer is returned. Otherwise, if the conversion overflows, the
|
4811 | 158142c2 | bellard | | largest integer with the same sign as `a' is returned.
|
4812 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
4813 | 158142c2 | bellard | |
4814 | 158142c2 | bellard | int32 float128_to_int32( float128 a STATUS_PARAM ) |
4815 | 158142c2 | bellard | { |
4816 | 158142c2 | bellard | flag aSign; |
4817 | 158142c2 | bellard | int32 aExp, shiftCount; |
4818 | bb98fe42 | Andreas Färber | uint64_t aSig0, aSig1; |
4819 | 158142c2 | bellard | |
4820 | 158142c2 | bellard | aSig1 = extractFloat128Frac1( a ); |
4821 | 158142c2 | bellard | aSig0 = extractFloat128Frac0( a ); |
4822 | 158142c2 | bellard | aExp = extractFloat128Exp( a ); |
4823 | 158142c2 | bellard | aSign = extractFloat128Sign( a ); |
4824 | 158142c2 | bellard | if ( ( aExp == 0x7FFF ) && ( aSig0 | aSig1 ) ) aSign = 0; |
4825 | 158142c2 | bellard | if ( aExp ) aSig0 |= LIT64( 0x0001000000000000 ); |
4826 | 158142c2 | bellard | aSig0 |= ( aSig1 != 0 );
|
4827 | 158142c2 | bellard | shiftCount = 0x4028 - aExp;
|
4828 | 158142c2 | bellard | if ( 0 < shiftCount ) shift64RightJamming( aSig0, shiftCount, &aSig0 ); |
4829 | 158142c2 | bellard | return roundAndPackInt32( aSign, aSig0 STATUS_VAR );
|
4830 | 158142c2 | bellard | |
4831 | 158142c2 | bellard | } |
4832 | 158142c2 | bellard | |
4833 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
4834 | 158142c2 | bellard | | Returns the result of converting the quadruple-precision floating-point
|
4835 | 158142c2 | bellard | | value `a' to the 32-bit two's complement integer format. The conversion
|
4836 | 158142c2 | bellard | | is performed according to the IEC/IEEE Standard for Binary Floating-Point
|
4837 | 158142c2 | bellard | | Arithmetic, except that the conversion is always rounded toward zero. If
|
4838 | 158142c2 | bellard | | `a' is a NaN, the largest positive integer is returned. Otherwise, if the
|
4839 | 158142c2 | bellard | | conversion overflows, the largest integer with the same sign as `a' is
|
4840 | 158142c2 | bellard | | returned.
|
4841 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
4842 | 158142c2 | bellard | |
4843 | 158142c2 | bellard | int32 float128_to_int32_round_to_zero( float128 a STATUS_PARAM ) |
4844 | 158142c2 | bellard | { |
4845 | 158142c2 | bellard | flag aSign; |
4846 | 158142c2 | bellard | int32 aExp, shiftCount; |
4847 | bb98fe42 | Andreas Färber | uint64_t aSig0, aSig1, savedASig; |
4848 | 158142c2 | bellard | int32 z; |
4849 | 158142c2 | bellard | |
4850 | 158142c2 | bellard | aSig1 = extractFloat128Frac1( a ); |
4851 | 158142c2 | bellard | aSig0 = extractFloat128Frac0( a ); |
4852 | 158142c2 | bellard | aExp = extractFloat128Exp( a ); |
4853 | 158142c2 | bellard | aSign = extractFloat128Sign( a ); |
4854 | 158142c2 | bellard | aSig0 |= ( aSig1 != 0 );
|
4855 | 158142c2 | bellard | if ( 0x401E < aExp ) { |
4856 | 158142c2 | bellard | if ( ( aExp == 0x7FFF ) && aSig0 ) aSign = 0; |
4857 | 158142c2 | bellard | goto invalid;
|
4858 | 158142c2 | bellard | } |
4859 | 158142c2 | bellard | else if ( aExp < 0x3FFF ) { |
4860 | 158142c2 | bellard | if ( aExp || aSig0 ) STATUS(float_exception_flags) |= float_flag_inexact;
|
4861 | 158142c2 | bellard | return 0; |
4862 | 158142c2 | bellard | } |
4863 | 158142c2 | bellard | aSig0 |= LIT64( 0x0001000000000000 );
|
4864 | 158142c2 | bellard | shiftCount = 0x402F - aExp;
|
4865 | 158142c2 | bellard | savedASig = aSig0; |
4866 | 158142c2 | bellard | aSig0 >>= shiftCount; |
4867 | 158142c2 | bellard | z = aSig0; |
4868 | 158142c2 | bellard | if ( aSign ) z = - z;
|
4869 | 158142c2 | bellard | if ( ( z < 0 ) ^ aSign ) { |
4870 | 158142c2 | bellard | invalid:
|
4871 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
4872 | bb98fe42 | Andreas Färber | return aSign ? (int32_t) 0x80000000 : 0x7FFFFFFF; |
4873 | 158142c2 | bellard | } |
4874 | 158142c2 | bellard | if ( ( aSig0<<shiftCount ) != savedASig ) {
|
4875 | 158142c2 | bellard | STATUS(float_exception_flags) |= float_flag_inexact; |
4876 | 158142c2 | bellard | } |
4877 | 158142c2 | bellard | return z;
|
4878 | 158142c2 | bellard | |
4879 | 158142c2 | bellard | } |
4880 | 158142c2 | bellard | |
4881 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
4882 | 158142c2 | bellard | | Returns the result of converting the quadruple-precision floating-point
|
4883 | 158142c2 | bellard | | value `a' to the 64-bit two's complement integer format. The conversion
|
4884 | 158142c2 | bellard | | is performed according to the IEC/IEEE Standard for Binary Floating-Point
|
4885 | 158142c2 | bellard | | Arithmetic---which means in particular that the conversion is rounded
|
4886 | 158142c2 | bellard | | according to the current rounding mode. If `a' is a NaN, the largest
|
4887 | 158142c2 | bellard | | positive integer is returned. Otherwise, if the conversion overflows, the
|
4888 | 158142c2 | bellard | | largest integer with the same sign as `a' is returned.
|
4889 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
4890 | 158142c2 | bellard | |
4891 | 158142c2 | bellard | int64 float128_to_int64( float128 a STATUS_PARAM ) |
4892 | 158142c2 | bellard | { |
4893 | 158142c2 | bellard | flag aSign; |
4894 | 158142c2 | bellard | int32 aExp, shiftCount; |
4895 | bb98fe42 | Andreas Färber | uint64_t aSig0, aSig1; |
4896 | 158142c2 | bellard | |
4897 | 158142c2 | bellard | aSig1 = extractFloat128Frac1( a ); |
4898 | 158142c2 | bellard | aSig0 = extractFloat128Frac0( a ); |
4899 | 158142c2 | bellard | aExp = extractFloat128Exp( a ); |
4900 | 158142c2 | bellard | aSign = extractFloat128Sign( a ); |
4901 | 158142c2 | bellard | if ( aExp ) aSig0 |= LIT64( 0x0001000000000000 ); |
4902 | 158142c2 | bellard | shiftCount = 0x402F - aExp;
|
4903 | 158142c2 | bellard | if ( shiftCount <= 0 ) { |
4904 | 158142c2 | bellard | if ( 0x403E < aExp ) { |
4905 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
4906 | 158142c2 | bellard | if ( ! aSign
|
4907 | 158142c2 | bellard | || ( ( aExp == 0x7FFF )
|
4908 | 158142c2 | bellard | && ( aSig1 || ( aSig0 != LIT64( 0x0001000000000000 ) ) )
|
4909 | 158142c2 | bellard | ) |
4910 | 158142c2 | bellard | ) { |
4911 | 158142c2 | bellard | return LIT64( 0x7FFFFFFFFFFFFFFF ); |
4912 | 158142c2 | bellard | } |
4913 | bb98fe42 | Andreas Färber | return (int64_t) LIT64( 0x8000000000000000 ); |
4914 | 158142c2 | bellard | } |
4915 | 158142c2 | bellard | shortShift128Left( aSig0, aSig1, - shiftCount, &aSig0, &aSig1 ); |
4916 | 158142c2 | bellard | } |
4917 | 158142c2 | bellard | else {
|
4918 | 158142c2 | bellard | shift64ExtraRightJamming( aSig0, aSig1, shiftCount, &aSig0, &aSig1 ); |
4919 | 158142c2 | bellard | } |
4920 | 158142c2 | bellard | return roundAndPackInt64( aSign, aSig0, aSig1 STATUS_VAR );
|
4921 | 158142c2 | bellard | |
4922 | 158142c2 | bellard | } |
4923 | 158142c2 | bellard | |
4924 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
4925 | 158142c2 | bellard | | Returns the result of converting the quadruple-precision floating-point
|
4926 | 158142c2 | bellard | | value `a' to the 64-bit two's complement integer format. The conversion
|
4927 | 158142c2 | bellard | | is performed according to the IEC/IEEE Standard for Binary Floating-Point
|
4928 | 158142c2 | bellard | | Arithmetic, except that the conversion is always rounded toward zero.
|
4929 | 158142c2 | bellard | | If `a' is a NaN, the largest positive integer is returned. Otherwise, if
|
4930 | 158142c2 | bellard | | the conversion overflows, the largest integer with the same sign as `a' is
|
4931 | 158142c2 | bellard | | returned.
|
4932 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
4933 | 158142c2 | bellard | |
4934 | 158142c2 | bellard | int64 float128_to_int64_round_to_zero( float128 a STATUS_PARAM ) |
4935 | 158142c2 | bellard | { |
4936 | 158142c2 | bellard | flag aSign; |
4937 | 158142c2 | bellard | int32 aExp, shiftCount; |
4938 | bb98fe42 | Andreas Färber | uint64_t aSig0, aSig1; |
4939 | 158142c2 | bellard | int64 z; |
4940 | 158142c2 | bellard | |
4941 | 158142c2 | bellard | aSig1 = extractFloat128Frac1( a ); |
4942 | 158142c2 | bellard | aSig0 = extractFloat128Frac0( a ); |
4943 | 158142c2 | bellard | aExp = extractFloat128Exp( a ); |
4944 | 158142c2 | bellard | aSign = extractFloat128Sign( a ); |
4945 | 158142c2 | bellard | if ( aExp ) aSig0 |= LIT64( 0x0001000000000000 ); |
4946 | 158142c2 | bellard | shiftCount = aExp - 0x402F;
|
4947 | 158142c2 | bellard | if ( 0 < shiftCount ) { |
4948 | 158142c2 | bellard | if ( 0x403E <= aExp ) { |
4949 | 158142c2 | bellard | aSig0 &= LIT64( 0x0000FFFFFFFFFFFF );
|
4950 | 158142c2 | bellard | if ( ( a.high == LIT64( 0xC03E000000000000 ) ) |
4951 | 158142c2 | bellard | && ( aSig1 < LIT64( 0x0002000000000000 ) ) ) {
|
4952 | 158142c2 | bellard | if ( aSig1 ) STATUS(float_exception_flags) |= float_flag_inexact;
|
4953 | 158142c2 | bellard | } |
4954 | 158142c2 | bellard | else {
|
4955 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
4956 | 158142c2 | bellard | if ( ! aSign || ( ( aExp == 0x7FFF ) && ( aSig0 | aSig1 ) ) ) { |
4957 | 158142c2 | bellard | return LIT64( 0x7FFFFFFFFFFFFFFF ); |
4958 | 158142c2 | bellard | } |
4959 | 158142c2 | bellard | } |
4960 | bb98fe42 | Andreas Färber | return (int64_t) LIT64( 0x8000000000000000 ); |
4961 | 158142c2 | bellard | } |
4962 | 158142c2 | bellard | z = ( aSig0<<shiftCount ) | ( aSig1>>( ( - shiftCount ) & 63 ) );
|
4963 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( aSig1<<shiftCount ) ) {
|
4964 | 158142c2 | bellard | STATUS(float_exception_flags) |= float_flag_inexact; |
4965 | 158142c2 | bellard | } |
4966 | 158142c2 | bellard | } |
4967 | 158142c2 | bellard | else {
|
4968 | 158142c2 | bellard | if ( aExp < 0x3FFF ) { |
4969 | 158142c2 | bellard | if ( aExp | aSig0 | aSig1 ) {
|
4970 | 158142c2 | bellard | STATUS(float_exception_flags) |= float_flag_inexact; |
4971 | 158142c2 | bellard | } |
4972 | 158142c2 | bellard | return 0; |
4973 | 158142c2 | bellard | } |
4974 | 158142c2 | bellard | z = aSig0>>( - shiftCount ); |
4975 | 158142c2 | bellard | if ( aSig1
|
4976 | bb98fe42 | Andreas Färber | || ( shiftCount && (uint64_t) ( aSig0<<( shiftCount & 63 ) ) ) ) {
|
4977 | 158142c2 | bellard | STATUS(float_exception_flags) |= float_flag_inexact; |
4978 | 158142c2 | bellard | } |
4979 | 158142c2 | bellard | } |
4980 | 158142c2 | bellard | if ( aSign ) z = - z;
|
4981 | 158142c2 | bellard | return z;
|
4982 | 158142c2 | bellard | |
4983 | 158142c2 | bellard | } |
4984 | 158142c2 | bellard | |
4985 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
4986 | 158142c2 | bellard | | Returns the result of converting the quadruple-precision floating-point
|
4987 | 158142c2 | bellard | | value `a' to the single-precision floating-point format. The conversion
|
4988 | 158142c2 | bellard | | is performed according to the IEC/IEEE Standard for Binary Floating-Point
|
4989 | 158142c2 | bellard | | Arithmetic.
|
4990 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
4991 | 158142c2 | bellard | |
4992 | 158142c2 | bellard | float32 float128_to_float32( float128 a STATUS_PARAM ) |
4993 | 158142c2 | bellard | { |
4994 | 158142c2 | bellard | flag aSign; |
4995 | 158142c2 | bellard | int32 aExp; |
4996 | bb98fe42 | Andreas Färber | uint64_t aSig0, aSig1; |
4997 | bb98fe42 | Andreas Färber | uint32_t zSig; |
4998 | 158142c2 | bellard | |
4999 | 158142c2 | bellard | aSig1 = extractFloat128Frac1( a ); |
5000 | 158142c2 | bellard | aSig0 = extractFloat128Frac0( a ); |
5001 | 158142c2 | bellard | aExp = extractFloat128Exp( a ); |
5002 | 158142c2 | bellard | aSign = extractFloat128Sign( a ); |
5003 | 158142c2 | bellard | if ( aExp == 0x7FFF ) { |
5004 | 158142c2 | bellard | if ( aSig0 | aSig1 ) {
|
5005 | bcd4d9af | Christophe Lyon | return commonNaNToFloat32( float128ToCommonNaN( a STATUS_VAR ) STATUS_VAR );
|
5006 | 158142c2 | bellard | } |
5007 | 158142c2 | bellard | return packFloat32( aSign, 0xFF, 0 ); |
5008 | 158142c2 | bellard | } |
5009 | 158142c2 | bellard | aSig0 |= ( aSig1 != 0 );
|
5010 | 158142c2 | bellard | shift64RightJamming( aSig0, 18, &aSig0 );
|
5011 | 158142c2 | bellard | zSig = aSig0; |
5012 | 158142c2 | bellard | if ( aExp || zSig ) {
|
5013 | 158142c2 | bellard | zSig |= 0x40000000;
|
5014 | 158142c2 | bellard | aExp -= 0x3F81;
|
5015 | 158142c2 | bellard | } |
5016 | 158142c2 | bellard | return roundAndPackFloat32( aSign, aExp, zSig STATUS_VAR );
|
5017 | 158142c2 | bellard | |
5018 | 158142c2 | bellard | } |
5019 | 158142c2 | bellard | |
5020 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
5021 | 158142c2 | bellard | | Returns the result of converting the quadruple-precision floating-point
|
5022 | 158142c2 | bellard | | value `a' to the double-precision floating-point format. The conversion
|
5023 | 158142c2 | bellard | | is performed according to the IEC/IEEE Standard for Binary Floating-Point
|
5024 | 158142c2 | bellard | | Arithmetic.
|
5025 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
5026 | 158142c2 | bellard | |
5027 | 158142c2 | bellard | float64 float128_to_float64( float128 a STATUS_PARAM ) |
5028 | 158142c2 | bellard | { |
5029 | 158142c2 | bellard | flag aSign; |
5030 | 158142c2 | bellard | int32 aExp; |
5031 | bb98fe42 | Andreas Färber | uint64_t aSig0, aSig1; |
5032 | 158142c2 | bellard | |
5033 | 158142c2 | bellard | aSig1 = extractFloat128Frac1( a ); |
5034 | 158142c2 | bellard | aSig0 = extractFloat128Frac0( a ); |
5035 | 158142c2 | bellard | aExp = extractFloat128Exp( a ); |
5036 | 158142c2 | bellard | aSign = extractFloat128Sign( a ); |
5037 | 158142c2 | bellard | if ( aExp == 0x7FFF ) { |
5038 | 158142c2 | bellard | if ( aSig0 | aSig1 ) {
|
5039 | bcd4d9af | Christophe Lyon | return commonNaNToFloat64( float128ToCommonNaN( a STATUS_VAR ) STATUS_VAR );
|
5040 | 158142c2 | bellard | } |
5041 | 158142c2 | bellard | return packFloat64( aSign, 0x7FF, 0 ); |
5042 | 158142c2 | bellard | } |
5043 | 158142c2 | bellard | shortShift128Left( aSig0, aSig1, 14, &aSig0, &aSig1 );
|
5044 | 158142c2 | bellard | aSig0 |= ( aSig1 != 0 );
|
5045 | 158142c2 | bellard | if ( aExp || aSig0 ) {
|
5046 | 158142c2 | bellard | aSig0 |= LIT64( 0x4000000000000000 );
|
5047 | 158142c2 | bellard | aExp -= 0x3C01;
|
5048 | 158142c2 | bellard | } |
5049 | 158142c2 | bellard | return roundAndPackFloat64( aSign, aExp, aSig0 STATUS_VAR );
|
5050 | 158142c2 | bellard | |
5051 | 158142c2 | bellard | } |
5052 | 158142c2 | bellard | |
5053 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
5054 | 158142c2 | bellard | | Returns the result of converting the quadruple-precision floating-point
|
5055 | 158142c2 | bellard | | value `a' to the extended double-precision floating-point format. The
|
5056 | 158142c2 | bellard | | conversion is performed according to the IEC/IEEE Standard for Binary
|
5057 | 158142c2 | bellard | | Floating-Point Arithmetic.
|
5058 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
5059 | 158142c2 | bellard | |
5060 | 158142c2 | bellard | floatx80 float128_to_floatx80( float128 a STATUS_PARAM ) |
5061 | 158142c2 | bellard | { |
5062 | 158142c2 | bellard | flag aSign; |
5063 | 158142c2 | bellard | int32 aExp; |
5064 | bb98fe42 | Andreas Färber | uint64_t aSig0, aSig1; |
5065 | 158142c2 | bellard | |
5066 | 158142c2 | bellard | aSig1 = extractFloat128Frac1( a ); |
5067 | 158142c2 | bellard | aSig0 = extractFloat128Frac0( a ); |
5068 | 158142c2 | bellard | aExp = extractFloat128Exp( a ); |
5069 | 158142c2 | bellard | aSign = extractFloat128Sign( a ); |
5070 | 158142c2 | bellard | if ( aExp == 0x7FFF ) { |
5071 | 158142c2 | bellard | if ( aSig0 | aSig1 ) {
|
5072 | bcd4d9af | Christophe Lyon | return commonNaNToFloatx80( float128ToCommonNaN( a STATUS_VAR ) STATUS_VAR );
|
5073 | 158142c2 | bellard | } |
5074 | 158142c2 | bellard | return packFloatx80( aSign, 0x7FFF, LIT64( 0x8000000000000000 ) ); |
5075 | 158142c2 | bellard | } |
5076 | 158142c2 | bellard | if ( aExp == 0 ) { |
5077 | 158142c2 | bellard | if ( ( aSig0 | aSig1 ) == 0 ) return packFloatx80( aSign, 0, 0 ); |
5078 | 158142c2 | bellard | normalizeFloat128Subnormal( aSig0, aSig1, &aExp, &aSig0, &aSig1 ); |
5079 | 158142c2 | bellard | } |
5080 | 158142c2 | bellard | else {
|
5081 | 158142c2 | bellard | aSig0 |= LIT64( 0x0001000000000000 );
|
5082 | 158142c2 | bellard | } |
5083 | 158142c2 | bellard | shortShift128Left( aSig0, aSig1, 15, &aSig0, &aSig1 );
|
5084 | 158142c2 | bellard | return roundAndPackFloatx80( 80, aSign, aExp, aSig0, aSig1 STATUS_VAR ); |
5085 | 158142c2 | bellard | |
5086 | 158142c2 | bellard | } |
5087 | 158142c2 | bellard | |
5088 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
5089 | 158142c2 | bellard | | Rounds the quadruple-precision floating-point value `a' to an integer, and
|
5090 | 158142c2 | bellard | | returns the result as a quadruple-precision floating-point value. The
|
5091 | 158142c2 | bellard | | operation is performed according to the IEC/IEEE Standard for Binary
|
5092 | 158142c2 | bellard | | Floating-Point Arithmetic.
|
5093 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
5094 | 158142c2 | bellard | |
5095 | 158142c2 | bellard | float128 float128_round_to_int( float128 a STATUS_PARAM ) |
5096 | 158142c2 | bellard | { |
5097 | 158142c2 | bellard | flag aSign; |
5098 | 158142c2 | bellard | int32 aExp; |
5099 | bb98fe42 | Andreas Färber | uint64_t lastBitMask, roundBitsMask; |
5100 | 158142c2 | bellard | int8 roundingMode; |
5101 | 158142c2 | bellard | float128 z; |
5102 | 158142c2 | bellard | |
5103 | 158142c2 | bellard | aExp = extractFloat128Exp( a ); |
5104 | 158142c2 | bellard | if ( 0x402F <= aExp ) { |
5105 | 158142c2 | bellard | if ( 0x406F <= aExp ) { |
5106 | 158142c2 | bellard | if ( ( aExp == 0x7FFF ) |
5107 | 158142c2 | bellard | && ( extractFloat128Frac0( a ) | extractFloat128Frac1( a ) ) |
5108 | 158142c2 | bellard | ) { |
5109 | 158142c2 | bellard | return propagateFloat128NaN( a, a STATUS_VAR );
|
5110 | 158142c2 | bellard | } |
5111 | 158142c2 | bellard | return a;
|
5112 | 158142c2 | bellard | } |
5113 | 158142c2 | bellard | lastBitMask = 1;
|
5114 | 158142c2 | bellard | lastBitMask = ( lastBitMask<<( 0x406E - aExp ) )<<1; |
5115 | 158142c2 | bellard | roundBitsMask = lastBitMask - 1;
|
5116 | 158142c2 | bellard | z = a; |
5117 | 158142c2 | bellard | roundingMode = STATUS(float_rounding_mode); |
5118 | 158142c2 | bellard | if ( roundingMode == float_round_nearest_even ) {
|
5119 | 158142c2 | bellard | if ( lastBitMask ) {
|
5120 | 158142c2 | bellard | add128( z.high, z.low, 0, lastBitMask>>1, &z.high, &z.low ); |
5121 | 158142c2 | bellard | if ( ( z.low & roundBitsMask ) == 0 ) z.low &= ~ lastBitMask; |
5122 | 158142c2 | bellard | } |
5123 | 158142c2 | bellard | else {
|
5124 | bb98fe42 | Andreas Färber | if ( (int64_t) z.low < 0 ) { |
5125 | 158142c2 | bellard | ++z.high; |
5126 | bb98fe42 | Andreas Färber | if ( (uint64_t) ( z.low<<1 ) == 0 ) z.high &= ~1; |
5127 | 158142c2 | bellard | } |
5128 | 158142c2 | bellard | } |
5129 | 158142c2 | bellard | } |
5130 | 158142c2 | bellard | else if ( roundingMode != float_round_to_zero ) { |
5131 | 158142c2 | bellard | if ( extractFloat128Sign( z )
|
5132 | 158142c2 | bellard | ^ ( roundingMode == float_round_up ) ) { |
5133 | 158142c2 | bellard | add128( z.high, z.low, 0, roundBitsMask, &z.high, &z.low );
|
5134 | 158142c2 | bellard | } |
5135 | 158142c2 | bellard | } |
5136 | 158142c2 | bellard | z.low &= ~ roundBitsMask; |
5137 | 158142c2 | bellard | } |
5138 | 158142c2 | bellard | else {
|
5139 | 158142c2 | bellard | if ( aExp < 0x3FFF ) { |
5140 | bb98fe42 | Andreas Färber | if ( ( ( (uint64_t) ( a.high<<1 ) ) | a.low ) == 0 ) return a; |
5141 | 158142c2 | bellard | STATUS(float_exception_flags) |= float_flag_inexact; |
5142 | 158142c2 | bellard | aSign = extractFloat128Sign( a ); |
5143 | 158142c2 | bellard | switch ( STATUS(float_rounding_mode) ) {
|
5144 | 158142c2 | bellard | case float_round_nearest_even:
|
5145 | 158142c2 | bellard | if ( ( aExp == 0x3FFE ) |
5146 | 158142c2 | bellard | && ( extractFloat128Frac0( a ) |
5147 | 158142c2 | bellard | | extractFloat128Frac1( a ) ) |
5148 | 158142c2 | bellard | ) { |
5149 | 158142c2 | bellard | return packFloat128( aSign, 0x3FFF, 0, 0 ); |
5150 | 158142c2 | bellard | } |
5151 | 158142c2 | bellard | break;
|
5152 | 158142c2 | bellard | case float_round_down:
|
5153 | 158142c2 | bellard | return
|
5154 | 158142c2 | bellard | aSign ? packFloat128( 1, 0x3FFF, 0, 0 ) |
5155 | 158142c2 | bellard | : packFloat128( 0, 0, 0, 0 ); |
5156 | 158142c2 | bellard | case float_round_up:
|
5157 | 158142c2 | bellard | return
|
5158 | 158142c2 | bellard | aSign ? packFloat128( 1, 0, 0, 0 ) |
5159 | 158142c2 | bellard | : packFloat128( 0, 0x3FFF, 0, 0 ); |
5160 | 158142c2 | bellard | } |
5161 | 158142c2 | bellard | return packFloat128( aSign, 0, 0, 0 ); |
5162 | 158142c2 | bellard | } |
5163 | 158142c2 | bellard | lastBitMask = 1;
|
5164 | 158142c2 | bellard | lastBitMask <<= 0x402F - aExp;
|
5165 | 158142c2 | bellard | roundBitsMask = lastBitMask - 1;
|
5166 | 158142c2 | bellard | z.low = 0;
|
5167 | 158142c2 | bellard | z.high = a.high; |
5168 | 158142c2 | bellard | roundingMode = STATUS(float_rounding_mode); |
5169 | 158142c2 | bellard | if ( roundingMode == float_round_nearest_even ) {
|
5170 | 158142c2 | bellard | z.high += lastBitMask>>1;
|
5171 | 158142c2 | bellard | if ( ( ( z.high & roundBitsMask ) | a.low ) == 0 ) { |
5172 | 158142c2 | bellard | z.high &= ~ lastBitMask; |
5173 | 158142c2 | bellard | } |
5174 | 158142c2 | bellard | } |
5175 | 158142c2 | bellard | else if ( roundingMode != float_round_to_zero ) { |
5176 | 158142c2 | bellard | if ( extractFloat128Sign( z )
|
5177 | 158142c2 | bellard | ^ ( roundingMode == float_round_up ) ) { |
5178 | 158142c2 | bellard | z.high |= ( a.low != 0 );
|
5179 | 158142c2 | bellard | z.high += roundBitsMask; |
5180 | 158142c2 | bellard | } |
5181 | 158142c2 | bellard | } |
5182 | 158142c2 | bellard | z.high &= ~ roundBitsMask; |
5183 | 158142c2 | bellard | } |
5184 | 158142c2 | bellard | if ( ( z.low != a.low ) || ( z.high != a.high ) ) {
|
5185 | 158142c2 | bellard | STATUS(float_exception_flags) |= float_flag_inexact; |
5186 | 158142c2 | bellard | } |
5187 | 158142c2 | bellard | return z;
|
5188 | 158142c2 | bellard | |
5189 | 158142c2 | bellard | } |
5190 | 158142c2 | bellard | |
5191 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
5192 | 158142c2 | bellard | | Returns the result of adding the absolute values of the quadruple-precision
|
5193 | 158142c2 | bellard | | floating-point values `a' and `b'. If `zSign' is 1, the sum is negated
|
5194 | 158142c2 | bellard | | before being returned. `zSign' is ignored if the result is a NaN.
|
5195 | 158142c2 | bellard | | The addition is performed according to the IEC/IEEE Standard for Binary
|
5196 | 158142c2 | bellard | | Floating-Point Arithmetic.
|
5197 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
5198 | 158142c2 | bellard | |
5199 | 158142c2 | bellard | static float128 addFloat128Sigs( float128 a, float128 b, flag zSign STATUS_PARAM)
|
5200 | 158142c2 | bellard | { |
5201 | 158142c2 | bellard | int32 aExp, bExp, zExp; |
5202 | bb98fe42 | Andreas Färber | uint64_t aSig0, aSig1, bSig0, bSig1, zSig0, zSig1, zSig2; |
5203 | 158142c2 | bellard | int32 expDiff; |
5204 | 158142c2 | bellard | |
5205 | 158142c2 | bellard | aSig1 = extractFloat128Frac1( a ); |
5206 | 158142c2 | bellard | aSig0 = extractFloat128Frac0( a ); |
5207 | 158142c2 | bellard | aExp = extractFloat128Exp( a ); |
5208 | 158142c2 | bellard | bSig1 = extractFloat128Frac1( b ); |
5209 | 158142c2 | bellard | bSig0 = extractFloat128Frac0( b ); |
5210 | 158142c2 | bellard | bExp = extractFloat128Exp( b ); |
5211 | 158142c2 | bellard | expDiff = aExp - bExp; |
5212 | 158142c2 | bellard | if ( 0 < expDiff ) { |
5213 | 158142c2 | bellard | if ( aExp == 0x7FFF ) { |
5214 | 158142c2 | bellard | if ( aSig0 | aSig1 ) return propagateFloat128NaN( a, b STATUS_VAR ); |
5215 | 158142c2 | bellard | return a;
|
5216 | 158142c2 | bellard | } |
5217 | 158142c2 | bellard | if ( bExp == 0 ) { |
5218 | 158142c2 | bellard | --expDiff; |
5219 | 158142c2 | bellard | } |
5220 | 158142c2 | bellard | else {
|
5221 | 158142c2 | bellard | bSig0 |= LIT64( 0x0001000000000000 );
|
5222 | 158142c2 | bellard | } |
5223 | 158142c2 | bellard | shift128ExtraRightJamming( |
5224 | 158142c2 | bellard | bSig0, bSig1, 0, expDiff, &bSig0, &bSig1, &zSig2 );
|
5225 | 158142c2 | bellard | zExp = aExp; |
5226 | 158142c2 | bellard | } |
5227 | 158142c2 | bellard | else if ( expDiff < 0 ) { |
5228 | 158142c2 | bellard | if ( bExp == 0x7FFF ) { |
5229 | 158142c2 | bellard | if ( bSig0 | bSig1 ) return propagateFloat128NaN( a, b STATUS_VAR ); |
5230 | 158142c2 | bellard | return packFloat128( zSign, 0x7FFF, 0, 0 ); |
5231 | 158142c2 | bellard | } |
5232 | 158142c2 | bellard | if ( aExp == 0 ) { |
5233 | 158142c2 | bellard | ++expDiff; |
5234 | 158142c2 | bellard | } |
5235 | 158142c2 | bellard | else {
|
5236 | 158142c2 | bellard | aSig0 |= LIT64( 0x0001000000000000 );
|
5237 | 158142c2 | bellard | } |
5238 | 158142c2 | bellard | shift128ExtraRightJamming( |
5239 | 158142c2 | bellard | aSig0, aSig1, 0, - expDiff, &aSig0, &aSig1, &zSig2 );
|
5240 | 158142c2 | bellard | zExp = bExp; |
5241 | 158142c2 | bellard | } |
5242 | 158142c2 | bellard | else {
|
5243 | 158142c2 | bellard | if ( aExp == 0x7FFF ) { |
5244 | 158142c2 | bellard | if ( aSig0 | aSig1 | bSig0 | bSig1 ) {
|
5245 | 158142c2 | bellard | return propagateFloat128NaN( a, b STATUS_VAR );
|
5246 | 158142c2 | bellard | } |
5247 | 158142c2 | bellard | return a;
|
5248 | 158142c2 | bellard | } |
5249 | 158142c2 | bellard | add128( aSig0, aSig1, bSig0, bSig1, &zSig0, &zSig1 ); |
5250 | fe76d976 | pbrook | if ( aExp == 0 ) { |
5251 | e6afc87f | Peter Maydell | if (STATUS(flush_to_zero)) {
|
5252 | e6afc87f | Peter Maydell | if (zSig0 | zSig1) {
|
5253 | e6afc87f | Peter Maydell | float_raise(float_flag_output_denormal STATUS_VAR); |
5254 | e6afc87f | Peter Maydell | } |
5255 | e6afc87f | Peter Maydell | return packFloat128(zSign, 0, 0, 0); |
5256 | e6afc87f | Peter Maydell | } |
5257 | fe76d976 | pbrook | return packFloat128( zSign, 0, zSig0, zSig1 ); |
5258 | fe76d976 | pbrook | } |
5259 | 158142c2 | bellard | zSig2 = 0;
|
5260 | 158142c2 | bellard | zSig0 |= LIT64( 0x0002000000000000 );
|
5261 | 158142c2 | bellard | zExp = aExp; |
5262 | 158142c2 | bellard | goto shiftRight1;
|
5263 | 158142c2 | bellard | } |
5264 | 158142c2 | bellard | aSig0 |= LIT64( 0x0001000000000000 );
|
5265 | 158142c2 | bellard | add128( aSig0, aSig1, bSig0, bSig1, &zSig0, &zSig1 ); |
5266 | 158142c2 | bellard | --zExp; |
5267 | 158142c2 | bellard | if ( zSig0 < LIT64( 0x0002000000000000 ) ) goto roundAndPack; |
5268 | 158142c2 | bellard | ++zExp; |
5269 | 158142c2 | bellard | shiftRight1:
|
5270 | 158142c2 | bellard | shift128ExtraRightJamming( |
5271 | 158142c2 | bellard | zSig0, zSig1, zSig2, 1, &zSig0, &zSig1, &zSig2 );
|
5272 | 158142c2 | bellard | roundAndPack:
|
5273 | 158142c2 | bellard | return roundAndPackFloat128( zSign, zExp, zSig0, zSig1, zSig2 STATUS_VAR );
|
5274 | 158142c2 | bellard | |
5275 | 158142c2 | bellard | } |
5276 | 158142c2 | bellard | |
5277 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
5278 | 158142c2 | bellard | | Returns the result of subtracting the absolute values of the quadruple-
|
5279 | 158142c2 | bellard | | precision floating-point values `a' and `b'. If `zSign' is 1, the
|
5280 | 158142c2 | bellard | | difference is negated before being returned. `zSign' is ignored if the
|
5281 | 158142c2 | bellard | | result is a NaN. The subtraction is performed according to the IEC/IEEE
|
5282 | 158142c2 | bellard | | Standard for Binary Floating-Point Arithmetic.
|
5283 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
5284 | 158142c2 | bellard | |
5285 | 158142c2 | bellard | static float128 subFloat128Sigs( float128 a, float128 b, flag zSign STATUS_PARAM)
|
5286 | 158142c2 | bellard | { |
5287 | 158142c2 | bellard | int32 aExp, bExp, zExp; |
5288 | bb98fe42 | Andreas Färber | uint64_t aSig0, aSig1, bSig0, bSig1, zSig0, zSig1; |
5289 | 158142c2 | bellard | int32 expDiff; |
5290 | 158142c2 | bellard | float128 z; |
5291 | 158142c2 | bellard | |
5292 | 158142c2 | bellard | aSig1 = extractFloat128Frac1( a ); |
5293 | 158142c2 | bellard | aSig0 = extractFloat128Frac0( a ); |
5294 | 158142c2 | bellard | aExp = extractFloat128Exp( a ); |
5295 | 158142c2 | bellard | bSig1 = extractFloat128Frac1( b ); |
5296 | 158142c2 | bellard | bSig0 = extractFloat128Frac0( b ); |
5297 | 158142c2 | bellard | bExp = extractFloat128Exp( b ); |
5298 | 158142c2 | bellard | expDiff = aExp - bExp; |
5299 | 158142c2 | bellard | shortShift128Left( aSig0, aSig1, 14, &aSig0, &aSig1 );
|
5300 | 158142c2 | bellard | shortShift128Left( bSig0, bSig1, 14, &bSig0, &bSig1 );
|
5301 | 158142c2 | bellard | if ( 0 < expDiff ) goto aExpBigger; |
5302 | 158142c2 | bellard | if ( expDiff < 0 ) goto bExpBigger; |
5303 | 158142c2 | bellard | if ( aExp == 0x7FFF ) { |
5304 | 158142c2 | bellard | if ( aSig0 | aSig1 | bSig0 | bSig1 ) {
|
5305 | 158142c2 | bellard | return propagateFloat128NaN( a, b STATUS_VAR );
|
5306 | 158142c2 | bellard | } |
5307 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
5308 | 158142c2 | bellard | z.low = float128_default_nan_low; |
5309 | 158142c2 | bellard | z.high = float128_default_nan_high; |
5310 | 158142c2 | bellard | return z;
|
5311 | 158142c2 | bellard | } |
5312 | 158142c2 | bellard | if ( aExp == 0 ) { |
5313 | 158142c2 | bellard | aExp = 1;
|
5314 | 158142c2 | bellard | bExp = 1;
|
5315 | 158142c2 | bellard | } |
5316 | 158142c2 | bellard | if ( bSig0 < aSig0 ) goto aBigger; |
5317 | 158142c2 | bellard | if ( aSig0 < bSig0 ) goto bBigger; |
5318 | 158142c2 | bellard | if ( bSig1 < aSig1 ) goto aBigger; |
5319 | 158142c2 | bellard | if ( aSig1 < bSig1 ) goto bBigger; |
5320 | 158142c2 | bellard | return packFloat128( STATUS(float_rounding_mode) == float_round_down, 0, 0, 0 ); |
5321 | 158142c2 | bellard | bExpBigger:
|
5322 | 158142c2 | bellard | if ( bExp == 0x7FFF ) { |
5323 | 158142c2 | bellard | if ( bSig0 | bSig1 ) return propagateFloat128NaN( a, b STATUS_VAR ); |
5324 | 158142c2 | bellard | return packFloat128( zSign ^ 1, 0x7FFF, 0, 0 ); |
5325 | 158142c2 | bellard | } |
5326 | 158142c2 | bellard | if ( aExp == 0 ) { |
5327 | 158142c2 | bellard | ++expDiff; |
5328 | 158142c2 | bellard | } |
5329 | 158142c2 | bellard | else {
|
5330 | 158142c2 | bellard | aSig0 |= LIT64( 0x4000000000000000 );
|
5331 | 158142c2 | bellard | } |
5332 | 158142c2 | bellard | shift128RightJamming( aSig0, aSig1, - expDiff, &aSig0, &aSig1 ); |
5333 | 158142c2 | bellard | bSig0 |= LIT64( 0x4000000000000000 );
|
5334 | 158142c2 | bellard | bBigger:
|
5335 | 158142c2 | bellard | sub128( bSig0, bSig1, aSig0, aSig1, &zSig0, &zSig1 ); |
5336 | 158142c2 | bellard | zExp = bExp; |
5337 | 158142c2 | bellard | zSign ^= 1;
|
5338 | 158142c2 | bellard | goto normalizeRoundAndPack;
|
5339 | 158142c2 | bellard | aExpBigger:
|
5340 | 158142c2 | bellard | if ( aExp == 0x7FFF ) { |
5341 | 158142c2 | bellard | if ( aSig0 | aSig1 ) return propagateFloat128NaN( a, b STATUS_VAR ); |
5342 | 158142c2 | bellard | return a;
|
5343 | 158142c2 | bellard | } |
5344 | 158142c2 | bellard | if ( bExp == 0 ) { |
5345 | 158142c2 | bellard | --expDiff; |
5346 | 158142c2 | bellard | } |
5347 | 158142c2 | bellard | else {
|
5348 | 158142c2 | bellard | bSig0 |= LIT64( 0x4000000000000000 );
|
5349 | 158142c2 | bellard | } |
5350 | 158142c2 | bellard | shift128RightJamming( bSig0, bSig1, expDiff, &bSig0, &bSig1 ); |
5351 | 158142c2 | bellard | aSig0 |= LIT64( 0x4000000000000000 );
|
5352 | 158142c2 | bellard | aBigger:
|
5353 | 158142c2 | bellard | sub128( aSig0, aSig1, bSig0, bSig1, &zSig0, &zSig1 ); |
5354 | 158142c2 | bellard | zExp = aExp; |
5355 | 158142c2 | bellard | normalizeRoundAndPack:
|
5356 | 158142c2 | bellard | --zExp; |
5357 | 158142c2 | bellard | return normalizeRoundAndPackFloat128( zSign, zExp - 14, zSig0, zSig1 STATUS_VAR ); |
5358 | 158142c2 | bellard | |
5359 | 158142c2 | bellard | } |
5360 | 158142c2 | bellard | |
5361 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
5362 | 158142c2 | bellard | | Returns the result of adding the quadruple-precision floating-point values
|
5363 | 158142c2 | bellard | | `a' and `b'. The operation is performed according to the IEC/IEEE Standard
|
5364 | 158142c2 | bellard | | for Binary Floating-Point Arithmetic.
|
5365 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
5366 | 158142c2 | bellard | |
5367 | 158142c2 | bellard | float128 float128_add( float128 a, float128 b STATUS_PARAM ) |
5368 | 158142c2 | bellard | { |
5369 | 158142c2 | bellard | flag aSign, bSign; |
5370 | 158142c2 | bellard | |
5371 | 158142c2 | bellard | aSign = extractFloat128Sign( a ); |
5372 | 158142c2 | bellard | bSign = extractFloat128Sign( b ); |
5373 | 158142c2 | bellard | if ( aSign == bSign ) {
|
5374 | 158142c2 | bellard | return addFloat128Sigs( a, b, aSign STATUS_VAR );
|
5375 | 158142c2 | bellard | } |
5376 | 158142c2 | bellard | else {
|
5377 | 158142c2 | bellard | return subFloat128Sigs( a, b, aSign STATUS_VAR );
|
5378 | 158142c2 | bellard | } |
5379 | 158142c2 | bellard | |
5380 | 158142c2 | bellard | } |
5381 | 158142c2 | bellard | |
5382 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
5383 | 158142c2 | bellard | | Returns the result of subtracting the quadruple-precision floating-point
|
5384 | 158142c2 | bellard | | values `a' and `b'. The operation is performed according to the IEC/IEEE
|
5385 | 158142c2 | bellard | | Standard for Binary Floating-Point Arithmetic.
|
5386 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
5387 | 158142c2 | bellard | |
5388 | 158142c2 | bellard | float128 float128_sub( float128 a, float128 b STATUS_PARAM ) |
5389 | 158142c2 | bellard | { |
5390 | 158142c2 | bellard | flag aSign, bSign; |
5391 | 158142c2 | bellard | |
5392 | 158142c2 | bellard | aSign = extractFloat128Sign( a ); |
5393 | 158142c2 | bellard | bSign = extractFloat128Sign( b ); |
5394 | 158142c2 | bellard | if ( aSign == bSign ) {
|
5395 | 158142c2 | bellard | return subFloat128Sigs( a, b, aSign STATUS_VAR );
|
5396 | 158142c2 | bellard | } |
5397 | 158142c2 | bellard | else {
|
5398 | 158142c2 | bellard | return addFloat128Sigs( a, b, aSign STATUS_VAR );
|
5399 | 158142c2 | bellard | } |
5400 | 158142c2 | bellard | |
5401 | 158142c2 | bellard | } |
5402 | 158142c2 | bellard | |
5403 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
5404 | 158142c2 | bellard | | Returns the result of multiplying the quadruple-precision floating-point
|
5405 | 158142c2 | bellard | | values `a' and `b'. The operation is performed according to the IEC/IEEE
|
5406 | 158142c2 | bellard | | Standard for Binary Floating-Point Arithmetic.
|
5407 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
5408 | 158142c2 | bellard | |
5409 | 158142c2 | bellard | float128 float128_mul( float128 a, float128 b STATUS_PARAM ) |
5410 | 158142c2 | bellard | { |
5411 | 158142c2 | bellard | flag aSign, bSign, zSign; |
5412 | 158142c2 | bellard | int32 aExp, bExp, zExp; |
5413 | bb98fe42 | Andreas Färber | uint64_t aSig0, aSig1, bSig0, bSig1, zSig0, zSig1, zSig2, zSig3; |
5414 | 158142c2 | bellard | float128 z; |
5415 | 158142c2 | bellard | |
5416 | 158142c2 | bellard | aSig1 = extractFloat128Frac1( a ); |
5417 | 158142c2 | bellard | aSig0 = extractFloat128Frac0( a ); |
5418 | 158142c2 | bellard | aExp = extractFloat128Exp( a ); |
5419 | 158142c2 | bellard | aSign = extractFloat128Sign( a ); |
5420 | 158142c2 | bellard | bSig1 = extractFloat128Frac1( b ); |
5421 | 158142c2 | bellard | bSig0 = extractFloat128Frac0( b ); |
5422 | 158142c2 | bellard | bExp = extractFloat128Exp( b ); |
5423 | 158142c2 | bellard | bSign = extractFloat128Sign( b ); |
5424 | 158142c2 | bellard | zSign = aSign ^ bSign; |
5425 | 158142c2 | bellard | if ( aExp == 0x7FFF ) { |
5426 | 158142c2 | bellard | if ( ( aSig0 | aSig1 )
|
5427 | 158142c2 | bellard | || ( ( bExp == 0x7FFF ) && ( bSig0 | bSig1 ) ) ) {
|
5428 | 158142c2 | bellard | return propagateFloat128NaN( a, b STATUS_VAR );
|
5429 | 158142c2 | bellard | } |
5430 | 158142c2 | bellard | if ( ( bExp | bSig0 | bSig1 ) == 0 ) goto invalid; |
5431 | 158142c2 | bellard | return packFloat128( zSign, 0x7FFF, 0, 0 ); |
5432 | 158142c2 | bellard | } |
5433 | 158142c2 | bellard | if ( bExp == 0x7FFF ) { |
5434 | 158142c2 | bellard | if ( bSig0 | bSig1 ) return propagateFloat128NaN( a, b STATUS_VAR ); |
5435 | 158142c2 | bellard | if ( ( aExp | aSig0 | aSig1 ) == 0 ) { |
5436 | 158142c2 | bellard | invalid:
|
5437 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
5438 | 158142c2 | bellard | z.low = float128_default_nan_low; |
5439 | 158142c2 | bellard | z.high = float128_default_nan_high; |
5440 | 158142c2 | bellard | return z;
|
5441 | 158142c2 | bellard | } |
5442 | 158142c2 | bellard | return packFloat128( zSign, 0x7FFF, 0, 0 ); |
5443 | 158142c2 | bellard | } |
5444 | 158142c2 | bellard | if ( aExp == 0 ) { |
5445 | 158142c2 | bellard | if ( ( aSig0 | aSig1 ) == 0 ) return packFloat128( zSign, 0, 0, 0 ); |
5446 | 158142c2 | bellard | normalizeFloat128Subnormal( aSig0, aSig1, &aExp, &aSig0, &aSig1 ); |
5447 | 158142c2 | bellard | } |
5448 | 158142c2 | bellard | if ( bExp == 0 ) { |
5449 | 158142c2 | bellard | if ( ( bSig0 | bSig1 ) == 0 ) return packFloat128( zSign, 0, 0, 0 ); |
5450 | 158142c2 | bellard | normalizeFloat128Subnormal( bSig0, bSig1, &bExp, &bSig0, &bSig1 ); |
5451 | 158142c2 | bellard | } |
5452 | 158142c2 | bellard | zExp = aExp + bExp - 0x4000;
|
5453 | 158142c2 | bellard | aSig0 |= LIT64( 0x0001000000000000 );
|
5454 | 158142c2 | bellard | shortShift128Left( bSig0, bSig1, 16, &bSig0, &bSig1 );
|
5455 | 158142c2 | bellard | mul128To256( aSig0, aSig1, bSig0, bSig1, &zSig0, &zSig1, &zSig2, &zSig3 ); |
5456 | 158142c2 | bellard | add128( zSig0, zSig1, aSig0, aSig1, &zSig0, &zSig1 ); |
5457 | 158142c2 | bellard | zSig2 |= ( zSig3 != 0 );
|
5458 | 158142c2 | bellard | if ( LIT64( 0x0002000000000000 ) <= zSig0 ) { |
5459 | 158142c2 | bellard | shift128ExtraRightJamming( |
5460 | 158142c2 | bellard | zSig0, zSig1, zSig2, 1, &zSig0, &zSig1, &zSig2 );
|
5461 | 158142c2 | bellard | ++zExp; |
5462 | 158142c2 | bellard | } |
5463 | 158142c2 | bellard | return roundAndPackFloat128( zSign, zExp, zSig0, zSig1, zSig2 STATUS_VAR );
|
5464 | 158142c2 | bellard | |
5465 | 158142c2 | bellard | } |
5466 | 158142c2 | bellard | |
5467 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
5468 | 158142c2 | bellard | | Returns the result of dividing the quadruple-precision floating-point value
|
5469 | 158142c2 | bellard | | `a' by the corresponding value `b'. The operation is performed according to
|
5470 | 158142c2 | bellard | | the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
5471 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
5472 | 158142c2 | bellard | |
5473 | 158142c2 | bellard | float128 float128_div( float128 a, float128 b STATUS_PARAM ) |
5474 | 158142c2 | bellard | { |
5475 | 158142c2 | bellard | flag aSign, bSign, zSign; |
5476 | 158142c2 | bellard | int32 aExp, bExp, zExp; |
5477 | bb98fe42 | Andreas Färber | uint64_t aSig0, aSig1, bSig0, bSig1, zSig0, zSig1, zSig2; |
5478 | bb98fe42 | Andreas Färber | uint64_t rem0, rem1, rem2, rem3, term0, term1, term2, term3; |
5479 | 158142c2 | bellard | float128 z; |
5480 | 158142c2 | bellard | |
5481 | 158142c2 | bellard | aSig1 = extractFloat128Frac1( a ); |
5482 | 158142c2 | bellard | aSig0 = extractFloat128Frac0( a ); |
5483 | 158142c2 | bellard | aExp = extractFloat128Exp( a ); |
5484 | 158142c2 | bellard | aSign = extractFloat128Sign( a ); |
5485 | 158142c2 | bellard | bSig1 = extractFloat128Frac1( b ); |
5486 | 158142c2 | bellard | bSig0 = extractFloat128Frac0( b ); |
5487 | 158142c2 | bellard | bExp = extractFloat128Exp( b ); |
5488 | 158142c2 | bellard | bSign = extractFloat128Sign( b ); |
5489 | 158142c2 | bellard | zSign = aSign ^ bSign; |
5490 | 158142c2 | bellard | if ( aExp == 0x7FFF ) { |
5491 | 158142c2 | bellard | if ( aSig0 | aSig1 ) return propagateFloat128NaN( a, b STATUS_VAR ); |
5492 | 158142c2 | bellard | if ( bExp == 0x7FFF ) { |
5493 | 158142c2 | bellard | if ( bSig0 | bSig1 ) return propagateFloat128NaN( a, b STATUS_VAR ); |
5494 | 158142c2 | bellard | goto invalid;
|
5495 | 158142c2 | bellard | } |
5496 | 158142c2 | bellard | return packFloat128( zSign, 0x7FFF, 0, 0 ); |
5497 | 158142c2 | bellard | } |
5498 | 158142c2 | bellard | if ( bExp == 0x7FFF ) { |
5499 | 158142c2 | bellard | if ( bSig0 | bSig1 ) return propagateFloat128NaN( a, b STATUS_VAR ); |
5500 | 158142c2 | bellard | return packFloat128( zSign, 0, 0, 0 ); |
5501 | 158142c2 | bellard | } |
5502 | 158142c2 | bellard | if ( bExp == 0 ) { |
5503 | 158142c2 | bellard | if ( ( bSig0 | bSig1 ) == 0 ) { |
5504 | 158142c2 | bellard | if ( ( aExp | aSig0 | aSig1 ) == 0 ) { |
5505 | 158142c2 | bellard | invalid:
|
5506 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
5507 | 158142c2 | bellard | z.low = float128_default_nan_low; |
5508 | 158142c2 | bellard | z.high = float128_default_nan_high; |
5509 | 158142c2 | bellard | return z;
|
5510 | 158142c2 | bellard | } |
5511 | 158142c2 | bellard | float_raise( float_flag_divbyzero STATUS_VAR); |
5512 | 158142c2 | bellard | return packFloat128( zSign, 0x7FFF, 0, 0 ); |
5513 | 158142c2 | bellard | } |
5514 | 158142c2 | bellard | normalizeFloat128Subnormal( bSig0, bSig1, &bExp, &bSig0, &bSig1 ); |
5515 | 158142c2 | bellard | } |
5516 | 158142c2 | bellard | if ( aExp == 0 ) { |
5517 | 158142c2 | bellard | if ( ( aSig0 | aSig1 ) == 0 ) return packFloat128( zSign, 0, 0, 0 ); |
5518 | 158142c2 | bellard | normalizeFloat128Subnormal( aSig0, aSig1, &aExp, &aSig0, &aSig1 ); |
5519 | 158142c2 | bellard | } |
5520 | 158142c2 | bellard | zExp = aExp - bExp + 0x3FFD;
|
5521 | 158142c2 | bellard | shortShift128Left( |
5522 | 158142c2 | bellard | aSig0 | LIT64( 0x0001000000000000 ), aSig1, 15, &aSig0, &aSig1 ); |
5523 | 158142c2 | bellard | shortShift128Left( |
5524 | 158142c2 | bellard | bSig0 | LIT64( 0x0001000000000000 ), bSig1, 15, &bSig0, &bSig1 ); |
5525 | 158142c2 | bellard | if ( le128( bSig0, bSig1, aSig0, aSig1 ) ) {
|
5526 | 158142c2 | bellard | shift128Right( aSig0, aSig1, 1, &aSig0, &aSig1 );
|
5527 | 158142c2 | bellard | ++zExp; |
5528 | 158142c2 | bellard | } |
5529 | 158142c2 | bellard | zSig0 = estimateDiv128To64( aSig0, aSig1, bSig0 ); |
5530 | 158142c2 | bellard | mul128By64To192( bSig0, bSig1, zSig0, &term0, &term1, &term2 ); |
5531 | 158142c2 | bellard | sub192( aSig0, aSig1, 0, term0, term1, term2, &rem0, &rem1, &rem2 );
|
5532 | bb98fe42 | Andreas Färber | while ( (int64_t) rem0 < 0 ) { |
5533 | 158142c2 | bellard | --zSig0; |
5534 | 158142c2 | bellard | add192( rem0, rem1, rem2, 0, bSig0, bSig1, &rem0, &rem1, &rem2 );
|
5535 | 158142c2 | bellard | } |
5536 | 158142c2 | bellard | zSig1 = estimateDiv128To64( rem1, rem2, bSig0 ); |
5537 | 158142c2 | bellard | if ( ( zSig1 & 0x3FFF ) <= 4 ) { |
5538 | 158142c2 | bellard | mul128By64To192( bSig0, bSig1, zSig1, &term1, &term2, &term3 ); |
5539 | 158142c2 | bellard | sub192( rem1, rem2, 0, term1, term2, term3, &rem1, &rem2, &rem3 );
|
5540 | bb98fe42 | Andreas Färber | while ( (int64_t) rem1 < 0 ) { |
5541 | 158142c2 | bellard | --zSig1; |
5542 | 158142c2 | bellard | add192( rem1, rem2, rem3, 0, bSig0, bSig1, &rem1, &rem2, &rem3 );
|
5543 | 158142c2 | bellard | } |
5544 | 158142c2 | bellard | zSig1 |= ( ( rem1 | rem2 | rem3 ) != 0 );
|
5545 | 158142c2 | bellard | } |
5546 | 158142c2 | bellard | shift128ExtraRightJamming( zSig0, zSig1, 0, 15, &zSig0, &zSig1, &zSig2 ); |
5547 | 158142c2 | bellard | return roundAndPackFloat128( zSign, zExp, zSig0, zSig1, zSig2 STATUS_VAR );
|
5548 | 158142c2 | bellard | |
5549 | 158142c2 | bellard | } |
5550 | 158142c2 | bellard | |
5551 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
5552 | 158142c2 | bellard | | Returns the remainder of the quadruple-precision floating-point value `a'
|
5553 | 158142c2 | bellard | | with respect to the corresponding value `b'. The operation is performed
|
5554 | 158142c2 | bellard | | according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
5555 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
5556 | 158142c2 | bellard | |
5557 | 158142c2 | bellard | float128 float128_rem( float128 a, float128 b STATUS_PARAM ) |
5558 | 158142c2 | bellard | { |
5559 | ed086f3d | Blue Swirl | flag aSign, zSign; |
5560 | 158142c2 | bellard | int32 aExp, bExp, expDiff; |
5561 | bb98fe42 | Andreas Färber | uint64_t aSig0, aSig1, bSig0, bSig1, q, term0, term1, term2; |
5562 | bb98fe42 | Andreas Färber | uint64_t allZero, alternateASig0, alternateASig1, sigMean1; |
5563 | bb98fe42 | Andreas Färber | int64_t sigMean0; |
5564 | 158142c2 | bellard | float128 z; |
5565 | 158142c2 | bellard | |
5566 | 158142c2 | bellard | aSig1 = extractFloat128Frac1( a ); |
5567 | 158142c2 | bellard | aSig0 = extractFloat128Frac0( a ); |
5568 | 158142c2 | bellard | aExp = extractFloat128Exp( a ); |
5569 | 158142c2 | bellard | aSign = extractFloat128Sign( a ); |
5570 | 158142c2 | bellard | bSig1 = extractFloat128Frac1( b ); |
5571 | 158142c2 | bellard | bSig0 = extractFloat128Frac0( b ); |
5572 | 158142c2 | bellard | bExp = extractFloat128Exp( b ); |
5573 | 158142c2 | bellard | if ( aExp == 0x7FFF ) { |
5574 | 158142c2 | bellard | if ( ( aSig0 | aSig1 )
|
5575 | 158142c2 | bellard | || ( ( bExp == 0x7FFF ) && ( bSig0 | bSig1 ) ) ) {
|
5576 | 158142c2 | bellard | return propagateFloat128NaN( a, b STATUS_VAR );
|
5577 | 158142c2 | bellard | } |
5578 | 158142c2 | bellard | goto invalid;
|
5579 | 158142c2 | bellard | } |
5580 | 158142c2 | bellard | if ( bExp == 0x7FFF ) { |
5581 | 158142c2 | bellard | if ( bSig0 | bSig1 ) return propagateFloat128NaN( a, b STATUS_VAR ); |
5582 | 158142c2 | bellard | return a;
|
5583 | 158142c2 | bellard | } |
5584 | 158142c2 | bellard | if ( bExp == 0 ) { |
5585 | 158142c2 | bellard | if ( ( bSig0 | bSig1 ) == 0 ) { |
5586 | 158142c2 | bellard | invalid:
|
5587 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
5588 | 158142c2 | bellard | z.low = float128_default_nan_low; |
5589 | 158142c2 | bellard | z.high = float128_default_nan_high; |
5590 | 158142c2 | bellard | return z;
|
5591 | 158142c2 | bellard | } |
5592 | 158142c2 | bellard | normalizeFloat128Subnormal( bSig0, bSig1, &bExp, &bSig0, &bSig1 ); |
5593 | 158142c2 | bellard | } |
5594 | 158142c2 | bellard | if ( aExp == 0 ) { |
5595 | 158142c2 | bellard | if ( ( aSig0 | aSig1 ) == 0 ) return a; |
5596 | 158142c2 | bellard | normalizeFloat128Subnormal( aSig0, aSig1, &aExp, &aSig0, &aSig1 ); |
5597 | 158142c2 | bellard | } |
5598 | 158142c2 | bellard | expDiff = aExp - bExp; |
5599 | 158142c2 | bellard | if ( expDiff < -1 ) return a; |
5600 | 158142c2 | bellard | shortShift128Left( |
5601 | 158142c2 | bellard | aSig0 | LIT64( 0x0001000000000000 ),
|
5602 | 158142c2 | bellard | aSig1, |
5603 | 158142c2 | bellard | 15 - ( expDiff < 0 ), |
5604 | 158142c2 | bellard | &aSig0, |
5605 | 158142c2 | bellard | &aSig1 |
5606 | 158142c2 | bellard | ); |
5607 | 158142c2 | bellard | shortShift128Left( |
5608 | 158142c2 | bellard | bSig0 | LIT64( 0x0001000000000000 ), bSig1, 15, &bSig0, &bSig1 ); |
5609 | 158142c2 | bellard | q = le128( bSig0, bSig1, aSig0, aSig1 ); |
5610 | 158142c2 | bellard | if ( q ) sub128( aSig0, aSig1, bSig0, bSig1, &aSig0, &aSig1 );
|
5611 | 158142c2 | bellard | expDiff -= 64;
|
5612 | 158142c2 | bellard | while ( 0 < expDiff ) { |
5613 | 158142c2 | bellard | q = estimateDiv128To64( aSig0, aSig1, bSig0 ); |
5614 | 158142c2 | bellard | q = ( 4 < q ) ? q - 4 : 0; |
5615 | 158142c2 | bellard | mul128By64To192( bSig0, bSig1, q, &term0, &term1, &term2 ); |
5616 | 158142c2 | bellard | shortShift192Left( term0, term1, term2, 61, &term1, &term2, &allZero );
|
5617 | 158142c2 | bellard | shortShift128Left( aSig0, aSig1, 61, &aSig0, &allZero );
|
5618 | 158142c2 | bellard | sub128( aSig0, 0, term1, term2, &aSig0, &aSig1 );
|
5619 | 158142c2 | bellard | expDiff -= 61;
|
5620 | 158142c2 | bellard | } |
5621 | 158142c2 | bellard | if ( -64 < expDiff ) { |
5622 | 158142c2 | bellard | q = estimateDiv128To64( aSig0, aSig1, bSig0 ); |
5623 | 158142c2 | bellard | q = ( 4 < q ) ? q - 4 : 0; |
5624 | 158142c2 | bellard | q >>= - expDiff; |
5625 | 158142c2 | bellard | shift128Right( bSig0, bSig1, 12, &bSig0, &bSig1 );
|
5626 | 158142c2 | bellard | expDiff += 52;
|
5627 | 158142c2 | bellard | if ( expDiff < 0 ) { |
5628 | 158142c2 | bellard | shift128Right( aSig0, aSig1, - expDiff, &aSig0, &aSig1 ); |
5629 | 158142c2 | bellard | } |
5630 | 158142c2 | bellard | else {
|
5631 | 158142c2 | bellard | shortShift128Left( aSig0, aSig1, expDiff, &aSig0, &aSig1 ); |
5632 | 158142c2 | bellard | } |
5633 | 158142c2 | bellard | mul128By64To192( bSig0, bSig1, q, &term0, &term1, &term2 ); |
5634 | 158142c2 | bellard | sub128( aSig0, aSig1, term1, term2, &aSig0, &aSig1 ); |
5635 | 158142c2 | bellard | } |
5636 | 158142c2 | bellard | else {
|
5637 | 158142c2 | bellard | shift128Right( aSig0, aSig1, 12, &aSig0, &aSig1 );
|
5638 | 158142c2 | bellard | shift128Right( bSig0, bSig1, 12, &bSig0, &bSig1 );
|
5639 | 158142c2 | bellard | } |
5640 | 158142c2 | bellard | do {
|
5641 | 158142c2 | bellard | alternateASig0 = aSig0; |
5642 | 158142c2 | bellard | alternateASig1 = aSig1; |
5643 | 158142c2 | bellard | ++q; |
5644 | 158142c2 | bellard | sub128( aSig0, aSig1, bSig0, bSig1, &aSig0, &aSig1 ); |
5645 | bb98fe42 | Andreas Färber | } while ( 0 <= (int64_t) aSig0 ); |
5646 | 158142c2 | bellard | add128( |
5647 | bb98fe42 | Andreas Färber | aSig0, aSig1, alternateASig0, alternateASig1, (uint64_t *)&sigMean0, &sigMean1 ); |
5648 | 158142c2 | bellard | if ( ( sigMean0 < 0 ) |
5649 | 158142c2 | bellard | || ( ( ( sigMean0 | sigMean1 ) == 0 ) && ( q & 1 ) ) ) { |
5650 | 158142c2 | bellard | aSig0 = alternateASig0; |
5651 | 158142c2 | bellard | aSig1 = alternateASig1; |
5652 | 158142c2 | bellard | } |
5653 | bb98fe42 | Andreas Färber | zSign = ( (int64_t) aSig0 < 0 );
|
5654 | 158142c2 | bellard | if ( zSign ) sub128( 0, 0, aSig0, aSig1, &aSig0, &aSig1 ); |
5655 | 158142c2 | bellard | return
|
5656 | 158142c2 | bellard | normalizeRoundAndPackFloat128( aSign ^ zSign, bExp - 4, aSig0, aSig1 STATUS_VAR );
|
5657 | 158142c2 | bellard | |
5658 | 158142c2 | bellard | } |
5659 | 158142c2 | bellard | |
5660 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
5661 | 158142c2 | bellard | | Returns the square root of the quadruple-precision floating-point value `a'.
|
5662 | 158142c2 | bellard | | The operation is performed according to the IEC/IEEE Standard for Binary
|
5663 | 158142c2 | bellard | | Floating-Point Arithmetic.
|
5664 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
5665 | 158142c2 | bellard | |
5666 | 158142c2 | bellard | float128 float128_sqrt( float128 a STATUS_PARAM ) |
5667 | 158142c2 | bellard | { |
5668 | 158142c2 | bellard | flag aSign; |
5669 | 158142c2 | bellard | int32 aExp, zExp; |
5670 | bb98fe42 | Andreas Färber | uint64_t aSig0, aSig1, zSig0, zSig1, zSig2, doubleZSig0; |
5671 | bb98fe42 | Andreas Färber | uint64_t rem0, rem1, rem2, rem3, term0, term1, term2, term3; |
5672 | 158142c2 | bellard | float128 z; |
5673 | 158142c2 | bellard | |
5674 | 158142c2 | bellard | aSig1 = extractFloat128Frac1( a ); |
5675 | 158142c2 | bellard | aSig0 = extractFloat128Frac0( a ); |
5676 | 158142c2 | bellard | aExp = extractFloat128Exp( a ); |
5677 | 158142c2 | bellard | aSign = extractFloat128Sign( a ); |
5678 | 158142c2 | bellard | if ( aExp == 0x7FFF ) { |
5679 | 158142c2 | bellard | if ( aSig0 | aSig1 ) return propagateFloat128NaN( a, a STATUS_VAR ); |
5680 | 158142c2 | bellard | if ( ! aSign ) return a; |
5681 | 158142c2 | bellard | goto invalid;
|
5682 | 158142c2 | bellard | } |
5683 | 158142c2 | bellard | if ( aSign ) {
|
5684 | 158142c2 | bellard | if ( ( aExp | aSig0 | aSig1 ) == 0 ) return a; |
5685 | 158142c2 | bellard | invalid:
|
5686 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
5687 | 158142c2 | bellard | z.low = float128_default_nan_low; |
5688 | 158142c2 | bellard | z.high = float128_default_nan_high; |
5689 | 158142c2 | bellard | return z;
|
5690 | 158142c2 | bellard | } |
5691 | 158142c2 | bellard | if ( aExp == 0 ) { |
5692 | 158142c2 | bellard | if ( ( aSig0 | aSig1 ) == 0 ) return packFloat128( 0, 0, 0, 0 ); |
5693 | 158142c2 | bellard | normalizeFloat128Subnormal( aSig0, aSig1, &aExp, &aSig0, &aSig1 ); |
5694 | 158142c2 | bellard | } |
5695 | 158142c2 | bellard | zExp = ( ( aExp - 0x3FFF )>>1 ) + 0x3FFE; |
5696 | 158142c2 | bellard | aSig0 |= LIT64( 0x0001000000000000 );
|
5697 | 158142c2 | bellard | zSig0 = estimateSqrt32( aExp, aSig0>>17 );
|
5698 | 158142c2 | bellard | shortShift128Left( aSig0, aSig1, 13 - ( aExp & 1 ), &aSig0, &aSig1 ); |
5699 | 158142c2 | bellard | zSig0 = estimateDiv128To64( aSig0, aSig1, zSig0<<32 ) + ( zSig0<<30 ); |
5700 | 158142c2 | bellard | doubleZSig0 = zSig0<<1;
|
5701 | 158142c2 | bellard | mul64To128( zSig0, zSig0, &term0, &term1 ); |
5702 | 158142c2 | bellard | sub128( aSig0, aSig1, term0, term1, &rem0, &rem1 ); |
5703 | bb98fe42 | Andreas Färber | while ( (int64_t) rem0 < 0 ) { |
5704 | 158142c2 | bellard | --zSig0; |
5705 | 158142c2 | bellard | doubleZSig0 -= 2;
|
5706 | 158142c2 | bellard | add128( rem0, rem1, zSig0>>63, doubleZSig0 | 1, &rem0, &rem1 ); |
5707 | 158142c2 | bellard | } |
5708 | 158142c2 | bellard | zSig1 = estimateDiv128To64( rem1, 0, doubleZSig0 );
|
5709 | 158142c2 | bellard | if ( ( zSig1 & 0x1FFF ) <= 5 ) { |
5710 | 158142c2 | bellard | if ( zSig1 == 0 ) zSig1 = 1; |
5711 | 158142c2 | bellard | mul64To128( doubleZSig0, zSig1, &term1, &term2 ); |
5712 | 158142c2 | bellard | sub128( rem1, 0, term1, term2, &rem1, &rem2 );
|
5713 | 158142c2 | bellard | mul64To128( zSig1, zSig1, &term2, &term3 ); |
5714 | 158142c2 | bellard | sub192( rem1, rem2, 0, 0, term2, term3, &rem1, &rem2, &rem3 ); |
5715 | bb98fe42 | Andreas Färber | while ( (int64_t) rem1 < 0 ) { |
5716 | 158142c2 | bellard | --zSig1; |
5717 | 158142c2 | bellard | shortShift128Left( 0, zSig1, 1, &term2, &term3 ); |
5718 | 158142c2 | bellard | term3 |= 1;
|
5719 | 158142c2 | bellard | term2 |= doubleZSig0; |
5720 | 158142c2 | bellard | add192( rem1, rem2, rem3, 0, term2, term3, &rem1, &rem2, &rem3 );
|
5721 | 158142c2 | bellard | } |
5722 | 158142c2 | bellard | zSig1 |= ( ( rem1 | rem2 | rem3 ) != 0 );
|
5723 | 158142c2 | bellard | } |
5724 | 158142c2 | bellard | shift128ExtraRightJamming( zSig0, zSig1, 0, 14, &zSig0, &zSig1, &zSig2 ); |
5725 | 158142c2 | bellard | return roundAndPackFloat128( 0, zExp, zSig0, zSig1, zSig2 STATUS_VAR ); |
5726 | 158142c2 | bellard | |
5727 | 158142c2 | bellard | } |
5728 | 158142c2 | bellard | |
5729 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
5730 | 158142c2 | bellard | | Returns 1 if the quadruple-precision floating-point value `a' is equal to
|
5731 | b689362d | Aurelien Jarno | | the corresponding value `b', and 0 otherwise. The invalid exception is
|
5732 | b689362d | Aurelien Jarno | | raised if either operand is a NaN. Otherwise, the comparison is performed
|
5733 | 158142c2 | bellard | | according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
5734 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
5735 | 158142c2 | bellard | |
5736 | b689362d | Aurelien Jarno | int float128_eq( float128 a, float128 b STATUS_PARAM )
|
5737 | 158142c2 | bellard | { |
5738 | 158142c2 | bellard | |
5739 | 158142c2 | bellard | if ( ( ( extractFloat128Exp( a ) == 0x7FFF ) |
5740 | 158142c2 | bellard | && ( extractFloat128Frac0( a ) | extractFloat128Frac1( a ) ) ) |
5741 | 158142c2 | bellard | || ( ( extractFloat128Exp( b ) == 0x7FFF )
|
5742 | 158142c2 | bellard | && ( extractFloat128Frac0( b ) | extractFloat128Frac1( b ) ) ) |
5743 | 158142c2 | bellard | ) { |
5744 | b689362d | Aurelien Jarno | float_raise( float_flag_invalid STATUS_VAR); |
5745 | 158142c2 | bellard | return 0; |
5746 | 158142c2 | bellard | } |
5747 | 158142c2 | bellard | return
|
5748 | 158142c2 | bellard | ( a.low == b.low ) |
5749 | 158142c2 | bellard | && ( ( a.high == b.high ) |
5750 | 158142c2 | bellard | || ( ( a.low == 0 )
|
5751 | bb98fe42 | Andreas Färber | && ( (uint64_t) ( ( a.high | b.high )<<1 ) == 0 ) ) |
5752 | 158142c2 | bellard | ); |
5753 | 158142c2 | bellard | |
5754 | 158142c2 | bellard | } |
5755 | 158142c2 | bellard | |
5756 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
5757 | 158142c2 | bellard | | Returns 1 if the quadruple-precision floating-point value `a' is less than
|
5758 | f5a64251 | Aurelien Jarno | | or equal to the corresponding value `b', and 0 otherwise. The invalid
|
5759 | f5a64251 | Aurelien Jarno | | exception is raised if either operand is a NaN. The comparison is performed
|
5760 | f5a64251 | Aurelien Jarno | | according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
5761 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
5762 | 158142c2 | bellard | |
5763 | 750afe93 | bellard | int float128_le( float128 a, float128 b STATUS_PARAM )
|
5764 | 158142c2 | bellard | { |
5765 | 158142c2 | bellard | flag aSign, bSign; |
5766 | 158142c2 | bellard | |
5767 | 158142c2 | bellard | if ( ( ( extractFloat128Exp( a ) == 0x7FFF ) |
5768 | 158142c2 | bellard | && ( extractFloat128Frac0( a ) | extractFloat128Frac1( a ) ) ) |
5769 | 158142c2 | bellard | || ( ( extractFloat128Exp( b ) == 0x7FFF )
|
5770 | 158142c2 | bellard | && ( extractFloat128Frac0( b ) | extractFloat128Frac1( b ) ) ) |
5771 | 158142c2 | bellard | ) { |
5772 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
5773 | 158142c2 | bellard | return 0; |
5774 | 158142c2 | bellard | } |
5775 | 158142c2 | bellard | aSign = extractFloat128Sign( a ); |
5776 | 158142c2 | bellard | bSign = extractFloat128Sign( b ); |
5777 | 158142c2 | bellard | if ( aSign != bSign ) {
|
5778 | 158142c2 | bellard | return
|
5779 | 158142c2 | bellard | aSign |
5780 | bb98fe42 | Andreas Färber | || ( ( ( (uint64_t) ( ( a.high | b.high )<<1 ) ) | a.low | b.low )
|
5781 | 158142c2 | bellard | == 0 );
|
5782 | 158142c2 | bellard | } |
5783 | 158142c2 | bellard | return
|
5784 | 158142c2 | bellard | aSign ? le128( b.high, b.low, a.high, a.low ) |
5785 | 158142c2 | bellard | : le128( a.high, a.low, b.high, b.low ); |
5786 | 158142c2 | bellard | |
5787 | 158142c2 | bellard | } |
5788 | 158142c2 | bellard | |
5789 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
5790 | 158142c2 | bellard | | Returns 1 if the quadruple-precision floating-point value `a' is less than
|
5791 | f5a64251 | Aurelien Jarno | | the corresponding value `b', and 0 otherwise. The invalid exception is
|
5792 | f5a64251 | Aurelien Jarno | | raised if either operand is a NaN. The comparison is performed according
|
5793 | f5a64251 | Aurelien Jarno | | to the IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
5794 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
5795 | 158142c2 | bellard | |
5796 | 750afe93 | bellard | int float128_lt( float128 a, float128 b STATUS_PARAM )
|
5797 | 158142c2 | bellard | { |
5798 | 158142c2 | bellard | flag aSign, bSign; |
5799 | 158142c2 | bellard | |
5800 | 158142c2 | bellard | if ( ( ( extractFloat128Exp( a ) == 0x7FFF ) |
5801 | 158142c2 | bellard | && ( extractFloat128Frac0( a ) | extractFloat128Frac1( a ) ) ) |
5802 | 158142c2 | bellard | || ( ( extractFloat128Exp( b ) == 0x7FFF )
|
5803 | 158142c2 | bellard | && ( extractFloat128Frac0( b ) | extractFloat128Frac1( b ) ) ) |
5804 | 158142c2 | bellard | ) { |
5805 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
5806 | 158142c2 | bellard | return 0; |
5807 | 158142c2 | bellard | } |
5808 | 158142c2 | bellard | aSign = extractFloat128Sign( a ); |
5809 | 158142c2 | bellard | bSign = extractFloat128Sign( b ); |
5810 | 158142c2 | bellard | if ( aSign != bSign ) {
|
5811 | 158142c2 | bellard | return
|
5812 | 158142c2 | bellard | aSign |
5813 | bb98fe42 | Andreas Färber | && ( ( ( (uint64_t) ( ( a.high | b.high )<<1 ) ) | a.low | b.low )
|
5814 | 158142c2 | bellard | != 0 );
|
5815 | 158142c2 | bellard | } |
5816 | 158142c2 | bellard | return
|
5817 | 158142c2 | bellard | aSign ? lt128( b.high, b.low, a.high, a.low ) |
5818 | 158142c2 | bellard | : lt128( a.high, a.low, b.high, b.low ); |
5819 | 158142c2 | bellard | |
5820 | 158142c2 | bellard | } |
5821 | 158142c2 | bellard | |
5822 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
5823 | 67b7861d | Aurelien Jarno | | Returns 1 if the quadruple-precision floating-point values `a' and `b' cannot
|
5824 | f5a64251 | Aurelien Jarno | | be compared, and 0 otherwise. The invalid exception is raised if either
|
5825 | f5a64251 | Aurelien Jarno | | operand is a NaN. The comparison is performed according to the IEC/IEEE
|
5826 | f5a64251 | Aurelien Jarno | | Standard for Binary Floating-Point Arithmetic.
|
5827 | 67b7861d | Aurelien Jarno | *----------------------------------------------------------------------------*/
|
5828 | 67b7861d | Aurelien Jarno | |
5829 | 67b7861d | Aurelien Jarno | int float128_unordered( float128 a, float128 b STATUS_PARAM )
|
5830 | 67b7861d | Aurelien Jarno | { |
5831 | 67b7861d | Aurelien Jarno | if ( ( ( extractFloat128Exp( a ) == 0x7FFF ) |
5832 | 67b7861d | Aurelien Jarno | && ( extractFloat128Frac0( a ) | extractFloat128Frac1( a ) ) ) |
5833 | 67b7861d | Aurelien Jarno | || ( ( extractFloat128Exp( b ) == 0x7FFF )
|
5834 | 67b7861d | Aurelien Jarno | && ( extractFloat128Frac0( b ) | extractFloat128Frac1( b ) ) ) |
5835 | 67b7861d | Aurelien Jarno | ) { |
5836 | 67b7861d | Aurelien Jarno | float_raise( float_flag_invalid STATUS_VAR); |
5837 | 67b7861d | Aurelien Jarno | return 1; |
5838 | 67b7861d | Aurelien Jarno | } |
5839 | 67b7861d | Aurelien Jarno | return 0; |
5840 | 67b7861d | Aurelien Jarno | } |
5841 | 67b7861d | Aurelien Jarno | |
5842 | 67b7861d | Aurelien Jarno | /*----------------------------------------------------------------------------
|
5843 | 158142c2 | bellard | | Returns 1 if the quadruple-precision floating-point value `a' is equal to
|
5844 | f5a64251 | Aurelien Jarno | | the corresponding value `b', and 0 otherwise. Quiet NaNs do not cause an
|
5845 | f5a64251 | Aurelien Jarno | | exception. The comparison is performed according to the IEC/IEEE Standard
|
5846 | f5a64251 | Aurelien Jarno | | for Binary Floating-Point Arithmetic.
|
5847 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
5848 | 158142c2 | bellard | |
5849 | b689362d | Aurelien Jarno | int float128_eq_quiet( float128 a, float128 b STATUS_PARAM )
|
5850 | 158142c2 | bellard | { |
5851 | 158142c2 | bellard | |
5852 | 158142c2 | bellard | if ( ( ( extractFloat128Exp( a ) == 0x7FFF ) |
5853 | 158142c2 | bellard | && ( extractFloat128Frac0( a ) | extractFloat128Frac1( a ) ) ) |
5854 | 158142c2 | bellard | || ( ( extractFloat128Exp( b ) == 0x7FFF )
|
5855 | 158142c2 | bellard | && ( extractFloat128Frac0( b ) | extractFloat128Frac1( b ) ) ) |
5856 | 158142c2 | bellard | ) { |
5857 | b689362d | Aurelien Jarno | if ( float128_is_signaling_nan( a )
|
5858 | b689362d | Aurelien Jarno | || float128_is_signaling_nan( b ) ) { |
5859 | b689362d | Aurelien Jarno | float_raise( float_flag_invalid STATUS_VAR); |
5860 | b689362d | Aurelien Jarno | } |
5861 | 158142c2 | bellard | return 0; |
5862 | 158142c2 | bellard | } |
5863 | 158142c2 | bellard | return
|
5864 | 158142c2 | bellard | ( a.low == b.low ) |
5865 | 158142c2 | bellard | && ( ( a.high == b.high ) |
5866 | 158142c2 | bellard | || ( ( a.low == 0 )
|
5867 | bb98fe42 | Andreas Färber | && ( (uint64_t) ( ( a.high | b.high )<<1 ) == 0 ) ) |
5868 | 158142c2 | bellard | ); |
5869 | 158142c2 | bellard | |
5870 | 158142c2 | bellard | } |
5871 | 158142c2 | bellard | |
5872 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
5873 | 158142c2 | bellard | | Returns 1 if the quadruple-precision floating-point value `a' is less than
|
5874 | 158142c2 | bellard | | or equal to the corresponding value `b', and 0 otherwise. Quiet NaNs do not
|
5875 | 158142c2 | bellard | | cause an exception. Otherwise, the comparison is performed according to the
|
5876 | 158142c2 | bellard | | IEC/IEEE Standard for Binary Floating-Point Arithmetic.
|
5877 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
5878 | 158142c2 | bellard | |
5879 | 750afe93 | bellard | int float128_le_quiet( float128 a, float128 b STATUS_PARAM )
|
5880 | 158142c2 | bellard | { |
5881 | 158142c2 | bellard | flag aSign, bSign; |
5882 | 158142c2 | bellard | |
5883 | 158142c2 | bellard | if ( ( ( extractFloat128Exp( a ) == 0x7FFF ) |
5884 | 158142c2 | bellard | && ( extractFloat128Frac0( a ) | extractFloat128Frac1( a ) ) ) |
5885 | 158142c2 | bellard | || ( ( extractFloat128Exp( b ) == 0x7FFF )
|
5886 | 158142c2 | bellard | && ( extractFloat128Frac0( b ) | extractFloat128Frac1( b ) ) ) |
5887 | 158142c2 | bellard | ) { |
5888 | 158142c2 | bellard | if ( float128_is_signaling_nan( a )
|
5889 | 158142c2 | bellard | || float128_is_signaling_nan( b ) ) { |
5890 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
5891 | 158142c2 | bellard | } |
5892 | 158142c2 | bellard | return 0; |
5893 | 158142c2 | bellard | } |
5894 | 158142c2 | bellard | aSign = extractFloat128Sign( a ); |
5895 | 158142c2 | bellard | bSign = extractFloat128Sign( b ); |
5896 | 158142c2 | bellard | if ( aSign != bSign ) {
|
5897 | 158142c2 | bellard | return
|
5898 | 158142c2 | bellard | aSign |
5899 | bb98fe42 | Andreas Färber | || ( ( ( (uint64_t) ( ( a.high | b.high )<<1 ) ) | a.low | b.low )
|
5900 | 158142c2 | bellard | == 0 );
|
5901 | 158142c2 | bellard | } |
5902 | 158142c2 | bellard | return
|
5903 | 158142c2 | bellard | aSign ? le128( b.high, b.low, a.high, a.low ) |
5904 | 158142c2 | bellard | : le128( a.high, a.low, b.high, b.low ); |
5905 | 158142c2 | bellard | |
5906 | 158142c2 | bellard | } |
5907 | 158142c2 | bellard | |
5908 | 158142c2 | bellard | /*----------------------------------------------------------------------------
|
5909 | 158142c2 | bellard | | Returns 1 if the quadruple-precision floating-point value `a' is less than
|
5910 | 158142c2 | bellard | | the corresponding value `b', and 0 otherwise. Quiet NaNs do not cause an
|
5911 | 158142c2 | bellard | | exception. Otherwise, the comparison is performed according to the IEC/IEEE
|
5912 | 158142c2 | bellard | | Standard for Binary Floating-Point Arithmetic.
|
5913 | 158142c2 | bellard | *----------------------------------------------------------------------------*/
|
5914 | 158142c2 | bellard | |
5915 | 750afe93 | bellard | int float128_lt_quiet( float128 a, float128 b STATUS_PARAM )
|
5916 | 158142c2 | bellard | { |
5917 | 158142c2 | bellard | flag aSign, bSign; |
5918 | 158142c2 | bellard | |
5919 | 158142c2 | bellard | if ( ( ( extractFloat128Exp( a ) == 0x7FFF ) |
5920 | 158142c2 | bellard | && ( extractFloat128Frac0( a ) | extractFloat128Frac1( a ) ) ) |
5921 | 158142c2 | bellard | || ( ( extractFloat128Exp( b ) == 0x7FFF )
|
5922 | 158142c2 | bellard | && ( extractFloat128Frac0( b ) | extractFloat128Frac1( b ) ) ) |
5923 | 158142c2 | bellard | ) { |
5924 | 158142c2 | bellard | if ( float128_is_signaling_nan( a )
|
5925 | 158142c2 | bellard | || float128_is_signaling_nan( b ) ) { |
5926 | 158142c2 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
5927 | 158142c2 | bellard | } |
5928 | 158142c2 | bellard | return 0; |
5929 | 158142c2 | bellard | } |
5930 | 158142c2 | bellard | aSign = extractFloat128Sign( a ); |
5931 | 158142c2 | bellard | bSign = extractFloat128Sign( b ); |
5932 | 158142c2 | bellard | if ( aSign != bSign ) {
|
5933 | 158142c2 | bellard | return
|
5934 | 158142c2 | bellard | aSign |
5935 | bb98fe42 | Andreas Färber | && ( ( ( (uint64_t) ( ( a.high | b.high )<<1 ) ) | a.low | b.low )
|
5936 | 158142c2 | bellard | != 0 );
|
5937 | 158142c2 | bellard | } |
5938 | 158142c2 | bellard | return
|
5939 | 158142c2 | bellard | aSign ? lt128( b.high, b.low, a.high, a.low ) |
5940 | 158142c2 | bellard | : lt128( a.high, a.low, b.high, b.low ); |
5941 | 158142c2 | bellard | |
5942 | 158142c2 | bellard | } |
5943 | 158142c2 | bellard | |
5944 | 67b7861d | Aurelien Jarno | /*----------------------------------------------------------------------------
|
5945 | 67b7861d | Aurelien Jarno | | Returns 1 if the quadruple-precision floating-point values `a' and `b' cannot
|
5946 | 67b7861d | Aurelien Jarno | | be compared, and 0 otherwise. Quiet NaNs do not cause an exception. The
|
5947 | 67b7861d | Aurelien Jarno | | comparison is performed according to the IEC/IEEE Standard for Binary
|
5948 | 67b7861d | Aurelien Jarno | | Floating-Point Arithmetic.
|
5949 | 67b7861d | Aurelien Jarno | *----------------------------------------------------------------------------*/
|
5950 | 67b7861d | Aurelien Jarno | |
5951 | 67b7861d | Aurelien Jarno | int float128_unordered_quiet( float128 a, float128 b STATUS_PARAM )
|
5952 | 67b7861d | Aurelien Jarno | { |
5953 | 67b7861d | Aurelien Jarno | if ( ( ( extractFloat128Exp( a ) == 0x7FFF ) |
5954 | 67b7861d | Aurelien Jarno | && ( extractFloat128Frac0( a ) | extractFloat128Frac1( a ) ) ) |
5955 | 67b7861d | Aurelien Jarno | || ( ( extractFloat128Exp( b ) == 0x7FFF )
|
5956 | 67b7861d | Aurelien Jarno | && ( extractFloat128Frac0( b ) | extractFloat128Frac1( b ) ) ) |
5957 | 67b7861d | Aurelien Jarno | ) { |
5958 | 67b7861d | Aurelien Jarno | if ( float128_is_signaling_nan( a )
|
5959 | 67b7861d | Aurelien Jarno | || float128_is_signaling_nan( b ) ) { |
5960 | 67b7861d | Aurelien Jarno | float_raise( float_flag_invalid STATUS_VAR); |
5961 | 67b7861d | Aurelien Jarno | } |
5962 | 67b7861d | Aurelien Jarno | return 1; |
5963 | 67b7861d | Aurelien Jarno | } |
5964 | 67b7861d | Aurelien Jarno | return 0; |
5965 | 67b7861d | Aurelien Jarno | } |
5966 | 67b7861d | Aurelien Jarno | |
5967 | 1d6bda35 | bellard | /* misc functions */
|
5968 | 1d6bda35 | bellard | float32 uint32_to_float32( unsigned int a STATUS_PARAM ) |
5969 | 1d6bda35 | bellard | { |
5970 | 1d6bda35 | bellard | return int64_to_float32(a STATUS_VAR);
|
5971 | 1d6bda35 | bellard | } |
5972 | 1d6bda35 | bellard | |
5973 | 1d6bda35 | bellard | float64 uint32_to_float64( unsigned int a STATUS_PARAM ) |
5974 | 1d6bda35 | bellard | { |
5975 | 1d6bda35 | bellard | return int64_to_float64(a STATUS_VAR);
|
5976 | 1d6bda35 | bellard | } |
5977 | 1d6bda35 | bellard | |
5978 | 1d6bda35 | bellard | unsigned int float32_to_uint32( float32 a STATUS_PARAM ) |
5979 | 1d6bda35 | bellard | { |
5980 | 1d6bda35 | bellard | int64_t v; |
5981 | 1d6bda35 | bellard | unsigned int res; |
5982 | 1d6bda35 | bellard | |
5983 | 1d6bda35 | bellard | v = float32_to_int64(a STATUS_VAR); |
5984 | 1d6bda35 | bellard | if (v < 0) { |
5985 | 1d6bda35 | bellard | res = 0;
|
5986 | 1d6bda35 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
5987 | 1d6bda35 | bellard | } else if (v > 0xffffffff) { |
5988 | 1d6bda35 | bellard | res = 0xffffffff;
|
5989 | 1d6bda35 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
5990 | 1d6bda35 | bellard | } else {
|
5991 | 1d6bda35 | bellard | res = v; |
5992 | 1d6bda35 | bellard | } |
5993 | 1d6bda35 | bellard | return res;
|
5994 | 1d6bda35 | bellard | } |
5995 | 1d6bda35 | bellard | |
5996 | 1d6bda35 | bellard | unsigned int float32_to_uint32_round_to_zero( float32 a STATUS_PARAM ) |
5997 | 1d6bda35 | bellard | { |
5998 | 1d6bda35 | bellard | int64_t v; |
5999 | 1d6bda35 | bellard | unsigned int res; |
6000 | 1d6bda35 | bellard | |
6001 | 1d6bda35 | bellard | v = float32_to_int64_round_to_zero(a STATUS_VAR); |
6002 | 1d6bda35 | bellard | if (v < 0) { |
6003 | 1d6bda35 | bellard | res = 0;
|
6004 | 1d6bda35 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
6005 | 1d6bda35 | bellard | } else if (v > 0xffffffff) { |
6006 | 1d6bda35 | bellard | res = 0xffffffff;
|
6007 | 1d6bda35 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
6008 | 1d6bda35 | bellard | } else {
|
6009 | 1d6bda35 | bellard | res = v; |
6010 | 1d6bda35 | bellard | } |
6011 | 1d6bda35 | bellard | return res;
|
6012 | 1d6bda35 | bellard | } |
6013 | 1d6bda35 | bellard | |
6014 | cbcef455 | Peter Maydell | unsigned int float32_to_uint16_round_to_zero( float32 a STATUS_PARAM ) |
6015 | cbcef455 | Peter Maydell | { |
6016 | cbcef455 | Peter Maydell | int64_t v; |
6017 | cbcef455 | Peter Maydell | unsigned int res; |
6018 | cbcef455 | Peter Maydell | |
6019 | cbcef455 | Peter Maydell | v = float32_to_int64_round_to_zero(a STATUS_VAR); |
6020 | cbcef455 | Peter Maydell | if (v < 0) { |
6021 | cbcef455 | Peter Maydell | res = 0;
|
6022 | cbcef455 | Peter Maydell | float_raise( float_flag_invalid STATUS_VAR); |
6023 | cbcef455 | Peter Maydell | } else if (v > 0xffff) { |
6024 | cbcef455 | Peter Maydell | res = 0xffff;
|
6025 | cbcef455 | Peter Maydell | float_raise( float_flag_invalid STATUS_VAR); |
6026 | cbcef455 | Peter Maydell | } else {
|
6027 | cbcef455 | Peter Maydell | res = v; |
6028 | cbcef455 | Peter Maydell | } |
6029 | cbcef455 | Peter Maydell | return res;
|
6030 | cbcef455 | Peter Maydell | } |
6031 | cbcef455 | Peter Maydell | |
6032 | 1d6bda35 | bellard | unsigned int float64_to_uint32( float64 a STATUS_PARAM ) |
6033 | 1d6bda35 | bellard | { |
6034 | 1d6bda35 | bellard | int64_t v; |
6035 | 1d6bda35 | bellard | unsigned int res; |
6036 | 1d6bda35 | bellard | |
6037 | 1d6bda35 | bellard | v = float64_to_int64(a STATUS_VAR); |
6038 | 1d6bda35 | bellard | if (v < 0) { |
6039 | 1d6bda35 | bellard | res = 0;
|
6040 | 1d6bda35 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
6041 | 1d6bda35 | bellard | } else if (v > 0xffffffff) { |
6042 | 1d6bda35 | bellard | res = 0xffffffff;
|
6043 | 1d6bda35 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
6044 | 1d6bda35 | bellard | } else {
|
6045 | 1d6bda35 | bellard | res = v; |
6046 | 1d6bda35 | bellard | } |
6047 | 1d6bda35 | bellard | return res;
|
6048 | 1d6bda35 | bellard | } |
6049 | 1d6bda35 | bellard | |
6050 | 1d6bda35 | bellard | unsigned int float64_to_uint32_round_to_zero( float64 a STATUS_PARAM ) |
6051 | 1d6bda35 | bellard | { |
6052 | 1d6bda35 | bellard | int64_t v; |
6053 | 1d6bda35 | bellard | unsigned int res; |
6054 | 1d6bda35 | bellard | |
6055 | 1d6bda35 | bellard | v = float64_to_int64_round_to_zero(a STATUS_VAR); |
6056 | 1d6bda35 | bellard | if (v < 0) { |
6057 | 1d6bda35 | bellard | res = 0;
|
6058 | 1d6bda35 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
6059 | 1d6bda35 | bellard | } else if (v > 0xffffffff) { |
6060 | 1d6bda35 | bellard | res = 0xffffffff;
|
6061 | 1d6bda35 | bellard | float_raise( float_flag_invalid STATUS_VAR); |
6062 | 1d6bda35 | bellard | } else {
|
6063 | 1d6bda35 | bellard | res = v; |
6064 | 1d6bda35 | bellard | } |
6065 | 1d6bda35 | bellard | return res;
|
6066 | 1d6bda35 | bellard | } |
6067 | 1d6bda35 | bellard | |
6068 | cbcef455 | Peter Maydell | unsigned int float64_to_uint16_round_to_zero( float64 a STATUS_PARAM ) |
6069 | cbcef455 | Peter Maydell | { |
6070 | cbcef455 | Peter Maydell | int64_t v; |
6071 | cbcef455 | Peter Maydell | unsigned int res; |
6072 | cbcef455 | Peter Maydell | |
6073 | cbcef455 | Peter Maydell | v = float64_to_int64_round_to_zero(a STATUS_VAR); |
6074 | cbcef455 | Peter Maydell | if (v < 0) { |
6075 | cbcef455 | Peter Maydell | res = 0;
|
6076 | cbcef455 | Peter Maydell | float_raise( float_flag_invalid STATUS_VAR); |
6077 | cbcef455 | Peter Maydell | } else if (v > 0xffff) { |
6078 | cbcef455 | Peter Maydell | res = 0xffff;
|
6079 | cbcef455 | Peter Maydell | float_raise( float_flag_invalid STATUS_VAR); |
6080 | cbcef455 | Peter Maydell | } else {
|
6081 | cbcef455 | Peter Maydell | res = v; |
6082 | cbcef455 | Peter Maydell | } |
6083 | cbcef455 | Peter Maydell | return res;
|
6084 | cbcef455 | Peter Maydell | } |
6085 | cbcef455 | Peter Maydell | |
6086 | f090c9d4 | pbrook | /* FIXME: This looks broken. */
|
6087 | 75d62a58 | j_mayer | uint64_t float64_to_uint64 (float64 a STATUS_PARAM) |
6088 | 75d62a58 | j_mayer | { |
6089 | 75d62a58 | j_mayer | int64_t v; |
6090 | 75d62a58 | j_mayer | |
6091 | f090c9d4 | pbrook | v = float64_val(int64_to_float64(INT64_MIN STATUS_VAR)); |
6092 | f090c9d4 | pbrook | v += float64_val(a); |
6093 | f090c9d4 | pbrook | v = float64_to_int64(make_float64(v) STATUS_VAR); |
6094 | 75d62a58 | j_mayer | |
6095 | 75d62a58 | j_mayer | return v - INT64_MIN;
|
6096 | 75d62a58 | j_mayer | } |
6097 | 75d62a58 | j_mayer | |
6098 | 75d62a58 | j_mayer | uint64_t float64_to_uint64_round_to_zero (float64 a STATUS_PARAM) |
6099 | 75d62a58 | j_mayer | { |
6100 | 75d62a58 | j_mayer | int64_t v; |
6101 | 75d62a58 | j_mayer | |
6102 | f090c9d4 | pbrook | v = float64_val(int64_to_float64(INT64_MIN STATUS_VAR)); |
6103 | f090c9d4 | pbrook | v += float64_val(a); |
6104 | f090c9d4 | pbrook | v = float64_to_int64_round_to_zero(make_float64(v) STATUS_VAR); |
6105 | 75d62a58 | j_mayer | |
6106 | 75d62a58 | j_mayer | return v - INT64_MIN;
|
6107 | 75d62a58 | j_mayer | } |
6108 | 75d62a58 | j_mayer | |
6109 | 1d6bda35 | bellard | #define COMPARE(s, nan_exp) \
|
6110 | 750afe93 | bellard | INLINE int float ## s ## _compare_internal( float ## s a, float ## s b, \ |
6111 | 1d6bda35 | bellard | int is_quiet STATUS_PARAM ) \
|
6112 | 1d6bda35 | bellard | { \ |
6113 | 1d6bda35 | bellard | flag aSign, bSign; \ |
6114 | bb98fe42 | Andreas Färber | uint ## s ## _t av, bv; \ |
6115 | 37d18660 | Peter Maydell | a = float ## s ## _squash_input_denormal(a STATUS_VAR); \ |
6116 | 37d18660 | Peter Maydell | b = float ## s ## _squash_input_denormal(b STATUS_VAR); \ |
6117 | 1d6bda35 | bellard | \ |
6118 | 1d6bda35 | bellard | if (( ( extractFloat ## s ## Exp( a ) == nan_exp ) && \ |
6119 | 1d6bda35 | bellard | extractFloat ## s ## Frac( a ) ) || \ |
6120 | 1d6bda35 | bellard | ( ( extractFloat ## s ## Exp( b ) == nan_exp ) && \ |
6121 | 1d6bda35 | bellard | extractFloat ## s ## Frac( b ) )) { \ |
6122 | 1d6bda35 | bellard | if (!is_quiet || \
|
6123 | 1d6bda35 | bellard | float ## s ## _is_signaling_nan( a ) || \ |
6124 | 1d6bda35 | bellard | float ## s ## _is_signaling_nan( b ) ) { \ |
6125 | 1d6bda35 | bellard | float_raise( float_flag_invalid STATUS_VAR); \ |
6126 | 1d6bda35 | bellard | } \ |
6127 | 1d6bda35 | bellard | return float_relation_unordered; \
|
6128 | 1d6bda35 | bellard | } \ |
6129 | 1d6bda35 | bellard | aSign = extractFloat ## s ## Sign( a ); \ |
6130 | 1d6bda35 | bellard | bSign = extractFloat ## s ## Sign( b ); \ |
6131 | f090c9d4 | pbrook | av = float ## s ## _val(a); \ |
6132 | cd8a2533 | blueswir1 | bv = float ## s ## _val(b); \ |
6133 | 1d6bda35 | bellard | if ( aSign != bSign ) { \
|
6134 | bb98fe42 | Andreas Färber | if ( (uint ## s ## _t) ( ( av | bv )<<1 ) == 0 ) { \ |
6135 | 1d6bda35 | bellard | /* zero case */ \
|
6136 | 1d6bda35 | bellard | return float_relation_equal; \
|
6137 | 1d6bda35 | bellard | } else { \
|
6138 | 1d6bda35 | bellard | return 1 - (2 * aSign); \ |
6139 | 1d6bda35 | bellard | } \ |
6140 | 1d6bda35 | bellard | } else { \
|
6141 | f090c9d4 | pbrook | if (av == bv) { \
|
6142 | 1d6bda35 | bellard | return float_relation_equal; \
|
6143 | 1d6bda35 | bellard | } else { \
|
6144 | f090c9d4 | pbrook | return 1 - 2 * (aSign ^ ( av < bv )); \ |
6145 | 1d6bda35 | bellard | } \ |
6146 | 1d6bda35 | bellard | } \ |
6147 | 1d6bda35 | bellard | } \ |
6148 | 1d6bda35 | bellard | \ |
6149 | 750afe93 | bellard | int float ## s ## _compare( float ## s a, float ## s b STATUS_PARAM ) \ |
6150 | 1d6bda35 | bellard | { \ |
6151 | 1d6bda35 | bellard | return float ## s ## _compare_internal(a, b, 0 STATUS_VAR); \ |
6152 | 1d6bda35 | bellard | } \ |
6153 | 1d6bda35 | bellard | \ |
6154 | 750afe93 | bellard | int float ## s ## _compare_quiet( float ## s a, float ## s b STATUS_PARAM ) \ |
6155 | 1d6bda35 | bellard | { \ |
6156 | 1d6bda35 | bellard | return float ## s ## _compare_internal(a, b, 1 STATUS_VAR); \ |
6157 | 1d6bda35 | bellard | } |
6158 | 1d6bda35 | bellard | |
6159 | 1d6bda35 | bellard | COMPARE(32, 0xff) |
6160 | 1d6bda35 | bellard | COMPARE(64, 0x7ff) |
6161 | 9ee6e8bb | pbrook | |
6162 | f6714d36 | Aurelien Jarno | INLINE int floatx80_compare_internal( floatx80 a, floatx80 b,
|
6163 | f6714d36 | Aurelien Jarno | int is_quiet STATUS_PARAM )
|
6164 | f6714d36 | Aurelien Jarno | { |
6165 | f6714d36 | Aurelien Jarno | flag aSign, bSign; |
6166 | f6714d36 | Aurelien Jarno | |
6167 | f6714d36 | Aurelien Jarno | if (( ( extractFloatx80Exp( a ) == 0x7fff ) && |
6168 | f6714d36 | Aurelien Jarno | ( extractFloatx80Frac( a )<<1 ) ) ||
|
6169 | f6714d36 | Aurelien Jarno | ( ( extractFloatx80Exp( b ) == 0x7fff ) &&
|
6170 | f6714d36 | Aurelien Jarno | ( extractFloatx80Frac( b )<<1 ) )) {
|
6171 | f6714d36 | Aurelien Jarno | if (!is_quiet ||
|
6172 | f6714d36 | Aurelien Jarno | floatx80_is_signaling_nan( a ) || |
6173 | f6714d36 | Aurelien Jarno | floatx80_is_signaling_nan( b ) ) { |
6174 | f6714d36 | Aurelien Jarno | float_raise( float_flag_invalid STATUS_VAR); |
6175 | f6714d36 | Aurelien Jarno | } |
6176 | f6714d36 | Aurelien Jarno | return float_relation_unordered;
|
6177 | f6714d36 | Aurelien Jarno | } |
6178 | f6714d36 | Aurelien Jarno | aSign = extractFloatx80Sign( a ); |
6179 | f6714d36 | Aurelien Jarno | bSign = extractFloatx80Sign( b ); |
6180 | f6714d36 | Aurelien Jarno | if ( aSign != bSign ) {
|
6181 | f6714d36 | Aurelien Jarno | |
6182 | f6714d36 | Aurelien Jarno | if ( ( ( (uint16_t) ( ( a.high | b.high ) << 1 ) ) == 0) && |
6183 | f6714d36 | Aurelien Jarno | ( ( a.low | b.low ) == 0 ) ) {
|
6184 | f6714d36 | Aurelien Jarno | /* zero case */
|
6185 | f6714d36 | Aurelien Jarno | return float_relation_equal;
|
6186 | f6714d36 | Aurelien Jarno | } else {
|
6187 | f6714d36 | Aurelien Jarno | return 1 - (2 * aSign); |
6188 | f6714d36 | Aurelien Jarno | } |
6189 | f6714d36 | Aurelien Jarno | } else {
|
6190 | f6714d36 | Aurelien Jarno | if (a.low == b.low && a.high == b.high) {
|
6191 | f6714d36 | Aurelien Jarno | return float_relation_equal;
|
6192 | f6714d36 | Aurelien Jarno | } else {
|
6193 | f6714d36 | Aurelien Jarno | return 1 - 2 * (aSign ^ ( lt128( a.high, a.low, b.high, b.low ) )); |
6194 | f6714d36 | Aurelien Jarno | } |
6195 | f6714d36 | Aurelien Jarno | } |
6196 | f6714d36 | Aurelien Jarno | } |
6197 | f6714d36 | Aurelien Jarno | |
6198 | f6714d36 | Aurelien Jarno | int floatx80_compare( floatx80 a, floatx80 b STATUS_PARAM )
|
6199 | f6714d36 | Aurelien Jarno | { |
6200 | f6714d36 | Aurelien Jarno | return floatx80_compare_internal(a, b, 0 STATUS_VAR); |
6201 | f6714d36 | Aurelien Jarno | } |
6202 | f6714d36 | Aurelien Jarno | |
6203 | f6714d36 | Aurelien Jarno | int floatx80_compare_quiet( floatx80 a, floatx80 b STATUS_PARAM )
|
6204 | f6714d36 | Aurelien Jarno | { |
6205 | f6714d36 | Aurelien Jarno | return floatx80_compare_internal(a, b, 1 STATUS_VAR); |
6206 | f6714d36 | Aurelien Jarno | } |
6207 | f6714d36 | Aurelien Jarno | |
6208 | 1f587329 | blueswir1 | INLINE int float128_compare_internal( float128 a, float128 b,
|
6209 | 1f587329 | blueswir1 | int is_quiet STATUS_PARAM )
|
6210 | 1f587329 | blueswir1 | { |
6211 | 1f587329 | blueswir1 | flag aSign, bSign; |
6212 | 1f587329 | blueswir1 | |
6213 | 1f587329 | blueswir1 | if (( ( extractFloat128Exp( a ) == 0x7fff ) && |
6214 | 1f587329 | blueswir1 | ( extractFloat128Frac0( a ) | extractFloat128Frac1( a ) ) ) || |
6215 | 1f587329 | blueswir1 | ( ( extractFloat128Exp( b ) == 0x7fff ) &&
|
6216 | 1f587329 | blueswir1 | ( extractFloat128Frac0( b ) | extractFloat128Frac1( b ) ) )) { |
6217 | 1f587329 | blueswir1 | if (!is_quiet ||
|
6218 | 1f587329 | blueswir1 | float128_is_signaling_nan( a ) || |
6219 | 1f587329 | blueswir1 | float128_is_signaling_nan( b ) ) { |
6220 | 1f587329 | blueswir1 | float_raise( float_flag_invalid STATUS_VAR); |
6221 | 1f587329 | blueswir1 | } |
6222 | 1f587329 | blueswir1 | return float_relation_unordered;
|
6223 | 1f587329 | blueswir1 | } |
6224 | 1f587329 | blueswir1 | aSign = extractFloat128Sign( a ); |
6225 | 1f587329 | blueswir1 | bSign = extractFloat128Sign( b ); |
6226 | 1f587329 | blueswir1 | if ( aSign != bSign ) {
|
6227 | 1f587329 | blueswir1 | if ( ( ( ( a.high | b.high )<<1 ) | a.low | b.low ) == 0 ) { |
6228 | 1f587329 | blueswir1 | /* zero case */
|
6229 | 1f587329 | blueswir1 | return float_relation_equal;
|
6230 | 1f587329 | blueswir1 | } else {
|
6231 | 1f587329 | blueswir1 | return 1 - (2 * aSign); |
6232 | 1f587329 | blueswir1 | } |
6233 | 1f587329 | blueswir1 | } else {
|
6234 | 1f587329 | blueswir1 | if (a.low == b.low && a.high == b.high) {
|
6235 | 1f587329 | blueswir1 | return float_relation_equal;
|
6236 | 1f587329 | blueswir1 | } else {
|
6237 | 1f587329 | blueswir1 | return 1 - 2 * (aSign ^ ( lt128( a.high, a.low, b.high, b.low ) )); |
6238 | 1f587329 | blueswir1 | } |
6239 | 1f587329 | blueswir1 | } |
6240 | 1f587329 | blueswir1 | } |
6241 | 1f587329 | blueswir1 | |
6242 | 1f587329 | blueswir1 | int float128_compare( float128 a, float128 b STATUS_PARAM )
|
6243 | 1f587329 | blueswir1 | { |
6244 | 1f587329 | blueswir1 | return float128_compare_internal(a, b, 0 STATUS_VAR); |
6245 | 1f587329 | blueswir1 | } |
6246 | 1f587329 | blueswir1 | |
6247 | 1f587329 | blueswir1 | int float128_compare_quiet( float128 a, float128 b STATUS_PARAM )
|
6248 | 1f587329 | blueswir1 | { |
6249 | 1f587329 | blueswir1 | return float128_compare_internal(a, b, 1 STATUS_VAR); |
6250 | 1f587329 | blueswir1 | } |
6251 | 1f587329 | blueswir1 | |
6252 | 274f1b04 | Peter Maydell | /* min() and max() functions. These can't be implemented as
|
6253 | 274f1b04 | Peter Maydell | * 'compare and pick one input' because that would mishandle
|
6254 | 274f1b04 | Peter Maydell | * NaNs and +0 vs -0.
|
6255 | 274f1b04 | Peter Maydell | */
|
6256 | 274f1b04 | Peter Maydell | #define MINMAX(s, nan_exp) \
|
6257 | 274f1b04 | Peter Maydell | INLINE float ## s float ## s ## _minmax(float ## s a, float ## s b, \ |
6258 | 274f1b04 | Peter Maydell | int ismin STATUS_PARAM ) \
|
6259 | 274f1b04 | Peter Maydell | { \ |
6260 | 274f1b04 | Peter Maydell | flag aSign, bSign; \ |
6261 | 274f1b04 | Peter Maydell | uint ## s ## _t av, bv; \ |
6262 | 274f1b04 | Peter Maydell | a = float ## s ## _squash_input_denormal(a STATUS_VAR); \ |
6263 | 274f1b04 | Peter Maydell | b = float ## s ## _squash_input_denormal(b STATUS_VAR); \ |
6264 | 274f1b04 | Peter Maydell | if (float ## s ## _is_any_nan(a) || \ |
6265 | 274f1b04 | Peter Maydell | float ## s ## _is_any_nan(b)) { \ |
6266 | 274f1b04 | Peter Maydell | return propagateFloat ## s ## NaN(a, b STATUS_VAR); \ |
6267 | 274f1b04 | Peter Maydell | } \ |
6268 | 274f1b04 | Peter Maydell | aSign = extractFloat ## s ## Sign(a); \ |
6269 | 274f1b04 | Peter Maydell | bSign = extractFloat ## s ## Sign(b); \ |
6270 | 274f1b04 | Peter Maydell | av = float ## s ## _val(a); \ |
6271 | 274f1b04 | Peter Maydell | bv = float ## s ## _val(b); \ |
6272 | 274f1b04 | Peter Maydell | if (aSign != bSign) { \
|
6273 | 274f1b04 | Peter Maydell | if (ismin) { \
|
6274 | 274f1b04 | Peter Maydell | return aSign ? a : b; \
|
6275 | 274f1b04 | Peter Maydell | } else { \
|
6276 | 274f1b04 | Peter Maydell | return aSign ? b : a; \
|
6277 | 274f1b04 | Peter Maydell | } \ |
6278 | 274f1b04 | Peter Maydell | } else { \
|
6279 | 274f1b04 | Peter Maydell | if (ismin) { \
|
6280 | 274f1b04 | Peter Maydell | return (aSign ^ (av < bv)) ? a : b; \
|
6281 | 274f1b04 | Peter Maydell | } else { \
|
6282 | 274f1b04 | Peter Maydell | return (aSign ^ (av < bv)) ? b : a; \
|
6283 | 274f1b04 | Peter Maydell | } \ |
6284 | 274f1b04 | Peter Maydell | } \ |
6285 | 274f1b04 | Peter Maydell | } \ |
6286 | 274f1b04 | Peter Maydell | \ |
6287 | 274f1b04 | Peter Maydell | float ## s float ## s ## _min(float ## s a, float ## s b STATUS_PARAM) \ |
6288 | 274f1b04 | Peter Maydell | { \ |
6289 | 274f1b04 | Peter Maydell | return float ## s ## _minmax(a, b, 1 STATUS_VAR); \ |
6290 | 274f1b04 | Peter Maydell | } \ |
6291 | 274f1b04 | Peter Maydell | \ |
6292 | 274f1b04 | Peter Maydell | float ## s float ## s ## _max(float ## s a, float ## s b STATUS_PARAM) \ |
6293 | 274f1b04 | Peter Maydell | { \ |
6294 | 274f1b04 | Peter Maydell | return float ## s ## _minmax(a, b, 0 STATUS_VAR); \ |
6295 | 274f1b04 | Peter Maydell | } |
6296 | 274f1b04 | Peter Maydell | |
6297 | 274f1b04 | Peter Maydell | MINMAX(32, 0xff) |
6298 | 274f1b04 | Peter Maydell | MINMAX(64, 0x7ff) |
6299 | 274f1b04 | Peter Maydell | |
6300 | 274f1b04 | Peter Maydell | |
6301 | 9ee6e8bb | pbrook | /* Multiply A by 2 raised to the power N. */
|
6302 | 9ee6e8bb | pbrook | float32 float32_scalbn( float32 a, int n STATUS_PARAM )
|
6303 | 9ee6e8bb | pbrook | { |
6304 | 9ee6e8bb | pbrook | flag aSign; |
6305 | 326b9e98 | Aurelien Jarno | int16_t aExp; |
6306 | bb98fe42 | Andreas Färber | uint32_t aSig; |
6307 | 9ee6e8bb | pbrook | |
6308 | 37d18660 | Peter Maydell | a = float32_squash_input_denormal(a STATUS_VAR); |
6309 | 9ee6e8bb | pbrook | aSig = extractFloat32Frac( a ); |
6310 | 9ee6e8bb | pbrook | aExp = extractFloat32Exp( a ); |
6311 | 9ee6e8bb | pbrook | aSign = extractFloat32Sign( a ); |
6312 | 9ee6e8bb | pbrook | |
6313 | 9ee6e8bb | pbrook | if ( aExp == 0xFF ) { |
6314 | 326b9e98 | Aurelien Jarno | if ( aSig ) {
|
6315 | 326b9e98 | Aurelien Jarno | return propagateFloat32NaN( a, a STATUS_VAR );
|
6316 | 326b9e98 | Aurelien Jarno | } |
6317 | 9ee6e8bb | pbrook | return a;
|
6318 | 9ee6e8bb | pbrook | } |
6319 | 69397542 | pbrook | if ( aExp != 0 ) |
6320 | 69397542 | pbrook | aSig |= 0x00800000;
|
6321 | 69397542 | pbrook | else if ( aSig == 0 ) |
6322 | 69397542 | pbrook | return a;
|
6323 | 69397542 | pbrook | |
6324 | 326b9e98 | Aurelien Jarno | if (n > 0x200) { |
6325 | 326b9e98 | Aurelien Jarno | n = 0x200;
|
6326 | 326b9e98 | Aurelien Jarno | } else if (n < -0x200) { |
6327 | 326b9e98 | Aurelien Jarno | n = -0x200;
|
6328 | 326b9e98 | Aurelien Jarno | } |
6329 | 326b9e98 | Aurelien Jarno | |
6330 | 69397542 | pbrook | aExp += n - 1;
|
6331 | 69397542 | pbrook | aSig <<= 7;
|
6332 | 69397542 | pbrook | return normalizeRoundAndPackFloat32( aSign, aExp, aSig STATUS_VAR );
|
6333 | 9ee6e8bb | pbrook | } |
6334 | 9ee6e8bb | pbrook | |
6335 | 9ee6e8bb | pbrook | float64 float64_scalbn( float64 a, int n STATUS_PARAM )
|
6336 | 9ee6e8bb | pbrook | { |
6337 | 9ee6e8bb | pbrook | flag aSign; |
6338 | 326b9e98 | Aurelien Jarno | int16_t aExp; |
6339 | bb98fe42 | Andreas Färber | uint64_t aSig; |
6340 | 9ee6e8bb | pbrook | |
6341 | 37d18660 | Peter Maydell | a = float64_squash_input_denormal(a STATUS_VAR); |
6342 | 9ee6e8bb | pbrook | aSig = extractFloat64Frac( a ); |
6343 | 9ee6e8bb | pbrook | aExp = extractFloat64Exp( a ); |
6344 | 9ee6e8bb | pbrook | aSign = extractFloat64Sign( a ); |
6345 | 9ee6e8bb | pbrook | |
6346 | 9ee6e8bb | pbrook | if ( aExp == 0x7FF ) { |
6347 | 326b9e98 | Aurelien Jarno | if ( aSig ) {
|
6348 | 326b9e98 | Aurelien Jarno | return propagateFloat64NaN( a, a STATUS_VAR );
|
6349 | 326b9e98 | Aurelien Jarno | } |
6350 | 9ee6e8bb | pbrook | return a;
|
6351 | 9ee6e8bb | pbrook | } |
6352 | 69397542 | pbrook | if ( aExp != 0 ) |
6353 | 69397542 | pbrook | aSig |= LIT64( 0x0010000000000000 );
|
6354 | 69397542 | pbrook | else if ( aSig == 0 ) |
6355 | 69397542 | pbrook | return a;
|
6356 | 69397542 | pbrook | |
6357 | 326b9e98 | Aurelien Jarno | if (n > 0x1000) { |
6358 | 326b9e98 | Aurelien Jarno | n = 0x1000;
|
6359 | 326b9e98 | Aurelien Jarno | } else if (n < -0x1000) { |
6360 | 326b9e98 | Aurelien Jarno | n = -0x1000;
|
6361 | 326b9e98 | Aurelien Jarno | } |
6362 | 326b9e98 | Aurelien Jarno | |
6363 | 69397542 | pbrook | aExp += n - 1;
|
6364 | 69397542 | pbrook | aSig <<= 10;
|
6365 | 69397542 | pbrook | return normalizeRoundAndPackFloat64( aSign, aExp, aSig STATUS_VAR );
|
6366 | 9ee6e8bb | pbrook | } |
6367 | 9ee6e8bb | pbrook | |
6368 | 9ee6e8bb | pbrook | floatx80 floatx80_scalbn( floatx80 a, int n STATUS_PARAM )
|
6369 | 9ee6e8bb | pbrook | { |
6370 | 9ee6e8bb | pbrook | flag aSign; |
6371 | 326b9e98 | Aurelien Jarno | int32_t aExp; |
6372 | bb98fe42 | Andreas Färber | uint64_t aSig; |
6373 | 9ee6e8bb | pbrook | |
6374 | 9ee6e8bb | pbrook | aSig = extractFloatx80Frac( a ); |
6375 | 9ee6e8bb | pbrook | aExp = extractFloatx80Exp( a ); |
6376 | 9ee6e8bb | pbrook | aSign = extractFloatx80Sign( a ); |
6377 | 9ee6e8bb | pbrook | |
6378 | 326b9e98 | Aurelien Jarno | if ( aExp == 0x7FFF ) { |
6379 | 326b9e98 | Aurelien Jarno | if ( aSig<<1 ) { |
6380 | 326b9e98 | Aurelien Jarno | return propagateFloatx80NaN( a, a STATUS_VAR );
|
6381 | 326b9e98 | Aurelien Jarno | } |
6382 | 9ee6e8bb | pbrook | return a;
|
6383 | 9ee6e8bb | pbrook | } |
6384 | 326b9e98 | Aurelien Jarno | |
6385 | 69397542 | pbrook | if (aExp == 0 && aSig == 0) |
6386 | 69397542 | pbrook | return a;
|
6387 | 69397542 | pbrook | |
6388 | 326b9e98 | Aurelien Jarno | if (n > 0x10000) { |
6389 | 326b9e98 | Aurelien Jarno | n = 0x10000;
|
6390 | 326b9e98 | Aurelien Jarno | } else if (n < -0x10000) { |
6391 | 326b9e98 | Aurelien Jarno | n = -0x10000;
|
6392 | 326b9e98 | Aurelien Jarno | } |
6393 | 326b9e98 | Aurelien Jarno | |
6394 | 9ee6e8bb | pbrook | aExp += n; |
6395 | 69397542 | pbrook | return normalizeRoundAndPackFloatx80( STATUS(floatx80_rounding_precision),
|
6396 | 69397542 | pbrook | aSign, aExp, aSig, 0 STATUS_VAR );
|
6397 | 9ee6e8bb | pbrook | } |
6398 | 9ee6e8bb | pbrook | |
6399 | 9ee6e8bb | pbrook | float128 float128_scalbn( float128 a, int n STATUS_PARAM )
|
6400 | 9ee6e8bb | pbrook | { |
6401 | 9ee6e8bb | pbrook | flag aSign; |
6402 | 326b9e98 | Aurelien Jarno | int32_t aExp; |
6403 | bb98fe42 | Andreas Färber | uint64_t aSig0, aSig1; |
6404 | 9ee6e8bb | pbrook | |
6405 | 9ee6e8bb | pbrook | aSig1 = extractFloat128Frac1( a ); |
6406 | 9ee6e8bb | pbrook | aSig0 = extractFloat128Frac0( a ); |
6407 | 9ee6e8bb | pbrook | aExp = extractFloat128Exp( a ); |
6408 | 9ee6e8bb | pbrook | aSign = extractFloat128Sign( a ); |
6409 | 9ee6e8bb | pbrook | if ( aExp == 0x7FFF ) { |
6410 | 326b9e98 | Aurelien Jarno | if ( aSig0 | aSig1 ) {
|
6411 | 326b9e98 | Aurelien Jarno | return propagateFloat128NaN( a, a STATUS_VAR );
|
6412 | 326b9e98 | Aurelien Jarno | } |
6413 | 9ee6e8bb | pbrook | return a;
|
6414 | 9ee6e8bb | pbrook | } |
6415 | 69397542 | pbrook | if ( aExp != 0 ) |
6416 | 69397542 | pbrook | aSig0 |= LIT64( 0x0001000000000000 );
|
6417 | 69397542 | pbrook | else if ( aSig0 == 0 && aSig1 == 0 ) |
6418 | 69397542 | pbrook | return a;
|
6419 | 69397542 | pbrook | |
6420 | 326b9e98 | Aurelien Jarno | if (n > 0x10000) { |
6421 | 326b9e98 | Aurelien Jarno | n = 0x10000;
|
6422 | 326b9e98 | Aurelien Jarno | } else if (n < -0x10000) { |
6423 | 326b9e98 | Aurelien Jarno | n = -0x10000;
|
6424 | 326b9e98 | Aurelien Jarno | } |
6425 | 326b9e98 | Aurelien Jarno | |
6426 | 69397542 | pbrook | aExp += n - 1;
|
6427 | 69397542 | pbrook | return normalizeRoundAndPackFloat128( aSign, aExp, aSig0, aSig1
|
6428 | 69397542 | pbrook | STATUS_VAR ); |
6429 | 9ee6e8bb | pbrook | |
6430 | 9ee6e8bb | pbrook | } |