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 #include "nir_control_flow.h"
  28
  29 struct exec_list;
  30
  31 typedef struct nir_builder {
  32    nir_cursor cursor;
  33
  34    nir_shader *shader;
  35    nir_function_impl *impl;
  36 } nir_builder;
  37
  38 static inline void
  39 nir_builder_init(nir_builder *build, nir_function_impl *impl)
  40 {
  41    memset(build, 0, sizeof(*build));
  42    build->impl = impl;
  43    build->shader = impl->overload->function->shader;
  44 }
  45
  46 static inline void
  47 nir_builder_instr_insert(nir_builder *build, nir_instr *instr)
  48 {
  49    nir_instr_insert(build->cursor, instr);
  50
  51    /* Move the cursor forward. */
  52    build->cursor = nir_after_instr(instr);
  53 }
  54
  55 static inline void
  56 nir_builder_cf_insert(nir_builder *build, nir_cf_node *cf)
  57 {
  58    nir_cf_node_insert(build->cursor, cf);
  59 }
  60
  61 static inline nir_ssa_def *
  62 nir_ssa_undef(nir_builder *build, unsigned num_components)
  63 {
  64    nir_ssa_undef_instr *undef =
  65       nir_ssa_undef_instr_create(build->shader, num_components);
  66    if (!undef)
  67       return NULL;
  68
  69    nir_instr_insert(nir_before_block(nir_start_block(build->impl)),
  70                     &undef->instr);
  71
  72    return &undef->def;
  73 }
  74
  75 static inline nir_ssa_def *
  76 nir_build_imm(nir_builder *build, unsigned num_components, nir_const_value value)
  77 {
  78    nir_load_const_instr *load_const =
  79       nir_load_const_instr_create(build->shader, num_components);
  80    if (!load_const)
  81       return NULL;
  82
  83    load_const->value = value;
  84
  85    nir_builder_instr_insert(build, &load_const->instr);
  86
  87    return &load_const->def;
  88 }
  89
  90 static inline nir_ssa_def *
  91 nir_imm_float(nir_builder *build, float x)
  92 {
  93    nir_const_value v;
  94
  95    memset(&v, 0, sizeof(v));
  96    v.f[0] = x;
  97
  98    return nir_build_imm(build, 1, v);
  99 }
 100
 101 static inline nir_ssa_def *
 102 nir_imm_vec4(nir_builder *build, float x, float y, float z, float w)
 103 {
 104    nir_const_value v;
 105
 106    memset(&v, 0, sizeof(v));
 107    v.f[0] = x;
 108    v.f[1] = y;
 109    v.f[2] = z;
 110    v.f[3] = w;
 111
 112    return nir_build_imm(build, 4, v);
 113 }
 114
 115 static inline nir_ssa_def *
 116 nir_imm_int(nir_builder *build, int x)
 117 {
 118    nir_const_value v;
 119
 120    memset(&v, 0, sizeof(v));
 121    v.i[0] = x;
 122
 123    return nir_build_imm(build, 1, v);
 124 }
 125
 126 static inline nir_ssa_def *
 127 nir_build_alu(nir_builder *build, nir_op op, nir_ssa_def *src0,
 128               nir_ssa_def *src1, nir_ssa_def *src2, nir_ssa_def *src3)
 129 {
 130    const nir_op_info *op_info = &nir_op_infos[op];
 131    nir_alu_instr *instr = nir_alu_instr_create(build->shader, op);
 132    if (!instr)
 133       return NULL;
 134
 135    instr->src[0].src = nir_src_for_ssa(src0);
 136    if (src1)
 137       instr->src[1].src = nir_src_for_ssa(src1);
 138    if (src2)
 139       instr->src[2].src = nir_src_for_ssa(src2);
 140    if (src3)
 141       instr->src[3].src = nir_src_for_ssa(src3);
 142
 143    /* Guess the number of components the destination temporary should have
 144     * based on our input sizes, if it's not fixed for the op.
 145     */
 146    unsigned num_components = op_info->output_size;
 147    if (num_components == 0) {
 148       for (unsigned i = 0; i < op_info->num_inputs; i++) {
 149          if (op_info->input_sizes[i] == 0)
 150             num_components = MAX2(num_components,
 151                                   instr->src[i].src.ssa->num_components);
 152       }
 153    }
 154    assert(num_components != 0);
 155
 156    /* Make sure we don't swizzle from outside of our source vector (like if a
 157     * scalar value was passed into a multiply with a vector).
 158     */
 159    for (unsigned i = 0; i < op_info->num_inputs; i++) {
 160       for (unsigned j = instr->src[i].src.ssa->num_components; j < 4; j++) {
 161          instr->src[i].swizzle[j] = instr->src[i].src.ssa->num_components - 1;
 162       }
 163    }
 164
 165    nir_ssa_dest_init(&instr->instr, &instr->dest.dest, num_components, NULL);
 166    instr->dest.write_mask = (1 << num_components) - 1;
 167
 168    nir_builder_instr_insert(build, &instr->instr);
 169
 170    return &instr->dest.dest.ssa;
 171 }
 172
 173 #define ALU1(op)                                                          \
 174 static inline nir_ssa_def *                                               \
 175 nir_##op(nir_builder *build, nir_ssa_def *src0)                           \
 176 {                                                                         \
 177    return nir_build_alu(build, nir_op_##op, src0, NULL, NULL, NULL);      \
 178 }
 179
 180 #define ALU2(op)                                                          \
 181 static inline nir_ssa_def *                                               \
 182 nir_##op(nir_builder *build, nir_ssa_def *src0, nir_ssa_def *src1)        \
 183 {                                                                         \
 184    return nir_build_alu(build, nir_op_##op, src0, src1, NULL, NULL);      \
 185 }
 186
 187 #define ALU3(op)                                                          \
 188 static inline nir_ssa_def *                                               \
 189 nir_##op(nir_builder *build, nir_ssa_def *src0,                           \
 190          nir_ssa_def *src1, nir_ssa_def *src2)                            \
 191 {                                                                         \
 192    return nir_build_alu(build, nir_op_##op, src0, src1, src2, NULL);      \
 193 }
 194
 195 #define ALU4(op)                                                          \
 196 static inline nir_ssa_def *                                               \
 197 nir_##op(nir_builder *build, nir_ssa_def *src0,                           \
 198          nir_ssa_def *src1, nir_ssa_def *src2, nir_ssa_def *src3)         \
 199 {                                                                         \
 200    return nir_build_alu(build, nir_op_##op, src0, src1, src2, src3);      \
 201 }
 202
 203 #include "nir_builder_opcodes.h"
 204
 205 static inline nir_ssa_def *
 206 nir_vec(nir_builder *build, nir_ssa_def **comp, unsigned num_components)
 207 {
 208    switch (num_components) {
 209    case 4:
 210       return nir_vec4(build, comp[0], comp[1], comp[2], comp[3]);
 211    case 3:
 212       return nir_vec3(build, comp[0], comp[1], comp[2]);
 213    case 2:
 214       return nir_vec2(build, comp[0], comp[1]);
 215    case 1:
 216       return comp[0];
 217    default:
 218       unreachable("bad component count");
 219       return NULL;
 220    }
 221 }
 222
 223 /**
 224  * Similar to nir_fmov, but takes a nir_alu_src instead of a nir_ssa_def.
 225  */
 226 static inline nir_ssa_def *
 227 nir_fmov_alu(nir_builder *build, nir_alu_src src, unsigned num_components)
 228 {
 229    nir_alu_instr *mov = nir_alu_instr_create(build->shader, nir_op_fmov);
 230    nir_ssa_dest_init(&mov->instr, &mov->dest.dest, num_components, NULL);
 231    mov->dest.write_mask = (1 << num_components) - 1;
 232    mov->src[0] = src;
 233    nir_builder_instr_insert(build, &mov->instr);
 234
 235    return &mov->dest.dest.ssa;
 236 }
 237
 238 static inline nir_ssa_def *
 239 nir_imov_alu(nir_builder *build, nir_alu_src src, unsigned num_components)
 240 {
 241    nir_alu_instr *mov = nir_alu_instr_create(build->shader, nir_op_imov);
 242    nir_ssa_dest_init(&mov->instr, &mov->dest.dest, num_components, NULL);
 243    mov->dest.write_mask = (1 << num_components) - 1;
 244    mov->src[0] = src;
 245    nir_builder_instr_insert(build, &mov->instr);
 246
 247    return &mov->dest.dest.ssa;
 248 }
 249
 250 /**
 251  * Construct an fmov or imov that reswizzles the source's components.
 252  */
 253 static inline nir_ssa_def *
 254 nir_swizzle(nir_builder *build, nir_ssa_def *src, unsigned swiz[4],
 255             unsigned num_components, bool use_fmov)
 256 {
 257    nir_alu_src alu_src = { NIR_SRC_INIT };
 258    alu_src.src = nir_src_for_ssa(src);
 259    for (int i = 0; i < 4; i++)
 260       alu_src.swizzle[i] = swiz[i];
 261
 262    return use_fmov ? nir_fmov_alu(build, alu_src, num_components) :
 263                      nir_imov_alu(build, alu_src, num_components);
 264 }
 265
 266 /* Selects the right fdot given the number of components in each source. */
 267 static inline nir_ssa_def *
 268 nir_fdot(nir_builder *build, nir_ssa_def *src0, nir_ssa_def *src1)
 269 {
 270    assert(src0->num_components == src1->num_components);
 271    switch (src0->num_components) {
 272    case 1: return nir_fmul(build, src0, src1);
 273    case 2: return nir_fdot2(build, src0, src1);
 274    case 3: return nir_fdot3(build, src0, src1);
 275    case 4: return nir_fdot4(build, src0, src1);
 276    default:
 277       unreachable("bad component size");
 278    }
 279
 280    return NULL;
 281 }
 282
 283 static inline nir_ssa_def *
 284 nir_channel(nir_builder *b, nir_ssa_def *def, unsigned c)
 285 {
 286    unsigned swizzle[4] = {c, c, c, c};
 287    return nir_swizzle(b, def, swizzle, 1, false);
 288 }
 289
 290 /**
 291  * Turns a nir_src into a nir_ssa_def * so it can be passed to
 292  * nir_build_alu()-based builder calls.
 293  */
 294 static inline nir_ssa_def *
 295 nir_ssa_for_src(nir_builder *build, nir_src src, int num_components)
 296 {
 297    if (src.is_ssa && src.ssa->num_components == num_components)
 298       return src.ssa;
 299
 300    nir_alu_src alu = { NIR_SRC_INIT };
 301    alu.src = src;
 302    for (int j = 0; j < 4; j++)
 303       alu.swizzle[j] = j;
 304
 305    return nir_imov_alu(build, alu, num_components);
 306 }
 307
 308 static inline nir_ssa_def *
 309 nir_load_var(nir_builder *build, nir_variable *var)
 310 {
 311    const unsigned num_components = glsl_get_vector_elements(var->type);
 312
 313    nir_intrinsic_instr *load =
 314       nir_intrinsic_instr_create(build->shader, nir_intrinsic_load_var);
 315    load->num_components = num_components;
 316    load->variables[0] = nir_deref_var_create(load, var);
 317    nir_ssa_dest_init(&load->instr, &load->dest, num_components, NULL);
 318    nir_builder_instr_insert(build, &load->instr);
 319    return &load->dest.ssa;
 320 }
 321
 322 static inline void
 323 nir_store_var(nir_builder *build, nir_variable *var, nir_ssa_def *value)
 324 {
 325    const unsigned num_components = glsl_get_vector_elements(var->type);
 326
 327    nir_intrinsic_instr *store =
 328       nir_intrinsic_instr_create(build->shader, nir_intrinsic_store_var);
 329    store->num_components = num_components;
 330    store->variables[0] = nir_deref_var_create(store, var);
 331    store->src[0] = nir_src_for_ssa(value);
 332    nir_builder_instr_insert(build, &store->instr);
 333 }
 334
 335 #endif /* NIR_BUILDER_H */