1/************************************************************************** 2 * 3 * Copyright 2007 VMware, Inc. 4 * All Rights Reserved. 5 * Copyright 2009 VMware, Inc. All Rights Reserved. 6 * 7 * Permission is hereby granted, free of charge, to any person obtaining a 8 * copy of this software and associated documentation files (the 9 * "Software"), to deal in the Software without restriction, including 10 * without limitation the rights to use, copy, modify, merge, publish, 11 * distribute, sub license, and/or sell copies of the Software, and to 12 * permit persons to whom the Software is furnished to do so, subject to 13 * the following conditions: 14 * 15 * The above copyright notice and this permission notice (including the 16 * next paragraph) shall be included in all copies or substantial portions 17 * of the Software. 18 * 19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. 22 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR 23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, 24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE 25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 26 * 27 **************************************************************************/ 28 29 /* 30 * Authors: 31 * Keith Whitwell <keithw@vmware.com> 32 * Brian Paul 33 * Michel Dänzer 34 */ 35 36#include "main/errors.h" 37#include "main/glheader.h" 38#include "main/accum.h" 39#include "main/formats.h" 40#include "main/framebuffer.h" 41#include "main/macros.h" 42#include "main/glformats.h" 43#include "program/prog_instruction.h" 44#include "st_context.h" 45#include "st_atom.h" 46#include "st_cb_bitmap.h" 47#include "st_cb_clear.h" 48#include "st_cb_fbo.h" 49#include "st_draw.h" 50#include "st_format.h" 51#include "st_nir.h" 52#include "st_program.h" 53#include "st_util.h" 54 55#include "pipe/p_context.h" 56#include "pipe/p_shader_tokens.h" 57#include "pipe/p_state.h" 58#include "pipe/p_defines.h" 59#include "util/u_format.h" 60#include "util/u_inlines.h" 61#include "util/u_simple_shaders.h" 62 63#include "cso_cache/cso_context.h" 64 65 66/** 67 * Do per-context initialization for glClear. 68 */ 69void 70st_init_clear(struct st_context *st) 71{ 72 memset(&st->clear, 0, sizeof(st->clear)); 73 74 st->clear.raster.half_pixel_center = 1; 75 st->clear.raster.bottom_edge_rule = 1; 76 st->clear.raster.depth_clip_near = 1; 77 st->clear.raster.depth_clip_far = 1; 78} 79 80 81/** 82 * Free per-context state for glClear. 83 */ 84void 85st_destroy_clear(struct st_context *st) 86{ 87 if (st->clear.fs) { 88 cso_delete_fragment_shader(st->cso_context, st->clear.fs); 89 st->clear.fs = NULL; 90 } 91 if (st->clear.vs) { 92 cso_delete_vertex_shader(st->cso_context, st->clear.vs); 93 st->clear.vs = NULL; 94 } 95 if (st->clear.vs_layered) { 96 cso_delete_vertex_shader(st->cso_context, st->clear.vs_layered); 97 st->clear.vs_layered = NULL; 98 } 99 if (st->clear.gs_layered) { 100 cso_delete_geometry_shader(st->cso_context, st->clear.gs_layered); 101 st->clear.gs_layered = NULL; 102 } 103} 104 105 106/** 107 * Helper function to set the fragment shaders. 108 */ 109static inline void 110set_fragment_shader(struct st_context *st) 111{ 112 struct pipe_screen *pscreen = st->pipe->screen; 113 bool use_nir = PIPE_SHADER_IR_NIR == 114 pscreen->get_shader_param(pscreen, PIPE_SHADER_VERTEX, 115 PIPE_SHADER_CAP_PREFERRED_IR); 116 117 if (!st->clear.fs) { 118 if (use_nir) { 119 unsigned inputs[] = { VARYING_SLOT_VAR0 }; 120 unsigned outputs[] = { FRAG_RESULT_COLOR }; 121 unsigned interpolation[] = { INTERP_MODE_FLAT }; 122 st->clear.fs = st_nir_make_passthrough_shader(st, "clear FS", 123 MESA_SHADER_FRAGMENT, 124 1, inputs, outputs, 125 interpolation, 0); 126 } else { 127 st->clear.fs = 128 util_make_fragment_passthrough_shader(st->pipe, 129 TGSI_SEMANTIC_GENERIC, 130 TGSI_INTERPOLATE_CONSTANT, 131 TRUE); 132 } 133 } 134 135 cso_set_fragment_shader_handle(st->cso_context, st->clear.fs); 136} 137 138 139static void * 140make_nir_clear_vertex_shader(struct st_context *st, bool layered) 141{ 142 const char *shader_name = layered ? "layered clear VS" : "clear VS"; 143 unsigned inputs[] = { 144 VERT_ATTRIB_POS, 145 VERT_ATTRIB_GENERIC0, 146 SYSTEM_VALUE_INSTANCE_ID, 147 }; 148 unsigned outputs[] = { 149 VARYING_SLOT_POS, 150 VARYING_SLOT_VAR0, 151 VARYING_SLOT_LAYER 152 }; 153 154 return st_nir_make_passthrough_shader(st, shader_name, MESA_SHADER_VERTEX, 155 layered ? 3 : 2, inputs, outputs, 156 NULL, (1 << 2)); 157} 158 159 160/** 161 * Helper function to set the vertex shader. 162 */ 163static inline void 164set_vertex_shader(struct st_context *st) 165{ 166 struct pipe_screen *pscreen = st->pipe->screen; 167 bool use_nir = PIPE_SHADER_IR_NIR == 168 pscreen->get_shader_param(pscreen, PIPE_SHADER_VERTEX, 169 PIPE_SHADER_CAP_PREFERRED_IR); 170 171 /* vertex shader - still required to provide the linkage between 172 * fragment shader input semantics and vertex_element/buffers. 173 */ 174 if (!st->clear.vs) 175 { 176 if (use_nir) { 177 st->clear.vs = make_nir_clear_vertex_shader(st, false); 178 } else { 179 const enum tgsi_semantic semantic_names[] = { 180 TGSI_SEMANTIC_POSITION, 181 TGSI_SEMANTIC_GENERIC 182 }; 183 const uint semantic_indexes[] = { 0, 0 }; 184 st->clear.vs = util_make_vertex_passthrough_shader(st->pipe, 2, 185 semantic_names, 186 semantic_indexes, 187 FALSE); 188 } 189 } 190 191 cso_set_vertex_shader_handle(st->cso_context, st->clear.vs); 192 cso_set_geometry_shader_handle(st->cso_context, NULL); 193} 194 195 196static void 197set_vertex_shader_layered(struct st_context *st) 198{ 199 struct pipe_context *pipe = st->pipe; 200 struct pipe_screen *pscreen = pipe->screen; 201 bool use_nir = PIPE_SHADER_IR_NIR == 202 pscreen->get_shader_param(pscreen, PIPE_SHADER_VERTEX, 203 PIPE_SHADER_CAP_PREFERRED_IR); 204 205 if (!pipe->screen->get_param(pipe->screen, PIPE_CAP_TGSI_INSTANCEID)) { 206 assert(!"Got layered clear, but VS instancing is unsupported"); 207 set_vertex_shader(st); 208 return; 209 } 210 211 if (!st->clear.vs_layered) { 212 bool vs_layer = 213 pipe->screen->get_param(pipe->screen, PIPE_CAP_TGSI_VS_LAYER_VIEWPORT); 214 if (vs_layer) { 215 st->clear.vs_layered = 216 use_nir ? make_nir_clear_vertex_shader(st, true) 217 : util_make_layered_clear_vertex_shader(pipe); 218 } else { 219 st->clear.vs_layered = util_make_layered_clear_helper_vertex_shader(pipe); 220 st->clear.gs_layered = util_make_layered_clear_geometry_shader(pipe); 221 } 222 } 223 224 cso_set_vertex_shader_handle(st->cso_context, st->clear.vs_layered); 225 cso_set_geometry_shader_handle(st->cso_context, st->clear.gs_layered); 226} 227 228 229/** 230 * Do glClear by drawing a quadrilateral. 231 * The vertices of the quad will be computed from the 232 * ctx->DrawBuffer->_X/Ymin/max fields. 233 */ 234static void 235clear_with_quad(struct gl_context *ctx, unsigned clear_buffers) 236{ 237 struct st_context *st = st_context(ctx); 238 struct cso_context *cso = st->cso_context; 239 const struct gl_framebuffer *fb = ctx->DrawBuffer; 240 const GLfloat fb_width = (GLfloat) fb->Width; 241 const GLfloat fb_height = (GLfloat) fb->Height; 242 243 _mesa_update_draw_buffer_bounds(ctx, ctx->DrawBuffer); 244 245 const GLfloat x0 = (GLfloat) ctx->DrawBuffer->_Xmin / fb_width * 2.0f - 1.0f; 246 const GLfloat x1 = (GLfloat) ctx->DrawBuffer->_Xmax / fb_width * 2.0f - 1.0f; 247 const GLfloat y0 = (GLfloat) ctx->DrawBuffer->_Ymin / fb_height * 2.0f - 1.0f; 248 const GLfloat y1 = (GLfloat) ctx->DrawBuffer->_Ymax / fb_height * 2.0f - 1.0f; 249 unsigned num_layers = st->state.fb_num_layers; 250 251 /* 252 printf("%s %s%s%s %f,%f %f,%f\n", __func__, 253 color ? "color, " : "", 254 depth ? "depth, " : "", 255 stencil ? "stencil" : "", 256 x0, y0, 257 x1, y1); 258 */ 259 260 cso_save_state(cso, (CSO_BIT_BLEND | 261 CSO_BIT_STENCIL_REF | 262 CSO_BIT_DEPTH_STENCIL_ALPHA | 263 CSO_BIT_RASTERIZER | 264 CSO_BIT_SAMPLE_MASK | 265 CSO_BIT_MIN_SAMPLES | 266 CSO_BIT_VIEWPORT | 267 CSO_BIT_STREAM_OUTPUTS | 268 CSO_BIT_VERTEX_ELEMENTS | 269 CSO_BIT_AUX_VERTEX_BUFFER_SLOT | 270 CSO_BIT_PAUSE_QUERIES | 271 CSO_BITS_ALL_SHADERS)); 272 273 /* blend state: RGBA masking */ 274 { 275 struct pipe_blend_state blend; 276 memset(&blend, 0, sizeof(blend)); 277 if (clear_buffers & PIPE_CLEAR_COLOR) { 278 int num_buffers = ctx->Extensions.EXT_draw_buffers2 ? 279 ctx->DrawBuffer->_NumColorDrawBuffers : 1; 280 int i; 281 282 blend.independent_blend_enable = num_buffers > 1; 283 284 for (i = 0; i < num_buffers; i++) { 285 if (!(clear_buffers & (PIPE_CLEAR_COLOR0 << i))) 286 continue; 287 288 blend.rt[i].colormask = GET_COLORMASK(ctx->Color.ColorMask, i); 289 } 290 291 if (ctx->Color.DitherFlag) 292 blend.dither = 1; 293 } 294 cso_set_blend(cso, &blend); 295 } 296 297 /* depth_stencil state: always pass/set to ref value */ 298 { 299 struct pipe_depth_stencil_alpha_state depth_stencil; 300 memset(&depth_stencil, 0, sizeof(depth_stencil)); 301 if (clear_buffers & PIPE_CLEAR_DEPTH) { 302 depth_stencil.depth.enabled = 1; 303 depth_stencil.depth.writemask = 1; 304 depth_stencil.depth.func = PIPE_FUNC_ALWAYS; 305 } 306 307 if (clear_buffers & PIPE_CLEAR_STENCIL) { 308 struct pipe_stencil_ref stencil_ref; 309 memset(&stencil_ref, 0, sizeof(stencil_ref)); 310 depth_stencil.stencil[0].enabled = 1; 311 depth_stencil.stencil[0].func = PIPE_FUNC_ALWAYS; 312 depth_stencil.stencil[0].fail_op = PIPE_STENCIL_OP_REPLACE; 313 depth_stencil.stencil[0].zpass_op = PIPE_STENCIL_OP_REPLACE; 314 depth_stencil.stencil[0].zfail_op = PIPE_STENCIL_OP_REPLACE; 315 depth_stencil.stencil[0].valuemask = 0xff; 316 depth_stencil.stencil[0].writemask = ctx->Stencil.WriteMask[0] & 0xff; 317 stencil_ref.ref_value[0] = ctx->Stencil.Clear; 318 cso_set_stencil_ref(cso, &stencil_ref); 319 } 320 321 cso_set_depth_stencil_alpha(cso, &depth_stencil); 322 } 323 324 cso_set_vertex_elements(cso, 2, st->util_velems); 325 cso_set_stream_outputs(cso, 0, NULL, NULL); 326 cso_set_sample_mask(cso, ~0); 327 cso_set_min_samples(cso, 1); 328 cso_set_rasterizer(cso, &st->clear.raster); 329 330 /* viewport state: viewport matching window dims */ 331 cso_set_viewport_dims(st->cso_context, fb_width, fb_height, 332 st_fb_orientation(fb) == Y_0_TOP); 333 334 set_fragment_shader(st); 335 cso_set_tessctrl_shader_handle(cso, NULL); 336 cso_set_tesseval_shader_handle(cso, NULL); 337 338 if (num_layers > 1) 339 set_vertex_shader_layered(st); 340 else 341 set_vertex_shader(st); 342 343 /* draw quad matching scissor rect. 344 * 345 * Note: if we're only clearing depth/stencil we still setup vertices 346 * with color, but they'll be ignored. 347 * 348 * We can't translate the clear color to the colorbuffer format, 349 * because different colorbuffers may have different formats. 350 */ 351 if (!st_draw_quad(st, x0, y0, x1, y1, 352 ctx->Depth.Clear * 2.0f - 1.0f, 353 0.0f, 0.0f, 0.0f, 0.0f, 354 (const float *) &ctx->Color.ClearColor.f, 355 num_layers)) { 356 _mesa_error(ctx, GL_OUT_OF_MEMORY, "glClear"); 357 } 358 359 /* Restore pipe state */ 360 cso_restore_state(cso); 361} 362 363 364/** 365 * Return if the scissor must be enabled during the clear. 366 */ 367static inline GLboolean 368is_scissor_enabled(struct gl_context *ctx, struct gl_renderbuffer *rb) 369{ 370 const struct gl_scissor_rect *scissor = &ctx->Scissor.ScissorArray[0]; 371 372 return (ctx->Scissor.EnableFlags & 1) && 373 (scissor->X > 0 || 374 scissor->Y > 0 || 375 scissor->X + scissor->Width < (int)rb->Width || 376 scissor->Y + scissor->Height < (int)rb->Height); 377} 378 379/** 380 * Return if window rectangles must be enabled during the clear. 381 */ 382static inline bool 383is_window_rectangle_enabled(struct gl_context *ctx) 384{ 385 if (ctx->DrawBuffer == ctx->WinSysDrawBuffer) 386 return false; 387 return ctx->Scissor.NumWindowRects > 0 || 388 ctx->Scissor.WindowRectMode == GL_INCLUSIVE_EXT; 389} 390 391 392/** 393 * Return if all of the stencil bits are masked. 394 */ 395static inline GLboolean 396is_stencil_disabled(struct gl_context *ctx, struct gl_renderbuffer *rb) 397{ 398 const GLuint stencilMax = 0xff; 399 400 assert(_mesa_get_format_bits(rb->Format, GL_STENCIL_BITS) > 0); 401 return (ctx->Stencil.WriteMask[0] & stencilMax) == 0; 402} 403 404 405/** 406 * Return if any of the stencil bits are masked. 407 */ 408static inline GLboolean 409is_stencil_masked(struct gl_context *ctx, struct gl_renderbuffer *rb) 410{ 411 const GLuint stencilMax = 0xff; 412 413 assert(_mesa_get_format_bits(rb->Format, GL_STENCIL_BITS) > 0); 414 return (ctx->Stencil.WriteMask[0] & stencilMax) != stencilMax; 415} 416 417 418/** 419 * Called via ctx->Driver.Clear() 420 */ 421static void 422st_Clear(struct gl_context *ctx, GLbitfield mask) 423{ 424 struct st_context *st = st_context(ctx); 425 struct gl_renderbuffer *depthRb 426 = ctx->DrawBuffer->Attachment[BUFFER_DEPTH].Renderbuffer; 427 struct gl_renderbuffer *stencilRb 428 = ctx->DrawBuffer->Attachment[BUFFER_STENCIL].Renderbuffer; 429 GLbitfield quad_buffers = 0x0; 430 GLbitfield clear_buffers = 0x0; 431 GLuint i; 432 433 st_flush_bitmap_cache(st); 434 st_invalidate_readpix_cache(st); 435 436 /* This makes sure the pipe has the latest scissor, etc values */ 437 st_validate_state(st, ST_PIPELINE_CLEAR); 438 439 if (mask & BUFFER_BITS_COLOR) { 440 for (i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; i++) { 441 gl_buffer_index b = ctx->DrawBuffer->_ColorDrawBufferIndexes[i]; 442 443 if (b != BUFFER_NONE && mask & (1 << b)) { 444 struct gl_renderbuffer *rb 445 = ctx->DrawBuffer->Attachment[b].Renderbuffer; 446 struct st_renderbuffer *strb = st_renderbuffer(rb); 447 int colormask_index = ctx->Extensions.EXT_draw_buffers2 ? i : 0; 448 449 if (!strb || !strb->surface) 450 continue; 451 452 unsigned colormask = 453 GET_COLORMASK(ctx->Color.ColorMask, colormask_index); 454 455 if (!colormask) 456 continue; 457 458 unsigned surf_colormask = 459 util_format_colormask(util_format_description(strb->surface->format)); 460 461 if (is_scissor_enabled(ctx, rb) || 462 is_window_rectangle_enabled(ctx) || 463 ((colormask & surf_colormask) != surf_colormask)) 464 quad_buffers |= PIPE_CLEAR_COLOR0 << i; 465 else 466 clear_buffers |= PIPE_CLEAR_COLOR0 << i; 467 } 468 } 469 } 470 471 if (mask & BUFFER_BIT_DEPTH) { 472 struct st_renderbuffer *strb = st_renderbuffer(depthRb); 473 474 if (strb->surface && ctx->Depth.Mask) { 475 if (is_scissor_enabled(ctx, depthRb) || 476 is_window_rectangle_enabled(ctx)) 477 quad_buffers |= PIPE_CLEAR_DEPTH; 478 else 479 clear_buffers |= PIPE_CLEAR_DEPTH; 480 } 481 } 482 if (mask & BUFFER_BIT_STENCIL) { 483 struct st_renderbuffer *strb = st_renderbuffer(stencilRb); 484 485 if (strb->surface && !is_stencil_disabled(ctx, stencilRb)) { 486 if (is_scissor_enabled(ctx, stencilRb) || 487 is_window_rectangle_enabled(ctx) || 488 is_stencil_masked(ctx, stencilRb)) 489 quad_buffers |= PIPE_CLEAR_STENCIL; 490 else 491 clear_buffers |= PIPE_CLEAR_STENCIL; 492 } 493 } 494 495 /* Always clear depth and stencil together. 496 * This can only happen when the stencil writemask is not a full mask. 497 */ 498 if (quad_buffers & PIPE_CLEAR_DEPTHSTENCIL && 499 clear_buffers & PIPE_CLEAR_DEPTHSTENCIL) { 500 quad_buffers |= clear_buffers & PIPE_CLEAR_DEPTHSTENCIL; 501 clear_buffers &= ~PIPE_CLEAR_DEPTHSTENCIL; 502 } 503 504 /* Only use quad-based clearing for the renderbuffers which cannot 505 * use pipe->clear. We want to always use pipe->clear for the other 506 * renderbuffers, because it's likely to be faster. 507 */ 508 if (clear_buffers) { 509 /* We can't translate the clear color to the colorbuffer format, 510 * because different colorbuffers may have different formats. 511 */ 512 st->pipe->clear(st->pipe, clear_buffers, 513 (union pipe_color_union*)&ctx->Color.ClearColor, 514 ctx->Depth.Clear, ctx->Stencil.Clear); 515 } 516 if (quad_buffers) { 517 clear_with_quad(ctx, quad_buffers); 518 } 519 if (mask & BUFFER_BIT_ACCUM) 520 _mesa_clear_accum_buffer(ctx); 521} 522 523 524void 525st_init_clear_functions(struct dd_function_table *functions) 526{ 527 functions->Clear = st_Clear; 528} 529