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
*tcs
= sh_prog
->_LinkedShaders
[MESA_SHADER_TESS_CTRL
];
46 struct gl_shader
*tes
= sh_prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
47 struct gl_shader
*gs
= sh_prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
];
48 struct gl_shader
*fs
= sh_prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
49 struct gl_shader
*cs
= sh_prog
->_LinkedShaders
[MESA_SHADER_COMPUTE
];
51 if (fs
&& !brw_fs_precompile(ctx
, sh_prog
, fs
->Program
))
54 if (gs
&& !brw_gs_precompile(ctx
, sh_prog
, gs
->Program
))
57 if (tes
&& !brw_tes_precompile(ctx
, sh_prog
, tes
->Program
))
60 if (tcs
&& !brw_tcs_precompile(ctx
, sh_prog
, tcs
->Program
))
63 if (vs
&& !brw_vs_precompile(ctx
, sh_prog
, vs
->Program
))
66 if (cs
&& !brw_cs_precompile(ctx
, sh_prog
, cs
->Program
))
73 brw_lower_packing_builtins(struct brw_context
*brw
,
74 gl_shader_stage shader_type
,
77 const struct brw_compiler
*compiler
= brw
->intelScreen
->compiler
;
79 int ops
= LOWER_PACK_SNORM_2x16
80 | LOWER_UNPACK_SNORM_2x16
81 | LOWER_PACK_UNORM_2x16
82 | LOWER_UNPACK_UNORM_2x16
;
84 if (compiler
->scalar_stage
[shader_type
]) {
85 ops
|= LOWER_UNPACK_UNORM_4x8
86 | LOWER_UNPACK_SNORM_4x8
87 | LOWER_PACK_UNORM_4x8
88 | LOWER_PACK_SNORM_4x8
;
92 /* Gen7 introduced the f32to16 and f16to32 instructions, which can be
93 * used to execute packHalf2x16 and unpackHalf2x16. For AOS code, no
94 * lowering is needed. For SOA code, the Half2x16 ops must be
97 if (compiler
->scalar_stage
[shader_type
]) {
98 ops
|= LOWER_PACK_HALF_2x16_TO_SPLIT
99 | LOWER_UNPACK_HALF_2x16_TO_SPLIT
;
102 ops
|= LOWER_PACK_HALF_2x16
103 | LOWER_UNPACK_HALF_2x16
;
106 lower_packing_builtins(ir
, ops
);
110 process_glsl_ir(gl_shader_stage stage
,
111 struct brw_context
*brw
,
112 struct gl_shader_program
*shader_prog
,
113 struct gl_shader
*shader
)
115 struct gl_context
*ctx
= &brw
->ctx
;
116 const struct brw_compiler
*compiler
= brw
->intelScreen
->compiler
;
117 const struct gl_shader_compiler_options
*options
=
118 &ctx
->Const
.ShaderCompilerOptions
[shader
->Stage
];
120 /* Temporary memory context for any new IR. */
121 void *mem_ctx
= ralloc_context(NULL
);
123 ralloc_adopt(mem_ctx
, shader
->ir
);
125 /* lower_packing_builtins() inserts arithmetic instructions, so it
126 * must precede lower_instructions().
128 brw_lower_packing_builtins(brw
, shader
->Stage
, shader
->ir
);
129 do_mat_op_to_vec(shader
->ir
);
130 const int bitfield_insert
= brw
->gen
>= 7 ? BITFIELD_INSERT_TO_BFM_BFI
: 0;
131 lower_instructions(shader
->ir
,
142 /* Pre-gen6 HW can only nest if-statements 16 deep. Beyond this,
143 * if-statements need to be flattened.
146 lower_if_to_cond_assign(shader
->ir
, 16);
148 do_lower_texture_projection(shader
->ir
);
149 brw_lower_texture_gradients(brw
, shader
->ir
);
150 do_vec_index_to_cond_assign(shader
->ir
);
151 lower_vector_insert(shader
->ir
, true);
152 lower_offset_arrays(shader
->ir
);
153 brw_do_lower_unnormalized_offset(shader
->ir
);
154 lower_noise(shader
->ir
);
155 lower_quadop_vector(shader
->ir
, false);
157 bool lowered_variable_indexing
=
158 lower_variable_index_to_cond_assign((gl_shader_stage
)stage
,
160 options
->EmitNoIndirectInput
,
161 options
->EmitNoIndirectOutput
,
162 options
->EmitNoIndirectTemp
,
163 options
->EmitNoIndirectUniform
);
165 if (unlikely(brw
->perf_debug
&& lowered_variable_indexing
)) {
166 perf_debug("Unsupported form of variable indexing in %s; falling "
167 "back to very inefficient code generation\n",
168 _mesa_shader_stage_to_abbrev(shader
->Stage
));
175 if (compiler
->scalar_stage
[shader
->Stage
]) {
176 brw_do_channel_expressions(shader
->ir
);
177 brw_do_vector_splitting(shader
->ir
);
180 progress
= do_lower_jumps(shader
->ir
, true, true,
181 true, /* main return */
182 false, /* continue */
186 progress
= do_common_optimization(shader
->ir
, true, true,
187 options
, ctx
->Const
.NativeIntegers
) || progress
;
190 validate_ir_tree(shader
->ir
);
192 /* Now that we've finished altering the linked IR, reparent any live IR back
193 * to the permanent memory context, and free the temporary one (discarding any
194 * junk we optimized away).
196 reparent_ir(shader
->ir
, shader
->ir
);
197 ralloc_free(mem_ctx
);
199 if (ctx
->_Shader
->Flags
& GLSL_DUMP
) {
200 fprintf(stderr
, "\n");
201 fprintf(stderr
, "GLSL IR for linked %s program %d:\n",
202 _mesa_shader_stage_to_string(shader
->Stage
),
204 _mesa_print_ir(stderr
, shader
->ir
, NULL
);
205 fprintf(stderr
, "\n");
210 brw_link_shader(struct gl_context
*ctx
, struct gl_shader_program
*shProg
)
212 struct brw_context
*brw
= brw_context(ctx
);
213 const struct brw_compiler
*compiler
= brw
->intelScreen
->compiler
;
216 for (stage
= 0; stage
< ARRAY_SIZE(shProg
->_LinkedShaders
); stage
++) {
217 struct gl_shader
*shader
= shProg
->_LinkedShaders
[stage
];
221 struct gl_program
*prog
=
222 ctx
->Driver
.NewProgram(ctx
, _mesa_shader_stage_to_program(stage
),
226 prog
->Parameters
= _mesa_new_parameter_list();
228 _mesa_copy_linked_program_data((gl_shader_stage
) stage
, shProg
, prog
);
230 process_glsl_ir((gl_shader_stage
) stage
, brw
, shProg
, shader
);
232 /* Make a pass over the IR to add state references for any built-in
233 * uniforms that are used. This has to be done now (during linking).
234 * Code generation doesn't happen until the first time this shader is
235 * used for rendering. Waiting until then to generate the parameters is
236 * too late. At that point, the values for the built-in uniforms won't
237 * get sent to the shader.
239 foreach_in_list(ir_instruction
, node
, shader
->ir
) {
240 ir_variable
*var
= node
->as_variable();
242 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
)
243 || (strncmp(var
->name
, "gl_", 3) != 0))
246 const ir_state_slot
*const slots
= var
->get_state_slots();
247 assert(slots
!= NULL
);
249 for (unsigned int i
= 0; i
< var
->get_num_state_slots(); i
++) {
250 _mesa_add_state_reference(prog
->Parameters
,
251 (gl_state_index
*) slots
[i
].tokens
);
255 do_set_program_inouts(shader
->ir
, prog
, shader
->Stage
);
257 prog
->SamplersUsed
= shader
->active_samplers
;
258 prog
->ShadowSamplers
= shader
->shadow_samplers
;
259 _mesa_update_shader_textures_used(shProg
, prog
);
261 _mesa_reference_program(ctx
, &shader
->Program
, prog
);
263 brw_add_texrect_params(prog
);
265 prog
->nir
= brw_create_nir(brw
, shProg
, prog
, (gl_shader_stage
) stage
,
266 compiler
->scalar_stage
[stage
]);
268 _mesa_reference_program(ctx
, &prog
, NULL
);
271 if ((ctx
->_Shader
->Flags
& GLSL_DUMP
) && shProg
->Name
!= 0) {
272 for (unsigned i
= 0; i
< shProg
->NumShaders
; i
++) {
273 const struct gl_shader
*sh
= shProg
->Shaders
[i
];
277 fprintf(stderr
, "GLSL %s shader %d source for linked program %d:\n",
278 _mesa_shader_stage_to_string(sh
->Stage
),
280 fprintf(stderr
, "%s", sh
->Source
);
281 fprintf(stderr
, "\n");
285 if (brw
->precompile
&& !brw_shader_precompile(ctx
, shProg
))