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
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
|
// Copyright 2015 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <smmintrin.h>
#include "common/x64/abi.h"
#include "common/x64/cpu_detect.h"
#include "common/x64/emitter.h"
#include "shader.h"
#include "shader_jit_x64.h"
#include "video_core/pica_state.h"
namespace Pica {
namespace Shader {
using namespace Gen;
typedef void (JitCompiler::*JitFunction)(Instruction instr);
const JitFunction instr_table[64] = {
&JitCompiler::Compile_ADD, // add
&JitCompiler::Compile_DP3, // dp3
&JitCompiler::Compile_DP4, // dp4
&JitCompiler::Compile_DPH, // dph
nullptr, // unknown
&JitCompiler::Compile_EX2, // ex2
&JitCompiler::Compile_LG2, // lg2
nullptr, // unknown
&JitCompiler::Compile_MUL, // mul
&JitCompiler::Compile_SGE, // sge
&JitCompiler::Compile_SLT, // slt
&JitCompiler::Compile_FLR, // flr
&JitCompiler::Compile_MAX, // max
&JitCompiler::Compile_MIN, // min
&JitCompiler::Compile_RCP, // rcp
&JitCompiler::Compile_RSQ, // rsq
nullptr, // unknown
nullptr, // unknown
&JitCompiler::Compile_MOVA, // mova
&JitCompiler::Compile_MOV, // mov
nullptr, // unknown
nullptr, // unknown
nullptr, // unknown
nullptr, // unknown
&JitCompiler::Compile_DPH, // dphi
nullptr, // unknown
&JitCompiler::Compile_SGE, // sgei
&JitCompiler::Compile_SLT, // slti
nullptr, // unknown
nullptr, // unknown
nullptr, // unknown
nullptr, // unknown
nullptr, // unknown
&JitCompiler::Compile_NOP, // nop
&JitCompiler::Compile_END, // end
nullptr, // break
&JitCompiler::Compile_CALL, // call
&JitCompiler::Compile_CALLC, // callc
&JitCompiler::Compile_CALLU, // callu
&JitCompiler::Compile_IF, // ifu
&JitCompiler::Compile_IF, // ifc
&JitCompiler::Compile_LOOP, // loop
nullptr, // emit
nullptr, // sete
&JitCompiler::Compile_JMP, // jmpc
&JitCompiler::Compile_JMP, // jmpu
&JitCompiler::Compile_CMP, // cmp
&JitCompiler::Compile_CMP, // cmp
&JitCompiler::Compile_MAD, // madi
&JitCompiler::Compile_MAD, // madi
&JitCompiler::Compile_MAD, // madi
&JitCompiler::Compile_MAD, // madi
&JitCompiler::Compile_MAD, // madi
&JitCompiler::Compile_MAD, // madi
&JitCompiler::Compile_MAD, // madi
&JitCompiler::Compile_MAD, // madi
&JitCompiler::Compile_MAD, // mad
&JitCompiler::Compile_MAD, // mad
&JitCompiler::Compile_MAD, // mad
&JitCompiler::Compile_MAD, // mad
&JitCompiler::Compile_MAD, // mad
&JitCompiler::Compile_MAD, // mad
&JitCompiler::Compile_MAD, // mad
&JitCompiler::Compile_MAD, // mad
};
// The following is used to alias some commonly used registers. Generally, RAX-RDX and XMM0-XMM3 can
// be used as scratch registers within a compiler function. The other registers have designated
// purposes, as documented below:
/// Pointer to the uniform memory
static const X64Reg UNIFORMS = R9;
/// The two 32-bit VS address offset registers set by the MOVA instruction
static const X64Reg ADDROFFS_REG_0 = R10;
static const X64Reg ADDROFFS_REG_1 = R11;
/// VS loop count register
static const X64Reg LOOPCOUNT_REG = R12;
/// Current VS loop iteration number (we could probably use LOOPCOUNT_REG, but this quicker)
static const X64Reg LOOPCOUNT = RSI;
/// Number to increment LOOPCOUNT_REG by on each loop iteration
static const X64Reg LOOPINC = RDI;
/// Result of the previous CMP instruction for the X-component comparison
static const X64Reg COND0 = R13;
/// Result of the previous CMP instruction for the Y-component comparison
static const X64Reg COND1 = R14;
/// Pointer to the UnitState instance for the current VS unit
static const X64Reg REGISTERS = R15;
/// SIMD scratch register
static const X64Reg SCRATCH = XMM0;
/// Loaded with the first swizzled source register, otherwise can be used as a scratch register
static const X64Reg SRC1 = XMM1;
/// Loaded with the second swizzled source register, otherwise can be used as a scratch register
static const X64Reg SRC2 = XMM2;
/// Loaded with the third swizzled source register, otherwise can be used as a scratch register
static const X64Reg SRC3 = XMM3;
/// Additional scratch register
static const X64Reg SCRATCH2 = XMM4;
/// Constant vector of [1.0f, 1.0f, 1.0f, 1.0f], used to efficiently set a vector to one
static const X64Reg ONE = XMM14;
/// Constant vector of [-0.f, -0.f, -0.f, -0.f], used to efficiently negate a vector with XOR
static const X64Reg NEGBIT = XMM15;
// State registers that must not be modified by external functions calls
// Scratch registers, e.g., SRC1 and SCRATCH, have to be saved on the side if needed
static const BitSet32 persistent_regs = {
UNIFORMS, REGISTERS, // Pointers to register blocks
ADDROFFS_REG_0, ADDROFFS_REG_1, LOOPCOUNT_REG, COND0, COND1, // Cached registers
ONE+16, NEGBIT+16, // Constants
};
/// Raw constant for the source register selector that indicates no swizzling is performed
static const u8 NO_SRC_REG_SWIZZLE = 0x1b;
/// Raw constant for the destination register enable mask that indicates all components are enabled
static const u8 NO_DEST_REG_MASK = 0xf;
/**
* Get the vertex shader instruction for a given offset in the current shader program
* @param offset Offset in the current shader program of the instruction
* @return Instruction at the specified offset
*/
static Instruction GetVertexShaderInstruction(size_t offset) {
return { g_state.vs.program_code[offset] };
}
static void LogCritical(const char* msg) {
LOG_CRITICAL(HW_GPU, msg);
}
void JitCompiler::RuntimeAssert(bool condition, const char* msg) {
if (!condition) {
ABI_CallFunctionP(reinterpret_cast<const void*>(LogCritical), const_cast<char*>(msg));
}
}
/**
* Loads and swizzles a source register into the specified XMM register.
* @param instr VS instruction, used for determining how to load the source register
* @param src_num Number indicating which source register to load (1 = src1, 2 = src2, 3 = src3)
* @param src_reg SourceRegister object corresponding to the source register to load
* @param dest Destination XMM register to store the loaded, swizzled source register
*/
void JitCompiler::Compile_SwizzleSrc(Instruction instr, unsigned src_num, SourceRegister src_reg, X64Reg dest) {
X64Reg src_ptr;
size_t src_offset;
if (src_reg.GetRegisterType() == RegisterType::FloatUniform) {
src_ptr = UNIFORMS;
src_offset = src_reg.GetIndex() * sizeof(float24) * 4;
} else {
src_ptr = REGISTERS;
src_offset = UnitState<false>::InputOffset(src_reg);
}
int src_offset_disp = (int)src_offset;
ASSERT_MSG(src_offset == src_offset_disp, "Source register offset too large for int type");
unsigned operand_desc_id;
const bool is_inverted = (0 != (instr.opcode.Value().GetInfo().subtype & OpCode::Info::SrcInversed));
unsigned address_register_index;
unsigned offset_src;
if (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::MAD ||
instr.opcode.Value().EffectiveOpCode() == OpCode::Id::MADI) {
operand_desc_id = instr.mad.operand_desc_id;
offset_src = is_inverted ? 3 : 2;
address_register_index = instr.mad.address_register_index;
} else {
operand_desc_id = instr.common.operand_desc_id;
offset_src = is_inverted ? 2 : 1;
address_register_index = instr.common.address_register_index;
}
if (src_num == offset_src && address_register_index != 0) {
switch (address_register_index) {
case 1: // address offset 1
MOVAPS(dest, MComplex(src_ptr, ADDROFFS_REG_0, SCALE_1, src_offset_disp));
break;
case 2: // address offset 2
MOVAPS(dest, MComplex(src_ptr, ADDROFFS_REG_1, SCALE_1, src_offset_disp));
break;
case 3: // address offset 3
MOVAPS(dest, MComplex(src_ptr, LOOPCOUNT_REG, SCALE_1, src_offset_disp));
break;
default:
UNREACHABLE();
break;
}
} else {
// Load the source
MOVAPS(dest, MDisp(src_ptr, src_offset_disp));
}
SwizzlePattern swiz = { g_state.vs.swizzle_data[operand_desc_id] };
// Generate instructions for source register swizzling as needed
u8 sel = swiz.GetRawSelector(src_num);
if (sel != NO_SRC_REG_SWIZZLE) {
// Selector component order needs to be reversed for the SHUFPS instruction
sel = ((sel & 0xc0) >> 6) | ((sel & 3) << 6) | ((sel & 0xc) << 2) | ((sel & 0x30) >> 2);
// Shuffle inputs for swizzle
SHUFPS(dest, R(dest), sel);
}
// If the source register should be negated, flip the negative bit using XOR
const bool negate[] = { swiz.negate_src1, swiz.negate_src2, swiz.negate_src3 };
if (negate[src_num - 1]) {
XORPS(dest, R(NEGBIT));
}
}
void JitCompiler::Compile_DestEnable(Instruction instr,X64Reg src) {
DestRegister dest;
unsigned operand_desc_id;
if (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::MAD ||
instr.opcode.Value().EffectiveOpCode() == OpCode::Id::MADI) {
operand_desc_id = instr.mad.operand_desc_id;
dest = instr.mad.dest.Value();
} else {
operand_desc_id = instr.common.operand_desc_id;
dest = instr.common.dest.Value();
}
SwizzlePattern swiz = { g_state.vs.swizzle_data[operand_desc_id] };
int dest_offset_disp = (int)UnitState<false>::OutputOffset(dest);
ASSERT_MSG(dest_offset_disp == UnitState<false>::OutputOffset(dest), "Destinaton offset too large for int type");
// If all components are enabled, write the result to the destination register
if (swiz.dest_mask == NO_DEST_REG_MASK) {
// Store dest back to memory
MOVAPS(MDisp(REGISTERS, dest_offset_disp), src);
} else {
// Not all components are enabled, so mask the result when storing to the destination register...
MOVAPS(SCRATCH, MDisp(REGISTERS, dest_offset_disp));
if (Common::GetCPUCaps().sse4_1) {
u8 mask = ((swiz.dest_mask & 1) << 3) | ((swiz.dest_mask & 8) >> 3) | ((swiz.dest_mask & 2) << 1) | ((swiz.dest_mask & 4) >> 1);
BLENDPS(SCRATCH, R(src), mask);
} else {
MOVAPS(SCRATCH2, R(src));
UNPCKHPS(SCRATCH2, R(SCRATCH)); // Unpack X/Y components of source and destination
UNPCKLPS(SCRATCH, R(src)); // Unpack Z/W components of source and destination
// Compute selector to selectively copy source components to destination for SHUFPS instruction
u8 sel = ((swiz.DestComponentEnabled(0) ? 1 : 0) << 0) |
((swiz.DestComponentEnabled(1) ? 3 : 2) << 2) |
((swiz.DestComponentEnabled(2) ? 0 : 1) << 4) |
((swiz.DestComponentEnabled(3) ? 2 : 3) << 6);
SHUFPS(SCRATCH, R(SCRATCH2), sel);
}
// Store dest back to memory
MOVAPS(MDisp(REGISTERS, dest_offset_disp), SCRATCH);
}
}
void JitCompiler::Compile_SanitizedMul(Gen::X64Reg src1, Gen::X64Reg src2, Gen::X64Reg scratch) {
MOVAPS(scratch, R(src1));
CMPPS(scratch, R(src2), CMP_ORD);
MULPS(src1, R(src2));
MOVAPS(src2, R(src1));
CMPPS(src2, R(src2), CMP_UNORD);
XORPS(scratch, R(src2));
ANDPS(src1, R(scratch));
}
void JitCompiler::Compile_EvaluateCondition(Instruction instr) {
// Note: NXOR is used below to check for equality
switch (instr.flow_control.op) {
case Instruction::FlowControlType::Or:
MOV(32, R(RAX), R(COND0));
MOV(32, R(RBX), R(COND1));
XOR(32, R(RAX), Imm32(instr.flow_control.refx.Value() ^ 1));
XOR(32, R(RBX), Imm32(instr.flow_control.refy.Value() ^ 1));
OR(32, R(RAX), R(RBX));
break;
case Instruction::FlowControlType::And:
MOV(32, R(RAX), R(COND0));
MOV(32, R(RBX), R(COND1));
XOR(32, R(RAX), Imm32(instr.flow_control.refx.Value() ^ 1));
XOR(32, R(RBX), Imm32(instr.flow_control.refy.Value() ^ 1));
AND(32, R(RAX), R(RBX));
break;
case Instruction::FlowControlType::JustX:
MOV(32, R(RAX), R(COND0));
XOR(32, R(RAX), Imm32(instr.flow_control.refx.Value() ^ 1));
break;
case Instruction::FlowControlType::JustY:
MOV(32, R(RAX), R(COND1));
XOR(32, R(RAX), Imm32(instr.flow_control.refy.Value() ^ 1));
break;
}
}
void JitCompiler::Compile_UniformCondition(Instruction instr) {
int offset = offsetof(decltype(g_state.vs.uniforms), b) + (instr.flow_control.bool_uniform_id * sizeof(bool));
CMP(sizeof(bool) * 8, MDisp(UNIFORMS, offset), Imm8(0));
}
BitSet32 JitCompiler::PersistentCallerSavedRegs() {
return persistent_regs & ABI_ALL_CALLER_SAVED;
}
void JitCompiler::Compile_ADD(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
ADDPS(SRC1, R(SRC2));
Compile_DestEnable(instr, SRC1);
}
void JitCompiler::Compile_DP3(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
Compile_SanitizedMul(SRC1, SRC2, SCRATCH);
MOVAPS(SRC2, R(SRC1));
SHUFPS(SRC2, R(SRC2), _MM_SHUFFLE(1, 1, 1, 1));
MOVAPS(SRC3, R(SRC1));
SHUFPS(SRC3, R(SRC3), _MM_SHUFFLE(2, 2, 2, 2));
SHUFPS(SRC1, R(SRC1), _MM_SHUFFLE(0, 0, 0, 0));
ADDPS(SRC1, R(SRC2));
ADDPS(SRC1, R(SRC3));
Compile_DestEnable(instr, SRC1);
}
void JitCompiler::Compile_DP4(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
Compile_SanitizedMul(SRC1, SRC2, SCRATCH);
MOVAPS(SRC2, R(SRC1));
SHUFPS(SRC1, R(SRC1), _MM_SHUFFLE(2, 3, 0, 1)); // XYZW -> ZWXY
ADDPS(SRC1, R(SRC2));
MOVAPS(SRC2, R(SRC1));
SHUFPS(SRC1, R(SRC1), _MM_SHUFFLE(0, 1, 2, 3)); // XYZW -> WZYX
ADDPS(SRC1, R(SRC2));
Compile_DestEnable(instr, SRC1);
}
void JitCompiler::Compile_DPH(Instruction instr) {
if (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::DPHI) {
Compile_SwizzleSrc(instr, 1, instr.common.src1i, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2i, SRC2);
} else {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
}
if (Common::GetCPUCaps().sse4_1) {
// Set 4th component to 1.0
BLENDPS(SRC1, R(ONE), 0x8); // 0b1000
} else {
// Set 4th component to 1.0
MOVAPS(SCRATCH, R(SRC1));
UNPCKHPS(SCRATCH, R(ONE)); // XYZW, 1111 -> Z1__
UNPCKLPD(SRC1, R(SCRATCH)); // XYZW, Z1__ -> XYZ1
}
Compile_SanitizedMul(SRC1, SRC2, SCRATCH);
MOVAPS(SRC2, R(SRC1));
SHUFPS(SRC1, R(SRC1), _MM_SHUFFLE(2, 3, 0, 1)); // XYZW -> ZWXY
ADDPS(SRC1, R(SRC2));
MOVAPS(SRC2, R(SRC1));
SHUFPS(SRC1, R(SRC1), _MM_SHUFFLE(0, 1, 2, 3)); // XYZW -> WZYX
ADDPS(SRC1, R(SRC2));
Compile_DestEnable(instr, SRC1);
}
void JitCompiler::Compile_EX2(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
MOVSS(XMM0, R(SRC1));
ABI_PushRegistersAndAdjustStack(PersistentCallerSavedRegs(), 0);
ABI_CallFunction(reinterpret_cast<const void*>(exp2f));
ABI_PopRegistersAndAdjustStack(PersistentCallerSavedRegs(), 0);
SHUFPS(XMM0, R(XMM0), _MM_SHUFFLE(0, 0, 0, 0));
MOVAPS(SRC1, R(XMM0));
Compile_DestEnable(instr, SRC1);
}
void JitCompiler::Compile_LG2(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
MOVSS(XMM0, R(SRC1));
ABI_PushRegistersAndAdjustStack(PersistentCallerSavedRegs(), 0);
ABI_CallFunction(reinterpret_cast<const void*>(log2f));
ABI_PopRegistersAndAdjustStack(PersistentCallerSavedRegs(), 0);
SHUFPS(XMM0, R(XMM0), _MM_SHUFFLE(0, 0, 0, 0));
MOVAPS(SRC1, R(XMM0));
Compile_DestEnable(instr, SRC1);
}
void JitCompiler::Compile_MUL(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
Compile_SanitizedMul(SRC1, SRC2, SCRATCH);
Compile_DestEnable(instr, SRC1);
}
void JitCompiler::Compile_SGE(Instruction instr) {
if (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::SGEI) {
Compile_SwizzleSrc(instr, 1, instr.common.src1i, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2i, SRC2);
} else {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
}
CMPPS(SRC2, R(SRC1), CMP_LE);
ANDPS(SRC2, R(ONE));
Compile_DestEnable(instr, SRC2);
}
void JitCompiler::Compile_SLT(Instruction instr) {
if (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::SLTI) {
Compile_SwizzleSrc(instr, 1, instr.common.src1i, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2i, SRC2);
} else {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
}
CMPPS(SRC1, R(SRC2), CMP_LT);
ANDPS(SRC1, R(ONE));
Compile_DestEnable(instr, SRC1);
}
void JitCompiler::Compile_FLR(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
if (Common::GetCPUCaps().sse4_1) {
ROUNDFLOORPS(SRC1, R(SRC1));
} else {
CVTPS2DQ(SRC1, R(SRC1));
CVTDQ2PS(SRC1, R(SRC1));
}
Compile_DestEnable(instr, SRC1);
}
void JitCompiler::Compile_MAX(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
// SSE semantics match PICA200 ones: In case of NaN, SRC2 is returned.
MAXPS(SRC1, R(SRC2));
Compile_DestEnable(instr, SRC1);
}
void JitCompiler::Compile_MIN(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
// SSE semantics match PICA200 ones: In case of NaN, SRC2 is returned.
MINPS(SRC1, R(SRC2));
Compile_DestEnable(instr, SRC1);
}
void JitCompiler::Compile_MOVA(Instruction instr) {
SwizzlePattern swiz = { g_state.vs.swizzle_data[instr.common.operand_desc_id] };
if (!swiz.DestComponentEnabled(0) && !swiz.DestComponentEnabled(1)) {
return; // NoOp
}
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
// Convert floats to integers using truncation (only care about X and Y components)
CVTTPS2DQ(SRC1, R(SRC1));
// Get result
MOVQ_xmm(R(RAX), SRC1);
// Handle destination enable
if (swiz.DestComponentEnabled(0) && swiz.DestComponentEnabled(1)) {
// Move and sign-extend low 32 bits
MOVSX(64, 32, ADDROFFS_REG_0, R(RAX));
// Move and sign-extend high 32 bits
SHR(64, R(RAX), Imm8(32));
MOVSX(64, 32, ADDROFFS_REG_1, R(RAX));
// Multiply by 16 to be used as an offset later
SHL(64, R(ADDROFFS_REG_0), Imm8(4));
SHL(64, R(ADDROFFS_REG_1), Imm8(4));
} else {
if (swiz.DestComponentEnabled(0)) {
// Move and sign-extend low 32 bits
MOVSX(64, 32, ADDROFFS_REG_0, R(RAX));
// Multiply by 16 to be used as an offset later
SHL(64, R(ADDROFFS_REG_0), Imm8(4));
} else if (swiz.DestComponentEnabled(1)) {
// Move and sign-extend high 32 bits
SHR(64, R(RAX), Imm8(32));
MOVSX(64, 32, ADDROFFS_REG_1, R(RAX));
// Multiply by 16 to be used as an offset later
SHL(64, R(ADDROFFS_REG_1), Imm8(4));
}
}
}
void JitCompiler::Compile_MOV(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_DestEnable(instr, SRC1);
}
void JitCompiler::Compile_RCP(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
// TODO(bunnei): RCPSS is a pretty rough approximation, this might cause problems if Pica
// performs this operation more accurately. This should be checked on hardware.
RCPSS(SRC1, R(SRC1));
SHUFPS(SRC1, R(SRC1), _MM_SHUFFLE(0, 0, 0, 0)); // XYWZ -> XXXX
Compile_DestEnable(instr, SRC1);
}
void JitCompiler::Compile_RSQ(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
// TODO(bunnei): RSQRTSS is a pretty rough approximation, this might cause problems if Pica
// performs this operation more accurately. This should be checked on hardware.
RSQRTSS(SRC1, R(SRC1));
SHUFPS(SRC1, R(SRC1), _MM_SHUFFLE(0, 0, 0, 0)); // XYWZ -> XXXX
Compile_DestEnable(instr, SRC1);
}
void JitCompiler::Compile_NOP(Instruction instr) {
}
void JitCompiler::Compile_END(Instruction instr) {
ABI_PopRegistersAndAdjustStack(ABI_ALL_CALLEE_SAVED, 8);
RET();
}
void JitCompiler::Compile_CALL(Instruction instr) {
// Push offset of the return
PUSH(64, Imm32(instr.flow_control.dest_offset + instr.flow_control.num_instructions));
// Call the subroutine
FixupBranch b = CALL();
fixup_branches.push_back({ b, instr.flow_control.dest_offset });
// Skip over the return offset that's on the stack
ADD(64, R(RSP), Imm32(8));
}
void JitCompiler::Compile_CALLC(Instruction instr) {
Compile_EvaluateCondition(instr);
FixupBranch b = J_CC(CC_Z, true);
Compile_CALL(instr);
SetJumpTarget(b);
}
void JitCompiler::Compile_CALLU(Instruction instr) {
Compile_UniformCondition(instr);
FixupBranch b = J_CC(CC_Z, true);
Compile_CALL(instr);
SetJumpTarget(b);
}
void JitCompiler::Compile_CMP(Instruction instr) {
using Op = Instruction::Common::CompareOpType::Op;
Op op_x = instr.common.compare_op.x;
Op op_y = instr.common.compare_op.y;
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
// SSE doesn't have greater-than (GT) or greater-equal (GE) comparison operators. You need to
// emulate them by swapping the lhs and rhs and using LT and LE. NLT and NLE can't be used here
// because they don't match when used with NaNs.
static const u8 cmp[] = { CMP_EQ, CMP_NEQ, CMP_LT, CMP_LE, CMP_LT, CMP_LE };
bool invert_op_x = (op_x == Op::GreaterThan || op_x == Op::GreaterEqual);
Gen::X64Reg lhs_x = invert_op_x ? SRC2 : SRC1;
Gen::X64Reg rhs_x = invert_op_x ? SRC1 : SRC2;
if (op_x == op_y) {
// Compare X-component and Y-component together
CMPPS(lhs_x, R(rhs_x), cmp[op_x]);
MOVQ_xmm(R(COND0), lhs_x);
MOV(64, R(COND1), R(COND0));
} else {
bool invert_op_y = (op_y == Op::GreaterThan || op_y == Op::GreaterEqual);
Gen::X64Reg lhs_y = invert_op_y ? SRC2 : SRC1;
Gen::X64Reg rhs_y = invert_op_y ? SRC1 : SRC2;
// Compare X-component
MOVAPS(SCRATCH, R(lhs_x));
CMPSS(SCRATCH, R(rhs_x), cmp[op_x]);
// Compare Y-component
CMPPS(lhs_y, R(rhs_y), cmp[op_y]);
MOVQ_xmm(R(COND0), SCRATCH);
MOVQ_xmm(R(COND1), lhs_y);
}
SHR(32, R(COND0), Imm8(31));
SHR(64, R(COND1), Imm8(63));
}
void JitCompiler::Compile_MAD(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.mad.src1, SRC1);
if (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::MADI) {
Compile_SwizzleSrc(instr, 2, instr.mad.src2i, SRC2);
Compile_SwizzleSrc(instr, 3, instr.mad.src3i, SRC3);
} else {
Compile_SwizzleSrc(instr, 2, instr.mad.src2, SRC2);
Compile_SwizzleSrc(instr, 3, instr.mad.src3, SRC3);
}
Compile_SanitizedMul(SRC1, SRC2, SCRATCH);
ADDPS(SRC1, R(SRC3));
Compile_DestEnable(instr, SRC1);
}
void JitCompiler::Compile_IF(Instruction instr) {
RuntimeAssert(instr.flow_control.dest_offset >= program_counter, "Backwards if-statements not supported");
// Evaluate the "IF" condition
if (instr.opcode.Value() == OpCode::Id::IFU) {
Compile_UniformCondition(instr);
} else if (instr.opcode.Value() == OpCode::Id::IFC) {
Compile_EvaluateCondition(instr);
}
FixupBranch b = J_CC(CC_Z, true);
// Compile the code that corresponds to the condition evaluating as true
Compile_Block(instr.flow_control.dest_offset);
// If there isn't an "ELSE" condition, we are done here
if (instr.flow_control.num_instructions == 0) {
SetJumpTarget(b);
return;
}
FixupBranch b2 = J(true);
SetJumpTarget(b);
// This code corresponds to the "ELSE" condition
// Comple the code that corresponds to the condition evaluating as false
Compile_Block(instr.flow_control.dest_offset + instr.flow_control.num_instructions);
SetJumpTarget(b2);
}
void JitCompiler::Compile_LOOP(Instruction instr) {
RuntimeAssert(instr.flow_control.dest_offset >= program_counter, "Backwards loops not supported");
RuntimeAssert(!looping, "Nested loops not supported");
looping = true;
int offset = offsetof(decltype(g_state.vs.uniforms), i) + (instr.flow_control.int_uniform_id * sizeof(Math::Vec4<u8>));
MOV(32, R(LOOPCOUNT), MDisp(UNIFORMS, offset));
MOV(32, R(LOOPCOUNT_REG), R(LOOPCOUNT));
SHR(32, R(LOOPCOUNT_REG), Imm8(8));
AND(32, R(LOOPCOUNT_REG), Imm32(0xff)); // Y-component is the start
MOV(32, R(LOOPINC), R(LOOPCOUNT));
SHR(32, R(LOOPINC), Imm8(16));
MOVZX(32, 8, LOOPINC, R(LOOPINC)); // Z-component is the incrementer
MOVZX(32, 8, LOOPCOUNT, R(LOOPCOUNT)); // X-component is iteration count
ADD(32, R(LOOPCOUNT), Imm8(1)); // Iteration count is X-component + 1
auto loop_start = GetCodePtr();
Compile_Block(instr.flow_control.dest_offset + 1);
ADD(32, R(LOOPCOUNT_REG), R(LOOPINC)); // Increment LOOPCOUNT_REG by Z-component
SUB(32, R(LOOPCOUNT), Imm8(1)); // Increment loop count by 1
J_CC(CC_NZ, loop_start); // Loop if not equal
looping = false;
}
void JitCompiler::Compile_JMP(Instruction instr) {
if (instr.opcode.Value() == OpCode::Id::JMPC)
Compile_EvaluateCondition(instr);
else if (instr.opcode.Value() == OpCode::Id::JMPU)
Compile_UniformCondition(instr);
else
UNREACHABLE();
bool inverted_condition = (instr.opcode.Value() == OpCode::Id::JMPU) &&
(instr.flow_control.num_instructions & 1);
FixupBranch b = J_CC(inverted_condition ? CC_Z : CC_NZ, true);
fixup_branches.push_back({ b, instr.flow_control.dest_offset });
}
void JitCompiler::Compile_Block(unsigned end) {
while (program_counter < end) {
Compile_NextInstr();
}
}
void JitCompiler::Compile_Return() {
// Peek return offset on the stack and check if we're at that offset
MOV(64, R(RAX), MDisp(RSP, 8));
CMP(32, R(RAX), Imm32(program_counter));
// If so, jump back to before CALL
FixupBranch b = J_CC(CC_NZ, true);
RET();
SetJumpTarget(b);
}
void JitCompiler::Compile_NextInstr() {
auto search = return_offsets.find(program_counter);
if (search != return_offsets.end()) {
Compile_Return();
}
ASSERT_MSG(code_ptr[program_counter] == nullptr, "Tried to compile already compiled shader location!");
code_ptr[program_counter] = GetCodePtr();
Instruction instr = GetVertexShaderInstruction(program_counter++);
OpCode::Id opcode = instr.opcode.Value();
auto instr_func = instr_table[static_cast<unsigned>(opcode)];
if (instr_func) {
// JIT the instruction!
((*this).*instr_func)(instr);
} else {
// Unhandled instruction
LOG_CRITICAL(HW_GPU, "Unhandled instruction: 0x%02x (0x%08x)",
instr.opcode.Value().EffectiveOpCode(), instr.hex);
}
}
void JitCompiler::FindReturnOffsets() {
return_offsets.clear();
for (size_t offset = 0; offset < g_state.vs.program_code.size(); ++offset) {
Instruction instr = GetVertexShaderInstruction(offset);
switch (instr.opcode.Value()) {
case OpCode::Id::CALL:
case OpCode::Id::CALLC:
case OpCode::Id::CALLU:
return_offsets.insert(instr.flow_control.dest_offset + instr.flow_control.num_instructions);
break;
}
}
}
void JitCompiler::Compile() {
// Reset flow control state
program = (CompiledShader*)GetCodePtr();
program_counter = 0;
looping = false;
code_ptr.fill(nullptr);
fixup_branches.clear();
// Find all `CALL` instructions and identify return locations
FindReturnOffsets();
// The stack pointer is 8 modulo 16 at the entry of a procedure
ABI_PushRegistersAndAdjustStack(ABI_ALL_CALLEE_SAVED, 8);
MOV(PTRBITS, R(REGISTERS), R(ABI_PARAM1));
MOV(PTRBITS, R(UNIFORMS), ImmPtr(&g_state.vs.uniforms));
// Zero address/loop registers
XOR(64, R(ADDROFFS_REG_0), R(ADDROFFS_REG_0));
XOR(64, R(ADDROFFS_REG_1), R(ADDROFFS_REG_1));
XOR(64, R(LOOPCOUNT_REG), R(LOOPCOUNT_REG));
// Used to set a register to one
static const __m128 one = { 1.f, 1.f, 1.f, 1.f };
MOV(PTRBITS, R(RAX), ImmPtr(&one));
MOVAPS(ONE, MatR(RAX));
// Used to negate registers
static const __m128 neg = { -0.f, -0.f, -0.f, -0.f };
MOV(PTRBITS, R(RAX), ImmPtr(&neg));
MOVAPS(NEGBIT, MatR(RAX));
// Jump to start of the shader program
JMPptr(R(ABI_PARAM2));
// Compile entire program
Compile_Block(static_cast<unsigned>(g_state.vs.program_code.size()));
// Set the target for any incomplete branches now that the entire shader program has been emitted
for (const auto& branch : fixup_branches) {
SetJumpTarget(branch.first, code_ptr[branch.second]);
}
uintptr_t size = reinterpret_cast<uintptr_t>(GetCodePtr()) - reinterpret_cast<uintptr_t>(program);
ASSERT_MSG(size <= MAX_SHADER_SIZE, "Compiled a shader that exceeds the allocated size!");
LOG_DEBUG(HW_GPU, "Compiled shader size=%d", size);
}
JitCompiler::JitCompiler() {
AllocCodeSpace(MAX_SHADER_SIZE);
}
} // namespace Shader
} // namespace Pica
|