src/freedreno/ir3/ir3_cf.c

   1 /*
   2  * Copyright (C) 2019 Google.
   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 FROM,
  20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  21  * SOFTWARE.
  22  */
  23
  24 #include "util/ralloc.h"
  25
  26 #include "ir3.h"
  27
  28 static bool
  29 is_fp16_conv(struct ir3_instruction *instr)
  30 {
  31         if (instr->opc != OPC_MOV)
  32                 return false;
  33
  34         struct ir3_register *dst = instr->regs[0];
  35         struct ir3_register *src = instr->regs[1];
  36
  37         /* disallow conversions that cannot be folded into
  38          * alu instructions:
  39          */
  40         if (dst->flags & (IR3_REG_EVEN | IR3_REG_POS_INF))
  41                 return false;
  42
  43         if (dst->flags & (IR3_REG_RELATIV | IR3_REG_ARRAY))
  44                 return false;
  45         if (src->flags & (IR3_REG_RELATIV | IR3_REG_ARRAY))
  46                 return false;
  47
  48         if (instr->cat1.src_type == TYPE_F32 &&
  49                         instr->cat1.dst_type == TYPE_F16)
  50                 return true;
  51
  52         if (instr->cat1.src_type == TYPE_F16 &&
  53                         instr->cat1.dst_type == TYPE_F32)
  54                 return true;
  55
  56         return false;
  57 }
  58
  59 static bool
  60 all_uses_fp16_conv(struct ir3_instruction *conv_src)
  61 {
  62         foreach_ssa_use (use, conv_src)
  63                 if (!is_fp16_conv(use))
  64                         return false;
  65         return true;
  66 }
  67
  68 static void
  69 rewrite_uses(struct ir3_instruction *conv, struct ir3_instruction *replace)
  70 {
  71         foreach_ssa_use (use, conv) {
  72                 foreach_ssa_src_n (src, n, use) {
  73                         if (src == conv)
  74                                 use->regs[n]->instr = replace;
  75                 }
  76         }
  77 }
  78
  79 static bool
  80 try_conversion_folding(struct ir3_instruction *conv)
  81 {
  82         struct ir3_instruction *src;
  83
  84         if (!is_fp16_conv(conv))
  85                 return false;
  86
  87         /* NOTE: we can have non-ssa srcs after copy propagation: */
  88         src = ssa(conv->regs[1]);
  89         if (!src)
  90                 return false;
  91
  92         if (!is_alu(src))
  93                 return false;
  94
  95         /* avoid folding f2f32(f2f16) together, in cases where this is legal to
  96          * do (glsl) nir should have handled that for us already:
  97          */
  98         if (is_fp16_conv(src))
  99                 return false;
 100
 101         switch (src->opc) {
 102         case OPC_SEL_B32:
 103         case OPC_SEL_B16:
 104         case OPC_MAX_F:
 105         case OPC_MIN_F:
 106         case OPC_SIGN_F:
 107         case OPC_ABSNEG_F:
 108                 return false;
 109         case OPC_MOV:
 110                 /* if src is a "cov" and type doesn't match, then it can't be folded
 111                  * for example cov.u32u16+cov.f16f32 can't be folded to cov.u32f32
 112                  */
 113                 if (src->cat1.dst_type != src->cat1.src_type &&
 114                         conv->cat1.src_type != src->cat1.dst_type)
 115                         return false;
 116         default:
 117                 break;
 118         }
 119
 120         if (!all_uses_fp16_conv(src))
 121                 return false;
 122
 123         if (src->opc == OPC_MOV) {
 124                 if (src->cat1.dst_type == src->cat1.src_type) {
 125                         /* If we're folding a conversion into a bitwise move, we need to
 126                          * change the dst type to F32 to get the right behavior, since we
 127                          * could be moving a float with a u32.u32 move.
 128                          */
 129                         src->cat1.dst_type = conv->cat1.dst_type;
 130                         src->cat1.src_type = conv->cat1.src_type;
 131                 } else {
 132                         /* Otherwise, for typechanging movs, we can just change the dst
 133                          * type to F16 to collaps the two conversions.  For example
 134                          * cov.s32f32 follwed by cov.f32f16 becomes cov.s32f16.
 135                          */
 136                         src->cat1.dst_type = conv->cat1.dst_type;
 137                 }
 138         }
 139
 140         if (conv->regs[0]->flags & IR3_REG_HALF) {
 141                 src->regs[0]->flags |= IR3_REG_HALF;
 142         } else {
 143                 src->regs[0]->flags &= ~IR3_REG_HALF;
 144         }
 145
 146         rewrite_uses(conv, src);
 147
 148         return true;
 149 }
 150
 151 bool
 152 ir3_cf(struct ir3 *ir)
 153 {
 154         void *mem_ctx = ralloc_context(NULL);
 155         bool progress = false;
 156
 157         ir3_find_ssa_uses(ir, mem_ctx, false);
 158
 159         foreach_block (block, &ir->block_list) {
 160                 foreach_instr_safe (instr, &block->instr_list) {
 161                         progress |= try_conversion_folding(instr);
 162                 }
 163         }
 164
 165         ralloc_free(mem_ctx);
 166
 167         return progress;
 168 }