drivers/gpu/drm/radeon/r600_blit_kms.c

   1 /*
   2  * Copyright 2009 Advanced Micro Devices, Inc.
   3  * Copyright 2009 Red Hat Inc.
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the "Software"),
   7  * to deal in the Software without restriction, including without limitation
   8  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   9  * and/or sell copies of the Software, and to permit persons to whom the
  10  * Software is furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice (including the next
  13  * paragraph) shall be included in all copies or substantial portions of the
  14  * Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  19  * THE COPYRIGHT HOLDER(S) AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
  20  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  21  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  22  * DEALINGS IN THE SOFTWARE.
  23  *
  24  */
  25
  26 #include "drmP.h"
  27 #include "radeon_drm.h"
  28 #include "radeon.h"
  29
  30 #include "r600d.h"
  31 #include "r600_blit_shaders.h"
  32 #include "radeon_blit_common.h"
  33
  34 /* emits 21 on rv770+, 23 on r600 */
  35 static void
  36 set_render_target(struct radeon_device *rdev, int format,
  37                   int w, int h, u64 gpu_addr)
  38 {
  39         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
  40         u32 cb_color_info;
  41         int pitch, slice;
  42
  43         h = ALIGN(h, 8);
  44         if (h < 8)
  45                 h = 8;
  46
  47         cb_color_info = CB_FORMAT(format) |
  48                 CB_SOURCE_FORMAT(CB_SF_EXPORT_NORM) |
  49                 CB_ARRAY_MODE(ARRAY_1D_TILED_THIN1);
  50         pitch = (w / 8) - 1;
  51         slice = ((w * h) / 64) - 1;
  52
  53         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 1));
  54         radeon_ring_write(ring, (CB_COLOR0_BASE - PACKET3_SET_CONTEXT_REG_OFFSET) >> 2);
  55         radeon_ring_write(ring, gpu_addr >> 8);
  56
  57         if (rdev->family > CHIP_R600 && rdev->family < CHIP_RV770) {
  58                 radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_BASE_UPDATE, 0));
  59                 radeon_ring_write(ring, 2 << 0);
  60         }
  61
  62         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 1));
  63         radeon_ring_write(ring, (CB_COLOR0_SIZE - PACKET3_SET_CONTEXT_REG_OFFSET) >> 2);
  64         radeon_ring_write(ring, (pitch << 0) | (slice << 10));
  65
  66         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 1));
  67         radeon_ring_write(ring, (CB_COLOR0_VIEW - PACKET3_SET_CONTEXT_REG_OFFSET) >> 2);
  68         radeon_ring_write(ring, 0);
  69
  70         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 1));
  71         radeon_ring_write(ring, (CB_COLOR0_INFO - PACKET3_SET_CONTEXT_REG_OFFSET) >> 2);
  72         radeon_ring_write(ring, cb_color_info);
  73
  74         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 1));
  75         radeon_ring_write(ring, (CB_COLOR0_TILE - PACKET3_SET_CONTEXT_REG_OFFSET) >> 2);
  76         radeon_ring_write(ring, 0);
  77
  78         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 1));
  79         radeon_ring_write(ring, (CB_COLOR0_FRAG - PACKET3_SET_CONTEXT_REG_OFFSET) >> 2);
  80         radeon_ring_write(ring, 0);
  81
  82         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 1));
  83         radeon_ring_write(ring, (CB_COLOR0_MASK - PACKET3_SET_CONTEXT_REG_OFFSET) >> 2);
  84         radeon_ring_write(ring, 0);
  85 }
  86
  87 /* emits 5dw */
  88 static void
  89 cp_set_surface_sync(struct radeon_device *rdev,
  90                     u32 sync_type, u32 size,
  91                     u64 mc_addr)
  92 {
  93         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
  94         u32 cp_coher_size;
  95
  96         if (size == 0xffffffff)
  97                 cp_coher_size = 0xffffffff;
  98         else
  99                 cp_coher_size = ((size + 255) >> 8);
 100
 101         radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
 102         radeon_ring_write(ring, sync_type);
 103         radeon_ring_write(ring, cp_coher_size);
 104         radeon_ring_write(ring, mc_addr >> 8);
 105         radeon_ring_write(ring, 10); /* poll interval */
 106 }
 107
 108 /* emits 21dw + 1 surface sync = 26dw */
 109 static void
 110 set_shaders(struct radeon_device *rdev)
 111 {
 112         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
 113         u64 gpu_addr;
 114         u32 sq_pgm_resources;
 115
 116         /* setup shader regs */
 117         sq_pgm_resources = (1 << 0);
 118
 119         /* VS */
 120         gpu_addr = rdev->r600_blit.shader_gpu_addr + rdev->r600_blit.vs_offset;
 121         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 1));
 122         radeon_ring_write(ring, (SQ_PGM_START_VS - PACKET3_SET_CONTEXT_REG_OFFSET) >> 2);
 123         radeon_ring_write(ring, gpu_addr >> 8);
 124
 125         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 1));
 126         radeon_ring_write(ring, (SQ_PGM_RESOURCES_VS - PACKET3_SET_CONTEXT_REG_OFFSET) >> 2);
 127         radeon_ring_write(ring, sq_pgm_resources);
 128
 129         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 1));
 130         radeon_ring_write(ring, (SQ_PGM_CF_OFFSET_VS - PACKET3_SET_CONTEXT_REG_OFFSET) >> 2);
 131         radeon_ring_write(ring, 0);
 132
 133         /* PS */
 134         gpu_addr = rdev->r600_blit.shader_gpu_addr + rdev->r600_blit.ps_offset;
 135         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 1));
 136         radeon_ring_write(ring, (SQ_PGM_START_PS - PACKET3_SET_CONTEXT_REG_OFFSET) >> 2);
 137         radeon_ring_write(ring, gpu_addr >> 8);
 138
 139         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 1));
 140         radeon_ring_write(ring, (SQ_PGM_RESOURCES_PS - PACKET3_SET_CONTEXT_REG_OFFSET) >> 2);
 141         radeon_ring_write(ring, sq_pgm_resources | (1 << 28));
 142
 143         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 1));
 144         radeon_ring_write(ring, (SQ_PGM_EXPORTS_PS - PACKET3_SET_CONTEXT_REG_OFFSET) >> 2);
 145         radeon_ring_write(ring, 2);
 146
 147         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 1));
 148         radeon_ring_write(ring, (SQ_PGM_CF_OFFSET_PS - PACKET3_SET_CONTEXT_REG_OFFSET) >> 2);
 149         radeon_ring_write(ring, 0);
 150
 151         gpu_addr = rdev->r600_blit.shader_gpu_addr + rdev->r600_blit.vs_offset;
 152         cp_set_surface_sync(rdev, PACKET3_SH_ACTION_ENA, 512, gpu_addr);
 153 }
 154
 155 /* emits 9 + 1 sync (5) = 14*/
 156 static void
 157 set_vtx_resource(struct radeon_device *rdev, u64 gpu_addr)
 158 {
 159         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
 160         u32 sq_vtx_constant_word2;
 161
 162         sq_vtx_constant_word2 = SQ_VTXC_BASE_ADDR_HI(upper_32_bits(gpu_addr) & 0xff) |
 163                 SQ_VTXC_STRIDE(16);
 164 #ifdef __BIG_ENDIAN
 165         sq_vtx_constant_word2 |=  SQ_VTXC_ENDIAN_SWAP(SQ_ENDIAN_8IN32);
 166 #endif
 167
 168         radeon_ring_write(ring, PACKET3(PACKET3_SET_RESOURCE, 7));
 169         radeon_ring_write(ring, 0x460);
 170         radeon_ring_write(ring, gpu_addr & 0xffffffff);
 171         radeon_ring_write(ring, 48 - 1);
 172         radeon_ring_write(ring, sq_vtx_constant_word2);
 173         radeon_ring_write(ring, 1 << 0);
 174         radeon_ring_write(ring, 0);
 175         radeon_ring_write(ring, 0);
 176         radeon_ring_write(ring, SQ_TEX_VTX_VALID_BUFFER << 30);
 177
 178         if ((rdev->family == CHIP_RV610) ||
 179             (rdev->family == CHIP_RV620) ||
 180             (rdev->family == CHIP_RS780) ||
 181             (rdev->family == CHIP_RS880) ||
 182             (rdev->family == CHIP_RV710))
 183                 cp_set_surface_sync(rdev,
 184                                     PACKET3_TC_ACTION_ENA, 48, gpu_addr);
 185         else
 186                 cp_set_surface_sync(rdev,
 187                                     PACKET3_VC_ACTION_ENA, 48, gpu_addr);
 188 }
 189
 190 /* emits 9 */
 191 static void
 192 set_tex_resource(struct radeon_device *rdev,
 193                  int format, int w, int h, int pitch,
 194                  u64 gpu_addr, u32 size)
 195 {
 196         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
 197         uint32_t sq_tex_resource_word0, sq_tex_resource_word1, sq_tex_resource_word4;
 198
 199         if (h < 1)
 200                 h = 1;
 201
 202         sq_tex_resource_word0 = S_038000_DIM(V_038000_SQ_TEX_DIM_2D) |
 203                 S_038000_TILE_MODE(V_038000_ARRAY_1D_TILED_THIN1);
 204         sq_tex_resource_word0 |= S_038000_PITCH((pitch >> 3) - 1) |
 205                 S_038000_TEX_WIDTH(w - 1);
 206
 207         sq_tex_resource_word1 = S_038004_DATA_FORMAT(format);
 208         sq_tex_resource_word1 |= S_038004_TEX_HEIGHT(h - 1);
 209
 210         sq_tex_resource_word4 = S_038010_REQUEST_SIZE(1) |
 211                 S_038010_DST_SEL_X(SQ_SEL_X) |
 212                 S_038010_DST_SEL_Y(SQ_SEL_Y) |
 213                 S_038010_DST_SEL_Z(SQ_SEL_Z) |
 214                 S_038010_DST_SEL_W(SQ_SEL_W);
 215
 216         cp_set_surface_sync(rdev,
 217                             PACKET3_TC_ACTION_ENA, size, gpu_addr);
 218
 219         radeon_ring_write(ring, PACKET3(PACKET3_SET_RESOURCE, 7));
 220         radeon_ring_write(ring, 0);
 221         radeon_ring_write(ring, sq_tex_resource_word0);
 222         radeon_ring_write(ring, sq_tex_resource_word1);
 223         radeon_ring_write(ring, gpu_addr >> 8);
 224         radeon_ring_write(ring, gpu_addr >> 8);
 225         radeon_ring_write(ring, sq_tex_resource_word4);
 226         radeon_ring_write(ring, 0);
 227         radeon_ring_write(ring, SQ_TEX_VTX_VALID_TEXTURE << 30);
 228 }
 229
 230 /* emits 12 */
 231 static void
 232 set_scissors(struct radeon_device *rdev, int x1, int y1,
 233              int x2, int y2)
 234 {
 235         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
 236         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
 237         radeon_ring_write(ring, (PA_SC_SCREEN_SCISSOR_TL - PACKET3_SET_CONTEXT_REG_OFFSET) >> 2);
 238         radeon_ring_write(ring, (x1 << 0) | (y1 << 16));
 239         radeon_ring_write(ring, (x2 << 0) | (y2 << 16));
 240
 241         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
 242         radeon_ring_write(ring, (PA_SC_GENERIC_SCISSOR_TL - PACKET3_SET_CONTEXT_REG_OFFSET) >> 2);
 243         radeon_ring_write(ring, (x1 << 0) | (y1 << 16) | (1 << 31));
 244         radeon_ring_write(ring, (x2 << 0) | (y2 << 16));
 245
 246         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
 247         radeon_ring_write(ring, (PA_SC_WINDOW_SCISSOR_TL - PACKET3_SET_CONTEXT_REG_OFFSET) >> 2);
 248         radeon_ring_write(ring, (x1 << 0) | (y1 << 16) | (1 << 31));
 249         radeon_ring_write(ring, (x2 << 0) | (y2 << 16));
 250 }
 251
 252 /* emits 10 */
 253 static void
 254 draw_auto(struct radeon_device *rdev)
 255 {
 256         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
 257         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
 258         radeon_ring_write(ring, (VGT_PRIMITIVE_TYPE - PACKET3_SET_CONFIG_REG_OFFSET) >> 2);
 259         radeon_ring_write(ring, DI_PT_RECTLIST);
 260
 261         radeon_ring_write(ring, PACKET3(PACKET3_INDEX_TYPE, 0));
 262         radeon_ring_write(ring,
 263 #ifdef __BIG_ENDIAN
 264                           (2 << 2) |
 265 #endif
 266                           DI_INDEX_SIZE_16_BIT);
 267
 268         radeon_ring_write(ring, PACKET3(PACKET3_NUM_INSTANCES, 0));
 269         radeon_ring_write(ring, 1);
 270
 271         radeon_ring_write(ring, PACKET3(PACKET3_DRAW_INDEX_AUTO, 1));
 272         radeon_ring_write(ring, 3);
 273         radeon_ring_write(ring, DI_SRC_SEL_AUTO_INDEX);
 274
 275 }
 276
 277 /* emits 14 */
 278 static void
 279 set_default_state(struct radeon_device *rdev)
 280 {
 281         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
 282         u32 sq_config, sq_gpr_resource_mgmt_1, sq_gpr_resource_mgmt_2;
 283         u32 sq_thread_resource_mgmt, sq_stack_resource_mgmt_1, sq_stack_resource_mgmt_2;
 284         int num_ps_gprs, num_vs_gprs, num_temp_gprs, num_gs_gprs, num_es_gprs;
 285         int num_ps_threads, num_vs_threads, num_gs_threads, num_es_threads;
 286         int num_ps_stack_entries, num_vs_stack_entries, num_gs_stack_entries, num_es_stack_entries;
 287         u64 gpu_addr;
 288         int dwords;
 289
 290         switch (rdev->family) {
 291         case CHIP_R600:
 292                 num_ps_gprs = 192;
 293                 num_vs_gprs = 56;
 294                 num_temp_gprs = 4;
 295                 num_gs_gprs = 0;
 296                 num_es_gprs = 0;
 297                 num_ps_threads = 136;
 298                 num_vs_threads = 48;
 299                 num_gs_threads = 4;
 300                 num_es_threads = 4;
 301                 num_ps_stack_entries = 128;
 302                 num_vs_stack_entries = 128;
 303                 num_gs_stack_entries = 0;
 304                 num_es_stack_entries = 0;
 305                 break;
 306         case CHIP_RV630:
 307         case CHIP_RV635:
 308                 num_ps_gprs = 84;
 309                 num_vs_gprs = 36;
 310                 num_temp_gprs = 4;
 311                 num_gs_gprs = 0;
 312                 num_es_gprs = 0;
 313                 num_ps_threads = 144;
 314                 num_vs_threads = 40;
 315                 num_gs_threads = 4;
 316                 num_es_threads = 4;
 317                 num_ps_stack_entries = 40;
 318                 num_vs_stack_entries = 40;
 319                 num_gs_stack_entries = 32;
 320                 num_es_stack_entries = 16;
 321                 break;
 322         case CHIP_RV610:
 323         case CHIP_RV620:
 324         case CHIP_RS780:
 325         case CHIP_RS880:
 326         default:
 327                 num_ps_gprs = 84;
 328                 num_vs_gprs = 36;
 329                 num_temp_gprs = 4;
 330                 num_gs_gprs = 0;
 331                 num_es_gprs = 0;
 332                 num_ps_threads = 136;
 333                 num_vs_threads = 48;
 334                 num_gs_threads = 4;
 335                 num_es_threads = 4;
 336                 num_ps_stack_entries = 40;
 337                 num_vs_stack_entries = 40;
 338                 num_gs_stack_entries = 32;
 339                 num_es_stack_entries = 16;
 340                 break;
 341         case CHIP_RV670:
 342                 num_ps_gprs = 144;
 343                 num_vs_gprs = 40;
 344                 num_temp_gprs = 4;
 345                 num_gs_gprs = 0;
 346                 num_es_gprs = 0;
 347                 num_ps_threads = 136;
 348                 num_vs_threads = 48;
 349                 num_gs_threads = 4;
 350                 num_es_threads = 4;
 351                 num_ps_stack_entries = 40;
 352                 num_vs_stack_entries = 40;
 353                 num_gs_stack_entries = 32;
 354                 num_es_stack_entries = 16;
 355                 break;
 356         case CHIP_RV770:
 357                 num_ps_gprs = 192;
 358                 num_vs_gprs = 56;
 359                 num_temp_gprs = 4;
 360                 num_gs_gprs = 0;
 361                 num_es_gprs = 0;
 362                 num_ps_threads = 188;
 363                 num_vs_threads = 60;
 364                 num_gs_threads = 0;
 365                 num_es_threads = 0;
 366                 num_ps_stack_entries = 256;
 367                 num_vs_stack_entries = 256;
 368                 num_gs_stack_entries = 0;
 369                 num_es_stack_entries = 0;
 370                 break;
 371         case CHIP_RV730:
 372         case CHIP_RV740:
 373                 num_ps_gprs = 84;
 374                 num_vs_gprs = 36;
 375                 num_temp_gprs = 4;
 376                 num_gs_gprs = 0;
 377                 num_es_gprs = 0;
 378                 num_ps_threads = 188;
 379                 num_vs_threads = 60;
 380                 num_gs_threads = 0;
 381                 num_es_threads = 0;
 382                 num_ps_stack_entries = 128;
 383                 num_vs_stack_entries = 128;
 384                 num_gs_stack_entries = 0;
 385                 num_es_stack_entries = 0;
 386                 break;
 387         case CHIP_RV710:
 388                 num_ps_gprs = 192;
 389                 num_vs_gprs = 56;
 390                 num_temp_gprs = 4;
 391                 num_gs_gprs = 0;
 392                 num_es_gprs = 0;
 393                 num_ps_threads = 144;
 394                 num_vs_threads = 48;
 395                 num_gs_threads = 0;
 396                 num_es_threads = 0;
 397                 num_ps_stack_entries = 128;
 398                 num_vs_stack_entries = 128;
 399                 num_gs_stack_entries = 0;
 400                 num_es_stack_entries = 0;
 401                 break;
 402         }
 403
 404         if ((rdev->family == CHIP_RV610) ||
 405             (rdev->family == CHIP_RV620) ||
 406             (rdev->family == CHIP_RS780) ||
 407             (rdev->family == CHIP_RS880) ||
 408             (rdev->family == CHIP_RV710))
 409                 sq_config = 0;
 410         else
 411                 sq_config = VC_ENABLE;
 412
 413         sq_config |= (DX9_CONSTS |
 414                       ALU_INST_PREFER_VECTOR |
 415                       PS_PRIO(0) |
 416                       VS_PRIO(1) |
 417                       GS_PRIO(2) |
 418                       ES_PRIO(3));
 419
 420         sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(num_ps_gprs) |
 421                                   NUM_VS_GPRS(num_vs_gprs) |
 422                                   NUM_CLAUSE_TEMP_GPRS(num_temp_gprs));
 423         sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(num_gs_gprs) |
 424                                   NUM_ES_GPRS(num_es_gprs));
 425         sq_thread_resource_mgmt = (NUM_PS_THREADS(num_ps_threads) |
 426                                    NUM_VS_THREADS(num_vs_threads) |
 427                                    NUM_GS_THREADS(num_gs_threads) |
 428                                    NUM_ES_THREADS(num_es_threads));
 429         sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(num_ps_stack_entries) |
 430                                     NUM_VS_STACK_ENTRIES(num_vs_stack_entries));
 431         sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(num_gs_stack_entries) |
 432                                     NUM_ES_STACK_ENTRIES(num_es_stack_entries));
 433
 434         /* emit an IB pointing at default state */
 435         dwords = ALIGN(rdev->r600_blit.state_len, 0x10);
 436         gpu_addr = rdev->r600_blit.shader_gpu_addr + rdev->r600_blit.state_offset;
 437         radeon_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
 438         radeon_ring_write(ring,
 439 #ifdef __BIG_ENDIAN
 440                           (2 << 0) |
 441 #endif
 442                           (gpu_addr & 0xFFFFFFFC));
 443         radeon_ring_write(ring, upper_32_bits(gpu_addr) & 0xFF);
 444         radeon_ring_write(ring, dwords);
 445
 446         /* SQ config */
 447         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 6));
 448         radeon_ring_write(ring, (SQ_CONFIG - PACKET3_SET_CONFIG_REG_OFFSET) >> 2);
 449         radeon_ring_write(ring, sq_config);
 450         radeon_ring_write(ring, sq_gpr_resource_mgmt_1);
 451         radeon_ring_write(ring, sq_gpr_resource_mgmt_2);
 452         radeon_ring_write(ring, sq_thread_resource_mgmt);
 453         radeon_ring_write(ring, sq_stack_resource_mgmt_1);
 454         radeon_ring_write(ring, sq_stack_resource_mgmt_2);
 455 }
 456
 457 #define I2F_MAX_BITS 15
 458 #define I2F_MAX_INPUT  ((1 << I2F_MAX_BITS) - 1)
 459 #define I2F_SHIFT (24 - I2F_MAX_BITS)
 460
 461 /*
 462  * Converts unsigned integer into 32-bit IEEE floating point representation.
 463  * Conversion is not universal and only works for the range from 0
 464  * to 2^I2F_MAX_BITS-1. Currently we only use it with inputs between
 465  * 0 and 16384 (inclusive), so I2F_MAX_BITS=15 is enough. If necessary,
 466  * I2F_MAX_BITS can be increased, but that will add to the loop iterations
 467  * and slow us down. Conversion is done by shifting the input and counting
 468  * down until the first 1 reaches bit position 23. The resulting counter
 469  * and the shifted input are, respectively, the exponent and the fraction.
 470  * The sign is always zero.
 471  */
 472 static uint32_t i2f(uint32_t input)
 473 {
 474         u32 result, i, exponent, fraction;
 475
 476         WARN_ON_ONCE(input > I2F_MAX_INPUT);
 477
 478         if ((input & I2F_MAX_INPUT) == 0)
 479                 result = 0;
 480         else {
 481                 exponent = 126 + I2F_MAX_BITS;
 482                 fraction = (input & I2F_MAX_INPUT) << I2F_SHIFT;
 483
 484                 for (i = 0; i < I2F_MAX_BITS; i++) {
 485                         if (fraction & 0x800000)
 486                                 break;
 487                         else {
 488                                 fraction = fraction << 1;
 489                                 exponent = exponent - 1;
 490                         }
 491                 }
 492                 result = exponent << 23 | (fraction & 0x7fffff);
 493         }
 494         return result;
 495 }
 496
 497 int r600_blit_init(struct radeon_device *rdev)
 498 {
 499         u32 obj_size;
 500         int i, r, dwords;
 501         void *ptr;
 502         u32 packet2s[16];
 503         int num_packet2s = 0;
 504
 505         rdev->r600_blit.primitives.set_render_target = set_render_target;
 506         rdev->r600_blit.primitives.cp_set_surface_sync = cp_set_surface_sync;
 507         rdev->r600_blit.primitives.set_shaders = set_shaders;
 508         rdev->r600_blit.primitives.set_vtx_resource = set_vtx_resource;
 509         rdev->r600_blit.primitives.set_tex_resource = set_tex_resource;
 510         rdev->r600_blit.primitives.set_scissors = set_scissors;
 511         rdev->r600_blit.primitives.draw_auto = draw_auto;
 512         rdev->r600_blit.primitives.set_default_state = set_default_state;
 513
 514         rdev->r600_blit.ring_size_common = 8; /* sync semaphore */
 515         rdev->r600_blit.ring_size_common += 40; /* shaders + def state */
 516         rdev->r600_blit.ring_size_common += 5; /* done copy */
 517         rdev->r600_blit.ring_size_common += 16; /* fence emit for done copy */
 518
 519         rdev->r600_blit.ring_size_per_loop = 76;
 520         /* set_render_target emits 2 extra dwords on rv6xx */
 521         if (rdev->family > CHIP_R600 && rdev->family < CHIP_RV770)
 522                 rdev->r600_blit.ring_size_per_loop += 2;
 523
 524         rdev->r600_blit.max_dim = 8192;
 525
 526         rdev->r600_blit.state_offset = 0;
 527
 528         if (rdev->family >= CHIP_RV770)
 529                 rdev->r600_blit.state_len = r7xx_default_size;
 530         else
 531                 rdev->r600_blit.state_len = r6xx_default_size;
 532
 533         dwords = rdev->r600_blit.state_len;
 534         while (dwords & 0xf) {
 535                 packet2s[num_packet2s++] = cpu_to_le32(PACKET2(0));
 536                 dwords++;
 537         }
 538
 539         obj_size = dwords * 4;
 540         obj_size = ALIGN(obj_size, 256);
 541
 542         rdev->r600_blit.vs_offset = obj_size;
 543         obj_size += r6xx_vs_size * 4;
 544         obj_size = ALIGN(obj_size, 256);
 545
 546         rdev->r600_blit.ps_offset = obj_size;
 547         obj_size += r6xx_ps_size * 4;
 548         obj_size = ALIGN(obj_size, 256);
 549
 550         /* pin copy shader into vram if not already initialized */
 551         if (rdev->r600_blit.shader_obj == NULL) {
 552                 r = radeon_bo_create(rdev, obj_size, PAGE_SIZE, true,
 553                                      RADEON_GEM_DOMAIN_VRAM,
 554                                      NULL, &rdev->r600_blit.shader_obj);
 555                 if (r) {
 556                         DRM_ERROR("r600 failed to allocate shader\n");
 557                         return r;
 558                 }
 559
 560                 r = radeon_bo_reserve(rdev->r600_blit.shader_obj, false);
 561                 if (unlikely(r != 0))
 562                         return r;
 563                 r = radeon_bo_pin(rdev->r600_blit.shader_obj, RADEON_GEM_DOMAIN_VRAM,
 564                                   &rdev->r600_blit.shader_gpu_addr);
 565                 radeon_bo_unreserve(rdev->r600_blit.shader_obj);
 566                 if (r) {
 567                         dev_err(rdev->dev, "(%d) pin blit object failed\n", r);
 568                         return r;
 569                 }
 570         }
 571
 572         DRM_DEBUG("r6xx blit allocated bo %08x vs %08x ps %08x\n",
 573                   obj_size,
 574                   rdev->r600_blit.vs_offset, rdev->r600_blit.ps_offset);
 575
 576         r = radeon_bo_reserve(rdev->r600_blit.shader_obj, false);
 577         if (unlikely(r != 0))
 578                 return r;
 579         r = radeon_bo_kmap(rdev->r600_blit.shader_obj, &ptr);
 580         if (r) {
 581                 DRM_ERROR("failed to map blit object %d\n", r);
 582                 return r;
 583         }
 584         if (rdev->family >= CHIP_RV770)
 585                 memcpy_toio(ptr + rdev->r600_blit.state_offset,
 586                             r7xx_default_state, rdev->r600_blit.state_len * 4);
 587         else
 588                 memcpy_toio(ptr + rdev->r600_blit.state_offset,
 589                             r6xx_default_state, rdev->r600_blit.state_len * 4);
 590         if (num_packet2s)
 591                 memcpy_toio(ptr + rdev->r600_blit.state_offset + (rdev->r600_blit.state_len * 4),
 592                             packet2s, num_packet2s * 4);
 593         for (i = 0; i < r6xx_vs_size; i++)
 594                 *(u32 *)((unsigned long)ptr + rdev->r600_blit.vs_offset + i * 4) = cpu_to_le32(r6xx_vs[i]);
 595         for (i = 0; i < r6xx_ps_size; i++)
 596                 *(u32 *)((unsigned long)ptr + rdev->r600_blit.ps_offset + i * 4) = cpu_to_le32(r6xx_ps[i]);
 597         radeon_bo_kunmap(rdev->r600_blit.shader_obj);
 598         radeon_bo_unreserve(rdev->r600_blit.shader_obj);
 599
 600         radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
 601         return 0;
 602 }
 603
 604 void r600_blit_fini(struct radeon_device *rdev)
 605 {
 606         int r;
 607
 608         radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
 609         if (rdev->r600_blit.shader_obj == NULL)
 610                 return;
 611         /* If we can't reserve the bo, unref should be enough to destroy
 612          * it when it becomes idle.
 613          */
 614         r = radeon_bo_reserve(rdev->r600_blit.shader_obj, false);
 615         if (!r) {
 616                 radeon_bo_unpin(rdev->r600_blit.shader_obj);
 617                 radeon_bo_unreserve(rdev->r600_blit.shader_obj);
 618         }
 619         radeon_bo_unref(&rdev->r600_blit.shader_obj);
 620 }
 621
 622 static unsigned r600_blit_create_rect(unsigned num_gpu_pages,
 623                                       int *width, int *height, int max_dim)
 624 {
 625         unsigned max_pages;
 626         unsigned pages = num_gpu_pages;
 627         int w, h;
 628
 629         if (num_gpu_pages == 0) {
 630                 /* not supposed to be called with no pages, but just in case */
 631                 h = 0;
 632                 w = 0;
 633                 pages = 0;
 634                 WARN_ON(1);
 635         } else {
 636                 int rect_order = 2;
 637                 h = RECT_UNIT_H;
 638                 while (num_gpu_pages / rect_order) {
 639                         h *= 2;
 640                         rect_order *= 4;
 641                         if (h >= max_dim) {
 642                                 h = max_dim;
 643                                 break;
 644                         }
 645                 }
 646                 max_pages = (max_dim * h) / (RECT_UNIT_W * RECT_UNIT_H);
 647                 if (pages > max_pages)
 648                         pages = max_pages;
 649                 w = (pages * RECT_UNIT_W * RECT_UNIT_H) / h;
 650                 w = (w / RECT_UNIT_W) * RECT_UNIT_W;
 651                 pages = (w * h) / (RECT_UNIT_W * RECT_UNIT_H);
 652                 BUG_ON(pages == 0);
 653         }
 654
 655
 656         DRM_DEBUG("blit_rectangle: h=%d, w=%d, pages=%d\n", h, w, pages);
 657
 658         /* return width and height only of the caller wants it */
 659         if (height)
 660                 *height = h;
 661         if (width)
 662                 *width = w;
 663
 664         return pages;
 665 }
 666
 667
 668 int r600_blit_prepare_copy(struct radeon_device *rdev, unsigned num_gpu_pages,
 669                            struct radeon_fence **fence, struct radeon_sa_bo **vb,
 670                            struct radeon_semaphore **sem)
 671 {
 672         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
 673         int r;
 674         int ring_size;
 675         int num_loops = 0;
 676         int dwords_per_loop = rdev->r600_blit.ring_size_per_loop;
 677
 678         /* num loops */
 679         while (num_gpu_pages) {
 680                 num_gpu_pages -=
 681                         r600_blit_create_rect(num_gpu_pages, NULL, NULL,
 682                                               rdev->r600_blit.max_dim);
 683                 num_loops++;
 684         }
 685
 686         /* 48 bytes for vertex per loop */
 687         r = radeon_sa_bo_new(rdev, &rdev->ring_tmp_bo, vb,
 688                              (num_loops*48)+256, 256, true);
 689         if (r) {
 690                 return r;
 691         }
 692
 693         r = radeon_semaphore_create(rdev, sem);
 694         if (r) {
 695                 radeon_sa_bo_free(rdev, vb, NULL);
 696                 return r;
 697         }
 698
 699         /* calculate number of loops correctly */
 700         ring_size = num_loops * dwords_per_loop;
 701         ring_size += rdev->r600_blit.ring_size_common;
 702         r = radeon_ring_lock(rdev, ring, ring_size);
 703         if (r) {
 704                 radeon_sa_bo_free(rdev, vb, NULL);
 705                 radeon_semaphore_free(rdev, sem, NULL);
 706                 return r;
 707         }
 708
 709         if (radeon_fence_need_sync(*fence, RADEON_RING_TYPE_GFX_INDEX)) {
 710                 radeon_semaphore_sync_rings(rdev, *sem, (*fence)->ring,
 711                                             RADEON_RING_TYPE_GFX_INDEX);
 712                 radeon_fence_note_sync(*fence, RADEON_RING_TYPE_GFX_INDEX);
 713         } else {
 714                 radeon_semaphore_free(rdev, sem, NULL);
 715         }
 716
 717         rdev->r600_blit.primitives.set_default_state(rdev);
 718         rdev->r600_blit.primitives.set_shaders(rdev);
 719         return 0;
 720 }
 721
 722 void r600_blit_done_copy(struct radeon_device *rdev, struct radeon_fence **fence,
 723                          struct radeon_sa_bo *vb, struct radeon_semaphore *sem)
 724 {
 725         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
 726         int r;
 727
 728         r = radeon_fence_emit(rdev, fence, RADEON_RING_TYPE_GFX_INDEX);
 729         if (r) {
 730                 radeon_ring_unlock_undo(rdev, ring);
 731                 return;
 732         }
 733
 734         radeon_ring_unlock_commit(rdev, ring);
 735         radeon_sa_bo_free(rdev, &vb, *fence);
 736         radeon_semaphore_free(rdev, &sem, *fence);
 737 }
 738
 739 void r600_kms_blit_copy(struct radeon_device *rdev,
 740                         u64 src_gpu_addr, u64 dst_gpu_addr,
 741                         unsigned num_gpu_pages,
 742                         struct radeon_sa_bo *vb)
 743 {
 744         u64 vb_gpu_addr;
 745         u32 *vb_cpu_addr;
 746
 747         DRM_DEBUG("emitting copy %16llx %16llx %d\n",
 748                   src_gpu_addr, dst_gpu_addr, num_gpu_pages);
 749         vb_cpu_addr = (u32 *)radeon_sa_bo_cpu_addr(vb);
 750         vb_gpu_addr = radeon_sa_bo_gpu_addr(vb);
 751
 752         while (num_gpu_pages) {
 753                 int w, h;
 754                 unsigned size_in_bytes;
 755                 unsigned pages_per_loop =
 756                         r600_blit_create_rect(num_gpu_pages, &w, &h,
 757                                               rdev->r600_blit.max_dim);
 758
 759                 size_in_bytes = pages_per_loop * RADEON_GPU_PAGE_SIZE;
 760                 DRM_DEBUG("rectangle w=%d h=%d\n", w, h);
 761
 762                 vb_cpu_addr[0] = 0;
 763                 vb_cpu_addr[1] = 0;
 764                 vb_cpu_addr[2] = 0;
 765                 vb_cpu_addr[3] = 0;
 766
 767                 vb_cpu_addr[4] = 0;
 768                 vb_cpu_addr[5] = i2f(h);
 769                 vb_cpu_addr[6] = 0;
 770                 vb_cpu_addr[7] = i2f(h);
 771
 772                 vb_cpu_addr[8] = i2f(w);
 773                 vb_cpu_addr[9] = i2f(h);
 774                 vb_cpu_addr[10] = i2f(w);
 775                 vb_cpu_addr[11] = i2f(h);
 776
 777                 rdev->r600_blit.primitives.set_tex_resource(rdev, FMT_8_8_8_8,
 778                                                             w, h, w, src_gpu_addr, size_in_bytes);
 779                 rdev->r600_blit.primitives.set_render_target(rdev, COLOR_8_8_8_8,
 780                                                              w, h, dst_gpu_addr);
 781                 rdev->r600_blit.primitives.set_scissors(rdev, 0, 0, w, h);
 782                 rdev->r600_blit.primitives.set_vtx_resource(rdev, vb_gpu_addr);
 783                 rdev->r600_blit.primitives.draw_auto(rdev);
 784                 rdev->r600_blit.primitives.cp_set_surface_sync(rdev,
 785                                     PACKET3_CB_ACTION_ENA | PACKET3_CB0_DEST_BASE_ENA,
 786                                     size_in_bytes, dst_gpu_addr);
 787
 788                 vb_cpu_addr += 12;
 789                 vb_gpu_addr += 4*12;
 790                 src_gpu_addr += size_in_bytes;
 791                 dst_gpu_addr += size_in_bytes;
 792                 num_gpu_pages -= pages_per_loop;
 793         }
 794 }