src/gallium/drivers/svga/svga_state_framebuffer.c

   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 "util/u_inlines.h"
  27 #include "pipe/p_defines.h"
  28 #include "util/u_math.h"
  29
  30 #include "svga_context.h"
  31 #include "svga_state.h"
  32 #include "svga_cmd.h"
  33 #include "svga_debug.h"
  34
  35
  36 /*
  37  * flush our command buffer after the 8th distinct render target
  38  *
  39  * This helps improve the surface cache behaviour in the face of the
  40  * large number of single-use render targets generated by EXA and the xorg
  41  * state tracker.  Without this we can reference hundreds of individual
  42  * render targets from a command buffer, which leaves little scope for
  43  * sharing or reuse of those targets.
  44  */
  45 #define MAX_RT_PER_BATCH 8
  46
  47
  48 /***********************************************************************
  49  * Hardware state update
  50  */
  51
  52
  53 static enum pipe_error
  54 emit_framebuffer( struct svga_context *svga,
  55                   unsigned dirty )
  56 {
  57    const struct pipe_framebuffer_state *curr = &svga->curr.framebuffer;
  58    struct pipe_framebuffer_state *hw = &svga->state.hw_clear.framebuffer;
  59    boolean reemit = svga->rebind.rendertargets;
  60    unsigned i;
  61    enum pipe_error ret;
  62
  63    /*
  64     * We need to reemit non-null surface bindings, even when they are not
  65     * dirty, to ensure that the resources are paged in.
  66     */
  67
  68    for(i = 0; i < PIPE_MAX_COLOR_BUFS; ++i) {
  69       if (curr->cbufs[i] != hw->cbufs[i] ||
  70           (reemit && hw->cbufs[i])) {
  71          if (svga->curr.nr_fbs++ > MAX_RT_PER_BATCH)
  72             return PIPE_ERROR_OUT_OF_MEMORY;
  73
  74          ret = SVGA3D_SetRenderTarget(svga->swc, SVGA3D_RT_COLOR0 + i, curr->cbufs[i]);
  75          if (ret != PIPE_OK)
  76             return ret;
  77
  78          pipe_surface_reference(&hw->cbufs[i], curr->cbufs[i]);
  79       }
  80    }
  81
  82
  83    if (curr->zsbuf != hw->zsbuf ||
  84        (reemit && hw->zsbuf)) {
  85       ret = SVGA3D_SetRenderTarget(svga->swc, SVGA3D_RT_DEPTH, curr->zsbuf);
  86       if (ret != PIPE_OK)
  87          return ret;
  88
  89       if (curr->zsbuf &&
  90           curr->zsbuf->format == PIPE_FORMAT_S8_UINT_Z24_UNORM) {
  91          ret = SVGA3D_SetRenderTarget(svga->swc, SVGA3D_RT_STENCIL, curr->zsbuf);
  92          if (ret != PIPE_OK)
  93             return ret;
  94       }
  95       else {
  96          ret = SVGA3D_SetRenderTarget(svga->swc, SVGA3D_RT_STENCIL, NULL);
  97          if (ret != PIPE_OK)
  98             return ret;
  99       }
 100
 101       pipe_surface_reference(&hw->zsbuf, curr->zsbuf);
 102    }
 103
 104    svga->rebind.rendertargets = FALSE;
 105
 106    return PIPE_OK;
 107 }
 108
 109
 110 /*
 111  * Rebind rendertargets.
 112  *
 113  * Similar to emit_framebuffer, but without any state checking/update.
 114  *
 115  * Called at the beginning of every new command buffer to ensure that
 116  * non-dirty rendertargets are properly paged-in.
 117  */
 118 enum pipe_error
 119 svga_reemit_framebuffer_bindings(struct svga_context *svga)
 120 {
 121    struct pipe_framebuffer_state *hw = &svga->state.hw_clear.framebuffer;
 122    unsigned i;
 123    enum pipe_error ret;
 124
 125    assert(svga->rebind.rendertargets);
 126
 127    for (i = 0; i < MIN2(PIPE_MAX_COLOR_BUFS, 8); ++i) {
 128       if (hw->cbufs[i]) {
 129          ret = SVGA3D_SetRenderTarget(svga->swc, SVGA3D_RT_COLOR0 + i, hw->cbufs[i]);
 130          if (ret != PIPE_OK) {
 131             return ret;
 132          }
 133       }
 134    }
 135
 136    if (hw->zsbuf) {
 137       ret = SVGA3D_SetRenderTarget(svga->swc, SVGA3D_RT_DEPTH, hw->zsbuf);
 138       if (ret != PIPE_OK) {
 139          return ret;
 140       }
 141
 142       if (hw->zsbuf &&
 143           hw->zsbuf->format == PIPE_FORMAT_S8_UINT_Z24_UNORM) {
 144          ret = SVGA3D_SetRenderTarget(svga->swc, SVGA3D_RT_STENCIL, hw->zsbuf);
 145          if (ret != PIPE_OK) {
 146             return ret;
 147          }
 148       }
 149       else {
 150          ret = SVGA3D_SetRenderTarget(svga->swc, SVGA3D_RT_STENCIL, NULL);
 151          if (ret != PIPE_OK) {
 152             return ret;
 153          }
 154       }
 155    }
 156
 157    svga->rebind.rendertargets = FALSE;
 158
 159    return PIPE_OK;
 160 }
 161
 162
 163 struct svga_tracked_state svga_hw_framebuffer =
 164 {
 165    "hw framebuffer state",
 166    SVGA_NEW_FRAME_BUFFER,
 167    emit_framebuffer
 168 };
 169
 170
 171
 172
 173 /***********************************************************************
 174  */
 175
 176 static enum pipe_error
 177 emit_viewport( struct svga_context *svga,
 178                unsigned dirty )
 179 {
 180    const struct pipe_viewport_state *viewport = &svga->curr.viewport;
 181    struct svga_prescale prescale;
 182    SVGA3dRect rect;
 183    /* Not sure if this state is relevant with POSITIONT.  Probably
 184     * not, but setting to 0,1 avoids some state pingponging.
 185     */
 186    float range_min = 0.0;
 187    float range_max = 1.0;
 188    float flip = -1.0;
 189    boolean degenerate = FALSE;
 190    boolean invertY = FALSE;
 191    enum pipe_error ret;
 192
 193    float fb_width = (float) svga->curr.framebuffer.width;
 194    float fb_height = (float) svga->curr.framebuffer.height;
 195
 196    float fx =        viewport->scale[0] * -1.0f + viewport->translate[0];
 197    float fy = flip * viewport->scale[1] * -1.0f + viewport->translate[1];
 198    float fw =        viewport->scale[0] * 2.0f;
 199    float fh = flip * viewport->scale[1] * 2.0f;
 200
 201    memset( &prescale, 0, sizeof(prescale) );
 202
 203    /* Examine gallium viewport transformation and produce a screen
 204     * rectangle and possibly vertex shader pre-transformation to
 205     * get the same results.
 206     */
 207
 208    SVGA_DBG(DEBUG_VIEWPORT,
 209             "\ninitial %f,%f %fx%f\n",
 210             fx,
 211             fy,
 212             fw,
 213             fh);
 214
 215    prescale.scale[0] = 1.0;
 216    prescale.scale[1] = 1.0;
 217    prescale.scale[2] = 1.0;
 218    prescale.scale[3] = 1.0;
 219    prescale.translate[0] = 0;
 220    prescale.translate[1] = 0;
 221    prescale.translate[2] = 0;
 222    prescale.translate[3] = 0;
 223    prescale.enabled = TRUE;
 224
 225
 226
 227    if (fw < 0) {
 228       prescale.scale[0] *= -1.0f;
 229       prescale.translate[0] += -fw;
 230       fw = -fw;
 231       fx =        viewport->scale[0] * 1.0f + viewport->translate[0];
 232    }
 233
 234    if (fh < 0.0) {
 235       prescale.translate[1] = fh - 1.0f + fy * 2.0f;
 236       fh = -fh;
 237       fy -= fh;
 238       prescale.scale[1] = -1.0f;
 239       invertY = TRUE;
 240    }
 241
 242    if (fx < 0) {
 243       prescale.translate[0] += fx;
 244       prescale.scale[0] *= fw / (fw + fx);
 245       fw += fx;
 246       fx = 0.0f;
 247    }
 248
 249    if (fy < 0) {
 250       if (invertY) {
 251          prescale.translate[1] -= fy;
 252       }
 253       else {
 254          prescale.translate[1] += fy;
 255       }
 256       prescale.scale[1] *= fh / (fh + fy);
 257       fh += fy;
 258       fy = 0.0f;
 259    }
 260
 261    if (fx + fw > fb_width) {
 262       prescale.scale[0] *= fw / (fb_width - fx);
 263       prescale.translate[0] -= fx * (fw / (fb_width - fx));
 264       prescale.translate[0] += fx;
 265       fw = fb_width - fx;
 266
 267    }
 268
 269    if (fy + fh > fb_height) {
 270       prescale.scale[1] *= fh / (fb_height - fy);
 271       if (invertY) {
 272          float in = fb_height - fy;       /* number of vp pixels inside view */
 273          float out = fy + fh - fb_height; /* number of vp pixels out of view */
 274          prescale.translate[1] += fy * out / in;
 275       }
 276       else {
 277          prescale.translate[1] -= fy * (fh / (fb_height - fy));
 278          prescale.translate[1] += fy;
 279       }
 280       fh = fb_height - fy;
 281    }
 282
 283    if (fw < 0 || fh < 0) {
 284       fw = fh = fx = fy = 0;
 285       degenerate = TRUE;
 286       goto out;
 287    }
 288
 289
 290    /* D3D viewport is integer space.  Convert fx,fy,etc. to
 291     * integers.
 292     *
 293     * TODO: adjust pretranslate correct for any subpixel error
 294     * introduced converting to integers.
 295     */
 296    rect.x = (uint32) fx;
 297    rect.y = (uint32) fy;
 298    rect.w = (uint32) fw;
 299    rect.h = (uint32) fh;
 300
 301    SVGA_DBG(DEBUG_VIEWPORT,
 302             "viewport error %f,%f %fx%f\n",
 303             fabs((float)rect.x - fx),
 304             fabs((float)rect.y - fy),
 305             fabs((float)rect.w - fw),
 306             fabs((float)rect.h - fh));
 307
 308    SVGA_DBG(DEBUG_VIEWPORT,
 309             "viewport %d,%d %dx%d\n",
 310             rect.x,
 311             rect.y,
 312             rect.w,
 313             rect.h);
 314
 315
 316    /* Finally, to get GL rasterization rules, need to tweak the
 317     * screen-space coordinates slightly relative to D3D which is
 318     * what hardware implements natively.
 319     */
 320    if (svga->curr.rast->templ.half_pixel_center) {
 321       float adjust_x = 0.0;
 322       float adjust_y = 0.0;
 323
 324       switch (svga->curr.reduced_prim) {
 325       case PIPE_PRIM_LINES:
 326          adjust_x = -0.5;
 327          adjust_y = 0;
 328          break;
 329       case PIPE_PRIM_POINTS:
 330       case PIPE_PRIM_TRIANGLES:
 331          adjust_x = -0.5;
 332          adjust_y = -0.5;
 333          break;
 334       }
 335
 336       if (invertY)
 337          adjust_y = -adjust_y;
 338
 339       prescale.translate[0] += adjust_x;
 340       prescale.translate[1] += adjust_y;
 341       prescale.translate[2] = 0.5; /* D3D clip space */
 342       prescale.scale[2]     = 0.5; /* D3D clip space */
 343    }
 344
 345
 346    range_min = viewport->scale[2] * -1.0f + viewport->translate[2];
 347    range_max = viewport->scale[2] *  1.0f + viewport->translate[2];
 348
 349    /* D3D (and by implication SVGA) doesn't like dealing with zmax
 350     * less than zmin.  Detect that case, flip the depth range and
 351     * invert our z-scale factor to achieve the same effect.
 352     */
 353    if (range_min > range_max) {
 354       float range_tmp;
 355       range_tmp = range_min;
 356       range_min = range_max;
 357       range_max = range_tmp;
 358       prescale.scale[2]     = -prescale.scale[2];
 359    }
 360
 361    if (prescale.enabled) {
 362       float H[2];
 363       float J[2];
 364       int i;
 365
 366       SVGA_DBG(DEBUG_VIEWPORT,
 367                "prescale %f,%f %fx%f\n",
 368                prescale.translate[0],
 369                prescale.translate[1],
 370                prescale.scale[0],
 371                prescale.scale[1]);
 372
 373       H[0] = (float)rect.w / 2.0f;
 374       H[1] = -(float)rect.h / 2.0f;
 375       J[0] = (float)rect.x + (float)rect.w / 2.0f;
 376       J[1] = (float)rect.y + (float)rect.h / 2.0f;
 377
 378       SVGA_DBG(DEBUG_VIEWPORT,
 379                "H %f,%f\n"
 380                "J %fx%f\n",
 381                H[0],
 382                H[1],
 383                J[0],
 384                J[1]);
 385
 386       /* Adjust prescale to take into account the fact that it is
 387        * going to be applied prior to the perspective divide and
 388        * viewport transformation.
 389        *
 390        * Vwin = H(Vc/Vc.w) + J
 391        *
 392        * We want to tweak Vwin with scale and translation from above,
 393        * as in:
 394        *
 395        * Vwin' = S Vwin + T
 396        *
 397        * But we can only modify the values at Vc.  Plugging all the
 398        * above together, and rearranging, eventually we get:
 399        *
 400        *   Vwin' = H(Vc'/Vc'.w) + J
 401        * where:
 402        *   Vc' = SVc + KVc.w
 403        *   K = (T + (S-1)J) / H
 404        *
 405        * Overwrite prescale.translate with values for K:
 406        */
 407       for (i = 0; i < 2; i++) {
 408          prescale.translate[i] = ((prescale.translate[i] +
 409                                    (prescale.scale[i] - 1.0f) * J[i]) / H[i]);
 410       }
 411
 412       SVGA_DBG(DEBUG_VIEWPORT,
 413                "clipspace %f,%f %fx%f\n",
 414                prescale.translate[0],
 415                prescale.translate[1],
 416                prescale.scale[0],
 417                prescale.scale[1]);
 418    }
 419
 420 out:
 421    if (degenerate) {
 422       rect.x = 0;
 423       rect.y = 0;
 424       rect.w = 1;
 425       rect.h = 1;
 426       prescale.enabled = FALSE;
 427    }
 428
 429    if (memcmp(&rect, &svga->state.hw_clear.viewport, sizeof(rect)) != 0) {
 430       ret = SVGA3D_SetViewport(svga->swc, &rect);
 431       if(ret != PIPE_OK)
 432          return ret;
 433
 434       memcpy(&svga->state.hw_clear.viewport, &rect, sizeof(rect));
 435       assert(sizeof(rect) == sizeof(svga->state.hw_clear.viewport));
 436    }
 437
 438    if (svga->state.hw_clear.depthrange.zmin != range_min ||
 439        svga->state.hw_clear.depthrange.zmax != range_max)
 440    {
 441       ret = SVGA3D_SetZRange(svga->swc, range_min, range_max );
 442       if(ret != PIPE_OK)
 443          return ret;
 444
 445       svga->state.hw_clear.depthrange.zmin = range_min;
 446       svga->state.hw_clear.depthrange.zmax = range_max;
 447    }
 448
 449    if (memcmp(&prescale, &svga->state.hw_clear.prescale, sizeof prescale) != 0) {
 450       svga->dirty |= SVGA_NEW_PRESCALE;
 451       svga->state.hw_clear.prescale = prescale;
 452    }
 453
 454    return PIPE_OK;
 455 }
 456
 457
 458 struct svga_tracked_state svga_hw_viewport =
 459 {
 460    "hw viewport state",
 461    ( SVGA_NEW_FRAME_BUFFER |
 462      SVGA_NEW_VIEWPORT |
 463      SVGA_NEW_RAST |
 464      SVGA_NEW_REDUCED_PRIMITIVE ),
 465    emit_viewport
 466 };
 467
 468
 469 /***********************************************************************
 470  * Scissor state
 471  */
 472 static enum pipe_error
 473 emit_scissor_rect( struct svga_context *svga,
 474                    unsigned dirty )
 475 {
 476    const struct pipe_scissor_state *scissor = &svga->curr.scissor;
 477    SVGA3dRect rect;
 478
 479    rect.x = scissor->minx;
 480    rect.y = scissor->miny;
 481    rect.w = scissor->maxx - scissor->minx; /* + 1 ?? */
 482    rect.h = scissor->maxy - scissor->miny; /* + 1 ?? */
 483
 484    return SVGA3D_SetScissorRect(svga->swc, &rect);
 485 }
 486
 487
 488 struct svga_tracked_state svga_hw_scissor =
 489 {
 490    "hw scissor state",
 491    SVGA_NEW_SCISSOR,
 492    emit_scissor_rect
 493 };
 494
 495
 496 /***********************************************************************
 497  * Userclip state
 498  */
 499
 500 static enum pipe_error
 501 emit_clip_planes( struct svga_context *svga,
 502                   unsigned dirty )
 503 {
 504    unsigned i;
 505    enum pipe_error ret;
 506
 507    /* TODO: just emit directly from svga_set_clip_state()?
 508     */
 509    for (i = 0; i < SVGA3D_MAX_CLIP_PLANES; i++) {
 510       /* need to express the plane in D3D-style coordinate space.
 511        * GL coords get converted to D3D coords with the matrix:
 512        * [ 1  0  0  0 ]
 513        * [ 0 -1  0  0 ]
 514        * [ 0  0  2  0 ]
 515        * [ 0  0 -1  1 ]
 516        * Apply that matrix to our plane equation, and invert Y.
 517        */
 518       float a = svga->curr.clip.ucp[i][0];
 519       float b = svga->curr.clip.ucp[i][1];
 520       float c = svga->curr.clip.ucp[i][2];
 521       float d = svga->curr.clip.ucp[i][3];
 522       float plane[4];
 523
 524       plane[0] = a;
 525       plane[1] = b;
 526       plane[2] = 2.0f * c;
 527       plane[3] = d - c;
 528
 529       ret = SVGA3D_SetClipPlane(svga->swc, i, plane);
 530       if(ret != PIPE_OK)
 531          return ret;
 532    }
 533
 534    return PIPE_OK;
 535 }
 536
 537
 538 struct svga_tracked_state svga_hw_clip_planes =
 539 {
 540    "hw viewport state",
 541    SVGA_NEW_CLIP,
 542    emit_clip_planes
 543 };