summaryrefslogtreecommitdiff
path: root/src/gallium/drivers/swr/rasterizer/jitter/builder_mem.cpp
blob: a9d649092d41337bd3bcd7ac93195f7ef26a1fc2 (plain)
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
/****************************************************************************
 * Copyright (C) 2014-2015 Intel Corporation.   All Rights Reserved.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 *
 * @file builder_misc.cpp
 *
 * @brief Implementation for miscellaneous builder functions
 *
 * Notes:
 *
 ******************************************************************************/
#include "jit_pch.hpp"
#include "builder.h"

#include <cstdarg>

namespace SwrJit
{
    void Builder::AssertMemoryUsageParams(Value* ptr, JIT_MEM_CLIENT usage)
    {
        SWR_ASSERT(
            ptr->getType() != mInt64Ty,
            "Address appears to be GFX access.  Requires translation through BuilderGfxMem.");
    }

    Value* Builder::GEP(Value* Ptr, Value* Idx, Type* Ty, const Twine& Name)
    {
        return IRB()->CreateGEP(Ptr, Idx, Name);
    }

    Value* Builder::GEP(Type* Ty, Value* Ptr, Value* Idx, const Twine& Name)
    {
        return IRB()->CreateGEP(Ty, Ptr, Idx, Name);
    }

    Value* Builder::GEP(Value* ptr, const std::initializer_list<Value*>& indexList, Type* Ty)
    {
        std::vector<Value*> indices;
        for (auto i : indexList)
            indices.push_back(i);
        return GEPA(ptr, indices);
    }

    Value* Builder::GEP(Value* ptr, const std::initializer_list<uint32_t>& indexList, Type* Ty)
    {
        std::vector<Value*> indices;
        for (auto i : indexList)
            indices.push_back(C(i));
        return GEPA(ptr, indices);
    }

    Value* Builder::GEPA(Value* Ptr, ArrayRef<Value*> IdxList, const Twine& Name)
    {
        return IRB()->CreateGEP(Ptr, IdxList, Name);
    }

    Value* Builder::GEPA(Type* Ty, Value* Ptr, ArrayRef<Value*> IdxList, const Twine& Name)
    {
        return IRB()->CreateGEP(Ty, Ptr, IdxList, Name);
    }

    Value* Builder::IN_BOUNDS_GEP(Value* ptr, const std::initializer_list<Value*>& indexList)
    {
        std::vector<Value*> indices;
        for (auto i : indexList)
            indices.push_back(i);
        return IN_BOUNDS_GEP(ptr, indices);
    }

    Value* Builder::IN_BOUNDS_GEP(Value* ptr, const std::initializer_list<uint32_t>& indexList)
    {
        std::vector<Value*> indices;
        for (auto i : indexList)
            indices.push_back(C(i));
        return IN_BOUNDS_GEP(ptr, indices);
    }

    LoadInst* Builder::LOAD(Value* Ptr, const char* Name, Type* Ty, JIT_MEM_CLIENT usage)
    {
        AssertMemoryUsageParams(Ptr, usage);
        return IRB()->CreateLoad(Ptr, Name);
    }

    LoadInst* Builder::LOAD(Value* Ptr, const Twine& Name, Type* Ty, JIT_MEM_CLIENT usage)
    {
        AssertMemoryUsageParams(Ptr, usage);
        return IRB()->CreateLoad(Ptr, Name);
    }

    LoadInst* Builder::LOAD(Type* Ty, Value* Ptr, const Twine& Name, JIT_MEM_CLIENT usage)
    {
        AssertMemoryUsageParams(Ptr, usage);
        return IRB()->CreateLoad(Ty, Ptr, Name);
    }

    LoadInst*
    Builder::LOAD(Value* Ptr, bool isVolatile, const Twine& Name, Type* Ty, JIT_MEM_CLIENT usage)
    {
        AssertMemoryUsageParams(Ptr, usage);
        return IRB()->CreateLoad(Ptr, isVolatile, Name);
    }

    LoadInst* Builder::LOAD(Value*                                 basePtr,
                            const std::initializer_list<uint32_t>& indices,
                            const llvm::Twine&                     name,
                            Type*                                  Ty,
                            JIT_MEM_CLIENT                         usage)
    {
        std::vector<Value*> valIndices;
        for (auto i : indices)
            valIndices.push_back(C(i));
        return Builder::LOAD(GEPA(basePtr, valIndices), name);
    }

    LoadInst* Builder::LOADV(Value*                               basePtr,
                             const std::initializer_list<Value*>& indices,
                             const llvm::Twine&                   name)
    {
        std::vector<Value*> valIndices;
        for (auto i : indices)
            valIndices.push_back(i);
        return LOAD(GEPA(basePtr, valIndices), name);
    }

    StoreInst*
    Builder::STORE(Value* val, Value* basePtr, const std::initializer_list<uint32_t>& indices)
    {
        std::vector<Value*> valIndices;
        for (auto i : indices)
            valIndices.push_back(C(i));
        return STORE(val, GEPA(basePtr, valIndices));
    }

    StoreInst*
    Builder::STOREV(Value* val, Value* basePtr, const std::initializer_list<Value*>& indices)
    {
        std::vector<Value*> valIndices;
        for (auto i : indices)
            valIndices.push_back(i);
        return STORE(val, GEPA(basePtr, valIndices));
    }

    Value* Builder::OFFSET_TO_NEXT_COMPONENT(Value* base, Constant* offset)
    {
        return GEP(base, offset);
    }

    Value* Builder::MEM_ADD(Value*                                 i32Incr,
                            Value*                                 basePtr,
                            const std::initializer_list<uint32_t>& indices,
                            const llvm::Twine&                     name)
    {
        Value* i32Value  = LOAD(GEP(basePtr, indices), name);
        Value* i32Result = ADD(i32Value, i32Incr);
        return STORE(i32Result, GEP(basePtr, indices));
    }

    //////////////////////////////////////////////////////////////////////////
    /// @brief Generate a masked gather operation in LLVM IR.  If not
    /// supported on the underlying platform, emulate it with loads
    /// @param vSrc - SIMD wide value that will be loaded if mask is invalid
    /// @param pBase - Int8* base VB address pointer value
    /// @param vIndices - SIMD wide value of VB byte offsets
    /// @param vMask - SIMD wide mask that controls whether to access memory or the src values
    /// @param scale - value to scale indices by
    Value* Builder::GATHERPS(Value*         vSrc,
                             Value*         pBase,
                             Value*         vIndices,
                             Value*         vMask,
                             uint8_t        scale,
                             JIT_MEM_CLIENT usage)
    {
        AssertMemoryUsageParams(pBase, usage);

        return VGATHERPS(vSrc, pBase, vIndices, vMask, C(scale));
    }

    //////////////////////////////////////////////////////////////////////////
    /// @brief Generate a masked gather operation in LLVM IR.  If not
    /// supported on the underlying platform, emulate it with loads
    /// @param vSrc - SIMD wide value that will be loaded if mask is invalid
    /// @param pBase - Int8* base VB address pointer value
    /// @param vIndices - SIMD wide value of VB byte offsets
    /// @param vMask - SIMD wide mask that controls whether to access memory or the src values
    /// @param scale - value to scale indices by
    Value* Builder::GATHERDD(Value*         vSrc,
                             Value*         pBase,
                             Value*         vIndices,
                             Value*         vMask,
                             uint8_t        scale,
                             JIT_MEM_CLIENT usage)
    {
        AssertMemoryUsageParams(pBase, usage);

        return VGATHERDD(vSrc, pBase, vIndices, vMask, C(scale));
    }

    //////////////////////////////////////////////////////////////////////////
    /// @brief Generate a masked gather operation in LLVM IR.  If not
    /// supported on the underlying platform, emulate it with loads
    /// @param vSrc - SIMD wide value that will be loaded if mask is invalid
    /// @param pBase - Int8* base VB address pointer value
    /// @param vIndices - SIMD wide value of VB byte offsets
    /// @param vMask - SIMD wide mask that controls whether to access memory or the src values
    /// @param scale - value to scale indices by
    Value*
    Builder::GATHERPD(Value* vSrc, Value* pBase, Value* vIndices, Value* vMask, uint8_t scale)
    {
        return VGATHERPD(vSrc, pBase, vIndices, vMask, C(scale));
    }

    //////////////////////////////////////////////////////////////////////////
    /// @brief Alternative masked gather where source is a vector of pointers
    /// @param pVecSrcPtr   - SIMD wide vector of pointers
    /// @param pVecMask     - SIMD active lanes
    /// @param pVecPassthru - SIMD wide vector of values to load when lane is inactive
    Value* Builder::GATHER_PTR(Value* pVecSrcPtr, Value* pVecMask, Value* pVecPassthru)
    {
        return MASKED_GATHER(pVecSrcPtr, 4, pVecMask, pVecPassthru);
    }

    void Builder::Gather4(const SWR_FORMAT format,
                          Value*           pSrcBase,
                          Value*           byteOffsets,
                          Value*           mask,
                          Value*           vGatherComponents[],
                          bool             bPackedOutput,
                          JIT_MEM_CLIENT   usage)
    {
        const SWR_FORMAT_INFO& info = GetFormatInfo(format);
        if (info.type[0] == SWR_TYPE_FLOAT && info.bpc[0] == 32)
        {
            GATHER4PS(info, pSrcBase, byteOffsets, mask, vGatherComponents, bPackedOutput, usage);
        }
        else
        {
            GATHER4DD(info, pSrcBase, byteOffsets, mask, vGatherComponents, bPackedOutput, usage);
        }
    }

    void Builder::GATHER4PS(const SWR_FORMAT_INFO& info,
                            Value*                 pSrcBase,
                            Value*                 byteOffsets,
                            Value*                 vMask,
                            Value*                 vGatherComponents[],
                            bool                   bPackedOutput,
                            JIT_MEM_CLIENT         usage)
    {
        switch (info.bpp / info.numComps)
        {
        case 16:
        {
            Value* vGatherResult[2];

            // TODO: vGatherMaskedVal
            Value* vGatherMaskedVal = VIMMED1((float)0);

            // always have at least one component out of x or y to fetch

            vGatherResult[0] = GATHERPS(vGatherMaskedVal, pSrcBase, byteOffsets, vMask, 1, usage);
            // e.g. result of first 8x32bit integer gather for 16bit components
            // 256i - 0    1    2    3    4    5    6    7
            //        xyxy xyxy xyxy xyxy xyxy xyxy xyxy xyxy
            //

            // if we have at least one component out of x or y to fetch
            if (info.numComps > 2)
            {
                // offset base to the next components(zw) in the vertex to gather
                pSrcBase = OFFSET_TO_NEXT_COMPONENT(pSrcBase, C((intptr_t)4));

                vGatherResult[1] =
                    GATHERPS(vGatherMaskedVal, pSrcBase, byteOffsets, vMask, 1, usage);
                // e.g. result of second 8x32bit integer gather for 16bit components
                // 256i - 0    1    2    3    4    5    6    7
                //        zwzw zwzw zwzw zwzw zwzw zwzw zwzw zwzw
                //
            }
            else
            {
                vGatherResult[1] = vGatherMaskedVal;
            }

            // Shuffle gathered components into place, each row is a component
            Shuffle16bpcGather4(info, vGatherResult, vGatherComponents, bPackedOutput);
        }
        break;
        case 32:
        {
            // apply defaults
            for (uint32_t i = 0; i < 4; ++i)
            {
                vGatherComponents[i] = VIMMED1(*(float*)&info.defaults[i]);
            }

            for (uint32_t i = 0; i < info.numComps; i++)
            {
                uint32_t swizzleIndex = info.swizzle[i];

                // Gather a SIMD of components
                vGatherComponents[swizzleIndex] = GATHERPS(
                    vGatherComponents[swizzleIndex], pSrcBase, byteOffsets, vMask, 1, usage);

                // offset base to the next component to gather
                pSrcBase = OFFSET_TO_NEXT_COMPONENT(pSrcBase, C((intptr_t)4));
            }
        }
        break;
        default:
            SWR_INVALID("Invalid float format");
            break;
        }
    }

    void Builder::GATHER4DD(const SWR_FORMAT_INFO& info,
                            Value*                 pSrcBase,
                            Value*                 byteOffsets,
                            Value*                 vMask,
                            Value*                 vGatherComponents[],
                            bool                   bPackedOutput,
                            JIT_MEM_CLIENT         usage)
    {
        switch (info.bpp / info.numComps)
        {
        case 8:
        {
            Value* vGatherMaskedVal = VIMMED1((int32_t)0);
            Value* vGatherResult =
                GATHERDD(vGatherMaskedVal, pSrcBase, byteOffsets, vMask, 1, usage);
            // e.g. result of an 8x32bit integer gather for 8bit components
            // 256i - 0    1    2    3    4    5    6    7
            //        xyzw xyzw xyzw xyzw xyzw xyzw xyzw xyzw

            Shuffle8bpcGather4(info, vGatherResult, vGatherComponents, bPackedOutput);
        }
        break;
        case 16:
        {
            Value* vGatherResult[2];

            // TODO: vGatherMaskedVal
            Value* vGatherMaskedVal = VIMMED1((int32_t)0);

            // always have at least one component out of x or y to fetch

            vGatherResult[0] = GATHERDD(vGatherMaskedVal, pSrcBase, byteOffsets, vMask, 1, usage);
            // e.g. result of first 8x32bit integer gather for 16bit components
            // 256i - 0    1    2    3    4    5    6    7
            //        xyxy xyxy xyxy xyxy xyxy xyxy xyxy xyxy
            //

            // if we have at least one component out of x or y to fetch
            if (info.numComps > 2)
            {
                // offset base to the next components(zw) in the vertex to gather
                pSrcBase = OFFSET_TO_NEXT_COMPONENT(pSrcBase, C((intptr_t)4));

                vGatherResult[1] =
                    GATHERDD(vGatherMaskedVal, pSrcBase, byteOffsets, vMask, 1, usage);
                // e.g. result of second 8x32bit integer gather for 16bit components
                // 256i - 0    1    2    3    4    5    6    7
                //        zwzw zwzw zwzw zwzw zwzw zwzw zwzw zwzw
                //
            }
            else
            {
                vGatherResult[1] = vGatherMaskedVal;
            }

            // Shuffle gathered components into place, each row is a component
            Shuffle16bpcGather4(info, vGatherResult, vGatherComponents, bPackedOutput);
        }
        break;
        case 32:
        {
            // apply defaults
            for (uint32_t i = 0; i < 4; ++i)
            {
                vGatherComponents[i] = VIMMED1((int)info.defaults[i]);
            }

            for (uint32_t i = 0; i < info.numComps; i++)
            {
                uint32_t swizzleIndex = info.swizzle[i];

                // Gather a SIMD of components
                vGatherComponents[swizzleIndex] = GATHERDD(
                    vGatherComponents[swizzleIndex], pSrcBase, byteOffsets, vMask, 1, usage);

                // offset base to the next component to gather
                pSrcBase = OFFSET_TO_NEXT_COMPONENT(pSrcBase, C((intptr_t)4));
            }
        }
        break;
        default:
            SWR_INVALID("unsupported format");
            break;
        }
    }

    void Builder::Shuffle16bpcGather4(const SWR_FORMAT_INFO& info,
                                      Value*                 vGatherInput[2],
                                      Value*                 vGatherOutput[4],
                                      bool                   bPackedOutput)
    {
        // cast types
        Type* vGatherTy = VectorType::get(IntegerType::getInt32Ty(JM()->mContext), mVWidth);
        Type* v32x8Ty   = VectorType::get(mInt8Ty, mVWidth * 4); // vwidth is units of 32 bits

        // input could either be float or int vector; do shuffle work in int
        vGatherInput[0] = BITCAST(vGatherInput[0], mSimdInt32Ty);
        vGatherInput[1] = BITCAST(vGatherInput[1], mSimdInt32Ty);

        if (bPackedOutput)
        {
            Type* v128bitTy = VectorType::get(IntegerType::getIntNTy(JM()->mContext, 128),
                                              mVWidth / 4); // vwidth is units of 32 bits

            // shuffle mask
            Value* vConstMask = C<char>({0, 1, 4, 5, 8, 9, 12, 13, 2, 3, 6, 7, 10, 11, 14, 15,
                                         0, 1, 4, 5, 8, 9, 12, 13, 2, 3, 6, 7, 10, 11, 14, 15});
            Value* vShufResult =
                BITCAST(PSHUFB(BITCAST(vGatherInput[0], v32x8Ty), vConstMask), vGatherTy);
            // after pshufb: group components together in each 128bit lane
            // 256i - 0    1    2    3    4    5    6    7
            //        xxxx xxxx yyyy yyyy xxxx xxxx yyyy yyyy

            Value* vi128XY =
                BITCAST(VPERMD(vShufResult, C<int32_t>({0, 1, 4, 5, 2, 3, 6, 7})), v128bitTy);
            // after PERMD: move and pack xy components into each 128bit lane
            // 256i - 0    1    2    3    4    5    6    7
            //        xxxx xxxx xxxx xxxx yyyy yyyy yyyy yyyy

            // do the same for zw components
            Value* vi128ZW = nullptr;
            if (info.numComps > 2)
            {
                Value* vShufResult =
                    BITCAST(PSHUFB(BITCAST(vGatherInput[1], v32x8Ty), vConstMask), vGatherTy);
                vi128ZW =
                    BITCAST(VPERMD(vShufResult, C<int32_t>({0, 1, 4, 5, 2, 3, 6, 7})), v128bitTy);
            }

            for (uint32_t i = 0; i < 4; i++)
            {
                uint32_t swizzleIndex = info.swizzle[i];
                // todo: fixed for packed
                Value* vGatherMaskedVal = VIMMED1((int32_t)(info.defaults[i]));
                if (i >= info.numComps)
                {
                    // set the default component val
                    vGatherOutput[swizzleIndex] = vGatherMaskedVal;
                    continue;
                }

                // if x or z, extract 128bits from lane 0, else for y or w, extract from lane 1
                uint32_t lane = ((i == 0) || (i == 2)) ? 0 : 1;
                // if x or y, use vi128XY permute result, else use vi128ZW
                Value* selectedPermute = (i < 2) ? vi128XY : vi128ZW;

                // extract packed component 128 bit lanes
                vGatherOutput[swizzleIndex] = VEXTRACT(selectedPermute, C(lane));
            }
        }
        else
        {
            // pshufb masks for each component
            Value* vConstMask[2];
            // x/z shuffle mask
            vConstMask[0] = C<char>({
                0, 1, -1, -1, 4, 5, -1, -1, 8, 9, -1, -1, 12, 13, -1, -1,
                0, 1, -1, -1, 4, 5, -1, -1, 8, 9, -1, -1, 12, 13, -1, -1,
            });

            // y/w shuffle mask
            vConstMask[1] = C<char>({2, 3, -1, -1, 6, 7, -1, -1, 10, 11, -1, -1, 14, 15, -1, -1,
                                     2, 3, -1, -1, 6, 7, -1, -1, 10, 11, -1, -1, 14, 15, -1, -1});

            // shuffle enabled components into lower word of each 32bit lane, 0 extending to 32 bits
            // apply defaults
            for (uint32_t i = 0; i < 4; ++i)
            {
                vGatherOutput[i] = VIMMED1((int32_t)info.defaults[i]);
            }

            for (uint32_t i = 0; i < info.numComps; i++)
            {
                uint32_t swizzleIndex = info.swizzle[i];

                // select correct constMask for x/z or y/w pshufb
                uint32_t selectedMask = ((i == 0) || (i == 2)) ? 0 : 1;
                // if x or y, use vi128XY permute result, else use vi128ZW
                uint32_t selectedGather = (i < 2) ? 0 : 1;

                vGatherOutput[swizzleIndex] =
                    BITCAST(PSHUFB(BITCAST(vGatherInput[selectedGather], v32x8Ty),
                                   vConstMask[selectedMask]),
                            vGatherTy);
                // after pshufb mask for x channel; z uses the same shuffle from the second gather
                // 256i - 0    1    2    3    4    5    6    7
                //        xx00 xx00 xx00 xx00 xx00 xx00 xx00 xx00
            }
        }
    }

    void Builder::Shuffle8bpcGather4(const SWR_FORMAT_INFO& info,
                                     Value*                 vGatherInput,
                                     Value*                 vGatherOutput[],
                                     bool                   bPackedOutput)
    {
        // cast types
        Type* vGatherTy = VectorType::get(IntegerType::getInt32Ty(JM()->mContext), mVWidth);
        Type* v32x8Ty   = VectorType::get(mInt8Ty, mVWidth * 4); // vwidth is units of 32 bits

        if (bPackedOutput)
        {
            Type* v128Ty = VectorType::get(IntegerType::getIntNTy(JM()->mContext, 128),
                                           mVWidth / 4); // vwidth is units of 32 bits
                                                         // shuffle mask
            Value* vConstMask = C<char>({0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15,
                                         0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15});
            Value* vShufResult =
                BITCAST(PSHUFB(BITCAST(vGatherInput, v32x8Ty), vConstMask), vGatherTy);
            // after pshufb: group components together in each 128bit lane
            // 256i - 0    1    2    3    4    5    6    7
            //        xxxx yyyy zzzz wwww xxxx yyyy zzzz wwww

            Value* vi128XY =
                BITCAST(VPERMD(vShufResult, C<int32_t>({0, 4, 0, 0, 1, 5, 0, 0})), v128Ty);
            // after PERMD: move and pack xy and zw components in low 64 bits of each 128bit lane
            // 256i - 0    1    2    3    4    5    6    7
            //        xxxx xxxx dcdc dcdc yyyy yyyy dcdc dcdc (dc - don't care)

            // do the same for zw components
            Value* vi128ZW = nullptr;
            if (info.numComps > 2)
            {
                vi128ZW =
                    BITCAST(VPERMD(vShufResult, C<int32_t>({2, 6, 0, 0, 3, 7, 0, 0})), v128Ty);
            }

            // sign extend all enabled components. If we have a fill vVertexElements, output to
            // current simdvertex
            for (uint32_t i = 0; i < 4; i++)
            {
                uint32_t swizzleIndex = info.swizzle[i];
                // todo: fix for packed
                Value* vGatherMaskedVal = VIMMED1((int32_t)(info.defaults[i]));
                if (i >= info.numComps)
                {
                    // set the default component val
                    vGatherOutput[swizzleIndex] = vGatherMaskedVal;
                    continue;
                }

                // if x or z, extract 128bits from lane 0, else for y or w, extract from lane 1
                uint32_t lane = ((i == 0) || (i == 2)) ? 0 : 1;
                // if x or y, use vi128XY permute result, else use vi128ZW
                Value* selectedPermute = (i < 2) ? vi128XY : vi128ZW;

                // sign extend
                vGatherOutput[swizzleIndex] = VEXTRACT(selectedPermute, C(lane));
            }
        }
        // else zero extend
        else
        {
            // shuffle enabled components into lower byte of each 32bit lane, 0 extending to 32 bits
            // apply defaults
            for (uint32_t i = 0; i < 4; ++i)
            {
                vGatherOutput[i] = VIMMED1((int32_t)info.defaults[i]);
            }

            for (uint32_t i = 0; i < info.numComps; i++)
            {
                uint32_t swizzleIndex = info.swizzle[i];

                // pshufb masks for each component
                Value* vConstMask;
                switch (i)
                {
                case 0:
                    // x shuffle mask
                    vConstMask =
                        C<char>({0, -1, -1, -1, 4, -1, -1, -1, 8, -1, -1, -1, 12, -1, -1, -1,
                                 0, -1, -1, -1, 4, -1, -1, -1, 8, -1, -1, -1, 12, -1, -1, -1});
                    break;
                case 1:
                    // y shuffle mask
                    vConstMask =
                        C<char>({1, -1, -1, -1, 5, -1, -1, -1, 9, -1, -1, -1, 13, -1, -1, -1,
                                 1, -1, -1, -1, 5, -1, -1, -1, 9, -1, -1, -1, 13, -1, -1, -1});
                    break;
                case 2:
                    // z shuffle mask
                    vConstMask =
                        C<char>({2, -1, -1, -1, 6, -1, -1, -1, 10, -1, -1, -1, 14, -1, -1, -1,
                                 2, -1, -1, -1, 6, -1, -1, -1, 10, -1, -1, -1, 14, -1, -1, -1});
                    break;
                case 3:
                    // w shuffle mask
                    vConstMask =
                        C<char>({3, -1, -1, -1, 7, -1, -1, -1, 11, -1, -1, -1, 15, -1, -1, -1,
                                 3, -1, -1, -1, 7, -1, -1, -1, 11, -1, -1, -1, 15, -1, -1, -1});
                    break;
                default:
                    vConstMask = nullptr;
                    break;
                }

                vGatherOutput[swizzleIndex] =
                    BITCAST(PSHUFB(BITCAST(vGatherInput, v32x8Ty), vConstMask), vGatherTy);
                // after pshufb for x channel
                // 256i - 0    1    2    3    4    5    6    7
                //        x000 x000 x000 x000 x000 x000 x000 x000
            }
        }
    }

    //////////////////////////////////////////////////////////////////////////
    /// @brief emulates a scatter operation.
    /// @param pDst - pointer to destination
    /// @param vSrc - vector of src data to scatter
    /// @param vOffsets - vector of byte offsets from pDst
    /// @param vMask - mask of valid lanes
    void Builder::SCATTERPS(
        Value* pDst, Value* vSrc, Value* vOffsets, Value* vMask, JIT_MEM_CLIENT usage)
    {
        AssertMemoryUsageParams(pDst, usage);

        /* Scatter algorithm

        while(Index = BitScanForward(mask))
        srcElem = srcVector[Index]
        offsetElem = offsetVector[Index]
        *(pDst + offsetElem) = srcElem
        Update mask (&= ~(1<<Index)

        */

        BasicBlock* pCurBB = IRB()->GetInsertBlock();
        Function*   pFunc  = pCurBB->getParent();
        Type*       pSrcTy = vSrc->getType()->getVectorElementType();

        // Store vectors on stack
        if (pScatterStackSrc == nullptr)
        {
            // Save off stack allocations and reuse per scatter. Significantly reduces stack
            // requirements for shaders with a lot of scatters.
            pScatterStackSrc     = CreateEntryAlloca(pFunc, mSimdInt64Ty);
            pScatterStackOffsets = CreateEntryAlloca(pFunc, mSimdInt32Ty);
        }

        Value* pSrcArrayPtr     = BITCAST(pScatterStackSrc, PointerType::get(vSrc->getType(), 0));
        Value* pOffsetsArrayPtr = pScatterStackOffsets;
        STORE(vSrc, pSrcArrayPtr);
        STORE(vOffsets, pOffsetsArrayPtr);

        // Cast to pointers for random access
        pSrcArrayPtr     = POINTER_CAST(pSrcArrayPtr, PointerType::get(pSrcTy, 0));
        pOffsetsArrayPtr = POINTER_CAST(pOffsetsArrayPtr, PointerType::get(mInt32Ty, 0));

        Value* pMask = VMOVMSK(vMask);

        // Setup loop basic block
        BasicBlock* pLoop = BasicBlock::Create(mpJitMgr->mContext, "Scatter_Loop", pFunc);

        // compute first set bit
        Value* pIndex = CTTZ(pMask, C(false));

        Value* pIsUndef = ICMP_EQ(pIndex, C(32));

        // Split current block or create new one if building inline
        BasicBlock* pPostLoop;
        if (pCurBB->getTerminator())
        {
            pPostLoop = pCurBB->splitBasicBlock(cast<Instruction>(pIsUndef)->getNextNode());

            // Remove unconditional jump created by splitBasicBlock
            pCurBB->getTerminator()->eraseFromParent();

            // Add terminator to end of original block
            IRB()->SetInsertPoint(pCurBB);

            // Add conditional branch
            COND_BR(pIsUndef, pPostLoop, pLoop);
        }
        else
        {
            pPostLoop = BasicBlock::Create(mpJitMgr->mContext, "PostScatter_Loop", pFunc);

            // Add conditional branch
            COND_BR(pIsUndef, pPostLoop, pLoop);
        }

        // Add loop basic block contents
        IRB()->SetInsertPoint(pLoop);
        PHINode* pIndexPhi = PHI(mInt32Ty, 2);
        PHINode* pMaskPhi  = PHI(mInt32Ty, 2);

        pIndexPhi->addIncoming(pIndex, pCurBB);
        pMaskPhi->addIncoming(pMask, pCurBB);

        // Extract elements for this index
        Value* pSrcElem    = LOADV(pSrcArrayPtr, {pIndexPhi});
        Value* pOffsetElem = LOADV(pOffsetsArrayPtr, {pIndexPhi});

        // GEP to this offset in dst
        Value* pCurDst = GEP(pDst, pOffsetElem, mInt8PtrTy);
        pCurDst        = POINTER_CAST(pCurDst, PointerType::get(pSrcTy, 0));
        STORE(pSrcElem, pCurDst);

        // Update the mask
        Value* pNewMask = AND(pMaskPhi, NOT(SHL(C(1), pIndexPhi)));

        // Terminator
        Value* pNewIndex = CTTZ(pNewMask, C(false));

        pIsUndef = ICMP_EQ(pNewIndex, C(32));
        COND_BR(pIsUndef, pPostLoop, pLoop);

        // Update phi edges
        pIndexPhi->addIncoming(pNewIndex, pLoop);
        pMaskPhi->addIncoming(pNewMask, pLoop);

        // Move builder to beginning of post loop
        IRB()->SetInsertPoint(pPostLoop, pPostLoop->begin());
    }
} // namespace SwrJit