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"
40 static struct v3d_compiled_shader
*
41 v3d_get_compiled_shader(struct v3d_context
*v3d
,
42 struct v3d_key
*key
, size_t key_size
);
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
, bool bindless
)
175 return glsl_count_attribute_slots(type
, false);
179 * Precompiles a shader variant at shader state creation time if
180 * V3D_DEBUG=precompile is set. Used for shader-db
181 * (https://gitlab.freedesktop.org/mesa/shader-db)
184 v3d_shader_precompile(struct v3d_context
*v3d
,
185 struct v3d_uncompiled_shader
*so
)
187 nir_shader
*s
= so
->base
.ir
.nir
;
189 if (s
->info
.stage
== MESA_SHADER_FRAGMENT
) {
190 struct v3d_fs_key key
= {
191 .base
.shader_state
= so
,
194 nir_foreach_variable(var
, &s
->outputs
) {
195 if (var
->data
.location
== FRAG_RESULT_COLOR
) {
197 } else if (var
->data
.location
>= FRAG_RESULT_DATA0
) {
198 key
.cbufs
|= 1 << (var
->data
.location
-
203 v3d_setup_shared_precompile_key(so
, &key
.base
);
204 v3d_get_compiled_shader(v3d
, &key
.base
, sizeof(key
));
206 struct v3d_vs_key key
= {
207 .base
.shader_state
= so
,
210 v3d_setup_shared_precompile_key(so
, &key
.base
);
212 /* Compile VS: All outputs */
213 nir_foreach_variable(var
, &s
->outputs
) {
214 unsigned array_len
= MAX2(glsl_get_length(var
->type
), 1);
215 assert(array_len
== 1);
218 int slot
= var
->data
.location
;
219 for (int i
= 0; i
< glsl_get_components(var
->type
); i
++) {
220 int swiz
= var
->data
.location_frac
+ i
;
221 key
.fs_inputs
[key
.num_fs_inputs
++] =
222 v3d_slot_from_slot_and_component(slot
,
227 v3d_get_compiled_shader(v3d
, &key
.base
, sizeof(key
));
229 /* Compile VS bin shader: only position (XXX: include TF) */
231 key
.num_fs_inputs
= 0;
232 for (int i
= 0; i
< 4; i
++) {
233 key
.fs_inputs
[key
.num_fs_inputs
++] =
234 v3d_slot_from_slot_and_component(VARYING_SLOT_POS
,
237 v3d_get_compiled_shader(v3d
, &key
.base
, sizeof(key
));
242 v3d_uncompiled_shader_create(struct pipe_context
*pctx
,
243 enum pipe_shader_ir type
, void *ir
)
245 struct v3d_context
*v3d
= v3d_context(pctx
);
246 struct v3d_uncompiled_shader
*so
= CALLOC_STRUCT(v3d_uncompiled_shader
);
250 so
->program_id
= v3d
->next_uncompiled_program_id
++;
254 if (type
== PIPE_SHADER_IR_NIR
) {
255 /* The backend takes ownership of the NIR shader on state
260 assert(type
== PIPE_SHADER_IR_TGSI
);
262 if (V3D_DEBUG
& V3D_DEBUG_TGSI
) {
263 fprintf(stderr
, "prog %d TGSI:\n",
266 fprintf(stderr
, "\n");
268 s
= tgsi_to_nir(ir
, pctx
->screen
);
271 nir_variable_mode lower_mode
= nir_var_all
& ~nir_var_uniform
;
272 if (s
->info
.stage
== MESA_SHADER_VERTEX
)
273 lower_mode
&= ~(nir_var_shader_in
| nir_var_shader_out
);
274 NIR_PASS_V(s
, nir_lower_io
, lower_mode
,
276 (nir_lower_io_options
)0);
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_function_temp
);
287 /* Garbage collect dead instructions */
290 so
->base
.type
= PIPE_SHADER_IR_NIR
;
293 if (V3D_DEBUG
& (V3D_DEBUG_NIR
|
294 v3d_debug_flag_for_shader_stage(s
->info
.stage
))) {
295 fprintf(stderr
, "%s prog %d NIR:\n",
296 gl_shader_stage_name(s
->info
.stage
),
298 nir_print_shader(s
, stderr
);
299 fprintf(stderr
, "\n");
302 if (V3D_DEBUG
& V3D_DEBUG_PRECOMPILE
)
303 v3d_shader_precompile(v3d
, so
);
309 v3d_shader_debug_output(const char *message
, void *data
)
311 struct v3d_context
*v3d
= data
;
313 pipe_debug_message(&v3d
->debug
, SHADER_INFO
, "%s", message
);
317 v3d_shader_state_create(struct pipe_context
*pctx
,
318 const struct pipe_shader_state
*cso
)
320 struct v3d_uncompiled_shader
*so
=
321 v3d_uncompiled_shader_create(pctx
,
323 (cso
->type
== PIPE_SHADER_IR_TGSI
?
324 (void *)cso
->tokens
:
327 v3d_set_transform_feedback_outputs(so
, &cso
->stream_output
);
332 struct v3d_compiled_shader
*
333 v3d_get_compiled_shader(struct v3d_context
*v3d
,
337 struct v3d_uncompiled_shader
*shader_state
= key
->shader_state
;
338 nir_shader
*s
= shader_state
->base
.ir
.nir
;
340 struct hash_table
*ht
= v3d
->prog
.cache
[s
->info
.stage
];
341 struct hash_entry
*entry
= _mesa_hash_table_search(ht
, key
);
345 struct v3d_compiled_shader
*shader
=
346 rzalloc(NULL
, struct v3d_compiled_shader
);
348 int program_id
= shader_state
->program_id
;
350 p_atomic_inc_return(&shader_state
->compiled_variant_count
);
352 uint32_t shader_size
;
354 qpu_insts
= v3d_compile(v3d
->screen
->compiler
, key
,
355 &shader
->prog_data
.base
, s
,
356 v3d_shader_debug_output
,
358 program_id
, variant_id
, &shader_size
);
359 ralloc_steal(shader
, shader
->prog_data
.base
);
361 v3d_set_shader_uniform_dirty_flags(shader
);
364 u_upload_data(v3d
->state_uploader
, 0, shader_size
, 8,
365 qpu_insts
, &shader
->offset
, &shader
->resource
);
371 struct v3d_key
*dup_key
;
372 dup_key
= ralloc_size(shader
, key_size
);
373 memcpy(dup_key
, key
, key_size
);
374 _mesa_hash_table_insert(ht
, dup_key
, shader
);
377 if (shader
->prog_data
.base
->spill_size
>
378 v3d
->prog
.spill_size_per_thread
) {
379 /* The TIDX register we use for choosing the area to access
380 * for scratch space is: (core << 6) | (qpu << 2) | thread.
381 * Even at minimum threadcount in a particular shader, that
382 * means we still multiply by qpus by 4.
384 int total_spill_size
= (v3d
->screen
->devinfo
.qpu_count
* 4 *
385 shader
->prog_data
.base
->spill_size
);
387 v3d_bo_unreference(&v3d
->prog
.spill_bo
);
388 v3d
->prog
.spill_bo
= v3d_bo_alloc(v3d
->screen
,
389 total_spill_size
, "spill");
390 v3d
->prog
.spill_size_per_thread
=
391 shader
->prog_data
.base
->spill_size
;
398 v3d_free_compiled_shader(struct v3d_compiled_shader
*shader
)
400 pipe_resource_reference(&shader
->resource
, NULL
);
405 v3d_setup_shared_key(struct v3d_context
*v3d
, struct v3d_key
*key
,
406 struct v3d_texture_stateobj
*texstate
)
408 const struct v3d_device_info
*devinfo
= &v3d
->screen
->devinfo
;
410 for (int i
= 0; i
< texstate
->num_textures
; i
++) {
411 struct pipe_sampler_view
*sampler
= texstate
->textures
[i
];
412 struct v3d_sampler_view
*v3d_sampler
= v3d_sampler_view(sampler
);
413 struct pipe_sampler_state
*sampler_state
=
414 texstate
->samplers
[i
];
419 key
->tex
[i
].return_size
=
420 v3d_get_tex_return_size(devinfo
,
422 sampler_state
->compare_mode
);
424 /* For 16-bit, we set up the sampler to always return 2
425 * channels (meaning no recompiles for most statechanges),
426 * while for 32 we actually scale the returns with channels.
428 if (key
->tex
[i
].return_size
== 16) {
429 key
->tex
[i
].return_channels
= 2;
430 } else if (devinfo
->ver
> 40) {
431 key
->tex
[i
].return_channels
= 4;
433 key
->tex
[i
].return_channels
=
434 v3d_get_tex_return_channels(devinfo
,
438 if (key
->tex
[i
].return_size
== 32 && devinfo
->ver
< 40) {
439 memcpy(key
->tex
[i
].swizzle
,
440 v3d_sampler
->swizzle
,
441 sizeof(v3d_sampler
->swizzle
));
443 /* For 16-bit returns, we let the sampler state handle
446 key
->tex
[i
].swizzle
[0] = PIPE_SWIZZLE_X
;
447 key
->tex
[i
].swizzle
[1] = PIPE_SWIZZLE_Y
;
448 key
->tex
[i
].swizzle
[2] = PIPE_SWIZZLE_Z
;
449 key
->tex
[i
].swizzle
[3] = PIPE_SWIZZLE_W
;
453 key
->tex
[i
].clamp_s
=
454 sampler_state
->wrap_s
== PIPE_TEX_WRAP_CLAMP
;
455 key
->tex
[i
].clamp_t
=
456 sampler_state
->wrap_t
== PIPE_TEX_WRAP_CLAMP
;
457 key
->tex
[i
].clamp_r
=
458 sampler_state
->wrap_r
== PIPE_TEX_WRAP_CLAMP
;
464 v3d_setup_shared_precompile_key(struct v3d_uncompiled_shader
*uncompiled
,
467 nir_shader
*s
= uncompiled
->base
.ir
.nir
;
469 for (int i
= 0; i
< s
->info
.num_textures
; i
++) {
470 key
->tex
[i
].return_size
= 16;
471 key
->tex
[i
].return_channels
= 2;
473 key
->tex
[i
].swizzle
[0] = PIPE_SWIZZLE_X
;
474 key
->tex
[i
].swizzle
[1] = PIPE_SWIZZLE_Y
;
475 key
->tex
[i
].swizzle
[2] = PIPE_SWIZZLE_Z
;
476 key
->tex
[i
].swizzle
[3] = PIPE_SWIZZLE_W
;
481 v3d_update_compiled_fs(struct v3d_context
*v3d
, uint8_t prim_mode
)
483 struct v3d_job
*job
= v3d
->job
;
484 struct v3d_fs_key local_key
;
485 struct v3d_fs_key
*key
= &local_key
;
486 nir_shader
*s
= v3d
->prog
.bind_fs
->base
.ir
.nir
;
488 if (!(v3d
->dirty
& (VC5_DIRTY_PRIM_MODE
|
490 VC5_DIRTY_FRAMEBUFFER
|
492 VC5_DIRTY_RASTERIZER
|
493 VC5_DIRTY_SAMPLE_STATE
|
495 VC5_DIRTY_UNCOMPILED_FS
))) {
499 memset(key
, 0, sizeof(*key
));
500 v3d_setup_shared_key(v3d
, &key
->base
, &v3d
->tex
[PIPE_SHADER_FRAGMENT
]);
501 key
->base
.shader_state
= v3d
->prog
.bind_fs
;
502 key
->base
.ucp_enables
= v3d
->rasterizer
->base
.clip_plane_enable
;
503 key
->is_points
= (prim_mode
== PIPE_PRIM_POINTS
);
504 key
->is_lines
= (prim_mode
>= PIPE_PRIM_LINES
&&
505 prim_mode
<= PIPE_PRIM_LINE_STRIP
);
506 key
->clamp_color
= v3d
->rasterizer
->base
.clamp_fragment_color
;
507 if (v3d
->blend
->base
.logicop_enable
) {
508 key
->logicop_func
= v3d
->blend
->base
.logicop_func
;
510 key
->logicop_func
= PIPE_LOGICOP_COPY
;
513 key
->msaa
= v3d
->rasterizer
->base
.multisample
;
514 key
->sample_coverage
= (v3d
->rasterizer
->base
.multisample
&&
515 v3d
->sample_mask
!= (1 << V3D_MAX_SAMPLES
) - 1);
516 key
->sample_alpha_to_coverage
= v3d
->blend
->base
.alpha_to_coverage
;
517 key
->sample_alpha_to_one
= v3d
->blend
->base
.alpha_to_one
;
520 key
->depth_enabled
= (v3d
->zsa
->base
.depth
.enabled
||
521 v3d
->zsa
->base
.stencil
[0].enabled
);
522 if (v3d
->zsa
->base
.alpha
.enabled
) {
523 key
->alpha_test
= true;
524 key
->alpha_test_func
= v3d
->zsa
->base
.alpha
.func
;
527 key
->swap_color_rb
= v3d
->swap_color_rb
;
529 for (int i
= 0; i
< v3d
->framebuffer
.nr_cbufs
; i
++) {
530 struct pipe_surface
*cbuf
= v3d
->framebuffer
.cbufs
[i
];
534 /* gl_FragColor's propagation to however many bound color
535 * buffers there are means that the shader compile needs to
536 * know what buffers are present.
538 key
->cbufs
|= 1 << i
;
540 const struct util_format_description
*desc
=
541 util_format_description(cbuf
->format
);
543 if (desc
->channel
[0].type
== UTIL_FORMAT_TYPE_FLOAT
&&
544 desc
->channel
[0].size
== 32) {
545 key
->f32_color_rb
|= 1 << i
;
548 if (s
->info
.fs
.untyped_color_outputs
) {
549 if (util_format_is_pure_uint(cbuf
->format
))
550 key
->uint_color_rb
|= 1 << i
;
551 else if (util_format_is_pure_sint(cbuf
->format
))
552 key
->int_color_rb
|= 1 << i
;
556 if (key
->is_points
) {
557 key
->point_sprite_mask
=
558 v3d
->rasterizer
->base
.sprite_coord_enable
;
559 key
->point_coord_upper_left
=
560 (v3d
->rasterizer
->base
.sprite_coord_mode
==
561 PIPE_SPRITE_COORD_UPPER_LEFT
);
564 key
->light_twoside
= v3d
->rasterizer
->base
.light_twoside
;
565 key
->shade_model_flat
= v3d
->rasterizer
->base
.flatshade
;
567 struct v3d_compiled_shader
*old_fs
= v3d
->prog
.fs
;
568 v3d
->prog
.fs
= v3d_get_compiled_shader(v3d
, &key
->base
, sizeof(*key
));
569 if (v3d
->prog
.fs
== old_fs
)
572 v3d
->dirty
|= VC5_DIRTY_COMPILED_FS
;
575 if (v3d
->prog
.fs
->prog_data
.fs
->flat_shade_flags
!=
576 old_fs
->prog_data
.fs
->flat_shade_flags
) {
577 v3d
->dirty
|= VC5_DIRTY_FLAT_SHADE_FLAGS
;
580 if (v3d
->prog
.fs
->prog_data
.fs
->noperspective_flags
!=
581 old_fs
->prog_data
.fs
->noperspective_flags
) {
582 v3d
->dirty
|= VC5_DIRTY_NOPERSPECTIVE_FLAGS
;
585 if (v3d
->prog
.fs
->prog_data
.fs
->centroid_flags
!=
586 old_fs
->prog_data
.fs
->centroid_flags
) {
587 v3d
->dirty
|= VC5_DIRTY_CENTROID_FLAGS
;
591 if (old_fs
&& memcmp(v3d
->prog
.fs
->prog_data
.fs
->input_slots
,
592 old_fs
->prog_data
.fs
->input_slots
,
593 sizeof(v3d
->prog
.fs
->prog_data
.fs
->input_slots
))) {
594 v3d
->dirty
|= VC5_DIRTY_FS_INPUTS
;
599 v3d_update_compiled_vs(struct v3d_context
*v3d
, uint8_t prim_mode
)
601 struct v3d_vs_key local_key
;
602 struct v3d_vs_key
*key
= &local_key
;
604 if (!(v3d
->dirty
& (VC5_DIRTY_PRIM_MODE
|
605 VC5_DIRTY_RASTERIZER
|
608 VC5_DIRTY_UNCOMPILED_VS
|
609 VC5_DIRTY_FS_INPUTS
))) {
613 memset(key
, 0, sizeof(*key
));
614 v3d_setup_shared_key(v3d
, &key
->base
, &v3d
->tex
[PIPE_SHADER_VERTEX
]);
615 key
->base
.shader_state
= v3d
->prog
.bind_vs
;
616 key
->base
.ucp_enables
= v3d
->rasterizer
->base
.clip_plane_enable
;
617 key
->num_fs_inputs
= v3d
->prog
.fs
->prog_data
.fs
->num_inputs
;
618 STATIC_ASSERT(sizeof(key
->fs_inputs
) ==
619 sizeof(v3d
->prog
.fs
->prog_data
.fs
->input_slots
));
620 memcpy(key
->fs_inputs
, v3d
->prog
.fs
->prog_data
.fs
->input_slots
,
621 sizeof(key
->fs_inputs
));
622 key
->clamp_color
= v3d
->rasterizer
->base
.clamp_vertex_color
;
624 key
->per_vertex_point_size
=
625 (prim_mode
== PIPE_PRIM_POINTS
&&
626 v3d
->rasterizer
->base
.point_size_per_vertex
);
628 struct v3d_compiled_shader
*vs
=
629 v3d_get_compiled_shader(v3d
, &key
->base
, sizeof(*key
));
630 if (vs
!= v3d
->prog
.vs
) {
632 v3d
->dirty
|= VC5_DIRTY_COMPILED_VS
;
635 key
->is_coord
= true;
636 /* Coord shaders only output varyings used by transform feedback. */
637 struct v3d_uncompiled_shader
*shader_state
= key
->base
.shader_state
;
638 memcpy(key
->fs_inputs
, shader_state
->tf_outputs
,
639 sizeof(*key
->fs_inputs
) * shader_state
->num_tf_outputs
);
640 if (shader_state
->num_tf_outputs
< key
->num_fs_inputs
) {
641 memset(&key
->fs_inputs
[shader_state
->num_tf_outputs
],
643 sizeof(*key
->fs_inputs
) * (key
->num_fs_inputs
-
644 shader_state
->num_tf_outputs
));
646 key
->num_fs_inputs
= shader_state
->num_tf_outputs
;
648 struct v3d_compiled_shader
*cs
=
649 v3d_get_compiled_shader(v3d
, &key
->base
, sizeof(*key
));
650 if (cs
!= v3d
->prog
.cs
) {
652 v3d
->dirty
|= VC5_DIRTY_COMPILED_CS
;
657 v3d_update_compiled_shaders(struct v3d_context
*v3d
, uint8_t prim_mode
)
659 v3d_update_compiled_fs(v3d
, prim_mode
);
660 v3d_update_compiled_vs(v3d
, prim_mode
);
664 v3d_update_compiled_cs(struct v3d_context
*v3d
)
666 struct v3d_key local_key
;
667 struct v3d_key
*key
= &local_key
;
669 if (!(v3d
->dirty
& (~0 | /* XXX */
671 VC5_DIRTY_UNCOMPILED_FS
))) {
675 memset(key
, 0, sizeof(*key
));
676 v3d_setup_shared_key(v3d
, key
, &v3d
->tex
[PIPE_SHADER_COMPUTE
]);
677 key
->shader_state
= v3d
->prog
.bind_compute
;
679 struct v3d_compiled_shader
*cs
=
680 v3d_get_compiled_shader(v3d
, key
, sizeof(*key
));
681 if (cs
!= v3d
->prog
.compute
) {
682 v3d
->prog
.compute
= cs
;
683 v3d
->dirty
|= VC5_DIRTY_COMPILED_CS
; /* XXX */
688 fs_cache_hash(const void *key
)
690 return _mesa_hash_data(key
, sizeof(struct v3d_fs_key
));
694 vs_cache_hash(const void *key
)
696 return _mesa_hash_data(key
, sizeof(struct v3d_vs_key
));
700 cs_cache_hash(const void *key
)
702 return _mesa_hash_data(key
, sizeof(struct v3d_key
));
706 fs_cache_compare(const void *key1
, const void *key2
)
708 return memcmp(key1
, key2
, sizeof(struct v3d_fs_key
)) == 0;
712 vs_cache_compare(const void *key1
, const void *key2
)
714 return memcmp(key1
, key2
, sizeof(struct v3d_vs_key
)) == 0;
718 cs_cache_compare(const void *key1
, const void *key2
)
720 return memcmp(key1
, key2
, sizeof(struct v3d_key
)) == 0;
724 v3d_shader_state_delete(struct pipe_context
*pctx
, void *hwcso
)
726 struct v3d_context
*v3d
= v3d_context(pctx
);
727 struct v3d_uncompiled_shader
*so
= hwcso
;
728 nir_shader
*s
= so
->base
.ir
.nir
;
730 hash_table_foreach(v3d
->prog
.cache
[s
->info
.stage
], entry
) {
731 const struct v3d_key
*key
= entry
->key
;
732 struct v3d_compiled_shader
*shader
= entry
->data
;
734 if (key
->shader_state
!= so
)
737 if (v3d
->prog
.fs
== shader
)
739 if (v3d
->prog
.vs
== shader
)
741 if (v3d
->prog
.cs
== shader
)
743 if (v3d
->prog
.compute
== shader
)
744 v3d
->prog
.compute
= NULL
;
746 _mesa_hash_table_remove(v3d
->prog
.cache
[s
->info
.stage
], entry
);
747 v3d_free_compiled_shader(shader
);
750 ralloc_free(so
->base
.ir
.nir
);
755 v3d_fp_state_bind(struct pipe_context
*pctx
, void *hwcso
)
757 struct v3d_context
*v3d
= v3d_context(pctx
);
758 v3d
->prog
.bind_fs
= hwcso
;
759 v3d
->dirty
|= VC5_DIRTY_UNCOMPILED_FS
;
763 v3d_vp_state_bind(struct pipe_context
*pctx
, void *hwcso
)
765 struct v3d_context
*v3d
= v3d_context(pctx
);
766 v3d
->prog
.bind_vs
= hwcso
;
767 v3d
->dirty
|= VC5_DIRTY_UNCOMPILED_VS
;
771 v3d_compute_state_bind(struct pipe_context
*pctx
, void *state
)
773 struct v3d_context
*v3d
= v3d_context(pctx
);
775 v3d
->prog
.bind_compute
= state
;
779 v3d_create_compute_state(struct pipe_context
*pctx
,
780 const struct pipe_compute_state
*cso
)
782 return v3d_uncompiled_shader_create(pctx
, cso
->ir_type
,
787 v3d_program_init(struct pipe_context
*pctx
)
789 struct v3d_context
*v3d
= v3d_context(pctx
);
791 pctx
->create_vs_state
= v3d_shader_state_create
;
792 pctx
->delete_vs_state
= v3d_shader_state_delete
;
794 pctx
->create_fs_state
= v3d_shader_state_create
;
795 pctx
->delete_fs_state
= v3d_shader_state_delete
;
797 pctx
->bind_fs_state
= v3d_fp_state_bind
;
798 pctx
->bind_vs_state
= v3d_vp_state_bind
;
800 if (v3d
->screen
->has_csd
) {
801 pctx
->create_compute_state
= v3d_create_compute_state
;
802 pctx
->delete_compute_state
= v3d_shader_state_delete
;
803 pctx
->bind_compute_state
= v3d_compute_state_bind
;
806 v3d
->prog
.cache
[MESA_SHADER_VERTEX
] =
807 _mesa_hash_table_create(pctx
, vs_cache_hash
, vs_cache_compare
);
808 v3d
->prog
.cache
[MESA_SHADER_FRAGMENT
] =
809 _mesa_hash_table_create(pctx
, fs_cache_hash
, fs_cache_compare
);
810 v3d
->prog
.cache
[MESA_SHADER_COMPUTE
] =
811 _mesa_hash_table_create(pctx
, cs_cache_hash
, cs_cache_compare
);
815 v3d_program_fini(struct pipe_context
*pctx
)
817 struct v3d_context
*v3d
= v3d_context(pctx
);
819 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
820 struct hash_table
*cache
= v3d
->prog
.cache
[i
];
824 hash_table_foreach(cache
, entry
) {
825 struct v3d_compiled_shader
*shader
= entry
->data
;
826 v3d_free_compiled_shader(shader
);
827 _mesa_hash_table_remove(cache
, entry
);
831 v3d_bo_unreference(&v3d
->prog
.spill_bo
);