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 set_output_usage_mask(const nir_shader
*nir
, const nir_intrinsic_instr
*instr
,
119 uint8_t *output_usage_mask
)
121 nir_deref_instr
*deref_instr
=
122 nir_instr_as_deref(instr
->src
[0].ssa
->parent_instr
);
123 nir_variable
*var
= nir_deref_instr_get_variable(deref_instr
);
124 unsigned attrib_count
= glsl_count_attribute_slots(var
->type
, false);
125 unsigned idx
= var
->data
.location
;
126 unsigned comp
= var
->data
.location_frac
;
127 unsigned const_offset
= 0;
129 get_deref_offset(deref_instr
, &const_offset
);
131 if (idx
== VARYING_SLOT_CLIP_DIST0
) {
132 /* Special case for clip/cull distances because there are
133 * combined into a single array that contains both.
135 output_usage_mask
[idx
] |= 1 << const_offset
;
139 for (unsigned i
= 0; i
< attrib_count
; i
++) {
140 output_usage_mask
[idx
+ i
+ const_offset
] |=
141 instr
->const_index
[0] << comp
;
146 gather_intrinsic_store_deref_info(const nir_shader
*nir
,
147 const nir_intrinsic_instr
*instr
,
148 struct radv_shader_info
*info
)
150 nir_variable
*var
= nir_deref_instr_get_variable(nir_instr_as_deref(instr
->src
[0].ssa
->parent_instr
));
152 if (var
->data
.mode
== nir_var_shader_out
) {
153 unsigned idx
= var
->data
.location
;
155 switch (nir
->info
.stage
) {
156 case MESA_SHADER_VERTEX
:
157 set_output_usage_mask(nir
, instr
,
158 info
->vs
.output_usage_mask
);
160 case MESA_SHADER_GEOMETRY
:
161 set_output_usage_mask(nir
, instr
,
162 info
->gs
.output_usage_mask
);
164 case MESA_SHADER_TESS_EVAL
:
165 set_output_usage_mask(nir
, instr
,
166 info
->tes
.output_usage_mask
);
168 case MESA_SHADER_TESS_CTRL
: {
169 unsigned param
= shader_io_get_unique_index(idx
);
170 const struct glsl_type
*type
= var
->type
;
172 if (!var
->data
.patch
)
173 type
= glsl_get_array_element(var
->type
);
176 var
->data
.compact
? DIV_ROUND_UP(glsl_get_length(type
), 4)
177 : glsl_count_attribute_slots(type
, false);
179 if (idx
== VARYING_SLOT_CLIP_DIST0
)
180 slots
= (nir
->info
.clip_distance_array_size
+
181 nir
->info
.cull_distance_array_size
> 4) ? 2 : 1;
183 mark_tess_output(info
, var
->data
.patch
, param
, slots
);
193 gather_intrinsic_info(const nir_shader
*nir
, const nir_intrinsic_instr
*instr
,
194 struct radv_shader_info
*info
)
196 switch (instr
->intrinsic
) {
197 case nir_intrinsic_interp_deref_at_sample
:
198 info
->ps
.needs_sample_positions
= true;
200 case nir_intrinsic_load_draw_id
:
201 info
->vs
.needs_draw_id
= true;
203 case nir_intrinsic_load_instance_id
:
204 info
->vs
.needs_instance_id
= true;
206 case nir_intrinsic_load_num_work_groups
:
207 info
->cs
.uses_grid_size
= true;
209 case nir_intrinsic_load_local_invocation_id
:
210 case nir_intrinsic_load_work_group_id
: {
211 unsigned mask
= nir_ssa_def_components_read(&instr
->dest
.ssa
);
213 unsigned i
= u_bit_scan(&mask
);
215 if (instr
->intrinsic
== nir_intrinsic_load_work_group_id
)
216 info
->cs
.uses_block_id
[i
] = true;
218 info
->cs
.uses_thread_id
[i
] = true;
222 case nir_intrinsic_load_local_invocation_index
:
223 case nir_intrinsic_load_subgroup_id
:
224 case nir_intrinsic_load_num_subgroups
:
225 info
->cs
.uses_local_invocation_idx
= true;
227 case nir_intrinsic_load_sample_id
:
228 info
->ps
.force_persample
= true;
230 case nir_intrinsic_load_sample_pos
:
231 info
->ps
.force_persample
= true;
233 case nir_intrinsic_load_view_index
:
234 info
->needs_multiview_view_index
= true;
235 if (nir
->info
.stage
== MESA_SHADER_FRAGMENT
)
236 info
->ps
.layer_input
= true;
238 case nir_intrinsic_load_invocation_id
:
239 info
->uses_invocation_id
= true;
241 case nir_intrinsic_load_primitive_id
:
242 info
->uses_prim_id
= true;
244 case nir_intrinsic_load_push_constant
:
245 info
->loads_push_constants
= true;
247 case nir_intrinsic_vulkan_resource_index
:
248 info
->desc_set_used_mask
|= (1 << nir_intrinsic_desc_set(instr
));
250 case nir_intrinsic_image_deref_load
:
251 case nir_intrinsic_image_deref_store
:
252 case nir_intrinsic_image_deref_atomic_add
:
253 case nir_intrinsic_image_deref_atomic_min
:
254 case nir_intrinsic_image_deref_atomic_max
:
255 case nir_intrinsic_image_deref_atomic_and
:
256 case nir_intrinsic_image_deref_atomic_or
:
257 case nir_intrinsic_image_deref_atomic_xor
:
258 case nir_intrinsic_image_deref_atomic_exchange
:
259 case nir_intrinsic_image_deref_atomic_comp_swap
:
260 case nir_intrinsic_image_deref_size
: {
261 nir_variable
*var
= nir_deref_instr_get_variable(nir_instr_as_deref(instr
->src
[0].ssa
->parent_instr
));
262 const struct glsl_type
*type
= glsl_without_array(var
->type
);
264 enum glsl_sampler_dim dim
= glsl_get_sampler_dim(type
);
265 if (dim
== GLSL_SAMPLER_DIM_SUBPASS
||
266 dim
== GLSL_SAMPLER_DIM_SUBPASS_MS
) {
267 info
->ps
.layer_input
= true;
268 info
->ps
.uses_input_attachments
= true;
270 mark_sampler_desc(var
, info
);
272 if (nir_intrinsic_image_deref_store
||
273 nir_intrinsic_image_deref_atomic_add
||
274 nir_intrinsic_image_deref_atomic_min
||
275 nir_intrinsic_image_deref_atomic_max
||
276 nir_intrinsic_image_deref_atomic_and
||
277 nir_intrinsic_image_deref_atomic_or
||
278 nir_intrinsic_image_deref_atomic_xor
||
279 nir_intrinsic_image_deref_atomic_exchange
||
280 nir_intrinsic_image_deref_atomic_comp_swap
) {
281 if (nir
->info
.stage
== MESA_SHADER_FRAGMENT
)
282 info
->ps
.writes_memory
= true;
286 case nir_intrinsic_store_ssbo
:
287 case nir_intrinsic_ssbo_atomic_add
:
288 case nir_intrinsic_ssbo_atomic_imin
:
289 case nir_intrinsic_ssbo_atomic_umin
:
290 case nir_intrinsic_ssbo_atomic_imax
:
291 case nir_intrinsic_ssbo_atomic_umax
:
292 case nir_intrinsic_ssbo_atomic_and
:
293 case nir_intrinsic_ssbo_atomic_or
:
294 case nir_intrinsic_ssbo_atomic_xor
:
295 case nir_intrinsic_ssbo_atomic_exchange
:
296 case nir_intrinsic_ssbo_atomic_comp_swap
:
297 if (nir
->info
.stage
== MESA_SHADER_FRAGMENT
)
298 info
->ps
.writes_memory
= true;
300 case nir_intrinsic_load_deref
:
301 gather_intrinsic_load_deref_info(nir
, instr
, info
);
303 case nir_intrinsic_store_deref
:
304 gather_intrinsic_store_deref_info(nir
, instr
, info
);
312 gather_tex_info(const nir_shader
*nir
, const nir_tex_instr
*instr
,
313 struct radv_shader_info
*info
)
315 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++) {
316 switch (instr
->src
[i
].src_type
) {
317 case nir_tex_src_texture_deref
:
318 mark_sampler_desc(nir_deref_instr_get_variable(nir_src_as_deref(instr
->src
[i
].src
)), info
);
320 case nir_tex_src_sampler_deref
:
321 mark_sampler_desc(nir_deref_instr_get_variable(nir_src_as_deref(instr
->src
[i
].src
)), info
);
330 gather_info_block(const nir_shader
*nir
, const nir_block
*block
,
331 struct radv_shader_info
*info
)
333 nir_foreach_instr(instr
, block
) {
334 switch (instr
->type
) {
335 case nir_instr_type_intrinsic
:
336 gather_intrinsic_info(nir
, nir_instr_as_intrinsic(instr
), info
);
338 case nir_instr_type_tex
:
339 gather_tex_info(nir
, nir_instr_as_tex(instr
), info
);
348 gather_info_input_decl_vs(const nir_shader
*nir
, const nir_variable
*var
,
349 struct radv_shader_info
*info
)
351 int idx
= var
->data
.location
;
353 if (idx
>= VERT_ATTRIB_GENERIC0
&& idx
<= VERT_ATTRIB_GENERIC15
)
354 info
->vs
.has_vertex_buffers
= true;
358 gather_info_input_decl_ps(const nir_shader
*nir
, const nir_variable
*var
,
359 struct radv_shader_info
*info
)
361 unsigned attrib_count
= glsl_count_attribute_slots(var
->type
, false);
362 const struct glsl_type
*type
= glsl_without_array(var
->type
);
363 int idx
= var
->data
.location
;
366 case VARYING_SLOT_PNTC
:
367 info
->ps
.has_pcoord
= true;
369 case VARYING_SLOT_PRIMITIVE_ID
:
370 info
->ps
.prim_id_input
= true;
372 case VARYING_SLOT_LAYER
:
373 info
->ps
.layer_input
= true;
375 case VARYING_SLOT_CLIP_DIST0
:
376 info
->ps
.num_input_clips_culls
= attrib_count
;
382 if (glsl_get_base_type(type
) == GLSL_TYPE_FLOAT
) {
383 if (var
->data
.sample
)
384 info
->ps
.force_persample
= true;
389 gather_info_input_decl(const nir_shader
*nir
, const nir_variable
*var
,
390 struct radv_shader_info
*info
)
392 switch (nir
->info
.stage
) {
393 case MESA_SHADER_VERTEX
:
394 gather_info_input_decl_vs(nir
, var
, info
);
396 case MESA_SHADER_FRAGMENT
:
397 gather_info_input_decl_ps(nir
, var
, info
);
405 gather_info_output_decl_ls(const nir_shader
*nir
, const nir_variable
*var
,
406 struct radv_shader_info
*info
)
408 int idx
= var
->data
.location
;
409 unsigned param
= shader_io_get_unique_index(idx
);
410 int num_slots
= glsl_count_attribute_slots(var
->type
, false);
411 if (idx
== VARYING_SLOT_CLIP_DIST0
)
412 num_slots
= (nir
->info
.clip_distance_array_size
+ nir
->info
.cull_distance_array_size
> 4) ? 2 : 1;
413 mark_ls_output(info
, param
, num_slots
);
417 gather_info_output_decl_ps(const nir_shader
*nir
, const nir_variable
*var
,
418 struct radv_shader_info
*info
)
420 int idx
= var
->data
.location
;
423 case FRAG_RESULT_DEPTH
:
424 info
->ps
.writes_z
= true;
426 case FRAG_RESULT_STENCIL
:
427 info
->ps
.writes_stencil
= true;
429 case FRAG_RESULT_SAMPLE_MASK
:
430 info
->ps
.writes_sample_mask
= true;
438 gather_info_output_decl_gs(const nir_shader
*nir
, const nir_variable
*var
,
439 struct radv_shader_info
*info
)
441 unsigned num_components
= glsl_get_component_slots(var
->type
);
442 unsigned stream
= var
->data
.stream
;
446 info
->gs
.max_stream
= MAX2(info
->gs
.max_stream
, stream
);
447 info
->gs
.num_stream_output_components
[stream
] += num_components
;
451 gather_info_output_decl(const nir_shader
*nir
, const nir_variable
*var
,
452 struct radv_shader_info
*info
,
453 const struct radv_nir_compiler_options
*options
)
455 switch (nir
->info
.stage
) {
456 case MESA_SHADER_FRAGMENT
:
457 gather_info_output_decl_ps(nir
, var
, info
);
459 case MESA_SHADER_VERTEX
:
460 if (options
->key
.vs
.as_ls
)
461 gather_info_output_decl_ls(nir
, var
, info
);
463 case MESA_SHADER_GEOMETRY
:
464 gather_info_output_decl_gs(nir
, var
, info
);
472 radv_nir_shader_info_pass(const struct nir_shader
*nir
,
473 const struct radv_nir_compiler_options
*options
,
474 struct radv_shader_info
*info
)
476 struct nir_function
*func
=
477 (struct nir_function
*)exec_list_get_head_const(&nir
->functions
);
479 if (options
->layout
&& options
->layout
->dynamic_offset_count
)
480 info
->loads_push_constants
= true;
482 nir_foreach_variable(variable
, &nir
->inputs
)
483 gather_info_input_decl(nir
, variable
, info
);
485 nir_foreach_block(block
, func
->impl
) {
486 gather_info_block(nir
, block
, info
);
489 nir_foreach_variable(variable
, &nir
->outputs
)
490 gather_info_output_decl(nir
, variable
, info
, options
);