src/gpu/gl/GrGLProgramDesc.cpp

/*
 * Copyright 2013 Google Inc.
 *
 * Use of this source code is governed by a BSD-style license that can be
 * found in the LICENSE file.
 */
#include "GrGLProgramDesc.h"

#include "GrGLProcessor.h"
#include "GrProcessor.h"
#include "GrGLGpu.h"
#include "GrOptDrawState.h"
#include "SkChecksum.h"
#include "gl/builders/GrGLFragmentShaderBuilder.h"

/**
 * Do we need to either map r,g,b->a or a->r. configComponentMask indicates which channels are
 * present in the texture's config. swizzleComponentMask indicates the channels present in the
 * shader swizzle.
 */
static bool swizzle_requires_alpha_remapping(const GrGLCaps& caps,
                                             uint32_t configComponentMask,
                                             uint32_t swizzleComponentMask) {
    if (caps.textureSwizzleSupport()) {
        // Any remapping is handled using texture swizzling not shader modifications.
        return false;
    }
    // check if the texture is alpha-only
    if (kA_GrColorComponentFlag == configComponentMask) {
        if (caps.textureRedSupport() && (kA_GrColorComponentFlag & swizzleComponentMask)) {
            // we must map the swizzle 'a's to 'r'.
            return true;
        }
        if (kRGB_GrColorComponentFlags & swizzleComponentMask) {
            // The 'r', 'g', and/or 'b's must be mapped to 'a' according to our semantics that
            // alpha-only textures smear alpha across all four channels when read.
            return true;
        }
    }
    return false;
}

/**
 * The key for an individual coord transform is made up of a matrix type, a precision, and a bit
 * that indicates the source of the input coords.
 */
enum {
    kMatrixTypeKeyBits   = 1,
    kMatrixTypeKeyMask   = (1 << kMatrixTypeKeyBits) - 1,
    
    kPrecisionBits       = 2,
    kPrecisionShift      = kMatrixTypeKeyBits,

    kPositionCoords_Flag = (1 << (kPrecisionShift + kPrecisionBits)),
    kDeviceCoords_Flag   = kPositionCoords_Flag + kPositionCoords_Flag,

    kTransformKeyBits    = kMatrixTypeKeyBits + kPrecisionBits + 2,
};

GR_STATIC_ASSERT(kHigh_GrSLPrecision < (1 << kPrecisionBits));

/**
 * We specialize the vertex code for each of these matrix types.
 */
enum MatrixType {
    kNoPersp_MatrixType  = 0,
    kGeneral_MatrixType  = 1,
};

static uint32_t gen_transform_key(const GrPendingFragmentStage& stage, bool useExplicitLocalCoords) {
    uint32_t totalKey = 0;
    int numTransforms = stage.processor()->numTransforms();
    for (int t = 0; t < numTransforms; ++t) {
        uint32_t key = 0;
        if (stage.isPerspectiveCoordTransform(t)) {
            key |= kGeneral_MatrixType;
        } else {
            key |= kNoPersp_MatrixType;
        }

        const GrCoordTransform& coordTransform = stage.processor()->coordTransform(t);
        if (kLocal_GrCoordSet == coordTransform.sourceCoords() && !useExplicitLocalCoords) {
            key |= kPositionCoords_Flag;
        } else if (kDevice_GrCoordSet == coordTransform.sourceCoords()) {
            key |= kDeviceCoords_Flag;
        }

        GR_STATIC_ASSERT(kGrSLPrecisionCount <= (1 << kPrecisionBits));
        key |= (coordTransform.precision() << kPrecisionShift);

        key <<= kTransformKeyBits * t;

        SkASSERT(0 == (totalKey & key)); // keys for each transform ought not to overlap
        totalKey |= key;
    }
    return totalKey;
}

static uint32_t gen_texture_key(const GrProcessor& proc, const GrGLCaps& caps) {
    uint32_t key = 0;
    int numTextures = proc.numTextures();
    for (int t = 0; t < numTextures; ++t) {
        const GrTextureAccess& access = proc.textureAccess(t);
        uint32_t configComponentMask = GrPixelConfigComponentMask(access.getTexture()->config());
        if (swizzle_requires_alpha_remapping(caps, configComponentMask, access.swizzleMask())) {
            key |= 1 << t;
        }
    }
    return key;
}

/**
 * A function which emits a meta key into the key builder.  This is required because shader code may
 * be dependent on properties of the effect that the effect itself doesn't use
 * in its key (e.g. the pixel format of textures used). So we create a meta-key for
 * every effect using this function. It is also responsible for inserting the effect's class ID
 * which must be different for every GrProcessor subclass. It can fail if an effect uses too many
 * textures, transforms, etc, for the space allotted in the meta-key.  NOTE, both FPs and GPs share
 * this function because it is hairy, though FPs do not have attribs, and GPs do not have transforms
 */
static bool get_meta_key(const GrProcessor& proc,
                         const GrGLCaps& caps,
                         uint32_t transformKey,
                         GrProcessorKeyBuilder* b) {
    size_t processorKeySize = b->size();
    uint32_t textureKey = gen_texture_key(proc, caps);
    uint32_t classID = proc.classID();

    // Currently we allow 16 bits for each of the above portions of the meta-key. Fail if they
    // don't fit.
    static const uint32_t kMetaKeyInvalidMask = ~((uint32_t) SK_MaxU16);
    if ((textureKey | transformKey | classID) & kMetaKeyInvalidMask) {
        return false;
    }
    if (processorKeySize > SK_MaxU16) {
        return false;
    }

    uint32_t* key = b->add32n(2);
    key[0] = (textureKey << 16 | transformKey);
    key[1] = (classID << 16 | SkToU16(processorKeySize));
    return true;
}

bool GrGLProgramDescBuilder::Build(const GrOptDrawState& optState,
                                   const GrProgramDesc::DescInfo& descInfo,
                                   GrGpu::DrawType drawType,
                                   GrGLGpu* gpu,
                                   GrProgramDesc* desc) {
    // The descriptor is used as a cache key. Thus when a field of the
    // descriptor will not affect program generation (because of the attribute
    // bindings in use or other descriptor field settings) it should be set
    // to a canonical value to avoid duplicate programs with different keys.

    bool requiresLocalCoordAttrib = descInfo.fRequiresLocalCoordAttrib;

    GR_STATIC_ASSERT(0 == kProcessorKeysOffset % sizeof(uint32_t));
    // Make room for everything up to the effect keys.
    desc->fKey.reset();
    desc->fKey.push_back_n(kProcessorKeysOffset);

    GrProcessorKeyBuilder b(&desc->fKey);

    const GrPrimitiveProcessor& primProc = *optState.getPrimitiveProcessor();
    primProc.getGLProcessorKey(optState.getBatchTracker(), gpu->glCaps(), &b);
    if (!get_meta_key(primProc, gpu->glCaps(), 0, &b)) {
        desc->fKey.reset();
        return false;
    }

    for (int s = 0; s < optState.numFragmentStages(); ++s) {
        const GrPendingFragmentStage& fps = optState.getFragmentStage(s);
        const GrFragmentProcessor& fp = *fps.processor();
        fp.getGLProcessorKey(gpu->glCaps(), &b);
        if (!get_meta_key(fp, gpu->glCaps(),
                          gen_transform_key(fps, requiresLocalCoordAttrib), &b)) {
            desc->fKey.reset();
            return false;
        }
    }

    const GrXferProcessor& xp = *optState.getXferProcessor();
    xp.getGLProcessorKey(gpu->glCaps(), &b);
    if (!get_meta_key(xp, gpu->glCaps(), 0, &b)) {
        desc->fKey.reset();
        return false;
    }

    // --------DO NOT MOVE HEADER ABOVE THIS LINE--------------------------------------------------
    // Because header is a pointer into the dynamic array, we can't push any new data into the key
    // below here.
    GLKeyHeader* header = desc->atOffset<GLKeyHeader, kHeaderOffset>();

    // make sure any padding in the header is zeroed.
    memset(header, 0, kHeaderSize);

    bool isPathRendering = GrGpu::IsPathRenderingDrawType(drawType);
    if (gpu->caps()->pathRenderingSupport() && isPathRendering) {
        header->fUseNvpr = true;
        SkASSERT(!optState.hasGeometryProcessor());
    } else {
        header->fUseNvpr = false;
    }

    if (descInfo.fReadsDst) {
        const GrDeviceCoordTexture* dstCopy = optState.getDstCopy();
        SkASSERT(dstCopy || gpu->caps()->dstReadInShaderSupport());
        const GrTexture* dstCopyTexture = NULL;
        if (dstCopy) {
            dstCopyTexture = dstCopy->texture();
        }
        header->fDstReadKey = GrGLFragmentShaderBuilder::KeyForDstRead(dstCopyTexture,
                                                                       gpu->glCaps());
        SkASSERT(0 != header->fDstReadKey);
    } else {
        header->fDstReadKey = 0;
    }

    if (descInfo.fReadsFragPosition) {
        header->fFragPosKey =
                GrGLFragmentShaderBuilder::KeyForFragmentPosition(optState.getRenderTarget(),
                                                                  gpu->glCaps());
    } else {
        header->fFragPosKey = 0;
    }

    header->fColorEffectCnt = optState.numColorStages();
    header->fCoverageEffectCnt = optState.numCoverageStages();
    desc->finalize();
    return true;
}