summaryrefslogtreecommitdiff
path: root/src/compiler/glsl/gl_nir_lower_samplers_as_deref.c
blob: 934369d74fe424a4c8459e76c147ce4d97376cb3 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
/*
 * Copyright (C) 2005-2007  Brian Paul   All Rights Reserved.
 * Copyright (C) 2008  VMware, Inc.   All Rights Reserved.
 * Copyright © 2014 Intel Corporation
 * Copyright © 2017 Advanced Micro Devices, Inc.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
 * DEALINGS IN THE SOFTWARE.
 */

/**
 * \file
 *
 * Lower sampler and image references of (non-bindless) uniforms by removing
 * struct dereferences, and synthesizing new uniform variables without structs
 * if required.
 *
 * This will allow backends to have a simple, uniform treatment of bindless and
 * non-bindless samplers and images.
 *
 * Example:
 *
 *   struct S {
 *      sampler2D tex[2];
 *      sampler2D other;
 *   };
 *   uniform S s[2];
 *
 *   tmp = texture(s[n].tex[m], coord);
 *
 * Becomes:
 *
 *   decl_var uniform INTERP_MODE_NONE sampler2D[2][2] lower@s.tex (...)
 *
 *   vec1 32 ssa_idx = $(2 * n + m)
 *   vec4 32 ssa_out = tex ssa_coord (coord), lower@s.tex[n][m] (texture), lower@s.tex[n][m] (sampler)
 *
 * and lower@s.tex has var->data.binding set to the base index as defined by
 * the opaque uniform mapping.
 */

#include "compiler/nir/nir.h"
#include "compiler/nir/nir_builder.h"
#include "compiler/nir/nir_deref.h"
#include "gl_nir.h"
#include "ir_uniform.h"

#include "util/compiler.h"
#include "main/shader_types.h"

struct lower_samplers_as_deref_state {
   nir_shader *shader;
   const struct gl_shader_program *shader_program;
   struct hash_table *remap_table;
};

/* Prepare for removing struct derefs.  This pre-pass generates the name
 * of the lowered deref, and calculates the lowered type and location.
 * After that, once looking up (or creating if needed) the lowered var,
 * constructing the new chain of deref instructions is a simple loop
 * that skips the struct deref's
 *
 * path:     appended to as we descend down the chain of deref instrs
 *           and remove struct derefs
 * location: increased as we descend down and remove struct derefs
 * type:     updated as we recurse back up the chain of deref instrs
 *           with the resulting type after removing struct derefs
 */
static void
remove_struct_derefs_prep(nir_deref_instr **p, char **name,
                          unsigned *location, const struct glsl_type **type)
{
   nir_deref_instr *cur = p[0], *next = p[1];

   if (!next) {
      *type = cur->type;
      return;
   }

   switch (next->deref_type) {
   case nir_deref_type_array: {
      unsigned length = glsl_get_length(cur->type);

      remove_struct_derefs_prep(&p[1], name, location, type);

      *type = glsl_array_type(*type, length, glsl_get_explicit_stride(cur->type));
      break;
   }

   case nir_deref_type_struct: {
      *location += glsl_get_struct_location_offset(cur->type, next->strct.index);
      ralloc_asprintf_append(name, ".%s",
                             glsl_get_struct_elem_name(cur->type, next->strct.index));

      remove_struct_derefs_prep(&p[1], name, location, type);
      break;
   }

   default:
      unreachable("Invalid deref type");
      break;
   }
}

static void
record_images_used(struct shader_info *info,
                   nir_intrinsic_instr *instr)
{
   nir_variable *var =
      nir_deref_instr_get_variable(nir_src_as_deref(instr->src[0]));

   /* Structs have been lowered already, so get_aoa_size is sufficient. */
   const unsigned size =
      glsl_type_is_array(var->type) ? glsl_get_aoa_size(var->type) : 1;

   BITSET_SET_RANGE(info->images_used, var->data.binding,
                    var->data.binding + (MAX2(size, 1) - 1));

   enum glsl_sampler_dim sampler_dim =
      glsl_get_sampler_dim(glsl_without_array(var->type));
   if (sampler_dim == GLSL_SAMPLER_DIM_BUF) {
      BITSET_SET_RANGE(info->image_buffers, var->data.binding,
                       var->data.binding + (MAX2(size, 1) - 1));
   }
   if (sampler_dim == GLSL_SAMPLER_DIM_MS) {
      BITSET_SET_RANGE(info->msaa_images, var->data.binding,
                       var->data.binding + (MAX2(size, 1) - 1));
   }
}


static nir_deref_instr *
lower_deref(nir_builder *b, struct lower_samplers_as_deref_state *state,
            nir_deref_instr *deref)
{
   nir_variable *var = nir_deref_instr_get_variable(deref);
   gl_shader_stage stage = state->shader->info.stage;

   if (!(var->data.mode & (nir_var_uniform | nir_var_image)) ||
       var->data.bindless)
      return NULL;

   nir_deref_path path;
   nir_deref_path_init(&path, deref, state->remap_table);
   assert(path.path[0]->deref_type == nir_deref_type_var);

   char *name = ralloc_asprintf(state->remap_table, "lower@%s", var->name);
   unsigned location = var->data.location;
   const struct glsl_type *type = NULL;
   unsigned binding;

   /*
    * We end up needing to do this in two passes, in order to generate
    * the name of the lowered var (and detecting whether there even are
    * any struct deref's), and then the second pass to construct the
    * actual deref instructions after looking up / generating a new
    * nir_variable (since we need to construct the deref_var first)
    */

   remove_struct_derefs_prep(path.path, &name, &location, &type);

   if (state->shader_program && var->data.how_declared != nir_var_hidden) {
      /* For GLSL programs, look up the bindings in the uniform storage. */
      assert(location < state->shader_program->data->NumUniformStorage &&
             state->shader_program->data->UniformStorage[location].opaque[stage].active);

      binding = state->shader_program->data->UniformStorage[location].opaque[stage].index;
   } else {
      /* For ARB programs, built-in shaders, or internally generated sampler
       * variables in GLSL programs, assume that whoever created the shader
       * set the bindings correctly already.
       */
      assert(var->data.explicit_binding);
      binding = var->data.binding;
   }

   if (var->type == type) {
      /* Fast path: We did not encounter any struct derefs. */
      var->data.binding = binding;
      return deref;
   }

   uint32_t hash = _mesa_hash_string(name);
   struct hash_entry *h =
      _mesa_hash_table_search_pre_hashed(state->remap_table, hash, name);

   if (h) {
      var = (nir_variable *)h->data;
   } else {
      var = nir_variable_create(state->shader, var->data.mode, type, name);
      var->data.binding = binding;

      /* Don't set var->data.location.  The old structure location could be
       * used to index into gl_uniform_storage, assuming the full structure
       * was walked in order.  With the new split variables, this invariant
       * no longer holds and there's no meaningful way to start from a base
       * location and access a particular array element.  Just leave it 0.
       */

      _mesa_hash_table_insert_pre_hashed(state->remap_table, hash, name, var);
   }

   /* construct a new deref based on lowered var (skipping the struct deref's
    * from the original deref:
    */
   nir_deref_instr *new_deref = nir_build_deref_var(b, var);
   for (nir_deref_instr **p = &path.path[1]; *p; p++) {
      if ((*p)->deref_type == nir_deref_type_struct)
         continue;

      assert((*p)->deref_type == nir_deref_type_array);

      new_deref = nir_build_deref_array(b, new_deref,
                                        nir_ssa_for_src(b, (*p)->arr.index, 1));
   }

   return new_deref;
}

static void
record_textures_used(struct shader_info *info,
                     nir_deref_instr *deref,
                     nir_texop op)
{
   nir_variable *var = nir_deref_instr_get_variable(deref);

   /* Structs have been lowered already, so get_aoa_size is sufficient. */
   const unsigned size =
      glsl_type_is_array(var->type) ? glsl_get_aoa_size(var->type) : 1;

   BITSET_SET_RANGE(info->textures_used, var->data.binding,
                    var->data.binding + (MAX2(size, 1) - 1));

   if (op == nir_texop_txf ||
       op == nir_texop_txf_ms ||
       op == nir_texop_txf_ms_mcs_intel) {
      BITSET_SET_RANGE(info->textures_used_by_txf, var->data.binding,
                       var->data.binding + (MAX2(size, 1) - 1));
   }
}

static void
record_samplers_used(struct shader_info *info,
                     nir_deref_instr *deref,
                     nir_texop op)
{
   nir_variable *var = nir_deref_instr_get_variable(deref);

   /* Structs have been lowered already, so get_aoa_size is sufficient. */
   const unsigned size =
      glsl_type_is_array(var->type) ? glsl_get_aoa_size(var->type) : 1;

   BITSET_SET_RANGE(info->samplers_used, var->data.binding,
                    var->data.binding + (MAX2(size, 1) - 1));
}

static bool
lower_sampler(nir_tex_instr *instr, struct lower_samplers_as_deref_state *state,
              nir_builder *b)
{
   int texture_idx =
      nir_tex_instr_src_index(instr, nir_tex_src_texture_deref);
   int sampler_idx =
      nir_tex_instr_src_index(instr, nir_tex_src_sampler_deref);

   b->cursor = nir_before_instr(&instr->instr);

   if (texture_idx >= 0) {
      assert(instr->src[texture_idx].src.is_ssa);

      nir_deref_instr *texture_deref =
         lower_deref(b, state, nir_src_as_deref(instr->src[texture_idx].src));
      /* only lower non-bindless: */
      if (texture_deref) {
         nir_instr_rewrite_src(&instr->instr, &instr->src[texture_idx].src,
                               nir_src_for_ssa(&texture_deref->dest.ssa));
         record_textures_used(&b->shader->info, texture_deref, instr->op);
      }
   }

   if (sampler_idx >= 0) {
      assert(instr->src[sampler_idx].src.is_ssa);
      nir_deref_instr *sampler_deref =
         lower_deref(b, state, nir_src_as_deref(instr->src[sampler_idx].src));
      /* only lower non-bindless: */
      if (sampler_deref) {
         nir_instr_rewrite_src(&instr->instr, &instr->src[sampler_idx].src,
                               nir_src_for_ssa(&sampler_deref->dest.ssa));
         record_samplers_used(&b->shader->info, sampler_deref, instr->op);
      }
   }

   return true;
}

static bool
lower_intrinsic(nir_intrinsic_instr *instr,
                struct lower_samplers_as_deref_state *state,
                nir_builder *b)
{
   if (instr->intrinsic == nir_intrinsic_image_deref_load ||
       instr->intrinsic == nir_intrinsic_image_deref_store ||
       instr->intrinsic == nir_intrinsic_image_deref_atomic_add ||
       instr->intrinsic == nir_intrinsic_image_deref_atomic_imin ||
       instr->intrinsic == nir_intrinsic_image_deref_atomic_umin ||
       instr->intrinsic == nir_intrinsic_image_deref_atomic_imax ||
       instr->intrinsic == nir_intrinsic_image_deref_atomic_umax ||
       instr->intrinsic == nir_intrinsic_image_deref_atomic_and ||
       instr->intrinsic == nir_intrinsic_image_deref_atomic_or ||
       instr->intrinsic == nir_intrinsic_image_deref_atomic_xor ||
       instr->intrinsic == nir_intrinsic_image_deref_atomic_exchange ||
       instr->intrinsic == nir_intrinsic_image_deref_atomic_comp_swap ||
       instr->intrinsic == nir_intrinsic_image_deref_atomic_fadd ||
       instr->intrinsic == nir_intrinsic_image_deref_size ||
       instr->intrinsic == nir_intrinsic_image_deref_samples_identical ||
       instr->intrinsic == nir_intrinsic_image_deref_descriptor_amd ||
       instr->intrinsic == nir_intrinsic_image_deref_samples) {

      b->cursor = nir_before_instr(&instr->instr);
      nir_deref_instr *deref =
         lower_deref(b, state, nir_src_as_deref(instr->src[0]));

      record_images_used(&state->shader->info, instr);

      /* don't lower bindless: */
      if (!deref)
         return false;
      nir_instr_rewrite_src(&instr->instr, &instr->src[0],
                            nir_src_for_ssa(&deref->dest.ssa));
      return true;
   }
   if (instr->intrinsic == nir_intrinsic_image_deref_order ||
       instr->intrinsic == nir_intrinsic_image_deref_format)
      unreachable("how did you even manage this?");

   return false;
}

static bool
lower_instr(nir_builder *b, nir_instr *instr, void *cb_data)
{
   struct lower_samplers_as_deref_state *state = cb_data;

   if (instr->type == nir_instr_type_tex)
      return lower_sampler(nir_instr_as_tex(instr), state, b);

   if (instr->type == nir_instr_type_intrinsic)
      return lower_intrinsic(nir_instr_as_intrinsic(instr), state, b);

   return false;
}

bool
gl_nir_lower_samplers_as_deref(nir_shader *shader,
                               const struct gl_shader_program *shader_program)
{
   struct lower_samplers_as_deref_state state;

   state.shader = shader;
   state.shader_program = shader_program;
   state.remap_table = _mesa_hash_table_create(NULL, _mesa_hash_string,
                                               _mesa_key_string_equal);

   bool progress = nir_shader_instructions_pass(shader, lower_instr,
                                                nir_metadata_block_index |
                                                nir_metadata_dominance,
                                                &state);

   if (progress) {
      nir_remove_dead_derefs(shader);
      if (!shader->info.internal && shader_program) {
         /* try to apply bindings for unused samplers to avoid index zero clobbering in backends */
         nir_foreach_uniform_variable(var, shader) {
            /* ignore hidden variables */
            if (!glsl_type_is_sampler(glsl_without_array(var->type)) ||
                var->data.how_declared == nir_var_hidden)
               continue;
            bool found = false;
            hash_table_foreach(state.remap_table, entry) {
               if (var == entry->data) {
                  found = true;
                  break;
               }
            }
            if (!found) {
               /* same as lower_deref() */
               var->data.binding = shader_program->data->UniformStorage[var->data.location].opaque[shader->info.stage].index;
            }
         }
      }
   }

   /* keys are freed automatically by ralloc */
   _mesa_hash_table_destroy(state.remap_table, NULL);

   return progress;
}