2 * Copyright © 2017 Red Hat
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 DEALINGS
23 #include "radv_private.h"
24 #include "radv_shader.h"
26 #include "nir/nir_deref.h"
28 static void mark_sampler_desc(const nir_variable
*var
,
29 struct radv_shader_info
*info
)
31 info
->desc_set_used_mask
|= (1 << var
->data
.descriptor_set
);
34 static void mark_ls_output(struct radv_shader_info
*info
,
35 uint32_t param
, int num_slots
)
37 uint64_t mask
= (1ull << num_slots
) - 1ull;
38 info
->vs
.ls_outputs_written
|= (mask
<< param
);
41 static void mark_tess_output(struct radv_shader_info
*info
,
42 bool is_patch
, uint32_t param
, int num_slots
)
44 uint64_t mask
= (1ull << num_slots
) - 1ull;
46 info
->tcs
.patch_outputs_written
|= (mask
<< param
);
48 info
->tcs
.outputs_written
|= (mask
<< param
);
52 get_deref_offset(nir_deref_instr
*instr
,
55 nir_variable
*var
= nir_deref_instr_get_variable(instr
);
59 if (var
->data
.compact
) {
60 assert(instr
->deref_type
== nir_deref_type_array
);
61 nir_const_value
*v
= nir_src_as_const_value(instr
->arr
.index
);
63 *const_out
= v
->u32
[0];
67 nir_deref_path_init(&path
, instr
, NULL
);
69 uint32_t const_offset
= 0;
71 for (; path
.path
[idx_lvl
]; ++idx_lvl
) {
72 const struct glsl_type
*parent_type
= path
.path
[idx_lvl
- 1]->type
;
73 if (path
.path
[idx_lvl
]->deref_type
== nir_deref_type_struct
) {
74 unsigned index
= path
.path
[idx_lvl
]->strct
.index
;
76 for (unsigned i
= 0; i
< index
; i
++) {
77 const struct glsl_type
*ft
= glsl_get_struct_field(parent_type
, i
);
78 const_offset
+= glsl_count_attribute_slots(ft
, false);
80 } else if(path
.path
[idx_lvl
]->deref_type
== nir_deref_type_array
) {
81 unsigned size
= glsl_count_attribute_slots(path
.path
[idx_lvl
]->type
, false);
82 nir_const_value
*v
= nir_src_as_const_value(path
.path
[idx_lvl
]->arr
.index
);
84 const_offset
+= v
->u32
[0] * size
;
86 unreachable("Uhandled deref type in get_deref_instr_offset");
89 *const_out
= const_offset
;
91 nir_deref_path_finish(&path
);
95 gather_intrinsic_load_deref_info(const nir_shader
*nir
,
96 const nir_intrinsic_instr
*instr
,
97 struct radv_shader_info
*info
)
99 switch (nir
->info
.stage
) {
100 case MESA_SHADER_VERTEX
: {
101 nir_variable
*var
= nir_deref_instr_get_variable(nir_instr_as_deref(instr
->src
[0].ssa
->parent_instr
));
103 if (var
->data
.mode
== nir_var_shader_in
) {
104 unsigned idx
= var
->data
.location
;
105 uint8_t mask
= nir_ssa_def_components_read(&instr
->dest
.ssa
);
107 info
->vs
.input_usage_mask
[idx
] |=
108 mask
<< var
->data
.location_frac
;
118 gather_intrinsic_store_deref_info(const nir_shader
*nir
,
119 const nir_intrinsic_instr
*instr
,
120 struct radv_shader_info
*info
)
122 nir_variable
*var
= nir_deref_instr_get_variable(nir_instr_as_deref(instr
->src
[0].ssa
->parent_instr
));
124 if (var
->data
.mode
== nir_var_shader_out
) {
125 unsigned attrib_count
= glsl_count_attribute_slots(var
->type
, false);
126 unsigned idx
= var
->data
.location
;
127 unsigned comp
= var
->data
.location_frac
;
128 unsigned const_offset
= 0;
130 get_deref_offset(nir_instr_as_deref(instr
->src
[0].ssa
->parent_instr
), &const_offset
);
132 switch (nir
->info
.stage
) {
133 case MESA_SHADER_VERTEX
:
134 for (unsigned i
= 0; i
< attrib_count
; i
++) {
135 info
->vs
.output_usage_mask
[idx
+ i
+ const_offset
] |=
136 instr
->const_index
[0] << comp
;
139 case MESA_SHADER_GEOMETRY
:
140 for (unsigned i
= 0; i
< attrib_count
; i
++) {
141 info
->gs
.output_usage_mask
[idx
+ i
+ const_offset
] |=
142 instr
->const_index
[0] << comp
;
145 case MESA_SHADER_TESS_EVAL
:
146 for (unsigned i
= 0; i
< attrib_count
; i
++) {
147 info
->tes
.output_usage_mask
[idx
+ i
+ const_offset
] |=
148 instr
->const_index
[0] << comp
;
151 case MESA_SHADER_TESS_CTRL
: {
152 unsigned param
= shader_io_get_unique_index(idx
);
153 const struct glsl_type
*type
= var
->type
;
155 if (!var
->data
.patch
)
156 type
= glsl_get_array_element(var
->type
);
159 var
->data
.compact
? DIV_ROUND_UP(glsl_get_length(type
), 4)
160 : glsl_count_attribute_slots(type
, false);
162 if (idx
== VARYING_SLOT_CLIP_DIST0
)
163 slots
= (nir
->info
.clip_distance_array_size
+
164 nir
->info
.cull_distance_array_size
> 4) ? 2 : 1;
166 mark_tess_output(info
, var
->data
.patch
, param
, slots
);
176 gather_intrinsic_info(const nir_shader
*nir
, const nir_intrinsic_instr
*instr
,
177 struct radv_shader_info
*info
)
179 switch (instr
->intrinsic
) {
180 case nir_intrinsic_interp_deref_at_sample
:
181 info
->ps
.needs_sample_positions
= true;
183 case nir_intrinsic_load_draw_id
:
184 info
->vs
.needs_draw_id
= true;
186 case nir_intrinsic_load_instance_id
:
187 info
->vs
.needs_instance_id
= true;
189 case nir_intrinsic_load_num_work_groups
:
190 info
->cs
.uses_grid_size
= true;
192 case nir_intrinsic_load_local_invocation_id
:
193 case nir_intrinsic_load_work_group_id
: {
194 unsigned mask
= nir_ssa_def_components_read(&instr
->dest
.ssa
);
196 unsigned i
= u_bit_scan(&mask
);
198 if (instr
->intrinsic
== nir_intrinsic_load_work_group_id
)
199 info
->cs
.uses_block_id
[i
] = true;
201 info
->cs
.uses_thread_id
[i
] = true;
205 case nir_intrinsic_load_local_invocation_index
:
206 case nir_intrinsic_load_subgroup_id
:
207 case nir_intrinsic_load_num_subgroups
:
208 info
->cs
.uses_local_invocation_idx
= true;
210 case nir_intrinsic_load_sample_id
:
211 info
->ps
.force_persample
= true;
213 case nir_intrinsic_load_sample_pos
:
214 info
->ps
.force_persample
= true;
216 case nir_intrinsic_load_view_index
:
217 info
->needs_multiview_view_index
= true;
218 if (nir
->info
.stage
== MESA_SHADER_FRAGMENT
)
219 info
->ps
.layer_input
= true;
221 case nir_intrinsic_load_invocation_id
:
222 info
->uses_invocation_id
= true;
224 case nir_intrinsic_load_primitive_id
:
225 info
->uses_prim_id
= true;
227 case nir_intrinsic_load_push_constant
:
228 info
->loads_push_constants
= true;
230 case nir_intrinsic_vulkan_resource_index
:
231 info
->desc_set_used_mask
|= (1 << nir_intrinsic_desc_set(instr
));
233 case nir_intrinsic_image_deref_load
:
234 case nir_intrinsic_image_deref_store
:
235 case nir_intrinsic_image_deref_atomic_add
:
236 case nir_intrinsic_image_deref_atomic_min
:
237 case nir_intrinsic_image_deref_atomic_max
:
238 case nir_intrinsic_image_deref_atomic_and
:
239 case nir_intrinsic_image_deref_atomic_or
:
240 case nir_intrinsic_image_deref_atomic_xor
:
241 case nir_intrinsic_image_deref_atomic_exchange
:
242 case nir_intrinsic_image_deref_atomic_comp_swap
:
243 case nir_intrinsic_image_deref_size
: {
244 nir_variable
*var
= nir_deref_instr_get_variable(nir_instr_as_deref(instr
->src
[0].ssa
->parent_instr
));
245 const struct glsl_type
*type
= glsl_without_array(var
->type
);
247 enum glsl_sampler_dim dim
= glsl_get_sampler_dim(type
);
248 if (dim
== GLSL_SAMPLER_DIM_SUBPASS
||
249 dim
== GLSL_SAMPLER_DIM_SUBPASS_MS
) {
250 info
->ps
.layer_input
= true;
251 info
->ps
.uses_input_attachments
= true;
253 mark_sampler_desc(var
, info
);
255 if (nir_intrinsic_image_deref_store
||
256 nir_intrinsic_image_deref_atomic_add
||
257 nir_intrinsic_image_deref_atomic_min
||
258 nir_intrinsic_image_deref_atomic_max
||
259 nir_intrinsic_image_deref_atomic_and
||
260 nir_intrinsic_image_deref_atomic_or
||
261 nir_intrinsic_image_deref_atomic_xor
||
262 nir_intrinsic_image_deref_atomic_exchange
||
263 nir_intrinsic_image_deref_atomic_comp_swap
) {
264 if (nir
->info
.stage
== MESA_SHADER_FRAGMENT
)
265 info
->ps
.writes_memory
= true;
269 case nir_intrinsic_store_ssbo
:
270 case nir_intrinsic_ssbo_atomic_add
:
271 case nir_intrinsic_ssbo_atomic_imin
:
272 case nir_intrinsic_ssbo_atomic_umin
:
273 case nir_intrinsic_ssbo_atomic_imax
:
274 case nir_intrinsic_ssbo_atomic_umax
:
275 case nir_intrinsic_ssbo_atomic_and
:
276 case nir_intrinsic_ssbo_atomic_or
:
277 case nir_intrinsic_ssbo_atomic_xor
:
278 case nir_intrinsic_ssbo_atomic_exchange
:
279 case nir_intrinsic_ssbo_atomic_comp_swap
:
280 if (nir
->info
.stage
== MESA_SHADER_FRAGMENT
)
281 info
->ps
.writes_memory
= true;
283 case nir_intrinsic_load_deref
:
284 gather_intrinsic_load_deref_info(nir
, instr
, info
);
286 case nir_intrinsic_store_deref
:
287 gather_intrinsic_store_deref_info(nir
, instr
, info
);
295 gather_tex_info(const nir_shader
*nir
, const nir_tex_instr
*instr
,
296 struct radv_shader_info
*info
)
298 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++) {
299 switch (instr
->src
[i
].src_type
) {
300 case nir_tex_src_texture_deref
:
301 mark_sampler_desc(nir_deref_instr_get_variable(nir_src_as_deref(instr
->src
[i
].src
)), info
);
303 case nir_tex_src_sampler_deref
:
304 mark_sampler_desc(nir_deref_instr_get_variable(nir_src_as_deref(instr
->src
[i
].src
)), info
);
313 gather_info_block(const nir_shader
*nir
, const nir_block
*block
,
314 struct radv_shader_info
*info
)
316 nir_foreach_instr(instr
, block
) {
317 switch (instr
->type
) {
318 case nir_instr_type_intrinsic
:
319 gather_intrinsic_info(nir
, nir_instr_as_intrinsic(instr
), info
);
321 case nir_instr_type_tex
:
322 gather_tex_info(nir
, nir_instr_as_tex(instr
), info
);
331 gather_info_input_decl_vs(const nir_shader
*nir
, const nir_variable
*var
,
332 struct radv_shader_info
*info
)
334 int idx
= var
->data
.location
;
336 if (idx
>= VERT_ATTRIB_GENERIC0
&& idx
<= VERT_ATTRIB_GENERIC15
)
337 info
->vs
.has_vertex_buffers
= true;
341 gather_info_input_decl_ps(const nir_shader
*nir
, const nir_variable
*var
,
342 struct radv_shader_info
*info
)
344 unsigned attrib_count
= glsl_count_attribute_slots(var
->type
, false);
345 const struct glsl_type
*type
= glsl_without_array(var
->type
);
346 int idx
= var
->data
.location
;
349 case VARYING_SLOT_PNTC
:
350 info
->ps
.has_pcoord
= true;
352 case VARYING_SLOT_PRIMITIVE_ID
:
353 info
->ps
.prim_id_input
= true;
355 case VARYING_SLOT_LAYER
:
356 info
->ps
.layer_input
= true;
358 case VARYING_SLOT_CLIP_DIST0
:
359 info
->ps
.num_input_clips_culls
= attrib_count
;
365 if (glsl_get_base_type(type
) == GLSL_TYPE_FLOAT
) {
366 if (var
->data
.sample
)
367 info
->ps
.force_persample
= true;
372 gather_info_input_decl(const nir_shader
*nir
, const nir_variable
*var
,
373 struct radv_shader_info
*info
)
375 switch (nir
->info
.stage
) {
376 case MESA_SHADER_VERTEX
:
377 gather_info_input_decl_vs(nir
, var
, info
);
379 case MESA_SHADER_FRAGMENT
:
380 gather_info_input_decl_ps(nir
, var
, info
);
388 gather_info_output_decl_ls(const nir_shader
*nir
, const nir_variable
*var
,
389 struct radv_shader_info
*info
)
391 int idx
= var
->data
.location
;
392 unsigned param
= shader_io_get_unique_index(idx
);
393 int num_slots
= glsl_count_attribute_slots(var
->type
, false);
394 if (idx
== VARYING_SLOT_CLIP_DIST0
)
395 num_slots
= (nir
->info
.clip_distance_array_size
+ nir
->info
.cull_distance_array_size
> 4) ? 2 : 1;
396 mark_ls_output(info
, param
, num_slots
);
400 gather_info_output_decl_ps(const nir_shader
*nir
, const nir_variable
*var
,
401 struct radv_shader_info
*info
)
403 int idx
= var
->data
.location
;
406 case FRAG_RESULT_DEPTH
:
407 info
->ps
.writes_z
= true;
409 case FRAG_RESULT_STENCIL
:
410 info
->ps
.writes_stencil
= true;
412 case FRAG_RESULT_SAMPLE_MASK
:
413 info
->ps
.writes_sample_mask
= true;
421 gather_info_output_decl(const nir_shader
*nir
, const nir_variable
*var
,
422 struct radv_shader_info
*info
,
423 const struct radv_nir_compiler_options
*options
)
425 switch (nir
->info
.stage
) {
426 case MESA_SHADER_FRAGMENT
:
427 gather_info_output_decl_ps(nir
, var
, info
);
429 case MESA_SHADER_VERTEX
:
430 if (options
->key
.vs
.as_ls
)
431 gather_info_output_decl_ls(nir
, var
, info
);
439 radv_nir_shader_info_pass(const struct nir_shader
*nir
,
440 const struct radv_nir_compiler_options
*options
,
441 struct radv_shader_info
*info
)
443 struct nir_function
*func
=
444 (struct nir_function
*)exec_list_get_head_const(&nir
->functions
);
446 if (options
->layout
&& options
->layout
->dynamic_offset_count
)
447 info
->loads_push_constants
= true;
449 nir_foreach_variable(variable
, &nir
->inputs
)
450 gather_info_input_decl(nir
, variable
, info
);
452 nir_foreach_block(block
, func
->impl
) {
453 gather_info_block(nir
, block
, info
);
456 nir_foreach_variable(variable
, &nir
->outputs
)
457 gather_info_output_decl(nir
, variable
, info
, options
);