drivers/gpu/drm/radeon/evergreen_blit_kms.c

   1 /*
   2  * Copyright 2010 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 (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE COPYRIGHT HOLDER(S) AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
  19  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  20  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  21  * DEALINGS IN THE SOFTWARE.
  22  *
  23  * Authors:
  24  *     Alex Deucher <alexander.deucher@amd.com>
  25  */
  26
  27 #include "drmP.h"
  28 #include "drm.h"
  29 #include "radeon_drm.h"
  30 #include "radeon.h"
  31
  32 #include "evergreend.h"
  33 #include "evergreen_blit_shaders.h"
  34 #include "cayman_blit_shaders.h"
  35
  36 #define DI_PT_RECTLIST        0x11
  37 #define DI_INDEX_SIZE_16_BIT  0x0
  38 #define DI_SRC_SEL_AUTO_INDEX 0x2
  39
  40 #define FMT_8                 0x1
  41 #define FMT_5_6_5             0x8
  42 #define FMT_8_8_8_8           0x1a
  43 #define COLOR_8               0x1
  44 #define COLOR_5_6_5           0x8
  45 #define COLOR_8_8_8_8         0x1a
  46
  47 /* emits 17 */
  48 static void
  49 set_render_target(struct radeon_device *rdev, int format,
  50                   int w, int h, u64 gpu_addr)
  51 {
  52         u32 cb_color_info;
  53         int pitch, slice;
  54
  55         h = ALIGN(h, 8);
  56         if (h < 8)
  57                 h = 8;
  58
  59         cb_color_info = ((format << 2) | (1 << 24) | (1 << 8));
  60         pitch = (w / 8) - 1;
  61         slice = ((w * h) / 64) - 1;
  62
  63         radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONTEXT_REG, 15));
  64         radeon_ring_write(rdev, (CB_COLOR0_BASE - PACKET3_SET_CONTEXT_REG_START) >> 2);
  65         radeon_ring_write(rdev, gpu_addr >> 8);
  66         radeon_ring_write(rdev, pitch);
  67         radeon_ring_write(rdev, slice);
  68         radeon_ring_write(rdev, 0);
  69         radeon_ring_write(rdev, cb_color_info);
  70         radeon_ring_write(rdev, (1 << 4));
  71         radeon_ring_write(rdev, (w - 1) | ((h - 1) << 16));
  72         radeon_ring_write(rdev, 0);
  73         radeon_ring_write(rdev, 0);
  74         radeon_ring_write(rdev, 0);
  75         radeon_ring_write(rdev, 0);
  76         radeon_ring_write(rdev, 0);
  77         radeon_ring_write(rdev, 0);
  78         radeon_ring_write(rdev, 0);
  79         radeon_ring_write(rdev, 0);
  80 }
  81
  82 /* emits 5dw */
  83 static void
  84 cp_set_surface_sync(struct radeon_device *rdev,
  85                     u32 sync_type, u32 size,
  86                     u64 mc_addr)
  87 {
  88         u32 cp_coher_size;
  89
  90         if (size == 0xffffffff)
  91                 cp_coher_size = 0xffffffff;
  92         else
  93                 cp_coher_size = ((size + 255) >> 8);
  94
  95         radeon_ring_write(rdev, PACKET3(PACKET3_SURFACE_SYNC, 3));
  96         radeon_ring_write(rdev, sync_type);
  97         radeon_ring_write(rdev, cp_coher_size);
  98         radeon_ring_write(rdev, mc_addr >> 8);
  99         radeon_ring_write(rdev, 10); /* poll interval */
 100 }
 101
 102 /* emits 11dw + 1 surface sync = 16dw */
 103 static void
 104 set_shaders(struct radeon_device *rdev)
 105 {
 106         u64 gpu_addr;
 107
 108         /* VS */
 109         gpu_addr = rdev->r600_blit.shader_gpu_addr + rdev->r600_blit.vs_offset;
 110         radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONTEXT_REG, 3));
 111         radeon_ring_write(rdev, (SQ_PGM_START_VS - PACKET3_SET_CONTEXT_REG_START) >> 2);
 112         radeon_ring_write(rdev, gpu_addr >> 8);
 113         radeon_ring_write(rdev, 2);
 114         radeon_ring_write(rdev, 0);
 115
 116         /* PS */
 117         gpu_addr = rdev->r600_blit.shader_gpu_addr + rdev->r600_blit.ps_offset;
 118         radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONTEXT_REG, 4));
 119         radeon_ring_write(rdev, (SQ_PGM_START_PS - PACKET3_SET_CONTEXT_REG_START) >> 2);
 120         radeon_ring_write(rdev, gpu_addr >> 8);
 121         radeon_ring_write(rdev, 1);
 122         radeon_ring_write(rdev, 0);
 123         radeon_ring_write(rdev, 2);
 124
 125         gpu_addr = rdev->r600_blit.shader_gpu_addr + rdev->r600_blit.vs_offset;
 126         cp_set_surface_sync(rdev, PACKET3_SH_ACTION_ENA, 512, gpu_addr);
 127 }
 128
 129 /* emits 10 + 1 sync (5) = 15 */
 130 static void
 131 set_vtx_resource(struct radeon_device *rdev, u64 gpu_addr)
 132 {
 133         u32 sq_vtx_constant_word2, sq_vtx_constant_word3;
 134
 135         /* high addr, stride */
 136         sq_vtx_constant_word2 = ((upper_32_bits(gpu_addr) & 0xff) | (16 << 8));
 137 #ifdef __BIG_ENDIAN
 138         sq_vtx_constant_word2 |= (2 << 30);
 139 #endif
 140         /* xyzw swizzles */
 141         sq_vtx_constant_word3 = (0 << 3) | (1 << 6) | (2 << 9) | (3 << 12);
 142
 143         radeon_ring_write(rdev, PACKET3(PACKET3_SET_RESOURCE, 8));
 144         radeon_ring_write(rdev, 0x580);
 145         radeon_ring_write(rdev, gpu_addr & 0xffffffff);
 146         radeon_ring_write(rdev, 48 - 1); /* size */
 147         radeon_ring_write(rdev, sq_vtx_constant_word2);
 148         radeon_ring_write(rdev, sq_vtx_constant_word3);
 149         radeon_ring_write(rdev, 0);
 150         radeon_ring_write(rdev, 0);
 151         radeon_ring_write(rdev, 0);
 152         radeon_ring_write(rdev, SQ_TEX_VTX_VALID_BUFFER << 30);
 153
 154         if ((rdev->family == CHIP_CEDAR) ||
 155             (rdev->family == CHIP_PALM) ||
 156             (rdev->family == CHIP_SUMO) ||
 157             (rdev->family == CHIP_SUMO2) ||
 158             (rdev->family == CHIP_CAICOS))
 159                 cp_set_surface_sync(rdev,
 160                                     PACKET3_TC_ACTION_ENA, 48, gpu_addr);
 161         else
 162                 cp_set_surface_sync(rdev,
 163                                     PACKET3_VC_ACTION_ENA, 48, gpu_addr);
 164
 165 }
 166
 167 /* emits 10 */
 168 static void
 169 set_tex_resource(struct radeon_device *rdev,
 170                  int format, int w, int h, int pitch,
 171                  u64 gpu_addr)
 172 {
 173         u32 sq_tex_resource_word0, sq_tex_resource_word1;
 174         u32 sq_tex_resource_word4, sq_tex_resource_word7;
 175
 176         if (h < 1)
 177                 h = 1;
 178
 179         sq_tex_resource_word0 = (1 << 0); /* 2D */
 180         sq_tex_resource_word0 |= ((((pitch >> 3) - 1) << 6) |
 181                                   ((w - 1) << 18));
 182         sq_tex_resource_word1 = ((h - 1) << 0) | (1 << 28);
 183         /* xyzw swizzles */
 184         sq_tex_resource_word4 = (0 << 16) | (1 << 19) | (2 << 22) | (3 << 25);
 185
 186         sq_tex_resource_word7 = format | (SQ_TEX_VTX_VALID_TEXTURE << 30);
 187
 188         radeon_ring_write(rdev, PACKET3(PACKET3_SET_RESOURCE, 8));
 189         radeon_ring_write(rdev, 0);
 190         radeon_ring_write(rdev, sq_tex_resource_word0);
 191         radeon_ring_write(rdev, sq_tex_resource_word1);
 192         radeon_ring_write(rdev, gpu_addr >> 8);
 193         radeon_ring_write(rdev, gpu_addr >> 8);
 194         radeon_ring_write(rdev, sq_tex_resource_word4);
 195         radeon_ring_write(rdev, 0);
 196         radeon_ring_write(rdev, 0);
 197         radeon_ring_write(rdev, sq_tex_resource_word7);
 198 }
 199
 200 /* emits 12 */
 201 static void
 202 set_scissors(struct radeon_device *rdev, int x1, int y1,
 203              int x2, int y2)
 204 {
 205         /* workaround some hw bugs */
 206         if (x2 == 0)
 207                 x1 = 1;
 208         if (y2 == 0)
 209                 y1 = 1;
 210         if (rdev->family == CHIP_CAYMAN) {
 211                 if ((x2 == 1) && (y2 == 1))
 212                         x2 = 2;
 213         }
 214
 215         radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
 216         radeon_ring_write(rdev, (PA_SC_SCREEN_SCISSOR_TL - PACKET3_SET_CONTEXT_REG_START) >> 2);
 217         radeon_ring_write(rdev, (x1 << 0) | (y1 << 16));
 218         radeon_ring_write(rdev, (x2 << 0) | (y2 << 16));
 219
 220         radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
 221         radeon_ring_write(rdev, (PA_SC_GENERIC_SCISSOR_TL - PACKET3_SET_CONTEXT_REG_START) >> 2);
 222         radeon_ring_write(rdev, (x1 << 0) | (y1 << 16) | (1 << 31));
 223         radeon_ring_write(rdev, (x2 << 0) | (y2 << 16));
 224
 225         radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
 226         radeon_ring_write(rdev, (PA_SC_WINDOW_SCISSOR_TL - PACKET3_SET_CONTEXT_REG_START) >> 2);
 227         radeon_ring_write(rdev, (x1 << 0) | (y1 << 16) | (1 << 31));
 228         radeon_ring_write(rdev, (x2 << 0) | (y2 << 16));
 229 }
 230
 231 /* emits 10 */
 232 static void
 233 draw_auto(struct radeon_device *rdev)
 234 {
 235         radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONFIG_REG, 1));
 236         radeon_ring_write(rdev, (VGT_PRIMITIVE_TYPE - PACKET3_SET_CONFIG_REG_START) >> 2);
 237         radeon_ring_write(rdev, DI_PT_RECTLIST);
 238
 239         radeon_ring_write(rdev, PACKET3(PACKET3_INDEX_TYPE, 0));
 240         radeon_ring_write(rdev,
 241 #ifdef __BIG_ENDIAN
 242                           (2 << 2) |
 243 #endif
 244                           DI_INDEX_SIZE_16_BIT);
 245
 246         radeon_ring_write(rdev, PACKET3(PACKET3_NUM_INSTANCES, 0));
 247         radeon_ring_write(rdev, 1);
 248
 249         radeon_ring_write(rdev, PACKET3(PACKET3_DRAW_INDEX_AUTO, 1));
 250         radeon_ring_write(rdev, 3);
 251         radeon_ring_write(rdev, DI_SRC_SEL_AUTO_INDEX);
 252
 253 }
 254
 255 /* emits 39 */
 256 static void
 257 set_default_state(struct radeon_device *rdev)
 258 {
 259         u32 sq_config, sq_gpr_resource_mgmt_1, sq_gpr_resource_mgmt_2, sq_gpr_resource_mgmt_3;
 260         u32 sq_thread_resource_mgmt, sq_thread_resource_mgmt_2;
 261         u32 sq_stack_resource_mgmt_1, sq_stack_resource_mgmt_2, sq_stack_resource_mgmt_3;
 262         int num_ps_gprs, num_vs_gprs, num_temp_gprs;
 263         int num_gs_gprs, num_es_gprs, num_hs_gprs, num_ls_gprs;
 264         int num_ps_threads, num_vs_threads, num_gs_threads, num_es_threads;
 265         int num_hs_threads, num_ls_threads;
 266         int num_ps_stack_entries, num_vs_stack_entries, num_gs_stack_entries, num_es_stack_entries;
 267         int num_hs_stack_entries, num_ls_stack_entries;
 268         u64 gpu_addr;
 269         int dwords;
 270
 271         /* set clear context state */
 272         radeon_ring_write(rdev, PACKET3(PACKET3_CLEAR_STATE, 0));
 273         radeon_ring_write(rdev, 0);
 274
 275         if (rdev->family < CHIP_CAYMAN) {
 276                 switch (rdev->family) {
 277                 case CHIP_CEDAR:
 278                 default:
 279                         num_ps_gprs = 93;
 280                         num_vs_gprs = 46;
 281                         num_temp_gprs = 4;
 282                         num_gs_gprs = 31;
 283                         num_es_gprs = 31;
 284                         num_hs_gprs = 23;
 285                         num_ls_gprs = 23;
 286                         num_ps_threads = 96;
 287                         num_vs_threads = 16;
 288                         num_gs_threads = 16;
 289                         num_es_threads = 16;
 290                         num_hs_threads = 16;
 291                         num_ls_threads = 16;
 292                         num_ps_stack_entries = 42;
 293                         num_vs_stack_entries = 42;
 294                         num_gs_stack_entries = 42;
 295                         num_es_stack_entries = 42;
 296                         num_hs_stack_entries = 42;
 297                         num_ls_stack_entries = 42;
 298                         break;
 299                 case CHIP_REDWOOD:
 300                         num_ps_gprs = 93;
 301                         num_vs_gprs = 46;
 302                         num_temp_gprs = 4;
 303                         num_gs_gprs = 31;
 304                         num_es_gprs = 31;
 305                         num_hs_gprs = 23;
 306                         num_ls_gprs = 23;
 307                         num_ps_threads = 128;
 308                         num_vs_threads = 20;
 309                         num_gs_threads = 20;
 310                         num_es_threads = 20;
 311                         num_hs_threads = 20;
 312                         num_ls_threads = 20;
 313                         num_ps_stack_entries = 42;
 314                         num_vs_stack_entries = 42;
 315                         num_gs_stack_entries = 42;
 316                         num_es_stack_entries = 42;
 317                         num_hs_stack_entries = 42;
 318                         num_ls_stack_entries = 42;
 319                         break;
 320                 case CHIP_JUNIPER:
 321                         num_ps_gprs = 93;
 322                         num_vs_gprs = 46;
 323                         num_temp_gprs = 4;
 324                         num_gs_gprs = 31;
 325                         num_es_gprs = 31;
 326                         num_hs_gprs = 23;
 327                         num_ls_gprs = 23;
 328                         num_ps_threads = 128;
 329                         num_vs_threads = 20;
 330                         num_gs_threads = 20;
 331                         num_es_threads = 20;
 332                         num_hs_threads = 20;
 333                         num_ls_threads = 20;
 334                         num_ps_stack_entries = 85;
 335                         num_vs_stack_entries = 85;
 336                         num_gs_stack_entries = 85;
 337                         num_es_stack_entries = 85;
 338                         num_hs_stack_entries = 85;
 339                         num_ls_stack_entries = 85;
 340                         break;
 341                 case CHIP_CYPRESS:
 342                 case CHIP_HEMLOCK:
 343                         num_ps_gprs = 93;
 344                         num_vs_gprs = 46;
 345                         num_temp_gprs = 4;
 346                         num_gs_gprs = 31;
 347                         num_es_gprs = 31;
 348                         num_hs_gprs = 23;
 349                         num_ls_gprs = 23;
 350                         num_ps_threads = 128;
 351                         num_vs_threads = 20;
 352                         num_gs_threads = 20;
 353                         num_es_threads = 20;
 354                         num_hs_threads = 20;
 355                         num_ls_threads = 20;
 356                         num_ps_stack_entries = 85;
 357                         num_vs_stack_entries = 85;
 358                         num_gs_stack_entries = 85;
 359                         num_es_stack_entries = 85;
 360                         num_hs_stack_entries = 85;
 361                         num_ls_stack_entries = 85;
 362                         break;
 363                 case CHIP_PALM:
 364                         num_ps_gprs = 93;
 365                         num_vs_gprs = 46;
 366                         num_temp_gprs = 4;
 367                         num_gs_gprs = 31;
 368                         num_es_gprs = 31;
 369                         num_hs_gprs = 23;
 370                         num_ls_gprs = 23;
 371                         num_ps_threads = 96;
 372                         num_vs_threads = 16;
 373                         num_gs_threads = 16;
 374                         num_es_threads = 16;
 375                         num_hs_threads = 16;
 376                         num_ls_threads = 16;
 377                         num_ps_stack_entries = 42;
 378                         num_vs_stack_entries = 42;
 379                         num_gs_stack_entries = 42;
 380                         num_es_stack_entries = 42;
 381                         num_hs_stack_entries = 42;
 382                         num_ls_stack_entries = 42;
 383                         break;
 384                 case CHIP_SUMO:
 385                         num_ps_gprs = 93;
 386                         num_vs_gprs = 46;
 387                         num_temp_gprs = 4;
 388                         num_gs_gprs = 31;
 389                         num_es_gprs = 31;
 390                         num_hs_gprs = 23;
 391                         num_ls_gprs = 23;
 392                         num_ps_threads = 96;
 393                         num_vs_threads = 25;
 394                         num_gs_threads = 25;
 395                         num_es_threads = 25;
 396                         num_hs_threads = 25;
 397                         num_ls_threads = 25;
 398                         num_ps_stack_entries = 42;
 399                         num_vs_stack_entries = 42;
 400                         num_gs_stack_entries = 42;
 401                         num_es_stack_entries = 42;
 402                         num_hs_stack_entries = 42;
 403                         num_ls_stack_entries = 42;
 404                         break;
 405                 case CHIP_SUMO2:
 406                         num_ps_gprs = 93;
 407                         num_vs_gprs = 46;
 408                         num_temp_gprs = 4;
 409                         num_gs_gprs = 31;
 410                         num_es_gprs = 31;
 411                         num_hs_gprs = 23;
 412                         num_ls_gprs = 23;
 413                         num_ps_threads = 96;
 414                         num_vs_threads = 25;
 415                         num_gs_threads = 25;
 416                         num_es_threads = 25;
 417                         num_hs_threads = 25;
 418                         num_ls_threads = 25;
 419                         num_ps_stack_entries = 85;
 420                         num_vs_stack_entries = 85;
 421                         num_gs_stack_entries = 85;
 422                         num_es_stack_entries = 85;
 423                         num_hs_stack_entries = 85;
 424                         num_ls_stack_entries = 85;
 425                         break;
 426                 case CHIP_BARTS:
 427                         num_ps_gprs = 93;
 428                         num_vs_gprs = 46;
 429                         num_temp_gprs = 4;
 430                         num_gs_gprs = 31;
 431                         num_es_gprs = 31;
 432                         num_hs_gprs = 23;
 433                         num_ls_gprs = 23;
 434                         num_ps_threads = 128;
 435                         num_vs_threads = 20;
 436                         num_gs_threads = 20;
 437                         num_es_threads = 20;
 438                         num_hs_threads = 20;
 439                         num_ls_threads = 20;
 440                         num_ps_stack_entries = 85;
 441                         num_vs_stack_entries = 85;
 442                         num_gs_stack_entries = 85;
 443                         num_es_stack_entries = 85;
 444                         num_hs_stack_entries = 85;
 445                         num_ls_stack_entries = 85;
 446                         break;
 447                 case CHIP_TURKS:
 448                         num_ps_gprs = 93;
 449                         num_vs_gprs = 46;
 450                         num_temp_gprs = 4;
 451                         num_gs_gprs = 31;
 452                         num_es_gprs = 31;
 453                         num_hs_gprs = 23;
 454                         num_ls_gprs = 23;
 455                         num_ps_threads = 128;
 456                         num_vs_threads = 20;
 457                         num_gs_threads = 20;
 458                         num_es_threads = 20;
 459                         num_hs_threads = 20;
 460                         num_ls_threads = 20;
 461                         num_ps_stack_entries = 42;
 462                         num_vs_stack_entries = 42;
 463                         num_gs_stack_entries = 42;
 464                         num_es_stack_entries = 42;
 465                         num_hs_stack_entries = 42;
 466                         num_ls_stack_entries = 42;
 467                         break;
 468                 case CHIP_CAICOS:
 469                         num_ps_gprs = 93;
 470                         num_vs_gprs = 46;
 471                         num_temp_gprs = 4;
 472                         num_gs_gprs = 31;
 473                         num_es_gprs = 31;
 474                         num_hs_gprs = 23;
 475                         num_ls_gprs = 23;
 476                         num_ps_threads = 128;
 477                         num_vs_threads = 10;
 478                         num_gs_threads = 10;
 479                         num_es_threads = 10;
 480                         num_hs_threads = 10;
 481                         num_ls_threads = 10;
 482                         num_ps_stack_entries = 42;
 483                         num_vs_stack_entries = 42;
 484                         num_gs_stack_entries = 42;
 485                         num_es_stack_entries = 42;
 486                         num_hs_stack_entries = 42;
 487                         num_ls_stack_entries = 42;
 488                         break;
 489                 }
 490
 491                 if ((rdev->family == CHIP_CEDAR) ||
 492                     (rdev->family == CHIP_PALM) ||
 493                     (rdev->family == CHIP_SUMO) ||
 494                     (rdev->family == CHIP_SUMO2) ||
 495                     (rdev->family == CHIP_CAICOS))
 496                         sq_config = 0;
 497                 else
 498                         sq_config = VC_ENABLE;
 499
 500                 sq_config |= (EXPORT_SRC_C |
 501                               CS_PRIO(0) |
 502                               LS_PRIO(0) |
 503                               HS_PRIO(0) |
 504                               PS_PRIO(0) |
 505                               VS_PRIO(1) |
 506                               GS_PRIO(2) |
 507                               ES_PRIO(3));
 508
 509                 sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(num_ps_gprs) |
 510                                           NUM_VS_GPRS(num_vs_gprs) |
 511                                           NUM_CLAUSE_TEMP_GPRS(num_temp_gprs));
 512                 sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(num_gs_gprs) |
 513                                           NUM_ES_GPRS(num_es_gprs));
 514                 sq_gpr_resource_mgmt_3 = (NUM_HS_GPRS(num_hs_gprs) |
 515                                           NUM_LS_GPRS(num_ls_gprs));
 516                 sq_thread_resource_mgmt = (NUM_PS_THREADS(num_ps_threads) |
 517                                            NUM_VS_THREADS(num_vs_threads) |
 518                                            NUM_GS_THREADS(num_gs_threads) |
 519                                            NUM_ES_THREADS(num_es_threads));
 520                 sq_thread_resource_mgmt_2 = (NUM_HS_THREADS(num_hs_threads) |
 521                                              NUM_LS_THREADS(num_ls_threads));
 522                 sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(num_ps_stack_entries) |
 523                                             NUM_VS_STACK_ENTRIES(num_vs_stack_entries));
 524                 sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(num_gs_stack_entries) |
 525                                             NUM_ES_STACK_ENTRIES(num_es_stack_entries));
 526                 sq_stack_resource_mgmt_3 = (NUM_HS_STACK_ENTRIES(num_hs_stack_entries) |
 527                                             NUM_LS_STACK_ENTRIES(num_ls_stack_entries));
 528
 529                 /* disable dyn gprs */
 530                 radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONFIG_REG, 1));
 531                 radeon_ring_write(rdev, (SQ_DYN_GPR_CNTL_PS_FLUSH_REQ - PACKET3_SET_CONFIG_REG_START) >> 2);
 532                 radeon_ring_write(rdev, 0);
 533
 534                 /* setup LDS */
 535                 radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONFIG_REG, 1));
 536                 radeon_ring_write(rdev, (SQ_LDS_RESOURCE_MGMT - PACKET3_SET_CONFIG_REG_START) >> 2);
 537                 radeon_ring_write(rdev, 0x10001000);
 538
 539                 /* SQ config */
 540                 radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONFIG_REG, 11));
 541                 radeon_ring_write(rdev, (SQ_CONFIG - PACKET3_SET_CONFIG_REG_START) >> 2);
 542                 radeon_ring_write(rdev, sq_config);
 543                 radeon_ring_write(rdev, sq_gpr_resource_mgmt_1);
 544                 radeon_ring_write(rdev, sq_gpr_resource_mgmt_2);
 545                 radeon_ring_write(rdev, sq_gpr_resource_mgmt_3);
 546                 radeon_ring_write(rdev, 0);
 547                 radeon_ring_write(rdev, 0);
 548                 radeon_ring_write(rdev, sq_thread_resource_mgmt);
 549                 radeon_ring_write(rdev, sq_thread_resource_mgmt_2);
 550                 radeon_ring_write(rdev, sq_stack_resource_mgmt_1);
 551                 radeon_ring_write(rdev, sq_stack_resource_mgmt_2);
 552                 radeon_ring_write(rdev, sq_stack_resource_mgmt_3);
 553         }
 554
 555         /* CONTEXT_CONTROL */
 556         radeon_ring_write(rdev, 0xc0012800);
 557         radeon_ring_write(rdev, 0x80000000);
 558         radeon_ring_write(rdev, 0x80000000);
 559
 560         /* SQ_VTX_BASE_VTX_LOC */
 561         radeon_ring_write(rdev, 0xc0026f00);
 562         radeon_ring_write(rdev, 0x00000000);
 563         radeon_ring_write(rdev, 0x00000000);
 564         radeon_ring_write(rdev, 0x00000000);
 565
 566         /* SET_SAMPLER */
 567         radeon_ring_write(rdev, 0xc0036e00);
 568         radeon_ring_write(rdev, 0x00000000);
 569         radeon_ring_write(rdev, 0x00000012);
 570         radeon_ring_write(rdev, 0x00000000);
 571         radeon_ring_write(rdev, 0x00000000);
 572
 573         /* set to DX10/11 mode */
 574         radeon_ring_write(rdev, PACKET3(PACKET3_MODE_CONTROL, 0));
 575         radeon_ring_write(rdev, 1);
 576
 577         /* emit an IB pointing at default state */
 578         dwords = ALIGN(rdev->r600_blit.state_len, 0x10);
 579         gpu_addr = rdev->r600_blit.shader_gpu_addr + rdev->r600_blit.state_offset;
 580         radeon_ring_write(rdev, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
 581         radeon_ring_write(rdev, gpu_addr & 0xFFFFFFFC);
 582         radeon_ring_write(rdev, upper_32_bits(gpu_addr) & 0xFF);
 583         radeon_ring_write(rdev, dwords);
 584
 585 }
 586
 587 static inline uint32_t i2f(uint32_t input)
 588 {
 589         u32 result, i, exponent, fraction;
 590
 591         if ((input & 0x3fff) == 0)
 592                 result = 0; /* 0 is a special case */
 593         else {
 594                 exponent = 140; /* exponent biased by 127; */
 595                 fraction = (input & 0x3fff) << 10; /* cheat and only
 596                                                       handle numbers below 2^^15 */
 597                 for (i = 0; i < 14; i++) {
 598                         if (fraction & 0x800000)
 599                                 break;
 600                         else {
 601                                 fraction = fraction << 1; /* keep
 602                                                              shifting left until top bit = 1 */
 603                                 exponent = exponent - 1;
 604                         }
 605                 }
 606                 result = exponent << 23 | (fraction & 0x7fffff); /* mask
 607                                                                     off top bit; assumed 1 */
 608         }
 609         return result;
 610 }
 611
 612 int evergreen_blit_init(struct radeon_device *rdev)
 613 {
 614         u32 obj_size;
 615         int i, r, dwords;
 616         void *ptr;
 617         u32 packet2s[16];
 618         int num_packet2s = 0;
 619
 620         /* pin copy shader into vram if already initialized */
 621         if (rdev->r600_blit.shader_obj)
 622                 goto done;
 623
 624         mutex_init(&rdev->r600_blit.mutex);
 625         rdev->r600_blit.state_offset = 0;
 626
 627         if (rdev->family < CHIP_CAYMAN)
 628                 rdev->r600_blit.state_len = evergreen_default_size;
 629         else
 630                 rdev->r600_blit.state_len = cayman_default_size;
 631
 632         dwords = rdev->r600_blit.state_len;
 633         while (dwords & 0xf) {
 634                 packet2s[num_packet2s++] = cpu_to_le32(PACKET2(0));
 635                 dwords++;
 636         }
 637
 638         obj_size = dwords * 4;
 639         obj_size = ALIGN(obj_size, 256);
 640
 641         rdev->r600_blit.vs_offset = obj_size;
 642         if (rdev->family < CHIP_CAYMAN)
 643                 obj_size += evergreen_vs_size * 4;
 644         else
 645                 obj_size += cayman_vs_size * 4;
 646         obj_size = ALIGN(obj_size, 256);
 647
 648         rdev->r600_blit.ps_offset = obj_size;
 649         if (rdev->family < CHIP_CAYMAN)
 650                 obj_size += evergreen_ps_size * 4;
 651         else
 652                 obj_size += cayman_ps_size * 4;
 653         obj_size = ALIGN(obj_size, 256);
 654
 655         r = radeon_bo_create(rdev, obj_size, PAGE_SIZE, true, RADEON_GEM_DOMAIN_VRAM,
 656                                 &rdev->r600_blit.shader_obj);
 657         if (r) {
 658                 DRM_ERROR("evergreen failed to allocate shader\n");
 659                 return r;
 660         }
 661
 662         DRM_DEBUG("evergreen blit allocated bo %08x vs %08x ps %08x\n",
 663                   obj_size,
 664                   rdev->r600_blit.vs_offset, rdev->r600_blit.ps_offset);
 665
 666         r = radeon_bo_reserve(rdev->r600_blit.shader_obj, false);
 667         if (unlikely(r != 0))
 668                 return r;
 669         r = radeon_bo_kmap(rdev->r600_blit.shader_obj, &ptr);
 670         if (r) {
 671                 DRM_ERROR("failed to map blit object %d\n", r);
 672                 return r;
 673         }
 674
 675         if (rdev->family < CHIP_CAYMAN) {
 676                 memcpy_toio(ptr + rdev->r600_blit.state_offset,
 677                             evergreen_default_state, rdev->r600_blit.state_len * 4);
 678
 679                 if (num_packet2s)
 680                         memcpy_toio(ptr + rdev->r600_blit.state_offset + (rdev->r600_blit.state_len * 4),
 681                                     packet2s, num_packet2s * 4);
 682                 for (i = 0; i < evergreen_vs_size; i++)
 683                         *(u32 *)((unsigned long)ptr + rdev->r600_blit.vs_offset + i * 4) = cpu_to_le32(evergreen_vs[i]);
 684                 for (i = 0; i < evergreen_ps_size; i++)
 685                         *(u32 *)((unsigned long)ptr + rdev->r600_blit.ps_offset + i * 4) = cpu_to_le32(evergreen_ps[i]);
 686         } else {
 687                 memcpy_toio(ptr + rdev->r600_blit.state_offset,
 688                             cayman_default_state, rdev->r600_blit.state_len * 4);
 689
 690                 if (num_packet2s)
 691                         memcpy_toio(ptr + rdev->r600_blit.state_offset + (rdev->r600_blit.state_len * 4),
 692                                     packet2s, num_packet2s * 4);
 693                 for (i = 0; i < cayman_vs_size; i++)
 694                         *(u32 *)((unsigned long)ptr + rdev->r600_blit.vs_offset + i * 4) = cpu_to_le32(cayman_vs[i]);
 695                 for (i = 0; i < cayman_ps_size; i++)
 696                         *(u32 *)((unsigned long)ptr + rdev->r600_blit.ps_offset + i * 4) = cpu_to_le32(cayman_ps[i]);
 697         }
 698         radeon_bo_kunmap(rdev->r600_blit.shader_obj);
 699         radeon_bo_unreserve(rdev->r600_blit.shader_obj);
 700
 701 done:
 702         r = radeon_bo_reserve(rdev->r600_blit.shader_obj, false);
 703         if (unlikely(r != 0))
 704                 return r;
 705         r = radeon_bo_pin(rdev->r600_blit.shader_obj, RADEON_GEM_DOMAIN_VRAM,
 706                           &rdev->r600_blit.shader_gpu_addr);
 707         radeon_bo_unreserve(rdev->r600_blit.shader_obj);
 708         if (r) {
 709                 dev_err(rdev->dev, "(%d) pin blit object failed\n", r);
 710                 return r;
 711         }
 712         radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
 713         return 0;
 714 }
 715
 716 void evergreen_blit_fini(struct radeon_device *rdev)
 717 {
 718         int r;
 719
 720         radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
 721         if (rdev->r600_blit.shader_obj == NULL)
 722                 return;
 723         /* If we can't reserve the bo, unref should be enough to destroy
 724          * it when it becomes idle.
 725          */
 726         r = radeon_bo_reserve(rdev->r600_blit.shader_obj, false);
 727         if (!r) {
 728                 radeon_bo_unpin(rdev->r600_blit.shader_obj);
 729                 radeon_bo_unreserve(rdev->r600_blit.shader_obj);
 730         }
 731         radeon_bo_unref(&rdev->r600_blit.shader_obj);
 732 }
 733
 734 static int evergreen_vb_ib_get(struct radeon_device *rdev)
 735 {
 736         int r;
 737         r = radeon_ib_get(rdev, &rdev->r600_blit.vb_ib);
 738         if (r) {
 739                 DRM_ERROR("failed to get IB for vertex buffer\n");
 740                 return r;
 741         }
 742
 743         rdev->r600_blit.vb_total = 64*1024;
 744         rdev->r600_blit.vb_used = 0;
 745         return 0;
 746 }
 747
 748 static void evergreen_vb_ib_put(struct radeon_device *rdev)
 749 {
 750         radeon_fence_emit(rdev, rdev->r600_blit.vb_ib->fence);
 751         radeon_ib_free(rdev, &rdev->r600_blit.vb_ib);
 752 }
 753
 754 int evergreen_blit_prepare_copy(struct radeon_device *rdev, int size_bytes)
 755 {
 756         int r;
 757         int ring_size, line_size;
 758         int max_size;
 759         /* loops of emits + fence emit possible */
 760         int dwords_per_loop = 74, num_loops;
 761
 762         r = evergreen_vb_ib_get(rdev);
 763         if (r)
 764                 return r;
 765
 766         /* 8 bpp vs 32 bpp for xfer unit */
 767         if (size_bytes & 3)
 768                 line_size = 8192;
 769         else
 770                 line_size = 8192 * 4;
 771
 772         max_size = 8192 * line_size;
 773
 774         /* major loops cover the max size transfer */
 775         num_loops = ((size_bytes + max_size) / max_size);
 776         /* minor loops cover the extra non aligned bits */
 777         num_loops += ((size_bytes % line_size) ? 1 : 0);
 778         /* calculate number of loops correctly */
 779         ring_size = num_loops * dwords_per_loop;
 780         /* set default  + shaders */
 781         ring_size += 55; /* shaders + def state */
 782         ring_size += 10; /* fence emit for VB IB */
 783         ring_size += 5; /* done copy */
 784         ring_size += 10; /* fence emit for done copy */
 785         r = radeon_ring_lock(rdev, ring_size);
 786         if (r)
 787                 return r;
 788
 789         set_default_state(rdev); /* 36 */
 790         set_shaders(rdev); /* 16 */
 791         return 0;
 792 }
 793
 794 void evergreen_blit_done_copy(struct radeon_device *rdev, struct radeon_fence *fence)
 795 {
 796         int r;
 797
 798         if (rdev->r600_blit.vb_ib)
 799                 evergreen_vb_ib_put(rdev);
 800
 801         if (fence)
 802                 r = radeon_fence_emit(rdev, fence);
 803
 804         radeon_ring_unlock_commit(rdev);
 805 }
 806
 807 void evergreen_kms_blit_copy(struct radeon_device *rdev,
 808                              u64 src_gpu_addr, u64 dst_gpu_addr,
 809                              int size_bytes)
 810 {
 811         int max_bytes;
 812         u64 vb_gpu_addr;
 813         u32 *vb;
 814
 815         DRM_DEBUG("emitting copy %16llx %16llx %d %d\n", src_gpu_addr, dst_gpu_addr,
 816                   size_bytes, rdev->r600_blit.vb_used);
 817         vb = (u32 *)(rdev->r600_blit.vb_ib->ptr + rdev->r600_blit.vb_used);
 818         if ((size_bytes & 3) || (src_gpu_addr & 3) || (dst_gpu_addr & 3)) {
 819                 max_bytes = 8192;
 820
 821                 while (size_bytes) {
 822                         int cur_size = size_bytes;
 823                         int src_x = src_gpu_addr & 255;
 824                         int dst_x = dst_gpu_addr & 255;
 825                         int h = 1;
 826                         src_gpu_addr = src_gpu_addr & ~255ULL;
 827                         dst_gpu_addr = dst_gpu_addr & ~255ULL;
 828
 829                         if (!src_x && !dst_x) {
 830                                 h = (cur_size / max_bytes);
 831                                 if (h > 8192)
 832                                         h = 8192;
 833                                 if (h == 0)
 834                                         h = 1;
 835                                 else
 836                                         cur_size = max_bytes;
 837                         } else {
 838                                 if (cur_size > max_bytes)
 839                                         cur_size = max_bytes;
 840                                 if (cur_size > (max_bytes - dst_x))
 841                                         cur_size = (max_bytes - dst_x);
 842                                 if (cur_size > (max_bytes - src_x))
 843                                         cur_size = (max_bytes - src_x);
 844                         }
 845
 846                         if ((rdev->r600_blit.vb_used + 48) > rdev->r600_blit.vb_total) {
 847                                 WARN_ON(1);
 848                         }
 849
 850                         vb[0] = i2f(dst_x);
 851                         vb[1] = 0;
 852                         vb[2] = i2f(src_x);
 853                         vb[3] = 0;
 854
 855                         vb[4] = i2f(dst_x);
 856                         vb[5] = i2f(h);
 857                         vb[6] = i2f(src_x);
 858                         vb[7] = i2f(h);
 859
 860                         vb[8] = i2f(dst_x + cur_size);
 861                         vb[9] = i2f(h);
 862                         vb[10] = i2f(src_x + cur_size);
 863                         vb[11] = i2f(h);
 864
 865                         /* src 10 */
 866                         set_tex_resource(rdev, FMT_8,
 867                                          src_x + cur_size, h, src_x + cur_size,
 868                                          src_gpu_addr);
 869
 870                         /* 5 */
 871                         cp_set_surface_sync(rdev,
 872                                             PACKET3_TC_ACTION_ENA, (src_x + cur_size * h), src_gpu_addr);
 873
 874
 875                         /* dst 17 */
 876                         set_render_target(rdev, COLOR_8,
 877                                           dst_x + cur_size, h,
 878                                           dst_gpu_addr);
 879
 880                         /* scissors 12 */
 881                         set_scissors(rdev, dst_x, 0, dst_x + cur_size, h);
 882
 883                         /* 15 */
 884                         vb_gpu_addr = rdev->r600_blit.vb_ib->gpu_addr + rdev->r600_blit.vb_used;
 885                         set_vtx_resource(rdev, vb_gpu_addr);
 886
 887                         /* draw 10 */
 888                         draw_auto(rdev);
 889
 890                         /* 5 */
 891                         cp_set_surface_sync(rdev,
 892                                             PACKET3_CB_ACTION_ENA | PACKET3_CB0_DEST_BASE_ENA,
 893                                             cur_size * h, dst_gpu_addr);
 894
 895                         vb += 12;
 896                         rdev->r600_blit.vb_used += 12 * 4;
 897
 898                         src_gpu_addr += cur_size * h;
 899                         dst_gpu_addr += cur_size * h;
 900                         size_bytes -= cur_size * h;
 901                 }
 902         } else {
 903                 max_bytes = 8192 * 4;
 904
 905                 while (size_bytes) {
 906                         int cur_size = size_bytes;
 907                         int src_x = (src_gpu_addr & 255);
 908                         int dst_x = (dst_gpu_addr & 255);
 909                         int h = 1;
 910                         src_gpu_addr = src_gpu_addr & ~255ULL;
 911                         dst_gpu_addr = dst_gpu_addr & ~255ULL;
 912
 913                         if (!src_x && !dst_x) {
 914                                 h = (cur_size / max_bytes);
 915                                 if (h > 8192)
 916                                         h = 8192;
 917                                 if (h == 0)
 918                                         h = 1;
 919                                 else
 920                                         cur_size = max_bytes;
 921                         } else {
 922                                 if (cur_size > max_bytes)
 923                                         cur_size = max_bytes;
 924                                 if (cur_size > (max_bytes - dst_x))
 925                                         cur_size = (max_bytes - dst_x);
 926                                 if (cur_size > (max_bytes - src_x))
 927                                         cur_size = (max_bytes - src_x);
 928                         }
 929
 930                         if ((rdev->r600_blit.vb_used + 48) > rdev->r600_blit.vb_total) {
 931                                 WARN_ON(1);
 932                         }
 933
 934                         vb[0] = i2f(dst_x / 4);
 935                         vb[1] = 0;
 936                         vb[2] = i2f(src_x / 4);
 937                         vb[3] = 0;
 938
 939                         vb[4] = i2f(dst_x / 4);
 940                         vb[5] = i2f(h);
 941                         vb[6] = i2f(src_x / 4);
 942                         vb[7] = i2f(h);
 943
 944                         vb[8] = i2f((dst_x + cur_size) / 4);
 945                         vb[9] = i2f(h);
 946                         vb[10] = i2f((src_x + cur_size) / 4);
 947                         vb[11] = i2f(h);
 948
 949                         /* src 10 */
 950                         set_tex_resource(rdev, FMT_8_8_8_8,
 951                                          (src_x + cur_size) / 4,
 952                                          h, (src_x + cur_size) / 4,
 953                                          src_gpu_addr);
 954                         /* 5 */
 955                         cp_set_surface_sync(rdev,
 956                                             PACKET3_TC_ACTION_ENA, (src_x + cur_size * h), src_gpu_addr);
 957
 958                         /* dst 17 */
 959                         set_render_target(rdev, COLOR_8_8_8_8,
 960                                           (dst_x + cur_size) / 4, h,
 961                                           dst_gpu_addr);
 962
 963                         /* scissors 12  */
 964                         set_scissors(rdev, (dst_x / 4), 0, (dst_x + cur_size / 4), h);
 965
 966                         /* Vertex buffer setup 15 */
 967                         vb_gpu_addr = rdev->r600_blit.vb_ib->gpu_addr + rdev->r600_blit.vb_used;
 968                         set_vtx_resource(rdev, vb_gpu_addr);
 969
 970                         /* draw 10 */
 971                         draw_auto(rdev);
 972
 973                         /* 5 */
 974                         cp_set_surface_sync(rdev,
 975                                             PACKET3_CB_ACTION_ENA | PACKET3_CB0_DEST_BASE_ENA,
 976                                             cur_size * h, dst_gpu_addr);
 977
 978                         /* 74 ring dwords per loop */
 979                         vb += 12;
 980                         rdev->r600_blit.vb_used += 12 * 4;
 981
 982                         src_gpu_addr += cur_size * h;
 983                         dst_gpu_addr += cur_size * h;
 984                         size_bytes -= cur_size * h;
 985                 }
 986         }
 987 }
 988