1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
|
//===-- R600CodeEmitter.cpp - Code Emitter for R600->Cayman GPU families --===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This code emitters outputs bytecode that is understood by the r600g driver
// in the Mesa [1] project. The bytecode is very similar to the hardware's ISA,
// except that the size of the instruction fields are rounded up to the
// nearest byte.
//
// [1] http://www.mesa3d.org/
//
//===----------------------------------------------------------------------===//
#include "AMDGPU.h"
#include "AMDGPUCodeEmitter.h"
#include "AMDGPUInstrInfo.h"
#include "AMDILUtilityFunctions.h"
#include "R600InstrInfo.h"
#include "R600RegisterInfo.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/Support/DataTypes.h"
#include "llvm/Support/FormattedStream.h"
#include "llvm/Target/TargetMachine.h"
#include <stdio.h>
#define SRC_BYTE_COUNT 11
#define DST_BYTE_COUNT 5
using namespace llvm;
namespace {
class R600CodeEmitter : public MachineFunctionPass, public AMDGPUCodeEmitter {
private:
static char ID;
formatted_raw_ostream &_OS;
const TargetMachine * TM;
const MachineRegisterInfo * MRI;
const R600RegisterInfo * TRI;
const R600InstrInfo * TII;
unsigned currentElement;
unsigned section_start;
public:
R600CodeEmitter(formatted_raw_ostream &OS) : MachineFunctionPass(ID),
_OS(OS), TM(NULL) { }
const char *getPassName() const { return "AMDGPU Machine Code Emitter"; }
bool runOnMachineFunction(MachineFunction &MF);
virtual uint64_t getMachineOpValue(const MachineInstr &MI,
const MachineOperand &MO) const;
private:
void EmitALUInstr(MachineInstr &MI);
void EmitSrc(const MachineOperand & MO, unsigned SrcIdx);
void EmitDst(const MachineOperand & MO);
void EmitALU(MachineInstr &MI, unsigned numSrc);
void EmitTexInstr(MachineInstr &MI);
void EmitFCInstr(MachineInstr &MI);
void EmitNullBytes(unsigned int byteCount);
void EmitByte(unsigned int byte);
void EmitTwoBytes(uint32_t bytes);
void Emit(uint32_t value);
void Emit(uint64_t value);
unsigned getHWReg(unsigned regNo) const;
};
} // End anonymous namespace
enum RegElement {
ELEMENT_X = 0,
ELEMENT_Y,
ELEMENT_Z,
ELEMENT_W
};
enum InstrTypes {
INSTR_ALU = 0,
INSTR_TEX,
INSTR_FC,
INSTR_NATIVE,
INSTR_VTX
};
enum FCInstr {
FC_IF = 0,
FC_IF_INT,
FC_ELSE,
FC_ENDIF,
FC_BGNLOOP,
FC_ENDLOOP,
FC_BREAK,
FC_BREAK_NZ_INT,
FC_CONTINUE,
FC_BREAK_Z_INT,
FC_BREAK_NZ
};
enum TextureTypes {
TEXTURE_1D = 1,
TEXTURE_2D,
TEXTURE_3D,
TEXTURE_CUBE,
TEXTURE_RECT,
TEXTURE_SHADOW1D,
TEXTURE_SHADOW2D,
TEXTURE_SHADOWRECT,
TEXTURE_1D_ARRAY,
TEXTURE_2D_ARRAY,
TEXTURE_SHADOW1D_ARRAY,
TEXTURE_SHADOW2D_ARRAY
};
char R600CodeEmitter::ID = 0;
FunctionPass *llvm::createR600CodeEmitterPass(formatted_raw_ostream &OS) {
return new R600CodeEmitter(OS);
}
bool R600CodeEmitter::runOnMachineFunction(MachineFunction &MF) {
TM = &MF.getTarget();
MRI = &MF.getRegInfo();
TRI = static_cast<const R600RegisterInfo *>(TM->getRegisterInfo());
TII = static_cast<const R600InstrInfo *>(TM->getInstrInfo());
const AMDGPUSubtarget &STM = TM->getSubtarget<AMDGPUSubtarget>();
std::string gpu = STM.getDeviceName();
if (STM.dumpCode()) {
MF.dump();
}
for (MachineFunction::iterator BB = MF.begin(), BB_E = MF.end();
BB != BB_E; ++BB) {
MachineBasicBlock &MBB = *BB;
for (MachineBasicBlock::instr_iterator I = MBB.instr_begin(),
E = MBB.instr_end(); I != E; ++I) {
MachineInstr &MI = *I;
if (MI.getNumOperands() > 1 && MI.getOperand(0).isReg() && MI.getOperand(0).isDead()) {
continue;
}
if (TII->isTexOp(MI.getOpcode())) {
EmitTexInstr(MI);
} else if (TII->isFCOp(MI.getOpcode())){
EmitFCInstr(MI);
} else if (MI.getOpcode() == AMDGPU::RETURN ||
MI.getOpcode() == AMDGPU::BUNDLE ||
MI.getOpcode() == AMDGPU::KILL) {
continue;
} else {
switch(MI.getOpcode()) {
case AMDGPU::RAT_WRITE_CACHELESS_eg:
{
uint64_t inst = getBinaryCodeForInstr(MI);
// Set End Of Program bit
// XXX: Need better check of end of program. EOP should be
// encoded in one of the operands of the MI, and it should be
// set in a prior pass.
MachineBasicBlock::iterator NextI = llvm::next(I);
MachineInstr &NextMI = *NextI;
if (NextMI.getOpcode() == AMDGPU::RETURN) {
inst |= (((uint64_t)1) << 53);
}
EmitByte(INSTR_NATIVE);
Emit(inst);
break;
}
case AMDGPU::VTX_READ_PARAM_i32_eg:
case AMDGPU::VTX_READ_PARAM_f32_eg:
case AMDGPU::VTX_READ_GLOBAL_i32_eg:
case AMDGPU::VTX_READ_GLOBAL_f32_eg:
case AMDGPU::VTX_READ_GLOBAL_v4i32_eg:
case AMDGPU::VTX_READ_GLOBAL_v4f32_eg:
{
uint64_t InstWord01 = getBinaryCodeForInstr(MI);
uint32_t InstWord2 = MI.getOperand(2).getImm(); // Offset
EmitByte(INSTR_VTX);
Emit(InstWord01);
Emit(InstWord2);
break;
}
default:
EmitALUInstr(MI);
break;
}
}
}
}
return false;
}
void R600CodeEmitter::EmitALUInstr(MachineInstr &MI)
{
unsigned numOperands = MI.getNumExplicitOperands();
if(MI.findFirstPredOperandIdx() > -1)
numOperands--;
if (TII->HasFlagOperand(MI))
numOperands--;
// Some instructions are just place holder instructions that represent
// operations that the GPU does automatically. They should be ignored.
if (TII->isPlaceHolderOpcode(MI.getOpcode())) {
return;
}
if(MI.getOpcode() == AMDGPU::PRED_X)
numOperands = 2;
// XXX Check if instruction writes a result
if (numOperands < 1) {
return;
}
const MachineOperand dstOp = MI.getOperand(0);
// Emit instruction type
EmitByte(0);
unsigned int opIndex;
for (opIndex = 1; opIndex < numOperands; opIndex++) {
// Literal constants are always stored as the last operand.
if (MI.getOperand(opIndex).isImm() || MI.getOperand(opIndex).isFPImm()) {
break;
}
EmitSrc(MI.getOperand(opIndex), opIndex);
}
// Emit zeros for unused sources
for ( ; opIndex < 4; opIndex++) {
EmitNullBytes(SRC_BYTE_COUNT);
}
EmitDst(dstOp);
EmitALU(MI, numOperands - 1);
}
void R600CodeEmitter::EmitSrc(const MachineOperand & MO, unsigned SrcIdx)
{
const MachineInstr *MI = MO.getParent();
uint32_t value = 0;
// Emit the source select (2 bytes). For GPRs, this is the register index.
// For other potential instruction operands, (e.g. constant registers) the
// value of the source select is defined in the r600isa docs.
if (MO.isReg()) {
unsigned reg = MO.getReg();
EmitTwoBytes(getHWReg(reg));
if (reg == AMDGPU::ALU_LITERAL_X) {
const MachineInstr * parent = MO.getParent();
unsigned immOpIndex = parent->getNumExplicitOperands() - 1;
MachineOperand immOp = parent->getOperand(immOpIndex);
if (immOp.isFPImm()) {
value = immOp.getFPImm()->getValueAPF().bitcastToAPInt().getZExtValue();
} else {
assert(immOp.isImm());
value = immOp.getImm();
}
}
} else {
// XXX: Handle other operand types.
EmitTwoBytes(0);
}
// Emit the source channel (1 byte)
if (MO.isReg()) {
EmitByte(TRI->getHWRegChan(MO.getReg()));
} else {
EmitByte(0);
}
// XXX: Emit isNegated (1 byte)
if ((!(TII->IsFlagSet(*MI, SrcIdx, MO_FLAG_ABS)))
&& (TII->IsFlagSet(*MI, SrcIdx, MO_FLAG_NEG) ||
(MO.isReg() &&
(MO.getReg() == AMDGPU::NEG_ONE || MO.getReg() == AMDGPU::NEG_HALF)))){
EmitByte(1);
} else {
EmitByte(0);
}
// Emit isAbsolute (1 byte)
if (TII->IsFlagSet(*MI, SrcIdx, MO_FLAG_ABS)) {
EmitByte(1);
} else {
EmitByte(0);
}
// XXX: Emit relative addressing mode (1 byte)
EmitByte(0);
// Emit kc_bank, This will be adjusted later by r600_asm
EmitByte(0);
// Emit the literal value, if applicable (4 bytes).
Emit(value);
}
void R600CodeEmitter::EmitDst(const MachineOperand & MO)
{
const MachineInstr *MI = MO.getParent();
if (MO.isReg() && MO.getReg() != AMDGPU::PREDICATE_BIT) {
// Emit the destination register index (1 byte)
EmitByte(getHWReg(MO.getReg()));
// Emit the element of the destination register (1 byte)
EmitByte(TRI->getHWRegChan(MO.getReg()));
// Emit isClamped (1 byte)
if (TII->IsFlagSet(*MI, 0, MO_FLAG_CLAMP)) {
EmitByte(1);
} else {
EmitByte(0);
}
// Emit writemask (1 byte).
if (TII->IsFlagSet(*MI, 0, MO_FLAG_MASK)) {
EmitByte(0);
} else {
EmitByte(1);
}
// XXX: Emit relative addressing mode
EmitByte(0);
} else {
// XXX: Handle other operand types. Are there any for destination regs?
EmitNullBytes(DST_BYTE_COUNT);
}
}
void R600CodeEmitter::EmitALU(MachineInstr &MI, unsigned numSrc)
{
// Emit the instruction (2 bytes)
EmitTwoBytes(getBinaryCodeForInstr(MI));
// Emit IsLast (for this instruction group) (1 byte)
if (MI.isInsideBundle() && !TII->IsFlagSet(MI, 0, MO_FLAG_LAST)) {
EmitByte(0);
} else {
EmitByte(1);
}
// Emit isOp3 (1 byte)
if (numSrc == 3) {
EmitByte(1);
} else {
EmitByte(0);
}
// XXX: Emit push modifier
if(MI.getOperand(1).getTargetFlags() & MO_FLAG_PUSH) {
EmitByte(1);
} else {
EmitByte(0);
}
// XXX: Emit predicate (1 byte)
int predidx = MI.findFirstPredOperandIdx();
if (predidx > -1)
switch(MI.getOperand(predidx).getReg()) {
case AMDGPU::PRED_SEL_ZERO:
EmitByte(2);
break;
case AMDGPU::PRED_SEL_ONE:
EmitByte(3);
break;
default:
EmitByte(0);
break;
}
else {
EmitByte(0);
}
// XXX: Emit bank swizzle. (1 byte) Do we need this? It looks like
// r600_asm.c sets it.
EmitByte(0);
// XXX: Emit bank_swizzle_force (1 byte) Not sure what this is for.
EmitByte(0);
// XXX: Emit OMOD (1 byte) Not implemented.
EmitByte(0);
// XXX: Emit index_mode. I think this is for indirect addressing, so we
// don't need to worry about it.
EmitByte(0);
}
void R600CodeEmitter::EmitTexInstr(MachineInstr &MI)
{
unsigned opcode = MI.getOpcode();
bool hasOffsets = (opcode == AMDGPU::TEX_LD);
unsigned op_offset = hasOffsets ? 3 : 0;
int64_t sampler = MI.getOperand(op_offset+2).getImm();
int64_t textureType = MI.getOperand(op_offset+3).getImm();
unsigned srcSelect[4] = {0, 1, 2, 3};
// Emit instruction type
EmitByte(1);
// Emit instruction
EmitByte(getBinaryCodeForInstr(MI));
// XXX: Emit resource id r600_shader.c uses sampler + 1. Why?
EmitByte(sampler + 1 + 1);
// Emit source register
EmitByte(getHWReg(MI.getOperand(1).getReg()));
// XXX: Emit src isRelativeAddress
EmitByte(0);
// Emit destination register
EmitByte(getHWReg(MI.getOperand(0).getReg()));
// XXX: Emit dst isRealtiveAddress
EmitByte(0);
// XXX: Emit dst select
EmitByte(0); // X
EmitByte(1); // Y
EmitByte(2); // Z
EmitByte(3); // W
// XXX: Emit lod bias
EmitByte(0);
// XXX: Emit coord types
unsigned coordType[4] = {1, 1, 1, 1};
if (textureType == TEXTURE_RECT
|| textureType == TEXTURE_SHADOWRECT) {
coordType[ELEMENT_X] = 0;
coordType[ELEMENT_Y] = 0;
}
if (textureType == TEXTURE_1D_ARRAY
|| textureType == TEXTURE_SHADOW1D_ARRAY) {
if (opcode == AMDGPU::TEX_SAMPLE_C_L || opcode == AMDGPU::TEX_SAMPLE_C_LB) {
coordType[ELEMENT_Y] = 0;
} else {
coordType[ELEMENT_Z] = 0;
srcSelect[ELEMENT_Z] = ELEMENT_Y;
}
} else if (textureType == TEXTURE_2D_ARRAY
|| textureType == TEXTURE_SHADOW2D_ARRAY) {
coordType[ELEMENT_Z] = 0;
}
for (unsigned i = 0; i < 4; i++) {
EmitByte(coordType[i]);
}
// XXX: Emit offsets
if (hasOffsets)
for (unsigned i = 2; i < 5; i++)
EmitByte(MI.getOperand(i).getImm()<<1);
else
EmitNullBytes(3);
// Emit sampler id
EmitByte(sampler);
// XXX:Emit source select
if ((textureType == TEXTURE_SHADOW1D
|| textureType == TEXTURE_SHADOW2D
|| textureType == TEXTURE_SHADOWRECT
|| textureType == TEXTURE_SHADOW1D_ARRAY)
&& opcode != AMDGPU::TEX_SAMPLE_C_L
&& opcode != AMDGPU::TEX_SAMPLE_C_LB) {
srcSelect[ELEMENT_W] = ELEMENT_Z;
}
for (unsigned i = 0; i < 4; i++) {
EmitByte(srcSelect[i]);
}
}
void R600CodeEmitter::EmitFCInstr(MachineInstr &MI)
{
// Emit instruction type
EmitByte(INSTR_FC);
// Emit SRC
unsigned numOperands = MI.getNumOperands();
if (numOperands > 0) {
assert(numOperands == 1);
EmitSrc(MI.getOperand(0), 0);
} else {
EmitNullBytes(SRC_BYTE_COUNT);
}
// Emit FC Instruction
enum FCInstr instr;
switch (MI.getOpcode()) {
case AMDGPU::BREAK_LOGICALZ_f32:
instr = FC_BREAK;
break;
case AMDGPU::BREAK_LOGICALNZ_f32:
instr = FC_BREAK_NZ;
break;
case AMDGPU::BREAK_LOGICALNZ_i32:
instr = FC_BREAK_NZ_INT;
break;
case AMDGPU::BREAK_LOGICALZ_i32:
instr = FC_BREAK_Z_INT;
break;
case AMDGPU::CONTINUE_LOGICALNZ_f32:
case AMDGPU::CONTINUE_LOGICALNZ_i32:
instr = FC_CONTINUE;
break;
case AMDGPU::IF_LOGICALNZ_f32:
instr = FC_IF;
case AMDGPU::IF_LOGICALNZ_i32:
instr = FC_IF_INT;
break;
case AMDGPU::IF_LOGICALZ_f32:
abort();
break;
case AMDGPU::ELSE:
instr = FC_ELSE;
break;
case AMDGPU::ENDIF:
instr = FC_ENDIF;
break;
case AMDGPU::ENDLOOP:
instr = FC_ENDLOOP;
break;
case AMDGPU::WHILELOOP:
instr = FC_BGNLOOP;
break;
default:
abort();
break;
}
EmitByte(instr);
}
void R600CodeEmitter::EmitNullBytes(unsigned int byteCount)
{
for (unsigned int i = 0; i < byteCount; i++) {
EmitByte(0);
}
}
void R600CodeEmitter::EmitByte(unsigned int byte)
{
_OS.write((uint8_t) byte & 0xff);
}
void R600CodeEmitter::EmitTwoBytes(unsigned int bytes)
{
_OS.write((uint8_t) (bytes & 0xff));
_OS.write((uint8_t) ((bytes >> 8) & 0xff));
}
void R600CodeEmitter::Emit(uint32_t value)
{
for (unsigned i = 0; i < 4; i++) {
_OS.write((uint8_t) ((value >> (8 * i)) & 0xff));
}
}
void R600CodeEmitter::Emit(uint64_t value)
{
for (unsigned i = 0; i < 8; i++) {
EmitByte((value >> (8 * i)) & 0xff);
}
}
unsigned R600CodeEmitter::getHWReg(unsigned regNo) const
{
unsigned HWReg;
HWReg = TRI->getHWRegIndex(regNo);
if (AMDGPU::R600_CReg32RegClass.contains(regNo)) {
HWReg += 512;
}
return HWReg;
}
uint64_t R600CodeEmitter::getMachineOpValue(const MachineInstr &MI,
const MachineOperand &MO) const
{
if (MO.isReg()) {
return getHWReg(MO.getReg());
} else {
return MO.getImm();
}
}
#include "AMDGPUGenCodeEmitter.inc"
|