1 /* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
4 * Copyright (C) 2014 Rob Clark <robclark@freedesktop.org>
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice (including the next
14 * paragraph) shall be included in all copies or substantial portions of the
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
26 * Rob Clark <robclark@freedesktop.org>
29 #include "pipe/p_state.h"
30 #include "util/u_string.h"
31 #include "util/u_memory.h"
32 #include "util/u_inlines.h"
33 #include "util/u_format.h"
34 #include "tgsi/tgsi_dump.h"
35 #include "tgsi/tgsi_parse.h"
37 #include "freedreno_context.h"
38 #include "freedreno_util.h"
40 #include "ir3_shader.h"
41 #include "ir3_compiler.h"
45 delete_variant(struct ir3_shader_variant
*v
)
54 assemble_variant(struct ir3_shader_variant
*v
)
56 struct fd_context
*ctx
= fd_context(v
->shader
->pctx
);
59 bin
= ir3_assemble(v
->ir
, &v
->info
, ctx
->screen
->gpu_id
);
60 sz
= v
->info
.sizedwords
* 4;
62 v
->bo
= fd_bo_new(ctx
->dev
, sz
,
63 DRM_FREEDRENO_GEM_CACHE_WCOMBINE
|
64 DRM_FREEDRENO_GEM_TYPE_KMEM
);
66 memcpy(fd_bo_map(v
->bo
), bin
, sz
);
70 if (ctx
->screen
->gpu_id
>= 400) {
71 v
->instrlen
= v
->info
.sizedwords
/ (2 * 16);
73 v
->instrlen
= v
->info
.sizedwords
/ (2 * 4);
76 /* NOTE: if relative addressing is used, we set constlen in
77 * the compiler (to worst-case value) since we don't know in
78 * the assembler what the max addr reg value can be:
80 v
->constlen
= MAX2(v
->constlen
, v
->info
.max_const
+ 1);
82 /* no need to keep the ir around beyond this point: */
87 /* for vertex shader, the inputs are loaded into registers before the shader
88 * is executed, so max_regs from the shader instructions might not properly
89 * reflect the # of registers actually used:
92 fixup_vp_regfootprint(struct ir3_shader_variant
*v
)
95 for (i
= 0; i
< v
->inputs_count
; i
++) {
96 if (v
->inputs
[i
].compmask
) {
97 int32_t regid
= (v
->inputs
[i
].regid
+ 3) >> 2;
98 v
->info
.max_reg
= MAX2(v
->info
.max_reg
, regid
);
101 for (i
= 0; i
< v
->outputs_count
; i
++) {
102 int32_t regid
= (v
->outputs
[i
].regid
+ 3) >> 2;
103 v
->info
.max_reg
= MAX2(v
->info
.max_reg
, regid
);
107 /* reset before attempting to compile again.. */
108 static void reset_variant(struct ir3_shader_variant
*v
, const char *msg
)
112 v
->outputs_count
= 0;
115 v
->immediates_count
= 0;
118 static struct ir3_shader_variant
*
119 create_variant(struct ir3_shader
*shader
, struct ir3_shader_key key
)
121 struct ir3_shader_variant
*v
= CALLOC_STRUCT(ir3_shader_variant
);
122 const struct tgsi_token
*tokens
= shader
->tokens
;
130 v
->type
= shader
->type
;
132 if (fd_mesa_debug
& FD_DBG_DISASM
) {
133 DBG("dump tgsi: type=%d, k={bp=%u,cts=%u,hp=%u}", shader
->type
,
134 key
.binning_pass
, key
.color_two_side
, key
.half_precision
);
135 tgsi_dump(tokens
, 0);
138 if (!(fd_mesa_debug
& FD_DBG_NOOPT
)) {
139 ret
= ir3_compile_shader(v
, tokens
, key
, true);
141 reset_variant(v
, "new compiler failed, trying without copy propagation!");
142 ret
= ir3_compile_shader(v
, tokens
, key
, false);
144 reset_variant(v
, "new compiler failed, trying fallback!");
147 ret
= -1; /* force fallback to old compiler */
151 ret
= ir3_compile_shader_old(v
, tokens
, key
);
154 debug_error("compile failed!");
160 debug_error("assemble failed!");
164 if (shader
->type
== SHADER_VERTEX
)
165 fixup_vp_regfootprint(v
);
167 if (fd_mesa_debug
& FD_DBG_DISASM
) {
168 DBG("disassemble: type=%d, k={bp=%u,cts=%u,hp=%u}", v
->type
,
169 key
.binning_pass
, key
.color_two_side
, key
.half_precision
);
170 disasm_a3xx(fd_bo_map(v
->bo
), v
->info
.sizedwords
, 0, v
->type
);
180 struct ir3_shader_variant
*
181 ir3_shader_variant(struct ir3_shader
*shader
, struct ir3_shader_key key
)
183 struct ir3_shader_variant
*v
;
185 /* some shader key values only apply to vertex or frag shader,
186 * so normalize the key to avoid constructing multiple identical
189 switch (shader
->type
) {
190 case SHADER_FRAGMENT
:
192 key
.binning_pass
= false;
193 if (key
.has_per_samp
) {
200 key
.color_two_side
= false;
201 key
.half_precision
= false;
203 if (key
.has_per_samp
) {
211 for (v
= shader
->variants
; v
; v
= v
->next
)
212 if (ir3_shader_key_equal(&key
, &v
->key
))
215 /* compile new variant if it doesn't exist already: */
216 v
= create_variant(shader
, key
);
217 v
->next
= shader
->variants
;
218 shader
->variants
= v
;
225 ir3_shader_destroy(struct ir3_shader
*shader
)
227 struct ir3_shader_variant
*v
, *t
;
228 for (v
= shader
->variants
; v
; ) {
233 free((void *)shader
->tokens
);
238 ir3_shader_create(struct pipe_context
*pctx
, const struct tgsi_token
*tokens
,
241 struct ir3_shader
*shader
= CALLOC_STRUCT(ir3_shader
);
244 shader
->tokens
= tgsi_dup_tokens(tokens
);