2 * Copyright © 2010 Intel Corporation
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 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * 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 NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 #include "brw_context.h"
29 #include "brw_vec4_tes.h"
30 #include "main/shaderobj.h"
31 #include "main/uniforms.h"
33 extern "C" struct gl_shader
*
34 brw_new_shader(struct gl_context
*ctx
, GLuint name
, GLuint type
)
36 struct brw_shader
*shader
;
38 shader
= rzalloc(NULL
, struct brw_shader
);
40 shader
->base
.Type
= type
;
41 shader
->base
.Stage
= _mesa_shader_enum_to_shader_stage(type
);
42 shader
->base
.Name
= name
;
43 _mesa_init_shader(ctx
, &shader
->base
);
50 brw_mark_surface_used(struct brw_stage_prog_data
*prog_data
,
53 assert(surf_index
< BRW_MAX_SURFACES
);
55 prog_data
->binding_table
.size_bytes
=
56 MAX2(prog_data
->binding_table
.size_bytes
, (surf_index
+ 1) * 4);
60 brw_type_for_base_type(const struct glsl_type
*type
)
62 switch (type
->base_type
) {
64 return BRW_REGISTER_TYPE_F
;
67 case GLSL_TYPE_SUBROUTINE
:
68 return BRW_REGISTER_TYPE_D
;
70 return BRW_REGISTER_TYPE_UD
;
72 return brw_type_for_base_type(type
->fields
.array
);
73 case GLSL_TYPE_STRUCT
:
74 case GLSL_TYPE_SAMPLER
:
75 case GLSL_TYPE_ATOMIC_UINT
:
76 /* These should be overridden with the type of the member when
77 * dereferenced into. BRW_REGISTER_TYPE_UD seems like a likely
78 * way to trip up if we don't.
80 return BRW_REGISTER_TYPE_UD
;
82 return BRW_REGISTER_TYPE_UD
;
83 case GLSL_TYPE_DOUBLE
:
84 return BRW_REGISTER_TYPE_DF
;
87 case GLSL_TYPE_INTERFACE
:
88 case GLSL_TYPE_FUNCTION
:
89 unreachable("not reached");
92 return BRW_REGISTER_TYPE_F
;
95 enum brw_conditional_mod
96 brw_conditional_for_comparison(unsigned int op
)
100 return BRW_CONDITIONAL_L
;
101 case ir_binop_greater
:
102 return BRW_CONDITIONAL_G
;
103 case ir_binop_lequal
:
104 return BRW_CONDITIONAL_LE
;
105 case ir_binop_gequal
:
106 return BRW_CONDITIONAL_GE
;
108 case ir_binop_all_equal
: /* same as equal for scalars */
109 return BRW_CONDITIONAL_Z
;
110 case ir_binop_nequal
:
111 case ir_binop_any_nequal
: /* same as nequal for scalars */
112 return BRW_CONDITIONAL_NZ
;
114 unreachable("not reached: bad operation for comparison");
119 brw_math_function(enum opcode op
)
122 case SHADER_OPCODE_RCP
:
123 return BRW_MATH_FUNCTION_INV
;
124 case SHADER_OPCODE_RSQ
:
125 return BRW_MATH_FUNCTION_RSQ
;
126 case SHADER_OPCODE_SQRT
:
127 return BRW_MATH_FUNCTION_SQRT
;
128 case SHADER_OPCODE_EXP2
:
129 return BRW_MATH_FUNCTION_EXP
;
130 case SHADER_OPCODE_LOG2
:
131 return BRW_MATH_FUNCTION_LOG
;
132 case SHADER_OPCODE_POW
:
133 return BRW_MATH_FUNCTION_POW
;
134 case SHADER_OPCODE_SIN
:
135 return BRW_MATH_FUNCTION_SIN
;
136 case SHADER_OPCODE_COS
:
137 return BRW_MATH_FUNCTION_COS
;
138 case SHADER_OPCODE_INT_QUOTIENT
:
139 return BRW_MATH_FUNCTION_INT_DIV_QUOTIENT
;
140 case SHADER_OPCODE_INT_REMAINDER
:
141 return BRW_MATH_FUNCTION_INT_DIV_REMAINDER
;
143 unreachable("not reached: unknown math function");
148 brw_texture_offset(int *offsets
, unsigned num_components
)
150 if (!offsets
) return 0; /* nonconstant offset; caller will handle it. */
152 /* Combine all three offsets into a single unsigned dword:
154 * bits 11:8 - U Offset (X component)
155 * bits 7:4 - V Offset (Y component)
156 * bits 3:0 - R Offset (Z component)
158 unsigned offset_bits
= 0;
159 for (unsigned i
= 0; i
< num_components
; i
++) {
160 const unsigned shift
= 4 * (2 - i
);
161 offset_bits
|= (offsets
[i
] << shift
) & (0xF << shift
);
167 brw_instruction_name(const struct brw_device_info
*devinfo
, enum opcode op
)
170 case BRW_OPCODE_ILLEGAL
... BRW_OPCODE_NOP
:
171 assert(brw_opcode_desc(devinfo
, op
)->name
);
172 return brw_opcode_desc(devinfo
, op
)->name
;
173 case FS_OPCODE_FB_WRITE
:
175 case FS_OPCODE_FB_WRITE_LOGICAL
:
176 return "fb_write_logical";
177 case FS_OPCODE_PACK_STENCIL_REF
:
178 return "pack_stencil_ref";
179 case FS_OPCODE_BLORP_FB_WRITE
:
180 return "blorp_fb_write";
181 case FS_OPCODE_REP_FB_WRITE
:
182 return "rep_fb_write";
184 case SHADER_OPCODE_RCP
:
186 case SHADER_OPCODE_RSQ
:
188 case SHADER_OPCODE_SQRT
:
190 case SHADER_OPCODE_EXP2
:
192 case SHADER_OPCODE_LOG2
:
194 case SHADER_OPCODE_POW
:
196 case SHADER_OPCODE_INT_QUOTIENT
:
198 case SHADER_OPCODE_INT_REMAINDER
:
200 case SHADER_OPCODE_SIN
:
202 case SHADER_OPCODE_COS
:
205 case SHADER_OPCODE_TEX
:
207 case SHADER_OPCODE_TEX_LOGICAL
:
208 return "tex_logical";
209 case SHADER_OPCODE_TXD
:
211 case SHADER_OPCODE_TXD_LOGICAL
:
212 return "txd_logical";
213 case SHADER_OPCODE_TXF
:
215 case SHADER_OPCODE_TXF_LOGICAL
:
216 return "txf_logical";
217 case SHADER_OPCODE_TXL
:
219 case SHADER_OPCODE_TXL_LOGICAL
:
220 return "txl_logical";
221 case SHADER_OPCODE_TXS
:
223 case SHADER_OPCODE_TXS_LOGICAL
:
224 return "txs_logical";
227 case FS_OPCODE_TXB_LOGICAL
:
228 return "txb_logical";
229 case SHADER_OPCODE_TXF_CMS
:
231 case SHADER_OPCODE_TXF_CMS_LOGICAL
:
232 return "txf_cms_logical";
233 case SHADER_OPCODE_TXF_CMS_W
:
235 case SHADER_OPCODE_TXF_CMS_W_LOGICAL
:
236 return "txf_cms_w_logical";
237 case SHADER_OPCODE_TXF_UMS
:
239 case SHADER_OPCODE_TXF_UMS_LOGICAL
:
240 return "txf_ums_logical";
241 case SHADER_OPCODE_TXF_MCS
:
243 case SHADER_OPCODE_TXF_MCS_LOGICAL
:
244 return "txf_mcs_logical";
245 case SHADER_OPCODE_LOD
:
247 case SHADER_OPCODE_LOD_LOGICAL
:
248 return "lod_logical";
249 case SHADER_OPCODE_TG4
:
251 case SHADER_OPCODE_TG4_LOGICAL
:
252 return "tg4_logical";
253 case SHADER_OPCODE_TG4_OFFSET
:
255 case SHADER_OPCODE_TG4_OFFSET_LOGICAL
:
256 return "tg4_offset_logical";
257 case SHADER_OPCODE_SAMPLEINFO
:
260 case SHADER_OPCODE_SHADER_TIME_ADD
:
261 return "shader_time_add";
263 case SHADER_OPCODE_UNTYPED_ATOMIC
:
264 return "untyped_atomic";
265 case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL
:
266 return "untyped_atomic_logical";
267 case SHADER_OPCODE_UNTYPED_SURFACE_READ
:
268 return "untyped_surface_read";
269 case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL
:
270 return "untyped_surface_read_logical";
271 case SHADER_OPCODE_UNTYPED_SURFACE_WRITE
:
272 return "untyped_surface_write";
273 case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL
:
274 return "untyped_surface_write_logical";
275 case SHADER_OPCODE_TYPED_ATOMIC
:
276 return "typed_atomic";
277 case SHADER_OPCODE_TYPED_ATOMIC_LOGICAL
:
278 return "typed_atomic_logical";
279 case SHADER_OPCODE_TYPED_SURFACE_READ
:
280 return "typed_surface_read";
281 case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL
:
282 return "typed_surface_read_logical";
283 case SHADER_OPCODE_TYPED_SURFACE_WRITE
:
284 return "typed_surface_write";
285 case SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL
:
286 return "typed_surface_write_logical";
287 case SHADER_OPCODE_MEMORY_FENCE
:
288 return "memory_fence";
290 case SHADER_OPCODE_LOAD_PAYLOAD
:
291 return "load_payload";
295 case SHADER_OPCODE_GEN4_SCRATCH_READ
:
296 return "gen4_scratch_read";
297 case SHADER_OPCODE_GEN4_SCRATCH_WRITE
:
298 return "gen4_scratch_write";
299 case SHADER_OPCODE_GEN7_SCRATCH_READ
:
300 return "gen7_scratch_read";
301 case SHADER_OPCODE_URB_WRITE_SIMD8
:
302 return "gen8_urb_write_simd8";
303 case SHADER_OPCODE_URB_WRITE_SIMD8_PER_SLOT
:
304 return "gen8_urb_write_simd8_per_slot";
305 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED
:
306 return "gen8_urb_write_simd8_masked";
307 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT
:
308 return "gen8_urb_write_simd8_masked_per_slot";
309 case SHADER_OPCODE_URB_READ_SIMD8
:
310 return "urb_read_simd8";
311 case SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT
:
312 return "urb_read_simd8_per_slot";
314 case SHADER_OPCODE_FIND_LIVE_CHANNEL
:
315 return "find_live_channel";
316 case SHADER_OPCODE_BROADCAST
:
319 case SHADER_OPCODE_EXTRACT_BYTE
:
320 return "extract_byte";
321 case SHADER_OPCODE_EXTRACT_WORD
:
322 return "extract_word";
323 case VEC4_OPCODE_MOV_BYTES
:
325 case VEC4_OPCODE_PACK_BYTES
:
327 case VEC4_OPCODE_UNPACK_UNIFORM
:
328 return "unpack_uniform";
330 case FS_OPCODE_DDX_COARSE
:
332 case FS_OPCODE_DDX_FINE
:
334 case FS_OPCODE_DDY_COARSE
:
336 case FS_OPCODE_DDY_FINE
:
339 case FS_OPCODE_CINTERP
:
341 case FS_OPCODE_LINTERP
:
344 case FS_OPCODE_PIXEL_X
:
346 case FS_OPCODE_PIXEL_Y
:
349 case FS_OPCODE_GET_BUFFER_SIZE
:
350 return "fs_get_buffer_size";
352 case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD
:
353 return "uniform_pull_const";
354 case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD_GEN7
:
355 return "uniform_pull_const_gen7";
356 case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD
:
357 return "varying_pull_const";
358 case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7
:
359 return "varying_pull_const_gen7";
361 case FS_OPCODE_MOV_DISPATCH_TO_FLAGS
:
362 return "mov_dispatch_to_flags";
363 case FS_OPCODE_DISCARD_JUMP
:
364 return "discard_jump";
366 case FS_OPCODE_SET_SAMPLE_ID
:
367 return "set_sample_id";
368 case FS_OPCODE_SET_SIMD4X2_OFFSET
:
369 return "set_simd4x2_offset";
371 case FS_OPCODE_PACK_HALF_2x16_SPLIT
:
372 return "pack_half_2x16_split";
373 case FS_OPCODE_UNPACK_HALF_2x16_SPLIT_X
:
374 return "unpack_half_2x16_split_x";
375 case FS_OPCODE_UNPACK_HALF_2x16_SPLIT_Y
:
376 return "unpack_half_2x16_split_y";
378 case FS_OPCODE_PLACEHOLDER_HALT
:
379 return "placeholder_halt";
381 case FS_OPCODE_INTERPOLATE_AT_CENTROID
:
382 return "interp_centroid";
383 case FS_OPCODE_INTERPOLATE_AT_SAMPLE
:
384 return "interp_sample";
385 case FS_OPCODE_INTERPOLATE_AT_SHARED_OFFSET
:
386 return "interp_shared_offset";
387 case FS_OPCODE_INTERPOLATE_AT_PER_SLOT_OFFSET
:
388 return "interp_per_slot_offset";
390 case VS_OPCODE_URB_WRITE
:
391 return "vs_urb_write";
392 case VS_OPCODE_PULL_CONSTANT_LOAD
:
393 return "pull_constant_load";
394 case VS_OPCODE_PULL_CONSTANT_LOAD_GEN7
:
395 return "pull_constant_load_gen7";
397 case VS_OPCODE_SET_SIMD4X2_HEADER_GEN9
:
398 return "set_simd4x2_header_gen9";
400 case VS_OPCODE_GET_BUFFER_SIZE
:
401 return "vs_get_buffer_size";
403 case VS_OPCODE_UNPACK_FLAGS_SIMD4X2
:
404 return "unpack_flags_simd4x2";
406 case GS_OPCODE_URB_WRITE
:
407 return "gs_urb_write";
408 case GS_OPCODE_URB_WRITE_ALLOCATE
:
409 return "gs_urb_write_allocate";
410 case GS_OPCODE_THREAD_END
:
411 return "gs_thread_end";
412 case GS_OPCODE_SET_WRITE_OFFSET
:
413 return "set_write_offset";
414 case GS_OPCODE_SET_VERTEX_COUNT
:
415 return "set_vertex_count";
416 case GS_OPCODE_SET_DWORD_2
:
417 return "set_dword_2";
418 case GS_OPCODE_PREPARE_CHANNEL_MASKS
:
419 return "prepare_channel_masks";
420 case GS_OPCODE_SET_CHANNEL_MASKS
:
421 return "set_channel_masks";
422 case GS_OPCODE_GET_INSTANCE_ID
:
423 return "get_instance_id";
424 case GS_OPCODE_FF_SYNC
:
426 case GS_OPCODE_SET_PRIMITIVE_ID
:
427 return "set_primitive_id";
428 case GS_OPCODE_SVB_WRITE
:
429 return "gs_svb_write";
430 case GS_OPCODE_SVB_SET_DST_INDEX
:
431 return "gs_svb_set_dst_index";
432 case GS_OPCODE_FF_SYNC_SET_PRIMITIVES
:
433 return "gs_ff_sync_set_primitives";
434 case CS_OPCODE_CS_TERMINATE
:
435 return "cs_terminate";
436 case SHADER_OPCODE_BARRIER
:
438 case SHADER_OPCODE_MULH
:
440 case SHADER_OPCODE_MOV_INDIRECT
:
441 return "mov_indirect";
443 case VEC4_OPCODE_URB_READ
:
445 case TCS_OPCODE_GET_INSTANCE_ID
:
446 return "tcs_get_instance_id";
447 case TCS_OPCODE_URB_WRITE
:
448 return "tcs_urb_write";
449 case TCS_OPCODE_SET_INPUT_URB_OFFSETS
:
450 return "tcs_set_input_urb_offsets";
451 case TCS_OPCODE_SET_OUTPUT_URB_OFFSETS
:
452 return "tcs_set_output_urb_offsets";
453 case TCS_OPCODE_GET_PRIMITIVE_ID
:
454 return "tcs_get_primitive_id";
455 case TCS_OPCODE_CREATE_BARRIER_HEADER
:
456 return "tcs_create_barrier_header";
457 case TCS_OPCODE_SRC0_010_IS_ZERO
:
458 return "tcs_src0<0,1,0>_is_zero";
459 case TCS_OPCODE_RELEASE_INPUT
:
460 return "tcs_release_input";
461 case TCS_OPCODE_THREAD_END
:
462 return "tcs_thread_end";
463 case TES_OPCODE_CREATE_INPUT_READ_HEADER
:
464 return "tes_create_input_read_header";
465 case TES_OPCODE_ADD_INDIRECT_URB_OFFSET
:
466 return "tes_add_indirect_urb_offset";
467 case TES_OPCODE_GET_PRIMITIVE_ID
:
468 return "tes_get_primitive_id";
471 unreachable("not reached");
475 brw_saturate_immediate(enum brw_reg_type type
, struct brw_reg
*reg
)
482 } imm
, sat_imm
= { 0 };
484 const unsigned size
= type_sz(type
);
486 /* We want to either do a 32-bit or 64-bit data copy, the type is otherwise
487 * irrelevant, so just check the size of the type and copy from/to an
488 * appropriately sized field.
496 case BRW_REGISTER_TYPE_UD
:
497 case BRW_REGISTER_TYPE_D
:
498 case BRW_REGISTER_TYPE_UW
:
499 case BRW_REGISTER_TYPE_W
:
500 case BRW_REGISTER_TYPE_UQ
:
501 case BRW_REGISTER_TYPE_Q
:
504 case BRW_REGISTER_TYPE_F
:
505 sat_imm
.f
= CLAMP(imm
.f
, 0.0f
, 1.0f
);
507 case BRW_REGISTER_TYPE_DF
:
508 sat_imm
.df
= CLAMP(imm
.df
, 0.0, 1.0);
510 case BRW_REGISTER_TYPE_UB
:
511 case BRW_REGISTER_TYPE_B
:
512 unreachable("no UB/B immediates");
513 case BRW_REGISTER_TYPE_V
:
514 case BRW_REGISTER_TYPE_UV
:
515 case BRW_REGISTER_TYPE_VF
:
516 unreachable("unimplemented: saturate vector immediate");
517 case BRW_REGISTER_TYPE_HF
:
518 unreachable("unimplemented: saturate HF immediate");
522 if (imm
.ud
!= sat_imm
.ud
) {
523 reg
->ud
= sat_imm
.ud
;
527 if (imm
.df
!= sat_imm
.df
) {
528 reg
->df
= sat_imm
.df
;
536 brw_negate_immediate(enum brw_reg_type type
, struct brw_reg
*reg
)
539 case BRW_REGISTER_TYPE_D
:
540 case BRW_REGISTER_TYPE_UD
:
543 case BRW_REGISTER_TYPE_W
:
544 case BRW_REGISTER_TYPE_UW
:
545 reg
->d
= -(int16_t)reg
->ud
;
547 case BRW_REGISTER_TYPE_F
:
550 case BRW_REGISTER_TYPE_VF
:
551 reg
->ud
^= 0x80808080;
553 case BRW_REGISTER_TYPE_DF
:
556 case BRW_REGISTER_TYPE_UB
:
557 case BRW_REGISTER_TYPE_B
:
558 unreachable("no UB/B immediates");
559 case BRW_REGISTER_TYPE_UV
:
560 case BRW_REGISTER_TYPE_V
:
561 assert(!"unimplemented: negate UV/V immediate");
562 case BRW_REGISTER_TYPE_UQ
:
563 case BRW_REGISTER_TYPE_Q
:
564 assert(!"unimplemented: negate UQ/Q immediate");
565 case BRW_REGISTER_TYPE_HF
:
566 assert(!"unimplemented: negate HF immediate");
573 brw_abs_immediate(enum brw_reg_type type
, struct brw_reg
*reg
)
576 case BRW_REGISTER_TYPE_D
:
577 reg
->d
= abs(reg
->d
);
579 case BRW_REGISTER_TYPE_W
:
580 reg
->d
= abs((int16_t)reg
->ud
);
582 case BRW_REGISTER_TYPE_F
:
583 reg
->f
= fabsf(reg
->f
);
585 case BRW_REGISTER_TYPE_DF
:
586 reg
->df
= fabs(reg
->df
);
588 case BRW_REGISTER_TYPE_VF
:
589 reg
->ud
&= ~0x80808080;
591 case BRW_REGISTER_TYPE_UB
:
592 case BRW_REGISTER_TYPE_B
:
593 unreachable("no UB/B immediates");
594 case BRW_REGISTER_TYPE_UQ
:
595 case BRW_REGISTER_TYPE_UD
:
596 case BRW_REGISTER_TYPE_UW
:
597 case BRW_REGISTER_TYPE_UV
:
598 /* Presumably the absolute value modifier on an unsigned source is a
599 * nop, but it would be nice to confirm.
601 assert(!"unimplemented: abs unsigned immediate");
602 case BRW_REGISTER_TYPE_V
:
603 assert(!"unimplemented: abs V immediate");
604 case BRW_REGISTER_TYPE_Q
:
605 assert(!"unimplemented: abs Q immediate");
606 case BRW_REGISTER_TYPE_HF
:
607 assert(!"unimplemented: abs HF immediate");
614 tesslevel_outer_components(GLenum tes_primitive_mode
)
616 switch (tes_primitive_mode
) {
624 unreachable("Bogus tessellation domain");
630 tesslevel_inner_components(GLenum tes_primitive_mode
)
632 switch (tes_primitive_mode
) {
640 unreachable("Bogus tessellation domain");
646 * Given a normal .xyzw writemask, convert it to a writemask for a vector
647 * that's stored backwards, i.e. .wzyx.
650 writemask_for_backwards_vector(unsigned mask
)
652 unsigned new_mask
= 0;
654 for (int i
= 0; i
< 4; i
++)
655 new_mask
|= ((mask
>> i
) & 1) << (3 - i
);
660 backend_shader::backend_shader(const struct brw_compiler
*compiler
,
663 const nir_shader
*shader
,
664 struct brw_stage_prog_data
*stage_prog_data
)
665 : compiler(compiler
),
667 devinfo(compiler
->devinfo
),
669 stage_prog_data(stage_prog_data
),
674 debug_enabled
= INTEL_DEBUG
& intel_debug_flag_for_shader_stage(stage
);
675 stage_name
= _mesa_shader_stage_to_string(stage
);
676 stage_abbrev
= _mesa_shader_stage_to_abbrev(stage
);
677 is_passthrough_shader
=
678 nir
->info
.name
&& strcmp(nir
->info
.name
, "passthrough") == 0;
682 backend_reg::equals(const backend_reg
&r
) const
684 return memcmp((brw_reg
*)this, (brw_reg
*)&r
, sizeof(brw_reg
)) == 0 &&
685 reg_offset
== r
.reg_offset
;
689 backend_reg::is_zero() const
695 case BRW_REGISTER_TYPE_F
:
697 case BRW_REGISTER_TYPE_DF
:
699 case BRW_REGISTER_TYPE_D
:
700 case BRW_REGISTER_TYPE_UD
:
708 backend_reg::is_one() const
714 case BRW_REGISTER_TYPE_F
:
716 case BRW_REGISTER_TYPE_DF
:
718 case BRW_REGISTER_TYPE_D
:
719 case BRW_REGISTER_TYPE_UD
:
727 backend_reg::is_negative_one() const
733 case BRW_REGISTER_TYPE_F
:
735 case BRW_REGISTER_TYPE_DF
:
737 case BRW_REGISTER_TYPE_D
:
745 backend_reg::is_null() const
747 return file
== ARF
&& nr
== BRW_ARF_NULL
;
752 backend_reg::is_accumulator() const
754 return file
== ARF
&& nr
== BRW_ARF_ACCUMULATOR
;
758 backend_reg::in_range(const backend_reg
&r
, unsigned n
) const
760 return (file
== r
.file
&&
762 reg_offset
>= r
.reg_offset
&&
763 reg_offset
< r
.reg_offset
+ n
);
767 backend_instruction::is_commutative() const
775 case SHADER_OPCODE_MULH
:
778 /* MIN and MAX are commutative. */
779 if (conditional_mod
== BRW_CONDITIONAL_GE
||
780 conditional_mod
== BRW_CONDITIONAL_L
) {
790 backend_instruction::is_3src(const struct brw_device_info
*devinfo
) const
792 return ::is_3src(devinfo
, opcode
);
796 backend_instruction::is_tex() const
798 return (opcode
== SHADER_OPCODE_TEX
||
799 opcode
== FS_OPCODE_TXB
||
800 opcode
== SHADER_OPCODE_TXD
||
801 opcode
== SHADER_OPCODE_TXF
||
802 opcode
== SHADER_OPCODE_TXF_CMS
||
803 opcode
== SHADER_OPCODE_TXF_CMS_W
||
804 opcode
== SHADER_OPCODE_TXF_UMS
||
805 opcode
== SHADER_OPCODE_TXF_MCS
||
806 opcode
== SHADER_OPCODE_TXL
||
807 opcode
== SHADER_OPCODE_TXS
||
808 opcode
== SHADER_OPCODE_LOD
||
809 opcode
== SHADER_OPCODE_TG4
||
810 opcode
== SHADER_OPCODE_TG4_OFFSET
||
811 opcode
== SHADER_OPCODE_SAMPLEINFO
);
815 backend_instruction::is_math() const
817 return (opcode
== SHADER_OPCODE_RCP
||
818 opcode
== SHADER_OPCODE_RSQ
||
819 opcode
== SHADER_OPCODE_SQRT
||
820 opcode
== SHADER_OPCODE_EXP2
||
821 opcode
== SHADER_OPCODE_LOG2
||
822 opcode
== SHADER_OPCODE_SIN
||
823 opcode
== SHADER_OPCODE_COS
||
824 opcode
== SHADER_OPCODE_INT_QUOTIENT
||
825 opcode
== SHADER_OPCODE_INT_REMAINDER
||
826 opcode
== SHADER_OPCODE_POW
);
830 backend_instruction::is_control_flow() const
834 case BRW_OPCODE_WHILE
:
836 case BRW_OPCODE_ELSE
:
837 case BRW_OPCODE_ENDIF
:
838 case BRW_OPCODE_BREAK
:
839 case BRW_OPCODE_CONTINUE
:
847 backend_instruction::can_do_source_mods() const
850 case BRW_OPCODE_ADDC
:
852 case BRW_OPCODE_BFI1
:
853 case BRW_OPCODE_BFI2
:
854 case BRW_OPCODE_BFREV
:
855 case BRW_OPCODE_CBIT
:
858 case BRW_OPCODE_SUBB
:
866 backend_instruction::can_do_saturate() const
876 case BRW_OPCODE_F16TO32
:
877 case BRW_OPCODE_F32TO16
:
878 case BRW_OPCODE_LINE
:
882 case BRW_OPCODE_MATH
:
885 case SHADER_OPCODE_MULH
:
887 case BRW_OPCODE_RNDD
:
888 case BRW_OPCODE_RNDE
:
889 case BRW_OPCODE_RNDU
:
890 case BRW_OPCODE_RNDZ
:
894 case FS_OPCODE_LINTERP
:
895 case SHADER_OPCODE_COS
:
896 case SHADER_OPCODE_EXP2
:
897 case SHADER_OPCODE_LOG2
:
898 case SHADER_OPCODE_POW
:
899 case SHADER_OPCODE_RCP
:
900 case SHADER_OPCODE_RSQ
:
901 case SHADER_OPCODE_SIN
:
902 case SHADER_OPCODE_SQRT
:
910 backend_instruction::can_do_cmod() const
914 case BRW_OPCODE_ADDC
:
919 case BRW_OPCODE_CMPN
:
924 case BRW_OPCODE_F16TO32
:
925 case BRW_OPCODE_F32TO16
:
927 case BRW_OPCODE_LINE
:
931 case BRW_OPCODE_MACH
:
938 case BRW_OPCODE_RNDD
:
939 case BRW_OPCODE_RNDE
:
940 case BRW_OPCODE_RNDU
:
941 case BRW_OPCODE_RNDZ
:
942 case BRW_OPCODE_SAD2
:
943 case BRW_OPCODE_SADA2
:
946 case BRW_OPCODE_SUBB
:
948 case FS_OPCODE_CINTERP
:
949 case FS_OPCODE_LINTERP
:
957 backend_instruction::reads_accumulator_implicitly() const
961 case BRW_OPCODE_MACH
:
962 case BRW_OPCODE_SADA2
:
970 backend_instruction::writes_accumulator_implicitly(const struct brw_device_info
*devinfo
) const
972 return writes_accumulator
||
974 ((opcode
>= BRW_OPCODE_ADD
&& opcode
< BRW_OPCODE_NOP
) ||
975 (opcode
>= FS_OPCODE_DDX_COARSE
&& opcode
<= FS_OPCODE_LINTERP
&&
976 opcode
!= FS_OPCODE_CINTERP
)));
980 backend_instruction::has_side_effects() const
983 case SHADER_OPCODE_UNTYPED_ATOMIC
:
984 case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL
:
985 case SHADER_OPCODE_GEN4_SCRATCH_WRITE
:
986 case SHADER_OPCODE_UNTYPED_SURFACE_WRITE
:
987 case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL
:
988 case SHADER_OPCODE_TYPED_ATOMIC
:
989 case SHADER_OPCODE_TYPED_ATOMIC_LOGICAL
:
990 case SHADER_OPCODE_TYPED_SURFACE_WRITE
:
991 case SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL
:
992 case SHADER_OPCODE_MEMORY_FENCE
:
993 case SHADER_OPCODE_URB_WRITE_SIMD8
:
994 case SHADER_OPCODE_URB_WRITE_SIMD8_PER_SLOT
:
995 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED
:
996 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT
:
997 case FS_OPCODE_FB_WRITE
:
998 case SHADER_OPCODE_BARRIER
:
999 case TCS_OPCODE_URB_WRITE
:
1000 case TCS_OPCODE_RELEASE_INPUT
:
1008 backend_instruction::is_volatile() const
1011 case SHADER_OPCODE_UNTYPED_SURFACE_READ
:
1012 case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL
:
1013 case SHADER_OPCODE_TYPED_SURFACE_READ
:
1014 case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL
:
1015 case SHADER_OPCODE_URB_READ_SIMD8
:
1016 case SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT
:
1017 case VEC4_OPCODE_URB_READ
:
1026 inst_is_in_block(const bblock_t
*block
, const backend_instruction
*inst
)
1029 foreach_inst_in_block (backend_instruction
, i
, block
) {
1039 adjust_later_block_ips(bblock_t
*start_block
, int ip_adjustment
)
1041 for (bblock_t
*block_iter
= start_block
->next();
1043 block_iter
= block_iter
->next()) {
1044 block_iter
->start_ip
+= ip_adjustment
;
1045 block_iter
->end_ip
+= ip_adjustment
;
1050 backend_instruction::insert_after(bblock_t
*block
, backend_instruction
*inst
)
1052 assert(this != inst
);
1054 if (!this->is_head_sentinel())
1055 assert(inst_is_in_block(block
, this) || !"Instruction not in block");
1059 adjust_later_block_ips(block
, 1);
1061 exec_node::insert_after(inst
);
1065 backend_instruction::insert_before(bblock_t
*block
, backend_instruction
*inst
)
1067 assert(this != inst
);
1069 if (!this->is_tail_sentinel())
1070 assert(inst_is_in_block(block
, this) || !"Instruction not in block");
1074 adjust_later_block_ips(block
, 1);
1076 exec_node::insert_before(inst
);
1080 backend_instruction::insert_before(bblock_t
*block
, exec_list
*list
)
1082 assert(inst_is_in_block(block
, this) || !"Instruction not in block");
1084 unsigned num_inst
= list
->length();
1086 block
->end_ip
+= num_inst
;
1088 adjust_later_block_ips(block
, num_inst
);
1090 exec_node::insert_before(list
);
1094 backend_instruction::remove(bblock_t
*block
)
1096 assert(inst_is_in_block(block
, this) || !"Instruction not in block");
1098 adjust_later_block_ips(block
, -1);
1100 if (block
->start_ip
== block
->end_ip
) {
1101 block
->cfg
->remove_block(block
);
1106 exec_node::remove();
1110 backend_shader::dump_instructions()
1112 dump_instructions(NULL
);
1116 backend_shader::dump_instructions(const char *name
)
1118 FILE *file
= stderr
;
1119 if (name
&& geteuid() != 0) {
1120 file
= fopen(name
, "w");
1127 foreach_block_and_inst(block
, backend_instruction
, inst
, cfg
) {
1128 if (!unlikely(INTEL_DEBUG
& DEBUG_OPTIMIZER
))
1129 fprintf(file
, "%4d: ", ip
++);
1130 dump_instruction(inst
, file
);
1134 foreach_in_list(backend_instruction
, inst
, &instructions
) {
1135 if (!unlikely(INTEL_DEBUG
& DEBUG_OPTIMIZER
))
1136 fprintf(file
, "%4d: ", ip
++);
1137 dump_instruction(inst
, file
);
1141 if (file
!= stderr
) {
1147 backend_shader::calculate_cfg()
1151 cfg
= new(mem_ctx
) cfg_t(&this->instructions
);
1155 * Sets up the starting offsets for the groups of binding table entries
1156 * commong to all pipeline stages.
1158 * Unused groups are initialized to 0xd0d0d0d0 to make it obvious that they're
1159 * unused but also make sure that addition of small offsets to them will
1160 * trigger some of our asserts that surface indices are < BRW_MAX_SURFACES.
1163 brw_assign_common_binding_table_offsets(gl_shader_stage stage
,
1164 const struct brw_device_info
*devinfo
,
1165 const struct gl_shader_program
*shader_prog
,
1166 const struct gl_program
*prog
,
1167 struct brw_stage_prog_data
*stage_prog_data
,
1168 uint32_t next_binding_table_offset
)
1170 const struct gl_shader
*shader
= NULL
;
1171 int num_textures
= _mesa_fls(prog
->SamplersUsed
);
1174 shader
= shader_prog
->_LinkedShaders
[stage
];
1176 stage_prog_data
->binding_table
.texture_start
= next_binding_table_offset
;
1177 next_binding_table_offset
+= num_textures
;
1180 assert(shader
->NumUniformBlocks
<= BRW_MAX_UBO
);
1181 stage_prog_data
->binding_table
.ubo_start
= next_binding_table_offset
;
1182 next_binding_table_offset
+= shader
->NumUniformBlocks
;
1184 assert(shader
->NumShaderStorageBlocks
<= BRW_MAX_SSBO
);
1185 stage_prog_data
->binding_table
.ssbo_start
= next_binding_table_offset
;
1186 next_binding_table_offset
+= shader
->NumShaderStorageBlocks
;
1188 stage_prog_data
->binding_table
.ubo_start
= 0xd0d0d0d0;
1189 stage_prog_data
->binding_table
.ssbo_start
= 0xd0d0d0d0;
1192 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
1193 stage_prog_data
->binding_table
.shader_time_start
= next_binding_table_offset
;
1194 next_binding_table_offset
++;
1196 stage_prog_data
->binding_table
.shader_time_start
= 0xd0d0d0d0;
1199 if (prog
->UsesGather
) {
1200 if (devinfo
->gen
>= 8) {
1201 stage_prog_data
->binding_table
.gather_texture_start
=
1202 stage_prog_data
->binding_table
.texture_start
;
1204 stage_prog_data
->binding_table
.gather_texture_start
= next_binding_table_offset
;
1205 next_binding_table_offset
+= num_textures
;
1208 stage_prog_data
->binding_table
.gather_texture_start
= 0xd0d0d0d0;
1211 if (shader
&& shader
->NumAtomicBuffers
) {
1212 stage_prog_data
->binding_table
.abo_start
= next_binding_table_offset
;
1213 next_binding_table_offset
+= shader
->NumAtomicBuffers
;
1215 stage_prog_data
->binding_table
.abo_start
= 0xd0d0d0d0;
1218 if (shader
&& shader
->NumImages
) {
1219 stage_prog_data
->binding_table
.image_start
= next_binding_table_offset
;
1220 next_binding_table_offset
+= shader
->NumImages
;
1222 stage_prog_data
->binding_table
.image_start
= 0xd0d0d0d0;
1225 /* This may or may not be used depending on how the compile goes. */
1226 stage_prog_data
->binding_table
.pull_constants_start
= next_binding_table_offset
;
1227 next_binding_table_offset
++;
1229 assert(next_binding_table_offset
<= BRW_MAX_SURFACES
);
1231 /* prog_data->base.binding_table.size will be set by brw_mark_surface_used. */
1235 setup_vec4_uniform_value(const gl_constant_value
**params
,
1236 const gl_constant_value
*values
,
1239 static const gl_constant_value zero
= { 0 };
1241 for (unsigned i
= 0; i
< n
; ++i
)
1242 params
[i
] = &values
[i
];
1244 for (unsigned i
= n
; i
< 4; ++i
)
1249 brw_setup_image_uniform_values(gl_shader_stage stage
,
1250 struct brw_stage_prog_data
*stage_prog_data
,
1251 unsigned param_start_index
,
1252 const gl_uniform_storage
*storage
)
1254 const gl_constant_value
**param
=
1255 &stage_prog_data
->param
[param_start_index
];
1257 for (unsigned i
= 0; i
< MAX2(storage
->array_elements
, 1); i
++) {
1258 const unsigned image_idx
= storage
->opaque
[stage
].index
+ i
;
1259 const brw_image_param
*image_param
=
1260 &stage_prog_data
->image_param
[image_idx
];
1262 /* Upload the brw_image_param structure. The order is expected to match
1263 * the BRW_IMAGE_PARAM_*_OFFSET defines.
1265 setup_vec4_uniform_value(param
+ BRW_IMAGE_PARAM_SURFACE_IDX_OFFSET
,
1266 (const gl_constant_value
*)&image_param
->surface_idx
, 1);
1267 setup_vec4_uniform_value(param
+ BRW_IMAGE_PARAM_OFFSET_OFFSET
,
1268 (const gl_constant_value
*)image_param
->offset
, 2);
1269 setup_vec4_uniform_value(param
+ BRW_IMAGE_PARAM_SIZE_OFFSET
,
1270 (const gl_constant_value
*)image_param
->size
, 3);
1271 setup_vec4_uniform_value(param
+ BRW_IMAGE_PARAM_STRIDE_OFFSET
,
1272 (const gl_constant_value
*)image_param
->stride
, 4);
1273 setup_vec4_uniform_value(param
+ BRW_IMAGE_PARAM_TILING_OFFSET
,
1274 (const gl_constant_value
*)image_param
->tiling
, 3);
1275 setup_vec4_uniform_value(param
+ BRW_IMAGE_PARAM_SWIZZLING_OFFSET
,
1276 (const gl_constant_value
*)image_param
->swizzling
, 2);
1277 param
+= BRW_IMAGE_PARAM_SIZE
;
1279 brw_mark_surface_used(
1281 stage_prog_data
->binding_table
.image_start
+ image_idx
);
1286 * Decide which set of clip planes should be used when clipping via
1287 * gl_Position or gl_ClipVertex.
1289 gl_clip_plane
*brw_select_clip_planes(struct gl_context
*ctx
)
1291 if (ctx
->_Shader
->CurrentProgram
[MESA_SHADER_VERTEX
]) {
1292 /* There is currently a GLSL vertex shader, so clip according to GLSL
1293 * rules, which means compare gl_ClipVertex (or gl_Position, if
1294 * gl_ClipVertex wasn't assigned) against the eye-coordinate clip planes
1295 * that were stored in EyeUserPlane at the time the clip planes were
1298 return ctx
->Transform
.EyeUserPlane
;
1300 /* Either we are using fixed function or an ARB vertex program. In
1301 * either case the clip planes are going to be compared against
1302 * gl_Position (which is in clip coordinates) so we have to clip using
1303 * _ClipUserPlane, which was transformed into clip coordinates by Mesa
1306 return ctx
->Transform
._ClipUserPlane
;
1310 extern "C" const unsigned *
1311 brw_compile_tes(const struct brw_compiler
*compiler
,
1314 const struct brw_tes_prog_key
*key
,
1315 struct brw_tes_prog_data
*prog_data
,
1316 const nir_shader
*src_shader
,
1317 struct gl_shader_program
*shader_prog
,
1318 int shader_time_index
,
1319 unsigned *final_assembly_size
,
1322 const struct brw_device_info
*devinfo
= compiler
->devinfo
;
1323 struct gl_shader
*shader
=
1324 shader_prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
1325 const bool is_scalar
= compiler
->scalar_stage
[MESA_SHADER_TESS_EVAL
];
1327 nir_shader
*nir
= nir_shader_clone(mem_ctx
, src_shader
);
1328 nir
->info
.inputs_read
= key
->inputs_read
;
1329 nir
->info
.patch_inputs_read
= key
->patch_inputs_read
;
1331 struct brw_vue_map input_vue_map
;
1332 brw_compute_tess_vue_map(&input_vue_map
,
1333 nir
->info
.inputs_read
& ~VARYING_BIT_PRIMITIVE_ID
,
1334 nir
->info
.patch_inputs_read
);
1336 nir
= brw_nir_apply_sampler_key(nir
, devinfo
, &key
->tex
, is_scalar
);
1337 brw_nir_lower_tes_inputs(nir
, &input_vue_map
);
1338 brw_nir_lower_vue_outputs(nir
, is_scalar
);
1339 nir
= brw_postprocess_nir(nir
, compiler
->devinfo
, is_scalar
);
1341 brw_compute_vue_map(devinfo
, &prog_data
->base
.vue_map
,
1342 nir
->info
.outputs_written
,
1343 nir
->info
.separate_shader
);
1345 unsigned output_size_bytes
= prog_data
->base
.vue_map
.num_slots
* 4 * 4;
1347 assert(output_size_bytes
>= 1);
1348 if (output_size_bytes
> GEN7_MAX_DS_URB_ENTRY_SIZE_BYTES
) {
1350 *error_str
= ralloc_strdup(mem_ctx
, "DS outputs exceed maximum size");
1354 /* URB entry sizes are stored as a multiple of 64 bytes. */
1355 prog_data
->base
.urb_entry_size
= ALIGN(output_size_bytes
, 64) / 64;
1357 bool need_patch_header
= nir
->info
.system_values_read
&
1358 (BITFIELD64_BIT(SYSTEM_VALUE_TESS_LEVEL_OUTER
) |
1359 BITFIELD64_BIT(SYSTEM_VALUE_TESS_LEVEL_INNER
));
1361 /* The TES will pull most inputs using URB read messages.
1363 * However, we push the patch header for TessLevel factors when required,
1364 * as it's a tiny amount of extra data.
1366 prog_data
->base
.urb_read_length
= need_patch_header
? 1 : 0;
1368 if (unlikely(INTEL_DEBUG
& DEBUG_TES
)) {
1369 fprintf(stderr
, "TES Input ");
1370 brw_print_vue_map(stderr
, &input_vue_map
);
1371 fprintf(stderr
, "TES Output ");
1372 brw_print_vue_map(stderr
, &prog_data
->base
.vue_map
);
1376 fs_visitor
v(compiler
, log_data
, mem_ctx
, (void *) key
,
1377 &prog_data
->base
.base
, shader
->Program
, nir
, 8,
1378 shader_time_index
, &input_vue_map
);
1381 *error_str
= ralloc_strdup(mem_ctx
, v
.fail_msg
);
1385 prog_data
->base
.dispatch_mode
= DISPATCH_MODE_SIMD8
;
1387 fs_generator
g(compiler
, log_data
, mem_ctx
, (void *) key
,
1388 &prog_data
->base
.base
, v
.promoted_constants
, false,
1389 MESA_SHADER_TESS_EVAL
);
1390 if (unlikely(INTEL_DEBUG
& DEBUG_TES
)) {
1391 g
.enable_debug(ralloc_asprintf(mem_ctx
,
1392 "%s tessellation evaluation shader %s",
1393 nir
->info
.label
? nir
->info
.label
1398 g
.generate_code(v
.cfg
, 8);
1400 return g
.get_assembly(final_assembly_size
);
1402 brw::vec4_tes_visitor
v(compiler
, log_data
, key
, prog_data
,
1403 nir
, mem_ctx
, shader_time_index
);
1406 *error_str
= ralloc_strdup(mem_ctx
, v
.fail_msg
);
1410 if (unlikely(INTEL_DEBUG
& DEBUG_TES
))
1411 v
.dump_instructions();
1413 return brw_vec4_generate_assembly(compiler
, log_data
, mem_ctx
, nir
,
1414 &prog_data
->base
, v
.cfg
,
1415 final_assembly_size
);