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/errors.h"
35 #include "main/imports.h"
36 #include "main/hash.h"
37 #include "main/mtypes.h"
38 #include "program/prog_parameter.h"
39 #include "program/prog_print.h"
40 #include "program/prog_to_nir.h"
41 #include "program/programopt.h"
43 #include "compiler/nir/nir.h"
45 #include "pipe/p_context.h"
46 #include "pipe/p_defines.h"
47 #include "pipe/p_shader_tokens.h"
48 #include "draw/draw_context.h"
49 #include "tgsi/tgsi_dump.h"
50 #include "tgsi/tgsi_emulate.h"
51 #include "tgsi/tgsi_parse.h"
52 #include "tgsi/tgsi_ureg.h"
55 #include "st_cb_bitmap.h"
56 #include "st_cb_drawpixels.h"
57 #include "st_context.h"
58 #include "st_tgsi_lower_depth_clamp.h"
59 #include "st_tgsi_lower_yuv.h"
60 #include "st_program.h"
61 #include "st_mesa_to_tgsi.h"
62 #include "st_atifs_to_tgsi.h"
64 #include "st_shader_cache.h"
65 #include "cso_cache/cso_context.h"
70 set_affected_state_flags(uint64_t *states
,
71 struct gl_program
*prog
,
72 uint64_t new_constants
,
73 uint64_t new_sampler_views
,
74 uint64_t new_samplers
,
80 if (prog
->Parameters
->NumParameters
)
81 *states
|= new_constants
;
83 if (prog
->info
.num_textures
)
84 *states
|= new_sampler_views
| new_samplers
;
86 if (prog
->info
.num_images
)
87 *states
|= new_images
;
89 if (prog
->info
.num_ubos
)
92 if (prog
->info
.num_ssbos
)
95 if (prog
->info
.num_abos
)
96 *states
|= new_atomics
;
100 * This determines which states will be updated when the shader is bound.
103 st_set_prog_affected_state_flags(struct gl_program
*prog
)
107 switch (prog
->info
.stage
) {
108 case MESA_SHADER_VERTEX
:
109 states
= &((struct st_vertex_program
*)prog
)->affected_states
;
111 *states
= ST_NEW_VS_STATE
|
113 ST_NEW_VERTEX_ARRAYS
;
115 set_affected_state_flags(states
, prog
,
117 ST_NEW_VS_SAMPLER_VIEWS
,
125 case MESA_SHADER_TESS_CTRL
:
126 states
= &(st_common_program(prog
))->affected_states
;
128 *states
= ST_NEW_TCS_STATE
;
130 set_affected_state_flags(states
, prog
,
131 ST_NEW_TCS_CONSTANTS
,
132 ST_NEW_TCS_SAMPLER_VIEWS
,
140 case MESA_SHADER_TESS_EVAL
:
141 states
= &(st_common_program(prog
))->affected_states
;
143 *states
= ST_NEW_TES_STATE
|
146 set_affected_state_flags(states
, prog
,
147 ST_NEW_TES_CONSTANTS
,
148 ST_NEW_TES_SAMPLER_VIEWS
,
156 case MESA_SHADER_GEOMETRY
:
157 states
= &(st_common_program(prog
))->affected_states
;
159 *states
= ST_NEW_GS_STATE
|
162 set_affected_state_flags(states
, prog
,
164 ST_NEW_GS_SAMPLER_VIEWS
,
172 case MESA_SHADER_FRAGMENT
:
173 states
= &((struct st_fragment_program
*)prog
)->affected_states
;
175 /* gl_FragCoord and glDrawPixels always use constants. */
176 *states
= ST_NEW_FS_STATE
|
177 ST_NEW_SAMPLE_SHADING
|
180 set_affected_state_flags(states
, prog
,
182 ST_NEW_FS_SAMPLER_VIEWS
,
190 case MESA_SHADER_COMPUTE
:
191 states
= &((struct st_common_program
*)prog
)->affected_states
;
193 *states
= ST_NEW_CS_STATE
;
195 set_affected_state_flags(states
, prog
,
197 ST_NEW_CS_SAMPLER_VIEWS
,
206 unreachable("unhandled shader stage");
211 delete_ir(struct pipe_shader_state
*ir
)
214 ureg_free_tokens(ir
->tokens
);
218 /* Note: Any setup of ->ir.nir that has had pipe->create_*_state called on
219 * it has resulted in the driver taking ownership of the NIR. Those
220 * callers should be NULLing out the nir field in any pipe_shader_state
221 * that might have this called in order to indicate that.
223 * GLSL IR and ARB programs will have set gl_program->nir to the same
224 * shader as ir->ir.nir, so it will be freed by _mesa_delete_program().
229 * Delete a vertex program variant. Note the caller must unlink
230 * the variant from the linked list.
233 delete_vp_variant(struct st_context
*st
, struct st_vp_variant
*vpv
)
235 if (vpv
->driver_shader
) {
236 if (st
->has_shareable_shaders
|| vpv
->key
.st
== st
) {
237 cso_delete_vertex_shader(st
->cso_context
, vpv
->driver_shader
);
239 st_save_zombie_shader(vpv
->key
.st
, PIPE_SHADER_VERTEX
,
244 if (vpv
->draw_shader
)
245 draw_delete_vertex_shader( st
->draw
, vpv
->draw_shader
);
247 delete_ir(&vpv
->tgsi
);
255 * Clean out any old compilations:
258 st_release_vp_variants( struct st_context
*st
,
259 struct st_vertex_program
*stvp
)
261 struct st_vp_variant
*vpv
;
263 for (vpv
= stvp
->variants
; vpv
; ) {
264 struct st_vp_variant
*next
= vpv
->next
;
265 delete_vp_variant(st
, vpv
);
269 stvp
->variants
= NULL
;
271 delete_ir(&stvp
->tgsi
);
277 * Delete a fragment program variant. Note the caller must unlink
278 * the variant from the linked list.
281 delete_fp_variant(struct st_context
*st
, struct st_fp_variant
*fpv
)
283 if (fpv
->driver_shader
) {
284 if (st
->has_shareable_shaders
|| fpv
->key
.st
== st
) {
285 cso_delete_fragment_shader(st
->cso_context
, fpv
->driver_shader
);
287 st_save_zombie_shader(fpv
->key
.st
, PIPE_SHADER_FRAGMENT
,
297 * Free all variants of a fragment program.
300 st_release_fp_variants(struct st_context
*st
, struct st_fragment_program
*stfp
)
302 struct st_fp_variant
*fpv
;
304 for (fpv
= stfp
->variants
; fpv
; ) {
305 struct st_fp_variant
*next
= fpv
->next
;
306 delete_fp_variant(st
, fpv
);
310 stfp
->variants
= NULL
;
312 delete_ir(&stfp
->tgsi
);
317 * Delete a basic program variant. Note the caller must unlink
318 * the variant from the linked list.
321 delete_basic_variant(struct st_context
*st
, struct st_basic_variant
*v
,
324 if (v
->driver_shader
) {
325 if (st
->has_shareable_shaders
|| v
->key
.st
== st
) {
326 /* The shader's context matches the calling context, or we
330 case GL_TESS_CONTROL_PROGRAM_NV
:
331 cso_delete_tessctrl_shader(st
->cso_context
, v
->driver_shader
);
333 case GL_TESS_EVALUATION_PROGRAM_NV
:
334 cso_delete_tesseval_shader(st
->cso_context
, v
->driver_shader
);
336 case GL_GEOMETRY_PROGRAM_NV
:
337 cso_delete_geometry_shader(st
->cso_context
, v
->driver_shader
);
339 case GL_COMPUTE_PROGRAM_NV
:
340 cso_delete_compute_shader(st
->cso_context
, v
->driver_shader
);
343 unreachable("bad shader type in delete_basic_variant");
346 /* We can't delete a shader with a context different from the one
347 * that created it. Add it to the creating context's zombie list.
349 enum pipe_shader_type type
;
351 case GL_TESS_CONTROL_PROGRAM_NV
:
352 type
= PIPE_SHADER_TESS_CTRL
;
354 case GL_TESS_EVALUATION_PROGRAM_NV
:
355 type
= PIPE_SHADER_TESS_EVAL
;
357 case GL_GEOMETRY_PROGRAM_NV
:
358 type
= PIPE_SHADER_GEOMETRY
;
363 st_save_zombie_shader(v
->key
.st
, type
, v
->driver_shader
);
372 * Free all basic program variants.
375 st_release_basic_variants(struct st_context
*st
, struct st_common_program
*p
)
377 struct st_basic_variant
*v
;
379 for (v
= p
->variants
; v
; ) {
380 struct st_basic_variant
*next
= v
->next
;
381 delete_basic_variant(st
, v
, p
->Base
.Target
);
391 * Translate ARB (asm) program to NIR
394 st_translate_prog_to_nir(struct st_context
*st
, struct gl_program
*prog
,
395 gl_shader_stage stage
)
397 const struct gl_shader_compiler_options
*options
=
398 &st
->ctx
->Const
.ShaderCompilerOptions
[stage
];
400 /* Translate to NIR */
401 nir_shader
*nir
= prog_to_nir(prog
, options
->NirOptions
);
402 NIR_PASS_V(nir
, nir_lower_regs_to_ssa
); /* turn registers into SSA */
403 nir_validate_shader(nir
, "after st/ptn lower_regs_to_ssa");
405 NIR_PASS_V(nir
, st_nir_lower_wpos_ytransform
, prog
, st
->pipe
->screen
);
406 NIR_PASS_V(nir
, nir_lower_system_values
);
409 NIR_PASS_V(nir
, nir_opt_constant_folding
);
411 nir_validate_shader(nir
, "after st/ptn NIR opts");
417 st_prepare_vertex_program(struct st_vertex_program
*stvp
)
419 stvp
->num_inputs
= 0;
420 memset(stvp
->input_to_index
, ~0, sizeof(stvp
->input_to_index
));
421 memset(stvp
->result_to_output
, ~0, sizeof(stvp
->result_to_output
));
423 /* Determine number of inputs, the mappings between VERT_ATTRIB_x
424 * and TGSI generic input indexes, plus input attrib semantic info.
426 for (unsigned attr
= 0; attr
< VERT_ATTRIB_MAX
; attr
++) {
427 if ((stvp
->Base
.info
.inputs_read
& BITFIELD64_BIT(attr
)) != 0) {
428 stvp
->input_to_index
[attr
] = stvp
->num_inputs
;
429 stvp
->index_to_input
[stvp
->num_inputs
] = attr
;
432 if ((stvp
->Base
.DualSlotInputs
& BITFIELD64_BIT(attr
)) != 0) {
433 /* add placeholder for second part of a double attribute */
434 stvp
->index_to_input
[stvp
->num_inputs
] = ST_DOUBLE_ATTRIB_PLACEHOLDER
;
439 /* pre-setup potentially unused edgeflag input */
440 stvp
->input_to_index
[VERT_ATTRIB_EDGEFLAG
] = stvp
->num_inputs
;
441 stvp
->index_to_input
[stvp
->num_inputs
] = VERT_ATTRIB_EDGEFLAG
;
443 /* Compute mapping of vertex program outputs to slots. */
444 unsigned num_outputs
= 0;
445 for (unsigned attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
446 if (stvp
->Base
.info
.outputs_written
& BITFIELD64_BIT(attr
))
447 stvp
->result_to_output
[attr
] = num_outputs
++;
449 /* pre-setup potentially unused edgeflag output */
450 stvp
->result_to_output
[VARYING_SLOT_EDGE
] = num_outputs
;
454 * Translate a vertex program.
457 st_translate_vertex_program(struct st_context
*st
,
458 struct st_vertex_program
*stvp
)
460 struct ureg_program
*ureg
;
461 enum pipe_error error
;
462 unsigned num_outputs
= 0;
464 ubyte output_semantic_name
[VARYING_SLOT_MAX
] = {0};
465 ubyte output_semantic_index
[VARYING_SLOT_MAX
] = {0};
467 if (stvp
->Base
.arb
.IsPositionInvariant
)
468 _mesa_insert_mvp_code(st
->ctx
, &stvp
->Base
);
470 st_prepare_vertex_program(stvp
);
472 /* Get semantic names and indices. */
473 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
474 if (stvp
->Base
.info
.outputs_written
& BITFIELD64_BIT(attr
)) {
475 unsigned slot
= num_outputs
++;
476 unsigned semantic_name
, semantic_index
;
477 tgsi_get_gl_varying_semantic(attr
, st
->needs_texcoord_semantic
,
478 &semantic_name
, &semantic_index
);
479 output_semantic_name
[slot
] = semantic_name
;
480 output_semantic_index
[slot
] = semantic_index
;
483 /* pre-setup potentially unused edgeflag output */
484 output_semantic_name
[num_outputs
] = TGSI_SEMANTIC_EDGEFLAG
;
485 output_semantic_index
[num_outputs
] = 0;
488 if (!stvp
->glsl_to_tgsi
&& !stvp
->shader_program
) {
489 _mesa_remove_output_reads(&stvp
->Base
, PROGRAM_OUTPUT
);
491 /* This determines which states will be updated when the assembly
494 stvp
->affected_states
= ST_NEW_VS_STATE
|
496 ST_NEW_VERTEX_ARRAYS
;
498 if (stvp
->Base
.Parameters
->NumParameters
)
499 stvp
->affected_states
|= ST_NEW_VS_CONSTANTS
;
501 /* No samplers are allowed in ARB_vp. */
504 if (stvp
->shader_program
) {
505 st_translate_stream_output_info(stvp
->Base
.sh
.LinkedTransformFeedback
,
506 stvp
->result_to_output
,
507 &stvp
->tgsi
.stream_output
);
509 st_store_ir_in_disk_cache(st
, &stvp
->Base
, true);
513 ureg
= ureg_create_with_screen(PIPE_SHADER_VERTEX
, st
->pipe
->screen
);
517 if (stvp
->Base
.info
.clip_distance_array_size
)
518 ureg_property(ureg
, TGSI_PROPERTY_NUM_CLIPDIST_ENABLED
,
519 stvp
->Base
.info
.clip_distance_array_size
);
520 if (stvp
->Base
.info
.cull_distance_array_size
)
521 ureg_property(ureg
, TGSI_PROPERTY_NUM_CULLDIST_ENABLED
,
522 stvp
->Base
.info
.cull_distance_array_size
);
524 if (ST_DEBUG
& DEBUG_MESA
) {
525 _mesa_print_program(&stvp
->Base
);
526 _mesa_print_program_parameters(st
->ctx
, &stvp
->Base
);
530 if (stvp
->glsl_to_tgsi
) {
531 error
= st_translate_program(st
->ctx
,
538 stvp
->input_to_index
,
539 NULL
, /* inputSlotToAttr */
540 NULL
, /* input semantic name */
541 NULL
, /* input semantic index */
542 NULL
, /* interp mode */
545 stvp
->result_to_output
,
546 output_semantic_name
,
547 output_semantic_index
);
549 st_translate_stream_output_info(stvp
->Base
.sh
.LinkedTransformFeedback
,
550 stvp
->result_to_output
,
551 &stvp
->tgsi
.stream_output
);
553 free_glsl_to_tgsi_visitor(stvp
->glsl_to_tgsi
);
555 error
= st_translate_mesa_program(st
->ctx
,
561 stvp
->input_to_index
,
562 NULL
, /* input semantic name */
563 NULL
, /* input semantic index */
567 stvp
->result_to_output
,
568 output_semantic_name
,
569 output_semantic_index
);
572 debug_printf("%s: failed to translate Mesa program:\n", __func__
);
573 _mesa_print_program(&stvp
->Base
);
578 stvp
->tgsi
.tokens
= ureg_get_tokens(ureg
, &stvp
->num_tgsi_tokens
);
581 if (stvp
->glsl_to_tgsi
) {
582 stvp
->glsl_to_tgsi
= NULL
;
583 st_store_ir_in_disk_cache(st
, &stvp
->Base
, false);
586 bool use_nir
= PIPE_SHADER_IR_NIR
==
587 st
->pipe
->screen
->get_shader_param(st
->pipe
->screen
, PIPE_SHADER_VERTEX
,
588 PIPE_SHADER_CAP_PREFERRED_IR
);
592 st_translate_prog_to_nir(st
, &stvp
->Base
, MESA_SHADER_VERTEX
);
594 if (stvp
->tgsi
.ir
.nir
)
595 ralloc_free(stvp
->tgsi
.ir
.nir
);
596 stvp
->tgsi
.type
= PIPE_SHADER_IR_NIR
;
597 stvp
->tgsi
.ir
.nir
= nir
;
598 stvp
->Base
.nir
= nir
;
602 return stvp
->tgsi
.tokens
!= NULL
;
605 static const gl_state_index16 depth_range_state
[STATE_LENGTH
] =
606 { STATE_DEPTH_RANGE
};
608 static struct st_vp_variant
*
609 st_create_vp_variant(struct st_context
*st
,
610 struct st_vertex_program
*stvp
,
611 const struct st_vp_variant_key
*key
)
613 struct st_vp_variant
*vpv
= CALLOC_STRUCT(st_vp_variant
);
614 struct pipe_context
*pipe
= st
->pipe
;
615 struct gl_program_parameter_list
*params
= stvp
->Base
.Parameters
;
618 vpv
->tgsi
.stream_output
= stvp
->tgsi
.stream_output
;
619 vpv
->num_inputs
= stvp
->num_inputs
;
621 /* When generating a NIR program, we usually don't have TGSI tokens.
622 * However, we do create them for ARB_vertex_program / fixed-function VS
623 * programs which we may need to use with the draw module for legacy
624 * feedback/select emulation. If they exist, copy them.
626 if (stvp
->tgsi
.tokens
)
627 vpv
->tgsi
.tokens
= tgsi_dup_tokens(stvp
->tgsi
.tokens
);
629 if (stvp
->tgsi
.type
== PIPE_SHADER_IR_NIR
) {
630 vpv
->tgsi
.type
= PIPE_SHADER_IR_NIR
;
631 vpv
->tgsi
.ir
.nir
= nir_shader_clone(NULL
, stvp
->tgsi
.ir
.nir
);
632 if (key
->clamp_color
)
633 NIR_PASS_V(vpv
->tgsi
.ir
.nir
, nir_lower_clamp_color_outputs
);
634 if (key
->passthrough_edgeflags
) {
635 NIR_PASS_V(vpv
->tgsi
.ir
.nir
, nir_lower_passthrough_edgeflags
);
639 st_finalize_nir(st
, &stvp
->Base
, stvp
->shader_program
,
642 vpv
->driver_shader
= pipe
->create_vs_state(pipe
, &vpv
->tgsi
);
643 /* driver takes ownership of IR: */
644 vpv
->tgsi
.ir
.nir
= NULL
;
648 /* Emulate features. */
649 if (key
->clamp_color
|| key
->passthrough_edgeflags
) {
650 const struct tgsi_token
*tokens
;
652 (key
->clamp_color
? TGSI_EMU_CLAMP_COLOR_OUTPUTS
: 0) |
653 (key
->passthrough_edgeflags
? TGSI_EMU_PASSTHROUGH_EDGEFLAG
: 0);
655 tokens
= tgsi_emulate(vpv
->tgsi
.tokens
, flags
);
658 tgsi_free_tokens(vpv
->tgsi
.tokens
);
659 vpv
->tgsi
.tokens
= tokens
;
661 if (key
->passthrough_edgeflags
)
664 fprintf(stderr
, "mesa: cannot emulate deprecated features\n");
667 if (key
->lower_depth_clamp
) {
668 unsigned depth_range_const
=
669 _mesa_add_state_reference(params
, depth_range_state
);
671 const struct tgsi_token
*tokens
;
672 tokens
= st_tgsi_lower_depth_clamp(vpv
->tgsi
.tokens
, depth_range_const
,
673 key
->clip_negative_one_to_one
);
674 if (tokens
!= vpv
->tgsi
.tokens
)
675 tgsi_free_tokens(vpv
->tgsi
.tokens
);
676 vpv
->tgsi
.tokens
= tokens
;
679 if (ST_DEBUG
& DEBUG_TGSI
) {
680 tgsi_dump(vpv
->tgsi
.tokens
, 0);
684 vpv
->driver_shader
= pipe
->create_vs_state(pipe
, &vpv
->tgsi
);
690 * Find/create a vertex program variant.
692 struct st_vp_variant
*
693 st_get_vp_variant(struct st_context
*st
,
694 struct st_vertex_program
*stvp
,
695 const struct st_vp_variant_key
*key
)
697 struct st_vp_variant
*vpv
;
699 /* Search for existing variant */
700 for (vpv
= stvp
->variants
; vpv
; vpv
= vpv
->next
) {
701 if (memcmp(&vpv
->key
, key
, sizeof(*key
)) == 0) {
708 vpv
= st_create_vp_variant(st
, stvp
, key
);
710 for (unsigned index
= 0; index
< vpv
->num_inputs
; ++index
) {
711 unsigned attr
= stvp
->index_to_input
[index
];
712 if (attr
== ST_DOUBLE_ATTRIB_PLACEHOLDER
)
714 vpv
->vert_attrib_mask
|= 1u << attr
;
717 /* insert into list */
718 vpv
->next
= stvp
->variants
;
719 stvp
->variants
= vpv
;
728 * Translate a Mesa fragment shader into a TGSI shader.
731 st_translate_fragment_program(struct st_context
*st
,
732 struct st_fragment_program
*stfp
)
734 /* We have already compiled to NIR so just return */
735 if (stfp
->shader_program
) {
736 st_store_ir_in_disk_cache(st
, &stfp
->Base
, true);
740 ubyte outputMapping
[2 * FRAG_RESULT_MAX
];
741 ubyte inputMapping
[VARYING_SLOT_MAX
];
742 ubyte inputSlotToAttr
[VARYING_SLOT_MAX
];
743 ubyte interpMode
[PIPE_MAX_SHADER_INPUTS
]; /* XXX size? */
745 GLbitfield64 inputsRead
;
746 struct ureg_program
*ureg
;
748 GLboolean write_all
= GL_FALSE
;
750 ubyte input_semantic_name
[PIPE_MAX_SHADER_INPUTS
];
751 ubyte input_semantic_index
[PIPE_MAX_SHADER_INPUTS
];
752 uint fs_num_inputs
= 0;
754 ubyte fs_output_semantic_name
[PIPE_MAX_SHADER_OUTPUTS
];
755 ubyte fs_output_semantic_index
[PIPE_MAX_SHADER_OUTPUTS
];
756 uint fs_num_outputs
= 0;
758 memset(inputSlotToAttr
, ~0, sizeof(inputSlotToAttr
));
760 /* Non-GLSL programs: */
761 if (!stfp
->glsl_to_tgsi
&& !stfp
->shader_program
) {
762 _mesa_remove_output_reads(&stfp
->Base
, PROGRAM_OUTPUT
);
763 if (st
->ctx
->Const
.GLSLFragCoordIsSysVal
)
764 _mesa_program_fragment_position_to_sysval(&stfp
->Base
);
766 /* This determines which states will be updated when the assembly
769 * fragment.position and glDrawPixels always use constants.
771 stfp
->affected_states
= ST_NEW_FS_STATE
|
772 ST_NEW_SAMPLE_SHADING
|
776 /* Just set them for ATI_fs unconditionally. */
777 stfp
->affected_states
|= ST_NEW_FS_SAMPLER_VIEWS
|
781 if (stfp
->Base
.SamplersUsed
)
782 stfp
->affected_states
|= ST_NEW_FS_SAMPLER_VIEWS
|
788 bool use_nir
= PIPE_SHADER_IR_NIR
==
789 st
->pipe
->screen
->get_shader_param(st
->pipe
->screen
,
790 PIPE_SHADER_FRAGMENT
,
791 PIPE_SHADER_CAP_PREFERRED_IR
);
793 if (use_nir
&& !stfp
->ati_fs
) {
795 st_translate_prog_to_nir(st
, &stfp
->Base
, MESA_SHADER_FRAGMENT
);
797 if (stfp
->tgsi
.ir
.nir
)
798 ralloc_free(stfp
->tgsi
.ir
.nir
);
799 stfp
->tgsi
.type
= PIPE_SHADER_IR_NIR
;
800 stfp
->tgsi
.ir
.nir
= nir
;
801 stfp
->Base
.nir
= nir
;
806 * Convert Mesa program inputs to TGSI input register semantics.
808 inputsRead
= stfp
->Base
.info
.inputs_read
;
809 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
810 if ((inputsRead
& BITFIELD64_BIT(attr
)) != 0) {
811 const GLuint slot
= fs_num_inputs
++;
813 inputMapping
[attr
] = slot
;
814 inputSlotToAttr
[slot
] = attr
;
817 case VARYING_SLOT_POS
:
818 input_semantic_name
[slot
] = TGSI_SEMANTIC_POSITION
;
819 input_semantic_index
[slot
] = 0;
820 interpMode
[slot
] = TGSI_INTERPOLATE_LINEAR
;
822 case VARYING_SLOT_COL0
:
823 input_semantic_name
[slot
] = TGSI_SEMANTIC_COLOR
;
824 input_semantic_index
[slot
] = 0;
825 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
826 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_COLOR
;
828 case VARYING_SLOT_COL1
:
829 input_semantic_name
[slot
] = TGSI_SEMANTIC_COLOR
;
830 input_semantic_index
[slot
] = 1;
831 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
832 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_COLOR
;
834 case VARYING_SLOT_FOGC
:
835 input_semantic_name
[slot
] = TGSI_SEMANTIC_FOG
;
836 input_semantic_index
[slot
] = 0;
837 interpMode
[slot
] = TGSI_INTERPOLATE_PERSPECTIVE
;
839 case VARYING_SLOT_FACE
:
840 input_semantic_name
[slot
] = TGSI_SEMANTIC_FACE
;
841 input_semantic_index
[slot
] = 0;
842 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
844 case VARYING_SLOT_PRIMITIVE_ID
:
845 input_semantic_name
[slot
] = TGSI_SEMANTIC_PRIMID
;
846 input_semantic_index
[slot
] = 0;
847 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
849 case VARYING_SLOT_LAYER
:
850 input_semantic_name
[slot
] = TGSI_SEMANTIC_LAYER
;
851 input_semantic_index
[slot
] = 0;
852 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
854 case VARYING_SLOT_VIEWPORT
:
855 input_semantic_name
[slot
] = TGSI_SEMANTIC_VIEWPORT_INDEX
;
856 input_semantic_index
[slot
] = 0;
857 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
859 case VARYING_SLOT_CLIP_DIST0
:
860 input_semantic_name
[slot
] = TGSI_SEMANTIC_CLIPDIST
;
861 input_semantic_index
[slot
] = 0;
862 interpMode
[slot
] = TGSI_INTERPOLATE_PERSPECTIVE
;
864 case VARYING_SLOT_CLIP_DIST1
:
865 input_semantic_name
[slot
] = TGSI_SEMANTIC_CLIPDIST
;
866 input_semantic_index
[slot
] = 1;
867 interpMode
[slot
] = TGSI_INTERPOLATE_PERSPECTIVE
;
869 case VARYING_SLOT_CULL_DIST0
:
870 case VARYING_SLOT_CULL_DIST1
:
871 /* these should have been lowered by GLSL */
874 /* In most cases, there is nothing special about these
875 * inputs, so adopt a convention to use the generic
876 * semantic name and the mesa VARYING_SLOT_ number as the
879 * All that is required is that the vertex shader labels
880 * its own outputs similarly, and that the vertex shader
881 * generates at least every output required by the
882 * fragment shader plus fixed-function hardware (such as
885 * However, some drivers may need us to identify the PNTC and TEXi
886 * varyings if, for example, their capability to replace them with
887 * sprite coordinates is limited.
889 case VARYING_SLOT_PNTC
:
890 if (st
->needs_texcoord_semantic
) {
891 input_semantic_name
[slot
] = TGSI_SEMANTIC_PCOORD
;
892 input_semantic_index
[slot
] = 0;
893 interpMode
[slot
] = TGSI_INTERPOLATE_LINEAR
;
897 case VARYING_SLOT_TEX0
:
898 case VARYING_SLOT_TEX1
:
899 case VARYING_SLOT_TEX2
:
900 case VARYING_SLOT_TEX3
:
901 case VARYING_SLOT_TEX4
:
902 case VARYING_SLOT_TEX5
:
903 case VARYING_SLOT_TEX6
:
904 case VARYING_SLOT_TEX7
:
905 if (st
->needs_texcoord_semantic
) {
906 input_semantic_name
[slot
] = TGSI_SEMANTIC_TEXCOORD
;
907 input_semantic_index
[slot
] = attr
- VARYING_SLOT_TEX0
;
908 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
909 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_PERSPECTIVE
;
913 case VARYING_SLOT_VAR0
:
915 /* Semantic indices should be zero-based because drivers may choose
916 * to assign a fixed slot determined by that index.
917 * This is useful because ARB_separate_shader_objects uses location
918 * qualifiers for linkage, and if the semantic index corresponds to
919 * these locations, linkage passes in the driver become unecessary.
921 * If needs_texcoord_semantic is true, no semantic indices will be
922 * consumed for the TEXi varyings, and we can base the locations of
923 * the user varyings on VAR0. Otherwise, we use TEX0 as base index.
925 assert(attr
>= VARYING_SLOT_VAR0
|| attr
== VARYING_SLOT_PNTC
||
926 (attr
>= VARYING_SLOT_TEX0
&& attr
<= VARYING_SLOT_TEX7
));
927 input_semantic_name
[slot
] = TGSI_SEMANTIC_GENERIC
;
928 input_semantic_index
[slot
] = st_get_generic_varying_index(st
, attr
);
929 if (attr
== VARYING_SLOT_PNTC
)
930 interpMode
[slot
] = TGSI_INTERPOLATE_LINEAR
;
932 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
933 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_PERSPECTIVE
;
939 inputMapping
[attr
] = -1;
944 * Semantics and mapping for outputs
946 GLbitfield64 outputsWritten
= stfp
->Base
.info
.outputs_written
;
948 /* if z is written, emit that first */
949 if (outputsWritten
& BITFIELD64_BIT(FRAG_RESULT_DEPTH
)) {
950 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_POSITION
;
951 fs_output_semantic_index
[fs_num_outputs
] = 0;
952 outputMapping
[FRAG_RESULT_DEPTH
] = fs_num_outputs
;
954 outputsWritten
&= ~(1 << FRAG_RESULT_DEPTH
);
957 if (outputsWritten
& BITFIELD64_BIT(FRAG_RESULT_STENCIL
)) {
958 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_STENCIL
;
959 fs_output_semantic_index
[fs_num_outputs
] = 0;
960 outputMapping
[FRAG_RESULT_STENCIL
] = fs_num_outputs
;
962 outputsWritten
&= ~(1 << FRAG_RESULT_STENCIL
);
965 if (outputsWritten
& BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK
)) {
966 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_SAMPLEMASK
;
967 fs_output_semantic_index
[fs_num_outputs
] = 0;
968 outputMapping
[FRAG_RESULT_SAMPLE_MASK
] = fs_num_outputs
;
970 outputsWritten
&= ~(1 << FRAG_RESULT_SAMPLE_MASK
);
973 /* handle remaining outputs (color) */
974 for (attr
= 0; attr
< ARRAY_SIZE(outputMapping
); attr
++) {
975 const GLbitfield64 written
= attr
< FRAG_RESULT_MAX
? outputsWritten
:
976 stfp
->Base
.SecondaryOutputsWritten
;
977 const unsigned loc
= attr
% FRAG_RESULT_MAX
;
979 if (written
& BITFIELD64_BIT(loc
)) {
981 case FRAG_RESULT_DEPTH
:
982 case FRAG_RESULT_STENCIL
:
983 case FRAG_RESULT_SAMPLE_MASK
:
987 case FRAG_RESULT_COLOR
:
988 write_all
= GL_TRUE
; /* fallthrough */
991 assert(loc
== FRAG_RESULT_COLOR
||
992 (FRAG_RESULT_DATA0
<= loc
&& loc
< FRAG_RESULT_MAX
));
994 index
= (loc
== FRAG_RESULT_COLOR
) ? 0 : (loc
- FRAG_RESULT_DATA0
);
996 if (attr
>= FRAG_RESULT_MAX
) {
997 /* Secondary color for dual source blending. */
1002 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_COLOR
;
1003 fs_output_semantic_index
[fs_num_outputs
] = index
;
1004 outputMapping
[attr
] = fs_num_outputs
;
1013 ureg
= ureg_create_with_screen(PIPE_SHADER_FRAGMENT
, st
->pipe
->screen
);
1017 if (ST_DEBUG
& DEBUG_MESA
) {
1018 _mesa_print_program(&stfp
->Base
);
1019 _mesa_print_program_parameters(st
->ctx
, &stfp
->Base
);
1022 if (write_all
== GL_TRUE
)
1023 ureg_property(ureg
, TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS
, 1);
1025 if (stfp
->Base
.info
.fs
.depth_layout
!= FRAG_DEPTH_LAYOUT_NONE
) {
1026 switch (stfp
->Base
.info
.fs
.depth_layout
) {
1027 case FRAG_DEPTH_LAYOUT_ANY
:
1028 ureg_property(ureg
, TGSI_PROPERTY_FS_DEPTH_LAYOUT
,
1029 TGSI_FS_DEPTH_LAYOUT_ANY
);
1031 case FRAG_DEPTH_LAYOUT_GREATER
:
1032 ureg_property(ureg
, TGSI_PROPERTY_FS_DEPTH_LAYOUT
,
1033 TGSI_FS_DEPTH_LAYOUT_GREATER
);
1035 case FRAG_DEPTH_LAYOUT_LESS
:
1036 ureg_property(ureg
, TGSI_PROPERTY_FS_DEPTH_LAYOUT
,
1037 TGSI_FS_DEPTH_LAYOUT_LESS
);
1039 case FRAG_DEPTH_LAYOUT_UNCHANGED
:
1040 ureg_property(ureg
, TGSI_PROPERTY_FS_DEPTH_LAYOUT
,
1041 TGSI_FS_DEPTH_LAYOUT_UNCHANGED
);
1048 if (stfp
->glsl_to_tgsi
) {
1049 st_translate_program(st
->ctx
,
1050 PIPE_SHADER_FRAGMENT
,
1058 input_semantic_name
,
1059 input_semantic_index
,
1064 fs_output_semantic_name
,
1065 fs_output_semantic_index
);
1067 free_glsl_to_tgsi_visitor(stfp
->glsl_to_tgsi
);
1068 } else if (stfp
->ati_fs
)
1069 st_translate_atifs_program(ureg
,
1075 input_semantic_name
,
1076 input_semantic_index
,
1081 fs_output_semantic_name
,
1082 fs_output_semantic_index
);
1084 st_translate_mesa_program(st
->ctx
,
1085 PIPE_SHADER_FRAGMENT
,
1091 input_semantic_name
,
1092 input_semantic_index
,
1097 fs_output_semantic_name
,
1098 fs_output_semantic_index
);
1100 stfp
->tgsi
.tokens
= ureg_get_tokens(ureg
, &stfp
->num_tgsi_tokens
);
1103 if (stfp
->glsl_to_tgsi
) {
1104 stfp
->glsl_to_tgsi
= NULL
;
1105 st_store_ir_in_disk_cache(st
, &stfp
->Base
, false);
1108 return stfp
->tgsi
.tokens
!= NULL
;
1111 static struct st_fp_variant
*
1112 st_create_fp_variant(struct st_context
*st
,
1113 struct st_fragment_program
*stfp
,
1114 const struct st_fp_variant_key
*key
)
1116 struct pipe_context
*pipe
= st
->pipe
;
1117 struct st_fp_variant
*variant
= CALLOC_STRUCT(st_fp_variant
);
1118 struct pipe_shader_state tgsi
= {0};
1119 struct gl_program_parameter_list
*params
= stfp
->Base
.Parameters
;
1120 static const gl_state_index16 texcoord_state
[STATE_LENGTH
] =
1121 { STATE_INTERNAL
, STATE_CURRENT_ATTRIB
, VERT_ATTRIB_TEX0
};
1122 static const gl_state_index16 scale_state
[STATE_LENGTH
] =
1123 { STATE_INTERNAL
, STATE_PT_SCALE
};
1124 static const gl_state_index16 bias_state
[STATE_LENGTH
] =
1125 { STATE_INTERNAL
, STATE_PT_BIAS
};
1130 if (stfp
->tgsi
.type
== PIPE_SHADER_IR_NIR
) {
1131 tgsi
.type
= PIPE_SHADER_IR_NIR
;
1132 tgsi
.ir
.nir
= nir_shader_clone(NULL
, stfp
->tgsi
.ir
.nir
);
1134 if (key
->clamp_color
)
1135 NIR_PASS_V(tgsi
.ir
.nir
, nir_lower_clamp_color_outputs
);
1137 if (key
->persample_shading
) {
1138 nir_shader
*shader
= tgsi
.ir
.nir
;
1139 nir_foreach_variable(var
, &shader
->inputs
)
1140 var
->data
.sample
= true;
1143 assert(!(key
->bitmap
&& key
->drawpixels
));
1147 nir_lower_bitmap_options options
= {0};
1149 variant
->bitmap_sampler
= ffs(~stfp
->Base
.SamplersUsed
) - 1;
1150 options
.sampler
= variant
->bitmap_sampler
;
1151 options
.swizzle_xxxx
= st
->bitmap
.tex_format
== PIPE_FORMAT_R8_UNORM
;
1153 NIR_PASS_V(tgsi
.ir
.nir
, nir_lower_bitmap
, &options
);
1156 /* glDrawPixels (color only) */
1157 if (key
->drawpixels
) {
1158 nir_lower_drawpixels_options options
= {{0}};
1159 unsigned samplers_used
= stfp
->Base
.SamplersUsed
;
1161 /* Find the first unused slot. */
1162 variant
->drawpix_sampler
= ffs(~samplers_used
) - 1;
1163 options
.drawpix_sampler
= variant
->drawpix_sampler
;
1164 samplers_used
|= (1 << variant
->drawpix_sampler
);
1166 options
.pixel_maps
= key
->pixelMaps
;
1167 if (key
->pixelMaps
) {
1168 variant
->pixelmap_sampler
= ffs(~samplers_used
) - 1;
1169 options
.pixelmap_sampler
= variant
->pixelmap_sampler
;
1172 options
.scale_and_bias
= key
->scaleAndBias
;
1173 if (key
->scaleAndBias
) {
1174 _mesa_add_state_reference(params
, scale_state
);
1175 memcpy(options
.scale_state_tokens
, scale_state
,
1176 sizeof(options
.scale_state_tokens
));
1177 _mesa_add_state_reference(params
, bias_state
);
1178 memcpy(options
.bias_state_tokens
, bias_state
,
1179 sizeof(options
.bias_state_tokens
));
1182 _mesa_add_state_reference(params
, texcoord_state
);
1183 memcpy(options
.texcoord_state_tokens
, texcoord_state
,
1184 sizeof(options
.texcoord_state_tokens
));
1186 NIR_PASS_V(tgsi
.ir
.nir
, nir_lower_drawpixels
, &options
);
1189 if (unlikely(key
->external
.lower_nv12
|| key
->external
.lower_iyuv
||
1190 key
->external
.lower_xy_uxvx
|| key
->external
.lower_yx_xuxv
||
1191 key
->external
.lower_ayuv
|| key
->external
.lower_xyuv
)) {
1192 nir_lower_tex_options options
= {0};
1193 options
.lower_y_uv_external
= key
->external
.lower_nv12
;
1194 options
.lower_y_u_v_external
= key
->external
.lower_iyuv
;
1195 options
.lower_xy_uxvx_external
= key
->external
.lower_xy_uxvx
;
1196 options
.lower_yx_xuxv_external
= key
->external
.lower_yx_xuxv
;
1197 options
.lower_ayuv_external
= key
->external
.lower_ayuv
;
1198 options
.lower_xyuv_external
= key
->external
.lower_xyuv
;
1199 NIR_PASS_V(tgsi
.ir
.nir
, nir_lower_tex
, &options
);
1202 st_finalize_nir(st
, &stfp
->Base
, stfp
->shader_program
, tgsi
.ir
.nir
);
1204 if (unlikely(key
->external
.lower_nv12
|| key
->external
.lower_iyuv
||
1205 key
->external
.lower_xy_uxvx
|| key
->external
.lower_yx_xuxv
)) {
1206 /* This pass needs to happen *after* nir_lower_sampler */
1207 NIR_PASS_V(tgsi
.ir
.nir
, st_nir_lower_tex_src_plane
,
1208 ~stfp
->Base
.SamplersUsed
,
1209 key
->external
.lower_nv12
|| key
->external
.lower_xy_uxvx
||
1210 key
->external
.lower_yx_xuxv
,
1211 key
->external
.lower_iyuv
);
1214 /* Some of the lowering above may have introduced new varyings */
1215 nir_shader_gather_info(tgsi
.ir
.nir
,
1216 nir_shader_get_entrypoint(tgsi
.ir
.nir
));
1218 variant
->driver_shader
= pipe
->create_fs_state(pipe
, &tgsi
);
1219 variant
->key
= *key
;
1224 tgsi
.tokens
= stfp
->tgsi
.tokens
;
1226 assert(!(key
->bitmap
&& key
->drawpixels
));
1228 /* Fix texture targets and add fog for ATI_fs */
1230 const struct tgsi_token
*tokens
= st_fixup_atifs(tgsi
.tokens
, key
);
1233 tgsi
.tokens
= tokens
;
1235 fprintf(stderr
, "mesa: cannot post-process ATI_fs\n");
1238 /* Emulate features. */
1239 if (key
->clamp_color
|| key
->persample_shading
) {
1240 const struct tgsi_token
*tokens
;
1242 (key
->clamp_color
? TGSI_EMU_CLAMP_COLOR_OUTPUTS
: 0) |
1243 (key
->persample_shading
? TGSI_EMU_FORCE_PERSAMPLE_INTERP
: 0);
1245 tokens
= tgsi_emulate(tgsi
.tokens
, flags
);
1248 if (tgsi
.tokens
!= stfp
->tgsi
.tokens
)
1249 tgsi_free_tokens(tgsi
.tokens
);
1250 tgsi
.tokens
= tokens
;
1252 fprintf(stderr
, "mesa: cannot emulate deprecated features\n");
1257 const struct tgsi_token
*tokens
;
1259 variant
->bitmap_sampler
= ffs(~stfp
->Base
.SamplersUsed
) - 1;
1261 tokens
= st_get_bitmap_shader(tgsi
.tokens
,
1262 st
->internal_target
,
1263 variant
->bitmap_sampler
,
1264 st
->needs_texcoord_semantic
,
1265 st
->bitmap
.tex_format
==
1266 PIPE_FORMAT_R8_UNORM
);
1269 if (tgsi
.tokens
!= stfp
->tgsi
.tokens
)
1270 tgsi_free_tokens(tgsi
.tokens
);
1271 tgsi
.tokens
= tokens
;
1273 fprintf(stderr
, "mesa: cannot create a shader for glBitmap\n");
1276 /* glDrawPixels (color only) */
1277 if (key
->drawpixels
) {
1278 const struct tgsi_token
*tokens
;
1279 unsigned scale_const
= 0, bias_const
= 0, texcoord_const
= 0;
1281 /* Find the first unused slot. */
1282 variant
->drawpix_sampler
= ffs(~stfp
->Base
.SamplersUsed
) - 1;
1284 if (key
->pixelMaps
) {
1285 unsigned samplers_used
= stfp
->Base
.SamplersUsed
|
1286 (1 << variant
->drawpix_sampler
);
1288 variant
->pixelmap_sampler
= ffs(~samplers_used
) - 1;
1291 if (key
->scaleAndBias
) {
1292 scale_const
= _mesa_add_state_reference(params
, scale_state
);
1293 bias_const
= _mesa_add_state_reference(params
, bias_state
);
1296 texcoord_const
= _mesa_add_state_reference(params
, texcoord_state
);
1298 tokens
= st_get_drawpix_shader(tgsi
.tokens
,
1299 st
->needs_texcoord_semantic
,
1300 key
->scaleAndBias
, scale_const
,
1301 bias_const
, key
->pixelMaps
,
1302 variant
->drawpix_sampler
,
1303 variant
->pixelmap_sampler
,
1304 texcoord_const
, st
->internal_target
);
1307 if (tgsi
.tokens
!= stfp
->tgsi
.tokens
)
1308 tgsi_free_tokens(tgsi
.tokens
);
1309 tgsi
.tokens
= tokens
;
1311 fprintf(stderr
, "mesa: cannot create a shader for glDrawPixels\n");
1314 if (unlikely(key
->external
.lower_nv12
|| key
->external
.lower_iyuv
||
1315 key
->external
.lower_xy_uxvx
|| key
->external
.lower_yx_xuxv
)) {
1316 const struct tgsi_token
*tokens
;
1318 /* samplers inserted would conflict, but this should be unpossible: */
1319 assert(!(key
->bitmap
|| key
->drawpixels
));
1321 tokens
= st_tgsi_lower_yuv(tgsi
.tokens
,
1322 ~stfp
->Base
.SamplersUsed
,
1323 key
->external
.lower_nv12
||
1324 key
->external
.lower_xy_uxvx
||
1325 key
->external
.lower_yx_xuxv
,
1326 key
->external
.lower_iyuv
);
1328 if (tgsi
.tokens
!= stfp
->tgsi
.tokens
)
1329 tgsi_free_tokens(tgsi
.tokens
);
1330 tgsi
.tokens
= tokens
;
1332 fprintf(stderr
, "mesa: cannot create a shader for samplerExternalOES\n");
1336 if (key
->lower_depth_clamp
) {
1337 unsigned depth_range_const
= _mesa_add_state_reference(params
, depth_range_state
);
1339 const struct tgsi_token
*tokens
;
1340 tokens
= st_tgsi_lower_depth_clamp_fs(tgsi
.tokens
, depth_range_const
);
1341 if (tgsi
.tokens
!= stfp
->tgsi
.tokens
)
1342 tgsi_free_tokens(tgsi
.tokens
);
1343 tgsi
.tokens
= tokens
;
1346 if (ST_DEBUG
& DEBUG_TGSI
) {
1347 tgsi_dump(tgsi
.tokens
, 0);
1351 /* fill in variant */
1352 variant
->driver_shader
= pipe
->create_fs_state(pipe
, &tgsi
);
1353 variant
->key
= *key
;
1355 if (tgsi
.tokens
!= stfp
->tgsi
.tokens
)
1356 tgsi_free_tokens(tgsi
.tokens
);
1361 * Translate fragment program if needed.
1363 struct st_fp_variant
*
1364 st_get_fp_variant(struct st_context
*st
,
1365 struct st_fragment_program
*stfp
,
1366 const struct st_fp_variant_key
*key
)
1368 struct st_fp_variant
*fpv
;
1370 /* Search for existing variant */
1371 for (fpv
= stfp
->variants
; fpv
; fpv
= fpv
->next
) {
1372 if (memcmp(&fpv
->key
, key
, sizeof(*key
)) == 0) {
1379 fpv
= st_create_fp_variant(st
, stfp
, key
);
1381 if (key
->bitmap
|| key
->drawpixels
) {
1382 /* Regular variants should always come before the
1383 * bitmap & drawpixels variants, (unless there
1384 * are no regular variants) so that
1385 * st_update_fp can take a fast path when
1386 * shader_has_one_variant is set.
1388 if (!stfp
->variants
) {
1389 stfp
->variants
= fpv
;
1391 /* insert into list after the first one */
1392 fpv
->next
= stfp
->variants
->next
;
1393 stfp
->variants
->next
= fpv
;
1396 /* insert into list */
1397 fpv
->next
= stfp
->variants
;
1398 stfp
->variants
= fpv
;
1407 * Update stream-output info for GS/TCS/TES. Normally this is done in
1408 * st_translate_program_common() but that is not called for glsl_to_nir
1412 st_translate_program_stream_output(struct gl_program
*prog
,
1413 struct pipe_stream_output_info
*stream_output
)
1415 if (!prog
->sh
.LinkedTransformFeedback
)
1418 ubyte outputMapping
[VARYING_SLOT_TESS_MAX
];
1420 uint num_outputs
= 0;
1422 memset(outputMapping
, 0, sizeof(outputMapping
));
1425 * Determine number of outputs, the (default) output register
1426 * mapping and the semantic information for each output.
1428 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
1429 if (prog
->info
.outputs_written
& BITFIELD64_BIT(attr
)) {
1430 GLuint slot
= num_outputs
++;
1432 outputMapping
[attr
] = slot
;
1436 st_translate_stream_output_info(prog
->sh
.LinkedTransformFeedback
,
1442 * Translate a program. This is common code for geometry and tessellation
1446 st_translate_common_program(struct st_context
*st
,
1447 struct st_common_program
*stcp
)
1449 /* We have already compiled to NIR so just return */
1450 if (stcp
->shader_program
) {
1452 st_finalize_nir(st
, &stcp
->Base
, stcp
->shader_program
,
1454 if (stcp
->Base
.info
.stage
== MESA_SHADER_TESS_EVAL
||
1455 stcp
->Base
.info
.stage
== MESA_SHADER_GEOMETRY
) {
1456 st_translate_program_stream_output(&stcp
->Base
,
1457 &stcp
->tgsi
.stream_output
);
1459 st_store_ir_in_disk_cache(st
, &stcp
->Base
, true);
1463 struct gl_program
*prog
= &stcp
->Base
;
1464 enum pipe_shader_type stage
=
1465 pipe_shader_type_from_mesa(stcp
->Base
.info
.stage
);
1466 struct ureg_program
*ureg
= ureg_create_with_screen(stage
, st
->pipe
->screen
);
1472 case PIPE_SHADER_TESS_CTRL
:
1473 ureg_property(ureg
, TGSI_PROPERTY_TCS_VERTICES_OUT
,
1474 stcp
->Base
.info
.tess
.tcs_vertices_out
);
1477 case PIPE_SHADER_TESS_EVAL
:
1478 if (stcp
->Base
.info
.tess
.primitive_mode
== GL_ISOLINES
)
1479 ureg_property(ureg
, TGSI_PROPERTY_TES_PRIM_MODE
, GL_LINES
);
1481 ureg_property(ureg
, TGSI_PROPERTY_TES_PRIM_MODE
,
1482 stcp
->Base
.info
.tess
.primitive_mode
);
1484 STATIC_ASSERT((TESS_SPACING_EQUAL
+ 1) % 3 == PIPE_TESS_SPACING_EQUAL
);
1485 STATIC_ASSERT((TESS_SPACING_FRACTIONAL_ODD
+ 1) % 3 ==
1486 PIPE_TESS_SPACING_FRACTIONAL_ODD
);
1487 STATIC_ASSERT((TESS_SPACING_FRACTIONAL_EVEN
+ 1) % 3 ==
1488 PIPE_TESS_SPACING_FRACTIONAL_EVEN
);
1490 ureg_property(ureg
, TGSI_PROPERTY_TES_SPACING
,
1491 (stcp
->Base
.info
.tess
.spacing
+ 1) % 3);
1493 ureg_property(ureg
, TGSI_PROPERTY_TES_VERTEX_ORDER_CW
,
1494 !stcp
->Base
.info
.tess
.ccw
);
1495 ureg_property(ureg
, TGSI_PROPERTY_TES_POINT_MODE
,
1496 stcp
->Base
.info
.tess
.point_mode
);
1499 case PIPE_SHADER_GEOMETRY
:
1500 ureg_property(ureg
, TGSI_PROPERTY_GS_INPUT_PRIM
,
1501 stcp
->Base
.info
.gs
.input_primitive
);
1502 ureg_property(ureg
, TGSI_PROPERTY_GS_OUTPUT_PRIM
,
1503 stcp
->Base
.info
.gs
.output_primitive
);
1504 ureg_property(ureg
, TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES
,
1505 stcp
->Base
.info
.gs
.vertices_out
);
1506 ureg_property(ureg
, TGSI_PROPERTY_GS_INVOCATIONS
,
1507 stcp
->Base
.info
.gs
.invocations
);
1514 ubyte inputSlotToAttr
[VARYING_SLOT_TESS_MAX
];
1515 ubyte inputMapping
[VARYING_SLOT_TESS_MAX
];
1516 ubyte outputMapping
[VARYING_SLOT_TESS_MAX
];
1519 ubyte input_semantic_name
[PIPE_MAX_SHADER_INPUTS
];
1520 ubyte input_semantic_index
[PIPE_MAX_SHADER_INPUTS
];
1521 uint num_inputs
= 0;
1523 ubyte output_semantic_name
[PIPE_MAX_SHADER_OUTPUTS
];
1524 ubyte output_semantic_index
[PIPE_MAX_SHADER_OUTPUTS
];
1525 uint num_outputs
= 0;
1529 memset(inputSlotToAttr
, 0, sizeof(inputSlotToAttr
));
1530 memset(inputMapping
, 0, sizeof(inputMapping
));
1531 memset(outputMapping
, 0, sizeof(outputMapping
));
1532 memset(&stcp
->tgsi
, 0, sizeof(stcp
->tgsi
));
1534 if (prog
->info
.clip_distance_array_size
)
1535 ureg_property(ureg
, TGSI_PROPERTY_NUM_CLIPDIST_ENABLED
,
1536 prog
->info
.clip_distance_array_size
);
1537 if (prog
->info
.cull_distance_array_size
)
1538 ureg_property(ureg
, TGSI_PROPERTY_NUM_CULLDIST_ENABLED
,
1539 prog
->info
.cull_distance_array_size
);
1542 * Convert Mesa program inputs to TGSI input register semantics.
1544 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
1545 if ((prog
->info
.inputs_read
& BITFIELD64_BIT(attr
)) == 0)
1548 unsigned slot
= num_inputs
++;
1550 inputMapping
[attr
] = slot
;
1551 inputSlotToAttr
[slot
] = attr
;
1553 unsigned semantic_name
, semantic_index
;
1554 tgsi_get_gl_varying_semantic(attr
, st
->needs_texcoord_semantic
,
1555 &semantic_name
, &semantic_index
);
1556 input_semantic_name
[slot
] = semantic_name
;
1557 input_semantic_index
[slot
] = semantic_index
;
1560 /* Also add patch inputs. */
1561 for (attr
= 0; attr
< 32; attr
++) {
1562 if (prog
->info
.patch_inputs_read
& (1u << attr
)) {
1563 GLuint slot
= num_inputs
++;
1564 GLuint patch_attr
= VARYING_SLOT_PATCH0
+ attr
;
1566 inputMapping
[patch_attr
] = slot
;
1567 inputSlotToAttr
[slot
] = patch_attr
;
1568 input_semantic_name
[slot
] = TGSI_SEMANTIC_PATCH
;
1569 input_semantic_index
[slot
] = attr
;
1573 /* initialize output semantics to defaults */
1574 for (i
= 0; i
< PIPE_MAX_SHADER_OUTPUTS
; i
++) {
1575 output_semantic_name
[i
] = TGSI_SEMANTIC_GENERIC
;
1576 output_semantic_index
[i
] = 0;
1580 * Determine number of outputs, the (default) output register
1581 * mapping and the semantic information for each output.
1583 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
1584 if (prog
->info
.outputs_written
& BITFIELD64_BIT(attr
)) {
1585 GLuint slot
= num_outputs
++;
1587 outputMapping
[attr
] = slot
;
1589 unsigned semantic_name
, semantic_index
;
1590 tgsi_get_gl_varying_semantic(attr
, st
->needs_texcoord_semantic
,
1591 &semantic_name
, &semantic_index
);
1592 output_semantic_name
[slot
] = semantic_name
;
1593 output_semantic_index
[slot
] = semantic_index
;
1597 /* Also add patch outputs. */
1598 for (attr
= 0; attr
< 32; attr
++) {
1599 if (prog
->info
.patch_outputs_written
& (1u << attr
)) {
1600 GLuint slot
= num_outputs
++;
1601 GLuint patch_attr
= VARYING_SLOT_PATCH0
+ attr
;
1603 outputMapping
[patch_attr
] = slot
;
1604 output_semantic_name
[slot
] = TGSI_SEMANTIC_PATCH
;
1605 output_semantic_index
[slot
] = attr
;
1609 st_translate_program(st
->ctx
,
1618 input_semantic_name
,
1619 input_semantic_index
,
1624 output_semantic_name
,
1625 output_semantic_index
);
1627 stcp
->tgsi
.tokens
= ureg_get_tokens(ureg
, &stcp
->num_tgsi_tokens
);
1631 st_translate_stream_output_info(prog
->sh
.LinkedTransformFeedback
,
1633 &stcp
->tgsi
.stream_output
);
1635 st_store_ir_in_disk_cache(st
, prog
, false);
1637 if ((ST_DEBUG
& DEBUG_TGSI
) && (ST_DEBUG
& DEBUG_MESA
)) {
1638 _mesa_print_program(prog
);
1642 if (ST_DEBUG
& DEBUG_TGSI
) {
1643 tgsi_dump(stcp
->tgsi
.tokens
, 0);
1647 free_glsl_to_tgsi_visitor(stcp
->glsl_to_tgsi
);
1648 stcp
->glsl_to_tgsi
= NULL
;
1654 * Get/create a basic program variant.
1656 struct st_basic_variant
*
1657 st_get_basic_variant(struct st_context
*st
,
1658 struct st_common_program
*prog
,
1659 const struct st_basic_variant_key
*key
)
1661 struct pipe_context
*pipe
= st
->pipe
;
1662 struct st_basic_variant
*v
;
1663 struct pipe_shader_state tgsi
= {0};
1665 /* Search for existing variant */
1666 for (v
= prog
->variants
; v
; v
= v
->next
) {
1667 if (memcmp(&v
->key
, key
, sizeof(*key
)) == 0) {
1674 v
= CALLOC_STRUCT(st_basic_variant
);
1677 if (prog
->tgsi
.type
== PIPE_SHADER_IR_NIR
) {
1678 tgsi
.type
= PIPE_SHADER_IR_NIR
;
1679 tgsi
.ir
.nir
= nir_shader_clone(NULL
, prog
->tgsi
.ir
.nir
);
1681 if (key
->clamp_color
)
1682 NIR_PASS_V(tgsi
.ir
.nir
, nir_lower_clamp_color_outputs
);
1684 tgsi
.stream_output
= prog
->tgsi
.stream_output
;
1686 if (key
->lower_depth_clamp
) {
1687 struct gl_program_parameter_list
*params
= prog
->Base
.Parameters
;
1689 unsigned depth_range_const
=
1690 _mesa_add_state_reference(params
, depth_range_state
);
1692 const struct tgsi_token
*tokens
;
1694 st_tgsi_lower_depth_clamp(prog
->tgsi
.tokens
,
1696 key
->clip_negative_one_to_one
);
1698 if (tokens
!= prog
->tgsi
.tokens
)
1699 tgsi_free_tokens(prog
->tgsi
.tokens
);
1701 prog
->tgsi
.tokens
= tokens
;
1702 prog
->num_tgsi_tokens
= tgsi_num_tokens(tokens
);
1706 /* fill in new variant */
1707 switch (prog
->Base
.info
.stage
) {
1708 case MESA_SHADER_TESS_CTRL
:
1709 v
->driver_shader
= pipe
->create_tcs_state(pipe
, &tgsi
);
1711 case MESA_SHADER_TESS_EVAL
:
1712 v
->driver_shader
= pipe
->create_tes_state(pipe
, &tgsi
);
1714 case MESA_SHADER_GEOMETRY
:
1715 v
->driver_shader
= pipe
->create_gs_state(pipe
, &tgsi
);
1717 case MESA_SHADER_COMPUTE
: {
1718 struct pipe_compute_state cs
= {0};
1719 cs
.ir_type
= tgsi
.type
;
1720 cs
.req_local_mem
= prog
->Base
.info
.cs
.shared_size
;
1722 if (tgsi
.type
== PIPE_SHADER_IR_NIR
)
1723 cs
.prog
= tgsi
.ir
.nir
;
1725 cs
.prog
= tgsi
.tokens
;
1727 v
->driver_shader
= pipe
->create_compute_state(pipe
, &cs
);
1731 assert(!"unhandled shader type");
1738 /* insert into list */
1739 v
->next
= prog
->variants
;
1749 * Vert/Geom/Frag programs have per-context variants. Free all the
1750 * variants attached to the given program which match the given context.
1753 destroy_program_variants(struct st_context
*st
, struct gl_program
*target
)
1755 if (!target
|| target
== &_mesa_DummyProgram
)
1758 switch (target
->Target
) {
1759 case GL_VERTEX_PROGRAM_ARB
:
1761 struct st_vertex_program
*stvp
= (struct st_vertex_program
*) target
;
1762 struct st_vp_variant
*vpv
, **prevPtr
= &stvp
->variants
;
1764 for (vpv
= stvp
->variants
; vpv
; ) {
1765 struct st_vp_variant
*next
= vpv
->next
;
1766 if (vpv
->key
.st
== st
) {
1767 /* unlink from list */
1769 /* destroy this variant */
1770 delete_vp_variant(st
, vpv
);
1773 prevPtr
= &vpv
->next
;
1779 case GL_FRAGMENT_PROGRAM_ARB
:
1781 struct st_fragment_program
*stfp
=
1782 (struct st_fragment_program
*) target
;
1783 struct st_fp_variant
*fpv
, **prevPtr
= &stfp
->variants
;
1785 for (fpv
= stfp
->variants
; fpv
; ) {
1786 struct st_fp_variant
*next
= fpv
->next
;
1787 if (fpv
->key
.st
== st
) {
1788 /* unlink from list */
1790 /* destroy this variant */
1791 delete_fp_variant(st
, fpv
);
1794 prevPtr
= &fpv
->next
;
1800 case GL_GEOMETRY_PROGRAM_NV
:
1801 case GL_TESS_CONTROL_PROGRAM_NV
:
1802 case GL_TESS_EVALUATION_PROGRAM_NV
:
1803 case GL_COMPUTE_PROGRAM_NV
:
1805 struct st_common_program
*p
= st_common_program(target
);
1806 struct st_basic_variant
*v
, **prevPtr
= &p
->variants
;
1808 for (v
= p
->variants
; v
; ) {
1809 struct st_basic_variant
*next
= v
->next
;
1810 if (v
->key
.st
== st
) {
1811 /* unlink from list */
1813 /* destroy this variant */
1814 delete_basic_variant(st
, v
, target
->Target
);
1824 _mesa_problem(NULL
, "Unexpected program target 0x%x in "
1825 "destroy_program_variants_cb()", target
->Target
);
1831 * Callback for _mesa_HashWalk. Free all the shader's program variants
1832 * which match the given context.
1835 destroy_shader_program_variants_cb(GLuint key
, void *data
, void *userData
)
1837 struct st_context
*st
= (struct st_context
*) userData
;
1838 struct gl_shader
*shader
= (struct gl_shader
*) data
;
1840 switch (shader
->Type
) {
1841 case GL_SHADER_PROGRAM_MESA
:
1843 struct gl_shader_program
*shProg
= (struct gl_shader_program
*) data
;
1846 for (i
= 0; i
< ARRAY_SIZE(shProg
->_LinkedShaders
); i
++) {
1847 if (shProg
->_LinkedShaders
[i
])
1848 destroy_program_variants(st
, shProg
->_LinkedShaders
[i
]->Program
);
1852 case GL_VERTEX_SHADER
:
1853 case GL_FRAGMENT_SHADER
:
1854 case GL_GEOMETRY_SHADER
:
1855 case GL_TESS_CONTROL_SHADER
:
1856 case GL_TESS_EVALUATION_SHADER
:
1857 case GL_COMPUTE_SHADER
:
1866 * Callback for _mesa_HashWalk. Free all the program variants which match
1867 * the given context.
1870 destroy_program_variants_cb(GLuint key
, void *data
, void *userData
)
1872 struct st_context
*st
= (struct st_context
*) userData
;
1873 struct gl_program
*program
= (struct gl_program
*) data
;
1874 destroy_program_variants(st
, program
);
1879 * Walk over all shaders and programs to delete any variants which
1880 * belong to the given context.
1881 * This is called during context tear-down.
1884 st_destroy_program_variants(struct st_context
*st
)
1886 /* If shaders can be shared with other contexts, the last context will
1887 * call DeleteProgram on all shaders, releasing everything.
1889 if (st
->has_shareable_shaders
)
1892 /* ARB vert/frag program */
1893 _mesa_HashWalk(st
->ctx
->Shared
->Programs
,
1894 destroy_program_variants_cb
, st
);
1896 /* GLSL vert/frag/geom shaders */
1897 _mesa_HashWalk(st
->ctx
->Shared
->ShaderObjects
,
1898 destroy_shader_program_variants_cb
, st
);
1903 * For debugging, print/dump the current vertex program.
1906 st_print_current_vertex_program(void)
1908 GET_CURRENT_CONTEXT(ctx
);
1910 if (ctx
->VertexProgram
._Current
) {
1911 struct st_vertex_program
*stvp
=
1912 (struct st_vertex_program
*) ctx
->VertexProgram
._Current
;
1913 struct st_vp_variant
*stv
;
1915 debug_printf("Vertex program %u\n", stvp
->Base
.Id
);
1917 for (stv
= stvp
->variants
; stv
; stv
= stv
->next
) {
1918 debug_printf("variant %p\n", stv
);
1919 tgsi_dump(stv
->tgsi
.tokens
, 0);
1926 * Compile one shader variant.
1929 st_precompile_shader_variant(struct st_context
*st
,
1930 struct gl_program
*prog
)
1932 switch (prog
->Target
) {
1933 case GL_VERTEX_PROGRAM_ARB
: {
1934 struct st_vertex_program
*p
= (struct st_vertex_program
*)prog
;
1935 struct st_vp_variant_key key
;
1937 memset(&key
, 0, sizeof(key
));
1939 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
1940 st_get_vp_variant(st
, p
, &key
);
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
));
1950 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
1951 st_get_fp_variant(st
, p
, &key
);
1955 case GL_TESS_CONTROL_PROGRAM_NV
:
1956 case GL_TESS_EVALUATION_PROGRAM_NV
:
1957 case GL_GEOMETRY_PROGRAM_NV
:
1958 case GL_COMPUTE_PROGRAM_NV
: {
1959 struct st_common_program
*p
= st_common_program(prog
);
1960 struct st_basic_variant_key key
;
1962 memset(&key
, 0, sizeof(key
));
1964 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
1965 st_get_basic_variant(st
, p
, &key
);