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
25 #include "main/macros.h"
26 #include "brw_context.h"
30 #include "glsl/ir_optimization.h"
31 #include "glsl/ir_print_visitor.h"
34 brw_new_shader(struct gl_context *ctx, GLuint name, GLuint type)
36 struct brw_shader *shader;
38 shader = rzalloc(NULL, struct brw_shader);
40 shader->base.Type = type;
41 shader->base.Name = name;
42 _mesa_init_shader(ctx, &shader->base);
48 struct gl_shader_program *
49 brw_new_shader_program(struct gl_context *ctx, GLuint name)
51 struct gl_shader_program *prog = rzalloc(NULL, struct gl_shader_program);
54 _mesa_init_shader_program(ctx, prog);
60 * Performs a compile of the shader stages even when we don't know
61 * what non-orthogonal state will be set, in the hope that it reflects
62 * the eventual NOS used, and thus allows us to produce link failures.
65 brw_shader_precompile(struct gl_context *ctx, struct gl_shader_program *prog)
67 struct brw_context *brw = brw_context(ctx);
69 if (brw->precompile && !brw_fs_precompile(ctx, prog))
72 if (brw->precompile && !brw_vs_precompile(ctx, prog))
79 brw_lower_packing_builtins(struct brw_context *brw,
80 gl_shader_type shader_type,
83 int ops = LOWER_PACK_SNORM_2x16
84 | LOWER_UNPACK_SNORM_2x16
85 | LOWER_PACK_UNORM_2x16
86 | LOWER_UNPACK_UNORM_2x16
87 | LOWER_PACK_SNORM_4x8
88 | LOWER_UNPACK_SNORM_4x8
89 | LOWER_PACK_UNORM_4x8
90 | LOWER_UNPACK_UNORM_4x8;
92 if (brw->intel.gen >= 7) {
93 /* Gen7 introduced the f32to16 and f16to32 instructions, which can be
94 * used to execute packHalf2x16 and unpackHalf2x16. For AOS code, no
95 * lowering is needed. For SOA code, the Half2x16 ops must be
98 if (shader_type == MESA_SHADER_FRAGMENT) {
99 ops |= LOWER_PACK_HALF_2x16_TO_SPLIT
100 | LOWER_UNPACK_HALF_2x16_TO_SPLIT;
103 ops |= LOWER_PACK_HALF_2x16
104 | LOWER_UNPACK_HALF_2x16;
107 lower_packing_builtins(ir, ops);
111 brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg)
113 struct brw_context *brw = brw_context(ctx);
114 struct intel_context *intel = &brw->intel;
116 static const char *target_strings[]
117 = { "vertex", "fragment", "geometry" };
119 for (stage = 0; stage < ARRAY_SIZE(shProg->_LinkedShaders); stage++) {
120 struct brw_shader *shader =
121 (struct brw_shader *)shProg->_LinkedShaders[stage];
122 static const GLenum targets[] = {
123 GL_VERTEX_PROGRAM_ARB,
124 GL_FRAGMENT_PROGRAM_ARB,
125 GL_GEOMETRY_PROGRAM_NV
131 struct gl_program *prog =
132 ctx->Driver.NewProgram(ctx, targets[stage], shader->base.Name);
135 prog->Parameters = _mesa_new_parameter_list();
138 struct gl_vertex_program *vp = (struct gl_vertex_program *) prog;
139 vp->UsesClipDistance = shProg->Vert.UsesClipDistance;
142 void *mem_ctx = ralloc_context(NULL);
146 ralloc_free(shader->ir);
147 shader->ir = new(shader) exec_list;
148 clone_ir_list(mem_ctx, shader->ir, shader->base.ir);
150 /* lower_packing_builtins() inserts arithmetic instructions, so it
151 * must precede lower_instructions().
153 brw_lower_packing_builtins(brw, (gl_shader_type) stage, shader->ir);
154 do_mat_op_to_vec(shader->ir);
155 const int lrp_to_arith = intel->gen < 6 ? LRP_TO_ARITH : 0;
156 lower_instructions(shader->ir,
164 /* Pre-gen6 HW can only nest if-statements 16 deep. Beyond this,
165 * if-statements need to be flattened.
168 lower_if_to_cond_assign(shader->ir, 16);
170 do_lower_texture_projection(shader->ir);
171 if (intel->gen < 8 && !intel->is_haswell)
172 brw_lower_texture_gradients(shader->ir);
173 do_vec_index_to_cond_assign(shader->ir);
174 brw_do_cubemap_normalize(shader->ir);
175 lower_noise(shader->ir);
176 lower_quadop_vector(shader->ir, false);
179 bool output = stage == MESA_SHADER_FRAGMENT;
180 bool temp = stage == MESA_SHADER_FRAGMENT;
181 bool uniform = false;
183 bool lowered_variable_indexing =
184 lower_variable_index_to_cond_assign(shader->ir,
185 input, output, temp, uniform);
187 if (unlikely((intel->perf_debug) && lowered_variable_indexing)) {
188 perf_debug("Unsupported form of variable indexing in FS; falling "
189 "back to very inefficient code generation\n");
192 /* FINISHME: Do this before the variable index lowering. */
193 lower_ubo_reference(&shader->base, shader->ir);
198 if (stage == MESA_SHADER_FRAGMENT) {
199 brw_do_channel_expressions(shader->ir);
200 brw_do_vector_splitting(shader->ir);
203 progress = do_lower_jumps(shader->ir, true, true,
204 true, /* main return */
205 false, /* continue */
209 progress = do_common_optimization(shader->ir, true, true, 32)
213 /* Make a pass over the IR to add state references for any built-in
214 * uniforms that are used. This has to be done now (during linking).
215 * Code generation doesn't happen until the first time this shader is
216 * used for rendering. Waiting until then to generate the parameters is
217 * too late. At that point, the values for the built-in uniforms won't
218 * get sent to the shader.
220 foreach_list(node, shader->ir) {
221 ir_variable *var = ((ir_instruction *) node)->as_variable();
223 if ((var == NULL) || (var->mode != ir_var_uniform)
224 || (strncmp(var->name, "gl_", 3) != 0))
227 const ir_state_slot *const slots = var->state_slots;
228 assert(var->state_slots != NULL);
230 for (unsigned int i = 0; i < var->num_state_slots; i++) {
231 _mesa_add_state_reference(prog->Parameters,
232 (gl_state_index *) slots[i].tokens);
236 validate_ir_tree(shader->ir);
238 reparent_ir(shader->ir, shader->ir);
239 ralloc_free(mem_ctx);
241 do_set_program_inouts(shader->ir, prog,
242 shader->base.Type == GL_FRAGMENT_SHADER);
244 prog->SamplersUsed = shader->base.active_samplers;
245 _mesa_update_shader_textures_used(shProg, prog);
247 _mesa_reference_program(ctx, &shader->base.Program, prog);
249 brw_add_texrect_params(prog);
251 /* This has to be done last. Any operation that can cause
252 * prog->ParameterValues to get reallocated (e.g., anything that adds a
253 * program constant) has to happen before creating this linkage.
255 _mesa_associate_uniform_storage(ctx, shProg, prog->Parameters);
257 _mesa_reference_program(ctx, &prog, NULL);
259 if (ctx->Shader.Flags & GLSL_DUMP) {
261 printf("GLSL IR for linked %s program %d:\n", target_strings[stage],
263 _mesa_print_ir(shader->base.ir, NULL);
268 if (ctx->Shader.Flags & GLSL_DUMP) {
269 for (unsigned i = 0; i < shProg->NumShaders; i++) {
270 const struct gl_shader *sh = shProg->Shaders[i];
274 printf("GLSL %s shader %d source for linked program %d:\n",
275 target_strings[_mesa_shader_type_to_index(sh->Type)],
278 printf("%s", sh->Source);
283 if (!brw_shader_precompile(ctx, shProg))
291 brw_type_for_base_type(const struct glsl_type *type)
293 switch (type->base_type) {
294 case GLSL_TYPE_FLOAT:
295 return BRW_REGISTER_TYPE_F;
298 return BRW_REGISTER_TYPE_D;
300 return BRW_REGISTER_TYPE_UD;
301 case GLSL_TYPE_ARRAY:
302 return brw_type_for_base_type(type->fields.array);
303 case GLSL_TYPE_STRUCT:
304 case GLSL_TYPE_SAMPLER:
305 /* These should be overridden with the type of the member when
306 * dereferenced into. BRW_REGISTER_TYPE_UD seems like a likely
307 * way to trip up if we don't.
309 return BRW_REGISTER_TYPE_UD;
311 case GLSL_TYPE_ERROR:
312 case GLSL_TYPE_INTERFACE:
313 assert(!"not reached");
317 return BRW_REGISTER_TYPE_F;
321 brw_conditional_for_comparison(unsigned int op)
325 return BRW_CONDITIONAL_L;
326 case ir_binop_greater:
327 return BRW_CONDITIONAL_G;
328 case ir_binop_lequal:
329 return BRW_CONDITIONAL_LE;
330 case ir_binop_gequal:
331 return BRW_CONDITIONAL_GE;
333 case ir_binop_all_equal: /* same as equal for scalars */
334 return BRW_CONDITIONAL_Z;
335 case ir_binop_nequal:
336 case ir_binop_any_nequal: /* same as nequal for scalars */
337 return BRW_CONDITIONAL_NZ;
339 assert(!"not reached: bad operation for comparison");
340 return BRW_CONDITIONAL_NZ;
345 brw_math_function(enum opcode op)
348 case SHADER_OPCODE_RCP:
349 return BRW_MATH_FUNCTION_INV;
350 case SHADER_OPCODE_RSQ:
351 return BRW_MATH_FUNCTION_RSQ;
352 case SHADER_OPCODE_SQRT:
353 return BRW_MATH_FUNCTION_SQRT;
354 case SHADER_OPCODE_EXP2:
355 return BRW_MATH_FUNCTION_EXP;
356 case SHADER_OPCODE_LOG2:
357 return BRW_MATH_FUNCTION_LOG;
358 case SHADER_OPCODE_POW:
359 return BRW_MATH_FUNCTION_POW;
360 case SHADER_OPCODE_SIN:
361 return BRW_MATH_FUNCTION_SIN;
362 case SHADER_OPCODE_COS:
363 return BRW_MATH_FUNCTION_COS;
364 case SHADER_OPCODE_INT_QUOTIENT:
365 return BRW_MATH_FUNCTION_INT_DIV_QUOTIENT;
366 case SHADER_OPCODE_INT_REMAINDER:
367 return BRW_MATH_FUNCTION_INT_DIV_REMAINDER;
369 assert(!"not reached: unknown math function");
375 brw_texture_offset(ir_constant *offset)
377 assert(offset != NULL);
379 signed char offsets[3];
380 for (unsigned i = 0; i < offset->type->vector_elements; i++)
381 offsets[i] = (signed char) offset->value.i[i];
383 /* Combine all three offsets into a single unsigned dword:
385 * bits 11:8 - U Offset (X component)
386 * bits 7:4 - V Offset (Y component)
387 * bits 3:0 - R Offset (Z component)
389 unsigned offset_bits = 0;
390 for (unsigned i = 0; i < offset->type->vector_elements; i++) {
391 const unsigned shift = 4 * (2 - i);
392 offset_bits |= (offsets[i] << shift) & (0xF << shift);
398 brw_instruction_name(enum opcode op)
402 if (op < ARRAY_SIZE(opcode_descs) && opcode_descs[op].name)
403 return opcode_descs[op].name;
406 case FS_OPCODE_FB_WRITE:
409 case SHADER_OPCODE_RCP:
411 case SHADER_OPCODE_RSQ:
413 case SHADER_OPCODE_SQRT:
415 case SHADER_OPCODE_EXP2:
417 case SHADER_OPCODE_LOG2:
419 case SHADER_OPCODE_POW:
421 case SHADER_OPCODE_INT_QUOTIENT:
423 case SHADER_OPCODE_INT_REMAINDER:
425 case SHADER_OPCODE_SIN:
427 case SHADER_OPCODE_COS:
430 case SHADER_OPCODE_TEX:
432 case SHADER_OPCODE_TXD:
434 case SHADER_OPCODE_TXF:
436 case SHADER_OPCODE_TXL:
438 case SHADER_OPCODE_TXS:
442 case SHADER_OPCODE_TXF_MS:
450 case FS_OPCODE_PIXEL_X:
452 case FS_OPCODE_PIXEL_Y:
455 case FS_OPCODE_CINTERP:
457 case FS_OPCODE_LINTERP:
460 case FS_OPCODE_SPILL:
462 case FS_OPCODE_UNSPILL:
465 case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD:
466 return "uniform_pull_const";
467 case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD_GEN7:
468 return "uniform_pull_const_gen7";
469 case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD:
470 return "varying_pull_const";
471 case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7:
472 return "varying_pull_const_gen7";
474 case FS_OPCODE_MOV_DISPATCH_TO_FLAGS:
475 return "mov_dispatch_to_flags";
476 case FS_OPCODE_DISCARD_JUMP:
477 return "discard_jump";
479 case FS_OPCODE_SET_SIMD4X2_OFFSET:
480 return "set_simd4x2_offset";
482 case FS_OPCODE_PACK_HALF_2x16_SPLIT:
483 return "pack_half_2x16_split";
484 case FS_OPCODE_UNPACK_HALF_2x16_SPLIT_X:
485 return "unpack_half_2x16_split_x";
486 case FS_OPCODE_UNPACK_HALF_2x16_SPLIT_Y:
487 return "unpack_half_2x16_split_y";
489 case FS_OPCODE_PLACEHOLDER_HALT:
490 return "placeholder_halt";
492 case VS_OPCODE_URB_WRITE:
494 case VS_OPCODE_SCRATCH_READ:
495 return "scratch_read";
496 case VS_OPCODE_SCRATCH_WRITE:
497 return "scratch_write";
498 case VS_OPCODE_PULL_CONSTANT_LOAD:
499 return "pull_constant_load";
500 case VS_OPCODE_PULL_CONSTANT_LOAD_GEN7:
501 return "pull_constant_load_gen7";
504 /* Yes, this leaks. It's in debug code, it should never occur, and if
505 * it does, you should just add the case to the list above.
507 asprintf(&fallback, "op%d", op);
513 backend_instruction::is_tex()
515 return (opcode == SHADER_OPCODE_TEX ||
516 opcode == FS_OPCODE_TXB ||
517 opcode == SHADER_OPCODE_TXD ||
518 opcode == SHADER_OPCODE_TXF ||
519 opcode == SHADER_OPCODE_TXF_MS ||
520 opcode == SHADER_OPCODE_TXL ||
521 opcode == SHADER_OPCODE_TXS ||
522 opcode == SHADER_OPCODE_LOD);
526 backend_instruction::is_math()
528 return (opcode == SHADER_OPCODE_RCP ||
529 opcode == SHADER_OPCODE_RSQ ||
530 opcode == SHADER_OPCODE_SQRT ||
531 opcode == SHADER_OPCODE_EXP2 ||
532 opcode == SHADER_OPCODE_LOG2 ||
533 opcode == SHADER_OPCODE_SIN ||
534 opcode == SHADER_OPCODE_COS ||
535 opcode == SHADER_OPCODE_INT_QUOTIENT ||
536 opcode == SHADER_OPCODE_INT_REMAINDER ||
537 opcode == SHADER_OPCODE_POW);
541 backend_instruction::is_control_flow()
545 case BRW_OPCODE_WHILE:
547 case BRW_OPCODE_ELSE:
548 case BRW_OPCODE_ENDIF:
549 case BRW_OPCODE_BREAK:
550 case BRW_OPCODE_CONTINUE: