2 * Copyright (c) 2017-2019 Lima Project
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, sub license,
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
12 * next paragraph) shall be included in all copies or substantial portions
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 NON-INFRINGEMENT. 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.
25 #include "util/u_memory.h"
26 #include "util/ralloc.h"
27 #include "util/u_debug.h"
29 #include "tgsi/tgsi_dump.h"
30 #include "compiler/nir/nir.h"
31 #include "nir/tgsi_to_nir.h"
33 #include "pipe/p_state.h"
35 #include "lima_screen.h"
36 #include "lima_context.h"
38 #include "lima_program.h"
40 #include "ir/lima_ir.h"
42 static const nir_shader_compiler_options vs_nir_options
= {
53 /* could be implemented by clamp */
61 static const nir_shader_compiler_options fs_nir_options
= {
74 .lower_vector_cmp
= true,
77 static const struct nir_lower_tex_options tex_options
= {
82 lima_program_get_compiler_options(enum pipe_shader_type shader
)
85 case PIPE_SHADER_VERTEX
:
86 return &vs_nir_options
;
87 case PIPE_SHADER_FRAGMENT
:
88 return &fs_nir_options
;
95 type_size(const struct glsl_type
*type
, bool bindless
)
97 return glsl_count_attribute_slots(type
, false);
101 lima_program_optimize_vs_nir(struct nir_shader
*s
)
105 NIR_PASS_V(s
, nir_lower_viewport_transform
);
106 NIR_PASS_V(s
, nir_lower_point_size
, 1.0f
, 100.0f
);
107 NIR_PASS_V(s
, nir_lower_io
, nir_var_all
, type_size
, 0);
108 NIR_PASS_V(s
, nir_lower_load_const_to_scalar
);
109 NIR_PASS_V(s
, lima_nir_lower_uniform_to_scalar
);
110 NIR_PASS_V(s
, nir_lower_io_to_scalar
,
111 nir_var_shader_in
|nir_var_shader_out
);
116 NIR_PASS_V(s
, nir_lower_vars_to_ssa
);
117 NIR_PASS(progress
, s
, nir_lower_alu_to_scalar
, NULL
, NULL
);
118 NIR_PASS(progress
, s
, nir_lower_phis_to_scalar
);
119 NIR_PASS(progress
, s
, nir_copy_prop
);
120 NIR_PASS(progress
, s
, nir_opt_remove_phis
);
121 NIR_PASS(progress
, s
, nir_opt_dce
);
122 NIR_PASS(progress
, s
, nir_opt_dead_cf
);
123 NIR_PASS(progress
, s
, nir_opt_cse
);
124 NIR_PASS(progress
, s
, nir_opt_peephole_select
, 8, true, true);
125 NIR_PASS(progress
, s
, nir_opt_algebraic
);
126 NIR_PASS(progress
, s
, nir_opt_constant_folding
);
127 NIR_PASS(progress
, s
, nir_opt_undef
);
128 NIR_PASS(progress
, s
, nir_opt_loop_unroll
,
131 nir_var_function_temp
);
134 NIR_PASS_V(s
, nir_lower_int_to_float
);
135 /* Run opt_algebraic between int_to_float and bool_to_float because
136 * int_to_float emits ftrunc, and ftrunc lowering generates bool ops
140 NIR_PASS(progress
, s
, nir_opt_algebraic
);
143 NIR_PASS_V(s
, nir_lower_bool_to_float
);
145 NIR_PASS_V(s
, nir_copy_prop
);
146 NIR_PASS_V(s
, nir_opt_dce
);
147 NIR_PASS_V(s
, nir_lower_locals_to_regs
);
148 NIR_PASS_V(s
, nir_convert_from_ssa
, true);
149 NIR_PASS_V(s
, nir_remove_dead_variables
, nir_var_function_temp
);
154 lima_alu_to_scalar_filter_cb(const nir_instr
*instr
, const void *data
)
156 if (instr
->type
!= nir_instr_type_alu
)
159 nir_alu_instr
*alu
= nir_instr_as_alu(instr
);
173 /* nir vec4 fcsel assumes that each component of the condition will be
174 * used to select the same component from the two options, but Utgard PP
175 * has only 1 component condition. If all condition components are not the
176 * same we need to lower it to scalar.
186 int num_components
= nir_dest_num_components(alu
->dest
.dest
);
188 uint8_t swizzle
= alu
->src
[0].swizzle
[0];
190 for (int i
= 1; i
< num_components
; i
++)
191 if (alu
->src
[0].swizzle
[i
] != swizzle
)
198 lima_program_optimize_fs_nir(struct nir_shader
*s
)
202 NIR_PASS_V(s
, nir_lower_fragcoord_wtrans
);
203 NIR_PASS_V(s
, nir_lower_io
, nir_var_all
, type_size
, 0);
204 NIR_PASS_V(s
, nir_lower_regs_to_ssa
);
205 NIR_PASS_V(s
, nir_lower_tex
, &tex_options
);
209 NIR_PASS(progress
, s
, nir_opt_vectorize
);
215 NIR_PASS_V(s
, nir_lower_vars_to_ssa
);
216 NIR_PASS(progress
, s
, nir_lower_alu_to_scalar
, lima_alu_to_scalar_filter_cb
, NULL
);
217 NIR_PASS(progress
, s
, nir_copy_prop
);
218 NIR_PASS(progress
, s
, nir_opt_remove_phis
);
219 NIR_PASS(progress
, s
, nir_opt_dce
);
220 NIR_PASS(progress
, s
, nir_opt_dead_cf
);
221 NIR_PASS(progress
, s
, nir_opt_cse
);
222 NIR_PASS(progress
, s
, nir_opt_peephole_select
, 8, true, true);
223 NIR_PASS(progress
, s
, nir_opt_algebraic
);
224 NIR_PASS(progress
, s
, nir_opt_constant_folding
);
225 NIR_PASS(progress
, s
, nir_opt_undef
);
226 NIR_PASS(progress
, s
, nir_opt_loop_unroll
,
229 nir_var_function_temp
);
230 NIR_PASS(progress
, s
, lima_nir_split_load_input
);
233 NIR_PASS_V(s
, nir_lower_int_to_float
);
234 NIR_PASS_V(s
, nir_lower_bool_to_float
);
236 /* Some ops must be lowered after being converted from int ops,
237 * so re-run nir_opt_algebraic after int lowering. */
240 NIR_PASS(progress
, s
, nir_opt_algebraic
);
243 /* Must be run after optimization loop */
244 NIR_PASS_V(s
, lima_nir_scale_trig
);
246 /* Lower modifiers */
247 NIR_PASS_V(s
, nir_lower_to_source_mods
, nir_lower_all_source_mods
);
248 NIR_PASS_V(s
, nir_copy_prop
);
249 NIR_PASS_V(s
, nir_opt_dce
);
251 NIR_PASS_V(s
, nir_lower_locals_to_regs
);
252 NIR_PASS_V(s
, nir_convert_from_ssa
, true);
253 NIR_PASS_V(s
, nir_remove_dead_variables
, nir_var_function_temp
);
255 NIR_PASS_V(s
, nir_move_vec_src_uses_to_dest
);
256 NIR_PASS_V(s
, nir_lower_vec_to_movs
);
262 lima_create_fs_state(struct pipe_context
*pctx
,
263 const struct pipe_shader_state
*cso
)
265 struct lima_context
*ctx
= lima_context(pctx
);
266 struct lima_screen
*screen
= lima_screen(pctx
->screen
);
267 struct lima_fs_shader_state
*so
= rzalloc(NULL
, struct lima_fs_shader_state
);
273 if (cso
->type
== PIPE_SHADER_IR_NIR
)
276 assert(cso
->type
== PIPE_SHADER_IR_TGSI
);
278 nir
= tgsi_to_nir(cso
->tokens
, pctx
->screen
);
281 lima_program_optimize_fs_nir(nir
);
283 if (lima_debug
& LIMA_DEBUG_PP
)
284 nir_print_shader(nir
, stdout
);
286 if (!ppir_compile_nir(so
, nir
, screen
->pp_ra
, &ctx
->debug
)) {
291 so
->uses_discard
= nir
->info
.fs
.uses_discard
;
297 lima_bind_fs_state(struct pipe_context
*pctx
, void *hwcso
)
299 struct lima_context
*ctx
= lima_context(pctx
);
302 ctx
->dirty
|= LIMA_CONTEXT_DIRTY_SHADER_FRAG
;
306 lima_delete_fs_state(struct pipe_context
*pctx
, void *hwcso
)
308 struct lima_fs_shader_state
*so
= hwcso
;
311 lima_bo_unreference(so
->bo
);
317 lima_update_vs_state(struct lima_context
*ctx
)
319 struct lima_vs_shader_state
*vs
= ctx
->vs
;
321 struct lima_screen
*screen
= lima_screen(ctx
->base
.screen
);
322 vs
->bo
= lima_bo_create(screen
, vs
->shader_size
, 0);
324 fprintf(stderr
, "lima: create vs shader bo fail\n");
328 memcpy(lima_bo_map(vs
->bo
), vs
->shader
, vs
->shader_size
);
329 ralloc_free(vs
->shader
);
337 lima_update_fs_state(struct lima_context
*ctx
)
339 struct lima_fs_shader_state
*fs
= ctx
->fs
;
341 struct lima_screen
*screen
= lima_screen(ctx
->base
.screen
);
342 fs
->bo
= lima_bo_create(screen
, fs
->shader_size
, 0);
344 fprintf(stderr
, "lima: create fs shader bo fail\n");
348 memcpy(lima_bo_map(fs
->bo
), fs
->shader
, fs
->shader_size
);
349 ralloc_free(fs
->shader
);
353 struct lima_job
*job
= lima_job_get(ctx
);
354 job
->pp_max_stack_size
= MAX2(job
->pp_max_stack_size
, ctx
->fs
->stack_size
);
360 lima_create_vs_state(struct pipe_context
*pctx
,
361 const struct pipe_shader_state
*cso
)
363 struct lima_context
*ctx
= lima_context(pctx
);
364 struct lima_vs_shader_state
*so
= rzalloc(NULL
, struct lima_vs_shader_state
);
370 if (cso
->type
== PIPE_SHADER_IR_NIR
)
373 assert(cso
->type
== PIPE_SHADER_IR_TGSI
);
375 nir
= tgsi_to_nir(cso
->tokens
, pctx
->screen
);
378 lima_program_optimize_vs_nir(nir
);
380 if (lima_debug
& LIMA_DEBUG_GP
)
381 nir_print_shader(nir
, stdout
);
383 if (!gpir_compile_nir(so
, nir
, &ctx
->debug
)) {
394 lima_bind_vs_state(struct pipe_context
*pctx
, void *hwcso
)
396 struct lima_context
*ctx
= lima_context(pctx
);
399 ctx
->dirty
|= LIMA_CONTEXT_DIRTY_SHADER_VERT
;
403 lima_delete_vs_state(struct pipe_context
*pctx
, void *hwcso
)
405 struct lima_vs_shader_state
*so
= hwcso
;
408 lima_bo_unreference(so
->bo
);
414 lima_program_init(struct lima_context
*ctx
)
416 ctx
->base
.create_fs_state
= lima_create_fs_state
;
417 ctx
->base
.bind_fs_state
= lima_bind_fs_state
;
418 ctx
->base
.delete_fs_state
= lima_delete_fs_state
;
420 ctx
->base
.create_vs_state
= lima_create_vs_state
;
421 ctx
->base
.bind_vs_state
= lima_bind_vs_state
;
422 ctx
->base
.delete_vs_state
= lima_delete_vs_state
;