d3d11_context.h revision 6ab6c43079083c811761241b901d07f9ad2c9aff
1/************************************************************************** 2 * 3 * Copyright 2010 Luca Barbieri 4 * 5 * Permission is hereby granted, free of charge, to any person obtaining 6 * a copy of this software and associated documentation files (the 7 * "Software"), to deal in the Software without restriction, including 8 * without limitation the rights to use, copy, modify, merge, publish, 9 * distribute, sublicense, and/or sell copies of the Software, and to 10 * permit persons to whom the Software is furnished to do so, subject to 11 * the following conditions: 12 * 13 * The above copyright notice and this permission notice (including the 14 * next paragraph) shall be included in all copies or substantial 15 * portions of the Software. 16 * 17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 18 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 19 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. 20 * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE 21 * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 22 * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 23 * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 24 * 25 **************************************************************************/ 26 27/* used to unbind things, we need 128 due to resources */ 28static const void* zero_data[128]; 29 30#define UPDATE_VIEWS_SHIFT (D3D11_STAGES * 0) 31#define UPDATE_SAMPLERS_SHIFT (D3D11_STAGES * 1) 32#define UPDATE_VERTEX_BUFFERS (1 << (D3D11_STAGES * 2)) 33 34#if API >= 11 35template<typename PtrTraits> 36struct GalliumD3D11DeviceContext : 37 public GalliumD3D11DeviceChild<ID3D11DeviceContext> 38{ 39#else 40template<bool threadsafe> 41struct GalliumD3D10Device : public GalliumD3D10ScreenImpl<threadsafe> 42{ 43 typedef simple_ptr_traits PtrTraits; 44 typedef GalliumD3D10Device GalliumD3D10DeviceContext; 45#endif 46 47 refcnt_ptr<GalliumD3D11Shader<>, PtrTraits> shaders[D3D11_STAGES]; 48 refcnt_ptr<GalliumD3D11InputLayout, PtrTraits> input_layout; 49 refcnt_ptr<GalliumD3D11Buffer, PtrTraits> index_buffer; 50 refcnt_ptr<GalliumD3D11RasterizerState, PtrTraits> rasterizer_state; 51 refcnt_ptr<GalliumD3D11DepthStencilState, PtrTraits> depth_stencil_state; 52 refcnt_ptr<GalliumD3D11BlendState, PtrTraits> blend_state; 53 refcnt_ptr<GalliumD3D11DepthStencilView, PtrTraits> depth_stencil_view; 54 refcnt_ptr<GalliumD3D11Predicate, PtrTraits> render_predicate; 55 56 refcnt_ptr<GalliumD3D11Buffer, PtrTraits> constant_buffers[D3D11_STAGES][D3D11_COMMONSHADER_CONSTANT_BUFFER_API_SLOT_COUNT]; 57 refcnt_ptr<GalliumD3D11ShaderResourceView, PtrTraits> shader_resource_views[D3D11_STAGES][D3D11_COMMONSHADER_INPUT_RESOURCE_SLOT_COUNT]; 58 refcnt_ptr<GalliumD3D11SamplerState, PtrTraits> samplers[D3D11_STAGES][D3D11_COMMONSHADER_SAMPLER_SLOT_COUNT]; 59 refcnt_ptr<GalliumD3D11Buffer, PtrTraits> input_buffers[D3D11_IA_VERTEX_INPUT_RESOURCE_SLOT_COUNT]; 60 refcnt_ptr<GalliumD3D11RenderTargetView, PtrTraits> render_target_views[D3D11_SIMULTANEOUS_RENDER_TARGET_COUNT]; 61 refcnt_ptr<GalliumD3D11Buffer, PtrTraits> so_targets[D3D11_SO_BUFFER_SLOT_COUNT]; 62 63#if API >= 11 64 refcnt_ptr<ID3D11UnorderedAccessView, PtrTraits> cs_unordered_access_views[D3D11_PS_CS_UAV_REGISTER_COUNT]; 65 refcnt_ptr<ID3D11UnorderedAccessView, PtrTraits> om_unordered_access_views[D3D11_SIMULTANEOUS_RENDER_TARGET_COUNT]; 66#endif 67 68 D3D11_VIEWPORT viewports[D3D11_VIEWPORT_AND_SCISSORRECT_OBJECT_COUNT_PER_PIPELINE]; 69 D3D11_RECT scissor_rects[D3D11_VIEWPORT_AND_SCISSORRECT_OBJECT_COUNT_PER_PIPELINE]; 70 unsigned so_offsets[D3D11_SO_BUFFER_SLOT_COUNT]; 71 D3D11_PRIMITIVE_TOPOLOGY primitive_topology; 72 DXGI_FORMAT index_format; 73 unsigned index_offset; 74 uint32_t strip_cut_index; 75 BOOL render_predicate_value; 76 float blend_color[4]; 77 unsigned sample_mask; 78 unsigned stencil_ref; 79 bool depth_clamp; 80 81 void* default_input_layout; 82 void* default_rasterizer; 83 void* default_depth_stencil; 84 void* default_blend; 85 void* default_sampler; 86 void* default_shaders[D3D11_STAGES]; 87 88 // derived state 89 int primitive_mode; 90 struct pipe_vertex_buffer vertex_buffers[D3D11_IA_VERTEX_INPUT_RESOURCE_SLOT_COUNT]; 91 struct pipe_resource* so_buffers[D3D11_SO_BUFFER_SLOT_COUNT]; 92 struct pipe_sampler_view* sampler_views[D3D11_STAGES][D3D11_COMMONSHADER_INPUT_RESOURCE_SLOT_COUNT]; 93 void* sampler_csos[D3D11_STAGES][D3D11_COMMONSHADER_SAMPLER_SLOT_COUNT]; 94 struct pipe_resource* buffers[D3D11_SO_BUFFER_SLOT_COUNT]; 95 unsigned num_shader_resource_views[D3D11_STAGES]; 96 unsigned num_samplers[D3D11_STAGES]; 97 unsigned num_vertex_buffers; 98 unsigned num_render_target_views; 99 unsigned num_viewports; 100 unsigned num_scissor_rects; 101 unsigned num_so_targets; 102 103 struct pipe_context* pipe; 104 unsigned update_flags; 105 106 bool owns_pipe; 107 unsigned context_flags; 108 109 GalliumD3D11Caps caps; 110 111 cso_context* cso_ctx; 112 gen_mipmap_state* gen_mipmap; 113 114#if API >= 11 115#define SYNCHRONIZED do {} while(0) 116 117 GalliumD3D11DeviceContext(GalliumD3D11Screen* device, pipe_context* pipe, bool owns_pipe, unsigned context_flags = 0) 118 : GalliumD3D11DeviceChild<ID3D11DeviceContext>(device), pipe(pipe), owns_pipe(owns_pipe), context_flags(context_flags) 119 { 120 caps = device->screen_caps; 121 init_context(); 122 } 123 124 ~GalliumD3D11DeviceContext() 125 { 126 destroy_context(); 127 } 128#else 129#define SYNCHRONIZED lock_t<maybe_mutex_t<threadsafe> > lock_(this->mutex) 130 131 GalliumD3D10Device(pipe_screen* screen, pipe_context* pipe, bool owns_pipe, unsigned creation_flags, IDXGIAdapter* adapter) 132 : GalliumD3D10ScreenImpl<threadsafe>(screen, pipe, owns_pipe, creation_flags, adapter), pipe(pipe), owns_pipe(owns_pipe), context_flags(0) 133 { 134 caps = this->screen_caps; 135 init_context(); 136 } 137 138 ~GalliumD3D10Device() 139 { 140 destroy_context(); 141 } 142#endif 143 144 void init_context() 145 { 146 if(!pipe->begin_query) 147 caps.queries = false; 148 if(!pipe->bind_gs_state) 149 { 150 caps.gs = false; 151 caps.stages = 2; 152 } 153 if(!pipe->set_stream_output_buffers) 154 caps.so = false; 155 if(!pipe->set_geometry_sampler_views) 156 caps.stages_with_sampling &=~ (1 << PIPE_SHADER_GEOMETRY); 157 if(!pipe->set_fragment_sampler_views) 158 caps.stages_with_sampling &=~ (1 << PIPE_SHADER_FRAGMENT); 159 if(!pipe->set_vertex_sampler_views) 160 caps.stages_with_sampling &=~ (1 << PIPE_SHADER_VERTEX); 161 162 update_flags = 0; 163 164 // pipeline state 165 memset(viewports, 0, sizeof(viewports)); 166 memset(scissor_rects, 0, sizeof(scissor_rects)); 167 memset(so_offsets, 0, sizeof(so_offsets)); 168 primitive_topology = D3D11_PRIMITIVE_TOPOLOGY_UNDEFINED; 169 index_format = DXGI_FORMAT_UNKNOWN; 170 index_offset = 0; 171 strip_cut_index = 0xffffffff; 172 render_predicate_value = 0; 173 memset(blend_color, 0, sizeof(blend_color)); 174 sample_mask = ~0; 175 stencil_ref = 0; 176 depth_clamp = 0; 177 178 // derived state 179 primitive_mode = 0; 180 memset(vertex_buffers, 0, sizeof(vertex_buffers)); 181 memset(so_buffers, 0, sizeof(so_buffers)); 182 memset(sampler_views, 0, sizeof(sampler_views)); 183 memset(sampler_csos, 0, sizeof(sampler_csos)); 184 memset(num_shader_resource_views, 0, sizeof(num_shader_resource_views)); 185 memset(num_samplers, 0, sizeof(num_samplers)); 186 num_vertex_buffers = 0; 187 num_render_target_views = 0; 188 num_viewports = 0; 189 num_scissor_rects = 0; 190 num_so_targets = 0; 191 192 default_input_layout = pipe->create_vertex_elements_state(pipe, 0, 0); 193 194 struct pipe_rasterizer_state rasterizerd; 195 memset(&rasterizerd, 0, sizeof(rasterizerd)); 196 rasterizerd.gl_rasterization_rules = 1; 197 rasterizerd.cull_face = PIPE_FACE_BACK; 198 rasterizerd.flatshade_first = 1; 199 rasterizerd.line_width = 1.0f; 200 rasterizerd.point_size = 1.0f; 201 default_rasterizer = pipe->create_rasterizer_state(pipe, &rasterizerd); 202 203 struct pipe_depth_stencil_alpha_state depth_stencild; 204 memset(&depth_stencild, 0, sizeof(depth_stencild)); 205 depth_stencild.depth.enabled = TRUE; 206 depth_stencild.depth.writemask = 1; 207 depth_stencild.depth.func = PIPE_FUNC_LESS; 208 default_depth_stencil = pipe->create_depth_stencil_alpha_state(pipe, &depth_stencild); 209 210 struct pipe_blend_state blendd; 211 memset(&blendd, 0, sizeof(blendd)); 212 blendd.rt[0].colormask = 0xf; 213 default_blend = pipe->create_blend_state(pipe, &blendd); 214 215 struct pipe_sampler_state samplerd; 216 memset(&samplerd, 0, sizeof(samplerd)); 217 samplerd.normalized_coords = 1; 218 samplerd.min_img_filter = PIPE_TEX_FILTER_LINEAR; 219 samplerd.mag_img_filter = PIPE_TEX_FILTER_LINEAR; 220 samplerd.min_mip_filter = PIPE_TEX_MIPFILTER_LINEAR; 221 samplerd.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 222 samplerd.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 223 samplerd.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE; 224 samplerd.border_color.f[0] = 1.0f; 225 samplerd.border_color.f[1] = 1.0f; 226 samplerd.border_color.f[2] = 1.0f; 227 samplerd.border_color.f[3] = 1.0f; 228 samplerd.min_lod = -FLT_MAX; 229 samplerd.max_lod = FLT_MAX; 230 samplerd.max_anisotropy = 1; 231 default_sampler = pipe->create_sampler_state(pipe, &samplerd); 232 233 memset(&samplerd, 0, sizeof(samplerd)); 234 samplerd.normalized_coords = 0; 235 samplerd.min_img_filter = PIPE_TEX_FILTER_NEAREST; 236 samplerd.mag_img_filter = PIPE_TEX_FILTER_NEAREST; 237 samplerd.min_mip_filter = PIPE_TEX_MIPFILTER_NONE; 238 samplerd.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_BORDER; 239 samplerd.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_BORDER; 240 samplerd.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_BORDER; 241 samplerd.min_lod = -FLT_MAX; 242 samplerd.max_lod = FLT_MAX; 243 samplerd.max_anisotropy = 1; 244 245 for(unsigned s = 0; s < D3D11_STAGES; ++s) 246 for(unsigned i = 0; i < D3D11_COMMONSHADER_SAMPLER_SLOT_COUNT; ++i) 247 sampler_csos[s][i] = default_sampler; 248 249 // TODO: should this really be empty shaders, or should they be all-passthrough? 250 memset(default_shaders, 0, sizeof(default_shaders)); 251 struct ureg_program *ureg; 252 ureg = ureg_create(TGSI_PROCESSOR_FRAGMENT); 253 ureg_END(ureg); 254 default_shaders[PIPE_SHADER_FRAGMENT] = ureg_create_shader_and_destroy(ureg, pipe); 255 256 ureg = ureg_create(TGSI_PROCESSOR_VERTEX); 257 ureg_END(ureg); 258 default_shaders[PIPE_SHADER_VERTEX] = ureg_create_shader_and_destroy(ureg, pipe); 259 260 cso_ctx = cso_create_context(pipe); 261 gen_mipmap = util_create_gen_mipmap(pipe, cso_ctx); 262 263 RestoreGalliumState(); 264 } 265 266 void destroy_context() 267 { 268 util_destroy_gen_mipmap(gen_mipmap); 269 cso_destroy_context(cso_ctx); 270 271 pipe->bind_vertex_elements_state(pipe, 0); 272 pipe->delete_vertex_elements_state(pipe, default_input_layout); 273 274 pipe->bind_rasterizer_state(pipe, 0); 275 pipe->delete_rasterizer_state(pipe, default_rasterizer); 276 277 pipe->bind_depth_stencil_alpha_state(pipe, 0); 278 pipe->delete_depth_stencil_alpha_state(pipe, default_depth_stencil); 279 280 pipe->bind_blend_state(pipe, 0); 281 pipe->delete_blend_state(pipe, default_blend); 282 283 pipe->bind_fragment_sampler_states(pipe, 0, 0); 284 pipe->bind_vertex_sampler_states(pipe, 0, 0); 285 if(pipe->bind_geometry_sampler_states) 286 pipe->bind_geometry_sampler_states(pipe, 0, 0); 287 pipe->delete_sampler_state(pipe, default_sampler); 288 289 pipe->bind_fs_state(pipe, 0); 290 pipe->delete_fs_state(pipe, default_shaders[PIPE_SHADER_FRAGMENT]); 291 292 pipe->bind_vs_state(pipe, 0); 293 pipe->delete_vs_state(pipe, default_shaders[PIPE_SHADER_VERTEX]); 294 295 if(owns_pipe) 296 pipe->destroy(pipe); 297 } 298 299 virtual unsigned STDMETHODCALLTYPE GetContextFlags(void) 300 { 301 return context_flags; 302 } 303#if API >= 11 304#define SET_SHADER_EXTRA_ARGS , \ 305 ID3D11ClassInstance *const *ppClassInstances, \ 306 unsigned count 307#define GET_SHADER_EXTRA_ARGS , \ 308 ID3D11ClassInstance **ppClassInstances, \ 309 unsigned *out_count 310#else 311#define SET_SHADER_EXTRA_ARGS 312#define GET_SHADER_EXTRA_ARGS 313#endif 314 315/* On Windows D3D11, SetConstantBuffers and SetShaderResources crash if passed a null pointer. 316 * Instead, you have to pass a pointer to nulls to unbind things. 317 * We do the same. 318 * TODO: is D3D10 the same? 319 */ 320 template<unsigned s> 321 void xs_set_shader(GalliumD3D11Shader<>* shader) 322 { 323 if(shader != shaders[s].p) 324 { 325 shaders[s] = shader; 326 void* shader_cso = shader ? shader->object : default_shaders[s]; 327 switch(s) 328 { 329 case PIPE_SHADER_VERTEX: 330 pipe->bind_vs_state(pipe, shader_cso); 331 break; 332 case PIPE_SHADER_FRAGMENT: 333 pipe->bind_fs_state(pipe, shader_cso); 334 break; 335 case PIPE_SHADER_GEOMETRY: 336 pipe->bind_gs_state(pipe, shader_cso); 337 break; 338 } 339 update_flags |= (1 << (UPDATE_SAMPLERS_SHIFT + s)) | (1 << (UPDATE_VIEWS_SHIFT + s)); 340 } 341 } 342 343 template<unsigned s> 344 void xs_set_constant_buffers(unsigned start, unsigned count, GalliumD3D11Buffer *const *constbufs) 345 { 346 for(unsigned i = 0; i < count; ++i) 347 { 348 if(constbufs[i] != constant_buffers[s][start + i].p) 349 { 350 constant_buffers[s][start + i] = constbufs[i]; 351 if(s < caps.stages && start + i < caps.constant_buffers[s]) 352 pipe->set_constant_buffer(pipe, s, start + i, constbufs[i] ? constbufs[i]->resource : NULL); 353 } 354 } 355 } 356 357 template<unsigned s> 358 void xs_set_shader_resources(unsigned start, unsigned count, GalliumD3D11ShaderResourceView *const *srvs) 359 { 360 int last_different = -1; 361 for(unsigned i = 0; i < count; ++i) 362 { 363 if(shader_resource_views[s][start + i].p != srvs[i]) 364 { 365 shader_resource_views[s][start + i] = srvs[i]; 366 sampler_views[s][start + i] = srvs[i] ? srvs[i]->object : 0; 367 last_different = i; 368 } 369 } 370 if(last_different >= 0) 371 { 372 num_shader_resource_views[s] = std::max(num_shader_resource_views[s], start + last_different + 1); 373 update_flags |= 1 << (UPDATE_VIEWS_SHIFT + s); 374 } 375 } 376 377 template<unsigned s> 378 void xs_set_samplers(unsigned start, unsigned count, GalliumD3D11SamplerState *const *samps) 379 { 380 int last_different = -1; 381 for(unsigned i = 0; i < count; ++i) 382 { 383 if(samplers[s][start + i].p != samps[i]) 384 { 385 samplers[s][start + i] = samps[i]; 386 sampler_csos[s][start + i] = samps[i] ? samps[i]->object : default_sampler; 387 last_different = i; 388 } 389 } 390 if(last_different >= 0) 391 { 392 num_samplers[s] = std::max(num_samplers[s], start + last_different + 1); 393 update_flags |= 1 << (UPDATE_SAMPLERS_SHIFT + s); 394 } 395 } 396 397#define IMPLEMENT_SHADER_STAGE(XS, Stage) \ 398 virtual void STDMETHODCALLTYPE XS##SetShader( \ 399 ID3D11##Stage##Shader *pShader \ 400 SET_SHADER_EXTRA_ARGS) \ 401 { \ 402 SYNCHRONIZED; \ 403 xs_set_shader<D3D11_STAGE_##XS>((GalliumD3D11Shader<>*)pShader); \ 404 } \ 405 virtual void STDMETHODCALLTYPE XS##GetShader(\ 406 ID3D11##Stage##Shader **ppShader \ 407 GET_SHADER_EXTRA_ARGS) \ 408 { \ 409 SYNCHRONIZED; \ 410 *ppShader = (ID3D11##Stage##Shader*)shaders[D3D11_STAGE_##XS].ref(); \ 411 } \ 412 virtual void STDMETHODCALLTYPE XS##SetConstantBuffers(\ 413 unsigned start, \ 414 unsigned count, \ 415 ID3D11Buffer *const* constant_buffers) \ 416 { \ 417 SYNCHRONIZED; \ 418 xs_set_constant_buffers<D3D11_STAGE_##XS>(start, count, (GalliumD3D11Buffer *const *)constant_buffers); \ 419 } \ 420 virtual void STDMETHODCALLTYPE XS##GetConstantBuffers(\ 421 unsigned start, \ 422 unsigned count, \ 423 ID3D11Buffer **out_constant_buffers) \ 424 { \ 425 SYNCHRONIZED; \ 426 for(unsigned i = 0; i < count; ++i) \ 427 out_constant_buffers[i] = constant_buffers[D3D11_STAGE_##XS][start + i].ref(); \ 428 } \ 429 virtual void STDMETHODCALLTYPE XS##SetShaderResources(\ 430 unsigned start, \ 431 unsigned count, \ 432 ID3D11ShaderResourceView *const *new_shader_resource_views) \ 433 { \ 434 SYNCHRONIZED; \ 435 xs_set_shader_resources<D3D11_STAGE_##XS>(start, count, (GalliumD3D11ShaderResourceView *const *)new_shader_resource_views); \ 436 } \ 437 virtual void STDMETHODCALLTYPE XS##GetShaderResources(\ 438 unsigned start, \ 439 unsigned count, \ 440 ID3D11ShaderResourceView **out_shader_resource_views) \ 441 { \ 442 SYNCHRONIZED; \ 443 for(unsigned i = 0; i < count; ++i) \ 444 out_shader_resource_views[i] = shader_resource_views[D3D11_STAGE_##XS][start + i].ref(); \ 445 } \ 446 virtual void STDMETHODCALLTYPE XS##SetSamplers(\ 447 unsigned start, \ 448 unsigned count, \ 449 ID3D11SamplerState *const *new_samplers) \ 450 { \ 451 SYNCHRONIZED; \ 452 xs_set_samplers<D3D11_STAGE_##XS>(start, count, (GalliumD3D11SamplerState *const *)new_samplers); \ 453 } \ 454 virtual void STDMETHODCALLTYPE XS##GetSamplers( \ 455 unsigned start, \ 456 unsigned count, \ 457 ID3D11SamplerState **out_samplers) \ 458 { \ 459 SYNCHRONIZED; \ 460 for(unsigned i = 0; i < count; ++i) \ 461 out_samplers[i] = samplers[D3D11_STAGE_##XS][start + i].ref(); \ 462 } 463 464#define DO_VS(x) x 465#define DO_GS(x) do {if(caps.gs) {x;}} while(0) 466#define DO_PS(x) x 467#define DO_HS(x) 468#define DO_DS(x) 469#define DO_CS(x) 470 IMPLEMENT_SHADER_STAGE(VS, Vertex) 471 IMPLEMENT_SHADER_STAGE(GS, Geometry) 472 IMPLEMENT_SHADER_STAGE(PS, Pixel) 473 474#if API >= 11 475 IMPLEMENT_SHADER_STAGE(HS, Hull) 476 IMPLEMENT_SHADER_STAGE(DS, Domain) 477 IMPLEMENT_SHADER_STAGE(CS, Compute) 478 479 virtual void STDMETHODCALLTYPE CSSetUnorderedAccessViews( 480 unsigned start, 481 unsigned count, 482 ID3D11UnorderedAccessView *const *new_unordered_access_views, 483 const unsigned *new_uav_initial_counts) 484 { 485 SYNCHRONIZED; 486 for(unsigned i = 0; i < count; ++i) 487 cs_unordered_access_views[start + i] = new_unordered_access_views[i]; 488 } 489 490 virtual void STDMETHODCALLTYPE CSGetUnorderedAccessViews( 491 unsigned start, 492 unsigned count, 493 ID3D11UnorderedAccessView **out_unordered_access_views) 494 { 495 SYNCHRONIZED; 496 for(unsigned i = 0; i < count; ++i) 497 out_unordered_access_views[i] = cs_unordered_access_views[start + i].ref(); 498 } 499#endif 500 501 template<unsigned s> 502 void update_stage() 503 { 504 if(update_flags & (1 << (UPDATE_VIEWS_SHIFT + s))) 505 { 506 while(num_shader_resource_views[s] && !sampler_views[s][num_shader_resource_views[s] - 1]) \ 507 --num_shader_resource_views[s]; 508 if((1 << s) & caps.stages_with_sampling) 509 { 510 const unsigned num_views_to_bind = num_shader_resource_views[s]; 511 switch(s) 512 { 513 case PIPE_SHADER_VERTEX: 514 pipe->set_vertex_sampler_views(pipe, num_views_to_bind, sampler_views[s]); 515 break; 516 case PIPE_SHADER_FRAGMENT: 517 pipe->set_fragment_sampler_views(pipe, num_views_to_bind, sampler_views[s]); 518 break; 519 case PIPE_SHADER_GEOMETRY: 520 pipe->set_geometry_sampler_views(pipe, num_views_to_bind, sampler_views[s]); 521 break; 522 } 523 } 524 } 525 526 if(update_flags & (1 << (UPDATE_SAMPLERS_SHIFT + s))) 527 { 528 while(num_samplers[s] && !sampler_csos[s][num_samplers[s] - 1]) 529 --num_samplers[s]; 530 if((1 << s) & caps.stages_with_sampling) 531 { 532 const unsigned num_samplers_to_bind = num_samplers[s]; 533 switch(s) 534 { 535 case PIPE_SHADER_VERTEX: 536 pipe->bind_vertex_sampler_states(pipe, num_samplers_to_bind, sampler_csos[s]); 537 break; 538 case PIPE_SHADER_FRAGMENT: 539 pipe->bind_fragment_sampler_states(pipe, num_samplers_to_bind, sampler_csos[s]); 540 break; 541 case PIPE_SHADER_GEOMETRY: 542 pipe->bind_geometry_sampler_states(pipe, num_samplers_to_bind, sampler_csos[s]); 543 break; 544 } 545 } 546 } 547 } 548 549 void update_state() 550 { 551 update_stage<D3D11_STAGE_PS>(); 552 update_stage<D3D11_STAGE_VS>(); 553 update_stage<D3D11_STAGE_GS>(); 554#if API >= 11 555 update_stage<D3D11_STAGE_HS>(); 556 update_stage<D3D11_STAGE_DS>(); 557 update_stage<D3D11_STAGE_CS>(); 558#endif 559 560 if(update_flags & UPDATE_VERTEX_BUFFERS) 561 { 562 while(num_vertex_buffers && !vertex_buffers[num_vertex_buffers - 1].buffer) 563 --num_vertex_buffers; 564 pipe->set_vertex_buffers(pipe, num_vertex_buffers, vertex_buffers); 565 } 566 567 update_flags = 0; 568 } 569 570 virtual void STDMETHODCALLTYPE IASetInputLayout( 571 ID3D11InputLayout *new_input_layout) 572 { 573 SYNCHRONIZED; 574 if(new_input_layout != input_layout.p) 575 { 576 input_layout = new_input_layout; 577 pipe->bind_vertex_elements_state(pipe, new_input_layout ? ((GalliumD3D11InputLayout*)new_input_layout)->object : default_input_layout); 578 } 579 } 580 581 virtual void STDMETHODCALLTYPE IAGetInputLayout( 582 ID3D11InputLayout **out_input_layout) 583 { 584 SYNCHRONIZED; 585 *out_input_layout = input_layout.ref(); 586 } 587 588 virtual void STDMETHODCALLTYPE IASetVertexBuffers( 589 unsigned start, 590 unsigned count, 591 ID3D11Buffer *const *new_vertex_buffers, 592 const unsigned *new_strides, 593 const unsigned *new_offsets) 594 { 595 SYNCHRONIZED; 596 int last_different = -1; 597 for(unsigned i = 0; i < count; ++i) 598 { 599 ID3D11Buffer* buffer = new_vertex_buffers[i]; 600 if(buffer != input_buffers[start + i].p 601 || vertex_buffers[start + i].buffer_offset != new_offsets[i] 602 || vertex_buffers[start + i].stride != new_offsets[i] 603 ) 604 { 605 input_buffers[start + i] = buffer; 606 vertex_buffers[start + i].buffer = buffer ? ((GalliumD3D11Buffer*)buffer)->resource : 0; 607 vertex_buffers[start + i].buffer_offset = new_offsets[i]; 608 vertex_buffers[start + i].stride = new_strides[i]; 609 last_different = i; 610 } 611 } 612 if(last_different >= 0) 613 { 614 num_vertex_buffers = std::max(num_vertex_buffers, start + count); 615 update_flags |= UPDATE_VERTEX_BUFFERS; 616 } 617 } 618 619 virtual void STDMETHODCALLTYPE IAGetVertexBuffers( 620 unsigned start, 621 unsigned count, 622 ID3D11Buffer **out_vertex_buffers, 623 unsigned *out_strides, 624 unsigned *out_offsets) 625 { 626 SYNCHRONIZED; 627 if(out_vertex_buffers) 628 { 629 for(unsigned i = 0; i < count; ++i) 630 out_vertex_buffers[i] = input_buffers[start + i].ref(); 631 } 632 633 if(out_offsets) 634 { 635 for(unsigned i = 0; i < count; ++i) 636 out_offsets[i] = vertex_buffers[start + i].buffer_offset; 637 } 638 639 if(out_strides) 640 { 641 for(unsigned i = 0; i < count; ++i) 642 out_strides[i] = vertex_buffers[start + i].stride; 643 } 644 } 645 646 void set_index_buffer() 647 { 648 pipe_index_buffer ib; 649 if(!index_buffer) 650 { 651 memset(&ib, 0, sizeof(ib)); 652 } 653 else 654 { 655 switch(index_format) { 656 case DXGI_FORMAT_R32_UINT: 657 ib.index_size = 4; 658 strip_cut_index = 0xffffffff; 659 break; 660 case DXGI_FORMAT_R16_UINT: 661 ib.index_size = 2; 662 strip_cut_index = 0xffff; 663 break; 664 default: 665 ib.index_size = 1; 666 strip_cut_index = 0xff; 667 break; 668 } 669 ib.offset = index_offset; 670 ib.buffer = index_buffer ? ((GalliumD3D11Buffer*)index_buffer.p)->resource : 0; 671 } 672 pipe->set_index_buffer(pipe, &ib); 673 } 674 675 virtual void STDMETHODCALLTYPE IASetIndexBuffer( 676 ID3D11Buffer *new_index_buffer, 677 DXGI_FORMAT new_index_format, 678 unsigned new_index_offset) 679 { 680 SYNCHRONIZED; 681 if(index_buffer.p != new_index_buffer || index_format != new_index_format || index_offset != new_index_offset) 682 { 683 index_buffer = new_index_buffer; 684 index_format = new_index_format; 685 index_offset = new_index_offset; 686 687 set_index_buffer(); 688 } 689 } 690 691 virtual void STDMETHODCALLTYPE IAGetIndexBuffer( 692 ID3D11Buffer **out_index_buffer, 693 DXGI_FORMAT *out_index_format, 694 unsigned *out_index_offset) 695 { 696 SYNCHRONIZED; 697 if(out_index_buffer) 698 *out_index_buffer = index_buffer.ref(); 699 if(out_index_format) 700 *out_index_format = index_format; 701 if(out_index_offset) 702 *out_index_offset = index_offset; 703 } 704 705 virtual void STDMETHODCALLTYPE IASetPrimitiveTopology( 706 D3D11_PRIMITIVE_TOPOLOGY new_primitive_topology) 707 { 708 SYNCHRONIZED; 709 if(primitive_topology != new_primitive_topology) 710 { 711 if(new_primitive_topology < D3D_PRIMITIVE_TOPOLOGY_COUNT) 712 primitive_mode = d3d_to_pipe_prim[new_primitive_topology]; 713 else 714 primitive_mode = 0; 715 primitive_topology = new_primitive_topology; 716 } 717 } 718 719 virtual void STDMETHODCALLTYPE IAGetPrimitiveTopology( 720 D3D11_PRIMITIVE_TOPOLOGY *out_primitive_topology) 721 { 722 SYNCHRONIZED; 723 *out_primitive_topology = primitive_topology; 724 } 725 726 virtual void STDMETHODCALLTYPE DrawIndexed( 727 unsigned index_count, 728 unsigned start_index_location, 729 int base_vertex_location) 730 { 731 SYNCHRONIZED; 732 if(update_flags) 733 update_state(); 734 735 pipe_draw_info info; 736 info.mode = primitive_mode; 737 info.indexed = TRUE; 738 info.count = index_count; 739 info.start = start_index_location; 740 info.index_bias = base_vertex_location; 741 info.min_index = 0; 742 info.max_index = ~0; 743 info.start_instance = 0; 744 info.instance_count = 1; 745 info.primitive_restart = FALSE; 746 747 pipe->draw_vbo(pipe, &info); 748 } 749 750 virtual void STDMETHODCALLTYPE Draw( 751 unsigned vertex_count, 752 unsigned start_vertex_location) 753 { 754 SYNCHRONIZED; 755 if(update_flags) 756 update_state(); 757 758 pipe_draw_info info; 759 info.mode = primitive_mode; 760 info.indexed = FALSE; 761 info.count = vertex_count; 762 info.start = start_vertex_location; 763 info.index_bias = 0; 764 info.min_index = 0; 765 info.max_index = ~0; 766 info.start_instance = 0; 767 info.instance_count = 1; 768 info.primitive_restart = TRUE; 769 info.restart_index = strip_cut_index; 770 771 pipe->draw_vbo(pipe, &info); 772 } 773 774 virtual void STDMETHODCALLTYPE DrawIndexedInstanced( 775 unsigned index_countPerInstance, 776 unsigned instance_count, 777 unsigned start_index_location, 778 int base_vertex_location, 779 unsigned start_instance_location) 780 { 781 SYNCHRONIZED; 782 if(update_flags) 783 update_state(); 784 785 pipe_draw_info info; 786 info.mode = primitive_mode; 787 info.indexed = TRUE; 788 info.count = index_countPerInstance; 789 info.start = start_index_location; 790 info.index_bias = base_vertex_location; 791 info.min_index = 0; 792 info.max_index = ~0; 793 info.start_instance = start_instance_location; 794 info.instance_count = instance_count; 795 info.primitive_restart = FALSE; 796 797 pipe->draw_vbo(pipe, &info); 798 } 799 800 virtual void STDMETHODCALLTYPE DrawInstanced( 801 unsigned vertex_countPerInstance, 802 unsigned instance_count, 803 unsigned start_vertex_location, 804 unsigned start_instance_location) 805 { 806 SYNCHRONIZED; 807 if(update_flags) 808 update_state(); 809 810 pipe_draw_info info; 811 info.mode = primitive_mode; 812 info.indexed = FALSE; 813 info.count = vertex_countPerInstance; 814 info.start = start_vertex_location; 815 info.index_bias = 0; 816 info.min_index = 0; 817 info.max_index = ~0; 818 info.start_instance = start_instance_location; 819 info.instance_count = instance_count; 820 info.primitive_restart = TRUE; 821 info.restart_index = strip_cut_index; 822 823 pipe->draw_vbo(pipe, &info); 824 } 825 826 virtual void STDMETHODCALLTYPE DrawAuto(void) 827 { 828 if(!caps.so) 829 return; 830 831 SYNCHRONIZED; 832 if(update_flags) 833 update_state(); 834 835 pipe->draw_stream_output(pipe, primitive_mode); 836 } 837 838 virtual void STDMETHODCALLTYPE DrawIndexedInstancedIndirect( 839 ID3D11Buffer *buffer, 840 unsigned aligned_byte_offset) 841 { 842 SYNCHRONIZED; 843 if(update_flags) 844 update_state(); 845 846 struct { 847 unsigned count; 848 unsigned instance_count; 849 unsigned start; 850 unsigned index_bias; 851 } data; 852 853 pipe_buffer_read(pipe, ((GalliumD3D11Buffer*)buffer)->resource, aligned_byte_offset, sizeof(data), &data); 854 855 pipe_draw_info info; 856 info.mode = primitive_mode; 857 info.indexed = TRUE; 858 info.start = data.start; 859 info.count = data.count; 860 info.index_bias = data.index_bias; 861 info.min_index = 0; 862 info.max_index = ~0; 863 info.start_instance = 0; 864 info.instance_count = data.instance_count; 865 info.primitive_restart = FALSE; 866 867 pipe->draw_vbo(pipe, &info); 868 } 869 870 virtual void STDMETHODCALLTYPE DrawInstancedIndirect( 871 ID3D11Buffer *buffer, 872 unsigned aligned_byte_offset) 873 { 874 SYNCHRONIZED; 875 if(update_flags) 876 update_state(); 877 878 struct { 879 unsigned count; 880 unsigned instance_count; 881 unsigned start; 882 } data; 883 884 pipe_buffer_read(pipe, ((GalliumD3D11Buffer*)buffer)->resource, aligned_byte_offset, sizeof(data), &data); 885 886 pipe_draw_info info; 887 info.mode = primitive_mode; 888 info.indexed = FALSE; 889 info.start = data.start; 890 info.count = data.count; 891 info.index_bias = 0; 892 info.min_index = 0; 893 info.max_index = ~0; 894 info.start_instance = 0; 895 info.instance_count = data.instance_count; 896 info.primitive_restart = TRUE; 897 info.restart_index = strip_cut_index; 898 899 pipe->draw_vbo(pipe, &info); 900 } 901 902#if API >= 11 903 virtual void STDMETHODCALLTYPE Dispatch( 904 unsigned thread_group_count_x, 905 unsigned thread_group_count_y, 906 unsigned thread_group_count_z) 907 { 908// uncomment this when this is implemented 909// SYNCHRONIZED; 910// if(update_flags) 911// update_state(); 912 } 913 914 virtual void STDMETHODCALLTYPE DispatchIndirect( 915 ID3D11Buffer *buffer, 916 unsigned aligned_byte_offset) 917 { 918// uncomment this when this is implemented 919// SYNCHRONIZED; 920// if(update_flags) 921// update_state(); 922 } 923#endif 924 925 void set_clip() 926 { 927 pipe_clip_state clip; 928 clip.nr = 0; 929 clip.depth_clamp = depth_clamp; 930 pipe->set_clip_state(pipe, &clip); 931 } 932 933 virtual void STDMETHODCALLTYPE RSSetState( 934 ID3D11RasterizerState *new_rasterizer_state) 935 { 936 SYNCHRONIZED; 937 if(new_rasterizer_state != rasterizer_state.p) 938 { 939 rasterizer_state = new_rasterizer_state; 940 pipe->bind_rasterizer_state(pipe, new_rasterizer_state ? ((GalliumD3D11RasterizerState*)new_rasterizer_state)->object : default_rasterizer); 941 bool new_depth_clamp = new_rasterizer_state ? ((GalliumD3D11RasterizerState*)new_rasterizer_state)->depth_clamp : false; 942 if(depth_clamp != new_depth_clamp) 943 { 944 depth_clamp = new_depth_clamp; 945 set_clip(); 946 } 947 } 948 } 949 950 virtual void STDMETHODCALLTYPE RSGetState( 951 ID3D11RasterizerState **out_rasterizer_state) 952 { 953 SYNCHRONIZED; 954 *out_rasterizer_state = rasterizer_state.ref(); 955 } 956 957 void set_viewport() 958 { 959 // TODO: is depth correct? it seems D3D10/11 uses a [-1,1]x[-1,1]x[0,1] cube 960 pipe_viewport_state viewport; 961 float half_width = viewports[0].Width * 0.5f; 962 float half_height = viewports[0].Height * 0.5f; 963 964 viewport.scale[0] = half_width; 965 viewport.scale[1] = -half_height; 966 viewport.scale[2] = (viewports[0].MaxDepth - viewports[0].MinDepth); 967 viewport.scale[3] = 1.0f; 968 viewport.translate[0] = half_width + viewports[0].TopLeftX; 969 viewport.translate[1] = half_height + viewports[0].TopLeftY; 970 viewport.translate[2] = viewports[0].MinDepth; 971 viewport.translate[3] = 1.0f; 972 pipe->set_viewport_state(pipe, &viewport); 973 } 974 975 virtual void STDMETHODCALLTYPE RSSetViewports( 976 unsigned count, 977 const D3D11_VIEWPORT *new_viewports) 978 { 979 SYNCHRONIZED; 980 if(count) 981 { 982 if(memcmp(&viewports[0], &new_viewports[0], sizeof(viewports[0]))) 983 { 984 viewports[0] = new_viewports[0]; 985 set_viewport(); 986 } 987 for(unsigned i = 1; i < count; ++i) 988 viewports[i] = new_viewports[i]; 989 } 990 else if(num_viewports) 991 { 992 // TODO: what should we do here? 993 memset(&viewports[0], 0, sizeof(viewports[0])); 994 set_viewport(); 995 } 996 num_viewports = count; 997 } 998 999 virtual void STDMETHODCALLTYPE RSGetViewports( 1000 unsigned *out_count, 1001 D3D11_VIEWPORT *out_viewports) 1002 { 1003 SYNCHRONIZED; 1004 if(out_viewports) 1005 { 1006 unsigned i; 1007 for(i = 0; i < std::min(*out_count, num_viewports); ++i) 1008 out_viewports[i] = viewports[i]; 1009 1010 memset(out_viewports + i, 0, (*out_count - i) * sizeof(D3D11_VIEWPORT)); 1011 } 1012 1013 *out_count = num_viewports; 1014 } 1015 1016 void set_scissor() 1017 { 1018 pipe_scissor_state scissor; 1019 scissor.minx = scissor_rects[0].left; 1020 scissor.miny = scissor_rects[0].top; 1021 scissor.maxx = scissor_rects[0].right; 1022 scissor.maxy = scissor_rects[0].bottom; 1023 pipe->set_scissor_state(pipe, &scissor); 1024 } 1025 1026 virtual void STDMETHODCALLTYPE RSSetScissorRects( 1027 unsigned count, 1028 const D3D11_RECT *new_rects) 1029 { 1030 SYNCHRONIZED; 1031 if(count) 1032 { 1033 if(memcmp(&scissor_rects[0], &new_rects[0], sizeof(scissor_rects[0]))) 1034 { 1035 scissor_rects[0] = new_rects[0]; 1036 set_scissor(); 1037 } 1038 for(unsigned i = 1; i < count; ++i) 1039 scissor_rects[i] = new_rects[i]; 1040 } 1041 else if(num_scissor_rects) 1042 { 1043 // TODO: what should we do here? 1044 memset(&scissor_rects[0], 0, sizeof(scissor_rects[0])); 1045 set_scissor(); 1046 } 1047 1048 num_scissor_rects = count; 1049 } 1050 1051 virtual void STDMETHODCALLTYPE RSGetScissorRects( 1052 unsigned *out_count, 1053 D3D11_RECT *out_rects) 1054 { 1055 SYNCHRONIZED; 1056 if(out_rects) 1057 { 1058 unsigned i; 1059 for(i = 0; i < std::min(*out_count, num_scissor_rects); ++i) 1060 out_rects[i] = scissor_rects[i]; 1061 1062 memset(out_rects + i, 0, (*out_count - i) * sizeof(D3D11_RECT)); 1063 } 1064 1065 *out_count = num_scissor_rects; 1066 } 1067 1068 virtual void STDMETHODCALLTYPE OMSetBlendState( 1069 ID3D11BlendState *new_blend_state, 1070 const float new_blend_factor[4], 1071 unsigned new_sample_mask) 1072 { 1073 SYNCHRONIZED; 1074 float white[4] = {1.0f, 1.0f, 1.0f, 1.0f}; 1075 1076 if(blend_state.p != new_blend_state) 1077 { 1078 pipe->bind_blend_state(pipe, new_blend_state ? ((GalliumD3D11BlendState*)new_blend_state)->object : default_blend); 1079 blend_state = new_blend_state; 1080 } 1081 1082 // Windows D3D11 does this, even though it's apparently undocumented 1083 if(!new_blend_factor) 1084 new_blend_factor = white; 1085 1086 if(memcmp(blend_color, new_blend_factor, sizeof(blend_color))) 1087 { 1088 pipe->set_blend_color(pipe, (struct pipe_blend_color*)new_blend_factor); 1089 memcpy(blend_color, new_blend_factor, sizeof(blend_color)); 1090 } 1091 1092 if(sample_mask != new_sample_mask) 1093 { 1094 pipe->set_sample_mask(pipe, new_sample_mask); 1095 sample_mask = new_sample_mask; 1096 } 1097 } 1098 1099 virtual void STDMETHODCALLTYPE OMGetBlendState( 1100 ID3D11BlendState **out_blend_state, 1101 float out_blend_factor[4], 1102 unsigned *out_sample_mask) 1103 { 1104 SYNCHRONIZED; 1105 if(out_blend_state) 1106 *out_blend_state = blend_state.ref(); 1107 if(out_blend_factor) 1108 memcpy(out_blend_factor, blend_color, sizeof(blend_color)); 1109 if(out_sample_mask) 1110 *out_sample_mask = sample_mask; 1111 } 1112 1113 void set_stencil_ref() 1114 { 1115 struct pipe_stencil_ref sref; 1116 sref.ref_value[0] = stencil_ref; 1117 sref.ref_value[1] = stencil_ref; 1118 pipe->set_stencil_ref(pipe, &sref); 1119 } 1120 1121 virtual void STDMETHODCALLTYPE OMSetDepthStencilState( 1122 ID3D11DepthStencilState *new_depth_stencil_state, 1123 unsigned new_stencil_ref) 1124 { 1125 SYNCHRONIZED; 1126 if(new_depth_stencil_state != depth_stencil_state.p) 1127 { 1128 pipe->bind_depth_stencil_alpha_state(pipe, new_depth_stencil_state ? ((GalliumD3D11DepthStencilState*)new_depth_stencil_state)->object : default_depth_stencil); 1129 depth_stencil_state = new_depth_stencil_state; 1130 } 1131 1132 if(new_stencil_ref != stencil_ref) 1133 { 1134 stencil_ref = new_stencil_ref; 1135 set_stencil_ref(); 1136 } 1137 } 1138 1139 virtual void STDMETHODCALLTYPE OMGetDepthStencilState( 1140 ID3D11DepthStencilState **out_depth_stencil_state, 1141 unsigned *out_stencil_ref) 1142 { 1143 SYNCHRONIZED; 1144 if(*out_depth_stencil_state) 1145 *out_depth_stencil_state = depth_stencil_state.ref(); 1146 if(out_stencil_ref) 1147 *out_stencil_ref = stencil_ref; 1148 } 1149 1150 void set_framebuffer() 1151 { 1152 struct pipe_framebuffer_state fb; 1153 memset(&fb, 0, sizeof(fb)); 1154 if(depth_stencil_view) 1155 { 1156 struct pipe_surface* surf = ((GalliumD3D11DepthStencilView*)depth_stencil_view.p)->object; 1157 fb.zsbuf = surf; 1158 if(surf->width > fb.width) 1159 fb.width = surf->width; 1160 if(surf->height > fb.height) 1161 fb.height = surf->height; 1162 } 1163 fb.nr_cbufs = num_render_target_views; 1164 unsigned i; 1165 for(i = 0; i < num_render_target_views; ++i) 1166 { 1167 if(render_target_views[i]) 1168 { 1169 struct pipe_surface* surf = ((GalliumD3D11RenderTargetView*)render_target_views[i].p)->object; 1170 fb.cbufs[i] = surf; 1171 if(surf->width > fb.width) 1172 fb.width = surf->width; 1173 if(surf->height > fb.height) 1174 fb.height = surf->height; 1175 } 1176 } 1177 1178 pipe->set_framebuffer_state(pipe, &fb); 1179 } 1180 1181 /* TODO: the docs say that we should unbind conflicting resources (e.g. those bound for read while we are binding them for write too), but we aren't. 1182 * Hopefully nobody relies on this happening 1183 */ 1184 1185 virtual void STDMETHODCALLTYPE OMSetRenderTargets( 1186 unsigned count, 1187 ID3D11RenderTargetView *const *new_render_target_views, 1188 ID3D11DepthStencilView *new_depth_stencil_view) 1189 { 1190 SYNCHRONIZED; 1191 if(!new_render_target_views) 1192 count = 0; 1193 if(count == num_render_target_views) 1194 { 1195 for(unsigned i = 0; i < count; ++i) 1196 { 1197 if(new_render_target_views[i] != render_target_views[i].p) 1198 goto changed; 1199 } 1200 return; 1201 } 1202changed: 1203 depth_stencil_view = new_depth_stencil_view; 1204 unsigned i; 1205 for(i = 0; i < count; ++i) 1206 { 1207 render_target_views[i] = new_render_target_views[i]; 1208#if API >= 11 1209 om_unordered_access_views[i] = (ID3D11UnorderedAccessView*)NULL; 1210#endif 1211 } 1212 for(; i < num_render_target_views; ++i) 1213 render_target_views[i] = (ID3D11RenderTargetView*)NULL; 1214 num_render_target_views = count; 1215 set_framebuffer(); 1216 } 1217 1218 virtual void STDMETHODCALLTYPE OMGetRenderTargets( 1219 unsigned count, 1220 ID3D11RenderTargetView **out_render_target_views, 1221 ID3D11DepthStencilView **out_depth_stencil_view) 1222 { 1223 SYNCHRONIZED; 1224 if(out_render_target_views) 1225 { 1226 unsigned i; 1227 for(i = 0; i < std::min(num_render_target_views, count); ++i) 1228 out_render_target_views[i] = render_target_views[i].ref(); 1229 1230 for(; i < count; ++i) 1231 out_render_target_views[i] = 0; 1232 } 1233 1234 if(out_depth_stencil_view) 1235 *out_depth_stencil_view = depth_stencil_view.ref(); 1236 } 1237 1238#if API >= 11 1239 /* TODO: what is this supposed to do _exactly_? are we doing the right thing? */ 1240 virtual void STDMETHODCALLTYPE OMSetRenderTargetsAndUnorderedAccessViews( 1241 unsigned rtv_count, 1242 ID3D11RenderTargetView *const *new_render_target_views, 1243 ID3D11DepthStencilView *new_depth_stencil_view, 1244 unsigned uav_start, 1245 unsigned uav_count, 1246 ID3D11UnorderedAccessView *const *new_unordered_access_views, 1247 const unsigned *new_uav_initial_counts) 1248 { 1249 SYNCHRONIZED; 1250 if(rtv_count != D3D11_KEEP_RENDER_TARGETS_AND_DEPTH_STENCIL) 1251 OMSetRenderTargets(rtv_count, new_render_target_views, new_depth_stencil_view); 1252 1253 if(uav_count != D3D11_KEEP_UNORDERED_ACCESS_VIEWS) 1254 { 1255 for(unsigned i = 0; i < uav_count; ++i) 1256 { 1257 om_unordered_access_views[uav_start + i] = new_unordered_access_views[i]; 1258 render_target_views[uav_start + i] = (ID3D11RenderTargetView*)0; 1259 } 1260 } 1261 } 1262 1263 virtual void STDMETHODCALLTYPE OMGetRenderTargetsAndUnorderedAccessViews( 1264 unsigned rtv_count, 1265 ID3D11RenderTargetView **out_render_target_views, 1266 ID3D11DepthStencilView **out_depth_stencil_view, 1267 unsigned uav_start, 1268 unsigned uav_count, 1269 ID3D11UnorderedAccessView **out_unordered_access_views) 1270 { 1271 SYNCHRONIZED; 1272 if(out_render_target_views) 1273 OMGetRenderTargets(rtv_count, out_render_target_views, out_depth_stencil_view); 1274 1275 if(out_unordered_access_views) 1276 { 1277 for(unsigned i = 0; i < uav_count; ++i) 1278 out_unordered_access_views[i] = om_unordered_access_views[uav_start + i].ref(); 1279 } 1280 } 1281#endif 1282 1283 virtual void STDMETHODCALLTYPE SOSetTargets( 1284 unsigned count, 1285 ID3D11Buffer *const *new_so_targets, 1286 const unsigned *new_offsets) 1287 { 1288 SYNCHRONIZED; 1289 unsigned i; 1290 if(!new_so_targets) 1291 count = 0; 1292 bool changed = false; 1293 for(i = 0; i < count; ++i) 1294 { 1295 ID3D11Buffer* buffer = new_so_targets[i]; 1296 if(buffer != so_targets[i].p || new_offsets[i] != so_offsets[i]) 1297 { 1298 so_buffers[i] = buffer ? ((GalliumD3D11Buffer*)buffer)->resource : 0; 1299 so_targets[i] = buffer; 1300 so_offsets[i] = new_offsets[i]; 1301 changed = true; 1302 } 1303 } 1304 for(; i < D3D11_SO_BUFFER_SLOT_COUNT; ++i) 1305 { 1306 if(so_targets[i].p || so_offsets[i]) 1307 { 1308 changed = true; 1309 so_targets[i] = (ID3D11Buffer*)0; 1310 so_offsets[i] = 0; 1311 } 1312 } 1313 num_so_targets = count; 1314 1315 if(changed && caps.so) 1316 pipe->set_stream_output_buffers(pipe, so_buffers, (int*)so_offsets, num_so_targets); 1317 } 1318 1319 virtual void STDMETHODCALLTYPE SOGetTargets( 1320 unsigned count, 1321 ID3D11Buffer **out_so_targets 1322#if API < 11 1323 , UINT *out_offsets 1324#endif 1325 ) 1326 { 1327 SYNCHRONIZED; 1328 for(unsigned i = 0; i < count; ++i) 1329 { 1330 out_so_targets[i] = so_targets[i].ref(); 1331#if API < 11 1332 out_offsets[i] = so_offsets[i]; 1333#endif 1334 } 1335 } 1336 1337 virtual void STDMETHODCALLTYPE Begin( 1338 ID3D11Asynchronous *async) 1339 { 1340 SYNCHRONIZED; 1341 if(caps.queries) 1342 pipe->begin_query(pipe, ((GalliumD3D11Asynchronous<>*)async)->query); 1343 } 1344 1345 virtual void STDMETHODCALLTYPE End( 1346 ID3D11Asynchronous *async) 1347 { 1348 SYNCHRONIZED; 1349 if(caps.queries) 1350 pipe->end_query(pipe, ((GalliumD3D11Asynchronous<>*)async)->query); 1351 } 1352 1353 virtual HRESULT STDMETHODCALLTYPE GetData( 1354 ID3D11Asynchronous *iasync, 1355 void *out_data, 1356 unsigned data_size, 1357 unsigned get_data_flags) 1358 { 1359 SYNCHRONIZED; 1360 if(!caps.queries) 1361 return E_NOTIMPL; 1362 1363 GalliumD3D11Asynchronous<>* async = (GalliumD3D11Asynchronous<>*)iasync; 1364 void* tmp_data = alloca(async->data_size); 1365 boolean ret = pipe->get_query_result(pipe, async->query, !(get_data_flags & D3D11_ASYNC_GETDATA_DONOTFLUSH), tmp_data); 1366 if(out_data) 1367 memcpy(out_data, tmp_data, std::min(async->data_size, data_size)); 1368 return ret ? S_OK : S_FALSE; 1369 } 1370 1371 void set_render_condition() 1372 { 1373 if(caps.render_condition) 1374 { 1375 if(!render_predicate) 1376 pipe->render_condition(pipe, 0, 0); 1377 else 1378 { 1379 GalliumD3D11Predicate* predicate = (GalliumD3D11Predicate*)render_predicate.p; 1380 if(!render_predicate_value && predicate->desc.Query == D3D11_QUERY_OCCLUSION_PREDICATE) 1381 { 1382 unsigned mode = (predicate->desc.MiscFlags & D3D11_QUERY_MISC_PREDICATEHINT) ? PIPE_RENDER_COND_NO_WAIT : PIPE_RENDER_COND_WAIT; 1383 pipe->render_condition(pipe, predicate->query, mode); 1384 } 1385 else 1386 { 1387 /* TODO: add inverted predication to Gallium*/ 1388 pipe->render_condition(pipe, 0, 0); 1389 } 1390 } 1391 } 1392 } 1393 1394 virtual void STDMETHODCALLTYPE SetPredication( 1395 ID3D11Predicate *new_predicate, 1396 BOOL new_predicate_value) 1397 { 1398 SYNCHRONIZED; 1399 if(render_predicate.p != new_predicate || render_predicate_value != new_predicate_value) 1400 { 1401 render_predicate = new_predicate; 1402 render_predicate_value = new_predicate_value; 1403 set_render_condition(); 1404 } 1405 } 1406 1407 virtual void STDMETHODCALLTYPE GetPredication( 1408 ID3D11Predicate **out_predicate, 1409 BOOL *out_predicate_value) 1410 { 1411 SYNCHRONIZED; 1412 if(out_predicate) 1413 *out_predicate = render_predicate.ref(); 1414 if(out_predicate_value) 1415 *out_predicate_value = render_predicate_value; 1416 } 1417 1418 static unsigned d3d11_subresource_to_level(struct pipe_resource* resource, unsigned subresource) 1419 { 1420 if(subresource <= resource->last_level) 1421 { 1422 return subresource; 1423 } 1424 else 1425 { 1426 unsigned levels = resource->last_level + 1; 1427 return subresource % levels; 1428 } 1429 } 1430 1431 static unsigned d3d11_subresource_to_layer(struct pipe_resource* resource, unsigned subresource) 1432 { 1433 if(subresource <= resource->last_level) 1434 { 1435 return 0; 1436 } 1437 else 1438 { 1439 unsigned levels = resource->last_level + 1; 1440 return subresource / levels; 1441 } 1442 } 1443 1444 1445 /* TODO: deferred contexts will need a different implementation of this, 1446 * because we can't put the transfer info into the resource itself. 1447 * Also, there are very different restrictions, for obvious reasons. 1448 */ 1449 virtual HRESULT STDMETHODCALLTYPE Map( 1450 ID3D11Resource *iresource, 1451 unsigned subresource, 1452 D3D11_MAP map_type, 1453 unsigned map_flags, 1454 D3D11_MAPPED_SUBRESOURCE *mapped_resource) 1455 { 1456 SYNCHRONIZED; 1457 GalliumD3D11Resource<>* resource = (GalliumD3D11Resource<>*)iresource; 1458 if(resource->transfers.count(subresource)) 1459 return E_FAIL; 1460 unsigned level = d3d11_subresource_to_level(resource->resource, subresource); 1461 unsigned layer = d3d11_subresource_to_layer(resource->resource, subresource); 1462 pipe_box box = d3d11_to_pipe_box(resource->resource, level, 0); 1463 box.z += layer; 1464 unsigned usage = 0; 1465 if(map_type == D3D11_MAP_READ) 1466 usage = PIPE_TRANSFER_READ; 1467 else if(map_type == D3D11_MAP_WRITE) 1468 usage = PIPE_TRANSFER_WRITE; 1469 else if(map_type == D3D11_MAP_READ_WRITE) 1470 usage = PIPE_TRANSFER_READ_WRITE; 1471 else if(map_type == D3D11_MAP_WRITE_DISCARD) 1472 usage = PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD; 1473 else if(map_type == D3D11_MAP_WRITE_NO_OVERWRITE) 1474 usage = PIPE_TRANSFER_WRITE | PIPE_TRANSFER_NOOVERWRITE; 1475 else 1476 return E_INVALIDARG; 1477 if(map_type & D3D10_MAP_FLAG_DO_NOT_WAIT) 1478 usage |= PIPE_TRANSFER_DONTBLOCK; 1479 struct pipe_transfer* transfer = pipe->get_transfer(pipe, resource->resource, level, usage, &box); 1480 if(!transfer) { 1481 if(map_type & D3D10_MAP_FLAG_DO_NOT_WAIT) 1482 return DXGI_ERROR_WAS_STILL_DRAWING; 1483 else 1484 return E_FAIL; 1485 } 1486 resource->transfers[subresource] = transfer; 1487 mapped_resource->pData = pipe->transfer_map(pipe, transfer); 1488 mapped_resource->RowPitch = transfer->stride; 1489 mapped_resource->DepthPitch = transfer->layer_stride; 1490 return S_OK; 1491 } 1492 1493 virtual void STDMETHODCALLTYPE Unmap( 1494 ID3D11Resource *iresource, 1495 unsigned subresource) 1496 { 1497 SYNCHRONIZED; 1498 GalliumD3D11Resource<>* resource = (GalliumD3D11Resource<>*)iresource; 1499 std::unordered_map<unsigned, pipe_transfer*>::iterator i = resource->transfers.find(subresource); 1500 if(i != resource->transfers.end()) 1501 { 1502 pipe->transfer_unmap(pipe, i->second); 1503 pipe->transfer_destroy(pipe, i->second); 1504 resource->transfers.erase(i); 1505 } 1506 } 1507 1508 virtual void STDMETHODCALLTYPE CopySubresourceRegion( 1509 ID3D11Resource *dst_resource, 1510 unsigned dst_subresource, 1511 unsigned dst_x, 1512 unsigned dst_y, 1513 unsigned dst_z, 1514 ID3D11Resource *src_resource, 1515 unsigned src_subresource, 1516 const D3D11_BOX *src_box) 1517 { 1518 SYNCHRONIZED; 1519 GalliumD3D11Resource<>* dst = (GalliumD3D11Resource<>*)dst_resource; 1520 GalliumD3D11Resource<>* src = (GalliumD3D11Resource<>*)src_resource; 1521 unsigned dst_level = d3d11_subresource_to_level(dst->resource, dst_subresource); 1522 unsigned dst_layer = d3d11_subresource_to_layer(dst->resource, dst_subresource); 1523 unsigned src_level = d3d11_subresource_to_level(src->resource, src_subresource); 1524 unsigned src_layer = d3d11_subresource_to_layer(src->resource, src_subresource); 1525 pipe_box box = d3d11_to_pipe_box(src->resource, src_level, src_box); 1526 dst_z += dst_layer; 1527 box.z += src_layer; 1528 { 1529 pipe->resource_copy_region(pipe, 1530 dst->resource, dst_level, dst_x, dst_y, dst_z, 1531 src->resource, src_level, &box); 1532 } 1533 } 1534 1535 virtual void STDMETHODCALLTYPE CopyResource( 1536 ID3D11Resource *dst_resource, 1537 ID3D11Resource *src_resource) 1538 { 1539 SYNCHRONIZED; 1540 GalliumD3D11Resource<>* dst = (GalliumD3D11Resource<>*)dst_resource; 1541 GalliumD3D11Resource<>* src = (GalliumD3D11Resource<>*)src_resource; 1542 unsigned level; 1543 for(level = 0; level <= dst->resource->last_level; ++level) 1544 { 1545 pipe_box box; 1546 box.x = box.y = box.z = 0; 1547 box.width = u_minify(dst->resource->width0, level); 1548 box.height = u_minify(dst->resource->height0, level); 1549 if(dst->resource->target == PIPE_TEXTURE_3D) 1550 box.depth = u_minify(dst->resource->depth0, level); 1551 else 1552 box.depth = dst->resource->array_size; 1553 pipe->resource_copy_region(pipe, 1554 dst->resource, level, 0, 0, 0, 1555 src->resource, level, &box); 1556 } 1557 } 1558 1559 virtual void STDMETHODCALLTYPE UpdateSubresource( 1560 ID3D11Resource *dst_resource, 1561 unsigned dst_subresource, 1562 const D3D11_BOX *pDstBox, 1563 const void *pSrcData, 1564 unsigned src_row_pitch, 1565 unsigned src_depth_pitch) 1566 { 1567 SYNCHRONIZED; 1568 GalliumD3D11Resource<>* dst = (GalliumD3D11Resource<>*)dst_resource; 1569 unsigned dst_level = d3d11_subresource_to_level(dst->resource, dst_subresource); 1570 unsigned dst_layer = d3d11_subresource_to_layer(dst->resource, dst_subresource); 1571 pipe_box box = d3d11_to_pipe_box(dst->resource, dst_level, pDstBox); 1572 box.z += dst_layer; 1573 pipe->transfer_inline_write(pipe, dst->resource, dst_level, PIPE_TRANSFER_WRITE, &box, pSrcData, src_row_pitch, src_depth_pitch); 1574 } 1575 1576#if API >= 11 1577 virtual void STDMETHODCALLTYPE CopyStructureCount( 1578 ID3D11Buffer *dst_buffer, 1579 unsigned dst_aligned_byte_offset, 1580 ID3D11UnorderedAccessView *src_view) 1581 { 1582 SYNCHRONIZED; 1583 } 1584#endif 1585 1586 virtual void STDMETHODCALLTYPE ClearRenderTargetView( 1587 ID3D11RenderTargetView *render_target_view, 1588 const float color[4]) 1589 { 1590 SYNCHRONIZED; 1591 GalliumD3D11RenderTargetView* view = ((GalliumD3D11RenderTargetView*)render_target_view); 1592 union pipe_color_union cc; 1593 cc.f[0] = color[0]; 1594 cc.f[1] = color[1]; 1595 cc.f[2] = color[2]; 1596 cc.f[3] = color[3]; 1597 pipe->clear_render_target(pipe, view->object, &cc, 0, 0, view->object->width, view->object->height); 1598 } 1599 1600 virtual void STDMETHODCALLTYPE ClearDepthStencilView( 1601 ID3D11DepthStencilView *depth_stencil_view, 1602 unsigned clear_flags, 1603 float depth, 1604 UINT8 stencil) 1605 { 1606 SYNCHRONIZED; 1607 GalliumD3D11DepthStencilView* view = ((GalliumD3D11DepthStencilView*)depth_stencil_view); 1608 unsigned flags = 0; 1609 if(clear_flags & D3D11_CLEAR_DEPTH) 1610 flags |= PIPE_CLEAR_DEPTH; 1611 if(clear_flags & D3D11_CLEAR_STENCIL) 1612 flags |= PIPE_CLEAR_STENCIL; 1613 pipe->clear_depth_stencil(pipe, view->object, flags, depth, stencil, 0, 0, view->object->width, view->object->height); 1614 } 1615 1616#if API >= 11 1617 virtual void STDMETHODCALLTYPE ClearUnorderedAccessViewUint( 1618 ID3D11UnorderedAccessView *unordered_access_view, 1619 const unsigned values[4]) 1620 { 1621 SYNCHRONIZED; 1622 } 1623 1624 virtual void STDMETHODCALLTYPE ClearUnorderedAccessViewFloat( 1625 ID3D11UnorderedAccessView *unordered_access_view, 1626 const float values[4]) 1627 { 1628 SYNCHRONIZED; 1629 } 1630#endif 1631 1632 void restore_gallium_state_blit_only() 1633 { 1634 pipe->bind_blend_state(pipe, blend_state.p ? blend_state.p->object : default_blend); 1635 pipe->bind_depth_stencil_alpha_state(pipe, depth_stencil_state.p ? depth_stencil_state.p->object : default_depth_stencil); 1636 pipe->bind_rasterizer_state(pipe, rasterizer_state.p ? rasterizer_state.p->object : default_rasterizer); 1637 pipe->bind_vertex_elements_state(pipe, input_layout.p ? input_layout.p->object : default_input_layout); 1638 pipe->bind_fs_state(pipe, shaders[D3D11_STAGE_PS].p ? shaders[D3D11_STAGE_PS].p->object : default_shaders[PIPE_SHADER_FRAGMENT]); 1639 pipe->bind_vs_state(pipe, shaders[D3D11_STAGE_VS].p ? shaders[D3D11_STAGE_VS].p->object : default_shaders[PIPE_SHADER_VERTEX]); 1640 if(caps.gs) 1641 pipe->bind_gs_state(pipe, shaders[D3D11_STAGE_GS].p ? shaders[D3D11_STAGE_GS].p->object : default_shaders[PIPE_SHADER_GEOMETRY]); 1642 set_framebuffer(); 1643 set_viewport(); 1644 set_clip(); 1645 set_render_condition(); 1646 // TODO: restore stream output 1647 1648 update_flags |= UPDATE_VERTEX_BUFFERS | (1 << (UPDATE_SAMPLERS_SHIFT + D3D11_STAGE_PS)) | (1 << (UPDATE_VIEWS_SHIFT + D3D11_STAGE_PS)); 1649 } 1650 1651 virtual void STDMETHODCALLTYPE RestoreGalliumStateBlitOnly() 1652 { 1653 SYNCHRONIZED; 1654 restore_gallium_state_blit_only(); 1655 } 1656 1657 virtual void STDMETHODCALLTYPE GenerateMips( 1658 ID3D11ShaderResourceView *shader_resource_view) 1659 { 1660 SYNCHRONIZED; 1661 1662 GalliumD3D11ShaderResourceView* view = (GalliumD3D11ShaderResourceView*)shader_resource_view; 1663 if(caps.gs) 1664 pipe->bind_gs_state(pipe, 0); 1665 if(caps.so) 1666 pipe->bind_stream_output_state(pipe, 0); 1667 if(pipe->render_condition) 1668 pipe->render_condition(pipe, 0, 0); 1669 for(unsigned layer = view->object->u.tex.first_layer; layer <= view->object->u.tex.last_layer; ++layer) 1670 util_gen_mipmap(gen_mipmap, view->object, layer, view->object->u.tex.first_level, view->object->u.tex.last_level, PIPE_TEX_FILTER_LINEAR); 1671 restore_gallium_state_blit_only(); 1672 } 1673 1674 virtual void STDMETHODCALLTYPE RestoreGalliumState() 1675 { 1676 SYNCHRONIZED; 1677 restore_gallium_state_blit_only(); 1678 1679 set_index_buffer(); 1680 set_stencil_ref(); 1681 pipe->set_blend_color(pipe, (struct pipe_blend_color*)blend_color); 1682 pipe->set_sample_mask(pipe, sample_mask); 1683 1684 for(unsigned s = 0; s < 3; ++s) 1685 { 1686 unsigned num = std::min(caps.constant_buffers[s], (unsigned)D3D11_COMMONSHADER_CONSTANT_BUFFER_API_SLOT_COUNT); 1687 for(unsigned i = 0; i < num; ++i) 1688 pipe->set_constant_buffer(pipe, s, i, constant_buffers[s][i].p ? constant_buffers[s][i].p->resource : 0); 1689 } 1690 1691 if(caps.so) 1692 pipe->set_stream_output_buffers(pipe, so_buffers, (int*)so_offsets, num_so_targets); 1693 1694 update_flags |= (1 << (UPDATE_SAMPLERS_SHIFT + D3D11_STAGE_VS)) | (1 << (UPDATE_VIEWS_SHIFT + D3D11_STAGE_VS)); 1695 update_flags |= (1 << (UPDATE_SAMPLERS_SHIFT + D3D11_STAGE_GS)) | (1 << (UPDATE_VIEWS_SHIFT + D3D11_STAGE_GS)); 1696 1697 set_scissor(); 1698 } 1699 1700#if API >= 11 1701 /* TODO: hack SRVs or sampler states to handle this, or add to Gallium */ 1702 virtual void STDMETHODCALLTYPE SetResourceMinLOD( 1703 ID3D11Resource *iresource, 1704 float min_lod) 1705 { 1706 SYNCHRONIZED; 1707 GalliumD3D11Resource<>* resource = (GalliumD3D11Resource<>*)iresource; 1708 if(resource->min_lod != min_lod) 1709 { 1710 // TODO: actually do anything? 1711 resource->min_lod = min_lod; 1712 } 1713 } 1714 1715 virtual float STDMETHODCALLTYPE GetResourceMinLOD( 1716 ID3D11Resource *iresource) 1717 { 1718 SYNCHRONIZED; 1719 GalliumD3D11Resource<>* resource = (GalliumD3D11Resource<>*)iresource; 1720 return resource->min_lod; 1721 } 1722#endif 1723 1724 virtual void STDMETHODCALLTYPE ResolveSubresource( 1725 ID3D11Resource *dst_resource, 1726 unsigned dst_subresource, 1727 ID3D11Resource *src_resource, 1728 unsigned src_subresource, 1729 DXGI_FORMAT format) 1730 { 1731 SYNCHRONIZED; 1732 GalliumD3D11Resource<>* dst = (GalliumD3D11Resource<>*)dst_resource; 1733 GalliumD3D11Resource<>* src = (GalliumD3D11Resource<>*)src_resource; 1734 struct pipe_resolve_info info; 1735 1736 info.dst.res = dst->resource; 1737 info.src.res = src->resource; 1738 info.dst.level = 0; 1739 info.dst.layer = d3d11_subresource_to_layer(dst->resource, dst_subresource); 1740 info.src.layer = d3d11_subresource_to_layer(src->resource, src_subresource); 1741 1742 info.src.x0 = 0; 1743 info.src.x1 = info.src.res->width0; 1744 info.src.y0 = 0; 1745 info.src.y1 = info.src.res->height0; 1746 info.dst.x0 = 0; 1747 info.dst.x1 = info.dst.res->width0; 1748 info.dst.y0 = 0; 1749 info.dst.y1 = info.dst.res->height0; 1750 1751 info.mask = PIPE_MASK_RGBA | PIPE_MASK_ZS; 1752 1753 pipe->resource_resolve(pipe, &info); 1754 } 1755 1756#if API >= 11 1757 virtual void STDMETHODCALLTYPE ExecuteCommandList( 1758 ID3D11CommandList *command_list, 1759 BOOL restore_context_state) 1760 { 1761 SYNCHRONIZED; 1762 } 1763 1764 virtual HRESULT STDMETHODCALLTYPE FinishCommandList( 1765 BOOL restore_deferred_context_state, 1766 ID3D11CommandList **out_command_list) 1767 { 1768 SYNCHRONIZED; 1769 return E_NOTIMPL; 1770 } 1771#endif 1772 1773 virtual void STDMETHODCALLTYPE ClearState(void) 1774 { 1775 /* we don't take a lock here because we would deadlock otherwise 1776 * TODO: this is probably incorrect, because ClearState should likely be atomic. 1777 * However, I can't think of any correct usage that would be affected by this 1778 * being non-atomic, and making this atomic is quite expensive and complicates 1779 * the code 1780 */ 1781 1782 // we qualify all calls so that we avoid virtual dispatch and might get them inlined 1783 // TODO: make sure all this gets inlined, which might require more compiler flags 1784 // TODO: optimize this 1785#if API >= 11 1786 GalliumD3D11DeviceContext::PSSetShader(0, 0, 0); 1787 GalliumD3D11DeviceContext::GSSetShader(0, 0, 0); 1788 GalliumD3D11DeviceContext::VSSetShader(0, 0, 0); 1789 GalliumD3D11DeviceContext::HSSetShader(0, 0, 0); 1790 GalliumD3D11DeviceContext::DSSetShader(0, 0, 0); 1791 GalliumD3D11DeviceContext::CSSetShader(0, 0, 0); 1792#else 1793 GalliumD3D11DeviceContext::PSSetShader(0); 1794 GalliumD3D11DeviceContext::GSSetShader(0); 1795 GalliumD3D11DeviceContext::VSSetShader(0); 1796#endif 1797 1798 GalliumD3D11DeviceContext::IASetInputLayout(0); 1799 GalliumD3D11DeviceContext::IASetIndexBuffer(0, DXGI_FORMAT_UNKNOWN, 0); 1800 GalliumD3D11DeviceContext::RSSetState(0); 1801 GalliumD3D11DeviceContext::OMSetDepthStencilState(0, 0); 1802 GalliumD3D11DeviceContext::OMSetBlendState(0, (float*)zero_data, ~0); 1803 GalliumD3D11DeviceContext::SetPredication(0, 0); 1804 GalliumD3D11DeviceContext::IASetPrimitiveTopology(D3D_PRIMITIVE_TOPOLOGY_UNDEFINED); 1805 1806 GalliumD3D11DeviceContext::PSSetConstantBuffers(0, D3D11_COMMONSHADER_CONSTANT_BUFFER_API_SLOT_COUNT, (ID3D11Buffer**)zero_data); 1807 GalliumD3D11DeviceContext::GSSetConstantBuffers(0, D3D11_COMMONSHADER_CONSTANT_BUFFER_API_SLOT_COUNT, (ID3D11Buffer**)zero_data); 1808 GalliumD3D11DeviceContext::VSSetConstantBuffers(0, D3D11_COMMONSHADER_CONSTANT_BUFFER_API_SLOT_COUNT, (ID3D11Buffer**)zero_data); 1809#if API >= 11 1810 GalliumD3D11DeviceContext::HSSetConstantBuffers(0, D3D11_COMMONSHADER_CONSTANT_BUFFER_API_SLOT_COUNT, (ID3D11Buffer**)zero_data); 1811 GalliumD3D11DeviceContext::DSSetConstantBuffers(0, D3D11_COMMONSHADER_CONSTANT_BUFFER_API_SLOT_COUNT, (ID3D11Buffer**)zero_data); 1812 GalliumD3D11DeviceContext::CSSetConstantBuffers(0, D3D11_COMMONSHADER_CONSTANT_BUFFER_API_SLOT_COUNT, (ID3D11Buffer**)zero_data); 1813#endif 1814 1815 GalliumD3D11DeviceContext::IASetVertexBuffers(0, num_vertex_buffers, (ID3D11Buffer**)zero_data, (unsigned*)zero_data, (unsigned*)zero_data); 1816#if API >= 11 1817 GalliumD3D11DeviceContext::OMSetRenderTargetsAndUnorderedAccessViews(0, 0, 0 , 0, 0, 0, 0); 1818#else 1819 GalliumD3D11DeviceContext::OMSetRenderTargets(0, 0, 0 ); 1820#endif 1821 GalliumD3D11DeviceContext::SOSetTargets(0, 0, 0); 1822 1823 GalliumD3D11DeviceContext::PSSetShaderResources(0, num_shader_resource_views[D3D11_STAGE_PS], (ID3D11ShaderResourceView**)zero_data); 1824 GalliumD3D11DeviceContext::GSSetShaderResources(0, num_shader_resource_views[D3D11_STAGE_GS], (ID3D11ShaderResourceView**)zero_data); 1825 GalliumD3D11DeviceContext::VSSetShaderResources(0, num_shader_resource_views[D3D11_STAGE_VS], (ID3D11ShaderResourceView**)zero_data); 1826#if API >= 11 1827 GalliumD3D11DeviceContext::HSSetShaderResources(0, num_shader_resource_views[D3D11_STAGE_HS], (ID3D11ShaderResourceView**)zero_data); 1828 GalliumD3D11DeviceContext::DSSetShaderResources(0, num_shader_resource_views[D3D11_STAGE_DS], (ID3D11ShaderResourceView**)zero_data); 1829 GalliumD3D11DeviceContext::CSSetShaderResources(0, num_shader_resource_views[D3D11_STAGE_CS], (ID3D11ShaderResourceView**)zero_data); 1830#endif 1831 1832 GalliumD3D11DeviceContext::PSSetSamplers(0, num_shader_resource_views[D3D11_STAGE_PS], (ID3D11SamplerState**)zero_data); 1833 GalliumD3D11DeviceContext::GSSetSamplers(0, num_shader_resource_views[D3D11_STAGE_GS], (ID3D11SamplerState**)zero_data); 1834 GalliumD3D11DeviceContext::VSSetSamplers(0, num_shader_resource_views[D3D11_STAGE_VS], (ID3D11SamplerState**)zero_data); 1835#if API >= 11 1836 GalliumD3D11DeviceContext::HSSetSamplers(0, num_shader_resource_views[D3D11_STAGE_HS], (ID3D11SamplerState**)zero_data); 1837 GalliumD3D11DeviceContext::DSSetSamplers(0, num_shader_resource_views[D3D11_STAGE_DS], (ID3D11SamplerState**)zero_data); 1838 GalliumD3D11DeviceContext::CSSetSamplers(0, num_shader_resource_views[D3D11_STAGE_CS], (ID3D11SamplerState**)zero_data); 1839#endif 1840 1841 GalliumD3D11DeviceContext::RSSetViewports(0, 0); 1842 GalliumD3D11DeviceContext::RSSetScissorRects(0, 0); 1843 } 1844 1845 virtual void STDMETHODCALLTYPE Flush(void) 1846 { 1847 SYNCHRONIZED; 1848 pipe->flush(pipe, 0); 1849 } 1850 1851 /* In Direct3D 10, if the reference count of an object drops to 0, it is automatically 1852 * cleanly unbound from the pipeline. 1853 * In Direct3D 11, the pipeline holds a reference. 1854 * 1855 * Note that instead of always scanning the pipeline on destruction, we could 1856 * maintain the internal reference count on DirectX 10 and use it to check if an 1857 * object is still bound. 1858 * Presumably, on average, scanning is faster if the application is well written. 1859 */ 1860#if API < 11 1861#define IMPLEMENT_SIMPLE_UNBIND(name, member, gallium, def) \ 1862 void Unbind##name(ID3D11##name* state) \ 1863 { \ 1864 SYNCHRONIZED; \ 1865 if((void*)state == (void*)member.p) \ 1866 { \ 1867 member.p = 0; \ 1868 pipe->bind_##gallium##_state(pipe, default_##def); \ 1869 } \ 1870 } 1871 IMPLEMENT_SIMPLE_UNBIND(BlendState, blend_state, blend, blend) 1872 IMPLEMENT_SIMPLE_UNBIND(RasterizerState, rasterizer_state, rasterizer, rasterizer) 1873 IMPLEMENT_SIMPLE_UNBIND(DepthStencilState, depth_stencil_state, depth_stencil_alpha, depth_stencil) 1874 IMPLEMENT_SIMPLE_UNBIND(InputLayout, input_layout, vertex_elements, input_layout) 1875 IMPLEMENT_SIMPLE_UNBIND(PixelShader, shaders[D3D11_STAGE_PS], fs, shaders[D3D11_STAGE_PS]) 1876 IMPLEMENT_SIMPLE_UNBIND(VertexShader, shaders[D3D11_STAGE_VS], vs, shaders[D3D11_STAGE_VS]) 1877 IMPLEMENT_SIMPLE_UNBIND(GeometryShader, shaders[D3D11_STAGE_GS], gs, shaders[D3D11_STAGE_GS]) 1878 1879 void UnbindPredicate(ID3D11Predicate* predicate) 1880 { 1881 SYNCHRONIZED; 1882 if(predicate == render_predicate) 1883 { 1884 render_predicate.p = NULL; 1885 render_predicate_value = 0; 1886 pipe->render_condition(pipe, 0, 0); 1887 } 1888 } 1889 1890 void UnbindSamplerState(ID3D11SamplerState* state) 1891 { 1892 SYNCHRONIZED; 1893 for(unsigned s = 0; s < D3D11_STAGES; ++s) 1894 { 1895 for(unsigned i = 0; i < num_samplers[s]; ++i) 1896 { 1897 if(samplers[s][i] == state) 1898 { 1899 samplers[s][i].p = NULL; 1900 sampler_csos[s][i] = NULL; 1901 update_flags |= (1 << (UPDATE_SAMPLERS_SHIFT + s)); 1902 } 1903 } 1904 } 1905 } 1906 1907 void UnbindBuffer(ID3D11Buffer* buffer) 1908 { 1909 SYNCHRONIZED; 1910 if(buffer == index_buffer) 1911 { 1912 index_buffer.p = 0; 1913 index_format = DXGI_FORMAT_UNKNOWN; 1914 index_offset = 0; 1915 struct pipe_index_buffer ib; 1916 memset(&ib, 0, sizeof(ib)); 1917 pipe->set_index_buffer(pipe, &ib); 1918 } 1919 1920 for(unsigned i = 0; i < num_vertex_buffers; ++i) 1921 { 1922 if(buffer == input_buffers[i]) 1923 { 1924 input_buffers[i].p = 0; 1925 memset(&vertex_buffers[num_vertex_buffers], 0, sizeof(vertex_buffers[num_vertex_buffers])); 1926 update_flags |= UPDATE_VERTEX_BUFFERS; 1927 } 1928 } 1929 1930 for(unsigned s = 0; s < D3D11_STAGES; ++s) 1931 { 1932 for(unsigned i = 0; i < sizeof(constant_buffers) / sizeof(constant_buffers[0]); ++i) 1933 { 1934 if(constant_buffers[s][i] == buffer) 1935 { 1936 constant_buffers[s][i] = (ID3D10Buffer*)NULL; 1937 pipe->set_constant_buffer(pipe, s, i, NULL); 1938 } 1939 } 1940 } 1941 } 1942 1943 void UnbindDepthStencilView(ID3D11DepthStencilView * view) 1944 { 1945 SYNCHRONIZED; 1946 if(view == depth_stencil_view) 1947 { 1948 depth_stencil_view.p = NULL; 1949 set_framebuffer(); 1950 } 1951 } 1952 1953 void UnbindRenderTargetView(ID3D11RenderTargetView* view) 1954 { 1955 SYNCHRONIZED; 1956 bool any_bound = false; 1957 for(unsigned i = 0; i < num_render_target_views; ++i) 1958 { 1959 if(render_target_views[i] == view) 1960 { 1961 render_target_views[i].p = NULL; 1962 any_bound = true; 1963 } 1964 } 1965 if(any_bound) 1966 set_framebuffer(); 1967 } 1968 1969 void UnbindShaderResourceView(ID3D11ShaderResourceView* view) 1970 { 1971 SYNCHRONIZED; 1972 for(unsigned s = 0; s < D3D11_STAGES; ++s) 1973 { 1974 for(unsigned i = 0; i < num_shader_resource_views[s]; ++i) 1975 { 1976 if(shader_resource_views[s][i] == view) 1977 { 1978 shader_resource_views[s][i].p = NULL; 1979 sampler_views[s][i] = NULL; 1980 update_flags |= (1 << (UPDATE_VIEWS_SHIFT + s)); 1981 } 1982 } 1983 } 1984 } 1985#endif 1986 1987#undef SYNCHRONIZED 1988}; 1989 1990#if API >= 11 1991/* This approach serves two purposes. 1992 * First, we don't want to do an atomic operation to manipulate the reference 1993 * count every time something is bound/unbound to the pipeline, since they are 1994 * expensive. 1995 * Fortunately, the immediate context can only be used by a single thread, so 1996 * we don't have to use them, as long as a separate reference count is used 1997 * (see dual_refcnt_t). 1998 * 1999 * Second, we want to avoid the Device -> DeviceContext -> bound DeviceChild -> Device 2000 * garbage cycle. 2001 * To avoid it, DeviceChild doesn't hold a reference to Device as usual, but adds 2002 * one for each external reference count, while internal nonatomic_add_ref doesn't 2003 * add any. 2004 * 2005 * Note that ideally we would to eliminate the non-atomic op too, but this is more 2006 * complicated, since we would either need to use garbage collection and give up 2007 * deterministic destruction (especially bad for large textures), or scan the whole 2008 * pipeline state every time the reference count of object drops to 0, which risks 2009 * pathological slowdowns. 2010 * 2011 * Since this microoptimization should matter relatively little, let's avoid it for now. 2012 * 2013 * Note that deferred contexts don't use this, since as a whole, they must thread-safe. 2014 * Eliminating the atomic ops for deferred contexts seems substantially harder. 2015 * This might be a problem if they are used in a one-shot multithreaded rendering 2016 * fashion, where SMP cacheline bouncing on the reference count may be visible. 2017 * 2018 * The idea would be to attach a structure of reference counts indexed by deferred 2019 * context id to each object. Ideally, this should be organized like ext2 block pointers. 2020 * 2021 * Every deferred context would get a reference count in its own cacheline. 2022 * The external count is protected by a lock bit, and there is also a "lock bit" in each 2023 * internal count. 2024 * 2025 * When the external count has to be dropped to 0, the lock bit is taken and all internal 2026 * reference counts are scanned, taking a count of them. A flag would also be set on them. 2027 * Deferred context manipulation would notice the flag, and update the count. 2028 * Once the count goes to zero, the object is freed. 2029 * 2030 * The problem of this is that if the external reference count ping-pongs between 2031 * zero and non-zero, the scans will take a lot of time. 2032 * 2033 * The idea to solve this is to compute the scans in a binary-tree like fashion, where 2034 * each binary tree node would have a "determined bit", which would be invalidated 2035 * by manipulations. 2036 * 2037 * However, all this complexity might actually be a loss in most cases, so let's just 2038 * stick to a single atomic refcnt for now. 2039 * 2040 * Also, we don't even support deferred contexts yet, so this can wait. 2041 */ 2042struct nonatomic_device_child_ptr_traits 2043{ 2044 static void add_ref(void* p) 2045 { 2046 if(p) 2047 ((GalliumD3D11DeviceChild<>*)p)->nonatomic_add_ref(); 2048 } 2049 2050 static void release(void* p) 2051 { 2052 if(p) 2053 ((GalliumD3D11DeviceChild<>*)p)->nonatomic_release(); 2054 } 2055}; 2056 2057struct GalliumD3D11ImmediateDeviceContext 2058 : public GalliumD3D11DeviceContext<nonatomic_device_child_ptr_traits> 2059{ 2060 GalliumD3D11ImmediateDeviceContext(GalliumD3D11Screen* device, pipe_context* pipe, unsigned context_flags = 0) 2061 : GalliumD3D11DeviceContext<nonatomic_device_child_ptr_traits>(device, pipe, context_flags) 2062 { 2063 // not necessary, but tests that the API at least basically works 2064 ClearState(); 2065 } 2066 2067 /* we do this since otherwise we would have a garbage cycle between this and the device */ 2068 virtual ULONG STDMETHODCALLTYPE AddRef() 2069 { 2070 return this->device->AddRef(); 2071 } 2072 2073 virtual ULONG STDMETHODCALLTYPE Release() 2074 { 2075 return this->device->Release(); 2076 } 2077 2078 virtual D3D11_DEVICE_CONTEXT_TYPE STDMETHODCALLTYPE GetType() 2079 { 2080 return D3D11_DEVICE_CONTEXT_IMMEDIATE; 2081 } 2082}; 2083 2084static ID3D11DeviceContext* GalliumD3D11ImmediateDeviceContext_Create(GalliumD3D11Screen* device, struct pipe_context* pipe, bool owns_pipe) 2085{ 2086 return new GalliumD3D11ImmediateDeviceContext(device, pipe, owns_pipe); 2087} 2088 2089static void GalliumD3D11ImmediateDeviceContext_RestoreGalliumState(ID3D11DeviceContext* context) 2090{ 2091 ((GalliumD3D11ImmediateDeviceContext*)context)->RestoreGalliumState(); 2092} 2093 2094static void GalliumD3D11ImmediateDeviceContext_RestoreGalliumStateBlitOnly(ID3D11DeviceContext* context) 2095{ 2096 ((GalliumD3D11ImmediateDeviceContext*)context)->RestoreGalliumStateBlitOnly(); 2097} 2098 2099static void GalliumD3D11ImmediateDeviceContext_Destroy(ID3D11DeviceContext* context) 2100{ 2101 delete (GalliumD3D11ImmediateDeviceContext*)context; 2102} 2103#endif 2104