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 46
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 /* all VS variants */
178 SI_SGPR_BASE_VERTEX
= SI_NUM_RESOURCE_SGPRS
,
179 SI_SGPR_START_INSTANCE
,
181 SI_SGPR_VS_STATE_BITS
,
184 SI_SGPR_VS_BLIT_DATA
= SI_SGPR_CONST_AND_SHADER_BUFFERS
,
187 SI_SGPR_TES_OFFCHIP_LAYOUT
= SI_NUM_RESOURCE_SGPRS
,
188 SI_SGPR_TES_OFFCHIP_ADDR
,
189 SI_TES_NUM_USER_SGPR
,
191 /* GFX6-8: TCS only */
192 GFX6_SGPR_TCS_OFFCHIP_LAYOUT
= SI_NUM_RESOURCE_SGPRS
,
193 GFX6_SGPR_TCS_OUT_OFFSETS
,
194 GFX6_SGPR_TCS_OUT_LAYOUT
,
195 GFX6_SGPR_TCS_IN_LAYOUT
,
196 GFX6_TCS_NUM_USER_SGPR
,
198 /* GFX9: Merged shaders. */
199 #if HAVE_32BIT_POINTERS
200 /* 2ND_CONST_AND_SHADER_BUFFERS is set in USER_DATA_ADDR_LO (SGPR0). */
201 /* 2ND_SAMPLERS_AND_IMAGES is set in USER_DATA_ADDR_HI (SGPR1). */
202 GFX9_MERGED_NUM_USER_SGPR
= SI_VS_NUM_USER_SGPR
,
204 /* 2ND_CONST_AND_SHADER_BUFFERS is set in USER_DATA_ADDR_LO/HI (SGPR[0:1]). */
205 GFX9_SGPR_2ND_SAMPLERS_AND_IMAGES
= SI_VS_NUM_USER_SGPR
,
206 GFX9_SGPR_2ND_SAMPLERS_AND_IMAGES_HI
,
207 GFX9_MERGED_NUM_USER_SGPR
,
210 /* GFX9: Merged LS-HS (VS-TCS) only. */
211 GFX9_SGPR_TCS_OFFCHIP_LAYOUT
= GFX9_MERGED_NUM_USER_SGPR
,
212 GFX9_SGPR_TCS_OUT_OFFSETS
,
213 GFX9_SGPR_TCS_OUT_LAYOUT
,
214 #if !HAVE_32BIT_POINTERS
215 GFX9_SGPR_align_for_vb_pointer
,
217 GFX9_TCS_NUM_USER_SGPR
,
220 GFX6_GS_NUM_USER_SGPR
= SI_NUM_RESOURCE_SGPRS
,
221 #if HAVE_32BIT_POINTERS
222 GFX9_VSGS_NUM_USER_SGPR
= SI_VS_NUM_USER_SGPR
,
223 GFX9_TESGS_NUM_USER_SGPR
= SI_TES_NUM_USER_SGPR
,
225 GFX9_VSGS_NUM_USER_SGPR
= GFX9_MERGED_NUM_USER_SGPR
,
226 GFX9_TESGS_NUM_USER_SGPR
= GFX9_MERGED_NUM_USER_SGPR
,
228 SI_GSCOPY_NUM_USER_SGPR
= SI_SGPR_RW_BUFFERS
+ (HAVE_32BIT_POINTERS
? 1 : 2),
231 SI_SGPR_ALPHA_REF
= SI_NUM_RESOURCE_SGPRS
,
235 /* LLVM function parameter indices */
237 SI_NUM_RESOURCE_PARAMS
= 4,
239 /* PS only parameters */
240 SI_PARAM_ALPHA_REF
= SI_NUM_RESOURCE_PARAMS
,
242 SI_PARAM_PERSP_SAMPLE
,
243 SI_PARAM_PERSP_CENTER
,
244 SI_PARAM_PERSP_CENTROID
,
245 SI_PARAM_PERSP_PULL_MODEL
,
246 SI_PARAM_LINEAR_SAMPLE
,
247 SI_PARAM_LINEAR_CENTER
,
248 SI_PARAM_LINEAR_CENTROID
,
249 SI_PARAM_LINE_STIPPLE_TEX
,
250 SI_PARAM_POS_X_FLOAT
,
251 SI_PARAM_POS_Y_FLOAT
,
252 SI_PARAM_POS_Z_FLOAT
,
253 SI_PARAM_POS_W_FLOAT
,
256 SI_PARAM_SAMPLE_COVERAGE
,
257 SI_PARAM_POS_FIXED_PT
,
259 SI_NUM_PARAMS
= SI_PARAM_POS_FIXED_PT
+ 9, /* +8 for COLOR[0..1] */
262 /* Fields of driver-defined VS state SGPR. */
263 /* Clamp vertex color output (only used in VS as VS). */
264 #define S_VS_STATE_CLAMP_VERTEX_COLOR(x) (((unsigned)(x) & 0x1) << 0)
265 #define C_VS_STATE_CLAMP_VERTEX_COLOR 0xFFFFFFFE
266 #define S_VS_STATE_INDEXED(x) (((unsigned)(x) & 0x1) << 1)
267 #define C_VS_STATE_INDEXED 0xFFFFFFFD
268 #define S_VS_STATE_LS_OUT_PATCH_SIZE(x) (((unsigned)(x) & 0x1FFF) << 8)
269 #define C_VS_STATE_LS_OUT_PATCH_SIZE 0xFFE000FF
270 #define S_VS_STATE_LS_OUT_VERTEX_SIZE(x) (((unsigned)(x) & 0xFF) << 24)
271 #define C_VS_STATE_LS_OUT_VERTEX_SIZE 0x00FFFFFF
273 /* SI-specific system values. */
275 TGSI_SEMANTIC_DEFAULT_TESSOUTER_SI
= TGSI_SEMANTIC_COUNT
,
276 TGSI_SEMANTIC_DEFAULT_TESSINNER_SI
,
280 /* Use a property enum that VS wouldn't use. */
281 TGSI_PROPERTY_VS_BLIT_SGPRS
= TGSI_PROPERTY_FS_COORD_ORIGIN
,
283 /* These represent the number of SGPRs the shader uses. */
284 SI_VS_BLIT_SGPRS_POS
= 3,
285 SI_VS_BLIT_SGPRS_POS_COLOR
= 7,
286 SI_VS_BLIT_SGPRS_POS_TEXCOORD
= 9,
289 /* For VS shader key fix_fetch. */
291 SI_FIX_FETCH_NONE
= 0,
292 SI_FIX_FETCH_A2_SNORM
,
293 SI_FIX_FETCH_A2_SSCALED
,
294 SI_FIX_FETCH_A2_SINT
,
295 SI_FIX_FETCH_RGBA_32_UNORM
,
296 SI_FIX_FETCH_RGBX_32_UNORM
,
297 SI_FIX_FETCH_RGBA_32_SNORM
,
298 SI_FIX_FETCH_RGBX_32_SNORM
,
299 SI_FIX_FETCH_RGBA_32_USCALED
,
300 SI_FIX_FETCH_RGBA_32_SSCALED
,
301 SI_FIX_FETCH_RGBA_32_FIXED
,
302 SI_FIX_FETCH_RGBX_32_FIXED
,
303 SI_FIX_FETCH_RG_64_FLOAT
,
304 SI_FIX_FETCH_RGB_64_FLOAT
,
305 SI_FIX_FETCH_RGBA_64_FLOAT
,
306 SI_FIX_FETCH_RGB_8
, /* A = 1.0 */
307 SI_FIX_FETCH_RGB_8_INT
, /* A = 1 */
309 SI_FIX_FETCH_RGB_16_INT
,
314 /* State of the context creating the shader object. */
315 struct si_compiler_ctx_state
{
316 /* Should only be used by si_init_shader_selector_async and
317 * si_build_shader_variant if thread_index == -1 (non-threaded). */
318 LLVMTargetMachineRef tm
;
320 /* Used if thread_index == -1 or if debug.async is true. */
321 struct pipe_debug_callback debug
;
323 /* Used for creating the log string for gallium/ddebug. */
324 bool is_debug_context
;
327 /* A shader selector is a gallium CSO and contains shader variants and
328 * binaries for one TGSI program. This can be shared by multiple contexts.
330 struct si_shader_selector
{
331 struct pipe_reference reference
;
332 struct si_screen
*screen
;
333 struct util_queue_fence ready
;
334 struct si_compiler_ctx_state compiler_ctx_state
;
337 struct si_shader
*first_variant
; /* immutable after the first variant */
338 struct si_shader
*last_variant
; /* mutable */
340 /* The compiled TGSI shader expecting a prolog and/or epilog (not
341 * uploaded to a buffer).
343 struct si_shader
*main_shader_part
;
344 struct si_shader
*main_shader_part_ls
; /* as_ls is set in the key */
345 struct si_shader
*main_shader_part_es
; /* as_es is set in the key */
347 struct si_shader
*gs_copy_shader
;
349 struct tgsi_token
*tokens
;
350 struct nir_shader
*nir
;
351 struct pipe_stream_output_info so
;
352 struct tgsi_shader_info info
;
353 struct tgsi_tessctrl_info tcs_info
;
355 /* PIPE_SHADER_[VERTEX|FRAGMENT|...] */
357 bool vs_needs_prolog
;
358 bool force_correct_derivs_after_kill
;
359 unsigned pa_cl_vs_out_cntl
;
364 unsigned esgs_itemsize
;
367 unsigned gs_input_verts_per_prim
;
368 unsigned gs_output_prim
;
369 unsigned gs_max_out_vertices
;
370 unsigned gs_num_invocations
;
371 unsigned max_gs_stream
; /* count - 1 */
372 unsigned gsvs_vertex_size
;
373 unsigned max_gsvs_emit_size
;
374 unsigned enabled_streamout_buffer_mask
;
377 unsigned color_attr_index
[2];
378 unsigned db_shader_control
;
379 /* Set 0xf or 0x0 (4 bits) per each written output.
380 * ANDed with spi_shader_col_format.
382 unsigned colors_written_4bit
;
387 uint64_t outputs_written
; /* "get_unique_index" bits */
388 uint32_t patch_outputs_written
; /* "get_unique_index_patch" bits */
390 uint64_t inputs_read
; /* "get_unique_index" bits */
392 /* bitmasks of used descriptor slots */
393 uint32_t active_const_and_shader_buffers
;
394 uint64_t active_samplers_and_images
;
397 /* Valid shader configurations:
399 * API shaders VS | TCS | TES | GS |pass| PS
400 * are compiled as: | | | |thru|
402 * Only VS & PS: VS | | | | | PS
403 * GFX6 - with GS: ES | | | GS | VS | PS
404 * - with tess: LS | HS | VS | | | PS
405 * - with both: LS | HS | ES | GS | VS | PS
406 * GFX9 - with GS: -> | | | GS | VS | PS
407 * - with tess: -> | HS | VS | | | PS
408 * - with both: -> | HS | -> | GS | VS | PS
410 * -> = merged with the next stage
413 /* Use the byte alignment for all following structure members for optimal
414 * shader key memory footprint.
416 #pragma pack(push, 1)
418 /* Common VS bits between the shader key and the prolog key. */
419 struct si_vs_prolog_bits
{
420 /* - If neither "is_one" nor "is_fetched" has a bit set, the instance
422 * - If "is_one" has a bit set, the instance divisor is 1.
423 * - If "is_fetched" has a bit set, the instance divisor will be loaded
424 * from the constant buffer.
426 uint16_t instance_divisor_is_one
; /* bitmask of inputs */
427 uint16_t instance_divisor_is_fetched
; /* bitmask of inputs */
428 unsigned ls_vgpr_fix
:1;
431 /* Common TCS bits between the shader key and the epilog key. */
432 struct si_tcs_epilog_bits
{
433 unsigned prim_mode
:3;
434 unsigned invoc0_tess_factors_are_def
:1;
435 unsigned tes_reads_tess_factors
:1;
438 struct si_gs_prolog_bits
{
439 unsigned tri_strip_adj_fix
:1;
440 unsigned gfx9_prev_is_vs
:1;
443 /* Common PS bits between the shader key and the prolog key. */
444 struct si_ps_prolog_bits
{
445 unsigned color_two_side
:1;
446 unsigned flatshade_colors
:1;
447 unsigned poly_stipple
:1;
448 unsigned force_persp_sample_interp
:1;
449 unsigned force_linear_sample_interp
:1;
450 unsigned force_persp_center_interp
:1;
451 unsigned force_linear_center_interp
:1;
452 unsigned bc_optimize_for_persp
:1;
453 unsigned bc_optimize_for_linear
:1;
454 unsigned samplemask_log_ps_iter
:3;
457 /* Common PS bits between the shader key and the epilog key. */
458 struct si_ps_epilog_bits
{
459 unsigned spi_shader_col_format
;
460 unsigned color_is_int8
:8;
461 unsigned color_is_int10
:8;
462 unsigned last_cbuf
:3;
463 unsigned alpha_func
:3;
464 unsigned alpha_to_one
:1;
465 unsigned poly_line_smoothing
:1;
466 unsigned clamp_color
:1;
469 union si_shader_part_key
{
471 struct si_vs_prolog_bits states
;
472 unsigned num_input_sgprs
:6;
473 /* For merged stages such as LS-HS, HS input VGPRs are first. */
474 unsigned num_merged_next_stage_vgprs
:3;
475 unsigned last_input
:4;
478 /* Prologs for monolithic shaders shouldn't set EXEC. */
479 unsigned is_monolithic
:1;
482 struct si_tcs_epilog_bits states
;
485 struct si_gs_prolog_bits states
;
486 /* Prologs of monolithic shaders shouldn't set EXEC. */
487 unsigned is_monolithic
:1;
490 struct si_ps_prolog_bits states
;
491 unsigned num_input_sgprs
:6;
492 unsigned num_input_vgprs
:5;
493 /* Color interpolation and two-side color selection. */
494 unsigned colors_read
:8; /* color input components read */
495 unsigned num_interp_inputs
:5; /* BCOLOR is at this location */
496 unsigned face_vgpr_index
:5;
497 unsigned ancillary_vgpr_index
:5;
499 char color_attr_index
[2];
500 char color_interp_vgpr_index
[2]; /* -1 == constant */
503 struct si_ps_epilog_bits states
;
504 unsigned colors_written
:8;
506 unsigned writes_stencil
:1;
507 unsigned writes_samplemask
:1;
511 struct si_shader_key
{
512 /* Prolog and epilog flags. */
515 struct si_vs_prolog_bits prolog
;
518 struct si_vs_prolog_bits ls_prolog
; /* for merged LS-HS */
519 struct si_shader_selector
*ls
; /* for merged LS-HS */
520 struct si_tcs_epilog_bits epilog
;
521 } tcs
; /* tessellation control shader */
523 struct si_vs_prolog_bits vs_prolog
; /* for merged ES-GS */
524 struct si_shader_selector
*es
; /* for merged ES-GS */
525 struct si_gs_prolog_bits prolog
;
528 struct si_ps_prolog_bits prolog
;
529 struct si_ps_epilog_bits epilog
;
533 /* These two are initially set according to the NEXT_SHADER property,
534 * or guessed if the property doesn't seem correct.
536 unsigned as_es
:1; /* export shader, which precedes GS */
537 unsigned as_ls
:1; /* local shader, which precedes TCS */
539 /* Flags for monolithic compilation only. */
541 /* One byte for every input: SI_FIX_FETCH_* enums. */
542 uint8_t vs_fix_fetch
[SI_MAX_ATTRIBS
];
545 uint64_t ff_tcs_inputs_to_copy
; /* for fixed-func TCS */
546 /* When PS needs PrimID and GS is disabled. */
547 unsigned vs_export_prim_id
:1;
549 unsigned interpolate_at_sample_force_center
:1;
550 unsigned fbfetch_msaa
;
551 unsigned fbfetch_is_1D
;
552 unsigned fbfetch_layered
;
557 /* Optimization flags for asynchronous compilation only. */
559 /* For HW VS (it can be VS, TES, GS) */
560 uint64_t kill_outputs
; /* "get_unique_index" bits */
561 unsigned clip_disable
:1;
563 /* For shaders where monolithic variants have better code.
565 * This is a flag that has no effect on code generation,
566 * but forces monolithic shaders to be used as soon as
567 * possible, because it's in the "opt" group.
569 unsigned prefer_mono
:1;
573 /* Restore the pack alignment to default. */
576 struct si_shader_config
{
579 unsigned spilled_sgprs
;
580 unsigned spilled_vgprs
;
581 unsigned private_mem_vgprs
;
583 unsigned max_simd_waves
;
584 unsigned spi_ps_input_ena
;
585 unsigned spi_ps_input_addr
;
587 unsigned scratch_bytes_per_wave
;
592 /* GCN-specific shader info. */
593 struct si_shader_info
{
594 ubyte vs_output_param_offset
[SI_MAX_VS_OUTPUTS
];
595 ubyte num_input_sgprs
;
596 ubyte num_input_vgprs
;
597 signed char face_vgpr_index
;
598 signed char ancillary_vgpr_index
;
599 bool uses_instanceid
;
600 ubyte nr_pos_exports
;
601 ubyte nr_param_exports
;
605 struct si_compiler_ctx_state compiler_ctx_state
;
607 struct si_shader_selector
*selector
;
608 struct si_shader_selector
*previous_stage_sel
; /* for refcounting */
609 struct si_shader
*next_variant
;
611 struct si_shader_part
*prolog
;
612 struct si_shader
*previous_stage
; /* for GFX9 */
613 struct si_shader_part
*prolog2
;
614 struct si_shader_part
*epilog
;
616 struct si_pm4_state
*pm4
;
617 struct r600_resource
*bo
;
618 struct r600_resource
*scratch_bo
;
619 struct si_shader_key key
;
620 struct util_queue_fence ready
;
621 bool compilation_failed
;
624 bool is_binary_shared
;
625 bool is_gs_copy_shader
;
627 /* The following data is all that's needed for binary shaders. */
628 struct ac_shader_binary binary
;
629 struct si_shader_config config
;
630 struct si_shader_info info
;
632 /* Shader key + LLVM IR + disassembly + statistics.
633 * Generated for debug contexts only.
636 size_t shader_log_size
;
639 struct si_shader_part
{
640 struct si_shader_part
*next
;
641 union si_shader_part_key key
;
642 struct ac_shader_binary binary
;
643 struct si_shader_config config
;
648 si_generate_gs_copy_shader(struct si_screen
*sscreen
,
649 LLVMTargetMachineRef tm
,
650 struct si_shader_selector
*gs_selector
,
651 struct pipe_debug_callback
*debug
);
652 int si_compile_tgsi_shader(struct si_screen
*sscreen
,
653 LLVMTargetMachineRef tm
,
654 struct si_shader
*shader
,
656 struct pipe_debug_callback
*debug
);
657 int si_shader_create(struct si_screen
*sscreen
, LLVMTargetMachineRef tm
,
658 struct si_shader
*shader
,
659 struct pipe_debug_callback
*debug
);
660 void si_shader_destroy(struct si_shader
*shader
);
661 unsigned si_shader_io_get_unique_index_patch(unsigned semantic_name
, unsigned index
);
662 unsigned si_shader_io_get_unique_index(unsigned semantic_name
, unsigned index
);
663 int si_shader_binary_upload(struct si_screen
*sscreen
, struct si_shader
*shader
);
664 void si_shader_dump(struct si_screen
*sscreen
, const struct si_shader
*shader
,
665 struct pipe_debug_callback
*debug
, unsigned processor
,
666 FILE *f
, bool check_debug_option
);
667 void si_shader_dump_stats_for_shader_db(const struct si_shader
*shader
,
668 struct pipe_debug_callback
*debug
);
669 void si_multiwave_lds_size_workaround(struct si_screen
*sscreen
,
671 void si_shader_apply_scratch_relocs(struct si_shader
*shader
,
672 uint64_t scratch_va
);
673 void si_shader_binary_read_config(struct ac_shader_binary
*binary
,
674 struct si_shader_config
*conf
,
675 unsigned symbol_offset
);
676 const char *si_get_shader_name(const struct si_shader
*shader
, unsigned processor
);
678 /* si_shader_nir.c */
679 void si_nir_scan_shader(const struct nir_shader
*nir
,
680 struct tgsi_shader_info
*info
);
681 void si_nir_scan_tess_ctrl(const struct nir_shader
*nir
,
682 const struct tgsi_shader_info
*info
,
683 struct tgsi_tessctrl_info
*out
);
684 void si_lower_nir(struct si_shader_selector
*sel
);
686 /* Inline helpers. */
688 /* Return the pointer to the main shader part's pointer. */
689 static inline struct si_shader
**
690 si_get_main_shader_part(struct si_shader_selector
*sel
,
691 struct si_shader_key
*key
)
694 return &sel
->main_shader_part_ls
;
696 return &sel
->main_shader_part_es
;
697 return &sel
->main_shader_part
;
701 si_shader_uses_bindless_samplers(struct si_shader_selector
*selector
)
703 return selector
? selector
->info
.uses_bindless_samplers
: false;
707 si_shader_uses_bindless_images(struct si_shader_selector
*selector
)
709 return selector
? selector
->info
.uses_bindless_images
: false;
712 void si_destroy_shader_selector(struct si_context
*sctx
,
713 struct si_shader_selector
*sel
);
716 si_shader_selector_reference(struct si_context
*sctx
,
717 struct si_shader_selector
**dst
,
718 struct si_shader_selector
*src
)
720 if (pipe_reference(&(*dst
)->reference
, &src
->reference
))
721 si_destroy_shader_selector(sctx
, *dst
);