aboutsummaryrefslogtreecommitdiffhomepage
path: root/src/gpu/effects/GrGaussianConvolutionFragmentProcessor.cpp
blob: f05bf42c92b418eb14cd63f57fcba9f33f71ec23 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
/*
 * Copyright 2012 Google Inc.
 *
 * Use of this source code is governed by a BSD-style license that can be
 * found in the LICENSE file.
 */

#include "GrGaussianConvolutionFragmentProcessor.h"

#include "GrTexture.h"
#include "GrTextureProxy.h"
#include "glsl/GrGLSLFragmentProcessor.h"
#include "glsl/GrGLSLFragmentShaderBuilder.h"
#include "glsl/GrGLSLProgramDataManager.h"
#include "glsl/GrGLSLUniformHandler.h"

// For brevity
using UniformHandle = GrGLSLProgramDataManager::UniformHandle;
using Direction = GrGaussianConvolutionFragmentProcessor::Direction;

class GrGLConvolutionEffect : public GrGLSLFragmentProcessor {
public:
    void emitCode(EmitArgs&) override;

    static inline void GenKey(const GrProcessor&, const GrShaderCaps&, GrProcessorKeyBuilder*);

protected:
    void onSetData(const GrGLSLProgramDataManager&, const GrFragmentProcessor&) override;

private:
    UniformHandle fKernelUni;
    UniformHandle fImageIncrementUni;
    UniformHandle fBoundsUni;

    typedef GrGLSLFragmentProcessor INHERITED;
};

void GrGLConvolutionEffect::emitCode(EmitArgs& args) {
    const GrGaussianConvolutionFragmentProcessor& ce =
            args.fFp.cast<GrGaussianConvolutionFragmentProcessor>();

    GrGLSLUniformHandler* uniformHandler = args.fUniformHandler;
    fImageIncrementUni = uniformHandler->addUniform(kFragment_GrShaderFlag, kHalf2_GrSLType,
                                                    "ImageIncrement");
    if (ce.useBounds()) {
        fBoundsUni = uniformHandler->addUniform(kFragment_GrShaderFlag, kHalf2_GrSLType,
                                                "Bounds");
    }

    int width = ce.width();

    int arrayCount = (width + 3) / 4;
    SkASSERT(4 * arrayCount >= width);

    fKernelUni = uniformHandler->addUniformArray(kFragment_GrShaderFlag, kHalf4_GrSLType,
                                                 "Kernel", arrayCount);

    GrGLSLFPFragmentBuilder* fragBuilder = args.fFragBuilder;
    SkString coords2D = fragBuilder->ensureCoords2D(args.fTransformedCoords[0]);

    fragBuilder->codeAppendf("%s = half4(0, 0, 0, 0);", args.fOutputColor);

    const GrShaderVar& kernel = uniformHandler->getUniformVariable(fKernelUni);
    const char* imgInc = uniformHandler->getUniformCStr(fImageIncrementUni);

    fragBuilder->codeAppendf("float2 coord = %s - %d.0 * %s;", coords2D.c_str(), ce.radius(), imgInc);
    fragBuilder->codeAppend("float2 coordSampled = half2(0, 0);");

    // Manually unroll loop because some drivers don't; yields 20-30% speedup.
    const char* kVecSuffix[4] = {".x", ".y", ".z", ".w"};
    for (int i = 0; i < width; i++) {
        SkString index;
        SkString kernelIndex;
        index.appendS32(i / 4);
        kernel.appendArrayAccess(index.c_str(), &kernelIndex);
        kernelIndex.append(kVecSuffix[i & 0x3]);

        fragBuilder->codeAppend("coordSampled = coord;");
        if (ce.useBounds()) {
            // We used to compute a bool indicating whether we're in bounds or not, cast it to a
            // float, and then mul weight*texture_sample by the float. However, the Adreno 430 seems
            // to have a bug that caused corruption.
            const char* bounds = uniformHandler->getUniformCStr(fBoundsUni);
            const char* component = ce.direction() == Direction::kY ? "y" : "x";

            switch (ce.mode()) {
                case GrTextureDomain::kClamp_Mode: {
                    fragBuilder->codeAppendf("coordSampled.%s = clamp(coord.%s, %s.x, %s.y);\n",
                                             component, component, bounds, bounds);
                    break;
                }
                case GrTextureDomain::kRepeat_Mode: {
                    fragBuilder->codeAppendf("coordSampled.%s = "
                                             "mod(coord.%s - %s.x, %s.y - %s.x) + %s.x;\n",
                                             component, component, bounds, bounds, bounds, bounds);
                    break;
                }
                case GrTextureDomain::kDecal_Mode: {
                    fragBuilder->codeAppendf("if (coord.%s >= %s.x && coord.%s <= %s.y) {",
                                             component, bounds, component, bounds);
                    break;
                }
                default: {
                    SK_ABORT("Unsupported operation.");
                }
            }
        }
        fragBuilder->codeAppendf("%s += ", args.fOutputColor);
        fragBuilder->appendTextureLookup(args.fTexSamplers[0], "coordSampled");
        fragBuilder->codeAppendf(" * %s;\n", kernelIndex.c_str());
        if (GrTextureDomain::kDecal_Mode == ce.mode()) {
            fragBuilder->codeAppend("}");
        }
        fragBuilder->codeAppendf("coord += %s;\n", imgInc);
    }
    fragBuilder->codeAppendf("%s *= %s;\n", args.fOutputColor, args.fInputColor);
}

void GrGLConvolutionEffect::onSetData(const GrGLSLProgramDataManager& pdman,
                                      const GrFragmentProcessor& processor) {
    const GrGaussianConvolutionFragmentProcessor& conv =
            processor.cast<GrGaussianConvolutionFragmentProcessor>();
    GrSurfaceProxy* proxy = conv.textureSampler(0).proxy();
    GrTexture& texture = *proxy->priv().peekTexture();

    float imageIncrement[2] = {0};
    float ySign = proxy->origin() != kTopLeft_GrSurfaceOrigin ? 1.0f : -1.0f;
    switch (conv.direction()) {
        case Direction::kX:
            imageIncrement[0] = 1.0f / texture.width();
            break;
        case Direction::kY:
            imageIncrement[1] = ySign / texture.height();
            break;
        default:
            SK_ABORT("Unknown filter direction.");
    }
    pdman.set2fv(fImageIncrementUni, 1, imageIncrement);
    if (conv.useBounds()) {
        float bounds[2] = {0};
        bounds[0] = conv.bounds()[0];
        bounds[1] = conv.bounds()[1];
        if (GrTextureDomain::kClamp_Mode == conv.mode()) {
            bounds[0] += SK_ScalarHalf;
            bounds[1] -= SK_ScalarHalf;
        }
        if (Direction::kX == conv.direction()) {
            SkScalar inv = SkScalarInvert(SkIntToScalar(texture.width()));
            bounds[0] *= inv;
            bounds[1] *= inv;
        } else {
            SkScalar inv = SkScalarInvert(SkIntToScalar(texture.height()));
            if (proxy->origin() != kTopLeft_GrSurfaceOrigin) {
                float tmp = bounds[0];
                bounds[0] = 1.0f - (inv * bounds[1]);
                bounds[1] = 1.0f - (inv * tmp);
            } else {
                bounds[0] *= inv;
                bounds[1] *= inv;
            }
        }

        SkASSERT(bounds[0] <= bounds[1]);
        pdman.set2f(fBoundsUni, bounds[0], bounds[1]);
    }
    int width = conv.width();

    int arrayCount = (width + 3) / 4;
    SkASSERT(4 * arrayCount >= width);
    pdman.set4fv(fKernelUni, arrayCount, conv.kernel());
}

void GrGLConvolutionEffect::GenKey(const GrProcessor& processor, const GrShaderCaps&,
                                   GrProcessorKeyBuilder* b) {
    const GrGaussianConvolutionFragmentProcessor& conv =
            processor.cast<GrGaussianConvolutionFragmentProcessor>();
    uint32_t key = conv.radius();
    key <<= 3;
    key |= Direction::kY == conv.direction() ? 0x4 : 0x0;
    key |= static_cast<uint32_t>(conv.mode());
    b->add32(key);
}

///////////////////////////////////////////////////////////////////////////////
static void fill_in_1D_gaussian_kernel(float* kernel, int width, float gaussianSigma, int radius) {
    const float twoSigmaSqrd = 2.0f * gaussianSigma * gaussianSigma;
    if (SkScalarNearlyZero(twoSigmaSqrd, SK_ScalarNearlyZero)) {
        for (int i = 0; i < width; ++i) {
            kernel[i] = 0.0f;
        }
        return;
    }

    const float denom = 1.0f / twoSigmaSqrd;

    float sum = 0.0f;
    for (int i = 0; i < width; ++i) {
        float x = static_cast<float>(i - radius);
        // Note that the constant term (1/(sqrt(2*pi*sigma^2)) of the Gaussian
        // is dropped here, since we renormalize the kernel below.
        kernel[i] = sk_float_exp(-x * x * denom);
        sum += kernel[i];
    }
    // Normalize the kernel
    float scale = 1.0f / sum;
    for (int i = 0; i < width; ++i) {
        kernel[i] *= scale;
    }
}

GrGaussianConvolutionFragmentProcessor::GrGaussianConvolutionFragmentProcessor(
                                                            sk_sp<GrTextureProxy> proxy,
                                                            Direction direction,
                                                            int radius,
                                                            float gaussianSigma,
                                                            GrTextureDomain::Mode mode,
                                                            int bounds[2])
        : INHERITED(kGrGaussianConvolutionFragmentProcessor_ClassID,
                    ModulateByConfigOptimizationFlags(proxy->config()))
        , fCoordTransform(proxy.get())
        , fTextureSampler(std::move(proxy))
        , fRadius(radius)
        , fDirection(direction)
        , fMode(mode) {
    this->addCoordTransform(&fCoordTransform);
    this->setTextureSamplerCnt(1);
    SkASSERT(radius <= kMaxKernelRadius);

    fill_in_1D_gaussian_kernel(fKernel, this->width(), gaussianSigma, this->radius());

    memcpy(fBounds, bounds, sizeof(fBounds));
}

GrGaussianConvolutionFragmentProcessor::GrGaussianConvolutionFragmentProcessor(
        const GrGaussianConvolutionFragmentProcessor& that)
        : INHERITED(kGrGaussianConvolutionFragmentProcessor_ClassID, that.optimizationFlags())
        , fCoordTransform(that.fCoordTransform)
        , fTextureSampler(that.fTextureSampler)
        , fRadius(that.fRadius)
        , fDirection(that.fDirection)
        , fMode(that.fMode) {
    this->addCoordTransform(&fCoordTransform);
    this->setTextureSamplerCnt(1);
    memcpy(fKernel, that.fKernel, that.width() * sizeof(float));
    memcpy(fBounds, that.fBounds, sizeof(fBounds));
}

void GrGaussianConvolutionFragmentProcessor::onGetGLSLProcessorKey(const GrShaderCaps& caps,
                                                                   GrProcessorKeyBuilder* b) const {
    GrGLConvolutionEffect::GenKey(*this, caps, b);
}

GrGLSLFragmentProcessor* GrGaussianConvolutionFragmentProcessor::onCreateGLSLInstance() const {
    return new GrGLConvolutionEffect;
}

bool GrGaussianConvolutionFragmentProcessor::onIsEqual(const GrFragmentProcessor& sBase) const {
    const GrGaussianConvolutionFragmentProcessor& s =
            sBase.cast<GrGaussianConvolutionFragmentProcessor>();
    return (this->radius() == s.radius() && this->direction() == s.direction() &&
            this->mode() == s.mode() &&
            0 == memcmp(fBounds, s.fBounds, sizeof(fBounds)) &&
            0 == memcmp(fKernel, s.fKernel, this->width() * sizeof(float)));
}

///////////////////////////////////////////////////////////////////////////////

GR_DEFINE_FRAGMENT_PROCESSOR_TEST(GrGaussianConvolutionFragmentProcessor);

#if GR_TEST_UTILS
std::unique_ptr<GrFragmentProcessor> GrGaussianConvolutionFragmentProcessor::TestCreate(
        GrProcessorTestData* d) {
    int texIdx = d->fRandom->nextBool() ? GrProcessorUnitTest::kSkiaPMTextureIdx
                                        : GrProcessorUnitTest::kAlphaTextureIdx;
    sk_sp<GrTextureProxy> proxy = d->textureProxy(texIdx);

    int bounds[2];
    int modeIdx = d->fRandom->nextRangeU(0, GrTextureDomain::kModeCount-1);

    Direction dir;
    if (d->fRandom->nextBool()) {
        dir = Direction::kX;
        bounds[0] = d->fRandom->nextRangeU(0, proxy->width()-2);
        bounds[1] = d->fRandom->nextRangeU(bounds[0]+1, proxy->width()-1);
    } else {
        dir = Direction::kY;
        bounds[0] = d->fRandom->nextRangeU(0, proxy->height()-2);
        bounds[1] = d->fRandom->nextRangeU(bounds[0]+1, proxy->height()-1);
    }

    int radius = d->fRandom->nextRangeU(1, kMaxKernelRadius);
    float sigma = radius / 3.f;

    return GrGaussianConvolutionFragmentProcessor::Make(
            d->textureProxy(texIdx),
            dir, radius, sigma, static_cast<GrTextureDomain::Mode>(modeIdx), bounds);
}
#endif