src/glsl/nir/nir_builder.h

   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 #ifndef NIR_BUILDER_H
  25 #define NIR_BUILDER_H
  26
  27 struct exec_list;
  28
  29 typedef struct nir_builder {
  30    struct exec_list *cf_node_list;
  31
  32    nir_block *before_block;
  33    nir_block *after_block;
  34
  35    nir_instr *before_instr;
  36    nir_instr *after_instr;
  37
  38    nir_shader *shader;
  39    nir_function_impl *impl;
  40 } nir_builder;
  41
  42 static inline void
  43 nir_builder_init(nir_builder *build, nir_function_impl *impl)
  44 {
  45    memset(build, 0, sizeof(*build));
  46    build->impl = impl;
  47    build->shader = impl->overload->function->shader;
  48 }
  49
  50 static inline void
  51 nir_builder_insert_after_cf_list(nir_builder *build,
  52                                  struct exec_list *cf_node_list)
  53 {
  54    build->cf_node_list = cf_node_list;
  55    build->before_block = NULL;
  56    build->after_block = NULL;
  57    build->before_instr = NULL;
  58    build->after_instr = NULL;
  59 }
  60
  61 static inline void
  62 nir_builder_insert_before_block(nir_builder *build,
  63                                 nir_block *block)
  64 {
  65    build->cf_node_list = NULL;
  66    build->before_block = block;
  67    build->after_block = NULL;
  68    build->before_instr = NULL;
  69    build->after_instr = NULL;
  70 }
  71
  72 static inline void
  73 nir_builder_insert_after_block(nir_builder *build,
  74                                 nir_block *block)
  75 {
  76    build->cf_node_list = NULL;
  77    build->before_block = NULL;
  78    build->after_block = block;
  79    build->before_instr = NULL;
  80    build->after_instr = NULL;
  81 }
  82
  83 static inline void
  84 nir_builder_insert_before_instr(nir_builder *build, nir_instr *before_instr)
  85 {
  86    build->cf_node_list = NULL;
  87    build->before_block = NULL;
  88    build->after_block = NULL;
  89    build->before_instr = before_instr;
  90    build->after_instr = NULL;
  91 }
  92
  93 static inline void
  94 nir_builder_insert_after_instr(nir_builder *build, nir_instr *after_instr)
  95 {
  96    build->cf_node_list = NULL;
  97    build->before_block = NULL;
  98    build->after_block = NULL;
  99    build->before_instr = NULL;
 100    build->after_instr = after_instr;
 101 }
 102
 103 static inline void
 104 nir_builder_instr_insert(nir_builder *build, nir_instr *instr)
 105 {
 106    if (build->cf_node_list) {
 107       nir_instr_insert_after_cf_list(build->cf_node_list, instr);
 108    } else if (build->before_block) {
 109       nir_instr_insert_before_block(build->before_block, instr);
 110    } else if (build->after_block) {
 111       nir_instr_insert_after_block(build->after_block, instr);
 112    } else if (build->before_instr) {
 113       nir_instr_insert_before(build->before_instr, instr);
 114    } else {
 115       assert(build->after_instr);
 116       nir_instr_insert_after(build->after_instr, instr);
 117       build->after_instr = instr;
 118    }
 119 }
 120
 121 static inline nir_ssa_def *
 122 nir_build_imm(nir_builder *build, unsigned num_components, nir_const_value value)
 123 {
 124    nir_load_const_instr *load_const =
 125       nir_load_const_instr_create(build->shader, num_components);
 126    if (!load_const)
 127       return NULL;
 128
 129    load_const->value = value;
 130
 131    nir_builder_instr_insert(build, &load_const->instr);
 132
 133    return &load_const->def;
 134 }
 135
 136 static inline nir_ssa_def *
 137 nir_imm_float(nir_builder *build, float x)
 138 {
 139    nir_const_value v = { { .f = {x, 0, 0, 0} } };
 140    return nir_build_imm(build, 1, v);
 141 }
 142
 143 static inline nir_ssa_def *
 144 nir_imm_vec4(nir_builder *build, float x, float y, float z, float w)
 145 {
 146    nir_const_value v = { { .f = {x, y, z, w} } };
 147    return nir_build_imm(build, 4, v);
 148 }
 149
 150 static inline nir_ssa_def *
 151 nir_imm_int(nir_builder *build, int x)
 152 {
 153    nir_const_value v = { { .i = {x, 0, 0, 0} } };
 154    return nir_build_imm(build, 1, v);
 155 }
 156
 157 static inline nir_ssa_def *
 158 nir_build_alu(nir_builder *build, nir_op op, nir_ssa_def *src0,
 159               nir_ssa_def *src1, nir_ssa_def *src2, nir_ssa_def *src3)
 160 {
 161    const nir_op_info *op_info = &nir_op_infos[op];
 162    nir_alu_instr *instr = nir_alu_instr_create(build->shader, op);
 163    if (!instr)
 164       return NULL;
 165
 166    instr->src[0].src = nir_src_for_ssa(src0);
 167    if (src1)
 168       instr->src[1].src = nir_src_for_ssa(src1);
 169    if (src2)
 170       instr->src[2].src = nir_src_for_ssa(src2);
 171    if (src3)
 172       instr->src[3].src = nir_src_for_ssa(src3);
 173
 174    /* Guess the number of components the destination temporary should have
 175     * based on our input sizes, if it's not fixed for the op.
 176     */
 177    unsigned num_components = op_info->output_size;
 178    if (num_components == 0) {
 179       for (unsigned i = 0; i < op_info->num_inputs; i++) {
 180          if (op_info->input_sizes[i] == 0)
 181             num_components = MAX2(num_components,
 182                                   instr->src[i].src.ssa->num_components);
 183       }
 184    }
 185    assert(num_components != 0);
 186
 187    /* Make sure we don't swizzle from outside of our source vector (like if a
 188     * scalar value was passed into a multiply with a vector).
 189     */
 190    for (unsigned i = 0; i < op_info->num_inputs; i++) {
 191       for (unsigned j = instr->src[i].src.ssa->num_components; j < 4; j++) {
 192          instr->src[i].swizzle[j] = instr->src[i].src.ssa->num_components - 1;
 193       }
 194    }
 195
 196    nir_ssa_dest_init(&instr->instr, &instr->dest.dest, num_components, NULL);
 197    instr->dest.write_mask = (1 << num_components) - 1;
 198
 199    nir_builder_instr_insert(build, &instr->instr);
 200
 201    return &instr->dest.dest.ssa;
 202 }
 203
 204 #define ALU1(op)                                                          \
 205 static inline nir_ssa_def *                                               \
 206 nir_##op(nir_builder *build, nir_ssa_def *src0)                           \
 207 {                                                                         \
 208    return nir_build_alu(build, nir_op_##op, src0, NULL, NULL, NULL);      \
 209 }
 210
 211 #define ALU2(op)                                                          \
 212 static inline nir_ssa_def *                                               \
 213 nir_##op(nir_builder *build, nir_ssa_def *src0, nir_ssa_def *src1)        \
 214 {                                                                         \
 215    return nir_build_alu(build, nir_op_##op, src0, src1, NULL, NULL);      \
 216 }
 217
 218 #define ALU3(op)                                                          \
 219 static inline nir_ssa_def *                                               \
 220 nir_##op(nir_builder *build, nir_ssa_def *src0,                           \
 221          nir_ssa_def *src1, nir_ssa_def *src2)                            \
 222 {                                                                         \
 223    return nir_build_alu(build, nir_op_##op, src0, src1, src2, NULL);      \
 224 }
 225
 226 #define ALU4(op)                                                          \
 227 static inline nir_ssa_def *                                               \
 228 nir_##op(nir_builder *build, nir_ssa_def *src0,                           \
 229          nir_ssa_def *src1, nir_ssa_def *src2, nir_ssa_def *src3)         \
 230 {                                                                         \
 231    return nir_build_alu(build, nir_op_##op, src0, src1, src2, src3);      \
 232 }
 233
 234 #include "nir_builder_opcodes.h"
 235
 236 /**
 237  * Similar to nir_fmov, but takes a nir_alu_src instead of a nir_ssa_def.
 238  */
 239 static inline nir_ssa_def *
 240 nir_fmov_alu(nir_builder *build, nir_alu_src src, unsigned num_components)
 241 {
 242    nir_alu_instr *mov = nir_alu_instr_create(build->shader, nir_op_fmov);
 243    nir_ssa_dest_init(&mov->instr, &mov->dest.dest, num_components, NULL);
 244    mov->dest.write_mask = (1 << num_components) - 1;
 245    mov->src[0] = src;
 246    nir_builder_instr_insert(build, &mov->instr);
 247
 248    return &mov->dest.dest.ssa;
 249 }
 250
 251 static inline nir_ssa_def *
 252 nir_imov_alu(nir_builder *build, nir_alu_src src, unsigned num_components)
 253 {
 254    nir_alu_instr *mov = nir_alu_instr_create(build->shader, nir_op_imov);
 255    nir_ssa_dest_init(&mov->instr, &mov->dest.dest, num_components, NULL);
 256    mov->dest.write_mask = (1 << num_components) - 1;
 257    mov->src[0] = src;
 258    nir_builder_instr_insert(build, &mov->instr);
 259
 260    return &mov->dest.dest.ssa;
 261 }
 262
 263 /**
 264  * Construct an fmov or imov that reswizzles the source's components.
 265  */
 266 static inline nir_ssa_def *
 267 nir_swizzle(nir_builder *build, nir_ssa_def *src, unsigned swiz[4],
 268             unsigned num_components, bool use_fmov)
 269 {
 270    nir_alu_src alu_src = { NIR_SRC_INIT };
 271    alu_src.src = nir_src_for_ssa(src);
 272    for (int i = 0; i < 4; i++)
 273       alu_src.swizzle[i] = swiz[i];
 274
 275    return use_fmov ? nir_fmov_alu(build, alu_src, num_components) :
 276                      nir_imov_alu(build, alu_src, num_components);
 277 }
 278
 279 /* Selects the right fdot given the number of components in each source. */
 280 static inline nir_ssa_def *
 281 nir_fdot(nir_builder *build, nir_ssa_def *src0, nir_ssa_def *src1)
 282 {
 283    assert(src0->num_components == src1->num_components);
 284    switch (src0->num_components) {
 285    case 1: return nir_fmul(build, src0, src1);
 286    case 2: return nir_fdot2(build, src0, src1);
 287    case 3: return nir_fdot3(build, src0, src1);
 288    case 4: return nir_fdot4(build, src0, src1);
 289    default:
 290       unreachable("bad component size");
 291    }
 292
 293    return NULL;
 294 }
 295
 296 /**
 297  * Turns a nir_src into a nir_ssa_def * so it can be passed to
 298  * nir_build_alu()-based builder calls.
 299  */
 300 static inline nir_ssa_def *
 301 nir_ssa_for_src(nir_builder *build, nir_src src, int num_components)
 302 {
 303    if (src.is_ssa && src.ssa->num_components == num_components)
 304       return src.ssa;
 305
 306    nir_alu_src alu = { NIR_SRC_INIT };
 307    alu.src = src;
 308    for (int j = 0; j < 4; j++)
 309       alu.swizzle[j] = j;
 310
 311    return nir_imov_alu(build, alu, num_components);
 312 }
 313
 314 #endif /* NIR_BUILDER_H */