2 * Copyright © 2019 Google LLC
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
21 * DEALINGS IN THE SOFTWARE.
24 #include "tu_private.h"
26 #include "spirv/nir_spirv.h"
27 #include "util/mesa-sha1.h"
29 #include "ir3/ir3_nir.h"
32 tu_spirv_to_nir(struct ir3_compiler
*compiler
,
33 const uint32_t *words
,
35 gl_shader_stage stage
,
36 const char *entry_point_name
,
37 const VkSpecializationInfo
*spec_info
)
39 /* TODO these are made-up */
40 const struct spirv_to_nir_options spirv_options
= {
41 .lower_workgroup_access_to_offsets
= true,
42 .lower_ubo_ssbo_access_to_offsets
= true,
45 const nir_shader_compiler_options
*nir_options
=
46 ir3_get_compiler_options(compiler
);
48 /* convert VkSpecializationInfo */
49 struct nir_spirv_specialization
*spec
= NULL
;
50 uint32_t num_spec
= 0;
51 if (spec_info
&& spec_info
->mapEntryCount
) {
52 spec
= malloc(sizeof(*spec
) * spec_info
->mapEntryCount
);
56 for (uint32_t i
= 0; i
< spec_info
->mapEntryCount
; i
++) {
57 const VkSpecializationMapEntry
*entry
= &spec_info
->pMapEntries
[i
];
58 const void *data
= spec_info
->pData
+ entry
->offset
;
59 assert(data
+ entry
->size
<= spec_info
->pData
+ spec_info
->dataSize
);
60 spec
[i
].id
= entry
->constantID
;
62 spec
[i
].data64
= *(const uint64_t *) data
;
64 spec
[i
].data32
= *(const uint32_t *) data
;
65 spec
[i
].defined_on_module
= false;
68 num_spec
= spec_info
->mapEntryCount
;
72 spirv_to_nir(words
, word_count
, spec
, num_spec
, stage
, entry_point_name
,
73 &spirv_options
, nir_options
);
77 assert(nir
->info
.stage
== stage
);
78 nir_validate_shader(nir
, "after spirv_to_nir");
84 tu_sort_variables_by_location(struct exec_list
*variables
)
86 struct exec_list sorted
;
87 exec_list_make_empty(&sorted
);
89 nir_foreach_variable_safe(var
, variables
)
91 exec_node_remove(&var
->node
);
93 /* insert the variable into the sorted list */
94 nir_variable
*next
= NULL
;
95 nir_foreach_variable(tmp
, &sorted
)
97 if (var
->data
.location
< tmp
->data
.location
) {
103 exec_node_insert_node_before(&next
->node
, &var
->node
);
105 exec_list_push_tail(&sorted
, &var
->node
);
108 exec_list_move_nodes_to(&sorted
, variables
);
112 tu_shader_create(struct tu_device
*dev
,
113 gl_shader_stage stage
,
114 const VkPipelineShaderStageCreateInfo
*stage_info
,
115 const VkAllocationCallbacks
*alloc
)
117 const struct tu_shader_module
*module
=
118 tu_shader_module_from_handle(stage_info
->module
);
119 struct tu_shader
*shader
;
121 const uint32_t max_variant_count
= (stage
== MESA_SHADER_VERTEX
) ? 2 : 1;
124 sizeof(*shader
) + sizeof(struct ir3_shader_variant
) * max_variant_count
,
125 8, VK_SYSTEM_ALLOCATION_SCOPE_COMMAND
);
129 /* translate SPIR-V to NIR */
130 assert(module
->code_size
% 4 == 0);
131 nir_shader
*nir
= tu_spirv_to_nir(
132 dev
->compiler
, (const uint32_t *) module
->code
, module
->code_size
/ 4,
133 stage
, stage_info
->pName
, stage_info
->pSpecializationInfo
);
135 vk_free2(&dev
->alloc
, alloc
, shader
);
139 if (unlikely(dev
->physical_device
->instance
->debug_flags
& TU_DEBUG_NIR
)) {
140 fprintf(stderr
, "translated nir:\n");
141 nir_print_shader(nir
, stderr
);
144 /* TODO what needs to happen? */
147 case MESA_SHADER_VERTEX
:
148 tu_sort_variables_by_location(&nir
->outputs
);
150 case MESA_SHADER_TESS_CTRL
:
151 case MESA_SHADER_TESS_EVAL
:
152 case MESA_SHADER_GEOMETRY
:
153 tu_sort_variables_by_location(&nir
->inputs
);
154 tu_sort_variables_by_location(&nir
->outputs
);
156 case MESA_SHADER_FRAGMENT
:
157 tu_sort_variables_by_location(&nir
->inputs
);
159 case MESA_SHADER_COMPUTE
:
162 unreachable("invalid gl_shader_stage");
166 nir_assign_var_locations(&nir
->inputs
, &nir
->num_inputs
,
168 nir_assign_var_locations(&nir
->outputs
, &nir
->num_outputs
,
170 nir_assign_var_locations(&nir
->uniforms
, &nir
->num_uniforms
,
173 NIR_PASS_V(nir
, nir_lower_system_values
);
174 NIR_PASS_V(nir
, nir_lower_frexp
);
175 NIR_PASS_V(nir
, nir_lower_io
, nir_var_all
, ir3_glsl_type_size
, 0);
177 nir_shader_gather_info(nir
, nir_shader_get_entrypoint(nir
));
179 shader
->ir3_shader
.compiler
= dev
->compiler
;
180 shader
->ir3_shader
.type
= stage
;
181 shader
->ir3_shader
.nir
= nir
;
187 tu_shader_destroy(struct tu_device
*dev
,
188 struct tu_shader
*shader
,
189 const VkAllocationCallbacks
*alloc
)
191 if (shader
->ir3_shader
.nir
)
192 ralloc_free(shader
->ir3_shader
.nir
);
194 for (uint32_t i
= 0; i
< 1 + shader
->has_binning_pass
; i
++) {
195 if (shader
->variants
[i
].ir
)
196 ir3_destroy(shader
->variants
[i
].ir
);
199 if (shader
->ir3_shader
.const_state
.immediates
)
200 free(shader
->ir3_shader
.const_state
.immediates
);
202 free(shader
->binary
);
203 if (shader
->binning_binary
)
204 free(shader
->binning_binary
);
206 vk_free2(&dev
->alloc
, alloc
, shader
);
210 tu_shader_compile_options_init(
211 struct tu_shader_compile_options
*options
,
212 const VkGraphicsPipelineCreateInfo
*pipeline_info
)
214 *options
= (struct tu_shader_compile_options
) {
217 .optimize
= !(pipeline_info
->flags
&
218 VK_PIPELINE_CREATE_DISABLE_OPTIMIZATION_BIT
),
219 .include_binning_pass
= true,
224 tu_compile_shader_variant(struct ir3_shader
*shader
,
225 const struct ir3_shader_key
*key
,
227 struct ir3_shader_variant
*variant
)
229 variant
->shader
= shader
;
230 variant
->type
= shader
->type
;
232 variant
->binning_pass
= binning_pass
;
234 int ret
= ir3_compile_shader_nir(shader
->compiler
, variant
);
238 /* when assemble fails, we rely on tu_shader_destroy to clean up the
241 return ir3_shader_assemble(variant
, shader
->compiler
->gpu_id
);
245 tu_shader_compile(struct tu_device
*dev
,
246 struct tu_shader
*shader
,
247 const struct tu_shader
*next_stage
,
248 const struct tu_shader_compile_options
*options
,
249 const VkAllocationCallbacks
*alloc
)
251 if (options
->optimize
) {
252 /* ignore the key for the first pass of optimization */
253 ir3_optimize_nir(&shader
->ir3_shader
, shader
->ir3_shader
.nir
, NULL
);
255 if (unlikely(dev
->physical_device
->instance
->debug_flags
&
257 fprintf(stderr
, "optimized nir:\n");
258 nir_print_shader(shader
->ir3_shader
.nir
, stderr
);
262 shader
->binary
= tu_compile_shader_variant(
263 &shader
->ir3_shader
, &options
->key
, false, &shader
->variants
[0]);
265 return VK_ERROR_OUT_OF_HOST_MEMORY
;
267 /* compile another variant for the binning pass */
268 if (options
->include_binning_pass
&&
269 shader
->ir3_shader
.type
== MESA_SHADER_VERTEX
) {
270 shader
->binning_binary
= tu_compile_shader_variant(
271 &shader
->ir3_shader
, &options
->key
, true, &shader
->variants
[1]);
272 if (!shader
->binning_binary
)
273 return VK_ERROR_OUT_OF_HOST_MEMORY
;
275 shader
->has_binning_pass
= true;
278 if (unlikely(dev
->physical_device
->instance
->debug_flags
& TU_DEBUG_IR3
)) {
279 fprintf(stderr
, "disassembled ir3:\n");
280 fprintf(stderr
, "shader: %s\n",
281 gl_shader_stage_name(shader
->ir3_shader
.type
));
282 ir3_shader_disasm(&shader
->variants
[0], shader
->binary
, stderr
);
284 if (shader
->has_binning_pass
) {
285 fprintf(stderr
, "disassembled ir3:\n");
286 fprintf(stderr
, "shader: %s (binning)\n",
287 gl_shader_stage_name(shader
->ir3_shader
.type
));
288 ir3_shader_disasm(&shader
->variants
[1], shader
->binning_binary
,
297 tu_CreateShaderModule(VkDevice _device
,
298 const VkShaderModuleCreateInfo
*pCreateInfo
,
299 const VkAllocationCallbacks
*pAllocator
,
300 VkShaderModule
*pShaderModule
)
302 TU_FROM_HANDLE(tu_device
, device
, _device
);
303 struct tu_shader_module
*module
;
305 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO
);
306 assert(pCreateInfo
->flags
== 0);
307 assert(pCreateInfo
->codeSize
% 4 == 0);
309 module
= vk_alloc2(&device
->alloc
, pAllocator
,
310 sizeof(*module
) + pCreateInfo
->codeSize
, 8,
311 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
313 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
315 module
->code_size
= pCreateInfo
->codeSize
;
316 memcpy(module
->code
, pCreateInfo
->pCode
, pCreateInfo
->codeSize
);
318 _mesa_sha1_compute(module
->code
, module
->code_size
, module
->sha1
);
320 *pShaderModule
= tu_shader_module_to_handle(module
);
326 tu_DestroyShaderModule(VkDevice _device
,
327 VkShaderModule _module
,
328 const VkAllocationCallbacks
*pAllocator
)
330 TU_FROM_HANDLE(tu_device
, device
, _device
);
331 TU_FROM_HANDLE(tu_shader_module
, module
, _module
);
336 vk_free2(&device
->alloc
, pAllocator
, module
);