1/* 2 * Copyright © 2018 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 24#include "nir.h" 25#include "nir_builder.h" 26#include "util/fast_idiv_by_const.h" 27#include "util/u_math.h" 28 29static nir_ssa_def * 30build_udiv(nir_builder *b, nir_ssa_def *n, uint64_t d) 31{ 32 if (d == 0) { 33 return nir_imm_intN_t(b, 0, n->bit_size); 34 } else if (util_is_power_of_two_or_zero64(d)) { 35 return nir_ushr_imm(b, n, util_logbase2_64(d)); 36 } else { 37 struct util_fast_udiv_info m = 38 util_compute_fast_udiv_info(d, n->bit_size, n->bit_size); 39 40 if (m.pre_shift) 41 n = nir_ushr_imm(b, n, m.pre_shift); 42 if (m.increment) 43 n = nir_uadd_sat(b, n, nir_imm_intN_t(b, m.increment, n->bit_size)); 44 n = nir_umul_high(b, n, nir_imm_intN_t(b, m.multiplier, n->bit_size)); 45 if (m.post_shift) 46 n = nir_ushr_imm(b, n, m.post_shift); 47 48 return n; 49 } 50} 51 52static nir_ssa_def * 53build_umod(nir_builder *b, nir_ssa_def *n, uint64_t d) 54{ 55 if (d == 0) { 56 return nir_imm_intN_t(b, 0, n->bit_size); 57 } else if (util_is_power_of_two_or_zero64(d)) { 58 return nir_iand(b, n, nir_imm_intN_t(b, d - 1, n->bit_size)); 59 } else { 60 return nir_isub(b, n, nir_imul(b, build_udiv(b, n, d), 61 nir_imm_intN_t(b, d, n->bit_size))); 62 } 63} 64 65static nir_ssa_def * 66build_idiv(nir_builder *b, nir_ssa_def *n, int64_t d) 67{ 68 int64_t int_min = u_intN_min(n->bit_size); 69 if (d == int_min) 70 return nir_b2i(b, nir_ieq_imm(b, n, int_min), n->bit_size); 71 72 uint64_t abs_d = d < 0 ? -d : d; 73 74 if (d == 0) { 75 return nir_imm_intN_t(b, 0, n->bit_size); 76 } else if (d == 1) { 77 return n; 78 } else if (d == -1) { 79 return nir_ineg(b, n); 80 } else if (util_is_power_of_two_or_zero64(abs_d)) { 81 nir_ssa_def *uq = nir_ushr_imm(b, nir_iabs(b, n), util_logbase2_64(abs_d)); 82 nir_ssa_def *n_neg = nir_ilt(b, n, nir_imm_intN_t(b, 0, n->bit_size)); 83 nir_ssa_def *neg = d < 0 ? nir_inot(b, n_neg) : n_neg; 84 return nir_bcsel(b, neg, nir_ineg(b, uq), uq); 85 } else { 86 struct util_fast_sdiv_info m = 87 util_compute_fast_sdiv_info(d, n->bit_size); 88 89 nir_ssa_def *res = 90 nir_imul_high(b, n, nir_imm_intN_t(b, m.multiplier, n->bit_size)); 91 if (d > 0 && m.multiplier < 0) 92 res = nir_iadd(b, res, n); 93 if (d < 0 && m.multiplier > 0) 94 res = nir_isub(b, res, n); 95 if (m.shift) 96 res = nir_ishr_imm(b, res, m.shift); 97 res = nir_iadd(b, res, nir_ushr_imm(b, res, n->bit_size - 1)); 98 99 return res; 100 } 101} 102 103static nir_ssa_def * 104build_irem(nir_builder *b, nir_ssa_def *n, int64_t d) 105{ 106 int64_t int_min = u_intN_min(n->bit_size); 107 if (d == 0) { 108 return nir_imm_intN_t(b, 0, n->bit_size); 109 } else if (d == int_min) { 110 return nir_bcsel(b, nir_ieq_imm(b, n, int_min), nir_imm_intN_t(b, 0, n->bit_size), n); 111 } else { 112 d = d < 0 ? -d : d; 113 if (util_is_power_of_two_or_zero64(d)) { 114 nir_ssa_def *tmp = nir_bcsel(b, nir_ilt(b, n, nir_imm_intN_t(b, 0, n->bit_size)), 115 nir_iadd_imm(b, n, d - 1), n); 116 return nir_isub(b, n, nir_iand_imm(b, tmp, -d)); 117 } else { 118 return nir_isub(b, n, nir_imul(b, build_idiv(b, n, d), 119 nir_imm_intN_t(b, d, n->bit_size))); 120 } 121 } 122} 123 124static nir_ssa_def * 125build_imod(nir_builder *b, nir_ssa_def *n, int64_t d) 126{ 127 int64_t int_min = u_intN_min(n->bit_size); 128 if (d == 0) { 129 return nir_imm_intN_t(b, 0, n->bit_size); 130 } else if (d == int_min) { 131 nir_ssa_def *int_min_def = nir_imm_intN_t(b, int_min, n->bit_size); 132 nir_ssa_def *is_neg_not_int_min = nir_ult(b, int_min_def, n); 133 nir_ssa_def *is_zero = nir_ieq_imm(b, n, 0); 134 return nir_bcsel(b, nir_ior(b, is_neg_not_int_min, is_zero), n, nir_iadd(b, int_min_def, n)); 135 } else if (d > 0 && util_is_power_of_two_or_zero64(d)) { 136 return nir_iand(b, n, nir_imm_intN_t(b, d - 1, n->bit_size)); 137 } else if (d < 0 && util_is_power_of_two_or_zero64(-d)) { 138 nir_ssa_def *d_def = nir_imm_intN_t(b, d, n->bit_size); 139 nir_ssa_def *res = nir_ior(b, n, d_def); 140 return nir_bcsel(b, nir_ieq(b, res, d_def), nir_imm_intN_t(b, 0, n->bit_size), res); 141 } else { 142 nir_ssa_def *rem = build_irem(b, n, d); 143 nir_ssa_def *zero = nir_imm_intN_t(b, 0, n->bit_size); 144 nir_ssa_def *sign_same = d < 0 ? nir_ilt(b, n, zero) : nir_ige(b, n, zero); 145 nir_ssa_def *rem_zero = nir_ieq(b, rem, zero); 146 return nir_bcsel(b, nir_ior(b, rem_zero, sign_same), rem, nir_iadd_imm(b, rem, d)); 147 } 148} 149 150static bool 151nir_opt_idiv_const_instr(nir_builder *b, nir_alu_instr *alu) 152{ 153 assert(alu->dest.dest.is_ssa); 154 assert(alu->src[0].src.is_ssa && alu->src[1].src.is_ssa); 155 156 if (!nir_src_is_const(alu->src[1].src)) 157 return false; 158 159 unsigned bit_size = alu->src[1].src.ssa->bit_size; 160 161 b->cursor = nir_before_instr(&alu->instr); 162 163 nir_ssa_def *q[NIR_MAX_VEC_COMPONENTS]; 164 for (unsigned comp = 0; comp < alu->dest.dest.ssa.num_components; comp++) { 165 /* Get the numerator for the channel */ 166 nir_ssa_def *n = nir_channel(b, alu->src[0].src.ssa, 167 alu->src[0].swizzle[comp]); 168 169 /* Get the denominator for the channel */ 170 int64_t d = nir_src_comp_as_int(alu->src[1].src, 171 alu->src[1].swizzle[comp]); 172 173 nir_alu_type d_type = nir_op_infos[alu->op].input_types[1]; 174 if (nir_alu_type_get_base_type(d_type) == nir_type_uint) { 175 /* The code above sign-extended. If we're lowering an unsigned op, 176 * we need to mask it off to the correct number of bits so that a 177 * cast to uint64_t will do the right thing. 178 */ 179 if (bit_size < 64) 180 d &= (1ull << bit_size) - 1; 181 } 182 183 switch (alu->op) { 184 case nir_op_udiv: 185 q[comp] = build_udiv(b, n, d); 186 break; 187 case nir_op_idiv: 188 q[comp] = build_idiv(b, n, d); 189 break; 190 case nir_op_umod: 191 q[comp] = build_umod(b, n, d); 192 break; 193 case nir_op_imod: 194 q[comp] = build_imod(b, n, d); 195 break; 196 case nir_op_irem: 197 q[comp] = build_irem(b, n, d); 198 break; 199 default: 200 unreachable("Unknown integer division op"); 201 } 202 } 203 204 nir_ssa_def *qvec = nir_vec(b, q, alu->dest.dest.ssa.num_components); 205 nir_ssa_def_rewrite_uses(&alu->dest.dest.ssa, qvec); 206 nir_instr_remove(&alu->instr); 207 208 return true; 209} 210 211static bool 212nir_opt_idiv_const_impl(nir_function_impl *impl, unsigned min_bit_size) 213{ 214 bool progress = false; 215 216 nir_builder b; 217 nir_builder_init(&b, impl); 218 219 nir_foreach_block(block, impl) { 220 nir_foreach_instr_safe(instr, block) { 221 if (instr->type != nir_instr_type_alu) 222 continue; 223 224 nir_alu_instr *alu = nir_instr_as_alu(instr); 225 if (alu->op != nir_op_udiv && 226 alu->op != nir_op_idiv && 227 alu->op != nir_op_umod && 228 alu->op != nir_op_imod && 229 alu->op != nir_op_irem) 230 continue; 231 232 assert(alu->dest.dest.is_ssa); 233 if (alu->dest.dest.ssa.bit_size < min_bit_size) 234 continue; 235 236 progress |= nir_opt_idiv_const_instr(&b, alu); 237 } 238 } 239 240 if (progress) { 241 nir_metadata_preserve(impl, nir_metadata_block_index | 242 nir_metadata_dominance); 243 } else { 244 nir_metadata_preserve(impl, nir_metadata_all); 245 } 246 247 return progress; 248} 249 250bool 251nir_opt_idiv_const(nir_shader *shader, unsigned min_bit_size) 252{ 253 bool progress = false; 254 255 nir_foreach_function(function, shader) { 256 if (function->impl) 257 progress |= nir_opt_idiv_const_impl(function->impl, min_bit_size); 258 } 259 260 return progress; 261} 262