2 * Copyright (C) 2015 Rob Clark <robclark@freedesktop.org>
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 FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
24 * Rob Clark <robclark@freedesktop.org>
28 #include "util/debug.h"
31 #include "ir3_compiler.h"
32 #include "ir3_shader.h"
34 #include "nir/tgsi_to_nir.h"
37 static const nir_shader_compiler_options options
= {
48 .native_integers
= true,
49 .vertex_id_zero_based
= true,
50 .lower_extract_byte
= true,
51 .lower_extract_word
= true,
52 .lower_all_io_to_temps
= true,
53 .lower_helper_invocation
= true,
57 ir3_tgsi_to_nir(const struct tgsi_token
*tokens
)
59 return tgsi_to_nir(tokens
, &options
);
62 const nir_shader_compiler_options
*
63 ir3_get_compiler_options(struct ir3_compiler
*compiler
)
68 /* for given shader key, are any steps handled in nir? */
70 ir3_key_lowers_nir(const struct ir3_shader_key
*key
)
72 return key
->fsaturate_s
| key
->fsaturate_t
| key
->fsaturate_r
|
73 key
->vsaturate_s
| key
->vsaturate_t
| key
->vsaturate_r
|
74 key
->ucp_enables
| key
->color_two_side
|
75 key
->fclamp_color
| key
->vclamp_color
;
78 #define OPT(nir, pass, ...) ({ \
79 bool this_progress = false; \
80 NIR_PASS(this_progress, nir, pass, ##__VA_ARGS__); \
84 #define OPT_V(nir, pass, ...) NIR_PASS_V(nir, pass, ##__VA_ARGS__)
87 ir3_optimize_loop(nir_shader
*s
)
93 OPT_V(s
, nir_lower_vars_to_ssa
);
94 progress
|= OPT(s
, nir_opt_copy_prop_vars
);
95 progress
|= OPT(s
, nir_opt_dead_write_vars
);
96 progress
|= OPT(s
, nir_lower_alu_to_scalar
);
97 progress
|= OPT(s
, nir_lower_phis_to_scalar
);
99 progress
|= OPT(s
, nir_copy_prop
);
100 progress
|= OPT(s
, nir_opt_dce
);
101 progress
|= OPT(s
, nir_opt_cse
);
104 gcm
= env_var_as_unsigned("GCM", 0);
106 progress
|= OPT(s
, nir_opt_gcm
, true);
108 progress
|= OPT(s
, nir_opt_gcm
, false);
109 progress
|= OPT(s
, nir_opt_peephole_select
, 16);
110 progress
|= OPT(s
, nir_opt_intrinsics
);
111 progress
|= OPT(s
, nir_opt_algebraic
);
112 progress
|= OPT(s
, nir_opt_constant_folding
);
113 progress
|= OPT(s
, nir_opt_dead_cf
);
114 if (OPT(s
, nir_opt_trivial_continues
)) {
116 /* If nir_opt_trivial_continues makes progress, then we need to clean
117 * things up if we want any hope of nir_opt_if or nir_opt_loop_unroll
120 OPT(s
, nir_copy_prop
);
123 progress
|= OPT(s
, nir_opt_if
);
124 progress
|= OPT(s
, nir_opt_remove_phis
);
125 progress
|= OPT(s
, nir_opt_undef
);
131 ir3_optimize_nir(struct ir3_shader
*shader
, nir_shader
*s
,
132 const struct ir3_shader_key
*key
)
134 struct nir_lower_tex_options tex_options
= {
139 switch (shader
->type
) {
140 case MESA_SHADER_FRAGMENT
:
141 tex_options
.saturate_s
= key
->fsaturate_s
;
142 tex_options
.saturate_t
= key
->fsaturate_t
;
143 tex_options
.saturate_r
= key
->fsaturate_r
;
145 case MESA_SHADER_VERTEX
:
146 tex_options
.saturate_s
= key
->vsaturate_s
;
147 tex_options
.saturate_t
= key
->vsaturate_t
;
148 tex_options
.saturate_r
= key
->vsaturate_r
;
156 if (shader
->compiler
->gpu_id
>= 400) {
157 /* a4xx seems to have *no* sam.p */
158 tex_options
.lower_txp
= ~0; /* lower all txp */
160 /* a3xx just needs to avoid sam.p for 3d tex */
161 tex_options
.lower_txp
= (1 << GLSL_SAMPLER_DIM_3D
);
164 if (ir3_shader_debug
& IR3_DBG_DISASM
) {
165 debug_printf("----------------------\n");
166 nir_print_shader(s
, stdout
);
167 debug_printf("----------------------\n");
170 OPT_V(s
, nir_opt_global_to_local
);
171 OPT_V(s
, nir_lower_regs_to_ssa
);
174 if (s
->info
.stage
== MESA_SHADER_VERTEX
) {
175 OPT_V(s
, nir_lower_clip_vs
, key
->ucp_enables
, false);
176 if (key
->vclamp_color
)
177 OPT_V(s
, nir_lower_clamp_color_outputs
);
178 } else if (s
->info
.stage
== MESA_SHADER_FRAGMENT
) {
179 OPT_V(s
, nir_lower_clip_fs
, key
->ucp_enables
);
180 if (key
->fclamp_color
)
181 OPT_V(s
, nir_lower_clamp_color_outputs
);
183 if (key
->color_two_side
) {
184 OPT_V(s
, nir_lower_two_sided_color
);
187 /* only want to do this the first time (when key is null)
188 * and not again on any potential 2nd variant lowering pass:
190 OPT_V(s
, ir3_nir_apply_trig_workarounds
);
193 OPT_V(s
, nir_lower_tex
, &tex_options
);
194 OPT_V(s
, nir_lower_load_const_to_scalar
);
195 if (shader
->compiler
->gpu_id
< 500)
196 OPT_V(s
, ir3_nir_lower_tg4_to_tex
);
198 ir3_optimize_loop(s
);
200 /* do idiv lowering after first opt loop to give a chance for
201 * divide by immed power-of-two to be caught first:
203 if (OPT(s
, nir_lower_idiv
))
204 ir3_optimize_loop(s
);
206 OPT_V(s
, nir_remove_dead_variables
, nir_var_local
);
208 OPT_V(s
, nir_move_load_const
);
210 if (ir3_shader_debug
& IR3_DBG_DISASM
) {
211 debug_printf("----------------------\n");
212 nir_print_shader(s
, stdout
);
213 debug_printf("----------------------\n");
222 ir3_nir_scan_driver_consts(nir_shader
*shader
,
223 struct ir3_driver_const_layout
*layout
)
225 nir_foreach_function(function
, shader
) {
229 nir_foreach_block(block
, function
->impl
) {
230 nir_foreach_instr(instr
, block
) {
231 if (instr
->type
!= nir_instr_type_intrinsic
)
234 nir_intrinsic_instr
*intr
=
235 nir_instr_as_intrinsic(instr
);
238 switch (intr
->intrinsic
) {
239 case nir_intrinsic_get_buffer_size
:
240 idx
= nir_src_as_const_value(intr
->src
[0])->u32
[0];
241 if (layout
->ssbo_size
.mask
& (1 << idx
))
243 layout
->ssbo_size
.mask
|= (1 << idx
);
244 layout
->ssbo_size
.off
[idx
] =
245 layout
->ssbo_size
.count
;
246 layout
->ssbo_size
.count
+= 1; /* one const per */
248 case nir_intrinsic_image_deref_atomic_add
:
249 case nir_intrinsic_image_deref_atomic_min
:
250 case nir_intrinsic_image_deref_atomic_max
:
251 case nir_intrinsic_image_deref_atomic_and
:
252 case nir_intrinsic_image_deref_atomic_or
:
253 case nir_intrinsic_image_deref_atomic_xor
:
254 case nir_intrinsic_image_deref_atomic_exchange
:
255 case nir_intrinsic_image_deref_atomic_comp_swap
:
256 case nir_intrinsic_image_deref_store
:
257 case nir_intrinsic_image_deref_size
:
258 idx
= nir_intrinsic_get_var(intr
, 0)->data
.driver_location
;
259 if (layout
->image_dims
.mask
& (1 << idx
))
261 layout
->image_dims
.mask
|= (1 << idx
);
262 layout
->image_dims
.off
[idx
] =
263 layout
->image_dims
.count
;
264 layout
->image_dims
.count
+= 3; /* three const per */