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 "program/prog_parameter.h"
35 #include "program/prog_print.h"
36 #include "program/prog_to_nir.h"
37 #include "program/program.h"
38 #include "program/programopt.h"
40 #include "util/ralloc.h"
41 #include "compiler/glsl/ir.h"
42 #include "compiler/glsl/glsl_to_nir.h"
44 #include "brw_program.h"
45 #include "brw_context.h"
47 #include "intel_batchbuffer.h"
50 brw_nir_lower_uniforms(nir_shader
*nir
, bool is_scalar
)
53 nir_assign_var_locations(&nir
->uniforms
, &nir
->num_uniforms
,
54 type_size_scalar_bytes
);
55 nir_lower_io(nir
, nir_var_uniform
, type_size_scalar_bytes
, 0);
57 nir_assign_var_locations(&nir
->uniforms
, &nir
->num_uniforms
,
58 type_size_vec4_bytes
);
59 nir_lower_io(nir
, nir_var_uniform
, type_size_vec4_bytes
, 0);
64 brw_create_nir(struct brw_context
*brw
,
65 const struct gl_shader_program
*shader_prog
,
66 struct gl_program
*prog
,
67 gl_shader_stage stage
,
70 struct gl_context
*ctx
= &brw
->ctx
;
71 const nir_shader_compiler_options
*options
=
72 ctx
->Const
.ShaderCompilerOptions
[stage
].NirOptions
;
76 /* First, lower the GLSL IR or Mesa IR to NIR */
78 nir
= glsl_to_nir(shader_prog
, stage
, options
);
79 nir_remove_dead_variables(nir
, nir_var_shader_in
| nir_var_shader_out
);
80 nir_lower_returns(nir
);
81 nir_validate_shader(nir
);
82 NIR_PASS_V(nir
, nir_lower_io_to_temporaries
,
83 nir_shader_get_entrypoint(nir
), true, false);
85 nir
= prog_to_nir(prog
, options
);
86 NIR_PASS_V(nir
, nir_lower_regs_to_ssa
); /* turn registers into SSA */
88 nir_validate_shader(nir
);
92 nir
= brw_preprocess_nir(brw
->screen
->compiler
, nir
);
94 if (stage
== MESA_SHADER_FRAGMENT
) {
95 static const struct nir_lower_wpos_ytransform_options wpos_options
= {
96 .state_tokens
= {STATE_INTERNAL
, STATE_FB_WPOS_Y_TRANSFORM
, 0, 0, 0},
97 .fs_coord_pixel_center_integer
= 1,
98 .fs_coord_origin_upper_left
= 1,
100 _mesa_add_state_reference(prog
->Parameters
,
101 (gl_state_index
*) wpos_options
.state_tokens
);
103 NIR_PASS(progress
, nir
, nir_lower_wpos_ytransform
, &wpos_options
);
106 NIR_PASS(progress
, nir
, nir_lower_system_values
);
107 NIR_PASS_V(nir
, brw_nir_lower_uniforms
, is_scalar
);
109 nir_shader_gather_info(nir
, nir_shader_get_entrypoint(nir
));
111 /* nir_shader may have been cloned so make sure shader_info is in sync */
112 if (nir
->info
!= &prog
->info
) {
113 const char *name
= prog
->info
.name
;
114 const char *label
= prog
->info
.label
;
115 prog
->info
= *nir
->info
;
116 prog
->info
.name
= name
;
117 prog
->info
.label
= label
;
121 NIR_PASS_V(nir
, nir_lower_samplers
, shader_prog
);
122 NIR_PASS_V(nir
, nir_lower_atomics
, shader_prog
);
129 get_new_program_id(struct intel_screen
*screen
)
131 static pthread_mutex_t m
= PTHREAD_MUTEX_INITIALIZER
;
132 pthread_mutex_lock(&m
);
133 unsigned id
= screen
->program_id
++;
134 pthread_mutex_unlock(&m
);
138 static struct gl_program
*brwNewProgram(struct gl_context
*ctx
, GLenum target
,
139 GLuint id
, bool is_arb_asm
)
141 struct brw_context
*brw
= brw_context(ctx
);
144 case GL_VERTEX_PROGRAM_ARB
:
145 case GL_TESS_CONTROL_PROGRAM_NV
:
146 case GL_TESS_EVALUATION_PROGRAM_NV
:
147 case GL_GEOMETRY_PROGRAM_NV
:
148 case GL_COMPUTE_PROGRAM_NV
: {
149 struct brw_program
*prog
= rzalloc(NULL
, struct brw_program
);
151 prog
->id
= get_new_program_id(brw
->screen
);
153 return _mesa_init_gl_program(&prog
->program
, target
, id
, is_arb_asm
);
159 case GL_FRAGMENT_PROGRAM_ARB
: {
160 struct brw_program
*prog
= rzalloc(NULL
, struct brw_program
);
163 prog
->id
= get_new_program_id(brw
->screen
);
165 return _mesa_init_gl_program(&prog
->program
, target
, id
, is_arb_asm
);
172 unreachable("Unsupported target in brwNewProgram()");
176 static void brwDeleteProgram( struct gl_context
*ctx
,
177 struct gl_program
*prog
)
179 struct brw_context
*brw
= brw_context(ctx
);
181 /* Beware! prog's refcount has reached zero, and it's about to be freed.
183 * In brw_upload_pipeline_state(), we compare brw->foo_program to
184 * ctx->FooProgram._Current, and flag BRW_NEW_FOO_PROGRAM if the
185 * pointer has changed.
187 * We cannot leave brw->foo_program as a dangling pointer to the dead
188 * program. malloc() may allocate the same memory for a new gl_program,
189 * causing us to see matching pointers...but totally different programs.
191 * We cannot set brw->foo_program to NULL, either. If we've deleted the
192 * active program, Mesa may set ctx->FooProgram._Current to NULL. That
193 * would cause us to see matching pointers (NULL == NULL), and fail to
194 * detect that a program has changed since our last draw.
196 * So, set it to a bogus gl_program pointer that will never match,
197 * causing us to properly reevaluate the state on our next draw.
199 * Getting this wrong causes heisenbugs which are very hard to catch,
200 * as you need a very specific allocation pattern to hit the problem.
202 static const struct gl_program deleted_program
;
204 if (brw
->vertex_program
== prog
)
205 brw
->vertex_program
= &deleted_program
;
207 if (brw
->tess_ctrl_program
== prog
)
208 brw
->tess_ctrl_program
= &deleted_program
;
210 if (brw
->tess_eval_program
== prog
)
211 brw
->tess_eval_program
= &deleted_program
;
213 if (brw
->geometry_program
== prog
)
214 brw
->geometry_program
= &deleted_program
;
216 if (brw
->fragment_program
== prog
)
217 brw
->fragment_program
= &deleted_program
;
219 if (brw
->compute_program
== prog
)
220 brw
->compute_program
= &deleted_program
;
222 _mesa_delete_program( ctx
, prog
);
227 brwProgramStringNotify(struct gl_context
*ctx
,
229 struct gl_program
*prog
)
231 assert(target
== GL_VERTEX_PROGRAM_ARB
|| !prog
->arb
.IsPositionInvariant
);
233 struct brw_context
*brw
= brw_context(ctx
);
234 const struct brw_compiler
*compiler
= brw
->screen
->compiler
;
237 case GL_FRAGMENT_PROGRAM_ARB
: {
238 struct brw_program
*newFP
= brw_program(prog
);
239 const struct brw_program
*curFP
=
240 brw_program_const(brw
->fragment_program
);
243 brw
->ctx
.NewDriverState
|= BRW_NEW_FRAGMENT_PROGRAM
;
244 newFP
->id
= get_new_program_id(brw
->screen
);
246 prog
->nir
= brw_create_nir(brw
, NULL
, prog
, MESA_SHADER_FRAGMENT
, true);
248 brw_fs_precompile(ctx
, prog
);
251 case GL_VERTEX_PROGRAM_ARB
: {
252 struct brw_program
*newVP
= brw_program(prog
);
253 const struct brw_program
*curVP
=
254 brw_program_const(brw
->vertex_program
);
257 brw
->ctx
.NewDriverState
|= BRW_NEW_VERTEX_PROGRAM
;
258 if (newVP
->program
.arb
.IsPositionInvariant
) {
259 _mesa_insert_mvp_code(ctx
, &newVP
->program
);
261 newVP
->id
= get_new_program_id(brw
->screen
);
263 /* Also tell tnl about it:
265 _tnl_program_string(ctx
, target
, prog
);
267 prog
->nir
= brw_create_nir(brw
, NULL
, prog
, MESA_SHADER_VERTEX
,
268 compiler
->scalar_stage
[MESA_SHADER_VERTEX
]);
270 brw_vs_precompile(ctx
, prog
);
275 * driver->ProgramStringNotify is only called for ARB programs, fixed
276 * function vertex programs, and ir_to_mesa (which isn't used by the
277 * i965 back-end). Therefore, even after geometry shaders are added,
278 * this function should only ever be called with a target of
279 * GL_VERTEX_PROGRAM_ARB or GL_FRAGMENT_PROGRAM_ARB.
281 unreachable("Unexpected target in brwProgramStringNotify");
288 brw_memory_barrier(struct gl_context
*ctx
, GLbitfield barriers
)
290 struct brw_context
*brw
= brw_context(ctx
);
291 unsigned bits
= (PIPE_CONTROL_DATA_CACHE_FLUSH
|
292 PIPE_CONTROL_NO_WRITE
|
293 PIPE_CONTROL_CS_STALL
);
294 assert(brw
->gen
>= 7 && brw
->gen
<= 9);
296 if (barriers
& (GL_VERTEX_ATTRIB_ARRAY_BARRIER_BIT
|
297 GL_ELEMENT_ARRAY_BARRIER_BIT
|
298 GL_COMMAND_BARRIER_BIT
))
299 bits
|= PIPE_CONTROL_VF_CACHE_INVALIDATE
;
301 if (barriers
& GL_UNIFORM_BARRIER_BIT
)
302 bits
|= (PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE
|
303 PIPE_CONTROL_CONST_CACHE_INVALIDATE
);
305 if (barriers
& GL_TEXTURE_FETCH_BARRIER_BIT
)
306 bits
|= PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE
;
308 if (barriers
& GL_TEXTURE_UPDATE_BARRIER_BIT
)
309 bits
|= PIPE_CONTROL_RENDER_TARGET_FLUSH
;
311 if (barriers
& GL_FRAMEBUFFER_BARRIER_BIT
)
312 bits
|= (PIPE_CONTROL_DEPTH_CACHE_FLUSH
|
313 PIPE_CONTROL_RENDER_TARGET_FLUSH
);
315 /* Typed surface messages are handled by the render cache on IVB, so we
316 * need to flush it too.
318 if (brw
->gen
== 7 && !brw
->is_haswell
)
319 bits
|= PIPE_CONTROL_RENDER_TARGET_FLUSH
;
321 brw_emit_pipe_control_flush(brw
, bits
);
325 brw_blend_barrier(struct gl_context
*ctx
)
327 struct brw_context
*brw
= brw_context(ctx
);
329 if (!ctx
->Extensions
.MESA_shader_framebuffer_fetch
) {
331 brw_emit_pipe_control_flush(brw
,
332 PIPE_CONTROL_RENDER_TARGET_FLUSH
|
333 PIPE_CONTROL_CS_STALL
);
334 brw_emit_pipe_control_flush(brw
,
335 PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE
);
337 brw_emit_pipe_control_flush(brw
,
338 PIPE_CONTROL_RENDER_TARGET_FLUSH
);
344 brw_get_scratch_bo(struct brw_context
*brw
,
345 drm_intel_bo
**scratch_bo
, int size
)
347 drm_intel_bo
*old_bo
= *scratch_bo
;
349 if (old_bo
&& old_bo
->size
< size
) {
350 drm_intel_bo_unreference(old_bo
);
355 *scratch_bo
= drm_intel_bo_alloc(brw
->bufmgr
, "scratch bo", size
, 4096);
360 * Reserve enough scratch space for the given stage to hold \p per_thread_size
361 * bytes times the given \p thread_count.
364 brw_alloc_stage_scratch(struct brw_context
*brw
,
365 struct brw_stage_state
*stage_state
,
366 unsigned per_thread_size
,
367 unsigned thread_count
)
369 if (stage_state
->per_thread_scratch
< per_thread_size
) {
370 stage_state
->per_thread_scratch
= per_thread_size
;
372 if (stage_state
->scratch_bo
)
373 drm_intel_bo_unreference(stage_state
->scratch_bo
);
375 stage_state
->scratch_bo
=
376 drm_intel_bo_alloc(brw
->bufmgr
, "shader scratch space",
377 per_thread_size
* thread_count
, 4096);
381 void brwInitFragProgFuncs( struct dd_function_table
*functions
)
383 assert(functions
->ProgramStringNotify
== _tnl_program_string
);
385 functions
->NewProgram
= brwNewProgram
;
386 functions
->DeleteProgram
= brwDeleteProgram
;
387 functions
->ProgramStringNotify
= brwProgramStringNotify
;
389 functions
->LinkShader
= brw_link_shader
;
391 functions
->MemoryBarrier
= brw_memory_barrier
;
392 functions
->BlendBarrier
= brw_blend_barrier
;
395 struct shader_times
{
402 brw_init_shader_time(struct brw_context
*brw
)
404 const int max_entries
= 2048;
405 brw
->shader_time
.bo
=
406 drm_intel_bo_alloc(brw
->bufmgr
, "shader time",
407 max_entries
* BRW_SHADER_TIME_STRIDE
* 3, 4096);
408 brw
->shader_time
.names
= rzalloc_array(brw
, const char *, max_entries
);
409 brw
->shader_time
.ids
= rzalloc_array(brw
, int, max_entries
);
410 brw
->shader_time
.types
= rzalloc_array(brw
, enum shader_time_shader_type
,
412 brw
->shader_time
.cumulative
= rzalloc_array(brw
, struct shader_times
,
414 brw
->shader_time
.max_entries
= max_entries
;
418 compare_time(const void *a
, const void *b
)
420 uint64_t * const *a_val
= a
;
421 uint64_t * const *b_val
= b
;
423 /* We don't just subtract because we're turning the value to an int. */
424 if (**a_val
< **b_val
)
426 else if (**a_val
== **b_val
)
433 print_shader_time_line(const char *stage
, const char *name
,
434 int shader_num
, uint64_t time
, uint64_t total
)
436 fprintf(stderr
, "%-6s%-18s", stage
, name
);
439 fprintf(stderr
, "%4d: ", shader_num
);
441 fprintf(stderr
, " : ");
443 fprintf(stderr
, "%16lld (%7.2f Gcycles) %4.1f%%\n",
445 (double)time
/ 1000000000.0,
446 (double)time
/ total
* 100.0);
450 brw_report_shader_time(struct brw_context
*brw
)
452 if (!brw
->shader_time
.bo
|| !brw
->shader_time
.num_entries
)
455 uint64_t scaled
[brw
->shader_time
.num_entries
];
456 uint64_t *sorted
[brw
->shader_time
.num_entries
];
457 uint64_t total_by_type
[ST_CS
+ 1];
458 memset(total_by_type
, 0, sizeof(total_by_type
));
460 for (int i
= 0; i
< brw
->shader_time
.num_entries
; i
++) {
461 uint64_t written
= 0, reset
= 0;
462 enum shader_time_shader_type type
= brw
->shader_time
.types
[i
];
464 sorted
[i
] = &scaled
[i
];
474 written
= brw
->shader_time
.cumulative
[i
].written
;
475 reset
= brw
->shader_time
.cumulative
[i
].reset
;
479 /* I sometimes want to print things that aren't the 3 shader times.
480 * Just print the sum in that case.
487 uint64_t time
= brw
->shader_time
.cumulative
[i
].time
;
489 scaled
[i
] = time
/ written
* (written
+ reset
);
502 total_by_type
[type
] += scaled
[i
];
512 fprintf(stderr
, "No shader time collected yet\n");
516 qsort(sorted
, brw
->shader_time
.num_entries
, sizeof(sorted
[0]), compare_time
);
518 fprintf(stderr
, "\n");
519 fprintf(stderr
, "type ID cycles spent %% of total\n");
520 for (int s
= 0; s
< brw
->shader_time
.num_entries
; s
++) {
522 /* Work back from the sorted pointers times to a time to print. */
523 int i
= sorted
[s
] - scaled
;
528 int shader_num
= brw
->shader_time
.ids
[i
];
529 const char *shader_name
= brw
->shader_time
.names
[i
];
531 switch (brw
->shader_time
.types
[i
]) {
558 print_shader_time_line(stage
, shader_name
, shader_num
,
562 fprintf(stderr
, "\n");
563 print_shader_time_line("total", "vs", 0, total_by_type
[ST_VS
], total
);
564 print_shader_time_line("total", "tcs", 0, total_by_type
[ST_TCS
], total
);
565 print_shader_time_line("total", "tes", 0, total_by_type
[ST_TES
], total
);
566 print_shader_time_line("total", "gs", 0, total_by_type
[ST_GS
], total
);
567 print_shader_time_line("total", "fs8", 0, total_by_type
[ST_FS8
], total
);
568 print_shader_time_line("total", "fs16", 0, total_by_type
[ST_FS16
], total
);
569 print_shader_time_line("total", "cs", 0, total_by_type
[ST_CS
], total
);
573 brw_collect_shader_time(struct brw_context
*brw
)
575 if (!brw
->shader_time
.bo
)
578 /* This probably stalls on the last rendering. We could fix that by
579 * delaying reading the reports, but it doesn't look like it's a big
580 * overhead compared to the cost of tracking the time in the first place.
582 drm_intel_bo_map(brw
->shader_time
.bo
, true);
583 void *bo_map
= brw
->shader_time
.bo
->virtual;
585 for (int i
= 0; i
< brw
->shader_time
.num_entries
; i
++) {
586 uint32_t *times
= bo_map
+ i
* 3 * BRW_SHADER_TIME_STRIDE
;
588 brw
->shader_time
.cumulative
[i
].time
+= times
[BRW_SHADER_TIME_STRIDE
* 0 / 4];
589 brw
->shader_time
.cumulative
[i
].written
+= times
[BRW_SHADER_TIME_STRIDE
* 1 / 4];
590 brw
->shader_time
.cumulative
[i
].reset
+= times
[BRW_SHADER_TIME_STRIDE
* 2 / 4];
593 /* Zero the BO out to clear it out for our next collection.
595 memset(bo_map
, 0, brw
->shader_time
.bo
->size
);
596 drm_intel_bo_unmap(brw
->shader_time
.bo
);
600 brw_collect_and_report_shader_time(struct brw_context
*brw
)
602 brw_collect_shader_time(brw
);
604 if (brw
->shader_time
.report_time
== 0 ||
605 get_time() - brw
->shader_time
.report_time
>= 1.0) {
606 brw_report_shader_time(brw
);
607 brw
->shader_time
.report_time
= get_time();
612 * Chooses an index in the shader_time buffer and sets up tracking information
615 * Note that this holds on to references to the underlying programs, which may
616 * change their lifetimes compared to normal operation.
619 brw_get_shader_time_index(struct brw_context
*brw
, struct gl_program
*prog
,
620 enum shader_time_shader_type type
, bool is_glsl_sh
)
622 int shader_time_index
= brw
->shader_time
.num_entries
++;
623 assert(shader_time_index
< brw
->shader_time
.max_entries
);
624 brw
->shader_time
.types
[shader_time_index
] = type
;
629 } else if (is_glsl_sh
) {
630 name
= prog
->info
.label
?
631 ralloc_strdup(brw
->shader_time
.names
, prog
->info
.label
) : "glsl";
636 brw
->shader_time
.names
[shader_time_index
] = name
;
637 brw
->shader_time
.ids
[shader_time_index
] = prog
->Id
;
639 return shader_time_index
;
643 brw_destroy_shader_time(struct brw_context
*brw
)
645 drm_intel_bo_unreference(brw
->shader_time
.bo
);
646 brw
->shader_time
.bo
= NULL
;
650 brw_stage_prog_data_free(const void *p
)
652 struct brw_stage_prog_data
*prog_data
= (struct brw_stage_prog_data
*)p
;
654 ralloc_free(prog_data
->param
);
655 ralloc_free(prog_data
->pull_param
);
656 ralloc_free(prog_data
->image_param
);
660 brw_dump_arb_asm(const char *stage
, struct gl_program
*prog
)
662 fprintf(stderr
, "ARB_%s_program %d ir for native %s shader\n",
663 stage
, prog
->Id
, stage
);
664 _mesa_print_program(prog
);
668 brw_setup_tex_for_precompile(struct brw_context
*brw
,
669 struct brw_sampler_prog_key_data
*tex
,
670 struct gl_program
*prog
)
672 const bool has_shader_channel_select
= brw
->is_haswell
|| brw
->gen
>= 8;
673 unsigned sampler_count
= util_last_bit(prog
->SamplersUsed
);
674 for (unsigned i
= 0; i
< sampler_count
; i
++) {
675 if (!has_shader_channel_select
&& (prog
->ShadowSamplers
& (1 << i
))) {
676 /* Assume DEPTH_TEXTURE_MODE is the default: X, X, X, 1 */
678 MAKE_SWIZZLE4(SWIZZLE_X
, SWIZZLE_X
, SWIZZLE_X
, SWIZZLE_ONE
);
680 /* Color sampler: assume no swizzling. */
681 tex
->swizzles
[i
] = SWIZZLE_XYZW
;
687 * Sets up the starting offsets for the groups of binding table entries
688 * common to all pipeline stages.
690 * Unused groups are initialized to 0xd0d0d0d0 to make it obvious that they're
691 * unused but also make sure that addition of small offsets to them will
692 * trigger some of our asserts that surface indices are < BRW_MAX_SURFACES.
695 brw_assign_common_binding_table_offsets(const struct gen_device_info
*devinfo
,
696 const struct gl_program
*prog
,
697 struct brw_stage_prog_data
*stage_prog_data
,
698 uint32_t next_binding_table_offset
)
700 int num_textures
= util_last_bit(prog
->SamplersUsed
);
702 stage_prog_data
->binding_table
.texture_start
= next_binding_table_offset
;
703 next_binding_table_offset
+= num_textures
;
705 if (prog
->info
.num_ubos
) {
706 assert(prog
->info
.num_ubos
<= BRW_MAX_UBO
);
707 stage_prog_data
->binding_table
.ubo_start
= next_binding_table_offset
;
708 next_binding_table_offset
+= prog
->info
.num_ubos
;
710 stage_prog_data
->binding_table
.ubo_start
= 0xd0d0d0d0;
713 if (prog
->info
.num_ssbos
) {
714 assert(prog
->info
.num_ssbos
<= BRW_MAX_SSBO
);
715 stage_prog_data
->binding_table
.ssbo_start
= next_binding_table_offset
;
716 next_binding_table_offset
+= prog
->info
.num_ssbos
;
718 stage_prog_data
->binding_table
.ssbo_start
= 0xd0d0d0d0;
721 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
722 stage_prog_data
->binding_table
.shader_time_start
= next_binding_table_offset
;
723 next_binding_table_offset
++;
725 stage_prog_data
->binding_table
.shader_time_start
= 0xd0d0d0d0;
728 if (prog
->nir
->info
->uses_texture_gather
) {
729 if (devinfo
->gen
>= 8) {
730 stage_prog_data
->binding_table
.gather_texture_start
=
731 stage_prog_data
->binding_table
.texture_start
;
733 stage_prog_data
->binding_table
.gather_texture_start
= next_binding_table_offset
;
734 next_binding_table_offset
+= num_textures
;
737 stage_prog_data
->binding_table
.gather_texture_start
= 0xd0d0d0d0;
740 if (prog
->info
.num_abos
) {
741 stage_prog_data
->binding_table
.abo_start
= next_binding_table_offset
;
742 next_binding_table_offset
+= prog
->info
.num_abos
;
744 stage_prog_data
->binding_table
.abo_start
= 0xd0d0d0d0;
747 if (prog
->info
.num_images
) {
748 stage_prog_data
->binding_table
.image_start
= next_binding_table_offset
;
749 next_binding_table_offset
+= prog
->info
.num_images
;
751 stage_prog_data
->binding_table
.image_start
= 0xd0d0d0d0;
754 /* This may or may not be used depending on how the compile goes. */
755 stage_prog_data
->binding_table
.pull_constants_start
= next_binding_table_offset
;
756 next_binding_table_offset
++;
758 /* Plane 0 is just the regular texture section */
759 stage_prog_data
->binding_table
.plane_start
[0] = stage_prog_data
->binding_table
.texture_start
;
761 stage_prog_data
->binding_table
.plane_start
[1] = next_binding_table_offset
;
762 next_binding_table_offset
+= num_textures
;
764 stage_prog_data
->binding_table
.plane_start
[2] = next_binding_table_offset
;
765 next_binding_table_offset
+= num_textures
;
767 /* prog_data->base.binding_table.size will be set by brw_mark_surface_used. */
769 assert(next_binding_table_offset
<= BRW_MAX_SURFACES
);
770 return next_binding_table_offset
;