r600_hw_context.c revision d6cd514edbeca0de38561f66189748078d0dc602
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 uint64_t va = 0; 982 983 /* if bo has already been flushed */ 984 if (!(~bo->cs_buf->last_flush & flush_flags)) { 985 bo->cs_buf->last_flush &= flush_mask; 986 return; 987 } 988 989 if ((ctx->screen->family < CHIP_RV770) && 990 (G_0085F0_CB_ACTION_ENA(flush_flags) || 991 G_0085F0_DB_ACTION_ENA(flush_flags))) { 992 if (ctx->flags & R600_CONTEXT_CHECK_EVENT_FLUSH) { 993 /* the rv670 seems to fail fbo-generatemipmap unless we flush the CB1 dest base ena */ 994 if ((bo->cs_buf->binding & BO_BOUND_TEXTURE) && 995 (flush_flags & S_0085F0_CB_ACTION_ENA(1))) { 996 if ((ctx->screen->family == CHIP_RV670) || 997 (ctx->screen->family == CHIP_RS780) || 998 (ctx->screen->family == CHIP_RS880)) { 999 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SURFACE_SYNC, 3, 0); 1000 ctx->pm4[ctx->pm4_cdwords++] = S_0085F0_CB1_DEST_BASE_ENA(1); /* CP_COHER_CNTL */ 1001 ctx->pm4[ctx->pm4_cdwords++] = 0xffffffff; /* CP_COHER_SIZE */ 1002 ctx->pm4[ctx->pm4_cdwords++] = 0; /* CP_COHER_BASE */ 1003 ctx->pm4[ctx->pm4_cdwords++] = 0x0000000A; /* POLL_INTERVAL */ 1004 } 1005 } 1006 1007 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, 0); 1008 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_EVENT) | EVENT_INDEX(0); 1009 ctx->flags &= ~R600_CONTEXT_CHECK_EVENT_FLUSH; 1010 } 1011 } else { 1012 va = r600_resource_va(&ctx->screen->screen, (void *)bo); 1013 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SURFACE_SYNC, 3, 0); 1014 ctx->pm4[ctx->pm4_cdwords++] = flush_flags; 1015 ctx->pm4[ctx->pm4_cdwords++] = (bo->buf->size + 255) >> 8; 1016 ctx->pm4[ctx->pm4_cdwords++] = va >> 8; 1017 ctx->pm4[ctx->pm4_cdwords++] = 0x0000000A; 1018 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); 1019 ctx->pm4[ctx->pm4_cdwords++] = r600_context_bo_reloc(ctx, bo, RADEON_USAGE_WRITE); 1020 } 1021 bo->cs_buf->last_flush = (bo->cs_buf->last_flush | flush_flags) & flush_mask; 1022} 1023 1024void r600_context_reg(struct r600_context *ctx, 1025 unsigned offset, unsigned value, 1026 unsigned mask) 1027{ 1028 struct r600_range *range; 1029 struct r600_block *block; 1030 unsigned id; 1031 unsigned new_val; 1032 int dirty; 1033 1034 range = &ctx->range[CTX_RANGE_ID(offset)]; 1035 block = range->blocks[CTX_BLOCK_ID(offset)]; 1036 id = (offset - block->start_offset) >> 2; 1037 1038 dirty = block->status & R600_BLOCK_STATUS_DIRTY; 1039 1040 new_val = block->reg[id]; 1041 new_val &= ~mask; 1042 new_val |= value; 1043 if (new_val != block->reg[id]) { 1044 dirty |= R600_BLOCK_STATUS_DIRTY; 1045 block->reg[id] = new_val; 1046 } 1047 if (dirty) 1048 r600_context_dirty_block(ctx, block, dirty, id); 1049} 1050 1051void r600_context_dirty_block(struct r600_context *ctx, 1052 struct r600_block *block, 1053 int dirty, int index) 1054{ 1055 if ((index + 1) > block->nreg_dirty) 1056 block->nreg_dirty = index + 1; 1057 1058 if ((dirty != (block->status & R600_BLOCK_STATUS_DIRTY)) || !(block->status & R600_BLOCK_STATUS_ENABLED)) { 1059 block->status |= R600_BLOCK_STATUS_DIRTY; 1060 ctx->pm4_dirty_cdwords += block->pm4_ndwords + block->pm4_flush_ndwords; 1061 if (!(block->status & R600_BLOCK_STATUS_ENABLED)) { 1062 block->status |= R600_BLOCK_STATUS_ENABLED; 1063 LIST_ADDTAIL(&block->enable_list, &ctx->enable_list); 1064 } 1065 LIST_ADDTAIL(&block->list,&ctx->dirty); 1066 1067 if (block->flags & REG_FLAG_FLUSH_CHANGE) { 1068 r600_context_ps_partial_flush(ctx); 1069 } 1070 } 1071} 1072 1073void r600_context_pipe_state_set(struct r600_context *ctx, struct r600_pipe_state *state) 1074{ 1075 struct r600_block *block; 1076 unsigned new_val; 1077 int dirty; 1078 for (int i = 0; i < state->nregs; i++) { 1079 unsigned id, reloc_id; 1080 struct r600_pipe_reg *reg = &state->regs[i]; 1081 1082 block = reg->block; 1083 id = reg->id; 1084 1085 dirty = block->status & R600_BLOCK_STATUS_DIRTY; 1086 1087 new_val = block->reg[id]; 1088 new_val &= ~reg->mask; 1089 new_val |= reg->value; 1090 if (new_val != block->reg[id]) { 1091 block->reg[id] = new_val; 1092 dirty |= R600_BLOCK_STATUS_DIRTY; 1093 } 1094 if (block->flags & REG_FLAG_DIRTY_ALWAYS) 1095 dirty |= R600_BLOCK_STATUS_DIRTY; 1096 if (block->pm4_bo_index[id]) { 1097 /* find relocation */ 1098 reloc_id = block->pm4_bo_index[id]; 1099 pipe_resource_reference((struct pipe_resource**)&block->reloc[reloc_id].bo, ®->bo->b.b.b); 1100 block->reloc[reloc_id].bo_usage = reg->bo_usage; 1101 /* always force dirty for relocs for now */ 1102 dirty |= R600_BLOCK_STATUS_DIRTY; 1103 } 1104 1105 if (dirty) 1106 r600_context_dirty_block(ctx, block, dirty, id); 1107 } 1108} 1109 1110static void r600_context_dirty_resource_block(struct r600_context *ctx, 1111 struct r600_block *block, 1112 int dirty, int index) 1113{ 1114 block->nreg_dirty = index + 1; 1115 1116 if ((dirty != (block->status & R600_BLOCK_STATUS_RESOURCE_DIRTY)) || !(block->status & R600_BLOCK_STATUS_ENABLED)) { 1117 block->status |= R600_BLOCK_STATUS_RESOURCE_DIRTY; 1118 ctx->pm4_dirty_cdwords += block->pm4_ndwords + block->pm4_flush_ndwords; 1119 if (!(block->status & R600_BLOCK_STATUS_ENABLED)) { 1120 block->status |= R600_BLOCK_STATUS_ENABLED; 1121 LIST_ADDTAIL(&block->enable_list, &ctx->enable_list); 1122 } 1123 LIST_ADDTAIL(&block->list,&ctx->resource_dirty); 1124 } 1125} 1126 1127void r600_context_pipe_state_set_resource(struct r600_context *ctx, struct r600_pipe_resource_state *state, struct r600_block *block) 1128{ 1129 int dirty; 1130 int num_regs = ctx->screen->chip_class >= EVERGREEN ? 8 : 7; 1131 boolean is_vertex; 1132 1133 if (state == NULL) { 1134 block->status &= ~(R600_BLOCK_STATUS_ENABLED | R600_BLOCK_STATUS_RESOURCE_DIRTY); 1135 if (block->reloc[1].bo) 1136 block->reloc[1].bo->cs_buf->binding &= ~BO_BOUND_TEXTURE; 1137 1138 pipe_resource_reference((struct pipe_resource**)&block->reloc[1].bo, NULL); 1139 pipe_resource_reference((struct pipe_resource**)&block->reloc[2].bo, NULL); 1140 LIST_DELINIT(&block->list); 1141 LIST_DELINIT(&block->enable_list); 1142 return; 1143 } 1144 1145 is_vertex = ((state->val[num_regs-1] & 0xc0000000) == 0xc0000000); 1146 dirty = block->status & R600_BLOCK_STATUS_RESOURCE_DIRTY; 1147 1148 if (memcmp(block->reg, state->val, num_regs*4)) { 1149 memcpy(block->reg, state->val, num_regs * 4); 1150 dirty |= R600_BLOCK_STATUS_RESOURCE_DIRTY; 1151 } 1152 1153 /* if no BOs on block, force dirty */ 1154 if (!block->reloc[1].bo || !block->reloc[2].bo) 1155 dirty |= R600_BLOCK_STATUS_RESOURCE_DIRTY; 1156 1157 if (!dirty) { 1158 if (is_vertex) { 1159 if (block->reloc[1].bo->buf != state->bo[0]->buf) 1160 dirty |= R600_BLOCK_STATUS_RESOURCE_DIRTY; 1161 } else { 1162 if ((block->reloc[1].bo->buf != state->bo[0]->buf) || 1163 (block->reloc[2].bo->buf != state->bo[1]->buf)) 1164 dirty |= R600_BLOCK_STATUS_RESOURCE_DIRTY; 1165 } 1166 } 1167 1168 if (dirty) { 1169 if (is_vertex) { 1170 /* VERTEX RESOURCE, we preted there is 2 bo to relocate so 1171 * we have single case btw VERTEX & TEXTURE resource 1172 */ 1173 pipe_resource_reference((struct pipe_resource**)&block->reloc[1].bo, &state->bo[0]->b.b.b); 1174 block->reloc[1].bo_usage = state->bo_usage[0]; 1175 pipe_resource_reference((struct pipe_resource**)&block->reloc[2].bo, NULL); 1176 } else { 1177 /* TEXTURE RESOURCE */ 1178 pipe_resource_reference((struct pipe_resource**)&block->reloc[1].bo, &state->bo[0]->b.b.b); 1179 block->reloc[1].bo_usage = state->bo_usage[0]; 1180 pipe_resource_reference((struct pipe_resource**)&block->reloc[2].bo, &state->bo[1]->b.b.b); 1181 block->reloc[2].bo_usage = state->bo_usage[1]; 1182 state->bo[0]->cs_buf->binding |= BO_BOUND_TEXTURE; 1183 } 1184 1185 if (is_vertex) 1186 block->status |= R600_BLOCK_STATUS_RESOURCE_VERTEX; 1187 else 1188 block->status &= ~R600_BLOCK_STATUS_RESOURCE_VERTEX; 1189 1190 r600_context_dirty_resource_block(ctx, block, dirty, num_regs - 1); 1191 } 1192} 1193 1194void r600_context_pipe_state_set_ps_resource(struct r600_context *ctx, struct r600_pipe_resource_state *state, unsigned rid) 1195{ 1196 struct r600_block *block = ctx->ps_resources.blocks[rid]; 1197 1198 r600_context_pipe_state_set_resource(ctx, state, block); 1199} 1200 1201void r600_context_pipe_state_set_vs_resource(struct r600_context *ctx, struct r600_pipe_resource_state *state, unsigned rid) 1202{ 1203 struct r600_block *block = ctx->vs_resources.blocks[rid]; 1204 1205 r600_context_pipe_state_set_resource(ctx, state, block); 1206} 1207 1208void r600_context_pipe_state_set_fs_resource(struct r600_context *ctx, struct r600_pipe_resource_state *state, unsigned rid) 1209{ 1210 struct r600_block *block = ctx->fs_resources.blocks[rid]; 1211 1212 r600_context_pipe_state_set_resource(ctx, state, block); 1213} 1214 1215static inline void r600_context_pipe_state_set_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned offset) 1216{ 1217 struct r600_range *range; 1218 struct r600_block *block; 1219 int i; 1220 int dirty; 1221 1222 range = &ctx->range[CTX_RANGE_ID(offset)]; 1223 block = range->blocks[CTX_BLOCK_ID(offset)]; 1224 if (state == NULL) { 1225 block->status &= ~(R600_BLOCK_STATUS_ENABLED | R600_BLOCK_STATUS_DIRTY); 1226 LIST_DELINIT(&block->list); 1227 LIST_DELINIT(&block->enable_list); 1228 return; 1229 } 1230 dirty = block->status & R600_BLOCK_STATUS_DIRTY; 1231 for (i = 0; i < 3; i++) { 1232 if (block->reg[i] != state->regs[i].value) { 1233 block->reg[i] = state->regs[i].value; 1234 dirty |= R600_BLOCK_STATUS_DIRTY; 1235 } 1236 } 1237 1238 if (dirty) 1239 r600_context_dirty_block(ctx, block, dirty, 2); 1240} 1241 1242 1243static inline void r600_context_pipe_state_set_sampler_border(struct r600_context *ctx, struct r600_pipe_state *state, unsigned offset) 1244{ 1245 struct r600_range *range; 1246 struct r600_block *block; 1247 int i; 1248 int dirty; 1249 1250 range = &ctx->range[CTX_RANGE_ID(offset)]; 1251 block = range->blocks[CTX_BLOCK_ID(offset)]; 1252 if (state == NULL) { 1253 block->status &= ~(R600_BLOCK_STATUS_ENABLED | R600_BLOCK_STATUS_DIRTY); 1254 LIST_DELINIT(&block->list); 1255 LIST_DELINIT(&block->enable_list); 1256 return; 1257 } 1258 if (state->nregs <= 3) { 1259 return; 1260 } 1261 dirty = block->status & R600_BLOCK_STATUS_DIRTY; 1262 for (i = 0; i < 4; i++) { 1263 if (block->reg[i] != state->regs[i + 3].value) { 1264 block->reg[i] = state->regs[i + 3].value; 1265 dirty |= R600_BLOCK_STATUS_DIRTY; 1266 } 1267 } 1268 1269 /* We have to flush the shaders before we change the border color 1270 * registers, or previous draw commands that haven't completed yet 1271 * will end up using the new border color. */ 1272 if (dirty & R600_BLOCK_STATUS_DIRTY) 1273 r600_context_ps_partial_flush(ctx); 1274 if (dirty) 1275 r600_context_dirty_block(ctx, block, dirty, 3); 1276} 1277 1278void r600_context_pipe_state_set_ps_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id) 1279{ 1280 unsigned offset; 1281 1282 offset = 0x0003C000 + id * 0xc; 1283 r600_context_pipe_state_set_sampler(ctx, state, offset); 1284 offset = 0x0000A400 + id * 0x10; 1285 r600_context_pipe_state_set_sampler_border(ctx, state, offset); 1286} 1287 1288void r600_context_pipe_state_set_vs_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id) 1289{ 1290 unsigned offset; 1291 1292 offset = 0x0003C0D8 + id * 0xc; 1293 r600_context_pipe_state_set_sampler(ctx, state, offset); 1294 offset = 0x0000A600 + id * 0x10; 1295 r600_context_pipe_state_set_sampler_border(ctx, state, offset); 1296} 1297 1298struct r600_resource *r600_context_reg_bo(struct r600_context *ctx, unsigned offset) 1299{ 1300 struct r600_range *range; 1301 struct r600_block *block; 1302 unsigned id; 1303 1304 range = &ctx->range[CTX_RANGE_ID(offset)]; 1305 block = range->blocks[CTX_BLOCK_ID(offset)]; 1306 offset -= block->start_offset; 1307 id = block->pm4_bo_index[offset >> 2]; 1308 if (block->reloc[id].bo) { 1309 return block->reloc[id].bo; 1310 } 1311 return NULL; 1312} 1313 1314void r600_context_block_emit_dirty(struct r600_context *ctx, struct r600_block *block) 1315{ 1316 int optional = block->nbo == 0 && !(block->flags & REG_FLAG_DIRTY_ALWAYS); 1317 int cp_dwords = block->pm4_ndwords, start_dword = 0; 1318 int new_dwords = 0; 1319 int nbo = block->nbo; 1320 1321 if (block->nreg_dirty == 0 && optional) { 1322 goto out; 1323 } 1324 1325 if (nbo) { 1326 ctx->flags |= R600_CONTEXT_CHECK_EVENT_FLUSH; 1327 1328 for (int j = 0; j < block->nreg; j++) { 1329 if (block->pm4_bo_index[j]) { 1330 /* find relocation */ 1331 struct r600_block_reloc *reloc = &block->reloc[block->pm4_bo_index[j]]; 1332 block->pm4[reloc->bo_pm4_index] = 1333 r600_context_bo_reloc(ctx, reloc->bo, reloc->bo_usage); 1334 r600_context_bo_flush(ctx, 1335 reloc->flush_flags, 1336 reloc->flush_mask, 1337 reloc->bo); 1338 nbo--; 1339 if (nbo == 0) 1340 break; 1341 } 1342 } 1343 ctx->flags &= ~R600_CONTEXT_CHECK_EVENT_FLUSH; 1344 } 1345 1346 optional &= (block->nreg_dirty != block->nreg); 1347 if (optional) { 1348 new_dwords = block->nreg_dirty; 1349 start_dword = ctx->pm4_cdwords; 1350 cp_dwords = new_dwords + 2; 1351 } 1352 memcpy(&ctx->pm4[ctx->pm4_cdwords], block->pm4, cp_dwords * 4); 1353 ctx->pm4_cdwords += cp_dwords; 1354 1355 if (optional) { 1356 uint32_t newword; 1357 1358 newword = ctx->pm4[start_dword]; 1359 newword &= PKT_COUNT_C; 1360 newword |= PKT_COUNT_S(new_dwords); 1361 ctx->pm4[start_dword] = newword; 1362 } 1363out: 1364 block->status ^= R600_BLOCK_STATUS_DIRTY; 1365 block->nreg_dirty = 0; 1366 LIST_DELINIT(&block->list); 1367} 1368 1369void r600_context_block_resource_emit_dirty(struct r600_context *ctx, struct r600_block *block) 1370{ 1371 int cp_dwords = block->pm4_ndwords; 1372 int nbo = block->nbo; 1373 1374 ctx->flags |= R600_CONTEXT_CHECK_EVENT_FLUSH; 1375 1376 if (block->status & R600_BLOCK_STATUS_RESOURCE_VERTEX) { 1377 nbo = 1; 1378 cp_dwords -= 2; /* don't copy the second NOP */ 1379 } 1380 1381 for (int j = 0; j < nbo; j++) { 1382 if (block->pm4_bo_index[j]) { 1383 /* find relocation */ 1384 struct r600_block_reloc *reloc = &block->reloc[block->pm4_bo_index[j]]; 1385 block->pm4[reloc->bo_pm4_index] = 1386 r600_context_bo_reloc(ctx, reloc->bo, reloc->bo_usage); 1387 r600_context_bo_flush(ctx, 1388 reloc->flush_flags, 1389 reloc->flush_mask, 1390 reloc->bo); 1391 } 1392 } 1393 ctx->flags &= ~R600_CONTEXT_CHECK_EVENT_FLUSH; 1394 1395 memcpy(&ctx->pm4[ctx->pm4_cdwords], block->pm4, cp_dwords * 4); 1396 ctx->pm4_cdwords += cp_dwords; 1397 1398 block->status ^= R600_BLOCK_STATUS_RESOURCE_DIRTY; 1399 block->nreg_dirty = 0; 1400 LIST_DELINIT(&block->list); 1401} 1402 1403void r600_context_flush_dest_caches(struct r600_context *ctx) 1404{ 1405 struct r600_resource *cb[8]; 1406 struct r600_resource *db; 1407 int i; 1408 1409 if (!(ctx->flags & R600_CONTEXT_DST_CACHES_DIRTY)) 1410 return; 1411 1412 db = r600_context_reg_bo(ctx, R_02800C_DB_DEPTH_BASE); 1413 cb[0] = r600_context_reg_bo(ctx, R_028040_CB_COLOR0_BASE); 1414 cb[1] = r600_context_reg_bo(ctx, R_028044_CB_COLOR1_BASE); 1415 cb[2] = r600_context_reg_bo(ctx, R_028048_CB_COLOR2_BASE); 1416 cb[3] = r600_context_reg_bo(ctx, R_02804C_CB_COLOR3_BASE); 1417 cb[4] = r600_context_reg_bo(ctx, R_028050_CB_COLOR4_BASE); 1418 cb[5] = r600_context_reg_bo(ctx, R_028054_CB_COLOR5_BASE); 1419 cb[6] = r600_context_reg_bo(ctx, R_028058_CB_COLOR6_BASE); 1420 cb[7] = r600_context_reg_bo(ctx, R_02805C_CB_COLOR7_BASE); 1421 1422 ctx->flags |= R600_CONTEXT_CHECK_EVENT_FLUSH; 1423 /* flush the color buffers */ 1424 for (i = 0; i < 8; i++) { 1425 if (!cb[i]) 1426 continue; 1427 1428 r600_context_bo_flush(ctx, 1429 (S_0085F0_CB0_DEST_BASE_ENA(1) << i) | 1430 S_0085F0_CB_ACTION_ENA(1), 1431 0, cb[i]); 1432 } 1433 if (db) { 1434 r600_context_bo_flush(ctx, S_0085F0_DB_ACTION_ENA(1) | S_0085F0_DB_DEST_BASE_ENA(1), 0, db); 1435 } 1436 ctx->flags &= ~R600_CONTEXT_CHECK_EVENT_FLUSH; 1437 ctx->flags &= ~R600_CONTEXT_DST_CACHES_DIRTY; 1438} 1439 1440void r600_context_draw(struct r600_context *ctx, const struct r600_draw *draw) 1441{ 1442 unsigned ndwords = 7; 1443 struct r600_block *dirty_block = NULL; 1444 struct r600_block *next_block; 1445 uint32_t *pm4; 1446 1447 if (draw->indices) { 1448 ndwords = 11; 1449 } 1450 /* when increasing ndwords, bump the max limit too */ 1451 assert(ndwords <= R600_MAX_DRAW_CS_DWORDS); 1452 1453 /* queries need some special values 1454 * (this is non-zero if any query is active) */ 1455 if (ctx->num_cs_dw_queries_suspend) { 1456 if (ctx->screen->family >= CHIP_RV770) { 1457 r600_context_reg(ctx, 1458 R_028D0C_DB_RENDER_CONTROL, 1459 S_028D0C_R700_PERFECT_ZPASS_COUNTS(1), 1460 S_028D0C_R700_PERFECT_ZPASS_COUNTS(1)); 1461 } 1462 r600_context_reg(ctx, 1463 R_028D10_DB_RENDER_OVERRIDE, 1464 S_028D10_NOOP_CULL_DISABLE(1), 1465 S_028D10_NOOP_CULL_DISABLE(1)); 1466 } 1467 1468 r600_need_cs_space(ctx, 0, TRUE); 1469 assert(ctx->pm4_cdwords + ctx->pm4_dirty_cdwords + ndwords < RADEON_MAX_CMDBUF_DWORDS); 1470 1471 /* enough room to copy packet */ 1472 LIST_FOR_EACH_ENTRY_SAFE(dirty_block, next_block, &ctx->dirty, list) { 1473 r600_context_block_emit_dirty(ctx, dirty_block); 1474 } 1475 1476 LIST_FOR_EACH_ENTRY_SAFE(dirty_block, next_block, &ctx->resource_dirty, list) { 1477 r600_context_block_resource_emit_dirty(ctx, dirty_block); 1478 } 1479 1480 /* Enable stream out if needed. */ 1481 if (ctx->streamout_start) { 1482 r600_context_streamout_begin(ctx); 1483 ctx->streamout_start = FALSE; 1484 } 1485 1486 /* draw packet */ 1487 pm4 = &ctx->pm4[ctx->pm4_cdwords]; 1488 1489 pm4[0] = PKT3(PKT3_INDEX_TYPE, 0, ctx->predicate_drawing); 1490 pm4[1] = draw->vgt_index_type; 1491 pm4[2] = PKT3(PKT3_NUM_INSTANCES, 0, ctx->predicate_drawing); 1492 pm4[3] = draw->vgt_num_instances; 1493 if (draw->indices) { 1494 pm4[4] = PKT3(PKT3_DRAW_INDEX, 3, ctx->predicate_drawing); 1495 pm4[5] = draw->indices_bo_offset; 1496 pm4[6] = 0; 1497 pm4[7] = draw->vgt_num_indices; 1498 pm4[8] = draw->vgt_draw_initiator; 1499 pm4[9] = PKT3(PKT3_NOP, 0, ctx->predicate_drawing); 1500 pm4[10] = r600_context_bo_reloc(ctx, draw->indices, RADEON_USAGE_READ); 1501 } else { 1502 pm4[4] = PKT3(PKT3_DRAW_INDEX_AUTO, 1, ctx->predicate_drawing); 1503 pm4[5] = draw->vgt_num_indices; 1504 pm4[6] = draw->vgt_draw_initiator; 1505 } 1506 ctx->pm4_cdwords += ndwords; 1507 1508 ctx->flags |= (R600_CONTEXT_DST_CACHES_DIRTY | R600_CONTEXT_DRAW_PENDING); 1509 1510 /* all dirty state have been scheduled in current cs */ 1511 ctx->pm4_dirty_cdwords = 0; 1512} 1513 1514void r600_context_flush(struct r600_context *ctx, unsigned flags) 1515{ 1516 struct r600_block *enable_block = NULL; 1517 bool queries_suspended = false; 1518 bool streamout_suspended = false; 1519 1520 if (ctx->pm4_cdwords == ctx->init_dwords) 1521 return; 1522 1523 /* suspend queries */ 1524 if (ctx->num_cs_dw_queries_suspend) { 1525 r600_context_queries_suspend(ctx); 1526 queries_suspended = true; 1527 } 1528 1529 if (ctx->num_cs_dw_streamout_end) { 1530 r600_context_streamout_end(ctx); 1531 streamout_suspended = true; 1532 } 1533 1534 if (ctx->screen->chip_class >= EVERGREEN) 1535 evergreen_context_flush_dest_caches(ctx); 1536 else 1537 r600_context_flush_dest_caches(ctx); 1538 1539 /* partial flush is needed to avoid lockups on some chips with user fences */ 1540 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, 0); 1541 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4); 1542 1543 /* Flush the CS. */ 1544 ctx->cs->cdw = ctx->pm4_cdwords; 1545 ctx->ws->cs_flush(ctx->cs, flags); 1546 1547 /* We need to get the pointer to the other CS, 1548 * the command streams are double-buffered. */ 1549 ctx->pm4 = ctx->cs->buf; 1550 1551 /* restart */ 1552 for (int i = 0; i < ctx->creloc; i++) { 1553 ctx->bo[i]->cs_buf->last_flush = 0; 1554 pipe_resource_reference((struct pipe_resource**)&ctx->bo[i], NULL); 1555 } 1556 ctx->creloc = 0; 1557 ctx->pm4_dirty_cdwords = 0; 1558 ctx->pm4_cdwords = 0; 1559 ctx->flags = 0; 1560 1561 r600_init_cs(ctx); 1562 1563 if (streamout_suspended) { 1564 ctx->streamout_start = TRUE; 1565 ctx->streamout_append_bitmask = ~0; 1566 } 1567 1568 /* resume queries */ 1569 if (queries_suspended) { 1570 r600_context_queries_resume(ctx); 1571 } 1572 1573 /* set all valid group as dirty so they get reemited on 1574 * next draw command 1575 */ 1576 LIST_FOR_EACH_ENTRY(enable_block, &ctx->enable_list, enable_list) { 1577 if (!(enable_block->flags & BLOCK_FLAG_RESOURCE)) { 1578 if(!(enable_block->status & R600_BLOCK_STATUS_DIRTY)) { 1579 LIST_ADDTAIL(&enable_block->list,&ctx->dirty); 1580 enable_block->status |= R600_BLOCK_STATUS_DIRTY; 1581 } 1582 } else { 1583 if(!(enable_block->status & R600_BLOCK_STATUS_RESOURCE_DIRTY)) { 1584 LIST_ADDTAIL(&enable_block->list,&ctx->resource_dirty); 1585 enable_block->status |= R600_BLOCK_STATUS_RESOURCE_DIRTY; 1586 } 1587 } 1588 ctx->pm4_dirty_cdwords += enable_block->pm4_ndwords + 1589 enable_block->pm4_flush_ndwords; 1590 enable_block->nreg_dirty = enable_block->nreg; 1591 } 1592} 1593 1594void r600_context_emit_fence(struct r600_context *ctx, struct r600_resource *fence_bo, unsigned offset, unsigned value) 1595{ 1596 uint64_t va; 1597 1598 r600_need_cs_space(ctx, 10, FALSE); 1599 1600 va = r600_resource_va(&ctx->screen->screen, (void*)fence_bo); 1601 va = va + (offset << 2); 1602 1603 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, 0); 1604 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4); 1605 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0); 1606 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5); 1607 ctx->pm4[ctx->pm4_cdwords++] = va & 0xFFFFFFFFUL; /* ADDRESS_LO */ 1608 /* DATA_SEL | INT_EN | ADDRESS_HI */ 1609 ctx->pm4[ctx->pm4_cdwords++] = (1 << 29) | (0 << 24) | ((va >> 32UL) & 0xFF); 1610 ctx->pm4[ctx->pm4_cdwords++] = value; /* DATA_LO */ 1611 ctx->pm4[ctx->pm4_cdwords++] = 0; /* DATA_HI */ 1612 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); 1613 ctx->pm4[ctx->pm4_cdwords++] = r600_context_bo_reloc(ctx, fence_bo, RADEON_USAGE_WRITE); 1614} 1615 1616static unsigned r600_query_read_result(char *map, unsigned start_index, unsigned end_index, 1617 bool test_status_bit) 1618{ 1619 uint32_t *current_result = (uint32_t*)map; 1620 uint64_t start, end; 1621 1622 start = (uint64_t)current_result[start_index] | 1623 (uint64_t)current_result[start_index+1] << 32; 1624 end = (uint64_t)current_result[end_index] | 1625 (uint64_t)current_result[end_index+1] << 32; 1626 1627 if (!test_status_bit || 1628 ((start & 0x8000000000000000UL) && (end & 0x8000000000000000UL))) { 1629 return end - start; 1630 } 1631 return 0; 1632} 1633 1634static boolean r600_query_result(struct r600_context *ctx, struct r600_query *query, boolean wait) 1635{ 1636 unsigned results_base = query->results_start; 1637 char *map; 1638 1639 map = ctx->ws->buffer_map(query->buffer->buf, ctx->cs, 1640 PIPE_TRANSFER_READ | 1641 (wait ? 0 : PIPE_TRANSFER_DONTBLOCK)); 1642 if (!map) 1643 return FALSE; 1644 1645 /* count all results across all data blocks */ 1646 switch (query->type) { 1647 case PIPE_QUERY_OCCLUSION_COUNTER: 1648 while (results_base != query->results_end) { 1649 query->result.u64 += 1650 r600_query_read_result(map + results_base, 0, 2, true); 1651 results_base = (results_base + 16) % query->buffer->b.b.b.width0; 1652 } 1653 break; 1654 case PIPE_QUERY_OCCLUSION_PREDICATE: 1655 while (results_base != query->results_end) { 1656 query->result.b = query->result.b || 1657 r600_query_read_result(map + results_base, 0, 2, true) != 0; 1658 results_base = (results_base + 16) % query->buffer->b.b.b.width0; 1659 } 1660 break; 1661 case PIPE_QUERY_TIME_ELAPSED: 1662 while (results_base != query->results_end) { 1663 query->result.u64 += 1664 r600_query_read_result(map + results_base, 0, 2, false); 1665 results_base = (results_base + query->result_size) % query->buffer->b.b.b.width0; 1666 } 1667 break; 1668 case PIPE_QUERY_PRIMITIVES_EMITTED: 1669 /* SAMPLE_STREAMOUTSTATS stores this structure: 1670 * { 1671 * u64 NumPrimitivesWritten; 1672 * u64 PrimitiveStorageNeeded; 1673 * } 1674 * We only need NumPrimitivesWritten here. */ 1675 while (results_base != query->results_end) { 1676 query->result.u64 += 1677 r600_query_read_result(map + results_base, 2, 6, true); 1678 results_base = (results_base + query->result_size) % query->buffer->b.b.b.width0; 1679 } 1680 break; 1681 case PIPE_QUERY_PRIMITIVES_GENERATED: 1682 /* Here we read PrimitiveStorageNeeded. */ 1683 while (results_base != query->results_end) { 1684 query->result.u64 += 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_STATISTICS: 1690 while (results_base != query->results_end) { 1691 query->result.so.num_primitives_written += 1692 r600_query_read_result(map + results_base, 2, 6, true); 1693 query->result.so.primitives_storage_needed += 1694 r600_query_read_result(map + results_base, 0, 4, true); 1695 results_base = (results_base + query->result_size) % query->buffer->b.b.b.width0; 1696 } 1697 break; 1698 case PIPE_QUERY_SO_OVERFLOW_PREDICATE: 1699 while (results_base != query->results_end) { 1700 query->result.b = query->result.b || 1701 r600_query_read_result(map + results_base, 2, 6, true) != 1702 r600_query_read_result(map + results_base, 0, 4, true); 1703 results_base = (results_base + query->result_size) % query->buffer->b.b.b.width0; 1704 } 1705 break; 1706 default: 1707 assert(0); 1708 } 1709 1710 query->results_start = query->results_end; 1711 ctx->ws->buffer_unmap(query->buffer->buf); 1712 return TRUE; 1713} 1714 1715void r600_query_begin(struct r600_context *ctx, struct r600_query *query) 1716{ 1717 unsigned new_results_end, i; 1718 u32 *results; 1719 uint64_t va; 1720 1721 r600_need_cs_space(ctx, query->num_cs_dw * 2, TRUE); 1722 1723 new_results_end = (query->results_end + query->result_size) % query->buffer->b.b.b.width0; 1724 1725 /* collect current results if query buffer is full */ 1726 if (new_results_end == query->results_start) { 1727 r600_query_result(ctx, query, TRUE); 1728 } 1729 1730 switch (query->type) { 1731 case PIPE_QUERY_OCCLUSION_COUNTER: 1732 case PIPE_QUERY_OCCLUSION_PREDICATE: 1733 results = ctx->ws->buffer_map(query->buffer->buf, ctx->cs, PIPE_TRANSFER_WRITE); 1734 if (results) { 1735 results = (u32*)((char*)results + query->results_end); 1736 memset(results, 0, query->result_size); 1737 1738 /* Set top bits for unused backends */ 1739 for (i = 0; i < ctx->max_db; i++) { 1740 if (!(ctx->backend_mask & (1<<i))) { 1741 results[(i * 4)+1] = 0x80000000; 1742 results[(i * 4)+3] = 0x80000000; 1743 } 1744 } 1745 ctx->ws->buffer_unmap(query->buffer->buf); 1746 } 1747 break; 1748 case PIPE_QUERY_TIME_ELAPSED: 1749 break; 1750 case PIPE_QUERY_PRIMITIVES_EMITTED: 1751 case PIPE_QUERY_PRIMITIVES_GENERATED: 1752 case PIPE_QUERY_SO_STATISTICS: 1753 case PIPE_QUERY_SO_OVERFLOW_PREDICATE: 1754 results = ctx->ws->buffer_map(query->buffer->buf, ctx->cs, PIPE_TRANSFER_WRITE); 1755 results = (u32*)((char*)results + query->results_end); 1756 memset(results, 0, query->result_size); 1757 ctx->ws->buffer_unmap(query->buffer->buf); 1758 break; 1759 default: 1760 assert(0); 1761 } 1762 1763 /* emit begin query */ 1764 va = r600_resource_va(&ctx->screen->screen, (void*)query->buffer); 1765 va += query->results_end; 1766 1767 switch (query->type) { 1768 case PIPE_QUERY_OCCLUSION_COUNTER: 1769 case PIPE_QUERY_OCCLUSION_PREDICATE: 1770 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 2, 0); 1771 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_ZPASS_DONE) | EVENT_INDEX(1); 1772 ctx->pm4[ctx->pm4_cdwords++] = va; 1773 ctx->pm4[ctx->pm4_cdwords++] = (va >> 32UL) & 0xFF; 1774 break; 1775 case PIPE_QUERY_PRIMITIVES_EMITTED: 1776 case PIPE_QUERY_PRIMITIVES_GENERATED: 1777 case PIPE_QUERY_SO_STATISTICS: 1778 case PIPE_QUERY_SO_OVERFLOW_PREDICATE: 1779 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 2, 0); 1780 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_SAMPLE_STREAMOUTSTATS) | EVENT_INDEX(3); 1781 ctx->pm4[ctx->pm4_cdwords++] = query->results_end; 1782 ctx->pm4[ctx->pm4_cdwords++] = 0; 1783 break; 1784 case PIPE_QUERY_TIME_ELAPSED: 1785 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0); 1786 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5); 1787 ctx->pm4[ctx->pm4_cdwords++] = va; 1788 ctx->pm4[ctx->pm4_cdwords++] = (3 << 29) | ((va >> 32UL) & 0xFF); 1789 ctx->pm4[ctx->pm4_cdwords++] = 0; 1790 ctx->pm4[ctx->pm4_cdwords++] = 0; 1791 break; 1792 default: 1793 assert(0); 1794 } 1795 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); 1796 ctx->pm4[ctx->pm4_cdwords++] = r600_context_bo_reloc(ctx, query->buffer, RADEON_USAGE_WRITE); 1797 1798 ctx->num_cs_dw_queries_suspend += query->num_cs_dw; 1799} 1800 1801void r600_query_end(struct r600_context *ctx, struct r600_query *query) 1802{ 1803 uint64_t va; 1804 1805 va = r600_resource_va(&ctx->screen->screen, (void*)query->buffer); 1806 /* emit end query */ 1807 switch (query->type) { 1808 case PIPE_QUERY_OCCLUSION_COUNTER: 1809 case PIPE_QUERY_OCCLUSION_PREDICATE: 1810 va += query->results_end + 8; 1811 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 2, 0); 1812 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_ZPASS_DONE) | EVENT_INDEX(1); 1813 ctx->pm4[ctx->pm4_cdwords++] = va; 1814 ctx->pm4[ctx->pm4_cdwords++] = (va >> 32UL) & 0xFF; 1815 break; 1816 case PIPE_QUERY_PRIMITIVES_EMITTED: 1817 case PIPE_QUERY_PRIMITIVES_GENERATED: 1818 case PIPE_QUERY_SO_STATISTICS: 1819 case PIPE_QUERY_SO_OVERFLOW_PREDICATE: 1820 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 2, 0); 1821 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_SAMPLE_STREAMOUTSTATS) | EVENT_INDEX(3); 1822 ctx->pm4[ctx->pm4_cdwords++] = query->results_end + query->result_size/2; 1823 ctx->pm4[ctx->pm4_cdwords++] = 0; 1824 break; 1825 case PIPE_QUERY_TIME_ELAPSED: 1826 va += query->results_end + query->result_size/2; 1827 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0); 1828 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5); 1829 ctx->pm4[ctx->pm4_cdwords++] = va; 1830 ctx->pm4[ctx->pm4_cdwords++] = (3 << 29) | ((va >> 32UL) & 0xFF); 1831 ctx->pm4[ctx->pm4_cdwords++] = 0; 1832 ctx->pm4[ctx->pm4_cdwords++] = 0; 1833 break; 1834 default: 1835 assert(0); 1836 } 1837 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); 1838 ctx->pm4[ctx->pm4_cdwords++] = r600_context_bo_reloc(ctx, query->buffer, RADEON_USAGE_WRITE); 1839 1840 query->results_end = (query->results_end + query->result_size) % query->buffer->b.b.b.width0; 1841 ctx->num_cs_dw_queries_suspend -= query->num_cs_dw; 1842} 1843 1844void r600_query_predication(struct r600_context *ctx, struct r600_query *query, int operation, 1845 int flag_wait) 1846{ 1847 uint64_t va; 1848 1849 if (operation == PREDICATION_OP_CLEAR) { 1850 r600_need_cs_space(ctx, 3, FALSE); 1851 1852 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_PREDICATION, 1, 0); 1853 ctx->pm4[ctx->pm4_cdwords++] = 0; 1854 ctx->pm4[ctx->pm4_cdwords++] = PRED_OP(PREDICATION_OP_CLEAR); 1855 } else { 1856 unsigned results_base = query->results_start; 1857 unsigned count; 1858 u32 op; 1859 1860 /* find count of the query data blocks */ 1861 count = (query->buffer->b.b.b.width0 + query->results_end - query->results_start) % query->buffer->b.b.b.width0; 1862 count /= query->result_size; 1863 1864 r600_need_cs_space(ctx, 5 * count, TRUE); 1865 1866 op = PRED_OP(operation) | PREDICATION_DRAW_VISIBLE | 1867 (flag_wait ? PREDICATION_HINT_WAIT : PREDICATION_HINT_NOWAIT_DRAW); 1868 va = r600_resource_va(&ctx->screen->screen, (void*)query->buffer); 1869 1870 /* emit predicate packets for all data blocks */ 1871 while (results_base != query->results_end) { 1872 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_PREDICATION, 1, 0); 1873 ctx->pm4[ctx->pm4_cdwords++] = (va + results_base) & 0xFFFFFFFFUL; 1874 ctx->pm4[ctx->pm4_cdwords++] = op | (((va + results_base) >> 32UL) & 0xFF); 1875 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); 1876 ctx->pm4[ctx->pm4_cdwords++] = r600_context_bo_reloc(ctx, query->buffer, 1877 RADEON_USAGE_READ); 1878 results_base = (results_base + query->result_size) % query->buffer->b.b.b.width0; 1879 1880 /* set CONTINUE bit for all packets except the first */ 1881 op |= PREDICATION_CONTINUE; 1882 } 1883 } 1884} 1885 1886struct r600_query *r600_context_query_create(struct r600_context *ctx, unsigned query_type) 1887{ 1888 struct r600_query *query; 1889 unsigned buffer_size = 4096; 1890 1891 query = CALLOC_STRUCT(r600_query); 1892 if (query == NULL) 1893 return NULL; 1894 1895 query->type = query_type; 1896 1897 switch (query_type) { 1898 case PIPE_QUERY_OCCLUSION_COUNTER: 1899 case PIPE_QUERY_OCCLUSION_PREDICATE: 1900 query->result_size = 16 * ctx->max_db; 1901 query->num_cs_dw = 6; 1902 break; 1903 case PIPE_QUERY_TIME_ELAPSED: 1904 query->result_size = 16; 1905 query->num_cs_dw = 8; 1906 break; 1907 case PIPE_QUERY_PRIMITIVES_EMITTED: 1908 case PIPE_QUERY_PRIMITIVES_GENERATED: 1909 case PIPE_QUERY_SO_STATISTICS: 1910 case PIPE_QUERY_SO_OVERFLOW_PREDICATE: 1911 /* NumPrimitivesWritten, PrimitiveStorageNeeded. */ 1912 query->result_size = 32; 1913 query->num_cs_dw = 6; 1914 break; 1915 default: 1916 assert(0); 1917 FREE(query); 1918 return NULL; 1919 } 1920 1921 /* adjust buffer size to simplify offsets wrapping math */ 1922 buffer_size -= buffer_size % query->result_size; 1923 1924 /* Queries are normally read by the CPU after 1925 * being written by the gpu, hence staging is probably a good 1926 * usage pattern. 1927 */ 1928 query->buffer = (struct r600_resource*) 1929 pipe_buffer_create(&ctx->screen->screen, PIPE_BIND_CUSTOM, PIPE_USAGE_STAGING, buffer_size); 1930 if (!query->buffer) { 1931 FREE(query); 1932 return NULL; 1933 } 1934 return query; 1935} 1936 1937void r600_context_query_destroy(struct r600_context *ctx, struct r600_query *query) 1938{ 1939 pipe_resource_reference((struct pipe_resource**)&query->buffer, NULL); 1940 free(query); 1941} 1942 1943boolean r600_context_query_result(struct r600_context *ctx, 1944 struct r600_query *query, 1945 boolean wait, void *vresult) 1946{ 1947 boolean *result_b = (boolean*)vresult; 1948 uint64_t *result_u64 = (uint64_t*)vresult; 1949 struct pipe_query_data_so_statistics *result_so = 1950 (struct pipe_query_data_so_statistics*)vresult; 1951 1952 if (!r600_query_result(ctx, query, wait)) 1953 return FALSE; 1954 1955 switch (query->type) { 1956 case PIPE_QUERY_OCCLUSION_COUNTER: 1957 case PIPE_QUERY_PRIMITIVES_EMITTED: 1958 case PIPE_QUERY_PRIMITIVES_GENERATED: 1959 *result_u64 = query->result.u64; 1960 break; 1961 case PIPE_QUERY_OCCLUSION_PREDICATE: 1962 case PIPE_QUERY_SO_OVERFLOW_PREDICATE: 1963 *result_b = query->result.b; 1964 break; 1965 case PIPE_QUERY_TIME_ELAPSED: 1966 *result_u64 = (1000000 * query->result.u64) / ctx->screen->info.r600_clock_crystal_freq; 1967 break; 1968 case PIPE_QUERY_SO_STATISTICS: 1969 *result_so = query->result.so; 1970 break; 1971 default: 1972 assert(0); 1973 } 1974 return TRUE; 1975} 1976 1977void r600_context_queries_suspend(struct r600_context *ctx) 1978{ 1979 struct r600_query *query; 1980 1981 LIST_FOR_EACH_ENTRY(query, &ctx->active_query_list, list) { 1982 r600_query_end(ctx, query); 1983 } 1984 assert(ctx->num_cs_dw_queries_suspend == 0); 1985} 1986 1987void r600_context_queries_resume(struct r600_context *ctx) 1988{ 1989 struct r600_query *query; 1990 1991 assert(ctx->num_cs_dw_queries_suspend == 0); 1992 1993 LIST_FOR_EACH_ENTRY(query, &ctx->active_query_list, list) { 1994 r600_query_begin(ctx, query); 1995 } 1996} 1997 1998static void r600_flush_vgt_streamout(struct r600_context *ctx) 1999{ 2000 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_CONFIG_REG, 1, 0); 2001 ctx->pm4[ctx->pm4_cdwords++] = (R_008490_CP_STRMOUT_CNTL - R600_CONFIG_REG_OFFSET) >> 2; 2002 ctx->pm4[ctx->pm4_cdwords++] = 0; 2003 2004 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, 0); 2005 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_SO_VGTSTREAMOUT_FLUSH) | EVENT_INDEX(0); 2006 2007 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_WAIT_REG_MEM, 5, 0); 2008 ctx->pm4[ctx->pm4_cdwords++] = WAIT_REG_MEM_EQUAL; /* wait until the register is equal to the reference value */ 2009 ctx->pm4[ctx->pm4_cdwords++] = R_008490_CP_STRMOUT_CNTL >> 2; /* register */ 2010 ctx->pm4[ctx->pm4_cdwords++] = 0; 2011 ctx->pm4[ctx->pm4_cdwords++] = S_008490_OFFSET_UPDATE_DONE(1); /* reference value */ 2012 ctx->pm4[ctx->pm4_cdwords++] = S_008490_OFFSET_UPDATE_DONE(1); /* mask */ 2013 ctx->pm4[ctx->pm4_cdwords++] = 4; /* poll interval */ 2014} 2015 2016static void r600_set_streamout_enable(struct r600_context *ctx, unsigned buffer_enable_bit) 2017{ 2018 if (buffer_enable_bit) { 2019 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0); 2020 ctx->pm4[ctx->pm4_cdwords++] = (R_028AB0_VGT_STRMOUT_EN - R600_CONTEXT_REG_OFFSET) >> 2; 2021 ctx->pm4[ctx->pm4_cdwords++] = S_028AB0_STREAMOUT(1); 2022 2023 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0); 2024 ctx->pm4[ctx->pm4_cdwords++] = (R_028B20_VGT_STRMOUT_BUFFER_EN - R600_CONTEXT_REG_OFFSET) >> 2; 2025 ctx->pm4[ctx->pm4_cdwords++] = S_028B20_BUFFER_0_EN(buffer_enable_bit); 2026 } else { 2027 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0); 2028 ctx->pm4[ctx->pm4_cdwords++] = (R_028AB0_VGT_STRMOUT_EN - R600_CONTEXT_REG_OFFSET) >> 2; 2029 ctx->pm4[ctx->pm4_cdwords++] = S_028AB0_STREAMOUT(0); 2030 } 2031} 2032 2033void r600_context_streamout_begin(struct r600_context *ctx) 2034{ 2035 struct r600_so_target **t = ctx->so_targets; 2036 unsigned *stride_in_dw = ctx->vs_so_stride_in_dw; 2037 unsigned buffer_en, i, update_flags = 0; 2038 uint64_t va; 2039 2040 buffer_en = (ctx->num_so_targets >= 1 && t[0] ? 1 : 0) | 2041 (ctx->num_so_targets >= 2 && t[1] ? 2 : 0) | 2042 (ctx->num_so_targets >= 3 && t[2] ? 4 : 0) | 2043 (ctx->num_so_targets >= 4 && t[3] ? 8 : 0); 2044 2045 ctx->num_cs_dw_streamout_end = 2046 12 + /* flush_vgt_streamout */ 2047 util_bitcount(buffer_en) * 8 + 2048 8; 2049 2050 r600_need_cs_space(ctx, 2051 12 + /* flush_vgt_streamout */ 2052 6 + /* enables */ 2053 util_bitcount(buffer_en & ctx->streamout_append_bitmask) * 8 + 2054 util_bitcount(buffer_en & ~ctx->streamout_append_bitmask) * 6 + 2055 (ctx->screen->family > CHIP_R600 && ctx->screen->family < CHIP_RV770 ? 2 : 0) + 2056 ctx->num_cs_dw_streamout_end, TRUE); 2057 2058 if (ctx->screen->chip_class >= EVERGREEN) { 2059 evergreen_flush_vgt_streamout(ctx); 2060 evergreen_set_streamout_enable(ctx, buffer_en); 2061 } else { 2062 r600_flush_vgt_streamout(ctx); 2063 r600_set_streamout_enable(ctx, buffer_en); 2064 } 2065 2066 for (i = 0; i < ctx->num_so_targets; i++) { 2067 if (t[i]) { 2068 t[i]->stride_in_dw = stride_in_dw[i]; 2069 t[i]->so_index = i; 2070 va = r600_resource_va(&ctx->screen->screen, 2071 (void*)t[i]->b.buffer); 2072 2073 update_flags |= SURFACE_BASE_UPDATE_STRMOUT(i); 2074 2075 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_CONTEXT_REG, 3, 0); 2076 ctx->pm4[ctx->pm4_cdwords++] = (R_028AD0_VGT_STRMOUT_BUFFER_SIZE_0 + 2077 16*i - R600_CONTEXT_REG_OFFSET) >> 2; 2078 ctx->pm4[ctx->pm4_cdwords++] = (t[i]->b.buffer_offset + 2079 t[i]->b.buffer_size) >> 2; /* BUFFER_SIZE (in DW) */ 2080 ctx->pm4[ctx->pm4_cdwords++] = stride_in_dw[i]; /* VTX_STRIDE (in DW) */ 2081 ctx->pm4[ctx->pm4_cdwords++] = va >> 8; /* BUFFER_BASE */ 2082 2083 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); 2084 ctx->pm4[ctx->pm4_cdwords++] = 2085 r600_context_bo_reloc(ctx, r600_resource(t[i]->b.buffer), 2086 RADEON_USAGE_WRITE); 2087 2088 if (ctx->streamout_append_bitmask & (1 << i)) { 2089 va = r600_resource_va(&ctx->screen->screen, 2090 (void*)t[i]->filled_size); 2091 /* Append. */ 2092 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0); 2093 ctx->pm4[ctx->pm4_cdwords++] = STRMOUT_SELECT_BUFFER(i) | 2094 STRMOUT_OFFSET_SOURCE(STRMOUT_OFFSET_FROM_MEM); /* control */ 2095 ctx->pm4[ctx->pm4_cdwords++] = 0; /* unused */ 2096 ctx->pm4[ctx->pm4_cdwords++] = 0; /* unused */ 2097 ctx->pm4[ctx->pm4_cdwords++] = va & 0xFFFFFFFFUL; /* src address lo */ 2098 ctx->pm4[ctx->pm4_cdwords++] = (va >> 32UL) & 0xFFUL; /* src address hi */ 2099 2100 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); 2101 ctx->pm4[ctx->pm4_cdwords++] = 2102 r600_context_bo_reloc(ctx, t[i]->filled_size, 2103 RADEON_USAGE_READ); 2104 } else { 2105 /* Start from the beginning. */ 2106 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0); 2107 ctx->pm4[ctx->pm4_cdwords++] = STRMOUT_SELECT_BUFFER(i) | 2108 STRMOUT_OFFSET_SOURCE(STRMOUT_OFFSET_FROM_PACKET); /* control */ 2109 ctx->pm4[ctx->pm4_cdwords++] = 0; /* unused */ 2110 ctx->pm4[ctx->pm4_cdwords++] = 0; /* unused */ 2111 ctx->pm4[ctx->pm4_cdwords++] = t[i]->b.buffer_offset >> 2; /* buffer offset in DW */ 2112 ctx->pm4[ctx->pm4_cdwords++] = 0; /* unused */ 2113 } 2114 } 2115 } 2116 2117 if (ctx->screen->family > CHIP_R600 && ctx->screen->family < CHIP_RV770) { 2118 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SURFACE_BASE_UPDATE, 0, 0); 2119 ctx->pm4[ctx->pm4_cdwords++] = update_flags; 2120 } 2121} 2122 2123void r600_context_streamout_end(struct r600_context *ctx) 2124{ 2125 struct r600_so_target **t = ctx->so_targets; 2126 unsigned i, flush_flags = 0; 2127 uint64_t va; 2128 2129 if (ctx->screen->chip_class >= EVERGREEN) { 2130 evergreen_flush_vgt_streamout(ctx); 2131 } else { 2132 r600_flush_vgt_streamout(ctx); 2133 } 2134 2135 for (i = 0; i < ctx->num_so_targets; i++) { 2136 if (t[i]) { 2137 va = r600_resource_va(&ctx->screen->screen, 2138 (void*)t[i]->filled_size); 2139 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0); 2140 ctx->pm4[ctx->pm4_cdwords++] = STRMOUT_SELECT_BUFFER(i) | 2141 STRMOUT_OFFSET_SOURCE(STRMOUT_OFFSET_NONE) | 2142 STRMOUT_STORE_BUFFER_FILLED_SIZE; /* control */ 2143 ctx->pm4[ctx->pm4_cdwords++] = va & 0xFFFFFFFFUL; /* dst address lo */ 2144 ctx->pm4[ctx->pm4_cdwords++] = (va >> 32UL) & 0xFFUL; /* dst address hi */ 2145 ctx->pm4[ctx->pm4_cdwords++] = 0; /* unused */ 2146 ctx->pm4[ctx->pm4_cdwords++] = 0; /* unused */ 2147 2148 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); 2149 ctx->pm4[ctx->pm4_cdwords++] = 2150 r600_context_bo_reloc(ctx, t[i]->filled_size, 2151 RADEON_USAGE_WRITE); 2152 2153 flush_flags |= S_0085F0_SO0_DEST_BASE_ENA(1) << i; 2154 } 2155 } 2156 2157 if (ctx->screen->chip_class >= EVERGREEN) { 2158 evergreen_set_streamout_enable(ctx, 0); 2159 } else { 2160 r600_set_streamout_enable(ctx, 0); 2161 } 2162 2163 if (ctx->screen->family < CHIP_RV770) { 2164 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, 0); 2165 ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_EVENT) | EVENT_INDEX(0); 2166 } else { 2167 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SURFACE_SYNC, 3, 0); 2168 ctx->pm4[ctx->pm4_cdwords++] = flush_flags; /* CP_COHER_CNTL */ 2169 ctx->pm4[ctx->pm4_cdwords++] = 0xffffffff; /* CP_COHER_SIZE */ 2170 ctx->pm4[ctx->pm4_cdwords++] = 0; /* CP_COHER_BASE */ 2171 ctx->pm4[ctx->pm4_cdwords++] = 0x0000000A; /* POLL_INTERVAL */ 2172 } 2173 2174 ctx->num_cs_dw_streamout_end = 0; 2175 2176 /* XXX print some debug info */ 2177 for (i = 0; i < ctx->num_so_targets; i++) { 2178 if (!t[i]) 2179 continue; 2180 2181 uint32_t *ptr = ctx->ws->buffer_map(t[i]->filled_size->buf, ctx->cs, RADEON_USAGE_READ); 2182 printf("FILLED_SIZE%i: %u\n", i, *ptr); 2183 ctx->ws->buffer_unmap(t[i]->filled_size->buf); 2184 } 2185} 2186 2187void r600_context_draw_opaque_count(struct r600_context *ctx, struct r600_so_target *t) 2188{ 2189 uint64_t va = r600_resource_va(&ctx->screen->screen, 2190 (void*)t->filled_size); 2191 2192 r600_need_cs_space(ctx, 14 + 21, TRUE); 2193 2194 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0); 2195 ctx->pm4[ctx->pm4_cdwords++] = (R_028B28_VGT_STRMOUT_DRAW_OPAQUE_OFFSET - R600_CONTEXT_REG_OFFSET) >> 2; 2196 ctx->pm4[ctx->pm4_cdwords++] = 0; 2197 2198 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0); 2199 ctx->pm4[ctx->pm4_cdwords++] = (R_028B30_VGT_STRMOUT_DRAW_OPAQUE_VERTEX_STRIDE - R600_CONTEXT_REG_OFFSET) >> 2; 2200 ctx->pm4[ctx->pm4_cdwords++] = t->stride_in_dw; 2201 2202 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_COPY_DW, 4, 0); 2203 ctx->pm4[ctx->pm4_cdwords++] = COPY_DW_SRC_IS_MEM | COPY_DW_DST_IS_REG; 2204 ctx->pm4[ctx->pm4_cdwords++] = va & 0xFFFFFFFFUL; /* src address lo */ 2205 ctx->pm4[ctx->pm4_cdwords++] = (va >> 32UL) & 0xFFUL; /* src address hi */ 2206 ctx->pm4[ctx->pm4_cdwords++] = R_028B2C_VGT_STRMOUT_DRAW_OPAQUE_BUFFER_FILLED_SIZE >> 2; /* dst register */ 2207 ctx->pm4[ctx->pm4_cdwords++] = 0; /* unused */ 2208 2209 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); 2210 ctx->pm4[ctx->pm4_cdwords++] = r600_context_bo_reloc(ctx, t->filled_size, 2211 RADEON_USAGE_READ); 2212 2213#if 0 /* I have not found this useful yet. */ 2214 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_COPY_DW, 4, 0); 2215 ctx->pm4[ctx->pm4_cdwords++] = COPY_DW_SRC_IS_REG | COPY_DW_DST_IS_REG; 2216 ctx->pm4[ctx->pm4_cdwords++] = R_028B2C_VGT_STRMOUT_DRAW_OPAQUE_BUFFER_FILLED_SIZE >> 2; /* src register */ 2217 ctx->pm4[ctx->pm4_cdwords++] = 0; /* unused */ 2218 ctx->pm4[ctx->pm4_cdwords++] = R_0085F4_CP_COHER_SIZE >> 2; /* dst register */ 2219 ctx->pm4[ctx->pm4_cdwords++] = 0; /* unused */ 2220 2221 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_CONFIG_REG, 1, 0); 2222 ctx->pm4[ctx->pm4_cdwords++] = (R_0085F0_CP_COHER_CNTL - R600_CONFIG_REG_OFFSET) >> 2; 2223 ctx->pm4[ctx->pm4_cdwords++] = S_0085F0_SO0_DEST_BASE_ENA(1) << t->so_index; 2224 2225 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_CONFIG_REG, 1, 0); 2226 ctx->pm4[ctx->pm4_cdwords++] = (R_0085F8_CP_COHER_BASE - R600_CONFIG_REG_OFFSET) >> 2; 2227 ctx->pm4[ctx->pm4_cdwords++] = t->b.buffer_offset >> 2; 2228 2229 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); 2230 ctx->pm4[ctx->pm4_cdwords++] = r600_context_bo_reloc(ctx, (struct r600_resource*)t->b.buffer, 2231 RADEON_USAGE_WRITE); 2232 2233 ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_WAIT_REG_MEM, 5, 0); 2234 ctx->pm4[ctx->pm4_cdwords++] = WAIT_REG_MEM_EQUAL; /* wait until the register is equal to the reference value */ 2235 ctx->pm4[ctx->pm4_cdwords++] = R_0085FC_CP_COHER_STATUS >> 2; /* register */ 2236 ctx->pm4[ctx->pm4_cdwords++] = 0; 2237 ctx->pm4[ctx->pm4_cdwords++] = 0; /* reference value */ 2238 ctx->pm4[ctx->pm4_cdwords++] = 0xffffffff; /* mask */ 2239 ctx->pm4[ctx->pm4_cdwords++] = 4; /* poll interval */ 2240#endif 2241} 2242