2 * Copyright 2012 Advanced Micro Devices, Inc.
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * on the rights to use, copy, modify, merge, publish, distribute, sub
9 * license, and/or sell copies of the Software, and to permit persons to whom
10 * the Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
20 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
21 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
22 * USE OR OTHER DEALINGS IN THE SOFTWARE.
25 /* The compiler middle-end architecture: Explaining (non-)monolithic shaders
26 * -------------------------------------------------------------------------
28 * Typically, there is one-to-one correspondence between API and HW shaders,
29 * that is, for every API shader, there is exactly one shader binary in
32 * The problem with that is that we also have to emulate some API states
33 * (e.g. alpha-test, and many others) in shaders too. The two obvious ways
34 * to deal with it are:
35 * - each shader has multiple variants for each combination of emulated states,
36 * and the variants are compiled on demand, possibly relying on a shader
37 * cache for good performance
38 * - patch shaders at the binary level
40 * This driver uses something completely different. The emulated states are
41 * usually implemented at the beginning or end of shaders. Therefore, we can
42 * split the shader into 3 parts:
43 * - prolog part (shader code dependent on states)
44 * - main part (the API shader)
45 * - epilog part (shader code dependent on states)
47 * Each part is compiled as a separate shader and the final binaries are
48 * concatenated. This type of shader is called non-monolithic, because it
49 * consists of multiple independent binaries. Creating a new shader variant
50 * is therefore only a concatenation of shader parts (binaries) and doesn't
51 * involve any compilation. The main shader parts are the only parts that are
52 * compiled when applications create shader objects. The prolog and epilog
53 * parts are compiled on the first use and saved, so that their binaries can
54 * be reused by many other shaders.
56 * One of the roles of the prolog part is to compute vertex buffer addresses
57 * for vertex shaders. A few of the roles of the epilog part are color buffer
58 * format conversions in pixel shaders that we have to do manually, and write
59 * tessellation factors in tessellation control shaders. The prolog and epilog
60 * have many other important responsibilities in various shader stages.
61 * They don't just "emulate legacy stuff".
63 * Monolithic shaders are shaders where the parts are combined before LLVM
64 * compilation, and the whole thing is compiled and optimized as one unit with
65 * one binary on the output. The result is the same as the non-monolithic
66 * shader, but the final code can be better, because LLVM can optimize across
67 * all shader parts. Monolithic shaders aren't usually used except for these
70 * 1) Some rarely-used states require modification of the main shader part
71 * itself, and in such cases, only the monolithic shader variant is
72 * compiled, and that's always done on the first use.
74 * 2) When we do cross-stage optimizations for separate shader objects and
75 * e.g. eliminate unused shader varyings, the resulting optimized shader
76 * variants are always compiled as monolithic shaders, and always
77 * asynchronously (i.e. not stalling ongoing rendering). We call them
78 * "optimized monolithic" shaders. The important property here is that
79 * the non-monolithic unoptimized shader variant is always available for use
80 * when the asynchronous compilation of the optimized shader is not done
83 * Starting with GFX9 chips, some shader stages are merged, and the number of
84 * shader parts per shader increased. The complete new list of shader parts is:
85 * - 1st shader: prolog part
86 * - 1st shader: main part
87 * - 2nd shader: prolog part
88 * - 2nd shader: main part
89 * - 2nd shader: epilog part
92 /* How linking shader inputs and outputs between vertex, tessellation, and
93 * geometry shaders works.
95 * Inputs and outputs between shaders are stored in a buffer. This buffer
96 * lives in LDS (typical case for tessellation), but it can also live
97 * in memory (ESGS). Each input or output has a fixed location within a vertex.
98 * The highest used input or output determines the stride between vertices.
100 * Since GS and tessellation are only possible in the OpenGL core profile,
101 * only these semantics are valid for per-vertex data:
108 * CULLDIST0..1 (not implemented)
111 * For example, a shader only writing GENERIC0 has the output stride of 5.
113 * Only these semantics are valid for per-patch data:
121 * That's how independent shaders agree on input and output locations.
122 * The si_shader_io_get_unique_index function assigns the locations.
124 * For tessellation, other required information for calculating the input and
125 * output addresses like the vertex stride, the patch stride, and the offsets
126 * where per-vertex and per-patch data start, is passed to the shader via
127 * user data SGPRs. The offsets and strides are calculated at draw time and
128 * aren't available at compile time.
134 #include <llvm-c/Core.h> /* LLVMModuleRef */
135 #include <llvm-c/TargetMachine.h>
136 #include "tgsi/tgsi_scan.h"
137 #include "util/u_inlines.h"
138 #include "util/u_queue.h"
140 #include "ac_binary.h"
141 #include "ac_llvm_build.h"
149 #define SI_MAX_ATTRIBS 16
150 #define SI_MAX_VS_OUTPUTS 40
152 /* Shader IO unique indices are supported for TGSI_SEMANTIC_GENERIC with an
153 * index smaller than this.
155 #define SI_MAX_IO_GENERIC 43
157 /* SGPR user data indices */
159 SI_SGPR_RW_BUFFERS
, /* rings (& stream-out, VS only) */
160 #if !HAVE_32BIT_POINTERS
161 SI_SGPR_RW_BUFFERS_HI
,
163 SI_SGPR_BINDLESS_SAMPLERS_AND_IMAGES
,
164 #if !HAVE_32BIT_POINTERS
165 SI_SGPR_BINDLESS_SAMPLERS_AND_IMAGES_HI
,
167 SI_SGPR_CONST_AND_SHADER_BUFFERS
, /* or just a constant buffer 0 pointer */
168 #if !HAVE_32BIT_POINTERS
169 SI_SGPR_CONST_AND_SHADER_BUFFERS_HI
,
171 SI_SGPR_SAMPLERS_AND_IMAGES
,
172 #if !HAVE_32BIT_POINTERS
173 SI_SGPR_SAMPLERS_AND_IMAGES_HI
,
175 SI_NUM_RESOURCE_SGPRS
,
177 /* API VS, TES without GS, GS copy shader */
178 SI_SGPR_VS_STATE_BITS
= SI_NUM_RESOURCE_SGPRS
,
179 SI_NUM_VS_STATE_RESOURCE_SGPRS
,
181 /* all VS variants */
182 SI_SGPR_BASE_VERTEX
= SI_NUM_VS_STATE_RESOURCE_SGPRS
,
183 SI_SGPR_START_INSTANCE
,
187 SI_SGPR_VS_BLIT_DATA
= SI_SGPR_CONST_AND_SHADER_BUFFERS
,
190 SI_SGPR_TES_OFFCHIP_LAYOUT
= SI_NUM_VS_STATE_RESOURCE_SGPRS
,
191 SI_SGPR_TES_OFFCHIP_ADDR
,
192 SI_TES_NUM_USER_SGPR
,
194 /* GFX6-8: TCS only */
195 GFX6_SGPR_TCS_OFFCHIP_LAYOUT
= SI_NUM_RESOURCE_SGPRS
,
196 GFX6_SGPR_TCS_OUT_OFFSETS
,
197 GFX6_SGPR_TCS_OUT_LAYOUT
,
198 GFX6_SGPR_TCS_IN_LAYOUT
,
199 GFX6_TCS_NUM_USER_SGPR
,
201 /* GFX9: Merged shaders. */
202 #if HAVE_32BIT_POINTERS
203 /* 2ND_CONST_AND_SHADER_BUFFERS is set in USER_DATA_ADDR_LO (SGPR0). */
204 /* 2ND_SAMPLERS_AND_IMAGES is set in USER_DATA_ADDR_HI (SGPR1). */
205 GFX9_MERGED_NUM_USER_SGPR
= SI_VS_NUM_USER_SGPR
,
207 /* 2ND_CONST_AND_SHADER_BUFFERS is set in USER_DATA_ADDR_LO/HI (SGPR[0:1]). */
208 GFX9_SGPR_2ND_SAMPLERS_AND_IMAGES
= SI_VS_NUM_USER_SGPR
,
209 GFX9_SGPR_2ND_SAMPLERS_AND_IMAGES_HI
,
210 GFX9_MERGED_NUM_USER_SGPR
,
213 /* GFX9: Merged LS-HS (VS-TCS) only. */
214 GFX9_SGPR_TCS_OFFCHIP_LAYOUT
= GFX9_MERGED_NUM_USER_SGPR
,
215 GFX9_SGPR_TCS_OUT_OFFSETS
,
216 GFX9_SGPR_TCS_OUT_LAYOUT
,
217 #if !HAVE_32BIT_POINTERS
218 GFX9_SGPR_align_for_vb_pointer
,
220 GFX9_TCS_NUM_USER_SGPR
,
223 GFX6_GS_NUM_USER_SGPR
= SI_NUM_RESOURCE_SGPRS
,
224 #if HAVE_32BIT_POINTERS
225 GFX9_VSGS_NUM_USER_SGPR
= SI_VS_NUM_USER_SGPR
,
226 GFX9_TESGS_NUM_USER_SGPR
= SI_TES_NUM_USER_SGPR
,
228 GFX9_VSGS_NUM_USER_SGPR
= GFX9_MERGED_NUM_USER_SGPR
,
229 GFX9_TESGS_NUM_USER_SGPR
= GFX9_MERGED_NUM_USER_SGPR
,
231 SI_GSCOPY_NUM_USER_SGPR
= SI_NUM_VS_STATE_RESOURCE_SGPRS
,
234 SI_SGPR_ALPHA_REF
= SI_NUM_RESOURCE_SGPRS
,
238 /* LLVM function parameter indices */
240 SI_NUM_RESOURCE_PARAMS
= 4,
242 /* PS only parameters */
243 SI_PARAM_ALPHA_REF
= SI_NUM_RESOURCE_PARAMS
,
245 SI_PARAM_PERSP_SAMPLE
,
246 SI_PARAM_PERSP_CENTER
,
247 SI_PARAM_PERSP_CENTROID
,
248 SI_PARAM_PERSP_PULL_MODEL
,
249 SI_PARAM_LINEAR_SAMPLE
,
250 SI_PARAM_LINEAR_CENTER
,
251 SI_PARAM_LINEAR_CENTROID
,
252 SI_PARAM_LINE_STIPPLE_TEX
,
253 SI_PARAM_POS_X_FLOAT
,
254 SI_PARAM_POS_Y_FLOAT
,
255 SI_PARAM_POS_Z_FLOAT
,
256 SI_PARAM_POS_W_FLOAT
,
259 SI_PARAM_SAMPLE_COVERAGE
,
260 SI_PARAM_POS_FIXED_PT
,
262 SI_NUM_PARAMS
= SI_PARAM_POS_FIXED_PT
+ 9, /* +8 for COLOR[0..1] */
265 /* Fields of driver-defined VS state SGPR. */
266 /* Clamp vertex color output (only used in VS as VS). */
267 #define S_VS_STATE_CLAMP_VERTEX_COLOR(x) (((unsigned)(x) & 0x1) << 0)
268 #define C_VS_STATE_CLAMP_VERTEX_COLOR 0xFFFFFFFE
269 #define S_VS_STATE_INDEXED(x) (((unsigned)(x) & 0x1) << 1)
270 #define C_VS_STATE_INDEXED 0xFFFFFFFD
271 #define S_VS_STATE_LS_OUT_PATCH_SIZE(x) (((unsigned)(x) & 0x1FFF) << 8)
272 #define C_VS_STATE_LS_OUT_PATCH_SIZE 0xFFE000FF
273 #define S_VS_STATE_LS_OUT_VERTEX_SIZE(x) (((unsigned)(x) & 0xFF) << 24)
274 #define C_VS_STATE_LS_OUT_VERTEX_SIZE 0x00FFFFFF
276 /* SI-specific system values. */
278 TGSI_SEMANTIC_DEFAULT_TESSOUTER_SI
= TGSI_SEMANTIC_COUNT
,
279 TGSI_SEMANTIC_DEFAULT_TESSINNER_SI
,
283 /* Use a property enum that CS wouldn't use. */
284 TGSI_PROPERTY_CS_LOCAL_SIZE
= TGSI_PROPERTY_FS_COORD_ORIGIN
,
286 /* Use a property enum that VS wouldn't use. */
287 TGSI_PROPERTY_VS_BLIT_SGPRS
= TGSI_PROPERTY_FS_COORD_ORIGIN
,
289 /* These represent the number of SGPRs the shader uses. */
290 SI_VS_BLIT_SGPRS_POS
= 3,
291 SI_VS_BLIT_SGPRS_POS_COLOR
= 7,
292 SI_VS_BLIT_SGPRS_POS_TEXCOORD
= 9,
295 /* For VS shader key fix_fetch. */
297 SI_FIX_FETCH_NONE
= 0,
298 SI_FIX_FETCH_A2_SNORM
,
299 SI_FIX_FETCH_A2_SSCALED
,
300 SI_FIX_FETCH_A2_SINT
,
301 SI_FIX_FETCH_RGBA_32_UNORM
,
302 SI_FIX_FETCH_RGBX_32_UNORM
,
303 SI_FIX_FETCH_RGBA_32_SNORM
,
304 SI_FIX_FETCH_RGBX_32_SNORM
,
305 SI_FIX_FETCH_RGBA_32_USCALED
,
306 SI_FIX_FETCH_RGBA_32_SSCALED
,
307 SI_FIX_FETCH_RGBA_32_FIXED
,
308 SI_FIX_FETCH_RGBX_32_FIXED
,
309 SI_FIX_FETCH_RG_64_FLOAT
,
310 SI_FIX_FETCH_RGB_64_FLOAT
,
311 SI_FIX_FETCH_RGBA_64_FLOAT
,
312 SI_FIX_FETCH_RGB_8
, /* A = 1.0 */
313 SI_FIX_FETCH_RGB_8_INT
, /* A = 1 */
315 SI_FIX_FETCH_RGB_16_INT
,
320 /* Per-thread persistent LLVM objects. */
322 LLVMTargetMachineRef tm
;
323 LLVMTargetLibraryInfoRef target_library_info
;
324 LLVMPassManagerRef passmgr
;
327 /* State of the context creating the shader object. */
328 struct si_compiler_ctx_state
{
329 /* Should only be used by si_init_shader_selector_async and
330 * si_build_shader_variant if thread_index == -1 (non-threaded). */
331 struct si_compiler
*compiler
;
333 /* Used if thread_index == -1 or if debug.async is true. */
334 struct pipe_debug_callback debug
;
336 /* Used for creating the log string for gallium/ddebug. */
337 bool is_debug_context
;
340 /* A shader selector is a gallium CSO and contains shader variants and
341 * binaries for one TGSI program. This can be shared by multiple contexts.
343 struct si_shader_selector
{
344 struct pipe_reference reference
;
345 struct si_screen
*screen
;
346 struct util_queue_fence ready
;
347 struct si_compiler_ctx_state compiler_ctx_state
;
350 struct si_shader
*first_variant
; /* immutable after the first variant */
351 struct si_shader
*last_variant
; /* mutable */
353 /* The compiled TGSI shader expecting a prolog and/or epilog (not
354 * uploaded to a buffer).
356 struct si_shader
*main_shader_part
;
357 struct si_shader
*main_shader_part_ls
; /* as_ls is set in the key */
358 struct si_shader
*main_shader_part_es
; /* as_es is set in the key */
360 struct si_shader
*gs_copy_shader
;
362 struct tgsi_token
*tokens
;
363 struct nir_shader
*nir
;
364 struct pipe_stream_output_info so
;
365 struct tgsi_shader_info info
;
366 struct tgsi_tessctrl_info tcs_info
;
368 /* PIPE_SHADER_[VERTEX|FRAGMENT|...] */
370 bool vs_needs_prolog
;
371 bool force_correct_derivs_after_kill
;
372 unsigned pa_cl_vs_out_cntl
;
377 unsigned esgs_itemsize
;
380 unsigned gs_input_verts_per_prim
;
381 unsigned gs_output_prim
;
382 unsigned gs_max_out_vertices
;
383 unsigned gs_num_invocations
;
384 unsigned max_gs_stream
; /* count - 1 */
385 unsigned gsvs_vertex_size
;
386 unsigned max_gsvs_emit_size
;
387 unsigned enabled_streamout_buffer_mask
;
390 unsigned color_attr_index
[2];
391 unsigned db_shader_control
;
392 /* Set 0xf or 0x0 (4 bits) per each written output.
393 * ANDed with spi_shader_col_format.
395 unsigned colors_written_4bit
;
397 uint64_t outputs_written_before_ps
; /* "get_unique_index" bits */
398 uint64_t outputs_written
; /* "get_unique_index" bits */
399 uint32_t patch_outputs_written
; /* "get_unique_index_patch" bits */
401 uint64_t inputs_read
; /* "get_unique_index" bits */
403 /* bitmasks of used descriptor slots */
404 uint32_t active_const_and_shader_buffers
;
405 uint64_t active_samplers_and_images
;
408 /* Valid shader configurations:
410 * API shaders VS | TCS | TES | GS |pass| PS
411 * are compiled as: | | | |thru|
413 * Only VS & PS: VS | | | | | PS
414 * GFX6 - with GS: ES | | | GS | VS | PS
415 * - with tess: LS | HS | VS | | | PS
416 * - with both: LS | HS | ES | GS | VS | PS
417 * GFX9 - with GS: -> | | | GS | VS | PS
418 * - with tess: -> | HS | VS | | | PS
419 * - with both: -> | HS | -> | GS | VS | PS
421 * -> = merged with the next stage
424 /* Use the byte alignment for all following structure members for optimal
425 * shader key memory footprint.
427 #pragma pack(push, 1)
429 /* Common VS bits between the shader key and the prolog key. */
430 struct si_vs_prolog_bits
{
431 /* - If neither "is_one" nor "is_fetched" has a bit set, the instance
433 * - If "is_one" has a bit set, the instance divisor is 1.
434 * - If "is_fetched" has a bit set, the instance divisor will be loaded
435 * from the constant buffer.
437 uint16_t instance_divisor_is_one
; /* bitmask of inputs */
438 uint16_t instance_divisor_is_fetched
; /* bitmask of inputs */
439 unsigned ls_vgpr_fix
:1;
442 /* Common TCS bits between the shader key and the epilog key. */
443 struct si_tcs_epilog_bits
{
444 unsigned prim_mode
:3;
445 unsigned invoc0_tess_factors_are_def
:1;
446 unsigned tes_reads_tess_factors
:1;
449 struct si_gs_prolog_bits
{
450 unsigned tri_strip_adj_fix
:1;
451 unsigned gfx9_prev_is_vs
:1;
454 /* Common PS bits between the shader key and the prolog key. */
455 struct si_ps_prolog_bits
{
456 unsigned color_two_side
:1;
457 unsigned flatshade_colors
:1;
458 unsigned poly_stipple
:1;
459 unsigned force_persp_sample_interp
:1;
460 unsigned force_linear_sample_interp
:1;
461 unsigned force_persp_center_interp
:1;
462 unsigned force_linear_center_interp
:1;
463 unsigned bc_optimize_for_persp
:1;
464 unsigned bc_optimize_for_linear
:1;
465 unsigned samplemask_log_ps_iter
:3;
468 /* Common PS bits between the shader key and the epilog key. */
469 struct si_ps_epilog_bits
{
470 unsigned spi_shader_col_format
;
471 unsigned color_is_int8
:8;
472 unsigned color_is_int10
:8;
473 unsigned last_cbuf
:3;
474 unsigned alpha_func
:3;
475 unsigned alpha_to_one
:1;
476 unsigned poly_line_smoothing
:1;
477 unsigned clamp_color
:1;
480 union si_shader_part_key
{
482 struct si_vs_prolog_bits states
;
483 unsigned num_input_sgprs
:6;
484 /* For merged stages such as LS-HS, HS input VGPRs are first. */
485 unsigned num_merged_next_stage_vgprs
:3;
486 unsigned last_input
:4;
489 /* Prologs for monolithic shaders shouldn't set EXEC. */
490 unsigned is_monolithic
:1;
493 struct si_tcs_epilog_bits states
;
496 struct si_gs_prolog_bits states
;
497 /* Prologs of monolithic shaders shouldn't set EXEC. */
498 unsigned is_monolithic
:1;
501 struct si_ps_prolog_bits states
;
502 unsigned num_input_sgprs
:6;
503 unsigned num_input_vgprs
:5;
504 /* Color interpolation and two-side color selection. */
505 unsigned colors_read
:8; /* color input components read */
506 unsigned num_interp_inputs
:5; /* BCOLOR is at this location */
507 unsigned face_vgpr_index
:5;
508 unsigned ancillary_vgpr_index
:5;
510 char color_attr_index
[2];
511 char color_interp_vgpr_index
[2]; /* -1 == constant */
514 struct si_ps_epilog_bits states
;
515 unsigned colors_written
:8;
517 unsigned writes_stencil
:1;
518 unsigned writes_samplemask
:1;
522 struct si_shader_key
{
523 /* Prolog and epilog flags. */
526 struct si_vs_prolog_bits prolog
;
529 struct si_vs_prolog_bits ls_prolog
; /* for merged LS-HS */
530 struct si_shader_selector
*ls
; /* for merged LS-HS */
531 struct si_tcs_epilog_bits epilog
;
532 } tcs
; /* tessellation control shader */
534 struct si_vs_prolog_bits vs_prolog
; /* for merged ES-GS */
535 struct si_shader_selector
*es
; /* for merged ES-GS */
536 struct si_gs_prolog_bits prolog
;
539 struct si_ps_prolog_bits prolog
;
540 struct si_ps_epilog_bits epilog
;
544 /* These two are initially set according to the NEXT_SHADER property,
545 * or guessed if the property doesn't seem correct.
547 unsigned as_es
:1; /* export shader, which precedes GS */
548 unsigned as_ls
:1; /* local shader, which precedes TCS */
550 /* Flags for monolithic compilation only. */
552 /* One byte for every input: SI_FIX_FETCH_* enums. */
553 uint8_t vs_fix_fetch
[SI_MAX_ATTRIBS
];
556 uint64_t ff_tcs_inputs_to_copy
; /* for fixed-func TCS */
557 /* When PS needs PrimID and GS is disabled. */
558 unsigned vs_export_prim_id
:1;
560 unsigned interpolate_at_sample_force_center
:1;
561 unsigned fbfetch_msaa
;
562 unsigned fbfetch_is_1D
;
563 unsigned fbfetch_layered
;
568 /* Optimization flags for asynchronous compilation only. */
570 /* For HW VS (it can be VS, TES, GS) */
571 uint64_t kill_outputs
; /* "get_unique_index" bits */
572 unsigned clip_disable
:1;
574 /* For shaders where monolithic variants have better code.
576 * This is a flag that has no effect on code generation,
577 * but forces monolithic shaders to be used as soon as
578 * possible, because it's in the "opt" group.
580 unsigned prefer_mono
:1;
584 /* Restore the pack alignment to default. */
587 struct si_shader_config
{
590 unsigned spilled_sgprs
;
591 unsigned spilled_vgprs
;
592 unsigned private_mem_vgprs
;
594 unsigned max_simd_waves
;
595 unsigned spi_ps_input_ena
;
596 unsigned spi_ps_input_addr
;
598 unsigned scratch_bytes_per_wave
;
603 /* GCN-specific shader info. */
604 struct si_shader_info
{
605 ubyte vs_output_param_offset
[SI_MAX_VS_OUTPUTS
];
606 ubyte num_input_sgprs
;
607 ubyte num_input_vgprs
;
608 signed char face_vgpr_index
;
609 signed char ancillary_vgpr_index
;
610 bool uses_instanceid
;
611 ubyte nr_pos_exports
;
612 ubyte nr_param_exports
;
616 struct si_compiler_ctx_state compiler_ctx_state
;
618 struct si_shader_selector
*selector
;
619 struct si_shader_selector
*previous_stage_sel
; /* for refcounting */
620 struct si_shader
*next_variant
;
622 struct si_shader_part
*prolog
;
623 struct si_shader
*previous_stage
; /* for GFX9 */
624 struct si_shader_part
*prolog2
;
625 struct si_shader_part
*epilog
;
627 struct si_pm4_state
*pm4
;
628 struct r600_resource
*bo
;
629 struct r600_resource
*scratch_bo
;
630 struct si_shader_key key
;
631 struct util_queue_fence ready
;
632 bool compilation_failed
;
635 bool is_binary_shared
;
636 bool is_gs_copy_shader
;
638 /* The following data is all that's needed for binary shaders. */
639 struct ac_shader_binary binary
;
640 struct si_shader_config config
;
641 struct si_shader_info info
;
643 /* Shader key + LLVM IR + disassembly + statistics.
644 * Generated for debug contexts only.
647 size_t shader_log_size
;
650 struct si_shader_part
{
651 struct si_shader_part
*next
;
652 union si_shader_part_key key
;
653 struct ac_shader_binary binary
;
654 struct si_shader_config config
;
659 si_generate_gs_copy_shader(struct si_screen
*sscreen
,
660 struct si_compiler
*compiler
,
661 struct si_shader_selector
*gs_selector
,
662 struct pipe_debug_callback
*debug
);
663 int si_compile_tgsi_shader(struct si_screen
*sscreen
,
664 struct si_compiler
*compiler
,
665 struct si_shader
*shader
,
666 struct pipe_debug_callback
*debug
);
667 int si_shader_create(struct si_screen
*sscreen
, struct si_compiler
*compiler
,
668 struct si_shader
*shader
,
669 struct pipe_debug_callback
*debug
);
670 void si_shader_destroy(struct si_shader
*shader
);
671 unsigned si_shader_io_get_unique_index_patch(unsigned semantic_name
, unsigned index
);
672 unsigned si_shader_io_get_unique_index(unsigned semantic_name
, unsigned index
,
673 unsigned is_varying
);
674 int si_shader_binary_upload(struct si_screen
*sscreen
, struct si_shader
*shader
);
675 void si_shader_dump(struct si_screen
*sscreen
, const struct si_shader
*shader
,
676 struct pipe_debug_callback
*debug
, unsigned processor
,
677 FILE *f
, bool check_debug_option
);
678 void si_shader_dump_stats_for_shader_db(const struct si_shader
*shader
,
679 struct pipe_debug_callback
*debug
);
680 void si_multiwave_lds_size_workaround(struct si_screen
*sscreen
,
682 void si_shader_apply_scratch_relocs(struct si_shader
*shader
,
683 uint64_t scratch_va
);
684 void si_shader_binary_read_config(struct ac_shader_binary
*binary
,
685 struct si_shader_config
*conf
,
686 unsigned symbol_offset
);
687 const char *si_get_shader_name(const struct si_shader
*shader
, unsigned processor
);
689 /* si_shader_nir.c */
690 void si_nir_scan_shader(const struct nir_shader
*nir
,
691 struct tgsi_shader_info
*info
);
692 void si_nir_scan_tess_ctrl(const struct nir_shader
*nir
,
693 const struct tgsi_shader_info
*info
,
694 struct tgsi_tessctrl_info
*out
);
695 void si_lower_nir(struct si_shader_selector
*sel
);
697 /* Inline helpers. */
699 /* Return the pointer to the main shader part's pointer. */
700 static inline struct si_shader
**
701 si_get_main_shader_part(struct si_shader_selector
*sel
,
702 struct si_shader_key
*key
)
705 return &sel
->main_shader_part_ls
;
707 return &sel
->main_shader_part_es
;
708 return &sel
->main_shader_part
;
712 si_shader_uses_bindless_samplers(struct si_shader_selector
*selector
)
714 return selector
? selector
->info
.uses_bindless_samplers
: false;
718 si_shader_uses_bindless_images(struct si_shader_selector
*selector
)
720 return selector
? selector
->info
.uses_bindless_images
: false;
723 void si_destroy_shader_selector(struct si_context
*sctx
,
724 struct si_shader_selector
*sel
);
727 si_shader_selector_reference(struct si_context
*sctx
,
728 struct si_shader_selector
**dst
,
729 struct si_shader_selector
*src
)
731 if (pipe_reference(&(*dst
)->reference
, &src
->reference
))
732 si_destroy_shader_selector(sctx
, *dst
);