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 /* nir_shader may have been cloned so make sure shader_info is in sync */
113 if (nir
->info
!= &prog
->info
) {
114 const char *name
= prog
->info
.name
;
115 const char *label
= prog
->info
.label
;
116 prog
->info
= *nir
->info
;
117 prog
->info
.name
= name
;
118 prog
->info
.label
= label
;
122 NIR_PASS_V(nir
, nir_lower_samplers
, shader_prog
);
123 NIR_PASS_V(nir
, nir_lower_atomics
, shader_prog
);
130 get_new_program_id(struct intel_screen
*screen
)
132 static pthread_mutex_t m
= PTHREAD_MUTEX_INITIALIZER
;
133 pthread_mutex_lock(&m
);
134 unsigned id
= screen
->program_id
++;
135 pthread_mutex_unlock(&m
);
139 static struct gl_program
*brwNewProgram(struct gl_context
*ctx
, GLenum target
,
140 GLuint id
, bool is_arb_asm
)
142 struct brw_context
*brw
= brw_context(ctx
);
145 case GL_VERTEX_PROGRAM_ARB
:
146 case GL_TESS_CONTROL_PROGRAM_NV
:
147 case GL_TESS_EVALUATION_PROGRAM_NV
:
148 case GL_GEOMETRY_PROGRAM_NV
:
149 case GL_COMPUTE_PROGRAM_NV
: {
150 struct brw_program
*prog
= rzalloc(NULL
, struct brw_program
);
152 prog
->id
= get_new_program_id(brw
->screen
);
154 return _mesa_init_gl_program(&prog
->program
, target
, id
, is_arb_asm
);
160 case GL_FRAGMENT_PROGRAM_ARB
: {
161 struct brw_program
*prog
= rzalloc(NULL
, struct brw_program
);
164 prog
->id
= get_new_program_id(brw
->screen
);
166 return _mesa_init_gl_program(&prog
->program
, target
, id
, is_arb_asm
);
173 unreachable("Unsupported target in brwNewProgram()");
177 static void brwDeleteProgram( struct gl_context
*ctx
,
178 struct gl_program
*prog
)
180 struct brw_context
*brw
= brw_context(ctx
);
182 /* Beware! prog's refcount has reached zero, and it's about to be freed.
184 * In brw_upload_pipeline_state(), we compare brw->foo_program to
185 * ctx->FooProgram._Current, and flag BRW_NEW_FOO_PROGRAM if the
186 * pointer has changed.
188 * We cannot leave brw->foo_program as a dangling pointer to the dead
189 * program. malloc() may allocate the same memory for a new gl_program,
190 * causing us to see matching pointers...but totally different programs.
192 * We cannot set brw->foo_program to NULL, either. If we've deleted the
193 * active program, Mesa may set ctx->FooProgram._Current to NULL. That
194 * would cause us to see matching pointers (NULL == NULL), and fail to
195 * detect that a program has changed since our last draw.
197 * So, set it to a bogus gl_program pointer that will never match,
198 * causing us to properly reevaluate the state on our next draw.
200 * Getting this wrong causes heisenbugs which are very hard to catch,
201 * as you need a very specific allocation pattern to hit the problem.
203 static const struct gl_program deleted_program
;
205 if (brw
->vertex_program
== prog
)
206 brw
->vertex_program
= &deleted_program
;
208 if (brw
->tess_ctrl_program
== prog
)
209 brw
->tess_ctrl_program
= &deleted_program
;
211 if (brw
->tess_eval_program
== prog
)
212 brw
->tess_eval_program
= &deleted_program
;
214 if (brw
->geometry_program
== prog
)
215 brw
->geometry_program
= &deleted_program
;
217 if (brw
->fragment_program
== prog
)
218 brw
->fragment_program
= &deleted_program
;
220 if (brw
->compute_program
== prog
)
221 brw
->compute_program
= &deleted_program
;
223 _mesa_delete_program( ctx
, prog
);
228 brwProgramStringNotify(struct gl_context
*ctx
,
230 struct gl_program
*prog
)
232 assert(target
== GL_VERTEX_PROGRAM_ARB
|| !prog
->arb
.IsPositionInvariant
);
234 struct brw_context
*brw
= brw_context(ctx
);
235 const struct brw_compiler
*compiler
= brw
->screen
->compiler
;
238 case GL_FRAGMENT_PROGRAM_ARB
: {
239 struct brw_program
*newFP
= brw_program(prog
);
240 const struct brw_program
*curFP
=
241 brw_program_const(brw
->fragment_program
);
244 brw
->ctx
.NewDriverState
|= BRW_NEW_FRAGMENT_PROGRAM
;
245 newFP
->id
= get_new_program_id(brw
->screen
);
247 prog
->nir
= brw_create_nir(brw
, NULL
, prog
, MESA_SHADER_FRAGMENT
, true);
249 brw_fs_precompile(ctx
, prog
);
252 case GL_VERTEX_PROGRAM_ARB
: {
253 struct brw_program
*newVP
= brw_program(prog
);
254 const struct brw_program
*curVP
=
255 brw_program_const(brw
->vertex_program
);
258 brw
->ctx
.NewDriverState
|= BRW_NEW_VERTEX_PROGRAM
;
259 if (newVP
->program
.arb
.IsPositionInvariant
) {
260 _mesa_insert_mvp_code(ctx
, &newVP
->program
);
262 newVP
->id
= get_new_program_id(brw
->screen
);
264 /* Also tell tnl about it:
266 _tnl_program_string(ctx
, target
, prog
);
268 prog
->nir
= brw_create_nir(brw
, NULL
, prog
, MESA_SHADER_VERTEX
,
269 compiler
->scalar_stage
[MESA_SHADER_VERTEX
]);
271 brw_vs_precompile(ctx
, prog
);
276 * driver->ProgramStringNotify is only called for ARB programs, fixed
277 * function vertex programs, and ir_to_mesa (which isn't used by the
278 * i965 back-end). Therefore, even after geometry shaders are added,
279 * this function should only ever be called with a target of
280 * GL_VERTEX_PROGRAM_ARB or GL_FRAGMENT_PROGRAM_ARB.
282 unreachable("Unexpected target in brwProgramStringNotify");
289 brw_memory_barrier(struct gl_context
*ctx
, GLbitfield barriers
)
291 struct brw_context
*brw
= brw_context(ctx
);
292 unsigned bits
= (PIPE_CONTROL_DATA_CACHE_FLUSH
|
293 PIPE_CONTROL_NO_WRITE
|
294 PIPE_CONTROL_CS_STALL
);
295 assert(brw
->gen
>= 7 && brw
->gen
<= 9);
297 if (barriers
& (GL_VERTEX_ATTRIB_ARRAY_BARRIER_BIT
|
298 GL_ELEMENT_ARRAY_BARRIER_BIT
|
299 GL_COMMAND_BARRIER_BIT
))
300 bits
|= PIPE_CONTROL_VF_CACHE_INVALIDATE
;
302 if (barriers
& GL_UNIFORM_BARRIER_BIT
)
303 bits
|= (PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE
|
304 PIPE_CONTROL_CONST_CACHE_INVALIDATE
);
306 if (barriers
& GL_TEXTURE_FETCH_BARRIER_BIT
)
307 bits
|= PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE
;
309 if (barriers
& GL_TEXTURE_UPDATE_BARRIER_BIT
)
310 bits
|= PIPE_CONTROL_RENDER_TARGET_FLUSH
;
312 if (barriers
& GL_FRAMEBUFFER_BARRIER_BIT
)
313 bits
|= (PIPE_CONTROL_DEPTH_CACHE_FLUSH
|
314 PIPE_CONTROL_RENDER_TARGET_FLUSH
);
316 /* Typed surface messages are handled by the render cache on IVB, so we
317 * need to flush it too.
319 if (brw
->gen
== 7 && !brw
->is_haswell
)
320 bits
|= PIPE_CONTROL_RENDER_TARGET_FLUSH
;
322 brw_emit_pipe_control_flush(brw
, bits
);
326 brw_blend_barrier(struct gl_context
*ctx
)
328 struct brw_context
*brw
= brw_context(ctx
);
330 if (!ctx
->Extensions
.MESA_shader_framebuffer_fetch
) {
332 brw_emit_pipe_control_flush(brw
,
333 PIPE_CONTROL_RENDER_TARGET_FLUSH
|
334 PIPE_CONTROL_CS_STALL
);
335 brw_emit_pipe_control_flush(brw
,
336 PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE
);
338 brw_emit_pipe_control_flush(brw
,
339 PIPE_CONTROL_RENDER_TARGET_FLUSH
);
345 brw_get_scratch_bo(struct brw_context
*brw
,
346 drm_bacon_bo
**scratch_bo
, int size
)
348 drm_bacon_bo
*old_bo
= *scratch_bo
;
350 if (old_bo
&& old_bo
->size
< size
) {
351 drm_bacon_bo_unreference(old_bo
);
356 *scratch_bo
= drm_bacon_bo_alloc(brw
->bufmgr
, "scratch bo", size
, 4096);
361 * Reserve enough scratch space for the given stage to hold \p per_thread_size
362 * bytes times the given \p thread_count.
365 brw_alloc_stage_scratch(struct brw_context
*brw
,
366 struct brw_stage_state
*stage_state
,
367 unsigned per_thread_size
,
368 unsigned thread_count
)
370 if (stage_state
->per_thread_scratch
< per_thread_size
) {
371 stage_state
->per_thread_scratch
= per_thread_size
;
373 if (stage_state
->scratch_bo
)
374 drm_bacon_bo_unreference(stage_state
->scratch_bo
);
376 stage_state
->scratch_bo
=
377 drm_bacon_bo_alloc(brw
->bufmgr
, "shader scratch space",
378 per_thread_size
* thread_count
, 4096);
382 void brwInitFragProgFuncs( struct dd_function_table
*functions
)
384 assert(functions
->ProgramStringNotify
== _tnl_program_string
);
386 functions
->NewProgram
= brwNewProgram
;
387 functions
->DeleteProgram
= brwDeleteProgram
;
388 functions
->ProgramStringNotify
= brwProgramStringNotify
;
390 functions
->LinkShader
= brw_link_shader
;
392 functions
->MemoryBarrier
= brw_memory_barrier
;
393 functions
->BlendBarrier
= brw_blend_barrier
;
396 struct shader_times
{
403 brw_init_shader_time(struct brw_context
*brw
)
405 const int max_entries
= 2048;
406 brw
->shader_time
.bo
=
407 drm_bacon_bo_alloc(brw
->bufmgr
, "shader time",
408 max_entries
* BRW_SHADER_TIME_STRIDE
* 3, 4096);
409 brw
->shader_time
.names
= rzalloc_array(brw
, const char *, max_entries
);
410 brw
->shader_time
.ids
= rzalloc_array(brw
, int, max_entries
);
411 brw
->shader_time
.types
= rzalloc_array(brw
, enum shader_time_shader_type
,
413 brw
->shader_time
.cumulative
= rzalloc_array(brw
, struct shader_times
,
415 brw
->shader_time
.max_entries
= max_entries
;
419 compare_time(const void *a
, const void *b
)
421 uint64_t * const *a_val
= a
;
422 uint64_t * const *b_val
= b
;
424 /* We don't just subtract because we're turning the value to an int. */
425 if (**a_val
< **b_val
)
427 else if (**a_val
== **b_val
)
434 print_shader_time_line(const char *stage
, const char *name
,
435 int shader_num
, uint64_t time
, uint64_t total
)
437 fprintf(stderr
, "%-6s%-18s", stage
, name
);
440 fprintf(stderr
, "%4d: ", shader_num
);
442 fprintf(stderr
, " : ");
444 fprintf(stderr
, "%16lld (%7.2f Gcycles) %4.1f%%\n",
446 (double)time
/ 1000000000.0,
447 (double)time
/ total
* 100.0);
451 brw_report_shader_time(struct brw_context
*brw
)
453 if (!brw
->shader_time
.bo
|| !brw
->shader_time
.num_entries
)
456 uint64_t scaled
[brw
->shader_time
.num_entries
];
457 uint64_t *sorted
[brw
->shader_time
.num_entries
];
458 uint64_t total_by_type
[ST_CS
+ 1];
459 memset(total_by_type
, 0, sizeof(total_by_type
));
461 for (int i
= 0; i
< brw
->shader_time
.num_entries
; i
++) {
462 uint64_t written
= 0, reset
= 0;
463 enum shader_time_shader_type type
= brw
->shader_time
.types
[i
];
465 sorted
[i
] = &scaled
[i
];
475 written
= brw
->shader_time
.cumulative
[i
].written
;
476 reset
= brw
->shader_time
.cumulative
[i
].reset
;
480 /* I sometimes want to print things that aren't the 3 shader times.
481 * Just print the sum in that case.
488 uint64_t time
= brw
->shader_time
.cumulative
[i
].time
;
490 scaled
[i
] = time
/ written
* (written
+ reset
);
503 total_by_type
[type
] += scaled
[i
];
513 fprintf(stderr
, "No shader time collected yet\n");
517 qsort(sorted
, brw
->shader_time
.num_entries
, sizeof(sorted
[0]), compare_time
);
519 fprintf(stderr
, "\n");
520 fprintf(stderr
, "type ID cycles spent %% of total\n");
521 for (int s
= 0; s
< brw
->shader_time
.num_entries
; s
++) {
523 /* Work back from the sorted pointers times to a time to print. */
524 int i
= sorted
[s
] - scaled
;
529 int shader_num
= brw
->shader_time
.ids
[i
];
530 const char *shader_name
= brw
->shader_time
.names
[i
];
532 switch (brw
->shader_time
.types
[i
]) {
559 print_shader_time_line(stage
, shader_name
, shader_num
,
563 fprintf(stderr
, "\n");
564 print_shader_time_line("total", "vs", 0, total_by_type
[ST_VS
], total
);
565 print_shader_time_line("total", "tcs", 0, total_by_type
[ST_TCS
], total
);
566 print_shader_time_line("total", "tes", 0, total_by_type
[ST_TES
], total
);
567 print_shader_time_line("total", "gs", 0, total_by_type
[ST_GS
], total
);
568 print_shader_time_line("total", "fs8", 0, total_by_type
[ST_FS8
], total
);
569 print_shader_time_line("total", "fs16", 0, total_by_type
[ST_FS16
], total
);
570 print_shader_time_line("total", "cs", 0, total_by_type
[ST_CS
], total
);
574 brw_collect_shader_time(struct brw_context
*brw
)
576 if (!brw
->shader_time
.bo
)
579 /* This probably stalls on the last rendering. We could fix that by
580 * delaying reading the reports, but it doesn't look like it's a big
581 * overhead compared to the cost of tracking the time in the first place.
583 drm_bacon_bo_map(brw
->shader_time
.bo
, true);
584 void *bo_map
= brw
->shader_time
.bo
->virtual;
586 for (int i
= 0; i
< brw
->shader_time
.num_entries
; i
++) {
587 uint32_t *times
= bo_map
+ i
* 3 * BRW_SHADER_TIME_STRIDE
;
589 brw
->shader_time
.cumulative
[i
].time
+= times
[BRW_SHADER_TIME_STRIDE
* 0 / 4];
590 brw
->shader_time
.cumulative
[i
].written
+= times
[BRW_SHADER_TIME_STRIDE
* 1 / 4];
591 brw
->shader_time
.cumulative
[i
].reset
+= times
[BRW_SHADER_TIME_STRIDE
* 2 / 4];
594 /* Zero the BO out to clear it out for our next collection.
596 memset(bo_map
, 0, brw
->shader_time
.bo
->size
);
597 drm_bacon_bo_unmap(brw
->shader_time
.bo
);
601 brw_collect_and_report_shader_time(struct brw_context
*brw
)
603 brw_collect_shader_time(brw
);
605 if (brw
->shader_time
.report_time
== 0 ||
606 get_time() - brw
->shader_time
.report_time
>= 1.0) {
607 brw_report_shader_time(brw
);
608 brw
->shader_time
.report_time
= get_time();
613 * Chooses an index in the shader_time buffer and sets up tracking information
616 * Note that this holds on to references to the underlying programs, which may
617 * change their lifetimes compared to normal operation.
620 brw_get_shader_time_index(struct brw_context
*brw
, struct gl_program
*prog
,
621 enum shader_time_shader_type type
, bool is_glsl_sh
)
623 int shader_time_index
= brw
->shader_time
.num_entries
++;
624 assert(shader_time_index
< brw
->shader_time
.max_entries
);
625 brw
->shader_time
.types
[shader_time_index
] = type
;
630 } else if (is_glsl_sh
) {
631 name
= prog
->info
.label
?
632 ralloc_strdup(brw
->shader_time
.names
, prog
->info
.label
) : "glsl";
637 brw
->shader_time
.names
[shader_time_index
] = name
;
638 brw
->shader_time
.ids
[shader_time_index
] = prog
->Id
;
640 return shader_time_index
;
644 brw_destroy_shader_time(struct brw_context
*brw
)
646 drm_bacon_bo_unreference(brw
->shader_time
.bo
);
647 brw
->shader_time
.bo
= NULL
;
651 brw_stage_prog_data_free(const void *p
)
653 struct brw_stage_prog_data
*prog_data
= (struct brw_stage_prog_data
*)p
;
655 ralloc_free(prog_data
->param
);
656 ralloc_free(prog_data
->pull_param
);
657 ralloc_free(prog_data
->image_param
);
661 brw_dump_arb_asm(const char *stage
, struct gl_program
*prog
)
663 fprintf(stderr
, "ARB_%s_program %d ir for native %s shader\n",
664 stage
, prog
->Id
, stage
);
665 _mesa_print_program(prog
);
669 brw_setup_tex_for_precompile(struct brw_context
*brw
,
670 struct brw_sampler_prog_key_data
*tex
,
671 struct gl_program
*prog
)
673 const bool has_shader_channel_select
= brw
->is_haswell
|| brw
->gen
>= 8;
674 unsigned sampler_count
= util_last_bit(prog
->SamplersUsed
);
675 for (unsigned i
= 0; i
< sampler_count
; i
++) {
676 if (!has_shader_channel_select
&& (prog
->ShadowSamplers
& (1 << i
))) {
677 /* Assume DEPTH_TEXTURE_MODE is the default: X, X, X, 1 */
679 MAKE_SWIZZLE4(SWIZZLE_X
, SWIZZLE_X
, SWIZZLE_X
, SWIZZLE_ONE
);
681 /* Color sampler: assume no swizzling. */
682 tex
->swizzles
[i
] = SWIZZLE_XYZW
;
688 * Sets up the starting offsets for the groups of binding table entries
689 * common to all pipeline stages.
691 * Unused groups are initialized to 0xd0d0d0d0 to make it obvious that they're
692 * unused but also make sure that addition of small offsets to them will
693 * trigger some of our asserts that surface indices are < BRW_MAX_SURFACES.
696 brw_assign_common_binding_table_offsets(const struct gen_device_info
*devinfo
,
697 const struct gl_program
*prog
,
698 struct brw_stage_prog_data
*stage_prog_data
,
699 uint32_t next_binding_table_offset
)
701 int num_textures
= util_last_bit(prog
->SamplersUsed
);
703 stage_prog_data
->binding_table
.texture_start
= next_binding_table_offset
;
704 next_binding_table_offset
+= num_textures
;
706 if (prog
->info
.num_ubos
) {
707 assert(prog
->info
.num_ubos
<= BRW_MAX_UBO
);
708 stage_prog_data
->binding_table
.ubo_start
= next_binding_table_offset
;
709 next_binding_table_offset
+= prog
->info
.num_ubos
;
711 stage_prog_data
->binding_table
.ubo_start
= 0xd0d0d0d0;
714 if (prog
->info
.num_ssbos
) {
715 assert(prog
->info
.num_ssbos
<= BRW_MAX_SSBO
);
716 stage_prog_data
->binding_table
.ssbo_start
= next_binding_table_offset
;
717 next_binding_table_offset
+= prog
->info
.num_ssbos
;
719 stage_prog_data
->binding_table
.ssbo_start
= 0xd0d0d0d0;
722 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
723 stage_prog_data
->binding_table
.shader_time_start
= next_binding_table_offset
;
724 next_binding_table_offset
++;
726 stage_prog_data
->binding_table
.shader_time_start
= 0xd0d0d0d0;
729 if (prog
->nir
->info
->uses_texture_gather
) {
730 if (devinfo
->gen
>= 8) {
731 stage_prog_data
->binding_table
.gather_texture_start
=
732 stage_prog_data
->binding_table
.texture_start
;
734 stage_prog_data
->binding_table
.gather_texture_start
= next_binding_table_offset
;
735 next_binding_table_offset
+= num_textures
;
738 stage_prog_data
->binding_table
.gather_texture_start
= 0xd0d0d0d0;
741 if (prog
->info
.num_abos
) {
742 stage_prog_data
->binding_table
.abo_start
= next_binding_table_offset
;
743 next_binding_table_offset
+= prog
->info
.num_abos
;
745 stage_prog_data
->binding_table
.abo_start
= 0xd0d0d0d0;
748 if (prog
->info
.num_images
) {
749 stage_prog_data
->binding_table
.image_start
= next_binding_table_offset
;
750 next_binding_table_offset
+= prog
->info
.num_images
;
752 stage_prog_data
->binding_table
.image_start
= 0xd0d0d0d0;
755 /* This may or may not be used depending on how the compile goes. */
756 stage_prog_data
->binding_table
.pull_constants_start
= next_binding_table_offset
;
757 next_binding_table_offset
++;
759 /* Plane 0 is just the regular texture section */
760 stage_prog_data
->binding_table
.plane_start
[0] = stage_prog_data
->binding_table
.texture_start
;
762 stage_prog_data
->binding_table
.plane_start
[1] = next_binding_table_offset
;
763 next_binding_table_offset
+= num_textures
;
765 stage_prog_data
->binding_table
.plane_start
[2] = next_binding_table_offset
;
766 next_binding_table_offset
+= num_textures
;
768 /* prog_data->base.binding_table.size will be set by brw_mark_surface_used. */
770 assert(next_binding_table_offset
<= BRW_MAX_SURFACES
);
771 return next_binding_table_offset
;