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
|
/*
* Copyright 2017 Google Inc.
*
* Use of this source code is governed by a BSD-style license that can be
* found in the LICENSE file.
*/
#include "SkCpu.h"
#include "SkJumper.h"
#include "SkRasterPipeline.h"
#include "SkTemplates.h"
// We'll use __has_feature(memory_sanitizer) to detect MSAN.
// SkJumper_generated.S is not compiled with MSAN, so MSAN would yell really loud.
#if !defined(__has_feature)
#define __has_feature(x) 0
#endif
// Stages expect these constants to be set to these values.
// It's fine to rearrange and add new ones if you update SkJumper_constants.
using K = const SkJumper_constants;
static K kConstants = {
1.0f, 0.5f, 255.0f, 1/255.0f, 0x000000ff,
{0.0f, 1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f},
0.0025f, 0.6975f, 0.3000f, 1/12.92f, 0.055f, // from_srgb
12.46f, 0.411192f, 0.689206f, -0.0988f, 0.0043f, // to_srgb
0x77800000, 0x07800000, 0x04000400, // fp16 <-> fp32
};
#define STAGES(M) \
M(seed_shader) \
M(constant_color) \
M(clear) \
M(plus_) \
M(srcover) \
M(dstover) \
M(clamp_0) \
M(clamp_1) \
M(clamp_a) \
M(swap) \
M(move_src_dst) \
M(move_dst_src) \
M(premul) \
M(unpremul) \
M(from_srgb) \
M(to_srgb) \
M(scale_u8) \
M(load_tables) \
M(load_8888) \
M(store_8888) \
M(load_f16) \
M(store_f16) \
M(matrix_2x3) \
M(matrix_3x4) \
M(clamp_x) \
M(clamp_y) \
M(linear_gradient_2stops)
// We can't express the real types of most stage functions portably, so we use a stand-in.
// We'll only ever call start_pipeline(), which then chains into the rest for us.
using StageFn = void(void);
// Some platforms expect C "name" maps to asm "_name", others to "name".
#if defined(__APPLE__)
#define ASM(name, suffix) sk_##name##_##suffix
#else
#define ASM(name, suffix) _sk_##name##_##suffix
#endif
extern "C" {
#if __has_feature(memory_sanitizer)
// We'll just run portable code.
#elif defined(__aarch64__)
size_t ASM(start_pipeline,aarch64)(size_t, void**, K*, size_t);
StageFn ASM(just_return,aarch64);
#define M(st) StageFn ASM(st,aarch64);
STAGES(M)
#undef M
#elif defined(__arm__)
size_t ASM(start_pipeline,vfp4)(size_t, void**, K*, size_t);
StageFn ASM(just_return,vfp4);
#define M(st) StageFn ASM(st,vfp4);
STAGES(M)
#undef M
#elif defined(__x86_64__) || defined(_M_X64)
size_t ASM(start_pipeline,hsw )(size_t, void**, K*, size_t);
size_t ASM(start_pipeline,sse41)(size_t, void**, K*, size_t);
size_t ASM(start_pipeline,sse2 )(size_t, void**, K*, size_t);
StageFn ASM(just_return,hsw),
ASM(just_return,sse41),
ASM(just_return,sse2);
#define M(st) StageFn ASM(st,hsw);
STAGES(M)
#undef M
#define M(st) StageFn ASM(st,sse41);
STAGES(M)
#undef M
#define M(st) StageFn ASM(st,sse2);
STAGES(M)
#undef M
#endif
// Portable, single-pixel stages.
size_t sk_start_pipeline(size_t, void**, K*, size_t);
StageFn sk_just_return;
#define M(st) StageFn sk_##st;
STAGES(M)
#undef M
}
// Translate SkRasterPipeline's StockStage enum to StageFn function pointers.
#if __has_feature(memory_sanitizer)
// We'll just run portable code.
#elif defined(__aarch64__)
static StageFn* lookup_aarch64(SkRasterPipeline::StockStage st) {
switch (st) {
default: return nullptr;
#define M(st) case SkRasterPipeline::st: return ASM(st,aarch64);
STAGES(M)
#undef M
}
}
#elif defined(__arm__)
static StageFn* lookup_vfp4(SkRasterPipeline::StockStage st) {
switch (st) {
default: return nullptr;
#define M(st) case SkRasterPipeline::st: return ASM(st,vfp4);
STAGES(M)
#undef M
}
}
#elif defined(__x86_64__) || defined(_M_X64)
static StageFn* lookup_hsw(SkRasterPipeline::StockStage st) {
switch (st) {
default: return nullptr;
#define M(st) case SkRasterPipeline::st: return ASM(st,hsw);
STAGES(M)
#undef M
}
}
static StageFn* lookup_sse41(SkRasterPipeline::StockStage st) {
switch (st) {
default: return nullptr;
#define M(st) case SkRasterPipeline::st: return ASM(st,sse41);
STAGES(M)
#undef M
}
}
static StageFn* lookup_sse2(SkRasterPipeline::StockStage st) {
switch (st) {
default: return nullptr;
#define M(st) case SkRasterPipeline::st: return ASM(st,sse2);
STAGES(M)
#undef M
}
}
#endif
static StageFn* lookup_portable(SkRasterPipeline::StockStage st) {
switch (st) {
default: return nullptr;
#define M(st) case SkRasterPipeline::st: return sk_##st;
STAGES(M)
#undef M
}
}
bool SkRasterPipeline::run_with_jumper(size_t x, size_t n) const {
SkAutoSTMalloc<64, void*> program(2*fStages.size() + 1);
const size_t limit = x+n;
auto build_and_run = [&](size_t stride,
StageFn* (*lookup)(SkRasterPipeline::StockStage),
StageFn* just_return,
size_t (*start_pipeline)(size_t, void**, K*, size_t)) {
if (x + stride <= limit) {
void** ip = program.get();
for (auto&& st : fStages) {
auto fn = lookup(st.stage);
if (!fn) {
return false;
}
*ip++ = (void*)fn;
*ip++ = st.ctx;
}
*ip = (void*)just_return;
x = start_pipeline(x, program.get(), &kConstants, limit);
}
return true;
};
// While possible, build and run at full vector stride.
#if __has_feature(memory_sanitizer)
// We'll just run portable code.
#elif defined(__aarch64__)
if (!build_and_run(4, lookup_aarch64, ASM(just_return,aarch64), ASM(start_pipeline,aarch64))) {
return false;
}
#elif defined(__arm__)
if (1 && SkCpu::Supports(SkCpu::NEON|SkCpu::NEON_FMA|SkCpu::VFP_FP16)) {
if (!build_and_run(2, lookup_vfp4, ASM(just_return,vfp4), ASM(start_pipeline,vfp4))) {
return false;
}
}
#elif defined(__x86_64__) || defined(_M_X64)
if (1 && SkCpu::Supports(SkCpu::HSW)) {
if (!build_and_run(8, lookup_hsw, ASM(just_return,hsw), ASM(start_pipeline,hsw))) {
return false;
}
}
if (1 && SkCpu::Supports(SkCpu::SSE41)) {
if (!build_and_run(4, lookup_sse41, ASM(just_return,sse41), ASM(start_pipeline,sse41))) {
return false;
}
}
if (1 && SkCpu::Supports(SkCpu::SSE2)) {
if (!build_and_run(4, lookup_sse2, ASM(just_return,sse2), ASM(start_pipeline,sse2))) {
return false;
}
}
#endif
// Finish up any leftover with portable code one pixel at a time.
return build_and_run(1, lookup_portable, sk_just_return, sk_start_pipeline);
}
|