1 /*
2 * Copyright (C) 2005-2007 Brian Paul All Rights Reserved.
3 * Copyright (C) 2008 VMware, Inc. All Rights Reserved.
4 * Copyright © 2014 Intel Corporation
5 * Copyright © 2017 Advanced Micro Devices, Inc.
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the "Software"),
9 * to deal in the Software without restriction, including without limitation
10 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
11 * and/or sell copies of the Software, and to permit persons to whom the
12 * Software is furnished to do so, subject to the following conditions:
13 *
14 * The above copyright notice and this permission notice (including the next
15 * paragraph) shall be included in all copies or substantial portions of the
16 * Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
23 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
24 * DEALINGS IN THE SOFTWARE.
25 */
26
27 /**
28 * \file
29 *
30 * Lower sampler and image references of (non-bindless) uniforms by removing
31 * struct dereferences, and synthesizing new uniform variables without structs
32 * if required.
33 *
34 * This will allow backends to have a simple, uniform treatment of bindless and
35 * non-bindless samplers and images.
36 *
37 * Example:
38 *
39 * struct S {
40 * sampler2D tex[2];
41 * sampler2D other;
42 * };
43 * uniform S s[2];
44 *
45 * tmp = texture(s[n].tex[m], coord);
46 *
47 * Becomes:
48 *
49 * decl_var uniform INTERP_MODE_NONE sampler2D[2][2] lower@s.tex (...)
50 *
51 * vec1 32 ssa_idx = $(2 * n + m)
52 * vec4 32 ssa_out = tex ssa_coord (coord), lower@s.tex[n][m] (texture), lower@s.tex[n][m] (sampler)
53 *
54 * and lower@s.tex has var->data.binding set to the base index as defined by
55 * the opaque uniform mapping.
56 */
57
58 #include "compiler/nir/nir.h"
59 #include "compiler/nir/nir_builder.h"
60 #include "compiler/nir/nir_deref.h"
61 #include "gl_nir.h"
62 #include "ir_uniform.h"
63
64 #include "util/compiler.h"
65 #include "main/shader_types.h"
66
67 struct lower_samplers_as_deref_state {
68 nir_shader *shader;
69 const struct gl_shader_program *shader_program;
70 struct hash_table *remap_table;
71 };
72
73 /* Prepare for removing struct derefs. This pre-pass generates the name
74 * of the lowered deref, and calculates the lowered type and location.
75 * After that, once looking up (or creating if needed) the lowered var,
76 * constructing the new chain of deref instructions is a simple loop
77 * that skips the struct deref's
78 *
79 * path: appended to as we descend down the chain of deref instrs
80 * and remove struct derefs
81 * location: increased as we descend down and remove struct derefs
82 * type: updated as we recurse back up the chain of deref instrs
83 * with the resulting type after removing struct derefs
84 */
85 static void
remove_struct_derefs_prep(nir_deref_instr ** p,char ** name,unsigned * location,const struct glsl_type ** type)86 remove_struct_derefs_prep(nir_deref_instr **p, char **name,
87 unsigned *location, const struct glsl_type **type)
88 {
89 nir_deref_instr *cur = p[0], *next = p[1];
90
91 if (!next) {
92 *type = cur->type;
93 return;
94 }
95
96 switch (next->deref_type) {
97 case nir_deref_type_array: {
98 unsigned length = glsl_get_length(cur->type);
99
100 remove_struct_derefs_prep(&p[1], name, location, type);
101
102 *type = glsl_array_type(*type, length, glsl_get_explicit_stride(cur->type));
103 break;
104 }
105
106 case nir_deref_type_struct: {
107 *location += glsl_get_struct_location_offset(cur->type, next->strct.index);
108 ralloc_asprintf_append(name, ".%s",
109 glsl_get_struct_elem_name(cur->type, next->strct.index));
110
111 remove_struct_derefs_prep(&p[1], name, location, type);
112 break;
113 }
114
115 default:
116 unreachable("Invalid deref type");
117 break;
118 }
119 }
120
121 static void
record_images_used(struct shader_info * info,nir_intrinsic_instr * instr)122 record_images_used(struct shader_info *info,
123 nir_intrinsic_instr *instr)
124 {
125 nir_variable *var =
126 nir_deref_instr_get_variable(nir_src_as_deref(instr->src[0]));
127
128 /* Structs have been lowered already, so get_aoa_size is sufficient. */
129 const unsigned size =
130 glsl_type_is_array(var->type) ? glsl_get_aoa_size(var->type) : 1;
131
132 BITSET_SET_RANGE(info->images_used, var->data.binding,
133 var->data.binding + (MAX2(size, 1) - 1));
134
135 enum glsl_sampler_dim sampler_dim =
136 glsl_get_sampler_dim(glsl_without_array(var->type));
137 if (sampler_dim == GLSL_SAMPLER_DIM_BUF) {
138 BITSET_SET_RANGE(info->image_buffers, var->data.binding,
139 var->data.binding + (MAX2(size, 1) - 1));
140 }
141 if (sampler_dim == GLSL_SAMPLER_DIM_MS) {
142 BITSET_SET_RANGE(info->msaa_images, var->data.binding,
143 var->data.binding + (MAX2(size, 1) - 1));
144 }
145 }
146
147
148 static nir_deref_instr *
lower_deref(nir_builder * b,struct lower_samplers_as_deref_state * state,nir_deref_instr * deref)149 lower_deref(nir_builder *b, struct lower_samplers_as_deref_state *state,
150 nir_deref_instr *deref)
151 {
152 nir_variable *var = nir_deref_instr_get_variable(deref);
153 gl_shader_stage stage = state->shader->info.stage;
154
155 if (!(var->data.mode & (nir_var_uniform | nir_var_image)) ||
156 var->data.bindless)
157 return NULL;
158
159 nir_deref_path path;
160 nir_deref_path_init(&path, deref, state->remap_table);
161 assert(path.path[0]->deref_type == nir_deref_type_var);
162
163 char *name = ralloc_asprintf(state->remap_table, "lower@%s", var->name);
164 unsigned location = var->data.location;
165 const struct glsl_type *type = NULL;
166 unsigned binding;
167
168 /*
169 * We end up needing to do this in two passes, in order to generate
170 * the name of the lowered var (and detecting whether there even are
171 * any struct deref's), and then the second pass to construct the
172 * actual deref instructions after looking up / generating a new
173 * nir_variable (since we need to construct the deref_var first)
174 */
175
176 remove_struct_derefs_prep(path.path, &name, &location, &type);
177
178 if (state->shader_program && var->data.how_declared != nir_var_hidden) {
179 /* For GLSL programs, look up the bindings in the uniform storage. */
180 assert(location < state->shader_program->data->NumUniformStorage &&
181 state->shader_program->data->UniformStorage[location].opaque[stage].active);
182
183 binding = state->shader_program->data->UniformStorage[location].opaque[stage].index;
184 } else {
185 /* For ARB programs, built-in shaders, or internally generated sampler
186 * variables in GLSL programs, assume that whoever created the shader
187 * set the bindings correctly already.
188 */
189 assert(var->data.explicit_binding);
190 binding = var->data.binding;
191 }
192
193 if (var->type == type) {
194 /* Fast path: We did not encounter any struct derefs. */
195 var->data.binding = binding;
196 return deref;
197 }
198
199 uint32_t hash = _mesa_hash_string(name);
200 struct hash_entry *h =
201 _mesa_hash_table_search_pre_hashed(state->remap_table, hash, name);
202
203 if (h) {
204 var = (nir_variable *)h->data;
205 } else {
206 var = nir_variable_create(state->shader, var->data.mode, type, name);
207 var->data.binding = binding;
208
209 /* Don't set var->data.location. The old structure location could be
210 * used to index into gl_uniform_storage, assuming the full structure
211 * was walked in order. With the new split variables, this invariant
212 * no longer holds and there's no meaningful way to start from a base
213 * location and access a particular array element. Just leave it 0.
214 */
215
216 _mesa_hash_table_insert_pre_hashed(state->remap_table, hash, name, var);
217 }
218
219 /* construct a new deref based on lowered var (skipping the struct deref's
220 * from the original deref:
221 */
222 nir_deref_instr *new_deref = nir_build_deref_var(b, var);
223 for (nir_deref_instr **p = &path.path[1]; *p; p++) {
224 if ((*p)->deref_type == nir_deref_type_struct)
225 continue;
226
227 assert((*p)->deref_type == nir_deref_type_array);
228
229 new_deref = nir_build_deref_array(b, new_deref,
230 nir_ssa_for_src(b, (*p)->arr.index, 1));
231 }
232
233 return new_deref;
234 }
235
236 static void
record_textures_used(struct shader_info * info,nir_deref_instr * deref,nir_texop op)237 record_textures_used(struct shader_info *info,
238 nir_deref_instr *deref,
239 nir_texop op)
240 {
241 nir_variable *var = nir_deref_instr_get_variable(deref);
242
243 /* Structs have been lowered already, so get_aoa_size is sufficient. */
244 const unsigned size =
245 glsl_type_is_array(var->type) ? glsl_get_aoa_size(var->type) : 1;
246
247 BITSET_SET_RANGE(info->textures_used, var->data.binding,
248 var->data.binding + (MAX2(size, 1) - 1));
249
250 if (op == nir_texop_txf ||
251 op == nir_texop_txf_ms ||
252 op == nir_texop_txf_ms_mcs_intel) {
253 BITSET_SET_RANGE(info->textures_used_by_txf, var->data.binding,
254 var->data.binding + (MAX2(size, 1) - 1));
255 }
256 }
257
258 static void
record_samplers_used(struct shader_info * info,nir_deref_instr * deref,nir_texop op)259 record_samplers_used(struct shader_info *info,
260 nir_deref_instr *deref,
261 nir_texop op)
262 {
263 nir_variable *var = nir_deref_instr_get_variable(deref);
264
265 /* Structs have been lowered already, so get_aoa_size is sufficient. */
266 const unsigned size =
267 glsl_type_is_array(var->type) ? glsl_get_aoa_size(var->type) : 1;
268
269 BITSET_SET_RANGE(info->samplers_used, var->data.binding,
270 var->data.binding + (MAX2(size, 1) - 1));
271 }
272
273 static bool
lower_sampler(nir_tex_instr * instr,struct lower_samplers_as_deref_state * state,nir_builder * b)274 lower_sampler(nir_tex_instr *instr, struct lower_samplers_as_deref_state *state,
275 nir_builder *b)
276 {
277 int texture_idx =
278 nir_tex_instr_src_index(instr, nir_tex_src_texture_deref);
279 int sampler_idx =
280 nir_tex_instr_src_index(instr, nir_tex_src_sampler_deref);
281
282 b->cursor = nir_before_instr(&instr->instr);
283
284 if (texture_idx >= 0) {
285 assert(instr->src[texture_idx].src.is_ssa);
286
287 nir_deref_instr *texture_deref =
288 lower_deref(b, state, nir_src_as_deref(instr->src[texture_idx].src));
289 /* only lower non-bindless: */
290 if (texture_deref) {
291 nir_instr_rewrite_src(&instr->instr, &instr->src[texture_idx].src,
292 nir_src_for_ssa(&texture_deref->dest.ssa));
293 record_textures_used(&b->shader->info, texture_deref, instr->op);
294 }
295 }
296
297 if (sampler_idx >= 0) {
298 assert(instr->src[sampler_idx].src.is_ssa);
299 nir_deref_instr *sampler_deref =
300 lower_deref(b, state, nir_src_as_deref(instr->src[sampler_idx].src));
301 /* only lower non-bindless: */
302 if (sampler_deref) {
303 nir_instr_rewrite_src(&instr->instr, &instr->src[sampler_idx].src,
304 nir_src_for_ssa(&sampler_deref->dest.ssa));
305 record_samplers_used(&b->shader->info, sampler_deref, instr->op);
306 }
307 }
308
309 return true;
310 }
311
312 static bool
lower_intrinsic(nir_intrinsic_instr * instr,struct lower_samplers_as_deref_state * state,nir_builder * b)313 lower_intrinsic(nir_intrinsic_instr *instr,
314 struct lower_samplers_as_deref_state *state,
315 nir_builder *b)
316 {
317 if (instr->intrinsic == nir_intrinsic_image_deref_load ||
318 instr->intrinsic == nir_intrinsic_image_deref_store ||
319 instr->intrinsic == nir_intrinsic_image_deref_atomic_add ||
320 instr->intrinsic == nir_intrinsic_image_deref_atomic_imin ||
321 instr->intrinsic == nir_intrinsic_image_deref_atomic_umin ||
322 instr->intrinsic == nir_intrinsic_image_deref_atomic_imax ||
323 instr->intrinsic == nir_intrinsic_image_deref_atomic_umax ||
324 instr->intrinsic == nir_intrinsic_image_deref_atomic_and ||
325 instr->intrinsic == nir_intrinsic_image_deref_atomic_or ||
326 instr->intrinsic == nir_intrinsic_image_deref_atomic_xor ||
327 instr->intrinsic == nir_intrinsic_image_deref_atomic_exchange ||
328 instr->intrinsic == nir_intrinsic_image_deref_atomic_comp_swap ||
329 instr->intrinsic == nir_intrinsic_image_deref_atomic_fadd ||
330 instr->intrinsic == nir_intrinsic_image_deref_size ||
331 instr->intrinsic == nir_intrinsic_image_deref_samples) {
332
333 b->cursor = nir_before_instr(&instr->instr);
334 nir_deref_instr *deref =
335 lower_deref(b, state, nir_src_as_deref(instr->src[0]));
336
337 record_images_used(&state->shader->info, instr);
338
339 /* don't lower bindless: */
340 if (!deref)
341 return false;
342 nir_instr_rewrite_src(&instr->instr, &instr->src[0],
343 nir_src_for_ssa(&deref->dest.ssa));
344 return true;
345 }
346 if (instr->intrinsic == nir_intrinsic_image_deref_order ||
347 instr->intrinsic == nir_intrinsic_image_deref_format)
348 unreachable("how did you even manage this?");
349
350 return false;
351 }
352
353 static bool
lower_impl(nir_function_impl * impl,struct lower_samplers_as_deref_state * state)354 lower_impl(nir_function_impl *impl, struct lower_samplers_as_deref_state *state)
355 {
356 nir_builder b;
357 nir_builder_init(&b, impl);
358 bool progress = false;
359
360 nir_foreach_block(block, impl) {
361 nir_foreach_instr(instr, block) {
362 if (instr->type == nir_instr_type_tex)
363 progress |= lower_sampler(nir_instr_as_tex(instr), state, &b);
364 else if (instr->type == nir_instr_type_intrinsic)
365 progress |= lower_intrinsic(nir_instr_as_intrinsic(instr), state, &b);
366 }
367 }
368
369 if (progress) {
370 nir_metadata_preserve(impl, nir_metadata_block_index |
371 nir_metadata_dominance);
372 } else {
373 nir_metadata_preserve(impl, nir_metadata_all);
374 }
375
376 return progress;
377 }
378
379 bool
gl_nir_lower_samplers_as_deref(nir_shader * shader,const struct gl_shader_program * shader_program)380 gl_nir_lower_samplers_as_deref(nir_shader *shader,
381 const struct gl_shader_program *shader_program)
382 {
383 bool progress = false;
384 struct lower_samplers_as_deref_state state;
385
386 state.shader = shader;
387 state.shader_program = shader_program;
388 state.remap_table = _mesa_hash_table_create(NULL, _mesa_hash_string,
389 _mesa_key_string_equal);
390
391 nir_foreach_function(function, shader) {
392 if (function->impl)
393 progress |= lower_impl(function->impl, &state);
394 }
395
396 /* keys are freed automatically by ralloc */
397 _mesa_hash_table_destroy(state.remap_table, NULL);
398
399 if (progress)
400 nir_remove_dead_derefs(shader);
401
402 return progress;
403 }
404