• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright © 2015-2016 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_compiler.h"
25 #include "brw_context.h"
26 #include "compiler/nir/nir.h"
27 #include "main/errors.h"
28 #include "util/debug.h"
29 
30 #define COMMON_OPTIONS                                                        \
31    .lower_sub = true,                                                         \
32    .lower_fdiv = true,                                                        \
33    .lower_scmp = true,                                                        \
34    .lower_fmod32 = true,                                                      \
35    .lower_fmod64 = false,                                                     \
36    .lower_bitfield_extract = true,                                            \
37    .lower_bitfield_insert = true,                                             \
38    .lower_uadd_carry = true,                                                  \
39    .lower_usub_borrow = true,                                                 \
40    .lower_fdiv = true,                                                        \
41    .lower_flrp64 = true,                                                      \
42    .native_integers = true,                                                   \
43    .use_interpolated_input_intrinsics = true,                                 \
44    .vertex_id_zero_based = true
45 
46 static const struct nir_shader_compiler_options scalar_nir_options = {
47    COMMON_OPTIONS,
48    .lower_pack_half_2x16 = true,
49    .lower_pack_snorm_2x16 = true,
50    .lower_pack_snorm_4x8 = true,
51    .lower_pack_unorm_2x16 = true,
52    .lower_pack_unorm_4x8 = true,
53    .lower_unpack_half_2x16 = true,
54    .lower_unpack_snorm_2x16 = true,
55    .lower_unpack_snorm_4x8 = true,
56    .lower_unpack_unorm_2x16 = true,
57    .lower_unpack_unorm_4x8 = true,
58    .max_unroll_iterations = 32,
59 };
60 
61 static const struct nir_shader_compiler_options vector_nir_options = {
62    COMMON_OPTIONS,
63 
64    /* In the vec4 backend, our dpN instruction replicates its result to all the
65     * components of a vec4.  We would like NIR to give us replicated fdot
66     * instructions because it can optimize better for us.
67     */
68    .fdot_replicates = true,
69 
70    /* Prior to Gen6, there are no three source operations for SIMD4x2. */
71    .lower_flrp32 = true,
72 
73    .lower_pack_snorm_2x16 = true,
74    .lower_pack_unorm_2x16 = true,
75    .lower_unpack_snorm_2x16 = true,
76    .lower_unpack_unorm_2x16 = true,
77    .lower_extract_byte = true,
78    .lower_extract_word = true,
79    .max_unroll_iterations = 32,
80 };
81 
82 static const struct nir_shader_compiler_options vector_nir_options_gen6 = {
83    COMMON_OPTIONS,
84 
85    /* In the vec4 backend, our dpN instruction replicates its result to all the
86     * components of a vec4.  We would like NIR to give us replicated fdot
87     * instructions because it can optimize better for us.
88     */
89    .fdot_replicates = true,
90 
91    .lower_pack_snorm_2x16 = true,
92    .lower_pack_unorm_2x16 = true,
93    .lower_unpack_snorm_2x16 = true,
94    .lower_unpack_unorm_2x16 = true,
95    .lower_extract_byte = true,
96    .lower_extract_word = true,
97    .max_unroll_iterations = 32,
98 };
99 
100 struct brw_compiler *
brw_compiler_create(void * mem_ctx,const struct gen_device_info * devinfo)101 brw_compiler_create(void *mem_ctx, const struct gen_device_info *devinfo)
102 {
103    struct brw_compiler *compiler = rzalloc(mem_ctx, struct brw_compiler);
104 
105    compiler->devinfo = devinfo;
106 
107    brw_fs_alloc_reg_sets(compiler);
108    brw_vec4_alloc_reg_set(compiler);
109 
110    compiler->precise_trig = env_var_as_boolean("INTEL_PRECISE_TRIG", false);
111 
112    compiler->scalar_stage[MESA_SHADER_VERTEX] =
113       devinfo->gen >= 8 && !(INTEL_DEBUG & DEBUG_VEC4VS);
114    compiler->scalar_stage[MESA_SHADER_TESS_CTRL] =
115       devinfo->gen >= 8 && env_var_as_boolean("INTEL_SCALAR_TCS", true);
116    compiler->scalar_stage[MESA_SHADER_TESS_EVAL] =
117       devinfo->gen >= 8 && env_var_as_boolean("INTEL_SCALAR_TES", true);
118    compiler->scalar_stage[MESA_SHADER_GEOMETRY] =
119       devinfo->gen >= 8 && env_var_as_boolean("INTEL_SCALAR_GS", true);
120    compiler->scalar_stage[MESA_SHADER_FRAGMENT] = true;
121    compiler->scalar_stage[MESA_SHADER_COMPUTE] = true;
122 
123    /* We want the GLSL compiler to emit code that uses condition codes */
124    for (int i = 0; i < MESA_SHADER_STAGES; i++) {
125       compiler->glsl_compiler_options[i].MaxUnrollIterations = 0;
126       compiler->glsl_compiler_options[i].MaxIfDepth =
127          devinfo->gen < 6 ? 16 : UINT_MAX;
128 
129       compiler->glsl_compiler_options[i].EmitNoIndirectInput = true;
130       compiler->glsl_compiler_options[i].EmitNoIndirectUniform = false;
131 
132       bool is_scalar = compiler->scalar_stage[i];
133 
134       compiler->glsl_compiler_options[i].EmitNoIndirectOutput = is_scalar;
135       compiler->glsl_compiler_options[i].EmitNoIndirectTemp = is_scalar;
136       compiler->glsl_compiler_options[i].OptimizeForAOS = !is_scalar;
137 
138       if (is_scalar) {
139          compiler->glsl_compiler_options[i].NirOptions = &scalar_nir_options;
140       } else {
141          compiler->glsl_compiler_options[i].NirOptions =
142             devinfo->gen < 6 ? &vector_nir_options : &vector_nir_options_gen6;
143       }
144 
145       compiler->glsl_compiler_options[i].LowerBufferInterfaceBlocks = true;
146       compiler->glsl_compiler_options[i].ClampBlockIndicesToArrayBounds = true;
147    }
148 
149    compiler->glsl_compiler_options[MESA_SHADER_TESS_CTRL].EmitNoIndirectInput = false;
150    compiler->glsl_compiler_options[MESA_SHADER_TESS_EVAL].EmitNoIndirectInput = false;
151    compiler->glsl_compiler_options[MESA_SHADER_TESS_CTRL].EmitNoIndirectOutput = false;
152 
153    if (compiler->scalar_stage[MESA_SHADER_GEOMETRY])
154       compiler->glsl_compiler_options[MESA_SHADER_GEOMETRY].EmitNoIndirectInput = false;
155 
156    return compiler;
157 }
158