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