Dave Airlie | eb9ad9f | 2016-03-22 07:59:35 +1000 | [diff] [blame^] | 1 | /* |
| 2 | * Copyright 2016 Red Hat. |
| 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 | * on the rights to use, copy, modify, merge, publish, distribute, sub |
| 8 | * license, and/or sell copies of the Software, and to permit persons to whom |
| 9 | * the 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 NON-INFRINGEMENT. IN NO EVENT SHALL |
| 18 | * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, |
| 19 | * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR |
| 20 | * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE |
| 21 | * USE OR OTHER DEALINGS IN THE SOFTWARE. |
| 22 | */ |
| 23 | |
| 24 | #include "sp_context.h" |
| 25 | #include "sp_image.h" |
| 26 | #include "sp_texture.h" |
| 27 | |
| 28 | #include "util/u_format.h" |
| 29 | |
| 30 | /* |
| 31 | * Get the offset into the base image |
| 32 | * first element for a buffer or layer/level for texture. |
| 33 | */ |
| 34 | static uint32_t |
| 35 | get_image_offset(const struct softpipe_resource *spr, |
| 36 | const struct pipe_image_view *iview, |
| 37 | enum pipe_format format, unsigned r_coord) |
| 38 | { |
| 39 | int base_layer = 0; |
| 40 | |
| 41 | if (spr->base.target == PIPE_BUFFER) |
| 42 | return iview->u.buf.first_element * util_format_get_blocksize(format); |
| 43 | |
| 44 | if (spr->base.target == PIPE_TEXTURE_1D_ARRAY || |
| 45 | spr->base.target == PIPE_TEXTURE_2D_ARRAY || |
| 46 | spr->base.target == PIPE_TEXTURE_CUBE_ARRAY || |
| 47 | spr->base.target == PIPE_TEXTURE_CUBE || |
| 48 | spr->base.target == PIPE_TEXTURE_3D) |
| 49 | base_layer = r_coord + iview->u.tex.first_layer; |
| 50 | return softpipe_get_tex_image_offset(spr, iview->u.tex.level, base_layer); |
| 51 | } |
| 52 | |
| 53 | /* |
| 54 | * Does this texture instruction have a layer or depth parameter. |
| 55 | */ |
| 56 | static inline bool |
| 57 | has_layer_or_depth(unsigned tgsi_tex_instr) |
| 58 | { |
| 59 | return (tgsi_tex_instr == TGSI_TEXTURE_3D || |
| 60 | tgsi_tex_instr == TGSI_TEXTURE_CUBE || |
| 61 | tgsi_tex_instr == TGSI_TEXTURE_1D_ARRAY || |
| 62 | tgsi_tex_instr == TGSI_TEXTURE_2D_ARRAY || |
| 63 | tgsi_tex_instr == TGSI_TEXTURE_CUBE_ARRAY || |
| 64 | tgsi_tex_instr == TGSI_TEXTURE_2D_ARRAY_MSAA); |
| 65 | } |
| 66 | |
| 67 | /* |
| 68 | * Is this texture instruction a single non-array coordinate. |
| 69 | */ |
| 70 | static inline bool |
| 71 | has_1coord(unsigned tgsi_tex_instr) |
| 72 | { |
| 73 | return (tgsi_tex_instr == TGSI_TEXTURE_BUFFER || |
| 74 | tgsi_tex_instr == TGSI_TEXTURE_1D || |
| 75 | tgsi_tex_instr == TGSI_TEXTURE_1D_ARRAY); |
| 76 | } |
| 77 | |
| 78 | /* |
| 79 | * check the bounds vs w/h/d |
| 80 | */ |
| 81 | static inline bool |
| 82 | bounds_check(int width, int height, int depth, |
| 83 | int s, int t, int r) |
| 84 | { |
| 85 | if (s < 0 || s >= width) |
| 86 | return false; |
| 87 | if (t < 0 || t >= height) |
| 88 | return false; |
| 89 | if (r < 0 || r >= depth) |
| 90 | return false; |
| 91 | return true; |
| 92 | } |
| 93 | |
| 94 | /* |
| 95 | * Checks if the texture target compatible with the image resource |
| 96 | * pipe target. |
| 97 | */ |
| 98 | static inline bool |
| 99 | has_compat_target(unsigned pipe_target, unsigned tgsi_target) |
| 100 | { |
| 101 | switch (pipe_target) { |
| 102 | case PIPE_TEXTURE_1D: |
| 103 | if (tgsi_target == TGSI_TEXTURE_1D) |
| 104 | return true; |
| 105 | break; |
| 106 | case PIPE_TEXTURE_2D: |
| 107 | if (tgsi_target == TGSI_TEXTURE_2D) |
| 108 | return true; |
| 109 | break; |
| 110 | case PIPE_TEXTURE_RECT: |
| 111 | if (tgsi_target == TGSI_TEXTURE_RECT) |
| 112 | return true; |
| 113 | break; |
| 114 | case PIPE_TEXTURE_3D: |
| 115 | if (tgsi_target == TGSI_TEXTURE_3D || |
| 116 | tgsi_target == TGSI_TEXTURE_2D) |
| 117 | return true; |
| 118 | break; |
| 119 | case PIPE_TEXTURE_CUBE: |
| 120 | if (tgsi_target == TGSI_TEXTURE_CUBE || |
| 121 | tgsi_target == TGSI_TEXTURE_2D) |
| 122 | return true; |
| 123 | break; |
| 124 | case PIPE_TEXTURE_1D_ARRAY: |
| 125 | if (tgsi_target == TGSI_TEXTURE_1D || |
| 126 | tgsi_target == TGSI_TEXTURE_1D_ARRAY) |
| 127 | return true; |
| 128 | break; |
| 129 | case PIPE_TEXTURE_2D_ARRAY: |
| 130 | if (tgsi_target == TGSI_TEXTURE_2D || |
| 131 | tgsi_target == TGSI_TEXTURE_2D_ARRAY) |
| 132 | return true; |
| 133 | break; |
| 134 | case PIPE_TEXTURE_CUBE_ARRAY: |
| 135 | if (tgsi_target == TGSI_TEXTURE_CUBE || |
| 136 | tgsi_target == TGSI_TEXTURE_CUBE_ARRAY || |
| 137 | tgsi_target == TGSI_TEXTURE_2D) |
| 138 | return true; |
| 139 | break; |
| 140 | case PIPE_BUFFER: |
| 141 | return (tgsi_target == TGSI_TEXTURE_BUFFER); |
| 142 | } |
| 143 | return false; |
| 144 | } |
| 145 | |
| 146 | static bool |
| 147 | get_dimensions(const struct pipe_image_view *iview, |
| 148 | const struct softpipe_resource *spr, |
| 149 | unsigned tgsi_tex_instr, |
| 150 | enum pipe_format pformat, |
| 151 | unsigned *width, |
| 152 | unsigned *height, |
| 153 | unsigned *depth) |
| 154 | { |
| 155 | if (tgsi_tex_instr == TGSI_TEXTURE_BUFFER) { |
| 156 | *width = iview->u.buf.last_element - iview->u.buf.first_element + 1; |
| 157 | *height = 1; |
| 158 | *depth = 1; |
| 159 | /* |
| 160 | * Bounds check the buffer size from the view |
| 161 | * and the buffer size from the underlying buffer. |
| 162 | */ |
| 163 | if (util_format_get_stride(pformat, *width) > |
| 164 | util_format_get_stride(spr->base.format, spr->base.width0)) |
| 165 | return false; |
| 166 | } else { |
| 167 | unsigned level; |
| 168 | |
| 169 | level = spr->base.target == PIPE_BUFFER ? 0 : iview->u.tex.level; |
| 170 | *width = u_minify(spr->base.width0, level); |
| 171 | *height = u_minify(spr->base.height0, level); |
| 172 | |
| 173 | if (spr->base.target == TGSI_TEXTURE_3D) |
| 174 | *depth = u_minify(spr->base.depth0, level); |
| 175 | else |
| 176 | *depth = spr->base.array_size; |
| 177 | |
| 178 | /* Make sure the resource and view have compatiable formats */ |
| 179 | if (util_format_get_blocksize(pformat) > |
| 180 | util_format_get_blocksize(spr->base.format)) |
| 181 | return false; |
| 182 | } |
| 183 | return true; |
| 184 | } |
| 185 | |
| 186 | static void |
| 187 | fill_coords(const struct tgsi_image_params *params, |
| 188 | unsigned index, |
| 189 | const int s[TGSI_QUAD_SIZE], |
| 190 | const int t[TGSI_QUAD_SIZE], |
| 191 | const int r[TGSI_QUAD_SIZE], |
| 192 | int *s_coord, int *t_coord, int *r_coord) |
| 193 | { |
| 194 | *s_coord = s[index]; |
| 195 | *t_coord = has_1coord(params->tgsi_tex_instr) ? 0 : t[index]; |
| 196 | *r_coord = has_layer_or_depth(params->tgsi_tex_instr) ? |
| 197 | (params->tgsi_tex_instr == TGSI_TEXTURE_1D_ARRAY ? t[index] : r[index]) : 0; |
| 198 | } |
| 199 | /* |
| 200 | * Implement the image LOAD operation. |
| 201 | */ |
| 202 | static void |
| 203 | sp_tgsi_load(const struct tgsi_image *image, |
| 204 | const struct tgsi_image_params *params, |
| 205 | const int s[TGSI_QUAD_SIZE], |
| 206 | const int t[TGSI_QUAD_SIZE], |
| 207 | const int r[TGSI_QUAD_SIZE], |
| 208 | const int sample[TGSI_QUAD_SIZE], |
| 209 | float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE]) |
| 210 | { |
| 211 | struct sp_tgsi_image *sp_img = (struct sp_tgsi_image *)image; |
| 212 | struct pipe_image_view *iview; |
| 213 | struct softpipe_resource *spr; |
| 214 | unsigned width, height, depth; |
| 215 | unsigned stride; |
| 216 | int c, j; |
| 217 | char *data_ptr; |
| 218 | unsigned offset = 0; |
| 219 | |
| 220 | if (params->unit > PIPE_MAX_SHADER_IMAGES) |
| 221 | goto fail_write_all_zero; |
| 222 | iview = &sp_img->sp_iview[params->unit]; |
| 223 | spr = (struct softpipe_resource *)iview->resource; |
| 224 | if (!spr) |
| 225 | goto fail_write_all_zero; |
| 226 | |
| 227 | if (!has_compat_target(spr->base.target, params->tgsi_tex_instr)) |
| 228 | goto fail_write_all_zero; |
| 229 | |
| 230 | if (!get_dimensions(iview, spr, params->tgsi_tex_instr, |
| 231 | params->format, &width, &height, &depth)) |
| 232 | return; |
| 233 | |
| 234 | stride = util_format_get_stride(params->format, width); |
| 235 | |
| 236 | for (j = 0; j < TGSI_QUAD_SIZE; j++) { |
| 237 | int s_coord, t_coord, r_coord; |
| 238 | bool fill_zero = false; |
| 239 | |
| 240 | if (!(params->execmask & (1 << j))) |
| 241 | fill_zero = true; |
| 242 | |
| 243 | fill_coords(params, j, s, t, r, &s_coord, &t_coord, &r_coord); |
| 244 | if (!bounds_check(width, height, depth, |
| 245 | s_coord, t_coord, r_coord)) |
| 246 | fill_zero = true; |
| 247 | |
| 248 | if (fill_zero) { |
| 249 | int nc = util_format_get_nr_components(params->format); |
| 250 | int ival = util_format_is_pure_integer(params->format); |
| 251 | for (c = 0; c < 4; c++) { |
| 252 | rgba[c][j] = 0; |
| 253 | if (c == 3 && nc < 4) { |
| 254 | if (ival) |
| 255 | ((int32_t *)rgba[c])[j] = 1; |
| 256 | else |
| 257 | rgba[c][j] = 1.0; |
| 258 | } |
| 259 | } |
| 260 | continue; |
| 261 | } |
| 262 | offset = get_image_offset(spr, iview, params->format, r_coord); |
| 263 | data_ptr = (char *)spr->data + offset; |
| 264 | |
| 265 | if (util_format_is_pure_sint(params->format)) { |
| 266 | int32_t sdata[4]; |
| 267 | |
| 268 | util_format_read_4i(params->format, |
| 269 | sdata, 0, |
| 270 | data_ptr, stride, |
| 271 | s_coord, t_coord, 1, 1); |
| 272 | for (c = 0; c < 4; c++) |
| 273 | ((int32_t *)rgba[c])[j] = sdata[c]; |
| 274 | } else if (util_format_is_pure_uint(params->format)) { |
| 275 | uint32_t sdata[4]; |
| 276 | util_format_read_4ui(params->format, |
| 277 | sdata, 0, |
| 278 | data_ptr, stride, |
| 279 | s_coord, t_coord, 1, 1); |
| 280 | for (c = 0; c < 4; c++) |
| 281 | ((uint32_t *)rgba[c])[j] = sdata[c]; |
| 282 | } else { |
| 283 | float sdata[4]; |
| 284 | util_format_read_4f(params->format, |
| 285 | sdata, 0, |
| 286 | data_ptr, stride, |
| 287 | s_coord, t_coord, 1, 1); |
| 288 | for (c = 0; c < 4; c++) |
| 289 | rgba[c][j] = sdata[c]; |
| 290 | } |
| 291 | } |
| 292 | return; |
| 293 | fail_write_all_zero: |
| 294 | for (j = 0; j < TGSI_QUAD_SIZE; j++) { |
| 295 | for (c = 0; c < 4; c++) |
| 296 | rgba[c][j] = 0; |
| 297 | } |
| 298 | return; |
| 299 | } |
| 300 | |
| 301 | /* |
| 302 | * Implement the image STORE operation. |
| 303 | */ |
| 304 | static void |
| 305 | sp_tgsi_store(const struct tgsi_image *image, |
| 306 | const struct tgsi_image_params *params, |
| 307 | const int s[TGSI_QUAD_SIZE], |
| 308 | const int t[TGSI_QUAD_SIZE], |
| 309 | const int r[TGSI_QUAD_SIZE], |
| 310 | const int sample[TGSI_QUAD_SIZE], |
| 311 | float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE]) |
| 312 | { |
| 313 | struct sp_tgsi_image *sp_img = (struct sp_tgsi_image *)image; |
| 314 | struct pipe_image_view *iview; |
| 315 | struct softpipe_resource *spr; |
| 316 | unsigned width, height, depth; |
| 317 | unsigned stride; |
| 318 | char *data_ptr; |
| 319 | int j, c; |
| 320 | unsigned offset = 0; |
| 321 | unsigned pformat = params->format; |
| 322 | |
| 323 | if (params->unit > PIPE_MAX_SHADER_IMAGES) |
| 324 | return; |
| 325 | iview = &sp_img->sp_iview[params->unit]; |
| 326 | spr = (struct softpipe_resource *)iview->resource; |
| 327 | if (!spr) |
| 328 | return; |
| 329 | if (!has_compat_target(spr->base.target, params->tgsi_tex_instr)) |
| 330 | return; |
| 331 | |
| 332 | if (params->format == PIPE_FORMAT_NONE) |
| 333 | pformat = spr->base.format; |
| 334 | |
| 335 | if (!get_dimensions(iview, spr, params->tgsi_tex_instr, |
| 336 | pformat, &width, &height, &depth)) |
| 337 | return; |
| 338 | |
| 339 | stride = util_format_get_stride(pformat, width); |
| 340 | |
| 341 | for (j = 0; j < TGSI_QUAD_SIZE; j++) { |
| 342 | int s_coord, t_coord, r_coord; |
| 343 | |
| 344 | if (!(params->execmask & (1 << j))) |
| 345 | continue; |
| 346 | |
| 347 | fill_coords(params, j, s, t, r, &s_coord, &t_coord, &r_coord); |
| 348 | if (!bounds_check(width, height, depth, |
| 349 | s_coord, t_coord, r_coord)) |
| 350 | continue; |
| 351 | |
| 352 | offset = get_image_offset(spr, iview, pformat, r_coord); |
| 353 | data_ptr = (char *)spr->data + offset; |
| 354 | |
| 355 | if (util_format_is_pure_sint(pformat)) { |
| 356 | int32_t sdata[4]; |
| 357 | for (c = 0; c < 4; c++) |
| 358 | sdata[c] = ((int32_t *)rgba[c])[j]; |
| 359 | util_format_write_4i(pformat, sdata, 0, data_ptr, stride, |
| 360 | s_coord, t_coord, 1, 1); |
| 361 | } else if (util_format_is_pure_uint(pformat)) { |
| 362 | uint32_t sdata[4]; |
| 363 | for (c = 0; c < 4; c++) |
| 364 | sdata[c] = ((uint32_t *)rgba[c])[j]; |
| 365 | util_format_write_4ui(pformat, sdata, 0, data_ptr, stride, |
| 366 | s_coord, t_coord, 1, 1); |
| 367 | } else { |
| 368 | float sdata[4]; |
| 369 | for (c = 0; c < 4; c++) |
| 370 | sdata[c] = rgba[c][j]; |
| 371 | util_format_write_4f(pformat, sdata, 0, data_ptr, stride, |
| 372 | s_coord, t_coord, 1, 1); |
| 373 | } |
| 374 | } |
| 375 | } |
| 376 | |
| 377 | /* |
| 378 | * Implement atomic operations on unsigned integers. |
| 379 | */ |
| 380 | static void |
| 381 | handle_op_uint(const struct pipe_image_view *iview, |
| 382 | const struct tgsi_image_params *params, |
| 383 | bool just_read, |
| 384 | char *data_ptr, |
| 385 | uint qi, |
| 386 | unsigned stride, |
| 387 | unsigned opcode, |
| 388 | int s, |
| 389 | int t, |
| 390 | float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE], |
| 391 | float rgba2[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE]) |
| 392 | { |
| 393 | uint c; |
| 394 | int nc = util_format_get_nr_components(params->format); |
| 395 | unsigned sdata[4]; |
| 396 | |
| 397 | util_format_read_4ui(params->format, |
| 398 | sdata, 0, |
| 399 | data_ptr, stride, |
| 400 | s, t, 1, 1); |
| 401 | |
| 402 | if (just_read) { |
| 403 | for (c = 0; c < nc; c++) { |
| 404 | ((uint32_t *)rgba[c])[qi] = sdata[c]; |
| 405 | } |
| 406 | return; |
| 407 | } |
| 408 | switch (opcode) { |
| 409 | case TGSI_OPCODE_ATOMUADD: |
| 410 | for (c = 0; c < nc; c++) { |
| 411 | unsigned temp = sdata[c]; |
| 412 | sdata[c] += ((uint32_t *)rgba[c])[qi]; |
| 413 | ((uint32_t *)rgba[c])[qi] = temp; |
| 414 | } |
| 415 | break; |
| 416 | case TGSI_OPCODE_ATOMXCHG: |
| 417 | for (c = 0; c < nc; c++) { |
| 418 | unsigned temp = sdata[c]; |
| 419 | sdata[c] = ((uint32_t *)rgba[c])[qi]; |
| 420 | ((uint32_t *)rgba[c])[qi] = temp; |
| 421 | } |
| 422 | break; |
| 423 | case TGSI_OPCODE_ATOMCAS: |
| 424 | for (c = 0; c < nc; c++) { |
| 425 | unsigned dst_x = sdata[c]; |
| 426 | unsigned cmp_x = ((uint32_t *)rgba[c])[qi]; |
| 427 | unsigned src_x = ((uint32_t *)rgba2[c])[qi]; |
| 428 | unsigned temp = sdata[c]; |
| 429 | sdata[c] = (dst_x == cmp_x) ? src_x : dst_x; |
| 430 | ((uint32_t *)rgba[c])[qi] = temp; |
| 431 | } |
| 432 | break; |
| 433 | case TGSI_OPCODE_ATOMAND: |
| 434 | for (c = 0; c < nc; c++) { |
| 435 | unsigned temp = sdata[c]; |
| 436 | sdata[c] &= ((uint32_t *)rgba[c])[qi]; |
| 437 | ((uint32_t *)rgba[c])[qi] = temp; |
| 438 | } |
| 439 | break; |
| 440 | case TGSI_OPCODE_ATOMOR: |
| 441 | for (c = 0; c < nc; c++) { |
| 442 | unsigned temp = sdata[c]; |
| 443 | sdata[c] |= ((uint32_t *)rgba[c])[qi]; |
| 444 | ((uint32_t *)rgba[c])[qi] = temp; |
| 445 | } |
| 446 | break; |
| 447 | case TGSI_OPCODE_ATOMXOR: |
| 448 | for (c = 0; c < nc; c++) { |
| 449 | unsigned temp = sdata[c]; |
| 450 | sdata[c] ^= ((uint32_t *)rgba[c])[qi]; |
| 451 | ((uint32_t *)rgba[c])[qi] = temp; |
| 452 | } |
| 453 | break; |
| 454 | case TGSI_OPCODE_ATOMUMIN: |
| 455 | for (c = 0; c < nc; c++) { |
| 456 | unsigned dst_x = sdata[c]; |
| 457 | unsigned src_x = ((uint32_t *)rgba[c])[qi]; |
| 458 | sdata[c] = MIN2(dst_x, src_x); |
| 459 | ((uint32_t *)rgba[c])[qi] = dst_x; |
| 460 | } |
| 461 | break; |
| 462 | case TGSI_OPCODE_ATOMUMAX: |
| 463 | for (c = 0; c < nc; c++) { |
| 464 | unsigned dst_x = sdata[c]; |
| 465 | unsigned src_x = ((uint32_t *)rgba[c])[qi]; |
| 466 | sdata[c] = MAX2(dst_x, src_x); |
| 467 | ((uint32_t *)rgba[c])[qi] = dst_x; |
| 468 | } |
| 469 | break; |
| 470 | case TGSI_OPCODE_ATOMIMIN: |
| 471 | for (c = 0; c < nc; c++) { |
| 472 | int dst_x = sdata[c]; |
| 473 | int src_x = ((uint32_t *)rgba[c])[qi]; |
| 474 | sdata[c] = MIN2(dst_x, src_x); |
| 475 | ((uint32_t *)rgba[c])[qi] = dst_x; |
| 476 | } |
| 477 | break; |
| 478 | case TGSI_OPCODE_ATOMIMAX: |
| 479 | for (c = 0; c < nc; c++) { |
| 480 | int dst_x = sdata[c]; |
| 481 | int src_x = ((uint32_t *)rgba[c])[qi]; |
| 482 | sdata[c] = MAX2(dst_x, src_x); |
| 483 | ((uint32_t *)rgba[c])[qi] = dst_x; |
| 484 | } |
| 485 | break; |
| 486 | default: |
| 487 | assert(!"Unexpected TGSI opcode in sp_tgsi_op"); |
| 488 | break; |
| 489 | } |
| 490 | util_format_write_4ui(params->format, sdata, 0, data_ptr, stride, |
| 491 | s, t, 1, 1); |
| 492 | } |
| 493 | |
| 494 | /* |
| 495 | * Implement atomic operations on signed integers. |
| 496 | */ |
| 497 | static void |
| 498 | handle_op_int(const struct pipe_image_view *iview, |
| 499 | const struct tgsi_image_params *params, |
| 500 | bool just_read, |
| 501 | char *data_ptr, |
| 502 | uint qi, |
| 503 | unsigned stride, |
| 504 | unsigned opcode, |
| 505 | int s, |
| 506 | int t, |
| 507 | float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE], |
| 508 | float rgba2[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE]) |
| 509 | { |
| 510 | uint c; |
| 511 | int nc = util_format_get_nr_components(params->format); |
| 512 | int sdata[4]; |
| 513 | util_format_read_4i(params->format, |
| 514 | sdata, 0, |
| 515 | data_ptr, stride, |
| 516 | s, t, 1, 1); |
| 517 | |
| 518 | if (just_read) { |
| 519 | for (c = 0; c < nc; c++) { |
| 520 | ((int32_t *)rgba[c])[qi] = sdata[c]; |
| 521 | } |
| 522 | return; |
| 523 | } |
| 524 | switch (opcode) { |
| 525 | case TGSI_OPCODE_ATOMUADD: |
| 526 | for (c = 0; c < nc; c++) { |
| 527 | int temp = sdata[c]; |
| 528 | sdata[c] += ((int32_t *)rgba[c])[qi]; |
| 529 | ((int32_t *)rgba[c])[qi] = temp; |
| 530 | } |
| 531 | break; |
| 532 | case TGSI_OPCODE_ATOMXCHG: |
| 533 | for (c = 0; c < nc; c++) { |
| 534 | int temp = sdata[c]; |
| 535 | sdata[c] = ((int32_t *)rgba[c])[qi]; |
| 536 | ((int32_t *)rgba[c])[qi] = temp; |
| 537 | } |
| 538 | break; |
| 539 | case TGSI_OPCODE_ATOMCAS: |
| 540 | for (c = 0; c < nc; c++) { |
| 541 | int dst_x = sdata[c]; |
| 542 | int cmp_x = ((int32_t *)rgba[c])[qi]; |
| 543 | int src_x = ((int32_t *)rgba2[c])[qi]; |
| 544 | int temp = sdata[c]; |
| 545 | sdata[c] = (dst_x == cmp_x) ? src_x : dst_x; |
| 546 | ((int32_t *)rgba[c])[qi] = temp; |
| 547 | } |
| 548 | break; |
| 549 | case TGSI_OPCODE_ATOMAND: |
| 550 | for (c = 0; c < nc; c++) { |
| 551 | int temp = sdata[c]; |
| 552 | sdata[c] &= ((int32_t *)rgba[c])[qi]; |
| 553 | ((int32_t *)rgba[c])[qi] = temp; |
| 554 | } |
| 555 | break; |
| 556 | case TGSI_OPCODE_ATOMOR: |
| 557 | for (c = 0; c < nc; c++) { |
| 558 | int temp = sdata[c]; |
| 559 | sdata[c] |= ((int32_t *)rgba[c])[qi]; |
| 560 | ((int32_t *)rgba[c])[qi] = temp; |
| 561 | } |
| 562 | break; |
| 563 | case TGSI_OPCODE_ATOMXOR: |
| 564 | for (c = 0; c < nc; c++) { |
| 565 | int temp = sdata[c]; |
| 566 | sdata[c] ^= ((int32_t *)rgba[c])[qi]; |
| 567 | ((int32_t *)rgba[c])[qi] = temp; |
| 568 | } |
| 569 | break; |
| 570 | case TGSI_OPCODE_ATOMUMIN: |
| 571 | for (c = 0; c < nc; c++) { |
| 572 | int dst_x = sdata[c]; |
| 573 | int src_x = ((int32_t *)rgba[c])[qi]; |
| 574 | sdata[c] = MIN2(dst_x, src_x); |
| 575 | ((int32_t *)rgba[c])[qi] = dst_x; |
| 576 | } |
| 577 | break; |
| 578 | case TGSI_OPCODE_ATOMUMAX: |
| 579 | for (c = 0; c < nc; c++) { |
| 580 | int dst_x = sdata[c]; |
| 581 | int src_x = ((int32_t *)rgba[c])[qi]; |
| 582 | sdata[c] = MAX2(dst_x, src_x); |
| 583 | ((int32_t *)rgba[c])[qi] = dst_x; |
| 584 | } |
| 585 | break; |
| 586 | case TGSI_OPCODE_ATOMIMIN: |
| 587 | for (c = 0; c < nc; c++) { |
| 588 | int dst_x = sdata[c]; |
| 589 | int src_x = ((int32_t *)rgba[c])[qi]; |
| 590 | sdata[c] = MIN2(dst_x, src_x); |
| 591 | ((int32_t *)rgba[c])[qi] = dst_x; |
| 592 | } |
| 593 | break; |
| 594 | case TGSI_OPCODE_ATOMIMAX: |
| 595 | for (c = 0; c < nc; c++) { |
| 596 | int dst_x = sdata[c]; |
| 597 | int src_x = ((int32_t *)rgba[c])[qi]; |
| 598 | sdata[c] = MAX2(dst_x, src_x); |
| 599 | ((int32_t *)rgba[c])[qi] = dst_x; |
| 600 | } |
| 601 | break; |
| 602 | default: |
| 603 | assert(!"Unexpected TGSI opcode in sp_tgsi_op"); |
| 604 | break; |
| 605 | } |
| 606 | util_format_write_4i(params->format, sdata, 0, data_ptr, stride, |
| 607 | s, t, 1, 1); |
| 608 | } |
| 609 | |
| 610 | /* |
| 611 | * Implement atomic image operations. |
| 612 | */ |
| 613 | static void |
| 614 | sp_tgsi_op(const struct tgsi_image *image, |
| 615 | const struct tgsi_image_params *params, |
| 616 | unsigned opcode, |
| 617 | const int s[TGSI_QUAD_SIZE], |
| 618 | const int t[TGSI_QUAD_SIZE], |
| 619 | const int r[TGSI_QUAD_SIZE], |
| 620 | const int sample[TGSI_QUAD_SIZE], |
| 621 | float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE], |
| 622 | float rgba2[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE]) |
| 623 | { |
| 624 | struct sp_tgsi_image *sp_img = (struct sp_tgsi_image *)image; |
| 625 | struct pipe_image_view *iview; |
| 626 | struct softpipe_resource *spr; |
| 627 | unsigned width, height, depth; |
| 628 | unsigned stride; |
| 629 | int j, c; |
| 630 | unsigned offset; |
| 631 | char *data_ptr; |
| 632 | |
| 633 | if (params->unit > PIPE_MAX_SHADER_IMAGES) |
| 634 | return; |
| 635 | iview = &sp_img->sp_iview[params->unit]; |
| 636 | spr = (struct softpipe_resource *)iview->resource; |
| 637 | if (!spr) |
| 638 | goto fail_write_all_zero; |
| 639 | if (!has_compat_target(spr->base.target, params->tgsi_tex_instr)) |
| 640 | goto fail_write_all_zero; |
| 641 | |
| 642 | if (!get_dimensions(iview, spr, params->tgsi_tex_instr, |
| 643 | params->format, &width, &height, &depth)) |
| 644 | goto fail_write_all_zero; |
| 645 | |
| 646 | stride = util_format_get_stride(spr->base.format, width); |
| 647 | |
| 648 | for (j = 0; j < TGSI_QUAD_SIZE; j++) { |
| 649 | int s_coord, t_coord, r_coord; |
| 650 | bool just_read = false; |
| 651 | |
| 652 | fill_coords(params, j, s, t, r, &s_coord, &t_coord, &r_coord); |
| 653 | if (!bounds_check(width, height, depth, |
| 654 | s_coord, t_coord, r_coord)) { |
| 655 | int nc = util_format_get_nr_components(params->format); |
| 656 | int ival = util_format_is_pure_integer(params->format); |
| 657 | int c; |
| 658 | for (c = 0; c < 4; c++) { |
| 659 | rgba[c][j] = 0; |
| 660 | if (c == 3 && nc < 4) { |
| 661 | if (ival) |
| 662 | ((int32_t *)rgba[c])[j] = 1; |
| 663 | else |
| 664 | rgba[c][j] = 1.0; |
| 665 | } |
| 666 | } |
| 667 | continue; |
| 668 | } |
| 669 | |
| 670 | /* just readback the value for atomic if execmask isn't set */ |
| 671 | if (!(params->execmask & (1 << j))) { |
| 672 | just_read = true; |
| 673 | } |
| 674 | |
| 675 | offset = get_image_offset(spr, iview, params->format, r_coord); |
| 676 | data_ptr = (char *)spr->data + offset; |
| 677 | |
| 678 | /* we should see atomic operations on r32 formats */ |
| 679 | if (util_format_is_pure_uint(params->format)) |
| 680 | handle_op_uint(iview, params, just_read, data_ptr, j, stride, |
| 681 | opcode, s_coord, t_coord, rgba, rgba2); |
| 682 | else if (util_format_is_pure_sint(params->format)) |
| 683 | handle_op_int(iview, params, just_read, data_ptr, j, stride, |
| 684 | opcode, s_coord, t_coord, rgba, rgba2); |
| 685 | else |
| 686 | assert(0); |
| 687 | } |
| 688 | return; |
| 689 | fail_write_all_zero: |
| 690 | for (j = 0; j < TGSI_QUAD_SIZE; j++) { |
| 691 | for (c = 0; c < 4; c++) |
| 692 | rgba[c][j] = 0; |
| 693 | } |
| 694 | return; |
| 695 | } |
| 696 | |
| 697 | static void |
| 698 | sp_tgsi_get_dims(const struct tgsi_image *image, |
| 699 | const struct tgsi_image_params *params, |
| 700 | int dims[4]) |
| 701 | { |
| 702 | struct sp_tgsi_image *sp_img = (struct sp_tgsi_image *)image; |
| 703 | struct pipe_image_view *iview; |
| 704 | struct softpipe_resource *spr; |
| 705 | int level; |
| 706 | |
| 707 | if (params->unit > PIPE_MAX_SHADER_IMAGES) |
| 708 | return; |
| 709 | iview = &sp_img->sp_iview[params->unit]; |
| 710 | spr = (struct softpipe_resource *)iview->resource; |
| 711 | if (!spr) |
| 712 | return; |
| 713 | |
| 714 | if (params->tgsi_tex_instr == TGSI_TEXTURE_BUFFER) { |
| 715 | dims[0] = iview->u.buf.last_element - iview->u.buf.first_element + 1; |
| 716 | dims[1] = dims[2] = dims[3] = 0; |
| 717 | return; |
| 718 | } |
| 719 | |
| 720 | level = iview->u.tex.level; |
| 721 | dims[0] = u_minify(spr->base.width0, level); |
| 722 | switch (params->tgsi_tex_instr) { |
| 723 | case TGSI_TEXTURE_1D_ARRAY: |
| 724 | dims[1] = iview->u.tex.last_layer - iview->u.tex.first_layer + 1; |
| 725 | /* fallthrough */ |
| 726 | case TGSI_TEXTURE_1D: |
| 727 | return; |
| 728 | case TGSI_TEXTURE_2D_ARRAY: |
| 729 | dims[2] = iview->u.tex.last_layer - iview->u.tex.first_layer + 1; |
| 730 | /* fallthrough */ |
| 731 | case TGSI_TEXTURE_2D: |
| 732 | case TGSI_TEXTURE_CUBE: |
| 733 | case TGSI_TEXTURE_RECT: |
| 734 | dims[1] = u_minify(spr->base.height0, level); |
| 735 | return; |
| 736 | case TGSI_TEXTURE_3D: |
| 737 | dims[1] = u_minify(spr->base.height0, level); |
| 738 | dims[2] = u_minify(spr->base.depth0, level); |
| 739 | return; |
| 740 | case TGSI_TEXTURE_CUBE_ARRAY: |
| 741 | dims[1] = u_minify(spr->base.height0, level); |
| 742 | dims[2] = (iview->u.tex.last_layer - iview->u.tex.first_layer + 1) / 6; |
| 743 | break; |
| 744 | default: |
| 745 | assert(!"unexpected texture target in sp_get_dims()"); |
| 746 | return; |
| 747 | } |
| 748 | } |
| 749 | |
| 750 | struct sp_tgsi_image * |
| 751 | sp_create_tgsi_image(void) |
| 752 | { |
| 753 | struct sp_tgsi_image *img = CALLOC_STRUCT(sp_tgsi_image); |
| 754 | if (!img) |
| 755 | return NULL; |
| 756 | |
| 757 | img->base.load = sp_tgsi_load; |
| 758 | img->base.store = sp_tgsi_store; |
| 759 | img->base.op = sp_tgsi_op; |
| 760 | img->base.get_dims = sp_tgsi_get_dims; |
| 761 | return img; |
| 762 | }; |