#define MASK 0xff
#endif
-static INLINE simdmask ComputeUserClipMask(uint8_t clipMask, float* pUserClipBuffer, simdscalar vI, simdscalar vJ)
+static INLINE simdmask ComputeUserClipMask(uint8_t clipMask, float* pUserClipBuffer, simdscalar const &vI, simdscalar const &vJ)
{
simdscalar vClipMask = _simd_setzero_ps();
uint32_t numClipDistance = _mm_popcnt_u32(clipMask);
simdscalar vB = _simd_broadcast_ss(pUserClipBuffer++);
simdscalar vC = _simd_broadcast_ss(pUserClipBuffer++);
+ simdscalar vK = _simd_sub_ps(_simd_sub_ps(_simd_set1_ps(1.0f), vI), vJ);
+
// interpolate
- simdscalar vInterp = vplaneps(vA, vB, vC, vI, vJ);
+ simdscalar vInterp = vplaneps(vA, vB, _simd_mul_ps(vK, vC), vI, vJ);
// clip if interpolated clip distance is < 0 || NAN
simdscalar vCull = _simd_cmp_ps(_simd_setzero_ps(), vInterp, _CMP_NLE_UQ);
template<typename T>
INLINE void CalcCentroidPos(SWR_PS_CONTEXT &psContext, const SWR_MULTISAMPLE_POS& samplePos,
const uint64_t *const coverageMask, const uint32_t sampleMask,
- const simdscalar vXSamplePosUL, const simdscalar vYSamplePosUL)
+ simdscalar const &vXSamplePosUL, simdscalar const &vYSamplePosUL)
{
uint32_t inputMask[KNOB_SIMD_WIDTH];
generateInputCoverage<T, T::InputCoverage>(coverageMask, inputMask, sampleMask);
}
INLINE void CalcCentroidBarycentrics(const BarycentricCoeffs& coeffs, SWR_PS_CONTEXT &psContext,
- const simdscalar vXSamplePosUL, const simdscalar vYSamplePosUL)
+ const simdscalar &vXSamplePosUL, const simdscalar &vYSamplePosUL)
{
// evaluate I,J
psContext.vI.centroid = vplaneps(coeffs.vIa, coeffs.vIb, coeffs.vIc, psContext.vX.centroid, psContext.vY.centroid);
psContext.vOneOverW.centroid = vplaneps(coeffs.vAOneOverW, coeffs.vBOneOverW, coeffs.vCOneOverW, psContext.vI.centroid, psContext.vJ.centroid);
}
-INLINE simdmask CalcDepthBoundsAcceptMask(simdscalar z, float minz, float maxz)
+INLINE simdmask CalcDepthBoundsAcceptMask(simdscalar const &z, float minz, float maxz)
{
const simdscalar minzMask = _simd_cmpge_ps(z, _simd_set1_ps(minz));
const simdscalar maxzMask = _simd_cmple_ps(z, _simd_set1_ps(maxz));
uint32_t operator()(simdscalar& activeLanes, SWR_PS_CONTEXT& psContext,
const CORE_BUCKETS BEDepthBucket, uint32_t currentSimdIn8x8 = 0)
{
- SWR_CONTEXT *pContext = pDC->pContext;
uint32_t statCount = 0;
simdscalar anyDepthSamplePassed = _simd_setzero_ps();
for(uint32_t sample = 0; sample < T::MultisampleT::numCoverageSamples; sample++)
{
const uint8_t *pCoverageMask = (uint8_t*)&work.coverageMask[sample];
- vCoverageMask[sample] = _simd_and_ps(activeLanes, vMask(pCoverageMask[currentSimdIn8x8] & MASK));
+ vCoverageMask[sample] = _simd_and_ps(activeLanes, _simd_vmask_ps(pCoverageMask[currentSimdIn8x8] & MASK));
if(!_simd_movemask_ps(vCoverageMask[sample]))
{
const float minz = state.depthBoundsState.depthBoundsTestMinValue;
const float maxz = state.depthBoundsState.depthBoundsTestMaxValue;
- vCoverageMask[sample] = _simd_and_ps(vCoverageMask[sample], vMask(CalcDepthBoundsAcceptMask(z, minz, maxz)));
+ vCoverageMask[sample] = _simd_and_ps(vCoverageMask[sample], _simd_vmask_ps(CalcDepthBoundsAcceptMask(z, minz, maxz)));
}
- AR_BEGIN(BEBarycentric, pDC->drawId);
+ RDTSC_BEGIN(BEBarycentric, pDC->drawId);
// calculate per sample positions
psContext.vX.sample = _simd_add_ps(psContext.vX.UL, samplePos.vX(sample));
vZ[sample] = state.pfnQuantizeDepth(vZ[sample]);
}
- AR_END(BEBarycentric, 0);
+ RDTSC_END(BEBarycentric, 0);
///@todo: perspective correct vs non-perspective correct clipping?
// if clip distances are enabled, we need to interpolate for each sample
{
uint8_t clipMask = ComputeUserClipMask(clipDistanceMask, work.pUserClipBuffer, psContext.vI.sample, psContext.vJ.sample);
- vCoverageMask[sample] = _simd_and_ps(vCoverageMask[sample], vMask(~clipMask));
+ vCoverageMask[sample] = _simd_and_ps(vCoverageMask[sample], _simd_vmask_ps(~clipMask));
}
// ZTest for this sample
///@todo Need to uncomment out this bucket.
- //AR_BEGIN(BEDepthBucket, pDC->drawId);
+ //RDTSC_BEGIN(BEDepthBucket, pDC->drawId);
depthPassMask[sample] = vCoverageMask[sample];
stencilPassMask[sample] = vCoverageMask[sample];
depthPassMask[sample] = DepthStencilTest(&state, work.triFlags.frontFacing, work.triFlags.viewportIndex,
vZ[sample], pDepthSample, vCoverageMask[sample],
pStencilSample, &stencilPassMask[sample]);
- //AR_END(BEDepthBucket, 0);
+ //RDTSC_END(BEDepthBucket, 0);
// early-exit if no pixels passed depth or earlyZ is forced on
if(psState.forceEarlyZ || !_simd_movemask_ps(depthPassMask[sample]))
// Merge Output to 4x2 SIMD Tile Format
INLINE void OutputMerger4x2(SWR_PS_CONTEXT &psContext, uint8_t* (&pColorBase)[SWR_NUM_RENDERTARGETS], uint32_t sample, const SWR_BLEND_STATE *pBlendState,
- const PFN_BLEND_JIT_FUNC (&pfnBlendFunc)[SWR_NUM_RENDERTARGETS], simdscalar &coverageMask, simdscalar depthPassMask, uint32_t renderTargetMask)
+ const PFN_BLEND_JIT_FUNC (&pfnBlendFunc)[SWR_NUM_RENDERTARGETS], simdscalar &coverageMask, simdscalar const &depthPassMask, uint32_t renderTargetMask)
{
// type safety guaranteed from template instantiation in BEChooser<>::GetFunc
const uint32_t rasterTileColorOffset = RasterTileColorOffset(sample);
#if USE_8x2_TILE_BACKEND
// Merge Output to 8x2 SIMD16 Tile Format
INLINE void OutputMerger8x2(SWR_PS_CONTEXT &psContext, uint8_t* (&pColorBase)[SWR_NUM_RENDERTARGETS], uint32_t sample, const SWR_BLEND_STATE *pBlendState,
- const PFN_BLEND_JIT_FUNC(&pfnBlendFunc)[SWR_NUM_RENDERTARGETS], simdscalar &coverageMask, simdscalar depthPassMask, uint32_t renderTargetMask, bool useAlternateOffset)
+ const PFN_BLEND_JIT_FUNC(&pfnBlendFunc)[SWR_NUM_RENDERTARGETS], simdscalar &coverageMask, simdscalar const &depthPassMask, uint32_t renderTargetMask, bool useAlternateOffset)
{
// type safety guaranteed from template instantiation in BEChooser<>::GetFunc
uint32_t rasterTileColorOffset = RasterTileColorOffset(sample);
///@todo: Need to move locals off stack to prevent __chkstk's from being generated for the backend
- SWR_CONTEXT *pContext = pDC->pContext;
-
- AR_BEGIN(BEPixelRateBackend, pDC->drawId);
- AR_BEGIN(BESetup, pDC->drawId);
+ RDTSC_BEGIN(BEPixelRateBackend, pDC->drawId);
+ RDTSC_BEGIN(BESetup, pDC->drawId);
const API_STATE &state = GetApiState(pDC);
uint8_t *pDepthBuffer, *pStencilBuffer;
SetupRenderBuffers(psContext.pColorBuffer, &pDepthBuffer, &pStencilBuffer, state.colorHottileEnable, renderBuffers);
- AR_END(BESetup, 0);
+ RDTSC_END(BESetup, 0);
- PixelRateZTestLoop<T> PixelRateZTest(pDC, workerId, work, coeffs, state, pDepthBuffer, pStencilBuffer, state.rastState.clipDistanceMask);
+ PixelRateZTestLoop<T> PixelRateZTest(pDC, workerId, work, coeffs, state, pDepthBuffer, pStencilBuffer, state.backendState.clipDistanceMask);
psContext.vY.UL = _simd_add_ps(vULOffsetsY, _simd_set1_ps(static_cast<float>(y)));
psContext.vY.center = _simd_add_ps(vCenterOffsetsY, _simd_set1_ps(static_cast<float>(y)));
#endif
simdscalar activeLanes;
if(!(work.anyCoveredSamples & MASK)) {goto Endtile;};
- activeLanes = vMask(work.anyCoveredSamples & MASK);
+ activeLanes = _simd_vmask_ps(work.anyCoveredSamples & MASK);
if (T::InputCoverage != SWR_INPUT_COVERAGE_NONE)
{
generateInputCoverage<T, T::InputCoverage>(pCoverageMask, psContext.inputMask, state.blendState.sampleMask);
}
- AR_BEGIN(BEBarycentric, pDC->drawId);
+ RDTSC_BEGIN(BEBarycentric, pDC->drawId);
CalcPixelBarycentrics(coeffs, psContext);
CalcCentroid<T, false>(&psContext, samplePos, coeffs, work.coverageMask, state.blendState.sampleMask);
- AR_END(BEBarycentric, 0);
+ RDTSC_END(BEBarycentric, 0);
if(T::bForcedSampleCount)
{
if(state.psState.usesSourceDepth)
{
- AR_BEGIN(BEBarycentric, pDC->drawId);
+ RDTSC_BEGIN(BEBarycentric, pDC->drawId);
// interpolate and quantize z
psContext.vZ = vplaneps(coeffs.vZa, coeffs.vZb, coeffs.vZc, psContext.vI.center, psContext.vJ.center);
psContext.vZ = state.pfnQuantizeDepth(psContext.vZ);
- AR_END(BEBarycentric, 0);
+ RDTSC_END(BEBarycentric, 0);
}
// pixels that are currently active
psContext.oMask = T::MultisampleT::FullSampleMask();
// execute pixel shader
- AR_BEGIN(BEPixelShader, pDC->drawId);
+ RDTSC_BEGIN(BEPixelShader, pDC->drawId);
state.psState.pfnPixelShader(GetPrivateState(pDC), &psContext);
UPDATE_STAT_BE(PsInvocations, _mm_popcnt_u32(_simd_movemask_ps(activeLanes)));
- AR_END(BEPixelShader, 0);
+ RDTSC_END(BEPixelShader, 0);
// update active lanes to remove any discarded or oMask'd pixels
activeLanes = _simd_castsi_ps(_simd_and_si(psContext.activeMask, _simd_cmpgt_epi32(psContext.oMask, _simd_setzero_si())));
// loop over all samples, broadcasting the results of the PS to all passing pixels
for(uint32_t sample = 0; sample < GetNumOMSamples<T>(state.blendState.sampleCount); sample++)
{
- AR_BEGIN(BEOutputMerger, pDC->drawId);
+ RDTSC_BEGIN(BEOutputMerger, pDC->drawId);
// center pattern does a single coverage/depth/stencil test, standard pattern tests all samples
uint32_t coverageSampleNum = (T::bIsCenterPattern) ? 0 : sample;
simdscalar coverageMask, depthMask;
if(!_simd_movemask_ps(depthMask))
{
// stencil should already have been written in early/lateZ tests
- AR_END(BEOutputMerger, 0);
+ RDTSC_END(BEOutputMerger, 0);
continue;
}
}
DepthStencilWrite(&state.vp[work.triFlags.viewportIndex], &state.depthStencilState, work.triFlags.frontFacing, PixelRateZTest.vZ[coverageSampleNum],
pDepthSample, depthMask, coverageMask, pStencilSample, PixelRateZTest.stencilPassMask[coverageSampleNum]);
}
- AR_END(BEOutputMerger, 0);
+ RDTSC_END(BEOutputMerger, 0);
}
Endtile:
- AR_BEGIN(BEEndTile, pDC->drawId);
+ RDTSC_BEGIN(BEEndTile, pDC->drawId);
for(uint32_t sample = 0; sample < T::MultisampleT::numCoverageSamples; sample++)
{
pDepthBuffer += (KNOB_SIMD_WIDTH * FormatTraits<KNOB_DEPTH_HOT_TILE_FORMAT>::bpp) / 8;
pStencilBuffer += (KNOB_SIMD_WIDTH * FormatTraits<KNOB_STENCIL_HOT_TILE_FORMAT>::bpp) / 8;
- AR_END(BEEndTile, 0);
+ RDTSC_END(BEEndTile, 0);
psContext.vX.UL = _simd_add_ps(psContext.vX.UL, dx);
psContext.vX.center = _simd_add_ps(psContext.vX.center, dx);
psContext.vY.center = _simd_add_ps(psContext.vY.center, dy);
}
- AR_END(BEPixelRateBackend, 0);
+ RDTSC_END(BEPixelRateBackend, 0);
}
template<uint32_t sampleCountT = SWR_MULTISAMPLE_1X, uint32_t isCenter = 0,
projects / mesa.git / blobdiff