src/mesa/drivers/dri/i965/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    assert(state->cs_prog_data->thread_local_id_index >= 0);
  43    state->cs_thread_id_used = true;
  44    const int id_index = state->cs_prog_data->thread_local_id_index;
  45
  46    nir_builder *b = &state->builder;
  47    nir_shader *nir = state->nir;
  48    nir_intrinsic_instr *load =
  49       nir_intrinsic_instr_create(nir, nir_intrinsic_load_uniform);
  50    load->num_components = 1;
  51    load->src[0] = nir_src_for_ssa(nir_imm_int(b, 0));
  52    nir_ssa_dest_init(&load->instr, &load->dest, 1, 32, NULL);
  53    nir_intrinsic_set_base(load, id_index * sizeof(uint32_t));
  54    nir_intrinsic_set_range(load, sizeof(uint32_t));
  55    nir_builder_instr_insert(b, &load->instr);
  56    return &load->dest.ssa;
  57 }
  58
  59 static bool
  60 lower_cs_intrinsics_convert_block(struct lower_intrinsics_state *state,
  61                                   nir_block *block)
  62 {
  63    bool progress = false;
  64    nir_builder *b = &state->builder;
  65    nir_shader *nir = state->nir;
  66
  67    nir_foreach_instr_safe(instr, block) {
  68       if (instr->type != nir_instr_type_intrinsic)
  69          continue;
  70
  71       nir_intrinsic_instr *intrinsic = nir_instr_as_intrinsic(instr);
  72
  73       b->cursor = nir_after_instr(&intrinsic->instr);
  74
  75       nir_ssa_def *sysval;
  76       switch (intrinsic->intrinsic) {
  77       case nir_intrinsic_load_local_invocation_index: {
  78          assert(nir->stage == MESA_SHADER_COMPUTE);
  79          /* We construct the local invocation index from:
  80           *
  81           *    gl_LocalInvocationIndex =
  82           *       cs_thread_local_id + channel_num;
  83           */
  84          nir_ssa_def *thread_local_id = read_thread_local_id(state);
  85          nir_ssa_def *channel =
  86             nir_load_system_value(b, nir_intrinsic_load_channel_num, 0);
  87          sysval = nir_iadd(b, channel, thread_local_id);
  88          break;
  89       }
  90
  91       case nir_intrinsic_load_local_invocation_id: {
  92          assert(nir->stage == MESA_SHADER_COMPUTE);
  93          /* We lower gl_LocalInvocationID from gl_LocalInvocationIndex based
  94           * on this formula:
  95           *
  96           *    gl_LocalInvocationID.x =
  97           *       gl_LocalInvocationIndex % gl_WorkGroupSize.x;
  98           *    gl_LocalInvocationID.y =
  99           *       (gl_LocalInvocationIndex / gl_WorkGroupSize.x) %
 100           *       gl_WorkGroupSize.y;
 101           *    gl_LocalInvocationID.z =
 102           *       (gl_LocalInvocationIndex /
 103           *        (gl_WorkGroupSize.x * gl_WorkGroupSize.y)) %
 104           *       gl_WorkGroupSize.z;
 105           */
 106          unsigned *size = nir->info.cs.local_size;
 107
 108          nir_ssa_def *local_index =
 109             nir_load_system_value(b, nir_intrinsic_load_local_invocation_index, 0);
 110
 111          nir_const_value uvec3;
 112          uvec3.u32[0] = 1;
 113          uvec3.u32[1] = size[0];
 114          uvec3.u32[2] = size[0] * size[1];
 115          nir_ssa_def *div_val = nir_build_imm(b, 3, 32, uvec3);
 116          uvec3.u32[0] = size[0];
 117          uvec3.u32[1] = size[1];
 118          uvec3.u32[2] = size[2];
 119          nir_ssa_def *mod_val = nir_build_imm(b, 3, 32, uvec3);
 120
 121          sysval = nir_imod(b, nir_idiv(b, local_index, div_val), mod_val);
 122          break;
 123       }
 124
 125       default:
 126          continue;
 127       }
 128
 129       nir_ssa_def_rewrite_uses(&intrinsic->dest.ssa, nir_src_for_ssa(sysval));
 130       nir_instr_remove(&intrinsic->instr);
 131
 132       state->progress = true;
 133    }
 134
 135    return progress;
 136 }
 137
 138 static void
 139 lower_cs_intrinsics_convert_impl(struct lower_intrinsics_state *state)
 140 {
 141    nir_builder_init(&state->builder, state->impl);
 142
 143    nir_foreach_block(block, state->impl) {
 144       lower_cs_intrinsics_convert_block(state, block);
 145    }
 146
 147    nir_metadata_preserve(state->impl,
 148                          nir_metadata_block_index | nir_metadata_dominance);
 149 }
 150
 151 bool
 152 brw_nir_lower_intrinsics(nir_shader *nir, struct brw_stage_prog_data *prog_data)
 153 {
 154    /* Currently we only lower intrinsics for compute shaders */
 155    if (nir->stage != MESA_SHADER_COMPUTE)
 156       return false;
 157
 158    bool progress = false;
 159    struct lower_intrinsics_state state;
 160    memset(&state, 0, sizeof(state));
 161    state.nir = nir;
 162    state.prog_data = prog_data;
 163
 164    /* Currently this pass only lowers intrinsics using the uniform specified
 165     * by thread_local_id_index.
 166     */
 167    if (nir->stage == MESA_SHADER_COMPUTE &&
 168        state.cs_prog_data->thread_local_id_index < 0)
 169       return false;
 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 }