Home | History | Annotate | Download | only in gpu 
      1 /*
      2  * Copyright 2016 Google Inc.
      3  *
      4  * Use of this source code is governed by a BSD-style license that can be
      5  * found in the LICENSE file.
      6  */
      7 #include "GrProgramDesc.h"
      8 #include "GrPipeline.h"
      9 #include "GrPrimitiveProcessor.h"
     10 #include "GrProcessor.h"
     11 #include "GrRenderTargetPriv.h"
     12 #include "GrShaderCaps.h"
     13 #include "GrTexturePriv.h"
     14 #include "SkChecksum.h"
     15 #include "glsl/GrGLSLFragmentProcessor.h"
     16 #include "glsl/GrGLSLFragmentShaderBuilder.h"
     17 
     18 enum {
     19     kSamplerOrImageTypeKeyBits = 4
     20 };
     21 
     22 static inline uint16_t image_storage_or_sampler_uniform_type_key(GrSLType type ) {
     23     int value = UINT16_MAX;
     24     switch (type) {
     25         case kTexture2DSampler_GrSLType:
     26             value = 0;
     27             break;
     28         case kITexture2DSampler_GrSLType:
     29             value = 1;
     30             break;
     31         case kTextureExternalSampler_GrSLType:
     32             value = 2;
     33             break;
     34         case kTexture2DRectSampler_GrSLType:
     35             value = 3;
     36             break;
     37         case kBufferSampler_GrSLType:
     38             value = 4;
     39             break;
     40 
     41         default:
     42             break;
     43     }
     44     SkASSERT((value & ((1 << kSamplerOrImageTypeKeyBits) - 1)) == value);
     45     return value;
     46 }
     47 
     48 static uint16_t sampler_key(GrSLType samplerType, GrPixelConfig config, GrShaderFlags visibility,
     49                             const GrShaderCaps& caps) {
     50     int samplerTypeKey = image_storage_or_sampler_uniform_type_key(samplerType);
     51 
     52     GR_STATIC_ASSERT(1 == sizeof(caps.configTextureSwizzle(config).asKey()));
     53     return SkToU16(samplerTypeKey |
     54                    caps.configTextureSwizzle(config).asKey() << kSamplerOrImageTypeKeyBits |
     55                    (GrSLSamplerPrecision(config) << (8 + kSamplerOrImageTypeKeyBits)));
     56 }
     57 
     58 static void add_sampler_and_image_keys(GrProcessorKeyBuilder* b, const GrResourceIOProcessor& proc,
     59                                        const GrShaderCaps& caps) {
     60     int numTextureSamplers = proc.numTextureSamplers();
     61     int numBuffers = proc.numBuffers();
     62     int numUniforms = numTextureSamplers + numBuffers;
     63     // Need two bytes per key.
     64     int word32Count = (numUniforms + 1) / 2;
     65     if (0 == word32Count) {
     66         return;
     67     }
     68     uint16_t* k16 = SkTCast<uint16_t*>(b->add32n(word32Count));
     69     int j = 0;
     70     for (int i = 0; i < numTextureSamplers; ++i, ++j) {
     71         const GrResourceIOProcessor::TextureSampler& sampler = proc.textureSampler(i);
     72         const GrTexture* tex = sampler.peekTexture();
     73 
     74         k16[j] = sampler_key(tex->texturePriv().samplerType(), tex->config(), sampler.visibility(),
     75                              caps);
     76     }
     77     for (int i = 0; i < numBuffers; ++i, ++j) {
     78         const GrResourceIOProcessor::BufferAccess& access = proc.bufferAccess(i);
     79         k16[j] = sampler_key(kBufferSampler_GrSLType, access.texelConfig(), access.visibility(),
     80                              caps);
     81     }
     82     // zero the last 16 bits if the number of uniforms for samplers and image storages is odd.
     83     if (numUniforms & 0x1) {
     84         k16[numUniforms] = 0;
     85     }
     86 }
     87 
     88 /**
     89  * A function which emits a meta key into the key builder.  This is required because shader code may
     90  * be dependent on properties of the effect that the effect itself doesn't use
     91  * in its key (e.g. the pixel format of textures used). So we create a meta-key for
     92  * every effect using this function. It is also responsible for inserting the effect's class ID
     93  * which must be different for every GrProcessor subclass. It can fail if an effect uses too many
     94  * transforms, etc, for the space allotted in the meta-key.  NOTE, both FPs and GPs share this
     95  * function because it is hairy, though FPs do not have attribs, and GPs do not have transforms
     96  */
     97 static bool gen_meta_key(const GrResourceIOProcessor& proc,
     98                          const GrShaderCaps& shaderCaps,
     99                          uint32_t transformKey,
    100                          GrProcessorKeyBuilder* b) {
    101     size_t processorKeySize = b->size();
    102     uint32_t classID = proc.classID();
    103 
    104     // Currently we allow 16 bits for the class id and the overall processor key size.
    105     static const uint32_t kMetaKeyInvalidMask = ~((uint32_t)SK_MaxU16);
    106     if ((processorKeySize | classID) & kMetaKeyInvalidMask) {
    107         return false;
    108     }
    109 
    110     add_sampler_and_image_keys(b, proc, shaderCaps);
    111 
    112     uint32_t* key = b->add32n(2);
    113     key[0] = (classID << 16) | SkToU32(processorKeySize);
    114     key[1] = transformKey;
    115     return true;
    116 }
    117 
    118 static bool gen_meta_key(const GrXferProcessor& xp,
    119                          const GrShaderCaps& shaderCaps,
    120                          GrProcessorKeyBuilder* b) {
    121     size_t processorKeySize = b->size();
    122     uint32_t classID = xp.classID();
    123 
    124     // Currently we allow 16 bits for the class id and the overall processor key size.
    125     static const uint32_t kMetaKeyInvalidMask = ~((uint32_t)SK_MaxU16);
    126     if ((processorKeySize | classID) & kMetaKeyInvalidMask) {
    127         return false;
    128     }
    129 
    130     b->add32((classID << 16) | SkToU32(processorKeySize));
    131     return true;
    132 }
    133 
    134 static bool gen_frag_proc_and_meta_keys(const GrPrimitiveProcessor& primProc,
    135                                         const GrFragmentProcessor& fp,
    136                                         const GrShaderCaps& shaderCaps,
    137                                         GrProcessorKeyBuilder* b) {
    138     for (int i = 0; i < fp.numChildProcessors(); ++i) {
    139         if (!gen_frag_proc_and_meta_keys(primProc, fp.childProcessor(i), shaderCaps, b)) {
    140             return false;
    141         }
    142     }
    143 
    144     fp.getGLSLProcessorKey(shaderCaps, b);
    145 
    146     return gen_meta_key(fp, shaderCaps, primProc.getTransformKey(fp.coordTransforms(),
    147                                                                  fp.numCoordTransforms()), b);
    148 }
    149 
    150 bool GrProgramDesc::Build(GrProgramDesc* desc,
    151                           const GrPrimitiveProcessor& primProc,
    152                           bool hasPointSize,
    153                           const GrPipeline& pipeline,
    154                           const GrShaderCaps& shaderCaps) {
    155     // The descriptor is used as a cache key. Thus when a field of the
    156     // descriptor will not affect program generation (because of the attribute
    157     // bindings in use or other descriptor field settings) it should be set
    158     // to a canonical value to avoid duplicate programs with different keys.
    159 
    160     GR_STATIC_ASSERT(0 == kProcessorKeysOffset % sizeof(uint32_t));
    161     // Make room for everything up to the effect keys.
    162     desc->key().reset();
    163     desc->key().push_back_n(kProcessorKeysOffset);
    164 
    165     GrProcessorKeyBuilder b(&desc->key());
    166 
    167     primProc.getGLSLProcessorKey(shaderCaps, &b);
    168     if (!gen_meta_key(primProc, shaderCaps, 0, &b)) {
    169         desc->key().reset();
    170         return false;
    171     }
    172     GrProcessor::RequiredFeatures requiredFeatures = primProc.requiredFeatures();
    173 
    174     for (int i = 0; i < pipeline.numFragmentProcessors(); ++i) {
    175         const GrFragmentProcessor& fp = pipeline.getFragmentProcessor(i);
    176         if (!gen_frag_proc_and_meta_keys(primProc, fp, shaderCaps, &b)) {
    177             desc->key().reset();
    178             return false;
    179         }
    180         requiredFeatures |= fp.requiredFeatures();
    181     }
    182 
    183     const GrXferProcessor& xp = pipeline.getXferProcessor();
    184     const GrSurfaceOrigin* originIfDstTexture = nullptr;
    185     GrSurfaceOrigin origin;
    186     if (pipeline.dstTextureProxy()) {
    187         origin = pipeline.dstTextureProxy()->origin();
    188         originIfDstTexture = &origin;
    189     }
    190     xp.getGLSLProcessorKey(shaderCaps, &b, originIfDstTexture);
    191     if (!gen_meta_key(xp, shaderCaps, &b)) {
    192         desc->key().reset();
    193         return false;
    194     }
    195     requiredFeatures |= xp.requiredFeatures();
    196 
    197     // --------DO NOT MOVE HEADER ABOVE THIS LINE--------------------------------------------------
    198     // Because header is a pointer into the dynamic array, we can't push any new data into the key
    199     // below here.
    200     KeyHeader* header = desc->atOffset<KeyHeader, kHeaderOffset>();
    201 
    202     // make sure any padding in the header is zeroed.
    203     memset(header, 0, kHeaderSize);
    204 
    205     GrRenderTargetProxy* proxy = pipeline.proxy();
    206 
    207     if (requiredFeatures & GrProcessor::kSampleLocations_RequiredFeature) {
    208         SkASSERT(pipeline.isHWAntialiasState());
    209 
    210         GrRenderTarget* rt = pipeline.renderTarget();
    211         header->fSamplePatternKey =
    212             rt->renderTargetPriv().getMultisampleSpecs(pipeline).fUniqueID;
    213     } else {
    214         header->fSamplePatternKey = 0;
    215     }
    216 
    217     header->fOutputSwizzle = shaderCaps.configOutputSwizzle(proxy->config()).asKey();
    218 
    219     header->fSnapVerticesToPixelCenters = pipeline.snapVerticesToPixelCenters();
    220     header->fColorFragmentProcessorCnt = pipeline.numColorFragmentProcessors();
    221     header->fCoverageFragmentProcessorCnt = pipeline.numCoverageFragmentProcessors();
    222     // Fail if the client requested more processors than the key can fit.
    223     if (header->fColorFragmentProcessorCnt != pipeline.numColorFragmentProcessors() ||
    224         header->fCoverageFragmentProcessorCnt != pipeline.numCoverageFragmentProcessors()) {
    225         return false;
    226     }
    227     header->fHasPointSize = hasPointSize ? 1 : 0;
    228     return true;
    229 }
    230