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 st_translate_stream_output_info(struct gl_program
*prog
)
456 struct gl_transform_feedback_info
*info
= prog
->sh
.LinkedTransformFeedback
;
460 /* Determine the (default) output register mapping for each output. */
461 unsigned num_outputs
= 0;
462 ubyte output_mapping
[VARYING_SLOT_TESS_MAX
];
463 memset(output_mapping
, 0, sizeof(output_mapping
));
465 for (unsigned attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
466 if (prog
->info
.outputs_written
& BITFIELD64_BIT(attr
))
467 output_mapping
[attr
] = num_outputs
++;
470 /* Translate stream output info. */
471 struct pipe_stream_output_info
*so_info
= NULL
;
472 if (prog
->info
.stage
== MESA_SHADER_VERTEX
)
473 so_info
= &((struct st_vertex_program
*)prog
)->tgsi
.stream_output
;
475 so_info
= &((struct st_common_program
*)prog
)->tgsi
.stream_output
;
477 for (unsigned i
= 0; i
< info
->NumOutputs
; i
++) {
478 so_info
->output
[i
].register_index
=
479 output_mapping
[info
->Outputs
[i
].OutputRegister
];
480 so_info
->output
[i
].start_component
= info
->Outputs
[i
].ComponentOffset
;
481 so_info
->output
[i
].num_components
= info
->Outputs
[i
].NumComponents
;
482 so_info
->output
[i
].output_buffer
= info
->Outputs
[i
].OutputBuffer
;
483 so_info
->output
[i
].dst_offset
= info
->Outputs
[i
].DstOffset
;
484 so_info
->output
[i
].stream
= info
->Outputs
[i
].StreamId
;
487 for (unsigned i
= 0; i
< PIPE_MAX_SO_BUFFERS
; i
++) {
488 so_info
->stride
[i
] = info
->Buffers
[i
].Stride
;
490 so_info
->num_outputs
= info
->NumOutputs
;
494 * Translate a vertex program.
497 st_translate_vertex_program(struct st_context
*st
,
498 struct st_vertex_program
*stvp
)
500 struct ureg_program
*ureg
;
501 enum pipe_error error
;
502 unsigned num_outputs
= 0;
504 ubyte output_semantic_name
[VARYING_SLOT_MAX
] = {0};
505 ubyte output_semantic_index
[VARYING_SLOT_MAX
] = {0};
507 if (stvp
->Base
.arb
.IsPositionInvariant
)
508 _mesa_insert_mvp_code(st
->ctx
, &stvp
->Base
);
510 st_prepare_vertex_program(stvp
);
512 /* Get semantic names and indices. */
513 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
514 if (stvp
->Base
.info
.outputs_written
& BITFIELD64_BIT(attr
)) {
515 unsigned slot
= num_outputs
++;
516 unsigned semantic_name
, semantic_index
;
517 tgsi_get_gl_varying_semantic(attr
, st
->needs_texcoord_semantic
,
518 &semantic_name
, &semantic_index
);
519 output_semantic_name
[slot
] = semantic_name
;
520 output_semantic_index
[slot
] = semantic_index
;
523 /* pre-setup potentially unused edgeflag output */
524 output_semantic_name
[num_outputs
] = TGSI_SEMANTIC_EDGEFLAG
;
525 output_semantic_index
[num_outputs
] = 0;
528 if (!stvp
->glsl_to_tgsi
) {
529 _mesa_remove_output_reads(&stvp
->Base
, PROGRAM_OUTPUT
);
531 /* This determines which states will be updated when the assembly
534 stvp
->affected_states
= ST_NEW_VS_STATE
|
536 ST_NEW_VERTEX_ARRAYS
;
538 if (stvp
->Base
.Parameters
->NumParameters
)
539 stvp
->affected_states
|= ST_NEW_VS_CONSTANTS
;
541 /* No samplers are allowed in ARB_vp. */
544 ureg
= ureg_create_with_screen(PIPE_SHADER_VERTEX
, st
->pipe
->screen
);
548 if (stvp
->Base
.info
.clip_distance_array_size
)
549 ureg_property(ureg
, TGSI_PROPERTY_NUM_CLIPDIST_ENABLED
,
550 stvp
->Base
.info
.clip_distance_array_size
);
551 if (stvp
->Base
.info
.cull_distance_array_size
)
552 ureg_property(ureg
, TGSI_PROPERTY_NUM_CULLDIST_ENABLED
,
553 stvp
->Base
.info
.cull_distance_array_size
);
555 if (ST_DEBUG
& DEBUG_MESA
) {
556 _mesa_print_program(&stvp
->Base
);
557 _mesa_print_program_parameters(st
->ctx
, &stvp
->Base
);
561 if (stvp
->glsl_to_tgsi
) {
562 error
= st_translate_program(st
->ctx
,
569 stvp
->input_to_index
,
570 NULL
, /* inputSlotToAttr */
571 NULL
, /* input semantic name */
572 NULL
, /* input semantic index */
573 NULL
, /* interp mode */
576 stvp
->result_to_output
,
577 output_semantic_name
,
578 output_semantic_index
);
580 st_translate_stream_output_info(&stvp
->Base
);
582 free_glsl_to_tgsi_visitor(stvp
->glsl_to_tgsi
);
584 error
= st_translate_mesa_program(st
->ctx
,
590 stvp
->input_to_index
,
591 NULL
, /* input semantic name */
592 NULL
, /* input semantic index */
596 stvp
->result_to_output
,
597 output_semantic_name
,
598 output_semantic_index
);
601 debug_printf("%s: failed to translate Mesa program:\n", __func__
);
602 _mesa_print_program(&stvp
->Base
);
607 stvp
->tgsi
.tokens
= ureg_get_tokens(ureg
, &stvp
->num_tgsi_tokens
);
610 if (stvp
->glsl_to_tgsi
) {
611 stvp
->glsl_to_tgsi
= NULL
;
612 st_store_ir_in_disk_cache(st
, &stvp
->Base
, false);
615 bool use_nir
= PIPE_SHADER_IR_NIR
==
616 st
->pipe
->screen
->get_shader_param(st
->pipe
->screen
, PIPE_SHADER_VERTEX
,
617 PIPE_SHADER_CAP_PREFERRED_IR
);
621 st_translate_prog_to_nir(st
, &stvp
->Base
, MESA_SHADER_VERTEX
);
623 if (stvp
->tgsi
.ir
.nir
)
624 ralloc_free(stvp
->tgsi
.ir
.nir
);
625 stvp
->tgsi
.type
= PIPE_SHADER_IR_NIR
;
626 stvp
->tgsi
.ir
.nir
= nir
;
627 stvp
->Base
.nir
= nir
;
631 return stvp
->tgsi
.tokens
!= NULL
;
634 static const gl_state_index16 depth_range_state
[STATE_LENGTH
] =
635 { STATE_DEPTH_RANGE
};
637 static struct st_vp_variant
*
638 st_create_vp_variant(struct st_context
*st
,
639 struct st_vertex_program
*stvp
,
640 const struct st_vp_variant_key
*key
)
642 struct st_vp_variant
*vpv
= CALLOC_STRUCT(st_vp_variant
);
643 struct pipe_context
*pipe
= st
->pipe
;
644 struct gl_program_parameter_list
*params
= stvp
->Base
.Parameters
;
647 vpv
->tgsi
.stream_output
= stvp
->tgsi
.stream_output
;
648 vpv
->num_inputs
= stvp
->num_inputs
;
650 /* When generating a NIR program, we usually don't have TGSI tokens.
651 * However, we do create them for ARB_vertex_program / fixed-function VS
652 * programs which we may need to use with the draw module for legacy
653 * feedback/select emulation. If they exist, copy them.
655 if (stvp
->tgsi
.tokens
)
656 vpv
->tgsi
.tokens
= tgsi_dup_tokens(stvp
->tgsi
.tokens
);
658 if (stvp
->tgsi
.type
== PIPE_SHADER_IR_NIR
) {
659 vpv
->tgsi
.type
= PIPE_SHADER_IR_NIR
;
660 vpv
->tgsi
.ir
.nir
= nir_shader_clone(NULL
, stvp
->tgsi
.ir
.nir
);
661 if (key
->clamp_color
)
662 NIR_PASS_V(vpv
->tgsi
.ir
.nir
, nir_lower_clamp_color_outputs
);
663 if (key
->passthrough_edgeflags
) {
664 NIR_PASS_V(vpv
->tgsi
.ir
.nir
, nir_lower_passthrough_edgeflags
);
668 st_finalize_nir(st
, &stvp
->Base
, stvp
->shader_program
,
671 vpv
->driver_shader
= pipe
->create_vs_state(pipe
, &vpv
->tgsi
);
672 /* driver takes ownership of IR: */
673 vpv
->tgsi
.ir
.nir
= NULL
;
677 /* Emulate features. */
678 if (key
->clamp_color
|| key
->passthrough_edgeflags
) {
679 const struct tgsi_token
*tokens
;
681 (key
->clamp_color
? TGSI_EMU_CLAMP_COLOR_OUTPUTS
: 0) |
682 (key
->passthrough_edgeflags
? TGSI_EMU_PASSTHROUGH_EDGEFLAG
: 0);
684 tokens
= tgsi_emulate(vpv
->tgsi
.tokens
, flags
);
687 tgsi_free_tokens(vpv
->tgsi
.tokens
);
688 vpv
->tgsi
.tokens
= tokens
;
690 if (key
->passthrough_edgeflags
)
693 fprintf(stderr
, "mesa: cannot emulate deprecated features\n");
696 if (key
->lower_depth_clamp
) {
697 unsigned depth_range_const
=
698 _mesa_add_state_reference(params
, depth_range_state
);
700 const struct tgsi_token
*tokens
;
701 tokens
= st_tgsi_lower_depth_clamp(vpv
->tgsi
.tokens
, depth_range_const
,
702 key
->clip_negative_one_to_one
);
703 if (tokens
!= vpv
->tgsi
.tokens
)
704 tgsi_free_tokens(vpv
->tgsi
.tokens
);
705 vpv
->tgsi
.tokens
= tokens
;
708 if (ST_DEBUG
& DEBUG_TGSI
) {
709 tgsi_dump(vpv
->tgsi
.tokens
, 0);
713 vpv
->driver_shader
= pipe
->create_vs_state(pipe
, &vpv
->tgsi
);
719 * Find/create a vertex program variant.
721 struct st_vp_variant
*
722 st_get_vp_variant(struct st_context
*st
,
723 struct st_vertex_program
*stvp
,
724 const struct st_vp_variant_key
*key
)
726 struct st_vp_variant
*vpv
;
728 /* Search for existing variant */
729 for (vpv
= stvp
->variants
; vpv
; vpv
= vpv
->next
) {
730 if (memcmp(&vpv
->key
, key
, sizeof(*key
)) == 0) {
737 vpv
= st_create_vp_variant(st
, stvp
, key
);
739 for (unsigned index
= 0; index
< vpv
->num_inputs
; ++index
) {
740 unsigned attr
= stvp
->index_to_input
[index
];
741 if (attr
== ST_DOUBLE_ATTRIB_PLACEHOLDER
)
743 vpv
->vert_attrib_mask
|= 1u << attr
;
746 /* insert into list */
747 vpv
->next
= stvp
->variants
;
748 stvp
->variants
= vpv
;
757 * Translate a Mesa fragment shader into a TGSI shader.
760 st_translate_fragment_program(struct st_context
*st
,
761 struct st_fragment_program
*stfp
)
763 ubyte outputMapping
[2 * FRAG_RESULT_MAX
];
764 ubyte inputMapping
[VARYING_SLOT_MAX
];
765 ubyte inputSlotToAttr
[VARYING_SLOT_MAX
];
766 ubyte interpMode
[PIPE_MAX_SHADER_INPUTS
]; /* XXX size? */
768 GLbitfield64 inputsRead
;
769 struct ureg_program
*ureg
;
771 GLboolean write_all
= GL_FALSE
;
773 ubyte input_semantic_name
[PIPE_MAX_SHADER_INPUTS
];
774 ubyte input_semantic_index
[PIPE_MAX_SHADER_INPUTS
];
775 uint fs_num_inputs
= 0;
777 ubyte fs_output_semantic_name
[PIPE_MAX_SHADER_OUTPUTS
];
778 ubyte fs_output_semantic_index
[PIPE_MAX_SHADER_OUTPUTS
];
779 uint fs_num_outputs
= 0;
781 memset(inputSlotToAttr
, ~0, sizeof(inputSlotToAttr
));
783 /* Non-GLSL programs: */
784 if (!stfp
->glsl_to_tgsi
) {
785 _mesa_remove_output_reads(&stfp
->Base
, PROGRAM_OUTPUT
);
786 if (st
->ctx
->Const
.GLSLFragCoordIsSysVal
)
787 _mesa_program_fragment_position_to_sysval(&stfp
->Base
);
789 /* This determines which states will be updated when the assembly
792 * fragment.position and glDrawPixels always use constants.
794 stfp
->affected_states
= ST_NEW_FS_STATE
|
795 ST_NEW_SAMPLE_SHADING
|
799 /* Just set them for ATI_fs unconditionally. */
800 stfp
->affected_states
|= ST_NEW_FS_SAMPLER_VIEWS
|
804 if (stfp
->Base
.SamplersUsed
)
805 stfp
->affected_states
|= ST_NEW_FS_SAMPLER_VIEWS
|
811 bool use_nir
= PIPE_SHADER_IR_NIR
==
812 st
->pipe
->screen
->get_shader_param(st
->pipe
->screen
,
813 PIPE_SHADER_FRAGMENT
,
814 PIPE_SHADER_CAP_PREFERRED_IR
);
816 if (use_nir
&& !stfp
->ati_fs
) {
818 st_translate_prog_to_nir(st
, &stfp
->Base
, MESA_SHADER_FRAGMENT
);
820 if (stfp
->tgsi
.ir
.nir
)
821 ralloc_free(stfp
->tgsi
.ir
.nir
);
822 stfp
->tgsi
.type
= PIPE_SHADER_IR_NIR
;
823 stfp
->tgsi
.ir
.nir
= nir
;
824 stfp
->Base
.nir
= nir
;
829 * Convert Mesa program inputs to TGSI input register semantics.
831 inputsRead
= stfp
->Base
.info
.inputs_read
;
832 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
833 if ((inputsRead
& BITFIELD64_BIT(attr
)) != 0) {
834 const GLuint slot
= fs_num_inputs
++;
836 inputMapping
[attr
] = slot
;
837 inputSlotToAttr
[slot
] = attr
;
840 case VARYING_SLOT_POS
:
841 input_semantic_name
[slot
] = TGSI_SEMANTIC_POSITION
;
842 input_semantic_index
[slot
] = 0;
843 interpMode
[slot
] = TGSI_INTERPOLATE_LINEAR
;
845 case VARYING_SLOT_COL0
:
846 input_semantic_name
[slot
] = TGSI_SEMANTIC_COLOR
;
847 input_semantic_index
[slot
] = 0;
848 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
849 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_COLOR
;
851 case VARYING_SLOT_COL1
:
852 input_semantic_name
[slot
] = TGSI_SEMANTIC_COLOR
;
853 input_semantic_index
[slot
] = 1;
854 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
855 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_COLOR
;
857 case VARYING_SLOT_FOGC
:
858 input_semantic_name
[slot
] = TGSI_SEMANTIC_FOG
;
859 input_semantic_index
[slot
] = 0;
860 interpMode
[slot
] = TGSI_INTERPOLATE_PERSPECTIVE
;
862 case VARYING_SLOT_FACE
:
863 input_semantic_name
[slot
] = TGSI_SEMANTIC_FACE
;
864 input_semantic_index
[slot
] = 0;
865 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
867 case VARYING_SLOT_PRIMITIVE_ID
:
868 input_semantic_name
[slot
] = TGSI_SEMANTIC_PRIMID
;
869 input_semantic_index
[slot
] = 0;
870 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
872 case VARYING_SLOT_LAYER
:
873 input_semantic_name
[slot
] = TGSI_SEMANTIC_LAYER
;
874 input_semantic_index
[slot
] = 0;
875 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
877 case VARYING_SLOT_VIEWPORT
:
878 input_semantic_name
[slot
] = TGSI_SEMANTIC_VIEWPORT_INDEX
;
879 input_semantic_index
[slot
] = 0;
880 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
882 case VARYING_SLOT_CLIP_DIST0
:
883 input_semantic_name
[slot
] = TGSI_SEMANTIC_CLIPDIST
;
884 input_semantic_index
[slot
] = 0;
885 interpMode
[slot
] = TGSI_INTERPOLATE_PERSPECTIVE
;
887 case VARYING_SLOT_CLIP_DIST1
:
888 input_semantic_name
[slot
] = TGSI_SEMANTIC_CLIPDIST
;
889 input_semantic_index
[slot
] = 1;
890 interpMode
[slot
] = TGSI_INTERPOLATE_PERSPECTIVE
;
892 case VARYING_SLOT_CULL_DIST0
:
893 case VARYING_SLOT_CULL_DIST1
:
894 /* these should have been lowered by GLSL */
897 /* In most cases, there is nothing special about these
898 * inputs, so adopt a convention to use the generic
899 * semantic name and the mesa VARYING_SLOT_ number as the
902 * All that is required is that the vertex shader labels
903 * its own outputs similarly, and that the vertex shader
904 * generates at least every output required by the
905 * fragment shader plus fixed-function hardware (such as
908 * However, some drivers may need us to identify the PNTC and TEXi
909 * varyings if, for example, their capability to replace them with
910 * sprite coordinates is limited.
912 case VARYING_SLOT_PNTC
:
913 if (st
->needs_texcoord_semantic
) {
914 input_semantic_name
[slot
] = TGSI_SEMANTIC_PCOORD
;
915 input_semantic_index
[slot
] = 0;
916 interpMode
[slot
] = TGSI_INTERPOLATE_LINEAR
;
920 case VARYING_SLOT_TEX0
:
921 case VARYING_SLOT_TEX1
:
922 case VARYING_SLOT_TEX2
:
923 case VARYING_SLOT_TEX3
:
924 case VARYING_SLOT_TEX4
:
925 case VARYING_SLOT_TEX5
:
926 case VARYING_SLOT_TEX6
:
927 case VARYING_SLOT_TEX7
:
928 if (st
->needs_texcoord_semantic
) {
929 input_semantic_name
[slot
] = TGSI_SEMANTIC_TEXCOORD
;
930 input_semantic_index
[slot
] = attr
- VARYING_SLOT_TEX0
;
931 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
932 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_PERSPECTIVE
;
936 case VARYING_SLOT_VAR0
:
938 /* Semantic indices should be zero-based because drivers may choose
939 * to assign a fixed slot determined by that index.
940 * This is useful because ARB_separate_shader_objects uses location
941 * qualifiers for linkage, and if the semantic index corresponds to
942 * these locations, linkage passes in the driver become unecessary.
944 * If needs_texcoord_semantic is true, no semantic indices will be
945 * consumed for the TEXi varyings, and we can base the locations of
946 * the user varyings on VAR0. Otherwise, we use TEX0 as base index.
948 assert(attr
>= VARYING_SLOT_VAR0
|| attr
== VARYING_SLOT_PNTC
||
949 (attr
>= VARYING_SLOT_TEX0
&& attr
<= VARYING_SLOT_TEX7
));
950 input_semantic_name
[slot
] = TGSI_SEMANTIC_GENERIC
;
951 input_semantic_index
[slot
] = st_get_generic_varying_index(st
, attr
);
952 if (attr
== VARYING_SLOT_PNTC
)
953 interpMode
[slot
] = TGSI_INTERPOLATE_LINEAR
;
955 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
956 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_PERSPECTIVE
;
962 inputMapping
[attr
] = -1;
967 * Semantics and mapping for outputs
969 GLbitfield64 outputsWritten
= stfp
->Base
.info
.outputs_written
;
971 /* if z is written, emit that first */
972 if (outputsWritten
& BITFIELD64_BIT(FRAG_RESULT_DEPTH
)) {
973 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_POSITION
;
974 fs_output_semantic_index
[fs_num_outputs
] = 0;
975 outputMapping
[FRAG_RESULT_DEPTH
] = fs_num_outputs
;
977 outputsWritten
&= ~(1 << FRAG_RESULT_DEPTH
);
980 if (outputsWritten
& BITFIELD64_BIT(FRAG_RESULT_STENCIL
)) {
981 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_STENCIL
;
982 fs_output_semantic_index
[fs_num_outputs
] = 0;
983 outputMapping
[FRAG_RESULT_STENCIL
] = fs_num_outputs
;
985 outputsWritten
&= ~(1 << FRAG_RESULT_STENCIL
);
988 if (outputsWritten
& BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK
)) {
989 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_SAMPLEMASK
;
990 fs_output_semantic_index
[fs_num_outputs
] = 0;
991 outputMapping
[FRAG_RESULT_SAMPLE_MASK
] = fs_num_outputs
;
993 outputsWritten
&= ~(1 << FRAG_RESULT_SAMPLE_MASK
);
996 /* handle remaining outputs (color) */
997 for (attr
= 0; attr
< ARRAY_SIZE(outputMapping
); attr
++) {
998 const GLbitfield64 written
= attr
< FRAG_RESULT_MAX
? outputsWritten
:
999 stfp
->Base
.SecondaryOutputsWritten
;
1000 const unsigned loc
= attr
% FRAG_RESULT_MAX
;
1002 if (written
& BITFIELD64_BIT(loc
)) {
1004 case FRAG_RESULT_DEPTH
:
1005 case FRAG_RESULT_STENCIL
:
1006 case FRAG_RESULT_SAMPLE_MASK
:
1010 case FRAG_RESULT_COLOR
:
1011 write_all
= GL_TRUE
; /* fallthrough */
1014 assert(loc
== FRAG_RESULT_COLOR
||
1015 (FRAG_RESULT_DATA0
<= loc
&& loc
< FRAG_RESULT_MAX
));
1017 index
= (loc
== FRAG_RESULT_COLOR
) ? 0 : (loc
- FRAG_RESULT_DATA0
);
1019 if (attr
>= FRAG_RESULT_MAX
) {
1020 /* Secondary color for dual source blending. */
1025 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_COLOR
;
1026 fs_output_semantic_index
[fs_num_outputs
] = index
;
1027 outputMapping
[attr
] = fs_num_outputs
;
1036 ureg
= ureg_create_with_screen(PIPE_SHADER_FRAGMENT
, st
->pipe
->screen
);
1040 if (ST_DEBUG
& DEBUG_MESA
) {
1041 _mesa_print_program(&stfp
->Base
);
1042 _mesa_print_program_parameters(st
->ctx
, &stfp
->Base
);
1045 if (write_all
== GL_TRUE
)
1046 ureg_property(ureg
, TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS
, 1);
1048 if (stfp
->Base
.info
.fs
.depth_layout
!= FRAG_DEPTH_LAYOUT_NONE
) {
1049 switch (stfp
->Base
.info
.fs
.depth_layout
) {
1050 case FRAG_DEPTH_LAYOUT_ANY
:
1051 ureg_property(ureg
, TGSI_PROPERTY_FS_DEPTH_LAYOUT
,
1052 TGSI_FS_DEPTH_LAYOUT_ANY
);
1054 case FRAG_DEPTH_LAYOUT_GREATER
:
1055 ureg_property(ureg
, TGSI_PROPERTY_FS_DEPTH_LAYOUT
,
1056 TGSI_FS_DEPTH_LAYOUT_GREATER
);
1058 case FRAG_DEPTH_LAYOUT_LESS
:
1059 ureg_property(ureg
, TGSI_PROPERTY_FS_DEPTH_LAYOUT
,
1060 TGSI_FS_DEPTH_LAYOUT_LESS
);
1062 case FRAG_DEPTH_LAYOUT_UNCHANGED
:
1063 ureg_property(ureg
, TGSI_PROPERTY_FS_DEPTH_LAYOUT
,
1064 TGSI_FS_DEPTH_LAYOUT_UNCHANGED
);
1071 if (stfp
->glsl_to_tgsi
) {
1072 st_translate_program(st
->ctx
,
1073 PIPE_SHADER_FRAGMENT
,
1081 input_semantic_name
,
1082 input_semantic_index
,
1087 fs_output_semantic_name
,
1088 fs_output_semantic_index
);
1090 free_glsl_to_tgsi_visitor(stfp
->glsl_to_tgsi
);
1091 } else if (stfp
->ati_fs
)
1092 st_translate_atifs_program(ureg
,
1098 input_semantic_name
,
1099 input_semantic_index
,
1104 fs_output_semantic_name
,
1105 fs_output_semantic_index
);
1107 st_translate_mesa_program(st
->ctx
,
1108 PIPE_SHADER_FRAGMENT
,
1114 input_semantic_name
,
1115 input_semantic_index
,
1120 fs_output_semantic_name
,
1121 fs_output_semantic_index
);
1123 stfp
->tgsi
.tokens
= ureg_get_tokens(ureg
, &stfp
->num_tgsi_tokens
);
1126 if (stfp
->glsl_to_tgsi
) {
1127 stfp
->glsl_to_tgsi
= NULL
;
1128 st_store_ir_in_disk_cache(st
, &stfp
->Base
, false);
1131 return stfp
->tgsi
.tokens
!= NULL
;
1134 static struct st_fp_variant
*
1135 st_create_fp_variant(struct st_context
*st
,
1136 struct st_fragment_program
*stfp
,
1137 const struct st_fp_variant_key
*key
)
1139 struct pipe_context
*pipe
= st
->pipe
;
1140 struct st_fp_variant
*variant
= CALLOC_STRUCT(st_fp_variant
);
1141 struct pipe_shader_state tgsi
= {0};
1142 struct gl_program_parameter_list
*params
= stfp
->Base
.Parameters
;
1143 static const gl_state_index16 texcoord_state
[STATE_LENGTH
] =
1144 { STATE_INTERNAL
, STATE_CURRENT_ATTRIB
, VERT_ATTRIB_TEX0
};
1145 static const gl_state_index16 scale_state
[STATE_LENGTH
] =
1146 { STATE_INTERNAL
, STATE_PT_SCALE
};
1147 static const gl_state_index16 bias_state
[STATE_LENGTH
] =
1148 { STATE_INTERNAL
, STATE_PT_BIAS
};
1153 if (stfp
->tgsi
.type
== PIPE_SHADER_IR_NIR
) {
1154 tgsi
.type
= PIPE_SHADER_IR_NIR
;
1155 tgsi
.ir
.nir
= nir_shader_clone(NULL
, stfp
->tgsi
.ir
.nir
);
1157 if (key
->clamp_color
)
1158 NIR_PASS_V(tgsi
.ir
.nir
, nir_lower_clamp_color_outputs
);
1160 if (key
->persample_shading
) {
1161 nir_shader
*shader
= tgsi
.ir
.nir
;
1162 nir_foreach_variable(var
, &shader
->inputs
)
1163 var
->data
.sample
= true;
1166 assert(!(key
->bitmap
&& key
->drawpixels
));
1170 nir_lower_bitmap_options options
= {0};
1172 variant
->bitmap_sampler
= ffs(~stfp
->Base
.SamplersUsed
) - 1;
1173 options
.sampler
= variant
->bitmap_sampler
;
1174 options
.swizzle_xxxx
= st
->bitmap
.tex_format
== PIPE_FORMAT_R8_UNORM
;
1176 NIR_PASS_V(tgsi
.ir
.nir
, nir_lower_bitmap
, &options
);
1179 /* glDrawPixels (color only) */
1180 if (key
->drawpixels
) {
1181 nir_lower_drawpixels_options options
= {{0}};
1182 unsigned samplers_used
= stfp
->Base
.SamplersUsed
;
1184 /* Find the first unused slot. */
1185 variant
->drawpix_sampler
= ffs(~samplers_used
) - 1;
1186 options
.drawpix_sampler
= variant
->drawpix_sampler
;
1187 samplers_used
|= (1 << variant
->drawpix_sampler
);
1189 options
.pixel_maps
= key
->pixelMaps
;
1190 if (key
->pixelMaps
) {
1191 variant
->pixelmap_sampler
= ffs(~samplers_used
) - 1;
1192 options
.pixelmap_sampler
= variant
->pixelmap_sampler
;
1195 options
.scale_and_bias
= key
->scaleAndBias
;
1196 if (key
->scaleAndBias
) {
1197 _mesa_add_state_reference(params
, scale_state
);
1198 memcpy(options
.scale_state_tokens
, scale_state
,
1199 sizeof(options
.scale_state_tokens
));
1200 _mesa_add_state_reference(params
, bias_state
);
1201 memcpy(options
.bias_state_tokens
, bias_state
,
1202 sizeof(options
.bias_state_tokens
));
1205 _mesa_add_state_reference(params
, texcoord_state
);
1206 memcpy(options
.texcoord_state_tokens
, texcoord_state
,
1207 sizeof(options
.texcoord_state_tokens
));
1209 NIR_PASS_V(tgsi
.ir
.nir
, nir_lower_drawpixels
, &options
);
1212 if (unlikely(key
->external
.lower_nv12
|| key
->external
.lower_iyuv
||
1213 key
->external
.lower_xy_uxvx
|| key
->external
.lower_yx_xuxv
||
1214 key
->external
.lower_ayuv
|| key
->external
.lower_xyuv
)) {
1215 nir_lower_tex_options options
= {0};
1216 options
.lower_y_uv_external
= key
->external
.lower_nv12
;
1217 options
.lower_y_u_v_external
= key
->external
.lower_iyuv
;
1218 options
.lower_xy_uxvx_external
= key
->external
.lower_xy_uxvx
;
1219 options
.lower_yx_xuxv_external
= key
->external
.lower_yx_xuxv
;
1220 options
.lower_ayuv_external
= key
->external
.lower_ayuv
;
1221 options
.lower_xyuv_external
= key
->external
.lower_xyuv
;
1222 NIR_PASS_V(tgsi
.ir
.nir
, nir_lower_tex
, &options
);
1225 st_finalize_nir(st
, &stfp
->Base
, stfp
->shader_program
, tgsi
.ir
.nir
);
1227 if (unlikely(key
->external
.lower_nv12
|| key
->external
.lower_iyuv
||
1228 key
->external
.lower_xy_uxvx
|| key
->external
.lower_yx_xuxv
)) {
1229 /* This pass needs to happen *after* nir_lower_sampler */
1230 NIR_PASS_V(tgsi
.ir
.nir
, st_nir_lower_tex_src_plane
,
1231 ~stfp
->Base
.SamplersUsed
,
1232 key
->external
.lower_nv12
|| key
->external
.lower_xy_uxvx
||
1233 key
->external
.lower_yx_xuxv
,
1234 key
->external
.lower_iyuv
);
1237 /* Some of the lowering above may have introduced new varyings */
1238 nir_shader_gather_info(tgsi
.ir
.nir
,
1239 nir_shader_get_entrypoint(tgsi
.ir
.nir
));
1241 variant
->driver_shader
= pipe
->create_fs_state(pipe
, &tgsi
);
1242 variant
->key
= *key
;
1247 tgsi
.tokens
= stfp
->tgsi
.tokens
;
1249 assert(!(key
->bitmap
&& key
->drawpixels
));
1251 /* Fix texture targets and add fog for ATI_fs */
1253 const struct tgsi_token
*tokens
= st_fixup_atifs(tgsi
.tokens
, key
);
1256 tgsi
.tokens
= tokens
;
1258 fprintf(stderr
, "mesa: cannot post-process ATI_fs\n");
1261 /* Emulate features. */
1262 if (key
->clamp_color
|| key
->persample_shading
) {
1263 const struct tgsi_token
*tokens
;
1265 (key
->clamp_color
? TGSI_EMU_CLAMP_COLOR_OUTPUTS
: 0) |
1266 (key
->persample_shading
? TGSI_EMU_FORCE_PERSAMPLE_INTERP
: 0);
1268 tokens
= tgsi_emulate(tgsi
.tokens
, flags
);
1271 if (tgsi
.tokens
!= stfp
->tgsi
.tokens
)
1272 tgsi_free_tokens(tgsi
.tokens
);
1273 tgsi
.tokens
= tokens
;
1275 fprintf(stderr
, "mesa: cannot emulate deprecated features\n");
1280 const struct tgsi_token
*tokens
;
1282 variant
->bitmap_sampler
= ffs(~stfp
->Base
.SamplersUsed
) - 1;
1284 tokens
= st_get_bitmap_shader(tgsi
.tokens
,
1285 st
->internal_target
,
1286 variant
->bitmap_sampler
,
1287 st
->needs_texcoord_semantic
,
1288 st
->bitmap
.tex_format
==
1289 PIPE_FORMAT_R8_UNORM
);
1292 if (tgsi
.tokens
!= stfp
->tgsi
.tokens
)
1293 tgsi_free_tokens(tgsi
.tokens
);
1294 tgsi
.tokens
= tokens
;
1296 fprintf(stderr
, "mesa: cannot create a shader for glBitmap\n");
1299 /* glDrawPixels (color only) */
1300 if (key
->drawpixels
) {
1301 const struct tgsi_token
*tokens
;
1302 unsigned scale_const
= 0, bias_const
= 0, texcoord_const
= 0;
1304 /* Find the first unused slot. */
1305 variant
->drawpix_sampler
= ffs(~stfp
->Base
.SamplersUsed
) - 1;
1307 if (key
->pixelMaps
) {
1308 unsigned samplers_used
= stfp
->Base
.SamplersUsed
|
1309 (1 << variant
->drawpix_sampler
);
1311 variant
->pixelmap_sampler
= ffs(~samplers_used
) - 1;
1314 if (key
->scaleAndBias
) {
1315 scale_const
= _mesa_add_state_reference(params
, scale_state
);
1316 bias_const
= _mesa_add_state_reference(params
, bias_state
);
1319 texcoord_const
= _mesa_add_state_reference(params
, texcoord_state
);
1321 tokens
= st_get_drawpix_shader(tgsi
.tokens
,
1322 st
->needs_texcoord_semantic
,
1323 key
->scaleAndBias
, scale_const
,
1324 bias_const
, key
->pixelMaps
,
1325 variant
->drawpix_sampler
,
1326 variant
->pixelmap_sampler
,
1327 texcoord_const
, st
->internal_target
);
1330 if (tgsi
.tokens
!= stfp
->tgsi
.tokens
)
1331 tgsi_free_tokens(tgsi
.tokens
);
1332 tgsi
.tokens
= tokens
;
1334 fprintf(stderr
, "mesa: cannot create a shader for glDrawPixels\n");
1337 if (unlikely(key
->external
.lower_nv12
|| key
->external
.lower_iyuv
||
1338 key
->external
.lower_xy_uxvx
|| key
->external
.lower_yx_xuxv
)) {
1339 const struct tgsi_token
*tokens
;
1341 /* samplers inserted would conflict, but this should be unpossible: */
1342 assert(!(key
->bitmap
|| key
->drawpixels
));
1344 tokens
= st_tgsi_lower_yuv(tgsi
.tokens
,
1345 ~stfp
->Base
.SamplersUsed
,
1346 key
->external
.lower_nv12
||
1347 key
->external
.lower_xy_uxvx
||
1348 key
->external
.lower_yx_xuxv
,
1349 key
->external
.lower_iyuv
);
1351 if (tgsi
.tokens
!= stfp
->tgsi
.tokens
)
1352 tgsi_free_tokens(tgsi
.tokens
);
1353 tgsi
.tokens
= tokens
;
1355 fprintf(stderr
, "mesa: cannot create a shader for samplerExternalOES\n");
1359 if (key
->lower_depth_clamp
) {
1360 unsigned depth_range_const
= _mesa_add_state_reference(params
, depth_range_state
);
1362 const struct tgsi_token
*tokens
;
1363 tokens
= st_tgsi_lower_depth_clamp_fs(tgsi
.tokens
, depth_range_const
);
1364 if (tgsi
.tokens
!= stfp
->tgsi
.tokens
)
1365 tgsi_free_tokens(tgsi
.tokens
);
1366 tgsi
.tokens
= tokens
;
1369 if (ST_DEBUG
& DEBUG_TGSI
) {
1370 tgsi_dump(tgsi
.tokens
, 0);
1374 /* fill in variant */
1375 variant
->driver_shader
= pipe
->create_fs_state(pipe
, &tgsi
);
1376 variant
->key
= *key
;
1378 if (tgsi
.tokens
!= stfp
->tgsi
.tokens
)
1379 tgsi_free_tokens(tgsi
.tokens
);
1384 * Translate fragment program if needed.
1386 struct st_fp_variant
*
1387 st_get_fp_variant(struct st_context
*st
,
1388 struct st_fragment_program
*stfp
,
1389 const struct st_fp_variant_key
*key
)
1391 struct st_fp_variant
*fpv
;
1393 /* Search for existing variant */
1394 for (fpv
= stfp
->variants
; fpv
; fpv
= fpv
->next
) {
1395 if (memcmp(&fpv
->key
, key
, sizeof(*key
)) == 0) {
1402 fpv
= st_create_fp_variant(st
, stfp
, key
);
1404 if (key
->bitmap
|| key
->drawpixels
) {
1405 /* Regular variants should always come before the
1406 * bitmap & drawpixels variants, (unless there
1407 * are no regular variants) so that
1408 * st_update_fp can take a fast path when
1409 * shader_has_one_variant is set.
1411 if (!stfp
->variants
) {
1412 stfp
->variants
= fpv
;
1414 /* insert into list after the first one */
1415 fpv
->next
= stfp
->variants
->next
;
1416 stfp
->variants
->next
= fpv
;
1419 /* insert into list */
1420 fpv
->next
= stfp
->variants
;
1421 stfp
->variants
= fpv
;
1430 * Translate a program. This is common code for geometry and tessellation
1434 st_translate_common_program(struct st_context
*st
,
1435 struct st_common_program
*stcp
)
1437 struct gl_program
*prog
= &stcp
->Base
;
1438 enum pipe_shader_type stage
=
1439 pipe_shader_type_from_mesa(stcp
->Base
.info
.stage
);
1440 struct ureg_program
*ureg
= ureg_create_with_screen(stage
, st
->pipe
->screen
);
1446 case PIPE_SHADER_TESS_CTRL
:
1447 ureg_property(ureg
, TGSI_PROPERTY_TCS_VERTICES_OUT
,
1448 stcp
->Base
.info
.tess
.tcs_vertices_out
);
1451 case PIPE_SHADER_TESS_EVAL
:
1452 if (stcp
->Base
.info
.tess
.primitive_mode
== GL_ISOLINES
)
1453 ureg_property(ureg
, TGSI_PROPERTY_TES_PRIM_MODE
, GL_LINES
);
1455 ureg_property(ureg
, TGSI_PROPERTY_TES_PRIM_MODE
,
1456 stcp
->Base
.info
.tess
.primitive_mode
);
1458 STATIC_ASSERT((TESS_SPACING_EQUAL
+ 1) % 3 == PIPE_TESS_SPACING_EQUAL
);
1459 STATIC_ASSERT((TESS_SPACING_FRACTIONAL_ODD
+ 1) % 3 ==
1460 PIPE_TESS_SPACING_FRACTIONAL_ODD
);
1461 STATIC_ASSERT((TESS_SPACING_FRACTIONAL_EVEN
+ 1) % 3 ==
1462 PIPE_TESS_SPACING_FRACTIONAL_EVEN
);
1464 ureg_property(ureg
, TGSI_PROPERTY_TES_SPACING
,
1465 (stcp
->Base
.info
.tess
.spacing
+ 1) % 3);
1467 ureg_property(ureg
, TGSI_PROPERTY_TES_VERTEX_ORDER_CW
,
1468 !stcp
->Base
.info
.tess
.ccw
);
1469 ureg_property(ureg
, TGSI_PROPERTY_TES_POINT_MODE
,
1470 stcp
->Base
.info
.tess
.point_mode
);
1473 case PIPE_SHADER_GEOMETRY
:
1474 ureg_property(ureg
, TGSI_PROPERTY_GS_INPUT_PRIM
,
1475 stcp
->Base
.info
.gs
.input_primitive
);
1476 ureg_property(ureg
, TGSI_PROPERTY_GS_OUTPUT_PRIM
,
1477 stcp
->Base
.info
.gs
.output_primitive
);
1478 ureg_property(ureg
, TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES
,
1479 stcp
->Base
.info
.gs
.vertices_out
);
1480 ureg_property(ureg
, TGSI_PROPERTY_GS_INVOCATIONS
,
1481 stcp
->Base
.info
.gs
.invocations
);
1488 ubyte inputSlotToAttr
[VARYING_SLOT_TESS_MAX
];
1489 ubyte inputMapping
[VARYING_SLOT_TESS_MAX
];
1490 ubyte outputMapping
[VARYING_SLOT_TESS_MAX
];
1493 ubyte input_semantic_name
[PIPE_MAX_SHADER_INPUTS
];
1494 ubyte input_semantic_index
[PIPE_MAX_SHADER_INPUTS
];
1495 uint num_inputs
= 0;
1497 ubyte output_semantic_name
[PIPE_MAX_SHADER_OUTPUTS
];
1498 ubyte output_semantic_index
[PIPE_MAX_SHADER_OUTPUTS
];
1499 uint num_outputs
= 0;
1503 memset(inputSlotToAttr
, 0, sizeof(inputSlotToAttr
));
1504 memset(inputMapping
, 0, sizeof(inputMapping
));
1505 memset(outputMapping
, 0, sizeof(outputMapping
));
1506 memset(&stcp
->tgsi
, 0, sizeof(stcp
->tgsi
));
1508 if (prog
->info
.clip_distance_array_size
)
1509 ureg_property(ureg
, TGSI_PROPERTY_NUM_CLIPDIST_ENABLED
,
1510 prog
->info
.clip_distance_array_size
);
1511 if (prog
->info
.cull_distance_array_size
)
1512 ureg_property(ureg
, TGSI_PROPERTY_NUM_CULLDIST_ENABLED
,
1513 prog
->info
.cull_distance_array_size
);
1516 * Convert Mesa program inputs to TGSI input register semantics.
1518 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
1519 if ((prog
->info
.inputs_read
& BITFIELD64_BIT(attr
)) == 0)
1522 unsigned slot
= num_inputs
++;
1524 inputMapping
[attr
] = slot
;
1525 inputSlotToAttr
[slot
] = attr
;
1527 unsigned semantic_name
, semantic_index
;
1528 tgsi_get_gl_varying_semantic(attr
, st
->needs_texcoord_semantic
,
1529 &semantic_name
, &semantic_index
);
1530 input_semantic_name
[slot
] = semantic_name
;
1531 input_semantic_index
[slot
] = semantic_index
;
1534 /* Also add patch inputs. */
1535 for (attr
= 0; attr
< 32; attr
++) {
1536 if (prog
->info
.patch_inputs_read
& (1u << attr
)) {
1537 GLuint slot
= num_inputs
++;
1538 GLuint patch_attr
= VARYING_SLOT_PATCH0
+ attr
;
1540 inputMapping
[patch_attr
] = slot
;
1541 inputSlotToAttr
[slot
] = patch_attr
;
1542 input_semantic_name
[slot
] = TGSI_SEMANTIC_PATCH
;
1543 input_semantic_index
[slot
] = attr
;
1547 /* initialize output semantics to defaults */
1548 for (i
= 0; i
< PIPE_MAX_SHADER_OUTPUTS
; i
++) {
1549 output_semantic_name
[i
] = TGSI_SEMANTIC_GENERIC
;
1550 output_semantic_index
[i
] = 0;
1554 * Determine number of outputs, the (default) output register
1555 * mapping and the semantic information for each output.
1557 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
1558 if (prog
->info
.outputs_written
& BITFIELD64_BIT(attr
)) {
1559 GLuint slot
= num_outputs
++;
1561 outputMapping
[attr
] = slot
;
1563 unsigned semantic_name
, semantic_index
;
1564 tgsi_get_gl_varying_semantic(attr
, st
->needs_texcoord_semantic
,
1565 &semantic_name
, &semantic_index
);
1566 output_semantic_name
[slot
] = semantic_name
;
1567 output_semantic_index
[slot
] = semantic_index
;
1571 /* Also add patch outputs. */
1572 for (attr
= 0; attr
< 32; attr
++) {
1573 if (prog
->info
.patch_outputs_written
& (1u << attr
)) {
1574 GLuint slot
= num_outputs
++;
1575 GLuint patch_attr
= VARYING_SLOT_PATCH0
+ attr
;
1577 outputMapping
[patch_attr
] = slot
;
1578 output_semantic_name
[slot
] = TGSI_SEMANTIC_PATCH
;
1579 output_semantic_index
[slot
] = attr
;
1583 st_translate_program(st
->ctx
,
1592 input_semantic_name
,
1593 input_semantic_index
,
1598 output_semantic_name
,
1599 output_semantic_index
);
1601 stcp
->tgsi
.tokens
= ureg_get_tokens(ureg
, &stcp
->num_tgsi_tokens
);
1605 st_translate_stream_output_info(prog
);
1607 st_store_ir_in_disk_cache(st
, prog
, false);
1609 if ((ST_DEBUG
& DEBUG_TGSI
) && (ST_DEBUG
& DEBUG_MESA
)) {
1610 _mesa_print_program(prog
);
1614 if (ST_DEBUG
& DEBUG_TGSI
) {
1615 tgsi_dump(stcp
->tgsi
.tokens
, 0);
1619 free_glsl_to_tgsi_visitor(stcp
->glsl_to_tgsi
);
1620 stcp
->glsl_to_tgsi
= NULL
;
1626 * Get/create a basic program variant.
1628 struct st_basic_variant
*
1629 st_get_basic_variant(struct st_context
*st
,
1630 struct st_common_program
*prog
,
1631 const struct st_basic_variant_key
*key
)
1633 struct pipe_context
*pipe
= st
->pipe
;
1634 struct st_basic_variant
*v
;
1635 struct pipe_shader_state tgsi
= {0};
1637 /* Search for existing variant */
1638 for (v
= prog
->variants
; v
; v
= v
->next
) {
1639 if (memcmp(&v
->key
, key
, sizeof(*key
)) == 0) {
1646 v
= CALLOC_STRUCT(st_basic_variant
);
1649 if (prog
->tgsi
.type
== PIPE_SHADER_IR_NIR
) {
1650 tgsi
.type
= PIPE_SHADER_IR_NIR
;
1651 tgsi
.ir
.nir
= nir_shader_clone(NULL
, prog
->tgsi
.ir
.nir
);
1653 if (key
->clamp_color
)
1654 NIR_PASS_V(tgsi
.ir
.nir
, nir_lower_clamp_color_outputs
);
1656 tgsi
.stream_output
= prog
->tgsi
.stream_output
;
1658 st_finalize_nir(st
, &prog
->Base
, prog
->shader_program
,
1661 if (key
->lower_depth_clamp
) {
1662 struct gl_program_parameter_list
*params
= prog
->Base
.Parameters
;
1664 unsigned depth_range_const
=
1665 _mesa_add_state_reference(params
, depth_range_state
);
1667 const struct tgsi_token
*tokens
;
1669 st_tgsi_lower_depth_clamp(prog
->tgsi
.tokens
,
1671 key
->clip_negative_one_to_one
);
1673 if (tokens
!= prog
->tgsi
.tokens
)
1674 tgsi_free_tokens(prog
->tgsi
.tokens
);
1676 prog
->tgsi
.tokens
= tokens
;
1677 prog
->num_tgsi_tokens
= tgsi_num_tokens(tokens
);
1681 /* fill in new variant */
1682 switch (prog
->Base
.info
.stage
) {
1683 case MESA_SHADER_TESS_CTRL
:
1684 v
->driver_shader
= pipe
->create_tcs_state(pipe
, &tgsi
);
1686 case MESA_SHADER_TESS_EVAL
:
1687 v
->driver_shader
= pipe
->create_tes_state(pipe
, &tgsi
);
1689 case MESA_SHADER_GEOMETRY
:
1690 v
->driver_shader
= pipe
->create_gs_state(pipe
, &tgsi
);
1692 case MESA_SHADER_COMPUTE
: {
1693 struct pipe_compute_state cs
= {0};
1694 cs
.ir_type
= tgsi
.type
;
1695 cs
.req_local_mem
= prog
->Base
.info
.cs
.shared_size
;
1697 if (tgsi
.type
== PIPE_SHADER_IR_NIR
)
1698 cs
.prog
= tgsi
.ir
.nir
;
1700 cs
.prog
= tgsi
.tokens
;
1702 v
->driver_shader
= pipe
->create_compute_state(pipe
, &cs
);
1706 assert(!"unhandled shader type");
1713 /* insert into list */
1714 v
->next
= prog
->variants
;
1724 * Vert/Geom/Frag programs have per-context variants. Free all the
1725 * variants attached to the given program which match the given context.
1728 destroy_program_variants(struct st_context
*st
, struct gl_program
*target
)
1730 if (!target
|| target
== &_mesa_DummyProgram
)
1733 switch (target
->Target
) {
1734 case GL_VERTEX_PROGRAM_ARB
:
1736 struct st_vertex_program
*stvp
= (struct st_vertex_program
*) target
;
1737 struct st_vp_variant
*vpv
, **prevPtr
= &stvp
->variants
;
1739 for (vpv
= stvp
->variants
; vpv
; ) {
1740 struct st_vp_variant
*next
= vpv
->next
;
1741 if (vpv
->key
.st
== st
) {
1742 /* unlink from list */
1744 /* destroy this variant */
1745 delete_vp_variant(st
, vpv
);
1748 prevPtr
= &vpv
->next
;
1754 case GL_FRAGMENT_PROGRAM_ARB
:
1756 struct st_fragment_program
*stfp
=
1757 (struct st_fragment_program
*) target
;
1758 struct st_fp_variant
*fpv
, **prevPtr
= &stfp
->variants
;
1760 for (fpv
= stfp
->variants
; fpv
; ) {
1761 struct st_fp_variant
*next
= fpv
->next
;
1762 if (fpv
->key
.st
== st
) {
1763 /* unlink from list */
1765 /* destroy this variant */
1766 delete_fp_variant(st
, fpv
);
1769 prevPtr
= &fpv
->next
;
1775 case GL_GEOMETRY_PROGRAM_NV
:
1776 case GL_TESS_CONTROL_PROGRAM_NV
:
1777 case GL_TESS_EVALUATION_PROGRAM_NV
:
1778 case GL_COMPUTE_PROGRAM_NV
:
1780 struct st_common_program
*p
= st_common_program(target
);
1781 struct st_basic_variant
*v
, **prevPtr
= &p
->variants
;
1783 for (v
= p
->variants
; v
; ) {
1784 struct st_basic_variant
*next
= v
->next
;
1785 if (v
->key
.st
== st
) {
1786 /* unlink from list */
1788 /* destroy this variant */
1789 delete_basic_variant(st
, v
, target
->Target
);
1799 _mesa_problem(NULL
, "Unexpected program target 0x%x in "
1800 "destroy_program_variants_cb()", target
->Target
);
1806 * Callback for _mesa_HashWalk. Free all the shader's program variants
1807 * which match the given context.
1810 destroy_shader_program_variants_cb(GLuint key
, void *data
, void *userData
)
1812 struct st_context
*st
= (struct st_context
*) userData
;
1813 struct gl_shader
*shader
= (struct gl_shader
*) data
;
1815 switch (shader
->Type
) {
1816 case GL_SHADER_PROGRAM_MESA
:
1818 struct gl_shader_program
*shProg
= (struct gl_shader_program
*) data
;
1821 for (i
= 0; i
< ARRAY_SIZE(shProg
->_LinkedShaders
); i
++) {
1822 if (shProg
->_LinkedShaders
[i
])
1823 destroy_program_variants(st
, shProg
->_LinkedShaders
[i
]->Program
);
1827 case GL_VERTEX_SHADER
:
1828 case GL_FRAGMENT_SHADER
:
1829 case GL_GEOMETRY_SHADER
:
1830 case GL_TESS_CONTROL_SHADER
:
1831 case GL_TESS_EVALUATION_SHADER
:
1832 case GL_COMPUTE_SHADER
:
1841 * Callback for _mesa_HashWalk. Free all the program variants which match
1842 * the given context.
1845 destroy_program_variants_cb(GLuint key
, void *data
, void *userData
)
1847 struct st_context
*st
= (struct st_context
*) userData
;
1848 struct gl_program
*program
= (struct gl_program
*) data
;
1849 destroy_program_variants(st
, program
);
1854 * Walk over all shaders and programs to delete any variants which
1855 * belong to the given context.
1856 * This is called during context tear-down.
1859 st_destroy_program_variants(struct st_context
*st
)
1861 /* If shaders can be shared with other contexts, the last context will
1862 * call DeleteProgram on all shaders, releasing everything.
1864 if (st
->has_shareable_shaders
)
1867 /* ARB vert/frag program */
1868 _mesa_HashWalk(st
->ctx
->Shared
->Programs
,
1869 destroy_program_variants_cb
, st
);
1871 /* GLSL vert/frag/geom shaders */
1872 _mesa_HashWalk(st
->ctx
->Shared
->ShaderObjects
,
1873 destroy_shader_program_variants_cb
, st
);
1878 * For debugging, print/dump the current vertex program.
1881 st_print_current_vertex_program(void)
1883 GET_CURRENT_CONTEXT(ctx
);
1885 if (ctx
->VertexProgram
._Current
) {
1886 struct st_vertex_program
*stvp
=
1887 (struct st_vertex_program
*) ctx
->VertexProgram
._Current
;
1888 struct st_vp_variant
*stv
;
1890 debug_printf("Vertex program %u\n", stvp
->Base
.Id
);
1892 for (stv
= stvp
->variants
; stv
; stv
= stv
->next
) {
1893 debug_printf("variant %p\n", stv
);
1894 tgsi_dump(stv
->tgsi
.tokens
, 0);
1901 * Compile one shader variant.
1904 st_precompile_shader_variant(struct st_context
*st
,
1905 struct gl_program
*prog
)
1907 switch (prog
->Target
) {
1908 case GL_VERTEX_PROGRAM_ARB
: {
1909 struct st_vertex_program
*p
= (struct st_vertex_program
*)prog
;
1910 struct st_vp_variant_key key
;
1912 memset(&key
, 0, sizeof(key
));
1914 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
1915 st_get_vp_variant(st
, p
, &key
);
1919 case GL_FRAGMENT_PROGRAM_ARB
: {
1920 struct st_fragment_program
*p
= (struct st_fragment_program
*)prog
;
1921 struct st_fp_variant_key key
;
1923 memset(&key
, 0, sizeof(key
));
1925 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
1926 st_get_fp_variant(st
, p
, &key
);
1930 case GL_TESS_CONTROL_PROGRAM_NV
:
1931 case GL_TESS_EVALUATION_PROGRAM_NV
:
1932 case GL_GEOMETRY_PROGRAM_NV
:
1933 case GL_COMPUTE_PROGRAM_NV
: {
1934 struct st_common_program
*p
= st_common_program(prog
);
1935 struct st_basic_variant_key key
;
1937 memset(&key
, 0, sizeof(key
));
1939 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
1940 st_get_basic_variant(st
, p
, &key
);