1 /*
2 * Copyright © 2010 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 #pragma once
25
26 #include <stdint.h>
27 #include "elk_cfg.h"
28 #include "elk_compiler.h"
29 #include "compiler/nir/nir.h"
30
31 #ifdef __cplusplus
32 #include "elk_ir_analysis.h"
33 #include "elk_ir_allocator.h"
34
35 enum instruction_scheduler_mode {
36 SCHEDULE_PRE,
37 SCHEDULE_PRE_NON_LIFO,
38 SCHEDULE_PRE_LIFO,
39 SCHEDULE_POST,
40 SCHEDULE_NONE,
41 };
42
43 #define UBO_START ((1 << 16) - 4)
44
45 struct elk_backend_shader {
46 protected:
47
48 elk_backend_shader(const struct elk_compiler *compiler,
49 const struct elk_compile_params *params,
50 const nir_shader *shader,
51 struct elk_stage_prog_data *stage_prog_data,
52 bool debug_enabled);
53
54 public:
55 virtual ~elk_backend_shader();
56
57 const struct elk_compiler *compiler;
58 void *log_data; /* Passed to compiler->*_log functions */
59
60 const struct intel_device_info * const devinfo;
61 const nir_shader *nir;
62 struct elk_stage_prog_data * const stage_prog_data;
63
64 /** ralloc context for temporary data used during compile */
65 void *mem_ctx;
66
67 /**
68 * List of either elk_fs_inst or vec4_instruction (inheriting from
69 * elk_backend_instruction)
70 */
71 exec_list instructions;
72
73 elk_cfg_t *cfg;
74 elk_analysis<elk::idom_tree, elk_backend_shader> idom_analysis;
75
76 gl_shader_stage stage;
77 bool debug_enabled;
78
79 elk::simple_allocator alloc;
80
81 virtual void dump_instruction_to_file(const elk_backend_instruction *inst, FILE *file) const = 0;
82 virtual void dump_instructions_to_file(FILE *file) const;
83
84 /* Convenience functions based on the above. */
85 void dump_instruction(const elk_backend_instruction *inst, FILE *file = stderr) const {
86 dump_instruction_to_file(inst, file);
87 }
88 void dump_instructions(const char *name = nullptr) const;
89
90 void calculate_cfg();
91
92 virtual void invalidate_analysis(elk::analysis_dependency_class c);
93 };
94
95 #else
96 struct elk_backend_shader;
97 #endif /* __cplusplus */
98
99 enum elk_reg_type elk_type_for_base_type(const struct glsl_type *type);
100 uint32_t elk_math_function(enum elk_opcode op);
101 const char *elk_instruction_name(const struct elk_isa_info *isa,
102 enum elk_opcode op);
103 bool elk_saturate_immediate(enum elk_reg_type type, struct elk_reg *reg);
104 bool elk_negate_immediate(enum elk_reg_type type, struct elk_reg *reg);
105 bool elk_abs_immediate(enum elk_reg_type type, struct elk_reg *reg);
106
107 bool elk_opt_predicated_break(struct elk_backend_shader *s);
108
109 #ifdef __cplusplus
110 extern "C" {
111 #endif
112
113 /* elk_fs_reg_allocate.cpp */
114 void elk_fs_alloc_reg_sets(struct elk_compiler *compiler);
115
116 /* elk_vec4_reg_allocate.cpp */
117 void elk_vec4_alloc_reg_set(struct elk_compiler *compiler);
118
119 /* elk_disasm.c */
120 extern const char *const elk_conditional_modifier[16];
121 extern const char *const elk_pred_ctrl_align16[16];
122
123 /* Per-thread scratch space is a power-of-two multiple of 1KB. */
124 static inline unsigned
elk_get_scratch_size(int size)125 elk_get_scratch_size(int size)
126 {
127 return MAX2(1024, util_next_power_of_two(size));
128 }
129
130
131 static inline nir_variable_mode
elk_nir_no_indirect_mask(const struct elk_compiler * compiler,gl_shader_stage stage)132 elk_nir_no_indirect_mask(const struct elk_compiler *compiler,
133 gl_shader_stage stage)
134 {
135 const struct intel_device_info *devinfo = compiler->devinfo;
136 const bool is_scalar = compiler->scalar_stage[stage];
137 nir_variable_mode indirect_mask = (nir_variable_mode) 0;
138
139 switch (stage) {
140 case MESA_SHADER_VERTEX:
141 case MESA_SHADER_FRAGMENT:
142 indirect_mask |= nir_var_shader_in;
143 break;
144
145 case MESA_SHADER_GEOMETRY:
146 if (!is_scalar)
147 indirect_mask |= nir_var_shader_in;
148 break;
149
150 default:
151 /* Everything else can handle indirect inputs */
152 break;
153 }
154
155 if (is_scalar && stage != MESA_SHADER_TESS_CTRL)
156 indirect_mask |= nir_var_shader_out;
157
158 /* On HSW+, we allow indirects in scalar shaders. They get implemented
159 * using nir_lower_vars_to_explicit_types and nir_lower_explicit_io in
160 * elk_postprocess_nir.
161 *
162 * We haven't plumbed through the indirect scratch messages on gfx6 or
163 * earlier so doing indirects via scratch doesn't work there. On gfx7 and
164 * earlier the scratch space size is limited to 12kB. If we allowed
165 * indirects as scratch all the time, we may easily exceed this limit
166 * without having any fallback.
167 */
168 if (is_scalar && devinfo->verx10 <= 70)
169 indirect_mask |= nir_var_function_temp;
170
171 return indirect_mask;
172 }
173
174 bool elk_texture_offset(const nir_tex_instr *tex, unsigned src,
175 uint32_t *offset_bits);
176
177 /**
178 * Scratch data used when compiling a GLSL geometry shader.
179 */
180 struct elk_gs_compile
181 {
182 struct elk_gs_prog_key key;
183 struct intel_vue_map input_vue_map;
184
185 unsigned control_data_bits_per_vertex;
186 unsigned control_data_header_size_bits;
187 };
188
189 #ifdef __cplusplus
190 }
191 #endif
192