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
28 #include "brw_vec4_tes.h"
29 #include "common/gen_debug.h"
30 #include "main/uniforms.h"
31 #include "util/macros.h"
34 brw_type_for_base_type(const struct glsl_type
*type
)
36 switch (type
->base_type
) {
37 case GLSL_TYPE_FLOAT16
:
38 return BRW_REGISTER_TYPE_HF
;
40 return BRW_REGISTER_TYPE_F
;
43 case GLSL_TYPE_SUBROUTINE
:
44 return BRW_REGISTER_TYPE_D
;
46 return BRW_REGISTER_TYPE_W
;
48 return BRW_REGISTER_TYPE_B
;
50 return BRW_REGISTER_TYPE_UD
;
51 case GLSL_TYPE_UINT16
:
52 return BRW_REGISTER_TYPE_UW
;
54 return BRW_REGISTER_TYPE_UB
;
56 return brw_type_for_base_type(type
->fields
.array
);
57 case GLSL_TYPE_STRUCT
:
58 case GLSL_TYPE_SAMPLER
:
59 case GLSL_TYPE_ATOMIC_UINT
:
60 /* These should be overridden with the type of the member when
61 * dereferenced into. BRW_REGISTER_TYPE_UD seems like a likely
62 * way to trip up if we don't.
64 return BRW_REGISTER_TYPE_UD
;
66 return BRW_REGISTER_TYPE_UD
;
67 case GLSL_TYPE_DOUBLE
:
68 return BRW_REGISTER_TYPE_DF
;
69 case GLSL_TYPE_UINT64
:
70 return BRW_REGISTER_TYPE_UQ
;
72 return BRW_REGISTER_TYPE_Q
;
75 case GLSL_TYPE_INTERFACE
:
76 case GLSL_TYPE_FUNCTION
:
77 unreachable("not reached");
80 return BRW_REGISTER_TYPE_F
;
83 enum brw_conditional_mod
84 brw_conditional_for_comparison(unsigned int op
)
88 return BRW_CONDITIONAL_L
;
90 return BRW_CONDITIONAL_GE
;
92 case ir_binop_all_equal
: /* same as equal for scalars */
93 return BRW_CONDITIONAL_Z
;
95 case ir_binop_any_nequal
: /* same as nequal for scalars */
96 return BRW_CONDITIONAL_NZ
;
98 unreachable("not reached: bad operation for comparison");
103 brw_math_function(enum opcode op
)
106 case SHADER_OPCODE_RCP
:
107 return BRW_MATH_FUNCTION_INV
;
108 case SHADER_OPCODE_RSQ
:
109 return BRW_MATH_FUNCTION_RSQ
;
110 case SHADER_OPCODE_SQRT
:
111 return BRW_MATH_FUNCTION_SQRT
;
112 case SHADER_OPCODE_EXP2
:
113 return BRW_MATH_FUNCTION_EXP
;
114 case SHADER_OPCODE_LOG2
:
115 return BRW_MATH_FUNCTION_LOG
;
116 case SHADER_OPCODE_POW
:
117 return BRW_MATH_FUNCTION_POW
;
118 case SHADER_OPCODE_SIN
:
119 return BRW_MATH_FUNCTION_SIN
;
120 case SHADER_OPCODE_COS
:
121 return BRW_MATH_FUNCTION_COS
;
122 case SHADER_OPCODE_INT_QUOTIENT
:
123 return BRW_MATH_FUNCTION_INT_DIV_QUOTIENT
;
124 case SHADER_OPCODE_INT_REMAINDER
:
125 return BRW_MATH_FUNCTION_INT_DIV_REMAINDER
;
127 unreachable("not reached: unknown math function");
132 brw_texture_offset(int *offsets
, unsigned num_components
, uint32_t *offset_bits
)
134 if (!offsets
) return false; /* nonconstant offset; caller will handle it. */
136 /* offset out of bounds; caller will handle it. */
137 for (unsigned i
= 0; i
< num_components
; i
++)
138 if (offsets
[i
] > 7 || offsets
[i
] < -8)
141 /* Combine all three offsets into a single unsigned dword:
143 * bits 11:8 - U Offset (X component)
144 * bits 7:4 - V Offset (Y component)
145 * bits 3:0 - R Offset (Z component)
148 for (unsigned i
= 0; i
< num_components
; i
++) {
149 const unsigned shift
= 4 * (2 - i
);
150 *offset_bits
|= (offsets
[i
] << shift
) & (0xF << shift
);
156 brw_instruction_name(const struct gen_device_info
*devinfo
, enum opcode op
)
159 case BRW_OPCODE_ILLEGAL
... BRW_OPCODE_NOP
:
160 /* The DO instruction doesn't exist on Gen6+, but we use it to mark the
161 * start of a loop in the IR.
163 if (devinfo
->gen
>= 6 && op
== BRW_OPCODE_DO
)
166 /* The following conversion opcodes doesn't exist on Gen8+, but we use
167 * then to mark that we want to do the conversion.
169 if (devinfo
->gen
> 7 && op
== BRW_OPCODE_F32TO16
)
172 if (devinfo
->gen
> 7 && op
== BRW_OPCODE_F16TO32
)
175 assert(brw_opcode_desc(devinfo
, op
)->name
);
176 return brw_opcode_desc(devinfo
, op
)->name
;
177 case FS_OPCODE_FB_WRITE
:
179 case FS_OPCODE_FB_WRITE_LOGICAL
:
180 return "fb_write_logical";
181 case FS_OPCODE_REP_FB_WRITE
:
182 return "rep_fb_write";
183 case FS_OPCODE_FB_READ
:
185 case FS_OPCODE_FB_READ_LOGICAL
:
186 return "fb_read_logical";
188 case SHADER_OPCODE_RCP
:
190 case SHADER_OPCODE_RSQ
:
192 case SHADER_OPCODE_SQRT
:
194 case SHADER_OPCODE_EXP2
:
196 case SHADER_OPCODE_LOG2
:
198 case SHADER_OPCODE_POW
:
200 case SHADER_OPCODE_INT_QUOTIENT
:
202 case SHADER_OPCODE_INT_REMAINDER
:
204 case SHADER_OPCODE_SIN
:
206 case SHADER_OPCODE_COS
:
209 case SHADER_OPCODE_SEND
:
212 case SHADER_OPCODE_TEX
:
214 case SHADER_OPCODE_TEX_LOGICAL
:
215 return "tex_logical";
216 case SHADER_OPCODE_TXD
:
218 case SHADER_OPCODE_TXD_LOGICAL
:
219 return "txd_logical";
220 case SHADER_OPCODE_TXF
:
222 case SHADER_OPCODE_TXF_LOGICAL
:
223 return "txf_logical";
224 case SHADER_OPCODE_TXF_LZ
:
226 case SHADER_OPCODE_TXL
:
228 case SHADER_OPCODE_TXL_LOGICAL
:
229 return "txl_logical";
230 case SHADER_OPCODE_TXL_LZ
:
232 case SHADER_OPCODE_TXS
:
234 case SHADER_OPCODE_TXS_LOGICAL
:
235 return "txs_logical";
238 case FS_OPCODE_TXB_LOGICAL
:
239 return "txb_logical";
240 case SHADER_OPCODE_TXF_CMS
:
242 case SHADER_OPCODE_TXF_CMS_LOGICAL
:
243 return "txf_cms_logical";
244 case SHADER_OPCODE_TXF_CMS_W
:
246 case SHADER_OPCODE_TXF_CMS_W_LOGICAL
:
247 return "txf_cms_w_logical";
248 case SHADER_OPCODE_TXF_UMS
:
250 case SHADER_OPCODE_TXF_UMS_LOGICAL
:
251 return "txf_ums_logical";
252 case SHADER_OPCODE_TXF_MCS
:
254 case SHADER_OPCODE_TXF_MCS_LOGICAL
:
255 return "txf_mcs_logical";
256 case SHADER_OPCODE_LOD
:
258 case SHADER_OPCODE_LOD_LOGICAL
:
259 return "lod_logical";
260 case SHADER_OPCODE_TG4
:
262 case SHADER_OPCODE_TG4_LOGICAL
:
263 return "tg4_logical";
264 case SHADER_OPCODE_TG4_OFFSET
:
266 case SHADER_OPCODE_TG4_OFFSET_LOGICAL
:
267 return "tg4_offset_logical";
268 case SHADER_OPCODE_SAMPLEINFO
:
270 case SHADER_OPCODE_SAMPLEINFO_LOGICAL
:
271 return "sampleinfo_logical";
273 case SHADER_OPCODE_IMAGE_SIZE_LOGICAL
:
274 return "image_size_logical";
276 case SHADER_OPCODE_SHADER_TIME_ADD
:
277 return "shader_time_add";
279 case SHADER_OPCODE_UNTYPED_ATOMIC
:
280 return "untyped_atomic";
281 case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL
:
282 return "untyped_atomic_logical";
283 case SHADER_OPCODE_UNTYPED_ATOMIC_FLOAT
:
284 return "untyped_atomic_float";
285 case SHADER_OPCODE_UNTYPED_ATOMIC_FLOAT_LOGICAL
:
286 return "untyped_atomic_float_logical";
287 case SHADER_OPCODE_UNTYPED_SURFACE_READ
:
288 return "untyped_surface_read";
289 case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL
:
290 return "untyped_surface_read_logical";
291 case SHADER_OPCODE_UNTYPED_SURFACE_WRITE
:
292 return "untyped_surface_write";
293 case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL
:
294 return "untyped_surface_write_logical";
295 case SHADER_OPCODE_A64_UNTYPED_READ_LOGICAL
:
296 return "a64_untyped_read_logical";
297 case SHADER_OPCODE_A64_UNTYPED_WRITE_LOGICAL
:
298 return "a64_untyped_write_logical";
299 case SHADER_OPCODE_A64_BYTE_SCATTERED_READ_LOGICAL
:
300 return "a64_byte_scattered_read_logical";
301 case SHADER_OPCODE_A64_BYTE_SCATTERED_WRITE_LOGICAL
:
302 return "a64_byte_scattered_write_logical";
303 case SHADER_OPCODE_A64_UNTYPED_ATOMIC_LOGICAL
:
304 return "a64_untyped_atomic_logical";
305 case SHADER_OPCODE_A64_UNTYPED_ATOMIC_FLOAT_LOGICAL
:
306 return "a64_untyped_atomic_float_logical";
307 case SHADER_OPCODE_TYPED_ATOMIC
:
308 return "typed_atomic";
309 case SHADER_OPCODE_TYPED_ATOMIC_LOGICAL
:
310 return "typed_atomic_logical";
311 case SHADER_OPCODE_TYPED_SURFACE_READ
:
312 return "typed_surface_read";
313 case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL
:
314 return "typed_surface_read_logical";
315 case SHADER_OPCODE_TYPED_SURFACE_WRITE
:
316 return "typed_surface_write";
317 case SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL
:
318 return "typed_surface_write_logical";
319 case SHADER_OPCODE_MEMORY_FENCE
:
320 return "memory_fence";
321 case SHADER_OPCODE_INTERLOCK
:
322 /* For an interlock we actually issue a memory fence via sendc. */
325 case SHADER_OPCODE_BYTE_SCATTERED_READ
:
326 return "byte_scattered_read";
327 case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL
:
328 return "byte_scattered_read_logical";
329 case SHADER_OPCODE_BYTE_SCATTERED_WRITE
:
330 return "byte_scattered_write";
331 case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL
:
332 return "byte_scattered_write_logical";
334 case SHADER_OPCODE_LOAD_PAYLOAD
:
335 return "load_payload";
339 case SHADER_OPCODE_GEN4_SCRATCH_READ
:
340 return "gen4_scratch_read";
341 case SHADER_OPCODE_GEN4_SCRATCH_WRITE
:
342 return "gen4_scratch_write";
343 case SHADER_OPCODE_GEN7_SCRATCH_READ
:
344 return "gen7_scratch_read";
345 case SHADER_OPCODE_URB_WRITE_SIMD8
:
346 return "gen8_urb_write_simd8";
347 case SHADER_OPCODE_URB_WRITE_SIMD8_PER_SLOT
:
348 return "gen8_urb_write_simd8_per_slot";
349 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED
:
350 return "gen8_urb_write_simd8_masked";
351 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT
:
352 return "gen8_urb_write_simd8_masked_per_slot";
353 case SHADER_OPCODE_URB_READ_SIMD8
:
354 return "urb_read_simd8";
355 case SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT
:
356 return "urb_read_simd8_per_slot";
358 case SHADER_OPCODE_FIND_LIVE_CHANNEL
:
359 return "find_live_channel";
360 case SHADER_OPCODE_BROADCAST
:
362 case SHADER_OPCODE_SHUFFLE
:
364 case SHADER_OPCODE_SEL_EXEC
:
366 case SHADER_OPCODE_QUAD_SWIZZLE
:
367 return "quad_swizzle";
368 case SHADER_OPCODE_CLUSTER_BROADCAST
:
369 return "cluster_broadcast";
371 case SHADER_OPCODE_GET_BUFFER_SIZE
:
372 return "get_buffer_size";
374 case VEC4_OPCODE_MOV_BYTES
:
376 case VEC4_OPCODE_PACK_BYTES
:
378 case VEC4_OPCODE_UNPACK_UNIFORM
:
379 return "unpack_uniform";
380 case VEC4_OPCODE_DOUBLE_TO_F32
:
381 return "double_to_f32";
382 case VEC4_OPCODE_DOUBLE_TO_D32
:
383 return "double_to_d32";
384 case VEC4_OPCODE_DOUBLE_TO_U32
:
385 return "double_to_u32";
386 case VEC4_OPCODE_TO_DOUBLE
:
387 return "single_to_double";
388 case VEC4_OPCODE_PICK_LOW_32BIT
:
389 return "pick_low_32bit";
390 case VEC4_OPCODE_PICK_HIGH_32BIT
:
391 return "pick_high_32bit";
392 case VEC4_OPCODE_SET_LOW_32BIT
:
393 return "set_low_32bit";
394 case VEC4_OPCODE_SET_HIGH_32BIT
:
395 return "set_high_32bit";
397 case FS_OPCODE_DDX_COARSE
:
399 case FS_OPCODE_DDX_FINE
:
401 case FS_OPCODE_DDY_COARSE
:
403 case FS_OPCODE_DDY_FINE
:
406 case FS_OPCODE_LINTERP
:
409 case FS_OPCODE_PIXEL_X
:
411 case FS_OPCODE_PIXEL_Y
:
414 case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD
:
415 return "uniform_pull_const";
416 case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD_GEN7
:
417 return "uniform_pull_const_gen7";
418 case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN4
:
419 return "varying_pull_const_gen4";
420 case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_LOGICAL
:
421 return "varying_pull_const_logical";
423 case FS_OPCODE_DISCARD_JUMP
:
424 return "discard_jump";
426 case FS_OPCODE_SET_SAMPLE_ID
:
427 return "set_sample_id";
429 case FS_OPCODE_PACK_HALF_2x16_SPLIT
:
430 return "pack_half_2x16_split";
432 case FS_OPCODE_PLACEHOLDER_HALT
:
433 return "placeholder_halt";
435 case FS_OPCODE_INTERPOLATE_AT_SAMPLE
:
436 return "interp_sample";
437 case FS_OPCODE_INTERPOLATE_AT_SHARED_OFFSET
:
438 return "interp_shared_offset";
439 case FS_OPCODE_INTERPOLATE_AT_PER_SLOT_OFFSET
:
440 return "interp_per_slot_offset";
442 case VS_OPCODE_URB_WRITE
:
443 return "vs_urb_write";
444 case VS_OPCODE_PULL_CONSTANT_LOAD
:
445 return "pull_constant_load";
446 case VS_OPCODE_PULL_CONSTANT_LOAD_GEN7
:
447 return "pull_constant_load_gen7";
449 case VS_OPCODE_SET_SIMD4X2_HEADER_GEN9
:
450 return "set_simd4x2_header_gen9";
452 case VS_OPCODE_UNPACK_FLAGS_SIMD4X2
:
453 return "unpack_flags_simd4x2";
455 case GS_OPCODE_URB_WRITE
:
456 return "gs_urb_write";
457 case GS_OPCODE_URB_WRITE_ALLOCATE
:
458 return "gs_urb_write_allocate";
459 case GS_OPCODE_THREAD_END
:
460 return "gs_thread_end";
461 case GS_OPCODE_SET_WRITE_OFFSET
:
462 return "set_write_offset";
463 case GS_OPCODE_SET_VERTEX_COUNT
:
464 return "set_vertex_count";
465 case GS_OPCODE_SET_DWORD_2
:
466 return "set_dword_2";
467 case GS_OPCODE_PREPARE_CHANNEL_MASKS
:
468 return "prepare_channel_masks";
469 case GS_OPCODE_SET_CHANNEL_MASKS
:
470 return "set_channel_masks";
471 case GS_OPCODE_GET_INSTANCE_ID
:
472 return "get_instance_id";
473 case GS_OPCODE_FF_SYNC
:
475 case GS_OPCODE_SET_PRIMITIVE_ID
:
476 return "set_primitive_id";
477 case GS_OPCODE_SVB_WRITE
:
478 return "gs_svb_write";
479 case GS_OPCODE_SVB_SET_DST_INDEX
:
480 return "gs_svb_set_dst_index";
481 case GS_OPCODE_FF_SYNC_SET_PRIMITIVES
:
482 return "gs_ff_sync_set_primitives";
483 case CS_OPCODE_CS_TERMINATE
:
484 return "cs_terminate";
485 case SHADER_OPCODE_BARRIER
:
487 case SHADER_OPCODE_MULH
:
489 case SHADER_OPCODE_MOV_INDIRECT
:
490 return "mov_indirect";
492 case VEC4_OPCODE_URB_READ
:
494 case TCS_OPCODE_GET_INSTANCE_ID
:
495 return "tcs_get_instance_id";
496 case TCS_OPCODE_URB_WRITE
:
497 return "tcs_urb_write";
498 case TCS_OPCODE_SET_INPUT_URB_OFFSETS
:
499 return "tcs_set_input_urb_offsets";
500 case TCS_OPCODE_SET_OUTPUT_URB_OFFSETS
:
501 return "tcs_set_output_urb_offsets";
502 case TCS_OPCODE_GET_PRIMITIVE_ID
:
503 return "tcs_get_primitive_id";
504 case TCS_OPCODE_CREATE_BARRIER_HEADER
:
505 return "tcs_create_barrier_header";
506 case TCS_OPCODE_SRC0_010_IS_ZERO
:
507 return "tcs_src0<0,1,0>_is_zero";
508 case TCS_OPCODE_RELEASE_INPUT
:
509 return "tcs_release_input";
510 case TCS_OPCODE_THREAD_END
:
511 return "tcs_thread_end";
512 case TES_OPCODE_CREATE_INPUT_READ_HEADER
:
513 return "tes_create_input_read_header";
514 case TES_OPCODE_ADD_INDIRECT_URB_OFFSET
:
515 return "tes_add_indirect_urb_offset";
516 case TES_OPCODE_GET_PRIMITIVE_ID
:
517 return "tes_get_primitive_id";
519 case SHADER_OPCODE_RND_MODE
:
523 unreachable("not reached");
527 brw_saturate_immediate(enum brw_reg_type type
, struct brw_reg
*reg
)
534 } imm
, sat_imm
= { 0 };
536 const unsigned size
= type_sz(type
);
538 /* We want to either do a 32-bit or 64-bit data copy, the type is otherwise
539 * irrelevant, so just check the size of the type and copy from/to an
540 * appropriately sized field.
548 case BRW_REGISTER_TYPE_UD
:
549 case BRW_REGISTER_TYPE_D
:
550 case BRW_REGISTER_TYPE_UW
:
551 case BRW_REGISTER_TYPE_W
:
552 case BRW_REGISTER_TYPE_UQ
:
553 case BRW_REGISTER_TYPE_Q
:
556 case BRW_REGISTER_TYPE_F
:
557 sat_imm
.f
= CLAMP(imm
.f
, 0.0f
, 1.0f
);
559 case BRW_REGISTER_TYPE_DF
:
560 sat_imm
.df
= CLAMP(imm
.df
, 0.0, 1.0);
562 case BRW_REGISTER_TYPE_UB
:
563 case BRW_REGISTER_TYPE_B
:
564 unreachable("no UB/B immediates");
565 case BRW_REGISTER_TYPE_V
:
566 case BRW_REGISTER_TYPE_UV
:
567 case BRW_REGISTER_TYPE_VF
:
568 unreachable("unimplemented: saturate vector immediate");
569 case BRW_REGISTER_TYPE_HF
:
570 unreachable("unimplemented: saturate HF immediate");
571 case BRW_REGISTER_TYPE_NF
:
572 unreachable("no NF immediates");
576 if (imm
.ud
!= sat_imm
.ud
) {
577 reg
->ud
= sat_imm
.ud
;
581 if (imm
.df
!= sat_imm
.df
) {
582 reg
->df
= sat_imm
.df
;
590 brw_negate_immediate(enum brw_reg_type type
, struct brw_reg
*reg
)
593 case BRW_REGISTER_TYPE_D
:
594 case BRW_REGISTER_TYPE_UD
:
597 case BRW_REGISTER_TYPE_W
:
598 case BRW_REGISTER_TYPE_UW
: {
599 uint16_t value
= -(int16_t)reg
->ud
;
600 reg
->ud
= value
| (uint32_t)value
<< 16;
603 case BRW_REGISTER_TYPE_F
:
606 case BRW_REGISTER_TYPE_VF
:
607 reg
->ud
^= 0x80808080;
609 case BRW_REGISTER_TYPE_DF
:
612 case BRW_REGISTER_TYPE_UQ
:
613 case BRW_REGISTER_TYPE_Q
:
614 reg
->d64
= -reg
->d64
;
616 case BRW_REGISTER_TYPE_UB
:
617 case BRW_REGISTER_TYPE_B
:
618 unreachable("no UB/B immediates");
619 case BRW_REGISTER_TYPE_UV
:
620 case BRW_REGISTER_TYPE_V
:
621 assert(!"unimplemented: negate UV/V immediate");
622 case BRW_REGISTER_TYPE_HF
:
623 reg
->ud
^= 0x80008000;
625 case BRW_REGISTER_TYPE_NF
:
626 unreachable("no NF immediates");
633 brw_abs_immediate(enum brw_reg_type type
, struct brw_reg
*reg
)
636 case BRW_REGISTER_TYPE_D
:
637 reg
->d
= abs(reg
->d
);
639 case BRW_REGISTER_TYPE_W
: {
640 uint16_t value
= abs((int16_t)reg
->ud
);
641 reg
->ud
= value
| (uint32_t)value
<< 16;
644 case BRW_REGISTER_TYPE_F
:
645 reg
->f
= fabsf(reg
->f
);
647 case BRW_REGISTER_TYPE_DF
:
648 reg
->df
= fabs(reg
->df
);
650 case BRW_REGISTER_TYPE_VF
:
651 reg
->ud
&= ~0x80808080;
653 case BRW_REGISTER_TYPE_Q
:
654 reg
->d64
= imaxabs(reg
->d64
);
656 case BRW_REGISTER_TYPE_UB
:
657 case BRW_REGISTER_TYPE_B
:
658 unreachable("no UB/B immediates");
659 case BRW_REGISTER_TYPE_UQ
:
660 case BRW_REGISTER_TYPE_UD
:
661 case BRW_REGISTER_TYPE_UW
:
662 case BRW_REGISTER_TYPE_UV
:
663 /* Presumably the absolute value modifier on an unsigned source is a
664 * nop, but it would be nice to confirm.
666 assert(!"unimplemented: abs unsigned immediate");
667 case BRW_REGISTER_TYPE_V
:
668 assert(!"unimplemented: abs V immediate");
669 case BRW_REGISTER_TYPE_HF
:
670 reg
->ud
&= ~0x80008000;
672 case BRW_REGISTER_TYPE_NF
:
673 unreachable("no NF immediates");
679 backend_shader::backend_shader(const struct brw_compiler
*compiler
,
682 const nir_shader
*shader
,
683 struct brw_stage_prog_data
*stage_prog_data
)
684 : compiler(compiler
),
686 devinfo(compiler
->devinfo
),
688 stage_prog_data(stage_prog_data
),
691 stage(shader
->info
.stage
)
693 debug_enabled
= INTEL_DEBUG
& intel_debug_flag_for_shader_stage(stage
);
694 stage_name
= _mesa_shader_stage_to_string(stage
);
695 stage_abbrev
= _mesa_shader_stage_to_abbrev(stage
);
698 backend_shader::~backend_shader()
703 backend_reg::equals(const backend_reg
&r
) const
705 return brw_regs_equal(this, &r
) && offset
== r
.offset
;
709 backend_reg::negative_equals(const backend_reg
&r
) const
711 return brw_regs_negative_equal(this, &r
) && offset
== r
.offset
;
715 backend_reg::is_zero() const
721 case BRW_REGISTER_TYPE_F
:
723 case BRW_REGISTER_TYPE_DF
:
725 case BRW_REGISTER_TYPE_D
:
726 case BRW_REGISTER_TYPE_UD
:
728 case BRW_REGISTER_TYPE_UQ
:
729 case BRW_REGISTER_TYPE_Q
:
737 backend_reg::is_one() const
743 case BRW_REGISTER_TYPE_F
:
745 case BRW_REGISTER_TYPE_DF
:
747 case BRW_REGISTER_TYPE_D
:
748 case BRW_REGISTER_TYPE_UD
:
750 case BRW_REGISTER_TYPE_UQ
:
751 case BRW_REGISTER_TYPE_Q
:
759 backend_reg::is_negative_one() const
765 case BRW_REGISTER_TYPE_F
:
767 case BRW_REGISTER_TYPE_DF
:
769 case BRW_REGISTER_TYPE_D
:
771 case BRW_REGISTER_TYPE_Q
:
779 backend_reg::is_null() const
781 return file
== ARF
&& nr
== BRW_ARF_NULL
;
786 backend_reg::is_accumulator() const
788 return file
== ARF
&& nr
== BRW_ARF_ACCUMULATOR
;
792 backend_instruction::is_commutative() const
800 case SHADER_OPCODE_MULH
:
803 /* MIN and MAX are commutative. */
804 if (conditional_mod
== BRW_CONDITIONAL_GE
||
805 conditional_mod
== BRW_CONDITIONAL_L
) {
815 backend_instruction::is_3src(const struct gen_device_info
*devinfo
) const
817 return ::is_3src(devinfo
, opcode
);
821 backend_instruction::is_tex() const
823 return (opcode
== SHADER_OPCODE_TEX
||
824 opcode
== FS_OPCODE_TXB
||
825 opcode
== SHADER_OPCODE_TXD
||
826 opcode
== SHADER_OPCODE_TXF
||
827 opcode
== SHADER_OPCODE_TXF_LZ
||
828 opcode
== SHADER_OPCODE_TXF_CMS
||
829 opcode
== SHADER_OPCODE_TXF_CMS_W
||
830 opcode
== SHADER_OPCODE_TXF_UMS
||
831 opcode
== SHADER_OPCODE_TXF_MCS
||
832 opcode
== SHADER_OPCODE_TXL
||
833 opcode
== SHADER_OPCODE_TXL_LZ
||
834 opcode
== SHADER_OPCODE_TXS
||
835 opcode
== SHADER_OPCODE_LOD
||
836 opcode
== SHADER_OPCODE_TG4
||
837 opcode
== SHADER_OPCODE_TG4_OFFSET
||
838 opcode
== SHADER_OPCODE_SAMPLEINFO
);
842 backend_instruction::is_math() const
844 return (opcode
== SHADER_OPCODE_RCP
||
845 opcode
== SHADER_OPCODE_RSQ
||
846 opcode
== SHADER_OPCODE_SQRT
||
847 opcode
== SHADER_OPCODE_EXP2
||
848 opcode
== SHADER_OPCODE_LOG2
||
849 opcode
== SHADER_OPCODE_SIN
||
850 opcode
== SHADER_OPCODE_COS
||
851 opcode
== SHADER_OPCODE_INT_QUOTIENT
||
852 opcode
== SHADER_OPCODE_INT_REMAINDER
||
853 opcode
== SHADER_OPCODE_POW
);
857 backend_instruction::is_control_flow() const
861 case BRW_OPCODE_WHILE
:
863 case BRW_OPCODE_ELSE
:
864 case BRW_OPCODE_ENDIF
:
865 case BRW_OPCODE_BREAK
:
866 case BRW_OPCODE_CONTINUE
:
874 backend_instruction::can_do_source_mods() const
877 case BRW_OPCODE_ADDC
:
879 case BRW_OPCODE_BFI1
:
880 case BRW_OPCODE_BFI2
:
881 case BRW_OPCODE_BFREV
:
882 case BRW_OPCODE_CBIT
:
885 case BRW_OPCODE_SUBB
:
886 case SHADER_OPCODE_BROADCAST
:
887 case SHADER_OPCODE_CLUSTER_BROADCAST
:
888 case SHADER_OPCODE_MOV_INDIRECT
:
896 backend_instruction::can_do_saturate() const
906 case BRW_OPCODE_F16TO32
:
907 case BRW_OPCODE_F32TO16
:
908 case BRW_OPCODE_LINE
:
912 case BRW_OPCODE_MATH
:
915 case SHADER_OPCODE_MULH
:
917 case BRW_OPCODE_RNDD
:
918 case BRW_OPCODE_RNDE
:
919 case BRW_OPCODE_RNDU
:
920 case BRW_OPCODE_RNDZ
:
924 case FS_OPCODE_LINTERP
:
925 case SHADER_OPCODE_COS
:
926 case SHADER_OPCODE_EXP2
:
927 case SHADER_OPCODE_LOG2
:
928 case SHADER_OPCODE_POW
:
929 case SHADER_OPCODE_RCP
:
930 case SHADER_OPCODE_RSQ
:
931 case SHADER_OPCODE_SIN
:
932 case SHADER_OPCODE_SQRT
:
940 backend_instruction::can_do_cmod() const
944 case BRW_OPCODE_ADDC
:
949 case BRW_OPCODE_CMPN
:
954 case BRW_OPCODE_F16TO32
:
955 case BRW_OPCODE_F32TO16
:
957 case BRW_OPCODE_LINE
:
961 case BRW_OPCODE_MACH
:
968 case BRW_OPCODE_RNDD
:
969 case BRW_OPCODE_RNDE
:
970 case BRW_OPCODE_RNDU
:
971 case BRW_OPCODE_RNDZ
:
972 case BRW_OPCODE_SAD2
:
973 case BRW_OPCODE_SADA2
:
976 case BRW_OPCODE_SUBB
:
978 case FS_OPCODE_LINTERP
:
986 backend_instruction::reads_accumulator_implicitly() const
990 case BRW_OPCODE_MACH
:
991 case BRW_OPCODE_SADA2
:
999 backend_instruction::writes_accumulator_implicitly(const struct gen_device_info
*devinfo
) const
1001 return writes_accumulator
||
1002 (devinfo
->gen
< 6 &&
1003 ((opcode
>= BRW_OPCODE_ADD
&& opcode
< BRW_OPCODE_NOP
) ||
1004 (opcode
>= FS_OPCODE_DDX_COARSE
&& opcode
<= FS_OPCODE_LINTERP
))) ||
1005 (opcode
== FS_OPCODE_LINTERP
&&
1006 (!devinfo
->has_pln
|| devinfo
->gen
<= 6));
1010 backend_instruction::has_side_effects() const
1013 case SHADER_OPCODE_SEND
:
1014 return send_has_side_effects
;
1016 case SHADER_OPCODE_UNTYPED_ATOMIC
:
1017 case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL
:
1018 case SHADER_OPCODE_UNTYPED_ATOMIC_FLOAT
:
1019 case SHADER_OPCODE_UNTYPED_ATOMIC_FLOAT_LOGICAL
:
1020 case SHADER_OPCODE_GEN4_SCRATCH_WRITE
:
1021 case SHADER_OPCODE_UNTYPED_SURFACE_WRITE
:
1022 case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL
:
1023 case SHADER_OPCODE_A64_UNTYPED_WRITE_LOGICAL
:
1024 case SHADER_OPCODE_A64_BYTE_SCATTERED_WRITE_LOGICAL
:
1025 case SHADER_OPCODE_A64_UNTYPED_ATOMIC_LOGICAL
:
1026 case SHADER_OPCODE_A64_UNTYPED_ATOMIC_FLOAT_LOGICAL
:
1027 case SHADER_OPCODE_BYTE_SCATTERED_WRITE
:
1028 case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL
:
1029 case SHADER_OPCODE_TYPED_ATOMIC
:
1030 case SHADER_OPCODE_TYPED_ATOMIC_LOGICAL
:
1031 case SHADER_OPCODE_TYPED_SURFACE_WRITE
:
1032 case SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL
:
1033 case SHADER_OPCODE_MEMORY_FENCE
:
1034 case SHADER_OPCODE_INTERLOCK
:
1035 case SHADER_OPCODE_URB_WRITE_SIMD8
:
1036 case SHADER_OPCODE_URB_WRITE_SIMD8_PER_SLOT
:
1037 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED
:
1038 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT
:
1039 case FS_OPCODE_FB_WRITE
:
1040 case FS_OPCODE_FB_WRITE_LOGICAL
:
1041 case FS_OPCODE_REP_FB_WRITE
:
1042 case SHADER_OPCODE_BARRIER
:
1043 case TCS_OPCODE_URB_WRITE
:
1044 case TCS_OPCODE_RELEASE_INPUT
:
1045 case SHADER_OPCODE_RND_MODE
:
1053 backend_instruction::is_volatile() const
1056 case SHADER_OPCODE_SEND
:
1057 return send_is_volatile
;
1059 case SHADER_OPCODE_UNTYPED_SURFACE_READ
:
1060 case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL
:
1061 case SHADER_OPCODE_TYPED_SURFACE_READ
:
1062 case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL
:
1063 case SHADER_OPCODE_BYTE_SCATTERED_READ
:
1064 case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL
:
1065 case SHADER_OPCODE_A64_UNTYPED_READ_LOGICAL
:
1066 case SHADER_OPCODE_A64_BYTE_SCATTERED_READ_LOGICAL
:
1067 case SHADER_OPCODE_URB_READ_SIMD8
:
1068 case SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT
:
1069 case VEC4_OPCODE_URB_READ
:
1078 inst_is_in_block(const bblock_t
*block
, const backend_instruction
*inst
)
1081 foreach_inst_in_block (backend_instruction
, i
, block
) {
1091 adjust_later_block_ips(bblock_t
*start_block
, int ip_adjustment
)
1093 for (bblock_t
*block_iter
= start_block
->next();
1095 block_iter
= block_iter
->next()) {
1096 block_iter
->start_ip
+= ip_adjustment
;
1097 block_iter
->end_ip
+= ip_adjustment
;
1102 backend_instruction::insert_after(bblock_t
*block
, backend_instruction
*inst
)
1104 assert(this != inst
);
1106 if (!this->is_head_sentinel())
1107 assert(inst_is_in_block(block
, this) || !"Instruction not in block");
1111 adjust_later_block_ips(block
, 1);
1113 exec_node::insert_after(inst
);
1117 backend_instruction::insert_before(bblock_t
*block
, backend_instruction
*inst
)
1119 assert(this != inst
);
1121 if (!this->is_tail_sentinel())
1122 assert(inst_is_in_block(block
, this) || !"Instruction not in block");
1126 adjust_later_block_ips(block
, 1);
1128 exec_node::insert_before(inst
);
1132 backend_instruction::insert_before(bblock_t
*block
, exec_list
*list
)
1134 assert(inst_is_in_block(block
, this) || !"Instruction not in block");
1136 unsigned num_inst
= list
->length();
1138 block
->end_ip
+= num_inst
;
1140 adjust_later_block_ips(block
, num_inst
);
1142 exec_node::insert_before(list
);
1146 backend_instruction::remove(bblock_t
*block
)
1148 assert(inst_is_in_block(block
, this) || !"Instruction not in block");
1150 adjust_later_block_ips(block
, -1);
1152 if (block
->start_ip
== block
->end_ip
) {
1153 block
->cfg
->remove_block(block
);
1158 exec_node::remove();
1162 backend_shader::dump_instructions()
1164 dump_instructions(NULL
);
1168 backend_shader::dump_instructions(const char *name
)
1170 FILE *file
= stderr
;
1171 if (name
&& geteuid() != 0) {
1172 file
= fopen(name
, "w");
1179 foreach_block_and_inst(block
, backend_instruction
, inst
, cfg
) {
1180 if (!unlikely(INTEL_DEBUG
& DEBUG_OPTIMIZER
))
1181 fprintf(file
, "%4d: ", ip
++);
1182 dump_instruction(inst
, file
);
1186 foreach_in_list(backend_instruction
, inst
, &instructions
) {
1187 if (!unlikely(INTEL_DEBUG
& DEBUG_OPTIMIZER
))
1188 fprintf(file
, "%4d: ", ip
++);
1189 dump_instruction(inst
, file
);
1193 if (file
!= stderr
) {
1199 backend_shader::calculate_cfg()
1203 cfg
= new(mem_ctx
) cfg_t(&this->instructions
);
1206 extern "C" const unsigned *
1207 brw_compile_tes(const struct brw_compiler
*compiler
,
1210 const struct brw_tes_prog_key
*key
,
1211 const struct brw_vue_map
*input_vue_map
,
1212 struct brw_tes_prog_data
*prog_data
,
1214 struct gl_program
*prog
,
1215 int shader_time_index
,
1218 const struct gen_device_info
*devinfo
= compiler
->devinfo
;
1219 const bool is_scalar
= compiler
->scalar_stage
[MESA_SHADER_TESS_EVAL
];
1220 const unsigned *assembly
;
1222 nir
->info
.inputs_read
= key
->inputs_read
;
1223 nir
->info
.patch_inputs_read
= key
->patch_inputs_read
;
1225 nir
= brw_nir_apply_sampler_key(nir
, compiler
, &key
->tex
, is_scalar
);
1226 brw_nir_lower_tes_inputs(nir
, input_vue_map
);
1227 brw_nir_lower_vue_outputs(nir
);
1228 nir
= brw_postprocess_nir(nir
, compiler
, is_scalar
);
1230 brw_compute_vue_map(devinfo
, &prog_data
->base
.vue_map
,
1231 nir
->info
.outputs_written
,
1232 nir
->info
.separate_shader
);
1234 unsigned output_size_bytes
= prog_data
->base
.vue_map
.num_slots
* 4 * 4;
1236 assert(output_size_bytes
>= 1);
1237 if (output_size_bytes
> GEN7_MAX_DS_URB_ENTRY_SIZE_BYTES
) {
1239 *error_str
= ralloc_strdup(mem_ctx
, "DS outputs exceed maximum size");
1243 prog_data
->base
.clip_distance_mask
=
1244 ((1 << nir
->info
.clip_distance_array_size
) - 1);
1245 prog_data
->base
.cull_distance_mask
=
1246 ((1 << nir
->info
.cull_distance_array_size
) - 1) <<
1247 nir
->info
.clip_distance_array_size
;
1249 /* URB entry sizes are stored as a multiple of 64 bytes. */
1250 prog_data
->base
.urb_entry_size
= ALIGN(output_size_bytes
, 64) / 64;
1252 /* On Cannonlake software shall not program an allocation size that
1253 * specifies a size that is a multiple of 3 64B (512-bit) cachelines.
1255 if (devinfo
->gen
== 10 &&
1256 prog_data
->base
.urb_entry_size
% 3 == 0)
1257 prog_data
->base
.urb_entry_size
++;
1259 prog_data
->base
.urb_read_length
= 0;
1261 STATIC_ASSERT(BRW_TESS_PARTITIONING_INTEGER
== TESS_SPACING_EQUAL
- 1);
1262 STATIC_ASSERT(BRW_TESS_PARTITIONING_ODD_FRACTIONAL
==
1263 TESS_SPACING_FRACTIONAL_ODD
- 1);
1264 STATIC_ASSERT(BRW_TESS_PARTITIONING_EVEN_FRACTIONAL
==
1265 TESS_SPACING_FRACTIONAL_EVEN
- 1);
1267 prog_data
->partitioning
=
1268 (enum brw_tess_partitioning
) (nir
->info
.tess
.spacing
- 1);
1270 switch (nir
->info
.tess
.primitive_mode
) {
1272 prog_data
->domain
= BRW_TESS_DOMAIN_QUAD
;
1275 prog_data
->domain
= BRW_TESS_DOMAIN_TRI
;
1278 prog_data
->domain
= BRW_TESS_DOMAIN_ISOLINE
;
1281 unreachable("invalid domain shader primitive mode");
1284 if (nir
->info
.tess
.point_mode
) {
1285 prog_data
->output_topology
= BRW_TESS_OUTPUT_TOPOLOGY_POINT
;
1286 } else if (nir
->info
.tess
.primitive_mode
== GL_ISOLINES
) {
1287 prog_data
->output_topology
= BRW_TESS_OUTPUT_TOPOLOGY_LINE
;
1289 /* Hardware winding order is backwards from OpenGL */
1290 prog_data
->output_topology
=
1291 nir
->info
.tess
.ccw
? BRW_TESS_OUTPUT_TOPOLOGY_TRI_CW
1292 : BRW_TESS_OUTPUT_TOPOLOGY_TRI_CCW
;
1295 if (unlikely(INTEL_DEBUG
& DEBUG_TES
)) {
1296 fprintf(stderr
, "TES Input ");
1297 brw_print_vue_map(stderr
, input_vue_map
);
1298 fprintf(stderr
, "TES Output ");
1299 brw_print_vue_map(stderr
, &prog_data
->base
.vue_map
);
1303 fs_visitor
v(compiler
, log_data
, mem_ctx
, (void *) key
,
1304 &prog_data
->base
.base
, NULL
, nir
, 8,
1305 shader_time_index
, input_vue_map
);
1308 *error_str
= ralloc_strdup(mem_ctx
, v
.fail_msg
);
1312 prog_data
->base
.base
.dispatch_grf_start_reg
= v
.payload
.num_regs
;
1313 prog_data
->base
.dispatch_mode
= DISPATCH_MODE_SIMD8
;
1315 fs_generator
g(compiler
, log_data
, mem_ctx
,
1316 &prog_data
->base
.base
, v
.promoted_constants
, false,
1317 MESA_SHADER_TESS_EVAL
);
1318 if (unlikely(INTEL_DEBUG
& DEBUG_TES
)) {
1319 g
.enable_debug(ralloc_asprintf(mem_ctx
,
1320 "%s tessellation evaluation shader %s",
1321 nir
->info
.label
? nir
->info
.label
1326 g
.generate_code(v
.cfg
, 8);
1328 assembly
= g
.get_assembly();
1330 brw::vec4_tes_visitor
v(compiler
, log_data
, key
, prog_data
,
1331 nir
, mem_ctx
, shader_time_index
);
1334 *error_str
= ralloc_strdup(mem_ctx
, v
.fail_msg
);
1338 if (unlikely(INTEL_DEBUG
& DEBUG_TES
))
1339 v
.dump_instructions();
1341 assembly
= brw_vec4_generate_assembly(compiler
, log_data
, mem_ctx
, nir
,
1342 &prog_data
->base
, v
.cfg
);