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
|
// Copyright 2020 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/logging/log.h"
#include "common/microprofile.h"
#include "common/x64/xbyak_util.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/macro/macro_interpreter.h"
#include "video_core/macro/macro_jit_x64.h"
MICROPROFILE_DEFINE(MacroJitCompile, "GPU", "Compile macro JIT", MP_RGB(173, 255, 47));
MICROPROFILE_DEFINE(MacroJitExecute, "GPU", "Execute macro JIT", MP_RGB(255, 255, 0));
namespace Tegra {
static const Xbyak::Reg64 STATE = Xbyak::util::rbx;
static const Xbyak::Reg32 RESULT = Xbyak::util::ebp;
static const Xbyak::Reg64 PARAMETERS = Xbyak::util::r12;
static const Xbyak::Reg32 METHOD_ADDRESS = Xbyak::util::r14d;
static const Xbyak::Reg64 BRANCH_HOLDER = Xbyak::util::r15;
static const std::bitset<32> PERSISTENT_REGISTERS = Common::X64::BuildRegSet({
STATE,
RESULT,
PARAMETERS,
METHOD_ADDRESS,
BRANCH_HOLDER,
});
MacroJITx64::MacroJITx64(Engines::Maxwell3D& maxwell3d) : maxwell3d(maxwell3d) {}
std::unique_ptr<CachedMacro> MacroJITx64::Compile(const std::vector<u32>& code) {
return std::make_unique<MacroJITx64Impl>(maxwell3d, code);
}
MacroJITx64Impl::MacroJITx64Impl(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& code)
: Xbyak::CodeGenerator(MAX_CODE_SIZE), code(code), maxwell3d(maxwell3d) {
Compile();
}
MacroJITx64Impl::~MacroJITx64Impl() = default;
void MacroJITx64Impl::Execute(const std::vector<u32>& parameters, u32 method) {
MICROPROFILE_SCOPE(MacroJitExecute);
ASSERT_OR_EXECUTE(program != nullptr, { return; });
JITState state{};
state.maxwell3d = &maxwell3d;
state.registers = {};
program(&state, parameters.data());
}
void MacroJITx64Impl::Compile_ALU(Macro::Opcode opcode) {
const bool is_a_zero = opcode.src_a == 0;
const bool is_b_zero = opcode.src_b == 0;
const bool valid_operation = !is_a_zero && !is_b_zero;
const bool is_move_operation = !is_a_zero && is_b_zero;
const bool has_zero_register = is_a_zero || is_b_zero;
Xbyak::Reg32 src_a;
Xbyak::Reg32 src_b;
if (!optimizer.zero_reg_skip) {
src_a = Compile_GetRegister(opcode.src_a, RESULT);
src_b = Compile_GetRegister(opcode.src_b, eax);
} else {
if (!is_a_zero) {
src_a = Compile_GetRegister(opcode.src_a, RESULT);
}
if (!is_b_zero) {
src_b = Compile_GetRegister(opcode.src_b, eax);
}
}
Xbyak::Label skip_carry{};
bool has_emitted = false;
switch (opcode.alu_operation) {
case Macro::ALUOperation::Add:
if (optimizer.zero_reg_skip) {
if (valid_operation) {
add(src_a, src_b);
}
} else {
add(src_a, src_b);
}
if (!optimizer.can_skip_carry) {
setc(byte[STATE + offsetof(JITState, carry_flag)]);
}
break;
case Macro::ALUOperation::AddWithCarry:
bt(dword[STATE + offsetof(JITState, carry_flag)], 0);
adc(src_a, src_b);
setc(byte[STATE + offsetof(JITState, carry_flag)]);
break;
case Macro::ALUOperation::Subtract:
if (optimizer.zero_reg_skip) {
if (valid_operation) {
sub(src_a, src_b);
has_emitted = true;
}
} else {
sub(src_a, src_b);
has_emitted = true;
}
if (!optimizer.can_skip_carry && has_emitted) {
setc(byte[STATE + offsetof(JITState, carry_flag)]);
}
break;
case Macro::ALUOperation::SubtractWithBorrow:
bt(dword[STATE + offsetof(JITState, carry_flag)], 0);
sbb(src_a, src_b);
setc(byte[STATE + offsetof(JITState, carry_flag)]);
break;
case Macro::ALUOperation::Xor:
if (optimizer.zero_reg_skip) {
if (valid_operation) {
xor_(src_a, src_b);
}
} else {
xor_(src_a, src_b);
}
break;
case Macro::ALUOperation::Or:
if (optimizer.zero_reg_skip) {
if (valid_operation) {
or_(src_a, src_b);
}
} else {
or_(src_a, src_b);
}
break;
case Macro::ALUOperation::And:
if (optimizer.zero_reg_skip) {
if (!has_zero_register) {
and_(src_a, src_b);
}
} else {
and_(src_a, src_b);
}
break;
case Macro::ALUOperation::AndNot:
if (optimizer.zero_reg_skip) {
if (!is_a_zero) {
not_(src_b);
and_(src_a, src_b);
}
} else {
not_(src_b);
and_(src_a, src_b);
}
break;
case Macro::ALUOperation::Nand:
if (optimizer.zero_reg_skip) {
if (!is_a_zero) {
and_(src_a, src_b);
not_(src_a);
}
} else {
and_(src_a, src_b);
not_(src_a);
}
break;
default:
UNIMPLEMENTED_MSG("Unimplemented ALU operation {}",
static_cast<std::size_t>(opcode.alu_operation.Value()));
break;
}
Compile_ProcessResult(opcode.result_operation, opcode.dst);
}
void MacroJITx64Impl::Compile_AddImmediate(Macro::Opcode opcode) {
if (optimizer.skip_dummy_addimmediate) {
// Games tend to use this as an exit instruction placeholder. It's to encode an instruction
// without doing anything. In our case we can just not emit anything.
if (opcode.result_operation == Macro::ResultOperation::Move && opcode.dst == 0) {
return;
}
}
// Check for redundant moves
if (optimizer.optimize_for_method_move &&
opcode.result_operation == Macro::ResultOperation::MoveAndSetMethod) {
if (next_opcode.has_value()) {
const auto next = *next_opcode;
if (next.result_operation == Macro::ResultOperation::MoveAndSetMethod) {
return;
}
}
}
if (optimizer.zero_reg_skip && opcode.src_a == 0) {
if (opcode.immediate == 0) {
xor_(RESULT, RESULT);
} else {
mov(RESULT, opcode.immediate);
}
} else {
auto result = Compile_GetRegister(opcode.src_a, RESULT);
if (opcode.immediate > 2) {
add(result, opcode.immediate);
} else if (opcode.immediate == 1) {
inc(result);
} else if (opcode.immediate < 0) {
sub(result, opcode.immediate * -1);
}
}
Compile_ProcessResult(opcode.result_operation, opcode.dst);
}
void MacroJITx64Impl::Compile_ExtractInsert(Macro::Opcode opcode) {
auto dst = Compile_GetRegister(opcode.src_a, RESULT);
auto src = Compile_GetRegister(opcode.src_b, eax);
if (opcode.bf_src_bit != 0 && opcode.bf_src_bit != 31) {
shr(src, opcode.bf_src_bit);
} else if (opcode.bf_src_bit == 31) {
xor_(src, src);
}
// Don't bother masking the whole register since we're using a 32 bit register
if (opcode.bf_size != 31 && opcode.bf_size != 0) {
and_(src, opcode.GetBitfieldMask());
} else if (opcode.bf_size == 0) {
xor_(src, src);
}
if (opcode.bf_dst_bit != 31 && opcode.bf_dst_bit != 0) {
shl(src, opcode.bf_dst_bit);
} else if (opcode.bf_dst_bit == 31) {
xor_(src, src);
}
const u32 mask = ~(opcode.GetBitfieldMask() << opcode.bf_dst_bit);
if (mask != 0xffffffff) {
and_(dst, mask);
}
or_(dst, src);
Compile_ProcessResult(opcode.result_operation, opcode.dst);
}
void MacroJITx64Impl::Compile_ExtractShiftLeftImmediate(Macro::Opcode opcode) {
auto dst = Compile_GetRegister(opcode.src_a, eax);
auto src = Compile_GetRegister(opcode.src_b, RESULT);
shr(src, al);
if (opcode.bf_size != 0 && opcode.bf_size != 31) {
and_(src, opcode.GetBitfieldMask());
} else if (opcode.bf_size == 0) {
xor_(src, src);
}
if (opcode.bf_dst_bit != 0 && opcode.bf_dst_bit != 31) {
shl(src, opcode.bf_dst_bit);
} else if (opcode.bf_dst_bit == 31) {
xor_(src, src);
}
Compile_ProcessResult(opcode.result_operation, opcode.dst);
}
void MacroJITx64Impl::Compile_ExtractShiftLeftRegister(Macro::Opcode opcode) {
auto dst = Compile_GetRegister(opcode.src_a, eax);
auto src = Compile_GetRegister(opcode.src_b, RESULT);
if (opcode.bf_src_bit != 0) {
shr(src, opcode.bf_src_bit);
}
if (opcode.bf_size != 31) {
and_(src, opcode.GetBitfieldMask());
}
shl(src, al);
Compile_ProcessResult(opcode.result_operation, opcode.dst);
}
static u32 Read(Engines::Maxwell3D* maxwell3d, u32 method) {
return maxwell3d->GetRegisterValue(method);
}
static void Send(Engines::Maxwell3D* maxwell3d, Macro::MethodAddress method_address, u32 value) {
maxwell3d->CallMethodFromMME(method_address.address, value);
}
void MacroJITx64Impl::Compile_Read(Macro::Opcode opcode) {
if (optimizer.zero_reg_skip && opcode.src_a == 0) {
if (opcode.immediate == 0) {
xor_(RESULT, RESULT);
} else {
mov(RESULT, opcode.immediate);
}
} else {
auto result = Compile_GetRegister(opcode.src_a, RESULT);
if (opcode.immediate > 2) {
add(result, opcode.immediate);
} else if (opcode.immediate == 1) {
inc(result);
} else if (opcode.immediate < 0) {
sub(result, opcode.immediate * -1);
}
}
Common::X64::ABI_PushRegistersAndAdjustStackGPS(*this, PersistentCallerSavedRegs(), 0);
mov(Common::X64::ABI_PARAM1, qword[STATE]);
mov(Common::X64::ABI_PARAM2, RESULT);
Common::X64::CallFarFunction(*this, &Read);
Common::X64::ABI_PopRegistersAndAdjustStackGPS(*this, PersistentCallerSavedRegs(), 0);
mov(RESULT, Common::X64::ABI_RETURN.cvt32());
Compile_ProcessResult(opcode.result_operation, opcode.dst);
}
void Tegra::MacroJITx64Impl::Compile_Send(Xbyak::Reg32 value) {
Common::X64::ABI_PushRegistersAndAdjustStackGPS(*this, PersistentCallerSavedRegs(), 0);
mov(Common::X64::ABI_PARAM1, qword[STATE]);
mov(Common::X64::ABI_PARAM2, METHOD_ADDRESS);
mov(Common::X64::ABI_PARAM3, value);
Common::X64::CallFarFunction(*this, &Send);
Common::X64::ABI_PopRegistersAndAdjustStackGPS(*this, PersistentCallerSavedRegs(), 0);
Xbyak::Label dont_process{};
// Get increment
test(METHOD_ADDRESS, 0x3f000);
// If zero, method address doesn't update
je(dont_process);
mov(ecx, METHOD_ADDRESS);
and_(METHOD_ADDRESS, 0xfff);
shr(ecx, 12);
and_(ecx, 0x3f);
lea(eax, ptr[rcx + METHOD_ADDRESS.cvt64()]);
sal(ecx, 12);
or_(eax, ecx);
mov(METHOD_ADDRESS, eax);
L(dont_process);
}
void Tegra::MacroJITx64Impl::Compile_Branch(Macro::Opcode opcode) {
ASSERT_MSG(!is_delay_slot, "Executing a branch in a delay slot is not valid");
const s32 jump_address =
static_cast<s32>(pc) + static_cast<s32>(opcode.GetBranchTarget() / sizeof(s32));
Xbyak::Label end;
auto value = Compile_GetRegister(opcode.src_a, eax);
test(value, value);
if (optimizer.has_delayed_pc) {
switch (opcode.branch_condition) {
case Macro::BranchCondition::Zero:
jne(end, T_NEAR);
break;
case Macro::BranchCondition::NotZero:
je(end, T_NEAR);
break;
}
if (opcode.branch_annul) {
xor_(BRANCH_HOLDER, BRANCH_HOLDER);
jmp(labels[jump_address], T_NEAR);
} else {
Xbyak::Label handle_post_exit{};
Xbyak::Label skip{};
jmp(skip, T_NEAR);
if (opcode.is_exit) {
L(handle_post_exit);
// Execute 1 instruction
mov(BRANCH_HOLDER, end_of_code);
// Jump to next instruction to skip delay slot check
jmp(labels[jump_address], T_NEAR);
} else {
L(handle_post_exit);
xor_(BRANCH_HOLDER, BRANCH_HOLDER);
jmp(labels[jump_address], T_NEAR);
}
L(skip);
mov(BRANCH_HOLDER, handle_post_exit);
jmp(delay_skip[pc], T_NEAR);
}
} else {
switch (opcode.branch_condition) {
case Macro::BranchCondition::Zero:
je(labels[jump_address], T_NEAR);
break;
case Macro::BranchCondition::NotZero:
jne(labels[jump_address], T_NEAR);
break;
}
}
L(end);
}
void Tegra::MacroJITx64Impl::Optimizer_ScanFlags() {
optimizer.can_skip_carry = true;
optimizer.has_delayed_pc = false;
for (auto raw_op : code) {
Macro::Opcode op{};
op.raw = raw_op;
if (op.operation == Macro::Operation::ALU) {
// Scan for any ALU operations which actually use the carry flag, if they don't exist in
// our current code we can skip emitting the carry flag handling operations
if (op.alu_operation == Macro::ALUOperation::AddWithCarry ||
op.alu_operation == Macro::ALUOperation::SubtractWithBorrow) {
optimizer.can_skip_carry = false;
}
}
if (op.operation == Macro::Operation::Branch) {
if (!op.branch_annul) {
optimizer.has_delayed_pc = true;
}
}
}
}
void MacroJITx64Impl::Compile() {
MICROPROFILE_SCOPE(MacroJitCompile);
bool keep_executing = true;
labels.fill(Xbyak::Label());
Common::X64::ABI_PushRegistersAndAdjustStackGPS(*this, Common::X64::ABI_ALL_CALLEE_SAVED, 8);
// JIT state
mov(STATE, Common::X64::ABI_PARAM1);
mov(PARAMETERS, Common::X64::ABI_PARAM2);
xor_(RESULT, RESULT);
xor_(METHOD_ADDRESS, METHOD_ADDRESS);
xor_(BRANCH_HOLDER, BRANCH_HOLDER);
mov(dword[STATE + offsetof(JITState, registers) + 4], Compile_FetchParameter());
// Track get register for zero registers and mark it as no-op
optimizer.zero_reg_skip = true;
// AddImmediate tends to be used as a NOP instruction, if we detect this we can
// completely skip the entire code path and no emit anything
optimizer.skip_dummy_addimmediate = true;
// SMO tends to emit a lot of unnecessary method moves, we can mitigate this by only emitting
// one if our register isn't "dirty"
optimizer.optimize_for_method_move = true;
// Check to see if we can skip emitting certain instructions
Optimizer_ScanFlags();
const u32 op_count = static_cast<u32>(code.size());
for (u32 i = 0; i < op_count; i++) {
if (i < op_count - 1) {
pc = i + 1;
next_opcode = GetOpCode();
} else {
next_opcode = {};
}
pc = i;
Compile_NextInstruction();
}
L(end_of_code);
Common::X64::ABI_PopRegistersAndAdjustStackGPS(*this, Common::X64::ABI_ALL_CALLEE_SAVED, 8);
ret();
ready();
program = getCode<ProgramType>();
}
bool MacroJITx64Impl::Compile_NextInstruction() {
const auto opcode = GetOpCode();
if (labels[pc].getAddress()) {
return false;
}
L(labels[pc]);
switch (opcode.operation) {
case Macro::Operation::ALU:
Compile_ALU(opcode);
break;
case Macro::Operation::AddImmediate:
Compile_AddImmediate(opcode);
break;
case Macro::Operation::ExtractInsert:
Compile_ExtractInsert(opcode);
break;
case Macro::Operation::ExtractShiftLeftImmediate:
Compile_ExtractShiftLeftImmediate(opcode);
break;
case Macro::Operation::ExtractShiftLeftRegister:
Compile_ExtractShiftLeftRegister(opcode);
break;
case Macro::Operation::Read:
Compile_Read(opcode);
break;
case Macro::Operation::Branch:
Compile_Branch(opcode);
break;
default:
UNIMPLEMENTED_MSG("Unimplemented opcode {}", opcode.operation.Value());
break;
}
if (optimizer.has_delayed_pc) {
if (opcode.is_exit) {
mov(rax, end_of_code);
test(BRANCH_HOLDER, BRANCH_HOLDER);
cmove(BRANCH_HOLDER, rax);
// Jump to next instruction to skip delay slot check
je(labels[pc + 1], T_NEAR);
} else {
// TODO(ogniK): Optimize delay slot branching
Xbyak::Label no_delay_slot{};
test(BRANCH_HOLDER, BRANCH_HOLDER);
je(no_delay_slot, T_NEAR);
mov(rax, BRANCH_HOLDER);
xor_(BRANCH_HOLDER, BRANCH_HOLDER);
jmp(rax);
L(no_delay_slot);
}
L(delay_skip[pc]);
if (opcode.is_exit) {
return false;
}
} else {
test(BRANCH_HOLDER, BRANCH_HOLDER);
jne(end_of_code, T_NEAR);
if (opcode.is_exit) {
inc(BRANCH_HOLDER);
return false;
}
}
return true;
}
Xbyak::Reg32 Tegra::MacroJITx64Impl::Compile_FetchParameter() {
mov(eax, dword[PARAMETERS]);
add(PARAMETERS, sizeof(u32));
return eax;
}
Xbyak::Reg32 MacroJITx64Impl::Compile_GetRegister(u32 index, Xbyak::Reg32 dst) {
if (index == 0) {
// Register 0 is always zero
xor_(dst, dst);
} else {
mov(dst, dword[STATE + offsetof(JITState, registers) + index * sizeof(u32)]);
}
return dst;
}
void MacroJITx64Impl::Compile_ProcessResult(Macro::ResultOperation operation, u32 reg) {
auto SetRegister = [=](u32 reg, Xbyak::Reg32 result) {
// Register 0 is supposed to always return 0. NOP is implemented as a store to the zero
// register.
if (reg == 0) {
return;
}
mov(dword[STATE + offsetof(JITState, registers) + reg * sizeof(u32)], result);
};
auto SetMethodAddress = [=](Xbyak::Reg32 reg) { mov(METHOD_ADDRESS, reg); };
switch (operation) {
case Macro::ResultOperation::IgnoreAndFetch:
SetRegister(reg, Compile_FetchParameter());
break;
case Macro::ResultOperation::Move:
SetRegister(reg, RESULT);
break;
case Macro::ResultOperation::MoveAndSetMethod:
SetRegister(reg, RESULT);
SetMethodAddress(RESULT);
break;
case Macro::ResultOperation::FetchAndSend:
// Fetch parameter and send result.
SetRegister(reg, Compile_FetchParameter());
Compile_Send(RESULT);
break;
case Macro::ResultOperation::MoveAndSend:
// Move and send result.
SetRegister(reg, RESULT);
Compile_Send(RESULT);
break;
case Macro::ResultOperation::FetchAndSetMethod:
// Fetch parameter and use result as Method Address.
SetRegister(reg, Compile_FetchParameter());
SetMethodAddress(RESULT);
break;
case Macro::ResultOperation::MoveAndSetMethodFetchAndSend:
// Move result and use as Method Address, then fetch and send parameter.
SetRegister(reg, RESULT);
SetMethodAddress(RESULT);
Compile_Send(Compile_FetchParameter());
break;
case Macro::ResultOperation::MoveAndSetMethodSend:
// Move result and use as Method Address, then send bits 12:17 of result.
SetRegister(reg, RESULT);
SetMethodAddress(RESULT);
shr(RESULT, 12);
and_(RESULT, 0b111111);
Compile_Send(RESULT);
break;
default:
UNIMPLEMENTED_MSG("Unimplemented macro operation {}", static_cast<std::size_t>(operation));
}
}
Macro::Opcode MacroJITx64Impl::GetOpCode() const {
ASSERT(pc < code.size());
return {code[pc]};
}
std::bitset<32> MacroJITx64Impl::PersistentCallerSavedRegs() const {
return PERSISTENT_REGISTERS & Common::X64::ABI_ALL_CALLER_SAVED;
}
} // namespace Tegra
|