1 /********************************************************** 2 * Copyright 2008-2009 VMware, Inc. All rights reserved. 3 * 4 * Permission is hereby granted, free of charge, to any person 5 * obtaining a copy of this software and associated documentation 6 * files (the "Software"), to deal in the Software without 7 * restriction, including without limitation the rights to use, copy, 8 * modify, merge, publish, distribute, sublicense, and/or sell copies 9 * of the Software, and to permit persons to whom the Software is 10 * furnished to do so, subject to the following conditions: 11 * 12 * The above copyright notice and this permission notice shall be 13 * included in all copies or substantial portions of the Software. 14 * 15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 16 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 17 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 18 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 19 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 20 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 21 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 22 * SOFTWARE. 23 * 24 **********************************************************/ 25 26 #include "pipe/p_defines.h" 27 #include "util/u_bitmask.h" 28 #include "util/u_format.h" 29 #include "util/u_inlines.h" 30 #include "util/u_memory.h" 31 #include "util/u_math.h" 32 #include "util/u_memory.h" 33 34 #include "svga_context.h" 35 #include "svga_screen.h" 36 #include "svga_state.h" 37 #include "svga_cmd.h" 38 #include "svga_format.h" 39 #include "svga_shader.h" 40 41 42 struct rs_queue { 43 unsigned rs_count; 44 SVGA3dRenderState rs[SVGA3D_RS_MAX]; 45 }; 46 47 48 #define EMIT_RS(svga, value, token, fail) \ 49 do { \ 50 STATIC_ASSERT(SVGA3D_RS_##token < ARRAY_SIZE(svga->state.hw_draw.rs)); \ 51 if (svga->state.hw_draw.rs[SVGA3D_RS_##token] != value) { \ 52 svga_queue_rs( &queue, SVGA3D_RS_##token, value ); \ 53 svga->state.hw_draw.rs[SVGA3D_RS_##token] = value; \ 54 } \ 55 } while (0) 56 57 #define EMIT_RS_FLOAT(svga, fvalue, token, fail) \ 58 do { \ 59 unsigned value = fui(fvalue); \ 60 STATIC_ASSERT(SVGA3D_RS_##token < ARRAY_SIZE(svga->state.hw_draw.rs)); \ 61 if (svga->state.hw_draw.rs[SVGA3D_RS_##token] != value) { \ 62 svga_queue_rs( &queue, SVGA3D_RS_##token, value ); \ 63 svga->state.hw_draw.rs[SVGA3D_RS_##token] = value; \ 64 } \ 65 } while (0) 66 67 68 static inline void 69 svga_queue_rs( struct rs_queue *q, 70 unsigned rss, 71 unsigned value ) 72 { 73 q->rs[q->rs_count].state = rss; 74 q->rs[q->rs_count].uintValue = value; 75 q->rs_count++; 76 } 77 78 79 /* Compare old and new render states and emit differences between them 80 * to hardware. Simplest implementation would be to emit the whole of 81 * the "to" state. 82 */ 83 static enum pipe_error 84 emit_rss_vgpu9(struct svga_context *svga, unsigned dirty) 85 { 86 struct svga_screen *screen = svga_screen(svga->pipe.screen); 87 struct rs_queue queue; 88 float point_size_min; 89 90 queue.rs_count = 0; 91 92 if (dirty & (SVGA_NEW_BLEND | SVGA_NEW_BLEND_COLOR)) { 93 const struct svga_blend_state *curr = svga->curr.blend; 94 95 EMIT_RS( svga, curr->rt[0].writemask, COLORWRITEENABLE, fail ); 96 EMIT_RS( svga, curr->rt[0].blend_enable, BLENDENABLE, fail ); 97 98 if (curr->rt[0].blend_enable) { 99 EMIT_RS( svga, curr->rt[0].srcblend, SRCBLEND, fail ); 100 EMIT_RS( svga, curr->rt[0].dstblend, DSTBLEND, fail ); 101 EMIT_RS( svga, curr->rt[0].blendeq, BLENDEQUATION, fail ); 102 103 EMIT_RS( svga, curr->rt[0].separate_alpha_blend_enable, 104 SEPARATEALPHABLENDENABLE, fail ); 105 106 if (curr->rt[0].separate_alpha_blend_enable) { 107 EMIT_RS( svga, curr->rt[0].srcblend_alpha, SRCBLENDALPHA, fail ); 108 EMIT_RS( svga, curr->rt[0].dstblend_alpha, DSTBLENDALPHA, fail ); 109 EMIT_RS( svga, curr->rt[0].blendeq_alpha, BLENDEQUATIONALPHA, fail ); 110 } 111 } 112 } 113 114 if (dirty & SVGA_NEW_BLEND_COLOR) { 115 uint32 color; 116 uint32 r = float_to_ubyte(svga->curr.blend_color.color[0]); 117 uint32 g = float_to_ubyte(svga->curr.blend_color.color[1]); 118 uint32 b = float_to_ubyte(svga->curr.blend_color.color[2]); 119 uint32 a = float_to_ubyte(svga->curr.blend_color.color[3]); 120 121 color = (a << 24) | (r << 16) | (g << 8) | b; 122 123 EMIT_RS( svga, color, BLENDCOLOR, fail ); 124 } 125 126 if (dirty & (SVGA_NEW_DEPTH_STENCIL_ALPHA | SVGA_NEW_RAST)) { 127 const struct svga_depth_stencil_state *curr = svga->curr.depth; 128 const struct svga_rasterizer_state *rast = svga->curr.rast; 129 130 if (!curr->stencil[0].enabled) 131 { 132 /* Stencil disabled 133 */ 134 EMIT_RS( svga, FALSE, STENCILENABLE, fail ); 135 EMIT_RS( svga, FALSE, STENCILENABLE2SIDED, fail ); 136 } 137 else if (curr->stencil[0].enabled && !curr->stencil[1].enabled) 138 { 139 /* Regular stencil 140 */ 141 EMIT_RS( svga, TRUE, STENCILENABLE, fail ); 142 EMIT_RS( svga, FALSE, STENCILENABLE2SIDED, fail ); 143 144 EMIT_RS( svga, curr->stencil[0].func, STENCILFUNC, fail ); 145 EMIT_RS( svga, curr->stencil[0].fail, STENCILFAIL, fail ); 146 EMIT_RS( svga, curr->stencil[0].zfail, STENCILZFAIL, fail ); 147 EMIT_RS( svga, curr->stencil[0].pass, STENCILPASS, fail ); 148 149 EMIT_RS( svga, curr->stencil_mask, STENCILMASK, fail ); 150 EMIT_RS( svga, curr->stencil_writemask, STENCILWRITEMASK, fail ); 151 } 152 else 153 { 154 int cw, ccw; 155 156 /* Hardware frontwinding is always CW, so if ours is also CW, 157 * then our definition of front face agrees with hardware. 158 * Otherwise need to flip. 159 */ 160 if (rast->templ.front_ccw) { 161 ccw = 0; 162 cw = 1; 163 } 164 else { 165 ccw = 1; 166 cw = 0; 167 } 168 169 /* Twoside stencil 170 */ 171 EMIT_RS( svga, TRUE, STENCILENABLE, fail ); 172 EMIT_RS( svga, TRUE, STENCILENABLE2SIDED, fail ); 173 174 EMIT_RS( svga, curr->stencil[cw].func, STENCILFUNC, fail ); 175 EMIT_RS( svga, curr->stencil[cw].fail, STENCILFAIL, fail ); 176 EMIT_RS( svga, curr->stencil[cw].zfail, STENCILZFAIL, fail ); 177 EMIT_RS( svga, curr->stencil[cw].pass, STENCILPASS, fail ); 178 179 EMIT_RS( svga, curr->stencil[ccw].func, CCWSTENCILFUNC, fail ); 180 EMIT_RS( svga, curr->stencil[ccw].fail, CCWSTENCILFAIL, fail ); 181 EMIT_RS( svga, curr->stencil[ccw].zfail, CCWSTENCILZFAIL, fail ); 182 EMIT_RS( svga, curr->stencil[ccw].pass, CCWSTENCILPASS, fail ); 183 184 EMIT_RS( svga, curr->stencil_mask, STENCILMASK, fail ); 185 EMIT_RS( svga, curr->stencil_writemask, STENCILWRITEMASK, fail ); 186 } 187 188 EMIT_RS( svga, curr->zenable, ZENABLE, fail ); 189 if (curr->zenable) { 190 EMIT_RS( svga, curr->zfunc, ZFUNC, fail ); 191 EMIT_RS( svga, curr->zwriteenable, ZWRITEENABLE, fail ); 192 } 193 194 EMIT_RS( svga, curr->alphatestenable, ALPHATESTENABLE, fail ); 195 if (curr->alphatestenable) { 196 EMIT_RS( svga, curr->alphafunc, ALPHAFUNC, fail ); 197 EMIT_RS_FLOAT( svga, curr->alpharef, ALPHAREF, fail ); 198 } 199 } 200 201 if (dirty & SVGA_NEW_STENCIL_REF) { 202 EMIT_RS( svga, svga->curr.stencil_ref.ref_value[0], STENCILREF, fail ); 203 } 204 205 if (dirty & (SVGA_NEW_RAST | SVGA_NEW_NEED_PIPELINE)) 206 { 207 const struct svga_rasterizer_state *curr = svga->curr.rast; 208 unsigned cullmode = curr->cullmode; 209 210 /* Shademode: still need to rearrange index list to move 211 * flat-shading PV first vertex. 212 */ 213 EMIT_RS( svga, curr->shademode, SHADEMODE, fail ); 214 215 /* Don't do culling while the software pipeline is active. It 216 * does it for us, and additionally introduces potentially 217 * back-facing triangles. 218 */ 219 if (svga->state.sw.need_pipeline) 220 cullmode = SVGA3D_FACE_NONE; 221 222 point_size_min = util_get_min_point_size(&curr->templ); 223 224 EMIT_RS( svga, cullmode, CULLMODE, fail ); 225 EMIT_RS( svga, curr->scissortestenable, SCISSORTESTENABLE, fail ); 226 EMIT_RS( svga, curr->multisampleantialias, MULTISAMPLEANTIALIAS, fail ); 227 EMIT_RS( svga, curr->lastpixel, LASTPIXEL, fail ); 228 EMIT_RS_FLOAT( svga, curr->pointsize, POINTSIZE, fail ); 229 EMIT_RS_FLOAT( svga, point_size_min, POINTSIZEMIN, fail ); 230 EMIT_RS_FLOAT( svga, screen->maxPointSize, POINTSIZEMAX, fail ); 231 EMIT_RS( svga, curr->pointsprite, POINTSPRITEENABLE, fail); 232 233 /* Emit line state, when the device understands it */ 234 if (screen->haveLineStipple) 235 EMIT_RS( svga, curr->linepattern, LINEPATTERN, fail ); 236 if (screen->haveLineSmooth) 237 EMIT_RS( svga, curr->antialiasedlineenable, ANTIALIASEDLINEENABLE, fail ); 238 if (screen->maxLineWidth > 1.0F) 239 EMIT_RS_FLOAT( svga, curr->linewidth, LINEWIDTH, fail ); 240 } 241 242 if (dirty & (SVGA_NEW_RAST | SVGA_NEW_FRAME_BUFFER | SVGA_NEW_NEED_PIPELINE)) 243 { 244 const struct svga_rasterizer_state *curr = svga->curr.rast; 245 float slope = 0.0; 246 float bias = 0.0; 247 248 /* Need to modify depth bias according to bound depthbuffer 249 * format. Don't do hardware depthbias while the software 250 * pipeline is active. 251 */ 252 if (!svga->state.sw.need_pipeline && 253 svga->curr.framebuffer.zsbuf) 254 { 255 slope = curr->slopescaledepthbias; 256 bias = svga->curr.depthscale * curr->depthbias; 257 } 258 259 EMIT_RS_FLOAT( svga, slope, SLOPESCALEDEPTHBIAS, fail ); 260 EMIT_RS_FLOAT( svga, bias, DEPTHBIAS, fail ); 261 } 262 263 if (dirty & SVGA_NEW_FRAME_BUFFER) { 264 /* XXX: we only look at the first color buffer's sRGB state */ 265 float gamma = 1.0f; 266 if (svga->curr.framebuffer.cbufs[0] && 267 util_format_is_srgb(svga->curr.framebuffer.cbufs[0]->format)) { 268 gamma = 2.2f; 269 } 270 EMIT_RS_FLOAT(svga, gamma, OUTPUTGAMMA, fail); 271 } 272 273 if (dirty & SVGA_NEW_RAST) { 274 /* bitmask of the enabled clip planes */ 275 unsigned enabled = svga->curr.rast->templ.clip_plane_enable; 276 EMIT_RS( svga, enabled, CLIPPLANEENABLE, fail ); 277 } 278 279 if (queue.rs_count) { 280 SVGA3dRenderState *rs; 281 282 if (SVGA3D_BeginSetRenderState( svga->swc, 283 &rs, 284 queue.rs_count ) != PIPE_OK) 285 goto fail; 286 287 memcpy( rs, 288 queue.rs, 289 queue.rs_count * sizeof queue.rs[0]); 290 291 SVGA_FIFOCommitAll( svga->swc ); 292 } 293 294 return PIPE_OK; 295 296 fail: 297 /* XXX: need to poison cached hardware state on failure to ensure 298 * dirty state gets re-emitted. Fix this by re-instating partial 299 * FIFOCommit command and only updating cached hw state once the 300 * initial allocation has succeeded. 301 */ 302 memset(svga->state.hw_draw.rs, 0xcd, sizeof(svga->state.hw_draw.rs)); 303 304 return PIPE_ERROR_OUT_OF_MEMORY; 305 } 306 307 /** Returns a non-culling rasterizer state object to be used with 308 * point sprite. 309 */ 310 static struct svga_rasterizer_state * 311 get_no_cull_rasterizer_state(struct svga_context *svga) 312 { 313 const struct svga_rasterizer_state *r = svga->curr.rast; 314 unsigned int aa_point = r->templ.point_smooth; 315 316 if (!svga->rasterizer_no_cull[aa_point]) { 317 struct pipe_rasterizer_state rast; 318 319 memset(&rast, 0, sizeof(rast)); 320 rast.flatshade = 1; 321 rast.front_ccw = 1; 322 rast.point_smooth = r->templ.point_smooth; 323 324 /* All rasterizer states have the same half_pixel_center, 325 * bottom_edge_rule and clip_halfz values since they are 326 * constant for a context. If we ever implement 327 * GL_ARB_clip_control, the clip_halfz field would have to be observed. 328 */ 329 rast.half_pixel_center = r->templ.half_pixel_center; 330 rast.bottom_edge_rule = r->templ.bottom_edge_rule; 331 rast.clip_halfz = r->templ.clip_halfz; 332 333 svga->rasterizer_no_cull[aa_point] = 334 svga->pipe.create_rasterizer_state(&svga->pipe, &rast); 335 } 336 return svga->rasterizer_no_cull[aa_point]; 337 } 338 339 static enum pipe_error 340 emit_rss_vgpu10(struct svga_context *svga, unsigned dirty) 341 { 342 enum pipe_error ret = PIPE_OK; 343 344 svga_hwtnl_flush_retry(svga); 345 346 if (dirty & (SVGA_NEW_BLEND | SVGA_NEW_BLEND_COLOR)) { 347 const struct svga_blend_state *curr; 348 float blend_factor[4]; 349 350 if (svga_has_any_integer_cbufs(svga)) { 351 /* Blending is not supported in integer-valued render targets. */ 352 curr = svga->noop_blend; 353 blend_factor[0] = 354 blend_factor[1] = 355 blend_factor[2] = 356 blend_factor[3] = 0; 357 } 358 else { 359 curr = svga->curr.blend; 360 361 if (curr->blend_color_alpha) { 362 blend_factor[0] = 363 blend_factor[1] = 364 blend_factor[2] = 365 blend_factor[3] = svga->curr.blend_color.color[3]; 366 } 367 else { 368 blend_factor[0] = svga->curr.blend_color.color[0]; 369 blend_factor[1] = svga->curr.blend_color.color[1]; 370 blend_factor[2] = svga->curr.blend_color.color[2]; 371 blend_factor[3] = svga->curr.blend_color.color[3]; 372 } 373 } 374 375 /* Set/bind the blend state object */ 376 if (svga->state.hw_draw.blend_id != curr->id || 377 svga->state.hw_draw.blend_factor[0] != blend_factor[0] || 378 svga->state.hw_draw.blend_factor[1] != blend_factor[1] || 379 svga->state.hw_draw.blend_factor[2] != blend_factor[2] || 380 svga->state.hw_draw.blend_factor[3] != blend_factor[3] || 381 svga->state.hw_draw.blend_sample_mask != svga->curr.sample_mask) { 382 ret = SVGA3D_vgpu10_SetBlendState(svga->swc, curr->id, 383 blend_factor, 384 svga->curr.sample_mask); 385 if (ret != PIPE_OK) 386 return ret; 387 388 svga->state.hw_draw.blend_id = curr->id; 389 svga->state.hw_draw.blend_factor[0] = blend_factor[0]; 390 svga->state.hw_draw.blend_factor[1] = blend_factor[1]; 391 svga->state.hw_draw.blend_factor[2] = blend_factor[2]; 392 svga->state.hw_draw.blend_factor[3] = blend_factor[3]; 393 svga->state.hw_draw.blend_sample_mask = svga->curr.sample_mask; 394 } 395 } 396 397 if (dirty & (SVGA_NEW_DEPTH_STENCIL_ALPHA | SVGA_NEW_STENCIL_REF)) { 398 const struct svga_depth_stencil_state *curr = svga->curr.depth; 399 unsigned curr_ref = svga->curr.stencil_ref.ref_value[0]; 400 401 if (curr->id != svga->state.hw_draw.depth_stencil_id || 402 curr_ref != svga->state.hw_draw.stencil_ref) { 403 /* Set/bind the depth/stencil state object */ 404 ret = SVGA3D_vgpu10_SetDepthStencilState(svga->swc, curr->id, 405 curr_ref); 406 if (ret != PIPE_OK) 407 return ret; 408 409 svga->state.hw_draw.depth_stencil_id = curr->id; 410 svga->state.hw_draw.stencil_ref = curr_ref; 411 } 412 } 413 414 if (dirty & (SVGA_NEW_REDUCED_PRIMITIVE | SVGA_NEW_RAST)) { 415 const struct svga_rasterizer_state *rast; 416 417 if (svga->curr.reduced_prim == PIPE_PRIM_POINTS && 418 svga->curr.gs && svga->curr.gs->wide_point) { 419 420 /* If we are drawing a point sprite, we will need to 421 * bind a non-culling rasterizer state object 422 */ 423 rast = get_no_cull_rasterizer_state(svga); 424 } 425 else { 426 rast = svga->curr.rast; 427 } 428 429 if (svga->state.hw_draw.rasterizer_id != rast->id) { 430 /* Set/bind the rasterizer state object */ 431 ret = SVGA3D_vgpu10_SetRasterizerState(svga->swc, rast->id); 432 if (ret != PIPE_OK) 433 return ret; 434 svga->state.hw_draw.rasterizer_id = rast->id; 435 } 436 } 437 return PIPE_OK; 438 } 439 440 441 static enum pipe_error 442 emit_rss(struct svga_context *svga, unsigned dirty) 443 { 444 if (svga_have_vgpu10(svga)) { 445 return emit_rss_vgpu10(svga, dirty); 446 } 447 else { 448 return emit_rss_vgpu9(svga, dirty); 449 } 450 } 451 452 453 struct svga_tracked_state svga_hw_rss = 454 { 455 "hw rss state", 456 457 (SVGA_NEW_BLEND | 458 SVGA_NEW_BLEND_COLOR | 459 SVGA_NEW_DEPTH_STENCIL_ALPHA | 460 SVGA_NEW_STENCIL_REF | 461 SVGA_NEW_RAST | 462 SVGA_NEW_FRAME_BUFFER | 463 SVGA_NEW_NEED_PIPELINE | 464 SVGA_NEW_REDUCED_PRIMITIVE), 465 466 emit_rss 467 }; 468