/*------------------------------------------------------------------------- * drawElements Quality Program Reference Renderer * ----------------------------------------------- * * Copyright 2014 The Android Open Source Project * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. * *//*! * \file * \brief Reference implementation for per-fragment operations. *//*--------------------------------------------------------------------*/ #include "rrFragmentOperations.hpp" #include "tcuVectorUtil.hpp" #include "tcuTextureUtil.hpp" using tcu::IVec2; using tcu::Vec3; using tcu::Vec4; using tcu::IVec4; using tcu::UVec4; using tcu::min; using tcu::max; using tcu::clamp; using de::min; using de::max; using de::clamp; namespace rr { // Return oldValue with the bits indicated by mask replaced by corresponding bits of newValue. static inline int maskedBitReplace (int oldValue, int newValue, deUint32 mask) { return (oldValue & ~mask) | (newValue & mask); } static inline bool isInsideRect (const IVec2& point, const WindowRectangle& rect) { return de::inBounds(point.x(), rect.left, rect.left + rect.width) && de::inBounds(point.y(), rect.bottom, rect.bottom + rect.height); } static inline Vec4 unpremultiply (const Vec4& v) { if (v.w() > 0.0f) return Vec4(v.x()/v.w(), v.y()/v.w(), v.z()/v.w(), v.w()); else { DE_ASSERT(v.x() == 0.0f && v.y() == 0.0f && v.z() == 0.0f); return Vec4(0.0f, 0.0f, 0.0f, 0.0f); } } void clearMultisampleColorBuffer (const tcu::PixelBufferAccess& dst, const Vec4& v, const WindowRectangle& r) { tcu::clear(tcu::getSubregion(dst, 0, r.left, r.bottom, dst.getWidth(), r.width, r.height), v); } void clearMultisampleColorBuffer (const tcu::PixelBufferAccess& dst, const IVec4& v, const WindowRectangle& r) { tcu::clear(tcu::getSubregion(dst, 0, r.left, r.bottom, dst.getWidth(), r.width, r.height), v); } void clearMultisampleColorBuffer (const tcu::PixelBufferAccess& dst, const UVec4& v, const WindowRectangle& r) { tcu::clear(tcu::getSubregion(dst, 0, r.left, r.bottom, dst.getWidth(), r.width, r.height), v.cast()); } void clearMultisampleDepthBuffer (const tcu::PixelBufferAccess& dst, float v, const WindowRectangle& r) { tcu::clearDepth(tcu::getSubregion(dst, 0, r.left, r.bottom, dst.getWidth(), r.width, r.height), v); } void clearMultisampleStencilBuffer (const tcu::PixelBufferAccess& dst, int v, const WindowRectangle& r) { tcu::clearStencil(tcu::getSubregion(dst, 0, r.left, r.bottom, dst.getWidth(), r.width, r.height), v); } FragmentProcessor::FragmentProcessor (void) : m_sampleRegister() { } void FragmentProcessor::executeScissorTest (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const WindowRectangle& scissorRect) { for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) { if (m_sampleRegister[regSampleNdx].isAlive) { int fragNdx = fragNdxOffset + regSampleNdx/numSamplesPerFragment; if (!isInsideRect(inputFragments[fragNdx].pixelCoord, scissorRect)) m_sampleRegister[regSampleNdx].isAlive = false; } } } void FragmentProcessor::executeStencilCompare (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const StencilState& stencilState, int numStencilBits, const tcu::ConstPixelBufferAccess& stencilBuffer) { #define SAMPLE_REGISTER_STENCIL_COMPARE(COMPARE_EXPRESSION) \ for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \ { \ if (m_sampleRegister[regSampleNdx].isAlive) \ { \ int fragSampleNdx = regSampleNdx % numSamplesPerFragment; \ const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment]; \ int stencilBufferValue = stencilBuffer.getPixStencil(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); \ int maskedRef = stencilState.compMask & clampedStencilRef; \ int maskedBuf = stencilState.compMask & stencilBufferValue; \ DE_UNREF(maskedRef); \ DE_UNREF(maskedBuf); \ \ m_sampleRegister[regSampleNdx].stencilPassed = (COMPARE_EXPRESSION); \ } \ } int clampedStencilRef = de::clamp(stencilState.ref, 0, (1< maskedBuf) break; case TESTFUNC_GEQUAL: SAMPLE_REGISTER_STENCIL_COMPARE(maskedRef >= maskedBuf) break; case TESTFUNC_EQUAL: SAMPLE_REGISTER_STENCIL_COMPARE(maskedRef == maskedBuf) break; case TESTFUNC_NOTEQUAL: SAMPLE_REGISTER_STENCIL_COMPARE(maskedRef != maskedBuf) break; default: DE_ASSERT(false); } #undef SAMPLE_REGISTER_STENCIL_COMPARE } void FragmentProcessor::executeStencilSFail (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const StencilState& stencilState, int numStencilBits, const tcu::PixelBufferAccess& stencilBuffer) { #define SAMPLE_REGISTER_SFAIL(SFAIL_EXPRESSION) \ for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \ { \ if (m_sampleRegister[regSampleNdx].isAlive && !m_sampleRegister[regSampleNdx].stencilPassed) \ { \ int fragSampleNdx = regSampleNdx % numSamplesPerFragment; \ const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment]; \ int stencilBufferValue = stencilBuffer.getPixStencil(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); \ \ stencilBuffer.setPixStencil(maskedBitReplace(stencilBufferValue, (SFAIL_EXPRESSION), stencilState.writeMask), fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); \ m_sampleRegister[regSampleNdx].isAlive = false; \ } \ } int clampedStencilRef = de::clamp(stencilState.ref, 0, (1<= (size_t)depthBuffer.getFormat().getPixelSize()); \ \ tcu::PixelBufferAccess access(depthBuffer.getFormat(), 1, 1, 1, &buffer); \ access.setPixDepth(sampleDepthFloat, 0, 0, 0); \ deUint32 sampleDepth = access.getPixelUint(0, 0, 0).x(); \ \ m_sampleRegister[regSampleNdx].depthPassed = (COMPARE_EXPRESSION); \ \ DE_UNREF(depthBufferValue); \ DE_UNREF(sampleDepth); \ } \ } if (depthBuffer.getFormat().type == tcu::TextureFormat::FLOAT || depthBuffer.getFormat().type == tcu::TextureFormat::FLOAT_UNSIGNED_INT_24_8_REV) { switch (depthFunc) { case TESTFUNC_NEVER: SAMPLE_REGISTER_DEPTH_COMPARE_F(false) break; case TESTFUNC_ALWAYS: SAMPLE_REGISTER_DEPTH_COMPARE_F(true) break; case TESTFUNC_LESS: SAMPLE_REGISTER_DEPTH_COMPARE_F(sampleDepth < depthBufferValue) break; case TESTFUNC_LEQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_F(sampleDepth <= depthBufferValue) break; case TESTFUNC_GREATER: SAMPLE_REGISTER_DEPTH_COMPARE_F(sampleDepth > depthBufferValue) break; case TESTFUNC_GEQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_F(sampleDepth >= depthBufferValue) break; case TESTFUNC_EQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_F(sampleDepth == depthBufferValue) break; case TESTFUNC_NOTEQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_F(sampleDepth != depthBufferValue) break; default: DE_ASSERT(false); } } else { switch (depthFunc) { case TESTFUNC_NEVER: SAMPLE_REGISTER_DEPTH_COMPARE_UI(false) break; case TESTFUNC_ALWAYS: SAMPLE_REGISTER_DEPTH_COMPARE_UI(true) break; case TESTFUNC_LESS: SAMPLE_REGISTER_DEPTH_COMPARE_UI(sampleDepth < depthBufferValue) break; case TESTFUNC_LEQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_UI(sampleDepth <= depthBufferValue) break; case TESTFUNC_GREATER: SAMPLE_REGISTER_DEPTH_COMPARE_UI(sampleDepth > depthBufferValue) break; case TESTFUNC_GEQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_UI(sampleDepth >= depthBufferValue) break; case TESTFUNC_EQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_UI(sampleDepth == depthBufferValue) break; case TESTFUNC_NOTEQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_UI(sampleDepth != depthBufferValue) break; default: DE_ASSERT(false); } } #undef SAMPLE_REGISTER_DEPTH_COMPARE_F #undef SAMPLE_REGISTER_DEPTH_COMPARE_UI } void FragmentProcessor::executeDepthWrite (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const tcu::PixelBufferAccess& depthBuffer) { for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) { if (m_sampleRegister[regSampleNdx].isAlive && m_sampleRegister[regSampleNdx].depthPassed) { int fragSampleNdx = regSampleNdx % numSamplesPerFragment; const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment]; const float clampedDepth = de::clamp(frag.sampleDepths[fragSampleNdx], 0.0f, 1.0f); depthBuffer.setPixDepth(clampedDepth, fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); } } } void FragmentProcessor::executeStencilDpFailAndPass (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const StencilState& stencilState, int numStencilBits, const tcu::PixelBufferAccess& stencilBuffer) { #define SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, EXPRESSION) \ for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \ { \ if (m_sampleRegister[regSampleNdx].isAlive && (CONDITION)) \ { \ int fragSampleNdx = regSampleNdx % numSamplesPerFragment; \ const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment]; \ int stencilBufferValue = stencilBuffer.getPixStencil(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); \ \ stencilBuffer.setPixStencil(maskedBitReplace(stencilBufferValue, (EXPRESSION), stencilState.writeMask), fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); \ } \ } #define SWITCH_DPFAIL_OR_DPPASS(OP_NAME, CONDITION) \ switch (stencilState.OP_NAME) \ { \ case STENCILOP_KEEP: SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, stencilBufferValue) break; \ case STENCILOP_ZERO: SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, 0) break; \ case STENCILOP_REPLACE: SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, clampedStencilRef) break; \ case STENCILOP_INCR: SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, de::clamp(stencilBufferValue+1, 0, (1<= 1.0f) return 1.0f; else if (src > 0.0f) return 1.0f - de::min(1.0f, (1.0f-dst)/src); else return 0.0f; } inline float hardlight (float src, float dst) { if (src <= 0.5f) return 2.0f*src*dst; else return 1.0f - 2.0f*(1.0f-src)*(1.0f-dst); } inline float softlight (float src, float dst) { if (src <= 0.5f) return dst - (1.0f - 2.0f*src)*dst*(1.0f-dst); else if (dst <= 0.25f) return dst + (2.0f*src - 1.0f)*dst*((16.0f*dst - 12.0f)*dst + 3.0f); else return dst + (2.0f*src - 1.0f)*(deFloatSqrt(dst)-dst); } inline float minComp (const Vec3& v) { return de::min(de::min(v.x(), v.y()), v.z()); } inline float maxComp (const Vec3& v) { return de::max(de::max(v.x(), v.y()), v.z()); } inline float luminosity (const Vec3& rgb) { return dot(rgb, Vec3(0.3f, 0.59f, 0.11f)); } inline float saturation (const Vec3& rgb) { return maxComp(rgb) - minComp(rgb); } Vec3 setLum (const Vec3& cbase, const Vec3& clum) { const float lbase = luminosity(cbase); const float llum = luminosity(clum); const float ldiff = llum - lbase; const Vec3 color = cbase + Vec3(ldiff); const float minC = minComp(color); const float maxC = maxComp(color); if (minC < 0.0f) return llum + ((color-llum)*llum / (llum != minC ? (llum-minC) : 1.0f)); else if (maxC > 1.0f) return llum + ((color-llum)*(1.0f-llum) / (llum != maxC ? (maxC-llum) : 1.0f)); else return color; } Vec3 setLumSat (const Vec3& cbase, const Vec3& csat, const Vec3& clum) { const float minbase = minComp(cbase); const float sbase = saturation(cbase); const float ssat = saturation(csat); Vec3 color = Vec3(0.0f); if (sbase > 0.0f) color = (cbase - minbase) * ssat / sbase; else color = color; return setLum(color, clum); } } // advblend void FragmentProcessor::executeAdvancedBlend (BlendEquationAdvanced equation) { using namespace advblend; #define SAMPLE_REGISTER_ADV_BLEND(FUNCTION_NAME) \ for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \ { \ if (m_sampleRegister[regSampleNdx].isAlive) \ { \ SampleData& sample = m_sampleRegister[regSampleNdx]; \ const Vec4& srcColor = sample.clampedBlendSrcColor; \ const Vec4& dstColor = sample.clampedBlendDstColor; \ const Vec3& bias = sample.blendSrcFactorRGB; \ const float p0 = sample.blendSrcFactorA; \ const float r = FUNCTION_NAME(srcColor[0], dstColor[0])*p0 + bias[0]; \ const float g = FUNCTION_NAME(srcColor[1], dstColor[1])*p0 + bias[1]; \ const float b = FUNCTION_NAME(srcColor[2], dstColor[2])*p0 + bias[2]; \ \ sample.blendedRGB = Vec3(r, g, b); \ } \ } #define SAMPLE_REGISTER_ADV_BLEND_HSL(COLOR_EXPRESSION) \ for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \ { \ if (m_sampleRegister[regSampleNdx].isAlive) \ { \ SampleData& sample = m_sampleRegister[regSampleNdx]; \ const Vec3 srcColor = sample.clampedBlendSrcColor.swizzle(0,1,2); \ const Vec3 dstColor = sample.clampedBlendDstColor.swizzle(0,1,2); \ const Vec3& bias = sample.blendSrcFactorRGB; \ const float p0 = sample.blendSrcFactorA; \ \ sample.blendedRGB = (COLOR_EXPRESSION)*p0 + bias; \ } \ } // Pre-compute factors & compute alpha \todo [2014-03-18 pyry] Re-using variable names. // \note clampedBlend*Color contains clamped & unpremultiplied colors for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) { if (m_sampleRegister[regSampleNdx].isAlive) { SampleData& sample = m_sampleRegister[regSampleNdx]; const Vec4& srcColor = sample.clampedBlendSrcColor; const Vec4& dstColor = sample.clampedBlendDstColor; const float srcA = srcColor.w(); const float dstA = dstColor.w(); const float p0 = srcA*dstA; const float p1 = srcA*(1.0f-dstA); const float p2 = dstA*(1.0f-srcA); const Vec3 bias (srcColor[0]*p1 + dstColor[0]*p2, srcColor[1]*p1 + dstColor[1]*p2, srcColor[2]*p1 + dstColor[2]*p2); sample.blendSrcFactorRGB = bias; sample.blendSrcFactorA = p0; sample.blendedA = p0 + p1 + p2; } } switch (equation) { case BLENDEQUATION_ADVANCED_MULTIPLY: SAMPLE_REGISTER_ADV_BLEND(multiply); break; case BLENDEQUATION_ADVANCED_SCREEN: SAMPLE_REGISTER_ADV_BLEND(screen); break; case BLENDEQUATION_ADVANCED_OVERLAY: SAMPLE_REGISTER_ADV_BLEND(overlay); break; case BLENDEQUATION_ADVANCED_DARKEN: SAMPLE_REGISTER_ADV_BLEND(darken); break; case BLENDEQUATION_ADVANCED_LIGHTEN: SAMPLE_REGISTER_ADV_BLEND(lighten); break; case BLENDEQUATION_ADVANCED_COLORDODGE: SAMPLE_REGISTER_ADV_BLEND(colordodge); break; case BLENDEQUATION_ADVANCED_COLORBURN: SAMPLE_REGISTER_ADV_BLEND(colorburn); break; case BLENDEQUATION_ADVANCED_HARDLIGHT: SAMPLE_REGISTER_ADV_BLEND(hardlight); break; case BLENDEQUATION_ADVANCED_SOFTLIGHT: SAMPLE_REGISTER_ADV_BLEND(softlight); break; case BLENDEQUATION_ADVANCED_DIFFERENCE: SAMPLE_REGISTER_ADV_BLEND(difference); break; case BLENDEQUATION_ADVANCED_EXCLUSION: SAMPLE_REGISTER_ADV_BLEND(exclusion); break; case BLENDEQUATION_ADVANCED_HSL_HUE: SAMPLE_REGISTER_ADV_BLEND_HSL(setLumSat(srcColor, dstColor, dstColor)); break; case BLENDEQUATION_ADVANCED_HSL_SATURATION: SAMPLE_REGISTER_ADV_BLEND_HSL(setLumSat(dstColor, srcColor, dstColor)); break; case BLENDEQUATION_ADVANCED_HSL_COLOR: SAMPLE_REGISTER_ADV_BLEND_HSL(setLum(srcColor, dstColor)); break; case BLENDEQUATION_ADVANCED_HSL_LUMINOSITY: SAMPLE_REGISTER_ADV_BLEND_HSL(setLum(dstColor, srcColor)); break; default: DE_ASSERT(false); } #undef SAMPLE_REGISTER_ADV_BLEND #undef SAMPLE_REGISTER_ADV_BLEND_HSL } void FragmentProcessor::executeColorWrite (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, bool isSRGB, const tcu::PixelBufferAccess& colorBuffer) { for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) { if (m_sampleRegister[regSampleNdx].isAlive) { int fragSampleNdx = regSampleNdx % numSamplesPerFragment; const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment]; Vec4 combinedColor; combinedColor.xyz() = m_sampleRegister[regSampleNdx].blendedRGB; combinedColor.w() = m_sampleRegister[regSampleNdx].blendedA; if (isSRGB) combinedColor = tcu::linearToSRGB(combinedColor); colorBuffer.setPixel(combinedColor, fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); } } } void FragmentProcessor::executeRGBA8ColorWrite (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const tcu::PixelBufferAccess& colorBuffer) { const int fragStride = 4; const int xStride = colorBuffer.getRowPitch(); const int yStride = colorBuffer.getSlicePitch(); deUint8* const basePtr = (deUint8*)colorBuffer.getDataPtr(); for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) { if (m_sampleRegister[regSampleNdx].isAlive) { const int fragSampleNdx = regSampleNdx % numSamplesPerFragment; const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment]; deUint8* dstPtr = basePtr + fragSampleNdx*fragStride + frag.pixelCoord.x()*xStride + frag.pixelCoord.y()*yStride; dstPtr[0] = tcu::floatToU8(m_sampleRegister[regSampleNdx].blendedRGB.x()); dstPtr[1] = tcu::floatToU8(m_sampleRegister[regSampleNdx].blendedRGB.y()); dstPtr[2] = tcu::floatToU8(m_sampleRegister[regSampleNdx].blendedRGB.z()); dstPtr[3] = tcu::floatToU8(m_sampleRegister[regSampleNdx].blendedA); } } } void FragmentProcessor::executeMaskedColorWrite (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const Vec4& colorMaskFactor, const Vec4& colorMaskNegationFactor, bool isSRGB, const tcu::PixelBufferAccess& colorBuffer) { for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) { if (m_sampleRegister[regSampleNdx].isAlive) { int fragSampleNdx = regSampleNdx % numSamplesPerFragment; const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment]; Vec4 originalColor = colorBuffer.getPixel(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); Vec4 newColor; newColor.xyz() = m_sampleRegister[regSampleNdx].blendedRGB; newColor.w() = m_sampleRegister[regSampleNdx].blendedA; if (isSRGB) newColor = tcu::linearToSRGB(newColor); newColor = colorMaskFactor*newColor + colorMaskNegationFactor*originalColor; colorBuffer.setPixel(newColor, fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); } } } void FragmentProcessor::executeSignedValueWrite (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const tcu::BVec4& colorMask, const tcu::PixelBufferAccess& colorBuffer) { for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) { if (m_sampleRegister[regSampleNdx].isAlive) { int fragSampleNdx = regSampleNdx % numSamplesPerFragment; const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment]; const IVec4 originalValue = colorBuffer.getPixelInt(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); colorBuffer.setPixel(tcu::select(m_sampleRegister[regSampleNdx].signedValue, originalValue, colorMask), fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); } } } void FragmentProcessor::executeUnsignedValueWrite (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const tcu::BVec4& colorMask, const tcu::PixelBufferAccess& colorBuffer) { for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) { if (m_sampleRegister[regSampleNdx].isAlive) { int fragSampleNdx = regSampleNdx % numSamplesPerFragment; const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment]; const UVec4 originalValue = colorBuffer.getPixelUint(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); colorBuffer.setPixel(tcu::select(m_sampleRegister[regSampleNdx].unsignedValue, originalValue, colorMask), fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); } } } void FragmentProcessor::render (const rr::MultisamplePixelBufferAccess& msColorBuffer, const rr::MultisamplePixelBufferAccess& msDepthBuffer, const rr::MultisamplePixelBufferAccess& msStencilBuffer, const Fragment* inputFragments, int numFragments, FaceType fragmentFacing, const FragmentOperationState& state) { DE_ASSERT(fragmentFacing < FACETYPE_LAST); DE_ASSERT(state.numStencilBits < 32); // code bitshifts numStencilBits, avoid undefined behavior const tcu::PixelBufferAccess& colorBuffer = msColorBuffer.raw(); const tcu::PixelBufferAccess& depthBuffer = msDepthBuffer.raw(); const tcu::PixelBufferAccess& stencilBuffer = msStencilBuffer.raw(); bool hasDepth = depthBuffer.getWidth() > 0 && depthBuffer.getHeight() > 0 && depthBuffer.getDepth() > 0; bool hasStencil = stencilBuffer.getWidth() > 0 && stencilBuffer.getHeight() > 0 && stencilBuffer.getDepth() > 0; bool doDepthTest = hasDepth && state.depthTestEnabled; bool doStencilTest = hasStencil && state.stencilTestEnabled; tcu::TextureChannelClass colorbufferClass = tcu::getTextureChannelClass(msColorBuffer.raw().getFormat().type); rr::GenericVecType fragmentDataType = (colorbufferClass == tcu::TEXTURECHANNELCLASS_SIGNED_INTEGER) ? (rr::GENERICVECTYPE_INT32) : ((colorbufferClass == tcu::TEXTURECHANNELCLASS_UNSIGNED_INTEGER) ? (rr::GENERICVECTYPE_UINT32) : (rr::GENERICVECTYPE_FLOAT)); DE_ASSERT((!hasDepth || colorBuffer.getWidth() == depthBuffer.getWidth()) && (!hasStencil || colorBuffer.getWidth() == stencilBuffer.getWidth())); DE_ASSERT((!hasDepth || colorBuffer.getHeight() == depthBuffer.getHeight()) && (!hasStencil || colorBuffer.getHeight() == stencilBuffer.getHeight())); DE_ASSERT((!hasDepth || colorBuffer.getDepth() == depthBuffer.getDepth()) && (!hasStencil || colorBuffer.getDepth() == stencilBuffer.getDepth())); int numSamplesPerFragment = colorBuffer.getWidth(); int totalNumSamples = numFragments*numSamplesPerFragment; int numSampleGroups = (totalNumSamples - 1) / SAMPLE_REGISTER_SIZE + 1; // \note totalNumSamples/SAMPLE_REGISTER_SIZE rounded up. const StencilState& stencilState = state.stencilStates[fragmentFacing]; Vec4 colorMaskFactor (state.colorMask[0] ? 1.0f : 0.0f, state.colorMask[1] ? 1.0f : 0.0f, state.colorMask[2] ? 1.0f : 0.0f, state.colorMask[3] ? 1.0f : 0.0f); Vec4 colorMaskNegationFactor (state.colorMask[0] ? 0.0f : 1.0f, state.colorMask[1] ? 0.0f : 1.0f, state.colorMask[2] ? 0.0f : 1.0f, state.colorMask[3] ? 0.0f : 1.0f); bool sRGBTarget = state.sRGBEnabled && tcu::isSRGB(colorBuffer.getFormat()); DE_ASSERT(SAMPLE_REGISTER_SIZE % numSamplesPerFragment == 0); // Divide the fragments' samples into groups of size SAMPLE_REGISTER_SIZE, and perform // the per-sample operations for one group at a time. for (int sampleGroupNdx = 0; sampleGroupNdx < numSampleGroups; sampleGroupNdx++) { // The index of the fragment of the sample at the beginning of m_sampleRegisters. int groupFirstFragNdx = (sampleGroupNdx*SAMPLE_REGISTER_SIZE) / numSamplesPerFragment; // Initialize sample data in the sample register. for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) { int fragNdx = groupFirstFragNdx + regSampleNdx/numSamplesPerFragment; int fragSampleNdx = regSampleNdx % numSamplesPerFragment; if (fragNdx < numFragments) { m_sampleRegister[regSampleNdx].isAlive = (inputFragments[fragNdx].coverage & (1u << fragSampleNdx)) != 0; m_sampleRegister[regSampleNdx].depthPassed = true; // \note This will stay true if depth test is disabled. } else m_sampleRegister[regSampleNdx].isAlive = false; } // Scissor test. if (state.scissorTestEnabled) executeScissorTest(groupFirstFragNdx, numSamplesPerFragment, inputFragments, state.scissorRectangle); // Stencil test. if (doStencilTest) { executeStencilCompare(groupFirstFragNdx, numSamplesPerFragment, inputFragments, stencilState, state.numStencilBits, stencilBuffer); executeStencilSFail(groupFirstFragNdx, numSamplesPerFragment, inputFragments, stencilState, state.numStencilBits, stencilBuffer); } // Depth test. // \note Current value of isAlive is needed for dpPass and dpFail, so it's only updated after them and not right after depth test. if (doDepthTest) { executeDepthCompare(groupFirstFragNdx, numSamplesPerFragment, inputFragments, state.depthFunc, depthBuffer); if (state.depthMask) executeDepthWrite(groupFirstFragNdx, numSamplesPerFragment, inputFragments, depthBuffer); } // Do dpFail and dpPass stencil writes. if (doStencilTest) executeStencilDpFailAndPass(groupFirstFragNdx, numSamplesPerFragment, inputFragments, stencilState, state.numStencilBits, stencilBuffer); // Kill the samples that failed depth test. if (doDepthTest) { for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) m_sampleRegister[regSampleNdx].isAlive = m_sampleRegister[regSampleNdx].isAlive && m_sampleRegister[regSampleNdx].depthPassed; } // Paint fragments to target switch (fragmentDataType) { case rr::GENERICVECTYPE_FLOAT: // Blend calculation - only if using blend. if (state.blendMode == BLENDMODE_STANDARD) { // Put dst color to register, doing srgb-to-linear conversion if needed. for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) { if (m_sampleRegister[regSampleNdx].isAlive) { int fragSampleNdx = regSampleNdx % numSamplesPerFragment; const Fragment& frag = inputFragments[groupFirstFragNdx + regSampleNdx/numSamplesPerFragment]; Vec4 dstColor = colorBuffer.getPixel(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); m_sampleRegister[regSampleNdx].clampedBlendSrcColor = clamp(frag.value.get(), Vec4(0.0f), Vec4(1.0f)); m_sampleRegister[regSampleNdx].clampedBlendSrc1Color = clamp(frag.value1.get(), Vec4(0.0f), Vec4(1.0f)); m_sampleRegister[regSampleNdx].clampedBlendDstColor = clamp(sRGBTarget ? tcu::sRGBToLinear(dstColor) : dstColor, Vec4(0.0f), Vec4(1.0f)); } } // Calculate blend factors to register. executeBlendFactorComputeRGB(state.blendColor, state.blendRGBState); executeBlendFactorComputeA(state.blendColor, state.blendAState); // Compute blended color. executeBlend(state.blendRGBState, state.blendAState); } else if (state.blendMode == BLENDMODE_ADVANCED) { // Unpremultiply colors for blending, and do sRGB->linear if necessary // \todo [2014-03-17 pyry] Re-consider clampedBlend*Color var names for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) { if (m_sampleRegister[regSampleNdx].isAlive) { int fragSampleNdx = regSampleNdx % numSamplesPerFragment; const Fragment& frag = inputFragments[groupFirstFragNdx + regSampleNdx/numSamplesPerFragment]; const Vec4 srcColor = frag.value.get(); const Vec4 dstColor = colorBuffer.getPixel(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); m_sampleRegister[regSampleNdx].clampedBlendSrcColor = unpremultiply(clamp(srcColor, Vec4(0.0f), Vec4(1.0f))); m_sampleRegister[regSampleNdx].clampedBlendDstColor = unpremultiply(clamp(sRGBTarget ? tcu::sRGBToLinear(dstColor) : dstColor, Vec4(0.0f), Vec4(1.0f))); } } executeAdvancedBlend(state.blendEquationAdvaced); } else { // Not using blend - just put values to register as-is. DE_ASSERT(state.blendMode == BLENDMODE_NONE); for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) { if (m_sampleRegister[regSampleNdx].isAlive) { const Fragment& frag = inputFragments[groupFirstFragNdx + regSampleNdx/numSamplesPerFragment]; m_sampleRegister[regSampleNdx].blendedRGB = frag.value.get().xyz(); m_sampleRegister[regSampleNdx].blendedA = frag.value.get().w(); } } } // Finally, write the colors to the color buffer. if (state.colorMask[0] && state.colorMask[1] && state.colorMask[2] && state.colorMask[3]) { if (colorBuffer.getFormat() == tcu::TextureFormat(tcu::TextureFormat::RGBA, tcu::TextureFormat::UNORM_INT8)) executeRGBA8ColorWrite(groupFirstFragNdx, numSamplesPerFragment, inputFragments, colorBuffer); else executeColorWrite(groupFirstFragNdx, numSamplesPerFragment, inputFragments, sRGBTarget, colorBuffer); } else if (state.colorMask[0] || state.colorMask[1] || state.colorMask[2] || state.colorMask[3]) executeMaskedColorWrite(groupFirstFragNdx, numSamplesPerFragment, inputFragments, colorMaskFactor, colorMaskNegationFactor, sRGBTarget, colorBuffer); break; case rr::GENERICVECTYPE_INT32: // Write fragments for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) { if (m_sampleRegister[regSampleNdx].isAlive) { const Fragment& frag = inputFragments[groupFirstFragNdx + regSampleNdx/numSamplesPerFragment]; m_sampleRegister[regSampleNdx].signedValue = frag.value.get(); } } if (state.colorMask[0] || state.colorMask[1] || state.colorMask[2] || state.colorMask[3]) executeSignedValueWrite(groupFirstFragNdx, numSamplesPerFragment, inputFragments, state.colorMask, colorBuffer); break; case rr::GENERICVECTYPE_UINT32: // Write fragments for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) { if (m_sampleRegister[regSampleNdx].isAlive) { const Fragment& frag = inputFragments[groupFirstFragNdx + regSampleNdx/numSamplesPerFragment]; m_sampleRegister[regSampleNdx].unsignedValue = frag.value.get(); } } if (state.colorMask[0] || state.colorMask[1] || state.colorMask[2] || state.colorMask[3]) executeUnsignedValueWrite(groupFirstFragNdx, numSamplesPerFragment, inputFragments, state.colorMask, colorBuffer); break; default: DE_ASSERT(DE_FALSE); } } } } // rr