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 ureg_free_tokens(stcp
->tgsi
.prog
);
364 stcp
->tgsi
.prog
= NULL
;
369 * Translate a vertex program.
372 st_translate_vertex_program(struct st_context
*st
,
373 struct st_vertex_program
*stvp
)
375 struct ureg_program
*ureg
;
376 enum pipe_error error
;
377 unsigned num_outputs
= 0;
379 ubyte input_to_index
[VERT_ATTRIB_MAX
] = {0};
380 ubyte output_semantic_name
[VARYING_SLOT_MAX
] = {0};
381 ubyte output_semantic_index
[VARYING_SLOT_MAX
] = {0};
383 stvp
->num_inputs
= 0;
385 if (stvp
->Base
.arb
.IsPositionInvariant
)
386 _mesa_insert_mvp_code(st
->ctx
, &stvp
->Base
);
389 * Determine number of inputs, the mappings between VERT_ATTRIB_x
390 * and TGSI generic input indexes, plus input attrib semantic info.
392 for (attr
= 0; attr
< VERT_ATTRIB_MAX
; attr
++) {
393 if ((stvp
->Base
.info
.inputs_read
& BITFIELD64_BIT(attr
)) != 0) {
394 input_to_index
[attr
] = stvp
->num_inputs
;
395 stvp
->index_to_input
[stvp
->num_inputs
] = attr
;
397 if ((stvp
->Base
.info
.double_inputs_read
&
398 BITFIELD64_BIT(attr
)) != 0) {
399 /* add placeholder for second part of a double attribute */
400 stvp
->index_to_input
[stvp
->num_inputs
] = ST_DOUBLE_ATTRIB_PLACEHOLDER
;
405 /* bit of a hack, presetup potentially unused edgeflag input */
406 input_to_index
[VERT_ATTRIB_EDGEFLAG
] = stvp
->num_inputs
;
407 stvp
->index_to_input
[stvp
->num_inputs
] = VERT_ATTRIB_EDGEFLAG
;
409 /* Compute mapping of vertex program outputs to slots.
411 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
412 if ((stvp
->Base
.info
.outputs_written
& BITFIELD64_BIT(attr
)) == 0) {
413 stvp
->result_to_output
[attr
] = ~0;
416 unsigned slot
= num_outputs
++;
418 stvp
->result_to_output
[attr
] = slot
;
420 unsigned semantic_name
, semantic_index
;
421 tgsi_get_gl_varying_semantic(attr
, st
->needs_texcoord_semantic
,
422 &semantic_name
, &semantic_index
);
423 output_semantic_name
[slot
] = semantic_name
;
424 output_semantic_index
[slot
] = semantic_index
;
427 /* similar hack to above, presetup potentially unused edgeflag output */
428 stvp
->result_to_output
[VARYING_SLOT_EDGE
] = num_outputs
;
429 output_semantic_name
[num_outputs
] = TGSI_SEMANTIC_EDGEFLAG
;
430 output_semantic_index
[num_outputs
] = 0;
433 if (!stvp
->glsl_to_tgsi
&& !stvp
->shader_program
) {
434 _mesa_remove_output_reads(&stvp
->Base
, PROGRAM_OUTPUT
);
436 /* This determines which states will be updated when the assembly
439 stvp
->affected_states
= ST_NEW_VS_STATE
|
441 ST_NEW_VERTEX_ARRAYS
;
443 if (stvp
->Base
.Parameters
->NumParameters
)
444 stvp
->affected_states
|= ST_NEW_VS_CONSTANTS
;
446 /* No samplers are allowed in ARB_vp. */
449 if (stvp
->shader_program
) {
450 nir_shader
*nir
= st_glsl_to_nir(st
, &stvp
->Base
, stvp
->shader_program
,
453 stvp
->tgsi
.type
= PIPE_SHADER_IR_NIR
;
454 stvp
->tgsi
.ir
.nir
= nir
;
456 struct gl_program
*prog
= stvp
->shader_program
->last_vert_prog
;
458 st_translate_stream_output_info2(prog
->sh
.LinkedTransformFeedback
,
459 stvp
->result_to_output
,
460 &stvp
->tgsi
.stream_output
);
466 ureg
= ureg_create_with_screen(PIPE_SHADER_VERTEX
, st
->pipe
->screen
);
470 if (stvp
->Base
.info
.clip_distance_array_size
)
471 ureg_property(ureg
, TGSI_PROPERTY_NUM_CLIPDIST_ENABLED
,
472 stvp
->Base
.info
.clip_distance_array_size
);
473 if (stvp
->Base
.info
.cull_distance_array_size
)
474 ureg_property(ureg
, TGSI_PROPERTY_NUM_CULLDIST_ENABLED
,
475 stvp
->Base
.info
.cull_distance_array_size
);
477 if (ST_DEBUG
& DEBUG_MESA
) {
478 _mesa_print_program(&stvp
->Base
);
479 _mesa_print_program_parameters(st
->ctx
, &stvp
->Base
);
483 if (stvp
->glsl_to_tgsi
) {
484 error
= st_translate_program(st
->ctx
,
492 NULL
, /* inputSlotToAttr */
493 NULL
, /* input semantic name */
494 NULL
, /* input semantic index */
495 NULL
, /* interp mode */
498 stvp
->result_to_output
,
499 output_semantic_name
,
500 output_semantic_index
);
502 st_translate_stream_output_info(stvp
->glsl_to_tgsi
,
503 stvp
->result_to_output
,
504 &stvp
->tgsi
.stream_output
);
506 free_glsl_to_tgsi_visitor(stvp
->glsl_to_tgsi
);
508 error
= st_translate_mesa_program(st
->ctx
,
515 NULL
, /* input semantic name */
516 NULL
, /* input semantic index */
520 stvp
->result_to_output
,
521 output_semantic_name
,
522 output_semantic_index
);
525 debug_printf("%s: failed to translate Mesa program:\n", __func__
);
526 _mesa_print_program(&stvp
->Base
);
532 stvp
->tgsi
.tokens
= ureg_get_tokens(ureg
, &num_tokens
);
535 if (stvp
->glsl_to_tgsi
) {
536 stvp
->glsl_to_tgsi
= NULL
;
537 st_store_tgsi_in_disk_cache(st
, &stvp
->Base
, NULL
, num_tokens
);
540 return stvp
->tgsi
.tokens
!= NULL
;
543 static struct st_vp_variant
*
544 st_create_vp_variant(struct st_context
*st
,
545 struct st_vertex_program
*stvp
,
546 const struct st_vp_variant_key
*key
)
548 struct st_vp_variant
*vpv
= CALLOC_STRUCT(st_vp_variant
);
549 struct pipe_context
*pipe
= st
->pipe
;
552 vpv
->tgsi
.stream_output
= stvp
->tgsi
.stream_output
;
553 vpv
->num_inputs
= stvp
->num_inputs
;
555 if (stvp
->tgsi
.type
== PIPE_SHADER_IR_NIR
) {
556 vpv
->tgsi
.type
= PIPE_SHADER_IR_NIR
;
557 vpv
->tgsi
.ir
.nir
= nir_shader_clone(NULL
, stvp
->tgsi
.ir
.nir
);
558 if (key
->clamp_color
)
559 NIR_PASS_V(vpv
->tgsi
.ir
.nir
, nir_lower_clamp_color_outputs
);
560 if (key
->passthrough_edgeflags
) {
561 NIR_PASS_V(vpv
->tgsi
.ir
.nir
, nir_lower_passthrough_edgeflags
);
565 st_finalize_nir(st
, &stvp
->Base
, vpv
->tgsi
.ir
.nir
);
567 vpv
->driver_shader
= pipe
->create_vs_state(pipe
, &vpv
->tgsi
);
568 /* driver takes ownership of IR: */
569 vpv
->tgsi
.ir
.nir
= NULL
;
573 vpv
->tgsi
.tokens
= tgsi_dup_tokens(stvp
->tgsi
.tokens
);
575 /* Emulate features. */
576 if (key
->clamp_color
|| key
->passthrough_edgeflags
) {
577 const struct tgsi_token
*tokens
;
579 (key
->clamp_color
? TGSI_EMU_CLAMP_COLOR_OUTPUTS
: 0) |
580 (key
->passthrough_edgeflags
? TGSI_EMU_PASSTHROUGH_EDGEFLAG
: 0);
582 tokens
= tgsi_emulate(vpv
->tgsi
.tokens
, flags
);
585 tgsi_free_tokens(vpv
->tgsi
.tokens
);
586 vpv
->tgsi
.tokens
= tokens
;
588 if (key
->passthrough_edgeflags
)
591 fprintf(stderr
, "mesa: cannot emulate deprecated features\n");
594 if (ST_DEBUG
& DEBUG_TGSI
) {
595 tgsi_dump(vpv
->tgsi
.tokens
, 0);
599 vpv
->driver_shader
= pipe
->create_vs_state(pipe
, &vpv
->tgsi
);
605 * Find/create a vertex program variant.
607 struct st_vp_variant
*
608 st_get_vp_variant(struct st_context
*st
,
609 struct st_vertex_program
*stvp
,
610 const struct st_vp_variant_key
*key
)
612 struct st_vp_variant
*vpv
;
614 /* Search for existing variant */
615 for (vpv
= stvp
->variants
; vpv
; vpv
= vpv
->next
) {
616 if (memcmp(&vpv
->key
, key
, sizeof(*key
)) == 0) {
623 vpv
= st_create_vp_variant(st
, stvp
, key
);
625 /* insert into list */
626 vpv
->next
= stvp
->variants
;
627 stvp
->variants
= vpv
;
636 * Translate a Mesa fragment shader into a TGSI shader.
639 st_translate_fragment_program(struct st_context
*st
,
640 struct st_fragment_program
*stfp
)
642 ubyte outputMapping
[2 * FRAG_RESULT_MAX
];
643 ubyte inputMapping
[VARYING_SLOT_MAX
];
644 ubyte inputSlotToAttr
[VARYING_SLOT_MAX
];
645 ubyte interpMode
[PIPE_MAX_SHADER_INPUTS
]; /* XXX size? */
647 GLbitfield64 inputsRead
;
648 struct ureg_program
*ureg
;
650 GLboolean write_all
= GL_FALSE
;
652 ubyte input_semantic_name
[PIPE_MAX_SHADER_INPUTS
];
653 ubyte input_semantic_index
[PIPE_MAX_SHADER_INPUTS
];
654 uint fs_num_inputs
= 0;
656 ubyte fs_output_semantic_name
[PIPE_MAX_SHADER_OUTPUTS
];
657 ubyte fs_output_semantic_index
[PIPE_MAX_SHADER_OUTPUTS
];
658 uint fs_num_outputs
= 0;
660 memset(inputSlotToAttr
, ~0, sizeof(inputSlotToAttr
));
662 /* Non-GLSL programs: */
663 if (!stfp
->glsl_to_tgsi
&& !stfp
->shader_program
) {
664 _mesa_remove_output_reads(&stfp
->Base
, PROGRAM_OUTPUT
);
665 if (st
->ctx
->Const
.GLSLFragCoordIsSysVal
)
666 _mesa_program_fragment_position_to_sysval(&stfp
->Base
);
668 /* This determines which states will be updated when the assembly
671 * fragment.position and glDrawPixels always use constants.
673 stfp
->affected_states
= ST_NEW_FS_STATE
|
674 ST_NEW_SAMPLE_SHADING
|
678 /* Just set them for ATI_fs unconditionally. */
679 stfp
->affected_states
|= ST_NEW_FS_SAMPLER_VIEWS
|
683 if (stfp
->Base
.SamplersUsed
)
684 stfp
->affected_states
|= ST_NEW_FS_SAMPLER_VIEWS
|
690 * Convert Mesa program inputs to TGSI input register semantics.
692 inputsRead
= stfp
->Base
.info
.inputs_read
;
693 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
694 if ((inputsRead
& BITFIELD64_BIT(attr
)) != 0) {
695 const GLuint slot
= fs_num_inputs
++;
697 inputMapping
[attr
] = slot
;
698 inputSlotToAttr
[slot
] = attr
;
701 case VARYING_SLOT_POS
:
702 input_semantic_name
[slot
] = TGSI_SEMANTIC_POSITION
;
703 input_semantic_index
[slot
] = 0;
704 interpMode
[slot
] = TGSI_INTERPOLATE_LINEAR
;
706 case VARYING_SLOT_COL0
:
707 input_semantic_name
[slot
] = TGSI_SEMANTIC_COLOR
;
708 input_semantic_index
[slot
] = 0;
709 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
710 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_COLOR
;
712 case VARYING_SLOT_COL1
:
713 input_semantic_name
[slot
] = TGSI_SEMANTIC_COLOR
;
714 input_semantic_index
[slot
] = 1;
715 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
716 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_COLOR
;
718 case VARYING_SLOT_FOGC
:
719 input_semantic_name
[slot
] = TGSI_SEMANTIC_FOG
;
720 input_semantic_index
[slot
] = 0;
721 interpMode
[slot
] = TGSI_INTERPOLATE_PERSPECTIVE
;
723 case VARYING_SLOT_FACE
:
724 input_semantic_name
[slot
] = TGSI_SEMANTIC_FACE
;
725 input_semantic_index
[slot
] = 0;
726 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
728 case VARYING_SLOT_PRIMITIVE_ID
:
729 input_semantic_name
[slot
] = TGSI_SEMANTIC_PRIMID
;
730 input_semantic_index
[slot
] = 0;
731 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
733 case VARYING_SLOT_LAYER
:
734 input_semantic_name
[slot
] = TGSI_SEMANTIC_LAYER
;
735 input_semantic_index
[slot
] = 0;
736 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
738 case VARYING_SLOT_VIEWPORT
:
739 input_semantic_name
[slot
] = TGSI_SEMANTIC_VIEWPORT_INDEX
;
740 input_semantic_index
[slot
] = 0;
741 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
743 case VARYING_SLOT_CLIP_DIST0
:
744 input_semantic_name
[slot
] = TGSI_SEMANTIC_CLIPDIST
;
745 input_semantic_index
[slot
] = 0;
746 interpMode
[slot
] = TGSI_INTERPOLATE_PERSPECTIVE
;
748 case VARYING_SLOT_CLIP_DIST1
:
749 input_semantic_name
[slot
] = TGSI_SEMANTIC_CLIPDIST
;
750 input_semantic_index
[slot
] = 1;
751 interpMode
[slot
] = TGSI_INTERPOLATE_PERSPECTIVE
;
753 case VARYING_SLOT_CULL_DIST0
:
754 case VARYING_SLOT_CULL_DIST1
:
755 /* these should have been lowered by GLSL */
758 /* In most cases, there is nothing special about these
759 * inputs, so adopt a convention to use the generic
760 * semantic name and the mesa VARYING_SLOT_ number as the
763 * All that is required is that the vertex shader labels
764 * its own outputs similarly, and that the vertex shader
765 * generates at least every output required by the
766 * fragment shader plus fixed-function hardware (such as
769 * However, some drivers may need us to identify the PNTC and TEXi
770 * varyings if, for example, their capability to replace them with
771 * sprite coordinates is limited.
773 case VARYING_SLOT_PNTC
:
774 if (st
->needs_texcoord_semantic
) {
775 input_semantic_name
[slot
] = TGSI_SEMANTIC_PCOORD
;
776 input_semantic_index
[slot
] = 0;
777 interpMode
[slot
] = TGSI_INTERPOLATE_LINEAR
;
781 case VARYING_SLOT_TEX0
:
782 case VARYING_SLOT_TEX1
:
783 case VARYING_SLOT_TEX2
:
784 case VARYING_SLOT_TEX3
:
785 case VARYING_SLOT_TEX4
:
786 case VARYING_SLOT_TEX5
:
787 case VARYING_SLOT_TEX6
:
788 case VARYING_SLOT_TEX7
:
789 if (st
->needs_texcoord_semantic
) {
790 input_semantic_name
[slot
] = TGSI_SEMANTIC_TEXCOORD
;
791 input_semantic_index
[slot
] = attr
- VARYING_SLOT_TEX0
;
792 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
793 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_PERSPECTIVE
;
797 case VARYING_SLOT_VAR0
:
799 /* Semantic indices should be zero-based because drivers may choose
800 * to assign a fixed slot determined by that index.
801 * This is useful because ARB_separate_shader_objects uses location
802 * qualifiers for linkage, and if the semantic index corresponds to
803 * these locations, linkage passes in the driver become unecessary.
805 * If needs_texcoord_semantic is true, no semantic indices will be
806 * consumed for the TEXi varyings, and we can base the locations of
807 * the user varyings on VAR0. Otherwise, we use TEX0 as base index.
809 assert(attr
>= VARYING_SLOT_VAR0
|| attr
== VARYING_SLOT_PNTC
||
810 (attr
>= VARYING_SLOT_TEX0
&& attr
<= VARYING_SLOT_TEX7
));
811 input_semantic_name
[slot
] = TGSI_SEMANTIC_GENERIC
;
812 input_semantic_index
[slot
] = st_get_generic_varying_index(st
, attr
);
813 if (attr
== VARYING_SLOT_PNTC
)
814 interpMode
[slot
] = TGSI_INTERPOLATE_LINEAR
;
816 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
817 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_PERSPECTIVE
;
823 inputMapping
[attr
] = -1;
828 * Semantics and mapping for outputs
830 GLbitfield64 outputsWritten
= stfp
->Base
.info
.outputs_written
;
832 /* if z is written, emit that first */
833 if (outputsWritten
& BITFIELD64_BIT(FRAG_RESULT_DEPTH
)) {
834 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_POSITION
;
835 fs_output_semantic_index
[fs_num_outputs
] = 0;
836 outputMapping
[FRAG_RESULT_DEPTH
] = fs_num_outputs
;
838 outputsWritten
&= ~(1 << FRAG_RESULT_DEPTH
);
841 if (outputsWritten
& BITFIELD64_BIT(FRAG_RESULT_STENCIL
)) {
842 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_STENCIL
;
843 fs_output_semantic_index
[fs_num_outputs
] = 0;
844 outputMapping
[FRAG_RESULT_STENCIL
] = fs_num_outputs
;
846 outputsWritten
&= ~(1 << FRAG_RESULT_STENCIL
);
849 if (outputsWritten
& BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK
)) {
850 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_SAMPLEMASK
;
851 fs_output_semantic_index
[fs_num_outputs
] = 0;
852 outputMapping
[FRAG_RESULT_SAMPLE_MASK
] = fs_num_outputs
;
854 outputsWritten
&= ~(1 << FRAG_RESULT_SAMPLE_MASK
);
857 /* handle remaining outputs (color) */
858 for (attr
= 0; attr
< ARRAY_SIZE(outputMapping
); attr
++) {
859 const GLbitfield64 written
= attr
< FRAG_RESULT_MAX
? outputsWritten
:
860 stfp
->Base
.SecondaryOutputsWritten
;
861 const unsigned loc
= attr
% FRAG_RESULT_MAX
;
863 if (written
& BITFIELD64_BIT(loc
)) {
865 case FRAG_RESULT_DEPTH
:
866 case FRAG_RESULT_STENCIL
:
867 case FRAG_RESULT_SAMPLE_MASK
:
871 case FRAG_RESULT_COLOR
:
872 write_all
= GL_TRUE
; /* fallthrough */
875 assert(loc
== FRAG_RESULT_COLOR
||
876 (FRAG_RESULT_DATA0
<= loc
&& loc
< FRAG_RESULT_MAX
));
878 index
= (loc
== FRAG_RESULT_COLOR
) ? 0 : (loc
- FRAG_RESULT_DATA0
);
880 if (attr
>= FRAG_RESULT_MAX
) {
881 /* Secondary color for dual source blending. */
886 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_COLOR
;
887 fs_output_semantic_index
[fs_num_outputs
] = index
;
888 outputMapping
[attr
] = fs_num_outputs
;
897 if (stfp
->shader_program
) {
898 nir_shader
*nir
= st_glsl_to_nir(st
, &stfp
->Base
, stfp
->shader_program
,
899 MESA_SHADER_FRAGMENT
);
901 stfp
->tgsi
.type
= PIPE_SHADER_IR_NIR
;
902 stfp
->tgsi
.ir
.nir
= nir
;
907 ureg
= ureg_create_with_screen(PIPE_SHADER_FRAGMENT
, st
->pipe
->screen
);
911 if (ST_DEBUG
& DEBUG_MESA
) {
912 _mesa_print_program(&stfp
->Base
);
913 _mesa_print_program_parameters(st
->ctx
, &stfp
->Base
);
916 if (write_all
== GL_TRUE
)
917 ureg_property(ureg
, TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS
, 1);
919 if (stfp
->Base
.info
.fs
.depth_layout
!= FRAG_DEPTH_LAYOUT_NONE
) {
920 switch (stfp
->Base
.info
.fs
.depth_layout
) {
921 case FRAG_DEPTH_LAYOUT_ANY
:
922 ureg_property(ureg
, TGSI_PROPERTY_FS_DEPTH_LAYOUT
,
923 TGSI_FS_DEPTH_LAYOUT_ANY
);
925 case FRAG_DEPTH_LAYOUT_GREATER
:
926 ureg_property(ureg
, TGSI_PROPERTY_FS_DEPTH_LAYOUT
,
927 TGSI_FS_DEPTH_LAYOUT_GREATER
);
929 case FRAG_DEPTH_LAYOUT_LESS
:
930 ureg_property(ureg
, TGSI_PROPERTY_FS_DEPTH_LAYOUT
,
931 TGSI_FS_DEPTH_LAYOUT_LESS
);
933 case FRAG_DEPTH_LAYOUT_UNCHANGED
:
934 ureg_property(ureg
, TGSI_PROPERTY_FS_DEPTH_LAYOUT
,
935 TGSI_FS_DEPTH_LAYOUT_UNCHANGED
);
942 if (stfp
->glsl_to_tgsi
) {
943 st_translate_program(st
->ctx
,
944 PIPE_SHADER_FRAGMENT
,
953 input_semantic_index
,
958 fs_output_semantic_name
,
959 fs_output_semantic_index
);
961 free_glsl_to_tgsi_visitor(stfp
->glsl_to_tgsi
);
962 } else if (stfp
->ati_fs
)
963 st_translate_atifs_program(ureg
,
970 input_semantic_index
,
975 fs_output_semantic_name
,
976 fs_output_semantic_index
);
978 st_translate_mesa_program(st
->ctx
,
979 PIPE_SHADER_FRAGMENT
,
986 input_semantic_index
,
991 fs_output_semantic_name
,
992 fs_output_semantic_index
);
995 stfp
->tgsi
.tokens
= ureg_get_tokens(ureg
, &num_tokens
);
998 if (stfp
->glsl_to_tgsi
) {
999 stfp
->glsl_to_tgsi
= NULL
;
1000 st_store_tgsi_in_disk_cache(st
, &stfp
->Base
, NULL
, num_tokens
);
1003 return stfp
->tgsi
.tokens
!= NULL
;
1006 static struct st_fp_variant
*
1007 st_create_fp_variant(struct st_context
*st
,
1008 struct st_fragment_program
*stfp
,
1009 const struct st_fp_variant_key
*key
)
1011 struct pipe_context
*pipe
= st
->pipe
;
1012 struct st_fp_variant
*variant
= CALLOC_STRUCT(st_fp_variant
);
1013 struct pipe_shader_state tgsi
= {0};
1014 struct gl_program_parameter_list
*params
= stfp
->Base
.Parameters
;
1015 static const gl_state_index texcoord_state
[STATE_LENGTH
] =
1016 { STATE_INTERNAL
, STATE_CURRENT_ATTRIB
, VERT_ATTRIB_TEX0
};
1017 static const gl_state_index scale_state
[STATE_LENGTH
] =
1018 { STATE_INTERNAL
, STATE_PT_SCALE
};
1019 static const gl_state_index bias_state
[STATE_LENGTH
] =
1020 { STATE_INTERNAL
, STATE_PT_BIAS
};
1025 if (stfp
->tgsi
.type
== PIPE_SHADER_IR_NIR
) {
1026 tgsi
.type
= PIPE_SHADER_IR_NIR
;
1027 tgsi
.ir
.nir
= nir_shader_clone(NULL
, stfp
->tgsi
.ir
.nir
);
1029 if (key
->clamp_color
)
1030 NIR_PASS_V(tgsi
.ir
.nir
, nir_lower_clamp_color_outputs
);
1032 if (key
->persample_shading
) {
1033 nir_shader
*shader
= tgsi
.ir
.nir
;
1034 nir_foreach_variable(var
, &shader
->inputs
)
1035 var
->data
.sample
= true;
1038 assert(!(key
->bitmap
&& key
->drawpixels
));
1042 nir_lower_bitmap_options options
= {0};
1044 variant
->bitmap_sampler
= ffs(~stfp
->Base
.SamplersUsed
) - 1;
1045 options
.sampler
= variant
->bitmap_sampler
;
1046 options
.swizzle_xxxx
= (st
->bitmap
.tex_format
== PIPE_FORMAT_L8_UNORM
);
1048 NIR_PASS_V(tgsi
.ir
.nir
, nir_lower_bitmap
, &options
);
1051 /* glDrawPixels (color only) */
1052 if (key
->drawpixels
) {
1053 nir_lower_drawpixels_options options
= {{0}};
1054 unsigned samplers_used
= stfp
->Base
.SamplersUsed
;
1056 /* Find the first unused slot. */
1057 variant
->drawpix_sampler
= ffs(~samplers_used
) - 1;
1058 options
.drawpix_sampler
= variant
->drawpix_sampler
;
1059 samplers_used
|= (1 << variant
->drawpix_sampler
);
1061 options
.pixel_maps
= key
->pixelMaps
;
1062 if (key
->pixelMaps
) {
1063 variant
->pixelmap_sampler
= ffs(~samplers_used
) - 1;
1064 options
.pixelmap_sampler
= variant
->pixelmap_sampler
;
1067 options
.scale_and_bias
= key
->scaleAndBias
;
1068 if (key
->scaleAndBias
) {
1069 _mesa_add_state_reference(params
, scale_state
);
1070 memcpy(options
.scale_state_tokens
, scale_state
,
1071 sizeof(options
.scale_state_tokens
));
1072 _mesa_add_state_reference(params
, bias_state
);
1073 memcpy(options
.bias_state_tokens
, bias_state
,
1074 sizeof(options
.bias_state_tokens
));
1077 _mesa_add_state_reference(params
, texcoord_state
);
1078 memcpy(options
.texcoord_state_tokens
, texcoord_state
,
1079 sizeof(options
.texcoord_state_tokens
));
1081 NIR_PASS_V(tgsi
.ir
.nir
, nir_lower_drawpixels
, &options
);
1084 if (unlikely(key
->external
.lower_nv12
|| key
->external
.lower_iyuv
)) {
1085 nir_lower_tex_options options
= {0};
1086 options
.lower_y_uv_external
= key
->external
.lower_nv12
;
1087 options
.lower_y_u_v_external
= key
->external
.lower_iyuv
;
1088 NIR_PASS_V(tgsi
.ir
.nir
, nir_lower_tex
, &options
);
1091 st_finalize_nir(st
, &stfp
->Base
, tgsi
.ir
.nir
);
1093 if (unlikely(key
->external
.lower_nv12
|| key
->external
.lower_iyuv
)) {
1094 /* This pass needs to happen *after* nir_lower_sampler */
1095 NIR_PASS_V(tgsi
.ir
.nir
, st_nir_lower_tex_src_plane
,
1096 ~stfp
->Base
.SamplersUsed
,
1097 key
->external
.lower_nv12
,
1098 key
->external
.lower_iyuv
);
1101 variant
->driver_shader
= pipe
->create_fs_state(pipe
, &tgsi
);
1102 variant
->key
= *key
;
1107 tgsi
.tokens
= stfp
->tgsi
.tokens
;
1109 assert(!(key
->bitmap
&& key
->drawpixels
));
1111 /* Fix texture targets and add fog for ATI_fs */
1113 const struct tgsi_token
*tokens
= st_fixup_atifs(tgsi
.tokens
, key
);
1116 tgsi
.tokens
= tokens
;
1118 fprintf(stderr
, "mesa: cannot post-process ATI_fs\n");
1121 /* Emulate features. */
1122 if (key
->clamp_color
|| key
->persample_shading
) {
1123 const struct tgsi_token
*tokens
;
1125 (key
->clamp_color
? TGSI_EMU_CLAMP_COLOR_OUTPUTS
: 0) |
1126 (key
->persample_shading
? TGSI_EMU_FORCE_PERSAMPLE_INTERP
: 0);
1128 tokens
= tgsi_emulate(tgsi
.tokens
, flags
);
1131 if (tgsi
.tokens
!= stfp
->tgsi
.tokens
)
1132 tgsi_free_tokens(tgsi
.tokens
);
1133 tgsi
.tokens
= tokens
;
1135 fprintf(stderr
, "mesa: cannot emulate deprecated features\n");
1140 const struct tgsi_token
*tokens
;
1142 variant
->bitmap_sampler
= ffs(~stfp
->Base
.SamplersUsed
) - 1;
1144 tokens
= st_get_bitmap_shader(tgsi
.tokens
,
1145 st
->internal_target
,
1146 variant
->bitmap_sampler
,
1147 st
->needs_texcoord_semantic
,
1148 st
->bitmap
.tex_format
==
1149 PIPE_FORMAT_L8_UNORM
);
1152 if (tgsi
.tokens
!= stfp
->tgsi
.tokens
)
1153 tgsi_free_tokens(tgsi
.tokens
);
1154 tgsi
.tokens
= tokens
;
1156 fprintf(stderr
, "mesa: cannot create a shader for glBitmap\n");
1159 /* glDrawPixels (color only) */
1160 if (key
->drawpixels
) {
1161 const struct tgsi_token
*tokens
;
1162 unsigned scale_const
= 0, bias_const
= 0, texcoord_const
= 0;
1164 /* Find the first unused slot. */
1165 variant
->drawpix_sampler
= ffs(~stfp
->Base
.SamplersUsed
) - 1;
1167 if (key
->pixelMaps
) {
1168 unsigned samplers_used
= stfp
->Base
.SamplersUsed
|
1169 (1 << variant
->drawpix_sampler
);
1171 variant
->pixelmap_sampler
= ffs(~samplers_used
) - 1;
1174 if (key
->scaleAndBias
) {
1175 scale_const
= _mesa_add_state_reference(params
, scale_state
);
1176 bias_const
= _mesa_add_state_reference(params
, bias_state
);
1179 texcoord_const
= _mesa_add_state_reference(params
, texcoord_state
);
1181 tokens
= st_get_drawpix_shader(tgsi
.tokens
,
1182 st
->needs_texcoord_semantic
,
1183 key
->scaleAndBias
, scale_const
,
1184 bias_const
, key
->pixelMaps
,
1185 variant
->drawpix_sampler
,
1186 variant
->pixelmap_sampler
,
1187 texcoord_const
, st
->internal_target
);
1190 if (tgsi
.tokens
!= stfp
->tgsi
.tokens
)
1191 tgsi_free_tokens(tgsi
.tokens
);
1192 tgsi
.tokens
= tokens
;
1194 fprintf(stderr
, "mesa: cannot create a shader for glDrawPixels\n");
1197 if (unlikely(key
->external
.lower_nv12
|| key
->external
.lower_iyuv
)) {
1198 const struct tgsi_token
*tokens
;
1200 /* samplers inserted would conflict, but this should be unpossible: */
1201 assert(!(key
->bitmap
|| key
->drawpixels
));
1203 tokens
= st_tgsi_lower_yuv(tgsi
.tokens
,
1204 ~stfp
->Base
.SamplersUsed
,
1205 key
->external
.lower_nv12
,
1206 key
->external
.lower_iyuv
);
1208 if (tgsi
.tokens
!= stfp
->tgsi
.tokens
)
1209 tgsi_free_tokens(tgsi
.tokens
);
1210 tgsi
.tokens
= tokens
;
1212 fprintf(stderr
, "mesa: cannot create a shader for samplerExternalOES\n");
1216 if (ST_DEBUG
& DEBUG_TGSI
) {
1217 tgsi_dump(tgsi
.tokens
, 0);
1221 /* fill in variant */
1222 variant
->driver_shader
= pipe
->create_fs_state(pipe
, &tgsi
);
1223 variant
->key
= *key
;
1225 if (tgsi
.tokens
!= stfp
->tgsi
.tokens
)
1226 tgsi_free_tokens(tgsi
.tokens
);
1231 * Translate fragment program if needed.
1233 struct st_fp_variant
*
1234 st_get_fp_variant(struct st_context
*st
,
1235 struct st_fragment_program
*stfp
,
1236 const struct st_fp_variant_key
*key
)
1238 struct st_fp_variant
*fpv
;
1240 /* Search for existing variant */
1241 for (fpv
= stfp
->variants
; fpv
; fpv
= fpv
->next
) {
1242 if (memcmp(&fpv
->key
, key
, sizeof(*key
)) == 0) {
1249 fpv
= st_create_fp_variant(st
, stfp
, key
);
1251 /* insert into list */
1252 fpv
->next
= stfp
->variants
;
1253 stfp
->variants
= fpv
;
1262 * Translate a program. This is common code for geometry and tessellation
1266 st_translate_program_common(struct st_context
*st
,
1267 struct gl_program
*prog
,
1268 struct glsl_to_tgsi_visitor
*glsl_to_tgsi
,
1269 struct ureg_program
*ureg
,
1270 unsigned tgsi_processor
,
1271 struct pipe_shader_state
*out_state
)
1273 ubyte inputSlotToAttr
[VARYING_SLOT_TESS_MAX
];
1274 ubyte inputMapping
[VARYING_SLOT_TESS_MAX
];
1275 ubyte outputMapping
[VARYING_SLOT_TESS_MAX
];
1278 ubyte input_semantic_name
[PIPE_MAX_SHADER_INPUTS
];
1279 ubyte input_semantic_index
[PIPE_MAX_SHADER_INPUTS
];
1280 uint num_inputs
= 0;
1282 ubyte output_semantic_name
[PIPE_MAX_SHADER_OUTPUTS
];
1283 ubyte output_semantic_index
[PIPE_MAX_SHADER_OUTPUTS
];
1284 uint num_outputs
= 0;
1288 memset(inputSlotToAttr
, 0, sizeof(inputSlotToAttr
));
1289 memset(inputMapping
, 0, sizeof(inputMapping
));
1290 memset(outputMapping
, 0, sizeof(outputMapping
));
1291 memset(out_state
, 0, sizeof(*out_state
));
1293 if (prog
->info
.clip_distance_array_size
)
1294 ureg_property(ureg
, TGSI_PROPERTY_NUM_CLIPDIST_ENABLED
,
1295 prog
->info
.clip_distance_array_size
);
1296 if (prog
->info
.cull_distance_array_size
)
1297 ureg_property(ureg
, TGSI_PROPERTY_NUM_CULLDIST_ENABLED
,
1298 prog
->info
.cull_distance_array_size
);
1301 * Convert Mesa program inputs to TGSI input register semantics.
1303 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
1304 if ((prog
->info
.inputs_read
& BITFIELD64_BIT(attr
)) != 0) {
1305 const GLuint slot
= num_inputs
++;
1307 inputMapping
[attr
] = slot
;
1308 inputSlotToAttr
[slot
] = attr
;
1311 case VARYING_SLOT_PRIMITIVE_ID
:
1312 assert(tgsi_processor
== PIPE_SHADER_GEOMETRY
);
1313 input_semantic_name
[slot
] = TGSI_SEMANTIC_PRIMID
;
1314 input_semantic_index
[slot
] = 0;
1316 case VARYING_SLOT_POS
:
1317 input_semantic_name
[slot
] = TGSI_SEMANTIC_POSITION
;
1318 input_semantic_index
[slot
] = 0;
1320 case VARYING_SLOT_COL0
:
1321 input_semantic_name
[slot
] = TGSI_SEMANTIC_COLOR
;
1322 input_semantic_index
[slot
] = 0;
1324 case VARYING_SLOT_COL1
:
1325 input_semantic_name
[slot
] = TGSI_SEMANTIC_COLOR
;
1326 input_semantic_index
[slot
] = 1;
1328 case VARYING_SLOT_FOGC
:
1329 input_semantic_name
[slot
] = TGSI_SEMANTIC_FOG
;
1330 input_semantic_index
[slot
] = 0;
1332 case VARYING_SLOT_CLIP_VERTEX
:
1333 input_semantic_name
[slot
] = TGSI_SEMANTIC_CLIPVERTEX
;
1334 input_semantic_index
[slot
] = 0;
1336 case VARYING_SLOT_CLIP_DIST0
:
1337 input_semantic_name
[slot
] = TGSI_SEMANTIC_CLIPDIST
;
1338 input_semantic_index
[slot
] = 0;
1340 case VARYING_SLOT_CLIP_DIST1
:
1341 input_semantic_name
[slot
] = TGSI_SEMANTIC_CLIPDIST
;
1342 input_semantic_index
[slot
] = 1;
1344 case VARYING_SLOT_CULL_DIST0
:
1345 case VARYING_SLOT_CULL_DIST1
:
1346 /* these should have been lowered by GLSL */
1349 case VARYING_SLOT_PSIZ
:
1350 input_semantic_name
[slot
] = TGSI_SEMANTIC_PSIZE
;
1351 input_semantic_index
[slot
] = 0;
1353 case VARYING_SLOT_TEX0
:
1354 case VARYING_SLOT_TEX1
:
1355 case VARYING_SLOT_TEX2
:
1356 case VARYING_SLOT_TEX3
:
1357 case VARYING_SLOT_TEX4
:
1358 case VARYING_SLOT_TEX5
:
1359 case VARYING_SLOT_TEX6
:
1360 case VARYING_SLOT_TEX7
:
1361 if (st
->needs_texcoord_semantic
) {
1362 input_semantic_name
[slot
] = TGSI_SEMANTIC_TEXCOORD
;
1363 input_semantic_index
[slot
] = attr
- VARYING_SLOT_TEX0
;
1367 case VARYING_SLOT_VAR0
:
1369 assert(attr
>= VARYING_SLOT_VAR0
||
1370 (attr
>= VARYING_SLOT_TEX0
&& attr
<= VARYING_SLOT_TEX7
));
1371 input_semantic_name
[slot
] = TGSI_SEMANTIC_GENERIC
;
1372 input_semantic_index
[slot
] =
1373 st_get_generic_varying_index(st
, attr
);
1379 /* Also add patch inputs. */
1380 for (attr
= 0; attr
< 32; attr
++) {
1381 if (prog
->info
.patch_inputs_read
& (1u << attr
)) {
1382 GLuint slot
= num_inputs
++;
1383 GLuint patch_attr
= VARYING_SLOT_PATCH0
+ attr
;
1385 inputMapping
[patch_attr
] = slot
;
1386 inputSlotToAttr
[slot
] = patch_attr
;
1387 input_semantic_name
[slot
] = TGSI_SEMANTIC_PATCH
;
1388 input_semantic_index
[slot
] = attr
;
1392 /* initialize output semantics to defaults */
1393 for (i
= 0; i
< PIPE_MAX_SHADER_OUTPUTS
; i
++) {
1394 output_semantic_name
[i
] = TGSI_SEMANTIC_GENERIC
;
1395 output_semantic_index
[i
] = 0;
1399 * Determine number of outputs, the (default) output register
1400 * mapping and the semantic information for each output.
1402 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
1403 if (prog
->info
.outputs_written
& BITFIELD64_BIT(attr
)) {
1404 GLuint slot
= num_outputs
++;
1406 outputMapping
[attr
] = slot
;
1409 case VARYING_SLOT_POS
:
1411 output_semantic_name
[slot
] = TGSI_SEMANTIC_POSITION
;
1412 output_semantic_index
[slot
] = 0;
1414 case VARYING_SLOT_COL0
:
1415 output_semantic_name
[slot
] = TGSI_SEMANTIC_COLOR
;
1416 output_semantic_index
[slot
] = 0;
1418 case VARYING_SLOT_COL1
:
1419 output_semantic_name
[slot
] = TGSI_SEMANTIC_COLOR
;
1420 output_semantic_index
[slot
] = 1;
1422 case VARYING_SLOT_BFC0
:
1423 output_semantic_name
[slot
] = TGSI_SEMANTIC_BCOLOR
;
1424 output_semantic_index
[slot
] = 0;
1426 case VARYING_SLOT_BFC1
:
1427 output_semantic_name
[slot
] = TGSI_SEMANTIC_BCOLOR
;
1428 output_semantic_index
[slot
] = 1;
1430 case VARYING_SLOT_FOGC
:
1431 output_semantic_name
[slot
] = TGSI_SEMANTIC_FOG
;
1432 output_semantic_index
[slot
] = 0;
1434 case VARYING_SLOT_PSIZ
:
1435 output_semantic_name
[slot
] = TGSI_SEMANTIC_PSIZE
;
1436 output_semantic_index
[slot
] = 0;
1438 case VARYING_SLOT_CLIP_VERTEX
:
1439 output_semantic_name
[slot
] = TGSI_SEMANTIC_CLIPVERTEX
;
1440 output_semantic_index
[slot
] = 0;
1442 case VARYING_SLOT_CLIP_DIST0
:
1443 output_semantic_name
[slot
] = TGSI_SEMANTIC_CLIPDIST
;
1444 output_semantic_index
[slot
] = 0;
1446 case VARYING_SLOT_CLIP_DIST1
:
1447 output_semantic_name
[slot
] = TGSI_SEMANTIC_CLIPDIST
;
1448 output_semantic_index
[slot
] = 1;
1450 case VARYING_SLOT_CULL_DIST0
:
1451 case VARYING_SLOT_CULL_DIST1
:
1452 /* these should have been lowered by GLSL */
1455 case VARYING_SLOT_LAYER
:
1456 output_semantic_name
[slot
] = TGSI_SEMANTIC_LAYER
;
1457 output_semantic_index
[slot
] = 0;
1459 case VARYING_SLOT_PRIMITIVE_ID
:
1460 output_semantic_name
[slot
] = TGSI_SEMANTIC_PRIMID
;
1461 output_semantic_index
[slot
] = 0;
1463 case VARYING_SLOT_VIEWPORT
:
1464 output_semantic_name
[slot
] = TGSI_SEMANTIC_VIEWPORT_INDEX
;
1465 output_semantic_index
[slot
] = 0;
1467 case VARYING_SLOT_TESS_LEVEL_OUTER
:
1468 output_semantic_name
[slot
] = TGSI_SEMANTIC_TESSOUTER
;
1469 output_semantic_index
[slot
] = 0;
1471 case VARYING_SLOT_TESS_LEVEL_INNER
:
1472 output_semantic_name
[slot
] = TGSI_SEMANTIC_TESSINNER
;
1473 output_semantic_index
[slot
] = 0;
1475 case VARYING_SLOT_TEX0
:
1476 case VARYING_SLOT_TEX1
:
1477 case VARYING_SLOT_TEX2
:
1478 case VARYING_SLOT_TEX3
:
1479 case VARYING_SLOT_TEX4
:
1480 case VARYING_SLOT_TEX5
:
1481 case VARYING_SLOT_TEX6
:
1482 case VARYING_SLOT_TEX7
:
1483 if (st
->needs_texcoord_semantic
) {
1484 output_semantic_name
[slot
] = TGSI_SEMANTIC_TEXCOORD
;
1485 output_semantic_index
[slot
] = attr
- VARYING_SLOT_TEX0
;
1489 case VARYING_SLOT_VAR0
:
1491 assert(slot
< ARRAY_SIZE(output_semantic_name
));
1492 assert(attr
>= VARYING_SLOT_VAR0
||
1493 (attr
>= VARYING_SLOT_TEX0
&& attr
<= VARYING_SLOT_TEX7
));
1494 output_semantic_name
[slot
] = TGSI_SEMANTIC_GENERIC
;
1495 output_semantic_index
[slot
] =
1496 st_get_generic_varying_index(st
, attr
);
1502 /* Also add patch outputs. */
1503 for (attr
= 0; attr
< 32; attr
++) {
1504 if (prog
->info
.patch_outputs_written
& (1u << attr
)) {
1505 GLuint slot
= num_outputs
++;
1506 GLuint patch_attr
= VARYING_SLOT_PATCH0
+ attr
;
1508 outputMapping
[patch_attr
] = slot
;
1509 output_semantic_name
[slot
] = TGSI_SEMANTIC_PATCH
;
1510 output_semantic_index
[slot
] = attr
;
1514 st_translate_program(st
->ctx
,
1523 input_semantic_name
,
1524 input_semantic_index
,
1529 output_semantic_name
,
1530 output_semantic_index
);
1532 unsigned num_tokens
;
1533 out_state
->tokens
= ureg_get_tokens(ureg
, &num_tokens
);
1536 st_translate_stream_output_info(glsl_to_tgsi
,
1538 &out_state
->stream_output
);
1540 st_store_tgsi_in_disk_cache(st
, prog
, out_state
, num_tokens
);
1542 if ((ST_DEBUG
& DEBUG_TGSI
) && (ST_DEBUG
& DEBUG_MESA
)) {
1543 _mesa_print_program(prog
);
1547 if (ST_DEBUG
& DEBUG_TGSI
) {
1548 tgsi_dump(out_state
->tokens
, 0);
1555 * Translate a geometry program to create a new variant.
1558 st_translate_geometry_program(struct st_context
*st
,
1559 struct st_common_program
*stgp
)
1561 struct ureg_program
*ureg
;
1563 ureg
= ureg_create_with_screen(PIPE_SHADER_GEOMETRY
, st
->pipe
->screen
);
1567 ureg_property(ureg
, TGSI_PROPERTY_GS_INPUT_PRIM
,
1568 stgp
->Base
.info
.gs
.input_primitive
);
1569 ureg_property(ureg
, TGSI_PROPERTY_GS_OUTPUT_PRIM
,
1570 stgp
->Base
.info
.gs
.output_primitive
);
1571 ureg_property(ureg
, TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES
,
1572 stgp
->Base
.info
.gs
.vertices_out
);
1573 ureg_property(ureg
, TGSI_PROPERTY_GS_INVOCATIONS
,
1574 stgp
->Base
.info
.gs
.invocations
);
1576 st_translate_program_common(st
, &stgp
->Base
, stgp
->glsl_to_tgsi
, ureg
,
1577 PIPE_SHADER_GEOMETRY
, &stgp
->tgsi
);
1579 free_glsl_to_tgsi_visitor(stgp
->glsl_to_tgsi
);
1580 stgp
->glsl_to_tgsi
= NULL
;
1586 * Get/create a basic program variant.
1588 struct st_basic_variant
*
1589 st_get_basic_variant(struct st_context
*st
,
1590 unsigned pipe_shader
,
1591 struct pipe_shader_state
*tgsi
,
1592 struct st_basic_variant
**variants
)
1594 struct pipe_context
*pipe
= st
->pipe
;
1595 struct st_basic_variant
*v
;
1596 struct st_basic_variant_key key
;
1598 memset(&key
, 0, sizeof(key
));
1599 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
1601 /* Search for existing variant */
1602 for (v
= *variants
; v
; v
= v
->next
) {
1603 if (memcmp(&v
->key
, &key
, sizeof(key
)) == 0) {
1610 v
= CALLOC_STRUCT(st_basic_variant
);
1612 /* fill in new variant */
1613 switch (pipe_shader
) {
1614 case PIPE_SHADER_TESS_CTRL
:
1615 v
->driver_shader
= pipe
->create_tcs_state(pipe
, tgsi
);
1617 case PIPE_SHADER_TESS_EVAL
:
1618 v
->driver_shader
= pipe
->create_tes_state(pipe
, tgsi
);
1620 case PIPE_SHADER_GEOMETRY
:
1621 v
->driver_shader
= pipe
->create_gs_state(pipe
, tgsi
);
1624 assert(!"unhandled shader type");
1631 /* insert into list */
1632 v
->next
= *variants
;
1642 * Translate a tessellation control program to create a new variant.
1645 st_translate_tessctrl_program(struct st_context
*st
,
1646 struct st_common_program
*sttcp
)
1648 struct ureg_program
*ureg
;
1650 ureg
= ureg_create_with_screen(PIPE_SHADER_TESS_CTRL
, st
->pipe
->screen
);
1654 ureg_property(ureg
, TGSI_PROPERTY_TCS_VERTICES_OUT
,
1655 sttcp
->Base
.info
.tess
.tcs_vertices_out
);
1657 st_translate_program_common(st
, &sttcp
->Base
, sttcp
->glsl_to_tgsi
, ureg
,
1658 PIPE_SHADER_TESS_CTRL
, &sttcp
->tgsi
);
1660 free_glsl_to_tgsi_visitor(sttcp
->glsl_to_tgsi
);
1661 sttcp
->glsl_to_tgsi
= NULL
;
1667 * Translate a tessellation evaluation program to create a new variant.
1670 st_translate_tesseval_program(struct st_context
*st
,
1671 struct st_common_program
*sttep
)
1673 struct ureg_program
*ureg
;
1675 ureg
= ureg_create_with_screen(PIPE_SHADER_TESS_EVAL
, st
->pipe
->screen
);
1679 if (sttep
->Base
.info
.tess
.primitive_mode
== GL_ISOLINES
)
1680 ureg_property(ureg
, TGSI_PROPERTY_TES_PRIM_MODE
, GL_LINES
);
1682 ureg_property(ureg
, TGSI_PROPERTY_TES_PRIM_MODE
,
1683 sttep
->Base
.info
.tess
.primitive_mode
);
1685 STATIC_ASSERT((TESS_SPACING_EQUAL
+ 1) % 3 == PIPE_TESS_SPACING_EQUAL
);
1686 STATIC_ASSERT((TESS_SPACING_FRACTIONAL_ODD
+ 1) % 3 ==
1687 PIPE_TESS_SPACING_FRACTIONAL_ODD
);
1688 STATIC_ASSERT((TESS_SPACING_FRACTIONAL_EVEN
+ 1) % 3 ==
1689 PIPE_TESS_SPACING_FRACTIONAL_EVEN
);
1691 ureg_property(ureg
, TGSI_PROPERTY_TES_SPACING
,
1692 (sttep
->Base
.info
.tess
.spacing
+ 1) % 3);
1694 ureg_property(ureg
, TGSI_PROPERTY_TES_VERTEX_ORDER_CW
,
1695 !sttep
->Base
.info
.tess
.ccw
);
1696 ureg_property(ureg
, TGSI_PROPERTY_TES_POINT_MODE
,
1697 sttep
->Base
.info
.tess
.point_mode
);
1699 st_translate_program_common(st
, &sttep
->Base
, sttep
->glsl_to_tgsi
,
1700 ureg
, PIPE_SHADER_TESS_EVAL
, &sttep
->tgsi
);
1702 free_glsl_to_tgsi_visitor(sttep
->glsl_to_tgsi
);
1703 sttep
->glsl_to_tgsi
= NULL
;
1709 * Translate a compute program to create a new variant.
1712 st_translate_compute_program(struct st_context
*st
,
1713 struct st_compute_program
*stcp
)
1715 struct ureg_program
*ureg
;
1716 struct pipe_shader_state prog
;
1718 if (stcp
->shader_program
) {
1719 nir_shader
*nir
= st_glsl_to_nir(st
, &stcp
->Base
, stcp
->shader_program
,
1720 MESA_SHADER_COMPUTE
);
1722 /* no compute variants: */
1723 st_finalize_nir(st
, &stcp
->Base
, nir
);
1725 stcp
->tgsi
.ir_type
= PIPE_SHADER_IR_NIR
;
1726 stcp
->tgsi
.prog
= nir
;
1731 ureg
= ureg_create_with_screen(PIPE_SHADER_COMPUTE
, st
->pipe
->screen
);
1735 st_translate_program_common(st
, &stcp
->Base
, stcp
->glsl_to_tgsi
, ureg
,
1736 PIPE_SHADER_COMPUTE
, &prog
);
1738 stcp
->tgsi
.ir_type
= PIPE_SHADER_IR_TGSI
;
1739 stcp
->tgsi
.prog
= prog
.tokens
;
1740 stcp
->tgsi
.req_local_mem
= stcp
->Base
.info
.cs
.shared_size
;
1741 stcp
->tgsi
.req_private_mem
= 0;
1742 stcp
->tgsi
.req_input_mem
= 0;
1744 free_glsl_to_tgsi_visitor(stcp
->glsl_to_tgsi
);
1745 stcp
->glsl_to_tgsi
= NULL
;
1751 * Get/create compute program variant.
1753 struct st_basic_variant
*
1754 st_get_cp_variant(struct st_context
*st
,
1755 struct pipe_compute_state
*tgsi
,
1756 struct st_basic_variant
**variants
)
1758 struct pipe_context
*pipe
= st
->pipe
;
1759 struct st_basic_variant
*v
;
1760 struct st_basic_variant_key key
;
1762 memset(&key
, 0, sizeof(key
));
1763 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
1765 /* Search for existing variant */
1766 for (v
= *variants
; v
; v
= v
->next
) {
1767 if (memcmp(&v
->key
, &key
, sizeof(key
)) == 0) {
1774 v
= CALLOC_STRUCT(st_basic_variant
);
1776 /* fill in new variant */
1777 v
->driver_shader
= pipe
->create_compute_state(pipe
, tgsi
);
1780 /* insert into list */
1781 v
->next
= *variants
;
1791 * Vert/Geom/Frag programs have per-context variants. Free all the
1792 * variants attached to the given program which match the given context.
1795 destroy_program_variants(struct st_context
*st
, struct gl_program
*target
)
1797 if (!target
|| target
== &_mesa_DummyProgram
)
1800 switch (target
->Target
) {
1801 case GL_VERTEX_PROGRAM_ARB
:
1803 struct st_vertex_program
*stvp
= (struct st_vertex_program
*) target
;
1804 struct st_vp_variant
*vpv
, **prevPtr
= &stvp
->variants
;
1806 for (vpv
= stvp
->variants
; vpv
; ) {
1807 struct st_vp_variant
*next
= vpv
->next
;
1808 if (vpv
->key
.st
== st
) {
1809 /* unlink from list */
1811 /* destroy this variant */
1812 delete_vp_variant(st
, vpv
);
1815 prevPtr
= &vpv
->next
;
1821 case GL_FRAGMENT_PROGRAM_ARB
:
1823 struct st_fragment_program
*stfp
=
1824 (struct st_fragment_program
*) target
;
1825 struct st_fp_variant
*fpv
, **prevPtr
= &stfp
->variants
;
1827 for (fpv
= stfp
->variants
; fpv
; ) {
1828 struct st_fp_variant
*next
= fpv
->next
;
1829 if (fpv
->key
.st
== st
) {
1830 /* unlink from list */
1832 /* destroy this variant */
1833 delete_fp_variant(st
, fpv
);
1836 prevPtr
= &fpv
->next
;
1842 case GL_GEOMETRY_PROGRAM_NV
:
1843 case GL_TESS_CONTROL_PROGRAM_NV
:
1844 case GL_TESS_EVALUATION_PROGRAM_NV
:
1845 case GL_COMPUTE_PROGRAM_NV
:
1847 struct st_common_program
*p
= st_common_program(target
);
1848 struct st_compute_program
*cp
= (struct st_compute_program
*)target
;
1849 struct st_basic_variant
**variants
=
1850 target
->Target
== GL_COMPUTE_PROGRAM_NV
? &cp
->variants
:
1852 struct st_basic_variant
*v
, **prevPtr
= variants
;
1854 for (v
= *variants
; v
; ) {
1855 struct st_basic_variant
*next
= v
->next
;
1856 if (v
->key
.st
== st
) {
1857 /* unlink from list */
1859 /* destroy this variant */
1860 delete_basic_variant(st
, v
, target
->Target
);
1870 _mesa_problem(NULL
, "Unexpected program target 0x%x in "
1871 "destroy_program_variants_cb()", target
->Target
);
1877 * Callback for _mesa_HashWalk. Free all the shader's program variants
1878 * which match the given context.
1881 destroy_shader_program_variants_cb(GLuint key
, void *data
, void *userData
)
1883 struct st_context
*st
= (struct st_context
*) userData
;
1884 struct gl_shader
*shader
= (struct gl_shader
*) data
;
1886 switch (shader
->Type
) {
1887 case GL_SHADER_PROGRAM_MESA
:
1889 struct gl_shader_program
*shProg
= (struct gl_shader_program
*) data
;
1892 for (i
= 0; i
< ARRAY_SIZE(shProg
->_LinkedShaders
); i
++) {
1893 if (shProg
->_LinkedShaders
[i
])
1894 destroy_program_variants(st
, shProg
->_LinkedShaders
[i
]->Program
);
1898 case GL_VERTEX_SHADER
:
1899 case GL_FRAGMENT_SHADER
:
1900 case GL_GEOMETRY_SHADER
:
1901 case GL_TESS_CONTROL_SHADER
:
1902 case GL_TESS_EVALUATION_SHADER
:
1903 case GL_COMPUTE_SHADER
:
1912 * Callback for _mesa_HashWalk. Free all the program variants which match
1913 * the given context.
1916 destroy_program_variants_cb(GLuint key
, void *data
, void *userData
)
1918 struct st_context
*st
= (struct st_context
*) userData
;
1919 struct gl_program
*program
= (struct gl_program
*) data
;
1920 destroy_program_variants(st
, program
);
1925 * Walk over all shaders and programs to delete any variants which
1926 * belong to the given context.
1927 * This is called during context tear-down.
1930 st_destroy_program_variants(struct st_context
*st
)
1932 /* If shaders can be shared with other contexts, the last context will
1933 * call DeleteProgram on all shaders, releasing everything.
1935 if (st
->has_shareable_shaders
)
1938 /* ARB vert/frag program */
1939 _mesa_HashWalk(st
->ctx
->Shared
->Programs
,
1940 destroy_program_variants_cb
, st
);
1942 /* GLSL vert/frag/geom shaders */
1943 _mesa_HashWalk(st
->ctx
->Shared
->ShaderObjects
,
1944 destroy_shader_program_variants_cb
, st
);
1949 * For debugging, print/dump the current vertex program.
1952 st_print_current_vertex_program(void)
1954 GET_CURRENT_CONTEXT(ctx
);
1956 if (ctx
->VertexProgram
._Current
) {
1957 struct st_vertex_program
*stvp
=
1958 (struct st_vertex_program
*) ctx
->VertexProgram
._Current
;
1959 struct st_vp_variant
*stv
;
1961 debug_printf("Vertex program %u\n", stvp
->Base
.Id
);
1963 for (stv
= stvp
->variants
; stv
; stv
= stv
->next
) {
1964 debug_printf("variant %p\n", stv
);
1965 tgsi_dump(stv
->tgsi
.tokens
, 0);
1972 * Compile one shader variant.
1975 st_precompile_shader_variant(struct st_context
*st
,
1976 struct gl_program
*prog
)
1978 switch (prog
->Target
) {
1979 case GL_VERTEX_PROGRAM_ARB
: {
1980 struct st_vertex_program
*p
= (struct st_vertex_program
*)prog
;
1981 struct st_vp_variant_key key
;
1983 memset(&key
, 0, sizeof(key
));
1984 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
1985 st_get_vp_variant(st
, p
, &key
);
1989 case GL_TESS_CONTROL_PROGRAM_NV
: {
1990 struct st_common_program
*p
= st_common_program(prog
);
1991 st_get_basic_variant(st
, PIPE_SHADER_TESS_CTRL
, &p
->tgsi
, &p
->variants
);
1995 case GL_TESS_EVALUATION_PROGRAM_NV
: {
1996 struct st_common_program
*p
= st_common_program(prog
);
1997 st_get_basic_variant(st
, PIPE_SHADER_TESS_EVAL
, &p
->tgsi
, &p
->variants
);
2001 case GL_GEOMETRY_PROGRAM_NV
: {
2002 struct st_common_program
*p
= st_common_program(prog
);
2003 st_get_basic_variant(st
, PIPE_SHADER_GEOMETRY
, &p
->tgsi
, &p
->variants
);
2007 case GL_FRAGMENT_PROGRAM_ARB
: {
2008 struct st_fragment_program
*p
= (struct st_fragment_program
*)prog
;
2009 struct st_fp_variant_key key
;
2011 memset(&key
, 0, sizeof(key
));
2012 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
2013 st_get_fp_variant(st
, p
, &key
);
2017 case GL_COMPUTE_PROGRAM_NV
: {
2018 struct st_compute_program
*p
= (struct st_compute_program
*)prog
;
2019 st_get_cp_variant(st
, &p
->tgsi
, &p
->variants
);