src/mesa/drivers/dri/r300/r300_draw.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2009 Maciej Cencora
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * 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, sub license, 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 portions
  15  * of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  18  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  19  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  20  * IN NO EVENT SHALL THE AUTHOR(S) AND/OR ITS SUPPLIERS BE LIABLE FOR
  21  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  22  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  23  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  24  *
  25  **************************************************************************/
  26
  27 #include <stdlib.h>
  28
  29 #include "main/glheader.h"
  30 #include "main/context.h"
  31 #include "main/state.h"
  32 #include "main/enums.h"
  33 #include "main/simple_list.h"
  34
  35 #include "r300_reg.h"
  36 #include "r300_context.h"
  37 #include "r300_emit.h"
  38 #include "r300_render.h"
  39 #include "r300_state.h"
  40 #include "r300_tex.h"
  41 #include "r300_cmdbuf.h"
  42
  43 #include "radeon_buffer_objects.h"
  44 #include "radeon_common_context.h"
  45
  46 #include "tnl/tnl.h"
  47 #include "tnl/t_vp_build.h"
  48 #include "vbo/vbo_context.h"
  49
  50
  51 static int getTypeSize(GLenum type)
  52 {
  53         switch (type) {
  54                 case GL_DOUBLE:
  55                         return sizeof(GLdouble);
  56                 case GL_HALF_FLOAT:
  57                         return sizeof(GLhalfARB);
  58                 case GL_FLOAT:
  59                         return sizeof(GLfloat);
  60                 case GL_INT:
  61                         return sizeof(GLint);
  62                 case GL_UNSIGNED_INT:
  63                         return sizeof(GLuint);
  64                 case GL_SHORT:
  65                         return sizeof(GLshort);
  66                 case GL_UNSIGNED_SHORT:
  67                         return sizeof(GLushort);
  68                 case GL_BYTE:
  69                         return sizeof(GLbyte);
  70                 case GL_UNSIGNED_BYTE:
  71                         return sizeof(GLubyte);
  72                 default:
  73                         assert(0);
  74                         return 0;
  75         }
  76 }
  77
  78 static void r300FixupIndexBuffer(struct gl_context *ctx, const struct _mesa_index_buffer *mesa_ind_buf)
  79 {
  80         r300ContextPtr r300 = R300_CONTEXT(ctx);
  81         GLvoid *src_ptr;
  82         GLuint *out;
  83         int i;
  84         GLboolean mapped_named_bo = GL_FALSE;
  85
  86         if (mesa_ind_buf->obj->Name && !mesa_ind_buf->obj->Pointer) {
  87                 ctx->Driver.MapBufferRange(ctx, 0, mesa_ind_buf->obj->Size,
  88                                            GL_MAP_READ_BIT, mesa_ind_buf->obj);
  89                 mapped_named_bo = GL_TRUE;
  90                 assert(mesa_ind_buf->obj->Pointer != NULL);
  91         }
  92         src_ptr = ADD_POINTERS(mesa_ind_buf->obj->Pointer, mesa_ind_buf->ptr);
  93
  94         radeon_print(RADEON_FALLBACKS, RADEON_IMPORTANT,
  95                         "%s: Fixing index buffer format. type %d\n",
  96                         __func__, mesa_ind_buf->type);
  97
  98         if (mesa_ind_buf->type == GL_UNSIGNED_BYTE) {
  99                 GLuint size = sizeof(GLushort) * ((mesa_ind_buf->count + 1) & ~1);
 100                 GLubyte *in = (GLubyte *)src_ptr;
 101
 102                 radeonAllocDmaRegion(&r300->radeon, &r300->ind_buf.bo, &r300->ind_buf.bo_offset, size, 4);
 103                 radeon_bo_map(r300->ind_buf.bo, 1);
 104                 assert(r300->ind_buf.bo->ptr != NULL);
 105                 out = (GLuint *)ADD_POINTERS(r300->ind_buf.bo->ptr, r300->ind_buf.bo_offset);
 106
 107                 for (i = 0; i + 1 < mesa_ind_buf->count; i += 2) {
 108                         *out++ = in[i] | in[i + 1] << 16;
 109                 }
 110
 111                 if (i < mesa_ind_buf->count) {
 112                         *out++ = in[i];
 113                 }
 114                 radeon_bo_unmap(r300->ind_buf.bo);
 115 #if MESA_BIG_ENDIAN
 116         } else { /* if (mesa_ind_buf->type == GL_UNSIGNED_SHORT) */
 117                 GLushort *in = (GLushort *)src_ptr;
 118                 GLuint size = sizeof(GLushort) * ((mesa_ind_buf->count + 1) & ~1);
 119
 120                 radeonAllocDmaRegion(&r300->radeon, &r300->ind_buf.bo,
 121                                      &r300->ind_buf.bo_offset, size, 4);
 122
 123                 radeon_bo_map(r300->ind_buf.bo, 1);
 124                 assert(r300->ind_buf.bo->ptr != NULL);
 125                 out = (GLuint *)ADD_POINTERS(r300->ind_buf.bo->ptr, r300->ind_buf.bo_offset);
 126
 127                 for (i = 0; i + 1 < mesa_ind_buf->count; i += 2) {
 128                         *out++ = in[i] | in[i + 1] << 16;
 129                 }
 130
 131                 if (i < mesa_ind_buf->count) {
 132                         *out++ = in[i];
 133                 }
 134                 radeon_bo_unmap(r300->ind_buf.bo);
 135 #endif
 136         }
 137
 138         r300->ind_buf.is_32bit = GL_FALSE;
 139         r300->ind_buf.count = mesa_ind_buf->count;
 140
 141         if (mapped_named_bo) {
 142                 ctx->Driver.UnmapBuffer(ctx, mesa_ind_buf->obj);
 143         }
 144 }
 145
 146
 147 static void r300SetupIndexBuffer(struct gl_context *ctx, const struct _mesa_index_buffer *mesa_ind_buf)
 148 {
 149         r300ContextPtr r300 = R300_CONTEXT(ctx);
 150
 151         if (!mesa_ind_buf) {
 152                 r300->ind_buf.bo = NULL;
 153                 return;
 154         }
 155         radeon_print(RADEON_RENDER, RADEON_TRACE, "%s\n", __func__);
 156
 157 #if MESA_BIG_ENDIAN
 158         if (mesa_ind_buf->type == GL_UNSIGNED_INT) {
 159 #else
 160         if (mesa_ind_buf->type != GL_UNSIGNED_BYTE) {
 161 #endif
 162                 const GLvoid *src_ptr;
 163                 GLvoid *dst_ptr;
 164                 GLboolean mapped_named_bo = GL_FALSE;
 165
 166                 if (mesa_ind_buf->obj->Name && !mesa_ind_buf->obj->Pointer) {
 167                         ctx->Driver.MapBufferRange(ctx, 0,
 168                                                    mesa_ind_buf->obj->Size,
 169                                                    GL_MAP_READ_BIT,
 170                                                    mesa_ind_buf->obj);
 171                         assert(mesa_ind_buf->obj->Pointer != NULL);
 172                         mapped_named_bo = GL_TRUE;
 173                 }
 174
 175                 src_ptr = ADD_POINTERS(mesa_ind_buf->obj->Pointer, mesa_ind_buf->ptr);
 176
 177                 const GLuint size = mesa_ind_buf->count * getTypeSize(mesa_ind_buf->type);
 178
 179                 radeonAllocDmaRegion(&r300->radeon, &r300->ind_buf.bo, &r300->ind_buf.bo_offset, size, 4);
 180
 181                 radeon_bo_map(r300->ind_buf.bo, 1);
 182                 assert(r300->ind_buf.bo->ptr != NULL);
 183                 dst_ptr = ADD_POINTERS(r300->ind_buf.bo->ptr, r300->ind_buf.bo_offset);
 184                 memcpy(dst_ptr, src_ptr, size);
 185
 186                 radeon_bo_unmap(r300->ind_buf.bo);
 187                 r300->ind_buf.is_32bit = (mesa_ind_buf->type == GL_UNSIGNED_INT);
 188                 r300->ind_buf.count = mesa_ind_buf->count;
 189
 190                 if (mapped_named_bo) {
 191                         ctx->Driver.UnmapBuffer(ctx, mesa_ind_buf->obj);
 192                 }
 193         } else {
 194                 r300FixupIndexBuffer(ctx, mesa_ind_buf);
 195         }
 196 }
 197
 198 #define CONVERT( TYPE, MACRO ) do {             \
 199         GLuint i, j, sz;                                \
 200         sz = input->Size;                               \
 201         if (input->Normalized) {                        \
 202                 for (i = 0; i < count; i++) {           \
 203                         const TYPE *in = (TYPE *)src_ptr;               \
 204                         for (j = 0; j < sz; j++) {              \
 205                                 *dst_ptr++ = MACRO(*in);                \
 206                                 in++;                           \
 207                         }                                       \
 208                         src_ptr += stride;                      \
 209                 }                                               \
 210         } else {                                        \
 211                 for (i = 0; i < count; i++) {           \
 212                         const TYPE *in = (TYPE *)src_ptr;               \
 213                         for (j = 0; j < sz; j++) {              \
 214                                 *dst_ptr++ = (GLfloat)(*in);            \
 215                                 in++;                           \
 216                         }                                       \
 217                         src_ptr += stride;                      \
 218                 }                                               \
 219         }                                               \
 220 } while (0)
 221
 222 /**
 223  * Convert attribute data type to float
 224  * If the attribute uses named buffer object replace the bo with newly allocated bo
 225  */
 226 static void r300ConvertAttrib(struct gl_context *ctx, int count, const struct gl_client_array *input, struct vertex_attribute *attr)
 227 {
 228         r300ContextPtr r300 = R300_CONTEXT(ctx);
 229         const GLvoid *src_ptr;
 230         GLboolean mapped_named_bo = GL_FALSE;
 231         GLfloat *dst_ptr;
 232         GLuint stride;
 233
 234         stride = (input->StrideB == 0) ? getTypeSize(input->Type) * input->Size : input->StrideB;
 235
 236         /* Convert value for first element only */
 237         if (input->StrideB == 0)
 238                 count = 1;
 239
 240         if (input->BufferObj->Name) {
 241                 if (!input->BufferObj->Pointer) {
 242                         ctx->Driver.MapBufferRange(ctx, 0, input->BufferObj->Size,
 243                                               GL_MAP_READ_BIT, input->BufferObj);
 244                         mapped_named_bo = GL_TRUE;
 245                 }
 246
 247                 src_ptr = ADD_POINTERS(input->BufferObj->Pointer, input->Ptr);
 248         } else {
 249                 src_ptr = input->Ptr;
 250         }
 251
 252         radeonAllocDmaRegion(&r300->radeon, &attr->bo, &attr->bo_offset, sizeof(GLfloat) * input->Size * count, 32);
 253         radeon_bo_map(attr->bo, 1);
 254         dst_ptr = (GLfloat *)ADD_POINTERS(attr->bo->ptr, attr->bo_offset);
 255
 256         radeon_print(RADEON_FALLBACKS, RADEON_IMPORTANT,
 257                         "%s: Converting vertex attributes, attribute data format %x,"
 258                         "stride %d, components %d\n"
 259                         , __FUNCTION__, input->Type
 260                         , stride, input->Size);
 261
 262         assert(src_ptr != NULL);
 263
 264         switch (input->Type) {
 265                 case GL_DOUBLE:
 266                         CONVERT(GLdouble, (GLfloat));
 267                         break;
 268                 case GL_UNSIGNED_INT:
 269                         CONVERT(GLuint, UINT_TO_FLOAT);
 270                         break;
 271                 case GL_INT:
 272                         CONVERT(GLint, INT_TO_FLOAT);
 273                         break;
 274                 case GL_UNSIGNED_SHORT:
 275                         CONVERT(GLushort, USHORT_TO_FLOAT);
 276                         break;
 277                 case GL_SHORT:
 278                         CONVERT(GLshort, SHORT_TO_FLOAT);
 279                         break;
 280                 case GL_UNSIGNED_BYTE:
 281                         assert(input->Format != GL_BGRA);
 282                         CONVERT(GLubyte, UBYTE_TO_FLOAT);
 283                         break;
 284                 case GL_BYTE:
 285                         CONVERT(GLbyte, BYTE_TO_FLOAT);
 286                         break;
 287                 default:
 288                         assert(0);
 289                         break;
 290         }
 291
 292         radeon_bo_unmap(attr->bo);
 293         if (mapped_named_bo) {
 294                 ctx->Driver.UnmapBuffer(ctx, input->BufferObj);
 295         }
 296 }
 297
 298 static void r300AlignDataToDword(struct gl_context *ctx, const struct gl_client_array *input, int count, struct vertex_attribute *attr)
 299 {
 300         r300ContextPtr r300 = R300_CONTEXT(ctx);
 301         const int dst_stride = (input->StrideB + 3) & ~3;
 302         const int size = getTypeSize(input->Type) * input->Size * count;
 303         GLboolean mapped_named_bo = GL_FALSE;
 304
 305         radeonAllocDmaRegion(&r300->radeon, &attr->bo, &attr->bo_offset, size, 32);
 306
 307         radeon_bo_map(attr->bo, 1);
 308
 309         if (!input->BufferObj->Pointer) {
 310                 ctx->Driver.MapBufferRange(ctx, 0, input->BufferObj->Size,
 311                                            GL_MAP_READ_BIT, input->BufferObj);
 312                 mapped_named_bo = GL_TRUE;
 313         }
 314
 315         radeon_print(RADEON_FALLBACKS, RADEON_IMPORTANT, "%s. Vertex alignment doesn't match hw requirements.\n", __func__);
 316
 317         {
 318                 GLvoid *src_ptr = ADD_POINTERS(input->BufferObj->Pointer, input->Ptr);
 319                 GLvoid *dst_ptr = ADD_POINTERS(attr->bo->ptr, attr->bo_offset);
 320                 int i;
 321
 322                 for (i = 0; i < count; ++i) {
 323                         memcpy(dst_ptr, src_ptr, input->StrideB);
 324                         src_ptr += input->StrideB;
 325                         dst_ptr += dst_stride;
 326                 }
 327         }
 328
 329         if (mapped_named_bo) {
 330                 ctx->Driver.UnmapBuffer(ctx, input->BufferObj);
 331         }
 332
 333         radeon_bo_unmap(attr->bo);
 334         attr->stride = dst_stride;
 335 }
 336
 337 static void r300TranslateAttrib(struct gl_context *ctx, GLuint attr, int count, const struct gl_client_array *input)
 338 {
 339         r300ContextPtr r300 = R300_CONTEXT(ctx);
 340         struct r300_vertex_buffer *vbuf = &r300->vbuf;
 341         struct vertex_attribute r300_attr = { 0 };
 342         GLenum type;
 343         GLuint stride;
 344
 345         radeon_print(RADEON_RENDER, RADEON_TRACE, "%s\n", __func__);
 346         stride = (input->StrideB == 0) ? getTypeSize(input->Type) * input->Size : input->StrideB;
 347
 348         if (input->Type == GL_DOUBLE || input->Type == GL_UNSIGNED_INT || input->Type == GL_INT ||
 349 #if MESA_BIG_ENDIAN
 350             getTypeSize(input->Type) != 4 ||
 351 #endif
 352             stride < 4) {
 353
 354                 type = GL_FLOAT;
 355
 356                 if (input->StrideB == 0) {
 357                         r300_attr.stride = 0;
 358                 } else {
 359                         r300_attr.stride = sizeof(GLfloat) * input->Size;
 360                 }
 361                 r300_attr.dwords = input->Size;
 362                 r300_attr.is_named_bo = GL_FALSE;
 363         } else {
 364                 type = input->Type;
 365                 r300_attr.dwords = (getTypeSize(type) * input->Size + 3)/ 4;
 366                 if (!input->BufferObj->Name) {
 367
 368                         if (input->StrideB == 0) {
 369                                 r300_attr.stride = 0;
 370                         } else {
 371                                 r300_attr.stride = (getTypeSize(type) * input->Size + 3) & ~3;
 372                         }
 373
 374                         r300_attr.is_named_bo = GL_FALSE;
 375                 }
 376         }
 377
 378         r300_attr.size = input->Size;
 379         r300_attr.element = attr;
 380         r300_attr.dst_loc = vbuf->num_attribs;
 381
 382         switch (type) {
 383                 case GL_FLOAT:
 384                         switch (input->Size) {
 385                                 case 1: r300_attr.data_type = R300_DATA_TYPE_FLOAT_1; break;
 386                                 case 2: r300_attr.data_type = R300_DATA_TYPE_FLOAT_2; break;
 387                                 case 3: r300_attr.data_type = R300_DATA_TYPE_FLOAT_3; break;
 388                                 case 4: r300_attr.data_type = R300_DATA_TYPE_FLOAT_4; break;
 389                         }
 390                         r300_attr._signed = 0;
 391                         r300_attr.normalize = 0;
 392                         break;
 393                 case GL_HALF_FLOAT:
 394                         switch (input->Size) {
 395                                 case 1:
 396                                 case 2:
 397                                         r300_attr.data_type = R300_DATA_TYPE_FLT16_2;
 398                                         break;
 399                                 case 3:
 400                                 case 4:
 401                                         r300_attr.data_type = R300_DATA_TYPE_FLT16_4;
 402                                         break;
 403                         }
 404                         break;
 405                 case GL_SHORT:
 406                         r300_attr._signed = 1;
 407                         r300_attr.normalize = input->Normalized;
 408                         switch (input->Size) {
 409                                 case 1:
 410                                 case 2:
 411                                         r300_attr.data_type = R300_DATA_TYPE_SHORT_2;
 412                                         break;
 413                                 case 3:
 414                                 case 4:
 415                                         r300_attr.data_type = R300_DATA_TYPE_SHORT_4;
 416                                         break;
 417                         }
 418                         break;
 419                 case GL_BYTE:
 420                         r300_attr._signed = 1;
 421                         r300_attr.normalize = input->Normalized;
 422                         r300_attr.data_type = R300_DATA_TYPE_BYTE;
 423                         break;
 424                 case GL_UNSIGNED_SHORT:
 425                         r300_attr._signed = 0;
 426                         r300_attr.normalize = input->Normalized;
 427                         switch (input->Size) {
 428                                 case 1:
 429                                 case 2:
 430                                         r300_attr.data_type = R300_DATA_TYPE_SHORT_2;
 431                                         break;
 432                                 case 3:
 433                                 case 4:
 434                                         r300_attr.data_type = R300_DATA_TYPE_SHORT_4;
 435                                         break;
 436                         }
 437                         break;
 438                 case GL_UNSIGNED_BYTE:
 439                         r300_attr._signed = 0;
 440                         r300_attr.normalize = input->Normalized;
 441                         if (input->Format == GL_BGRA)
 442                                 r300_attr.data_type = R300_DATA_TYPE_D3DCOLOR;
 443                         else
 444                                 r300_attr.data_type = R300_DATA_TYPE_BYTE;
 445                         break;
 446
 447                 default:
 448                 case GL_DOUBLE:
 449                 case GL_INT:
 450                 case GL_UNSIGNED_INT:
 451                         assert(0);
 452                         break;
 453         }
 454
 455         switch (input->Size) {
 456                 case 4:
 457                         r300_attr.swizzle = SWIZZLE_XYZW;
 458                         break;
 459                 case 3:
 460                         r300_attr.swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ONE);
 461                         break;
 462                 case 2:
 463                         r300_attr.swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_ZERO, SWIZZLE_ONE);
 464                         break;
 465                 case 1:
 466                         r300_attr.swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ONE);
 467                         break;
 468         }
 469
 470         r300_attr.write_mask = MASK_XYZW;
 471
 472         vbuf->attribs[vbuf->num_attribs] = r300_attr;
 473         ++vbuf->num_attribs;
 474 }
 475
 476 static void r300SetVertexFormat(struct gl_context *ctx, const struct gl_client_array *arrays[], int count)
 477 {
 478         r300ContextPtr r300 = R300_CONTEXT(ctx);
 479         struct r300_vertex_buffer *vbuf = &r300->vbuf;
 480         radeon_print(RADEON_RENDER, RADEON_VERBOSE, "%s\n", __func__);
 481         {
 482                 int i, tmp;
 483
 484                 tmp = r300->selected_vp->code.InputsRead;
 485                 i = 0;
 486                 vbuf->num_attribs = 0;
 487                 while (tmp) {
 488                         /* find first enabled bit */
 489                         while (!(tmp & 1)) {
 490                                 tmp >>= 1;
 491                                 ++i;
 492                         }
 493
 494                         r300TranslateAttrib(ctx, i, count, arrays[i]);
 495
 496                         tmp >>= 1;
 497                         ++i;
 498                 }
 499         }
 500
 501         r300SwitchFallback(ctx, R300_FALLBACK_AOS_LIMIT, vbuf->num_attribs > R300_MAX_AOS_ARRAYS);
 502         if (r300->fallback)
 503                 return;
 504 }
 505
 506 static void r300AllocDmaRegions(struct gl_context *ctx, const struct gl_client_array *input[], int count)
 507 {
 508         r300ContextPtr r300 = R300_CONTEXT(ctx);
 509         struct r300_vertex_buffer *vbuf = &r300->vbuf;
 510         GLuint stride;
 511         int ret;
 512         int i, index;
 513         radeon_print(RADEON_RENDER, RADEON_VERBOSE,
 514                         "%s: count %d num_attribs %d\n",
 515                         __func__, count, vbuf->num_attribs);
 516
 517         for (index = 0; index < vbuf->num_attribs; index++) {
 518                 struct radeon_aos *aos = &r300->radeon.tcl.aos[index];
 519                 i = vbuf->attribs[index].element;
 520
 521                 stride = (input[i]->StrideB == 0) ? getTypeSize(input[i]->Type) * input[i]->Size : input[i]->StrideB;
 522
 523                 if (input[i]->Type == GL_DOUBLE || input[i]->Type == GL_UNSIGNED_INT || input[i]->Type == GL_INT ||
 524 #if MESA_BIG_ENDIAN
 525                                 getTypeSize(input[i]->Type) != 4 ||
 526 #endif
 527                                 stride < 4) {
 528
 529                         r300ConvertAttrib(ctx, count, input[i], &vbuf->attribs[index]);
 530                 } else {
 531                         if (input[i]->BufferObj->Name) {
 532                                 if (stride % 4 != 0 || (intptr_t)input[i]->Ptr % 4 != 0) {
 533                                         r300AlignDataToDword(ctx, input[i], count, &vbuf->attribs[index]);
 534                                         vbuf->attribs[index].is_named_bo = GL_FALSE;
 535                                 } else {
 536                                         vbuf->attribs[index].stride = input[i]->StrideB;
 537                                         vbuf->attribs[index].bo_offset = (intptr_t) input[i]->Ptr;
 538                                         vbuf->attribs[index].bo = get_radeon_buffer_object(input[i]->BufferObj)->bo;
 539                                         vbuf->attribs[index].is_named_bo = GL_TRUE;
 540                                 }
 541                         } else {
 542
 543                                 int size;
 544                                 int local_count = count;
 545                                 uint32_t *dst;
 546
 547                                 if (input[i]->StrideB == 0) {
 548                                         size = getTypeSize(input[i]->Type) * input[i]->Size;
 549                                         local_count = 1;
 550                                 } else {
 551                                         size = getTypeSize(input[i]->Type) * input[i]->Size * local_count;
 552                                 }
 553
 554                                 radeonAllocDmaRegion(&r300->radeon, &vbuf->attribs[index].bo, &vbuf->attribs[index].bo_offset, size, 32);
 555                                 radeon_bo_map(vbuf->attribs[index].bo, 1);
 556                                 assert(vbuf->attribs[index].bo->ptr != NULL);
 557                                 dst = (uint32_t *)ADD_POINTERS(vbuf->attribs[index].bo->ptr, vbuf->attribs[index].bo_offset);
 558                                 switch (vbuf->attribs[index].dwords) {
 559                                         case 1: radeonEmitVec4(dst, input[i]->Ptr, input[i]->StrideB, local_count); break;
 560                                         case 2: radeonEmitVec8(dst, input[i]->Ptr, input[i]->StrideB, local_count); break;
 561                                         case 3: radeonEmitVec12(dst, input[i]->Ptr, input[i]->StrideB, local_count); break;
 562                                         case 4: radeonEmitVec16(dst, input[i]->Ptr, input[i]->StrideB, local_count); break;
 563                                         default: assert(0); break;
 564                                 }
 565                                 radeon_bo_unmap(vbuf->attribs[index].bo);
 566
 567                         }
 568                 }
 569
 570                 aos->count = vbuf->attribs[index].stride == 0 ? 1 : count;
 571                 aos->stride = vbuf->attribs[index].stride / sizeof(float);
 572                 aos->components = vbuf->attribs[index].dwords;
 573                 aos->bo = vbuf->attribs[index].bo;
 574                 aos->offset = vbuf->attribs[index].bo_offset;
 575
 576                 if (vbuf->attribs[index].is_named_bo) {
 577                         radeon_cs_space_add_persistent_bo(r300->radeon.cmdbuf.cs, r300->vbuf.attribs[index].bo, RADEON_GEM_DOMAIN_GTT, 0);
 578                 }
 579         }
 580
 581         r300->radeon.tcl.aos_count = vbuf->num_attribs;
 582         ret = radeon_cs_space_check_with_bo(r300->radeon.cmdbuf.cs, first_elem(&r300->radeon.dma.reserved)->bo, RADEON_GEM_DOMAIN_GTT, 0);
 583         r300SwitchFallback(ctx, R300_FALLBACK_INVALID_BUFFERS, ret);
 584
 585 }
 586
 587 static void r300FreeData(struct gl_context *ctx)
 588 {
 589         /* Need to zero tcl.aos[n].bo and tcl.elt_dma_bo
 590          * to prevent double unref in radeonReleaseArrays
 591          * called during context destroy
 592          */
 593         radeon_print(RADEON_RENDER, RADEON_VERBOSE, "%s\n", __func__);
 594         r300ContextPtr r300 = R300_CONTEXT(ctx);
 595         {
 596                 int i;
 597
 598                 for (i = 0; i < r300->vbuf.num_attribs; i++) {
 599                         if (!r300->vbuf.attribs[i].is_named_bo) {
 600                                 radeon_bo_unref(r300->vbuf.attribs[i].bo);
 601                         }
 602                         r300->radeon.tcl.aos[i].bo = NULL;
 603                 }
 604         }
 605
 606         {
 607                 if (r300->ind_buf.bo != NULL) {
 608                         radeon_bo_unref(r300->ind_buf.bo);
 609                 }
 610         }
 611 }
 612
 613 static GLuint r300PredictTryDrawPrimsSize(struct gl_context *ctx,
 614                 GLuint nr_prims, const struct _mesa_prim *prim)
 615 {
 616         struct r300_context *r300 = R300_CONTEXT(ctx);
 617         struct r300_vertex_buffer *vbuf = &r300->vbuf;
 618         GLboolean flushed;
 619         GLuint dwords;
 620         GLuint state_size;
 621         int i;
 622         GLuint extra_prims = 0;
 623
 624         /* Check for primitive splitting. */
 625         for (i = 0; i < nr_prims; ++i) {
 626                 const GLuint num_verts =  r300NumVerts(r300, prim[i].count, prim[i].mode);
 627                 extra_prims += num_verts/(65535 - 32);
 628         }
 629         nr_prims += extra_prims;
 630
 631         dwords = 2*CACHE_FLUSH_BUFSZ;
 632         dwords += PRE_EMIT_STATE_BUFSZ;
 633         dwords += (AOS_BUFSZ(vbuf->num_attribs)
 634                 + SCISSORS_BUFSZ*2
 635                 + FIREAOS_BUFSZ )*nr_prims;
 636
 637         state_size = radeonCountStateEmitSize(&r300->radeon);
 638         flushed = rcommonEnsureCmdBufSpace(&r300->radeon,
 639                         dwords + state_size,
 640                         __FUNCTION__);
 641         if (flushed)
 642                 dwords += radeonCountStateEmitSize(&r300->radeon);
 643         else
 644                 dwords += state_size;
 645
 646         radeon_print(RADEON_RENDER, RADEON_VERBOSE, "%s: total prediction size is %d.\n", __FUNCTION__, dwords);
 647         return dwords;
 648 }
 649
 650 static GLboolean r300TryDrawPrims(struct gl_context *ctx,
 651                                          const struct gl_client_array *arrays[],
 652                                          const struct _mesa_prim *prim,
 653                                          GLuint nr_prims,
 654                                          const struct _mesa_index_buffer *ib,
 655                                          GLuint min_index,
 656                                          GLuint max_index )
 657 {
 658         struct r300_context *r300 = R300_CONTEXT(ctx);
 659         GLuint i;
 660
 661         radeon_print(RADEON_RENDER, RADEON_NORMAL, "%s: %u (%d-%d) cs begin at %d\n",
 662                                 __FUNCTION__, nr_prims, min_index, max_index, r300->radeon.cmdbuf.cs->cdw );
 663
 664         if (ctx->NewState)
 665                 _mesa_update_state( ctx );
 666
 667         if (r300->options.hw_tcl_enabled)
 668                 _tnl_UpdateFixedFunctionProgram(ctx);
 669
 670         r300UpdateShaders(r300);
 671
 672         r300SwitchFallback(ctx, R300_FALLBACK_INVALID_BUFFERS, !r300ValidateBuffers(ctx));
 673
 674         r300SetVertexFormat(ctx, arrays, max_index + 1);
 675
 676         if (r300->fallback)
 677                 return GL_FALSE;
 678
 679         r300SetupVAP(ctx, r300->selected_vp->code.InputsRead, r300->selected_vp->code.OutputsWritten);
 680
 681         r300UpdateShaderStates(r300);
 682
 683         /* ensure we have the cmd buf space in advance to cover
 684          * the state + DMA AOS pointers */
 685         GLuint emit_end = r300PredictTryDrawPrimsSize(ctx, nr_prims, prim)
 686                 + r300->radeon.cmdbuf.cs->cdw;
 687
 688         r300SetupIndexBuffer(ctx, ib);
 689
 690         r300AllocDmaRegions(ctx, arrays, max_index + 1);
 691
 692         if (r300->fallback)
 693                 return GL_FALSE;
 694
 695         r300EmitCacheFlush(r300);
 696         radeonEmitState(&r300->radeon);
 697
 698         for (i = 0; i < nr_prims; ++i) {
 699                 r300RunRenderPrimitive(ctx, prim[i].start, prim[i].start + prim[i].count, prim[i].mode);
 700         }
 701
 702         r300EmitCacheFlush(r300);
 703
 704         r300FreeData(ctx);
 705
 706         radeon_print(RADEON_RENDER, RADEON_VERBOSE, "%s: %u (%d-%d) cs ending at %d\n",
 707                         __FUNCTION__, nr_prims, min_index, max_index, r300->radeon.cmdbuf.cs->cdw );
 708
 709         if (emit_end < r300->radeon.cmdbuf.cs->cdw)
 710                 WARN_ONCE("Rendering was %d commands larger than predicted size."
 711                                 " We might overflow  command buffer.\n", r300->radeon.cmdbuf.cs->cdw - emit_end);
 712
 713         return GL_TRUE;
 714 }
 715
 716 static void r300DrawPrims(struct gl_context *ctx,
 717                          const struct gl_client_array *arrays[],
 718                          const struct _mesa_prim *prim,
 719                          GLuint nr_prims,
 720                          const struct _mesa_index_buffer *ib,
 721                          GLboolean index_bounds_valid,
 722                          GLuint min_index,
 723                          GLuint max_index)
 724 {
 725         GLboolean retval;
 726         struct r300_context *r300 = R300_CONTEXT(ctx);
 727         radeonContextPtr radeon = &r300->radeon;
 728
 729         radeon_prepare_render(radeon);
 730
 731         /* This check should get folded into just the places that
 732          * min/max index are really needed.
 733          */
 734         if (!index_bounds_valid) {
 735                 vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index);
 736         }
 737
 738         if (min_index) {
 739                 radeon_print(RADEON_FALLBACKS, RADEON_IMPORTANT,
 740                                 "%s: Rebasing primitives. %p nr_prims %d min_index %u max_index %u\n",
 741                                 __func__, prim, nr_prims, min_index, max_index);
 742                 vbo_rebase_prims( ctx, arrays, prim, nr_prims, ib, min_index, max_index, r300DrawPrims );
 743                 return;
 744         }
 745
 746         /* Make an attempt at drawing */
 747         retval = r300TryDrawPrims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 748
 749         /* If failed run tnl pipeline - it should take care of fallbacks */
 750         if (!retval)
 751                 _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 752 }
 753
 754 void r300InitDraw(struct gl_context *ctx)
 755 {
 756         struct vbo_context *vbo = vbo_context(ctx);
 757
 758         vbo->draw_prims = r300DrawPrims;
 759 }