-
Notifications
You must be signed in to change notification settings - Fork 3
/
PipelineThread.cpp
1565 lines (1256 loc) · 78.8 KB
/
PipelineThread.cpp
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
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
#include "PipelineThread.h"
#include "RenderEngine.h"
#include "RenderState.h"
namespace tyler
{
PipelineThread::PipelineThread(RenderEngine* pRenderEngine, uint32_t threadIdx)
:
m_pRenderEngine(pRenderEngine),
m_RenderConfig(m_pRenderEngine->m_RenderConfig),
m_ThreadIdx(threadIdx),
m_CurrentState(ThreadStatus::IDLE)
{
m_WorkerThread = std::thread(&PipelineThread::Run, this);
}
PipelineThread::~PipelineThread()
{
m_CurrentState.store(ThreadStatus::TERMINATED);
ASSERT(m_WorkerThread.joinable());
m_WorkerThread.join();
}
void PipelineThread::Run()
{
while (m_CurrentState.load(std::memory_order_relaxed) != ThreadStatus::TERMINATED)
{
if (m_CurrentState.load(std::memory_order_relaxed) == ThreadStatus::DRAWCALL_TOP)
{
while (!m_pRenderEngine->m_DrawcallSetupComplete.load(std::memory_order_relaxed))
{
// Don't start processing a drawcall until all threads have draw parameters assigned
}
// Drawcall received, switch to processing it
if (m_ActiveDrawParams.m_IsIndexed)
{
ProcessDrawcall<true>();
}
else
{
ProcessDrawcall<false>();
}
}
std::this_thread::yield();
}
}
template<bool IsIndexed>
void PipelineThread::ProcessDrawcall()
{
LOG("Thread %d drawcall processing begins\n", m_ThreadIdx);
ASSERT(m_pRenderEngine->m_DrawcallSetupComplete.load());
// Drawcall starts with geometry processing
m_CurrentState.store(ThreadStatus::DRAWCALL_GEOMETRY, std::memory_order_relaxed);
LOG("Thread %d processing geometry...\n", m_ThreadIdx);
// Iterate over triangles in assigned drawcall range
for (uint32_t drawIdx = m_ActiveDrawParams.m_ElemsStart, primIdx = m_ActiveDrawParams.m_ElemsStart % m_RenderConfig.m_MaxDrawIterationSize;
drawIdx < m_ActiveDrawParams.m_ElemsEnd;
drawIdx++, primIdx++)
{
// drawIdx = Assigned prim indices which will be only used to fetch indices
// primIdx = Prim index relative to current iteration
// Clip-space vertices to be retrieved from VS
glm::vec4 v0Clip, v1Clip, v2Clip;
// VS
ExecuteVertexShader<IsIndexed>(drawIdx, primIdx, &v0Clip, &v1Clip, &v2Clip);
// Bbox of the primitive which will be computed during clipping
Rect2D bbox;
// CLIPPER
if (!ExecuteFullTriangleClipping(primIdx, v0Clip, v1Clip, v2Clip, &bbox))
{
// Triangle clipped, proceed iteration with next primitive
continue;
}
// TRIANGLE SETUP & CULL
if (!ExecuteTriangleSetupAndCull(primIdx, v0Clip, v1Clip, v2Clip))
{
// Triangle culled, proceed iteration with next primitive
continue;
}
// BINNER
ExecuteBinner(primIdx, v0Clip, v1Clip, v2Clip, bbox);
}
ASSERT(m_CurrentState.load() <= ThreadStatus::DRAWCALL_BINNING);
LOG("Thread %d post-binning sync point\n", m_ThreadIdx);
// To preserve rendering order, we must ensure that all threads finish binning primitives to tiles
// before rasterization is started. To do that, we will stall all threads to sync @DRAWCALL_RASTERIZATION
// Set state to post binning and stall until all PipelineThreads complete binning
m_CurrentState.store(ThreadStatus::DRAWCALL_SYNC_POINT_POST_BINNER, std::memory_order_release);
m_pRenderEngine->WaitForPipelineThreadsToCompleteBinning();
LOG("Thread %d post-binning sync point reached!\n", m_ThreadIdx);
// State must have been set to rasterization by RenderEngine
// when binnnig is "signaled" to have ended
ASSERT(m_CurrentState.load() < ThreadStatus::DRAWCALL_SYNC_POINT_POST_RASTER);
LOG("Thread %d rasterizing...\n", m_ThreadIdx);
// RASTERIZATION
ExecuteRasterizer();
LOG("Thread %d post-raster sync point\n", m_ThreadIdx);
// Rasterization completed, set state to post raster and
// stall until all PipelineThreads complete rasterization.
// We need this sync because when (N-x) threads finish rasterization and
// reach the end of tile queue while x threads are still busy rasterizing tile blocks,
// we must ensure that none of the (N-x) non-busy threads will go ahead and start fragment-shading tiles
// whose blocks could be currently still rasterized by x remaining threads
m_CurrentState.store(ThreadStatus::DRAWCALL_SYNC_POINT_POST_RASTER, std::memory_order_release);
m_pRenderEngine->WaitForPipelineThreadsToCompleteRasterization();
LOG("Thread %d post-raster sync point reached!\n", m_ThreadIdx);
// State must have been set to fragment shader by RenderEngine
// when rasterization is "signaled" to have ended
ASSERT(m_CurrentState.load() == ThreadStatus::DRAWCALL_FRAGMENTSHADER);
LOG("Thread %d fragment-shading...\n", m_ThreadIdx);
// FS
ExecuteFragmentShader();
LOG("Thread %d drawcall ended\n", m_ThreadIdx);
// Draw iteration completed
m_CurrentState.store(ThreadStatus::DRAWCALL_BOTTOM, std::memory_order_relaxed);
}
template<bool IsIndexed>
void PipelineThread::ExecuteVertexShader(uint32_t drawIdx, uint32_t primIdx, glm::vec4* pV0Clip, glm::vec4* pV1Clip, glm::vec4* pV2Clip)
{
uint8_t* pVertexBuffer = static_cast<uint8_t*>(m_pRenderEngine->m_pVertexBuffer);
IndexBuffer* pIndexBuffer = m_pRenderEngine->m_pIndexBuffer;
ASSERT((pVertexBuffer != nullptr) && (!IsIndexed || (pIndexBuffer != nullptr)));
ConstantBuffer* pConstantBuffer = m_pRenderEngine->m_pConstantBuffer;
uint32_t vertexStride = m_pRenderEngine->m_VertexInputStride;
uint32_t vertexOffset = m_ActiveDrawParams.m_VertexOffset;
VertexAttributes* pTempVertexAttrib0 = &m_TempVertexAttributes[0];
VertexAttributes* pTempVertexAttrib1 = &m_TempVertexAttributes[1];
VertexAttributes* pTempVertexAttrib2 = &m_TempVertexAttributes[2];
VertexShader VS = m_pRenderEngine->m_VertexShader;
ASSERT(VS != nullptr);
if constexpr (g_scVertexShaderCacheEnabled && IsIndexed)
{
uint32_t cacheEntry0 = UINT32_MAX;
uint32_t cacheEntry1 = UINT32_MAX;
uint32_t cacheEntry2 = UINT32_MAX;
uint32_t vertexIdx0 = pIndexBuffer[vertexOffset + (3 * drawIdx + 0)];
uint32_t vertexIdx1 = pIndexBuffer[vertexOffset + (3 * drawIdx + 1)];
uint32_t vertexIdx2 = pIndexBuffer[vertexOffset + (3 * drawIdx + 2)];
if (PerformVertexCacheLookup(vertexIdx0, &cacheEntry0))
{
// Vertex 0 is found in the cache, skip VS and fetch cached data
CopyVertexData(cacheEntry0, pV0Clip, pTempVertexAttrib0);
}
else
{
// Vertex 0 is not found in the cache,
// first invoke VS and then cache the clip-space position & vertex attributes
uint8_t* pVertIn0 = &pVertexBuffer[vertexStride * vertexIdx0];
*pV0Clip = VS(pVertIn0, pTempVertexAttrib0, pConstantBuffer);
CacheVertexData(vertexIdx0, *pV0Clip, *pTempVertexAttrib0);
}
if (PerformVertexCacheLookup(vertexIdx1, &cacheEntry1))
{
// Vertex 1 is found in the cache, skip VS and fetch cached data
CopyVertexData(cacheEntry1, pV1Clip, pTempVertexAttrib1);
}
else
{
// Vertex 1 is not found in the cache,
// first invoke VS and then cache the clip-space position & vertex attributes
uint8_t* pVertIn1 = &pVertexBuffer[vertexStride * vertexIdx1];
*pV1Clip = VS(pVertIn1, pTempVertexAttrib1, pConstantBuffer);
CacheVertexData(vertexIdx1, *pV1Clip, *pTempVertexAttrib1);
}
if (PerformVertexCacheLookup(vertexIdx2, &cacheEntry2))
{
// Vertex 2 is found in the cache, skip VS and fetch cached data
CopyVertexData(cacheEntry2, pV2Clip, pTempVertexAttrib2);
}
else
{
// Vertex 2 is not found in the cache,
// first invoke VS and then cache the clip-space position & vertex attributes
uint8_t* pVertIn2 = &pVertexBuffer[vertexStride * vertexIdx2];
*pV2Clip = VS(pVertIn2, pTempVertexAttrib2, pConstantBuffer);
CacheVertexData(vertexIdx0, *pV2Clip, *pTempVertexAttrib2);
}
}
else if (IsIndexed)
{
// VS$ disabled, don't look up vertices in the cache
// Fetch pointers to vertex input that'll be passed to vertex shader
uint8_t* pVertIn0 = &pVertexBuffer[vertexStride * pIndexBuffer[vertexOffset + (3 * drawIdx + 0)]];
uint8_t* pVertIn1 = &pVertexBuffer[vertexStride * pIndexBuffer[vertexOffset + (3 * drawIdx + 1)]];
uint8_t* pVertIn2 = &pVertexBuffer[vertexStride * pIndexBuffer[vertexOffset + (3 * drawIdx + 2)]];
// Invoke vertex shader with vertex attributes payload
*pV0Clip = VS(pVertIn0, pTempVertexAttrib0, pConstantBuffer);
*pV1Clip = VS(pVertIn1, pTempVertexAttrib1, pConstantBuffer);
*pV2Clip = VS(pVertIn2, pTempVertexAttrib2, pConstantBuffer);
}
else
{
// Fetch pointers to vertex input that'll be passed to vertex shader
uint8_t* pVertIn0 = &pVertexBuffer[vertexOffset + vertexStride * (3 * drawIdx + 0)];
uint8_t* pVertIn1 = &pVertexBuffer[vertexOffset + vertexStride * (3 * drawIdx + 1)];
uint8_t* pVertIn2 = &pVertexBuffer[vertexOffset + vertexStride * (3 * drawIdx + 2)];
// Invoke vertex shader with vertex attributes payload
*pV0Clip = VS(pVertIn0, pTempVertexAttrib0, pConstantBuffer);
*pV1Clip = VS(pVertIn1, pTempVertexAttrib1, pConstantBuffer);
*pV2Clip = VS(pVertIn2, pTempVertexAttrib2, pConstantBuffer);
}
// Calculate interpolation data for active vertex attributes
CalculateInterpolationCoefficients(primIdx, *pTempVertexAttrib0, *pTempVertexAttrib1, *pTempVertexAttrib2);
}
void PipelineThread::CopyVertexData(uint32_t cacheEntry, glm::vec4* pVClip, VertexAttributes* pTempVertexAttrib)
{
ASSERT((pVClip != nullptr) && (pTempVertexAttrib != nullptr));
// Copy cached clip-space positions
*pVClip = m_VertexCacheEntries[cacheEntry].m_ClipPos;
// Copy vertex (only active!) attributes
memcpy(
pTempVertexAttrib->m_Attributes2,
m_VertexCacheEntries[cacheEntry].m_VertexAttribs.m_Attributes2,
sizeof(glm::vec2) * m_pRenderEngine->m_ShaderMetadata.m_NumVec2Attributes);
memcpy(
pTempVertexAttrib->m_Attributes3,
m_VertexCacheEntries[cacheEntry].m_VertexAttribs.m_Attributes3,
sizeof(glm::vec3) * m_pRenderEngine->m_ShaderMetadata.m_NumVec3Attributes);
memcpy(
pTempVertexAttrib->m_Attributes4,
m_VertexCacheEntries[cacheEntry].m_VertexAttribs.m_Attributes4,
sizeof(glm::vec4) * m_pRenderEngine->m_ShaderMetadata.m_NumVec4Attributes);
}
void PipelineThread::CacheVertexData(uint32_t vertexIdx, const glm::vec4& vClip, const tyler::VertexAttributes& tempVertexAttrib)
{
// Check if VS$ cache has space and if so, append new vertex data
if (m_NumVertexCacheEntries < g_scVertexShaderCacheSize)
{
m_CachedVertexIndices[m_NumVertexCacheEntries] = vertexIdx;
m_VertexCacheEntries[m_NumVertexCacheEntries].m_ClipPos = vClip;
m_VertexCacheEntries[m_NumVertexCacheEntries].m_VertexAttribs = tempVertexAttrib;
++m_NumVertexCacheEntries;
}
}
bool PipelineThread::PerformVertexCacheLookup(uint32_t vertexIdx, uint32_t* pCachedIdx)
{
// Go through all cached entries
for (uint32_t idx = 0; idx < m_NumVertexCacheEntries; idx++)
{
if (m_CachedVertexIndices[idx] == vertexIdx)
{
// Vertex is found in VS$, just return its entry index within the cache
LOG("Vertex %d found in the VS$\n", vertexIdx);
*pCachedIdx = idx;
return true;
}
}
return false;
}
bool PipelineThread::ExecuteFullTriangleClipping(uint32_t primIdx, const glm::vec4& v0Clip, const glm::vec4& v1Clip, const glm::vec4& v2Clip, Rect2D* pBbox)
{
ASSERT(pBbox != nullptr);
if constexpr (g_scFullTriangleClippingEnabled)
{
// Clip-space positions are to be bounded by:
// -w < x < w -> LEFT/RIGHT
// -w < y < w -> TOP/BOTTOM
// 0 < z < w -> NEAR/FAR
// However, we will only clip primitives that are *completely* outside of any of clipping planes.
// This means that, triangles intersecting view frustum are passed as-is, to be rasterized as usual.
// Because we're utilizing homogeneous rasterization, we don't need to do explcit line-clipping here.
// Clip against w+x=0 left plane
bool allOutsideLeftPlane =
(v0Clip.x < -v0Clip.w) &&
(v1Clip.x < -v1Clip.w) &&
(v2Clip.x < -v2Clip.w);
bool allInsideLeftPlane =
(v0Clip.x >= -v0Clip.w) &&
(v1Clip.x >= -v1Clip.w) &&
(v2Clip.x >= -v2Clip.w);
// Clip against w-x=0 right plane
bool allOutsideRightPlane =
(v0Clip.x > v0Clip.w) &&
(v1Clip.x > v1Clip.w) &&
(v2Clip.x > v2Clip.w);
bool allInsideRightPlane =
(v0Clip.x <= v0Clip.w) &&
(v1Clip.x <= v1Clip.w) &&
(v2Clip.x <= v2Clip.w);
// Clip against w+y top plane
bool allOutsideBottomPlane =
(v0Clip.y < -v0Clip.w) &&
(v1Clip.y < -v1Clip.w) &&
(v2Clip.y < -v2Clip.w);
bool allInsideBottomPlane =
(v0Clip.y >= -v0Clip.w) &&
(v1Clip.y >= -v1Clip.w) &&
(v2Clip.y >= -v2Clip.w);
// Clip against w-y bottom plane
bool allOutsideTopPlane =
(v0Clip.y > v0Clip.w) &&
(v1Clip.y > v1Clip.w) &&
(v2Clip.y > v2Clip.w);
bool allInsideTopPlane =
(v0Clip.y <= v0Clip.w) &&
(v1Clip.y <= v1Clip.w) &&
(v2Clip.y <= v2Clip.w);
// Clip against 0<z near plane
bool allOutsideNearPlane =
(v0Clip.z < 0.f) &&
(v1Clip.z < 0.f) &&
(v2Clip.z < 0.f);
bool allInsideNearPlane =
(v0Clip.z >= 0.f) &&
(v1Clip.z >= 0.f) &&
(v2Clip.z >= 0.f);
// Clip against z>w far plane
bool allOutsideFarPlane =
(v0Clip.z > v0Clip.w) &&
(v1Clip.z > v1Clip.w) &&
(v2Clip.z > v2Clip.w);
bool allInsideFarPlane =
(v0Clip.z <= v0Clip.w) &&
(v1Clip.z <= v1Clip.w) &&
(v2Clip.z <= v2Clip.w);
float width = static_cast<float>(m_pRenderEngine->m_Framebuffer.m_Width);
float height = static_cast<float>(m_pRenderEngine->m_Framebuffer.m_Height);
if (allOutsideLeftPlane ||
allOutsideRightPlane ||
allOutsideBottomPlane ||
allOutsideTopPlane ||
allOutsideNearPlane ||
allOutsideFarPlane)
{
// TRIVIALREJECT case
LOG("Prim %d TR'd in FT-clipper by thread %d", primIdx, m_ThreadIdx);
// Primitive completely outside of one of the clip planes, discard it
return false;
}
else if (allInsideLeftPlane &&
allInsideRightPlane &&
allInsideBottomPlane &&
allInsideTopPlane &&
allInsideNearPlane &&
allInsideFarPlane)
{
// TRIVIALACCEPT
LOG("Prim %d TA'd in FT-clipper by thread %d", primIdx, m_ThreadIdx);
// Primitive is completely inside view frustum
// Compute bounding box
Rect2D bbox = ComputeBoundingBox(v0Clip, v1Clip, v2Clip, width, height);
// Clamp bbox to screen extents
bbox.m_MinX = glm::max(0.f, bbox.m_MinX);
bbox.m_MaxX = glm::min(width, bbox.m_MaxX);
bbox.m_MinY = glm::max(0.f, bbox.m_MinY);
bbox.m_MaxY = glm::min(height, bbox.m_MaxY);
*pBbox = bbox;
// Cache bbox of the primitive
m_pRenderEngine->m_SetupBuffers.m_pPrimBBoxes[primIdx] = bbox;
return true;
}
else
{
// MUSTCLIP
LOG("Prim %d MUSTCLIP'd by thread %d", primIdx, m_ThreadIdx);
// Primitive is partially inside view frustum, but we don't clip for this
// so we must be conservative and return the whole range to rasterize further.
// Note that is *overly* conservative in practice; we could do better by implementing
// Blinn's method of screen coverage calculation properly but it's an overkill here.
Rect2D bbox =
{
0.f,
0.f,
width,
height
};
*pBbox = bbox;
// Cache bbox of the primitive
m_pRenderEngine->m_SetupBuffers.m_pPrimBBoxes[primIdx] = bbox;
return true;
}
}
else
{
// FT clipping disabled
// Compute bounding box
float width = static_cast<float>(m_pRenderEngine->m_Framebuffer.m_Width);
float height = static_cast<float>(m_pRenderEngine->m_Framebuffer.m_Height);
Rect2D bbox = ComputeBoundingBox(v0Clip, v1Clip, v2Clip, width, height);
if ((bbox.m_MinX >= width) ||
(bbox.m_MaxX < 0.f) ||
(bbox.m_MinY >= height) ||
(bbox.m_MaxY < 0.f))
{
// If tri's bbox exceeds screen bounds, discard it
return false;
}
else
{
// Clamp bbox to screen bounds
bbox.m_MinX = glm::max(0.f, bbox.m_MinX);
bbox.m_MaxX = glm::min(width, bbox.m_MaxX);
bbox.m_MinY = glm::max(0.f, bbox.m_MinY);
bbox.m_MaxY = glm::min(height, bbox.m_MaxY);
*pBbox = bbox;
// Cache bbox of the primitive
m_pRenderEngine->m_SetupBuffers.m_pPrimBBoxes[primIdx] = bbox;
// No clipping
return true;
}
}
}
bool PipelineThread::ExecuteTriangleSetupAndCull(uint32_t primIdx, const glm::vec4& v0Clip, const glm::vec4& v1Clip, const glm::vec4& v2Clip)
{
//TODO: Cull degenerate and back-facing primitives if culling is enabled
// Transform a given vertex in clip-space [-w,w] to device-space homogeneous coordinates [0, {w|h}]
#define TO_HOMOGEN(clipPos, width, height) glm::vec4((width * (clipPos.x + clipPos.w) * 0.5f), (height * (clipPos.y + clipPos.w) * 0.5f), clipPos.z, clipPos.w)
float fbWidth = static_cast<float>(m_pRenderEngine->m_Framebuffer.m_Width);
float fbHeight = static_cast<float>(m_pRenderEngine->m_Framebuffer.m_Height);
// First, transform clip-space (x, y, z, w) vertices to device-space 2D homogeneous coordinates (x, y, w)
const glm::vec4 v0Homogen = TO_HOMOGEN(v0Clip, fbWidth, fbHeight);
const glm::vec4 v1Homogen = TO_HOMOGEN(v1Clip, fbWidth, fbHeight);
const glm::vec4 v2Homogen = TO_HOMOGEN(v2Clip, fbWidth, fbHeight);
// To calculate EE coefficients, we need to set up a "vertex matrix" and invert it
// M = | x0 x1 x2 |
// | y0 y1 y2 |
// | w0 w1 w2 |
// Alternatively, we can rely on the following relation between an inverse and adjoint of a matrix: inv(M) = adj(M)/det(M)
// Since we use homogeneous coordinates, it's sufficient to only compute adjoint matrix:
// A = | a0 b0 c0 |
// | a1 b1 c1 |
// | a2 b2 c2 |
float a0 = (v2Homogen.y * v1Homogen.w) - (v1Homogen.y * v2Homogen.w);
float a1 = (v0Homogen.y * v2Homogen.w) - (v2Homogen.y * v0Homogen.w);
float a2 = (v1Homogen.y * v0Homogen.w) - (v0Homogen.y * v1Homogen.w);
float b0 = (v1Homogen.x * v2Homogen.w) - (v2Homogen.x * v1Homogen.w);
float b1 = (v2Homogen.x * v0Homogen.w) - (v0Homogen.x * v2Homogen.w);
float b2 = (v0Homogen.x * v1Homogen.w) - (v1Homogen.x * v0Homogen.w);
float c0 = (v2Homogen.x * v1Homogen.y) - (v1Homogen.x * v2Homogen.y);
float c1 = (v0Homogen.x * v2Homogen.y) - (v2Homogen.x * v0Homogen.y);
float c2 = (v1Homogen.x * v0Homogen.y) - (v0Homogen.x * v1Homogen.y);
// Additionally,
// det(M) == 0 -> degenerate/zero-area triangle
// det(M) < 0 -> back-facing triangle
float detM = (c0 * v0Homogen.w) + (c1 * v1Homogen.w) + (c2 * v2Homogen.w);
// Assign computed EE coefficients for given primitive
m_pRenderEngine->m_SetupBuffers.m_pEdgeCoefficients[3 * primIdx + 0] = { a0, b0, c0 };
m_pRenderEngine->m_SetupBuffers.m_pEdgeCoefficients[3 * primIdx + 1] = { a1, b1, c1 };
m_pRenderEngine->m_SetupBuffers.m_pEdgeCoefficients[3 * primIdx + 2] = { a2, b2, c2 };
// Store clip-space Z interpolation deltas in the setup buffer that will be used for perspective-correct interpolation of Z
m_pRenderEngine->m_SetupBuffers.m_pInterpolatedZValues[primIdx] = { (v0Clip.z - v2Clip.z), (v1Clip.z - v2Clip.z), v2Clip.z };
//TODO: Proper culling? Render back-facing tris by flipping sign of EEs?!
// Return whether the primitive should be culled
return detM > 0.f;
}
void PipelineThread::ExecuteBinner(uint32_t primIdx, const Rect2D& bbox)
{
LOG("Thread %d binning prim %d\n", m_ThreadIdx, primIdx);
// Binning in progress now
m_CurrentState.store(ThreadStatus::DRAWCALL_BINNING, std::memory_order_relaxed);
float fbWidth = static_cast<float>(m_pRenderEngine->m_Framebuffer.m_Width);
float fbHeight = static_cast<float>(m_pRenderEngine->m_Framebuffer.m_Height);
// FT clipper must have clamped bbox to screen extents!
ASSERT((bbox.m_MinX >= 0.f) && (bbox.m_MaxX >= 0.f) && (bbox.m_MinY >= 0.f) && (bbox.m_MaxY >= 0.f));
ASSERT((bbox.m_MinX <= bbox.m_MaxX) && (bbox.m_MinY <= bbox.m_MaxY));
// Given a tile size and frame buffer dimensions, find min/max range of the tiles that fall within bbox computed above
// which we're going to iterate over, in order to determine if the primitive should be binned or not
// Use floor(), min indices are inclusive
uint32_t minTileX = static_cast<uint32_t>(glm::floor(bbox.m_MinX / m_RenderConfig.m_TileSize));
uint32_t minTileY = static_cast<uint32_t>(glm::floor(bbox.m_MinY / m_RenderConfig.m_TileSize));
// Use ceil(), max indices are exclusive
uint32_t maxTileX = static_cast<uint32_t>(glm::ceil(bbox.m_MaxX / m_RenderConfig.m_TileSize));
uint32_t maxTileY = static_cast<uint32_t>(glm::ceil(bbox.m_MaxY / m_RenderConfig.m_TileSize));
ASSERT((minTileX <= maxTileX) && (maxTileX <= m_pRenderEngine->m_NumTilePerRow));
ASSERT((minTileY <= maxTileY) && (maxTileY <= m_pRenderEngine->m_NumTilePerColumn));
// Fetch edge equation coefficients computed in triangle setup
glm::vec3 ee0 = m_pRenderEngine->m_SetupBuffers.m_pEdgeCoefficients[3 * primIdx + 0];
glm::vec3 ee1 = m_pRenderEngine->m_SetupBuffers.m_pEdgeCoefficients[3 * primIdx + 1];
glm::vec3 ee2 = m_pRenderEngine->m_SetupBuffers.m_pEdgeCoefficients[3 * primIdx + 2];
// Normalize edge functions
ee0 /= (glm::abs(ee0.x) + glm::abs(ee0.y));
ee1 /= (glm::abs(ee1.x) + glm::abs(ee1.y));
ee2 /= (glm::abs(ee2.x) + glm::abs(ee2.y));
// Indices of tile corners:
// LL -> 0 LR -> 1
// UL -> 2 UR -> 3
static const glm::vec2 scTileCornerOffsets[] =
{
{ 0.f, 0.f}, // LL (origin)
{ m_RenderConfig.m_TileSize, 0.f }, // LR
{ 0.f, m_RenderConfig.m_TileSize }, // UL
{ m_RenderConfig.m_TileSize, m_RenderConfig.m_TileSize} // UR
};
// (x, y) -> sample location | (a, b, c) -> edge equation coefficients
// E(x, y) = (a * x) + (b * y) + c
// E(x + s, y + t) = E(x, y) + (a * s) + (b * t)
// Based on edge normal n=(a, b), set up tile TR corners for each edge
const uint8_t edge0TRCorner = (ee0.y >= 0.f) ? ((ee0.x >= 0.f) ? 3u : 2u) : (ee0.x >= 0.f) ? 1u : 0u;
const uint8_t edge1TRCorner = (ee1.y >= 0.f) ? ((ee1.x >= 0.f) ? 3u : 2u) : (ee1.x >= 0.f) ? 1u : 0u;
const uint8_t edge2TRCorner = (ee2.y >= 0.f) ? ((ee2.x >= 0.f) ? 3u : 2u) : (ee2.x >= 0.f) ? 1u : 0u;
// TA corner is the one diagonal from TR corner calculated above
const uint8_t edge0TACorner = 3u - edge0TRCorner;
const uint8_t edge1TACorner = 3u - edge1TRCorner;
const uint8_t edge2TACorner = 3u - edge2TRCorner;
// Evaluate edge function for the first tile within [minTile, maxTile] region
// once and re-use it by stepping from it within following nested loop
// Tile origin
const float tilePosX = glm::min(fbWidth, static_cast<float>(minTileX * m_RenderConfig.m_TileSize));
const float tilePosY = glm::min(fbHeight, static_cast<float>(minTileY * m_RenderConfig.m_TileSize));
// No need to fetch tile positions from tiles, ensure the calculations match nevertheless
ASSERT((tilePosX == m_pRenderEngine->m_TileList[m_pRenderEngine->GetGlobalTileIndex(minTileX, minTileY)].m_PosX));
ASSERT((tilePosY == m_pRenderEngine->m_TileList[m_pRenderEngine->GetGlobalTileIndex(minTileX, minTileY)].m_PosY));
// Evaluaate edge equation at first tile origin
const float edgeFunc0 = ee0.z + ((ee0.x * tilePosX) + (ee0.y * tilePosY));
const float edgeFunc1 = ee1.z + ((ee1.x * tilePosX) + (ee1.y * tilePosY));
const float edgeFunc2 = ee2.z + ((ee2.x * tilePosX) + (ee2.y * tilePosY));
// Iterate over calculated range of tiles
for (uint32_t ty = minTileY, tyy = 0; ty < maxTileY; ty++, tyy++)
{
for (uint32_t tx = minTileX, txx = 0; tx < maxTileX; tx++, txx++)
{
// Using EE coefficients calculated in TriangleSetup stage and positive half-space tests, determine one of three cases possible for each tile:
// 1) TrivialReject -- tile within tri's bbox does not intersect tri -> move on
// 2) TrivialAccept -- tile within tri's bbox is completely within tri -> emit a full-tile coverage mask
// 3) Overlap -- tile within tri's bbox intersects tri -> bin the triangle to given tile for further rasterization where block/pixel-level coverage masks will be emitted
// (txx, tyy) = how many steps are done per dimension
const float txxOffset = static_cast<float>(txx * m_RenderConfig.m_TileSize);
const float tyyOffset = static_cast<float>(tyy * m_RenderConfig.m_TileSize);
// Step from edge function computed above for the first tile in bbox
float edgeFuncTR0 = edgeFunc0 + ((ee0.x * (scTileCornerOffsets[edge0TRCorner].x + txxOffset)) + (ee0.y * (scTileCornerOffsets[edge0TRCorner].y + tyyOffset)));
float edgeFuncTR1 = edgeFunc1 + ((ee1.x * (scTileCornerOffsets[edge1TRCorner].x + txxOffset)) + (ee1.y * (scTileCornerOffsets[edge1TRCorner].y + tyyOffset)));
float edgeFuncTR2 = edgeFunc2 + ((ee2.x * (scTileCornerOffsets[edge2TRCorner].x + txxOffset)) + (ee2.y * (scTileCornerOffsets[edge2TRCorner].y + tyyOffset)));
// If TR corner of the tile is outside any edge, reject whole tile
bool TRForEdge0 = (edgeFuncTR0 < 0.f);
bool TRForEdge1 = (edgeFuncTR1 < 0.f);
bool TRForEdge2 = (edgeFuncTR2 < 0.f);
if (TRForEdge0 || TRForEdge1 || TRForEdge2)
{
LOG("Tile %d TR'd by thread %d\n", m_pRenderEngine->GetGlobalTileIndex(tx, ty), m_ThreadIdx);
// TrivialReject
// Tile is completely outside of one or more edges
continue;
}
else
{
// Tile is partially or completely inside one or more edges, do TrivialAccept tests first
// Compute edge functions at TA corners based on edge function at first tile origin
float edgeFuncTA0 = edgeFunc0 + ((ee0.x * (scTileCornerOffsets[edge0TACorner].x + txxOffset)) + (ee0.y * (scTileCornerOffsets[edge0TACorner].y + tyyOffset)));
float edgeFuncTA1 = edgeFunc1 + ((ee1.x * (scTileCornerOffsets[edge1TACorner].x + txxOffset)) + (ee1.y * (scTileCornerOffsets[edge1TACorner].y + tyyOffset)));
float edgeFuncTA2 = edgeFunc2 + ((ee2.x * (scTileCornerOffsets[edge2TACorner].x + txxOffset)) + (ee2.y * (scTileCornerOffsets[edge2TACorner].y + tyyOffset)));
// If TA corner of the tile is outside all edges, accept whole tile
bool TAForEdge0 = (edgeFuncTA0 >= 0.f);
bool TAForEdge1 = (edgeFuncTA1 >= 0.f);
bool TAForEdge2 = (edgeFuncTA2 >= 0.f);
if (TAForEdge0 && TAForEdge1 && TAForEdge2)
{
// TrivialAccept
// Tile is completely inside of the triangle, no further rasterization is needed,
// whole tile will be fragment-shaded!
LOG("Tile %d TA'd by thread %d\n", m_pRenderEngine->GetGlobalTileIndex(tx, ty), m_ThreadIdx);
// Append tile to the rasterizer queue
m_pRenderEngine->EnqueueTileForRasterization(m_pRenderEngine->GetGlobalTileIndex(tx, ty));
CoverageMask mask;
mask.m_SampleX = static_cast<uint32_t>(tilePosX + txxOffset); // Based off of first tile position calculated above
mask.m_SampleY = static_cast<uint32_t>(tilePosY + tyyOffset); // Based off of first tile position calculated above
mask.m_PrimIdx = primIdx;
mask.m_Type = CoverageMaskType::TILE;
// Emit full-tile coverage mask
m_pRenderEngine->AppendCoverageMask(
m_ThreadIdx,
m_pRenderEngine->GetGlobalTileIndex(tx, ty),
mask);
}
else
{
LOG("Tile %d binned by thread %d\n", m_pRenderEngine->GetGlobalTileIndex(tx, ty), m_ThreadIdx);
// Overlap
// Tile is partially covered by the triangle, bin the triangle for the tile
m_pRenderEngine->BinPrimitiveForTile(
m_ThreadIdx,
m_pRenderEngine->GetGlobalTileIndex(tx, ty),
primIdx);
}
}
}
}
}
void PipelineThread::ExecuteRasterizer()
{
// Request next (global) index of the tile to be rasterized at block level from RenderEngine
uint32_t nextTileIdx;
while ((nextTileIdx = m_pRenderEngine->FetchNextTileForRasterization()) != g_scInvalidTileIndex)
{
LOG("Thread %d rasterizing tile %d\n", m_ThreadIdx, nextTileIdx);
ASSERT(nextTileIdx < (m_pRenderEngine->m_NumTilePerRow * m_pRenderEngine->m_NumTilePerColumn));
// Grabbed next tile from the queue, scan through its per-thread bins to rasterize the primitives
ASSERT(m_pRenderEngine->m_BinList[nextTileIdx].size() == m_RenderConfig.m_NumPipelineThreads);
// Tile must have been appended to the rasterizer queue, otherwise binning was incorrectly done for primitive!
ASSERT(m_pRenderEngine->m_TileList[nextTileIdx].m_IsTileQueued.test_and_set());
// Tile origin
const float tilePosX = m_pRenderEngine->m_TileList[nextTileIdx].m_PosX;
const float tilePosY = m_pRenderEngine->m_TileList[nextTileIdx].m_PosY;
// Go through all per-thread bins in-order
for (uint32_t i = 0; i < m_RenderConfig.m_NumPipelineThreads; i++)
{
// If a tile was trivially accepted, its bin will be empty
const std::vector<uint32_t>& perThreadBin = m_pRenderEngine->m_BinList[nextTileIdx][i];
LOG("Tile %d thread %d bin size: %d\n", nextTileIdx, i, perThreadBin.size());
// Go through all primitives in current per-thread bin in-order
for (uint32_t p = 0; p < perThreadBin.size(); p++)
{
// Get next (global) primitive index to be rasterized
uint32_t primIdx = perThreadBin[p];
// Copy prim's bbox to clamp it to the tile edges
Rect2D bbox = m_pRenderEngine->m_SetupBuffers.m_pPrimBBoxes[primIdx];
bbox.m_MinX = glm::max(bbox.m_MinX, tilePosX);
bbox.m_MinY = glm::max(bbox.m_MinY, tilePosY);
bbox.m_MaxX = glm::min(bbox.m_MaxX, tilePosX + m_RenderConfig.m_TileSize);
bbox.m_MaxY = glm::min(bbox.m_MaxY, tilePosY + m_RenderConfig.m_TileSize);
// In case bbox is screwed up after clamping to the tile edges
ASSERT((bbox.m_MinX <= bbox.m_MaxX) && (bbox.m_MinY <= bbox.m_MaxY));
// Given a fixed 8x8 block and tile size, find min/max range of the blocks that fall within bbox computed above
// which we're going to iterate over, in order to determine how blocks within tile are to be rasterized
// Use floor(), min indices are inclusive
uint32_t minBlockX = static_cast<uint32_t>(glm::floor((bbox.m_MinX - tilePosX) / g_scPixelBlockSize));
uint32_t minBlockY = static_cast<uint32_t>(glm::floor((bbox.m_MinY - tilePosY) / g_scPixelBlockSize));
// Use ceil(), max indices are exclusive
uint32_t maxBlockX = static_cast<uint32_t>(glm::ceil((bbox.m_MaxX - tilePosX) / g_scPixelBlockSize));
uint32_t maxBlockY = static_cast<uint32_t>(glm::ceil((bbox.m_MaxY - tilePosY) / g_scPixelBlockSize));
ASSERT((minBlockX <= maxBlockX) && (maxBlockX <= m_RenderConfig.m_TileSize / g_scPixelBlockSize));
ASSERT((minBlockY <= maxBlockY) && (maxBlockY <= m_RenderConfig.m_TileSize / g_scPixelBlockSize));
// Use EE coefficients calculated in TriangleSetup again to rasterize primitive at the 8x8 block level
glm::vec3 ee0 = m_pRenderEngine->m_SetupBuffers.m_pEdgeCoefficients[3 * primIdx + 0];
glm::vec3 ee1 = m_pRenderEngine->m_SetupBuffers.m_pEdgeCoefficients[3 * primIdx + 1];
glm::vec3 ee2 = m_pRenderEngine->m_SetupBuffers.m_pEdgeCoefficients[3 * primIdx + 2];
// Normalize edge functions
ee0 /= (glm::abs(ee0.x) + glm::abs(ee0.y));
ee1 /= (glm::abs(ee1.x) + glm::abs(ee1.y));
ee2 /= (glm::abs(ee2.x) + glm::abs(ee2.y));
static constexpr glm::vec2 scBlockCornerOffsets[] =
{
{ 0.f, 0.f}, // LL (origin)
{ g_scPixelBlockSize, 0.f }, // LR
{ 0.f, g_scPixelBlockSize }, // UL
{ g_scPixelBlockSize, g_scPixelBlockSize} // UR
};
// (x, y) -> sample location | (a, b, c) -> edge equation coefficients
// E(x, y) = (a * x) + (b * y) + c
// E(x + s, y + t) = E(x, y) + (a * s) + (b * t)
// Based on edge normal n=(a, b), set up block TR corners for each edge once
const uint8_t edge0TRCorner = (ee0.y >= 0.f) ? ((ee0.x >= 0.f) ? 3u : 2u) : (ee0.x >= 0.f) ? 1u : 0u;
const uint8_t edge1TRCorner = (ee1.y >= 0.f) ? ((ee1.x >= 0.f) ? 3u : 2u) : (ee1.x >= 0.f) ? 1u : 0u;
const uint8_t edge2TRCorner = (ee2.y >= 0.f) ? ((ee2.x >= 0.f) ? 3u : 2u) : (ee2.x >= 0.f) ? 1u : 0u;
const uint8_t edge0TACorner = 3u - edge0TRCorner;
const uint8_t edge1TACorner = 3u - edge1TRCorner;
const uint8_t edge2TACorner = 3u - edge2TRCorner;
// Evaluate edge function for the first block within [minBlock, maxBlock] region
// once and re-use it by stepping from it within following nested loop
const float firstBlockWithinBBoxX = tilePosX + minBlockX * g_scPixelBlockSize;
const float firstBlockWithinBBoxY = tilePosY + minBlockY * g_scPixelBlockSize;
// Evaluate edge equation at first block origin
const float edgeFunc0 = ee0.z + ((ee0.x * firstBlockWithinBBoxX) + (ee0.y * firstBlockWithinBBoxY));
const float edgeFunc1 = ee1.z + ((ee1.x * firstBlockWithinBBoxX) + (ee1.y * firstBlockWithinBBoxY));
const float edgeFunc2 = ee2.z + ((ee2.x * firstBlockWithinBBoxX) + (ee2.y * firstBlockWithinBBoxY));
// Iterate over calculated range of blocks within the tile
for (uint32_t by = minBlockY, byy = 0; by < maxBlockY; by++, byy++)
{
for (uint32_t bx = minBlockX, bxx = 0; bx < maxBlockX; bx++, bxx++)
{
// Using EE coefficients calculated in TriangleSetup stage and positive half-space tests, determine one of three cases possible for each block:
// 1) TrivialReject -- block within tri's bbox does not intersect tri -> move on
// 2) TrivialAccept -- block within tri's bbox is completely within tri -> emit a full-block coverage mask
// 3) Overlap -- block within tri's bbox intersects tri -> descend into block level to emit coverage masks at pixel granularity
// (bxx, byy) = How many steps are done per dimension
const float bxxOffset = static_cast<float>(bxx * g_scPixelBlockSize);
const float byyOffset = static_cast<float>(byy * g_scPixelBlockSize);
// Step down from edge function computed above for the first block in bbox
float edgeFuncTR0 = edgeFunc0 + ((ee0.x * (scBlockCornerOffsets[edge0TRCorner].x + bxxOffset)) + (ee0.y * (scBlockCornerOffsets[edge0TRCorner].y + byyOffset)));
float edgeFuncTR1 = edgeFunc1 + ((ee1.x * (scBlockCornerOffsets[edge1TRCorner].x + bxxOffset)) + (ee1.y * (scBlockCornerOffsets[edge1TRCorner].y + byyOffset)));
float edgeFuncTR2 = edgeFunc2 + ((ee2.x * (scBlockCornerOffsets[edge2TRCorner].x + bxxOffset)) + (ee2.y * (scBlockCornerOffsets[edge2TRCorner].y + byyOffset)));
// If TR corner of the block is outside an edge, reject whole block
bool TRForEdge0 = (edgeFuncTR0 < 0.f);
bool TRForEdge1 = (edgeFuncTR1 < 0.f);
bool TRForEdge2 = (edgeFuncTR2 < 0.f);
if (TRForEdge0 || TRForEdge1 || TRForEdge2)
{
LOG("Tile %d block (%d, %d) TR'd by thread %d\n", nextTileIdx, bx, by, m_ThreadIdx);
// TrivialReject
// Block is completely outside of one or more edges
continue;
}
else
{
// Block is partially or completely inside one or more edges, do TrivialAccept tests first
// Compute edge functions at TA corners by stepping from first block position calculated above
float edgeFuncTA0 = edgeFunc0 + ((ee0.x * (scBlockCornerOffsets[edge0TACorner].x + bxxOffset)) + (ee0.y * (scBlockCornerOffsets[edge0TACorner].y + byyOffset)));
float edgeFuncTA1 = edgeFunc1 + ((ee1.x * (scBlockCornerOffsets[edge1TACorner].x + bxxOffset)) + (ee1.y * (scBlockCornerOffsets[edge1TACorner].y + byyOffset)));
float edgeFuncTA2 = edgeFunc2 + ((ee2.x * (scBlockCornerOffsets[edge2TACorner].x + bxxOffset)) + (ee2.y * (scBlockCornerOffsets[edge2TACorner].y + byyOffset)));
// If TA corner of the block is inside all edges, accept whole block
bool TAForEdge0 = (edgeFuncTA0 >= 0.f);
bool TAForEdge1 = (edgeFuncTA1 >= 0.f);
bool TAForEdge2 = (edgeFuncTA2 >= 0.f);
if (TAForEdge0 && TAForEdge1 && TAForEdge2)
{
// TrivialAccept
// Block is completely inside of the triangle, emit a full-block coverage mask
LOG("Tile %d block (%d, %d) TA'd by thread %d\n", nextTileIdx, bx, by, m_ThreadIdx);
CoverageMask mask;
mask.m_SampleX = static_cast<uint32_t>(firstBlockWithinBBoxX + bxxOffset); // Based off of first block position calculated above
mask.m_SampleY = static_cast<uint32_t>(firstBlockWithinBBoxY + byyOffset); // Based off of first block position calculated above
mask.m_PrimIdx = primIdx;
mask.m_Type = CoverageMaskType::BLOCK;
// Emit full-block coverage mask
m_pRenderEngine->AppendCoverageMask(
m_ThreadIdx,
nextTileIdx,
mask);
}
else
{
// Overlap
// Block is partially covered by the triangle, descend into pixel level and perform edge tests
LOG("Tile %d block (%d, %d) overlapping tests by thread %d\n", nextTileIdx, bx, by, m_ThreadIdx);
// Position of the block that we're testing at pixel level
float blockPosX = (firstBlockWithinBBoxX + bxxOffset);
float blockPosY = (firstBlockWithinBBoxY + byyOffset);
// Compute E(x, y) = (x * a) + (y * b) c at block origin once
__m128 sseEdge0FuncAtBlockOrigin = _mm_set1_ps(ee0.z + ((ee0.x * blockPosX) + (ee0.y * blockPosY)));
__m128 sseEdge1FuncAtBlockOrigin = _mm_set1_ps(ee1.z + ((ee1.x * blockPosX) + (ee1.y * blockPosY)));
__m128 sseEdge2FuncAtBlockOrigin = _mm_set1_ps(ee2.z + ((ee2.x * blockPosX) + (ee2.y * blockPosY)));
// Store edge 0 equation coefficients
__m128 sseEdge0A4 = _mm_set_ps1(ee0.x);
__m128 sseEdge0B4 = _mm_set_ps1(ee0.y);
// Store edge 1 equation coefficients
__m128 sseEdge1A4 = _mm_set_ps1(ee1.x);
__m128 sseEdge1B4 = _mm_set_ps1(ee1.y);
// Store edge 2 equation coefficients
__m128 sseEdge2A4 = _mm_set_ps1(ee2.x);
__m128 sseEdge2B4 = _mm_set_ps1(ee2.y);
// Generate masks used for tie-breaking rules (not to double-shade along shared edges)
__m128 sseEdge0A4PositiveOrB4NonNegativeA4Zero = _mm_or_ps(_mm_cmpgt_ps(sseEdge0A4, _mm_setzero_ps()),
_mm_and_ps(_mm_cmpge_ps(sseEdge0B4, _mm_setzero_ps()), _mm_cmpeq_ps(sseEdge0A4, _mm_setzero_ps())));
__m128 sseEdge1A4PositiveOrB4NonNegativeA4Zero = _mm_or_ps(_mm_cmpgt_ps(sseEdge1A4, _mm_setzero_ps()),
_mm_and_ps(_mm_cmpge_ps(sseEdge1B4, _mm_setzero_ps()), _mm_cmpeq_ps(sseEdge1A4, _mm_setzero_ps())));
__m128 sseEdge2A4PositiveOrB4NonNegativeA4Zero = _mm_or_ps(_mm_cmpgt_ps(sseEdge2A4, _mm_setzero_ps()),
_mm_and_ps(_mm_cmpge_ps(sseEdge2B4, _mm_setzero_ps()), _mm_cmpeq_ps(sseEdge2A4, _mm_setzero_ps())));
for (uint32_t py = 0; py < g_scPixelBlockSize; py++)
{
// Store Y positions in current row (all samples on the same row has the same Y position)
__m128 sseY4 = _mm_set_ps1(py + 0.5f);
for (uint32_t px = 0; px < g_scNumEdgeTestsPerRow; px++)
{
// E(x, y) = (x * a) + (y * b) + c
// E(x + s, y + t) = E(x, y) + s * a + t * b
#ifdef _DEBUG
int32_t debugMaskScalar = 0;
{
// Debug for SSE edge tests
float edge0FuncAtBlockOrigin = ee0.z + (ee0.x * blockPosX) + (ee0.y * blockPosY);
float edge1FuncAtBlockOrigin = ee1.z + (ee1.x * blockPosX) + (ee1.y * blockPosY);
float edge2FuncAtBlockOrigin = ee2.z + (ee2.x * blockPosX) + (ee2.y * blockPosY);
// 4 Sample locations
glm::vec2 sample0 = { g_scSIMDWidth * px + 0.5f, py + 0.5f };
glm::vec2 sample1 = { g_scSIMDWidth * px + 1.5f, py + 0.5f };
glm::vec2 sample2 = { g_scSIMDWidth * px + 2.5f, py + 0.5f };
glm::vec2 sample3 = { g_scSIMDWidth * px + 3.5f, py + 0.5f };
bool inside0 =
EvaluateEdgeFunctionIncremental(ee0, sample0, edge0FuncAtBlockOrigin) &&
EvaluateEdgeFunctionIncremental(ee1, sample0, edge1FuncAtBlockOrigin) &&
EvaluateEdgeFunctionIncremental(ee2, sample0, edge2FuncAtBlockOrigin);
bool inside1 =
EvaluateEdgeFunctionIncremental(ee0, sample1, edge0FuncAtBlockOrigin) &&
EvaluateEdgeFunctionIncremental(ee1, sample1, edge1FuncAtBlockOrigin) &&
EvaluateEdgeFunctionIncremental(ee2, sample1, edge2FuncAtBlockOrigin);
bool inside2 =
EvaluateEdgeFunctionIncremental(ee0, sample2, edge0FuncAtBlockOrigin) &&
EvaluateEdgeFunctionIncremental(ee1, sample2, edge1FuncAtBlockOrigin) &&
EvaluateEdgeFunctionIncremental(ee2, sample2, edge2FuncAtBlockOrigin);
bool inside3 =
EvaluateEdgeFunctionIncremental(ee0, sample3, edge0FuncAtBlockOrigin) &&
EvaluateEdgeFunctionIncremental(ee1, sample3, edge1FuncAtBlockOrigin) &&
EvaluateEdgeFunctionIncremental(ee2, sample3, edge2FuncAtBlockOrigin);
if (inside0) debugMaskScalar |= g_scQuadMask0;
if (inside1) debugMaskScalar |= g_scQuadMask1;
if (inside2) debugMaskScalar |= g_scQuadMask2;
if (inside3) debugMaskScalar |= g_scQuadMask3;
}
#endif
// Store X positions of 4 consecutive samples
__m128 sseX4 = _mm_setr_ps(
g_scSIMDWidth * px + 0.5f,
g_scSIMDWidth * px + 1.5f,
g_scSIMDWidth * px + 2.5f,
g_scSIMDWidth * px + 3.5f);
// a * s
__m128 sseEdge0TermA = _mm_mul_ps(sseEdge0A4, sseX4);
__m128 sseEdge1TermA = _mm_mul_ps(sseEdge1A4, sseX4);
__m128 sseEdge2TermA = _mm_mul_ps(sseEdge2A4, sseX4);
// b * t
__m128 sseEdge0TermB = _mm_mul_ps(sseEdge0B4, sseY4);
__m128 sseEdge1TermB = _mm_mul_ps(sseEdge1B4, sseY4);
__m128 sseEdge2TermB = _mm_mul_ps(sseEdge2B4, sseY4);
// E(x+s, y+t) = E(x,y) + a*s + t*b
__m128 sseEdgeFunc0 = _mm_add_ps(sseEdge0FuncAtBlockOrigin, _mm_add_ps(sseEdge0TermA, sseEdge0TermB));
__m128 sseEdgeFunc1 = _mm_add_ps(sseEdge1FuncAtBlockOrigin, _mm_add_ps(sseEdge1TermA, sseEdge1TermB));
__m128 sseEdgeFunc2 = _mm_add_ps(sseEdge2FuncAtBlockOrigin, _mm_add_ps(sseEdge2TermA, sseEdge2TermB));
#ifdef EDGE_TEST_SHARED_EDGES
//E(x, y):
// E(x, y) > 0