2 * Copyright (C) 2011 Francisco Jerez.
5 * Permission is hereby granted, free of charge, to any person obtaining
6 * a copy of this software and associated documentation files (the
7 * "Software"), to deal in the Software without restriction, including
8 * without limitation the rights to use, copy, modify, merge, publish,
9 * distribute, sublicense, and/or sell copies of the Software, and to
10 * permit persons to whom the Software is furnished to do so, subject to
11 * the following conditions:
13 * The above copyright notice and this permission notice (including the
14 * next paragraph) shall be included in all copies or substantial
15 * portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
18 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
19 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
20 * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
21 * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
22 * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
23 * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
31 #include "pipe/p_state.h"
32 #include "pipe/p_context.h"
33 #include "pipe/p_screen.h"
34 #include "pipe/p_defines.h"
35 #include "pipe/p_shader_tokens.h"
36 #include "util/u_memory.h"
37 #include "util/u_inlines.h"
38 #include "util/u_sampler.h"
39 #include "util/u_format.h"
40 #include "tgsi/tgsi_text.h"
41 #include "pipe-loader/pipe_loader.h"
43 #define MAX_RESOURCES 4
46 struct pipe_loader_device
*dev
;
47 struct pipe_screen
*screen
;
48 struct pipe_context
*pipe
;
50 void *hwsmp
[MAX_RESOURCES
];
51 struct pipe_resource
*tex
[MAX_RESOURCES
];
52 bool tex_rw
[MAX_RESOURCES
];
53 struct pipe_sampler_view
*view
[MAX_RESOURCES
];
54 struct pipe_surface
*surf
[MAX_RESOURCES
];
57 #define DUMP_COMPUTE_PARAM(p, c) do { \
61 __n = ctx->screen->get_compute_param(ctx->screen, c, __v); \
62 printf("%s: {", #c); \
64 for (__i = 0; __i < __n / sizeof(*__v); ++__i) \
65 printf(" %"PRIu64, __v[__i]); \
70 static void init_ctx(struct context
*ctx
)
74 ret
= pipe_loader_probe(&ctx
->dev
, 1);
77 ctx
->screen
= pipe_loader_create_screen(ctx
->dev
, PIPE_SEARCH_DIR
);
80 ctx
->pipe
= ctx
->screen
->context_create(ctx
->screen
, NULL
);
83 DUMP_COMPUTE_PARAM(p
, PIPE_COMPUTE_CAP_GRID_DIMENSION
);
84 DUMP_COMPUTE_PARAM(p
, PIPE_COMPUTE_CAP_MAX_GRID_SIZE
);
85 DUMP_COMPUTE_PARAM(p
, PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE
);
88 static void destroy_ctx(struct context
*ctx
)
90 ctx
->pipe
->destroy(ctx
->pipe
);
91 ctx
->screen
->destroy(ctx
->screen
);
92 pipe_loader_release(&ctx
->dev
, 1);
97 preprocess_prog(struct context
*ctx
, const char *src
, const char *defs
)
100 "#define RGLOBAL RES[32767]\n"
101 "#define RLOCAL RES[32766]\n"
102 "#define RPRIVATE RES[32765]\n"
103 "#define RINPUT RES[32764]\n";
105 char tmp
[] = "/tmp/test-compute.tgsi-XXXXXX";
111 /* Open a temporary file */
114 snprintf(cmd
, sizeof(cmd
), "cpp -P -nostdinc -undef %s > %s",
115 defs
? defs
: "", tmp
);
119 fwrite(header
, strlen(header
), 1, p
);
120 fwrite(src
, strlen(src
), 1, p
);
125 ret
= fstat(fd
, &st
);
128 buf
= malloc(st
.st_size
+ 1);
129 ret
= read(fd
, buf
, st
.st_size
);
130 assert(ret
== st
.st_size
);
140 static void init_prog(struct context
*ctx
, unsigned local_sz
,
141 unsigned private_sz
, unsigned input_sz
,
142 const char *src
, const char *defs
)
144 struct pipe_context
*pipe
= ctx
->pipe
;
145 struct tgsi_token prog
[1024];
146 struct pipe_compute_state cs
= {
148 .req_local_mem
= local_sz
,
149 .req_private_mem
= private_sz
,
150 .req_input_mem
= input_sz
152 char *psrc
= preprocess_prog(ctx
, src
, defs
);
155 ret
= tgsi_text_translate(psrc
, prog
, Elements(prog
));
159 ctx
->hwcs
= pipe
->create_compute_state(pipe
, &cs
);
162 pipe
->bind_compute_state(pipe
, ctx
->hwcs
);
165 static void destroy_prog(struct context
*ctx
)
167 struct pipe_context
*pipe
= ctx
->pipe
;
169 pipe
->delete_compute_state(pipe
, ctx
->hwcs
);
173 static void init_tex(struct context
*ctx
, int slot
,
174 enum pipe_texture_target target
, bool rw
,
175 enum pipe_format format
, int w
, int h
,
176 void (*init
)(void *, int, int, int))
178 struct pipe_context
*pipe
= ctx
->pipe
;
179 struct pipe_resource
**tex
= &ctx
->tex
[slot
];
180 struct pipe_resource ttex
= {
187 .bind
= (PIPE_BIND_SAMPLER_VIEW
|
188 PIPE_BIND_COMPUTE_RESOURCE
|
191 int dx
= util_format_get_blocksize(format
);
192 int dy
= util_format_get_stride(format
, w
);
193 int nx
= (target
== PIPE_BUFFER
? (w
/ dx
) :
194 util_format_get_nblocksx(format
, w
));
195 int ny
= (target
== PIPE_BUFFER
? 1 :
196 util_format_get_nblocksy(format
, h
));
197 struct pipe_transfer
*xfer
;
201 *tex
= ctx
->screen
->resource_create(ctx
->screen
, &ttex
);
204 map
= pipe
->transfer_map(pipe
, *tex
, 0, PIPE_TRANSFER_WRITE
,
205 &(struct pipe_box
) { .width
= w
,
207 .depth
= 1 }, &xfer
);
211 for (y
= 0; y
< ny
; ++y
) {
212 for (x
= 0; x
< nx
; ++x
) {
213 init(map
+ y
* dy
+ x
* dx
, slot
, x
, y
);
217 pipe
->transfer_unmap(pipe
, xfer
);
219 ctx
->tex_rw
[slot
] = rw
;
222 static bool default_check(void *x
, void *y
, int sz
) {
223 return !memcmp(x
, y
, sz
);
226 static void check_tex(struct context
*ctx
, int slot
,
227 void (*expect
)(void *, int, int, int),
228 bool (*check
)(void *, void *, int))
230 struct pipe_context
*pipe
= ctx
->pipe
;
231 struct pipe_resource
*tex
= ctx
->tex
[slot
];
232 int dx
= util_format_get_blocksize(tex
->format
);
233 int dy
= util_format_get_stride(tex
->format
, tex
->width0
);
234 int nx
= (tex
->target
== PIPE_BUFFER
? (tex
->width0
/ dx
) :
235 util_format_get_nblocksx(tex
->format
, tex
->width0
));
236 int ny
= (tex
->target
== PIPE_BUFFER
? 1 :
237 util_format_get_nblocksy(tex
->format
, tex
->height0
));
238 struct pipe_transfer
*xfer
;
244 check
= default_check
;
246 map
= pipe
->transfer_map(pipe
, tex
, 0, PIPE_TRANSFER_READ
,
247 &(struct pipe_box
) { .width
= tex
->width0
,
248 .height
= tex
->height0
,
249 .depth
= 1 }, &xfer
);
253 for (y
= 0; y
< ny
; ++y
) {
254 for (x
= 0; x
< nx
; ++x
) {
256 uint32_t *res
= (uint32_t *)(map
+ y
* dy
+ x
* dx
);
258 expect(exp
, slot
, x
, y
);
259 if (check(res
, exp
, dx
) || (++err
) > 20)
263 uint32_t u
= 0, v
= 0;
265 for (i
= 0; i
< dx
; i
++) {
266 u
|= ((uint8_t *)exp
)[i
] << (8 * i
);
267 v
|= ((uint8_t *)res
)[i
] << (8 * i
);
269 printf("(%d, %d): got 0x%x, expected 0x%x\n",
272 for (i
= 0; i
< dx
/ 4; i
++) {
273 printf("(%d, %d)[%d]: got 0x%x/%f,"
274 " expected 0x%x/%f\n", x
, y
, i
,
275 res
[i
], ((float *)res
)[i
],
276 exp
[i
], ((float *)exp
)[i
]);
282 pipe
->transfer_unmap(pipe
, xfer
);
285 printf("(%d, %d): \x1b[31mFAIL\x1b[0m (%d)\n", x
, y
, err
);
287 printf("(%d, %d): \x1b[32mOK\x1b[0m\n", x
, y
);
290 static void destroy_tex(struct context
*ctx
)
294 for (i
= 0; i
< MAX_RESOURCES
; ++i
) {
296 pipe_resource_reference(&ctx
->tex
[i
], NULL
);
300 static void init_sampler_views(struct context
*ctx
, const int *slots
)
302 struct pipe_context
*pipe
= ctx
->pipe
;
303 struct pipe_sampler_view tview
;
306 for (i
= 0; *slots
>= 0; ++i
, ++slots
) {
307 u_sampler_view_default_template(&tview
, ctx
->tex
[*slots
],
308 ctx
->tex
[*slots
]->format
);
310 ctx
->view
[i
] = pipe
->create_sampler_view(pipe
, ctx
->tex
[*slots
],
312 assert(ctx
->view
[i
]);
315 pipe
->set_sampler_views(pipe
, PIPE_SHADER_COMPUTE
, 0, i
, ctx
->view
);
318 static void destroy_sampler_views(struct context
*ctx
)
320 struct pipe_context
*pipe
= ctx
->pipe
;
323 pipe
->set_sampler_views(pipe
, PIPE_SHADER_COMPUTE
, 0, MAX_RESOURCES
, NULL
);
325 for (i
= 0; i
< MAX_RESOURCES
; ++i
) {
327 pipe
->sampler_view_destroy(pipe
, ctx
->view
[i
]);
333 static void init_compute_resources(struct context
*ctx
, const int *slots
)
335 struct pipe_context
*pipe
= ctx
->pipe
;
338 for (i
= 0; *slots
>= 0; ++i
, ++slots
) {
339 struct pipe_surface tsurf
= {
340 .format
= ctx
->tex
[*slots
]->format
,
341 .writable
= ctx
->tex_rw
[*slots
]
344 if (ctx
->tex
[*slots
]->target
== PIPE_BUFFER
)
345 tsurf
.u
.buf
.last_element
= ctx
->tex
[*slots
]->width0
- 1;
347 ctx
->surf
[i
] = pipe
->create_surface(pipe
, ctx
->tex
[*slots
],
349 assert(ctx
->surf
[i
]);
352 pipe
->set_compute_resources(pipe
, 0, i
, ctx
->surf
);
355 static void destroy_compute_resources(struct context
*ctx
)
357 struct pipe_context
*pipe
= ctx
->pipe
;
360 pipe
->set_compute_resources(pipe
, 0, MAX_RESOURCES
, NULL
);
362 for (i
= 0; i
< MAX_RESOURCES
; ++i
) {
364 pipe
->surface_destroy(pipe
, ctx
->surf
[i
]);
370 static void init_sampler_states(struct context
*ctx
, int n
)
372 struct pipe_context
*pipe
= ctx
->pipe
;
373 struct pipe_sampler_state smp
= {
374 .normalized_coords
= 1,
378 for (i
= 0; i
< n
; ++i
) {
379 ctx
->hwsmp
[i
] = pipe
->create_sampler_state(pipe
, &smp
);
380 assert(ctx
->hwsmp
[i
]);
383 pipe
->bind_sampler_states(pipe
, PIPE_SHADER_COMPUTE
, 0, i
, ctx
->hwsmp
);
386 static void destroy_sampler_states(struct context
*ctx
)
388 struct pipe_context
*pipe
= ctx
->pipe
;
391 pipe
->bind_sampler_states(pipe
, PIPE_SHADER_COMPUTE
,
392 0, MAX_RESOURCES
, NULL
);
394 for (i
= 0; i
< MAX_RESOURCES
; ++i
) {
396 pipe
->delete_sampler_state(pipe
, ctx
->hwsmp
[i
]);
397 ctx
->hwsmp
[i
] = NULL
;
402 static void init_globals(struct context
*ctx
, const int *slots
,
405 struct pipe_context
*pipe
= ctx
->pipe
;
406 struct pipe_resource
*res
[MAX_RESOURCES
];
409 for (i
= 0; *slots
>= 0; ++i
, ++slots
)
410 res
[i
] = ctx
->tex
[*slots
];
412 pipe
->set_global_binding(pipe
, 0, i
, res
, handles
);
415 static void destroy_globals(struct context
*ctx
)
417 struct pipe_context
*pipe
= ctx
->pipe
;
419 pipe
->set_global_binding(pipe
, 0, MAX_RESOURCES
, NULL
, NULL
);
422 static void launch_grid(struct context
*ctx
, const uint
*block_layout
,
423 const uint
*grid_layout
, uint32_t pc
,
426 struct pipe_context
*pipe
= ctx
->pipe
;
428 pipe
->launch_grid(pipe
, block_layout
, grid_layout
, pc
, input
);
431 static void test_system_values(struct context
*ctx
)
433 const char *src
= "COMP\n"
434 "DCL RES[0], BUFFER, RAW, WR\n"
435 "DCL SV[0], BLOCK_ID[0]\n"
436 "DCL SV[1], BLOCK_SIZE[0]\n"
437 "DCL SV[2], GRID_SIZE[0]\n"
438 "DCL SV[3], THREAD_ID[0]\n"
439 "DCL TEMP[0], LOCAL\n"
440 "DCL TEMP[1], LOCAL\n"
441 "IMM UINT32 { 64, 0, 0, 0 }\n"
442 "IMM UINT32 { 16, 0, 0, 0 }\n"
443 "IMM UINT32 { 0, 0, 0, 0 }\n"
446 " UMUL TEMP[0], SV[0], SV[1]\n"
447 " UADD TEMP[0], TEMP[0], SV[3]\n"
448 " UMUL TEMP[1], SV[1], SV[2]\n"
449 " UMUL TEMP[0].w, TEMP[0], TEMP[1].zzzz\n"
450 " UMUL TEMP[0].zw, TEMP[0], TEMP[1].yyyy\n"
451 " UMUL TEMP[0].yzw, TEMP[0], TEMP[1].xxxx\n"
452 " UADD TEMP[0].xy, TEMP[0].xyxy, TEMP[0].zwzw\n"
453 " UADD TEMP[0].x, TEMP[0].xxxx, TEMP[0].yyyy\n"
454 " UMUL TEMP[0].x, TEMP[0], IMM[0]\n"
455 " STORE RES[0].xyzw, TEMP[0], SV[0]\n"
456 " UADD TEMP[0].x, TEMP[0], IMM[1]\n"
457 " STORE RES[0].xyzw, TEMP[0], SV[1]\n"
458 " UADD TEMP[0].x, TEMP[0], IMM[1]\n"
459 " STORE RES[0].xyzw, TEMP[0], SV[2]\n"
460 " UADD TEMP[0].x, TEMP[0], IMM[1]\n"
461 " STORE RES[0].xyzw, TEMP[0], SV[3]\n"
464 void init(void *p
, int s
, int x
, int y
) {
465 *(uint32_t *)p
= 0xdeadbeef;
467 void expect(void *p
, int s
, int x
, int y
) {
468 int id
= x
/ 16, sv
= (x
% 16) / 4, c
= x
% 4;
469 int tid
[] = { id
% 20, (id
% 240) / 20, id
/ 240, 0 };
470 int bsz
[] = { 4, 3, 5, 1};
471 int gsz
[] = { 5, 4, 1, 1};
475 *(uint32_t *)p
= tid
[c
] / bsz
[c
];
478 *(uint32_t *)p
= bsz
[c
];
481 *(uint32_t *)p
= gsz
[c
];
484 *(uint32_t *)p
= tid
[c
] % bsz
[c
];
489 printf("- %s\n", __func__
);
491 init_prog(ctx
, 0, 0, 0, src
, NULL
);
492 init_tex(ctx
, 0, PIPE_BUFFER
, true, PIPE_FORMAT_R32_FLOAT
,
494 init_compute_resources(ctx
, (int []) { 0, -1 });
495 launch_grid(ctx
, (uint
[]){4, 3, 5}, (uint
[]){5, 4, 1}, 0, NULL
);
496 check_tex(ctx
, 0, expect
, NULL
);
497 destroy_compute_resources(ctx
);
502 static void test_resource_access(struct context
*ctx
)
504 const char *src
= "COMP\n"
505 "DCL RES[0], BUFFER, RAW, WR\n"
506 "DCL RES[1], 2D, RAW, WR\n"
507 "DCL SV[0], BLOCK_ID[0]\n"
508 "DCL TEMP[0], LOCAL\n"
509 "DCL TEMP[1], LOCAL\n"
510 "IMM UINT32 { 15, 0, 0, 0 }\n"
511 "IMM UINT32 { 16, 1, 0, 0 }\n"
514 " UADD TEMP[0].x, SV[0].xxxx, SV[0].yyyy\n"
515 " AND TEMP[0].x, TEMP[0], IMM[0]\n"
516 " UMUL TEMP[0].x, TEMP[0], IMM[1]\n"
517 " LOAD TEMP[0].xyzw, RES[0], TEMP[0]\n"
518 " UMUL TEMP[1], SV[0], IMM[1]\n"
519 " STORE RES[1].xyzw, TEMP[1], TEMP[0]\n"
522 void init0(void *p
, int s
, int x
, int y
) {
523 *(float *)p
= 8.0 - (float)x
;
525 void init1(void *p
, int s
, int x
, int y
) {
526 *(uint32_t *)p
= 0xdeadbeef;
528 void expect(void *p
, int s
, int x
, int y
) {
529 *(float *)p
= 8.0 - (float)((x
+ 4*y
) & 0x3f);
532 printf("- %s\n", __func__
);
534 init_prog(ctx
, 0, 0, 0, src
, NULL
);
535 init_tex(ctx
, 0, PIPE_BUFFER
, true, PIPE_FORMAT_R32_FLOAT
,
537 init_tex(ctx
, 1, PIPE_TEXTURE_2D
, true, PIPE_FORMAT_R32_FLOAT
,
539 init_compute_resources(ctx
, (int []) { 0, 1, -1 });
540 launch_grid(ctx
, (uint
[]){1, 1, 1}, (uint
[]){15, 12, 1}, 0, NULL
);
541 check_tex(ctx
, 1, expect
, NULL
);
542 destroy_compute_resources(ctx
);
547 static void test_function_calls(struct context
*ctx
)
549 const char *src
= "COMP\n"
550 "DCL RES[0], 2D, RAW, WR\n"
551 "DCL SV[0], BLOCK_ID[0]\n"
552 "DCL SV[1], BLOCK_SIZE[0]\n"
553 "DCL SV[2], GRID_SIZE[0]\n"
554 "DCL SV[3], THREAD_ID[0]\n"
557 "DCL TEMP[2], LOCAL\n"
558 "IMM UINT32 { 0, 11, 22, 33 }\n"
559 "IMM FLT32 { 11, 33, 55, 99 }\n"
560 "IMM UINT32 { 4, 1, 0, 0 }\n"
561 "IMM UINT32 { 12, 0, 0, 0 }\n"
564 "01: UMUL TEMP[0].x, TEMP[0], TEMP[0]\n"
565 "02: UADD TEMP[1].x, TEMP[1], IMM[2].yyyy\n"
566 "03: USLT TEMP[0].x, TEMP[0], IMM[0]\n"
570 "07: UMUL TEMP[0].x, TEMP[0], TEMP[0]\n"
571 "08: UADD TEMP[1].x, TEMP[1], IMM[2].yyyy\n"
572 "09: USLT TEMP[0].x, TEMP[0], IMM[0].yyyy\n"
573 "10: IF TEMP[0].xxxx\n"
579 "16: UMUL TEMP[2], SV[0], SV[1]\n"
580 "17: UADD TEMP[2], TEMP[2], SV[3]\n"
581 "18: UMUL TEMP[2], TEMP[2], IMM[2]\n"
582 "00: MOV TEMP[1].x, IMM[2].wwww\n"
583 "19: LOAD TEMP[0].x, RES[0].xxxx, TEMP[2]\n"
585 "21: STORE RES[0].x, TEMP[2], TEMP[1].xxxx\n"
588 void init(void *p
, int s
, int x
, int y
) {
589 *(uint32_t *)p
= 15 * y
+ x
;
591 void expect(void *p
, int s
, int x
, int y
) {
592 *(uint32_t *)p
= (15 * y
+ x
) < 4 ? 2 : 1 ;
595 printf("- %s\n", __func__
);
597 init_prog(ctx
, 0, 0, 0, src
, NULL
);
598 init_tex(ctx
, 0, PIPE_TEXTURE_2D
, true, PIPE_FORMAT_R32_FLOAT
,
600 init_compute_resources(ctx
, (int []) { 0, -1 });
601 launch_grid(ctx
, (uint
[]){3, 3, 3}, (uint
[]){5, 4, 1}, 15, NULL
);
602 check_tex(ctx
, 0, expect
, NULL
);
603 destroy_compute_resources(ctx
);
608 static void test_input_global(struct context
*ctx
)
610 const char *src
= "COMP\n"
611 "DCL SV[0], THREAD_ID[0]\n"
612 "DCL TEMP[0], LOCAL\n"
613 "DCL TEMP[1], LOCAL\n"
614 "IMM UINT32 { 8, 0, 0, 0 }\n"
617 " UMUL TEMP[0], SV[0], IMM[0]\n"
618 " LOAD TEMP[1].xy, RINPUT, TEMP[0]\n"
619 " LOAD TEMP[0].x, RGLOBAL, TEMP[1].yyyy\n"
620 " UADD TEMP[1].x, TEMP[0], -TEMP[1]\n"
621 " STORE RGLOBAL.x, TEMP[1].yyyy, TEMP[1]\n"
624 void init(void *p
, int s
, int x
, int y
) {
625 *(uint32_t *)p
= 0xdeadbeef;
627 void expect(void *p
, int s
, int x
, int y
) {
628 *(uint32_t *)p
= 0xdeadbeef - (x
== 0 ? 0x10001 + 2 * s
: 0);
630 uint32_t input
[8] = { 0x10001, 0x10002, 0x10003, 0x10004,
631 0x10005, 0x10006, 0x10007, 0x10008 };
633 printf("- %s\n", __func__
);
635 init_prog(ctx
, 0, 0, 32, src
, NULL
);
636 init_tex(ctx
, 0, PIPE_BUFFER
, true, PIPE_FORMAT_R32_FLOAT
, 32, 0, init
);
637 init_tex(ctx
, 1, PIPE_BUFFER
, true, PIPE_FORMAT_R32_FLOAT
, 32, 0, init
);
638 init_tex(ctx
, 2, PIPE_BUFFER
, true, PIPE_FORMAT_R32_FLOAT
, 32, 0, init
);
639 init_tex(ctx
, 3, PIPE_BUFFER
, true, PIPE_FORMAT_R32_FLOAT
, 32, 0, init
);
640 init_globals(ctx
, (int []){ 0, 1, 2, 3, -1 },
641 (uint32_t *[]){ &input
[1], &input
[3],
642 &input
[5], &input
[7] });
643 launch_grid(ctx
, (uint
[]){4, 1, 1}, (uint
[]){1, 1, 1}, 0, input
);
644 check_tex(ctx
, 0, expect
, NULL
);
645 check_tex(ctx
, 1, expect
, NULL
);
646 check_tex(ctx
, 2, expect
, NULL
);
647 check_tex(ctx
, 3, expect
, NULL
);
648 destroy_globals(ctx
);
653 static void test_private(struct context
*ctx
)
655 const char *src
= "COMP\n"
656 "DCL RES[0], BUFFER, RAW, WR\n"
657 "DCL SV[0], BLOCK_ID[0]\n"
658 "DCL SV[1], BLOCK_SIZE[0]\n"
659 "DCL SV[2], THREAD_ID[0]\n"
660 "DCL TEMP[0], LOCAL\n"
661 "DCL TEMP[1], LOCAL\n"
662 "DCL TEMP[2], LOCAL\n"
663 "IMM UINT32 { 128, 0, 0, 0 }\n"
664 "IMM UINT32 { 4, 0, 0, 0 }\n"
667 " UMUL TEMP[0].x, SV[0], SV[1]\n"
668 " UADD TEMP[0].x, TEMP[0], SV[2]\n"
669 " MOV TEMP[1].x, IMM[0].wwww\n"
671 " USEQ TEMP[2].x, TEMP[1], IMM[0]\n"
675 " UDIV TEMP[2].x, TEMP[1], IMM[1]\n"
676 " UADD TEMP[2].x, TEMP[2], TEMP[0]\n"
677 " STORE RPRIVATE.x, TEMP[1], TEMP[2]\n"
678 " UADD TEMP[1].x, TEMP[1], IMM[1]\n"
680 " MOV TEMP[1].x, IMM[0].wwww\n"
681 " UMUL TEMP[0].x, TEMP[0], IMM[0]\n"
683 " USEQ TEMP[2].x, TEMP[1], IMM[0]\n"
687 " LOAD TEMP[2].x, RPRIVATE, TEMP[1]\n"
688 " STORE RES[0].x, TEMP[0], TEMP[2]\n"
689 " UADD TEMP[0].x, TEMP[0], IMM[1]\n"
690 " UADD TEMP[1].x, TEMP[1], IMM[1]\n"
694 void init(void *p
, int s
, int x
, int y
) {
695 *(uint32_t *)p
= 0xdeadbeef;
697 void expect(void *p
, int s
, int x
, int y
) {
698 *(uint32_t *)p
= (x
/ 32) + x
% 32;
701 printf("- %s\n", __func__
);
703 init_prog(ctx
, 0, 128, 0, src
, NULL
);
704 init_tex(ctx
, 0, PIPE_BUFFER
, true, PIPE_FORMAT_R32_FLOAT
,
706 init_compute_resources(ctx
, (int []) { 0, -1 });
707 launch_grid(ctx
, (uint
[]){16, 1, 1}, (uint
[]){16, 1, 1}, 0, NULL
);
708 check_tex(ctx
, 0, expect
, NULL
);
709 destroy_compute_resources(ctx
);
714 static void test_local(struct context
*ctx
)
716 const char *src
= "COMP\n"
717 "DCL RES[0], BUFFER, RAW, WR\n"
718 "DCL SV[0], BLOCK_ID[0]\n"
719 "DCL SV[1], BLOCK_SIZE[0]\n"
720 "DCL SV[2], THREAD_ID[0]\n"
721 "DCL TEMP[0], LOCAL\n"
722 "DCL TEMP[1], LOCAL\n"
723 "DCL TEMP[2], LOCAL\n"
724 "IMM UINT32 { 1, 0, 0, 0 }\n"
725 "IMM UINT32 { 2, 0, 0, 0 }\n"
726 "IMM UINT32 { 4, 0, 0, 0 }\n"
727 "IMM UINT32 { 32, 0, 0, 0 }\n"
728 "IMM UINT32 { 128, 0, 0, 0 }\n"
731 " UMUL TEMP[0].x, SV[2], IMM[2]\n"
732 " STORE RLOCAL.x, TEMP[0], IMM[0].wwww\n"
734 " USLT TEMP[1].x, SV[2], IMM[3]\n"
736 " UADD TEMP[1].x, TEMP[0], IMM[4]\n"
738 " LOAD TEMP[2].x, RLOCAL, TEMP[1]\n"
739 " USEQ TEMP[2].x, TEMP[2], IMM[0]\n"
744 " STORE RLOCAL.x, TEMP[0], IMM[0]\n"
747 " LOAD TEMP[2].x, RLOCAL, TEMP[1]\n"
748 " USEQ TEMP[2].x, TEMP[2], IMM[1]\n"
754 " UADD TEMP[1].x, TEMP[0], -IMM[4]\n"
756 " LOAD TEMP[2].x, RLOCAL, TEMP[1]\n"
757 " USEQ TEMP[2].x, TEMP[2], IMM[0].wwww\n"
762 " STORE RLOCAL.x, TEMP[0], IMM[0]\n"
765 " LOAD TEMP[2].x, RLOCAL, TEMP[1]\n"
766 " USEQ TEMP[2].x, TEMP[2], IMM[0]\n"
771 " STORE RLOCAL.x, TEMP[0], IMM[1]\n"
774 " UMUL TEMP[1].x, SV[0], SV[1]\n"
775 " UMUL TEMP[1].x, TEMP[1], IMM[2]\n"
776 " UADD TEMP[1].x, TEMP[1], TEMP[0]\n"
777 " LOAD TEMP[0].x, RLOCAL, TEMP[0]\n"
778 " STORE RES[0].x, TEMP[1], TEMP[0]\n"
781 void init(void *p
, int s
, int x
, int y
) {
782 *(uint32_t *)p
= 0xdeadbeef;
784 void expect(void *p
, int s
, int x
, int y
) {
785 *(uint32_t *)p
= x
& 0x20 ? 2 : 1;
788 printf("- %s\n", __func__
);
790 init_prog(ctx
, 256, 0, 0, src
, NULL
);
791 init_tex(ctx
, 0, PIPE_BUFFER
, true, PIPE_FORMAT_R32_FLOAT
,
793 init_compute_resources(ctx
, (int []) { 0, -1 });
794 launch_grid(ctx
, (uint
[]){64, 1, 1}, (uint
[]){16, 1, 1}, 0, NULL
);
795 check_tex(ctx
, 0, expect
, NULL
);
796 destroy_compute_resources(ctx
);
801 static void test_sample(struct context
*ctx
)
803 const char *src
= "COMP\n"
804 "DCL SVIEW[0], 2D, FLOAT\n"
805 "DCL RES[0], 2D, RAW, WR\n"
807 "DCL SV[0], BLOCK_ID[0]\n"
808 "DCL TEMP[0], LOCAL\n"
809 "DCL TEMP[1], LOCAL\n"
810 "IMM UINT32 { 16, 1, 0, 0 }\n"
811 "IMM FLT32 { 128, 32, 0, 0 }\n"
814 " I2F TEMP[1], SV[0]\n"
815 " DIV TEMP[1], TEMP[1], IMM[1]\n"
816 " SAMPLE TEMP[1], TEMP[1], SVIEW[0], SAMP[0]\n"
817 " UMUL TEMP[0], SV[0], IMM[0]\n"
818 " STORE RES[0].xyzw, TEMP[0], TEMP[1]\n"
821 void init(void *p
, int s
, int x
, int y
) {
822 *(float *)p
= s
? 1 : x
* y
;
824 void expect(void *p
, int s
, int x
, int y
) {
827 *(float *)p
= x
/ 4 * y
;
839 printf("- %s\n", __func__
);
841 init_prog(ctx
, 0, 0, 0, src
, NULL
);
842 init_tex(ctx
, 0, PIPE_TEXTURE_2D
, true, PIPE_FORMAT_R32_FLOAT
,
844 init_tex(ctx
, 1, PIPE_TEXTURE_2D
, true, PIPE_FORMAT_R32_FLOAT
,
846 init_compute_resources(ctx
, (int []) { 1, -1 });
847 init_sampler_views(ctx
, (int []) { 0, -1 });
848 init_sampler_states(ctx
, 2);
849 launch_grid(ctx
, (uint
[]){1, 1, 1}, (uint
[]){128, 32, 1}, 0, NULL
);
850 check_tex(ctx
, 1, expect
, NULL
);
851 destroy_sampler_states(ctx
);
852 destroy_sampler_views(ctx
);
853 destroy_compute_resources(ctx
);
858 static void test_many_kern(struct context
*ctx
)
860 const char *src
= "COMP\n"
861 "DCL RES[0], BUFFER, RAW, WR\n"
862 "DCL TEMP[0], LOCAL\n"
863 "IMM UINT32 { 0, 1, 2, 3 }\n"
864 "IMM UINT32 { 4, 0, 0, 0 }\n"
867 " UMUL TEMP[0].x, IMM[0].xxxx, IMM[1].xxxx\n"
868 " STORE RES[0].x, TEMP[0], IMM[0].xxxx\n"
872 " UMUL TEMP[0].x, IMM[0].yyyy, IMM[1].xxxx\n"
873 " STORE RES[0].x, TEMP[0], IMM[0].yyyy\n"
877 " UMUL TEMP[0].x, IMM[0].zzzz, IMM[1].xxxx\n"
878 " STORE RES[0].x, TEMP[0], IMM[0].zzzz\n"
882 " UMUL TEMP[0].x, IMM[0].wwww, IMM[1].xxxx\n"
883 " STORE RES[0].x, TEMP[0], IMM[0].wwww\n"
886 void init(void *p
, int s
, int x
, int y
) {
887 *(uint32_t *)p
= 0xdeadbeef;
889 void expect(void *p
, int s
, int x
, int y
) {
893 printf("- %s\n", __func__
);
895 init_prog(ctx
, 0, 0, 0, src
, NULL
);
896 init_tex(ctx
, 0, PIPE_BUFFER
, true, PIPE_FORMAT_R32_FLOAT
,
898 init_compute_resources(ctx
, (int []) { 0, -1 });
899 launch_grid(ctx
, (uint
[]){1, 1, 1}, (uint
[]){1, 1, 1}, 0, NULL
);
900 launch_grid(ctx
, (uint
[]){1, 1, 1}, (uint
[]){1, 1, 1}, 5, NULL
);
901 launch_grid(ctx
, (uint
[]){1, 1, 1}, (uint
[]){1, 1, 1}, 10, NULL
);
902 launch_grid(ctx
, (uint
[]){1, 1, 1}, (uint
[]){1, 1, 1}, 15, NULL
);
903 check_tex(ctx
, 0, expect
, NULL
);
904 destroy_compute_resources(ctx
);
909 static void test_constant(struct context
*ctx
)
911 const char *src
= "COMP\n"
912 "DCL RES[0], BUFFER, RAW\n"
913 "DCL RES[1], BUFFER, RAW, WR\n"
914 "DCL SV[0], BLOCK_ID[0]\n"
915 "DCL TEMP[0], LOCAL\n"
916 "DCL TEMP[1], LOCAL\n"
917 "IMM UINT32 { 4, 0, 0, 0 }\n"
920 " UMUL TEMP[0].x, SV[0], IMM[0]\n"
921 " LOAD TEMP[1].x, RES[0], TEMP[0]\n"
922 " STORE RES[1].x, TEMP[0], TEMP[1]\n"
925 void init(void *p
, int s
, int x
, int y
) {
926 *(float *)p
= s
? 0xdeadbeef : 8.0 - (float)x
;
928 void expect(void *p
, int s
, int x
, int y
) {
929 *(float *)p
= 8.0 - (float)x
;
932 printf("- %s\n", __func__
);
934 init_prog(ctx
, 0, 0, 0, src
, NULL
);
935 init_tex(ctx
, 0, PIPE_BUFFER
, false, PIPE_FORMAT_R32_FLOAT
,
937 init_tex(ctx
, 1, PIPE_BUFFER
, true, PIPE_FORMAT_R32_FLOAT
,
939 init_compute_resources(ctx
, (int []) { 0, 1, -1 });
940 launch_grid(ctx
, (uint
[]){1, 1, 1}, (uint
[]){64, 1, 1}, 0, NULL
);
941 check_tex(ctx
, 1, expect
, NULL
);
942 destroy_compute_resources(ctx
);
947 static void test_resource_indirect(struct context
*ctx
)
949 const char *src
= "COMP\n"
950 "DCL RES[0], BUFFER, RAW, WR\n"
951 "DCL RES[1..3], BUFFER, RAW\n"
952 "DCL SV[0], BLOCK_ID[0]\n"
953 "DCL TEMP[0], LOCAL\n"
954 "DCL TEMP[1], LOCAL\n"
955 "IMM UINT32 { 4, 0, 0, 0 }\n"
958 " UMUL TEMP[0].x, SV[0], IMM[0]\n"
959 " LOAD TEMP[1].x, RES[1], TEMP[0]\n"
960 " LOAD TEMP[1].x, RES[TEMP[1].x+2], TEMP[0]\n"
961 " STORE RES[0].x, TEMP[0], TEMP[1]\n"
964 void init(void *p
, int s
, int x
, int y
) {
965 *(uint32_t *)p
= s
== 0 ? 0xdeadbeef :
970 void expect(void *p
, int s
, int x
, int y
) {
971 *(uint32_t *)p
= 2 * x
+ (x
% 2 ? 1 : 0);
974 printf("- %s\n", __func__
);
976 init_prog(ctx
, 0, 0, 0, src
, NULL
);
977 init_tex(ctx
, 0, PIPE_BUFFER
, true, PIPE_FORMAT_R32_FLOAT
,
979 init_tex(ctx
, 1, PIPE_BUFFER
, false, PIPE_FORMAT_R32_FLOAT
,
981 init_tex(ctx
, 2, PIPE_BUFFER
, false, PIPE_FORMAT_R32_FLOAT
,
983 init_tex(ctx
, 3, PIPE_BUFFER
, false, PIPE_FORMAT_R32_FLOAT
,
985 init_compute_resources(ctx
, (int []) { 0, 1, 2, 3, -1 });
986 launch_grid(ctx
, (uint
[]){1, 1, 1}, (uint
[]){64, 1, 1}, 0, NULL
);
987 check_tex(ctx
, 0, expect
, NULL
);
988 destroy_compute_resources(ctx
);
993 enum pipe_format surface_fmts
[] = {
994 PIPE_FORMAT_B8G8R8A8_UNORM
,
995 PIPE_FORMAT_B8G8R8X8_UNORM
,
996 PIPE_FORMAT_A8R8G8B8_UNORM
,
997 PIPE_FORMAT_X8R8G8B8_UNORM
,
998 PIPE_FORMAT_X8R8G8B8_UNORM
,
999 PIPE_FORMAT_L8_UNORM
,
1000 PIPE_FORMAT_A8_UNORM
,
1001 PIPE_FORMAT_I8_UNORM
,
1002 PIPE_FORMAT_L8A8_UNORM
,
1003 PIPE_FORMAT_R32_FLOAT
,
1004 PIPE_FORMAT_R32G32_FLOAT
,
1005 PIPE_FORMAT_R32G32B32A32_FLOAT
,
1006 PIPE_FORMAT_R32_UNORM
,
1007 PIPE_FORMAT_R32G32_UNORM
,
1008 PIPE_FORMAT_R32G32B32A32_UNORM
,
1009 PIPE_FORMAT_R32_SNORM
,
1010 PIPE_FORMAT_R32G32_SNORM
,
1011 PIPE_FORMAT_R32G32B32A32_SNORM
,
1012 PIPE_FORMAT_R8_UINT
,
1013 PIPE_FORMAT_R8G8_UINT
,
1014 PIPE_FORMAT_R8G8B8A8_UINT
,
1015 PIPE_FORMAT_R8_SINT
,
1016 PIPE_FORMAT_R8G8_SINT
,
1017 PIPE_FORMAT_R8G8B8A8_SINT
,
1018 PIPE_FORMAT_R32_UINT
,
1019 PIPE_FORMAT_R32G32_UINT
,
1020 PIPE_FORMAT_R32G32B32A32_UINT
,
1021 PIPE_FORMAT_R32_SINT
,
1022 PIPE_FORMAT_R32G32_SINT
,
1023 PIPE_FORMAT_R32G32B32A32_SINT
1026 static void test_surface_ld(struct context
*ctx
)
1028 const char *src
= "COMP\n"
1030 "DCL RES[1], 2D, RAW, WR\n"
1031 "DCL SV[0], BLOCK_ID[0]\n"
1032 "DCL TEMP[0], LOCAL\n"
1033 "DCL TEMP[1], LOCAL\n"
1034 "IMM UINT32 { 16, 1, 0, 0 }\n"
1037 " LOAD TEMP[1], RES[0], SV[0]\n"
1038 " UMUL TEMP[0], SV[0], IMM[0]\n"
1039 " STORE RES[1].xyzw, TEMP[0], TEMP[1]\n"
1043 void init0f(void *p
, int s
, int x
, int y
) {
1044 float v
[] = { 1.0, -.75, .50, -.25 };
1045 util_format_write_4f(surface_fmts
[i
], v
, 0,
1048 void init0i(void *p
, int s
, int x
, int y
) {
1049 int v
[] = { 0xffffffff, 0xffff, 0xff, 0xf };
1050 util_format_write_4i(surface_fmts
[i
], v
, 0,
1053 void init1(void *p
, int s
, int x
, int y
) {
1054 *(uint32_t *)p
= 0xdeadbeef;
1056 void expectf(void *p
, int s
, int x
, int y
) {
1058 init0f(v
, s
, x
/ 4, y
);
1059 util_format_read_4f(surface_fmts
[i
], w
, 0,
1061 *(float *)p
= w
[x
% 4];
1063 void expecti(void *p
, int s
, int x
, int y
) {
1065 init0i(v
, s
, x
/ 4, y
);
1066 util_format_read_4i(surface_fmts
[i
], w
, 0,
1068 *(uint32_t *)p
= w
[x
% 4];
1071 printf("- %s\n", __func__
);
1073 init_prog(ctx
, 0, 0, 0, src
, NULL
);
1075 for (i
= 0; i
< Elements(surface_fmts
); i
++) {
1076 bool is_int
= util_format_is_pure_integer(surface_fmts
[i
]);
1078 printf(" - %s\n", util_format_name(surface_fmts
[i
]));
1080 if (!ctx
->screen
->is_format_supported(ctx
->screen
,
1081 surface_fmts
[i
], PIPE_TEXTURE_2D
, 1,
1082 PIPE_BIND_COMPUTE_RESOURCE
)) {
1083 printf("(unsupported)\n");
1087 init_tex(ctx
, 0, PIPE_TEXTURE_2D
, true, surface_fmts
[i
],
1088 128, 32, (is_int
? init0i
: init0f
));
1089 init_tex(ctx
, 1, PIPE_TEXTURE_2D
, true, PIPE_FORMAT_R32_FLOAT
,
1091 init_compute_resources(ctx
, (int []) { 0, 1, -1 });
1092 init_sampler_states(ctx
, 2);
1093 launch_grid(ctx
, (uint
[]){1, 1, 1}, (uint
[]){128, 32, 1}, 0,
1095 check_tex(ctx
, 1, (is_int
? expecti
: expectf
), NULL
);
1096 destroy_sampler_states(ctx
);
1097 destroy_compute_resources(ctx
);
1104 static void test_surface_st(struct context
*ctx
)
1106 const char *src
= "COMP\n"
1107 "DCL RES[0], 2D, RAW\n"
1108 "DCL RES[1], 2D, WR\n"
1109 "DCL SV[0], BLOCK_ID[0]\n"
1110 "DCL TEMP[0], LOCAL\n"
1111 "DCL TEMP[1], LOCAL\n"
1112 "IMM UINT32 { 16, 1, 0, 0 }\n"
1115 " UMUL TEMP[0], SV[0], IMM[0]\n"
1116 " LOAD TEMP[1], RES[0], TEMP[0]\n"
1117 " STORE RES[1], SV[0], TEMP[1]\n"
1121 void init0f(void *p
, int s
, int x
, int y
) {
1122 float v
[] = { 1.0, -.75, 0.5, -.25 };
1123 *(float *)p
= v
[x
% 4];
1125 void init0i(void *p
, int s
, int x
, int y
) {
1126 int v
[] = { 0xffffffff, 0xffff, 0xff, 0xf };
1127 *(int32_t *)p
= v
[x
% 4];
1129 void init1(void *p
, int s
, int x
, int y
) {
1130 memset(p
, 1, util_format_get_blocksize(surface_fmts
[i
]));
1132 void expectf(void *p
, int s
, int x
, int y
) {
1136 for (j
= 0; j
< 4; j
++)
1137 init0f(&vf
[j
], s
, 4 * x
+ j
, y
);
1138 util_format_write_4f(surface_fmts
[i
], vf
, 0,
1141 void expects(void *p
, int s
, int x
, int y
) {
1145 for (j
= 0; j
< 4; j
++)
1146 init0i(&v
[j
], s
, 4 * x
+ j
, y
);
1147 util_format_write_4i(surface_fmts
[i
], v
, 0,
1150 void expectu(void *p
, int s
, int x
, int y
) {
1154 for (j
= 0; j
< 4; j
++)
1155 init0i(&v
[j
], s
, 4 * x
+ j
, y
);
1156 util_format_write_4ui(surface_fmts
[i
], v
, 0,
1159 bool check(void *x
, void *y
, int sz
) {
1162 if (util_format_is_float(surface_fmts
[i
])) {
1163 return fabs(*(float *)x
- *(float *)y
) < 3.92156863e-3;
1165 } else if ((sz
% 4) == 0) {
1166 for (j
= 0; j
< sz
/ 4; j
++)
1167 if (abs(((uint32_t *)x
)[j
] -
1168 ((uint32_t *)y
)[j
]) > 1)
1172 return !memcmp(x
, y
, sz
);
1176 printf("- %s\n", __func__
);
1178 init_prog(ctx
, 0, 0, 0, src
, NULL
);
1180 for (i
= 0; i
< Elements(surface_fmts
); i
++) {
1181 bool is_signed
= (util_format_description(surface_fmts
[i
])
1182 ->channel
[0].type
== UTIL_FORMAT_TYPE_SIGNED
);
1183 bool is_int
= util_format_is_pure_integer(surface_fmts
[i
]);
1185 printf(" - %s\n", util_format_name(surface_fmts
[i
]));
1187 if (!ctx
->screen
->is_format_supported(ctx
->screen
,
1188 surface_fmts
[i
], PIPE_TEXTURE_2D
, 1,
1189 PIPE_BIND_COMPUTE_RESOURCE
)) {
1190 printf("(unsupported)\n");
1194 init_tex(ctx
, 0, PIPE_TEXTURE_2D
, true, PIPE_FORMAT_R32_FLOAT
,
1195 512, 32, (is_int
? init0i
: init0f
));
1196 init_tex(ctx
, 1, PIPE_TEXTURE_2D
, true, surface_fmts
[i
],
1198 init_compute_resources(ctx
, (int []) { 0, 1, -1 });
1199 init_sampler_states(ctx
, 2);
1200 launch_grid(ctx
, (uint
[]){1, 1, 1}, (uint
[]){128, 32, 1}, 0,
1202 check_tex(ctx
, 1, (is_int
&& is_signed
? expects
:
1203 is_int
&& !is_signed
? expectu
:
1205 destroy_sampler_states(ctx
);
1206 destroy_compute_resources(ctx
);
1213 static void test_barrier(struct context
*ctx
)
1215 const char *src
= "COMP\n"
1216 "DCL RES[0], BUFFER, RAW, WR\n"
1217 "DCL SV[0], BLOCK_ID[0]\n"
1218 "DCL SV[1], BLOCK_SIZE[0]\n"
1219 "DCL SV[2], THREAD_ID[0]\n"
1220 "DCL TEMP[0], LOCAL\n"
1221 "DCL TEMP[1], LOCAL\n"
1222 "DCL TEMP[2], LOCAL\n"
1223 "DCL TEMP[3], LOCAL\n"
1224 "IMM UINT32 { 1, 0, 0, 0 }\n"
1225 "IMM UINT32 { 4, 0, 0, 0 }\n"
1226 "IMM UINT32 { 32, 0, 0, 0 }\n"
1229 " UMUL TEMP[0].x, SV[2], IMM[1]\n"
1230 " MOV TEMP[1].x, IMM[0].wwww\n"
1233 " STORE RLOCAL.x, TEMP[0], TEMP[1]\n"
1235 " MOV TEMP[2].x, IMM[0].wwww\n"
1237 " UMUL TEMP[3].x, TEMP[2], IMM[1]\n"
1238 " LOAD TEMP[3].x, RLOCAL, TEMP[3]\n"
1239 " USNE TEMP[3].x, TEMP[3], TEMP[1]\n"
1243 " UADD TEMP[2].x, TEMP[2], IMM[0]\n"
1244 " USEQ TEMP[3].x, TEMP[2], SV[1]\n"
1249 " UADD TEMP[1].x, TEMP[1], IMM[0]\n"
1250 " USEQ TEMP[2].x, TEMP[1], IMM[2]\n"
1255 " UMUL TEMP[1].x, SV[0], SV[1]\n"
1256 " UMUL TEMP[1].x, TEMP[1], IMM[1]\n"
1257 " UADD TEMP[1].x, TEMP[1], TEMP[0]\n"
1258 " LOAD TEMP[0].x, RLOCAL, TEMP[0]\n"
1259 " STORE RES[0].x, TEMP[1], TEMP[0]\n"
1262 void init(void *p
, int s
, int x
, int y
) {
1263 *(uint32_t *)p
= 0xdeadbeef;
1265 void expect(void *p
, int s
, int x
, int y
) {
1266 *(uint32_t *)p
= 31;
1269 printf("- %s\n", __func__
);
1271 init_prog(ctx
, 256, 0, 0, src
, NULL
);
1272 init_tex(ctx
, 0, PIPE_BUFFER
, true, PIPE_FORMAT_R32_FLOAT
,
1274 init_compute_resources(ctx
, (int []) { 0, -1 });
1275 launch_grid(ctx
, (uint
[]){64, 1, 1}, (uint
[]){16, 1, 1}, 0, NULL
);
1276 check_tex(ctx
, 0, expect
, NULL
);
1277 destroy_compute_resources(ctx
);
1282 static void test_atom_ops(struct context
*ctx
, bool global
)
1284 const char *src
= "COMP\n"
1285 "#ifdef TARGET_GLOBAL\n"
1286 "#define target RES[0]\n"
1288 "#define target RLOCAL\n"
1291 "DCL RES[0], BUFFER, RAW, WR\n"
1292 "#define threadid SV[0]\n"
1293 "DCL threadid, THREAD_ID[0]\n"
1295 "#define offset TEMP[0]\n"
1296 "DCL offset, LOCAL\n"
1297 "#define tmp TEMP[1]\n"
1300 "#define k0 IMM[0]\n"
1301 "IMM UINT32 { 0, 0, 0, 0 }\n"
1302 "#define k1 IMM[1]\n"
1303 "IMM UINT32 { 1, 0, 0, 0 }\n"
1304 "#define k2 IMM[2]\n"
1305 "IMM UINT32 { 2, 0, 0, 0 }\n"
1306 "#define k3 IMM[3]\n"
1307 "IMM UINT32 { 3, 0, 0, 0 }\n"
1308 "#define k4 IMM[4]\n"
1309 "IMM UINT32 { 4, 0, 0, 0 }\n"
1310 "#define k5 IMM[5]\n"
1311 "IMM UINT32 { 5, 0, 0, 0 }\n"
1312 "#define k6 IMM[6]\n"
1313 "IMM UINT32 { 6, 0, 0, 0 }\n"
1314 "#define k7 IMM[7]\n"
1315 "IMM UINT32 { 7, 0, 0, 0 }\n"
1316 "#define k8 IMM[8]\n"
1317 "IMM UINT32 { 8, 0, 0, 0 }\n"
1318 "#define k9 IMM[9]\n"
1319 "IMM UINT32 { 9, 0, 0, 0 }\n"
1320 "#define korig IMM[10].xxxx\n"
1321 "#define karg IMM[10].yyyy\n"
1322 "IMM UINT32 { 3735928559, 286331153, 0, 0 }\n"
1325 " UMUL offset.x, threadid, k4\n"
1326 " STORE target.x, offset, korig\n"
1327 " USEQ tmp.x, threadid, k0\n"
1329 " ATOMUADD tmp.x, target, offset, karg\n"
1330 " ATOMUADD tmp.x, target, offset, tmp\n"
1332 " USEQ tmp.x, threadid, k1\n"
1334 " ATOMXCHG tmp.x, target, offset, karg\n"
1335 " ATOMXCHG tmp.x, target, offset, tmp\n"
1337 " USEQ tmp.x, threadid, k2\n"
1339 " ATOMCAS tmp.x, target, offset, korig, karg\n"
1340 " ATOMCAS tmp.x, target, offset, tmp, k0\n"
1342 " USEQ tmp.x, threadid, k3\n"
1344 " ATOMAND tmp.x, target, offset, karg\n"
1345 " ATOMAND tmp.x, target, offset, tmp\n"
1347 " USEQ tmp.x, threadid, k4\n"
1349 " ATOMOR tmp.x, target, offset, karg\n"
1350 " ATOMOR tmp.x, target, offset, tmp\n"
1352 " USEQ tmp.x, threadid, k5\n"
1354 " ATOMXOR tmp.x, target, offset, karg\n"
1355 " ATOMXOR tmp.x, target, offset, tmp\n"
1357 " USEQ tmp.x, threadid, k6\n"
1359 " ATOMUMIN tmp.x, target, offset, karg\n"
1360 " ATOMUMIN tmp.x, target, offset, tmp\n"
1362 " USEQ tmp.x, threadid, k7\n"
1364 " ATOMUMAX tmp.x, target, offset, karg\n"
1365 " ATOMUMAX tmp.x, target, offset, tmp\n"
1367 " USEQ tmp.x, threadid, k8\n"
1369 " ATOMIMIN tmp.x, target, offset, karg\n"
1370 " ATOMIMIN tmp.x, target, offset, tmp\n"
1372 " USEQ tmp.x, threadid, k9\n"
1374 " ATOMIMAX tmp.x, target, offset, karg\n"
1375 " ATOMIMAX tmp.x, target, offset, tmp\n"
1377 "#ifdef TARGET_LOCAL\n"
1378 " LOAD tmp.x, RLOCAL, offset\n"
1379 " STORE RES[0].x, offset, tmp\n"
1384 void init(void *p
, int s
, int x
, int y
) {
1385 *(uint32_t *)p
= 0xbad;
1387 void expect(void *p
, int s
, int x
, int y
) {
1390 *(uint32_t *)p
= 0xce6c8eef;
1393 *(uint32_t *)p
= 0xdeadbeef;
1396 *(uint32_t *)p
= 0x11111111;
1399 *(uint32_t *)p
= 0x10011001;
1402 *(uint32_t *)p
= 0xdfbdbfff;
1405 *(uint32_t *)p
= 0x11111111;
1408 *(uint32_t *)p
= 0x11111111;
1411 *(uint32_t *)p
= 0xdeadbeef;
1414 *(uint32_t *)p
= 0xdeadbeef;
1417 *(uint32_t *)p
= 0x11111111;
1422 printf("- %s (%s)\n", __func__
, global
? "global" : "local");
1424 init_prog(ctx
, 40, 0, 0, src
,
1425 (global
? "-DTARGET_GLOBAL" : "-DTARGET_LOCAL"));
1426 init_tex(ctx
, 0, PIPE_BUFFER
, true, PIPE_FORMAT_R32_FLOAT
,
1428 init_compute_resources(ctx
, (int []) { 0, -1 });
1429 launch_grid(ctx
, (uint
[]){10, 1, 1}, (uint
[]){1, 1, 1}, 0, NULL
);
1430 check_tex(ctx
, 0, expect
, NULL
);
1431 destroy_compute_resources(ctx
);
1436 static void test_atom_race(struct context
*ctx
, bool global
)
1438 const char *src
= "COMP\n"
1439 "#ifdef TARGET_GLOBAL\n"
1440 "#define target RES[0]\n"
1442 "#define target RLOCAL\n"
1445 "DCL RES[0], BUFFER, RAW, WR\n"
1447 "#define blockid SV[0]\n"
1448 "DCL blockid, BLOCK_ID[0]\n"
1449 "#define blocksz SV[1]\n"
1450 "DCL blocksz, BLOCK_SIZE[0]\n"
1451 "#define threadid SV[2]\n"
1452 "DCL threadid, THREAD_ID[0]\n"
1454 "#define offset TEMP[0]\n"
1455 "DCL offset, LOCAL\n"
1456 "#define arg TEMP[1]\n"
1458 "#define count TEMP[2]\n"
1459 "DCL count, LOCAL\n"
1460 "#define vlocal TEMP[3]\n"
1461 "DCL vlocal, LOCAL\n"
1462 "#define vshared TEMP[4]\n"
1463 "DCL vshared, LOCAL\n"
1464 "#define last TEMP[5]\n"
1466 "#define tmp0 TEMP[6]\n"
1468 "#define tmp1 TEMP[7]\n"
1471 "#define k0 IMM[0]\n"
1472 "IMM UINT32 { 0, 0, 0, 0 }\n"
1473 "#define k1 IMM[1]\n"
1474 "IMM UINT32 { 1, 0, 0, 0 }\n"
1475 "#define k4 IMM[2]\n"
1476 "IMM UINT32 { 4, 0, 0, 0 }\n"
1477 "#define k32 IMM[3]\n"
1478 "IMM UINT32 { 32, 0, 0, 0 }\n"
1479 "#define k128 IMM[4]\n"
1480 "IMM UINT32 { 128, 0, 0, 0 }\n"
1481 "#define kdeadcafe IMM[5]\n"
1482 "IMM UINT32 { 3735931646, 0, 0, 0 }\n"
1483 "#define kallowed_set IMM[6]\n"
1484 "IMM UINT32 { 559035650, 0, 0, 0 }\n"
1485 "#define k11111111 IMM[7]\n"
1486 "IMM UINT32 { 286331153, 0, 0, 0 }\n"
1489 " MOV offset.x, threadid\n"
1490 "#ifdef TARGET_GLOBAL\n"
1491 " UMUL tmp0.x, blockid, blocksz\n"
1492 " UADD offset.x, offset, tmp0\n"
1494 " UMUL offset.x, offset, k4\n"
1495 " USLT tmp0.x, threadid, k32\n"
1496 " STORE target.x, offset, k0\n"
1499 " MOV vlocal.x, k0\n"
1500 " MOV arg.x, kdeadcafe\n"
1502 " INEG arg.x, arg\n"
1503 " ATOMUADD vshared.x, target, offset, arg\n"
1505 " USNE tmp0.x, vshared, vlocal\n"
1509 " UADD vlocal.x, vlocal, arg\n"
1511 " UADD vlocal.x, vshared, arg\n"
1512 " LOAD vshared.x, target, offset\n"
1513 " USEQ tmp0.x, vshared, vlocal\n"
1514 " STORE target.x, offset, tmp0\n"
1516 " UADD offset.x, offset, -k128\n"
1517 " MOV count.x, k0\n"
1520 " LOAD vshared.x, target, offset\n"
1521 " USEQ tmp0.x, vshared, kallowed_set.xxxx\n"
1522 " USEQ tmp1.x, vshared, kallowed_set.yyyy\n"
1523 " OR tmp0.x, tmp0, tmp1\n"
1525 " USEQ tmp0.x, vshared, last\n"
1529 " MOV last.x, vshared\n"
1533 " UADD count.x, count, k1\n"
1534 " USEQ tmp0.x, count, k128\n"
1539 " ATOMXCHG tmp0.x, target, offset, k11111111\n"
1540 " UADD offset.x, offset, k128\n"
1541 " ATOMXCHG tmp0.x, target, offset, k11111111\n"
1544 "#ifdef TARGET_LOCAL\n"
1545 " LOAD tmp0.x, RLOCAL, offset\n"
1546 " UMUL tmp1.x, blockid, blocksz\n"
1547 " UMUL tmp1.x, tmp1, k4\n"
1548 " UADD offset.x, offset, tmp1\n"
1549 " STORE RES[0].x, offset, tmp0\n"
1554 void init(void *p
, int s
, int x
, int y
) {
1555 *(uint32_t *)p
= 0xdeadbeef;
1557 void expect(void *p
, int s
, int x
, int y
) {
1558 *(uint32_t *)p
= x
& 0x20 ? 0x11111111 : 0xffffffff;
1561 printf("- %s (%s)\n", __func__
, global
? "global" : "local");
1563 init_prog(ctx
, 256, 0, 0, src
,
1564 (global
? "-DTARGET_GLOBAL" : "-DTARGET_LOCAL"));
1565 init_tex(ctx
, 0, PIPE_BUFFER
, true, PIPE_FORMAT_R32_FLOAT
,
1567 init_compute_resources(ctx
, (int []) { 0, -1 });
1568 launch_grid(ctx
, (uint
[]){64, 1, 1}, (uint
[]){16, 1, 1}, 0, NULL
);
1569 check_tex(ctx
, 0, expect
, NULL
);
1570 destroy_compute_resources(ctx
);
1575 int main(int argc
, char *argv
[])
1577 struct context
*ctx
= CALLOC_STRUCT(context
);
1579 unsigned tests
= (argc
> 1) ? strtoul(argv
[1], NULL
, 0) : ~0;
1583 if (tests
& (1 << 0))
1584 test_system_values(ctx
);
1585 if (tests
& (1 << 1))
1586 test_resource_access(ctx
);
1587 if (tests
& (1 << 2))
1588 test_function_calls(ctx
);
1589 if (tests
& (1 << 3))
1590 test_input_global(ctx
);
1591 if (tests
& (1 << 4))
1593 if (tests
& (1 << 5))
1595 if (tests
& (1 << 6))
1597 if (tests
& (1 << 7))
1598 test_many_kern(ctx
);
1599 if (tests
& (1 << 8))
1601 if (tests
& (1 << 9))
1602 test_resource_indirect(ctx
);
1603 if (tests
& (1 << 10))
1604 test_surface_ld(ctx
);
1605 if (tests
& (1 << 11))
1606 test_surface_st(ctx
);
1607 if (tests
& (1 << 12))
1609 if (tests
& (1 << 13))
1610 test_atom_ops(ctx
, true);
1611 if (tests
& (1 << 14))
1612 test_atom_race(ctx
, true);
1613 if (tests
& (1 << 15))
1614 test_atom_ops(ctx
, false);
1615 if (tests
& (1 << 16))
1616 test_atom_race(ctx
, false);