2 * Copyright 2012 Advanced Micro Devices, Inc.
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
24 /* The compiler middle-end architecture: Explaining (non-)monolithic shaders
25 * -------------------------------------------------------------------------
27 * Typically, there is one-to-one correspondence between API and HW shaders,
28 * that is, for every API shader, there is exactly one shader binary in
31 * The problem with that is that we also have to emulate some API states
32 * (e.g. alpha-test, and many others) in shaders too. The two obvious ways
33 * to deal with it are:
34 * - each shader has multiple variants for each combination of emulated states,
35 * and the variants are compiled on demand, possibly relying on a shader
36 * cache for good performance
37 * - patch shaders at the binary level
39 * This driver uses something completely different. The emulated states are
40 * usually implemented at the beginning or end of shaders. Therefore, we can
41 * split the shader into 3 parts:
42 * - prolog part (shader code dependent on states)
43 * - main part (the API shader)
44 * - epilog part (shader code dependent on states)
46 * Each part is compiled as a separate shader and the final binaries are
47 * concatenated. This type of shader is called non-monolithic, because it
48 * consists of multiple independent binaries. Creating a new shader variant
49 * is therefore only a concatenation of shader parts (binaries) and doesn't
50 * involve any compilation. The main shader parts are the only parts that are
51 * compiled when applications create shader objects. The prolog and epilog
52 * parts are compiled on the first use and saved, so that their binaries can
53 * be reused by many other shaders.
55 * One of the roles of the prolog part is to compute vertex buffer addresses
56 * for vertex shaders. A few of the roles of the epilog part are color buffer
57 * format conversions in pixel shaders that we have to do manually, and write
58 * tessellation factors in tessellation control shaders. The prolog and epilog
59 * have many other important responsibilities in various shader stages.
60 * They don't just "emulate legacy stuff".
62 * Monolithic shaders are shaders where the parts are combined before LLVM
63 * compilation, and the whole thing is compiled and optimized as one unit with
64 * one binary on the output. The result is the same as the non-monolithic
65 * shader, but the final code can be better, because LLVM can optimize across
66 * all shader parts. Monolithic shaders aren't usually used except for these
69 * 1) Some rarely-used states require modification of the main shader part
70 * itself, and in such cases, only the monolithic shader variant is
71 * compiled, and that's always done on the first use.
73 * 2) When we do cross-stage optimizations for separate shader objects and
74 * e.g. eliminate unused shader varyings, the resulting optimized shader
75 * variants are always compiled as monolithic shaders, and always
76 * asynchronously (i.e. not stalling ongoing rendering). We call them
77 * "optimized monolithic" shaders. The important property here is that
78 * the non-monolithic unoptimized shader variant is always available for use
79 * when the asynchronous compilation of the optimized shader is not done
82 * Starting with GFX9 chips, some shader stages are merged, and the number of
83 * shader parts per shader increased. The complete new list of shader parts is:
84 * - 1st shader: prolog part
85 * - 1st shader: main part
86 * - 2nd shader: prolog part
87 * - 2nd shader: main part
88 * - 2nd shader: epilog part
91 /* How linking shader inputs and outputs between vertex, tessellation, and
92 * geometry shaders works.
94 * Inputs and outputs between shaders are stored in a buffer. This buffer
95 * lives in LDS (typical case for tessellation), but it can also live
96 * in memory (ESGS). Each input or output has a fixed location within a vertex.
97 * The highest used input or output determines the stride between vertices.
99 * Since GS and tessellation are only possible in the OpenGL core profile,
100 * only these semantics are valid for per-vertex data:
107 * CULLDIST0..1 (not implemented)
110 * For example, a shader only writing GENERIC0 has the output stride of 5.
112 * Only these semantics are valid for per-patch data:
120 * That's how independent shaders agree on input and output locations.
121 * The si_shader_io_get_unique_index function assigns the locations.
123 * For tessellation, other required information for calculating the input and
124 * output addresses like the vertex stride, the patch stride, and the offsets
125 * where per-vertex and per-patch data start, is passed to the shader via
126 * user data SGPRs. The offsets and strides are calculated at draw time and
127 * aren't available at compile time.
133 #include <llvm-c/Core.h> /* LLVMModuleRef */
134 #include <llvm-c/TargetMachine.h>
135 #include "tgsi/tgsi_scan.h"
136 #include "util/u_queue.h"
138 #include "ac_binary.h"
139 #include "ac_llvm_build.h"
140 #include "si_state.h"
144 #define SI_MAX_VS_OUTPUTS 40
146 /* Shader IO unique indices are supported for TGSI_SEMANTIC_GENERIC with an
147 * index smaller than this.
149 #define SI_MAX_IO_GENERIC 46
151 /* SGPR user data indices */
153 SI_SGPR_RW_BUFFERS
, /* rings (& stream-out, VS only) */
154 #if !HAVE_32BIT_POINTERS
155 SI_SGPR_RW_BUFFERS_HI
,
157 SI_SGPR_BINDLESS_SAMPLERS_AND_IMAGES
,
158 #if !HAVE_32BIT_POINTERS
159 SI_SGPR_BINDLESS_SAMPLERS_AND_IMAGES_HI
,
161 SI_SGPR_CONST_AND_SHADER_BUFFERS
, /* or just a constant buffer 0 pointer */
162 #if !HAVE_32BIT_POINTERS
163 SI_SGPR_CONST_AND_SHADER_BUFFERS_HI
,
165 SI_SGPR_SAMPLERS_AND_IMAGES
,
166 #if !HAVE_32BIT_POINTERS
167 SI_SGPR_SAMPLERS_AND_IMAGES_HI
,
169 SI_NUM_RESOURCE_SGPRS
,
171 /* all VS variants */
172 SI_SGPR_BASE_VERTEX
= SI_NUM_RESOURCE_SGPRS
,
173 SI_SGPR_START_INSTANCE
,
175 SI_SGPR_VS_STATE_BITS
,
178 SI_SGPR_VS_BLIT_DATA
= SI_SGPR_CONST_AND_SHADER_BUFFERS
,
181 SI_SGPR_TES_OFFCHIP_LAYOUT
= SI_NUM_RESOURCE_SGPRS
,
182 SI_SGPR_TES_OFFCHIP_ADDR
,
183 SI_TES_NUM_USER_SGPR
,
185 /* GFX6-8: TCS only */
186 GFX6_SGPR_TCS_OFFCHIP_LAYOUT
= SI_NUM_RESOURCE_SGPRS
,
187 GFX6_SGPR_TCS_OUT_OFFSETS
,
188 GFX6_SGPR_TCS_OUT_LAYOUT
,
189 GFX6_SGPR_TCS_IN_LAYOUT
,
190 GFX6_TCS_NUM_USER_SGPR
,
192 /* GFX9: Merged shaders. */
193 #if HAVE_32BIT_POINTERS
194 /* 2ND_CONST_AND_SHADER_BUFFERS is set in USER_DATA_ADDR_LO (SGPR0). */
195 /* 2ND_SAMPLERS_AND_IMAGES is set in USER_DATA_ADDR_HI (SGPR1). */
196 GFX9_MERGED_NUM_USER_SGPR
= SI_VS_NUM_USER_SGPR
,
198 /* 2ND_CONST_AND_SHADER_BUFFERS is set in USER_DATA_ADDR_LO/HI (SGPR[0:1]). */
199 GFX9_SGPR_2ND_SAMPLERS_AND_IMAGES
= SI_VS_NUM_USER_SGPR
,
200 GFX9_SGPR_2ND_SAMPLERS_AND_IMAGES_HI
,
201 GFX9_MERGED_NUM_USER_SGPR
,
204 /* GFX9: Merged LS-HS (VS-TCS) only. */
205 GFX9_SGPR_TCS_OFFCHIP_LAYOUT
= GFX9_MERGED_NUM_USER_SGPR
,
206 GFX9_SGPR_TCS_OUT_OFFSETS
,
207 GFX9_SGPR_TCS_OUT_LAYOUT
,
208 #if !HAVE_32BIT_POINTERS
209 GFX9_SGPR_align_for_vb_pointer
,
211 GFX9_TCS_NUM_USER_SGPR
,
214 GFX6_GS_NUM_USER_SGPR
= SI_NUM_RESOURCE_SGPRS
,
215 GFX9_GS_NUM_USER_SGPR
= GFX9_MERGED_NUM_USER_SGPR
,
216 SI_GSCOPY_NUM_USER_SGPR
= SI_SGPR_RW_BUFFERS
+ (HAVE_32BIT_POINTERS
? 1 : 2),
219 SI_SGPR_ALPHA_REF
= SI_NUM_RESOURCE_SGPRS
,
223 /* LLVM function parameter indices */
225 SI_NUM_RESOURCE_PARAMS
= 4,
227 /* PS only parameters */
228 SI_PARAM_ALPHA_REF
= SI_NUM_RESOURCE_PARAMS
,
230 SI_PARAM_PERSP_SAMPLE
,
231 SI_PARAM_PERSP_CENTER
,
232 SI_PARAM_PERSP_CENTROID
,
233 SI_PARAM_PERSP_PULL_MODEL
,
234 SI_PARAM_LINEAR_SAMPLE
,
235 SI_PARAM_LINEAR_CENTER
,
236 SI_PARAM_LINEAR_CENTROID
,
237 SI_PARAM_LINE_STIPPLE_TEX
,
238 SI_PARAM_POS_X_FLOAT
,
239 SI_PARAM_POS_Y_FLOAT
,
240 SI_PARAM_POS_Z_FLOAT
,
241 SI_PARAM_POS_W_FLOAT
,
244 SI_PARAM_SAMPLE_COVERAGE
,
245 SI_PARAM_POS_FIXED_PT
,
247 SI_NUM_PARAMS
= SI_PARAM_POS_FIXED_PT
+ 9, /* +8 for COLOR[0..1] */
250 /* Fields of driver-defined VS state SGPR. */
251 /* Clamp vertex color output (only used in VS as VS). */
252 #define S_VS_STATE_CLAMP_VERTEX_COLOR(x) (((unsigned)(x) & 0x1) << 0)
253 #define C_VS_STATE_CLAMP_VERTEX_COLOR 0xFFFFFFFE
254 #define S_VS_STATE_INDEXED(x) (((unsigned)(x) & 0x1) << 1)
255 #define C_VS_STATE_INDEXED 0xFFFFFFFD
256 #define S_VS_STATE_LS_OUT_PATCH_SIZE(x) (((unsigned)(x) & 0x1FFF) << 8)
257 #define C_VS_STATE_LS_OUT_PATCH_SIZE 0xFFE000FF
258 #define S_VS_STATE_LS_OUT_VERTEX_SIZE(x) (((unsigned)(x) & 0xFF) << 24)
259 #define C_VS_STATE_LS_OUT_VERTEX_SIZE 0x00FFFFFF
261 /* SI-specific system values. */
263 TGSI_SEMANTIC_DEFAULT_TESSOUTER_SI
= TGSI_SEMANTIC_COUNT
,
264 TGSI_SEMANTIC_DEFAULT_TESSINNER_SI
,
268 /* Use a property enum that VS wouldn't use. */
269 TGSI_PROPERTY_VS_BLIT_SGPRS
= TGSI_PROPERTY_FS_COORD_ORIGIN
,
271 /* These represent the number of SGPRs the shader uses. */
272 SI_VS_BLIT_SGPRS_POS
= 3,
273 SI_VS_BLIT_SGPRS_POS_COLOR
= 7,
274 SI_VS_BLIT_SGPRS_POS_TEXCOORD
= 9,
277 /* For VS shader key fix_fetch. */
279 SI_FIX_FETCH_NONE
= 0,
280 SI_FIX_FETCH_A2_SNORM
,
281 SI_FIX_FETCH_A2_SSCALED
,
282 SI_FIX_FETCH_A2_SINT
,
283 SI_FIX_FETCH_RGBA_32_UNORM
,
284 SI_FIX_FETCH_RGBX_32_UNORM
,
285 SI_FIX_FETCH_RGBA_32_SNORM
,
286 SI_FIX_FETCH_RGBX_32_SNORM
,
287 SI_FIX_FETCH_RGBA_32_USCALED
,
288 SI_FIX_FETCH_RGBA_32_SSCALED
,
289 SI_FIX_FETCH_RGBA_32_FIXED
,
290 SI_FIX_FETCH_RGBX_32_FIXED
,
291 SI_FIX_FETCH_RG_64_FLOAT
,
292 SI_FIX_FETCH_RGB_64_FLOAT
,
293 SI_FIX_FETCH_RGBA_64_FLOAT
,
294 SI_FIX_FETCH_RGB_8
, /* A = 1.0 */
295 SI_FIX_FETCH_RGB_8_INT
, /* A = 1 */
297 SI_FIX_FETCH_RGB_16_INT
,
302 /* State of the context creating the shader object. */
303 struct si_compiler_ctx_state
{
304 /* Should only be used by si_init_shader_selector_async and
305 * si_build_shader_variant if thread_index == -1 (non-threaded). */
306 LLVMTargetMachineRef tm
;
308 /* Used if thread_index == -1 or if debug.async is true. */
309 struct pipe_debug_callback debug
;
311 /* Used for creating the log string for gallium/ddebug. */
312 bool is_debug_context
;
315 /* A shader selector is a gallium CSO and contains shader variants and
316 * binaries for one TGSI program. This can be shared by multiple contexts.
318 struct si_shader_selector
{
319 struct pipe_reference reference
;
320 struct si_screen
*screen
;
321 struct util_queue_fence ready
;
322 struct si_compiler_ctx_state compiler_ctx_state
;
325 struct si_shader
*first_variant
; /* immutable after the first variant */
326 struct si_shader
*last_variant
; /* mutable */
328 /* The compiled TGSI shader expecting a prolog and/or epilog (not
329 * uploaded to a buffer).
331 struct si_shader
*main_shader_part
;
332 struct si_shader
*main_shader_part_ls
; /* as_ls is set in the key */
333 struct si_shader
*main_shader_part_es
; /* as_es is set in the key */
335 struct si_shader
*gs_copy_shader
;
337 struct tgsi_token
*tokens
;
338 struct nir_shader
*nir
;
339 struct pipe_stream_output_info so
;
340 struct tgsi_shader_info info
;
341 struct tgsi_tessctrl_info tcs_info
;
343 /* PIPE_SHADER_[VERTEX|FRAGMENT|...] */
345 bool vs_needs_prolog
;
346 bool force_correct_derivs_after_kill
;
347 unsigned pa_cl_vs_out_cntl
;
352 unsigned esgs_itemsize
;
355 unsigned gs_input_verts_per_prim
;
356 unsigned gs_output_prim
;
357 unsigned gs_max_out_vertices
;
358 unsigned gs_num_invocations
;
359 unsigned max_gs_stream
; /* count - 1 */
360 unsigned gsvs_vertex_size
;
361 unsigned max_gsvs_emit_size
;
362 unsigned enabled_streamout_buffer_mask
;
365 unsigned color_attr_index
[2];
366 unsigned db_shader_control
;
367 /* Set 0xf or 0x0 (4 bits) per each written output.
368 * ANDed with spi_shader_col_format.
370 unsigned colors_written_4bit
;
375 uint64_t outputs_written
; /* "get_unique_index" bits */
376 uint32_t patch_outputs_written
; /* "get_unique_index_patch" bits */
378 uint64_t inputs_read
; /* "get_unique_index" bits */
380 /* bitmasks of used descriptor slots */
381 uint32_t active_const_and_shader_buffers
;
382 uint64_t active_samplers_and_images
;
385 /* Valid shader configurations:
387 * API shaders VS | TCS | TES | GS |pass| PS
388 * are compiled as: | | | |thru|
390 * Only VS & PS: VS | | | | | PS
391 * GFX6 - with GS: ES | | | GS | VS | PS
392 * - with tess: LS | HS | VS | | | PS
393 * - with both: LS | HS | ES | GS | VS | PS
394 * GFX9 - with GS: -> | | | GS | VS | PS
395 * - with tess: -> | HS | VS | | | PS
396 * - with both: -> | HS | -> | GS | VS | PS
398 * -> = merged with the next stage
401 /* Use the byte alignment for all following structure members for optimal
402 * shader key memory footprint.
404 #pragma pack(push, 1)
406 /* Common VS bits between the shader key and the prolog key. */
407 struct si_vs_prolog_bits
{
408 /* - If neither "is_one" nor "is_fetched" has a bit set, the instance
410 * - If "is_one" has a bit set, the instance divisor is 1.
411 * - If "is_fetched" has a bit set, the instance divisor will be loaded
412 * from the constant buffer.
414 uint16_t instance_divisor_is_one
; /* bitmask of inputs */
415 uint16_t instance_divisor_is_fetched
; /* bitmask of inputs */
416 unsigned ls_vgpr_fix
:1;
419 /* Common TCS bits between the shader key and the epilog key. */
420 struct si_tcs_epilog_bits
{
421 unsigned prim_mode
:3;
422 unsigned invoc0_tess_factors_are_def
:1;
423 unsigned tes_reads_tess_factors
:1;
426 struct si_gs_prolog_bits
{
427 unsigned tri_strip_adj_fix
:1;
430 /* Common PS bits between the shader key and the prolog key. */
431 struct si_ps_prolog_bits
{
432 unsigned color_two_side
:1;
433 unsigned flatshade_colors
:1;
434 unsigned poly_stipple
:1;
435 unsigned force_persp_sample_interp
:1;
436 unsigned force_linear_sample_interp
:1;
437 unsigned force_persp_center_interp
:1;
438 unsigned force_linear_center_interp
:1;
439 unsigned bc_optimize_for_persp
:1;
440 unsigned bc_optimize_for_linear
:1;
441 unsigned samplemask_log_ps_iter
:3;
444 /* Common PS bits between the shader key and the epilog key. */
445 struct si_ps_epilog_bits
{
446 unsigned spi_shader_col_format
;
447 unsigned color_is_int8
:8;
448 unsigned color_is_int10
:8;
449 unsigned last_cbuf
:3;
450 unsigned alpha_func
:3;
451 unsigned alpha_to_one
:1;
452 unsigned poly_line_smoothing
:1;
453 unsigned clamp_color
:1;
456 union si_shader_part_key
{
458 struct si_vs_prolog_bits states
;
459 unsigned num_input_sgprs
:6;
460 /* For merged stages such as LS-HS, HS input VGPRs are first. */
461 unsigned num_merged_next_stage_vgprs
:3;
462 unsigned last_input
:4;
465 /* Prologs for monolithic shaders shouldn't set EXEC. */
466 unsigned is_monolithic
:1;
469 struct si_tcs_epilog_bits states
;
472 struct si_gs_prolog_bits states
;
473 /* Prologs of monolithic shaders shouldn't set EXEC. */
474 unsigned is_monolithic
:1;
477 struct si_ps_prolog_bits states
;
478 unsigned num_input_sgprs
:6;
479 unsigned num_input_vgprs
:5;
480 /* Color interpolation and two-side color selection. */
481 unsigned colors_read
:8; /* color input components read */
482 unsigned num_interp_inputs
:5; /* BCOLOR is at this location */
483 unsigned face_vgpr_index
:5;
484 unsigned ancillary_vgpr_index
:5;
486 char color_attr_index
[2];
487 char color_interp_vgpr_index
[2]; /* -1 == constant */
490 struct si_ps_epilog_bits states
;
491 unsigned colors_written
:8;
493 unsigned writes_stencil
:1;
494 unsigned writes_samplemask
:1;
498 struct si_shader_key
{
499 /* Prolog and epilog flags. */
502 struct si_vs_prolog_bits prolog
;
505 struct si_vs_prolog_bits ls_prolog
; /* for merged LS-HS */
506 struct si_shader_selector
*ls
; /* for merged LS-HS */
507 struct si_tcs_epilog_bits epilog
;
508 } tcs
; /* tessellation control shader */
510 struct si_vs_prolog_bits vs_prolog
; /* for merged ES-GS */
511 struct si_shader_selector
*es
; /* for merged ES-GS */
512 struct si_gs_prolog_bits prolog
;
515 struct si_ps_prolog_bits prolog
;
516 struct si_ps_epilog_bits epilog
;
520 /* These two are initially set according to the NEXT_SHADER property,
521 * or guessed if the property doesn't seem correct.
523 unsigned as_es
:1; /* export shader, which precedes GS */
524 unsigned as_ls
:1; /* local shader, which precedes TCS */
526 /* Flags for monolithic compilation only. */
528 /* One byte for every input: SI_FIX_FETCH_* enums. */
529 uint8_t vs_fix_fetch
[SI_MAX_ATTRIBS
];
532 uint64_t ff_tcs_inputs_to_copy
; /* for fixed-func TCS */
533 /* When PS needs PrimID and GS is disabled. */
534 unsigned vs_export_prim_id
:1;
536 unsigned interpolate_at_sample_force_center
:1;
541 /* Optimization flags for asynchronous compilation only. */
543 /* For HW VS (it can be VS, TES, GS) */
544 uint64_t kill_outputs
; /* "get_unique_index" bits */
545 unsigned clip_disable
:1;
547 /* For shaders where monolithic variants have better code.
549 * This is a flag that has no effect on code generation,
550 * but forces monolithic shaders to be used as soon as
551 * possible, because it's in the "opt" group.
553 unsigned prefer_mono
:1;
557 /* Restore the pack alignment to default. */
560 struct si_shader_config
{
563 unsigned spilled_sgprs
;
564 unsigned spilled_vgprs
;
565 unsigned private_mem_vgprs
;
567 unsigned max_simd_waves
;
568 unsigned spi_ps_input_ena
;
569 unsigned spi_ps_input_addr
;
571 unsigned scratch_bytes_per_wave
;
576 /* GCN-specific shader info. */
577 struct si_shader_info
{
578 ubyte vs_output_param_offset
[SI_MAX_VS_OUTPUTS
];
579 ubyte num_input_sgprs
;
580 ubyte num_input_vgprs
;
581 signed char face_vgpr_index
;
582 signed char ancillary_vgpr_index
;
583 bool uses_instanceid
;
584 ubyte nr_pos_exports
;
585 ubyte nr_param_exports
;
589 struct si_compiler_ctx_state compiler_ctx_state
;
591 struct si_shader_selector
*selector
;
592 struct si_shader_selector
*previous_stage_sel
; /* for refcounting */
593 struct si_shader
*next_variant
;
595 struct si_shader_part
*prolog
;
596 struct si_shader
*previous_stage
; /* for GFX9 */
597 struct si_shader_part
*prolog2
;
598 struct si_shader_part
*epilog
;
600 struct si_pm4_state
*pm4
;
601 struct r600_resource
*bo
;
602 struct r600_resource
*scratch_bo
;
603 struct si_shader_key key
;
604 struct util_queue_fence ready
;
605 bool compilation_failed
;
608 bool is_binary_shared
;
609 bool is_gs_copy_shader
;
611 /* The following data is all that's needed for binary shaders. */
612 struct ac_shader_binary binary
;
613 struct si_shader_config config
;
614 struct si_shader_info info
;
616 /* Shader key + LLVM IR + disassembly + statistics.
617 * Generated for debug contexts only.
620 size_t shader_log_size
;
623 struct si_shader_part
{
624 struct si_shader_part
*next
;
625 union si_shader_part_key key
;
626 struct ac_shader_binary binary
;
627 struct si_shader_config config
;
632 si_generate_gs_copy_shader(struct si_screen
*sscreen
,
633 LLVMTargetMachineRef tm
,
634 struct si_shader_selector
*gs_selector
,
635 struct pipe_debug_callback
*debug
);
636 int si_compile_tgsi_shader(struct si_screen
*sscreen
,
637 LLVMTargetMachineRef tm
,
638 struct si_shader
*shader
,
640 struct pipe_debug_callback
*debug
);
641 int si_shader_create(struct si_screen
*sscreen
, LLVMTargetMachineRef tm
,
642 struct si_shader
*shader
,
643 struct pipe_debug_callback
*debug
);
644 void si_shader_destroy(struct si_shader
*shader
);
645 unsigned si_shader_io_get_unique_index_patch(unsigned semantic_name
, unsigned index
);
646 unsigned si_shader_io_get_unique_index(unsigned semantic_name
, unsigned index
);
647 int si_shader_binary_upload(struct si_screen
*sscreen
, struct si_shader
*shader
);
648 void si_shader_dump(struct si_screen
*sscreen
, const struct si_shader
*shader
,
649 struct pipe_debug_callback
*debug
, unsigned processor
,
650 FILE *f
, bool check_debug_option
);
651 void si_shader_dump_stats_for_shader_db(const struct si_shader
*shader
,
652 struct pipe_debug_callback
*debug
);
653 void si_multiwave_lds_size_workaround(struct si_screen
*sscreen
,
655 void si_shader_apply_scratch_relocs(struct si_shader
*shader
,
656 uint64_t scratch_va
);
657 void si_shader_binary_read_config(struct ac_shader_binary
*binary
,
658 struct si_shader_config
*conf
,
659 unsigned symbol_offset
);
660 const char *si_get_shader_name(const struct si_shader
*shader
, unsigned processor
);
662 /* si_shader_nir.c */
663 void si_nir_scan_shader(const struct nir_shader
*nir
,
664 struct tgsi_shader_info
*info
);
665 void si_nir_scan_tess_ctrl(const struct nir_shader
*nir
,
666 const struct tgsi_shader_info
*info
,
667 struct tgsi_tessctrl_info
*out
);
668 void si_lower_nir(struct si_shader_selector
*sel
);
670 /* Inline helpers. */
672 /* Return the pointer to the main shader part's pointer. */
673 static inline struct si_shader
**
674 si_get_main_shader_part(struct si_shader_selector
*sel
,
675 struct si_shader_key
*key
)
678 return &sel
->main_shader_part_ls
;
680 return &sel
->main_shader_part_es
;
681 return &sel
->main_shader_part
;
685 si_shader_uses_bindless_samplers(struct si_shader_selector
*selector
)
687 return selector
? selector
->info
.uses_bindless_samplers
: false;
691 si_shader_uses_bindless_images(struct si_shader_selector
*selector
)
693 return selector
? selector
->info
.uses_bindless_images
: false;
696 void si_destroy_shader_selector(struct si_context
*sctx
,
697 struct si_shader_selector
*sel
);
700 si_shader_selector_reference(struct si_context
*sctx
,
701 struct si_shader_selector
**dst
,
702 struct si_shader_selector
*src
)
704 if (pipe_reference(&(*dst
)->reference
, &src
->reference
))
705 si_destroy_shader_selector(sctx
, *dst
);