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_VERTEX_BUFFERS
= SI_NUM_RESOURCE_SGPRS
,
173 #if !HAVE_32BIT_POINTERS
174 SI_SGPR_VERTEX_BUFFERS_HI
,
177 SI_SGPR_START_INSTANCE
,
179 SI_SGPR_VS_STATE_BITS
,
182 SI_SGPR_VS_BLIT_DATA
= SI_SGPR_CONST_AND_SHADER_BUFFERS
,
185 SI_SGPR_TES_OFFCHIP_LAYOUT
= SI_NUM_RESOURCE_SGPRS
,
186 SI_SGPR_TES_OFFCHIP_ADDR
,
187 SI_TES_NUM_USER_SGPR
,
189 /* GFX6-8: TCS only */
190 GFX6_SGPR_TCS_OFFCHIP_LAYOUT
= SI_NUM_RESOURCE_SGPRS
,
191 GFX6_SGPR_TCS_OUT_OFFSETS
,
192 GFX6_SGPR_TCS_OUT_LAYOUT
,
193 GFX6_SGPR_TCS_IN_LAYOUT
,
194 GFX6_TCS_NUM_USER_SGPR
,
196 /* GFX9: Merged shaders. */
197 #if HAVE_32BIT_POINTERS
198 /* 2ND_CONST_AND_SHADER_BUFFERS is set in USER_DATA_ADDR_LO (SGPR0). */
199 /* 2ND_SAMPLERS_AND_IMAGES is set in USER_DATA_ADDR_HI (SGPR1). */
200 GFX9_MERGED_NUM_USER_SGPR
= SI_VS_NUM_USER_SGPR
,
202 /* 2ND_CONST_AND_SHADER_BUFFERS is set in USER_DATA_ADDR_LO/HI (SGPR[0:1]). */
203 GFX9_SGPR_2ND_SAMPLERS_AND_IMAGES
= SI_VS_NUM_USER_SGPR
,
204 GFX9_SGPR_2ND_SAMPLERS_AND_IMAGES_HI
,
205 GFX9_MERGED_NUM_USER_SGPR
,
208 /* GFX9: Merged LS-HS (VS-TCS) only. */
209 GFX9_SGPR_TCS_OFFCHIP_LAYOUT
= GFX9_MERGED_NUM_USER_SGPR
,
210 GFX9_SGPR_TCS_OUT_OFFSETS
,
211 GFX9_SGPR_TCS_OUT_LAYOUT
,
212 GFX9_TCS_NUM_USER_SGPR
,
215 GFX6_GS_NUM_USER_SGPR
= SI_NUM_RESOURCE_SGPRS
,
216 GFX9_GS_NUM_USER_SGPR
= GFX9_MERGED_NUM_USER_SGPR
,
217 SI_GSCOPY_NUM_USER_SGPR
= SI_SGPR_RW_BUFFERS
+ (HAVE_32BIT_POINTERS
? 1 : 2),
220 SI_SGPR_ALPHA_REF
= SI_NUM_RESOURCE_SGPRS
,
224 /* LLVM function parameter indices */
226 SI_NUM_RESOURCE_PARAMS
= 4,
228 /* PS only parameters */
229 SI_PARAM_ALPHA_REF
= SI_NUM_RESOURCE_PARAMS
,
231 SI_PARAM_PERSP_SAMPLE
,
232 SI_PARAM_PERSP_CENTER
,
233 SI_PARAM_PERSP_CENTROID
,
234 SI_PARAM_PERSP_PULL_MODEL
,
235 SI_PARAM_LINEAR_SAMPLE
,
236 SI_PARAM_LINEAR_CENTER
,
237 SI_PARAM_LINEAR_CENTROID
,
238 SI_PARAM_LINE_STIPPLE_TEX
,
239 SI_PARAM_POS_X_FLOAT
,
240 SI_PARAM_POS_Y_FLOAT
,
241 SI_PARAM_POS_Z_FLOAT
,
242 SI_PARAM_POS_W_FLOAT
,
245 SI_PARAM_SAMPLE_COVERAGE
,
246 SI_PARAM_POS_FIXED_PT
,
248 SI_NUM_PARAMS
= SI_PARAM_POS_FIXED_PT
+ 9, /* +8 for COLOR[0..1] */
251 /* Fields of driver-defined VS state SGPR. */
252 /* Clamp vertex color output (only used in VS as VS). */
253 #define S_VS_STATE_CLAMP_VERTEX_COLOR(x) (((unsigned)(x) & 0x1) << 0)
254 #define C_VS_STATE_CLAMP_VERTEX_COLOR 0xFFFFFFFE
255 #define S_VS_STATE_INDEXED(x) (((unsigned)(x) & 0x1) << 1)
256 #define C_VS_STATE_INDEXED 0xFFFFFFFD
257 #define S_VS_STATE_LS_OUT_PATCH_SIZE(x) (((unsigned)(x) & 0x1FFF) << 8)
258 #define C_VS_STATE_LS_OUT_PATCH_SIZE 0xFFE000FF
259 #define S_VS_STATE_LS_OUT_VERTEX_SIZE(x) (((unsigned)(x) & 0xFF) << 24)
260 #define C_VS_STATE_LS_OUT_VERTEX_SIZE 0x00FFFFFF
262 /* SI-specific system values. */
264 TGSI_SEMANTIC_DEFAULT_TESSOUTER_SI
= TGSI_SEMANTIC_COUNT
,
265 TGSI_SEMANTIC_DEFAULT_TESSINNER_SI
,
269 /* Use a property enum that VS wouldn't use. */
270 TGSI_PROPERTY_VS_BLIT_SGPRS
= TGSI_PROPERTY_FS_COORD_ORIGIN
,
272 /* These represent the number of SGPRs the shader uses. */
273 SI_VS_BLIT_SGPRS_POS
= 3,
274 SI_VS_BLIT_SGPRS_POS_COLOR
= 7,
275 SI_VS_BLIT_SGPRS_POS_TEXCOORD
= 9,
278 /* For VS shader key fix_fetch. */
280 SI_FIX_FETCH_NONE
= 0,
281 SI_FIX_FETCH_A2_SNORM
,
282 SI_FIX_FETCH_A2_SSCALED
,
283 SI_FIX_FETCH_A2_SINT
,
284 SI_FIX_FETCH_RGBA_32_UNORM
,
285 SI_FIX_FETCH_RGBX_32_UNORM
,
286 SI_FIX_FETCH_RGBA_32_SNORM
,
287 SI_FIX_FETCH_RGBX_32_SNORM
,
288 SI_FIX_FETCH_RGBA_32_USCALED
,
289 SI_FIX_FETCH_RGBA_32_SSCALED
,
290 SI_FIX_FETCH_RGBA_32_FIXED
,
291 SI_FIX_FETCH_RGBX_32_FIXED
,
292 SI_FIX_FETCH_RG_64_FLOAT
,
293 SI_FIX_FETCH_RGB_64_FLOAT
,
294 SI_FIX_FETCH_RGBA_64_FLOAT
,
295 SI_FIX_FETCH_RGB_8
, /* A = 1.0 */
296 SI_FIX_FETCH_RGB_8_INT
, /* A = 1 */
298 SI_FIX_FETCH_RGB_16_INT
,
303 /* State of the context creating the shader object. */
304 struct si_compiler_ctx_state
{
305 /* Should only be used by si_init_shader_selector_async and
306 * si_build_shader_variant if thread_index == -1 (non-threaded). */
307 LLVMTargetMachineRef tm
;
309 /* Used if thread_index == -1 or if debug.async is true. */
310 struct pipe_debug_callback debug
;
312 /* Used for creating the log string for gallium/ddebug. */
313 bool is_debug_context
;
316 /* A shader selector is a gallium CSO and contains shader variants and
317 * binaries for one TGSI program. This can be shared by multiple contexts.
319 struct si_shader_selector
{
320 struct pipe_reference reference
;
321 struct si_screen
*screen
;
322 struct util_queue_fence ready
;
323 struct si_compiler_ctx_state compiler_ctx_state
;
326 struct si_shader
*first_variant
; /* immutable after the first variant */
327 struct si_shader
*last_variant
; /* mutable */
329 /* The compiled TGSI shader expecting a prolog and/or epilog (not
330 * uploaded to a buffer).
332 struct si_shader
*main_shader_part
;
333 struct si_shader
*main_shader_part_ls
; /* as_ls is set in the key */
334 struct si_shader
*main_shader_part_es
; /* as_es is set in the key */
336 struct si_shader
*gs_copy_shader
;
338 struct tgsi_token
*tokens
;
339 struct nir_shader
*nir
;
340 struct pipe_stream_output_info so
;
341 struct tgsi_shader_info info
;
342 struct tgsi_tessctrl_info tcs_info
;
344 /* PIPE_SHADER_[VERTEX|FRAGMENT|...] */
346 bool vs_needs_prolog
;
347 bool force_correct_derivs_after_kill
;
348 unsigned pa_cl_vs_out_cntl
;
353 unsigned esgs_itemsize
;
356 unsigned gs_input_verts_per_prim
;
357 unsigned gs_output_prim
;
358 unsigned gs_max_out_vertices
;
359 unsigned gs_num_invocations
;
360 unsigned max_gs_stream
; /* count - 1 */
361 unsigned gsvs_vertex_size
;
362 unsigned max_gsvs_emit_size
;
363 unsigned enabled_streamout_buffer_mask
;
366 unsigned color_attr_index
[2];
367 unsigned db_shader_control
;
368 /* Set 0xf or 0x0 (4 bits) per each written output.
369 * ANDed with spi_shader_col_format.
371 unsigned colors_written_4bit
;
376 uint64_t outputs_written
; /* "get_unique_index" bits */
377 uint32_t patch_outputs_written
; /* "get_unique_index_patch" bits */
379 uint64_t inputs_read
; /* "get_unique_index" bits */
381 /* bitmasks of used descriptor slots */
382 uint32_t active_const_and_shader_buffers
;
383 uint64_t active_samplers_and_images
;
386 /* Valid shader configurations:
388 * API shaders VS | TCS | TES | GS |pass| PS
389 * are compiled as: | | | |thru|
391 * Only VS & PS: VS | | | | | PS
392 * GFX6 - with GS: ES | | | GS | VS | PS
393 * - with tess: LS | HS | VS | | | PS
394 * - with both: LS | HS | ES | GS | VS | PS
395 * GFX9 - with GS: -> | | | GS | VS | PS
396 * - with tess: -> | HS | VS | | | PS
397 * - with both: -> | HS | -> | GS | VS | PS
399 * -> = merged with the next stage
402 /* Use the byte alignment for all following structure members for optimal
403 * shader key memory footprint.
405 #pragma pack(push, 1)
407 /* Common VS bits between the shader key and the prolog key. */
408 struct si_vs_prolog_bits
{
409 /* - If neither "is_one" nor "is_fetched" has a bit set, the instance
411 * - If "is_one" has a bit set, the instance divisor is 1.
412 * - If "is_fetched" has a bit set, the instance divisor will be loaded
413 * from the constant buffer.
415 uint16_t instance_divisor_is_one
; /* bitmask of inputs */
416 uint16_t instance_divisor_is_fetched
; /* bitmask of inputs */
417 unsigned ls_vgpr_fix
:1;
420 /* Common TCS bits between the shader key and the epilog key. */
421 struct si_tcs_epilog_bits
{
422 unsigned prim_mode
:3;
423 unsigned invoc0_tess_factors_are_def
:1;
424 unsigned tes_reads_tess_factors
:1;
427 struct si_gs_prolog_bits
{
428 unsigned tri_strip_adj_fix
:1;
431 /* Common PS bits between the shader key and the prolog key. */
432 struct si_ps_prolog_bits
{
433 unsigned color_two_side
:1;
434 unsigned flatshade_colors
:1;
435 unsigned poly_stipple
:1;
436 unsigned force_persp_sample_interp
:1;
437 unsigned force_linear_sample_interp
:1;
438 unsigned force_persp_center_interp
:1;
439 unsigned force_linear_center_interp
:1;
440 unsigned bc_optimize_for_persp
:1;
441 unsigned bc_optimize_for_linear
:1;
442 unsigned samplemask_log_ps_iter
:3;
445 /* Common PS bits between the shader key and the epilog key. */
446 struct si_ps_epilog_bits
{
447 unsigned spi_shader_col_format
;
448 unsigned color_is_int8
:8;
449 unsigned color_is_int10
:8;
450 unsigned last_cbuf
:3;
451 unsigned alpha_func
:3;
452 unsigned alpha_to_one
:1;
453 unsigned poly_line_smoothing
:1;
454 unsigned clamp_color
:1;
457 union si_shader_part_key
{
459 struct si_vs_prolog_bits states
;
460 unsigned num_input_sgprs
:6;
461 /* For merged stages such as LS-HS, HS input VGPRs are first. */
462 unsigned num_merged_next_stage_vgprs
:3;
463 unsigned last_input
:4;
466 /* Prologs for monolithic shaders shouldn't set EXEC. */
467 unsigned is_monolithic
:1;
470 struct si_tcs_epilog_bits states
;
473 struct si_gs_prolog_bits states
;
474 /* Prologs of monolithic shaders shouldn't set EXEC. */
475 unsigned is_monolithic
:1;
478 struct si_ps_prolog_bits states
;
479 unsigned num_input_sgprs
:6;
480 unsigned num_input_vgprs
:5;
481 /* Color interpolation and two-side color selection. */
482 unsigned colors_read
:8; /* color input components read */
483 unsigned num_interp_inputs
:5; /* BCOLOR is at this location */
484 unsigned face_vgpr_index
:5;
485 unsigned ancillary_vgpr_index
:5;
487 char color_attr_index
[2];
488 char color_interp_vgpr_index
[2]; /* -1 == constant */
491 struct si_ps_epilog_bits states
;
492 unsigned colors_written
:8;
494 unsigned writes_stencil
:1;
495 unsigned writes_samplemask
:1;
499 struct si_shader_key
{
500 /* Prolog and epilog flags. */
503 struct si_vs_prolog_bits prolog
;
506 struct si_vs_prolog_bits ls_prolog
; /* for merged LS-HS */
507 struct si_shader_selector
*ls
; /* for merged LS-HS */
508 struct si_tcs_epilog_bits epilog
;
509 } tcs
; /* tessellation control shader */
511 struct si_vs_prolog_bits vs_prolog
; /* for merged ES-GS */
512 struct si_shader_selector
*es
; /* for merged ES-GS */
513 struct si_gs_prolog_bits prolog
;
516 struct si_ps_prolog_bits prolog
;
517 struct si_ps_epilog_bits epilog
;
521 /* These two are initially set according to the NEXT_SHADER property,
522 * or guessed if the property doesn't seem correct.
524 unsigned as_es
:1; /* export shader, which precedes GS */
525 unsigned as_ls
:1; /* local shader, which precedes TCS */
527 /* Flags for monolithic compilation only. */
529 /* One byte for every input: SI_FIX_FETCH_* enums. */
530 uint8_t vs_fix_fetch
[SI_MAX_ATTRIBS
];
533 uint64_t ff_tcs_inputs_to_copy
; /* for fixed-func TCS */
534 /* When PS needs PrimID and GS is disabled. */
535 unsigned vs_export_prim_id
:1;
537 unsigned interpolate_at_sample_force_center
:1;
542 /* Optimization flags for asynchronous compilation only. */
544 /* For HW VS (it can be VS, TES, GS) */
545 uint64_t kill_outputs
; /* "get_unique_index" bits */
546 unsigned clip_disable
:1;
548 /* For shaders where monolithic variants have better code.
550 * This is a flag that has no effect on code generation,
551 * but forces monolithic shaders to be used as soon as
552 * possible, because it's in the "opt" group.
554 unsigned prefer_mono
:1;
558 /* Restore the pack alignment to default. */
561 struct si_shader_config
{
564 unsigned spilled_sgprs
;
565 unsigned spilled_vgprs
;
566 unsigned private_mem_vgprs
;
568 unsigned max_simd_waves
;
569 unsigned spi_ps_input_ena
;
570 unsigned spi_ps_input_addr
;
572 unsigned scratch_bytes_per_wave
;
577 /* GCN-specific shader info. */
578 struct si_shader_info
{
579 ubyte vs_output_param_offset
[SI_MAX_VS_OUTPUTS
];
580 ubyte num_input_sgprs
;
581 ubyte num_input_vgprs
;
582 signed char face_vgpr_index
;
583 signed char ancillary_vgpr_index
;
584 bool uses_instanceid
;
585 ubyte nr_pos_exports
;
586 ubyte nr_param_exports
;
590 struct si_compiler_ctx_state compiler_ctx_state
;
592 struct si_shader_selector
*selector
;
593 struct si_shader_selector
*previous_stage_sel
; /* for refcounting */
594 struct si_shader
*next_variant
;
596 struct si_shader_part
*prolog
;
597 struct si_shader
*previous_stage
; /* for GFX9 */
598 struct si_shader_part
*prolog2
;
599 struct si_shader_part
*epilog
;
601 struct si_pm4_state
*pm4
;
602 struct r600_resource
*bo
;
603 struct r600_resource
*scratch_bo
;
604 struct si_shader_key key
;
605 struct util_queue_fence ready
;
606 bool compilation_failed
;
609 bool is_binary_shared
;
610 bool is_gs_copy_shader
;
612 /* The following data is all that's needed for binary shaders. */
613 struct ac_shader_binary binary
;
614 struct si_shader_config config
;
615 struct si_shader_info info
;
617 /* Shader key + LLVM IR + disassembly + statistics.
618 * Generated for debug contexts only.
621 size_t shader_log_size
;
624 struct si_shader_part
{
625 struct si_shader_part
*next
;
626 union si_shader_part_key key
;
627 struct ac_shader_binary binary
;
628 struct si_shader_config config
;
633 si_generate_gs_copy_shader(struct si_screen
*sscreen
,
634 LLVMTargetMachineRef tm
,
635 struct si_shader_selector
*gs_selector
,
636 struct pipe_debug_callback
*debug
);
637 int si_compile_tgsi_shader(struct si_screen
*sscreen
,
638 LLVMTargetMachineRef tm
,
639 struct si_shader
*shader
,
641 struct pipe_debug_callback
*debug
);
642 int si_shader_create(struct si_screen
*sscreen
, LLVMTargetMachineRef tm
,
643 struct si_shader
*shader
,
644 struct pipe_debug_callback
*debug
);
645 void si_shader_destroy(struct si_shader
*shader
);
646 unsigned si_shader_io_get_unique_index_patch(unsigned semantic_name
, unsigned index
);
647 unsigned si_shader_io_get_unique_index(unsigned semantic_name
, unsigned index
);
648 int si_shader_binary_upload(struct si_screen
*sscreen
, struct si_shader
*shader
);
649 void si_shader_dump(struct si_screen
*sscreen
, const struct si_shader
*shader
,
650 struct pipe_debug_callback
*debug
, unsigned processor
,
651 FILE *f
, bool check_debug_option
);
652 void si_shader_dump_stats_for_shader_db(const struct si_shader
*shader
,
653 struct pipe_debug_callback
*debug
);
654 void si_multiwave_lds_size_workaround(struct si_screen
*sscreen
,
656 void si_shader_apply_scratch_relocs(struct si_shader
*shader
,
657 uint64_t scratch_va
);
658 void si_shader_binary_read_config(struct ac_shader_binary
*binary
,
659 struct si_shader_config
*conf
,
660 unsigned symbol_offset
);
661 const char *si_get_shader_name(const struct si_shader
*shader
, unsigned processor
);
663 /* si_shader_nir.c */
664 void si_nir_scan_shader(const struct nir_shader
*nir
,
665 struct tgsi_shader_info
*info
);
666 void si_nir_scan_tess_ctrl(const struct nir_shader
*nir
,
667 const struct tgsi_shader_info
*info
,
668 struct tgsi_tessctrl_info
*out
);
669 void si_lower_nir(struct si_shader_selector
*sel
);
671 /* Inline helpers. */
673 /* Return the pointer to the main shader part's pointer. */
674 static inline struct si_shader
**
675 si_get_main_shader_part(struct si_shader_selector
*sel
,
676 struct si_shader_key
*key
)
679 return &sel
->main_shader_part_ls
;
681 return &sel
->main_shader_part_es
;
682 return &sel
->main_shader_part
;
686 si_shader_uses_bindless_samplers(struct si_shader_selector
*selector
)
688 return selector
? selector
->info
.uses_bindless_samplers
: false;
692 si_shader_uses_bindless_images(struct si_shader_selector
*selector
)
694 return selector
? selector
->info
.uses_bindless_images
: false;
697 void si_destroy_shader_selector(struct si_context
*sctx
,
698 struct si_shader_selector
*sel
);
701 si_shader_selector_reference(struct si_context
*sctx
,
702 struct si_shader_selector
**dst
,
703 struct si_shader_selector
*src
)
705 if (pipe_reference(&(*dst
)->reference
, &src
->reference
))
706 si_destroy_shader_selector(sctx
, *dst
);