2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics to
4 develop this 3D driver.
6 Permission is hereby granted, free of charge, to any person obtaining
7 a copy of this software and associated documentation files (the
8 "Software"), to deal in the Software without restriction, including
9 without limitation the rights to use, copy, modify, merge, publish,
10 distribute, sublicense, and/or sell copies of the Software, and to
11 permit persons to whom the Software is furnished to do so, subject to
12 the following conditions:
14 The above copyright notice and this permission notice (including the
15 next paragraph) shall be included in all copies or substantial
16 portions of the Software.
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **********************************************************************/
29 * Keith Whitwell <keithw@vmware.com>
33 #include "main/imports.h"
34 #include "main/enums.h"
35 #include "main/shaderobj.h"
36 #include "program/prog_parameter.h"
37 #include "program/prog_print.h"
38 #include "program/program.h"
39 #include "program/programopt.h"
41 #include "util/ralloc.h"
44 #include "brw_context.h"
48 get_new_program_id(struct intel_screen
*screen
)
50 static pthread_mutex_t m
= PTHREAD_MUTEX_INITIALIZER
;
51 pthread_mutex_lock(&m
);
52 unsigned id
= screen
->program_id
++;
53 pthread_mutex_unlock(&m
);
57 static struct gl_program
*brwNewProgram( struct gl_context
*ctx
,
61 struct brw_context
*brw
= brw_context(ctx
);
64 case GL_VERTEX_PROGRAM_ARB
: {
65 struct brw_vertex_program
*prog
= CALLOC_STRUCT(brw_vertex_program
);
67 prog
->id
= get_new_program_id(brw
->intelScreen
);
69 return _mesa_init_vertex_program( ctx
, &prog
->program
,
76 case GL_FRAGMENT_PROGRAM_ARB
: {
77 struct brw_fragment_program
*prog
= CALLOC_STRUCT(brw_fragment_program
);
79 prog
->id
= get_new_program_id(brw
->intelScreen
);
81 return _mesa_init_fragment_program( ctx
, &prog
->program
,
88 case MESA_GEOMETRY_PROGRAM
: {
89 struct brw_geometry_program
*prog
= CALLOC_STRUCT(brw_geometry_program
);
91 prog
->id
= get_new_program_id(brw
->intelScreen
);
93 return _mesa_init_geometry_program(ctx
, &prog
->program
, target
, id
);
99 case GL_COMPUTE_PROGRAM_NV
: {
100 struct brw_compute_program
*prog
= CALLOC_STRUCT(brw_compute_program
);
102 prog
->id
= get_new_program_id(brw
->intelScreen
);
104 return _mesa_init_compute_program(ctx
, &prog
->program
, target
, id
);
111 unreachable("Unsupported target in brwNewProgram()");
115 static void brwDeleteProgram( struct gl_context
*ctx
,
116 struct gl_program
*prog
)
118 _mesa_delete_program( ctx
, prog
);
123 brwIsProgramNative(struct gl_context
*ctx
,
125 struct gl_program
*prog
)
131 brwProgramStringNotify(struct gl_context
*ctx
,
133 struct gl_program
*prog
)
135 struct brw_context
*brw
= brw_context(ctx
);
138 case GL_FRAGMENT_PROGRAM_ARB
: {
139 struct gl_fragment_program
*fprog
= (struct gl_fragment_program
*) prog
;
140 struct brw_fragment_program
*newFP
= brw_fragment_program(fprog
);
141 const struct brw_fragment_program
*curFP
=
142 brw_fragment_program_const(brw
->fragment_program
);
145 brw
->state
.dirty
.brw
|= BRW_NEW_FRAGMENT_PROGRAM
;
146 newFP
->id
= get_new_program_id(brw
->intelScreen
);
149 case GL_VERTEX_PROGRAM_ARB
: {
150 struct gl_vertex_program
*vprog
= (struct gl_vertex_program
*) prog
;
151 struct brw_vertex_program
*newVP
= brw_vertex_program(vprog
);
152 const struct brw_vertex_program
*curVP
=
153 brw_vertex_program_const(brw
->vertex_program
);
156 brw
->state
.dirty
.brw
|= BRW_NEW_VERTEX_PROGRAM
;
157 if (newVP
->program
.IsPositionInvariant
) {
158 _mesa_insert_mvp_code(ctx
, &newVP
->program
);
160 newVP
->id
= get_new_program_id(brw
->intelScreen
);
162 /* Also tell tnl about it:
164 _tnl_program_string(ctx
, target
, prog
);
169 * driver->ProgramStringNotify is only called for ARB programs, fixed
170 * function vertex programs, and ir_to_mesa (which isn't used by the
171 * i965 back-end). Therefore, even after geometry shaders are added,
172 * this function should only ever be called with a target of
173 * GL_VERTEX_PROGRAM_ARB or GL_FRAGMENT_PROGRAM_ARB.
175 unreachable("Unexpected target in brwProgramStringNotify");
178 brw_add_texrect_params(prog
);
184 brw_add_texrect_params(struct gl_program
*prog
)
186 for (int texunit
= 0; texunit
< BRW_MAX_TEX_UNIT
; texunit
++) {
187 if (!(prog
->TexturesUsed
[texunit
] & (1 << TEXTURE_RECT_INDEX
)))
190 int tokens
[STATE_LENGTH
] = {
198 _mesa_add_state_reference(prog
->Parameters
, (gl_state_index
*)tokens
);
202 /* Per-thread scratch space is a power-of-two multiple of 1KB. */
204 brw_get_scratch_size(int size
)
208 for (i
= 1024; i
< size
; i
*= 2)
215 brw_get_scratch_bo(struct brw_context
*brw
,
216 drm_intel_bo
**scratch_bo
, int size
)
218 drm_intel_bo
*old_bo
= *scratch_bo
;
220 if (old_bo
&& old_bo
->size
< size
) {
221 drm_intel_bo_unreference(old_bo
);
226 *scratch_bo
= drm_intel_bo_alloc(brw
->bufmgr
, "scratch bo", size
, 4096);
230 void brwInitFragProgFuncs( struct dd_function_table
*functions
)
232 assert(functions
->ProgramStringNotify
== _tnl_program_string
);
234 functions
->NewProgram
= brwNewProgram
;
235 functions
->DeleteProgram
= brwDeleteProgram
;
236 functions
->IsProgramNative
= brwIsProgramNative
;
237 functions
->ProgramStringNotify
= brwProgramStringNotify
;
239 functions
->NewShader
= brw_new_shader
;
240 functions
->LinkShader
= brw_link_shader
;
244 brw_init_shader_time(struct brw_context
*brw
)
246 const int max_entries
= 4096;
247 brw
->shader_time
.bo
= drm_intel_bo_alloc(brw
->bufmgr
, "shader time",
248 max_entries
* SHADER_TIME_STRIDE
,
250 brw
->shader_time
.shader_programs
= rzalloc_array(brw
, struct gl_shader_program
*,
252 brw
->shader_time
.programs
= rzalloc_array(brw
, struct gl_program
*,
254 brw
->shader_time
.types
= rzalloc_array(brw
, enum shader_time_shader_type
,
256 brw
->shader_time
.cumulative
= rzalloc_array(brw
, uint64_t,
258 brw
->shader_time
.max_entries
= max_entries
;
262 compare_time(const void *a
, const void *b
)
264 uint64_t * const *a_val
= a
;
265 uint64_t * const *b_val
= b
;
267 /* We don't just subtract because we're turning the value to an int. */
268 if (**a_val
< **b_val
)
270 else if (**a_val
== **b_val
)
277 get_written_and_reset(struct brw_context
*brw
, int i
,
278 uint64_t *written
, uint64_t *reset
)
280 enum shader_time_shader_type type
= brw
->shader_time
.types
[i
];
281 assert(type
== ST_VS
|| type
== ST_GS
|| type
== ST_FS8
|| type
== ST_FS16
);
283 /* Find where we recorded written and reset. */
286 for (wi
= i
; brw
->shader_time
.types
[wi
] != type
+ 1; wi
++)
289 for (ri
= i
; brw
->shader_time
.types
[ri
] != type
+ 2; ri
++)
292 *written
= brw
->shader_time
.cumulative
[wi
];
293 *reset
= brw
->shader_time
.cumulative
[ri
];
297 print_shader_time_line(const char *stage
, const char *name
,
298 int shader_num
, uint64_t time
, uint64_t total
)
300 fprintf(stderr
, "%-6s%-18s", stage
, name
);
302 if (shader_num
!= -1)
303 fprintf(stderr
, "%4d: ", shader_num
);
305 fprintf(stderr
, " : ");
307 fprintf(stderr
, "%16lld (%7.2f Gcycles) %4.1f%%\n",
309 (double)time
/ 1000000000.0,
310 (double)time
/ total
* 100.0);
314 brw_report_shader_time(struct brw_context
*brw
)
316 if (!brw
->shader_time
.bo
|| !brw
->shader_time
.num_entries
)
319 uint64_t scaled
[brw
->shader_time
.num_entries
];
320 uint64_t *sorted
[brw
->shader_time
.num_entries
];
321 uint64_t total_by_type
[ST_FS16
+ 1];
322 memset(total_by_type
, 0, sizeof(total_by_type
));
324 for (int i
= 0; i
< brw
->shader_time
.num_entries
; i
++) {
325 uint64_t written
= 0, reset
= 0;
326 enum shader_time_shader_type type
= brw
->shader_time
.types
[i
];
328 sorted
[i
] = &scaled
[i
];
337 case ST_FS16_WRITTEN
:
339 /* We'll handle these when along with the time. */
347 get_written_and_reset(brw
, i
, &written
, &reset
);
351 /* I sometimes want to print things that aren't the 3 shader times.
352 * Just print the sum in that case.
359 uint64_t time
= brw
->shader_time
.cumulative
[i
];
361 scaled
[i
] = time
/ written
* (written
+ reset
);
371 total_by_type
[type
] += scaled
[i
];
381 fprintf(stderr
, "No shader time collected yet\n");
385 qsort(sorted
, brw
->shader_time
.num_entries
, sizeof(sorted
[0]), compare_time
);
387 fprintf(stderr
, "\n");
388 fprintf(stderr
, "type ID cycles spent %% of total\n");
389 for (int s
= 0; s
< brw
->shader_time
.num_entries
; s
++) {
390 const char *shader_name
;
392 /* Work back from the sorted pointers times to a time to print. */
393 int i
= sorted
[s
] - scaled
;
394 struct gl_shader_program
*prog
= brw
->shader_time
.shader_programs
[i
];
401 shader_num
= prog
->Name
;
403 /* The fixed function fragment shader generates GLSL IR with a Name
404 * of 0, and nothing else does.
407 shader_name
= prog
->Label
;
408 } else if (shader_num
== 0 &&
409 (brw
->shader_time
.types
[i
] == ST_FS8
||
410 brw
->shader_time
.types
[i
] == ST_FS16
)) {
414 shader_name
= "glsl";
416 } else if (brw
->shader_time
.programs
[i
]) {
417 shader_num
= brw
->shader_time
.programs
[i
]->Id
;
418 if (shader_num
== 0) {
422 shader_name
= "prog";
425 shader_name
= "other";
428 switch (brw
->shader_time
.types
[i
]) {
446 print_shader_time_line(stage
, shader_name
, shader_num
,
450 fprintf(stderr
, "\n");
451 print_shader_time_line("total", "vs", -1, total_by_type
[ST_VS
], total
);
452 print_shader_time_line("total", "gs", -1, total_by_type
[ST_GS
], total
);
453 print_shader_time_line("total", "fs8", -1, total_by_type
[ST_FS8
], total
);
454 print_shader_time_line("total", "fs16", -1, total_by_type
[ST_FS16
], total
);
458 brw_collect_shader_time(struct brw_context
*brw
)
460 if (!brw
->shader_time
.bo
)
463 /* This probably stalls on the last rendering. We could fix that by
464 * delaying reading the reports, but it doesn't look like it's a big
465 * overhead compared to the cost of tracking the time in the first place.
467 drm_intel_bo_map(brw
->shader_time
.bo
, true);
469 uint32_t *times
= brw
->shader_time
.bo
->virtual;
471 for (int i
= 0; i
< brw
->shader_time
.num_entries
; i
++) {
472 brw
->shader_time
.cumulative
[i
] += times
[i
* SHADER_TIME_STRIDE
/ 4];
475 /* Zero the BO out to clear it out for our next collection.
477 memset(times
, 0, brw
->shader_time
.bo
->size
);
478 drm_intel_bo_unmap(brw
->shader_time
.bo
);
482 brw_collect_and_report_shader_time(struct brw_context
*brw
)
484 brw_collect_shader_time(brw
);
486 if (brw
->shader_time
.report_time
== 0 ||
487 get_time() - brw
->shader_time
.report_time
>= 1.0) {
488 brw_report_shader_time(brw
);
489 brw
->shader_time
.report_time
= get_time();
494 * Chooses an index in the shader_time buffer and sets up tracking information
497 * Note that this holds on to references to the underlying programs, which may
498 * change their lifetimes compared to normal operation.
501 brw_get_shader_time_index(struct brw_context
*brw
,
502 struct gl_shader_program
*shader_prog
,
503 struct gl_program
*prog
,
504 enum shader_time_shader_type type
)
506 struct gl_context
*ctx
= &brw
->ctx
;
508 int shader_time_index
= brw
->shader_time
.num_entries
++;
509 assert(shader_time_index
< brw
->shader_time
.max_entries
);
510 brw
->shader_time
.types
[shader_time_index
] = type
;
512 _mesa_reference_shader_program(ctx
,
513 &brw
->shader_time
.shader_programs
[shader_time_index
],
516 _mesa_reference_program(ctx
,
517 &brw
->shader_time
.programs
[shader_time_index
],
520 return shader_time_index
;
524 brw_destroy_shader_time(struct brw_context
*brw
)
526 drm_intel_bo_unreference(brw
->shader_time
.bo
);
527 brw
->shader_time
.bo
= NULL
;
531 brw_mark_surface_used(struct brw_stage_prog_data
*prog_data
,
534 assert(surf_index
< BRW_MAX_SURFACES
);
536 prog_data
->binding_table
.size_bytes
=
537 MAX2(prog_data
->binding_table
.size_bytes
, (surf_index
+ 1) * 4);
541 brw_stage_prog_data_compare(const struct brw_stage_prog_data
*a
,
542 const struct brw_stage_prog_data
*b
)
544 /* Compare all the struct up to the pointers. */
545 if (memcmp(a
, b
, offsetof(struct brw_stage_prog_data
, param
)))
548 if (memcmp(a
->param
, b
->param
, a
->nr_params
* sizeof(void *)))
551 if (memcmp(a
->pull_param
, b
->pull_param
, a
->nr_pull_params
* sizeof(void *)))
558 brw_stage_prog_data_free(const void *p
)
560 struct brw_stage_prog_data
*prog_data
= (struct brw_stage_prog_data
*)p
;
562 ralloc_free(prog_data
->param
);
563 ralloc_free(prog_data
->pull_param
);
567 brw_dump_ir(struct brw_context
*brw
, const char *stage
,
568 struct gl_shader_program
*shader_prog
,
569 struct gl_shader
*shader
, struct gl_program
*prog
)
573 "GLSL IR for native %s shader %d:\n", stage
, shader_prog
->Name
);
574 _mesa_print_ir(stderr
, shader
->ir
, NULL
);
575 fprintf(stderr
, "\n\n");
577 fprintf(stderr
, "ARB_%s_program %d ir for native %s shader\n",
578 stage
, prog
->Id
, stage
);
579 _mesa_print_program(prog
);