Damien Lespiau | 4ca1a04 | 2013-01-19 00:30:18 +0000 | [diff] [blame] | 1 | /* |
| 2 | Copyright (C) Intel Corp. 2006. All Rights Reserved. |
| 3 | Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to |
| 4 | develop this 3D driver. |
| 5 | |
| 6 | Permission is hereby granted, free of charge, to any person obtaining |
| 7 | a copy of this software and associated documentation files (the |
| 8 | "Software"), to deal in the Software without restriction, including |
| 9 | without limitation the rights to use, copy, modify, merge, publish, |
| 10 | distribute, sublicense, and/or sell copies of the Software, and to |
| 11 | permit persons to whom the Software is furnished to do so, subject to |
| 12 | the following conditions: |
| 13 | |
| 14 | The above copyright notice and this permission notice (including the |
| 15 | next paragraph) shall be included in all copies or substantial |
| 16 | portions of the Software. |
| 17 | |
| 18 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, |
| 19 | EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF |
| 20 | MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. |
| 21 | IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE |
| 22 | LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION |
| 23 | OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION |
| 24 | WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. |
| 25 | |
| 26 | **********************************************************************/ |
| 27 | /* |
| 28 | * Authors: |
| 29 | * Keith Whitwell <keith@tungstengraphics.com> |
| 30 | */ |
| 31 | |
| 32 | /** @file brw_reg.h |
| 33 | * |
| 34 | * This file defines struct brw_reg, which is our representation for EU |
| 35 | * registers. They're not a hardware specific format, just an abstraction |
| 36 | * that intends to capture the full flexibility of the hardware registers. |
| 37 | * |
| 38 | * The brw_eu_emit.c layer's brw_set_dest/brw_set_src[01] functions encode |
| 39 | * the abstract brw_reg type into the actual hardware instruction encoding. |
| 40 | */ |
| 41 | |
| 42 | #ifndef BRW_REG_H |
| 43 | #define BRW_REG_H |
| 44 | |
| 45 | #include <stdbool.h> |
| 46 | #include <assert.h> |
| 47 | #include "brw_defines.h" |
| 48 | |
| 49 | #ifdef __cplusplus |
| 50 | extern "C" { |
| 51 | #endif |
| 52 | |
| 53 | /** Number of general purpose registers (VS, WM, etc) */ |
| 54 | #define BRW_MAX_GRF 128 |
| 55 | |
| 56 | /** |
| 57 | * First GRF used for the MRF hack. |
| 58 | * |
| 59 | * On gen7, MRFs are no longer used, and contiguous GRFs are used instead. We |
| 60 | * haven't converted our compiler to be aware of this, so it asks for MRFs and |
| 61 | * brw_eu_emit.c quietly converts them to be accesses of the top GRFs. The |
| 62 | * register allocators have to be careful of this to avoid corrupting the "MRF"s |
| 63 | * with actual GRF allocations. |
| 64 | */ |
| 65 | #define GEN7_MRF_HACK_START 112 |
| 66 | |
| 67 | /** Number of message register file registers */ |
| 68 | #define BRW_MAX_MRF 16 |
| 69 | |
| 70 | #define BRW_SWIZZLE4(a,b,c,d) (((a)<<0) | ((b)<<2) | ((c)<<4) | ((d)<<6)) |
| 71 | #define BRW_GET_SWZ(swz, idx) (((swz) >> ((idx)*2)) & 0x3) |
| 72 | |
| 73 | #define BRW_SWIZZLE_NOOP BRW_SWIZZLE4(0,1,2,3) |
| 74 | #define BRW_SWIZZLE_XYZW BRW_SWIZZLE4(0,1,2,3) |
| 75 | #define BRW_SWIZZLE_XXXX BRW_SWIZZLE4(0,0,0,0) |
| 76 | #define BRW_SWIZZLE_YYYY BRW_SWIZZLE4(1,1,1,1) |
| 77 | #define BRW_SWIZZLE_ZZZZ BRW_SWIZZLE4(2,2,2,2) |
| 78 | #define BRW_SWIZZLE_WWWW BRW_SWIZZLE4(3,3,3,3) |
| 79 | #define BRW_SWIZZLE_XYXY BRW_SWIZZLE4(0,1,0,1) |
| 80 | |
| 81 | static inline bool |
| 82 | brw_is_single_value_swizzle(int swiz) |
| 83 | { |
| 84 | return (swiz == BRW_SWIZZLE_XXXX || |
| 85 | swiz == BRW_SWIZZLE_YYYY || |
| 86 | swiz == BRW_SWIZZLE_ZZZZ || |
| 87 | swiz == BRW_SWIZZLE_WWWW); |
| 88 | } |
| 89 | |
| 90 | #define BRW_WRITEMASK_X 0x1 |
| 91 | #define BRW_WRITEMASK_Y 0x2 |
| 92 | #define BRW_WRITEMASK_Z 0x4 |
| 93 | #define BRW_WRITEMASK_W 0x8 |
| 94 | |
| 95 | #define BRW_WRITEMASK_XY (BRW_WRITEMASK_X | BRW_WRITEMASK_Y) |
| 96 | #define BRW_WRITEMASK_XZ (BRW_WRITEMASK_X | BRW_WRITEMASK_Z) |
| 97 | #define BRW_WRITEMASK_XW (BRW_WRITEMASK_X | BRW_WRITEMASK_W) |
| 98 | #define BRW_WRITEMASK_YW (BRW_WRITEMASK_Y | BRW_WRITEMASK_W) |
| 99 | #define BRW_WRITEMASK_ZW (BRW_WRITEMASK_Z | BRW_WRITEMASK_W) |
| 100 | #define BRW_WRITEMASK_XYZ (BRW_WRITEMASK_X | BRW_WRITEMASK_Y | BRW_WRITEMASK_Z) |
| 101 | #define BRW_WRITEMASK_XYZW (BRW_WRITEMASK_X | BRW_WRITEMASK_Y | \ |
| 102 | BRW_WRITEMASK_Z | BRW_WRITEMASK_W) |
| 103 | |
| 104 | #define REG_SIZE (8*4) |
| 105 | |
| 106 | /* These aren't hardware structs, just something useful for us to pass around: |
| 107 | * |
| 108 | * Align1 operation has a lot of control over input ranges. Used in |
| 109 | * WM programs to implement shaders decomposed into "channel serial" |
| 110 | * or "structure of array" form: |
| 111 | */ |
| 112 | struct brw_reg { |
| 113 | unsigned type:4; |
| 114 | unsigned file:2; |
| 115 | unsigned nr:8; |
| 116 | unsigned subnr:5; /* :1 in align16 */ |
| 117 | unsigned negate:1; /* source only */ |
| 118 | unsigned abs:1; /* source only */ |
| 119 | unsigned vstride:4; /* source only */ |
| 120 | unsigned width:3; /* src only, align1 only */ |
| 121 | unsigned hstride:2; /* align1 only */ |
| 122 | unsigned address_mode:1; /* relative addressing, hopefully! */ |
| 123 | unsigned pad0:1; |
| 124 | |
| 125 | union { |
| 126 | struct { |
| 127 | unsigned swizzle:8; /* src only, align16 only */ |
| 128 | unsigned writemask:4; /* dest only, align16 only */ |
| 129 | int indirect_offset:10; /* relative addressing offset */ |
| 130 | unsigned pad1:10; /* two dwords total */ |
| 131 | } bits; |
| 132 | |
| 133 | float f; |
| 134 | int d; |
| 135 | unsigned ud; |
| 136 | } dw1; |
| 137 | }; |
| 138 | |
| 139 | |
| 140 | struct brw_indirect { |
| 141 | unsigned addr_subnr:4; |
| 142 | int addr_offset:10; |
| 143 | unsigned pad:18; |
| 144 | }; |
| 145 | |
| 146 | |
| 147 | static inline int |
| 148 | type_sz(unsigned type) |
| 149 | { |
| 150 | switch(type) { |
| 151 | case BRW_REGISTER_TYPE_UD: |
| 152 | case BRW_REGISTER_TYPE_D: |
| 153 | case BRW_REGISTER_TYPE_F: |
| 154 | return 4; |
| 155 | case BRW_REGISTER_TYPE_HF: |
| 156 | case BRW_REGISTER_TYPE_UW: |
| 157 | case BRW_REGISTER_TYPE_W: |
| 158 | return 2; |
| 159 | case BRW_REGISTER_TYPE_UB: |
| 160 | case BRW_REGISTER_TYPE_B: |
| 161 | return 1; |
| 162 | default: |
| 163 | return 0; |
| 164 | } |
| 165 | } |
| 166 | |
| 167 | /** |
| 168 | * Construct a brw_reg. |
| 169 | * \param file one of the BRW_x_REGISTER_FILE values |
| 170 | * \param nr register number/index |
| 171 | * \param subnr register sub number |
| 172 | * \param type one of BRW_REGISTER_TYPE_x |
| 173 | * \param vstride one of BRW_VERTICAL_STRIDE_x |
| 174 | * \param width one of BRW_WIDTH_x |
| 175 | * \param hstride one of BRW_HORIZONTAL_STRIDE_x |
| 176 | * \param swizzle one of BRW_SWIZZLE_x |
| 177 | * \param writemask BRW_WRITEMASK_X/Y/Z/W bitfield |
| 178 | */ |
| 179 | static inline struct brw_reg |
| 180 | brw_reg(unsigned file, |
| 181 | unsigned nr, |
| 182 | unsigned subnr, |
| 183 | unsigned type, |
| 184 | unsigned vstride, |
| 185 | unsigned width, |
| 186 | unsigned hstride, |
| 187 | unsigned swizzle, |
| 188 | unsigned writemask) |
| 189 | { |
| 190 | struct brw_reg reg; |
| 191 | if (file == BRW_GENERAL_REGISTER_FILE) |
| 192 | assert(nr < BRW_MAX_GRF); |
| 193 | else if (file == BRW_MESSAGE_REGISTER_FILE) |
| 194 | assert((nr & ~(1 << 7)) < BRW_MAX_MRF); |
| 195 | else if (file == BRW_ARCHITECTURE_REGISTER_FILE) |
| 196 | assert(nr <= BRW_ARF_TIMESTAMP); |
| 197 | |
| 198 | reg.type = type; |
| 199 | reg.file = file; |
| 200 | reg.nr = nr; |
| 201 | reg.subnr = subnr * type_sz(type); |
| 202 | reg.negate = 0; |
| 203 | reg.abs = 0; |
| 204 | reg.vstride = vstride; |
| 205 | reg.width = width; |
| 206 | reg.hstride = hstride; |
| 207 | reg.address_mode = BRW_ADDRESS_DIRECT; |
| 208 | reg.pad0 = 0; |
| 209 | |
| 210 | /* Could do better: If the reg is r5.3<0;1,0>, we probably want to |
| 211 | * set swizzle and writemask to W, as the lower bits of subnr will |
| 212 | * be lost when converted to align16. This is probably too much to |
| 213 | * keep track of as you'd want it adjusted by suboffset(), etc. |
| 214 | * Perhaps fix up when converting to align16? |
| 215 | */ |
| 216 | reg.dw1.bits.swizzle = swizzle; |
| 217 | reg.dw1.bits.writemask = writemask; |
| 218 | reg.dw1.bits.indirect_offset = 0; |
| 219 | reg.dw1.bits.pad1 = 0; |
| 220 | return reg; |
| 221 | } |
| 222 | |
| 223 | /** Construct float[16] register */ |
| 224 | static inline struct brw_reg |
| 225 | brw_vec16_reg(unsigned file, unsigned nr, unsigned subnr) |
| 226 | { |
| 227 | return brw_reg(file, |
| 228 | nr, |
| 229 | subnr, |
| 230 | BRW_REGISTER_TYPE_F, |
| 231 | BRW_VERTICAL_STRIDE_16, |
| 232 | BRW_WIDTH_16, |
| 233 | BRW_HORIZONTAL_STRIDE_1, |
| 234 | BRW_SWIZZLE_XYZW, |
| 235 | BRW_WRITEMASK_XYZW); |
| 236 | } |
| 237 | |
| 238 | /** Construct float[8] register */ |
| 239 | static inline struct brw_reg |
| 240 | brw_vec8_reg(unsigned file, unsigned nr, unsigned subnr) |
| 241 | { |
| 242 | return brw_reg(file, |
| 243 | nr, |
| 244 | subnr, |
| 245 | BRW_REGISTER_TYPE_F, |
| 246 | BRW_VERTICAL_STRIDE_8, |
| 247 | BRW_WIDTH_8, |
| 248 | BRW_HORIZONTAL_STRIDE_1, |
| 249 | BRW_SWIZZLE_XYZW, |
| 250 | BRW_WRITEMASK_XYZW); |
| 251 | } |
| 252 | |
| 253 | /** Construct float[4] register */ |
| 254 | static inline struct brw_reg |
| 255 | brw_vec4_reg(unsigned file, unsigned nr, unsigned subnr) |
| 256 | { |
| 257 | return brw_reg(file, |
| 258 | nr, |
| 259 | subnr, |
| 260 | BRW_REGISTER_TYPE_F, |
| 261 | BRW_VERTICAL_STRIDE_4, |
| 262 | BRW_WIDTH_4, |
| 263 | BRW_HORIZONTAL_STRIDE_1, |
| 264 | BRW_SWIZZLE_XYZW, |
| 265 | BRW_WRITEMASK_XYZW); |
| 266 | } |
| 267 | |
| 268 | /** Construct float[2] register */ |
| 269 | static inline struct brw_reg |
| 270 | brw_vec2_reg(unsigned file, unsigned nr, unsigned subnr) |
| 271 | { |
| 272 | return brw_reg(file, |
| 273 | nr, |
| 274 | subnr, |
| 275 | BRW_REGISTER_TYPE_F, |
| 276 | BRW_VERTICAL_STRIDE_2, |
| 277 | BRW_WIDTH_2, |
| 278 | BRW_HORIZONTAL_STRIDE_1, |
| 279 | BRW_SWIZZLE_XYXY, |
| 280 | BRW_WRITEMASK_XY); |
| 281 | } |
| 282 | |
| 283 | /** Construct float[1] register */ |
| 284 | static inline struct brw_reg |
| 285 | brw_vec1_reg(unsigned file, unsigned nr, unsigned subnr) |
| 286 | { |
| 287 | return brw_reg(file, |
| 288 | nr, |
| 289 | subnr, |
| 290 | BRW_REGISTER_TYPE_F, |
| 291 | BRW_VERTICAL_STRIDE_0, |
| 292 | BRW_WIDTH_1, |
| 293 | BRW_HORIZONTAL_STRIDE_0, |
| 294 | BRW_SWIZZLE_XXXX, |
| 295 | BRW_WRITEMASK_X); |
| 296 | } |
| 297 | |
| 298 | |
| 299 | static inline struct brw_reg |
| 300 | retype(struct brw_reg reg, unsigned type) |
| 301 | { |
| 302 | reg.type = type; |
| 303 | return reg; |
| 304 | } |
| 305 | |
| 306 | static inline struct brw_reg |
| 307 | sechalf(struct brw_reg reg) |
| 308 | { |
| 309 | if (reg.vstride) |
| 310 | reg.nr++; |
| 311 | return reg; |
| 312 | } |
| 313 | |
| 314 | static inline struct brw_reg |
| 315 | suboffset(struct brw_reg reg, unsigned delta) |
| 316 | { |
| 317 | reg.subnr += delta * type_sz(reg.type); |
| 318 | return reg; |
| 319 | } |
| 320 | |
| 321 | |
| 322 | static inline struct brw_reg |
| 323 | offset(struct brw_reg reg, unsigned delta) |
| 324 | { |
| 325 | reg.nr += delta; |
| 326 | return reg; |
| 327 | } |
| 328 | |
| 329 | |
| 330 | static inline struct brw_reg |
| 331 | byte_offset(struct brw_reg reg, unsigned bytes) |
| 332 | { |
| 333 | unsigned newoffset = reg.nr * REG_SIZE + reg.subnr + bytes; |
| 334 | reg.nr = newoffset / REG_SIZE; |
| 335 | reg.subnr = newoffset % REG_SIZE; |
| 336 | return reg; |
| 337 | } |
| 338 | |
| 339 | |
| 340 | /** Construct unsigned word[16] register */ |
| 341 | static inline struct brw_reg |
| 342 | brw_uw16_reg(unsigned file, unsigned nr, unsigned subnr) |
| 343 | { |
| 344 | return suboffset(retype(brw_vec16_reg(file, nr, 0), BRW_REGISTER_TYPE_UW), subnr); |
| 345 | } |
| 346 | |
| 347 | /** Construct unsigned word[8] register */ |
| 348 | static inline struct brw_reg |
| 349 | brw_uw8_reg(unsigned file, unsigned nr, unsigned subnr) |
| 350 | { |
| 351 | return suboffset(retype(brw_vec8_reg(file, nr, 0), BRW_REGISTER_TYPE_UW), subnr); |
| 352 | } |
| 353 | |
| 354 | /** Construct unsigned word[1] register */ |
| 355 | static inline struct brw_reg |
| 356 | brw_uw1_reg(unsigned file, unsigned nr, unsigned subnr) |
| 357 | { |
| 358 | return suboffset(retype(brw_vec1_reg(file, nr, 0), BRW_REGISTER_TYPE_UW), subnr); |
| 359 | } |
| 360 | |
| 361 | static inline struct brw_reg |
| 362 | brw_imm_reg(unsigned type) |
| 363 | { |
| 364 | return brw_reg(BRW_IMMEDIATE_VALUE, |
| 365 | 0, |
| 366 | 0, |
| 367 | type, |
| 368 | BRW_VERTICAL_STRIDE_0, |
| 369 | BRW_WIDTH_1, |
| 370 | BRW_HORIZONTAL_STRIDE_0, |
| 371 | 0, |
| 372 | 0); |
| 373 | } |
| 374 | |
| 375 | /** Construct float immediate register */ |
| 376 | static inline struct brw_reg |
| 377 | brw_imm_f(float f) |
| 378 | { |
| 379 | struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_F); |
| 380 | imm.dw1.f = f; |
| 381 | return imm; |
| 382 | } |
| 383 | |
| 384 | /** Construct integer immediate register */ |
| 385 | static inline struct brw_reg |
| 386 | brw_imm_d(int d) |
| 387 | { |
| 388 | struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_D); |
| 389 | imm.dw1.d = d; |
| 390 | return imm; |
| 391 | } |
| 392 | |
| 393 | /** Construct uint immediate register */ |
| 394 | static inline struct brw_reg |
| 395 | brw_imm_ud(unsigned ud) |
| 396 | { |
| 397 | struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_UD); |
| 398 | imm.dw1.ud = ud; |
| 399 | return imm; |
| 400 | } |
| 401 | |
| 402 | /** Construct ushort immediate register */ |
| 403 | static inline struct brw_reg |
| 404 | brw_imm_uw(uint16_t uw) |
| 405 | { |
| 406 | struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_UW); |
| 407 | imm.dw1.ud = uw | (uw << 16); |
| 408 | return imm; |
| 409 | } |
| 410 | |
| 411 | /** Construct short immediate register */ |
| 412 | static inline struct brw_reg |
| 413 | brw_imm_w(int16_t w) |
| 414 | { |
| 415 | struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_W); |
| 416 | imm.dw1.d = w | (w << 16); |
| 417 | return imm; |
| 418 | } |
| 419 | |
| 420 | /* brw_imm_b and brw_imm_ub aren't supported by hardware - the type |
| 421 | * numbers alias with _V and _VF below: |
| 422 | */ |
| 423 | |
| 424 | /** Construct vector of eight signed half-byte values */ |
| 425 | static inline struct brw_reg |
| 426 | brw_imm_v(unsigned v) |
| 427 | { |
| 428 | struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_V); |
| 429 | imm.vstride = BRW_VERTICAL_STRIDE_0; |
| 430 | imm.width = BRW_WIDTH_8; |
| 431 | imm.hstride = BRW_HORIZONTAL_STRIDE_1; |
| 432 | imm.dw1.ud = v; |
| 433 | return imm; |
| 434 | } |
| 435 | |
| 436 | /** Construct vector of four 8-bit float values */ |
| 437 | static inline struct brw_reg |
| 438 | brw_imm_vf(unsigned v) |
| 439 | { |
| 440 | struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_VF); |
| 441 | imm.vstride = BRW_VERTICAL_STRIDE_0; |
| 442 | imm.width = BRW_WIDTH_4; |
| 443 | imm.hstride = BRW_HORIZONTAL_STRIDE_1; |
| 444 | imm.dw1.ud = v; |
| 445 | return imm; |
| 446 | } |
| 447 | |
| 448 | #define VF_ZERO 0x0 |
| 449 | #define VF_ONE 0x30 |
| 450 | #define VF_NEG (1<<7) |
| 451 | |
| 452 | static inline struct brw_reg |
| 453 | brw_imm_vf4(unsigned v0, unsigned v1, unsigned v2, unsigned v3) |
| 454 | { |
| 455 | struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_VF); |
| 456 | imm.vstride = BRW_VERTICAL_STRIDE_0; |
| 457 | imm.width = BRW_WIDTH_4; |
| 458 | imm.hstride = BRW_HORIZONTAL_STRIDE_1; |
| 459 | imm.dw1.ud = ((v0 << 0) | (v1 << 8) | (v2 << 16) | (v3 << 24)); |
| 460 | return imm; |
| 461 | } |
| 462 | |
| 463 | |
| 464 | static inline struct brw_reg |
| 465 | brw_address(struct brw_reg reg) |
| 466 | { |
| 467 | return brw_imm_uw(reg.nr * REG_SIZE + reg.subnr); |
| 468 | } |
| 469 | |
| 470 | /** Construct float[1] general-purpose register */ |
| 471 | static inline struct brw_reg |
| 472 | brw_vec1_grf(unsigned nr, unsigned subnr) |
| 473 | { |
| 474 | return brw_vec1_reg(BRW_GENERAL_REGISTER_FILE, nr, subnr); |
| 475 | } |
| 476 | |
| 477 | /** Construct float[2] general-purpose register */ |
| 478 | static inline struct brw_reg |
| 479 | brw_vec2_grf(unsigned nr, unsigned subnr) |
| 480 | { |
| 481 | return brw_vec2_reg(BRW_GENERAL_REGISTER_FILE, nr, subnr); |
| 482 | } |
| 483 | |
| 484 | /** Construct float[4] general-purpose register */ |
| 485 | static inline struct brw_reg |
| 486 | brw_vec4_grf(unsigned nr, unsigned subnr) |
| 487 | { |
| 488 | return brw_vec4_reg(BRW_GENERAL_REGISTER_FILE, nr, subnr); |
| 489 | } |
| 490 | |
| 491 | /** Construct float[8] general-purpose register */ |
| 492 | static inline struct brw_reg |
| 493 | brw_vec8_grf(unsigned nr, unsigned subnr) |
| 494 | { |
| 495 | return brw_vec8_reg(BRW_GENERAL_REGISTER_FILE, nr, subnr); |
| 496 | } |
| 497 | |
| 498 | |
| 499 | static inline struct brw_reg |
| 500 | brw_uw8_grf(unsigned nr, unsigned subnr) |
| 501 | { |
| 502 | return brw_uw8_reg(BRW_GENERAL_REGISTER_FILE, nr, subnr); |
| 503 | } |
| 504 | |
| 505 | static inline struct brw_reg |
| 506 | brw_uw16_grf(unsigned nr, unsigned subnr) |
| 507 | { |
| 508 | return brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, nr, subnr); |
| 509 | } |
| 510 | |
| 511 | |
| 512 | /** Construct null register (usually used for setting condition codes) */ |
| 513 | static inline struct brw_reg |
| 514 | brw_null_reg(void) |
| 515 | { |
| 516 | return brw_vec8_reg(BRW_ARCHITECTURE_REGISTER_FILE, BRW_ARF_NULL, 0); |
| 517 | } |
| 518 | |
| 519 | static inline struct brw_reg |
| 520 | brw_address_reg(unsigned subnr) |
| 521 | { |
| 522 | return brw_uw1_reg(BRW_ARCHITECTURE_REGISTER_FILE, BRW_ARF_ADDRESS, subnr); |
| 523 | } |
| 524 | |
| 525 | /* If/else instructions break in align16 mode if writemask & swizzle |
| 526 | * aren't xyzw. This goes against the convention for other scalar |
| 527 | * regs: |
| 528 | */ |
| 529 | static inline struct brw_reg |
| 530 | brw_ip_reg(void) |
| 531 | { |
| 532 | return brw_reg(BRW_ARCHITECTURE_REGISTER_FILE, |
| 533 | BRW_ARF_IP, |
| 534 | 0, |
| 535 | BRW_REGISTER_TYPE_UD, |
| 536 | BRW_VERTICAL_STRIDE_4, /* ? */ |
| 537 | BRW_WIDTH_1, |
| 538 | BRW_HORIZONTAL_STRIDE_0, |
| 539 | BRW_SWIZZLE_XYZW, /* NOTE! */ |
| 540 | BRW_WRITEMASK_XYZW); /* NOTE! */ |
| 541 | } |
| 542 | |
| 543 | static inline struct brw_reg |
| 544 | brw_acc_reg(void) |
| 545 | { |
| 546 | return brw_vec8_reg(BRW_ARCHITECTURE_REGISTER_FILE, BRW_ARF_ACCUMULATOR, 0); |
| 547 | } |
| 548 | |
| 549 | static inline struct brw_reg |
| 550 | brw_notification_1_reg(void) |
| 551 | { |
| 552 | |
| 553 | return brw_reg(BRW_ARCHITECTURE_REGISTER_FILE, |
| 554 | BRW_ARF_NOTIFICATION_COUNT, |
| 555 | 1, |
| 556 | BRW_REGISTER_TYPE_UD, |
| 557 | BRW_VERTICAL_STRIDE_0, |
| 558 | BRW_WIDTH_1, |
| 559 | BRW_HORIZONTAL_STRIDE_0, |
| 560 | BRW_SWIZZLE_XXXX, |
| 561 | BRW_WRITEMASK_X); |
| 562 | } |
| 563 | |
| 564 | |
| 565 | static inline struct brw_reg |
| 566 | brw_flag_reg(int reg, int subreg) |
| 567 | { |
| 568 | return brw_uw1_reg(BRW_ARCHITECTURE_REGISTER_FILE, |
| 569 | BRW_ARF_FLAG + reg, subreg); |
| 570 | } |
| 571 | |
| 572 | |
| 573 | static inline struct brw_reg |
| 574 | brw_mask_reg(unsigned subnr) |
| 575 | { |
| 576 | return brw_uw1_reg(BRW_ARCHITECTURE_REGISTER_FILE, BRW_ARF_MASK, subnr); |
| 577 | } |
| 578 | |
| 579 | static inline struct brw_reg |
| 580 | brw_message_reg(unsigned nr) |
| 581 | { |
| 582 | assert((nr & ~(1 << 7)) < BRW_MAX_MRF); |
| 583 | return brw_vec8_reg(BRW_MESSAGE_REGISTER_FILE, nr, 0); |
| 584 | } |
| 585 | |
| 586 | |
| 587 | /* This is almost always called with a numeric constant argument, so |
| 588 | * make things easy to evaluate at compile time: |
| 589 | */ |
| 590 | static inline unsigned cvt(unsigned val) |
| 591 | { |
| 592 | switch (val) { |
| 593 | case 0: return 0; |
| 594 | case 1: return 1; |
| 595 | case 2: return 2; |
| 596 | case 4: return 3; |
| 597 | case 8: return 4; |
| 598 | case 16: return 5; |
| 599 | case 32: return 6; |
| 600 | } |
| 601 | return 0; |
| 602 | } |
| 603 | |
| 604 | static inline struct brw_reg |
| 605 | stride(struct brw_reg reg, unsigned vstride, unsigned width, unsigned hstride) |
| 606 | { |
| 607 | reg.vstride = cvt(vstride); |
| 608 | reg.width = cvt(width) - 1; |
| 609 | reg.hstride = cvt(hstride); |
| 610 | return reg; |
| 611 | } |
| 612 | |
| 613 | |
| 614 | static inline struct brw_reg |
| 615 | vec16(struct brw_reg reg) |
| 616 | { |
| 617 | return stride(reg, 16,16,1); |
| 618 | } |
| 619 | |
| 620 | static inline struct brw_reg |
| 621 | vec8(struct brw_reg reg) |
| 622 | { |
| 623 | return stride(reg, 8,8,1); |
| 624 | } |
| 625 | |
| 626 | static inline struct brw_reg |
| 627 | vec4(struct brw_reg reg) |
| 628 | { |
| 629 | return stride(reg, 4,4,1); |
| 630 | } |
| 631 | |
| 632 | static inline struct brw_reg |
| 633 | vec2(struct brw_reg reg) |
| 634 | { |
| 635 | return stride(reg, 2,2,1); |
| 636 | } |
| 637 | |
| 638 | static inline struct brw_reg |
| 639 | vec1(struct brw_reg reg) |
| 640 | { |
| 641 | return stride(reg, 0,1,0); |
| 642 | } |
| 643 | |
| 644 | |
| 645 | static inline struct brw_reg |
| 646 | get_element(struct brw_reg reg, unsigned elt) |
| 647 | { |
| 648 | return vec1(suboffset(reg, elt)); |
| 649 | } |
| 650 | |
| 651 | static inline struct brw_reg |
| 652 | get_element_ud(struct brw_reg reg, unsigned elt) |
| 653 | { |
| 654 | return vec1(suboffset(retype(reg, BRW_REGISTER_TYPE_UD), elt)); |
| 655 | } |
| 656 | |
| 657 | static inline struct brw_reg |
| 658 | get_element_d(struct brw_reg reg, unsigned elt) |
| 659 | { |
| 660 | return vec1(suboffset(retype(reg, BRW_REGISTER_TYPE_D), elt)); |
| 661 | } |
| 662 | |
| 663 | |
| 664 | static inline struct brw_reg |
| 665 | brw_swizzle(struct brw_reg reg, unsigned x, unsigned y, unsigned z, unsigned w) |
| 666 | { |
| 667 | assert(reg.file != BRW_IMMEDIATE_VALUE); |
| 668 | |
| 669 | reg.dw1.bits.swizzle = BRW_SWIZZLE4(BRW_GET_SWZ(reg.dw1.bits.swizzle, x), |
| 670 | BRW_GET_SWZ(reg.dw1.bits.swizzle, y), |
| 671 | BRW_GET_SWZ(reg.dw1.bits.swizzle, z), |
| 672 | BRW_GET_SWZ(reg.dw1.bits.swizzle, w)); |
| 673 | return reg; |
| 674 | } |
| 675 | |
| 676 | |
| 677 | static inline struct brw_reg |
| 678 | brw_swizzle1(struct brw_reg reg, unsigned x) |
| 679 | { |
| 680 | return brw_swizzle(reg, x, x, x, x); |
| 681 | } |
| 682 | |
| 683 | static inline struct brw_reg |
| 684 | brw_writemask(struct brw_reg reg, unsigned mask) |
| 685 | { |
| 686 | assert(reg.file != BRW_IMMEDIATE_VALUE); |
| 687 | reg.dw1.bits.writemask &= mask; |
| 688 | return reg; |
| 689 | } |
| 690 | |
| 691 | static inline struct brw_reg |
| 692 | brw_set_writemask(struct brw_reg reg, unsigned mask) |
| 693 | { |
| 694 | assert(reg.file != BRW_IMMEDIATE_VALUE); |
| 695 | reg.dw1.bits.writemask = mask; |
| 696 | return reg; |
| 697 | } |
| 698 | |
| 699 | static inline struct brw_reg |
| 700 | negate(struct brw_reg reg) |
| 701 | { |
| 702 | reg.negate ^= 1; |
| 703 | return reg; |
| 704 | } |
| 705 | |
| 706 | static inline struct brw_reg |
| 707 | brw_abs(struct brw_reg reg) |
| 708 | { |
| 709 | reg.abs = 1; |
| 710 | reg.negate = 0; |
| 711 | return reg; |
| 712 | } |
| 713 | |
| 714 | /************************************************************************/ |
| 715 | |
| 716 | static inline struct brw_reg |
| 717 | brw_vec4_indirect(unsigned subnr, int offset) |
| 718 | { |
| 719 | struct brw_reg reg = brw_vec4_grf(0, 0); |
| 720 | reg.subnr = subnr; |
| 721 | reg.address_mode = BRW_ADDRESS_REGISTER_INDIRECT_REGISTER; |
| 722 | reg.dw1.bits.indirect_offset = offset; |
| 723 | return reg; |
| 724 | } |
| 725 | |
| 726 | static inline struct brw_reg |
| 727 | brw_vec1_indirect(unsigned subnr, int offset) |
| 728 | { |
| 729 | struct brw_reg reg = brw_vec1_grf(0, 0); |
| 730 | reg.subnr = subnr; |
| 731 | reg.address_mode = BRW_ADDRESS_REGISTER_INDIRECT_REGISTER; |
| 732 | reg.dw1.bits.indirect_offset = offset; |
| 733 | return reg; |
| 734 | } |
| 735 | |
| 736 | static inline struct brw_reg |
| 737 | deref_4f(struct brw_indirect ptr, int offset) |
| 738 | { |
| 739 | return brw_vec4_indirect(ptr.addr_subnr, ptr.addr_offset + offset); |
| 740 | } |
| 741 | |
| 742 | static inline struct brw_reg |
| 743 | deref_1f(struct brw_indirect ptr, int offset) |
| 744 | { |
| 745 | return brw_vec1_indirect(ptr.addr_subnr, ptr.addr_offset + offset); |
| 746 | } |
| 747 | |
| 748 | static inline struct brw_reg |
| 749 | deref_4b(struct brw_indirect ptr, int offset) |
| 750 | { |
| 751 | return retype(deref_4f(ptr, offset), BRW_REGISTER_TYPE_B); |
| 752 | } |
| 753 | |
| 754 | static inline struct brw_reg |
| 755 | deref_1uw(struct brw_indirect ptr, int offset) |
| 756 | { |
| 757 | return retype(deref_1f(ptr, offset), BRW_REGISTER_TYPE_UW); |
| 758 | } |
| 759 | |
| 760 | static inline struct brw_reg |
| 761 | deref_1d(struct brw_indirect ptr, int offset) |
| 762 | { |
| 763 | return retype(deref_1f(ptr, offset), BRW_REGISTER_TYPE_D); |
| 764 | } |
| 765 | |
| 766 | static inline struct brw_reg |
| 767 | deref_1ud(struct brw_indirect ptr, int offset) |
| 768 | { |
| 769 | return retype(deref_1f(ptr, offset), BRW_REGISTER_TYPE_UD); |
| 770 | } |
| 771 | |
| 772 | static inline struct brw_reg |
| 773 | get_addr_reg(struct brw_indirect ptr) |
| 774 | { |
| 775 | return brw_address_reg(ptr.addr_subnr); |
| 776 | } |
| 777 | |
| 778 | static inline struct brw_indirect |
| 779 | brw_indirect_offset(struct brw_indirect ptr, int offset) |
| 780 | { |
| 781 | ptr.addr_offset += offset; |
| 782 | return ptr; |
| 783 | } |
| 784 | |
| 785 | static inline struct brw_indirect |
| 786 | brw_indirect(unsigned addr_subnr, int offset) |
| 787 | { |
| 788 | struct brw_indirect ptr; |
| 789 | ptr.addr_subnr = addr_subnr; |
| 790 | ptr.addr_offset = offset; |
| 791 | ptr.pad = 0; |
| 792 | return ptr; |
| 793 | } |
| 794 | |
| 795 | /** Do two brw_regs refer to the same register? */ |
| 796 | static inline bool |
| 797 | brw_same_reg(struct brw_reg r1, struct brw_reg r2) |
| 798 | { |
| 799 | return r1.file == r2.file && r1.nr == r2.nr; |
| 800 | } |
| 801 | |
| 802 | void brw_print_reg(struct brw_reg reg); |
| 803 | |
| 804 | #ifdef __cplusplus |
| 805 | } |
| 806 | #endif |
| 807 | |
| 808 | #endif |