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"
44 #include "draw/draw_context.h"
46 #include "pipe/p_context.h"
47 #include "pipe/p_defines.h"
48 #include "pipe/p_shader_tokens.h"
49 #include "draw/draw_context.h"
50 #include "tgsi/tgsi_dump.h"
51 #include "tgsi/tgsi_emulate.h"
52 #include "tgsi/tgsi_parse.h"
53 #include "tgsi/tgsi_ureg.h"
56 #include "st_cb_bitmap.h"
57 #include "st_cb_drawpixels.h"
58 #include "st_context.h"
59 #include "st_tgsi_lower_depth_clamp.h"
60 #include "st_tgsi_lower_yuv.h"
61 #include "st_program.h"
62 #include "st_mesa_to_tgsi.h"
63 #include "st_atifs_to_tgsi.h"
65 #include "st_shader_cache.h"
67 #include "cso_cache/cso_context.h"
72 set_affected_state_flags(uint64_t *states
,
73 struct gl_program
*prog
,
74 uint64_t new_constants
,
75 uint64_t new_sampler_views
,
76 uint64_t new_samplers
,
82 if (prog
->Parameters
->NumParameters
)
83 *states
|= new_constants
;
85 if (prog
->info
.num_textures
)
86 *states
|= new_sampler_views
| new_samplers
;
88 if (prog
->info
.num_images
)
89 *states
|= new_images
;
91 if (prog
->info
.num_ubos
)
94 if (prog
->info
.num_ssbos
)
97 if (prog
->info
.num_abos
)
98 *states
|= new_atomics
;
102 * This determines which states will be updated when the shader is bound.
105 st_set_prog_affected_state_flags(struct gl_program
*prog
)
109 switch (prog
->info
.stage
) {
110 case MESA_SHADER_VERTEX
:
111 states
= &((struct st_program
*)prog
)->affected_states
;
113 *states
= ST_NEW_VS_STATE
|
115 ST_NEW_VERTEX_ARRAYS
;
117 set_affected_state_flags(states
, prog
,
119 ST_NEW_VS_SAMPLER_VIEWS
,
127 case MESA_SHADER_TESS_CTRL
:
128 states
= &(st_program(prog
))->affected_states
;
130 *states
= ST_NEW_TCS_STATE
;
132 set_affected_state_flags(states
, prog
,
133 ST_NEW_TCS_CONSTANTS
,
134 ST_NEW_TCS_SAMPLER_VIEWS
,
142 case MESA_SHADER_TESS_EVAL
:
143 states
= &(st_program(prog
))->affected_states
;
145 *states
= ST_NEW_TES_STATE
|
148 set_affected_state_flags(states
, prog
,
149 ST_NEW_TES_CONSTANTS
,
150 ST_NEW_TES_SAMPLER_VIEWS
,
158 case MESA_SHADER_GEOMETRY
:
159 states
= &(st_program(prog
))->affected_states
;
161 *states
= ST_NEW_GS_STATE
|
164 set_affected_state_flags(states
, prog
,
166 ST_NEW_GS_SAMPLER_VIEWS
,
174 case MESA_SHADER_FRAGMENT
:
175 states
= &((struct st_program
*)prog
)->affected_states
;
177 /* gl_FragCoord and glDrawPixels always use constants. */
178 *states
= ST_NEW_FS_STATE
|
179 ST_NEW_SAMPLE_SHADING
|
182 set_affected_state_flags(states
, prog
,
184 ST_NEW_FS_SAMPLER_VIEWS
,
192 case MESA_SHADER_COMPUTE
:
193 states
= &((struct st_program
*)prog
)->affected_states
;
195 *states
= ST_NEW_CS_STATE
;
197 set_affected_state_flags(states
, prog
,
199 ST_NEW_CS_SAMPLER_VIEWS
,
208 unreachable("unhandled shader stage");
214 * Delete a shader variant. Note the caller must unlink the variant from
218 delete_variant(struct st_context
*st
, struct st_variant
*v
, GLenum target
)
220 if (v
->driver_shader
) {
221 if (st
->has_shareable_shaders
|| v
->st
== st
) {
222 /* The shader's context matches the calling context, or we
226 case GL_VERTEX_PROGRAM_ARB
:
227 cso_delete_vertex_shader(st
->cso_context
, v
->driver_shader
);
229 case GL_TESS_CONTROL_PROGRAM_NV
:
230 cso_delete_tessctrl_shader(st
->cso_context
, v
->driver_shader
);
232 case GL_TESS_EVALUATION_PROGRAM_NV
:
233 cso_delete_tesseval_shader(st
->cso_context
, v
->driver_shader
);
235 case GL_GEOMETRY_PROGRAM_NV
:
236 cso_delete_geometry_shader(st
->cso_context
, v
->driver_shader
);
238 case GL_FRAGMENT_PROGRAM_ARB
:
239 cso_delete_fragment_shader(st
->cso_context
, v
->driver_shader
);
241 case GL_COMPUTE_PROGRAM_NV
:
242 cso_delete_compute_shader(st
->cso_context
, v
->driver_shader
);
245 unreachable("bad shader type in delete_basic_variant");
248 /* We can't delete a shader with a context different from the one
249 * that created it. Add it to the creating context's zombie list.
251 enum pipe_shader_type type
=
252 pipe_shader_type_from_mesa(_mesa_program_enum_to_shader_stage(target
));
254 st_save_zombie_shader(v
->st
, type
, v
->driver_shader
);
263 * Free all basic program variants.
266 st_release_variants(struct st_context
*st
, struct st_program
*p
)
268 struct st_variant
*v
;
270 for (v
= p
->variants
; v
; ) {
271 struct st_variant
*next
= v
->next
;
272 delete_variant(st
, v
, p
->Base
.Target
);
278 if (p
->state
.tokens
) {
279 ureg_free_tokens(p
->state
.tokens
);
280 p
->state
.tokens
= NULL
;
283 /* Note: Any setup of ->ir.nir that has had pipe->create_*_state called on
284 * it has resulted in the driver taking ownership of the NIR. Those
285 * callers should be NULLing out the nir field in any pipe_shader_state
286 * that might have this called in order to indicate that.
288 * GLSL IR and ARB programs will have set gl_program->nir to the same
289 * shader as ir->ir.nir, so it will be freed by _mesa_delete_program().
294 st_finalize_nir_before_variants(struct nir_shader
*nir
)
296 NIR_PASS_V(nir
, nir_opt_access
);
298 NIR_PASS_V(nir
, nir_split_var_copies
);
299 NIR_PASS_V(nir
, nir_lower_var_copies
);
300 if (nir
->options
->lower_all_io_to_temps
||
301 nir
->options
->lower_all_io_to_elements
||
302 nir
->info
.stage
== MESA_SHADER_VERTEX
||
303 nir
->info
.stage
== MESA_SHADER_GEOMETRY
) {
304 NIR_PASS_V(nir
, nir_lower_io_arrays_to_elements_no_indirects
, false);
305 } else if (nir
->info
.stage
== MESA_SHADER_FRAGMENT
) {
306 NIR_PASS_V(nir
, nir_lower_io_arrays_to_elements_no_indirects
, true);
309 st_nir_assign_vs_in_locations(nir
);
313 * Translate ARB (asm) program to NIR
316 st_translate_prog_to_nir(struct st_context
*st
, struct gl_program
*prog
,
317 gl_shader_stage stage
)
319 struct pipe_screen
*screen
= st
->pipe
->screen
;
320 const struct gl_shader_compiler_options
*options
=
321 &st
->ctx
->Const
.ShaderCompilerOptions
[stage
];
323 /* Translate to NIR */
324 nir_shader
*nir
= prog_to_nir(prog
, options
->NirOptions
);
325 NIR_PASS_V(nir
, nir_lower_regs_to_ssa
); /* turn registers into SSA */
326 nir_validate_shader(nir
, "after st/ptn lower_regs_to_ssa");
328 NIR_PASS_V(nir
, st_nir_lower_wpos_ytransform
, prog
, screen
);
329 NIR_PASS_V(nir
, nir_lower_system_values
);
332 NIR_PASS_V(nir
, nir_opt_constant_folding
);
334 st_finalize_nir_before_variants(nir
);
336 if (st
->allow_st_finalize_nir_twice
)
337 st_finalize_nir(st
, prog
, NULL
, nir
, true);
339 nir_validate_shader(nir
, "after st/glsl finalize_nir");
345 st_prepare_vertex_program(struct st_program
*stp
)
347 struct st_vertex_program
*stvp
= (struct st_vertex_program
*)stp
;
349 stvp
->num_inputs
= 0;
350 memset(stvp
->input_to_index
, ~0, sizeof(stvp
->input_to_index
));
351 memset(stvp
->result_to_output
, ~0, sizeof(stvp
->result_to_output
));
353 /* Determine number of inputs, the mappings between VERT_ATTRIB_x
354 * and TGSI generic input indexes, plus input attrib semantic info.
356 for (unsigned attr
= 0; attr
< VERT_ATTRIB_MAX
; attr
++) {
357 if ((stp
->Base
.info
.inputs_read
& BITFIELD64_BIT(attr
)) != 0) {
358 stvp
->input_to_index
[attr
] = stvp
->num_inputs
;
359 stvp
->index_to_input
[stvp
->num_inputs
] = attr
;
362 if ((stp
->Base
.DualSlotInputs
& BITFIELD64_BIT(attr
)) != 0) {
363 /* add placeholder for second part of a double attribute */
364 stvp
->index_to_input
[stvp
->num_inputs
] = ST_DOUBLE_ATTRIB_PLACEHOLDER
;
369 /* pre-setup potentially unused edgeflag input */
370 stvp
->input_to_index
[VERT_ATTRIB_EDGEFLAG
] = stvp
->num_inputs
;
371 stvp
->index_to_input
[stvp
->num_inputs
] = VERT_ATTRIB_EDGEFLAG
;
373 /* Compute mapping of vertex program outputs to slots. */
374 unsigned num_outputs
= 0;
375 for (unsigned attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
376 if (stp
->Base
.info
.outputs_written
& BITFIELD64_BIT(attr
))
377 stvp
->result_to_output
[attr
] = num_outputs
++;
379 /* pre-setup potentially unused edgeflag output */
380 stvp
->result_to_output
[VARYING_SLOT_EDGE
] = num_outputs
;
384 st_translate_stream_output_info(struct gl_program
*prog
)
386 struct gl_transform_feedback_info
*info
= prog
->sh
.LinkedTransformFeedback
;
390 /* Determine the (default) output register mapping for each output. */
391 unsigned num_outputs
= 0;
392 ubyte output_mapping
[VARYING_SLOT_TESS_MAX
];
393 memset(output_mapping
, 0, sizeof(output_mapping
));
395 for (unsigned attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
396 if (prog
->info
.outputs_written
& BITFIELD64_BIT(attr
))
397 output_mapping
[attr
] = num_outputs
++;
400 /* Translate stream output info. */
401 struct pipe_stream_output_info
*so_info
=
402 &((struct st_program
*)prog
)->state
.stream_output
;
404 for (unsigned i
= 0; i
< info
->NumOutputs
; i
++) {
405 so_info
->output
[i
].register_index
=
406 output_mapping
[info
->Outputs
[i
].OutputRegister
];
407 so_info
->output
[i
].start_component
= info
->Outputs
[i
].ComponentOffset
;
408 so_info
->output
[i
].num_components
= info
->Outputs
[i
].NumComponents
;
409 so_info
->output
[i
].output_buffer
= info
->Outputs
[i
].OutputBuffer
;
410 so_info
->output
[i
].dst_offset
= info
->Outputs
[i
].DstOffset
;
411 so_info
->output
[i
].stream
= info
->Outputs
[i
].StreamId
;
414 for (unsigned i
= 0; i
< PIPE_MAX_SO_BUFFERS
; i
++) {
415 so_info
->stride
[i
] = info
->Buffers
[i
].Stride
;
417 so_info
->num_outputs
= info
->NumOutputs
;
421 * Translate a vertex program.
424 st_translate_vertex_program(struct st_context
*st
,
425 struct st_program
*stp
)
427 struct ureg_program
*ureg
;
428 enum pipe_error error
;
429 unsigned num_outputs
= 0;
431 ubyte output_semantic_name
[VARYING_SLOT_MAX
] = {0};
432 ubyte output_semantic_index
[VARYING_SLOT_MAX
] = {0};
434 if (stp
->Base
.arb
.IsPositionInvariant
)
435 _mesa_insert_mvp_code(st
->ctx
, &stp
->Base
);
437 st_prepare_vertex_program(stp
);
440 if (!stp
->glsl_to_tgsi
) {
441 _mesa_remove_output_reads(&stp
->Base
, PROGRAM_OUTPUT
);
443 /* This determines which states will be updated when the assembly
446 stp
->affected_states
= ST_NEW_VS_STATE
|
448 ST_NEW_VERTEX_ARRAYS
;
450 if (stp
->Base
.Parameters
->NumParameters
)
451 stp
->affected_states
|= ST_NEW_VS_CONSTANTS
;
453 /* Translate to NIR if preferred. */
454 if (st
->pipe
->screen
->get_shader_param(st
->pipe
->screen
,
456 PIPE_SHADER_CAP_PREFERRED_IR
)) {
457 assert(!stp
->glsl_to_tgsi
);
460 ralloc_free(stp
->Base
.nir
);
462 stp
->state
.type
= PIPE_SHADER_IR_NIR
;
463 stp
->Base
.nir
= st_translate_prog_to_nir(st
, &stp
->Base
,
465 /* For st_draw_feedback, we need to generate TGSI too if draw doesn't
473 /* Get semantic names and indices. */
474 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
475 if (stp
->Base
.info
.outputs_written
& BITFIELD64_BIT(attr
)) {
476 unsigned slot
= num_outputs
++;
477 unsigned semantic_name
, semantic_index
;
478 tgsi_get_gl_varying_semantic(attr
, st
->needs_texcoord_semantic
,
479 &semantic_name
, &semantic_index
);
480 output_semantic_name
[slot
] = semantic_name
;
481 output_semantic_index
[slot
] = semantic_index
;
484 /* pre-setup potentially unused edgeflag output */
485 output_semantic_name
[num_outputs
] = TGSI_SEMANTIC_EDGEFLAG
;
486 output_semantic_index
[num_outputs
] = 0;
488 ureg
= ureg_create_with_screen(PIPE_SHADER_VERTEX
, st
->pipe
->screen
);
492 if (stp
->Base
.info
.clip_distance_array_size
)
493 ureg_property(ureg
, TGSI_PROPERTY_NUM_CLIPDIST_ENABLED
,
494 stp
->Base
.info
.clip_distance_array_size
);
495 if (stp
->Base
.info
.cull_distance_array_size
)
496 ureg_property(ureg
, TGSI_PROPERTY_NUM_CULLDIST_ENABLED
,
497 stp
->Base
.info
.cull_distance_array_size
);
499 if (ST_DEBUG
& DEBUG_MESA
) {
500 _mesa_print_program(&stp
->Base
);
501 _mesa_print_program_parameters(st
->ctx
, &stp
->Base
);
505 struct st_vertex_program
*stvp
= (struct st_vertex_program
*)stp
;
507 if (stp
->glsl_to_tgsi
) {
508 error
= st_translate_program(st
->ctx
,
515 stvp
->input_to_index
,
516 NULL
, /* inputSlotToAttr */
517 NULL
, /* input semantic name */
518 NULL
, /* input semantic index */
519 NULL
, /* interp mode */
522 stvp
->result_to_output
,
523 output_semantic_name
,
524 output_semantic_index
);
526 st_translate_stream_output_info(&stp
->Base
);
528 free_glsl_to_tgsi_visitor(stp
->glsl_to_tgsi
);
530 error
= st_translate_mesa_program(st
->ctx
,
536 stvp
->input_to_index
,
537 NULL
, /* input semantic name */
538 NULL
, /* input semantic index */
542 stvp
->result_to_output
,
543 output_semantic_name
,
544 output_semantic_index
);
547 debug_printf("%s: failed to translate Mesa program:\n", __func__
);
548 _mesa_print_program(&stp
->Base
);
553 stp
->state
.tokens
= ureg_get_tokens(ureg
, NULL
);
556 if (stp
->glsl_to_tgsi
) {
557 stp
->glsl_to_tgsi
= NULL
;
558 st_store_ir_in_disk_cache(st
, &stp
->Base
, false);
561 return stp
->state
.tokens
!= NULL
;
564 static const gl_state_index16 depth_range_state
[STATE_LENGTH
] =
565 { STATE_DEPTH_RANGE
};
567 static struct st_common_variant
*
568 st_create_vp_variant(struct st_context
*st
,
569 struct st_program
*stvp
,
570 const struct st_common_variant_key
*key
)
572 struct st_common_variant
*vpv
= CALLOC_STRUCT(st_common_variant
);
573 struct pipe_context
*pipe
= st
->pipe
;
574 struct pipe_screen
*screen
= pipe
->screen
;
575 struct pipe_shader_state state
= {0};
577 static const gl_state_index16 point_size_state
[STATE_LENGTH
] =
578 { STATE_INTERNAL
, STATE_POINT_SIZE_CLAMPED
, 0 };
579 struct gl_program_parameter_list
*params
= stvp
->Base
.Parameters
;
583 state
.stream_output
= stvp
->state
.stream_output
;
585 if (stvp
->state
.type
== PIPE_SHADER_IR_NIR
&&
586 (!key
->is_draw_shader
|| draw_has_llvm())) {
587 bool finalize
= false;
589 state
.type
= PIPE_SHADER_IR_NIR
;
590 state
.ir
.nir
= nir_shader_clone(NULL
, stvp
->Base
.nir
);
591 if (key
->clamp_color
) {
592 NIR_PASS_V(state
.ir
.nir
, nir_lower_clamp_color_outputs
);
595 if (key
->passthrough_edgeflags
) {
596 NIR_PASS_V(state
.ir
.nir
, nir_lower_passthrough_edgeflags
);
600 if (key
->lower_point_size
) {
601 _mesa_add_state_reference(params
, point_size_state
);
602 NIR_PASS_V(state
.ir
.nir
, nir_lower_point_size_mov
,
607 if (key
->lower_ucp
) {
608 bool can_compact
= screen
->get_param(screen
,
609 PIPE_CAP_NIR_COMPACT_ARRAYS
);
611 bool use_eye
= st
->ctx
->_Shader
->CurrentProgram
[MESA_SHADER_VERTEX
] != NULL
;
612 gl_state_index16 clipplane_state
[MAX_CLIP_PLANES
][STATE_LENGTH
];
613 for (int i
= 0; i
< MAX_CLIP_PLANES
; ++i
) {
615 clipplane_state
[i
][0] = STATE_CLIPPLANE
;
616 clipplane_state
[i
][1] = i
;
618 clipplane_state
[i
][0] = STATE_INTERNAL
;
619 clipplane_state
[i
][1] = STATE_CLIP_INTERNAL
;
620 clipplane_state
[i
][2] = i
;
622 _mesa_add_state_reference(params
, clipplane_state
[i
]);
625 NIR_PASS_V(state
.ir
.nir
, nir_lower_clip_vs
, key
->lower_ucp
,
626 true, can_compact
, clipplane_state
);
627 NIR_PASS_V(state
.ir
.nir
, nir_lower_io_to_temporaries
,
628 nir_shader_get_entrypoint(state
.ir
.nir
), true, false);
629 NIR_PASS_V(state
.ir
.nir
, nir_lower_global_vars_to_local
);
633 if (finalize
|| !st
->allow_st_finalize_nir_twice
) {
634 st_finalize_nir(st
, &stvp
->Base
, stvp
->shader_program
, state
.ir
.nir
,
637 /* Some of the lowering above may have introduced new varyings */
638 nir_shader_gather_info(state
.ir
.nir
,
639 nir_shader_get_entrypoint(state
.ir
.nir
));
642 if (ST_DEBUG
& DEBUG_PRINT_IR
)
643 nir_print_shader(state
.ir
.nir
, stderr
);
645 if (key
->is_draw_shader
)
646 vpv
->base
.driver_shader
= draw_create_vertex_shader(st
->draw
, &state
);
648 vpv
->base
.driver_shader
= pipe
->create_vs_state(pipe
, &state
);
653 state
.type
= PIPE_SHADER_IR_TGSI
;
654 state
.tokens
= tgsi_dup_tokens(stvp
->state
.tokens
);
656 /* Emulate features. */
657 if (key
->clamp_color
|| key
->passthrough_edgeflags
) {
658 const struct tgsi_token
*tokens
;
660 (key
->clamp_color
? TGSI_EMU_CLAMP_COLOR_OUTPUTS
: 0) |
661 (key
->passthrough_edgeflags
? TGSI_EMU_PASSTHROUGH_EDGEFLAG
: 0);
663 tokens
= tgsi_emulate(state
.tokens
, flags
);
666 tgsi_free_tokens(state
.tokens
);
667 state
.tokens
= tokens
;
669 fprintf(stderr
, "mesa: cannot emulate deprecated features\n");
673 if (key
->lower_depth_clamp
) {
674 unsigned depth_range_const
=
675 _mesa_add_state_reference(params
, depth_range_state
);
677 const struct tgsi_token
*tokens
;
678 tokens
= st_tgsi_lower_depth_clamp(state
.tokens
, depth_range_const
,
679 key
->clip_negative_one_to_one
);
680 if (tokens
!= state
.tokens
)
681 tgsi_free_tokens(state
.tokens
);
682 state
.tokens
= tokens
;
685 if (ST_DEBUG
& DEBUG_PRINT_IR
)
686 tgsi_dump(state
.tokens
, 0);
688 if (key
->is_draw_shader
)
689 vpv
->base
.driver_shader
= draw_create_vertex_shader(st
->draw
, &state
);
691 vpv
->base
.driver_shader
= pipe
->create_vs_state(pipe
, &state
);
698 * Find/create a vertex program variant.
700 struct st_common_variant
*
701 st_get_vp_variant(struct st_context
*st
,
702 struct st_program
*stp
,
703 const struct st_common_variant_key
*key
)
705 struct st_vertex_program
*stvp
= (struct st_vertex_program
*)stp
;
706 struct st_common_variant
*vpv
;
708 /* Search for existing variant */
709 for (vpv
= st_common_variant(stp
->variants
); vpv
;
710 vpv
= st_common_variant(vpv
->base
.next
)) {
711 if (memcmp(&vpv
->key
, key
, sizeof(*key
)) == 0) {
718 vpv
= st_create_vp_variant(st
, stp
, key
);
720 vpv
->base
.st
= key
->st
;
722 unsigned num_inputs
= stvp
->num_inputs
+ key
->passthrough_edgeflags
;
723 for (unsigned index
= 0; index
< num_inputs
; ++index
) {
724 unsigned attr
= stvp
->index_to_input
[index
];
725 if (attr
== ST_DOUBLE_ATTRIB_PLACEHOLDER
)
727 vpv
->vert_attrib_mask
|= 1u << attr
;
730 /* insert into list */
731 vpv
->base
.next
= stp
->variants
;
732 stp
->variants
= &vpv
->base
;
741 * Translate a Mesa fragment shader into a TGSI shader.
744 st_translate_fragment_program(struct st_context
*st
,
745 struct st_program
*stfp
)
747 /* Non-GLSL programs: */
748 if (!stfp
->glsl_to_tgsi
) {
749 _mesa_remove_output_reads(&stfp
->Base
, PROGRAM_OUTPUT
);
750 if (st
->ctx
->Const
.GLSLFragCoordIsSysVal
)
751 _mesa_program_fragment_position_to_sysval(&stfp
->Base
);
753 /* This determines which states will be updated when the assembly
756 * fragment.position and glDrawPixels always use constants.
758 stfp
->affected_states
= ST_NEW_FS_STATE
|
759 ST_NEW_SAMPLE_SHADING
|
763 /* Just set them for ATI_fs unconditionally. */
764 stfp
->affected_states
|= ST_NEW_FS_SAMPLER_VIEWS
|
768 if (stfp
->Base
.SamplersUsed
)
769 stfp
->affected_states
|= ST_NEW_FS_SAMPLER_VIEWS
|
773 /* Translate to NIR. */
775 st
->pipe
->screen
->get_shader_param(st
->pipe
->screen
,
776 PIPE_SHADER_FRAGMENT
,
777 PIPE_SHADER_CAP_PREFERRED_IR
)) {
779 st_translate_prog_to_nir(st
, &stfp
->Base
, MESA_SHADER_FRAGMENT
);
782 ralloc_free(stfp
->Base
.nir
);
783 stfp
->state
.type
= PIPE_SHADER_IR_NIR
;
784 stfp
->Base
.nir
= nir
;
789 ubyte outputMapping
[2 * FRAG_RESULT_MAX
];
790 ubyte inputMapping
[VARYING_SLOT_MAX
];
791 ubyte inputSlotToAttr
[VARYING_SLOT_MAX
];
792 ubyte interpMode
[PIPE_MAX_SHADER_INPUTS
]; /* XXX size? */
794 GLbitfield64 inputsRead
;
795 struct ureg_program
*ureg
;
797 GLboolean write_all
= GL_FALSE
;
799 ubyte input_semantic_name
[PIPE_MAX_SHADER_INPUTS
];
800 ubyte input_semantic_index
[PIPE_MAX_SHADER_INPUTS
];
801 uint fs_num_inputs
= 0;
803 ubyte fs_output_semantic_name
[PIPE_MAX_SHADER_OUTPUTS
];
804 ubyte fs_output_semantic_index
[PIPE_MAX_SHADER_OUTPUTS
];
805 uint fs_num_outputs
= 0;
807 memset(inputSlotToAttr
, ~0, sizeof(inputSlotToAttr
));
810 * Convert Mesa program inputs to TGSI input register semantics.
812 inputsRead
= stfp
->Base
.info
.inputs_read
;
813 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
814 if ((inputsRead
& BITFIELD64_BIT(attr
)) != 0) {
815 const GLuint slot
= fs_num_inputs
++;
817 inputMapping
[attr
] = slot
;
818 inputSlotToAttr
[slot
] = attr
;
821 case VARYING_SLOT_POS
:
822 input_semantic_name
[slot
] = TGSI_SEMANTIC_POSITION
;
823 input_semantic_index
[slot
] = 0;
824 interpMode
[slot
] = TGSI_INTERPOLATE_LINEAR
;
826 case VARYING_SLOT_COL0
:
827 input_semantic_name
[slot
] = TGSI_SEMANTIC_COLOR
;
828 input_semantic_index
[slot
] = 0;
829 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
830 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_COLOR
;
832 case VARYING_SLOT_COL1
:
833 input_semantic_name
[slot
] = TGSI_SEMANTIC_COLOR
;
834 input_semantic_index
[slot
] = 1;
835 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
836 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_COLOR
;
838 case VARYING_SLOT_FOGC
:
839 input_semantic_name
[slot
] = TGSI_SEMANTIC_FOG
;
840 input_semantic_index
[slot
] = 0;
841 interpMode
[slot
] = TGSI_INTERPOLATE_PERSPECTIVE
;
843 case VARYING_SLOT_FACE
:
844 input_semantic_name
[slot
] = TGSI_SEMANTIC_FACE
;
845 input_semantic_index
[slot
] = 0;
846 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
848 case VARYING_SLOT_PRIMITIVE_ID
:
849 input_semantic_name
[slot
] = TGSI_SEMANTIC_PRIMID
;
850 input_semantic_index
[slot
] = 0;
851 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
853 case VARYING_SLOT_LAYER
:
854 input_semantic_name
[slot
] = TGSI_SEMANTIC_LAYER
;
855 input_semantic_index
[slot
] = 0;
856 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
858 case VARYING_SLOT_VIEWPORT
:
859 input_semantic_name
[slot
] = TGSI_SEMANTIC_VIEWPORT_INDEX
;
860 input_semantic_index
[slot
] = 0;
861 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
863 case VARYING_SLOT_CLIP_DIST0
:
864 input_semantic_name
[slot
] = TGSI_SEMANTIC_CLIPDIST
;
865 input_semantic_index
[slot
] = 0;
866 interpMode
[slot
] = TGSI_INTERPOLATE_PERSPECTIVE
;
868 case VARYING_SLOT_CLIP_DIST1
:
869 input_semantic_name
[slot
] = TGSI_SEMANTIC_CLIPDIST
;
870 input_semantic_index
[slot
] = 1;
871 interpMode
[slot
] = TGSI_INTERPOLATE_PERSPECTIVE
;
873 case VARYING_SLOT_CULL_DIST0
:
874 case VARYING_SLOT_CULL_DIST1
:
875 /* these should have been lowered by GLSL */
878 /* In most cases, there is nothing special about these
879 * inputs, so adopt a convention to use the generic
880 * semantic name and the mesa VARYING_SLOT_ number as the
883 * All that is required is that the vertex shader labels
884 * its own outputs similarly, and that the vertex shader
885 * generates at least every output required by the
886 * fragment shader plus fixed-function hardware (such as
889 * However, some drivers may need us to identify the PNTC and TEXi
890 * varyings if, for example, their capability to replace them with
891 * sprite coordinates is limited.
893 case VARYING_SLOT_PNTC
:
894 if (st
->needs_texcoord_semantic
) {
895 input_semantic_name
[slot
] = TGSI_SEMANTIC_PCOORD
;
896 input_semantic_index
[slot
] = 0;
897 interpMode
[slot
] = TGSI_INTERPOLATE_LINEAR
;
901 case VARYING_SLOT_TEX0
:
902 case VARYING_SLOT_TEX1
:
903 case VARYING_SLOT_TEX2
:
904 case VARYING_SLOT_TEX3
:
905 case VARYING_SLOT_TEX4
:
906 case VARYING_SLOT_TEX5
:
907 case VARYING_SLOT_TEX6
:
908 case VARYING_SLOT_TEX7
:
909 if (st
->needs_texcoord_semantic
) {
910 input_semantic_name
[slot
] = TGSI_SEMANTIC_TEXCOORD
;
911 input_semantic_index
[slot
] = attr
- VARYING_SLOT_TEX0
;
912 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
913 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_PERSPECTIVE
;
917 case VARYING_SLOT_VAR0
:
919 /* Semantic indices should be zero-based because drivers may choose
920 * to assign a fixed slot determined by that index.
921 * This is useful because ARB_separate_shader_objects uses location
922 * qualifiers for linkage, and if the semantic index corresponds to
923 * these locations, linkage passes in the driver become unecessary.
925 * If needs_texcoord_semantic is true, no semantic indices will be
926 * consumed for the TEXi varyings, and we can base the locations of
927 * the user varyings on VAR0. Otherwise, we use TEX0 as base index.
929 assert(attr
>= VARYING_SLOT_VAR0
|| attr
== VARYING_SLOT_PNTC
||
930 (attr
>= VARYING_SLOT_TEX0
&& attr
<= VARYING_SLOT_TEX7
));
931 input_semantic_name
[slot
] = TGSI_SEMANTIC_GENERIC
;
932 input_semantic_index
[slot
] = st_get_generic_varying_index(st
, attr
);
933 if (attr
== VARYING_SLOT_PNTC
)
934 interpMode
[slot
] = TGSI_INTERPOLATE_LINEAR
;
936 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
937 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_PERSPECTIVE
;
943 inputMapping
[attr
] = -1;
948 * Semantics and mapping for outputs
950 GLbitfield64 outputsWritten
= stfp
->Base
.info
.outputs_written
;
952 /* if z is written, emit that first */
953 if (outputsWritten
& BITFIELD64_BIT(FRAG_RESULT_DEPTH
)) {
954 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_POSITION
;
955 fs_output_semantic_index
[fs_num_outputs
] = 0;
956 outputMapping
[FRAG_RESULT_DEPTH
] = fs_num_outputs
;
958 outputsWritten
&= ~(1 << FRAG_RESULT_DEPTH
);
961 if (outputsWritten
& BITFIELD64_BIT(FRAG_RESULT_STENCIL
)) {
962 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_STENCIL
;
963 fs_output_semantic_index
[fs_num_outputs
] = 0;
964 outputMapping
[FRAG_RESULT_STENCIL
] = fs_num_outputs
;
966 outputsWritten
&= ~(1 << FRAG_RESULT_STENCIL
);
969 if (outputsWritten
& BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK
)) {
970 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_SAMPLEMASK
;
971 fs_output_semantic_index
[fs_num_outputs
] = 0;
972 outputMapping
[FRAG_RESULT_SAMPLE_MASK
] = fs_num_outputs
;
974 outputsWritten
&= ~(1 << FRAG_RESULT_SAMPLE_MASK
);
977 /* handle remaining outputs (color) */
978 for (attr
= 0; attr
< ARRAY_SIZE(outputMapping
); attr
++) {
979 const GLbitfield64 written
= attr
< FRAG_RESULT_MAX
? outputsWritten
:
980 stfp
->Base
.SecondaryOutputsWritten
;
981 const unsigned loc
= attr
% FRAG_RESULT_MAX
;
983 if (written
& BITFIELD64_BIT(loc
)) {
985 case FRAG_RESULT_DEPTH
:
986 case FRAG_RESULT_STENCIL
:
987 case FRAG_RESULT_SAMPLE_MASK
:
991 case FRAG_RESULT_COLOR
:
992 write_all
= GL_TRUE
; /* fallthrough */
995 assert(loc
== FRAG_RESULT_COLOR
||
996 (FRAG_RESULT_DATA0
<= loc
&& loc
< FRAG_RESULT_MAX
));
998 index
= (loc
== FRAG_RESULT_COLOR
) ? 0 : (loc
- FRAG_RESULT_DATA0
);
1000 if (attr
>= FRAG_RESULT_MAX
) {
1001 /* Secondary color for dual source blending. */
1006 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_COLOR
;
1007 fs_output_semantic_index
[fs_num_outputs
] = index
;
1008 outputMapping
[attr
] = fs_num_outputs
;
1017 ureg
= ureg_create_with_screen(PIPE_SHADER_FRAGMENT
, st
->pipe
->screen
);
1021 if (ST_DEBUG
& DEBUG_MESA
) {
1022 _mesa_print_program(&stfp
->Base
);
1023 _mesa_print_program_parameters(st
->ctx
, &stfp
->Base
);
1026 if (write_all
== GL_TRUE
)
1027 ureg_property(ureg
, TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS
, 1);
1029 if (stfp
->Base
.info
.fs
.depth_layout
!= FRAG_DEPTH_LAYOUT_NONE
) {
1030 switch (stfp
->Base
.info
.fs
.depth_layout
) {
1031 case FRAG_DEPTH_LAYOUT_ANY
:
1032 ureg_property(ureg
, TGSI_PROPERTY_FS_DEPTH_LAYOUT
,
1033 TGSI_FS_DEPTH_LAYOUT_ANY
);
1035 case FRAG_DEPTH_LAYOUT_GREATER
:
1036 ureg_property(ureg
, TGSI_PROPERTY_FS_DEPTH_LAYOUT
,
1037 TGSI_FS_DEPTH_LAYOUT_GREATER
);
1039 case FRAG_DEPTH_LAYOUT_LESS
:
1040 ureg_property(ureg
, TGSI_PROPERTY_FS_DEPTH_LAYOUT
,
1041 TGSI_FS_DEPTH_LAYOUT_LESS
);
1043 case FRAG_DEPTH_LAYOUT_UNCHANGED
:
1044 ureg_property(ureg
, TGSI_PROPERTY_FS_DEPTH_LAYOUT
,
1045 TGSI_FS_DEPTH_LAYOUT_UNCHANGED
);
1052 if (stfp
->glsl_to_tgsi
) {
1053 st_translate_program(st
->ctx
,
1054 PIPE_SHADER_FRAGMENT
,
1062 input_semantic_name
,
1063 input_semantic_index
,
1068 fs_output_semantic_name
,
1069 fs_output_semantic_index
);
1071 free_glsl_to_tgsi_visitor(stfp
->glsl_to_tgsi
);
1072 } else if (stfp
->ati_fs
)
1073 st_translate_atifs_program(ureg
,
1079 input_semantic_name
,
1080 input_semantic_index
,
1085 fs_output_semantic_name
,
1086 fs_output_semantic_index
);
1088 st_translate_mesa_program(st
->ctx
,
1089 PIPE_SHADER_FRAGMENT
,
1095 input_semantic_name
,
1096 input_semantic_index
,
1101 fs_output_semantic_name
,
1102 fs_output_semantic_index
);
1104 stfp
->state
.tokens
= ureg_get_tokens(ureg
, NULL
);
1107 if (stfp
->glsl_to_tgsi
) {
1108 stfp
->glsl_to_tgsi
= NULL
;
1109 st_store_ir_in_disk_cache(st
, &stfp
->Base
, false);
1112 return stfp
->state
.tokens
!= NULL
;
1115 static struct st_fp_variant
*
1116 st_create_fp_variant(struct st_context
*st
,
1117 struct st_program
*stfp
,
1118 const struct st_fp_variant_key
*key
)
1120 struct pipe_context
*pipe
= st
->pipe
;
1121 struct st_fp_variant
*variant
= CALLOC_STRUCT(st_fp_variant
);
1122 struct pipe_shader_state state
= {0};
1123 struct gl_program_parameter_list
*params
= stfp
->Base
.Parameters
;
1124 static const gl_state_index16 texcoord_state
[STATE_LENGTH
] =
1125 { STATE_INTERNAL
, STATE_CURRENT_ATTRIB
, VERT_ATTRIB_TEX0
};
1126 static const gl_state_index16 scale_state
[STATE_LENGTH
] =
1127 { STATE_INTERNAL
, STATE_PT_SCALE
};
1128 static const gl_state_index16 bias_state
[STATE_LENGTH
] =
1129 { STATE_INTERNAL
, STATE_PT_BIAS
};
1130 static const gl_state_index16 alpha_ref_state
[STATE_LENGTH
] =
1131 { STATE_INTERNAL
, STATE_ALPHA_REF
};
1136 if (stfp
->state
.type
== PIPE_SHADER_IR_NIR
) {
1137 bool finalize
= false;
1139 state
.type
= PIPE_SHADER_IR_NIR
;
1140 state
.ir
.nir
= nir_shader_clone(NULL
, stfp
->Base
.nir
);
1142 if (key
->clamp_color
) {
1143 NIR_PASS_V(state
.ir
.nir
, nir_lower_clamp_color_outputs
);
1147 if (key
->lower_flatshade
) {
1148 NIR_PASS_V(state
.ir
.nir
, nir_lower_flatshade
);
1152 if (key
->lower_alpha_func
!= COMPARE_FUNC_NEVER
) {
1153 _mesa_add_state_reference(params
, alpha_ref_state
);
1154 NIR_PASS_V(state
.ir
.nir
, nir_lower_alpha_test
, key
->lower_alpha_func
,
1155 false, alpha_ref_state
);
1159 if (key
->lower_two_sided_color
) {
1160 NIR_PASS_V(state
.ir
.nir
, nir_lower_two_sided_color
);
1164 if (key
->persample_shading
) {
1165 nir_shader
*shader
= state
.ir
.nir
;
1166 nir_foreach_variable(var
, &shader
->inputs
)
1167 var
->data
.sample
= true;
1171 assert(!(key
->bitmap
&& key
->drawpixels
));
1175 nir_lower_bitmap_options options
= {0};
1177 variant
->bitmap_sampler
= ffs(~stfp
->Base
.SamplersUsed
) - 1;
1178 options
.sampler
= variant
->bitmap_sampler
;
1179 options
.swizzle_xxxx
= st
->bitmap
.tex_format
== PIPE_FORMAT_R8_UNORM
;
1181 NIR_PASS_V(state
.ir
.nir
, nir_lower_bitmap
, &options
);
1185 /* glDrawPixels (color only) */
1186 if (key
->drawpixels
) {
1187 nir_lower_drawpixels_options options
= {{0}};
1188 unsigned samplers_used
= stfp
->Base
.SamplersUsed
;
1190 /* Find the first unused slot. */
1191 variant
->drawpix_sampler
= ffs(~samplers_used
) - 1;
1192 options
.drawpix_sampler
= variant
->drawpix_sampler
;
1193 samplers_used
|= (1 << variant
->drawpix_sampler
);
1195 options
.pixel_maps
= key
->pixelMaps
;
1196 if (key
->pixelMaps
) {
1197 variant
->pixelmap_sampler
= ffs(~samplers_used
) - 1;
1198 options
.pixelmap_sampler
= variant
->pixelmap_sampler
;
1201 options
.scale_and_bias
= key
->scaleAndBias
;
1202 if (key
->scaleAndBias
) {
1203 _mesa_add_state_reference(params
, scale_state
);
1204 memcpy(options
.scale_state_tokens
, scale_state
,
1205 sizeof(options
.scale_state_tokens
));
1206 _mesa_add_state_reference(params
, bias_state
);
1207 memcpy(options
.bias_state_tokens
, bias_state
,
1208 sizeof(options
.bias_state_tokens
));
1211 _mesa_add_state_reference(params
, texcoord_state
);
1212 memcpy(options
.texcoord_state_tokens
, texcoord_state
,
1213 sizeof(options
.texcoord_state_tokens
));
1215 NIR_PASS_V(state
.ir
.nir
, nir_lower_drawpixels
, &options
);
1219 if (unlikely(key
->external
.lower_nv12
|| key
->external
.lower_iyuv
||
1220 key
->external
.lower_xy_uxvx
|| key
->external
.lower_yx_xuxv
||
1221 key
->external
.lower_ayuv
|| key
->external
.lower_xyuv
)) {
1222 nir_lower_tex_options options
= {0};
1223 options
.lower_y_uv_external
= key
->external
.lower_nv12
;
1224 options
.lower_y_u_v_external
= key
->external
.lower_iyuv
;
1225 options
.lower_xy_uxvx_external
= key
->external
.lower_xy_uxvx
;
1226 options
.lower_yx_xuxv_external
= key
->external
.lower_yx_xuxv
;
1227 options
.lower_ayuv_external
= key
->external
.lower_ayuv
;
1228 options
.lower_xyuv_external
= key
->external
.lower_xyuv
;
1229 NIR_PASS_V(state
.ir
.nir
, nir_lower_tex
, &options
);
1233 if (finalize
|| !st
->allow_st_finalize_nir_twice
) {
1234 st_finalize_nir(st
, &stfp
->Base
, stfp
->shader_program
, state
.ir
.nir
,
1238 /* This pass needs to happen *after* nir_lower_sampler */
1239 if (unlikely(key
->external
.lower_nv12
|| key
->external
.lower_iyuv
||
1240 key
->external
.lower_xy_uxvx
|| key
->external
.lower_yx_xuxv
)) {
1241 NIR_PASS_V(state
.ir
.nir
, st_nir_lower_tex_src_plane
,
1242 ~stfp
->Base
.SamplersUsed
,
1243 key
->external
.lower_nv12
|| key
->external
.lower_xy_uxvx
||
1244 key
->external
.lower_yx_xuxv
,
1245 key
->external
.lower_iyuv
);
1249 if (finalize
|| !st
->allow_st_finalize_nir_twice
) {
1250 /* Some of the lowering above may have introduced new varyings */
1251 nir_shader_gather_info(state
.ir
.nir
,
1252 nir_shader_get_entrypoint(state
.ir
.nir
));
1254 struct pipe_screen
*screen
= pipe
->screen
;
1255 if (screen
->finalize_nir
)
1256 screen
->finalize_nir(screen
, state
.ir
.nir
, false);
1259 if (ST_DEBUG
& DEBUG_PRINT_IR
)
1260 nir_print_shader(state
.ir
.nir
, stderr
);
1262 variant
->base
.driver_shader
= pipe
->create_fs_state(pipe
, &state
);
1263 variant
->key
= *key
;
1268 state
.tokens
= stfp
->state
.tokens
;
1270 assert(!(key
->bitmap
&& key
->drawpixels
));
1272 /* Fix texture targets and add fog for ATI_fs */
1274 const struct tgsi_token
*tokens
= st_fixup_atifs(state
.tokens
, key
);
1277 state
.tokens
= tokens
;
1279 fprintf(stderr
, "mesa: cannot post-process ATI_fs\n");
1282 /* Emulate features. */
1283 if (key
->clamp_color
|| key
->persample_shading
) {
1284 const struct tgsi_token
*tokens
;
1286 (key
->clamp_color
? TGSI_EMU_CLAMP_COLOR_OUTPUTS
: 0) |
1287 (key
->persample_shading
? TGSI_EMU_FORCE_PERSAMPLE_INTERP
: 0);
1289 tokens
= tgsi_emulate(state
.tokens
, flags
);
1292 if (state
.tokens
!= stfp
->state
.tokens
)
1293 tgsi_free_tokens(state
.tokens
);
1294 state
.tokens
= tokens
;
1296 fprintf(stderr
, "mesa: cannot emulate deprecated features\n");
1301 const struct tgsi_token
*tokens
;
1303 variant
->bitmap_sampler
= ffs(~stfp
->Base
.SamplersUsed
) - 1;
1305 tokens
= st_get_bitmap_shader(state
.tokens
,
1306 st
->internal_target
,
1307 variant
->bitmap_sampler
,
1308 st
->needs_texcoord_semantic
,
1309 st
->bitmap
.tex_format
==
1310 PIPE_FORMAT_R8_UNORM
);
1313 if (state
.tokens
!= stfp
->state
.tokens
)
1314 tgsi_free_tokens(state
.tokens
);
1315 state
.tokens
= tokens
;
1317 fprintf(stderr
, "mesa: cannot create a shader for glBitmap\n");
1320 /* glDrawPixels (color only) */
1321 if (key
->drawpixels
) {
1322 const struct tgsi_token
*tokens
;
1323 unsigned scale_const
= 0, bias_const
= 0, texcoord_const
= 0;
1325 /* Find the first unused slot. */
1326 variant
->drawpix_sampler
= ffs(~stfp
->Base
.SamplersUsed
) - 1;
1328 if (key
->pixelMaps
) {
1329 unsigned samplers_used
= stfp
->Base
.SamplersUsed
|
1330 (1 << variant
->drawpix_sampler
);
1332 variant
->pixelmap_sampler
= ffs(~samplers_used
) - 1;
1335 if (key
->scaleAndBias
) {
1336 scale_const
= _mesa_add_state_reference(params
, scale_state
);
1337 bias_const
= _mesa_add_state_reference(params
, bias_state
);
1340 texcoord_const
= _mesa_add_state_reference(params
, texcoord_state
);
1342 tokens
= st_get_drawpix_shader(state
.tokens
,
1343 st
->needs_texcoord_semantic
,
1344 key
->scaleAndBias
, scale_const
,
1345 bias_const
, key
->pixelMaps
,
1346 variant
->drawpix_sampler
,
1347 variant
->pixelmap_sampler
,
1348 texcoord_const
, st
->internal_target
);
1351 if (state
.tokens
!= stfp
->state
.tokens
)
1352 tgsi_free_tokens(state
.tokens
);
1353 state
.tokens
= tokens
;
1355 fprintf(stderr
, "mesa: cannot create a shader for glDrawPixels\n");
1358 if (unlikely(key
->external
.lower_nv12
|| key
->external
.lower_iyuv
||
1359 key
->external
.lower_xy_uxvx
|| key
->external
.lower_yx_xuxv
)) {
1360 const struct tgsi_token
*tokens
;
1362 /* samplers inserted would conflict, but this should be unpossible: */
1363 assert(!(key
->bitmap
|| key
->drawpixels
));
1365 tokens
= st_tgsi_lower_yuv(state
.tokens
,
1366 ~stfp
->Base
.SamplersUsed
,
1367 key
->external
.lower_nv12
||
1368 key
->external
.lower_xy_uxvx
||
1369 key
->external
.lower_yx_xuxv
,
1370 key
->external
.lower_iyuv
);
1372 if (state
.tokens
!= stfp
->state
.tokens
)
1373 tgsi_free_tokens(state
.tokens
);
1374 state
.tokens
= tokens
;
1376 fprintf(stderr
, "mesa: cannot create a shader for samplerExternalOES\n");
1380 if (key
->lower_depth_clamp
) {
1381 unsigned depth_range_const
= _mesa_add_state_reference(params
, depth_range_state
);
1383 const struct tgsi_token
*tokens
;
1384 tokens
= st_tgsi_lower_depth_clamp_fs(state
.tokens
, depth_range_const
);
1385 if (state
.tokens
!= stfp
->state
.tokens
)
1386 tgsi_free_tokens(state
.tokens
);
1387 state
.tokens
= tokens
;
1390 if (ST_DEBUG
& DEBUG_PRINT_IR
)
1391 tgsi_dump(state
.tokens
, 0);
1393 /* fill in variant */
1394 variant
->base
.driver_shader
= pipe
->create_fs_state(pipe
, &state
);
1395 variant
->key
= *key
;
1397 if (state
.tokens
!= stfp
->state
.tokens
)
1398 tgsi_free_tokens(state
.tokens
);
1403 * Translate fragment program if needed.
1405 struct st_fp_variant
*
1406 st_get_fp_variant(struct st_context
*st
,
1407 struct st_program
*stfp
,
1408 const struct st_fp_variant_key
*key
)
1410 struct st_fp_variant
*fpv
;
1412 /* Search for existing variant */
1413 for (fpv
= st_fp_variant(stfp
->variants
); fpv
;
1414 fpv
= st_fp_variant(fpv
->base
.next
)) {
1415 if (memcmp(&fpv
->key
, key
, sizeof(*key
)) == 0) {
1422 fpv
= st_create_fp_variant(st
, stfp
, key
);
1424 fpv
->base
.st
= key
->st
;
1426 if (key
->bitmap
|| key
->drawpixels
) {
1427 /* Regular variants should always come before the
1428 * bitmap & drawpixels variants, (unless there
1429 * are no regular variants) so that
1430 * st_update_fp can take a fast path when
1431 * shader_has_one_variant is set.
1433 if (!stfp
->variants
) {
1434 stfp
->variants
= &fpv
->base
;
1436 /* insert into list after the first one */
1437 fpv
->base
.next
= stfp
->variants
->next
;
1438 stfp
->variants
->next
= &fpv
->base
;
1441 /* insert into list */
1442 fpv
->base
.next
= stfp
->variants
;
1443 stfp
->variants
= &fpv
->base
;
1452 * Translate a program. This is common code for geometry and tessellation
1456 st_translate_common_program(struct st_context
*st
,
1457 struct st_program
*stp
)
1459 struct gl_program
*prog
= &stp
->Base
;
1460 enum pipe_shader_type stage
=
1461 pipe_shader_type_from_mesa(stp
->Base
.info
.stage
);
1462 struct ureg_program
*ureg
= ureg_create_with_screen(stage
, st
->pipe
->screen
);
1468 case PIPE_SHADER_TESS_CTRL
:
1469 ureg_property(ureg
, TGSI_PROPERTY_TCS_VERTICES_OUT
,
1470 stp
->Base
.info
.tess
.tcs_vertices_out
);
1473 case PIPE_SHADER_TESS_EVAL
:
1474 if (stp
->Base
.info
.tess
.primitive_mode
== GL_ISOLINES
)
1475 ureg_property(ureg
, TGSI_PROPERTY_TES_PRIM_MODE
, GL_LINES
);
1477 ureg_property(ureg
, TGSI_PROPERTY_TES_PRIM_MODE
,
1478 stp
->Base
.info
.tess
.primitive_mode
);
1480 STATIC_ASSERT((TESS_SPACING_EQUAL
+ 1) % 3 == PIPE_TESS_SPACING_EQUAL
);
1481 STATIC_ASSERT((TESS_SPACING_FRACTIONAL_ODD
+ 1) % 3 ==
1482 PIPE_TESS_SPACING_FRACTIONAL_ODD
);
1483 STATIC_ASSERT((TESS_SPACING_FRACTIONAL_EVEN
+ 1) % 3 ==
1484 PIPE_TESS_SPACING_FRACTIONAL_EVEN
);
1486 ureg_property(ureg
, TGSI_PROPERTY_TES_SPACING
,
1487 (stp
->Base
.info
.tess
.spacing
+ 1) % 3);
1489 ureg_property(ureg
, TGSI_PROPERTY_TES_VERTEX_ORDER_CW
,
1490 !stp
->Base
.info
.tess
.ccw
);
1491 ureg_property(ureg
, TGSI_PROPERTY_TES_POINT_MODE
,
1492 stp
->Base
.info
.tess
.point_mode
);
1495 case PIPE_SHADER_GEOMETRY
:
1496 ureg_property(ureg
, TGSI_PROPERTY_GS_INPUT_PRIM
,
1497 stp
->Base
.info
.gs
.input_primitive
);
1498 ureg_property(ureg
, TGSI_PROPERTY_GS_OUTPUT_PRIM
,
1499 stp
->Base
.info
.gs
.output_primitive
);
1500 ureg_property(ureg
, TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES
,
1501 stp
->Base
.info
.gs
.vertices_out
);
1502 ureg_property(ureg
, TGSI_PROPERTY_GS_INVOCATIONS
,
1503 stp
->Base
.info
.gs
.invocations
);
1510 ubyte inputSlotToAttr
[VARYING_SLOT_TESS_MAX
];
1511 ubyte inputMapping
[VARYING_SLOT_TESS_MAX
];
1512 ubyte outputMapping
[VARYING_SLOT_TESS_MAX
];
1515 ubyte input_semantic_name
[PIPE_MAX_SHADER_INPUTS
];
1516 ubyte input_semantic_index
[PIPE_MAX_SHADER_INPUTS
];
1517 uint num_inputs
= 0;
1519 ubyte output_semantic_name
[PIPE_MAX_SHADER_OUTPUTS
];
1520 ubyte output_semantic_index
[PIPE_MAX_SHADER_OUTPUTS
];
1521 uint num_outputs
= 0;
1525 memset(inputSlotToAttr
, 0, sizeof(inputSlotToAttr
));
1526 memset(inputMapping
, 0, sizeof(inputMapping
));
1527 memset(outputMapping
, 0, sizeof(outputMapping
));
1528 memset(&stp
->state
, 0, sizeof(stp
->state
));
1530 if (prog
->info
.clip_distance_array_size
)
1531 ureg_property(ureg
, TGSI_PROPERTY_NUM_CLIPDIST_ENABLED
,
1532 prog
->info
.clip_distance_array_size
);
1533 if (prog
->info
.cull_distance_array_size
)
1534 ureg_property(ureg
, TGSI_PROPERTY_NUM_CULLDIST_ENABLED
,
1535 prog
->info
.cull_distance_array_size
);
1538 * Convert Mesa program inputs to TGSI input register semantics.
1540 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
1541 if ((prog
->info
.inputs_read
& BITFIELD64_BIT(attr
)) == 0)
1544 unsigned slot
= num_inputs
++;
1546 inputMapping
[attr
] = slot
;
1547 inputSlotToAttr
[slot
] = attr
;
1549 unsigned semantic_name
, semantic_index
;
1550 tgsi_get_gl_varying_semantic(attr
, st
->needs_texcoord_semantic
,
1551 &semantic_name
, &semantic_index
);
1552 input_semantic_name
[slot
] = semantic_name
;
1553 input_semantic_index
[slot
] = semantic_index
;
1556 /* Also add patch inputs. */
1557 for (attr
= 0; attr
< 32; attr
++) {
1558 if (prog
->info
.patch_inputs_read
& (1u << attr
)) {
1559 GLuint slot
= num_inputs
++;
1560 GLuint patch_attr
= VARYING_SLOT_PATCH0
+ attr
;
1562 inputMapping
[patch_attr
] = slot
;
1563 inputSlotToAttr
[slot
] = patch_attr
;
1564 input_semantic_name
[slot
] = TGSI_SEMANTIC_PATCH
;
1565 input_semantic_index
[slot
] = attr
;
1569 /* initialize output semantics to defaults */
1570 for (i
= 0; i
< PIPE_MAX_SHADER_OUTPUTS
; i
++) {
1571 output_semantic_name
[i
] = TGSI_SEMANTIC_GENERIC
;
1572 output_semantic_index
[i
] = 0;
1576 * Determine number of outputs, the (default) output register
1577 * mapping and the semantic information for each output.
1579 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
1580 if (prog
->info
.outputs_written
& BITFIELD64_BIT(attr
)) {
1581 GLuint slot
= num_outputs
++;
1583 outputMapping
[attr
] = slot
;
1585 unsigned semantic_name
, semantic_index
;
1586 tgsi_get_gl_varying_semantic(attr
, st
->needs_texcoord_semantic
,
1587 &semantic_name
, &semantic_index
);
1588 output_semantic_name
[slot
] = semantic_name
;
1589 output_semantic_index
[slot
] = semantic_index
;
1593 /* Also add patch outputs. */
1594 for (attr
= 0; attr
< 32; attr
++) {
1595 if (prog
->info
.patch_outputs_written
& (1u << attr
)) {
1596 GLuint slot
= num_outputs
++;
1597 GLuint patch_attr
= VARYING_SLOT_PATCH0
+ attr
;
1599 outputMapping
[patch_attr
] = slot
;
1600 output_semantic_name
[slot
] = TGSI_SEMANTIC_PATCH
;
1601 output_semantic_index
[slot
] = attr
;
1605 st_translate_program(st
->ctx
,
1614 input_semantic_name
,
1615 input_semantic_index
,
1620 output_semantic_name
,
1621 output_semantic_index
);
1623 stp
->state
.tokens
= ureg_get_tokens(ureg
, NULL
);
1627 st_translate_stream_output_info(prog
);
1629 st_store_ir_in_disk_cache(st
, prog
, false);
1631 if (ST_DEBUG
& DEBUG_PRINT_IR
&& ST_DEBUG
& DEBUG_MESA
)
1632 _mesa_print_program(prog
);
1634 free_glsl_to_tgsi_visitor(stp
->glsl_to_tgsi
);
1635 stp
->glsl_to_tgsi
= NULL
;
1641 * Get/create a basic program variant.
1644 st_get_common_variant(struct st_context
*st
,
1645 struct st_program
*prog
,
1646 const struct st_common_variant_key
*key
)
1648 struct pipe_context
*pipe
= st
->pipe
;
1649 struct st_variant
*v
;
1650 struct pipe_shader_state state
= {0};
1652 /* Search for existing variant */
1653 for (v
= prog
->variants
; v
; v
= v
->next
) {
1654 if (memcmp(&st_common_variant(v
)->key
, key
, sizeof(*key
)) == 0)
1660 v
= (struct st_variant
*)CALLOC_STRUCT(st_common_variant
);
1662 if (prog
->state
.type
== PIPE_SHADER_IR_NIR
) {
1663 bool finalize
= false;
1665 state
.type
= PIPE_SHADER_IR_NIR
;
1666 state
.ir
.nir
= nir_shader_clone(NULL
, prog
->Base
.nir
);
1668 if (key
->clamp_color
) {
1669 NIR_PASS_V(state
.ir
.nir
, nir_lower_clamp_color_outputs
);
1673 state
.stream_output
= prog
->state
.stream_output
;
1675 if (finalize
|| !st
->allow_st_finalize_nir_twice
) {
1676 st_finalize_nir(st
, &prog
->Base
, prog
->shader_program
,
1677 state
.ir
.nir
, true);
1680 if (ST_DEBUG
& DEBUG_PRINT_IR
)
1681 nir_print_shader(state
.ir
.nir
, stderr
);
1683 if (key
->lower_depth_clamp
) {
1684 struct gl_program_parameter_list
*params
= prog
->Base
.Parameters
;
1686 unsigned depth_range_const
=
1687 _mesa_add_state_reference(params
, depth_range_state
);
1689 const struct tgsi_token
*tokens
;
1691 st_tgsi_lower_depth_clamp(prog
->state
.tokens
,
1693 key
->clip_negative_one_to_one
);
1695 if (tokens
!= prog
->state
.tokens
)
1696 tgsi_free_tokens(prog
->state
.tokens
);
1698 prog
->state
.tokens
= tokens
;
1700 state
= prog
->state
;
1702 if (ST_DEBUG
& DEBUG_PRINT_IR
)
1703 tgsi_dump(state
.tokens
, 0);
1705 /* fill in new variant */
1706 switch (prog
->Base
.info
.stage
) {
1707 case MESA_SHADER_TESS_CTRL
:
1708 v
->driver_shader
= pipe
->create_tcs_state(pipe
, &state
);
1710 case MESA_SHADER_TESS_EVAL
:
1711 v
->driver_shader
= pipe
->create_tes_state(pipe
, &state
);
1713 case MESA_SHADER_GEOMETRY
:
1714 v
->driver_shader
= pipe
->create_gs_state(pipe
, &state
);
1716 case MESA_SHADER_COMPUTE
: {
1717 struct pipe_compute_state cs
= {0};
1718 cs
.ir_type
= state
.type
;
1719 cs
.req_local_mem
= prog
->Base
.info
.cs
.shared_size
;
1721 if (state
.type
== PIPE_SHADER_IR_NIR
)
1722 cs
.prog
= state
.ir
.nir
;
1724 cs
.prog
= state
.tokens
;
1726 v
->driver_shader
= pipe
->create_compute_state(pipe
, &cs
);
1730 assert(!"unhandled shader type");
1735 st_common_variant(v
)->key
= *key
;
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 struct st_program
*p
= st_program(target
);
1759 struct st_variant
*v
, **prevPtr
= &p
->variants
;
1761 for (v
= p
->variants
; v
; ) {
1762 struct st_variant
*next
= v
->next
;
1764 /* unlink from list */
1766 /* destroy this variant */
1767 delete_variant(st
, v
, target
->Target
);
1778 * Callback for _mesa_HashWalk. Free all the shader's program variants
1779 * which match the given context.
1782 destroy_shader_program_variants_cb(GLuint key
, void *data
, void *userData
)
1784 struct st_context
*st
= (struct st_context
*) userData
;
1785 struct gl_shader
*shader
= (struct gl_shader
*) data
;
1787 switch (shader
->Type
) {
1788 case GL_SHADER_PROGRAM_MESA
:
1790 struct gl_shader_program
*shProg
= (struct gl_shader_program
*) data
;
1793 for (i
= 0; i
< ARRAY_SIZE(shProg
->_LinkedShaders
); i
++) {
1794 if (shProg
->_LinkedShaders
[i
])
1795 destroy_program_variants(st
, shProg
->_LinkedShaders
[i
]->Program
);
1799 case GL_VERTEX_SHADER
:
1800 case GL_FRAGMENT_SHADER
:
1801 case GL_GEOMETRY_SHADER
:
1802 case GL_TESS_CONTROL_SHADER
:
1803 case GL_TESS_EVALUATION_SHADER
:
1804 case GL_COMPUTE_SHADER
:
1813 * Callback for _mesa_HashWalk. Free all the program variants which match
1814 * the given context.
1817 destroy_program_variants_cb(GLuint key
, void *data
, void *userData
)
1819 struct st_context
*st
= (struct st_context
*) userData
;
1820 struct gl_program
*program
= (struct gl_program
*) data
;
1821 destroy_program_variants(st
, program
);
1826 * Walk over all shaders and programs to delete any variants which
1827 * belong to the given context.
1828 * This is called during context tear-down.
1831 st_destroy_program_variants(struct st_context
*st
)
1833 /* If shaders can be shared with other contexts, the last context will
1834 * call DeleteProgram on all shaders, releasing everything.
1836 if (st
->has_shareable_shaders
)
1839 /* ARB vert/frag program */
1840 _mesa_HashWalk(st
->ctx
->Shared
->Programs
,
1841 destroy_program_variants_cb
, st
);
1843 /* GLSL vert/frag/geom shaders */
1844 _mesa_HashWalk(st
->ctx
->Shared
->ShaderObjects
,
1845 destroy_shader_program_variants_cb
, st
);
1850 * Compile one shader variant.
1853 st_precompile_shader_variant(struct st_context
*st
,
1854 struct gl_program
*prog
)
1856 switch (prog
->Target
) {
1857 case GL_VERTEX_PROGRAM_ARB
: {
1858 struct st_program
*p
= (struct st_program
*)prog
;
1859 struct st_common_variant_key key
;
1861 memset(&key
, 0, sizeof(key
));
1863 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
1864 st_get_vp_variant(st
, p
, &key
);
1868 case GL_FRAGMENT_PROGRAM_ARB
: {
1869 struct st_program
*p
= (struct st_program
*)prog
;
1870 struct st_fp_variant_key key
;
1872 memset(&key
, 0, sizeof(key
));
1874 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
1875 st_get_fp_variant(st
, p
, &key
);
1879 case GL_TESS_CONTROL_PROGRAM_NV
:
1880 case GL_TESS_EVALUATION_PROGRAM_NV
:
1881 case GL_GEOMETRY_PROGRAM_NV
:
1882 case GL_COMPUTE_PROGRAM_NV
: {
1883 struct st_program
*p
= st_program(prog
);
1884 struct st_common_variant_key key
;
1886 memset(&key
, 0, sizeof(key
));
1888 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
1889 st_get_common_variant(st
, p
, &key
);
1899 st_finalize_program(struct st_context
*st
, struct gl_program
*prog
)
1901 if (st
->current_program
[prog
->info
.stage
] == prog
) {
1902 if (prog
->info
.stage
== MESA_SHADER_VERTEX
)
1903 st
->dirty
|= ST_NEW_VERTEX_PROGRAM(st
, (struct st_program
*)prog
);
1905 st
->dirty
|= ((struct st_program
*)prog
)->affected_states
;
1909 nir_sweep(prog
->nir
);
1911 /* Create Gallium shaders now instead of on demand. */
1912 if (ST_DEBUG
& DEBUG_PRECOMPILE
||
1913 st
->shader_has_one_variant
[prog
->info
.stage
])
1914 st_precompile_shader_variant(st
, prog
);