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
|
/*
* Copyright 2016 Google Inc.
*
* Use of this source code is governed by a BSD-style license that can be
* found in the LICENSE file.
*/
#include <tuple>
#include "Benchmark.h"
#include "Resources.h"
#include "SkCpu.h"
#include "SkImage.h"
#include "SkImage_Base.h"
#include "SkNx.h"
#include "SkOpts.h"
#include "SkPM4fPriv.h"
#include "SkString.h"
#define INNER_LOOPS 10
static inline void brute_srcover_srgb_srgb_1(uint32_t* dst, uint32_t src) {
auto d = Sk4f_fromS32(*dst),
s = Sk4f_fromS32( src);
*dst = Sk4f_toS32(s + d * (1.0f - s[3]));
}
static inline void srcover_srgb_srgb_1(uint32_t* dst, uint32_t src) {
if (src >= 0xFF000000) {
*dst = src;
return;
}
brute_srcover_srgb_srgb_1(dst, src);
}
static void brute_force_srcover_srgb_srgb(
uint32_t* dst, const uint32_t* const src, int ndst, const int nsrc) {
while (ndst > 0) {
int n = SkTMin(ndst, nsrc);
for (int i = 0; i < n; i++) {
brute_srcover_srgb_srgb_1(dst++, src[i]);
}
ndst -= n;
}
}
static void trivial_srcover_srgb_srgb(
uint32_t* dst, const uint32_t* const src, int ndst, const int nsrc) {
while (ndst > 0) {
int n = SkTMin(ndst, nsrc);
for (int i = 0; i < n; i++) {
srcover_srgb_srgb_1(dst++, src[i]);
}
ndst -= n;
}
}
static void best_non_simd_srcover_srgb_srgb(
uint32_t* dst, const uint32_t* const src, int ndst, const int nsrc) {
uint64_t* ddst = reinterpret_cast<uint64_t*>(dst);
auto srcover_srgb_srgb_2 = [](uint32_t* dst, const uint32_t* src) {
srcover_srgb_srgb_1(dst++, *src++);
srcover_srgb_srgb_1(dst, *src);
};
while (ndst >0) {
int count = SkTMin(ndst, nsrc);
ndst -= count;
const uint64_t* dsrc = reinterpret_cast<const uint64_t*>(src);
const uint64_t* end = dsrc + (count >> 1);
do {
if ((~*dsrc & 0xFF000000FF000000) == 0) {
do {
*ddst++ = *dsrc++;
} while (dsrc < end && (~*dsrc & 0xFF000000FF000000) == 0);
} else if ((*dsrc & 0xFF000000FF000000) == 0) {
do {
dsrc++;
ddst++;
} while (dsrc < end && (*dsrc & 0xFF000000FF000000) == 0);
} else {
srcover_srgb_srgb_2(reinterpret_cast<uint32_t*>(ddst++),
reinterpret_cast<const uint32_t*>(dsrc++));
}
} while (dsrc < end);
if ((count & 1) != 0) {
srcover_srgb_srgb_1(reinterpret_cast<uint32_t*>(ddst),
*reinterpret_cast<const uint32_t*>(dsrc));
}
}
}
class SrcOverVSkOptsBruteForce {
public:
static SkString Name() { return SkString{"VSkOptsBruteForce"}; }
static void BlendN(uint32_t* dst, const uint32_t* src, int count) {
brute_force_srcover_srgb_srgb(dst, src, count, count);
}
};
class SrcOverVSkOptsTrivial {
public:
static SkString Name() { return SkString{"VSkOptsTrivial"}; }
static void BlendN(uint32_t* dst, const uint32_t* src, int count) {
trivial_srcover_srgb_srgb(dst, src, count, count);
}
};
class SrcOverVSkOptsNonSimdCore {
public:
static SkString Name() { return SkString{"VSkOptsNonSimdCore"}; }
static void BlendN(uint32_t* dst, const uint32_t* src, int count) {
best_non_simd_srcover_srgb_srgb(dst, src, count, count);
}
};
class SrcOverVSkOptsDefault {
public:
static SkString Name() { return SkString{"VSkOptsDefault"}; }
static void BlendN(uint32_t* dst, const uint32_t* src, int count) {
SkOpts::srcover_srgb_srgb(dst, src, count, count);
}
};
///////////////////////////////////////////////////////////////////////////////////////////////////
template <typename Blender>
class LinearSrcOverBench : public Benchmark {
public:
LinearSrcOverBench(const char* fileName) : fFileName(fileName) {
fName = "LinearSrcOver_";
fName.append(fileName);
fName.append(Blender::Name());
}
protected:
bool isSuitableFor(Backend backend) override { return backend == kNonRendering_Backend; }
const char* onGetName() override { return fName.c_str(); }
void onPreDraw(SkCanvas*) override {
if (!fPixmap.addr()) {
sk_sp<SkImage> image = GetResourceAsImage(fFileName.c_str());
SkBitmap bm;
if (!as_IB(image)->getROPixels(&bm)) {
SkFAIL("Could not read resource");
}
bm.peekPixels(&fPixmap);
fCount = fPixmap.rowBytesAsPixels();
fDst.reset(fCount);
sk_bzero(fDst.get(), fPixmap.rowBytes());
}
}
void onDraw(int loops, SkCanvas*) override {
SkASSERT(fPixmap.colorType() == kN32_SkColorType);
const int width = fPixmap.rowBytesAsPixels();
for (int i = 0; i < loops * INNER_LOOPS; ++i) {
const uint32_t* src = fPixmap.addr32();
for (int y = 0; y < fPixmap.height(); y++) {
Blender::BlendN(fDst.get(), src, width);
src += width;
}
}
}
void onPostDraw(SkCanvas*) override {
// Make sure the compiler does not optimize away the operation.
volatile uint32_t v = 0;
for (int i = 0; i < fCount; i++) {
v ^= fDst[i];
}
}
private:
int fCount;
SkAutoTArray<uint32_t> fDst;
SkString fFileName;
SkString fName;
SkPixmap fPixmap;
typedef Benchmark INHERITED;
};
#define BENCHES(fileName) \
DEF_BENCH( return new LinearSrcOverBench<SrcOverVSkOptsBruteForce>(fileName); ) \
DEF_BENCH( return new LinearSrcOverBench<SrcOverVSkOptsTrivial>(fileName); ) \
DEF_BENCH( return new LinearSrcOverBench<SrcOverVSkOptsNonSimdCore>(fileName); ) \
DEF_BENCH( return new LinearSrcOverBench<SrcOverVSkOptsDefault>(fileName); )
BENCHES("yellow_rose.png")
BENCHES("baby_tux.png")
BENCHES("plane.png")
BENCHES("mandrill_512.png")
BENCHES("iconstrip.png")
|