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"
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 "compiler/nir/nir_serialize.h"
45 #include "draw/draw_context.h"
47 #include "pipe/p_context.h"
48 #include "pipe/p_defines.h"
49 #include "pipe/p_shader_tokens.h"
50 #include "draw/draw_context.h"
51 #include "tgsi/tgsi_dump.h"
52 #include "tgsi/tgsi_emulate.h"
53 #include "tgsi/tgsi_parse.h"
54 #include "tgsi/tgsi_ureg.h"
56 #include "util/u_memory.h"
59 #include "st_cb_bitmap.h"
60 #include "st_cb_drawpixels.h"
61 #include "st_context.h"
62 #include "st_tgsi_lower_depth_clamp.h"
63 #include "st_tgsi_lower_yuv.h"
64 #include "st_program.h"
65 #include "st_mesa_to_tgsi.h"
66 #include "st_atifs_to_tgsi.h"
68 #include "st_shader_cache.h"
70 #include "cso_cache/cso_context.h"
74 destroy_program_variants(struct st_context
*st
, struct gl_program
*target
);
77 set_affected_state_flags(uint64_t *states
,
78 struct gl_program
*prog
,
79 uint64_t new_constants
,
80 uint64_t new_sampler_views
,
81 uint64_t new_samplers
,
87 if (prog
->Parameters
->NumParameters
)
88 *states
|= new_constants
;
90 if (prog
->info
.num_textures
)
91 *states
|= new_sampler_views
| new_samplers
;
93 if (prog
->info
.num_images
)
94 *states
|= new_images
;
96 if (prog
->info
.num_ubos
)
99 if (prog
->info
.num_ssbos
)
100 *states
|= new_ssbos
;
102 if (prog
->info
.num_abos
)
103 *states
|= new_atomics
;
107 * This determines which states will be updated when the shader is bound.
110 st_set_prog_affected_state_flags(struct gl_program
*prog
)
114 switch (prog
->info
.stage
) {
115 case MESA_SHADER_VERTEX
:
116 states
= &((struct st_program
*)prog
)->affected_states
;
118 *states
= ST_NEW_VS_STATE
|
120 ST_NEW_VERTEX_ARRAYS
;
122 set_affected_state_flags(states
, prog
,
124 ST_NEW_VS_SAMPLER_VIEWS
,
132 case MESA_SHADER_TESS_CTRL
:
133 states
= &(st_program(prog
))->affected_states
;
135 *states
= ST_NEW_TCS_STATE
;
137 set_affected_state_flags(states
, prog
,
138 ST_NEW_TCS_CONSTANTS
,
139 ST_NEW_TCS_SAMPLER_VIEWS
,
147 case MESA_SHADER_TESS_EVAL
:
148 states
= &(st_program(prog
))->affected_states
;
150 *states
= ST_NEW_TES_STATE
|
153 set_affected_state_flags(states
, prog
,
154 ST_NEW_TES_CONSTANTS
,
155 ST_NEW_TES_SAMPLER_VIEWS
,
163 case MESA_SHADER_GEOMETRY
:
164 states
= &(st_program(prog
))->affected_states
;
166 *states
= ST_NEW_GS_STATE
|
169 set_affected_state_flags(states
, prog
,
171 ST_NEW_GS_SAMPLER_VIEWS
,
179 case MESA_SHADER_FRAGMENT
:
180 states
= &((struct st_program
*)prog
)->affected_states
;
182 /* gl_FragCoord and glDrawPixels always use constants. */
183 *states
= ST_NEW_FS_STATE
|
184 ST_NEW_SAMPLE_SHADING
|
187 set_affected_state_flags(states
, prog
,
189 ST_NEW_FS_SAMPLER_VIEWS
,
197 case MESA_SHADER_COMPUTE
:
198 states
= &((struct st_program
*)prog
)->affected_states
;
200 *states
= ST_NEW_CS_STATE
;
202 set_affected_state_flags(states
, prog
,
204 ST_NEW_CS_SAMPLER_VIEWS
,
213 unreachable("unhandled shader stage");
219 * Delete a shader variant. Note the caller must unlink the variant from
223 delete_variant(struct st_context
*st
, struct st_variant
*v
, GLenum target
)
225 if (v
->driver_shader
) {
226 if (target
== GL_VERTEX_PROGRAM_ARB
&&
227 ((struct st_common_variant
*)v
)->key
.is_draw_shader
) {
229 draw_delete_vertex_shader(st
->draw
, v
->driver_shader
);
230 } else if (st
->has_shareable_shaders
|| v
->st
== st
) {
231 /* The shader's context matches the calling context, or we
235 case GL_VERTEX_PROGRAM_ARB
:
236 st
->pipe
->delete_vs_state(st
->pipe
, v
->driver_shader
);
238 case GL_TESS_CONTROL_PROGRAM_NV
:
239 st
->pipe
->delete_tcs_state(st
->pipe
, v
->driver_shader
);
241 case GL_TESS_EVALUATION_PROGRAM_NV
:
242 st
->pipe
->delete_tes_state(st
->pipe
, v
->driver_shader
);
244 case GL_GEOMETRY_PROGRAM_NV
:
245 st
->pipe
->delete_gs_state(st
->pipe
, v
->driver_shader
);
247 case GL_FRAGMENT_PROGRAM_ARB
:
248 st
->pipe
->delete_fs_state(st
->pipe
, v
->driver_shader
);
250 case GL_COMPUTE_PROGRAM_NV
:
251 st
->pipe
->delete_compute_state(st
->pipe
, v
->driver_shader
);
254 unreachable("bad shader type in delete_basic_variant");
257 /* We can't delete a shader with a context different from the one
258 * that created it. Add it to the creating context's zombie list.
260 enum pipe_shader_type type
=
261 pipe_shader_type_from_mesa(_mesa_program_enum_to_shader_stage(target
));
263 st_save_zombie_shader(v
->st
, type
, v
->driver_shader
);
271 st_unbind_program(struct st_context
*st
, struct st_program
*p
)
273 /* Unbind the shader in cso_context and re-bind in st/mesa. */
274 switch (p
->Base
.info
.stage
) {
275 case MESA_SHADER_VERTEX
:
276 cso_set_vertex_shader_handle(st
->cso_context
, NULL
);
277 st
->dirty
|= ST_NEW_VS_STATE
;
279 case MESA_SHADER_TESS_CTRL
:
280 cso_set_tessctrl_shader_handle(st
->cso_context
, NULL
);
281 st
->dirty
|= ST_NEW_TCS_STATE
;
283 case MESA_SHADER_TESS_EVAL
:
284 cso_set_tesseval_shader_handle(st
->cso_context
, NULL
);
285 st
->dirty
|= ST_NEW_TES_STATE
;
287 case MESA_SHADER_GEOMETRY
:
288 cso_set_geometry_shader_handle(st
->cso_context
, NULL
);
289 st
->dirty
|= ST_NEW_GS_STATE
;
291 case MESA_SHADER_FRAGMENT
:
292 cso_set_fragment_shader_handle(st
->cso_context
, NULL
);
293 st
->dirty
|= ST_NEW_FS_STATE
;
295 case MESA_SHADER_COMPUTE
:
296 cso_set_compute_shader_handle(st
->cso_context
, NULL
);
297 st
->dirty
|= ST_NEW_CS_STATE
;
300 unreachable("invalid shader type");
305 * Free all basic program variants.
308 st_release_variants(struct st_context
*st
, struct st_program
*p
)
310 struct st_variant
*v
;
312 /* If we are releasing shaders, re-bind them, because we don't
313 * know which shaders are bound in the driver.
316 st_unbind_program(st
, p
);
318 for (v
= p
->variants
; v
; ) {
319 struct st_variant
*next
= v
->next
;
320 delete_variant(st
, v
, p
->Base
.Target
);
326 if (p
->state
.tokens
) {
327 ureg_free_tokens(p
->state
.tokens
);
328 p
->state
.tokens
= NULL
;
331 /* Note: Any setup of ->ir.nir that has had pipe->create_*_state called on
332 * it has resulted in the driver taking ownership of the NIR. Those
333 * callers should be NULLing out the nir field in any pipe_shader_state
334 * that might have this called in order to indicate that.
336 * GLSL IR and ARB programs will have set gl_program->nir to the same
337 * shader as ir->ir.nir, so it will be freed by _mesa_delete_program().
342 * Free all basic program variants and unref program.
345 st_release_program(struct st_context
*st
, struct st_program
**p
)
350 destroy_program_variants(st
, &((*p
)->Base
));
351 st_reference_prog(st
, p
, NULL
);
355 st_finalize_nir_before_variants(struct nir_shader
*nir
)
357 NIR_PASS_V(nir
, nir_opt_access
);
359 NIR_PASS_V(nir
, nir_split_var_copies
);
360 NIR_PASS_V(nir
, nir_lower_var_copies
);
361 if (nir
->options
->lower_all_io_to_temps
||
362 nir
->options
->lower_all_io_to_elements
||
363 nir
->info
.stage
== MESA_SHADER_VERTEX
||
364 nir
->info
.stage
== MESA_SHADER_GEOMETRY
) {
365 NIR_PASS_V(nir
, nir_lower_io_arrays_to_elements_no_indirects
, false);
366 } else if (nir
->info
.stage
== MESA_SHADER_FRAGMENT
) {
367 NIR_PASS_V(nir
, nir_lower_io_arrays_to_elements_no_indirects
, true);
370 st_nir_assign_vs_in_locations(nir
);
374 * Translate ARB (asm) program to NIR
377 st_translate_prog_to_nir(struct st_context
*st
, struct gl_program
*prog
,
378 gl_shader_stage stage
)
380 struct pipe_screen
*screen
= st
->pipe
->screen
;
381 const struct gl_shader_compiler_options
*options
=
382 &st
->ctx
->Const
.ShaderCompilerOptions
[stage
];
384 /* Translate to NIR */
385 nir_shader
*nir
= prog_to_nir(prog
, options
->NirOptions
);
386 NIR_PASS_V(nir
, nir_lower_regs_to_ssa
); /* turn registers into SSA */
387 nir_validate_shader(nir
, "after st/ptn lower_regs_to_ssa");
389 NIR_PASS_V(nir
, st_nir_lower_wpos_ytransform
, prog
, screen
);
390 NIR_PASS_V(nir
, nir_lower_system_values
);
391 NIR_PASS_V(nir
, nir_lower_compute_system_values
, NULL
);
394 NIR_PASS_V(nir
, nir_opt_constant_folding
);
396 st_finalize_nir_before_variants(nir
);
398 if (st
->allow_st_finalize_nir_twice
)
399 st_finalize_nir(st
, prog
, NULL
, nir
, true);
401 nir_validate_shader(nir
, "after st/glsl finalize_nir");
407 st_prepare_vertex_program(struct st_program
*stp
)
409 struct st_vertex_program
*stvp
= (struct st_vertex_program
*)stp
;
411 stvp
->num_inputs
= 0;
412 memset(stvp
->input_to_index
, ~0, sizeof(stvp
->input_to_index
));
413 memset(stvp
->result_to_output
, ~0, sizeof(stvp
->result_to_output
));
415 /* Determine number of inputs, the mappings between VERT_ATTRIB_x
416 * and TGSI generic input indexes, plus input attrib semantic info.
418 for (unsigned attr
= 0; attr
< VERT_ATTRIB_MAX
; attr
++) {
419 if ((stp
->Base
.info
.inputs_read
& BITFIELD64_BIT(attr
)) != 0) {
420 stvp
->input_to_index
[attr
] = stvp
->num_inputs
;
421 stvp
->index_to_input
[stvp
->num_inputs
] = attr
;
424 if ((stp
->Base
.DualSlotInputs
& BITFIELD64_BIT(attr
)) != 0) {
425 /* add placeholder for second part of a double attribute */
426 stvp
->index_to_input
[stvp
->num_inputs
] = ST_DOUBLE_ATTRIB_PLACEHOLDER
;
431 /* pre-setup potentially unused edgeflag input */
432 stvp
->input_to_index
[VERT_ATTRIB_EDGEFLAG
] = stvp
->num_inputs
;
433 stvp
->index_to_input
[stvp
->num_inputs
] = VERT_ATTRIB_EDGEFLAG
;
435 /* Compute mapping of vertex program outputs to slots. */
436 unsigned num_outputs
= 0;
437 for (unsigned attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
438 if (stp
->Base
.info
.outputs_written
& BITFIELD64_BIT(attr
))
439 stvp
->result_to_output
[attr
] = num_outputs
++;
441 /* pre-setup potentially unused edgeflag output */
442 stvp
->result_to_output
[VARYING_SLOT_EDGE
] = num_outputs
;
446 st_translate_stream_output_info(struct gl_program
*prog
)
448 struct gl_transform_feedback_info
*info
= prog
->sh
.LinkedTransformFeedback
;
452 /* Determine the (default) output register mapping for each output. */
453 unsigned num_outputs
= 0;
454 ubyte output_mapping
[VARYING_SLOT_TESS_MAX
];
455 memset(output_mapping
, 0, sizeof(output_mapping
));
457 for (unsigned attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
458 if (prog
->info
.outputs_written
& BITFIELD64_BIT(attr
))
459 output_mapping
[attr
] = num_outputs
++;
462 /* Translate stream output info. */
463 struct pipe_stream_output_info
*so_info
=
464 &((struct st_program
*)prog
)->state
.stream_output
;
466 for (unsigned i
= 0; i
< info
->NumOutputs
; i
++) {
467 so_info
->output
[i
].register_index
=
468 output_mapping
[info
->Outputs
[i
].OutputRegister
];
469 so_info
->output
[i
].start_component
= info
->Outputs
[i
].ComponentOffset
;
470 so_info
->output
[i
].num_components
= info
->Outputs
[i
].NumComponents
;
471 so_info
->output
[i
].output_buffer
= info
->Outputs
[i
].OutputBuffer
;
472 so_info
->output
[i
].dst_offset
= info
->Outputs
[i
].DstOffset
;
473 so_info
->output
[i
].stream
= info
->Outputs
[i
].StreamId
;
476 for (unsigned i
= 0; i
< PIPE_MAX_SO_BUFFERS
; i
++) {
477 so_info
->stride
[i
] = info
->Buffers
[i
].Stride
;
479 so_info
->num_outputs
= info
->NumOutputs
;
483 * Translate a vertex program.
486 st_translate_vertex_program(struct st_context
*st
,
487 struct st_program
*stp
)
489 struct ureg_program
*ureg
;
490 enum pipe_error error
;
491 unsigned num_outputs
= 0;
493 ubyte output_semantic_name
[VARYING_SLOT_MAX
] = {0};
494 ubyte output_semantic_index
[VARYING_SLOT_MAX
] = {0};
496 if (stp
->Base
.arb
.IsPositionInvariant
)
497 _mesa_insert_mvp_code(st
->ctx
, &stp
->Base
);
500 if (!stp
->glsl_to_tgsi
) {
501 _mesa_remove_output_reads(&stp
->Base
, PROGRAM_OUTPUT
);
503 /* This determines which states will be updated when the assembly
506 stp
->affected_states
= ST_NEW_VS_STATE
|
508 ST_NEW_VERTEX_ARRAYS
;
510 if (stp
->Base
.Parameters
->NumParameters
)
511 stp
->affected_states
|= ST_NEW_VS_CONSTANTS
;
513 /* Translate to NIR if preferred. */
514 if (PIPE_SHADER_IR_NIR
==
515 st
->pipe
->screen
->get_shader_param(st
->pipe
->screen
,
517 PIPE_SHADER_CAP_PREFERRED_IR
)) {
518 assert(!stp
->glsl_to_tgsi
);
521 ralloc_free(stp
->Base
.nir
);
523 if (stp
->serialized_nir
) {
524 free(stp
->serialized_nir
);
525 stp
->serialized_nir
= NULL
;
528 stp
->state
.type
= PIPE_SHADER_IR_NIR
;
529 stp
->Base
.nir
= st_translate_prog_to_nir(st
, &stp
->Base
,
532 /* We must update stp->Base.info after translation and before
533 * st_prepare_vertex_program is called, because inputs_read
534 * may become outdated after NIR optimization passes.
536 * For ffvp/ARB_vp inputs_read is populated based
537 * on declared attributes without taking their usage into
538 * consideration. When creating shader variants we expect
539 * that their inputs_read would match the base ones for
540 * input mapping to work properly.
542 nir_shader_gather_info(stp
->Base
.nir
,
543 nir_shader_get_entrypoint(stp
->Base
.nir
));
544 st_nir_assign_vs_in_locations(stp
->Base
.nir
);
545 stp
->Base
.info
= stp
->Base
.nir
->info
;
547 /* For st_draw_feedback, we need to generate TGSI too if draw doesn't
550 /* TODO: Draw can't handle lowered IO. */
551 if (draw_has_llvm() && !stp
->Base
.info
.io_lowered
) {
552 st_prepare_vertex_program(stp
);
558 st_prepare_vertex_program(stp
);
560 /* Get semantic names and indices. */
561 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
562 if (stp
->Base
.info
.outputs_written
& BITFIELD64_BIT(attr
)) {
563 unsigned slot
= num_outputs
++;
564 unsigned semantic_name
, semantic_index
;
565 tgsi_get_gl_varying_semantic(attr
, st
->needs_texcoord_semantic
,
566 &semantic_name
, &semantic_index
);
567 output_semantic_name
[slot
] = semantic_name
;
568 output_semantic_index
[slot
] = semantic_index
;
571 /* pre-setup potentially unused edgeflag output */
572 output_semantic_name
[num_outputs
] = TGSI_SEMANTIC_EDGEFLAG
;
573 output_semantic_index
[num_outputs
] = 0;
575 ureg
= ureg_create_with_screen(PIPE_SHADER_VERTEX
, st
->pipe
->screen
);
579 ureg_setup_shader_info(ureg
, &stp
->Base
.info
);
581 if (ST_DEBUG
& DEBUG_MESA
) {
582 _mesa_print_program(&stp
->Base
);
583 _mesa_print_program_parameters(st
->ctx
, &stp
->Base
);
587 struct st_vertex_program
*stvp
= (struct st_vertex_program
*)stp
;
589 if (stp
->glsl_to_tgsi
) {
590 error
= st_translate_program(st
->ctx
,
597 stvp
->input_to_index
,
598 NULL
, /* inputSlotToAttr */
599 NULL
, /* input semantic name */
600 NULL
, /* input semantic index */
601 NULL
, /* interp mode */
604 stvp
->result_to_output
,
605 output_semantic_name
,
606 output_semantic_index
);
608 st_translate_stream_output_info(&stp
->Base
);
610 free_glsl_to_tgsi_visitor(stp
->glsl_to_tgsi
);
612 error
= st_translate_mesa_program(st
->ctx
,
618 stvp
->input_to_index
,
619 NULL
, /* input semantic name */
620 NULL
, /* input semantic index */
624 stvp
->result_to_output
,
625 output_semantic_name
,
626 output_semantic_index
);
629 debug_printf("%s: failed to translate Mesa program:\n", __func__
);
630 _mesa_print_program(&stp
->Base
);
635 stp
->state
.tokens
= ureg_get_tokens(ureg
, NULL
);
638 if (stp
->glsl_to_tgsi
) {
639 stp
->glsl_to_tgsi
= NULL
;
640 st_store_ir_in_disk_cache(st
, &stp
->Base
, false);
643 return stp
->state
.tokens
!= NULL
;
646 static struct nir_shader
*
647 get_nir_shader(struct st_context
*st
, struct st_program
*stp
)
650 nir_shader
*nir
= stp
->Base
.nir
;
652 /* The first shader variant takes ownership of NIR, so that there is
653 * no cloning. Additional shader variants are always generated from
654 * serialized NIR to save memory.
656 stp
->Base
.nir
= NULL
;
657 assert(stp
->serialized_nir
&& stp
->serialized_nir_size
);
661 struct blob_reader blob_reader
;
662 const struct nir_shader_compiler_options
*options
=
663 st
->ctx
->Const
.ShaderCompilerOptions
[stp
->Base
.info
.stage
].NirOptions
;
665 blob_reader_init(&blob_reader
, stp
->serialized_nir
, stp
->serialized_nir_size
);
666 return nir_deserialize(NULL
, options
, &blob_reader
);
670 lower_ucp(struct st_context
*st
,
671 struct nir_shader
*nir
,
672 unsigned ucp_enables
,
673 struct gl_program_parameter_list
*params
)
675 if (nir
->info
.outputs_written
& VARYING_BIT_CLIP_DIST0
)
676 NIR_PASS_V(nir
, nir_lower_clip_disable
, ucp_enables
);
678 struct pipe_screen
*screen
= st
->pipe
->screen
;
679 bool can_compact
= screen
->get_param(screen
,
680 PIPE_CAP_NIR_COMPACT_ARRAYS
);
681 bool use_eye
= st
->ctx
->_Shader
->CurrentProgram
[MESA_SHADER_VERTEX
] != NULL
;
683 gl_state_index16 clipplane_state
[MAX_CLIP_PLANES
][STATE_LENGTH
];
684 for (int i
= 0; i
< MAX_CLIP_PLANES
; ++i
) {
686 clipplane_state
[i
][0] = STATE_CLIPPLANE
;
687 clipplane_state
[i
][1] = i
;
689 clipplane_state
[i
][0] = STATE_INTERNAL
;
690 clipplane_state
[i
][1] = STATE_CLIP_INTERNAL
;
691 clipplane_state
[i
][2] = i
;
693 _mesa_add_state_reference(params
, clipplane_state
[i
]);
696 if (nir
->info
.stage
== MESA_SHADER_VERTEX
) {
697 NIR_PASS_V(nir
, nir_lower_clip_vs
, ucp_enables
,
698 true, can_compact
, clipplane_state
);
699 } else if (nir
->info
.stage
== MESA_SHADER_GEOMETRY
) {
700 NIR_PASS_V(nir
, nir_lower_clip_gs
, ucp_enables
,
701 can_compact
, clipplane_state
);
704 NIR_PASS_V(nir
, nir_lower_io_to_temporaries
,
705 nir_shader_get_entrypoint(nir
), true, false);
706 NIR_PASS_V(nir
, nir_lower_global_vars_to_local
);
710 static const gl_state_index16 depth_range_state
[STATE_LENGTH
] =
711 { STATE_DEPTH_RANGE
};
713 static struct st_common_variant
*
714 st_create_vp_variant(struct st_context
*st
,
715 struct st_program
*stvp
,
716 const struct st_common_variant_key
*key
)
718 struct st_common_variant
*vpv
= CALLOC_STRUCT(st_common_variant
);
719 struct pipe_context
*pipe
= st
->pipe
;
720 struct pipe_shader_state state
= {0};
722 static const gl_state_index16 point_size_state
[STATE_LENGTH
] =
723 { STATE_INTERNAL
, STATE_POINT_SIZE_CLAMPED
, 0 };
724 struct gl_program_parameter_list
*params
= stvp
->Base
.Parameters
;
728 state
.stream_output
= stvp
->state
.stream_output
;
730 if (stvp
->state
.type
== PIPE_SHADER_IR_NIR
&&
731 (!key
->is_draw_shader
||
732 /* TODO: Draw can't handle lowered IO. */
733 (draw_has_llvm() && !stvp
->Base
.info
.io_lowered
))) {
734 bool finalize
= false;
736 state
.type
= PIPE_SHADER_IR_NIR
;
737 state
.ir
.nir
= get_nir_shader(st
, stvp
);
738 if (key
->clamp_color
) {
739 NIR_PASS_V(state
.ir
.nir
, nir_lower_clamp_color_outputs
);
742 if (key
->passthrough_edgeflags
) {
743 NIR_PASS_V(state
.ir
.nir
, nir_lower_passthrough_edgeflags
);
747 if (key
->lower_point_size
) {
748 _mesa_add_state_reference(params
, point_size_state
);
749 NIR_PASS_V(state
.ir
.nir
, nir_lower_point_size_mov
,
754 if (key
->lower_ucp
) {
755 lower_ucp(st
, state
.ir
.nir
, key
->lower_ucp
, params
);
759 if (finalize
|| !st
->allow_st_finalize_nir_twice
) {
760 st_finalize_nir(st
, &stvp
->Base
, stvp
->shader_program
, state
.ir
.nir
,
763 /* Some of the lowering above may have introduced new varyings */
764 nir_shader_gather_info(state
.ir
.nir
,
765 nir_shader_get_entrypoint(state
.ir
.nir
));
768 if (ST_DEBUG
& DEBUG_PRINT_IR
)
769 nir_print_shader(state
.ir
.nir
, stderr
);
771 if (key
->is_draw_shader
)
772 vpv
->base
.driver_shader
= draw_create_vertex_shader(st
->draw
, &state
);
774 vpv
->base
.driver_shader
= pipe
->create_vs_state(pipe
, &state
);
779 state
.type
= PIPE_SHADER_IR_TGSI
;
780 state
.tokens
= tgsi_dup_tokens(stvp
->state
.tokens
);
782 /* Emulate features. */
783 if (key
->clamp_color
|| key
->passthrough_edgeflags
) {
784 const struct tgsi_token
*tokens
;
786 (key
->clamp_color
? TGSI_EMU_CLAMP_COLOR_OUTPUTS
: 0) |
787 (key
->passthrough_edgeflags
? TGSI_EMU_PASSTHROUGH_EDGEFLAG
: 0);
789 tokens
= tgsi_emulate(state
.tokens
, flags
);
792 tgsi_free_tokens(state
.tokens
);
793 state
.tokens
= tokens
;
795 fprintf(stderr
, "mesa: cannot emulate deprecated features\n");
799 if (key
->lower_depth_clamp
) {
800 unsigned depth_range_const
=
801 _mesa_add_state_reference(params
, depth_range_state
);
803 const struct tgsi_token
*tokens
;
804 tokens
= st_tgsi_lower_depth_clamp(state
.tokens
, depth_range_const
,
805 key
->clip_negative_one_to_one
);
806 if (tokens
!= state
.tokens
)
807 tgsi_free_tokens(state
.tokens
);
808 state
.tokens
= tokens
;
811 if (ST_DEBUG
& DEBUG_PRINT_IR
)
812 tgsi_dump(state
.tokens
, 0);
814 if (key
->is_draw_shader
)
815 vpv
->base
.driver_shader
= draw_create_vertex_shader(st
->draw
, &state
);
817 vpv
->base
.driver_shader
= pipe
->create_vs_state(pipe
, &state
);
820 tgsi_free_tokens(state
.tokens
);
828 * Find/create a vertex program variant.
830 struct st_common_variant
*
831 st_get_vp_variant(struct st_context
*st
,
832 struct st_program
*stp
,
833 const struct st_common_variant_key
*key
)
835 struct st_vertex_program
*stvp
= (struct st_vertex_program
*)stp
;
836 struct st_common_variant
*vpv
;
838 /* Search for existing variant */
839 for (vpv
= st_common_variant(stp
->variants
); vpv
;
840 vpv
= st_common_variant(vpv
->base
.next
)) {
841 if (memcmp(&vpv
->key
, key
, sizeof(*key
)) == 0) {
848 vpv
= st_create_vp_variant(st
, stp
, key
);
850 vpv
->base
.st
= key
->st
;
852 unsigned num_inputs
= stvp
->num_inputs
+ key
->passthrough_edgeflags
;
853 for (unsigned index
= 0; index
< num_inputs
; ++index
) {
854 unsigned attr
= stvp
->index_to_input
[index
];
855 if (attr
== ST_DOUBLE_ATTRIB_PLACEHOLDER
)
857 vpv
->vert_attrib_mask
|= 1u << attr
;
860 /* insert into list */
861 vpv
->base
.next
= stp
->variants
;
862 stp
->variants
= &vpv
->base
;
871 * Translate a Mesa fragment shader into a TGSI shader.
874 st_translate_fragment_program(struct st_context
*st
,
875 struct st_program
*stfp
)
877 /* Non-GLSL programs: */
878 if (!stfp
->glsl_to_tgsi
) {
879 _mesa_remove_output_reads(&stfp
->Base
, PROGRAM_OUTPUT
);
880 if (st
->ctx
->Const
.GLSLFragCoordIsSysVal
)
881 _mesa_program_fragment_position_to_sysval(&stfp
->Base
);
883 /* This determines which states will be updated when the assembly
886 * fragment.position and glDrawPixels always use constants.
888 stfp
->affected_states
= ST_NEW_FS_STATE
|
889 ST_NEW_SAMPLE_SHADING
|
893 /* Just set them for ATI_fs unconditionally. */
894 stfp
->affected_states
|= ST_NEW_FS_SAMPLER_VIEWS
|
898 if (stfp
->Base
.SamplersUsed
)
899 stfp
->affected_states
|= ST_NEW_FS_SAMPLER_VIEWS
|
903 /* Translate to NIR. */
905 PIPE_SHADER_IR_NIR
==
906 st
->pipe
->screen
->get_shader_param(st
->pipe
->screen
,
907 PIPE_SHADER_FRAGMENT
,
908 PIPE_SHADER_CAP_PREFERRED_IR
)) {
910 st_translate_prog_to_nir(st
, &stfp
->Base
, MESA_SHADER_FRAGMENT
);
913 ralloc_free(stfp
->Base
.nir
);
914 if (stfp
->serialized_nir
) {
915 free(stfp
->serialized_nir
);
916 stfp
->serialized_nir
= NULL
;
918 stfp
->state
.type
= PIPE_SHADER_IR_NIR
;
919 stfp
->Base
.nir
= nir
;
924 ubyte outputMapping
[2 * FRAG_RESULT_MAX
];
925 ubyte inputMapping
[VARYING_SLOT_MAX
];
926 ubyte inputSlotToAttr
[VARYING_SLOT_MAX
];
927 ubyte interpMode
[PIPE_MAX_SHADER_INPUTS
]; /* XXX size? */
929 GLbitfield64 inputsRead
;
930 struct ureg_program
*ureg
;
932 GLboolean write_all
= GL_FALSE
;
934 ubyte input_semantic_name
[PIPE_MAX_SHADER_INPUTS
];
935 ubyte input_semantic_index
[PIPE_MAX_SHADER_INPUTS
];
936 uint fs_num_inputs
= 0;
938 ubyte fs_output_semantic_name
[PIPE_MAX_SHADER_OUTPUTS
];
939 ubyte fs_output_semantic_index
[PIPE_MAX_SHADER_OUTPUTS
];
940 uint fs_num_outputs
= 0;
942 memset(inputSlotToAttr
, ~0, sizeof(inputSlotToAttr
));
945 * Convert Mesa program inputs to TGSI input register semantics.
947 inputsRead
= stfp
->Base
.info
.inputs_read
;
948 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
949 if ((inputsRead
& BITFIELD64_BIT(attr
)) != 0) {
950 const GLuint slot
= fs_num_inputs
++;
952 inputMapping
[attr
] = slot
;
953 inputSlotToAttr
[slot
] = attr
;
956 case VARYING_SLOT_POS
:
957 input_semantic_name
[slot
] = TGSI_SEMANTIC_POSITION
;
958 input_semantic_index
[slot
] = 0;
959 interpMode
[slot
] = TGSI_INTERPOLATE_LINEAR
;
961 case VARYING_SLOT_COL0
:
962 input_semantic_name
[slot
] = TGSI_SEMANTIC_COLOR
;
963 input_semantic_index
[slot
] = 0;
964 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
965 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_COLOR
;
967 case VARYING_SLOT_COL1
:
968 input_semantic_name
[slot
] = TGSI_SEMANTIC_COLOR
;
969 input_semantic_index
[slot
] = 1;
970 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
971 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_COLOR
;
973 case VARYING_SLOT_FOGC
:
974 input_semantic_name
[slot
] = TGSI_SEMANTIC_FOG
;
975 input_semantic_index
[slot
] = 0;
976 interpMode
[slot
] = TGSI_INTERPOLATE_PERSPECTIVE
;
978 case VARYING_SLOT_FACE
:
979 input_semantic_name
[slot
] = TGSI_SEMANTIC_FACE
;
980 input_semantic_index
[slot
] = 0;
981 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
983 case VARYING_SLOT_PRIMITIVE_ID
:
984 input_semantic_name
[slot
] = TGSI_SEMANTIC_PRIMID
;
985 input_semantic_index
[slot
] = 0;
986 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
988 case VARYING_SLOT_LAYER
:
989 input_semantic_name
[slot
] = TGSI_SEMANTIC_LAYER
;
990 input_semantic_index
[slot
] = 0;
991 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
993 case VARYING_SLOT_VIEWPORT
:
994 input_semantic_name
[slot
] = TGSI_SEMANTIC_VIEWPORT_INDEX
;
995 input_semantic_index
[slot
] = 0;
996 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
998 case VARYING_SLOT_CLIP_DIST0
:
999 input_semantic_name
[slot
] = TGSI_SEMANTIC_CLIPDIST
;
1000 input_semantic_index
[slot
] = 0;
1001 interpMode
[slot
] = TGSI_INTERPOLATE_PERSPECTIVE
;
1003 case VARYING_SLOT_CLIP_DIST1
:
1004 input_semantic_name
[slot
] = TGSI_SEMANTIC_CLIPDIST
;
1005 input_semantic_index
[slot
] = 1;
1006 interpMode
[slot
] = TGSI_INTERPOLATE_PERSPECTIVE
;
1008 case VARYING_SLOT_CULL_DIST0
:
1009 case VARYING_SLOT_CULL_DIST1
:
1010 /* these should have been lowered by GLSL */
1013 /* In most cases, there is nothing special about these
1014 * inputs, so adopt a convention to use the generic
1015 * semantic name and the mesa VARYING_SLOT_ number as the
1018 * All that is required is that the vertex shader labels
1019 * its own outputs similarly, and that the vertex shader
1020 * generates at least every output required by the
1021 * fragment shader plus fixed-function hardware (such as
1024 * However, some drivers may need us to identify the PNTC and TEXi
1025 * varyings if, for example, their capability to replace them with
1026 * sprite coordinates is limited.
1028 case VARYING_SLOT_PNTC
:
1029 if (st
->needs_texcoord_semantic
) {
1030 input_semantic_name
[slot
] = TGSI_SEMANTIC_PCOORD
;
1031 input_semantic_index
[slot
] = 0;
1032 interpMode
[slot
] = TGSI_INTERPOLATE_LINEAR
;
1036 case VARYING_SLOT_TEX0
:
1037 case VARYING_SLOT_TEX1
:
1038 case VARYING_SLOT_TEX2
:
1039 case VARYING_SLOT_TEX3
:
1040 case VARYING_SLOT_TEX4
:
1041 case VARYING_SLOT_TEX5
:
1042 case VARYING_SLOT_TEX6
:
1043 case VARYING_SLOT_TEX7
:
1044 if (st
->needs_texcoord_semantic
) {
1045 input_semantic_name
[slot
] = TGSI_SEMANTIC_TEXCOORD
;
1046 input_semantic_index
[slot
] = attr
- VARYING_SLOT_TEX0
;
1047 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
1048 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_PERSPECTIVE
;
1052 case VARYING_SLOT_VAR0
:
1054 /* Semantic indices should be zero-based because drivers may choose
1055 * to assign a fixed slot determined by that index.
1056 * This is useful because ARB_separate_shader_objects uses location
1057 * qualifiers for linkage, and if the semantic index corresponds to
1058 * these locations, linkage passes in the driver become unecessary.
1060 * If needs_texcoord_semantic is true, no semantic indices will be
1061 * consumed for the TEXi varyings, and we can base the locations of
1062 * the user varyings on VAR0. Otherwise, we use TEX0 as base index.
1064 assert(attr
>= VARYING_SLOT_VAR0
|| attr
== VARYING_SLOT_PNTC
||
1065 (attr
>= VARYING_SLOT_TEX0
&& attr
<= VARYING_SLOT_TEX7
));
1066 input_semantic_name
[slot
] = TGSI_SEMANTIC_GENERIC
;
1067 input_semantic_index
[slot
] = st_get_generic_varying_index(st
, attr
);
1068 if (attr
== VARYING_SLOT_PNTC
)
1069 interpMode
[slot
] = TGSI_INTERPOLATE_LINEAR
;
1071 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
1072 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_PERSPECTIVE
;
1078 inputMapping
[attr
] = -1;
1083 * Semantics and mapping for outputs
1085 GLbitfield64 outputsWritten
= stfp
->Base
.info
.outputs_written
;
1087 /* if z is written, emit that first */
1088 if (outputsWritten
& BITFIELD64_BIT(FRAG_RESULT_DEPTH
)) {
1089 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_POSITION
;
1090 fs_output_semantic_index
[fs_num_outputs
] = 0;
1091 outputMapping
[FRAG_RESULT_DEPTH
] = fs_num_outputs
;
1093 outputsWritten
&= ~(1 << FRAG_RESULT_DEPTH
);
1096 if (outputsWritten
& BITFIELD64_BIT(FRAG_RESULT_STENCIL
)) {
1097 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_STENCIL
;
1098 fs_output_semantic_index
[fs_num_outputs
] = 0;
1099 outputMapping
[FRAG_RESULT_STENCIL
] = fs_num_outputs
;
1101 outputsWritten
&= ~(1 << FRAG_RESULT_STENCIL
);
1104 if (outputsWritten
& BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK
)) {
1105 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_SAMPLEMASK
;
1106 fs_output_semantic_index
[fs_num_outputs
] = 0;
1107 outputMapping
[FRAG_RESULT_SAMPLE_MASK
] = fs_num_outputs
;
1109 outputsWritten
&= ~(1 << FRAG_RESULT_SAMPLE_MASK
);
1112 /* handle remaining outputs (color) */
1113 for (attr
= 0; attr
< ARRAY_SIZE(outputMapping
); attr
++) {
1114 const GLbitfield64 written
= attr
< FRAG_RESULT_MAX
? outputsWritten
:
1115 stfp
->Base
.SecondaryOutputsWritten
;
1116 const unsigned loc
= attr
% FRAG_RESULT_MAX
;
1118 if (written
& BITFIELD64_BIT(loc
)) {
1120 case FRAG_RESULT_DEPTH
:
1121 case FRAG_RESULT_STENCIL
:
1122 case FRAG_RESULT_SAMPLE_MASK
:
1126 case FRAG_RESULT_COLOR
:
1127 write_all
= GL_TRUE
; /* fallthrough */
1130 assert(loc
== FRAG_RESULT_COLOR
||
1131 (FRAG_RESULT_DATA0
<= loc
&& loc
< FRAG_RESULT_MAX
));
1133 index
= (loc
== FRAG_RESULT_COLOR
) ? 0 : (loc
- FRAG_RESULT_DATA0
);
1135 if (attr
>= FRAG_RESULT_MAX
) {
1136 /* Secondary color for dual source blending. */
1141 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_COLOR
;
1142 fs_output_semantic_index
[fs_num_outputs
] = index
;
1143 outputMapping
[attr
] = fs_num_outputs
;
1152 ureg
= ureg_create_with_screen(PIPE_SHADER_FRAGMENT
, st
->pipe
->screen
);
1156 ureg_setup_shader_info(ureg
, &stfp
->Base
.info
);
1158 if (ST_DEBUG
& DEBUG_MESA
) {
1159 _mesa_print_program(&stfp
->Base
);
1160 _mesa_print_program_parameters(st
->ctx
, &stfp
->Base
);
1163 if (write_all
== GL_TRUE
)
1164 ureg_property(ureg
, TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS
, 1);
1166 if (stfp
->glsl_to_tgsi
) {
1167 st_translate_program(st
->ctx
,
1168 PIPE_SHADER_FRAGMENT
,
1176 input_semantic_name
,
1177 input_semantic_index
,
1182 fs_output_semantic_name
,
1183 fs_output_semantic_index
);
1185 free_glsl_to_tgsi_visitor(stfp
->glsl_to_tgsi
);
1186 } else if (stfp
->ati_fs
)
1187 st_translate_atifs_program(ureg
,
1193 input_semantic_name
,
1194 input_semantic_index
,
1199 fs_output_semantic_name
,
1200 fs_output_semantic_index
);
1202 st_translate_mesa_program(st
->ctx
,
1203 PIPE_SHADER_FRAGMENT
,
1209 input_semantic_name
,
1210 input_semantic_index
,
1215 fs_output_semantic_name
,
1216 fs_output_semantic_index
);
1218 stfp
->state
.tokens
= ureg_get_tokens(ureg
, NULL
);
1221 if (stfp
->glsl_to_tgsi
) {
1222 stfp
->glsl_to_tgsi
= NULL
;
1223 st_store_ir_in_disk_cache(st
, &stfp
->Base
, false);
1226 return stfp
->state
.tokens
!= NULL
;
1229 static struct st_fp_variant
*
1230 st_create_fp_variant(struct st_context
*st
,
1231 struct st_program
*stfp
,
1232 const struct st_fp_variant_key
*key
)
1234 struct pipe_context
*pipe
= st
->pipe
;
1235 struct st_fp_variant
*variant
= CALLOC_STRUCT(st_fp_variant
);
1236 struct pipe_shader_state state
= {0};
1237 struct gl_program_parameter_list
*params
= stfp
->Base
.Parameters
;
1238 static const gl_state_index16 texcoord_state
[STATE_LENGTH
] =
1239 { STATE_INTERNAL
, STATE_CURRENT_ATTRIB
, VERT_ATTRIB_TEX0
};
1240 static const gl_state_index16 scale_state
[STATE_LENGTH
] =
1241 { STATE_INTERNAL
, STATE_PT_SCALE
};
1242 static const gl_state_index16 bias_state
[STATE_LENGTH
] =
1243 { STATE_INTERNAL
, STATE_PT_BIAS
};
1244 static const gl_state_index16 alpha_ref_state
[STATE_LENGTH
] =
1245 { STATE_INTERNAL
, STATE_ALPHA_REF
};
1250 if (stfp
->state
.type
== PIPE_SHADER_IR_NIR
) {
1251 bool finalize
= false;
1253 state
.type
= PIPE_SHADER_IR_NIR
;
1254 state
.ir
.nir
= get_nir_shader(st
, stfp
);
1256 if (key
->clamp_color
) {
1257 NIR_PASS_V(state
.ir
.nir
, nir_lower_clamp_color_outputs
);
1261 if (key
->lower_flatshade
) {
1262 NIR_PASS_V(state
.ir
.nir
, nir_lower_flatshade
);
1266 if (key
->lower_alpha_func
!= COMPARE_FUNC_NEVER
) {
1267 _mesa_add_state_reference(params
, alpha_ref_state
);
1268 NIR_PASS_V(state
.ir
.nir
, nir_lower_alpha_test
, key
->lower_alpha_func
,
1269 false, alpha_ref_state
);
1273 if (key
->lower_two_sided_color
) {
1274 bool face_sysval
= st
->ctx
->Const
.GLSLFrontFacingIsSysVal
;
1275 NIR_PASS_V(state
.ir
.nir
, nir_lower_two_sided_color
, face_sysval
);
1279 if (key
->persample_shading
) {
1280 nir_shader
*shader
= state
.ir
.nir
;
1281 nir_foreach_shader_in_variable(var
, shader
)
1282 var
->data
.sample
= true;
1286 assert(!(key
->bitmap
&& key
->drawpixels
));
1290 nir_lower_bitmap_options options
= {0};
1292 variant
->bitmap_sampler
= ffs(~stfp
->Base
.SamplersUsed
) - 1;
1293 options
.sampler
= variant
->bitmap_sampler
;
1294 options
.swizzle_xxxx
= st
->bitmap
.tex_format
== PIPE_FORMAT_R8_UNORM
;
1296 NIR_PASS_V(state
.ir
.nir
, nir_lower_bitmap
, &options
);
1300 /* glDrawPixels (color only) */
1301 if (key
->drawpixels
) {
1302 nir_lower_drawpixels_options options
= {{0}};
1303 unsigned samplers_used
= stfp
->Base
.SamplersUsed
;
1305 /* Find the first unused slot. */
1306 variant
->drawpix_sampler
= ffs(~samplers_used
) - 1;
1307 options
.drawpix_sampler
= variant
->drawpix_sampler
;
1308 samplers_used
|= (1 << variant
->drawpix_sampler
);
1310 options
.pixel_maps
= key
->pixelMaps
;
1311 if (key
->pixelMaps
) {
1312 variant
->pixelmap_sampler
= ffs(~samplers_used
) - 1;
1313 options
.pixelmap_sampler
= variant
->pixelmap_sampler
;
1316 options
.scale_and_bias
= key
->scaleAndBias
;
1317 if (key
->scaleAndBias
) {
1318 _mesa_add_state_reference(params
, scale_state
);
1319 memcpy(options
.scale_state_tokens
, scale_state
,
1320 sizeof(options
.scale_state_tokens
));
1321 _mesa_add_state_reference(params
, bias_state
);
1322 memcpy(options
.bias_state_tokens
, bias_state
,
1323 sizeof(options
.bias_state_tokens
));
1326 _mesa_add_state_reference(params
, texcoord_state
);
1327 memcpy(options
.texcoord_state_tokens
, texcoord_state
,
1328 sizeof(options
.texcoord_state_tokens
));
1330 NIR_PASS_V(state
.ir
.nir
, nir_lower_drawpixels
, &options
);
1334 if (unlikely(key
->external
.lower_nv12
|| key
->external
.lower_iyuv
||
1335 key
->external
.lower_xy_uxvx
|| key
->external
.lower_yx_xuxv
||
1336 key
->external
.lower_ayuv
|| key
->external
.lower_xyuv
)) {
1338 st_nir_lower_samplers(pipe
->screen
, state
.ir
.nir
,
1339 stfp
->shader_program
, &stfp
->Base
);
1341 nir_lower_tex_options options
= {0};
1342 options
.lower_y_uv_external
= key
->external
.lower_nv12
;
1343 options
.lower_y_u_v_external
= key
->external
.lower_iyuv
;
1344 options
.lower_xy_uxvx_external
= key
->external
.lower_xy_uxvx
;
1345 options
.lower_yx_xuxv_external
= key
->external
.lower_yx_xuxv
;
1346 options
.lower_ayuv_external
= key
->external
.lower_ayuv
;
1347 options
.lower_xyuv_external
= key
->external
.lower_xyuv
;
1348 NIR_PASS_V(state
.ir
.nir
, nir_lower_tex
, &options
);
1352 if (finalize
|| !st
->allow_st_finalize_nir_twice
) {
1353 st_finalize_nir(st
, &stfp
->Base
, stfp
->shader_program
, state
.ir
.nir
,
1357 /* This pass needs to happen *after* nir_lower_sampler */
1358 if (unlikely(key
->external
.lower_nv12
|| key
->external
.lower_iyuv
||
1359 key
->external
.lower_xy_uxvx
|| key
->external
.lower_yx_xuxv
||
1360 key
->external
.lower_ayuv
|| key
->external
.lower_xyuv
)) {
1361 NIR_PASS_V(state
.ir
.nir
, st_nir_lower_tex_src_plane
,
1362 ~stfp
->Base
.SamplersUsed
,
1363 key
->external
.lower_nv12
|| key
->external
.lower_xy_uxvx
||
1364 key
->external
.lower_yx_xuxv
,
1365 key
->external
.lower_iyuv
);
1369 if (finalize
|| !st
->allow_st_finalize_nir_twice
) {
1370 /* Some of the lowering above may have introduced new varyings */
1371 nir_shader_gather_info(state
.ir
.nir
,
1372 nir_shader_get_entrypoint(state
.ir
.nir
));
1374 struct pipe_screen
*screen
= pipe
->screen
;
1375 if (screen
->finalize_nir
)
1376 screen
->finalize_nir(screen
, state
.ir
.nir
, false);
1379 if (ST_DEBUG
& DEBUG_PRINT_IR
)
1380 nir_print_shader(state
.ir
.nir
, stderr
);
1382 variant
->base
.driver_shader
= pipe
->create_fs_state(pipe
, &state
);
1383 variant
->key
= *key
;
1388 state
.tokens
= stfp
->state
.tokens
;
1390 assert(!(key
->bitmap
&& key
->drawpixels
));
1392 /* Fix texture targets and add fog for ATI_fs */
1394 const struct tgsi_token
*tokens
= st_fixup_atifs(state
.tokens
, key
);
1397 state
.tokens
= tokens
;
1399 fprintf(stderr
, "mesa: cannot post-process ATI_fs\n");
1402 /* Emulate features. */
1403 if (key
->clamp_color
|| key
->persample_shading
) {
1404 const struct tgsi_token
*tokens
;
1406 (key
->clamp_color
? TGSI_EMU_CLAMP_COLOR_OUTPUTS
: 0) |
1407 (key
->persample_shading
? TGSI_EMU_FORCE_PERSAMPLE_INTERP
: 0);
1409 tokens
= tgsi_emulate(state
.tokens
, flags
);
1412 if (state
.tokens
!= stfp
->state
.tokens
)
1413 tgsi_free_tokens(state
.tokens
);
1414 state
.tokens
= tokens
;
1416 fprintf(stderr
, "mesa: cannot emulate deprecated features\n");
1421 const struct tgsi_token
*tokens
;
1423 variant
->bitmap_sampler
= ffs(~stfp
->Base
.SamplersUsed
) - 1;
1425 tokens
= st_get_bitmap_shader(state
.tokens
,
1426 st
->internal_target
,
1427 variant
->bitmap_sampler
,
1428 st
->needs_texcoord_semantic
,
1429 st
->bitmap
.tex_format
==
1430 PIPE_FORMAT_R8_UNORM
);
1433 if (state
.tokens
!= stfp
->state
.tokens
)
1434 tgsi_free_tokens(state
.tokens
);
1435 state
.tokens
= tokens
;
1437 fprintf(stderr
, "mesa: cannot create a shader for glBitmap\n");
1440 /* glDrawPixels (color only) */
1441 if (key
->drawpixels
) {
1442 const struct tgsi_token
*tokens
;
1443 unsigned scale_const
= 0, bias_const
= 0, texcoord_const
= 0;
1445 /* Find the first unused slot. */
1446 variant
->drawpix_sampler
= ffs(~stfp
->Base
.SamplersUsed
) - 1;
1448 if (key
->pixelMaps
) {
1449 unsigned samplers_used
= stfp
->Base
.SamplersUsed
|
1450 (1 << variant
->drawpix_sampler
);
1452 variant
->pixelmap_sampler
= ffs(~samplers_used
) - 1;
1455 if (key
->scaleAndBias
) {
1456 scale_const
= _mesa_add_state_reference(params
, scale_state
);
1457 bias_const
= _mesa_add_state_reference(params
, bias_state
);
1460 texcoord_const
= _mesa_add_state_reference(params
, texcoord_state
);
1462 tokens
= st_get_drawpix_shader(state
.tokens
,
1463 st
->needs_texcoord_semantic
,
1464 key
->scaleAndBias
, scale_const
,
1465 bias_const
, key
->pixelMaps
,
1466 variant
->drawpix_sampler
,
1467 variant
->pixelmap_sampler
,
1468 texcoord_const
, st
->internal_target
);
1471 if (state
.tokens
!= stfp
->state
.tokens
)
1472 tgsi_free_tokens(state
.tokens
);
1473 state
.tokens
= tokens
;
1475 fprintf(stderr
, "mesa: cannot create a shader for glDrawPixels\n");
1478 if (unlikely(key
->external
.lower_nv12
|| key
->external
.lower_iyuv
||
1479 key
->external
.lower_xy_uxvx
|| key
->external
.lower_yx_xuxv
)) {
1480 const struct tgsi_token
*tokens
;
1482 /* samplers inserted would conflict, but this should be unpossible: */
1483 assert(!(key
->bitmap
|| key
->drawpixels
));
1485 tokens
= st_tgsi_lower_yuv(state
.tokens
,
1486 ~stfp
->Base
.SamplersUsed
,
1487 key
->external
.lower_nv12
||
1488 key
->external
.lower_xy_uxvx
||
1489 key
->external
.lower_yx_xuxv
,
1490 key
->external
.lower_iyuv
);
1492 if (state
.tokens
!= stfp
->state
.tokens
)
1493 tgsi_free_tokens(state
.tokens
);
1494 state
.tokens
= tokens
;
1496 fprintf(stderr
, "mesa: cannot create a shader for samplerExternalOES\n");
1500 if (key
->lower_depth_clamp
) {
1501 unsigned depth_range_const
= _mesa_add_state_reference(params
, depth_range_state
);
1503 const struct tgsi_token
*tokens
;
1504 tokens
= st_tgsi_lower_depth_clamp_fs(state
.tokens
, depth_range_const
);
1505 if (state
.tokens
!= stfp
->state
.tokens
)
1506 tgsi_free_tokens(state
.tokens
);
1507 state
.tokens
= tokens
;
1510 if (ST_DEBUG
& DEBUG_PRINT_IR
)
1511 tgsi_dump(state
.tokens
, 0);
1513 /* fill in variant */
1514 variant
->base
.driver_shader
= pipe
->create_fs_state(pipe
, &state
);
1515 variant
->key
= *key
;
1517 if (state
.tokens
!= stfp
->state
.tokens
)
1518 tgsi_free_tokens(state
.tokens
);
1523 * Translate fragment program if needed.
1525 struct st_fp_variant
*
1526 st_get_fp_variant(struct st_context
*st
,
1527 struct st_program
*stfp
,
1528 const struct st_fp_variant_key
*key
)
1530 struct st_fp_variant
*fpv
;
1532 /* Search for existing variant */
1533 for (fpv
= st_fp_variant(stfp
->variants
); fpv
;
1534 fpv
= st_fp_variant(fpv
->base
.next
)) {
1535 if (memcmp(&fpv
->key
, key
, sizeof(*key
)) == 0) {
1542 fpv
= st_create_fp_variant(st
, stfp
, key
);
1544 fpv
->base
.st
= key
->st
;
1546 if (key
->bitmap
|| key
->drawpixels
) {
1547 /* Regular variants should always come before the
1548 * bitmap & drawpixels variants, (unless there
1549 * are no regular variants) so that
1550 * st_update_fp can take a fast path when
1551 * shader_has_one_variant is set.
1553 if (!stfp
->variants
) {
1554 stfp
->variants
= &fpv
->base
;
1556 /* insert into list after the first one */
1557 fpv
->base
.next
= stfp
->variants
->next
;
1558 stfp
->variants
->next
= &fpv
->base
;
1561 /* insert into list */
1562 fpv
->base
.next
= stfp
->variants
;
1563 stfp
->variants
= &fpv
->base
;
1572 * Translate a program. This is common code for geometry and tessellation
1576 st_translate_common_program(struct st_context
*st
,
1577 struct st_program
*stp
)
1579 struct gl_program
*prog
= &stp
->Base
;
1580 enum pipe_shader_type stage
=
1581 pipe_shader_type_from_mesa(stp
->Base
.info
.stage
);
1582 struct ureg_program
*ureg
= ureg_create_with_screen(stage
, st
->pipe
->screen
);
1587 ureg_setup_shader_info(ureg
, &stp
->Base
.info
);
1589 ubyte inputSlotToAttr
[VARYING_SLOT_TESS_MAX
];
1590 ubyte inputMapping
[VARYING_SLOT_TESS_MAX
];
1591 ubyte outputMapping
[VARYING_SLOT_TESS_MAX
];
1594 ubyte input_semantic_name
[PIPE_MAX_SHADER_INPUTS
];
1595 ubyte input_semantic_index
[PIPE_MAX_SHADER_INPUTS
];
1596 uint num_inputs
= 0;
1598 ubyte output_semantic_name
[PIPE_MAX_SHADER_OUTPUTS
];
1599 ubyte output_semantic_index
[PIPE_MAX_SHADER_OUTPUTS
];
1600 uint num_outputs
= 0;
1604 memset(inputSlotToAttr
, 0, sizeof(inputSlotToAttr
));
1605 memset(inputMapping
, 0, sizeof(inputMapping
));
1606 memset(outputMapping
, 0, sizeof(outputMapping
));
1607 memset(&stp
->state
, 0, sizeof(stp
->state
));
1610 * Convert Mesa program inputs to TGSI input register semantics.
1612 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
1613 if ((prog
->info
.inputs_read
& BITFIELD64_BIT(attr
)) == 0)
1616 unsigned slot
= num_inputs
++;
1618 inputMapping
[attr
] = slot
;
1619 inputSlotToAttr
[slot
] = attr
;
1621 unsigned semantic_name
, semantic_index
;
1622 tgsi_get_gl_varying_semantic(attr
, st
->needs_texcoord_semantic
,
1623 &semantic_name
, &semantic_index
);
1624 input_semantic_name
[slot
] = semantic_name
;
1625 input_semantic_index
[slot
] = semantic_index
;
1628 /* Also add patch inputs. */
1629 for (attr
= 0; attr
< 32; attr
++) {
1630 if (prog
->info
.patch_inputs_read
& (1u << attr
)) {
1631 GLuint slot
= num_inputs
++;
1632 GLuint patch_attr
= VARYING_SLOT_PATCH0
+ attr
;
1634 inputMapping
[patch_attr
] = slot
;
1635 inputSlotToAttr
[slot
] = patch_attr
;
1636 input_semantic_name
[slot
] = TGSI_SEMANTIC_PATCH
;
1637 input_semantic_index
[slot
] = attr
;
1641 /* initialize output semantics to defaults */
1642 for (i
= 0; i
< PIPE_MAX_SHADER_OUTPUTS
; i
++) {
1643 output_semantic_name
[i
] = TGSI_SEMANTIC_GENERIC
;
1644 output_semantic_index
[i
] = 0;
1648 * Determine number of outputs, the (default) output register
1649 * mapping and the semantic information for each output.
1651 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
1652 if (prog
->info
.outputs_written
& BITFIELD64_BIT(attr
)) {
1653 GLuint slot
= num_outputs
++;
1655 outputMapping
[attr
] = slot
;
1657 unsigned semantic_name
, semantic_index
;
1658 tgsi_get_gl_varying_semantic(attr
, st
->needs_texcoord_semantic
,
1659 &semantic_name
, &semantic_index
);
1660 output_semantic_name
[slot
] = semantic_name
;
1661 output_semantic_index
[slot
] = semantic_index
;
1665 /* Also add patch outputs. */
1666 for (attr
= 0; attr
< 32; attr
++) {
1667 if (prog
->info
.patch_outputs_written
& (1u << attr
)) {
1668 GLuint slot
= num_outputs
++;
1669 GLuint patch_attr
= VARYING_SLOT_PATCH0
+ attr
;
1671 outputMapping
[patch_attr
] = slot
;
1672 output_semantic_name
[slot
] = TGSI_SEMANTIC_PATCH
;
1673 output_semantic_index
[slot
] = attr
;
1677 st_translate_program(st
->ctx
,
1686 input_semantic_name
,
1687 input_semantic_index
,
1692 output_semantic_name
,
1693 output_semantic_index
);
1695 stp
->state
.tokens
= ureg_get_tokens(ureg
, NULL
);
1699 st_translate_stream_output_info(prog
);
1701 st_store_ir_in_disk_cache(st
, prog
, false);
1703 if (ST_DEBUG
& DEBUG_PRINT_IR
&& ST_DEBUG
& DEBUG_MESA
)
1704 _mesa_print_program(prog
);
1706 free_glsl_to_tgsi_visitor(stp
->glsl_to_tgsi
);
1707 stp
->glsl_to_tgsi
= NULL
;
1713 * Get/create a basic program variant.
1716 st_get_common_variant(struct st_context
*st
,
1717 struct st_program
*prog
,
1718 const struct st_common_variant_key
*key
)
1720 struct pipe_context
*pipe
= st
->pipe
;
1721 struct st_variant
*v
;
1722 struct pipe_shader_state state
= {0};
1723 struct gl_program_parameter_list
*params
= prog
->Base
.Parameters
;
1725 /* Search for existing variant */
1726 for (v
= prog
->variants
; v
; v
= v
->next
) {
1727 if (memcmp(&st_common_variant(v
)->key
, key
, sizeof(*key
)) == 0)
1733 v
= (struct st_variant
*)CALLOC_STRUCT(st_common_variant
);
1735 if (prog
->state
.type
== PIPE_SHADER_IR_NIR
) {
1736 bool finalize
= false;
1738 state
.type
= PIPE_SHADER_IR_NIR
;
1739 state
.ir
.nir
= get_nir_shader(st
, prog
);
1741 if (key
->clamp_color
) {
1742 NIR_PASS_V(state
.ir
.nir
, nir_lower_clamp_color_outputs
);
1746 if (key
->lower_ucp
) {
1747 lower_ucp(st
, state
.ir
.nir
, key
->lower_ucp
, params
);
1751 state
.stream_output
= prog
->state
.stream_output
;
1753 if (finalize
|| !st
->allow_st_finalize_nir_twice
) {
1754 st_finalize_nir(st
, &prog
->Base
, prog
->shader_program
,
1755 state
.ir
.nir
, true);
1758 if (ST_DEBUG
& DEBUG_PRINT_IR
)
1759 nir_print_shader(state
.ir
.nir
, stderr
);
1761 if (key
->lower_depth_clamp
) {
1762 struct gl_program_parameter_list
*params
= prog
->Base
.Parameters
;
1764 unsigned depth_range_const
=
1765 _mesa_add_state_reference(params
, depth_range_state
);
1767 const struct tgsi_token
*tokens
;
1769 st_tgsi_lower_depth_clamp(prog
->state
.tokens
,
1771 key
->clip_negative_one_to_one
);
1773 if (tokens
!= prog
->state
.tokens
)
1774 tgsi_free_tokens(prog
->state
.tokens
);
1776 prog
->state
.tokens
= tokens
;
1778 state
= prog
->state
;
1780 if (ST_DEBUG
& DEBUG_PRINT_IR
)
1781 tgsi_dump(state
.tokens
, 0);
1783 /* fill in new variant */
1784 switch (prog
->Base
.info
.stage
) {
1785 case MESA_SHADER_TESS_CTRL
:
1786 v
->driver_shader
= pipe
->create_tcs_state(pipe
, &state
);
1788 case MESA_SHADER_TESS_EVAL
:
1789 v
->driver_shader
= pipe
->create_tes_state(pipe
, &state
);
1791 case MESA_SHADER_GEOMETRY
:
1792 v
->driver_shader
= pipe
->create_gs_state(pipe
, &state
);
1794 case MESA_SHADER_COMPUTE
: {
1795 struct pipe_compute_state cs
= {0};
1796 cs
.ir_type
= state
.type
;
1797 cs
.req_local_mem
= prog
->Base
.info
.cs
.shared_size
;
1799 if (state
.type
== PIPE_SHADER_IR_NIR
)
1800 cs
.prog
= state
.ir
.nir
;
1802 cs
.prog
= state
.tokens
;
1804 v
->driver_shader
= pipe
->create_compute_state(pipe
, &cs
);
1808 assert(!"unhandled shader type");
1813 st_common_variant(v
)->key
= *key
;
1816 /* insert into list */
1817 v
->next
= prog
->variants
;
1827 * Vert/Geom/Frag programs have per-context variants. Free all the
1828 * variants attached to the given program which match the given context.
1831 destroy_program_variants(struct st_context
*st
, struct gl_program
*target
)
1833 if (!target
|| target
== &_mesa_DummyProgram
)
1836 struct st_program
*p
= st_program(target
);
1837 struct st_variant
*v
, **prevPtr
= &p
->variants
;
1838 bool unbound
= false;
1840 for (v
= p
->variants
; v
; ) {
1841 struct st_variant
*next
= v
->next
;
1844 st_unbind_program(st
, p
);
1848 /* unlink from list */
1850 /* destroy this variant */
1851 delete_variant(st
, v
, target
->Target
);
1862 * Callback for _mesa_HashWalk. Free all the shader's program variants
1863 * which match the given context.
1866 destroy_shader_program_variants_cb(GLuint key
, void *data
, void *userData
)
1868 struct st_context
*st
= (struct st_context
*) userData
;
1869 struct gl_shader
*shader
= (struct gl_shader
*) data
;
1871 switch (shader
->Type
) {
1872 case GL_SHADER_PROGRAM_MESA
:
1874 struct gl_shader_program
*shProg
= (struct gl_shader_program
*) data
;
1877 for (i
= 0; i
< ARRAY_SIZE(shProg
->_LinkedShaders
); i
++) {
1878 if (shProg
->_LinkedShaders
[i
])
1879 destroy_program_variants(st
, shProg
->_LinkedShaders
[i
]->Program
);
1883 case GL_VERTEX_SHADER
:
1884 case GL_FRAGMENT_SHADER
:
1885 case GL_GEOMETRY_SHADER
:
1886 case GL_TESS_CONTROL_SHADER
:
1887 case GL_TESS_EVALUATION_SHADER
:
1888 case GL_COMPUTE_SHADER
:
1897 * Callback for _mesa_HashWalk. Free all the program variants which match
1898 * the given context.
1901 destroy_program_variants_cb(GLuint key
, void *data
, void *userData
)
1903 struct st_context
*st
= (struct st_context
*) userData
;
1904 struct gl_program
*program
= (struct gl_program
*) data
;
1905 destroy_program_variants(st
, program
);
1910 * Walk over all shaders and programs to delete any variants which
1911 * belong to the given context.
1912 * This is called during context tear-down.
1915 st_destroy_program_variants(struct st_context
*st
)
1917 /* If shaders can be shared with other contexts, the last context will
1918 * call DeleteProgram on all shaders, releasing everything.
1920 if (st
->has_shareable_shaders
)
1923 /* ARB vert/frag program */
1924 _mesa_HashWalk(st
->ctx
->Shared
->Programs
,
1925 destroy_program_variants_cb
, st
);
1927 /* GLSL vert/frag/geom shaders */
1928 _mesa_HashWalk(st
->ctx
->Shared
->ShaderObjects
,
1929 destroy_shader_program_variants_cb
, st
);
1934 * Compile one shader variant.
1937 st_precompile_shader_variant(struct st_context
*st
,
1938 struct gl_program
*prog
)
1940 switch (prog
->Target
) {
1941 case GL_VERTEX_PROGRAM_ARB
: {
1942 struct st_program
*p
= (struct st_program
*)prog
;
1943 struct st_common_variant_key key
;
1945 memset(&key
, 0, sizeof(key
));
1947 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
1948 st_get_vp_variant(st
, p
, &key
);
1952 case GL_FRAGMENT_PROGRAM_ARB
: {
1953 struct st_program
*p
= (struct st_program
*)prog
;
1954 struct st_fp_variant_key key
;
1956 memset(&key
, 0, sizeof(key
));
1958 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
1959 st_get_fp_variant(st
, p
, &key
);
1963 case GL_TESS_CONTROL_PROGRAM_NV
:
1964 case GL_TESS_EVALUATION_PROGRAM_NV
:
1965 case GL_GEOMETRY_PROGRAM_NV
:
1966 case GL_COMPUTE_PROGRAM_NV
: {
1967 struct st_program
*p
= st_program(prog
);
1968 struct st_common_variant_key key
;
1970 memset(&key
, 0, sizeof(key
));
1972 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
1973 st_get_common_variant(st
, p
, &key
);
1983 st_serialize_nir(struct st_program
*stp
)
1985 if (!stp
->serialized_nir
) {
1990 nir_serialize(&blob
, stp
->Base
.nir
, false);
1991 blob_finish_get_buffer(&blob
, &stp
->serialized_nir
, &size
);
1992 stp
->serialized_nir_size
= size
;
1997 st_finalize_program(struct st_context
*st
, struct gl_program
*prog
)
1999 if (st
->current_program
[prog
->info
.stage
] == prog
) {
2000 if (prog
->info
.stage
== MESA_SHADER_VERTEX
)
2001 st
->dirty
|= ST_NEW_VERTEX_PROGRAM(st
, (struct st_program
*)prog
);
2003 st
->dirty
|= ((struct st_program
*)prog
)->affected_states
;
2007 nir_sweep(prog
->nir
);
2009 /* This is only needed for ARB_vp/fp programs and when the disk cache
2010 * is disabled. If the disk cache is enabled, GLSL programs are
2011 * serialized in write_nir_to_cache.
2013 st_serialize_nir(st_program(prog
));
2016 /* Create Gallium shaders now instead of on demand. */
2017 if (ST_DEBUG
& DEBUG_PRECOMPILE
||
2018 st
->shader_has_one_variant
[prog
->info
.stage
])
2019 st_precompile_shader_variant(st
, prog
);