src/intel/compiler/brw_nir_intrinsics.c

   1 /*
   2  * Copyright (c) 2016 Intel Corporation
   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
  24 #include "brw_nir.h"
  25 #include "compiler/nir/nir_builder.h"
  26
  27 struct lower_intrinsics_state {
  28    nir_shader *nir;
  29    union {
  30       struct brw_stage_prog_data *prog_data;
  31       struct brw_cs_prog_data *cs_prog_data;
  32    };
  33    nir_function_impl *impl;
  34    bool progress;
  35    nir_builder builder;
  36    bool cs_thread_id_used;
  37 };
  38
  39 static nir_ssa_def *
  40 read_thread_local_id(struct lower_intrinsics_state *state)
  41 {
  42    nir_builder *b = &state->builder;
  43    nir_shader *nir = state->nir;
  44    const unsigned *sizes = nir->info->cs.local_size;
  45    const unsigned group_size = sizes[0] * sizes[1] * sizes[2];
  46
  47    /* Some programs have local_size dimensions so small that the thread local
  48     * ID will always be 0.
  49     */
  50    if (group_size <= 8)
  51       return nir_imm_int(b, 0);
  52
  53    assert(state->cs_prog_data->thread_local_id_index >= 0);
  54    state->cs_thread_id_used = true;
  55    const int id_index = state->cs_prog_data->thread_local_id_index;
  56
  57    nir_intrinsic_instr *load =
  58       nir_intrinsic_instr_create(nir, nir_intrinsic_load_uniform);
  59    load->num_components = 1;
  60    load->src[0] = nir_src_for_ssa(nir_imm_int(b, 0));
  61    nir_ssa_dest_init(&load->instr, &load->dest, 1, 32, NULL);
  62    nir_intrinsic_set_base(load, id_index * sizeof(uint32_t));
  63    nir_intrinsic_set_range(load, sizeof(uint32_t));
  64    nir_builder_instr_insert(b, &load->instr);
  65    return &load->dest.ssa;
  66 }
  67
  68 static bool
  69 lower_cs_intrinsics_convert_block(struct lower_intrinsics_state *state,
  70                                   nir_block *block)
  71 {
  72    bool progress = false;
  73    nir_builder *b = &state->builder;
  74    nir_shader *nir = state->nir;
  75
  76    nir_foreach_instr_safe(instr, block) {
  77       if (instr->type != nir_instr_type_intrinsic)
  78          continue;
  79
  80       nir_intrinsic_instr *intrinsic = nir_instr_as_intrinsic(instr);
  81
  82       b->cursor = nir_after_instr(&intrinsic->instr);
  83
  84       nir_ssa_def *sysval;
  85       switch (intrinsic->intrinsic) {
  86       case nir_intrinsic_load_local_invocation_index: {
  87          assert(nir->stage == MESA_SHADER_COMPUTE);
  88          /* We construct the local invocation index from:
  89           *
  90           *    gl_LocalInvocationIndex =
  91           *       cs_thread_local_id + channel_num;
  92           */
  93          nir_ssa_def *thread_local_id = read_thread_local_id(state);
  94          nir_ssa_def *channel = nir_load_channel_num(b);
  95          sysval = nir_iadd(b, channel, thread_local_id);
  96          break;
  97       }
  98
  99       case nir_intrinsic_load_local_invocation_id: {
 100          assert(nir->stage == MESA_SHADER_COMPUTE);
 101          /* We lower gl_LocalInvocationID from gl_LocalInvocationIndex based
 102           * on this formula:
 103           *
 104           *    gl_LocalInvocationID.x =
 105           *       gl_LocalInvocationIndex % gl_WorkGroupSize.x;
 106           *    gl_LocalInvocationID.y =
 107           *       (gl_LocalInvocationIndex / gl_WorkGroupSize.x) %
 108           *       gl_WorkGroupSize.y;
 109           *    gl_LocalInvocationID.z =
 110           *       (gl_LocalInvocationIndex /
 111           *        (gl_WorkGroupSize.x * gl_WorkGroupSize.y)) %
 112           *       gl_WorkGroupSize.z;
 113           */
 114          unsigned *size = nir->info->cs.local_size;
 115
 116          nir_ssa_def *local_index = nir_load_local_invocation_index(b);
 117
 118          nir_const_value uvec3;
 119          uvec3.u32[0] = 1;
 120          uvec3.u32[1] = size[0];
 121          uvec3.u32[2] = size[0] * size[1];
 122          nir_ssa_def *div_val = nir_build_imm(b, 3, 32, uvec3);
 123          uvec3.u32[0] = size[0];
 124          uvec3.u32[1] = size[1];
 125          uvec3.u32[2] = size[2];
 126          nir_ssa_def *mod_val = nir_build_imm(b, 3, 32, uvec3);
 127
 128          sysval = nir_umod(b, nir_udiv(b, local_index, div_val), mod_val);
 129          break;
 130       }
 131
 132       default:
 133          continue;
 134       }
 135
 136       nir_ssa_def_rewrite_uses(&intrinsic->dest.ssa, nir_src_for_ssa(sysval));
 137       nir_instr_remove(&intrinsic->instr);
 138
 139       state->progress = true;
 140    }
 141
 142    return progress;
 143 }
 144
 145 static void
 146 lower_cs_intrinsics_convert_impl(struct lower_intrinsics_state *state)
 147 {
 148    nir_builder_init(&state->builder, state->impl);
 149
 150    nir_foreach_block(block, state->impl) {
 151       lower_cs_intrinsics_convert_block(state, block);
 152    }
 153
 154    nir_metadata_preserve(state->impl,
 155                          nir_metadata_block_index | nir_metadata_dominance);
 156 }
 157
 158 bool
 159 brw_nir_lower_intrinsics(nir_shader *nir, struct brw_stage_prog_data *prog_data)
 160 {
 161    /* Currently we only lower intrinsics for compute shaders */
 162    if (nir->stage != MESA_SHADER_COMPUTE)
 163       return false;
 164
 165    bool progress = false;
 166    struct lower_intrinsics_state state;
 167    memset(&state, 0, sizeof(state));
 168    state.nir = nir;
 169    state.prog_data = prog_data;
 170
 171    do {
 172       state.progress = false;
 173       nir_foreach_function(function, nir) {
 174          if (function->impl) {
 175             state.impl = function->impl;
 176             lower_cs_intrinsics_convert_impl(&state);
 177          }
 178       }
 179       progress |= state.progress;
 180    } while (state.progress);
 181
 182    if (nir->stage == MESA_SHADER_COMPUTE && !state.cs_thread_id_used)
 183       state.cs_prog_data->thread_local_id_index = -1;
 184
 185    return progress;
 186 }