src/compiler/nir/nir_lower_atomics_to_ssbo.c

   1 /*
   2  * Copyright © 2017 Red Hat
   3  *
   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:
  10  *
  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
  13  * Software.
  14  *
  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
  21  * IN THE SOFTWARE.
  22  *
  23  * Authors:
  24  *    Rob Clark <robclark@freedesktop.org>
  25  */
  26
  27 #include "nir.h"
  28 #include "nir_builder.h"
  29
  30 #if defined(_WIN32) && !defined(snprintf)
  31 #define snprintf _snprintf
  32 #endif
  33
  34 /*
  35  * Remap atomic counters to SSBOs.  Atomic counters get remapped to
  36  * SSBO binding points [0..ssbo_offset) and the original SSBOs are
  37  * remapped to [ssbo_offset..n) (mostly to align with what mesa/st
  38  * does.
  39  */
  40
  41 static bool
  42 lower_instr(nir_intrinsic_instr *instr, unsigned ssbo_offset, nir_builder *b)
  43 {
  44    nir_intrinsic_op op;
  45    int idx_src;
  46
  47    b->cursor = nir_before_instr(&instr->instr);
  48
  49    switch (instr->intrinsic) {
  50    case nir_intrinsic_memory_barrier_atomic_counter:
  51       /* Atomic counters are now SSBOs so memoryBarrierAtomicCounter() is now
  52        * memoryBarrierBuffer().
  53        */
  54       instr->intrinsic = nir_intrinsic_memory_barrier_buffer;
  55       return true;
  56
  57    case nir_intrinsic_ssbo_atomic_add:
  58    case nir_intrinsic_ssbo_atomic_imin:
  59    case nir_intrinsic_ssbo_atomic_umin:
  60    case nir_intrinsic_ssbo_atomic_imax:
  61    case nir_intrinsic_ssbo_atomic_umax:
  62    case nir_intrinsic_ssbo_atomic_and:
  63    case nir_intrinsic_ssbo_atomic_or:
  64    case nir_intrinsic_ssbo_atomic_xor:
  65    case nir_intrinsic_ssbo_atomic_exchange:
  66    case nir_intrinsic_ssbo_atomic_comp_swap:
  67    case nir_intrinsic_ssbo_atomic_fadd:
  68    case nir_intrinsic_ssbo_atomic_fmin:
  69    case nir_intrinsic_ssbo_atomic_fmax:
  70    case nir_intrinsic_ssbo_atomic_fcomp_swap:
  71    case nir_intrinsic_store_ssbo:
  72    case nir_intrinsic_load_ssbo:
  73    case nir_intrinsic_get_buffer_size:
  74       /* easy case, keep same opcode and just remap SSBO buffer index: */
  75       op = instr->intrinsic;
  76       idx_src = (op == nir_intrinsic_store_ssbo) ? 1 : 0;
  77       nir_ssa_def *old_idx = nir_ssa_for_src(b, instr->src[idx_src], 1);
  78       nir_ssa_def *new_idx = nir_iadd(b, old_idx, nir_imm_int(b, ssbo_offset));
  79       nir_instr_rewrite_src(&instr->instr,
  80                             &instr->src[idx_src],
  81                             nir_src_for_ssa(new_idx));
  82       return true;
  83    case nir_intrinsic_atomic_counter_inc:
  84    case nir_intrinsic_atomic_counter_add:
  85    case nir_intrinsic_atomic_counter_pre_dec:
  86    case nir_intrinsic_atomic_counter_post_dec:
  87       /* inc and dec get remapped to add: */
  88       op = nir_intrinsic_ssbo_atomic_add;
  89       break;
  90    case nir_intrinsic_atomic_counter_read:
  91       op = nir_intrinsic_load_ssbo;
  92       break;
  93    case nir_intrinsic_atomic_counter_min:
  94       op = nir_intrinsic_ssbo_atomic_umin;
  95       break;
  96    case nir_intrinsic_atomic_counter_max:
  97       op = nir_intrinsic_ssbo_atomic_umax;
  98       break;
  99    case nir_intrinsic_atomic_counter_and:
 100       op = nir_intrinsic_ssbo_atomic_and;
 101       break;
 102    case nir_intrinsic_atomic_counter_or:
 103       op = nir_intrinsic_ssbo_atomic_or;
 104       break;
 105    case nir_intrinsic_atomic_counter_xor:
 106       op = nir_intrinsic_ssbo_atomic_xor;
 107       break;
 108    case nir_intrinsic_atomic_counter_exchange:
 109       op = nir_intrinsic_ssbo_atomic_exchange;
 110       break;
 111    case nir_intrinsic_atomic_counter_comp_swap:
 112       op = nir_intrinsic_ssbo_atomic_comp_swap;
 113       break;
 114    default:
 115       return false;
 116    }
 117
 118    nir_ssa_def *buffer = nir_imm_int(b, nir_intrinsic_base(instr));
 119    nir_ssa_def *temp = NULL;
 120    nir_intrinsic_instr *new_instr =
 121          nir_intrinsic_instr_create(ralloc_parent(instr), op);
 122
 123    /* a couple instructions need special handling since they don't map
 124     * 1:1 with ssbo atomics
 125     */
 126    switch (instr->intrinsic) {
 127    case nir_intrinsic_atomic_counter_inc:
 128       /* remapped to ssbo_atomic_add: { buffer_idx, offset, +1 } */
 129       temp = nir_imm_int(b, +1);
 130       new_instr->src[0] = nir_src_for_ssa(buffer);
 131       nir_src_copy(&new_instr->src[1], &instr->src[0], new_instr);
 132       new_instr->src[2] = nir_src_for_ssa(temp);
 133       break;
 134    case nir_intrinsic_atomic_counter_pre_dec:
 135    case nir_intrinsic_atomic_counter_post_dec:
 136       /* remapped to ssbo_atomic_add: { buffer_idx, offset, -1 } */
 137       /* NOTE semantic difference so we adjust the return value below */
 138       temp = nir_imm_int(b, -1);
 139       new_instr->src[0] = nir_src_for_ssa(buffer);
 140       nir_src_copy(&new_instr->src[1], &instr->src[0], new_instr);
 141       new_instr->src[2] = nir_src_for_ssa(temp);
 142       break;
 143    case nir_intrinsic_atomic_counter_read:
 144       /* remapped to load_ssbo: { buffer_idx, offset } */
 145       new_instr->src[0] = nir_src_for_ssa(buffer);
 146       nir_src_copy(&new_instr->src[1], &instr->src[0], new_instr);
 147       break;
 148    default:
 149       /* remapped to ssbo_atomic_x: { buffer_idx, offset, data, (compare)? } */
 150       new_instr->src[0] = nir_src_for_ssa(buffer);
 151       nir_src_copy(&new_instr->src[1], &instr->src[0], new_instr);
 152       nir_src_copy(&new_instr->src[2], &instr->src[1], new_instr);
 153       if (op == nir_intrinsic_ssbo_atomic_comp_swap ||
 154           op == nir_intrinsic_ssbo_atomic_fcomp_swap)
 155          nir_src_copy(&new_instr->src[3], &instr->src[2], new_instr);
 156       break;
 157    }
 158
 159    if (new_instr->intrinsic == nir_intrinsic_load_ssbo ||
 160        new_instr->intrinsic == nir_intrinsic_store_ssbo)
 161       nir_intrinsic_set_align(new_instr, 4, 0);
 162
 163    nir_ssa_dest_init(&new_instr->instr, &new_instr->dest,
 164                      instr->dest.ssa.num_components,
 165                      instr->dest.ssa.bit_size, NULL);
 166    nir_instr_insert_before(&instr->instr, &new_instr->instr);
 167    nir_instr_remove(&instr->instr);
 168
 169    if (instr->intrinsic == nir_intrinsic_atomic_counter_pre_dec) {
 170       b->cursor = nir_after_instr(&new_instr->instr);
 171       nir_ssa_def *result = nir_iadd(b, &new_instr->dest.ssa, temp);
 172       nir_ssa_def_rewrite_uses(&instr->dest.ssa, nir_src_for_ssa(result));
 173    } else {
 174       nir_ssa_def_rewrite_uses(&instr->dest.ssa, nir_src_for_ssa(&new_instr->dest.ssa));
 175    }
 176
 177    /* we could be replacing an intrinsic with fixed # of dest num_components
 178     * with one that has variable number.  So best to take this from the dest:
 179     */
 180    new_instr->num_components = instr->dest.ssa.num_components;
 181
 182    return true;
 183 }
 184
 185 static bool
 186 is_atomic_uint(const struct glsl_type *type)
 187 {
 188    if (glsl_get_base_type(type) == GLSL_TYPE_ARRAY)
 189       return is_atomic_uint(glsl_get_array_element(type));
 190    return glsl_get_base_type(type) == GLSL_TYPE_ATOMIC_UINT;
 191 }
 192
 193 bool
 194 nir_lower_atomics_to_ssbo(nir_shader *shader, unsigned ssbo_offset)
 195 {
 196    bool progress = false;
 197
 198    nir_foreach_function(function, shader) {
 199       if (function->impl) {
 200          nir_builder builder;
 201          nir_builder_init(&builder, function->impl);
 202          nir_foreach_block(block, function->impl) {
 203             nir_foreach_instr_safe(instr, block) {
 204                if (instr->type == nir_instr_type_intrinsic)
 205                   progress |= lower_instr(nir_instr_as_intrinsic(instr),
 206                                           ssbo_offset, &builder);
 207             }
 208          }
 209
 210          nir_metadata_preserve(function->impl, nir_metadata_block_index |
 211                                                nir_metadata_dominance);
 212       }
 213    }
 214
 215    if (progress) {
 216       /* replace atomic_uint uniforms with ssbo's: */
 217       unsigned replaced = 0;
 218       nir_foreach_variable_safe(var, &shader->uniforms) {
 219          if (is_atomic_uint(var->type)) {
 220             exec_node_remove(&var->node);
 221
 222             if (replaced & (1 << var->data.binding))
 223                continue;
 224
 225             nir_variable *ssbo;
 226             char name[16];
 227
 228             /* A length of 0 is used to denote unsized arrays */
 229             const struct glsl_type *type = glsl_array_type(glsl_uint_type(), 0, 0);
 230
 231             snprintf(name, sizeof(name), "counter%d", var->data.binding);
 232
 233             ssbo = nir_variable_create(shader, nir_var_mem_ssbo, type, name);
 234             ssbo->data.binding = var->data.binding;
 235
 236             struct glsl_struct_field field = {
 237                   .type = type,
 238                   .name = "counters",
 239                   .location = -1,
 240             };
 241
 242             ssbo->interface_type =
 243                   glsl_interface_type(&field, 1, GLSL_INTERFACE_PACKING_STD430,
 244                                       false, "counters");
 245
 246             replaced |= (1 << var->data.binding);
 247          }
 248       }
 249
 250       /* Make sure that shader->info.num_ssbos still reflects the maximum SSBO
 251        * index that can be used in the shader.
 252        */
 253       if (shader->info.num_ssbos > 0) {
 254          shader->info.num_ssbos += ssbo_offset;
 255       } else {
 256          /* We can't use num_abos, because it only represents the number of
 257           * active atomic counters, and currently unlike SSBO's they aren't
 258           * compacted so num_abos actually isn't a bound on the index passed
 259           * to nir_intrinsic_atomic_counter_*. e.g. if we have a single atomic
 260           * counter declared like:
 261           *
 262           * layout(binding=1) atomic_uint counter0;
 263           *
 264           * then when we lower accesses to it the atomic_counter_* intrinsics
 265           * will have 1 as the index but num_abos will still be 1.
 266           * */
 267          shader->info.num_ssbos = util_last_bit(replaced);
 268       }
 269       shader->info.num_abos = 0;
 270    }
 271
 272    return progress;
 273 }
 274