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 "dev/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_INTERFACE
:
59 case GLSL_TYPE_SAMPLER
:
60 case GLSL_TYPE_ATOMIC_UINT
:
61 /* These should be overridden with the type of the member when
62 * dereferenced into. BRW_REGISTER_TYPE_UD seems like a likely
63 * way to trip up if we don't.
65 return BRW_REGISTER_TYPE_UD
;
67 return BRW_REGISTER_TYPE_UD
;
68 case GLSL_TYPE_DOUBLE
:
69 return BRW_REGISTER_TYPE_DF
;
70 case GLSL_TYPE_UINT64
:
71 return BRW_REGISTER_TYPE_UQ
;
73 return BRW_REGISTER_TYPE_Q
;
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(const nir_tex_instr
*tex
, unsigned src
,
133 uint32_t *offset_bits_out
)
135 if (!nir_src_is_const(tex
->src
[src
].src
))
138 const unsigned num_components
= nir_tex_instr_src_size(tex
, src
);
140 /* Combine all three offsets into a single unsigned dword:
142 * bits 11:8 - U Offset (X component)
143 * bits 7:4 - V Offset (Y component)
144 * bits 3:0 - R Offset (Z component)
146 uint32_t offset_bits
= 0;
147 for (unsigned i
= 0; i
< num_components
; i
++) {
148 int offset
= nir_src_comp_as_int(tex
->src
[src
].src
, i
);
150 /* offset out of bounds; caller will handle it. */
151 if (offset
> 7 || offset
< -8)
154 const unsigned shift
= 4 * (2 - i
);
155 offset_bits
|= (offset
<< shift
) & (0xF << shift
);
158 *offset_bits_out
= offset_bits
;
164 brw_instruction_name(const struct gen_device_info
*devinfo
, enum opcode op
)
167 case BRW_OPCODE_ILLEGAL
... BRW_OPCODE_NOP
:
168 /* The DO instruction doesn't exist on Gen6+, but we use it to mark the
169 * start of a loop in the IR.
171 if (devinfo
->gen
>= 6 && op
== BRW_OPCODE_DO
)
174 /* The following conversion opcodes doesn't exist on Gen8+, but we use
175 * then to mark that we want to do the conversion.
177 if (devinfo
->gen
> 7 && op
== BRW_OPCODE_F32TO16
)
180 if (devinfo
->gen
> 7 && op
== BRW_OPCODE_F16TO32
)
183 assert(brw_opcode_desc(devinfo
, op
)->name
);
184 return brw_opcode_desc(devinfo
, op
)->name
;
185 case FS_OPCODE_FB_WRITE
:
187 case FS_OPCODE_FB_WRITE_LOGICAL
:
188 return "fb_write_logical";
189 case FS_OPCODE_REP_FB_WRITE
:
190 return "rep_fb_write";
191 case FS_OPCODE_FB_READ
:
193 case FS_OPCODE_FB_READ_LOGICAL
:
194 return "fb_read_logical";
196 case SHADER_OPCODE_RCP
:
198 case SHADER_OPCODE_RSQ
:
200 case SHADER_OPCODE_SQRT
:
202 case SHADER_OPCODE_EXP2
:
204 case SHADER_OPCODE_LOG2
:
206 case SHADER_OPCODE_POW
:
208 case SHADER_OPCODE_INT_QUOTIENT
:
210 case SHADER_OPCODE_INT_REMAINDER
:
212 case SHADER_OPCODE_SIN
:
214 case SHADER_OPCODE_COS
:
217 case SHADER_OPCODE_SEND
:
220 case SHADER_OPCODE_TEX
:
222 case SHADER_OPCODE_TEX_LOGICAL
:
223 return "tex_logical";
224 case SHADER_OPCODE_TXD
:
226 case SHADER_OPCODE_TXD_LOGICAL
:
227 return "txd_logical";
228 case SHADER_OPCODE_TXF
:
230 case SHADER_OPCODE_TXF_LOGICAL
:
231 return "txf_logical";
232 case SHADER_OPCODE_TXF_LZ
:
234 case SHADER_OPCODE_TXL
:
236 case SHADER_OPCODE_TXL_LOGICAL
:
237 return "txl_logical";
238 case SHADER_OPCODE_TXL_LZ
:
240 case SHADER_OPCODE_TXS
:
242 case SHADER_OPCODE_TXS_LOGICAL
:
243 return "txs_logical";
246 case FS_OPCODE_TXB_LOGICAL
:
247 return "txb_logical";
248 case SHADER_OPCODE_TXF_CMS
:
250 case SHADER_OPCODE_TXF_CMS_LOGICAL
:
251 return "txf_cms_logical";
252 case SHADER_OPCODE_TXF_CMS_W
:
254 case SHADER_OPCODE_TXF_CMS_W_LOGICAL
:
255 return "txf_cms_w_logical";
256 case SHADER_OPCODE_TXF_UMS
:
258 case SHADER_OPCODE_TXF_UMS_LOGICAL
:
259 return "txf_ums_logical";
260 case SHADER_OPCODE_TXF_MCS
:
262 case SHADER_OPCODE_TXF_MCS_LOGICAL
:
263 return "txf_mcs_logical";
264 case SHADER_OPCODE_LOD
:
266 case SHADER_OPCODE_LOD_LOGICAL
:
267 return "lod_logical";
268 case SHADER_OPCODE_TG4
:
270 case SHADER_OPCODE_TG4_LOGICAL
:
271 return "tg4_logical";
272 case SHADER_OPCODE_TG4_OFFSET
:
274 case SHADER_OPCODE_TG4_OFFSET_LOGICAL
:
275 return "tg4_offset_logical";
276 case SHADER_OPCODE_SAMPLEINFO
:
278 case SHADER_OPCODE_SAMPLEINFO_LOGICAL
:
279 return "sampleinfo_logical";
281 case SHADER_OPCODE_IMAGE_SIZE_LOGICAL
:
282 return "image_size_logical";
284 case SHADER_OPCODE_SHADER_TIME_ADD
:
285 return "shader_time_add";
287 case VEC4_OPCODE_UNTYPED_ATOMIC
:
288 return "untyped_atomic";
289 case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL
:
290 return "untyped_atomic_logical";
291 case SHADER_OPCODE_UNTYPED_ATOMIC_FLOAT_LOGICAL
:
292 return "untyped_atomic_float_logical";
293 case VEC4_OPCODE_UNTYPED_SURFACE_READ
:
294 return "untyped_surface_read";
295 case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL
:
296 return "untyped_surface_read_logical";
297 case VEC4_OPCODE_UNTYPED_SURFACE_WRITE
:
298 return "untyped_surface_write";
299 case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL
:
300 return "untyped_surface_write_logical";
301 case SHADER_OPCODE_A64_UNTYPED_READ_LOGICAL
:
302 return "a64_untyped_read_logical";
303 case SHADER_OPCODE_A64_UNTYPED_WRITE_LOGICAL
:
304 return "a64_untyped_write_logical";
305 case SHADER_OPCODE_A64_BYTE_SCATTERED_READ_LOGICAL
:
306 return "a64_byte_scattered_read_logical";
307 case SHADER_OPCODE_A64_BYTE_SCATTERED_WRITE_LOGICAL
:
308 return "a64_byte_scattered_write_logical";
309 case SHADER_OPCODE_A64_UNTYPED_ATOMIC_LOGICAL
:
310 return "a64_untyped_atomic_logical";
311 case SHADER_OPCODE_A64_UNTYPED_ATOMIC_INT64_LOGICAL
:
312 return "a64_untyped_atomic_int64_logical";
313 case SHADER_OPCODE_A64_UNTYPED_ATOMIC_FLOAT_LOGICAL
:
314 return "a64_untyped_atomic_float_logical";
315 case SHADER_OPCODE_TYPED_ATOMIC_LOGICAL
:
316 return "typed_atomic_logical";
317 case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL
:
318 return "typed_surface_read_logical";
319 case SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL
:
320 return "typed_surface_write_logical";
321 case SHADER_OPCODE_MEMORY_FENCE
:
322 return "memory_fence";
323 case SHADER_OPCODE_INTERLOCK
:
324 /* For an interlock we actually issue a memory fence via sendc. */
327 case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL
:
328 return "byte_scattered_read_logical";
329 case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL
:
330 return "byte_scattered_write_logical";
332 case SHADER_OPCODE_LOAD_PAYLOAD
:
333 return "load_payload";
337 case SHADER_OPCODE_GEN4_SCRATCH_READ
:
338 return "gen4_scratch_read";
339 case SHADER_OPCODE_GEN4_SCRATCH_WRITE
:
340 return "gen4_scratch_write";
341 case SHADER_OPCODE_GEN7_SCRATCH_READ
:
342 return "gen7_scratch_read";
343 case SHADER_OPCODE_URB_WRITE_SIMD8
:
344 return "gen8_urb_write_simd8";
345 case SHADER_OPCODE_URB_WRITE_SIMD8_PER_SLOT
:
346 return "gen8_urb_write_simd8_per_slot";
347 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED
:
348 return "gen8_urb_write_simd8_masked";
349 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT
:
350 return "gen8_urb_write_simd8_masked_per_slot";
351 case SHADER_OPCODE_URB_READ_SIMD8
:
352 return "urb_read_simd8";
353 case SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT
:
354 return "urb_read_simd8_per_slot";
356 case SHADER_OPCODE_FIND_LIVE_CHANNEL
:
357 return "find_live_channel";
358 case SHADER_OPCODE_BROADCAST
:
360 case SHADER_OPCODE_SHUFFLE
:
362 case SHADER_OPCODE_SEL_EXEC
:
364 case SHADER_OPCODE_QUAD_SWIZZLE
:
365 return "quad_swizzle";
366 case SHADER_OPCODE_CLUSTER_BROADCAST
:
367 return "cluster_broadcast";
369 case SHADER_OPCODE_GET_BUFFER_SIZE
:
370 return "get_buffer_size";
372 case VEC4_OPCODE_MOV_BYTES
:
374 case VEC4_OPCODE_PACK_BYTES
:
376 case VEC4_OPCODE_UNPACK_UNIFORM
:
377 return "unpack_uniform";
378 case VEC4_OPCODE_DOUBLE_TO_F32
:
379 return "double_to_f32";
380 case VEC4_OPCODE_DOUBLE_TO_D32
:
381 return "double_to_d32";
382 case VEC4_OPCODE_DOUBLE_TO_U32
:
383 return "double_to_u32";
384 case VEC4_OPCODE_TO_DOUBLE
:
385 return "single_to_double";
386 case VEC4_OPCODE_PICK_LOW_32BIT
:
387 return "pick_low_32bit";
388 case VEC4_OPCODE_PICK_HIGH_32BIT
:
389 return "pick_high_32bit";
390 case VEC4_OPCODE_SET_LOW_32BIT
:
391 return "set_low_32bit";
392 case VEC4_OPCODE_SET_HIGH_32BIT
:
393 return "set_high_32bit";
395 case FS_OPCODE_DDX_COARSE
:
397 case FS_OPCODE_DDX_FINE
:
399 case FS_OPCODE_DDY_COARSE
:
401 case FS_OPCODE_DDY_FINE
:
404 case FS_OPCODE_LINTERP
:
407 case FS_OPCODE_PIXEL_X
:
409 case FS_OPCODE_PIXEL_Y
:
412 case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD
:
413 return "uniform_pull_const";
414 case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD_GEN7
:
415 return "uniform_pull_const_gen7";
416 case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN4
:
417 return "varying_pull_const_gen4";
418 case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_LOGICAL
:
419 return "varying_pull_const_logical";
421 case FS_OPCODE_DISCARD_JUMP
:
422 return "discard_jump";
424 case FS_OPCODE_SET_SAMPLE_ID
:
425 return "set_sample_id";
427 case FS_OPCODE_PACK_HALF_2x16_SPLIT
:
428 return "pack_half_2x16_split";
430 case FS_OPCODE_PLACEHOLDER_HALT
:
431 return "placeholder_halt";
433 case FS_OPCODE_INTERPOLATE_AT_SAMPLE
:
434 return "interp_sample";
435 case FS_OPCODE_INTERPOLATE_AT_SHARED_OFFSET
:
436 return "interp_shared_offset";
437 case FS_OPCODE_INTERPOLATE_AT_PER_SLOT_OFFSET
:
438 return "interp_per_slot_offset";
440 case VS_OPCODE_URB_WRITE
:
441 return "vs_urb_write";
442 case VS_OPCODE_PULL_CONSTANT_LOAD
:
443 return "pull_constant_load";
444 case VS_OPCODE_PULL_CONSTANT_LOAD_GEN7
:
445 return "pull_constant_load_gen7";
447 case VS_OPCODE_SET_SIMD4X2_HEADER_GEN9
:
448 return "set_simd4x2_header_gen9";
450 case VS_OPCODE_UNPACK_FLAGS_SIMD4X2
:
451 return "unpack_flags_simd4x2";
453 case GS_OPCODE_URB_WRITE
:
454 return "gs_urb_write";
455 case GS_OPCODE_URB_WRITE_ALLOCATE
:
456 return "gs_urb_write_allocate";
457 case GS_OPCODE_THREAD_END
:
458 return "gs_thread_end";
459 case GS_OPCODE_SET_WRITE_OFFSET
:
460 return "set_write_offset";
461 case GS_OPCODE_SET_VERTEX_COUNT
:
462 return "set_vertex_count";
463 case GS_OPCODE_SET_DWORD_2
:
464 return "set_dword_2";
465 case GS_OPCODE_PREPARE_CHANNEL_MASKS
:
466 return "prepare_channel_masks";
467 case GS_OPCODE_SET_CHANNEL_MASKS
:
468 return "set_channel_masks";
469 case GS_OPCODE_GET_INSTANCE_ID
:
470 return "get_instance_id";
471 case GS_OPCODE_FF_SYNC
:
473 case GS_OPCODE_SET_PRIMITIVE_ID
:
474 return "set_primitive_id";
475 case GS_OPCODE_SVB_WRITE
:
476 return "gs_svb_write";
477 case GS_OPCODE_SVB_SET_DST_INDEX
:
478 return "gs_svb_set_dst_index";
479 case GS_OPCODE_FF_SYNC_SET_PRIMITIVES
:
480 return "gs_ff_sync_set_primitives";
481 case CS_OPCODE_CS_TERMINATE
:
482 return "cs_terminate";
483 case SHADER_OPCODE_BARRIER
:
485 case SHADER_OPCODE_MULH
:
487 case SHADER_OPCODE_MOV_INDIRECT
:
488 return "mov_indirect";
490 case VEC4_OPCODE_URB_READ
:
492 case TCS_OPCODE_GET_INSTANCE_ID
:
493 return "tcs_get_instance_id";
494 case TCS_OPCODE_URB_WRITE
:
495 return "tcs_urb_write";
496 case TCS_OPCODE_SET_INPUT_URB_OFFSETS
:
497 return "tcs_set_input_urb_offsets";
498 case TCS_OPCODE_SET_OUTPUT_URB_OFFSETS
:
499 return "tcs_set_output_urb_offsets";
500 case TCS_OPCODE_GET_PRIMITIVE_ID
:
501 return "tcs_get_primitive_id";
502 case TCS_OPCODE_CREATE_BARRIER_HEADER
:
503 return "tcs_create_barrier_header";
504 case TCS_OPCODE_SRC0_010_IS_ZERO
:
505 return "tcs_src0<0,1,0>_is_zero";
506 case TCS_OPCODE_RELEASE_INPUT
:
507 return "tcs_release_input";
508 case TCS_OPCODE_THREAD_END
:
509 return "tcs_thread_end";
510 case TES_OPCODE_CREATE_INPUT_READ_HEADER
:
511 return "tes_create_input_read_header";
512 case TES_OPCODE_ADD_INDIRECT_URB_OFFSET
:
513 return "tes_add_indirect_urb_offset";
514 case TES_OPCODE_GET_PRIMITIVE_ID
:
515 return "tes_get_primitive_id";
517 case SHADER_OPCODE_RND_MODE
:
521 unreachable("not reached");
525 brw_saturate_immediate(enum brw_reg_type type
, struct brw_reg
*reg
)
532 } imm
, sat_imm
= { 0 };
534 const unsigned size
= type_sz(type
);
536 /* We want to either do a 32-bit or 64-bit data copy, the type is otherwise
537 * irrelevant, so just check the size of the type and copy from/to an
538 * appropriately sized field.
546 case BRW_REGISTER_TYPE_UD
:
547 case BRW_REGISTER_TYPE_D
:
548 case BRW_REGISTER_TYPE_UW
:
549 case BRW_REGISTER_TYPE_W
:
550 case BRW_REGISTER_TYPE_UQ
:
551 case BRW_REGISTER_TYPE_Q
:
554 case BRW_REGISTER_TYPE_F
:
555 sat_imm
.f
= CLAMP(imm
.f
, 0.0f
, 1.0f
);
557 case BRW_REGISTER_TYPE_DF
:
558 sat_imm
.df
= CLAMP(imm
.df
, 0.0, 1.0);
560 case BRW_REGISTER_TYPE_UB
:
561 case BRW_REGISTER_TYPE_B
:
562 unreachable("no UB/B immediates");
563 case BRW_REGISTER_TYPE_V
:
564 case BRW_REGISTER_TYPE_UV
:
565 case BRW_REGISTER_TYPE_VF
:
566 unreachable("unimplemented: saturate vector immediate");
567 case BRW_REGISTER_TYPE_HF
:
568 unreachable("unimplemented: saturate HF immediate");
569 case BRW_REGISTER_TYPE_NF
:
570 unreachable("no NF immediates");
574 if (imm
.ud
!= sat_imm
.ud
) {
575 reg
->ud
= sat_imm
.ud
;
579 if (imm
.df
!= sat_imm
.df
) {
580 reg
->df
= sat_imm
.df
;
588 brw_negate_immediate(enum brw_reg_type type
, struct brw_reg
*reg
)
591 case BRW_REGISTER_TYPE_D
:
592 case BRW_REGISTER_TYPE_UD
:
595 case BRW_REGISTER_TYPE_W
:
596 case BRW_REGISTER_TYPE_UW
: {
597 uint16_t value
= -(int16_t)reg
->ud
;
598 reg
->ud
= value
| (uint32_t)value
<< 16;
601 case BRW_REGISTER_TYPE_F
:
604 case BRW_REGISTER_TYPE_VF
:
605 reg
->ud
^= 0x80808080;
607 case BRW_REGISTER_TYPE_DF
:
610 case BRW_REGISTER_TYPE_UQ
:
611 case BRW_REGISTER_TYPE_Q
:
612 reg
->d64
= -reg
->d64
;
614 case BRW_REGISTER_TYPE_UB
:
615 case BRW_REGISTER_TYPE_B
:
616 unreachable("no UB/B immediates");
617 case BRW_REGISTER_TYPE_UV
:
618 case BRW_REGISTER_TYPE_V
:
619 assert(!"unimplemented: negate UV/V immediate");
620 case BRW_REGISTER_TYPE_HF
:
621 reg
->ud
^= 0x80008000;
623 case BRW_REGISTER_TYPE_NF
:
624 unreachable("no NF immediates");
631 brw_abs_immediate(enum brw_reg_type type
, struct brw_reg
*reg
)
634 case BRW_REGISTER_TYPE_D
:
635 reg
->d
= abs(reg
->d
);
637 case BRW_REGISTER_TYPE_W
: {
638 uint16_t value
= abs((int16_t)reg
->ud
);
639 reg
->ud
= value
| (uint32_t)value
<< 16;
642 case BRW_REGISTER_TYPE_F
:
643 reg
->f
= fabsf(reg
->f
);
645 case BRW_REGISTER_TYPE_DF
:
646 reg
->df
= fabs(reg
->df
);
648 case BRW_REGISTER_TYPE_VF
:
649 reg
->ud
&= ~0x80808080;
651 case BRW_REGISTER_TYPE_Q
:
652 reg
->d64
= imaxabs(reg
->d64
);
654 case BRW_REGISTER_TYPE_UB
:
655 case BRW_REGISTER_TYPE_B
:
656 unreachable("no UB/B immediates");
657 case BRW_REGISTER_TYPE_UQ
:
658 case BRW_REGISTER_TYPE_UD
:
659 case BRW_REGISTER_TYPE_UW
:
660 case BRW_REGISTER_TYPE_UV
:
661 /* Presumably the absolute value modifier on an unsigned source is a
662 * nop, but it would be nice to confirm.
664 assert(!"unimplemented: abs unsigned immediate");
665 case BRW_REGISTER_TYPE_V
:
666 assert(!"unimplemented: abs V immediate");
667 case BRW_REGISTER_TYPE_HF
:
668 reg
->ud
&= ~0x80008000;
670 case BRW_REGISTER_TYPE_NF
:
671 unreachable("no NF immediates");
677 backend_shader::backend_shader(const struct brw_compiler
*compiler
,
680 const nir_shader
*shader
,
681 struct brw_stage_prog_data
*stage_prog_data
)
682 : compiler(compiler
),
684 devinfo(compiler
->devinfo
),
686 stage_prog_data(stage_prog_data
),
689 stage(shader
->info
.stage
)
691 debug_enabled
= INTEL_DEBUG
& intel_debug_flag_for_shader_stage(stage
);
692 stage_name
= _mesa_shader_stage_to_string(stage
);
693 stage_abbrev
= _mesa_shader_stage_to_abbrev(stage
);
696 backend_shader::~backend_shader()
701 backend_reg::equals(const backend_reg
&r
) const
703 return brw_regs_equal(this, &r
) && offset
== r
.offset
;
707 backend_reg::negative_equals(const backend_reg
&r
) const
709 return brw_regs_negative_equal(this, &r
) && offset
== r
.offset
;
713 backend_reg::is_zero() const
718 assert(type_sz(type
) > 1);
721 case BRW_REGISTER_TYPE_HF
:
722 assert((d
& 0xffff) == ((d
>> 16) & 0xffff));
723 return (d
& 0xffff) == 0 || (d
& 0xffff) == 0x8000;
724 case BRW_REGISTER_TYPE_F
:
726 case BRW_REGISTER_TYPE_DF
:
728 case BRW_REGISTER_TYPE_W
:
729 case BRW_REGISTER_TYPE_UW
:
730 assert((d
& 0xffff) == ((d
>> 16) & 0xffff));
731 return (d
& 0xffff) == 0;
732 case BRW_REGISTER_TYPE_D
:
733 case BRW_REGISTER_TYPE_UD
:
735 case BRW_REGISTER_TYPE_UQ
:
736 case BRW_REGISTER_TYPE_Q
:
744 backend_reg::is_one() const
749 assert(type_sz(type
) > 1);
752 case BRW_REGISTER_TYPE_HF
:
753 assert((d
& 0xffff) == ((d
>> 16) & 0xffff));
754 return (d
& 0xffff) == 0x3c00;
755 case BRW_REGISTER_TYPE_F
:
757 case BRW_REGISTER_TYPE_DF
:
759 case BRW_REGISTER_TYPE_W
:
760 case BRW_REGISTER_TYPE_UW
:
761 assert((d
& 0xffff) == ((d
>> 16) & 0xffff));
762 return (d
& 0xffff) == 1;
763 case BRW_REGISTER_TYPE_D
:
764 case BRW_REGISTER_TYPE_UD
:
766 case BRW_REGISTER_TYPE_UQ
:
767 case BRW_REGISTER_TYPE_Q
:
775 backend_reg::is_negative_one() const
780 assert(type_sz(type
) > 1);
783 case BRW_REGISTER_TYPE_HF
:
784 assert((d
& 0xffff) == ((d
>> 16) & 0xffff));
785 return (d
& 0xffff) == 0xbc00;
786 case BRW_REGISTER_TYPE_F
:
788 case BRW_REGISTER_TYPE_DF
:
790 case BRW_REGISTER_TYPE_W
:
791 assert((d
& 0xffff) == ((d
>> 16) & 0xffff));
792 return (d
& 0xffff) == 0xffff;
793 case BRW_REGISTER_TYPE_D
:
795 case BRW_REGISTER_TYPE_Q
:
803 backend_reg::is_null() const
805 return file
== ARF
&& nr
== BRW_ARF_NULL
;
810 backend_reg::is_accumulator() const
812 return file
== ARF
&& nr
== BRW_ARF_ACCUMULATOR
;
816 backend_instruction::is_commutative() const
824 case SHADER_OPCODE_MULH
:
827 /* MIN and MAX are commutative. */
828 if (conditional_mod
== BRW_CONDITIONAL_GE
||
829 conditional_mod
== BRW_CONDITIONAL_L
) {
839 backend_instruction::is_3src(const struct gen_device_info
*devinfo
) const
841 return ::is_3src(devinfo
, opcode
);
845 backend_instruction::is_tex() const
847 return (opcode
== SHADER_OPCODE_TEX
||
848 opcode
== FS_OPCODE_TXB
||
849 opcode
== SHADER_OPCODE_TXD
||
850 opcode
== SHADER_OPCODE_TXF
||
851 opcode
== SHADER_OPCODE_TXF_LZ
||
852 opcode
== SHADER_OPCODE_TXF_CMS
||
853 opcode
== SHADER_OPCODE_TXF_CMS_W
||
854 opcode
== SHADER_OPCODE_TXF_UMS
||
855 opcode
== SHADER_OPCODE_TXF_MCS
||
856 opcode
== SHADER_OPCODE_TXL
||
857 opcode
== SHADER_OPCODE_TXL_LZ
||
858 opcode
== SHADER_OPCODE_TXS
||
859 opcode
== SHADER_OPCODE_LOD
||
860 opcode
== SHADER_OPCODE_TG4
||
861 opcode
== SHADER_OPCODE_TG4_OFFSET
||
862 opcode
== SHADER_OPCODE_SAMPLEINFO
);
866 backend_instruction::is_math() const
868 return (opcode
== SHADER_OPCODE_RCP
||
869 opcode
== SHADER_OPCODE_RSQ
||
870 opcode
== SHADER_OPCODE_SQRT
||
871 opcode
== SHADER_OPCODE_EXP2
||
872 opcode
== SHADER_OPCODE_LOG2
||
873 opcode
== SHADER_OPCODE_SIN
||
874 opcode
== SHADER_OPCODE_COS
||
875 opcode
== SHADER_OPCODE_INT_QUOTIENT
||
876 opcode
== SHADER_OPCODE_INT_REMAINDER
||
877 opcode
== SHADER_OPCODE_POW
);
881 backend_instruction::is_control_flow() const
885 case BRW_OPCODE_WHILE
:
887 case BRW_OPCODE_ELSE
:
888 case BRW_OPCODE_ENDIF
:
889 case BRW_OPCODE_BREAK
:
890 case BRW_OPCODE_CONTINUE
:
898 backend_instruction::can_do_source_mods() const
901 case BRW_OPCODE_ADDC
:
903 case BRW_OPCODE_BFI1
:
904 case BRW_OPCODE_BFI2
:
905 case BRW_OPCODE_BFREV
:
906 case BRW_OPCODE_CBIT
:
909 case BRW_OPCODE_SUBB
:
910 case SHADER_OPCODE_BROADCAST
:
911 case SHADER_OPCODE_CLUSTER_BROADCAST
:
912 case SHADER_OPCODE_MOV_INDIRECT
:
920 backend_instruction::can_do_saturate() const
930 case BRW_OPCODE_F16TO32
:
931 case BRW_OPCODE_F32TO16
:
932 case BRW_OPCODE_LINE
:
936 case BRW_OPCODE_MATH
:
939 case SHADER_OPCODE_MULH
:
941 case BRW_OPCODE_RNDD
:
942 case BRW_OPCODE_RNDE
:
943 case BRW_OPCODE_RNDU
:
944 case BRW_OPCODE_RNDZ
:
948 case FS_OPCODE_LINTERP
:
949 case SHADER_OPCODE_COS
:
950 case SHADER_OPCODE_EXP2
:
951 case SHADER_OPCODE_LOG2
:
952 case SHADER_OPCODE_POW
:
953 case SHADER_OPCODE_RCP
:
954 case SHADER_OPCODE_RSQ
:
955 case SHADER_OPCODE_SIN
:
956 case SHADER_OPCODE_SQRT
:
964 backend_instruction::can_do_cmod() const
968 case BRW_OPCODE_ADDC
:
973 case BRW_OPCODE_CMPN
:
978 case BRW_OPCODE_F16TO32
:
979 case BRW_OPCODE_F32TO16
:
981 case BRW_OPCODE_LINE
:
985 case BRW_OPCODE_MACH
:
992 case BRW_OPCODE_RNDD
:
993 case BRW_OPCODE_RNDE
:
994 case BRW_OPCODE_RNDU
:
995 case BRW_OPCODE_RNDZ
:
996 case BRW_OPCODE_SAD2
:
997 case BRW_OPCODE_SADA2
:
1000 case BRW_OPCODE_SUBB
:
1001 case BRW_OPCODE_XOR
:
1002 case FS_OPCODE_LINTERP
:
1010 backend_instruction::reads_accumulator_implicitly() const
1013 case BRW_OPCODE_MAC
:
1014 case BRW_OPCODE_MACH
:
1015 case BRW_OPCODE_SADA2
:
1023 backend_instruction::writes_accumulator_implicitly(const struct gen_device_info
*devinfo
) const
1025 return writes_accumulator
||
1026 (devinfo
->gen
< 6 &&
1027 ((opcode
>= BRW_OPCODE_ADD
&& opcode
< BRW_OPCODE_NOP
) ||
1028 (opcode
>= FS_OPCODE_DDX_COARSE
&& opcode
<= FS_OPCODE_LINTERP
))) ||
1029 (opcode
== FS_OPCODE_LINTERP
&&
1030 (!devinfo
->has_pln
|| devinfo
->gen
<= 6));
1034 backend_instruction::has_side_effects() const
1037 case SHADER_OPCODE_SEND
:
1038 return send_has_side_effects
;
1040 case VEC4_OPCODE_UNTYPED_ATOMIC
:
1041 case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL
:
1042 case SHADER_OPCODE_UNTYPED_ATOMIC_FLOAT_LOGICAL
:
1043 case SHADER_OPCODE_GEN4_SCRATCH_WRITE
:
1044 case VEC4_OPCODE_UNTYPED_SURFACE_WRITE
:
1045 case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL
:
1046 case SHADER_OPCODE_A64_UNTYPED_WRITE_LOGICAL
:
1047 case SHADER_OPCODE_A64_BYTE_SCATTERED_WRITE_LOGICAL
:
1048 case SHADER_OPCODE_A64_UNTYPED_ATOMIC_LOGICAL
:
1049 case SHADER_OPCODE_A64_UNTYPED_ATOMIC_INT64_LOGICAL
:
1050 case SHADER_OPCODE_A64_UNTYPED_ATOMIC_FLOAT_LOGICAL
:
1051 case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL
:
1052 case SHADER_OPCODE_TYPED_ATOMIC_LOGICAL
:
1053 case SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL
:
1054 case SHADER_OPCODE_MEMORY_FENCE
:
1055 case SHADER_OPCODE_INTERLOCK
:
1056 case SHADER_OPCODE_URB_WRITE_SIMD8
:
1057 case SHADER_OPCODE_URB_WRITE_SIMD8_PER_SLOT
:
1058 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED
:
1059 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT
:
1060 case FS_OPCODE_FB_WRITE
:
1061 case FS_OPCODE_FB_WRITE_LOGICAL
:
1062 case FS_OPCODE_REP_FB_WRITE
:
1063 case SHADER_OPCODE_BARRIER
:
1064 case TCS_OPCODE_URB_WRITE
:
1065 case TCS_OPCODE_RELEASE_INPUT
:
1066 case SHADER_OPCODE_RND_MODE
:
1074 backend_instruction::is_volatile() const
1077 case SHADER_OPCODE_SEND
:
1078 return send_is_volatile
;
1080 case VEC4_OPCODE_UNTYPED_SURFACE_READ
:
1081 case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL
:
1082 case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL
:
1083 case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL
:
1084 case SHADER_OPCODE_A64_UNTYPED_READ_LOGICAL
:
1085 case SHADER_OPCODE_A64_BYTE_SCATTERED_READ_LOGICAL
:
1086 case SHADER_OPCODE_URB_READ_SIMD8
:
1087 case SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT
:
1088 case VEC4_OPCODE_URB_READ
:
1097 inst_is_in_block(const bblock_t
*block
, const backend_instruction
*inst
)
1100 foreach_inst_in_block (backend_instruction
, i
, block
) {
1110 adjust_later_block_ips(bblock_t
*start_block
, int ip_adjustment
)
1112 for (bblock_t
*block_iter
= start_block
->next();
1114 block_iter
= block_iter
->next()) {
1115 block_iter
->start_ip
+= ip_adjustment
;
1116 block_iter
->end_ip
+= ip_adjustment
;
1121 backend_instruction::insert_after(bblock_t
*block
, backend_instruction
*inst
)
1123 assert(this != inst
);
1125 if (!this->is_head_sentinel())
1126 assert(inst_is_in_block(block
, this) || !"Instruction not in block");
1130 adjust_later_block_ips(block
, 1);
1132 exec_node::insert_after(inst
);
1136 backend_instruction::insert_before(bblock_t
*block
, backend_instruction
*inst
)
1138 assert(this != inst
);
1140 if (!this->is_tail_sentinel())
1141 assert(inst_is_in_block(block
, this) || !"Instruction not in block");
1145 adjust_later_block_ips(block
, 1);
1147 exec_node::insert_before(inst
);
1151 backend_instruction::insert_before(bblock_t
*block
, exec_list
*list
)
1153 assert(inst_is_in_block(block
, this) || !"Instruction not in block");
1155 unsigned num_inst
= list
->length();
1157 block
->end_ip
+= num_inst
;
1159 adjust_later_block_ips(block
, num_inst
);
1161 exec_node::insert_before(list
);
1165 backend_instruction::remove(bblock_t
*block
)
1167 assert(inst_is_in_block(block
, this) || !"Instruction not in block");
1169 adjust_later_block_ips(block
, -1);
1171 if (block
->start_ip
== block
->end_ip
) {
1172 block
->cfg
->remove_block(block
);
1177 exec_node::remove();
1181 backend_shader::dump_instructions()
1183 dump_instructions(NULL
);
1187 backend_shader::dump_instructions(const char *name
)
1189 FILE *file
= stderr
;
1190 if (name
&& geteuid() != 0) {
1191 file
= fopen(name
, "w");
1198 foreach_block_and_inst(block
, backend_instruction
, inst
, cfg
) {
1199 if (!unlikely(INTEL_DEBUG
& DEBUG_OPTIMIZER
))
1200 fprintf(file
, "%4d: ", ip
++);
1201 dump_instruction(inst
, file
);
1205 foreach_in_list(backend_instruction
, inst
, &instructions
) {
1206 if (!unlikely(INTEL_DEBUG
& DEBUG_OPTIMIZER
))
1207 fprintf(file
, "%4d: ", ip
++);
1208 dump_instruction(inst
, file
);
1212 if (file
!= stderr
) {
1218 backend_shader::calculate_cfg()
1222 cfg
= new(mem_ctx
) cfg_t(&this->instructions
);
1225 extern "C" const unsigned *
1226 brw_compile_tes(const struct brw_compiler
*compiler
,
1229 const struct brw_tes_prog_key
*key
,
1230 const struct brw_vue_map
*input_vue_map
,
1231 struct brw_tes_prog_data
*prog_data
,
1233 struct gl_program
*prog
,
1234 int shader_time_index
,
1237 const struct gen_device_info
*devinfo
= compiler
->devinfo
;
1238 const bool is_scalar
= compiler
->scalar_stage
[MESA_SHADER_TESS_EVAL
];
1239 const unsigned *assembly
;
1241 nir
->info
.inputs_read
= key
->inputs_read
;
1242 nir
->info
.patch_inputs_read
= key
->patch_inputs_read
;
1244 nir
= brw_nir_apply_sampler_key(nir
, compiler
, &key
->tex
, is_scalar
);
1245 brw_nir_lower_tes_inputs(nir
, input_vue_map
);
1246 brw_nir_lower_vue_outputs(nir
);
1247 nir
= brw_postprocess_nir(nir
, compiler
, is_scalar
);
1249 brw_compute_vue_map(devinfo
, &prog_data
->base
.vue_map
,
1250 nir
->info
.outputs_written
,
1251 nir
->info
.separate_shader
);
1253 unsigned output_size_bytes
= prog_data
->base
.vue_map
.num_slots
* 4 * 4;
1255 assert(output_size_bytes
>= 1);
1256 if (output_size_bytes
> GEN7_MAX_DS_URB_ENTRY_SIZE_BYTES
) {
1258 *error_str
= ralloc_strdup(mem_ctx
, "DS outputs exceed maximum size");
1262 prog_data
->base
.clip_distance_mask
=
1263 ((1 << nir
->info
.clip_distance_array_size
) - 1);
1264 prog_data
->base
.cull_distance_mask
=
1265 ((1 << nir
->info
.cull_distance_array_size
) - 1) <<
1266 nir
->info
.clip_distance_array_size
;
1268 /* URB entry sizes are stored as a multiple of 64 bytes. */
1269 prog_data
->base
.urb_entry_size
= ALIGN(output_size_bytes
, 64) / 64;
1271 /* On Cannonlake software shall not program an allocation size that
1272 * specifies a size that is a multiple of 3 64B (512-bit) cachelines.
1274 if (devinfo
->gen
== 10 &&
1275 prog_data
->base
.urb_entry_size
% 3 == 0)
1276 prog_data
->base
.urb_entry_size
++;
1278 prog_data
->base
.urb_read_length
= 0;
1280 STATIC_ASSERT(BRW_TESS_PARTITIONING_INTEGER
== TESS_SPACING_EQUAL
- 1);
1281 STATIC_ASSERT(BRW_TESS_PARTITIONING_ODD_FRACTIONAL
==
1282 TESS_SPACING_FRACTIONAL_ODD
- 1);
1283 STATIC_ASSERT(BRW_TESS_PARTITIONING_EVEN_FRACTIONAL
==
1284 TESS_SPACING_FRACTIONAL_EVEN
- 1);
1286 prog_data
->partitioning
=
1287 (enum brw_tess_partitioning
) (nir
->info
.tess
.spacing
- 1);
1289 switch (nir
->info
.tess
.primitive_mode
) {
1291 prog_data
->domain
= BRW_TESS_DOMAIN_QUAD
;
1294 prog_data
->domain
= BRW_TESS_DOMAIN_TRI
;
1297 prog_data
->domain
= BRW_TESS_DOMAIN_ISOLINE
;
1300 unreachable("invalid domain shader primitive mode");
1303 if (nir
->info
.tess
.point_mode
) {
1304 prog_data
->output_topology
= BRW_TESS_OUTPUT_TOPOLOGY_POINT
;
1305 } else if (nir
->info
.tess
.primitive_mode
== GL_ISOLINES
) {
1306 prog_data
->output_topology
= BRW_TESS_OUTPUT_TOPOLOGY_LINE
;
1308 /* Hardware winding order is backwards from OpenGL */
1309 prog_data
->output_topology
=
1310 nir
->info
.tess
.ccw
? BRW_TESS_OUTPUT_TOPOLOGY_TRI_CW
1311 : BRW_TESS_OUTPUT_TOPOLOGY_TRI_CCW
;
1314 if (unlikely(INTEL_DEBUG
& DEBUG_TES
)) {
1315 fprintf(stderr
, "TES Input ");
1316 brw_print_vue_map(stderr
, input_vue_map
);
1317 fprintf(stderr
, "TES Output ");
1318 brw_print_vue_map(stderr
, &prog_data
->base
.vue_map
);
1322 fs_visitor
v(compiler
, log_data
, mem_ctx
, (void *) key
,
1323 &prog_data
->base
.base
, NULL
, nir
, 8,
1324 shader_time_index
, input_vue_map
);
1327 *error_str
= ralloc_strdup(mem_ctx
, v
.fail_msg
);
1331 prog_data
->base
.base
.dispatch_grf_start_reg
= v
.payload
.num_regs
;
1332 prog_data
->base
.dispatch_mode
= DISPATCH_MODE_SIMD8
;
1334 fs_generator
g(compiler
, log_data
, mem_ctx
,
1335 &prog_data
->base
.base
, v
.promoted_constants
, false,
1336 MESA_SHADER_TESS_EVAL
);
1337 if (unlikely(INTEL_DEBUG
& DEBUG_TES
)) {
1338 g
.enable_debug(ralloc_asprintf(mem_ctx
,
1339 "%s tessellation evaluation shader %s",
1340 nir
->info
.label
? nir
->info
.label
1345 g
.generate_code(v
.cfg
, 8);
1347 assembly
= g
.get_assembly();
1349 brw::vec4_tes_visitor
v(compiler
, log_data
, key
, prog_data
,
1350 nir
, mem_ctx
, shader_time_index
);
1353 *error_str
= ralloc_strdup(mem_ctx
, v
.fail_msg
);
1357 if (unlikely(INTEL_DEBUG
& DEBUG_TES
))
1358 v
.dump_instructions();
1360 assembly
= brw_vec4_generate_assembly(compiler
, log_data
, mem_ctx
, nir
,
1361 &prog_data
->base
, v
.cfg
);