src/gallium/drivers/swr/rasterizer/core/backend_sample.cpp

   1 /****************************************************************************
   2 * Copyright (C) 2014-2015 Intel Corporation.   All Rights Reserved.
   3 *
   4 * Permission is hereby granted, free of charge, to any person obtaining a
   5 * copy of this software and associated documentation files (the "Software"),
   6 * to deal in the Software without restriction, including without limitation
   7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8 * and/or sell copies of the Software, and to permit persons to whom the
   9 * Software is furnished to do so, subject to the following conditions:
  10 *
  11 * The above copyright notice and this permission notice (including the next
  12 * paragraph) shall be included in all copies or substantial portions of the
  13 * Software.
  14 *
  15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21 * IN THE SOFTWARE.
  22 *
  23 * @file backend.cpp
  24 *
  25 * @brief Backend handles rasterization, pixel shading and output merger
  26 *        operations.
  27 *
  28 ******************************************************************************/
  29
  30 #include <smmintrin.h>
  31
  32 #include "backend.h"
  33 #include "backend_impl.h"
  34 #include "tilemgr.h"
  35 #include "memory/tilingtraits.h"
  36 #include "core/multisample.h"
  37
  38 #include <algorithm>
  39
  40 template<typename T>
  41 void BackendSampleRate(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t x, uint32_t y, SWR_TRIANGLE_DESC &work, RenderOutputBuffers &renderBuffers)
  42 {
  43     SWR_CONTEXT *pContext = pDC->pContext;
  44
  45     AR_BEGIN(BESampleRateBackend, pDC->drawId);
  46     AR_BEGIN(BESetup, pDC->drawId);
  47
  48     const API_STATE &state = GetApiState(pDC);
  49
  50     BarycentricCoeffs coeffs;
  51     SetupBarycentricCoeffs(&coeffs, work);
  52
  53     uint8_t *pColorBuffer[SWR_NUM_RENDERTARGETS], *pDepthBuffer, *pStencilBuffer;
  54     SetupRenderBuffers(pColorBuffer, &pDepthBuffer, &pStencilBuffer, state.psState.numRenderTargets, renderBuffers);
  55
  56     SWR_PS_CONTEXT psContext;
  57     const SWR_MULTISAMPLE_POS& samplePos = state.rastState.samplePositions;
  58     SetupPixelShaderContext<T>(&psContext, samplePos, work);
  59
  60     AR_END(BESetup, 0);
  61
  62     psContext.vY.UL = _simd_add_ps(vULOffsetsY, _simd_set1_ps(static_cast<float>(y)));
  63     psContext.vY.center = _simd_add_ps(vCenterOffsetsY, _simd_set1_ps(static_cast<float>(y)));
  64
  65     const simdscalar dy = _simd_set1_ps(static_cast<float>(SIMD_TILE_Y_DIM));
  66
  67     for (uint32_t yy = y; yy < y + KNOB_TILE_Y_DIM; yy += SIMD_TILE_Y_DIM)
  68     {
  69         psContext.vX.UL = _simd_add_ps(vULOffsetsX, _simd_set1_ps(static_cast<float>(x)));
  70         psContext.vX.center = _simd_add_ps(vCenterOffsetsX, _simd_set1_ps(static_cast<float>(x)));
  71
  72         const simdscalar dx = _simd_set1_ps(static_cast<float>(SIMD_TILE_X_DIM));
  73
  74         for (uint32_t xx = x; xx < x + KNOB_TILE_X_DIM; xx += SIMD_TILE_X_DIM)
  75         {
  76 #if USE_8x2_TILE_BACKEND
  77             const bool useAlternateOffset = ((xx & SIMD_TILE_X_DIM) != 0);
  78
  79 #endif
  80             if (T::InputCoverage != SWR_INPUT_COVERAGE_NONE)
  81             {
  82                 const uint64_t* pCoverageMask = (T::InputCoverage == SWR_INPUT_COVERAGE_INNER_CONSERVATIVE) ? &work.innerCoverageMask : &work.coverageMask[0];
  83
  84                 generateInputCoverage<T, T::InputCoverage>(pCoverageMask, psContext.inputMask, state.blendState.sampleMask);
  85             }
  86
  87             AR_BEGIN(BEBarycentric, pDC->drawId);
  88
  89             CalcPixelBarycentrics(coeffs, psContext);
  90
  91             CalcCentroid<T, false>(&psContext, samplePos, coeffs, work.coverageMask, state.blendState.sampleMask);
  92
  93             AR_END(BEBarycentric, 0);
  94
  95             for (uint32_t sample = 0; sample < T::MultisampleT::numSamples; sample++)
  96             {
  97                 simdmask coverageMask = work.coverageMask[sample] & MASK;
  98
  99                 if (coverageMask)
 100                 {
 101                     // offset depth/stencil buffers current sample
 102                     uint8_t *pDepthSample = pDepthBuffer + RasterTileDepthOffset(sample);
 103                     uint8_t *pStencilSample = pStencilBuffer + RasterTileStencilOffset(sample);
 104
 105                     if (state.depthHottileEnable && state.depthBoundsState.depthBoundsTestEnable)
 106                     {
 107                         static_assert(KNOB_DEPTH_HOT_TILE_FORMAT == R32_FLOAT, "Unsupported depth hot tile format");
 108
 109                         const simdscalar z = _simd_load_ps(reinterpret_cast<const float *>(pDepthSample));
 110
 111                         const float minz = state.depthBoundsState.depthBoundsTestMinValue;
 112                         const float maxz = state.depthBoundsState.depthBoundsTestMaxValue;
 113
 114                         coverageMask &= CalcDepthBoundsAcceptMask(z, minz, maxz);
 115                     }
 116
 117                     AR_BEGIN(BEBarycentric, pDC->drawId);
 118
 119                     // calculate per sample positions
 120                     psContext.vX.sample = _simd_add_ps(psContext.vX.UL, samplePos.vX(sample));
 121                     psContext.vY.sample = _simd_add_ps(psContext.vY.UL, samplePos.vY(sample));
 122
 123                     CalcSampleBarycentrics(coeffs, psContext);
 124
 125                     // interpolate and quantize z
 126                     psContext.vZ = vplaneps(coeffs.vZa, coeffs.vZb, coeffs.vZc, psContext.vI.sample, psContext.vJ.sample);
 127                     psContext.vZ = state.pfnQuantizeDepth(psContext.vZ);
 128
 129                     AR_END(BEBarycentric, 0);
 130
 131                     // interpolate user clip distance if available
 132                     if (state.rastState.clipDistanceMask)
 133                     {
 134                         coverageMask &= ~ComputeUserClipMask(state.rastState.clipDistanceMask, work.pUserClipBuffer, psContext.vI.sample, psContext.vJ.sample);
 135                     }
 136
 137                     simdscalar vCoverageMask = vMask(coverageMask);
 138                     simdscalar depthPassMask = vCoverageMask;
 139                     simdscalar stencilPassMask = vCoverageMask;
 140
 141                     // Early-Z?
 142                     if (T::bCanEarlyZ)
 143                     {
 144                         AR_BEGIN(BEEarlyDepthTest, pDC->drawId);
 145                         depthPassMask = DepthStencilTest(&state, work.triFlags.frontFacing, work.triFlags.viewportIndex,
 146                             psContext.vZ, pDepthSample, vCoverageMask, pStencilSample, &stencilPassMask);
 147                         AR_EVENT(EarlyDepthStencilInfoSampleRate(_simd_movemask_ps(depthPassMask), _simd_movemask_ps(stencilPassMask), _simd_movemask_ps(vCoverageMask)));
 148                         AR_END(BEEarlyDepthTest, 0);
 149
 150                         // early-exit if no samples passed depth or earlyZ is forced on.
 151                         if (state.psState.forceEarlyZ || !_simd_movemask_ps(depthPassMask))
 152                         {
 153                             DepthStencilWrite(&state.vp[work.triFlags.viewportIndex], &state.depthStencilState, work.triFlags.frontFacing, psContext.vZ,
 154                                 pDepthSample, depthPassMask, vCoverageMask, pStencilSample, stencilPassMask);
 155
 156                             if (!_simd_movemask_ps(depthPassMask))
 157                             {
 158                                 work.coverageMask[sample] >>= (SIMD_TILE_Y_DIM * SIMD_TILE_X_DIM);
 159                                 continue;
 160                             }
 161                         }
 162                     }
 163
 164                     psContext.sampleIndex = sample;
 165                     psContext.activeMask = _simd_castps_si(vCoverageMask);
 166
 167                     // execute pixel shader
 168                     AR_BEGIN(BEPixelShader, pDC->drawId);
 169                     UPDATE_STAT_BE(PsInvocations, _mm_popcnt_u32(_simd_movemask_ps(vCoverageMask)));
 170                     state.psState.pfnPixelShader(GetPrivateState(pDC), &psContext);
 171                     AR_END(BEPixelShader, 0);
 172
 173                     vCoverageMask = _simd_castsi_ps(psContext.activeMask);
 174
 175                     // late-Z
 176                     if (!T::bCanEarlyZ)
 177                     {
 178                         AR_BEGIN(BELateDepthTest, pDC->drawId);
 179                         depthPassMask = DepthStencilTest(&state, work.triFlags.frontFacing, work.triFlags.viewportIndex,
 180                             psContext.vZ, pDepthSample, vCoverageMask, pStencilSample, &stencilPassMask);
 181                         AR_EVENT(LateDepthStencilInfoSampleRate(_simd_movemask_ps(depthPassMask), _simd_movemask_ps(stencilPassMask), _simd_movemask_ps(vCoverageMask)));
 182                         AR_END(BELateDepthTest, 0);
 183
 184                         if (!_simd_movemask_ps(depthPassMask))
 185                         {
 186                             // need to call depth/stencil write for stencil write
 187                             DepthStencilWrite(&state.vp[work.triFlags.viewportIndex], &state.depthStencilState, work.triFlags.frontFacing, psContext.vZ,
 188                                 pDepthSample, depthPassMask, vCoverageMask, pStencilSample, stencilPassMask);
 189
 190                             work.coverageMask[sample] >>= (SIMD_TILE_Y_DIM * SIMD_TILE_X_DIM);
 191                             continue;
 192                         }
 193                     }
 194
 195                     uint32_t statMask = _simd_movemask_ps(depthPassMask);
 196                     uint32_t statCount = _mm_popcnt_u32(statMask);
 197                     UPDATE_STAT_BE(DepthPassCount, statCount);
 198
 199                     // output merger
 200                     AR_BEGIN(BEOutputMerger, pDC->drawId);
 201 #if USE_8x2_TILE_BACKEND
 202                     OutputMerger8x2(psContext, pColorBuffer, sample, &state.blendState, state.pfnBlendFunc, vCoverageMask, depthPassMask, state.psState.numRenderTargets, state.colorHottileEnable, useAlternateOffset);
 203 #else
 204                     OutputMerger4x2(psContext, pColorBuffer, sample, &state.blendState, state.pfnBlendFunc, vCoverageMask, depthPassMask, state.psState.numRenderTargets);
 205 #endif
 206
 207                     // do final depth write after all pixel kills
 208                     if (!state.psState.forceEarlyZ)
 209                     {
 210                         DepthStencilWrite(&state.vp[work.triFlags.viewportIndex], &state.depthStencilState, work.triFlags.frontFacing, psContext.vZ,
 211                             pDepthSample, depthPassMask, vCoverageMask, pStencilSample, stencilPassMask);
 212                     }
 213                     AR_END(BEOutputMerger, 0);
 214                 }
 215                 work.coverageMask[sample] >>= (SIMD_TILE_Y_DIM * SIMD_TILE_X_DIM);
 216             }
 217
 218         Endtile:
 219             ATTR_UNUSED;
 220
 221             AR_BEGIN(BEEndTile, pDC->drawId);
 222
 223             if (T::InputCoverage == SWR_INPUT_COVERAGE_INNER_CONSERVATIVE)
 224             {
 225                 work.innerCoverageMask >>= (SIMD_TILE_Y_DIM * SIMD_TILE_X_DIM);
 226             }
 227
 228 #if USE_8x2_TILE_BACKEND
 229             if (useAlternateOffset)
 230             {
 231                 for (uint32_t rt = 0; rt < state.psState.numRenderTargets; ++rt)
 232                 {
 233                     pColorBuffer[rt] += (2 * KNOB_SIMD_WIDTH * FormatTraits<KNOB_COLOR_HOT_TILE_FORMAT>::bpp) / 8;
 234                 }
 235             }
 236 #else
 237             for (uint32_t rt = 0; rt < state.psState.numRenderTargets; ++rt)
 238             {
 239                 pColorBuffer[rt] += (KNOB_SIMD_WIDTH * FormatTraits<KNOB_COLOR_HOT_TILE_FORMAT>::bpp) / 8;
 240             }
 241 #endif
 242             pDepthBuffer += (KNOB_SIMD_WIDTH * FormatTraits<KNOB_DEPTH_HOT_TILE_FORMAT>::bpp) / 8;
 243             pStencilBuffer += (KNOB_SIMD_WIDTH * FormatTraits<KNOB_STENCIL_HOT_TILE_FORMAT>::bpp) / 8;
 244
 245             AR_END(BEEndTile, 0);
 246
 247             psContext.vX.UL = _simd_add_ps(psContext.vX.UL, dx);
 248             psContext.vX.center = _simd_add_ps(psContext.vX.center, dx);
 249         }
 250
 251         psContext.vY.UL = _simd_add_ps(psContext.vY.UL, dy);
 252         psContext.vY.center = _simd_add_ps(psContext.vY.center, dy);
 253     }
 254
 255     AR_END(BESampleRateBackend, 0);
 256 }
 257
 258 // Recursive template used to auto-nest conditionals.  Converts dynamic enum function
 259 // arguments to static template arguments.
 260 template <uint32_t... ArgsT>
 261 struct BEChooserSampleRate
 262 {
 263     // Last Arg Terminator
 264     static PFN_BACKEND_FUNC GetFunc(SWR_BACKEND_FUNCS tArg)
 265     {
 266         switch (tArg)
 267         {
 268         case SWR_BACKEND_MSAA_SAMPLE_RATE: return BackendSampleRate<SwrBackendTraits<ArgsT...>>; break;
 269         case SWR_BACKEND_SINGLE_SAMPLE:
 270         case SWR_BACKEND_MSAA_PIXEL_RATE:
 271             SWR_ASSERT(0 && "Invalid backend func\n");
 272             return nullptr;
 273             break;
 274         default:
 275             SWR_ASSERT(0 && "Invalid backend func\n");
 276             return nullptr;
 277             break;
 278         }
 279     }
 280
 281     // Recursively parse args
 282     template <typename... TArgsT>
 283     static PFN_BACKEND_FUNC GetFunc(SWR_INPUT_COVERAGE tArg, TArgsT... remainingArgs)
 284     {
 285         switch (tArg)
 286         {
 287         case SWR_INPUT_COVERAGE_NONE: return BEChooserSampleRate<ArgsT..., SWR_INPUT_COVERAGE_NONE>::GetFunc(remainingArgs...); break;
 288         case SWR_INPUT_COVERAGE_NORMAL: return BEChooserSampleRate<ArgsT..., SWR_INPUT_COVERAGE_NORMAL>::GetFunc(remainingArgs...); break;
 289         case SWR_INPUT_COVERAGE_INNER_CONSERVATIVE: return BEChooserSampleRate<ArgsT..., SWR_INPUT_COVERAGE_INNER_CONSERVATIVE>::GetFunc(remainingArgs...); break;
 290         default:
 291             SWR_ASSERT(0 && "Invalid sample pattern\n");
 292             return BEChooserSampleRate<ArgsT..., SWR_INPUT_COVERAGE_NONE>::GetFunc(remainingArgs...);
 293             break;
 294         }
 295     }
 296
 297     // Recursively parse args
 298     template <typename... TArgsT>
 299     static PFN_BACKEND_FUNC GetFunc(SWR_MULTISAMPLE_COUNT tArg, TArgsT... remainingArgs)
 300     {
 301         switch (tArg)
 302         {
 303         case SWR_MULTISAMPLE_1X: return BEChooserSampleRate<ArgsT..., SWR_MULTISAMPLE_1X>::GetFunc(remainingArgs...); break;
 304         case SWR_MULTISAMPLE_2X: return BEChooserSampleRate<ArgsT..., SWR_MULTISAMPLE_2X>::GetFunc(remainingArgs...); break;
 305         case SWR_MULTISAMPLE_4X: return BEChooserSampleRate<ArgsT..., SWR_MULTISAMPLE_4X>::GetFunc(remainingArgs...); break;
 306         case SWR_MULTISAMPLE_8X: return BEChooserSampleRate<ArgsT..., SWR_MULTISAMPLE_8X>::GetFunc(remainingArgs...); break;
 307         case SWR_MULTISAMPLE_16X: return BEChooserSampleRate<ArgsT..., SWR_MULTISAMPLE_16X>::GetFunc(remainingArgs...); break;
 308         default:
 309             SWR_ASSERT(0 && "Invalid sample count\n");
 310             return BEChooserSampleRate<ArgsT..., SWR_MULTISAMPLE_1X>::GetFunc(remainingArgs...);
 311             break;
 312         }
 313     }
 314
 315     // Recursively parse args
 316     template <typename... TArgsT>
 317     static PFN_BACKEND_FUNC GetFunc(bool tArg, TArgsT... remainingArgs)
 318     {
 319         if (tArg == true)
 320         {
 321             return BEChooserSampleRate<ArgsT..., 1>::GetFunc(remainingArgs...);
 322         }
 323
 324         return BEChooserSampleRate<ArgsT..., 0>::GetFunc(remainingArgs...);
 325     }
 326 };
 327
 328 void InitBackendSampleFuncTable(PFN_BACKEND_FUNC(&table)[SWR_MULTISAMPLE_TYPE_COUNT][SWR_INPUT_COVERAGE_COUNT][2][2])
 329 {
 330     for (uint32_t sampleCount = SWR_MULTISAMPLE_1X; sampleCount < SWR_MULTISAMPLE_TYPE_COUNT; sampleCount++)
 331     {
 332         for (uint32_t inputCoverage = 0; inputCoverage < SWR_INPUT_COVERAGE_COUNT; inputCoverage++)
 333         {
 334             for (uint32_t centroid = 0; centroid < 2; centroid++)
 335             {
 336                 for (uint32_t canEarlyZ = 0; canEarlyZ < 2; canEarlyZ++)
 337                 {
 338                     table[sampleCount][inputCoverage][centroid][canEarlyZ] =
 339                         BEChooserSampleRate<>::GetFunc((SWR_MULTISAMPLE_COUNT)sampleCount, false, (SWR_INPUT_COVERAGE)inputCoverage,
 340                         (centroid > 0), false, (canEarlyZ > 0), (SWR_BACKEND_FUNCS)SWR_BACKEND_MSAA_SAMPLE_RATE);
 341                 }
 342             }
 343         }
 344     }
 345 }