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_UD;
49 case GLSL_TYPE_UINT16:
50 return BRW_REGISTER_TYPE_UW;
52 return brw_type_for_base_type(type->fields.array);
53 case GLSL_TYPE_STRUCT:
54 case GLSL_TYPE_SAMPLER:
55 case GLSL_TYPE_ATOMIC_UINT:
56 /* These should be overridden with the type of the member when
57 * dereferenced into. BRW_REGISTER_TYPE_UD seems like a likely
58 * way to trip up if we don't.
60 return BRW_REGISTER_TYPE_UD;
62 return BRW_REGISTER_TYPE_UD;
63 case GLSL_TYPE_DOUBLE:
64 return BRW_REGISTER_TYPE_DF;
65 case GLSL_TYPE_UINT64:
66 return BRW_REGISTER_TYPE_UQ;
68 return BRW_REGISTER_TYPE_Q;
71 case GLSL_TYPE_INTERFACE:
72 case GLSL_TYPE_FUNCTION:
73 unreachable("not reached");
76 return BRW_REGISTER_TYPE_F;
79 enum brw_conditional_mod
80 brw_conditional_for_comparison(unsigned int op)
84 return BRW_CONDITIONAL_L;
86 return BRW_CONDITIONAL_GE;
88 case ir_binop_all_equal: /* same as equal for scalars */
89 return BRW_CONDITIONAL_Z;
91 case ir_binop_any_nequal: /* same as nequal for scalars */
92 return BRW_CONDITIONAL_NZ;
94 unreachable("not reached: bad operation for comparison");
99 brw_math_function(enum opcode op)
102 case SHADER_OPCODE_RCP:
103 return BRW_MATH_FUNCTION_INV;
104 case SHADER_OPCODE_RSQ:
105 return BRW_MATH_FUNCTION_RSQ;
106 case SHADER_OPCODE_SQRT:
107 return BRW_MATH_FUNCTION_SQRT;
108 case SHADER_OPCODE_EXP2:
109 return BRW_MATH_FUNCTION_EXP;
110 case SHADER_OPCODE_LOG2:
111 return BRW_MATH_FUNCTION_LOG;
112 case SHADER_OPCODE_POW:
113 return BRW_MATH_FUNCTION_POW;
114 case SHADER_OPCODE_SIN:
115 return BRW_MATH_FUNCTION_SIN;
116 case SHADER_OPCODE_COS:
117 return BRW_MATH_FUNCTION_COS;
118 case SHADER_OPCODE_INT_QUOTIENT:
119 return BRW_MATH_FUNCTION_INT_DIV_QUOTIENT;
120 case SHADER_OPCODE_INT_REMAINDER:
121 return BRW_MATH_FUNCTION_INT_DIV_REMAINDER;
123 unreachable("not reached: unknown math function");
128 brw_texture_offset(int *offsets, unsigned num_components, uint32_t *offset_bits)
130 if (!offsets) return false; /* nonconstant offset; caller will handle it. */
132 /* offset out of bounds; caller will handle it. */
133 for (unsigned i = 0; i < num_components; i++)
134 if (offsets[i] > 7 || offsets[i] < -8)
137 /* Combine all three offsets into a single unsigned dword:
139 * bits 11:8 - U Offset (X component)
140 * bits 7:4 - V Offset (Y component)
141 * bits 3:0 - R Offset (Z component)
144 for (unsigned i = 0; i < num_components; i++) {
145 const unsigned shift = 4 * (2 - i);
146 *offset_bits |= (offsets[i] << shift) & (0xF << shift);
152 brw_instruction_name(const struct gen_device_info *devinfo, enum opcode op)
155 case BRW_OPCODE_ILLEGAL ... BRW_OPCODE_NOP:
156 /* The DO instruction doesn't exist on Gen6+, but we use it to mark the
157 * start of a loop in the IR.
159 if (devinfo->gen >= 6 && op == BRW_OPCODE_DO)
162 /* The following conversion opcodes doesn't exist on Gen8+, but we use
163 * then to mark that we want to do the conversion.
165 if (devinfo->gen > 7 && op == BRW_OPCODE_F32TO16)
168 if (devinfo->gen > 7 && op == BRW_OPCODE_F16TO32)
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_REP_FB_WRITE:
178 return "rep_fb_write";
179 case FS_OPCODE_FB_READ:
181 case FS_OPCODE_FB_READ_LOGICAL:
182 return "fb_read_logical";
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_TXF_LZ:
219 case SHADER_OPCODE_TXL:
221 case SHADER_OPCODE_TXL_LOGICAL:
222 return "txl_logical";
223 case SHADER_OPCODE_TXL_LZ:
225 case SHADER_OPCODE_TXS:
227 case SHADER_OPCODE_TXS_LOGICAL:
228 return "txs_logical";
231 case FS_OPCODE_TXB_LOGICAL:
232 return "txb_logical";
233 case SHADER_OPCODE_TXF_CMS:
235 case SHADER_OPCODE_TXF_CMS_LOGICAL:
236 return "txf_cms_logical";
237 case SHADER_OPCODE_TXF_CMS_W:
239 case SHADER_OPCODE_TXF_CMS_W_LOGICAL:
240 return "txf_cms_w_logical";
241 case SHADER_OPCODE_TXF_UMS:
243 case SHADER_OPCODE_TXF_UMS_LOGICAL:
244 return "txf_ums_logical";
245 case SHADER_OPCODE_TXF_MCS:
247 case SHADER_OPCODE_TXF_MCS_LOGICAL:
248 return "txf_mcs_logical";
249 case SHADER_OPCODE_LOD:
251 case SHADER_OPCODE_LOD_LOGICAL:
252 return "lod_logical";
253 case SHADER_OPCODE_TG4:
255 case SHADER_OPCODE_TG4_LOGICAL:
256 return "tg4_logical";
257 case SHADER_OPCODE_TG4_OFFSET:
259 case SHADER_OPCODE_TG4_OFFSET_LOGICAL:
260 return "tg4_offset_logical";
261 case SHADER_OPCODE_SAMPLEINFO:
263 case SHADER_OPCODE_SAMPLEINFO_LOGICAL:
264 return "sampleinfo_logical";
266 case SHADER_OPCODE_SHADER_TIME_ADD:
267 return "shader_time_add";
269 case SHADER_OPCODE_UNTYPED_ATOMIC:
270 return "untyped_atomic";
271 case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL:
272 return "untyped_atomic_logical";
273 case SHADER_OPCODE_UNTYPED_SURFACE_READ:
274 return "untyped_surface_read";
275 case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL:
276 return "untyped_surface_read_logical";
277 case SHADER_OPCODE_UNTYPED_SURFACE_WRITE:
278 return "untyped_surface_write";
279 case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL:
280 return "untyped_surface_write_logical";
281 case SHADER_OPCODE_TYPED_ATOMIC:
282 return "typed_atomic";
283 case SHADER_OPCODE_TYPED_ATOMIC_LOGICAL:
284 return "typed_atomic_logical";
285 case SHADER_OPCODE_TYPED_SURFACE_READ:
286 return "typed_surface_read";
287 case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL:
288 return "typed_surface_read_logical";
289 case SHADER_OPCODE_TYPED_SURFACE_WRITE:
290 return "typed_surface_write";
291 case SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL:
292 return "typed_surface_write_logical";
293 case SHADER_OPCODE_MEMORY_FENCE:
294 return "memory_fence";
296 case SHADER_OPCODE_BYTE_SCATTERED_READ:
297 return "byte_scattered_read";
298 case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL:
299 return "byte_scattered_read_logical";
300 case SHADER_OPCODE_BYTE_SCATTERED_WRITE:
301 return "byte_scattered_write";
302 case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL:
303 return "byte_scattered_write_logical";
305 case SHADER_OPCODE_LOAD_PAYLOAD:
306 return "load_payload";
310 case SHADER_OPCODE_GEN4_SCRATCH_READ:
311 return "gen4_scratch_read";
312 case SHADER_OPCODE_GEN4_SCRATCH_WRITE:
313 return "gen4_scratch_write";
314 case SHADER_OPCODE_GEN7_SCRATCH_READ:
315 return "gen7_scratch_read";
316 case SHADER_OPCODE_URB_WRITE_SIMD8:
317 return "gen8_urb_write_simd8";
318 case SHADER_OPCODE_URB_WRITE_SIMD8_PER_SLOT:
319 return "gen8_urb_write_simd8_per_slot";
320 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED:
321 return "gen8_urb_write_simd8_masked";
322 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT:
323 return "gen8_urb_write_simd8_masked_per_slot";
324 case SHADER_OPCODE_URB_READ_SIMD8:
325 return "urb_read_simd8";
326 case SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT:
327 return "urb_read_simd8_per_slot";
329 case SHADER_OPCODE_FIND_LIVE_CHANNEL:
330 return "find_live_channel";
331 case SHADER_OPCODE_BROADCAST:
334 case SHADER_OPCODE_GET_BUFFER_SIZE:
335 return "get_buffer_size";
337 case VEC4_OPCODE_MOV_BYTES:
339 case VEC4_OPCODE_PACK_BYTES:
341 case VEC4_OPCODE_UNPACK_UNIFORM:
342 return "unpack_uniform";
343 case VEC4_OPCODE_DOUBLE_TO_F32:
344 return "double_to_f32";
345 case VEC4_OPCODE_DOUBLE_TO_D32:
346 return "double_to_d32";
347 case VEC4_OPCODE_DOUBLE_TO_U32:
348 return "double_to_u32";
349 case VEC4_OPCODE_TO_DOUBLE:
350 return "single_to_double";
351 case VEC4_OPCODE_PICK_LOW_32BIT:
352 return "pick_low_32bit";
353 case VEC4_OPCODE_PICK_HIGH_32BIT:
354 return "pick_high_32bit";
355 case VEC4_OPCODE_SET_LOW_32BIT:
356 return "set_low_32bit";
357 case VEC4_OPCODE_SET_HIGH_32BIT:
358 return "set_high_32bit";
360 case FS_OPCODE_DDX_COARSE:
362 case FS_OPCODE_DDX_FINE:
364 case FS_OPCODE_DDY_COARSE:
366 case FS_OPCODE_DDY_FINE:
369 case FS_OPCODE_CINTERP:
371 case FS_OPCODE_LINTERP:
374 case FS_OPCODE_PIXEL_X:
376 case FS_OPCODE_PIXEL_Y:
379 case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD:
380 return "uniform_pull_const";
381 case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD_GEN7:
382 return "uniform_pull_const_gen7";
383 case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN4:
384 return "varying_pull_const_gen4";
385 case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7:
386 return "varying_pull_const_gen7";
387 case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_LOGICAL:
388 return "varying_pull_const_logical";
390 case FS_OPCODE_MOV_DISPATCH_TO_FLAGS:
391 return "mov_dispatch_to_flags";
392 case FS_OPCODE_DISCARD_JUMP:
393 return "discard_jump";
395 case FS_OPCODE_SET_SAMPLE_ID:
396 return "set_sample_id";
398 case FS_OPCODE_PACK_HALF_2x16_SPLIT:
399 return "pack_half_2x16_split";
400 case FS_OPCODE_UNPACK_HALF_2x16_SPLIT_X:
401 return "unpack_half_2x16_split_x";
402 case FS_OPCODE_UNPACK_HALF_2x16_SPLIT_Y:
403 return "unpack_half_2x16_split_y";
405 case FS_OPCODE_PLACEHOLDER_HALT:
406 return "placeholder_halt";
408 case FS_OPCODE_INTERPOLATE_AT_SAMPLE:
409 return "interp_sample";
410 case FS_OPCODE_INTERPOLATE_AT_SHARED_OFFSET:
411 return "interp_shared_offset";
412 case FS_OPCODE_INTERPOLATE_AT_PER_SLOT_OFFSET:
413 return "interp_per_slot_offset";
415 case VS_OPCODE_URB_WRITE:
416 return "vs_urb_write";
417 case VS_OPCODE_PULL_CONSTANT_LOAD:
418 return "pull_constant_load";
419 case VS_OPCODE_PULL_CONSTANT_LOAD_GEN7:
420 return "pull_constant_load_gen7";
422 case VS_OPCODE_SET_SIMD4X2_HEADER_GEN9:
423 return "set_simd4x2_header_gen9";
425 case VS_OPCODE_UNPACK_FLAGS_SIMD4X2:
426 return "unpack_flags_simd4x2";
428 case GS_OPCODE_URB_WRITE:
429 return "gs_urb_write";
430 case GS_OPCODE_URB_WRITE_ALLOCATE:
431 return "gs_urb_write_allocate";
432 case GS_OPCODE_THREAD_END:
433 return "gs_thread_end";
434 case GS_OPCODE_SET_WRITE_OFFSET:
435 return "set_write_offset";
436 case GS_OPCODE_SET_VERTEX_COUNT:
437 return "set_vertex_count";
438 case GS_OPCODE_SET_DWORD_2:
439 return "set_dword_2";
440 case GS_OPCODE_PREPARE_CHANNEL_MASKS:
441 return "prepare_channel_masks";
442 case GS_OPCODE_SET_CHANNEL_MASKS:
443 return "set_channel_masks";
444 case GS_OPCODE_GET_INSTANCE_ID:
445 return "get_instance_id";
446 case GS_OPCODE_FF_SYNC:
448 case GS_OPCODE_SET_PRIMITIVE_ID:
449 return "set_primitive_id";
450 case GS_OPCODE_SVB_WRITE:
451 return "gs_svb_write";
452 case GS_OPCODE_SVB_SET_DST_INDEX:
453 return "gs_svb_set_dst_index";
454 case GS_OPCODE_FF_SYNC_SET_PRIMITIVES:
455 return "gs_ff_sync_set_primitives";
456 case CS_OPCODE_CS_TERMINATE:
457 return "cs_terminate";
458 case SHADER_OPCODE_BARRIER:
460 case SHADER_OPCODE_MULH:
462 case SHADER_OPCODE_MOV_INDIRECT:
463 return "mov_indirect";
465 case VEC4_OPCODE_URB_READ:
467 case TCS_OPCODE_GET_INSTANCE_ID:
468 return "tcs_get_instance_id";
469 case TCS_OPCODE_URB_WRITE:
470 return "tcs_urb_write";
471 case TCS_OPCODE_SET_INPUT_URB_OFFSETS:
472 return "tcs_set_input_urb_offsets";
473 case TCS_OPCODE_SET_OUTPUT_URB_OFFSETS:
474 return "tcs_set_output_urb_offsets";
475 case TCS_OPCODE_GET_PRIMITIVE_ID:
476 return "tcs_get_primitive_id";
477 case TCS_OPCODE_CREATE_BARRIER_HEADER:
478 return "tcs_create_barrier_header";
479 case TCS_OPCODE_SRC0_010_IS_ZERO:
480 return "tcs_src0<0,1,0>_is_zero";
481 case TCS_OPCODE_RELEASE_INPUT:
482 return "tcs_release_input";
483 case TCS_OPCODE_THREAD_END:
484 return "tcs_thread_end";
485 case TES_OPCODE_CREATE_INPUT_READ_HEADER:
486 return "tes_create_input_read_header";
487 case TES_OPCODE_ADD_INDIRECT_URB_OFFSET:
488 return "tes_add_indirect_urb_offset";
489 case TES_OPCODE_GET_PRIMITIVE_ID:
490 return "tes_get_primitive_id";
492 case SHADER_OPCODE_RND_MODE:
496 unreachable("not reached");
500 brw_saturate_immediate(enum brw_reg_type type, struct brw_reg *reg)
507 } imm, sat_imm = { 0 };
509 const unsigned size = type_sz(type);
511 /* We want to either do a 32-bit or 64-bit data copy, the type is otherwise
512 * irrelevant, so just check the size of the type and copy from/to an
513 * appropriately sized field.
521 case BRW_REGISTER_TYPE_UD:
522 case BRW_REGISTER_TYPE_D:
523 case BRW_REGISTER_TYPE_UW:
524 case BRW_REGISTER_TYPE_W:
525 case BRW_REGISTER_TYPE_UQ:
526 case BRW_REGISTER_TYPE_Q:
529 case BRW_REGISTER_TYPE_F:
530 sat_imm.f = CLAMP(imm.f, 0.0f, 1.0f);
532 case BRW_REGISTER_TYPE_DF:
533 sat_imm.df = CLAMP(imm.df, 0.0, 1.0);
535 case BRW_REGISTER_TYPE_UB:
536 case BRW_REGISTER_TYPE_B:
537 unreachable("no UB/B immediates");
538 case BRW_REGISTER_TYPE_V:
539 case BRW_REGISTER_TYPE_UV:
540 case BRW_REGISTER_TYPE_VF:
541 unreachable("unimplemented: saturate vector immediate");
542 case BRW_REGISTER_TYPE_HF:
543 unreachable("unimplemented: saturate HF immediate");
544 case BRW_REGISTER_TYPE_NF:
545 unreachable("no NF immediates");
549 if (imm.ud != sat_imm.ud) {
550 reg->ud = sat_imm.ud;
554 if (imm.df != sat_imm.df) {
555 reg->df = sat_imm.df;
563 brw_negate_immediate(enum brw_reg_type type, struct brw_reg *reg)
566 case BRW_REGISTER_TYPE_D:
567 case BRW_REGISTER_TYPE_UD:
570 case BRW_REGISTER_TYPE_W:
571 case BRW_REGISTER_TYPE_UW:
572 reg->d = -(int16_t)reg->ud;
574 case BRW_REGISTER_TYPE_F:
577 case BRW_REGISTER_TYPE_VF:
578 reg->ud ^= 0x80808080;
580 case BRW_REGISTER_TYPE_DF:
583 case BRW_REGISTER_TYPE_UQ:
584 case BRW_REGISTER_TYPE_Q:
585 reg->d64 = -reg->d64;
587 case BRW_REGISTER_TYPE_UB:
588 case BRW_REGISTER_TYPE_B:
589 unreachable("no UB/B immediates");
590 case BRW_REGISTER_TYPE_UV:
591 case BRW_REGISTER_TYPE_V:
592 assert(!"unimplemented: negate UV/V immediate");
593 case BRW_REGISTER_TYPE_HF:
594 assert(!"unimplemented: negate HF immediate");
595 case BRW_REGISTER_TYPE_NF:
596 unreachable("no NF immediates");
603 brw_abs_immediate(enum brw_reg_type type, struct brw_reg *reg)
606 case BRW_REGISTER_TYPE_D:
607 reg->d = abs(reg->d);
609 case BRW_REGISTER_TYPE_W:
610 reg->d = abs((int16_t)reg->ud);
612 case BRW_REGISTER_TYPE_F:
613 reg->f = fabsf(reg->f);
615 case BRW_REGISTER_TYPE_DF:
616 reg->df = fabs(reg->df);
618 case BRW_REGISTER_TYPE_VF:
619 reg->ud &= ~0x80808080;
621 case BRW_REGISTER_TYPE_Q:
622 reg->d64 = imaxabs(reg->d64);
624 case BRW_REGISTER_TYPE_UB:
625 case BRW_REGISTER_TYPE_B:
626 unreachable("no UB/B immediates");
627 case BRW_REGISTER_TYPE_UQ:
628 case BRW_REGISTER_TYPE_UD:
629 case BRW_REGISTER_TYPE_UW:
630 case BRW_REGISTER_TYPE_UV:
631 /* Presumably the absolute value modifier on an unsigned source is a
632 * nop, but it would be nice to confirm.
634 assert(!"unimplemented: abs unsigned immediate");
635 case BRW_REGISTER_TYPE_V:
636 assert(!"unimplemented: abs V immediate");
637 case BRW_REGISTER_TYPE_HF:
638 assert(!"unimplemented: abs HF immediate");
639 case BRW_REGISTER_TYPE_NF:
640 unreachable("no NF immediates");
646 backend_shader::backend_shader(const struct brw_compiler *compiler,
649 const nir_shader *shader,
650 struct brw_stage_prog_data *stage_prog_data)
651 : compiler(compiler),
653 devinfo(compiler->devinfo),
655 stage_prog_data(stage_prog_data),
658 stage(shader->info.stage)
660 debug_enabled = INTEL_DEBUG & intel_debug_flag_for_shader_stage(stage);
661 stage_name = _mesa_shader_stage_to_string(stage);
662 stage_abbrev = _mesa_shader_stage_to_abbrev(stage);
665 backend_shader::~backend_shader()
670 backend_reg::equals(const backend_reg &r) const
672 return brw_regs_equal(this, &r) && offset == r.offset;
676 backend_reg::is_zero() const
682 case BRW_REGISTER_TYPE_F:
684 case BRW_REGISTER_TYPE_DF:
686 case BRW_REGISTER_TYPE_D:
687 case BRW_REGISTER_TYPE_UD:
689 case BRW_REGISTER_TYPE_UQ:
690 case BRW_REGISTER_TYPE_Q:
698 backend_reg::is_one() const
704 case BRW_REGISTER_TYPE_F:
706 case BRW_REGISTER_TYPE_DF:
708 case BRW_REGISTER_TYPE_D:
709 case BRW_REGISTER_TYPE_UD:
711 case BRW_REGISTER_TYPE_UQ:
712 case BRW_REGISTER_TYPE_Q:
720 backend_reg::is_negative_one() const
726 case BRW_REGISTER_TYPE_F:
728 case BRW_REGISTER_TYPE_DF:
730 case BRW_REGISTER_TYPE_D:
732 case BRW_REGISTER_TYPE_Q:
740 backend_reg::is_null() const
742 return file == ARF && nr == BRW_ARF_NULL;
747 backend_reg::is_accumulator() const
749 return file == ARF && nr == BRW_ARF_ACCUMULATOR;
753 backend_instruction::is_commutative() const
761 case SHADER_OPCODE_MULH:
764 /* MIN and MAX are commutative. */
765 if (conditional_mod == BRW_CONDITIONAL_GE ||
766 conditional_mod == BRW_CONDITIONAL_L) {
776 backend_instruction::is_3src(const struct gen_device_info *devinfo) const
778 return ::is_3src(devinfo, opcode);
782 backend_instruction::is_tex() const
784 return (opcode == SHADER_OPCODE_TEX ||
785 opcode == FS_OPCODE_TXB ||
786 opcode == SHADER_OPCODE_TXD ||
787 opcode == SHADER_OPCODE_TXF ||
788 opcode == SHADER_OPCODE_TXF_LZ ||
789 opcode == SHADER_OPCODE_TXF_CMS ||
790 opcode == SHADER_OPCODE_TXF_CMS_W ||
791 opcode == SHADER_OPCODE_TXF_UMS ||
792 opcode == SHADER_OPCODE_TXF_MCS ||
793 opcode == SHADER_OPCODE_TXL ||
794 opcode == SHADER_OPCODE_TXL_LZ ||
795 opcode == SHADER_OPCODE_TXS ||
796 opcode == SHADER_OPCODE_LOD ||
797 opcode == SHADER_OPCODE_TG4 ||
798 opcode == SHADER_OPCODE_TG4_OFFSET ||
799 opcode == SHADER_OPCODE_SAMPLEINFO);
803 backend_instruction::is_math() const
805 return (opcode == SHADER_OPCODE_RCP ||
806 opcode == SHADER_OPCODE_RSQ ||
807 opcode == SHADER_OPCODE_SQRT ||
808 opcode == SHADER_OPCODE_EXP2 ||
809 opcode == SHADER_OPCODE_LOG2 ||
810 opcode == SHADER_OPCODE_SIN ||
811 opcode == SHADER_OPCODE_COS ||
812 opcode == SHADER_OPCODE_INT_QUOTIENT ||
813 opcode == SHADER_OPCODE_INT_REMAINDER ||
814 opcode == SHADER_OPCODE_POW);
818 backend_instruction::is_control_flow() const
822 case BRW_OPCODE_WHILE:
824 case BRW_OPCODE_ELSE:
825 case BRW_OPCODE_ENDIF:
826 case BRW_OPCODE_BREAK:
827 case BRW_OPCODE_CONTINUE:
835 backend_instruction::can_do_source_mods() const
838 case BRW_OPCODE_ADDC:
840 case BRW_OPCODE_BFI1:
841 case BRW_OPCODE_BFI2:
842 case BRW_OPCODE_BFREV:
843 case BRW_OPCODE_CBIT:
846 case BRW_OPCODE_SUBB:
847 case SHADER_OPCODE_BROADCAST:
848 case SHADER_OPCODE_MOV_INDIRECT:
856 backend_instruction::can_do_saturate() const
866 case BRW_OPCODE_F16TO32:
867 case BRW_OPCODE_F32TO16:
868 case BRW_OPCODE_LINE:
872 case BRW_OPCODE_MATH:
875 case SHADER_OPCODE_MULH:
877 case BRW_OPCODE_RNDD:
878 case BRW_OPCODE_RNDE:
879 case BRW_OPCODE_RNDU:
880 case BRW_OPCODE_RNDZ:
884 case FS_OPCODE_LINTERP:
885 case SHADER_OPCODE_COS:
886 case SHADER_OPCODE_EXP2:
887 case SHADER_OPCODE_LOG2:
888 case SHADER_OPCODE_POW:
889 case SHADER_OPCODE_RCP:
890 case SHADER_OPCODE_RSQ:
891 case SHADER_OPCODE_SIN:
892 case SHADER_OPCODE_SQRT:
900 backend_instruction::can_do_cmod() const
904 case BRW_OPCODE_ADDC:
909 case BRW_OPCODE_CMPN:
914 case BRW_OPCODE_F16TO32:
915 case BRW_OPCODE_F32TO16:
917 case BRW_OPCODE_LINE:
921 case BRW_OPCODE_MACH:
928 case BRW_OPCODE_RNDD:
929 case BRW_OPCODE_RNDE:
930 case BRW_OPCODE_RNDU:
931 case BRW_OPCODE_RNDZ:
932 case BRW_OPCODE_SAD2:
933 case BRW_OPCODE_SADA2:
936 case BRW_OPCODE_SUBB:
938 case FS_OPCODE_CINTERP:
939 case FS_OPCODE_LINTERP:
947 backend_instruction::reads_accumulator_implicitly() const
951 case BRW_OPCODE_MACH:
952 case BRW_OPCODE_SADA2:
960 backend_instruction::writes_accumulator_implicitly(const struct gen_device_info *devinfo) const
962 return writes_accumulator ||
964 ((opcode >= BRW_OPCODE_ADD && opcode < BRW_OPCODE_NOP) ||
965 (opcode >= FS_OPCODE_DDX_COARSE && opcode <= FS_OPCODE_LINTERP &&
966 opcode != FS_OPCODE_CINTERP)));
970 backend_instruction::has_side_effects() const
973 case SHADER_OPCODE_UNTYPED_ATOMIC:
974 case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL:
975 case SHADER_OPCODE_GEN4_SCRATCH_WRITE:
976 case SHADER_OPCODE_UNTYPED_SURFACE_WRITE:
977 case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL:
978 case SHADER_OPCODE_BYTE_SCATTERED_WRITE:
979 case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL:
980 case SHADER_OPCODE_TYPED_ATOMIC:
981 case SHADER_OPCODE_TYPED_ATOMIC_LOGICAL:
982 case SHADER_OPCODE_TYPED_SURFACE_WRITE:
983 case SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL:
984 case SHADER_OPCODE_MEMORY_FENCE:
985 case SHADER_OPCODE_URB_WRITE_SIMD8:
986 case SHADER_OPCODE_URB_WRITE_SIMD8_PER_SLOT:
987 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED:
988 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT:
989 case FS_OPCODE_FB_WRITE:
990 case FS_OPCODE_FB_WRITE_LOGICAL:
991 case SHADER_OPCODE_BARRIER:
992 case TCS_OPCODE_URB_WRITE:
993 case TCS_OPCODE_RELEASE_INPUT:
994 case SHADER_OPCODE_RND_MODE:
1002 backend_instruction::is_volatile() const
1005 case SHADER_OPCODE_UNTYPED_SURFACE_READ:
1006 case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL:
1007 case SHADER_OPCODE_TYPED_SURFACE_READ:
1008 case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL:
1009 case SHADER_OPCODE_BYTE_SCATTERED_READ:
1010 case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL:
1011 case SHADER_OPCODE_URB_READ_SIMD8:
1012 case SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT:
1013 case VEC4_OPCODE_URB_READ:
1022 inst_is_in_block(const bblock_t *block, const backend_instruction *inst)
1025 foreach_inst_in_block (backend_instruction, i, block) {
1035 adjust_later_block_ips(bblock_t *start_block, int ip_adjustment)
1037 for (bblock_t *block_iter = start_block->next();
1039 block_iter = block_iter->next()) {
1040 block_iter->start_ip += ip_adjustment;
1041 block_iter->end_ip += ip_adjustment;
1046 backend_instruction::insert_after(bblock_t *block, backend_instruction *inst)
1048 assert(this != inst);
1050 if (!this->is_head_sentinel())
1051 assert(inst_is_in_block(block, this) || !"Instruction not in block");
1055 adjust_later_block_ips(block, 1);
1057 exec_node::insert_after(inst);
1061 backend_instruction::insert_before(bblock_t *block, backend_instruction *inst)
1063 assert(this != inst);
1065 if (!this->is_tail_sentinel())
1066 assert(inst_is_in_block(block, this) || !"Instruction not in block");
1070 adjust_later_block_ips(block, 1);
1072 exec_node::insert_before(inst);
1076 backend_instruction::insert_before(bblock_t *block, exec_list *list)
1078 assert(inst_is_in_block(block, this) || !"Instruction not in block");
1080 unsigned num_inst = list->length();
1082 block->end_ip += num_inst;
1084 adjust_later_block_ips(block, num_inst);
1086 exec_node::insert_before(list);
1090 backend_instruction::remove(bblock_t *block)
1092 assert(inst_is_in_block(block, this) || !"Instruction not in block");
1094 adjust_later_block_ips(block, -1);
1096 if (block->start_ip == block->end_ip) {
1097 block->cfg->remove_block(block);
1102 exec_node::remove();
1106 backend_shader::dump_instructions()
1108 dump_instructions(NULL);
1112 backend_shader::dump_instructions(const char *name)
1114 FILE *file = stderr;
1115 if (name && geteuid() != 0) {
1116 file = fopen(name, "w");
1123 foreach_block_and_inst(block, backend_instruction, inst, cfg) {
1124 if (!unlikely(INTEL_DEBUG & DEBUG_OPTIMIZER))
1125 fprintf(file, "%4d: ", ip++);
1126 dump_instruction(inst, file);
1130 foreach_in_list(backend_instruction, inst, &instructions) {
1131 if (!unlikely(INTEL_DEBUG & DEBUG_OPTIMIZER))
1132 fprintf(file, "%4d: ", ip++);
1133 dump_instruction(inst, file);
1137 if (file != stderr) {
1143 backend_shader::calculate_cfg()
1147 cfg = new(mem_ctx) cfg_t(&this->instructions);
1150 extern "C" const unsigned *
1151 brw_compile_tes(const struct brw_compiler *compiler,
1154 const struct brw_tes_prog_key *key,
1155 const struct brw_vue_map *input_vue_map,
1156 struct brw_tes_prog_data *prog_data,
1157 const nir_shader *src_shader,
1158 struct gl_program *prog,
1159 int shader_time_index,
1162 const struct gen_device_info *devinfo = compiler->devinfo;
1163 const bool is_scalar = compiler->scalar_stage[MESA_SHADER_TESS_EVAL];
1164 const unsigned *assembly;
1166 nir_shader *nir = nir_shader_clone(mem_ctx, src_shader);
1167 nir->info.inputs_read = key->inputs_read;
1168 nir->info.patch_inputs_read = key->patch_inputs_read;
1170 nir = brw_nir_apply_sampler_key(nir, compiler, &key->tex, is_scalar);
1171 brw_nir_lower_tes_inputs(nir, input_vue_map);
1172 brw_nir_lower_vue_outputs(nir, is_scalar);
1173 nir = brw_postprocess_nir(nir, compiler, is_scalar);
1175 brw_compute_vue_map(devinfo, &prog_data->base.vue_map,
1176 nir->info.outputs_written,
1177 nir->info.separate_shader);
1179 unsigned output_size_bytes = prog_data->base.vue_map.num_slots * 4 * 4;
1181 assert(output_size_bytes >= 1);
1182 if (output_size_bytes > GEN7_MAX_DS_URB_ENTRY_SIZE_BYTES) {
1184 *error_str = ralloc_strdup(mem_ctx, "DS outputs exceed maximum size");
1188 prog_data->base.clip_distance_mask =
1189 ((1 << nir->info.clip_distance_array_size) - 1);
1190 prog_data->base.cull_distance_mask =
1191 ((1 << nir->info.cull_distance_array_size) - 1) <<
1192 nir->info.clip_distance_array_size;
1194 /* URB entry sizes are stored as a multiple of 64 bytes. */
1195 prog_data->base.urb_entry_size = ALIGN(output_size_bytes, 64) / 64;
1197 /* On Cannonlake software shall not program an allocation size that
1198 * specifies a size that is a multiple of 3 64B (512-bit) cachelines.
1200 if (devinfo->gen == 10 &&
1201 prog_data->base.urb_entry_size % 3 == 0)
1202 prog_data->base.urb_entry_size++;
1204 prog_data->base.urb_read_length = 0;
1206 STATIC_ASSERT(BRW_TESS_PARTITIONING_INTEGER == TESS_SPACING_EQUAL - 1);
1207 STATIC_ASSERT(BRW_TESS_PARTITIONING_ODD_FRACTIONAL ==
1208 TESS_SPACING_FRACTIONAL_ODD - 1);
1209 STATIC_ASSERT(BRW_TESS_PARTITIONING_EVEN_FRACTIONAL ==
1210 TESS_SPACING_FRACTIONAL_EVEN - 1);
1212 prog_data->partitioning =
1213 (enum brw_tess_partitioning) (nir->info.tess.spacing - 1);
1215 switch (nir->info.tess.primitive_mode) {
1217 prog_data->domain = BRW_TESS_DOMAIN_QUAD;
1220 prog_data->domain = BRW_TESS_DOMAIN_TRI;
1223 prog_data->domain = BRW_TESS_DOMAIN_ISOLINE;
1226 unreachable("invalid domain shader primitive mode");
1229 if (nir->info.tess.point_mode) {
1230 prog_data->output_topology = BRW_TESS_OUTPUT_TOPOLOGY_POINT;
1231 } else if (nir->info.tess.primitive_mode == GL_ISOLINES) {
1232 prog_data->output_topology = BRW_TESS_OUTPUT_TOPOLOGY_LINE;
1234 /* Hardware winding order is backwards from OpenGL */
1235 prog_data->output_topology =
1236 nir->info.tess.ccw ? BRW_TESS_OUTPUT_TOPOLOGY_TRI_CW
1237 : BRW_TESS_OUTPUT_TOPOLOGY_TRI_CCW;
1240 if (unlikely(INTEL_DEBUG & DEBUG_TES)) {
1241 fprintf(stderr, "TES Input ");
1242 brw_print_vue_map(stderr, input_vue_map);
1243 fprintf(stderr, "TES Output ");
1244 brw_print_vue_map(stderr, &prog_data->base.vue_map);
1248 fs_visitor v(compiler, log_data, mem_ctx, (void *) key,
1249 &prog_data->base.base, NULL, nir, 8,
1250 shader_time_index, input_vue_map);
1253 *error_str = ralloc_strdup(mem_ctx, v.fail_msg);
1257 prog_data->base.base.dispatch_grf_start_reg = v.payload.num_regs;
1258 prog_data->base.dispatch_mode = DISPATCH_MODE_SIMD8;
1260 fs_generator g(compiler, log_data, mem_ctx, (void *) key,
1261 &prog_data->base.base, v.promoted_constants, false,
1262 MESA_SHADER_TESS_EVAL);
1263 if (unlikely(INTEL_DEBUG & DEBUG_TES)) {
1264 g.enable_debug(ralloc_asprintf(mem_ctx,
1265 "%s tessellation evaluation shader %s",
1266 nir->info.label ? nir->info.label
1271 g.generate_code(v.cfg, 8);
1273 assembly = g.get_assembly();
1275 brw::vec4_tes_visitor v(compiler, log_data, key, prog_data,
1276 nir, mem_ctx, shader_time_index);
1279 *error_str = ralloc_strdup(mem_ctx, v.fail_msg);
1283 if (unlikely(INTEL_DEBUG & DEBUG_TES))
1284 v.dump_instructions();
1286 assembly = brw_vec4_generate_assembly(compiler, log_data, mem_ctx, nir,
1287 &prog_data->base, v.cfg);