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 key
.logicop_func
= PIPE_LOGICOP_COPY
;
205 v3d_setup_shared_precompile_key(so
, &key
.base
);
206 v3d_get_compiled_shader(v3d
, &key
.base
, sizeof(key
));
208 struct v3d_vs_key key
= {
209 .base
.shader_state
= so
,
212 v3d_setup_shared_precompile_key(so
, &key
.base
);
214 /* Compile VS: All outputs */
215 nir_foreach_variable(var
, &s
->outputs
) {
216 unsigned array_len
= MAX2(glsl_get_length(var
->type
), 1);
217 assert(array_len
== 1);
220 int slot
= var
->data
.location
;
221 for (int i
= 0; i
< glsl_get_components(var
->type
); i
++) {
222 int swiz
= var
->data
.location_frac
+ i
;
223 key
.fs_inputs
[key
.num_fs_inputs
++] =
224 v3d_slot_from_slot_and_component(slot
,
229 v3d_get_compiled_shader(v3d
, &key
.base
, sizeof(key
));
231 /* Compile VS bin shader: only position (XXX: include TF) */
233 key
.num_fs_inputs
= 0;
234 for (int i
= 0; i
< 4; i
++) {
235 key
.fs_inputs
[key
.num_fs_inputs
++] =
236 v3d_slot_from_slot_and_component(VARYING_SLOT_POS
,
239 v3d_get_compiled_shader(v3d
, &key
.base
, sizeof(key
));
244 v3d_uncompiled_shader_create(struct pipe_context
*pctx
,
245 enum pipe_shader_ir type
, void *ir
)
247 struct v3d_context
*v3d
= v3d_context(pctx
);
248 struct v3d_uncompiled_shader
*so
= CALLOC_STRUCT(v3d_uncompiled_shader
);
252 so
->program_id
= v3d
->next_uncompiled_program_id
++;
256 if (type
== PIPE_SHADER_IR_NIR
) {
257 /* The backend takes ownership of the NIR shader on state
262 assert(type
== PIPE_SHADER_IR_TGSI
);
264 if (V3D_DEBUG
& V3D_DEBUG_TGSI
) {
265 fprintf(stderr
, "prog %d TGSI:\n",
268 fprintf(stderr
, "\n");
270 s
= tgsi_to_nir(ir
, pctx
->screen
);
273 nir_variable_mode lower_mode
= nir_var_all
& ~nir_var_uniform
;
274 if (s
->info
.stage
== MESA_SHADER_VERTEX
)
275 lower_mode
&= ~(nir_var_shader_in
| nir_var_shader_out
);
276 NIR_PASS_V(s
, nir_lower_io
, lower_mode
,
278 (nir_lower_io_options
)0);
280 NIR_PASS_V(s
, nir_lower_regs_to_ssa
);
281 NIR_PASS_V(s
, nir_normalize_cubemap_coords
);
283 NIR_PASS_V(s
, nir_lower_load_const_to_scalar
);
287 NIR_PASS_V(s
, nir_remove_dead_variables
, nir_var_function_temp
);
289 /* Garbage collect dead instructions */
292 so
->base
.type
= PIPE_SHADER_IR_NIR
;
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
);
319 v3d_shader_state_create(struct pipe_context
*pctx
,
320 const struct pipe_shader_state
*cso
)
322 struct v3d_uncompiled_shader
*so
=
323 v3d_uncompiled_shader_create(pctx
,
325 (cso
->type
== PIPE_SHADER_IR_TGSI
?
326 (void *)cso
->tokens
:
329 v3d_set_transform_feedback_outputs(so
, &cso
->stream_output
);
334 struct v3d_compiled_shader
*
335 v3d_get_compiled_shader(struct v3d_context
*v3d
,
339 struct v3d_uncompiled_shader
*shader_state
= key
->shader_state
;
340 nir_shader
*s
= shader_state
->base
.ir
.nir
;
342 struct hash_table
*ht
= v3d
->prog
.cache
[s
->info
.stage
];
343 struct hash_entry
*entry
= _mesa_hash_table_search(ht
, key
);
347 struct v3d_compiled_shader
*shader
=
348 rzalloc(NULL
, struct v3d_compiled_shader
);
350 int program_id
= shader_state
->program_id
;
352 p_atomic_inc_return(&shader_state
->compiled_variant_count
);
354 uint32_t shader_size
;
356 qpu_insts
= v3d_compile(v3d
->screen
->compiler
, key
,
357 &shader
->prog_data
.base
, s
,
358 v3d_shader_debug_output
,
360 program_id
, variant_id
, &shader_size
);
361 ralloc_steal(shader
, shader
->prog_data
.base
);
363 v3d_set_shader_uniform_dirty_flags(shader
);
366 u_upload_data(v3d
->state_uploader
, 0, shader_size
, 8,
367 qpu_insts
, &shader
->offset
, &shader
->resource
);
373 struct v3d_key
*dup_key
;
374 dup_key
= ralloc_size(shader
, key_size
);
375 memcpy(dup_key
, key
, key_size
);
376 _mesa_hash_table_insert(ht
, dup_key
, shader
);
379 if (shader
->prog_data
.base
->spill_size
>
380 v3d
->prog
.spill_size_per_thread
) {
381 /* The TIDX register we use for choosing the area to access
382 * for scratch space is: (core << 6) | (qpu << 2) | thread.
383 * Even at minimum threadcount in a particular shader, that
384 * means we still multiply by qpus by 4.
386 int total_spill_size
= (v3d
->screen
->devinfo
.qpu_count
* 4 *
387 shader
->prog_data
.base
->spill_size
);
389 v3d_bo_unreference(&v3d
->prog
.spill_bo
);
390 v3d
->prog
.spill_bo
= v3d_bo_alloc(v3d
->screen
,
391 total_spill_size
, "spill");
392 v3d
->prog
.spill_size_per_thread
=
393 shader
->prog_data
.base
->spill_size
;
400 v3d_free_compiled_shader(struct v3d_compiled_shader
*shader
)
402 pipe_resource_reference(&shader
->resource
, NULL
);
407 v3d_setup_shared_key(struct v3d_context
*v3d
, struct v3d_key
*key
,
408 struct v3d_texture_stateobj
*texstate
)
410 const struct v3d_device_info
*devinfo
= &v3d
->screen
->devinfo
;
412 for (int i
= 0; i
< texstate
->num_textures
; i
++) {
413 struct pipe_sampler_view
*sampler
= texstate
->textures
[i
];
414 struct v3d_sampler_view
*v3d_sampler
= v3d_sampler_view(sampler
);
415 struct pipe_sampler_state
*sampler_state
=
416 texstate
->samplers
[i
];
421 key
->tex
[i
].return_size
=
422 v3d_get_tex_return_size(devinfo
,
424 sampler_state
->compare_mode
);
426 /* For 16-bit, we set up the sampler to always return 2
427 * channels (meaning no recompiles for most statechanges),
428 * while for 32 we actually scale the returns with channels.
430 if (key
->tex
[i
].return_size
== 16) {
431 key
->tex
[i
].return_channels
= 2;
432 } else if (devinfo
->ver
> 40) {
433 key
->tex
[i
].return_channels
= 4;
435 key
->tex
[i
].return_channels
=
436 v3d_get_tex_return_channels(devinfo
,
440 if (key
->tex
[i
].return_size
== 32 && devinfo
->ver
< 40) {
441 memcpy(key
->tex
[i
].swizzle
,
442 v3d_sampler
->swizzle
,
443 sizeof(v3d_sampler
->swizzle
));
445 /* For 16-bit returns, we let the sampler state handle
448 key
->tex
[i
].swizzle
[0] = PIPE_SWIZZLE_X
;
449 key
->tex
[i
].swizzle
[1] = PIPE_SWIZZLE_Y
;
450 key
->tex
[i
].swizzle
[2] = PIPE_SWIZZLE_Z
;
451 key
->tex
[i
].swizzle
[3] = PIPE_SWIZZLE_W
;
455 key
->tex
[i
].clamp_s
=
456 sampler_state
->wrap_s
== PIPE_TEX_WRAP_CLAMP
;
457 key
->tex
[i
].clamp_t
=
458 sampler_state
->wrap_t
== PIPE_TEX_WRAP_CLAMP
;
459 key
->tex
[i
].clamp_r
=
460 sampler_state
->wrap_r
== PIPE_TEX_WRAP_CLAMP
;
466 v3d_setup_shared_precompile_key(struct v3d_uncompiled_shader
*uncompiled
,
469 nir_shader
*s
= uncompiled
->base
.ir
.nir
;
471 for (int i
= 0; i
< s
->info
.num_textures
; i
++) {
472 key
->tex
[i
].return_size
= 16;
473 key
->tex
[i
].return_channels
= 2;
475 key
->tex
[i
].swizzle
[0] = PIPE_SWIZZLE_X
;
476 key
->tex
[i
].swizzle
[1] = PIPE_SWIZZLE_Y
;
477 key
->tex
[i
].swizzle
[2] = PIPE_SWIZZLE_Z
;
478 key
->tex
[i
].swizzle
[3] = PIPE_SWIZZLE_W
;
483 v3d_update_compiled_fs(struct v3d_context
*v3d
, uint8_t prim_mode
)
485 struct v3d_job
*job
= v3d
->job
;
486 struct v3d_fs_key local_key
;
487 struct v3d_fs_key
*key
= &local_key
;
488 nir_shader
*s
= v3d
->prog
.bind_fs
->base
.ir
.nir
;
490 if (!(v3d
->dirty
& (VC5_DIRTY_PRIM_MODE
|
492 VC5_DIRTY_FRAMEBUFFER
|
494 VC5_DIRTY_RASTERIZER
|
495 VC5_DIRTY_SAMPLE_STATE
|
497 VC5_DIRTY_UNCOMPILED_FS
))) {
501 memset(key
, 0, sizeof(*key
));
502 v3d_setup_shared_key(v3d
, &key
->base
, &v3d
->tex
[PIPE_SHADER_FRAGMENT
]);
503 key
->base
.shader_state
= v3d
->prog
.bind_fs
;
504 key
->base
.ucp_enables
= v3d
->rasterizer
->base
.clip_plane_enable
;
505 key
->is_points
= (prim_mode
== PIPE_PRIM_POINTS
);
506 key
->is_lines
= (prim_mode
>= PIPE_PRIM_LINES
&&
507 prim_mode
<= PIPE_PRIM_LINE_STRIP
);
508 key
->clamp_color
= v3d
->rasterizer
->base
.clamp_fragment_color
;
509 if (v3d
->blend
->base
.logicop_enable
) {
510 key
->logicop_func
= v3d
->blend
->base
.logicop_func
;
512 key
->logicop_func
= PIPE_LOGICOP_COPY
;
515 key
->msaa
= v3d
->rasterizer
->base
.multisample
;
516 key
->sample_coverage
= (v3d
->rasterizer
->base
.multisample
&&
517 v3d
->sample_mask
!= (1 << V3D_MAX_SAMPLES
) - 1);
518 key
->sample_alpha_to_coverage
= v3d
->blend
->base
.alpha_to_coverage
;
519 key
->sample_alpha_to_one
= v3d
->blend
->base
.alpha_to_one
;
522 key
->depth_enabled
= (v3d
->zsa
->base
.depth
.enabled
||
523 v3d
->zsa
->base
.stencil
[0].enabled
);
524 if (v3d
->zsa
->base
.alpha
.enabled
) {
525 key
->alpha_test
= true;
526 key
->alpha_test_func
= v3d
->zsa
->base
.alpha
.func
;
529 key
->swap_color_rb
= v3d
->swap_color_rb
;
531 for (int i
= 0; i
< v3d
->framebuffer
.nr_cbufs
; i
++) {
532 struct pipe_surface
*cbuf
= v3d
->framebuffer
.cbufs
[i
];
536 /* gl_FragColor's propagation to however many bound color
537 * buffers there are means that the shader compile needs to
538 * know what buffers are present.
540 key
->cbufs
|= 1 << i
;
542 /* If logic operations are enabled then we might emit color
543 * reads and we need to know the color buffer format and
546 if (key
->logicop_func
!= PIPE_LOGICOP_COPY
) {
547 key
->color_fmt
[i
].format
= cbuf
->format
;
548 key
->color_fmt
[i
].swizzle
=
549 v3d_get_format_swizzle(&v3d
->screen
->devinfo
,
553 const struct util_format_description
*desc
=
554 util_format_description(cbuf
->format
);
556 if (desc
->channel
[0].type
== UTIL_FORMAT_TYPE_FLOAT
&&
557 desc
->channel
[0].size
== 32) {
558 key
->f32_color_rb
|= 1 << i
;
561 if (s
->info
.fs
.untyped_color_outputs
) {
562 if (util_format_is_pure_uint(cbuf
->format
))
563 key
->uint_color_rb
|= 1 << i
;
564 else if (util_format_is_pure_sint(cbuf
->format
))
565 key
->int_color_rb
|= 1 << i
;
569 if (key
->is_points
) {
570 key
->point_sprite_mask
=
571 v3d
->rasterizer
->base
.sprite_coord_enable
;
572 key
->point_coord_upper_left
=
573 (v3d
->rasterizer
->base
.sprite_coord_mode
==
574 PIPE_SPRITE_COORD_UPPER_LEFT
);
577 key
->light_twoside
= v3d
->rasterizer
->base
.light_twoside
;
578 key
->shade_model_flat
= v3d
->rasterizer
->base
.flatshade
;
580 struct v3d_compiled_shader
*old_fs
= v3d
->prog
.fs
;
581 v3d
->prog
.fs
= v3d_get_compiled_shader(v3d
, &key
->base
, sizeof(*key
));
582 if (v3d
->prog
.fs
== old_fs
)
585 v3d
->dirty
|= VC5_DIRTY_COMPILED_FS
;
588 if (v3d
->prog
.fs
->prog_data
.fs
->flat_shade_flags
!=
589 old_fs
->prog_data
.fs
->flat_shade_flags
) {
590 v3d
->dirty
|= VC5_DIRTY_FLAT_SHADE_FLAGS
;
593 if (v3d
->prog
.fs
->prog_data
.fs
->noperspective_flags
!=
594 old_fs
->prog_data
.fs
->noperspective_flags
) {
595 v3d
->dirty
|= VC5_DIRTY_NOPERSPECTIVE_FLAGS
;
598 if (v3d
->prog
.fs
->prog_data
.fs
->centroid_flags
!=
599 old_fs
->prog_data
.fs
->centroid_flags
) {
600 v3d
->dirty
|= VC5_DIRTY_CENTROID_FLAGS
;
604 if (old_fs
&& memcmp(v3d
->prog
.fs
->prog_data
.fs
->input_slots
,
605 old_fs
->prog_data
.fs
->input_slots
,
606 sizeof(v3d
->prog
.fs
->prog_data
.fs
->input_slots
))) {
607 v3d
->dirty
|= VC5_DIRTY_FS_INPUTS
;
612 v3d_update_compiled_vs(struct v3d_context
*v3d
, uint8_t prim_mode
)
614 struct v3d_vs_key local_key
;
615 struct v3d_vs_key
*key
= &local_key
;
617 if (!(v3d
->dirty
& (VC5_DIRTY_PRIM_MODE
|
618 VC5_DIRTY_RASTERIZER
|
621 VC5_DIRTY_UNCOMPILED_VS
|
622 VC5_DIRTY_FS_INPUTS
))) {
626 memset(key
, 0, sizeof(*key
));
627 v3d_setup_shared_key(v3d
, &key
->base
, &v3d
->tex
[PIPE_SHADER_VERTEX
]);
628 key
->base
.shader_state
= v3d
->prog
.bind_vs
;
629 key
->base
.ucp_enables
= v3d
->rasterizer
->base
.clip_plane_enable
;
630 key
->num_fs_inputs
= v3d
->prog
.fs
->prog_data
.fs
->num_inputs
;
631 STATIC_ASSERT(sizeof(key
->fs_inputs
) ==
632 sizeof(v3d
->prog
.fs
->prog_data
.fs
->input_slots
));
633 memcpy(key
->fs_inputs
, v3d
->prog
.fs
->prog_data
.fs
->input_slots
,
634 sizeof(key
->fs_inputs
));
635 key
->clamp_color
= v3d
->rasterizer
->base
.clamp_vertex_color
;
637 key
->per_vertex_point_size
=
638 (prim_mode
== PIPE_PRIM_POINTS
&&
639 v3d
->rasterizer
->base
.point_size_per_vertex
);
641 struct v3d_compiled_shader
*vs
=
642 v3d_get_compiled_shader(v3d
, &key
->base
, sizeof(*key
));
643 if (vs
!= v3d
->prog
.vs
) {
645 v3d
->dirty
|= VC5_DIRTY_COMPILED_VS
;
648 key
->is_coord
= true;
649 /* Coord shaders only output varyings used by transform feedback. */
650 struct v3d_uncompiled_shader
*shader_state
= key
->base
.shader_state
;
651 memcpy(key
->fs_inputs
, shader_state
->tf_outputs
,
652 sizeof(*key
->fs_inputs
) * shader_state
->num_tf_outputs
);
653 if (shader_state
->num_tf_outputs
< key
->num_fs_inputs
) {
654 memset(&key
->fs_inputs
[shader_state
->num_tf_outputs
],
656 sizeof(*key
->fs_inputs
) * (key
->num_fs_inputs
-
657 shader_state
->num_tf_outputs
));
659 key
->num_fs_inputs
= shader_state
->num_tf_outputs
;
661 struct v3d_compiled_shader
*cs
=
662 v3d_get_compiled_shader(v3d
, &key
->base
, sizeof(*key
));
663 if (cs
!= v3d
->prog
.cs
) {
665 v3d
->dirty
|= VC5_DIRTY_COMPILED_CS
;
670 v3d_update_compiled_shaders(struct v3d_context
*v3d
, uint8_t prim_mode
)
672 v3d_update_compiled_fs(v3d
, prim_mode
);
673 v3d_update_compiled_vs(v3d
, prim_mode
);
677 v3d_update_compiled_cs(struct v3d_context
*v3d
)
679 struct v3d_key local_key
;
680 struct v3d_key
*key
= &local_key
;
682 if (!(v3d
->dirty
& (VC5_DIRTY_UNCOMPILED_CS
|
683 VC5_DIRTY_COMPTEX
))) {
687 memset(key
, 0, sizeof(*key
));
688 v3d_setup_shared_key(v3d
, key
, &v3d
->tex
[PIPE_SHADER_COMPUTE
]);
689 key
->shader_state
= v3d
->prog
.bind_compute
;
691 struct v3d_compiled_shader
*cs
=
692 v3d_get_compiled_shader(v3d
, key
, sizeof(*key
));
693 if (cs
!= v3d
->prog
.compute
) {
694 v3d
->prog
.compute
= cs
;
695 v3d
->dirty
|= VC5_DIRTY_COMPILED_CS
; /* XXX */
700 fs_cache_hash(const void *key
)
702 return _mesa_hash_data(key
, sizeof(struct v3d_fs_key
));
706 vs_cache_hash(const void *key
)
708 return _mesa_hash_data(key
, sizeof(struct v3d_vs_key
));
712 cs_cache_hash(const void *key
)
714 return _mesa_hash_data(key
, sizeof(struct v3d_key
));
718 fs_cache_compare(const void *key1
, const void *key2
)
720 return memcmp(key1
, key2
, sizeof(struct v3d_fs_key
)) == 0;
724 vs_cache_compare(const void *key1
, const void *key2
)
726 return memcmp(key1
, key2
, sizeof(struct v3d_vs_key
)) == 0;
730 cs_cache_compare(const void *key1
, const void *key2
)
732 return memcmp(key1
, key2
, sizeof(struct v3d_key
)) == 0;
736 v3d_shader_state_delete(struct pipe_context
*pctx
, void *hwcso
)
738 struct v3d_context
*v3d
= v3d_context(pctx
);
739 struct v3d_uncompiled_shader
*so
= hwcso
;
740 nir_shader
*s
= so
->base
.ir
.nir
;
742 hash_table_foreach(v3d
->prog
.cache
[s
->info
.stage
], entry
) {
743 const struct v3d_key
*key
= entry
->key
;
744 struct v3d_compiled_shader
*shader
= entry
->data
;
746 if (key
->shader_state
!= so
)
749 if (v3d
->prog
.fs
== shader
)
751 if (v3d
->prog
.vs
== shader
)
753 if (v3d
->prog
.cs
== shader
)
755 if (v3d
->prog
.compute
== shader
)
756 v3d
->prog
.compute
= NULL
;
758 _mesa_hash_table_remove(v3d
->prog
.cache
[s
->info
.stage
], entry
);
759 v3d_free_compiled_shader(shader
);
762 ralloc_free(so
->base
.ir
.nir
);
767 v3d_fp_state_bind(struct pipe_context
*pctx
, void *hwcso
)
769 struct v3d_context
*v3d
= v3d_context(pctx
);
770 v3d
->prog
.bind_fs
= hwcso
;
771 v3d
->dirty
|= VC5_DIRTY_UNCOMPILED_FS
;
775 v3d_vp_state_bind(struct pipe_context
*pctx
, void *hwcso
)
777 struct v3d_context
*v3d
= v3d_context(pctx
);
778 v3d
->prog
.bind_vs
= hwcso
;
779 v3d
->dirty
|= VC5_DIRTY_UNCOMPILED_VS
;
783 v3d_compute_state_bind(struct pipe_context
*pctx
, void *state
)
785 struct v3d_context
*v3d
= v3d_context(pctx
);
787 v3d
->prog
.bind_compute
= state
;
788 v3d
->dirty
|= VC5_DIRTY_UNCOMPILED_CS
;
792 v3d_create_compute_state(struct pipe_context
*pctx
,
793 const struct pipe_compute_state
*cso
)
795 return v3d_uncompiled_shader_create(pctx
, cso
->ir_type
,
800 v3d_program_init(struct pipe_context
*pctx
)
802 struct v3d_context
*v3d
= v3d_context(pctx
);
804 pctx
->create_vs_state
= v3d_shader_state_create
;
805 pctx
->delete_vs_state
= v3d_shader_state_delete
;
807 pctx
->create_fs_state
= v3d_shader_state_create
;
808 pctx
->delete_fs_state
= v3d_shader_state_delete
;
810 pctx
->bind_fs_state
= v3d_fp_state_bind
;
811 pctx
->bind_vs_state
= v3d_vp_state_bind
;
813 if (v3d
->screen
->has_csd
) {
814 pctx
->create_compute_state
= v3d_create_compute_state
;
815 pctx
->delete_compute_state
= v3d_shader_state_delete
;
816 pctx
->bind_compute_state
= v3d_compute_state_bind
;
819 v3d
->prog
.cache
[MESA_SHADER_VERTEX
] =
820 _mesa_hash_table_create(pctx
, vs_cache_hash
, vs_cache_compare
);
821 v3d
->prog
.cache
[MESA_SHADER_FRAGMENT
] =
822 _mesa_hash_table_create(pctx
, fs_cache_hash
, fs_cache_compare
);
823 v3d
->prog
.cache
[MESA_SHADER_COMPUTE
] =
824 _mesa_hash_table_create(pctx
, cs_cache_hash
, cs_cache_compare
);
828 v3d_program_fini(struct pipe_context
*pctx
)
830 struct v3d_context
*v3d
= v3d_context(pctx
);
832 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
833 struct hash_table
*cache
= v3d
->prog
.cache
[i
];
837 hash_table_foreach(cache
, entry
) {
838 struct v3d_compiled_shader
*shader
= entry
->data
;
839 v3d_free_compiled_shader(shader
);
840 _mesa_hash_table_remove(cache
, entry
);
844 v3d_bo_unreference(&v3d
->prog
.spill_bo
);