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 = { { .f = {x, 0, 0, 0} } };
  80    return nir_build_imm(build, 1, v);
  81 }
  82
  83 static inline nir_ssa_def *
  84 nir_imm_vec4(nir_builder *build, float x, float y, float z, float w)
  85 {
  86    nir_const_value v = { { .f = {x, y, z, w} } };
  87    return nir_build_imm(build, 4, v);
  88 }
  89
  90 static inline nir_ssa_def *
  91 nir_imm_int(nir_builder *build, int x)
  92 {
  93    nir_const_value v = { { .i = {x, 0, 0, 0} } };
  94    return nir_build_imm(build, 1, v);
  95 }
  96
  97 static inline nir_ssa_def *
  98 nir_build_alu(nir_builder *build, nir_op op, nir_ssa_def *src0,
  99               nir_ssa_def *src1, nir_ssa_def *src2, nir_ssa_def *src3)
 100 {
 101    const nir_op_info *op_info = &nir_op_infos[op];
 102    nir_alu_instr *instr = nir_alu_instr_create(build->shader, op);
 103    if (!instr)
 104       return NULL;
 105
 106    instr->src[0].src = nir_src_for_ssa(src0);
 107    if (src1)
 108       instr->src[1].src = nir_src_for_ssa(src1);
 109    if (src2)
 110       instr->src[2].src = nir_src_for_ssa(src2);
 111    if (src3)
 112       instr->src[3].src = nir_src_for_ssa(src3);
 113
 114    /* Guess the number of components the destination temporary should have
 115     * based on our input sizes, if it's not fixed for the op.
 116     */
 117    unsigned num_components = op_info->output_size;
 118    if (num_components == 0) {
 119       for (unsigned i = 0; i < op_info->num_inputs; i++) {
 120          if (op_info->input_sizes[i] == 0)
 121             num_components = MAX2(num_components,
 122                                   instr->src[i].src.ssa->num_components);
 123       }
 124    }
 125    assert(num_components != 0);
 126
 127    /* Make sure we don't swizzle from outside of our source vector (like if a
 128     * scalar value was passed into a multiply with a vector).
 129     */
 130    for (unsigned i = 0; i < op_info->num_inputs; i++) {
 131       for (unsigned j = instr->src[i].src.ssa->num_components; j < 4; j++) {
 132          instr->src[i].swizzle[j] = instr->src[i].src.ssa->num_components - 1;
 133       }
 134    }
 135
 136    nir_ssa_dest_init(&instr->instr, &instr->dest.dest, num_components, NULL);
 137    instr->dest.write_mask = (1 << num_components) - 1;
 138
 139    nir_builder_instr_insert(build, &instr->instr);
 140
 141    return &instr->dest.dest.ssa;
 142 }
 143
 144 #define ALU1(op)                                                          \
 145 static inline nir_ssa_def *                                               \
 146 nir_##op(nir_builder *build, nir_ssa_def *src0)                           \
 147 {                                                                         \
 148    return nir_build_alu(build, nir_op_##op, src0, NULL, NULL, NULL);      \
 149 }
 150
 151 #define ALU2(op)                                                          \
 152 static inline nir_ssa_def *                                               \
 153 nir_##op(nir_builder *build, nir_ssa_def *src0, nir_ssa_def *src1)        \
 154 {                                                                         \
 155    return nir_build_alu(build, nir_op_##op, src0, src1, NULL, NULL);      \
 156 }
 157
 158 #define ALU3(op)                                                          \
 159 static inline nir_ssa_def *                                               \
 160 nir_##op(nir_builder *build, nir_ssa_def *src0,                           \
 161          nir_ssa_def *src1, nir_ssa_def *src2)                            \
 162 {                                                                         \
 163    return nir_build_alu(build, nir_op_##op, src0, src1, src2, NULL);      \
 164 }
 165
 166 #define ALU4(op)                                                          \
 167 static inline nir_ssa_def *                                               \
 168 nir_##op(nir_builder *build, nir_ssa_def *src0,                           \
 169          nir_ssa_def *src1, nir_ssa_def *src2, nir_ssa_def *src3)         \
 170 {                                                                         \
 171    return nir_build_alu(build, nir_op_##op, src0, src1, src2, src3);      \
 172 }
 173
 174 #include "nir_builder_opcodes.h"
 175
 176 /**
 177  * Similar to nir_fmov, but takes a nir_alu_src instead of a nir_ssa_def.
 178  */
 179 static inline nir_ssa_def *
 180 nir_fmov_alu(nir_builder *build, nir_alu_src src, unsigned num_components)
 181 {
 182    nir_alu_instr *mov = nir_alu_instr_create(build->shader, nir_op_fmov);
 183    nir_ssa_dest_init(&mov->instr, &mov->dest.dest, num_components, NULL);
 184    mov->dest.write_mask = (1 << num_components) - 1;
 185    mov->src[0] = src;
 186    nir_builder_instr_insert(build, &mov->instr);
 187
 188    return &mov->dest.dest.ssa;
 189 }
 190
 191 static inline nir_ssa_def *
 192 nir_imov_alu(nir_builder *build, nir_alu_src src, unsigned num_components)
 193 {
 194    nir_alu_instr *mov = nir_alu_instr_create(build->shader, nir_op_imov);
 195    nir_ssa_dest_init(&mov->instr, &mov->dest.dest, num_components, NULL);
 196    mov->dest.write_mask = (1 << num_components) - 1;
 197    mov->src[0] = src;
 198    nir_builder_instr_insert(build, &mov->instr);
 199
 200    return &mov->dest.dest.ssa;
 201 }
 202
 203 /**
 204  * Construct an fmov or imov that reswizzles the source's components.
 205  */
 206 static inline nir_ssa_def *
 207 nir_swizzle(nir_builder *build, nir_ssa_def *src, unsigned swiz[4],
 208             unsigned num_components, bool use_fmov)
 209 {
 210    nir_alu_src alu_src = { NIR_SRC_INIT };
 211    alu_src.src = nir_src_for_ssa(src);
 212    for (int i = 0; i < 4; i++)
 213       alu_src.swizzle[i] = swiz[i];
 214
 215    return use_fmov ? nir_fmov_alu(build, alu_src, num_components) :
 216                      nir_imov_alu(build, alu_src, num_components);
 217 }
 218
 219 /* Selects the right fdot given the number of components in each source. */
 220 static inline nir_ssa_def *
 221 nir_fdot(nir_builder *build, nir_ssa_def *src0, nir_ssa_def *src1)
 222 {
 223    assert(src0->num_components == src1->num_components);
 224    switch (src0->num_components) {
 225    case 1: return nir_fmul(build, src0, src1);
 226    case 2: return nir_fdot2(build, src0, src1);
 227    case 3: return nir_fdot3(build, src0, src1);
 228    case 4: return nir_fdot4(build, src0, src1);
 229    default:
 230       unreachable("bad component size");
 231    }
 232
 233    return NULL;
 234 }
 235
 236 /**
 237  * Turns a nir_src into a nir_ssa_def * so it can be passed to
 238  * nir_build_alu()-based builder calls.
 239  */
 240 static inline nir_ssa_def *
 241 nir_ssa_for_src(nir_builder *build, nir_src src, int num_components)
 242 {
 243    if (src.is_ssa && src.ssa->num_components == num_components)
 244       return src.ssa;
 245
 246    nir_alu_src alu = { NIR_SRC_INIT };
 247    alu.src = src;
 248    for (int j = 0; j < 4; j++)
 249       alu.swizzle[j] = j;
 250
 251    return nir_imov_alu(build, alu, num_components);
 252 }
 253
 254 #endif /* NIR_BUILDER_H */