projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
swr/rast: allow early-z if shader uses depth value
[mesa.git] / src / gallium / drivers / swr / rasterizer / core / clip.h
1 /****************************************************************************
2 * Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved.
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 *
23 * @file clip.h
24 *
25 * @brief Definitions for clipping
26 *
27 ******************************************************************************/
28 #pragma once
29
30 #include "common/simdintrin.h"
31 #include "core/context.h"
32 #include "core/pa.h"
33 #include "rdtsc_core.h"
34
35 // Temp storage used by the clipper
36 extern THREAD simdvertex tlsTempVertices[7];
37 #if USE_SIMD16_FRONTEND
38 extern THREAD simd16vertex tlsTempVertices_simd16[7];
39 #endif
40
41 enum SWR_CLIPCODES
42 {
43 // Shift clip codes out of the mantissa to prevent denormalized values when used in float compare.
44 // Guardband is able to use a single high-bit with 4 separate LSBs, because it computes a union, rather than intersection, of clipcodes.
45 #define CLIPCODE_SHIFT 23
46 FRUSTUM_LEFT = (0x01 << CLIPCODE_SHIFT),
47 FRUSTUM_TOP = (0x02 << CLIPCODE_SHIFT),
48 FRUSTUM_RIGHT = (0x04 << CLIPCODE_SHIFT),
49 FRUSTUM_BOTTOM = (0x08 << CLIPCODE_SHIFT),
50
51 FRUSTUM_NEAR = (0x10 << CLIPCODE_SHIFT),
52 FRUSTUM_FAR = (0x20 << CLIPCODE_SHIFT),
53
54 NEGW = (0x40 << CLIPCODE_SHIFT),
55
56 GUARDBAND_LEFT = (0x80 << CLIPCODE_SHIFT | 0x1),
57 GUARDBAND_TOP = (0x80 << CLIPCODE_SHIFT | 0x2),
58 GUARDBAND_RIGHT = (0x80 << CLIPCODE_SHIFT | 0x4),
59 GUARDBAND_BOTTOM = (0x80 << CLIPCODE_SHIFT | 0x8)
60 };
61
62 #define GUARDBAND_CLIP_MASK (FRUSTUM_NEAR|FRUSTUM_FAR|GUARDBAND_LEFT|GUARDBAND_TOP|GUARDBAND_RIGHT|GUARDBAND_BOTTOM|NEGW)
63
64 INLINE
65 void ComputeClipCodes(const API_STATE& state, const simdvector& vertex, simdscalar& clipCodes, simdscalari viewportIndexes)
66 {
67 clipCodes = _simd_setzero_ps();
68
69 // -w
70 simdscalar vNegW = _simd_mul_ps(vertex.w, _simd_set1_ps(-1.0f));
71
72 // FRUSTUM_LEFT
73 simdscalar vRes = _simd_cmplt_ps(vertex.x, vNegW);
74 clipCodes = _simd_and_ps(vRes, _simd_castsi_ps(_simd_set1_epi32(FRUSTUM_LEFT)));
75
76 // FRUSTUM_TOP
77 vRes = _simd_cmplt_ps(vertex.y, vNegW);
78 clipCodes = _simd_or_ps(clipCodes, _simd_and_ps(vRes, _simd_castsi_ps(_simd_set1_epi32(FRUSTUM_TOP))));
79
80 // FRUSTUM_RIGHT
81 vRes = _simd_cmpgt_ps(vertex.x, vertex.w);
82 clipCodes = _simd_or_ps(clipCodes, _simd_and_ps(vRes, _simd_castsi_ps(_simd_set1_epi32(FRUSTUM_RIGHT))));
83
84 // FRUSTUM_BOTTOM
85 vRes = _simd_cmpgt_ps(vertex.y, vertex.w);
86 clipCodes = _simd_or_ps(clipCodes, _simd_and_ps(vRes, _simd_castsi_ps(_simd_set1_epi32(FRUSTUM_BOTTOM))));
87
88 if (state.rastState.depthClipEnable)
89 {
90 // FRUSTUM_NEAR
91 // DX clips depth [0..w], GL clips [-w..w]
92 if (state.rastState.clipHalfZ)
93 {
94 vRes = _simd_cmplt_ps(vertex.z, _simd_setzero_ps());
95 }
96 else
97 {
98 vRes = _simd_cmplt_ps(vertex.z, vNegW);
99 }
100 clipCodes = _simd_or_ps(clipCodes, _simd_and_ps(vRes, _simd_castsi_ps(_simd_set1_epi32(FRUSTUM_NEAR))));
101
102 // FRUSTUM_FAR
103 vRes = _simd_cmpgt_ps(vertex.z, vertex.w);
104 clipCodes = _simd_or_ps(clipCodes, _simd_and_ps(vRes, _simd_castsi_ps(_simd_set1_epi32(FRUSTUM_FAR))));
105 }
106
107 // NEGW
108 vRes = _simd_cmple_ps(vertex.w, _simd_setzero_ps());
109 clipCodes = _simd_or_ps(clipCodes, _simd_and_ps(vRes, _simd_castsi_ps(_simd_set1_epi32(NEGW))));
110
111 // GUARDBAND_LEFT
112 simdscalar gbMult = _simd_mul_ps(vNegW, _simd_i32gather_ps(&state.gbState.left[0], viewportIndexes, 4));
113 vRes = _simd_cmplt_ps(vertex.x, gbMult);
114 clipCodes = _simd_or_ps(clipCodes, _simd_and_ps(vRes, _simd_castsi_ps(_simd_set1_epi32(GUARDBAND_LEFT))));
115
116 // GUARDBAND_TOP
117 gbMult = _simd_mul_ps(vNegW, _simd_i32gather_ps(&state.gbState.top[0], viewportIndexes, 4));
118 vRes = _simd_cmplt_ps(vertex.y, gbMult);
119 clipCodes = _simd_or_ps(clipCodes, _simd_and_ps(vRes, _simd_castsi_ps(_simd_set1_epi32(GUARDBAND_TOP))));
120
121 // GUARDBAND_RIGHT
122 gbMult = _simd_mul_ps(vertex.w, _simd_i32gather_ps(&state.gbState.right[0], viewportIndexes, 4));
123 vRes = _simd_cmpgt_ps(vertex.x, gbMult);
124 clipCodes = _simd_or_ps(clipCodes, _simd_and_ps(vRes, _simd_castsi_ps(_simd_set1_epi32(GUARDBAND_RIGHT))));
125
126 // GUARDBAND_BOTTOM
127 gbMult = _simd_mul_ps(vertex.w, _simd_i32gather_ps(&state.gbState.bottom[0], viewportIndexes, 4));
128 vRes = _simd_cmpgt_ps(vertex.y, gbMult);
129 clipCodes = _simd_or_ps(clipCodes, _simd_and_ps(vRes, _simd_castsi_ps(_simd_set1_epi32(GUARDBAND_BOTTOM))));
130 }
131
132 #if USE_SIMD16_FRONTEND
133 INLINE
134 void ComputeClipCodes(const API_STATE& state, const simd16vector& vertex, simd16scalar& clipCodes, simd16scalari viewportIndexes)
135 {
136 clipCodes = _simd16_setzero_ps();
137
138 // -w
139 simd16scalar vNegW = _simd16_mul_ps(vertex.w, _simd16_set1_ps(-1.0f));
140
141 // FRUSTUM_LEFT
142 simd16scalar vRes = _simd16_cmplt_ps(vertex.x, vNegW);
143 clipCodes = _simd16_and_ps(vRes, _simd16_castsi_ps(_simd16_set1_epi32(FRUSTUM_LEFT)));
144
145 // FRUSTUM_TOP
146 vRes = _simd16_cmplt_ps(vertex.y, vNegW);
147 clipCodes = _simd16_or_ps(clipCodes, _simd16_and_ps(vRes, _simd16_castsi_ps(_simd16_set1_epi32(FRUSTUM_TOP))));
148
149 // FRUSTUM_RIGHT
150 vRes = _simd16_cmpgt_ps(vertex.x, vertex.w);
151 clipCodes = _simd16_or_ps(clipCodes, _simd16_and_ps(vRes, _simd16_castsi_ps(_simd16_set1_epi32(FRUSTUM_RIGHT))));
152
153 // FRUSTUM_BOTTOM
154 vRes = _simd16_cmpgt_ps(vertex.y, vertex.w);
155 clipCodes = _simd16_or_ps(clipCodes, _simd16_and_ps(vRes, _simd16_castsi_ps(_simd16_set1_epi32(FRUSTUM_BOTTOM))));
156
157 if (state.rastState.depthClipEnable)
158 {
159 // FRUSTUM_NEAR
160 // DX clips depth [0..w], GL clips [-w..w]
161 if (state.rastState.clipHalfZ)
162 {
163 vRes = _simd16_cmplt_ps(vertex.z, _simd16_setzero_ps());
164 }
165 else
166 {
167 vRes = _simd16_cmplt_ps(vertex.z, vNegW);
168 }
169 clipCodes = _simd16_or_ps(clipCodes, _simd16_and_ps(vRes, _simd16_castsi_ps(_simd16_set1_epi32(FRUSTUM_NEAR))));
170
171 // FRUSTUM_FAR
172 vRes = _simd16_cmpgt_ps(vertex.z, vertex.w);
173 clipCodes = _simd16_or_ps(clipCodes, _simd16_and_ps(vRes, _simd16_castsi_ps(_simd16_set1_epi32(FRUSTUM_FAR))));
174 }
175
176 // NEGW
177 vRes = _simd16_cmple_ps(vertex.w, _simd16_setzero_ps());
178 clipCodes = _simd16_or_ps(clipCodes, _simd16_and_ps(vRes, _simd16_castsi_ps(_simd16_set1_epi32(NEGW))));
179
180 // GUARDBAND_LEFT
181 simd16scalar gbMult = _simd16_mul_ps(vNegW, _simd16_i32gather_ps(&state.gbState.left[0], viewportIndexes, 4));
182 vRes = _simd16_cmplt_ps(vertex.x, gbMult);
183 clipCodes = _simd16_or_ps(clipCodes, _simd16_and_ps(vRes, _simd16_castsi_ps(_simd16_set1_epi32(GUARDBAND_LEFT))));
184
185 // GUARDBAND_TOP
186 gbMult = _simd16_mul_ps(vNegW, _simd16_i32gather_ps(&state.gbState.top[0], viewportIndexes, 4));
187 vRes = _simd16_cmplt_ps(vertex.y, gbMult);
188 clipCodes = _simd16_or_ps(clipCodes, _simd16_and_ps(vRes, _simd16_castsi_ps(_simd16_set1_epi32(GUARDBAND_TOP))));
189
190 // GUARDBAND_RIGHT
191 gbMult = _simd16_mul_ps(vertex.w, _simd16_i32gather_ps(&state.gbState.right[0], viewportIndexes, 4));
192 vRes = _simd16_cmpgt_ps(vertex.x, gbMult);
193 clipCodes = _simd16_or_ps(clipCodes, _simd16_and_ps(vRes, _simd16_castsi_ps(_simd16_set1_epi32(GUARDBAND_RIGHT))));
194
195 // GUARDBAND_BOTTOM
196 gbMult = _simd16_mul_ps(vertex.w, _simd16_i32gather_ps(&state.gbState.bottom[0], viewportIndexes, 4));
197 vRes = _simd16_cmpgt_ps(vertex.y, gbMult);
198 clipCodes = _simd16_or_ps(clipCodes, _simd16_and_ps(vRes, _simd16_castsi_ps(_simd16_set1_epi32(GUARDBAND_BOTTOM))));
199 }
200
201 #endif
202 template<uint32_t NumVertsPerPrim>
203 class Clipper
204 {
205 public:
206 Clipper(uint32_t in_workerId, DRAW_CONTEXT* in_pDC) :
207 workerId(in_workerId), pDC(in_pDC), state(GetApiState(in_pDC))
208 {
209 static_assert(NumVertsPerPrim >= 1 && NumVertsPerPrim <= 3, "Invalid NumVertsPerPrim");
210 }
211
212 void ComputeClipCodes(simdvector vertex[], simdscalari viewportIndexes)
213 {
214 for (uint32_t i = 0; i < NumVertsPerPrim; ++i)
215 {
216 ::ComputeClipCodes(this->state, vertex[i], this->clipCodes[i], viewportIndexes);
217 }
218 }
219
220 #if USE_SIMD16_FRONTEND
221 void ComputeClipCodes(simd16vector vertex[], simd16scalari viewportIndexes)
222 {
223 for (uint32_t i = 0; i < NumVertsPerPrim; ++i)
224 {
225 ::ComputeClipCodes(this->state, vertex[i], this->clipCodes_simd16[i], viewportIndexes);
226 }
227 }
228
229 #endif
230 simdscalar ComputeClipCodeIntersection()
231 {
232 simdscalar result = this->clipCodes[0];
233 for (uint32_t i = 1; i < NumVertsPerPrim; ++i)
234 {
235 result = _simd_and_ps(result, this->clipCodes[i]);
236 }
237 return result;
238 }
239
240 #if USE_SIMD16_FRONTEND
241 simd16scalar ComputeClipCodeIntersection_simd16()
242 {
243 simd16scalar result = this->clipCodes_simd16[0];
244 for (uint32_t i = 1; i < NumVertsPerPrim; ++i)
245 {
246 result = _simd16_and_ps(result, this->clipCodes_simd16[i]);
247 }
248 return result;
249 }
250
251 #endif
252 simdscalar ComputeClipCodeUnion()
253 {
254 simdscalar result = this->clipCodes[0];
255 for (uint32_t i = 1; i < NumVertsPerPrim; ++i)
256 {
257 result = _simd_or_ps(result, this->clipCodes[i]);
258 }
259 return result;
260 }
261
262 #if USE_SIMD16_FRONTEND
263 simd16scalar ComputeClipCodeUnion_simd16()
264 {
265 simd16scalar result = this->clipCodes_simd16[0];
266 for (uint32_t i = 1; i < NumVertsPerPrim; ++i)
267 {
268 result = _simd16_or_ps(result, this->clipCodes_simd16[i]);
269 }
270 return result;
271 }
272
273 #endif
274 int ComputeNegWMask()
275 {
276 simdscalar clipCodeUnion = ComputeClipCodeUnion();
277 clipCodeUnion = _simd_and_ps(clipCodeUnion, _simd_castsi_ps(_simd_set1_epi32(NEGW)));
278 return _simd_movemask_ps(_simd_cmpneq_ps(clipCodeUnion, _simd_setzero_ps()));
279 }
280
281 int ComputeClipMask()
282 {
283 simdscalar clipUnion = ComputeClipCodeUnion();
284 clipUnion = _simd_and_ps(clipUnion, _simd_castsi_ps(_simd_set1_epi32(GUARDBAND_CLIP_MASK)));
285 return _simd_movemask_ps(_simd_cmpneq_ps(clipUnion, _simd_setzero_ps()));
286 }
287
288 #if USE_SIMD16_FRONTEND
289 int ComputeClipMask_simd16()
290 {
291 simd16scalar clipUnion = ComputeClipCodeUnion_simd16();
292 clipUnion = _simd16_and_ps(clipUnion, _simd16_castsi_ps(_simd16_set1_epi32(GUARDBAND_CLIP_MASK)));
293 return _simd16_movemask_ps(_simd16_cmpneq_ps(clipUnion, _simd16_setzero_ps()));
294 }
295
296 #endif
297 // clipper is responsible for culling any prims with NAN coordinates
298 int ComputeNaNMask(simdvector prim[])
299 {
300 simdscalar vNanMask = _simd_setzero_ps();
301 for (uint32_t e = 0; e < NumVertsPerPrim; ++e)
302 {
303 simdscalar vNan01 = _simd_cmp_ps(prim[e].v[0], prim[e].v[1], _CMP_UNORD_Q);
304 vNanMask = _simd_or_ps(vNanMask, vNan01);
305 simdscalar vNan23 = _simd_cmp_ps(prim[e].v[2], prim[e].v[3], _CMP_UNORD_Q);
306 vNanMask = _simd_or_ps(vNanMask, vNan23);
307 }
308
309 return _simd_movemask_ps(vNanMask);
310 }
311
312 #if USE_SIMD16_FRONTEND
313 int ComputeNaNMask(simd16vector prim[])
314 {
315 simd16scalar vNanMask = _simd16_setzero_ps();
316 for (uint32_t e = 0; e < NumVertsPerPrim; ++e)
317 {
318 simd16scalar vNan01 = _simd16_cmp_ps(prim[e].v[0], prim[e].v[1], _CMP_UNORD_Q);
319 vNanMask = _simd16_or_ps(vNanMask, vNan01);
320 simd16scalar vNan23 = _simd16_cmp_ps(prim[e].v[2], prim[e].v[3], _CMP_UNORD_Q);
321 vNanMask = _simd16_or_ps(vNanMask, vNan23);
322 }
323
324 return _simd16_movemask_ps(vNanMask);
325 }
326
327 #endif
328 int ComputeUserClipCullMask(PA_STATE& pa, simdvector prim[])
329 {
330 uint8_t cullMask = this->state.rastState.cullDistanceMask;
331 simdscalar vClipCullMask = _simd_setzero_ps();
332 DWORD index;
333
334 simdvector vClipCullDistLo[3];
335 simdvector vClipCullDistHi[3];
336
337 pa.Assemble(VERTEX_CLIPCULL_DIST_LO_SLOT, vClipCullDistLo);
338 pa.Assemble(VERTEX_CLIPCULL_DIST_HI_SLOT, vClipCullDistHi);
339 while (_BitScanForward(&index, cullMask))
340 {
341 cullMask &= ~(1 << index);
342 uint32_t slot = index >> 2;
343 uint32_t component = index & 0x3;
344
345 simdscalar vCullMaskElem = _simd_set1_ps(-1.0f);
346 for (uint32_t e = 0; e < NumVertsPerPrim; ++e)
347 {
348 simdscalar vCullComp;
349 if (slot == 0)
350 {
351 vCullComp = vClipCullDistLo[e][component];
352 }
353 else
354 {
355 vCullComp = vClipCullDistHi[e][component];
356 }
357
358 // cull if cull distance < 0 || NAN
359 simdscalar vCull = _simd_cmp_ps(_mm256_setzero_ps(), vCullComp, _CMP_NLE_UQ);
360 vCullMaskElem = _simd_and_ps(vCullMaskElem, vCull);
361 }
362 vClipCullMask = _simd_or_ps(vClipCullMask, vCullMaskElem);
363 }
364
365 // clipper should also discard any primitive with NAN clip distance
366 uint8_t clipMask = this->state.rastState.clipDistanceMask;
367 while (_BitScanForward(&index, clipMask))
368 {
369 clipMask &= ~(1 << index);
370 uint32_t slot = index >> 2;
371 uint32_t component = index & 0x3;
372
373 for (uint32_t e = 0; e < NumVertsPerPrim; ++e)
374 {
375 simdscalar vClipComp;
376 if (slot == 0)
377 {
378 vClipComp = vClipCullDistLo[e][component];
379 }
380 else
381 {
382 vClipComp = vClipCullDistHi[e][component];
383 }
384
385 simdscalar vClip = _simd_cmp_ps(vClipComp, vClipComp, _CMP_UNORD_Q);
386 vClipCullMask = _simd_or_ps(vClipCullMask, vClip);
387 }
388 }
389
390 return _simd_movemask_ps(vClipCullMask);
391 }
392
393 #if USE_SIMD16_FRONTEND
394 int ComputeUserClipCullMask(PA_STATE& pa, simd16vector prim[])
395 {
396 uint8_t cullMask = this->state.rastState.cullDistanceMask;
397 simd16scalar vClipCullMask = _simd16_setzero_ps();
398
399 simd16vector vClipCullDistLo[3];
400 simd16vector vClipCullDistHi[3];
401
402 pa.Assemble_simd16(VERTEX_CLIPCULL_DIST_LO_SLOT, vClipCullDistLo);
403 pa.Assemble_simd16(VERTEX_CLIPCULL_DIST_HI_SLOT, vClipCullDistHi);
404
405 DWORD index;
406 while (_BitScanForward(&index, cullMask))
407 {
408 cullMask &= ~(1 << index);
409 uint32_t slot = index >> 2;
410 uint32_t component = index & 0x3;
411
412 simd16scalar vCullMaskElem = _simd16_set1_ps(-1.0f);
413 for (uint32_t e = 0; e < NumVertsPerPrim; ++e)
414 {
415 simd16scalar vCullComp;
416 if (slot == 0)
417 {
418 vCullComp = vClipCullDistLo[e][component];
419 }
420 else
421 {
422 vCullComp = vClipCullDistHi[e][component];
423 }
424
425 // cull if cull distance < 0 || NAN
426 simd16scalar vCull = _simd16_cmp_ps(_simd16_setzero_ps(), vCullComp, _CMP_NLE_UQ);
427 vCullMaskElem = _simd16_and_ps(vCullMaskElem, vCull);
428 }
429 vClipCullMask = _simd16_or_ps(vClipCullMask, vCullMaskElem);
430 }
431
432 // clipper should also discard any primitive with NAN clip distance
433 uint8_t clipMask = this->state.rastState.clipDistanceMask;
434 while (_BitScanForward(&index, clipMask))
435 {
436 clipMask &= ~(1 << index);
437 uint32_t slot = index >> 2;
438 uint32_t component = index & 0x3;
439
440 for (uint32_t e = 0; e < NumVertsPerPrim; ++e)
441 {
442 simd16scalar vClipComp;
443 if (slot == 0)
444 {
445 vClipComp = vClipCullDistLo[e][component];
446 }
447 else
448 {
449 vClipComp = vClipCullDistHi[e][component];
450 }
451
452 simd16scalar vClip = _simd16_cmp_ps(vClipComp, vClipComp, _CMP_UNORD_Q);
453 vClipCullMask = _simd16_or_ps(vClipCullMask, vClip);
454 }
455 }
456
457 return _simd16_movemask_ps(vClipCullMask);
458 }
459
460 #endif
461 // clip SIMD primitives
462 void ClipSimd(const simdscalar& vPrimMask, const simdscalar& vClipMask, PA_STATE& pa, const simdscalari& vPrimId, const simdscalari& vViewportIdx)
463 {
464 // input/output vertex store for clipper
465 simdvertex vertices[7]; // maximum 7 verts generated per triangle
466
467 LONG constantInterpMask = this->state.backendState.constantInterpolationMask;
468 uint32_t provokingVertex = 0;
469 if(pa.binTopology == TOP_TRIANGLE_FAN)
470 {
471 provokingVertex = this->state.frontendState.provokingVertex.triFan;
472 }
473 ///@todo: line topology for wireframe?
474
475 // assemble pos
476 simdvector tmpVector[NumVertsPerPrim];
477 pa.Assemble(VERTEX_POSITION_SLOT, tmpVector);
478 for (uint32_t i = 0; i < NumVertsPerPrim; ++i)
479 {
480 vertices[i].attrib[VERTEX_POSITION_SLOT] = tmpVector[i];
481 }
482
483 // assemble attribs
484 const SWR_BACKEND_STATE& backendState = this->state.backendState;
485
486 int32_t maxSlot = -1;
487 for (uint32_t slot = 0; slot < backendState.numAttributes; ++slot)
488 {
489 // Compute absolute attrib slot in vertex array
490 uint32_t mapSlot = backendState.swizzleEnable ? backendState.swizzleMap[slot].sourceAttrib : slot;
491 maxSlot = std::max<int32_t>(maxSlot, mapSlot);
492 uint32_t inputSlot = VERTEX_ATTRIB_START_SLOT + mapSlot;
493
494 pa.Assemble(inputSlot, tmpVector);
495
496 // if constant interpolation enabled for this attribute, assign the provoking
497 // vertex values to all edges
498 if (_bittest(&constantInterpMask, slot))
499 {
500 for (uint32_t i = 0; i < NumVertsPerPrim; ++i)
501 {
502 vertices[i].attrib[inputSlot] = tmpVector[provokingVertex];
503 }
504 }
505 else
506 {
507 for (uint32_t i = 0; i < NumVertsPerPrim; ++i)
508 {
509 vertices[i].attrib[inputSlot] = tmpVector[i];
510 }
511 }
512 }
513
514 // assemble user clip distances if enabled
515 if (this->state.rastState.clipDistanceMask & 0xf)
516 {
517 pa.Assemble(VERTEX_CLIPCULL_DIST_LO_SLOT, tmpVector);
518 for (uint32_t i = 0; i < NumVertsPerPrim; ++i)
519 {
520 vertices[i].attrib[VERTEX_CLIPCULL_DIST_LO_SLOT] = tmpVector[i];
521 }
522 }
523
524 if (this->state.rastState.clipDistanceMask & 0xf0)
525 {
526 pa.Assemble(VERTEX_CLIPCULL_DIST_HI_SLOT, tmpVector);
527 for (uint32_t i = 0; i < NumVertsPerPrim; ++i)
528 {
529 vertices[i].attrib[VERTEX_CLIPCULL_DIST_HI_SLOT] = tmpVector[i];
530 }
531 }
532
533 uint32_t numAttribs = maxSlot + 1;
534
535 simdscalari vNumClippedVerts = ClipPrims((float*)&vertices[0], vPrimMask, vClipMask, numAttribs);
536
537 // set up new PA for binning clipped primitives
538 PFN_PROCESS_PRIMS pfnBinFunc = nullptr;
539 PRIMITIVE_TOPOLOGY clipTopology = TOP_UNKNOWN;
540 if (NumVertsPerPrim == 3)
541 {
542 pfnBinFunc = GetBinTrianglesFunc((pa.pDC->pState->state.rastState.conservativeRast > 0));
543 clipTopology = TOP_TRIANGLE_FAN;
544
545 // so that the binner knows to bloat wide points later
546 if (pa.binTopology == TOP_POINT_LIST)
547 clipTopology = TOP_POINT_LIST;
548
549 }
550 else if (NumVertsPerPrim == 2)
551 {
552 pfnBinFunc = BinLines;
553 clipTopology = TOP_LINE_LIST;
554 }
555 else
556 {
557 SWR_ASSERT(0 && "Unexpected points in clipper.");
558 }
559
560 uint32_t* pVertexCount = (uint32_t*)&vNumClippedVerts;
561 uint32_t* pPrimitiveId = (uint32_t*)&vPrimId;
562 uint32_t* pViewportIdx = (uint32_t*)&vViewportIdx;
563
564 const simdscalari vOffsets = _mm256_set_epi32(
565 0 * sizeof(simdvertex), // unused lane
566 6 * sizeof(simdvertex),
567 5 * sizeof(simdvertex),
568 4 * sizeof(simdvertex),
569 3 * sizeof(simdvertex),
570 2 * sizeof(simdvertex),
571 1 * sizeof(simdvertex),
572 0 * sizeof(simdvertex));
573
574 // only need to gather 7 verts
575 // @todo dynamic mask based on actual # of verts generated per lane
576 const simdscalar vMask = _mm256_set_ps(0, -1, -1, -1, -1, -1, -1, -1);
577
578 uint32_t numClippedPrims = 0;
579 #if USE_SIMD16_FRONTEND
580 const uint32_t numPrims = pa.NumPrims();
581 const uint32_t numPrims_lo = std::min<uint32_t>(numPrims, KNOB_SIMD_WIDTH);
582
583 SWR_ASSERT(numPrims <= numPrims_lo);
584
585 for (uint32_t inputPrim = 0; inputPrim < numPrims_lo; ++inputPrim)
586 #else
587 for (uint32_t inputPrim = 0; inputPrim < pa.NumPrims(); ++inputPrim)
588 #endif
589 {
590 uint32_t numEmittedVerts = pVertexCount[inputPrim];
591 if (numEmittedVerts < NumVertsPerPrim)
592 {
593 continue;
594 }
595 SWR_ASSERT(numEmittedVerts <= 7, "Unexpected vertex count from clipper.");
596
597 uint32_t numEmittedPrims = GetNumPrims(clipTopology, numEmittedVerts);
598 numClippedPrims += numEmittedPrims;
599
600 // tranpose clipper output so that each lane's vertices are in SIMD order
601 // set aside space for 2 vertices, as the PA will try to read up to 16 verts
602 // for triangle fan
603 #if USE_SIMD16_FRONTEND
604 simd16vertex transposedPrims[2];
605 #else
606 simdvertex transposedPrims[2];
607 #endif
608
609 // transpose pos
610 uint8_t* pBase = (uint8_t*)(&vertices[0].attrib[VERTEX_POSITION_SLOT]) + sizeof(float) * inputPrim;
611
612 #if USE_SIMD16_FRONTEND
613 // TEMPORARY WORKAROUND for bizarre VS2015 code-gen bug
614 static const float *dummy = reinterpret_cast<const float *>(pBase);
615 #endif
616
617 for (uint32_t c = 0; c < 4; ++c)
618 {
619 #if USE_SIMD16_FRONTEND
620 simdscalar temp = _simd_mask_i32gather_ps(_simd_setzero_ps(), (const float *)pBase, vOffsets, vMask, 1);
621 transposedPrims[0].attrib[VERTEX_POSITION_SLOT][c] = _simd16_insert_ps(_simd16_setzero_ps(), temp, 0);
622 #else
623 transposedPrims[0].attrib[VERTEX_POSITION_SLOT][c] = _simd_mask_i32gather_ps(_mm256_undefined_ps(), (const float*)pBase, vOffsets, vMask, 1);
624 #endif
625 pBase += sizeof(simdscalar);
626 }
627
628 // transpose attribs
629 pBase = (uint8_t*)(&vertices[0].attrib[VERTEX_ATTRIB_START_SLOT]) + sizeof(float) * inputPrim;
630 for (uint32_t attrib = 0; attrib < numAttribs; ++attrib)
631 {
632 uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + attrib;
633 for (uint32_t c = 0; c < 4; ++c)
634 {
635 #if USE_SIMD16_FRONTEND
636 simdscalar temp = _simd_mask_i32gather_ps(_simd_setzero_ps(), (const float *)pBase, vOffsets, vMask, 1);
637 transposedPrims[0].attrib[attribSlot][c] = _simd16_insert_ps(_simd16_setzero_ps(), temp, 0);
638 #else
639 transposedPrims[0].attrib[attribSlot][c] = _simd_mask_i32gather_ps(_mm256_undefined_ps(), (const float*)pBase, vOffsets, vMask, 1);
640 #endif
641 pBase += sizeof(simdscalar);
642 }
643 }
644
645 // transpose user clip distances if enabled
646 if (this->state.rastState.clipDistanceMask & 0xf)
647 {
648 pBase = (uint8_t*)(&vertices[0].attrib[VERTEX_CLIPCULL_DIST_LO_SLOT]) + sizeof(float) * inputPrim;
649 for (uint32_t c = 0; c < 4; ++c)
650 {
651 #if USE_SIMD16_FRONTEND
652 simdscalar temp = _simd_mask_i32gather_ps(_simd_setzero_ps(), (const float *)pBase, vOffsets, vMask, 1);
653 transposedPrims[0].attrib[VERTEX_CLIPCULL_DIST_LO_SLOT][c] = _simd16_insert_ps(_simd16_setzero_ps(), temp, 0);
654 #else
655 transposedPrims[0].attrib[VERTEX_CLIPCULL_DIST_LO_SLOT][c] = _simd_mask_i32gather_ps(_mm256_undefined_ps(), (const float*)pBase, vOffsets, vMask, 1);
656 #endif
657 pBase += sizeof(simdscalar);
658 }
659 }
660
661 if (this->state.rastState.clipDistanceMask & 0xf0)
662 {
663 pBase = (uint8_t*)(&vertices[0].attrib[VERTEX_CLIPCULL_DIST_HI_SLOT]) + sizeof(float) * inputPrim;
664 for (uint32_t c = 0; c < 4; ++c)
665 {
666 #if USE_SIMD16_FRONTEND
667 simdscalar temp = _simd_mask_i32gather_ps(_simd_setzero_ps(), (const float *)pBase, vOffsets, vMask, 1);
668 transposedPrims[0].attrib[VERTEX_CLIPCULL_DIST_HI_SLOT][c] = _simd16_insert_ps(_simd16_setzero_ps(), temp, 0);
669 #else
670 transposedPrims[0].attrib[VERTEX_CLIPCULL_DIST_HI_SLOT][c] = _simd_mask_i32gather_ps(_mm256_undefined_ps(), (const float*)pBase, vOffsets, vMask, 1);
671 #endif
672 pBase += sizeof(simdscalar);
673 }
674 }
675
676 PA_STATE_OPT clipPa(this->pDC, numEmittedPrims, (uint8_t*)&transposedPrims[0], numEmittedVerts, true, clipTopology);
677
678 while (clipPa.GetNextStreamOutput())
679 {
680 do
681 {
682 #if USE_SIMD16_FRONTEND
683 simd16vector attrib_simd16[NumVertsPerPrim];
684 bool assemble = clipPa.Assemble_simd16(VERTEX_POSITION_SLOT, attrib_simd16);
685
686 if (assemble)
687 {
688 static const uint32_t primMaskMap[] = { 0x0, 0x1, 0x3, 0x7, 0xf, 0x1f, 0x3f, 0x7f, 0xff };
689
690 simdvector attrib[NumVertsPerPrim];
691 for (uint32_t i = 0; i < NumVertsPerPrim; i += 1)
692 {
693 for (uint32_t j = 0; j < 4; j += 1)
694 {
695 attrib[i][j] = _simd16_extract_ps(attrib_simd16[i][j], 0);
696 }
697 }
698
699 clipPa.useAlternateOffset = false;
700 pfnBinFunc(this->pDC, clipPa, this->workerId, attrib, primMaskMap[numEmittedPrims], _simd_set1_epi32(pPrimitiveId[inputPrim]), _simd_set1_epi32(pViewportIdx[inputPrim]));
701 }
702 #else
703 simdvector attrib[NumVertsPerPrim];
704 bool assemble = clipPa.Assemble(VERTEX_POSITION_SLOT, attrib);
705 if (assemble)
706 {
707 static const uint32_t primMaskMap[] = { 0x0, 0x1, 0x3, 0x7, 0xf, 0x1f, 0x3f, 0x7f, 0xff };
708 pfnBinFunc(this->pDC, clipPa, this->workerId, attrib, primMaskMap[numEmittedPrims], _simd_set1_epi32(pPrimitiveId[inputPrim]), _simd_set1_epi32(pViewportIdx[inputPrim]));
709 }
710 #endif
711 } while (clipPa.NextPrim());
712 }
713 }
714
715 // update global pipeline stat
716 UPDATE_STAT_FE(CPrimitives, numClippedPrims);
717 }
718
719 #if USE_SIMD16_FRONTEND
720 void ClipSimd(const simd16scalar& vPrimMask, const simd16scalar& vClipMask, PA_STATE& pa, const simd16scalari& vPrimId, const simd16scalari& vViewportIdx)
721 {
722 // input/output vertex store for clipper
723 simd16vertex vertices[7]; // maximum 7 verts generated per triangle
724
725 LONG constantInterpMask = this->state.backendState.constantInterpolationMask;
726 uint32_t provokingVertex = 0;
727 if (pa.binTopology == TOP_TRIANGLE_FAN)
728 {
729 provokingVertex = this->state.frontendState.provokingVertex.triFan;
730 }
731 ///@todo: line topology for wireframe?
732
733 // assemble pos
734 simd16vector tmpVector[NumVertsPerPrim];
735 pa.Assemble_simd16(VERTEX_POSITION_SLOT, tmpVector);
736 for (uint32_t i = 0; i < NumVertsPerPrim; ++i)
737 {
738 vertices[i].attrib[VERTEX_POSITION_SLOT] = tmpVector[i];
739 }
740
741 // assemble attribs
742 const SWR_BACKEND_STATE& backendState = this->state.backendState;
743
744 int32_t maxSlot = -1;
745 for (uint32_t slot = 0; slot < backendState.numAttributes; ++slot)
746 {
747 // Compute absolute attrib slot in vertex array
748 uint32_t mapSlot = backendState.swizzleEnable ? backendState.swizzleMap[slot].sourceAttrib : slot;
749 maxSlot = std::max<int32_t>(maxSlot, mapSlot);
750 uint32_t inputSlot = VERTEX_ATTRIB_START_SLOT + mapSlot;
751
752 pa.Assemble_simd16(inputSlot, tmpVector);
753
754 // if constant interpolation enabled for this attribute, assign the provoking
755 // vertex values to all edges
756 if (_bittest(&constantInterpMask, slot))
757 {
758 for (uint32_t i = 0; i < NumVertsPerPrim; ++i)
759 {
760 vertices[i].attrib[inputSlot] = tmpVector[provokingVertex];
761 }
762 }
763 else
764 {
765 for (uint32_t i = 0; i < NumVertsPerPrim; ++i)
766 {
767 vertices[i].attrib[inputSlot] = tmpVector[i];
768 }
769 }
770 }
771
772 // assemble user clip distances if enabled
773 if (this->state.rastState.clipDistanceMask & 0xf)
774 {
775 pa.Assemble_simd16(VERTEX_CLIPCULL_DIST_LO_SLOT, tmpVector);
776 for (uint32_t i = 0; i < NumVertsPerPrim; ++i)
777 {
778 vertices[i].attrib[VERTEX_CLIPCULL_DIST_LO_SLOT] = tmpVector[i];
779 }
780 }
781
782 if (this->state.rastState.clipDistanceMask & 0xf0)
783 {
784 pa.Assemble_simd16(VERTEX_CLIPCULL_DIST_HI_SLOT, tmpVector);
785 for (uint32_t i = 0; i < NumVertsPerPrim; ++i)
786 {
787 vertices[i].attrib[VERTEX_CLIPCULL_DIST_HI_SLOT] = tmpVector[i];
788 }
789 }
790
791 uint32_t numAttribs = maxSlot + 1;
792
793 simd16scalari vNumClippedVerts = ClipPrims((float*)&vertices[0], vPrimMask, vClipMask, numAttribs);
794
795 // set up new PA for binning clipped primitives
796 PFN_PROCESS_PRIMS_SIMD16 pfnBinFunc = nullptr;
797 PRIMITIVE_TOPOLOGY clipTopology = TOP_UNKNOWN;
798 if (NumVertsPerPrim == 3)
799 {
800 pfnBinFunc = GetBinTrianglesFunc_simd16((pa.pDC->pState->state.rastState.conservativeRast > 0));
801 clipTopology = TOP_TRIANGLE_FAN;
802
803 // so that the binner knows to bloat wide points later
804 if (pa.binTopology == TOP_POINT_LIST)
805 clipTopology = TOP_POINT_LIST;
806
807 }
808 else if (NumVertsPerPrim == 2)
809 {
810 pfnBinFunc = BinLines_simd16;
811 clipTopology = TOP_LINE_LIST;
812 }
813 else
814 {
815 SWR_ASSERT(0 && "Unexpected points in clipper.");
816 }
817
818 uint32_t* pVertexCount = (uint32_t*)&vNumClippedVerts;
819 uint32_t* pPrimitiveId = (uint32_t*)&vPrimId;
820 uint32_t* pViewportIdx = (uint32_t*)&vViewportIdx;
821
822 const simdscalari vOffsets = _simd_set_epi32(
823 0 * sizeof(simd16vertex), // unused lane
824 6 * sizeof(simd16vertex),
825 5 * sizeof(simd16vertex),
826 4 * sizeof(simd16vertex),
827 3 * sizeof(simd16vertex),
828 2 * sizeof(simd16vertex),
829 1 * sizeof(simd16vertex),
830 0 * sizeof(simd16vertex));
831
832 // only need to gather 7 verts
833 // @todo dynamic mask based on actual # of verts generated per lane
834 const simdscalar vMask = _mm256_set_ps(0, -1, -1, -1, -1, -1, -1, -1);
835
836 uint32_t numClippedPrims = 0;
837
838 // tranpose clipper output so that each lane's vertices are in SIMD order
839 // set aside space for 2 vertices, as the PA will try to read up to 16 verts
840 // for triangle fan
841
842 #if defined(_DEBUG)
843 // TODO: need to increase stack size, allocating SIMD16-widened transposedPrims causes stack overflow in debug builds
844 simd16vertex *transposedPrims = reinterpret_cast<simd16vertex *>(malloc(sizeof(simd16vertex) * 2));
845
846 #else
847 simd16vertex transposedPrims[2];
848
849 #endif
850 for (uint32_t inputPrim = 0; inputPrim < pa.NumPrims(); ++inputPrim)
851 {
852 uint32_t numEmittedVerts = pVertexCount[inputPrim];
853 if (numEmittedVerts < NumVertsPerPrim)
854 {
855 continue;
856 }
857 SWR_ASSERT(numEmittedVerts <= 7, "Unexpected vertex count from clipper.");
858
859 uint32_t numEmittedPrims = GetNumPrims(clipTopology, numEmittedVerts);
860 numClippedPrims += numEmittedPrims;
861
862 // tranpose clipper output so that each lane's vertices are in SIMD order
863 // set aside space for 2 vertices, as the PA will try to read up to 16 verts
864 // for triangle fan
865
866 // transpose pos
867 uint8_t* pBase = (uint8_t*)(&vertices[0].attrib[VERTEX_POSITION_SLOT]) + sizeof(float) * inputPrim;
868
869 #if 0
870 // TEMPORARY WORKAROUND for bizarre VS2015 code-gen bug
871 static const float *dummy = reinterpret_cast<const float *>(pBase);
872 #endif
873
874 for (uint32_t c = 0; c < 4; ++c)
875 {
876 simdscalar temp = _simd_mask_i32gather_ps(_simd_setzero_ps(), (const float *)pBase, vOffsets, vMask, 1);
877 transposedPrims[0].attrib[VERTEX_POSITION_SLOT][c] = _simd16_insert_ps(_simd16_setzero_ps(), temp, 0);
878 pBase += sizeof(simd16scalar);
879 }
880
881 // transpose attribs
882 pBase = (uint8_t*)(&vertices[0].attrib[VERTEX_ATTRIB_START_SLOT]) + sizeof(float) * inputPrim;
883 for (uint32_t attrib = 0; attrib < numAttribs; ++attrib)
884 {
885 uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + attrib;
886 for (uint32_t c = 0; c < 4; ++c)
887 {
888 simdscalar temp = _simd_mask_i32gather_ps(_simd_setzero_ps(), (const float *)pBase, vOffsets, vMask, 1);
889 transposedPrims[0].attrib[attribSlot][c] = _simd16_insert_ps(_simd16_setzero_ps(), temp, 0);
890 pBase += sizeof(simd16scalar);
891 }
892 }
893
894 // transpose user clip distances if enabled
895 if (this->state.rastState.clipDistanceMask & 0xf)
896 {
897 pBase = (uint8_t*)(&vertices[0].attrib[VERTEX_CLIPCULL_DIST_LO_SLOT]) + sizeof(float) * inputPrim;
898 for (uint32_t c = 0; c < 4; ++c)
899 {
900 simdscalar temp = _simd_mask_i32gather_ps(_simd_setzero_ps(), (const float *)pBase, vOffsets, vMask, 1);
901 transposedPrims[0].attrib[VERTEX_CLIPCULL_DIST_LO_SLOT][c] = _simd16_insert_ps(_simd16_setzero_ps(), temp, 0);
902 pBase += sizeof(simd16scalar);
903 }
904 }
905
906 if (this->state.rastState.clipDistanceMask & 0xf0)
907 {
908 pBase = (uint8_t*)(&vertices[0].attrib[VERTEX_CLIPCULL_DIST_HI_SLOT]) + sizeof(float) * inputPrim;
909 for (uint32_t c = 0; c < 4; ++c)
910 {
911 simdscalar temp = _simd_mask_i32gather_ps(_simd_setzero_ps(), (const float *)pBase, vOffsets, vMask, 1);
912 transposedPrims[0].attrib[VERTEX_CLIPCULL_DIST_HI_SLOT][c] = _simd16_insert_ps(_simd16_setzero_ps(), temp, 0);
913 pBase += sizeof(simd16scalar);
914 }
915 }
916
917 PA_STATE_OPT clipPa(this->pDC, numEmittedPrims, (uint8_t*)&transposedPrims[0], numEmittedVerts, true, clipTopology);
918
919 while (clipPa.GetNextStreamOutput())
920 {
921 do
922 {
923 simd16vector attrib[NumVertsPerPrim];
924 bool assemble = clipPa.Assemble_simd16(VERTEX_POSITION_SLOT, attrib);
925
926 if (assemble)
927 {
928 static const uint32_t primMaskMap[] = { 0x0, 0x1, 0x3, 0x7, 0xf, 0x1f, 0x3f, 0x7f, 0xff, 0x1ff, 0x3ff, 0x7ff, 0xfff, 0x1fff, 0x3fff, 0x7fff, 0xffff };
929
930 clipPa.useAlternateOffset = false;
931 pfnBinFunc(this->pDC, clipPa, this->workerId, attrib, primMaskMap[numEmittedPrims], _simd16_set1_epi32(pPrimitiveId[inputPrim]), _simd16_set1_epi32(pViewportIdx[inputPrim]));
932 }
933
934 } while (clipPa.NextPrim());
935 }
936 }
937
938 #if defined(_DEBUG)
939 free(transposedPrims);
940
941 #endif
942 // update global pipeline stat
943 UPDATE_STAT_FE(CPrimitives, numClippedPrims);
944 }
945
946 #endif
947 // execute the clipper stage
948 void ExecuteStage(PA_STATE& pa, simdvector prim[], uint32_t primMask, simdscalari primId, simdscalari viewportIdx)
949 {
950 SWR_ASSERT(pa.pDC != nullptr);
951 SWR_CONTEXT* pContext = pa.pDC->pContext;
952
953 // set up binner based on PA state
954 PFN_PROCESS_PRIMS pfnBinner;
955 switch (pa.binTopology)
956 {
957 case TOP_POINT_LIST:
958 pfnBinner = BinPoints;
959 break;
960 case TOP_LINE_LIST:
961 case TOP_LINE_STRIP:
962 case TOP_LINE_LOOP:
963 case TOP_LINE_LIST_ADJ:
964 case TOP_LISTSTRIP_ADJ:
965 pfnBinner = BinLines;
966 break;
967 default:
968 pfnBinner = GetBinTrianglesFunc((pa.pDC->pState->state.rastState.conservativeRast > 0));
969 break;
970 };
971
972 // update clipper invocations pipeline stat
973 uint32_t numInvoc = _mm_popcnt_u32(primMask);
974 UPDATE_STAT_FE(CInvocations, numInvoc);
975
976 ComputeClipCodes(prim, viewportIdx);
977
978 // cull prims with NAN coords
979 primMask &= ~ComputeNaNMask(prim);
980
981 // user cull distance cull
982 if (this->state.rastState.cullDistanceMask)
983 {
984 primMask &= ~ComputeUserClipCullMask(pa, prim);
985 }
986
987 // cull prims outside view frustum
988 simdscalar clipIntersection = ComputeClipCodeIntersection();
989 int validMask = primMask & _simd_movemask_ps(_simd_cmpeq_ps(clipIntersection, _simd_setzero_ps()));
990
991 // skip clipping for points
992 uint32_t clipMask = 0;
993 if (NumVertsPerPrim != 1)
994 {
995 clipMask = primMask & ComputeClipMask();
996 }
997
998 if (clipMask)
999 {
1000 AR_BEGIN(FEGuardbandClip, pa.pDC->drawId);
1001 // we have to clip tris, execute the clipper, which will also
1002 // call the binner
1003 ClipSimd(vMask(primMask), vMask(clipMask), pa, primId, viewportIdx);
1004 AR_END(FEGuardbandClip, 1);
1005 }
1006 else if (validMask)
1007 {
1008 // update CPrimitives pipeline state
1009 UPDATE_STAT_FE(CPrimitives, _mm_popcnt_u32(validMask));
1010
1011 // forward valid prims directly to binner
1012 pfnBinner(this->pDC, pa, this->workerId, prim, validMask, primId, viewportIdx);
1013 }
1014 }
1015
1016 #if USE_SIMD16_FRONTEND
1017 void ExecuteStage(PA_STATE& pa, simd16vector prim[], uint32_t primMask, simd16scalari primId, simd16scalari viewportIdx)
1018 {
1019 SWR_ASSERT(pa.pDC != nullptr);
1020 SWR_CONTEXT* pContext = pa.pDC->pContext;
1021
1022 // set up binner based on PA state
1023 PFN_PROCESS_PRIMS_SIMD16 pfnBinner;
1024 switch (pa.binTopology)
1025 {
1026 case TOP_POINT_LIST:
1027 pfnBinner = BinPoints_simd16;
1028 break;
1029 case TOP_LINE_LIST:
1030 case TOP_LINE_STRIP:
1031 case TOP_LINE_LOOP:
1032 case TOP_LINE_LIST_ADJ:
1033 case TOP_LISTSTRIP_ADJ:
1034 pfnBinner = BinLines_simd16;
1035 break;
1036 default:
1037 pfnBinner = GetBinTrianglesFunc_simd16((pa.pDC->pState->state.rastState.conservativeRast > 0));
1038 break;
1039 };
1040
1041 // update clipper invocations pipeline stat
1042 uint32_t numInvoc = _mm_popcnt_u32(primMask);
1043 UPDATE_STAT_FE(CInvocations, numInvoc);
1044
1045 ComputeClipCodes(prim, viewportIdx);
1046
1047 // cull prims with NAN coords
1048 primMask &= ~ComputeNaNMask(prim);
1049
1050 // user cull distance cull
1051 if (this->state.rastState.cullDistanceMask)
1052 {
1053 primMask &= ~ComputeUserClipCullMask(pa, prim);
1054 }
1055
1056 // cull prims outside view frustum
1057 simd16scalar clipIntersection = ComputeClipCodeIntersection_simd16();
1058 int validMask = primMask & _simd16_movemask_ps(_simd16_cmpeq_ps(clipIntersection, _simd16_setzero_ps()));
1059
1060 // skip clipping for points
1061 uint32_t clipMask = 0;
1062 if (NumVertsPerPrim != 1)
1063 {
1064 clipMask = primMask & ComputeClipMask_simd16();
1065 }
1066
1067 if (clipMask)
1068 {
1069 AR_BEGIN(FEGuardbandClip, pa.pDC->drawId);
1070 // we have to clip tris, execute the clipper, which will also
1071 // call the binner
1072 ClipSimd(vMask16(primMask), vMask16(clipMask), pa, primId, viewportIdx);
1073 AR_END(FEGuardbandClip, 1);
1074 }
1075 else if (validMask)
1076 {
1077 // update CPrimitives pipeline state
1078 UPDATE_STAT_FE(CPrimitives, _mm_popcnt_u32(validMask));
1079
1080 // forward valid prims directly to binner
1081 pfnBinner(this->pDC, pa, this->workerId, prim, validMask, primId, viewportIdx);
1082 }
1083 }
1084
1085 #endif
1086 private:
1087 inline simdscalar ComputeInterpFactor(simdscalar boundaryCoord0, simdscalar boundaryCoord1)
1088 {
1089 return _simd_div_ps(boundaryCoord0, _simd_sub_ps(boundaryCoord0, boundaryCoord1));
1090 }
1091
1092 #if USE_SIMD16_FRONTEND
1093 inline simd16scalar ComputeInterpFactor(simd16scalar boundaryCoord0, simd16scalar boundaryCoord1)
1094 {
1095 return _simd16_div_ps(boundaryCoord0, _simd16_sub_ps(boundaryCoord0, boundaryCoord1));
1096 }
1097
1098 #endif
1099 inline simdscalari ComputeOffsets(uint32_t attrib, simdscalari vIndices, uint32_t component)
1100 {
1101 const uint32_t simdVertexStride = sizeof(simdvertex);
1102 const uint32_t componentStride = sizeof(simdscalar);
1103 const uint32_t attribStride = sizeof(simdvector);
1104 const __m256i vElemOffset = _mm256_set_epi32(7 * sizeof(float), 6 * sizeof(float), 5 * sizeof(float), 4 * sizeof(float),
1105 3 * sizeof(float), 2 * sizeof(float), 1 * sizeof(float), 0 * sizeof(float));
1106
1107 // step to the simdvertex
1108 simdscalari vOffsets = _simd_mullo_epi32(vIndices, _simd_set1_epi32(simdVertexStride));
1109
1110 // step to the attribute and component
1111 vOffsets = _simd_add_epi32(vOffsets, _simd_set1_epi32(attribStride * attrib + componentStride * component));
1112
1113 // step to the lane
1114 vOffsets = _simd_add_epi32(vOffsets, vElemOffset);
1115
1116 return vOffsets;
1117 }
1118
1119 #if USE_SIMD16_FRONTEND
1120 inline simd16scalari ComputeOffsets(uint32_t attrib, simd16scalari vIndices, uint32_t component)
1121 {
1122 const uint32_t simdVertexStride = sizeof(simd16vertex);
1123 const uint32_t componentStride = sizeof(simd16scalar);
1124 const uint32_t attribStride = sizeof(simd16vector);
1125 const simd16scalari vElemOffset = _simd16_set_epi32(
1126 15 * sizeof(float), 14 * sizeof(float), 13 * sizeof(float), 12 * sizeof(float),
1127 11 * sizeof(float), 10 * sizeof(float), 9 * sizeof(float), 8 * sizeof(float),
1128 7 * sizeof(float), 6 * sizeof(float), 5 * sizeof(float), 4 * sizeof(float),
1129 3 * sizeof(float), 2 * sizeof(float), 1 * sizeof(float), 0 * sizeof(float));
1130
1131 // step to the simdvertex
1132 simd16scalari vOffsets = _simd16_mullo_epi32(vIndices, _simd16_set1_epi32(simdVertexStride));
1133
1134 // step to the attribute and component
1135 vOffsets = _simd16_add_epi32(vOffsets, _simd16_set1_epi32(attribStride * attrib + componentStride * component));
1136
1137 // step to the lane
1138 vOffsets = _simd16_add_epi32(vOffsets, vElemOffset);
1139
1140 return vOffsets;
1141 }
1142
1143 #endif
1144 // gathers a single component for a given attribute for each SIMD lane
1145 inline simdscalar GatherComponent(const float* pBuffer, uint32_t attrib, simdscalar vMask, simdscalari vIndices, uint32_t component)
1146 {
1147 simdscalari vOffsets = ComputeOffsets(attrib, vIndices, component);
1148 simdscalar vSrc = _mm256_undefined_ps();
1149 return _simd_mask_i32gather_ps(vSrc, pBuffer, vOffsets, vMask, 1);
1150 }
1151
1152 #if USE_SIMD16_FRONTEND
1153 inline simd16scalar GatherComponent(const float* pBuffer, uint32_t attrib, simd16scalar vMask, simd16scalari vIndices, uint32_t component)
1154 {
1155 simd16scalari vOffsets = ComputeOffsets(attrib, vIndices, component);
1156 simd16scalar vSrc = _simd16_setzero_ps();
1157 return _simd16_mask_i32gather_ps(vSrc, pBuffer, vOffsets, _simd16_castps_si(vMask), 1);
1158 }
1159
1160 #endif
1161 inline void ScatterComponent(const float* pBuffer, uint32_t attrib, simdscalar vMask, simdscalari vIndices, uint32_t component, simdscalar vSrc)
1162 {
1163 simdscalari vOffsets = ComputeOffsets(attrib, vIndices, component);
1164
1165 uint32_t* pOffsets = (uint32_t*)&vOffsets;
1166 float* pSrc = (float*)&vSrc;
1167 uint32_t mask = _simd_movemask_ps(vMask);
1168 DWORD lane;
1169 while (_BitScanForward(&lane, mask))
1170 {
1171 mask &= ~(1 << lane);
1172 uint8_t* pBuf = (uint8_t*)pBuffer + pOffsets[lane];
1173 *(float*)pBuf = pSrc[lane];
1174 }
1175 }
1176
1177 #if USE_SIMD16_FRONTEND
1178 inline void ScatterComponent(const float* pBuffer, uint32_t attrib, simd16scalar vMask, simd16scalari vIndices, uint32_t component, simd16scalar vSrc)
1179 {
1180 simd16scalari vOffsets = ComputeOffsets(attrib, vIndices, component);
1181
1182 uint32_t* pOffsets = (uint32_t*)&vOffsets;
1183 float* pSrc = (float*)&vSrc;
1184 uint32_t mask = _simd16_movemask_ps(vMask);
1185 DWORD lane;
1186 while (_BitScanForward(&lane, mask))
1187 {
1188 mask &= ~(1 << lane);
1189 uint8_t* pBuf = (uint8_t*)pBuffer + pOffsets[lane];
1190 *(float*)pBuf = pSrc[lane];
1191 }
1192 }
1193
1194 #endif
1195 template<SWR_CLIPCODES ClippingPlane>
1196 inline void intersect(
1197 const simdscalar& vActiveMask, // active lanes to operate on
1198 const simdscalari& s, // index to first edge vertex v0 in pInPts.
1199 const simdscalari& p, // index to second edge vertex v1 in pInPts.
1200 const simdvector& v1, // vertex 0 position
1201 const simdvector& v2, // vertex 1 position
1202 simdscalari& outIndex, // output index.
1203 const float *pInVerts, // array of all the input positions.
1204 uint32_t numInAttribs, // number of attributes per vertex.
1205 float *pOutVerts) // array of output positions. We'll write our new intersection point at i*4.
1206 {
1207 // compute interpolation factor
1208 simdscalar t;
1209 switch (ClippingPlane)
1210 {
1211 case FRUSTUM_LEFT: t = ComputeInterpFactor(_simd_add_ps(v1[3], v1[0]), _simd_add_ps(v2[3], v2[0])); break;
1212 case FRUSTUM_RIGHT: t = ComputeInterpFactor(_simd_sub_ps(v1[3], v1[0]), _simd_sub_ps(v2[3], v2[0])); break;
1213 case FRUSTUM_TOP: t = ComputeInterpFactor(_simd_add_ps(v1[3], v1[1]), _simd_add_ps(v2[3], v2[1])); break;
1214 case FRUSTUM_BOTTOM: t = ComputeInterpFactor(_simd_sub_ps(v1[3], v1[1]), _simd_sub_ps(v2[3], v2[1])); break;
1215 case FRUSTUM_NEAR:
1216 // DX Znear plane is 0, GL is -w
1217 if (this->state.rastState.clipHalfZ)
1218 {
1219 t = ComputeInterpFactor(v1[2], v2[2]);
1220 }
1221 else
1222 {
1223 t = ComputeInterpFactor(_simd_add_ps(v1[3], v1[2]), _simd_add_ps(v2[3], v2[2]));
1224 }
1225 break;
1226 case FRUSTUM_FAR: t = ComputeInterpFactor(_simd_sub_ps(v1[3], v1[2]), _simd_sub_ps(v2[3], v2[2])); break;
1227 default: SWR_INVALID("invalid clipping plane: %d", ClippingPlane);
1228 };
1229
1230 // interpolate position and store
1231 for (uint32_t c = 0; c < 4; ++c)
1232 {
1233 simdscalar vOutPos = _simd_fmadd_ps(_simd_sub_ps(v2[c], v1[c]), t, v1[c]);
1234 ScatterComponent(pOutVerts, VERTEX_POSITION_SLOT, vActiveMask, outIndex, c, vOutPos);
1235 }
1236
1237 // interpolate attributes and store
1238 for (uint32_t a = 0; a < numInAttribs; ++a)
1239 {
1240 uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + a;
1241 for (uint32_t c = 0; c < 4; ++c)
1242 {
1243 simdscalar vAttrib0 = GatherComponent(pInVerts, attribSlot, vActiveMask, s, c);
1244 simdscalar vAttrib1 = GatherComponent(pInVerts, attribSlot, vActiveMask, p, c);
1245 simdscalar vOutAttrib = _simd_fmadd_ps(_simd_sub_ps(vAttrib1, vAttrib0), t, vAttrib0);
1246 ScatterComponent(pOutVerts, attribSlot, vActiveMask, outIndex, c, vOutAttrib);
1247 }
1248 }
1249
1250 // interpolate clip distance if enabled
1251 if (this->state.rastState.clipDistanceMask & 0xf)
1252 {
1253 uint32_t attribSlot = VERTEX_CLIPCULL_DIST_LO_SLOT;
1254 for (uint32_t c = 0; c < 4; ++c)
1255 {
1256 simdscalar vAttrib0 = GatherComponent(pInVerts, attribSlot, vActiveMask, s, c);
1257 simdscalar vAttrib1 = GatherComponent(pInVerts, attribSlot, vActiveMask, p, c);
1258 simdscalar vOutAttrib = _simd_fmadd_ps(_simd_sub_ps(vAttrib1, vAttrib0), t, vAttrib0);
1259 ScatterComponent(pOutVerts, attribSlot, vActiveMask, outIndex, c, vOutAttrib);
1260 }
1261 }
1262
1263 if (this->state.rastState.clipDistanceMask & 0xf0)
1264 {
1265 uint32_t attribSlot = VERTEX_CLIPCULL_DIST_HI_SLOT;
1266 for (uint32_t c = 0; c < 4; ++c)
1267 {
1268 simdscalar vAttrib0 = GatherComponent(pInVerts, attribSlot, vActiveMask, s, c);
1269 simdscalar vAttrib1 = GatherComponent(pInVerts, attribSlot, vActiveMask, p, c);
1270 simdscalar vOutAttrib = _simd_fmadd_ps(_simd_sub_ps(vAttrib1, vAttrib0), t, vAttrib0);
1271 ScatterComponent(pOutVerts, attribSlot, vActiveMask, outIndex, c, vOutAttrib);
1272 }
1273 }
1274 }
1275
1276 #if USE_SIMD16_FRONTEND
1277 template<SWR_CLIPCODES ClippingPlane>
1278 inline void intersect(
1279 const simd16scalar& vActiveMask,// active lanes to operate on
1280 const simd16scalari& s, // index to first edge vertex v0 in pInPts.
1281 const simd16scalari& p, // index to second edge vertex v1 in pInPts.
1282 const simd16vector& v1, // vertex 0 position
1283 const simd16vector& v2, // vertex 1 position
1284 simd16scalari& outIndex, // output index.
1285 const float *pInVerts, // array of all the input positions.
1286 uint32_t numInAttribs, // number of attributes per vertex.
1287 float *pOutVerts) // array of output positions. We'll write our new intersection point at i*4.
1288 {
1289 // compute interpolation factor
1290 simd16scalar t;
1291 switch (ClippingPlane)
1292 {
1293 case FRUSTUM_LEFT: t = ComputeInterpFactor(_simd16_add_ps(v1[3], v1[0]), _simd16_add_ps(v2[3], v2[0])); break;
1294 case FRUSTUM_RIGHT: t = ComputeInterpFactor(_simd16_sub_ps(v1[3], v1[0]), _simd16_sub_ps(v2[3], v2[0])); break;
1295 case FRUSTUM_TOP: t = ComputeInterpFactor(_simd16_add_ps(v1[3], v1[1]), _simd16_add_ps(v2[3], v2[1])); break;
1296 case FRUSTUM_BOTTOM: t = ComputeInterpFactor(_simd16_sub_ps(v1[3], v1[1]), _simd16_sub_ps(v2[3], v2[1])); break;
1297 case FRUSTUM_NEAR:
1298 // DX Znear plane is 0, GL is -w
1299 if (this->state.rastState.clipHalfZ)
1300 {
1301 t = ComputeInterpFactor(v1[2], v2[2]);
1302 }
1303 else
1304 {
1305 t = ComputeInterpFactor(_simd16_add_ps(v1[3], v1[2]), _simd16_add_ps(v2[3], v2[2]));
1306 }
1307 break;
1308 case FRUSTUM_FAR: t = ComputeInterpFactor(_simd16_sub_ps(v1[3], v1[2]), _simd16_sub_ps(v2[3], v2[2])); break;
1309 default: SWR_INVALID("invalid clipping plane: %d", ClippingPlane);
1310 };
1311
1312 // interpolate position and store
1313 for (uint32_t c = 0; c < 4; ++c)
1314 {
1315 simd16scalar vOutPos = _simd16_fmadd_ps(_simd16_sub_ps(v2[c], v1[c]), t, v1[c]);
1316 ScatterComponent(pOutVerts, VERTEX_POSITION_SLOT, vActiveMask, outIndex, c, vOutPos);
1317 }
1318
1319 // interpolate attributes and store
1320 for (uint32_t a = 0; a < numInAttribs; ++a)
1321 {
1322 uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + a;
1323 for (uint32_t c = 0; c < 4; ++c)
1324 {
1325 simd16scalar vAttrib0 = GatherComponent(pInVerts, attribSlot, vActiveMask, s, c);
1326 simd16scalar vAttrib1 = GatherComponent(pInVerts, attribSlot, vActiveMask, p, c);
1327 simd16scalar vOutAttrib = _simd16_fmadd_ps(_simd16_sub_ps(vAttrib1, vAttrib0), t, vAttrib0);
1328 ScatterComponent(pOutVerts, attribSlot, vActiveMask, outIndex, c, vOutAttrib);
1329 }
1330 }
1331
1332 // interpolate clip distance if enabled
1333 if (this->state.rastState.clipDistanceMask & 0xf)
1334 {
1335 uint32_t attribSlot = VERTEX_CLIPCULL_DIST_LO_SLOT;
1336 for (uint32_t c = 0; c < 4; ++c)
1337 {
1338 simd16scalar vAttrib0 = GatherComponent(pInVerts, attribSlot, vActiveMask, s, c);
1339 simd16scalar vAttrib1 = GatherComponent(pInVerts, attribSlot, vActiveMask, p, c);
1340 simd16scalar vOutAttrib = _simd16_fmadd_ps(_simd16_sub_ps(vAttrib1, vAttrib0), t, vAttrib0);
1341 ScatterComponent(pOutVerts, attribSlot, vActiveMask, outIndex, c, vOutAttrib);
1342 }
1343 }
1344
1345 if (this->state.rastState.clipDistanceMask & 0xf0)
1346 {
1347 uint32_t attribSlot = VERTEX_CLIPCULL_DIST_HI_SLOT;
1348 for (uint32_t c = 0; c < 4; ++c)
1349 {
1350 simd16scalar vAttrib0 = GatherComponent(pInVerts, attribSlot, vActiveMask, s, c);
1351 simd16scalar vAttrib1 = GatherComponent(pInVerts, attribSlot, vActiveMask, p, c);
1352 simd16scalar vOutAttrib = _simd16_fmadd_ps(_simd16_sub_ps(vAttrib1, vAttrib0), t, vAttrib0);
1353 ScatterComponent(pOutVerts, attribSlot, vActiveMask, outIndex, c, vOutAttrib);
1354 }
1355 }
1356 }
1357
1358 #endif
1359 template<SWR_CLIPCODES ClippingPlane>
1360 inline simdscalar inside(const simdvector& v)
1361 {
1362 switch (ClippingPlane)
1363 {
1364 case FRUSTUM_LEFT: return _simd_cmpge_ps(v[0], _simd_mul_ps(v[3], _simd_set1_ps(-1.0f)));
1365 case FRUSTUM_RIGHT: return _simd_cmple_ps(v[0], v[3]);
1366 case FRUSTUM_TOP: return _simd_cmpge_ps(v[1], _simd_mul_ps(v[3], _simd_set1_ps(-1.0f)));
1367 case FRUSTUM_BOTTOM: return _simd_cmple_ps(v[1], v[3]);
1368 case FRUSTUM_NEAR: return _simd_cmpge_ps(v[2], this->state.rastState.clipHalfZ ? _simd_setzero_ps() : _simd_mul_ps(v[3], _simd_set1_ps(-1.0f)));
1369 case FRUSTUM_FAR: return _simd_cmple_ps(v[2], v[3]);
1370 default:
1371 SWR_INVALID("invalid clipping plane: %d", ClippingPlane);
1372 return _simd_setzero_ps();
1373 }
1374 }
1375
1376 #if USE_SIMD16_FRONTEND
1377 template<SWR_CLIPCODES ClippingPlane>
1378 inline simd16scalar inside(const simd16vector& v)
1379 {
1380 switch (ClippingPlane)
1381 {
1382 case FRUSTUM_LEFT: return _simd16_cmpge_ps(v[0], _simd16_mul_ps(v[3], _simd16_set1_ps(-1.0f)));
1383 case FRUSTUM_RIGHT: return _simd16_cmple_ps(v[0], v[3]);
1384 case FRUSTUM_TOP: return _simd16_cmpge_ps(v[1], _simd16_mul_ps(v[3], _simd16_set1_ps(-1.0f)));
1385 case FRUSTUM_BOTTOM: return _simd16_cmple_ps(v[1], v[3]);
1386 case FRUSTUM_NEAR: return _simd16_cmpge_ps(v[2], this->state.rastState.clipHalfZ ? _simd16_setzero_ps() : _simd16_mul_ps(v[3], _simd16_set1_ps(-1.0f)));
1387 case FRUSTUM_FAR: return _simd16_cmple_ps(v[2], v[3]);
1388 default:
1389 SWR_INVALID("invalid clipping plane: %d", ClippingPlane);
1390 return _simd16_setzero_ps();
1391 }
1392 }
1393
1394 #endif
1395 template<SWR_CLIPCODES ClippingPlane>
1396 simdscalari ClipTriToPlane(const float* pInVerts, const simdscalari& vNumInPts, uint32_t numInAttribs, float* pOutVerts)
1397 {
1398 simdscalari vCurIndex = _simd_setzero_si();
1399 simdscalari vOutIndex = _simd_setzero_si();
1400 simdscalar vActiveMask = _simd_castsi_ps(_simd_cmplt_epi32(vCurIndex, vNumInPts));
1401
1402 while (!_simd_testz_ps(vActiveMask, vActiveMask)) // loop until activeMask is empty
1403 {
1404 simdscalari s = vCurIndex;
1405 simdscalari p = _simd_add_epi32(s, _simd_set1_epi32(1));
1406 simdscalari underFlowMask = _simd_cmpgt_epi32(vNumInPts, p);
1407 p = _simd_castps_si(_simd_blendv_ps(_simd_setzero_ps(), _simd_castsi_ps(p), _simd_castsi_ps(underFlowMask)));
1408
1409 // gather position
1410 simdvector vInPos0, vInPos1;
1411 for (uint32_t c = 0; c < 4; ++c)
1412 {
1413 vInPos0[c] = GatherComponent(pInVerts, VERTEX_POSITION_SLOT, vActiveMask, s, c);
1414 vInPos1[c] = GatherComponent(pInVerts, VERTEX_POSITION_SLOT, vActiveMask, p, c);
1415 }
1416
1417 // compute inside mask
1418 simdscalar s_in = inside<ClippingPlane>(vInPos0);
1419 simdscalar p_in = inside<ClippingPlane>(vInPos1);
1420
1421 // compute intersection mask (s_in != p_in)
1422 simdscalar intersectMask = _simd_xor_ps(s_in, p_in);
1423 intersectMask = _simd_and_ps(intersectMask, vActiveMask);
1424
1425 // store s if inside
1426 s_in = _simd_and_ps(s_in, vActiveMask);
1427 if (!_simd_testz_ps(s_in, s_in))
1428 {
1429 // store position
1430 for (uint32_t c = 0; c < 4; ++c)
1431 {
1432 ScatterComponent(pOutVerts, VERTEX_POSITION_SLOT, s_in, vOutIndex, c, vInPos0[c]);
1433 }
1434
1435 // store attribs
1436 for (uint32_t a = 0; a < numInAttribs; ++a)
1437 {
1438 uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + a;
1439 for (uint32_t c = 0; c < 4; ++c)
1440 {
1441 simdscalar vAttrib = GatherComponent(pInVerts, attribSlot, s_in, s, c);
1442 ScatterComponent(pOutVerts, attribSlot, s_in, vOutIndex, c, vAttrib);
1443 }
1444 }
1445
1446 // store clip distance if enabled
1447 if (this->state.rastState.clipDistanceMask & 0xf)
1448 {
1449 uint32_t attribSlot = VERTEX_CLIPCULL_DIST_LO_SLOT;
1450 for (uint32_t c = 0; c < 4; ++c)
1451 {
1452 simdscalar vAttrib = GatherComponent(pInVerts, attribSlot, s_in, s, c);
1453 ScatterComponent(pOutVerts, attribSlot, s_in, vOutIndex, c, vAttrib);
1454 }
1455 }
1456
1457 if (this->state.rastState.clipDistanceMask & 0xf0)
1458 {
1459 uint32_t attribSlot = VERTEX_CLIPCULL_DIST_HI_SLOT;
1460 for (uint32_t c = 0; c < 4; ++c)
1461 {
1462 simdscalar vAttrib = GatherComponent(pInVerts, attribSlot, s_in, s, c);
1463 ScatterComponent(pOutVerts, attribSlot, s_in, vOutIndex, c, vAttrib);
1464 }
1465 }
1466
1467 // increment outIndex
1468 vOutIndex = _simd_blendv_epi32(vOutIndex, _simd_add_epi32(vOutIndex, _simd_set1_epi32(1)), s_in);
1469 }
1470
1471 // compute and store intersection
1472 if (!_simd_testz_ps(intersectMask, intersectMask))
1473 {
1474 intersect<ClippingPlane>(intersectMask, s, p, vInPos0, vInPos1, vOutIndex, pInVerts, numInAttribs, pOutVerts);
1475
1476 // increment outIndex for active lanes
1477 vOutIndex = _simd_blendv_epi32(vOutIndex, _simd_add_epi32(vOutIndex, _simd_set1_epi32(1)), intersectMask);
1478 }
1479
1480 // increment loop index and update active mask
1481 vCurIndex = _simd_add_epi32(vCurIndex, _simd_set1_epi32(1));
1482 vActiveMask = _simd_castsi_ps(_simd_cmplt_epi32(vCurIndex, vNumInPts));
1483 }
1484
1485 return vOutIndex;
1486 }
1487
1488 #if USE_SIMD16_FRONTEND
1489 template<SWR_CLIPCODES ClippingPlane>
1490 simd16scalari ClipTriToPlane(const float* pInVerts, const simd16scalari& vNumInPts, uint32_t numInAttribs, float* pOutVerts)
1491 {
1492 simd16scalari vCurIndex = _simd16_setzero_si();
1493 simd16scalari vOutIndex = _simd16_setzero_si();
1494 simd16scalar vActiveMask = _simd16_castsi_ps(_simd16_cmplt_epi32(vCurIndex, vNumInPts));
1495
1496 while (!_simd16_testz_ps(vActiveMask, vActiveMask)) // loop until activeMask is empty
1497 {
1498 simd16scalari s = vCurIndex;
1499 simd16scalari p = _simd16_add_epi32(s, _simd16_set1_epi32(1));
1500 simd16scalari underFlowMask = _simd16_cmpgt_epi32(vNumInPts, p);
1501 p = _simd16_castps_si(_simd16_blendv_ps(_simd16_setzero_ps(), _simd16_castsi_ps(p), _simd16_castsi_ps(underFlowMask)));
1502
1503 // gather position
1504 simd16vector vInPos0, vInPos1;
1505 for (uint32_t c = 0; c < 4; ++c)
1506 {
1507 vInPos0[c] = GatherComponent(pInVerts, VERTEX_POSITION_SLOT, vActiveMask, s, c);
1508 vInPos1[c] = GatherComponent(pInVerts, VERTEX_POSITION_SLOT, vActiveMask, p, c);
1509 }
1510
1511 // compute inside mask
1512 simd16scalar s_in = inside<ClippingPlane>(vInPos0);
1513 simd16scalar p_in = inside<ClippingPlane>(vInPos1);
1514
1515 // compute intersection mask (s_in != p_in)
1516 simd16scalar intersectMask = _simd16_xor_ps(s_in, p_in);
1517 intersectMask = _simd16_and_ps(intersectMask, vActiveMask);
1518
1519 // store s if inside
1520 s_in = _simd16_and_ps(s_in, vActiveMask);
1521 if (!_simd16_testz_ps(s_in, s_in))
1522 {
1523 // store position
1524 for (uint32_t c = 0; c < 4; ++c)
1525 {
1526 ScatterComponent(pOutVerts, VERTEX_POSITION_SLOT, s_in, vOutIndex, c, vInPos0[c]);
1527 }
1528
1529 // store attribs
1530 for (uint32_t a = 0; a < numInAttribs; ++a)
1531 {
1532 uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + a;
1533 for (uint32_t c = 0; c < 4; ++c)
1534 {
1535 simd16scalar vAttrib = GatherComponent(pInVerts, attribSlot, s_in, s, c);
1536 ScatterComponent(pOutVerts, attribSlot, s_in, vOutIndex, c, vAttrib);
1537 }
1538 }
1539
1540 // store clip distance if enabled
1541 if (this->state.rastState.clipDistanceMask & 0xf)
1542 {
1543 uint32_t attribSlot = VERTEX_CLIPCULL_DIST_LO_SLOT;
1544 for (uint32_t c = 0; c < 4; ++c)
1545 {
1546 simd16scalar vAttrib = GatherComponent(pInVerts, attribSlot, s_in, s, c);
1547 ScatterComponent(pOutVerts, attribSlot, s_in, vOutIndex, c, vAttrib);
1548 }
1549 }
1550
1551 if (this->state.rastState.clipDistanceMask & 0xf0)
1552 {
1553 uint32_t attribSlot = VERTEX_CLIPCULL_DIST_HI_SLOT;
1554 for (uint32_t c = 0; c < 4; ++c)
1555 {
1556 simd16scalar vAttrib = GatherComponent(pInVerts, attribSlot, s_in, s, c);
1557 ScatterComponent(pOutVerts, attribSlot, s_in, vOutIndex, c, vAttrib);
1558 }
1559 }
1560
1561 // increment outIndex
1562 vOutIndex = _simd16_blendv_epi32(vOutIndex, _simd16_add_epi32(vOutIndex, _simd16_set1_epi32(1)), s_in);
1563 }
1564
1565 // compute and store intersection
1566 if (!_simd16_testz_ps(intersectMask, intersectMask))
1567 {
1568 intersect<ClippingPlane>(intersectMask, s, p, vInPos0, vInPos1, vOutIndex, pInVerts, numInAttribs, pOutVerts);
1569
1570 // increment outIndex for active lanes
1571 vOutIndex = _simd16_blendv_epi32(vOutIndex, _simd16_add_epi32(vOutIndex, _simd16_set1_epi32(1)), intersectMask);
1572 }
1573
1574 // increment loop index and update active mask
1575 vCurIndex = _simd16_add_epi32(vCurIndex, _simd16_set1_epi32(1));
1576 vActiveMask = _simd16_castsi_ps(_simd16_cmplt_epi32(vCurIndex, vNumInPts));
1577 }
1578
1579 return vOutIndex;
1580 }
1581
1582 #endif
1583 template<SWR_CLIPCODES ClippingPlane>
1584 simdscalari ClipLineToPlane(const float* pInVerts, const simdscalari& vNumInPts, uint32_t numInAttribs, float* pOutVerts)
1585 {
1586 simdscalari vCurIndex = _simd_setzero_si();
1587 simdscalari vOutIndex = _simd_setzero_si();
1588 simdscalar vActiveMask = _simd_castsi_ps(_simd_cmplt_epi32(vCurIndex, vNumInPts));
1589
1590 if (!_simd_testz_ps(vActiveMask, vActiveMask))
1591 {
1592 simdscalari s = vCurIndex;
1593 simdscalari p = _simd_add_epi32(s, _simd_set1_epi32(1));
1594
1595 // gather position
1596 simdvector vInPos0, vInPos1;
1597 for (uint32_t c = 0; c < 4; ++c)
1598 {
1599 vInPos0[c] = GatherComponent(pInVerts, VERTEX_POSITION_SLOT, vActiveMask, s, c);
1600 vInPos1[c] = GatherComponent(pInVerts, VERTEX_POSITION_SLOT, vActiveMask, p, c);
1601 }
1602
1603 // compute inside mask
1604 simdscalar s_in = inside<ClippingPlane>(vInPos0);
1605 simdscalar p_in = inside<ClippingPlane>(vInPos1);
1606
1607 // compute intersection mask (s_in != p_in)
1608 simdscalar intersectMask = _simd_xor_ps(s_in, p_in);
1609 intersectMask = _simd_and_ps(intersectMask, vActiveMask);
1610
1611 // store s if inside
1612 s_in = _simd_and_ps(s_in, vActiveMask);
1613 if (!_simd_testz_ps(s_in, s_in))
1614 {
1615 for (uint32_t c = 0; c < 4; ++c)
1616 {
1617 ScatterComponent(pOutVerts, VERTEX_POSITION_SLOT, s_in, vOutIndex, c, vInPos0[c]);
1618 }
1619
1620 // interpolate attributes and store
1621 for (uint32_t a = 0; a < numInAttribs; ++a)
1622 {
1623 uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + a;
1624 for (uint32_t c = 0; c < 4; ++c)
1625 {
1626 simdscalar vAttrib = GatherComponent(pInVerts, attribSlot, s_in, s, c);
1627 ScatterComponent(pOutVerts, attribSlot, s_in, vOutIndex, c, vAttrib);
1628 }
1629 }
1630
1631 // increment outIndex
1632 vOutIndex = _simd_blendv_epi32(vOutIndex, _simd_add_epi32(vOutIndex, _simd_set1_epi32(1)), s_in);
1633 }
1634
1635 // compute and store intersection
1636 if (!_simd_testz_ps(intersectMask, intersectMask))
1637 {
1638 intersect<ClippingPlane>(intersectMask, s, p, vInPos0, vInPos1, vOutIndex, pInVerts, numInAttribs, pOutVerts);
1639
1640 // increment outIndex for active lanes
1641 vOutIndex = _simd_blendv_epi32(vOutIndex, _simd_add_epi32(vOutIndex, _simd_set1_epi32(1)), intersectMask);
1642 }
1643
1644 // store p if inside
1645 p_in = _simd_and_ps(p_in, vActiveMask);
1646 if (!_simd_testz_ps(p_in, p_in))
1647 {
1648 for (uint32_t c = 0; c < 4; ++c)
1649 {
1650 ScatterComponent(pOutVerts, VERTEX_POSITION_SLOT, p_in, vOutIndex, c, vInPos1[c]);
1651 }
1652
1653 // interpolate attributes and store
1654 for (uint32_t a = 0; a < numInAttribs; ++a)
1655 {
1656 uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + a;
1657 for (uint32_t c = 0; c < 4; ++c)
1658 {
1659 simdscalar vAttrib = GatherComponent(pInVerts, attribSlot, p_in, p, c);
1660 ScatterComponent(pOutVerts, attribSlot, p_in, vOutIndex, c, vAttrib);
1661 }
1662 }
1663
1664 // increment outIndex
1665 vOutIndex = _simd_blendv_epi32(vOutIndex, _simd_add_epi32(vOutIndex, _simd_set1_epi32(1)), p_in);
1666 }
1667 }
1668
1669 return vOutIndex;
1670 }
1671
1672 #if USE_SIMD16_FRONTEND
1673 template<SWR_CLIPCODES ClippingPlane>
1674 simd16scalari ClipLineToPlane(const float* pInVerts, const simd16scalari& vNumInPts, uint32_t numInAttribs, float* pOutVerts)
1675 {
1676 simd16scalari vCurIndex = _simd16_setzero_si();
1677 simd16scalari vOutIndex = _simd16_setzero_si();
1678 simd16scalar vActiveMask = _simd16_castsi_ps(_simd16_cmplt_epi32(vCurIndex, vNumInPts));
1679
1680 if (!_simd16_testz_ps(vActiveMask, vActiveMask))
1681 {
1682 simd16scalari s = vCurIndex;
1683 simd16scalari p = _simd16_add_epi32(s, _simd16_set1_epi32(1));
1684
1685 // gather position
1686 simd16vector vInPos0, vInPos1;
1687 for (uint32_t c = 0; c < 4; ++c)
1688 {
1689 vInPos0[c] = GatherComponent(pInVerts, VERTEX_POSITION_SLOT, vActiveMask, s, c);
1690 vInPos1[c] = GatherComponent(pInVerts, VERTEX_POSITION_SLOT, vActiveMask, p, c);
1691 }
1692
1693 // compute inside mask
1694 simd16scalar s_in = inside<ClippingPlane>(vInPos0);
1695 simd16scalar p_in = inside<ClippingPlane>(vInPos1);
1696
1697 // compute intersection mask (s_in != p_in)
1698 simd16scalar intersectMask = _simd16_xor_ps(s_in, p_in);
1699 intersectMask = _simd16_and_ps(intersectMask, vActiveMask);
1700
1701 // store s if inside
1702 s_in = _simd16_and_ps(s_in, vActiveMask);
1703 if (!_simd16_testz_ps(s_in, s_in))
1704 {
1705 for (uint32_t c = 0; c < 4; ++c)
1706 {
1707 ScatterComponent(pOutVerts, VERTEX_POSITION_SLOT, s_in, vOutIndex, c, vInPos0[c]);
1708 }
1709
1710 // interpolate attributes and store
1711 for (uint32_t a = 0; a < numInAttribs; ++a)
1712 {
1713 uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + a;
1714 for (uint32_t c = 0; c < 4; ++c)
1715 {
1716 simd16scalar vAttrib = GatherComponent(pInVerts, attribSlot, s_in, s, c);
1717 ScatterComponent(pOutVerts, attribSlot, s_in, vOutIndex, c, vAttrib);
1718 }
1719 }
1720
1721 // increment outIndex
1722 vOutIndex = _simd16_blendv_epi32(vOutIndex, _simd16_add_epi32(vOutIndex, _simd16_set1_epi32(1)), s_in);
1723 }
1724
1725 // compute and store intersection
1726 if (!_simd16_testz_ps(intersectMask, intersectMask))
1727 {
1728 intersect<ClippingPlane>(intersectMask, s, p, vInPos0, vInPos1, vOutIndex, pInVerts, numInAttribs, pOutVerts);
1729
1730 // increment outIndex for active lanes
1731 vOutIndex = _simd16_blendv_epi32(vOutIndex, _simd16_add_epi32(vOutIndex, _simd16_set1_epi32(1)), intersectMask);
1732 }
1733
1734 // store p if inside
1735 p_in = _simd16_and_ps(p_in, vActiveMask);
1736 if (!_simd16_testz_ps(p_in, p_in))
1737 {
1738 for (uint32_t c = 0; c < 4; ++c)
1739 {
1740 ScatterComponent(pOutVerts, VERTEX_POSITION_SLOT, p_in, vOutIndex, c, vInPos1[c]);
1741 }
1742
1743 // interpolate attributes and store
1744 for (uint32_t a = 0; a < numInAttribs; ++a)
1745 {
1746 uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + a;
1747 for (uint32_t c = 0; c < 4; ++c)
1748 {
1749 simd16scalar vAttrib = GatherComponent(pInVerts, attribSlot, p_in, p, c);
1750 ScatterComponent(pOutVerts, attribSlot, p_in, vOutIndex, c, vAttrib);
1751 }
1752 }
1753
1754 // increment outIndex
1755 vOutIndex = _simd16_blendv_epi32(vOutIndex, _simd16_add_epi32(vOutIndex, _simd16_set1_epi32(1)), p_in);
1756 }
1757 }
1758
1759 return vOutIndex;
1760 }
1761 #endif
1762 //////////////////////////////////////////////////////////////////////////
1763 /// @brief Vertical clipper. Clips SIMD primitives at a time
1764 /// @param pVertices - pointer to vertices in SOA form. Clipper will read input and write results to this buffer
1765 /// @param vPrimMask - mask of valid input primitives, including non-clipped prims
1766 /// @param numAttribs - number of valid input attribs, including position
1767 simdscalari ClipPrims(float* pVertices, const simdscalar& vPrimMask, const simdscalar& vClipMask, int numAttribs)
1768 {
1769 // temp storage
1770 float* pTempVerts = (float*)&tlsTempVertices[0];
1771
1772 // zero out num input verts for non-active lanes
1773 simdscalari vNumInPts = _simd_set1_epi32(NumVertsPerPrim);
1774 vNumInPts = _simd_blendv_epi32(_simd_setzero_si(), vNumInPts, vClipMask);
1775
1776 // clip prims to frustum
1777 simdscalari vNumOutPts;
1778 if (NumVertsPerPrim == 3)
1779 {
1780 vNumOutPts = ClipTriToPlane<FRUSTUM_NEAR>(pVertices, vNumInPts, numAttribs, pTempVerts);
1781 vNumOutPts = ClipTriToPlane<FRUSTUM_FAR>(pTempVerts, vNumOutPts, numAttribs, pVertices);
1782 vNumOutPts = ClipTriToPlane<FRUSTUM_LEFT>(pVertices, vNumOutPts, numAttribs, pTempVerts);
1783 vNumOutPts = ClipTriToPlane<FRUSTUM_RIGHT>(pTempVerts, vNumOutPts, numAttribs, pVertices);
1784 vNumOutPts = ClipTriToPlane<FRUSTUM_BOTTOM>(pVertices, vNumOutPts, numAttribs, pTempVerts);
1785 vNumOutPts = ClipTriToPlane<FRUSTUM_TOP>(pTempVerts, vNumOutPts, numAttribs, pVertices);
1786 }
1787 else
1788 {
1789 SWR_ASSERT(NumVertsPerPrim == 2);
1790 vNumOutPts = ClipLineToPlane<FRUSTUM_NEAR>(pVertices, vNumInPts, numAttribs, pTempVerts);
1791 vNumOutPts = ClipLineToPlane<FRUSTUM_FAR>(pTempVerts, vNumOutPts, numAttribs, pVertices);
1792 vNumOutPts = ClipLineToPlane<FRUSTUM_LEFT>(pVertices, vNumOutPts, numAttribs, pTempVerts);
1793 vNumOutPts = ClipLineToPlane<FRUSTUM_RIGHT>(pTempVerts, vNumOutPts, numAttribs, pVertices);
1794 vNumOutPts = ClipLineToPlane<FRUSTUM_BOTTOM>(pVertices, vNumOutPts, numAttribs, pTempVerts);
1795 vNumOutPts = ClipLineToPlane<FRUSTUM_TOP>(pTempVerts, vNumOutPts, numAttribs, pVertices);
1796 }
1797
1798 // restore num verts for non-clipped, active lanes
1799 simdscalar vNonClippedMask = _simd_andnot_ps(vClipMask, vPrimMask);
1800 vNumOutPts = _simd_blendv_epi32(vNumOutPts, _simd_set1_epi32(NumVertsPerPrim), vNonClippedMask);
1801
1802 return vNumOutPts;
1803 }
1804
1805 #if USE_SIMD16_FRONTEND
1806 simd16scalari ClipPrims(float* pVertices, const simd16scalar& vPrimMask, const simd16scalar& vClipMask, int numAttribs)
1807 {
1808 // temp storage
1809 float* pTempVerts = (float*)&tlsTempVertices_simd16[0];
1810
1811 // zero out num input verts for non-active lanes
1812 simd16scalari vNumInPts = _simd16_set1_epi32(NumVertsPerPrim);
1813 vNumInPts = _simd16_blendv_epi32(_simd16_setzero_si(), vNumInPts, vClipMask);
1814
1815 // clip prims to frustum
1816 simd16scalari vNumOutPts;
1817 if (NumVertsPerPrim == 3)
1818 {
1819 vNumOutPts = ClipTriToPlane<FRUSTUM_NEAR>(pVertices, vNumInPts, numAttribs, pTempVerts);
1820 vNumOutPts = ClipTriToPlane<FRUSTUM_FAR>(pTempVerts, vNumOutPts, numAttribs, pVertices);
1821 vNumOutPts = ClipTriToPlane<FRUSTUM_LEFT>(pVertices, vNumOutPts, numAttribs, pTempVerts);
1822 vNumOutPts = ClipTriToPlane<FRUSTUM_RIGHT>(pTempVerts, vNumOutPts, numAttribs, pVertices);
1823 vNumOutPts = ClipTriToPlane<FRUSTUM_BOTTOM>(pVertices, vNumOutPts, numAttribs, pTempVerts);
1824 vNumOutPts = ClipTriToPlane<FRUSTUM_TOP>(pTempVerts, vNumOutPts, numAttribs, pVertices);
1825 }
1826 else
1827 {
1828 SWR_ASSERT(NumVertsPerPrim == 2);
1829 vNumOutPts = ClipLineToPlane<FRUSTUM_NEAR>(pVertices, vNumInPts, numAttribs, pTempVerts);
1830 vNumOutPts = ClipLineToPlane<FRUSTUM_FAR>(pTempVerts, vNumOutPts, numAttribs, pVertices);
1831 vNumOutPts = ClipLineToPlane<FRUSTUM_LEFT>(pVertices, vNumOutPts, numAttribs, pTempVerts);
1832 vNumOutPts = ClipLineToPlane<FRUSTUM_RIGHT>(pTempVerts, vNumOutPts, numAttribs, pVertices);
1833 vNumOutPts = ClipLineToPlane<FRUSTUM_BOTTOM>(pVertices, vNumOutPts, numAttribs, pTempVerts);
1834 vNumOutPts = ClipLineToPlane<FRUSTUM_TOP>(pTempVerts, vNumOutPts, numAttribs, pVertices);
1835 }
1836
1837 // restore num verts for non-clipped, active lanes
1838 simd16scalar vNonClippedMask = _simd16_andnot_ps(vClipMask, vPrimMask);
1839 vNumOutPts = _simd16_blendv_epi32(vNumOutPts, _simd16_set1_epi32(NumVertsPerPrim), vNonClippedMask);
1840
1841 return vNumOutPts;
1842 }
1843
1844 #endif
1845 const uint32_t workerId{ 0 };
1846 DRAW_CONTEXT* pDC{ nullptr };
1847 const API_STATE& state;
1848 simdscalar clipCodes[NumVertsPerPrim];
1849 #if USE_SIMD16_FRONTEND
1850 simd16scalar clipCodes_simd16[NumVertsPerPrim];
1851 #endif
1852 };
1853
1854
1855 // pipeline stage functions
1856 void ClipTriangles(DRAW_CONTEXT *pDC, PA_STATE& pa, uint32_t workerId, simdvector prims[], uint32_t primMask, simdscalari primId, simdscalari viewportIdx);
1857 void ClipLines(DRAW_CONTEXT *pDC, PA_STATE& pa, uint32_t workerId, simdvector prims[], uint32_t primMask, simdscalari primId, simdscalari viewportIdx);
1858 void ClipPoints(DRAW_CONTEXT *pDC, PA_STATE& pa, uint32_t workerId, simdvector prims[], uint32_t primMask, simdscalari primId, simdscalari viewportIdx);
1859 #if USE_SIMD16_FRONTEND
1860 void SIMDAPI ClipTriangles_simd16(DRAW_CONTEXT *pDC, PA_STATE& pa, uint32_t workerId, simd16vector prims[], uint32_t primMask, simd16scalari primId, simd16scalari viewportIdx);
1861 void SIMDAPI ClipLines_simd16(DRAW_CONTEXT *pDC, PA_STATE& pa, uint32_t workerId, simd16vector prims[], uint32_t primMask, simd16scalari primId, simd16scalari viewportIdx);
1862 void SIMDAPI ClipPoints_simd16(DRAW_CONTEXT *pDC, PA_STATE& pa, uint32_t workerId, simd16vector prims[], uint32_t primMask, simd16scalari primId, simd16scalari viewportIdx);
1863 #endif
1864