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