1 /*-------------------------------------------------------------------------
2 * drawElements Quality Program Reference Renderer
3 * -----------------------------------------------
5 * Copyright 2014 The Android Open Source Project
7 * Licensed under the Apache License, Version 2.0 (the "License");
8 * you may not use this file except in compliance with the License.
9 * You may obtain a copy of the License at
11 * http://www.apache.org/licenses/LICENSE-2.0
13 * Unless required by applicable law or agreed to in writing, software
14 * distributed under the License is distributed on an "AS IS" BASIS,
15 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
16 * See the License for the specific language governing permissions and
17 * limitations under the License.
21 * \brief Reference implementation for per-fragment operations.
22 *//*--------------------------------------------------------------------*/
24 #include "rrFragmentOperations.hpp"
25 #include "tcuVectorUtil.hpp"
26 #include "tcuTextureUtil.hpp"
44 // Return oldValue with the bits indicated by mask replaced by corresponding bits of newValue.
45 static inline int maskedBitReplace (int oldValue, int newValue, deUint32 mask)
47 return (oldValue & ~mask) | (newValue & mask);
50 static inline bool isInsideRect (const IVec2& point, const WindowRectangle& rect)
52 return de::inBounds(point.x(), rect.left, rect.left + rect.width) &&
53 de::inBounds(point.y(), rect.bottom, rect.bottom + rect.height);
56 static inline Vec4 unpremultiply (const Vec4& v)
59 return Vec4(v.x()/v.w(), v.y()/v.w(), v.z()/v.w(), v.w());
62 DE_ASSERT(v.x() == 0.0f && v.y() == 0.0f && v.z() == 0.0f);
63 return Vec4(0.0f, 0.0f, 0.0f, 0.0f);
67 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); }
68 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); }
69 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<int>()); }
70 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); }
71 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); }
73 FragmentProcessor::FragmentProcessor (void)
78 void FragmentProcessor::executeScissorTest (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const WindowRectangle& scissorRect)
80 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
82 if (m_sampleRegister[regSampleNdx].isAlive)
84 int fragNdx = fragNdxOffset + regSampleNdx/numSamplesPerFragment;
86 if (!isInsideRect(inputFragments[fragNdx].pixelCoord, scissorRect))
87 m_sampleRegister[regSampleNdx].isAlive = false;
92 void FragmentProcessor::executeStencilCompare (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const StencilState& stencilState, int numStencilBits, const tcu::ConstPixelBufferAccess& stencilBuffer)
94 #define SAMPLE_REGISTER_STENCIL_COMPARE(COMPARE_EXPRESSION) \
95 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \
97 if (m_sampleRegister[regSampleNdx].isAlive) \
99 int fragSampleNdx = regSampleNdx % numSamplesPerFragment; \
100 const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment]; \
101 int stencilBufferValue = stencilBuffer.getPixStencil(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); \
102 int maskedRef = stencilState.compMask & clampedStencilRef; \
103 int maskedBuf = stencilState.compMask & stencilBufferValue; \
104 DE_UNREF(maskedRef); \
105 DE_UNREF(maskedBuf); \
107 m_sampleRegister[regSampleNdx].stencilPassed = (COMPARE_EXPRESSION); \
111 int clampedStencilRef = de::clamp(stencilState.ref, 0, (1<<numStencilBits)-1);
113 switch (stencilState.func)
115 case TESTFUNC_NEVER: SAMPLE_REGISTER_STENCIL_COMPARE(false) break;
116 case TESTFUNC_ALWAYS: SAMPLE_REGISTER_STENCIL_COMPARE(true) break;
117 case TESTFUNC_LESS: SAMPLE_REGISTER_STENCIL_COMPARE(maskedRef < maskedBuf) break;
118 case TESTFUNC_LEQUAL: SAMPLE_REGISTER_STENCIL_COMPARE(maskedRef <= maskedBuf) break;
119 case TESTFUNC_GREATER: SAMPLE_REGISTER_STENCIL_COMPARE(maskedRef > maskedBuf) break;
120 case TESTFUNC_GEQUAL: SAMPLE_REGISTER_STENCIL_COMPARE(maskedRef >= maskedBuf) break;
121 case TESTFUNC_EQUAL: SAMPLE_REGISTER_STENCIL_COMPARE(maskedRef == maskedBuf) break;
122 case TESTFUNC_NOTEQUAL: SAMPLE_REGISTER_STENCIL_COMPARE(maskedRef != maskedBuf) break;
127 #undef SAMPLE_REGISTER_STENCIL_COMPARE
130 void FragmentProcessor::executeStencilSFail (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const StencilState& stencilState, int numStencilBits, const tcu::PixelBufferAccess& stencilBuffer)
132 #define SAMPLE_REGISTER_SFAIL(SFAIL_EXPRESSION) \
133 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \
135 if (m_sampleRegister[regSampleNdx].isAlive && !m_sampleRegister[regSampleNdx].stencilPassed) \
137 int fragSampleNdx = regSampleNdx % numSamplesPerFragment; \
138 const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment]; \
139 int stencilBufferValue = stencilBuffer.getPixStencil(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); \
141 stencilBuffer.setPixStencil(maskedBitReplace(stencilBufferValue, (SFAIL_EXPRESSION), stencilState.writeMask), fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); \
142 m_sampleRegister[regSampleNdx].isAlive = false; \
146 int clampedStencilRef = de::clamp(stencilState.ref, 0, (1<<numStencilBits)-1);
148 switch (stencilState.sFail)
150 case STENCILOP_KEEP: SAMPLE_REGISTER_SFAIL(stencilBufferValue) break;
151 case STENCILOP_ZERO: SAMPLE_REGISTER_SFAIL(0) break;
152 case STENCILOP_REPLACE: SAMPLE_REGISTER_SFAIL(clampedStencilRef) break;
153 case STENCILOP_INCR: SAMPLE_REGISTER_SFAIL(de::clamp(stencilBufferValue+1, 0, (1<<numStencilBits) - 1)) break;
154 case STENCILOP_DECR: SAMPLE_REGISTER_SFAIL(de::clamp(stencilBufferValue-1, 0, (1<<numStencilBits) - 1)) break;
155 case STENCILOP_INCR_WRAP: SAMPLE_REGISTER_SFAIL((stencilBufferValue + 1) & ((1<<numStencilBits) - 1)) break;
156 case STENCILOP_DECR_WRAP: SAMPLE_REGISTER_SFAIL((stencilBufferValue - 1) & ((1<<numStencilBits) - 1)) break;
157 case STENCILOP_INVERT: SAMPLE_REGISTER_SFAIL((~stencilBufferValue) & ((1<<numStencilBits) - 1)) break;
162 #undef SAMPLE_REGISTER_SFAIL
166 void FragmentProcessor::executeDepthBoundsTest (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const float minDepthBound, const float maxDepthBound, const tcu::ConstPixelBufferAccess& depthBuffer)
168 if (depthBuffer.getFormat().type == tcu::TextureFormat::FLOAT || depthBuffer.getFormat().type == tcu::TextureFormat::FLOAT_UNSIGNED_INT_24_8_REV)
170 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; ++regSampleNdx)
172 if (m_sampleRegister[regSampleNdx].isAlive)
174 const int fragSampleNdx = regSampleNdx % numSamplesPerFragment;
175 const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment];
176 const float depthBufferValue = depthBuffer.getPixDepth(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y());
178 if (!de::inRange(depthBufferValue, minDepthBound, maxDepthBound))
179 m_sampleRegister[regSampleNdx].isAlive = false;
185 /* Convert float bounds to target buffer format for comparison */
187 deUint32 minDepthBoundUint, maxDepthBoundUint;
190 DE_ASSERT(sizeof(buffer) >= (size_t)depthBuffer.getFormat().getPixelSize());
192 tcu::PixelBufferAccess access(depthBuffer.getFormat(), 1, 1, 1, &buffer);
193 access.setPixDepth(minDepthBound, 0, 0, 0);
194 minDepthBoundUint = access.getPixelUint(0, 0, 0).x();
199 tcu::PixelBufferAccess access(depthBuffer.getFormat(), 1, 1, 1, &buffer);
200 access.setPixDepth(maxDepthBound, 0, 0, 0);
201 maxDepthBoundUint = access.getPixelUint(0, 0, 0).x();
204 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; ++regSampleNdx)
206 if (m_sampleRegister[regSampleNdx].isAlive)
208 const int fragSampleNdx = regSampleNdx % numSamplesPerFragment;
209 const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx / numSamplesPerFragment];
210 const deUint32 depthBufferValue = depthBuffer.getPixelUint(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()).x();
212 if (!de::inRange(depthBufferValue, minDepthBoundUint, maxDepthBoundUint))
213 m_sampleRegister[regSampleNdx].isAlive = false;
219 void FragmentProcessor::executeDepthCompare (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, TestFunc depthFunc, const tcu::ConstPixelBufferAccess& depthBuffer)
221 #define SAMPLE_REGISTER_DEPTH_COMPARE_F(COMPARE_EXPRESSION) \
222 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \
224 if (m_sampleRegister[regSampleNdx].isAlive) \
226 int fragSampleNdx = regSampleNdx % numSamplesPerFragment; \
227 const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment]; \
228 float depthBufferValue = depthBuffer.getPixDepth(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); \
229 float sampleDepthFloat = frag.sampleDepths[fragSampleNdx]; \
230 float sampleDepth = de::clamp(sampleDepthFloat, 0.0f, 1.0f); \
232 m_sampleRegister[regSampleNdx].depthPassed = (COMPARE_EXPRESSION); \
234 DE_UNREF(depthBufferValue); \
235 DE_UNREF(sampleDepth); \
239 #define SAMPLE_REGISTER_DEPTH_COMPARE_UI(COMPARE_EXPRESSION) \
240 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \
242 if (m_sampleRegister[regSampleNdx].isAlive) \
244 int fragSampleNdx = regSampleNdx % numSamplesPerFragment; \
245 const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment]; \
246 deUint32 depthBufferValue = depthBuffer.getPixelUint(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()).x(); \
247 float sampleDepthFloat = frag.sampleDepths[fragSampleNdx]; \
249 /* Convert input float to target buffer format for comparison */ \
251 deUint32 buffer[2]; \
253 DE_ASSERT(sizeof(buffer) >= (size_t)depthBuffer.getFormat().getPixelSize()); \
255 tcu::PixelBufferAccess access(depthBuffer.getFormat(), 1, 1, 1, &buffer); \
256 access.setPixDepth(sampleDepthFloat, 0, 0, 0); \
257 deUint32 sampleDepth = access.getPixelUint(0, 0, 0).x(); \
259 m_sampleRegister[regSampleNdx].depthPassed = (COMPARE_EXPRESSION); \
261 DE_UNREF(depthBufferValue); \
262 DE_UNREF(sampleDepth); \
266 if (depthBuffer.getFormat().type == tcu::TextureFormat::FLOAT || depthBuffer.getFormat().type == tcu::TextureFormat::FLOAT_UNSIGNED_INT_24_8_REV)
271 case TESTFUNC_NEVER: SAMPLE_REGISTER_DEPTH_COMPARE_F(false) break;
272 case TESTFUNC_ALWAYS: SAMPLE_REGISTER_DEPTH_COMPARE_F(true) break;
273 case TESTFUNC_LESS: SAMPLE_REGISTER_DEPTH_COMPARE_F(sampleDepth < depthBufferValue) break;
274 case TESTFUNC_LEQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_F(sampleDepth <= depthBufferValue) break;
275 case TESTFUNC_GREATER: SAMPLE_REGISTER_DEPTH_COMPARE_F(sampleDepth > depthBufferValue) break;
276 case TESTFUNC_GEQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_F(sampleDepth >= depthBufferValue) break;
277 case TESTFUNC_EQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_F(sampleDepth == depthBufferValue) break;
278 case TESTFUNC_NOTEQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_F(sampleDepth != depthBufferValue) break;
288 case TESTFUNC_NEVER: SAMPLE_REGISTER_DEPTH_COMPARE_UI(false) break;
289 case TESTFUNC_ALWAYS: SAMPLE_REGISTER_DEPTH_COMPARE_UI(true) break;
290 case TESTFUNC_LESS: SAMPLE_REGISTER_DEPTH_COMPARE_UI(sampleDepth < depthBufferValue) break;
291 case TESTFUNC_LEQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_UI(sampleDepth <= depthBufferValue) break;
292 case TESTFUNC_GREATER: SAMPLE_REGISTER_DEPTH_COMPARE_UI(sampleDepth > depthBufferValue) break;
293 case TESTFUNC_GEQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_UI(sampleDepth >= depthBufferValue) break;
294 case TESTFUNC_EQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_UI(sampleDepth == depthBufferValue) break;
295 case TESTFUNC_NOTEQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_UI(sampleDepth != depthBufferValue) break;
301 #undef SAMPLE_REGISTER_DEPTH_COMPARE_F
302 #undef SAMPLE_REGISTER_DEPTH_COMPARE_UI
305 void FragmentProcessor::executeDepthWrite (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const tcu::PixelBufferAccess& depthBuffer)
307 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
309 if (m_sampleRegister[regSampleNdx].isAlive && m_sampleRegister[regSampleNdx].depthPassed)
311 int fragSampleNdx = regSampleNdx % numSamplesPerFragment;
312 const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment];
313 const float clampedDepth = de::clamp(frag.sampleDepths[fragSampleNdx], 0.0f, 1.0f);
315 depthBuffer.setPixDepth(clampedDepth, fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y());
320 void FragmentProcessor::executeStencilDpFailAndPass (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const StencilState& stencilState, int numStencilBits, const tcu::PixelBufferAccess& stencilBuffer)
322 #define SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, EXPRESSION) \
323 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \
325 if (m_sampleRegister[regSampleNdx].isAlive && (CONDITION)) \
327 int fragSampleNdx = regSampleNdx % numSamplesPerFragment; \
328 const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment]; \
329 int stencilBufferValue = stencilBuffer.getPixStencil(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); \
331 stencilBuffer.setPixStencil(maskedBitReplace(stencilBufferValue, (EXPRESSION), stencilState.writeMask), fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); \
335 #define SWITCH_DPFAIL_OR_DPPASS(OP_NAME, CONDITION) \
336 switch (stencilState.OP_NAME) \
338 case STENCILOP_KEEP: SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, stencilBufferValue) break; \
339 case STENCILOP_ZERO: SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, 0) break; \
340 case STENCILOP_REPLACE: SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, clampedStencilRef) break; \
341 case STENCILOP_INCR: SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, de::clamp(stencilBufferValue+1, 0, (1<<numStencilBits) - 1)) break; \
342 case STENCILOP_DECR: SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, de::clamp(stencilBufferValue-1, 0, (1<<numStencilBits) - 1)) break; \
343 case STENCILOP_INCR_WRAP: SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, (stencilBufferValue + 1) & ((1<<numStencilBits) - 1)) break; \
344 case STENCILOP_DECR_WRAP: SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, (stencilBufferValue - 1) & ((1<<numStencilBits) - 1)) break; \
345 case STENCILOP_INVERT: SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, (~stencilBufferValue) & ((1<<numStencilBits) - 1)) break; \
350 int clampedStencilRef = de::clamp(stencilState.ref, 0, (1<<numStencilBits)-1);
352 SWITCH_DPFAIL_OR_DPPASS(dpFail, !m_sampleRegister[regSampleNdx].depthPassed)
353 SWITCH_DPFAIL_OR_DPPASS(dpPass, m_sampleRegister[regSampleNdx].depthPassed)
355 #undef SWITCH_DPFAIL_OR_DPPASS
356 #undef SAMPLE_REGISTER_DPFAIL_OR_DPPASS
359 void FragmentProcessor::executeBlendFactorComputeRGB (const Vec4& blendColor, const BlendState& blendRGBState)
361 #define SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, FACTOR_EXPRESSION) \
362 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \
364 if (m_sampleRegister[regSampleNdx].isAlive) \
366 const Vec4& src = m_sampleRegister[regSampleNdx].clampedBlendSrcColor; \
367 const Vec4& src1 = m_sampleRegister[regSampleNdx].clampedBlendSrc1Color; \
368 const Vec4& dst = m_sampleRegister[regSampleNdx].clampedBlendDstColor; \
373 m_sampleRegister[regSampleNdx].FACTOR_NAME = (FACTOR_EXPRESSION); \
377 #define SWITCH_SRC_OR_DST_FACTOR_RGB(FUNC_NAME, FACTOR_NAME) \
378 switch (blendRGBState.FUNC_NAME) \
380 case BLENDFUNC_ZERO: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(0.0f)) break; \
381 case BLENDFUNC_ONE: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(1.0f)) break; \
382 case BLENDFUNC_SRC_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, src.swizzle(0,1,2)) break; \
383 case BLENDFUNC_ONE_MINUS_SRC_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(1.0f) - src.swizzle(0,1,2)) break; \
384 case BLENDFUNC_DST_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, dst.swizzle(0,1,2)) break; \
385 case BLENDFUNC_ONE_MINUS_DST_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(1.0f) - dst.swizzle(0,1,2)) break; \
386 case BLENDFUNC_SRC_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(src.w())) break; \
387 case BLENDFUNC_ONE_MINUS_SRC_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(1.0f - src.w())) break; \
388 case BLENDFUNC_DST_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(dst.w())) break; \
389 case BLENDFUNC_ONE_MINUS_DST_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(1.0f - dst.w())) break; \
390 case BLENDFUNC_CONSTANT_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, blendColor.swizzle(0,1,2)) break; \
391 case BLENDFUNC_ONE_MINUS_CONSTANT_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(1.0f) - blendColor.swizzle(0,1,2)) break; \
392 case BLENDFUNC_CONSTANT_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(blendColor.w())) break; \
393 case BLENDFUNC_ONE_MINUS_CONSTANT_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(1.0f - blendColor.w())) break; \
394 case BLENDFUNC_SRC_ALPHA_SATURATE: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(de::min(src.w(), 1.0f - dst.w()))) break; \
395 case BLENDFUNC_SRC1_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, src1.swizzle(0,1,2)) break; \
396 case BLENDFUNC_ONE_MINUS_SRC1_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(1.0f) - src1.swizzle(0,1,2)) break; \
397 case BLENDFUNC_SRC1_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(src1.w())) break; \
398 case BLENDFUNC_ONE_MINUS_SRC1_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(1.0f - src1.w())) break; \
403 SWITCH_SRC_OR_DST_FACTOR_RGB(srcFunc, blendSrcFactorRGB)
404 SWITCH_SRC_OR_DST_FACTOR_RGB(dstFunc, blendDstFactorRGB)
406 #undef SWITCH_SRC_OR_DST_FACTOR_RGB
407 #undef SAMPLE_REGISTER_BLEND_FACTOR
410 void FragmentProcessor::executeBlendFactorComputeA (const Vec4& blendColor, const BlendState& blendAState)
412 #define SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, FACTOR_EXPRESSION) \
413 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \
415 if (m_sampleRegister[regSampleNdx].isAlive) \
417 const Vec4& src = m_sampleRegister[regSampleNdx].clampedBlendSrcColor; \
418 const Vec4& src1 = m_sampleRegister[regSampleNdx].clampedBlendSrc1Color; \
419 const Vec4& dst = m_sampleRegister[regSampleNdx].clampedBlendDstColor; \
424 m_sampleRegister[regSampleNdx].FACTOR_NAME = (FACTOR_EXPRESSION); \
428 #define SWITCH_SRC_OR_DST_FACTOR_A(FUNC_NAME, FACTOR_NAME) \
429 switch (blendAState.FUNC_NAME) \
431 case BLENDFUNC_ZERO: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, 0.0f) break; \
432 case BLENDFUNC_ONE: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, 1.0f) break; \
433 case BLENDFUNC_SRC_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, src.w()) break; \
434 case BLENDFUNC_ONE_MINUS_SRC_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, 1.0f - src.w()) break; \
435 case BLENDFUNC_DST_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, dst.w()) break; \
436 case BLENDFUNC_ONE_MINUS_DST_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, 1.0f - dst.w()) break; \
437 case BLENDFUNC_SRC_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, src.w()) break; \
438 case BLENDFUNC_ONE_MINUS_SRC_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, 1.0f - src.w()) break; \
439 case BLENDFUNC_DST_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, dst.w()) break; \
440 case BLENDFUNC_ONE_MINUS_DST_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, 1.0f - dst.w()) break; \
441 case BLENDFUNC_CONSTANT_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, blendColor.w()) break; \
442 case BLENDFUNC_ONE_MINUS_CONSTANT_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, 1.0f - blendColor.w()) break; \
443 case BLENDFUNC_CONSTANT_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, blendColor.w()) break; \
444 case BLENDFUNC_ONE_MINUS_CONSTANT_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, 1.0f - blendColor.w()) break; \
445 case BLENDFUNC_SRC_ALPHA_SATURATE: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, 1.0f) break; \
446 case BLENDFUNC_SRC1_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, src1.w()) break; \
447 case BLENDFUNC_ONE_MINUS_SRC1_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, 1.0f - src1.w()) break; \
448 case BLENDFUNC_SRC1_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, src1.w()) break; \
449 case BLENDFUNC_ONE_MINUS_SRC1_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, 1.0f - src1.w()) break; \
454 SWITCH_SRC_OR_DST_FACTOR_A(srcFunc, blendSrcFactorA)
455 SWITCH_SRC_OR_DST_FACTOR_A(dstFunc, blendDstFactorA)
457 #undef SWITCH_SRC_OR_DST_FACTOR_A
458 #undef SAMPLE_REGISTER_BLEND_FACTOR
461 void FragmentProcessor::executeBlend (const BlendState& blendRGBState, const BlendState& blendAState)
463 #define SAMPLE_REGISTER_BLENDED_COLOR(COLOR_NAME, COLOR_EXPRESSION) \
464 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \
466 if (m_sampleRegister[regSampleNdx].isAlive) \
468 SampleData& sample = m_sampleRegister[regSampleNdx]; \
469 const Vec4& srcColor = sample.clampedBlendSrcColor; \
470 const Vec4& dstColor = sample.clampedBlendDstColor; \
472 sample.COLOR_NAME = (COLOR_EXPRESSION); \
476 switch (blendRGBState.equation)
478 case BLENDEQUATION_ADD: SAMPLE_REGISTER_BLENDED_COLOR(blendedRGB, srcColor.swizzle(0,1,2)*sample.blendSrcFactorRGB + dstColor.swizzle(0,1,2)*sample.blendDstFactorRGB) break;
479 case BLENDEQUATION_SUBTRACT: SAMPLE_REGISTER_BLENDED_COLOR(blendedRGB, srcColor.swizzle(0,1,2)*sample.blendSrcFactorRGB - dstColor.swizzle(0,1,2)*sample.blendDstFactorRGB) break;
480 case BLENDEQUATION_REVERSE_SUBTRACT: SAMPLE_REGISTER_BLENDED_COLOR(blendedRGB, dstColor.swizzle(0,1,2)*sample.blendDstFactorRGB - srcColor.swizzle(0,1,2)*sample.blendSrcFactorRGB) break;
481 case BLENDEQUATION_MIN: SAMPLE_REGISTER_BLENDED_COLOR(blendedRGB, min(srcColor.swizzle(0,1,2), dstColor.swizzle(0,1,2))) break;
482 case BLENDEQUATION_MAX: SAMPLE_REGISTER_BLENDED_COLOR(blendedRGB, max(srcColor.swizzle(0,1,2), dstColor.swizzle(0,1,2))) break;
487 switch (blendAState.equation)
489 case BLENDEQUATION_ADD: SAMPLE_REGISTER_BLENDED_COLOR(blendedA, srcColor.w()*sample.blendSrcFactorA + dstColor.w()*sample.blendDstFactorA) break;
490 case BLENDEQUATION_SUBTRACT: SAMPLE_REGISTER_BLENDED_COLOR(blendedA, srcColor.w()*sample.blendSrcFactorA - dstColor.w()*sample.blendDstFactorA) break;
491 case BLENDEQUATION_REVERSE_SUBTRACT: SAMPLE_REGISTER_BLENDED_COLOR(blendedA, dstColor.w()*sample.blendDstFactorA - srcColor.w()*sample.blendSrcFactorA) break;
492 case BLENDEQUATION_MIN: SAMPLE_REGISTER_BLENDED_COLOR(blendedA, min(srcColor.w(), dstColor.w())) break;
493 case BLENDEQUATION_MAX: SAMPLE_REGISTER_BLENDED_COLOR(blendedA, max(srcColor.w(), dstColor.w())) break;
497 #undef SAMPLE_REGISTER_BLENDED_COLOR
503 inline float multiply (float src, float dst) { return src*dst; }
504 inline float screen (float src, float dst) { return src + dst - src*dst; }
505 inline float darken (float src, float dst) { return de::min(src, dst); }
506 inline float lighten (float src, float dst) { return de::max(src, dst); }
507 inline float difference (float src, float dst) { return de::abs(dst-src); }
508 inline float exclusion (float src, float dst) { return src + dst - 2.0f*src*dst; }
510 inline float overlay (float src, float dst)
515 return 1.0f - 2.0f*(1.0f-src)*(1.0f-dst);
518 inline float colordodge (float src, float dst)
523 return de::min(1.0f, dst/(1.0f-src));
528 inline float colorburn (float src, float dst)
533 return 1.0f - de::min(1.0f, (1.0f-dst)/src);
538 inline float hardlight (float src, float dst)
543 return 1.0f - 2.0f*(1.0f-src)*(1.0f-dst);
546 inline float softlight (float src, float dst)
549 return dst - (1.0f - 2.0f*src)*dst*(1.0f-dst);
550 else if (dst <= 0.25f)
551 return dst + (2.0f*src - 1.0f)*dst*((16.0f*dst - 12.0f)*dst + 3.0f);
553 return dst + (2.0f*src - 1.0f)*(deFloatSqrt(dst)-dst);
556 inline float minComp (const Vec3& v)
558 return de::min(de::min(v.x(), v.y()), v.z());
561 inline float maxComp (const Vec3& v)
563 return de::max(de::max(v.x(), v.y()), v.z());
566 inline float luminosity (const Vec3& rgb)
568 return dot(rgb, Vec3(0.3f, 0.59f, 0.11f));
571 inline float saturation (const Vec3& rgb)
573 return maxComp(rgb) - minComp(rgb);
576 Vec3 setLum (const Vec3& cbase, const Vec3& clum)
578 const float lbase = luminosity(cbase);
579 const float llum = luminosity(clum);
580 const float ldiff = llum - lbase;
581 const Vec3 color = cbase + Vec3(ldiff);
582 const float minC = minComp(color);
583 const float maxC = maxComp(color);
586 return llum + ((color-llum)*llum / (llum != minC ? (llum-minC) : 1.0f));
587 else if (maxC > 1.0f)
588 return llum + ((color-llum)*(1.0f-llum) / (llum != maxC ? (maxC-llum) : 1.0f));
593 Vec3 setLumSat (const Vec3& cbase, const Vec3& csat, const Vec3& clum)
595 const float minbase = minComp(cbase);
596 const float sbase = saturation(cbase);
597 const float ssat = saturation(csat);
598 Vec3 color = Vec3(0.0f);
601 color = (cbase - minbase) * ssat / sbase;
605 return setLum(color, clum);
610 void FragmentProcessor::executeAdvancedBlend (BlendEquationAdvanced equation)
612 using namespace advblend;
614 #define SAMPLE_REGISTER_ADV_BLEND(FUNCTION_NAME) \
615 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \
617 if (m_sampleRegister[regSampleNdx].isAlive) \
619 SampleData& sample = m_sampleRegister[regSampleNdx]; \
620 const Vec4& srcColor = sample.clampedBlendSrcColor; \
621 const Vec4& dstColor = sample.clampedBlendDstColor; \
622 const Vec3& bias = sample.blendSrcFactorRGB; \
623 const float p0 = sample.blendSrcFactorA; \
624 const float r = FUNCTION_NAME(srcColor[0], dstColor[0])*p0 + bias[0]; \
625 const float g = FUNCTION_NAME(srcColor[1], dstColor[1])*p0 + bias[1]; \
626 const float b = FUNCTION_NAME(srcColor[2], dstColor[2])*p0 + bias[2]; \
628 sample.blendedRGB = Vec3(r, g, b); \
632 #define SAMPLE_REGISTER_ADV_BLEND_HSL(COLOR_EXPRESSION) \
633 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \
635 if (m_sampleRegister[regSampleNdx].isAlive) \
637 SampleData& sample = m_sampleRegister[regSampleNdx]; \
638 const Vec3 srcColor = sample.clampedBlendSrcColor.swizzle(0,1,2); \
639 const Vec3 dstColor = sample.clampedBlendDstColor.swizzle(0,1,2); \
640 const Vec3& bias = sample.blendSrcFactorRGB; \
641 const float p0 = sample.blendSrcFactorA; \
643 sample.blendedRGB = (COLOR_EXPRESSION)*p0 + bias; \
647 // Pre-compute factors & compute alpha \todo [2014-03-18 pyry] Re-using variable names.
648 // \note clampedBlend*Color contains clamped & unpremultiplied colors
649 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
651 if (m_sampleRegister[regSampleNdx].isAlive)
653 SampleData& sample = m_sampleRegister[regSampleNdx];
654 const Vec4& srcColor = sample.clampedBlendSrcColor;
655 const Vec4& dstColor = sample.clampedBlendDstColor;
656 const float srcA = srcColor.w();
657 const float dstA = dstColor.w();
658 const float p0 = srcA*dstA;
659 const float p1 = srcA*(1.0f-dstA);
660 const float p2 = dstA*(1.0f-srcA);
661 const Vec3 bias (srcColor[0]*p1 + dstColor[0]*p2,
662 srcColor[1]*p1 + dstColor[1]*p2,
663 srcColor[2]*p1 + dstColor[2]*p2);
665 sample.blendSrcFactorRGB = bias;
666 sample.blendSrcFactorA = p0;
667 sample.blendedA = p0 + p1 + p2;
673 case BLENDEQUATION_ADVANCED_MULTIPLY: SAMPLE_REGISTER_ADV_BLEND(multiply); break;
674 case BLENDEQUATION_ADVANCED_SCREEN: SAMPLE_REGISTER_ADV_BLEND(screen); break;
675 case BLENDEQUATION_ADVANCED_OVERLAY: SAMPLE_REGISTER_ADV_BLEND(overlay); break;
676 case BLENDEQUATION_ADVANCED_DARKEN: SAMPLE_REGISTER_ADV_BLEND(darken); break;
677 case BLENDEQUATION_ADVANCED_LIGHTEN: SAMPLE_REGISTER_ADV_BLEND(lighten); break;
678 case BLENDEQUATION_ADVANCED_COLORDODGE: SAMPLE_REGISTER_ADV_BLEND(colordodge); break;
679 case BLENDEQUATION_ADVANCED_COLORBURN: SAMPLE_REGISTER_ADV_BLEND(colorburn); break;
680 case BLENDEQUATION_ADVANCED_HARDLIGHT: SAMPLE_REGISTER_ADV_BLEND(hardlight); break;
681 case BLENDEQUATION_ADVANCED_SOFTLIGHT: SAMPLE_REGISTER_ADV_BLEND(softlight); break;
682 case BLENDEQUATION_ADVANCED_DIFFERENCE: SAMPLE_REGISTER_ADV_BLEND(difference); break;
683 case BLENDEQUATION_ADVANCED_EXCLUSION: SAMPLE_REGISTER_ADV_BLEND(exclusion); break;
684 case BLENDEQUATION_ADVANCED_HSL_HUE: SAMPLE_REGISTER_ADV_BLEND_HSL(setLumSat(srcColor, dstColor, dstColor)); break;
685 case BLENDEQUATION_ADVANCED_HSL_SATURATION: SAMPLE_REGISTER_ADV_BLEND_HSL(setLumSat(dstColor, srcColor, dstColor)); break;
686 case BLENDEQUATION_ADVANCED_HSL_COLOR: SAMPLE_REGISTER_ADV_BLEND_HSL(setLum(srcColor, dstColor)); break;
687 case BLENDEQUATION_ADVANCED_HSL_LUMINOSITY: SAMPLE_REGISTER_ADV_BLEND_HSL(setLum(dstColor, srcColor)); break;
692 #undef SAMPLE_REGISTER_ADV_BLEND
693 #undef SAMPLE_REGISTER_ADV_BLEND_HSL
696 void FragmentProcessor::executeColorWrite (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, bool isSRGB, const tcu::PixelBufferAccess& colorBuffer)
698 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
700 if (m_sampleRegister[regSampleNdx].isAlive)
702 int fragSampleNdx = regSampleNdx % numSamplesPerFragment;
703 const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment];
706 combinedColor.xyz() = m_sampleRegister[regSampleNdx].blendedRGB;
707 combinedColor.w() = m_sampleRegister[regSampleNdx].blendedA;
710 combinedColor = tcu::linearToSRGB(combinedColor);
712 colorBuffer.setPixel(combinedColor, fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y());
717 void FragmentProcessor::executeRGBA8ColorWrite (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const tcu::PixelBufferAccess& colorBuffer)
719 const int fragStride = 4;
720 const int xStride = colorBuffer.getRowPitch();
721 const int yStride = colorBuffer.getSlicePitch();
722 deUint8* const basePtr = (deUint8*)colorBuffer.getDataPtr();
724 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
726 if (m_sampleRegister[regSampleNdx].isAlive)
728 const int fragSampleNdx = regSampleNdx % numSamplesPerFragment;
729 const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment];
730 deUint8* dstPtr = basePtr + fragSampleNdx*fragStride + frag.pixelCoord.x()*xStride + frag.pixelCoord.y()*yStride;
732 dstPtr[0] = tcu::floatToU8(m_sampleRegister[regSampleNdx].blendedRGB.x());
733 dstPtr[1] = tcu::floatToU8(m_sampleRegister[regSampleNdx].blendedRGB.y());
734 dstPtr[2] = tcu::floatToU8(m_sampleRegister[regSampleNdx].blendedRGB.z());
735 dstPtr[3] = tcu::floatToU8(m_sampleRegister[regSampleNdx].blendedA);
740 void FragmentProcessor::executeMaskedColorWrite (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const Vec4& colorMaskFactor, const Vec4& colorMaskNegationFactor, bool isSRGB, const tcu::PixelBufferAccess& colorBuffer)
742 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
744 if (m_sampleRegister[regSampleNdx].isAlive)
746 int fragSampleNdx = regSampleNdx % numSamplesPerFragment;
747 const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment];
748 Vec4 originalColor = colorBuffer.getPixel(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y());
751 newColor.xyz() = m_sampleRegister[regSampleNdx].blendedRGB;
752 newColor.w() = m_sampleRegister[regSampleNdx].blendedA;
755 newColor = tcu::linearToSRGB(newColor);
757 newColor = colorMaskFactor*newColor + colorMaskNegationFactor*originalColor;
759 colorBuffer.setPixel(newColor, fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y());
764 void FragmentProcessor::executeSignedValueWrite (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const tcu::BVec4& colorMask, const tcu::PixelBufferAccess& colorBuffer)
766 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
768 if (m_sampleRegister[regSampleNdx].isAlive)
770 int fragSampleNdx = regSampleNdx % numSamplesPerFragment;
771 const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment];
772 const IVec4 originalValue = colorBuffer.getPixelInt(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y());
774 colorBuffer.setPixel(tcu::select(m_sampleRegister[regSampleNdx].signedValue, originalValue, colorMask), fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y());
779 void FragmentProcessor::executeUnsignedValueWrite (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const tcu::BVec4& colorMask, const tcu::PixelBufferAccess& colorBuffer)
781 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
783 if (m_sampleRegister[regSampleNdx].isAlive)
785 int fragSampleNdx = regSampleNdx % numSamplesPerFragment;
786 const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment];
787 const UVec4 originalValue = colorBuffer.getPixelUint(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y());
789 colorBuffer.setPixel(tcu::select(m_sampleRegister[regSampleNdx].unsignedValue, originalValue, colorMask), fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y());
794 void FragmentProcessor::render (const rr::MultisamplePixelBufferAccess& msColorBuffer,
795 const rr::MultisamplePixelBufferAccess& msDepthBuffer,
796 const rr::MultisamplePixelBufferAccess& msStencilBuffer,
797 const Fragment* inputFragments,
799 FaceType fragmentFacing,
800 const FragmentOperationState& state)
802 DE_ASSERT(fragmentFacing < FACETYPE_LAST);
803 DE_ASSERT(state.numStencilBits < 32); // code bitshifts numStencilBits, avoid undefined behavior
805 const tcu::PixelBufferAccess& colorBuffer = msColorBuffer.raw();
806 const tcu::PixelBufferAccess& depthBuffer = msDepthBuffer.raw();
807 const tcu::PixelBufferAccess& stencilBuffer = msStencilBuffer.raw();
809 bool hasDepth = depthBuffer.getWidth() > 0 && depthBuffer.getHeight() > 0 && depthBuffer.getDepth() > 0;
810 bool hasStencil = stencilBuffer.getWidth() > 0 && stencilBuffer.getHeight() > 0 && stencilBuffer.getDepth() > 0;
811 bool doDepthBoundsTest = hasDepth && state.depthBoundsTestEnabled;
812 bool doDepthTest = hasDepth && state.depthTestEnabled;
813 bool doStencilTest = hasStencil && state.stencilTestEnabled;
815 tcu::TextureChannelClass colorbufferClass = tcu::getTextureChannelClass(msColorBuffer.raw().getFormat().type);
816 rr::GenericVecType fragmentDataType = (colorbufferClass == tcu::TEXTURECHANNELCLASS_SIGNED_INTEGER) ? (rr::GENERICVECTYPE_INT32) : ((colorbufferClass == tcu::TEXTURECHANNELCLASS_UNSIGNED_INTEGER) ? (rr::GENERICVECTYPE_UINT32) : (rr::GENERICVECTYPE_FLOAT));
818 DE_ASSERT((!hasDepth || colorBuffer.getWidth() == depthBuffer.getWidth()) && (!hasStencil || colorBuffer.getWidth() == stencilBuffer.getWidth()));
819 DE_ASSERT((!hasDepth || colorBuffer.getHeight() == depthBuffer.getHeight()) && (!hasStencil || colorBuffer.getHeight() == stencilBuffer.getHeight()));
820 DE_ASSERT((!hasDepth || colorBuffer.getDepth() == depthBuffer.getDepth()) && (!hasStencil || colorBuffer.getDepth() == stencilBuffer.getDepth()));
822 // Combined formats must be separated beforehand
823 DE_ASSERT(!hasDepth || (!tcu::isCombinedDepthStencilType(depthBuffer.getFormat().type) && depthBuffer.getFormat().order == tcu::TextureFormat::D));
824 DE_ASSERT(!hasStencil || (!tcu::isCombinedDepthStencilType(stencilBuffer.getFormat().type) && stencilBuffer.getFormat().order == tcu::TextureFormat::S));
826 int numSamplesPerFragment = colorBuffer.getWidth();
827 int totalNumSamples = numFragments*numSamplesPerFragment;
828 int numSampleGroups = (totalNumSamples - 1) / SAMPLE_REGISTER_SIZE + 1; // \note totalNumSamples/SAMPLE_REGISTER_SIZE rounded up.
829 const StencilState& stencilState = state.stencilStates[fragmentFacing];
830 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);
831 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);
832 bool sRGBTarget = state.sRGBEnabled && tcu::isSRGB(colorBuffer.getFormat());
834 DE_ASSERT(SAMPLE_REGISTER_SIZE % numSamplesPerFragment == 0);
836 // Divide the fragments' samples into groups of size SAMPLE_REGISTER_SIZE, and perform
837 // the per-sample operations for one group at a time.
839 for (int sampleGroupNdx = 0; sampleGroupNdx < numSampleGroups; sampleGroupNdx++)
841 // The index of the fragment of the sample at the beginning of m_sampleRegisters.
842 int groupFirstFragNdx = (sampleGroupNdx*SAMPLE_REGISTER_SIZE) / numSamplesPerFragment;
844 // Initialize sample data in the sample register.
846 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
848 int fragNdx = groupFirstFragNdx + regSampleNdx/numSamplesPerFragment;
849 int fragSampleNdx = regSampleNdx % numSamplesPerFragment;
851 if (fragNdx < numFragments)
853 m_sampleRegister[regSampleNdx].isAlive = (inputFragments[fragNdx].coverage & (1u << fragSampleNdx)) != 0;
854 m_sampleRegister[regSampleNdx].depthPassed = true; // \note This will stay true if depth test is disabled.
857 m_sampleRegister[regSampleNdx].isAlive = false;
862 if (state.scissorTestEnabled)
863 executeScissorTest(groupFirstFragNdx, numSamplesPerFragment, inputFragments, state.scissorRectangle);
865 // Depth bounds test.
867 if (doDepthBoundsTest)
868 executeDepthBoundsTest(groupFirstFragNdx, numSamplesPerFragment, inputFragments, state.minDepthBound, state.maxDepthBound, depthBuffer);
874 executeStencilCompare(groupFirstFragNdx, numSamplesPerFragment, inputFragments, stencilState, state.numStencilBits, stencilBuffer);
875 executeStencilSFail(groupFirstFragNdx, numSamplesPerFragment, inputFragments, stencilState, state.numStencilBits, stencilBuffer);
879 // \note Current value of isAlive is needed for dpPass and dpFail, so it's only updated after them and not right after depth test.
883 executeDepthCompare(groupFirstFragNdx, numSamplesPerFragment, inputFragments, state.depthFunc, depthBuffer);
886 executeDepthWrite(groupFirstFragNdx, numSamplesPerFragment, inputFragments, depthBuffer);
889 // Do dpFail and dpPass stencil writes.
892 executeStencilDpFailAndPass(groupFirstFragNdx, numSamplesPerFragment, inputFragments, stencilState, state.numStencilBits, stencilBuffer);
894 // Kill the samples that failed depth test.
898 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
899 m_sampleRegister[regSampleNdx].isAlive = m_sampleRegister[regSampleNdx].isAlive && m_sampleRegister[regSampleNdx].depthPassed;
902 // Paint fragments to target
904 switch (fragmentDataType)
906 case rr::GENERICVECTYPE_FLOAT:
908 // Select min/max clamping values for blending factors and operands
912 if (colorbufferClass == tcu::TEXTURECHANNELCLASS_UNSIGNED_FIXED_POINT)
914 minClampValue = Vec4(0.0f);
915 maxClampValue = Vec4(1.0f);
917 else if (colorbufferClass == tcu::TEXTURECHANNELCLASS_SIGNED_FIXED_POINT)
919 minClampValue = Vec4(-1.0f);
920 maxClampValue = Vec4(1.0f);
925 minClampValue = Vec4(-std::numeric_limits<float>::infinity());
926 maxClampValue = Vec4(std::numeric_limits<float>::infinity());
929 // Blend calculation - only if using blend.
930 if (state.blendMode == BLENDMODE_STANDARD)
932 // Put dst color to register, doing srgb-to-linear conversion if needed.
933 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
935 if (m_sampleRegister[regSampleNdx].isAlive)
937 int fragSampleNdx = regSampleNdx % numSamplesPerFragment;
938 const Fragment& frag = inputFragments[groupFirstFragNdx + regSampleNdx/numSamplesPerFragment];
939 Vec4 dstColor = colorBuffer.getPixel(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y());
941 m_sampleRegister[regSampleNdx].clampedBlendSrcColor = clamp(frag.value.get<float>(), minClampValue, maxClampValue);
942 m_sampleRegister[regSampleNdx].clampedBlendSrc1Color = clamp(frag.value1.get<float>(), minClampValue, maxClampValue);
943 m_sampleRegister[regSampleNdx].clampedBlendDstColor = clamp(sRGBTarget ? tcu::sRGBToLinear(dstColor) : dstColor, minClampValue, maxClampValue);
947 // Calculate blend factors to register.
948 executeBlendFactorComputeRGB(state.blendColor, state.blendRGBState);
949 executeBlendFactorComputeA(state.blendColor, state.blendAState);
951 // Compute blended color.
952 executeBlend(state.blendRGBState, state.blendAState);
954 else if (state.blendMode == BLENDMODE_ADVANCED)
956 // Unpremultiply colors for blending, and do sRGB->linear if necessary
957 // \todo [2014-03-17 pyry] Re-consider clampedBlend*Color var names
958 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
960 if (m_sampleRegister[regSampleNdx].isAlive)
962 int fragSampleNdx = regSampleNdx % numSamplesPerFragment;
963 const Fragment& frag = inputFragments[groupFirstFragNdx + regSampleNdx/numSamplesPerFragment];
964 const Vec4 srcColor = frag.value.get<float>();
965 const Vec4 dstColor = colorBuffer.getPixel(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y());
967 m_sampleRegister[regSampleNdx].clampedBlendSrcColor = unpremultiply(clamp(srcColor, minClampValue, maxClampValue));
968 m_sampleRegister[regSampleNdx].clampedBlendDstColor = unpremultiply(clamp(sRGBTarget ? tcu::sRGBToLinear(dstColor) : dstColor, minClampValue, maxClampValue));
972 executeAdvancedBlend(state.blendEquationAdvaced);
976 // Not using blend - just put values to register as-is.
977 DE_ASSERT(state.blendMode == BLENDMODE_NONE);
979 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
981 if (m_sampleRegister[regSampleNdx].isAlive)
983 const Fragment& frag = inputFragments[groupFirstFragNdx + regSampleNdx/numSamplesPerFragment];
985 m_sampleRegister[regSampleNdx].blendedRGB = frag.value.get<float>().xyz();
986 m_sampleRegister[regSampleNdx].blendedA = frag.value.get<float>().w();
991 // Clamp result values in sample register
992 if (colorbufferClass != tcu::TEXTURECHANNELCLASS_FLOATING_POINT)
994 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
996 if (m_sampleRegister[regSampleNdx].isAlive)
998 m_sampleRegister[regSampleNdx].blendedRGB = clamp(m_sampleRegister[regSampleNdx].blendedRGB, minClampValue.swizzle(0, 1, 2), maxClampValue.swizzle(0, 1, 2));
999 m_sampleRegister[regSampleNdx].blendedA = clamp(m_sampleRegister[regSampleNdx].blendedA, minClampValue.w(), maxClampValue.w());
1004 // Finally, write the colors to the color buffer.
1006 if (state.colorMask[0] && state.colorMask[1] && state.colorMask[2] && state.colorMask[3])
1008 if (colorBuffer.getFormat() == tcu::TextureFormat(tcu::TextureFormat::RGBA, tcu::TextureFormat::UNORM_INT8))
1009 executeRGBA8ColorWrite(groupFirstFragNdx, numSamplesPerFragment, inputFragments, colorBuffer);
1011 executeColorWrite(groupFirstFragNdx, numSamplesPerFragment, inputFragments, sRGBTarget, colorBuffer);
1013 else if (state.colorMask[0] || state.colorMask[1] || state.colorMask[2] || state.colorMask[3])
1014 executeMaskedColorWrite(groupFirstFragNdx, numSamplesPerFragment, inputFragments, colorMaskFactor, colorMaskNegationFactor, sRGBTarget, colorBuffer);
1017 case rr::GENERICVECTYPE_INT32:
1019 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
1021 if (m_sampleRegister[regSampleNdx].isAlive)
1023 const Fragment& frag = inputFragments[groupFirstFragNdx + regSampleNdx/numSamplesPerFragment];
1025 m_sampleRegister[regSampleNdx].signedValue = frag.value.get<deInt32>();
1029 if (state.colorMask[0] || state.colorMask[1] || state.colorMask[2] || state.colorMask[3])
1030 executeSignedValueWrite(groupFirstFragNdx, numSamplesPerFragment, inputFragments, state.colorMask, colorBuffer);
1033 case rr::GENERICVECTYPE_UINT32:
1035 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
1037 if (m_sampleRegister[regSampleNdx].isAlive)
1039 const Fragment& frag = inputFragments[groupFirstFragNdx + regSampleNdx/numSamplesPerFragment];
1041 m_sampleRegister[regSampleNdx].unsignedValue = frag.value.get<deUint32>();
1045 if (state.colorMask[0] || state.colorMask[1] || state.colorMask[2] || state.colorMask[3])
1046 executeUnsignedValueWrite(groupFirstFragNdx, numSamplesPerFragment, inputFragments, state.colorMask, colorBuffer);
1050 DE_ASSERT(DE_FALSE);