r600_hw_context.c revision 543b2331d7b45a29ccd3530daa2389e87e65d89b
1/* 2 * Copyright 2010 Jerome Glisse <glisse@freedesktop.org> 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * on the rights to use, copy, modify, merge, publish, distribute, sub 8 * license, and/or sell copies of the Software, and to permit persons to whom 9 * the Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice (including the next 12 * paragraph) shall be included in all copies or substantial portions of the 13 * Software. 14 * 15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL 18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, 19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 21 * USE OR OTHER DEALINGS IN THE SOFTWARE. 22 * 23 * Authors: 24 * Jerome Glisse 25 */ 26#include "r600_hw_context_priv.h" 27#include "r600_pipe.h" 28#include "r600d.h" 29#include "util/u_memory.h" 30#include <errno.h> 31 32#define GROUP_FORCE_NEW_BLOCK 0 33 34/* Get backends mask */ 35void r600_get_backend_mask(struct r600_context *ctx) 36{ 37 struct r600_resource *buffer; 38 u32 *results; 39 unsigned num_backends = ctx->screen->info.r600_num_backends; 40 unsigned i, mask = 0; 41 42 /* if backend_map query is supported by the kernel */ 43 if (ctx->screen->info.r600_backend_map_valid) { 44 unsigned num_tile_pipes = ctx->screen->info.r600_num_tile_pipes; 45 unsigned backend_map = ctx->screen->info.r600_backend_map; 46 unsigned item_width, item_mask; 47 48 if (ctx->screen->chip_class >= EVERGREEN) { 49 item_width = 4; 50 item_mask = 0x7; 51 } else { 52 item_width = 2; 53 item_mask = 0x3; 54 } 55 56 while(num_tile_pipes--) { 57 i = backend_map & item_mask; 58 mask |= (1<<i); 59 backend_map >>= item_width; 60 } 61 if (mask != 0) { 62 ctx->backend_mask = mask; 63 return; 64 } 65 } 66 67 /* otherwise backup path for older kernels */ 68 69 /* create buffer for event data */ 70 buffer = (struct r600_resource*) 71 pipe_buffer_create(&ctx->screen->screen, PIPE_BIND_CUSTOM, 72 PIPE_USAGE_STAGING, ctx->max_db*16); 73 if (!buffer) 74 goto err; 75 76 /* initialize buffer with zeroes */ 77 results = ctx->ws->buffer_map(buffer->buf, ctx->cs, PIPE_TRANSFER_WRITE); 78 if (results) { 79 memset(results, 0, ctx->max_db * 4 * 4); 80 ctx->ws->buffer_unmap(buffer->buf); 81 82 /* emit EVENT_WRITE for ZPASS_DONE */ 83 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 2, 0); 84 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_ZPASS_DONE) | EVENT_INDEX(1); 85 ctx->pm4[ctx->pm4_cdwords++] = 0; 86 ctx->pm4[ctx->pm4_cdwords++] = 0; 87 88 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); 89 ctx->pm4[ctx->pm4_cdwords++] = r600_context_bo_reloc(ctx, buffer, RADEON_USAGE_WRITE); 90 91 /* analyze results */ 92 results = ctx->ws->buffer_map(buffer->buf, ctx->cs, PIPE_TRANSFER_READ); 93 if (results) { 94 for(i = 0; i < ctx->max_db; i++) { 95 /* at least highest bit will be set if backend is used */ 96 if (results[i*4 + 1]) 97 mask |= (1<<i); 98 } 99 ctx->ws->buffer_unmap(buffer->buf); 100 } 101 } 102 103 pipe_resource_reference((struct pipe_resource**)&buffer, NULL); 104 105 if (mask != 0) { 106 ctx->backend_mask = mask; 107 return; 108 } 109 110err: 111 /* fallback to old method - set num_backends lower bits to 1 */ 112 ctx->backend_mask = (~((u32)0))>>(32-num_backends); 113 return; 114} 115 116static inline void r600_context_ps_partial_flush(struct r600_context *ctx) 117{ 118 if (!(ctx->flags & R600_CONTEXT_DRAW_PENDING)) 119 return; 120 121 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, 0); 122 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4); 123 124 ctx->flags &= ~R600_CONTEXT_DRAW_PENDING; 125} 126 127void r600_init_cs(struct r600_context *ctx) 128{ 129 /* R6xx requires this packet at the start of each command buffer */ 130 if (ctx->screen->family < CHIP_RV770) { 131 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_START_3D_CMDBUF, 0, 0); 132 ctx->pm4[ctx->pm4_cdwords++] = 0x00000000; 133 } 134 /* All asics require this one */ 135 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_CONTEXT_CONTROL, 1, 0); 136 ctx->pm4[ctx->pm4_cdwords++] = 0x80000000; 137 ctx->pm4[ctx->pm4_cdwords++] = 0x80000000; 138 139 ctx->init_dwords = ctx->pm4_cdwords; 140} 141 142static void r600_init_block(struct r600_context *ctx, 143 struct r600_block *block, 144 const struct r600_reg *reg, int index, int nreg, 145 unsigned opcode, unsigned offset_base) 146{ 147 int i = index; 148 int j, n = nreg; 149 150 /* initialize block */ 151 if (opcode == PKT3_SET_RESOURCE) { 152 block->flags = BLOCK_FLAG_RESOURCE; 153 block->status |= R600_BLOCK_STATUS_RESOURCE_DIRTY; /* dirty all blocks at start */ 154 } else { 155 block->flags = 0; 156 block->status |= R600_BLOCK_STATUS_DIRTY; /* dirty all blocks at start */ 157 } 158 block->start_offset = reg[i].offset; 159 block->pm4[block->pm4_ndwords++] = PKT3(opcode, n, 0); 160 block->pm4[block->pm4_ndwords++] = (block->start_offset - offset_base) >> 2; 161 block->reg = &block->pm4[block->pm4_ndwords]; 162 block->pm4_ndwords += n; 163 block->nreg = n; 164 block->nreg_dirty = n; 165 LIST_INITHEAD(&block->list); 166 LIST_INITHEAD(&block->enable_list); 167 168 for (j = 0; j < n; j++) { 169 if (reg[i+j].flags & REG_FLAG_DIRTY_ALWAYS) { 170 block->flags |= REG_FLAG_DIRTY_ALWAYS; 171 } 172 if (reg[i+j].flags & REG_FLAG_ENABLE_ALWAYS) { 173 if (!(block->status & R600_BLOCK_STATUS_ENABLED)) { 174 block->status |= R600_BLOCK_STATUS_ENABLED; 175 LIST_ADDTAIL(&block->enable_list, &ctx->enable_list); 176 LIST_ADDTAIL(&block->list,&ctx->dirty); 177 } 178 } 179 if (reg[i+j].flags & REG_FLAG_FLUSH_CHANGE) { 180 block->flags |= REG_FLAG_FLUSH_CHANGE; 181 } 182 183 if (reg[i+j].flags & REG_FLAG_NEED_BO) { 184 block->nbo++; 185 assert(block->nbo < R600_BLOCK_MAX_BO); 186 block->pm4_bo_index[j] = block->nbo; 187 block->pm4[block->pm4_ndwords++] = PKT3(PKT3_NOP, 0, 0); 188 block->pm4[block->pm4_ndwords++] = 0x00000000; 189 if (reg[i+j].flags & REG_FLAG_RV6XX_SBU) { 190 block->reloc[block->nbo].flush_flags = 0; 191 block->reloc[block->nbo].flush_mask = 0; 192 } else { 193 block->reloc[block->nbo].flush_flags = reg[i+j].flush_flags; 194 block->reloc[block->nbo].flush_mask = reg[i+j].flush_mask; 195 } 196 block->reloc[block->nbo].bo_pm4_index = block->pm4_ndwords - 1; 197 } 198 if ((ctx->screen->family > CHIP_R600) && 199 (ctx->screen->family < CHIP_RV770) && reg[i+j].flags & REG_FLAG_RV6XX_SBU) { 200 block->pm4[block->pm4_ndwords++] = PKT3(PKT3_SURFACE_BASE_UPDATE, 0, 0); 201 block->pm4[block->pm4_ndwords++] = reg[i+j].flush_flags; 202 } 203 } 204 for (j = 0; j < n; j++) { 205 if (reg[i+j].flush_flags) { 206 block->pm4_flush_ndwords += 7; 207 } 208 } 209 /* check that we stay in limit */ 210 assert(block->pm4_ndwords < R600_BLOCK_MAX_REG); 211} 212 213int r600_context_add_block(struct r600_context *ctx, const struct r600_reg *reg, unsigned nreg, 214 unsigned opcode, unsigned offset_base) 215{ 216 struct r600_block *block; 217 struct r600_range *range; 218 int offset; 219 220 for (unsigned i = 0, n = 0; i < nreg; i += n) { 221 /* ignore new block balise */ 222 if (reg[i].offset == GROUP_FORCE_NEW_BLOCK) { 223 n = 1; 224 continue; 225 } 226 227 /* ignore regs not on R600 on R600 */ 228 if ((reg[i].flags & REG_FLAG_NOT_R600) && ctx->screen->family == CHIP_R600) { 229 n = 1; 230 continue; 231 } 232 233 /* register that need relocation are in their own group */ 234 /* find number of consecutive registers */ 235 n = 0; 236 offset = reg[i].offset; 237 while (reg[i + n].offset == offset) { 238 n++; 239 offset += 4; 240 if ((n + i) >= nreg) 241 break; 242 if (n >= (R600_BLOCK_MAX_REG - 2)) 243 break; 244 } 245 246 /* allocate new block */ 247 block = calloc(1, sizeof(struct r600_block)); 248 if (block == NULL) { 249 return -ENOMEM; 250 } 251 ctx->nblocks++; 252 for (int j = 0; j < n; j++) { 253 range = &ctx->range[CTX_RANGE_ID(reg[i + j].offset)]; 254 /* create block table if it doesn't exist */ 255 if (!range->blocks) 256 range->blocks = calloc(1 << HASH_SHIFT, sizeof(void *)); 257 if (!range->blocks) 258 return -1; 259 260 range->blocks[CTX_BLOCK_ID(reg[i + j].offset)] = block; 261 } 262 263 r600_init_block(ctx, block, reg, i, n, opcode, offset_base); 264 265 } 266 return 0; 267} 268 269/* R600/R700 configuration */ 270static const struct r600_reg r600_config_reg_list[] = { 271 {R_008958_VGT_PRIMITIVE_TYPE, 0, 0, 0}, 272 {R_008C00_SQ_CONFIG, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0, 0}, 273 {R_008C04_SQ_GPR_RESOURCE_MGMT_1, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0, 0}, 274 {R_008C08_SQ_GPR_RESOURCE_MGMT_2, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0, 0}, 275 {R_008C0C_SQ_THREAD_RESOURCE_MGMT, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0, 0}, 276 {R_008C10_SQ_STACK_RESOURCE_MGMT_1, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0, 0}, 277 {R_008C14_SQ_STACK_RESOURCE_MGMT_2, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0, 0}, 278 {R_008D8C_SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0, 0}, 279 {R_009508_TA_CNTL_AUX, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0, 0}, 280 {R_009714_VC_ENHANCE, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0, 0}, 281 {R_009830_DB_DEBUG, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0, 0}, 282 {R_009838_DB_WATERMARKS, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0, 0}, 283}; 284 285static const struct r600_reg r600_ctl_const_list[] = { 286 {R_03CFF0_SQ_VTX_BASE_VTX_LOC, 0, 0, 0}, 287 {R_03CFF4_SQ_VTX_START_INST_LOC, 0, 0, 0}, 288}; 289 290static const struct r600_reg r600_context_reg_list[] = { 291 {R_028350_SX_MISC, 0, 0, 0}, 292 {R_0286C8_SPI_THREAD_GROUPING, 0, 0, 0}, 293 {R_0288A8_SQ_ESGS_RING_ITEMSIZE, 0, 0, 0}, 294 {R_0288AC_SQ_GSVS_RING_ITEMSIZE, 0, 0, 0}, 295 {R_0288B0_SQ_ESTMP_RING_ITEMSIZE, 0, 0, 0}, 296 {R_0288B4_SQ_GSTMP_RING_ITEMSIZE, 0, 0, 0}, 297 {R_0288B8_SQ_VSTMP_RING_ITEMSIZE, 0, 0, 0}, 298 {R_0288BC_SQ_PSTMP_RING_ITEMSIZE, 0, 0, 0}, 299 {R_0288C0_SQ_FBUF_RING_ITEMSIZE, 0, 0, 0}, 300 {R_0288C4_SQ_REDUC_RING_ITEMSIZE, 0, 0, 0}, 301 {R_0288C8_SQ_GS_VERT_ITEMSIZE, 0, 0, 0}, 302 {R_028A10_VGT_OUTPUT_PATH_CNTL, 0, 0, 0}, 303 {R_028A14_VGT_HOS_CNTL, 0, 0, 0}, 304 {R_028A18_VGT_HOS_MAX_TESS_LEVEL, 0, 0, 0}, 305 {R_028A1C_VGT_HOS_MIN_TESS_LEVEL, 0, 0, 0}, 306 {R_028A20_VGT_HOS_REUSE_DEPTH, 0, 0, 0}, 307 {R_028A24_VGT_GROUP_PRIM_TYPE, 0, 0, 0}, 308 {R_028A28_VGT_GROUP_FIRST_DECR, 0, 0, 0}, 309 {R_028A2C_VGT_GROUP_DECR, 0, 0, 0}, 310 {R_028A30_VGT_GROUP_VECT_0_CNTL, 0, 0, 0}, 311 {R_028A34_VGT_GROUP_VECT_1_CNTL, 0, 0, 0}, 312 {R_028A38_VGT_GROUP_VECT_0_FMT_CNTL, 0, 0, 0}, 313 {R_028A3C_VGT_GROUP_VECT_1_FMT_CNTL, 0, 0, 0}, 314 {R_028A40_VGT_GS_MODE, 0, 0, 0}, 315 {R_028A4C_PA_SC_MODE_CNTL, 0, 0, 0}, 316 {R_028AB0_VGT_STRMOUT_EN, 0, 0, 0}, 317 {R_028AB4_VGT_REUSE_OFF, 0, 0, 0}, 318 {R_028AB8_VGT_VTX_CNT_EN, 0, 0, 0}, 319 {R_028B20_VGT_STRMOUT_BUFFER_EN, 0, 0, 0}, 320 {R_028028_DB_STENCIL_CLEAR, 0, 0, 0}, 321 {R_02802C_DB_DEPTH_CLEAR, 0, 0, 0}, 322 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 323 {R_028040_CB_COLOR0_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(0), 0}, 324 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 325 {R_0280A0_CB_COLOR0_INFO, REG_FLAG_NEED_BO, 0, 0xFFFFFFFF}, 326 {R_028060_CB_COLOR0_SIZE, 0, 0, 0}, 327 {R_028080_CB_COLOR0_VIEW, 0, 0, 0}, 328 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 329 {R_0280E0_CB_COLOR0_FRAG, REG_FLAG_NEED_BO, 0, 0}, 330 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 331 {R_0280C0_CB_COLOR0_TILE, REG_FLAG_NEED_BO, 0, 0}, 332 {R_028100_CB_COLOR0_MASK, 0, 0, 0}, 333 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 334 {R_028044_CB_COLOR1_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(1), 0}, 335 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 336 {R_0280A4_CB_COLOR1_INFO, REG_FLAG_NEED_BO, 0, 0xFFFFFFFF}, 337 {R_028064_CB_COLOR1_SIZE, 0, 0, 0}, 338 {R_028084_CB_COLOR1_VIEW, 0, 0, 0}, 339 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 340 {R_0280E4_CB_COLOR1_FRAG, REG_FLAG_NEED_BO, 0, 0}, 341 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 342 {R_0280C4_CB_COLOR1_TILE, REG_FLAG_NEED_BO, 0, 0}, 343 {R_028104_CB_COLOR1_MASK, 0, 0, 0}, 344 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 345 {R_028048_CB_COLOR2_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(2), 0}, 346 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 347 {R_0280A8_CB_COLOR2_INFO, REG_FLAG_NEED_BO, 0, 0xFFFFFFFF}, 348 {R_028068_CB_COLOR2_SIZE, 0, 0, 0}, 349 {R_028088_CB_COLOR2_VIEW, 0, 0, 0}, 350 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 351 {R_0280E8_CB_COLOR2_FRAG, REG_FLAG_NEED_BO, 0, 0}, 352 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 353 {R_0280C8_CB_COLOR2_TILE, REG_FLAG_NEED_BO, 0, 0}, 354 {R_028108_CB_COLOR2_MASK, 0, 0, 0}, 355 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 356 {R_02804C_CB_COLOR3_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(3), 0}, 357 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 358 {R_0280AC_CB_COLOR3_INFO, REG_FLAG_NEED_BO, 0, 0xFFFFFFFF}, 359 {R_02806C_CB_COLOR3_SIZE, 0, 0, 0}, 360 {R_02808C_CB_COLOR3_VIEW, 0, 0, 0}, 361 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 362 {R_0280EC_CB_COLOR3_FRAG, REG_FLAG_NEED_BO, 0, 0}, 363 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 364 {R_0280CC_CB_COLOR3_TILE, REG_FLAG_NEED_BO, 0, 0}, 365 {R_02810C_CB_COLOR3_MASK, 0, 0, 0}, 366 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 367 {R_028050_CB_COLOR4_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(4), 0}, 368 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 369 {R_0280B0_CB_COLOR4_INFO, REG_FLAG_NEED_BO, 0, 0xFFFFFFFF}, 370 {R_028070_CB_COLOR4_SIZE, 0, 0, 0}, 371 {R_028090_CB_COLOR4_VIEW, 0, 0, 0}, 372 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 373 {R_0280F0_CB_COLOR4_FRAG, REG_FLAG_NEED_BO, 0, 0}, 374 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 375 {R_0280D0_CB_COLOR4_TILE, REG_FLAG_NEED_BO, 0, 0}, 376 {R_028110_CB_COLOR4_MASK, 0, 0, 0}, 377 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 378 {R_028054_CB_COLOR5_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(5), 0}, 379 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 380 {R_0280B4_CB_COLOR5_INFO, REG_FLAG_NEED_BO, 0, 0xFFFFFFFF}, 381 {R_028074_CB_COLOR5_SIZE, 0, 0, 0}, 382 {R_028094_CB_COLOR5_VIEW, 0, 0, 0}, 383 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 384 {R_0280F4_CB_COLOR5_FRAG, REG_FLAG_NEED_BO, 0, 0}, 385 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 386 {R_0280D4_CB_COLOR5_TILE, REG_FLAG_NEED_BO, 0, 0}, 387 {R_028114_CB_COLOR5_MASK, 0, 0, 0}, 388 {R_028058_CB_COLOR6_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(6), 0}, 389 {R_0280B8_CB_COLOR6_INFO, REG_FLAG_NEED_BO, 0, 0xFFFFFFFF}, 390 {R_028078_CB_COLOR6_SIZE, 0, 0, 0}, 391 {R_028098_CB_COLOR6_VIEW, 0, 0, 0}, 392 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 393 {R_0280F8_CB_COLOR6_FRAG, REG_FLAG_NEED_BO, 0, 0}, 394 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 395 {R_0280D8_CB_COLOR6_TILE, REG_FLAG_NEED_BO, 0, 0}, 396 {R_028118_CB_COLOR6_MASK, 0, 0, 0}, 397 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 398 {R_02805C_CB_COLOR7_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_COLOR(7), 0}, 399 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 400 {R_0280BC_CB_COLOR7_INFO, REG_FLAG_NEED_BO, 0, 0xFFFFFFFF}, 401 {R_02807C_CB_COLOR7_SIZE, 0, 0, 0}, 402 {R_02809C_CB_COLOR7_VIEW, 0, 0, 0}, 403 {R_0280FC_CB_COLOR7_FRAG, REG_FLAG_NEED_BO, 0, 0}, 404 {R_0280DC_CB_COLOR7_TILE, REG_FLAG_NEED_BO, 0, 0}, 405 {R_02811C_CB_COLOR7_MASK, 0, 0, 0}, 406 {R_028120_CB_CLEAR_RED, 0, 0, 0}, 407 {R_028124_CB_CLEAR_GREEN, 0, 0, 0}, 408 {R_028128_CB_CLEAR_BLUE, 0, 0, 0}, 409 {R_02812C_CB_CLEAR_ALPHA, 0, 0, 0}, 410 {R_028140_ALU_CONST_BUFFER_SIZE_PS_0, REG_FLAG_DIRTY_ALWAYS, 0, 0}, 411 {R_028180_ALU_CONST_BUFFER_SIZE_VS_0, REG_FLAG_DIRTY_ALWAYS, 0, 0}, 412 {R_028940_ALU_CONST_CACHE_PS_0, REG_FLAG_NEED_BO, S_0085F0_SH_ACTION_ENA(1), 0xFFFFFFFF}, 413 {R_028980_ALU_CONST_CACHE_VS_0, REG_FLAG_NEED_BO, S_0085F0_SH_ACTION_ENA(1), 0xFFFFFFFF}, 414 {R_02823C_CB_SHADER_MASK, 0, 0, 0}, 415 {R_028238_CB_TARGET_MASK, 0, 0, 0}, 416 {R_028410_SX_ALPHA_TEST_CONTROL, 0, 0, 0}, 417 {R_028414_CB_BLEND_RED, 0, 0, 0}, 418 {R_028418_CB_BLEND_GREEN, 0, 0, 0}, 419 {R_02841C_CB_BLEND_BLUE, 0, 0, 0}, 420 {R_028420_CB_BLEND_ALPHA, 0, 0, 0}, 421 {R_028424_CB_FOG_RED, 0, 0, 0}, 422 {R_028428_CB_FOG_GREEN, 0, 0, 0}, 423 {R_02842C_CB_FOG_BLUE, 0, 0, 0}, 424 {R_028430_DB_STENCILREFMASK, 0, 0, 0}, 425 {R_028434_DB_STENCILREFMASK_BF, 0, 0, 0}, 426 {R_028438_SX_ALPHA_REF, 0, 0, 0}, 427 {R_0286DC_SPI_FOG_CNTL, 0, 0, 0}, 428 {R_0286E0_SPI_FOG_FUNC_SCALE, 0, 0, 0}, 429 {R_0286E4_SPI_FOG_FUNC_BIAS, 0, 0, 0}, 430 {R_028780_CB_BLEND0_CONTROL, REG_FLAG_NOT_R600, 0, 0}, 431 {R_028784_CB_BLEND1_CONTROL, REG_FLAG_NOT_R600, 0, 0}, 432 {R_028788_CB_BLEND2_CONTROL, REG_FLAG_NOT_R600, 0, 0}, 433 {R_02878C_CB_BLEND3_CONTROL, REG_FLAG_NOT_R600, 0, 0}, 434 {R_028790_CB_BLEND4_CONTROL, REG_FLAG_NOT_R600, 0, 0}, 435 {R_028794_CB_BLEND5_CONTROL, REG_FLAG_NOT_R600, 0, 0}, 436 {R_028798_CB_BLEND6_CONTROL, REG_FLAG_NOT_R600, 0, 0}, 437 {R_02879C_CB_BLEND7_CONTROL, REG_FLAG_NOT_R600, 0, 0}, 438 {R_0287A0_CB_SHADER_CONTROL, 0, 0, 0}, 439 {R_028800_DB_DEPTH_CONTROL, 0, 0, 0}, 440 {R_028804_CB_BLEND_CONTROL, 0, 0, 0}, 441 {R_028808_CB_COLOR_CONTROL, 0, 0, 0}, 442 {R_02880C_DB_SHADER_CONTROL, 0, 0, 0}, 443 {R_028C04_PA_SC_AA_CONFIG, 0, 0, 0}, 444 {R_028C1C_PA_SC_AA_SAMPLE_LOCS_MCTX, 0, 0, 0}, 445 {R_028C20_PA_SC_AA_SAMPLE_LOCS_8S_WD1_MCTX, 0, 0, 0}, 446 {R_028C30_CB_CLRCMP_CONTROL, 0, 0, 0}, 447 {R_028C34_CB_CLRCMP_SRC, 0, 0, 0}, 448 {R_028C38_CB_CLRCMP_DST, 0, 0, 0}, 449 {R_028C3C_CB_CLRCMP_MSK, 0, 0, 0}, 450 {R_028C48_PA_SC_AA_MASK, 0, 0, 0}, 451 {R_028D2C_DB_SRESULTS_COMPARE_STATE1, 0, 0, 0}, 452 {R_028D44_DB_ALPHA_TO_MASK, 0, 0, 0}, 453 {R_02800C_DB_DEPTH_BASE, REG_FLAG_NEED_BO|REG_FLAG_RV6XX_SBU, SURFACE_BASE_UPDATE_DEPTH, 0}, 454 {R_028000_DB_DEPTH_SIZE, 0, 0, 0}, 455 {R_028004_DB_DEPTH_VIEW, 0, 0, 0}, 456 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 457 {R_028010_DB_DEPTH_INFO, REG_FLAG_NEED_BO, 0, 0}, 458 {R_028D0C_DB_RENDER_CONTROL, 0, 0, 0}, 459 {R_028D10_DB_RENDER_OVERRIDE, 0, 0, 0}, 460 {R_028D24_DB_HTILE_SURFACE, 0, 0, 0}, 461 {R_028D30_DB_PRELOAD_CONTROL, 0, 0, 0}, 462 {R_028D34_DB_PREFETCH_LIMIT, 0, 0, 0}, 463 {R_028030_PA_SC_SCREEN_SCISSOR_TL, 0, 0, 0}, 464 {R_028034_PA_SC_SCREEN_SCISSOR_BR, 0, 0, 0}, 465 {R_028200_PA_SC_WINDOW_OFFSET, 0, 0, 0}, 466 {R_028204_PA_SC_WINDOW_SCISSOR_TL, 0, 0, 0}, 467 {R_028208_PA_SC_WINDOW_SCISSOR_BR, 0, 0, 0}, 468 {R_02820C_PA_SC_CLIPRECT_RULE, 0, 0, 0}, 469 {R_028210_PA_SC_CLIPRECT_0_TL, 0, 0, 0}, 470 {R_028214_PA_SC_CLIPRECT_0_BR, 0, 0, 0}, 471 {R_028218_PA_SC_CLIPRECT_1_TL, 0, 0, 0}, 472 {R_02821C_PA_SC_CLIPRECT_1_BR, 0, 0, 0}, 473 {R_028220_PA_SC_CLIPRECT_2_TL, 0, 0, 0}, 474 {R_028224_PA_SC_CLIPRECT_2_BR, 0, 0, 0}, 475 {R_028228_PA_SC_CLIPRECT_3_TL, 0, 0, 0}, 476 {R_02822C_PA_SC_CLIPRECT_3_BR, 0, 0, 0}, 477 {R_028230_PA_SC_EDGERULE, 0, 0, 0}, 478 {R_028240_PA_SC_GENERIC_SCISSOR_TL, 0, 0, 0}, 479 {R_028244_PA_SC_GENERIC_SCISSOR_BR, 0, 0, 0}, 480 {R_028250_PA_SC_VPORT_SCISSOR_0_TL, 0, 0, 0}, 481 {R_028254_PA_SC_VPORT_SCISSOR_0_BR, 0, 0, 0}, 482 {R_0282D0_PA_SC_VPORT_ZMIN_0, 0, 0, 0}, 483 {R_0282D4_PA_SC_VPORT_ZMAX_0, 0, 0, 0}, 484 {R_02843C_PA_CL_VPORT_XSCALE_0, 0, 0, 0}, 485 {R_028440_PA_CL_VPORT_XOFFSET_0, 0, 0, 0}, 486 {R_028444_PA_CL_VPORT_YSCALE_0, 0, 0, 0}, 487 {R_028448_PA_CL_VPORT_YOFFSET_0, 0, 0, 0}, 488 {R_02844C_PA_CL_VPORT_ZSCALE_0, 0, 0, 0}, 489 {R_028450_PA_CL_VPORT_ZOFFSET_0, 0, 0, 0}, 490 {R_0286D4_SPI_INTERP_CONTROL_0, 0, 0, 0}, 491 {R_028810_PA_CL_CLIP_CNTL, 0, 0, 0}, 492 {R_028814_PA_SU_SC_MODE_CNTL, 0, 0, 0}, 493 {R_028818_PA_CL_VTE_CNTL, 0, 0, 0}, 494 {R_02881C_PA_CL_VS_OUT_CNTL, 0, 0, 0}, 495 {R_028820_PA_CL_NANINF_CNTL, 0, 0, 0}, 496 {R_028A00_PA_SU_POINT_SIZE, 0, 0, 0}, 497 {R_028A04_PA_SU_POINT_MINMAX, 0, 0, 0}, 498 {R_028A08_PA_SU_LINE_CNTL, 0, 0, 0}, 499 {R_028A0C_PA_SC_LINE_STIPPLE, 0, 0, 0}, 500 {R_028A48_PA_SC_MPASS_PS_CNTL, 0, 0, 0}, 501 {R_028C00_PA_SC_LINE_CNTL, 0, 0, 0}, 502 {R_028C08_PA_SU_VTX_CNTL, 0, 0, 0}, 503 {R_028C0C_PA_CL_GB_VERT_CLIP_ADJ, 0, 0, 0}, 504 {R_028C10_PA_CL_GB_VERT_DISC_ADJ, 0, 0, 0}, 505 {R_028C14_PA_CL_GB_HORZ_CLIP_ADJ, 0, 0, 0}, 506 {R_028C18_PA_CL_GB_HORZ_DISC_ADJ, 0, 0, 0}, 507 {R_028DF8_PA_SU_POLY_OFFSET_DB_FMT_CNTL, 0, 0, 0}, 508 {R_028DFC_PA_SU_POLY_OFFSET_CLAMP, 0, 0, 0}, 509 {R_028E00_PA_SU_POLY_OFFSET_FRONT_SCALE, 0, 0, 0}, 510 {R_028E04_PA_SU_POLY_OFFSET_FRONT_OFFSET, 0, 0, 0}, 511 {R_028E08_PA_SU_POLY_OFFSET_BACK_SCALE, 0, 0, 0}, 512 {R_028E0C_PA_SU_POLY_OFFSET_BACK_OFFSET, 0, 0, 0}, 513 {R_028E20_PA_CL_UCP0_X, 0, 0, 0}, 514 {R_028E24_PA_CL_UCP0_Y, 0, 0, 0}, 515 {R_028E28_PA_CL_UCP0_Z, 0, 0, 0}, 516 {R_028E2C_PA_CL_UCP0_W, 0, 0, 0}, 517 {R_028E30_PA_CL_UCP1_X, 0, 0, 0}, 518 {R_028E34_PA_CL_UCP1_Y, 0, 0, 0}, 519 {R_028E38_PA_CL_UCP1_Z, 0, 0, 0}, 520 {R_028E3C_PA_CL_UCP1_W, 0, 0, 0}, 521 {R_028E40_PA_CL_UCP2_X, 0, 0, 0}, 522 {R_028E44_PA_CL_UCP2_Y, 0, 0, 0}, 523 {R_028E48_PA_CL_UCP2_Z, 0, 0, 0}, 524 {R_028E4C_PA_CL_UCP2_W, 0, 0, 0}, 525 {R_028E50_PA_CL_UCP3_X, 0, 0, 0}, 526 {R_028E54_PA_CL_UCP3_Y, 0, 0, 0}, 527 {R_028E58_PA_CL_UCP3_Z, 0, 0, 0}, 528 {R_028E5C_PA_CL_UCP3_W, 0, 0, 0}, 529 {R_028E60_PA_CL_UCP4_X, 0, 0, 0}, 530 {R_028E64_PA_CL_UCP4_Y, 0, 0, 0}, 531 {R_028E68_PA_CL_UCP4_Z, 0, 0, 0}, 532 {R_028E6C_PA_CL_UCP4_W, 0, 0, 0}, 533 {R_028E70_PA_CL_UCP5_X, 0, 0, 0}, 534 {R_028E74_PA_CL_UCP5_Y, 0, 0, 0}, 535 {R_028E78_PA_CL_UCP5_Z, 0, 0, 0}, 536 {R_028E7C_PA_CL_UCP5_W, 0, 0, 0}, 537 {R_028380_SQ_VTX_SEMANTIC_0, 0, 0, 0}, 538 {R_028384_SQ_VTX_SEMANTIC_1, 0, 0, 0}, 539 {R_028388_SQ_VTX_SEMANTIC_2, 0, 0, 0}, 540 {R_02838C_SQ_VTX_SEMANTIC_3, 0, 0, 0}, 541 {R_028390_SQ_VTX_SEMANTIC_4, 0, 0, 0}, 542 {R_028394_SQ_VTX_SEMANTIC_5, 0, 0, 0}, 543 {R_028398_SQ_VTX_SEMANTIC_6, 0, 0, 0}, 544 {R_02839C_SQ_VTX_SEMANTIC_7, 0, 0, 0}, 545 {R_0283A0_SQ_VTX_SEMANTIC_8, 0, 0, 0}, 546 {R_0283A4_SQ_VTX_SEMANTIC_9, 0, 0, 0}, 547 {R_0283A8_SQ_VTX_SEMANTIC_10, 0, 0, 0}, 548 {R_0283AC_SQ_VTX_SEMANTIC_11, 0, 0, 0}, 549 {R_0283B0_SQ_VTX_SEMANTIC_12, 0, 0, 0}, 550 {R_0283B4_SQ_VTX_SEMANTIC_13, 0, 0, 0}, 551 {R_0283B8_SQ_VTX_SEMANTIC_14, 0, 0, 0}, 552 {R_0283BC_SQ_VTX_SEMANTIC_15, 0, 0, 0}, 553 {R_0283C0_SQ_VTX_SEMANTIC_16, 0, 0, 0}, 554 {R_0283C4_SQ_VTX_SEMANTIC_17, 0, 0, 0}, 555 {R_0283C8_SQ_VTX_SEMANTIC_18, 0, 0, 0}, 556 {R_0283CC_SQ_VTX_SEMANTIC_19, 0, 0, 0}, 557 {R_0283D0_SQ_VTX_SEMANTIC_20, 0, 0, 0}, 558 {R_0283D4_SQ_VTX_SEMANTIC_21, 0, 0, 0}, 559 {R_0283D8_SQ_VTX_SEMANTIC_22, 0, 0, 0}, 560 {R_0283DC_SQ_VTX_SEMANTIC_23, 0, 0, 0}, 561 {R_0283E0_SQ_VTX_SEMANTIC_24, 0, 0, 0}, 562 {R_0283E4_SQ_VTX_SEMANTIC_25, 0, 0, 0}, 563 {R_0283E8_SQ_VTX_SEMANTIC_26, 0, 0, 0}, 564 {R_0283EC_SQ_VTX_SEMANTIC_27, 0, 0, 0}, 565 {R_0283F0_SQ_VTX_SEMANTIC_28, 0, 0, 0}, 566 {R_0283F4_SQ_VTX_SEMANTIC_29, 0, 0, 0}, 567 {R_0283F8_SQ_VTX_SEMANTIC_30, 0, 0, 0}, 568 {R_0283FC_SQ_VTX_SEMANTIC_31, 0, 0, 0}, 569 {R_028614_SPI_VS_OUT_ID_0, 0, 0, 0}, 570 {R_028618_SPI_VS_OUT_ID_1, 0, 0, 0}, 571 {R_02861C_SPI_VS_OUT_ID_2, 0, 0, 0}, 572 {R_028620_SPI_VS_OUT_ID_3, 0, 0, 0}, 573 {R_028624_SPI_VS_OUT_ID_4, 0, 0, 0}, 574 {R_028628_SPI_VS_OUT_ID_5, 0, 0, 0}, 575 {R_02862C_SPI_VS_OUT_ID_6, 0, 0, 0}, 576 {R_028630_SPI_VS_OUT_ID_7, 0, 0, 0}, 577 {R_028634_SPI_VS_OUT_ID_8, 0, 0, 0}, 578 {R_028638_SPI_VS_OUT_ID_9, 0, 0, 0}, 579 {R_0286C4_SPI_VS_OUT_CONFIG, 0, 0, 0}, 580 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 581 {R_028858_SQ_PGM_START_VS, REG_FLAG_NEED_BO, S_0085F0_SH_ACTION_ENA(1), 0xFFFFFFFF}, 582 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 583 {R_028868_SQ_PGM_RESOURCES_VS, 0, 0, 0}, 584 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 585 {R_028894_SQ_PGM_START_FS, REG_FLAG_NEED_BO, S_0085F0_SH_ACTION_ENA(1), 0xFFFFFFFF}, 586 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 587 {R_0288A4_SQ_PGM_RESOURCES_FS, 0, 0, 0}, 588 {R_0288D0_SQ_PGM_CF_OFFSET_VS, 0, 0, 0}, 589 {R_0288DC_SQ_PGM_CF_OFFSET_FS, 0, 0, 0}, 590 {R_028644_SPI_PS_INPUT_CNTL_0, 0, 0, 0}, 591 {R_028648_SPI_PS_INPUT_CNTL_1, 0, 0, 0}, 592 {R_02864C_SPI_PS_INPUT_CNTL_2, 0, 0, 0}, 593 {R_028650_SPI_PS_INPUT_CNTL_3, 0, 0, 0}, 594 {R_028654_SPI_PS_INPUT_CNTL_4, 0, 0, 0}, 595 {R_028658_SPI_PS_INPUT_CNTL_5, 0, 0, 0}, 596 {R_02865C_SPI_PS_INPUT_CNTL_6, 0, 0, 0}, 597 {R_028660_SPI_PS_INPUT_CNTL_7, 0, 0, 0}, 598 {R_028664_SPI_PS_INPUT_CNTL_8, 0, 0, 0}, 599 {R_028668_SPI_PS_INPUT_CNTL_9, 0, 0, 0}, 600 {R_02866C_SPI_PS_INPUT_CNTL_10, 0, 0, 0}, 601 {R_028670_SPI_PS_INPUT_CNTL_11, 0, 0, 0}, 602 {R_028674_SPI_PS_INPUT_CNTL_12, 0, 0, 0}, 603 {R_028678_SPI_PS_INPUT_CNTL_13, 0, 0, 0}, 604 {R_02867C_SPI_PS_INPUT_CNTL_14, 0, 0, 0}, 605 {R_028680_SPI_PS_INPUT_CNTL_15, 0, 0, 0}, 606 {R_028684_SPI_PS_INPUT_CNTL_16, 0, 0, 0}, 607 {R_028688_SPI_PS_INPUT_CNTL_17, 0, 0, 0}, 608 {R_02868C_SPI_PS_INPUT_CNTL_18, 0, 0, 0}, 609 {R_028690_SPI_PS_INPUT_CNTL_19, 0, 0, 0}, 610 {R_028694_SPI_PS_INPUT_CNTL_20, 0, 0, 0}, 611 {R_028698_SPI_PS_INPUT_CNTL_21, 0, 0, 0}, 612 {R_02869C_SPI_PS_INPUT_CNTL_22, 0, 0, 0}, 613 {R_0286A0_SPI_PS_INPUT_CNTL_23, 0, 0, 0}, 614 {R_0286A4_SPI_PS_INPUT_CNTL_24, 0, 0, 0}, 615 {R_0286A8_SPI_PS_INPUT_CNTL_25, 0, 0, 0}, 616 {R_0286AC_SPI_PS_INPUT_CNTL_26, 0, 0, 0}, 617 {R_0286B0_SPI_PS_INPUT_CNTL_27, 0, 0, 0}, 618 {R_0286B4_SPI_PS_INPUT_CNTL_28, 0, 0, 0}, 619 {R_0286B8_SPI_PS_INPUT_CNTL_29, 0, 0, 0}, 620 {R_0286BC_SPI_PS_INPUT_CNTL_30, 0, 0, 0}, 621 {R_0286C0_SPI_PS_INPUT_CNTL_31, 0, 0, 0}, 622 {R_0286CC_SPI_PS_IN_CONTROL_0, 0, 0, 0}, 623 {R_0286D0_SPI_PS_IN_CONTROL_1, 0, 0, 0}, 624 {R_0286D8_SPI_INPUT_Z, 0, 0, 0}, 625 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 626 {R_028840_SQ_PGM_START_PS, REG_FLAG_NEED_BO, S_0085F0_SH_ACTION_ENA(1), 0xFFFFFFFF}, 627 {GROUP_FORCE_NEW_BLOCK, 0, 0, 0}, 628 {R_028850_SQ_PGM_RESOURCES_PS, 0, 0, 0}, 629 {R_028854_SQ_PGM_EXPORTS_PS, 0, 0, 0}, 630 {R_0288CC_SQ_PGM_CF_OFFSET_PS, 0, 0, 0}, 631 {R_028400_VGT_MAX_VTX_INDX, 0, 0, 0}, 632 {R_028404_VGT_MIN_VTX_INDX, 0, 0, 0}, 633 {R_028408_VGT_INDX_OFFSET, 0, 0, 0}, 634 {R_02840C_VGT_MULTI_PRIM_IB_RESET_INDX, 0, 0, 0}, 635 {R_028A84_VGT_PRIMITIVEID_EN, 0, 0, 0}, 636 {R_028A94_VGT_MULTI_PRIM_IB_RESET_EN, 0, 0, 0}, 637 {R_028AA0_VGT_INSTANCE_STEP_RATE_0, 0, 0, 0}, 638 {R_028AA4_VGT_INSTANCE_STEP_RATE_1, 0, 0, 0}, 639}; 640 641/* SHADER RESOURCE R600/R700 */ 642int r600_resource_init(struct r600_context *ctx, struct r600_range *range, unsigned offset, unsigned nblocks, unsigned stride, struct r600_reg *reg, int nreg, unsigned offset_base) 643{ 644 int i; 645 struct r600_block *block; 646 range->blocks = calloc(nblocks, sizeof(struct r600_block *)); 647 if (range->blocks == NULL) 648 return -ENOMEM; 649 650 reg[0].offset += offset; 651 for (i = 0; i < nblocks; i++) { 652 block = calloc(1, sizeof(struct r600_block)); 653 if (block == NULL) { 654 return -ENOMEM; 655 } 656 ctx->nblocks++; 657 range->blocks[i] = block; 658 r600_init_block(ctx, block, reg, 0, nreg, PKT3_SET_RESOURCE, offset_base); 659 660 reg[0].offset += stride; 661 } 662 return 0; 663} 664 665 666static int r600_resource_range_init(struct r600_context *ctx, struct r600_range *range, unsigned offset, unsigned nblocks, unsigned stride) 667{ 668 struct r600_reg r600_shader_resource[] = { 669 {R_038000_RESOURCE0_WORD0, REG_FLAG_NEED_BO, S_0085F0_TC_ACTION_ENA(1) | S_0085F0_VC_ACTION_ENA(1), 0xFFFFFFFF}, 670 {R_038004_RESOURCE0_WORD1, REG_FLAG_NEED_BO, S_0085F0_TC_ACTION_ENA(1) | S_0085F0_VC_ACTION_ENA(1), 0xFFFFFFFF}, 671 {R_038008_RESOURCE0_WORD2, 0, 0, 0}, 672 {R_03800C_RESOURCE0_WORD3, 0, 0, 0}, 673 {R_038010_RESOURCE0_WORD4, 0, 0, 0}, 674 {R_038014_RESOURCE0_WORD5, 0, 0, 0}, 675 {R_038018_RESOURCE0_WORD6, 0, 0, 0}, 676 }; 677 unsigned nreg = Elements(r600_shader_resource); 678 679 return r600_resource_init(ctx, range, offset, nblocks, stride, r600_shader_resource, nreg, R600_RESOURCE_OFFSET); 680} 681 682/* SHADER SAMPLER R600/R700 */ 683static int r600_state_sampler_init(struct r600_context *ctx, u32 offset) 684{ 685 struct r600_reg r600_shader_sampler[] = { 686 {R_03C000_SQ_TEX_SAMPLER_WORD0_0, 0, 0, 0}, 687 {R_03C004_SQ_TEX_SAMPLER_WORD1_0, 0, 0, 0}, 688 {R_03C008_SQ_TEX_SAMPLER_WORD2_0, 0, 0, 0}, 689 }; 690 unsigned nreg = Elements(r600_shader_sampler); 691 692 for (int i = 0; i < nreg; i++) { 693 r600_shader_sampler[i].offset += offset; 694 } 695 return r600_context_add_block(ctx, r600_shader_sampler, nreg, PKT3_SET_SAMPLER, R600_SAMPLER_OFFSET); 696} 697 698/* SHADER SAMPLER BORDER R600/R700 */ 699static int r600_state_sampler_border_init(struct r600_context *ctx, u32 offset) 700{ 701 struct r600_reg r600_shader_sampler_border[] = { 702 {R_00A400_TD_PS_SAMPLER0_BORDER_RED, 0, 0, 0}, 703 {R_00A404_TD_PS_SAMPLER0_BORDER_GREEN, 0, 0, 0}, 704 {R_00A408_TD_PS_SAMPLER0_BORDER_BLUE, 0, 0, 0}, 705 {R_00A40C_TD_PS_SAMPLER0_BORDER_ALPHA, 0, 0, 0}, 706 }; 707 unsigned nreg = Elements(r600_shader_sampler_border); 708 709 for (int i = 0; i < nreg; i++) { 710 r600_shader_sampler_border[i].offset += offset; 711 } 712 return r600_context_add_block(ctx, r600_shader_sampler_border, nreg, PKT3_SET_CONFIG_REG, R600_CONFIG_REG_OFFSET); 713} 714 715static int r600_loop_const_init(struct r600_context *ctx, u32 offset) 716{ 717 unsigned nreg = 32; 718 struct r600_reg r600_loop_consts[32]; 719 int i; 720 721 for (i = 0; i < nreg; i++) { 722 r600_loop_consts[i].offset = R600_LOOP_CONST_OFFSET + ((offset + i) * 4); 723 r600_loop_consts[i].flags = REG_FLAG_DIRTY_ALWAYS; 724 r600_loop_consts[i].flush_flags = 0; 725 r600_loop_consts[i].flush_mask = 0; 726 } 727 return r600_context_add_block(ctx, r600_loop_consts, nreg, PKT3_SET_LOOP_CONST, R600_LOOP_CONST_OFFSET); 728} 729 730static void r600_free_resource_range(struct r600_context *ctx, struct r600_range *range, int nblocks) 731{ 732 struct r600_block *block; 733 int i; 734 for (i = 0; i < nblocks; i++) { 735 block = range->blocks[i]; 736 if (block) { 737 for (int k = 1; k <= block->nbo; k++) 738 pipe_resource_reference((struct pipe_resource**)&block->reloc[k].bo, NULL); 739 free(block); 740 } 741 } 742 free(range->blocks); 743 744} 745 746/* initialize */ 747void r600_context_fini(struct r600_context *ctx) 748{ 749 struct r600_block *block; 750 struct r600_range *range; 751 752 for (int i = 0; i < NUM_RANGES; i++) { 753 if (!ctx->range[i].blocks) 754 continue; 755 for (int j = 0; j < (1 << HASH_SHIFT); j++) { 756 block = ctx->range[i].blocks[j]; 757 if (block) { 758 for (int k = 0, offset = block->start_offset; k < block->nreg; k++, offset += 4) { 759 range = &ctx->range[CTX_RANGE_ID(offset)]; 760 range->blocks[CTX_BLOCK_ID(offset)] = NULL; 761 } 762 for (int k = 1; k <= block->nbo; k++) { 763 pipe_resource_reference((struct pipe_resource**)&block->reloc[k].bo, NULL); 764 } 765 free(block); 766 } 767 } 768 free(ctx->range[i].blocks); 769 } 770 r600_free_resource_range(ctx, &ctx->ps_resources, ctx->num_ps_resources); 771 r600_free_resource_range(ctx, &ctx->vs_resources, ctx->num_vs_resources); 772 r600_free_resource_range(ctx, &ctx->fs_resources, ctx->num_fs_resources); 773 free(ctx->range); 774 free(ctx->blocks); 775 free(ctx->bo); 776 ctx->ws->cs_destroy(ctx->cs); 777 778 memset(ctx, 0, sizeof(struct r600_context)); 779} 780 781static void r600_add_resource_block(struct r600_context *ctx, struct r600_range *range, int num_blocks, int *index) 782{ 783 int c = *index; 784 for (int j = 0; j < num_blocks; j++) { 785 if (!range->blocks[j]) 786 continue; 787 788 ctx->blocks[c++] = range->blocks[j]; 789 } 790 *index = c; 791} 792 793int r600_setup_block_table(struct r600_context *ctx) 794{ 795 /* setup block table */ 796 int c = 0; 797 ctx->blocks = calloc(ctx->nblocks, sizeof(void*)); 798 if (!ctx->blocks) 799 return -ENOMEM; 800 for (int i = 0; i < NUM_RANGES; i++) { 801 if (!ctx->range[i].blocks) 802 continue; 803 for (int j = 0, add; j < (1 << HASH_SHIFT); j++) { 804 if (!ctx->range[i].blocks[j]) 805 continue; 806 807 add = 1; 808 for (int k = 0; k < c; k++) { 809 if (ctx->blocks[k] == ctx->range[i].blocks[j]) { 810 add = 0; 811 break; 812 } 813 } 814 if (add) { 815 assert(c < ctx->nblocks); 816 ctx->blocks[c++] = ctx->range[i].blocks[j]; 817 j += (ctx->range[i].blocks[j]->nreg) - 1; 818 } 819 } 820 } 821 822 r600_add_resource_block(ctx, &ctx->ps_resources, ctx->num_ps_resources, &c); 823 r600_add_resource_block(ctx, &ctx->vs_resources, ctx->num_vs_resources, &c); 824 r600_add_resource_block(ctx, &ctx->fs_resources, ctx->num_fs_resources, &c); 825 return 0; 826} 827 828int r600_context_init(struct r600_context *ctx, struct r600_screen *screen) 829{ 830 int r; 831 832 memset(ctx, 0, sizeof(struct r600_context)); 833 ctx->screen = screen; 834 ctx->ws = screen->ws; 835 836 LIST_INITHEAD(&ctx->active_query_list); 837 838 /* init dirty list */ 839 LIST_INITHEAD(&ctx->dirty); 840 LIST_INITHEAD(&ctx->resource_dirty); 841 LIST_INITHEAD(&ctx->enable_list); 842 843 ctx->range = calloc(NUM_RANGES, sizeof(struct r600_range)); 844 if (!ctx->range) { 845 r = -ENOMEM; 846 goto out_err; 847 } 848 849 /* add blocks */ 850 r = r600_context_add_block(ctx, r600_config_reg_list, 851 Elements(r600_config_reg_list), PKT3_SET_CONFIG_REG, R600_CONFIG_REG_OFFSET); 852 if (r) 853 goto out_err; 854 r = r600_context_add_block(ctx, r600_context_reg_list, 855 Elements(r600_context_reg_list), PKT3_SET_CONTEXT_REG, R600_CONTEXT_REG_OFFSET); 856 if (r) 857 goto out_err; 858 r = r600_context_add_block(ctx, r600_ctl_const_list, 859 Elements(r600_ctl_const_list), PKT3_SET_CTL_CONST, R600_CTL_CONST_OFFSET); 860 if (r) 861 goto out_err; 862 863 /* PS SAMPLER BORDER */ 864 for (int j = 0, offset = 0; j < 18; j++, offset += 0x10) { 865 r = r600_state_sampler_border_init(ctx, offset); 866 if (r) 867 goto out_err; 868 } 869 870 /* VS SAMPLER BORDER */ 871 for (int j = 0, offset = 0x200; j < 18; j++, offset += 0x10) { 872 r = r600_state_sampler_border_init(ctx, offset); 873 if (r) 874 goto out_err; 875 } 876 /* PS SAMPLER */ 877 for (int j = 0, offset = 0; j < 18; j++, offset += 0xC) { 878 r = r600_state_sampler_init(ctx, offset); 879 if (r) 880 goto out_err; 881 } 882 /* VS SAMPLER */ 883 for (int j = 0, offset = 0xD8; j < 18; j++, offset += 0xC) { 884 r = r600_state_sampler_init(ctx, offset); 885 if (r) 886 goto out_err; 887 } 888 889 ctx->num_ps_resources = 160; 890 ctx->num_vs_resources = 160; 891 ctx->num_fs_resources = 16; 892 r = r600_resource_range_init(ctx, &ctx->ps_resources, 0, 160, 0x1c); 893 if (r) 894 goto out_err; 895 r = r600_resource_range_init(ctx, &ctx->vs_resources, 0x1180, 160, 0x1c); 896 if (r) 897 goto out_err; 898 r = r600_resource_range_init(ctx, &ctx->fs_resources, 0x2300, 16, 0x1c); 899 if (r) 900 goto out_err; 901 902 /* PS loop const */ 903 r600_loop_const_init(ctx, 0); 904 /* VS loop const */ 905 r600_loop_const_init(ctx, 32); 906 907 r = r600_setup_block_table(ctx); 908 if (r) 909 goto out_err; 910 911 ctx->cs = screen->ws->cs_create(screen->ws); 912 913 /* allocate cs variables */ 914 ctx->bo = calloc(RADEON_MAX_CMDBUF_DWORDS, sizeof(void *)); 915 if (ctx->bo == NULL) { 916 r = -ENOMEM; 917 goto out_err; 918 } 919 ctx->pm4 = ctx->cs->buf; 920 921 r600_init_cs(ctx); 922 ctx->max_db = 4; 923 return 0; 924out_err: 925 r600_context_fini(ctx); 926 return r; 927} 928 929void r600_need_cs_space(struct r600_context *ctx, unsigned num_dw, 930 boolean count_draw_in) 931{ 932 /* The number of dwords we already used in the CS so far. */ 933 num_dw += ctx->pm4_cdwords; 934 935 if (count_draw_in) { 936 /* The number of dwords all the dirty states would take. */ 937 num_dw += ctx->pm4_dirty_cdwords; 938 939 /* The upper-bound of how much a draw command would take. */ 940 num_dw += R600_MAX_DRAW_CS_DWORDS; 941 } 942 943 /* Count in queries_suspend. */ 944 num_dw += ctx->num_cs_dw_queries_suspend; 945 946 /* Count in streamout_end at the end of CS. */ 947 num_dw += ctx->num_cs_dw_streamout_end; 948 949 /* Count in render_condition(NULL) at the end of CS. */ 950 if (ctx->predicate_drawing) { 951 num_dw += 3; 952 } 953 954 /* Count in framebuffer cache flushes at the end of CS. */ 955 num_dw += ctx->num_dest_buffers * 7; 956 957 /* Save 16 dwords for the fence mechanism. */ 958 num_dw += 16; 959 960 /* Flush if there's not enough space. */ 961 if (num_dw > RADEON_MAX_CMDBUF_DWORDS) { 962 ctx->flush(ctx->pipe, RADEON_FLUSH_ASYNC); 963 } 964} 965 966/* Flushes all surfaces */ 967void r600_context_flush_all(struct r600_context *ctx, unsigned flush_flags) 968{ 969 r600_need_cs_space(ctx, 5, FALSE); 970 971 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SURFACE_SYNC, 3, 0); 972 ctx->pm4[ctx->pm4_cdwords++] = flush_flags; /* CP_COHER_CNTL */ 973 ctx->pm4[ctx->pm4_cdwords++] = 0xffffffff; /* CP_COHER_SIZE */ 974 ctx->pm4[ctx->pm4_cdwords++] = 0; /* CP_COHER_BASE */ 975 ctx->pm4[ctx->pm4_cdwords++] = 0x0000000A; /* POLL_INTERVAL */ 976} 977 978void r600_context_bo_flush(struct r600_context *ctx, unsigned flush_flags, 979 unsigned flush_mask, struct r600_resource *bo) 980{ 981 /* if bo has already been flushed */ 982 if (!(~bo->cs_buf->last_flush & flush_flags)) { 983 bo->cs_buf->last_flush &= flush_mask; 984 return; 985 } 986 987 if ((ctx->screen->family < CHIP_RV770) && 988 (G_0085F0_CB_ACTION_ENA(flush_flags) || 989 G_0085F0_DB_ACTION_ENA(flush_flags))) { 990 if (ctx->flags & R600_CONTEXT_CHECK_EVENT_FLUSH) { 991 /* the rv670 seems to fail fbo-generatemipmap unless we flush the CB1 dest base ena */ 992 if ((bo->cs_buf->binding & BO_BOUND_TEXTURE) && 993 (flush_flags & S_0085F0_CB_ACTION_ENA(1))) { 994 if ((ctx->screen->family == CHIP_RV670) || 995 (ctx->screen->family == CHIP_RS780) || 996 (ctx->screen->family == CHIP_RS880)) { 997 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SURFACE_SYNC, 3, 0); 998 ctx->pm4[ctx->pm4_cdwords++] = S_0085F0_CB1_DEST_BASE_ENA(1); /* CP_COHER_CNTL */ 999 ctx->pm4[ctx->pm4_cdwords++] = 0xffffffff; /* CP_COHER_SIZE */ 1000 ctx->pm4[ctx->pm4_cdwords++] = 0; /* CP_COHER_BASE */ 1001 ctx->pm4[ctx->pm4_cdwords++] = 0x0000000A; /* POLL_INTERVAL */ 1002 } 1003 } 1004 1005 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, 0); 1006 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_EVENT) | EVENT_INDEX(0); 1007 ctx->flags &= ~R600_CONTEXT_CHECK_EVENT_FLUSH; 1008 } 1009 } else { 1010 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SURFACE_SYNC, 3, 0); 1011 ctx->pm4[ctx->pm4_cdwords++] = flush_flags; 1012 ctx->pm4[ctx->pm4_cdwords++] = (bo->buf->size + 255) >> 8; 1013 ctx->pm4[ctx->pm4_cdwords++] = 0x00000000; 1014 ctx->pm4[ctx->pm4_cdwords++] = 0x0000000A; 1015 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); 1016 ctx->pm4[ctx->pm4_cdwords++] = r600_context_bo_reloc(ctx, bo, RADEON_USAGE_WRITE); 1017 } 1018 bo->cs_buf->last_flush = (bo->cs_buf->last_flush | flush_flags) & flush_mask; 1019} 1020 1021void r600_context_reg(struct r600_context *ctx, 1022 unsigned offset, unsigned value, 1023 unsigned mask) 1024{ 1025 struct r600_range *range; 1026 struct r600_block *block; 1027 unsigned id; 1028 unsigned new_val; 1029 int dirty; 1030 1031 range = &ctx->range[CTX_RANGE_ID(offset)]; 1032 block = range->blocks[CTX_BLOCK_ID(offset)]; 1033 id = (offset - block->start_offset) >> 2; 1034 1035 dirty = block->status & R600_BLOCK_STATUS_DIRTY; 1036 1037 new_val = block->reg[id]; 1038 new_val &= ~mask; 1039 new_val |= value; 1040 if (new_val != block->reg[id]) { 1041 dirty |= R600_BLOCK_STATUS_DIRTY; 1042 block->reg[id] = new_val; 1043 } 1044 if (dirty) 1045 r600_context_dirty_block(ctx, block, dirty, id); 1046} 1047 1048void r600_context_dirty_block(struct r600_context *ctx, 1049 struct r600_block *block, 1050 int dirty, int index) 1051{ 1052 if ((index + 1) > block->nreg_dirty) 1053 block->nreg_dirty = index + 1; 1054 1055 if ((dirty != (block->status & R600_BLOCK_STATUS_DIRTY)) || !(block->status & R600_BLOCK_STATUS_ENABLED)) { 1056 block->status |= R600_BLOCK_STATUS_DIRTY; 1057 ctx->pm4_dirty_cdwords += block->pm4_ndwords + block->pm4_flush_ndwords; 1058 if (!(block->status & R600_BLOCK_STATUS_ENABLED)) { 1059 block->status |= R600_BLOCK_STATUS_ENABLED; 1060 LIST_ADDTAIL(&block->enable_list, &ctx->enable_list); 1061 } 1062 LIST_ADDTAIL(&block->list,&ctx->dirty); 1063 1064 if (block->flags & REG_FLAG_FLUSH_CHANGE) { 1065 r600_context_ps_partial_flush(ctx); 1066 } 1067 } 1068} 1069 1070void r600_context_pipe_state_set(struct r600_context *ctx, struct r600_pipe_state *state) 1071{ 1072 struct r600_block *block; 1073 unsigned new_val; 1074 int dirty; 1075 for (int i = 0; i < state->nregs; i++) { 1076 unsigned id, reloc_id; 1077 struct r600_pipe_reg *reg = &state->regs[i]; 1078 1079 block = reg->block; 1080 id = reg->id; 1081 1082 dirty = block->status & R600_BLOCK_STATUS_DIRTY; 1083 1084 new_val = block->reg[id]; 1085 new_val &= ~reg->mask; 1086 new_val |= reg->value; 1087 if (new_val != block->reg[id]) { 1088 block->reg[id] = new_val; 1089 dirty |= R600_BLOCK_STATUS_DIRTY; 1090 } 1091 if (block->flags & REG_FLAG_DIRTY_ALWAYS) 1092 dirty |= R600_BLOCK_STATUS_DIRTY; 1093 if (block->pm4_bo_index[id]) { 1094 /* find relocation */ 1095 reloc_id = block->pm4_bo_index[id]; 1096 pipe_resource_reference((struct pipe_resource**)&block->reloc[reloc_id].bo, ®->bo->b.b.b); 1097 block->reloc[reloc_id].bo_usage = reg->bo_usage; 1098 /* always force dirty for relocs for now */ 1099 dirty |= R600_BLOCK_STATUS_DIRTY; 1100 } 1101 1102 if (dirty) 1103 r600_context_dirty_block(ctx, block, dirty, id); 1104 } 1105} 1106 1107static void r600_context_dirty_resource_block(struct r600_context *ctx, 1108 struct r600_block *block, 1109 int dirty, int index) 1110{ 1111 block->nreg_dirty = index + 1; 1112 1113 if ((dirty != (block->status & R600_BLOCK_STATUS_RESOURCE_DIRTY)) || !(block->status & R600_BLOCK_STATUS_ENABLED)) { 1114 block->status |= R600_BLOCK_STATUS_RESOURCE_DIRTY; 1115 ctx->pm4_dirty_cdwords += block->pm4_ndwords + block->pm4_flush_ndwords; 1116 if (!(block->status & R600_BLOCK_STATUS_ENABLED)) { 1117 block->status |= R600_BLOCK_STATUS_ENABLED; 1118 LIST_ADDTAIL(&block->enable_list, &ctx->enable_list); 1119 } 1120 LIST_ADDTAIL(&block->list,&ctx->resource_dirty); 1121 } 1122} 1123 1124void r600_context_pipe_state_set_resource(struct r600_context *ctx, struct r600_pipe_resource_state *state, struct r600_block *block) 1125{ 1126 int dirty; 1127 int num_regs = ctx->screen->chip_class >= EVERGREEN ? 8 : 7; 1128 boolean is_vertex; 1129 1130 if (state == NULL) { 1131 block->status &= ~(R600_BLOCK_STATUS_ENABLED | R600_BLOCK_STATUS_RESOURCE_DIRTY); 1132 if (block->reloc[1].bo) 1133 block->reloc[1].bo->cs_buf->binding &= ~BO_BOUND_TEXTURE; 1134 1135 pipe_resource_reference((struct pipe_resource**)&block->reloc[1].bo, NULL); 1136 pipe_resource_reference((struct pipe_resource**)&block->reloc[2].bo, NULL); 1137 LIST_DELINIT(&block->list); 1138 LIST_DELINIT(&block->enable_list); 1139 return; 1140 } 1141 1142 is_vertex = ((state->val[num_regs-1] & 0xc0000000) == 0xc0000000); 1143 dirty = block->status & R600_BLOCK_STATUS_RESOURCE_DIRTY; 1144 1145 if (memcmp(block->reg, state->val, num_regs*4)) { 1146 memcpy(block->reg, state->val, num_regs * 4); 1147 dirty |= R600_BLOCK_STATUS_RESOURCE_DIRTY; 1148 } 1149 1150 /* if no BOs on block, force dirty */ 1151 if (!block->reloc[1].bo || !block->reloc[2].bo) 1152 dirty |= R600_BLOCK_STATUS_RESOURCE_DIRTY; 1153 1154 if (!dirty) { 1155 if (is_vertex) { 1156 if (block->reloc[1].bo->buf != state->bo[0]->buf) 1157 dirty |= R600_BLOCK_STATUS_RESOURCE_DIRTY; 1158 } else { 1159 if ((block->reloc[1].bo->buf != state->bo[0]->buf) || 1160 (block->reloc[2].bo->buf != state->bo[1]->buf)) 1161 dirty |= R600_BLOCK_STATUS_RESOURCE_DIRTY; 1162 } 1163 } 1164 1165 if (dirty) { 1166 if (is_vertex) { 1167 /* VERTEX RESOURCE, we preted there is 2 bo to relocate so 1168 * we have single case btw VERTEX & TEXTURE resource 1169 */ 1170 pipe_resource_reference((struct pipe_resource**)&block->reloc[1].bo, &state->bo[0]->b.b.b); 1171 block->reloc[1].bo_usage = state->bo_usage[0]; 1172 pipe_resource_reference((struct pipe_resource**)&block->reloc[2].bo, NULL); 1173 } else { 1174 /* TEXTURE RESOURCE */ 1175 pipe_resource_reference((struct pipe_resource**)&block->reloc[1].bo, &state->bo[0]->b.b.b); 1176 block->reloc[1].bo_usage = state->bo_usage[0]; 1177 pipe_resource_reference((struct pipe_resource**)&block->reloc[2].bo, &state->bo[1]->b.b.b); 1178 block->reloc[2].bo_usage = state->bo_usage[1]; 1179 state->bo[0]->cs_buf->binding |= BO_BOUND_TEXTURE; 1180 } 1181 1182 if (is_vertex) 1183 block->status |= R600_BLOCK_STATUS_RESOURCE_VERTEX; 1184 else 1185 block->status &= ~R600_BLOCK_STATUS_RESOURCE_VERTEX; 1186 1187 r600_context_dirty_resource_block(ctx, block, dirty, num_regs - 1); 1188 } 1189} 1190 1191void r600_context_pipe_state_set_ps_resource(struct r600_context *ctx, struct r600_pipe_resource_state *state, unsigned rid) 1192{ 1193 struct r600_block *block = ctx->ps_resources.blocks[rid]; 1194 1195 r600_context_pipe_state_set_resource(ctx, state, block); 1196} 1197 1198void r600_context_pipe_state_set_vs_resource(struct r600_context *ctx, struct r600_pipe_resource_state *state, unsigned rid) 1199{ 1200 struct r600_block *block = ctx->vs_resources.blocks[rid]; 1201 1202 r600_context_pipe_state_set_resource(ctx, state, block); 1203} 1204 1205void r600_context_pipe_state_set_fs_resource(struct r600_context *ctx, struct r600_pipe_resource_state *state, unsigned rid) 1206{ 1207 struct r600_block *block = ctx->fs_resources.blocks[rid]; 1208 1209 r600_context_pipe_state_set_resource(ctx, state, block); 1210} 1211 1212static inline void r600_context_pipe_state_set_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned offset) 1213{ 1214 struct r600_range *range; 1215 struct r600_block *block; 1216 int i; 1217 int dirty; 1218 1219 range = &ctx->range[CTX_RANGE_ID(offset)]; 1220 block = range->blocks[CTX_BLOCK_ID(offset)]; 1221 if (state == NULL) { 1222 block->status &= ~(R600_BLOCK_STATUS_ENABLED | R600_BLOCK_STATUS_DIRTY); 1223 LIST_DELINIT(&block->list); 1224 LIST_DELINIT(&block->enable_list); 1225 return; 1226 } 1227 dirty = block->status & R600_BLOCK_STATUS_DIRTY; 1228 for (i = 0; i < 3; i++) { 1229 if (block->reg[i] != state->regs[i].value) { 1230 block->reg[i] = state->regs[i].value; 1231 dirty |= R600_BLOCK_STATUS_DIRTY; 1232 } 1233 } 1234 1235 if (dirty) 1236 r600_context_dirty_block(ctx, block, dirty, 2); 1237} 1238 1239 1240static inline void r600_context_pipe_state_set_sampler_border(struct r600_context *ctx, struct r600_pipe_state *state, unsigned offset) 1241{ 1242 struct r600_range *range; 1243 struct r600_block *block; 1244 int i; 1245 int dirty; 1246 1247 range = &ctx->range[CTX_RANGE_ID(offset)]; 1248 block = range->blocks[CTX_BLOCK_ID(offset)]; 1249 if (state == NULL) { 1250 block->status &= ~(R600_BLOCK_STATUS_ENABLED | R600_BLOCK_STATUS_DIRTY); 1251 LIST_DELINIT(&block->list); 1252 LIST_DELINIT(&block->enable_list); 1253 return; 1254 } 1255 if (state->nregs <= 3) { 1256 return; 1257 } 1258 dirty = block->status & R600_BLOCK_STATUS_DIRTY; 1259 for (i = 0; i < 4; i++) { 1260 if (block->reg[i] != state->regs[i + 3].value) { 1261 block->reg[i] = state->regs[i + 3].value; 1262 dirty |= R600_BLOCK_STATUS_DIRTY; 1263 } 1264 } 1265 1266 /* We have to flush the shaders before we change the border color 1267 * registers, or previous draw commands that haven't completed yet 1268 * will end up using the new border color. */ 1269 if (dirty & R600_BLOCK_STATUS_DIRTY) 1270 r600_context_ps_partial_flush(ctx); 1271 if (dirty) 1272 r600_context_dirty_block(ctx, block, dirty, 3); 1273} 1274 1275void r600_context_pipe_state_set_ps_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id) 1276{ 1277 unsigned offset; 1278 1279 offset = 0x0003C000 + id * 0xc; 1280 r600_context_pipe_state_set_sampler(ctx, state, offset); 1281 offset = 0x0000A400 + id * 0x10; 1282 r600_context_pipe_state_set_sampler_border(ctx, state, offset); 1283} 1284 1285void r600_context_pipe_state_set_vs_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id) 1286{ 1287 unsigned offset; 1288 1289 offset = 0x0003C0D8 + id * 0xc; 1290 r600_context_pipe_state_set_sampler(ctx, state, offset); 1291 offset = 0x0000A600 + id * 0x10; 1292 r600_context_pipe_state_set_sampler_border(ctx, state, offset); 1293} 1294 1295struct r600_resource *r600_context_reg_bo(struct r600_context *ctx, unsigned offset) 1296{ 1297 struct r600_range *range; 1298 struct r600_block *block; 1299 unsigned id; 1300 1301 range = &ctx->range[CTX_RANGE_ID(offset)]; 1302 block = range->blocks[CTX_BLOCK_ID(offset)]; 1303 offset -= block->start_offset; 1304 id = block->pm4_bo_index[offset >> 2]; 1305 if (block->reloc[id].bo) { 1306 return block->reloc[id].bo; 1307 } 1308 return NULL; 1309} 1310 1311void r600_context_block_emit_dirty(struct r600_context *ctx, struct r600_block *block) 1312{ 1313 int optional = block->nbo == 0 && !(block->flags & REG_FLAG_DIRTY_ALWAYS); 1314 int cp_dwords = block->pm4_ndwords, start_dword = 0; 1315 int new_dwords = 0; 1316 int nbo = block->nbo; 1317 1318 if (block->nreg_dirty == 0 && optional) { 1319 goto out; 1320 } 1321 1322 if (nbo) { 1323 ctx->flags |= R600_CONTEXT_CHECK_EVENT_FLUSH; 1324 1325 for (int j = 0; j < block->nreg; j++) { 1326 if (block->pm4_bo_index[j]) { 1327 /* find relocation */ 1328 struct r600_block_reloc *reloc = &block->reloc[block->pm4_bo_index[j]]; 1329 block->pm4[reloc->bo_pm4_index] = 1330 r600_context_bo_reloc(ctx, reloc->bo, reloc->bo_usage); 1331 r600_context_bo_flush(ctx, 1332 reloc->flush_flags, 1333 reloc->flush_mask, 1334 reloc->bo); 1335 nbo--; 1336 if (nbo == 0) 1337 break; 1338 } 1339 } 1340 ctx->flags &= ~R600_CONTEXT_CHECK_EVENT_FLUSH; 1341 } 1342 1343 optional &= (block->nreg_dirty != block->nreg); 1344 if (optional) { 1345 new_dwords = block->nreg_dirty; 1346 start_dword = ctx->pm4_cdwords; 1347 cp_dwords = new_dwords + 2; 1348 } 1349 memcpy(&ctx->pm4[ctx->pm4_cdwords], block->pm4, cp_dwords * 4); 1350 ctx->pm4_cdwords += cp_dwords; 1351 1352 if (optional) { 1353 uint32_t newword; 1354 1355 newword = ctx->pm4[start_dword]; 1356 newword &= PKT_COUNT_C; 1357 newword |= PKT_COUNT_S(new_dwords); 1358 ctx->pm4[start_dword] = newword; 1359 } 1360out: 1361 block->status ^= R600_BLOCK_STATUS_DIRTY; 1362 block->nreg_dirty = 0; 1363 LIST_DELINIT(&block->list); 1364} 1365 1366void r600_context_block_resource_emit_dirty(struct r600_context *ctx, struct r600_block *block) 1367{ 1368 int cp_dwords = block->pm4_ndwords; 1369 int nbo = block->nbo; 1370 1371 ctx->flags |= R600_CONTEXT_CHECK_EVENT_FLUSH; 1372 1373 if (block->status & R600_BLOCK_STATUS_RESOURCE_VERTEX) { 1374 nbo = 1; 1375 cp_dwords -= 2; /* don't copy the second NOP */ 1376 } 1377 1378 for (int j = 0; j < nbo; j++) { 1379 if (block->pm4_bo_index[j]) { 1380 /* find relocation */ 1381 struct r600_block_reloc *reloc = &block->reloc[block->pm4_bo_index[j]]; 1382 block->pm4[reloc->bo_pm4_index] = 1383 r600_context_bo_reloc(ctx, reloc->bo, reloc->bo_usage); 1384 r600_context_bo_flush(ctx, 1385 reloc->flush_flags, 1386 reloc->flush_mask, 1387 reloc->bo); 1388 } 1389 } 1390 ctx->flags &= ~R600_CONTEXT_CHECK_EVENT_FLUSH; 1391 1392 memcpy(&ctx->pm4[ctx->pm4_cdwords], block->pm4, cp_dwords * 4); 1393 ctx->pm4_cdwords += cp_dwords; 1394 1395 block->status ^= R600_BLOCK_STATUS_RESOURCE_DIRTY; 1396 block->nreg_dirty = 0; 1397 LIST_DELINIT(&block->list); 1398} 1399 1400void r600_context_flush_dest_caches(struct r600_context *ctx) 1401{ 1402 struct r600_resource *cb[8]; 1403 struct r600_resource *db; 1404 int i; 1405 1406 if (!(ctx->flags & R600_CONTEXT_DST_CACHES_DIRTY)) 1407 return; 1408 1409 db = r600_context_reg_bo(ctx, R_02800C_DB_DEPTH_BASE); 1410 cb[0] = r600_context_reg_bo(ctx, R_028040_CB_COLOR0_BASE); 1411 cb[1] = r600_context_reg_bo(ctx, R_028044_CB_COLOR1_BASE); 1412 cb[2] = r600_context_reg_bo(ctx, R_028048_CB_COLOR2_BASE); 1413 cb[3] = r600_context_reg_bo(ctx, R_02804C_CB_COLOR3_BASE); 1414 cb[4] = r600_context_reg_bo(ctx, R_028050_CB_COLOR4_BASE); 1415 cb[5] = r600_context_reg_bo(ctx, R_028054_CB_COLOR5_BASE); 1416 cb[6] = r600_context_reg_bo(ctx, R_028058_CB_COLOR6_BASE); 1417 cb[7] = r600_context_reg_bo(ctx, R_02805C_CB_COLOR7_BASE); 1418 1419 ctx->flags |= R600_CONTEXT_CHECK_EVENT_FLUSH; 1420 /* flush the color buffers */ 1421 for (i = 0; i < 8; i++) { 1422 if (!cb[i]) 1423 continue; 1424 1425 r600_context_bo_flush(ctx, 1426 (S_0085F0_CB0_DEST_BASE_ENA(1) << i) | 1427 S_0085F0_CB_ACTION_ENA(1), 1428 0, cb[i]); 1429 } 1430 if (db) { 1431 r600_context_bo_flush(ctx, S_0085F0_DB_ACTION_ENA(1) | S_0085F0_DB_DEST_BASE_ENA(1), 0, db); 1432 } 1433 ctx->flags &= ~R600_CONTEXT_CHECK_EVENT_FLUSH; 1434 ctx->flags &= ~R600_CONTEXT_DST_CACHES_DIRTY; 1435} 1436 1437void r600_context_draw(struct r600_context *ctx, const struct r600_draw *draw) 1438{ 1439 unsigned ndwords = 7; 1440 struct r600_block *dirty_block = NULL; 1441 struct r600_block *next_block; 1442 uint32_t *pm4; 1443 1444 if (draw->indices) { 1445 ndwords = 11; 1446 } 1447 /* when increasing ndwords, bump the max limit too */ 1448 assert(ndwords <= R600_MAX_DRAW_CS_DWORDS); 1449 1450 /* queries need some special values 1451 * (this is non-zero if any query is active) */ 1452 if (ctx->num_cs_dw_queries_suspend) { 1453 if (ctx->screen->family >= CHIP_RV770) { 1454 r600_context_reg(ctx, 1455 R_028D0C_DB_RENDER_CONTROL, 1456 S_028D0C_R700_PERFECT_ZPASS_COUNTS(1), 1457 S_028D0C_R700_PERFECT_ZPASS_COUNTS(1)); 1458 } 1459 r600_context_reg(ctx, 1460 R_028D10_DB_RENDER_OVERRIDE, 1461 S_028D10_NOOP_CULL_DISABLE(1), 1462 S_028D10_NOOP_CULL_DISABLE(1)); 1463 } 1464 1465 r600_need_cs_space(ctx, 0, TRUE); 1466 assert(ctx->pm4_cdwords + ctx->pm4_dirty_cdwords + ndwords < RADEON_MAX_CMDBUF_DWORDS); 1467 1468 /* enough room to copy packet */ 1469 LIST_FOR_EACH_ENTRY_SAFE(dirty_block, next_block, &ctx->dirty, list) { 1470 r600_context_block_emit_dirty(ctx, dirty_block); 1471 } 1472 1473 LIST_FOR_EACH_ENTRY_SAFE(dirty_block, next_block, &ctx->resource_dirty, list) { 1474 r600_context_block_resource_emit_dirty(ctx, dirty_block); 1475 } 1476 1477 /* Enable stream out if needed. */ 1478 if (ctx->streamout_start) { 1479 r600_context_streamout_begin(ctx); 1480 ctx->streamout_start = FALSE; 1481 } 1482 1483 /* draw packet */ 1484 pm4 = &ctx->pm4[ctx->pm4_cdwords]; 1485 1486 pm4[0] = PKT3(PKT3_INDEX_TYPE, 0, ctx->predicate_drawing); 1487 pm4[1] = draw->vgt_index_type; 1488 pm4[2] = PKT3(PKT3_NUM_INSTANCES, 0, ctx->predicate_drawing); 1489 pm4[3] = draw->vgt_num_instances; 1490 if (draw->indices) { 1491 pm4[4] = PKT3(PKT3_DRAW_INDEX, 3, ctx->predicate_drawing); 1492 pm4[5] = draw->indices_bo_offset; 1493 pm4[6] = 0; 1494 pm4[7] = draw->vgt_num_indices; 1495 pm4[8] = draw->vgt_draw_initiator; 1496 pm4[9] = PKT3(PKT3_NOP, 0, ctx->predicate_drawing); 1497 pm4[10] = r600_context_bo_reloc(ctx, draw->indices, RADEON_USAGE_READ); 1498 } else { 1499 pm4[4] = PKT3(PKT3_DRAW_INDEX_AUTO, 1, ctx->predicate_drawing); 1500 pm4[5] = draw->vgt_num_indices; 1501 pm4[6] = draw->vgt_draw_initiator; 1502 } 1503 ctx->pm4_cdwords += ndwords; 1504 1505 ctx->flags |= (R600_CONTEXT_DST_CACHES_DIRTY | R600_CONTEXT_DRAW_PENDING); 1506 1507 /* all dirty state have been scheduled in current cs */ 1508 ctx->pm4_dirty_cdwords = 0; 1509} 1510 1511void r600_context_flush(struct r600_context *ctx, unsigned flags) 1512{ 1513 struct r600_block *enable_block = NULL; 1514 bool queries_suspended = false; 1515 bool streamout_suspended = false; 1516 1517 if (ctx->pm4_cdwords == ctx->init_dwords) 1518 return; 1519 1520 /* suspend queries */ 1521 if (ctx->num_cs_dw_queries_suspend) { 1522 r600_context_queries_suspend(ctx); 1523 queries_suspended = true; 1524 } 1525 1526 if (ctx->num_cs_dw_streamout_end) { 1527 r600_context_streamout_end(ctx); 1528 streamout_suspended = true; 1529 } 1530 1531 if (ctx->screen->chip_class >= EVERGREEN) 1532 evergreen_context_flush_dest_caches(ctx); 1533 else 1534 r600_context_flush_dest_caches(ctx); 1535 1536 /* partial flush is needed to avoid lockups on some chips with user fences */ 1537 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, 0); 1538 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4); 1539 1540 /* Flush the CS. */ 1541 ctx->cs->cdw = ctx->pm4_cdwords; 1542 ctx->ws->cs_flush(ctx->cs, flags); 1543 1544 /* We need to get the pointer to the other CS, 1545 * the command streams are double-buffered. */ 1546 ctx->pm4 = ctx->cs->buf; 1547 1548 /* restart */ 1549 for (int i = 0; i < ctx->creloc; i++) { 1550 ctx->bo[i]->cs_buf->last_flush = 0; 1551 pipe_resource_reference((struct pipe_resource**)&ctx->bo[i], NULL); 1552 } 1553 ctx->creloc = 0; 1554 ctx->pm4_dirty_cdwords = 0; 1555 ctx->pm4_cdwords = 0; 1556 ctx->flags = 0; 1557 1558 r600_init_cs(ctx); 1559 1560 if (streamout_suspended) { 1561 ctx->streamout_start = TRUE; 1562 ctx->streamout_append_bitmask = ~0; 1563 } 1564 1565 /* resume queries */ 1566 if (queries_suspended) { 1567 r600_context_queries_resume(ctx); 1568 } 1569 1570 /* set all valid group as dirty so they get reemited on 1571 * next draw command 1572 */ 1573 LIST_FOR_EACH_ENTRY(enable_block, &ctx->enable_list, enable_list) { 1574 if (!(enable_block->flags & BLOCK_FLAG_RESOURCE)) { 1575 if(!(enable_block->status & R600_BLOCK_STATUS_DIRTY)) { 1576 LIST_ADDTAIL(&enable_block->list,&ctx->dirty); 1577 enable_block->status |= R600_BLOCK_STATUS_DIRTY; 1578 } 1579 } else { 1580 if(!(enable_block->status & R600_BLOCK_STATUS_RESOURCE_DIRTY)) { 1581 LIST_ADDTAIL(&enable_block->list,&ctx->resource_dirty); 1582 enable_block->status |= R600_BLOCK_STATUS_RESOURCE_DIRTY; 1583 } 1584 } 1585 ctx->pm4_dirty_cdwords += enable_block->pm4_ndwords + 1586 enable_block->pm4_flush_ndwords; 1587 enable_block->nreg_dirty = enable_block->nreg; 1588 } 1589} 1590 1591void r600_context_emit_fence(struct r600_context *ctx, struct r600_resource *fence_bo, unsigned offset, unsigned value) 1592{ 1593 r600_need_cs_space(ctx, 10, FALSE); 1594 1595 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, 0); 1596 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4); 1597 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0); 1598 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5); 1599 ctx->pm4[ctx->pm4_cdwords++] = offset << 2; /* ADDRESS_LO */ 1600 ctx->pm4[ctx->pm4_cdwords++] = (1 << 29) | (0 << 24); /* DATA_SEL | INT_EN | ADDRESS_HI */ 1601 ctx->pm4[ctx->pm4_cdwords++] = value; /* DATA_LO */ 1602 ctx->pm4[ctx->pm4_cdwords++] = 0; /* DATA_HI */ 1603 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); 1604 ctx->pm4[ctx->pm4_cdwords++] = r600_context_bo_reloc(ctx, fence_bo, RADEON_USAGE_WRITE); 1605} 1606 1607static unsigned r600_query_read_result(char *map, unsigned start_index, unsigned end_index, 1608 bool test_status_bit) 1609{ 1610 uint32_t *current_result = (uint32_t*)map; 1611 uint64_t start, end; 1612 1613 start = (uint64_t)current_result[start_index] | 1614 (uint64_t)current_result[start_index+1] << 32; 1615 end = (uint64_t)current_result[end_index] | 1616 (uint64_t)current_result[end_index+1] << 32; 1617 1618 if (!test_status_bit || 1619 ((start & 0x8000000000000000UL) && (end & 0x8000000000000000UL))) { 1620 return end - start; 1621 } 1622 return 0; 1623} 1624 1625static boolean r600_query_result(struct r600_context *ctx, struct r600_query *query, boolean wait) 1626{ 1627 unsigned results_base = query->results_start; 1628 char *map; 1629 1630 map = ctx->ws->buffer_map(query->buffer->buf, ctx->cs, 1631 PIPE_TRANSFER_READ | 1632 (wait ? 0 : PIPE_TRANSFER_DONTBLOCK)); 1633 if (!map) 1634 return FALSE; 1635 1636 /* count all results across all data blocks */ 1637 switch (query->type) { 1638 case PIPE_QUERY_OCCLUSION_COUNTER: 1639 while (results_base != query->results_end) { 1640 query->result.u64 += 1641 r600_query_read_result(map + results_base, 0, 2, true); 1642 results_base = (results_base + 16) % query->buffer->b.b.b.width0; 1643 } 1644 break; 1645 case PIPE_QUERY_OCCLUSION_PREDICATE: 1646 while (results_base != query->results_end) { 1647 query->result.b = query->result.b || 1648 r600_query_read_result(map + results_base, 0, 2, true) != 0; 1649 results_base = (results_base + 16) % query->buffer->b.b.b.width0; 1650 } 1651 break; 1652 case PIPE_QUERY_TIME_ELAPSED: 1653 while (results_base != query->results_end) { 1654 query->result.u64 += 1655 r600_query_read_result(map + results_base, 0, 2, false); 1656 results_base = (results_base + query->result_size) % query->buffer->b.b.b.width0; 1657 } 1658 break; 1659 case PIPE_QUERY_PRIMITIVES_EMITTED: 1660 /* SAMPLE_STREAMOUTSTATS stores this structure: 1661 * { 1662 * u64 NumPrimitivesWritten; 1663 * u64 PrimitiveStorageNeeded; 1664 * } 1665 * We only need NumPrimitivesWritten here. */ 1666 while (results_base != query->results_end) { 1667 query->result.u64 += 1668 r600_query_read_result(map + results_base, 2, 6, true); 1669 results_base = (results_base + query->result_size) % query->buffer->b.b.b.width0; 1670 } 1671 break; 1672 case PIPE_QUERY_PRIMITIVES_GENERATED: 1673 /* Here we read PrimitiveStorageNeeded. */ 1674 while (results_base != query->results_end) { 1675 query->result.u64 += 1676 r600_query_read_result(map + results_base, 0, 4, true); 1677 results_base = (results_base + query->result_size) % query->buffer->b.b.b.width0; 1678 } 1679 break; 1680 case PIPE_QUERY_SO_STATISTICS: 1681 while (results_base != query->results_end) { 1682 query->result.so.num_primitives_written += 1683 r600_query_read_result(map + results_base, 2, 6, true); 1684 query->result.so.primitives_storage_needed += 1685 r600_query_read_result(map + results_base, 0, 4, true); 1686 results_base = (results_base + query->result_size) % query->buffer->b.b.b.width0; 1687 } 1688 break; 1689 case PIPE_QUERY_SO_OVERFLOW_PREDICATE: 1690 while (results_base != query->results_end) { 1691 query->result.b = query->result.b || 1692 r600_query_read_result(map + results_base, 2, 6, true) != 1693 r600_query_read_result(map + results_base, 0, 4, true); 1694 results_base = (results_base + query->result_size) % query->buffer->b.b.b.width0; 1695 } 1696 break; 1697 default: 1698 assert(0); 1699 } 1700 1701 query->results_start = query->results_end; 1702 ctx->ws->buffer_unmap(query->buffer->buf); 1703 return TRUE; 1704} 1705 1706void r600_query_begin(struct r600_context *ctx, struct r600_query *query) 1707{ 1708 unsigned new_results_end, i; 1709 u32 *results; 1710 1711 r600_need_cs_space(ctx, query->num_cs_dw * 2, TRUE); 1712 1713 new_results_end = (query->results_end + query->result_size) % query->buffer->b.b.b.width0; 1714 1715 /* collect current results if query buffer is full */ 1716 if (new_results_end == query->results_start) { 1717 r600_query_result(ctx, query, TRUE); 1718 } 1719 1720 switch (query->type) { 1721 case PIPE_QUERY_OCCLUSION_COUNTER: 1722 case PIPE_QUERY_OCCLUSION_PREDICATE: 1723 results = ctx->ws->buffer_map(query->buffer->buf, ctx->cs, PIPE_TRANSFER_WRITE); 1724 if (results) { 1725 results = (u32*)((char*)results + query->results_end); 1726 memset(results, 0, query->result_size); 1727 1728 /* Set top bits for unused backends */ 1729 for (i = 0; i < ctx->max_db; i++) { 1730 if (!(ctx->backend_mask & (1<<i))) { 1731 results[(i * 4)+1] = 0x80000000; 1732 results[(i * 4)+3] = 0x80000000; 1733 } 1734 } 1735 ctx->ws->buffer_unmap(query->buffer->buf); 1736 } 1737 break; 1738 case PIPE_QUERY_TIME_ELAPSED: 1739 break; 1740 case PIPE_QUERY_PRIMITIVES_EMITTED: 1741 case PIPE_QUERY_PRIMITIVES_GENERATED: 1742 case PIPE_QUERY_SO_STATISTICS: 1743 case PIPE_QUERY_SO_OVERFLOW_PREDICATE: 1744 results = ctx->ws->buffer_map(query->buffer->buf, ctx->cs, PIPE_TRANSFER_WRITE); 1745 results = (u32*)((char*)results + query->results_end); 1746 memset(results, 0, query->result_size); 1747 ctx->ws->buffer_unmap(query->buffer->buf); 1748 break; 1749 default: 1750 assert(0); 1751 } 1752 1753 /* emit begin query */ 1754 switch (query->type) { 1755 case PIPE_QUERY_OCCLUSION_COUNTER: 1756 case PIPE_QUERY_OCCLUSION_PREDICATE: 1757 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 2, 0); 1758 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_ZPASS_DONE) | EVENT_INDEX(1); 1759 ctx->pm4[ctx->pm4_cdwords++] = query->results_end; 1760 ctx->pm4[ctx->pm4_cdwords++] = 0; 1761 break; 1762 case PIPE_QUERY_PRIMITIVES_EMITTED: 1763 case PIPE_QUERY_PRIMITIVES_GENERATED: 1764 case PIPE_QUERY_SO_STATISTICS: 1765 case PIPE_QUERY_SO_OVERFLOW_PREDICATE: 1766 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 2, 0); 1767 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_SAMPLE_STREAMOUTSTATS) | EVENT_INDEX(3); 1768 ctx->pm4[ctx->pm4_cdwords++] = query->results_end; 1769 ctx->pm4[ctx->pm4_cdwords++] = 0; 1770 break; 1771 case PIPE_QUERY_TIME_ELAPSED: 1772 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0); 1773 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5); 1774 ctx->pm4[ctx->pm4_cdwords++] = query->results_end; 1775 ctx->pm4[ctx->pm4_cdwords++] = (3 << 29); 1776 ctx->pm4[ctx->pm4_cdwords++] = 0; 1777 ctx->pm4[ctx->pm4_cdwords++] = 0; 1778 break; 1779 default: 1780 assert(0); 1781 } 1782 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); 1783 ctx->pm4[ctx->pm4_cdwords++] = r600_context_bo_reloc(ctx, query->buffer, RADEON_USAGE_WRITE); 1784 1785 ctx->num_cs_dw_queries_suspend += query->num_cs_dw; 1786} 1787 1788void r600_query_end(struct r600_context *ctx, struct r600_query *query) 1789{ 1790 /* emit end query */ 1791 switch (query->type) { 1792 case PIPE_QUERY_OCCLUSION_COUNTER: 1793 case PIPE_QUERY_OCCLUSION_PREDICATE: 1794 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 2, 0); 1795 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_ZPASS_DONE) | EVENT_INDEX(1); 1796 ctx->pm4[ctx->pm4_cdwords++] = query->results_end + 8; 1797 ctx->pm4[ctx->pm4_cdwords++] = 0; 1798 break; 1799 case PIPE_QUERY_PRIMITIVES_EMITTED: 1800 case PIPE_QUERY_PRIMITIVES_GENERATED: 1801 case PIPE_QUERY_SO_STATISTICS: 1802 case PIPE_QUERY_SO_OVERFLOW_PREDICATE: 1803 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 2, 0); 1804 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_SAMPLE_STREAMOUTSTATS) | EVENT_INDEX(3); 1805 ctx->pm4[ctx->pm4_cdwords++] = query->results_end + query->result_size/2; 1806 ctx->pm4[ctx->pm4_cdwords++] = 0; 1807 break; 1808 case PIPE_QUERY_TIME_ELAPSED: 1809 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0); 1810 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5); 1811 ctx->pm4[ctx->pm4_cdwords++] = query->results_end + query->result_size/2; 1812 ctx->pm4[ctx->pm4_cdwords++] = (3 << 29); 1813 ctx->pm4[ctx->pm4_cdwords++] = 0; 1814 ctx->pm4[ctx->pm4_cdwords++] = 0; 1815 break; 1816 default: 1817 assert(0); 1818 } 1819 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); 1820 ctx->pm4[ctx->pm4_cdwords++] = r600_context_bo_reloc(ctx, query->buffer, RADEON_USAGE_WRITE); 1821 1822 query->results_end = (query->results_end + query->result_size) % query->buffer->b.b.b.width0; 1823 ctx->num_cs_dw_queries_suspend -= query->num_cs_dw; 1824} 1825 1826void r600_query_predication(struct r600_context *ctx, struct r600_query *query, int operation, 1827 int flag_wait) 1828{ 1829 if (operation == PREDICATION_OP_CLEAR) { 1830 r600_need_cs_space(ctx, 3, FALSE); 1831 1832 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_PREDICATION, 1, 0); 1833 ctx->pm4[ctx->pm4_cdwords++] = 0; 1834 ctx->pm4[ctx->pm4_cdwords++] = PRED_OP(PREDICATION_OP_CLEAR); 1835 } else { 1836 unsigned results_base = query->results_start; 1837 unsigned count; 1838 u32 op; 1839 1840 /* find count of the query data blocks */ 1841 count = (query->buffer->b.b.b.width0 + query->results_end - query->results_start) % query->buffer->b.b.b.width0; 1842 count /= query->result_size; 1843 1844 r600_need_cs_space(ctx, 5 * count, TRUE); 1845 1846 op = PRED_OP(operation) | PREDICATION_DRAW_VISIBLE | 1847 (flag_wait ? PREDICATION_HINT_WAIT : PREDICATION_HINT_NOWAIT_DRAW); 1848 1849 /* emit predicate packets for all data blocks */ 1850 while (results_base != query->results_end) { 1851 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_PREDICATION, 1, 0); 1852 ctx->pm4[ctx->pm4_cdwords++] = results_base; 1853 ctx->pm4[ctx->pm4_cdwords++] = op; 1854 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); 1855 ctx->pm4[ctx->pm4_cdwords++] = r600_context_bo_reloc(ctx, query->buffer, 1856 RADEON_USAGE_READ); 1857 results_base = (results_base + query->result_size) % query->buffer->b.b.b.width0; 1858 1859 /* set CONTINUE bit for all packets except the first */ 1860 op |= PREDICATION_CONTINUE; 1861 } 1862 } 1863} 1864 1865struct r600_query *r600_context_query_create(struct r600_context *ctx, unsigned query_type) 1866{ 1867 struct r600_query *query; 1868 unsigned buffer_size = 4096; 1869 1870 query = CALLOC_STRUCT(r600_query); 1871 if (query == NULL) 1872 return NULL; 1873 1874 query->type = query_type; 1875 1876 switch (query_type) { 1877 case PIPE_QUERY_OCCLUSION_COUNTER: 1878 case PIPE_QUERY_OCCLUSION_PREDICATE: 1879 query->result_size = 16 * ctx->max_db; 1880 query->num_cs_dw = 6; 1881 break; 1882 case PIPE_QUERY_TIME_ELAPSED: 1883 query->result_size = 16; 1884 query->num_cs_dw = 8; 1885 break; 1886 case PIPE_QUERY_PRIMITIVES_EMITTED: 1887 case PIPE_QUERY_PRIMITIVES_GENERATED: 1888 case PIPE_QUERY_SO_STATISTICS: 1889 case PIPE_QUERY_SO_OVERFLOW_PREDICATE: 1890 /* NumPrimitivesWritten, PrimitiveStorageNeeded. */ 1891 query->result_size = 32; 1892 query->num_cs_dw = 6; 1893 break; 1894 default: 1895 assert(0); 1896 FREE(query); 1897 return NULL; 1898 } 1899 1900 /* adjust buffer size to simplify offsets wrapping math */ 1901 buffer_size -= buffer_size % query->result_size; 1902 1903 /* Queries are normally read by the CPU after 1904 * being written by the gpu, hence staging is probably a good 1905 * usage pattern. 1906 */ 1907 query->buffer = (struct r600_resource*) 1908 pipe_buffer_create(&ctx->screen->screen, PIPE_BIND_CUSTOM, PIPE_USAGE_STAGING, buffer_size); 1909 if (!query->buffer) { 1910 FREE(query); 1911 return NULL; 1912 } 1913 return query; 1914} 1915 1916void r600_context_query_destroy(struct r600_context *ctx, struct r600_query *query) 1917{ 1918 pipe_resource_reference((struct pipe_resource**)&query->buffer, NULL); 1919 free(query); 1920} 1921 1922boolean r600_context_query_result(struct r600_context *ctx, 1923 struct r600_query *query, 1924 boolean wait, void *vresult) 1925{ 1926 boolean *result_b = (boolean*)vresult; 1927 uint64_t *result_u64 = (uint64_t*)vresult; 1928 struct pipe_query_data_so_statistics *result_so = 1929 (struct pipe_query_data_so_statistics*)vresult; 1930 1931 if (!r600_query_result(ctx, query, wait)) 1932 return FALSE; 1933 1934 switch (query->type) { 1935 case PIPE_QUERY_OCCLUSION_COUNTER: 1936 case PIPE_QUERY_PRIMITIVES_EMITTED: 1937 case PIPE_QUERY_PRIMITIVES_GENERATED: 1938 *result_u64 = query->result.u64; 1939 break; 1940 case PIPE_QUERY_OCCLUSION_PREDICATE: 1941 case PIPE_QUERY_SO_OVERFLOW_PREDICATE: 1942 *result_b = query->result.b; 1943 break; 1944 case PIPE_QUERY_TIME_ELAPSED: 1945 *result_u64 = (1000000 * query->result.u64) / ctx->screen->info.r600_clock_crystal_freq; 1946 break; 1947 case PIPE_QUERY_SO_STATISTICS: 1948 *result_so = query->result.so; 1949 break; 1950 default: 1951 assert(0); 1952 } 1953 return TRUE; 1954} 1955 1956void r600_context_queries_suspend(struct r600_context *ctx) 1957{ 1958 struct r600_query *query; 1959 1960 LIST_FOR_EACH_ENTRY(query, &ctx->active_query_list, list) { 1961 r600_query_end(ctx, query); 1962 } 1963 assert(ctx->num_cs_dw_queries_suspend == 0); 1964} 1965 1966void r600_context_queries_resume(struct r600_context *ctx) 1967{ 1968 struct r600_query *query; 1969 1970 assert(ctx->num_cs_dw_queries_suspend == 0); 1971 1972 LIST_FOR_EACH_ENTRY(query, &ctx->active_query_list, list) { 1973 r600_query_begin(ctx, query); 1974 } 1975} 1976 1977static void r600_flush_vgt_streamout(struct r600_context *ctx) 1978{ 1979 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_CONFIG_REG, 1, 0); 1980 ctx->pm4[ctx->pm4_cdwords++] = (R_008490_CP_STRMOUT_CNTL - R600_CONFIG_REG_OFFSET) >> 2; 1981 ctx->pm4[ctx->pm4_cdwords++] = 0; 1982 1983 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, 0); 1984 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_SO_VGTSTREAMOUT_FLUSH) | EVENT_INDEX(0); 1985 1986 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_WAIT_REG_MEM, 5, 0); 1987 ctx->pm4[ctx->pm4_cdwords++] = WAIT_REG_MEM_EQUAL; /* wait until the register is equal to the reference value */ 1988 ctx->pm4[ctx->pm4_cdwords++] = R_008490_CP_STRMOUT_CNTL >> 2; /* register */ 1989 ctx->pm4[ctx->pm4_cdwords++] = 0; 1990 ctx->pm4[ctx->pm4_cdwords++] = S_008490_OFFSET_UPDATE_DONE(1); /* reference value */ 1991 ctx->pm4[ctx->pm4_cdwords++] = S_008490_OFFSET_UPDATE_DONE(1); /* mask */ 1992 ctx->pm4[ctx->pm4_cdwords++] = 4; /* poll interval */ 1993} 1994 1995static void r600_set_streamout_enable(struct r600_context *ctx, unsigned buffer_enable_bit) 1996{ 1997 if (buffer_enable_bit) { 1998 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0); 1999 ctx->pm4[ctx->pm4_cdwords++] = (R_028AB0_VGT_STRMOUT_EN - R600_CONTEXT_REG_OFFSET) >> 2; 2000 ctx->pm4[ctx->pm4_cdwords++] = S_028AB0_STREAMOUT(1); 2001 2002 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0); 2003 ctx->pm4[ctx->pm4_cdwords++] = (R_028B20_VGT_STRMOUT_BUFFER_EN - R600_CONTEXT_REG_OFFSET) >> 2; 2004 ctx->pm4[ctx->pm4_cdwords++] = buffer_enable_bit; 2005 } else { 2006 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0); 2007 ctx->pm4[ctx->pm4_cdwords++] = (R_028AB0_VGT_STRMOUT_EN - R600_CONTEXT_REG_OFFSET) >> 2; 2008 ctx->pm4[ctx->pm4_cdwords++] = S_028AB0_STREAMOUT(0); 2009 } 2010} 2011 2012void r600_context_streamout_begin(struct r600_context *ctx) 2013{ 2014 struct r600_so_target **t = ctx->so_targets; 2015 unsigned *strides = ctx->vs_shader_so_strides; 2016 unsigned buffer_en, i, update_flags = 0; 2017 2018 buffer_en = (ctx->num_so_targets >= 1 && t[0] ? 1 : 0) | 2019 (ctx->num_so_targets >= 2 && t[1] ? 2 : 0) | 2020 (ctx->num_so_targets >= 3 && t[2] ? 4 : 0) | 2021 (ctx->num_so_targets >= 4 && t[3] ? 8 : 0); 2022 2023 ctx->num_cs_dw_streamout_end = 2024 12 + /* flush_vgt_streamout */ 2025 util_bitcount(buffer_en) * 8 + 2026 8; 2027 2028 r600_need_cs_space(ctx, 2029 12 + /* flush_vgt_streamout */ 2030 6 + /* enables */ 2031 util_bitcount(buffer_en & ctx->streamout_append_bitmask) * 8 + 2032 util_bitcount(buffer_en & ~ctx->streamout_append_bitmask) * 6 + 2033 (ctx->screen->family > CHIP_R600 && ctx->screen->family < CHIP_RV770 ? 2 : 0) + 2034 ctx->num_cs_dw_streamout_end, TRUE); 2035 2036 if (ctx->screen->chip_class >= EVERGREEN) { 2037 evergreen_flush_vgt_streamout(ctx); 2038 evergreen_set_streamout_enable(ctx, buffer_en); 2039 } else { 2040 r600_flush_vgt_streamout(ctx); 2041 r600_set_streamout_enable(ctx, buffer_en); 2042 } 2043 2044 for (i = 0; i < ctx->num_so_targets; i++) { 2045 if (t[i]) { 2046 t[i]->stride = strides[i]; 2047 t[i]->so_index = i; 2048 2049 update_flags |= SURFACE_BASE_UPDATE_STRMOUT(i); 2050 2051 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_CONTEXT_REG, 3, 0); 2052 ctx->pm4[ctx->pm4_cdwords++] = (R_028AD0_VGT_STRMOUT_BUFFER_SIZE_0 + 2053 16*i - R600_CONTEXT_REG_OFFSET) >> 2; 2054 ctx->pm4[ctx->pm4_cdwords++] = (t[i]->b.buffer_offset + 2055 t[i]->b.buffer_size) >> 2; /* BUFFER_SIZE (in DW) */ 2056 ctx->pm4[ctx->pm4_cdwords++] = strides[i] >> 2; /* VTX_STRIDE (in DW) */ 2057 ctx->pm4[ctx->pm4_cdwords++] = 0; /* BUFFER_BASE */ 2058 2059 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); 2060 ctx->pm4[ctx->pm4_cdwords++] = 2061 r600_context_bo_reloc(ctx, r600_resource(t[i]->b.buffer), 2062 RADEON_USAGE_WRITE); 2063 2064 if (ctx->streamout_append_bitmask & (1 << i)) { 2065 /* Append. */ 2066 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0); 2067 ctx->pm4[ctx->pm4_cdwords++] = STRMOUT_SELECT_BUFFER(i) | 2068 STRMOUT_OFFSET_SOURCE(STRMOUT_OFFSET_FROM_MEM); /* control */ 2069 ctx->pm4[ctx->pm4_cdwords++] = 0; /* unused */ 2070 ctx->pm4[ctx->pm4_cdwords++] = 0; /* unused */ 2071 ctx->pm4[ctx->pm4_cdwords++] = 0; /* src address lo */ 2072 ctx->pm4[ctx->pm4_cdwords++] = 0; /* src address hi */ 2073 2074 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); 2075 ctx->pm4[ctx->pm4_cdwords++] = 2076 r600_context_bo_reloc(ctx, t[i]->filled_size, 2077 RADEON_USAGE_READ); 2078 } else { 2079 /* Start from the beginning. */ 2080 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0); 2081 ctx->pm4[ctx->pm4_cdwords++] = STRMOUT_SELECT_BUFFER(i) | 2082 STRMOUT_OFFSET_SOURCE(STRMOUT_OFFSET_FROM_PACKET); /* control */ 2083 ctx->pm4[ctx->pm4_cdwords++] = 0; /* unused */ 2084 ctx->pm4[ctx->pm4_cdwords++] = 0; /* unused */ 2085 ctx->pm4[ctx->pm4_cdwords++] = t[i]->b.buffer_offset >> 2; /* buffer offset in DW */ 2086 ctx->pm4[ctx->pm4_cdwords++] = 0; /* unused */ 2087 } 2088 } 2089 } 2090 2091 if (ctx->screen->family > CHIP_R600 && ctx->screen->family < CHIP_RV770) { 2092 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SURFACE_BASE_UPDATE, 0, 0); 2093 ctx->pm4[ctx->pm4_cdwords++] = update_flags; 2094 } 2095} 2096 2097void r600_context_streamout_end(struct r600_context *ctx) 2098{ 2099 struct r600_so_target **t = ctx->so_targets; 2100 unsigned i, flush_flags = 0; 2101 2102 if (ctx->screen->chip_class >= EVERGREEN) { 2103 evergreen_flush_vgt_streamout(ctx); 2104 } else { 2105 r600_flush_vgt_streamout(ctx); 2106 } 2107 2108 for (i = 0; i < ctx->num_so_targets; i++) { 2109 if (t[i]) { 2110 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0); 2111 ctx->pm4[ctx->pm4_cdwords++] = STRMOUT_SELECT_BUFFER(i) | 2112 STRMOUT_OFFSET_SOURCE(STRMOUT_OFFSET_NONE) | 2113 STRMOUT_STORE_BUFFER_FILLED_SIZE; /* control */ 2114 ctx->pm4[ctx->pm4_cdwords++] = 0; /* dst address lo */ 2115 ctx->pm4[ctx->pm4_cdwords++] = 0; /* dst address hi */ 2116 ctx->pm4[ctx->pm4_cdwords++] = 0; /* unused */ 2117 ctx->pm4[ctx->pm4_cdwords++] = 0; /* unused */ 2118 2119 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); 2120 ctx->pm4[ctx->pm4_cdwords++] = 2121 r600_context_bo_reloc(ctx, t[i]->filled_size, 2122 RADEON_USAGE_WRITE); 2123 2124 flush_flags |= S_0085F0_SO0_DEST_BASE_ENA(1) << i; 2125 } 2126 } 2127 2128 if (ctx->screen->chip_class >= EVERGREEN) { 2129 evergreen_set_streamout_enable(ctx, 0); 2130 } else { 2131 r600_set_streamout_enable(ctx, 0); 2132 } 2133 2134 if (ctx->screen->family < CHIP_RV770) { 2135 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, 0); 2136 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_EVENT) | EVENT_INDEX(0); 2137 } else { 2138 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SURFACE_SYNC, 3, 0); 2139 ctx->pm4[ctx->pm4_cdwords++] = flush_flags; /* CP_COHER_CNTL */ 2140 ctx->pm4[ctx->pm4_cdwords++] = 0xffffffff; /* CP_COHER_SIZE */ 2141 ctx->pm4[ctx->pm4_cdwords++] = 0; /* CP_COHER_BASE */ 2142 ctx->pm4[ctx->pm4_cdwords++] = 0x0000000A; /* POLL_INTERVAL */ 2143 } 2144 2145 ctx->num_cs_dw_streamout_end = 0; 2146 2147 /* XXX print some debug info */ 2148 for (i = 0; i < ctx->num_so_targets; i++) { 2149 if (!t[i]) 2150 continue; 2151 2152 uint32_t *ptr = ctx->ws->buffer_map(t[i]->filled_size->buf, ctx->cs, RADEON_USAGE_READ); 2153 printf("FILLED_SIZE%i: %u\n", i, *ptr); 2154 ctx->ws->buffer_unmap(t[i]->filled_size->buf); 2155 } 2156} 2157 2158void r600_context_draw_opaque_count(struct r600_context *ctx, struct r600_so_target *t) 2159{ 2160 r600_need_cs_space(ctx, 14 + 21, TRUE); 2161 2162 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0); 2163 ctx->pm4[ctx->pm4_cdwords++] = (R_028B28_VGT_STRMOUT_DRAW_OPAQUE_OFFSET - R600_CONTEXT_REG_OFFSET) >> 2; 2164 ctx->pm4[ctx->pm4_cdwords++] = 0; 2165 2166 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0); 2167 ctx->pm4[ctx->pm4_cdwords++] = (R_028B30_VGT_STRMOUT_DRAW_OPAQUE_VERTEX_STRIDE - R600_CONTEXT_REG_OFFSET) >> 2; 2168 ctx->pm4[ctx->pm4_cdwords++] = t->stride >> 2; 2169 2170 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_COPY_DW, 4, 0); 2171 ctx->pm4[ctx->pm4_cdwords++] = COPY_DW_SRC_IS_MEM | COPY_DW_DST_IS_REG; 2172 ctx->pm4[ctx->pm4_cdwords++] = 0; /* src address lo */ 2173 ctx->pm4[ctx->pm4_cdwords++] = 0; /* src address hi */ 2174 ctx->pm4[ctx->pm4_cdwords++] = R_028B2C_VGT_STRMOUT_DRAW_OPAQUE_BUFFER_FILLED_SIZE >> 2; /* dst register */ 2175 ctx->pm4[ctx->pm4_cdwords++] = 0; /* unused */ 2176 2177 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); 2178 ctx->pm4[ctx->pm4_cdwords++] = r600_context_bo_reloc(ctx, t->filled_size, 2179 RADEON_USAGE_READ); 2180 2181#if 0 /* I have not found this useful yet. */ 2182 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_COPY_DW, 4, 0); 2183 ctx->pm4[ctx->pm4_cdwords++] = COPY_DW_SRC_IS_REG | COPY_DW_DST_IS_REG; 2184 ctx->pm4[ctx->pm4_cdwords++] = R_028B2C_VGT_STRMOUT_DRAW_OPAQUE_BUFFER_FILLED_SIZE >> 2; /* src register */ 2185 ctx->pm4[ctx->pm4_cdwords++] = 0; /* unused */ 2186 ctx->pm4[ctx->pm4_cdwords++] = R_0085F4_CP_COHER_SIZE >> 2; /* dst register */ 2187 ctx->pm4[ctx->pm4_cdwords++] = 0; /* unused */ 2188 2189 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_CONFIG_REG, 1, 0); 2190 ctx->pm4[ctx->pm4_cdwords++] = (R_0085F0_CP_COHER_CNTL - R600_CONFIG_REG_OFFSET) >> 2; 2191 ctx->pm4[ctx->pm4_cdwords++] = S_0085F0_SO0_DEST_BASE_ENA(1) << t->so_index; 2192 2193 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_CONFIG_REG, 1, 0); 2194 ctx->pm4[ctx->pm4_cdwords++] = (R_0085F8_CP_COHER_BASE - R600_CONFIG_REG_OFFSET) >> 2; 2195 ctx->pm4[ctx->pm4_cdwords++] = t->b.buffer_offset >> 2; 2196 2197 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); 2198 ctx->pm4[ctx->pm4_cdwords++] = r600_context_bo_reloc(ctx, (struct r600_resource*)t->b.buffer, 2199 RADEON_USAGE_WRITE); 2200 2201 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_WAIT_REG_MEM, 5, 0); 2202 ctx->pm4[ctx->pm4_cdwords++] = WAIT_REG_MEM_EQUAL; /* wait until the register is equal to the reference value */ 2203 ctx->pm4[ctx->pm4_cdwords++] = R_0085FC_CP_COHER_STATUS >> 2; /* register */ 2204 ctx->pm4[ctx->pm4_cdwords++] = 0; 2205 ctx->pm4[ctx->pm4_cdwords++] = 0; /* reference value */ 2206 ctx->pm4[ctx->pm4_cdwords++] = 0xffffffff; /* mask */ 2207 ctx->pm4[ctx->pm4_cdwords++] = 4; /* poll interval */ 2208#endif 2209} 2210