Merge branch 'vulkan' into 'vulkan'
[mesa.git] / src / mesa / drivers / dri / i965 / brw_link.cpp
1 /*
2 * Copyright © 2015 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_shader.h"
26 #include "brw_fs.h"
27 #include "brw_nir.h"
28 #include "brw_program.h"
29 #include "glsl/ir_optimization.h"
30 #include "program/program.h"
31 #include "main/shaderapi.h"
32 #include "main/uniforms.h"
33
34 /**
35 * Performs a compile of the shader stages even when we don't know
36 * what non-orthogonal state will be set, in the hope that it reflects
37 * the eventual NOS used, and thus allows us to produce link failures.
38 */
39 static bool
40 brw_shader_precompile(struct gl_context *ctx,
41 struct gl_shader_program *sh_prog)
42 {
43 struct gl_shader *vs = sh_prog->_LinkedShaders[MESA_SHADER_VERTEX];
44 struct gl_shader *tcs = sh_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL];
45 struct gl_shader *tes = sh_prog->_LinkedShaders[MESA_SHADER_TESS_EVAL];
46 struct gl_shader *gs = sh_prog->_LinkedShaders[MESA_SHADER_GEOMETRY];
47 struct gl_shader *fs = sh_prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
48 struct gl_shader *cs = sh_prog->_LinkedShaders[MESA_SHADER_COMPUTE];
49
50 if (fs && !brw_fs_precompile(ctx, sh_prog, fs->Program))
51 return false;
52
53 if (gs && !brw_gs_precompile(ctx, sh_prog, gs->Program))
54 return false;
55
56 if (tes && !brw_tes_precompile(ctx, sh_prog, tes->Program))
57 return false;
58
59 if (tcs && !brw_tcs_precompile(ctx, sh_prog, tcs->Program))
60 return false;
61
62 if (vs && !brw_vs_precompile(ctx, sh_prog, vs->Program))
63 return false;
64
65 if (cs && !brw_cs_precompile(ctx, sh_prog, cs->Program))
66 return false;
67
68 return true;
69 }
70
71 static void
72 brw_lower_packing_builtins(struct brw_context *brw,
73 gl_shader_stage shader_type,
74 exec_list *ir)
75 {
76 /* Gens < 7 don't have instructions to convert to or from half-precision,
77 * and Gens < 6 don't expose that functionality.
78 */
79 if (brw->gen != 6)
80 return;
81
82 lower_packing_builtins(ir, LOWER_PACK_HALF_2x16 | LOWER_UNPACK_HALF_2x16);
83 }
84
85 static void
86 process_glsl_ir(gl_shader_stage stage,
87 struct brw_context *brw,
88 struct gl_shader_program *shader_prog,
89 struct gl_shader *shader)
90 {
91 struct gl_context *ctx = &brw->ctx;
92 const struct brw_compiler *compiler = brw->intelScreen->compiler;
93 const struct gl_shader_compiler_options *options =
94 &ctx->Const.ShaderCompilerOptions[shader->Stage];
95
96 /* Temporary memory context for any new IR. */
97 void *mem_ctx = ralloc_context(NULL);
98
99 ralloc_adopt(mem_ctx, shader->ir);
100
101 /* lower_packing_builtins() inserts arithmetic instructions, so it
102 * must precede lower_instructions().
103 */
104 brw_lower_packing_builtins(brw, shader->Stage, shader->ir);
105 do_mat_op_to_vec(shader->ir);
106 lower_instructions(shader->ir,
107 MOD_TO_FLOOR |
108 DIV_TO_MUL_RCP |
109 SUB_TO_ADD_NEG |
110 EXP_TO_EXP2 |
111 LOG_TO_LOG2 |
112 LDEXP_TO_ARITH |
113 CARRY_TO_ARITH |
114 BORROW_TO_ARITH);
115
116 /* Pre-gen6 HW can only nest if-statements 16 deep. Beyond this,
117 * if-statements need to be flattened.
118 */
119 if (brw->gen < 6)
120 lower_if_to_cond_assign(shader->ir, 16);
121
122 do_lower_texture_projection(shader->ir);
123 brw_lower_texture_gradients(brw, shader->ir);
124 do_vec_index_to_cond_assign(shader->ir);
125 lower_vector_insert(shader->ir, true);
126 lower_offset_arrays(shader->ir);
127 brw_do_lower_unnormalized_offset(shader->ir);
128 lower_noise(shader->ir);
129 lower_quadop_vector(shader->ir, false);
130
131 bool lowered_variable_indexing =
132 lower_variable_index_to_cond_assign((gl_shader_stage)stage,
133 shader->ir,
134 options->EmitNoIndirectInput,
135 options->EmitNoIndirectOutput,
136 options->EmitNoIndirectTemp,
137 options->EmitNoIndirectUniform);
138
139 if (unlikely(brw->perf_debug && lowered_variable_indexing)) {
140 perf_debug("Unsupported form of variable indexing in %s; falling "
141 "back to very inefficient code generation\n",
142 _mesa_shader_stage_to_abbrev(shader->Stage));
143 }
144
145 bool progress;
146 do {
147 progress = false;
148
149 if (compiler->scalar_stage[shader->Stage]) {
150 brw_do_channel_expressions(shader->ir);
151 brw_do_vector_splitting(shader->ir);
152 }
153
154 progress = do_lower_jumps(shader->ir, true, true,
155 true, /* main return */
156 false, /* continue */
157 false /* loops */
158 ) || progress;
159
160 progress = do_common_optimization(shader->ir, true, true,
161 options, ctx->Const.NativeIntegers) || progress;
162 } while (progress);
163
164 validate_ir_tree(shader->ir);
165
166 /* Now that we've finished altering the linked IR, reparent any live IR back
167 * to the permanent memory context, and free the temporary one (discarding any
168 * junk we optimized away).
169 */
170 reparent_ir(shader->ir, shader->ir);
171 ralloc_free(mem_ctx);
172
173 if (ctx->_Shader->Flags & GLSL_DUMP) {
174 fprintf(stderr, "\n");
175 fprintf(stderr, "GLSL IR for linked %s program %d:\n",
176 _mesa_shader_stage_to_string(shader->Stage),
177 shader_prog->Name);
178 _mesa_print_ir(stderr, shader->ir, NULL);
179 fprintf(stderr, "\n");
180 }
181 }
182
183 extern "C" GLboolean
184 brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg)
185 {
186 struct brw_context *brw = brw_context(ctx);
187 const struct brw_compiler *compiler = brw->intelScreen->compiler;
188 unsigned int stage;
189
190 for (stage = 0; stage < ARRAY_SIZE(shProg->_LinkedShaders); stage++) {
191 struct gl_shader *shader = shProg->_LinkedShaders[stage];
192 if (!shader)
193 continue;
194
195 struct gl_program *prog =
196 ctx->Driver.NewProgram(ctx, _mesa_shader_stage_to_program(stage),
197 shader->Name);
198 if (!prog)
199 return false;
200 prog->Parameters = _mesa_new_parameter_list();
201
202 _mesa_copy_linked_program_data((gl_shader_stage) stage, shProg, prog);
203
204 process_glsl_ir((gl_shader_stage) stage, brw, shProg, shader);
205
206 /* Make a pass over the IR to add state references for any built-in
207 * uniforms that are used. This has to be done now (during linking).
208 * Code generation doesn't happen until the first time this shader is
209 * used for rendering. Waiting until then to generate the parameters is
210 * too late. At that point, the values for the built-in uniforms won't
211 * get sent to the shader.
212 */
213 foreach_in_list(ir_instruction, node, shader->ir) {
214 ir_variable *var = node->as_variable();
215
216 if ((var == NULL) || (var->data.mode != ir_var_uniform)
217 || (strncmp(var->name, "gl_", 3) != 0))
218 continue;
219
220 const ir_state_slot *const slots = var->get_state_slots();
221 assert(slots != NULL);
222
223 for (unsigned int i = 0; i < var->get_num_state_slots(); i++) {
224 _mesa_add_state_reference(prog->Parameters,
225 (gl_state_index *) slots[i].tokens);
226 }
227 }
228
229 do_set_program_inouts(shader->ir, prog, shader->Stage);
230
231 prog->SamplersUsed = shader->active_samplers;
232 prog->ShadowSamplers = shader->shadow_samplers;
233 _mesa_update_shader_textures_used(shProg, prog);
234
235 _mesa_reference_program(ctx, &shader->Program, prog);
236
237 brw_add_texrect_params(prog);
238
239 prog->nir = brw_create_nir(brw, shProg, prog, (gl_shader_stage) stage,
240 compiler->scalar_stage[stage]);
241
242 _mesa_reference_program(ctx, &prog, NULL);
243 }
244
245 if ((ctx->_Shader->Flags & GLSL_DUMP) && shProg->Name != 0) {
246 for (unsigned i = 0; i < shProg->NumShaders; i++) {
247 const struct gl_shader *sh = shProg->Shaders[i];
248 if (!sh)
249 continue;
250
251 fprintf(stderr, "GLSL %s shader %d source for linked program %d:\n",
252 _mesa_shader_stage_to_string(sh->Stage),
253 i, shProg->Name);
254 fprintf(stderr, "%s", sh->Source);
255 fprintf(stderr, "\n");
256 }
257 }
258
259 if (brw->precompile && !brw_shader_precompile(ctx, shProg))
260 return false;
261
262 return true;
263 }