1 /* 2 * Copyright 2010 Red Hat Inc. 3 * 2010 Jerome Glisse 4 * 5 * Permission is hereby granted, free of charge, to any person obtaining a 6 * copy of this software and associated documentation files (the "Software"), 7 * to deal in the Software without restriction, including without limitation 8 * on the rights to use, copy, modify, merge, publish, distribute, sub 9 * license, and/or sell copies of the Software, and to permit persons to whom 10 * the Software is furnished to do so, subject to the following conditions: 11 * 12 * The above copyright notice and this permission notice (including the next 13 * paragraph) shall be included in all copies or substantial portions of the 14 * Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL 19 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, 20 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 21 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 22 * USE OR OTHER DEALINGS IN THE SOFTWARE. 23 * 24 * Authors: Dave Airlie <airlied (at) redhat.com> 25 * Jerome Glisse <jglisse (at) redhat.com> 26 */ 27 #include "r600_formats.h" 28 #include "r600_shader.h" 29 #include "r600d.h" 30 31 #include "util/u_format_s3tc.h" 32 #include "util/u_index_modify.h" 33 #include "util/u_memory.h" 34 #include "util/u_upload_mgr.h" 35 #include "util/u_math.h" 36 #include "tgsi/tgsi_parse.h" 37 #include "tgsi/tgsi_scan.h" 38 #include "tgsi/tgsi_ureg.h" 39 40 void r600_init_command_buffer(struct r600_command_buffer *cb, unsigned num_dw) 41 { 42 assert(!cb->buf); 43 cb->buf = CALLOC(1, 4 * num_dw); 44 cb->max_num_dw = num_dw; 45 } 46 47 void r600_release_command_buffer(struct r600_command_buffer *cb) 48 { 49 FREE(cb->buf); 50 } 51 52 void r600_add_atom(struct r600_context *rctx, 53 struct r600_atom *atom, 54 unsigned id) 55 { 56 assert(id < R600_NUM_ATOMS); 57 assert(rctx->atoms[id] == NULL); 58 rctx->atoms[id] = atom; 59 atom->id = id; 60 } 61 62 void r600_init_atom(struct r600_context *rctx, 63 struct r600_atom *atom, 64 unsigned id, 65 void (*emit)(struct r600_context *ctx, struct r600_atom *state), 66 unsigned num_dw) 67 { 68 atom->emit = (void*)emit; 69 atom->num_dw = num_dw; 70 r600_add_atom(rctx, atom, id); 71 } 72 73 void r600_emit_cso_state(struct r600_context *rctx, struct r600_atom *atom) 74 { 75 r600_emit_command_buffer(rctx->b.gfx.cs, ((struct r600_cso_state*)atom)->cb); 76 } 77 78 void r600_emit_alphatest_state(struct r600_context *rctx, struct r600_atom *atom) 79 { 80 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 81 struct r600_alphatest_state *a = (struct r600_alphatest_state*)atom; 82 unsigned alpha_ref = a->sx_alpha_ref; 83 84 if (rctx->b.chip_class >= EVERGREEN && a->cb0_export_16bpc) { 85 alpha_ref &= ~0x1FFF; 86 } 87 88 radeon_set_context_reg(cs, R_028410_SX_ALPHA_TEST_CONTROL, 89 a->sx_alpha_test_control | 90 S_028410_ALPHA_TEST_BYPASS(a->bypass)); 91 radeon_set_context_reg(cs, R_028438_SX_ALPHA_REF, alpha_ref); 92 } 93 94 static void r600_memory_barrier(struct pipe_context *ctx, unsigned flags) 95 { 96 struct r600_context *rctx = (struct r600_context *)ctx; 97 if (flags & PIPE_BARRIER_CONSTANT_BUFFER) 98 rctx->b.flags |= R600_CONTEXT_INV_CONST_CACHE; 99 100 if (flags & (PIPE_BARRIER_VERTEX_BUFFER | 101 PIPE_BARRIER_SHADER_BUFFER | 102 PIPE_BARRIER_TEXTURE | 103 PIPE_BARRIER_IMAGE | 104 PIPE_BARRIER_STREAMOUT_BUFFER | 105 PIPE_BARRIER_GLOBAL_BUFFER)) { 106 rctx->b.flags |= R600_CONTEXT_INV_VERTEX_CACHE| 107 R600_CONTEXT_INV_TEX_CACHE; 108 } 109 110 if (flags & (PIPE_BARRIER_FRAMEBUFFER| 111 PIPE_BARRIER_IMAGE)) 112 rctx->b.flags |= R600_CONTEXT_FLUSH_AND_INV; 113 114 rctx->b.flags |= R600_CONTEXT_WAIT_3D_IDLE; 115 } 116 117 static void r600_texture_barrier(struct pipe_context *ctx, unsigned flags) 118 { 119 struct r600_context *rctx = (struct r600_context *)ctx; 120 121 rctx->b.flags |= R600_CONTEXT_INV_TEX_CACHE | 122 R600_CONTEXT_FLUSH_AND_INV_CB | 123 R600_CONTEXT_FLUSH_AND_INV | 124 R600_CONTEXT_WAIT_3D_IDLE; 125 rctx->framebuffer.do_update_surf_dirtiness = true; 126 } 127 128 static unsigned r600_conv_pipe_prim(unsigned prim) 129 { 130 static const unsigned prim_conv[] = { 131 [PIPE_PRIM_POINTS] = V_008958_DI_PT_POINTLIST, 132 [PIPE_PRIM_LINES] = V_008958_DI_PT_LINELIST, 133 [PIPE_PRIM_LINE_LOOP] = V_008958_DI_PT_LINELOOP, 134 [PIPE_PRIM_LINE_STRIP] = V_008958_DI_PT_LINESTRIP, 135 [PIPE_PRIM_TRIANGLES] = V_008958_DI_PT_TRILIST, 136 [PIPE_PRIM_TRIANGLE_STRIP] = V_008958_DI_PT_TRISTRIP, 137 [PIPE_PRIM_TRIANGLE_FAN] = V_008958_DI_PT_TRIFAN, 138 [PIPE_PRIM_QUADS] = V_008958_DI_PT_QUADLIST, 139 [PIPE_PRIM_QUAD_STRIP] = V_008958_DI_PT_QUADSTRIP, 140 [PIPE_PRIM_POLYGON] = V_008958_DI_PT_POLYGON, 141 [PIPE_PRIM_LINES_ADJACENCY] = V_008958_DI_PT_LINELIST_ADJ, 142 [PIPE_PRIM_LINE_STRIP_ADJACENCY] = V_008958_DI_PT_LINESTRIP_ADJ, 143 [PIPE_PRIM_TRIANGLES_ADJACENCY] = V_008958_DI_PT_TRILIST_ADJ, 144 [PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY] = V_008958_DI_PT_TRISTRIP_ADJ, 145 [PIPE_PRIM_PATCHES] = V_008958_DI_PT_PATCH, 146 [R600_PRIM_RECTANGLE_LIST] = V_008958_DI_PT_RECTLIST 147 }; 148 assert(prim < ARRAY_SIZE(prim_conv)); 149 return prim_conv[prim]; 150 } 151 152 unsigned r600_conv_prim_to_gs_out(unsigned mode) 153 { 154 static const int prim_conv[] = { 155 [PIPE_PRIM_POINTS] = V_028A6C_OUTPRIM_TYPE_POINTLIST, 156 [PIPE_PRIM_LINES] = V_028A6C_OUTPRIM_TYPE_LINESTRIP, 157 [PIPE_PRIM_LINE_LOOP] = V_028A6C_OUTPRIM_TYPE_LINESTRIP, 158 [PIPE_PRIM_LINE_STRIP] = V_028A6C_OUTPRIM_TYPE_LINESTRIP, 159 [PIPE_PRIM_TRIANGLES] = V_028A6C_OUTPRIM_TYPE_TRISTRIP, 160 [PIPE_PRIM_TRIANGLE_STRIP] = V_028A6C_OUTPRIM_TYPE_TRISTRIP, 161 [PIPE_PRIM_TRIANGLE_FAN] = V_028A6C_OUTPRIM_TYPE_TRISTRIP, 162 [PIPE_PRIM_QUADS] = V_028A6C_OUTPRIM_TYPE_TRISTRIP, 163 [PIPE_PRIM_QUAD_STRIP] = V_028A6C_OUTPRIM_TYPE_TRISTRIP, 164 [PIPE_PRIM_POLYGON] = V_028A6C_OUTPRIM_TYPE_TRISTRIP, 165 [PIPE_PRIM_LINES_ADJACENCY] = V_028A6C_OUTPRIM_TYPE_LINESTRIP, 166 [PIPE_PRIM_LINE_STRIP_ADJACENCY] = V_028A6C_OUTPRIM_TYPE_LINESTRIP, 167 [PIPE_PRIM_TRIANGLES_ADJACENCY] = V_028A6C_OUTPRIM_TYPE_TRISTRIP, 168 [PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY] = V_028A6C_OUTPRIM_TYPE_TRISTRIP, 169 [PIPE_PRIM_PATCHES] = V_028A6C_OUTPRIM_TYPE_POINTLIST, 170 [R600_PRIM_RECTANGLE_LIST] = V_028A6C_OUTPRIM_TYPE_TRISTRIP 171 }; 172 assert(mode < ARRAY_SIZE(prim_conv)); 173 174 return prim_conv[mode]; 175 } 176 177 /* common state between evergreen and r600 */ 178 179 static void r600_bind_blend_state_internal(struct r600_context *rctx, 180 struct r600_blend_state *blend, bool blend_disable) 181 { 182 unsigned color_control; 183 bool update_cb = false; 184 185 rctx->alpha_to_one = blend->alpha_to_one; 186 rctx->dual_src_blend = blend->dual_src_blend; 187 188 if (!blend_disable) { 189 r600_set_cso_state_with_cb(rctx, &rctx->blend_state, blend, &blend->buffer); 190 color_control = blend->cb_color_control; 191 } else { 192 /* Blending is disabled. */ 193 r600_set_cso_state_with_cb(rctx, &rctx->blend_state, blend, &blend->buffer_no_blend); 194 color_control = blend->cb_color_control_no_blend; 195 } 196 197 /* Update derived states. */ 198 if (rctx->cb_misc_state.blend_colormask != blend->cb_target_mask) { 199 rctx->cb_misc_state.blend_colormask = blend->cb_target_mask; 200 update_cb = true; 201 } 202 if (rctx->b.chip_class <= R700 && 203 rctx->cb_misc_state.cb_color_control != color_control) { 204 rctx->cb_misc_state.cb_color_control = color_control; 205 update_cb = true; 206 } 207 if (rctx->cb_misc_state.dual_src_blend != blend->dual_src_blend) { 208 rctx->cb_misc_state.dual_src_blend = blend->dual_src_blend; 209 update_cb = true; 210 } 211 if (update_cb) { 212 r600_mark_atom_dirty(rctx, &rctx->cb_misc_state.atom); 213 } 214 if (rctx->framebuffer.dual_src_blend != blend->dual_src_blend) { 215 rctx->framebuffer.dual_src_blend = blend->dual_src_blend; 216 r600_mark_atom_dirty(rctx, &rctx->framebuffer.atom); 217 } 218 } 219 220 static void r600_bind_blend_state(struct pipe_context *ctx, void *state) 221 { 222 struct r600_context *rctx = (struct r600_context *)ctx; 223 struct r600_blend_state *blend = (struct r600_blend_state *)state; 224 225 if (!blend) { 226 r600_set_cso_state_with_cb(rctx, &rctx->blend_state, NULL, NULL); 227 return; 228 } 229 230 r600_bind_blend_state_internal(rctx, blend, rctx->force_blend_disable); 231 } 232 233 static void r600_set_blend_color(struct pipe_context *ctx, 234 const struct pipe_blend_color *state) 235 { 236 struct r600_context *rctx = (struct r600_context *)ctx; 237 238 rctx->blend_color.state = *state; 239 r600_mark_atom_dirty(rctx, &rctx->blend_color.atom); 240 } 241 242 void r600_emit_blend_color(struct r600_context *rctx, struct r600_atom *atom) 243 { 244 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 245 struct pipe_blend_color *state = &rctx->blend_color.state; 246 247 radeon_set_context_reg_seq(cs, R_028414_CB_BLEND_RED, 4); 248 radeon_emit(cs, fui(state->color[0])); /* R_028414_CB_BLEND_RED */ 249 radeon_emit(cs, fui(state->color[1])); /* R_028418_CB_BLEND_GREEN */ 250 radeon_emit(cs, fui(state->color[2])); /* R_02841C_CB_BLEND_BLUE */ 251 radeon_emit(cs, fui(state->color[3])); /* R_028420_CB_BLEND_ALPHA */ 252 } 253 254 void r600_emit_vgt_state(struct r600_context *rctx, struct r600_atom *atom) 255 { 256 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 257 struct r600_vgt_state *a = (struct r600_vgt_state *)atom; 258 259 radeon_set_context_reg(cs, R_028A94_VGT_MULTI_PRIM_IB_RESET_EN, a->vgt_multi_prim_ib_reset_en); 260 radeon_set_context_reg_seq(cs, R_028408_VGT_INDX_OFFSET, 2); 261 radeon_emit(cs, a->vgt_indx_offset); /* R_028408_VGT_INDX_OFFSET */ 262 radeon_emit(cs, a->vgt_multi_prim_ib_reset_indx); /* R_02840C_VGT_MULTI_PRIM_IB_RESET_INDX */ 263 if (a->last_draw_was_indirect) { 264 a->last_draw_was_indirect = false; 265 radeon_set_ctl_const(cs, R_03CFF0_SQ_VTX_BASE_VTX_LOC, 0); 266 } 267 } 268 269 static void r600_set_clip_state(struct pipe_context *ctx, 270 const struct pipe_clip_state *state) 271 { 272 struct r600_context *rctx = (struct r600_context *)ctx; 273 274 rctx->clip_state.state = *state; 275 r600_mark_atom_dirty(rctx, &rctx->clip_state.atom); 276 rctx->driver_consts[PIPE_SHADER_VERTEX].vs_ucp_dirty = true; 277 } 278 279 static void r600_set_stencil_ref(struct pipe_context *ctx, 280 const struct r600_stencil_ref *state) 281 { 282 struct r600_context *rctx = (struct r600_context *)ctx; 283 284 rctx->stencil_ref.state = *state; 285 r600_mark_atom_dirty(rctx, &rctx->stencil_ref.atom); 286 } 287 288 void r600_emit_stencil_ref(struct r600_context *rctx, struct r600_atom *atom) 289 { 290 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 291 struct r600_stencil_ref_state *a = (struct r600_stencil_ref_state*)atom; 292 293 radeon_set_context_reg_seq(cs, R_028430_DB_STENCILREFMASK, 2); 294 radeon_emit(cs, /* R_028430_DB_STENCILREFMASK */ 295 S_028430_STENCILREF(a->state.ref_value[0]) | 296 S_028430_STENCILMASK(a->state.valuemask[0]) | 297 S_028430_STENCILWRITEMASK(a->state.writemask[0])); 298 radeon_emit(cs, /* R_028434_DB_STENCILREFMASK_BF */ 299 S_028434_STENCILREF_BF(a->state.ref_value[1]) | 300 S_028434_STENCILMASK_BF(a->state.valuemask[1]) | 301 S_028434_STENCILWRITEMASK_BF(a->state.writemask[1])); 302 } 303 304 static void r600_set_pipe_stencil_ref(struct pipe_context *ctx, 305 const struct pipe_stencil_ref *state) 306 { 307 struct r600_context *rctx = (struct r600_context *)ctx; 308 struct r600_dsa_state *dsa = (struct r600_dsa_state*)rctx->dsa_state.cso; 309 struct r600_stencil_ref ref; 310 311 rctx->stencil_ref.pipe_state = *state; 312 313 if (!dsa) 314 return; 315 316 ref.ref_value[0] = state->ref_value[0]; 317 ref.ref_value[1] = state->ref_value[1]; 318 ref.valuemask[0] = dsa->valuemask[0]; 319 ref.valuemask[1] = dsa->valuemask[1]; 320 ref.writemask[0] = dsa->writemask[0]; 321 ref.writemask[1] = dsa->writemask[1]; 322 323 r600_set_stencil_ref(ctx, &ref); 324 } 325 326 static void r600_bind_dsa_state(struct pipe_context *ctx, void *state) 327 { 328 struct r600_context *rctx = (struct r600_context *)ctx; 329 struct r600_dsa_state *dsa = state; 330 struct r600_stencil_ref ref; 331 332 if (!state) { 333 r600_set_cso_state_with_cb(rctx, &rctx->dsa_state, NULL, NULL); 334 return; 335 } 336 337 r600_set_cso_state_with_cb(rctx, &rctx->dsa_state, dsa, &dsa->buffer); 338 339 ref.ref_value[0] = rctx->stencil_ref.pipe_state.ref_value[0]; 340 ref.ref_value[1] = rctx->stencil_ref.pipe_state.ref_value[1]; 341 ref.valuemask[0] = dsa->valuemask[0]; 342 ref.valuemask[1] = dsa->valuemask[1]; 343 ref.writemask[0] = dsa->writemask[0]; 344 ref.writemask[1] = dsa->writemask[1]; 345 if (rctx->zwritemask != dsa->zwritemask) { 346 rctx->zwritemask = dsa->zwritemask; 347 if (rctx->b.chip_class >= EVERGREEN) { 348 /* work around some issue when not writing to zbuffer 349 * we are having lockup on evergreen so do not enable 350 * hyperz when not writing zbuffer 351 */ 352 r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom); 353 } 354 } 355 356 r600_set_stencil_ref(ctx, &ref); 357 358 /* Update alphatest state. */ 359 if (rctx->alphatest_state.sx_alpha_test_control != dsa->sx_alpha_test_control || 360 rctx->alphatest_state.sx_alpha_ref != dsa->alpha_ref) { 361 rctx->alphatest_state.sx_alpha_test_control = dsa->sx_alpha_test_control; 362 rctx->alphatest_state.sx_alpha_ref = dsa->alpha_ref; 363 r600_mark_atom_dirty(rctx, &rctx->alphatest_state.atom); 364 } 365 } 366 367 static void r600_bind_rs_state(struct pipe_context *ctx, void *state) 368 { 369 struct r600_rasterizer_state *rs = (struct r600_rasterizer_state *)state; 370 struct r600_context *rctx = (struct r600_context *)ctx; 371 372 if (!state) 373 return; 374 375 rctx->rasterizer = rs; 376 377 r600_set_cso_state_with_cb(rctx, &rctx->rasterizer_state, rs, &rs->buffer); 378 379 if (rs->offset_enable && 380 (rs->offset_units != rctx->poly_offset_state.offset_units || 381 rs->offset_scale != rctx->poly_offset_state.offset_scale || 382 rs->offset_units_unscaled != rctx->poly_offset_state.offset_units_unscaled)) { 383 rctx->poly_offset_state.offset_units = rs->offset_units; 384 rctx->poly_offset_state.offset_scale = rs->offset_scale; 385 rctx->poly_offset_state.offset_units_unscaled = rs->offset_units_unscaled; 386 r600_mark_atom_dirty(rctx, &rctx->poly_offset_state.atom); 387 } 388 389 /* Update clip_misc_state. */ 390 if (rctx->clip_misc_state.pa_cl_clip_cntl != rs->pa_cl_clip_cntl || 391 rctx->clip_misc_state.clip_plane_enable != rs->clip_plane_enable) { 392 rctx->clip_misc_state.pa_cl_clip_cntl = rs->pa_cl_clip_cntl; 393 rctx->clip_misc_state.clip_plane_enable = rs->clip_plane_enable; 394 r600_mark_atom_dirty(rctx, &rctx->clip_misc_state.atom); 395 } 396 397 r600_viewport_set_rast_deps(&rctx->b, rs->scissor_enable, rs->clip_halfz); 398 399 /* Re-emit PA_SC_LINE_STIPPLE. */ 400 rctx->last_primitive_type = -1; 401 } 402 403 static void r600_delete_rs_state(struct pipe_context *ctx, void *state) 404 { 405 struct r600_rasterizer_state *rs = (struct r600_rasterizer_state *)state; 406 407 r600_release_command_buffer(&rs->buffer); 408 FREE(rs); 409 } 410 411 static void r600_sampler_view_destroy(struct pipe_context *ctx, 412 struct pipe_sampler_view *state) 413 { 414 struct r600_pipe_sampler_view *view = (struct r600_pipe_sampler_view *)state; 415 416 if (view->tex_resource->gpu_address && 417 view->tex_resource->b.b.target == PIPE_BUFFER) 418 LIST_DELINIT(&view->list); 419 420 pipe_resource_reference(&state->texture, NULL); 421 FREE(view); 422 } 423 424 void r600_sampler_states_dirty(struct r600_context *rctx, 425 struct r600_sampler_states *state) 426 { 427 if (state->dirty_mask) { 428 if (state->dirty_mask & state->has_bordercolor_mask) { 429 rctx->b.flags |= R600_CONTEXT_WAIT_3D_IDLE; 430 } 431 state->atom.num_dw = 432 util_bitcount(state->dirty_mask & state->has_bordercolor_mask) * 11 + 433 util_bitcount(state->dirty_mask & ~state->has_bordercolor_mask) * 5; 434 r600_mark_atom_dirty(rctx, &state->atom); 435 } 436 } 437 438 static void r600_bind_sampler_states(struct pipe_context *pipe, 439 enum pipe_shader_type shader, 440 unsigned start, 441 unsigned count, void **states) 442 { 443 struct r600_context *rctx = (struct r600_context *)pipe; 444 struct r600_textures_info *dst = &rctx->samplers[shader]; 445 struct r600_pipe_sampler_state **rstates = (struct r600_pipe_sampler_state**)states; 446 int seamless_cube_map = -1; 447 unsigned i; 448 /* This sets 1-bit for states with index >= count. */ 449 uint32_t disable_mask = ~((1ull << count) - 1); 450 /* These are the new states set by this function. */ 451 uint32_t new_mask = 0; 452 453 assert(start == 0); /* XXX fix below */ 454 455 if (!states) { 456 disable_mask = ~0u; 457 count = 0; 458 } 459 460 for (i = 0; i < count; i++) { 461 struct r600_pipe_sampler_state *rstate = rstates[i]; 462 463 if (rstate == dst->states.states[i]) { 464 continue; 465 } 466 467 if (rstate) { 468 if (rstate->border_color_use) { 469 dst->states.has_bordercolor_mask |= 1 << i; 470 } else { 471 dst->states.has_bordercolor_mask &= ~(1 << i); 472 } 473 seamless_cube_map = rstate->seamless_cube_map; 474 475 new_mask |= 1 << i; 476 } else { 477 disable_mask |= 1 << i; 478 } 479 } 480 481 memcpy(dst->states.states, rstates, sizeof(void*) * count); 482 memset(dst->states.states + count, 0, sizeof(void*) * (NUM_TEX_UNITS - count)); 483 484 dst->states.enabled_mask &= ~disable_mask; 485 dst->states.dirty_mask &= dst->states.enabled_mask; 486 dst->states.enabled_mask |= new_mask; 487 dst->states.dirty_mask |= new_mask; 488 dst->states.has_bordercolor_mask &= dst->states.enabled_mask; 489 490 r600_sampler_states_dirty(rctx, &dst->states); 491 492 /* Seamless cubemap state. */ 493 if (rctx->b.chip_class <= R700 && 494 seamless_cube_map != -1 && 495 seamless_cube_map != rctx->seamless_cube_map.enabled) { 496 /* change in TA_CNTL_AUX need a pipeline flush */ 497 rctx->b.flags |= R600_CONTEXT_WAIT_3D_IDLE; 498 rctx->seamless_cube_map.enabled = seamless_cube_map; 499 r600_mark_atom_dirty(rctx, &rctx->seamless_cube_map.atom); 500 } 501 } 502 503 static void r600_delete_sampler_state(struct pipe_context *ctx, void *state) 504 { 505 free(state); 506 } 507 508 static void r600_delete_blend_state(struct pipe_context *ctx, void *state) 509 { 510 struct r600_context *rctx = (struct r600_context *)ctx; 511 struct r600_blend_state *blend = (struct r600_blend_state*)state; 512 513 if (rctx->blend_state.cso == state) { 514 ctx->bind_blend_state(ctx, NULL); 515 } 516 517 r600_release_command_buffer(&blend->buffer); 518 r600_release_command_buffer(&blend->buffer_no_blend); 519 FREE(blend); 520 } 521 522 static void r600_delete_dsa_state(struct pipe_context *ctx, void *state) 523 { 524 struct r600_context *rctx = (struct r600_context *)ctx; 525 struct r600_dsa_state *dsa = (struct r600_dsa_state *)state; 526 527 if (rctx->dsa_state.cso == state) { 528 ctx->bind_depth_stencil_alpha_state(ctx, NULL); 529 } 530 531 r600_release_command_buffer(&dsa->buffer); 532 free(dsa); 533 } 534 535 static void r600_bind_vertex_elements(struct pipe_context *ctx, void *state) 536 { 537 struct r600_context *rctx = (struct r600_context *)ctx; 538 539 r600_set_cso_state(rctx, &rctx->vertex_fetch_shader, state); 540 } 541 542 static void r600_delete_vertex_elements(struct pipe_context *ctx, void *state) 543 { 544 struct r600_fetch_shader *shader = (struct r600_fetch_shader*)state; 545 r600_resource_reference(&shader->buffer, NULL); 546 FREE(shader); 547 } 548 549 void r600_vertex_buffers_dirty(struct r600_context *rctx) 550 { 551 if (rctx->vertex_buffer_state.dirty_mask) { 552 rctx->vertex_buffer_state.atom.num_dw = (rctx->b.chip_class >= EVERGREEN ? 12 : 11) * 553 util_bitcount(rctx->vertex_buffer_state.dirty_mask); 554 r600_mark_atom_dirty(rctx, &rctx->vertex_buffer_state.atom); 555 } 556 } 557 558 static void r600_set_vertex_buffers(struct pipe_context *ctx, 559 unsigned start_slot, unsigned count, 560 const struct pipe_vertex_buffer *input) 561 { 562 struct r600_context *rctx = (struct r600_context *)ctx; 563 struct r600_vertexbuf_state *state = &rctx->vertex_buffer_state; 564 struct pipe_vertex_buffer *vb = state->vb + start_slot; 565 unsigned i; 566 uint32_t disable_mask = 0; 567 /* These are the new buffers set by this function. */ 568 uint32_t new_buffer_mask = 0; 569 570 /* Set vertex buffers. */ 571 if (input) { 572 for (i = 0; i < count; i++) { 573 if (memcmp(&input[i], &vb[i], sizeof(struct pipe_vertex_buffer))) { 574 if (input[i].buffer.resource) { 575 vb[i].stride = input[i].stride; 576 vb[i].buffer_offset = input[i].buffer_offset; 577 pipe_resource_reference(&vb[i].buffer.resource, input[i].buffer.resource); 578 new_buffer_mask |= 1 << i; 579 r600_context_add_resource_size(ctx, input[i].buffer.resource); 580 } else { 581 pipe_resource_reference(&vb[i].buffer.resource, NULL); 582 disable_mask |= 1 << i; 583 } 584 } 585 } 586 } else { 587 for (i = 0; i < count; i++) { 588 pipe_resource_reference(&vb[i].buffer.resource, NULL); 589 } 590 disable_mask = ((1ull << count) - 1); 591 } 592 593 disable_mask <<= start_slot; 594 new_buffer_mask <<= start_slot; 595 596 rctx->vertex_buffer_state.enabled_mask &= ~disable_mask; 597 rctx->vertex_buffer_state.dirty_mask &= rctx->vertex_buffer_state.enabled_mask; 598 rctx->vertex_buffer_state.enabled_mask |= new_buffer_mask; 599 rctx->vertex_buffer_state.dirty_mask |= new_buffer_mask; 600 601 r600_vertex_buffers_dirty(rctx); 602 } 603 604 void r600_sampler_views_dirty(struct r600_context *rctx, 605 struct r600_samplerview_state *state) 606 { 607 if (state->dirty_mask) { 608 state->atom.num_dw = (rctx->b.chip_class >= EVERGREEN ? 14 : 13) * 609 util_bitcount(state->dirty_mask); 610 r600_mark_atom_dirty(rctx, &state->atom); 611 } 612 } 613 614 static void r600_set_sampler_views(struct pipe_context *pipe, 615 enum pipe_shader_type shader, 616 unsigned start, unsigned count, 617 struct pipe_sampler_view **views) 618 { 619 struct r600_context *rctx = (struct r600_context *) pipe; 620 struct r600_textures_info *dst = &rctx->samplers[shader]; 621 struct r600_pipe_sampler_view **rviews = (struct r600_pipe_sampler_view **)views; 622 uint32_t dirty_sampler_states_mask = 0; 623 unsigned i; 624 /* This sets 1-bit for textures with index >= count. */ 625 uint32_t disable_mask = ~((1ull << count) - 1); 626 /* These are the new textures set by this function. */ 627 uint32_t new_mask = 0; 628 629 /* Set textures with index >= count to NULL. */ 630 uint32_t remaining_mask; 631 632 assert(start == 0); /* XXX fix below */ 633 634 if (!views) { 635 disable_mask = ~0u; 636 count = 0; 637 } 638 639 remaining_mask = dst->views.enabled_mask & disable_mask; 640 641 while (remaining_mask) { 642 i = u_bit_scan(&remaining_mask); 643 assert(dst->views.views[i]); 644 645 pipe_sampler_view_reference((struct pipe_sampler_view **)&dst->views.views[i], NULL); 646 } 647 648 for (i = 0; i < count; i++) { 649 if (rviews[i] == dst->views.views[i]) { 650 continue; 651 } 652 653 if (rviews[i]) { 654 struct r600_texture *rtex = 655 (struct r600_texture*)rviews[i]->base.texture; 656 bool is_buffer = rviews[i]->base.texture->target == PIPE_BUFFER; 657 658 if (!is_buffer && rtex->db_compatible) { 659 dst->views.compressed_depthtex_mask |= 1 << i; 660 } else { 661 dst->views.compressed_depthtex_mask &= ~(1 << i); 662 } 663 664 /* Track compressed colorbuffers. */ 665 if (!is_buffer && rtex->cmask.size) { 666 dst->views.compressed_colortex_mask |= 1 << i; 667 } else { 668 dst->views.compressed_colortex_mask &= ~(1 << i); 669 } 670 671 /* Changing from array to non-arrays textures and vice versa requires 672 * updating TEX_ARRAY_OVERRIDE in sampler states on R6xx-R7xx. */ 673 if (rctx->b.chip_class <= R700 && 674 (dst->states.enabled_mask & (1 << i)) && 675 (rviews[i]->base.texture->target == PIPE_TEXTURE_1D_ARRAY || 676 rviews[i]->base.texture->target == PIPE_TEXTURE_2D_ARRAY) != dst->is_array_sampler[i]) { 677 dirty_sampler_states_mask |= 1 << i; 678 } 679 680 pipe_sampler_view_reference((struct pipe_sampler_view **)&dst->views.views[i], views[i]); 681 new_mask |= 1 << i; 682 r600_context_add_resource_size(pipe, views[i]->texture); 683 } else { 684 pipe_sampler_view_reference((struct pipe_sampler_view **)&dst->views.views[i], NULL); 685 disable_mask |= 1 << i; 686 } 687 } 688 689 dst->views.enabled_mask &= ~disable_mask; 690 dst->views.dirty_mask &= dst->views.enabled_mask; 691 dst->views.enabled_mask |= new_mask; 692 dst->views.dirty_mask |= new_mask; 693 dst->views.compressed_depthtex_mask &= dst->views.enabled_mask; 694 dst->views.compressed_colortex_mask &= dst->views.enabled_mask; 695 dst->views.dirty_buffer_constants = TRUE; 696 r600_sampler_views_dirty(rctx, &dst->views); 697 698 if (dirty_sampler_states_mask) { 699 dst->states.dirty_mask |= dirty_sampler_states_mask; 700 r600_sampler_states_dirty(rctx, &dst->states); 701 } 702 } 703 704 static void r600_update_compressed_colortex_mask(struct r600_samplerview_state *views) 705 { 706 uint32_t mask = views->enabled_mask; 707 708 while (mask) { 709 unsigned i = u_bit_scan(&mask); 710 struct pipe_resource *res = views->views[i]->base.texture; 711 712 if (res && res->target != PIPE_BUFFER) { 713 struct r600_texture *rtex = (struct r600_texture *)res; 714 715 if (rtex->cmask.size) { 716 views->compressed_colortex_mask |= 1 << i; 717 } else { 718 views->compressed_colortex_mask &= ~(1 << i); 719 } 720 } 721 } 722 } 723 724 static int r600_get_hw_atomic_count(const struct pipe_context *ctx, 725 enum pipe_shader_type shader) 726 { 727 const struct r600_context *rctx = (struct r600_context *)ctx; 728 int value = 0; 729 switch (shader) { 730 case PIPE_SHADER_FRAGMENT: 731 case PIPE_SHADER_COMPUTE: 732 default: 733 break; 734 case PIPE_SHADER_VERTEX: 735 value = rctx->ps_shader->info.file_count[TGSI_FILE_HW_ATOMIC]; 736 break; 737 case PIPE_SHADER_GEOMETRY: 738 value = rctx->ps_shader->info.file_count[TGSI_FILE_HW_ATOMIC] + 739 rctx->vs_shader->info.file_count[TGSI_FILE_HW_ATOMIC]; 740 break; 741 case PIPE_SHADER_TESS_EVAL: 742 value = rctx->ps_shader->info.file_count[TGSI_FILE_HW_ATOMIC] + 743 rctx->vs_shader->info.file_count[TGSI_FILE_HW_ATOMIC] + 744 (rctx->gs_shader ? rctx->gs_shader->info.file_count[TGSI_FILE_HW_ATOMIC] : 0); 745 break; 746 case PIPE_SHADER_TESS_CTRL: 747 value = rctx->ps_shader->info.file_count[TGSI_FILE_HW_ATOMIC] + 748 rctx->vs_shader->info.file_count[TGSI_FILE_HW_ATOMIC] + 749 (rctx->gs_shader ? rctx->gs_shader->info.file_count[TGSI_FILE_HW_ATOMIC] : 0) + 750 rctx->tes_shader->info.file_count[TGSI_FILE_HW_ATOMIC]; 751 break; 752 } 753 return value; 754 } 755 756 static void r600_update_compressed_colortex_mask_images(struct r600_image_state *images) 757 { 758 uint32_t mask = images->enabled_mask; 759 760 while (mask) { 761 unsigned i = u_bit_scan(&mask); 762 struct pipe_resource *res = images->views[i].base.resource; 763 764 if (res && res->target != PIPE_BUFFER) { 765 struct r600_texture *rtex = (struct r600_texture *)res; 766 767 if (rtex->cmask.size) { 768 images->compressed_colortex_mask |= 1 << i; 769 } else { 770 images->compressed_colortex_mask &= ~(1 << i); 771 } 772 } 773 } 774 } 775 776 /* Compute the key for the hw shader variant */ 777 static inline void r600_shader_selector_key(const struct pipe_context *ctx, 778 const struct r600_pipe_shader_selector *sel, 779 union r600_shader_key *key) 780 { 781 const struct r600_context *rctx = (struct r600_context *)ctx; 782 memset(key, 0, sizeof(*key)); 783 784 switch (sel->type) { 785 case PIPE_SHADER_VERTEX: { 786 key->vs.as_ls = (rctx->tes_shader != NULL); 787 if (!key->vs.as_ls) 788 key->vs.as_es = (rctx->gs_shader != NULL); 789 790 if (rctx->ps_shader->current->shader.gs_prim_id_input && !rctx->gs_shader) { 791 key->vs.as_gs_a = true; 792 key->vs.prim_id_out = rctx->ps_shader->current->shader.input[rctx->ps_shader->current->shader.ps_prim_id_input].spi_sid; 793 } 794 key->vs.first_atomic_counter = r600_get_hw_atomic_count(ctx, PIPE_SHADER_VERTEX); 795 break; 796 } 797 case PIPE_SHADER_GEOMETRY: 798 key->gs.first_atomic_counter = r600_get_hw_atomic_count(ctx, PIPE_SHADER_GEOMETRY); 799 key->gs.tri_strip_adj_fix = rctx->gs_tri_strip_adj_fix; 800 break; 801 case PIPE_SHADER_FRAGMENT: { 802 if (rctx->ps_shader->info.images_declared) 803 key->ps.image_size_const_offset = util_last_bit(rctx->samplers[PIPE_SHADER_FRAGMENT].views.enabled_mask); 804 key->ps.first_atomic_counter = r600_get_hw_atomic_count(ctx, PIPE_SHADER_FRAGMENT); 805 key->ps.color_two_side = rctx->rasterizer && rctx->rasterizer->two_side; 806 key->ps.alpha_to_one = rctx->alpha_to_one && 807 rctx->rasterizer && rctx->rasterizer->multisample_enable && 808 !rctx->framebuffer.cb0_is_integer; 809 key->ps.nr_cbufs = rctx->framebuffer.state.nr_cbufs; 810 /* Dual-source blending only makes sense with nr_cbufs == 1. */ 811 if (key->ps.nr_cbufs == 1 && rctx->dual_src_blend) 812 key->ps.nr_cbufs = 2; 813 break; 814 } 815 case PIPE_SHADER_TESS_EVAL: 816 key->tes.as_es = (rctx->gs_shader != NULL); 817 key->tes.first_atomic_counter = r600_get_hw_atomic_count(ctx, PIPE_SHADER_TESS_EVAL); 818 break; 819 case PIPE_SHADER_TESS_CTRL: 820 key->tcs.prim_mode = rctx->tes_shader->info.properties[TGSI_PROPERTY_TES_PRIM_MODE]; 821 key->tcs.first_atomic_counter = r600_get_hw_atomic_count(ctx, PIPE_SHADER_TESS_CTRL); 822 break; 823 case PIPE_SHADER_COMPUTE: 824 break; 825 default: 826 assert(0); 827 } 828 } 829 830 /* Select the hw shader variant depending on the current state. 831 * (*dirty) is set to 1 if current variant was changed */ 832 int r600_shader_select(struct pipe_context *ctx, 833 struct r600_pipe_shader_selector* sel, 834 bool *dirty) 835 { 836 union r600_shader_key key; 837 struct r600_pipe_shader * shader = NULL; 838 int r; 839 840 r600_shader_selector_key(ctx, sel, &key); 841 842 /* Check if we don't need to change anything. 843 * This path is also used for most shaders that don't need multiple 844 * variants, it will cost just a computation of the key and this 845 * test. */ 846 if (likely(sel->current && memcmp(&sel->current->key, &key, sizeof(key)) == 0)) { 847 return 0; 848 } 849 850 /* lookup if we have other variants in the list */ 851 if (sel->num_shaders > 1) { 852 struct r600_pipe_shader *p = sel->current, *c = p->next_variant; 853 854 while (c && memcmp(&c->key, &key, sizeof(key)) != 0) { 855 p = c; 856 c = c->next_variant; 857 } 858 859 if (c) { 860 p->next_variant = c->next_variant; 861 shader = c; 862 } 863 } 864 865 if (unlikely(!shader)) { 866 shader = CALLOC(1, sizeof(struct r600_pipe_shader)); 867 shader->selector = sel; 868 869 r = r600_pipe_shader_create(ctx, shader, key); 870 if (unlikely(r)) { 871 R600_ERR("Failed to build shader variant (type=%u) %d\n", 872 sel->type, r); 873 sel->current = NULL; 874 FREE(shader); 875 return r; 876 } 877 878 /* We don't know the value of nr_ps_max_color_exports until we built 879 * at least one variant, so we may need to recompute the key after 880 * building first variant. */ 881 if (sel->type == PIPE_SHADER_FRAGMENT && 882 sel->num_shaders == 0) { 883 sel->nr_ps_max_color_exports = shader->shader.nr_ps_max_color_exports; 884 r600_shader_selector_key(ctx, sel, &key); 885 } 886 887 memcpy(&shader->key, &key, sizeof(key)); 888 sel->num_shaders++; 889 } 890 891 if (dirty) 892 *dirty = true; 893 894 shader->next_variant = sel->current; 895 sel->current = shader; 896 897 return 0; 898 } 899 900 struct r600_pipe_shader_selector *r600_create_shader_state_tokens(struct pipe_context *ctx, 901 const struct tgsi_token *tokens, 902 unsigned pipe_shader_type) 903 { 904 struct r600_pipe_shader_selector *sel = CALLOC_STRUCT(r600_pipe_shader_selector); 905 906 sel->type = pipe_shader_type; 907 sel->tokens = tgsi_dup_tokens(tokens); 908 tgsi_scan_shader(tokens, &sel->info); 909 return sel; 910 } 911 912 static void *r600_create_shader_state(struct pipe_context *ctx, 913 const struct pipe_shader_state *state, 914 unsigned pipe_shader_type) 915 { 916 int i; 917 struct r600_pipe_shader_selector *sel = r600_create_shader_state_tokens(ctx, state->tokens, pipe_shader_type); 918 919 sel->so = state->stream_output; 920 921 switch (pipe_shader_type) { 922 case PIPE_SHADER_GEOMETRY: 923 sel->gs_output_prim = 924 sel->info.properties[TGSI_PROPERTY_GS_OUTPUT_PRIM]; 925 sel->gs_max_out_vertices = 926 sel->info.properties[TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES]; 927 sel->gs_num_invocations = 928 sel->info.properties[TGSI_PROPERTY_GS_INVOCATIONS]; 929 break; 930 case PIPE_SHADER_VERTEX: 931 case PIPE_SHADER_TESS_CTRL: 932 sel->lds_patch_outputs_written_mask = 0; 933 sel->lds_outputs_written_mask = 0; 934 935 for (i = 0; i < sel->info.num_outputs; i++) { 936 unsigned name = sel->info.output_semantic_name[i]; 937 unsigned index = sel->info.output_semantic_index[i]; 938 939 switch (name) { 940 case TGSI_SEMANTIC_TESSINNER: 941 case TGSI_SEMANTIC_TESSOUTER: 942 case TGSI_SEMANTIC_PATCH: 943 sel->lds_patch_outputs_written_mask |= 944 1ull << r600_get_lds_unique_index(name, index); 945 break; 946 default: 947 sel->lds_outputs_written_mask |= 948 1ull << r600_get_lds_unique_index(name, index); 949 } 950 } 951 break; 952 default: 953 break; 954 } 955 956 return sel; 957 } 958 959 static void *r600_create_ps_state(struct pipe_context *ctx, 960 const struct pipe_shader_state *state) 961 { 962 return r600_create_shader_state(ctx, state, PIPE_SHADER_FRAGMENT); 963 } 964 965 static void *r600_create_vs_state(struct pipe_context *ctx, 966 const struct pipe_shader_state *state) 967 { 968 return r600_create_shader_state(ctx, state, PIPE_SHADER_VERTEX); 969 } 970 971 static void *r600_create_gs_state(struct pipe_context *ctx, 972 const struct pipe_shader_state *state) 973 { 974 return r600_create_shader_state(ctx, state, PIPE_SHADER_GEOMETRY); 975 } 976 977 static void *r600_create_tcs_state(struct pipe_context *ctx, 978 const struct pipe_shader_state *state) 979 { 980 return r600_create_shader_state(ctx, state, PIPE_SHADER_TESS_CTRL); 981 } 982 983 static void *r600_create_tes_state(struct pipe_context *ctx, 984 const struct pipe_shader_state *state) 985 { 986 return r600_create_shader_state(ctx, state, PIPE_SHADER_TESS_EVAL); 987 } 988 989 static void r600_bind_ps_state(struct pipe_context *ctx, void *state) 990 { 991 struct r600_context *rctx = (struct r600_context *)ctx; 992 993 if (!state) 994 state = rctx->dummy_pixel_shader; 995 996 rctx->ps_shader = (struct r600_pipe_shader_selector *)state; 997 } 998 999 static struct tgsi_shader_info *r600_get_vs_info(struct r600_context *rctx) 1000 { 1001 if (rctx->gs_shader) 1002 return &rctx->gs_shader->info; 1003 else if (rctx->tes_shader) 1004 return &rctx->tes_shader->info; 1005 else if (rctx->vs_shader) 1006 return &rctx->vs_shader->info; 1007 else 1008 return NULL; 1009 } 1010 1011 static void r600_bind_vs_state(struct pipe_context *ctx, void *state) 1012 { 1013 struct r600_context *rctx = (struct r600_context *)ctx; 1014 1015 if (!state || rctx->vs_shader == state) 1016 return; 1017 1018 rctx->vs_shader = (struct r600_pipe_shader_selector *)state; 1019 r600_update_vs_writes_viewport_index(&rctx->b, r600_get_vs_info(rctx)); 1020 rctx->b.streamout.stride_in_dw = rctx->vs_shader->so.stride; 1021 } 1022 1023 static void r600_bind_gs_state(struct pipe_context *ctx, void *state) 1024 { 1025 struct r600_context *rctx = (struct r600_context *)ctx; 1026 1027 if (state == rctx->gs_shader) 1028 return; 1029 1030 rctx->gs_shader = (struct r600_pipe_shader_selector *)state; 1031 r600_update_vs_writes_viewport_index(&rctx->b, r600_get_vs_info(rctx)); 1032 1033 if (!state) 1034 return; 1035 rctx->b.streamout.stride_in_dw = rctx->gs_shader->so.stride; 1036 } 1037 1038 static void r600_bind_tcs_state(struct pipe_context *ctx, void *state) 1039 { 1040 struct r600_context *rctx = (struct r600_context *)ctx; 1041 1042 rctx->tcs_shader = (struct r600_pipe_shader_selector *)state; 1043 } 1044 1045 static void r600_bind_tes_state(struct pipe_context *ctx, void *state) 1046 { 1047 struct r600_context *rctx = (struct r600_context *)ctx; 1048 1049 if (state == rctx->tes_shader) 1050 return; 1051 1052 rctx->tes_shader = (struct r600_pipe_shader_selector *)state; 1053 r600_update_vs_writes_viewport_index(&rctx->b, r600_get_vs_info(rctx)); 1054 1055 if (!state) 1056 return; 1057 rctx->b.streamout.stride_in_dw = rctx->tes_shader->so.stride; 1058 } 1059 1060 void r600_delete_shader_selector(struct pipe_context *ctx, 1061 struct r600_pipe_shader_selector *sel) 1062 { 1063 struct r600_pipe_shader *p = sel->current, *c; 1064 while (p) { 1065 c = p->next_variant; 1066 r600_pipe_shader_destroy(ctx, p); 1067 free(p); 1068 p = c; 1069 } 1070 1071 free(sel->tokens); 1072 free(sel); 1073 } 1074 1075 1076 static void r600_delete_ps_state(struct pipe_context *ctx, void *state) 1077 { 1078 struct r600_context *rctx = (struct r600_context *)ctx; 1079 struct r600_pipe_shader_selector *sel = (struct r600_pipe_shader_selector *)state; 1080 1081 if (rctx->ps_shader == sel) { 1082 rctx->ps_shader = NULL; 1083 } 1084 1085 r600_delete_shader_selector(ctx, sel); 1086 } 1087 1088 static void r600_delete_vs_state(struct pipe_context *ctx, void *state) 1089 { 1090 struct r600_context *rctx = (struct r600_context *)ctx; 1091 struct r600_pipe_shader_selector *sel = (struct r600_pipe_shader_selector *)state; 1092 1093 if (rctx->vs_shader == sel) { 1094 rctx->vs_shader = NULL; 1095 } 1096 1097 r600_delete_shader_selector(ctx, sel); 1098 } 1099 1100 1101 static void r600_delete_gs_state(struct pipe_context *ctx, void *state) 1102 { 1103 struct r600_context *rctx = (struct r600_context *)ctx; 1104 struct r600_pipe_shader_selector *sel = (struct r600_pipe_shader_selector *)state; 1105 1106 if (rctx->gs_shader == sel) { 1107 rctx->gs_shader = NULL; 1108 } 1109 1110 r600_delete_shader_selector(ctx, sel); 1111 } 1112 1113 static void r600_delete_tcs_state(struct pipe_context *ctx, void *state) 1114 { 1115 struct r600_context *rctx = (struct r600_context *)ctx; 1116 struct r600_pipe_shader_selector *sel = (struct r600_pipe_shader_selector *)state; 1117 1118 if (rctx->tcs_shader == sel) { 1119 rctx->tcs_shader = NULL; 1120 } 1121 1122 r600_delete_shader_selector(ctx, sel); 1123 } 1124 1125 static void r600_delete_tes_state(struct pipe_context *ctx, void *state) 1126 { 1127 struct r600_context *rctx = (struct r600_context *)ctx; 1128 struct r600_pipe_shader_selector *sel = (struct r600_pipe_shader_selector *)state; 1129 1130 if (rctx->tes_shader == sel) { 1131 rctx->tes_shader = NULL; 1132 } 1133 1134 r600_delete_shader_selector(ctx, sel); 1135 } 1136 1137 void r600_constant_buffers_dirty(struct r600_context *rctx, struct r600_constbuf_state *state) 1138 { 1139 if (state->dirty_mask) { 1140 state->atom.num_dw = rctx->b.chip_class >= EVERGREEN ? util_bitcount(state->dirty_mask)*20 1141 : util_bitcount(state->dirty_mask)*19; 1142 r600_mark_atom_dirty(rctx, &state->atom); 1143 } 1144 } 1145 1146 static void r600_set_constant_buffer(struct pipe_context *ctx, 1147 enum pipe_shader_type shader, uint index, 1148 const struct pipe_constant_buffer *input) 1149 { 1150 struct r600_context *rctx = (struct r600_context *)ctx; 1151 struct r600_constbuf_state *state = &rctx->constbuf_state[shader]; 1152 struct pipe_constant_buffer *cb; 1153 const uint8_t *ptr; 1154 1155 /* Note that the state tracker can unbind constant buffers by 1156 * passing NULL here. 1157 */ 1158 if (unlikely(!input || (!input->buffer && !input->user_buffer))) { 1159 state->enabled_mask &= ~(1 << index); 1160 state->dirty_mask &= ~(1 << index); 1161 pipe_resource_reference(&state->cb[index].buffer, NULL); 1162 return; 1163 } 1164 1165 cb = &state->cb[index]; 1166 cb->buffer_size = input->buffer_size; 1167 1168 ptr = input->user_buffer; 1169 1170 if (ptr) { 1171 /* Upload the user buffer. */ 1172 if (R600_BIG_ENDIAN) { 1173 uint32_t *tmpPtr; 1174 unsigned i, size = input->buffer_size; 1175 1176 if (!(tmpPtr = malloc(size))) { 1177 R600_ERR("Failed to allocate BE swap buffer.\n"); 1178 return; 1179 } 1180 1181 for (i = 0; i < size / 4; ++i) { 1182 tmpPtr[i] = util_cpu_to_le32(((uint32_t *)ptr)[i]); 1183 } 1184 1185 u_upload_data(ctx->stream_uploader, 0, size, 256, 1186 tmpPtr, &cb->buffer_offset, &cb->buffer); 1187 free(tmpPtr); 1188 } else { 1189 u_upload_data(ctx->stream_uploader, 0, 1190 input->buffer_size, 256, ptr, 1191 &cb->buffer_offset, &cb->buffer); 1192 } 1193 /* account it in gtt */ 1194 rctx->b.gtt += input->buffer_size; 1195 } else { 1196 /* Setup the hw buffer. */ 1197 cb->buffer_offset = input->buffer_offset; 1198 pipe_resource_reference(&cb->buffer, input->buffer); 1199 r600_context_add_resource_size(ctx, input->buffer); 1200 } 1201 1202 state->enabled_mask |= 1 << index; 1203 state->dirty_mask |= 1 << index; 1204 r600_constant_buffers_dirty(rctx, state); 1205 } 1206 1207 static void r600_set_sample_mask(struct pipe_context *pipe, unsigned sample_mask) 1208 { 1209 struct r600_context *rctx = (struct r600_context*)pipe; 1210 1211 if (rctx->sample_mask.sample_mask == (uint16_t)sample_mask) 1212 return; 1213 1214 rctx->sample_mask.sample_mask = sample_mask; 1215 r600_mark_atom_dirty(rctx, &rctx->sample_mask.atom); 1216 } 1217 1218 void r600_update_driver_const_buffers(struct r600_context *rctx, bool compute_only) 1219 { 1220 int sh, size; 1221 void *ptr; 1222 struct pipe_constant_buffer cb; 1223 int start, end; 1224 1225 start = compute_only ? PIPE_SHADER_COMPUTE : 0; 1226 end = compute_only ? PIPE_SHADER_TYPES : PIPE_SHADER_COMPUTE; 1227 1228 for (sh = start; sh < end; sh++) { 1229 struct r600_shader_driver_constants_info *info = &rctx->driver_consts[sh]; 1230 if (!info->vs_ucp_dirty && 1231 !info->texture_const_dirty && 1232 !info->ps_sample_pos_dirty && 1233 !info->tcs_default_levels_dirty && 1234 !info->cs_block_grid_size_dirty) 1235 continue; 1236 1237 ptr = info->constants; 1238 size = info->alloc_size; 1239 if (info->vs_ucp_dirty) { 1240 assert(sh == PIPE_SHADER_VERTEX); 1241 if (!size) { 1242 ptr = rctx->clip_state.state.ucp; 1243 size = R600_UCP_SIZE; 1244 } else { 1245 memcpy(ptr, rctx->clip_state.state.ucp, R600_UCP_SIZE); 1246 } 1247 info->vs_ucp_dirty = false; 1248 } 1249 1250 else if (info->ps_sample_pos_dirty) { 1251 assert(sh == PIPE_SHADER_FRAGMENT); 1252 if (!size) { 1253 ptr = rctx->sample_positions; 1254 size = R600_UCP_SIZE; 1255 } else { 1256 memcpy(ptr, rctx->sample_positions, R600_UCP_SIZE); 1257 } 1258 info->ps_sample_pos_dirty = false; 1259 } 1260 1261 else if (info->cs_block_grid_size_dirty) { 1262 assert(sh == PIPE_SHADER_COMPUTE); 1263 if (!size) { 1264 ptr = rctx->cs_block_grid_sizes; 1265 size = R600_CS_BLOCK_GRID_SIZE; 1266 } else { 1267 memcpy(ptr, rctx->cs_block_grid_sizes, R600_CS_BLOCK_GRID_SIZE); 1268 } 1269 info->cs_block_grid_size_dirty = false; 1270 } 1271 1272 else if (info->tcs_default_levels_dirty) { 1273 /* 1274 * We'd only really need this for default tcs shader. 1275 */ 1276 assert(sh == PIPE_SHADER_TESS_CTRL); 1277 if (!size) { 1278 ptr = rctx->tess_state; 1279 size = R600_TCS_DEFAULT_LEVELS_SIZE; 1280 } else { 1281 memcpy(ptr, rctx->tess_state, R600_TCS_DEFAULT_LEVELS_SIZE); 1282 } 1283 info->tcs_default_levels_dirty = false; 1284 } 1285 1286 if (info->texture_const_dirty) { 1287 assert (ptr); 1288 assert (size); 1289 if (sh == PIPE_SHADER_VERTEX) 1290 memcpy(ptr, rctx->clip_state.state.ucp, R600_UCP_SIZE); 1291 if (sh == PIPE_SHADER_FRAGMENT) 1292 memcpy(ptr, rctx->sample_positions, R600_UCP_SIZE); 1293 if (sh == PIPE_SHADER_COMPUTE) 1294 memcpy(ptr, rctx->cs_block_grid_sizes, R600_CS_BLOCK_GRID_SIZE); 1295 if (sh == PIPE_SHADER_TESS_CTRL) 1296 memcpy(ptr, rctx->tess_state, R600_TCS_DEFAULT_LEVELS_SIZE); 1297 } 1298 info->texture_const_dirty = false; 1299 1300 cb.buffer = NULL; 1301 cb.user_buffer = ptr; 1302 cb.buffer_offset = 0; 1303 cb.buffer_size = size; 1304 rctx->b.b.set_constant_buffer(&rctx->b.b, sh, R600_BUFFER_INFO_CONST_BUFFER, &cb); 1305 pipe_resource_reference(&cb.buffer, NULL); 1306 } 1307 } 1308 1309 static void *r600_alloc_buf_consts(struct r600_context *rctx, int shader_type, 1310 int array_size, uint32_t *base_offset) 1311 { 1312 struct r600_shader_driver_constants_info *info = &rctx->driver_consts[shader_type]; 1313 if (array_size + R600_UCP_SIZE > info->alloc_size) { 1314 info->constants = realloc(info->constants, array_size + R600_UCP_SIZE); 1315 info->alloc_size = array_size + R600_UCP_SIZE; 1316 } 1317 memset(info->constants + (R600_UCP_SIZE / 4), 0, array_size); 1318 info->texture_const_dirty = true; 1319 *base_offset = R600_UCP_SIZE; 1320 return info->constants; 1321 } 1322 /* 1323 * On r600/700 hw we don't have vertex fetch swizzle, though TBO 1324 * doesn't require full swizzles it does need masking and setting alpha 1325 * to one, so we setup a set of 5 constants with the masks + alpha value 1326 * then in the shader, we AND the 4 components with 0xffffffff or 0, 1327 * then OR the alpha with the value given here. 1328 * We use a 6th constant to store the txq buffer size in 1329 * we use 7th slot for number of cube layers in a cube map array. 1330 */ 1331 static void r600_setup_buffer_constants(struct r600_context *rctx, int shader_type) 1332 { 1333 struct r600_textures_info *samplers = &rctx->samplers[shader_type]; 1334 int bits; 1335 uint32_t array_size; 1336 int i, j; 1337 uint32_t *constants; 1338 uint32_t base_offset; 1339 if (!samplers->views.dirty_buffer_constants) 1340 return; 1341 1342 samplers->views.dirty_buffer_constants = FALSE; 1343 1344 bits = util_last_bit(samplers->views.enabled_mask); 1345 array_size = bits * 8 * sizeof(uint32_t); 1346 1347 constants = r600_alloc_buf_consts(rctx, shader_type, array_size, &base_offset); 1348 1349 for (i = 0; i < bits; i++) { 1350 if (samplers->views.enabled_mask & (1 << i)) { 1351 int offset = (base_offset / 4) + i * 8; 1352 const struct util_format_description *desc; 1353 desc = util_format_description(samplers->views.views[i]->base.format); 1354 1355 for (j = 0; j < 4; j++) 1356 if (j < desc->nr_channels) 1357 constants[offset+j] = 0xffffffff; 1358 else 1359 constants[offset+j] = 0x0; 1360 if (desc->nr_channels < 4) { 1361 if (desc->channel[0].pure_integer) 1362 constants[offset+4] = 1; 1363 else 1364 constants[offset+4] = fui(1.0); 1365 } else 1366 constants[offset + 4] = 0; 1367 1368 constants[offset + 5] = samplers->views.views[i]->base.u.buf.size / 1369 util_format_get_blocksize(samplers->views.views[i]->base.format); 1370 constants[offset + 6] = samplers->views.views[i]->base.texture->array_size / 6; 1371 } 1372 } 1373 1374 } 1375 1376 /* On evergreen we store one value 1377 * 1. number of cube layers in a cube map array. 1378 */ 1379 void eg_setup_buffer_constants(struct r600_context *rctx, int shader_type) 1380 { 1381 struct r600_textures_info *samplers = &rctx->samplers[shader_type]; 1382 struct r600_image_state *images = NULL; 1383 int bits, sview_bits, img_bits; 1384 uint32_t array_size; 1385 int i; 1386 uint32_t *constants; 1387 uint32_t base_offset; 1388 1389 if (shader_type == PIPE_SHADER_FRAGMENT) { 1390 images = &rctx->fragment_images; 1391 } else if (shader_type == PIPE_SHADER_COMPUTE) { 1392 images = &rctx->compute_images; 1393 } 1394 1395 if (!samplers->views.dirty_buffer_constants && 1396 !(images && images->dirty_buffer_constants)) 1397 return; 1398 1399 if (images) 1400 images->dirty_buffer_constants = FALSE; 1401 samplers->views.dirty_buffer_constants = FALSE; 1402 1403 bits = sview_bits = util_last_bit(samplers->views.enabled_mask); 1404 if (images) 1405 bits += util_last_bit(images->enabled_mask); 1406 img_bits = bits; 1407 1408 array_size = bits * sizeof(uint32_t); 1409 1410 constants = r600_alloc_buf_consts(rctx, shader_type, array_size, 1411 &base_offset); 1412 1413 for (i = 0; i < sview_bits; i++) { 1414 if (samplers->views.enabled_mask & (1 << i)) { 1415 uint32_t offset = (base_offset / 4) + i; 1416 constants[offset] = samplers->views.views[i]->base.texture->array_size / 6; 1417 } 1418 } 1419 if (images) { 1420 for (i = sview_bits; i < img_bits; i++) { 1421 int idx = i - sview_bits; 1422 if (images->enabled_mask & (1 << idx)) { 1423 uint32_t offset = (base_offset / 4) + i; 1424 constants[offset] = images->views[i].base.resource->array_size / 6; 1425 } 1426 } 1427 } 1428 } 1429 1430 /* set sample xy locations as array of fragment shader constants */ 1431 void r600_set_sample_locations_constant_buffer(struct r600_context *rctx) 1432 { 1433 int i; 1434 struct pipe_context *ctx = &rctx->b.b; 1435 1436 assert(rctx->framebuffer.nr_samples < R600_UCP_SIZE); 1437 assert(rctx->framebuffer.nr_samples <= ARRAY_SIZE(rctx->sample_positions)/4); 1438 1439 memset(rctx->sample_positions, 0, 4 * 4 * 16); 1440 for (i = 0; i < rctx->framebuffer.nr_samples; i++) { 1441 ctx->get_sample_position(ctx, rctx->framebuffer.nr_samples, i, &rctx->sample_positions[4*i]); 1442 /* Also fill in center-zeroed positions used for interpolateAtSample */ 1443 rctx->sample_positions[4*i + 2] = rctx->sample_positions[4*i + 0] - 0.5f; 1444 rctx->sample_positions[4*i + 3] = rctx->sample_positions[4*i + 1] - 0.5f; 1445 } 1446 1447 rctx->driver_consts[PIPE_SHADER_FRAGMENT].ps_sample_pos_dirty = true; 1448 } 1449 1450 static void update_shader_atom(struct pipe_context *ctx, 1451 struct r600_shader_state *state, 1452 struct r600_pipe_shader *shader) 1453 { 1454 struct r600_context *rctx = (struct r600_context *)ctx; 1455 1456 state->shader = shader; 1457 if (shader) { 1458 state->atom.num_dw = shader->command_buffer.num_dw; 1459 r600_context_add_resource_size(ctx, (struct pipe_resource *)shader->bo); 1460 } else { 1461 state->atom.num_dw = 0; 1462 } 1463 r600_mark_atom_dirty(rctx, &state->atom); 1464 } 1465 1466 static void update_gs_block_state(struct r600_context *rctx, unsigned enable) 1467 { 1468 if (rctx->shader_stages.geom_enable != enable) { 1469 rctx->shader_stages.geom_enable = enable; 1470 r600_mark_atom_dirty(rctx, &rctx->shader_stages.atom); 1471 } 1472 1473 if (rctx->gs_rings.enable != enable) { 1474 rctx->gs_rings.enable = enable; 1475 r600_mark_atom_dirty(rctx, &rctx->gs_rings.atom); 1476 1477 if (enable && !rctx->gs_rings.esgs_ring.buffer) { 1478 unsigned size = 0x1C000; 1479 rctx->gs_rings.esgs_ring.buffer = 1480 pipe_buffer_create(rctx->b.b.screen, 0, 1481 PIPE_USAGE_DEFAULT, size); 1482 rctx->gs_rings.esgs_ring.buffer_size = size; 1483 1484 size = 0x4000000; 1485 1486 rctx->gs_rings.gsvs_ring.buffer = 1487 pipe_buffer_create(rctx->b.b.screen, 0, 1488 PIPE_USAGE_DEFAULT, size); 1489 rctx->gs_rings.gsvs_ring.buffer_size = size; 1490 } 1491 1492 if (enable) { 1493 r600_set_constant_buffer(&rctx->b.b, PIPE_SHADER_GEOMETRY, 1494 R600_GS_RING_CONST_BUFFER, &rctx->gs_rings.esgs_ring); 1495 if (rctx->tes_shader) { 1496 r600_set_constant_buffer(&rctx->b.b, PIPE_SHADER_TESS_EVAL, 1497 R600_GS_RING_CONST_BUFFER, &rctx->gs_rings.gsvs_ring); 1498 } else { 1499 r600_set_constant_buffer(&rctx->b.b, PIPE_SHADER_VERTEX, 1500 R600_GS_RING_CONST_BUFFER, &rctx->gs_rings.gsvs_ring); 1501 } 1502 } else { 1503 r600_set_constant_buffer(&rctx->b.b, PIPE_SHADER_GEOMETRY, 1504 R600_GS_RING_CONST_BUFFER, NULL); 1505 r600_set_constant_buffer(&rctx->b.b, PIPE_SHADER_VERTEX, 1506 R600_GS_RING_CONST_BUFFER, NULL); 1507 r600_set_constant_buffer(&rctx->b.b, PIPE_SHADER_TESS_EVAL, 1508 R600_GS_RING_CONST_BUFFER, NULL); 1509 } 1510 } 1511 } 1512 1513 static void r600_update_clip_state(struct r600_context *rctx, 1514 struct r600_pipe_shader *current) 1515 { 1516 if (current->pa_cl_vs_out_cntl != rctx->clip_misc_state.pa_cl_vs_out_cntl || 1517 current->shader.clip_dist_write != rctx->clip_misc_state.clip_dist_write || 1518 current->shader.cull_dist_write != rctx->clip_misc_state.cull_dist_write || 1519 current->shader.vs_position_window_space != rctx->clip_misc_state.clip_disable || 1520 current->shader.vs_out_viewport != rctx->clip_misc_state.vs_out_viewport) { 1521 rctx->clip_misc_state.pa_cl_vs_out_cntl = current->pa_cl_vs_out_cntl; 1522 rctx->clip_misc_state.clip_dist_write = current->shader.clip_dist_write; 1523 rctx->clip_misc_state.cull_dist_write = current->shader.cull_dist_write; 1524 rctx->clip_misc_state.clip_disable = current->shader.vs_position_window_space; 1525 rctx->clip_misc_state.vs_out_viewport = current->shader.vs_out_viewport; 1526 r600_mark_atom_dirty(rctx, &rctx->clip_misc_state.atom); 1527 } 1528 } 1529 1530 static void r600_generate_fixed_func_tcs(struct r600_context *rctx) 1531 { 1532 struct ureg_src const0, const1; 1533 struct ureg_dst tessouter, tessinner; 1534 struct ureg_program *ureg = ureg_create(PIPE_SHADER_TESS_CTRL); 1535 1536 if (!ureg) 1537 return; /* if we get here, we're screwed */ 1538 1539 assert(!rctx->fixed_func_tcs_shader); 1540 1541 ureg_DECL_constant2D(ureg, 0, 1, R600_BUFFER_INFO_CONST_BUFFER); 1542 const0 = ureg_src_dimension(ureg_src_register(TGSI_FILE_CONSTANT, 0), 1543 R600_BUFFER_INFO_CONST_BUFFER); 1544 const1 = ureg_src_dimension(ureg_src_register(TGSI_FILE_CONSTANT, 1), 1545 R600_BUFFER_INFO_CONST_BUFFER); 1546 1547 tessouter = ureg_DECL_output(ureg, TGSI_SEMANTIC_TESSOUTER, 0); 1548 tessinner = ureg_DECL_output(ureg, TGSI_SEMANTIC_TESSINNER, 0); 1549 1550 ureg_MOV(ureg, tessouter, const0); 1551 ureg_MOV(ureg, tessinner, const1); 1552 ureg_END(ureg); 1553 1554 rctx->fixed_func_tcs_shader = 1555 ureg_create_shader_and_destroy(ureg, &rctx->b.b); 1556 } 1557 1558 void r600_update_compressed_resource_state(struct r600_context *rctx, bool compute_only) 1559 { 1560 unsigned i; 1561 unsigned counter; 1562 1563 counter = p_atomic_read(&rctx->screen->b.compressed_colortex_counter); 1564 if (counter != rctx->b.last_compressed_colortex_counter) { 1565 rctx->b.last_compressed_colortex_counter = counter; 1566 1567 if (compute_only) { 1568 r600_update_compressed_colortex_mask(&rctx->samplers[PIPE_SHADER_COMPUTE].views); 1569 } else { 1570 for (i = 0; i < PIPE_SHADER_TYPES; ++i) { 1571 r600_update_compressed_colortex_mask(&rctx->samplers[i].views); 1572 } 1573 } 1574 if (!compute_only) 1575 r600_update_compressed_colortex_mask_images(&rctx->fragment_images); 1576 r600_update_compressed_colortex_mask_images(&rctx->compute_images); 1577 } 1578 1579 /* Decompress textures if needed. */ 1580 for (i = 0; i < PIPE_SHADER_TYPES; i++) { 1581 struct r600_samplerview_state *views = &rctx->samplers[i].views; 1582 1583 if (compute_only) 1584 if (i != PIPE_SHADER_COMPUTE) 1585 continue; 1586 if (views->compressed_depthtex_mask) { 1587 r600_decompress_depth_textures(rctx, views); 1588 } 1589 if (views->compressed_colortex_mask) { 1590 r600_decompress_color_textures(rctx, views); 1591 } 1592 } 1593 1594 { 1595 struct r600_image_state *istate; 1596 1597 if (!compute_only) { 1598 istate = &rctx->fragment_images; 1599 if (istate->compressed_depthtex_mask) 1600 r600_decompress_depth_images(rctx, istate); 1601 if (istate->compressed_colortex_mask) 1602 r600_decompress_color_images(rctx, istate); 1603 } 1604 1605 istate = &rctx->compute_images; 1606 if (istate->compressed_depthtex_mask) 1607 r600_decompress_depth_images(rctx, istate); 1608 if (istate->compressed_colortex_mask) 1609 r600_decompress_color_images(rctx, istate); 1610 } 1611 } 1612 1613 #define SELECT_SHADER_OR_FAIL(x) do { \ 1614 r600_shader_select(ctx, rctx->x##_shader, &x##_dirty); \ 1615 if (unlikely(!rctx->x##_shader->current)) \ 1616 return false; \ 1617 } while(0) 1618 1619 #define UPDATE_SHADER(hw, sw) do { \ 1620 if (sw##_dirty || (rctx->hw_shader_stages[(hw)].shader != rctx->sw##_shader->current)) \ 1621 update_shader_atom(ctx, &rctx->hw_shader_stages[(hw)], rctx->sw##_shader->current); \ 1622 } while(0) 1623 1624 #define UPDATE_SHADER_CLIP(hw, sw) do { \ 1625 if (sw##_dirty || (rctx->hw_shader_stages[(hw)].shader != rctx->sw##_shader->current)) { \ 1626 update_shader_atom(ctx, &rctx->hw_shader_stages[(hw)], rctx->sw##_shader->current); \ 1627 clip_so_current = rctx->sw##_shader->current; \ 1628 } \ 1629 } while(0) 1630 1631 #define UPDATE_SHADER_GS(hw, hw2, sw) do { \ 1632 if (sw##_dirty || (rctx->hw_shader_stages[(hw)].shader != rctx->sw##_shader->current)) { \ 1633 update_shader_atom(ctx, &rctx->hw_shader_stages[(hw)], rctx->sw##_shader->current); \ 1634 update_shader_atom(ctx, &rctx->hw_shader_stages[(hw2)], rctx->sw##_shader->current->gs_copy_shader); \ 1635 clip_so_current = rctx->sw##_shader->current->gs_copy_shader; \ 1636 } \ 1637 } while(0) 1638 1639 #define SET_NULL_SHADER(hw) do { \ 1640 if (rctx->hw_shader_stages[(hw)].shader) \ 1641 update_shader_atom(ctx, &rctx->hw_shader_stages[(hw)], NULL); \ 1642 } while (0) 1643 1644 static bool r600_update_derived_state(struct r600_context *rctx) 1645 { 1646 struct pipe_context * ctx = (struct pipe_context*)rctx; 1647 bool ps_dirty = false, vs_dirty = false, gs_dirty = false; 1648 bool tcs_dirty = false, tes_dirty = false, fixed_func_tcs_dirty = false; 1649 bool blend_disable; 1650 bool need_buf_const; 1651 struct r600_pipe_shader *clip_so_current = NULL; 1652 1653 if (!rctx->blitter->running) 1654 r600_update_compressed_resource_state(rctx, false); 1655 1656 SELECT_SHADER_OR_FAIL(ps); 1657 1658 r600_mark_atom_dirty(rctx, &rctx->shader_stages.atom); 1659 1660 update_gs_block_state(rctx, rctx->gs_shader != NULL); 1661 1662 if (rctx->gs_shader) 1663 SELECT_SHADER_OR_FAIL(gs); 1664 1665 /* Hull Shader */ 1666 if (rctx->tcs_shader) { 1667 SELECT_SHADER_OR_FAIL(tcs); 1668 1669 UPDATE_SHADER(EG_HW_STAGE_HS, tcs); 1670 } else if (rctx->tes_shader) { 1671 if (!rctx->fixed_func_tcs_shader) { 1672 r600_generate_fixed_func_tcs(rctx); 1673 if (!rctx->fixed_func_tcs_shader) 1674 return false; 1675 1676 } 1677 SELECT_SHADER_OR_FAIL(fixed_func_tcs); 1678 1679 UPDATE_SHADER(EG_HW_STAGE_HS, fixed_func_tcs); 1680 } else 1681 SET_NULL_SHADER(EG_HW_STAGE_HS); 1682 1683 if (rctx->tes_shader) { 1684 SELECT_SHADER_OR_FAIL(tes); 1685 } 1686 1687 SELECT_SHADER_OR_FAIL(vs); 1688 1689 if (rctx->gs_shader) { 1690 if (!rctx->shader_stages.geom_enable) { 1691 rctx->shader_stages.geom_enable = true; 1692 r600_mark_atom_dirty(rctx, &rctx->shader_stages.atom); 1693 } 1694 1695 /* gs_shader provides GS and VS (copy shader) */ 1696 UPDATE_SHADER_GS(R600_HW_STAGE_GS, R600_HW_STAGE_VS, gs); 1697 1698 /* vs_shader is used as ES */ 1699 1700 if (rctx->tes_shader) { 1701 /* VS goes to LS, TES goes to ES */ 1702 UPDATE_SHADER(R600_HW_STAGE_ES, tes); 1703 UPDATE_SHADER(EG_HW_STAGE_LS, vs); 1704 } else { 1705 /* vs_shader is used as ES */ 1706 UPDATE_SHADER(R600_HW_STAGE_ES, vs); 1707 SET_NULL_SHADER(EG_HW_STAGE_LS); 1708 } 1709 } else { 1710 if (unlikely(rctx->hw_shader_stages[R600_HW_STAGE_GS].shader)) { 1711 SET_NULL_SHADER(R600_HW_STAGE_GS); 1712 SET_NULL_SHADER(R600_HW_STAGE_ES); 1713 rctx->shader_stages.geom_enable = false; 1714 r600_mark_atom_dirty(rctx, &rctx->shader_stages.atom); 1715 } 1716 1717 if (rctx->tes_shader) { 1718 /* if TES is loaded and no geometry, TES runs on hw VS, VS runs on hw LS */ 1719 UPDATE_SHADER_CLIP(R600_HW_STAGE_VS, tes); 1720 UPDATE_SHADER(EG_HW_STAGE_LS, vs); 1721 } else { 1722 SET_NULL_SHADER(EG_HW_STAGE_LS); 1723 UPDATE_SHADER_CLIP(R600_HW_STAGE_VS, vs); 1724 } 1725 } 1726 1727 /* 1728 * XXX: I believe there's some fatal flaw in the dirty state logic when 1729 * enabling/disabling tes. 1730 * VS/ES share all buffer/resource/sampler slots. If TES is enabled, 1731 * it will therefore overwrite the VS slots. If it now gets disabled, 1732 * the VS needs to rebind all buffer/resource/sampler slots - not only 1733 * has TES overwritten the corresponding slots, but when the VS was 1734 * operating as LS the things with correpsonding dirty bits got bound 1735 * to LS slots and won't reflect what is dirty as VS stage even if the 1736 * TES didn't overwrite it. The story for re-enabled TES is similar. 1737 * In any case, we're not allowed to submit any TES state when 1738 * TES is disabled (the state tracker may not do this but this looks 1739 * like an optimization to me, not something which can be relied on). 1740 */ 1741 1742 /* Update clip misc state. */ 1743 if (clip_so_current) { 1744 r600_update_clip_state(rctx, clip_so_current); 1745 rctx->b.streamout.enabled_stream_buffers_mask = clip_so_current->enabled_stream_buffers_mask; 1746 } 1747 1748 if (unlikely(ps_dirty || rctx->hw_shader_stages[R600_HW_STAGE_PS].shader != rctx->ps_shader->current || 1749 rctx->rasterizer->sprite_coord_enable != rctx->ps_shader->current->sprite_coord_enable || 1750 rctx->rasterizer->flatshade != rctx->ps_shader->current->flatshade)) { 1751 1752 if (rctx->cb_misc_state.nr_ps_color_outputs != rctx->ps_shader->current->nr_ps_color_outputs) { 1753 rctx->cb_misc_state.nr_ps_color_outputs = rctx->ps_shader->current->nr_ps_color_outputs; 1754 r600_mark_atom_dirty(rctx, &rctx->cb_misc_state.atom); 1755 } 1756 1757 if (rctx->b.chip_class <= R700) { 1758 bool multiwrite = rctx->ps_shader->current->shader.fs_write_all; 1759 1760 if (rctx->cb_misc_state.multiwrite != multiwrite) { 1761 rctx->cb_misc_state.multiwrite = multiwrite; 1762 r600_mark_atom_dirty(rctx, &rctx->cb_misc_state.atom); 1763 } 1764 } 1765 1766 if (unlikely(!ps_dirty && rctx->ps_shader && rctx->rasterizer && 1767 ((rctx->rasterizer->sprite_coord_enable != rctx->ps_shader->current->sprite_coord_enable) || 1768 (rctx->rasterizer->flatshade != rctx->ps_shader->current->flatshade)))) { 1769 1770 if (rctx->b.chip_class >= EVERGREEN) 1771 evergreen_update_ps_state(ctx, rctx->ps_shader->current); 1772 else 1773 r600_update_ps_state(ctx, rctx->ps_shader->current); 1774 } 1775 1776 r600_mark_atom_dirty(rctx, &rctx->shader_stages.atom); 1777 } 1778 UPDATE_SHADER(R600_HW_STAGE_PS, ps); 1779 1780 if (rctx->b.chip_class >= EVERGREEN) { 1781 evergreen_update_db_shader_control(rctx); 1782 } else { 1783 r600_update_db_shader_control(rctx); 1784 } 1785 1786 /* on R600 we stuff masks + txq info into one constant buffer */ 1787 /* on evergreen we only need a txq info one */ 1788 if (rctx->ps_shader) { 1789 need_buf_const = rctx->ps_shader->current->shader.uses_tex_buffers || rctx->ps_shader->current->shader.has_txq_cube_array_z_comp; 1790 if (need_buf_const) { 1791 if (rctx->b.chip_class < EVERGREEN) 1792 r600_setup_buffer_constants(rctx, PIPE_SHADER_FRAGMENT); 1793 else 1794 eg_setup_buffer_constants(rctx, PIPE_SHADER_FRAGMENT); 1795 } 1796 } 1797 1798 if (rctx->vs_shader) { 1799 need_buf_const = rctx->vs_shader->current->shader.uses_tex_buffers || rctx->vs_shader->current->shader.has_txq_cube_array_z_comp; 1800 if (need_buf_const) { 1801 if (rctx->b.chip_class < EVERGREEN) 1802 r600_setup_buffer_constants(rctx, PIPE_SHADER_VERTEX); 1803 else 1804 eg_setup_buffer_constants(rctx, PIPE_SHADER_VERTEX); 1805 } 1806 } 1807 1808 if (rctx->gs_shader) { 1809 need_buf_const = rctx->gs_shader->current->shader.uses_tex_buffers || rctx->gs_shader->current->shader.has_txq_cube_array_z_comp; 1810 if (need_buf_const) { 1811 if (rctx->b.chip_class < EVERGREEN) 1812 r600_setup_buffer_constants(rctx, PIPE_SHADER_GEOMETRY); 1813 else 1814 eg_setup_buffer_constants(rctx, PIPE_SHADER_GEOMETRY); 1815 } 1816 } 1817 1818 if (rctx->tes_shader) { 1819 assert(rctx->b.chip_class >= EVERGREEN); 1820 need_buf_const = rctx->tes_shader->current->shader.uses_tex_buffers || 1821 rctx->tes_shader->current->shader.has_txq_cube_array_z_comp; 1822 if (need_buf_const) { 1823 eg_setup_buffer_constants(rctx, PIPE_SHADER_TESS_EVAL); 1824 } 1825 if (rctx->tcs_shader) { 1826 need_buf_const = rctx->tcs_shader->current->shader.uses_tex_buffers || 1827 rctx->tcs_shader->current->shader.has_txq_cube_array_z_comp; 1828 if (need_buf_const) { 1829 eg_setup_buffer_constants(rctx, PIPE_SHADER_TESS_CTRL); 1830 } 1831 } 1832 } 1833 1834 r600_update_driver_const_buffers(rctx, false); 1835 1836 if (rctx->b.chip_class < EVERGREEN && rctx->ps_shader && rctx->vs_shader) { 1837 if (!r600_adjust_gprs(rctx)) { 1838 /* discard rendering */ 1839 return false; 1840 } 1841 } 1842 1843 if (rctx->b.chip_class == EVERGREEN) { 1844 if (!evergreen_adjust_gprs(rctx)) { 1845 /* discard rendering */ 1846 return false; 1847 } 1848 } 1849 1850 blend_disable = (rctx->dual_src_blend && 1851 rctx->ps_shader->current->nr_ps_color_outputs < 2); 1852 1853 if (blend_disable != rctx->force_blend_disable) { 1854 rctx->force_blend_disable = blend_disable; 1855 r600_bind_blend_state_internal(rctx, 1856 rctx->blend_state.cso, 1857 blend_disable); 1858 } 1859 1860 return true; 1861 } 1862 1863 void r600_emit_clip_misc_state(struct r600_context *rctx, struct r600_atom *atom) 1864 { 1865 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 1866 struct r600_clip_misc_state *state = &rctx->clip_misc_state; 1867 1868 radeon_set_context_reg(cs, R_028810_PA_CL_CLIP_CNTL, 1869 state->pa_cl_clip_cntl | 1870 (state->clip_dist_write ? 0 : state->clip_plane_enable & 0x3F) | 1871 S_028810_CLIP_DISABLE(state->clip_disable)); 1872 radeon_set_context_reg(cs, R_02881C_PA_CL_VS_OUT_CNTL, 1873 state->pa_cl_vs_out_cntl | 1874 (state->clip_plane_enable & state->clip_dist_write) | 1875 (state->cull_dist_write << 8)); 1876 /* reuse needs to be set off if we write oViewport */ 1877 if (rctx->b.chip_class >= EVERGREEN) 1878 radeon_set_context_reg(cs, R_028AB4_VGT_REUSE_OFF, 1879 S_028AB4_REUSE_OFF(state->vs_out_viewport)); 1880 } 1881 1882 /* rast_prim is the primitive type after GS. */ 1883 static inline void r600_emit_rasterizer_prim_state(struct r600_context *rctx) 1884 { 1885 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 1886 enum pipe_prim_type rast_prim = rctx->current_rast_prim; 1887 1888 /* Skip this if not rendering lines. */ 1889 if (rast_prim != PIPE_PRIM_LINES && 1890 rast_prim != PIPE_PRIM_LINE_LOOP && 1891 rast_prim != PIPE_PRIM_LINE_STRIP && 1892 rast_prim != PIPE_PRIM_LINES_ADJACENCY && 1893 rast_prim != PIPE_PRIM_LINE_STRIP_ADJACENCY) 1894 return; 1895 1896 if (rast_prim == rctx->last_rast_prim) 1897 return; 1898 1899 /* For lines, reset the stipple pattern at each primitive. Otherwise, 1900 * reset the stipple pattern at each packet (line strips, line loops). 1901 */ 1902 radeon_set_context_reg(cs, R_028A0C_PA_SC_LINE_STIPPLE, 1903 S_028A0C_AUTO_RESET_CNTL(rast_prim == PIPE_PRIM_LINES ? 1 : 2) | 1904 (rctx->rasterizer ? rctx->rasterizer->pa_sc_line_stipple : 0)); 1905 rctx->last_rast_prim = rast_prim; 1906 } 1907 1908 static void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info) 1909 { 1910 struct r600_context *rctx = (struct r600_context *)ctx; 1911 struct pipe_resource *indexbuf = info->has_user_indices ? NULL : info->index.resource; 1912 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 1913 bool render_cond_bit = rctx->b.render_cond && !rctx->b.render_cond_force_off; 1914 bool has_user_indices = info->has_user_indices; 1915 uint64_t mask; 1916 unsigned num_patches, dirty_tex_counter, index_offset = 0; 1917 unsigned index_size = info->index_size; 1918 int index_bias; 1919 struct r600_shader_atomic combined_atomics[8]; 1920 uint8_t atomic_used_mask; 1921 1922 if (!info->indirect && !info->count && (index_size || !info->count_from_stream_output)) { 1923 return; 1924 } 1925 1926 if (unlikely(!rctx->vs_shader)) { 1927 assert(0); 1928 return; 1929 } 1930 if (unlikely(!rctx->ps_shader && 1931 (!rctx->rasterizer || !rctx->rasterizer->rasterizer_discard))) { 1932 assert(0); 1933 return; 1934 } 1935 1936 /* make sure that the gfx ring is only one active */ 1937 if (radeon_emitted(rctx->b.dma.cs, 0)) { 1938 rctx->b.dma.flush(rctx, PIPE_FLUSH_ASYNC, NULL); 1939 } 1940 1941 if (rctx->cmd_buf_is_compute) { 1942 rctx->b.gfx.flush(rctx, PIPE_FLUSH_ASYNC, NULL); 1943 rctx->cmd_buf_is_compute = false; 1944 } 1945 1946 /* Re-emit the framebuffer state if needed. */ 1947 dirty_tex_counter = p_atomic_read(&rctx->b.screen->dirty_tex_counter); 1948 if (unlikely(dirty_tex_counter != rctx->b.last_dirty_tex_counter)) { 1949 rctx->b.last_dirty_tex_counter = dirty_tex_counter; 1950 r600_mark_atom_dirty(rctx, &rctx->framebuffer.atom); 1951 rctx->framebuffer.do_update_surf_dirtiness = true; 1952 } 1953 1954 if (rctx->gs_shader) { 1955 /* Determine whether the GS triangle strip adjacency fix should 1956 * be applied. Rotate every other triangle if 1957 * - triangle strips with adjacency are fed to the GS and 1958 * - primitive restart is disabled (the rotation doesn't help 1959 * when the restart occurs after an odd number of triangles). 1960 */ 1961 bool gs_tri_strip_adj_fix = 1962 !rctx->tes_shader && 1963 info->mode == PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY && 1964 !info->primitive_restart; 1965 if (gs_tri_strip_adj_fix != rctx->gs_tri_strip_adj_fix) 1966 rctx->gs_tri_strip_adj_fix = gs_tri_strip_adj_fix; 1967 } 1968 if (!r600_update_derived_state(rctx)) { 1969 /* useless to render because current rendering command 1970 * can't be achieved 1971 */ 1972 return; 1973 } 1974 1975 rctx->current_rast_prim = (rctx->gs_shader)? rctx->gs_shader->gs_output_prim 1976 : (rctx->tes_shader)? rctx->tes_shader->info.properties[TGSI_PROPERTY_TES_PRIM_MODE] 1977 : info->mode; 1978 1979 if (rctx->b.chip_class >= EVERGREEN) 1980 evergreen_emit_atomic_buffer_setup(rctx, NULL, combined_atomics, &atomic_used_mask); 1981 1982 if (index_size) { 1983 index_offset += info->start * index_size; 1984 1985 /* Translate 8-bit indices to 16-bit. */ 1986 if (unlikely(index_size == 1)) { 1987 struct pipe_resource *out_buffer = NULL; 1988 unsigned out_offset; 1989 void *ptr; 1990 unsigned start, count; 1991 1992 if (likely(!info->indirect)) { 1993 start = 0; 1994 count = info->count; 1995 } 1996 else { 1997 /* Have to get start/count from indirect buffer, slow path ahead... */ 1998 struct r600_resource *indirect_resource = (struct r600_resource *)info->indirect->buffer; 1999 unsigned *data = r600_buffer_map_sync_with_rings(&rctx->b, indirect_resource, 2000 PIPE_TRANSFER_READ); 2001 if (data) { 2002 data += info->indirect->offset / sizeof(unsigned); 2003 start = data[2] * index_size; 2004 count = data[0]; 2005 } 2006 else { 2007 start = 0; 2008 count = 0; 2009 } 2010 } 2011 2012 u_upload_alloc(ctx->stream_uploader, start, count * 2, 2013 256, &out_offset, &out_buffer, &ptr); 2014 if (unlikely(!ptr)) 2015 return; 2016 2017 util_shorten_ubyte_elts_to_userptr( 2018 &rctx->b.b, info, 0, 0, index_offset, count, ptr); 2019 2020 indexbuf = out_buffer; 2021 index_offset = out_offset; 2022 index_size = 2; 2023 has_user_indices = false; 2024 } 2025 2026 /* Upload the index buffer. 2027 * The upload is skipped for small index counts on little-endian machines 2028 * and the indices are emitted via PKT3_DRAW_INDEX_IMMD. 2029 * Indirect draws never use immediate indices. 2030 * Note: Instanced rendering in combination with immediate indices hangs. */ 2031 if (has_user_indices && (R600_BIG_ENDIAN || info->indirect || 2032 info->instance_count > 1 || 2033 info->count*index_size > 20)) { 2034 indexbuf = NULL; 2035 u_upload_data(ctx->stream_uploader, 0, 2036 info->count * index_size, 256, 2037 info->index.user, &index_offset, &indexbuf); 2038 has_user_indices = false; 2039 } 2040 index_bias = info->index_bias; 2041 } else { 2042 index_bias = info->start; 2043 } 2044 2045 /* Set the index offset and primitive restart. */ 2046 if (rctx->vgt_state.vgt_multi_prim_ib_reset_en != info->primitive_restart || 2047 rctx->vgt_state.vgt_multi_prim_ib_reset_indx != info->restart_index || 2048 rctx->vgt_state.vgt_indx_offset != index_bias || 2049 (rctx->vgt_state.last_draw_was_indirect && !info->indirect)) { 2050 rctx->vgt_state.vgt_multi_prim_ib_reset_en = info->primitive_restart; 2051 rctx->vgt_state.vgt_multi_prim_ib_reset_indx = info->restart_index; 2052 rctx->vgt_state.vgt_indx_offset = index_bias; 2053 r600_mark_atom_dirty(rctx, &rctx->vgt_state.atom); 2054 } 2055 2056 /* Workaround for hardware deadlock on certain R600 ASICs: write into a CB register. */ 2057 if (rctx->b.chip_class == R600) { 2058 rctx->b.flags |= R600_CONTEXT_PS_PARTIAL_FLUSH; 2059 r600_mark_atom_dirty(rctx, &rctx->cb_misc_state.atom); 2060 } 2061 2062 if (rctx->b.chip_class >= EVERGREEN) 2063 evergreen_setup_tess_constants(rctx, info, &num_patches); 2064 2065 /* Emit states. */ 2066 r600_need_cs_space(rctx, has_user_indices ? 5 : 0, TRUE); 2067 r600_flush_emit(rctx); 2068 2069 mask = rctx->dirty_atoms; 2070 while (mask != 0) { 2071 r600_emit_atom(rctx, rctx->atoms[u_bit_scan64(&mask)]); 2072 } 2073 2074 if (rctx->b.chip_class == CAYMAN) { 2075 /* Copied from radeonsi. */ 2076 unsigned primgroup_size = 128; /* recommended without a GS */ 2077 bool ia_switch_on_eop = false; 2078 bool partial_vs_wave = false; 2079 2080 if (rctx->gs_shader) 2081 primgroup_size = 64; /* recommended with a GS */ 2082 2083 if ((rctx->rasterizer && rctx->rasterizer->pa_sc_line_stipple) || 2084 (rctx->b.screen->debug_flags & DBG_SWITCH_ON_EOP)) { 2085 ia_switch_on_eop = true; 2086 } 2087 2088 if (r600_get_strmout_en(&rctx->b)) 2089 partial_vs_wave = true; 2090 2091 radeon_set_context_reg(cs, CM_R_028AA8_IA_MULTI_VGT_PARAM, 2092 S_028AA8_SWITCH_ON_EOP(ia_switch_on_eop) | 2093 S_028AA8_PARTIAL_VS_WAVE_ON(partial_vs_wave) | 2094 S_028AA8_PRIMGROUP_SIZE(primgroup_size - 1)); 2095 } 2096 2097 if (rctx->b.chip_class >= EVERGREEN) { 2098 uint32_t ls_hs_config = evergreen_get_ls_hs_config(rctx, info, 2099 num_patches); 2100 2101 evergreen_set_ls_hs_config(rctx, cs, ls_hs_config); 2102 evergreen_set_lds_alloc(rctx, cs, rctx->lds_alloc); 2103 } 2104 2105 /* On R6xx, CULL_FRONT=1 culls all points, lines, and rectangles, 2106 * even though it should have no effect on those. */ 2107 if (rctx->b.chip_class == R600 && rctx->rasterizer) { 2108 unsigned su_sc_mode_cntl = rctx->rasterizer->pa_su_sc_mode_cntl; 2109 unsigned prim = info->mode; 2110 2111 if (rctx->gs_shader) { 2112 prim = rctx->gs_shader->gs_output_prim; 2113 } 2114 prim = r600_conv_prim_to_gs_out(prim); /* decrease the number of types to 3 */ 2115 2116 if (prim == V_028A6C_OUTPRIM_TYPE_POINTLIST || 2117 prim == V_028A6C_OUTPRIM_TYPE_LINESTRIP || 2118 info->mode == R600_PRIM_RECTANGLE_LIST) { 2119 su_sc_mode_cntl &= C_028814_CULL_FRONT; 2120 } 2121 radeon_set_context_reg(cs, R_028814_PA_SU_SC_MODE_CNTL, su_sc_mode_cntl); 2122 } 2123 2124 /* Update start instance. */ 2125 if (!info->indirect && rctx->last_start_instance != info->start_instance) { 2126 radeon_set_ctl_const(cs, R_03CFF4_SQ_VTX_START_INST_LOC, info->start_instance); 2127 rctx->last_start_instance = info->start_instance; 2128 } 2129 2130 /* Update the primitive type. */ 2131 if (rctx->last_primitive_type != info->mode) { 2132 r600_emit_rasterizer_prim_state(rctx); 2133 radeon_set_config_reg(cs, R_008958_VGT_PRIMITIVE_TYPE, 2134 r600_conv_pipe_prim(info->mode)); 2135 2136 rctx->last_primitive_type = info->mode; 2137 } 2138 2139 /* Draw packets. */ 2140 if (likely(!info->indirect)) { 2141 radeon_emit(cs, PKT3(PKT3_NUM_INSTANCES, 0, 0)); 2142 radeon_emit(cs, info->instance_count); 2143 } else { 2144 uint64_t va = r600_resource(info->indirect->buffer)->gpu_address; 2145 assert(rctx->b.chip_class >= EVERGREEN); 2146 2147 // Invalidate so non-indirect draw calls reset this state 2148 rctx->vgt_state.last_draw_was_indirect = true; 2149 rctx->last_start_instance = -1; 2150 2151 radeon_emit(cs, PKT3(EG_PKT3_SET_BASE, 2, 0)); 2152 radeon_emit(cs, EG_DRAW_INDEX_INDIRECT_PATCH_TABLE_BASE); 2153 radeon_emit(cs, va); 2154 radeon_emit(cs, (va >> 32UL) & 0xFF); 2155 2156 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); 2157 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, 2158 (struct r600_resource*)info->indirect->buffer, 2159 RADEON_USAGE_READ, 2160 RADEON_PRIO_DRAW_INDIRECT)); 2161 } 2162 2163 if (index_size) { 2164 radeon_emit(cs, PKT3(PKT3_INDEX_TYPE, 0, 0)); 2165 radeon_emit(cs, index_size == 4 ? 2166 (VGT_INDEX_32 | (R600_BIG_ENDIAN ? VGT_DMA_SWAP_32_BIT : 0)) : 2167 (VGT_INDEX_16 | (R600_BIG_ENDIAN ? VGT_DMA_SWAP_16_BIT : 0))); 2168 2169 if (has_user_indices) { 2170 unsigned size_bytes = info->count*index_size; 2171 unsigned size_dw = align(size_bytes, 4) / 4; 2172 radeon_emit(cs, PKT3(PKT3_DRAW_INDEX_IMMD, 1 + size_dw, render_cond_bit)); 2173 radeon_emit(cs, info->count); 2174 radeon_emit(cs, V_0287F0_DI_SRC_SEL_IMMEDIATE); 2175 radeon_emit_array(cs, info->index.user, size_dw); 2176 } else { 2177 uint64_t va = r600_resource(indexbuf)->gpu_address + index_offset; 2178 2179 if (likely(!info->indirect)) { 2180 radeon_emit(cs, PKT3(PKT3_DRAW_INDEX, 3, render_cond_bit)); 2181 radeon_emit(cs, va); 2182 radeon_emit(cs, (va >> 32UL) & 0xFF); 2183 radeon_emit(cs, info->count); 2184 radeon_emit(cs, V_0287F0_DI_SRC_SEL_DMA); 2185 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); 2186 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, 2187 (struct r600_resource*)indexbuf, 2188 RADEON_USAGE_READ, 2189 RADEON_PRIO_INDEX_BUFFER)); 2190 } 2191 else { 2192 uint32_t max_size = (indexbuf->width0 - index_offset) / index_size; 2193 2194 radeon_emit(cs, PKT3(EG_PKT3_INDEX_BASE, 1, 0)); 2195 radeon_emit(cs, va); 2196 radeon_emit(cs, (va >> 32UL) & 0xFF); 2197 2198 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); 2199 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, 2200 (struct r600_resource*)indexbuf, 2201 RADEON_USAGE_READ, 2202 RADEON_PRIO_INDEX_BUFFER)); 2203 2204 radeon_emit(cs, PKT3(EG_PKT3_INDEX_BUFFER_SIZE, 0, 0)); 2205 radeon_emit(cs, max_size); 2206 2207 radeon_emit(cs, PKT3(EG_PKT3_DRAW_INDEX_INDIRECT, 1, render_cond_bit)); 2208 radeon_emit(cs, info->indirect->offset); 2209 radeon_emit(cs, V_0287F0_DI_SRC_SEL_DMA); 2210 } 2211 } 2212 } else { 2213 if (unlikely(info->count_from_stream_output)) { 2214 struct r600_so_target *t = (struct r600_so_target*)info->count_from_stream_output; 2215 uint64_t va = t->buf_filled_size->gpu_address + t->buf_filled_size_offset; 2216 2217 radeon_set_context_reg(cs, R_028B30_VGT_STRMOUT_DRAW_OPAQUE_VERTEX_STRIDE, t->stride_in_dw); 2218 2219 radeon_emit(cs, PKT3(PKT3_COPY_DW, 4, 0)); 2220 radeon_emit(cs, COPY_DW_SRC_IS_MEM | COPY_DW_DST_IS_REG); 2221 radeon_emit(cs, va & 0xFFFFFFFFUL); /* src address lo */ 2222 radeon_emit(cs, (va >> 32UL) & 0xFFUL); /* src address hi */ 2223 radeon_emit(cs, R_028B2C_VGT_STRMOUT_DRAW_OPAQUE_BUFFER_FILLED_SIZE >> 2); /* dst register */ 2224 radeon_emit(cs, 0); /* unused */ 2225 2226 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); 2227 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, 2228 t->buf_filled_size, RADEON_USAGE_READ, 2229 RADEON_PRIO_SO_FILLED_SIZE)); 2230 } 2231 2232 if (likely(!info->indirect)) { 2233 radeon_emit(cs, PKT3(PKT3_DRAW_INDEX_AUTO, 1, render_cond_bit)); 2234 radeon_emit(cs, info->count); 2235 } 2236 else { 2237 radeon_emit(cs, PKT3(EG_PKT3_DRAW_INDIRECT, 1, render_cond_bit)); 2238 radeon_emit(cs, info->indirect->offset); 2239 } 2240 radeon_emit(cs, V_0287F0_DI_SRC_SEL_AUTO_INDEX | 2241 (info->count_from_stream_output ? S_0287F0_USE_OPAQUE(1) : 0)); 2242 } 2243 2244 /* SMX returns CONTEXT_DONE too early workaround */ 2245 if (rctx->b.family == CHIP_R600 || 2246 rctx->b.family == CHIP_RV610 || 2247 rctx->b.family == CHIP_RV630 || 2248 rctx->b.family == CHIP_RV635) { 2249 /* if we have gs shader or streamout 2250 we need to do a wait idle after every draw */ 2251 if (rctx->gs_shader || r600_get_strmout_en(&rctx->b)) { 2252 radeon_set_config_reg(cs, R_008040_WAIT_UNTIL, S_008040_WAIT_3D_IDLE(1)); 2253 } 2254 } 2255 2256 /* ES ring rolling over at EOP - workaround */ 2257 if (rctx->b.chip_class == R600) { 2258 radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0)); 2259 radeon_emit(cs, EVENT_TYPE(EVENT_TYPE_SQ_NON_EVENT)); 2260 } 2261 2262 2263 if (rctx->b.chip_class >= EVERGREEN) 2264 evergreen_emit_atomic_buffer_save(rctx, false, combined_atomics, &atomic_used_mask); 2265 2266 if (rctx->trace_buf) 2267 eg_trace_emit(rctx); 2268 2269 if (rctx->framebuffer.do_update_surf_dirtiness) { 2270 /* Set the depth buffer as dirty. */ 2271 if (rctx->framebuffer.state.zsbuf) { 2272 struct pipe_surface *surf = rctx->framebuffer.state.zsbuf; 2273 struct r600_texture *rtex = (struct r600_texture *)surf->texture; 2274 2275 rtex->dirty_level_mask |= 1 << surf->u.tex.level; 2276 2277 if (rtex->surface.has_stencil) 2278 rtex->stencil_dirty_level_mask |= 1 << surf->u.tex.level; 2279 } 2280 if (rctx->framebuffer.compressed_cb_mask) { 2281 struct pipe_surface *surf; 2282 struct r600_texture *rtex; 2283 unsigned mask = rctx->framebuffer.compressed_cb_mask; 2284 2285 do { 2286 unsigned i = u_bit_scan(&mask); 2287 surf = rctx->framebuffer.state.cbufs[i]; 2288 rtex = (struct r600_texture*)surf->texture; 2289 2290 rtex->dirty_level_mask |= 1 << surf->u.tex.level; 2291 2292 } while (mask); 2293 } 2294 rctx->framebuffer.do_update_surf_dirtiness = false; 2295 } 2296 2297 if (index_size && indexbuf != info->index.resource) 2298 pipe_resource_reference(&indexbuf, NULL); 2299 rctx->b.num_draw_calls++; 2300 } 2301 2302 uint32_t r600_translate_stencil_op(int s_op) 2303 { 2304 switch (s_op) { 2305 case PIPE_STENCIL_OP_KEEP: 2306 return V_028800_STENCIL_KEEP; 2307 case PIPE_STENCIL_OP_ZERO: 2308 return V_028800_STENCIL_ZERO; 2309 case PIPE_STENCIL_OP_REPLACE: 2310 return V_028800_STENCIL_REPLACE; 2311 case PIPE_STENCIL_OP_INCR: 2312 return V_028800_STENCIL_INCR; 2313 case PIPE_STENCIL_OP_DECR: 2314 return V_028800_STENCIL_DECR; 2315 case PIPE_STENCIL_OP_INCR_WRAP: 2316 return V_028800_STENCIL_INCR_WRAP; 2317 case PIPE_STENCIL_OP_DECR_WRAP: 2318 return V_028800_STENCIL_DECR_WRAP; 2319 case PIPE_STENCIL_OP_INVERT: 2320 return V_028800_STENCIL_INVERT; 2321 default: 2322 R600_ERR("Unknown stencil op %d", s_op); 2323 assert(0); 2324 break; 2325 } 2326 return 0; 2327 } 2328 2329 uint32_t r600_translate_fill(uint32_t func) 2330 { 2331 switch(func) { 2332 case PIPE_POLYGON_MODE_FILL: 2333 return 2; 2334 case PIPE_POLYGON_MODE_LINE: 2335 return 1; 2336 case PIPE_POLYGON_MODE_POINT: 2337 return 0; 2338 default: 2339 assert(0); 2340 return 0; 2341 } 2342 } 2343 2344 unsigned r600_tex_wrap(unsigned wrap) 2345 { 2346 switch (wrap) { 2347 default: 2348 case PIPE_TEX_WRAP_REPEAT: 2349 return V_03C000_SQ_TEX_WRAP; 2350 case PIPE_TEX_WRAP_CLAMP: 2351 return V_03C000_SQ_TEX_CLAMP_HALF_BORDER; 2352 case PIPE_TEX_WRAP_CLAMP_TO_EDGE: 2353 return V_03C000_SQ_TEX_CLAMP_LAST_TEXEL; 2354 case PIPE_TEX_WRAP_CLAMP_TO_BORDER: 2355 return V_03C000_SQ_TEX_CLAMP_BORDER; 2356 case PIPE_TEX_WRAP_MIRROR_REPEAT: 2357 return V_03C000_SQ_TEX_MIRROR; 2358 case PIPE_TEX_WRAP_MIRROR_CLAMP: 2359 return V_03C000_SQ_TEX_MIRROR_ONCE_HALF_BORDER; 2360 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE: 2361 return V_03C000_SQ_TEX_MIRROR_ONCE_LAST_TEXEL; 2362 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER: 2363 return V_03C000_SQ_TEX_MIRROR_ONCE_BORDER; 2364 } 2365 } 2366 2367 unsigned r600_tex_mipfilter(unsigned filter) 2368 { 2369 switch (filter) { 2370 case PIPE_TEX_MIPFILTER_NEAREST: 2371 return V_03C000_SQ_TEX_Z_FILTER_POINT; 2372 case PIPE_TEX_MIPFILTER_LINEAR: 2373 return V_03C000_SQ_TEX_Z_FILTER_LINEAR; 2374 default: 2375 case PIPE_TEX_MIPFILTER_NONE: 2376 return V_03C000_SQ_TEX_Z_FILTER_NONE; 2377 } 2378 } 2379 2380 unsigned r600_tex_compare(unsigned compare) 2381 { 2382 switch (compare) { 2383 default: 2384 case PIPE_FUNC_NEVER: 2385 return V_03C000_SQ_TEX_DEPTH_COMPARE_NEVER; 2386 case PIPE_FUNC_LESS: 2387 return V_03C000_SQ_TEX_DEPTH_COMPARE_LESS; 2388 case PIPE_FUNC_EQUAL: 2389 return V_03C000_SQ_TEX_DEPTH_COMPARE_EQUAL; 2390 case PIPE_FUNC_LEQUAL: 2391 return V_03C000_SQ_TEX_DEPTH_COMPARE_LESSEQUAL; 2392 case PIPE_FUNC_GREATER: 2393 return V_03C000_SQ_TEX_DEPTH_COMPARE_GREATER; 2394 case PIPE_FUNC_NOTEQUAL: 2395 return V_03C000_SQ_TEX_DEPTH_COMPARE_NOTEQUAL; 2396 case PIPE_FUNC_GEQUAL: 2397 return V_03C000_SQ_TEX_DEPTH_COMPARE_GREATEREQUAL; 2398 case PIPE_FUNC_ALWAYS: 2399 return V_03C000_SQ_TEX_DEPTH_COMPARE_ALWAYS; 2400 } 2401 } 2402 2403 static bool wrap_mode_uses_border_color(unsigned wrap, bool linear_filter) 2404 { 2405 return wrap == PIPE_TEX_WRAP_CLAMP_TO_BORDER || 2406 wrap == PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER || 2407 (linear_filter && 2408 (wrap == PIPE_TEX_WRAP_CLAMP || 2409 wrap == PIPE_TEX_WRAP_MIRROR_CLAMP)); 2410 } 2411 2412 bool sampler_state_needs_border_color(const struct pipe_sampler_state *state) 2413 { 2414 bool linear_filter = state->min_img_filter != PIPE_TEX_FILTER_NEAREST || 2415 state->mag_img_filter != PIPE_TEX_FILTER_NEAREST; 2416 2417 return (state->border_color.ui[0] || state->border_color.ui[1] || 2418 state->border_color.ui[2] || state->border_color.ui[3]) && 2419 (wrap_mode_uses_border_color(state->wrap_s, linear_filter) || 2420 wrap_mode_uses_border_color(state->wrap_t, linear_filter) || 2421 wrap_mode_uses_border_color(state->wrap_r, linear_filter)); 2422 } 2423 2424 void r600_emit_shader(struct r600_context *rctx, struct r600_atom *a) 2425 { 2426 2427 struct radeon_winsys_cs *cs = rctx->b.gfx.cs; 2428 struct r600_pipe_shader *shader = ((struct r600_shader_state*)a)->shader; 2429 2430 if (!shader) 2431 return; 2432 2433 r600_emit_command_buffer(cs, &shader->command_buffer); 2434 radeon_emit(cs, PKT3(PKT3_NOP, 0, 0)); 2435 radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx, shader->bo, 2436 RADEON_USAGE_READ, RADEON_PRIO_SHADER_BINARY)); 2437 } 2438 2439 unsigned r600_get_swizzle_combined(const unsigned char *swizzle_format, 2440 const unsigned char *swizzle_view, 2441 boolean vtx) 2442 { 2443 unsigned i; 2444 unsigned char swizzle[4]; 2445 unsigned result = 0; 2446 const uint32_t tex_swizzle_shift[4] = { 2447 16, 19, 22, 25, 2448 }; 2449 const uint32_t vtx_swizzle_shift[4] = { 2450 3, 6, 9, 12, 2451 }; 2452 const uint32_t swizzle_bit[4] = { 2453 0, 1, 2, 3, 2454 }; 2455 const uint32_t *swizzle_shift = tex_swizzle_shift; 2456 2457 if (vtx) 2458 swizzle_shift = vtx_swizzle_shift; 2459 2460 if (swizzle_view) { 2461 util_format_compose_swizzles(swizzle_format, swizzle_view, swizzle); 2462 } else { 2463 memcpy(swizzle, swizzle_format, 4); 2464 } 2465 2466 /* Get swizzle. */ 2467 for (i = 0; i < 4; i++) { 2468 switch (swizzle[i]) { 2469 case PIPE_SWIZZLE_Y: 2470 result |= swizzle_bit[1] << swizzle_shift[i]; 2471 break; 2472 case PIPE_SWIZZLE_Z: 2473 result |= swizzle_bit[2] << swizzle_shift[i]; 2474 break; 2475 case PIPE_SWIZZLE_W: 2476 result |= swizzle_bit[3] << swizzle_shift[i]; 2477 break; 2478 case PIPE_SWIZZLE_0: 2479 result |= V_038010_SQ_SEL_0 << swizzle_shift[i]; 2480 break; 2481 case PIPE_SWIZZLE_1: 2482 result |= V_038010_SQ_SEL_1 << swizzle_shift[i]; 2483 break; 2484 default: /* PIPE_SWIZZLE_X */ 2485 result |= swizzle_bit[0] << swizzle_shift[i]; 2486 } 2487 } 2488 return result; 2489 } 2490 2491 /* texture format translate */ 2492 uint32_t r600_translate_texformat(struct pipe_screen *screen, 2493 enum pipe_format format, 2494 const unsigned char *swizzle_view, 2495 uint32_t *word4_p, uint32_t *yuv_format_p, 2496 bool do_endian_swap) 2497 { 2498 struct r600_screen *rscreen = (struct r600_screen *)screen; 2499 uint32_t result = 0, word4 = 0, yuv_format = 0; 2500 const struct util_format_description *desc; 2501 boolean uniform = TRUE; 2502 bool is_srgb_valid = FALSE; 2503 const unsigned char swizzle_xxxx[4] = {0, 0, 0, 0}; 2504 const unsigned char swizzle_yyyy[4] = {1, 1, 1, 1}; 2505 const unsigned char swizzle_xxxy[4] = {0, 0, 0, 1}; 2506 const unsigned char swizzle_zyx1[4] = {2, 1, 0, 5}; 2507 const unsigned char swizzle_zyxw[4] = {2, 1, 0, 3}; 2508 2509 int i; 2510 const uint32_t sign_bit[4] = { 2511 S_038010_FORMAT_COMP_X(V_038010_SQ_FORMAT_COMP_SIGNED), 2512 S_038010_FORMAT_COMP_Y(V_038010_SQ_FORMAT_COMP_SIGNED), 2513 S_038010_FORMAT_COMP_Z(V_038010_SQ_FORMAT_COMP_SIGNED), 2514 S_038010_FORMAT_COMP_W(V_038010_SQ_FORMAT_COMP_SIGNED) 2515 }; 2516 2517 /* Need to replace the specified texture formats in case of big-endian. 2518 * These formats are formats that have channels with number of bits 2519 * not divisible by 8. 2520 * Mesa conversion functions don't swap bits for those formats, and because 2521 * we transmit this over a serial bus to the GPU (PCIe), the 2522 * bit-endianess is important!!! 2523 * In case we have an "opposite" format, just use that for the swizzling 2524 * information. If we don't have such an "opposite" format, we need 2525 * to use a fixed swizzle info instead (see below) 2526 */ 2527 if (format == PIPE_FORMAT_R4A4_UNORM && do_endian_swap) 2528 format = PIPE_FORMAT_A4R4_UNORM; 2529 2530 desc = util_format_description(format); 2531 if (!desc) 2532 goto out_unknown; 2533 2534 /* Depth and stencil swizzling is handled separately. */ 2535 if (desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS) { 2536 /* Need to check for specific texture formats that don't have 2537 * an "opposite" format we can use. For those formats, we directly 2538 * specify the swizzling, which is the LE swizzling as defined in 2539 * u_format.csv 2540 */ 2541 if (do_endian_swap) { 2542 if (format == PIPE_FORMAT_L4A4_UNORM) 2543 word4 |= r600_get_swizzle_combined(swizzle_xxxy, swizzle_view, FALSE); 2544 else if (format == PIPE_FORMAT_B4G4R4A4_UNORM) 2545 word4 |= r600_get_swizzle_combined(swizzle_zyxw, swizzle_view, FALSE); 2546 else if (format == PIPE_FORMAT_B4G4R4X4_UNORM || format == PIPE_FORMAT_B5G6R5_UNORM) 2547 word4 |= r600_get_swizzle_combined(swizzle_zyx1, swizzle_view, FALSE); 2548 else 2549 word4 |= r600_get_swizzle_combined(desc->swizzle, swizzle_view, FALSE); 2550 } else { 2551 word4 |= r600_get_swizzle_combined(desc->swizzle, swizzle_view, FALSE); 2552 } 2553 } 2554 2555 /* Colorspace (return non-RGB formats directly). */ 2556 switch (desc->colorspace) { 2557 /* Depth stencil formats */ 2558 case UTIL_FORMAT_COLORSPACE_ZS: 2559 switch (format) { 2560 /* Depth sampler formats. */ 2561 case PIPE_FORMAT_Z16_UNORM: 2562 word4 |= r600_get_swizzle_combined(swizzle_xxxx, swizzle_view, FALSE); 2563 result = FMT_16; 2564 goto out_word4; 2565 case PIPE_FORMAT_Z24X8_UNORM: 2566 case PIPE_FORMAT_Z24_UNORM_S8_UINT: 2567 word4 |= r600_get_swizzle_combined(swizzle_xxxx, swizzle_view, FALSE); 2568 result = FMT_8_24; 2569 goto out_word4; 2570 case PIPE_FORMAT_X8Z24_UNORM: 2571 case PIPE_FORMAT_S8_UINT_Z24_UNORM: 2572 if (rscreen->b.chip_class < EVERGREEN) 2573 goto out_unknown; 2574 word4 |= r600_get_swizzle_combined(swizzle_yyyy, swizzle_view, FALSE); 2575 result = FMT_24_8; 2576 goto out_word4; 2577 case PIPE_FORMAT_Z32_FLOAT: 2578 word4 |= r600_get_swizzle_combined(swizzle_xxxx, swizzle_view, FALSE); 2579 result = FMT_32_FLOAT; 2580 goto out_word4; 2581 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT: 2582 word4 |= r600_get_swizzle_combined(swizzle_xxxx, swizzle_view, FALSE); 2583 result = FMT_X24_8_32_FLOAT; 2584 goto out_word4; 2585 /* Stencil sampler formats. */ 2586 case PIPE_FORMAT_S8_UINT: 2587 word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT); 2588 word4 |= r600_get_swizzle_combined(swizzle_xxxx, swizzle_view, FALSE); 2589 result = FMT_8; 2590 goto out_word4; 2591 case PIPE_FORMAT_X24S8_UINT: 2592 word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT); 2593 word4 |= r600_get_swizzle_combined(swizzle_yyyy, swizzle_view, FALSE); 2594 result = FMT_8_24; 2595 goto out_word4; 2596 case PIPE_FORMAT_S8X24_UINT: 2597 if (rscreen->b.chip_class < EVERGREEN) 2598 goto out_unknown; 2599 word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT); 2600 word4 |= r600_get_swizzle_combined(swizzle_xxxx, swizzle_view, FALSE); 2601 result = FMT_24_8; 2602 goto out_word4; 2603 case PIPE_FORMAT_X32_S8X24_UINT: 2604 word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT); 2605 word4 |= r600_get_swizzle_combined(swizzle_yyyy, swizzle_view, FALSE); 2606 result = FMT_X24_8_32_FLOAT; 2607 goto out_word4; 2608 default: 2609 goto out_unknown; 2610 } 2611 2612 case UTIL_FORMAT_COLORSPACE_YUV: 2613 yuv_format |= (1 << 30); 2614 switch (format) { 2615 case PIPE_FORMAT_UYVY: 2616 case PIPE_FORMAT_YUYV: 2617 default: 2618 break; 2619 } 2620 goto out_unknown; /* XXX */ 2621 2622 case UTIL_FORMAT_COLORSPACE_SRGB: 2623 word4 |= S_038010_FORCE_DEGAMMA(1); 2624 break; 2625 2626 default: 2627 break; 2628 } 2629 2630 if (desc->layout == UTIL_FORMAT_LAYOUT_RGTC) { 2631 switch (format) { 2632 case PIPE_FORMAT_RGTC1_SNORM: 2633 case PIPE_FORMAT_LATC1_SNORM: 2634 word4 |= sign_bit[0]; 2635 case PIPE_FORMAT_RGTC1_UNORM: 2636 case PIPE_FORMAT_LATC1_UNORM: 2637 result = FMT_BC4; 2638 goto out_word4; 2639 case PIPE_FORMAT_RGTC2_SNORM: 2640 case PIPE_FORMAT_LATC2_SNORM: 2641 word4 |= sign_bit[0] | sign_bit[1]; 2642 case PIPE_FORMAT_RGTC2_UNORM: 2643 case PIPE_FORMAT_LATC2_UNORM: 2644 result = FMT_BC5; 2645 goto out_word4; 2646 default: 2647 goto out_unknown; 2648 } 2649 } 2650 2651 if (desc->layout == UTIL_FORMAT_LAYOUT_S3TC) { 2652 switch (format) { 2653 case PIPE_FORMAT_DXT1_RGB: 2654 case PIPE_FORMAT_DXT1_RGBA: 2655 case PIPE_FORMAT_DXT1_SRGB: 2656 case PIPE_FORMAT_DXT1_SRGBA: 2657 result = FMT_BC1; 2658 is_srgb_valid = TRUE; 2659 goto out_word4; 2660 case PIPE_FORMAT_DXT3_RGBA: 2661 case PIPE_FORMAT_DXT3_SRGBA: 2662 result = FMT_BC2; 2663 is_srgb_valid = TRUE; 2664 goto out_word4; 2665 case PIPE_FORMAT_DXT5_RGBA: 2666 case PIPE_FORMAT_DXT5_SRGBA: 2667 result = FMT_BC3; 2668 is_srgb_valid = TRUE; 2669 goto out_word4; 2670 default: 2671 goto out_unknown; 2672 } 2673 } 2674 2675 if (desc->layout == UTIL_FORMAT_LAYOUT_BPTC) { 2676 if (rscreen->b.chip_class < EVERGREEN) 2677 goto out_unknown; 2678 2679 switch (format) { 2680 case PIPE_FORMAT_BPTC_RGBA_UNORM: 2681 case PIPE_FORMAT_BPTC_SRGBA: 2682 result = FMT_BC7; 2683 is_srgb_valid = TRUE; 2684 goto out_word4; 2685 case PIPE_FORMAT_BPTC_RGB_FLOAT: 2686 word4 |= sign_bit[0] | sign_bit[1] | sign_bit[2]; 2687 /* fall through */ 2688 case PIPE_FORMAT_BPTC_RGB_UFLOAT: 2689 result = FMT_BC6; 2690 goto out_word4; 2691 default: 2692 goto out_unknown; 2693 } 2694 } 2695 2696 if (desc->layout == UTIL_FORMAT_LAYOUT_SUBSAMPLED) { 2697 switch (format) { 2698 case PIPE_FORMAT_R8G8_B8G8_UNORM: 2699 case PIPE_FORMAT_G8R8_B8R8_UNORM: 2700 result = FMT_GB_GR; 2701 goto out_word4; 2702 case PIPE_FORMAT_G8R8_G8B8_UNORM: 2703 case PIPE_FORMAT_R8G8_R8B8_UNORM: 2704 result = FMT_BG_RG; 2705 goto out_word4; 2706 default: 2707 goto out_unknown; 2708 } 2709 } 2710 2711 if (format == PIPE_FORMAT_R9G9B9E5_FLOAT) { 2712 result = FMT_5_9_9_9_SHAREDEXP; 2713 goto out_word4; 2714 } else if (format == PIPE_FORMAT_R11G11B10_FLOAT) { 2715 result = FMT_10_11_11_FLOAT; 2716 goto out_word4; 2717 } 2718 2719 2720 for (i = 0; i < desc->nr_channels; i++) { 2721 if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) { 2722 word4 |= sign_bit[i]; 2723 } 2724 } 2725 2726 /* R8G8Bx_SNORM - XXX CxV8U8 */ 2727 2728 /* See whether the components are of the same size. */ 2729 for (i = 1; i < desc->nr_channels; i++) { 2730 uniform = uniform && desc->channel[0].size == desc->channel[i].size; 2731 } 2732 2733 /* Non-uniform formats. */ 2734 if (!uniform) { 2735 if (desc->colorspace != UTIL_FORMAT_COLORSPACE_SRGB && 2736 desc->channel[0].pure_integer) 2737 word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT); 2738 switch(desc->nr_channels) { 2739 case 3: 2740 if (desc->channel[0].size == 5 && 2741 desc->channel[1].size == 6 && 2742 desc->channel[2].size == 5) { 2743 result = FMT_5_6_5; 2744 goto out_word4; 2745 } 2746 goto out_unknown; 2747 case 4: 2748 if (desc->channel[0].size == 5 && 2749 desc->channel[1].size == 5 && 2750 desc->channel[2].size == 5 && 2751 desc->channel[3].size == 1) { 2752 result = FMT_1_5_5_5; 2753 goto out_word4; 2754 } 2755 if (desc->channel[0].size == 10 && 2756 desc->channel[1].size == 10 && 2757 desc->channel[2].size == 10 && 2758 desc->channel[3].size == 2) { 2759 result = FMT_2_10_10_10; 2760 goto out_word4; 2761 } 2762 goto out_unknown; 2763 } 2764 goto out_unknown; 2765 } 2766 2767 /* Find the first non-VOID channel. */ 2768 for (i = 0; i < 4; i++) { 2769 if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) { 2770 break; 2771 } 2772 } 2773 2774 if (i == 4) 2775 goto out_unknown; 2776 2777 /* uniform formats */ 2778 switch (desc->channel[i].type) { 2779 case UTIL_FORMAT_TYPE_UNSIGNED: 2780 case UTIL_FORMAT_TYPE_SIGNED: 2781 #if 0 2782 if (!desc->channel[i].normalized && 2783 desc->colorspace != UTIL_FORMAT_COLORSPACE_SRGB) { 2784 goto out_unknown; 2785 } 2786 #endif 2787 if (desc->colorspace != UTIL_FORMAT_COLORSPACE_SRGB && 2788 desc->channel[i].pure_integer) 2789 word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT); 2790 2791 switch (desc->channel[i].size) { 2792 case 4: 2793 switch (desc->nr_channels) { 2794 case 2: 2795 result = FMT_4_4; 2796 goto out_word4; 2797 case 4: 2798 result = FMT_4_4_4_4; 2799 goto out_word4; 2800 } 2801 goto out_unknown; 2802 case 8: 2803 switch (desc->nr_channels) { 2804 case 1: 2805 result = FMT_8; 2806 goto out_word4; 2807 case 2: 2808 result = FMT_8_8; 2809 goto out_word4; 2810 case 4: 2811 result = FMT_8_8_8_8; 2812 is_srgb_valid = TRUE; 2813 goto out_word4; 2814 } 2815 goto out_unknown; 2816 case 16: 2817 switch (desc->nr_channels) { 2818 case 1: 2819 result = FMT_16; 2820 goto out_word4; 2821 case 2: 2822 result = FMT_16_16; 2823 goto out_word4; 2824 case 4: 2825 result = FMT_16_16_16_16; 2826 goto out_word4; 2827 } 2828 goto out_unknown; 2829 case 32: 2830 switch (desc->nr_channels) { 2831 case 1: 2832 result = FMT_32; 2833 goto out_word4; 2834 case 2: 2835 result = FMT_32_32; 2836 goto out_word4; 2837 case 4: 2838 result = FMT_32_32_32_32; 2839 goto out_word4; 2840 } 2841 } 2842 goto out_unknown; 2843 2844 case UTIL_FORMAT_TYPE_FLOAT: 2845 switch (desc->channel[i].size) { 2846 case 16: 2847 switch (desc->nr_channels) { 2848 case 1: 2849 result = FMT_16_FLOAT; 2850 goto out_word4; 2851 case 2: 2852 result = FMT_16_16_FLOAT; 2853 goto out_word4; 2854 case 4: 2855 result = FMT_16_16_16_16_FLOAT; 2856 goto out_word4; 2857 } 2858 goto out_unknown; 2859 case 32: 2860 switch (desc->nr_channels) { 2861 case 1: 2862 result = FMT_32_FLOAT; 2863 goto out_word4; 2864 case 2: 2865 result = FMT_32_32_FLOAT; 2866 goto out_word4; 2867 case 4: 2868 result = FMT_32_32_32_32_FLOAT; 2869 goto out_word4; 2870 } 2871 } 2872 goto out_unknown; 2873 } 2874 2875 out_word4: 2876 2877 if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB && !is_srgb_valid) 2878 return ~0; 2879 if (word4_p) 2880 *word4_p = word4; 2881 if (yuv_format_p) 2882 *yuv_format_p = yuv_format; 2883 return result; 2884 out_unknown: 2885 /* R600_ERR("Unable to handle texformat %d %s\n", format, util_format_name(format)); */ 2886 return ~0; 2887 } 2888 2889 uint32_t r600_translate_colorformat(enum chip_class chip, enum pipe_format format, 2890 bool do_endian_swap) 2891 { 2892 const struct util_format_description *desc = util_format_description(format); 2893 int channel = util_format_get_first_non_void_channel(format); 2894 bool is_float; 2895 if (!desc) 2896 return ~0U; 2897 2898 #define HAS_SIZE(x,y,z,w) \ 2899 (desc->channel[0].size == (x) && desc->channel[1].size == (y) && \ 2900 desc->channel[2].size == (z) && desc->channel[3].size == (w)) 2901 2902 if (format == PIPE_FORMAT_R11G11B10_FLOAT) /* isn't plain */ 2903 return V_0280A0_COLOR_10_11_11_FLOAT; 2904 2905 if (desc->layout != UTIL_FORMAT_LAYOUT_PLAIN || 2906 channel == -1) 2907 return ~0U; 2908 2909 is_float = desc->channel[channel].type == UTIL_FORMAT_TYPE_FLOAT; 2910 2911 switch (desc->nr_channels) { 2912 case 1: 2913 switch (desc->channel[0].size) { 2914 case 8: 2915 return V_0280A0_COLOR_8; 2916 case 16: 2917 if (is_float) 2918 return V_0280A0_COLOR_16_FLOAT; 2919 else 2920 return V_0280A0_COLOR_16; 2921 case 32: 2922 if (is_float) 2923 return V_0280A0_COLOR_32_FLOAT; 2924 else 2925 return V_0280A0_COLOR_32; 2926 } 2927 break; 2928 case 2: 2929 if (desc->channel[0].size == desc->channel[1].size) { 2930 switch (desc->channel[0].size) { 2931 case 4: 2932 if (chip <= R700) 2933 return V_0280A0_COLOR_4_4; 2934 else 2935 return ~0U; /* removed on Evergreen */ 2936 case 8: 2937 return V_0280A0_COLOR_8_8; 2938 case 16: 2939 if (is_float) 2940 return V_0280A0_COLOR_16_16_FLOAT; 2941 else 2942 return V_0280A0_COLOR_16_16; 2943 case 32: 2944 if (is_float) 2945 return V_0280A0_COLOR_32_32_FLOAT; 2946 else 2947 return V_0280A0_COLOR_32_32; 2948 } 2949 } else if (HAS_SIZE(8,24,0,0)) { 2950 return (do_endian_swap ? V_0280A0_COLOR_8_24 : V_0280A0_COLOR_24_8); 2951 } else if (HAS_SIZE(24,8,0,0)) { 2952 return V_0280A0_COLOR_8_24; 2953 } 2954 break; 2955 case 3: 2956 if (HAS_SIZE(5,6,5,0)) { 2957 return V_0280A0_COLOR_5_6_5; 2958 } else if (HAS_SIZE(32,8,24,0)) { 2959 return V_0280A0_COLOR_X24_8_32_FLOAT; 2960 } 2961 break; 2962 case 4: 2963 if (desc->channel[0].size == desc->channel[1].size && 2964 desc->channel[0].size == desc->channel[2].size && 2965 desc->channel[0].size == desc->channel[3].size) { 2966 switch (desc->channel[0].size) { 2967 case 4: 2968 return V_0280A0_COLOR_4_4_4_4; 2969 case 8: 2970 return V_0280A0_COLOR_8_8_8_8; 2971 case 16: 2972 if (is_float) 2973 return V_0280A0_COLOR_16_16_16_16_FLOAT; 2974 else 2975 return V_0280A0_COLOR_16_16_16_16; 2976 case 32: 2977 if (is_float) 2978 return V_0280A0_COLOR_32_32_32_32_FLOAT; 2979 else 2980 return V_0280A0_COLOR_32_32_32_32; 2981 } 2982 } else if (HAS_SIZE(5,5,5,1)) { 2983 return V_0280A0_COLOR_1_5_5_5; 2984 } else if (HAS_SIZE(10,10,10,2)) { 2985 return V_0280A0_COLOR_2_10_10_10; 2986 } 2987 break; 2988 } 2989 return ~0U; 2990 } 2991 2992 uint32_t r600_colorformat_endian_swap(uint32_t colorformat, bool do_endian_swap) 2993 { 2994 if (R600_BIG_ENDIAN) { 2995 switch(colorformat) { 2996 /* 8-bit buffers. */ 2997 case V_0280A0_COLOR_4_4: 2998 case V_0280A0_COLOR_8: 2999 return ENDIAN_NONE; 3000 3001 /* 16-bit buffers. */ 3002 case V_0280A0_COLOR_8_8: 3003 /* 3004 * No need to do endian swaps on array formats, 3005 * as mesa<-->pipe formats conversion take into account 3006 * the endianess 3007 */ 3008 return ENDIAN_NONE; 3009 3010 case V_0280A0_COLOR_5_6_5: 3011 case V_0280A0_COLOR_1_5_5_5: 3012 case V_0280A0_COLOR_4_4_4_4: 3013 case V_0280A0_COLOR_16: 3014 return (do_endian_swap ? ENDIAN_8IN16 : ENDIAN_NONE); 3015 3016 /* 32-bit buffers. */ 3017 case V_0280A0_COLOR_8_8_8_8: 3018 /* 3019 * No need to do endian swaps on array formats, 3020 * as mesa<-->pipe formats conversion take into account 3021 * the endianess 3022 */ 3023 return ENDIAN_NONE; 3024 3025 case V_0280A0_COLOR_2_10_10_10: 3026 case V_0280A0_COLOR_8_24: 3027 case V_0280A0_COLOR_24_8: 3028 case V_0280A0_COLOR_32_FLOAT: 3029 return (do_endian_swap ? ENDIAN_8IN32 : ENDIAN_NONE); 3030 3031 case V_0280A0_COLOR_16_16_FLOAT: 3032 case V_0280A0_COLOR_16_16: 3033 return ENDIAN_8IN16; 3034 3035 /* 64-bit buffers. */ 3036 case V_0280A0_COLOR_16_16_16_16: 3037 case V_0280A0_COLOR_16_16_16_16_FLOAT: 3038 return ENDIAN_8IN16; 3039 3040 case V_0280A0_COLOR_32_32_FLOAT: 3041 case V_0280A0_COLOR_32_32: 3042 case V_0280A0_COLOR_X24_8_32_FLOAT: 3043 return ENDIAN_8IN32; 3044 3045 /* 128-bit buffers. */ 3046 case V_0280A0_COLOR_32_32_32_32_FLOAT: 3047 case V_0280A0_COLOR_32_32_32_32: 3048 return ENDIAN_8IN32; 3049 default: 3050 return ENDIAN_NONE; /* Unsupported. */ 3051 } 3052 } else { 3053 return ENDIAN_NONE; 3054 } 3055 } 3056 3057 static void r600_invalidate_buffer(struct pipe_context *ctx, struct pipe_resource *buf) 3058 { 3059 struct r600_context *rctx = (struct r600_context*)ctx; 3060 struct r600_resource *rbuffer = r600_resource(buf); 3061 unsigned i, shader, mask; 3062 struct r600_pipe_sampler_view *view; 3063 3064 /* Reallocate the buffer in the same pipe_resource. */ 3065 r600_alloc_resource(&rctx->screen->b, rbuffer); 3066 3067 /* We changed the buffer, now we need to bind it where the old one was bound. */ 3068 /* Vertex buffers. */ 3069 mask = rctx->vertex_buffer_state.enabled_mask; 3070 while (mask) { 3071 i = u_bit_scan(&mask); 3072 if (rctx->vertex_buffer_state.vb[i].buffer.resource == &rbuffer->b.b) { 3073 rctx->vertex_buffer_state.dirty_mask |= 1 << i; 3074 r600_vertex_buffers_dirty(rctx); 3075 } 3076 } 3077 /* Streamout buffers. */ 3078 for (i = 0; i < rctx->b.streamout.num_targets; i++) { 3079 if (rctx->b.streamout.targets[i] && 3080 rctx->b.streamout.targets[i]->b.buffer == &rbuffer->b.b) { 3081 if (rctx->b.streamout.begin_emitted) { 3082 r600_emit_streamout_end(&rctx->b); 3083 } 3084 rctx->b.streamout.append_bitmask = rctx->b.streamout.enabled_mask; 3085 r600_streamout_buffers_dirty(&rctx->b); 3086 } 3087 } 3088 3089 /* Constant buffers. */ 3090 for (shader = 0; shader < PIPE_SHADER_TYPES; shader++) { 3091 struct r600_constbuf_state *state = &rctx->constbuf_state[shader]; 3092 bool found = false; 3093 uint32_t mask = state->enabled_mask; 3094 3095 while (mask) { 3096 unsigned i = u_bit_scan(&mask); 3097 if (state->cb[i].buffer == &rbuffer->b.b) { 3098 found = true; 3099 state->dirty_mask |= 1 << i; 3100 } 3101 } 3102 if (found) { 3103 r600_constant_buffers_dirty(rctx, state); 3104 } 3105 } 3106 3107 /* Texture buffer objects - update the virtual addresses in descriptors. */ 3108 LIST_FOR_EACH_ENTRY(view, &rctx->texture_buffers, list) { 3109 if (view->base.texture == &rbuffer->b.b) { 3110 uint64_t offset = view->base.u.buf.offset; 3111 uint64_t va = rbuffer->gpu_address + offset; 3112 3113 view->tex_resource_words[0] = va; 3114 view->tex_resource_words[2] &= C_038008_BASE_ADDRESS_HI; 3115 view->tex_resource_words[2] |= S_038008_BASE_ADDRESS_HI(va >> 32); 3116 } 3117 } 3118 /* Texture buffer objects - make bindings dirty if needed. */ 3119 for (shader = 0; shader < PIPE_SHADER_TYPES; shader++) { 3120 struct r600_samplerview_state *state = &rctx->samplers[shader].views; 3121 bool found = false; 3122 uint32_t mask = state->enabled_mask; 3123 3124 while (mask) { 3125 unsigned i = u_bit_scan(&mask); 3126 if (state->views[i]->base.texture == &rbuffer->b.b) { 3127 found = true; 3128 state->dirty_mask |= 1 << i; 3129 } 3130 } 3131 if (found) { 3132 r600_sampler_views_dirty(rctx, state); 3133 } 3134 } 3135 3136 /* SSBOs */ 3137 struct r600_image_state *istate = &rctx->fragment_buffers; 3138 { 3139 uint32_t mask = istate->enabled_mask; 3140 bool found = false; 3141 while (mask) { 3142 unsigned i = u_bit_scan(&mask); 3143 if (istate->views[i].base.resource == &rbuffer->b.b) { 3144 found = true; 3145 istate->dirty_mask |= 1 << i; 3146 } 3147 } 3148 if (found) { 3149 r600_mark_atom_dirty(rctx, &istate->atom); 3150 } 3151 } 3152 3153 } 3154 3155 static void r600_set_active_query_state(struct pipe_context *ctx, boolean enable) 3156 { 3157 struct r600_context *rctx = (struct r600_context*)ctx; 3158 3159 /* Pipeline stat & streamout queries. */ 3160 if (enable) { 3161 rctx->b.flags &= ~R600_CONTEXT_STOP_PIPELINE_STATS; 3162 rctx->b.flags |= R600_CONTEXT_START_PIPELINE_STATS; 3163 } else { 3164 rctx->b.flags &= ~R600_CONTEXT_START_PIPELINE_STATS; 3165 rctx->b.flags |= R600_CONTEXT_STOP_PIPELINE_STATS; 3166 } 3167 3168 /* Occlusion queries. */ 3169 if (rctx->db_misc_state.occlusion_queries_disabled != !enable) { 3170 rctx->db_misc_state.occlusion_queries_disabled = !enable; 3171 r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom); 3172 } 3173 } 3174 3175 static void r600_need_gfx_cs_space(struct pipe_context *ctx, unsigned num_dw, 3176 bool include_draw_vbo) 3177 { 3178 r600_need_cs_space((struct r600_context*)ctx, num_dw, include_draw_vbo); 3179 } 3180 3181 /* keep this at the end of this file, please */ 3182 void r600_init_common_state_functions(struct r600_context *rctx) 3183 { 3184 rctx->b.b.create_fs_state = r600_create_ps_state; 3185 rctx->b.b.create_vs_state = r600_create_vs_state; 3186 rctx->b.b.create_gs_state = r600_create_gs_state; 3187 rctx->b.b.create_tcs_state = r600_create_tcs_state; 3188 rctx->b.b.create_tes_state = r600_create_tes_state; 3189 rctx->b.b.create_vertex_elements_state = r600_create_vertex_fetch_shader; 3190 rctx->b.b.bind_blend_state = r600_bind_blend_state; 3191 rctx->b.b.bind_depth_stencil_alpha_state = r600_bind_dsa_state; 3192 rctx->b.b.bind_sampler_states = r600_bind_sampler_states; 3193 rctx->b.b.bind_fs_state = r600_bind_ps_state; 3194 rctx->b.b.bind_rasterizer_state = r600_bind_rs_state; 3195 rctx->b.b.bind_vertex_elements_state = r600_bind_vertex_elements; 3196 rctx->b.b.bind_vs_state = r600_bind_vs_state; 3197 rctx->b.b.bind_gs_state = r600_bind_gs_state; 3198 rctx->b.b.bind_tcs_state = r600_bind_tcs_state; 3199 rctx->b.b.bind_tes_state = r600_bind_tes_state; 3200 rctx->b.b.delete_blend_state = r600_delete_blend_state; 3201 rctx->b.b.delete_depth_stencil_alpha_state = r600_delete_dsa_state; 3202 rctx->b.b.delete_fs_state = r600_delete_ps_state; 3203 rctx->b.b.delete_rasterizer_state = r600_delete_rs_state; 3204 rctx->b.b.delete_sampler_state = r600_delete_sampler_state; 3205 rctx->b.b.delete_vertex_elements_state = r600_delete_vertex_elements; 3206 rctx->b.b.delete_vs_state = r600_delete_vs_state; 3207 rctx->b.b.delete_gs_state = r600_delete_gs_state; 3208 rctx->b.b.delete_tcs_state = r600_delete_tcs_state; 3209 rctx->b.b.delete_tes_state = r600_delete_tes_state; 3210 rctx->b.b.set_blend_color = r600_set_blend_color; 3211 rctx->b.b.set_clip_state = r600_set_clip_state; 3212 rctx->b.b.set_constant_buffer = r600_set_constant_buffer; 3213 rctx->b.b.set_sample_mask = r600_set_sample_mask; 3214 rctx->b.b.set_stencil_ref = r600_set_pipe_stencil_ref; 3215 rctx->b.b.set_vertex_buffers = r600_set_vertex_buffers; 3216 rctx->b.b.set_sampler_views = r600_set_sampler_views; 3217 rctx->b.b.sampler_view_destroy = r600_sampler_view_destroy; 3218 rctx->b.b.memory_barrier = r600_memory_barrier; 3219 rctx->b.b.texture_barrier = r600_texture_barrier; 3220 rctx->b.b.set_stream_output_targets = r600_set_streamout_targets; 3221 rctx->b.b.set_active_query_state = r600_set_active_query_state; 3222 rctx->b.b.draw_vbo = r600_draw_vbo; 3223 rctx->b.invalidate_buffer = r600_invalidate_buffer; 3224 rctx->b.need_gfx_cs_space = r600_need_gfx_cs_space; 3225 } 3226