src/gallium/drivers/svga/svga_draw.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 "pipe/p_compiler.h"
  27 #include "util/u_inlines.h"
  28 #include "pipe/p_defines.h"
  29 #include "util/u_memory.h"
  30 #include "util/u_math.h"
  31 #include "util/u_upload_mgr.h"
  32
  33 #include "svga_context.h"
  34 #include "svga_draw.h"
  35 #include "svga_draw_private.h"
  36 #include "svga_debug.h"
  37 #include "svga_screen.h"
  38 #include "svga_resource_buffer.h"
  39 #include "svga_resource_texture.h"
  40 #include "svga_surface.h"
  41 #include "svga_winsys.h"
  42 #include "svga_cmd.h"
  43
  44
  45 struct svga_hwtnl *
  46 svga_hwtnl_create(struct svga_context *svga)
  47 {
  48    struct svga_hwtnl *hwtnl = CALLOC_STRUCT(svga_hwtnl);
  49    if (hwtnl == NULL)
  50       goto fail;
  51
  52    hwtnl->svga = svga;
  53
  54    hwtnl->cmd.swc = svga->swc;
  55
  56    return hwtnl;
  57
  58 fail:
  59    return NULL;
  60 }
  61
  62
  63 void
  64 svga_hwtnl_destroy(struct svga_hwtnl *hwtnl)
  65 {
  66    unsigned i, j;
  67
  68    for (i = 0; i < PIPE_PRIM_MAX; i++) {
  69       for (j = 0; j < IDX_CACHE_MAX; j++) {
  70          pipe_resource_reference(&hwtnl->index_cache[i][j].buffer, NULL);
  71       }
  72    }
  73
  74    for (i = 0; i < hwtnl->cmd.vdecl_count; i++)
  75       pipe_resource_reference(&hwtnl->cmd.vdecl_vb[i], NULL);
  76
  77    for (i = 0; i < hwtnl->cmd.prim_count; i++)
  78       pipe_resource_reference(&hwtnl->cmd.prim_ib[i], NULL);
  79
  80    FREE(hwtnl);
  81 }
  82
  83
  84 void
  85 svga_hwtnl_set_flatshade(struct svga_hwtnl *hwtnl,
  86                          boolean flatshade, boolean flatshade_first)
  87 {
  88    hwtnl->hw_pv = PV_FIRST;
  89    hwtnl->api_pv = (flatshade && !flatshade_first) ? PV_LAST : PV_FIRST;
  90 }
  91
  92
  93 void
  94 svga_hwtnl_set_unfilled(struct svga_hwtnl *hwtnl, unsigned mode)
  95 {
  96    hwtnl->api_fillmode = mode;
  97 }
  98
  99
 100 void
 101 svga_hwtnl_reset_vdecl(struct svga_hwtnl *hwtnl, unsigned count)
 102 {
 103    unsigned i;
 104
 105    assert(hwtnl->cmd.prim_count == 0);
 106
 107    for (i = count; i < hwtnl->cmd.vdecl_count; i++) {
 108       pipe_resource_reference(&hwtnl->cmd.vdecl_vb[i], NULL);
 109    }
 110
 111    hwtnl->cmd.vdecl_count = count;
 112 }
 113
 114
 115 void
 116 svga_hwtnl_vdecl(struct svga_hwtnl *hwtnl,
 117                  unsigned i,
 118                  const SVGA3dVertexDecl * decl, struct pipe_resource *vb)
 119 {
 120    assert(hwtnl->cmd.prim_count == 0);
 121
 122    assert(i < hwtnl->cmd.vdecl_count);
 123
 124    hwtnl->cmd.vdecl[i] = *decl;
 125
 126    pipe_resource_reference(&hwtnl->cmd.vdecl_vb[i], vb);
 127 }
 128
 129
 130 /**
 131  * Determine whether the specified buffer is referred in the primitive queue,
 132  * for which no commands have been written yet.
 133  */
 134 boolean
 135 svga_hwtnl_is_buffer_referred(struct svga_hwtnl *hwtnl,
 136                               struct pipe_resource *buffer)
 137 {
 138    unsigned i;
 139
 140    if (svga_buffer_is_user_buffer(buffer)) {
 141       return FALSE;
 142    }
 143
 144    if (!hwtnl->cmd.prim_count) {
 145       return FALSE;
 146    }
 147
 148    for (i = 0; i < hwtnl->cmd.vdecl_count; ++i) {
 149       if (hwtnl->cmd.vdecl_vb[i] == buffer) {
 150          return TRUE;
 151       }
 152    }
 153
 154    for (i = 0; i < hwtnl->cmd.prim_count; ++i) {
 155       if (hwtnl->cmd.prim_ib[i] == buffer) {
 156          return TRUE;
 157       }
 158    }
 159
 160    return FALSE;
 161 }
 162
 163
 164 enum pipe_error
 165 svga_hwtnl_flush(struct svga_hwtnl *hwtnl)
 166 {
 167    struct svga_winsys_context *swc = hwtnl->cmd.swc;
 168    struct svga_context *svga = hwtnl->svga;
 169    enum pipe_error ret;
 170
 171    if (hwtnl->cmd.prim_count) {
 172       struct svga_winsys_surface *vb_handle[SVGA3D_INPUTREG_MAX];
 173       struct svga_winsys_surface *ib_handle[QSZ];
 174       struct svga_winsys_surface *handle;
 175       SVGA3dVertexDecl *vdecl;
 176       SVGA3dPrimitiveRange *prim;
 177       unsigned i;
 178
 179       for (i = 0; i < hwtnl->cmd.vdecl_count; i++) {
 180          assert(!svga_buffer_is_user_buffer(hwtnl->cmd.vdecl_vb[i]));
 181          handle = svga_buffer_handle(svga, hwtnl->cmd.vdecl_vb[i]);
 182          if (handle == NULL)
 183             return PIPE_ERROR_OUT_OF_MEMORY;
 184
 185          vb_handle[i] = handle;
 186       }
 187
 188       for (i = 0; i < hwtnl->cmd.prim_count; i++) {
 189          if (hwtnl->cmd.prim_ib[i]) {
 190             assert(!svga_buffer_is_user_buffer(hwtnl->cmd.prim_ib[i]));
 191             handle = svga_buffer_handle(svga, hwtnl->cmd.prim_ib[i]);
 192             if (handle == NULL)
 193                return PIPE_ERROR_OUT_OF_MEMORY;
 194          }
 195          else {
 196             handle = NULL;
 197          }
 198
 199          ib_handle[i] = handle;
 200       }
 201
 202       if (svga->rebind.rendertargets) {
 203          ret = svga_reemit_framebuffer_bindings(svga);
 204          if (ret != PIPE_OK) {
 205             return ret;
 206          }
 207       }
 208
 209       if (svga->rebind.texture_samplers) {
 210          ret = svga_reemit_tss_bindings(svga);
 211          if (ret != PIPE_OK) {
 212             return ret;
 213          }
 214       }
 215
 216       SVGA_DBG(DEBUG_DMA, "draw to sid %p, %d prims\n",
 217                svga->curr.framebuffer.cbufs[0] ?
 218                svga_surface(svga->curr.framebuffer.cbufs[0])->handle : NULL,
 219                hwtnl->cmd.prim_count);
 220
 221       ret = SVGA3D_BeginDrawPrimitives(swc, &vdecl, hwtnl->cmd.vdecl_count,
 222                                        &prim, hwtnl->cmd.prim_count);
 223       if (ret != PIPE_OK)
 224          return ret;
 225
 226       memcpy(vdecl, hwtnl->cmd.vdecl,
 227              hwtnl->cmd.vdecl_count * sizeof hwtnl->cmd.vdecl[0]);
 228
 229       for (i = 0; i < hwtnl->cmd.vdecl_count; i++) {
 230          /* Given rangeHint is considered to be relative to indexBias, and
 231           * indexBias varies per primitive, we cannot accurately supply an
 232           * rangeHint when emitting more than one primitive per draw command.
 233           */
 234          if (hwtnl->cmd.prim_count == 1) {
 235             vdecl[i].rangeHint.first = hwtnl->cmd.min_index[0];
 236             vdecl[i].rangeHint.last = hwtnl->cmd.max_index[0] + 1;
 237          }
 238          else {
 239             vdecl[i].rangeHint.first = 0;
 240             vdecl[i].rangeHint.last = 0;
 241          }
 242
 243          swc->surface_relocation(swc, &vdecl[i].array.surfaceId, NULL,
 244                                  vb_handle[i], SVGA_RELOC_READ);
 245       }
 246
 247       memcpy(prim, hwtnl->cmd.prim,
 248              hwtnl->cmd.prim_count * sizeof hwtnl->cmd.prim[0]);
 249
 250       for (i = 0; i < hwtnl->cmd.prim_count; i++) {
 251          swc->surface_relocation(swc, &prim[i].indexArray.surfaceId, NULL,
 252                                  ib_handle[i], SVGA_RELOC_READ);
 253          pipe_resource_reference(&hwtnl->cmd.prim_ib[i], NULL);
 254       }
 255
 256       SVGA_FIFOCommitAll(swc);
 257       hwtnl->cmd.prim_count = 0;
 258    }
 259
 260    return PIPE_OK;
 261 }
 262
 263
 264 void
 265 svga_hwtnl_set_index_bias(struct svga_hwtnl *hwtnl, int index_bias)
 266 {
 267    hwtnl->index_bias = index_bias;
 268 }
 269
 270
 271
 272 /***********************************************************************
 273  * Internal functions:
 274  */
 275
 276 enum pipe_error
 277 svga_hwtnl_prim(struct svga_hwtnl *hwtnl,
 278                 const SVGA3dPrimitiveRange * range,
 279                 unsigned min_index,
 280                 unsigned max_index, struct pipe_resource *ib)
 281 {
 282    enum pipe_error ret = PIPE_OK;
 283
 284 #ifdef DEBUG
 285    {
 286       unsigned i;
 287       for (i = 0; i < hwtnl->cmd.vdecl_count; i++) {
 288          struct pipe_resource *vb = hwtnl->cmd.vdecl_vb[i];
 289          unsigned size = vb ? vb->width0 : 0;
 290          unsigned offset = hwtnl->cmd.vdecl[i].array.offset;
 291          unsigned stride = hwtnl->cmd.vdecl[i].array.stride;
 292          int index_bias = (int) range->indexBias + hwtnl->index_bias;
 293          unsigned width;
 294
 295          assert(vb);
 296          assert(size);
 297          assert(offset < size);
 298          assert(min_index <= max_index);
 299
 300          switch (hwtnl->cmd.vdecl[i].identity.type) {
 301          case SVGA3D_DECLTYPE_FLOAT1:
 302             width = 4;
 303             break;
 304          case SVGA3D_DECLTYPE_FLOAT2:
 305             width = 4 * 2;
 306             break;
 307          case SVGA3D_DECLTYPE_FLOAT3:
 308             width = 4 * 3;
 309             break;
 310          case SVGA3D_DECLTYPE_FLOAT4:
 311             width = 4 * 4;
 312             break;
 313          case SVGA3D_DECLTYPE_D3DCOLOR:
 314             width = 4;
 315             break;
 316          case SVGA3D_DECLTYPE_UBYTE4:
 317             width = 1 * 4;
 318             break;
 319          case SVGA3D_DECLTYPE_SHORT2:
 320             width = 2 * 2;
 321             break;
 322          case SVGA3D_DECLTYPE_SHORT4:
 323             width = 2 * 4;
 324             break;
 325          case SVGA3D_DECLTYPE_UBYTE4N:
 326             width = 1 * 4;
 327             break;
 328          case SVGA3D_DECLTYPE_SHORT2N:
 329             width = 2 * 2;
 330             break;
 331          case SVGA3D_DECLTYPE_SHORT4N:
 332             width = 2 * 4;
 333             break;
 334          case SVGA3D_DECLTYPE_USHORT2N:
 335             width = 2 * 2;
 336             break;
 337          case SVGA3D_DECLTYPE_USHORT4N:
 338             width = 2 * 4;
 339             break;
 340          case SVGA3D_DECLTYPE_UDEC3:
 341             width = 4;
 342             break;
 343          case SVGA3D_DECLTYPE_DEC3N:
 344             width = 4;
 345             break;
 346          case SVGA3D_DECLTYPE_FLOAT16_2:
 347             width = 2 * 2;
 348             break;
 349          case SVGA3D_DECLTYPE_FLOAT16_4:
 350             width = 2 * 4;
 351             break;
 352          default:
 353             assert(0);
 354             width = 0;
 355             break;
 356          }
 357
 358          if (index_bias >= 0) {
 359             assert(offset + index_bias * stride + width <= size);
 360          }
 361
 362          /*
 363           * min_index/max_index are merely conservative guesses, so we can't
 364           * make buffer overflow detection based on their values.
 365           */
 366       }
 367
 368       assert(range->indexWidth == range->indexArray.stride);
 369
 370       if (ib) {
 371          unsigned size = ib->width0;
 372          unsigned offset = range->indexArray.offset;
 373          unsigned stride = range->indexArray.stride;
 374          unsigned count;
 375
 376          assert(size);
 377          assert(offset < size);
 378          assert(stride);
 379
 380          switch (range->primType) {
 381          case SVGA3D_PRIMITIVE_POINTLIST:
 382             count = range->primitiveCount;
 383             break;
 384          case SVGA3D_PRIMITIVE_LINELIST:
 385             count = range->primitiveCount * 2;
 386             break;
 387          case SVGA3D_PRIMITIVE_LINESTRIP:
 388             count = range->primitiveCount + 1;
 389             break;
 390          case SVGA3D_PRIMITIVE_TRIANGLELIST:
 391             count = range->primitiveCount * 3;
 392             break;
 393          case SVGA3D_PRIMITIVE_TRIANGLESTRIP:
 394             count = range->primitiveCount + 2;
 395             break;
 396          case SVGA3D_PRIMITIVE_TRIANGLEFAN:
 397             count = range->primitiveCount + 2;
 398             break;
 399          default:
 400             assert(0);
 401             count = 0;
 402             break;
 403          }
 404
 405          assert(offset + count * stride <= size);
 406       }
 407    }
 408 #endif
 409
 410    if (hwtnl->cmd.prim_count + 1 >= QSZ) {
 411       ret = svga_hwtnl_flush(hwtnl);
 412       if (ret != PIPE_OK)
 413          return ret;
 414    }
 415
 416    /* min/max indices are relative to bias */
 417    hwtnl->cmd.min_index[hwtnl->cmd.prim_count] = min_index;
 418    hwtnl->cmd.max_index[hwtnl->cmd.prim_count] = max_index;
 419
 420    hwtnl->cmd.prim[hwtnl->cmd.prim_count] = *range;
 421    hwtnl->cmd.prim[hwtnl->cmd.prim_count].indexBias += hwtnl->index_bias;
 422
 423    pipe_resource_reference(&hwtnl->cmd.prim_ib[hwtnl->cmd.prim_count], ib);
 424    hwtnl->cmd.prim_count++;
 425
 426    return ret;
 427 }