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
|
#include "util/u_inlines.h"
#include "util/u_math.h"
#include "util/u_memory.h"
#include "util/u_pstipple.h"
#include "pipe/p_shader_tokens.h"
#include "draw/draw_context.h"
#include "draw/draw_vertex.h"
#include "sp_context.h"
#include "sp_screen.h"
#include "sp_state.h"
#include "sp_texture.h"
#include "sp_tex_sample.h"
#include "sp_tex_tile_cache.h"
#include "tgsi/tgsi_parse.h"
static void
cs_prepare(const struct sp_compute_shader *cs,
struct tgsi_exec_machine *machine,
int w, int h, int d,
int g_w, int g_h, int g_d,
int b_w, int b_h, int b_d,
struct tgsi_sampler *sampler,
struct tgsi_image *image,
struct tgsi_buffer *buffer )
{
int j;
/*
* Bind tokens/shader to the interpreter's machine state.
*/
tgsi_exec_machine_bind_shader(machine,
cs->tokens,
sampler, image, buffer);
if (machine->SysSemanticToIndex[TGSI_SEMANTIC_THREAD_ID] != -1) {
unsigned i = machine->SysSemanticToIndex[TGSI_SEMANTIC_THREAD_ID];
for (j = 0; j < TGSI_QUAD_SIZE; j++) {
machine->SystemValue[i].xyzw[0].i[j] = w;
machine->SystemValue[i].xyzw[1].i[j] = h;
machine->SystemValue[i].xyzw[2].i[j] = d;
}
}
if (machine->SysSemanticToIndex[TGSI_SEMANTIC_GRID_SIZE] != -1) {
unsigned i = machine->SysSemanticToIndex[TGSI_SEMANTIC_GRID_SIZE];
for (j = 0; j < TGSI_QUAD_SIZE; j++) {
machine->SystemValue[i].xyzw[0].i[j] = g_w;
machine->SystemValue[i].xyzw[1].i[j] = g_h;
machine->SystemValue[i].xyzw[2].i[j] = g_d;
}
}
if (machine->SysSemanticToIndex[TGSI_SEMANTIC_BLOCK_SIZE] != -1) {
unsigned i = machine->SysSemanticToIndex[TGSI_SEMANTIC_BLOCK_SIZE];
for (j = 0; j < TGSI_QUAD_SIZE; j++) {
machine->SystemValue[i].xyzw[0].i[j] = b_w;
machine->SystemValue[i].xyzw[1].i[j] = b_h;
machine->SystemValue[i].xyzw[2].i[j] = b_d;
}
}
}
static bool
cs_run(const struct sp_compute_shader *cs,
int g_w, int g_h, int g_d,
struct tgsi_exec_machine *machine, bool restart)
{
if (!restart) {
if (machine->SysSemanticToIndex[TGSI_SEMANTIC_BLOCK_ID] != -1) {
unsigned i = machine->SysSemanticToIndex[TGSI_SEMANTIC_BLOCK_ID];
int j;
for (j = 0; j < TGSI_QUAD_SIZE; j++) {
machine->SystemValue[i].xyzw[0].i[j] = g_w;
machine->SystemValue[i].xyzw[1].i[j] = g_h;
machine->SystemValue[i].xyzw[2].i[j] = g_d;
}
}
machine->NonHelperMask = (1 << 1) - 1;
}
tgsi_exec_machine_run(machine, restart ? machine->pc : 0);
if (machine->pc != -1)
return true;
return false;
}
static void
run_workgroup(const struct sp_compute_shader *cs,
int g_w, int g_h, int g_d, int num_threads,
struct tgsi_exec_machine **machines)
{
int i;
bool grp_hit_barrier, restart_threads = false;
do {
grp_hit_barrier = false;
for (i = 0; i < num_threads; i++) {
grp_hit_barrier |= cs_run(cs, g_w, g_h, g_d, machines[i], restart_threads);
}
restart_threads = false;
if (grp_hit_barrier) {
grp_hit_barrier = false;
restart_threads = true;
}
} while (restart_threads);
}
static void
cs_delete(const struct sp_compute_shader *cs,
struct tgsi_exec_machine *machine)
{
if (machine->Tokens == cs->tokens) {
tgsi_exec_machine_bind_shader(machine, NULL, NULL, NULL, NULL);
}
}
static void
fill_grid_size(struct pipe_context *context,
const struct pipe_grid_info *info,
uint32_t grid_size[3])
{
struct pipe_transfer *transfer;
uint32_t *params;
if (!info->indirect) {
grid_size[0] = info->grid[0];
grid_size[1] = info->grid[1];
grid_size[2] = info->grid[2];
return;
}
params = pipe_buffer_map_range(context, info->indirect,
info->indirect_offset,
3 * sizeof(uint32_t),
PIPE_TRANSFER_READ,
&transfer);
if (!transfer)
return;
grid_size[0] = params[0];
grid_size[1] = params[1];
grid_size[2] = params[2];
pipe_buffer_unmap(context, transfer);
}
void
softpipe_launch_grid(struct pipe_context *context,
const struct pipe_grid_info *info)
{
struct softpipe_context *softpipe = softpipe_context(context);
struct sp_compute_shader *cs = softpipe->cs;
int num_threads_in_group;
struct tgsi_exec_machine **machines;
int bwidth, bheight, bdepth;
int w, h, d, i;
int g_w, g_h, g_d;
uint32_t grid_size[3];
void *local_mem = NULL;
bwidth = cs->info.properties[TGSI_PROPERTY_CS_FIXED_BLOCK_WIDTH];
bheight = cs->info.properties[TGSI_PROPERTY_CS_FIXED_BLOCK_HEIGHT];
bdepth = cs->info.properties[TGSI_PROPERTY_CS_FIXED_BLOCK_DEPTH];
num_threads_in_group = bwidth * bheight * bdepth;
fill_grid_size(context, info, grid_size);
if (cs->shader.req_local_mem) {
local_mem = CALLOC(1, cs->shader.req_local_mem);
}
machines = CALLOC(sizeof(struct tgsi_exec_machine *), num_threads_in_group);
if (!machines)
return;
/* initialise machines + GRID_SIZE + THREAD_ID + BLOCK_SIZE */
for (d = 0; d < bdepth; d++) {
for (h = 0; h < bheight; h++) {
for (w = 0; w < bwidth; w++) {
int idx = w + (h * bwidth) + (d * bheight * bwidth);
machines[idx] = tgsi_exec_machine_create(PIPE_SHADER_COMPUTE);
machines[idx]->LocalMem = local_mem;
machines[idx]->LocalMemSize = cs->shader.req_local_mem;
cs_prepare(cs, machines[idx],
w, h, d,
grid_size[0], grid_size[1], grid_size[2],
bwidth, bheight, bdepth,
(struct tgsi_sampler *)softpipe->tgsi.sampler[PIPE_SHADER_COMPUTE],
(struct tgsi_image *)softpipe->tgsi.image[PIPE_SHADER_COMPUTE],
(struct tgsi_buffer *)softpipe->tgsi.buffer[PIPE_SHADER_COMPUTE]);
tgsi_exec_set_constant_buffers(machines[idx], PIPE_MAX_CONSTANT_BUFFERS,
softpipe->mapped_constants[PIPE_SHADER_COMPUTE],
softpipe->const_buffer_size[PIPE_SHADER_COMPUTE]);
}
}
}
for (g_d = 0; g_d < grid_size[2]; g_d++) {
for (g_h = 0; g_h < grid_size[1]; g_h++) {
for (g_w = 0; g_w < grid_size[0]; g_w++) {
run_workgroup(cs, g_w, g_h, g_d, num_threads_in_group, machines);
}
}
}
for (i = 0; i < num_threads_in_group; i++) {
cs_delete(cs, machines[i]);
tgsi_exec_machine_destroy(machines[i]);
}
FREE(local_mem);
FREE(machines);
}
|
