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"
46 #include "compiler/brw_nir.h"
47 #include "brw_defines.h"
48 #include "intel_batchbuffer.h"
51 brw_nir_lower_uniforms(nir_shader
*nir
, bool is_scalar
)
54 nir_assign_var_locations(&nir
->uniforms
, &nir
->num_uniforms
,
55 type_size_scalar_bytes
);
56 return nir_lower_io(nir
, nir_var_uniform
, type_size_scalar_bytes
, 0);
58 nir_assign_var_locations(&nir
->uniforms
, &nir
->num_uniforms
,
59 type_size_vec4_bytes
);
60 return nir_lower_io(nir
, nir_var_uniform
, type_size_vec4_bytes
, 0);
65 brw_create_nir(struct brw_context
*brw
,
66 const struct gl_shader_program
*shader_prog
,
67 struct gl_program
*prog
,
68 gl_shader_stage stage
,
71 struct gl_context
*ctx
= &brw
->ctx
;
72 const nir_shader_compiler_options
*options
=
73 ctx
->Const
.ShaderCompilerOptions
[stage
].NirOptions
;
77 /* First, lower the GLSL IR or Mesa IR to NIR */
79 nir
= glsl_to_nir(shader_prog
, stage
, options
);
80 nir_remove_dead_variables(nir
, nir_var_shader_in
| nir_var_shader_out
);
81 nir_lower_returns(nir
);
82 nir_validate_shader(nir
);
83 NIR_PASS_V(nir
, nir_lower_io_to_temporaries
,
84 nir_shader_get_entrypoint(nir
), true, false);
86 nir
= prog_to_nir(prog
, options
);
87 NIR_PASS_V(nir
, nir_lower_regs_to_ssa
); /* turn registers into SSA */
89 nir_validate_shader(nir
);
93 nir
= brw_preprocess_nir(brw
->screen
->compiler
, nir
);
95 if (stage
== MESA_SHADER_FRAGMENT
) {
96 static const struct nir_lower_wpos_ytransform_options wpos_options
= {
97 .state_tokens
= {STATE_INTERNAL
, STATE_FB_WPOS_Y_TRANSFORM
, 0, 0, 0},
98 .fs_coord_pixel_center_integer
= 1,
99 .fs_coord_origin_upper_left
= 1,
101 _mesa_add_state_reference(prog
->Parameters
,
102 (gl_state_index
*) wpos_options
.state_tokens
);
104 NIR_PASS(progress
, nir
, nir_lower_wpos_ytransform
, &wpos_options
);
107 NIR_PASS(progress
, nir
, nir_lower_system_values
);
108 NIR_PASS_V(nir
, brw_nir_lower_uniforms
, is_scalar
);
110 nir_shader_gather_info(nir
, nir_shader_get_entrypoint(nir
));
112 /* Copy the info we just generated back into the gl_program */
113 const char *prog_name
= prog
->info
.name
;
114 const char *prog_label
= prog
->info
.label
;
115 prog
->info
= nir
->info
;
116 prog
->info
.name
= prog_name
;
117 prog
->info
.label
= prog_label
;
120 NIR_PASS_V(nir
, nir_lower_samplers
, shader_prog
);
121 NIR_PASS_V(nir
, nir_lower_atomics
, shader_prog
);
128 get_new_program_id(struct intel_screen
*screen
)
130 static pthread_mutex_t m
= PTHREAD_MUTEX_INITIALIZER
;
131 pthread_mutex_lock(&m
);
132 unsigned id
= screen
->program_id
++;
133 pthread_mutex_unlock(&m
);
137 static struct gl_program
*brwNewProgram(struct gl_context
*ctx
, GLenum target
,
138 GLuint id
, bool is_arb_asm
)
140 struct brw_context
*brw
= brw_context(ctx
);
143 case GL_VERTEX_PROGRAM_ARB
:
144 case GL_TESS_CONTROL_PROGRAM_NV
:
145 case GL_TESS_EVALUATION_PROGRAM_NV
:
146 case GL_GEOMETRY_PROGRAM_NV
:
147 case GL_COMPUTE_PROGRAM_NV
: {
148 struct brw_program
*prog
= rzalloc(NULL
, struct brw_program
);
150 prog
->id
= get_new_program_id(brw
->screen
);
152 return _mesa_init_gl_program(&prog
->program
, target
, id
, is_arb_asm
);
158 case GL_FRAGMENT_PROGRAM_ARB
: {
159 struct brw_program
*prog
= rzalloc(NULL
, struct brw_program
);
162 prog
->id
= get_new_program_id(brw
->screen
);
164 return _mesa_init_gl_program(&prog
->program
, target
, id
, is_arb_asm
);
171 unreachable("Unsupported target in brwNewProgram()");
175 static void brwDeleteProgram( struct gl_context
*ctx
,
176 struct gl_program
*prog
)
178 struct brw_context
*brw
= brw_context(ctx
);
180 /* Beware! prog's refcount has reached zero, and it's about to be freed.
182 * In brw_upload_pipeline_state(), we compare brw->foo_program to
183 * ctx->FooProgram._Current, and flag BRW_NEW_FOO_PROGRAM if the
184 * pointer has changed.
186 * We cannot leave brw->foo_program as a dangling pointer to the dead
187 * program. malloc() may allocate the same memory for a new gl_program,
188 * causing us to see matching pointers...but totally different programs.
190 * We cannot set brw->foo_program to NULL, either. If we've deleted the
191 * active program, Mesa may set ctx->FooProgram._Current to NULL. That
192 * would cause us to see matching pointers (NULL == NULL), and fail to
193 * detect that a program has changed since our last draw.
195 * So, set it to a bogus gl_program pointer that will never match,
196 * causing us to properly reevaluate the state on our next draw.
198 * Getting this wrong causes heisenbugs which are very hard to catch,
199 * as you need a very specific allocation pattern to hit the problem.
201 static const struct gl_program deleted_program
;
203 if (brw
->vertex_program
== prog
)
204 brw
->vertex_program
= &deleted_program
;
206 if (brw
->tess_ctrl_program
== prog
)
207 brw
->tess_ctrl_program
= &deleted_program
;
209 if (brw
->tess_eval_program
== prog
)
210 brw
->tess_eval_program
= &deleted_program
;
212 if (brw
->geometry_program
== prog
)
213 brw
->geometry_program
= &deleted_program
;
215 if (brw
->fragment_program
== prog
)
216 brw
->fragment_program
= &deleted_program
;
218 if (brw
->compute_program
== prog
)
219 brw
->compute_program
= &deleted_program
;
221 _mesa_delete_program( ctx
, prog
);
226 brwProgramStringNotify(struct gl_context
*ctx
,
228 struct gl_program
*prog
)
230 assert(target
== GL_VERTEX_PROGRAM_ARB
|| !prog
->arb
.IsPositionInvariant
);
232 struct brw_context
*brw
= brw_context(ctx
);
233 const struct brw_compiler
*compiler
= brw
->screen
->compiler
;
236 case GL_FRAGMENT_PROGRAM_ARB
: {
237 struct brw_program
*newFP
= brw_program(prog
);
238 const struct brw_program
*curFP
=
239 brw_program_const(brw
->fragment_program
);
242 brw
->ctx
.NewDriverState
|= BRW_NEW_FRAGMENT_PROGRAM
;
243 newFP
->id
= get_new_program_id(brw
->screen
);
245 prog
->nir
= brw_create_nir(brw
, NULL
, prog
, MESA_SHADER_FRAGMENT
, true);
247 brw_fs_precompile(ctx
, prog
);
250 case GL_VERTEX_PROGRAM_ARB
: {
251 struct brw_program
*newVP
= brw_program(prog
);
252 const struct brw_program
*curVP
=
253 brw_program_const(brw
->vertex_program
);
256 brw
->ctx
.NewDriverState
|= BRW_NEW_VERTEX_PROGRAM
;
257 if (newVP
->program
.arb
.IsPositionInvariant
) {
258 _mesa_insert_mvp_code(ctx
, &newVP
->program
);
260 newVP
->id
= get_new_program_id(brw
->screen
);
262 /* Also tell tnl about it:
264 _tnl_program_string(ctx
, target
, prog
);
266 prog
->nir
= brw_create_nir(brw
, NULL
, prog
, MESA_SHADER_VERTEX
,
267 compiler
->scalar_stage
[MESA_SHADER_VERTEX
]);
269 brw_vs_precompile(ctx
, prog
);
274 * driver->ProgramStringNotify is only called for ARB programs, fixed
275 * function vertex programs, and ir_to_mesa (which isn't used by the
276 * i965 back-end). Therefore, even after geometry shaders are added,
277 * this function should only ever be called with a target of
278 * GL_VERTEX_PROGRAM_ARB or GL_FRAGMENT_PROGRAM_ARB.
280 unreachable("Unexpected target in brwProgramStringNotify");
287 brw_memory_barrier(struct gl_context
*ctx
, GLbitfield barriers
)
289 struct brw_context
*brw
= brw_context(ctx
);
290 unsigned bits
= (PIPE_CONTROL_DATA_CACHE_FLUSH
|
291 PIPE_CONTROL_NO_WRITE
|
292 PIPE_CONTROL_CS_STALL
);
293 assert(brw
->gen
>= 7 && brw
->gen
<= 9);
295 if (barriers
& (GL_VERTEX_ATTRIB_ARRAY_BARRIER_BIT
|
296 GL_ELEMENT_ARRAY_BARRIER_BIT
|
297 GL_COMMAND_BARRIER_BIT
))
298 bits
|= PIPE_CONTROL_VF_CACHE_INVALIDATE
;
300 if (barriers
& GL_UNIFORM_BARRIER_BIT
)
301 bits
|= (PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE
|
302 PIPE_CONTROL_CONST_CACHE_INVALIDATE
);
304 if (barriers
& GL_TEXTURE_FETCH_BARRIER_BIT
)
305 bits
|= PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE
;
307 if (barriers
& GL_TEXTURE_UPDATE_BARRIER_BIT
)
308 bits
|= PIPE_CONTROL_RENDER_TARGET_FLUSH
;
310 if (barriers
& GL_FRAMEBUFFER_BARRIER_BIT
)
311 bits
|= (PIPE_CONTROL_DEPTH_CACHE_FLUSH
|
312 PIPE_CONTROL_RENDER_TARGET_FLUSH
);
314 /* Typed surface messages are handled by the render cache on IVB, so we
315 * need to flush it too.
317 if (brw
->gen
== 7 && !brw
->is_haswell
)
318 bits
|= PIPE_CONTROL_RENDER_TARGET_FLUSH
;
320 brw_emit_pipe_control_flush(brw
, bits
);
324 brw_blend_barrier(struct gl_context
*ctx
)
326 struct brw_context
*brw
= brw_context(ctx
);
328 if (!ctx
->Extensions
.MESA_shader_framebuffer_fetch
) {
330 brw_emit_pipe_control_flush(brw
,
331 PIPE_CONTROL_RENDER_TARGET_FLUSH
|
332 PIPE_CONTROL_CS_STALL
);
333 brw_emit_pipe_control_flush(brw
,
334 PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE
);
336 brw_emit_pipe_control_flush(brw
,
337 PIPE_CONTROL_RENDER_TARGET_FLUSH
);
343 brw_get_scratch_bo(struct brw_context
*brw
,
344 struct brw_bo
**scratch_bo
, int size
)
346 struct brw_bo
*old_bo
= *scratch_bo
;
348 if (old_bo
&& old_bo
->size
< size
) {
349 brw_bo_unreference(old_bo
);
354 *scratch_bo
= brw_bo_alloc(brw
->bufmgr
, "scratch bo", size
, 4096);
359 * Reserve enough scratch space for the given stage to hold \p per_thread_size
360 * bytes times the given \p thread_count.
363 brw_alloc_stage_scratch(struct brw_context
*brw
,
364 struct brw_stage_state
*stage_state
,
365 unsigned per_thread_size
,
366 unsigned thread_count
)
368 if (stage_state
->per_thread_scratch
< per_thread_size
) {
369 stage_state
->per_thread_scratch
= per_thread_size
;
371 if (stage_state
->scratch_bo
)
372 brw_bo_unreference(stage_state
->scratch_bo
);
374 stage_state
->scratch_bo
=
375 brw_bo_alloc(brw
->bufmgr
, "shader scratch space",
376 per_thread_size
* thread_count
, 4096);
380 void brwInitFragProgFuncs( struct dd_function_table
*functions
)
382 assert(functions
->ProgramStringNotify
== _tnl_program_string
);
384 functions
->NewProgram
= brwNewProgram
;
385 functions
->DeleteProgram
= brwDeleteProgram
;
386 functions
->ProgramStringNotify
= brwProgramStringNotify
;
388 functions
->LinkShader
= brw_link_shader
;
390 functions
->MemoryBarrier
= brw_memory_barrier
;
391 functions
->BlendBarrier
= brw_blend_barrier
;
394 struct shader_times
{
401 brw_init_shader_time(struct brw_context
*brw
)
403 const int max_entries
= 2048;
404 brw
->shader_time
.bo
=
405 brw_bo_alloc(brw
->bufmgr
, "shader time",
406 max_entries
* BRW_SHADER_TIME_STRIDE
* 3, 4096);
407 brw
->shader_time
.names
= rzalloc_array(brw
, const char *, max_entries
);
408 brw
->shader_time
.ids
= rzalloc_array(brw
, int, max_entries
);
409 brw
->shader_time
.types
= rzalloc_array(brw
, enum shader_time_shader_type
,
411 brw
->shader_time
.cumulative
= rzalloc_array(brw
, struct shader_times
,
413 brw
->shader_time
.max_entries
= max_entries
;
417 compare_time(const void *a
, const void *b
)
419 uint64_t * const *a_val
= a
;
420 uint64_t * const *b_val
= b
;
422 /* We don't just subtract because we're turning the value to an int. */
423 if (**a_val
< **b_val
)
425 else if (**a_val
== **b_val
)
432 print_shader_time_line(const char *stage
, const char *name
,
433 int shader_num
, uint64_t time
, uint64_t total
)
435 fprintf(stderr
, "%-6s%-18s", stage
, name
);
438 fprintf(stderr
, "%4d: ", shader_num
);
440 fprintf(stderr
, " : ");
442 fprintf(stderr
, "%16lld (%7.2f Gcycles) %4.1f%%\n",
444 (double)time
/ 1000000000.0,
445 (double)time
/ total
* 100.0);
449 brw_report_shader_time(struct brw_context
*brw
)
451 if (!brw
->shader_time
.bo
|| !brw
->shader_time
.num_entries
)
454 uint64_t scaled
[brw
->shader_time
.num_entries
];
455 uint64_t *sorted
[brw
->shader_time
.num_entries
];
456 uint64_t total_by_type
[ST_CS
+ 1];
457 memset(total_by_type
, 0, sizeof(total_by_type
));
459 for (int i
= 0; i
< brw
->shader_time
.num_entries
; i
++) {
460 uint64_t written
= 0, reset
= 0;
461 enum shader_time_shader_type type
= brw
->shader_time
.types
[i
];
463 sorted
[i
] = &scaled
[i
];
473 written
= brw
->shader_time
.cumulative
[i
].written
;
474 reset
= brw
->shader_time
.cumulative
[i
].reset
;
478 /* I sometimes want to print things that aren't the 3 shader times.
479 * Just print the sum in that case.
486 uint64_t time
= brw
->shader_time
.cumulative
[i
].time
;
488 scaled
[i
] = time
/ written
* (written
+ reset
);
501 total_by_type
[type
] += scaled
[i
];
511 fprintf(stderr
, "No shader time collected yet\n");
515 qsort(sorted
, brw
->shader_time
.num_entries
, sizeof(sorted
[0]), compare_time
);
517 fprintf(stderr
, "\n");
518 fprintf(stderr
, "type ID cycles spent %% of total\n");
519 for (int s
= 0; s
< brw
->shader_time
.num_entries
; s
++) {
521 /* Work back from the sorted pointers times to a time to print. */
522 int i
= sorted
[s
] - scaled
;
527 int shader_num
= brw
->shader_time
.ids
[i
];
528 const char *shader_name
= brw
->shader_time
.names
[i
];
530 switch (brw
->shader_time
.types
[i
]) {
557 print_shader_time_line(stage
, shader_name
, shader_num
,
561 fprintf(stderr
, "\n");
562 print_shader_time_line("total", "vs", 0, total_by_type
[ST_VS
], total
);
563 print_shader_time_line("total", "tcs", 0, total_by_type
[ST_TCS
], total
);
564 print_shader_time_line("total", "tes", 0, total_by_type
[ST_TES
], total
);
565 print_shader_time_line("total", "gs", 0, total_by_type
[ST_GS
], total
);
566 print_shader_time_line("total", "fs8", 0, total_by_type
[ST_FS8
], total
);
567 print_shader_time_line("total", "fs16", 0, total_by_type
[ST_FS16
], total
);
568 print_shader_time_line("total", "cs", 0, total_by_type
[ST_CS
], total
);
572 brw_collect_shader_time(struct brw_context
*brw
)
574 if (!brw
->shader_time
.bo
)
577 /* This probably stalls on the last rendering. We could fix that by
578 * delaying reading the reports, but it doesn't look like it's a big
579 * overhead compared to the cost of tracking the time in the first place.
581 void *bo_map
= brw_bo_map(brw
, brw
->shader_time
.bo
, MAP_READ
| MAP_WRITE
);
583 for (int i
= 0; i
< brw
->shader_time
.num_entries
; i
++) {
584 uint32_t *times
= bo_map
+ i
* 3 * BRW_SHADER_TIME_STRIDE
;
586 brw
->shader_time
.cumulative
[i
].time
+= times
[BRW_SHADER_TIME_STRIDE
* 0 / 4];
587 brw
->shader_time
.cumulative
[i
].written
+= times
[BRW_SHADER_TIME_STRIDE
* 1 / 4];
588 brw
->shader_time
.cumulative
[i
].reset
+= times
[BRW_SHADER_TIME_STRIDE
* 2 / 4];
591 /* Zero the BO out to clear it out for our next collection.
593 memset(bo_map
, 0, brw
->shader_time
.bo
->size
);
594 brw_bo_unmap(brw
->shader_time
.bo
);
598 brw_collect_and_report_shader_time(struct brw_context
*brw
)
600 brw_collect_shader_time(brw
);
602 if (brw
->shader_time
.report_time
== 0 ||
603 get_time() - brw
->shader_time
.report_time
>= 1.0) {
604 brw_report_shader_time(brw
);
605 brw
->shader_time
.report_time
= get_time();
610 * Chooses an index in the shader_time buffer and sets up tracking information
613 * Note that this holds on to references to the underlying programs, which may
614 * change their lifetimes compared to normal operation.
617 brw_get_shader_time_index(struct brw_context
*brw
, struct gl_program
*prog
,
618 enum shader_time_shader_type type
, bool is_glsl_sh
)
620 int shader_time_index
= brw
->shader_time
.num_entries
++;
621 assert(shader_time_index
< brw
->shader_time
.max_entries
);
622 brw
->shader_time
.types
[shader_time_index
] = type
;
627 } else if (is_glsl_sh
) {
628 name
= prog
->info
.label
?
629 ralloc_strdup(brw
->shader_time
.names
, prog
->info
.label
) : "glsl";
634 brw
->shader_time
.names
[shader_time_index
] = name
;
635 brw
->shader_time
.ids
[shader_time_index
] = prog
->Id
;
637 return shader_time_index
;
641 brw_destroy_shader_time(struct brw_context
*brw
)
643 brw_bo_unreference(brw
->shader_time
.bo
);
644 brw
->shader_time
.bo
= NULL
;
648 brw_stage_prog_data_free(const void *p
)
650 struct brw_stage_prog_data
*prog_data
= (struct brw_stage_prog_data
*)p
;
652 ralloc_free(prog_data
->param
);
653 ralloc_free(prog_data
->pull_param
);
654 ralloc_free(prog_data
->image_param
);
658 brw_dump_arb_asm(const char *stage
, struct gl_program
*prog
)
660 fprintf(stderr
, "ARB_%s_program %d ir for native %s shader\n",
661 stage
, prog
->Id
, stage
);
662 _mesa_print_program(prog
);
666 brw_setup_tex_for_precompile(struct brw_context
*brw
,
667 struct brw_sampler_prog_key_data
*tex
,
668 struct gl_program
*prog
)
670 const bool has_shader_channel_select
= brw
->is_haswell
|| brw
->gen
>= 8;
671 unsigned sampler_count
= util_last_bit(prog
->SamplersUsed
);
672 for (unsigned i
= 0; i
< sampler_count
; i
++) {
673 if (!has_shader_channel_select
&& (prog
->ShadowSamplers
& (1 << i
))) {
674 /* Assume DEPTH_TEXTURE_MODE is the default: X, X, X, 1 */
676 MAKE_SWIZZLE4(SWIZZLE_X
, SWIZZLE_X
, SWIZZLE_X
, SWIZZLE_ONE
);
678 /* Color sampler: assume no swizzling. */
679 tex
->swizzles
[i
] = SWIZZLE_XYZW
;
685 * Sets up the starting offsets for the groups of binding table entries
686 * common to all pipeline stages.
688 * Unused groups are initialized to 0xd0d0d0d0 to make it obvious that they're
689 * unused but also make sure that addition of small offsets to them will
690 * trigger some of our asserts that surface indices are < BRW_MAX_SURFACES.
693 brw_assign_common_binding_table_offsets(const struct gen_device_info
*devinfo
,
694 const struct gl_program
*prog
,
695 struct brw_stage_prog_data
*stage_prog_data
,
696 uint32_t next_binding_table_offset
)
698 int num_textures
= util_last_bit(prog
->SamplersUsed
);
700 stage_prog_data
->binding_table
.texture_start
= next_binding_table_offset
;
701 next_binding_table_offset
+= num_textures
;
703 if (prog
->info
.num_ubos
) {
704 assert(prog
->info
.num_ubos
<= BRW_MAX_UBO
);
705 stage_prog_data
->binding_table
.ubo_start
= next_binding_table_offset
;
706 next_binding_table_offset
+= prog
->info
.num_ubos
;
708 stage_prog_data
->binding_table
.ubo_start
= 0xd0d0d0d0;
711 if (prog
->info
.num_ssbos
) {
712 assert(prog
->info
.num_ssbos
<= BRW_MAX_SSBO
);
713 stage_prog_data
->binding_table
.ssbo_start
= next_binding_table_offset
;
714 next_binding_table_offset
+= prog
->info
.num_ssbos
;
716 stage_prog_data
->binding_table
.ssbo_start
= 0xd0d0d0d0;
719 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
720 stage_prog_data
->binding_table
.shader_time_start
= next_binding_table_offset
;
721 next_binding_table_offset
++;
723 stage_prog_data
->binding_table
.shader_time_start
= 0xd0d0d0d0;
726 if (prog
->nir
->info
.uses_texture_gather
) {
727 if (devinfo
->gen
>= 8) {
728 stage_prog_data
->binding_table
.gather_texture_start
=
729 stage_prog_data
->binding_table
.texture_start
;
731 stage_prog_data
->binding_table
.gather_texture_start
= next_binding_table_offset
;
732 next_binding_table_offset
+= num_textures
;
735 stage_prog_data
->binding_table
.gather_texture_start
= 0xd0d0d0d0;
738 if (prog
->info
.num_abos
) {
739 stage_prog_data
->binding_table
.abo_start
= next_binding_table_offset
;
740 next_binding_table_offset
+= prog
->info
.num_abos
;
742 stage_prog_data
->binding_table
.abo_start
= 0xd0d0d0d0;
745 if (prog
->info
.num_images
) {
746 stage_prog_data
->binding_table
.image_start
= next_binding_table_offset
;
747 next_binding_table_offset
+= prog
->info
.num_images
;
749 stage_prog_data
->binding_table
.image_start
= 0xd0d0d0d0;
752 /* This may or may not be used depending on how the compile goes. */
753 stage_prog_data
->binding_table
.pull_constants_start
= next_binding_table_offset
;
754 next_binding_table_offset
++;
756 /* Plane 0 is just the regular texture section */
757 stage_prog_data
->binding_table
.plane_start
[0] = stage_prog_data
->binding_table
.texture_start
;
759 stage_prog_data
->binding_table
.plane_start
[1] = next_binding_table_offset
;
760 next_binding_table_offset
+= num_textures
;
762 stage_prog_data
->binding_table
.plane_start
[2] = next_binding_table_offset
;
763 next_binding_table_offset
+= num_textures
;
765 /* prog_data->base.binding_table.size will be set by brw_mark_surface_used. */
767 assert(next_binding_table_offset
<= BRW_MAX_SURFACES
);
768 return next_binding_table_offset
;