src/glsl/nir/nir_opt_peephole_select.c

   1 /*
   2  * Copyright © 2014 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  * Authors:
  24  *    Jason Ekstrand (jason@jlekstrand.net)
  25  *
  26  */
  27
  28 #include "nir.h"
  29
  30 /*
  31  * Implements a small peephole optimization that looks for
  32  *
  33  * if (cond) {
  34  *    <empty>
  35  * } else {
  36  *    <empty>
  37  * }
  38  * phi
  39  * ...
  40  * phi
  41  *
  42  * and replaces it with a series of selects.  It can also handle the case
  43  * where, instead of being empty, the if may contain some move operations
  44  * whose only use is one of the following phi nodes.  This happens all the
  45  * time when the SSA form comes from a conditional assignment with a
  46  * swizzle.
  47  */
  48
  49 struct peephole_select_state {
  50    void *mem_ctx;
  51    bool progress;
  52 };
  53
  54 static bool
  55 block_check_for_allowed_instrs(nir_block *block)
  56 {
  57    nir_foreach_instr(block, instr) {
  58       switch (instr->type) {
  59       case nir_instr_type_intrinsic: {
  60          nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
  61
  62          switch (intrin->intrinsic) {
  63          case nir_intrinsic_load_var:
  64             switch (intrin->variables[0]->var->data.mode) {
  65             case nir_var_shader_in:
  66             case nir_var_uniform:
  67                break;
  68
  69             default:
  70                return false;
  71             }
  72             break;
  73
  74          default:
  75             return false;
  76          }
  77
  78          break;
  79       }
  80
  81       case nir_instr_type_load_const:
  82          break;
  83
  84       case nir_instr_type_alu: {
  85          /* It must be a move operation */
  86          nir_alu_instr *mov = nir_instr_as_alu(instr);
  87          if (mov->op != nir_op_fmov && mov->op != nir_op_imov &&
  88              mov->op != nir_op_fneg && mov->op != nir_op_ineg &&
  89              mov->op != nir_op_fabs && mov->op != nir_op_iabs)
  90             return false;
  91
  92          /* Can't handle saturate */
  93          if (mov->dest.saturate)
  94             return false;
  95
  96          /* It must be SSA */
  97          if (!mov->dest.dest.is_ssa)
  98             return false;
  99
 100          /* It cannot have any if-uses */
 101          if (!list_empty(&mov->dest.dest.ssa.if_uses))
 102             return false;
 103
 104          /* The only uses of this definition must be phi's in the successor */
 105          nir_foreach_use(&mov->dest.dest.ssa, use) {
 106             if (use->parent_instr->type != nir_instr_type_phi ||
 107                 use->parent_instr->block != block->successors[0])
 108                return false;
 109          }
 110          break;
 111       }
 112
 113       default:
 114          return false;
 115       }
 116    }
 117
 118    return true;
 119 }
 120
 121 static bool
 122 nir_opt_peephole_select_block(nir_block *block, void *void_state)
 123 {
 124    struct peephole_select_state *state = void_state;
 125
 126    /* If the block is empty, then it certainly doesn't have any phi nodes,
 127     * so we can skip it.  This also ensures that we do an early skip on the
 128     * end block of the function which isn't actually attached to the CFG.
 129     */
 130    if (exec_list_is_empty(&block->instr_list))
 131       return true;
 132
 133    if (nir_cf_node_is_first(&block->cf_node))
 134       return true;
 135
 136    nir_cf_node *prev_node = nir_cf_node_prev(&block->cf_node);
 137    if (prev_node->type != nir_cf_node_if)
 138       return true;
 139
 140    nir_if *if_stmt = nir_cf_node_as_if(prev_node);
 141    nir_cf_node *then_node = nir_if_first_then_node(if_stmt);
 142    nir_cf_node *else_node = nir_if_first_else_node(if_stmt);
 143
 144    /* We can only have one block in each side ... */
 145    if (nir_if_last_then_node(if_stmt) != then_node ||
 146        nir_if_last_else_node(if_stmt) != else_node)
 147       return true;
 148
 149    nir_block *then_block = nir_cf_node_as_block(then_node);
 150    nir_block *else_block = nir_cf_node_as_block(else_node);
 151
 152    /* ... and those blocks must only contain "allowed" instructions. */
 153    if (!block_check_for_allowed_instrs(then_block) ||
 154        !block_check_for_allowed_instrs(else_block))
 155       return true;
 156
 157    /* At this point, we know that the previous CFG node is an if-then
 158     * statement containing only moves to phi nodes in this block.  We can
 159     * just remove that entire CF node and replace all of the phi nodes with
 160     * selects.
 161     */
 162
 163    nir_block *prev_block = nir_cf_node_as_block(nir_cf_node_prev(prev_node));
 164    assert(prev_block->cf_node.type == nir_cf_node_block);
 165
 166    /* First, we move the remaining instructions from the blocks to the
 167     * block before.  We have already guaranteed that this is safe by
 168     * calling block_check_for_allowed_instrs()
 169     */
 170    nir_foreach_instr_safe(then_block, instr) {
 171       exec_node_remove(&instr->node);
 172       instr->block = prev_block;
 173       exec_list_push_tail(&prev_block->instr_list, &instr->node);
 174    }
 175
 176    nir_foreach_instr_safe(else_block, instr) {
 177       exec_node_remove(&instr->node);
 178       instr->block = prev_block;
 179       exec_list_push_tail(&prev_block->instr_list, &instr->node);
 180    }
 181
 182    nir_foreach_instr_safe(block, instr) {
 183       if (instr->type != nir_instr_type_phi)
 184          break;
 185
 186       nir_phi_instr *phi = nir_instr_as_phi(instr);
 187       nir_alu_instr *sel = nir_alu_instr_create(state->mem_ctx, nir_op_bcsel);
 188       nir_src_copy(&sel->src[0].src, &if_stmt->condition, state->mem_ctx);
 189       /* Splat the condition to all channels */
 190       memset(sel->src[0].swizzle, 0, sizeof sel->src[0].swizzle);
 191
 192       assert(exec_list_length(&phi->srcs) == 2);
 193       nir_foreach_phi_src(phi, src) {
 194          assert(src->pred == then_block || src->pred == else_block);
 195          assert(src->src.is_ssa);
 196
 197          unsigned idx = src->pred == then_block ? 1 : 2;
 198          nir_src_copy(&sel->src[idx].src, &src->src, state->mem_ctx);
 199       }
 200
 201       nir_ssa_dest_init(&sel->instr, &sel->dest.dest,
 202                         phi->dest.ssa.num_components, phi->dest.ssa.name);
 203       sel->dest.write_mask = (1 << phi->dest.ssa.num_components) - 1;
 204
 205       nir_ssa_def_rewrite_uses(&phi->dest.ssa,
 206                                nir_src_for_ssa(&sel->dest.dest.ssa),
 207                                state->mem_ctx);
 208
 209       nir_instr_insert_before(&phi->instr, &sel->instr);
 210       nir_instr_remove(&phi->instr);
 211    }
 212
 213    nir_cf_node_remove(&if_stmt->cf_node);
 214    state->progress = true;
 215
 216    return true;
 217 }
 218
 219 static bool
 220 nir_opt_peephole_select_impl(nir_function_impl *impl)
 221 {
 222    struct peephole_select_state state;
 223
 224    state.mem_ctx = ralloc_parent(impl);
 225    state.progress = false;
 226
 227    nir_foreach_block(impl, nir_opt_peephole_select_block, &state);
 228
 229    if (state.progress)
 230       nir_metadata_preserve(impl, nir_metadata_none);
 231
 232    return state.progress;
 233 }
 234
 235 bool
 236 nir_opt_peephole_select(nir_shader *shader)
 237 {
 238    bool progress = false;
 239
 240    nir_foreach_overload(shader, overload) {
 241       if (overload->impl)
 242          progress |= nir_opt_peephole_select_impl(overload->impl);
 243    }
 244
 245    return progress;
 246 }