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 #include "nir_control_flow.h"
  30
  31 /*
  32  * Implements a small peephole optimization that looks for
  33  *
  34  * if (cond) {
  35  *    <empty>
  36  * } else {
  37  *    <empty>
  38  * }
  39  * phi
  40  * ...
  41  * phi
  42  *
  43  * and replaces it with a series of selects.  It can also handle the case
  44  * where, instead of being empty, the if may contain some move operations
  45  * whose only use is one of the following phi nodes.  This happens all the
  46  * time when the SSA form comes from a conditional assignment with a
  47  * swizzle.
  48  */
  49
  50 struct peephole_select_state {
  51    void *mem_ctx;
  52    bool progress;
  53 };
  54
  55 static bool
  56 block_check_for_allowed_instrs(nir_block *block)
  57 {
  58    nir_foreach_instr(block, instr) {
  59       switch (instr->type) {
  60       case nir_instr_type_intrinsic: {
  61          nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
  62
  63          switch (intrin->intrinsic) {
  64          case nir_intrinsic_load_var:
  65             switch (intrin->variables[0]->var->data.mode) {
  66             case nir_var_shader_in:
  67             case nir_var_uniform:
  68                break;
  69
  70             default:
  71                return false;
  72             }
  73             break;
  74
  75          default:
  76             return false;
  77          }
  78
  79          break;
  80       }
  81
  82       case nir_instr_type_load_const:
  83          break;
  84
  85       case nir_instr_type_alu: {
  86          nir_alu_instr *mov = nir_instr_as_alu(instr);
  87          switch (mov->op) {
  88          case nir_op_fmov:
  89          case nir_op_imov:
  90          case nir_op_fneg:
  91          case nir_op_ineg:
  92          case nir_op_fabs:
  93          case nir_op_iabs:
  94          case nir_op_vec2:
  95          case nir_op_vec3:
  96          case nir_op_vec4:
  97             /* It must be a move-like operation. */
  98             break;
  99          default:
 100             return false;
 101          }
 102
 103          /* Can't handle saturate */
 104          if (mov->dest.saturate)
 105             return false;
 106
 107          /* It must be SSA */
 108          if (!mov->dest.dest.is_ssa)
 109             return false;
 110
 111          /* It cannot have any if-uses */
 112          if (!list_empty(&mov->dest.dest.ssa.if_uses))
 113             return false;
 114
 115          /* The only uses of this definition must be phi's in the successor */
 116          nir_foreach_use(&mov->dest.dest.ssa, use) {
 117             if (use->parent_instr->type != nir_instr_type_phi ||
 118                 use->parent_instr->block != block->successors[0])
 119                return false;
 120          }
 121          break;
 122       }
 123
 124       default:
 125          return false;
 126       }
 127    }
 128
 129    return true;
 130 }
 131
 132 static bool
 133 nir_opt_peephole_select_block(nir_block *block, void *void_state)
 134 {
 135    struct peephole_select_state *state = void_state;
 136
 137    /* If the block is empty, then it certainly doesn't have any phi nodes,
 138     * so we can skip it.  This also ensures that we do an early skip on the
 139     * end block of the function which isn't actually attached to the CFG.
 140     */
 141    if (exec_list_is_empty(&block->instr_list))
 142       return true;
 143
 144    if (nir_cf_node_is_first(&block->cf_node))
 145       return true;
 146
 147    nir_cf_node *prev_node = nir_cf_node_prev(&block->cf_node);
 148    if (prev_node->type != nir_cf_node_if)
 149       return true;
 150
 151    nir_if *if_stmt = nir_cf_node_as_if(prev_node);
 152    nir_cf_node *then_node = nir_if_first_then_node(if_stmt);
 153    nir_cf_node *else_node = nir_if_first_else_node(if_stmt);
 154
 155    /* We can only have one block in each side ... */
 156    if (nir_if_last_then_node(if_stmt) != then_node ||
 157        nir_if_last_else_node(if_stmt) != else_node)
 158       return true;
 159
 160    nir_block *then_block = nir_cf_node_as_block(then_node);
 161    nir_block *else_block = nir_cf_node_as_block(else_node);
 162
 163    /* ... and those blocks must only contain "allowed" instructions. */
 164    if (!block_check_for_allowed_instrs(then_block) ||
 165        !block_check_for_allowed_instrs(else_block))
 166       return true;
 167
 168    /* At this point, we know that the previous CFG node is an if-then
 169     * statement containing only moves to phi nodes in this block.  We can
 170     * just remove that entire CF node and replace all of the phi nodes with
 171     * selects.
 172     */
 173
 174    nir_block *prev_block = nir_cf_node_as_block(nir_cf_node_prev(prev_node));
 175    assert(prev_block->cf_node.type == nir_cf_node_block);
 176
 177    /* First, we move the remaining instructions from the blocks to the
 178     * block before.  We have already guaranteed that this is safe by
 179     * calling block_check_for_allowed_instrs()
 180     */
 181    nir_foreach_instr_safe(then_block, instr) {
 182       exec_node_remove(&instr->node);
 183       instr->block = prev_block;
 184       exec_list_push_tail(&prev_block->instr_list, &instr->node);
 185    }
 186
 187    nir_foreach_instr_safe(else_block, instr) {
 188       exec_node_remove(&instr->node);
 189       instr->block = prev_block;
 190       exec_list_push_tail(&prev_block->instr_list, &instr->node);
 191    }
 192
 193    nir_foreach_instr_safe(block, instr) {
 194       if (instr->type != nir_instr_type_phi)
 195          break;
 196
 197       nir_phi_instr *phi = nir_instr_as_phi(instr);
 198       nir_alu_instr *sel = nir_alu_instr_create(state->mem_ctx, nir_op_bcsel);
 199       nir_src_copy(&sel->src[0].src, &if_stmt->condition, sel);
 200       /* Splat the condition to all channels */
 201       memset(sel->src[0].swizzle, 0, sizeof sel->src[0].swizzle);
 202
 203       assert(exec_list_length(&phi->srcs) == 2);
 204       nir_foreach_phi_src(phi, src) {
 205          assert(src->pred == then_block || src->pred == else_block);
 206          assert(src->src.is_ssa);
 207
 208          unsigned idx = src->pred == then_block ? 1 : 2;
 209          nir_src_copy(&sel->src[idx].src, &src->src, sel);
 210       }
 211
 212       nir_ssa_dest_init(&sel->instr, &sel->dest.dest,
 213                         phi->dest.ssa.num_components, phi->dest.ssa.name);
 214       sel->dest.write_mask = (1 << phi->dest.ssa.num_components) - 1;
 215
 216       nir_ssa_def_rewrite_uses(&phi->dest.ssa,
 217                                nir_src_for_ssa(&sel->dest.dest.ssa));
 218
 219       nir_instr_insert_before(&phi->instr, &sel->instr);
 220       nir_instr_remove(&phi->instr);
 221    }
 222
 223    nir_cf_node_remove(&if_stmt->cf_node);
 224    state->progress = true;
 225
 226    return true;
 227 }
 228
 229 static bool
 230 nir_opt_peephole_select_impl(nir_function_impl *impl)
 231 {
 232    struct peephole_select_state state;
 233
 234    state.mem_ctx = ralloc_parent(impl);
 235    state.progress = false;
 236
 237    nir_foreach_block(impl, nir_opt_peephole_select_block, &state);
 238
 239    if (state.progress)
 240       nir_metadata_preserve(impl, nir_metadata_none);
 241
 242    return state.progress;
 243 }
 244
 245 bool
 246 nir_opt_peephole_select(nir_shader *shader)
 247 {
 248    bool progress = false;
 249
 250    nir_foreach_overload(shader, overload) {
 251       if (overload->impl)
 252          progress |= nir_opt_peephole_select_impl(overload->impl);
 253    }
 254
 255    return progress;
 256 }