1 /****************************************************************************
2 * Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved.
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:
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
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
25 * @brief Definitions for clipping
27 ******************************************************************************/
30 #include "common/simdintrin.h"
31 #include "core/context.h"
33 #include "rdtsc_core.h"
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];
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
),
51 FRUSTUM_NEAR
= (0x10 << CLIPCODE_SHIFT
),
52 FRUSTUM_FAR
= (0x20 << CLIPCODE_SHIFT
),
54 NEGW
= (0x40 << CLIPCODE_SHIFT
),
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)
62 #define GUARDBAND_CLIP_MASK (FRUSTUM_NEAR|FRUSTUM_FAR|GUARDBAND_LEFT|GUARDBAND_TOP|GUARDBAND_RIGHT|GUARDBAND_BOTTOM|NEGW)
65 void ComputeClipCodes(const API_STATE
& state
, const simdvector
& vertex
, simdscalar
& clipCodes
, simdscalari viewportIndexes
)
67 clipCodes
= _simd_setzero_ps();
70 simdscalar vNegW
= _simd_mul_ps(vertex
.w
, _simd_set1_ps(-1.0f
));
73 simdscalar vRes
= _simd_cmplt_ps(vertex
.x
, vNegW
);
74 clipCodes
= _simd_and_ps(vRes
, _simd_castsi_ps(_simd_set1_epi32(FRUSTUM_LEFT
)));
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
))));
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
))));
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
))));
88 if (state
.rastState
.depthClipEnable
)
91 // DX clips depth [0..w], GL clips [-w..w]
92 if (state
.rastState
.clipHalfZ
)
94 vRes
= _simd_cmplt_ps(vertex
.z
, _simd_setzero_ps());
98 vRes
= _simd_cmplt_ps(vertex
.z
, vNegW
);
100 clipCodes
= _simd_or_ps(clipCodes
, _simd_and_ps(vRes
, _simd_castsi_ps(_simd_set1_epi32(FRUSTUM_NEAR
))));
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
))));
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
))));
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
))));
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
))));
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
))));
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
))));
132 #if USE_SIMD16_FRONTEND
134 void ComputeClipCodes(const API_STATE
& state
, const simd16vector
& vertex
, simd16scalar
& clipCodes
, simd16scalari viewportIndexes
)
136 clipCodes
= _simd16_setzero_ps();
139 simd16scalar vNegW
= _simd16_mul_ps(vertex
.w
, _simd16_set1_ps(-1.0f
));
142 simd16scalar vRes
= _simd16_cmplt_ps(vertex
.x
, vNegW
);
143 clipCodes
= _simd16_and_ps(vRes
, _simd16_castsi_ps(_simd16_set1_epi32(FRUSTUM_LEFT
)));
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
))));
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
))));
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
))));
157 if (state
.rastState
.depthClipEnable
)
160 // DX clips depth [0..w], GL clips [-w..w]
161 if (state
.rastState
.clipHalfZ
)
163 vRes
= _simd16_cmplt_ps(vertex
.z
, _simd16_setzero_ps());
167 vRes
= _simd16_cmplt_ps(vertex
.z
, vNegW
);
169 clipCodes
= _simd16_or_ps(clipCodes
, _simd16_and_ps(vRes
, _simd16_castsi_ps(_simd16_set1_epi32(FRUSTUM_NEAR
))));
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
))));
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
))));
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
))));
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
))));
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
))));
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
))));
202 template<uint32_t NumVertsPerPrim
>
206 Clipper(uint32_t in_workerId
, DRAW_CONTEXT
* in_pDC
) :
207 workerId(in_workerId
), pDC(in_pDC
), state(GetApiState(in_pDC
))
209 static_assert(NumVertsPerPrim
>= 1 && NumVertsPerPrim
<= 3, "Invalid NumVertsPerPrim");
212 void ComputeClipCodes(simdvector vertex
[], simdscalari viewportIndexes
)
214 for (uint32_t i
= 0; i
< NumVertsPerPrim
; ++i
)
216 ::ComputeClipCodes(this->state
, vertex
[i
], this->clipCodes
[i
], viewportIndexes
);
220 #if USE_SIMD16_FRONTEND
221 void ComputeClipCodes(simd16vector vertex
[], simd16scalari viewportIndexes
)
223 for (uint32_t i
= 0; i
< NumVertsPerPrim
; ++i
)
225 ::ComputeClipCodes(this->state
, vertex
[i
], this->clipCodes_simd16
[i
], viewportIndexes
);
230 simdscalar
ComputeClipCodeIntersection()
232 simdscalar result
= this->clipCodes
[0];
233 for (uint32_t i
= 1; i
< NumVertsPerPrim
; ++i
)
235 result
= _simd_and_ps(result
, this->clipCodes
[i
]);
240 #if USE_SIMD16_FRONTEND
241 simd16scalar
ComputeClipCodeIntersection_simd16()
243 simd16scalar result
= this->clipCodes_simd16
[0];
244 for (uint32_t i
= 1; i
< NumVertsPerPrim
; ++i
)
246 result
= _simd16_and_ps(result
, this->clipCodes_simd16
[i
]);
252 simdscalar
ComputeClipCodeUnion()
254 simdscalar result
= this->clipCodes
[0];
255 for (uint32_t i
= 1; i
< NumVertsPerPrim
; ++i
)
257 result
= _simd_or_ps(result
, this->clipCodes
[i
]);
262 #if USE_SIMD16_FRONTEND
263 simd16scalar
ComputeClipCodeUnion_simd16()
265 simd16scalar result
= this->clipCodes_simd16
[0];
266 for (uint32_t i
= 1; i
< NumVertsPerPrim
; ++i
)
268 result
= _simd16_or_ps(result
, this->clipCodes_simd16
[i
]);
274 int ComputeNegWMask()
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()));
281 int ComputeClipMask()
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()));
288 #if USE_SIMD16_FRONTEND
289 int ComputeClipMask_simd16()
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()));
297 // clipper is responsible for culling any prims with NAN coordinates
298 int ComputeNaNMask(simdvector prim
[])
300 simdscalar vNanMask
= _simd_setzero_ps();
301 for (uint32_t e
= 0; e
< NumVertsPerPrim
; ++e
)
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
);
309 return _simd_movemask_ps(vNanMask
);
312 #if USE_SIMD16_FRONTEND
313 int ComputeNaNMask(simd16vector prim
[])
315 simd16scalar vNanMask
= _simd16_setzero_ps();
316 for (uint32_t e
= 0; e
< NumVertsPerPrim
; ++e
)
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
);
324 return _simd16_movemask_ps(vNanMask
);
328 int ComputeUserClipCullMask(PA_STATE
& pa
, simdvector prim
[])
330 uint8_t cullMask
= this->state
.rastState
.cullDistanceMask
;
331 simdscalar vClipCullMask
= _simd_setzero_ps();
334 simdvector vClipCullDistLo
[3];
335 simdvector vClipCullDistHi
[3];
337 pa
.Assemble(VERTEX_CLIPCULL_DIST_LO_SLOT
, vClipCullDistLo
);
338 pa
.Assemble(VERTEX_CLIPCULL_DIST_HI_SLOT
, vClipCullDistHi
);
339 while (_BitScanForward(&index
, cullMask
))
341 cullMask
&= ~(1 << index
);
342 uint32_t slot
= index
>> 2;
343 uint32_t component
= index
& 0x3;
345 simdscalar vCullMaskElem
= _simd_set1_ps(-1.0f
);
346 for (uint32_t e
= 0; e
< NumVertsPerPrim
; ++e
)
348 simdscalar vCullComp
;
351 vCullComp
= vClipCullDistLo
[e
][component
];
355 vCullComp
= vClipCullDistHi
[e
][component
];
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
);
362 vClipCullMask
= _simd_or_ps(vClipCullMask
, vCullMaskElem
);
365 // clipper should also discard any primitive with NAN clip distance
366 uint8_t clipMask
= this->state
.rastState
.clipDistanceMask
;
367 while (_BitScanForward(&index
, clipMask
))
369 clipMask
&= ~(1 << index
);
370 uint32_t slot
= index
>> 2;
371 uint32_t component
= index
& 0x3;
373 for (uint32_t e
= 0; e
< NumVertsPerPrim
; ++e
)
375 simdscalar vClipComp
;
378 vClipComp
= vClipCullDistLo
[e
][component
];
382 vClipComp
= vClipCullDistHi
[e
][component
];
385 simdscalar vClip
= _simd_cmp_ps(vClipComp
, vClipComp
, _CMP_UNORD_Q
);
386 vClipCullMask
= _simd_or_ps(vClipCullMask
, vClip
);
390 return _simd_movemask_ps(vClipCullMask
);
393 #if USE_SIMD16_FRONTEND
394 int ComputeUserClipCullMask(PA_STATE
& pa
, simd16vector prim
[])
396 uint8_t cullMask
= this->state
.rastState
.cullDistanceMask
;
397 simd16scalar vClipCullMask
= _simd16_setzero_ps();
399 simd16vector vClipCullDistLo
[3];
400 simd16vector vClipCullDistHi
[3];
402 pa
.Assemble_simd16(VERTEX_CLIPCULL_DIST_LO_SLOT
, vClipCullDistLo
);
403 pa
.Assemble_simd16(VERTEX_CLIPCULL_DIST_HI_SLOT
, vClipCullDistHi
);
406 while (_BitScanForward(&index
, cullMask
))
408 cullMask
&= ~(1 << index
);
409 uint32_t slot
= index
>> 2;
410 uint32_t component
= index
& 0x3;
412 simd16scalar vCullMaskElem
= _simd16_set1_ps(-1.0f
);
413 for (uint32_t e
= 0; e
< NumVertsPerPrim
; ++e
)
415 simd16scalar vCullComp
;
418 vCullComp
= vClipCullDistLo
[e
][component
];
422 vCullComp
= vClipCullDistHi
[e
][component
];
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
);
429 vClipCullMask
= _simd16_or_ps(vClipCullMask
, vCullMaskElem
);
432 // clipper should also discard any primitive with NAN clip distance
433 uint8_t clipMask
= this->state
.rastState
.clipDistanceMask
;
434 while (_BitScanForward(&index
, clipMask
))
436 clipMask
&= ~(1 << index
);
437 uint32_t slot
= index
>> 2;
438 uint32_t component
= index
& 0x3;
440 for (uint32_t e
= 0; e
< NumVertsPerPrim
; ++e
)
442 simd16scalar vClipComp
;
445 vClipComp
= vClipCullDistLo
[e
][component
];
449 vClipComp
= vClipCullDistHi
[e
][component
];
452 simd16scalar vClip
= _simd16_cmp_ps(vClipComp
, vClipComp
, _CMP_UNORD_Q
);
453 vClipCullMask
= _simd16_or_ps(vClipCullMask
, vClip
);
457 return _simd16_movemask_ps(vClipCullMask
);
461 // clip SIMD primitives
462 void ClipSimd(const simdscalar
& vPrimMask
, const simdscalar
& vClipMask
, PA_STATE
& pa
, const simdscalari
& vPrimId
, const simdscalari
& vViewportIdx
)
464 // input/output vertex store for clipper
465 simdvertex vertices
[7]; // maximum 7 verts generated per triangle
467 LONG constantInterpMask
= this->state
.backendState
.constantInterpolationMask
;
468 uint32_t provokingVertex
= 0;
469 if(pa
.binTopology
== TOP_TRIANGLE_FAN
)
471 provokingVertex
= this->state
.frontendState
.provokingVertex
.triFan
;
473 ///@todo: line topology for wireframe?
476 simdvector tmpVector
[NumVertsPerPrim
];
477 pa
.Assemble(VERTEX_POSITION_SLOT
, tmpVector
);
478 for (uint32_t i
= 0; i
< NumVertsPerPrim
; ++i
)
480 vertices
[i
].attrib
[VERTEX_POSITION_SLOT
] = tmpVector
[i
];
484 const SWR_BACKEND_STATE
& backendState
= this->state
.backendState
;
486 int32_t maxSlot
= -1;
487 for (uint32_t slot
= 0; slot
< backendState
.numAttributes
; ++slot
)
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
;
494 pa
.Assemble(inputSlot
, tmpVector
);
496 // if constant interpolation enabled for this attribute, assign the provoking
497 // vertex values to all edges
498 if (_bittest(&constantInterpMask
, slot
))
500 for (uint32_t i
= 0; i
< NumVertsPerPrim
; ++i
)
502 vertices
[i
].attrib
[inputSlot
] = tmpVector
[provokingVertex
];
507 for (uint32_t i
= 0; i
< NumVertsPerPrim
; ++i
)
509 vertices
[i
].attrib
[inputSlot
] = tmpVector
[i
];
514 // assemble user clip distances if enabled
515 if (this->state
.rastState
.clipDistanceMask
& 0xf)
517 pa
.Assemble(VERTEX_CLIPCULL_DIST_LO_SLOT
, tmpVector
);
518 for (uint32_t i
= 0; i
< NumVertsPerPrim
; ++i
)
520 vertices
[i
].attrib
[VERTEX_CLIPCULL_DIST_LO_SLOT
] = tmpVector
[i
];
524 if (this->state
.rastState
.clipDistanceMask
& 0xf0)
526 pa
.Assemble(VERTEX_CLIPCULL_DIST_HI_SLOT
, tmpVector
);
527 for (uint32_t i
= 0; i
< NumVertsPerPrim
; ++i
)
529 vertices
[i
].attrib
[VERTEX_CLIPCULL_DIST_HI_SLOT
] = tmpVector
[i
];
533 uint32_t numAttribs
= maxSlot
+ 1;
535 simdscalari vNumClippedVerts
= ClipPrims((float*)&vertices
[0], vPrimMask
, vClipMask
, numAttribs
);
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)
542 pfnBinFunc
= GetBinTrianglesFunc((pa
.pDC
->pState
->state
.rastState
.conservativeRast
> 0));
543 clipTopology
= TOP_TRIANGLE_FAN
;
545 // so that the binner knows to bloat wide points later
546 if (pa
.binTopology
== TOP_POINT_LIST
)
547 clipTopology
= TOP_POINT_LIST
;
550 else if (NumVertsPerPrim
== 2)
552 pfnBinFunc
= BinLines
;
553 clipTopology
= TOP_LINE_LIST
;
557 SWR_ASSERT(0 && "Unexpected points in clipper.");
560 uint32_t* pVertexCount
= (uint32_t*)&vNumClippedVerts
;
561 uint32_t* pPrimitiveId
= (uint32_t*)&vPrimId
;
562 uint32_t* pViewportIdx
= (uint32_t*)&vViewportIdx
;
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
));
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);
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
);
583 SWR_ASSERT(numPrims
<= numPrims_lo
);
585 for (uint32_t inputPrim
= 0; inputPrim
< numPrims_lo
; ++inputPrim
)
587 for (uint32_t inputPrim
= 0; inputPrim
< pa
.NumPrims(); ++inputPrim
)
590 uint32_t numEmittedVerts
= pVertexCount
[inputPrim
];
591 if (numEmittedVerts
< NumVertsPerPrim
)
595 SWR_ASSERT(numEmittedVerts
<= 7, "Unexpected vertex count from clipper.");
597 uint32_t numEmittedPrims
= GetNumPrims(clipTopology
, numEmittedVerts
);
598 numClippedPrims
+= numEmittedPrims
;
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
603 #if USE_SIMD16_FRONTEND
604 simd16vertex transposedPrims
[2];
606 simdvertex transposedPrims
[2];
610 uint8_t* pBase
= (uint8_t*)(&vertices
[0].attrib
[VERTEX_POSITION_SLOT
]) + sizeof(float) * inputPrim
;
612 #if USE_SIMD16_FRONTEND
613 // TEMPORARY WORKAROUND for bizarre VS2015 code-gen bug
614 static const float *dummy
= reinterpret_cast<const float *>(pBase
);
617 for (uint32_t c
= 0; c
< 4; ++c
)
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);
623 transposedPrims
[0].attrib
[VERTEX_POSITION_SLOT
][c
] = _simd_mask_i32gather_ps(_mm256_undefined_ps(), (const float*)pBase
, vOffsets
, vMask
, 1);
625 pBase
+= sizeof(simdscalar
);
629 pBase
= (uint8_t*)(&vertices
[0].attrib
[VERTEX_ATTRIB_START_SLOT
]) + sizeof(float) * inputPrim
;
630 for (uint32_t attrib
= 0; attrib
< numAttribs
; ++attrib
)
632 uint32_t attribSlot
= VERTEX_ATTRIB_START_SLOT
+ attrib
;
633 for (uint32_t c
= 0; c
< 4; ++c
)
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);
639 transposedPrims
[0].attrib
[attribSlot
][c
] = _simd_mask_i32gather_ps(_mm256_undefined_ps(), (const float*)pBase
, vOffsets
, vMask
, 1);
641 pBase
+= sizeof(simdscalar
);
645 // transpose user clip distances if enabled
646 if (this->state
.rastState
.clipDistanceMask
& 0xf)
648 pBase
= (uint8_t*)(&vertices
[0].attrib
[VERTEX_CLIPCULL_DIST_LO_SLOT
]) + sizeof(float) * inputPrim
;
649 for (uint32_t c
= 0; c
< 4; ++c
)
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);
655 transposedPrims
[0].attrib
[VERTEX_CLIPCULL_DIST_LO_SLOT
][c
] = _simd_mask_i32gather_ps(_mm256_undefined_ps(), (const float*)pBase
, vOffsets
, vMask
, 1);
657 pBase
+= sizeof(simdscalar
);
661 if (this->state
.rastState
.clipDistanceMask
& 0xf0)
663 pBase
= (uint8_t*)(&vertices
[0].attrib
[VERTEX_CLIPCULL_DIST_HI_SLOT
]) + sizeof(float) * inputPrim
;
664 for (uint32_t c
= 0; c
< 4; ++c
)
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);
670 transposedPrims
[0].attrib
[VERTEX_CLIPCULL_DIST_HI_SLOT
][c
] = _simd_mask_i32gather_ps(_mm256_undefined_ps(), (const float*)pBase
, vOffsets
, vMask
, 1);
672 pBase
+= sizeof(simdscalar
);
676 PA_STATE_OPT
clipPa(this->pDC
, numEmittedPrims
, (uint8_t*)&transposedPrims
[0], numEmittedVerts
, true, clipTopology
);
678 while (clipPa
.GetNextStreamOutput())
682 #if USE_SIMD16_FRONTEND
683 simd16vector attrib_simd16
[NumVertsPerPrim
];
684 bool assemble
= clipPa
.Assemble_simd16(VERTEX_POSITION_SLOT
, attrib_simd16
);
688 static const uint32_t primMaskMap
[] = { 0x0, 0x1, 0x3, 0x7, 0xf, 0x1f, 0x3f, 0x7f, 0xff };
690 simdvector attrib
[NumVertsPerPrim
];
691 for (uint32_t i
= 0; i
< NumVertsPerPrim
; i
+= 1)
693 for (uint32_t j
= 0; j
< 4; j
+= 1)
695 attrib
[i
][j
] = _simd16_extract_ps(attrib_simd16
[i
][j
], 0);
699 clipPa
.useAlternateOffset
= false;
700 pfnBinFunc(this->pDC
, clipPa
, this->workerId
, attrib
, primMaskMap
[numEmittedPrims
], _simd_set1_epi32(pPrimitiveId
[inputPrim
]), _simd_set1_epi32(pViewportIdx
[inputPrim
]));
703 simdvector attrib
[NumVertsPerPrim
];
704 bool assemble
= clipPa
.Assemble(VERTEX_POSITION_SLOT
, attrib
);
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
]));
711 } while (clipPa
.NextPrim());
715 // update global pipeline stat
716 UPDATE_STAT_FE(CPrimitives
, numClippedPrims
);
719 #if USE_SIMD16_FRONTEND
720 void ClipSimd(const simd16scalar
& vPrimMask
, const simd16scalar
& vClipMask
, PA_STATE
& pa
, const simd16scalari
& vPrimId
, const simd16scalari
& vViewportIdx
)
722 // input/output vertex store for clipper
723 simd16vertex vertices
[7]; // maximum 7 verts generated per triangle
725 LONG constantInterpMask
= this->state
.backendState
.constantInterpolationMask
;
726 uint32_t provokingVertex
= 0;
727 if (pa
.binTopology
== TOP_TRIANGLE_FAN
)
729 provokingVertex
= this->state
.frontendState
.provokingVertex
.triFan
;
731 ///@todo: line topology for wireframe?
734 simd16vector tmpVector
[NumVertsPerPrim
];
735 pa
.Assemble_simd16(VERTEX_POSITION_SLOT
, tmpVector
);
736 for (uint32_t i
= 0; i
< NumVertsPerPrim
; ++i
)
738 vertices
[i
].attrib
[VERTEX_POSITION_SLOT
] = tmpVector
[i
];
742 const SWR_BACKEND_STATE
& backendState
= this->state
.backendState
;
744 int32_t maxSlot
= -1;
745 for (uint32_t slot
= 0; slot
< backendState
.numAttributes
; ++slot
)
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
;
752 pa
.Assemble_simd16(inputSlot
, tmpVector
);
754 // if constant interpolation enabled for this attribute, assign the provoking
755 // vertex values to all edges
756 if (_bittest(&constantInterpMask
, slot
))
758 for (uint32_t i
= 0; i
< NumVertsPerPrim
; ++i
)
760 vertices
[i
].attrib
[inputSlot
] = tmpVector
[provokingVertex
];
765 for (uint32_t i
= 0; i
< NumVertsPerPrim
; ++i
)
767 vertices
[i
].attrib
[inputSlot
] = tmpVector
[i
];
772 // assemble user clip distances if enabled
773 if (this->state
.rastState
.clipDistanceMask
& 0xf)
775 pa
.Assemble_simd16(VERTEX_CLIPCULL_DIST_LO_SLOT
, tmpVector
);
776 for (uint32_t i
= 0; i
< NumVertsPerPrim
; ++i
)
778 vertices
[i
].attrib
[VERTEX_CLIPCULL_DIST_LO_SLOT
] = tmpVector
[i
];
782 if (this->state
.rastState
.clipDistanceMask
& 0xf0)
784 pa
.Assemble_simd16(VERTEX_CLIPCULL_DIST_HI_SLOT
, tmpVector
);
785 for (uint32_t i
= 0; i
< NumVertsPerPrim
; ++i
)
787 vertices
[i
].attrib
[VERTEX_CLIPCULL_DIST_HI_SLOT
] = tmpVector
[i
];
791 uint32_t numAttribs
= maxSlot
+ 1;
793 simd16scalari vNumClippedVerts
= ClipPrims((float*)&vertices
[0], vPrimMask
, vClipMask
, numAttribs
);
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)
800 pfnBinFunc
= GetBinTrianglesFunc_simd16((pa
.pDC
->pState
->state
.rastState
.conservativeRast
> 0));
801 clipTopology
= TOP_TRIANGLE_FAN
;
803 // so that the binner knows to bloat wide points later
804 if (pa
.binTopology
== TOP_POINT_LIST
)
805 clipTopology
= TOP_POINT_LIST
;
808 else if (NumVertsPerPrim
== 2)
810 pfnBinFunc
= BinLines_simd16
;
811 clipTopology
= TOP_LINE_LIST
;
815 SWR_ASSERT(0 && "Unexpected points in clipper.");
818 uint32_t* pVertexCount
= (uint32_t*)&vNumClippedVerts
;
819 uint32_t* pPrimitiveId
= (uint32_t*)&vPrimId
;
820 uint32_t* pViewportIdx
= (uint32_t*)&vViewportIdx
;
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
));
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);
836 uint32_t numClippedPrims
= 0;
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
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));
847 simd16vertex transposedPrims
[2];
850 for (uint32_t inputPrim
= 0; inputPrim
< pa
.NumPrims(); ++inputPrim
)
852 uint32_t numEmittedVerts
= pVertexCount
[inputPrim
];
853 if (numEmittedVerts
< NumVertsPerPrim
)
857 SWR_ASSERT(numEmittedVerts
<= 7, "Unexpected vertex count from clipper.");
859 uint32_t numEmittedPrims
= GetNumPrims(clipTopology
, numEmittedVerts
);
860 numClippedPrims
+= numEmittedPrims
;
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
867 uint8_t* pBase
= (uint8_t*)(&vertices
[0].attrib
[VERTEX_POSITION_SLOT
]) + sizeof(float) * inputPrim
;
870 // TEMPORARY WORKAROUND for bizarre VS2015 code-gen bug
871 static const float *dummy
= reinterpret_cast<const float *>(pBase
);
874 for (uint32_t c
= 0; c
< 4; ++c
)
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
);
882 pBase
= (uint8_t*)(&vertices
[0].attrib
[VERTEX_ATTRIB_START_SLOT
]) + sizeof(float) * inputPrim
;
883 for (uint32_t attrib
= 0; attrib
< numAttribs
; ++attrib
)
885 uint32_t attribSlot
= VERTEX_ATTRIB_START_SLOT
+ attrib
;
886 for (uint32_t c
= 0; c
< 4; ++c
)
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
);
894 // transpose user clip distances if enabled
895 if (this->state
.rastState
.clipDistanceMask
& 0xf)
897 pBase
= (uint8_t*)(&vertices
[0].attrib
[VERTEX_CLIPCULL_DIST_LO_SLOT
]) + sizeof(float) * inputPrim
;
898 for (uint32_t c
= 0; c
< 4; ++c
)
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
);
906 if (this->state
.rastState
.clipDistanceMask
& 0xf0)
908 pBase
= (uint8_t*)(&vertices
[0].attrib
[VERTEX_CLIPCULL_DIST_HI_SLOT
]) + sizeof(float) * inputPrim
;
909 for (uint32_t c
= 0; c
< 4; ++c
)
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
);
917 PA_STATE_OPT
clipPa(this->pDC
, numEmittedPrims
, (uint8_t*)&transposedPrims
[0], numEmittedVerts
, true, clipTopology
);
919 while (clipPa
.GetNextStreamOutput())
923 simd16vector attrib
[NumVertsPerPrim
];
924 bool assemble
= clipPa
.Assemble_simd16(VERTEX_POSITION_SLOT
, attrib
);
928 static const uint32_t primMaskMap
[] = { 0x0, 0x1, 0x3, 0x7, 0xf, 0x1f, 0x3f, 0x7f, 0xff, 0x1ff, 0x3ff, 0x7ff, 0xfff, 0x1fff, 0x3fff, 0x7fff, 0xffff };
930 clipPa
.useAlternateOffset
= false;
931 pfnBinFunc(this->pDC
, clipPa
, this->workerId
, attrib
, primMaskMap
[numEmittedPrims
], _simd16_set1_epi32(pPrimitiveId
[inputPrim
]), _simd16_set1_epi32(pViewportIdx
[inputPrim
]));
934 } while (clipPa
.NextPrim());
939 free(transposedPrims
);
942 // update global pipeline stat
943 UPDATE_STAT_FE(CPrimitives
, numClippedPrims
);
947 // execute the clipper stage
948 void ExecuteStage(PA_STATE
& pa
, simdvector prim
[], uint32_t primMask
, simdscalari primId
, simdscalari viewportIdx
)
950 SWR_ASSERT(this->pDC
!= nullptr);
951 SWR_CONTEXT
* pContext
= this->pDC
->pContext
;
952 const API_STATE
& apiState
= this->pDC
->pState
->state
;
954 // set up binner based on PA state
955 PFN_PROCESS_PRIMS pfnBinner
;
956 switch (pa
.binTopology
)
959 pfnBinner
= BinPoints
;
964 case TOP_LINE_LIST_ADJ
:
965 case TOP_LISTSTRIP_ADJ
:
966 pfnBinner
= BinLines
;
969 pfnBinner
= GetBinTrianglesFunc((apiState
.rastState
.conservativeRast
> 0));
973 // update clipper invocations pipeline stat
974 uint32_t numInvoc
= _mm_popcnt_u32(primMask
);
975 UPDATE_STAT_FE(CInvocations
, numInvoc
);
977 ComputeClipCodes(prim
, viewportIdx
);
979 // cull prims with NAN coords
980 primMask
&= ~ComputeNaNMask(prim
);
982 // user cull distance cull
983 if (this->state
.rastState
.cullDistanceMask
)
985 primMask
&= ~ComputeUserClipCullMask(pa
, prim
);
988 // cull prims outside view frustum
989 simdscalar clipIntersection
= ComputeClipCodeIntersection();
990 int validMask
= primMask
& _simd_movemask_ps(_simd_cmpeq_ps(clipIntersection
, _simd_setzero_ps()));
992 // skip clipping for points
993 uint32_t clipMask
= 0;
994 if (NumVertsPerPrim
!= 1)
996 clipMask
= primMask
& ComputeClipMask();
1001 AR_BEGIN(FEGuardbandClip
, pa
.pDC
->drawId
);
1002 // we have to clip tris, execute the clipper, which will also
1004 ClipSimd(vMask(primMask
), vMask(clipMask
), pa
, primId
, viewportIdx
);
1005 AR_END(FEGuardbandClip
, 1);
1009 // update CPrimitives pipeline state
1010 UPDATE_STAT_FE(CPrimitives
, _mm_popcnt_u32(validMask
));
1012 // forward valid prims directly to binner
1013 pfnBinner(this->pDC
, pa
, this->workerId
, prim
, validMask
, primId
, viewportIdx
);
1017 #if USE_SIMD16_FRONTEND
1018 void ExecuteStage(PA_STATE
& pa
, simd16vector prim
[], uint32_t primMask
, simd16scalari primId
, simd16scalari viewportIdx
)
1020 SWR_ASSERT(pa
.pDC
!= nullptr);
1021 SWR_CONTEXT
* pContext
= pa
.pDC
->pContext
;
1023 // set up binner based on PA state
1024 PFN_PROCESS_PRIMS_SIMD16 pfnBinner
;
1025 switch (pa
.binTopology
)
1027 case TOP_POINT_LIST
:
1028 pfnBinner
= BinPoints_simd16
;
1031 case TOP_LINE_STRIP
:
1033 case TOP_LINE_LIST_ADJ
:
1034 case TOP_LISTSTRIP_ADJ
:
1035 pfnBinner
= BinLines_simd16
;
1038 pfnBinner
= GetBinTrianglesFunc_simd16((pa
.pDC
->pState
->state
.rastState
.conservativeRast
> 0));
1042 // update clipper invocations pipeline stat
1043 uint32_t numInvoc
= _mm_popcnt_u32(primMask
);
1044 UPDATE_STAT_FE(CInvocations
, numInvoc
);
1046 ComputeClipCodes(prim
, viewportIdx
);
1048 // cull prims with NAN coords
1049 primMask
&= ~ComputeNaNMask(prim
);
1051 // user cull distance cull
1052 if (this->state
.rastState
.cullDistanceMask
)
1054 primMask
&= ~ComputeUserClipCullMask(pa
, prim
);
1057 // cull prims outside view frustum
1058 simd16scalar clipIntersection
= ComputeClipCodeIntersection_simd16();
1059 int validMask
= primMask
& _simd16_movemask_ps(_simd16_cmpeq_ps(clipIntersection
, _simd16_setzero_ps()));
1061 // skip clipping for points
1062 uint32_t clipMask
= 0;
1063 if (NumVertsPerPrim
!= 1)
1065 clipMask
= primMask
& ComputeClipMask_simd16();
1070 AR_BEGIN(FEGuardbandClip
, pa
.pDC
->drawId
);
1071 // we have to clip tris, execute the clipper, which will also
1073 ClipSimd(vMask16(primMask
), vMask16(clipMask
), pa
, primId
, viewportIdx
);
1074 AR_END(FEGuardbandClip
, 1);
1078 // update CPrimitives pipeline state
1079 UPDATE_STAT_FE(CPrimitives
, _mm_popcnt_u32(validMask
));
1081 // forward valid prims directly to binner
1082 pfnBinner(this->pDC
, pa
, this->workerId
, prim
, validMask
, primId
, viewportIdx
);
1088 inline simdscalar
ComputeInterpFactor(simdscalar boundaryCoord0
, simdscalar boundaryCoord1
)
1090 return _simd_div_ps(boundaryCoord0
, _simd_sub_ps(boundaryCoord0
, boundaryCoord1
));
1093 #if USE_SIMD16_FRONTEND
1094 inline simd16scalar
ComputeInterpFactor(simd16scalar boundaryCoord0
, simd16scalar boundaryCoord1
)
1096 return _simd16_div_ps(boundaryCoord0
, _simd16_sub_ps(boundaryCoord0
, boundaryCoord1
));
1100 inline simdscalari
ComputeOffsets(uint32_t attrib
, simdscalari vIndices
, uint32_t component
)
1102 const uint32_t simdVertexStride
= sizeof(simdvertex
);
1103 const uint32_t componentStride
= sizeof(simdscalar
);
1104 const uint32_t attribStride
= sizeof(simdvector
);
1105 const __m256i vElemOffset
= _mm256_set_epi32(7 * sizeof(float), 6 * sizeof(float), 5 * sizeof(float), 4 * sizeof(float),
1106 3 * sizeof(float), 2 * sizeof(float), 1 * sizeof(float), 0 * sizeof(float));
1108 // step to the simdvertex
1109 simdscalari vOffsets
= _simd_mullo_epi32(vIndices
, _simd_set1_epi32(simdVertexStride
));
1111 // step to the attribute and component
1112 vOffsets
= _simd_add_epi32(vOffsets
, _simd_set1_epi32(attribStride
* attrib
+ componentStride
* component
));
1115 vOffsets
= _simd_add_epi32(vOffsets
, vElemOffset
);
1120 #if USE_SIMD16_FRONTEND
1121 inline simd16scalari
ComputeOffsets(uint32_t attrib
, simd16scalari vIndices
, uint32_t component
)
1123 const uint32_t simdVertexStride
= sizeof(simd16vertex
);
1124 const uint32_t componentStride
= sizeof(simd16scalar
);
1125 const uint32_t attribStride
= sizeof(simd16vector
);
1126 const simd16scalari vElemOffset
= _simd16_set_epi32(
1127 15 * sizeof(float), 14 * sizeof(float), 13 * sizeof(float), 12 * sizeof(float),
1128 11 * sizeof(float), 10 * sizeof(float), 9 * sizeof(float), 8 * sizeof(float),
1129 7 * sizeof(float), 6 * sizeof(float), 5 * sizeof(float), 4 * sizeof(float),
1130 3 * sizeof(float), 2 * sizeof(float), 1 * sizeof(float), 0 * sizeof(float));
1132 // step to the simdvertex
1133 simd16scalari vOffsets
= _simd16_mullo_epi32(vIndices
, _simd16_set1_epi32(simdVertexStride
));
1135 // step to the attribute and component
1136 vOffsets
= _simd16_add_epi32(vOffsets
, _simd16_set1_epi32(attribStride
* attrib
+ componentStride
* component
));
1139 vOffsets
= _simd16_add_epi32(vOffsets
, vElemOffset
);
1145 // gathers a single component for a given attribute for each SIMD lane
1146 inline simdscalar
GatherComponent(const float* pBuffer
, uint32_t attrib
, simdscalar vMask
, simdscalari vIndices
, uint32_t component
)
1148 simdscalari vOffsets
= ComputeOffsets(attrib
, vIndices
, component
);
1149 simdscalar vSrc
= _mm256_undefined_ps();
1150 return _simd_mask_i32gather_ps(vSrc
, pBuffer
, vOffsets
, vMask
, 1);
1153 #if USE_SIMD16_FRONTEND
1154 inline simd16scalar
GatherComponent(const float* pBuffer
, uint32_t attrib
, simd16scalar vMask
, simd16scalari vIndices
, uint32_t component
)
1156 simd16scalari vOffsets
= ComputeOffsets(attrib
, vIndices
, component
);
1157 simd16scalar vSrc
= _simd16_setzero_ps();
1158 return _simd16_mask_i32gather_ps(vSrc
, pBuffer
, vOffsets
, _simd16_castps_si(vMask
), 1);
1162 inline void ScatterComponent(const float* pBuffer
, uint32_t attrib
, simdscalar vMask
, simdscalari vIndices
, uint32_t component
, simdscalar vSrc
)
1164 simdscalari vOffsets
= ComputeOffsets(attrib
, vIndices
, component
);
1166 uint32_t* pOffsets
= (uint32_t*)&vOffsets
;
1167 float* pSrc
= (float*)&vSrc
;
1168 uint32_t mask
= _simd_movemask_ps(vMask
);
1170 while (_BitScanForward(&lane
, mask
))
1172 mask
&= ~(1 << lane
);
1173 uint8_t* pBuf
= (uint8_t*)pBuffer
+ pOffsets
[lane
];
1174 *(float*)pBuf
= pSrc
[lane
];
1178 #if USE_SIMD16_FRONTEND
1179 inline void ScatterComponent(const float* pBuffer
, uint32_t attrib
, simd16scalar vMask
, simd16scalari vIndices
, uint32_t component
, simd16scalar vSrc
)
1181 simd16scalari vOffsets
= ComputeOffsets(attrib
, vIndices
, component
);
1183 uint32_t* pOffsets
= (uint32_t*)&vOffsets
;
1184 float* pSrc
= (float*)&vSrc
;
1185 uint32_t mask
= _simd16_movemask_ps(vMask
);
1187 while (_BitScanForward(&lane
, mask
))
1189 mask
&= ~(1 << lane
);
1190 uint8_t* pBuf
= (uint8_t*)pBuffer
+ pOffsets
[lane
];
1191 *(float*)pBuf
= pSrc
[lane
];
1196 template<SWR_CLIPCODES ClippingPlane
>
1197 inline void intersect(
1198 const simdscalar
& vActiveMask
, // active lanes to operate on
1199 const simdscalari
& s
, // index to first edge vertex v0 in pInPts.
1200 const simdscalari
& p
, // index to second edge vertex v1 in pInPts.
1201 const simdvector
& v1
, // vertex 0 position
1202 const simdvector
& v2
, // vertex 1 position
1203 simdscalari
& outIndex
, // output index.
1204 const float *pInVerts
, // array of all the input positions.
1205 uint32_t numInAttribs
, // number of attributes per vertex.
1206 float *pOutVerts
) // array of output positions. We'll write our new intersection point at i*4.
1208 // compute interpolation factor
1210 switch (ClippingPlane
)
1212 case FRUSTUM_LEFT
: t
= ComputeInterpFactor(_simd_add_ps(v1
[3], v1
[0]), _simd_add_ps(v2
[3], v2
[0])); break;
1213 case FRUSTUM_RIGHT
: t
= ComputeInterpFactor(_simd_sub_ps(v1
[3], v1
[0]), _simd_sub_ps(v2
[3], v2
[0])); break;
1214 case FRUSTUM_TOP
: t
= ComputeInterpFactor(_simd_add_ps(v1
[3], v1
[1]), _simd_add_ps(v2
[3], v2
[1])); break;
1215 case FRUSTUM_BOTTOM
: t
= ComputeInterpFactor(_simd_sub_ps(v1
[3], v1
[1]), _simd_sub_ps(v2
[3], v2
[1])); break;
1217 // DX Znear plane is 0, GL is -w
1218 if (this->state
.rastState
.clipHalfZ
)
1220 t
= ComputeInterpFactor(v1
[2], v2
[2]);
1224 t
= ComputeInterpFactor(_simd_add_ps(v1
[3], v1
[2]), _simd_add_ps(v2
[3], v2
[2]));
1227 case FRUSTUM_FAR
: t
= ComputeInterpFactor(_simd_sub_ps(v1
[3], v1
[2]), _simd_sub_ps(v2
[3], v2
[2])); break;
1228 default: SWR_INVALID("invalid clipping plane: %d", ClippingPlane
);
1231 // interpolate position and store
1232 for (uint32_t c
= 0; c
< 4; ++c
)
1234 simdscalar vOutPos
= _simd_fmadd_ps(_simd_sub_ps(v2
[c
], v1
[c
]), t
, v1
[c
]);
1235 ScatterComponent(pOutVerts
, VERTEX_POSITION_SLOT
, vActiveMask
, outIndex
, c
, vOutPos
);
1238 // interpolate attributes and store
1239 for (uint32_t a
= 0; a
< numInAttribs
; ++a
)
1241 uint32_t attribSlot
= VERTEX_ATTRIB_START_SLOT
+ a
;
1242 for (uint32_t c
= 0; c
< 4; ++c
)
1244 simdscalar vAttrib0
= GatherComponent(pInVerts
, attribSlot
, vActiveMask
, s
, c
);
1245 simdscalar vAttrib1
= GatherComponent(pInVerts
, attribSlot
, vActiveMask
, p
, c
);
1246 simdscalar vOutAttrib
= _simd_fmadd_ps(_simd_sub_ps(vAttrib1
, vAttrib0
), t
, vAttrib0
);
1247 ScatterComponent(pOutVerts
, attribSlot
, vActiveMask
, outIndex
, c
, vOutAttrib
);
1251 // interpolate clip distance if enabled
1252 if (this->state
.rastState
.clipDistanceMask
& 0xf)
1254 uint32_t attribSlot
= VERTEX_CLIPCULL_DIST_LO_SLOT
;
1255 for (uint32_t c
= 0; c
< 4; ++c
)
1257 simdscalar vAttrib0
= GatherComponent(pInVerts
, attribSlot
, vActiveMask
, s
, c
);
1258 simdscalar vAttrib1
= GatherComponent(pInVerts
, attribSlot
, vActiveMask
, p
, c
);
1259 simdscalar vOutAttrib
= _simd_fmadd_ps(_simd_sub_ps(vAttrib1
, vAttrib0
), t
, vAttrib0
);
1260 ScatterComponent(pOutVerts
, attribSlot
, vActiveMask
, outIndex
, c
, vOutAttrib
);
1264 if (this->state
.rastState
.clipDistanceMask
& 0xf0)
1266 uint32_t attribSlot
= VERTEX_CLIPCULL_DIST_HI_SLOT
;
1267 for (uint32_t c
= 0; c
< 4; ++c
)
1269 simdscalar vAttrib0
= GatherComponent(pInVerts
, attribSlot
, vActiveMask
, s
, c
);
1270 simdscalar vAttrib1
= GatherComponent(pInVerts
, attribSlot
, vActiveMask
, p
, c
);
1271 simdscalar vOutAttrib
= _simd_fmadd_ps(_simd_sub_ps(vAttrib1
, vAttrib0
), t
, vAttrib0
);
1272 ScatterComponent(pOutVerts
, attribSlot
, vActiveMask
, outIndex
, c
, vOutAttrib
);
1277 #if USE_SIMD16_FRONTEND
1278 template<SWR_CLIPCODES ClippingPlane
>
1279 inline void intersect(
1280 const simd16scalar
& vActiveMask
,// active lanes to operate on
1281 const simd16scalari
& s
, // index to first edge vertex v0 in pInPts.
1282 const simd16scalari
& p
, // index to second edge vertex v1 in pInPts.
1283 const simd16vector
& v1
, // vertex 0 position
1284 const simd16vector
& v2
, // vertex 1 position
1285 simd16scalari
& outIndex
, // output index.
1286 const float *pInVerts
, // array of all the input positions.
1287 uint32_t numInAttribs
, // number of attributes per vertex.
1288 float *pOutVerts
) // array of output positions. We'll write our new intersection point at i*4.
1290 // compute interpolation factor
1292 switch (ClippingPlane
)
1294 case FRUSTUM_LEFT
: t
= ComputeInterpFactor(_simd16_add_ps(v1
[3], v1
[0]), _simd16_add_ps(v2
[3], v2
[0])); break;
1295 case FRUSTUM_RIGHT
: t
= ComputeInterpFactor(_simd16_sub_ps(v1
[3], v1
[0]), _simd16_sub_ps(v2
[3], v2
[0])); break;
1296 case FRUSTUM_TOP
: t
= ComputeInterpFactor(_simd16_add_ps(v1
[3], v1
[1]), _simd16_add_ps(v2
[3], v2
[1])); break;
1297 case FRUSTUM_BOTTOM
: t
= ComputeInterpFactor(_simd16_sub_ps(v1
[3], v1
[1]), _simd16_sub_ps(v2
[3], v2
[1])); break;
1299 // DX Znear plane is 0, GL is -w
1300 if (this->state
.rastState
.clipHalfZ
)
1302 t
= ComputeInterpFactor(v1
[2], v2
[2]);
1306 t
= ComputeInterpFactor(_simd16_add_ps(v1
[3], v1
[2]), _simd16_add_ps(v2
[3], v2
[2]));
1309 case FRUSTUM_FAR
: t
= ComputeInterpFactor(_simd16_sub_ps(v1
[3], v1
[2]), _simd16_sub_ps(v2
[3], v2
[2])); break;
1310 default: SWR_INVALID("invalid clipping plane: %d", ClippingPlane
);
1313 // interpolate position and store
1314 for (uint32_t c
= 0; c
< 4; ++c
)
1316 simd16scalar vOutPos
= _simd16_fmadd_ps(_simd16_sub_ps(v2
[c
], v1
[c
]), t
, v1
[c
]);
1317 ScatterComponent(pOutVerts
, VERTEX_POSITION_SLOT
, vActiveMask
, outIndex
, c
, vOutPos
);
1320 // interpolate attributes and store
1321 for (uint32_t a
= 0; a
< numInAttribs
; ++a
)
1323 uint32_t attribSlot
= VERTEX_ATTRIB_START_SLOT
+ a
;
1324 for (uint32_t c
= 0; c
< 4; ++c
)
1326 simd16scalar vAttrib0
= GatherComponent(pInVerts
, attribSlot
, vActiveMask
, s
, c
);
1327 simd16scalar vAttrib1
= GatherComponent(pInVerts
, attribSlot
, vActiveMask
, p
, c
);
1328 simd16scalar vOutAttrib
= _simd16_fmadd_ps(_simd16_sub_ps(vAttrib1
, vAttrib0
), t
, vAttrib0
);
1329 ScatterComponent(pOutVerts
, attribSlot
, vActiveMask
, outIndex
, c
, vOutAttrib
);
1333 // interpolate clip distance if enabled
1334 if (this->state
.rastState
.clipDistanceMask
& 0xf)
1336 uint32_t attribSlot
= VERTEX_CLIPCULL_DIST_LO_SLOT
;
1337 for (uint32_t c
= 0; c
< 4; ++c
)
1339 simd16scalar vAttrib0
= GatherComponent(pInVerts
, attribSlot
, vActiveMask
, s
, c
);
1340 simd16scalar vAttrib1
= GatherComponent(pInVerts
, attribSlot
, vActiveMask
, p
, c
);
1341 simd16scalar vOutAttrib
= _simd16_fmadd_ps(_simd16_sub_ps(vAttrib1
, vAttrib0
), t
, vAttrib0
);
1342 ScatterComponent(pOutVerts
, attribSlot
, vActiveMask
, outIndex
, c
, vOutAttrib
);
1346 if (this->state
.rastState
.clipDistanceMask
& 0xf0)
1348 uint32_t attribSlot
= VERTEX_CLIPCULL_DIST_HI_SLOT
;
1349 for (uint32_t c
= 0; c
< 4; ++c
)
1351 simd16scalar vAttrib0
= GatherComponent(pInVerts
, attribSlot
, vActiveMask
, s
, c
);
1352 simd16scalar vAttrib1
= GatherComponent(pInVerts
, attribSlot
, vActiveMask
, p
, c
);
1353 simd16scalar vOutAttrib
= _simd16_fmadd_ps(_simd16_sub_ps(vAttrib1
, vAttrib0
), t
, vAttrib0
);
1354 ScatterComponent(pOutVerts
, attribSlot
, vActiveMask
, outIndex
, c
, vOutAttrib
);
1360 template<SWR_CLIPCODES ClippingPlane
>
1361 inline simdscalar
inside(const simdvector
& v
)
1363 switch (ClippingPlane
)
1365 case FRUSTUM_LEFT
: return _simd_cmpge_ps(v
[0], _simd_mul_ps(v
[3], _simd_set1_ps(-1.0f
)));
1366 case FRUSTUM_RIGHT
: return _simd_cmple_ps(v
[0], v
[3]);
1367 case FRUSTUM_TOP
: return _simd_cmpge_ps(v
[1], _simd_mul_ps(v
[3], _simd_set1_ps(-1.0f
)));
1368 case FRUSTUM_BOTTOM
: return _simd_cmple_ps(v
[1], v
[3]);
1369 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
)));
1370 case FRUSTUM_FAR
: return _simd_cmple_ps(v
[2], v
[3]);
1372 SWR_INVALID("invalid clipping plane: %d", ClippingPlane
);
1373 return _simd_setzero_ps();
1377 #if USE_SIMD16_FRONTEND
1378 template<SWR_CLIPCODES ClippingPlane
>
1379 inline simd16scalar
inside(const simd16vector
& v
)
1381 switch (ClippingPlane
)
1383 case FRUSTUM_LEFT
: return _simd16_cmpge_ps(v
[0], _simd16_mul_ps(v
[3], _simd16_set1_ps(-1.0f
)));
1384 case FRUSTUM_RIGHT
: return _simd16_cmple_ps(v
[0], v
[3]);
1385 case FRUSTUM_TOP
: return _simd16_cmpge_ps(v
[1], _simd16_mul_ps(v
[3], _simd16_set1_ps(-1.0f
)));
1386 case FRUSTUM_BOTTOM
: return _simd16_cmple_ps(v
[1], v
[3]);
1387 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
)));
1388 case FRUSTUM_FAR
: return _simd16_cmple_ps(v
[2], v
[3]);
1390 SWR_INVALID("invalid clipping plane: %d", ClippingPlane
);
1391 return _simd16_setzero_ps();
1396 template<SWR_CLIPCODES ClippingPlane
>
1397 simdscalari
ClipTriToPlane(const float* pInVerts
, const simdscalari
& vNumInPts
, uint32_t numInAttribs
, float* pOutVerts
)
1399 simdscalari vCurIndex
= _simd_setzero_si();
1400 simdscalari vOutIndex
= _simd_setzero_si();
1401 simdscalar vActiveMask
= _simd_castsi_ps(_simd_cmplt_epi32(vCurIndex
, vNumInPts
));
1403 while (!_simd_testz_ps(vActiveMask
, vActiveMask
)) // loop until activeMask is empty
1405 simdscalari s
= vCurIndex
;
1406 simdscalari p
= _simd_add_epi32(s
, _simd_set1_epi32(1));
1407 simdscalari underFlowMask
= _simd_cmpgt_epi32(vNumInPts
, p
);
1408 p
= _simd_castps_si(_simd_blendv_ps(_simd_setzero_ps(), _simd_castsi_ps(p
), _simd_castsi_ps(underFlowMask
)));
1411 simdvector vInPos0
, vInPos1
;
1412 for (uint32_t c
= 0; c
< 4; ++c
)
1414 vInPos0
[c
] = GatherComponent(pInVerts
, VERTEX_POSITION_SLOT
, vActiveMask
, s
, c
);
1415 vInPos1
[c
] = GatherComponent(pInVerts
, VERTEX_POSITION_SLOT
, vActiveMask
, p
, c
);
1418 // compute inside mask
1419 simdscalar s_in
= inside
<ClippingPlane
>(vInPos0
);
1420 simdscalar p_in
= inside
<ClippingPlane
>(vInPos1
);
1422 // compute intersection mask (s_in != p_in)
1423 simdscalar intersectMask
= _simd_xor_ps(s_in
, p_in
);
1424 intersectMask
= _simd_and_ps(intersectMask
, vActiveMask
);
1426 // store s if inside
1427 s_in
= _simd_and_ps(s_in
, vActiveMask
);
1428 if (!_simd_testz_ps(s_in
, s_in
))
1431 for (uint32_t c
= 0; c
< 4; ++c
)
1433 ScatterComponent(pOutVerts
, VERTEX_POSITION_SLOT
, s_in
, vOutIndex
, c
, vInPos0
[c
]);
1437 for (uint32_t a
= 0; a
< numInAttribs
; ++a
)
1439 uint32_t attribSlot
= VERTEX_ATTRIB_START_SLOT
+ a
;
1440 for (uint32_t c
= 0; c
< 4; ++c
)
1442 simdscalar vAttrib
= GatherComponent(pInVerts
, attribSlot
, s_in
, s
, c
);
1443 ScatterComponent(pOutVerts
, attribSlot
, s_in
, vOutIndex
, c
, vAttrib
);
1447 // store clip distance if enabled
1448 if (this->state
.rastState
.clipDistanceMask
& 0xf)
1450 uint32_t attribSlot
= VERTEX_CLIPCULL_DIST_LO_SLOT
;
1451 for (uint32_t c
= 0; c
< 4; ++c
)
1453 simdscalar vAttrib
= GatherComponent(pInVerts
, attribSlot
, s_in
, s
, c
);
1454 ScatterComponent(pOutVerts
, attribSlot
, s_in
, vOutIndex
, c
, vAttrib
);
1458 if (this->state
.rastState
.clipDistanceMask
& 0xf0)
1460 uint32_t attribSlot
= VERTEX_CLIPCULL_DIST_HI_SLOT
;
1461 for (uint32_t c
= 0; c
< 4; ++c
)
1463 simdscalar vAttrib
= GatherComponent(pInVerts
, attribSlot
, s_in
, s
, c
);
1464 ScatterComponent(pOutVerts
, attribSlot
, s_in
, vOutIndex
, c
, vAttrib
);
1468 // increment outIndex
1469 vOutIndex
= _simd_blendv_epi32(vOutIndex
, _simd_add_epi32(vOutIndex
, _simd_set1_epi32(1)), s_in
);
1472 // compute and store intersection
1473 if (!_simd_testz_ps(intersectMask
, intersectMask
))
1475 intersect
<ClippingPlane
>(intersectMask
, s
, p
, vInPos0
, vInPos1
, vOutIndex
, pInVerts
, numInAttribs
, pOutVerts
);
1477 // increment outIndex for active lanes
1478 vOutIndex
= _simd_blendv_epi32(vOutIndex
, _simd_add_epi32(vOutIndex
, _simd_set1_epi32(1)), intersectMask
);
1481 // increment loop index and update active mask
1482 vCurIndex
= _simd_add_epi32(vCurIndex
, _simd_set1_epi32(1));
1483 vActiveMask
= _simd_castsi_ps(_simd_cmplt_epi32(vCurIndex
, vNumInPts
));
1489 #if USE_SIMD16_FRONTEND
1490 template<SWR_CLIPCODES ClippingPlane
>
1491 simd16scalari
ClipTriToPlane(const float* pInVerts
, const simd16scalari
& vNumInPts
, uint32_t numInAttribs
, float* pOutVerts
)
1493 simd16scalari vCurIndex
= _simd16_setzero_si();
1494 simd16scalari vOutIndex
= _simd16_setzero_si();
1495 simd16scalar vActiveMask
= _simd16_castsi_ps(_simd16_cmplt_epi32(vCurIndex
, vNumInPts
));
1497 while (!_simd16_testz_ps(vActiveMask
, vActiveMask
)) // loop until activeMask is empty
1499 simd16scalari s
= vCurIndex
;
1500 simd16scalari p
= _simd16_add_epi32(s
, _simd16_set1_epi32(1));
1501 simd16scalari underFlowMask
= _simd16_cmpgt_epi32(vNumInPts
, p
);
1502 p
= _simd16_castps_si(_simd16_blendv_ps(_simd16_setzero_ps(), _simd16_castsi_ps(p
), _simd16_castsi_ps(underFlowMask
)));
1505 simd16vector vInPos0
, vInPos1
;
1506 for (uint32_t c
= 0; c
< 4; ++c
)
1508 vInPos0
[c
] = GatherComponent(pInVerts
, VERTEX_POSITION_SLOT
, vActiveMask
, s
, c
);
1509 vInPos1
[c
] = GatherComponent(pInVerts
, VERTEX_POSITION_SLOT
, vActiveMask
, p
, c
);
1512 // compute inside mask
1513 simd16scalar s_in
= inside
<ClippingPlane
>(vInPos0
);
1514 simd16scalar p_in
= inside
<ClippingPlane
>(vInPos1
);
1516 // compute intersection mask (s_in != p_in)
1517 simd16scalar intersectMask
= _simd16_xor_ps(s_in
, p_in
);
1518 intersectMask
= _simd16_and_ps(intersectMask
, vActiveMask
);
1520 // store s if inside
1521 s_in
= _simd16_and_ps(s_in
, vActiveMask
);
1522 if (!_simd16_testz_ps(s_in
, s_in
))
1525 for (uint32_t c
= 0; c
< 4; ++c
)
1527 ScatterComponent(pOutVerts
, VERTEX_POSITION_SLOT
, s_in
, vOutIndex
, c
, vInPos0
[c
]);
1531 for (uint32_t a
= 0; a
< numInAttribs
; ++a
)
1533 uint32_t attribSlot
= VERTEX_ATTRIB_START_SLOT
+ a
;
1534 for (uint32_t c
= 0; c
< 4; ++c
)
1536 simd16scalar vAttrib
= GatherComponent(pInVerts
, attribSlot
, s_in
, s
, c
);
1537 ScatterComponent(pOutVerts
, attribSlot
, s_in
, vOutIndex
, c
, vAttrib
);
1541 // store clip distance if enabled
1542 if (this->state
.rastState
.clipDistanceMask
& 0xf)
1544 uint32_t attribSlot
= VERTEX_CLIPCULL_DIST_LO_SLOT
;
1545 for (uint32_t c
= 0; c
< 4; ++c
)
1547 simd16scalar vAttrib
= GatherComponent(pInVerts
, attribSlot
, s_in
, s
, c
);
1548 ScatterComponent(pOutVerts
, attribSlot
, s_in
, vOutIndex
, c
, vAttrib
);
1552 if (this->state
.rastState
.clipDistanceMask
& 0xf0)
1554 uint32_t attribSlot
= VERTEX_CLIPCULL_DIST_HI_SLOT
;
1555 for (uint32_t c
= 0; c
< 4; ++c
)
1557 simd16scalar vAttrib
= GatherComponent(pInVerts
, attribSlot
, s_in
, s
, c
);
1558 ScatterComponent(pOutVerts
, attribSlot
, s_in
, vOutIndex
, c
, vAttrib
);
1562 // increment outIndex
1563 vOutIndex
= _simd16_blendv_epi32(vOutIndex
, _simd16_add_epi32(vOutIndex
, _simd16_set1_epi32(1)), s_in
);
1566 // compute and store intersection
1567 if (!_simd16_testz_ps(intersectMask
, intersectMask
))
1569 intersect
<ClippingPlane
>(intersectMask
, s
, p
, vInPos0
, vInPos1
, vOutIndex
, pInVerts
, numInAttribs
, pOutVerts
);
1571 // increment outIndex for active lanes
1572 vOutIndex
= _simd16_blendv_epi32(vOutIndex
, _simd16_add_epi32(vOutIndex
, _simd16_set1_epi32(1)), intersectMask
);
1575 // increment loop index and update active mask
1576 vCurIndex
= _simd16_add_epi32(vCurIndex
, _simd16_set1_epi32(1));
1577 vActiveMask
= _simd16_castsi_ps(_simd16_cmplt_epi32(vCurIndex
, vNumInPts
));
1584 template<SWR_CLIPCODES ClippingPlane
>
1585 simdscalari
ClipLineToPlane(const float* pInVerts
, const simdscalari
& vNumInPts
, uint32_t numInAttribs
, float* pOutVerts
)
1587 simdscalari vCurIndex
= _simd_setzero_si();
1588 simdscalari vOutIndex
= _simd_setzero_si();
1589 simdscalar vActiveMask
= _simd_castsi_ps(_simd_cmplt_epi32(vCurIndex
, vNumInPts
));
1591 if (!_simd_testz_ps(vActiveMask
, vActiveMask
))
1593 simdscalari s
= vCurIndex
;
1594 simdscalari p
= _simd_add_epi32(s
, _simd_set1_epi32(1));
1597 simdvector vInPos0
, vInPos1
;
1598 for (uint32_t c
= 0; c
< 4; ++c
)
1600 vInPos0
[c
] = GatherComponent(pInVerts
, VERTEX_POSITION_SLOT
, vActiveMask
, s
, c
);
1601 vInPos1
[c
] = GatherComponent(pInVerts
, VERTEX_POSITION_SLOT
, vActiveMask
, p
, c
);
1604 // compute inside mask
1605 simdscalar s_in
= inside
<ClippingPlane
>(vInPos0
);
1606 simdscalar p_in
= inside
<ClippingPlane
>(vInPos1
);
1608 // compute intersection mask (s_in != p_in)
1609 simdscalar intersectMask
= _simd_xor_ps(s_in
, p_in
);
1610 intersectMask
= _simd_and_ps(intersectMask
, vActiveMask
);
1612 // store s if inside
1613 s_in
= _simd_and_ps(s_in
, vActiveMask
);
1614 if (!_simd_testz_ps(s_in
, s_in
))
1616 for (uint32_t c
= 0; c
< 4; ++c
)
1618 ScatterComponent(pOutVerts
, VERTEX_POSITION_SLOT
, s_in
, vOutIndex
, c
, vInPos0
[c
]);
1621 // interpolate attributes and store
1622 for (uint32_t a
= 0; a
< numInAttribs
; ++a
)
1624 uint32_t attribSlot
= VERTEX_ATTRIB_START_SLOT
+ a
;
1625 for (uint32_t c
= 0; c
< 4; ++c
)
1627 simdscalar vAttrib
= GatherComponent(pInVerts
, attribSlot
, s_in
, s
, c
);
1628 ScatterComponent(pOutVerts
, attribSlot
, s_in
, vOutIndex
, c
, vAttrib
);
1632 // increment outIndex
1633 vOutIndex
= _simd_blendv_epi32(vOutIndex
, _simd_add_epi32(vOutIndex
, _simd_set1_epi32(1)), s_in
);
1636 // compute and store intersection
1637 if (!_simd_testz_ps(intersectMask
, intersectMask
))
1639 intersect
<ClippingPlane
>(intersectMask
, s
, p
, vInPos0
, vInPos1
, vOutIndex
, pInVerts
, numInAttribs
, pOutVerts
);
1641 // increment outIndex for active lanes
1642 vOutIndex
= _simd_blendv_epi32(vOutIndex
, _simd_add_epi32(vOutIndex
, _simd_set1_epi32(1)), intersectMask
);
1645 // store p if inside
1646 p_in
= _simd_and_ps(p_in
, vActiveMask
);
1647 if (!_simd_testz_ps(p_in
, p_in
))
1649 for (uint32_t c
= 0; c
< 4; ++c
)
1651 ScatterComponent(pOutVerts
, VERTEX_POSITION_SLOT
, p_in
, vOutIndex
, c
, vInPos1
[c
]);
1654 // interpolate attributes and store
1655 for (uint32_t a
= 0; a
< numInAttribs
; ++a
)
1657 uint32_t attribSlot
= VERTEX_ATTRIB_START_SLOT
+ a
;
1658 for (uint32_t c
= 0; c
< 4; ++c
)
1660 simdscalar vAttrib
= GatherComponent(pInVerts
, attribSlot
, p_in
, p
, c
);
1661 ScatterComponent(pOutVerts
, attribSlot
, p_in
, vOutIndex
, c
, vAttrib
);
1665 // increment outIndex
1666 vOutIndex
= _simd_blendv_epi32(vOutIndex
, _simd_add_epi32(vOutIndex
, _simd_set1_epi32(1)), p_in
);
1673 #if USE_SIMD16_FRONTEND
1674 template<SWR_CLIPCODES ClippingPlane
>
1675 simd16scalari
ClipLineToPlane(const float* pInVerts
, const simd16scalari
& vNumInPts
, uint32_t numInAttribs
, float* pOutVerts
)
1677 simd16scalari vCurIndex
= _simd16_setzero_si();
1678 simd16scalari vOutIndex
= _simd16_setzero_si();
1679 simd16scalar vActiveMask
= _simd16_castsi_ps(_simd16_cmplt_epi32(vCurIndex
, vNumInPts
));
1681 if (!_simd16_testz_ps(vActiveMask
, vActiveMask
))
1683 simd16scalari s
= vCurIndex
;
1684 simd16scalari p
= _simd16_add_epi32(s
, _simd16_set1_epi32(1));
1687 simd16vector vInPos0
, vInPos1
;
1688 for (uint32_t c
= 0; c
< 4; ++c
)
1690 vInPos0
[c
] = GatherComponent(pInVerts
, VERTEX_POSITION_SLOT
, vActiveMask
, s
, c
);
1691 vInPos1
[c
] = GatherComponent(pInVerts
, VERTEX_POSITION_SLOT
, vActiveMask
, p
, c
);
1694 // compute inside mask
1695 simd16scalar s_in
= inside
<ClippingPlane
>(vInPos0
);
1696 simd16scalar p_in
= inside
<ClippingPlane
>(vInPos1
);
1698 // compute intersection mask (s_in != p_in)
1699 simd16scalar intersectMask
= _simd16_xor_ps(s_in
, p_in
);
1700 intersectMask
= _simd16_and_ps(intersectMask
, vActiveMask
);
1702 // store s if inside
1703 s_in
= _simd16_and_ps(s_in
, vActiveMask
);
1704 if (!_simd16_testz_ps(s_in
, s_in
))
1706 for (uint32_t c
= 0; c
< 4; ++c
)
1708 ScatterComponent(pOutVerts
, VERTEX_POSITION_SLOT
, s_in
, vOutIndex
, c
, vInPos0
[c
]);
1711 // interpolate attributes and store
1712 for (uint32_t a
= 0; a
< numInAttribs
; ++a
)
1714 uint32_t attribSlot
= VERTEX_ATTRIB_START_SLOT
+ a
;
1715 for (uint32_t c
= 0; c
< 4; ++c
)
1717 simd16scalar vAttrib
= GatherComponent(pInVerts
, attribSlot
, s_in
, s
, c
);
1718 ScatterComponent(pOutVerts
, attribSlot
, s_in
, vOutIndex
, c
, vAttrib
);
1722 // increment outIndex
1723 vOutIndex
= _simd16_blendv_epi32(vOutIndex
, _simd16_add_epi32(vOutIndex
, _simd16_set1_epi32(1)), s_in
);
1726 // compute and store intersection
1727 if (!_simd16_testz_ps(intersectMask
, intersectMask
))
1729 intersect
<ClippingPlane
>(intersectMask
, s
, p
, vInPos0
, vInPos1
, vOutIndex
, pInVerts
, numInAttribs
, pOutVerts
);
1731 // increment outIndex for active lanes
1732 vOutIndex
= _simd16_blendv_epi32(vOutIndex
, _simd16_add_epi32(vOutIndex
, _simd16_set1_epi32(1)), intersectMask
);
1735 // store p if inside
1736 p_in
= _simd16_and_ps(p_in
, vActiveMask
);
1737 if (!_simd16_testz_ps(p_in
, p_in
))
1739 for (uint32_t c
= 0; c
< 4; ++c
)
1741 ScatterComponent(pOutVerts
, VERTEX_POSITION_SLOT
, p_in
, vOutIndex
, c
, vInPos1
[c
]);
1744 // interpolate attributes and store
1745 for (uint32_t a
= 0; a
< numInAttribs
; ++a
)
1747 uint32_t attribSlot
= VERTEX_ATTRIB_START_SLOT
+ a
;
1748 for (uint32_t c
= 0; c
< 4; ++c
)
1750 simd16scalar vAttrib
= GatherComponent(pInVerts
, attribSlot
, p_in
, p
, c
);
1751 ScatterComponent(pOutVerts
, attribSlot
, p_in
, vOutIndex
, c
, vAttrib
);
1755 // increment outIndex
1756 vOutIndex
= _simd16_blendv_epi32(vOutIndex
, _simd16_add_epi32(vOutIndex
, _simd16_set1_epi32(1)), p_in
);
1763 //////////////////////////////////////////////////////////////////////////
1764 /// @brief Vertical clipper. Clips SIMD primitives at a time
1765 /// @param pVertices - pointer to vertices in SOA form. Clipper will read input and write results to this buffer
1766 /// @param vPrimMask - mask of valid input primitives, including non-clipped prims
1767 /// @param numAttribs - number of valid input attribs, including position
1768 simdscalari
ClipPrims(float* pVertices
, const simdscalar
& vPrimMask
, const simdscalar
& vClipMask
, int numAttribs
)
1771 float* pTempVerts
= (float*)&tlsTempVertices
[0];
1773 // zero out num input verts for non-active lanes
1774 simdscalari vNumInPts
= _simd_set1_epi32(NumVertsPerPrim
);
1775 vNumInPts
= _simd_blendv_epi32(_simd_setzero_si(), vNumInPts
, vClipMask
);
1777 // clip prims to frustum
1778 simdscalari vNumOutPts
;
1779 if (NumVertsPerPrim
== 3)
1781 vNumOutPts
= ClipTriToPlane
<FRUSTUM_NEAR
>(pVertices
, vNumInPts
, numAttribs
, pTempVerts
);
1782 vNumOutPts
= ClipTriToPlane
<FRUSTUM_FAR
>(pTempVerts
, vNumOutPts
, numAttribs
, pVertices
);
1783 vNumOutPts
= ClipTriToPlane
<FRUSTUM_LEFT
>(pVertices
, vNumOutPts
, numAttribs
, pTempVerts
);
1784 vNumOutPts
= ClipTriToPlane
<FRUSTUM_RIGHT
>(pTempVerts
, vNumOutPts
, numAttribs
, pVertices
);
1785 vNumOutPts
= ClipTriToPlane
<FRUSTUM_BOTTOM
>(pVertices
, vNumOutPts
, numAttribs
, pTempVerts
);
1786 vNumOutPts
= ClipTriToPlane
<FRUSTUM_TOP
>(pTempVerts
, vNumOutPts
, numAttribs
, pVertices
);
1790 SWR_ASSERT(NumVertsPerPrim
== 2);
1791 vNumOutPts
= ClipLineToPlane
<FRUSTUM_NEAR
>(pVertices
, vNumInPts
, numAttribs
, pTempVerts
);
1792 vNumOutPts
= ClipLineToPlane
<FRUSTUM_FAR
>(pTempVerts
, vNumOutPts
, numAttribs
, pVertices
);
1793 vNumOutPts
= ClipLineToPlane
<FRUSTUM_LEFT
>(pVertices
, vNumOutPts
, numAttribs
, pTempVerts
);
1794 vNumOutPts
= ClipLineToPlane
<FRUSTUM_RIGHT
>(pTempVerts
, vNumOutPts
, numAttribs
, pVertices
);
1795 vNumOutPts
= ClipLineToPlane
<FRUSTUM_BOTTOM
>(pVertices
, vNumOutPts
, numAttribs
, pTempVerts
);
1796 vNumOutPts
= ClipLineToPlane
<FRUSTUM_TOP
>(pTempVerts
, vNumOutPts
, numAttribs
, pVertices
);
1799 // restore num verts for non-clipped, active lanes
1800 simdscalar vNonClippedMask
= _simd_andnot_ps(vClipMask
, vPrimMask
);
1801 vNumOutPts
= _simd_blendv_epi32(vNumOutPts
, _simd_set1_epi32(NumVertsPerPrim
), vNonClippedMask
);
1806 #if USE_SIMD16_FRONTEND
1807 simd16scalari
ClipPrims(float* pVertices
, const simd16scalar
& vPrimMask
, const simd16scalar
& vClipMask
, int numAttribs
)
1810 float* pTempVerts
= (float*)&tlsTempVertices_simd16
[0];
1812 // zero out num input verts for non-active lanes
1813 simd16scalari vNumInPts
= _simd16_set1_epi32(NumVertsPerPrim
);
1814 vNumInPts
= _simd16_blendv_epi32(_simd16_setzero_si(), vNumInPts
, vClipMask
);
1816 // clip prims to frustum
1817 simd16scalari vNumOutPts
;
1818 if (NumVertsPerPrim
== 3)
1820 vNumOutPts
= ClipTriToPlane
<FRUSTUM_NEAR
>(pVertices
, vNumInPts
, numAttribs
, pTempVerts
);
1821 vNumOutPts
= ClipTriToPlane
<FRUSTUM_FAR
>(pTempVerts
, vNumOutPts
, numAttribs
, pVertices
);
1822 vNumOutPts
= ClipTriToPlane
<FRUSTUM_LEFT
>(pVertices
, vNumOutPts
, numAttribs
, pTempVerts
);
1823 vNumOutPts
= ClipTriToPlane
<FRUSTUM_RIGHT
>(pTempVerts
, vNumOutPts
, numAttribs
, pVertices
);
1824 vNumOutPts
= ClipTriToPlane
<FRUSTUM_BOTTOM
>(pVertices
, vNumOutPts
, numAttribs
, pTempVerts
);
1825 vNumOutPts
= ClipTriToPlane
<FRUSTUM_TOP
>(pTempVerts
, vNumOutPts
, numAttribs
, pVertices
);
1829 SWR_ASSERT(NumVertsPerPrim
== 2);
1830 vNumOutPts
= ClipLineToPlane
<FRUSTUM_NEAR
>(pVertices
, vNumInPts
, numAttribs
, pTempVerts
);
1831 vNumOutPts
= ClipLineToPlane
<FRUSTUM_FAR
>(pTempVerts
, vNumOutPts
, numAttribs
, pVertices
);
1832 vNumOutPts
= ClipLineToPlane
<FRUSTUM_LEFT
>(pVertices
, vNumOutPts
, numAttribs
, pTempVerts
);
1833 vNumOutPts
= ClipLineToPlane
<FRUSTUM_RIGHT
>(pTempVerts
, vNumOutPts
, numAttribs
, pVertices
);
1834 vNumOutPts
= ClipLineToPlane
<FRUSTUM_BOTTOM
>(pVertices
, vNumOutPts
, numAttribs
, pTempVerts
);
1835 vNumOutPts
= ClipLineToPlane
<FRUSTUM_TOP
>(pTempVerts
, vNumOutPts
, numAttribs
, pVertices
);
1838 // restore num verts for non-clipped, active lanes
1839 simd16scalar vNonClippedMask
= _simd16_andnot_ps(vClipMask
, vPrimMask
);
1840 vNumOutPts
= _simd16_blendv_epi32(vNumOutPts
, _simd16_set1_epi32(NumVertsPerPrim
), vNonClippedMask
);
1846 const uint32_t workerId
{ 0 };
1847 DRAW_CONTEXT
* pDC
{ nullptr };
1848 const API_STATE
& state
;
1849 simdscalar clipCodes
[NumVertsPerPrim
];
1850 #if USE_SIMD16_FRONTEND
1851 simd16scalar clipCodes_simd16
[NumVertsPerPrim
];
1856 // pipeline stage functions
1857 void ClipTriangles(DRAW_CONTEXT
*pDC
, PA_STATE
& pa
, uint32_t workerId
, simdvector prims
[], uint32_t primMask
, simdscalari primId
, simdscalari viewportIdx
);
1858 void ClipLines(DRAW_CONTEXT
*pDC
, PA_STATE
& pa
, uint32_t workerId
, simdvector prims
[], uint32_t primMask
, simdscalari primId
, simdscalari viewportIdx
);
1859 void ClipPoints(DRAW_CONTEXT
*pDC
, PA_STATE
& pa
, uint32_t workerId
, simdvector prims
[], uint32_t primMask
, simdscalari primId
, simdscalari viewportIdx
);
1860 #if USE_SIMD16_FRONTEND
1861 void SIMDAPI
ClipTriangles_simd16(DRAW_CONTEXT
*pDC
, PA_STATE
& pa
, uint32_t workerId
, simd16vector prims
[], uint32_t primMask
, simd16scalari primId
, simd16scalari viewportIdx
);
1862 void SIMDAPI
ClipLines_simd16(DRAW_CONTEXT
*pDC
, PA_STATE
& pa
, uint32_t workerId
, simd16vector prims
[], uint32_t primMask
, simd16scalari primId
, simd16scalari viewportIdx
);
1863 void SIMDAPI
ClipPoints_simd16(DRAW_CONTEXT
*pDC
, PA_STATE
& pa
, uint32_t workerId
, simd16vector prims
[], uint32_t primMask
, simd16scalari primId
, simd16scalari viewportIdx
);