1 /*
2 * Copyright © 2010 Intel Corporation
3 *
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:
10 *
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
13 * Software.
14 *
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
21 * IN THE SOFTWARE.
22 */
23
24 #include "brw_context.h"
25 #include "brw_cfg.h"
26 #include "brw_eu.h"
27 #include "brw_fs.h"
28 #include "brw_nir.h"
29 #include "brw_vec4_tes.h"
30 #include "main/uniforms.h"
31
32 extern "C" void
brw_mark_surface_used(struct brw_stage_prog_data * prog_data,unsigned surf_index)33 brw_mark_surface_used(struct brw_stage_prog_data *prog_data,
34 unsigned surf_index)
35 {
36 assert(surf_index < BRW_MAX_SURFACES);
37
38 prog_data->binding_table.size_bytes =
39 MAX2(prog_data->binding_table.size_bytes, (surf_index + 1) * 4);
40 }
41
42 enum brw_reg_type
brw_type_for_base_type(const struct glsl_type * type)43 brw_type_for_base_type(const struct glsl_type *type)
44 {
45 switch (type->base_type) {
46 case GLSL_TYPE_FLOAT:
47 return BRW_REGISTER_TYPE_F;
48 case GLSL_TYPE_INT:
49 case GLSL_TYPE_BOOL:
50 case GLSL_TYPE_SUBROUTINE:
51 return BRW_REGISTER_TYPE_D;
52 case GLSL_TYPE_UINT:
53 return BRW_REGISTER_TYPE_UD;
54 case GLSL_TYPE_ARRAY:
55 return brw_type_for_base_type(type->fields.array);
56 case GLSL_TYPE_STRUCT:
57 case GLSL_TYPE_SAMPLER:
58 case GLSL_TYPE_ATOMIC_UINT:
59 /* These should be overridden with the type of the member when
60 * dereferenced into. BRW_REGISTER_TYPE_UD seems like a likely
61 * way to trip up if we don't.
62 */
63 return BRW_REGISTER_TYPE_UD;
64 case GLSL_TYPE_IMAGE:
65 return BRW_REGISTER_TYPE_UD;
66 case GLSL_TYPE_DOUBLE:
67 return BRW_REGISTER_TYPE_DF;
68 case GLSL_TYPE_VOID:
69 case GLSL_TYPE_ERROR:
70 case GLSL_TYPE_INTERFACE:
71 case GLSL_TYPE_FUNCTION:
72 unreachable("not reached");
73 }
74
75 return BRW_REGISTER_TYPE_F;
76 }
77
78 enum brw_conditional_mod
brw_conditional_for_comparison(unsigned int op)79 brw_conditional_for_comparison(unsigned int op)
80 {
81 switch (op) {
82 case ir_binop_less:
83 return BRW_CONDITIONAL_L;
84 case ir_binop_greater:
85 return BRW_CONDITIONAL_G;
86 case ir_binop_lequal:
87 return BRW_CONDITIONAL_LE;
88 case ir_binop_gequal:
89 return BRW_CONDITIONAL_GE;
90 case ir_binop_equal:
91 case ir_binop_all_equal: /* same as equal for scalars */
92 return BRW_CONDITIONAL_Z;
93 case ir_binop_nequal:
94 case ir_binop_any_nequal: /* same as nequal for scalars */
95 return BRW_CONDITIONAL_NZ;
96 default:
97 unreachable("not reached: bad operation for comparison");
98 }
99 }
100
101 uint32_t
brw_math_function(enum opcode op)102 brw_math_function(enum opcode op)
103 {
104 switch (op) {
105 case SHADER_OPCODE_RCP:
106 return BRW_MATH_FUNCTION_INV;
107 case SHADER_OPCODE_RSQ:
108 return BRW_MATH_FUNCTION_RSQ;
109 case SHADER_OPCODE_SQRT:
110 return BRW_MATH_FUNCTION_SQRT;
111 case SHADER_OPCODE_EXP2:
112 return BRW_MATH_FUNCTION_EXP;
113 case SHADER_OPCODE_LOG2:
114 return BRW_MATH_FUNCTION_LOG;
115 case SHADER_OPCODE_POW:
116 return BRW_MATH_FUNCTION_POW;
117 case SHADER_OPCODE_SIN:
118 return BRW_MATH_FUNCTION_SIN;
119 case SHADER_OPCODE_COS:
120 return BRW_MATH_FUNCTION_COS;
121 case SHADER_OPCODE_INT_QUOTIENT:
122 return BRW_MATH_FUNCTION_INT_DIV_QUOTIENT;
123 case SHADER_OPCODE_INT_REMAINDER:
124 return BRW_MATH_FUNCTION_INT_DIV_REMAINDER;
125 default:
126 unreachable("not reached: unknown math function");
127 }
128 }
129
130 bool
brw_texture_offset(int * offsets,unsigned num_components,uint32_t * offset_bits)131 brw_texture_offset(int *offsets, unsigned num_components, uint32_t *offset_bits)
132 {
133 if (!offsets) return false; /* nonconstant offset; caller will handle it. */
134
135 /* offset out of bounds; caller will handle it. */
136 for (unsigned i = 0; i < num_components; i++)
137 if (offsets[i] > 7 || offsets[i] < -8)
138 return false;
139
140 /* Combine all three offsets into a single unsigned dword:
141 *
142 * bits 11:8 - U Offset (X component)
143 * bits 7:4 - V Offset (Y component)
144 * bits 3:0 - R Offset (Z component)
145 */
146 *offset_bits = 0;
147 for (unsigned i = 0; i < num_components; i++) {
148 const unsigned shift = 4 * (2 - i);
149 *offset_bits |= (offsets[i] << shift) & (0xF << shift);
150 }
151 return true;
152 }
153
154 const char *
brw_instruction_name(const struct gen_device_info * devinfo,enum opcode op)155 brw_instruction_name(const struct gen_device_info *devinfo, enum opcode op)
156 {
157 switch (op) {
158 case BRW_OPCODE_ILLEGAL ... BRW_OPCODE_NOP:
159 /* The DO instruction doesn't exist on Gen6+, but we use it to mark the
160 * start of a loop in the IR.
161 */
162 if (devinfo->gen >= 6 && op == BRW_OPCODE_DO)
163 return "do";
164
165 assert(brw_opcode_desc(devinfo, op)->name);
166 return brw_opcode_desc(devinfo, op)->name;
167 case FS_OPCODE_FB_WRITE:
168 return "fb_write";
169 case FS_OPCODE_FB_WRITE_LOGICAL:
170 return "fb_write_logical";
171 case FS_OPCODE_REP_FB_WRITE:
172 return "rep_fb_write";
173 case FS_OPCODE_FB_READ:
174 return "fb_read";
175 case FS_OPCODE_FB_READ_LOGICAL:
176 return "fb_read_logical";
177
178 case SHADER_OPCODE_RCP:
179 return "rcp";
180 case SHADER_OPCODE_RSQ:
181 return "rsq";
182 case SHADER_OPCODE_SQRT:
183 return "sqrt";
184 case SHADER_OPCODE_EXP2:
185 return "exp2";
186 case SHADER_OPCODE_LOG2:
187 return "log2";
188 case SHADER_OPCODE_POW:
189 return "pow";
190 case SHADER_OPCODE_INT_QUOTIENT:
191 return "int_quot";
192 case SHADER_OPCODE_INT_REMAINDER:
193 return "int_rem";
194 case SHADER_OPCODE_SIN:
195 return "sin";
196 case SHADER_OPCODE_COS:
197 return "cos";
198
199 case SHADER_OPCODE_TEX:
200 return "tex";
201 case SHADER_OPCODE_TEX_LOGICAL:
202 return "tex_logical";
203 case SHADER_OPCODE_TXD:
204 return "txd";
205 case SHADER_OPCODE_TXD_LOGICAL:
206 return "txd_logical";
207 case SHADER_OPCODE_TXF:
208 return "txf";
209 case SHADER_OPCODE_TXF_LOGICAL:
210 return "txf_logical";
211 case SHADER_OPCODE_TXF_LZ:
212 return "txf_lz";
213 case SHADER_OPCODE_TXL:
214 return "txl";
215 case SHADER_OPCODE_TXL_LOGICAL:
216 return "txl_logical";
217 case SHADER_OPCODE_TXL_LZ:
218 return "txl_lz";
219 case SHADER_OPCODE_TXS:
220 return "txs";
221 case SHADER_OPCODE_TXS_LOGICAL:
222 return "txs_logical";
223 case FS_OPCODE_TXB:
224 return "txb";
225 case FS_OPCODE_TXB_LOGICAL:
226 return "txb_logical";
227 case SHADER_OPCODE_TXF_CMS:
228 return "txf_cms";
229 case SHADER_OPCODE_TXF_CMS_LOGICAL:
230 return "txf_cms_logical";
231 case SHADER_OPCODE_TXF_CMS_W:
232 return "txf_cms_w";
233 case SHADER_OPCODE_TXF_CMS_W_LOGICAL:
234 return "txf_cms_w_logical";
235 case SHADER_OPCODE_TXF_UMS:
236 return "txf_ums";
237 case SHADER_OPCODE_TXF_UMS_LOGICAL:
238 return "txf_ums_logical";
239 case SHADER_OPCODE_TXF_MCS:
240 return "txf_mcs";
241 case SHADER_OPCODE_TXF_MCS_LOGICAL:
242 return "txf_mcs_logical";
243 case SHADER_OPCODE_LOD:
244 return "lod";
245 case SHADER_OPCODE_LOD_LOGICAL:
246 return "lod_logical";
247 case SHADER_OPCODE_TG4:
248 return "tg4";
249 case SHADER_OPCODE_TG4_LOGICAL:
250 return "tg4_logical";
251 case SHADER_OPCODE_TG4_OFFSET:
252 return "tg4_offset";
253 case SHADER_OPCODE_TG4_OFFSET_LOGICAL:
254 return "tg4_offset_logical";
255 case SHADER_OPCODE_SAMPLEINFO:
256 return "sampleinfo";
257 case SHADER_OPCODE_SAMPLEINFO_LOGICAL:
258 return "sampleinfo_logical";
259
260 case SHADER_OPCODE_SHADER_TIME_ADD:
261 return "shader_time_add";
262
263 case SHADER_OPCODE_UNTYPED_ATOMIC:
264 return "untyped_atomic";
265 case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL:
266 return "untyped_atomic_logical";
267 case SHADER_OPCODE_UNTYPED_SURFACE_READ:
268 return "untyped_surface_read";
269 case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL:
270 return "untyped_surface_read_logical";
271 case SHADER_OPCODE_UNTYPED_SURFACE_WRITE:
272 return "untyped_surface_write";
273 case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL:
274 return "untyped_surface_write_logical";
275 case SHADER_OPCODE_TYPED_ATOMIC:
276 return "typed_atomic";
277 case SHADER_OPCODE_TYPED_ATOMIC_LOGICAL:
278 return "typed_atomic_logical";
279 case SHADER_OPCODE_TYPED_SURFACE_READ:
280 return "typed_surface_read";
281 case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL:
282 return "typed_surface_read_logical";
283 case SHADER_OPCODE_TYPED_SURFACE_WRITE:
284 return "typed_surface_write";
285 case SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL:
286 return "typed_surface_write_logical";
287 case SHADER_OPCODE_MEMORY_FENCE:
288 return "memory_fence";
289
290 case SHADER_OPCODE_LOAD_PAYLOAD:
291 return "load_payload";
292 case FS_OPCODE_PACK:
293 return "pack";
294
295 case SHADER_OPCODE_GEN4_SCRATCH_READ:
296 return "gen4_scratch_read";
297 case SHADER_OPCODE_GEN4_SCRATCH_WRITE:
298 return "gen4_scratch_write";
299 case SHADER_OPCODE_GEN7_SCRATCH_READ:
300 return "gen7_scratch_read";
301 case SHADER_OPCODE_URB_WRITE_SIMD8:
302 return "gen8_urb_write_simd8";
303 case SHADER_OPCODE_URB_WRITE_SIMD8_PER_SLOT:
304 return "gen8_urb_write_simd8_per_slot";
305 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED:
306 return "gen8_urb_write_simd8_masked";
307 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT:
308 return "gen8_urb_write_simd8_masked_per_slot";
309 case SHADER_OPCODE_URB_READ_SIMD8:
310 return "urb_read_simd8";
311 case SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT:
312 return "urb_read_simd8_per_slot";
313
314 case SHADER_OPCODE_FIND_LIVE_CHANNEL:
315 return "find_live_channel";
316 case SHADER_OPCODE_BROADCAST:
317 return "broadcast";
318
319 case VEC4_OPCODE_MOV_BYTES:
320 return "mov_bytes";
321 case VEC4_OPCODE_PACK_BYTES:
322 return "pack_bytes";
323 case VEC4_OPCODE_UNPACK_UNIFORM:
324 return "unpack_uniform";
325 case VEC4_OPCODE_FROM_DOUBLE:
326 return "double_to_single";
327 case VEC4_OPCODE_TO_DOUBLE:
328 return "single_to_double";
329 case VEC4_OPCODE_PICK_LOW_32BIT:
330 return "pick_low_32bit";
331 case VEC4_OPCODE_PICK_HIGH_32BIT:
332 return "pick_high_32bit";
333 case VEC4_OPCODE_SET_LOW_32BIT:
334 return "set_low_32bit";
335 case VEC4_OPCODE_SET_HIGH_32BIT:
336 return "set_high_32bit";
337
338 case FS_OPCODE_DDX_COARSE:
339 return "ddx_coarse";
340 case FS_OPCODE_DDX_FINE:
341 return "ddx_fine";
342 case FS_OPCODE_DDY_COARSE:
343 return "ddy_coarse";
344 case FS_OPCODE_DDY_FINE:
345 return "ddy_fine";
346
347 case FS_OPCODE_CINTERP:
348 return "cinterp";
349 case FS_OPCODE_LINTERP:
350 return "linterp";
351
352 case FS_OPCODE_PIXEL_X:
353 return "pixel_x";
354 case FS_OPCODE_PIXEL_Y:
355 return "pixel_y";
356
357 case FS_OPCODE_GET_BUFFER_SIZE:
358 return "fs_get_buffer_size";
359
360 case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD:
361 return "uniform_pull_const";
362 case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD_GEN7:
363 return "uniform_pull_const_gen7";
364 case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN4:
365 return "varying_pull_const_gen4";
366 case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7:
367 return "varying_pull_const_gen7";
368 case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_LOGICAL:
369 return "varying_pull_const_logical";
370
371 case FS_OPCODE_MOV_DISPATCH_TO_FLAGS:
372 return "mov_dispatch_to_flags";
373 case FS_OPCODE_DISCARD_JUMP:
374 return "discard_jump";
375
376 case FS_OPCODE_SET_SAMPLE_ID:
377 return "set_sample_id";
378
379 case FS_OPCODE_PACK_HALF_2x16_SPLIT:
380 return "pack_half_2x16_split";
381 case FS_OPCODE_UNPACK_HALF_2x16_SPLIT_X:
382 return "unpack_half_2x16_split_x";
383 case FS_OPCODE_UNPACK_HALF_2x16_SPLIT_Y:
384 return "unpack_half_2x16_split_y";
385
386 case FS_OPCODE_PLACEHOLDER_HALT:
387 return "placeholder_halt";
388
389 case FS_OPCODE_INTERPOLATE_AT_SAMPLE:
390 return "interp_sample";
391 case FS_OPCODE_INTERPOLATE_AT_SHARED_OFFSET:
392 return "interp_shared_offset";
393 case FS_OPCODE_INTERPOLATE_AT_PER_SLOT_OFFSET:
394 return "interp_per_slot_offset";
395
396 case VS_OPCODE_URB_WRITE:
397 return "vs_urb_write";
398 case VS_OPCODE_PULL_CONSTANT_LOAD:
399 return "pull_constant_load";
400 case VS_OPCODE_PULL_CONSTANT_LOAD_GEN7:
401 return "pull_constant_load_gen7";
402
403 case VS_OPCODE_SET_SIMD4X2_HEADER_GEN9:
404 return "set_simd4x2_header_gen9";
405
406 case VS_OPCODE_GET_BUFFER_SIZE:
407 return "vs_get_buffer_size";
408
409 case VS_OPCODE_UNPACK_FLAGS_SIMD4X2:
410 return "unpack_flags_simd4x2";
411
412 case GS_OPCODE_URB_WRITE:
413 return "gs_urb_write";
414 case GS_OPCODE_URB_WRITE_ALLOCATE:
415 return "gs_urb_write_allocate";
416 case GS_OPCODE_THREAD_END:
417 return "gs_thread_end";
418 case GS_OPCODE_SET_WRITE_OFFSET:
419 return "set_write_offset";
420 case GS_OPCODE_SET_VERTEX_COUNT:
421 return "set_vertex_count";
422 case GS_OPCODE_SET_DWORD_2:
423 return "set_dword_2";
424 case GS_OPCODE_PREPARE_CHANNEL_MASKS:
425 return "prepare_channel_masks";
426 case GS_OPCODE_SET_CHANNEL_MASKS:
427 return "set_channel_masks";
428 case GS_OPCODE_GET_INSTANCE_ID:
429 return "get_instance_id";
430 case GS_OPCODE_FF_SYNC:
431 return "ff_sync";
432 case GS_OPCODE_SET_PRIMITIVE_ID:
433 return "set_primitive_id";
434 case GS_OPCODE_SVB_WRITE:
435 return "gs_svb_write";
436 case GS_OPCODE_SVB_SET_DST_INDEX:
437 return "gs_svb_set_dst_index";
438 case GS_OPCODE_FF_SYNC_SET_PRIMITIVES:
439 return "gs_ff_sync_set_primitives";
440 case CS_OPCODE_CS_TERMINATE:
441 return "cs_terminate";
442 case SHADER_OPCODE_BARRIER:
443 return "barrier";
444 case SHADER_OPCODE_MULH:
445 return "mulh";
446 case SHADER_OPCODE_MOV_INDIRECT:
447 return "mov_indirect";
448
449 case VEC4_OPCODE_URB_READ:
450 return "urb_read";
451 case TCS_OPCODE_GET_INSTANCE_ID:
452 return "tcs_get_instance_id";
453 case TCS_OPCODE_URB_WRITE:
454 return "tcs_urb_write";
455 case TCS_OPCODE_SET_INPUT_URB_OFFSETS:
456 return "tcs_set_input_urb_offsets";
457 case TCS_OPCODE_SET_OUTPUT_URB_OFFSETS:
458 return "tcs_set_output_urb_offsets";
459 case TCS_OPCODE_GET_PRIMITIVE_ID:
460 return "tcs_get_primitive_id";
461 case TCS_OPCODE_CREATE_BARRIER_HEADER:
462 return "tcs_create_barrier_header";
463 case TCS_OPCODE_SRC0_010_IS_ZERO:
464 return "tcs_src0<0,1,0>_is_zero";
465 case TCS_OPCODE_RELEASE_INPUT:
466 return "tcs_release_input";
467 case TCS_OPCODE_THREAD_END:
468 return "tcs_thread_end";
469 case TES_OPCODE_CREATE_INPUT_READ_HEADER:
470 return "tes_create_input_read_header";
471 case TES_OPCODE_ADD_INDIRECT_URB_OFFSET:
472 return "tes_add_indirect_urb_offset";
473 case TES_OPCODE_GET_PRIMITIVE_ID:
474 return "tes_get_primitive_id";
475 }
476
477 unreachable("not reached");
478 }
479
480 bool
brw_saturate_immediate(enum brw_reg_type type,struct brw_reg * reg)481 brw_saturate_immediate(enum brw_reg_type type, struct brw_reg *reg)
482 {
483 union {
484 unsigned ud;
485 int d;
486 float f;
487 double df;
488 } imm, sat_imm = { 0 };
489
490 const unsigned size = type_sz(type);
491
492 /* We want to either do a 32-bit or 64-bit data copy, the type is otherwise
493 * irrelevant, so just check the size of the type and copy from/to an
494 * appropriately sized field.
495 */
496 if (size < 8)
497 imm.ud = reg->ud;
498 else
499 imm.df = reg->df;
500
501 switch (type) {
502 case BRW_REGISTER_TYPE_UD:
503 case BRW_REGISTER_TYPE_D:
504 case BRW_REGISTER_TYPE_UW:
505 case BRW_REGISTER_TYPE_W:
506 case BRW_REGISTER_TYPE_UQ:
507 case BRW_REGISTER_TYPE_Q:
508 /* Nothing to do. */
509 return false;
510 case BRW_REGISTER_TYPE_F:
511 sat_imm.f = CLAMP(imm.f, 0.0f, 1.0f);
512 break;
513 case BRW_REGISTER_TYPE_DF:
514 sat_imm.df = CLAMP(imm.df, 0.0, 1.0);
515 break;
516 case BRW_REGISTER_TYPE_UB:
517 case BRW_REGISTER_TYPE_B:
518 unreachable("no UB/B immediates");
519 case BRW_REGISTER_TYPE_V:
520 case BRW_REGISTER_TYPE_UV:
521 case BRW_REGISTER_TYPE_VF:
522 unreachable("unimplemented: saturate vector immediate");
523 case BRW_REGISTER_TYPE_HF:
524 unreachable("unimplemented: saturate HF immediate");
525 }
526
527 if (size < 8) {
528 if (imm.ud != sat_imm.ud) {
529 reg->ud = sat_imm.ud;
530 return true;
531 }
532 } else {
533 if (imm.df != sat_imm.df) {
534 reg->df = sat_imm.df;
535 return true;
536 }
537 }
538 return false;
539 }
540
541 bool
brw_negate_immediate(enum brw_reg_type type,struct brw_reg * reg)542 brw_negate_immediate(enum brw_reg_type type, struct brw_reg *reg)
543 {
544 switch (type) {
545 case BRW_REGISTER_TYPE_D:
546 case BRW_REGISTER_TYPE_UD:
547 reg->d = -reg->d;
548 return true;
549 case BRW_REGISTER_TYPE_W:
550 case BRW_REGISTER_TYPE_UW:
551 reg->d = -(int16_t)reg->ud;
552 return true;
553 case BRW_REGISTER_TYPE_F:
554 reg->f = -reg->f;
555 return true;
556 case BRW_REGISTER_TYPE_VF:
557 reg->ud ^= 0x80808080;
558 return true;
559 case BRW_REGISTER_TYPE_DF:
560 reg->df = -reg->df;
561 return true;
562 case BRW_REGISTER_TYPE_UB:
563 case BRW_REGISTER_TYPE_B:
564 unreachable("no UB/B immediates");
565 case BRW_REGISTER_TYPE_UV:
566 case BRW_REGISTER_TYPE_V:
567 assert(!"unimplemented: negate UV/V immediate");
568 case BRW_REGISTER_TYPE_UQ:
569 case BRW_REGISTER_TYPE_Q:
570 assert(!"unimplemented: negate UQ/Q immediate");
571 case BRW_REGISTER_TYPE_HF:
572 assert(!"unimplemented: negate HF immediate");
573 }
574
575 return false;
576 }
577
578 bool
brw_abs_immediate(enum brw_reg_type type,struct brw_reg * reg)579 brw_abs_immediate(enum brw_reg_type type, struct brw_reg *reg)
580 {
581 switch (type) {
582 case BRW_REGISTER_TYPE_D:
583 reg->d = abs(reg->d);
584 return true;
585 case BRW_REGISTER_TYPE_W:
586 reg->d = abs((int16_t)reg->ud);
587 return true;
588 case BRW_REGISTER_TYPE_F:
589 reg->f = fabsf(reg->f);
590 return true;
591 case BRW_REGISTER_TYPE_DF:
592 reg->df = fabs(reg->df);
593 return true;
594 case BRW_REGISTER_TYPE_VF:
595 reg->ud &= ~0x80808080;
596 return true;
597 case BRW_REGISTER_TYPE_UB:
598 case BRW_REGISTER_TYPE_B:
599 unreachable("no UB/B immediates");
600 case BRW_REGISTER_TYPE_UQ:
601 case BRW_REGISTER_TYPE_UD:
602 case BRW_REGISTER_TYPE_UW:
603 case BRW_REGISTER_TYPE_UV:
604 /* Presumably the absolute value modifier on an unsigned source is a
605 * nop, but it would be nice to confirm.
606 */
607 assert(!"unimplemented: abs unsigned immediate");
608 case BRW_REGISTER_TYPE_V:
609 assert(!"unimplemented: abs V immediate");
610 case BRW_REGISTER_TYPE_Q:
611 assert(!"unimplemented: abs Q immediate");
612 case BRW_REGISTER_TYPE_HF:
613 assert(!"unimplemented: abs HF immediate");
614 }
615
616 return false;
617 }
618
619 /**
620 * Get the appropriate atomic op for an image atomic intrinsic.
621 */
622 unsigned
get_atomic_counter_op(nir_intrinsic_op op)623 get_atomic_counter_op(nir_intrinsic_op op)
624 {
625 switch (op) {
626 case nir_intrinsic_atomic_counter_inc:
627 return BRW_AOP_INC;
628 case nir_intrinsic_atomic_counter_dec:
629 return BRW_AOP_PREDEC;
630 case nir_intrinsic_atomic_counter_add:
631 return BRW_AOP_ADD;
632 case nir_intrinsic_atomic_counter_min:
633 return BRW_AOP_UMIN;
634 case nir_intrinsic_atomic_counter_max:
635 return BRW_AOP_UMAX;
636 case nir_intrinsic_atomic_counter_and:
637 return BRW_AOP_AND;
638 case nir_intrinsic_atomic_counter_or:
639 return BRW_AOP_OR;
640 case nir_intrinsic_atomic_counter_xor:
641 return BRW_AOP_XOR;
642 case nir_intrinsic_atomic_counter_exchange:
643 return BRW_AOP_MOV;
644 case nir_intrinsic_atomic_counter_comp_swap:
645 return BRW_AOP_CMPWR;
646 default:
647 unreachable("Not reachable.");
648 }
649 }
650
backend_shader(const struct brw_compiler * compiler,void * log_data,void * mem_ctx,const nir_shader * shader,struct brw_stage_prog_data * stage_prog_data)651 backend_shader::backend_shader(const struct brw_compiler *compiler,
652 void *log_data,
653 void *mem_ctx,
654 const nir_shader *shader,
655 struct brw_stage_prog_data *stage_prog_data)
656 : compiler(compiler),
657 log_data(log_data),
658 devinfo(compiler->devinfo),
659 nir(shader),
660 stage_prog_data(stage_prog_data),
661 mem_ctx(mem_ctx),
662 cfg(NULL),
663 stage(shader->stage)
664 {
665 debug_enabled = INTEL_DEBUG & intel_debug_flag_for_shader_stage(stage);
666 stage_name = _mesa_shader_stage_to_string(stage);
667 stage_abbrev = _mesa_shader_stage_to_abbrev(stage);
668 }
669
670 bool
equals(const backend_reg & r) const671 backend_reg::equals(const backend_reg &r) const
672 {
673 return brw_regs_equal(this, &r) && offset == r.offset;
674 }
675
676 bool
is_zero() const677 backend_reg::is_zero() const
678 {
679 if (file != IMM)
680 return false;
681
682 switch (type) {
683 case BRW_REGISTER_TYPE_F:
684 return f == 0;
685 case BRW_REGISTER_TYPE_DF:
686 return df == 0;
687 case BRW_REGISTER_TYPE_D:
688 case BRW_REGISTER_TYPE_UD:
689 return d == 0;
690 default:
691 return false;
692 }
693 }
694
695 bool
is_one() const696 backend_reg::is_one() const
697 {
698 if (file != IMM)
699 return false;
700
701 switch (type) {
702 case BRW_REGISTER_TYPE_F:
703 return f == 1.0f;
704 case BRW_REGISTER_TYPE_DF:
705 return df == 1.0;
706 case BRW_REGISTER_TYPE_D:
707 case BRW_REGISTER_TYPE_UD:
708 return d == 1;
709 default:
710 return false;
711 }
712 }
713
714 bool
is_negative_one() const715 backend_reg::is_negative_one() const
716 {
717 if (file != IMM)
718 return false;
719
720 switch (type) {
721 case BRW_REGISTER_TYPE_F:
722 return f == -1.0;
723 case BRW_REGISTER_TYPE_DF:
724 return df == -1.0;
725 case BRW_REGISTER_TYPE_D:
726 return d == -1;
727 default:
728 return false;
729 }
730 }
731
732 bool
is_null() const733 backend_reg::is_null() const
734 {
735 return file == ARF && nr == BRW_ARF_NULL;
736 }
737
738
739 bool
is_accumulator() const740 backend_reg::is_accumulator() const
741 {
742 return file == ARF && nr == BRW_ARF_ACCUMULATOR;
743 }
744
745 bool
is_commutative() const746 backend_instruction::is_commutative() const
747 {
748 switch (opcode) {
749 case BRW_OPCODE_AND:
750 case BRW_OPCODE_OR:
751 case BRW_OPCODE_XOR:
752 case BRW_OPCODE_ADD:
753 case BRW_OPCODE_MUL:
754 case SHADER_OPCODE_MULH:
755 return true;
756 case BRW_OPCODE_SEL:
757 /* MIN and MAX are commutative. */
758 if (conditional_mod == BRW_CONDITIONAL_GE ||
759 conditional_mod == BRW_CONDITIONAL_L) {
760 return true;
761 }
762 /* fallthrough */
763 default:
764 return false;
765 }
766 }
767
768 bool
is_3src(const struct gen_device_info * devinfo) const769 backend_instruction::is_3src(const struct gen_device_info *devinfo) const
770 {
771 return ::is_3src(devinfo, opcode);
772 }
773
774 bool
is_tex() const775 backend_instruction::is_tex() const
776 {
777 return (opcode == SHADER_OPCODE_TEX ||
778 opcode == FS_OPCODE_TXB ||
779 opcode == SHADER_OPCODE_TXD ||
780 opcode == SHADER_OPCODE_TXF ||
781 opcode == SHADER_OPCODE_TXF_LZ ||
782 opcode == SHADER_OPCODE_TXF_CMS ||
783 opcode == SHADER_OPCODE_TXF_CMS_W ||
784 opcode == SHADER_OPCODE_TXF_UMS ||
785 opcode == SHADER_OPCODE_TXF_MCS ||
786 opcode == SHADER_OPCODE_TXL ||
787 opcode == SHADER_OPCODE_TXL_LZ ||
788 opcode == SHADER_OPCODE_TXS ||
789 opcode == SHADER_OPCODE_LOD ||
790 opcode == SHADER_OPCODE_TG4 ||
791 opcode == SHADER_OPCODE_TG4_OFFSET ||
792 opcode == SHADER_OPCODE_SAMPLEINFO);
793 }
794
795 bool
is_math() const796 backend_instruction::is_math() const
797 {
798 return (opcode == SHADER_OPCODE_RCP ||
799 opcode == SHADER_OPCODE_RSQ ||
800 opcode == SHADER_OPCODE_SQRT ||
801 opcode == SHADER_OPCODE_EXP2 ||
802 opcode == SHADER_OPCODE_LOG2 ||
803 opcode == SHADER_OPCODE_SIN ||
804 opcode == SHADER_OPCODE_COS ||
805 opcode == SHADER_OPCODE_INT_QUOTIENT ||
806 opcode == SHADER_OPCODE_INT_REMAINDER ||
807 opcode == SHADER_OPCODE_POW);
808 }
809
810 bool
is_control_flow() const811 backend_instruction::is_control_flow() const
812 {
813 switch (opcode) {
814 case BRW_OPCODE_DO:
815 case BRW_OPCODE_WHILE:
816 case BRW_OPCODE_IF:
817 case BRW_OPCODE_ELSE:
818 case BRW_OPCODE_ENDIF:
819 case BRW_OPCODE_BREAK:
820 case BRW_OPCODE_CONTINUE:
821 return true;
822 default:
823 return false;
824 }
825 }
826
827 bool
can_do_source_mods() const828 backend_instruction::can_do_source_mods() const
829 {
830 switch (opcode) {
831 case BRW_OPCODE_ADDC:
832 case BRW_OPCODE_BFE:
833 case BRW_OPCODE_BFI1:
834 case BRW_OPCODE_BFI2:
835 case BRW_OPCODE_BFREV:
836 case BRW_OPCODE_CBIT:
837 case BRW_OPCODE_FBH:
838 case BRW_OPCODE_FBL:
839 case BRW_OPCODE_SUBB:
840 return false;
841 default:
842 return true;
843 }
844 }
845
846 bool
can_do_saturate() const847 backend_instruction::can_do_saturate() const
848 {
849 switch (opcode) {
850 case BRW_OPCODE_ADD:
851 case BRW_OPCODE_ASR:
852 case BRW_OPCODE_AVG:
853 case BRW_OPCODE_DP2:
854 case BRW_OPCODE_DP3:
855 case BRW_OPCODE_DP4:
856 case BRW_OPCODE_DPH:
857 case BRW_OPCODE_F16TO32:
858 case BRW_OPCODE_F32TO16:
859 case BRW_OPCODE_LINE:
860 case BRW_OPCODE_LRP:
861 case BRW_OPCODE_MAC:
862 case BRW_OPCODE_MAD:
863 case BRW_OPCODE_MATH:
864 case BRW_OPCODE_MOV:
865 case BRW_OPCODE_MUL:
866 case SHADER_OPCODE_MULH:
867 case BRW_OPCODE_PLN:
868 case BRW_OPCODE_RNDD:
869 case BRW_OPCODE_RNDE:
870 case BRW_OPCODE_RNDU:
871 case BRW_OPCODE_RNDZ:
872 case BRW_OPCODE_SEL:
873 case BRW_OPCODE_SHL:
874 case BRW_OPCODE_SHR:
875 case FS_OPCODE_LINTERP:
876 case SHADER_OPCODE_COS:
877 case SHADER_OPCODE_EXP2:
878 case SHADER_OPCODE_LOG2:
879 case SHADER_OPCODE_POW:
880 case SHADER_OPCODE_RCP:
881 case SHADER_OPCODE_RSQ:
882 case SHADER_OPCODE_SIN:
883 case SHADER_OPCODE_SQRT:
884 return true;
885 default:
886 return false;
887 }
888 }
889
890 bool
can_do_cmod() const891 backend_instruction::can_do_cmod() const
892 {
893 switch (opcode) {
894 case BRW_OPCODE_ADD:
895 case BRW_OPCODE_ADDC:
896 case BRW_OPCODE_AND:
897 case BRW_OPCODE_ASR:
898 case BRW_OPCODE_AVG:
899 case BRW_OPCODE_CMP:
900 case BRW_OPCODE_CMPN:
901 case BRW_OPCODE_DP2:
902 case BRW_OPCODE_DP3:
903 case BRW_OPCODE_DP4:
904 case BRW_OPCODE_DPH:
905 case BRW_OPCODE_F16TO32:
906 case BRW_OPCODE_F32TO16:
907 case BRW_OPCODE_FRC:
908 case BRW_OPCODE_LINE:
909 case BRW_OPCODE_LRP:
910 case BRW_OPCODE_LZD:
911 case BRW_OPCODE_MAC:
912 case BRW_OPCODE_MACH:
913 case BRW_OPCODE_MAD:
914 case BRW_OPCODE_MOV:
915 case BRW_OPCODE_MUL:
916 case BRW_OPCODE_NOT:
917 case BRW_OPCODE_OR:
918 case BRW_OPCODE_PLN:
919 case BRW_OPCODE_RNDD:
920 case BRW_OPCODE_RNDE:
921 case BRW_OPCODE_RNDU:
922 case BRW_OPCODE_RNDZ:
923 case BRW_OPCODE_SAD2:
924 case BRW_OPCODE_SADA2:
925 case BRW_OPCODE_SHL:
926 case BRW_OPCODE_SHR:
927 case BRW_OPCODE_SUBB:
928 case BRW_OPCODE_XOR:
929 case FS_OPCODE_CINTERP:
930 case FS_OPCODE_LINTERP:
931 return true;
932 default:
933 return false;
934 }
935 }
936
937 bool
reads_accumulator_implicitly() const938 backend_instruction::reads_accumulator_implicitly() const
939 {
940 switch (opcode) {
941 case BRW_OPCODE_MAC:
942 case BRW_OPCODE_MACH:
943 case BRW_OPCODE_SADA2:
944 return true;
945 default:
946 return false;
947 }
948 }
949
950 bool
writes_accumulator_implicitly(const struct gen_device_info * devinfo) const951 backend_instruction::writes_accumulator_implicitly(const struct gen_device_info *devinfo) const
952 {
953 return writes_accumulator ||
954 (devinfo->gen < 6 &&
955 ((opcode >= BRW_OPCODE_ADD && opcode < BRW_OPCODE_NOP) ||
956 (opcode >= FS_OPCODE_DDX_COARSE && opcode <= FS_OPCODE_LINTERP &&
957 opcode != FS_OPCODE_CINTERP)));
958 }
959
960 bool
has_side_effects() const961 backend_instruction::has_side_effects() const
962 {
963 switch (opcode) {
964 case SHADER_OPCODE_UNTYPED_ATOMIC:
965 case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL:
966 case SHADER_OPCODE_GEN4_SCRATCH_WRITE:
967 case SHADER_OPCODE_UNTYPED_SURFACE_WRITE:
968 case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL:
969 case SHADER_OPCODE_TYPED_ATOMIC:
970 case SHADER_OPCODE_TYPED_ATOMIC_LOGICAL:
971 case SHADER_OPCODE_TYPED_SURFACE_WRITE:
972 case SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL:
973 case SHADER_OPCODE_MEMORY_FENCE:
974 case SHADER_OPCODE_URB_WRITE_SIMD8:
975 case SHADER_OPCODE_URB_WRITE_SIMD8_PER_SLOT:
976 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED:
977 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT:
978 case FS_OPCODE_FB_WRITE:
979 case FS_OPCODE_FB_WRITE_LOGICAL:
980 case SHADER_OPCODE_BARRIER:
981 case TCS_OPCODE_URB_WRITE:
982 case TCS_OPCODE_RELEASE_INPUT:
983 return true;
984 default:
985 return false;
986 }
987 }
988
989 bool
is_volatile() const990 backend_instruction::is_volatile() const
991 {
992 switch (opcode) {
993 case SHADER_OPCODE_UNTYPED_SURFACE_READ:
994 case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL:
995 case SHADER_OPCODE_TYPED_SURFACE_READ:
996 case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL:
997 case SHADER_OPCODE_URB_READ_SIMD8:
998 case SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT:
999 case VEC4_OPCODE_URB_READ:
1000 return true;
1001 default:
1002 return false;
1003 }
1004 }
1005
1006 #ifndef NDEBUG
1007 static bool
inst_is_in_block(const bblock_t * block,const backend_instruction * inst)1008 inst_is_in_block(const bblock_t *block, const backend_instruction *inst)
1009 {
1010 bool found = false;
1011 foreach_inst_in_block (backend_instruction, i, block) {
1012 if (inst == i) {
1013 found = true;
1014 }
1015 }
1016 return found;
1017 }
1018 #endif
1019
1020 static void
adjust_later_block_ips(bblock_t * start_block,int ip_adjustment)1021 adjust_later_block_ips(bblock_t *start_block, int ip_adjustment)
1022 {
1023 for (bblock_t *block_iter = start_block->next();
1024 block_iter;
1025 block_iter = block_iter->next()) {
1026 block_iter->start_ip += ip_adjustment;
1027 block_iter->end_ip += ip_adjustment;
1028 }
1029 }
1030
1031 void
insert_after(bblock_t * block,backend_instruction * inst)1032 backend_instruction::insert_after(bblock_t *block, backend_instruction *inst)
1033 {
1034 assert(this != inst);
1035
1036 if (!this->is_head_sentinel())
1037 assert(inst_is_in_block(block, this) || !"Instruction not in block");
1038
1039 block->end_ip++;
1040
1041 adjust_later_block_ips(block, 1);
1042
1043 exec_node::insert_after(inst);
1044 }
1045
1046 void
insert_before(bblock_t * block,backend_instruction * inst)1047 backend_instruction::insert_before(bblock_t *block, backend_instruction *inst)
1048 {
1049 assert(this != inst);
1050
1051 if (!this->is_tail_sentinel())
1052 assert(inst_is_in_block(block, this) || !"Instruction not in block");
1053
1054 block->end_ip++;
1055
1056 adjust_later_block_ips(block, 1);
1057
1058 exec_node::insert_before(inst);
1059 }
1060
1061 void
insert_before(bblock_t * block,exec_list * list)1062 backend_instruction::insert_before(bblock_t *block, exec_list *list)
1063 {
1064 assert(inst_is_in_block(block, this) || !"Instruction not in block");
1065
1066 unsigned num_inst = list->length();
1067
1068 block->end_ip += num_inst;
1069
1070 adjust_later_block_ips(block, num_inst);
1071
1072 exec_node::insert_before(list);
1073 }
1074
1075 void
remove(bblock_t * block)1076 backend_instruction::remove(bblock_t *block)
1077 {
1078 assert(inst_is_in_block(block, this) || !"Instruction not in block");
1079
1080 adjust_later_block_ips(block, -1);
1081
1082 if (block->start_ip == block->end_ip) {
1083 block->cfg->remove_block(block);
1084 } else {
1085 block->end_ip--;
1086 }
1087
1088 exec_node::remove();
1089 }
1090
1091 void
dump_instructions()1092 backend_shader::dump_instructions()
1093 {
1094 dump_instructions(NULL);
1095 }
1096
1097 void
dump_instructions(const char * name)1098 backend_shader::dump_instructions(const char *name)
1099 {
1100 FILE *file = stderr;
1101 if (name && geteuid() != 0) {
1102 file = fopen(name, "w");
1103 if (!file)
1104 file = stderr;
1105 }
1106
1107 if (cfg) {
1108 int ip = 0;
1109 foreach_block_and_inst(block, backend_instruction, inst, cfg) {
1110 if (!unlikely(INTEL_DEBUG & DEBUG_OPTIMIZER))
1111 fprintf(file, "%4d: ", ip++);
1112 dump_instruction(inst, file);
1113 }
1114 } else {
1115 int ip = 0;
1116 foreach_in_list(backend_instruction, inst, &instructions) {
1117 if (!unlikely(INTEL_DEBUG & DEBUG_OPTIMIZER))
1118 fprintf(file, "%4d: ", ip++);
1119 dump_instruction(inst, file);
1120 }
1121 }
1122
1123 if (file != stderr) {
1124 fclose(file);
1125 }
1126 }
1127
1128 void
calculate_cfg()1129 backend_shader::calculate_cfg()
1130 {
1131 if (this->cfg)
1132 return;
1133 cfg = new(mem_ctx) cfg_t(&this->instructions);
1134 }
1135
1136 /**
1137 * Sets up the starting offsets for the groups of binding table entries
1138 * commong to all pipeline stages.
1139 *
1140 * Unused groups are initialized to 0xd0d0d0d0 to make it obvious that they're
1141 * unused but also make sure that addition of small offsets to them will
1142 * trigger some of our asserts that surface indices are < BRW_MAX_SURFACES.
1143 */
1144 uint32_t
brw_assign_common_binding_table_offsets(const struct gen_device_info * devinfo,const struct gl_program * prog,struct brw_stage_prog_data * stage_prog_data,uint32_t next_binding_table_offset)1145 brw_assign_common_binding_table_offsets(const struct gen_device_info *devinfo,
1146 const struct gl_program *prog,
1147 struct brw_stage_prog_data *stage_prog_data,
1148 uint32_t next_binding_table_offset)
1149 {
1150 int num_textures = util_last_bit(prog->SamplersUsed);
1151
1152 stage_prog_data->binding_table.texture_start = next_binding_table_offset;
1153 next_binding_table_offset += num_textures;
1154
1155 if (prog->info.num_ubos) {
1156 assert(prog->info.num_ubos <= BRW_MAX_UBO);
1157 stage_prog_data->binding_table.ubo_start = next_binding_table_offset;
1158 next_binding_table_offset += prog->info.num_ubos;
1159 } else {
1160 stage_prog_data->binding_table.ubo_start = 0xd0d0d0d0;
1161 }
1162
1163 if (prog->info.num_ssbos) {
1164 assert(prog->info.num_ssbos <= BRW_MAX_SSBO);
1165 stage_prog_data->binding_table.ssbo_start = next_binding_table_offset;
1166 next_binding_table_offset += prog->info.num_ssbos;
1167 } else {
1168 stage_prog_data->binding_table.ssbo_start = 0xd0d0d0d0;
1169 }
1170
1171 if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
1172 stage_prog_data->binding_table.shader_time_start = next_binding_table_offset;
1173 next_binding_table_offset++;
1174 } else {
1175 stage_prog_data->binding_table.shader_time_start = 0xd0d0d0d0;
1176 }
1177
1178 if (prog->nir->info->uses_texture_gather) {
1179 if (devinfo->gen >= 8) {
1180 stage_prog_data->binding_table.gather_texture_start =
1181 stage_prog_data->binding_table.texture_start;
1182 } else {
1183 stage_prog_data->binding_table.gather_texture_start = next_binding_table_offset;
1184 next_binding_table_offset += num_textures;
1185 }
1186 } else {
1187 stage_prog_data->binding_table.gather_texture_start = 0xd0d0d0d0;
1188 }
1189
1190 if (prog->info.num_abos) {
1191 stage_prog_data->binding_table.abo_start = next_binding_table_offset;
1192 next_binding_table_offset += prog->info.num_abos;
1193 } else {
1194 stage_prog_data->binding_table.abo_start = 0xd0d0d0d0;
1195 }
1196
1197 if (prog->info.num_images) {
1198 stage_prog_data->binding_table.image_start = next_binding_table_offset;
1199 next_binding_table_offset += prog->info.num_images;
1200 } else {
1201 stage_prog_data->binding_table.image_start = 0xd0d0d0d0;
1202 }
1203
1204 /* This may or may not be used depending on how the compile goes. */
1205 stage_prog_data->binding_table.pull_constants_start = next_binding_table_offset;
1206 next_binding_table_offset++;
1207
1208 /* Plane 0 is just the regular texture section */
1209 stage_prog_data->binding_table.plane_start[0] = stage_prog_data->binding_table.texture_start;
1210
1211 stage_prog_data->binding_table.plane_start[1] = next_binding_table_offset;
1212 next_binding_table_offset += num_textures;
1213
1214 stage_prog_data->binding_table.plane_start[2] = next_binding_table_offset;
1215 next_binding_table_offset += num_textures;
1216
1217 /* prog_data->base.binding_table.size will be set by brw_mark_surface_used. */
1218
1219 assert(next_binding_table_offset <= BRW_MAX_SURFACES);
1220 return next_binding_table_offset;
1221 }
1222
1223 static void
setup_vec4_uniform_value(const gl_constant_value ** params,const gl_constant_value * values,unsigned n)1224 setup_vec4_uniform_value(const gl_constant_value **params,
1225 const gl_constant_value *values,
1226 unsigned n)
1227 {
1228 static const gl_constant_value zero = { 0 };
1229
1230 for (unsigned i = 0; i < n; ++i)
1231 params[i] = &values[i];
1232
1233 for (unsigned i = n; i < 4; ++i)
1234 params[i] = &zero;
1235 }
1236
1237 void
brw_setup_image_uniform_values(gl_shader_stage stage,struct brw_stage_prog_data * stage_prog_data,unsigned param_start_index,const gl_uniform_storage * storage)1238 brw_setup_image_uniform_values(gl_shader_stage stage,
1239 struct brw_stage_prog_data *stage_prog_data,
1240 unsigned param_start_index,
1241 const gl_uniform_storage *storage)
1242 {
1243 const gl_constant_value **param =
1244 &stage_prog_data->param[param_start_index];
1245
1246 for (unsigned i = 0; i < MAX2(storage->array_elements, 1); i++) {
1247 const unsigned image_idx = storage->opaque[stage].index + i;
1248 const brw_image_param *image_param =
1249 &stage_prog_data->image_param[image_idx];
1250
1251 /* Upload the brw_image_param structure. The order is expected to match
1252 * the BRW_IMAGE_PARAM_*_OFFSET defines.
1253 */
1254 setup_vec4_uniform_value(param + BRW_IMAGE_PARAM_SURFACE_IDX_OFFSET,
1255 (const gl_constant_value *)&image_param->surface_idx, 1);
1256 setup_vec4_uniform_value(param + BRW_IMAGE_PARAM_OFFSET_OFFSET,
1257 (const gl_constant_value *)image_param->offset, 2);
1258 setup_vec4_uniform_value(param + BRW_IMAGE_PARAM_SIZE_OFFSET,
1259 (const gl_constant_value *)image_param->size, 3);
1260 setup_vec4_uniform_value(param + BRW_IMAGE_PARAM_STRIDE_OFFSET,
1261 (const gl_constant_value *)image_param->stride, 4);
1262 setup_vec4_uniform_value(param + BRW_IMAGE_PARAM_TILING_OFFSET,
1263 (const gl_constant_value *)image_param->tiling, 3);
1264 setup_vec4_uniform_value(param + BRW_IMAGE_PARAM_SWIZZLING_OFFSET,
1265 (const gl_constant_value *)image_param->swizzling, 2);
1266 param += BRW_IMAGE_PARAM_SIZE;
1267
1268 brw_mark_surface_used(
1269 stage_prog_data,
1270 stage_prog_data->binding_table.image_start + image_idx);
1271 }
1272 }
1273
1274 /**
1275 * Decide which set of clip planes should be used when clipping via
1276 * gl_Position or gl_ClipVertex.
1277 */
brw_select_clip_planes(struct gl_context * ctx)1278 gl_clip_plane *brw_select_clip_planes(struct gl_context *ctx)
1279 {
1280 if (ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX]) {
1281 /* There is currently a GLSL vertex shader, so clip according to GLSL
1282 * rules, which means compare gl_ClipVertex (or gl_Position, if
1283 * gl_ClipVertex wasn't assigned) against the eye-coordinate clip planes
1284 * that were stored in EyeUserPlane at the time the clip planes were
1285 * specified.
1286 */
1287 return ctx->Transform.EyeUserPlane;
1288 } else {
1289 /* Either we are using fixed function or an ARB vertex program. In
1290 * either case the clip planes are going to be compared against
1291 * gl_Position (which is in clip coordinates) so we have to clip using
1292 * _ClipUserPlane, which was transformed into clip coordinates by Mesa
1293 * core.
1294 */
1295 return ctx->Transform._ClipUserPlane;
1296 }
1297 }
1298
1299 extern "C" const unsigned *
brw_compile_tes(const struct brw_compiler * compiler,void * log_data,void * mem_ctx,const struct brw_tes_prog_key * key,const struct brw_vue_map * input_vue_map,struct brw_tes_prog_data * prog_data,const nir_shader * src_shader,struct gl_program * prog,int shader_time_index,unsigned * final_assembly_size,char ** error_str)1300 brw_compile_tes(const struct brw_compiler *compiler,
1301 void *log_data,
1302 void *mem_ctx,
1303 const struct brw_tes_prog_key *key,
1304 const struct brw_vue_map *input_vue_map,
1305 struct brw_tes_prog_data *prog_data,
1306 const nir_shader *src_shader,
1307 struct gl_program *prog,
1308 int shader_time_index,
1309 unsigned *final_assembly_size,
1310 char **error_str)
1311 {
1312 const struct gen_device_info *devinfo = compiler->devinfo;
1313 const bool is_scalar = compiler->scalar_stage[MESA_SHADER_TESS_EVAL];
1314
1315 nir_shader *nir = nir_shader_clone(mem_ctx, src_shader);
1316 nir->info->inputs_read = key->inputs_read;
1317 nir->info->patch_inputs_read = key->patch_inputs_read;
1318
1319 nir = brw_nir_apply_sampler_key(nir, compiler, &key->tex, is_scalar);
1320 brw_nir_lower_tes_inputs(nir, input_vue_map);
1321 brw_nir_lower_vue_outputs(nir, is_scalar);
1322 nir = brw_postprocess_nir(nir, compiler, is_scalar);
1323
1324 brw_compute_vue_map(devinfo, &prog_data->base.vue_map,
1325 nir->info->outputs_written,
1326 nir->info->separate_shader);
1327
1328 unsigned output_size_bytes = prog_data->base.vue_map.num_slots * 4 * 4;
1329
1330 assert(output_size_bytes >= 1);
1331 if (output_size_bytes > GEN7_MAX_DS_URB_ENTRY_SIZE_BYTES) {
1332 if (error_str)
1333 *error_str = ralloc_strdup(mem_ctx, "DS outputs exceed maximum size");
1334 return NULL;
1335 }
1336
1337 prog_data->base.clip_distance_mask =
1338 ((1 << nir->info->clip_distance_array_size) - 1);
1339 prog_data->base.cull_distance_mask =
1340 ((1 << nir->info->cull_distance_array_size) - 1) <<
1341 nir->info->clip_distance_array_size;
1342
1343 /* URB entry sizes are stored as a multiple of 64 bytes. */
1344 prog_data->base.urb_entry_size = ALIGN(output_size_bytes, 64) / 64;
1345 prog_data->base.urb_read_length = 0;
1346
1347 STATIC_ASSERT(BRW_TESS_PARTITIONING_INTEGER == TESS_SPACING_EQUAL - 1);
1348 STATIC_ASSERT(BRW_TESS_PARTITIONING_ODD_FRACTIONAL ==
1349 TESS_SPACING_FRACTIONAL_ODD - 1);
1350 STATIC_ASSERT(BRW_TESS_PARTITIONING_EVEN_FRACTIONAL ==
1351 TESS_SPACING_FRACTIONAL_EVEN - 1);
1352
1353 prog_data->partitioning =
1354 (enum brw_tess_partitioning) (nir->info->tess.spacing - 1);
1355
1356 switch (nir->info->tess.primitive_mode) {
1357 case GL_QUADS:
1358 prog_data->domain = BRW_TESS_DOMAIN_QUAD;
1359 break;
1360 case GL_TRIANGLES:
1361 prog_data->domain = BRW_TESS_DOMAIN_TRI;
1362 break;
1363 case GL_ISOLINES:
1364 prog_data->domain = BRW_TESS_DOMAIN_ISOLINE;
1365 break;
1366 default:
1367 unreachable("invalid domain shader primitive mode");
1368 }
1369
1370 if (nir->info->tess.point_mode) {
1371 prog_data->output_topology = BRW_TESS_OUTPUT_TOPOLOGY_POINT;
1372 } else if (nir->info->tess.primitive_mode == GL_ISOLINES) {
1373 prog_data->output_topology = BRW_TESS_OUTPUT_TOPOLOGY_LINE;
1374 } else {
1375 /* Hardware winding order is backwards from OpenGL */
1376 prog_data->output_topology =
1377 nir->info->tess.ccw ? BRW_TESS_OUTPUT_TOPOLOGY_TRI_CW
1378 : BRW_TESS_OUTPUT_TOPOLOGY_TRI_CCW;
1379 }
1380
1381 if (unlikely(INTEL_DEBUG & DEBUG_TES)) {
1382 fprintf(stderr, "TES Input ");
1383 brw_print_vue_map(stderr, input_vue_map);
1384 fprintf(stderr, "TES Output ");
1385 brw_print_vue_map(stderr, &prog_data->base.vue_map);
1386 }
1387
1388 if (is_scalar) {
1389 fs_visitor v(compiler, log_data, mem_ctx, (void *) key,
1390 &prog_data->base.base, NULL, nir, 8,
1391 shader_time_index, input_vue_map);
1392 if (!v.run_tes()) {
1393 if (error_str)
1394 *error_str = ralloc_strdup(mem_ctx, v.fail_msg);
1395 return NULL;
1396 }
1397
1398 prog_data->base.base.dispatch_grf_start_reg = v.payload.num_regs;
1399 prog_data->base.dispatch_mode = DISPATCH_MODE_SIMD8;
1400
1401 fs_generator g(compiler, log_data, mem_ctx, (void *) key,
1402 &prog_data->base.base, v.promoted_constants, false,
1403 MESA_SHADER_TESS_EVAL);
1404 if (unlikely(INTEL_DEBUG & DEBUG_TES)) {
1405 g.enable_debug(ralloc_asprintf(mem_ctx,
1406 "%s tessellation evaluation shader %s",
1407 nir->info->label ? nir->info->label
1408 : "unnamed",
1409 nir->info->name));
1410 }
1411
1412 g.generate_code(v.cfg, 8);
1413
1414 return g.get_assembly(final_assembly_size);
1415 } else {
1416 brw::vec4_tes_visitor v(compiler, log_data, key, prog_data,
1417 nir, mem_ctx, shader_time_index);
1418 if (!v.run()) {
1419 if (error_str)
1420 *error_str = ralloc_strdup(mem_ctx, v.fail_msg);
1421 return NULL;
1422 }
1423
1424 if (unlikely(INTEL_DEBUG & DEBUG_TES))
1425 v.dump_instructions();
1426
1427 return brw_vec4_generate_assembly(compiler, log_data, mem_ctx, nir,
1428 &prog_data->base, v.cfg,
1429 final_assembly_size);
1430 }
1431 }
1432