src/compiler/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 #include "nir_control_flow.h"
  28
  29 struct exec_list;
  30
  31 typedef struct nir_builder {
  32    nir_cursor cursor;
  33
  34    /* Whether new ALU instructions will be marked "exact" */
  35    bool exact;
  36
  37    nir_shader *shader;
  38    nir_function_impl *impl;
  39 } nir_builder;
  40
  41 static inline void
  42 nir_builder_init(nir_builder *build, nir_function_impl *impl)
  43 {
  44    memset(build, 0, sizeof(*build));
  45    build->exact = false;
  46    build->impl = impl;
  47    build->shader = impl->function->shader;
  48 }
  49
  50 static inline void
  51 nir_builder_init_simple_shader(nir_builder *build, void *mem_ctx,
  52                                gl_shader_stage stage,
  53                                const nir_shader_compiler_options *options)
  54 {
  55    build->shader = nir_shader_create(mem_ctx, stage, options);
  56    nir_function *func = nir_function_create(build->shader, "main");
  57    build->exact = false;
  58    build->impl = nir_function_impl_create(func);
  59    build->cursor = nir_after_cf_list(&build->impl->body);
  60 }
  61
  62 static inline void
  63 nir_builder_instr_insert(nir_builder *build, nir_instr *instr)
  64 {
  65    nir_instr_insert(build->cursor, instr);
  66
  67    /* Move the cursor forward. */
  68    build->cursor = nir_after_instr(instr);
  69 }
  70
  71 static inline void
  72 nir_builder_cf_insert(nir_builder *build, nir_cf_node *cf)
  73 {
  74    nir_cf_node_insert(build->cursor, cf);
  75 }
  76
  77 static inline nir_ssa_def *
  78 nir_ssa_undef(nir_builder *build, unsigned num_components, unsigned bit_size)
  79 {
  80    nir_ssa_undef_instr *undef =
  81       nir_ssa_undef_instr_create(build->shader, num_components, bit_size);
  82    if (!undef)
  83       return NULL;
  84
  85    nir_instr_insert(nir_before_cf_list(&build->impl->body), &undef->instr);
  86
  87    return &undef->def;
  88 }
  89
  90 static inline nir_ssa_def *
  91 nir_build_imm(nir_builder *build, unsigned num_components,
  92               unsigned bit_size, nir_const_value value)
  93 {
  94    nir_load_const_instr *load_const =
  95       nir_load_const_instr_create(build->shader, num_components, bit_size);
  96    if (!load_const)
  97       return NULL;
  98
  99    load_const->value = value;
 100
 101    nir_builder_instr_insert(build, &load_const->instr);
 102
 103    return &load_const->def;
 104 }
 105
 106 static inline nir_ssa_def *
 107 nir_imm_float(nir_builder *build, float x)
 108 {
 109    nir_const_value v;
 110
 111    memset(&v, 0, sizeof(v));
 112    v.f32[0] = x;
 113
 114    return nir_build_imm(build, 1, 32, v);
 115 }
 116
 117 static inline nir_ssa_def *
 118 nir_imm_double(nir_builder *build, double x)
 119 {
 120    nir_const_value v = { { .f64 = {x, 0, 0, 0} } };
 121    nir_ssa_def *def = nir_build_imm(build, 1, 64, v);
 122    return def;
 123 }
 124
 125 static inline nir_ssa_def *
 126 nir_imm_vec4(nir_builder *build, float x, float y, float z, float w)
 127 {
 128    nir_const_value v;
 129
 130    memset(&v, 0, sizeof(v));
 131    v.f32[0] = x;
 132    v.f32[1] = y;
 133    v.f32[2] = z;
 134    v.f32[3] = w;
 135
 136    return nir_build_imm(build, 4, 32, v);
 137 }
 138
 139 static inline nir_ssa_def *
 140 nir_imm_int(nir_builder *build, int x)
 141 {
 142    nir_const_value v;
 143
 144    memset(&v, 0, sizeof(v));
 145    v.i32[0] = x;
 146
 147    return nir_build_imm(build, 1, 32, v);
 148 }
 149
 150 static inline nir_ssa_def *
 151 nir_imm_ivec4(nir_builder *build, int x, int y, int z, int w)
 152 {
 153    nir_const_value v;
 154
 155    memset(&v, 0, sizeof(v));
 156    v.i32[0] = x;
 157    v.i32[1] = y;
 158    v.i32[2] = z;
 159    v.i32[3] = w;
 160
 161    return nir_build_imm(build, 4, 32, v);
 162 }
 163
 164 static inline nir_ssa_def *
 165 nir_build_alu(nir_builder *build, nir_op op, nir_ssa_def *src0,
 166               nir_ssa_def *src1, nir_ssa_def *src2, nir_ssa_def *src3)
 167 {
 168    const nir_op_info *op_info = &nir_op_infos[op];
 169    nir_alu_instr *instr = nir_alu_instr_create(build->shader, op);
 170    if (!instr)
 171       return NULL;
 172
 173    instr->exact = build->exact;
 174
 175    instr->src[0].src = nir_src_for_ssa(src0);
 176    if (src1)
 177       instr->src[1].src = nir_src_for_ssa(src1);
 178    if (src2)
 179       instr->src[2].src = nir_src_for_ssa(src2);
 180    if (src3)
 181       instr->src[3].src = nir_src_for_ssa(src3);
 182
 183    /* Guess the number of components the destination temporary should have
 184     * based on our input sizes, if it's not fixed for the op.
 185     */
 186    unsigned num_components = op_info->output_size;
 187    if (num_components == 0) {
 188       for (unsigned i = 0; i < op_info->num_inputs; i++) {
 189          if (op_info->input_sizes[i] == 0)
 190             num_components = MAX2(num_components,
 191                                   instr->src[i].src.ssa->num_components);
 192       }
 193    }
 194    assert(num_components != 0);
 195
 196    /* Figure out the bitwidth based on the source bitwidth if the instruction
 197     * is variable-width.
 198     */
 199    unsigned bit_size = nir_alu_type_get_type_size(op_info->output_type);
 200    if (bit_size == 0) {
 201       for (unsigned i = 0; i < op_info->num_inputs; i++) {
 202          unsigned src_bit_size = instr->src[i].src.ssa->bit_size;
 203          if (nir_alu_type_get_type_size(op_info->input_types[i]) == 0) {
 204             if (bit_size)
 205                assert(src_bit_size == bit_size);
 206             else
 207                bit_size = src_bit_size;
 208          } else {
 209             assert(src_bit_size ==
 210                nir_alu_type_get_type_size(op_info->input_types[i]));
 211          }
 212       }
 213    }
 214
 215    /* Make sure we don't swizzle from outside of our source vector (like if a
 216     * scalar value was passed into a multiply with a vector).
 217     */
 218    for (unsigned i = 0; i < op_info->num_inputs; i++) {
 219       for (unsigned j = instr->src[i].src.ssa->num_components; j < 4; j++) {
 220          instr->src[i].swizzle[j] = instr->src[i].src.ssa->num_components - 1;
 221       }
 222    }
 223
 224    nir_ssa_dest_init(&instr->instr, &instr->dest.dest, num_components,
 225                      bit_size, NULL);
 226    instr->dest.write_mask = (1 << num_components) - 1;
 227
 228    nir_builder_instr_insert(build, &instr->instr);
 229
 230    return &instr->dest.dest.ssa;
 231 }
 232
 233 #define ALU1(op)                                                          \
 234 static inline nir_ssa_def *                                               \
 235 nir_##op(nir_builder *build, nir_ssa_def *src0)                           \
 236 {                                                                         \
 237    return nir_build_alu(build, nir_op_##op, src0, NULL, NULL, NULL);      \
 238 }
 239
 240 #define ALU2(op)                                                          \
 241 static inline nir_ssa_def *                                               \
 242 nir_##op(nir_builder *build, nir_ssa_def *src0, nir_ssa_def *src1)        \
 243 {                                                                         \
 244    return nir_build_alu(build, nir_op_##op, src0, src1, NULL, NULL);      \
 245 }
 246
 247 #define ALU3(op)                                                          \
 248 static inline nir_ssa_def *                                               \
 249 nir_##op(nir_builder *build, nir_ssa_def *src0,                           \
 250          nir_ssa_def *src1, nir_ssa_def *src2)                            \
 251 {                                                                         \
 252    return nir_build_alu(build, nir_op_##op, src0, src1, src2, NULL);      \
 253 }
 254
 255 #define ALU4(op)                                                          \
 256 static inline nir_ssa_def *                                               \
 257 nir_##op(nir_builder *build, nir_ssa_def *src0,                           \
 258          nir_ssa_def *src1, nir_ssa_def *src2, nir_ssa_def *src3)         \
 259 {                                                                         \
 260    return nir_build_alu(build, nir_op_##op, src0, src1, src2, src3);      \
 261 }
 262
 263 #include "nir_builder_opcodes.h"
 264
 265 static inline nir_ssa_def *
 266 nir_vec(nir_builder *build, nir_ssa_def **comp, unsigned num_components)
 267 {
 268    switch (num_components) {
 269    case 4:
 270       return nir_vec4(build, comp[0], comp[1], comp[2], comp[3]);
 271    case 3:
 272       return nir_vec3(build, comp[0], comp[1], comp[2]);
 273    case 2:
 274       return nir_vec2(build, comp[0], comp[1]);
 275    case 1:
 276       return comp[0];
 277    default:
 278       unreachable("bad component count");
 279       return NULL;
 280    }
 281 }
 282
 283 /**
 284  * Similar to nir_fmov, but takes a nir_alu_src instead of a nir_ssa_def.
 285  */
 286 static inline nir_ssa_def *
 287 nir_fmov_alu(nir_builder *build, nir_alu_src src, unsigned num_components)
 288 {
 289    nir_alu_instr *mov = nir_alu_instr_create(build->shader, nir_op_fmov);
 290    nir_ssa_dest_init(&mov->instr, &mov->dest.dest, num_components,
 291                      nir_src_bit_size(src.src), NULL);
 292    mov->exact = build->exact;
 293    mov->dest.write_mask = (1 << num_components) - 1;
 294    mov->src[0] = src;
 295    nir_builder_instr_insert(build, &mov->instr);
 296
 297    return &mov->dest.dest.ssa;
 298 }
 299
 300 static inline nir_ssa_def *
 301 nir_imov_alu(nir_builder *build, nir_alu_src src, unsigned num_components)
 302 {
 303    nir_alu_instr *mov = nir_alu_instr_create(build->shader, nir_op_imov);
 304    nir_ssa_dest_init(&mov->instr, &mov->dest.dest, num_components,
 305                      nir_src_bit_size(src.src), NULL);
 306    mov->exact = build->exact;
 307    mov->dest.write_mask = (1 << num_components) - 1;
 308    mov->src[0] = src;
 309    nir_builder_instr_insert(build, &mov->instr);
 310
 311    return &mov->dest.dest.ssa;
 312 }
 313
 314 /**
 315  * Construct an fmov or imov that reswizzles the source's components.
 316  */
 317 static inline nir_ssa_def *
 318 nir_swizzle(nir_builder *build, nir_ssa_def *src, const unsigned swiz[4],
 319             unsigned num_components, bool use_fmov)
 320 {
 321    nir_alu_src alu_src = { NIR_SRC_INIT };
 322    alu_src.src = nir_src_for_ssa(src);
 323    for (unsigned i = 0; i < num_components; i++)
 324       alu_src.swizzle[i] = swiz[i];
 325
 326    return use_fmov ? nir_fmov_alu(build, alu_src, num_components) :
 327                      nir_imov_alu(build, alu_src, num_components);
 328 }
 329
 330 /* Selects the right fdot given the number of components in each source. */
 331 static inline nir_ssa_def *
 332 nir_fdot(nir_builder *build, nir_ssa_def *src0, nir_ssa_def *src1)
 333 {
 334    assert(src0->num_components == src1->num_components);
 335    switch (src0->num_components) {
 336    case 1: return nir_fmul(build, src0, src1);
 337    case 2: return nir_fdot2(build, src0, src1);
 338    case 3: return nir_fdot3(build, src0, src1);
 339    case 4: return nir_fdot4(build, src0, src1);
 340    default:
 341       unreachable("bad component size");
 342    }
 343
 344    return NULL;
 345 }
 346
 347 static inline nir_ssa_def *
 348 nir_channel(nir_builder *b, nir_ssa_def *def, unsigned c)
 349 {
 350    unsigned swizzle[4] = {c, c, c, c};
 351    return nir_swizzle(b, def, swizzle, 1, false);
 352 }
 353
 354 /**
 355  * Turns a nir_src into a nir_ssa_def * so it can be passed to
 356  * nir_build_alu()-based builder calls.
 357  *
 358  * See nir_ssa_for_alu_src() for alu instructions.
 359  */
 360 static inline nir_ssa_def *
 361 nir_ssa_for_src(nir_builder *build, nir_src src, int num_components)
 362 {
 363    if (src.is_ssa && src.ssa->num_components == num_components)
 364       return src.ssa;
 365
 366    nir_alu_src alu = { NIR_SRC_INIT };
 367    alu.src = src;
 368    for (int j = 0; j < 4; j++)
 369       alu.swizzle[j] = j;
 370
 371    return nir_imov_alu(build, alu, num_components);
 372 }
 373
 374 /**
 375  * Similar to nir_ssa_for_src(), but for alu src's, respecting the
 376  * nir_alu_src's swizzle.
 377  */
 378 static inline nir_ssa_def *
 379 nir_ssa_for_alu_src(nir_builder *build, nir_alu_instr *instr, unsigned srcn)
 380 {
 381    static uint8_t trivial_swizzle[4] = { 0, 1, 2, 3 };
 382    nir_alu_src *src = &instr->src[srcn];
 383    unsigned num_components = nir_ssa_alu_instr_src_components(instr, srcn);
 384
 385    if (src->src.is_ssa && (src->src.ssa->num_components == num_components) &&
 386        !src->abs && !src->negate &&
 387        (memcmp(src->swizzle, trivial_swizzle, num_components) == 0))
 388       return src->src.ssa;
 389
 390    return nir_imov_alu(build, *src, num_components);
 391 }
 392
 393 static inline nir_ssa_def *
 394 nir_load_var(nir_builder *build, nir_variable *var)
 395 {
 396    const unsigned num_components = glsl_get_vector_elements(var->type);
 397
 398    nir_intrinsic_instr *load =
 399       nir_intrinsic_instr_create(build->shader, nir_intrinsic_load_var);
 400    load->num_components = num_components;
 401    load->variables[0] = nir_deref_var_create(load, var);
 402    nir_ssa_dest_init(&load->instr, &load->dest, num_components,
 403                      glsl_get_bit_size(glsl_get_base_type(var->type)), NULL);
 404    nir_builder_instr_insert(build, &load->instr);
 405    return &load->dest.ssa;
 406 }
 407
 408 static inline void
 409 nir_store_var(nir_builder *build, nir_variable *var, nir_ssa_def *value,
 410               unsigned writemask)
 411 {
 412    const unsigned num_components = glsl_get_vector_elements(var->type);
 413
 414    nir_intrinsic_instr *store =
 415       nir_intrinsic_instr_create(build->shader, nir_intrinsic_store_var);
 416    store->num_components = num_components;
 417    nir_intrinsic_set_write_mask(store, writemask);
 418    store->variables[0] = nir_deref_var_create(store, var);
 419    store->src[0] = nir_src_for_ssa(value);
 420    nir_builder_instr_insert(build, &store->instr);
 421 }
 422
 423 static inline void
 424 nir_store_deref_var(nir_builder *build, nir_deref_var *deref,
 425                     nir_ssa_def *value, unsigned writemask)
 426 {
 427    const unsigned num_components =
 428       glsl_get_vector_elements(nir_deref_tail(&deref->deref)->type);
 429
 430    nir_intrinsic_instr *store =
 431       nir_intrinsic_instr_create(build->shader, nir_intrinsic_store_var);
 432    store->num_components = num_components;
 433    store->const_index[0] = writemask & ((1 << num_components) - 1);
 434    store->variables[0] = nir_deref_as_var(nir_copy_deref(store, &deref->deref));
 435    store->src[0] = nir_src_for_ssa(value);
 436    nir_builder_instr_insert(build, &store->instr);
 437 }
 438
 439 static inline void
 440 nir_copy_deref_var(nir_builder *build, nir_deref_var *dest, nir_deref_var *src)
 441 {
 442    assert(nir_deref_tail(&dest->deref)->type ==
 443           nir_deref_tail(&src->deref)->type);
 444
 445    nir_intrinsic_instr *copy =
 446       nir_intrinsic_instr_create(build->shader, nir_intrinsic_copy_var);
 447    copy->variables[0] = nir_deref_as_var(nir_copy_deref(copy, &dest->deref));
 448    copy->variables[1] = nir_deref_as_var(nir_copy_deref(copy, &src->deref));
 449    nir_builder_instr_insert(build, &copy->instr);
 450 }
 451
 452 static inline void
 453 nir_copy_var(nir_builder *build, nir_variable *dest, nir_variable *src)
 454 {
 455    nir_intrinsic_instr *copy =
 456       nir_intrinsic_instr_create(build->shader, nir_intrinsic_copy_var);
 457    copy->variables[0] = nir_deref_var_create(copy, dest);
 458    copy->variables[1] = nir_deref_var_create(copy, src);
 459    nir_builder_instr_insert(build, &copy->instr);
 460 }
 461
 462 static inline nir_ssa_def *
 463 nir_load_system_value(nir_builder *build, nir_intrinsic_op op, int index)
 464 {
 465    nir_intrinsic_instr *load = nir_intrinsic_instr_create(build->shader, op);
 466    load->num_components = nir_intrinsic_infos[op].dest_components;
 467    load->const_index[0] = index;
 468    nir_ssa_dest_init(&load->instr, &load->dest,
 469                      nir_intrinsic_infos[op].dest_components, 32, NULL);
 470    nir_builder_instr_insert(build, &load->instr);
 471    return &load->dest.ssa;
 472 }
 473
 474 static inline void
 475 nir_jump(nir_builder *build, nir_jump_type jump_type)
 476 {
 477    nir_jump_instr *jump = nir_jump_instr_create(build->shader, jump_type);
 478    nir_builder_instr_insert(build, &jump->instr);
 479 }
 480
 481 #endif /* NIR_BUILDER_H */