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 requires correct shader info. */
371 nir_shader_gather_info(nir
, nir_shader_get_entrypoint(nir
));
373 st_nir_assign_vs_in_locations(nir
);
377 * Translate ARB (asm) program to NIR
380 st_translate_prog_to_nir(struct st_context
*st
, struct gl_program
*prog
,
381 gl_shader_stage stage
)
383 struct pipe_screen
*screen
= st
->pipe
->screen
;
384 const struct gl_shader_compiler_options
*options
=
385 &st
->ctx
->Const
.ShaderCompilerOptions
[stage
];
387 /* Translate to NIR */
388 nir_shader
*nir
= prog_to_nir(prog
, options
->NirOptions
);
389 NIR_PASS_V(nir
, nir_lower_regs_to_ssa
); /* turn registers into SSA */
390 nir_validate_shader(nir
, "after st/ptn lower_regs_to_ssa");
392 NIR_PASS_V(nir
, st_nir_lower_wpos_ytransform
, prog
, screen
);
393 NIR_PASS_V(nir
, nir_lower_system_values
);
394 NIR_PASS_V(nir
, nir_lower_compute_system_values
, NULL
);
397 NIR_PASS_V(nir
, nir_opt_constant_folding
);
399 st_finalize_nir_before_variants(nir
);
401 if (st
->allow_st_finalize_nir_twice
)
402 st_finalize_nir(st
, prog
, NULL
, nir
, true);
404 nir_validate_shader(nir
, "after st/glsl finalize_nir");
410 st_prepare_vertex_program(struct st_program
*stp
)
412 struct st_vertex_program
*stvp
= (struct st_vertex_program
*)stp
;
414 stvp
->num_inputs
= 0;
415 memset(stvp
->input_to_index
, ~0, sizeof(stvp
->input_to_index
));
416 memset(stvp
->result_to_output
, ~0, sizeof(stvp
->result_to_output
));
418 /* Determine number of inputs, the mappings between VERT_ATTRIB_x
419 * and TGSI generic input indexes, plus input attrib semantic info.
421 for (unsigned attr
= 0; attr
< VERT_ATTRIB_MAX
; attr
++) {
422 if ((stp
->Base
.info
.inputs_read
& BITFIELD64_BIT(attr
)) != 0) {
423 stvp
->input_to_index
[attr
] = stvp
->num_inputs
;
424 stvp
->index_to_input
[stvp
->num_inputs
] = attr
;
427 if ((stp
->Base
.DualSlotInputs
& BITFIELD64_BIT(attr
)) != 0) {
428 /* add placeholder for second part of a double attribute */
429 stvp
->index_to_input
[stvp
->num_inputs
] = ST_DOUBLE_ATTRIB_PLACEHOLDER
;
434 /* pre-setup potentially unused edgeflag input */
435 stvp
->input_to_index
[VERT_ATTRIB_EDGEFLAG
] = stvp
->num_inputs
;
436 stvp
->index_to_input
[stvp
->num_inputs
] = VERT_ATTRIB_EDGEFLAG
;
438 /* Compute mapping of vertex program outputs to slots. */
439 unsigned num_outputs
= 0;
440 for (unsigned attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
441 if (stp
->Base
.info
.outputs_written
& BITFIELD64_BIT(attr
))
442 stvp
->result_to_output
[attr
] = num_outputs
++;
444 /* pre-setup potentially unused edgeflag output */
445 stvp
->result_to_output
[VARYING_SLOT_EDGE
] = num_outputs
;
449 st_translate_stream_output_info(struct gl_program
*prog
)
451 struct gl_transform_feedback_info
*info
= prog
->sh
.LinkedTransformFeedback
;
455 /* Determine the (default) output register mapping for each output. */
456 unsigned num_outputs
= 0;
457 ubyte output_mapping
[VARYING_SLOT_TESS_MAX
];
458 memset(output_mapping
, 0, sizeof(output_mapping
));
460 for (unsigned attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
461 if (prog
->info
.outputs_written
& BITFIELD64_BIT(attr
))
462 output_mapping
[attr
] = num_outputs
++;
465 /* Translate stream output info. */
466 struct pipe_stream_output_info
*so_info
=
467 &((struct st_program
*)prog
)->state
.stream_output
;
469 for (unsigned i
= 0; i
< info
->NumOutputs
; i
++) {
470 so_info
->output
[i
].register_index
=
471 output_mapping
[info
->Outputs
[i
].OutputRegister
];
472 so_info
->output
[i
].start_component
= info
->Outputs
[i
].ComponentOffset
;
473 so_info
->output
[i
].num_components
= info
->Outputs
[i
].NumComponents
;
474 so_info
->output
[i
].output_buffer
= info
->Outputs
[i
].OutputBuffer
;
475 so_info
->output
[i
].dst_offset
= info
->Outputs
[i
].DstOffset
;
476 so_info
->output
[i
].stream
= info
->Outputs
[i
].StreamId
;
479 for (unsigned i
= 0; i
< PIPE_MAX_SO_BUFFERS
; i
++) {
480 so_info
->stride
[i
] = info
->Buffers
[i
].Stride
;
482 so_info
->num_outputs
= info
->NumOutputs
;
486 * Translate a vertex program.
489 st_translate_vertex_program(struct st_context
*st
,
490 struct st_program
*stp
)
492 struct ureg_program
*ureg
;
493 enum pipe_error error
;
494 unsigned num_outputs
= 0;
496 ubyte output_semantic_name
[VARYING_SLOT_MAX
] = {0};
497 ubyte output_semantic_index
[VARYING_SLOT_MAX
] = {0};
499 if (stp
->Base
.arb
.IsPositionInvariant
)
500 _mesa_insert_mvp_code(st
->ctx
, &stp
->Base
);
503 if (!stp
->glsl_to_tgsi
) {
504 _mesa_remove_output_reads(&stp
->Base
, PROGRAM_OUTPUT
);
506 /* This determines which states will be updated when the assembly
509 stp
->affected_states
= ST_NEW_VS_STATE
|
511 ST_NEW_VERTEX_ARRAYS
;
513 if (stp
->Base
.Parameters
->NumParameters
)
514 stp
->affected_states
|= ST_NEW_VS_CONSTANTS
;
516 /* Translate to NIR if preferred. */
517 if (PIPE_SHADER_IR_NIR
==
518 st
->pipe
->screen
->get_shader_param(st
->pipe
->screen
,
520 PIPE_SHADER_CAP_PREFERRED_IR
)) {
521 assert(!stp
->glsl_to_tgsi
);
524 ralloc_free(stp
->Base
.nir
);
526 if (stp
->serialized_nir
) {
527 free(stp
->serialized_nir
);
528 stp
->serialized_nir
= NULL
;
531 stp
->state
.type
= PIPE_SHADER_IR_NIR
;
532 stp
->Base
.nir
= st_translate_prog_to_nir(st
, &stp
->Base
,
534 stp
->Base
.info
= stp
->Base
.nir
->info
;
536 /* For st_draw_feedback, we need to generate TGSI too if draw doesn't
539 /* TODO: Draw can't handle lowered IO. */
540 if (draw_has_llvm() && !stp
->Base
.info
.io_lowered
) {
541 st_prepare_vertex_program(stp
);
547 st_prepare_vertex_program(stp
);
549 /* Get semantic names and indices. */
550 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
551 if (stp
->Base
.info
.outputs_written
& BITFIELD64_BIT(attr
)) {
552 unsigned slot
= num_outputs
++;
553 unsigned semantic_name
, semantic_index
;
554 tgsi_get_gl_varying_semantic(attr
, st
->needs_texcoord_semantic
,
555 &semantic_name
, &semantic_index
);
556 output_semantic_name
[slot
] = semantic_name
;
557 output_semantic_index
[slot
] = semantic_index
;
560 /* pre-setup potentially unused edgeflag output */
561 output_semantic_name
[num_outputs
] = TGSI_SEMANTIC_EDGEFLAG
;
562 output_semantic_index
[num_outputs
] = 0;
564 ureg
= ureg_create_with_screen(PIPE_SHADER_VERTEX
, st
->pipe
->screen
);
568 ureg_setup_shader_info(ureg
, &stp
->Base
.info
);
570 if (ST_DEBUG
& DEBUG_MESA
) {
571 _mesa_print_program(&stp
->Base
);
572 _mesa_print_program_parameters(st
->ctx
, &stp
->Base
);
576 struct st_vertex_program
*stvp
= (struct st_vertex_program
*)stp
;
578 if (stp
->glsl_to_tgsi
) {
579 error
= st_translate_program(st
->ctx
,
586 stvp
->input_to_index
,
587 NULL
, /* inputSlotToAttr */
588 NULL
, /* input semantic name */
589 NULL
, /* input semantic index */
590 NULL
, /* interp mode */
593 stvp
->result_to_output
,
594 output_semantic_name
,
595 output_semantic_index
);
597 st_translate_stream_output_info(&stp
->Base
);
599 free_glsl_to_tgsi_visitor(stp
->glsl_to_tgsi
);
601 error
= st_translate_mesa_program(st
->ctx
,
607 stvp
->input_to_index
,
608 NULL
, /* input semantic name */
609 NULL
, /* input semantic index */
613 stvp
->result_to_output
,
614 output_semantic_name
,
615 output_semantic_index
);
618 debug_printf("%s: failed to translate Mesa program:\n", __func__
);
619 _mesa_print_program(&stp
->Base
);
624 stp
->state
.tokens
= ureg_get_tokens(ureg
, NULL
);
627 if (stp
->glsl_to_tgsi
) {
628 stp
->glsl_to_tgsi
= NULL
;
629 st_store_ir_in_disk_cache(st
, &stp
->Base
, false);
632 return stp
->state
.tokens
!= NULL
;
635 static struct nir_shader
*
636 get_nir_shader(struct st_context
*st
, struct st_program
*stp
)
639 nir_shader
*nir
= stp
->Base
.nir
;
641 /* The first shader variant takes ownership of NIR, so that there is
642 * no cloning. Additional shader variants are always generated from
643 * serialized NIR to save memory.
645 stp
->Base
.nir
= NULL
;
646 assert(stp
->serialized_nir
&& stp
->serialized_nir_size
);
650 struct blob_reader blob_reader
;
651 const struct nir_shader_compiler_options
*options
=
652 st
->ctx
->Const
.ShaderCompilerOptions
[stp
->Base
.info
.stage
].NirOptions
;
654 blob_reader_init(&blob_reader
, stp
->serialized_nir
, stp
->serialized_nir_size
);
655 return nir_deserialize(NULL
, options
, &blob_reader
);
659 lower_ucp(struct st_context
*st
,
660 struct nir_shader
*nir
,
661 unsigned ucp_enables
,
662 struct gl_program_parameter_list
*params
)
664 if (nir
->info
.outputs_written
& VARYING_BIT_CLIP_DIST0
)
665 NIR_PASS_V(nir
, nir_lower_clip_disable
, ucp_enables
);
667 struct pipe_screen
*screen
= st
->pipe
->screen
;
668 bool can_compact
= screen
->get_param(screen
,
669 PIPE_CAP_NIR_COMPACT_ARRAYS
);
670 bool use_eye
= st
->ctx
->_Shader
->CurrentProgram
[MESA_SHADER_VERTEX
] != NULL
;
672 gl_state_index16 clipplane_state
[MAX_CLIP_PLANES
][STATE_LENGTH
];
673 for (int i
= 0; i
< MAX_CLIP_PLANES
; ++i
) {
675 clipplane_state
[i
][0] = STATE_CLIPPLANE
;
676 clipplane_state
[i
][1] = i
;
678 clipplane_state
[i
][0] = STATE_INTERNAL
;
679 clipplane_state
[i
][1] = STATE_CLIP_INTERNAL
;
680 clipplane_state
[i
][2] = i
;
682 _mesa_add_state_reference(params
, clipplane_state
[i
]);
685 if (nir
->info
.stage
== MESA_SHADER_VERTEX
) {
686 NIR_PASS_V(nir
, nir_lower_clip_vs
, ucp_enables
,
687 true, can_compact
, clipplane_state
);
688 } else if (nir
->info
.stage
== MESA_SHADER_GEOMETRY
) {
689 NIR_PASS_V(nir
, nir_lower_clip_gs
, ucp_enables
,
690 can_compact
, clipplane_state
);
693 NIR_PASS_V(nir
, nir_lower_io_to_temporaries
,
694 nir_shader_get_entrypoint(nir
), true, false);
695 NIR_PASS_V(nir
, nir_lower_global_vars_to_local
);
699 static const gl_state_index16 depth_range_state
[STATE_LENGTH
] =
700 { STATE_DEPTH_RANGE
};
702 static struct st_common_variant
*
703 st_create_vp_variant(struct st_context
*st
,
704 struct st_program
*stvp
,
705 const struct st_common_variant_key
*key
)
707 struct st_common_variant
*vpv
= CALLOC_STRUCT(st_common_variant
);
708 struct pipe_context
*pipe
= st
->pipe
;
709 struct pipe_shader_state state
= {0};
711 static const gl_state_index16 point_size_state
[STATE_LENGTH
] =
712 { STATE_INTERNAL
, STATE_POINT_SIZE_CLAMPED
, 0 };
713 struct gl_program_parameter_list
*params
= stvp
->Base
.Parameters
;
717 state
.stream_output
= stvp
->state
.stream_output
;
719 if (stvp
->state
.type
== PIPE_SHADER_IR_NIR
&&
720 (!key
->is_draw_shader
||
721 /* TODO: Draw can't handle lowered IO. */
722 (draw_has_llvm() && !stvp
->Base
.info
.io_lowered
))) {
723 bool finalize
= false;
725 state
.type
= PIPE_SHADER_IR_NIR
;
726 state
.ir
.nir
= get_nir_shader(st
, stvp
);
727 if (key
->clamp_color
) {
728 NIR_PASS_V(state
.ir
.nir
, nir_lower_clamp_color_outputs
);
731 if (key
->passthrough_edgeflags
) {
732 NIR_PASS_V(state
.ir
.nir
, nir_lower_passthrough_edgeflags
);
736 if (key
->lower_point_size
) {
737 _mesa_add_state_reference(params
, point_size_state
);
738 NIR_PASS_V(state
.ir
.nir
, nir_lower_point_size_mov
,
743 if (key
->lower_ucp
) {
744 lower_ucp(st
, state
.ir
.nir
, key
->lower_ucp
, params
);
748 if (finalize
|| !st
->allow_st_finalize_nir_twice
) {
749 st_finalize_nir(st
, &stvp
->Base
, stvp
->shader_program
, state
.ir
.nir
,
752 /* Some of the lowering above may have introduced new varyings */
753 nir_shader_gather_info(state
.ir
.nir
,
754 nir_shader_get_entrypoint(state
.ir
.nir
));
757 if (ST_DEBUG
& DEBUG_PRINT_IR
)
758 nir_print_shader(state
.ir
.nir
, stderr
);
760 if (key
->is_draw_shader
)
761 vpv
->base
.driver_shader
= draw_create_vertex_shader(st
->draw
, &state
);
763 vpv
->base
.driver_shader
= pipe
->create_vs_state(pipe
, &state
);
768 state
.type
= PIPE_SHADER_IR_TGSI
;
769 state
.tokens
= tgsi_dup_tokens(stvp
->state
.tokens
);
771 /* Emulate features. */
772 if (key
->clamp_color
|| key
->passthrough_edgeflags
) {
773 const struct tgsi_token
*tokens
;
775 (key
->clamp_color
? TGSI_EMU_CLAMP_COLOR_OUTPUTS
: 0) |
776 (key
->passthrough_edgeflags
? TGSI_EMU_PASSTHROUGH_EDGEFLAG
: 0);
778 tokens
= tgsi_emulate(state
.tokens
, flags
);
781 tgsi_free_tokens(state
.tokens
);
782 state
.tokens
= tokens
;
784 fprintf(stderr
, "mesa: cannot emulate deprecated features\n");
788 if (key
->lower_depth_clamp
) {
789 unsigned depth_range_const
=
790 _mesa_add_state_reference(params
, depth_range_state
);
792 const struct tgsi_token
*tokens
;
793 tokens
= st_tgsi_lower_depth_clamp(state
.tokens
, depth_range_const
,
794 key
->clip_negative_one_to_one
);
795 if (tokens
!= state
.tokens
)
796 tgsi_free_tokens(state
.tokens
);
797 state
.tokens
= tokens
;
800 if (ST_DEBUG
& DEBUG_PRINT_IR
)
801 tgsi_dump(state
.tokens
, 0);
803 if (key
->is_draw_shader
)
804 vpv
->base
.driver_shader
= draw_create_vertex_shader(st
->draw
, &state
);
806 vpv
->base
.driver_shader
= pipe
->create_vs_state(pipe
, &state
);
809 tgsi_free_tokens(state
.tokens
);
817 * Find/create a vertex program variant.
819 struct st_common_variant
*
820 st_get_vp_variant(struct st_context
*st
,
821 struct st_program
*stp
,
822 const struct st_common_variant_key
*key
)
824 struct st_vertex_program
*stvp
= (struct st_vertex_program
*)stp
;
825 struct st_common_variant
*vpv
;
827 /* Search for existing variant */
828 for (vpv
= st_common_variant(stp
->variants
); vpv
;
829 vpv
= st_common_variant(vpv
->base
.next
)) {
830 if (memcmp(&vpv
->key
, key
, sizeof(*key
)) == 0) {
837 vpv
= st_create_vp_variant(st
, stp
, key
);
839 vpv
->base
.st
= key
->st
;
841 unsigned num_inputs
= stvp
->num_inputs
+ key
->passthrough_edgeflags
;
842 for (unsigned index
= 0; index
< num_inputs
; ++index
) {
843 unsigned attr
= stvp
->index_to_input
[index
];
844 if (attr
== ST_DOUBLE_ATTRIB_PLACEHOLDER
)
846 vpv
->vert_attrib_mask
|= 1u << attr
;
849 /* insert into list */
850 vpv
->base
.next
= stp
->variants
;
851 stp
->variants
= &vpv
->base
;
860 * Translate a Mesa fragment shader into a TGSI shader.
863 st_translate_fragment_program(struct st_context
*st
,
864 struct st_program
*stfp
)
866 /* Non-GLSL programs: */
867 if (!stfp
->glsl_to_tgsi
) {
868 _mesa_remove_output_reads(&stfp
->Base
, PROGRAM_OUTPUT
);
869 if (st
->ctx
->Const
.GLSLFragCoordIsSysVal
)
870 _mesa_program_fragment_position_to_sysval(&stfp
->Base
);
872 /* This determines which states will be updated when the assembly
875 * fragment.position and glDrawPixels always use constants.
877 stfp
->affected_states
= ST_NEW_FS_STATE
|
878 ST_NEW_SAMPLE_SHADING
|
882 /* Just set them for ATI_fs unconditionally. */
883 stfp
->affected_states
|= ST_NEW_FS_SAMPLER_VIEWS
|
887 if (stfp
->Base
.SamplersUsed
)
888 stfp
->affected_states
|= ST_NEW_FS_SAMPLER_VIEWS
|
892 /* Translate to NIR. */
894 PIPE_SHADER_IR_NIR
==
895 st
->pipe
->screen
->get_shader_param(st
->pipe
->screen
,
896 PIPE_SHADER_FRAGMENT
,
897 PIPE_SHADER_CAP_PREFERRED_IR
)) {
899 st_translate_prog_to_nir(st
, &stfp
->Base
, MESA_SHADER_FRAGMENT
);
902 ralloc_free(stfp
->Base
.nir
);
903 if (stfp
->serialized_nir
) {
904 free(stfp
->serialized_nir
);
905 stfp
->serialized_nir
= NULL
;
907 stfp
->state
.type
= PIPE_SHADER_IR_NIR
;
908 stfp
->Base
.nir
= nir
;
913 ubyte outputMapping
[2 * FRAG_RESULT_MAX
];
914 ubyte inputMapping
[VARYING_SLOT_MAX
];
915 ubyte inputSlotToAttr
[VARYING_SLOT_MAX
];
916 ubyte interpMode
[PIPE_MAX_SHADER_INPUTS
]; /* XXX size? */
918 GLbitfield64 inputsRead
;
919 struct ureg_program
*ureg
;
921 GLboolean write_all
= GL_FALSE
;
923 ubyte input_semantic_name
[PIPE_MAX_SHADER_INPUTS
];
924 ubyte input_semantic_index
[PIPE_MAX_SHADER_INPUTS
];
925 uint fs_num_inputs
= 0;
927 ubyte fs_output_semantic_name
[PIPE_MAX_SHADER_OUTPUTS
];
928 ubyte fs_output_semantic_index
[PIPE_MAX_SHADER_OUTPUTS
];
929 uint fs_num_outputs
= 0;
931 memset(inputSlotToAttr
, ~0, sizeof(inputSlotToAttr
));
934 * Convert Mesa program inputs to TGSI input register semantics.
936 inputsRead
= stfp
->Base
.info
.inputs_read
;
937 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
938 if ((inputsRead
& BITFIELD64_BIT(attr
)) != 0) {
939 const GLuint slot
= fs_num_inputs
++;
941 inputMapping
[attr
] = slot
;
942 inputSlotToAttr
[slot
] = attr
;
945 case VARYING_SLOT_POS
:
946 input_semantic_name
[slot
] = TGSI_SEMANTIC_POSITION
;
947 input_semantic_index
[slot
] = 0;
948 interpMode
[slot
] = TGSI_INTERPOLATE_LINEAR
;
950 case VARYING_SLOT_COL0
:
951 input_semantic_name
[slot
] = TGSI_SEMANTIC_COLOR
;
952 input_semantic_index
[slot
] = 0;
953 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
954 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_COLOR
;
956 case VARYING_SLOT_COL1
:
957 input_semantic_name
[slot
] = TGSI_SEMANTIC_COLOR
;
958 input_semantic_index
[slot
] = 1;
959 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
960 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_COLOR
;
962 case VARYING_SLOT_FOGC
:
963 input_semantic_name
[slot
] = TGSI_SEMANTIC_FOG
;
964 input_semantic_index
[slot
] = 0;
965 interpMode
[slot
] = TGSI_INTERPOLATE_PERSPECTIVE
;
967 case VARYING_SLOT_FACE
:
968 input_semantic_name
[slot
] = TGSI_SEMANTIC_FACE
;
969 input_semantic_index
[slot
] = 0;
970 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
972 case VARYING_SLOT_PRIMITIVE_ID
:
973 input_semantic_name
[slot
] = TGSI_SEMANTIC_PRIMID
;
974 input_semantic_index
[slot
] = 0;
975 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
977 case VARYING_SLOT_LAYER
:
978 input_semantic_name
[slot
] = TGSI_SEMANTIC_LAYER
;
979 input_semantic_index
[slot
] = 0;
980 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
982 case VARYING_SLOT_VIEWPORT
:
983 input_semantic_name
[slot
] = TGSI_SEMANTIC_VIEWPORT_INDEX
;
984 input_semantic_index
[slot
] = 0;
985 interpMode
[slot
] = TGSI_INTERPOLATE_CONSTANT
;
987 case VARYING_SLOT_CLIP_DIST0
:
988 input_semantic_name
[slot
] = TGSI_SEMANTIC_CLIPDIST
;
989 input_semantic_index
[slot
] = 0;
990 interpMode
[slot
] = TGSI_INTERPOLATE_PERSPECTIVE
;
992 case VARYING_SLOT_CLIP_DIST1
:
993 input_semantic_name
[slot
] = TGSI_SEMANTIC_CLIPDIST
;
994 input_semantic_index
[slot
] = 1;
995 interpMode
[slot
] = TGSI_INTERPOLATE_PERSPECTIVE
;
997 case VARYING_SLOT_CULL_DIST0
:
998 case VARYING_SLOT_CULL_DIST1
:
999 /* these should have been lowered by GLSL */
1002 /* In most cases, there is nothing special about these
1003 * inputs, so adopt a convention to use the generic
1004 * semantic name and the mesa VARYING_SLOT_ number as the
1007 * All that is required is that the vertex shader labels
1008 * its own outputs similarly, and that the vertex shader
1009 * generates at least every output required by the
1010 * fragment shader plus fixed-function hardware (such as
1013 * However, some drivers may need us to identify the PNTC and TEXi
1014 * varyings if, for example, their capability to replace them with
1015 * sprite coordinates is limited.
1017 case VARYING_SLOT_PNTC
:
1018 if (st
->needs_texcoord_semantic
) {
1019 input_semantic_name
[slot
] = TGSI_SEMANTIC_PCOORD
;
1020 input_semantic_index
[slot
] = 0;
1021 interpMode
[slot
] = TGSI_INTERPOLATE_LINEAR
;
1025 case VARYING_SLOT_TEX0
:
1026 case VARYING_SLOT_TEX1
:
1027 case VARYING_SLOT_TEX2
:
1028 case VARYING_SLOT_TEX3
:
1029 case VARYING_SLOT_TEX4
:
1030 case VARYING_SLOT_TEX5
:
1031 case VARYING_SLOT_TEX6
:
1032 case VARYING_SLOT_TEX7
:
1033 if (st
->needs_texcoord_semantic
) {
1034 input_semantic_name
[slot
] = TGSI_SEMANTIC_TEXCOORD
;
1035 input_semantic_index
[slot
] = attr
- VARYING_SLOT_TEX0
;
1036 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
1037 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_PERSPECTIVE
;
1041 case VARYING_SLOT_VAR0
:
1043 /* Semantic indices should be zero-based because drivers may choose
1044 * to assign a fixed slot determined by that index.
1045 * This is useful because ARB_separate_shader_objects uses location
1046 * qualifiers for linkage, and if the semantic index corresponds to
1047 * these locations, linkage passes in the driver become unecessary.
1049 * If needs_texcoord_semantic is true, no semantic indices will be
1050 * consumed for the TEXi varyings, and we can base the locations of
1051 * the user varyings on VAR0. Otherwise, we use TEX0 as base index.
1053 assert(attr
>= VARYING_SLOT_VAR0
|| attr
== VARYING_SLOT_PNTC
||
1054 (attr
>= VARYING_SLOT_TEX0
&& attr
<= VARYING_SLOT_TEX7
));
1055 input_semantic_name
[slot
] = TGSI_SEMANTIC_GENERIC
;
1056 input_semantic_index
[slot
] = st_get_generic_varying_index(st
, attr
);
1057 if (attr
== VARYING_SLOT_PNTC
)
1058 interpMode
[slot
] = TGSI_INTERPOLATE_LINEAR
;
1060 interpMode
[slot
] = stfp
->glsl_to_tgsi
?
1061 TGSI_INTERPOLATE_COUNT
: TGSI_INTERPOLATE_PERSPECTIVE
;
1067 inputMapping
[attr
] = -1;
1072 * Semantics and mapping for outputs
1074 GLbitfield64 outputsWritten
= stfp
->Base
.info
.outputs_written
;
1076 /* if z is written, emit that first */
1077 if (outputsWritten
& BITFIELD64_BIT(FRAG_RESULT_DEPTH
)) {
1078 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_POSITION
;
1079 fs_output_semantic_index
[fs_num_outputs
] = 0;
1080 outputMapping
[FRAG_RESULT_DEPTH
] = fs_num_outputs
;
1082 outputsWritten
&= ~(1 << FRAG_RESULT_DEPTH
);
1085 if (outputsWritten
& BITFIELD64_BIT(FRAG_RESULT_STENCIL
)) {
1086 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_STENCIL
;
1087 fs_output_semantic_index
[fs_num_outputs
] = 0;
1088 outputMapping
[FRAG_RESULT_STENCIL
] = fs_num_outputs
;
1090 outputsWritten
&= ~(1 << FRAG_RESULT_STENCIL
);
1093 if (outputsWritten
& BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK
)) {
1094 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_SAMPLEMASK
;
1095 fs_output_semantic_index
[fs_num_outputs
] = 0;
1096 outputMapping
[FRAG_RESULT_SAMPLE_MASK
] = fs_num_outputs
;
1098 outputsWritten
&= ~(1 << FRAG_RESULT_SAMPLE_MASK
);
1101 /* handle remaining outputs (color) */
1102 for (attr
= 0; attr
< ARRAY_SIZE(outputMapping
); attr
++) {
1103 const GLbitfield64 written
= attr
< FRAG_RESULT_MAX
? outputsWritten
:
1104 stfp
->Base
.SecondaryOutputsWritten
;
1105 const unsigned loc
= attr
% FRAG_RESULT_MAX
;
1107 if (written
& BITFIELD64_BIT(loc
)) {
1109 case FRAG_RESULT_DEPTH
:
1110 case FRAG_RESULT_STENCIL
:
1111 case FRAG_RESULT_SAMPLE_MASK
:
1115 case FRAG_RESULT_COLOR
:
1116 write_all
= GL_TRUE
; /* fallthrough */
1119 assert(loc
== FRAG_RESULT_COLOR
||
1120 (FRAG_RESULT_DATA0
<= loc
&& loc
< FRAG_RESULT_MAX
));
1122 index
= (loc
== FRAG_RESULT_COLOR
) ? 0 : (loc
- FRAG_RESULT_DATA0
);
1124 if (attr
>= FRAG_RESULT_MAX
) {
1125 /* Secondary color for dual source blending. */
1130 fs_output_semantic_name
[fs_num_outputs
] = TGSI_SEMANTIC_COLOR
;
1131 fs_output_semantic_index
[fs_num_outputs
] = index
;
1132 outputMapping
[attr
] = fs_num_outputs
;
1141 ureg
= ureg_create_with_screen(PIPE_SHADER_FRAGMENT
, st
->pipe
->screen
);
1145 ureg_setup_shader_info(ureg
, &stfp
->Base
.info
);
1147 if (ST_DEBUG
& DEBUG_MESA
) {
1148 _mesa_print_program(&stfp
->Base
);
1149 _mesa_print_program_parameters(st
->ctx
, &stfp
->Base
);
1152 if (write_all
== GL_TRUE
)
1153 ureg_property(ureg
, TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS
, 1);
1155 if (stfp
->glsl_to_tgsi
) {
1156 st_translate_program(st
->ctx
,
1157 PIPE_SHADER_FRAGMENT
,
1165 input_semantic_name
,
1166 input_semantic_index
,
1171 fs_output_semantic_name
,
1172 fs_output_semantic_index
);
1174 free_glsl_to_tgsi_visitor(stfp
->glsl_to_tgsi
);
1175 } else if (stfp
->ati_fs
)
1176 st_translate_atifs_program(ureg
,
1182 input_semantic_name
,
1183 input_semantic_index
,
1188 fs_output_semantic_name
,
1189 fs_output_semantic_index
);
1191 st_translate_mesa_program(st
->ctx
,
1192 PIPE_SHADER_FRAGMENT
,
1198 input_semantic_name
,
1199 input_semantic_index
,
1204 fs_output_semantic_name
,
1205 fs_output_semantic_index
);
1207 stfp
->state
.tokens
= ureg_get_tokens(ureg
, NULL
);
1210 if (stfp
->glsl_to_tgsi
) {
1211 stfp
->glsl_to_tgsi
= NULL
;
1212 st_store_ir_in_disk_cache(st
, &stfp
->Base
, false);
1215 return stfp
->state
.tokens
!= NULL
;
1218 static struct st_fp_variant
*
1219 st_create_fp_variant(struct st_context
*st
,
1220 struct st_program
*stfp
,
1221 const struct st_fp_variant_key
*key
)
1223 struct pipe_context
*pipe
= st
->pipe
;
1224 struct st_fp_variant
*variant
= CALLOC_STRUCT(st_fp_variant
);
1225 struct pipe_shader_state state
= {0};
1226 struct gl_program_parameter_list
*params
= stfp
->Base
.Parameters
;
1227 static const gl_state_index16 texcoord_state
[STATE_LENGTH
] =
1228 { STATE_INTERNAL
, STATE_CURRENT_ATTRIB
, VERT_ATTRIB_TEX0
};
1229 static const gl_state_index16 scale_state
[STATE_LENGTH
] =
1230 { STATE_INTERNAL
, STATE_PT_SCALE
};
1231 static const gl_state_index16 bias_state
[STATE_LENGTH
] =
1232 { STATE_INTERNAL
, STATE_PT_BIAS
};
1233 static const gl_state_index16 alpha_ref_state
[STATE_LENGTH
] =
1234 { STATE_INTERNAL
, STATE_ALPHA_REF
};
1239 if (stfp
->state
.type
== PIPE_SHADER_IR_NIR
) {
1240 bool finalize
= false;
1242 state
.type
= PIPE_SHADER_IR_NIR
;
1243 state
.ir
.nir
= get_nir_shader(st
, stfp
);
1245 if (key
->clamp_color
) {
1246 NIR_PASS_V(state
.ir
.nir
, nir_lower_clamp_color_outputs
);
1250 if (key
->lower_flatshade
) {
1251 NIR_PASS_V(state
.ir
.nir
, nir_lower_flatshade
);
1255 if (key
->lower_alpha_func
!= COMPARE_FUNC_NEVER
) {
1256 _mesa_add_state_reference(params
, alpha_ref_state
);
1257 NIR_PASS_V(state
.ir
.nir
, nir_lower_alpha_test
, key
->lower_alpha_func
,
1258 false, alpha_ref_state
);
1262 if (key
->lower_two_sided_color
) {
1263 bool face_sysval
= st
->ctx
->Const
.GLSLFrontFacingIsSysVal
;
1264 NIR_PASS_V(state
.ir
.nir
, nir_lower_two_sided_color
, face_sysval
);
1268 if (key
->persample_shading
) {
1269 nir_shader
*shader
= state
.ir
.nir
;
1270 nir_foreach_shader_in_variable(var
, shader
)
1271 var
->data
.sample
= true;
1275 assert(!(key
->bitmap
&& key
->drawpixels
));
1279 nir_lower_bitmap_options options
= {0};
1281 variant
->bitmap_sampler
= ffs(~stfp
->Base
.SamplersUsed
) - 1;
1282 options
.sampler
= variant
->bitmap_sampler
;
1283 options
.swizzle_xxxx
= st
->bitmap
.tex_format
== PIPE_FORMAT_R8_UNORM
;
1285 NIR_PASS_V(state
.ir
.nir
, nir_lower_bitmap
, &options
);
1289 /* glDrawPixels (color only) */
1290 if (key
->drawpixels
) {
1291 nir_lower_drawpixels_options options
= {{0}};
1292 unsigned samplers_used
= stfp
->Base
.SamplersUsed
;
1294 /* Find the first unused slot. */
1295 variant
->drawpix_sampler
= ffs(~samplers_used
) - 1;
1296 options
.drawpix_sampler
= variant
->drawpix_sampler
;
1297 samplers_used
|= (1 << variant
->drawpix_sampler
);
1299 options
.pixel_maps
= key
->pixelMaps
;
1300 if (key
->pixelMaps
) {
1301 variant
->pixelmap_sampler
= ffs(~samplers_used
) - 1;
1302 options
.pixelmap_sampler
= variant
->pixelmap_sampler
;
1305 options
.scale_and_bias
= key
->scaleAndBias
;
1306 if (key
->scaleAndBias
) {
1307 _mesa_add_state_reference(params
, scale_state
);
1308 memcpy(options
.scale_state_tokens
, scale_state
,
1309 sizeof(options
.scale_state_tokens
));
1310 _mesa_add_state_reference(params
, bias_state
);
1311 memcpy(options
.bias_state_tokens
, bias_state
,
1312 sizeof(options
.bias_state_tokens
));
1315 _mesa_add_state_reference(params
, texcoord_state
);
1316 memcpy(options
.texcoord_state_tokens
, texcoord_state
,
1317 sizeof(options
.texcoord_state_tokens
));
1319 NIR_PASS_V(state
.ir
.nir
, nir_lower_drawpixels
, &options
);
1323 if (unlikely(key
->external
.lower_nv12
|| key
->external
.lower_iyuv
||
1324 key
->external
.lower_xy_uxvx
|| key
->external
.lower_yx_xuxv
||
1325 key
->external
.lower_ayuv
|| key
->external
.lower_xyuv
)) {
1327 st_nir_lower_samplers(pipe
->screen
, state
.ir
.nir
,
1328 stfp
->shader_program
, &stfp
->Base
);
1330 nir_lower_tex_options options
= {0};
1331 options
.lower_y_uv_external
= key
->external
.lower_nv12
;
1332 options
.lower_y_u_v_external
= key
->external
.lower_iyuv
;
1333 options
.lower_xy_uxvx_external
= key
->external
.lower_xy_uxvx
;
1334 options
.lower_yx_xuxv_external
= key
->external
.lower_yx_xuxv
;
1335 options
.lower_ayuv_external
= key
->external
.lower_ayuv
;
1336 options
.lower_xyuv_external
= key
->external
.lower_xyuv
;
1337 NIR_PASS_V(state
.ir
.nir
, nir_lower_tex
, &options
);
1341 if (finalize
|| !st
->allow_st_finalize_nir_twice
) {
1342 st_finalize_nir(st
, &stfp
->Base
, stfp
->shader_program
, state
.ir
.nir
,
1346 /* This pass needs to happen *after* nir_lower_sampler */
1347 if (unlikely(key
->external
.lower_nv12
|| key
->external
.lower_iyuv
||
1348 key
->external
.lower_xy_uxvx
|| key
->external
.lower_yx_xuxv
||
1349 key
->external
.lower_ayuv
|| key
->external
.lower_xyuv
)) {
1350 NIR_PASS_V(state
.ir
.nir
, st_nir_lower_tex_src_plane
,
1351 ~stfp
->Base
.SamplersUsed
,
1352 key
->external
.lower_nv12
|| key
->external
.lower_xy_uxvx
||
1353 key
->external
.lower_yx_xuxv
,
1354 key
->external
.lower_iyuv
);
1358 if (finalize
|| !st
->allow_st_finalize_nir_twice
) {
1359 /* Some of the lowering above may have introduced new varyings */
1360 nir_shader_gather_info(state
.ir
.nir
,
1361 nir_shader_get_entrypoint(state
.ir
.nir
));
1363 struct pipe_screen
*screen
= pipe
->screen
;
1364 if (screen
->finalize_nir
)
1365 screen
->finalize_nir(screen
, state
.ir
.nir
, false);
1368 if (ST_DEBUG
& DEBUG_PRINT_IR
)
1369 nir_print_shader(state
.ir
.nir
, stderr
);
1371 variant
->base
.driver_shader
= pipe
->create_fs_state(pipe
, &state
);
1372 variant
->key
= *key
;
1377 state
.tokens
= stfp
->state
.tokens
;
1379 assert(!(key
->bitmap
&& key
->drawpixels
));
1381 /* Fix texture targets and add fog for ATI_fs */
1383 const struct tgsi_token
*tokens
= st_fixup_atifs(state
.tokens
, key
);
1386 state
.tokens
= tokens
;
1388 fprintf(stderr
, "mesa: cannot post-process ATI_fs\n");
1391 /* Emulate features. */
1392 if (key
->clamp_color
|| key
->persample_shading
) {
1393 const struct tgsi_token
*tokens
;
1395 (key
->clamp_color
? TGSI_EMU_CLAMP_COLOR_OUTPUTS
: 0) |
1396 (key
->persample_shading
? TGSI_EMU_FORCE_PERSAMPLE_INTERP
: 0);
1398 tokens
= tgsi_emulate(state
.tokens
, flags
);
1401 if (state
.tokens
!= stfp
->state
.tokens
)
1402 tgsi_free_tokens(state
.tokens
);
1403 state
.tokens
= tokens
;
1405 fprintf(stderr
, "mesa: cannot emulate deprecated features\n");
1410 const struct tgsi_token
*tokens
;
1412 variant
->bitmap_sampler
= ffs(~stfp
->Base
.SamplersUsed
) - 1;
1414 tokens
= st_get_bitmap_shader(state
.tokens
,
1415 st
->internal_target
,
1416 variant
->bitmap_sampler
,
1417 st
->needs_texcoord_semantic
,
1418 st
->bitmap
.tex_format
==
1419 PIPE_FORMAT_R8_UNORM
);
1422 if (state
.tokens
!= stfp
->state
.tokens
)
1423 tgsi_free_tokens(state
.tokens
);
1424 state
.tokens
= tokens
;
1426 fprintf(stderr
, "mesa: cannot create a shader for glBitmap\n");
1429 /* glDrawPixels (color only) */
1430 if (key
->drawpixels
) {
1431 const struct tgsi_token
*tokens
;
1432 unsigned scale_const
= 0, bias_const
= 0, texcoord_const
= 0;
1434 /* Find the first unused slot. */
1435 variant
->drawpix_sampler
= ffs(~stfp
->Base
.SamplersUsed
) - 1;
1437 if (key
->pixelMaps
) {
1438 unsigned samplers_used
= stfp
->Base
.SamplersUsed
|
1439 (1 << variant
->drawpix_sampler
);
1441 variant
->pixelmap_sampler
= ffs(~samplers_used
) - 1;
1444 if (key
->scaleAndBias
) {
1445 scale_const
= _mesa_add_state_reference(params
, scale_state
);
1446 bias_const
= _mesa_add_state_reference(params
, bias_state
);
1449 texcoord_const
= _mesa_add_state_reference(params
, texcoord_state
);
1451 tokens
= st_get_drawpix_shader(state
.tokens
,
1452 st
->needs_texcoord_semantic
,
1453 key
->scaleAndBias
, scale_const
,
1454 bias_const
, key
->pixelMaps
,
1455 variant
->drawpix_sampler
,
1456 variant
->pixelmap_sampler
,
1457 texcoord_const
, st
->internal_target
);
1460 if (state
.tokens
!= stfp
->state
.tokens
)
1461 tgsi_free_tokens(state
.tokens
);
1462 state
.tokens
= tokens
;
1464 fprintf(stderr
, "mesa: cannot create a shader for glDrawPixels\n");
1467 if (unlikely(key
->external
.lower_nv12
|| key
->external
.lower_iyuv
||
1468 key
->external
.lower_xy_uxvx
|| key
->external
.lower_yx_xuxv
)) {
1469 const struct tgsi_token
*tokens
;
1471 /* samplers inserted would conflict, but this should be unpossible: */
1472 assert(!(key
->bitmap
|| key
->drawpixels
));
1474 tokens
= st_tgsi_lower_yuv(state
.tokens
,
1475 ~stfp
->Base
.SamplersUsed
,
1476 key
->external
.lower_nv12
||
1477 key
->external
.lower_xy_uxvx
||
1478 key
->external
.lower_yx_xuxv
,
1479 key
->external
.lower_iyuv
);
1481 if (state
.tokens
!= stfp
->state
.tokens
)
1482 tgsi_free_tokens(state
.tokens
);
1483 state
.tokens
= tokens
;
1485 fprintf(stderr
, "mesa: cannot create a shader for samplerExternalOES\n");
1489 if (key
->lower_depth_clamp
) {
1490 unsigned depth_range_const
= _mesa_add_state_reference(params
, depth_range_state
);
1492 const struct tgsi_token
*tokens
;
1493 tokens
= st_tgsi_lower_depth_clamp_fs(state
.tokens
, depth_range_const
);
1494 if (state
.tokens
!= stfp
->state
.tokens
)
1495 tgsi_free_tokens(state
.tokens
);
1496 state
.tokens
= tokens
;
1499 if (ST_DEBUG
& DEBUG_PRINT_IR
)
1500 tgsi_dump(state
.tokens
, 0);
1502 /* fill in variant */
1503 variant
->base
.driver_shader
= pipe
->create_fs_state(pipe
, &state
);
1504 variant
->key
= *key
;
1506 if (state
.tokens
!= stfp
->state
.tokens
)
1507 tgsi_free_tokens(state
.tokens
);
1512 * Translate fragment program if needed.
1514 struct st_fp_variant
*
1515 st_get_fp_variant(struct st_context
*st
,
1516 struct st_program
*stfp
,
1517 const struct st_fp_variant_key
*key
)
1519 struct st_fp_variant
*fpv
;
1521 /* Search for existing variant */
1522 for (fpv
= st_fp_variant(stfp
->variants
); fpv
;
1523 fpv
= st_fp_variant(fpv
->base
.next
)) {
1524 if (memcmp(&fpv
->key
, key
, sizeof(*key
)) == 0) {
1531 fpv
= st_create_fp_variant(st
, stfp
, key
);
1533 fpv
->base
.st
= key
->st
;
1535 if (key
->bitmap
|| key
->drawpixels
) {
1536 /* Regular variants should always come before the
1537 * bitmap & drawpixels variants, (unless there
1538 * are no regular variants) so that
1539 * st_update_fp can take a fast path when
1540 * shader_has_one_variant is set.
1542 if (!stfp
->variants
) {
1543 stfp
->variants
= &fpv
->base
;
1545 /* insert into list after the first one */
1546 fpv
->base
.next
= stfp
->variants
->next
;
1547 stfp
->variants
->next
= &fpv
->base
;
1550 /* insert into list */
1551 fpv
->base
.next
= stfp
->variants
;
1552 stfp
->variants
= &fpv
->base
;
1561 * Translate a program. This is common code for geometry and tessellation
1565 st_translate_common_program(struct st_context
*st
,
1566 struct st_program
*stp
)
1568 struct gl_program
*prog
= &stp
->Base
;
1569 enum pipe_shader_type stage
=
1570 pipe_shader_type_from_mesa(stp
->Base
.info
.stage
);
1571 struct ureg_program
*ureg
= ureg_create_with_screen(stage
, st
->pipe
->screen
);
1576 ureg_setup_shader_info(ureg
, &stp
->Base
.info
);
1578 ubyte inputSlotToAttr
[VARYING_SLOT_TESS_MAX
];
1579 ubyte inputMapping
[VARYING_SLOT_TESS_MAX
];
1580 ubyte outputMapping
[VARYING_SLOT_TESS_MAX
];
1583 ubyte input_semantic_name
[PIPE_MAX_SHADER_INPUTS
];
1584 ubyte input_semantic_index
[PIPE_MAX_SHADER_INPUTS
];
1585 uint num_inputs
= 0;
1587 ubyte output_semantic_name
[PIPE_MAX_SHADER_OUTPUTS
];
1588 ubyte output_semantic_index
[PIPE_MAX_SHADER_OUTPUTS
];
1589 uint num_outputs
= 0;
1593 memset(inputSlotToAttr
, 0, sizeof(inputSlotToAttr
));
1594 memset(inputMapping
, 0, sizeof(inputMapping
));
1595 memset(outputMapping
, 0, sizeof(outputMapping
));
1596 memset(&stp
->state
, 0, sizeof(stp
->state
));
1599 * Convert Mesa program inputs to TGSI input register semantics.
1601 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
1602 if ((prog
->info
.inputs_read
& BITFIELD64_BIT(attr
)) == 0)
1605 unsigned slot
= num_inputs
++;
1607 inputMapping
[attr
] = slot
;
1608 inputSlotToAttr
[slot
] = attr
;
1610 unsigned semantic_name
, semantic_index
;
1611 tgsi_get_gl_varying_semantic(attr
, st
->needs_texcoord_semantic
,
1612 &semantic_name
, &semantic_index
);
1613 input_semantic_name
[slot
] = semantic_name
;
1614 input_semantic_index
[slot
] = semantic_index
;
1617 /* Also add patch inputs. */
1618 for (attr
= 0; attr
< 32; attr
++) {
1619 if (prog
->info
.patch_inputs_read
& (1u << attr
)) {
1620 GLuint slot
= num_inputs
++;
1621 GLuint patch_attr
= VARYING_SLOT_PATCH0
+ attr
;
1623 inputMapping
[patch_attr
] = slot
;
1624 inputSlotToAttr
[slot
] = patch_attr
;
1625 input_semantic_name
[slot
] = TGSI_SEMANTIC_PATCH
;
1626 input_semantic_index
[slot
] = attr
;
1630 /* initialize output semantics to defaults */
1631 for (i
= 0; i
< PIPE_MAX_SHADER_OUTPUTS
; i
++) {
1632 output_semantic_name
[i
] = TGSI_SEMANTIC_GENERIC
;
1633 output_semantic_index
[i
] = 0;
1637 * Determine number of outputs, the (default) output register
1638 * mapping and the semantic information for each output.
1640 for (attr
= 0; attr
< VARYING_SLOT_MAX
; attr
++) {
1641 if (prog
->info
.outputs_written
& BITFIELD64_BIT(attr
)) {
1642 GLuint slot
= num_outputs
++;
1644 outputMapping
[attr
] = slot
;
1646 unsigned semantic_name
, semantic_index
;
1647 tgsi_get_gl_varying_semantic(attr
, st
->needs_texcoord_semantic
,
1648 &semantic_name
, &semantic_index
);
1649 output_semantic_name
[slot
] = semantic_name
;
1650 output_semantic_index
[slot
] = semantic_index
;
1654 /* Also add patch outputs. */
1655 for (attr
= 0; attr
< 32; attr
++) {
1656 if (prog
->info
.patch_outputs_written
& (1u << attr
)) {
1657 GLuint slot
= num_outputs
++;
1658 GLuint patch_attr
= VARYING_SLOT_PATCH0
+ attr
;
1660 outputMapping
[patch_attr
] = slot
;
1661 output_semantic_name
[slot
] = TGSI_SEMANTIC_PATCH
;
1662 output_semantic_index
[slot
] = attr
;
1666 st_translate_program(st
->ctx
,
1675 input_semantic_name
,
1676 input_semantic_index
,
1681 output_semantic_name
,
1682 output_semantic_index
);
1684 stp
->state
.tokens
= ureg_get_tokens(ureg
, NULL
);
1688 st_translate_stream_output_info(prog
);
1690 st_store_ir_in_disk_cache(st
, prog
, false);
1692 if (ST_DEBUG
& DEBUG_PRINT_IR
&& ST_DEBUG
& DEBUG_MESA
)
1693 _mesa_print_program(prog
);
1695 free_glsl_to_tgsi_visitor(stp
->glsl_to_tgsi
);
1696 stp
->glsl_to_tgsi
= NULL
;
1702 * Get/create a basic program variant.
1705 st_get_common_variant(struct st_context
*st
,
1706 struct st_program
*prog
,
1707 const struct st_common_variant_key
*key
)
1709 struct pipe_context
*pipe
= st
->pipe
;
1710 struct st_variant
*v
;
1711 struct pipe_shader_state state
= {0};
1712 struct gl_program_parameter_list
*params
= prog
->Base
.Parameters
;
1714 /* Search for existing variant */
1715 for (v
= prog
->variants
; v
; v
= v
->next
) {
1716 if (memcmp(&st_common_variant(v
)->key
, key
, sizeof(*key
)) == 0)
1722 v
= (struct st_variant
*)CALLOC_STRUCT(st_common_variant
);
1724 if (prog
->state
.type
== PIPE_SHADER_IR_NIR
) {
1725 bool finalize
= false;
1727 state
.type
= PIPE_SHADER_IR_NIR
;
1728 state
.ir
.nir
= get_nir_shader(st
, prog
);
1730 if (key
->clamp_color
) {
1731 NIR_PASS_V(state
.ir
.nir
, nir_lower_clamp_color_outputs
);
1735 if (key
->lower_ucp
) {
1736 lower_ucp(st
, state
.ir
.nir
, key
->lower_ucp
, params
);
1740 state
.stream_output
= prog
->state
.stream_output
;
1742 if (finalize
|| !st
->allow_st_finalize_nir_twice
) {
1743 st_finalize_nir(st
, &prog
->Base
, prog
->shader_program
,
1744 state
.ir
.nir
, true);
1747 if (ST_DEBUG
& DEBUG_PRINT_IR
)
1748 nir_print_shader(state
.ir
.nir
, stderr
);
1750 if (key
->lower_depth_clamp
) {
1751 struct gl_program_parameter_list
*params
= prog
->Base
.Parameters
;
1753 unsigned depth_range_const
=
1754 _mesa_add_state_reference(params
, depth_range_state
);
1756 const struct tgsi_token
*tokens
;
1758 st_tgsi_lower_depth_clamp(prog
->state
.tokens
,
1760 key
->clip_negative_one_to_one
);
1762 if (tokens
!= prog
->state
.tokens
)
1763 tgsi_free_tokens(prog
->state
.tokens
);
1765 prog
->state
.tokens
= tokens
;
1767 state
= prog
->state
;
1769 if (ST_DEBUG
& DEBUG_PRINT_IR
)
1770 tgsi_dump(state
.tokens
, 0);
1772 /* fill in new variant */
1773 switch (prog
->Base
.info
.stage
) {
1774 case MESA_SHADER_TESS_CTRL
:
1775 v
->driver_shader
= pipe
->create_tcs_state(pipe
, &state
);
1777 case MESA_SHADER_TESS_EVAL
:
1778 v
->driver_shader
= pipe
->create_tes_state(pipe
, &state
);
1780 case MESA_SHADER_GEOMETRY
:
1781 v
->driver_shader
= pipe
->create_gs_state(pipe
, &state
);
1783 case MESA_SHADER_COMPUTE
: {
1784 struct pipe_compute_state cs
= {0};
1785 cs
.ir_type
= state
.type
;
1786 cs
.req_local_mem
= prog
->Base
.info
.cs
.shared_size
;
1788 if (state
.type
== PIPE_SHADER_IR_NIR
)
1789 cs
.prog
= state
.ir
.nir
;
1791 cs
.prog
= state
.tokens
;
1793 v
->driver_shader
= pipe
->create_compute_state(pipe
, &cs
);
1797 assert(!"unhandled shader type");
1802 st_common_variant(v
)->key
= *key
;
1805 /* insert into list */
1806 v
->next
= prog
->variants
;
1816 * Vert/Geom/Frag programs have per-context variants. Free all the
1817 * variants attached to the given program which match the given context.
1820 destroy_program_variants(struct st_context
*st
, struct gl_program
*target
)
1822 if (!target
|| target
== &_mesa_DummyProgram
)
1825 struct st_program
*p
= st_program(target
);
1826 struct st_variant
*v
, **prevPtr
= &p
->variants
;
1827 bool unbound
= false;
1829 for (v
= p
->variants
; v
; ) {
1830 struct st_variant
*next
= v
->next
;
1833 st_unbind_program(st
, p
);
1837 /* unlink from list */
1839 /* destroy this variant */
1840 delete_variant(st
, v
, target
->Target
);
1851 * Callback for _mesa_HashWalk. Free all the shader's program variants
1852 * which match the given context.
1855 destroy_shader_program_variants_cb(GLuint key
, void *data
, void *userData
)
1857 struct st_context
*st
= (struct st_context
*) userData
;
1858 struct gl_shader
*shader
= (struct gl_shader
*) data
;
1860 switch (shader
->Type
) {
1861 case GL_SHADER_PROGRAM_MESA
:
1863 struct gl_shader_program
*shProg
= (struct gl_shader_program
*) data
;
1866 for (i
= 0; i
< ARRAY_SIZE(shProg
->_LinkedShaders
); i
++) {
1867 if (shProg
->_LinkedShaders
[i
])
1868 destroy_program_variants(st
, shProg
->_LinkedShaders
[i
]->Program
);
1872 case GL_VERTEX_SHADER
:
1873 case GL_FRAGMENT_SHADER
:
1874 case GL_GEOMETRY_SHADER
:
1875 case GL_TESS_CONTROL_SHADER
:
1876 case GL_TESS_EVALUATION_SHADER
:
1877 case GL_COMPUTE_SHADER
:
1886 * Callback for _mesa_HashWalk. Free all the program variants which match
1887 * the given context.
1890 destroy_program_variants_cb(GLuint key
, void *data
, void *userData
)
1892 struct st_context
*st
= (struct st_context
*) userData
;
1893 struct gl_program
*program
= (struct gl_program
*) data
;
1894 destroy_program_variants(st
, program
);
1899 * Walk over all shaders and programs to delete any variants which
1900 * belong to the given context.
1901 * This is called during context tear-down.
1904 st_destroy_program_variants(struct st_context
*st
)
1906 /* If shaders can be shared with other contexts, the last context will
1907 * call DeleteProgram on all shaders, releasing everything.
1909 if (st
->has_shareable_shaders
)
1912 /* ARB vert/frag program */
1913 _mesa_HashWalk(st
->ctx
->Shared
->Programs
,
1914 destroy_program_variants_cb
, st
);
1916 /* GLSL vert/frag/geom shaders */
1917 _mesa_HashWalk(st
->ctx
->Shared
->ShaderObjects
,
1918 destroy_shader_program_variants_cb
, st
);
1923 * Compile one shader variant.
1926 st_precompile_shader_variant(struct st_context
*st
,
1927 struct gl_program
*prog
)
1929 switch (prog
->Target
) {
1930 case GL_VERTEX_PROGRAM_ARB
: {
1931 struct st_program
*p
= (struct st_program
*)prog
;
1932 struct st_common_variant_key key
;
1934 memset(&key
, 0, sizeof(key
));
1936 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
1937 st_get_vp_variant(st
, p
, &key
);
1941 case GL_FRAGMENT_PROGRAM_ARB
: {
1942 struct st_program
*p
= (struct st_program
*)prog
;
1943 struct st_fp_variant_key key
;
1945 memset(&key
, 0, sizeof(key
));
1947 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
1948 st_get_fp_variant(st
, p
, &key
);
1952 case GL_TESS_CONTROL_PROGRAM_NV
:
1953 case GL_TESS_EVALUATION_PROGRAM_NV
:
1954 case GL_GEOMETRY_PROGRAM_NV
:
1955 case GL_COMPUTE_PROGRAM_NV
: {
1956 struct st_program
*p
= st_program(prog
);
1957 struct st_common_variant_key key
;
1959 memset(&key
, 0, sizeof(key
));
1961 key
.st
= st
->has_shareable_shaders
? NULL
: st
;
1962 st_get_common_variant(st
, p
, &key
);
1972 st_serialize_nir(struct st_program
*stp
)
1974 if (!stp
->serialized_nir
) {
1979 nir_serialize(&blob
, stp
->Base
.nir
, false);
1980 blob_finish_get_buffer(&blob
, &stp
->serialized_nir
, &size
);
1981 stp
->serialized_nir_size
= size
;
1986 st_finalize_program(struct st_context
*st
, struct gl_program
*prog
)
1988 if (st
->current_program
[prog
->info
.stage
] == prog
) {
1989 if (prog
->info
.stage
== MESA_SHADER_VERTEX
)
1990 st
->dirty
|= ST_NEW_VERTEX_PROGRAM(st
, (struct st_program
*)prog
);
1992 st
->dirty
|= ((struct st_program
*)prog
)->affected_states
;
1996 nir_sweep(prog
->nir
);
1998 /* This is only needed for ARB_vp/fp programs and when the disk cache
1999 * is disabled. If the disk cache is enabled, GLSL programs are
2000 * serialized in write_nir_to_cache.
2002 st_serialize_nir(st_program(prog
));
2005 /* Create Gallium shaders now instead of on demand. */
2006 if (ST_DEBUG
& DEBUG_PRECOMPILE
||
2007 st
->shader_has_one_variant
[prog
->info
.stage
])
2008 st_precompile_shader_variant(st
, prog
);