src/glsl/nir/nir_opcodes.h

   1 /*
   2  * Copyright © 2014 Intel Corporation
   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  * Authors:
  24  *    Connor Abbott (cwabbott0@gmail.com)
  25  *
  26  */
  27
  28 /**
  29  * This header file defines all the available opcodes in one place. It expands
  30  * to a list of macros of the form:
  31  *
  32  * OPCODE(name, num_inputs, per_component, output_size, output_type,
  33  *        input_sizes, input_types)
  34  *
  35  * Which should correspond one-to-one with the nir_op_info structure. It is
  36  * included in both ir.h to create the nir_op enum (with members of the form
  37  * nir_op_(name)) and and in opcodes.c to create nir_op_infos, which is a
  38  * const array of nir_op_info structures for each opcode.
  39  */
  40
  41 #define ARR(...) { __VA_ARGS__ }
  42
  43 #define UNOP(name, type) OPCODE(name, 1, false, 0, type, ARR(0), ARR(type))
  44 #define UNOP_CONVERT(name, in_type, out_type) \
  45    OPCODE(name, 1, false, 0, out_type, ARR(0), ARR(in_type))
  46 #define UNOP_HORIZ(name, output_size, output_type, input_size, input_type) \
  47    OPCODE(name, 1, true, output_size, output_type, ARR(input_size), \
  48           ARR(input_type))
  49
  50 #define UNOP_REDUCE(name, output_size, output_type, input_type) \
  51    UNOP_HORIZ(name##2, output_size, output_type, 2, input_type) \
  52    UNOP_HORIZ(name##3, output_size, output_type, 3, input_type) \
  53    UNOP_HORIZ(name##4, output_size, output_type, 4, input_type)
  54
  55 /**
  56  * These two move instructions differ in what modifiers they support and what
  57  * the negate modifier means. Otherwise, they are identical.
  58  */
  59 UNOP(fmov, nir_type_float)
  60 UNOP(imov, nir_type_int)
  61
  62 UNOP(inot, nir_type_int) /* invert every bit of the integer */
  63 UNOP(fnot, nir_type_float) /* (src == 0.0) ? 1.0 : 0.0 */
  64 UNOP(fsign, nir_type_float)
  65 UNOP(isign, nir_type_int)
  66 UNOP(frcp, nir_type_float)
  67 UNOP(frsq, nir_type_float)
  68 UNOP(fsqrt, nir_type_float)
  69 UNOP(fexp, nir_type_float) /* < e^x */
  70 UNOP(flog, nir_type_float) /* log base e */
  71 UNOP(fexp2, nir_type_float)
  72 UNOP(flog2, nir_type_float)
  73 UNOP_CONVERT(f2i, nir_type_float, nir_type_int)       /**< Float-to-integer conversion. */
  74 UNOP_CONVERT(f2u, nir_type_float, nir_type_unsigned)  /**< Float-to-unsigned conversion. */
  75 UNOP_CONVERT(i2f, nir_type_int, nir_type_float)       /**< Integer-to-float conversion. */
  76 UNOP_CONVERT(f2b, nir_type_float, nir_type_bool)      /**< Float-to-boolean conversion */
  77 UNOP_CONVERT(b2f, nir_type_bool, nir_type_float)      /**< Boolean-to-float conversion */
  78 UNOP_CONVERT(i2b, nir_type_int, nir_type_bool)        /**< int-to-boolean conversion */
  79 UNOP_CONVERT(b2i, nir_type_bool, nir_type_int)        /**< Boolean-to-int conversion */
  80 UNOP_CONVERT(u2f, nir_type_unsigned, nir_type_float)  /**< Unsigned-to-float conversion. */
  81
  82 UNOP_REDUCE(bany, 1, nir_type_bool, nir_type_bool) /* returns ~0 if any component of src[0] != 0 */
  83 UNOP_REDUCE(ball, 1, nir_type_bool, nir_type_bool) /* returns ~0 if all components of src[0] != 0 */
  84 UNOP_REDUCE(fany, 1, nir_type_float, nir_type_float) /* returns 1.0 if any component of src[0] != 0 */
  85 UNOP_REDUCE(fall, 1, nir_type_float, nir_type_float) /* returns 1.0 if all components of src[0] != 0 */
  86
  87 /**
  88  * \name Unary floating-point rounding operations.
  89  */
  90 /*@{*/
  91 UNOP(ftrunc, nir_type_float)
  92 UNOP(fceil, nir_type_float)
  93 UNOP(ffloor, nir_type_float)
  94 UNOP(ffract, nir_type_float)
  95 UNOP(fround_even, nir_type_float)
  96 /*@}*/
  97
  98 /**
  99  * \name Trigonometric operations.
 100  */
 101 /*@{*/
 102 UNOP(fsin, nir_type_float)
 103 UNOP(fcos, nir_type_float)
 104 UNOP(fsin_reduced, nir_type_float)
 105 UNOP(fcos_reduced, nir_type_float)
 106 /*@}*/
 107
 108 /**
 109  * \name Partial derivatives.
 110  */
 111 /*@{*/
 112 UNOP(fddx, nir_type_float)
 113 UNOP(fddy, nir_type_float)
 114 /*@}*/
 115
 116 /**
 117  * \name Floating point pack and unpack operations.
 118  */
 119 /*@{*/
 120 UNOP_HORIZ(pack_snorm_2x16, 1, nir_type_unsigned, 2, nir_type_float)
 121 UNOP_HORIZ(pack_snorm_4x8, 1, nir_type_unsigned, 4, nir_type_float)
 122 UNOP_HORIZ(pack_unorm_2x16, 1, nir_type_unsigned, 2, nir_type_float)
 123 UNOP_HORIZ(pack_unorm_4x8, 1, nir_type_unsigned, 4, nir_type_float)
 124 UNOP_HORIZ(pack_half_2x16, 1, nir_type_unsigned, 2, nir_type_float)
 125 UNOP_HORIZ(unpack_snorm_2x16, 2, nir_type_float, 1, nir_type_unsigned)
 126 UNOP_HORIZ(unpack_snorm_4x8, 4, nir_type_float, 1, nir_type_unsigned)
 127 UNOP_HORIZ(unpack_unorm_2x16, 2, nir_type_float, 1, nir_type_unsigned)
 128 UNOP_HORIZ(unpack_unorm_4x8, 4, nir_type_float, 1, nir_type_unsigned)
 129 UNOP_HORIZ(unpack_half_2x16, 2, nir_type_float, 1, nir_type_unsigned)
 130 /*@}*/
 131
 132 /**
 133  * \name Lowered floating point unpacking operations.
 134  */
 135 /*@{*/
 136 UNOP_HORIZ(unpack_half_2x16_split_x, 1, nir_type_float, 1, nir_type_unsigned)
 137 UNOP_HORIZ(unpack_half_2x16_split_y, 1, nir_type_float, 1, nir_type_unsigned)
 138 /*@}*/
 139
 140 /**
 141  * \name Bit operations, part of ARB_gpu_shader5.
 142  */
 143 /*@{*/
 144 UNOP(bitfield_reverse, nir_type_unsigned)
 145 UNOP(bit_count, nir_type_unsigned)
 146 UNOP(find_msb, nir_type_unsigned)
 147 UNOP(find_lsb, nir_type_unsigned)
 148 /*@}*/
 149
 150 UNOP_HORIZ(fnoise1_1, 1, nir_type_float, 1, nir_type_float)
 151 UNOP_HORIZ(fnoise1_2, 1, nir_type_float, 2, nir_type_float)
 152 UNOP_HORIZ(fnoise1_3, 1, nir_type_float, 3, nir_type_float)
 153 UNOP_HORIZ(fnoise1_4, 1, nir_type_float, 4, nir_type_float)
 154 UNOP_HORIZ(fnoise2_1, 2, nir_type_float, 1, nir_type_float)
 155 UNOP_HORIZ(fnoise2_2, 2, nir_type_float, 2, nir_type_float)
 156 UNOP_HORIZ(fnoise2_3, 2, nir_type_float, 3, nir_type_float)
 157 UNOP_HORIZ(fnoise2_4, 2, nir_type_float, 4, nir_type_float)
 158 UNOP_HORIZ(fnoise3_1, 3, nir_type_float, 1, nir_type_float)
 159 UNOP_HORIZ(fnoise3_2, 3, nir_type_float, 2, nir_type_float)
 160 UNOP_HORIZ(fnoise3_3, 3, nir_type_float, 3, nir_type_float)
 161 UNOP_HORIZ(fnoise3_4, 3, nir_type_float, 4, nir_type_float)
 162 UNOP_HORIZ(fnoise4_1, 4, nir_type_float, 1, nir_type_float)
 163 UNOP_HORIZ(fnoise4_2, 4, nir_type_float, 2, nir_type_float)
 164 UNOP_HORIZ(fnoise4_3, 4, nir_type_float, 3, nir_type_float)
 165 UNOP_HORIZ(fnoise4_4, 4, nir_type_float, 4, nir_type_float)
 166
 167 #define BINOP(name, type) \
 168    OPCODE(name, 2, true, 0, type, ARR(0, 0), ARR(type, type))
 169 #define BINOP_CONVERT(name, out_type, in_type) \
 170    OPCODE(name, 2, true, 0, out_type, ARR(0, 0), ARR(in_type, in_type))
 171 #define BINOP_COMPARE(name, type) BINOP_CONVERT(name, nir_type_bool, type)
 172 #define BINOP_HORIZ(name, output_size, output_type, src1_size, src1_type, \
 173                     src2_size, src2_type) \
 174    OPCODE(name, 2, true, output_size, output_type, ARR(src1_size, src2_size), \
 175           ARR(src1_type, src2_type))
 176 #define BINOP_REDUCE(name, output_size, output_type, src_type) \
 177    BINOP_HORIZ(name##2, output_size, output_type, 2, src_type, 2, src_type) \
 178    BINOP_HORIZ(name##3, output_size, output_type, 3, src_type, 3, src_type) \
 179    BINOP_HORIZ(name##4, output_size, output_type, 4, src_type, 4, src_type) \
 180
 181 BINOP(fadd, nir_type_float)
 182 BINOP(iadd, nir_type_int)
 183 BINOP(fsub, nir_type_float)
 184 BINOP(isub, nir_type_int)
 185
 186 BINOP(fmul, nir_type_float)
 187 BINOP(imul, nir_type_int) /* low 32-bits of signed/unsigned integer multiply */
 188 BINOP(imul_high, nir_type_int) /* high 32-bits of signed integer multiply */
 189 BINOP(umul_high, nir_type_unsigned) /* high 32-bits of unsigned integer multiply */
 190
 191 BINOP(fdiv, nir_type_float)
 192 BINOP(idiv, nir_type_int)
 193 BINOP(udiv, nir_type_unsigned)
 194
 195 /**
 196  * returns a boolean representing the carry resulting from the addition of
 197  * the two unsigned arguments.
 198  */
 199 BINOP_CONVERT(uadd_carry, nir_type_bool, nir_type_unsigned)
 200
 201 /**
 202  * returns a boolean representing the borrow resulting from the subtraction
 203  * of the two unsigned arguments.
 204  */
 205 BINOP_CONVERT(usub_borrow, nir_type_bool, nir_type_unsigned)
 206
 207 BINOP(fmod, nir_type_float)
 208 BINOP(umod, nir_type_unsigned)
 209
 210 /**
 211  * \name comparisons
 212  */
 213 /*@{*/
 214
 215 /**
 216  * these integer-aware comparisons return a boolean (0 or ~0)
 217  */
 218 BINOP_COMPARE(flt, nir_type_float)
 219 BINOP_COMPARE(fge, nir_type_float)
 220 BINOP_COMPARE(feq, nir_type_float)
 221 BINOP_COMPARE(fne, nir_type_float)
 222 BINOP_COMPARE(ilt, nir_type_int)
 223 BINOP_COMPARE(ige, nir_type_int)
 224 BINOP_COMPARE(ieq, nir_type_int)
 225 BINOP_COMPARE(ine, nir_type_int)
 226 BINOP_COMPARE(ult, nir_type_unsigned)
 227 BINOP_COMPARE(uge, nir_type_unsigned)
 228
 229 /** integer-aware GLSL-style comparisons that compare floats and ints */
 230 BINOP_REDUCE(ball_fequal,  1, nir_type_bool, nir_type_float)
 231 BINOP_REDUCE(bany_fnequal, 1, nir_type_bool, nir_type_float)
 232 BINOP_REDUCE(ball_iequal,  1, nir_type_bool, nir_type_int)
 233 BINOP_REDUCE(bany_inequal, 1, nir_type_bool, nir_type_int)
 234
 235 /** non-integer-aware GLSL-style comparisons that return 0.0 or 1.0 */
 236 BINOP_REDUCE(fall_equal,  1, nir_type_float, nir_type_float)
 237 BINOP_REDUCE(fany_nequal, 1, nir_type_float, nir_type_float)
 238
 239 /**
 240  * These comparisons for integer-less hardware return 1.0 and 0.0 for true
 241  * and false respectively
 242  */
 243 BINOP(slt, nir_type_float) /* Set on Less Than */
 244 BINOP(sge, nir_type_float) /* Set on Greater Than or Equal */
 245 BINOP(seq, nir_type_float) /* Set on Equal */
 246 BINOP(sne, nir_type_float) /* Set on Not Equal */
 247
 248 /*@}*/
 249
 250 BINOP(ishl, nir_type_int)
 251 BINOP(ishr, nir_type_int)
 252 BINOP(ushr, nir_type_unsigned)
 253
 254 /**
 255  * \name bitwise logic operators
 256  *
 257  * These are also used as boolean and, or, xor for hardware supporting
 258  * integers.
 259  */
 260 /*@{*/
 261 BINOP(iand, nir_type_unsigned)
 262 BINOP(ior, nir_type_unsigned)
 263 BINOP(ixor, nir_type_unsigned)
 264 /*@{*/
 265
 266 /**
 267  * \name floating point logic operators
 268  *
 269  * These use (src != 0.0) for testing the truth of the input, and output 1.0
 270  * for true and 0.0 for false
 271  */
 272 BINOP(fand, nir_type_float)
 273 BINOP(for, nir_type_float)
 274 BINOP(fxor, nir_type_float)
 275
 276 BINOP_REDUCE(fdot, 1, nir_type_float, nir_type_float)
 277
 278 BINOP(fmin, nir_type_float)
 279 BINOP(imin, nir_type_int)
 280 BINOP(umin, nir_type_unsigned)
 281 BINOP(fmax, nir_type_float)
 282 BINOP(imax, nir_type_int)
 283 BINOP(umax, nir_type_unsigned)
 284
 285 BINOP(fpow, nir_type_float)
 286
 287 BINOP_HORIZ(pack_half_2x16_split, 1, nir_type_unsigned, 1, nir_type_float, 1, nir_type_float)
 288
 289 BINOP(bfm, nir_type_unsigned)
 290
 291 BINOP(ldexp, nir_type_unsigned)
 292
 293 /**
 294  * Combines the first component of each input to make a 2-component vector.
 295  */
 296 BINOP_HORIZ(vec2, 2, nir_type_unsigned, 1, nir_type_unsigned, 1, nir_type_unsigned)
 297
 298 #define TRIOP(name, type) \
 299    OPCODE(name, 3, true, 0, type, ARR(0, 0, 0), ARR(type, type, type))
 300 #define TRIOP_HORIZ(name, output_size, src1_size, src2_size, src3_size) \
 301    OPCODE(name, 3, false, output_size, nir_type_unsigned, \
 302    ARR(src1_size, src2_size, src3_size), \
 303    ARR(nir_type_unsigned, nir_type_unsigned, nir_type_unsigned))
 304
 305 TRIOP(ffma, nir_type_float)
 306
 307 TRIOP(flrp, nir_type_float)
 308
 309 /**
 310  * \name Conditional Select
 311  *
 312  * A vector conditional select instruction (like ?:, but operating per-
 313  * component on vectors). There are two versions, one for floating point
 314  * bools (0.0 vs 1.0) and one for integer bools (0 vs ~0).
 315  */
 316
 317 OPCODE(fcsel, 3, true, 0, nir_type_float, ARR(1, 0, 0),
 318        ARR(nir_type_float, nir_type_float, nir_type_float))
 319 OPCODE(bcsel, 3, true, 0, nir_type_unsigned, ARR(1, 0, 0),
 320        ARR(nir_type_bool, nir_type_unsigned, nir_type_unsigned))
 321
 322 TRIOP(bfi, nir_type_unsigned)
 323
 324 TRIOP(ubitfield_extract, nir_type_unsigned)
 325 OPCODE(ibitfield_extract, 3, true, 0, nir_type_int, ARR(0, 0, 0),
 326        ARR(nir_type_int, nir_type_unsigned, nir_type_unsigned))
 327
 328 /**
 329  * Combines the first component of each input to make a 3-component vector.
 330  */
 331 TRIOP_HORIZ(vec3, 3, 1, 1, 1)
 332
 333 #define QUADOP(name) \
 334    OPCODE(name, 4, true, 0, nir_type_unsigned, ARR(0, 0, 0, 0), \
 335    ARR(nir_type_unsigned, nir_type_unsigned, nir_type_unsigned, nir_type_unsigned))
 336 #define QUADOP_HORIZ(name, output_size, src1_size, src2_size, src3_size, \
 337                      src4_size) \
 338    OPCODE(name, 4, false, output_size, nir_type_unsigned, \
 339           ARR(src1_size, src2_size, src3_size, src4_size), \
 340           ARR(nir_type_unsigned, nir_type_unsigned, nir_type_unsigned, nir_type_unsigned))
 341
 342 QUADOP(bitfield_insert)
 343
 344 QUADOP_HORIZ(vec4, 4, 1, 1, 1, 1)
 345
 346 LAST_OPCODE(vec4)