From: Jan Zielinski Date: Fri, 26 Jul 2019 14:43:50 +0000 (+0200) Subject: swr/rasterizer: enable using AOS vertex data format X-Git-Tag: upstream/19.3.0~3135 X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=365ad367f11692c1637ce4c0c4d06f5a4d4bcf38;p=platform%2Fupstream%2Fmesa.git swr/rasterizer: enable using AOS vertex data format Reviewed-by: Alok Hota --- diff --git a/src/gallium/drivers/swr/rasterizer/core/frontend.cpp b/src/gallium/drivers/swr/rasterizer/core/frontend.cpp index d8703e5..a27b33d 100644 --- a/src/gallium/drivers/swr/rasterizer/core/frontend.cpp +++ b/src/gallium/drivers/swr/rasterizer/core/frontend.cpp @@ -42,15 +42,6 @@ #include ////////////////////////////////////////////////////////////////////////// -/// @brief Helper macro to generate a bitmask -static INLINE uint32_t GenMask(uint32_t numBits) -{ - SWR_ASSERT( - numBits <= (sizeof(uint32_t) * 8), "Too many bits (%d) for %s", numBits, __FUNCTION__); - return ((1U << numBits) - 1); -} - -////////////////////////////////////////////////////////////////////////// /// @brief FE handler for SwrSync. /// @param pContext - pointer to SWR context. /// @param pDC - pointer to draw context. @@ -400,7 +391,7 @@ uint32_t GetNumVerts(PRIMITIVE_TOPOLOGY mode, uint32_t numPrims) /// @brief Return number of verts per primitive. /// @param topology - topology /// @param includeAdjVerts - include adjacent verts in primitive vertices -INLINE uint32_t NumVertsPerPrim(PRIMITIVE_TOPOLOGY topology, bool includeAdjVerts) +uint32_t NumVertsPerPrim(PRIMITIVE_TOPOLOGY topology, bool includeAdjVerts) { uint32_t numVerts = 0; switch (topology) diff --git a/src/gallium/drivers/swr/rasterizer/core/frontend.h b/src/gallium/drivers/swr/rasterizer/core/frontend.h index 38fe77e..a6d9fb5 100644 --- a/src/gallium/drivers/swr/rasterizer/core/frontend.h +++ b/src/gallium/drivers/swr/rasterizer/core/frontend.h @@ -31,6 +31,16 @@ #include "common/simdintrin.h" #include +////////////////////////////////////////////////////////////////////////// +/// @brief Helper macro to generate a bitmask +static INLINE uint32_t + GenMask(uint32_t numBits) +{ + SWR_ASSERT( + numBits <= (sizeof(uint32_t) * 8), "Too many bits (%d) for %s", numBits, __FUNCTION__); + return ((1U << numBits) - 1); +} + // Calculates the A and B coefficients for the 3 edges of the triangle // // maths for edge equations: diff --git a/src/gallium/drivers/swr/rasterizer/core/pa.h b/src/gallium/drivers/swr/rasterizer/core/pa.h index e19c8ea..635bf19 100644 --- a/src/gallium/drivers/swr/rasterizer/core/pa.h +++ b/src/gallium/drivers/swr/rasterizer/core/pa.h @@ -1282,11 +1282,12 @@ struct PA_TESS : PA_STATE uint32_t* (&in_ppIndices)[3], uint32_t in_numPrims, PRIMITIVE_TOPOLOGY in_binTopology, - uint32_t numVertsPerPrim) : + uint32_t numVertsPerPrim, + bool SOA = true) : PA_STATE(in_pDC, nullptr, 0, in_vertexStride, numVertsPerPrim), m_pVertexData(in_pVertData), m_attributeStrideInVectors(in_attributeStrideInVectors), - m_numAttributes(in_numAttributes), m_numPrims(in_numPrims) + m_numAttributes(in_numAttributes), m_numPrims(in_numPrims), m_SOA(SOA) { #if USE_SIMD16_FRONTEND m_vPrimId = _simd16_setzero_si(); @@ -1363,8 +1364,17 @@ struct PA_TESS : PA_STATE SIMDSCALARI mask = GenPrimMask(numPrimsToAssemble); - const float* pBaseAttrib = - (const float*)&m_pVertexData[slot * m_attributeStrideInVectors * 4]; + const float* pBaseAttrib; + if (m_SOA) + { + pBaseAttrib = (const float*)&m_pVertexData[slot * m_attributeStrideInVectors * 4]; + } + else + { + const float* pVertData = (const float*)m_pVertexData; + pBaseAttrib = pVertData + slot * 4; + } + for (uint32_t i = 0; i < m_numVertsPerPrim; ++i) { #if USE_SIMD16_FRONTEND @@ -1393,7 +1403,14 @@ struct PA_TESS : PA_STATE _simd_castsi_ps(mask), 4); // gcc doesn't like sizeof(float) #endif - pBase += m_attributeStrideInVectors * SIMD_WIDTH; + if (m_SOA) + { + pBase += m_attributeStrideInVectors * SIMD_WIDTH; + } + else + { + pBase += sizeof(float); + } } } @@ -1413,12 +1430,25 @@ struct PA_TESS : PA_STATE SIMDSCALARI mask = GenPrimMask(numPrimsToAssemble); - const float* pBaseAttrib = - (const float*)&m_pVertexData[slot * m_attributeStrideInVectors * 4]; + const float* pBaseAttrib; + if (m_SOA) + { + pBaseAttrib = (const float*)&m_pVertexData[slot * m_attributeStrideInVectors * 4]; + } + else + { + const float* pVertData = (const float*)m_pVertexData; + pBaseAttrib = pVertData + slot * 4; + } + for (uint32_t i = 0; i < m_numVertsPerPrim; ++i) { #if USE_SIMD16_FRONTEND SIMDSCALARI indices = _simd16_load_si((const SIMDSCALARI*)m_ppIndices[i]); + if (!m_SOA) + { + indices = _simd16_mul_epi32(indices, _simd16_set1_epi32(vertexStride / 4)); + } #else SIMDSCALARI indices = _simd_load_si((const SIMDSCALARI*)m_ppIndices[i]); #endif @@ -1440,7 +1470,14 @@ struct PA_TESS : PA_STATE 4 /* gcc doesn't like sizeof(float) */); verts[i].v[c] = _simd16_insert_ps(_simd16_setzero_ps(), temp, 0); #endif - pBase += m_attributeStrideInVectors * SIMD_WIDTH; + if (m_SOA) + { + pBase += m_attributeStrideInVectors * SIMD_WIDTH; + } + else + { + pBase++; + } } } @@ -1455,13 +1492,25 @@ struct PA_TESS : PA_STATE SWR_ASSERT(primIndex < PA_TESS::NumPrims()); - const float* pVertDataBase = - (const float*)&m_pVertexData[slot * m_attributeStrideInVectors * 4]; + const float* pVertDataBase; + if (m_SOA) + { + pVertDataBase = (const float*)&m_pVertexData[slot * m_attributeStrideInVectors * 4]; + } + else + { + const float* pVertData = (const float*)m_pVertexData; + pVertDataBase = pVertData + slot * 4; + }; for (uint32_t i = 0; i < m_numVertsPerPrim; ++i) { #if USE_SIMD16_FRONTEND uint32_t index = useAlternateOffset ? m_ppIndices[i][primIndex + SIMD_WIDTH_DIV2] : m_ppIndices[i][primIndex]; + if (!m_SOA) + { + index *= (vertexStride / 4); + } #else uint32_t index = m_ppIndices[i][primIndex]; #endif @@ -1471,8 +1520,16 @@ struct PA_TESS : PA_STATE for (uint32_t c = 0; c < 4; ++c) { pVert[c] = pVertData[index]; - pVertData += m_attributeStrideInVectors * SIMD_WIDTH; + if (m_SOA) + { + pVertData += m_attributeStrideInVectors * SIMD_WIDTH; + } + else + { + pVertData++; + } } + } } @@ -1535,6 +1592,8 @@ private: #endif SIMDVERTEX junkVertex; // junk SIMDVERTEX for unimplemented API SIMDMASK junkIndices; // temporary index store for unused virtual function + + bool m_SOA; }; // Primitive Assembler factory class, responsible for creating and initializing the correct