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_ssbo_atomic_add:
  51    case nir_intrinsic_ssbo_atomic_imin:
  52    case nir_intrinsic_ssbo_atomic_umin:
  53    case nir_intrinsic_ssbo_atomic_imax:
  54    case nir_intrinsic_ssbo_atomic_umax:
  55    case nir_intrinsic_ssbo_atomic_and:
  56    case nir_intrinsic_ssbo_atomic_or:
  57    case nir_intrinsic_ssbo_atomic_xor:
  58    case nir_intrinsic_ssbo_atomic_exchange:
  59    case nir_intrinsic_ssbo_atomic_comp_swap:
  60    case nir_intrinsic_store_ssbo:
  61    case nir_intrinsic_load_ssbo:
  62    case nir_intrinsic_get_buffer_size:
  63       /* easy case, keep same opcode and just remap SSBO buffer index: */
  64       op = instr->intrinsic;
  65       idx_src = (op == nir_intrinsic_store_ssbo) ? 1 : 0;
  66       nir_ssa_def *old_idx = nir_ssa_for_src(b, instr->src[idx_src], 1);
  67       nir_ssa_def *new_idx = nir_iadd(b, old_idx, nir_imm_int(b, ssbo_offset));
  68       nir_instr_rewrite_src(&instr->instr,
  69                             &instr->src[idx_src],
  70                             nir_src_for_ssa(new_idx));
  71       return true;
  72    case nir_intrinsic_atomic_counter_inc:
  73    case nir_intrinsic_atomic_counter_add:
  74    case nir_intrinsic_atomic_counter_pre_dec:
  75    case nir_intrinsic_atomic_counter_post_dec:
  76       /* inc and dec get remapped to add: */
  77       op = nir_intrinsic_ssbo_atomic_add;
  78       break;
  79    case nir_intrinsic_atomic_counter_read:
  80       op = nir_intrinsic_load_ssbo;
  81       break;
  82    case nir_intrinsic_atomic_counter_min:
  83       op = nir_intrinsic_ssbo_atomic_umin;
  84       break;
  85    case nir_intrinsic_atomic_counter_max:
  86       op = nir_intrinsic_ssbo_atomic_umax;
  87       break;
  88    case nir_intrinsic_atomic_counter_and:
  89       op = nir_intrinsic_ssbo_atomic_and;
  90       break;
  91    case nir_intrinsic_atomic_counter_or:
  92       op = nir_intrinsic_ssbo_atomic_or;
  93       break;
  94    case nir_intrinsic_atomic_counter_xor:
  95       op = nir_intrinsic_ssbo_atomic_xor;
  96       break;
  97    case nir_intrinsic_atomic_counter_exchange:
  98       op = nir_intrinsic_ssbo_atomic_exchange;
  99       break;
 100    case nir_intrinsic_atomic_counter_comp_swap:
 101       op = nir_intrinsic_ssbo_atomic_comp_swap;
 102       break;
 103    default:
 104       return false;
 105    }
 106
 107    nir_ssa_def *buffer = nir_imm_int(b, nir_intrinsic_base(instr));
 108    nir_ssa_def *temp = NULL;
 109    nir_intrinsic_instr *new_instr =
 110          nir_intrinsic_instr_create(ralloc_parent(instr), op);
 111
 112    /* a couple instructions need special handling since they don't map
 113     * 1:1 with ssbo atomics
 114     */
 115    switch (instr->intrinsic) {
 116    case nir_intrinsic_atomic_counter_inc:
 117       /* remapped to ssbo_atomic_add: { buffer_idx, offset, +1 } */
 118       temp = nir_imm_int(b, +1);
 119       new_instr->src[0] = nir_src_for_ssa(buffer);
 120       nir_src_copy(&new_instr->src[1], &instr->src[0], new_instr);
 121       new_instr->src[2] = nir_src_for_ssa(temp);
 122       break;
 123    case nir_intrinsic_atomic_counter_pre_dec:
 124    case nir_intrinsic_atomic_counter_post_dec:
 125       /* remapped to ssbo_atomic_add: { buffer_idx, offset, -1 } */
 126       /* NOTE semantic difference so we adjust the return value below */
 127       temp = nir_imm_int(b, -1);
 128       new_instr->src[0] = nir_src_for_ssa(buffer);
 129       nir_src_copy(&new_instr->src[1], &instr->src[0], new_instr);
 130       new_instr->src[2] = nir_src_for_ssa(temp);
 131       break;
 132    case nir_intrinsic_atomic_counter_read:
 133       /* remapped to load_ssbo: { buffer_idx, offset } */
 134       new_instr->src[0] = nir_src_for_ssa(buffer);
 135       nir_src_copy(&new_instr->src[1], &instr->src[0], new_instr);
 136       break;
 137    default:
 138       /* remapped to ssbo_atomic_x: { buffer_idx, offset, data, (compare)? } */
 139       new_instr->src[0] = nir_src_for_ssa(buffer);
 140       nir_src_copy(&new_instr->src[1], &instr->src[0], new_instr);
 141       nir_src_copy(&new_instr->src[2], &instr->src[1], new_instr);
 142       if (op == nir_intrinsic_ssbo_atomic_comp_swap)
 143          nir_src_copy(&new_instr->src[3], &instr->src[2], new_instr);
 144       break;
 145    }
 146
 147    nir_ssa_dest_init(&new_instr->instr, &new_instr->dest,
 148                      instr->dest.ssa.num_components,
 149                      instr->dest.ssa.bit_size, NULL);
 150    nir_instr_insert_before(&instr->instr, &new_instr->instr);
 151    nir_instr_remove(&instr->instr);
 152
 153    if (instr->intrinsic == nir_intrinsic_atomic_counter_pre_dec) {
 154       b->cursor = nir_after_instr(&new_instr->instr);
 155       nir_ssa_def *result = nir_iadd(b, &new_instr->dest.ssa, temp);
 156       nir_ssa_def_rewrite_uses(&instr->dest.ssa, nir_src_for_ssa(result));
 157    } else {
 158       nir_ssa_def_rewrite_uses(&instr->dest.ssa, nir_src_for_ssa(&new_instr->dest.ssa));
 159    }
 160
 161    /* we could be replacing an intrinsic with fixed # of dest num_components
 162     * with one that has variable number.  So best to take this from the dest:
 163     */
 164    new_instr->num_components = instr->dest.ssa.num_components;
 165
 166    return true;
 167 }
 168
 169 static bool
 170 is_atomic_uint(const struct glsl_type *type)
 171 {
 172    if (glsl_get_base_type(type) == GLSL_TYPE_ARRAY)
 173       return is_atomic_uint(glsl_get_array_element(type));
 174    return glsl_get_base_type(type) == GLSL_TYPE_ATOMIC_UINT;
 175 }
 176
 177 bool
 178 nir_lower_atomics_to_ssbo(nir_shader *shader, unsigned ssbo_offset)
 179 {
 180    bool progress = false;
 181
 182    nir_foreach_function(function, shader) {
 183       if (function->impl) {
 184          nir_builder builder;
 185          nir_builder_init(&builder, function->impl);
 186          nir_foreach_block(block, function->impl) {
 187             nir_foreach_instr_safe(instr, block) {
 188                if (instr->type == nir_instr_type_intrinsic)
 189                   progress |= lower_instr(nir_instr_as_intrinsic(instr),
 190                                           ssbo_offset, &builder);
 191             }
 192          }
 193
 194          nir_metadata_preserve(function->impl, nir_metadata_block_index |
 195                                                nir_metadata_dominance);
 196       }
 197    }
 198
 199    if (progress) {
 200       /* replace atomic_uint uniforms with ssbo's: */
 201       unsigned replaced = 0;
 202       nir_foreach_variable_safe(var, &shader->uniforms) {
 203          if (is_atomic_uint(var->type)) {
 204             exec_node_remove(&var->node);
 205
 206             if (replaced & (1 << var->data.binding))
 207                continue;
 208
 209             nir_variable *ssbo;
 210             char name[16];
 211
 212             /* A length of 0 is used to denote unsized arrays */
 213             const struct glsl_type *type = glsl_array_type(glsl_uint_type(), 0);
 214
 215             snprintf(name, sizeof(name), "counter%d", var->data.binding);
 216
 217             ssbo = nir_variable_create(shader, nir_var_shader_storage,
 218                                        type, name);
 219             ssbo->data.binding = var->data.binding;
 220
 221             struct glsl_struct_field field = {
 222                   .type = type,
 223                   .name = "counters",
 224                   .location = -1,
 225             };
 226
 227             ssbo->interface_type =
 228                   glsl_interface_type(&field, 1, GLSL_INTERFACE_PACKING_STD430,
 229                                       false, "counters");
 230
 231             replaced |= (1 << var->data.binding);
 232          }
 233       }
 234    }
 235
 236    return progress;
 237 }
 238