src/glsl/nir/nir_lower_idiv.c

   1 /*
   2  * Copyright © 2015 Red Hat
   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  *    Rob Clark <robclark@freedesktop.org>
  25  */
  26
  27 #include "nir.h"
  28 #include "nir_builder.h"
  29
  30 /* Lowers idiv/udiv/umod
  31  * Based on NV50LegalizeSSA::handleDIV()
  32  *
  33  * Note that this is probably not enough precision for compute shaders.
  34  * Perhaps we want a second higher precision (looping) version of this?
  35  * Or perhaps we assume if you can do compute shaders you can also
  36  * branch out to a pre-optimized shader library routine..
  37  */
  38
  39 static void
  40 convert_instr(nir_builder *bld, nir_alu_instr *alu)
  41 {
  42    nir_ssa_def *numer, *denom, *af, *bf, *a, *b, *q, *r;
  43    nir_op op = alu->op;
  44    bool is_signed;
  45
  46    if ((op != nir_op_idiv) &&
  47        (op != nir_op_udiv) &&
  48        (op != nir_op_umod))
  49       return;
  50
  51    is_signed = (op == nir_op_idiv);
  52
  53    nir_builder_insert_before_instr(bld, &alu->instr);
  54
  55    numer = nir_ssa_for_src(bld, alu->src[0].src,
  56                            nir_ssa_alu_instr_src_components(alu, 0));
  57    denom = nir_ssa_for_src(bld, alu->src[1].src,
  58                            nir_ssa_alu_instr_src_components(alu, 1));
  59
  60    if (is_signed) {
  61       af = nir_i2f(bld, numer);
  62       bf = nir_i2f(bld, denom);
  63       af = nir_fabs(bld, af);
  64       bf = nir_fabs(bld, bf);
  65       a  = nir_iabs(bld, numer);
  66       b  = nir_iabs(bld, denom);
  67    } else {
  68       af = nir_u2f(bld, numer);
  69       bf = nir_u2f(bld, denom);
  70       a  = numer;
  71       b  = denom;
  72    }
  73
  74    /* get first result: */
  75    bf = nir_frcp(bld, bf);
  76    bf = nir_isub(bld, bf, nir_imm_int(bld, 2));  /* yes, really */
  77    q  = nir_fmul(bld, af, bf);
  78
  79    if (is_signed) {
  80       q = nir_f2i(bld, q);
  81    } else {
  82       q = nir_f2u(bld, q);
  83    }
  84
  85    /* get error of first result: */
  86    r = nir_imul(bld, q, b);
  87    r = nir_isub(bld, a, r);
  88    r = nir_u2f(bld, r);
  89    r = nir_fmul(bld, r, bf);
  90    r = nir_f2u(bld, r);
  91
  92    /* add quotients: */
  93    q = nir_iadd(bld, q, r);
  94
  95    /* correction: if modulus >= divisor, add 1 */
  96    r = nir_imul(bld, q, b);
  97    r = nir_isub(bld, a, r);
  98
  99    r = nir_ige(bld, r, b);
 100    r = nir_b2i(bld, r);
 101
 102    q = nir_iadd(bld, q, r);
 103    if (is_signed)  {
 104       /* fix the sign: */
 105       r = nir_ixor(bld, numer, denom);
 106       r = nir_ushr(bld, r, nir_imm_int(bld, 31));
 107       r = nir_i2b(bld, r);
 108       b = nir_ineg(bld, q);
 109       q = nir_bcsel(bld, r, b, q);
 110    }
 111
 112    if (op == nir_op_umod) {
 113       /* division result in q */
 114       r = nir_imul(bld, q, b);
 115       q = nir_isub(bld, a, r);
 116    }
 117
 118    assert(alu->dest.dest.is_ssa);
 119    nir_ssa_def_rewrite_uses(&alu->dest.dest.ssa,
 120                             nir_src_for_ssa(q),
 121                             ralloc_parent(alu));
 122 }
 123
 124 static bool
 125 convert_block(nir_block *block, void *state)
 126 {
 127    nir_builder *b = state;
 128
 129    nir_foreach_instr_safe(block, instr) {
 130       if (instr->type == nir_instr_type_alu)
 131          convert_instr(b, nir_instr_as_alu(instr));
 132    }
 133
 134    return true;
 135 }
 136
 137 static void
 138 convert_impl(nir_function_impl *impl)
 139 {
 140    nir_builder b;
 141    nir_builder_init(&b, impl);
 142
 143    nir_foreach_block(impl, convert_block, &b);
 144    nir_metadata_preserve(impl, nir_metadata_block_index |
 145                                nir_metadata_dominance);
 146 }
 147
 148 void
 149 nir_lower_idiv(nir_shader *shader)
 150 {
 151    nir_foreach_overload(shader, overload) {
 152       if (overload->impl)
 153          convert_impl(overload->impl);
 154    }
 155 }