src/compiler/nir/nir_lower_alu_to_scalar.c

   1 /*
   2  * Copyright © 2014-2015 Broadcom
   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 "nir.h"
  25 #include "nir_builder.h"
  26
  27 /** @file nir_lower_alu_to_scalar.c
  28  *
  29  * Replaces nir_alu_instr operations with more than one channel used in the
  30  * arguments with individual per-channel operations.
  31  */
  32
  33 static void
  34 nir_alu_ssa_dest_init(nir_alu_instr *instr, unsigned num_components,
  35                       unsigned bit_size)
  36 {
  37    nir_ssa_dest_init(&instr->instr, &instr->dest.dest, num_components,
  38                      bit_size, NULL);
  39    instr->dest.write_mask = (1 << num_components) - 1;
  40 }
  41
  42 static void
  43 lower_reduction(nir_alu_instr *instr, nir_op chan_op, nir_op merge_op,
  44                 nir_builder *builder)
  45 {
  46    unsigned num_components = nir_op_infos[instr->op].input_sizes[0];
  47
  48    nir_ssa_def *last = NULL;
  49    for (unsigned i = 0; i < num_components; i++) {
  50       nir_alu_instr *chan = nir_alu_instr_create(builder->shader, chan_op);
  51       nir_alu_ssa_dest_init(chan, 1, instr->dest.dest.ssa.bit_size);
  52       nir_alu_src_copy(&chan->src[0], &instr->src[0], chan);
  53       chan->src[0].swizzle[0] = chan->src[0].swizzle[i];
  54       if (nir_op_infos[chan_op].num_inputs > 1) {
  55          assert(nir_op_infos[chan_op].num_inputs == 2);
  56          nir_alu_src_copy(&chan->src[1], &instr->src[1], chan);
  57          chan->src[1].swizzle[0] = chan->src[1].swizzle[i];
  58       }
  59       chan->exact = instr->exact;
  60
  61       nir_builder_instr_insert(builder, &chan->instr);
  62
  63       if (i == 0) {
  64          last = &chan->dest.dest.ssa;
  65       } else {
  66          last = nir_build_alu(builder, merge_op,
  67                               last, &chan->dest.dest.ssa, NULL, NULL);
  68       }
  69    }
  70
  71    assert(instr->dest.write_mask == 1);
  72    nir_ssa_def_rewrite_uses(&instr->dest.dest.ssa, nir_src_for_ssa(last));
  73    nir_instr_remove(&instr->instr);
  74 }
  75
  76 static bool
  77 lower_alu_instr_scalar(nir_alu_instr *instr, nir_builder *b)
  78 {
  79    unsigned num_src = nir_op_infos[instr->op].num_inputs;
  80    unsigned i, chan;
  81
  82    assert(instr->dest.dest.is_ssa);
  83    assert(instr->dest.write_mask != 0);
  84
  85    b->cursor = nir_before_instr(&instr->instr);
  86    b->exact = instr->exact;
  87
  88 #define LOWER_REDUCTION(name, chan, merge) \
  89    case name##2: \
  90    case name##3: \
  91    case name##4: \
  92       lower_reduction(instr, chan, merge, b); \
  93       return true;
  94
  95    switch (instr->op) {
  96    case nir_op_vec4:
  97    case nir_op_vec3:
  98    case nir_op_vec2:
  99       /* We don't need to scalarize these ops, they're the ones generated to
 100        * group up outputs into a value that can be SSAed.
 101        */
 102       return false;
 103
 104    case nir_op_pack_half_2x16:
 105       if (!b->shader->options->lower_pack_half_2x16)
 106          return false;
 107
 108       nir_ssa_def *val =
 109          nir_pack_half_2x16_split(b, nir_channel(b, instr->src[0].src.ssa,
 110                                                  instr->src[0].swizzle[0]),
 111                                      nir_channel(b, instr->src[0].src.ssa,
 112                                                  instr->src[0].swizzle[1]));
 113
 114       nir_ssa_def_rewrite_uses(&instr->dest.dest.ssa, nir_src_for_ssa(val));
 115       nir_instr_remove(&instr->instr);
 116       return true;
 117
 118    case nir_op_unpack_unorm_4x8:
 119    case nir_op_unpack_snorm_4x8:
 120    case nir_op_unpack_unorm_2x16:
 121    case nir_op_unpack_snorm_2x16:
 122       /* There is no scalar version of these ops, unless we were to break it
 123        * down to bitshifts and math (which is definitely not intended).
 124        */
 125       return false;
 126
 127    case nir_op_unpack_half_2x16: {
 128       if (!b->shader->options->lower_unpack_half_2x16)
 129          return false;
 130
 131       nir_ssa_def *comps[2];
 132       comps[0] = nir_unpack_half_2x16_split_x(b, instr->src[0].src.ssa);
 133       comps[1] = nir_unpack_half_2x16_split_y(b, instr->src[0].src.ssa);
 134       nir_ssa_def *vec = nir_vec(b, comps, 2);
 135
 136       nir_ssa_def_rewrite_uses(&instr->dest.dest.ssa, nir_src_for_ssa(vec));
 137       nir_instr_remove(&instr->instr);
 138       return true;
 139    }
 140
 141    case nir_op_pack_uvec2_to_uint: {
 142       assert(b->shader->options->lower_pack_snorm_2x16 ||
 143              b->shader->options->lower_pack_unorm_2x16);
 144
 145       nir_ssa_def *word =
 146          nir_extract_u16(b, instr->src[0].src.ssa, nir_imm_int(b, 0));
 147       nir_ssa_def *val =
 148          nir_ior(b, nir_ishl(b, nir_channel(b, word, 1), nir_imm_int(b, 16)),
 149                                 nir_channel(b, word, 0));
 150
 151       nir_ssa_def_rewrite_uses(&instr->dest.dest.ssa, nir_src_for_ssa(val));
 152       nir_instr_remove(&instr->instr);
 153       break;
 154    }
 155
 156    case nir_op_pack_uvec4_to_uint: {
 157       assert(b->shader->options->lower_pack_snorm_4x8 ||
 158              b->shader->options->lower_pack_unorm_4x8);
 159
 160       nir_ssa_def *byte =
 161          nir_extract_u8(b, instr->src[0].src.ssa, nir_imm_int(b, 0));
 162       nir_ssa_def *val =
 163          nir_ior(b, nir_ior(b, nir_ishl(b, nir_channel(b, byte, 3), nir_imm_int(b, 24)),
 164                                nir_ishl(b, nir_channel(b, byte, 2), nir_imm_int(b, 16))),
 165                     nir_ior(b, nir_ishl(b, nir_channel(b, byte, 1), nir_imm_int(b, 8)),
 166                                nir_channel(b, byte, 0)));
 167
 168       nir_ssa_def_rewrite_uses(&instr->dest.dest.ssa, nir_src_for_ssa(val));
 169       nir_instr_remove(&instr->instr);
 170       break;
 171    }
 172
 173    case nir_op_fdph: {
 174       nir_ssa_def *sum[4];
 175       for (unsigned i = 0; i < 3; i++) {
 176          sum[i] = nir_fmul(b, nir_channel(b, instr->src[0].src.ssa,
 177                                           instr->src[0].swizzle[i]),
 178                               nir_channel(b, instr->src[1].src.ssa,
 179                                           instr->src[1].swizzle[i]));
 180       }
 181       sum[3] = nir_channel(b, instr->src[1].src.ssa, instr->src[1].swizzle[3]);
 182
 183       nir_ssa_def *val = nir_fadd(b, nir_fadd(b, sum[0], sum[1]),
 184                                      nir_fadd(b, sum[2], sum[3]));
 185
 186       nir_ssa_def_rewrite_uses(&instr->dest.dest.ssa, nir_src_for_ssa(val));
 187       nir_instr_remove(&instr->instr);
 188       return true;
 189    }
 190
 191    case nir_op_unpack_double_2x32:
 192       return false;
 193
 194       LOWER_REDUCTION(nir_op_fdot, nir_op_fmul, nir_op_fadd);
 195       LOWER_REDUCTION(nir_op_ball_fequal, nir_op_feq, nir_op_iand);
 196       LOWER_REDUCTION(nir_op_ball_iequal, nir_op_ieq, nir_op_iand);
 197       LOWER_REDUCTION(nir_op_bany_fnequal, nir_op_fne, nir_op_ior);
 198       LOWER_REDUCTION(nir_op_bany_inequal, nir_op_ine, nir_op_ior);
 199       LOWER_REDUCTION(nir_op_fall_equal, nir_op_seq, nir_op_fand);
 200       LOWER_REDUCTION(nir_op_fany_nequal, nir_op_sne, nir_op_for);
 201
 202    default:
 203       break;
 204    }
 205
 206    if (instr->dest.dest.ssa.num_components == 1)
 207       return false;
 208
 209    unsigned num_components = instr->dest.dest.ssa.num_components;
 210    nir_ssa_def *comps[] = { NULL, NULL, NULL, NULL };
 211
 212    for (chan = 0; chan < 4; chan++) {
 213       if (!(instr->dest.write_mask & (1 << chan)))
 214          continue;
 215
 216       nir_alu_instr *lower = nir_alu_instr_create(b->shader, instr->op);
 217       for (i = 0; i < num_src; i++) {
 218          /* We only handle same-size-as-dest (input_sizes[] == 0) or scalar
 219           * args (input_sizes[] == 1).
 220           */
 221          assert(nir_op_infos[instr->op].input_sizes[i] < 2);
 222          unsigned src_chan = (nir_op_infos[instr->op].input_sizes[i] == 1 ?
 223                               0 : chan);
 224
 225          nir_alu_src_copy(&lower->src[i], &instr->src[i], lower);
 226          for (int j = 0; j < 4; j++)
 227             lower->src[i].swizzle[j] = instr->src[i].swizzle[src_chan];
 228       }
 229
 230       nir_alu_ssa_dest_init(lower, 1, instr->dest.dest.ssa.bit_size);
 231       lower->dest.saturate = instr->dest.saturate;
 232       comps[chan] = &lower->dest.dest.ssa;
 233       lower->exact = instr->exact;
 234
 235       nir_builder_instr_insert(b, &lower->instr);
 236    }
 237
 238    nir_ssa_def *vec = nir_vec(b, comps, num_components);
 239
 240    nir_ssa_def_rewrite_uses(&instr->dest.dest.ssa, nir_src_for_ssa(vec));
 241
 242    nir_instr_remove(&instr->instr);
 243    return true;
 244 }
 245
 246 static bool
 247 nir_lower_alu_to_scalar_impl(nir_function_impl *impl)
 248 {
 249    nir_builder builder;
 250    nir_builder_init(&builder, impl);
 251    bool progress = false;
 252
 253    nir_foreach_block(block, impl) {
 254       nir_foreach_instr_safe(instr, block) {
 255          if (instr->type == nir_instr_type_alu) {
 256             progress = lower_alu_instr_scalar(nir_instr_as_alu(instr),
 257                                               &builder) || progress;
 258          }
 259       }
 260    }
 261
 262    nir_metadata_preserve(impl, nir_metadata_block_index |
 263                                nir_metadata_dominance);
 264
 265    return progress;
 266 }
 267
 268 bool
 269 nir_lower_alu_to_scalar(nir_shader *shader)
 270 {
 271    bool progress = false;
 272
 273    nir_foreach_function(function, shader) {
 274       if (function->impl)
 275          progress = nir_lower_alu_to_scalar_impl(function->impl) || progress;
 276    }
 277
 278    return progress;
 279 }