summaryrefslogtreecommitdiffstats
path: root/src/gallium
diff options
context:
space:
mode:
authorTim Rowley <[email protected]>2017-02-16 10:53:01 -0800
committerTim Rowley <[email protected]>2017-03-20 18:04:53 -0500
commitf445b6de9cc416ba3f3a900e98baa57e090c39ed (patch)
tree1e358d3955dba3eb440357dd5d6914e60761bb76 /src/gallium
parente4d1294afbd798fa145d68a1c4f4ce80622a62dc (diff)
swr: [rasterizer] Backend code adjustments
Reviewed-by: Bruce Cherniak <[email protected]>
Diffstat (limited to 'src/gallium')
-rw-r--r--src/gallium/drivers/swr/rasterizer/common/simdintrin.h9
-rw-r--r--src/gallium/drivers/swr/rasterizer/core/api.cpp5
-rw-r--r--src/gallium/drivers/swr/rasterizer/core/backend.cpp1
-rw-r--r--src/gallium/drivers/swr/rasterizer/core/backend.h96
-rw-r--r--src/gallium/drivers/swr/rasterizer/core/state.h4
5 files changed, 70 insertions, 45 deletions
diff --git a/src/gallium/drivers/swr/rasterizer/common/simdintrin.h b/src/gallium/drivers/swr/rasterizer/common/simdintrin.h
index ea79902a002..562408db8db 100644
--- a/src/gallium/drivers/swr/rasterizer/common/simdintrin.h
+++ b/src/gallium/drivers/swr/rasterizer/common/simdintrin.h
@@ -660,6 +660,15 @@ simdscalar vMask(int32_t mask)
}
INLINE
+simdscalari vMaski(int32_t mask)
+{
+ __m256i vec = _mm256_set1_epi32(mask);
+ const __m256i bit = _mm256_set_epi32(0x80, 0x40, 0x20, 0x10, 0x08, 0x04, 0x02, 0x01);
+ vec = _simd_and_si(vec, bit);
+ return _simd_cmplt_epi32(_mm256_setzero_si256(), vec);
+}
+
+INLINE
void _simd_mov(simdscalar &r, unsigned int rlane, simdscalar& s, unsigned int slane)
{
OSALIGNSIMD(float) rArray[KNOB_SIMD_WIDTH], sArray[KNOB_SIMD_WIDTH];
diff --git a/src/gallium/drivers/swr/rasterizer/core/api.cpp b/src/gallium/drivers/swr/rasterizer/core/api.cpp
index 90503ba6de4..f622e94d91f 100644
--- a/src/gallium/drivers/swr/rasterizer/core/api.cpp
+++ b/src/gallium/drivers/swr/rasterizer/core/api.cpp
@@ -806,7 +806,6 @@ void SetupPipeline(DRAW_CONTEXT *pDC)
const bool bMultisampleEnable = ((rastState.sampleCount > SWR_MULTISAMPLE_1X) || rastState.forcedSampleCount) ? 1 : 0;
const uint32_t centroid = ((psState.barycentricsMask & SWR_BARYCENTRIC_CENTROID_MASK) > 0) ? 1 : 0;
const uint32_t canEarlyZ = (psState.forceEarlyZ || (!psState.writesODepth && !psState.usesSourceDepth && !psState.usesUAV)) ? 1 : 0;
-
SWR_BARYCENTRICS_MASK barycentricsMask = (SWR_BARYCENTRICS_MASK)psState.barycentricsMask;
// select backend function
@@ -817,7 +816,9 @@ void SetupPipeline(DRAW_CONTEXT *pDC)
{
// always need to generate I & J per sample for Z interpolation
barycentricsMask = (SWR_BARYCENTRICS_MASK)(barycentricsMask | SWR_BARYCENTRIC_PER_SAMPLE_MASK);
- backendFuncs.pfnBackend = gBackendPixelRateTable[rastState.sampleCount][rastState.samplePattern][psState.inputCoverage][centroid][forcedSampleCount][canEarlyZ];
+ backendFuncs.pfnBackend = gBackendPixelRateTable[rastState.sampleCount][rastState.samplePattern][psState.inputCoverage]
+ [centroid][forcedSampleCount][canEarlyZ]
+ ;
}
else
{
diff --git a/src/gallium/drivers/swr/rasterizer/core/backend.cpp b/src/gallium/drivers/swr/rasterizer/core/backend.cpp
index b1bcdb0b393..b915e327426 100644
--- a/src/gallium/drivers/swr/rasterizer/core/backend.cpp
+++ b/src/gallium/drivers/swr/rasterizer/core/backend.cpp
@@ -39,6 +39,7 @@
typedef void(*PFN_CLEAR_TILES)(DRAW_CONTEXT*, SWR_RENDERTARGET_ATTACHMENT rt, uint32_t, uint32_t, DWORD[4], const SWR_RECT& rect);
static PFN_CLEAR_TILES sClearTilesTable[NUM_SWR_FORMATS];
+
//////////////////////////////////////////////////////////////////////////
/// @brief Process compute work.
/// @param pDC - pointer to draw context (dispatch).
diff --git a/src/gallium/drivers/swr/rasterizer/core/backend.h b/src/gallium/drivers/swr/rasterizer/core/backend.h
index c3585cc930c..f022990bf26 100644
--- a/src/gallium/drivers/swr/rasterizer/core/backend.h
+++ b/src/gallium/drivers/swr/rasterizer/core/backend.h
@@ -605,8 +605,10 @@ struct PixelRateZTestLoop
if(psState.writesODepth)
{
- // broadcast and test oDepth(psContext.vZ) written from the PS for each sample
- vZ[sample] = psContext.vZ;
+ {
+ // broadcast and test oDepth(psContext.vZ) written from the PS for each sample
+ vZ[sample] = psContext.vZ;
+ }
}
else
{
@@ -713,23 +715,26 @@ INLINE void OutputMerger4x2(SWR_PS_CONTEXT &psContext, uint8_t* (&pColorBase)[SW
uint8_t *pColorSample = pColorBase[rt] + rasterTileColorOffset;
const SWR_RENDER_TARGET_BLEND_STATE *pRTBlend = &pBlendState->renderTarget[rt];
- // pfnBlendFunc may not update all channels. Initialize with PS output.
- /// TODO: move this into the blend JIT.
- blendOut = psContext.shaded[rt];
- // Blend outputs and update coverage mask for alpha test
- if(pfnBlendFunc[rt] != nullptr)
{
- pfnBlendFunc[rt](
- pBlendState,
- psContext.shaded[rt],
- psContext.shaded[1],
- psContext.shaded[0].w,
- sample,
- pColorSample,
- blendOut,
- &psContext.oMask,
- (simdscalari*)&coverageMask);
+ // pfnBlendFunc may not update all channels. Initialize with PS output.
+ /// TODO: move this into the blend JIT.
+ blendOut = psContext.shaded[rt];
+
+ // Blend outputs and update coverage mask for alpha test
+ if(pfnBlendFunc[rt] != nullptr)
+ {
+ pfnBlendFunc[rt](
+ pBlendState,
+ psContext.shaded[rt],
+ psContext.shaded[1],
+ psContext.shaded[0].w,
+ sample,
+ pColorSample,
+ blendOut,
+ &psContext.oMask,
+ (simdscalari*)&coverageMask);
+ }
}
// final write mask
@@ -782,9 +787,6 @@ INLINE void OutputMerger8x2(SWR_PS_CONTEXT &psContext, uint8_t* (&pColorBase)[SW
simdscalar *pColorSample = reinterpret_cast<simdscalar *>(pColorBase[rt] + rasterTileColorOffset);
const SWR_RENDER_TARGET_BLEND_STATE *pRTBlend = &pBlendState->renderTarget[rt];
- // pfnBlendFunc may not update all channels. Initialize with PS output.
- /// TODO: move this into the blend JIT.
- blendOut = psContext.shaded[rt];
if (colorBufferBit & colorBufferEnableMask)
{
@@ -794,19 +796,25 @@ INLINE void OutputMerger8x2(SWR_PS_CONTEXT &psContext, uint8_t* (&pColorBase)[SW
blendSrc[3] = pColorSample[6];
}
- // Blend outputs and update coverage mask for alpha test
- if (pfnBlendFunc[rt] != nullptr)
{
- pfnBlendFunc[rt](
- pBlendState,
- psContext.shaded[rt],
- psContext.shaded[1],
- psContext.shaded[0].w,
- sample,
- reinterpret_cast<uint8_t *>(&blendSrc),
- blendOut,
- &psContext.oMask,
- reinterpret_cast<simdscalari *>(&coverageMask));
+ // pfnBlendFunc may not update all channels. Initialize with PS output.
+ /// TODO: move this into the blend JIT.
+ blendOut = psContext.shaded[rt];
+
+ // Blend outputs and update coverage mask for alpha test
+ if(pfnBlendFunc[rt] != nullptr)
+ {
+ pfnBlendFunc[rt](
+ pBlendState,
+ psContext.shaded[rt],
+ psContext.shaded[1],
+ psContext.shaded[0].w,
+ sample,
+ reinterpret_cast<uint8_t *>(&blendSrc),
+ blendOut,
+ &psContext.oMask,
+ reinterpret_cast<simdscalari *>(&coverageMask));
+ }
}
// final write mask
@@ -840,6 +848,9 @@ INLINE void OutputMerger8x2(SWR_PS_CONTEXT &psContext, uint8_t* (&pColorBase)[SW
template<typename T>
void BackendPixelRate(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t x, uint32_t y, SWR_TRIANGLE_DESC &work, RenderOutputBuffers &renderBuffers)
{
+ ///@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);
@@ -850,12 +861,12 @@ void BackendPixelRate(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t x, uint32_t
BarycentricCoeffs coeffs;
SetupBarycentricCoeffs(&coeffs, work);
- uint8_t *pColorBuffer[SWR_NUM_RENDERTARGETS], *pDepthBuffer, *pStencilBuffer;
- SetupRenderBuffers(pColorBuffer, &pDepthBuffer, &pStencilBuffer, state.psState.numRenderTargets, renderBuffers);
-
SWR_PS_CONTEXT psContext;
SetupPixelShaderContext<T>(&psContext, work);
+ uint8_t *pDepthBuffer, *pStencilBuffer;
+ SetupRenderBuffers(psContext.pColorBuffer, &pDepthBuffer, &pStencilBuffer, state.psState.numRenderTargets, renderBuffers);
+
AR_END(BESetup, 0);
PixelRateZTestLoop<T> PixelRateZTest(pDC, workerId, work, coeffs, state, pDepthBuffer, pStencilBuffer, state.rastState.clipDistanceMask);
@@ -975,10 +986,10 @@ void BackendPixelRate(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t x, uint32_t
// broadcast the results of the PS to all passing pixels
#if USE_8x2_TILE_BACKEND
- OutputMerger8x2(psContext, pColorBuffer, sample, &state.blendState, state.pfnBlendFunc, coverageMask, depthMask, state.psState.numRenderTargets, state.colorHottileEnable, useAlternateOffset);
-#else
- OutputMerger4x2(psContext, pColorBuffer, sample, &state.blendState, state.pfnBlendFunc, coverageMask, depthMask, state.psState.numRenderTargets);
-#endif
+ OutputMerger8x2(psContext, psContext.pColorBuffer, sample, &state.blendState,state.pfnBlendFunc, coverageMask, depthMask, state.psState.numRenderTargets, state.colorHottileEnable, useAlternateOffset);
+#else // USE_8x2_TILE_BACKEND
+ OutputMerger4x2(psContext, psContext.pColorBuffer, sample, &state.blendState, state.pfnBlendFunc, coverageMask, depthMask, state.psState.numRenderTargets);
+#endif // USE_8x2_TILE_BACKEND
if(!state.psState.forceEarlyZ && !T::bForcedSampleCount)
{
@@ -1009,13 +1020,13 @@ Endtile:
{
for (uint32_t rt = 0; rt < state.psState.numRenderTargets; ++rt)
{
- pColorBuffer[rt] += (2 * KNOB_SIMD_WIDTH * FormatTraits<KNOB_COLOR_HOT_TILE_FORMAT>::bpp) / 8;
+ psContext.pColorBuffer[rt] += (2 * KNOB_SIMD_WIDTH * FormatTraits<KNOB_COLOR_HOT_TILE_FORMAT>::bpp) / 8;
}
}
#else
for(uint32_t rt = 0; rt < state.psState.numRenderTargets; ++rt)
{
- pColorBuffer[rt] += (KNOB_SIMD_WIDTH * FormatTraits<KNOB_COLOR_HOT_TILE_FORMAT>::bpp) / 8;
+ psContext.pColorBuffer[rt] += (KNOB_SIMD_WIDTH * FormatTraits<KNOB_COLOR_HOT_TILE_FORMAT>::bpp) / 8;
}
pDepthBuffer += (KNOB_SIMD_WIDTH * FormatTraits<KNOB_DEPTH_HOT_TILE_FORMAT>::bpp) / 8;
pStencilBuffer += (KNOB_SIMD_WIDTH * FormatTraits<KNOB_STENCIL_HOT_TILE_FORMAT>::bpp) / 8;
@@ -1035,7 +1046,8 @@ Endtile:
}
template<uint32_t sampleCountT = SWR_MULTISAMPLE_1X, uint32_t samplePattern = SWR_MSAA_STANDARD_PATTERN,
- uint32_t coverage = 0, uint32_t centroid = 0, uint32_t forced = 0, uint32_t canEarlyZ = 0>
+ uint32_t coverage = 0, uint32_t centroid = 0, uint32_t forced = 0, uint32_t canEarlyZ = 0
+ >
struct SwrBackendTraits
{
static const bool bIsStandardPattern = (samplePattern == SWR_MSAA_STANDARD_PATTERN);
diff --git a/src/gallium/drivers/swr/rasterizer/core/state.h b/src/gallium/drivers/swr/rasterizer/core/state.h
index 05347dce986..bb1336c429f 100644
--- a/src/gallium/drivers/swr/rasterizer/core/state.h
+++ b/src/gallium/drivers/swr/rasterizer/core/state.h
@@ -330,6 +330,8 @@ struct SWR_PS_CONTEXT
uint32_t rasterizerSampleCount; // IN: sample count used by the rasterizer
+ uint8_t* pColorBuffer[SWR_NUM_RENDERTARGETS];
+ // IN: Pointers to render target hottiles
};
//////////////////////////////////////////////////////////////////////////
@@ -511,6 +513,7 @@ struct SWR_SURFACE_STATE
uint8_t *pAuxBaseAddress; // Used for compression, append/consume counter, etc.
SWR_AUX_MODE auxMode; // @llvm_enum
+
bool bInterleavedSamples; // are MSAA samples stored interleaved or planar
};
@@ -1087,7 +1090,6 @@ struct SWR_PS_STATE
uint32_t barycentricsMask : 3; // which type(s) of barycentric coords does the PS interpolate attributes with
uint32_t usesUAV : 1; // pixel shader accesses UAV
uint32_t forceEarlyZ : 1; // force execution of early depth/stencil test
-
};
// depth bounds state