2 * Copyright © 2014-2017 Broadcom
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
25 #include "util/u_format.h"
26 #include "util/u_math.h"
27 #include "util/u_memory.h"
28 #include "util/ralloc.h"
29 #include "util/hash_table.h"
30 #include "util/u_upload_mgr.h"
31 #include "tgsi/tgsi_dump.h"
32 #include "tgsi/tgsi_parse.h"
33 #include "compiler/nir/nir.h"
34 #include "compiler/nir/nir_builder.h"
35 #include "nir/tgsi_to_nir.h"
36 #include "compiler/v3d_compiler.h"
37 #include "v3d_context.h"
38 #include "broadcom/cle/v3d_packet_v33_pack.h"
39 #include "mesa/state_tracker/st_glsl_types.h"
41 static struct v3d_compiled_shader
*
42 v3d_get_compiled_shader(struct v3d_context
*v3d
, struct v3d_key
*key
);
44 v3d_setup_shared_precompile_key(struct v3d_uncompiled_shader
*uncompiled
,
47 static gl_varying_slot
48 v3d_get_slot_for_driver_location(nir_shader
*s
, uint32_t driver_location
)
50 nir_foreach_variable(var
, &s
->outputs
) {
51 if (var
->data
.driver_location
== driver_location
) {
52 return var
->data
.location
;
60 * Precomputes the TRANSFORM_FEEDBACK_OUTPUT_DATA_SPEC array for the shader.
62 * A shader can have 16 of these specs, and each one of them can write up to
63 * 16 dwords. Since we allow a total of 64 transform feedback output
64 * components (not 16 vectors), we have to group the writes of multiple
65 * varyings together in a single data spec.
68 v3d_set_transform_feedback_outputs(struct v3d_uncompiled_shader
*so
,
69 const struct pipe_stream_output_info
*stream_output
)
71 if (!stream_output
->num_outputs
)
74 struct v3d_varying_slot slots
[PIPE_MAX_SO_OUTPUTS
* 4];
77 for (int buffer
= 0; buffer
< PIPE_MAX_SO_BUFFERS
; buffer
++) {
78 uint32_t buffer_offset
= 0;
79 uint32_t vpm_start
= slot_count
;
81 for (int i
= 0; i
< stream_output
->num_outputs
; i
++) {
82 const struct pipe_stream_output
*output
=
83 &stream_output
->output
[i
];
85 if (output
->output_buffer
!= buffer
)
88 /* We assume that the SO outputs appear in increasing
89 * order in the buffer.
91 assert(output
->dst_offset
>= buffer_offset
);
93 /* Pad any undefined slots in the output */
94 for (int j
= buffer_offset
; j
< output
->dst_offset
; j
++) {
96 v3d_slot_from_slot_and_component(VARYING_SLOT_POS
, 0);
101 /* Set the coordinate shader up to output the
102 * components of this varying.
104 for (int j
= 0; j
< output
->num_components
; j
++) {
105 gl_varying_slot slot
=
106 v3d_get_slot_for_driver_location(so
->base
.ir
.nir
, output
->register_index
);
109 v3d_slot_from_slot_and_component(slot
,
110 output
->start_component
+ j
);
116 uint32_t vpm_size
= slot_count
- vpm_start
;
120 uint32_t vpm_start_offset
= vpm_start
+ 6;
123 uint32_t write_size
= MIN2(vpm_size
, 1 << 4);
125 struct V3D33_TRANSFORM_FEEDBACK_OUTPUT_DATA_SPEC unpacked
= {
126 /* We need the offset from the coordinate shader's VPM
127 * output block, which has the [X, Y, Z, W, Xs, Ys]
128 * values at the start.
130 .first_shaded_vertex_value_to_output
= vpm_start_offset
,
131 .number_of_consecutive_vertex_values_to_output_as_32_bit_values
= write_size
,
132 .output_buffer_to_write_to
= buffer
,
136 assert(unpacked
.first_shaded_vertex_value_to_output
!= 8 ||
137 so
->num_tf_specs
!= 0);
139 assert(so
->num_tf_specs
!= ARRAY_SIZE(so
->tf_specs
));
140 V3D33_TRANSFORM_FEEDBACK_OUTPUT_DATA_SPEC_pack(NULL
,
141 (void *)&so
->tf_specs
[so
->num_tf_specs
],
144 /* If point size is being written by the shader, then
145 * all the VPM start offsets are shifted up by one.
146 * We won't know that until the variant is compiled,
149 unpacked
.first_shaded_vertex_value_to_output
++;
152 assert(unpacked
.first_shaded_vertex_value_to_output
!= 8 ||
153 so
->num_tf_specs
!= 0);
155 V3D33_TRANSFORM_FEEDBACK_OUTPUT_DATA_SPEC_pack(NULL
,
156 (void *)&so
->tf_specs_psiz
[so
->num_tf_specs
],
159 vpm_start_offset
+= write_size
;
160 vpm_size
-= write_size
;
162 so
->base
.stream_output
.stride
[buffer
] =
163 stream_output
->stride
[buffer
];
166 so
->num_tf_outputs
= slot_count
;
167 so
->tf_outputs
= ralloc_array(so
->base
.ir
.nir
, struct v3d_varying_slot
,
169 memcpy(so
->tf_outputs
, slots
, sizeof(*slots
) * slot_count
);
173 type_size(const struct glsl_type
*type
)
175 return glsl_count_attribute_slots(type
, false);
179 uniforms_type_size(const struct glsl_type
*type
)
181 return st_glsl_storage_type_size(type
, false);
185 * Precompiles a shader variant at shader state creation time if
186 * V3D_DEBUG=precompile is set. Used for shader-db
187 * (https://gitlab.freedesktop.org/mesa/shader-db)
190 v3d_shader_precompile(struct v3d_context
*v3d
,
191 struct v3d_uncompiled_shader
*so
)
193 nir_shader
*s
= so
->base
.ir
.nir
;
195 if (s
->info
.stage
== MESA_SHADER_FRAGMENT
) {
196 struct v3d_fs_key key
= {
197 .base
.shader_state
= so
,
200 v3d_setup_shared_precompile_key(so
, &key
.base
);
201 v3d_get_compiled_shader(v3d
, &key
.base
);
203 struct v3d_vs_key key
= {
204 .base
.shader_state
= so
,
207 v3d_setup_shared_precompile_key(so
, &key
.base
);
209 /* Compile VS: All outputs */
210 for (int vary
= 0; vary
< 64; vary
++) {
211 if (!(s
->info
.outputs_written
& (1ull << vary
)))
213 for (int i
= 0; i
< 4; i
++) {
214 key
.fs_inputs
[key
.num_fs_inputs
++] =
215 v3d_slot_from_slot_and_component(vary
,
220 v3d_get_compiled_shader(v3d
, &key
.base
);
222 /* Compile VS bin shader: only position (XXX: include TF) */
224 key
.num_fs_inputs
= 0;
225 for (int i
= 0; i
< 4; i
++) {
226 key
.fs_inputs
[key
.num_fs_inputs
++] =
227 v3d_slot_from_slot_and_component(VARYING_SLOT_POS
,
230 v3d_get_compiled_shader(v3d
, &key
.base
);
235 v3d_shader_state_create(struct pipe_context
*pctx
,
236 const struct pipe_shader_state
*cso
)
238 struct v3d_context
*v3d
= v3d_context(pctx
);
239 struct v3d_uncompiled_shader
*so
= CALLOC_STRUCT(v3d_uncompiled_shader
);
243 so
->program_id
= v3d
->next_uncompiled_program_id
++;
247 if (cso
->type
== PIPE_SHADER_IR_NIR
) {
248 /* The backend takes ownership of the NIR shader on state
253 NIR_PASS_V(s
, nir_lower_io
, nir_var_uniform
,
255 (nir_lower_io_options
)0);
257 assert(cso
->type
== PIPE_SHADER_IR_TGSI
);
259 if (V3D_DEBUG
& V3D_DEBUG_TGSI
) {
260 fprintf(stderr
, "prog %d TGSI:\n",
262 tgsi_dump(cso
->tokens
, 0);
263 fprintf(stderr
, "\n");
265 s
= tgsi_to_nir(cso
->tokens
, &v3d_nir_options
);
270 nir_variable_mode lower_mode
= nir_var_all
& ~nir_var_uniform
;
271 if (s
->info
.stage
== MESA_SHADER_VERTEX
)
272 lower_mode
&= ~(nir_var_shader_in
| nir_var_shader_out
);
273 NIR_PASS_V(s
, nir_lower_io
, lower_mode
,
275 (nir_lower_io_options
)0);
277 NIR_PASS_V(s
, nir_opt_global_to_local
);
278 NIR_PASS_V(s
, nir_lower_regs_to_ssa
);
279 NIR_PASS_V(s
, nir_normalize_cubemap_coords
);
281 NIR_PASS_V(s
, nir_lower_load_const_to_scalar
);
285 NIR_PASS_V(s
, nir_remove_dead_variables
, nir_var_local
);
287 /* Garbage collect dead instructions */
290 so
->base
.type
= PIPE_SHADER_IR_NIR
;
293 v3d_set_transform_feedback_outputs(so
, &cso
->stream_output
);
295 if (V3D_DEBUG
& (V3D_DEBUG_NIR
|
296 v3d_debug_flag_for_shader_stage(s
->info
.stage
))) {
297 fprintf(stderr
, "%s prog %d NIR:\n",
298 gl_shader_stage_name(s
->info
.stage
),
300 nir_print_shader(s
, stderr
);
301 fprintf(stderr
, "\n");
304 if (V3D_DEBUG
& V3D_DEBUG_PRECOMPILE
)
305 v3d_shader_precompile(v3d
, so
);
311 v3d_shader_debug_output(const char *message
, void *data
)
313 struct v3d_context
*v3d
= data
;
315 pipe_debug_message(&v3d
->debug
, SHADER_INFO
, "%s", message
);
318 static struct v3d_compiled_shader
*
319 v3d_get_compiled_shader(struct v3d_context
*v3d
, struct v3d_key
*key
)
321 struct v3d_uncompiled_shader
*shader_state
= key
->shader_state
;
322 nir_shader
*s
= shader_state
->base
.ir
.nir
;
324 struct hash_table
*ht
;
326 if (s
->info
.stage
== MESA_SHADER_FRAGMENT
) {
328 key_size
= sizeof(struct v3d_fs_key
);
331 key_size
= sizeof(struct v3d_vs_key
);
334 struct hash_entry
*entry
= _mesa_hash_table_search(ht
, key
);
338 struct v3d_compiled_shader
*shader
=
339 rzalloc(NULL
, struct v3d_compiled_shader
);
341 int program_id
= shader_state
->program_id
;
343 p_atomic_inc_return(&shader_state
->compiled_variant_count
);
345 uint32_t shader_size
;
347 switch (s
->info
.stage
) {
348 case MESA_SHADER_VERTEX
:
349 shader
->prog_data
.vs
= rzalloc(shader
, struct v3d_vs_prog_data
);
351 qpu_insts
= v3d_compile_vs(v3d
->screen
->compiler
,
352 (struct v3d_vs_key
*)key
,
353 shader
->prog_data
.vs
, s
,
354 v3d_shader_debug_output
,
356 program_id
, variant_id
,
359 case MESA_SHADER_FRAGMENT
:
360 shader
->prog_data
.fs
= rzalloc(shader
, struct v3d_fs_prog_data
);
362 qpu_insts
= v3d_compile_fs(v3d
->screen
->compiler
,
363 (struct v3d_fs_key
*)key
,
364 shader
->prog_data
.fs
, s
,
365 v3d_shader_debug_output
,
367 program_id
, variant_id
,
371 unreachable("bad stage");
374 v3d_set_shader_uniform_dirty_flags(shader
);
376 u_upload_data(v3d
->state_uploader
, 0, shader_size
, 8,
377 qpu_insts
, &shader
->offset
, &shader
->resource
);
381 struct v3d_key
*dup_key
;
382 dup_key
= ralloc_size(shader
, key_size
);
383 memcpy(dup_key
, key
, key_size
);
384 _mesa_hash_table_insert(ht
, dup_key
, shader
);
386 if (shader
->prog_data
.base
->spill_size
>
387 v3d
->prog
.spill_size_per_thread
) {
388 /* Max 4 QPUs per slice, 3 slices per core. We only do single
389 * core so far. This overallocates memory on smaller cores.
391 int total_spill_size
=
392 4 * 3 * shader
->prog_data
.base
->spill_size
;
394 v3d_bo_unreference(&v3d
->prog
.spill_bo
);
395 v3d
->prog
.spill_bo
= v3d_bo_alloc(v3d
->screen
,
396 total_spill_size
, "spill");
397 v3d
->prog
.spill_size_per_thread
=
398 shader
->prog_data
.base
->spill_size
;
405 v3d_free_compiled_shader(struct v3d_compiled_shader
*shader
)
407 pipe_resource_reference(&shader
->resource
, NULL
);
412 v3d_setup_shared_key(struct v3d_context
*v3d
, struct v3d_key
*key
,
413 struct v3d_texture_stateobj
*texstate
)
415 const struct v3d_device_info
*devinfo
= &v3d
->screen
->devinfo
;
417 for (int i
= 0; i
< texstate
->num_textures
; i
++) {
418 struct pipe_sampler_view
*sampler
= texstate
->textures
[i
];
419 struct v3d_sampler_view
*v3d_sampler
= v3d_sampler_view(sampler
);
420 struct pipe_sampler_state
*sampler_state
=
421 texstate
->samplers
[i
];
426 key
->tex
[i
].return_size
=
427 v3d_get_tex_return_size(devinfo
,
429 sampler_state
->compare_mode
);
431 /* For 16-bit, we set up the sampler to always return 2
432 * channels (meaning no recompiles for most statechanges),
433 * while for 32 we actually scale the returns with channels.
435 if (key
->tex
[i
].return_size
== 16) {
436 key
->tex
[i
].return_channels
= 2;
437 } else if (devinfo
->ver
> 40) {
438 key
->tex
[i
].return_channels
= 4;
440 key
->tex
[i
].return_channels
=
441 v3d_get_tex_return_channels(devinfo
,
445 if (key
->tex
[i
].return_size
== 32 && devinfo
->ver
< 40) {
446 memcpy(key
->tex
[i
].swizzle
,
447 v3d_sampler
->swizzle
,
448 sizeof(v3d_sampler
->swizzle
));
450 /* For 16-bit returns, we let the sampler state handle
453 key
->tex
[i
].swizzle
[0] = PIPE_SWIZZLE_X
;
454 key
->tex
[i
].swizzle
[1] = PIPE_SWIZZLE_Y
;
455 key
->tex
[i
].swizzle
[2] = PIPE_SWIZZLE_Z
;
456 key
->tex
[i
].swizzle
[3] = PIPE_SWIZZLE_W
;
460 key
->tex
[i
].clamp_s
=
461 sampler_state
->wrap_s
== PIPE_TEX_WRAP_CLAMP
;
462 key
->tex
[i
].clamp_t
=
463 sampler_state
->wrap_t
== PIPE_TEX_WRAP_CLAMP
;
464 key
->tex
[i
].clamp_r
=
465 sampler_state
->wrap_r
== PIPE_TEX_WRAP_CLAMP
;
469 key
->ucp_enables
= v3d
->rasterizer
->base
.clip_plane_enable
;
473 v3d_setup_shared_precompile_key(struct v3d_uncompiled_shader
*uncompiled
,
476 nir_shader
*s
= uncompiled
->base
.ir
.nir
;
478 for (int i
= 0; i
< s
->info
.num_textures
; i
++) {
479 key
->tex
[i
].return_size
= 16;
480 key
->tex
[i
].return_channels
= 2;
482 key
->tex
[i
].swizzle
[0] = PIPE_SWIZZLE_X
;
483 key
->tex
[i
].swizzle
[1] = PIPE_SWIZZLE_Y
;
484 key
->tex
[i
].swizzle
[2] = PIPE_SWIZZLE_Z
;
485 key
->tex
[i
].swizzle
[3] = PIPE_SWIZZLE_W
;
490 v3d_update_compiled_fs(struct v3d_context
*v3d
, uint8_t prim_mode
)
492 struct v3d_job
*job
= v3d
->job
;
493 struct v3d_fs_key local_key
;
494 struct v3d_fs_key
*key
= &local_key
;
496 if (!(v3d
->dirty
& (VC5_DIRTY_PRIM_MODE
|
498 VC5_DIRTY_FRAMEBUFFER
|
500 VC5_DIRTY_RASTERIZER
|
501 VC5_DIRTY_SAMPLE_STATE
|
503 VC5_DIRTY_UNCOMPILED_FS
))) {
507 memset(key
, 0, sizeof(*key
));
508 v3d_setup_shared_key(v3d
, &key
->base
, &v3d
->tex
[PIPE_SHADER_FRAGMENT
]);
509 key
->base
.shader_state
= v3d
->prog
.bind_fs
;
510 key
->is_points
= (prim_mode
== PIPE_PRIM_POINTS
);
511 key
->is_lines
= (prim_mode
>= PIPE_PRIM_LINES
&&
512 prim_mode
<= PIPE_PRIM_LINE_STRIP
);
513 key
->clamp_color
= v3d
->rasterizer
->base
.clamp_fragment_color
;
514 if (v3d
->blend
->base
.logicop_enable
) {
515 key
->logicop_func
= v3d
->blend
->base
.logicop_func
;
517 key
->logicop_func
= PIPE_LOGICOP_COPY
;
520 key
->msaa
= v3d
->rasterizer
->base
.multisample
;
521 key
->sample_coverage
= (v3d
->rasterizer
->base
.multisample
&&
522 v3d
->sample_mask
!= (1 << VC5_MAX_SAMPLES
) - 1);
523 key
->sample_alpha_to_coverage
= v3d
->blend
->base
.alpha_to_coverage
;
524 key
->sample_alpha_to_one
= v3d
->blend
->base
.alpha_to_one
;
527 key
->depth_enabled
= (v3d
->zsa
->base
.depth
.enabled
||
528 v3d
->zsa
->base
.stencil
[0].enabled
);
529 if (v3d
->zsa
->base
.alpha
.enabled
) {
530 key
->alpha_test
= true;
531 key
->alpha_test_func
= v3d
->zsa
->base
.alpha
.func
;
534 /* gl_FragColor's propagation to however many bound color buffers
535 * there are means that the buffer count needs to be in the key.
537 key
->nr_cbufs
= v3d
->framebuffer
.nr_cbufs
;
538 key
->swap_color_rb
= v3d
->swap_color_rb
;
540 for (int i
= 0; i
< key
->nr_cbufs
; i
++) {
541 struct pipe_surface
*cbuf
= v3d
->framebuffer
.cbufs
[i
];
545 const struct util_format_description
*desc
=
546 util_format_description(cbuf
->format
);
548 if (desc
->channel
[0].type
== UTIL_FORMAT_TYPE_FLOAT
&&
549 desc
->channel
[0].size
== 32) {
550 key
->f32_color_rb
|= 1 << i
;
553 if (v3d
->prog
.bind_fs
->was_tgsi
) {
554 if (util_format_is_pure_uint(cbuf
->format
))
555 key
->uint_color_rb
|= 1 << i
;
556 else if (util_format_is_pure_sint(cbuf
->format
))
557 key
->int_color_rb
|= 1 << i
;
561 if (key
->is_points
) {
562 key
->point_sprite_mask
=
563 v3d
->rasterizer
->base
.sprite_coord_enable
;
564 key
->point_coord_upper_left
=
565 (v3d
->rasterizer
->base
.sprite_coord_mode
==
566 PIPE_SPRITE_COORD_UPPER_LEFT
);
569 key
->light_twoside
= v3d
->rasterizer
->base
.light_twoside
;
570 key
->shade_model_flat
= v3d
->rasterizer
->base
.flatshade
;
572 struct v3d_compiled_shader
*old_fs
= v3d
->prog
.fs
;
573 v3d
->prog
.fs
= v3d_get_compiled_shader(v3d
, &key
->base
);
574 if (v3d
->prog
.fs
== old_fs
)
577 v3d
->dirty
|= VC5_DIRTY_COMPILED_FS
;
580 if (v3d
->prog
.fs
->prog_data
.fs
->flat_shade_flags
!=
581 old_fs
->prog_data
.fs
->flat_shade_flags
) {
582 v3d
->dirty
|= VC5_DIRTY_FLAT_SHADE_FLAGS
;
585 if (v3d
->prog
.fs
->prog_data
.fs
->noperspective_flags
!=
586 old_fs
->prog_data
.fs
->noperspective_flags
) {
587 v3d
->dirty
|= VC5_DIRTY_NOPERSPECTIVE_FLAGS
;
590 if (v3d
->prog
.fs
->prog_data
.fs
->centroid_flags
!=
591 old_fs
->prog_data
.fs
->centroid_flags
) {
592 v3d
->dirty
|= VC5_DIRTY_CENTROID_FLAGS
;
596 if (old_fs
&& memcmp(v3d
->prog
.fs
->prog_data
.fs
->input_slots
,
597 old_fs
->prog_data
.fs
->input_slots
,
598 sizeof(v3d
->prog
.fs
->prog_data
.fs
->input_slots
))) {
599 v3d
->dirty
|= VC5_DIRTY_FS_INPUTS
;
604 v3d_update_compiled_vs(struct v3d_context
*v3d
, uint8_t prim_mode
)
606 struct v3d_vs_key local_key
;
607 struct v3d_vs_key
*key
= &local_key
;
609 if (!(v3d
->dirty
& (VC5_DIRTY_PRIM_MODE
|
610 VC5_DIRTY_RASTERIZER
|
613 VC5_DIRTY_UNCOMPILED_VS
|
614 VC5_DIRTY_FS_INPUTS
))) {
618 memset(key
, 0, sizeof(*key
));
619 v3d_setup_shared_key(v3d
, &key
->base
, &v3d
->tex
[PIPE_SHADER_VERTEX
]);
620 key
->base
.shader_state
= v3d
->prog
.bind_vs
;
621 key
->num_fs_inputs
= v3d
->prog
.fs
->prog_data
.fs
->base
.num_inputs
;
622 STATIC_ASSERT(sizeof(key
->fs_inputs
) ==
623 sizeof(v3d
->prog
.fs
->prog_data
.fs
->input_slots
));
624 memcpy(key
->fs_inputs
, v3d
->prog
.fs
->prog_data
.fs
->input_slots
,
625 sizeof(key
->fs_inputs
));
626 key
->clamp_color
= v3d
->rasterizer
->base
.clamp_vertex_color
;
628 key
->per_vertex_point_size
=
629 (prim_mode
== PIPE_PRIM_POINTS
&&
630 v3d
->rasterizer
->base
.point_size_per_vertex
);
632 struct v3d_compiled_shader
*vs
=
633 v3d_get_compiled_shader(v3d
, &key
->base
);
634 if (vs
!= v3d
->prog
.vs
) {
636 v3d
->dirty
|= VC5_DIRTY_COMPILED_VS
;
639 key
->is_coord
= true;
640 /* Coord shaders only output varyings used by transform feedback. */
641 struct v3d_uncompiled_shader
*shader_state
= key
->base
.shader_state
;
642 memcpy(key
->fs_inputs
, shader_state
->tf_outputs
,
643 sizeof(*key
->fs_inputs
) * shader_state
->num_tf_outputs
);
644 if (shader_state
->num_tf_outputs
< key
->num_fs_inputs
) {
645 memset(&key
->fs_inputs
[shader_state
->num_tf_outputs
],
647 sizeof(*key
->fs_inputs
) * (key
->num_fs_inputs
-
648 shader_state
->num_tf_outputs
));
650 key
->num_fs_inputs
= shader_state
->num_tf_outputs
;
652 struct v3d_compiled_shader
*cs
=
653 v3d_get_compiled_shader(v3d
, &key
->base
);
654 if (cs
!= v3d
->prog
.cs
) {
656 v3d
->dirty
|= VC5_DIRTY_COMPILED_CS
;
661 v3d_update_compiled_shaders(struct v3d_context
*v3d
, uint8_t prim_mode
)
663 v3d_update_compiled_fs(v3d
, prim_mode
);
664 v3d_update_compiled_vs(v3d
, prim_mode
);
668 fs_cache_hash(const void *key
)
670 return _mesa_hash_data(key
, sizeof(struct v3d_fs_key
));
674 vs_cache_hash(const void *key
)
676 return _mesa_hash_data(key
, sizeof(struct v3d_vs_key
));
680 fs_cache_compare(const void *key1
, const void *key2
)
682 return memcmp(key1
, key2
, sizeof(struct v3d_fs_key
)) == 0;
686 vs_cache_compare(const void *key1
, const void *key2
)
688 return memcmp(key1
, key2
, sizeof(struct v3d_vs_key
)) == 0;
692 delete_from_cache_if_matches(struct hash_table
*ht
,
693 struct v3d_compiled_shader
**last_compile
,
694 struct hash_entry
*entry
,
695 struct v3d_uncompiled_shader
*so
)
697 const struct v3d_key
*key
= entry
->key
;
699 if (key
->shader_state
== so
) {
700 struct v3d_compiled_shader
*shader
= entry
->data
;
701 _mesa_hash_table_remove(ht
, entry
);
703 if (shader
== *last_compile
)
704 *last_compile
= NULL
;
706 v3d_free_compiled_shader(shader
);
711 v3d_shader_state_delete(struct pipe_context
*pctx
, void *hwcso
)
713 struct v3d_context
*v3d
= v3d_context(pctx
);
714 struct v3d_uncompiled_shader
*so
= hwcso
;
716 hash_table_foreach(v3d
->fs_cache
, entry
) {
717 delete_from_cache_if_matches(v3d
->fs_cache
, &v3d
->prog
.fs
,
720 hash_table_foreach(v3d
->vs_cache
, entry
) {
721 delete_from_cache_if_matches(v3d
->vs_cache
, &v3d
->prog
.vs
,
725 ralloc_free(so
->base
.ir
.nir
);
730 v3d_fp_state_bind(struct pipe_context
*pctx
, void *hwcso
)
732 struct v3d_context
*v3d
= v3d_context(pctx
);
733 v3d
->prog
.bind_fs
= hwcso
;
734 v3d
->dirty
|= VC5_DIRTY_UNCOMPILED_FS
;
738 v3d_vp_state_bind(struct pipe_context
*pctx
, void *hwcso
)
740 struct v3d_context
*v3d
= v3d_context(pctx
);
741 v3d
->prog
.bind_vs
= hwcso
;
742 v3d
->dirty
|= VC5_DIRTY_UNCOMPILED_VS
;
746 v3d_program_init(struct pipe_context
*pctx
)
748 struct v3d_context
*v3d
= v3d_context(pctx
);
750 pctx
->create_vs_state
= v3d_shader_state_create
;
751 pctx
->delete_vs_state
= v3d_shader_state_delete
;
753 pctx
->create_fs_state
= v3d_shader_state_create
;
754 pctx
->delete_fs_state
= v3d_shader_state_delete
;
756 pctx
->bind_fs_state
= v3d_fp_state_bind
;
757 pctx
->bind_vs_state
= v3d_vp_state_bind
;
759 v3d
->fs_cache
= _mesa_hash_table_create(pctx
, fs_cache_hash
,
761 v3d
->vs_cache
= _mesa_hash_table_create(pctx
, vs_cache_hash
,
766 v3d_program_fini(struct pipe_context
*pctx
)
768 struct v3d_context
*v3d
= v3d_context(pctx
);
770 hash_table_foreach(v3d
->fs_cache
, entry
) {
771 struct v3d_compiled_shader
*shader
= entry
->data
;
772 v3d_free_compiled_shader(shader
);
773 _mesa_hash_table_remove(v3d
->fs_cache
, entry
);
776 hash_table_foreach(v3d
->vs_cache
, entry
) {
777 struct v3d_compiled_shader
*shader
= entry
->data
;
778 v3d_free_compiled_shader(shader
);
779 _mesa_hash_table_remove(v3d
->vs_cache
, entry
);
782 v3d_bo_unreference(&v3d
->prog
.spill_bo
);