1 /**************************************************************************
3 * Copyright 2007 VMware, Inc.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * 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, sub license, 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 portions
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
29 * Keith Whitwell <keithw@vmware.com>
34 #include "main/imports.h"
35 #include "main/hash.h"
36 #include "main/mtypes.h"
37 #include "program/prog_parameter.h"
38 #include "program/prog_print.h"
39 #include "program/programopt.h"
41 #include "compiler/nir/nir.h"
43 #include "pipe/p_context.h"
44 #include "pipe/p_defines.h"
45 #include "pipe/p_shader_tokens.h"
46 #include "draw/draw_context.h"
47 #include "tgsi/tgsi_dump.h"
48 #include "tgsi/tgsi_emulate.h"
49 #include "tgsi/tgsi_parse.h"
50 #include "tgsi/tgsi_ureg.h"
53 #include "st_cb_bitmap.h"
54 #include "st_cb_drawpixels.h"
55 #include "st_context.h"
56 #include "st_tgsi_lower_yuv.h"
57 #include "st_program.h"
58 #include "st_mesa_to_tgsi.h"
59 #include "st_atifs_to_tgsi.h"
61 #include "st_shader_cache.h"
62 #include "cso_cache/cso_context.h"
67 set_affected_state_flags(uint64_t *states
,
68 struct gl_program
*prog
,
69 uint64_t new_constants
,
70 uint64_t new_sampler_views
,
71 uint64_t new_samplers
,
77 if (prog
->Parameters
->NumParameters
)
78 *states
|= new_constants
;
80 if (prog
->info
.num_textures
)
81 *states
|= new_sampler_views
| new_samplers
;
83 if (prog
->info
.num_images
)
84 *states
|= new_images
;
86 if (prog
->info
.num_ubos
)
89 if (prog
->info
.num_ssbos
)
92 if (prog
->info
.num_abos
)
93 *states
|= new_atomics
;
97 * This determines which states will be updated when the shader is bound.
100 st_set_prog_affected_state_flags(struct gl_program
*prog
)
104 switch (prog
->info
.stage
) {
105 case MESA_SHADER_VERTEX
:
106 states
= &((struct st_vertex_program
*)prog
)->affected_states
;
108 *states
= ST_NEW_VS_STATE
|
110 ST_NEW_VERTEX_ARRAYS
;
112 set_affected_state_flags(states
, prog
,
114 ST_NEW_VS_SAMPLER_VIEWS
,
122 case MESA_SHADER_TESS_CTRL
:
123 states
= &(st_common_program(prog
))->affected_states
;
125 *states
= ST_NEW_TCS_STATE
;
127 set_affected_state_flags(states
, prog
,
128 ST_NEW_TCS_CONSTANTS
,
129 ST_NEW_TCS_SAMPLER_VIEWS
,
137 case MESA_SHADER_TESS_EVAL
:
138 states
= &(st_common_program(prog
))->affected_states
;
140 *states
= ST_NEW_TES_STATE
|
143 set_affected_state_flags(states
, prog
,
144 ST_NEW_TES_CONSTANTS
,
145 ST_NEW_TES_SAMPLER_VIEWS
,
153 case MESA_SHADER_GEOMETRY
:
154 states
= &(st_common_program(prog
))->affected_states
;
156 *states
= ST_NEW_GS_STATE
|
159 set_affected_state_flags(states
, prog
,
161 ST_NEW_GS_SAMPLER_VIEWS
,
169 case MESA_SHADER_FRAGMENT
:
170 states
= &((struct st_fragment_program
*)prog
)->affected_states
;
172 /* gl_FragCoord and glDrawPixels always use constants. */
173 *states
= ST_NEW_FS_STATE
|
174 ST_NEW_SAMPLE_SHADING
|
177 set_affected_state_flags(states
, prog
,
179 ST_NEW_FS_SAMPLER_VIEWS
,
187 case MESA_SHADER_COMPUTE
:
188 states
= &((struct st_compute_program
*)prog
)->affected_states
;
190 *states
= ST_NEW_CS_STATE
;
192 set_affected_state_flags(states
, prog
,
194 ST_NEW_CS_SAMPLER_VIEWS
,
203 unreachable("unhandled shader stage");
208 * Delete a vertex program variant. Note the caller must unlink
209 * the variant from the linked list.
212 delete_vp_variant(struct st_context
*st
, struct st_vp_variant
*vpv
)
214 if (vpv
->driver_shader
)
215 cso_delete_vertex_shader(st
->cso_context
, vpv
->driver_shader
);
217 if (vpv
->draw_shader
)
218 draw_delete_vertex_shader( st
->draw
, vpv
->draw_shader
);
220 if (((vpv
->tgsi
.type
== PIPE_SHADER_IR_TGSI
)) && vpv
->tgsi
.tokens
)
221 ureg_free_tokens(vpv
->tgsi
.tokens
);
229 * Clean out any old compilations:
232 st_release_vp_variants( struct st_context
*st
,
233 struct st_vertex_program
*stvp
)
235 struct st_vp_variant
*vpv
;
237 for (vpv
= stvp
->variants
; vpv
; ) {
238 struct st_vp_variant
*next
= vpv
->next
;
239 delete_vp_variant(st
, vpv
);
243 stvp
->variants
= NULL
;
245 if ((stvp
->tgsi
.type
== PIPE_SHADER_IR_TGSI
) && stvp
->tgsi
.tokens
) {
246 tgsi_free_tokens(stvp
->tgsi
.tokens
);
247 stvp
->tgsi
.tokens
= NULL
;
254 * Delete a fragment program variant. Note the caller must unlink
255 * the variant from the linked list.
258 delete_fp_variant(struct st_context
*st
, struct st_fp_variant
*fpv
)
260 if (fpv
->driver_shader
)
261 cso_delete_fragment_shader(st
->cso_context
, fpv
->driver_shader
);
267 * Free all variants of a fragment program.
270 st_release_fp_variants(struct st_context
*st
, struct st_fragment_program
*stfp
)
272 struct st_fp_variant
*fpv
;
274 for (fpv
= stfp
->variants
; fpv
; ) {
275 struct st_fp_variant
*next
= fpv
->next
;
276 delete_fp_variant(st
, fpv
);
280 stfp
->variants
= NULL
;
282 if ((stfp
->tgsi
.type
== PIPE_SHADER_IR_TGSI
) && stfp
->tgsi
.tokens
) {
283 ureg_free_tokens(stfp
->tgsi
.tokens
);
284 stfp
->tgsi
.tokens
= NULL
;
290 * Delete a basic program variant. Note the caller must unlink
291 * the variant from the linked list.
294 delete_basic_variant(struct st_context
*st
, struct st_basic_variant
*v
,
297 if (v
->driver_shader
) {
299 case GL_TESS_CONTROL_PROGRAM_NV
:
300 cso_delete_tessctrl_shader(st
->cso_context
, v
->driver_shader
);
302 case GL_TESS_EVALUATION_PROGRAM_NV
:
303 cso_delete_tesseval_shader(st
->cso_context
, v
->driver_shader
);
305 case GL_GEOMETRY_PROGRAM_NV
:
306 cso_delete_geometry_shader(st
->cso_context
, v
->driver_shader
);
308 case GL_COMPUTE_PROGRAM_NV
:
309 cso_delete_compute_shader(st
->cso_context
, v
->driver_shader
);
312 assert(!"this shouldn't occur");
321 * Free all basic program variants.
324 st_release_basic_variants(struct st_context
*st
, GLenum target
,
325 struct st_basic_variant
**variants
,
326 struct pipe_shader_state
*tgsi
)
328 struct st_basic_variant
*v
;
330 for (v
= *variants
; v
; ) {
331 struct st_basic_variant
*next
= v
->next
;
332 delete_basic_variant(st
, v
, target
);
339 ureg_free_tokens(tgsi
->tokens
);
346 * Free all variants of a compute program.
349 st_release_cp_variants(struct st_context
*st
, struct st_compute_program
*stcp
)
351 struct st_basic_variant
**variants
= &stcp
->variants
;
352 struct st_basic_variant
*v
;
354 for (v
= *variants
; v
; ) {
355 struct st_basic_variant
*next
= v
->next
;
356 delete_basic_variant(st
, v
, stcp
->Base
.Target
);
362 if (stcp
->tgsi
.prog
) {
363 switch (stcp
->tgsi
.ir_type
) {
364 case PIPE_SHADER_IR_TGSI
:
365 ureg_free_tokens(stcp
->tgsi
.prog
);
366 stcp
->tgsi
.prog
= NULL
;
368 case PIPE_SHADER_IR_NIR
:
369 /* pipe driver took ownership of prog */
371 case PIPE_SHADER_IR_NATIVE
:
373 stcp
->tgsi
.prog
= NULL
;
380 * Translate a vertex program.
383 st_translate_vertex_program(struct st_context
*st
,
384 struct st_vertex_program
*stvp
)
386 struct ureg_program
*ureg
;
387 enum pipe_error error
;
388 unsigned num_outputs
= 0;
390 ubyte input_to_index
[VERT_ATTRIB_MAX
] = {0};
391 ubyte output_semantic_name
[VARYING_SLOT_MAX
] = {0};
392 ubyte output_semantic_index
[VARYING_SLOT_MAX
] = {0};
394 stvp
->num_inputs
= 0;
396 if (stvp
->Base
.arb
.IsPositionInvariant
)
397 _mesa_insert_mvp_code(st
->ctx
, &stvp
->Base
);
400 * Determine number of inputs, the mappings between VERT_ATTRIB_x
401 * and TGSI generic input indexes, plus input attrib semantic info.
403 for (attr
= 0; attr
< VERT_ATTRIB_MAX
; attr
++) {
404 if ((stvp
->Base
.info
.inputs_read
& BITFIELD64_BIT(attr
)) != 0) {
405 input_to_index
[attr
] = stvp
->num_inputs
;
406 stvp
->index_to_input
[stvp
->num_inputs
] = attr
;
408 if ((stvp
->Base
.info
.vs
.double_inputs_read
&
409 BITFIELD64_BIT(attr
)) != 0) {
410 /* add placeholder for second part of a double attribute */
411 stvp
->index_to_input
[stvp
->num_inputs
] = ST_DOUBLE_ATTRIB_PLACEHOLDER
;
416 /* bit of a hack, presetup potentially unused edgeflag input */
417 input_to_index
[VERT_ATTRIB_EDGEFLAG
] = stvp
->num_inputs
;
418 stvp
->index_to_input
[stvp
->num_inputs
] = VERT_ATTRIB_EDGEFLAG
;
420 /* Compute mapping of vertex program outputs to slots.
422 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
423 if ((stvp
->Base
.info
.outputs_written
& BITFIELD64_BIT(attr
)) == 0) {
424 stvp
->result_to_output
[attr
] = ~0;
427 unsigned slot
= num_outputs
++;
429 stvp
->result_to_output
[attr
] = slot
;
431 unsigned semantic_name
, semantic_index
;
432 tgsi_get_gl_varying_semantic(attr
, st
->needs_texcoord_semantic
,
433 &semantic_name
, &semantic_index
);
434 output_semantic_name
[slot
] = semantic_name
;
435 output_semantic_index
[slot
] = semantic_index
;
438 /* similar hack to above, presetup potentially unused edgeflag output */
439 stvp
->result_to_output
[VARYING_SLOT_EDGE
] = num_outputs
;
440 output_semantic_name
[num_outputs
] = TGSI_SEMANTIC_EDGEFLAG
;
441 output_semantic_index
[num_outputs
] = 0;
444 if (!stvp
->glsl_to_tgsi
&& !stvp
->shader_program
) {
445 _mesa_remove_output_reads(&stvp
->Base
, PROGRAM_OUTPUT
);
447 /* This determines which states will be updated when the assembly
450 stvp
->affected_states
= ST_NEW_VS_STATE
|
452 ST_NEW_VERTEX_ARRAYS
;
454 if (stvp
->Base
.Parameters
->NumParameters
)
455 stvp
->affected_states
|= ST_NEW_VS_CONSTANTS
;
457 /* No samplers are allowed in ARB_vp. */
460 if (stvp
->shader_program
) {
461 struct gl_program
*prog
= stvp
->shader_program
->last_vert_prog
;
463 st_translate_stream_output_info2(prog
->sh
.LinkedTransformFeedback
,
464 stvp
->result_to_output
,
465 &stvp
->tgsi
.stream_output
);
468 st_store_ir_in_disk_cache(st
, &stvp
->Base
, true);
472 ureg
= ureg_create_with_screen(PIPE_SHADER_VERTEX
, st
->pipe
->screen
);
476 if (stvp
->Base
.info
.clip_distance_array_size
)
477 ureg_property(ureg
, TGSI_PROPERTY_NUM_CLIPDIST_ENABLED
,
478 stvp
->Base
.info
.clip_distance_array_size
);
479 if (stvp
->Base
.info
.cull_distance_array_size
)
480 ureg_property(ureg
, TGSI_PROPERTY_NUM_CULLDIST_ENABLED
,
481 stvp
->Base
.info
.cull_distance_array_size
);
483 if (ST_DEBUG
& DEBUG_MESA
) {
484 _mesa_print_program(&stvp
->Base
);
485 _mesa_print_program_parameters(st
->ctx
, &stvp
->Base
);
489 if (stvp
->glsl_to_tgsi
) {
490 error
= st_translate_program(st
->ctx
,
498 NULL
, /* inputSlotToAttr */
499 NULL
, /* input semantic name */
500 NULL
, /* input semantic index */
501 NULL
, /* interp mode */
504 stvp
->result_to_output
,
505 output_semantic_name
,
506 output_semantic_index
);
508 st_translate_stream_output_info(stvp
->glsl_to_tgsi
,
509 stvp
->result_to_output
,
510 &stvp
->tgsi
.stream_output
);
512 free_glsl_to_tgsi_visitor(stvp
->glsl_to_tgsi
);
514 error
= st_translate_mesa_program(st
->ctx
,
521 NULL
, /* input semantic name */
522 NULL
, /* input semantic index */
526 stvp
->result_to_output
,
527 output_semantic_name
,
528 output_semantic_index
);
531 debug_printf("%s: failed to translate Mesa program:\n", __func__
);
532 _mesa_print_program(&stvp
->Base
);
537 stvp
->tgsi
.tokens
= ureg_get_tokens(ureg
, &stvp
->num_tgsi_tokens
);
540 if (stvp
->glsl_to_tgsi
) {
541 stvp
->glsl_to_tgsi
= NULL
;
542 st_store_ir_in_disk_cache(st
, &stvp
->Base
, false);
545 return stvp
->tgsi
.tokens
!= NULL
;
548 static struct st_vp_variant
*
549 st_create_vp_variant(struct st_context
*st
,
550 struct st_vertex_program
*stvp
,
551 const struct st_vp_variant_key
*key
)
553 struct st_vp_variant
*vpv
= CALLOC_STRUCT(st_vp_variant
);
554 struct pipe_context
*pipe
= st
->pipe
;
557 vpv
->tgsi
.stream_output
= stvp
->tgsi
.stream_output
;
558 vpv
->num_inputs
= stvp
->num_inputs
;
560 if (stvp
->tgsi
.type
== PIPE_SHADER_IR_NIR
) {
561 vpv
->tgsi
.type
= PIPE_SHADER_IR_NIR
;
562 vpv
->tgsi
.ir
.nir
= nir_shader_clone(NULL
, stvp
->tgsi
.ir
.nir
);
563 if (key
->clamp_color
)
564 NIR_PASS_V(vpv
->tgsi
.ir
.nir
, nir_lower_clamp_color_outputs
);
565 if (key
->passthrough_edgeflags
) {
566 NIR_PASS_V(vpv
->tgsi
.ir
.nir
, nir_lower_passthrough_edgeflags
);
570 st_finalize_nir(st
, &stvp
->Base
, stvp
->shader_program
,
573 vpv
->driver_shader
= pipe
->create_vs_state(pipe
, &vpv
->tgsi
);
574 /* driver takes ownership of IR: */
575 vpv
->tgsi
.ir
.nir
= NULL
;
579 vpv
->tgsi
.tokens
= tgsi_dup_tokens(stvp
->tgsi
.tokens
);
581 /* Emulate features. */
582 if (key
->clamp_color
|| key
->passthrough_edgeflags
) {
583 const struct tgsi_token
*tokens
;
585 (key
->clamp_color
? TGSI_EMU_CLAMP_COLOR_OUTPUTS
: 0) |
586 (key
->passthrough_edgeflags
? TGSI_EMU_PASSTHROUGH_EDGEFLAG
: 0);
588 tokens
= tgsi_emulate(vpv
->tgsi
.tokens
, flags
);
591 tgsi_free_tokens(vpv
->tgsi
.tokens
);
592 vpv
->tgsi
.tokens
= tokens
;
594 if (key
->passthrough_edgeflags
)
597 fprintf(stderr
, "mesa: cannot emulate deprecated features\n");
600 if (ST_DEBUG
& DEBUG_TGSI
) {
601 tgsi_dump(vpv
->tgsi
.tokens
, 0);
605 vpv
->driver_shader
= pipe
->create_vs_state(pipe
, &vpv
->tgsi
);
611 * Find/create a vertex program variant.
613 struct st_vp_variant
*
614 st_get_vp_variant(struct st_context
*st
,
615 struct st_vertex_program
*stvp
,
616 const struct st_vp_variant_key
*key
)
618 struct st_vp_variant
*vpv
;
620 /* Search for existing variant */
621 for (vpv
= stvp
->variants
; vpv
; vpv
= vpv
->next
) {
622 if (memcmp(&vpv
->key
, key
, sizeof(*key
)) == 0) {
629 vpv
= st_create_vp_variant(st
, stvp
, key
);
631 /* insert into list */
632 vpv
->next
= stvp
->variants
;
633 stvp
->variants
= vpv
;
642 * Translate a Mesa fragment shader into a TGSI shader.
645 st_translate_fragment_program(struct st_context
*st
,
646 struct st_fragment_program
*stfp
)
648 ubyte outputMapping
[2 * FRAG_RESULT_MAX
];
649 ubyte inputMapping
[VARYING_SLOT_MAX
];
650 ubyte inputSlotToAttr
[VARYING_SLOT_MAX
];
651 ubyte interpMode
[PIPE_MAX_SHADER_INPUTS
]; /* XXX size? */
653 GLbitfield64 inputsRead
;
654 struct ureg_program
*ureg
;
656 GLboolean write_all
= GL_FALSE
;
658 ubyte input_semantic_name
[PIPE_MAX_SHADER_INPUTS
];
659 ubyte input_semantic_index
[PIPE_MAX_SHADER_INPUTS
];
660 uint fs_num_inputs
= 0;
662 ubyte fs_output_semantic_name
[PIPE_MAX_SHADER_OUTPUTS
];
663 ubyte fs_output_semantic_index
[PIPE_MAX_SHADER_OUTPUTS
];
664 uint fs_num_outputs
= 0;
666 memset(inputSlotToAttr
, ~0, sizeof(inputSlotToAttr
));
668 /* Non-GLSL programs: */
669 if (!stfp
->glsl_to_tgsi
&& !stfp
->shader_program
) {
670 _mesa_remove_output_reads(&stfp
->Base
, PROGRAM_OUTPUT
);
671 if (st
->ctx
->Const
.GLSLFragCoordIsSysVal
)
672 _mesa_program_fragment_position_to_sysval(&stfp
->Base
);
674 /* This determines which states will be updated when the assembly
677 * fragment.position and glDrawPixels always use constants.
679 stfp
->affected_states
= ST_NEW_FS_STATE
|
680 ST_NEW_SAMPLE_SHADING
|
684 /* Just set them for ATI_fs unconditionally. */
685 stfp
->affected_states
|= ST_NEW_FS_SAMPLER_VIEWS
|
689 if (stfp
->Base
.SamplersUsed
)
690 stfp
->affected_states
|= ST_NEW_FS_SAMPLER_VIEWS
|
696 * Convert Mesa program inputs to TGSI input register semantics.
698 inputsRead
= stfp
->Base
.info
.inputs_read
;
699 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
700 if ((inputsRead
& BITFIELD64_BIT(attr
)) != 0) {
701 const GLuint slot
= fs_num_inputs
++;
703 inputMapping
[attr
] = slot
;
704 inputSlotToAttr
[slot
] = attr
;
707 case VARYING_SLOT_POS
:
708 input_semantic_name
[slot
] = TGSI_SEMANTIC_POSITION
;
709 input_semantic_index
[slot
] = 0;
710 interpMode
[slot
] = TGSI_INTERPOLATE_LINEAR
;
712 case VARYING_SLOT_COL0
:
713 input_semantic_name
[slot
] = TGSI_SEMANTIC_COLOR
;
714 input_semantic_index
[slot
] = 0;
715 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
716 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_COLOR
;
718 case VARYING_SLOT_COL1
:
719 input_semantic_name
[slot
] = TGSI_SEMANTIC_COLOR
;
720 input_semantic_index
[slot
] = 1;
721 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
722 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_COLOR
;
724 case VARYING_SLOT_FOGC
:
725 input_semantic_name
[slot
] = TGSI_SEMANTIC_FOG
;
726 input_semantic_index
[slot
] = 0;
727 interpMode
[slot
] = TGSI_INTERPOLATE_PERSPECTIVE
;
729 case VARYING_SLOT_FACE
:
730 input_semantic_name
[slot
] = TGSI_SEMANTIC_FACE
;
731 input_semantic_index
[slot
] = 0;
732 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
734 case VARYING_SLOT_PRIMITIVE_ID
:
735 input_semantic_name
[slot
] = TGSI_SEMANTIC_PRIMID
;
736 input_semantic_index
[slot
] = 0;
737 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
739 case VARYING_SLOT_LAYER
:
740 input_semantic_name
[slot
] = TGSI_SEMANTIC_LAYER
;
741 input_semantic_index
[slot
] = 0;
742 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
744 case VARYING_SLOT_VIEWPORT
:
745 input_semantic_name
[slot
] = TGSI_SEMANTIC_VIEWPORT_INDEX
;
746 input_semantic_index
[slot
] = 0;
747 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
749 case VARYING_SLOT_CLIP_DIST0
:
750 input_semantic_name
[slot
] = TGSI_SEMANTIC_CLIPDIST
;
751 input_semantic_index
[slot
] = 0;
752 interpMode
[slot
] = TGSI_INTERPOLATE_PERSPECTIVE
;
754 case VARYING_SLOT_CLIP_DIST1
:
755 input_semantic_name
[slot
] = TGSI_SEMANTIC_CLIPDIST
;
756 input_semantic_index
[slot
] = 1;
757 interpMode
[slot
] = TGSI_INTERPOLATE_PERSPECTIVE
;
759 case VARYING_SLOT_CULL_DIST0
:
760 case VARYING_SLOT_CULL_DIST1
:
761 /* these should have been lowered by GLSL */
764 /* In most cases, there is nothing special about these
765 * inputs, so adopt a convention to use the generic
766 * semantic name and the mesa VARYING_SLOT_ number as the
769 * All that is required is that the vertex shader labels
770 * its own outputs similarly, and that the vertex shader
771 * generates at least every output required by the
772 * fragment shader plus fixed-function hardware (such as
775 * However, some drivers may need us to identify the PNTC and TEXi
776 * varyings if, for example, their capability to replace them with
777 * sprite coordinates is limited.
779 case VARYING_SLOT_PNTC
:
780 if (st
->needs_texcoord_semantic
) {
781 input_semantic_name
[slot
] = TGSI_SEMANTIC_PCOORD
;
782 input_semantic_index
[slot
] = 0;
783 interpMode
[slot
] = TGSI_INTERPOLATE_LINEAR
;
787 case VARYING_SLOT_TEX0
:
788 case VARYING_SLOT_TEX1
:
789 case VARYING_SLOT_TEX2
:
790 case VARYING_SLOT_TEX3
:
791 case VARYING_SLOT_TEX4
:
792 case VARYING_SLOT_TEX5
:
793 case VARYING_SLOT_TEX6
:
794 case VARYING_SLOT_TEX7
:
795 if (st
->needs_texcoord_semantic
) {
796 input_semantic_name
[slot
] = TGSI_SEMANTIC_TEXCOORD
;
797 input_semantic_index
[slot
] = attr
- VARYING_SLOT_TEX0
;
798 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
799 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_PERSPECTIVE
;
803 case VARYING_SLOT_VAR0
:
805 /* Semantic indices should be zero-based because drivers may choose
806 * to assign a fixed slot determined by that index.
807 * This is useful because ARB_separate_shader_objects uses location
808 * qualifiers for linkage, and if the semantic index corresponds to
809 * these locations, linkage passes in the driver become unecessary.
811 * If needs_texcoord_semantic is true, no semantic indices will be
812 * consumed for the TEXi varyings, and we can base the locations of
813 * the user varyings on VAR0. Otherwise, we use TEX0 as base index.
815 assert(attr
>= VARYING_SLOT_VAR0
|| attr
== VARYING_SLOT_PNTC
||
816 (attr
>= VARYING_SLOT_TEX0
&& attr
<= VARYING_SLOT_TEX7
));
817 input_semantic_name
[slot
] = TGSI_SEMANTIC_GENERIC
;
818 input_semantic_index
[slot
] = st_get_generic_varying_index(st
, attr
);
819 if (attr
== VARYING_SLOT_PNTC
)
820 interpMode
[slot
] = TGSI_INTERPOLATE_LINEAR
;
822 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
823 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_PERSPECTIVE
;
829 inputMapping
[attr
] = -1;
834 * Semantics and mapping for outputs
836 GLbitfield64 outputsWritten
= stfp
->Base
.info
.outputs_written
;
838 /* if z is written, emit that first */
839 if (outputsWritten
& BITFIELD64_BIT(FRAG_RESULT_DEPTH
)) {
840 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_POSITION
;
841 fs_output_semantic_index
[fs_num_outputs
] = 0;
842 outputMapping
[FRAG_RESULT_DEPTH
] = fs_num_outputs
;
844 outputsWritten
&= ~(1 << FRAG_RESULT_DEPTH
);
847 if (outputsWritten
& BITFIELD64_BIT(FRAG_RESULT_STENCIL
)) {
848 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_STENCIL
;
849 fs_output_semantic_index
[fs_num_outputs
] = 0;
850 outputMapping
[FRAG_RESULT_STENCIL
] = fs_num_outputs
;
852 outputsWritten
&= ~(1 << FRAG_RESULT_STENCIL
);
855 if (outputsWritten
& BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK
)) {
856 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_SAMPLEMASK
;
857 fs_output_semantic_index
[fs_num_outputs
] = 0;
858 outputMapping
[FRAG_RESULT_SAMPLE_MASK
] = fs_num_outputs
;
860 outputsWritten
&= ~(1 << FRAG_RESULT_SAMPLE_MASK
);
863 /* handle remaining outputs (color) */
864 for (attr
= 0; attr
< ARRAY_SIZE(outputMapping
); attr
++) {
865 const GLbitfield64 written
= attr
< FRAG_RESULT_MAX
? outputsWritten
:
866 stfp
->Base
.SecondaryOutputsWritten
;
867 const unsigned loc
= attr
% FRAG_RESULT_MAX
;
869 if (written
& BITFIELD64_BIT(loc
)) {
871 case FRAG_RESULT_DEPTH
:
872 case FRAG_RESULT_STENCIL
:
873 case FRAG_RESULT_SAMPLE_MASK
:
877 case FRAG_RESULT_COLOR
:
878 write_all
= GL_TRUE
; /* fallthrough */
881 assert(loc
== FRAG_RESULT_COLOR
||
882 (FRAG_RESULT_DATA0
<= loc
&& loc
< FRAG_RESULT_MAX
));
884 index
= (loc
== FRAG_RESULT_COLOR
) ? 0 : (loc
- FRAG_RESULT_DATA0
);
886 if (attr
>= FRAG_RESULT_MAX
) {
887 /* Secondary color for dual source blending. */
892 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_COLOR
;
893 fs_output_semantic_index
[fs_num_outputs
] = index
;
894 outputMapping
[attr
] = fs_num_outputs
;
903 /* We have already compiled to NIR so just return */
904 if (stfp
->shader_program
) {
905 st_store_ir_in_disk_cache(st
, &stfp
->Base
, true);
909 ureg
= ureg_create_with_screen(PIPE_SHADER_FRAGMENT
, st
->pipe
->screen
);
913 if (ST_DEBUG
& DEBUG_MESA
) {
914 _mesa_print_program(&stfp
->Base
);
915 _mesa_print_program_parameters(st
->ctx
, &stfp
->Base
);
918 if (write_all
== GL_TRUE
)
919 ureg_property(ureg
, TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS
, 1);
921 if (stfp
->Base
.info
.fs
.depth_layout
!= FRAG_DEPTH_LAYOUT_NONE
) {
922 switch (stfp
->Base
.info
.fs
.depth_layout
) {
923 case FRAG_DEPTH_LAYOUT_ANY
:
924 ureg_property(ureg
, TGSI_PROPERTY_FS_DEPTH_LAYOUT
,
925 TGSI_FS_DEPTH_LAYOUT_ANY
);
927 case FRAG_DEPTH_LAYOUT_GREATER
:
928 ureg_property(ureg
, TGSI_PROPERTY_FS_DEPTH_LAYOUT
,
929 TGSI_FS_DEPTH_LAYOUT_GREATER
);
931 case FRAG_DEPTH_LAYOUT_LESS
:
932 ureg_property(ureg
, TGSI_PROPERTY_FS_DEPTH_LAYOUT
,
933 TGSI_FS_DEPTH_LAYOUT_LESS
);
935 case FRAG_DEPTH_LAYOUT_UNCHANGED
:
936 ureg_property(ureg
, TGSI_PROPERTY_FS_DEPTH_LAYOUT
,
937 TGSI_FS_DEPTH_LAYOUT_UNCHANGED
);
944 if (stfp
->glsl_to_tgsi
) {
945 st_translate_program(st
->ctx
,
946 PIPE_SHADER_FRAGMENT
,
955 input_semantic_index
,
960 fs_output_semantic_name
,
961 fs_output_semantic_index
);
963 free_glsl_to_tgsi_visitor(stfp
->glsl_to_tgsi
);
964 } else if (stfp
->ati_fs
)
965 st_translate_atifs_program(ureg
,
972 input_semantic_index
,
977 fs_output_semantic_name
,
978 fs_output_semantic_index
);
980 st_translate_mesa_program(st
->ctx
,
981 PIPE_SHADER_FRAGMENT
,
988 input_semantic_index
,
993 fs_output_semantic_name
,
994 fs_output_semantic_index
);
996 stfp
->tgsi
.tokens
= ureg_get_tokens(ureg
, &stfp
->num_tgsi_tokens
);
999 if (stfp
->glsl_to_tgsi
) {
1000 stfp
->glsl_to_tgsi
= NULL
;
1001 st_store_ir_in_disk_cache(st
, &stfp
->Base
, false);
1004 return stfp
->tgsi
.tokens
!= NULL
;
1007 static struct st_fp_variant
*
1008 st_create_fp_variant(struct st_context
*st
,
1009 struct st_fragment_program
*stfp
,
1010 const struct st_fp_variant_key
*key
)
1012 struct pipe_context
*pipe
= st
->pipe
;
1013 struct st_fp_variant
*variant
= CALLOC_STRUCT(st_fp_variant
);
1014 struct pipe_shader_state tgsi
= {0};
1015 struct gl_program_parameter_list
*params
= stfp
->Base
.Parameters
;
1016 static const gl_state_index16 texcoord_state
[STATE_LENGTH
] =
1017 { STATE_INTERNAL
, STATE_CURRENT_ATTRIB
, VERT_ATTRIB_TEX0
};
1018 static const gl_state_index16 scale_state
[STATE_LENGTH
] =
1019 { STATE_INTERNAL
, STATE_PT_SCALE
};
1020 static const gl_state_index16 bias_state
[STATE_LENGTH
] =
1021 { STATE_INTERNAL
, STATE_PT_BIAS
};
1026 if (stfp
->tgsi
.type
== PIPE_SHADER_IR_NIR
) {
1027 tgsi
.type
= PIPE_SHADER_IR_NIR
;
1028 tgsi
.ir
.nir
= nir_shader_clone(NULL
, stfp
->tgsi
.ir
.nir
);
1030 if (key
->clamp_color
)
1031 NIR_PASS_V(tgsi
.ir
.nir
, nir_lower_clamp_color_outputs
);
1033 if (key
->persample_shading
) {
1034 nir_shader
*shader
= tgsi
.ir
.nir
;
1035 nir_foreach_variable(var
, &shader
->inputs
)
1036 var
->data
.sample
= true;
1039 assert(!(key
->bitmap
&& key
->drawpixels
));
1043 nir_lower_bitmap_options options
= {0};
1045 variant
->bitmap_sampler
= ffs(~stfp
->Base
.SamplersUsed
) - 1;
1046 options
.sampler
= variant
->bitmap_sampler
;
1047 options
.swizzle_xxxx
= (st
->bitmap
.tex_format
== PIPE_FORMAT_L8_UNORM
);
1049 NIR_PASS_V(tgsi
.ir
.nir
, nir_lower_bitmap
, &options
);
1052 /* glDrawPixels (color only) */
1053 if (key
->drawpixels
) {
1054 nir_lower_drawpixels_options options
= {{0}};
1055 unsigned samplers_used
= stfp
->Base
.SamplersUsed
;
1057 /* Find the first unused slot. */
1058 variant
->drawpix_sampler
= ffs(~samplers_used
) - 1;
1059 options
.drawpix_sampler
= variant
->drawpix_sampler
;
1060 samplers_used
|= (1 << variant
->drawpix_sampler
);
1062 options
.pixel_maps
= key
->pixelMaps
;
1063 if (key
->pixelMaps
) {
1064 variant
->pixelmap_sampler
= ffs(~samplers_used
) - 1;
1065 options
.pixelmap_sampler
= variant
->pixelmap_sampler
;
1068 options
.scale_and_bias
= key
->scaleAndBias
;
1069 if (key
->scaleAndBias
) {
1070 _mesa_add_state_reference(params
, scale_state
);
1071 memcpy(options
.scale_state_tokens
, scale_state
,
1072 sizeof(options
.scale_state_tokens
));
1073 _mesa_add_state_reference(params
, bias_state
);
1074 memcpy(options
.bias_state_tokens
, bias_state
,
1075 sizeof(options
.bias_state_tokens
));
1078 _mesa_add_state_reference(params
, texcoord_state
);
1079 memcpy(options
.texcoord_state_tokens
, texcoord_state
,
1080 sizeof(options
.texcoord_state_tokens
));
1082 NIR_PASS_V(tgsi
.ir
.nir
, nir_lower_drawpixels
, &options
);
1085 if (unlikely(key
->external
.lower_nv12
|| key
->external
.lower_iyuv
)) {
1086 nir_lower_tex_options options
= {0};
1087 options
.lower_y_uv_external
= key
->external
.lower_nv12
;
1088 options
.lower_y_u_v_external
= key
->external
.lower_iyuv
;
1089 NIR_PASS_V(tgsi
.ir
.nir
, nir_lower_tex
, &options
);
1092 st_finalize_nir(st
, &stfp
->Base
, stfp
->shader_program
, tgsi
.ir
.nir
);
1094 if (unlikely(key
->external
.lower_nv12
|| key
->external
.lower_iyuv
)) {
1095 /* This pass needs to happen *after* nir_lower_sampler */
1096 NIR_PASS_V(tgsi
.ir
.nir
, st_nir_lower_tex_src_plane
,
1097 ~stfp
->Base
.SamplersUsed
,
1098 key
->external
.lower_nv12
,
1099 key
->external
.lower_iyuv
);
1102 variant
->driver_shader
= pipe
->create_fs_state(pipe
, &tgsi
);
1103 variant
->key
= *key
;
1108 tgsi
.tokens
= stfp
->tgsi
.tokens
;
1110 assert(!(key
->bitmap
&& key
->drawpixels
));
1112 /* Fix texture targets and add fog for ATI_fs */
1114 const struct tgsi_token
*tokens
= st_fixup_atifs(tgsi
.tokens
, key
);
1117 tgsi
.tokens
= tokens
;
1119 fprintf(stderr
, "mesa: cannot post-process ATI_fs\n");
1122 /* Emulate features. */
1123 if (key
->clamp_color
|| key
->persample_shading
) {
1124 const struct tgsi_token
*tokens
;
1126 (key
->clamp_color
? TGSI_EMU_CLAMP_COLOR_OUTPUTS
: 0) |
1127 (key
->persample_shading
? TGSI_EMU_FORCE_PERSAMPLE_INTERP
: 0);
1129 tokens
= tgsi_emulate(tgsi
.tokens
, flags
);
1132 if (tgsi
.tokens
!= stfp
->tgsi
.tokens
)
1133 tgsi_free_tokens(tgsi
.tokens
);
1134 tgsi
.tokens
= tokens
;
1136 fprintf(stderr
, "mesa: cannot emulate deprecated features\n");
1141 const struct tgsi_token
*tokens
;
1143 variant
->bitmap_sampler
= ffs(~stfp
->Base
.SamplersUsed
) - 1;
1145 tokens
= st_get_bitmap_shader(tgsi
.tokens
,
1146 st
->internal_target
,
1147 variant
->bitmap_sampler
,
1148 st
->needs_texcoord_semantic
,
1149 st
->bitmap
.tex_format
==
1150 PIPE_FORMAT_L8_UNORM
);
1153 if (tgsi
.tokens
!= stfp
->tgsi
.tokens
)
1154 tgsi_free_tokens(tgsi
.tokens
);
1155 tgsi
.tokens
= tokens
;
1157 fprintf(stderr
, "mesa: cannot create a shader for glBitmap\n");
1160 /* glDrawPixels (color only) */
1161 if (key
->drawpixels
) {
1162 const struct tgsi_token
*tokens
;
1163 unsigned scale_const
= 0, bias_const
= 0, texcoord_const
= 0;
1165 /* Find the first unused slot. */
1166 variant
->drawpix_sampler
= ffs(~stfp
->Base
.SamplersUsed
) - 1;
1168 if (key
->pixelMaps
) {
1169 unsigned samplers_used
= stfp
->Base
.SamplersUsed
|
1170 (1 << variant
->drawpix_sampler
);
1172 variant
->pixelmap_sampler
= ffs(~samplers_used
) - 1;
1175 if (key
->scaleAndBias
) {
1176 scale_const
= _mesa_add_state_reference(params
, scale_state
);
1177 bias_const
= _mesa_add_state_reference(params
, bias_state
);
1180 texcoord_const
= _mesa_add_state_reference(params
, texcoord_state
);
1182 tokens
= st_get_drawpix_shader(tgsi
.tokens
,
1183 st
->needs_texcoord_semantic
,
1184 key
->scaleAndBias
, scale_const
,
1185 bias_const
, key
->pixelMaps
,
1186 variant
->drawpix_sampler
,
1187 variant
->pixelmap_sampler
,
1188 texcoord_const
, st
->internal_target
);
1191 if (tgsi
.tokens
!= stfp
->tgsi
.tokens
)
1192 tgsi_free_tokens(tgsi
.tokens
);
1193 tgsi
.tokens
= tokens
;
1195 fprintf(stderr
, "mesa: cannot create a shader for glDrawPixels\n");
1198 if (unlikely(key
->external
.lower_nv12
|| key
->external
.lower_iyuv
)) {
1199 const struct tgsi_token
*tokens
;
1201 /* samplers inserted would conflict, but this should be unpossible: */
1202 assert(!(key
->bitmap
|| key
->drawpixels
));
1204 tokens
= st_tgsi_lower_yuv(tgsi
.tokens
,
1205 ~stfp
->Base
.SamplersUsed
,
1206 key
->external
.lower_nv12
,
1207 key
->external
.lower_iyuv
);
1209 if (tgsi
.tokens
!= stfp
->tgsi
.tokens
)
1210 tgsi_free_tokens(tgsi
.tokens
);
1211 tgsi
.tokens
= tokens
;
1213 fprintf(stderr
, "mesa: cannot create a shader for samplerExternalOES\n");
1217 if (ST_DEBUG
& DEBUG_TGSI
) {
1218 tgsi_dump(tgsi
.tokens
, 0);
1222 /* fill in variant */
1223 variant
->driver_shader
= pipe
->create_fs_state(pipe
, &tgsi
);
1224 variant
->key
= *key
;
1226 if (tgsi
.tokens
!= stfp
->tgsi
.tokens
)
1227 tgsi_free_tokens(tgsi
.tokens
);
1232 * Translate fragment program if needed.
1234 struct st_fp_variant
*
1235 st_get_fp_variant(struct st_context
*st
,
1236 struct st_fragment_program
*stfp
,
1237 const struct st_fp_variant_key
*key
)
1239 struct st_fp_variant
*fpv
;
1241 /* Search for existing variant */
1242 for (fpv
= stfp
->variants
; fpv
; fpv
= fpv
->next
) {
1243 if (memcmp(&fpv
->key
, key
, sizeof(*key
)) == 0) {
1250 fpv
= st_create_fp_variant(st
, stfp
, key
);
1252 if (key
->bitmap
|| key
->drawpixels
) {
1253 /* Regular variants should always come before the
1254 * bitmap & drawpixels variants, (unless there
1255 * are no regular variants) so that
1256 * st_update_fp can take a fast path when
1257 * shader_has_one_variant is set.
1259 if (!stfp
->variants
) {
1260 stfp
->variants
= fpv
;
1262 /* insert into list after the first one */
1263 fpv
->next
= stfp
->variants
->next
;
1264 stfp
->variants
->next
= fpv
;
1267 /* insert into list */
1268 fpv
->next
= stfp
->variants
;
1269 stfp
->variants
= fpv
;
1279 * Translate a program. This is common code for geometry and tessellation
1283 st_translate_program_common(struct st_context
*st
,
1284 struct gl_program
*prog
,
1285 struct glsl_to_tgsi_visitor
*glsl_to_tgsi
,
1286 struct ureg_program
*ureg
,
1287 unsigned tgsi_processor
,
1288 struct pipe_shader_state
*out_state
)
1290 ubyte inputSlotToAttr
[VARYING_SLOT_TESS_MAX
];
1291 ubyte inputMapping
[VARYING_SLOT_TESS_MAX
];
1292 ubyte outputMapping
[VARYING_SLOT_TESS_MAX
];
1295 ubyte input_semantic_name
[PIPE_MAX_SHADER_INPUTS
];
1296 ubyte input_semantic_index
[PIPE_MAX_SHADER_INPUTS
];
1297 uint num_inputs
= 0;
1299 ubyte output_semantic_name
[PIPE_MAX_SHADER_OUTPUTS
];
1300 ubyte output_semantic_index
[PIPE_MAX_SHADER_OUTPUTS
];
1301 uint num_outputs
= 0;
1305 memset(inputSlotToAttr
, 0, sizeof(inputSlotToAttr
));
1306 memset(inputMapping
, 0, sizeof(inputMapping
));
1307 memset(outputMapping
, 0, sizeof(outputMapping
));
1308 memset(out_state
, 0, sizeof(*out_state
));
1310 if (prog
->info
.clip_distance_array_size
)
1311 ureg_property(ureg
, TGSI_PROPERTY_NUM_CLIPDIST_ENABLED
,
1312 prog
->info
.clip_distance_array_size
);
1313 if (prog
->info
.cull_distance_array_size
)
1314 ureg_property(ureg
, TGSI_PROPERTY_NUM_CULLDIST_ENABLED
,
1315 prog
->info
.cull_distance_array_size
);
1318 * Convert Mesa program inputs to TGSI input register semantics.
1320 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
1321 if ((prog
->info
.inputs_read
& BITFIELD64_BIT(attr
)) == 0)
1324 unsigned slot
= num_inputs
++;
1326 inputMapping
[attr
] = slot
;
1327 inputSlotToAttr
[slot
] = attr
;
1329 unsigned semantic_name
, semantic_index
;
1330 tgsi_get_gl_varying_semantic(attr
, st
->needs_texcoord_semantic
,
1331 &semantic_name
, &semantic_index
);
1332 input_semantic_name
[slot
] = semantic_name
;
1333 input_semantic_index
[slot
] = semantic_index
;
1336 /* Also add patch inputs. */
1337 for (attr
= 0; attr
< 32; attr
++) {
1338 if (prog
->info
.patch_inputs_read
& (1u << attr
)) {
1339 GLuint slot
= num_inputs
++;
1340 GLuint patch_attr
= VARYING_SLOT_PATCH0
+ attr
;
1342 inputMapping
[patch_attr
] = slot
;
1343 inputSlotToAttr
[slot
] = patch_attr
;
1344 input_semantic_name
[slot
] = TGSI_SEMANTIC_PATCH
;
1345 input_semantic_index
[slot
] = attr
;
1349 /* initialize output semantics to defaults */
1350 for (i
= 0; i
< PIPE_MAX_SHADER_OUTPUTS
; i
++) {
1351 output_semantic_name
[i
] = TGSI_SEMANTIC_GENERIC
;
1352 output_semantic_index
[i
] = 0;
1356 * Determine number of outputs, the (default) output register
1357 * mapping and the semantic information for each output.
1359 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
1360 if (prog
->info
.outputs_written
& BITFIELD64_BIT(attr
)) {
1361 GLuint slot
= num_outputs
++;
1363 outputMapping
[attr
] = slot
;
1365 unsigned semantic_name
, semantic_index
;
1366 tgsi_get_gl_varying_semantic(attr
, st
->needs_texcoord_semantic
,
1367 &semantic_name
, &semantic_index
);
1368 output_semantic_name
[slot
] = semantic_name
;
1369 output_semantic_index
[slot
] = semantic_index
;
1373 /* Also add patch outputs. */
1374 for (attr
= 0; attr
< 32; attr
++) {
1375 if (prog
->info
.patch_outputs_written
& (1u << attr
)) {
1376 GLuint slot
= num_outputs
++;
1377 GLuint patch_attr
= VARYING_SLOT_PATCH0
+ attr
;
1379 outputMapping
[patch_attr
] = slot
;
1380 output_semantic_name
[slot
] = TGSI_SEMANTIC_PATCH
;
1381 output_semantic_index
[slot
] = attr
;
1385 st_translate_program(st
->ctx
,
1394 input_semantic_name
,
1395 input_semantic_index
,
1400 output_semantic_name
,
1401 output_semantic_index
);
1403 if (tgsi_processor
== PIPE_SHADER_COMPUTE
) {
1404 struct st_compute_program
*stcp
= (struct st_compute_program
*) prog
;
1405 out_state
->tokens
= ureg_get_tokens(ureg
, &stcp
->num_tgsi_tokens
);
1406 stcp
->tgsi
.prog
= out_state
->tokens
;
1408 struct st_common_program
*stcp
= (struct st_common_program
*) prog
;
1409 out_state
->tokens
= ureg_get_tokens(ureg
, &stcp
->num_tgsi_tokens
);
1413 st_translate_stream_output_info(glsl_to_tgsi
,
1415 &out_state
->stream_output
);
1417 st_store_ir_in_disk_cache(st
, prog
, false);
1419 if ((ST_DEBUG
& DEBUG_TGSI
) && (ST_DEBUG
& DEBUG_MESA
)) {
1420 _mesa_print_program(prog
);
1424 if (ST_DEBUG
& DEBUG_TGSI
) {
1425 tgsi_dump(out_state
->tokens
, 0);
1431 * Update stream-output info for GS/TCS/TES. Normally this is done in
1432 * st_translate_program_common() but that is not called for glsl_to_nir
1436 st_translate_program_stream_output(struct gl_program
*prog
,
1437 struct pipe_stream_output_info
*stream_output
)
1439 if (!prog
->sh
.LinkedTransformFeedback
)
1442 ubyte outputMapping
[VARYING_SLOT_TESS_MAX
];
1444 uint num_outputs
= 0;
1446 memset(outputMapping
, 0, sizeof(outputMapping
));
1449 * Determine number of outputs, the (default) output register
1450 * mapping and the semantic information for each output.
1452 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
1453 if (prog
->info
.outputs_written
& BITFIELD64_BIT(attr
)) {
1454 GLuint slot
= num_outputs
++;
1456 outputMapping
[attr
] = slot
;
1460 st_translate_stream_output_info2(prog
->sh
.LinkedTransformFeedback
,
1466 * Translate a geometry program to create a new variant.
1469 st_translate_geometry_program(struct st_context
*st
,
1470 struct st_common_program
*stgp
)
1472 struct ureg_program
*ureg
;
1474 /* We have already compiled to NIR so just return */
1475 if (stgp
->shader_program
) {
1476 st_translate_program_stream_output(&stgp
->Base
, &stgp
->tgsi
.stream_output
);
1477 st_store_ir_in_disk_cache(st
, &stgp
->Base
, true);
1481 ureg
= ureg_create_with_screen(PIPE_SHADER_GEOMETRY
, st
->pipe
->screen
);
1485 ureg_property(ureg
, TGSI_PROPERTY_GS_INPUT_PRIM
,
1486 stgp
->Base
.info
.gs
.input_primitive
);
1487 ureg_property(ureg
, TGSI_PROPERTY_GS_OUTPUT_PRIM
,
1488 stgp
->Base
.info
.gs
.output_primitive
);
1489 ureg_property(ureg
, TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES
,
1490 stgp
->Base
.info
.gs
.vertices_out
);
1491 ureg_property(ureg
, TGSI_PROPERTY_GS_INVOCATIONS
,
1492 stgp
->Base
.info
.gs
.invocations
);
1494 st_translate_program_common(st
, &stgp
->Base
, stgp
->glsl_to_tgsi
, ureg
,
1495 PIPE_SHADER_GEOMETRY
, &stgp
->tgsi
);
1497 free_glsl_to_tgsi_visitor(stgp
->glsl_to_tgsi
);
1498 stgp
->glsl_to_tgsi
= NULL
;
1504 * Get/create a basic program variant.
1506 struct st_basic_variant
*
1507 st_get_basic_variant(struct st_context
*st
,
1508 unsigned pipe_shader
,
1509 struct st_common_program
*prog
)
1511 struct pipe_context
*pipe
= st
->pipe
;
1512 struct st_basic_variant
*v
;
1513 struct st_basic_variant_key key
;
1514 struct pipe_shader_state tgsi
= {0};
1515 memset(&key
, 0, sizeof(key
));
1516 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
1518 /* Search for existing variant */
1519 for (v
= prog
->variants
; v
; v
= v
->next
) {
1520 if (memcmp(&v
->key
, &key
, sizeof(key
)) == 0) {
1527 v
= CALLOC_STRUCT(st_basic_variant
);
1530 if (prog
->tgsi
.type
== PIPE_SHADER_IR_NIR
) {
1531 tgsi
.type
= PIPE_SHADER_IR_NIR
;
1532 tgsi
.ir
.nir
= nir_shader_clone(NULL
, prog
->tgsi
.ir
.nir
);
1533 st_finalize_nir(st
, &prog
->Base
, prog
->shader_program
,
1535 tgsi
.stream_output
= prog
->tgsi
.stream_output
;
1538 /* fill in new variant */
1539 switch (pipe_shader
) {
1540 case PIPE_SHADER_TESS_CTRL
:
1541 v
->driver_shader
= pipe
->create_tcs_state(pipe
, &tgsi
);
1543 case PIPE_SHADER_TESS_EVAL
:
1544 v
->driver_shader
= pipe
->create_tes_state(pipe
, &tgsi
);
1546 case PIPE_SHADER_GEOMETRY
:
1547 v
->driver_shader
= pipe
->create_gs_state(pipe
, &tgsi
);
1550 assert(!"unhandled shader type");
1557 /* insert into list */
1558 v
->next
= prog
->variants
;
1568 * Translate a tessellation control program to create a new variant.
1571 st_translate_tessctrl_program(struct st_context
*st
,
1572 struct st_common_program
*sttcp
)
1574 struct ureg_program
*ureg
;
1576 /* We have already compiled to NIR so just return */
1577 if (sttcp
->shader_program
) {
1578 st_store_ir_in_disk_cache(st
, &sttcp
->Base
, true);
1582 ureg
= ureg_create_with_screen(PIPE_SHADER_TESS_CTRL
, st
->pipe
->screen
);
1586 ureg_property(ureg
, TGSI_PROPERTY_TCS_VERTICES_OUT
,
1587 sttcp
->Base
.info
.tess
.tcs_vertices_out
);
1589 st_translate_program_common(st
, &sttcp
->Base
, sttcp
->glsl_to_tgsi
, ureg
,
1590 PIPE_SHADER_TESS_CTRL
, &sttcp
->tgsi
);
1592 free_glsl_to_tgsi_visitor(sttcp
->glsl_to_tgsi
);
1593 sttcp
->glsl_to_tgsi
= NULL
;
1599 * Translate a tessellation evaluation program to create a new variant.
1602 st_translate_tesseval_program(struct st_context
*st
,
1603 struct st_common_program
*sttep
)
1605 struct ureg_program
*ureg
;
1607 /* We have already compiled to NIR so just return */
1608 if (sttep
->shader_program
) {
1609 st_translate_program_stream_output(&sttep
->Base
, &sttep
->tgsi
.stream_output
);
1610 st_store_ir_in_disk_cache(st
, &sttep
->Base
, true);
1614 ureg
= ureg_create_with_screen(PIPE_SHADER_TESS_EVAL
, st
->pipe
->screen
);
1618 if (sttep
->Base
.info
.tess
.primitive_mode
== GL_ISOLINES
)
1619 ureg_property(ureg
, TGSI_PROPERTY_TES_PRIM_MODE
, GL_LINES
);
1621 ureg_property(ureg
, TGSI_PROPERTY_TES_PRIM_MODE
,
1622 sttep
->Base
.info
.tess
.primitive_mode
);
1624 STATIC_ASSERT((TESS_SPACING_EQUAL
+ 1) % 3 == PIPE_TESS_SPACING_EQUAL
);
1625 STATIC_ASSERT((TESS_SPACING_FRACTIONAL_ODD
+ 1) % 3 ==
1626 PIPE_TESS_SPACING_FRACTIONAL_ODD
);
1627 STATIC_ASSERT((TESS_SPACING_FRACTIONAL_EVEN
+ 1) % 3 ==
1628 PIPE_TESS_SPACING_FRACTIONAL_EVEN
);
1630 ureg_property(ureg
, TGSI_PROPERTY_TES_SPACING
,
1631 (sttep
->Base
.info
.tess
.spacing
+ 1) % 3);
1633 ureg_property(ureg
, TGSI_PROPERTY_TES_VERTEX_ORDER_CW
,
1634 !sttep
->Base
.info
.tess
.ccw
);
1635 ureg_property(ureg
, TGSI_PROPERTY_TES_POINT_MODE
,
1636 sttep
->Base
.info
.tess
.point_mode
);
1638 st_translate_program_common(st
, &sttep
->Base
, sttep
->glsl_to_tgsi
,
1639 ureg
, PIPE_SHADER_TESS_EVAL
, &sttep
->tgsi
);
1641 free_glsl_to_tgsi_visitor(sttep
->glsl_to_tgsi
);
1642 sttep
->glsl_to_tgsi
= NULL
;
1648 * Translate a compute program to create a new variant.
1651 st_translate_compute_program(struct st_context
*st
,
1652 struct st_compute_program
*stcp
)
1654 struct ureg_program
*ureg
;
1655 struct pipe_shader_state prog
;
1657 stcp
->tgsi
.req_local_mem
= stcp
->Base
.info
.cs
.shared_size
;
1659 if (stcp
->shader_program
) {
1660 /* no compute variants: */
1661 st_finalize_nir(st
, &stcp
->Base
, stcp
->shader_program
,
1662 (struct nir_shader
*) stcp
->tgsi
.prog
);
1663 st_store_ir_in_disk_cache(st
, &stcp
->Base
, true);
1667 ureg
= ureg_create_with_screen(PIPE_SHADER_COMPUTE
, st
->pipe
->screen
);
1671 st_translate_program_common(st
, &stcp
->Base
, stcp
->glsl_to_tgsi
, ureg
,
1672 PIPE_SHADER_COMPUTE
, &prog
);
1674 stcp
->tgsi
.ir_type
= PIPE_SHADER_IR_TGSI
;
1675 stcp
->tgsi
.req_private_mem
= 0;
1676 stcp
->tgsi
.req_input_mem
= 0;
1678 free_glsl_to_tgsi_visitor(stcp
->glsl_to_tgsi
);
1679 stcp
->glsl_to_tgsi
= NULL
;
1685 * Get/create compute program variant.
1687 struct st_basic_variant
*
1688 st_get_cp_variant(struct st_context
*st
,
1689 struct pipe_compute_state
*tgsi
,
1690 struct st_basic_variant
**variants
)
1692 struct pipe_context
*pipe
= st
->pipe
;
1693 struct st_basic_variant
*v
;
1694 struct st_basic_variant_key key
;
1696 memset(&key
, 0, sizeof(key
));
1697 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
1699 /* Search for existing variant */
1700 for (v
= *variants
; v
; v
= v
->next
) {
1701 if (memcmp(&v
->key
, &key
, sizeof(key
)) == 0) {
1708 v
= CALLOC_STRUCT(st_basic_variant
);
1710 /* fill in new variant */
1711 struct pipe_compute_state cs
= *tgsi
;
1712 if (tgsi
->ir_type
== PIPE_SHADER_IR_NIR
)
1713 cs
.prog
= nir_shader_clone(NULL
, tgsi
->prog
);
1714 v
->driver_shader
= pipe
->create_compute_state(pipe
, &cs
);
1717 /* insert into list */
1718 v
->next
= *variants
;
1728 * Vert/Geom/Frag programs have per-context variants. Free all the
1729 * variants attached to the given program which match the given context.
1732 destroy_program_variants(struct st_context
*st
, struct gl_program
*target
)
1734 if (!target
|| target
== &_mesa_DummyProgram
)
1737 switch (target
->Target
) {
1738 case GL_VERTEX_PROGRAM_ARB
:
1740 struct st_vertex_program
*stvp
= (struct st_vertex_program
*) target
;
1741 struct st_vp_variant
*vpv
, **prevPtr
= &stvp
->variants
;
1743 for (vpv
= stvp
->variants
; vpv
; ) {
1744 struct st_vp_variant
*next
= vpv
->next
;
1745 if (vpv
->key
.st
== st
) {
1746 /* unlink from list */
1748 /* destroy this variant */
1749 delete_vp_variant(st
, vpv
);
1752 prevPtr
= &vpv
->next
;
1758 case GL_FRAGMENT_PROGRAM_ARB
:
1760 struct st_fragment_program
*stfp
=
1761 (struct st_fragment_program
*) target
;
1762 struct st_fp_variant
*fpv
, **prevPtr
= &stfp
->variants
;
1764 for (fpv
= stfp
->variants
; fpv
; ) {
1765 struct st_fp_variant
*next
= fpv
->next
;
1766 if (fpv
->key
.st
== st
) {
1767 /* unlink from list */
1769 /* destroy this variant */
1770 delete_fp_variant(st
, fpv
);
1773 prevPtr
= &fpv
->next
;
1779 case GL_GEOMETRY_PROGRAM_NV
:
1780 case GL_TESS_CONTROL_PROGRAM_NV
:
1781 case GL_TESS_EVALUATION_PROGRAM_NV
:
1782 case GL_COMPUTE_PROGRAM_NV
:
1784 struct st_common_program
*p
= st_common_program(target
);
1785 struct st_compute_program
*cp
= (struct st_compute_program
*)target
;
1786 struct st_basic_variant
**variants
=
1787 target
->Target
== GL_COMPUTE_PROGRAM_NV
? &cp
->variants
:
1789 struct st_basic_variant
*v
, **prevPtr
= variants
;
1791 for (v
= *variants
; v
; ) {
1792 struct st_basic_variant
*next
= v
->next
;
1793 if (v
->key
.st
== st
) {
1794 /* unlink from list */
1796 /* destroy this variant */
1797 delete_basic_variant(st
, v
, target
->Target
);
1807 _mesa_problem(NULL
, "Unexpected program target 0x%x in "
1808 "destroy_program_variants_cb()", target
->Target
);
1814 * Callback for _mesa_HashWalk. Free all the shader's program variants
1815 * which match the given context.
1818 destroy_shader_program_variants_cb(GLuint key
, void *data
, void *userData
)
1820 struct st_context
*st
= (struct st_context
*) userData
;
1821 struct gl_shader
*shader
= (struct gl_shader
*) data
;
1823 switch (shader
->Type
) {
1824 case GL_SHADER_PROGRAM_MESA
:
1826 struct gl_shader_program
*shProg
= (struct gl_shader_program
*) data
;
1829 for (i
= 0; i
< ARRAY_SIZE(shProg
->_LinkedShaders
); i
++) {
1830 if (shProg
->_LinkedShaders
[i
])
1831 destroy_program_variants(st
, shProg
->_LinkedShaders
[i
]->Program
);
1835 case GL_VERTEX_SHADER
:
1836 case GL_FRAGMENT_SHADER
:
1837 case GL_GEOMETRY_SHADER
:
1838 case GL_TESS_CONTROL_SHADER
:
1839 case GL_TESS_EVALUATION_SHADER
:
1840 case GL_COMPUTE_SHADER
:
1849 * Callback for _mesa_HashWalk. Free all the program variants which match
1850 * the given context.
1853 destroy_program_variants_cb(GLuint key
, void *data
, void *userData
)
1855 struct st_context
*st
= (struct st_context
*) userData
;
1856 struct gl_program
*program
= (struct gl_program
*) data
;
1857 destroy_program_variants(st
, program
);
1862 * Walk over all shaders and programs to delete any variants which
1863 * belong to the given context.
1864 * This is called during context tear-down.
1867 st_destroy_program_variants(struct st_context
*st
)
1869 /* If shaders can be shared with other contexts, the last context will
1870 * call DeleteProgram on all shaders, releasing everything.
1872 if (st
->has_shareable_shaders
)
1875 /* ARB vert/frag program */
1876 _mesa_HashWalk(st
->ctx
->Shared
->Programs
,
1877 destroy_program_variants_cb
, st
);
1879 /* GLSL vert/frag/geom shaders */
1880 _mesa_HashWalk(st
->ctx
->Shared
->ShaderObjects
,
1881 destroy_shader_program_variants_cb
, st
);
1886 * For debugging, print/dump the current vertex program.
1889 st_print_current_vertex_program(void)
1891 GET_CURRENT_CONTEXT(ctx
);
1893 if (ctx
->VertexProgram
._Current
) {
1894 struct st_vertex_program
*stvp
=
1895 (struct st_vertex_program
*) ctx
->VertexProgram
._Current
;
1896 struct st_vp_variant
*stv
;
1898 debug_printf("Vertex program %u\n", stvp
->Base
.Id
);
1900 for (stv
= stvp
->variants
; stv
; stv
= stv
->next
) {
1901 debug_printf("variant %p\n", stv
);
1902 tgsi_dump(stv
->tgsi
.tokens
, 0);
1909 * Compile one shader variant.
1912 st_precompile_shader_variant(struct st_context
*st
,
1913 struct gl_program
*prog
)
1915 switch (prog
->Target
) {
1916 case GL_VERTEX_PROGRAM_ARB
: {
1917 struct st_vertex_program
*p
= (struct st_vertex_program
*)prog
;
1918 struct st_vp_variant_key key
;
1920 memset(&key
, 0, sizeof(key
));
1921 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
1922 st_get_vp_variant(st
, p
, &key
);
1926 case GL_TESS_CONTROL_PROGRAM_NV
: {
1927 struct st_common_program
*p
= st_common_program(prog
);
1928 st_get_basic_variant(st
, PIPE_SHADER_TESS_CTRL
, p
);
1932 case GL_TESS_EVALUATION_PROGRAM_NV
: {
1933 struct st_common_program
*p
= st_common_program(prog
);
1934 st_get_basic_variant(st
, PIPE_SHADER_TESS_EVAL
, p
);
1938 case GL_GEOMETRY_PROGRAM_NV
: {
1939 struct st_common_program
*p
= st_common_program(prog
);
1940 st_get_basic_variant(st
, PIPE_SHADER_GEOMETRY
, p
);
1944 case GL_FRAGMENT_PROGRAM_ARB
: {
1945 struct st_fragment_program
*p
= (struct st_fragment_program
*)prog
;
1946 struct st_fp_variant_key key
;
1948 memset(&key
, 0, sizeof(key
));
1949 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
1950 st_get_fp_variant(st
, p
, &key
);
1954 case GL_COMPUTE_PROGRAM_NV
: {
1955 struct st_compute_program
*p
= (struct st_compute_program
*)prog
;
1956 st_get_cp_variant(st
, &p
->tgsi
, &p
->variants
);