aboutsummaryrefslogtreecommitdiffhomepage
path: root/bench/MatrixBench.cpp
blob: a2e459adc4c77e67af0e62eadea8054a0215580e (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
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368

/*
 * Copyright 2011 Google Inc.
 *
 * Use of this source code is governed by a BSD-style license that can be
 * found in the LICENSE file.
 */
#include "SkBenchmark.h"
#include "SkMatrix.h"
#include "SkRandom.h"
#include "SkString.h"

class MatrixBench : public SkBenchmark {
    SkString    fName;
    enum { N = 100000 };
public:
    MatrixBench(void* param, const char name[]) : INHERITED(param) {
        fName.printf("matrix_%s", name);
    }

    virtual void performTest() = 0;

protected:
    virtual int mulLoopCount() const { return 1; }

    virtual const char* onGetName() {
        return fName.c_str();
    }

    virtual void onDraw(SkCanvas* canvas) {
        int n = SkBENCHLOOP(N * this->mulLoopCount());
        for (int i = 0; i < n; i++) {
            this->performTest();
        }
    }

private:
    typedef SkBenchmark INHERITED;
};

// we want to stop the compiler from eliminating code that it thinks is a no-op
// so we have a non-static global we increment, hoping that will convince the
// compiler to execute everything
int gMatrixBench_NonStaticGlobal;

#define always_do(pred)                     \
    do {                                    \
        if (pred) {                         \
            ++gMatrixBench_NonStaticGlobal; \
        }                                   \
    } while (0)

class EqualsMatrixBench : public MatrixBench {
public:
    EqualsMatrixBench(void* param) : INHERITED(param, "equals") {}
protected:
    virtual void performTest() {
        SkMatrix m0, m1, m2;

        m0.reset();
        m1.reset();
        m2.reset();
        always_do(m0 == m1);
        always_do(m1 == m2);
        always_do(m2 == m0);
    }
private:
    typedef MatrixBench INHERITED;
};

class ScaleMatrixBench : public MatrixBench {
public:
    ScaleMatrixBench(void* param) : INHERITED(param, "scale") {
        fSX = fSY = SkFloatToScalar(1.5f);
        fM0.reset();
        fM1.setScale(fSX, fSY);
        fM2.setTranslate(fSX, fSY);
    }
protected:
    virtual void performTest() {
        SkMatrix m;
        m = fM0; m.preScale(fSX, fSY);
        m = fM1; m.preScale(fSX, fSY);
        m = fM2; m.preScale(fSX, fSY);
    }
private:
    SkMatrix fM0, fM1, fM2;
    SkScalar fSX, fSY;
    typedef MatrixBench INHERITED;
};

// having unknown values in our arrays can throw off the timing a lot, perhaps
// handling NaN values is a lot slower. Anyway, this guy is just meant to put
// reasonable values in our arrays.
template <typename T> void init9(T array[9]) {
    SkRandom rand;
    for (int i = 0; i < 9; i++) {
        array[i] = rand.nextSScalar1();
    }
}

// Test the performance of setConcat() non-perspective case:
// using floating point precision only.
class FloatConcatMatrixBench : public MatrixBench {
public:
    FloatConcatMatrixBench(void* p) : INHERITED(p, "concat_floatfloat") {
        init9(mya);
        init9(myb);
        init9(myr);
    }
protected:
    virtual int mulLoopCount() const { return 4; }

    static inline void muladdmul(float a, float b, float c, float d,
                                   float* result) {
      *result = a * b + c * d;
    }
    virtual void performTest() {
        const float* a = mya;
        const float* b = myb;
        float* r = myr;
        muladdmul(a[0], b[0], a[1], b[3], &r[0]);
        muladdmul(a[0], b[1], a[1], b[4], &r[1]);
        muladdmul(a[0], b[2], a[1], b[5], &r[2]);
        r[2] += a[2];
        muladdmul(a[3], b[0], a[4], b[3], &r[3]);
        muladdmul(a[3], b[1], a[4], b[4], &r[4]);
        muladdmul(a[3], b[2], a[4], b[5], &r[5]);
        r[5] += a[5];
        r[6] = r[7] = 0.0f;
        r[8] = 1.0f;
    }
private:
    float mya [9];
    float myb [9];
    float myr [9];
    typedef MatrixBench INHERITED;
};

static inline float SkDoubleToFloat(double x) {
    return static_cast<float>(x);
}

// Test the performance of setConcat() non-perspective case:
// using floating point precision but casting up to float for
// intermediate results during computations.
class FloatDoubleConcatMatrixBench : public MatrixBench {
public:
    FloatDoubleConcatMatrixBench(void* p) : INHERITED(p, "concat_floatdouble") {
        init9(mya);
        init9(myb);
        init9(myr);
    }
protected:
    virtual int mulLoopCount() const { return 4; }

    static inline void muladdmul(float a, float b, float c, float d,
                                   float* result) {
      *result = SkDoubleToFloat((double)a * b + (double)c * d);
    }
    virtual void performTest() {
        const float* a = mya;
        const float* b = myb;
        float* r = myr;
        muladdmul(a[0], b[0], a[1], b[3], &r[0]);
        muladdmul(a[0], b[1], a[1], b[4], &r[1]);
        muladdmul(a[0], b[2], a[1], b[5], &r[2]);
        r[2] += a[2];
        muladdmul(a[3], b[0], a[4], b[3], &r[3]);
        muladdmul(a[3], b[1], a[4], b[4], &r[4]);
        muladdmul(a[3], b[2], a[4], b[5], &r[5]);
        r[5] += a[5];
        r[6] = r[7] = 0.0f;
        r[8] = 1.0f;
    }
private:
    float mya [9];
    float myb [9];
    float myr [9];
    typedef MatrixBench INHERITED;
};

// Test the performance of setConcat() non-perspective case:
// using double precision only.
class DoubleConcatMatrixBench : public MatrixBench {
public:
    DoubleConcatMatrixBench(void* p) : INHERITED(p, "concat_double") {
        init9(mya);
        init9(myb);
        init9(myr);
    }
protected:
    virtual int mulLoopCount() const { return 4; }

    static inline void muladdmul(double a, double b, double c, double d,
                                   double* result) {
      *result = a * b + c * d;
    }
    virtual void performTest() {
        const double* a = mya;
        const double* b = myb;
        double* r = myr;
        muladdmul(a[0], b[0], a[1], b[3], &r[0]);
        muladdmul(a[0], b[1], a[1], b[4], &r[1]);
        muladdmul(a[0], b[2], a[1], b[5], &r[2]);
        r[2] += a[2];
        muladdmul(a[3], b[0], a[4], b[3], &r[3]);
        muladdmul(a[3], b[1], a[4], b[4], &r[4]);
        muladdmul(a[3], b[2], a[4], b[5], &r[5]);
        r[5] += a[5];
        r[6] = r[7] = 0.0;
        r[8] = 1.0;
    }
private:
    double mya [9];
    double myb [9];
    double myr [9];
    typedef MatrixBench INHERITED;
};

class GetTypeMatrixBench : public MatrixBench {
public:
    GetTypeMatrixBench(void* param)
        : INHERITED(param, "gettype") {
        fArray[0] = (float) fRnd.nextS();
        fArray[1] = (float) fRnd.nextS();
        fArray[2] = (float) fRnd.nextS();
        fArray[3] = (float) fRnd.nextS();
        fArray[4] = (float) fRnd.nextS();
        fArray[5] = (float) fRnd.nextS();
        fArray[6] = (float) fRnd.nextS();
        fArray[7] = (float) fRnd.nextS();
        fArray[8] = (float) fRnd.nextS();
    }
protected:
    // Putting random generation of the matrix inside performTest()
    // would help us avoid anomalous runs, but takes up 25% or
    // more of the function time.
    virtual void performTest() {
        fMatrix.setAll(fArray[0], fArray[1], fArray[2],
                       fArray[3], fArray[4], fArray[5],
                       fArray[6], fArray[7], fArray[8]);
        always_do(fMatrix.getType());
        fMatrix.dirtyMatrixTypeCache();
        always_do(fMatrix.getType());
        fMatrix.dirtyMatrixTypeCache();
        always_do(fMatrix.getType());
        fMatrix.dirtyMatrixTypeCache();
        always_do(fMatrix.getType());
        fMatrix.dirtyMatrixTypeCache();
        always_do(fMatrix.getType());
        fMatrix.dirtyMatrixTypeCache();
        always_do(fMatrix.getType());
        fMatrix.dirtyMatrixTypeCache();
        always_do(fMatrix.getType());
        fMatrix.dirtyMatrixTypeCache();
        always_do(fMatrix.getType());
    }
private:
    SkMatrix fMatrix;
    float fArray[9];
    SkRandom fRnd;
    typedef MatrixBench INHERITED;
};

#ifdef SK_SCALAR_IS_FLOAT
class ScaleTransMixedMatrixBench : public MatrixBench {
 public:
    ScaleTransMixedMatrixBench(void* p) : INHERITED(p, "scaletrans_mixed"), fCount (16) {
        fMatrix.setAll(fRandom.nextSScalar1(), fRandom.nextSScalar1(), fRandom.nextSScalar1(),
                       fRandom.nextSScalar1(), fRandom.nextSScalar1(), fRandom.nextSScalar1(),
                       fRandom.nextSScalar1(), fRandom.nextSScalar1(), fRandom.nextSScalar1());
        int i;
        for (i = 0; i < SkBENCHLOOP(fCount); i++) {
            fSrc[i].fX = fRandom.nextSScalar1();
            fSrc[i].fY = fRandom.nextSScalar1();
            fDst[i].fX = fRandom.nextSScalar1();
            fDst[i].fY = fRandom.nextSScalar1();
        }
    }
 protected:
    virtual void performTest() {
        SkPoint* dst = fDst;
        const SkPoint* src = fSrc;
        int count = SkBENCHLOOP(fCount);
        float mx = fMatrix[SkMatrix::kMScaleX];
        float my = fMatrix[SkMatrix::kMScaleY];
        float tx = fMatrix[SkMatrix::kMTransX];
        float ty = fMatrix[SkMatrix::kMTransY];
        do {
            dst->fY = SkScalarMulAdd(src->fY, my, ty);
            dst->fX = SkScalarMulAdd(src->fX, mx, tx);
            src += 1;
            dst += 1;
        } while (--count);
    }
 private:
    SkMatrix fMatrix;
    SkPoint fSrc [16];
    SkPoint fDst [16];
    int fCount;
    SkRandom fRandom;
    typedef MatrixBench INHERITED;
};

class ScaleTransDoubleMatrixBench : public MatrixBench {
 public:
    ScaleTransDoubleMatrixBench(void* p) : INHERITED(p, "scaletrans_double"), fCount (16) {
        init9(fMatrix);
        int i;
        for (i = 0; i < SkBENCHLOOP(fCount); i++) {
            fSrc[i].fX = fRandom.nextSScalar1();
            fSrc[i].fY = fRandom.nextSScalar1();
            fDst[i].fX = fRandom.nextSScalar1();
            fDst[i].fY = fRandom.nextSScalar1();
        }
    }
 protected:
    virtual void performTest() {
        SkPoint* dst = fDst;
        const SkPoint* src = fSrc;
        int count = SkBENCHLOOP(fCount);
        // As doubles, on Z600 Linux systems this is 2.5x as expensive as mixed mode
        float mx = (float) fMatrix[SkMatrix::kMScaleX];
        float my = (float) fMatrix[SkMatrix::kMScaleY];
        float tx = (float) fMatrix[SkMatrix::kMTransX];
        float ty = (float) fMatrix[SkMatrix::kMTransY];
        do {
            dst->fY = src->fY * my + ty;
            dst->fX = src->fX * mx + tx;
            src += 1;
            dst += 1;
        } while (--count);
    }
 private:
    double fMatrix [9];
    SkPoint fSrc [16];
    SkPoint fDst [16];
    int fCount;
    SkRandom fRandom;
    typedef MatrixBench INHERITED;
};
#endif





static SkBenchmark* M0(void* p) { return new EqualsMatrixBench(p); }
static SkBenchmark* M1(void* p) { return new ScaleMatrixBench(p); }
static SkBenchmark* M2(void* p) { return new FloatConcatMatrixBench(p); }
static SkBenchmark* M3(void* p) { return new FloatDoubleConcatMatrixBench(p); }
static SkBenchmark* M4(void* p) { return new DoubleConcatMatrixBench(p); }
static SkBenchmark* M5(void* p) { return new GetTypeMatrixBench(p); }

static BenchRegistry gReg0(M0);
static BenchRegistry gReg1(M1);
static BenchRegistry gReg2(M2);
static BenchRegistry gReg3(M3);
static BenchRegistry gReg4(M4);
static BenchRegistry gReg5(M5);

#ifdef SK_SCALAR_IS_FLOAT
static SkBenchmark* FlM0(void* p) { return new ScaleTransMixedMatrixBench(p); }
static SkBenchmark* FlM1(void* p) { return new ScaleTransDoubleMatrixBench(p); }
static BenchRegistry gFlReg5(FlM0);
static BenchRegistry gFlReg6(FlM1);
#endif