src/gallium/drivers/freedreno/ir3/ir3_flatten.c

   1 /* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
   2
   3 /*
   4  * Copyright (C) 2014 Rob Clark <robclark@freedesktop.org>
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the "Software"),
   8  * to deal in the Software without restriction, including without limitation
   9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  10  * and/or sell copies of the Software, and to permit persons to whom the
  11  * Software is furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice (including the next
  14  * paragraph) shall be included in all copies or substantial portions of the
  15  * Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  23  * SOFTWARE.
  24  *
  25  * Authors:
  26  *    Rob Clark <robclark@freedesktop.org>
  27  */
  28
  29 #include <stdarg.h>
  30
  31 #include "ir3.h"
  32
  33 /*
  34  * Flatten: flatten out legs of if/else, etc
  35  *
  36  * TODO probably should use some heuristic to decide to not flatten
  37  * if one side of the other is too large / deeply nested / whatever?
  38  */
  39
  40 struct ir3_flatten_ctx {
  41         struct ir3_block *block;
  42         unsigned cnt;
  43 };
  44
  45 static struct ir3_register *unwrap(struct ir3_register *reg)
  46 {
  47
  48         if (reg->flags & IR3_REG_SSA) {
  49                 struct ir3_instruction *instr = reg->instr;
  50                 if (is_meta(instr)) {
  51                         switch (instr->opc) {
  52                         case OPC_META_OUTPUT:
  53                         case OPC_META_FLOW:
  54                                 if (instr->regs_count > 1)
  55                                         return instr->regs[1];
  56                                 return NULL;
  57                         default:
  58                                 break;
  59                         }
  60                 }
  61         }
  62         return reg;
  63 }
  64
  65 static void ir3_instr_flatten(struct ir3_flatten_ctx *ctx,
  66                 struct ir3_instruction *instr)
  67 {
  68         struct ir3_instruction *src;
  69
  70         /* if we've already visited this instruction, bail now: */
  71         if (ir3_instr_check_mark(instr))
  72                 return;
  73
  74         instr->block = ctx->block;
  75
  76         /* TODO: maybe some threshold to decide whether to
  77          * flatten or not??
  78          */
  79         if (is_meta(instr)) {
  80                 if (instr->opc == OPC_META_PHI) {
  81                         struct ir3_register *cond, *t, *f;
  82
  83                         cond = unwrap(instr->regs[1]);
  84                         t    = unwrap(instr->regs[2]);  /* true val */
  85                         f    = unwrap(instr->regs[3]);  /* false val */
  86
  87                         /* must have cond, but t or f may be null if only written
  88                          * one one side of the if/else (in which case we can just
  89                          * convert the PHI to a simple move).
  90                          */
  91                         assert(cond);
  92                         assert(t || f);
  93
  94                         if (t && f) {
  95                                 /* convert the PHI instruction to sel.{b16,b32} */
  96                                 instr->category = 3;
  97
  98                                 /* instruction type based on dst size: */
  99                                 if (instr->regs[0]->flags & IR3_REG_HALF)
 100                                         instr->opc = OPC_SEL_B16;
 101                                 else
 102                                         instr->opc = OPC_SEL_B32;
 103
 104                                 instr->regs[1] = t;
 105                                 instr->regs[2] = cond;
 106                                 instr->regs[3] = f;
 107                         } else {
 108                                 /* convert to simple mov: */
 109                                 instr->category = 1;
 110                                 instr->cat1.dst_type = TYPE_F32;
 111                                 instr->cat1.src_type = TYPE_F32;
 112                                 instr->regs_count = 2;
 113                                 instr->regs[1] = t ? t : f;
 114                         }
 115
 116                         ctx->cnt++;
 117                 } else if ((instr->opc == OPC_META_INPUT) &&
 118                                 (instr->regs_count == 2)) {
 119                         type_t ftype;
 120
 121                         if (instr->regs[0]->flags & IR3_REG_HALF)
 122                                 ftype = TYPE_F16;
 123                         else
 124                                 ftype = TYPE_F32;
 125
 126                         /* convert meta:input to mov: */
 127                         instr->category = 1;
 128                         instr->cat1.src_type = ftype;
 129                         instr->cat1.dst_type = ftype;
 130                 }
 131         }
 132
 133         /* recursively visit children: */
 134         foreach_ssa_src(src, instr)
 135                 ir3_instr_flatten(ctx, src);
 136 }
 137
 138 /* return >= 0 is # of phi's flattened, < 0 is error */
 139 int ir3_block_flatten(struct ir3_block *block)
 140 {
 141         struct ir3_flatten_ctx ctx = {
 142                         .block = block,
 143         };
 144         unsigned i;
 145
 146         ir3_clear_mark(block->shader);
 147         for(i = 0; i < block->noutputs; i++)
 148                 if (block->outputs[i])
 149                         ir3_instr_flatten(&ctx, block->outputs[i]);
 150
 151         return ctx.cnt;
 152 }