1 /* 2 * Copyright 2010 Jerome Glisse <glisse (at) 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 #include "r600_formats.h" 24 #include "r600_shader.h" 25 #include "evergreend.h" 26 27 #include "pipe/p_shader_tokens.h" 28 #include "util/u_pack_color.h" 29 #include "util/u_memory.h" 30 #include "util/u_framebuffer.h" 31 #include "util/u_dual_blend.h" 32 #include "evergreen_compute.h" 33 #include "util/u_math.h" 34 35 static inline unsigned evergreen_array_mode(unsigned mode) 36 { 37 switch (mode) { 38 default: 39 case RADEON_SURF_MODE_LINEAR_ALIGNED: return V_028C70_ARRAY_LINEAR_ALIGNED; 40 break; 41 case RADEON_SURF_MODE_1D: return V_028C70_ARRAY_1D_TILED_THIN1; 42 break; 43 case RADEON_SURF_MODE_2D: return V_028C70_ARRAY_2D_TILED_THIN1; 44 } 45 } 46 47 static uint32_t eg_num_banks(uint32_t nbanks) 48 { 49 switch (nbanks) { 50 case 2: 51 return 0; 52 case 4: 53 return 1; 54 case 8: 55 default: 56 return 2; 57 case 16: 58 return 3; 59 } 60 } 61 62 63 static unsigned eg_tile_split(unsigned tile_split) 64 { 65 switch (tile_split) { 66 case 64: tile_split = 0; break; 67 case 128: tile_split = 1; break; 68 case 256: tile_split = 2; break; 69 case 512: tile_split = 3; break; 70 default: 71 case 1024: tile_split = 4; break; 72 case 2048: tile_split = 5; break; 73 case 4096: tile_split = 6; break; 74 } 75 return tile_split; 76 } 77 78 static unsigned eg_macro_tile_aspect(unsigned macro_tile_aspect) 79 { 80 switch (macro_tile_aspect) { 81 default: 82 case 1: macro_tile_aspect = 0; break; 83 case 2: macro_tile_aspect = 1; break; 84 case 4: macro_tile_aspect = 2; break; 85 case 8: macro_tile_aspect = 3; break; 86 } 87 return macro_tile_aspect; 88 } 89 90 static unsigned eg_bank_wh(unsigned bankwh) 91 { 92 switch (bankwh) { 93 default: 94 case 1: bankwh = 0; break; 95 case 2: bankwh = 1; break; 96 case 4: bankwh = 2; break; 97 case 8: bankwh = 3; break; 98 } 99 return bankwh; 100 } 101 102 static uint32_t r600_translate_blend_function(int blend_func) 103 { 104 switch (blend_func) { 105 case PIPE_BLEND_ADD: 106 return V_028780_COMB_DST_PLUS_SRC; 107 case PIPE_BLEND_SUBTRACT: 108 return V_028780_COMB_SRC_MINUS_DST; 109 case PIPE_BLEND_REVERSE_SUBTRACT: 110 return V_028780_COMB_DST_MINUS_SRC; 111 case PIPE_BLEND_MIN: 112 return V_028780_COMB_MIN_DST_SRC; 113 case PIPE_BLEND_MAX: 114 return V_028780_COMB_MAX_DST_SRC; 115 default: 116 R600_ERR("Unknown blend function %d\n", blend_func); 117 assert(0); 118 break; 119 } 120 return 0; 121 } 122 123 static uint32_t r600_translate_blend_factor(int blend_fact) 124 { 125 switch (blend_fact) { 126 case PIPE_BLENDFACTOR_ONE: 127 return V_028780_BLEND_ONE; 128 case PIPE_BLENDFACTOR_SRC_COLOR: 129 return V_028780_BLEND_SRC_COLOR; 130 case PIPE_BLENDFACTOR_SRC_ALPHA: 131 return V_028780_BLEND_SRC_ALPHA; 132 case PIPE_BLENDFACTOR_DST_ALPHA: 133 return V_028780_BLEND_DST_ALPHA; 134 case PIPE_BLENDFACTOR_DST_COLOR: 135 return V_028780_BLEND_DST_COLOR; 136 case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: 137 return V_028780_BLEND_SRC_ALPHA_SATURATE; 138 case PIPE_BLENDFACTOR_CONST_COLOR: 139 return V_028780_BLEND_CONST_COLOR; 140 case PIPE_BLENDFACTOR_CONST_ALPHA: 141 return V_028780_BLEND_CONST_ALPHA; 142 case PIPE_BLENDFACTOR_ZERO: 143 return V_028780_BLEND_ZERO; 144 case PIPE_BLENDFACTOR_INV_SRC_COLOR: 145 return V_028780_BLEND_ONE_MINUS_SRC_COLOR; 146 case PIPE_BLENDFACTOR_INV_SRC_ALPHA: 147 return V_028780_BLEND_ONE_MINUS_SRC_ALPHA; 148 case PIPE_BLENDFACTOR_INV_DST_ALPHA: 149 return V_028780_BLEND_ONE_MINUS_DST_ALPHA; 150 case PIPE_BLENDFACTOR_INV_DST_COLOR: 151 return V_028780_BLEND_ONE_MINUS_DST_COLOR; 152 case PIPE_BLENDFACTOR_INV_CONST_COLOR: 153 return V_028780_BLEND_ONE_MINUS_CONST_COLOR; 154 case PIPE_BLENDFACTOR_INV_CONST_ALPHA: 155 return V_028780_BLEND_ONE_MINUS_CONST_ALPHA; 156 case PIPE_BLENDFACTOR_SRC1_COLOR: 157 return V_028780_BLEND_SRC1_COLOR; 158 case PIPE_BLENDFACTOR_SRC1_ALPHA: 159 return V_028780_BLEND_SRC1_ALPHA; 160 case PIPE_BLENDFACTOR_INV_SRC1_COLOR: 161 return V_028780_BLEND_INV_SRC1_COLOR; 162 case PIPE_BLENDFACTOR_INV_SRC1_ALPHA: 163 return V_028780_BLEND_INV_SRC1_ALPHA; 164 default: 165 R600_ERR("Bad blend factor %d not supported!\n", blend_fact); 166 assert(0); 167 break; 168 } 169 return 0; 170 } 171 172 static unsigned r600_tex_dim(unsigned dim, unsigned nr_samples) 173 { 174 switch (dim) { 175 default: 176 case PIPE_TEXTURE_1D: 177 return V_030000_SQ_TEX_DIM_1D; 178 case PIPE_TEXTURE_1D_ARRAY: 179 return V_030000_SQ_TEX_DIM_1D_ARRAY; 180 case PIPE_TEXTURE_2D: 181 case PIPE_TEXTURE_RECT: 182 return nr_samples > 1 ? V_030000_SQ_TEX_DIM_2D_MSAA : 183 V_030000_SQ_TEX_DIM_2D; 184 case PIPE_TEXTURE_2D_ARRAY: 185 return nr_samples > 1 ? V_030000_SQ_TEX_DIM_2D_ARRAY_MSAA : 186 V_030000_SQ_TEX_DIM_2D_ARRAY; 187 case PIPE_TEXTURE_3D: 188 return V_030000_SQ_TEX_DIM_3D; 189 case PIPE_TEXTURE_CUBE: 190 case PIPE_TEXTURE_CUBE_ARRAY: 191 return V_030000_SQ_TEX_DIM_CUBEMAP; 192 } 193 } 194 195 static uint32_t r600_translate_dbformat(enum pipe_format format) 196 { 197 switch (format) { 198 case PIPE_FORMAT_Z16_UNORM: 199 return V_028040_Z_16; 200 case PIPE_FORMAT_Z24X8_UNORM: 201 case PIPE_FORMAT_Z24_UNORM_S8_UINT: 202 case PIPE_FORMAT_X8Z24_UNORM: 203 case PIPE_FORMAT_S8_UINT_Z24_UNORM: 204 return V_028040_Z_24; 205 case PIPE_FORMAT_Z32_FLOAT: 206 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT: 207 return V_028040_Z_32_FLOAT; 208 default: 209 return ~0U; 210 } 211 } 212 213 static bool r600_is_sampler_format_supported(struct pipe_screen *screen, enum pipe_format format) 214 { 215 return r600_translate_texformat(screen, format, NULL, NULL, NULL, 216 FALSE) != ~0U; 217 } 218 219 static bool r600_is_colorbuffer_format_supported(enum chip_class chip, enum pipe_format format) 220 { 221 return r600_translate_colorformat(chip, format, FALSE) != ~0U && 222 r600_translate_colorswap(format, FALSE) != ~0U; 223 } 224 225 static bool r600_is_zs_format_supported(enum pipe_format format) 226 { 227 return r600_translate_dbformat(format) != ~0U; 228 } 229 230 boolean evergreen_is_format_supported(struct pipe_screen *screen, 231 enum pipe_format format, 232 enum pipe_texture_target target, 233 unsigned sample_count, 234 unsigned usage) 235 { 236 struct r600_screen *rscreen = (struct r600_screen*)screen; 237 unsigned retval = 0; 238 239 if (target >= PIPE_MAX_TEXTURE_TYPES) { 240 R600_ERR("r600: unsupported texture type %d\n", target); 241 return FALSE; 242 } 243 244 if (!util_format_is_supported(format, usage)) 245 return FALSE; 246 247 if (sample_count > 1) { 248 if (!rscreen->has_msaa) 249 return FALSE; 250 251 switch (sample_count) { 252 case 2: 253 case 4: 254 case 8: 255 break; 256 default: 257 return FALSE; 258 } 259 } 260 261 if (usage & PIPE_BIND_SAMPLER_VIEW) { 262 if (target == PIPE_BUFFER) { 263 if (r600_is_vertex_format_supported(format)) 264 retval |= PIPE_BIND_SAMPLER_VIEW; 265 } else { 266 if (r600_is_sampler_format_supported(screen, format)) 267 retval |= PIPE_BIND_SAMPLER_VIEW; 268 } 269 } 270 271 if ((usage & (PIPE_BIND_RENDER_TARGET | 272 PIPE_BIND_DISPLAY_TARGET | 273 PIPE_BIND_SCANOUT | 274 PIPE_BIND_SHARED | 275 PIPE_BIND_BLENDABLE)) && 276 r600_is_colorbuffer_format_supported(rscreen->b.chip_class, format)) { 277 retval |= usage & 278 (PIPE_BIND_RENDER_TARGET | 279 PIPE_BIND_DISPLAY_TARGET | 280 PIPE_BIND_SCANOUT | 281 PIPE_BIND_SHARED); 282 if (!util_format_is_pure_integer(format) && 283 !util_format_is_depth_or_stencil(format)) 284 retval |= usage & PIPE_BIND_BLENDABLE; 285 } 286 287 if ((usage & PIPE_BIND_DEPTH_STENCIL) && 288 r600_is_zs_format_supported(format)) { 289 retval |= PIPE_BIND_DEPTH_STENCIL; 290 } 291 292 if ((usage & PIPE_BIND_VERTEX_BUFFER) && 293 r600_is_vertex_format_supported(format)) { 294 retval |= PIPE_BIND_VERTEX_BUFFER; 295 } 296 297 if ((usage & PIPE_BIND_LINEAR) && 298 !util_format_is_compressed(format) && 299 !(usage & PIPE_BIND_DEPTH_STENCIL)) 300 retval |= PIPE_BIND_LINEAR; 301 302 return retval == usage; 303 } 304 305 static void *evergreen_create_blend_state_mode(struct pipe_context *ctx, 306 const struct pipe_blend_state *state, int mode) 307 { 308 uint32_t color_control = 0, target_mask = 0; 309 struct r600_blend_state *blend = CALLOC_STRUCT(r600_blend_state); 310 311 if (!blend) { 312 return NULL; 313 } 314 315 r600_init_command_buffer(&blend->buffer, 20); 316 r600_init_command_buffer(&blend->buffer_no_blend, 20); 317 318 if (state->logicop_enable) { 319 color_control |= (state->logicop_func << 16) | (state->logicop_func << 20); 320 } else { 321 color_control |= (0xcc << 16); 322 } 323 /* we pretend 8 buffer are used, CB_SHADER_MASK will disable unused one */ 324 if (state->independent_blend_enable) { 325 for (int i = 0; i < 8; i++) { 326 target_mask |= (state->rt[i].colormask << (4 * i)); 327 } 328 } else { 329 for (int i = 0; i < 8; i++) { 330 target_mask |= (state->rt[0].colormask << (4 * i)); 331 } 332 } 333 334 /* only have dual source on MRT0 */ 335 blend->dual_src_blend = util_blend_state_is_dual(state, 0); 336 blend->cb_target_mask = target_mask; 337 blend->alpha_to_one = state->alpha_to_one; 338 339 if (target_mask) 340 color_control |= S_028808_MODE(mode); 341 else 342 color_control |= S_028808_MODE(V_028808_CB_DISABLE); 343 344 345 r600_store_context_reg(&blend->buffer, R_028808_CB_COLOR_CONTROL, color_control); 346 r600_store_context_reg(&blend->buffer, R_028B70_DB_ALPHA_TO_MASK, 347 S_028B70_ALPHA_TO_MASK_ENABLE(state->alpha_to_coverage) | 348 S_028B70_ALPHA_TO_MASK_OFFSET0(2) | 349 S_028B70_ALPHA_TO_MASK_OFFSET1(2) | 350 S_028B70_ALPHA_TO_MASK_OFFSET2(2) | 351 S_028B70_ALPHA_TO_MASK_OFFSET3(2)); 352 r600_store_context_reg_seq(&blend->buffer, R_028780_CB_BLEND0_CONTROL, 8); 353 354 /* Copy over the dwords set so far into buffer_no_blend. 355 * Only the CB_BLENDi_CONTROL registers must be set after this. */ 356 memcpy(blend->buffer_no_blend.buf, blend->buffer.buf, blend->buffer.num_dw * 4); 357 blend->buffer_no_blend.num_dw = blend->buffer.num_dw; 358 359 for (int i = 0; i < 8; i++) { 360 /* state->rt entries > 0 only written if independent blending */ 361 const int j = state->independent_blend_enable ? i : 0; 362 363 unsigned eqRGB = state->rt[j].rgb_func; 364 unsigned srcRGB = state->rt[j].rgb_src_factor; 365 unsigned dstRGB = state->rt[j].rgb_dst_factor; 366 unsigned eqA = state->rt[j].alpha_func; 367 unsigned srcA = state->rt[j].alpha_src_factor; 368 unsigned dstA = state->rt[j].alpha_dst_factor; 369 uint32_t bc = 0; 370 371 r600_store_value(&blend->buffer_no_blend, 0); 372 373 if (!state->rt[j].blend_enable) { 374 r600_store_value(&blend->buffer, 0); 375 continue; 376 } 377 378 bc |= S_028780_BLEND_CONTROL_ENABLE(1); 379 bc |= S_028780_COLOR_COMB_FCN(r600_translate_blend_function(eqRGB)); 380 bc |= S_028780_COLOR_SRCBLEND(r600_translate_blend_factor(srcRGB)); 381 bc |= S_028780_COLOR_DESTBLEND(r600_translate_blend_factor(dstRGB)); 382 383 if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) { 384 bc |= S_028780_SEPARATE_ALPHA_BLEND(1); 385 bc |= S_028780_ALPHA_COMB_FCN(r600_translate_blend_function(eqA)); 386 bc |= S_028780_ALPHA_SRCBLEND(r600_translate_blend_factor(srcA)); 387 bc |= S_028780_ALPHA_DESTBLEND(r600_translate_blend_factor(dstA)); 388 } 389 r600_store_value(&blend->buffer, bc); 390 } 391 return blend; 392 } 393 394 static void *evergreen_create_blend_state(struct pipe_context *ctx, 395 const struct pipe_blend_state *state) 396 { 397 398 return evergreen_create_blend_state_mode(ctx, state, V_028808_CB_NORMAL); 399 } 400 401 static void *evergreen_create_dsa_state(struct pipe_context *ctx, 402 const struct pipe_depth_stencil_alpha_state *state) 403 { 404 unsigned db_depth_control, alpha_test_control, alpha_ref; 405 struct r600_dsa_state *dsa = CALLOC_STRUCT(r600_dsa_state); 406 407 if (!dsa) { 408 return NULL; 409 } 410 411 r600_init_command_buffer(&dsa->buffer, 3); 412 413 dsa->valuemask[0] = state->stencil[0].valuemask; 414 dsa->valuemask[1] = state->stencil[1].valuemask; 415 dsa->writemask[0] = state->stencil[0].writemask; 416 dsa->writemask[1] = state->stencil[1].writemask; 417 dsa->zwritemask = state->depth.writemask; 418 419 db_depth_control = S_028800_Z_ENABLE(state->depth.enabled) | 420 S_028800_Z_WRITE_ENABLE(state->depth.writemask) | 421 S_028800_ZFUNC(state->depth.func); 422 423 /* stencil */ 424 if (state->stencil[0].enabled) { 425 db_depth_control |= S_028800_STENCIL_ENABLE(1); 426 db_depth_control |= S_028800_STENCILFUNC(state->stencil[0].func); /* translates straight */ 427 db_depth_control |= S_028800_STENCILFAIL(r600_translate_stencil_op(state->stencil[0].fail_op)); 428 db_depth_control |= S_028800_STENCILZPASS(r600_translate_stencil_op(state->stencil[0].zpass_op)); 429 db_depth_control |= S_028800_STENCILZFAIL(r600_translate_stencil_op(state->stencil[0].zfail_op)); 430 431 if (state->stencil[1].enabled) { 432 db_depth_control |= S_028800_BACKFACE_ENABLE(1); 433 db_depth_control |= S_028800_STENCILFUNC_BF(state->stencil[1].func); /* translates straight */ 434 db_depth_control |= S_028800_STENCILFAIL_BF(r600_translate_stencil_op(state->stencil[1].fail_op)); 435 db_depth_control |= S_028800_STENCILZPASS_BF(r600_translate_stencil_op(state->stencil[1].zpass_op)); 436 db_depth_control |= S_028800_STENCILZFAIL_BF(r600_translate_stencil_op(state->stencil[1].zfail_op)); 437 } 438 } 439 440 /* alpha */ 441 alpha_test_control = 0; 442 alpha_ref = 0; 443 if (state->alpha.enabled) { 444 alpha_test_control = S_028410_ALPHA_FUNC(state->alpha.func); 445 alpha_test_control |= S_028410_ALPHA_TEST_ENABLE(1); 446 alpha_ref = fui(state->alpha.ref_value); 447 } 448 dsa->sx_alpha_test_control = alpha_test_control & 0xff; 449 dsa->alpha_ref = alpha_ref; 450 451 /* misc */ 452 r600_store_context_reg(&dsa->buffer, R_028800_DB_DEPTH_CONTROL, db_depth_control); 453 return dsa; 454 } 455 456 static void *evergreen_create_rs_state(struct pipe_context *ctx, 457 const struct pipe_rasterizer_state *state) 458 { 459 struct r600_context *rctx = (struct r600_context *)ctx; 460 unsigned tmp, spi_interp; 461 float psize_min, psize_max; 462 struct r600_rasterizer_state *rs = CALLOC_STRUCT(r600_rasterizer_state); 463 464 if (!rs) { 465 return NULL; 466 } 467 468 r600_init_command_buffer(&rs->buffer, 30); 469 470 rs->scissor_enable = state->scissor; 471 rs->clip_halfz = state->clip_halfz; 472 rs->flatshade = state->flatshade; 473 rs->sprite_coord_enable = state->sprite_coord_enable; 474 rs->two_side = state->light_twoside; 475 rs->clip_plane_enable = state->clip_plane_enable; 476 rs->pa_sc_line_stipple = state->line_stipple_enable ? 477 S_028A0C_LINE_PATTERN(state->line_stipple_pattern) | 478 S_028A0C_REPEAT_COUNT(state->line_stipple_factor) : 0; 479 rs->pa_cl_clip_cntl = 480 S_028810_DX_CLIP_SPACE_DEF(state->clip_halfz) | 481 S_028810_ZCLIP_NEAR_DISABLE(!state->depth_clip) | 482 S_028810_ZCLIP_FAR_DISABLE(!state->depth_clip) | 483 S_028810_DX_LINEAR_ATTR_CLIP_ENA(1) | 484 S_028810_DX_RASTERIZATION_KILL(state->rasterizer_discard); 485 rs->multisample_enable = state->multisample; 486 487 /* offset */ 488 rs->offset_units = state->offset_units; 489 rs->offset_scale = state->offset_scale * 16.0f; 490 rs->offset_enable = state->offset_point || state->offset_line || state->offset_tri; 491 rs->offset_units_unscaled = state->offset_units_unscaled; 492 493 if (state->point_size_per_vertex) { 494 psize_min = util_get_min_point_size(state); 495 psize_max = 8192; 496 } else { 497 /* Force the point size to be as if the vertex output was disabled. */ 498 psize_min = state->point_size; 499 psize_max = state->point_size; 500 } 501 502 spi_interp = S_0286D4_FLAT_SHADE_ENA(1); 503 if (state->sprite_coord_enable) { 504 spi_interp |= S_0286D4_PNT_SPRITE_ENA(1) | 505 S_0286D4_PNT_SPRITE_OVRD_X(2) | 506 S_0286D4_PNT_SPRITE_OVRD_Y(3) | 507 S_0286D4_PNT_SPRITE_OVRD_Z(0) | 508 S_0286D4_PNT_SPRITE_OVRD_W(1); 509 if (state->sprite_coord_mode != PIPE_SPRITE_COORD_UPPER_LEFT) { 510 spi_interp |= S_0286D4_PNT_SPRITE_TOP_1(1); 511 } 512 } 513 514 r600_store_context_reg_seq(&rs->buffer, R_028A00_PA_SU_POINT_SIZE, 3); 515 /* point size 12.4 fixed point (divide by two, because 0.5 = 1 pixel) */ 516 tmp = r600_pack_float_12p4(state->point_size/2); 517 r600_store_value(&rs->buffer, /* R_028A00_PA_SU_POINT_SIZE */ 518 S_028A00_HEIGHT(tmp) | S_028A00_WIDTH(tmp)); 519 r600_store_value(&rs->buffer, /* R_028A04_PA_SU_POINT_MINMAX */ 520 S_028A04_MIN_SIZE(r600_pack_float_12p4(psize_min/2)) | 521 S_028A04_MAX_SIZE(r600_pack_float_12p4(psize_max/2))); 522 r600_store_value(&rs->buffer, /* R_028A08_PA_SU_LINE_CNTL */ 523 S_028A08_WIDTH((unsigned)(state->line_width * 8))); 524 525 r600_store_context_reg(&rs->buffer, R_0286D4_SPI_INTERP_CONTROL_0, spi_interp); 526 r600_store_context_reg(&rs->buffer, R_028A48_PA_SC_MODE_CNTL_0, 527 S_028A48_MSAA_ENABLE(state->multisample) | 528 S_028A48_VPORT_SCISSOR_ENABLE(1) | 529 S_028A48_LINE_STIPPLE_ENABLE(state->line_stipple_enable)); 530 531 if (rctx->b.chip_class == CAYMAN) { 532 r600_store_context_reg(&rs->buffer, CM_R_028BE4_PA_SU_VTX_CNTL, 533 S_028C08_PIX_CENTER_HALF(state->half_pixel_center) | 534 S_028C08_QUANT_MODE(V_028C08_X_1_256TH)); 535 } else { 536 r600_store_context_reg(&rs->buffer, R_028C08_PA_SU_VTX_CNTL, 537 S_028C08_PIX_CENTER_HALF(state->half_pixel_center) | 538 S_028C08_QUANT_MODE(V_028C08_X_1_256TH)); 539 } 540 541 r600_store_context_reg(&rs->buffer, R_028B7C_PA_SU_POLY_OFFSET_CLAMP, fui(state->offset_clamp)); 542 r600_store_context_reg(&rs->buffer, R_028814_PA_SU_SC_MODE_CNTL, 543 S_028814_PROVOKING_VTX_LAST(!state->flatshade_first) | 544 S_028814_CULL_FRONT((state->cull_face & PIPE_FACE_FRONT) ? 1 : 0) | 545 S_028814_CULL_BACK((state->cull_face & PIPE_FACE_BACK) ? 1 : 0) | 546 S_028814_FACE(!state->front_ccw) | 547 S_028814_POLY_OFFSET_FRONT_ENABLE(util_get_offset(state, state->fill_front)) | 548 S_028814_POLY_OFFSET_BACK_ENABLE(util_get_offset(state, state->fill_back)) | 549 S_028814_POLY_OFFSET_PARA_ENABLE(state->offset_point || state->offset_line) | 550 S_028814_POLY_MODE(state->fill_front != PIPE_POLYGON_MODE_FILL || 551 state->fill_back != PIPE_POLYGON_MODE_FILL) | 552 S_028814_POLYMODE_FRONT_PTYPE(r600_translate_fill(state->fill_front)) | 553 S_028814_POLYMODE_BACK_PTYPE(r600_translate_fill(state->fill_back))); 554 return rs; 555 } 556 557 static void *evergreen_create_sampler_state(struct pipe_context *ctx, 558 const struct pipe_sampler_state *state) 559 { 560 struct r600_common_screen *rscreen = (struct r600_common_screen*)ctx->screen; 561 struct r600_pipe_sampler_state *ss = CALLOC_STRUCT(r600_pipe_sampler_state); 562 unsigned max_aniso = rscreen->force_aniso >= 0 ? rscreen->force_aniso 563 : state->max_anisotropy; 564 unsigned max_aniso_ratio = r600_tex_aniso_filter(max_aniso); 565 566 if (!ss) { 567 return NULL; 568 } 569 570 ss->border_color_use = sampler_state_needs_border_color(state); 571 572 /* R_03C000_SQ_TEX_SAMPLER_WORD0_0 */ 573 ss->tex_sampler_words[0] = 574 S_03C000_CLAMP_X(r600_tex_wrap(state->wrap_s)) | 575 S_03C000_CLAMP_Y(r600_tex_wrap(state->wrap_t)) | 576 S_03C000_CLAMP_Z(r600_tex_wrap(state->wrap_r)) | 577 S_03C000_XY_MAG_FILTER(eg_tex_filter(state->mag_img_filter, max_aniso)) | 578 S_03C000_XY_MIN_FILTER(eg_tex_filter(state->min_img_filter, max_aniso)) | 579 S_03C000_MIP_FILTER(r600_tex_mipfilter(state->min_mip_filter)) | 580 S_03C000_MAX_ANISO_RATIO(max_aniso_ratio) | 581 S_03C000_DEPTH_COMPARE_FUNCTION(r600_tex_compare(state->compare_func)) | 582 S_03C000_BORDER_COLOR_TYPE(ss->border_color_use ? V_03C000_SQ_TEX_BORDER_COLOR_REGISTER : 0); 583 /* R_03C004_SQ_TEX_SAMPLER_WORD1_0 */ 584 ss->tex_sampler_words[1] = 585 S_03C004_MIN_LOD(S_FIXED(CLAMP(state->min_lod, 0, 15), 8)) | 586 S_03C004_MAX_LOD(S_FIXED(CLAMP(state->max_lod, 0, 15), 8)); 587 /* R_03C008_SQ_TEX_SAMPLER_WORD2_0 */ 588 ss->tex_sampler_words[2] = 589 S_03C008_LOD_BIAS(S_FIXED(CLAMP(state->lod_bias, -16, 16), 8)) | 590 (state->seamless_cube_map ? 0 : S_03C008_DISABLE_CUBE_WRAP(1)) | 591 S_03C008_TYPE(1); 592 593 if (ss->border_color_use) { 594 memcpy(&ss->border_color, &state->border_color, sizeof(state->border_color)); 595 } 596 return ss; 597 } 598 599 static struct pipe_sampler_view * 600 texture_buffer_sampler_view(struct r600_context *rctx, 601 struct r600_pipe_sampler_view *view, 602 unsigned width0, unsigned height0) 603 604 { 605 struct r600_texture *tmp = (struct r600_texture*)view->base.texture; 606 uint64_t va; 607 int stride = util_format_get_blocksize(view->base.format); 608 unsigned format, num_format, format_comp, endian; 609 unsigned swizzle_res; 610 unsigned char swizzle[4]; 611 const struct util_format_description *desc; 612 unsigned offset = view->base.u.buf.offset; 613 unsigned size = view->base.u.buf.size; 614 615 swizzle[0] = view->base.swizzle_r; 616 swizzle[1] = view->base.swizzle_g; 617 swizzle[2] = view->base.swizzle_b; 618 swizzle[3] = view->base.swizzle_a; 619 620 r600_vertex_data_type(view->base.format, 621 &format, &num_format, &format_comp, 622 &endian); 623 624 desc = util_format_description(view->base.format); 625 626 swizzle_res = r600_get_swizzle_combined(desc->swizzle, swizzle, TRUE); 627 628 va = tmp->resource.gpu_address + offset; 629 view->tex_resource = &tmp->resource; 630 631 view->skip_mip_address_reloc = true; 632 view->tex_resource_words[0] = va; 633 view->tex_resource_words[1] = size - 1; 634 view->tex_resource_words[2] = S_030008_BASE_ADDRESS_HI(va >> 32UL) | 635 S_030008_STRIDE(stride) | 636 S_030008_DATA_FORMAT(format) | 637 S_030008_NUM_FORMAT_ALL(num_format) | 638 S_030008_FORMAT_COMP_ALL(format_comp) | 639 S_030008_ENDIAN_SWAP(endian); 640 view->tex_resource_words[3] = swizzle_res; 641 /* 642 * in theory dword 4 is for number of elements, for use with resinfo, 643 * but it seems to utterly fail to work, the amd gpu shader analyser 644 * uses a const buffer to store the element sizes for buffer txq 645 */ 646 view->tex_resource_words[4] = 0; 647 view->tex_resource_words[5] = view->tex_resource_words[6] = 0; 648 view->tex_resource_words[7] = S_03001C_TYPE(V_03001C_SQ_TEX_VTX_VALID_BUFFER); 649 650 if (tmp->resource.gpu_address) 651 LIST_ADDTAIL(&view->list, &rctx->texture_buffers); 652 return &view->base; 653 } 654 655 struct pipe_sampler_view * 656 evergreen_create_sampler_view_custom(struct pipe_context *ctx, 657 struct pipe_resource *texture, 658 const struct pipe_sampler_view *state, 659 unsigned width0, unsigned height0, 660 unsigned force_level) 661 { 662 struct r600_context *rctx = (struct r600_context*)ctx; 663 struct r600_screen *rscreen = (struct r600_screen*)ctx->screen; 664 struct r600_pipe_sampler_view *view = CALLOC_STRUCT(r600_pipe_sampler_view); 665 struct r600_texture *tmp = (struct r600_texture*)texture; 666 unsigned format, endian; 667 uint32_t word4 = 0, yuv_format = 0, pitch = 0; 668 unsigned char swizzle[4], array_mode = 0, non_disp_tiling = 0; 669 unsigned height, depth, width; 670 unsigned macro_aspect, tile_split, bankh, bankw, nbanks, fmask_bankh; 671 enum pipe_format pipe_format = state->format; 672 struct radeon_surf_level *surflevel; 673 unsigned base_level, first_level, last_level; 674 unsigned dim, last_layer; 675 uint64_t va; 676 bool do_endian_swap = FALSE; 677 678 if (!view) 679 return NULL; 680 681 /* initialize base object */ 682 view->base = *state; 683 view->base.texture = NULL; 684 pipe_reference(NULL, &texture->reference); 685 view->base.texture = texture; 686 view->base.reference.count = 1; 687 view->base.context = ctx; 688 689 if (state->target == PIPE_BUFFER) 690 return texture_buffer_sampler_view(rctx, view, width0, height0); 691 692 swizzle[0] = state->swizzle_r; 693 swizzle[1] = state->swizzle_g; 694 swizzle[2] = state->swizzle_b; 695 swizzle[3] = state->swizzle_a; 696 697 tile_split = tmp->surface.tile_split; 698 surflevel = tmp->surface.level; 699 700 /* Texturing with separate depth and stencil. */ 701 if (tmp->db_compatible) { 702 switch (pipe_format) { 703 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT: 704 pipe_format = PIPE_FORMAT_Z32_FLOAT; 705 break; 706 case PIPE_FORMAT_X8Z24_UNORM: 707 case PIPE_FORMAT_S8_UINT_Z24_UNORM: 708 /* Z24 is always stored like this for DB 709 * compatibility. 710 */ 711 pipe_format = PIPE_FORMAT_Z24X8_UNORM; 712 break; 713 case PIPE_FORMAT_X24S8_UINT: 714 case PIPE_FORMAT_S8X24_UINT: 715 case PIPE_FORMAT_X32_S8X24_UINT: 716 pipe_format = PIPE_FORMAT_S8_UINT; 717 tile_split = tmp->surface.stencil_tile_split; 718 surflevel = tmp->surface.stencil_level; 719 break; 720 default:; 721 } 722 } 723 724 if (R600_BIG_ENDIAN) 725 do_endian_swap = !tmp->db_compatible; 726 727 format = r600_translate_texformat(ctx->screen, pipe_format, 728 swizzle, 729 &word4, &yuv_format, do_endian_swap); 730 assert(format != ~0); 731 if (format == ~0) { 732 FREE(view); 733 return NULL; 734 } 735 736 endian = r600_colorformat_endian_swap(format, do_endian_swap); 737 738 base_level = 0; 739 first_level = state->u.tex.first_level; 740 last_level = state->u.tex.last_level; 741 width = width0; 742 height = height0; 743 depth = texture->depth0; 744 745 if (force_level) { 746 base_level = force_level; 747 first_level = 0; 748 last_level = 0; 749 width = u_minify(width, force_level); 750 height = u_minify(height, force_level); 751 depth = u_minify(depth, force_level); 752 } 753 754 pitch = surflevel[base_level].nblk_x * util_format_get_blockwidth(pipe_format); 755 non_disp_tiling = tmp->non_disp_tiling; 756 757 switch (surflevel[base_level].mode) { 758 default: 759 case RADEON_SURF_MODE_LINEAR_ALIGNED: 760 array_mode = V_028C70_ARRAY_LINEAR_ALIGNED; 761 break; 762 case RADEON_SURF_MODE_2D: 763 array_mode = V_028C70_ARRAY_2D_TILED_THIN1; 764 break; 765 case RADEON_SURF_MODE_1D: 766 array_mode = V_028C70_ARRAY_1D_TILED_THIN1; 767 break; 768 } 769 macro_aspect = tmp->surface.mtilea; 770 bankw = tmp->surface.bankw; 771 bankh = tmp->surface.bankh; 772 tile_split = eg_tile_split(tile_split); 773 macro_aspect = eg_macro_tile_aspect(macro_aspect); 774 bankw = eg_bank_wh(bankw); 775 bankh = eg_bank_wh(bankh); 776 fmask_bankh = eg_bank_wh(tmp->fmask.bank_height); 777 778 /* 128 bit formats require tile type = 1 */ 779 if (rscreen->b.chip_class == CAYMAN) { 780 if (util_format_get_blocksize(pipe_format) >= 16) 781 non_disp_tiling = 1; 782 } 783 nbanks = eg_num_banks(rscreen->b.info.r600_num_banks); 784 785 if (state->target == PIPE_TEXTURE_1D_ARRAY) { 786 height = 1; 787 depth = texture->array_size; 788 } else if (state->target == PIPE_TEXTURE_2D_ARRAY) { 789 depth = texture->array_size; 790 } else if (state->target == PIPE_TEXTURE_CUBE_ARRAY) 791 depth = texture->array_size / 6; 792 793 va = tmp->resource.gpu_address; 794 795 if (state->format == PIPE_FORMAT_X24S8_UINT || 796 state->format == PIPE_FORMAT_S8X24_UINT || 797 state->format == PIPE_FORMAT_X32_S8X24_UINT || 798 state->format == PIPE_FORMAT_S8_UINT) 799 view->is_stencil_sampler = true; 800 801 view->tex_resource = &tmp->resource; 802 803 /* array type views and views into array types need to use layer offset */ 804 dim = state->target; 805 if (state->target != PIPE_TEXTURE_CUBE) 806 dim = MAX2(state->target, texture->target); 807 808 view->tex_resource_words[0] = (S_030000_DIM(r600_tex_dim(dim, texture->nr_samples)) | 809 S_030000_PITCH((pitch / 8) - 1) | 810 S_030000_TEX_WIDTH(width - 1)); 811 if (rscreen->b.chip_class == CAYMAN) 812 view->tex_resource_words[0] |= CM_S_030000_NON_DISP_TILING_ORDER(non_disp_tiling); 813 else 814 view->tex_resource_words[0] |= S_030000_NON_DISP_TILING_ORDER(non_disp_tiling); 815 view->tex_resource_words[1] = (S_030004_TEX_HEIGHT(height - 1) | 816 S_030004_TEX_DEPTH(depth - 1) | 817 S_030004_ARRAY_MODE(array_mode)); 818 view->tex_resource_words[2] = (surflevel[base_level].offset + va) >> 8; 819 820 /* TEX_RESOURCE_WORD3.MIP_ADDRESS */ 821 if (texture->nr_samples > 1 && rscreen->has_compressed_msaa_texturing) { 822 if (tmp->is_depth) { 823 /* disable FMASK (0 = disabled) */ 824 view->tex_resource_words[3] = 0; 825 view->skip_mip_address_reloc = true; 826 } else { 827 /* FMASK should be in MIP_ADDRESS for multisample textures */ 828 view->tex_resource_words[3] = (tmp->fmask.offset + va) >> 8; 829 } 830 } else if (last_level && texture->nr_samples <= 1) { 831 view->tex_resource_words[3] = (surflevel[1].offset + va) >> 8; 832 } else { 833 view->tex_resource_words[3] = (surflevel[base_level].offset + va) >> 8; 834 } 835 836 last_layer = state->u.tex.last_layer; 837 if (state->target != texture->target && depth == 1) { 838 last_layer = state->u.tex.first_layer; 839 } 840 view->tex_resource_words[4] = (word4 | 841 S_030010_ENDIAN_SWAP(endian)); 842 view->tex_resource_words[5] = S_030014_BASE_ARRAY(state->u.tex.first_layer) | 843 S_030014_LAST_ARRAY(last_layer); 844 view->tex_resource_words[6] = S_030018_TILE_SPLIT(tile_split); 845 846 if (texture->nr_samples > 1) { 847 unsigned log_samples = util_logbase2(texture->nr_samples); 848 if (rscreen->b.chip_class == CAYMAN) { 849 view->tex_resource_words[4] |= S_030010_LOG2_NUM_FRAGMENTS(log_samples); 850 } 851 /* LAST_LEVEL holds log2(nr_samples) for multisample textures */ 852 view->tex_resource_words[5] |= S_030014_LAST_LEVEL(log_samples); 853 view->tex_resource_words[6] |= S_030018_FMASK_BANK_HEIGHT(fmask_bankh); 854 } else { 855 bool no_mip = first_level == last_level; 856 857 view->tex_resource_words[4] |= S_030010_BASE_LEVEL(first_level); 858 view->tex_resource_words[5] |= S_030014_LAST_LEVEL(last_level); 859 /* aniso max 16 samples */ 860 view->tex_resource_words[6] |= S_030018_MAX_ANISO_RATIO(no_mip ? 0 : 4); 861 } 862 863 view->tex_resource_words[7] = S_03001C_DATA_FORMAT(format) | 864 S_03001C_TYPE(V_03001C_SQ_TEX_VTX_VALID_TEXTURE) | 865 S_03001C_BANK_WIDTH(bankw) | 866 S_03001C_BANK_HEIGHT(bankh) | 867 S_03001C_MACRO_TILE_ASPECT(macro_aspect) | 868 S_03001C_NUM_BANKS(nbanks) | 869 S_03001C_DEPTH_SAMPLE_ORDER(tmp->db_compatible); 870 return &view->base; 871 } 872 873 static struct pipe_sampler_view * 874 evergreen_create_sampler_view(struct pipe_context *ctx, 875 struct pipe_resource *tex, 876 const struct pipe_sampler_view *state) 877 { 878 return evergreen_create_sampler_view_custom(ctx, tex, state, 879 tex->width0, tex->height0, 0); 880 } 881 882 static void evergreen_emit_config_state(struct r600_context *rctx, struct r600_atom *atom) 883 { 884 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 885 struct r600_config_state *a = (struct r600_config_state*)atom; 886 887 radeon_set_config_reg_seq(cs, R_008C04_SQ_GPR_RESOURCE_MGMT_1, 3); 888 if (a->dyn_gpr_enabled) { 889 radeon_emit(cs, S_008C04_NUM_CLAUSE_TEMP_GPRS(rctx->r6xx_num_clause_temp_gprs)); 890 radeon_emit(cs, 0); 891 radeon_emit(cs, 0); 892 } else { 893 radeon_emit(cs, a->sq_gpr_resource_mgmt_1); 894 radeon_emit(cs, a->sq_gpr_resource_mgmt_2); 895 radeon_emit(cs, a->sq_gpr_resource_mgmt_3); 896 } 897 radeon_set_config_reg(cs, R_008D8C_SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, (a->dyn_gpr_enabled << 8)); 898 if (a->dyn_gpr_enabled) { 899 radeon_set_context_reg(cs, R_028838_SQ_DYN_GPR_RESOURCE_LIMIT_1, 900 S_028838_PS_GPRS(0x1e) | 901 S_028838_VS_GPRS(0x1e) | 902 S_028838_GS_GPRS(0x1e) | 903 S_028838_ES_GPRS(0x1e) | 904 S_028838_HS_GPRS(0x1e) | 905 S_028838_LS_GPRS(0x1e)); /* workaround for hw issues with dyn gpr - must set all limits to 240 instead of 0, 0x1e == 240 / 8*/ 906 } 907 } 908 909 static void evergreen_emit_clip_state(struct r600_context *rctx, struct r600_atom *atom) 910 { 911 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 912 struct pipe_clip_state *state = &rctx->clip_state.state; 913 914 radeon_set_context_reg_seq(cs, R_0285BC_PA_CL_UCP0_X, 6*4); 915 radeon_emit_array(cs, (unsigned*)state, 6*4); 916 } 917 918 static void evergreen_set_polygon_stipple(struct pipe_context *ctx, 919 const struct pipe_poly_stipple *state) 920 { 921 } 922 923 static void evergreen_get_scissor_rect(struct r600_context *rctx, 924 unsigned tl_x, unsigned tl_y, unsigned br_x, unsigned br_y, 925 uint32_t *tl, uint32_t *br) 926 { 927 struct pipe_scissor_state scissor = {tl_x, tl_y, br_x, br_y}; 928 929 evergreen_apply_scissor_bug_workaround(&rctx->b, &scissor); 930 931 *tl = S_028240_TL_X(scissor.minx) | S_028240_TL_Y(scissor.miny); 932 *br = S_028244_BR_X(scissor.maxx) | S_028244_BR_Y(scissor.maxy); 933 } 934 935 /** 936 * This function intializes the CB* register values for RATs. It is meant 937 * to be used for 1D aligned buffers that do not have an associated 938 * radeon_surf. 939 */ 940 void evergreen_init_color_surface_rat(struct r600_context *rctx, 941 struct r600_surface *surf) 942 { 943 struct pipe_resource *pipe_buffer = surf->base.texture; 944 unsigned format = r600_translate_colorformat(rctx->b.chip_class, 945 surf->base.format, FALSE); 946 unsigned endian = r600_colorformat_endian_swap(format, FALSE); 947 unsigned swap = r600_translate_colorswap(surf->base.format, FALSE); 948 unsigned block_size = 949 align(util_format_get_blocksize(pipe_buffer->format), 4); 950 unsigned pitch_alignment = 951 MAX2(64, rctx->screen->b.info.pipe_interleave_bytes / block_size); 952 unsigned pitch = align(pipe_buffer->width0, pitch_alignment); 953 954 surf->cb_color_base = r600_resource(pipe_buffer)->gpu_address >> 8; 955 956 surf->cb_color_pitch = (pitch / 8) - 1; 957 958 surf->cb_color_slice = 0; 959 960 surf->cb_color_view = 0; 961 962 surf->cb_color_info = 963 S_028C70_ENDIAN(endian) 964 | S_028C70_FORMAT(format) 965 | S_028C70_ARRAY_MODE(V_028C70_ARRAY_LINEAR_ALIGNED) 966 | S_028C70_NUMBER_TYPE(V_028C70_NUMBER_UINT) 967 | S_028C70_COMP_SWAP(swap) 968 | S_028C70_BLEND_BYPASS(1) /* We must set this bit because we 969 * are using NUMBER_UINT */ 970 | S_028C70_RAT(1) 971 ; 972 973 surf->cb_color_attrib = S_028C74_NON_DISP_TILING_ORDER(1); 974 975 /* For buffers, CB_COLOR0_DIM needs to be set to the number of 976 * elements. */ 977 surf->cb_color_dim = pipe_buffer->width0; 978 979 /* Set the buffer range the GPU will have access to: */ 980 util_range_add(&r600_resource(pipe_buffer)->valid_buffer_range, 981 0, pipe_buffer->width0); 982 983 surf->cb_color_fmask = surf->cb_color_base; 984 surf->cb_color_fmask_slice = 0; 985 } 986 987 void evergreen_init_color_surface(struct r600_context *rctx, 988 struct r600_surface *surf) 989 { 990 struct r600_screen *rscreen = rctx->screen; 991 struct r600_texture *rtex = (struct r600_texture*)surf->base.texture; 992 unsigned level = surf->base.u.tex.level; 993 unsigned pitch, slice; 994 unsigned color_info, color_attrib, color_dim = 0, color_view; 995 unsigned format, swap, ntype, endian; 996 uint64_t offset, base_offset; 997 unsigned non_disp_tiling, macro_aspect, tile_split, bankh, bankw, fmask_bankh, nbanks; 998 const struct util_format_description *desc; 999 int i; 1000 bool blend_clamp = 0, blend_bypass = 0, do_endian_swap = FALSE; 1001 1002 offset = rtex->surface.level[level].offset; 1003 color_view = S_028C6C_SLICE_START(surf->base.u.tex.first_layer) | 1004 S_028C6C_SLICE_MAX(surf->base.u.tex.last_layer); 1005 1006 pitch = (rtex->surface.level[level].nblk_x) / 8 - 1; 1007 slice = (rtex->surface.level[level].nblk_x * rtex->surface.level[level].nblk_y) / 64; 1008 if (slice) { 1009 slice = slice - 1; 1010 } 1011 color_info = 0; 1012 switch (rtex->surface.level[level].mode) { 1013 default: 1014 case RADEON_SURF_MODE_LINEAR_ALIGNED: 1015 color_info = S_028C70_ARRAY_MODE(V_028C70_ARRAY_LINEAR_ALIGNED); 1016 non_disp_tiling = 1; 1017 break; 1018 case RADEON_SURF_MODE_1D: 1019 color_info = S_028C70_ARRAY_MODE(V_028C70_ARRAY_1D_TILED_THIN1); 1020 non_disp_tiling = rtex->non_disp_tiling; 1021 break; 1022 case RADEON_SURF_MODE_2D: 1023 color_info = S_028C70_ARRAY_MODE(V_028C70_ARRAY_2D_TILED_THIN1); 1024 non_disp_tiling = rtex->non_disp_tiling; 1025 break; 1026 } 1027 tile_split = rtex->surface.tile_split; 1028 macro_aspect = rtex->surface.mtilea; 1029 bankw = rtex->surface.bankw; 1030 bankh = rtex->surface.bankh; 1031 if (rtex->fmask.size) 1032 fmask_bankh = rtex->fmask.bank_height; 1033 else 1034 fmask_bankh = rtex->surface.bankh; 1035 tile_split = eg_tile_split(tile_split); 1036 macro_aspect = eg_macro_tile_aspect(macro_aspect); 1037 bankw = eg_bank_wh(bankw); 1038 bankh = eg_bank_wh(bankh); 1039 fmask_bankh = eg_bank_wh(fmask_bankh); 1040 1041 /* 128 bit formats require tile type = 1 */ 1042 if (rscreen->b.chip_class == CAYMAN) { 1043 if (util_format_get_blocksize(surf->base.format) >= 16) 1044 non_disp_tiling = 1; 1045 } 1046 nbanks = eg_num_banks(rscreen->b.info.r600_num_banks); 1047 desc = util_format_description(surf->base.format); 1048 for (i = 0; i < 4; i++) { 1049 if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) { 1050 break; 1051 } 1052 } 1053 1054 color_attrib = S_028C74_TILE_SPLIT(tile_split)| 1055 S_028C74_NUM_BANKS(nbanks) | 1056 S_028C74_BANK_WIDTH(bankw) | 1057 S_028C74_BANK_HEIGHT(bankh) | 1058 S_028C74_MACRO_TILE_ASPECT(macro_aspect) | 1059 S_028C74_NON_DISP_TILING_ORDER(non_disp_tiling) | 1060 S_028C74_FMASK_BANK_HEIGHT(fmask_bankh); 1061 1062 if (rctx->b.chip_class == CAYMAN) { 1063 color_attrib |= S_028C74_FORCE_DST_ALPHA_1(desc->swizzle[3] == 1064 PIPE_SWIZZLE_1); 1065 1066 if (rtex->resource.b.b.nr_samples > 1) { 1067 unsigned log_samples = util_logbase2(rtex->resource.b.b.nr_samples); 1068 color_attrib |= S_028C74_NUM_SAMPLES(log_samples) | 1069 S_028C74_NUM_FRAGMENTS(log_samples); 1070 } 1071 } 1072 1073 ntype = V_028C70_NUMBER_UNORM; 1074 if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB) 1075 ntype = V_028C70_NUMBER_SRGB; 1076 else if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) { 1077 if (desc->channel[i].normalized) 1078 ntype = V_028C70_NUMBER_SNORM; 1079 else if (desc->channel[i].pure_integer) 1080 ntype = V_028C70_NUMBER_SINT; 1081 } else if (desc->channel[i].type == UTIL_FORMAT_TYPE_UNSIGNED) { 1082 if (desc->channel[i].normalized) 1083 ntype = V_028C70_NUMBER_UNORM; 1084 else if (desc->channel[i].pure_integer) 1085 ntype = V_028C70_NUMBER_UINT; 1086 } 1087 1088 if (R600_BIG_ENDIAN) 1089 do_endian_swap = !rtex->db_compatible; 1090 1091 format = r600_translate_colorformat(rctx->b.chip_class, surf->base.format, 1092 do_endian_swap); 1093 assert(format != ~0); 1094 1095 swap = r600_translate_colorswap(surf->base.format, do_endian_swap); 1096 assert(swap != ~0); 1097 1098 endian = r600_colorformat_endian_swap(format, do_endian_swap); 1099 1100 /* blend clamp should be set for all NORM/SRGB types */ 1101 if (ntype == V_028C70_NUMBER_UNORM || ntype == V_028C70_NUMBER_SNORM || 1102 ntype == V_028C70_NUMBER_SRGB) 1103 blend_clamp = 1; 1104 1105 /* set blend bypass according to docs if SINT/UINT or 1106 8/24 COLOR variants */ 1107 if (ntype == V_028C70_NUMBER_UINT || ntype == V_028C70_NUMBER_SINT || 1108 format == V_028C70_COLOR_8_24 || format == V_028C70_COLOR_24_8 || 1109 format == V_028C70_COLOR_X24_8_32_FLOAT) { 1110 blend_clamp = 0; 1111 blend_bypass = 1; 1112 } 1113 1114 surf->alphatest_bypass = ntype == V_028C70_NUMBER_UINT || ntype == V_028C70_NUMBER_SINT; 1115 1116 color_info |= S_028C70_FORMAT(format) | 1117 S_028C70_COMP_SWAP(swap) | 1118 S_028C70_BLEND_CLAMP(blend_clamp) | 1119 S_028C70_BLEND_BYPASS(blend_bypass) | 1120 S_028C70_NUMBER_TYPE(ntype) | 1121 S_028C70_ENDIAN(endian); 1122 1123 /* EXPORT_NORM is an optimzation that can be enabled for better 1124 * performance in certain cases. 1125 * EXPORT_NORM can be enabled if: 1126 * - 11-bit or smaller UNORM/SNORM/SRGB 1127 * - 16-bit or smaller FLOAT 1128 */ 1129 if (desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS && 1130 ((desc->channel[i].size < 12 && 1131 desc->channel[i].type != UTIL_FORMAT_TYPE_FLOAT && 1132 ntype != V_028C70_NUMBER_UINT && ntype != V_028C70_NUMBER_SINT) || 1133 (desc->channel[i].size < 17 && 1134 desc->channel[i].type == UTIL_FORMAT_TYPE_FLOAT))) { 1135 color_info |= S_028C70_SOURCE_FORMAT(V_028C70_EXPORT_4C_16BPC); 1136 surf->export_16bpc = true; 1137 } 1138 1139 if (rtex->fmask.size) { 1140 color_info |= S_028C70_COMPRESSION(1); 1141 } 1142 1143 base_offset = rtex->resource.gpu_address; 1144 1145 /* XXX handle enabling of CB beyond BASE8 which has different offset */ 1146 surf->cb_color_base = (base_offset + offset) >> 8; 1147 surf->cb_color_dim = color_dim; 1148 surf->cb_color_info = color_info; 1149 surf->cb_color_pitch = S_028C64_PITCH_TILE_MAX(pitch); 1150 surf->cb_color_slice = S_028C68_SLICE_TILE_MAX(slice); 1151 surf->cb_color_view = color_view; 1152 surf->cb_color_attrib = color_attrib; 1153 if (rtex->fmask.size) { 1154 surf->cb_color_fmask = (base_offset + rtex->fmask.offset) >> 8; 1155 surf->cb_color_fmask_slice = S_028C88_TILE_MAX(rtex->fmask.slice_tile_max); 1156 } else { 1157 surf->cb_color_fmask = surf->cb_color_base; 1158 surf->cb_color_fmask_slice = S_028C88_TILE_MAX(slice); 1159 } 1160 1161 surf->color_initialized = true; 1162 } 1163 1164 static void evergreen_init_depth_surface(struct r600_context *rctx, 1165 struct r600_surface *surf) 1166 { 1167 struct r600_screen *rscreen = rctx->screen; 1168 struct r600_texture *rtex = (struct r600_texture*)surf->base.texture; 1169 unsigned level = surf->base.u.tex.level; 1170 struct radeon_surf_level *levelinfo = &rtex->surface.level[level]; 1171 uint64_t offset; 1172 unsigned format, array_mode; 1173 unsigned macro_aspect, tile_split, bankh, bankw, nbanks; 1174 1175 1176 format = r600_translate_dbformat(surf->base.format); 1177 assert(format != ~0); 1178 1179 offset = rtex->resource.gpu_address; 1180 offset += rtex->surface.level[level].offset; 1181 1182 switch (rtex->surface.level[level].mode) { 1183 case RADEON_SURF_MODE_2D: 1184 array_mode = V_028C70_ARRAY_2D_TILED_THIN1; 1185 break; 1186 case RADEON_SURF_MODE_1D: 1187 case RADEON_SURF_MODE_LINEAR_ALIGNED: 1188 default: 1189 array_mode = V_028C70_ARRAY_1D_TILED_THIN1; 1190 break; 1191 } 1192 tile_split = rtex->surface.tile_split; 1193 macro_aspect = rtex->surface.mtilea; 1194 bankw = rtex->surface.bankw; 1195 bankh = rtex->surface.bankh; 1196 tile_split = eg_tile_split(tile_split); 1197 macro_aspect = eg_macro_tile_aspect(macro_aspect); 1198 bankw = eg_bank_wh(bankw); 1199 bankh = eg_bank_wh(bankh); 1200 nbanks = eg_num_banks(rscreen->b.info.r600_num_banks); 1201 offset >>= 8; 1202 1203 surf->db_z_info = S_028040_ARRAY_MODE(array_mode) | 1204 S_028040_FORMAT(format) | 1205 S_028040_TILE_SPLIT(tile_split)| 1206 S_028040_NUM_BANKS(nbanks) | 1207 S_028040_BANK_WIDTH(bankw) | 1208 S_028040_BANK_HEIGHT(bankh) | 1209 S_028040_MACRO_TILE_ASPECT(macro_aspect); 1210 if (rscreen->b.chip_class == CAYMAN && rtex->resource.b.b.nr_samples > 1) { 1211 surf->db_z_info |= S_028040_NUM_SAMPLES(util_logbase2(rtex->resource.b.b.nr_samples)); 1212 } 1213 1214 assert(levelinfo->nblk_x % 8 == 0 && levelinfo->nblk_y % 8 == 0); 1215 1216 surf->db_depth_base = offset; 1217 surf->db_depth_view = S_028008_SLICE_START(surf->base.u.tex.first_layer) | 1218 S_028008_SLICE_MAX(surf->base.u.tex.last_layer); 1219 surf->db_depth_size = S_028058_PITCH_TILE_MAX(levelinfo->nblk_x / 8 - 1) | 1220 S_028058_HEIGHT_TILE_MAX(levelinfo->nblk_y / 8 - 1); 1221 surf->db_depth_slice = S_02805C_SLICE_TILE_MAX(levelinfo->nblk_x * 1222 levelinfo->nblk_y / 64 - 1); 1223 1224 if (rtex->surface.flags & RADEON_SURF_SBUFFER) { 1225 uint64_t stencil_offset; 1226 unsigned stile_split = rtex->surface.stencil_tile_split; 1227 1228 stile_split = eg_tile_split(stile_split); 1229 1230 stencil_offset = rtex->surface.stencil_level[level].offset; 1231 stencil_offset += rtex->resource.gpu_address; 1232 1233 surf->db_stencil_base = stencil_offset >> 8; 1234 surf->db_stencil_info = S_028044_FORMAT(V_028044_STENCIL_8) | 1235 S_028044_TILE_SPLIT(stile_split); 1236 } else { 1237 surf->db_stencil_base = offset; 1238 /* DRM 2.6.18 allows the INVALID format to disable stencil. 1239 * Older kernels are out of luck. */ 1240 surf->db_stencil_info = rctx->screen->b.info.drm_minor >= 18 ? 1241 S_028044_FORMAT(V_028044_STENCIL_INVALID) : 1242 S_028044_FORMAT(V_028044_STENCIL_8); 1243 } 1244 1245 /* use htile only for first level */ 1246 if (rtex->htile_buffer && !level) { 1247 uint64_t va = rtex->htile_buffer->gpu_address; 1248 surf->db_htile_data_base = va >> 8; 1249 surf->db_htile_surface = S_028ABC_HTILE_WIDTH(1) | 1250 S_028ABC_HTILE_HEIGHT(1) | 1251 S_028ABC_FULL_CACHE(1); 1252 surf->db_z_info |= S_028040_TILE_SURFACE_ENABLE(1); 1253 surf->db_preload_control = 0; 1254 } 1255 1256 surf->depth_initialized = true; 1257 } 1258 1259 static void evergreen_set_framebuffer_state(struct pipe_context *ctx, 1260 const struct pipe_framebuffer_state *state) 1261 { 1262 struct r600_context *rctx = (struct r600_context *)ctx; 1263 struct r600_surface *surf; 1264 struct r600_texture *rtex; 1265 uint32_t i, log_samples; 1266 1267 /* Flush TC when changing the framebuffer state, because the only 1268 * client not using TC that can change textures is the framebuffer. 1269 * Other places don't typically have to flush TC. 1270 */ 1271 rctx->b.flags |= R600_CONTEXT_WAIT_3D_IDLE | 1272 R600_CONTEXT_FLUSH_AND_INV | 1273 R600_CONTEXT_FLUSH_AND_INV_CB | 1274 R600_CONTEXT_FLUSH_AND_INV_CB_META | 1275 R600_CONTEXT_FLUSH_AND_INV_DB | 1276 R600_CONTEXT_FLUSH_AND_INV_DB_META | 1277 R600_CONTEXT_INV_TEX_CACHE; 1278 1279 util_copy_framebuffer_state(&rctx->framebuffer.state, state); 1280 1281 /* Colorbuffers. */ 1282 rctx->framebuffer.export_16bpc = state->nr_cbufs != 0; 1283 rctx->framebuffer.cb0_is_integer = state->nr_cbufs && state->cbufs[0] && 1284 util_format_is_pure_integer(state->cbufs[0]->format); 1285 rctx->framebuffer.compressed_cb_mask = 0; 1286 rctx->framebuffer.nr_samples = util_framebuffer_get_num_samples(state); 1287 1288 for (i = 0; i < state->nr_cbufs; i++) { 1289 surf = (struct r600_surface*)state->cbufs[i]; 1290 if (!surf) 1291 continue; 1292 1293 rtex = (struct r600_texture*)surf->base.texture; 1294 1295 r600_context_add_resource_size(ctx, state->cbufs[i]->texture); 1296 1297 if (!surf->color_initialized) { 1298 evergreen_init_color_surface(rctx, surf); 1299 } 1300 1301 if (!surf->export_16bpc) { 1302 rctx->framebuffer.export_16bpc = false; 1303 } 1304 1305 if (rtex->fmask.size) { 1306 rctx->framebuffer.compressed_cb_mask |= 1 << i; 1307 } 1308 } 1309 1310 /* Update alpha-test state dependencies. 1311 * Alpha-test is done on the first colorbuffer only. */ 1312 if (state->nr_cbufs) { 1313 bool alphatest_bypass = false; 1314 bool export_16bpc = true; 1315 1316 surf = (struct r600_surface*)state->cbufs[0]; 1317 if (surf) { 1318 alphatest_bypass = surf->alphatest_bypass; 1319 export_16bpc = surf->export_16bpc; 1320 } 1321 1322 if (rctx->alphatest_state.bypass != alphatest_bypass) { 1323 rctx->alphatest_state.bypass = alphatest_bypass; 1324 r600_mark_atom_dirty(rctx, &rctx->alphatest_state.atom); 1325 } 1326 if (rctx->alphatest_state.cb0_export_16bpc != export_16bpc) { 1327 rctx->alphatest_state.cb0_export_16bpc = export_16bpc; 1328 r600_mark_atom_dirty(rctx, &rctx->alphatest_state.atom); 1329 } 1330 } 1331 1332 /* ZS buffer. */ 1333 if (state->zsbuf) { 1334 surf = (struct r600_surface*)state->zsbuf; 1335 1336 r600_context_add_resource_size(ctx, state->zsbuf->texture); 1337 1338 if (!surf->depth_initialized) { 1339 evergreen_init_depth_surface(rctx, surf); 1340 } 1341 1342 if (state->zsbuf->format != rctx->poly_offset_state.zs_format) { 1343 rctx->poly_offset_state.zs_format = state->zsbuf->format; 1344 r600_mark_atom_dirty(rctx, &rctx->poly_offset_state.atom); 1345 } 1346 1347 if (rctx->db_state.rsurf != surf) { 1348 rctx->db_state.rsurf = surf; 1349 r600_mark_atom_dirty(rctx, &rctx->db_state.atom); 1350 r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom); 1351 } 1352 } else if (rctx->db_state.rsurf) { 1353 rctx->db_state.rsurf = NULL; 1354 r600_mark_atom_dirty(rctx, &rctx->db_state.atom); 1355 r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom); 1356 } 1357 1358 if (rctx->cb_misc_state.nr_cbufs != state->nr_cbufs) { 1359 rctx->cb_misc_state.nr_cbufs = state->nr_cbufs; 1360 r600_mark_atom_dirty(rctx, &rctx->cb_misc_state.atom); 1361 } 1362 1363 if (state->nr_cbufs == 0 && rctx->alphatest_state.bypass) { 1364 rctx->alphatest_state.bypass = false; 1365 r600_mark_atom_dirty(rctx, &rctx->alphatest_state.atom); 1366 } 1367 1368 log_samples = util_logbase2(rctx->framebuffer.nr_samples); 1369 /* This is for Cayman to program SAMPLE_RATE, and for RV770 to fix a hw bug. */ 1370 if ((rctx->b.chip_class == CAYMAN || 1371 rctx->b.family == CHIP_RV770) && 1372 rctx->db_misc_state.log_samples != log_samples) { 1373 rctx->db_misc_state.log_samples = log_samples; 1374 r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom); 1375 } 1376 1377 1378 /* Calculate the CS size. */ 1379 rctx->framebuffer.atom.num_dw = 4; /* SCISSOR */ 1380 1381 /* MSAA. */ 1382 if (rctx->b.chip_class == EVERGREEN) 1383 rctx->framebuffer.atom.num_dw += 17; /* Evergreen */ 1384 else 1385 rctx->framebuffer.atom.num_dw += 28; /* Cayman */ 1386 1387 /* Colorbuffers. */ 1388 rctx->framebuffer.atom.num_dw += state->nr_cbufs * 23; 1389 rctx->framebuffer.atom.num_dw += state->nr_cbufs * 2; 1390 rctx->framebuffer.atom.num_dw += (12 - state->nr_cbufs) * 3; 1391 1392 /* ZS buffer. */ 1393 if (state->zsbuf) { 1394 rctx->framebuffer.atom.num_dw += 24; 1395 rctx->framebuffer.atom.num_dw += 2; 1396 } else if (rctx->screen->b.info.drm_minor >= 18) { 1397 rctx->framebuffer.atom.num_dw += 4; 1398 } 1399 1400 r600_mark_atom_dirty(rctx, &rctx->framebuffer.atom); 1401 1402 r600_set_sample_locations_constant_buffer(rctx); 1403 } 1404 1405 static void evergreen_set_min_samples(struct pipe_context *ctx, unsigned min_samples) 1406 { 1407 struct r600_context *rctx = (struct r600_context *)ctx; 1408 1409 if (rctx->ps_iter_samples == min_samples) 1410 return; 1411 1412 rctx->ps_iter_samples = min_samples; 1413 if (rctx->framebuffer.nr_samples > 1) { 1414 r600_mark_atom_dirty(rctx, &rctx->framebuffer.atom); 1415 } 1416 } 1417 1418 /* 8xMSAA */ 1419 static uint32_t sample_locs_8x[] = { 1420 FILL_SREG(-1, 1, 1, 5, 3, -5, 5, 3), 1421 FILL_SREG(-7, -1, -3, -7, 7, -3, -5, 7), 1422 FILL_SREG(-1, 1, 1, 5, 3, -5, 5, 3), 1423 FILL_SREG(-7, -1, -3, -7, 7, -3, -5, 7), 1424 FILL_SREG(-1, 1, 1, 5, 3, -5, 5, 3), 1425 FILL_SREG(-7, -1, -3, -7, 7, -3, -5, 7), 1426 FILL_SREG(-1, 1, 1, 5, 3, -5, 5, 3), 1427 FILL_SREG(-7, -1, -3, -7, 7, -3, -5, 7), 1428 }; 1429 static unsigned max_dist_8x = 7; 1430 1431 static void evergreen_get_sample_position(struct pipe_context *ctx, 1432 unsigned sample_count, 1433 unsigned sample_index, 1434 float *out_value) 1435 { 1436 int offset, index; 1437 struct { 1438 int idx:4; 1439 } val; 1440 switch (sample_count) { 1441 case 1: 1442 default: 1443 out_value[0] = out_value[1] = 0.5; 1444 break; 1445 case 2: 1446 offset = 4 * (sample_index * 2); 1447 val.idx = (eg_sample_locs_2x[0] >> offset) & 0xf; 1448 out_value[0] = (float)(val.idx + 8) / 16.0f; 1449 val.idx = (eg_sample_locs_2x[0] >> (offset + 4)) & 0xf; 1450 out_value[1] = (float)(val.idx + 8) / 16.0f; 1451 break; 1452 case 4: 1453 offset = 4 * (sample_index * 2); 1454 val.idx = (eg_sample_locs_4x[0] >> offset) & 0xf; 1455 out_value[0] = (float)(val.idx + 8) / 16.0f; 1456 val.idx = (eg_sample_locs_4x[0] >> (offset + 4)) & 0xf; 1457 out_value[1] = (float)(val.idx + 8) / 16.0f; 1458 break; 1459 case 8: 1460 offset = 4 * (sample_index % 4 * 2); 1461 index = (sample_index / 4); 1462 val.idx = (sample_locs_8x[index] >> offset) & 0xf; 1463 out_value[0] = (float)(val.idx + 8) / 16.0f; 1464 val.idx = (sample_locs_8x[index] >> (offset + 4)) & 0xf; 1465 out_value[1] = (float)(val.idx + 8) / 16.0f; 1466 break; 1467 } 1468 } 1469 1470 static void evergreen_emit_msaa_state(struct r600_context *rctx, int nr_samples, int ps_iter_samples) 1471 { 1472 1473 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 1474 unsigned max_dist = 0; 1475 1476 switch (nr_samples) { 1477 default: 1478 nr_samples = 0; 1479 break; 1480 case 2: 1481 radeon_set_context_reg_seq(cs, R_028C1C_PA_SC_AA_SAMPLE_LOCS_0, ARRAY_SIZE(eg_sample_locs_2x)); 1482 radeon_emit_array(cs, eg_sample_locs_2x, ARRAY_SIZE(eg_sample_locs_2x)); 1483 max_dist = eg_max_dist_2x; 1484 break; 1485 case 4: 1486 radeon_set_context_reg_seq(cs, R_028C1C_PA_SC_AA_SAMPLE_LOCS_0, ARRAY_SIZE(eg_sample_locs_4x)); 1487 radeon_emit_array(cs, eg_sample_locs_4x, ARRAY_SIZE(eg_sample_locs_4x)); 1488 max_dist = eg_max_dist_4x; 1489 break; 1490 case 8: 1491 radeon_set_context_reg_seq(cs, R_028C1C_PA_SC_AA_SAMPLE_LOCS_0, ARRAY_SIZE(sample_locs_8x)); 1492 radeon_emit_array(cs, sample_locs_8x, ARRAY_SIZE(sample_locs_8x)); 1493 max_dist = max_dist_8x; 1494 break; 1495 } 1496 1497 if (nr_samples > 1) { 1498 radeon_set_context_reg_seq(cs, R_028C00_PA_SC_LINE_CNTL, 2); 1499 radeon_emit(cs, S_028C00_LAST_PIXEL(1) | 1500 S_028C00_EXPAND_LINE_WIDTH(1)); /* R_028C00_PA_SC_LINE_CNTL */ 1501 radeon_emit(cs, S_028C04_MSAA_NUM_SAMPLES(util_logbase2(nr_samples)) | 1502 S_028C04_MAX_SAMPLE_DIST(max_dist)); /* R_028C04_PA_SC_AA_CONFIG */ 1503 radeon_set_context_reg(cs, EG_R_028A4C_PA_SC_MODE_CNTL_1, 1504 EG_S_028A4C_PS_ITER_SAMPLE(ps_iter_samples > 1) | 1505 EG_S_028A4C_FORCE_EOV_CNTDWN_ENABLE(1) | 1506 EG_S_028A4C_FORCE_EOV_REZ_ENABLE(1)); 1507 } else { 1508 radeon_set_context_reg_seq(cs, R_028C00_PA_SC_LINE_CNTL, 2); 1509 radeon_emit(cs, S_028C00_LAST_PIXEL(1)); /* R_028C00_PA_SC_LINE_CNTL */ 1510 radeon_emit(cs, 0); /* R_028C04_PA_SC_AA_CONFIG */ 1511 radeon_set_context_reg(cs, EG_R_028A4C_PA_SC_MODE_CNTL_1, 1512 EG_S_028A4C_FORCE_EOV_CNTDWN_ENABLE(1) | 1513 EG_S_028A4C_FORCE_EOV_REZ_ENABLE(1)); 1514 } 1515 } 1516 1517 static void evergreen_emit_framebuffer_state(struct r600_context *rctx, struct r600_atom *atom) 1518 { 1519 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 1520 struct pipe_framebuffer_state *state = &rctx->framebuffer.state; 1521 unsigned nr_cbufs = state->nr_cbufs; 1522 unsigned i, tl, br; 1523 struct r600_texture *tex = NULL; 1524 struct r600_surface *cb = NULL; 1525 1526 /* XXX support more colorbuffers once we need them */ 1527 assert(nr_cbufs <= 8); 1528 if (nr_cbufs > 8) 1529 nr_cbufs = 8; 1530 1531 /* Colorbuffers. */ 1532 for (i = 0; i < nr_cbufs; i++) { 1533 unsigned reloc, cmask_reloc; 1534 1535 cb = (struct r600_surface*)state->cbufs[i]; 1536 if (!cb) { 1537 radeon_set_context_reg(cs, R_028C70_CB_COLOR0_INFO + i * 0x3C, 1538 S_028C70_FORMAT(V_028C70_COLOR_INVALID)); 1539 continue; 1540 } 1541 1542 tex = (struct r600_texture *)cb->base.texture; 1543 reloc = radeon_add_to_buffer_list(&rctx->b, 1544 &rctx->b.gfx, 1545 (struct r600_resource*)cb->base.texture, 1546 RADEON_USAGE_READWRITE, 1547 tex->resource.b.b.nr_samples > 1 ? 1548 RADEON_PRIO_COLOR_BUFFER_MSAA : 1549 RADEON_PRIO_COLOR_BUFFER); 1550 1551 if (tex->cmask_buffer && tex->cmask_buffer != &tex->resource) { 1552 cmask_reloc = radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, 1553 tex->cmask_buffer, RADEON_USAGE_READWRITE, 1554 RADEON_PRIO_CMASK); 1555 } else { 1556 cmask_reloc = reloc; 1557 } 1558 1559 radeon_set_context_reg_seq(cs, R_028C60_CB_COLOR0_BASE + i * 0x3C, 13); 1560 radeon_emit(cs, cb->cb_color_base); /* R_028C60_CB_COLOR0_BASE */ 1561 radeon_emit(cs, cb->cb_color_pitch); /* R_028C64_CB_COLOR0_PITCH */ 1562 radeon_emit(cs, cb->cb_color_slice); /* R_028C68_CB_COLOR0_SLICE */ 1563 radeon_emit(cs, cb->cb_color_view); /* R_028C6C_CB_COLOR0_VIEW */ 1564 radeon_emit(cs, cb->cb_color_info | tex->cb_color_info); /* R_028C70_CB_COLOR0_INFO */ 1565 radeon_emit(cs, cb->cb_color_attrib); /* R_028C74_CB_COLOR0_ATTRIB */ 1566 radeon_emit(cs, cb->cb_color_dim); /* R_028C78_CB_COLOR0_DIM */ 1567 radeon_emit(cs, tex->cmask.base_address_reg); /* R_028C7C_CB_COLOR0_CMASK */ 1568 radeon_emit(cs, tex->cmask.slice_tile_max); /* R_028C80_CB_COLOR0_CMASK_SLICE */ 1569 radeon_emit(cs, cb->cb_color_fmask); /* R_028C84_CB_COLOR0_FMASK */ 1570 radeon_emit(cs, cb->cb_color_fmask_slice); /* R_028C88_CB_COLOR0_FMASK_SLICE */ 1571 radeon_emit(cs, tex->color_clear_value[0]); /* R_028C8C_CB_COLOR0_CLEAR_WORD0 */ 1572 radeon_emit(cs, tex->color_clear_value[1]); /* R_028C90_CB_COLOR0_CLEAR_WORD1 */ 1573 1574 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028C60_CB_COLOR0_BASE */ 1575 radeon_emit(cs, reloc); 1576 1577 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028C74_CB_COLOR0_ATTRIB */ 1578 radeon_emit(cs, reloc); 1579 1580 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028C7C_CB_COLOR0_CMASK */ 1581 radeon_emit(cs, cmask_reloc); 1582 1583 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028C84_CB_COLOR0_FMASK */ 1584 radeon_emit(cs, reloc); 1585 } 1586 /* set CB_COLOR1_INFO for possible dual-src blending */ 1587 if (i == 1 && state->cbufs[0]) { 1588 radeon_set_context_reg(cs, R_028C70_CB_COLOR0_INFO + 1 * 0x3C, 1589 cb->cb_color_info | tex->cb_color_info); 1590 i++; 1591 } 1592 for (; i < 8 ; i++) 1593 radeon_set_context_reg(cs, R_028C70_CB_COLOR0_INFO + i * 0x3C, 0); 1594 for (; i < 12; i++) 1595 radeon_set_context_reg(cs, R_028E50_CB_COLOR8_INFO + (i - 8) * 0x1C, 0); 1596 1597 /* ZS buffer. */ 1598 if (state->zsbuf) { 1599 struct r600_surface *zb = (struct r600_surface*)state->zsbuf; 1600 unsigned reloc = radeon_add_to_buffer_list(&rctx->b, 1601 &rctx->b.gfx, 1602 (struct r600_resource*)state->zsbuf->texture, 1603 RADEON_USAGE_READWRITE, 1604 zb->base.texture->nr_samples > 1 ? 1605 RADEON_PRIO_DEPTH_BUFFER_MSAA : 1606 RADEON_PRIO_DEPTH_BUFFER); 1607 1608 radeon_set_context_reg(cs, R_028008_DB_DEPTH_VIEW, zb->db_depth_view); 1609 1610 radeon_set_context_reg_seq(cs, R_028040_DB_Z_INFO, 8); 1611 radeon_emit(cs, zb->db_z_info); /* R_028040_DB_Z_INFO */ 1612 radeon_emit(cs, zb->db_stencil_info); /* R_028044_DB_STENCIL_INFO */ 1613 radeon_emit(cs, zb->db_depth_base); /* R_028048_DB_Z_READ_BASE */ 1614 radeon_emit(cs, zb->db_stencil_base); /* R_02804C_DB_STENCIL_READ_BASE */ 1615 radeon_emit(cs, zb->db_depth_base); /* R_028050_DB_Z_WRITE_BASE */ 1616 radeon_emit(cs, zb->db_stencil_base); /* R_028054_DB_STENCIL_WRITE_BASE */ 1617 radeon_emit(cs, zb->db_depth_size); /* R_028058_DB_DEPTH_SIZE */ 1618 radeon_emit(cs, zb->db_depth_slice); /* R_02805C_DB_DEPTH_SLICE */ 1619 1620 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028048_DB_Z_READ_BASE */ 1621 radeon_emit(cs, reloc); 1622 1623 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); /* R_02804C_DB_STENCIL_READ_BASE */ 1624 radeon_emit(cs, reloc); 1625 1626 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028050_DB_Z_WRITE_BASE */ 1627 radeon_emit(cs, reloc); 1628 1629 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); /* R_028054_DB_STENCIL_WRITE_BASE */ 1630 radeon_emit(cs, reloc); 1631 } else if (rctx->screen->b.info.drm_minor >= 18) { 1632 /* DRM 2.6.18 allows the INVALID format to disable depth/stencil. 1633 * Older kernels are out of luck. */ 1634 radeon_set_context_reg_seq(cs, R_028040_DB_Z_INFO, 2); 1635 radeon_emit(cs, S_028040_FORMAT(V_028040_Z_INVALID)); /* R_028040_DB_Z_INFO */ 1636 radeon_emit(cs, S_028044_FORMAT(V_028044_STENCIL_INVALID)); /* R_028044_DB_STENCIL_INFO */ 1637 } 1638 1639 /* Framebuffer dimensions. */ 1640 evergreen_get_scissor_rect(rctx, 0, 0, state->width, state->height, &tl, &br); 1641 1642 radeon_set_context_reg_seq(cs, R_028204_PA_SC_WINDOW_SCISSOR_TL, 2); 1643 radeon_emit(cs, tl); /* R_028204_PA_SC_WINDOW_SCISSOR_TL */ 1644 radeon_emit(cs, br); /* R_028208_PA_SC_WINDOW_SCISSOR_BR */ 1645 1646 if (rctx->b.chip_class == EVERGREEN) { 1647 evergreen_emit_msaa_state(rctx, rctx->framebuffer.nr_samples, rctx->ps_iter_samples); 1648 } else { 1649 unsigned sc_mode_cntl_1 = 1650 EG_S_028A4C_FORCE_EOV_CNTDWN_ENABLE(1) | 1651 EG_S_028A4C_FORCE_EOV_REZ_ENABLE(1); 1652 1653 if (rctx->framebuffer.nr_samples > 1) 1654 cayman_emit_msaa_sample_locs(cs, rctx->framebuffer.nr_samples); 1655 cayman_emit_msaa_config(cs, rctx->framebuffer.nr_samples, 1656 rctx->ps_iter_samples, 0, sc_mode_cntl_1); 1657 } 1658 } 1659 1660 static void evergreen_emit_polygon_offset(struct r600_context *rctx, struct r600_atom *a) 1661 { 1662 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 1663 struct r600_poly_offset_state *state = (struct r600_poly_offset_state*)a; 1664 float offset_units = state->offset_units; 1665 float offset_scale = state->offset_scale; 1666 uint32_t pa_su_poly_offset_db_fmt_cntl = 0; 1667 1668 if (!state->offset_units_unscaled) { 1669 switch (state->zs_format) { 1670 case PIPE_FORMAT_Z24X8_UNORM: 1671 case PIPE_FORMAT_Z24_UNORM_S8_UINT: 1672 case PIPE_FORMAT_X8Z24_UNORM: 1673 case PIPE_FORMAT_S8_UINT_Z24_UNORM: 1674 offset_units *= 2.0f; 1675 pa_su_poly_offset_db_fmt_cntl = 1676 S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS((char)-24); 1677 break; 1678 case PIPE_FORMAT_Z16_UNORM: 1679 offset_units *= 4.0f; 1680 pa_su_poly_offset_db_fmt_cntl = 1681 S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS((char)-16); 1682 break; 1683 default: 1684 pa_su_poly_offset_db_fmt_cntl = 1685 S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS((char)-23) | 1686 S_028B78_POLY_OFFSET_DB_IS_FLOAT_FMT(1); 1687 } 1688 } 1689 1690 radeon_set_context_reg_seq(cs, R_028B80_PA_SU_POLY_OFFSET_FRONT_SCALE, 4); 1691 radeon_emit(cs, fui(offset_scale)); 1692 radeon_emit(cs, fui(offset_units)); 1693 radeon_emit(cs, fui(offset_scale)); 1694 radeon_emit(cs, fui(offset_units)); 1695 1696 radeon_set_context_reg(cs, R_028B78_PA_SU_POLY_OFFSET_DB_FMT_CNTL, 1697 pa_su_poly_offset_db_fmt_cntl); 1698 } 1699 1700 static void evergreen_emit_cb_misc_state(struct r600_context *rctx, struct r600_atom *atom) 1701 { 1702 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 1703 struct r600_cb_misc_state *a = (struct r600_cb_misc_state*)atom; 1704 unsigned fb_colormask = (1ULL << ((unsigned)a->nr_cbufs * 4)) - 1; 1705 unsigned ps_colormask = (1ULL << ((unsigned)a->nr_ps_color_outputs * 4)) - 1; 1706 1707 radeon_set_context_reg_seq(cs, R_028238_CB_TARGET_MASK, 2); 1708 radeon_emit(cs, a->blend_colormask & fb_colormask); /* R_028238_CB_TARGET_MASK */ 1709 /* This must match the used export instructions exactly. 1710 * Other values may lead to undefined behavior and hangs. 1711 */ 1712 radeon_emit(cs, ps_colormask); /* R_02823C_CB_SHADER_MASK */ 1713 } 1714 1715 static void evergreen_emit_db_state(struct r600_context *rctx, struct r600_atom *atom) 1716 { 1717 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 1718 struct r600_db_state *a = (struct r600_db_state*)atom; 1719 1720 if (a->rsurf && a->rsurf->db_htile_surface) { 1721 struct r600_texture *rtex = (struct r600_texture *)a->rsurf->base.texture; 1722 unsigned reloc_idx; 1723 1724 radeon_set_context_reg(cs, R_02802C_DB_DEPTH_CLEAR, fui(rtex->depth_clear_value)); 1725 radeon_set_context_reg(cs, R_028ABC_DB_HTILE_SURFACE, a->rsurf->db_htile_surface); 1726 radeon_set_context_reg(cs, R_028AC8_DB_PRELOAD_CONTROL, a->rsurf->db_preload_control); 1727 radeon_set_context_reg(cs, R_028014_DB_HTILE_DATA_BASE, a->rsurf->db_htile_data_base); 1728 reloc_idx = radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rtex->htile_buffer, 1729 RADEON_USAGE_READWRITE, RADEON_PRIO_HTILE); 1730 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); 1731 radeon_emit(cs, reloc_idx); 1732 } else { 1733 radeon_set_context_reg(cs, R_028ABC_DB_HTILE_SURFACE, 0); 1734 radeon_set_context_reg(cs, R_028AC8_DB_PRELOAD_CONTROL, 0); 1735 } 1736 } 1737 1738 static void evergreen_emit_db_misc_state(struct r600_context *rctx, struct r600_atom *atom) 1739 { 1740 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 1741 struct r600_db_misc_state *a = (struct r600_db_misc_state*)atom; 1742 unsigned db_render_control = 0; 1743 unsigned db_count_control = 0; 1744 unsigned db_render_override = 1745 S_02800C_FORCE_HIS_ENABLE0(V_02800C_FORCE_DISABLE) | 1746 S_02800C_FORCE_HIS_ENABLE1(V_02800C_FORCE_DISABLE); 1747 1748 if (rctx->b.num_occlusion_queries > 0 && 1749 !a->occlusion_queries_disabled) { 1750 db_count_control |= S_028004_PERFECT_ZPASS_COUNTS(1); 1751 if (rctx->b.chip_class == CAYMAN) { 1752 db_count_control |= S_028004_SAMPLE_RATE(a->log_samples); 1753 } 1754 db_render_override |= S_02800C_NOOP_CULL_DISABLE(1); 1755 } else { 1756 db_count_control |= S_028004_ZPASS_INCREMENT_DISABLE(1); 1757 } 1758 1759 /* This is to fix a lockup when hyperz and alpha test are enabled at 1760 * the same time somehow GPU get confuse on which order to pick for 1761 * z test 1762 */ 1763 if (rctx->alphatest_state.sx_alpha_test_control) 1764 db_render_override |= S_02800C_FORCE_SHADER_Z_ORDER(1); 1765 1766 if (a->flush_depthstencil_through_cb) { 1767 assert(a->copy_depth || a->copy_stencil); 1768 1769 db_render_control |= S_028000_DEPTH_COPY_ENABLE(a->copy_depth) | 1770 S_028000_STENCIL_COPY_ENABLE(a->copy_stencil) | 1771 S_028000_COPY_CENTROID(1) | 1772 S_028000_COPY_SAMPLE(a->copy_sample); 1773 } else if (a->flush_depth_inplace || a->flush_stencil_inplace) { 1774 db_render_control |= S_028000_DEPTH_COMPRESS_DISABLE(a->flush_depth_inplace) | 1775 S_028000_STENCIL_COMPRESS_DISABLE(a->flush_stencil_inplace); 1776 db_render_override |= S_02800C_DISABLE_PIXEL_RATE_TILES(1); 1777 } 1778 if (a->htile_clear) { 1779 /* FIXME we might want to disable cliprect here */ 1780 db_render_control |= S_028000_DEPTH_CLEAR_ENABLE(1); 1781 } 1782 1783 radeon_set_context_reg_seq(cs, R_028000_DB_RENDER_CONTROL, 2); 1784 radeon_emit(cs, db_render_control); /* R_028000_DB_RENDER_CONTROL */ 1785 radeon_emit(cs, db_count_control); /* R_028004_DB_COUNT_CONTROL */ 1786 radeon_set_context_reg(cs, R_02800C_DB_RENDER_OVERRIDE, db_render_override); 1787 radeon_set_context_reg(cs, R_02880C_DB_SHADER_CONTROL, a->db_shader_control); 1788 } 1789 1790 static void evergreen_emit_vertex_buffers(struct r600_context *rctx, 1791 struct r600_vertexbuf_state *state, 1792 unsigned resource_offset, 1793 unsigned pkt_flags) 1794 { 1795 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 1796 uint32_t dirty_mask = state->dirty_mask; 1797 1798 while (dirty_mask) { 1799 struct pipe_vertex_buffer *vb; 1800 struct r600_resource *rbuffer; 1801 uint64_t va; 1802 unsigned buffer_index = u_bit_scan(&dirty_mask); 1803 1804 vb = &state->vb[buffer_index]; 1805 rbuffer = (struct r600_resource*)vb->buffer; 1806 assert(rbuffer); 1807 1808 va = rbuffer->gpu_address + vb->buffer_offset; 1809 1810 /* fetch resources start at index 992 */ 1811 radeon_emit(cs, PKT3(PKT3_SET_RESOURCE, 8, 0) | pkt_flags); 1812 radeon_emit(cs, (resource_offset + buffer_index) * 8); 1813 radeon_emit(cs, va); /* RESOURCEi_WORD0 */ 1814 radeon_emit(cs, rbuffer->b.b.width0 - vb->buffer_offset - 1); /* RESOURCEi_WORD1 */ 1815 radeon_emit(cs, /* RESOURCEi_WORD2 */ 1816 S_030008_ENDIAN_SWAP(r600_endian_swap(32)) | 1817 S_030008_STRIDE(vb->stride) | 1818 S_030008_BASE_ADDRESS_HI(va >> 32UL)); 1819 radeon_emit(cs, /* RESOURCEi_WORD3 */ 1820 S_03000C_DST_SEL_X(V_03000C_SQ_SEL_X) | 1821 S_03000C_DST_SEL_Y(V_03000C_SQ_SEL_Y) | 1822 S_03000C_DST_SEL_Z(V_03000C_SQ_SEL_Z) | 1823 S_03000C_DST_SEL_W(V_03000C_SQ_SEL_W)); 1824 radeon_emit(cs, 0); /* RESOURCEi_WORD4 */ 1825 radeon_emit(cs, 0); /* RESOURCEi_WORD5 */ 1826 radeon_emit(cs, 0); /* RESOURCEi_WORD6 */ 1827 radeon_emit(cs, 0xc0000000); /* RESOURCEi_WORD7 */ 1828 1829 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0) | pkt_flags); 1830 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer, 1831 RADEON_USAGE_READ, RADEON_PRIO_VERTEX_BUFFER)); 1832 } 1833 state->dirty_mask = 0; 1834 } 1835 1836 static void evergreen_fs_emit_vertex_buffers(struct r600_context *rctx, struct r600_atom * atom) 1837 { 1838 evergreen_emit_vertex_buffers(rctx, &rctx->vertex_buffer_state, EG_FETCH_CONSTANTS_OFFSET_FS, 0); 1839 } 1840 1841 static void evergreen_cs_emit_vertex_buffers(struct r600_context *rctx, struct r600_atom * atom) 1842 { 1843 evergreen_emit_vertex_buffers(rctx, &rctx->cs_vertex_buffer_state, EG_FETCH_CONSTANTS_OFFSET_CS, 1844 RADEON_CP_PACKET3_COMPUTE_MODE); 1845 } 1846 1847 static void evergreen_emit_constant_buffers(struct r600_context *rctx, 1848 struct r600_constbuf_state *state, 1849 unsigned buffer_id_base, 1850 unsigned reg_alu_constbuf_size, 1851 unsigned reg_alu_const_cache, 1852 unsigned pkt_flags) 1853 { 1854 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 1855 uint32_t dirty_mask = state->dirty_mask; 1856 1857 while (dirty_mask) { 1858 struct pipe_constant_buffer *cb; 1859 struct r600_resource *rbuffer; 1860 uint64_t va; 1861 unsigned buffer_index = ffs(dirty_mask) - 1; 1862 unsigned gs_ring_buffer = (buffer_index == R600_GS_RING_CONST_BUFFER); 1863 1864 cb = &state->cb[buffer_index]; 1865 rbuffer = (struct r600_resource*)cb->buffer; 1866 assert(rbuffer); 1867 1868 va = rbuffer->gpu_address + cb->buffer_offset; 1869 1870 if (!gs_ring_buffer) { 1871 radeon_set_context_reg_flag(cs, reg_alu_constbuf_size + buffer_index * 4, 1872 DIV_ROUND_UP(cb->buffer_size, 256), pkt_flags); 1873 radeon_set_context_reg_flag(cs, reg_alu_const_cache + buffer_index * 4, va >> 8, 1874 pkt_flags); 1875 } 1876 1877 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0) | pkt_flags); 1878 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer, 1879 RADEON_USAGE_READ, RADEON_PRIO_CONST_BUFFER)); 1880 1881 radeon_emit(cs, PKT3(PKT3_SET_RESOURCE, 8, 0) | pkt_flags); 1882 radeon_emit(cs, (buffer_id_base + buffer_index) * 8); 1883 radeon_emit(cs, va); /* RESOURCEi_WORD0 */ 1884 radeon_emit(cs, rbuffer->b.b.width0 - cb->buffer_offset - 1); /* RESOURCEi_WORD1 */ 1885 radeon_emit(cs, /* RESOURCEi_WORD2 */ 1886 S_030008_ENDIAN_SWAP(gs_ring_buffer ? ENDIAN_NONE : r600_endian_swap(32)) | 1887 S_030008_STRIDE(gs_ring_buffer ? 4 : 16) | 1888 S_030008_BASE_ADDRESS_HI(va >> 32UL) | 1889 S_030008_DATA_FORMAT(FMT_32_32_32_32_FLOAT)); 1890 radeon_emit(cs, /* RESOURCEi_WORD3 */ 1891 S_03000C_UNCACHED(gs_ring_buffer ? 1 : 0) | 1892 S_03000C_DST_SEL_X(V_03000C_SQ_SEL_X) | 1893 S_03000C_DST_SEL_Y(V_03000C_SQ_SEL_Y) | 1894 S_03000C_DST_SEL_Z(V_03000C_SQ_SEL_Z) | 1895 S_03000C_DST_SEL_W(V_03000C_SQ_SEL_W)); 1896 radeon_emit(cs, 0); /* RESOURCEi_WORD4 */ 1897 radeon_emit(cs, 0); /* RESOURCEi_WORD5 */ 1898 radeon_emit(cs, 0); /* RESOURCEi_WORD6 */ 1899 radeon_emit(cs, /* RESOURCEi_WORD7 */ 1900 S_03001C_TYPE(V_03001C_SQ_TEX_VTX_VALID_BUFFER)); 1901 1902 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0) | pkt_flags); 1903 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer, 1904 RADEON_USAGE_READ, RADEON_PRIO_CONST_BUFFER)); 1905 1906 dirty_mask &= ~(1 << buffer_index); 1907 } 1908 state->dirty_mask = 0; 1909 } 1910 1911 /* VS constants can be in VS/ES (same space) or LS if tess is enabled */ 1912 static void evergreen_emit_vs_constant_buffers(struct r600_context *rctx, struct r600_atom *atom) 1913 { 1914 if (rctx->vs_shader->current->shader.vs_as_ls) { 1915 evergreen_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_VERTEX], 1916 EG_FETCH_CONSTANTS_OFFSET_LS, 1917 R_028FC0_ALU_CONST_BUFFER_SIZE_LS_0, 1918 R_028F40_ALU_CONST_CACHE_LS_0, 1919 0 /* PKT3 flags */); 1920 } else { 1921 evergreen_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_VERTEX], 1922 EG_FETCH_CONSTANTS_OFFSET_VS, 1923 R_028180_ALU_CONST_BUFFER_SIZE_VS_0, 1924 R_028980_ALU_CONST_CACHE_VS_0, 1925 0 /* PKT3 flags */); 1926 } 1927 } 1928 1929 static void evergreen_emit_gs_constant_buffers(struct r600_context *rctx, struct r600_atom *atom) 1930 { 1931 evergreen_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_GEOMETRY], 1932 EG_FETCH_CONSTANTS_OFFSET_GS, 1933 R_0281C0_ALU_CONST_BUFFER_SIZE_GS_0, 1934 R_0289C0_ALU_CONST_CACHE_GS_0, 1935 0 /* PKT3 flags */); 1936 } 1937 1938 static void evergreen_emit_ps_constant_buffers(struct r600_context *rctx, struct r600_atom *atom) 1939 { 1940 evergreen_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_FRAGMENT], 1941 EG_FETCH_CONSTANTS_OFFSET_PS, 1942 R_028140_ALU_CONST_BUFFER_SIZE_PS_0, 1943 R_028940_ALU_CONST_CACHE_PS_0, 1944 0 /* PKT3 flags */); 1945 } 1946 1947 static void evergreen_emit_cs_constant_buffers(struct r600_context *rctx, struct r600_atom *atom) 1948 { 1949 evergreen_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_COMPUTE], 1950 EG_FETCH_CONSTANTS_OFFSET_CS, 1951 R_028FC0_ALU_CONST_BUFFER_SIZE_LS_0, 1952 R_028F40_ALU_CONST_CACHE_LS_0, 1953 RADEON_CP_PACKET3_COMPUTE_MODE); 1954 } 1955 1956 /* tes constants can be emitted to VS or ES - which are common */ 1957 static void evergreen_emit_tes_constant_buffers(struct r600_context *rctx, struct r600_atom *atom) 1958 { 1959 if (!rctx->tes_shader) 1960 return; 1961 evergreen_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_TESS_EVAL], 1962 EG_FETCH_CONSTANTS_OFFSET_VS, 1963 R_028180_ALU_CONST_BUFFER_SIZE_VS_0, 1964 R_028980_ALU_CONST_CACHE_VS_0, 1965 0); 1966 } 1967 1968 static void evergreen_emit_tcs_constant_buffers(struct r600_context *rctx, struct r600_atom *atom) 1969 { 1970 if (!rctx->tes_shader) 1971 return; 1972 evergreen_emit_constant_buffers(rctx, &rctx->constbuf_state[PIPE_SHADER_TESS_CTRL], 1973 EG_FETCH_CONSTANTS_OFFSET_HS, 1974 R_028F80_ALU_CONST_BUFFER_SIZE_HS_0, 1975 R_028F00_ALU_CONST_CACHE_HS_0, 1976 0); 1977 } 1978 1979 static void evergreen_emit_sampler_views(struct r600_context *rctx, 1980 struct r600_samplerview_state *state, 1981 unsigned resource_id_base, unsigned pkt_flags) 1982 { 1983 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 1984 uint32_t dirty_mask = state->dirty_mask; 1985 1986 while (dirty_mask) { 1987 struct r600_pipe_sampler_view *rview; 1988 unsigned resource_index = u_bit_scan(&dirty_mask); 1989 unsigned reloc; 1990 1991 rview = state->views[resource_index]; 1992 assert(rview); 1993 1994 radeon_emit(cs, PKT3(PKT3_SET_RESOURCE, 8, 0) | pkt_flags); 1995 radeon_emit(cs, (resource_id_base + resource_index) * 8); 1996 radeon_emit_array(cs, rview->tex_resource_words, 8); 1997 1998 reloc = radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rview->tex_resource, 1999 RADEON_USAGE_READ, 2000 r600_get_sampler_view_priority(rview->tex_resource)); 2001 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0) | pkt_flags); 2002 radeon_emit(cs, reloc); 2003 2004 if (!rview->skip_mip_address_reloc) { 2005 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0) | pkt_flags); 2006 radeon_emit(cs, reloc); 2007 } 2008 } 2009 state->dirty_mask = 0; 2010 } 2011 2012 static void evergreen_emit_vs_sampler_views(struct r600_context *rctx, struct r600_atom *atom) 2013 { 2014 if (rctx->vs_shader->current->shader.vs_as_ls) { 2015 evergreen_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_VERTEX].views, 2016 EG_FETCH_CONSTANTS_OFFSET_LS + R600_MAX_CONST_BUFFERS, 0); 2017 } else { 2018 evergreen_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_VERTEX].views, 2019 EG_FETCH_CONSTANTS_OFFSET_VS + R600_MAX_CONST_BUFFERS, 0); 2020 } 2021 } 2022 2023 static void evergreen_emit_gs_sampler_views(struct r600_context *rctx, struct r600_atom *atom) 2024 { 2025 evergreen_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY].views, 2026 EG_FETCH_CONSTANTS_OFFSET_GS + R600_MAX_CONST_BUFFERS, 0); 2027 } 2028 2029 static void evergreen_emit_tcs_sampler_views(struct r600_context *rctx, struct r600_atom *atom) 2030 { 2031 evergreen_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_TESS_CTRL].views, 2032 EG_FETCH_CONSTANTS_OFFSET_HS + R600_MAX_CONST_BUFFERS, 0); 2033 } 2034 2035 static void evergreen_emit_tes_sampler_views(struct r600_context *rctx, struct r600_atom *atom) 2036 { 2037 evergreen_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_TESS_EVAL].views, 2038 EG_FETCH_CONSTANTS_OFFSET_VS + R600_MAX_CONST_BUFFERS, 0); 2039 } 2040 2041 static void evergreen_emit_ps_sampler_views(struct r600_context *rctx, struct r600_atom *atom) 2042 { 2043 evergreen_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT].views, 2044 EG_FETCH_CONSTANTS_OFFSET_PS + R600_MAX_CONST_BUFFERS, 0); 2045 } 2046 2047 static void evergreen_emit_cs_sampler_views(struct r600_context *rctx, struct r600_atom *atom) 2048 { 2049 evergreen_emit_sampler_views(rctx, &rctx->samplers[PIPE_SHADER_COMPUTE].views, 2050 EG_FETCH_CONSTANTS_OFFSET_CS + 2, RADEON_CP_PACKET3_COMPUTE_MODE); 2051 } 2052 2053 static void evergreen_emit_sampler_states(struct r600_context *rctx, 2054 struct r600_textures_info *texinfo, 2055 unsigned resource_id_base, 2056 unsigned border_index_reg, 2057 unsigned pkt_flags) 2058 { 2059 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 2060 uint32_t dirty_mask = texinfo->states.dirty_mask; 2061 2062 while (dirty_mask) { 2063 struct r600_pipe_sampler_state *rstate; 2064 unsigned i = u_bit_scan(&dirty_mask); 2065 2066 rstate = texinfo->states.states[i]; 2067 assert(rstate); 2068 2069 radeon_emit(cs, PKT3(PKT3_SET_SAMPLER, 3, 0) | pkt_flags); 2070 radeon_emit(cs, (resource_id_base + i) * 3); 2071 radeon_emit_array(cs, rstate->tex_sampler_words, 3); 2072 2073 if (rstate->border_color_use) { 2074 radeon_set_config_reg_seq(cs, border_index_reg, 5); 2075 radeon_emit(cs, i); 2076 radeon_emit_array(cs, rstate->border_color.ui, 4); 2077 } 2078 } 2079 texinfo->states.dirty_mask = 0; 2080 } 2081 2082 static void evergreen_emit_vs_sampler_states(struct r600_context *rctx, struct r600_atom *atom) 2083 { 2084 if (rctx->vs_shader->current->shader.vs_as_ls) { 2085 evergreen_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_VERTEX], 72, 2086 R_00A450_TD_LS_SAMPLER0_BORDER_COLOR_INDEX, 0); 2087 } else { 2088 evergreen_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_VERTEX], 18, 2089 R_00A414_TD_VS_SAMPLER0_BORDER_INDEX, 0); 2090 } 2091 } 2092 2093 static void evergreen_emit_gs_sampler_states(struct r600_context *rctx, struct r600_atom *atom) 2094 { 2095 evergreen_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY], 36, 2096 R_00A428_TD_GS_SAMPLER0_BORDER_INDEX, 0); 2097 } 2098 2099 static void evergreen_emit_tcs_sampler_states(struct r600_context *rctx, struct r600_atom *atom) 2100 { 2101 evergreen_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_TESS_CTRL], 54, 2102 R_00A43C_TD_HS_SAMPLER0_BORDER_COLOR_INDEX, 0); 2103 } 2104 2105 static void evergreen_emit_tes_sampler_states(struct r600_context *rctx, struct r600_atom *atom) 2106 { 2107 evergreen_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_TESS_EVAL], 18, 2108 R_00A414_TD_VS_SAMPLER0_BORDER_INDEX, 0); 2109 } 2110 2111 static void evergreen_emit_ps_sampler_states(struct r600_context *rctx, struct r600_atom *atom) 2112 { 2113 evergreen_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT], 0, 2114 R_00A400_TD_PS_SAMPLER0_BORDER_INDEX, 0); 2115 } 2116 2117 static void evergreen_emit_cs_sampler_states(struct r600_context *rctx, struct r600_atom *atom) 2118 { 2119 evergreen_emit_sampler_states(rctx, &rctx->samplers[PIPE_SHADER_COMPUTE], 90, 2120 R_00A464_TD_CS_SAMPLER0_BORDER_INDEX, 2121 RADEON_CP_PACKET3_COMPUTE_MODE); 2122 } 2123 2124 static void evergreen_emit_sample_mask(struct r600_context *rctx, struct r600_atom *a) 2125 { 2126 struct r600_sample_mask *s = (struct r600_sample_mask*)a; 2127 uint8_t mask = s->sample_mask; 2128 2129 radeon_set_context_reg(rctx->b.gfx.cs, R_028C3C_PA_SC_AA_MASK, 2130 mask | (mask << 8) | (mask << 16) | (mask << 24)); 2131 } 2132 2133 static void cayman_emit_sample_mask(struct r600_context *rctx, struct r600_atom *a) 2134 { 2135 struct r600_sample_mask *s = (struct r600_sample_mask*)a; 2136 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 2137 uint16_t mask = s->sample_mask; 2138 2139 radeon_set_context_reg_seq(cs, CM_R_028C38_PA_SC_AA_MASK_X0Y0_X1Y0, 2); 2140 radeon_emit(cs, mask | (mask << 16)); /* X0Y0_X1Y0 */ 2141 radeon_emit(cs, mask | (mask << 16)); /* X0Y1_X1Y1 */ 2142 } 2143 2144 static void evergreen_emit_vertex_fetch_shader(struct r600_context *rctx, struct r600_atom *a) 2145 { 2146 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 2147 struct r600_cso_state *state = (struct r600_cso_state*)a; 2148 struct r600_fetch_shader *shader = (struct r600_fetch_shader*)state->cso; 2149 2150 radeon_set_context_reg(cs, R_0288A4_SQ_PGM_START_FS, 2151 (shader->buffer->gpu_address + shader->offset) >> 8); 2152 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); 2153 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, shader->buffer, 2154 RADEON_USAGE_READ, 2155 RADEON_PRIO_SHADER_BINARY)); 2156 } 2157 2158 static void evergreen_emit_shader_stages(struct r600_context *rctx, struct r600_atom *a) 2159 { 2160 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 2161 struct r600_shader_stages_state *state = (struct r600_shader_stages_state*)a; 2162 2163 uint32_t v = 0, v2 = 0, primid = 0, tf_param = 0; 2164 2165 if (rctx->vs_shader->current->shader.vs_as_gs_a) { 2166 v2 = S_028A40_MODE(V_028A40_GS_SCENARIO_A); 2167 primid = 1; 2168 } 2169 2170 if (state->geom_enable) { 2171 uint32_t cut_val; 2172 2173 if (rctx->gs_shader->gs_max_out_vertices <= 128) 2174 cut_val = V_028A40_GS_CUT_128; 2175 else if (rctx->gs_shader->gs_max_out_vertices <= 256) 2176 cut_val = V_028A40_GS_CUT_256; 2177 else if (rctx->gs_shader->gs_max_out_vertices <= 512) 2178 cut_val = V_028A40_GS_CUT_512; 2179 else 2180 cut_val = V_028A40_GS_CUT_1024; 2181 2182 v = S_028B54_GS_EN(1) | 2183 S_028B54_VS_EN(V_028B54_VS_STAGE_COPY_SHADER); 2184 if (!rctx->tes_shader) 2185 v |= S_028B54_ES_EN(V_028B54_ES_STAGE_REAL); 2186 2187 v2 = S_028A40_MODE(V_028A40_GS_SCENARIO_G) | 2188 S_028A40_CUT_MODE(cut_val); 2189 2190 if (rctx->gs_shader->current->shader.gs_prim_id_input) 2191 primid = 1; 2192 } 2193 2194 if (rctx->tes_shader) { 2195 uint32_t type, partitioning, topology; 2196 struct tgsi_shader_info *info = &rctx->tes_shader->current->selector->info; 2197 unsigned tes_prim_mode = info->properties[TGSI_PROPERTY_TES_PRIM_MODE]; 2198 unsigned tes_spacing = info->properties[TGSI_PROPERTY_TES_SPACING]; 2199 bool tes_vertex_order_cw = info->properties[TGSI_PROPERTY_TES_VERTEX_ORDER_CW]; 2200 bool tes_point_mode = info->properties[TGSI_PROPERTY_TES_POINT_MODE]; 2201 switch (tes_prim_mode) { 2202 case PIPE_PRIM_LINES: 2203 type = V_028B6C_TESS_ISOLINE; 2204 break; 2205 case PIPE_PRIM_TRIANGLES: 2206 type = V_028B6C_TESS_TRIANGLE; 2207 break; 2208 case PIPE_PRIM_QUADS: 2209 type = V_028B6C_TESS_QUAD; 2210 break; 2211 default: 2212 assert(0); 2213 return; 2214 } 2215 2216 switch (tes_spacing) { 2217 case PIPE_TESS_SPACING_FRACTIONAL_ODD: 2218 partitioning = V_028B6C_PART_FRAC_ODD; 2219 break; 2220 case PIPE_TESS_SPACING_FRACTIONAL_EVEN: 2221 partitioning = V_028B6C_PART_FRAC_EVEN; 2222 break; 2223 case PIPE_TESS_SPACING_EQUAL: 2224 partitioning = V_028B6C_PART_INTEGER; 2225 break; 2226 default: 2227 assert(0); 2228 return; 2229 } 2230 2231 if (tes_point_mode) 2232 topology = V_028B6C_OUTPUT_POINT; 2233 else if (tes_prim_mode == PIPE_PRIM_LINES) 2234 topology = V_028B6C_OUTPUT_LINE; 2235 else if (tes_vertex_order_cw) 2236 /* XXX follow radeonsi and invert */ 2237 topology = V_028B6C_OUTPUT_TRIANGLE_CCW; 2238 else 2239 topology = V_028B6C_OUTPUT_TRIANGLE_CW; 2240 2241 tf_param = S_028B6C_TYPE(type) | 2242 S_028B6C_PARTITIONING(partitioning) | 2243 S_028B6C_TOPOLOGY(topology); 2244 } 2245 2246 if (rctx->tes_shader) { 2247 v |= S_028B54_LS_EN(V_028B54_LS_STAGE_ON) | 2248 S_028B54_HS_EN(1); 2249 if (!state->geom_enable) 2250 v |= S_028B54_VS_EN(V_028B54_VS_STAGE_DS); 2251 else 2252 v |= S_028B54_ES_EN(V_028B54_ES_STAGE_DS); 2253 } 2254 2255 radeon_set_context_reg(cs, R_028AB8_VGT_VTX_CNT_EN, v ? 1 : 0 ); 2256 radeon_set_context_reg(cs, R_028B54_VGT_SHADER_STAGES_EN, v); 2257 radeon_set_context_reg(cs, R_028A40_VGT_GS_MODE, v2); 2258 radeon_set_context_reg(cs, R_028A84_VGT_PRIMITIVEID_EN, primid); 2259 radeon_set_context_reg(cs, R_028B6C_VGT_TF_PARAM, tf_param); 2260 } 2261 2262 static void evergreen_emit_gs_rings(struct r600_context *rctx, struct r600_atom *a) 2263 { 2264 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 2265 struct r600_gs_rings_state *state = (struct r600_gs_rings_state*)a; 2266 struct r600_resource *rbuffer; 2267 2268 radeon_set_config_reg(cs, R_008040_WAIT_UNTIL, S_008040_WAIT_3D_IDLE(1)); 2269 radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0)); 2270 radeon_emit(cs, EVENT_TYPE(EVENT_TYPE_VGT_FLUSH)); 2271 2272 if (state->enable) { 2273 rbuffer =(struct r600_resource*)state->esgs_ring.buffer; 2274 radeon_set_config_reg(cs, R_008C40_SQ_ESGS_RING_BASE, 2275 rbuffer->gpu_address >> 8); 2276 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); 2277 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer, 2278 RADEON_USAGE_READWRITE, 2279 RADEON_PRIO_SHADER_RINGS)); 2280 radeon_set_config_reg(cs, R_008C44_SQ_ESGS_RING_SIZE, 2281 state->esgs_ring.buffer_size >> 8); 2282 2283 rbuffer =(struct r600_resource*)state->gsvs_ring.buffer; 2284 radeon_set_config_reg(cs, R_008C48_SQ_GSVS_RING_BASE, 2285 rbuffer->gpu_address >> 8); 2286 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); 2287 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, rbuffer, 2288 RADEON_USAGE_READWRITE, 2289 RADEON_PRIO_SHADER_RINGS)); 2290 radeon_set_config_reg(cs, R_008C4C_SQ_GSVS_RING_SIZE, 2291 state->gsvs_ring.buffer_size >> 8); 2292 } else { 2293 radeon_set_config_reg(cs, R_008C44_SQ_ESGS_RING_SIZE, 0); 2294 radeon_set_config_reg(cs, R_008C4C_SQ_GSVS_RING_SIZE, 0); 2295 } 2296 2297 radeon_set_config_reg(cs, R_008040_WAIT_UNTIL, S_008040_WAIT_3D_IDLE(1)); 2298 radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0)); 2299 radeon_emit(cs, EVENT_TYPE(EVENT_TYPE_VGT_FLUSH)); 2300 } 2301 2302 void cayman_init_common_regs(struct r600_command_buffer *cb, 2303 enum chip_class ctx_chip_class, 2304 enum radeon_family ctx_family, 2305 int ctx_drm_minor) 2306 { 2307 r600_store_config_reg_seq(cb, R_008C00_SQ_CONFIG, 2); 2308 r600_store_value(cb, S_008C00_EXPORT_SRC_C(1)); /* R_008C00_SQ_CONFIG */ 2309 /* always set the temp clauses */ 2310 r600_store_value(cb, S_008C04_NUM_CLAUSE_TEMP_GPRS(4)); /* R_008C04_SQ_GPR_RESOURCE_MGMT_1 */ 2311 2312 r600_store_config_reg_seq(cb, R_008C10_SQ_GLOBAL_GPR_RESOURCE_MGMT_1, 2); 2313 r600_store_value(cb, 0); /* R_008C10_SQ_GLOBAL_GPR_RESOURCE_MGMT_1 */ 2314 r600_store_value(cb, 0); /* R_008C14_SQ_GLOBAL_GPR_RESOURCE_MGMT_2 */ 2315 2316 r600_store_config_reg(cb, R_008D8C_SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, (1 << 8)); 2317 2318 r600_store_context_reg_seq(cb, R_028350_SX_MISC, 2); 2319 r600_store_value(cb, 0); 2320 r600_store_value(cb, S_028354_SURFACE_SYNC_MASK(0xf)); 2321 2322 r600_store_context_reg(cb, R_028800_DB_DEPTH_CONTROL, 0); 2323 } 2324 2325 static void cayman_init_atom_start_cs(struct r600_context *rctx) 2326 { 2327 struct r600_command_buffer *cb = &rctx->start_cs_cmd; 2328 int i; 2329 2330 r600_init_command_buffer(cb, 338); 2331 2332 /* This must be first. */ 2333 r600_store_value(cb, PKT3(PKT3_CONTEXT_CONTROL, 1, 0)); 2334 r600_store_value(cb, 0x80000000); 2335 r600_store_value(cb, 0x80000000); 2336 2337 /* We're setting config registers here. */ 2338 r600_store_value(cb, PKT3(PKT3_EVENT_WRITE, 0, 0)); 2339 r600_store_value(cb, EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4)); 2340 2341 /* This enables pipeline stat & streamout queries. 2342 * They are only disabled by blits. 2343 */ 2344 r600_store_value(cb, PKT3(PKT3_EVENT_WRITE, 0, 0)); 2345 r600_store_value(cb, EVENT_TYPE(EVENT_TYPE_PIPELINESTAT_START) | EVENT_INDEX(0)); 2346 2347 cayman_init_common_regs(cb, rctx->b.chip_class, 2348 rctx->b.family, rctx->screen->b.info.drm_minor); 2349 2350 r600_store_config_reg(cb, R_009100_SPI_CONFIG_CNTL, 0); 2351 r600_store_config_reg(cb, R_00913C_SPI_CONFIG_CNTL_1, S_00913C_VTX_DONE_DELAY(4)); 2352 2353 /* remove LS/HS from one SIMD for hw workaround */ 2354 r600_store_config_reg_seq(cb, R_008E20_SQ_STATIC_THREAD_MGMT1, 3); 2355 r600_store_value(cb, 0xffffffff); 2356 r600_store_value(cb, 0xffffffff); 2357 r600_store_value(cb, 0xfffffffe); 2358 2359 r600_store_context_reg_seq(cb, R_028900_SQ_ESGS_RING_ITEMSIZE, 6); 2360 r600_store_value(cb, 0); /* R_028900_SQ_ESGS_RING_ITEMSIZE */ 2361 r600_store_value(cb, 0); /* R_028904_SQ_GSVS_RING_ITEMSIZE */ 2362 r600_store_value(cb, 0); /* R_028908_SQ_ESTMP_RING_ITEMSIZE */ 2363 r600_store_value(cb, 0); /* R_02890C_SQ_GSTMP_RING_ITEMSIZE */ 2364 r600_store_value(cb, 0); /* R_028910_SQ_VSTMP_RING_ITEMSIZE */ 2365 r600_store_value(cb, 0); /* R_028914_SQ_PSTMP_RING_ITEMSIZE */ 2366 2367 r600_store_context_reg_seq(cb, R_02891C_SQ_GS_VERT_ITEMSIZE, 4); 2368 r600_store_value(cb, 0); /* R_02891C_SQ_GS_VERT_ITEMSIZE */ 2369 r600_store_value(cb, 0); /* R_028920_SQ_GS_VERT_ITEMSIZE_1 */ 2370 r600_store_value(cb, 0); /* R_028924_SQ_GS_VERT_ITEMSIZE_2 */ 2371 r600_store_value(cb, 0); /* R_028928_SQ_GS_VERT_ITEMSIZE_3 */ 2372 2373 r600_store_context_reg_seq(cb, R_028A10_VGT_OUTPUT_PATH_CNTL, 13); 2374 r600_store_value(cb, 0); /* R_028A10_VGT_OUTPUT_PATH_CNTL */ 2375 r600_store_value(cb, 0); /* R_028A14_VGT_HOS_CNTL */ 2376 r600_store_value(cb, fui(64)); /* R_028A18_VGT_HOS_MAX_TESS_LEVEL */ 2377 r600_store_value(cb, fui(0)); /* R_028A1C_VGT_HOS_MIN_TESS_LEVEL */ 2378 r600_store_value(cb, 16); /* R_028A20_VGT_HOS_REUSE_DEPTH */ 2379 r600_store_value(cb, 0); /* R_028A24_VGT_GROUP_PRIM_TYPE */ 2380 r600_store_value(cb, 0); /* R_028A28_VGT_GROUP_FIRST_DECR */ 2381 r600_store_value(cb, 0); /* R_028A2C_VGT_GROUP_DECR */ 2382 r600_store_value(cb, 0); /* R_028A30_VGT_GROUP_VECT_0_CNTL */ 2383 r600_store_value(cb, 0); /* R_028A34_VGT_GROUP_VECT_1_CNTL */ 2384 r600_store_value(cb, 0); /* R_028A38_VGT_GROUP_VECT_0_FMT_CNTL */ 2385 r600_store_value(cb, 0); /* R_028A3C_VGT_GROUP_VECT_1_FMT_CNTL */ 2386 r600_store_value(cb, 0); /* R_028A40_VGT_GS_MODE */ 2387 2388 r600_store_context_reg(cb, R_028B98_VGT_STRMOUT_BUFFER_CONFIG, 0); 2389 2390 r600_store_config_reg(cb, R_008A14_PA_CL_ENHANCE, (3 << 1) | 1); 2391 2392 r600_store_context_reg_seq(cb, CM_R_028BD4_PA_SC_CENTROID_PRIORITY_0, 2); 2393 r600_store_value(cb, 0x76543210); /* CM_R_028BD4_PA_SC_CENTROID_PRIORITY_0 */ 2394 r600_store_value(cb, 0xfedcba98); /* CM_R_028BD8_PA_SC_CENTROID_PRIORITY_1 */ 2395 2396 r600_store_context_reg_seq(cb, R_0288E8_SQ_LDS_ALLOC, 2); 2397 r600_store_value(cb, 0); /* R_0288E8_SQ_LDS_ALLOC */ 2398 r600_store_value(cb, 0); /* R_0288EC_SQ_LDS_ALLOC_PS */ 2399 2400 r600_store_context_reg(cb, R_0288F0_SQ_VTX_SEMANTIC_CLEAR, ~0); 2401 2402 r600_store_context_reg_seq(cb, R_028400_VGT_MAX_VTX_INDX, 2); 2403 r600_store_value(cb, ~0); /* R_028400_VGT_MAX_VTX_INDX */ 2404 r600_store_value(cb, 0); /* R_028404_VGT_MIN_VTX_INDX */ 2405 2406 r600_store_ctl_const(cb, R_03CFF0_SQ_VTX_BASE_VTX_LOC, 0); 2407 2408 r600_store_context_reg(cb, R_028028_DB_STENCIL_CLEAR, 0); 2409 2410 r600_store_context_reg(cb, R_0286DC_SPI_FOG_CNTL, 0); 2411 2412 r600_store_context_reg_seq(cb, R_028AC0_DB_SRESULTS_COMPARE_STATE0, 3); 2413 r600_store_value(cb, 0); /* R_028AC0_DB_SRESULTS_COMPARE_STATE0 */ 2414 r600_store_value(cb, 0); /* R_028AC4_DB_SRESULTS_COMPARE_STATE1 */ 2415 r600_store_value(cb, 0); /* R_028AC8_DB_PRELOAD_CONTROL */ 2416 2417 r600_store_context_reg(cb, R_028200_PA_SC_WINDOW_OFFSET, 0); 2418 r600_store_context_reg(cb, R_02820C_PA_SC_CLIPRECT_RULE, 0xFFFF); 2419 2420 r600_store_context_reg(cb, R_028230_PA_SC_EDGERULE, 0xAAAAAAAA); 2421 r600_store_context_reg(cb, R_028820_PA_CL_NANINF_CNTL, 0); 2422 2423 r600_store_context_reg_seq(cb, R_028240_PA_SC_GENERIC_SCISSOR_TL, 2); 2424 r600_store_value(cb, 0); /* R_028240_PA_SC_GENERIC_SCISSOR_TL */ 2425 r600_store_value(cb, S_028244_BR_X(16384) | S_028244_BR_Y(16384)); /* R_028244_PA_SC_GENERIC_SCISSOR_BR */ 2426 2427 r600_store_context_reg_seq(cb, R_028030_PA_SC_SCREEN_SCISSOR_TL, 2); 2428 r600_store_value(cb, 0); /* R_028030_PA_SC_SCREEN_SCISSOR_TL */ 2429 r600_store_value(cb, S_028034_BR_X(16384) | S_028034_BR_Y(16384)); /* R_028034_PA_SC_SCREEN_SCISSOR_BR */ 2430 2431 r600_store_context_reg(cb, R_028848_SQ_PGM_RESOURCES_2_PS, S_028848_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN)); 2432 r600_store_context_reg(cb, R_028864_SQ_PGM_RESOURCES_2_VS, S_028864_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN)); 2433 r600_store_context_reg(cb, R_02887C_SQ_PGM_RESOURCES_2_GS, S_028848_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN)); 2434 r600_store_context_reg(cb, R_028894_SQ_PGM_RESOURCES_2_ES, S_028848_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN)); 2435 r600_store_context_reg(cb, R_0288C0_SQ_PGM_RESOURCES_2_HS, S_028848_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN)); 2436 r600_store_context_reg(cb, R_0288D8_SQ_PGM_RESOURCES_2_LS, S_028848_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN)); 2437 2438 r600_store_context_reg(cb, R_0288A8_SQ_PGM_RESOURCES_FS, 0); 2439 2440 /* to avoid GPU doing any preloading of constant from random address */ 2441 r600_store_context_reg_seq(cb, R_028140_ALU_CONST_BUFFER_SIZE_PS_0, 16); 2442 for (i = 0; i < 16; i++) 2443 r600_store_value(cb, 0); 2444 2445 r600_store_context_reg_seq(cb, R_028180_ALU_CONST_BUFFER_SIZE_VS_0, 16); 2446 for (i = 0; i < 16; i++) 2447 r600_store_value(cb, 0); 2448 2449 r600_store_context_reg_seq(cb, R_0281C0_ALU_CONST_BUFFER_SIZE_GS_0, 16); 2450 for (i = 0; i < 16; i++) 2451 r600_store_value(cb, 0); 2452 2453 r600_store_context_reg_seq(cb, R_028FC0_ALU_CONST_BUFFER_SIZE_LS_0, 16); 2454 for (i = 0; i < 16; i++) 2455 r600_store_value(cb, 0); 2456 2457 r600_store_context_reg_seq(cb, R_028F80_ALU_CONST_BUFFER_SIZE_HS_0, 16); 2458 for (i = 0; i < 16; i++) 2459 r600_store_value(cb, 0); 2460 2461 if (rctx->screen->b.has_streamout) { 2462 r600_store_context_reg(cb, R_028B28_VGT_STRMOUT_DRAW_OPAQUE_OFFSET, 0); 2463 } 2464 2465 r600_store_context_reg(cb, R_028010_DB_RENDER_OVERRIDE2, 0); 2466 r600_store_context_reg(cb, R_028234_PA_SU_HARDWARE_SCREEN_OFFSET, 0); 2467 r600_store_context_reg(cb, R_0286C8_SPI_THREAD_GROUPING, 0); 2468 r600_store_context_reg_seq(cb, R_0286E4_SPI_PS_IN_CONTROL_2, 2); 2469 r600_store_value(cb, 0); /* R_0286E4_SPI_PS_IN_CONTROL_2 */ 2470 r600_store_value(cb, 0); /* R_0286E8_SPI_COMPUTE_INPUT_CNTL */ 2471 2472 r600_store_context_reg_seq(cb, R_028B54_VGT_SHADER_STAGES_EN, 2); 2473 r600_store_value(cb, 0); /* R028B54_VGT_SHADER_STAGES_EN */ 2474 r600_store_value(cb, 0); /* R028B58_VGT_LS_HS_CONFIG */ 2475 r600_store_context_reg(cb, R_028B6C_VGT_TF_PARAM, 0); 2476 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0, 0x01000FFF); 2477 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0 + (32 * 4), 0x01000FFF); 2478 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0 + (64 * 4), 0x01000FFF); 2479 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0 + (96 * 4), 0x01000FFF); 2480 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0 + (128 * 4), 0x01000FFF); 2481 } 2482 2483 void evergreen_init_common_regs(struct r600_context *rctx, struct r600_command_buffer *cb, 2484 enum chip_class ctx_chip_class, 2485 enum radeon_family ctx_family, 2486 int ctx_drm_minor) 2487 { 2488 int ps_prio; 2489 int vs_prio; 2490 int gs_prio; 2491 int es_prio; 2492 2493 int hs_prio; 2494 int cs_prio; 2495 int ls_prio; 2496 2497 unsigned tmp; 2498 2499 ps_prio = 0; 2500 vs_prio = 1; 2501 gs_prio = 2; 2502 es_prio = 3; 2503 hs_prio = 3; 2504 ls_prio = 3; 2505 cs_prio = 0; 2506 2507 rctx->default_gprs[R600_HW_STAGE_PS] = 93; 2508 rctx->default_gprs[R600_HW_STAGE_VS] = 46; 2509 rctx->r6xx_num_clause_temp_gprs = 4; 2510 rctx->default_gprs[R600_HW_STAGE_GS] = 31; 2511 rctx->default_gprs[R600_HW_STAGE_ES] = 31; 2512 rctx->default_gprs[EG_HW_STAGE_HS] = 23; 2513 rctx->default_gprs[EG_HW_STAGE_LS] = 23; 2514 2515 tmp = 0; 2516 switch (ctx_family) { 2517 case CHIP_CEDAR: 2518 case CHIP_PALM: 2519 case CHIP_SUMO: 2520 case CHIP_SUMO2: 2521 case CHIP_CAICOS: 2522 break; 2523 default: 2524 tmp |= S_008C00_VC_ENABLE(1); 2525 break; 2526 } 2527 tmp |= S_008C00_EXPORT_SRC_C(1); 2528 tmp |= S_008C00_CS_PRIO(cs_prio); 2529 tmp |= S_008C00_LS_PRIO(ls_prio); 2530 tmp |= S_008C00_HS_PRIO(hs_prio); 2531 tmp |= S_008C00_PS_PRIO(ps_prio); 2532 tmp |= S_008C00_VS_PRIO(vs_prio); 2533 tmp |= S_008C00_GS_PRIO(gs_prio); 2534 tmp |= S_008C00_ES_PRIO(es_prio); 2535 2536 r600_store_config_reg_seq(cb, R_008C00_SQ_CONFIG, 1); 2537 r600_store_value(cb, tmp); /* R_008C00_SQ_CONFIG */ 2538 2539 r600_store_config_reg_seq(cb, R_008C10_SQ_GLOBAL_GPR_RESOURCE_MGMT_1, 2); 2540 r600_store_value(cb, 0); /* R_008C10_SQ_GLOBAL_GPR_RESOURCE_MGMT_1 */ 2541 r600_store_value(cb, 0); /* R_008C14_SQ_GLOBAL_GPR_RESOURCE_MGMT_2 */ 2542 2543 /* The cs checker requires this register to be set. */ 2544 r600_store_context_reg(cb, R_028800_DB_DEPTH_CONTROL, 0); 2545 2546 r600_store_context_reg_seq(cb, R_028350_SX_MISC, 2); 2547 r600_store_value(cb, 0); 2548 r600_store_value(cb, S_028354_SURFACE_SYNC_MASK(0xf)); 2549 2550 return; 2551 } 2552 2553 void evergreen_init_atom_start_cs(struct r600_context *rctx) 2554 { 2555 struct r600_command_buffer *cb = &rctx->start_cs_cmd; 2556 int num_ps_threads; 2557 int num_vs_threads; 2558 int num_gs_threads; 2559 int num_es_threads; 2560 int num_hs_threads; 2561 int num_ls_threads; 2562 2563 int num_ps_stack_entries; 2564 int num_vs_stack_entries; 2565 int num_gs_stack_entries; 2566 int num_es_stack_entries; 2567 int num_hs_stack_entries; 2568 int num_ls_stack_entries; 2569 enum radeon_family family; 2570 unsigned tmp, i; 2571 2572 if (rctx->b.chip_class == CAYMAN) { 2573 cayman_init_atom_start_cs(rctx); 2574 return; 2575 } 2576 2577 r600_init_command_buffer(cb, 338); 2578 2579 /* This must be first. */ 2580 r600_store_value(cb, PKT3(PKT3_CONTEXT_CONTROL, 1, 0)); 2581 r600_store_value(cb, 0x80000000); 2582 r600_store_value(cb, 0x80000000); 2583 2584 /* We're setting config registers here. */ 2585 r600_store_value(cb, PKT3(PKT3_EVENT_WRITE, 0, 0)); 2586 r600_store_value(cb, EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4)); 2587 2588 /* This enables pipeline stat & streamout queries. 2589 * They are only disabled by blits. 2590 */ 2591 r600_store_value(cb, PKT3(PKT3_EVENT_WRITE, 0, 0)); 2592 r600_store_value(cb, EVENT_TYPE(EVENT_TYPE_PIPELINESTAT_START) | EVENT_INDEX(0)); 2593 2594 evergreen_init_common_regs(rctx, cb, rctx->b.chip_class, 2595 rctx->b.family, rctx->screen->b.info.drm_minor); 2596 2597 family = rctx->b.family; 2598 switch (family) { 2599 case CHIP_CEDAR: 2600 default: 2601 num_ps_threads = 96; 2602 num_vs_threads = 16; 2603 num_gs_threads = 16; 2604 num_es_threads = 16; 2605 num_hs_threads = 16; 2606 num_ls_threads = 16; 2607 num_ps_stack_entries = 42; 2608 num_vs_stack_entries = 42; 2609 num_gs_stack_entries = 42; 2610 num_es_stack_entries = 42; 2611 num_hs_stack_entries = 42; 2612 num_ls_stack_entries = 42; 2613 break; 2614 case CHIP_REDWOOD: 2615 num_ps_threads = 128; 2616 num_vs_threads = 20; 2617 num_gs_threads = 20; 2618 num_es_threads = 20; 2619 num_hs_threads = 20; 2620 num_ls_threads = 20; 2621 num_ps_stack_entries = 42; 2622 num_vs_stack_entries = 42; 2623 num_gs_stack_entries = 42; 2624 num_es_stack_entries = 42; 2625 num_hs_stack_entries = 42; 2626 num_ls_stack_entries = 42; 2627 break; 2628 case CHIP_JUNIPER: 2629 num_ps_threads = 128; 2630 num_vs_threads = 20; 2631 num_gs_threads = 20; 2632 num_es_threads = 20; 2633 num_hs_threads = 20; 2634 num_ls_threads = 20; 2635 num_ps_stack_entries = 85; 2636 num_vs_stack_entries = 85; 2637 num_gs_stack_entries = 85; 2638 num_es_stack_entries = 85; 2639 num_hs_stack_entries = 85; 2640 num_ls_stack_entries = 85; 2641 break; 2642 case CHIP_CYPRESS: 2643 case CHIP_HEMLOCK: 2644 num_ps_threads = 128; 2645 num_vs_threads = 20; 2646 num_gs_threads = 20; 2647 num_es_threads = 20; 2648 num_hs_threads = 20; 2649 num_ls_threads = 20; 2650 num_ps_stack_entries = 85; 2651 num_vs_stack_entries = 85; 2652 num_gs_stack_entries = 85; 2653 num_es_stack_entries = 85; 2654 num_hs_stack_entries = 85; 2655 num_ls_stack_entries = 85; 2656 break; 2657 case CHIP_PALM: 2658 num_ps_threads = 96; 2659 num_vs_threads = 16; 2660 num_gs_threads = 16; 2661 num_es_threads = 16; 2662 num_hs_threads = 16; 2663 num_ls_threads = 16; 2664 num_ps_stack_entries = 42; 2665 num_vs_stack_entries = 42; 2666 num_gs_stack_entries = 42; 2667 num_es_stack_entries = 42; 2668 num_hs_stack_entries = 42; 2669 num_ls_stack_entries = 42; 2670 break; 2671 case CHIP_SUMO: 2672 num_ps_threads = 96; 2673 num_vs_threads = 25; 2674 num_gs_threads = 25; 2675 num_es_threads = 25; 2676 num_hs_threads = 16; 2677 num_ls_threads = 16; 2678 num_ps_stack_entries = 42; 2679 num_vs_stack_entries = 42; 2680 num_gs_stack_entries = 42; 2681 num_es_stack_entries = 42; 2682 num_hs_stack_entries = 42; 2683 num_ls_stack_entries = 42; 2684 break; 2685 case CHIP_SUMO2: 2686 num_ps_threads = 96; 2687 num_vs_threads = 25; 2688 num_gs_threads = 25; 2689 num_es_threads = 25; 2690 num_hs_threads = 16; 2691 num_ls_threads = 16; 2692 num_ps_stack_entries = 85; 2693 num_vs_stack_entries = 85; 2694 num_gs_stack_entries = 85; 2695 num_es_stack_entries = 85; 2696 num_hs_stack_entries = 85; 2697 num_ls_stack_entries = 85; 2698 break; 2699 case CHIP_BARTS: 2700 num_ps_threads = 128; 2701 num_vs_threads = 20; 2702 num_gs_threads = 20; 2703 num_es_threads = 20; 2704 num_hs_threads = 20; 2705 num_ls_threads = 20; 2706 num_ps_stack_entries = 85; 2707 num_vs_stack_entries = 85; 2708 num_gs_stack_entries = 85; 2709 num_es_stack_entries = 85; 2710 num_hs_stack_entries = 85; 2711 num_ls_stack_entries = 85; 2712 break; 2713 case CHIP_TURKS: 2714 num_ps_threads = 128; 2715 num_vs_threads = 20; 2716 num_gs_threads = 20; 2717 num_es_threads = 20; 2718 num_hs_threads = 20; 2719 num_ls_threads = 20; 2720 num_ps_stack_entries = 42; 2721 num_vs_stack_entries = 42; 2722 num_gs_stack_entries = 42; 2723 num_es_stack_entries = 42; 2724 num_hs_stack_entries = 42; 2725 num_ls_stack_entries = 42; 2726 break; 2727 case CHIP_CAICOS: 2728 num_ps_threads = 96; 2729 num_vs_threads = 10; 2730 num_gs_threads = 10; 2731 num_es_threads = 10; 2732 num_hs_threads = 10; 2733 num_ls_threads = 10; 2734 num_ps_stack_entries = 42; 2735 num_vs_stack_entries = 42; 2736 num_gs_stack_entries = 42; 2737 num_es_stack_entries = 42; 2738 num_hs_stack_entries = 42; 2739 num_ls_stack_entries = 42; 2740 break; 2741 } 2742 2743 tmp = S_008C18_NUM_PS_THREADS(num_ps_threads); 2744 tmp |= S_008C18_NUM_VS_THREADS(num_vs_threads); 2745 tmp |= S_008C18_NUM_GS_THREADS(num_gs_threads); 2746 tmp |= S_008C18_NUM_ES_THREADS(num_es_threads); 2747 2748 r600_store_config_reg_seq(cb, R_008C18_SQ_THREAD_RESOURCE_MGMT_1, 5); 2749 r600_store_value(cb, tmp); /* R_008C18_SQ_THREAD_RESOURCE_MGMT_1 */ 2750 2751 tmp = S_008C1C_NUM_HS_THREADS(num_hs_threads); 2752 tmp |= S_008C1C_NUM_LS_THREADS(num_ls_threads); 2753 r600_store_value(cb, tmp); /* R_008C1C_SQ_THREAD_RESOURCE_MGMT_2 */ 2754 2755 tmp = S_008C20_NUM_PS_STACK_ENTRIES(num_ps_stack_entries); 2756 tmp |= S_008C20_NUM_VS_STACK_ENTRIES(num_vs_stack_entries); 2757 r600_store_value(cb, tmp); /* R_008C20_SQ_STACK_RESOURCE_MGMT_1 */ 2758 2759 tmp = S_008C24_NUM_GS_STACK_ENTRIES(num_gs_stack_entries); 2760 tmp |= S_008C24_NUM_ES_STACK_ENTRIES(num_es_stack_entries); 2761 r600_store_value(cb, tmp); /* R_008C24_SQ_STACK_RESOURCE_MGMT_2 */ 2762 2763 tmp = S_008C28_NUM_HS_STACK_ENTRIES(num_hs_stack_entries); 2764 tmp |= S_008C28_NUM_LS_STACK_ENTRIES(num_ls_stack_entries); 2765 r600_store_value(cb, tmp); /* R_008C28_SQ_STACK_RESOURCE_MGMT_3 */ 2766 2767 r600_store_config_reg(cb, R_008E2C_SQ_LDS_RESOURCE_MGMT, 2768 S_008E2C_NUM_PS_LDS(0x1000) | S_008E2C_NUM_LS_LDS(0x1000)); 2769 2770 /* remove LS/HS from one SIMD for hw workaround */ 2771 r600_store_config_reg_seq(cb, R_008E20_SQ_STATIC_THREAD_MGMT1, 3); 2772 r600_store_value(cb, 0xffffffff); 2773 r600_store_value(cb, 0xffffffff); 2774 r600_store_value(cb, 0xfffffffe); 2775 2776 r600_store_config_reg(cb, R_009100_SPI_CONFIG_CNTL, 0); 2777 r600_store_config_reg(cb, R_00913C_SPI_CONFIG_CNTL_1, S_00913C_VTX_DONE_DELAY(4)); 2778 2779 r600_store_context_reg_seq(cb, R_028900_SQ_ESGS_RING_ITEMSIZE, 6); 2780 r600_store_value(cb, 0); /* R_028900_SQ_ESGS_RING_ITEMSIZE */ 2781 r600_store_value(cb, 0); /* R_028904_SQ_GSVS_RING_ITEMSIZE */ 2782 r600_store_value(cb, 0); /* R_028908_SQ_ESTMP_RING_ITEMSIZE */ 2783 r600_store_value(cb, 0); /* R_02890C_SQ_GSTMP_RING_ITEMSIZE */ 2784 r600_store_value(cb, 0); /* R_028910_SQ_VSTMP_RING_ITEMSIZE */ 2785 r600_store_value(cb, 0); /* R_028914_SQ_PSTMP_RING_ITEMSIZE */ 2786 2787 r600_store_context_reg_seq(cb, R_02891C_SQ_GS_VERT_ITEMSIZE, 4); 2788 r600_store_value(cb, 0); /* R_02891C_SQ_GS_VERT_ITEMSIZE */ 2789 r600_store_value(cb, 0); /* R_028920_SQ_GS_VERT_ITEMSIZE_1 */ 2790 r600_store_value(cb, 0); /* R_028924_SQ_GS_VERT_ITEMSIZE_2 */ 2791 r600_store_value(cb, 0); /* R_028928_SQ_GS_VERT_ITEMSIZE_3 */ 2792 2793 r600_store_context_reg_seq(cb, R_028A10_VGT_OUTPUT_PATH_CNTL, 13); 2794 r600_store_value(cb, 0); /* R_028A10_VGT_OUTPUT_PATH_CNTL */ 2795 r600_store_value(cb, 0); /* R_028A14_VGT_HOS_CNTL */ 2796 r600_store_value(cb, fui(64)); /* R_028A18_VGT_HOS_MAX_TESS_LEVEL */ 2797 r600_store_value(cb, fui(1.0)); /* R_028A1C_VGT_HOS_MIN_TESS_LEVEL */ 2798 r600_store_value(cb, 16); /* R_028A20_VGT_HOS_REUSE_DEPTH */ 2799 r600_store_value(cb, 0); /* R_028A24_VGT_GROUP_PRIM_TYPE */ 2800 r600_store_value(cb, 0); /* R_028A28_VGT_GROUP_FIRST_DECR */ 2801 r600_store_value(cb, 0); /* R_028A2C_VGT_GROUP_DECR */ 2802 r600_store_value(cb, 0); /* R_028A30_VGT_GROUP_VECT_0_CNTL */ 2803 r600_store_value(cb, 0); /* R_028A34_VGT_GROUP_VECT_1_CNTL */ 2804 r600_store_value(cb, 0); /* R_028A38_VGT_GROUP_VECT_0_FMT_CNTL */ 2805 r600_store_value(cb, 0); /* R_028A3C_VGT_GROUP_VECT_1_FMT_CNTL */ 2806 r600_store_value(cb, 0); /* R_028A40_VGT_GS_MODE */ 2807 2808 r600_store_config_reg(cb, R_008A14_PA_CL_ENHANCE, (3 << 1) | 1); 2809 2810 r600_store_context_reg(cb, R_0288F0_SQ_VTX_SEMANTIC_CLEAR, ~0); 2811 2812 r600_store_context_reg_seq(cb, R_028400_VGT_MAX_VTX_INDX, 2); 2813 r600_store_value(cb, ~0); /* R_028400_VGT_MAX_VTX_INDX */ 2814 r600_store_value(cb, 0); /* R_028404_VGT_MIN_VTX_INDX */ 2815 2816 r600_store_ctl_const(cb, R_03CFF0_SQ_VTX_BASE_VTX_LOC, 0); 2817 2818 r600_store_context_reg(cb, R_028028_DB_STENCIL_CLEAR, 0); 2819 2820 r600_store_context_reg(cb, R_028200_PA_SC_WINDOW_OFFSET, 0); 2821 r600_store_context_reg(cb, R_02820C_PA_SC_CLIPRECT_RULE, 0xFFFF); 2822 r600_store_context_reg(cb, R_028230_PA_SC_EDGERULE, 0xAAAAAAAA); 2823 2824 r600_store_context_reg(cb, R_0286DC_SPI_FOG_CNTL, 0); 2825 r600_store_context_reg(cb, R_028820_PA_CL_NANINF_CNTL, 0); 2826 2827 r600_store_context_reg_seq(cb, R_028AC0_DB_SRESULTS_COMPARE_STATE0, 3); 2828 r600_store_value(cb, 0); /* R_028AC0_DB_SRESULTS_COMPARE_STATE0 */ 2829 r600_store_value(cb, 0); /* R_028AC4_DB_SRESULTS_COMPARE_STATE1 */ 2830 r600_store_value(cb, 0); /* R_028AC8_DB_PRELOAD_CONTROL */ 2831 2832 r600_store_context_reg_seq(cb, R_028240_PA_SC_GENERIC_SCISSOR_TL, 2); 2833 r600_store_value(cb, 0); /* R_028240_PA_SC_GENERIC_SCISSOR_TL */ 2834 r600_store_value(cb, S_028244_BR_X(16384) | S_028244_BR_Y(16384)); /* R_028244_PA_SC_GENERIC_SCISSOR_BR */ 2835 2836 r600_store_context_reg_seq(cb, R_028030_PA_SC_SCREEN_SCISSOR_TL, 2); 2837 r600_store_value(cb, 0); /* R_028030_PA_SC_SCREEN_SCISSOR_TL */ 2838 r600_store_value(cb, S_028034_BR_X(16384) | S_028034_BR_Y(16384)); /* R_028034_PA_SC_SCREEN_SCISSOR_BR */ 2839 2840 r600_store_context_reg(cb, R_028848_SQ_PGM_RESOURCES_2_PS, S_028848_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN)); 2841 r600_store_context_reg(cb, R_028864_SQ_PGM_RESOURCES_2_VS, S_028864_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN)); 2842 r600_store_context_reg(cb, R_02887C_SQ_PGM_RESOURCES_2_GS, S_028848_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN)); 2843 r600_store_context_reg(cb, R_028894_SQ_PGM_RESOURCES_2_ES, S_028848_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN)); 2844 r600_store_context_reg(cb, R_0288A8_SQ_PGM_RESOURCES_FS, 0); 2845 r600_store_context_reg(cb, R_0288C0_SQ_PGM_RESOURCES_2_HS, S_028848_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN)); 2846 r600_store_context_reg(cb, R_0288D8_SQ_PGM_RESOURCES_2_LS, S_028848_SINGLE_ROUND(V_SQ_ROUND_NEAREST_EVEN)); 2847 2848 /* to avoid GPU doing any preloading of constant from random address */ 2849 r600_store_context_reg_seq(cb, R_028140_ALU_CONST_BUFFER_SIZE_PS_0, 16); 2850 for (i = 0; i < 16; i++) 2851 r600_store_value(cb, 0); 2852 2853 r600_store_context_reg_seq(cb, R_028180_ALU_CONST_BUFFER_SIZE_VS_0, 16); 2854 for (i = 0; i < 16; i++) 2855 r600_store_value(cb, 0); 2856 2857 r600_store_context_reg_seq(cb, R_0281C0_ALU_CONST_BUFFER_SIZE_GS_0, 16); 2858 for (i = 0; i < 16; i++) 2859 r600_store_value(cb, 0); 2860 2861 r600_store_context_reg_seq(cb, R_028FC0_ALU_CONST_BUFFER_SIZE_LS_0, 16); 2862 for (i = 0; i < 16; i++) 2863 r600_store_value(cb, 0); 2864 2865 r600_store_context_reg_seq(cb, R_028F80_ALU_CONST_BUFFER_SIZE_HS_0, 16); 2866 for (i = 0; i < 16; i++) 2867 r600_store_value(cb, 0); 2868 2869 r600_store_context_reg(cb, R_028B98_VGT_STRMOUT_BUFFER_CONFIG, 0); 2870 2871 if (rctx->screen->b.has_streamout) { 2872 r600_store_context_reg(cb, R_028B28_VGT_STRMOUT_DRAW_OPAQUE_OFFSET, 0); 2873 } 2874 2875 r600_store_context_reg(cb, R_028010_DB_RENDER_OVERRIDE2, 0); 2876 r600_store_context_reg(cb, R_028234_PA_SU_HARDWARE_SCREEN_OFFSET, 0); 2877 r600_store_context_reg(cb, R_0286C8_SPI_THREAD_GROUPING, 0); 2878 r600_store_context_reg_seq(cb, R_0286E4_SPI_PS_IN_CONTROL_2, 2); 2879 r600_store_value(cb, 0); /* R_0286E4_SPI_PS_IN_CONTROL_2 */ 2880 r600_store_value(cb, 0); /* R_0286E8_SPI_COMPUTE_INPUT_CNTL */ 2881 2882 r600_store_context_reg_seq(cb, R_0288E8_SQ_LDS_ALLOC, 2); 2883 r600_store_value(cb, 0); /* R_0288E8_SQ_LDS_ALLOC */ 2884 r600_store_value(cb, 0); /* R_0288EC_SQ_LDS_ALLOC_PS */ 2885 2886 if (rctx->b.family == CHIP_CAICOS) { 2887 r600_store_context_reg_seq(cb, R_028B54_VGT_SHADER_STAGES_EN, 2); 2888 r600_store_value(cb, 0); /* R028B54_VGT_SHADER_STAGES_EN */ 2889 r600_store_value(cb, 0); /* R028B58_VGT_LS_HS_CONFIG */ 2890 r600_store_context_reg(cb, R_028B6C_VGT_TF_PARAM, 0); 2891 } else { 2892 r600_store_context_reg_seq(cb, R_028B54_VGT_SHADER_STAGES_EN, 7); 2893 r600_store_value(cb, 0); /* R028B54_VGT_SHADER_STAGES_EN */ 2894 r600_store_value(cb, 0); /* R028B58_VGT_LS_HS_CONFIG */ 2895 r600_store_value(cb, 0); /* R028B5C_VGT_LS_SIZE */ 2896 r600_store_value(cb, 0); /* R028B60_VGT_HS_SIZE */ 2897 r600_store_value(cb, 0); /* R028B64_VGT_LS_HS_ALLOC */ 2898 r600_store_value(cb, 0); /* R028B68_VGT_HS_PATCH_CONST */ 2899 r600_store_value(cb, 0); /* R028B68_VGT_TF_PARAM */ 2900 } 2901 2902 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0, 0x01000FFF); 2903 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0 + (32 * 4), 0x01000FFF); 2904 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0 + (64 * 4), 0x01000FFF); 2905 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0 + (96 * 4), 0x01000FFF); 2906 eg_store_loop_const(cb, R_03A200_SQ_LOOP_CONST_0 + (128 * 4), 0x01000FFF); 2907 } 2908 2909 void evergreen_update_ps_state(struct pipe_context *ctx, struct r600_pipe_shader *shader) 2910 { 2911 struct r600_context *rctx = (struct r600_context *)ctx; 2912 struct r600_command_buffer *cb = &shader->command_buffer; 2913 struct r600_shader *rshader = &shader->shader; 2914 unsigned i, exports_ps, num_cout, spi_ps_in_control_0, spi_input_z, spi_ps_in_control_1, db_shader_control = 0; 2915 int pos_index = -1, face_index = -1, fixed_pt_position_index = -1; 2916 int ninterp = 0; 2917 boolean have_perspective = FALSE, have_linear = FALSE; 2918 static const unsigned spi_baryc_enable_bit[6] = { 2919 S_0286E0_PERSP_SAMPLE_ENA(1), 2920 S_0286E0_PERSP_CENTER_ENA(1), 2921 S_0286E0_PERSP_CENTROID_ENA(1), 2922 S_0286E0_LINEAR_SAMPLE_ENA(1), 2923 S_0286E0_LINEAR_CENTER_ENA(1), 2924 S_0286E0_LINEAR_CENTROID_ENA(1) 2925 }; 2926 unsigned spi_baryc_cntl = 0, sid, tmp, num = 0; 2927 unsigned z_export = 0, stencil_export = 0, mask_export = 0; 2928 unsigned sprite_coord_enable = rctx->rasterizer ? rctx->rasterizer->sprite_coord_enable : 0; 2929 uint32_t spi_ps_input_cntl[32]; 2930 2931 if (!cb->buf) { 2932 r600_init_command_buffer(cb, 64); 2933 } else { 2934 cb->num_dw = 0; 2935 } 2936 2937 for (i = 0; i < rshader->ninput; i++) { 2938 /* evergreen NUM_INTERP only contains values interpolated into the LDS, 2939 POSITION goes via GPRs from the SC so isn't counted */ 2940 if (rshader->input[i].name == TGSI_SEMANTIC_POSITION) 2941 pos_index = i; 2942 else if (rshader->input[i].name == TGSI_SEMANTIC_FACE) { 2943 if (face_index == -1) 2944 face_index = i; 2945 } 2946 else if (rshader->input[i].name == TGSI_SEMANTIC_SAMPLEMASK) { 2947 if (face_index == -1) 2948 face_index = i; /* lives in same register, same enable bit */ 2949 } 2950 else if (rshader->input[i].name == TGSI_SEMANTIC_SAMPLEID) { 2951 fixed_pt_position_index = i; 2952 } 2953 else { 2954 ninterp++; 2955 int k = eg_get_interpolator_index( 2956 rshader->input[i].interpolate, 2957 rshader->input[i].interpolate_location); 2958 if (k >= 0) { 2959 spi_baryc_cntl |= spi_baryc_enable_bit[k]; 2960 have_perspective |= k < 3; 2961 have_linear |= !(k < 3); 2962 } 2963 } 2964 2965 sid = rshader->input[i].spi_sid; 2966 2967 if (sid) { 2968 tmp = S_028644_SEMANTIC(sid); 2969 2970 /* D3D 9 behaviour. GL is undefined */ 2971 if (rshader->input[i].name == TGSI_SEMANTIC_COLOR && rshader->input[i].sid == 0) 2972 tmp |= S_028644_DEFAULT_VAL(3); 2973 2974 if (rshader->input[i].name == TGSI_SEMANTIC_POSITION || 2975 rshader->input[i].interpolate == TGSI_INTERPOLATE_CONSTANT || 2976 (rshader->input[i].interpolate == TGSI_INTERPOLATE_COLOR && 2977 rctx->rasterizer && rctx->rasterizer->flatshade)) { 2978 tmp |= S_028644_FLAT_SHADE(1); 2979 } 2980 2981 if (rshader->input[i].name == TGSI_SEMANTIC_GENERIC && 2982 (sprite_coord_enable & (1 << rshader->input[i].sid))) { 2983 tmp |= S_028644_PT_SPRITE_TEX(1); 2984 } 2985 2986 spi_ps_input_cntl[num++] = tmp; 2987 } 2988 } 2989 2990 r600_store_context_reg_seq(cb, R_028644_SPI_PS_INPUT_CNTL_0, num); 2991 r600_store_array(cb, num, spi_ps_input_cntl); 2992 2993 for (i = 0; i < rshader->noutput; i++) { 2994 if (rshader->output[i].name == TGSI_SEMANTIC_POSITION) 2995 z_export = 1; 2996 if (rshader->output[i].name == TGSI_SEMANTIC_STENCIL) 2997 stencil_export = 1; 2998 if (rshader->output[i].name == TGSI_SEMANTIC_SAMPLEMASK && 2999 rctx->framebuffer.nr_samples > 1 && rctx->ps_iter_samples > 0) 3000 mask_export = 1; 3001 } 3002 if (rshader->uses_kill) 3003 db_shader_control |= S_02880C_KILL_ENABLE(1); 3004 3005 db_shader_control |= S_02880C_Z_EXPORT_ENABLE(z_export); 3006 db_shader_control |= S_02880C_STENCIL_EXPORT_ENABLE(stencil_export); 3007 db_shader_control |= S_02880C_MASK_EXPORT_ENABLE(mask_export); 3008 3009 switch (rshader->ps_conservative_z) { 3010 default: /* fall through */ 3011 case TGSI_FS_DEPTH_LAYOUT_ANY: 3012 db_shader_control |= S_02880C_CONSERVATIVE_Z_EXPORT(V_02880C_EXPORT_ANY_Z); 3013 break; 3014 case TGSI_FS_DEPTH_LAYOUT_GREATER: 3015 db_shader_control |= S_02880C_CONSERVATIVE_Z_EXPORT(V_02880C_EXPORT_GREATER_THAN_Z); 3016 break; 3017 case TGSI_FS_DEPTH_LAYOUT_LESS: 3018 db_shader_control |= S_02880C_CONSERVATIVE_Z_EXPORT(V_02880C_EXPORT_LESS_THAN_Z); 3019 break; 3020 } 3021 3022 exports_ps = 0; 3023 for (i = 0; i < rshader->noutput; i++) { 3024 if (rshader->output[i].name == TGSI_SEMANTIC_POSITION || 3025 rshader->output[i].name == TGSI_SEMANTIC_STENCIL || 3026 rshader->output[i].name == TGSI_SEMANTIC_SAMPLEMASK) 3027 exports_ps |= 1; 3028 } 3029 3030 num_cout = rshader->nr_ps_color_exports; 3031 3032 exports_ps |= S_02884C_EXPORT_COLORS(num_cout); 3033 if (!exports_ps) { 3034 /* always at least export 1 component per pixel */ 3035 exports_ps = 2; 3036 } 3037 shader->nr_ps_color_outputs = num_cout; 3038 if (ninterp == 0) { 3039 ninterp = 1; 3040 have_perspective = TRUE; 3041 } 3042 if (!spi_baryc_cntl) 3043 spi_baryc_cntl |= spi_baryc_enable_bit[0]; 3044 3045 if (!have_perspective && !have_linear) 3046 have_perspective = TRUE; 3047 3048 spi_ps_in_control_0 = S_0286CC_NUM_INTERP(ninterp) | 3049 S_0286CC_PERSP_GRADIENT_ENA(have_perspective) | 3050 S_0286CC_LINEAR_GRADIENT_ENA(have_linear); 3051 spi_input_z = 0; 3052 if (pos_index != -1) { 3053 spi_ps_in_control_0 |= S_0286CC_POSITION_ENA(1) | 3054 S_0286CC_POSITION_CENTROID(rshader->input[pos_index].interpolate_location == TGSI_INTERPOLATE_LOC_CENTROID) | 3055 S_0286CC_POSITION_ADDR(rshader->input[pos_index].gpr); 3056 spi_input_z |= S_0286D8_PROVIDE_Z_TO_SPI(1); 3057 } 3058 3059 spi_ps_in_control_1 = 0; 3060 if (face_index != -1) { 3061 spi_ps_in_control_1 |= S_0286D0_FRONT_FACE_ENA(1) | 3062 S_0286D0_FRONT_FACE_ADDR(rshader->input[face_index].gpr); 3063 } 3064 if (fixed_pt_position_index != -1) { 3065 spi_ps_in_control_1 |= S_0286D0_FIXED_PT_POSITION_ENA(1) | 3066 S_0286D0_FIXED_PT_POSITION_ADDR(rshader->input[fixed_pt_position_index].gpr); 3067 } 3068 3069 r600_store_context_reg_seq(cb, R_0286CC_SPI_PS_IN_CONTROL_0, 2); 3070 r600_store_value(cb, spi_ps_in_control_0); /* R_0286CC_SPI_PS_IN_CONTROL_0 */ 3071 r600_store_value(cb, spi_ps_in_control_1); /* R_0286D0_SPI_PS_IN_CONTROL_1 */ 3072 3073 r600_store_context_reg(cb, R_0286E0_SPI_BARYC_CNTL, spi_baryc_cntl); 3074 r600_store_context_reg(cb, R_0286D8_SPI_INPUT_Z, spi_input_z); 3075 r600_store_context_reg(cb, R_02884C_SQ_PGM_EXPORTS_PS, exports_ps); 3076 3077 r600_store_context_reg_seq(cb, R_028840_SQ_PGM_START_PS, 2); 3078 r600_store_value(cb, shader->bo->gpu_address >> 8); 3079 r600_store_value(cb, /* R_028844_SQ_PGM_RESOURCES_PS */ 3080 S_028844_NUM_GPRS(rshader->bc.ngpr) | 3081 S_028844_PRIME_CACHE_ON_DRAW(1) | 3082 S_028844_STACK_SIZE(rshader->bc.nstack)); 3083 /* After that, the NOP relocation packet must be emitted (shader->bo, RADEON_USAGE_READ). */ 3084 3085 shader->db_shader_control = db_shader_control; 3086 shader->ps_depth_export = z_export | stencil_export | mask_export; 3087 3088 shader->sprite_coord_enable = sprite_coord_enable; 3089 if (rctx->rasterizer) 3090 shader->flatshade = rctx->rasterizer->flatshade; 3091 } 3092 3093 void evergreen_update_es_state(struct pipe_context *ctx, struct r600_pipe_shader *shader) 3094 { 3095 struct r600_command_buffer *cb = &shader->command_buffer; 3096 struct r600_shader *rshader = &shader->shader; 3097 3098 r600_init_command_buffer(cb, 32); 3099 3100 r600_store_context_reg(cb, R_028890_SQ_PGM_RESOURCES_ES, 3101 S_028890_NUM_GPRS(rshader->bc.ngpr) | 3102 S_028890_STACK_SIZE(rshader->bc.nstack)); 3103 r600_store_context_reg(cb, R_02888C_SQ_PGM_START_ES, 3104 shader->bo->gpu_address >> 8); 3105 /* After that, the NOP relocation packet must be emitted (shader->bo, RADEON_USAGE_READ). */ 3106 } 3107 3108 void evergreen_update_gs_state(struct pipe_context *ctx, struct r600_pipe_shader *shader) 3109 { 3110 struct r600_context *rctx = (struct r600_context *)ctx; 3111 struct r600_command_buffer *cb = &shader->command_buffer; 3112 struct r600_shader *rshader = &shader->shader; 3113 struct r600_shader *cp_shader = &shader->gs_copy_shader->shader; 3114 unsigned gsvs_itemsizes[4] = { 3115 (cp_shader->ring_item_sizes[0] * shader->selector->gs_max_out_vertices) >> 2, 3116 (cp_shader->ring_item_sizes[1] * shader->selector->gs_max_out_vertices) >> 2, 3117 (cp_shader->ring_item_sizes[2] * shader->selector->gs_max_out_vertices) >> 2, 3118 (cp_shader->ring_item_sizes[3] * shader->selector->gs_max_out_vertices) >> 2 3119 }; 3120 3121 r600_init_command_buffer(cb, 64); 3122 3123 /* VGT_GS_MODE is written by evergreen_emit_shader_stages */ 3124 3125 3126 r600_store_context_reg(cb, R_028B38_VGT_GS_MAX_VERT_OUT, 3127 S_028B38_MAX_VERT_OUT(shader->selector->gs_max_out_vertices)); 3128 r600_store_context_reg(cb, R_028A6C_VGT_GS_OUT_PRIM_TYPE, 3129 r600_conv_prim_to_gs_out(shader->selector->gs_output_prim)); 3130 3131 if (rctx->screen->b.info.drm_minor >= 35) { 3132 r600_store_context_reg(cb, R_028B90_VGT_GS_INSTANCE_CNT, 3133 S_028B90_CNT(MIN2(shader->selector->gs_num_invocations, 127)) | 3134 S_028B90_ENABLE(shader->selector->gs_num_invocations > 0)); 3135 } 3136 r600_store_context_reg_seq(cb, R_02891C_SQ_GS_VERT_ITEMSIZE, 4); 3137 r600_store_value(cb, cp_shader->ring_item_sizes[0] >> 2); 3138 r600_store_value(cb, cp_shader->ring_item_sizes[1] >> 2); 3139 r600_store_value(cb, cp_shader->ring_item_sizes[2] >> 2); 3140 r600_store_value(cb, cp_shader->ring_item_sizes[3] >> 2); 3141 3142 r600_store_context_reg(cb, R_028900_SQ_ESGS_RING_ITEMSIZE, 3143 (rshader->ring_item_sizes[0]) >> 2); 3144 3145 r600_store_context_reg(cb, R_028904_SQ_GSVS_RING_ITEMSIZE, 3146 gsvs_itemsizes[0] + 3147 gsvs_itemsizes[1] + 3148 gsvs_itemsizes[2] + 3149 gsvs_itemsizes[3]); 3150 3151 r600_store_context_reg_seq(cb, R_02892C_SQ_GSVS_RING_OFFSET_1, 3); 3152 r600_store_value(cb, gsvs_itemsizes[0]); 3153 r600_store_value(cb, gsvs_itemsizes[0] + gsvs_itemsizes[1]); 3154 r600_store_value(cb, gsvs_itemsizes[0] + gsvs_itemsizes[1] + gsvs_itemsizes[2]); 3155 3156 /* FIXME calculate these values somehow ??? */ 3157 r600_store_context_reg_seq(cb, R_028A54_GS_PER_ES, 3); 3158 r600_store_value(cb, 0x80); /* GS_PER_ES */ 3159 r600_store_value(cb, 0x100); /* ES_PER_GS */ 3160 r600_store_value(cb, 0x2); /* GS_PER_VS */ 3161 3162 r600_store_context_reg(cb, R_028878_SQ_PGM_RESOURCES_GS, 3163 S_028878_NUM_GPRS(rshader->bc.ngpr) | 3164 S_028878_STACK_SIZE(rshader->bc.nstack)); 3165 r600_store_context_reg(cb, R_028874_SQ_PGM_START_GS, 3166 shader->bo->gpu_address >> 8); 3167 /* After that, the NOP relocation packet must be emitted (shader->bo, RADEON_USAGE_READ). */ 3168 } 3169 3170 3171 void evergreen_update_vs_state(struct pipe_context *ctx, struct r600_pipe_shader *shader) 3172 { 3173 struct r600_command_buffer *cb = &shader->command_buffer; 3174 struct r600_shader *rshader = &shader->shader; 3175 unsigned spi_vs_out_id[10] = {}; 3176 unsigned i, tmp, nparams = 0; 3177 3178 for (i = 0; i < rshader->noutput; i++) { 3179 if (rshader->output[i].spi_sid) { 3180 tmp = rshader->output[i].spi_sid << ((nparams & 3) * 8); 3181 spi_vs_out_id[nparams / 4] |= tmp; 3182 nparams++; 3183 } 3184 } 3185 3186 r600_init_command_buffer(cb, 32); 3187 3188 r600_store_context_reg_seq(cb, R_02861C_SPI_VS_OUT_ID_0, 10); 3189 for (i = 0; i < 10; i++) { 3190 r600_store_value(cb, spi_vs_out_id[i]); 3191 } 3192 3193 /* Certain attributes (position, psize, etc.) don't count as params. 3194 * VS is required to export at least one param and r600_shader_from_tgsi() 3195 * takes care of adding a dummy export. 3196 */ 3197 if (nparams < 1) 3198 nparams = 1; 3199 3200 r600_store_context_reg(cb, R_0286C4_SPI_VS_OUT_CONFIG, 3201 S_0286C4_VS_EXPORT_COUNT(nparams - 1)); 3202 r600_store_context_reg(cb, R_028860_SQ_PGM_RESOURCES_VS, 3203 S_028860_NUM_GPRS(rshader->bc.ngpr) | 3204 S_028860_STACK_SIZE(rshader->bc.nstack)); 3205 if (rshader->vs_position_window_space) { 3206 r600_store_context_reg(cb, R_028818_PA_CL_VTE_CNTL, 3207 S_028818_VTX_XY_FMT(1) | S_028818_VTX_Z_FMT(1)); 3208 } else { 3209 r600_store_context_reg(cb, R_028818_PA_CL_VTE_CNTL, 3210 S_028818_VTX_W0_FMT(1) | 3211 S_028818_VPORT_X_SCALE_ENA(1) | S_028818_VPORT_X_OFFSET_ENA(1) | 3212 S_028818_VPORT_Y_SCALE_ENA(1) | S_028818_VPORT_Y_OFFSET_ENA(1) | 3213 S_028818_VPORT_Z_SCALE_ENA(1) | S_028818_VPORT_Z_OFFSET_ENA(1)); 3214 3215 } 3216 r600_store_context_reg(cb, R_02885C_SQ_PGM_START_VS, 3217 shader->bo->gpu_address >> 8); 3218 /* After that, the NOP relocation packet must be emitted (shader->bo, RADEON_USAGE_READ). */ 3219 3220 shader->pa_cl_vs_out_cntl = 3221 S_02881C_VS_OUT_CCDIST0_VEC_ENA((rshader->clip_dist_write & 0x0F) != 0) | 3222 S_02881C_VS_OUT_CCDIST1_VEC_ENA((rshader->clip_dist_write & 0xF0) != 0) | 3223 S_02881C_VS_OUT_MISC_VEC_ENA(rshader->vs_out_misc_write) | 3224 S_02881C_USE_VTX_POINT_SIZE(rshader->vs_out_point_size) | 3225 S_02881C_USE_VTX_EDGE_FLAG(rshader->vs_out_edgeflag) | 3226 S_02881C_USE_VTX_VIEWPORT_INDX(rshader->vs_out_viewport) | 3227 S_02881C_USE_VTX_RENDER_TARGET_INDX(rshader->vs_out_layer); 3228 } 3229 3230 void evergreen_update_hs_state(struct pipe_context *ctx, struct r600_pipe_shader *shader) 3231 { 3232 struct r600_command_buffer *cb = &shader->command_buffer; 3233 struct r600_shader *rshader = &shader->shader; 3234 3235 r600_init_command_buffer(cb, 32); 3236 r600_store_context_reg(cb, R_0288BC_SQ_PGM_RESOURCES_HS, 3237 S_0288BC_NUM_GPRS(rshader->bc.ngpr) | 3238 S_0288BC_STACK_SIZE(rshader->bc.nstack)); 3239 r600_store_context_reg(cb, R_0288B8_SQ_PGM_START_HS, 3240 shader->bo->gpu_address >> 8); 3241 } 3242 3243 void evergreen_update_ls_state(struct pipe_context *ctx, struct r600_pipe_shader *shader) 3244 { 3245 struct r600_command_buffer *cb = &shader->command_buffer; 3246 struct r600_shader *rshader = &shader->shader; 3247 3248 r600_init_command_buffer(cb, 32); 3249 r600_store_context_reg(cb, R_0288D4_SQ_PGM_RESOURCES_LS, 3250 S_0288D4_NUM_GPRS(rshader->bc.ngpr) | 3251 S_0288D4_STACK_SIZE(rshader->bc.nstack)); 3252 r600_store_context_reg(cb, R_0288D0_SQ_PGM_START_LS, 3253 shader->bo->gpu_address >> 8); 3254 } 3255 void *evergreen_create_resolve_blend(struct r600_context *rctx) 3256 { 3257 struct pipe_blend_state blend; 3258 3259 memset(&blend, 0, sizeof(blend)); 3260 blend.independent_blend_enable = true; 3261 blend.rt[0].colormask = 0xf; 3262 return evergreen_create_blend_state_mode(&rctx->b.b, &blend, V_028808_CB_RESOLVE); 3263 } 3264 3265 void *evergreen_create_decompress_blend(struct r600_context *rctx) 3266 { 3267 struct pipe_blend_state blend; 3268 unsigned mode = rctx->screen->has_compressed_msaa_texturing ? 3269 V_028808_CB_FMASK_DECOMPRESS : V_028808_CB_DECOMPRESS; 3270 3271 memset(&blend, 0, sizeof(blend)); 3272 blend.independent_blend_enable = true; 3273 blend.rt[0].colormask = 0xf; 3274 return evergreen_create_blend_state_mode(&rctx->b.b, &blend, mode); 3275 } 3276 3277 void *evergreen_create_fastclear_blend(struct r600_context *rctx) 3278 { 3279 struct pipe_blend_state blend; 3280 unsigned mode = V_028808_CB_ELIMINATE_FAST_CLEAR; 3281 3282 memset(&blend, 0, sizeof(blend)); 3283 blend.independent_blend_enable = true; 3284 blend.rt[0].colormask = 0xf; 3285 return evergreen_create_blend_state_mode(&rctx->b.b, &blend, mode); 3286 } 3287 3288 void *evergreen_create_db_flush_dsa(struct r600_context *rctx) 3289 { 3290 struct pipe_depth_stencil_alpha_state dsa = {{0}}; 3291 3292 return rctx->b.b.create_depth_stencil_alpha_state(&rctx->b.b, &dsa); 3293 } 3294 3295 void evergreen_update_db_shader_control(struct r600_context * rctx) 3296 { 3297 bool dual_export; 3298 unsigned db_shader_control; 3299 3300 if (!rctx->ps_shader) { 3301 return; 3302 } 3303 3304 dual_export = rctx->framebuffer.export_16bpc && 3305 !rctx->ps_shader->current->ps_depth_export; 3306 3307 db_shader_control = rctx->ps_shader->current->db_shader_control | 3308 S_02880C_DUAL_EXPORT_ENABLE(dual_export) | 3309 S_02880C_DB_SOURCE_FORMAT(dual_export ? V_02880C_EXPORT_DB_TWO : 3310 V_02880C_EXPORT_DB_FULL) | 3311 S_02880C_ALPHA_TO_MASK_DISABLE(rctx->framebuffer.cb0_is_integer); 3312 3313 /* When alpha test is enabled we can't trust the hw to make the proper 3314 * decision on the order in which ztest should be run related to fragment 3315 * shader execution. 3316 * 3317 * If alpha test is enabled perform early z rejection (RE_Z) but don't early 3318 * write to the zbuffer. Write to zbuffer is delayed after fragment shader 3319 * execution and thus after alpha test so if discarded by the alpha test 3320 * the z value is not written. 3321 * If ReZ is enabled, and the zfunc/zenable/zwrite values change you can 3322 * get a hang unless you flush the DB in between. For now just use 3323 * LATE_Z. 3324 */ 3325 if (rctx->alphatest_state.sx_alpha_test_control) { 3326 db_shader_control |= S_02880C_Z_ORDER(V_02880C_LATE_Z); 3327 } else { 3328 db_shader_control |= S_02880C_Z_ORDER(V_02880C_EARLY_Z_THEN_LATE_Z); 3329 } 3330 3331 if (db_shader_control != rctx->db_misc_state.db_shader_control) { 3332 rctx->db_misc_state.db_shader_control = db_shader_control; 3333 r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom); 3334 } 3335 } 3336 3337 static void evergreen_dma_copy_tile(struct r600_context *rctx, 3338 struct pipe_resource *dst, 3339 unsigned dst_level, 3340 unsigned dst_x, 3341 unsigned dst_y, 3342 unsigned dst_z, 3343 struct pipe_resource *src, 3344 unsigned src_level, 3345 unsigned src_x, 3346 unsigned src_y, 3347 unsigned src_z, 3348 unsigned copy_height, 3349 unsigned pitch, 3350 unsigned bpp) 3351 { 3352 struct radeon_winsys_cs *cs = rctx->b.dma.cs; 3353 struct r600_texture *rsrc = (struct r600_texture*)src; 3354 struct r600_texture *rdst = (struct r600_texture*)dst; 3355 unsigned array_mode, lbpp, pitch_tile_max, slice_tile_max, size; 3356 unsigned ncopy, height, cheight, detile, i, x, y, z, src_mode, dst_mode; 3357 unsigned sub_cmd, bank_h, bank_w, mt_aspect, nbanks, tile_split, non_disp_tiling = 0; 3358 uint64_t base, addr; 3359 3360 dst_mode = rdst->surface.level[dst_level].mode; 3361 src_mode = rsrc->surface.level[src_level].mode; 3362 assert(dst_mode != src_mode); 3363 3364 /* non_disp_tiling bit needs to be set for depth, stencil, and fmask surfaces */ 3365 if (util_format_has_depth(util_format_description(src->format))) 3366 non_disp_tiling = 1; 3367 3368 y = 0; 3369 sub_cmd = EG_DMA_COPY_TILED; 3370 lbpp = util_logbase2(bpp); 3371 pitch_tile_max = ((pitch / bpp) / 8) - 1; 3372 nbanks = eg_num_banks(rctx->screen->b.info.r600_num_banks); 3373 3374 if (dst_mode == RADEON_SURF_MODE_LINEAR_ALIGNED) { 3375 /* T2L */ 3376 array_mode = evergreen_array_mode(src_mode); 3377 slice_tile_max = (rsrc->surface.level[src_level].nblk_x * rsrc->surface.level[src_level].nblk_y) / (8*8); 3378 slice_tile_max = slice_tile_max ? slice_tile_max - 1 : 0; 3379 /* linear height must be the same as the slice tile max height, it's ok even 3380 * if the linear destination/source have smaller heigh as the size of the 3381 * dma packet will be using the copy_height which is always smaller or equal 3382 * to the linear height 3383 */ 3384 height = u_minify(rsrc->resource.b.b.height0, src_level); 3385 detile = 1; 3386 x = src_x; 3387 y = src_y; 3388 z = src_z; 3389 base = rsrc->surface.level[src_level].offset; 3390 addr = rdst->surface.level[dst_level].offset; 3391 addr += rdst->surface.level[dst_level].slice_size * dst_z; 3392 addr += dst_y * pitch + dst_x * bpp; 3393 bank_h = eg_bank_wh(rsrc->surface.bankh); 3394 bank_w = eg_bank_wh(rsrc->surface.bankw); 3395 mt_aspect = eg_macro_tile_aspect(rsrc->surface.mtilea); 3396 tile_split = eg_tile_split(rsrc->surface.tile_split); 3397 base += rsrc->resource.gpu_address; 3398 addr += rdst->resource.gpu_address; 3399 } else { 3400 /* L2T */ 3401 array_mode = evergreen_array_mode(dst_mode); 3402 slice_tile_max = (rdst->surface.level[dst_level].nblk_x * rdst->surface.level[dst_level].nblk_y) / (8*8); 3403 slice_tile_max = slice_tile_max ? slice_tile_max - 1 : 0; 3404 /* linear height must be the same as the slice tile max height, it's ok even 3405 * if the linear destination/source have smaller heigh as the size of the 3406 * dma packet will be using the copy_height which is always smaller or equal 3407 * to the linear height 3408 */ 3409 height = u_minify(rdst->resource.b.b.height0, dst_level); 3410 detile = 0; 3411 x = dst_x; 3412 y = dst_y; 3413 z = dst_z; 3414 base = rdst->surface.level[dst_level].offset; 3415 addr = rsrc->surface.level[src_level].offset; 3416 addr += rsrc->surface.level[src_level].slice_size * src_z; 3417 addr += src_y * pitch + src_x * bpp; 3418 bank_h = eg_bank_wh(rdst->surface.bankh); 3419 bank_w = eg_bank_wh(rdst->surface.bankw); 3420 mt_aspect = eg_macro_tile_aspect(rdst->surface.mtilea); 3421 tile_split = eg_tile_split(rdst->surface.tile_split); 3422 base += rdst->resource.gpu_address; 3423 addr += rsrc->resource.gpu_address; 3424 } 3425 3426 size = (copy_height * pitch) / 4; 3427 ncopy = (size / EG_DMA_COPY_MAX_SIZE) + !!(size % EG_DMA_COPY_MAX_SIZE); 3428 r600_need_dma_space(&rctx->b, ncopy * 9, &rdst->resource, &rsrc->resource); 3429 3430 for (i = 0; i < ncopy; i++) { 3431 cheight = copy_height; 3432 if (((cheight * pitch) / 4) > EG_DMA_COPY_MAX_SIZE) { 3433 cheight = (EG_DMA_COPY_MAX_SIZE * 4) / pitch; 3434 } 3435 size = (cheight * pitch) / 4; 3436 /* emit reloc before writing cs so that cs is always in consistent state */ 3437 radeon_add_to_buffer_list(&rctx->b, &rctx->b.dma, &rsrc->resource, 3438 RADEON_USAGE_READ, RADEON_PRIO_SDMA_TEXTURE); 3439 radeon_add_to_buffer_list(&rctx->b, &rctx->b.dma, &rdst->resource, 3440 RADEON_USAGE_WRITE, RADEON_PRIO_SDMA_TEXTURE); 3441 radeon_emit(cs, DMA_PACKET(DMA_PACKET_COPY, sub_cmd, size)); 3442 radeon_emit(cs, base >> 8); 3443 radeon_emit(cs, (detile << 31) | (array_mode << 27) | 3444 (lbpp << 24) | (bank_h << 21) | 3445 (bank_w << 18) | (mt_aspect << 16)); 3446 radeon_emit(cs, (pitch_tile_max << 0) | ((height - 1) << 16)); 3447 radeon_emit(cs, (slice_tile_max << 0)); 3448 radeon_emit(cs, (x << 0) | (z << 18)); 3449 radeon_emit(cs, (y << 0) | (tile_split << 21) | (nbanks << 25) | (non_disp_tiling << 28)); 3450 radeon_emit(cs, addr & 0xfffffffc); 3451 radeon_emit(cs, (addr >> 32UL) & 0xff); 3452 copy_height -= cheight; 3453 addr += cheight * pitch; 3454 y += cheight; 3455 } 3456 } 3457 3458 static void evergreen_dma_copy(struct pipe_context *ctx, 3459 struct pipe_resource *dst, 3460 unsigned dst_level, 3461 unsigned dstx, unsigned dsty, unsigned dstz, 3462 struct pipe_resource *src, 3463 unsigned src_level, 3464 const struct pipe_box *src_box) 3465 { 3466 struct r600_context *rctx = (struct r600_context *)ctx; 3467 struct r600_texture *rsrc = (struct r600_texture*)src; 3468 struct r600_texture *rdst = (struct r600_texture*)dst; 3469 unsigned dst_pitch, src_pitch, bpp, dst_mode, src_mode, copy_height; 3470 unsigned src_w, dst_w; 3471 unsigned src_x, src_y; 3472 unsigned dst_x = dstx, dst_y = dsty, dst_z = dstz; 3473 3474 if (rctx->b.dma.cs == NULL) { 3475 goto fallback; 3476 } 3477 3478 if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) { 3479 evergreen_dma_copy_buffer(rctx, dst, src, dst_x, src_box->x, src_box->width); 3480 return; 3481 } 3482 3483 if (src_box->depth > 1 || 3484 !r600_prepare_for_dma_blit(&rctx->b, rdst, dst_level, dstx, dsty, 3485 dstz, rsrc, src_level, src_box)) 3486 goto fallback; 3487 3488 src_x = util_format_get_nblocksx(src->format, src_box->x); 3489 dst_x = util_format_get_nblocksx(src->format, dst_x); 3490 src_y = util_format_get_nblocksy(src->format, src_box->y); 3491 dst_y = util_format_get_nblocksy(src->format, dst_y); 3492 3493 bpp = rdst->surface.bpe; 3494 dst_pitch = rdst->surface.level[dst_level].nblk_x * rdst->surface.bpe; 3495 src_pitch = rsrc->surface.level[src_level].nblk_x * rsrc->surface.bpe; 3496 src_w = u_minify(rsrc->resource.b.b.width0, src_level); 3497 dst_w = u_minify(rdst->resource.b.b.width0, dst_level); 3498 copy_height = src_box->height / rsrc->surface.blk_h; 3499 3500 dst_mode = rdst->surface.level[dst_level].mode; 3501 src_mode = rsrc->surface.level[src_level].mode; 3502 3503 if (src_pitch != dst_pitch || src_box->x || dst_x || src_w != dst_w) { 3504 /* FIXME evergreen can do partial blit */ 3505 goto fallback; 3506 } 3507 /* the x test here are currently useless (because we don't support partial blit) 3508 * but keep them around so we don't forget about those 3509 */ 3510 if (src_pitch % 8 || src_box->x % 8 || dst_x % 8 || src_box->y % 8 || dst_y % 8) { 3511 goto fallback; 3512 } 3513 3514 /* 128 bpp surfaces require non_disp_tiling for both 3515 * tiled and linear buffers on cayman. However, async 3516 * DMA only supports it on the tiled side. As such 3517 * the tile order is backwards after a L2T/T2L packet. 3518 */ 3519 if ((rctx->b.chip_class == CAYMAN) && 3520 (src_mode != dst_mode) && 3521 (util_format_get_blocksize(src->format) >= 16)) { 3522 goto fallback; 3523 } 3524 3525 if (src_mode == dst_mode) { 3526 uint64_t dst_offset, src_offset; 3527 /* simple dma blit would do NOTE code here assume : 3528 * src_box.x/y == 0 3529 * dst_x/y == 0 3530 * dst_pitch == src_pitch 3531 */ 3532 src_offset= rsrc->surface.level[src_level].offset; 3533 src_offset += rsrc->surface.level[src_level].slice_size * src_box->z; 3534 src_offset += src_y * src_pitch + src_x * bpp; 3535 dst_offset = rdst->surface.level[dst_level].offset; 3536 dst_offset += rdst->surface.level[dst_level].slice_size * dst_z; 3537 dst_offset += dst_y * dst_pitch + dst_x * bpp; 3538 evergreen_dma_copy_buffer(rctx, dst, src, dst_offset, src_offset, 3539 src_box->height * src_pitch); 3540 } else { 3541 evergreen_dma_copy_tile(rctx, dst, dst_level, dst_x, dst_y, dst_z, 3542 src, src_level, src_x, src_y, src_box->z, 3543 copy_height, dst_pitch, bpp); 3544 } 3545 return; 3546 3547 fallback: 3548 r600_resource_copy_region(ctx, dst, dst_level, dstx, dsty, dstz, 3549 src, src_level, src_box); 3550 } 3551 3552 static void evergreen_set_tess_state(struct pipe_context *ctx, 3553 const float default_outer_level[4], 3554 const float default_inner_level[2]) 3555 { 3556 struct r600_context *rctx = (struct r600_context *)ctx; 3557 3558 memcpy(rctx->tess_state, default_outer_level, sizeof(float) * 4); 3559 memcpy(rctx->tess_state+4, default_inner_level, sizeof(float) * 2); 3560 rctx->tess_state_dirty = true; 3561 } 3562 3563 void evergreen_init_state_functions(struct r600_context *rctx) 3564 { 3565 unsigned id = 1; 3566 unsigned i; 3567 /* !!! 3568 * To avoid GPU lockup registers must be emitted in a specific order 3569 * (no kidding ...). The order below is important and have been 3570 * partially inferred from analyzing fglrx command stream. 3571 * 3572 * Don't reorder atom without carefully checking the effect (GPU lockup 3573 * or piglit regression). 3574 * !!! 3575 */ 3576 if (rctx->b.chip_class == EVERGREEN) { 3577 r600_init_atom(rctx, &rctx->config_state.atom, id++, evergreen_emit_config_state, 11); 3578 rctx->config_state.dyn_gpr_enabled = true; 3579 } 3580 r600_init_atom(rctx, &rctx->framebuffer.atom, id++, evergreen_emit_framebuffer_state, 0); 3581 /* shader const */ 3582 r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_VERTEX].atom, id++, evergreen_emit_vs_constant_buffers, 0); 3583 r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_GEOMETRY].atom, id++, evergreen_emit_gs_constant_buffers, 0); 3584 r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_FRAGMENT].atom, id++, evergreen_emit_ps_constant_buffers, 0); 3585 r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_TESS_CTRL].atom, id++, evergreen_emit_tcs_constant_buffers, 0); 3586 r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_TESS_EVAL].atom, id++, evergreen_emit_tes_constant_buffers, 0); 3587 r600_init_atom(rctx, &rctx->constbuf_state[PIPE_SHADER_COMPUTE].atom, id++, evergreen_emit_cs_constant_buffers, 0); 3588 /* shader program */ 3589 r600_init_atom(rctx, &rctx->cs_shader_state.atom, id++, evergreen_emit_cs_shader, 0); 3590 /* sampler */ 3591 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_VERTEX].states.atom, id++, evergreen_emit_vs_sampler_states, 0); 3592 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY].states.atom, id++, evergreen_emit_gs_sampler_states, 0); 3593 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_TESS_CTRL].states.atom, id++, evergreen_emit_tcs_sampler_states, 0); 3594 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_TESS_EVAL].states.atom, id++, evergreen_emit_tes_sampler_states, 0); 3595 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT].states.atom, id++, evergreen_emit_ps_sampler_states, 0); 3596 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_COMPUTE].states.atom, id++, evergreen_emit_cs_sampler_states, 0); 3597 /* resources */ 3598 r600_init_atom(rctx, &rctx->vertex_buffer_state.atom, id++, evergreen_fs_emit_vertex_buffers, 0); 3599 r600_init_atom(rctx, &rctx->cs_vertex_buffer_state.atom, id++, evergreen_cs_emit_vertex_buffers, 0); 3600 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_VERTEX].views.atom, id++, evergreen_emit_vs_sampler_views, 0); 3601 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_GEOMETRY].views.atom, id++, evergreen_emit_gs_sampler_views, 0); 3602 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_TESS_CTRL].views.atom, id++, evergreen_emit_tcs_sampler_views, 0); 3603 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_TESS_EVAL].views.atom, id++, evergreen_emit_tes_sampler_views, 0); 3604 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_FRAGMENT].views.atom, id++, evergreen_emit_ps_sampler_views, 0); 3605 r600_init_atom(rctx, &rctx->samplers[PIPE_SHADER_COMPUTE].views.atom, id++, evergreen_emit_cs_sampler_views, 0); 3606 3607 r600_init_atom(rctx, &rctx->vgt_state.atom, id++, r600_emit_vgt_state, 10); 3608 3609 if (rctx->b.chip_class == EVERGREEN) { 3610 r600_init_atom(rctx, &rctx->sample_mask.atom, id++, evergreen_emit_sample_mask, 3); 3611 } else { 3612 r600_init_atom(rctx, &rctx->sample_mask.atom, id++, cayman_emit_sample_mask, 4); 3613 } 3614 rctx->sample_mask.sample_mask = ~0; 3615 3616 r600_init_atom(rctx, &rctx->alphatest_state.atom, id++, r600_emit_alphatest_state, 6); 3617 r600_init_atom(rctx, &rctx->blend_color.atom, id++, r600_emit_blend_color, 6); 3618 r600_init_atom(rctx, &rctx->blend_state.atom, id++, r600_emit_cso_state, 0); 3619 r600_init_atom(rctx, &rctx->cb_misc_state.atom, id++, evergreen_emit_cb_misc_state, 4); 3620 r600_init_atom(rctx, &rctx->clip_misc_state.atom, id++, r600_emit_clip_misc_state, 9); 3621 r600_init_atom(rctx, &rctx->clip_state.atom, id++, evergreen_emit_clip_state, 26); 3622 r600_init_atom(rctx, &rctx->db_misc_state.atom, id++, evergreen_emit_db_misc_state, 10); 3623 r600_init_atom(rctx, &rctx->db_state.atom, id++, evergreen_emit_db_state, 14); 3624 r600_init_atom(rctx, &rctx->dsa_state.atom, id++, r600_emit_cso_state, 0); 3625 r600_init_atom(rctx, &rctx->poly_offset_state.atom, id++, evergreen_emit_polygon_offset, 9); 3626 r600_init_atom(rctx, &rctx->rasterizer_state.atom, id++, r600_emit_cso_state, 0); 3627 r600_add_atom(rctx, &rctx->b.scissors.atom, id++); 3628 r600_add_atom(rctx, &rctx->b.viewports.atom, id++); 3629 r600_init_atom(rctx, &rctx->stencil_ref.atom, id++, r600_emit_stencil_ref, 4); 3630 r600_init_atom(rctx, &rctx->vertex_fetch_shader.atom, id++, evergreen_emit_vertex_fetch_shader, 5); 3631 r600_add_atom(rctx, &rctx->b.render_cond_atom, id++); 3632 r600_add_atom(rctx, &rctx->b.streamout.begin_atom, id++); 3633 r600_add_atom(rctx, &rctx->b.streamout.enable_atom, id++); 3634 for (i = 0; i < EG_NUM_HW_STAGES; i++) 3635 r600_init_atom(rctx, &rctx->hw_shader_stages[i].atom, id++, r600_emit_shader, 0); 3636 r600_init_atom(rctx, &rctx->shader_stages.atom, id++, evergreen_emit_shader_stages, 15); 3637 r600_init_atom(rctx, &rctx->gs_rings.atom, id++, evergreen_emit_gs_rings, 26); 3638 3639 rctx->b.b.create_blend_state = evergreen_create_blend_state; 3640 rctx->b.b.create_depth_stencil_alpha_state = evergreen_create_dsa_state; 3641 rctx->b.b.create_rasterizer_state = evergreen_create_rs_state; 3642 rctx->b.b.create_sampler_state = evergreen_create_sampler_state; 3643 rctx->b.b.create_sampler_view = evergreen_create_sampler_view; 3644 rctx->b.b.set_framebuffer_state = evergreen_set_framebuffer_state; 3645 rctx->b.b.set_polygon_stipple = evergreen_set_polygon_stipple; 3646 rctx->b.b.set_min_samples = evergreen_set_min_samples; 3647 rctx->b.b.set_tess_state = evergreen_set_tess_state; 3648 if (rctx->b.chip_class == EVERGREEN) 3649 rctx->b.b.get_sample_position = evergreen_get_sample_position; 3650 else 3651 rctx->b.b.get_sample_position = cayman_get_sample_position; 3652 rctx->b.dma_copy = evergreen_dma_copy; 3653 3654 evergreen_init_compute_state_functions(rctx); 3655 } 3656 3657 /** 3658 * This calculates the LDS size for tessellation shaders (VS, TCS, TES). 3659 * 3660 * The information about LDS and other non-compile-time parameters is then 3661 * written to the const buffer. 3662 3663 * const buffer contains - 3664 * uint32_t input_patch_size 3665 * uint32_t input_vertex_size 3666 * uint32_t num_tcs_input_cp 3667 * uint32_t num_tcs_output_cp; 3668 * uint32_t output_patch_size 3669 * uint32_t output_vertex_size 3670 * uint32_t output_patch0_offset 3671 * uint32_t perpatch_output_offset 3672 * and the same constbuf is bound to LS/HS/VS(ES). 3673 */ 3674 void evergreen_setup_tess_constants(struct r600_context *rctx, const struct pipe_draw_info *info, unsigned *num_patches) 3675 { 3676 struct pipe_constant_buffer constbuf = {0}; 3677 struct r600_pipe_shader_selector *tcs = rctx->tcs_shader ? rctx->tcs_shader : rctx->tes_shader; 3678 struct r600_pipe_shader_selector *ls = rctx->vs_shader; 3679 unsigned num_tcs_input_cp = info->vertices_per_patch; 3680 unsigned num_tcs_outputs; 3681 unsigned num_tcs_output_cp; 3682 unsigned num_tcs_patch_outputs; 3683 unsigned num_tcs_inputs; 3684 unsigned input_vertex_size, output_vertex_size; 3685 unsigned input_patch_size, pervertex_output_patch_size, output_patch_size; 3686 unsigned output_patch0_offset, perpatch_output_offset, lds_size; 3687 uint32_t values[16]; 3688 unsigned num_waves; 3689 unsigned num_pipes = rctx->screen->b.info.r600_max_quad_pipes; 3690 unsigned wave_divisor = (16 * num_pipes); 3691 3692 *num_patches = 1; 3693 3694 if (!rctx->tes_shader) { 3695 rctx->lds_alloc = 0; 3696 rctx->b.b.set_constant_buffer(&rctx->b.b, PIPE_SHADER_VERTEX, 3697 R600_LDS_INFO_CONST_BUFFER, NULL); 3698 rctx->b.b.set_constant_buffer(&rctx->b.b, PIPE_SHADER_TESS_CTRL, 3699 R600_LDS_INFO_CONST_BUFFER, NULL); 3700 rctx->b.b.set_constant_buffer(&rctx->b.b, PIPE_SHADER_TESS_EVAL, 3701 R600_LDS_INFO_CONST_BUFFER, NULL); 3702 return; 3703 } 3704 3705 if (rctx->lds_alloc != 0 && 3706 rctx->last_ls == ls && 3707 !rctx->tess_state_dirty && 3708 rctx->last_num_tcs_input_cp == num_tcs_input_cp && 3709 rctx->last_tcs == tcs) 3710 return; 3711 3712 num_tcs_inputs = util_last_bit64(ls->lds_outputs_written_mask); 3713 3714 if (rctx->tcs_shader) { 3715 num_tcs_outputs = util_last_bit64(tcs->lds_outputs_written_mask); 3716 num_tcs_output_cp = tcs->info.properties[TGSI_PROPERTY_TCS_VERTICES_OUT]; 3717 num_tcs_patch_outputs = util_last_bit64(tcs->lds_patch_outputs_written_mask); 3718 } else { 3719 num_tcs_outputs = num_tcs_inputs; 3720 num_tcs_output_cp = num_tcs_input_cp; 3721 num_tcs_patch_outputs = 2; /* TESSINNER + TESSOUTER */ 3722 } 3723 3724 /* size in bytes */ 3725 input_vertex_size = num_tcs_inputs * 16; 3726 output_vertex_size = num_tcs_outputs * 16; 3727 3728 input_patch_size = num_tcs_input_cp * input_vertex_size; 3729 3730 pervertex_output_patch_size = num_tcs_output_cp * output_vertex_size; 3731 output_patch_size = pervertex_output_patch_size + num_tcs_patch_outputs * 16; 3732 3733 output_patch0_offset = rctx->tcs_shader ? input_patch_size * *num_patches : 0; 3734 perpatch_output_offset = output_patch0_offset + pervertex_output_patch_size; 3735 3736 lds_size = output_patch0_offset + output_patch_size * *num_patches; 3737 3738 values[0] = input_patch_size; 3739 values[1] = input_vertex_size; 3740 values[2] = num_tcs_input_cp; 3741 values[3] = num_tcs_output_cp; 3742 3743 values[4] = output_patch_size; 3744 values[5] = output_vertex_size; 3745 values[6] = output_patch0_offset; 3746 values[7] = perpatch_output_offset; 3747 3748 /* docs say HS_NUM_WAVES - CEIL((LS_HS_CONFIG.NUM_PATCHES * 3749 LS_HS_CONFIG.HS_NUM_OUTPUT_CP) / (NUM_GOOD_PIPES * 16)) */ 3750 num_waves = ceilf((float)(*num_patches * num_tcs_output_cp) / (float)wave_divisor); 3751 3752 rctx->lds_alloc = (lds_size | (num_waves << 14)); 3753 3754 memcpy(&values[8], rctx->tess_state, 6 * sizeof(float)); 3755 values[14] = 0; 3756 values[15] = 0; 3757 3758 rctx->tess_state_dirty = false; 3759 rctx->last_ls = ls; 3760 rctx->last_tcs = tcs; 3761 rctx->last_num_tcs_input_cp = num_tcs_input_cp; 3762 3763 constbuf.user_buffer = values; 3764 constbuf.buffer_size = 16 * 4; 3765 3766 rctx->b.b.set_constant_buffer(&rctx->b.b, PIPE_SHADER_VERTEX, 3767 R600_LDS_INFO_CONST_BUFFER, &constbuf); 3768 rctx->b.b.set_constant_buffer(&rctx->b.b, PIPE_SHADER_TESS_CTRL, 3769 R600_LDS_INFO_CONST_BUFFER, &constbuf); 3770 rctx->b.b.set_constant_buffer(&rctx->b.b, PIPE_SHADER_TESS_EVAL, 3771 R600_LDS_INFO_CONST_BUFFER, &constbuf); 3772 pipe_resource_reference(&constbuf.buffer, NULL); 3773 } 3774 3775 uint32_t evergreen_get_ls_hs_config(struct r600_context *rctx, 3776 const struct pipe_draw_info *info, 3777 unsigned num_patches) 3778 { 3779 unsigned num_output_cp; 3780 3781 if (!rctx->tes_shader) 3782 return 0; 3783 3784 num_output_cp = rctx->tcs_shader ? 3785 rctx->tcs_shader->info.properties[TGSI_PROPERTY_TCS_VERTICES_OUT] : 3786 info->vertices_per_patch; 3787 3788 return S_028B58_NUM_PATCHES(num_patches) | 3789 S_028B58_HS_NUM_INPUT_CP(info->vertices_per_patch) | 3790 S_028B58_HS_NUM_OUTPUT_CP(num_output_cp); 3791 } 3792 3793 void evergreen_set_ls_hs_config(struct r600_context *rctx, 3794 struct radeon_winsys_cs *cs, 3795 uint32_t ls_hs_config) 3796 { 3797 radeon_set_context_reg(cs, R_028B58_VGT_LS_HS_CONFIG, ls_hs_config); 3798 } 3799 3800 void evergreen_set_lds_alloc(struct r600_context *rctx, 3801 struct radeon_winsys_cs *cs, 3802 uint32_t lds_alloc) 3803 { 3804 radeon_set_context_reg(cs, R_0288E8_SQ_LDS_ALLOC, lds_alloc); 3805 } 3806 3807 /* on evergreen if you are running tessellation you need to disable dynamic 3808 GPRs to workaround a hardware bug.*/ 3809 bool evergreen_adjust_gprs(struct r600_context *rctx) 3810 { 3811 unsigned num_gprs[EG_NUM_HW_STAGES]; 3812 unsigned def_gprs[EG_NUM_HW_STAGES]; 3813 unsigned cur_gprs[EG_NUM_HW_STAGES]; 3814 unsigned new_gprs[EG_NUM_HW_STAGES]; 3815 unsigned def_num_clause_temp_gprs = rctx->r6xx_num_clause_temp_gprs; 3816 unsigned max_gprs; 3817 unsigned i; 3818 unsigned total_gprs; 3819 unsigned tmp[3]; 3820 bool rework = false, set_default = false, set_dirty = false; 3821 max_gprs = 0; 3822 for (i = 0; i < EG_NUM_HW_STAGES; i++) { 3823 def_gprs[i] = rctx->default_gprs[i]; 3824 max_gprs += def_gprs[i]; 3825 } 3826 max_gprs += def_num_clause_temp_gprs * 2; 3827 3828 /* if we have no TESS and dyn gpr is enabled then do nothing. */ 3829 if (!rctx->hw_shader_stages[EG_HW_STAGE_HS].shader) { 3830 if (rctx->config_state.dyn_gpr_enabled) 3831 return true; 3832 3833 /* transition back to dyn gpr enabled state */ 3834 rctx->config_state.dyn_gpr_enabled = true; 3835 r600_mark_atom_dirty(rctx, &rctx->config_state.atom); 3836 rctx->b.flags |= R600_CONTEXT_WAIT_3D_IDLE; 3837 return true; 3838 } 3839 3840 3841 /* gather required shader gprs */ 3842 for (i = 0; i < EG_NUM_HW_STAGES; i++) { 3843 if (rctx->hw_shader_stages[i].shader) 3844 num_gprs[i] = rctx->hw_shader_stages[i].shader->shader.bc.ngpr; 3845 else 3846 num_gprs[i] = 0; 3847 } 3848 3849 cur_gprs[R600_HW_STAGE_PS] = G_008C04_NUM_PS_GPRS(rctx->config_state.sq_gpr_resource_mgmt_1); 3850 cur_gprs[R600_HW_STAGE_VS] = G_008C04_NUM_VS_GPRS(rctx->config_state.sq_gpr_resource_mgmt_1); 3851 cur_gprs[R600_HW_STAGE_GS] = G_008C08_NUM_GS_GPRS(rctx->config_state.sq_gpr_resource_mgmt_2); 3852 cur_gprs[R600_HW_STAGE_ES] = G_008C08_NUM_ES_GPRS(rctx->config_state.sq_gpr_resource_mgmt_2); 3853 cur_gprs[EG_HW_STAGE_LS] = G_008C0C_NUM_LS_GPRS(rctx->config_state.sq_gpr_resource_mgmt_3); 3854 cur_gprs[EG_HW_STAGE_HS] = G_008C0C_NUM_HS_GPRS(rctx->config_state.sq_gpr_resource_mgmt_3); 3855 3856 total_gprs = 0; 3857 for (i = 0; i < EG_NUM_HW_STAGES; i++) { 3858 new_gprs[i] = num_gprs[i]; 3859 total_gprs += num_gprs[i]; 3860 } 3861 3862 if (total_gprs > (max_gprs - (2 * def_num_clause_temp_gprs))) 3863 return false; 3864 3865 for (i = 0; i < EG_NUM_HW_STAGES; i++) { 3866 if (new_gprs[i] > cur_gprs[i]) { 3867 rework = true; 3868 break; 3869 } 3870 } 3871 3872 if (rctx->config_state.dyn_gpr_enabled) { 3873 set_dirty = true; 3874 rctx->config_state.dyn_gpr_enabled = false; 3875 } 3876 3877 if (rework) { 3878 set_default = true; 3879 for (i = 0; i < EG_NUM_HW_STAGES; i++) { 3880 if (new_gprs[i] > def_gprs[i]) 3881 set_default = false; 3882 } 3883 3884 if (set_default) { 3885 for (i = 0; i < EG_NUM_HW_STAGES; i++) { 3886 new_gprs[i] = def_gprs[i]; 3887 } 3888 } else { 3889 unsigned ps_value = max_gprs; 3890 3891 ps_value -= (def_num_clause_temp_gprs * 2); 3892 for (i = R600_HW_STAGE_VS; i < EG_NUM_HW_STAGES; i++) 3893 ps_value -= new_gprs[i]; 3894 3895 new_gprs[R600_HW_STAGE_PS] = ps_value; 3896 } 3897 3898 tmp[0] = S_008C04_NUM_PS_GPRS(new_gprs[R600_HW_STAGE_PS]) | 3899 S_008C04_NUM_VS_GPRS(new_gprs[R600_HW_STAGE_VS]) | 3900 S_008C04_NUM_CLAUSE_TEMP_GPRS(def_num_clause_temp_gprs); 3901 3902 tmp[1] = S_008C08_NUM_ES_GPRS(new_gprs[R600_HW_STAGE_ES]) | 3903 S_008C08_NUM_GS_GPRS(new_gprs[R600_HW_STAGE_GS]); 3904 3905 tmp[2] = S_008C0C_NUM_HS_GPRS(new_gprs[EG_HW_STAGE_HS]) | 3906 S_008C0C_NUM_LS_GPRS(new_gprs[EG_HW_STAGE_LS]); 3907 3908 if (rctx->config_state.sq_gpr_resource_mgmt_1 != tmp[0] || 3909 rctx->config_state.sq_gpr_resource_mgmt_2 != tmp[1] || 3910 rctx->config_state.sq_gpr_resource_mgmt_3 != tmp[2]) { 3911 rctx->config_state.sq_gpr_resource_mgmt_1 = tmp[0]; 3912 rctx->config_state.sq_gpr_resource_mgmt_2 = tmp[1]; 3913 rctx->config_state.sq_gpr_resource_mgmt_3 = tmp[2]; 3914 set_dirty = true; 3915 } 3916 } 3917 3918 3919 if (set_dirty) { 3920 r600_mark_atom_dirty(rctx, &rctx->config_state.atom); 3921 rctx->b.flags |= R600_CONTEXT_WAIT_3D_IDLE; 3922 } 3923 return true; 3924 } 3925