swr/rast: Remove deprecated 4x2 backend code

- Use 8x2 tiling by default
  - Remove associated macros
- Use SIMDLIB emulation for SIMD16 on SIMD8 hardware
- Remove code rot in Load/StoreTile

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

8 files changed, 19 insertions, 834 deletions

diff --git a/src/gallium/drivers/swr/rasterizer/core/backend_clear.cpp b/src/gallium/drivers/swr/rasterizer/core/backend_clear.cpp
index 623ebc81d77..5750ceac7f0 100644
--- a/src/gallium/drivers/swr/rasterizer/core/backend_clear.cpp
+++ b/src/gallium/drivers/swr/rasterizer/core/backend_clear.cpp
@@ -38,28 +38,10 @@
 #include <algorithm>
 
 template <SWR_FORMAT format>
-void ClearRasterTile(uint8_t* pTileBuffer, simdvector& value)
-{
-    auto lambda = [&](int32_t comp) {
-        FormatTraits<format>::storeSOA(comp, pTileBuffer, value.v[comp]);
-
-        pTileBuffer += (KNOB_SIMD_WIDTH * FormatTraits<format>::GetBPC(comp) / 8);
-    };
-
-    const uint32_t numIter =
-        (KNOB_TILE_Y_DIM / SIMD_TILE_Y_DIM) * (KNOB_TILE_X_DIM / SIMD_TILE_X_DIM);
-
-    for (uint32_t i = 0; i < numIter; ++i)
-    {
-        UnrollerL<0, FormatTraits<format>::numComps, 1>::step(lambda);
-    }
-}
-
-#if USE_8x2_TILE_BACKEND
-template <SWR_FORMAT format>
 void ClearRasterTile(uint8_t* pTileBuffer, simd16vector& value)
 {
-    auto lambda = [&](int32_t comp) {
+    auto lambda = [&](int32_t comp)
+    {
         FormatTraits<format>::storeSOA(comp, pTileBuffer, value.v[comp]);
 
         pTileBuffer += (KNOB_SIMD16_WIDTH * FormatTraits<format>::GetBPC(comp) / 8);
@@ -74,7 +56,6 @@ void ClearRasterTile(uint8_t* pTileBuffer, simd16vector& value)
     }
 }
 
-#endif
 template <SWR_FORMAT format>
 INLINE void ClearMacroTile(DRAW_CONTEXT*               pDC,
                            HANDLE                      hWorkerPrivateData,
@@ -86,37 +67,22 @@ INLINE void ClearMacroTile(DRAW_CONTEXT*               pDC,
 {
     // convert clear color to hottile format
     // clear color is in RGBA float/uint32
-#if USE_8x2_TILE_BACKEND
+
     simd16vector vClear;
     for (uint32_t comp = 0; comp < FormatTraits<format>::numComps; ++comp)
     {
-        simd16scalar vComp;
-        vComp = _simd16_load1_ps((const float*)&clear[comp]);
+        simd16scalar vComp = _simd16_load1_ps((const float*)&clear[comp]);
+
         if (FormatTraits<format>::isNormalized(comp))
         {
             vComp = _simd16_mul_ps(vComp, _simd16_set1_ps(FormatTraits<format>::fromFloat(comp)));
             vComp = _simd16_castsi_ps(_simd16_cvtps_epi32(vComp));
         }
-        vComp                                         = FormatTraits<format>::pack(comp, vComp);
-        vClear.v[FormatTraits<format>::swizzle(comp)] = vComp;
-    }
+        vComp = FormatTraits<format>::pack(comp, vComp);
 
-#else
-    simdvector vClear;
-    for (uint32_t comp = 0; comp < FormatTraits<format>::numComps; ++comp)
-    {
-        simdscalar vComp;
-        vComp = _simd_load1_ps((const float*)&clear[comp]);
-        if (FormatTraits<format>::isNormalized(comp))
-        {
-            vComp = _simd_mul_ps(vComp, _simd_set1_ps(FormatTraits<format>::fromFloat(comp)));
-            vComp = _simd_castsi_ps(_simd_cvtps_epi32(vComp));
-        }
-        vComp                                         = FormatTraits<format>::pack(comp, vComp);
         vClear.v[FormatTraits<format>::swizzle(comp)] = vComp;
     }
 
-#endif
     uint32_t tileX, tileY;
     MacroTileMgr::getTileIndices(macroTile, tileX, tileY);
 
diff --git a/src/gallium/drivers/swr/rasterizer/core/backend_impl.h b/src/gallium/drivers/swr/rasterizer/core/backend_impl.h
index d556c549704..83d662bd9a7 100644
--- a/src/gallium/drivers/swr/rasterizer/core/backend_impl.h
+++ b/src/gallium/drivers/swr/rasterizer/core/backend_impl.h
@@ -894,87 +894,6 @@ static INLINE void CalcSampleBarycentrics(const BarycentricCoeffs& coeffs,
                                           psContext.vJ.sample);
 }
 
-// Merge Output to 4x2 SIMD Tile Format
-INLINE void OutputMerger4x2(DRAW_CONTEXT*   pDC,
-                            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 const& depthPassMask,
-                            uint32_t          renderTargetMask,
-                            uint32_t          workerId)
-{
-    // type safety guaranteed from template instantiation in BEChooser<>::GetFunc
-    const uint32_t rasterTileColorOffset = RasterTileColorOffset(sample);
-    simdvector     blendOut;
-
-    DWORD rt = 0;
-    while (_BitScanForward(&rt, renderTargetMask))
-    {
-        renderTargetMask &= ~(1 << rt);
-        uint8_t* pColorSample = pColorBase[rt] + rasterTileColorOffset;
-
-        const SWR_RENDER_TARGET_BLEND_STATE* pRTBlend = &pBlendState->renderTarget[rt];
-
-        SWR_BLEND_CONTEXT blendContext = {0};
-        {
-            // pfnBlendFunc may not update all channels.  Initialize with PS output.
-            /// TODO: move this into the blend JIT.
-            blendOut = psContext.shaded[rt];
-
-            blendContext.pBlendState = pBlendState;
-            blendContext.src         = &psContext.shaded[rt];
-            blendContext.src1        = &psContext.shaded[1];
-            blendContext.src0alpha   = reinterpret_cast<simdvector*>(&psContext.shaded[0].w);
-            blendContext.sampleNum   = sample;
-            blendContext.pDst        = (simdvector*)&pColorSample;
-            blendContext.result      = &blendOut;
-            blendContext.oMask       = &psContext.oMask;
-            blendContext.pMask       = reinterpret_cast<simdscalari*>(&coverageMask);
-
-            // Blend outputs and update coverage mask for alpha test
-            if (pfnBlendFunc[rt] != nullptr)
-            {
-                pfnBlendFunc[rt](&blendContext);
-            }
-        }
-
-        // Track alpha events
-        AR_EVENT(
-            AlphaInfoEvent(pDC->drawId, blendContext.isAlphaTested, blendContext.isAlphaBlended));
-
-        // final write mask
-        simdscalari outputMask = _simd_castps_si(_simd_and_ps(coverageMask, depthPassMask));
-
-        ///@todo can only use maskstore fast path if bpc is 32. Assuming hot tile is RGBA32_FLOAT.
-        static_assert(KNOB_COLOR_HOT_TILE_FORMAT == R32G32B32A32_FLOAT,
-                      "Unsupported hot tile format");
-
-        const uint32_t simd = KNOB_SIMD_WIDTH * sizeof(float);
-
-        // store with color mask
-        if (!pRTBlend->writeDisableRed)
-        {
-            _simd_maskstore_ps((float*)pColorSample, outputMask, blendOut.x);
-        }
-        if (!pRTBlend->writeDisableGreen)
-        {
-            _simd_maskstore_ps((float*)(pColorSample + simd), outputMask, blendOut.y);
-        }
-        if (!pRTBlend->writeDisableBlue)
-        {
-            _simd_maskstore_ps((float*)(pColorSample + simd * 2), outputMask, blendOut.z);
-        }
-        if (!pRTBlend->writeDisableAlpha)
-        {
-            _simd_maskstore_ps((float*)(pColorSample + simd * 3), outputMask, blendOut.w);
-        }
-    }
-}
-
-#if USE_8x2_TILE_BACKEND
 // Merge Output to 8x2 SIMD16 Tile Format
 INLINE void OutputMerger8x2(DRAW_CONTEXT*   pDC,
                             SWR_PS_CONTEXT& psContext,
@@ -1076,8 +995,6 @@ INLINE void OutputMerger8x2(DRAW_CONTEXT*   pDC,
     }
 }
 
-#endif
-
 template <typename T>
 void BackendPixelRate(DRAW_CONTEXT*        pDC,
                       uint32_t             workerId,
@@ -1137,9 +1054,9 @@ void BackendPixelRate(DRAW_CONTEXT*        pDC,
 
         for (uint32_t xx = x; xx < x + KNOB_TILE_X_DIM; xx += SIMD_TILE_X_DIM)
         {
-#if USE_8x2_TILE_BACKEND
             const bool useAlternateOffset = ((xx & SIMD_TILE_X_DIM) != 0);
-#endif
+
+
             simdscalar activeLanes;
             if (!(work.anyCoveredSamples & MASK))
             {
@@ -1264,7 +1181,7 @@ void BackendPixelRate(DRAW_CONTEXT*        pDC,
                 }
 
                 // broadcast the results of the PS to all passing pixels
-#if USE_8x2_TILE_BACKEND
+
                 OutputMerger8x2(pDC,
                                 psContext,
                                 psContext.pColorBuffer,
@@ -1276,18 +1193,7 @@ void BackendPixelRate(DRAW_CONTEXT*        pDC,
                                 state.psState.renderTargetMask,
                                 useAlternateOffset,
                                 workerId);
-#else  // USE_8x2_TILE_BACKEND
-                OutputMerger4x2(pDC,
-                                psContext,
-                                psContext.pColorBuffer,
-                                sample,
-                                &state.blendState,
-                                state.pfnBlendFunc,
-                                coverageMask,
-                                depthMask,
-                                state.psState.renderTargetMask,
-                                workerId);
-#endif // USE_8x2_TILE_BACKEND
+
 
                 if (!state.psState.forceEarlyZ && !T::bForcedSampleCount)
                 {
@@ -1320,7 +1226,6 @@ void BackendPixelRate(DRAW_CONTEXT*        pDC,
             }
             work.anyCoveredSamples >>= (SIMD_TILE_Y_DIM * SIMD_TILE_X_DIM);
 
-#if USE_8x2_TILE_BACKEND
             if (useAlternateOffset)
             {
                 DWORD    rt;
@@ -1332,16 +1237,7 @@ void BackendPixelRate(DRAW_CONTEXT*        pDC,
                         (2 * KNOB_SIMD_WIDTH * FormatTraits<KNOB_COLOR_HOT_TILE_FORMAT>::bpp) / 8;
                 }
             }
-#else
-            DWORD    rt;
-            uint32_t rtMask = state.colorHottileEnable;
-            while (_BitScanForward(&rt, rtMask))
-            {
-                rtMask &= ~(1 << rt);
-                psContext.pColorBuffer[rt] +=
-                    (KNOB_SIMD_WIDTH * FormatTraits<KNOB_COLOR_HOT_TILE_FORMAT>::bpp) / 8;
-            }
-#endif
+
             pDepthBuffer += (KNOB_SIMD_WIDTH * FormatTraits<KNOB_DEPTH_HOT_TILE_FORMAT>::bpp) / 8;
             pStencilBuffer +=
                 (KNOB_SIMD_WIDTH * FormatTraits<KNOB_STENCIL_HOT_TILE_FORMAT>::bpp) / 8;
diff --git a/src/gallium/drivers/swr/rasterizer/core/backend_sample.cpp b/src/gallium/drivers/swr/rasterizer/core/backend_sample.cpp
index ff09cc6caa7..9b0b80f766f 100644
--- a/src/gallium/drivers/swr/rasterizer/core/backend_sample.cpp
+++ b/src/gallium/drivers/swr/rasterizer/core/backend_sample.cpp
@@ -81,9 +81,9 @@ void BackendSampleRate(DRAW_CONTEXT*        pDC,
 
         for (uint32_t xx = x; xx < x + KNOB_TILE_X_DIM; xx += SIMD_TILE_X_DIM)
         {
-#if USE_8x2_TILE_BACKEND
             const bool useAlternateOffset = ((xx & SIMD_TILE_X_DIM) != 0);
-#endif
+
+
             if (T::InputCoverage != SWR_INPUT_COVERAGE_NONE)
             {
                 const uint64_t* pCoverageMask =
@@ -252,7 +252,7 @@ void BackendSampleRate(DRAW_CONTEXT*        pDC,
 
                     // output merger
                     RDTSC_BEGIN(BEOutputMerger, pDC->drawId);
-#if USE_8x2_TILE_BACKEND
+
                     OutputMerger8x2(pDC,
                                     psContext,
                                     psContext.pColorBuffer,
@@ -264,18 +264,6 @@ void BackendSampleRate(DRAW_CONTEXT*        pDC,
                                     state.psState.renderTargetMask,
                                     useAlternateOffset,
                                     workerId);
-#else
-                    OutputMerger4x2(pDC,
-                                    psContext,
-                                    psContext.pColorBuffer,
-                                    sample,
-                                    &state.blendState,
-                                    state.pfnBlendFunc,
-                                    vCoverageMask,
-                                    depthPassMask,
-                                    state.psState.renderTargetMask,
-                                    workerId);
-#endif
 
                     // do final depth write after all pixel kills
                     if (!state.psState.forceEarlyZ)
@@ -305,7 +293,6 @@ void BackendSampleRate(DRAW_CONTEXT*        pDC,
                 work.innerCoverageMask >>= (SIMD_TILE_Y_DIM * SIMD_TILE_X_DIM);
             }
 
-#if USE_8x2_TILE_BACKEND
             if (useAlternateOffset)
             {
                 DWORD    rt;
@@ -317,16 +304,7 @@ void BackendSampleRate(DRAW_CONTEXT*        pDC,
                         (2 * KNOB_SIMD_WIDTH * FormatTraits<KNOB_COLOR_HOT_TILE_FORMAT>::bpp) / 8;
                 }
             }
-#else
-            DWORD rt;
-            uint32_t rtMask = state.colorHottileEnable;
-            while (_BitScanForward(&rt, rtMask))
-            {
-                rtMask &= ~(1 << rt);
-                psContext.pColorBuffer[rt] +=
-                    (KNOB_SIMD_WIDTH * FormatTraits<KNOB_COLOR_HOT_TILE_FORMAT>::bpp) / 8;
-            }
-#endif
+
             pDepthBuffer += (KNOB_SIMD_WIDTH * FormatTraits<KNOB_DEPTH_HOT_TILE_FORMAT>::bpp) / 8;
             pStencilBuffer +=
                 (KNOB_SIMD_WIDTH * FormatTraits<KNOB_STENCIL_HOT_TILE_FORMAT>::bpp) / 8;
diff --git a/src/gallium/drivers/swr/rasterizer/core/backend_singlesample.cpp b/src/gallium/drivers/swr/rasterizer/core/backend_singlesample.cpp
index 1c065ab14bf..46aabcdf34b 100644
--- a/src/gallium/drivers/swr/rasterizer/core/backend_singlesample.cpp
+++ b/src/gallium/drivers/swr/rasterizer/core/backend_singlesample.cpp
@@ -82,9 +82,9 @@ void BackendSingleSample(DRAW_CONTEXT*        pDC,
 
         for (uint32_t xx = x; xx < x + KNOB_TILE_X_DIM; xx += SIMD_TILE_X_DIM)
         {
-#if USE_8x2_TILE_BACKEND
             const bool useAlternateOffset = ((xx & SIMD_TILE_X_DIM) != 0);
-#endif
+
+
             simdmask coverageMask = work.coverageMask[0] & MASK;
 
             if (coverageMask)
@@ -237,7 +237,7 @@ void BackendSingleSample(DRAW_CONTEXT*        pDC,
 
                 // output merger
                 RDTSC_BEGIN(BEOutputMerger, pDC->drawId);
-#if USE_8x2_TILE_BACKEND
+
                 OutputMerger8x2(pDC,
                                 psContext,
                                 psContext.pColorBuffer,
@@ -249,19 +249,6 @@ void BackendSingleSample(DRAW_CONTEXT*        pDC,
                                 state.psState.renderTargetMask,
                                 useAlternateOffset,
                                 workerId);
-#else
-                OutputMerger4x2(pDC,
-                                psContext,
-                                psContext.pColorBuffer,
-                                0,
-                                &state.blendState,
-                                state.pfnBlendFunc,
-                                vCoverageMask,
-                                depthPassMask,
-                                state.psState.renderTargetMask,
-                                workerId,
-                                workerId);
-#endif
 
                 // do final depth write after all pixel kills
                 if (!state.psState.forceEarlyZ)
@@ -288,7 +275,6 @@ void BackendSingleSample(DRAW_CONTEXT*        pDC,
                 work.innerCoverageMask >>= (SIMD_TILE_Y_DIM * SIMD_TILE_X_DIM);
             }
 
-#if USE_8x2_TILE_BACKEND
             if (useAlternateOffset)
             {
                 DWORD    rt;
@@ -300,16 +286,7 @@ void BackendSingleSample(DRAW_CONTEXT*        pDC,
                         (2 * KNOB_SIMD_WIDTH * FormatTraits<KNOB_COLOR_HOT_TILE_FORMAT>::bpp) / 8;
                 }
             }
-#else
-            DWORD rt;
-            uint32_t rtMask = state.colorHottileEnable;
-            while (_BitScanForward(&rt, rtMask))
-            {
-                rtMask &= ~(1 << rt);
-                psContext.pColorBuffer[rt] +=
-                    (KNOB_SIMD_WIDTH * FormatTraits<KNOB_COLOR_HOT_TILE_FORMAT>::bpp) / 8;
-            }
-#endif
+
             pDepthBuffer += (KNOB_SIMD_WIDTH * FormatTraits<KNOB_DEPTH_HOT_TILE_FORMAT>::bpp) / 8;
             pStencilBuffer +=
                 (KNOB_SIMD_WIDTH * FormatTraits<KNOB_STENCIL_HOT_TILE_FORMAT>::bpp) / 8;
diff --git a/src/gallium/drivers/swr/rasterizer/core/knobs.h b/src/gallium/drivers/swr/rasterizer/core/knobs.h
index b52accbbab3..8cccbf416af 100644
--- a/src/gallium/drivers/swr/rasterizer/core/knobs.h
+++ b/src/gallium/drivers/swr/rasterizer/core/knobs.h
@@ -39,7 +39,6 @@
 ///////////////////////////////////////////////////////////////////////////////
 
 #define ENABLE_AVX512_SIMD16 1
-#define USE_8x2_TILE_BACKEND 1
 #define USE_SIMD16_FRONTEND 1
 #define USE_SIMD16_SHADERS 1 // requires USE_SIMD16_FRONTEND
 #define USE_SIMD16_VS 1      // requires USE_SIMD16_SHADERS
diff --git a/src/gallium/drivers/swr/rasterizer/core/tilemgr.cpp b/src/gallium/drivers/swr/rasterizer/core/tilemgr.cpp
index 87d5373a215..1ea1c4b1a6f 100644
--- a/src/gallium/drivers/swr/rasterizer/core/tilemgr.cpp
+++ b/src/gallium/drivers/swr/rasterizer/core/tilemgr.cpp
@@ -244,7 +244,6 @@ HOTTILE* HotTileMgr::GetHotTileNoLoad(SWR_CONTEXT*                pContext,
     return &hotTile;
 }
 
-#if USE_8x2_TILE_BACKEND
 void HotTileMgr::ClearColorHotTile(
     const HOTTILE* pHotTile) // clear a macro tile from float4 clear data.
 {
@@ -330,91 +329,6 @@ void HotTileMgr::ClearStencilHotTile(const HOTTILE* pHotTile)
     }
 }
 
-#else
-void HotTileMgr::ClearColorHotTile(
-    const HOTTILE* pHotTile) // clear a macro tile from float4 clear data.
-{
-    // Load clear color into SIMD register...
-    float*     pClearData = (float*)(pHotTile->clearData);
-    simdscalar valR       = _simd_broadcast_ss(&pClearData[0]);
-    simdscalar valG       = _simd_broadcast_ss(&pClearData[1]);
-    simdscalar valB       = _simd_broadcast_ss(&pClearData[2]);
-    simdscalar valA       = _simd_broadcast_ss(&pClearData[3]);
-
-    float*   pfBuf      = (float*)pHotTile->pBuffer;
-    uint32_t numSamples = pHotTile->numSamples;
-
-    for (uint32_t row = 0; row < KNOB_MACROTILE_Y_DIM; row += KNOB_TILE_Y_DIM)
-    {
-        for (uint32_t col = 0; col < KNOB_MACROTILE_X_DIM; col += KNOB_TILE_X_DIM)
-        {
-            for (uint32_t si = 0; si < (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * numSamples);
-                 si +=
-                 SIMD_TILE_X_DIM * SIMD_TILE_Y_DIM) // SIMD_TILE_X_DIM * SIMD_TILE_Y_DIM); si++)
-            {
-                _simd_store_ps(pfBuf, valR);
-                pfBuf += KNOB_SIMD_WIDTH;
-                _simd_store_ps(pfBuf, valG);
-                pfBuf += KNOB_SIMD_WIDTH;
-                _simd_store_ps(pfBuf, valB);
-                pfBuf += KNOB_SIMD_WIDTH;
-                _simd_store_ps(pfBuf, valA);
-                pfBuf += KNOB_SIMD_WIDTH;
-            }
-        }
-    }
-}
-
-void HotTileMgr::ClearDepthHotTile(
-    const HOTTILE* pHotTile) // clear a macro tile from float4 clear data.
-{
-    // Load clear color into SIMD register...
-    float*     pClearData = (float*)(pHotTile->clearData);
-    simdscalar valZ       = _simd_broadcast_ss(&pClearData[0]);
-
-    float*   pfBuf      = (float*)pHotTile->pBuffer;
-    uint32_t numSamples = pHotTile->numSamples;
-
-    for (uint32_t row = 0; row < KNOB_MACROTILE_Y_DIM; row += KNOB_TILE_Y_DIM)
-    {
-        for (uint32_t col = 0; col < KNOB_MACROTILE_X_DIM; col += KNOB_TILE_X_DIM)
-        {
-            for (uint32_t si = 0; si < (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * numSamples);
-                 si += SIMD_TILE_X_DIM * SIMD_TILE_Y_DIM)
-            {
-                _simd_store_ps(pfBuf, valZ);
-                pfBuf += KNOB_SIMD_WIDTH;
-            }
-        }
-    }
-}
-
-void HotTileMgr::ClearStencilHotTile(const HOTTILE* pHotTile)
-{
-    // convert from F32 to U8.
-    uint8_t clearVal = (uint8_t)(pHotTile->clearData[0]);
-    // broadcast 32x into __m256i...
-    simdscalari valS = _simd_set1_epi8(clearVal);
-
-    simdscalari* pBuf       = (simdscalari*)pHotTile->pBuffer;
-    uint32_t     numSamples = pHotTile->numSamples;
-
-    for (uint32_t row = 0; row < KNOB_MACROTILE_Y_DIM; row += KNOB_TILE_Y_DIM)
-    {
-        for (uint32_t col = 0; col < KNOB_MACROTILE_X_DIM; col += KNOB_TILE_X_DIM)
-        {
-            // We're putting 4 pixels in each of the 32-bit slots, so increment 4 times as quickly.
-            for (uint32_t si = 0; si < (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * numSamples);
-                 si += SIMD_TILE_X_DIM * SIMD_TILE_Y_DIM * 4)
-            {
-                _simd_store_si(pBuf, valS);
-                pBuf += 1;
-            }
-        }
-    }
-}
-
-#endif
 //////////////////////////////////////////////////////////////////////////
 /// @brief InitializeHotTiles
 /// for draw calls, we initialize the active hot tiles and perform deferred
diff --git a/src/gallium/drivers/swr/rasterizer/memory/LoadTile.h b/src/gallium/drivers/swr/rasterizer/memory/LoadTile.h
index d1cc3ed207f..d85a3353526 100644
--- a/src/gallium/drivers/swr/rasterizer/memory/LoadTile.h
+++ b/src/gallium/drivers/swr/rasterizer/memory/LoadTile.h
@@ -67,7 +67,6 @@ struct LoadRasterTile
         uint32_t x, uint32_t y,
         uint8_t* pDst)
     {
-#if USE_8x2_TILE_BACKEND
         typedef SimdTile_16<DstFormat, SrcFormat> SimdT;
 
         SimdT* pDstSimdTiles = (SimdT*)pDst;
@@ -81,21 +80,6 @@ struct LoadRasterTile
         uint32_t simdOffset = (y % SIMD16_TILE_Y_DIM) * SIMD16_TILE_X_DIM + (x % SIMD16_TILE_X_DIM);
 
         pSimdTile->SetSwizzledColor(simdOffset, srcColor);
-#else
-        typedef SimdTile<DstFormat, SrcFormat> SimdT;
-
-        SimdT* pDstSimdTiles = (SimdT*)pDst;
-
-        // Compute which simd tile we're accessing within 8x8 tile.
-        //   i.e. Compute linear simd tile coordinate given (x, y) in pixel coordinates.
-        uint32_t simdIndex = (y / SIMD_TILE_Y_DIM) * (KNOB_TILE_X_DIM / SIMD_TILE_X_DIM) + (x / SIMD_TILE_X_DIM);
-
-        SimdT* pSimdTile = &pDstSimdTiles[simdIndex];
-
-        uint32_t simdOffset = (y % SIMD_TILE_Y_DIM) * SIMD_TILE_X_DIM + (x % SIMD_TILE_X_DIM);
-
-        pSimdTile->SetSwizzledColor(simdOffset, srcColor);
-#endif
     }
 
     //////////////////////////////////////////////////////////////////////////
diff --git a/src/gallium/drivers/swr/rasterizer/memory/StoreTile.h b/src/gallium/drivers/swr/rasterizer/memory/StoreTile.h
index 29717261e7f..407cefae54e 100644
--- a/src/gallium/drivers/swr/rasterizer/memory/StoreTile.h
+++ b/src/gallium/drivers/swr/rasterizer/memory/StoreTile.h
@@ -104,7 +104,6 @@ struct StorePixels<8, 2>
     }
 };
 
-#if USE_8x2_TILE_BACKEND
 template <>
 struct StorePixels<8, 4>
 {
@@ -130,7 +129,6 @@ struct StorePixels<8, 4>
     }
 };
 
-#endif
 //////////////////////////////////////////////////////////////////////////
 /// StorePixels (32-bit pixel specialization)
 /// @brief Stores a 4x2 (AVX) raster-tile to two rows.
@@ -159,7 +157,6 @@ struct StorePixels<16, 2>
     }
 };
 
-#if USE_8x2_TILE_BACKEND
 template <>
 struct StorePixels<16, 4>
 {
@@ -185,7 +182,6 @@ struct StorePixels<16, 4>
     }
 };
 
-#endif
 //////////////////////////////////////////////////////////////////////////
 /// StorePixels (32-bit pixel specialization)
 /// @brief Stores a 4x2 (AVX) raster-tile to two rows.
@@ -213,7 +209,6 @@ struct StorePixels<32, 2>
     }
 };
 
-#if USE_8x2_TILE_BACKEND
 template <>
 struct StorePixels<32, 4>
 {
@@ -237,7 +232,6 @@ struct StorePixels<32, 4>
     }
 };
 
-#endif
 //////////////////////////////////////////////////////////////////////////
 /// StorePixels (32-bit pixel specialization)
 /// @brief Stores a 4x2 (AVX) raster-tile to two rows.
@@ -264,7 +258,6 @@ struct StorePixels<64, 4>
     }
 };
 
-#if USE_8x2_TILE_BACKEND
 template <>
 struct StorePixels<64, 8>
 {
@@ -287,7 +280,6 @@ struct StorePixels<64, 8>
     }
 };
 
-#endif
 //////////////////////////////////////////////////////////////////////////
 /// StorePixels (32-bit pixel specialization)
 /// @brief Stores a 4x2 (AVX) raster-tile to two rows.
@@ -318,7 +310,6 @@ struct StorePixels<128, 8>
     }
 };
 
-#if USE_8x2_TILE_BACKEND
 template <>
 struct StorePixels<128, 16>
 {
@@ -339,7 +330,6 @@ struct StorePixels<128, 16>
     }
 };
 
-#endif
 //////////////////////////////////////////////////////////////////////////
 /// ConvertPixelsSOAtoAOS - Conversion for SIMD pixel (4x2 or 2x2)
 //////////////////////////////////////////////////////////////////////////
@@ -354,7 +344,6 @@ struct ConvertPixelsSOAtoAOS
     template <size_t NumDests>
     INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests])
     {
-#if USE_8x2_TILE_BACKEND
         static const uint32_t MAX_RASTER_TILE_BYTES = 16 * 16; // 16 pixels * 16 bytes per pixel
 
         OSALIGNSIMD16(uint8_t) soaTile[MAX_RASTER_TILE_BYTES];
@@ -368,21 +357,6 @@ struct ConvertPixelsSOAtoAOS
         // Convert from SOA --> AOS
         FormatTraits<DstFormat>::TransposeT::Transpose_16(soaTile, aosTile);
 
-#else
-        static const uint32_t MAX_RASTER_TILE_BYTES = 128; // 8 pixels * 16 bytes per pixel
-
-        OSALIGNSIMD(uint8_t) soaTile[MAX_RASTER_TILE_BYTES];
-        OSALIGNSIMD(uint8_t) aosTile[MAX_RASTER_TILE_BYTES];
-
-        // Convert from SrcFormat --> DstFormat
-        simdvector src;
-        LoadSOA<SrcFormat>(pSrc, src);
-        StoreSOA<DstFormat>(src, soaTile);
-
-        // Convert from SOA --> AOS
-        FormatTraits<DstFormat>::TransposeT::Transpose(soaTile, aosTile);
-
-#endif
         // Store data into destination
         StorePixels<FormatTraits<DstFormat>::bpp, NumDests>::Store(aosTile, ppDsts);
     }
@@ -403,7 +377,6 @@ struct ConvertPixelsSOAtoAOS<Format, Format>
     template <size_t NumDests>
     INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests])
     {
-#if USE_8x2_TILE_BACKEND
         static const uint32_t MAX_RASTER_TILE_BYTES = 16 * 16; // 16 pixels * 16 bytes per pixel
 
         OSALIGNSIMD16(uint8_t) aosTile[MAX_RASTER_TILE_BYTES];
@@ -411,15 +384,6 @@ struct ConvertPixelsSOAtoAOS<Format, Format>
         // Convert from SOA --> AOS
         FormatTraits<Format>::TransposeT::Transpose_16(pSrc, aosTile);
 
-#else
-        static const uint32_t MAX_RASTER_TILE_BYTES = 128; // 8 pixels * 16 bytes per pixel
-
-        OSALIGNSIMD(uint8_t) aosTile[MAX_RASTER_TILE_BYTES];
-
-        // Convert from SOA --> AOS
-        FormatTraits<Format>::TransposeT::Transpose(pSrc, aosTile);
-
-#endif
         // Store data into destination
         StorePixels<FormatTraits<Format>::bpp, NumDests>::Store(aosTile, ppDsts);
     }
@@ -439,7 +403,6 @@ struct ConvertPixelsSOAtoAOS < R32G32B32A32_FLOAT, B5G6R5_UNORM >
     template <size_t NumDests>
     INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests])
     {
-#if USE_8x2_TILE_BACKEND
         static const SWR_FORMAT SrcFormat = R32G32B32A32_FLOAT;
         static const SWR_FORMAT DstFormat = B5G6R5_UNORM;
 
@@ -483,47 +446,6 @@ struct ConvertPixelsSOAtoAOS < R32G32B32A32_FLOAT, B5G6R5_UNORM >
             *pAosTile++ = *pPacked++;
         }
 
-#else
-        static const SWR_FORMAT SrcFormat = R32G32B32A32_FLOAT;
-        static const SWR_FORMAT DstFormat = B5G6R5_UNORM;
-        static const uint32_t MAX_RASTER_TILE_BYTES = 128; // 8 pixels * 16 bytes per pixel
-
-        OSALIGNSIMD(uint8_t) aosTile[MAX_RASTER_TILE_BYTES];
-
-        // Load hot-tile
-        simdvector src, dst;
-        LoadSOA<SrcFormat>(pSrc, src);
-
-        // deswizzle
-        dst.x = src[FormatTraits<DstFormat>::swizzle(0)];
-        dst.y = src[FormatTraits<DstFormat>::swizzle(1)];
-        dst.z = src[FormatTraits<DstFormat>::swizzle(2)];
-
-        // clamp
-        dst.x = Clamp<DstFormat>(dst.x, 0);
-        dst.y = Clamp<DstFormat>(dst.y, 1);
-        dst.z = Clamp<DstFormat>(dst.z, 2);
-
-        // normalize
-        dst.x = Normalize<DstFormat>(dst.x, 0);
-        dst.y = Normalize<DstFormat>(dst.y, 1);
-        dst.z = Normalize<DstFormat>(dst.z, 2);
-
-        // pack
-        simdscalari packed = _simd_castps_si(dst.x);
-        packed = _simd_or_si(packed, _simd_slli_epi32(_simd_castps_si(dst.y), FormatTraits<DstFormat>::GetConstBPC(0)));
-        packed = _simd_or_si(packed, _simd_slli_epi32(_simd_castps_si(dst.z), FormatTraits<DstFormat>::GetConstBPC(0) +
-                                                                              FormatTraits<DstFormat>::GetConstBPC(1)));
-
-        // pack low 16 bits of each 32 bit lane to low 128 bits of dst
-        uint32_t *pPacked = (uint32_t*)&packed;
-        uint16_t *pAosTile = (uint16_t*)&aosTile[0];
-        for (uint32_t t = 0; t < KNOB_SIMD_WIDTH; ++t)
-        {
-            *pAosTile++ = *pPacked++;
-        }
-
-#endif
         // Store data into destination
         StorePixels<FormatTraits<DstFormat>::bpp, NumDests>::Store(aosTile, ppDsts);
     }
@@ -546,7 +468,6 @@ struct ConvertPixelsSOAtoAOS<R32_FLOAT, R24_UNORM_X8_TYPELESS>
     template <size_t NumDests>
     INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests])
     {
-#if USE_8x2_TILE_BACKEND
         simd16scalar comp = _simd16_load_ps(reinterpret_cast<const float *>(pSrc));
 
         // clamp
@@ -579,46 +500,9 @@ struct ConvertPixelsSOAtoAOS<R32_FLOAT, R24_UNORM_X8_TYPELESS>
 
         _simd_storeu2_si(reinterpret_cast<simd4scalari *>(ppDsts[1]), reinterpret_cast<simd4scalari *>(ppDsts[0]), _simd16_extract_si(dest, 0));
         _simd_storeu2_si(reinterpret_cast<simd4scalari *>(ppDsts[3]), reinterpret_cast<simd4scalari *>(ppDsts[2]), _simd16_extract_si(dest, 1));
-#else
-        static const uint32_t MAX_RASTER_TILE_BYTES = 128; // 8 pixels * 16 bytes per pixel
-
-        OSALIGNSIMD(uint8_t) soaTile[MAX_RASTER_TILE_BYTES];
-        OSALIGNSIMD(uint8_t) aosTile[MAX_RASTER_TILE_BYTES];
-
-        // Convert from SrcFormat --> DstFormat
-        simdvector src;
-        LoadSOA<SrcFormat>(pSrc, src);
-        StoreSOA<DstFormat>(src, soaTile);
-
-        // Convert from SOA --> AOS
-        FormatTraits<DstFormat>::TransposeT::Transpose(soaTile, aosTile);
-
-        // Store data into destination but don't overwrite the X8 bits
-        // Each 4-pixel row is 16-bytes
-        simd4scalari *pZRow01 = (simd4scalari*)aosTile;
-        simd4scalari vQuad00 = SIMD128::load_si(pZRow01);
-        simd4scalari vQuad01 = SIMD128::load_si(pZRow01 + 1);
-
-        simd4scalari vRow00 = SIMD128::unpacklo_epi64(vQuad00, vQuad01);
-        simd4scalari vRow10 = SIMD128::unpackhi_epi64(vQuad00, vQuad01);
-
-        simd4scalari vDst0 = SIMD128::loadu_si((const simd4scalari*)ppDsts[0]);
-        simd4scalari vDst1 = SIMD128::loadu_si((const simd4scalari*)ppDsts[1]);
-
-        simd4scalari vMask = _mm_set1_epi32(0xFFFFFF);
-
-        vDst0 = SIMD128::andnot_si(vMask, vDst0);
-        vDst0 = SIMD128::or_si(vDst0, SIMD128::and_si(vRow00, vMask));
-        vDst1 = SIMD128::andnot_si(vMask, vDst1);
-        vDst1 = SIMD128::or_si(vDst1, SIMD128::and_si(vRow10, vMask));
-
-        SIMD128::storeu_si((simd4scalari*)ppDsts[0], vDst0);
-        SIMD128::storeu_si((simd4scalari*)ppDsts[1], vDst1);
-#endif
     }
 };
 
-#if USE_8x2_TILE_BACKEND
 template<SWR_FORMAT DstFormat>
 INLINE static void FlatConvert(const uint8_t* pSrc, uint8_t* pDst0, uint8_t* pDst1, uint8_t* pDst2, uint8_t* pDst3)
 {
@@ -689,7 +573,6 @@ INLINE static void FlatConvert(const uint8_t* pSrc, uint8_t* pDst0, uint8_t* pDs
     _simd_storeu2_si(reinterpret_cast<simd4scalari *>(pDst3), reinterpret_cast<simd4scalari *>(pDst2), _simd16_extract_si(final, 1));
 }
 
-#endif
 template<SWR_FORMAT DstFormat>
 INLINE static void FlatConvert(const uint8_t* pSrc, uint8_t* pDst, uint8_t* pDst1)
 {
@@ -790,7 +673,6 @@ INLINE static void FlatConvert(const uint8_t* pSrc, uint8_t* pDst, uint8_t* pDst
     _simd_storeu2_si((simd4scalari*)pDst1, (simd4scalari*)pDst, final);
 }
 
-#if USE_8x2_TILE_BACKEND
 template<SWR_FORMAT DstFormat>
 INLINE static void FlatConvertNoAlpha(const uint8_t* pSrc, uint8_t* pDst0, uint8_t* pDst1, uint8_t* pDst2, uint8_t* pDst3)
 {
@@ -854,7 +736,6 @@ INLINE static void FlatConvertNoAlpha(const uint8_t* pSrc, uint8_t* pDst0, uint8
     _simd_storeu2_si(reinterpret_cast<simd4scalari *>(pDst3), reinterpret_cast<simd4scalari *>(pDst2), _simd16_extract_si(final, 1));
 }
 
-#endif
 template<SWR_FORMAT DstFormat>
 INLINE static void FlatConvertNoAlpha(const uint8_t* pSrc, uint8_t* pDst, uint8_t* pDst1)
 {
@@ -947,11 +828,7 @@ struct ConvertPixelsSOAtoAOS<R32G32B32A32_FLOAT, B8G8R8A8_UNORM>
     template <size_t NumDests>
     INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests])
     {
-#if USE_8x2_TILE_BACKEND
         FlatConvert<B8G8R8A8_UNORM>(pSrc, ppDsts[0], ppDsts[1], ppDsts[2], ppDsts[3]);
-#else
-        FlatConvert<B8G8R8A8_UNORM>(pSrc, ppDsts[0], ppDsts[1]);
-#endif
     }
 };
 
@@ -961,11 +838,7 @@ struct ConvertPixelsSOAtoAOS<R32G32B32A32_FLOAT, B8G8R8X8_UNORM>
     template <size_t NumDests>
     INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests])
     {
-#if USE_8x2_TILE_BACKEND
         FlatConvertNoAlpha<B8G8R8X8_UNORM>(pSrc, ppDsts[0], ppDsts[1], ppDsts[2], ppDsts[3]);
-#else
-        FlatConvertNoAlpha<B8G8R8X8_UNORM>(pSrc, ppDsts[0], ppDsts[1]);
-#endif
     }
 };
 
@@ -975,11 +848,7 @@ struct ConvertPixelsSOAtoAOS < R32G32B32A32_FLOAT, B8G8R8A8_UNORM_SRGB >
     template <size_t NumDests>
     INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests])
     {
-#if USE_8x2_TILE_BACKEND
         FlatConvert<B8G8R8A8_UNORM_SRGB>(pSrc, ppDsts[0], ppDsts[1], ppDsts[2], ppDsts[3]);
-#else
-        FlatConvert<B8G8R8A8_UNORM_SRGB>(pSrc, ppDsts[0], ppDsts[1]);
-#endif
     }
 };
 
@@ -989,11 +858,7 @@ struct ConvertPixelsSOAtoAOS < R32G32B32A32_FLOAT, B8G8R8X8_UNORM_SRGB >
     template <size_t NumDests>
     INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests])
     {
-#if USE_8x2_TILE_BACKEND
         FlatConvertNoAlpha<B8G8R8X8_UNORM_SRGB>(pSrc, ppDsts[0], ppDsts[1], ppDsts[2], ppDsts[3]);
-#else
-        FlatConvertNoAlpha<B8G8R8X8_UNORM_SRGB>(pSrc, ppDsts[0], ppDsts[1]);
-#endif
     }
 };
 
@@ -1003,11 +868,7 @@ struct ConvertPixelsSOAtoAOS < R32G32B32A32_FLOAT, R8G8B8A8_UNORM >
     template <size_t NumDests>
     INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests])
     {
-#if USE_8x2_TILE_BACKEND
         FlatConvert<R8G8B8A8_UNORM>(pSrc, ppDsts[0], ppDsts[1], ppDsts[2], ppDsts[3]);
-#else
-        FlatConvert<R8G8B8A8_UNORM>(pSrc, ppDsts[0], ppDsts[1]);
-#endif
     }
 };
 
@@ -1017,11 +878,7 @@ struct ConvertPixelsSOAtoAOS < R32G32B32A32_FLOAT, R8G8B8X8_UNORM >
     template <size_t NumDests>
     INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests])
     {
-#if USE_8x2_TILE_BACKEND
         FlatConvertNoAlpha<R8G8B8X8_UNORM>(pSrc, ppDsts[0], ppDsts[1], ppDsts[2], ppDsts[3]);
-#else
-        FlatConvertNoAlpha<R8G8B8X8_UNORM>(pSrc, ppDsts[0], ppDsts[1]);
-#endif
     }
 };
 
@@ -1031,11 +888,7 @@ struct ConvertPixelsSOAtoAOS < R32G32B32A32_FLOAT, R8G8B8A8_UNORM_SRGB >
     template <size_t NumDests>
     INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests])
     {
-#if USE_8x2_TILE_BACKEND
         FlatConvert<R8G8B8A8_UNORM_SRGB>(pSrc, ppDsts[0], ppDsts[1], ppDsts[2], ppDsts[3]);
-#else
-        FlatConvert<R8G8B8A8_UNORM_SRGB>(pSrc, ppDsts[0], ppDsts[1]);
-#endif
     }
 };
 
@@ -1045,11 +898,7 @@ struct ConvertPixelsSOAtoAOS < R32G32B32A32_FLOAT, R8G8B8X8_UNORM_SRGB >
     template <size_t NumDests>
     INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests])
     {
-#if USE_8x2_TILE_BACKEND
         FlatConvertNoAlpha<R8G8B8X8_UNORM_SRGB>(pSrc, ppDsts[0], ppDsts[1], ppDsts[2], ppDsts[3]);
-#else
-        FlatConvertNoAlpha<R8G8B8X8_UNORM_SRGB>(pSrc, ppDsts[0], ppDsts[1]);
-#endif
     }
 };
 
@@ -1069,7 +918,6 @@ struct StoreRasterTile
         uint32_t x, uint32_t y,
         float outputColor[4])
     {
-#if USE_8x2_TILE_BACKEND
         typedef SimdTile_16<SrcFormat, DstFormat> SimdT;
 
         SimdT *pSrcSimdTiles = reinterpret_cast<SimdT *>(pSrc);
@@ -1083,21 +931,6 @@ struct StoreRasterTile
         uint32_t simdOffset = (y % SIMD16_TILE_Y_DIM) * SIMD16_TILE_X_DIM + (x % SIMD16_TILE_X_DIM);
 
         pSimdTile->GetSwizzledColor(simdOffset, outputColor);
-#else
-        typedef SimdTile<SrcFormat, DstFormat> SimdT;
-
-        SimdT* pSrcSimdTiles = (SimdT*)pSrc;
-
-        // Compute which simd tile we're accessing within 8x8 tile.
-        //   i.e. Compute linear simd tile coordinate given (x, y) in pixel coordinates.
-        uint32_t simdIndex = (y / SIMD_TILE_Y_DIM) * (KNOB_TILE_X_DIM / SIMD_TILE_X_DIM) + (x / SIMD_TILE_X_DIM);
-
-        SimdT* pSimdTile = &pSrcSimdTiles[simdIndex];
-
-        uint32_t simdOffset = (y % SIMD_TILE_Y_DIM) * SIMD_TILE_X_DIM + (x % SIMD_TILE_X_DIM);
-
-        pSimdTile->GetSwizzledColor(simdOffset, outputColor);
-#endif
     }
 
     //////////////////////////////////////////////////////////////////////////
@@ -1230,7 +1063,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_NONE, 8>, SrcFormat, DstFormat>
 
         uint8_t *pDst = (uint8_t*)ComputeSurfaceAddress<false, false>(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex, 
             pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface);
-#if USE_8x2_TILE_BACKEND
 
         const uint32_t dx = SIMD16_TILE_X_DIM * DST_BYTES_PER_PIXEL;
         const uint32_t dy = SIMD16_TILE_Y_DIM * pDstSurface->pitch - KNOB_TILE_X_DIM * DST_BYTES_PER_PIXEL;
@@ -1262,27 +1094,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_NONE, 8>, SrcFormat, DstFormat>
             ppDsts[2] += dy;
             ppDsts[3] += dy;
         }
-#else
-        uint8_t* ppRows[] = { pDst, pDst + pDstSurface->pitch };
-
-        for (uint32_t row = 0; row < KNOB_TILE_Y_DIM / SIMD_TILE_Y_DIM; ++row)
-        {
-            uint8_t* ppStartRows[] = { ppRows[0], ppRows[1] };
-
-            for (uint32_t col = 0; col < KNOB_TILE_X_DIM / SIMD_TILE_X_DIM; ++col)
-            {
-                // Format conversion and convert from SOA to AOS, and store the rows.
-                ConvertPixelsSOAtoAOS<SrcFormat, DstFormat>::Convert(pSrc, ppRows);
-
-                ppRows[0] += KNOB_SIMD_WIDTH * DST_BYTES_PER_PIXEL / 2;
-                ppRows[1] += KNOB_SIMD_WIDTH * DST_BYTES_PER_PIXEL / 2;
-                pSrc += SRC_BYTES_PER_PIXEL * KNOB_SIMD_WIDTH;
-            }
-
-            ppRows[0] = ppStartRows[0] + 2 * pDstSurface->pitch;
-            ppRows[1] = ppStartRows[1] + 2 * pDstSurface->pitch;
-        }
-#endif
     }
 };
 
@@ -1317,7 +1128,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_NONE, 16>, SrcFormat, DstFormat
 
         uint8_t *pDst = (uint8_t*)ComputeSurfaceAddress<false, false>(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex, 
             pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface);
-#if USE_8x2_TILE_BACKEND
 
         const uint32_t dx = SIMD16_TILE_X_DIM * DST_BYTES_PER_PIXEL;
         const uint32_t dy = SIMD16_TILE_Y_DIM * pDstSurface->pitch - KNOB_TILE_X_DIM * DST_BYTES_PER_PIXEL;
@@ -1349,27 +1159,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_NONE, 16>, SrcFormat, DstFormat
             ppDsts[2] += dy;
             ppDsts[3] += dy;
         }
-#else
-        uint8_t* ppRows[] = { pDst, pDst + pDstSurface->pitch };
-
-        for (uint32_t row = 0; row < KNOB_TILE_Y_DIM / SIMD_TILE_Y_DIM; ++row)
-        {
-            uint8_t* ppStartRows[] = { ppRows[0], ppRows[1] };
-
-            for (uint32_t col = 0; col < KNOB_TILE_X_DIM / SIMD_TILE_X_DIM; ++col)
-            {
-                // Format conversion and convert from SOA to AOS, and store the rows.
-                ConvertPixelsSOAtoAOS<SrcFormat, DstFormat>::Convert(pSrc, ppRows);
-
-                ppRows[0] += KNOB_SIMD_WIDTH * DST_BYTES_PER_PIXEL / 2;
-                ppRows[1] += KNOB_SIMD_WIDTH * DST_BYTES_PER_PIXEL / 2;
-                pSrc += SRC_BYTES_PER_PIXEL * KNOB_SIMD_WIDTH;
-            }
-
-            ppRows[0] = ppStartRows[0] + 2 * pDstSurface->pitch;
-            ppRows[1] = ppStartRows[1] + 2 * pDstSurface->pitch;
-        }
-#endif
     }
 };
 
@@ -1404,7 +1193,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_NONE, 32>, SrcFormat, DstFormat
 
         uint8_t *pDst = (uint8_t*)ComputeSurfaceAddress<false, false>(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex,
             pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface);
-#if USE_8x2_TILE_BACKEND
 
         const uint32_t dx = SIMD16_TILE_X_DIM * DST_BYTES_PER_PIXEL;
         const uint32_t dy = SIMD16_TILE_Y_DIM * pDstSurface->pitch - KNOB_TILE_X_DIM * DST_BYTES_PER_PIXEL;
@@ -1436,27 +1224,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_NONE, 32>, SrcFormat, DstFormat
             ppDsts[2] += dy;
             ppDsts[3] += dy;
         }
-#else
-        uint8_t* ppRows[] = { pDst, pDst + pDstSurface->pitch };
-
-        for (uint32_t row = 0; row < KNOB_TILE_Y_DIM / SIMD_TILE_Y_DIM; ++row)
-        {
-            uint8_t* ppStartRows[] = { ppRows[0], ppRows[1] };
-
-            for (uint32_t col = 0; col < KNOB_TILE_X_DIM / SIMD_TILE_X_DIM; ++col)
-            {
-                // Format conversion and convert from SOA to AOS, and store the rows.
-                ConvertPixelsSOAtoAOS<SrcFormat, DstFormat>::Convert(pSrc, ppRows);
-
-                ppRows[0] += KNOB_SIMD_WIDTH * DST_BYTES_PER_PIXEL / 2;
-                ppRows[1] += KNOB_SIMD_WIDTH * DST_BYTES_PER_PIXEL / 2;
-                pSrc += SRC_BYTES_PER_PIXEL * KNOB_SIMD_WIDTH;
-            }
-
-            ppRows[0] = ppStartRows[0] + 2 * pDstSurface->pitch;
-            ppRows[1] = ppStartRows[1] + 2 * pDstSurface->pitch;
-        }
-#endif
     }
 };
 
@@ -1470,10 +1237,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_NONE, 64>, SrcFormat, DstFormat
     static const size_t SRC_BYTES_PER_PIXEL = FormatTraits<SrcFormat>::bpp / 8;
     static const size_t DST_BYTES_PER_PIXEL = FormatTraits<DstFormat>::bpp / 8;
     static const size_t MAX_DST_COLUMN_BYTES = 16;
-#if !USE_8x2_TILE_BACKEND
-    static const size_t SRC_COLUMN_BYTES = KNOB_SIMD_WIDTH * SRC_BYTES_PER_PIXEL;
-    static const size_t DST_COLUMN_BYTES_PER_SRC = KNOB_SIMD_WIDTH * DST_BYTES_PER_PIXEL / 2;
-#endif
 
     //////////////////////////////////////////////////////////////////////////
     /// @brief Stores an 8x8 raster tile to the destination surface.
@@ -1496,7 +1259,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_NONE, 64>, SrcFormat, DstFormat
 
         uint8_t *pDst = (uint8_t*)ComputeSurfaceAddress<false, false>(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex,
             pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface);
-#if USE_8x2_TILE_BACKEND
 
         const uint32_t dx = SIMD16_TILE_X_DIM * DST_BYTES_PER_PIXEL;
         const uint32_t dy = SIMD16_TILE_Y_DIM * pDstSurface->pitch;
@@ -1530,43 +1292,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_NONE, 64>, SrcFormat, DstFormat
                 ppDsts[i] += dy;
             }
         }
-#else
-        uint8_t* ppDsts[] =
-        {
-            pDst,                                               // row 0, col 0
-            pDst + pDstSurface->pitch,                          // row 1, col 0
-            pDst + MAX_DST_COLUMN_BYTES,                        // row 0, col 1
-            pDst + pDstSurface->pitch + MAX_DST_COLUMN_BYTES,   // row 1, col 1
-        };
-
-        for (uint32_t row = 0; row < KNOB_TILE_Y_DIM / SIMD_TILE_Y_DIM; ++row)
-        {
-            uint8_t* ppStartRows[] =
-            {
-                ppDsts[0],
-                ppDsts[1],
-                ppDsts[2],
-                ppDsts[3],
-            };
-
-            for (uint32_t col = 0; col < KNOB_TILE_X_DIM / SIMD_TILE_X_DIM; ++col)
-            {
-                // Format conversion and convert from SOA to AOS, and store the rows.
-                ConvertPixelsSOAtoAOS<SrcFormat, DstFormat>::Convert(pSrc, ppDsts);
-
-                ppDsts[0] += DST_COLUMN_BYTES_PER_SRC;
-                ppDsts[1] += DST_COLUMN_BYTES_PER_SRC;
-                ppDsts[2] += DST_COLUMN_BYTES_PER_SRC;
-                ppDsts[3] += DST_COLUMN_BYTES_PER_SRC;
-                pSrc += SRC_COLUMN_BYTES;
-            }
-
-            ppDsts[0] = ppStartRows[0] + 2 * pDstSurface->pitch;
-            ppDsts[1] = ppStartRows[1] + 2 * pDstSurface->pitch;
-            ppDsts[2] = ppStartRows[2] + 2 * pDstSurface->pitch;
-            ppDsts[3] = ppStartRows[3] + 2 * pDstSurface->pitch;
-        }
-#endif
     }
 };
 
@@ -1580,10 +1305,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_NONE, 128>, SrcFormat, DstForma
     static const size_t SRC_BYTES_PER_PIXEL = FormatTraits<SrcFormat>::bpp / 8;
     static const size_t DST_BYTES_PER_PIXEL = FormatTraits<DstFormat>::bpp / 8;
     static const size_t MAX_DST_COLUMN_BYTES = 16;
-#if !USE_8x2_TILE_BACKEND
-    static const size_t SRC_COLUMN_BYTES = KNOB_SIMD_WIDTH * SRC_BYTES_PER_PIXEL;
-    static const size_t DST_COLUMN_BYTES_PER_SRC = KNOB_SIMD_WIDTH * DST_BYTES_PER_PIXEL / 2;
-#endif
 
     //////////////////////////////////////////////////////////////////////////
     /// @brief Stores an 8x8 raster tile to the destination surface.
@@ -1606,7 +1327,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_NONE, 128>, SrcFormat, DstForma
 
         uint8_t *pDst = (uint8_t*)ComputeSurfaceAddress<false, false>(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex,
             pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface);
-#if USE_8x2_TILE_BACKEND
 
         const uint32_t dx = SIMD16_TILE_X_DIM * DST_BYTES_PER_PIXEL;
         const uint32_t dy = SIMD16_TILE_Y_DIM * pDstSurface->pitch;
@@ -1648,51 +1368,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_NONE, 128>, SrcFormat, DstForma
                 ppDsts[i] += dy;
             }
         }
-#else
-        struct DstPtrs
-        {
-            uint8_t* ppDsts[8];
-        } ptrs;
-
-        // Need 8 pointers, 4 columns of 2 rows each
-        for (uint32_t y = 0; y < 2; ++y)
-        {
-            for (uint32_t x = 0; x < 4; ++x)
-            {
-                ptrs.ppDsts[x * 2 + y] = pDst + y * pDstSurface->pitch + x * MAX_DST_COLUMN_BYTES;
-            }
-        }
-
-        for (uint32_t row = 0; row < KNOB_TILE_Y_DIM / SIMD_TILE_Y_DIM; ++row)
-        {
-            DstPtrs startPtrs = ptrs;
-
-            for (uint32_t col = 0; col < KNOB_TILE_X_DIM / SIMD_TILE_X_DIM; ++col)
-            {
-                // Format conversion and convert from SOA to AOS, and store the rows.
-                ConvertPixelsSOAtoAOS<SrcFormat, DstFormat>::Convert(pSrc, ptrs.ppDsts);
-
-                ptrs.ppDsts[0] += DST_COLUMN_BYTES_PER_SRC;
-                ptrs.ppDsts[1] += DST_COLUMN_BYTES_PER_SRC;
-                ptrs.ppDsts[2] += DST_COLUMN_BYTES_PER_SRC;
-                ptrs.ppDsts[3] += DST_COLUMN_BYTES_PER_SRC;
-                ptrs.ppDsts[4] += DST_COLUMN_BYTES_PER_SRC;
-                ptrs.ppDsts[5] += DST_COLUMN_BYTES_PER_SRC;
-                ptrs.ppDsts[6] += DST_COLUMN_BYTES_PER_SRC;
-                ptrs.ppDsts[7] += DST_COLUMN_BYTES_PER_SRC;
-                pSrc += SRC_COLUMN_BYTES;
-            }
-
-            ptrs.ppDsts[0] = startPtrs.ppDsts[0] + 2 * pDstSurface->pitch;
-            ptrs.ppDsts[1] = startPtrs.ppDsts[1] + 2 * pDstSurface->pitch;
-            ptrs.ppDsts[2] = startPtrs.ppDsts[2] + 2 * pDstSurface->pitch;
-            ptrs.ppDsts[3] = startPtrs.ppDsts[3] + 2 * pDstSurface->pitch;
-            ptrs.ppDsts[4] = startPtrs.ppDsts[4] + 2 * pDstSurface->pitch;
-            ptrs.ppDsts[5] = startPtrs.ppDsts[5] + 2 * pDstSurface->pitch;
-            ptrs.ppDsts[6] = startPtrs.ppDsts[6] + 2 * pDstSurface->pitch;
-            ptrs.ppDsts[7] = startPtrs.ppDsts[7] + 2 * pDstSurface->pitch;
-        }
-#endif
     }
 };
 
@@ -1728,7 +1403,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_MODE_YMAJOR, 8>, SrcFormat, Dst
 
         // TileY is a column-major tiling mode where each 4KB tile consist of 8 columns of 32 x 16B rows.
         // We can compute the offsets to each column within the raster tile once and increment from these.
-#if USE_8x2_TILE_BACKEND
         // There will be 4 8x2 simd tiles in an 8x8 raster tile.
         uint8_t *pDst = (uint8_t*)ComputeSurfaceAddress<false, false>(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex,
             pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface);
@@ -1758,32 +1432,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_MODE_YMAJOR, 8>, SrcFormat, Dst
             ppDsts[2] += dy;
             ppDsts[3] += dy;
         }
-#else
-        // There will be 8 4x2 simd tiles in an 8x8 raster tile.
-        uint8_t* pCol0 = (uint8_t*)ComputeSurfaceAddress<false, false>(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex,
-            pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface);
-
-        // Increment by a whole SIMD. 4x2 for AVX. 2x2 for SSE.
-        uint32_t pSrcInc = (FormatTraits<SrcFormat>::bpp * KNOB_SIMD_WIDTH) / 8;
-
-        // The Hot Tile uses a row-major tiling mode and has a larger memory footprint. So we iterate in a row-major pattern.
-        for (uint32_t row = 0; row < KNOB_TILE_Y_DIM; row += SIMD_TILE_Y_DIM)
-        {
-            uint32_t rowOffset = row * DestRowWidthBytes;
-
-            uint8_t* pRow = pCol0 + rowOffset;
-            uint8_t* ppDsts[] = { pRow, pRow + DestRowWidthBytes };
-
-            ConvertPixelsSOAtoAOS<SrcFormat, DstFormat>::Convert(pSrc, ppDsts);
-            pSrc += pSrcInc;
-
-            ppDsts[0] += DestRowWidthBytes / 4;
-            ppDsts[1] += DestRowWidthBytes / 4;
-
-            ConvertPixelsSOAtoAOS<SrcFormat, DstFormat>::Convert(pSrc, ppDsts);
-            pSrc += pSrcInc;
-        }
-#endif
     }
 };
 
@@ -1819,7 +1467,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_MODE_YMAJOR, 16>, SrcFormat, Ds
 
         // TileY is a column-major tiling mode where each 4KB tile consist of 8 columns of 32 x 16B rows.
         // We can compute the offsets to each column within the raster tile once and increment from these.
-#if USE_8x2_TILE_BACKEND
         // There will be 4 8x2 simd tiles in an 8x8 raster tile.
         uint8_t *pDst = (uint8_t*)ComputeSurfaceAddress<false, false>(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex,
             pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface);
@@ -1849,32 +1496,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_MODE_YMAJOR, 16>, SrcFormat, Ds
             ppDsts[2] += dy;
             ppDsts[3] += dy;
         }
-#else
-        // There will be 8 4x2 simd tiles in an 8x8 raster tile.
-        uint8_t* pCol0 = (uint8_t*)ComputeSurfaceAddress<false, false>(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex,
-            pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface);
-
-        // Increment by a whole SIMD. 4x2 for AVX. 2x2 for SSE.
-        uint32_t pSrcInc = (FormatTraits<SrcFormat>::bpp * KNOB_SIMD_WIDTH) / 8;
-
-        // The Hot Tile uses a row-major tiling mode and has a larger memory footprint. So we iterate in a row-major pattern.
-        for (uint32_t row = 0; row < KNOB_TILE_Y_DIM; row += SIMD_TILE_Y_DIM)
-        {
-            uint32_t rowOffset = row * DestRowWidthBytes;
-
-            uint8_t* pRow = pCol0 + rowOffset;
-            uint8_t* ppDsts[] = { pRow, pRow + DestRowWidthBytes };
-
-            ConvertPixelsSOAtoAOS<SrcFormat, DstFormat>::Convert(pSrc, ppDsts);
-            pSrc += pSrcInc;
-
-            ppDsts[0] += DestRowWidthBytes / 2;
-            ppDsts[1] += DestRowWidthBytes / 2;
-
-            ConvertPixelsSOAtoAOS<SrcFormat, DstFormat>::Convert(pSrc, ppDsts);
-            pSrc += pSrcInc;
-        }
-#endif
     }
 };
 
@@ -1911,7 +1532,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_MODE_XMAJOR, 32>, SrcFormat, Ds
 
         // TileX is a row-major tiling mode where each 4KB tile consist of 8 x 512B rows.
         // We can compute the offsets to each column within the raster tile once and increment from these.
-#if USE_8x2_TILE_BACKEND
         uint8_t *pDst = (uint8_t*)ComputeSurfaceAddress<false, false>(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex,
             pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface);
 
@@ -1945,28 +1565,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_MODE_XMAJOR, 32>, SrcFormat, Ds
             ppDsts[2] += dy;
             ppDsts[3] += dy;
         }
-#else
-        uint8_t *pRow0 = (uint8_t*)ComputeSurfaceAddress<false, false>(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex,
-            pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface);
-        uint8_t* pRow1 = pRow0 + DestRowWidthBytes;
-
-        for (uint32_t row = 0; row < KNOB_TILE_Y_DIM; row += SIMD_TILE_Y_DIM)
-        {
-            for (uint32_t col = 0; col < KNOB_TILE_X_DIM; col += SIMD_TILE_X_DIM)
-            {
-                uint32_t xRowOffset = col * (FormatTraits<DstFormat>::bpp / 8);
-
-                uint8_t* ppDsts[] = { pRow0 + xRowOffset, pRow1 + xRowOffset };
-                ConvertPixelsSOAtoAOS<SrcFormat, DstFormat>::Convert(pSrc, ppDsts);
-
-                // Increment by a whole SIMD. 4x2 for AVX. 2x2 for SSE.
-                pSrc += (FormatTraits<SrcFormat>::bpp * KNOB_SIMD_WIDTH) / 8;
-            }
-
-            pRow0 += (DestRowWidthBytes * 2);
-            pRow1 += (DestRowWidthBytes * 2);
-        }
-#endif
     }
 };
 
@@ -2003,7 +1601,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_MODE_YMAJOR, 32>, SrcFormat, Ds
 
         // TileY is a column-major tiling mode where each 4KB tile consist of 8 columns of 32 x 16B rows.
         // We can compute the offsets to each column within the raster tile once and increment from these.
-#if USE_8x2_TILE_BACKEND
         // There will be 4 8x2 simd tiles in an 8x8 raster tile.
         uint8_t *pDst = (uint8_t*)ComputeSurfaceAddress<false, false>(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex,
             pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface);
@@ -2034,32 +1631,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_MODE_YMAJOR, 32>, SrcFormat, Ds
             ppDsts[2] += dy;
             ppDsts[3] += dy;
         }
-#else
-        // There will be 8 4x2 simd tiles in an 8x8 raster tile.
-        uint8_t* pCol0 = (uint8_t*)ComputeSurfaceAddress<false, false>(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex,
-            pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface);
-
-        // Increment by a whole SIMD. 4x2 for AVX. 2x2 for SSE.
-        uint32_t pSrcInc = (FormatTraits<SrcFormat>::bpp * KNOB_SIMD_WIDTH) / 8;
-
-        // The Hot Tile uses a row-major tiling mode and has a larger memory footprint. So we iterate in a row-major pattern.
-        for (uint32_t row = 0; row < KNOB_TILE_Y_DIM; row += SIMD_TILE_Y_DIM)
-        {
-            uint32_t rowOffset = row * DestRowWidthBytes;
-
-            uint8_t* pRow = pCol0 + rowOffset;
-            uint8_t* ppDsts[] = { pRow, pRow + DestRowWidthBytes };
-
-            ConvertPixelsSOAtoAOS<SrcFormat, DstFormat>::Convert(pSrc, ppDsts);
-            pSrc += pSrcInc;
-
-            ppDsts[0] += DestColumnBytes;
-            ppDsts[1] += DestColumnBytes;
-
-            ConvertPixelsSOAtoAOS<SrcFormat, DstFormat>::Convert(pSrc, ppDsts);
-            pSrc += pSrcInc;
-        }
-#endif
     }
 };
 
@@ -2096,7 +1667,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_MODE_YMAJOR, 64>, SrcFormat, Ds
 
         // TileY is a column-major tiling mode where each 4KB tile consist of 8 columns of 32 x 16B rows.
         // We can compute the offsets to each column within the raster tile once and increment from these.
-#if USE_8x2_TILE_BACKEND
         // There will be 4 8x2 simd tiles in an 8x8 raster tile.
         uint8_t *pDst = (uint8_t*)ComputeSurfaceAddress<false, false>(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex,
             pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface);
@@ -2131,40 +1701,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_MODE_YMAJOR, 64>, SrcFormat, Ds
                 ppDsts[i] += dy;
             }
         }
-#else
-        // There will be 8 4x2 simd tiles in an 8x8 raster tile.
-        uint8_t* pCol0 = (uint8_t*)ComputeSurfaceAddress<false, false>(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex,
-            pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface);
-        uint8_t* pCol1 = pCol0 + DestColumnBytes;
-
-        // There are 4 columns, each 2 pixels wide when we have 64bpp pixels.
-        // Increment by a whole SIMD. 4x2 for AVX. 2x2 for SSE.
-        uint32_t pSrcInc = (FormatTraits<SrcFormat>::bpp * KNOB_SIMD_WIDTH) / 8;
-
-        // The Hot Tile uses a row-major tiling mode and has a larger memory footprint. So we iterate in a row-major pattern.
-        for (uint32_t row = 0; row < KNOB_TILE_Y_DIM; row += SIMD_TILE_Y_DIM)
-        {
-            uint32_t rowOffset = row * DestRowWidthBytes;
-            uint8_t* ppDsts[] =
-            {
-                pCol0 + rowOffset,
-                pCol0 + rowOffset + DestRowWidthBytes,
-                pCol1 + rowOffset,
-                pCol1 + rowOffset + DestRowWidthBytes,
-            };
-
-            ConvertPixelsSOAtoAOS<SrcFormat, DstFormat>::Convert(pSrc, ppDsts);
-            pSrc += pSrcInc;
-
-            ppDsts[0] += DestColumnBytes * 2;
-            ppDsts[1] += DestColumnBytes * 2;
-            ppDsts[2] += DestColumnBytes * 2;
-            ppDsts[3] += DestColumnBytes * 2;
-
-            ConvertPixelsSOAtoAOS<SrcFormat, DstFormat>::Convert(pSrc, ppDsts);
-            pSrc += pSrcInc;
-        }
-#endif
     }
 };
 
@@ -2175,22 +1711,8 @@ template<SWR_FORMAT SrcFormat, SWR_FORMAT DstFormat>
 struct OptStoreRasterTile< TilingTraits<SWR_TILE_MODE_YMAJOR, 128>, SrcFormat, DstFormat>
 {
     typedef StoreRasterTile<TilingTraits<SWR_TILE_MODE_YMAJOR, 128>, SrcFormat, DstFormat> GenericStoreTile;
-#if USE_8x2_TILE_BACKEND
-    static const size_t SRC_BYTES_PER_PIXEL = FormatTraits<SrcFormat>::bpp / 8;
-
-#else
-    static const size_t TILE_Y_COL_WIDTH_BYTES = 16;
-    static const size_t TILE_Y_ROWS = 32;
-    static const size_t TILE_Y_COL_BYTES = TILE_Y_ROWS * TILE_Y_COL_WIDTH_BYTES;
-
-    static const size_t DST_BYTES_PER_PIXEL = FormatTraits<DstFormat>::bpp / 8;
     static const size_t SRC_BYTES_PER_PIXEL = FormatTraits<SrcFormat>::bpp / 8;
-    static const size_t MAX_DST_COLUMN_BYTES = 16;
-
-    static const size_t SRC_COLUMN_BYTES = KNOB_SIMD_WIDTH * SRC_BYTES_PER_PIXEL;
-    static const size_t DST_COLUMN_BYTES_PER_SRC = TILE_Y_COL_BYTES * 4;
 
-#endif
     //////////////////////////////////////////////////////////////////////////
     /// @brief Stores an 8x8 raster tile to the destination surface.
     /// @param pSrc - Pointer to raster tile.
@@ -2201,10 +1723,8 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_MODE_YMAJOR, 128>, SrcFormat, D
         SWR_SURFACE_STATE* pDstSurface,
         uint32_t x, uint32_t y, uint32_t sampleNum, uint32_t renderTargetArrayIndex)
     {
-#if USE_8x2_TILE_BACKEND
         static const uint32_t DestRowWidthBytes = 16;                    // 16B rows
         static const uint32_t DestColumnBytes = DestRowWidthBytes * 32;  // 16B x 32 rows.
-#endif
 
         // Punt non-full tiles to generic store
         uint32_t lodWidth = std::max(pDstSurface->width >> pDstSurface->lod, 1U);
@@ -2217,7 +1737,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_MODE_YMAJOR, 128>, SrcFormat, D
 
         // TileY is a column-major tiling mode where each 4KB tile consist of 8 columns of 32 x 16B rows.
         // We can compute the offsets to each column within the raster tile once and increment from these.
-#if USE_8x2_TILE_BACKEND
         // There will be 4 8x2 simd tiles in an 8x8 raster tile.
         uint8_t *pDst = (uint8_t*)ComputeSurfaceAddress<false, false>(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex,
             pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface);
@@ -2260,54 +1779,6 @@ struct OptStoreRasterTile< TilingTraits<SWR_TILE_MODE_YMAJOR, 128>, SrcFormat, D
                 ppDsts[i] += dy;
             }
         }
-#else
-        // There will be 8 4x2 simd tiles in an 8x8 raster tile.
-        uint8_t* pDst = (uint8_t*)ComputeSurfaceAddress<false, false>(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex,
-            pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface);
-        struct DstPtrs
-        {
-            uint8_t* ppDsts[8];
-        } ptrs;
-
-        // Need 8 pointers, 4 columns of 2 rows each
-        for (uint32_t y = 0; y < 2; ++y)
-        {
-            for (uint32_t x = 0; x < 4; ++x)
-            {
-                ptrs.ppDsts[x * 2 + y] = pDst + y * TILE_Y_COL_WIDTH_BYTES + x * TILE_Y_COL_BYTES;
-            }
-        }
-
-        for (uint32_t row = 0; row < KNOB_TILE_Y_DIM / SIMD_TILE_Y_DIM; ++row)
-        {
-            DstPtrs startPtrs = ptrs;
-
-            for (uint32_t col = 0; col < KNOB_TILE_X_DIM / SIMD_TILE_X_DIM; ++col)
-            {
-                // Format conversion and convert from SOA to AOS, and store the rows.
-                ConvertPixelsSOAtoAOS<SrcFormat, DstFormat>::Convert(pSrc, ptrs.ppDsts);
-
-                ptrs.ppDsts[0] += DST_COLUMN_BYTES_PER_SRC;
-                ptrs.ppDsts[1] += DST_COLUMN_BYTES_PER_SRC;
-                ptrs.ppDsts[2] += DST_COLUMN_BYTES_PER_SRC;
-                ptrs.ppDsts[3] += DST_COLUMN_BYTES_PER_SRC;
-                ptrs.ppDsts[4] += DST_COLUMN_BYTES_PER_SRC;
-                ptrs.ppDsts[5] += DST_COLUMN_BYTES_PER_SRC;
-                ptrs.ppDsts[6] += DST_COLUMN_BYTES_PER_SRC;
-                ptrs.ppDsts[7] += DST_COLUMN_BYTES_PER_SRC;
-                pSrc += SRC_COLUMN_BYTES;
-            }
-
-            ptrs.ppDsts[0] = startPtrs.ppDsts[0] + 2 * TILE_Y_COL_WIDTH_BYTES;
-            ptrs.ppDsts[1] = startPtrs.ppDsts[1] + 2 * TILE_Y_COL_WIDTH_BYTES;
-            ptrs.ppDsts[2] = startPtrs.ppDsts[2] + 2 * TILE_Y_COL_WIDTH_BYTES;
-            ptrs.ppDsts[3] = startPtrs.ppDsts[3] + 2 * TILE_Y_COL_WIDTH_BYTES;
-            ptrs.ppDsts[4] = startPtrs.ppDsts[4] + 2 * TILE_Y_COL_WIDTH_BYTES;
-            ptrs.ppDsts[5] = startPtrs.ppDsts[5] + 2 * TILE_Y_COL_WIDTH_BYTES;
-            ptrs.ppDsts[6] = startPtrs.ppDsts[6] + 2 * TILE_Y_COL_WIDTH_BYTES;
-            ptrs.ppDsts[7] = startPtrs.ppDsts[7] + 2 * TILE_Y_COL_WIDTH_BYTES;
-        }
-#endif
     }
 };