1 /*-------------------------------------------------------------------------
2 * drawElements Quality Program Reference Renderer
3 * -----------------------------------------------
4 *
5 * Copyright 2014 The Android Open Source Project
6 *
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
10 *
11 * http://www.apache.org/licenses/LICENSE-2.0
12 *
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.
18 *
19 *//*!
20 * \file
21 * \brief Reference implementation for per-fragment operations.
22 *//*--------------------------------------------------------------------*/
23
24 #include "rrFragmentOperations.hpp"
25 #include "tcuVectorUtil.hpp"
26 #include "tcuTextureUtil.hpp"
27 #include <limits>
28
29 using tcu::IVec2;
30 using tcu::Vec3;
31 using tcu::Vec4;
32 using tcu::IVec4;
33 using tcu::UVec4;
34 using tcu::min;
35 using tcu::max;
36 using tcu::clamp;
37 using de::min;
38 using de::max;
39 using de::clamp;
40
41 namespace rr
42 {
43
44 // Return oldValue with the bits indicated by mask replaced by corresponding bits of newValue.
maskedBitReplace(int oldValue,int newValue,deUint32 mask)45 static inline int maskedBitReplace (int oldValue, int newValue, deUint32 mask)
46 {
47 return (oldValue & ~mask) | (newValue & mask);
48 }
49
isInsideRect(const IVec2 & point,const WindowRectangle & rect)50 static inline bool isInsideRect (const IVec2& point, const WindowRectangle& rect)
51 {
52 return de::inBounds(point.x(), rect.left, rect.left + rect.width) &&
53 de::inBounds(point.y(), rect.bottom, rect.bottom + rect.height);
54 }
55
unpremultiply(const Vec4 & v)56 static inline Vec4 unpremultiply (const Vec4& v)
57 {
58 if (v.w() > 0.0f)
59 return Vec4(v.x()/v.w(), v.y()/v.w(), v.z()/v.w(), v.w());
60 else
61 {
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);
64 }
65 }
66
clearMultisampleColorBuffer(const tcu::PixelBufferAccess & dst,const Vec4 & v,const WindowRectangle & r)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); }
clearMultisampleColorBuffer(const tcu::PixelBufferAccess & dst,const IVec4 & v,const WindowRectangle & r)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); }
clearMultisampleColorBuffer(const tcu::PixelBufferAccess & dst,const UVec4 & v,const WindowRectangle & r)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>()); }
clearMultisampleDepthBuffer(const tcu::PixelBufferAccess & dst,float v,const WindowRectangle & r)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); }
clearMultisampleStencilBuffer(const tcu::PixelBufferAccess & dst,int v,const WindowRectangle & r)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); }
72
FragmentProcessor(void)73 FragmentProcessor::FragmentProcessor (void)
74 : m_sampleRegister()
75 {
76 }
77
executeScissorTest(int fragNdxOffset,int numSamplesPerFragment,const Fragment * inputFragments,const WindowRectangle & scissorRect)78 void FragmentProcessor::executeScissorTest (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const WindowRectangle& scissorRect)
79 {
80 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
81 {
82 if (m_sampleRegister[regSampleNdx].isAlive)
83 {
84 int fragNdx = fragNdxOffset + regSampleNdx/numSamplesPerFragment;
85
86 if (!isInsideRect(inputFragments[fragNdx].pixelCoord, scissorRect))
87 m_sampleRegister[regSampleNdx].isAlive = false;
88 }
89 }
90 }
91
executeStencilCompare(int fragNdxOffset,int numSamplesPerFragment,const Fragment * inputFragments,const StencilState & stencilState,int numStencilBits,const tcu::ConstPixelBufferAccess & stencilBuffer)92 void FragmentProcessor::executeStencilCompare (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const StencilState& stencilState, int numStencilBits, const tcu::ConstPixelBufferAccess& stencilBuffer)
93 {
94 #define SAMPLE_REGISTER_STENCIL_COMPARE(COMPARE_EXPRESSION) \
95 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \
96 { \
97 if (m_sampleRegister[regSampleNdx].isAlive) \
98 { \
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); \
106 \
107 m_sampleRegister[regSampleNdx].stencilPassed = (COMPARE_EXPRESSION); \
108 } \
109 }
110
111 int clampedStencilRef = de::clamp(stencilState.ref, 0, (1<<numStencilBits)-1);
112
113 switch (stencilState.func)
114 {
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;
123 default:
124 DE_ASSERT(false);
125 }
126
127 #undef SAMPLE_REGISTER_STENCIL_COMPARE
128 }
129
executeStencilSFail(int fragNdxOffset,int numSamplesPerFragment,const Fragment * inputFragments,const StencilState & stencilState,int numStencilBits,const tcu::PixelBufferAccess & stencilBuffer)130 void FragmentProcessor::executeStencilSFail (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const StencilState& stencilState, int numStencilBits, const tcu::PixelBufferAccess& stencilBuffer)
131 {
132 #define SAMPLE_REGISTER_SFAIL(SFAIL_EXPRESSION) \
133 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \
134 { \
135 if (m_sampleRegister[regSampleNdx].isAlive && !m_sampleRegister[regSampleNdx].stencilPassed) \
136 { \
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()); \
140 \
141 stencilBuffer.setPixStencil(maskedBitReplace(stencilBufferValue, (SFAIL_EXPRESSION), stencilState.writeMask), fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); \
142 m_sampleRegister[regSampleNdx].isAlive = false; \
143 } \
144 }
145
146 int clampedStencilRef = de::clamp(stencilState.ref, 0, (1<<numStencilBits)-1);
147
148 switch (stencilState.sFail)
149 {
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;
158 default:
159 DE_ASSERT(false);
160 }
161
162 #undef SAMPLE_REGISTER_SFAIL
163 }
164
executeDepthCompare(int fragNdxOffset,int numSamplesPerFragment,const Fragment * inputFragments,TestFunc depthFunc,const tcu::ConstPixelBufferAccess & depthBuffer)165 void FragmentProcessor::executeDepthCompare (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, TestFunc depthFunc, const tcu::ConstPixelBufferAccess& depthBuffer)
166 {
167 #define SAMPLE_REGISTER_DEPTH_COMPARE_F(COMPARE_EXPRESSION) \
168 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \
169 { \
170 if (m_sampleRegister[regSampleNdx].isAlive) \
171 { \
172 int fragSampleNdx = regSampleNdx % numSamplesPerFragment; \
173 const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment]; \
174 float depthBufferValue = depthBuffer.getPixDepth(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); \
175 float sampleDepthFloat = frag.sampleDepths[fragSampleNdx]; \
176 float sampleDepth = de::clamp(sampleDepthFloat, 0.0f, 1.0f); \
177 \
178 m_sampleRegister[regSampleNdx].depthPassed = (COMPARE_EXPRESSION); \
179 \
180 DE_UNREF(depthBufferValue); \
181 DE_UNREF(sampleDepth); \
182 } \
183 }
184
185 #define SAMPLE_REGISTER_DEPTH_COMPARE_UI(COMPARE_EXPRESSION) \
186 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \
187 { \
188 if (m_sampleRegister[regSampleNdx].isAlive) \
189 { \
190 int fragSampleNdx = regSampleNdx % numSamplesPerFragment; \
191 const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment]; \
192 deUint32 depthBufferValue = depthBuffer.getPixelUint(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()).x(); \
193 float sampleDepthFloat = frag.sampleDepths[fragSampleNdx]; \
194 \
195 /* Convert input float to target buffer format for comparison */ \
196 \
197 deUint32 buffer[2]; \
198 \
199 DE_ASSERT(sizeof(buffer) >= (size_t)depthBuffer.getFormat().getPixelSize()); \
200 \
201 tcu::PixelBufferAccess access(depthBuffer.getFormat(), 1, 1, 1, &buffer); \
202 access.setPixDepth(sampleDepthFloat, 0, 0, 0); \
203 deUint32 sampleDepth = access.getPixelUint(0, 0, 0).x(); \
204 \
205 m_sampleRegister[regSampleNdx].depthPassed = (COMPARE_EXPRESSION); \
206 \
207 DE_UNREF(depthBufferValue); \
208 DE_UNREF(sampleDepth); \
209 } \
210 }
211
212 if (depthBuffer.getFormat().type == tcu::TextureFormat::FLOAT || depthBuffer.getFormat().type == tcu::TextureFormat::FLOAT_UNSIGNED_INT_24_8_REV)
213 {
214
215 switch (depthFunc)
216 {
217 case TESTFUNC_NEVER: SAMPLE_REGISTER_DEPTH_COMPARE_F(false) break;
218 case TESTFUNC_ALWAYS: SAMPLE_REGISTER_DEPTH_COMPARE_F(true) break;
219 case TESTFUNC_LESS: SAMPLE_REGISTER_DEPTH_COMPARE_F(sampleDepth < depthBufferValue) break;
220 case TESTFUNC_LEQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_F(sampleDepth <= depthBufferValue) break;
221 case TESTFUNC_GREATER: SAMPLE_REGISTER_DEPTH_COMPARE_F(sampleDepth > depthBufferValue) break;
222 case TESTFUNC_GEQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_F(sampleDepth >= depthBufferValue) break;
223 case TESTFUNC_EQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_F(sampleDepth == depthBufferValue) break;
224 case TESTFUNC_NOTEQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_F(sampleDepth != depthBufferValue) break;
225 default:
226 DE_ASSERT(false);
227 }
228
229 }
230 else
231 {
232 switch (depthFunc)
233 {
234 case TESTFUNC_NEVER: SAMPLE_REGISTER_DEPTH_COMPARE_UI(false) break;
235 case TESTFUNC_ALWAYS: SAMPLE_REGISTER_DEPTH_COMPARE_UI(true) break;
236 case TESTFUNC_LESS: SAMPLE_REGISTER_DEPTH_COMPARE_UI(sampleDepth < depthBufferValue) break;
237 case TESTFUNC_LEQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_UI(sampleDepth <= depthBufferValue) break;
238 case TESTFUNC_GREATER: SAMPLE_REGISTER_DEPTH_COMPARE_UI(sampleDepth > depthBufferValue) break;
239 case TESTFUNC_GEQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_UI(sampleDepth >= depthBufferValue) break;
240 case TESTFUNC_EQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_UI(sampleDepth == depthBufferValue) break;
241 case TESTFUNC_NOTEQUAL: SAMPLE_REGISTER_DEPTH_COMPARE_UI(sampleDepth != depthBufferValue) break;
242 default:
243 DE_ASSERT(false);
244 }
245 }
246
247 #undef SAMPLE_REGISTER_DEPTH_COMPARE_F
248 #undef SAMPLE_REGISTER_DEPTH_COMPARE_UI
249 }
250
executeDepthWrite(int fragNdxOffset,int numSamplesPerFragment,const Fragment * inputFragments,const tcu::PixelBufferAccess & depthBuffer)251 void FragmentProcessor::executeDepthWrite (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const tcu::PixelBufferAccess& depthBuffer)
252 {
253 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
254 {
255 if (m_sampleRegister[regSampleNdx].isAlive && m_sampleRegister[regSampleNdx].depthPassed)
256 {
257 int fragSampleNdx = regSampleNdx % numSamplesPerFragment;
258 const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment];
259 const float clampedDepth = de::clamp(frag.sampleDepths[fragSampleNdx], 0.0f, 1.0f);
260
261 depthBuffer.setPixDepth(clampedDepth, fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y());
262 }
263 }
264 }
265
executeStencilDpFailAndPass(int fragNdxOffset,int numSamplesPerFragment,const Fragment * inputFragments,const StencilState & stencilState,int numStencilBits,const tcu::PixelBufferAccess & stencilBuffer)266 void FragmentProcessor::executeStencilDpFailAndPass (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const StencilState& stencilState, int numStencilBits, const tcu::PixelBufferAccess& stencilBuffer)
267 {
268 #define SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, EXPRESSION) \
269 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \
270 { \
271 if (m_sampleRegister[regSampleNdx].isAlive && (CONDITION)) \
272 { \
273 int fragSampleNdx = regSampleNdx % numSamplesPerFragment; \
274 const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment]; \
275 int stencilBufferValue = stencilBuffer.getPixStencil(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); \
276 \
277 stencilBuffer.setPixStencil(maskedBitReplace(stencilBufferValue, (EXPRESSION), stencilState.writeMask), fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y()); \
278 } \
279 }
280
281 #define SWITCH_DPFAIL_OR_DPPASS(OP_NAME, CONDITION) \
282 switch (stencilState.OP_NAME) \
283 { \
284 case STENCILOP_KEEP: SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, stencilBufferValue) break; \
285 case STENCILOP_ZERO: SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, 0) break; \
286 case STENCILOP_REPLACE: SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, clampedStencilRef) break; \
287 case STENCILOP_INCR: SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, de::clamp(stencilBufferValue+1, 0, (1<<numStencilBits) - 1)) break; \
288 case STENCILOP_DECR: SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, de::clamp(stencilBufferValue-1, 0, (1<<numStencilBits) - 1)) break; \
289 case STENCILOP_INCR_WRAP: SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, (stencilBufferValue + 1) & ((1<<numStencilBits) - 1)) break; \
290 case STENCILOP_DECR_WRAP: SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, (stencilBufferValue - 1) & ((1<<numStencilBits) - 1)) break; \
291 case STENCILOP_INVERT: SAMPLE_REGISTER_DPFAIL_OR_DPPASS(CONDITION, (~stencilBufferValue) & ((1<<numStencilBits) - 1)) break; \
292 default: \
293 DE_ASSERT(false); \
294 }
295
296 int clampedStencilRef = de::clamp(stencilState.ref, 0, (1<<numStencilBits)-1);
297
298 SWITCH_DPFAIL_OR_DPPASS(dpFail, !m_sampleRegister[regSampleNdx].depthPassed)
299 SWITCH_DPFAIL_OR_DPPASS(dpPass, m_sampleRegister[regSampleNdx].depthPassed)
300
301 #undef SWITCH_DPFAIL_OR_DPPASS
302 #undef SAMPLE_REGISTER_DPFAIL_OR_DPPASS
303 }
304
executeBlendFactorComputeRGB(const Vec4 & blendColor,const BlendState & blendRGBState)305 void FragmentProcessor::executeBlendFactorComputeRGB (const Vec4& blendColor, const BlendState& blendRGBState)
306 {
307 #define SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, FACTOR_EXPRESSION) \
308 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \
309 { \
310 if (m_sampleRegister[regSampleNdx].isAlive) \
311 { \
312 const Vec4& src = m_sampleRegister[regSampleNdx].clampedBlendSrcColor; \
313 const Vec4& src1 = m_sampleRegister[regSampleNdx].clampedBlendSrc1Color; \
314 const Vec4& dst = m_sampleRegister[regSampleNdx].clampedBlendDstColor; \
315 DE_UNREF(src); \
316 DE_UNREF(src1); \
317 DE_UNREF(dst); \
318 \
319 m_sampleRegister[regSampleNdx].FACTOR_NAME = (FACTOR_EXPRESSION); \
320 } \
321 }
322
323 #define SWITCH_SRC_OR_DST_FACTOR_RGB(FUNC_NAME, FACTOR_NAME) \
324 switch (blendRGBState.FUNC_NAME) \
325 { \
326 case BLENDFUNC_ZERO: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(0.0f)) break; \
327 case BLENDFUNC_ONE: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(1.0f)) break; \
328 case BLENDFUNC_SRC_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, src.swizzle(0,1,2)) break; \
329 case BLENDFUNC_ONE_MINUS_SRC_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(1.0f) - src.swizzle(0,1,2)) break; \
330 case BLENDFUNC_DST_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, dst.swizzle(0,1,2)) break; \
331 case BLENDFUNC_ONE_MINUS_DST_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(1.0f) - dst.swizzle(0,1,2)) break; \
332 case BLENDFUNC_SRC_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(src.w())) break; \
333 case BLENDFUNC_ONE_MINUS_SRC_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(1.0f - src.w())) break; \
334 case BLENDFUNC_DST_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(dst.w())) break; \
335 case BLENDFUNC_ONE_MINUS_DST_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(1.0f - dst.w())) break; \
336 case BLENDFUNC_CONSTANT_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, blendColor.swizzle(0,1,2)) break; \
337 case BLENDFUNC_ONE_MINUS_CONSTANT_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(1.0f) - blendColor.swizzle(0,1,2)) break; \
338 case BLENDFUNC_CONSTANT_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(blendColor.w())) break; \
339 case BLENDFUNC_ONE_MINUS_CONSTANT_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(1.0f - blendColor.w())) break; \
340 case BLENDFUNC_SRC_ALPHA_SATURATE: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(de::min(src.w(), 1.0f - dst.w()))) break; \
341 case BLENDFUNC_SRC1_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, src1.swizzle(0,1,2)) break; \
342 case BLENDFUNC_ONE_MINUS_SRC1_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(1.0f) - src1.swizzle(0,1,2)) break; \
343 case BLENDFUNC_SRC1_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(src1.w())) break; \
344 case BLENDFUNC_ONE_MINUS_SRC1_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, Vec3(1.0f - src1.w())) break; \
345 default: \
346 DE_ASSERT(false); \
347 }
348
349 SWITCH_SRC_OR_DST_FACTOR_RGB(srcFunc, blendSrcFactorRGB)
350 SWITCH_SRC_OR_DST_FACTOR_RGB(dstFunc, blendDstFactorRGB)
351
352 #undef SWITCH_SRC_OR_DST_FACTOR_RGB
353 #undef SAMPLE_REGISTER_BLEND_FACTOR
354 }
355
executeBlendFactorComputeA(const Vec4 & blendColor,const BlendState & blendAState)356 void FragmentProcessor::executeBlendFactorComputeA (const Vec4& blendColor, const BlendState& blendAState)
357 {
358 #define SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, FACTOR_EXPRESSION) \
359 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \
360 { \
361 if (m_sampleRegister[regSampleNdx].isAlive) \
362 { \
363 const Vec4& src = m_sampleRegister[regSampleNdx].clampedBlendSrcColor; \
364 const Vec4& src1 = m_sampleRegister[regSampleNdx].clampedBlendSrc1Color; \
365 const Vec4& dst = m_sampleRegister[regSampleNdx].clampedBlendDstColor; \
366 DE_UNREF(src); \
367 DE_UNREF(src1); \
368 DE_UNREF(dst); \
369 \
370 m_sampleRegister[regSampleNdx].FACTOR_NAME = (FACTOR_EXPRESSION); \
371 } \
372 }
373
374 #define SWITCH_SRC_OR_DST_FACTOR_A(FUNC_NAME, FACTOR_NAME) \
375 switch (blendAState.FUNC_NAME) \
376 { \
377 case BLENDFUNC_ZERO: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, 0.0f) break; \
378 case BLENDFUNC_ONE: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, 1.0f) break; \
379 case BLENDFUNC_SRC_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, src.w()) break; \
380 case BLENDFUNC_ONE_MINUS_SRC_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, 1.0f - src.w()) break; \
381 case BLENDFUNC_DST_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, dst.w()) break; \
382 case BLENDFUNC_ONE_MINUS_DST_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, 1.0f - dst.w()) break; \
383 case BLENDFUNC_SRC_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, src.w()) break; \
384 case BLENDFUNC_ONE_MINUS_SRC_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, 1.0f - src.w()) break; \
385 case BLENDFUNC_DST_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, dst.w()) break; \
386 case BLENDFUNC_ONE_MINUS_DST_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, 1.0f - dst.w()) break; \
387 case BLENDFUNC_CONSTANT_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, blendColor.w()) break; \
388 case BLENDFUNC_ONE_MINUS_CONSTANT_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, 1.0f - blendColor.w()) break; \
389 case BLENDFUNC_CONSTANT_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, blendColor.w()) break; \
390 case BLENDFUNC_ONE_MINUS_CONSTANT_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, 1.0f - blendColor.w()) break; \
391 case BLENDFUNC_SRC_ALPHA_SATURATE: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, 1.0f) break; \
392 case BLENDFUNC_SRC1_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, src1.w()) break; \
393 case BLENDFUNC_ONE_MINUS_SRC1_COLOR: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, 1.0f - src1.w()) break; \
394 case BLENDFUNC_SRC1_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, src1.w()) break; \
395 case BLENDFUNC_ONE_MINUS_SRC1_ALPHA: SAMPLE_REGISTER_BLEND_FACTOR(FACTOR_NAME, 1.0f - src1.w()) break; \
396 default: \
397 DE_ASSERT(false); \
398 }
399
400 SWITCH_SRC_OR_DST_FACTOR_A(srcFunc, blendSrcFactorA)
401 SWITCH_SRC_OR_DST_FACTOR_A(dstFunc, blendDstFactorA)
402
403 #undef SWITCH_SRC_OR_DST_FACTOR_A
404 #undef SAMPLE_REGISTER_BLEND_FACTOR
405 }
406
executeBlend(const BlendState & blendRGBState,const BlendState & blendAState)407 void FragmentProcessor::executeBlend (const BlendState& blendRGBState, const BlendState& blendAState)
408 {
409 #define SAMPLE_REGISTER_BLENDED_COLOR(COLOR_NAME, COLOR_EXPRESSION) \
410 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \
411 { \
412 if (m_sampleRegister[regSampleNdx].isAlive) \
413 { \
414 SampleData& sample = m_sampleRegister[regSampleNdx]; \
415 const Vec4& srcColor = sample.clampedBlendSrcColor; \
416 const Vec4& dstColor = sample.clampedBlendDstColor; \
417 \
418 sample.COLOR_NAME = (COLOR_EXPRESSION); \
419 } \
420 }
421
422 switch (blendRGBState.equation)
423 {
424 case BLENDEQUATION_ADD: SAMPLE_REGISTER_BLENDED_COLOR(blendedRGB, srcColor.swizzle(0,1,2)*sample.blendSrcFactorRGB + dstColor.swizzle(0,1,2)*sample.blendDstFactorRGB) break;
425 case BLENDEQUATION_SUBTRACT: SAMPLE_REGISTER_BLENDED_COLOR(blendedRGB, srcColor.swizzle(0,1,2)*sample.blendSrcFactorRGB - dstColor.swizzle(0,1,2)*sample.blendDstFactorRGB) break;
426 case BLENDEQUATION_REVERSE_SUBTRACT: SAMPLE_REGISTER_BLENDED_COLOR(blendedRGB, dstColor.swizzle(0,1,2)*sample.blendDstFactorRGB - srcColor.swizzle(0,1,2)*sample.blendSrcFactorRGB) break;
427 case BLENDEQUATION_MIN: SAMPLE_REGISTER_BLENDED_COLOR(blendedRGB, min(srcColor.swizzle(0,1,2), dstColor.swizzle(0,1,2))) break;
428 case BLENDEQUATION_MAX: SAMPLE_REGISTER_BLENDED_COLOR(blendedRGB, max(srcColor.swizzle(0,1,2), dstColor.swizzle(0,1,2))) break;
429 default:
430 DE_ASSERT(false);
431 }
432
433 switch (blendAState.equation)
434 {
435 case BLENDEQUATION_ADD: SAMPLE_REGISTER_BLENDED_COLOR(blendedA, srcColor.w()*sample.blendSrcFactorA + dstColor.w()*sample.blendDstFactorA) break;
436 case BLENDEQUATION_SUBTRACT: SAMPLE_REGISTER_BLENDED_COLOR(blendedA, srcColor.w()*sample.blendSrcFactorA - dstColor.w()*sample.blendDstFactorA) break;
437 case BLENDEQUATION_REVERSE_SUBTRACT: SAMPLE_REGISTER_BLENDED_COLOR(blendedA, dstColor.w()*sample.blendDstFactorA - srcColor.w()*sample.blendSrcFactorA) break;
438 case BLENDEQUATION_MIN: SAMPLE_REGISTER_BLENDED_COLOR(blendedA, min(srcColor.w(), dstColor.w())) break;
439 case BLENDEQUATION_MAX: SAMPLE_REGISTER_BLENDED_COLOR(blendedA, max(srcColor.w(), dstColor.w())) break;
440 default:
441 DE_ASSERT(false);
442 }
443 #undef SAMPLE_REGISTER_BLENDED_COLOR
444 }
445
446 namespace advblend
447 {
448
multiply(float src,float dst)449 inline float multiply (float src, float dst) { return src*dst; }
screen(float src,float dst)450 inline float screen (float src, float dst) { return src + dst - src*dst; }
darken(float src,float dst)451 inline float darken (float src, float dst) { return de::min(src, dst); }
lighten(float src,float dst)452 inline float lighten (float src, float dst) { return de::max(src, dst); }
difference(float src,float dst)453 inline float difference (float src, float dst) { return de::abs(dst-src); }
exclusion(float src,float dst)454 inline float exclusion (float src, float dst) { return src + dst - 2.0f*src*dst; }
455
overlay(float src,float dst)456 inline float overlay (float src, float dst)
457 {
458 if (dst <= 0.5f)
459 return 2.0f*src*dst;
460 else
461 return 1.0f - 2.0f*(1.0f-src)*(1.0f-dst);
462 }
463
colordodge(float src,float dst)464 inline float colordodge (float src, float dst)
465 {
466 if (dst <= 0.0f)
467 return 0.0f;
468 else if (src < 1.0f)
469 return de::min(1.0f, dst/(1.0f-src));
470 else
471 return 1.0f;
472 }
473
colorburn(float src,float dst)474 inline float colorburn (float src, float dst)
475 {
476 if (dst >= 1.0f)
477 return 1.0f;
478 else if (src > 0.0f)
479 return 1.0f - de::min(1.0f, (1.0f-dst)/src);
480 else
481 return 0.0f;
482 }
483
hardlight(float src,float dst)484 inline float hardlight (float src, float dst)
485 {
486 if (src <= 0.5f)
487 return 2.0f*src*dst;
488 else
489 return 1.0f - 2.0f*(1.0f-src)*(1.0f-dst);
490 }
491
softlight(float src,float dst)492 inline float softlight (float src, float dst)
493 {
494 if (src <= 0.5f)
495 return dst - (1.0f - 2.0f*src)*dst*(1.0f-dst);
496 else if (dst <= 0.25f)
497 return dst + (2.0f*src - 1.0f)*dst*((16.0f*dst - 12.0f)*dst + 3.0f);
498 else
499 return dst + (2.0f*src - 1.0f)*(deFloatSqrt(dst)-dst);
500 }
501
minComp(const Vec3 & v)502 inline float minComp (const Vec3& v)
503 {
504 return de::min(de::min(v.x(), v.y()), v.z());
505 }
506
maxComp(const Vec3 & v)507 inline float maxComp (const Vec3& v)
508 {
509 return de::max(de::max(v.x(), v.y()), v.z());
510 }
511
luminosity(const Vec3 & rgb)512 inline float luminosity (const Vec3& rgb)
513 {
514 return dot(rgb, Vec3(0.3f, 0.59f, 0.11f));
515 }
516
saturation(const Vec3 & rgb)517 inline float saturation (const Vec3& rgb)
518 {
519 return maxComp(rgb) - minComp(rgb);
520 }
521
setLum(const Vec3 & cbase,const Vec3 & clum)522 Vec3 setLum (const Vec3& cbase, const Vec3& clum)
523 {
524 const float lbase = luminosity(cbase);
525 const float llum = luminosity(clum);
526 const float ldiff = llum - lbase;
527 const Vec3 color = cbase + Vec3(ldiff);
528 const float minC = minComp(color);
529 const float maxC = maxComp(color);
530
531 if (minC < 0.0f)
532 return llum + ((color-llum)*llum / (llum != minC ? (llum-minC) : 1.0f));
533 else if (maxC > 1.0f)
534 return llum + ((color-llum)*(1.0f-llum) / (llum != maxC ? (maxC-llum) : 1.0f));
535 else
536 return color;
537 }
538
setLumSat(const Vec3 & cbase,const Vec3 & csat,const Vec3 & clum)539 Vec3 setLumSat (const Vec3& cbase, const Vec3& csat, const Vec3& clum)
540 {
541 const float minbase = minComp(cbase);
542 const float sbase = saturation(cbase);
543 const float ssat = saturation(csat);
544 Vec3 color = Vec3(0.0f);
545
546 if (sbase > 0.0f)
547 color = (cbase - minbase) * ssat / sbase;
548 else
549 color = color;
550
551 return setLum(color, clum);
552 }
553
554 } // advblend
555
executeAdvancedBlend(BlendEquationAdvanced equation)556 void FragmentProcessor::executeAdvancedBlend (BlendEquationAdvanced equation)
557 {
558 using namespace advblend;
559
560 #define SAMPLE_REGISTER_ADV_BLEND(FUNCTION_NAME) \
561 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \
562 { \
563 if (m_sampleRegister[regSampleNdx].isAlive) \
564 { \
565 SampleData& sample = m_sampleRegister[regSampleNdx]; \
566 const Vec4& srcColor = sample.clampedBlendSrcColor; \
567 const Vec4& dstColor = sample.clampedBlendDstColor; \
568 const Vec3& bias = sample.blendSrcFactorRGB; \
569 const float p0 = sample.blendSrcFactorA; \
570 const float r = FUNCTION_NAME(srcColor[0], dstColor[0])*p0 + bias[0]; \
571 const float g = FUNCTION_NAME(srcColor[1], dstColor[1])*p0 + bias[1]; \
572 const float b = FUNCTION_NAME(srcColor[2], dstColor[2])*p0 + bias[2]; \
573 \
574 sample.blendedRGB = Vec3(r, g, b); \
575 } \
576 }
577
578 #define SAMPLE_REGISTER_ADV_BLEND_HSL(COLOR_EXPRESSION) \
579 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++) \
580 { \
581 if (m_sampleRegister[regSampleNdx].isAlive) \
582 { \
583 SampleData& sample = m_sampleRegister[regSampleNdx]; \
584 const Vec3 srcColor = sample.clampedBlendSrcColor.swizzle(0,1,2); \
585 const Vec3 dstColor = sample.clampedBlendDstColor.swizzle(0,1,2); \
586 const Vec3& bias = sample.blendSrcFactorRGB; \
587 const float p0 = sample.blendSrcFactorA; \
588 \
589 sample.blendedRGB = (COLOR_EXPRESSION)*p0 + bias; \
590 } \
591 }
592
593 // Pre-compute factors & compute alpha \todo [2014-03-18 pyry] Re-using variable names.
594 // \note clampedBlend*Color contains clamped & unpremultiplied colors
595 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
596 {
597 if (m_sampleRegister[regSampleNdx].isAlive)
598 {
599 SampleData& sample = m_sampleRegister[regSampleNdx];
600 const Vec4& srcColor = sample.clampedBlendSrcColor;
601 const Vec4& dstColor = sample.clampedBlendDstColor;
602 const float srcA = srcColor.w();
603 const float dstA = dstColor.w();
604 const float p0 = srcA*dstA;
605 const float p1 = srcA*(1.0f-dstA);
606 const float p2 = dstA*(1.0f-srcA);
607 const Vec3 bias (srcColor[0]*p1 + dstColor[0]*p2,
608 srcColor[1]*p1 + dstColor[1]*p2,
609 srcColor[2]*p1 + dstColor[2]*p2);
610
611 sample.blendSrcFactorRGB = bias;
612 sample.blendSrcFactorA = p0;
613 sample.blendedA = p0 + p1 + p2;
614 }
615 }
616
617 switch (equation)
618 {
619 case BLENDEQUATION_ADVANCED_MULTIPLY: SAMPLE_REGISTER_ADV_BLEND(multiply); break;
620 case BLENDEQUATION_ADVANCED_SCREEN: SAMPLE_REGISTER_ADV_BLEND(screen); break;
621 case BLENDEQUATION_ADVANCED_OVERLAY: SAMPLE_REGISTER_ADV_BLEND(overlay); break;
622 case BLENDEQUATION_ADVANCED_DARKEN: SAMPLE_REGISTER_ADV_BLEND(darken); break;
623 case BLENDEQUATION_ADVANCED_LIGHTEN: SAMPLE_REGISTER_ADV_BLEND(lighten); break;
624 case BLENDEQUATION_ADVANCED_COLORDODGE: SAMPLE_REGISTER_ADV_BLEND(colordodge); break;
625 case BLENDEQUATION_ADVANCED_COLORBURN: SAMPLE_REGISTER_ADV_BLEND(colorburn); break;
626 case BLENDEQUATION_ADVANCED_HARDLIGHT: SAMPLE_REGISTER_ADV_BLEND(hardlight); break;
627 case BLENDEQUATION_ADVANCED_SOFTLIGHT: SAMPLE_REGISTER_ADV_BLEND(softlight); break;
628 case BLENDEQUATION_ADVANCED_DIFFERENCE: SAMPLE_REGISTER_ADV_BLEND(difference); break;
629 case BLENDEQUATION_ADVANCED_EXCLUSION: SAMPLE_REGISTER_ADV_BLEND(exclusion); break;
630 case BLENDEQUATION_ADVANCED_HSL_HUE: SAMPLE_REGISTER_ADV_BLEND_HSL(setLumSat(srcColor, dstColor, dstColor)); break;
631 case BLENDEQUATION_ADVANCED_HSL_SATURATION: SAMPLE_REGISTER_ADV_BLEND_HSL(setLumSat(dstColor, srcColor, dstColor)); break;
632 case BLENDEQUATION_ADVANCED_HSL_COLOR: SAMPLE_REGISTER_ADV_BLEND_HSL(setLum(srcColor, dstColor)); break;
633 case BLENDEQUATION_ADVANCED_HSL_LUMINOSITY: SAMPLE_REGISTER_ADV_BLEND_HSL(setLum(dstColor, srcColor)); break;
634 default:
635 DE_ASSERT(false);
636 }
637
638 #undef SAMPLE_REGISTER_ADV_BLEND
639 #undef SAMPLE_REGISTER_ADV_BLEND_HSL
640 }
641
executeColorWrite(int fragNdxOffset,int numSamplesPerFragment,const Fragment * inputFragments,bool isSRGB,const tcu::PixelBufferAccess & colorBuffer)642 void FragmentProcessor::executeColorWrite (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, bool isSRGB, const tcu::PixelBufferAccess& colorBuffer)
643 {
644 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
645 {
646 if (m_sampleRegister[regSampleNdx].isAlive)
647 {
648 int fragSampleNdx = regSampleNdx % numSamplesPerFragment;
649 const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment];
650 Vec4 combinedColor;
651
652 combinedColor.xyz() = m_sampleRegister[regSampleNdx].blendedRGB;
653 combinedColor.w() = m_sampleRegister[regSampleNdx].blendedA;
654
655 if (isSRGB)
656 combinedColor = tcu::linearToSRGB(combinedColor);
657
658 colorBuffer.setPixel(combinedColor, fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y());
659 }
660 }
661 }
662
executeRGBA8ColorWrite(int fragNdxOffset,int numSamplesPerFragment,const Fragment * inputFragments,const tcu::PixelBufferAccess & colorBuffer)663 void FragmentProcessor::executeRGBA8ColorWrite (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const tcu::PixelBufferAccess& colorBuffer)
664 {
665 const int fragStride = 4;
666 const int xStride = colorBuffer.getRowPitch();
667 const int yStride = colorBuffer.getSlicePitch();
668 deUint8* const basePtr = (deUint8*)colorBuffer.getDataPtr();
669
670 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
671 {
672 if (m_sampleRegister[regSampleNdx].isAlive)
673 {
674 const int fragSampleNdx = regSampleNdx % numSamplesPerFragment;
675 const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment];
676 deUint8* dstPtr = basePtr + fragSampleNdx*fragStride + frag.pixelCoord.x()*xStride + frag.pixelCoord.y()*yStride;
677
678 dstPtr[0] = tcu::floatToU8(m_sampleRegister[regSampleNdx].blendedRGB.x());
679 dstPtr[1] = tcu::floatToU8(m_sampleRegister[regSampleNdx].blendedRGB.y());
680 dstPtr[2] = tcu::floatToU8(m_sampleRegister[regSampleNdx].blendedRGB.z());
681 dstPtr[3] = tcu::floatToU8(m_sampleRegister[regSampleNdx].blendedA);
682 }
683 }
684 }
685
executeMaskedColorWrite(int fragNdxOffset,int numSamplesPerFragment,const Fragment * inputFragments,const Vec4 & colorMaskFactor,const Vec4 & colorMaskNegationFactor,bool isSRGB,const tcu::PixelBufferAccess & colorBuffer)686 void FragmentProcessor::executeMaskedColorWrite (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const Vec4& colorMaskFactor, const Vec4& colorMaskNegationFactor, bool isSRGB, const tcu::PixelBufferAccess& colorBuffer)
687 {
688 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
689 {
690 if (m_sampleRegister[regSampleNdx].isAlive)
691 {
692 int fragSampleNdx = regSampleNdx % numSamplesPerFragment;
693 const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment];
694 Vec4 originalColor = colorBuffer.getPixel(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y());
695 Vec4 newColor;
696
697 newColor.xyz() = m_sampleRegister[regSampleNdx].blendedRGB;
698 newColor.w() = m_sampleRegister[regSampleNdx].blendedA;
699
700 if (isSRGB)
701 newColor = tcu::linearToSRGB(newColor);
702
703 newColor = colorMaskFactor*newColor + colorMaskNegationFactor*originalColor;
704
705 colorBuffer.setPixel(newColor, fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y());
706 }
707 }
708 }
709
executeSignedValueWrite(int fragNdxOffset,int numSamplesPerFragment,const Fragment * inputFragments,const tcu::BVec4 & colorMask,const tcu::PixelBufferAccess & colorBuffer)710 void FragmentProcessor::executeSignedValueWrite (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const tcu::BVec4& colorMask, const tcu::PixelBufferAccess& colorBuffer)
711 {
712 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
713 {
714 if (m_sampleRegister[regSampleNdx].isAlive)
715 {
716 int fragSampleNdx = regSampleNdx % numSamplesPerFragment;
717 const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment];
718 const IVec4 originalValue = colorBuffer.getPixelInt(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y());
719
720 colorBuffer.setPixel(tcu::select(m_sampleRegister[regSampleNdx].signedValue, originalValue, colorMask), fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y());
721 }
722 }
723 }
724
executeUnsignedValueWrite(int fragNdxOffset,int numSamplesPerFragment,const Fragment * inputFragments,const tcu::BVec4 & colorMask,const tcu::PixelBufferAccess & colorBuffer)725 void FragmentProcessor::executeUnsignedValueWrite (int fragNdxOffset, int numSamplesPerFragment, const Fragment* inputFragments, const tcu::BVec4& colorMask, const tcu::PixelBufferAccess& colorBuffer)
726 {
727 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
728 {
729 if (m_sampleRegister[regSampleNdx].isAlive)
730 {
731 int fragSampleNdx = regSampleNdx % numSamplesPerFragment;
732 const Fragment& frag = inputFragments[fragNdxOffset + regSampleNdx/numSamplesPerFragment];
733 const UVec4 originalValue = colorBuffer.getPixelUint(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y());
734
735 colorBuffer.setPixel(tcu::select(m_sampleRegister[regSampleNdx].unsignedValue, originalValue, colorMask), fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y());
736 }
737 }
738 }
739
render(const rr::MultisamplePixelBufferAccess & msColorBuffer,const rr::MultisamplePixelBufferAccess & msDepthBuffer,const rr::MultisamplePixelBufferAccess & msStencilBuffer,const Fragment * inputFragments,int numFragments,FaceType fragmentFacing,const FragmentOperationState & state)740 void FragmentProcessor::render (const rr::MultisamplePixelBufferAccess& msColorBuffer,
741 const rr::MultisamplePixelBufferAccess& msDepthBuffer,
742 const rr::MultisamplePixelBufferAccess& msStencilBuffer,
743 const Fragment* inputFragments,
744 int numFragments,
745 FaceType fragmentFacing,
746 const FragmentOperationState& state)
747 {
748 DE_ASSERT(fragmentFacing < FACETYPE_LAST);
749 DE_ASSERT(state.numStencilBits < 32); // code bitshifts numStencilBits, avoid undefined behavior
750
751 const tcu::PixelBufferAccess& colorBuffer = msColorBuffer.raw();
752 const tcu::PixelBufferAccess& depthBuffer = msDepthBuffer.raw();
753 const tcu::PixelBufferAccess& stencilBuffer = msStencilBuffer.raw();
754
755 bool hasDepth = depthBuffer.getWidth() > 0 && depthBuffer.getHeight() > 0 && depthBuffer.getDepth() > 0;
756 bool hasStencil = stencilBuffer.getWidth() > 0 && stencilBuffer.getHeight() > 0 && stencilBuffer.getDepth() > 0;
757 bool doDepthTest = hasDepth && state.depthTestEnabled;
758 bool doStencilTest = hasStencil && state.stencilTestEnabled;
759
760 tcu::TextureChannelClass colorbufferClass = tcu::getTextureChannelClass(msColorBuffer.raw().getFormat().type);
761 rr::GenericVecType fragmentDataType = (colorbufferClass == tcu::TEXTURECHANNELCLASS_SIGNED_INTEGER) ? (rr::GENERICVECTYPE_INT32) : ((colorbufferClass == tcu::TEXTURECHANNELCLASS_UNSIGNED_INTEGER) ? (rr::GENERICVECTYPE_UINT32) : (rr::GENERICVECTYPE_FLOAT));
762
763 DE_ASSERT((!hasDepth || colorBuffer.getWidth() == depthBuffer.getWidth()) && (!hasStencil || colorBuffer.getWidth() == stencilBuffer.getWidth()));
764 DE_ASSERT((!hasDepth || colorBuffer.getHeight() == depthBuffer.getHeight()) && (!hasStencil || colorBuffer.getHeight() == stencilBuffer.getHeight()));
765 DE_ASSERT((!hasDepth || colorBuffer.getDepth() == depthBuffer.getDepth()) && (!hasStencil || colorBuffer.getDepth() == stencilBuffer.getDepth()));
766
767 // Combined formats must be separated beforehand
768 DE_ASSERT(!hasDepth || (!tcu::isCombinedDepthStencilType(depthBuffer.getFormat().type) && depthBuffer.getFormat().order == tcu::TextureFormat::D));
769 DE_ASSERT(!hasStencil || (!tcu::isCombinedDepthStencilType(stencilBuffer.getFormat().type) && stencilBuffer.getFormat().order == tcu::TextureFormat::S));
770
771 int numSamplesPerFragment = colorBuffer.getWidth();
772 int totalNumSamples = numFragments*numSamplesPerFragment;
773 int numSampleGroups = (totalNumSamples - 1) / SAMPLE_REGISTER_SIZE + 1; // \note totalNumSamples/SAMPLE_REGISTER_SIZE rounded up.
774 const StencilState& stencilState = state.stencilStates[fragmentFacing];
775 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);
776 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);
777 bool sRGBTarget = state.sRGBEnabled && tcu::isSRGB(colorBuffer.getFormat());
778
779 DE_ASSERT(SAMPLE_REGISTER_SIZE % numSamplesPerFragment == 0);
780
781 // Divide the fragments' samples into groups of size SAMPLE_REGISTER_SIZE, and perform
782 // the per-sample operations for one group at a time.
783
784 for (int sampleGroupNdx = 0; sampleGroupNdx < numSampleGroups; sampleGroupNdx++)
785 {
786 // The index of the fragment of the sample at the beginning of m_sampleRegisters.
787 int groupFirstFragNdx = (sampleGroupNdx*SAMPLE_REGISTER_SIZE) / numSamplesPerFragment;
788
789 // Initialize sample data in the sample register.
790
791 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
792 {
793 int fragNdx = groupFirstFragNdx + regSampleNdx/numSamplesPerFragment;
794 int fragSampleNdx = regSampleNdx % numSamplesPerFragment;
795
796 if (fragNdx < numFragments)
797 {
798 m_sampleRegister[regSampleNdx].isAlive = (inputFragments[fragNdx].coverage & (1u << fragSampleNdx)) != 0;
799 m_sampleRegister[regSampleNdx].depthPassed = true; // \note This will stay true if depth test is disabled.
800 }
801 else
802 m_sampleRegister[regSampleNdx].isAlive = false;
803 }
804
805 // Scissor test.
806
807 if (state.scissorTestEnabled)
808 executeScissorTest(groupFirstFragNdx, numSamplesPerFragment, inputFragments, state.scissorRectangle);
809
810 // Stencil test.
811
812 if (doStencilTest)
813 {
814 executeStencilCompare(groupFirstFragNdx, numSamplesPerFragment, inputFragments, stencilState, state.numStencilBits, stencilBuffer);
815 executeStencilSFail(groupFirstFragNdx, numSamplesPerFragment, inputFragments, stencilState, state.numStencilBits, stencilBuffer);
816 }
817
818 // Depth test.
819 // \note Current value of isAlive is needed for dpPass and dpFail, so it's only updated after them and not right after depth test.
820
821 if (doDepthTest)
822 {
823 executeDepthCompare(groupFirstFragNdx, numSamplesPerFragment, inputFragments, state.depthFunc, depthBuffer);
824
825 if (state.depthMask)
826 executeDepthWrite(groupFirstFragNdx, numSamplesPerFragment, inputFragments, depthBuffer);
827 }
828
829 // Do dpFail and dpPass stencil writes.
830
831 if (doStencilTest)
832 executeStencilDpFailAndPass(groupFirstFragNdx, numSamplesPerFragment, inputFragments, stencilState, state.numStencilBits, stencilBuffer);
833
834 // Kill the samples that failed depth test.
835
836 if (doDepthTest)
837 {
838 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
839 m_sampleRegister[regSampleNdx].isAlive = m_sampleRegister[regSampleNdx].isAlive && m_sampleRegister[regSampleNdx].depthPassed;
840 }
841
842 // Paint fragments to target
843
844 switch (fragmentDataType)
845 {
846 case rr::GENERICVECTYPE_FLOAT:
847 {
848 // Select min/max clamping values for blending factors and operands
849 Vec4 minClampValue;
850 Vec4 maxClampValue;
851
852 if (colorbufferClass == tcu::TEXTURECHANNELCLASS_UNSIGNED_FIXED_POINT)
853 {
854 minClampValue = Vec4(0.0f);
855 maxClampValue = Vec4(1.0f);
856 }
857 else if (colorbufferClass == tcu::TEXTURECHANNELCLASS_SIGNED_FIXED_POINT)
858 {
859 minClampValue = Vec4(-1.0f);
860 maxClampValue = Vec4(1.0f);
861 }
862 else
863 {
864 // No clamping
865 minClampValue = Vec4(-std::numeric_limits<float>::infinity());
866 maxClampValue = Vec4(std::numeric_limits<float>::infinity());
867 }
868
869 // Blend calculation - only if using blend.
870 if (state.blendMode == BLENDMODE_STANDARD)
871 {
872 // Put dst color to register, doing srgb-to-linear conversion if needed.
873 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
874 {
875 if (m_sampleRegister[regSampleNdx].isAlive)
876 {
877 int fragSampleNdx = regSampleNdx % numSamplesPerFragment;
878 const Fragment& frag = inputFragments[groupFirstFragNdx + regSampleNdx/numSamplesPerFragment];
879 Vec4 dstColor = colorBuffer.getPixel(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y());
880
881 m_sampleRegister[regSampleNdx].clampedBlendSrcColor = clamp(frag.value.get<float>(), minClampValue, maxClampValue);
882 m_sampleRegister[regSampleNdx].clampedBlendSrc1Color = clamp(frag.value1.get<float>(), minClampValue, maxClampValue);
883 m_sampleRegister[regSampleNdx].clampedBlendDstColor = clamp(sRGBTarget ? tcu::sRGBToLinear(dstColor) : dstColor, minClampValue, maxClampValue);
884 }
885 }
886
887 // Calculate blend factors to register.
888 executeBlendFactorComputeRGB(state.blendColor, state.blendRGBState);
889 executeBlendFactorComputeA(state.blendColor, state.blendAState);
890
891 // Compute blended color.
892 executeBlend(state.blendRGBState, state.blendAState);
893 }
894 else if (state.blendMode == BLENDMODE_ADVANCED)
895 {
896 // Unpremultiply colors for blending, and do sRGB->linear if necessary
897 // \todo [2014-03-17 pyry] Re-consider clampedBlend*Color var names
898 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
899 {
900 if (m_sampleRegister[regSampleNdx].isAlive)
901 {
902 int fragSampleNdx = regSampleNdx % numSamplesPerFragment;
903 const Fragment& frag = inputFragments[groupFirstFragNdx + regSampleNdx/numSamplesPerFragment];
904 const Vec4 srcColor = frag.value.get<float>();
905 const Vec4 dstColor = colorBuffer.getPixel(fragSampleNdx, frag.pixelCoord.x(), frag.pixelCoord.y());
906
907 m_sampleRegister[regSampleNdx].clampedBlendSrcColor = unpremultiply(clamp(srcColor, minClampValue, maxClampValue));
908 m_sampleRegister[regSampleNdx].clampedBlendDstColor = unpremultiply(clamp(sRGBTarget ? tcu::sRGBToLinear(dstColor) : dstColor, minClampValue, maxClampValue));
909 }
910 }
911
912 executeAdvancedBlend(state.blendEquationAdvaced);
913 }
914 else
915 {
916 // Not using blend - just put values to register as-is.
917 DE_ASSERT(state.blendMode == BLENDMODE_NONE);
918
919 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
920 {
921 if (m_sampleRegister[regSampleNdx].isAlive)
922 {
923 const Fragment& frag = inputFragments[groupFirstFragNdx + regSampleNdx/numSamplesPerFragment];
924
925 m_sampleRegister[regSampleNdx].blendedRGB = frag.value.get<float>().xyz();
926 m_sampleRegister[regSampleNdx].blendedA = frag.value.get<float>().w();
927 }
928 }
929 }
930
931 // Clamp result values in sample register
932 if (colorbufferClass != tcu::TEXTURECHANNELCLASS_FLOATING_POINT)
933 {
934 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
935 {
936 if (m_sampleRegister[regSampleNdx].isAlive)
937 {
938 m_sampleRegister[regSampleNdx].blendedRGB = clamp(m_sampleRegister[regSampleNdx].blendedRGB, minClampValue.swizzle(0, 1, 2), maxClampValue.swizzle(0, 1, 2));
939 m_sampleRegister[regSampleNdx].blendedA = clamp(m_sampleRegister[regSampleNdx].blendedA, minClampValue.w(), maxClampValue.w());
940 }
941 }
942 }
943
944 // Finally, write the colors to the color buffer.
945
946 if (state.colorMask[0] && state.colorMask[1] && state.colorMask[2] && state.colorMask[3])
947 {
948 if (colorBuffer.getFormat() == tcu::TextureFormat(tcu::TextureFormat::RGBA, tcu::TextureFormat::UNORM_INT8))
949 executeRGBA8ColorWrite(groupFirstFragNdx, numSamplesPerFragment, inputFragments, colorBuffer);
950 else
951 executeColorWrite(groupFirstFragNdx, numSamplesPerFragment, inputFragments, sRGBTarget, colorBuffer);
952 }
953 else if (state.colorMask[0] || state.colorMask[1] || state.colorMask[2] || state.colorMask[3])
954 executeMaskedColorWrite(groupFirstFragNdx, numSamplesPerFragment, inputFragments, colorMaskFactor, colorMaskNegationFactor, sRGBTarget, colorBuffer);
955 break;
956 }
957 case rr::GENERICVECTYPE_INT32:
958 // Write fragments
959 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
960 {
961 if (m_sampleRegister[regSampleNdx].isAlive)
962 {
963 const Fragment& frag = inputFragments[groupFirstFragNdx + regSampleNdx/numSamplesPerFragment];
964
965 m_sampleRegister[regSampleNdx].signedValue = frag.value.get<deInt32>();
966 }
967 }
968
969 if (state.colorMask[0] || state.colorMask[1] || state.colorMask[2] || state.colorMask[3])
970 executeSignedValueWrite(groupFirstFragNdx, numSamplesPerFragment, inputFragments, state.colorMask, colorBuffer);
971 break;
972
973 case rr::GENERICVECTYPE_UINT32:
974 // Write fragments
975 for (int regSampleNdx = 0; regSampleNdx < SAMPLE_REGISTER_SIZE; regSampleNdx++)
976 {
977 if (m_sampleRegister[regSampleNdx].isAlive)
978 {
979 const Fragment& frag = inputFragments[groupFirstFragNdx + regSampleNdx/numSamplesPerFragment];
980
981 m_sampleRegister[regSampleNdx].unsignedValue = frag.value.get<deUint32>();
982 }
983 }
984
985 if (state.colorMask[0] || state.colorMask[1] || state.colorMask[2] || state.colorMask[3])
986 executeUnsignedValueWrite(groupFirstFragNdx, numSamplesPerFragment, inputFragments, state.colorMask, colorBuffer);
987 break;
988
989 default:
990 DE_ASSERT(DE_FALSE);
991 }
992 }
993 }
994
995 } // rr
996