1 /* 2 * Copyright 2017 Advanced Micro Devices, Inc. 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 shall be included in 12 * all copies or substantial portions of the Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR 18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 20 * OTHER DEALINGS IN THE SOFTWARE. 21 * 22 * Authors: AMD 23 * 24 */ 25 26 27 #include "../display_mode_lib.h" 28 #include "../display_mode_vba.h" 29 #include "../dml_inline_defs.h" 30 #include "display_rq_dlg_calc_21.h" 31 32 /* 33 * NOTE: 34 * This file is gcc-parseable HW gospel, coming straight from HW engineers. 35 * 36 * It doesn't adhere to Linux kernel style and sometimes will do things in odd 37 * ways. Unless there is something clearly wrong with it the code should 38 * remain as-is as it provides us with a guarantee from HW that it is correct. 39 */ 40 41 static void calculate_ttu_cursor( 42 struct display_mode_lib *mode_lib, 43 double *refcyc_per_req_delivery_pre_cur, 44 double *refcyc_per_req_delivery_cur, 45 double refclk_freq_in_mhz, 46 double ref_freq_to_pix_freq, 47 double hscale_pixel_rate_l, 48 double hscl_ratio, 49 double vratio_pre_l, 50 double vratio_l, 51 unsigned int cur_width, 52 enum cursor_bpp cur_bpp); 53 54 static unsigned int get_bytes_per_element(enum source_format_class source_format, bool is_chroma) 55 { 56 unsigned int ret_val = 0; 57 58 if (source_format == dm_444_16) { 59 if (!is_chroma) 60 ret_val = 2; 61 } else if (source_format == dm_444_32) { 62 if (!is_chroma) 63 ret_val = 4; 64 } else if (source_format == dm_444_64) { 65 if (!is_chroma) 66 ret_val = 8; 67 } else if (source_format == dm_420_8) { 68 if (is_chroma) 69 ret_val = 2; 70 else 71 ret_val = 1; 72 } else if (source_format == dm_420_10) { 73 if (is_chroma) 74 ret_val = 4; 75 else 76 ret_val = 2; 77 } else if (source_format == dm_444_8) { 78 ret_val = 1; 79 } 80 return ret_val; 81 } 82 83 static bool is_dual_plane(enum source_format_class source_format) 84 { 85 bool ret_val = false; 86 87 if ((source_format == dm_420_8) || (source_format == dm_420_10)) 88 ret_val = true; 89 90 return ret_val; 91 } 92 93 static double get_refcyc_per_delivery( 94 struct display_mode_lib *mode_lib, 95 double refclk_freq_in_mhz, 96 double pclk_freq_in_mhz, 97 bool odm_combine, 98 unsigned int recout_width, 99 unsigned int hactive, 100 double vratio, 101 double hscale_pixel_rate, 102 unsigned int delivery_width, 103 unsigned int req_per_swath_ub) 104 { 105 double refcyc_per_delivery = 0.0; 106 107 if (vratio <= 1.0) { 108 if (odm_combine) 109 refcyc_per_delivery = (double) refclk_freq_in_mhz 110 * dml_min((double) recout_width, (double) hactive / 2.0) 111 / pclk_freq_in_mhz / (double) req_per_swath_ub; 112 else 113 refcyc_per_delivery = (double) refclk_freq_in_mhz * (double) recout_width 114 / pclk_freq_in_mhz / (double) req_per_swath_ub; 115 } else { 116 refcyc_per_delivery = (double) refclk_freq_in_mhz * (double) delivery_width 117 / (double) hscale_pixel_rate / (double) req_per_swath_ub; 118 } 119 120 dml_print("DML_DLG: %s: refclk_freq_in_mhz = %3.2f\n", __func__, refclk_freq_in_mhz); 121 dml_print("DML_DLG: %s: pclk_freq_in_mhz = %3.2f\n", __func__, pclk_freq_in_mhz); 122 dml_print("DML_DLG: %s: recout_width = %d\n", __func__, recout_width); 123 dml_print("DML_DLG: %s: vratio = %3.2f\n", __func__, vratio); 124 dml_print("DML_DLG: %s: req_per_swath_ub = %d\n", __func__, req_per_swath_ub); 125 dml_print("DML_DLG: %s: refcyc_per_delivery= %3.2f\n", __func__, refcyc_per_delivery); 126 127 return refcyc_per_delivery; 128 129 } 130 131 static unsigned int get_blk_size_bytes(const enum source_macro_tile_size tile_size) 132 { 133 if (tile_size == dm_256k_tile) 134 return (256 * 1024); 135 else if (tile_size == dm_64k_tile) 136 return (64 * 1024); 137 else 138 return (4 * 1024); 139 } 140 141 static void extract_rq_sizing_regs( 142 struct display_mode_lib *mode_lib, 143 display_data_rq_regs_st *rq_regs, 144 const display_data_rq_sizing_params_st *rq_sizing) 145 { 146 dml_print("DML_DLG: %s: rq_sizing param\n", __func__); 147 print__data_rq_sizing_params_st(mode_lib, rq_sizing); 148 149 rq_regs->chunk_size = dml_log2(rq_sizing->chunk_bytes) - 10; 150 151 if (rq_sizing->min_chunk_bytes == 0) 152 rq_regs->min_chunk_size = 0; 153 else 154 rq_regs->min_chunk_size = dml_log2(rq_sizing->min_chunk_bytes) - 8 + 1; 155 156 rq_regs->meta_chunk_size = dml_log2(rq_sizing->meta_chunk_bytes) - 10; 157 if (rq_sizing->min_meta_chunk_bytes == 0) 158 rq_regs->min_meta_chunk_size = 0; 159 else 160 rq_regs->min_meta_chunk_size = dml_log2(rq_sizing->min_meta_chunk_bytes) - 6 + 1; 161 162 rq_regs->dpte_group_size = dml_log2(rq_sizing->dpte_group_bytes) - 6; 163 rq_regs->mpte_group_size = dml_log2(rq_sizing->mpte_group_bytes) - 6; 164 } 165 166 static void extract_rq_regs( 167 struct display_mode_lib *mode_lib, 168 display_rq_regs_st *rq_regs, 169 const display_rq_params_st *rq_param) 170 { 171 unsigned int detile_buf_size_in_bytes = mode_lib->ip.det_buffer_size_kbytes * 1024; 172 unsigned int detile_buf_plane1_addr = 0; 173 174 extract_rq_sizing_regs(mode_lib, &(rq_regs->rq_regs_l), &rq_param->sizing.rq_l); 175 176 rq_regs->rq_regs_l.pte_row_height_linear = dml_floor( 177 dml_log2(rq_param->dlg.rq_l.dpte_row_height), 178 1) - 3; 179 180 if (rq_param->yuv420) { 181 extract_rq_sizing_regs(mode_lib, &(rq_regs->rq_regs_c), &rq_param->sizing.rq_c); 182 rq_regs->rq_regs_c.pte_row_height_linear = dml_floor( 183 dml_log2(rq_param->dlg.rq_c.dpte_row_height), 184 1) - 3; 185 } 186 187 rq_regs->rq_regs_l.swath_height = dml_log2(rq_param->dlg.rq_l.swath_height); 188 rq_regs->rq_regs_c.swath_height = dml_log2(rq_param->dlg.rq_c.swath_height); 189 190 // FIXME: take the max between luma, chroma chunk size? 191 // okay for now, as we are setting chunk_bytes to 8kb anyways 192 if (rq_param->sizing.rq_l.chunk_bytes >= 32 * 1024) { //32kb 193 rq_regs->drq_expansion_mode = 0; 194 } else { 195 rq_regs->drq_expansion_mode = 2; 196 } 197 rq_regs->prq_expansion_mode = 1; 198 rq_regs->mrq_expansion_mode = 1; 199 rq_regs->crq_expansion_mode = 1; 200 201 if (rq_param->yuv420) { 202 if ((double) rq_param->misc.rq_l.stored_swath_bytes 203 / (double) rq_param->misc.rq_c.stored_swath_bytes <= 1.5) { 204 detile_buf_plane1_addr = (detile_buf_size_in_bytes / 2.0 / 64.0); // half to chroma 205 } else { 206 detile_buf_plane1_addr = dml_round_to_multiple( 207 (unsigned int) ((2.0 * detile_buf_size_in_bytes) / 3.0), 208 256, 209 0) / 64.0; // 2/3 to chroma 210 } 211 } 212 rq_regs->plane1_base_address = detile_buf_plane1_addr; 213 } 214 215 static void handle_det_buf_split( 216 struct display_mode_lib *mode_lib, 217 display_rq_params_st *rq_param, 218 const display_pipe_source_params_st *pipe_src_param) 219 { 220 unsigned int total_swath_bytes = 0; 221 unsigned int swath_bytes_l = 0; 222 unsigned int swath_bytes_c = 0; 223 unsigned int full_swath_bytes_packed_l = 0; 224 unsigned int full_swath_bytes_packed_c = 0; 225 bool req128_l = false; 226 bool req128_c = false; 227 bool surf_linear = (pipe_src_param->sw_mode == dm_sw_linear); 228 bool surf_vert = (pipe_src_param->source_scan == dm_vert); 229 unsigned int log2_swath_height_l = 0; 230 unsigned int log2_swath_height_c = 0; 231 unsigned int detile_buf_size_in_bytes = mode_lib->ip.det_buffer_size_kbytes * 1024; 232 233 full_swath_bytes_packed_l = rq_param->misc.rq_l.full_swath_bytes; 234 full_swath_bytes_packed_c = rq_param->misc.rq_c.full_swath_bytes; 235 236 if (rq_param->yuv420_10bpc) { 237 full_swath_bytes_packed_l = dml_round_to_multiple( 238 rq_param->misc.rq_l.full_swath_bytes * 2 / 3, 239 256, 240 1) + 256; 241 full_swath_bytes_packed_c = dml_round_to_multiple( 242 rq_param->misc.rq_c.full_swath_bytes * 2 / 3, 243 256, 244 1) + 256; 245 } 246 247 if (rq_param->yuv420) { 248 total_swath_bytes = 2 * full_swath_bytes_packed_l + 2 * full_swath_bytes_packed_c; 249 250 if (total_swath_bytes <= detile_buf_size_in_bytes) { //full 256b request 251 req128_l = false; 252 req128_c = false; 253 swath_bytes_l = full_swath_bytes_packed_l; 254 swath_bytes_c = full_swath_bytes_packed_c; 255 } else { //128b request (for luma only for yuv420 8bpc) 256 req128_l = true; 257 req128_c = false; 258 swath_bytes_l = full_swath_bytes_packed_l / 2; 259 swath_bytes_c = full_swath_bytes_packed_c; 260 } 261 // Note: assumption, the config that pass in will fit into 262 // the detiled buffer. 263 } else { 264 total_swath_bytes = 2 * full_swath_bytes_packed_l; 265 266 if (total_swath_bytes <= detile_buf_size_in_bytes) 267 req128_l = false; 268 else 269 req128_l = true; 270 271 swath_bytes_l = total_swath_bytes; 272 swath_bytes_c = 0; 273 } 274 rq_param->misc.rq_l.stored_swath_bytes = swath_bytes_l; 275 rq_param->misc.rq_c.stored_swath_bytes = swath_bytes_c; 276 277 if (surf_linear) { 278 log2_swath_height_l = 0; 279 log2_swath_height_c = 0; 280 } else { 281 unsigned int swath_height_l; 282 unsigned int swath_height_c; 283 284 if (!surf_vert) { 285 swath_height_l = rq_param->misc.rq_l.blk256_height; 286 swath_height_c = rq_param->misc.rq_c.blk256_height; 287 } else { 288 swath_height_l = rq_param->misc.rq_l.blk256_width; 289 swath_height_c = rq_param->misc.rq_c.blk256_width; 290 } 291 292 if (swath_height_l > 0) 293 log2_swath_height_l = dml_log2(swath_height_l); 294 295 if (req128_l && log2_swath_height_l > 0) 296 log2_swath_height_l -= 1; 297 298 if (swath_height_c > 0) 299 log2_swath_height_c = dml_log2(swath_height_c); 300 301 if (req128_c && log2_swath_height_c > 0) 302 log2_swath_height_c -= 1; 303 } 304 305 rq_param->dlg.rq_l.swath_height = 1 << log2_swath_height_l; 306 rq_param->dlg.rq_c.swath_height = 1 << log2_swath_height_c; 307 308 dml_print("DML_DLG: %s: req128_l = %0d\n", __func__, req128_l); 309 dml_print("DML_DLG: %s: req128_c = %0d\n", __func__, req128_c); 310 dml_print( 311 "DML_DLG: %s: full_swath_bytes_packed_l = %0d\n", 312 __func__, 313 full_swath_bytes_packed_l); 314 dml_print( 315 "DML_DLG: %s: full_swath_bytes_packed_c = %0d\n", 316 __func__, 317 full_swath_bytes_packed_c); 318 } 319 320 static void get_meta_and_pte_attr( 321 struct display_mode_lib *mode_lib, 322 display_data_rq_dlg_params_st *rq_dlg_param, 323 display_data_rq_misc_params_st *rq_misc_param, 324 display_data_rq_sizing_params_st *rq_sizing_param, 325 unsigned int vp_width, 326 unsigned int vp_height, 327 unsigned int data_pitch, 328 unsigned int meta_pitch, 329 unsigned int source_format, 330 unsigned int tiling, 331 unsigned int macro_tile_size, 332 unsigned int source_scan, 333 unsigned int hostvm_enable, 334 unsigned int is_chroma) 335 { 336 bool surf_linear = (tiling == dm_sw_linear); 337 bool surf_vert = (source_scan == dm_vert); 338 339 unsigned int bytes_per_element; 340 unsigned int bytes_per_element_y = get_bytes_per_element( 341 (enum source_format_class) (source_format), 342 false); 343 unsigned int bytes_per_element_c = get_bytes_per_element( 344 (enum source_format_class) (source_format), 345 true); 346 347 unsigned int blk256_width = 0; 348 unsigned int blk256_height = 0; 349 350 unsigned int blk256_width_y = 0; 351 unsigned int blk256_height_y = 0; 352 unsigned int blk256_width_c = 0; 353 unsigned int blk256_height_c = 0; 354 unsigned int log2_bytes_per_element; 355 unsigned int log2_blk256_width; 356 unsigned int log2_blk256_height; 357 unsigned int blk_bytes; 358 unsigned int log2_blk_bytes; 359 unsigned int log2_blk_height; 360 unsigned int log2_blk_width; 361 unsigned int log2_meta_req_bytes; 362 unsigned int log2_meta_req_height; 363 unsigned int log2_meta_req_width; 364 unsigned int meta_req_width; 365 unsigned int meta_req_height; 366 unsigned int log2_meta_row_height; 367 unsigned int meta_row_width_ub; 368 unsigned int log2_meta_chunk_bytes; 369 unsigned int log2_meta_chunk_height; 370 371 //full sized meta chunk width in unit of data elements 372 unsigned int log2_meta_chunk_width; 373 unsigned int log2_min_meta_chunk_bytes; 374 unsigned int min_meta_chunk_width; 375 unsigned int meta_chunk_width; 376 unsigned int meta_chunk_per_row_int; 377 unsigned int meta_row_remainder; 378 unsigned int meta_chunk_threshold; 379 unsigned int meta_blk_bytes; 380 unsigned int meta_blk_height; 381 unsigned int meta_blk_width; 382 unsigned int meta_surface_bytes; 383 unsigned int vmpg_bytes; 384 unsigned int meta_pte_req_per_frame_ub; 385 unsigned int meta_pte_bytes_per_frame_ub; 386 const unsigned int log2_vmpg_bytes = dml_log2(mode_lib->soc.vmm_page_size_bytes); 387 const unsigned int dpte_buf_in_pte_reqs = 388 mode_lib->ip.dpte_buffer_size_in_pte_reqs_luma + mode_lib->ip.dpte_buffer_size_in_pte_reqs_chroma; 389 const unsigned int pde_proc_buffer_size_64k_reqs = 390 mode_lib->ip.pde_proc_buffer_size_64k_reqs; 391 392 unsigned int log2_vmpg_height = 0; 393 unsigned int log2_vmpg_width = 0; 394 unsigned int log2_dpte_req_height_ptes = 0; 395 unsigned int log2_dpte_req_height = 0; 396 unsigned int log2_dpte_req_width = 0; 397 unsigned int log2_dpte_row_height_linear = 0; 398 unsigned int log2_dpte_row_height = 0; 399 unsigned int log2_dpte_group_width = 0; 400 unsigned int dpte_row_width_ub = 0; 401 unsigned int dpte_req_height = 0; 402 unsigned int dpte_req_width = 0; 403 unsigned int dpte_group_width = 0; 404 unsigned int log2_dpte_group_bytes = 0; 405 unsigned int log2_dpte_group_length = 0; 406 unsigned int pde_buf_entries; 407 bool yuv420 = (source_format == dm_420_8 || source_format == dm_420_10); 408 409 Calculate256BBlockSizes( 410 (enum source_format_class) (source_format), 411 (enum dm_swizzle_mode) (tiling), 412 bytes_per_element_y, 413 bytes_per_element_c, 414 &blk256_height_y, 415 &blk256_height_c, 416 &blk256_width_y, 417 &blk256_width_c); 418 419 if (!is_chroma) { 420 blk256_width = blk256_width_y; 421 blk256_height = blk256_height_y; 422 bytes_per_element = bytes_per_element_y; 423 } else { 424 blk256_width = blk256_width_c; 425 blk256_height = blk256_height_c; 426 bytes_per_element = bytes_per_element_c; 427 } 428 429 log2_bytes_per_element = dml_log2(bytes_per_element); 430 431 dml_print("DML_DLG: %s: surf_linear = %d\n", __func__, surf_linear); 432 dml_print("DML_DLG: %s: surf_vert = %d\n", __func__, surf_vert); 433 dml_print("DML_DLG: %s: blk256_width = %d\n", __func__, blk256_width); 434 dml_print("DML_DLG: %s: blk256_height = %d\n", __func__, blk256_height); 435 436 log2_blk256_width = dml_log2((double) blk256_width); 437 log2_blk256_height = dml_log2((double) blk256_height); 438 blk_bytes = surf_linear ? 439 256 : get_blk_size_bytes((enum source_macro_tile_size) macro_tile_size); 440 log2_blk_bytes = dml_log2((double) blk_bytes); 441 log2_blk_height = 0; 442 log2_blk_width = 0; 443 444 // remember log rule 445 // "+" in log is multiply 446 // "-" in log is divide 447 // "/2" is like square root 448 // blk is vertical biased 449 if (tiling != dm_sw_linear) 450 log2_blk_height = log2_blk256_height 451 + dml_ceil((double) (log2_blk_bytes - 8) / 2.0, 1); 452 else 453 log2_blk_height = 0; // blk height of 1 454 455 log2_blk_width = log2_blk_bytes - log2_bytes_per_element - log2_blk_height; 456 457 if (!surf_vert) { 458 rq_dlg_param->swath_width_ub = dml_round_to_multiple(vp_width - 1, blk256_width, 1) 459 + blk256_width; 460 rq_dlg_param->req_per_swath_ub = rq_dlg_param->swath_width_ub >> log2_blk256_width; 461 } else { 462 rq_dlg_param->swath_width_ub = dml_round_to_multiple( 463 vp_height - 1, 464 blk256_height, 465 1) + blk256_height; 466 rq_dlg_param->req_per_swath_ub = rq_dlg_param->swath_width_ub >> log2_blk256_height; 467 } 468 469 if (!surf_vert) 470 rq_misc_param->full_swath_bytes = rq_dlg_param->swath_width_ub * blk256_height 471 * bytes_per_element; 472 else 473 rq_misc_param->full_swath_bytes = rq_dlg_param->swath_width_ub * blk256_width 474 * bytes_per_element; 475 476 rq_misc_param->blk256_height = blk256_height; 477 rq_misc_param->blk256_width = blk256_width; 478 479 // ------- 480 // meta 481 // ------- 482 log2_meta_req_bytes = 6; // meta request is 64b and is 8x8byte meta element 483 484 // each 64b meta request for dcn is 8x8 meta elements and 485 // a meta element covers one 256b block of the data surface. 486 log2_meta_req_height = log2_blk256_height + 3; // meta req is 8x8 byte, each byte represent 1 blk256 487 log2_meta_req_width = log2_meta_req_bytes + 8 - log2_bytes_per_element 488 - log2_meta_req_height; 489 meta_req_width = 1 << log2_meta_req_width; 490 meta_req_height = 1 << log2_meta_req_height; 491 log2_meta_row_height = 0; 492 meta_row_width_ub = 0; 493 494 // the dimensions of a meta row are meta_row_width x meta_row_height in elements. 495 // calculate upper bound of the meta_row_width 496 if (!surf_vert) { 497 log2_meta_row_height = log2_meta_req_height; 498 meta_row_width_ub = dml_round_to_multiple(vp_width - 1, meta_req_width, 1) 499 + meta_req_width; 500 rq_dlg_param->meta_req_per_row_ub = meta_row_width_ub / meta_req_width; 501 } else { 502 log2_meta_row_height = log2_meta_req_width; 503 meta_row_width_ub = dml_round_to_multiple(vp_height - 1, meta_req_height, 1) 504 + meta_req_height; 505 rq_dlg_param->meta_req_per_row_ub = meta_row_width_ub / meta_req_height; 506 } 507 rq_dlg_param->meta_bytes_per_row_ub = rq_dlg_param->meta_req_per_row_ub * 64; 508 509 rq_dlg_param->meta_row_height = 1 << log2_meta_row_height; 510 511 log2_meta_chunk_bytes = dml_log2(rq_sizing_param->meta_chunk_bytes); 512 log2_meta_chunk_height = log2_meta_row_height; 513 514 //full sized meta chunk width in unit of data elements 515 log2_meta_chunk_width = log2_meta_chunk_bytes + 8 - log2_bytes_per_element 516 - log2_meta_chunk_height; 517 log2_min_meta_chunk_bytes = dml_log2(rq_sizing_param->min_meta_chunk_bytes); 518 min_meta_chunk_width = 1 519 << (log2_min_meta_chunk_bytes + 8 - log2_bytes_per_element 520 - log2_meta_chunk_height); 521 meta_chunk_width = 1 << log2_meta_chunk_width; 522 meta_chunk_per_row_int = (unsigned int) (meta_row_width_ub / meta_chunk_width); 523 meta_row_remainder = meta_row_width_ub % meta_chunk_width; 524 meta_chunk_threshold = 0; 525 meta_blk_bytes = 4096; 526 meta_blk_height = blk256_height * 64; 527 meta_blk_width = meta_blk_bytes * 256 / bytes_per_element / meta_blk_height; 528 meta_surface_bytes = meta_pitch 529 * (dml_round_to_multiple(vp_height - 1, meta_blk_height, 1) 530 + meta_blk_height) * bytes_per_element / 256; 531 vmpg_bytes = mode_lib->soc.vmm_page_size_bytes; 532 meta_pte_req_per_frame_ub = (dml_round_to_multiple( 533 meta_surface_bytes - vmpg_bytes, 534 8 * vmpg_bytes, 535 1) + 8 * vmpg_bytes) / (8 * vmpg_bytes); 536 meta_pte_bytes_per_frame_ub = meta_pte_req_per_frame_ub * 64; //64B mpte request 537 rq_dlg_param->meta_pte_bytes_per_frame_ub = meta_pte_bytes_per_frame_ub; 538 539 dml_print("DML_DLG: %s: meta_blk_height = %d\n", __func__, meta_blk_height); 540 dml_print("DML_DLG: %s: meta_blk_width = %d\n", __func__, meta_blk_width); 541 dml_print("DML_DLG: %s: meta_surface_bytes = %d\n", __func__, meta_surface_bytes); 542 dml_print( 543 "DML_DLG: %s: meta_pte_req_per_frame_ub = %d\n", 544 __func__, 545 meta_pte_req_per_frame_ub); 546 dml_print( 547 "DML_DLG: %s: meta_pte_bytes_per_frame_ub = %d\n", 548 __func__, 549 meta_pte_bytes_per_frame_ub); 550 551 if (!surf_vert) 552 meta_chunk_threshold = 2 * min_meta_chunk_width - meta_req_width; 553 else 554 meta_chunk_threshold = 2 * min_meta_chunk_width - meta_req_height; 555 556 if (meta_row_remainder <= meta_chunk_threshold) 557 rq_dlg_param->meta_chunks_per_row_ub = meta_chunk_per_row_int + 1; 558 else 559 rq_dlg_param->meta_chunks_per_row_ub = meta_chunk_per_row_int + 2; 560 561 // ------ 562 // dpte 563 // ------ 564 if (surf_linear) { 565 log2_vmpg_height = 0; // one line high 566 } else { 567 log2_vmpg_height = (log2_vmpg_bytes - 8) / 2 + log2_blk256_height; 568 } 569 log2_vmpg_width = log2_vmpg_bytes - log2_bytes_per_element - log2_vmpg_height; 570 571 // only 3 possible shapes for dpte request in dimensions of ptes: 8x1, 4x2, 2x4. 572 if (surf_linear) { //one 64B PTE request returns 8 PTEs 573 log2_dpte_req_height_ptes = 0; 574 log2_dpte_req_width = log2_vmpg_width + 3; 575 log2_dpte_req_height = 0; 576 } else if (log2_blk_bytes == 12) { //4KB tile means 4kB page size 577 //one 64B req gives 8x1 PTEs for 4KB tile 578 log2_dpte_req_height_ptes = 0; 579 log2_dpte_req_width = log2_blk_width + 3; 580 log2_dpte_req_height = log2_blk_height + 0; 581 } else if ((log2_blk_bytes >= 16) && (log2_vmpg_bytes == 12)) { // tile block >= 64KB 582 //two 64B reqs of 2x4 PTEs give 16 PTEs to cover 64KB 583 log2_dpte_req_height_ptes = 4; 584 log2_dpte_req_width = log2_blk256_width + 4; // log2_64KB_width 585 log2_dpte_req_height = log2_blk256_height + 4; // log2_64KB_height 586 } else { //64KB page size and must 64KB tile block 587 //one 64B req gives 8x1 PTEs for 64KB tile 588 log2_dpte_req_height_ptes = 0; 589 log2_dpte_req_width = log2_blk_width + 3; 590 log2_dpte_req_height = log2_blk_height + 0; 591 } 592 593 // The dpte request dimensions in data elements is dpte_req_width x dpte_req_height 594 // log2_vmpg_width is how much 1 pte represent, now calculating how much a 64b pte req represent 595 // That depends on the pte shape (i.e. 8x1, 4x2, 2x4) 596 //log2_dpte_req_height = log2_vmpg_height + log2_dpte_req_height_ptes; 597 //log2_dpte_req_width = log2_vmpg_width + log2_dpte_req_width_ptes; 598 dpte_req_height = 1 << log2_dpte_req_height; 599 dpte_req_width = 1 << log2_dpte_req_width; 600 601 // calculate pitch dpte row buffer can hold 602 // round the result down to a power of two. 603 pde_buf_entries = 604 yuv420 ? (pde_proc_buffer_size_64k_reqs >> 1) : pde_proc_buffer_size_64k_reqs; 605 if (surf_linear) { 606 unsigned int dpte_row_height; 607 608 log2_dpte_row_height_linear = dml_floor( 609 dml_log2( 610 dml_min( 611 64 * 1024 * pde_buf_entries 612 / bytes_per_element, 613 dpte_buf_in_pte_reqs 614 * dpte_req_width) 615 / data_pitch), 616 1); 617 618 ASSERT(log2_dpte_row_height_linear >= 3); 619 620 if (log2_dpte_row_height_linear > 7) 621 log2_dpte_row_height_linear = 7; 622 623 log2_dpte_row_height = log2_dpte_row_height_linear; 624 // For linear, the dpte row is pitch dependent and the pte requests wrap at the pitch boundary. 625 // the dpte_row_width_ub is the upper bound of data_pitch*dpte_row_height in elements with this unique buffering. 626 dpte_row_height = 1 << log2_dpte_row_height; 627 dpte_row_width_ub = dml_round_to_multiple( 628 data_pitch * dpte_row_height - 1, 629 dpte_req_width, 630 1) + dpte_req_width; 631 rq_dlg_param->dpte_req_per_row_ub = dpte_row_width_ub / dpte_req_width; 632 } else { 633 // the upper bound of the dpte_row_width without dependency on viewport position follows. 634 // for tiled mode, row height is the same as req height and row store up to vp size upper bound 635 if (!surf_vert) { 636 log2_dpte_row_height = log2_dpte_req_height; 637 dpte_row_width_ub = dml_round_to_multiple(vp_width - 1, dpte_req_width, 1) 638 + dpte_req_width; 639 rq_dlg_param->dpte_req_per_row_ub = dpte_row_width_ub / dpte_req_width; 640 } else { 641 log2_dpte_row_height = 642 (log2_blk_width < log2_dpte_req_width) ? 643 log2_blk_width : log2_dpte_req_width; 644 dpte_row_width_ub = dml_round_to_multiple(vp_height - 1, dpte_req_height, 1) 645 + dpte_req_height; 646 rq_dlg_param->dpte_req_per_row_ub = dpte_row_width_ub / dpte_req_height; 647 } 648 } 649 if (log2_blk_bytes >= 16 && log2_vmpg_bytes == 12) // tile block >= 64KB 650 rq_dlg_param->dpte_bytes_per_row_ub = rq_dlg_param->dpte_req_per_row_ub * 128; //2*64B dpte request 651 else 652 rq_dlg_param->dpte_bytes_per_row_ub = rq_dlg_param->dpte_req_per_row_ub * 64; //64B dpte request 653 654 rq_dlg_param->dpte_row_height = 1 << log2_dpte_row_height; 655 656 // the dpte_group_bytes is reduced for the specific case of vertical 657 // access of a tile surface that has dpte request of 8x1 ptes. 658 659 if (hostvm_enable) 660 rq_sizing_param->dpte_group_bytes = 512; 661 else { 662 if (!surf_linear & (log2_dpte_req_height_ptes == 0) & surf_vert) //reduced, in this case, will have page fault within a group 663 rq_sizing_param->dpte_group_bytes = 512; 664 else 665 //full size 666 rq_sizing_param->dpte_group_bytes = 2048; 667 } 668 669 //since pte request size is 64byte, the number of data pte requests per full sized group is as follows. 670 log2_dpte_group_bytes = dml_log2(rq_sizing_param->dpte_group_bytes); 671 log2_dpte_group_length = log2_dpte_group_bytes - 6; //length in 64b requests 672 673 // full sized data pte group width in elements 674 if (!surf_vert) 675 log2_dpte_group_width = log2_dpte_group_length + log2_dpte_req_width; 676 else 677 log2_dpte_group_width = log2_dpte_group_length + log2_dpte_req_height; 678 679 //But if the tile block >=64KB and the page size is 4KB, then each dPTE request is 2*64B 680 if ((log2_blk_bytes >= 16) && (log2_vmpg_bytes == 12)) // tile block >= 64KB 681 log2_dpte_group_width = log2_dpte_group_width - 1; 682 683 dpte_group_width = 1 << log2_dpte_group_width; 684 685 // since dpte groups are only aligned to dpte_req_width and not dpte_group_width, 686 // the upper bound for the dpte groups per row is as follows. 687 rq_dlg_param->dpte_groups_per_row_ub = dml_ceil( 688 (double) dpte_row_width_ub / dpte_group_width, 689 1); 690 } 691 692 static void get_surf_rq_param( 693 struct display_mode_lib *mode_lib, 694 display_data_rq_sizing_params_st *rq_sizing_param, 695 display_data_rq_dlg_params_st *rq_dlg_param, 696 display_data_rq_misc_params_st *rq_misc_param, 697 const display_pipe_params_st *pipe_param, 698 bool is_chroma) 699 { 700 bool mode_422 = false; 701 unsigned int vp_width = 0; 702 unsigned int vp_height = 0; 703 unsigned int data_pitch = 0; 704 unsigned int meta_pitch = 0; 705 unsigned int ppe = mode_422 ? 2 : 1; 706 707 // FIXME check if ppe apply for both luma and chroma in 422 case 708 if (is_chroma) { 709 vp_width = pipe_param->src.viewport_width_c / ppe; 710 vp_height = pipe_param->src.viewport_height_c; 711 data_pitch = pipe_param->src.data_pitch_c; 712 meta_pitch = pipe_param->src.meta_pitch_c; 713 } else { 714 vp_width = pipe_param->src.viewport_width / ppe; 715 vp_height = pipe_param->src.viewport_height; 716 data_pitch = pipe_param->src.data_pitch; 717 meta_pitch = pipe_param->src.meta_pitch; 718 } 719 720 if (pipe_param->dest.odm_combine) { 721 unsigned int access_dir; 722 unsigned int full_src_vp_width; 723 unsigned int hactive_half; 724 unsigned int src_hactive_half; 725 access_dir = (pipe_param->src.source_scan == dm_vert); // vp access direction: horizontal or vertical accessed 726 hactive_half = pipe_param->dest.hactive / 2; 727 if (is_chroma) { 728 full_src_vp_width = pipe_param->scale_ratio_depth.hscl_ratio_c * pipe_param->dest.full_recout_width; 729 src_hactive_half = pipe_param->scale_ratio_depth.hscl_ratio_c * hactive_half; 730 } else { 731 full_src_vp_width = pipe_param->scale_ratio_depth.hscl_ratio * pipe_param->dest.full_recout_width; 732 src_hactive_half = pipe_param->scale_ratio_depth.hscl_ratio * hactive_half; 733 } 734 735 if (access_dir == 0) { 736 vp_width = dml_min(full_src_vp_width, src_hactive_half); 737 dml_print("DML_DLG: %s: vp_width = %d\n", __func__, vp_width); 738 } else { 739 vp_height = dml_min(full_src_vp_width, src_hactive_half); 740 dml_print("DML_DLG: %s: vp_height = %d\n", __func__, vp_height); 741 742 } 743 dml_print("DML_DLG: %s: full_src_vp_width = %d\n", __func__, full_src_vp_width); 744 dml_print("DML_DLG: %s: hactive_half = %d\n", __func__, hactive_half); 745 dml_print("DML_DLG: %s: src_hactive_half = %d\n", __func__, src_hactive_half); 746 } 747 rq_sizing_param->chunk_bytes = 8192; 748 749 if (rq_sizing_param->chunk_bytes == 64 * 1024) 750 rq_sizing_param->min_chunk_bytes = 0; 751 else 752 rq_sizing_param->min_chunk_bytes = 1024; 753 754 rq_sizing_param->meta_chunk_bytes = 2048; 755 rq_sizing_param->min_meta_chunk_bytes = 256; 756 757 if (pipe_param->src.hostvm) 758 rq_sizing_param->mpte_group_bytes = 512; 759 else 760 rq_sizing_param->mpte_group_bytes = 2048; 761 762 get_meta_and_pte_attr( 763 mode_lib, 764 rq_dlg_param, 765 rq_misc_param, 766 rq_sizing_param, 767 vp_width, 768 vp_height, 769 data_pitch, 770 meta_pitch, 771 pipe_param->src.source_format, 772 pipe_param->src.sw_mode, 773 pipe_param->src.macro_tile_size, 774 pipe_param->src.source_scan, 775 pipe_param->src.hostvm, 776 is_chroma); 777 } 778 779 static void dml_rq_dlg_get_rq_params( 780 struct display_mode_lib *mode_lib, 781 display_rq_params_st *rq_param, 782 const display_pipe_params_st *pipe_param) 783 { 784 // get param for luma surface 785 rq_param->yuv420 = pipe_param->src.source_format == dm_420_8 786 || pipe_param->src.source_format == dm_420_10; 787 rq_param->yuv420_10bpc = pipe_param->src.source_format == dm_420_10; 788 789 get_surf_rq_param( 790 mode_lib, 791 &(rq_param->sizing.rq_l), 792 &(rq_param->dlg.rq_l), 793 &(rq_param->misc.rq_l), 794 pipe_param, 795 0); 796 797 if (is_dual_plane((enum source_format_class) (pipe_param->src.source_format))) { 798 // get param for chroma surface 799 get_surf_rq_param( 800 mode_lib, 801 &(rq_param->sizing.rq_c), 802 &(rq_param->dlg.rq_c), 803 &(rq_param->misc.rq_c), 804 pipe_param, 805 1); 806 } 807 808 // calculate how to split the det buffer space between luma and chroma 809 handle_det_buf_split(mode_lib, rq_param, &pipe_param->src); 810 print__rq_params_st(mode_lib, rq_param); 811 } 812 813 void dml21_rq_dlg_get_rq_reg( 814 struct display_mode_lib *mode_lib, 815 display_rq_regs_st *rq_regs, 816 const display_pipe_params_st *pipe_param) 817 { 818 display_rq_params_st rq_param = {0}; 819 820 memset(rq_regs, 0, sizeof(*rq_regs)); 821 dml_rq_dlg_get_rq_params(mode_lib, &rq_param, pipe_param); 822 extract_rq_regs(mode_lib, rq_regs, &rq_param); 823 824 print__rq_regs_st(mode_lib, rq_regs); 825 } 826 827 // Note: currently taken in as is. 828 // Nice to decouple code from hw register implement and extract code that are repeated for luma and chroma. 829 static void dml_rq_dlg_get_dlg_params( 830 struct display_mode_lib *mode_lib, 831 const display_e2e_pipe_params_st *e2e_pipe_param, 832 const unsigned int num_pipes, 833 const unsigned int pipe_idx, 834 display_dlg_regs_st *disp_dlg_regs, 835 display_ttu_regs_st *disp_ttu_regs, 836 const display_rq_dlg_params_st *rq_dlg_param, 837 const display_dlg_sys_params_st *dlg_sys_param, 838 const bool cstate_en, 839 const bool pstate_en) 840 { 841 const display_pipe_source_params_st *src = &e2e_pipe_param[pipe_idx].pipe.src; 842 const display_pipe_dest_params_st *dst = &e2e_pipe_param[pipe_idx].pipe.dest; 843 const display_output_params_st *dout = &e2e_pipe_param[pipe_idx].dout; 844 const display_clocks_and_cfg_st *clks = &e2e_pipe_param[pipe_idx].clks_cfg; 845 const scaler_ratio_depth_st *scl = &e2e_pipe_param[pipe_idx].pipe.scale_ratio_depth; 846 const scaler_taps_st *taps = &e2e_pipe_param[pipe_idx].pipe.scale_taps; 847 848 // ------------------------- 849 // Section 1.15.2.1: OTG dependent Params 850 // ------------------------- 851 // Timing 852 unsigned int htotal = dst->htotal; 853 // unsigned int hblank_start = dst.hblank_start; // TODO: Remove 854 unsigned int hblank_end = dst->hblank_end; 855 unsigned int vblank_start = dst->vblank_start; 856 unsigned int vblank_end = dst->vblank_end; 857 unsigned int min_vblank = mode_lib->ip.min_vblank_lines; 858 859 double dppclk_freq_in_mhz = clks->dppclk_mhz; 860 double dispclk_freq_in_mhz = clks->dispclk_mhz; 861 double refclk_freq_in_mhz = clks->refclk_mhz; 862 double pclk_freq_in_mhz = dst->pixel_rate_mhz; 863 bool interlaced = dst->interlaced; 864 865 double ref_freq_to_pix_freq = refclk_freq_in_mhz / pclk_freq_in_mhz; 866 867 double min_dcfclk_mhz; 868 double t_calc_us; 869 double min_ttu_vblank; 870 871 double min_dst_y_ttu_vblank; 872 unsigned int dlg_vblank_start; 873 bool dual_plane; 874 bool mode_422; 875 unsigned int access_dir; 876 unsigned int vp_height_l; 877 unsigned int vp_width_l; 878 unsigned int vp_height_c; 879 unsigned int vp_width_c; 880 881 // Scaling 882 unsigned int htaps_l; 883 unsigned int htaps_c; 884 double hratio_l; 885 double hratio_c; 886 double vratio_l; 887 double vratio_c; 888 bool scl_enable; 889 890 double line_time_in_us; 891 // double vinit_l; 892 // double vinit_c; 893 // double vinit_bot_l; 894 // double vinit_bot_c; 895 896 // unsigned int swath_height_l; 897 unsigned int swath_width_ub_l; 898 // unsigned int dpte_bytes_per_row_ub_l; 899 unsigned int dpte_groups_per_row_ub_l; 900 // unsigned int meta_pte_bytes_per_frame_ub_l; 901 // unsigned int meta_bytes_per_row_ub_l; 902 903 // unsigned int swath_height_c; 904 unsigned int swath_width_ub_c; 905 // unsigned int dpte_bytes_per_row_ub_c; 906 unsigned int dpte_groups_per_row_ub_c; 907 908 unsigned int meta_chunks_per_row_ub_l; 909 unsigned int meta_chunks_per_row_ub_c; 910 unsigned int vupdate_offset; 911 unsigned int vupdate_width; 912 unsigned int vready_offset; 913 914 unsigned int dppclk_delay_subtotal; 915 unsigned int dispclk_delay_subtotal; 916 unsigned int pixel_rate_delay_subtotal; 917 918 unsigned int vstartup_start; 919 unsigned int dst_x_after_scaler; 920 unsigned int dst_y_after_scaler; 921 double line_wait; 922 double dst_y_prefetch; 923 double dst_y_per_vm_vblank; 924 double dst_y_per_row_vblank; 925 double dst_y_per_vm_flip; 926 double dst_y_per_row_flip; 927 double max_dst_y_per_vm_vblank; 928 double max_dst_y_per_row_vblank; 929 double lsw; 930 double vratio_pre_l; 931 double vratio_pre_c; 932 unsigned int req_per_swath_ub_l; 933 unsigned int req_per_swath_ub_c; 934 unsigned int meta_row_height_l; 935 unsigned int meta_row_height_c; 936 unsigned int swath_width_pixels_ub_l; 937 unsigned int swath_width_pixels_ub_c; 938 unsigned int scaler_rec_in_width_l; 939 unsigned int scaler_rec_in_width_c; 940 unsigned int dpte_row_height_l; 941 unsigned int dpte_row_height_c; 942 double hscale_pixel_rate_l; 943 double hscale_pixel_rate_c; 944 double min_hratio_fact_l; 945 double min_hratio_fact_c; 946 double refcyc_per_line_delivery_pre_l; 947 double refcyc_per_line_delivery_pre_c; 948 double refcyc_per_line_delivery_l; 949 double refcyc_per_line_delivery_c; 950 951 double refcyc_per_req_delivery_pre_l; 952 double refcyc_per_req_delivery_pre_c; 953 double refcyc_per_req_delivery_l; 954 double refcyc_per_req_delivery_c; 955 956 unsigned int full_recout_width; 957 double refcyc_per_req_delivery_pre_cur0; 958 double refcyc_per_req_delivery_cur0; 959 double refcyc_per_req_delivery_pre_cur1; 960 double refcyc_per_req_delivery_cur1; 961 962 memset(disp_dlg_regs, 0, sizeof(*disp_dlg_regs)); 963 memset(disp_ttu_regs, 0, sizeof(*disp_ttu_regs)); 964 965 dml_print("DML_DLG: %s: cstate_en = %d\n", __func__, cstate_en); 966 dml_print("DML_DLG: %s: pstate_en = %d\n", __func__, pstate_en); 967 968 dml_print("DML_DLG: %s: dppclk_freq_in_mhz = %3.2f\n", __func__, dppclk_freq_in_mhz); 969 dml_print("DML_DLG: %s: dispclk_freq_in_mhz = %3.2f\n", __func__, dispclk_freq_in_mhz); 970 dml_print("DML_DLG: %s: refclk_freq_in_mhz = %3.2f\n", __func__, refclk_freq_in_mhz); 971 dml_print("DML_DLG: %s: pclk_freq_in_mhz = %3.2f\n", __func__, pclk_freq_in_mhz); 972 dml_print("DML_DLG: %s: interlaced = %d\n", __func__, interlaced); 973 ASSERT(ref_freq_to_pix_freq < 4.0); 974 975 disp_dlg_regs->ref_freq_to_pix_freq = 976 (unsigned int) (ref_freq_to_pix_freq * dml_pow(2, 19)); 977 disp_dlg_regs->refcyc_per_htotal = (unsigned int) (ref_freq_to_pix_freq * (double) htotal 978 * dml_pow(2, 8)); 979 disp_dlg_regs->dlg_vblank_end = interlaced ? (vblank_end / 2) : vblank_end; // 15 bits 980 disp_dlg_regs->refcyc_h_blank_end = (unsigned int) ((double) hblank_end 981 * (double) ref_freq_to_pix_freq); 982 ASSERT(disp_dlg_regs->refcyc_h_blank_end < (unsigned int)dml_pow(2, 13)); 983 984 min_dcfclk_mhz = dlg_sys_param->deepsleep_dcfclk_mhz; 985 t_calc_us = get_tcalc(mode_lib, e2e_pipe_param, num_pipes); 986 min_ttu_vblank = get_min_ttu_vblank(mode_lib, e2e_pipe_param, num_pipes, pipe_idx); 987 988 min_dst_y_ttu_vblank = min_ttu_vblank * pclk_freq_in_mhz / (double) htotal; 989 dlg_vblank_start = interlaced ? (vblank_start / 2) : vblank_start; 990 991 disp_dlg_regs->min_dst_y_next_start = (unsigned int) (((double) dlg_vblank_start) * dml_pow(2, 2)); 992 ASSERT(disp_dlg_regs->min_dst_y_next_start < (unsigned int)dml_pow(2, 18)); 993 994 dml_print( 995 "DML_DLG: %s: min_dcfclk_mhz = %3.2f\n", 996 __func__, 997 min_dcfclk_mhz); 998 dml_print( 999 "DML_DLG: %s: min_ttu_vblank = %3.2f\n", 1000 __func__, 1001 min_ttu_vblank); 1002 dml_print( 1003 "DML_DLG: %s: min_dst_y_ttu_vblank = %3.2f\n", 1004 __func__, 1005 min_dst_y_ttu_vblank); 1006 dml_print( 1007 "DML_DLG: %s: t_calc_us = %3.2f\n", 1008 __func__, 1009 t_calc_us); 1010 dml_print( 1011 "DML_DLG: %s: disp_dlg_regs->min_dst_y_next_start = 0x%0x\n", 1012 __func__, 1013 disp_dlg_regs->min_dst_y_next_start); 1014 dml_print( 1015 "DML_DLG: %s: ref_freq_to_pix_freq = %3.2f\n", 1016 __func__, 1017 ref_freq_to_pix_freq); 1018 1019 // ------------------------- 1020 // Section 1.15.2.2: Prefetch, Active and TTU 1021 // ------------------------- 1022 // Prefetch Calc 1023 // Source 1024 // dcc_en = src.dcc; 1025 dual_plane = is_dual_plane((enum source_format_class) (src->source_format)); 1026 mode_422 = false; // FIXME 1027 access_dir = (src->source_scan == dm_vert); // vp access direction: horizontal or vertical accessed 1028 // bytes_per_element_l = get_bytes_per_element(source_format_class(src.source_format), 0); 1029 // bytes_per_element_c = get_bytes_per_element(source_format_class(src.source_format), 1); 1030 vp_height_l = src->viewport_height; 1031 vp_width_l = src->viewport_width; 1032 vp_height_c = src->viewport_height_c; 1033 vp_width_c = src->viewport_width_c; 1034 1035 // Scaling 1036 htaps_l = taps->htaps; 1037 htaps_c = taps->htaps_c; 1038 hratio_l = scl->hscl_ratio; 1039 hratio_c = scl->hscl_ratio_c; 1040 vratio_l = scl->vscl_ratio; 1041 vratio_c = scl->vscl_ratio_c; 1042 scl_enable = scl->scl_enable; 1043 1044 line_time_in_us = (htotal / pclk_freq_in_mhz); 1045 swath_width_ub_l = rq_dlg_param->rq_l.swath_width_ub; 1046 dpte_groups_per_row_ub_l = rq_dlg_param->rq_l.dpte_groups_per_row_ub; 1047 swath_width_ub_c = rq_dlg_param->rq_c.swath_width_ub; 1048 dpte_groups_per_row_ub_c = rq_dlg_param->rq_c.dpte_groups_per_row_ub; 1049 1050 meta_chunks_per_row_ub_l = rq_dlg_param->rq_l.meta_chunks_per_row_ub; 1051 meta_chunks_per_row_ub_c = rq_dlg_param->rq_c.meta_chunks_per_row_ub; 1052 vupdate_offset = dst->vupdate_offset; 1053 vupdate_width = dst->vupdate_width; 1054 vready_offset = dst->vready_offset; 1055 1056 dppclk_delay_subtotal = mode_lib->ip.dppclk_delay_subtotal; 1057 dispclk_delay_subtotal = mode_lib->ip.dispclk_delay_subtotal; 1058 1059 if (scl_enable) 1060 dppclk_delay_subtotal += mode_lib->ip.dppclk_delay_scl; 1061 else 1062 dppclk_delay_subtotal += mode_lib->ip.dppclk_delay_scl_lb_only; 1063 1064 dppclk_delay_subtotal += mode_lib->ip.dppclk_delay_cnvc_formatter 1065 + src->num_cursors * mode_lib->ip.dppclk_delay_cnvc_cursor; 1066 1067 if (dout->dsc_enable) { 1068 double dsc_delay = get_dsc_delay(mode_lib, e2e_pipe_param, num_pipes, pipe_idx); 1069 1070 dispclk_delay_subtotal += dsc_delay; 1071 } 1072 1073 pixel_rate_delay_subtotal = dppclk_delay_subtotal * pclk_freq_in_mhz / dppclk_freq_in_mhz 1074 + dispclk_delay_subtotal * pclk_freq_in_mhz / dispclk_freq_in_mhz; 1075 1076 vstartup_start = dst->vstartup_start; 1077 if (interlaced) { 1078 if (vstartup_start / 2.0 1079 - (double) (vready_offset + vupdate_width + vupdate_offset) / htotal 1080 <= vblank_end / 2.0) 1081 disp_dlg_regs->vready_after_vcount0 = 1; 1082 else 1083 disp_dlg_regs->vready_after_vcount0 = 0; 1084 } else { 1085 if (vstartup_start 1086 - (double) (vready_offset + vupdate_width + vupdate_offset) / htotal 1087 <= vblank_end) 1088 disp_dlg_regs->vready_after_vcount0 = 1; 1089 else 1090 disp_dlg_regs->vready_after_vcount0 = 0; 1091 } 1092 1093 // TODO: Where is this coming from? 1094 if (interlaced) 1095 vstartup_start = vstartup_start / 2; 1096 1097 // TODO: What if this min_vblank doesn't match the value in the dml_config_settings.cpp? 1098 if (vstartup_start >= min_vblank) { 1099 dml_print( 1100 "WARNING: DML_DLG: %s: vblank_start=%d vblank_end=%d\n", 1101 __func__, 1102 vblank_start, 1103 vblank_end); 1104 dml_print( 1105 "WARNING: DML_DLG: %s: vstartup_start=%d should be less than min_vblank=%d\n", 1106 __func__, 1107 vstartup_start, 1108 min_vblank); 1109 min_vblank = vstartup_start + 1; 1110 dml_print( 1111 "WARNING: DML_DLG: %s: vstartup_start=%d should be less than min_vblank=%d\n", 1112 __func__, 1113 vstartup_start, 1114 min_vblank); 1115 } 1116 1117 dst_x_after_scaler = get_dst_x_after_scaler(mode_lib, e2e_pipe_param, num_pipes, pipe_idx); 1118 dst_y_after_scaler = get_dst_y_after_scaler(mode_lib, e2e_pipe_param, num_pipes, pipe_idx); 1119 1120 dml_print("DML_DLG: %s: htotal = %d\n", __func__, htotal); 1121 dml_print( 1122 "DML_DLG: %s: pixel_rate_delay_subtotal = %d\n", 1123 __func__, 1124 pixel_rate_delay_subtotal); 1125 dml_print( 1126 "DML_DLG: %s: dst_x_after_scaler = %d\n", 1127 __func__, 1128 dst_x_after_scaler); 1129 dml_print( 1130 "DML_DLG: %s: dst_y_after_scaler = %d\n", 1131 __func__, 1132 dst_y_after_scaler); 1133 1134 // Lwait 1135 // TODO: Should this be urgent_latency_pixel_mixed_with_vm_data_us? 1136 line_wait = mode_lib->soc.urgent_latency_pixel_data_only_us; 1137 if (cstate_en) 1138 line_wait = dml_max(mode_lib->soc.sr_enter_plus_exit_time_us, line_wait); 1139 if (pstate_en) 1140 line_wait = dml_max( 1141 mode_lib->soc.dram_clock_change_latency_us 1142 + mode_lib->soc.urgent_latency_pixel_data_only_us, // TODO: Should this be urgent_latency_pixel_mixed_with_vm_data_us? 1143 line_wait); 1144 line_wait = line_wait / line_time_in_us; 1145 1146 dst_y_prefetch = get_dst_y_prefetch(mode_lib, e2e_pipe_param, num_pipes, pipe_idx); 1147 dml_print("DML_DLG: %s: dst_y_prefetch (after rnd) = %3.2f\n", __func__, dst_y_prefetch); 1148 1149 dst_y_per_vm_vblank = get_dst_y_per_vm_vblank( 1150 mode_lib, 1151 e2e_pipe_param, 1152 num_pipes, 1153 pipe_idx); 1154 dst_y_per_row_vblank = get_dst_y_per_row_vblank( 1155 mode_lib, 1156 e2e_pipe_param, 1157 num_pipes, 1158 pipe_idx); 1159 dst_y_per_vm_flip = get_dst_y_per_vm_flip(mode_lib, e2e_pipe_param, num_pipes, pipe_idx); 1160 dst_y_per_row_flip = get_dst_y_per_row_flip(mode_lib, e2e_pipe_param, num_pipes, pipe_idx); 1161 1162 max_dst_y_per_vm_vblank = 32.0; 1163 max_dst_y_per_row_vblank = 16.0; 1164 1165 // magic! 1166 if (htotal <= 75) { 1167 min_vblank = 300; 1168 max_dst_y_per_vm_vblank = 100.0; 1169 max_dst_y_per_row_vblank = 100.0; 1170 } 1171 1172 dml_print("DML_DLG: %s: dst_y_per_vm_flip = %3.2f\n", __func__, dst_y_per_vm_flip); 1173 dml_print("DML_DLG: %s: dst_y_per_row_flip = %3.2f\n", __func__, dst_y_per_row_flip); 1174 dml_print("DML_DLG: %s: dst_y_per_vm_vblank = %3.2f\n", __func__, dst_y_per_vm_vblank); 1175 dml_print("DML_DLG: %s: dst_y_per_row_vblank = %3.2f\n", __func__, dst_y_per_row_vblank); 1176 1177 ASSERT(dst_y_per_vm_vblank < max_dst_y_per_vm_vblank); 1178 ASSERT(dst_y_per_row_vblank < max_dst_y_per_row_vblank); 1179 1180 ASSERT(dst_y_prefetch > (dst_y_per_vm_vblank + dst_y_per_row_vblank)); 1181 lsw = dst_y_prefetch - (dst_y_per_vm_vblank + dst_y_per_row_vblank); 1182 1183 dml_print("DML_DLG: %s: lsw = %3.2f\n", __func__, lsw); 1184 1185 vratio_pre_l = get_vratio_prefetch_l(mode_lib, e2e_pipe_param, num_pipes, pipe_idx); 1186 vratio_pre_c = get_vratio_prefetch_c(mode_lib, e2e_pipe_param, num_pipes, pipe_idx); 1187 1188 dml_print("DML_DLG: %s: vratio_pre_l=%3.2f\n", __func__, vratio_pre_l); 1189 dml_print("DML_DLG: %s: vratio_pre_c=%3.2f\n", __func__, vratio_pre_c); 1190 1191 // Active 1192 req_per_swath_ub_l = rq_dlg_param->rq_l.req_per_swath_ub; 1193 req_per_swath_ub_c = rq_dlg_param->rq_c.req_per_swath_ub; 1194 meta_row_height_l = rq_dlg_param->rq_l.meta_row_height; 1195 meta_row_height_c = rq_dlg_param->rq_c.meta_row_height; 1196 swath_width_pixels_ub_l = 0; 1197 swath_width_pixels_ub_c = 0; 1198 scaler_rec_in_width_l = 0; 1199 scaler_rec_in_width_c = 0; 1200 dpte_row_height_l = rq_dlg_param->rq_l.dpte_row_height; 1201 dpte_row_height_c = rq_dlg_param->rq_c.dpte_row_height; 1202 1203 if (mode_422) { 1204 swath_width_pixels_ub_l = swath_width_ub_l * 2; // *2 for 2 pixel per element 1205 swath_width_pixels_ub_c = swath_width_ub_c * 2; 1206 } else { 1207 swath_width_pixels_ub_l = swath_width_ub_l * 1; 1208 swath_width_pixels_ub_c = swath_width_ub_c * 1; 1209 } 1210 1211 hscale_pixel_rate_l = 0.; 1212 hscale_pixel_rate_c = 0.; 1213 min_hratio_fact_l = 1.0; 1214 min_hratio_fact_c = 1.0; 1215 1216 if (hratio_l <= 1) 1217 min_hratio_fact_l = 2.0; 1218 else if (htaps_l <= 6) { 1219 if ((hratio_l * 2.0) > 4.0) 1220 min_hratio_fact_l = 4.0; 1221 else 1222 min_hratio_fact_l = hratio_l * 2.0; 1223 } else { 1224 if (hratio_l > 4.0) 1225 min_hratio_fact_l = 4.0; 1226 else 1227 min_hratio_fact_l = hratio_l; 1228 } 1229 1230 hscale_pixel_rate_l = min_hratio_fact_l * dppclk_freq_in_mhz; 1231 1232 if (hratio_c <= 1) 1233 min_hratio_fact_c = 2.0; 1234 else if (htaps_c <= 6) { 1235 if ((hratio_c * 2.0) > 4.0) 1236 min_hratio_fact_c = 4.0; 1237 else 1238 min_hratio_fact_c = hratio_c * 2.0; 1239 } else { 1240 if (hratio_c > 4.0) 1241 min_hratio_fact_c = 4.0; 1242 else 1243 min_hratio_fact_c = hratio_c; 1244 } 1245 1246 hscale_pixel_rate_c = min_hratio_fact_c * dppclk_freq_in_mhz; 1247 1248 refcyc_per_line_delivery_pre_l = 0.; 1249 refcyc_per_line_delivery_pre_c = 0.; 1250 refcyc_per_line_delivery_l = 0.; 1251 refcyc_per_line_delivery_c = 0.; 1252 1253 refcyc_per_req_delivery_pre_l = 0.; 1254 refcyc_per_req_delivery_pre_c = 0.; 1255 refcyc_per_req_delivery_l = 0.; 1256 refcyc_per_req_delivery_c = 0.; 1257 1258 full_recout_width = 0; 1259 // In ODM 1260 if (src->is_hsplit) { 1261 // This "hack" is only allowed (and valid) for MPC combine. In ODM 1262 // combine, you MUST specify the full_recout_width...according to Oswin 1263 if (dst->full_recout_width == 0 && !dst->odm_combine) { 1264 dml_print( 1265 "DML_DLG: %s: Warning: full_recout_width not set in hsplit mode\n", 1266 __func__); 1267 full_recout_width = dst->recout_width * 2; // assume half split for dcn1 1268 } else 1269 full_recout_width = dst->full_recout_width; 1270 } else 1271 full_recout_width = dst->recout_width; 1272 1273 // As of DCN2, mpc_combine and odm_combine are mutually exclusive 1274 refcyc_per_line_delivery_pre_l = get_refcyc_per_delivery( 1275 mode_lib, 1276 refclk_freq_in_mhz, 1277 pclk_freq_in_mhz, 1278 dst->odm_combine, 1279 full_recout_width, 1280 dst->hactive, 1281 vratio_pre_l, 1282 hscale_pixel_rate_l, 1283 swath_width_pixels_ub_l, 1284 1); // per line 1285 1286 refcyc_per_line_delivery_l = get_refcyc_per_delivery( 1287 mode_lib, 1288 refclk_freq_in_mhz, 1289 pclk_freq_in_mhz, 1290 dst->odm_combine, 1291 full_recout_width, 1292 dst->hactive, 1293 vratio_l, 1294 hscale_pixel_rate_l, 1295 swath_width_pixels_ub_l, 1296 1); // per line 1297 1298 dml_print("DML_DLG: %s: full_recout_width = %d\n", __func__, full_recout_width); 1299 dml_print( 1300 "DML_DLG: %s: hscale_pixel_rate_l = %3.2f\n", 1301 __func__, 1302 hscale_pixel_rate_l); 1303 dml_print( 1304 "DML_DLG: %s: refcyc_per_line_delivery_pre_l = %3.2f\n", 1305 __func__, 1306 refcyc_per_line_delivery_pre_l); 1307 dml_print( 1308 "DML_DLG: %s: refcyc_per_line_delivery_l = %3.2f\n", 1309 __func__, 1310 refcyc_per_line_delivery_l); 1311 1312 if (dual_plane) { 1313 refcyc_per_line_delivery_pre_c = get_refcyc_per_delivery( 1314 mode_lib, 1315 refclk_freq_in_mhz, 1316 pclk_freq_in_mhz, 1317 dst->odm_combine, 1318 full_recout_width, 1319 dst->hactive, 1320 vratio_pre_c, 1321 hscale_pixel_rate_c, 1322 swath_width_pixels_ub_c, 1323 1); // per line 1324 1325 refcyc_per_line_delivery_c = get_refcyc_per_delivery( 1326 mode_lib, 1327 refclk_freq_in_mhz, 1328 pclk_freq_in_mhz, 1329 dst->odm_combine, 1330 full_recout_width, 1331 dst->hactive, 1332 vratio_c, 1333 hscale_pixel_rate_c, 1334 swath_width_pixels_ub_c, 1335 1); // per line 1336 1337 dml_print( 1338 "DML_DLG: %s: refcyc_per_line_delivery_pre_c = %3.2f\n", 1339 __func__, 1340 refcyc_per_line_delivery_pre_c); 1341 dml_print( 1342 "DML_DLG: %s: refcyc_per_line_delivery_c = %3.2f\n", 1343 __func__, 1344 refcyc_per_line_delivery_c); 1345 } 1346 1347 // TTU - Luma / Chroma 1348 if (access_dir) { // vertical access 1349 scaler_rec_in_width_l = vp_height_l; 1350 scaler_rec_in_width_c = vp_height_c; 1351 } else { 1352 scaler_rec_in_width_l = vp_width_l; 1353 scaler_rec_in_width_c = vp_width_c; 1354 } 1355 1356 refcyc_per_req_delivery_pre_l = get_refcyc_per_delivery( 1357 mode_lib, 1358 refclk_freq_in_mhz, 1359 pclk_freq_in_mhz, 1360 dst->odm_combine, 1361 full_recout_width, 1362 dst->hactive, 1363 vratio_pre_l, 1364 hscale_pixel_rate_l, 1365 scaler_rec_in_width_l, 1366 req_per_swath_ub_l); // per req 1367 refcyc_per_req_delivery_l = get_refcyc_per_delivery( 1368 mode_lib, 1369 refclk_freq_in_mhz, 1370 pclk_freq_in_mhz, 1371 dst->odm_combine, 1372 full_recout_width, 1373 dst->hactive, 1374 vratio_l, 1375 hscale_pixel_rate_l, 1376 scaler_rec_in_width_l, 1377 req_per_swath_ub_l); // per req 1378 1379 dml_print( 1380 "DML_DLG: %s: refcyc_per_req_delivery_pre_l = %3.2f\n", 1381 __func__, 1382 refcyc_per_req_delivery_pre_l); 1383 dml_print( 1384 "DML_DLG: %s: refcyc_per_req_delivery_l = %3.2f\n", 1385 __func__, 1386 refcyc_per_req_delivery_l); 1387 1388 ASSERT(refcyc_per_req_delivery_pre_l < dml_pow(2, 13)); 1389 ASSERT(refcyc_per_req_delivery_l < dml_pow(2, 13)); 1390 1391 if (dual_plane) { 1392 refcyc_per_req_delivery_pre_c = get_refcyc_per_delivery( 1393 mode_lib, 1394 refclk_freq_in_mhz, 1395 pclk_freq_in_mhz, 1396 dst->odm_combine, 1397 full_recout_width, 1398 dst->hactive, 1399 vratio_pre_c, 1400 hscale_pixel_rate_c, 1401 scaler_rec_in_width_c, 1402 req_per_swath_ub_c); // per req 1403 refcyc_per_req_delivery_c = get_refcyc_per_delivery( 1404 mode_lib, 1405 refclk_freq_in_mhz, 1406 pclk_freq_in_mhz, 1407 dst->odm_combine, 1408 full_recout_width, 1409 dst->hactive, 1410 vratio_c, 1411 hscale_pixel_rate_c, 1412 scaler_rec_in_width_c, 1413 req_per_swath_ub_c); // per req 1414 1415 dml_print( 1416 "DML_DLG: %s: refcyc_per_req_delivery_pre_c = %3.2f\n", 1417 __func__, 1418 refcyc_per_req_delivery_pre_c); 1419 dml_print( 1420 "DML_DLG: %s: refcyc_per_req_delivery_c = %3.2f\n", 1421 __func__, 1422 refcyc_per_req_delivery_c); 1423 1424 ASSERT(refcyc_per_req_delivery_pre_c < dml_pow(2, 13)); 1425 ASSERT(refcyc_per_req_delivery_c < dml_pow(2, 13)); 1426 } 1427 1428 // TTU - Cursor 1429 refcyc_per_req_delivery_pre_cur0 = 0.0; 1430 refcyc_per_req_delivery_cur0 = 0.0; 1431 if (src->num_cursors > 0) { 1432 calculate_ttu_cursor( 1433 mode_lib, 1434 &refcyc_per_req_delivery_pre_cur0, 1435 &refcyc_per_req_delivery_cur0, 1436 refclk_freq_in_mhz, 1437 ref_freq_to_pix_freq, 1438 hscale_pixel_rate_l, 1439 scl->hscl_ratio, 1440 vratio_pre_l, 1441 vratio_l, 1442 src->cur0_src_width, 1443 (enum cursor_bpp) (src->cur0_bpp)); 1444 } 1445 1446 refcyc_per_req_delivery_pre_cur1 = 0.0; 1447 refcyc_per_req_delivery_cur1 = 0.0; 1448 if (src->num_cursors > 1) { 1449 calculate_ttu_cursor( 1450 mode_lib, 1451 &refcyc_per_req_delivery_pre_cur1, 1452 &refcyc_per_req_delivery_cur1, 1453 refclk_freq_in_mhz, 1454 ref_freq_to_pix_freq, 1455 hscale_pixel_rate_l, 1456 scl->hscl_ratio, 1457 vratio_pre_l, 1458 vratio_l, 1459 src->cur1_src_width, 1460 (enum cursor_bpp) (src->cur1_bpp)); 1461 } 1462 1463 // TTU - Misc 1464 // all hard-coded 1465 1466 // Assignment to register structures 1467 disp_dlg_regs->dst_y_after_scaler = dst_y_after_scaler; // in terms of line 1468 disp_dlg_regs->refcyc_x_after_scaler = dst_x_after_scaler * ref_freq_to_pix_freq; // in terms of refclk 1469 ASSERT(disp_dlg_regs->refcyc_x_after_scaler < (unsigned int)dml_pow(2, 13)); 1470 disp_dlg_regs->dst_y_prefetch = (unsigned int) (dst_y_prefetch * dml_pow(2, 2)); 1471 disp_dlg_regs->dst_y_per_vm_vblank = (unsigned int) (dst_y_per_vm_vblank * dml_pow(2, 2)); 1472 disp_dlg_regs->dst_y_per_row_vblank = (unsigned int) (dst_y_per_row_vblank * dml_pow(2, 2)); 1473 disp_dlg_regs->dst_y_per_vm_flip = (unsigned int) (dst_y_per_vm_flip * dml_pow(2, 2)); 1474 disp_dlg_regs->dst_y_per_row_flip = (unsigned int) (dst_y_per_row_flip * dml_pow(2, 2)); 1475 1476 disp_dlg_regs->vratio_prefetch = (unsigned int) (vratio_pre_l * dml_pow(2, 19)); 1477 disp_dlg_regs->vratio_prefetch_c = (unsigned int) (vratio_pre_c * dml_pow(2, 19)); 1478 1479 dml_print("DML_DLG: %s: disp_dlg_regs->dst_y_per_vm_vblank = 0x%x\n", __func__, disp_dlg_regs->dst_y_per_vm_vblank); 1480 dml_print("DML_DLG: %s: disp_dlg_regs->dst_y_per_row_vblank = 0x%x\n", __func__, disp_dlg_regs->dst_y_per_row_vblank); 1481 dml_print("DML_DLG: %s: disp_dlg_regs->dst_y_per_vm_flip = 0x%x\n", __func__, disp_dlg_regs->dst_y_per_vm_flip); 1482 dml_print("DML_DLG: %s: disp_dlg_regs->dst_y_per_row_flip = 0x%x\n", __func__, disp_dlg_regs->dst_y_per_row_flip); 1483 1484 disp_dlg_regs->refcyc_per_pte_group_vblank_l = 1485 (unsigned int) (dst_y_per_row_vblank * (double) htotal 1486 * ref_freq_to_pix_freq / (double) dpte_groups_per_row_ub_l); 1487 if ((refclk_freq_in_mhz / ref_freq_to_pix_freq < 28) && 1488 disp_dlg_regs->refcyc_per_pte_group_vblank_l >= (unsigned int)dml_pow(2, 13)) 1489 disp_dlg_regs->refcyc_per_pte_group_vblank_l = (1 << 13) - 1; 1490 else 1491 ASSERT(disp_dlg_regs->refcyc_per_pte_group_vblank_l < (unsigned int)dml_pow(2, 13)); 1492 1493 if (dual_plane) { 1494 disp_dlg_regs->refcyc_per_pte_group_vblank_c = (unsigned int) (dst_y_per_row_vblank 1495 * (double) htotal * ref_freq_to_pix_freq 1496 / (double) dpte_groups_per_row_ub_c); 1497 if ((refclk_freq_in_mhz / ref_freq_to_pix_freq < 28) && 1498 disp_dlg_regs->refcyc_per_pte_group_vblank_c >= (unsigned int)dml_pow(2, 13)) 1499 disp_dlg_regs->refcyc_per_pte_group_vblank_c = (1 << 13) - 1; 1500 else 1501 ASSERT(disp_dlg_regs->refcyc_per_pte_group_vblank_c 1502 < (unsigned int)dml_pow(2, 13)); 1503 } 1504 1505 if (src->dcc) 1506 disp_dlg_regs->refcyc_per_meta_chunk_vblank_l = 1507 (unsigned int) (dst_y_per_row_vblank * (double) htotal 1508 * ref_freq_to_pix_freq / (double) meta_chunks_per_row_ub_l); 1509 else 1510 disp_dlg_regs->refcyc_per_meta_chunk_vblank_l = 0; 1511 ASSERT(disp_dlg_regs->refcyc_per_meta_chunk_vblank_l < (unsigned int)dml_pow(2, 13)); 1512 1513 disp_dlg_regs->refcyc_per_meta_chunk_vblank_c = 1514 disp_dlg_regs->refcyc_per_meta_chunk_vblank_l; // dcc for 4:2:0 is not supported in dcn1.0. assigned to be the same as _l for now 1515 1516 disp_dlg_regs->refcyc_per_pte_group_flip_l = (unsigned int) (dst_y_per_row_flip * htotal 1517 * ref_freq_to_pix_freq) / dpte_groups_per_row_ub_l; 1518 disp_dlg_regs->refcyc_per_meta_chunk_flip_l = (unsigned int) (dst_y_per_row_flip * htotal 1519 * ref_freq_to_pix_freq) / meta_chunks_per_row_ub_l; 1520 1521 if (dual_plane) { 1522 disp_dlg_regs->refcyc_per_pte_group_flip_c = (unsigned int) (dst_y_per_row_flip 1523 * htotal * ref_freq_to_pix_freq) / dpte_groups_per_row_ub_c; 1524 disp_dlg_regs->refcyc_per_meta_chunk_flip_c = (unsigned int) (dst_y_per_row_flip 1525 * htotal * ref_freq_to_pix_freq) / meta_chunks_per_row_ub_c; 1526 } 1527 1528 disp_dlg_regs->refcyc_per_vm_group_vblank = get_refcyc_per_vm_group_vblank(mode_lib, e2e_pipe_param, num_pipes, pipe_idx) * refclk_freq_in_mhz; 1529 disp_dlg_regs->refcyc_per_vm_group_flip = get_refcyc_per_vm_group_flip(mode_lib, e2e_pipe_param, num_pipes, pipe_idx) * refclk_freq_in_mhz; 1530 disp_dlg_regs->refcyc_per_vm_req_vblank = get_refcyc_per_vm_req_vblank(mode_lib, e2e_pipe_param, num_pipes, pipe_idx) * refclk_freq_in_mhz * dml_pow(2, 10); 1531 disp_dlg_regs->refcyc_per_vm_req_flip = get_refcyc_per_vm_req_flip(mode_lib, e2e_pipe_param, num_pipes, pipe_idx) * refclk_freq_in_mhz * dml_pow(2, 10); 1532 1533 // Clamp to max for now 1534 if (disp_dlg_regs->refcyc_per_vm_group_vblank >= (unsigned int)dml_pow(2, 23)) 1535 disp_dlg_regs->refcyc_per_vm_group_vblank = dml_pow(2, 23) - 1; 1536 1537 if (disp_dlg_regs->refcyc_per_vm_group_flip >= (unsigned int)dml_pow(2, 23)) 1538 disp_dlg_regs->refcyc_per_vm_group_flip = dml_pow(2, 23) - 1; 1539 1540 if (disp_dlg_regs->refcyc_per_vm_req_vblank >= (unsigned int)dml_pow(2, 23)) 1541 disp_dlg_regs->refcyc_per_vm_req_vblank = dml_pow(2, 23) - 1; 1542 1543 if (disp_dlg_regs->refcyc_per_vm_req_flip >= (unsigned int)dml_pow(2, 23)) 1544 disp_dlg_regs->refcyc_per_vm_req_flip = dml_pow(2, 23) - 1; 1545 disp_dlg_regs->dst_y_per_pte_row_nom_l = (unsigned int) ((double) dpte_row_height_l 1546 / (double) vratio_l * dml_pow(2, 2)); 1547 ASSERT(disp_dlg_regs->dst_y_per_pte_row_nom_l < (unsigned int)dml_pow(2, 17)); 1548 1549 if (dual_plane) { 1550 disp_dlg_regs->dst_y_per_pte_row_nom_c = (unsigned int) ((double) dpte_row_height_c 1551 / (double) vratio_c * dml_pow(2, 2)); 1552 if (disp_dlg_regs->dst_y_per_pte_row_nom_c >= (unsigned int) dml_pow(2, 17)) { 1553 dml_print( 1554 "DML_DLG: %s: Warning dst_y_per_pte_row_nom_c %u larger than supported by register format U15.2 %u\n", 1555 __func__, 1556 disp_dlg_regs->dst_y_per_pte_row_nom_c, 1557 (unsigned int)dml_pow(2, 17) - 1); 1558 } 1559 } 1560 1561 disp_dlg_regs->dst_y_per_meta_row_nom_l = (unsigned int) ((double) meta_row_height_l 1562 / (double) vratio_l * dml_pow(2, 2)); 1563 ASSERT(disp_dlg_regs->dst_y_per_meta_row_nom_l < (unsigned int)dml_pow(2, 17)); 1564 1565 disp_dlg_regs->dst_y_per_meta_row_nom_c = disp_dlg_regs->dst_y_per_meta_row_nom_l; // TODO: dcc for 4:2:0 is not supported in dcn1.0. assigned to be the same as _l for now 1566 1567 dml_print( 1568 "DML: Trow: %fus\n", 1569 line_time_in_us * (double)dpte_row_height_l / (double)vratio_l); 1570 1571 disp_dlg_regs->refcyc_per_pte_group_nom_l = (unsigned int) ((double) dpte_row_height_l 1572 / (double) vratio_l * (double) htotal * ref_freq_to_pix_freq 1573 / (double) dpte_groups_per_row_ub_l); 1574 if (disp_dlg_regs->refcyc_per_pte_group_nom_l >= (unsigned int) dml_pow(2, 23)) 1575 disp_dlg_regs->refcyc_per_pte_group_nom_l = dml_pow(2, 23) - 1; 1576 disp_dlg_regs->refcyc_per_meta_chunk_nom_l = (unsigned int) ((double) meta_row_height_l 1577 / (double) vratio_l * (double) htotal * ref_freq_to_pix_freq 1578 / (double) meta_chunks_per_row_ub_l); 1579 if (disp_dlg_regs->refcyc_per_meta_chunk_nom_l >= (unsigned int) dml_pow(2, 23)) 1580 disp_dlg_regs->refcyc_per_meta_chunk_nom_l = dml_pow(2, 23) - 1; 1581 1582 if (dual_plane) { 1583 disp_dlg_regs->refcyc_per_pte_group_nom_c = 1584 (unsigned int) ((double) dpte_row_height_c / (double) vratio_c 1585 * (double) htotal * ref_freq_to_pix_freq 1586 / (double) dpte_groups_per_row_ub_c); 1587 if (disp_dlg_regs->refcyc_per_pte_group_nom_c >= (unsigned int) dml_pow(2, 23)) 1588 disp_dlg_regs->refcyc_per_pte_group_nom_c = dml_pow(2, 23) - 1; 1589 1590 // TODO: Is this the right calculation? Does htotal need to be halved? 1591 disp_dlg_regs->refcyc_per_meta_chunk_nom_c = 1592 (unsigned int) ((double) meta_row_height_c / (double) vratio_c 1593 * (double) htotal * ref_freq_to_pix_freq 1594 / (double) meta_chunks_per_row_ub_c); 1595 if (disp_dlg_regs->refcyc_per_meta_chunk_nom_c >= (unsigned int) dml_pow(2, 23)) 1596 disp_dlg_regs->refcyc_per_meta_chunk_nom_c = dml_pow(2, 23) - 1; 1597 } 1598 1599 disp_dlg_regs->refcyc_per_line_delivery_pre_l = (unsigned int) dml_floor( 1600 refcyc_per_line_delivery_pre_l, 1); 1601 disp_dlg_regs->refcyc_per_line_delivery_l = (unsigned int) dml_floor( 1602 refcyc_per_line_delivery_l, 1); 1603 ASSERT(disp_dlg_regs->refcyc_per_line_delivery_pre_l < (unsigned int)dml_pow(2, 13)); 1604 ASSERT(disp_dlg_regs->refcyc_per_line_delivery_l < (unsigned int)dml_pow(2, 13)); 1605 1606 disp_dlg_regs->refcyc_per_line_delivery_pre_c = (unsigned int) dml_floor( 1607 refcyc_per_line_delivery_pre_c, 1); 1608 disp_dlg_regs->refcyc_per_line_delivery_c = (unsigned int) dml_floor( 1609 refcyc_per_line_delivery_c, 1); 1610 ASSERT(disp_dlg_regs->refcyc_per_line_delivery_pre_c < (unsigned int)dml_pow(2, 13)); 1611 ASSERT(disp_dlg_regs->refcyc_per_line_delivery_c < (unsigned int)dml_pow(2, 13)); 1612 1613 disp_dlg_regs->chunk_hdl_adjust_cur0 = 3; 1614 disp_dlg_regs->dst_y_offset_cur0 = 0; 1615 disp_dlg_regs->chunk_hdl_adjust_cur1 = 3; 1616 disp_dlg_regs->dst_y_offset_cur1 = 0; 1617 1618 disp_dlg_regs->dst_y_delta_drq_limit = 0x7fff; // off 1619 1620 disp_ttu_regs->refcyc_per_req_delivery_pre_l = (unsigned int) (refcyc_per_req_delivery_pre_l 1621 * dml_pow(2, 10)); 1622 disp_ttu_regs->refcyc_per_req_delivery_l = (unsigned int) (refcyc_per_req_delivery_l 1623 * dml_pow(2, 10)); 1624 disp_ttu_regs->refcyc_per_req_delivery_pre_c = (unsigned int) (refcyc_per_req_delivery_pre_c 1625 * dml_pow(2, 10)); 1626 disp_ttu_regs->refcyc_per_req_delivery_c = (unsigned int) (refcyc_per_req_delivery_c 1627 * dml_pow(2, 10)); 1628 disp_ttu_regs->refcyc_per_req_delivery_pre_cur0 = 1629 (unsigned int) (refcyc_per_req_delivery_pre_cur0 * dml_pow(2, 10)); 1630 disp_ttu_regs->refcyc_per_req_delivery_cur0 = (unsigned int) (refcyc_per_req_delivery_cur0 1631 * dml_pow(2, 10)); 1632 disp_ttu_regs->refcyc_per_req_delivery_pre_cur1 = 1633 (unsigned int) (refcyc_per_req_delivery_pre_cur1 * dml_pow(2, 10)); 1634 disp_ttu_regs->refcyc_per_req_delivery_cur1 = (unsigned int) (refcyc_per_req_delivery_cur1 1635 * dml_pow(2, 10)); 1636 disp_ttu_regs->qos_level_low_wm = 0; 1637 ASSERT(disp_ttu_regs->qos_level_low_wm < dml_pow(2, 14)); 1638 disp_ttu_regs->qos_level_high_wm = (unsigned int) (4.0 * (double) htotal 1639 * ref_freq_to_pix_freq); 1640 ASSERT(disp_ttu_regs->qos_level_high_wm < dml_pow(2, 14)); 1641 1642 disp_ttu_regs->qos_level_flip = 14; 1643 disp_ttu_regs->qos_level_fixed_l = 8; 1644 disp_ttu_regs->qos_level_fixed_c = 8; 1645 disp_ttu_regs->qos_level_fixed_cur0 = 8; 1646 disp_ttu_regs->qos_ramp_disable_l = 0; 1647 disp_ttu_regs->qos_ramp_disable_c = 0; 1648 disp_ttu_regs->qos_ramp_disable_cur0 = 0; 1649 1650 disp_ttu_regs->min_ttu_vblank = min_ttu_vblank * refclk_freq_in_mhz; 1651 ASSERT(disp_ttu_regs->min_ttu_vblank < dml_pow(2, 24)); 1652 1653 print__ttu_regs_st(mode_lib, disp_ttu_regs); 1654 print__dlg_regs_st(mode_lib, disp_dlg_regs); 1655 } 1656 1657 void dml21_rq_dlg_get_dlg_reg( 1658 struct display_mode_lib *mode_lib, 1659 display_dlg_regs_st *dlg_regs, 1660 display_ttu_regs_st *ttu_regs, 1661 const display_e2e_pipe_params_st *e2e_pipe_param, 1662 const unsigned int num_pipes, 1663 const unsigned int pipe_idx, 1664 const bool cstate_en, 1665 const bool pstate_en, 1666 const bool vm_en, 1667 const bool ignore_viewport_pos, 1668 const bool immediate_flip_support) 1669 { 1670 display_rq_params_st rq_param = {0}; 1671 display_dlg_sys_params_st dlg_sys_param = {0}; 1672 1673 // Get watermark and Tex. 1674 dlg_sys_param.t_urg_wm_us = get_wm_urgent(mode_lib, e2e_pipe_param, num_pipes); 1675 dlg_sys_param.deepsleep_dcfclk_mhz = get_clk_dcf_deepsleep( 1676 mode_lib, 1677 e2e_pipe_param, 1678 num_pipes); 1679 dlg_sys_param.t_extra_us = get_urgent_extra_latency(mode_lib, e2e_pipe_param, num_pipes); 1680 dlg_sys_param.mem_trip_us = get_wm_memory_trip(mode_lib, e2e_pipe_param, num_pipes); 1681 dlg_sys_param.t_mclk_wm_us = get_wm_dram_clock_change(mode_lib, e2e_pipe_param, num_pipes); 1682 dlg_sys_param.t_sr_wm_us = get_wm_stutter_enter_exit(mode_lib, e2e_pipe_param, num_pipes); 1683 dlg_sys_param.total_flip_bw = get_total_immediate_flip_bw( 1684 mode_lib, 1685 e2e_pipe_param, 1686 num_pipes); 1687 dlg_sys_param.total_flip_bytes = get_total_immediate_flip_bytes( 1688 mode_lib, 1689 e2e_pipe_param, 1690 num_pipes); 1691 1692 print__dlg_sys_params_st(mode_lib, &dlg_sys_param); 1693 1694 // system parameter calculation done 1695 1696 dml_print("DML_DLG: Calculation for pipe[%d] start\n\n", pipe_idx); 1697 dml_rq_dlg_get_rq_params(mode_lib, &rq_param, &e2e_pipe_param[pipe_idx].pipe); 1698 dml_rq_dlg_get_dlg_params( 1699 mode_lib, 1700 e2e_pipe_param, 1701 num_pipes, 1702 pipe_idx, 1703 dlg_regs, 1704 ttu_regs, 1705 &rq_param.dlg, 1706 &dlg_sys_param, 1707 cstate_en, 1708 pstate_en); 1709 dml_print("DML_DLG: Calculation for pipe[%d] end\n", pipe_idx); 1710 } 1711 1712 static void calculate_ttu_cursor( 1713 struct display_mode_lib *mode_lib, 1714 double *refcyc_per_req_delivery_pre_cur, 1715 double *refcyc_per_req_delivery_cur, 1716 double refclk_freq_in_mhz, 1717 double ref_freq_to_pix_freq, 1718 double hscale_pixel_rate_l, 1719 double hscl_ratio, 1720 double vratio_pre_l, 1721 double vratio_l, 1722 unsigned int cur_width, 1723 enum cursor_bpp cur_bpp) 1724 { 1725 unsigned int cur_src_width = cur_width; 1726 unsigned int cur_req_size = 0; 1727 unsigned int cur_req_width = 0; 1728 double cur_width_ub = 0.0; 1729 double cur_req_per_width = 0.0; 1730 double hactive_cur = 0.0; 1731 1732 ASSERT(cur_src_width <= 256); 1733 1734 *refcyc_per_req_delivery_pre_cur = 0.0; 1735 *refcyc_per_req_delivery_cur = 0.0; 1736 if (cur_src_width > 0) { 1737 unsigned int cur_bit_per_pixel = 0; 1738 1739 if (cur_bpp == dm_cur_2bit) { 1740 cur_req_size = 64; // byte 1741 cur_bit_per_pixel = 2; 1742 } else { // 32bit 1743 cur_bit_per_pixel = 32; 1744 if (cur_src_width >= 1 && cur_src_width <= 16) 1745 cur_req_size = 64; 1746 else if (cur_src_width >= 17 && cur_src_width <= 31) 1747 cur_req_size = 128; 1748 else 1749 cur_req_size = 256; 1750 } 1751 1752 cur_req_width = (double) cur_req_size / ((double) cur_bit_per_pixel / 8.0); 1753 cur_width_ub = dml_ceil((double) cur_src_width / (double) cur_req_width, 1) 1754 * (double) cur_req_width; 1755 cur_req_per_width = cur_width_ub / (double) cur_req_width; 1756 hactive_cur = (double) cur_src_width / hscl_ratio; // FIXME: oswin to think about what to do for cursor 1757 1758 if (vratio_pre_l <= 1.0) { 1759 *refcyc_per_req_delivery_pre_cur = hactive_cur * ref_freq_to_pix_freq 1760 / (double) cur_req_per_width; 1761 } else { 1762 *refcyc_per_req_delivery_pre_cur = (double) refclk_freq_in_mhz 1763 * (double) cur_src_width / hscale_pixel_rate_l 1764 / (double) cur_req_per_width; 1765 } 1766 1767 ASSERT(*refcyc_per_req_delivery_pre_cur < dml_pow(2, 13)); 1768 1769 if (vratio_l <= 1.0) { 1770 *refcyc_per_req_delivery_cur = hactive_cur * ref_freq_to_pix_freq 1771 / (double) cur_req_per_width; 1772 } else { 1773 *refcyc_per_req_delivery_cur = (double) refclk_freq_in_mhz 1774 * (double) cur_src_width / hscale_pixel_rate_l 1775 / (double) cur_req_per_width; 1776 } 1777 1778 dml_print( 1779 "DML_DLG: %s: cur_req_width = %d\n", 1780 __func__, 1781 cur_req_width); 1782 dml_print( 1783 "DML_DLG: %s: cur_width_ub = %3.2f\n", 1784 __func__, 1785 cur_width_ub); 1786 dml_print( 1787 "DML_DLG: %s: cur_req_per_width = %3.2f\n", 1788 __func__, 1789 cur_req_per_width); 1790 dml_print( 1791 "DML_DLG: %s: hactive_cur = %3.2f\n", 1792 __func__, 1793 hactive_cur); 1794 dml_print( 1795 "DML_DLG: %s: refcyc_per_req_delivery_pre_cur = %3.2f\n", 1796 __func__, 1797 *refcyc_per_req_delivery_pre_cur); 1798 dml_print( 1799 "DML_DLG: %s: refcyc_per_req_delivery_cur = %3.2f\n", 1800 __func__, 1801 *refcyc_per_req_delivery_cur); 1802 1803 ASSERT(*refcyc_per_req_delivery_cur < dml_pow(2, 13)); 1804 } 1805 } 1806 1807