Statistics
| Branch: | Revision:

root / tcg / README @ 4e17eae9

History | View | Annotate | Download (13.6 kB)

1
Tiny Code Generator - Fabrice Bellard.
2

    
3
1) Introduction
4

    
5
TCG (Tiny Code Generator) began as a generic backend for a C
6
compiler. It was simplified to be used in QEMU. It also has its roots
7
in the QOP code generator written by Paul Brook. 
8

    
9
2) Definitions
10

    
11
The TCG "target" is the architecture for which we generate the
12
code. It is of course not the same as the "target" of QEMU which is
13
the emulated architecture. As TCG started as a generic C backend used
14
for cross compiling, it is assumed that the TCG target is different
15
from the host, although it is never the case for QEMU.
16

    
17
A TCG "function" corresponds to a QEMU Translated Block (TB).
18

    
19
A TCG "temporary" is a variable only live in a basic
20
block. Temporaries are allocated explicitly in each function.
21

    
22
A TCG "local temporary" is a variable only live in a function. Local
23
temporaries are allocated explicitly in each function.
24

    
25
A TCG "global" is a variable which is live in all the functions
26
(equivalent of a C global variable). They are defined before the
27
functions defined. A TCG global can be a memory location (e.g. a QEMU
28
CPU register), a fixed host register (e.g. the QEMU CPU state pointer)
29
or a memory location which is stored in a register outside QEMU TBs
30
(not implemented yet).
31

    
32
A TCG "basic block" corresponds to a list of instructions terminated
33
by a branch instruction. 
34

    
35
3) Intermediate representation
36

    
37
3.1) Introduction
38

    
39
TCG instructions operate on variables which are temporaries, local
40
temporaries or globals. TCG instructions and variables are strongly
41
typed. Two types are supported: 32 bit integers and 64 bit
42
integers. Pointers are defined as an alias to 32 bit or 64 bit
43
integers depending on the TCG target word size.
44

    
45
Each instruction has a fixed number of output variable operands, input
46
variable operands and always constant operands.
47

    
48
The notable exception is the call instruction which has a variable
49
number of outputs and inputs.
50

    
51
In the textual form, output operands usually come first, followed by
52
input operands, followed by constant operands. The output type is
53
included in the instruction name. Constants are prefixed with a '$'.
54

    
55
add_i32 t0, t1, t2  (t0 <- t1 + t2)
56

    
57
3.2) Assumptions
58

    
59
* Basic blocks
60

    
61
- Basic blocks end after branches (e.g. brcond_i32 instruction),
62
  goto_tb and exit_tb instructions.
63
- Basic blocks start after the end of a previous basic block, or at a
64
  set_label instruction.
65

    
66
After the end of a basic block, the content of temporaries is
67
destroyed, but local temporaries and globals are preserved.
68

    
69
* Floating point types are not supported yet
70

    
71
* Pointers: depending on the TCG target, pointer size is 32 bit or 64
72
  bit. The type TCG_TYPE_PTR is an alias to TCG_TYPE_I32 or
73
  TCG_TYPE_I64.
74

    
75
* Helpers:
76

    
77
Using the tcg_gen_helper_x_y it is possible to call any function
78
taking i32, i64 or pointer types. Before calling an helper, all
79
globals are stored at their canonical location and it is assumed that
80
the function can modify them. In the future, function modifiers will
81
be allowed to tell that the helper does not read or write some globals.
82

    
83
On some TCG targets (e.g. x86), several calling conventions are
84
supported.
85

    
86
* Branches:
87

    
88
Use the instruction 'br' to jump to a label. Use 'jmp' to jump to an
89
explicit address. Conditional branches can only jump to labels.
90

    
91
3.3) Code Optimizations
92

    
93
When generating instructions, you can count on at least the following
94
optimizations:
95

    
96
- Single instructions are simplified, e.g.
97

    
98
   and_i32 t0, t0, $0xffffffff
99
    
100
  is suppressed.
101

    
102
- A liveness analysis is done at the basic block level. The
103
  information is used to suppress moves from a dead variable to
104
  another one. It is also used to remove instructions which compute
105
  dead results. The later is especially useful for condition code
106
  optimization in QEMU.
107

    
108
  In the following example:
109

    
110
  add_i32 t0, t1, t2
111
  add_i32 t0, t0, $1
112
  mov_i32 t0, $1
113

    
114
  only the last instruction is kept.
115

    
116
3.4) Instruction Reference
117

    
118
********* Function call
119

    
120
* call <ret> <params> ptr
121

    
122
call function 'ptr' (pointer type)
123

    
124
<ret> optional 32 bit or 64 bit return value
125
<params> optional 32 bit or 64 bit parameters
126

    
127
********* Jumps/Labels
128

    
129
* jmp t0
130

    
131
Absolute jump to address t0 (pointer type).
132

    
133
* set_label $label
134

    
135
Define label 'label' at the current program point.
136

    
137
* br $label
138

    
139
Jump to label.
140

    
141
* brcond_i32/i64 cond, t0, t1, label
142

    
143
Conditional jump if t0 cond t1 is true. cond can be:
144
    TCG_COND_EQ
145
    TCG_COND_NE
146
    TCG_COND_LT /* signed */
147
    TCG_COND_GE /* signed */
148
    TCG_COND_LE /* signed */
149
    TCG_COND_GT /* signed */
150
    TCG_COND_LTU /* unsigned */
151
    TCG_COND_GEU /* unsigned */
152
    TCG_COND_LEU /* unsigned */
153
    TCG_COND_GTU /* unsigned */
154

    
155
********* Arithmetic
156

    
157
* add_i32/i64 t0, t1, t2
158

    
159
t0=t1+t2
160

    
161
* sub_i32/i64 t0, t1, t2
162

    
163
t0=t1-t2
164

    
165
* neg_i32/i64 t0, t1
166

    
167
t0=-t1 (two's complement)
168

    
169
* mul_i32/i64 t0, t1, t2
170

    
171
t0=t1*t2
172

    
173
* div_i32/i64 t0, t1, t2
174

    
175
t0=t1/t2 (signed). Undefined behavior if division by zero or overflow.
176

    
177
* divu_i32/i64 t0, t1, t2
178

    
179
t0=t1/t2 (unsigned). Undefined behavior if division by zero.
180

    
181
* rem_i32/i64 t0, t1, t2
182

    
183
t0=t1%t2 (signed). Undefined behavior if division by zero or overflow.
184

    
185
* remu_i32/i64 t0, t1, t2
186

    
187
t0=t1%t2 (unsigned). Undefined behavior if division by zero.
188

    
189
********* Logical
190

    
191
* and_i32/i64 t0, t1, t2
192

    
193
t0=t1&t2
194

    
195
* or_i32/i64 t0, t1, t2
196

    
197
t0=t1|t2
198

    
199
* xor_i32/i64 t0, t1, t2
200

    
201
t0=t1^t2
202

    
203
* not_i32/i64 t0, t1
204

    
205
t0=~t1
206

    
207
* andc_i32/i64 t0, t1, t2
208

    
209
t0=t1&~t2
210

    
211
* eqv_i32/i64 t0, t1, t2
212

    
213
t0=~(t1^t2)
214

    
215
* nand_i32/i64 t0, t1, t2
216

    
217
t0=~(t1&t2)
218

    
219
* nor_i32/i64 t0, t1, t2
220

    
221
t0=~(t1|t2)
222

    
223
* orc_i32/i64 t0, t1, t2
224

    
225
t0=t1|~t2
226

    
227
********* Shifts/Rotates
228

    
229
* shl_i32/i64 t0, t1, t2
230

    
231
t0=t1 << t2. Undefined behavior if t2 < 0 or t2 >= 32 (resp 64)
232

    
233
* shr_i32/i64 t0, t1, t2
234

    
235
t0=t1 >> t2 (unsigned). Undefined behavior if t2 < 0 or t2 >= 32 (resp 64)
236

    
237
* sar_i32/i64 t0, t1, t2
238

    
239
t0=t1 >> t2 (signed). Undefined behavior if t2 < 0 or t2 >= 32 (resp 64)
240

    
241
* rotl_i32/i64 t0, t1, t2
242

    
243
Rotation of t2 bits to the left. Undefined behavior if t2 < 0 or t2 >= 32 (resp 64)
244

    
245
* rotr_i32/i64 t0, t1, t2
246

    
247
Rotation of t2 bits to the right. Undefined behavior if t2 < 0 or t2 >= 32 (resp 64)
248

    
249
********* Misc
250

    
251
* mov_i32/i64 t0, t1
252

    
253
t0 = t1
254

    
255
Move t1 to t0 (both operands must have the same type).
256

    
257
* ext8s_i32/i64 t0, t1
258
ext8u_i32/i64 t0, t1
259
ext16s_i32/i64 t0, t1
260
ext16u_i32/i64 t0, t1
261
ext32s_i64 t0, t1
262
ext32u_i64 t0, t1
263

    
264
8, 16 or 32 bit sign/zero extension (both operands must have the same type)
265

    
266
* bswap16_i32/i64 t0, t1
267

    
268
16 bit byte swap on a 32/64 bit value. The two/six high order bytes must be
269
set to zero.
270

    
271
* bswap32_i32/i64 t0, t1
272

    
273
32 bit byte swap on a 32/64 bit value. With a 64 bit value, the four high
274
order bytes must be set to zero.
275

    
276
* bswap64_i64 t0, t1
277

    
278
64 bit byte swap
279

    
280
* discard_i32/i64 t0
281

    
282
Indicate that the value of t0 won't be used later. It is useful to
283
force dead code elimination.
284

    
285
********* Conditional moves
286

    
287
* setcond_i32/i64 cond, dest, t1, t2
288

    
289
dest = (t1 cond t2)
290

    
291
Set DEST to 1 if (T1 cond T2) is true, otherwise set to 0.
292

    
293
********* Type conversions
294

    
295
* ext_i32_i64 t0, t1
296
Convert t1 (32 bit) to t0 (64 bit) and does sign extension
297

    
298
* extu_i32_i64 t0, t1
299
Convert t1 (32 bit) to t0 (64 bit) and does zero extension
300

    
301
* trunc_i64_i32 t0, t1
302
Truncate t1 (64 bit) to t0 (32 bit)
303

    
304
* concat_i32_i64 t0, t1, t2
305
Construct t0 (64-bit) taking the low half from t1 (32 bit) and the high half
306
from t2 (32 bit).
307

    
308
* concat32_i64 t0, t1, t2
309
Construct t0 (64-bit) taking the low half from t1 (64 bit) and the high half
310
from t2 (64 bit).
311

    
312
********* Load/Store
313

    
314
* ld_i32/i64 t0, t1, offset
315
ld8s_i32/i64 t0, t1, offset
316
ld8u_i32/i64 t0, t1, offset
317
ld16s_i32/i64 t0, t1, offset
318
ld16u_i32/i64 t0, t1, offset
319
ld32s_i64 t0, t1, offset
320
ld32u_i64 t0, t1, offset
321

    
322
t0 = read(t1 + offset)
323
Load 8, 16, 32 or 64 bits with or without sign extension from host memory. 
324
offset must be a constant.
325

    
326
* st_i32/i64 t0, t1, offset
327
st8_i32/i64 t0, t1, offset
328
st16_i32/i64 t0, t1, offset
329
st32_i64 t0, t1, offset
330

    
331
write(t0, t1 + offset)
332
Write 8, 16, 32 or 64 bits to host memory.
333

    
334
********* 64-bit target on 32-bit host support
335

    
336
The following opcodes are internal to TCG.  Thus they are to be implemented by
337
32-bit host code generators, but are not to be emitted by guest translators.
338
They are emitted as needed by inline functions within "tcg-op.h".
339

    
340
* brcond2_i32 cond, t0_low, t0_high, t1_low, t1_high, label
341

    
342
Similar to brcond, except that the 64-bit values T0 and T1
343
are formed from two 32-bit arguments.
344

    
345
* add2_i32 t0_low, t0_high, t1_low, t1_high, t2_low, t2_high
346
* sub2_i32 t0_low, t0_high, t1_low, t1_high, t2_low, t2_high
347

    
348
Similar to add/sub, except that the 64-bit inputs T1 and T2 are
349
formed from two 32-bit arguments, and the 64-bit output T0
350
is returned in two 32-bit outputs.
351

    
352
* mulu2_i32 t0_low, t0_high, t1, t2
353

    
354
Similar to mul, except two 32-bit (unsigned) inputs T1 and T2 yielding
355
the full 64-bit product T0.  The later is returned in two 32-bit outputs.
356

    
357
* setcond2_i32 cond, dest, t1_low, t1_high, t2_low, t2_high
358

    
359
Similar to setcond, except that the 64-bit values T1 and T2 are
360
formed from two 32-bit arguments.  The result is a 32-bit value.
361

    
362
********* QEMU specific operations
363

    
364
* tb_exit t0
365

    
366
Exit the current TB and return the value t0 (word type).
367

    
368
* goto_tb index
369

    
370
Exit the current TB and jump to the TB index 'index' (constant) if the
371
current TB was linked to this TB. Otherwise execute the next
372
instructions.
373

    
374
* qemu_ld8u t0, t1, flags
375
qemu_ld8s t0, t1, flags
376
qemu_ld16u t0, t1, flags
377
qemu_ld16s t0, t1, flags
378
qemu_ld32u t0, t1, flags
379
qemu_ld32s t0, t1, flags
380
qemu_ld64 t0, t1, flags
381

    
382
Load data at the QEMU CPU address t1 into t0. t1 has the QEMU CPU
383
address type. 'flags' contains the QEMU memory index (selects user or
384
kernel access) for example.
385

    
386
* qemu_st8 t0, t1, flags
387
qemu_st16 t0, t1, flags
388
qemu_st32 t0, t1, flags
389
qemu_st64 t0, t1, flags
390

    
391
Store the data t0 at the QEMU CPU Address t1. t1 has the QEMU CPU
392
address type. 'flags' contains the QEMU memory index (selects user or
393
kernel access) for example.
394

    
395
Note 1: Some shortcuts are defined when the last operand is known to be
396
a constant (e.g. addi for add, movi for mov).
397

    
398
Note 2: When using TCG, the opcodes must never be generated directly
399
as some of them may not be available as "real" opcodes. Always use the
400
function tcg_gen_xxx(args).
401

    
402
4) Backend
403

    
404
tcg-target.h contains the target specific definitions. tcg-target.c
405
contains the target specific code.
406

    
407
4.1) Assumptions
408

    
409
The target word size (TCG_TARGET_REG_BITS) is expected to be 32 bit or
410
64 bit. It is expected that the pointer has the same size as the word.
411

    
412
On a 32 bit target, all 64 bit operations are converted to 32 bits. A
413
few specific operations must be implemented to allow it (see add2_i32,
414
sub2_i32, brcond2_i32).
415

    
416
Floating point operations are not supported in this version. A
417
previous incarnation of the code generator had full support of them,
418
but it is better to concentrate on integer operations first.
419

    
420
On a 64 bit target, no assumption is made in TCG about the storage of
421
the 32 bit values in 64 bit registers.
422

    
423
4.2) Constraints
424

    
425
GCC like constraints are used to define the constraints of every
426
instruction. Memory constraints are not supported in this
427
version. Aliases are specified in the input operands as for GCC.
428

    
429
The same register may be used for both an input and an output, even when
430
they are not explicitly aliased.  If an op expands to multiple target
431
instructions then care must be taken to avoid clobbering input values.
432
GCC style "early clobber" outputs are not currently supported.
433

    
434
A target can define specific register or constant constraints. If an
435
operation uses a constant input constraint which does not allow all
436
constants, it must also accept registers in order to have a fallback.
437

    
438
The movi_i32 and movi_i64 operations must accept any constants.
439

    
440
The mov_i32 and mov_i64 operations must accept any registers of the
441
same type.
442

    
443
The ld/st instructions must accept signed 32 bit constant offsets. It
444
can be implemented by reserving a specific register to compute the
445
address if the offset is too big.
446

    
447
The ld/st instructions must accept any destination (ld) or source (st)
448
register.
449

    
450
4.3) Function call assumptions
451

    
452
- The only supported types for parameters and return value are: 32 and
453
  64 bit integers and pointer.
454
- The stack grows downwards.
455
- The first N parameters are passed in registers.
456
- The next parameters are passed on the stack by storing them as words.
457
- Some registers are clobbered during the call. 
458
- The function can return 0 or 1 value in registers. On a 32 bit
459
  target, functions must be able to return 2 values in registers for
460
  64 bit return type.
461

    
462
5) Recommended coding rules for best performance
463

    
464
- Use globals to represent the parts of the QEMU CPU state which are
465
  often modified, e.g. the integer registers and the condition
466
  codes. TCG will be able to use host registers to store them.
467

    
468
- Avoid globals stored in fixed registers. They must be used only to
469
  store the pointer to the CPU state and possibly to store a pointer
470
  to a register window.
471

    
472
- Use temporaries. Use local temporaries only when really needed,
473
  e.g. when you need to use a value after a jump. Local temporaries
474
  introduce a performance hit in the current TCG implementation: their
475
  content is saved to memory at end of each basic block.
476

    
477
- Free temporaries and local temporaries when they are no longer used
478
  (tcg_temp_free). Since tcg_const_x() also creates a temporary, you
479
  should free it after it is used. Freeing temporaries does not yield
480
  a better generated code, but it reduces the memory usage of TCG and
481
  the speed of the translation.
482

    
483
- Don't hesitate to use helpers for complicated or seldom used target
484
  intructions. There is little performance advantage in using TCG to
485
  implement target instructions taking more than about twenty TCG
486
  instructions.
487

    
488
- Use the 'discard' instruction if you know that TCG won't be able to
489
  prove that a given global is "dead" at a given program point. The
490
  x86 target uses it to improve the condition codes optimisation.