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
|
/*
* Copyright © 2017 Broadcom
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
* IN THE SOFTWARE.
*/
/** @file v3d_cpu_tiling.h
*
* Contains load/store functions common to both v3d and vc4. The utile layout
* stayed the same, though the way utiles get laid out has changed.
*/
static inline void
v3d_load_utile(void *cpu, uint32_t cpu_stride,
void *gpu, uint32_t gpu_stride)
{
#if defined(V3D_BUILD_NEON) && defined(PIPE_ARCH_ARM)
if (gpu_stride == 8) {
__asm__ volatile (
/* Load from the GPU in one shot, no interleave, to
* d0-d7.
*/
"vldm %0, {q0, q1, q2, q3}\n"
/* Store each 8-byte line to cpu-side destination,
* incrementing it by the stride each time.
*/
"vst1.8 d0, [%1], %2\n"
"vst1.8 d1, [%1], %2\n"
"vst1.8 d2, [%1], %2\n"
"vst1.8 d3, [%1], %2\n"
"vst1.8 d4, [%1], %2\n"
"vst1.8 d5, [%1], %2\n"
"vst1.8 d6, [%1], %2\n"
"vst1.8 d7, [%1]\n"
:
: "r"(gpu), "r"(cpu), "r"(cpu_stride)
: "q0", "q1", "q2", "q3");
return;
} else if (gpu_stride == 16) {
__asm__ volatile (
/* Load from the GPU in one shot, no interleave, to
* d0-d7.
*/
"vldm %0, {q0, q1, q2, q3};\n"
/* Store each 16-byte line in 2 parts to the cpu-side
* destination. (vld1 can only store one d-register
* at a time).
*/
"vst1.8 d0, [%1], %3\n"
"vst1.8 d1, [%2], %3\n"
"vst1.8 d2, [%1], %3\n"
"vst1.8 d3, [%2], %3\n"
"vst1.8 d4, [%1], %3\n"
"vst1.8 d5, [%2], %3\n"
"vst1.8 d6, [%1]\n"
"vst1.8 d7, [%2]\n"
:
: "r"(gpu), "r"(cpu), "r"(cpu + 8), "r"(cpu_stride)
: "q0", "q1", "q2", "q3");
return;
}
#elif defined (PIPE_ARCH_AARCH64)
if (gpu_stride == 8) {
__asm__ volatile (
/* Load from the GPU in one shot, no interleave, to
* d0-d7.
*/
"ld1 {v0.2d, v1.2d, v2.2d, v3.2d}, [%0]\n"
/* Store each 8-byte line to cpu-side destination,
* incrementing it by the stride each time.
*/
"st1 {v0.D}[0], [%1], %2\n"
"st1 {v0.D}[1], [%1], %2\n"
"st1 {v1.D}[0], [%1], %2\n"
"st1 {v1.D}[1], [%1], %2\n"
"st1 {v2.D}[0], [%1], %2\n"
"st1 {v2.D}[1], [%1], %2\n"
"st1 {v3.D}[0], [%1], %2\n"
"st1 {v3.D}[1], [%1]\n"
:
: "r"(gpu), "r"(cpu), "r"(cpu_stride)
: "v0", "v1", "v2", "v3");
return;
} else if (gpu_stride == 16) {
__asm__ volatile (
/* Load from the GPU in one shot, no interleave, to
* d0-d7.
*/
"ld1 {v0.2d, v1.2d, v2.2d, v3.2d}, [%0]\n"
/* Store each 16-byte line in 2 parts to the cpu-side
* destination. (vld1 can only store one d-register
* at a time).
*/
"st1 {v0.D}[0], [%1], %3\n"
"st1 {v0.D}[1], [%2], %3\n"
"st1 {v1.D}[0], [%1], %3\n"
"st1 {v1.D}[1], [%2], %3\n"
"st1 {v2.D}[0], [%1], %3\n"
"st1 {v2.D}[1], [%2], %3\n"
"st1 {v3.D}[0], [%1]\n"
"st1 {v3.D}[1], [%2]\n"
:
: "r"(gpu), "r"(cpu), "r"(cpu + 8), "r"(cpu_stride)
: "v0", "v1", "v2", "v3");
return;
}
#endif
for (uint32_t gpu_offset = 0; gpu_offset < 64; gpu_offset += gpu_stride) {
memcpy(cpu, gpu + gpu_offset, gpu_stride);
cpu += cpu_stride;
}
}
static inline void
v3d_store_utile(void *gpu, uint32_t gpu_stride,
void *cpu, uint32_t cpu_stride)
{
#if defined(V3D_BUILD_NEON) && defined(PIPE_ARCH_ARM)
if (gpu_stride == 8) {
__asm__ volatile (
/* Load each 8-byte line from cpu-side source,
* incrementing it by the stride each time.
*/
"vld1.8 d0, [%1], %2\n"
"vld1.8 d1, [%1], %2\n"
"vld1.8 d2, [%1], %2\n"
"vld1.8 d3, [%1], %2\n"
"vld1.8 d4, [%1], %2\n"
"vld1.8 d5, [%1], %2\n"
"vld1.8 d6, [%1], %2\n"
"vld1.8 d7, [%1]\n"
/* Load from the GPU in one shot, no interleave, to
* d0-d7.
*/
"vstm %0, {q0, q1, q2, q3}\n"
:
: "r"(gpu), "r"(cpu), "r"(cpu_stride)
: "q0", "q1", "q2", "q3");
return;
} else if (gpu_stride == 16) {
__asm__ volatile (
/* Load each 16-byte line in 2 parts from the cpu-side
* destination. (vld1 can only store one d-register
* at a time).
*/
"vld1.8 d0, [%1], %3\n"
"vld1.8 d1, [%2], %3\n"
"vld1.8 d2, [%1], %3\n"
"vld1.8 d3, [%2], %3\n"
"vld1.8 d4, [%1], %3\n"
"vld1.8 d5, [%2], %3\n"
"vld1.8 d6, [%1]\n"
"vld1.8 d7, [%2]\n"
/* Store to the GPU in one shot, no interleave. */
"vstm %0, {q0, q1, q2, q3}\n"
:
: "r"(gpu), "r"(cpu), "r"(cpu + 8), "r"(cpu_stride)
: "q0", "q1", "q2", "q3");
return;
}
#elif defined (PIPE_ARCH_AARCH64)
if (gpu_stride == 8) {
__asm__ volatile (
/* Load each 8-byte line from cpu-side source,
* incrementing it by the stride each time.
*/
"ld1 {v0.D}[0], [%1], %2\n"
"ld1 {v0.D}[1], [%1], %2\n"
"ld1 {v1.D}[0], [%1], %2\n"
"ld1 {v1.D}[1], [%1], %2\n"
"ld1 {v2.D}[0], [%1], %2\n"
"ld1 {v2.D}[1], [%1], %2\n"
"ld1 {v3.D}[0], [%1], %2\n"
"ld1 {v3.D}[1], [%1]\n"
/* Store to the GPU in one shot, no interleave. */
"st1 {v0.2d, v1.2d, v2.2d, v3.2d}, [%0]\n"
:
: "r"(gpu), "r"(cpu), "r"(cpu_stride)
: "v0", "v1", "v2", "v3");
return;
} else if (gpu_stride == 16) {
__asm__ volatile (
/* Load each 16-byte line in 2 parts from the cpu-side
* destination. (vld1 can only store one d-register
* at a time).
*/
"ld1 {v0.D}[0], [%1], %3\n"
"ld1 {v0.D}[1], [%2], %3\n"
"ld1 {v1.D}[0], [%1], %3\n"
"ld1 {v1.D}[1], [%2], %3\n"
"ld1 {v2.D}[0], [%1], %3\n"
"ld1 {v2.D}[1], [%2], %3\n"
"ld1 {v3.D}[0], [%1]\n"
"ld1 {v3.D}[1], [%2]\n"
/* Store to the GPU in one shot, no interleave. */
"st1 {v0.2d, v1.2d, v2.2d, v3.2d}, [%0]\n"
:
: "r"(gpu), "r"(cpu), "r"(cpu + 8), "r"(cpu_stride)
: "v0", "v1", "v2", "v3");
return;
}
#endif
for (uint32_t gpu_offset = 0; gpu_offset < 64; gpu_offset += gpu_stride) {
memcpy(gpu + gpu_offset, cpu, gpu_stride);
cpu += cpu_stride;
}
}
|