2 * Copyright © 2015 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
24 #include "brw_context.h"
25 #include "brw_shader.h"
28 #include "brw_program.h"
29 #include "glsl/ir_optimization.h"
30 #include "glsl/glsl_parser_extras.h"
31 #include "program/program.h"
32 #include "main/shaderapi.h"
33 #include "main/uniforms.h"
36 * Performs a compile of the shader stages even when we don't know
37 * what non-orthogonal state will be set, in the hope that it reflects
38 * the eventual NOS used, and thus allows us to produce link failures.
41 brw_shader_precompile(struct gl_context
*ctx
,
42 struct gl_shader_program
*sh_prog
)
44 struct gl_shader
*vs
= sh_prog
->_LinkedShaders
[MESA_SHADER_VERTEX
];
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
];
50 if (fs
&& !brw_fs_precompile(ctx
, sh_prog
, fs
->Program
))
53 if (gs
&& !brw_gs_precompile(ctx
, sh_prog
, gs
->Program
))
56 if (tes
&& !brw_tes_precompile(ctx
, sh_prog
, tes
->Program
))
59 if (vs
&& !brw_vs_precompile(ctx
, sh_prog
, vs
->Program
))
62 if (cs
&& !brw_cs_precompile(ctx
, sh_prog
, cs
->Program
))
69 brw_lower_packing_builtins(struct brw_context
*brw
,
70 gl_shader_stage shader_type
,
73 const struct brw_compiler
*compiler
= brw
->intelScreen
->compiler
;
75 int ops
= LOWER_PACK_SNORM_2x16
76 | LOWER_UNPACK_SNORM_2x16
77 | LOWER_PACK_UNORM_2x16
78 | LOWER_UNPACK_UNORM_2x16
;
80 if (compiler
->scalar_stage
[shader_type
]) {
81 ops
|= LOWER_UNPACK_UNORM_4x8
82 | LOWER_UNPACK_SNORM_4x8
83 | LOWER_PACK_UNORM_4x8
84 | LOWER_PACK_SNORM_4x8
;
88 /* Gen7 introduced the f32to16 and f16to32 instructions, which can be
89 * used to execute packHalf2x16 and unpackHalf2x16. For AOS code, no
90 * lowering is needed. For SOA code, the Half2x16 ops must be
93 if (compiler
->scalar_stage
[shader_type
]) {
94 ops
|= LOWER_PACK_HALF_2x16_TO_SPLIT
95 | LOWER_UNPACK_HALF_2x16_TO_SPLIT
;
98 ops
|= LOWER_PACK_HALF_2x16
99 | LOWER_UNPACK_HALF_2x16
;
102 lower_packing_builtins(ir
, ops
);
106 process_glsl_ir(gl_shader_stage stage
,
107 struct brw_context
*brw
,
108 struct gl_shader_program
*shader_prog
,
109 struct gl_shader
*shader
)
111 struct gl_context
*ctx
= &brw
->ctx
;
112 const struct brw_compiler
*compiler
= brw
->intelScreen
->compiler
;
113 const struct gl_shader_compiler_options
*options
=
114 &ctx
->Const
.ShaderCompilerOptions
[shader
->Stage
];
116 /* Temporary memory context for any new IR. */
117 void *mem_ctx
= ralloc_context(NULL
);
119 ralloc_adopt(mem_ctx
, shader
->ir
);
121 /* lower_packing_builtins() inserts arithmetic instructions, so it
122 * must precede lower_instructions().
124 brw_lower_packing_builtins(brw
, shader
->Stage
, shader
->ir
);
125 do_mat_op_to_vec(shader
->ir
);
126 const int bitfield_insert
= brw
->gen
>= 7 ? BITFIELD_INSERT_TO_BFM_BFI
: 0;
127 lower_instructions(shader
->ir
,
138 /* Pre-gen6 HW can only nest if-statements 16 deep. Beyond this,
139 * if-statements need to be flattened.
142 lower_if_to_cond_assign(shader
->ir
, 16);
144 do_lower_texture_projection(shader
->ir
);
145 brw_lower_texture_gradients(brw
, shader
->ir
);
146 do_vec_index_to_cond_assign(shader
->ir
);
147 lower_vector_insert(shader
->ir
, true);
148 lower_offset_arrays(shader
->ir
);
149 brw_do_lower_unnormalized_offset(shader
->ir
);
150 lower_noise(shader
->ir
);
151 lower_quadop_vector(shader
->ir
, false);
153 bool lowered_variable_indexing
=
154 lower_variable_index_to_cond_assign((gl_shader_stage
)stage
,
156 options
->EmitNoIndirectInput
,
157 options
->EmitNoIndirectOutput
,
158 options
->EmitNoIndirectTemp
,
159 options
->EmitNoIndirectUniform
);
161 if (unlikely(brw
->perf_debug
&& lowered_variable_indexing
)) {
162 perf_debug("Unsupported form of variable indexing in %s; falling "
163 "back to very inefficient code generation\n",
164 _mesa_shader_stage_to_abbrev(shader
->Stage
));
171 if (compiler
->scalar_stage
[shader
->Stage
]) {
172 brw_do_channel_expressions(shader
->ir
);
173 brw_do_vector_splitting(shader
->ir
);
176 progress
= do_lower_jumps(shader
->ir
, true, true,
177 true, /* main return */
178 false, /* continue */
182 progress
= do_common_optimization(shader
->ir
, true, true,
183 options
, ctx
->Const
.NativeIntegers
) || progress
;
186 validate_ir_tree(shader
->ir
);
188 /* Now that we've finished altering the linked IR, reparent any live IR back
189 * to the permanent memory context, and free the temporary one (discarding any
190 * junk we optimized away).
192 reparent_ir(shader
->ir
, shader
->ir
);
193 ralloc_free(mem_ctx
);
195 if (ctx
->_Shader
->Flags
& GLSL_DUMP
) {
196 fprintf(stderr
, "\n");
197 fprintf(stderr
, "GLSL IR for linked %s program %d:\n",
198 _mesa_shader_stage_to_string(shader
->Stage
),
200 _mesa_print_ir(stderr
, shader
->ir
, NULL
);
201 fprintf(stderr
, "\n");
206 brw_link_shader(struct gl_context
*ctx
, struct gl_shader_program
*shProg
)
208 struct brw_context
*brw
= brw_context(ctx
);
209 const struct brw_compiler
*compiler
= brw
->intelScreen
->compiler
;
212 for (stage
= 0; stage
< ARRAY_SIZE(shProg
->_LinkedShaders
); stage
++) {
213 struct gl_shader
*shader
= shProg
->_LinkedShaders
[stage
];
217 struct gl_program
*prog
=
218 ctx
->Driver
.NewProgram(ctx
, _mesa_shader_stage_to_program(stage
),
222 prog
->Parameters
= _mesa_new_parameter_list();
224 _mesa_copy_linked_program_data((gl_shader_stage
) stage
, shProg
, prog
);
226 process_glsl_ir((gl_shader_stage
) stage
, brw
, shProg
, shader
);
228 /* Make a pass over the IR to add state references for any built-in
229 * uniforms that are used. This has to be done now (during linking).
230 * Code generation doesn't happen until the first time this shader is
231 * used for rendering. Waiting until then to generate the parameters is
232 * too late. At that point, the values for the built-in uniforms won't
233 * get sent to the shader.
235 foreach_in_list(ir_instruction
, node
, shader
->ir
) {
236 ir_variable
*var
= node
->as_variable();
238 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
)
239 || (strncmp(var
->name
, "gl_", 3) != 0))
242 const ir_state_slot
*const slots
= var
->get_state_slots();
243 assert(slots
!= NULL
);
245 for (unsigned int i
= 0; i
< var
->get_num_state_slots(); i
++) {
246 _mesa_add_state_reference(prog
->Parameters
,
247 (gl_state_index
*) slots
[i
].tokens
);
251 do_set_program_inouts(shader
->ir
, prog
, shader
->Stage
);
253 prog
->SamplersUsed
= shader
->active_samplers
;
254 prog
->ShadowSamplers
= shader
->shadow_samplers
;
255 _mesa_update_shader_textures_used(shProg
, prog
);
257 _mesa_reference_program(ctx
, &shader
->Program
, prog
);
259 brw_add_texrect_params(prog
);
261 prog
->nir
= brw_create_nir(brw
, shProg
, prog
, (gl_shader_stage
) stage
,
262 compiler
->scalar_stage
[stage
]);
264 _mesa_reference_program(ctx
, &prog
, NULL
);
267 if ((ctx
->_Shader
->Flags
& GLSL_DUMP
) && shProg
->Name
!= 0) {
268 for (unsigned i
= 0; i
< shProg
->NumShaders
; i
++) {
269 const struct gl_shader
*sh
= shProg
->Shaders
[i
];
273 fprintf(stderr
, "GLSL %s shader %d source for linked program %d:\n",
274 _mesa_shader_stage_to_string(sh
->Stage
),
276 fprintf(stderr
, "%s", sh
->Source
);
277 fprintf(stderr
, "\n");
281 if (brw
->precompile
&& !brw_shader_precompile(ctx
, shProg
))