Mesa-git.patch

   1 diff --git a/configs/default b/configs/default
   2 index 2ca6fe4..a4069cb 100644
   3 --- a/configs/default
   4 +++ b/configs/default
   5 @@ -10,7 +10,7 @@ CONFIG_NAME = default
   6  # Version info
   7  MESA_MAJOR=8
   8  MESA_MINOR=0
   9 -MESA_TINY=0
  10 +MESA_TINY=2
  11  MESA_VERSION = $(MESA_MAJOR).$(MESA_MINOR).$(MESA_TINY)
  12
  13  # external projects.  This should be useless now that we use libdrm.
  14 diff --git a/docs/relnotes-8.0.2.html b/docs/relnotes-8.0.2.html
  15 index ad1818c..d73ba9f 100644
  16 --- a/docs/relnotes-8.0.2.html
  17 +++ b/docs/relnotes-8.0.2.html
  18 @@ -28,7 +28,9 @@ for DRI hardware acceleration.
  19
  20  <h2>MD5 checksums</h2>
  21  <pre>
  22 -tdb
  23 +70eb3dc74fbfcd72f6776268ee1db52e  MesaLib-8.0.2.tar.gz
  24 +a368104e5700707048dc3e8691a9a7a1  MesaLib-8.0.2.tar.bz2
  25 +d5e5cdb85d2afdbcd1c0623d3ed1c54d  MesaLib-8.0.2.zip
  26  </pre>
  27
  28  <h2>New features</h2>
  29 diff --git a/src/egl/main/eglimage.c b/src/egl/main/eglimage.c
  30 index d5deae7..1174d0a 100644
  31 --- a/src/egl/main/eglimage.c
  32 +++ b/src/egl/main/eglimage.c
  33 @@ -45,7 +45,7 @@ _eglParseImageAttribList(_EGLImageAttribs *attrs, _EGLDisplay *dpy,
  34
  35     (void) dpy;
  36
  37 -   memset(attrs, 0, sizeof(attrs));
  38 +   memset(attrs, 0, sizeof(*attrs));
  39     attrs->ImagePreserved = EGL_FALSE;
  40     attrs->GLTextureLevel = 0;
  41     attrs->GLTextureZOffset = 0;
  42 diff --git a/src/gallium/auxiliary/util/u_double_list.h b/src/gallium/auxiliary/util/u_double_list.h
  43 index 2384c36..9d1129b 100644
  44 --- a/src/gallium/auxiliary/util/u_double_list.h
  45 +++ b/src/gallium/auxiliary/util/u_double_list.h
  46 @@ -105,6 +105,11 @@ static INLINE void list_delinit(struct list_head *item)
  47  #define LIST_IS_EMPTY(__list)                   \
  48      ((__list)->next == (__list))
  49
  50 +/**
  51 + * Cast from a pointer to a member of a struct back to the containing struct.
  52 + *
  53 + * 'sample' MUST be initialized, or else the result is undefined!
  54 + */
  55  #ifndef container_of
  56  #define container_of(ptr, sample, member)                              \
  57      (void *)((char *)(ptr)                                             \
  58 @@ -112,29 +117,29 @@ static INLINE void list_delinit(struct list_head *item)
  59  #endif
  60
  61  #define LIST_FOR_EACH_ENTRY(pos, head, member)                         \
  62 -   for (pos = container_of((head)->next, pos, member);                 \
  63 +   for (pos = NULL, pos = container_of((head)->next, pos, member);     \
  64         &pos->member != (head);                                         \
  65         pos = container_of(pos->member.next, pos, member))
  66
  67  #define LIST_FOR_EACH_ENTRY_SAFE(pos, storage, head, member)   \
  68 -   for (pos = container_of((head)->next, pos, member),                 \
  69 +   for (pos = NULL, pos = container_of((head)->next, pos, member),     \
  70         storage = container_of(pos->member.next, pos, member);  \
  71         &pos->member != (head);                                         \
  72         pos = storage, storage = container_of(storage->member.next, storage, member))
  73
  74  #define LIST_FOR_EACH_ENTRY_SAFE_REV(pos, storage, head, member)       \
  75 -   for (pos = container_of((head)->prev, pos, member),                 \
  76 +   for (pos = NULL, pos = container_of((head)->prev, pos, member),     \
  77         storage = container_of(pos->member.prev, pos, member);          \
  78         &pos->member != (head);                                         \
  79         pos = storage, storage = container_of(storage->member.prev, storage, member))
  80
  81  #define LIST_FOR_EACH_ENTRY_FROM(pos, start, head, member)             \
  82 -   for (pos = container_of((start), pos, member);                      \
  83 +   for (pos = NULL, pos = container_of((start), pos, member);          \
  84         &pos->member != (head);                                         \
  85         pos = container_of(pos->member.next, pos, member))
  86
  87  #define LIST_FOR_EACH_ENTRY_FROM_REV(pos, start, head, member)         \
  88 -   for (pos = container_of((start), pos, member);                      \
  89 +   for (pos = NULL, pos = container_of((start), pos, member);          \
  90         &pos->member != (head);                                         \
  91         pos = container_of(pos->member.prev, pos, member))
  92
  93 diff --git a/src/gallium/auxiliary/util/u_linkage.h b/src/gallium/auxiliary/util/u_linkage.h
  94 index 43ec917..7b23123 100644
  95 --- a/src/gallium/auxiliary/util/u_linkage.h
  96 +++ b/src/gallium/auxiliary/util/u_linkage.h
  97 @@ -49,15 +49,16 @@ unsigned util_semantic_set_from_program_file(struct util_semantic_set *set, cons
  98   *
  99   * num_slots is the size of the layout array and hardware limit instead.
 100   *
 101 - * efficient_slots == 0 or efficient_solts == num_slots are typical settings.
 102 + * efficient_slots == 0 or efficient_slots == num_slots are typical settings.
 103   */
 104  void util_semantic_layout_from_set(unsigned char *layout, const struct util_semantic_set *set, unsigned efficient_slots, unsigned num_slots);
 105
 106  static INLINE void
 107 -util_semantic_table_from_layout(unsigned char *table, unsigned char *layout, unsigned char first_slot_value, unsigned char num_slots)
 108 +util_semantic_table_from_layout(unsigned char *table, size_t table_size, unsigned char *layout,
 109 +                                unsigned char first_slot_value, unsigned char num_slots)
 110  {
 111 -   int i;
 112 -   memset(table, 0xff, sizeof(table));
 113 +   unsigned char i;
 114 +   memset(table, 0xff, table_size);
 115
 116     for(i = 0; i < num_slots; ++i)
 117        table[layout[i]] = first_slot_value + i;
 118 diff --git a/src/gallium/drivers/nvfx/nvfx_fragprog.c b/src/gallium/drivers/nvfx/nvfx_fragprog.c
 119 index dbd7c77..0babcbb 100644
 120 --- a/src/gallium/drivers/nvfx/nvfx_fragprog.c
 121 +++ b/src/gallium/drivers/nvfx/nvfx_fragprog.c
 122 @@ -977,7 +977,8 @@ nvfx_fragprog_prepare(struct nvfx_context* nvfx, struct nvfx_fpc *fpc)
 123         if(fpc->fp->num_slots > num_texcoords)
 124                 return FALSE;
 125         util_semantic_layout_from_set(fpc->fp->slot_to_generic, &set, 0, num_texcoords);
 126 -       util_semantic_table_from_layout(fpc->generic_to_slot, fpc->fp->slot_to_generic, 0, num_texcoords);
 127 +       util_semantic_table_from_layout(fpc->generic_to_slot, sizeof fpc->generic_to_slot,
 128 +                                        fpc->fp->slot_to_generic, 0, num_texcoords);
 129
 130         memset(fpc->fp->slot_to_fp_input, 0xff, sizeof(fpc->fp->slot_to_fp_input));
 131
 132 diff --git a/src/gallium/drivers/r300/compiler/radeon_program_alu.c b/src/gallium/drivers/r300/compiler/radeon_program_alu.c
 133 index c48f936..b3da311 100644
 134 --- a/src/gallium/drivers/r300/compiler/radeon_program_alu.c
 135 +++ b/src/gallium/drivers/r300/compiler/radeon_program_alu.c
 136 @@ -41,13 +41,16 @@
 137
 138  static struct rc_instruction *emit1(
 139         struct radeon_compiler * c, struct rc_instruction * after,
 140 -       rc_opcode Opcode, rc_saturate_mode Saturate, struct rc_dst_register DstReg,
 141 -       struct rc_src_register SrcReg)
 142 +       rc_opcode Opcode, struct rc_sub_instruction * base,
 143 +       struct rc_dst_register DstReg, struct rc_src_register SrcReg)
 144  {
 145         struct rc_instruction *fpi = rc_insert_new_instruction(c, after);
 146
 147 +       if (base) {
 148 +               memcpy(&fpi->U.I, base, sizeof(struct rc_sub_instruction));
 149 +       }
 150 +
 151         fpi->U.I.Opcode = Opcode;
 152 -       fpi->U.I.SaturateMode = Saturate;
 153         fpi->U.I.DstReg = DstReg;
 154         fpi->U.I.SrcReg[0] = SrcReg;
 155         return fpi;
 156 @@ -55,13 +58,17 @@ static struct rc_instruction *emit1(
 157
 158  static struct rc_instruction *emit2(
 159         struct radeon_compiler * c, struct rc_instruction * after,
 160 -       rc_opcode Opcode, rc_saturate_mode Saturate, struct rc_dst_register DstReg,
 161 +       rc_opcode Opcode, struct rc_sub_instruction * base,
 162 +       struct rc_dst_register DstReg,
 163         struct rc_src_register SrcReg0, struct rc_src_register SrcReg1)
 164  {
 165         struct rc_instruction *fpi = rc_insert_new_instruction(c, after);
 166
 167 +       if (base) {
 168 +               memcpy(&fpi->U.I, base, sizeof(struct rc_sub_instruction));
 169 +       }
 170 +
 171         fpi->U.I.Opcode = Opcode;
 172 -       fpi->U.I.SaturateMode = Saturate;
 173         fpi->U.I.DstReg = DstReg;
 174         fpi->U.I.SrcReg[0] = SrcReg0;
 175         fpi->U.I.SrcReg[1] = SrcReg1;
 176 @@ -70,14 +77,18 @@ static struct rc_instruction *emit2(
 177
 178  static struct rc_instruction *emit3(
 179         struct radeon_compiler * c, struct rc_instruction * after,
 180 -       rc_opcode Opcode, rc_saturate_mode Saturate, struct rc_dst_register DstReg,
 181 +       rc_opcode Opcode, struct rc_sub_instruction * base,
 182 +       struct rc_dst_register DstReg,
 183         struct rc_src_register SrcReg0, struct rc_src_register SrcReg1,
 184         struct rc_src_register SrcReg2)
 185  {
 186         struct rc_instruction *fpi = rc_insert_new_instruction(c, after);
 187
 188 +       if (base) {
 189 +               memcpy(&fpi->U.I, base, sizeof(struct rc_sub_instruction));
 190 +       }
 191 +
 192         fpi->U.I.Opcode = Opcode;
 193 -       fpi->U.I.SaturateMode = Saturate;
 194         fpi->U.I.DstReg = DstReg;
 195         fpi->U.I.SrcReg[0] = SrcReg0;
 196         fpi->U.I.SrcReg[1] = SrcReg1;
 197 @@ -221,7 +232,7 @@ static void transform_ABS(struct radeon_compiler* c,
 198         struct rc_src_register src = inst->U.I.SrcReg[0];
 199         src.Abs = 1;
 200         src.Negate = RC_MASK_NONE;
 201 -       emit1(c, inst->Prev, RC_OPCODE_MOV, inst->U.I.SaturateMode, inst->U.I.DstReg, src);
 202 +       emit1(c, inst->Prev, RC_OPCODE_MOV, &inst->U.I, inst->U.I.DstReg, src);
 203         rc_remove_instruction(inst);
 204  }
 205
 206 @@ -240,7 +251,7 @@ static void transform_CEIL(struct radeon_compiler* c,
 207
 208         struct rc_dst_register dst = try_to_reuse_dst(c, inst);
 209         emit1(c, inst->Prev, RC_OPCODE_FRC, 0, dst, negate(inst->U.I.SrcReg[0]));
 210 -       emit2(c, inst->Prev, RC_OPCODE_ADD, inst->U.I.SaturateMode, inst->U.I.DstReg,
 211 +       emit2(c, inst->Prev, RC_OPCODE_ADD, &inst->U.I, inst->U.I.DstReg,
 212                 inst->U.I.SrcReg[0], srcreg(RC_FILE_TEMPORARY, dst.Index));
 213         rc_remove_instruction(inst);
 214  }
 215 @@ -256,7 +267,7 @@ static void transform_CLAMP(struct radeon_compiler *c,
 216         struct rc_dst_register dst = try_to_reuse_dst(c, inst);
 217         emit2(c, inst->Prev, RC_OPCODE_MIN, 0, dst,
 218                 inst->U.I.SrcReg[0], inst->U.I.SrcReg[2]);
 219 -       emit2(c, inst->Prev, RC_OPCODE_MAX, inst->U.I.SaturateMode, inst->U.I.DstReg,
 220 +       emit2(c, inst->Prev, RC_OPCODE_MAX, &inst->U.I, inst->U.I.DstReg,
 221                 srcreg(RC_FILE_TEMPORARY, dst.Index), inst->U.I.SrcReg[1]);
 222         rc_remove_instruction(inst);
 223  }
 224 @@ -272,7 +283,7 @@ static void transform_DP2(struct radeon_compiler* c,
 225         src1.Negate &= ~(RC_MASK_Z | RC_MASK_W);
 226         src1.Swizzle &= ~(63 << (3 * 2));
 227         src1.Swizzle |= (RC_SWIZZLE_ZERO << (3 * 2)) | (RC_SWIZZLE_ZERO << (3 * 3));
 228 -       emit2(c, inst->Prev, RC_OPCODE_DP3, inst->U.I.SaturateMode, inst->U.I.DstReg, src0, src1);
 229 +       emit2(c, inst->Prev, RC_OPCODE_DP3, &inst->U.I, inst->U.I.DstReg, src0, src1);
 230         rc_remove_instruction(inst);
 231  }
 232
 233 @@ -283,7 +294,7 @@ static void transform_DPH(struct radeon_compiler* c,
 234         src0.Negate &= ~RC_MASK_W;
 235         src0.Swizzle &= ~(7 << (3 * 3));
 236         src0.Swizzle |= RC_SWIZZLE_ONE << (3 * 3);
 237 -       emit2(c, inst->Prev, RC_OPCODE_DP4, inst->U.I.SaturateMode, inst->U.I.DstReg, src0, inst->U.I.SrcReg[1]);
 238 +       emit2(c, inst->Prev, RC_OPCODE_DP4, &inst->U.I, inst->U.I.DstReg, src0, inst->U.I.SrcReg[1]);
 239         rc_remove_instruction(inst);
 240  }
 241
 242 @@ -294,7 +305,7 @@ static void transform_DPH(struct radeon_compiler* c,
 243  static void transform_DST(struct radeon_compiler* c,
 244         struct rc_instruction* inst)
 245  {
 246 -       emit2(c, inst->Prev, RC_OPCODE_MUL, inst->U.I.SaturateMode, inst->U.I.DstReg,
 247 +       emit2(c, inst->Prev, RC_OPCODE_MUL, &inst->U.I, inst->U.I.DstReg,
 248                 swizzle(inst->U.I.SrcReg[0], RC_SWIZZLE_ONE, RC_SWIZZLE_Y, RC_SWIZZLE_Z, RC_SWIZZLE_ONE),
 249                 swizzle(inst->U.I.SrcReg[1], RC_SWIZZLE_ONE, RC_SWIZZLE_Y, RC_SWIZZLE_ONE, RC_SWIZZLE_W));
 250         rc_remove_instruction(inst);
 251 @@ -305,7 +316,7 @@ static void transform_FLR(struct radeon_compiler* c,
 252  {
 253         struct rc_dst_register dst = try_to_reuse_dst(c, inst);
 254         emit1(c, inst->Prev, RC_OPCODE_FRC, 0, dst, inst->U.I.SrcReg[0]);
 255 -       emit2(c, inst->Prev, RC_OPCODE_ADD, inst->U.I.SaturateMode, inst->U.I.DstReg,
 256 +       emit2(c, inst->Prev, RC_OPCODE_ADD, &inst->U.I, inst->U.I.DstReg,
 257                 inst->U.I.SrcReg[0], negate(srcreg(RC_FILE_TEMPORARY, dst.Index)));
 258         rc_remove_instruction(inst);
 259  }
 260 @@ -379,14 +390,14 @@ static void transform_LIT(struct radeon_compiler* c,
 261                 swizzle_wwww(srctemp));
 262
 263         /* tmp.z = (tmp.x > 0) ? tmp.w : 0.0 */
 264 -       emit3(c, inst->Prev, RC_OPCODE_CMP, inst->U.I.SaturateMode,
 265 +       emit3(c, inst->Prev, RC_OPCODE_CMP, &inst->U.I,
 266                 dstregtmpmask(temp, RC_MASK_Z),
 267                 negate(swizzle_xxxx(srctemp)),
 268                 swizzle_wwww(srctemp),
 269                 builtin_zero);
 270
 271         /* tmp.x, tmp.y, tmp.w = 1.0, tmp.x, 1.0 */
 272 -       emit1(c, inst->Prev, RC_OPCODE_MOV, inst->U.I.SaturateMode,
 273 +       emit1(c, inst->Prev, RC_OPCODE_MOV, &inst->U.I,
 274                 dstregtmpmask(temp, RC_MASK_XYW),
 275                 swizzle(srctemp, RC_SWIZZLE_ONE, RC_SWIZZLE_X, RC_SWIZZLE_ONE, RC_SWIZZLE_ONE));
 276
 277 @@ -401,7 +412,7 @@ static void transform_LRP(struct radeon_compiler* c,
 278         emit2(c, inst->Prev, RC_OPCODE_ADD, 0,
 279                 dst,
 280                 inst->U.I.SrcReg[1], negate(inst->U.I.SrcReg[2]));
 281 -       emit3(c, inst->Prev, RC_OPCODE_MAD, inst->U.I.SaturateMode,
 282 +       emit3(c, inst->Prev, RC_OPCODE_MAD, &inst->U.I,
 283                 inst->U.I.DstReg,
 284                 inst->U.I.SrcReg[0], srcreg(RC_FILE_TEMPORARY, dst.Index), inst->U.I.SrcReg[2]);
 285
 286 @@ -418,7 +429,7 @@ static void transform_POW(struct radeon_compiler* c,
 287
 288         emit1(c, inst->Prev, RC_OPCODE_LG2, 0, tempdst, swizzle_xxxx(inst->U.I.SrcReg[0]));
 289         emit2(c, inst->Prev, RC_OPCODE_MUL, 0, tempdst, tempsrc, swizzle_xxxx(inst->U.I.SrcReg[1]));
 290 -       emit1(c, inst->Prev, RC_OPCODE_EX2, inst->U.I.SaturateMode, inst->U.I.DstReg, tempsrc);
 291 +       emit1(c, inst->Prev, RC_OPCODE_EX2, &inst->U.I, inst->U.I.DstReg, tempsrc);
 292
 293         rc_remove_instruction(inst);
 294  }
 295 @@ -472,7 +483,7 @@ static void transform_SEQ(struct radeon_compiler* c,
 296         struct rc_dst_register dst = try_to_reuse_dst(c, inst);
 297
 298         emit2(c, inst->Prev, RC_OPCODE_ADD, 0, dst, inst->U.I.SrcReg[0], negate(inst->U.I.SrcReg[1]));
 299 -       emit3(c, inst->Prev, RC_OPCODE_CMP, inst->U.I.SaturateMode, inst->U.I.DstReg,
 300 +       emit3(c, inst->Prev, RC_OPCODE_CMP, &inst->U.I, inst->U.I.DstReg,
 301                 negate(absolute(srcreg(RC_FILE_TEMPORARY, dst.Index))), builtin_zero, builtin_one);
 302
 303         rc_remove_instruction(inst);
 304 @@ -481,7 +492,7 @@ static void transform_SEQ(struct radeon_compiler* c,
 305  static void transform_SFL(struct radeon_compiler* c,
 306         struct rc_instruction* inst)
 307  {
 308 -       emit1(c, inst->Prev, RC_OPCODE_MOV, inst->U.I.SaturateMode, inst->U.I.DstReg, builtin_zero);
 309 +       emit1(c, inst->Prev, RC_OPCODE_MOV, &inst->U.I, inst->U.I.DstReg, builtin_zero);
 310         rc_remove_instruction(inst);
 311  }
 312
 313 @@ -491,7 +502,7 @@ static void transform_SGE(struct radeon_compiler* c,
 314         struct rc_dst_register dst = try_to_reuse_dst(c, inst);
 315
 316         emit2(c, inst->Prev, RC_OPCODE_ADD, 0, dst, inst->U.I.SrcReg[0], negate(inst->U.I.SrcReg[1]));
 317 -       emit3(c, inst->Prev, RC_OPCODE_CMP, inst->U.I.SaturateMode, inst->U.I.DstReg,
 318 +       emit3(c, inst->Prev, RC_OPCODE_CMP, &inst->U.I, inst->U.I.DstReg,
 319                 srcreg(RC_FILE_TEMPORARY, dst.Index), builtin_zero, builtin_one);
 320
 321         rc_remove_instruction(inst);
 322 @@ -503,7 +514,7 @@ static void transform_SGT(struct radeon_compiler* c,
 323         struct rc_dst_register dst = try_to_reuse_dst(c, inst);
 324
 325         emit2(c, inst->Prev, RC_OPCODE_ADD, 0, dst, negate(inst->U.I.SrcReg[0]), inst->U.I.SrcReg[1]);
 326 -       emit3(c, inst->Prev, RC_OPCODE_CMP, inst->U.I.SaturateMode, inst->U.I.DstReg,
 327 +       emit3(c, inst->Prev, RC_OPCODE_CMP, &inst->U.I, inst->U.I.DstReg,
 328                 srcreg(RC_FILE_TEMPORARY, dst.Index), builtin_one, builtin_zero);
 329
 330         rc_remove_instruction(inst);
 331 @@ -515,7 +526,7 @@ static void transform_SLE(struct radeon_compiler* c,
 332         struct rc_dst_register dst = try_to_reuse_dst(c, inst);
 333
 334         emit2(c, inst->Prev, RC_OPCODE_ADD, 0, dst, negate(inst->U.I.SrcReg[0]), inst->U.I.SrcReg[1]);
 335 -       emit3(c, inst->Prev, RC_OPCODE_CMP, inst->U.I.SaturateMode, inst->U.I.DstReg,
 336 +       emit3(c, inst->Prev, RC_OPCODE_CMP, &inst->U.I, inst->U.I.DstReg,
 337                 srcreg(RC_FILE_TEMPORARY, dst.Index), builtin_zero, builtin_one);
 338
 339         rc_remove_instruction(inst);
 340 @@ -527,7 +538,7 @@ static void transform_SLT(struct radeon_compiler* c,
 341         struct rc_dst_register dst = try_to_reuse_dst(c, inst);
 342
 343         emit2(c, inst->Prev, RC_OPCODE_ADD, 0, dst, inst->U.I.SrcReg[0], negate(inst->U.I.SrcReg[1]));
 344 -       emit3(c, inst->Prev, RC_OPCODE_CMP, inst->U.I.SaturateMode, inst->U.I.DstReg,
 345 +       emit3(c, inst->Prev, RC_OPCODE_CMP, &inst->U.I, inst->U.I.DstReg,
 346                 srcreg(RC_FILE_TEMPORARY, dst.Index), builtin_one, builtin_zero);
 347
 348         rc_remove_instruction(inst);
 349 @@ -539,7 +550,7 @@ static void transform_SNE(struct radeon_compiler* c,
 350         struct rc_dst_register dst = try_to_reuse_dst(c, inst);
 351
 352         emit2(c, inst->Prev, RC_OPCODE_ADD, 0, dst, inst->U.I.SrcReg[0], negate(inst->U.I.SrcReg[1]));
 353 -       emit3(c, inst->Prev, RC_OPCODE_CMP, inst->U.I.SaturateMode, inst->U.I.DstReg,
 354 +       emit3(c, inst->Prev, RC_OPCODE_CMP, &inst->U.I, inst->U.I.DstReg,
 355                 negate(absolute(srcreg(RC_FILE_TEMPORARY, dst.Index))), builtin_one, builtin_zero);
 356
 357         rc_remove_instruction(inst);
 358 @@ -604,7 +615,7 @@ static void transform_XPD(struct radeon_compiler* c,
 359         emit2(c, inst->Prev, RC_OPCODE_MUL, 0, dst,
 360                 swizzle(inst->U.I.SrcReg[0], RC_SWIZZLE_Z, RC_SWIZZLE_X, RC_SWIZZLE_Y, RC_SWIZZLE_W),
 361                 swizzle(inst->U.I.SrcReg[1], RC_SWIZZLE_Y, RC_SWIZZLE_Z, RC_SWIZZLE_X, RC_SWIZZLE_W));
 362 -       emit3(c, inst->Prev, RC_OPCODE_MAD, inst->U.I.SaturateMode, inst->U.I.DstReg,
 363 +       emit3(c, inst->Prev, RC_OPCODE_MAD, &inst->U.I, inst->U.I.DstReg,
 364                 swizzle(inst->U.I.SrcReg[0], RC_SWIZZLE_Y, RC_SWIZZLE_Z, RC_SWIZZLE_X, RC_SWIZZLE_W),
 365                 swizzle(inst->U.I.SrcReg[1], RC_SWIZZLE_Z, RC_SWIZZLE_X, RC_SWIZZLE_Y, RC_SWIZZLE_W),
 366                 negate(srcreg(RC_FILE_TEMPORARY, dst.Index)));
 367 @@ -719,7 +730,7 @@ static void transform_r300_vertex_DP3(struct radeon_compiler* c,
 368         src1.Negate &= ~RC_MASK_W;
 369         src1.Swizzle &= ~(7 << (3 * 3));
 370         src1.Swizzle |= RC_SWIZZLE_ZERO << (3 * 3);
 371 -       emit2(c, inst->Prev, RC_OPCODE_DP4, inst->U.I.SaturateMode, inst->U.I.DstReg, src0, src1);
 372 +       emit2(c, inst->Prev, RC_OPCODE_DP4, &inst->U.I, inst->U.I.DstReg, src0, src1);
 373         rc_remove_instruction(inst);
 374  }
 375
 376 @@ -1043,22 +1054,22 @@ static void r300_transform_SIN_COS_SCS(struct radeon_compiler *c,
 377         unsigned srctmp)
 378  {
 379         if (inst->U.I.Opcode == RC_OPCODE_COS) {
 380 -               emit1(c, inst->Prev, RC_OPCODE_COS, inst->U.I.SaturateMode, inst->U.I.DstReg,
 381 +               emit1(c, inst->Prev, RC_OPCODE_COS, &inst->U.I, inst->U.I.DstReg,
 382                         srcregswz(RC_FILE_TEMPORARY, srctmp, RC_SWIZZLE_WWWW));
 383         } else if (inst->U.I.Opcode == RC_OPCODE_SIN) {
 384 -               emit1(c, inst->Prev, RC_OPCODE_SIN, inst->U.I.SaturateMode,
 385 +               emit1(c, inst->Prev, RC_OPCODE_SIN, &inst->U.I,
 386                         inst->U.I.DstReg, srcregswz(RC_FILE_TEMPORARY, srctmp, RC_SWIZZLE_WWWW));
 387         } else if (inst->U.I.Opcode == RC_OPCODE_SCS) {
 388                 struct rc_dst_register moddst = inst->U.I.DstReg;
 389
 390                 if (inst->U.I.DstReg.WriteMask & RC_MASK_X) {
 391                         moddst.WriteMask = RC_MASK_X;
 392 -                       emit1(c, inst->Prev, RC_OPCODE_COS, inst->U.I.SaturateMode, moddst,
 393 +                       emit1(c, inst->Prev, RC_OPCODE_COS, &inst->U.I, moddst,
 394                                 srcregswz(RC_FILE_TEMPORARY, srctmp, RC_SWIZZLE_WWWW));
 395                 }
 396                 if (inst->U.I.DstReg.WriteMask & RC_MASK_Y) {
 397                         moddst.WriteMask = RC_MASK_Y;
 398 -                       emit1(c, inst->Prev, RC_OPCODE_SIN, inst->U.I.SaturateMode, moddst,
 399 +                       emit1(c, inst->Prev, RC_OPCODE_SIN, &inst->U.I, moddst,
 400                                 srcregswz(RC_FILE_TEMPORARY, srctmp, RC_SWIZZLE_WWWW));
 401                 }
 402         }
 403 diff --git a/src/gallium/drivers/r300/r300_blit.c b/src/gallium/drivers/r300/r300_blit.c
 404 index d132638..920612b 100644
 405 --- a/src/gallium/drivers/r300/r300_blit.c
 406 +++ b/src/gallium/drivers/r300/r300_blit.c
 407 @@ -63,8 +63,13 @@ static void r300_blitter_begin(struct r300_context* r300, enum r300_blitter_op o
 408      util_blitter_save_vertex_shader(r300->blitter, r300->vs_state.state);
 409      util_blitter_save_viewport(r300->blitter, &r300->viewport);
 410      util_blitter_save_vertex_elements(r300->blitter, r300->velems);
 411 -    util_blitter_save_vertex_buffers(r300->blitter, r300->vbuf_mgr->nr_vertex_buffers,
 412 -                                     r300->vbuf_mgr->vertex_buffer);
 413 +    if (r300->vbuf_mgr) {
 414 +        util_blitter_save_vertex_buffers(r300->blitter, r300->vbuf_mgr->nr_vertex_buffers,
 415 +                                         r300->vbuf_mgr->vertex_buffer);
 416 +    } else {
 417 +        util_blitter_save_vertex_buffers(r300->blitter, r300->swtcl_nr_vertex_buffers,
 418 +                                         r300->swtcl_vertex_buffer);
 419 +    }
 420
 421      if (op & R300_SAVE_FRAMEBUFFER) {
 422          util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state);
 423 diff --git a/src/gallium/drivers/r300/r300_context.c b/src/gallium/drivers/r300/r300_context.c
 424 index 7d289ca..1626768 100644
 425 --- a/src/gallium/drivers/r300/r300_context.c
 426 +++ b/src/gallium/drivers/r300/r300_context.c
 427 @@ -419,17 +419,19 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
 428      r300_init_query_functions(r300);
 429      r300_init_state_functions(r300);
 430      r300_init_resource_functions(r300);
 431 -
 432 +
 433      r300->context.create_video_decoder = vl_create_decoder;
 434      r300->context.create_video_buffer = vl_video_buffer_create;
 435
 436 -    r300->vbuf_mgr = u_vbuf_create(&r300->context, 1024 * 1024, 16,
 437 +    if (r300->screen->caps.has_tcl) {
 438 +        r300->vbuf_mgr = u_vbuf_create(&r300->context, 1024 * 1024, 16,
 439                                         PIPE_BIND_VERTEX_BUFFER |
 440                                         PIPE_BIND_INDEX_BUFFER,
 441                                         U_VERTEX_FETCH_DWORD_ALIGNED);
 442 -    if (!r300->vbuf_mgr)
 443 -        goto fail;
 444 -    r300->vbuf_mgr->caps.format_fixed32 = 0;
 445 +        if (!r300->vbuf_mgr)
 446 +            goto fail;
 447 +        r300->vbuf_mgr->caps.format_fixed32 = 0;
 448 +    }
 449
 450      r300->blitter = util_blitter_create(&r300->context);
 451      if (r300->blitter == NULL)
 452 diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h
 453 index e40b7af..8264b28 100644
 454 --- a/src/gallium/drivers/r300/r300_context.h
 455 +++ b/src/gallium/drivers/r300/r300_context.h
 456 @@ -581,6 +581,9 @@ struct r300_context {
 457      void *dsa_decompress_zmask;
 458
 459      struct u_vbuf *vbuf_mgr;
 460 +    struct pipe_index_buffer swtcl_index_buffer;
 461 +    struct pipe_vertex_buffer swtcl_vertex_buffer[PIPE_MAX_ATTRIBS];
 462 +    unsigned swtcl_nr_vertex_buffers;
 463
 464      struct util_slab_mempool pool_transfers;
 465
 466 diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
 467 index 3897e99..e4afe78 100644
 468 --- a/src/gallium/drivers/r300/r300_emit.c
 469 +++ b/src/gallium/drivers/r300/r300_emit.c
 470 @@ -1030,20 +1030,18 @@ void r300_emit_vs_state(struct r300_context* r300, unsigned size, void* state)
 471              R300_PVS_VF_MAX_VTX_NUM(12) |
 472              (r300screen->caps.is_r500 ? R500_TCL_STATE_OPTIMIZATION : 0));
 473
 474 -    /* Emit flow control instructions. */
 475 -    if (code->num_fc_ops) {
 476 -
 477 -        OUT_CS_REG(R300_VAP_PVS_FLOW_CNTL_OPC, code->fc_ops);
 478 -        if (r300screen->caps.is_r500) {
 479 -            OUT_CS_REG_SEQ(R500_VAP_PVS_FLOW_CNTL_ADDRS_LW_0, code->num_fc_ops * 2);
 480 -            OUT_CS_TABLE(code->fc_op_addrs.r500, code->num_fc_ops * 2);
 481 -        } else {
 482 -            OUT_CS_REG_SEQ(R300_VAP_PVS_FLOW_CNTL_ADDRS_0, code->num_fc_ops);
 483 -            OUT_CS_TABLE(code->fc_op_addrs.r300, code->num_fc_ops);
 484 -        }
 485 -        OUT_CS_REG_SEQ(R300_VAP_PVS_FLOW_CNTL_LOOP_INDEX_0, code->num_fc_ops);
 486 -        OUT_CS_TABLE(code->fc_loop_index, code->num_fc_ops);
 487 +    /* Emit flow control instructions.  Even if there are no fc instructions,
 488 +     * we still need to write the registers to make sure they are cleared. */
 489 +    OUT_CS_REG(R300_VAP_PVS_FLOW_CNTL_OPC, code->fc_ops);
 490 +    if (r300screen->caps.is_r500) {
 491 +        OUT_CS_REG_SEQ(R500_VAP_PVS_FLOW_CNTL_ADDRS_LW_0, R300_VS_MAX_FC_OPS * 2);
 492 +        OUT_CS_TABLE(code->fc_op_addrs.r500, R300_VS_MAX_FC_OPS * 2);
 493 +    } else {
 494 +        OUT_CS_REG_SEQ(R300_VAP_PVS_FLOW_CNTL_ADDRS_0, R300_VS_MAX_FC_OPS);
 495 +        OUT_CS_TABLE(code->fc_op_addrs.r300, R300_VS_MAX_FC_OPS);
 496      }
 497 +    OUT_CS_REG_SEQ(R300_VAP_PVS_FLOW_CNTL_LOOP_INDEX_0, R300_VS_MAX_FC_OPS);
 498 +    OUT_CS_TABLE(code->fc_loop_index, R300_VS_MAX_FC_OPS);
 499
 500      END_CS;
 501  }
 502 diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c
 503 index 83cad42..1542648 100644
 504 --- a/src/gallium/drivers/r300/r300_render.c
 505 +++ b/src/gallium/drivers/r300/r300_render.c
 506 @@ -818,7 +818,7 @@ static void r300_swtcl_draw_vbo(struct pipe_context* pipe,
 507      struct pipe_transfer *ib_transfer = NULL;
 508      int i;
 509      void *indices = NULL;
 510 -    boolean indexed = info->indexed && r300->vbuf_mgr->index_buffer.buffer;
 511 +    boolean indexed = info->indexed && r300->swtcl_index_buffer.buffer;
 512
 513      if (r300->skip_rendering) {
 514          return;
 515 @@ -831,10 +831,10 @@ static void r300_swtcl_draw_vbo(struct pipe_context* pipe,
 516              (indexed ? PREP_INDEXED : 0),
 517              indexed ? 256 : 6);
 518
 519 -    for (i = 0; i < r300->vbuf_mgr->nr_vertex_buffers; i++) {
 520 -        if (r300->vbuf_mgr->vertex_buffer[i].buffer) {
 521 +    for (i = 0; i < r300->swtcl_nr_vertex_buffers; i++) {
 522 +        if (r300->swtcl_vertex_buffer[i].buffer) {
 523              void *buf = pipe_buffer_map(pipe,
 524 -                                  r300->vbuf_mgr->vertex_buffer[i].buffer,
 525 +                                  r300->swtcl_vertex_buffer[i].buffer,
 526                                    PIPE_TRANSFER_READ |
 527                                    PIPE_TRANSFER_UNSYNCHRONIZED,
 528                                    &vb_transfer[i]);
 529 @@ -843,7 +843,7 @@ static void r300_swtcl_draw_vbo(struct pipe_context* pipe,
 530      }
 531
 532      if (indexed) {
 533 -        indices = pipe_buffer_map(pipe, r300->vbuf_mgr->index_buffer.buffer,
 534 +        indices = pipe_buffer_map(pipe, r300->swtcl_index_buffer.buffer,
 535                                    PIPE_TRANSFER_READ |
 536                                    PIPE_TRANSFER_UNSYNCHRONIZED, &ib_transfer);
 537      }
 538 @@ -856,8 +856,8 @@ static void r300_swtcl_draw_vbo(struct pipe_context* pipe,
 539      draw_flush(r300->draw);
 540      r300->draw_vbo_locked = FALSE;
 541
 542 -    for (i = 0; i < r300->vbuf_mgr->nr_vertex_buffers; i++) {
 543 -        if (r300->vbuf_mgr->vertex_buffer[i].buffer) {
 544 +    for (i = 0; i < r300->swtcl_nr_vertex_buffers; i++) {
 545 +        if (r300->swtcl_vertex_buffer[i].buffer) {
 546              pipe_buffer_unmap(pipe, vb_transfer[i]);
 547              draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 548          }
 549 diff --git a/src/gallium/drivers/r300/r300_screen.c b/src/gallium/drivers/r300/r300_screen.c
 550 index 763321b..f28b0be 100644
 551 --- a/src/gallium/drivers/r300/r300_screen.c
 552 +++ b/src/gallium/drivers/r300/r300_screen.c
 553 @@ -212,6 +212,7 @@ static int r300_get_shader_param(struct pipe_screen *pscreen, unsigned shader, e
 554          switch (param)
 555          {
 556          case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 557 +        case PIPE_SHADER_CAP_SUBROUTINES:
 558              return 0;
 559          default:;
 560          }
 561 diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
 562 index 2bc7036..8a656e6 100644
 563 --- a/src/gallium/drivers/r300/r300_state.c
 564 +++ b/src/gallium/drivers/r300/r300_state.c
 565 @@ -1048,6 +1048,10 @@ static void* r300_create_rs_state(struct pipe_context* pipe,
 566
 567      /* Override some states for Draw. */
 568      rs->rs_draw.sprite_coord_enable = 0; /* We can do this in HW. */
 569 +    rs->rs_draw.offset_point = 0;
 570 +    rs->rs_draw.offset_line = 0;
 571 +    rs->rs_draw.offset_tri = 0;
 572 +    rs->rs_draw.offset_clamp = 0;
 573
 574  #ifdef PIPE_ARCH_LITTLE_ENDIAN
 575      vap_control_status = R300_VC_NO_SWAP;
 576 @@ -1595,7 +1599,6 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe,
 577                                      const struct pipe_vertex_buffer* buffers)
 578  {
 579      struct r300_context* r300 = r300_context(pipe);
 580 -    unsigned i;
 581      struct pipe_vertex_buffer dummy_vb = {0};
 582
 583      /* There must be at least one vertex buffer set, otherwise it locks up. */
 584 @@ -1605,18 +1608,13 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe,
 585          count = 1;
 586      }
 587
 588 -    u_vbuf_set_vertex_buffers(r300->vbuf_mgr, count, buffers);
 589 -
 590      if (r300->screen->caps.has_tcl) {
 591 -        /* HW TCL. */
 592 -        for (i = 0; i < count; i++) {
 593 -            if (buffers[i].buffer &&
 594 -               !r300_resource(buffers[i].buffer)->b.user_ptr) {
 595 -            }
 596 -        }
 597 +        u_vbuf_set_vertex_buffers(r300->vbuf_mgr, count, buffers);
 598          r300->vertex_arrays_dirty = TRUE;
 599      } else {
 600 -        /* SW TCL. */
 601 +        util_copy_vertex_buffers(r300->swtcl_vertex_buffer,
 602 +                                 &r300->swtcl_nr_vertex_buffers,
 603 +                                 buffers, count);
 604          draw_set_vertex_buffers(r300->draw, count, buffers);
 605      }
 606  }
 607 @@ -1626,9 +1624,15 @@ static void r300_set_index_buffer(struct pipe_context* pipe,
 608  {
 609      struct r300_context* r300 = r300_context(pipe);
 610
 611 -    u_vbuf_set_index_buffer(r300->vbuf_mgr, ib);
 612 -
 613 -    if (!r300->screen->caps.has_tcl) {
 614 +    if (r300->screen->caps.has_tcl) {
 615 +        u_vbuf_set_index_buffer(r300->vbuf_mgr, ib);
 616 +    } else {
 617 +        if (ib) {
 618 +            pipe_resource_reference(&r300->swtcl_index_buffer.buffer, ib->buffer);
 619 +            memcpy(&r300->swtcl_index_buffer, ib, sizeof(*ib));
 620 +        } else {
 621 +            pipe_resource_reference(&r300->swtcl_index_buffer.buffer, NULL);
 622 +        }
 623          draw_set_index_buffer(r300->draw, ib);
 624      }
 625  }
 626 @@ -1702,11 +1706,11 @@ static void* r300_create_vertex_elements_state(struct pipe_context* pipe,
 627          return NULL;
 628
 629      velems->count = count;
 630 -    velems->vmgr_elements =
 631 -        u_vbuf_create_vertex_elements(r300->vbuf_mgr, count, attribs,
 632 -                                          velems->velem);
 633
 634      if (r300_screen(pipe->screen)->caps.has_tcl) {
 635 +        velems->vmgr_elements =
 636 +            u_vbuf_create_vertex_elements(r300->vbuf_mgr, count, attribs,
 637 +                                          velems->velem);
 638          /* Setup PSC.
 639           * The unused components will be replaced by (..., 0, 1). */
 640          r300_vertex_psc(velems);
 641 @@ -1716,6 +1720,8 @@ static void* r300_create_vertex_elements_state(struct pipe_context* pipe,
 642                  align(util_format_get_blocksize(velems->velem[i].src_format), 4);
 643              velems->vertex_size_dwords += velems->format_size[i] / 4;
 644          }
 645 +    } else {
 646 +        memcpy(velems->velem, attribs, count * sizeof(struct pipe_vertex_element));
 647      }
 648
 649      return velems;
 650 @@ -1733,9 +1739,9 @@ static void r300_bind_vertex_elements_state(struct pipe_context *pipe,
 651
 652      r300->velems = velems;
 653
 654 -    u_vbuf_bind_vertex_elements(r300->vbuf_mgr, state, velems->vmgr_elements);
 655 -
 656 -    if (r300->draw) {
 657 +    if (r300->screen->caps.has_tcl) {
 658 +        u_vbuf_bind_vertex_elements(r300->vbuf_mgr, state, velems->vmgr_elements);
 659 +    } else {
 660          draw_set_vertex_elements(r300->draw, velems->count, velems->velem);
 661          return;
 662      }
 663 @@ -1750,7 +1756,9 @@ static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *s
 664      struct r300_context *r300 = r300_context(pipe);
 665      struct r300_vertex_element_state *velems = state;
 666
 667 -    u_vbuf_destroy_vertex_elements(r300->vbuf_mgr, velems->vmgr_elements);
 668 +    if (r300->screen->caps.has_tcl) {
 669 +        u_vbuf_destroy_vertex_elements(r300->vbuf_mgr, velems->vmgr_elements);
 670 +    }
 671      FREE(state);
 672  }
 673
 674 @@ -1765,10 +1773,10 @@ static void* r300_create_vs_state(struct pipe_context* pipe,
 675      vs->state.tokens = tgsi_dup_tokens(shader->tokens);
 676
 677      if (r300->screen->caps.has_tcl) {
 678 -        r300_init_vs_outputs(vs);
 679 +        r300_init_vs_outputs(r300, vs);
 680          r300_translate_vertex_shader(r300, vs);
 681      } else {
 682 -        r300_draw_init_vertex_shader(r300->draw, vs);
 683 +        r300_draw_init_vertex_shader(r300, vs);
 684      }
 685
 686      return vs;
 687 @@ -1794,9 +1802,8 @@ static void r300_bind_vs_state(struct pipe_context* pipe, void* shader)
 688      if (r300->screen->caps.has_tcl) {
 689          unsigned fc_op_dwords = r300->screen->caps.is_r500 ? 3 : 2;
 690          r300_mark_atom_dirty(r300, &r300->vs_state);
 691 -        r300->vs_state.size =
 692 -                vs->code.length + 9 +
 693 -        (vs->code.num_fc_ops ? vs->code.num_fc_ops * fc_op_dwords + 4 : 0);
 694 +        r300->vs_state.size = vs->code.length + 9 +
 695 +                       (R300_VS_MAX_FC_OPS * fc_op_dwords + 4);
 696
 697          r300_mark_atom_dirty(r300, &r300->vs_constants);
 698          r300->vs_constants.size =
 699 diff --git a/src/gallium/drivers/r300/r300_vs.c b/src/gallium/drivers/r300/r300_vs.c
 700 index 1eef071..4faf2b5 100644
 701 --- a/src/gallium/drivers/r300/r300_vs.c
 702 +++ b/src/gallium/drivers/r300/r300_vs.c
 703 @@ -36,6 +36,7 @@
 704
 705  /* Convert info about VS output semantics into r300_shader_semantics. */
 706  static void r300_shader_read_vs_outputs(
 707 +    struct r300_context *r300,
 708      struct tgsi_shader_info* info,
 709      struct r300_shader_semantics* vs_outputs)
 710  {
 711 @@ -83,6 +84,14 @@ static void r300_shader_read_vs_outputs(
 712                  fprintf(stderr, "r300 VP: cannot handle edgeflag output.\n");
 713                  break;
 714
 715 +            case TGSI_SEMANTIC_CLIPVERTEX:
 716 +                assert(index == 0);
 717 +                /* Draw does clip vertex for us. */
 718 +                if (r300->screen->caps.has_tcl) {
 719 +                    fprintf(stderr, "r300 VP: cannot handle clip vertex output.\n");
 720 +                }
 721 +                break;
 722 +
 723              default:
 724                  fprintf(stderr, "r300 VP: unknown vertex output semantic: %i.\n",
 725                          info->output_semantic_name[i]);
 726 @@ -160,10 +169,11 @@ static void set_vertex_inputs_outputs(struct r300_vertex_program_compiler * c)
 727      c->code->outputs[outputs->wpos] = reg++;
 728  }
 729
 730 -void r300_init_vs_outputs(struct r300_vertex_shader *vs)
 731 +void r300_init_vs_outputs(struct r300_context *r300,
 732 +                          struct r300_vertex_shader *vs)
 733  {
 734      tgsi_scan_shader(vs->state.tokens, &vs->info);
 735 -    r300_shader_read_vs_outputs(&vs->info, &vs->outputs);
 736 +    r300_shader_read_vs_outputs(r300, &vs->info, &vs->outputs);
 737  }
 738
 739  static void r300_dummy_vertex_shader(
 740 @@ -187,7 +197,7 @@ static void r300_dummy_vertex_shader(
 741      ureg_destroy(ureg);
 742
 743      shader->dummy = TRUE;
 744 -    r300_init_vs_outputs(shader);
 745 +    r300_init_vs_outputs(r300, shader);
 746      r300_translate_vertex_shader(r300, shader);
 747  }
 748
 749 diff --git a/src/gallium/drivers/r300/r300_vs.h b/src/gallium/drivers/r300/r300_vs.h
 750 index a482ddc..b02d5d7 100644
 751 --- a/src/gallium/drivers/r300/r300_vs.h
 752 +++ b/src/gallium/drivers/r300/r300_vs.h
 753 @@ -56,12 +56,13 @@ struct r300_vertex_shader {
 754      void *draw_vs;
 755  };
 756
 757 -void r300_init_vs_outputs(struct r300_vertex_shader *vs);
 758 +void r300_init_vs_outputs(struct r300_context *r300,
 759 +                          struct r300_vertex_shader *vs);
 760
 761  void r300_translate_vertex_shader(struct r300_context *r300,
 762                                    struct r300_vertex_shader *vs);
 763
 764 -void r300_draw_init_vertex_shader(struct draw_context *draw,
 765 +void r300_draw_init_vertex_shader(struct r300_context *r300,
 766                                    struct r300_vertex_shader *vs);
 767
 768  #endif /* R300_VS_H */
 769 diff --git a/src/gallium/drivers/r300/r300_vs_draw.c b/src/gallium/drivers/r300/r300_vs_draw.c
 770 index 2939963..69d6758 100644
 771 --- a/src/gallium/drivers/r300/r300_vs_draw.c
 772 +++ b/src/gallium/drivers/r300/r300_vs_draw.c
 773 @@ -29,7 +29,7 @@
 774   *
 775   * Transformations:
 776   * 1) If the secondary color output is present, the primary color must be
 777 - *    inserted before it.
 778 + *    present too.
 779   * 2) If any back-face color output is present, there must be all 4 color
 780   *    outputs and missing ones must be inserted.
 781   * 3) Insert a trailing texcoord output containing a copy of POS, for WPOS.
 782 @@ -52,7 +52,6 @@ struct vs_transform_context {
 783
 784      boolean color_used[2];
 785      boolean bcolor_used[2];
 786 -    boolean temp_used[128];
 787
 788      /* Index of the pos output, typically 0. */
 789      unsigned pos_output;
 790 @@ -72,6 +71,8 @@ struct vs_transform_context {
 791      boolean first_instruction;
 792      /* End instruction processed? */
 793      boolean end_instruction;
 794 +
 795 +    boolean temp_used[1024];
 796  };
 797
 798  static void emit_temp(struct tgsi_transform_context *ctx, unsigned reg)
 799 @@ -102,9 +103,9 @@ static void emit_output(struct tgsi_transform_context *ctx,
 800      ++vsctx->num_outputs;
 801  }
 802
 803 -static void insert_output(struct tgsi_transform_context *ctx,
 804 -                          struct tgsi_full_declaration *before,
 805 -                          unsigned name, unsigned index, unsigned interp)
 806 +static void insert_output_before(struct tgsi_transform_context *ctx,
 807 +                                 struct tgsi_full_declaration *before,
 808 +                                 unsigned name, unsigned index, unsigned interp)
 809  {
 810      struct vs_transform_context *vsctx = (struct vs_transform_context *)ctx;
 811      unsigned i;
 812 @@ -115,28 +116,29 @@ static void insert_output(struct tgsi_transform_context *ctx,
 813      }
 814
 815      /* Insert the new output. */
 816 -    emit_output(ctx, name, index, interp, before->Range.First);
 817 +    emit_output(ctx, name, index, interp,
 818 +                before->Range.First + vsctx->decl_shift);
 819
 820      ++vsctx->decl_shift;
 821  }
 822
 823 -static void insert_trailing_bcolor(struct tgsi_transform_context *ctx,
 824 -                                   struct tgsi_full_declaration *before)
 825 +static void insert_output_after(struct tgsi_transform_context *ctx,
 826 +                                struct tgsi_full_declaration *after,
 827 +                                unsigned name, unsigned index, unsigned interp)
 828  {
 829      struct vs_transform_context *vsctx = (struct vs_transform_context *)ctx;
 830 +    unsigned i;
 831
 832 -    /* If BCOLOR0 is used, make sure BCOLOR1 is present too. Otherwise
 833 -     * the rasterizer doesn't do the color selection correctly. */
 834 -    if (vsctx->bcolor_used[0] && !vsctx->bcolor_used[1]) {
 835 -        if (before) {
 836 -            insert_output(ctx, before, TGSI_SEMANTIC_BCOLOR, 1,
 837 -                          TGSI_INTERPOLATE_LINEAR);
 838 -        } else {
 839 -            emit_output(ctx, TGSI_SEMANTIC_BCOLOR, 1,
 840 -                        TGSI_INTERPOLATE_LINEAR, vsctx->num_outputs);
 841 -        }
 842 -        vsctx->bcolor_used[1] = TRUE;
 843 +    /* Make a place for the new output. */
 844 +    for (i = after->Range.First+1; i < Elements(vsctx->out_remap); i++) {
 845 +        ++vsctx->out_remap[i];
 846      }
 847 +
 848 +    /* Insert the new output. */
 849 +    emit_output(ctx, name, index, interp,
 850 +                after->Range.First + 1);
 851 +
 852 +    ++vsctx->decl_shift;
 853  }
 854
 855  static void transform_decl(struct tgsi_transform_context *ctx,
 856 @@ -153,41 +155,38 @@ static void transform_decl(struct tgsi_transform_context *ctx,
 857
 858              case TGSI_SEMANTIC_COLOR:
 859                  assert(decl->Semantic.Index < 2);
 860 -                vsctx->color_used[decl->Semantic.Index] = TRUE;
 861
 862                  /* We must rasterize the first color if the second one is
 863                   * used, otherwise the rasterizer doesn't do the color
 864                   * selection correctly. Declare it, but don't write to it. */
 865                  if (decl->Semantic.Index == 1 && !vsctx->color_used[0]) {
 866 -                    insert_output(ctx, decl, TGSI_SEMANTIC_COLOR, 0,
 867 -                                  TGSI_INTERPOLATE_LINEAR);
 868 +                    insert_output_before(ctx, decl, TGSI_SEMANTIC_COLOR, 0,
 869 +                                         TGSI_INTERPOLATE_LINEAR);
 870                      vsctx->color_used[0] = TRUE;
 871                  }
 872                  break;
 873
 874              case TGSI_SEMANTIC_BCOLOR:
 875                  assert(decl->Semantic.Index < 2);
 876 -                vsctx->bcolor_used[decl->Semantic.Index] = TRUE;
 877
 878                  /* We must rasterize all 4 colors if back-face colors are
 879                   * used, otherwise the rasterizer doesn't do the color
 880                   * selection correctly. Declare it, but don't write to it. */
 881                  if (!vsctx->color_used[0]) {
 882 -                    insert_output(ctx, decl, TGSI_SEMANTIC_COLOR, 0,
 883 -                                  TGSI_INTERPOLATE_LINEAR);
 884 +                    insert_output_before(ctx, decl, TGSI_SEMANTIC_COLOR, 0,
 885 +                                         TGSI_INTERPOLATE_LINEAR);
 886                      vsctx->color_used[0] = TRUE;
 887                  }
 888                  if (!vsctx->color_used[1]) {
 889 -                    insert_output(ctx, decl, TGSI_SEMANTIC_COLOR, 1,
 890 -                                  TGSI_INTERPOLATE_LINEAR);
 891 +                    insert_output_before(ctx, decl, TGSI_SEMANTIC_COLOR, 1,
 892 +                                         TGSI_INTERPOLATE_LINEAR);
 893                      vsctx->color_used[1] = TRUE;
 894                  }
 895                  if (decl->Semantic.Index == 1 && !vsctx->bcolor_used[0]) {
 896 -                    insert_output(ctx, decl, TGSI_SEMANTIC_BCOLOR, 0,
 897 -                                  TGSI_INTERPOLATE_LINEAR);
 898 +                    insert_output_before(ctx, decl, TGSI_SEMANTIC_BCOLOR, 0,
 899 +                                         TGSI_INTERPOLATE_LINEAR);
 900                      vsctx->bcolor_used[0] = TRUE;
 901                  }
 902 -                /* One more case is handled in insert_trailing_bcolor. */
 903                  break;
 904
 905              case TGSI_SEMANTIC_GENERIC:
 906 @@ -195,11 +194,6 @@ static void transform_decl(struct tgsi_transform_context *ctx,
 907                  break;
 908          }
 909
 910 -        if (decl->Semantic.Name != TGSI_SEMANTIC_BCOLOR) {
 911 -            /* Insert it as soon as possible. */
 912 -            insert_trailing_bcolor(ctx, decl);
 913 -        }
 914 -
 915          /* Since we're inserting new outputs in between, the following outputs
 916           * should be moved to the right so that they don't overlap with
 917           * the newly added ones. */
 918 @@ -214,6 +208,14 @@ static void transform_decl(struct tgsi_transform_context *ctx,
 919      }
 920
 921      ctx->emit_declaration(ctx, decl);
 922 +
 923 +    /* Insert BCOLOR1 if needed. */
 924 +    if (decl->Declaration.File == TGSI_FILE_OUTPUT &&
 925 +        decl->Semantic.Name == TGSI_SEMANTIC_BCOLOR &&
 926 +        !vsctx->bcolor_used[1]) {
 927 +        insert_output_after(ctx, decl, TGSI_SEMANTIC_BCOLOR, 1,
 928 +                            TGSI_INTERPOLATE_LINEAR);
 929 +    }
 930  }
 931
 932  static void transform_inst(struct tgsi_transform_context *ctx,
 933 @@ -226,10 +228,6 @@ static void transform_inst(struct tgsi_transform_context *ctx,
 934      if (!vsctx->first_instruction) {
 935          vsctx->first_instruction = TRUE;
 936
 937 -        /* The trailing BCOLOR should be inserted before the code
 938 -         * if it hasn't already been done so. */
 939 -        insert_trailing_bcolor(ctx, NULL);
 940 -
 941          /* Insert the generic output for WPOS. */
 942          emit_output(ctx, TGSI_SEMANTIC_GENERIC, vsctx->last_generic + 1,
 943                      TGSI_INTERPOLATE_PERSPECTIVE, vsctx->num_outputs);
 944 @@ -309,14 +307,18 @@ static void transform_inst(struct tgsi_transform_context *ctx,
 945      ctx->emit_instruction(ctx, inst);
 946  }
 947
 948 -void r300_draw_init_vertex_shader(struct draw_context *draw,
 949 +void r300_draw_init_vertex_shader(struct r300_context *r300,
 950                                    struct r300_vertex_shader *vs)
 951  {
 952 +    struct draw_context *draw = r300->draw;
 953      struct pipe_shader_state new_vs;
 954 +    struct tgsi_shader_info info;
 955      struct vs_transform_context transform;
 956      const uint newLen = tgsi_num_tokens(vs->state.tokens) + 100 /* XXX */;
 957      unsigned i;
 958
 959 +    tgsi_scan_shader(vs->state.tokens, &info);
 960 +
 961      new_vs.tokens = tgsi_alloc_tokens(newLen);
 962      if (new_vs.tokens == NULL)
 963          return;
 964 @@ -329,6 +331,22 @@ void r300_draw_init_vertex_shader(struct draw_context *draw,
 965      transform.base.transform_instruction = transform_inst;
 966      transform.base.transform_declaration = transform_decl;
 967
 968 +    for (i = 0; i < info.num_outputs; i++) {
 969 +        unsigned index = info.output_semantic_index[i];
 970 +
 971 +        switch (info.output_semantic_name[i]) {
 972 +            case TGSI_SEMANTIC_COLOR:
 973 +                assert(index < 2);
 974 +                transform.color_used[index] = TRUE;
 975 +                break;
 976 +
 977 +            case TGSI_SEMANTIC_BCOLOR:
 978 +                assert(index < 2);
 979 +                transform.bcolor_used[index] = TRUE;
 980 +                break;
 981 +        }
 982 +    }
 983 +
 984      tgsi_transform_shader(vs->state.tokens,
 985                            (struct tgsi_token*)new_vs.tokens,
 986                            newLen, &transform.base);
 987 @@ -350,7 +368,7 @@ void r300_draw_init_vertex_shader(struct draw_context *draw,
 988      vs->state.tokens = new_vs.tokens;
 989
 990      /* Init the VS output table for the rasterizer. */
 991 -    r300_init_vs_outputs(vs);
 992 +    r300_init_vs_outputs(r300, vs);
 993
 994      /* Make the last generic be WPOS. */
 995      vs->outputs.wpos = vs->outputs.generic[transform.last_generic + 1];
 996 diff --git a/src/gallium/state_trackers/vega/text.c b/src/gallium/state_trackers/vega/text.c
 997 index a183933..27d461c 100644
 998 --- a/src/gallium/state_trackers/vega/text.c
 999 +++ b/src/gallium/state_trackers/vega/text.c
1000 @@ -73,8 +73,8 @@ static void add_glyph(struct vg_font *font,
1001     glyph = CALLOC_STRUCT(vg_glyph);
1002     glyph->object = obj;
1003     glyph->is_hinted = isHinted;
1004 -   memcpy(glyph->glyph_origin, glyphOrigin, sizeof(glyphOrigin));
1005 -   memcpy(glyph->escapement, escapement, sizeof(escapement));
1006 +   memcpy(glyph->glyph_origin, glyphOrigin, sizeof(glyph->glyph_origin));
1007 +   memcpy(glyph->escapement, escapement, sizeof(glyph->glyph_origin));
1008
1009     cso_hash_insert(font->glyphs, (unsigned) glyphIndex, glyph);
1010  }
1011 diff --git a/src/gallium/targets/egl-static/egl_st.c b/src/gallium/targets/egl-static/egl_st.c
1012 index 81d7bb4..67e3c29 100644
1013 --- a/src/gallium/targets/egl-static/egl_st.c
1014 +++ b/src/gallium/targets/egl-static/egl_st.c
1015 @@ -54,8 +54,9 @@ dlopen_gl_lib_cb(const char *dir, size_t len, void *callback_data)
1016     int ret;
1017
1018     if (len) {
1019 +      assert(len <= INT_MAX && "path is insanely long!");
1020        ret = util_snprintf(path, sizeof(path), "%.*s/%s" UTIL_DL_EXT,
1021 -            len, dir, name);
1022 +            (int)len, dir, name);
1023     }
1024     else {
1025        ret = util_snprintf(path, sizeof(path), "%s" UTIL_DL_EXT, name);
1026 diff --git a/src/glsl/Android.mk b/src/glsl/Android.mk
1027 index d7d17dd..84a8655 100644
1028 --- a/src/glsl/Android.mk
1029 +++ b/src/glsl/Android.mk
1030 @@ -39,6 +39,7 @@ LOCAL_SRC_FILES := \
1031         $(LIBGLSL_CXX_SOURCES)
1032
1033  LOCAL_C_INCLUDES := \
1034 +       external/astl/include \
1035         $(MESA_TOP)/src/mapi \
1036         $(MESA_TOP)/src/mesa
1037
1038 diff --git a/src/glx/apple/Makefile b/src/glx/apple/Makefile
1039 index dc64295..68fe6ad 100644
1040 --- a/src/glx/apple/Makefile
1041 +++ b/src/glx/apple/Makefile
1042 @@ -26,6 +26,7 @@ SOURCES = \
1043         apple_glx.c \
1044         apple_glx_context.c \
1045         apple_glx_drawable.c \
1046 +       apple_glx_log.c \
1047         apple_glx_pbuffer.c \
1048         apple_glx_pixmap.c \
1049         apple_glx_surface.c \
1050 diff --git a/src/glx/apple/apple_glx.c b/src/glx/apple/apple_glx.c
1051 index d94c1e0..56cff64 100644
1052 --- a/src/glx/apple/apple_glx.c
1053 +++ b/src/glx/apple/apple_glx.c
1054 @@ -33,6 +33,8 @@
1055  #include <assert.h>
1056  #include <stdarg.h>
1057  #include <dlfcn.h>
1058 +#include <pthread.h>
1059 +#include <inttypes.h>
1060  #include "appledri.h"
1061  #include "apple_glx.h"
1062  #include "apple_glx_context.h"
1063 @@ -43,22 +45,6 @@ static int dri_event_base = 0;
1064
1065  const GLuint __glXDefaultPixelStore[9] = { 0, 0, 0, 0, 0, 0, 0, 0, 1 };
1066
1067 -static bool diagnostic = false;
1068 -
1069 -void
1070 -apple_glx_diagnostic(const char *fmt, ...)
1071 -{
1072 -   va_list vl;
1073 -
1074 -   if (diagnostic) {
1075 -      fprintf(stderr, "DIAG: ");
1076 -
1077 -      va_start(vl, fmt);
1078 -      vfprintf(stderr, fmt, vl);
1079 -      va_end(vl);
1080 -   }
1081 -}
1082 -
1083  int
1084  apple_get_dri_event_base(void)
1085  {
1086 @@ -125,10 +111,9 @@ apple_init_glx(Display * dpy)
1087     if (initialized)
1088        return false;
1089
1090 -   if (getenv("LIBGL_DIAGNOSTIC")) {
1091 -      printf("initializing libGL in %s\n", __func__);
1092 -      diagnostic = true;
1093 -   }
1094 +   apple_glx_log_init();
1095 +
1096 +   apple_glx_log(ASL_LEVEL_INFO, "Initializing libGL.");
1097
1098     apple_cgl_init();
1099     (void) apple_glx_get_client_id();
1100 diff --git a/src/glx/apple/apple_glx.h b/src/glx/apple/apple_glx.h
1101 index ce8c488..0967f18 100644
1102 --- a/src/glx/apple/apple_glx.h
1103 +++ b/src/glx/apple/apple_glx.h
1104 @@ -38,7 +38,8 @@
1105  #define XP_NO_X_HEADERS
1106  #include <Xplugin.h>
1107
1108 -void apple_glx_diagnostic(const char *fmt, ...);
1109 +#include "apple_glx_log.h"
1110 +
1111  xp_client_id apple_glx_get_client_id(void);
1112  bool apple_init_glx(Display * dpy);
1113  void apple_glx_swap_buffers(void *ptr);
1114 diff --git a/src/glx/apple/apple_glx_context.c b/src/glx/apple/apple_glx_context.c
1115 index c58d05a..0bb25b4 100644
1116 --- a/src/glx/apple/apple_glx_context.c
1117 +++ b/src/glx/apple/apple_glx_context.c
1118 @@ -421,7 +421,7 @@ apple_glx_make_current_context(Display * dpy, void *oldptr, void *ptr,
1119      */
1120
1121     if (same_drawable && ac->is_current) {
1122 -      apple_glx_diagnostic("%s: same_drawable and ac->is_current\n");
1123 +      apple_glx_diagnostic("same_drawable and ac->is_current\n");
1124        return false;
1125     }
1126
1127 diff --git a/src/glx/apple/apple_glx_drawable.c b/src/glx/apple/apple_glx_drawable.c
1128 index 5530224..3f84d56 100644
1129 --- a/src/glx/apple/apple_glx_drawable.c
1130 +++ b/src/glx/apple/apple_glx_drawable.c
1131 @@ -32,6 +32,7 @@
1132  #include <stdlib.h>
1133  #include <assert.h>
1134  #include <pthread.h>
1135 +#include <string.h>
1136  #include "apple_glx.h"
1137  #include "apple_glx_context.h"
1138  #include "apple_glx_drawable.h"
1139 @@ -48,8 +49,8 @@ lock_drawables_list(void)
1140     err = pthread_mutex_lock(&drawables_lock);
1141
1142     if (err) {
1143 -      fprintf(stderr, "pthread_mutex_lock failure in %s: %d\n",
1144 -              __func__, err);
1145 +      fprintf(stderr, "pthread_mutex_lock failure in %s: %s\n",
1146 +              __func__, strerror(err));
1147        abort();
1148     }
1149  }
1150 @@ -62,8 +63,8 @@ unlock_drawables_list(void)
1151     err = pthread_mutex_unlock(&drawables_lock);
1152
1153     if (err) {
1154 -      fprintf(stderr, "pthread_mutex_unlock failure in %s: %d\n",
1155 -              __func__, err);
1156 +      fprintf(stderr, "pthread_mutex_unlock failure in %s: %s\n",
1157 +              __func__, strerror(err));
1158        abort();
1159     }
1160  }
1161 @@ -95,7 +96,7 @@ drawable_lock(struct apple_glx_drawable *agd)
1162     err = pthread_mutex_lock(&agd->mutex);
1163
1164     if (err) {
1165 -      fprintf(stderr, "pthread_mutex_lock error: %d\n", err);
1166 +      fprintf(stderr, "pthread_mutex_lock error: %s\n", strerror(err));
1167        abort();
1168     }
1169  }
1170 @@ -108,7 +109,7 @@ drawable_unlock(struct apple_glx_drawable *d)
1171     err = pthread_mutex_unlock(&d->mutex);
1172
1173     if (err) {
1174 -      fprintf(stderr, "pthread_mutex_unlock error: %d\n", err);
1175 +      fprintf(stderr, "pthread_mutex_unlock error: %s\n", strerror(err));
1176        abort();
1177     }
1178  }
1179 @@ -135,6 +136,7 @@ release_drawable(struct apple_glx_drawable *d)
1180  static bool
1181  destroy_drawable(struct apple_glx_drawable *d)
1182  {
1183 +   int err;
1184
1185     d->lock(d);
1186
1187 @@ -172,6 +174,12 @@ destroy_drawable(struct apple_glx_drawable *d)
1188
1189     apple_glx_diagnostic("%s: freeing %p\n", __func__, (void *) d);
1190
1191 +   err = pthread_mutex_destroy(&d->mutex);
1192 +   if (err) {
1193 +      fprintf(stderr, "pthread_mutex_destroy error: %s\n", strerror(err));
1194 +      abort();
1195 +   }
1196 +
1197     free(d);
1198
1199     /* So that the locks are balanced and the caller correctly unlocks. */
1200 @@ -238,7 +246,7 @@ common_init(Display * dpy, GLXDrawable drawable, struct apple_glx_drawable *d)
1201     err = pthread_mutexattr_init(&attr);
1202
1203     if (err) {
1204 -      fprintf(stderr, "pthread_mutexattr_init error: %d\n", err);
1205 +      fprintf(stderr, "pthread_mutexattr_init error: %s\n", strerror(err));
1206        abort();
1207     }
1208
1209 @@ -250,14 +258,14 @@ common_init(Display * dpy, GLXDrawable drawable, struct apple_glx_drawable *d)
1210     err = pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_RECURSIVE);
1211
1212     if (err) {
1213 -      fprintf(stderr, "error: setting pthread mutex type: %d\n", err);
1214 +      fprintf(stderr, "error: setting pthread mutex type: %s\n", strerror(err));
1215        abort();
1216     }
1217
1218     err = pthread_mutex_init(&d->mutex, &attr);
1219
1220     if (err) {
1221 -      fprintf(stderr, "pthread_mutex_init error: %d\n", err);
1222 +      fprintf(stderr, "pthread_mutex_init error: %s\n", strerror(err));
1223        abort();
1224     }
1225
1226 diff --git a/src/glx/apple/apple_glx_log.c b/src/glx/apple/apple_glx_log.c
1227 new file mode 100644
1228 index 0000000..9ebf666
1229 --- /dev/null
1230 +++ b/src/glx/apple/apple_glx_log.c
1231 @@ -0,0 +1,118 @@
1232 +/*
1233 + * Copyright (c) 2012 Apple Inc.
1234 + *
1235 + * Permission is hereby granted, free of charge, to any person
1236 + * obtaining a copy of this software and associated documentation files
1237 + * (the "Software"), to deal in the Software without restriction,
1238 + * including without limitation the rights to use, copy, modify, merge,
1239 + * publish, distribute, sublicense, and/or sell copies of the Software,
1240 + * and to permit persons to whom the Software is furnished to do so,
1241 + * subject to the following conditions:
1242 + *
1243 + * The above copyright notice and this permission notice shall be
1244 + * included in all copies or substantial portions of the Software.
1245 + *
1246 + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
1247 + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
1248 + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
1249 + * NONINFRINGEMENT.  IN NO EVENT SHALL THE ABOVE LISTED COPYRIGHT
1250 + * HOLDER(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
1251 + * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
1252 + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
1253 + * DEALINGS IN THE SOFTWARE.
1254 + *
1255 + * Except as contained in this notice, the name(s) of the above
1256 + * copyright holders shall not be used in advertising or otherwise to
1257 + * promote the sale, use or other dealings in this Software without
1258 + * prior written authorization.
1259 + */
1260 +
1261 +#include <sys/cdefs.h>
1262 +#include <asl.h>
1263 +#include <stdio.h>
1264 +#include <stdbool.h>
1265 +#include <stdint.h>
1266 +#include <stdlib.h>
1267 +#include <inttypes.h>
1268 +#include <pthread.h>
1269 +#include "apple_glx_log.h"
1270 +
1271 +static bool diagnostic = false;
1272 +static aslclient aslc;
1273 +
1274 +void apple_glx_log_init(void) {
1275 +    if (getenv("LIBGL_DIAGNOSTIC")) {
1276 +        diagnostic = true;
1277 +    }
1278 +
1279 +    aslc = asl_open(NULL, NULL, 0);
1280 +}
1281 +
1282 +void _apple_glx_log(int level, const char *file, const char *function,
1283 +                    int line, const char *fmt, ...) {
1284 +    va_list v;
1285 +    va_start(v, fmt);
1286 +    _apple_glx_vlog(level, file, function, line, fmt, v);
1287 +    va_end(v);
1288 +}
1289 +
1290 +static const char *
1291 +_asl_level_string(int level)
1292 +{
1293 +        if (level == ASL_LEVEL_EMERG) return ASL_STRING_EMERG;
1294 +        if (level == ASL_LEVEL_ALERT) return ASL_STRING_ALERT;
1295 +        if (level == ASL_LEVEL_CRIT) return ASL_STRING_CRIT;
1296 +        if (level == ASL_LEVEL_ERR) return ASL_STRING_ERR;
1297 +        if (level == ASL_LEVEL_WARNING) return ASL_STRING_WARNING;
1298 +        if (level == ASL_LEVEL_NOTICE) return ASL_STRING_NOTICE;
1299 +        if (level == ASL_LEVEL_INFO) return ASL_STRING_INFO;
1300 +        if (level == ASL_LEVEL_DEBUG) return ASL_STRING_DEBUG;
1301 +        return "unknown";
1302 +}
1303 +
1304 +void _apple_glx_vlog(int level, const char *file, const char *function,
1305 +                     int line, const char *fmt, va_list args) {
1306 +    aslmsg msg;
1307 +    uint64_t thread = 0;
1308 +
1309 +    if (pthread_is_threaded_np()) {
1310 +        pthread_threadid_np(NULL, &thread);
1311 +    }
1312 +
1313 +    if (diagnostic) {
1314 +        va_list args2;
1315 +        va_copy(args2, args);
1316 +
1317 +        fprintf(stderr, "%-9s %24s:%-4d %s(%"PRIu64"): ",
1318 +                _asl_level_string(level), file, line, function, thread);
1319 +        vfprintf(stderr, fmt, args2);
1320 +    }
1321 +
1322 +    msg = asl_new(ASL_TYPE_MSG);
1323 +    if (msg) {
1324 +        if (file)
1325 +            asl_set(msg, "File", file);
1326 +        if (function)
1327 +            asl_set(msg, "Function", function);
1328 +        if (line) {
1329 +            char *_line;
1330 +            asprintf(&_line, "%d", line);
1331 +            if (_line) {
1332 +                asl_set(msg, "Line", _line);
1333 +                free(_line);
1334 +            }
1335 +        }
1336 +        if (pthread_is_threaded_np()) {
1337 +            char *_thread;
1338 +            asprintf(&_thread, "%"PRIu64, thread);
1339 +            if (_thread) {
1340 +                asl_set(msg, "Thread", _thread);
1341 +                free(_thread);
1342 +            }
1343 +        }
1344 +    }
1345 +
1346 +    asl_vlog(aslc, msg, level, fmt, args);
1347 +    if (msg)
1348 +        asl_free(msg);
1349 +}
1350 diff --git a/src/glx/apple/apple_glx_log.h b/src/glx/apple/apple_glx_log.h
1351 new file mode 100644
1352 index 0000000..4b1c531
1353 --- /dev/null
1354 +++ b/src/glx/apple/apple_glx_log.h
1355 @@ -0,0 +1,57 @@
1356 +/*
1357 + * Copyright (c) 2012 Apple Inc.
1358 + *
1359 + * Permission is hereby granted, free of charge, to any person
1360 + * obtaining a copy of this software and associated documentation files
1361 + * (the "Software"), to deal in the Software without restriction,
1362 + * including without limitation the rights to use, copy, modify, merge,
1363 + * publish, distribute, sublicense, and/or sell copies of the Software,
1364 + * and to permit persons to whom the Software is furnished to do so,
1365 + * subject to the following conditions:
1366 + *
1367 + * The above copyright notice and this permission notice shall be
1368 + * included in all copies or substantial portions of the Software.
1369 + *
1370 + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
1371 + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
1372 + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
1373 + * NONINFRINGEMENT.  IN NO EVENT SHALL THE ABOVE LISTED COPYRIGHT
1374 + * HOLDER(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
1375 + * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
1376 + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
1377 + * DEALINGS IN THE SOFTWARE.
1378 + *
1379 + * Except as contained in this notice, the name(s) of the above
1380 + * copyright holders shall not be used in advertising or otherwise to
1381 + * promote the sale, use or other dealings in this Software without
1382 + * prior written authorization.
1383 + */
1384 +
1385 +#ifndef APPLE_GLX_LOG_H
1386 +#define APPLE_GLX_LOG_H
1387 +
1388 +#include <sys/cdefs.h>
1389 +#include <asl.h>
1390 +
1391 +void apple_glx_log_init(void);
1392 +
1393 +__printflike(5, 6)
1394 +void _apple_glx_log(int level, const char *file, const char *function,
1395 +                    int line, const char *fmt, ...);
1396 +#define apple_glx_log(l, f, args ...) \
1397 +    _apple_glx_log(l, __FILE__, __FUNCTION__, __LINE__, f, ## args)
1398 +
1399 +
1400 +__printflike(5, 0)
1401 +void _apple_glx_vlog(int level, const char *file, const char *function,
1402 +                     int line, const char *fmt, va_list v);
1403 +#define apple_glx_vlog(l, f, v) \
1404 +    _apple_glx_vlog(l, __FILE__, __FUNCTION__, __LINE__, f, v)
1405 +
1406 +/* This is just here to help the transition.
1407 + * TODO: Replace calls to apple_glx_diagnostic
1408 + */
1409 +#define apple_glx_diagnostic(f, args ...) \
1410 +    apple_glx_log(ASL_LEVEL_DEBUG, f, ## args)
1411 +
1412 +#endif
1413 diff --git a/src/glx/apple/apple_glx_surface.c b/src/glx/apple/apple_glx_surface.c
1414 index 39f5130..d42fa3b 100644
1415 --- a/src/glx/apple/apple_glx_surface.c
1416 +++ b/src/glx/apple/apple_glx_surface.c
1417 @@ -206,6 +206,10 @@ apple_glx_surface_destroy(unsigned int uid)
1418     if (d) {
1419        d->types.surface.pending_destroy = true;
1420        d->release(d);
1421 +
1422 +      /* apple_glx_drawable_find_by_uid returns a locked drawable */
1423 +      d->unlock(d);
1424 +
1425        /*
1426         * We release 2 references to the surface.  One was acquired by
1427         * the find, and the other was leftover from a context, or
1428 @@ -217,7 +221,5 @@ apple_glx_surface_destroy(unsigned int uid)
1429         * by a glViewport callback (see apple_glx_context_update()).
1430         */
1431        d->destroy(d);
1432 -
1433 -      d->unlock(d);
1434     }
1435  }
1436 diff --git a/src/mapi/glapi/glapi_gentable.c b/src/mapi/glapi/glapi_gentable.c
1437 index 5c04801..640c495 100644
1438 --- a/src/mapi/glapi/glapi_gentable.c
1439 +++ b/src/mapi/glapi/glapi_gentable.c
1440 @@ -105,7 +105,7 @@ __glapi_gentable_set_remaining_noop(struct _glapi_table *disp) {
1441
1442  struct _glapi_table *
1443  _glapi_create_table_from_handle(void *handle, const char *symbol_prefix) {
1444 -    struct _glapi_table *disp = calloc(1, sizeof(struct _glapi_table));
1445 +    struct _glapi_table *disp = calloc(_glapi_get_dispatch_table_size(), sizeof(void *));
1446      char symboln[512];
1447
1448      if(!disp)
1449 diff --git a/src/mesa/drivers/dri/i915/i915_context.c b/src/mesa/drivers/dri/i915/i915_context.c
1450 index 36563ef..dc32292 100644
1451 --- a/src/mesa/drivers/dri/i915/i915_context.c
1452 +++ b/src/mesa/drivers/dri/i915/i915_context.c
1453 @@ -76,6 +76,8 @@ i915InvalidateState(struct gl_context * ctx, GLuint new_state)
1454         i915_update_provoking_vertex(ctx);
1455     if (new_state & (_NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS))
1456         i915_update_program(ctx);
1457 +   if (new_state & (_NEW_PROGRAM | _NEW_POINT))
1458 +       i915_update_sprite_point_enable(ctx);
1459  }
1460
1461
1462 diff --git a/src/mesa/drivers/dri/i915/i915_context.h b/src/mesa/drivers/dri/i915/i915_context.h
1463 index 8167137..7037465 100644
1464 --- a/src/mesa/drivers/dri/i915/i915_context.h
1465 +++ b/src/mesa/drivers/dri/i915/i915_context.h
1466 @@ -40,6 +40,7 @@
1467  #define I915_FALLBACK_POINT_SMOOTH      0x80000
1468  #define I915_FALLBACK_POINT_SPRITE_COORD_ORIGIN         0x100000
1469  #define I915_FALLBACK_DRAW_OFFSET       0x200000
1470 +#define I915_FALLBACK_COORD_REPLACE     0x400000
1471
1472  #define I915_UPLOAD_CTX              0x1
1473  #define I915_UPLOAD_BUFFERS          0x2
1474 @@ -338,6 +339,7 @@ extern void i915InitStateFunctions(struct dd_function_table *functions);
1475  extern void i915InitState(struct i915_context *i915);
1476  extern void i915_update_stencil(struct gl_context * ctx);
1477  extern void i915_update_provoking_vertex(struct gl_context *ctx);
1478 +extern void i915_update_sprite_point_enable(struct gl_context *ctx);
1479
1480
1481  /*======================================================================
1482 diff --git a/src/mesa/drivers/dri/i915/i915_state.c b/src/mesa/drivers/dri/i915/i915_state.c
1483 index 756001f..94c7327 100644
1484 --- a/src/mesa/drivers/dri/i915/i915_state.c
1485 +++ b/src/mesa/drivers/dri/i915/i915_state.c
1486 @@ -652,6 +652,48 @@ i915PointParameterfv(struct gl_context * ctx, GLenum pname, const GLfloat *param
1487     }
1488  }
1489
1490 +void
1491 +i915_update_sprite_point_enable(struct gl_context *ctx)
1492 +{
1493 +   struct intel_context *intel = intel_context(ctx);
1494 +   /* _NEW_PROGRAM */
1495 +   struct i915_fragment_program *p =
1496 +      (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1497 +   const GLbitfield64 inputsRead = p->FragProg.Base.InputsRead;
1498 +   struct i915_context *i915 = i915_context(ctx);
1499 +   GLuint s4 = i915->state.Ctx[I915_CTXREG_LIS4] & ~S4_VFMT_MASK;
1500 +   int i;
1501 +   GLuint coord_replace_bits = 0x0;
1502 +   GLuint tex_coord_unit_bits = 0x0;
1503 +
1504 +   for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
1505 +      /* _NEW_POINT */
1506 +      if (ctx->Point.CoordReplace[i] && ctx->Point.PointSprite)
1507 +         coord_replace_bits |= (1 << i);
1508 +      if (inputsRead & FRAG_BIT_TEX(i))
1509 +         tex_coord_unit_bits |= (1 << i);
1510 +   }
1511 +
1512 +   /*
1513 +    * Here we can't enable the SPRITE_POINT_ENABLE bit when the mis-match
1514 +    * of tex_coord_unit_bits and coord_replace_bits, or this will make all
1515 +    * the other non-point-sprite coords(like varying inputs, as we now use
1516 +    * tex coord to implement varying inputs) be replaced to value (0, 0)-(1, 1).
1517 +    *
1518 +    * Thus, do fallback when needed.
1519 +    */
1520 +   FALLBACK(intel, I915_FALLBACK_COORD_REPLACE,
1521 +            coord_replace_bits && coord_replace_bits != tex_coord_unit_bits);
1522 +
1523 +   s4 &= ~S4_SPRITE_POINT_ENABLE;
1524 +   s4 |= (coord_replace_bits && coord_replace_bits == tex_coord_unit_bits) ?
1525 +         S4_SPRITE_POINT_ENABLE : 0;
1526 +   if (s4 != i915->state.Ctx[I915_CTXREG_LIS4]) {
1527 +      i915->state.Ctx[I915_CTXREG_LIS4] = s4;
1528 +      I915_STATECHANGE(i915, I915_UPLOAD_CTX);
1529 +   }
1530 +}
1531 +
1532
1533  /* =============================================================
1534   * Color masks
1535 @@ -869,18 +911,7 @@ i915Enable(struct gl_context * ctx, GLenum cap, GLboolean state)
1536        break;
1537
1538     case GL_POINT_SPRITE:
1539 -      /* This state change is handled in i915_reduced_primitive_state because
1540 -       * the hardware bit should only be set when rendering points.
1541 -       */
1542 -        dw = i915->state.Ctx[I915_CTXREG_LIS4];
1543 -      if (state)
1544 -        dw |= S4_SPRITE_POINT_ENABLE;
1545 -      else
1546 -        dw &= ~S4_SPRITE_POINT_ENABLE;
1547 -      if (dw != i915->state.Ctx[I915_CTXREG_LIS4]) {
1548 -        i915->state.Ctx[I915_CTXREG_LIS4] = dw;
1549 -        I915_STATECHANGE(i915, I915_UPLOAD_CTX);
1550 -      }
1551 +      /* Handle it at i915_update_sprite_point_enable () */
1552        break;
1553
1554     case GL_POINT_SMOOTH:
1555 diff --git a/src/mesa/drivers/dri/i915/intel_tris.c b/src/mesa/drivers/dri/i915/intel_tris.c
1556 index a36011a..68f0e05 100644
1557 --- a/src/mesa/drivers/dri/i915/intel_tris.c
1558 +++ b/src/mesa/drivers/dri/i915/intel_tris.c
1559 @@ -1198,6 +1198,7 @@ static char *fallbackStrings[] = {
1560     [19] = "Smooth point",
1561     [20] = "point sprite coord origin",
1562     [21] = "depth/color drawing offset",
1563 +   [22] = "coord replace(SPRITE POINT ENABLE)",
1564  };
1565
1566
1567 diff --git a/src/mesa/drivers/dri/i965/brw_eu.h b/src/mesa/drivers/dri/i965/brw_eu.h
1568 index f660222..5064c18 100644
1569 --- a/src/mesa/drivers/dri/i965/brw_eu.h
1570 +++ b/src/mesa/drivers/dri/i965/brw_eu.h
1571 @@ -1048,6 +1048,7 @@ struct brw_instruction *brw_WHILE(struct brw_compile *p);
1572  struct brw_instruction *brw_BREAK(struct brw_compile *p);
1573  struct brw_instruction *brw_CONT(struct brw_compile *p);
1574  struct brw_instruction *gen6_CONT(struct brw_compile *p);
1575 +struct brw_instruction *gen6_HALT(struct brw_compile *p);
1576  /* Forward jumps:
1577   */
1578  void brw_land_fwd_jump(struct brw_compile *p, int jmp_insn_idx);
1579 diff --git a/src/mesa/drivers/dri/i965/brw_eu_emit.c b/src/mesa/drivers/dri/i965/brw_eu_emit.c
1580 index b2581da..21d3c5a 100644
1581 --- a/src/mesa/drivers/dri/i965/brw_eu_emit.c
1582 +++ b/src/mesa/drivers/dri/i965/brw_eu_emit.c
1583 @@ -1339,6 +1339,20 @@ struct brw_instruction *brw_CONT(struct brw_compile *p)
1584     return insn;
1585  }
1586
1587 +struct brw_instruction *gen6_HALT(struct brw_compile *p)
1588 +{
1589 +   struct brw_instruction *insn;
1590 +
1591 +   insn = next_insn(p, BRW_OPCODE_HALT);
1592 +   brw_set_dest(p, insn, retype(brw_null_reg(), BRW_REGISTER_TYPE_D));
1593 +   brw_set_src0(p, insn, retype(brw_null_reg(), BRW_REGISTER_TYPE_D));
1594 +   brw_set_src1(p, insn, brw_imm_d(0x0)); /* UIP and JIP, updated later. */
1595 +
1596 +   insn->header.compression_control = BRW_COMPRESSION_NONE;
1597 +   insn->header.execution_size = BRW_EXECUTE_8;
1598 +   return insn;
1599 +}
1600 +
1601  /* DO/WHILE loop:
1602   *
1603   * The DO/WHILE is just an unterminated loop -- break or continue are
1604 @@ -2395,8 +2409,8 @@ brw_find_next_block_end(struct brw_compile *p, int start)
1605          return ip;
1606        }
1607     }
1608 -   assert(!"not reached");
1609 -   return start + 1;
1610 +
1611 +   return 0;
1612  }
1613
1614  /* There is no DO instruction on gen6, so to find the end of the loop
1615 @@ -2425,7 +2439,7 @@ brw_find_loop_end(struct brw_compile *p, int start)
1616  }
1617
1618  /* After program generation, go back and update the UIP and JIP of
1619 - * BREAK and CONT instructions to their correct locations.
1620 + * BREAK, CONT, and HALT instructions to their correct locations.
1621   */
1622  void
1623  brw_set_uip_jip(struct brw_compile *p)
1624 @@ -2439,21 +2453,50 @@ brw_set_uip_jip(struct brw_compile *p)
1625
1626     for (ip = 0; ip < p->nr_insn; ip++) {
1627        struct brw_instruction *insn = &p->store[ip];
1628 +      int block_end_ip = 0;
1629 +
1630 +      if (insn->header.opcode == BRW_OPCODE_BREAK ||
1631 +         insn->header.opcode == BRW_OPCODE_CONTINUE ||
1632 +         insn->header.opcode == BRW_OPCODE_HALT) {
1633 +        block_end_ip = brw_find_next_block_end(p, ip);
1634 +      }
1635
1636        switch (insn->header.opcode) {
1637        case BRW_OPCODE_BREAK:
1638 -        insn->bits3.break_cont.jip = br * (brw_find_next_block_end(p, ip) - ip);
1639 +        assert(block_end_ip != 0);
1640 +        insn->bits3.break_cont.jip = br * (block_end_ip - ip);
1641          /* Gen7 UIP points to WHILE; Gen6 points just after it */
1642          insn->bits3.break_cont.uip =
1643             br * (brw_find_loop_end(p, ip) - ip + (intel->gen == 6 ? 1 : 0));
1644          break;
1645        case BRW_OPCODE_CONTINUE:
1646 -        insn->bits3.break_cont.jip = br * (brw_find_next_block_end(p, ip) - ip);
1647 +        assert(block_end_ip != 0);
1648 +        insn->bits3.break_cont.jip = br * (block_end_ip - ip);
1649          insn->bits3.break_cont.uip = br * (brw_find_loop_end(p, ip) - ip);
1650
1651          assert(insn->bits3.break_cont.uip != 0);
1652          assert(insn->bits3.break_cont.jip != 0);
1653          break;
1654 +      case BRW_OPCODE_HALT:
1655 +        /* From the Sandy Bridge PRM (volume 4, part 2, section 8.3.19):
1656 +         *
1657 +         *    "In case of the halt instruction not inside any conditional code
1658 +         *     block, the value of <JIP> and <UIP> should be the same. In case
1659 +         *     of the halt instruction inside conditional code block, the <UIP>
1660 +         *     should be the end of the program, and the <JIP> should be end of
1661 +         *     the most inner conditional code block."
1662 +         *
1663 +         * The uip will have already been set by whoever set up the
1664 +         * instruction.
1665 +         */
1666 +        if (block_end_ip == 0) {
1667 +           insn->bits3.break_cont.jip = insn->bits3.break_cont.uip;
1668 +        } else {
1669 +           insn->bits3.break_cont.jip = br * (block_end_ip - ip);
1670 +        }
1671 +        assert(insn->bits3.break_cont.uip != 0);
1672 +        assert(insn->bits3.break_cont.jip != 0);
1673 +        break;
1674        }
1675     }
1676  }
1677 diff --git a/src/mesa/drivers/dri/i965/brw_fs.h b/src/mesa/drivers/dri/i965/brw_fs.h
1678 index 9a2cc08..b9cd42f 100644
1679 --- a/src/mesa/drivers/dri/i965/brw_fs.h
1680 +++ b/src/mesa/drivers/dri/i965/brw_fs.h
1681 @@ -171,6 +171,26 @@ static const fs_reg reg_undef;
1682  static const fs_reg reg_null_f(ARF, BRW_ARF_NULL, BRW_REGISTER_TYPE_F);
1683  static const fs_reg reg_null_d(ARF, BRW_ARF_NULL, BRW_REGISTER_TYPE_D);
1684
1685 +class ip_record : public exec_node {
1686 +public:
1687 +   static void* operator new(size_t size, void *ctx)
1688 +   {
1689 +      void *node;
1690 +
1691 +      node = rzalloc_size(ctx, size);
1692 +      assert(node != NULL);
1693 +
1694 +      return node;
1695 +   }
1696 +
1697 +   ip_record(int ip)
1698 +   {
1699 +      this->ip = ip;
1700 +   }
1701 +
1702 +   int ip;
1703 +};
1704 +
1705  class fs_inst : public exec_node {
1706  public:
1707     /* Callers of this ralloc-based new need not call delete. It's
1708 @@ -489,6 +509,7 @@ public:
1709     bool remove_duplicate_mrf_writes();
1710     bool virtual_grf_interferes(int a, int b);
1711     void schedule_instructions();
1712 +   void patch_discard_jumps_to_fb_writes();
1713     void fail(const char *msg, ...);
1714
1715     void push_force_uncompressed();
1716 @@ -571,6 +592,7 @@ public:
1717     struct gl_shader_program *prog;
1718     void *mem_ctx;
1719     exec_list instructions;
1720 +   exec_list discard_halt_patches;
1721
1722     /* Delayed setup of c->prog_data.params[] due to realloc of
1723      * ParamValues[] during compile.
1724 diff --git a/src/mesa/drivers/dri/i965/brw_fs_emit.cpp b/src/mesa/drivers/dri/i965/brw_fs_emit.cpp
1725 index b68d8cb..cc70904 100644
1726 --- a/src/mesa/drivers/dri/i965/brw_fs_emit.cpp
1727 +++ b/src/mesa/drivers/dri/i965/brw_fs_emit.cpp
1728 @@ -37,11 +37,55 @@ extern "C" {
1729  #include "glsl/ir_print_visitor.h"
1730
1731  void
1732 +fs_visitor::patch_discard_jumps_to_fb_writes()
1733 +{
1734 +   if (intel->gen < 6 || this->discard_halt_patches.is_empty())
1735 +      return;
1736 +
1737 +   /* There is a somewhat strange undocumented requirement of using
1738 +    * HALT, according to the simulator.  If some channel has HALTed to
1739 +    * a particular UIP, then by the end of the program, every channel
1740 +    * must have HALTed to that UIP.  Furthermore, the tracking is a
1741 +    * stack, so you can't do the final halt of a UIP after starting
1742 +    * halting to a new UIP.
1743 +    *
1744 +    * Symptoms of not emitting this instruction on actual hardware
1745 +    * included GPU hangs and sparkly rendering on the piglit discard
1746 +    * tests.
1747 +    */
1748 +   struct brw_instruction *last_halt = gen6_HALT(p);
1749 +   last_halt->bits3.break_cont.uip = 2;
1750 +   last_halt->bits3.break_cont.jip = 2;
1751 +
1752 +   int ip = p->nr_insn;
1753 +
1754 +   foreach_list(node, &this->discard_halt_patches) {
1755 +      ip_record *patch_ip = (ip_record *)node;
1756 +      struct brw_instruction *patch = &p->store[patch_ip->ip];
1757 +      int br = (intel->gen >= 5) ? 2 : 1;
1758 +
1759 +      /* HALT takes a distance from the pre-incremented IP, so '1'
1760 +       * would be the next instruction after jmpi.
1761 +       */
1762 +      assert(patch->header.opcode == BRW_OPCODE_HALT);
1763 +      patch->bits3.break_cont.uip = (ip - patch_ip->ip) * br;
1764 +   }
1765 +
1766 +   this->discard_halt_patches.make_empty();
1767 +}
1768 +
1769 +void
1770  fs_visitor::generate_fb_write(fs_inst *inst)
1771  {
1772     bool eot = inst->eot;
1773     struct brw_reg implied_header;
1774
1775 +   /* Note that the jumps emitted to this point mean that the g0 ->
1776 +    * base_mrf setup must be inside of this function, so that we jump
1777 +    * to a point containing it.
1778 +    */
1779 +   patch_discard_jumps_to_fb_writes();
1780 +
1781     /* Header is 2 regs, g0 and g1 are the contents. g0 will be implied
1782      * move, here's g1.
1783      */
1784 @@ -482,6 +526,17 @@ fs_visitor::generate_discard(fs_inst *inst)
1785        brw_set_mask_control(p, BRW_MASK_DISABLE);
1786        brw_AND(p, g1, f0, g1);
1787        brw_pop_insn_state(p);
1788 +
1789 +      /* GLSL 1.30+ say that discarded channels should stop executing
1790 +       * (so, for example, an infinite loop that would otherwise in
1791 +       * just that channel does not occur.
1792 +       *
1793 +       * This HALT will be patched up at FB write time to point UIP at
1794 +       * the end of the program, and at brw_uip_jip() JIP will be set
1795 +       * to the end of the current block (or the program).
1796 +       */
1797 +      this->discard_halt_patches.push_tail(new(mem_ctx) ip_record(p->nr_insn));
1798 +      gen6_HALT(p);
1799     } else {
1800        struct brw_reg g0 = retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UW);
1801
1802 diff --git a/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp b/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
1803 index 0632052..cec1e95 100644
1804 --- a/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
1805 +++ b/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
1806 @@ -1921,7 +1921,10 @@ fs_visitor::emit_fb_writes()
1807  {
1808     this->current_annotation = "FB write header";
1809     bool header_present = true;
1810 -   int base_mrf = 2;
1811 +   /* We can potentially have a message length of up to 15, so we have to set
1812 +    * base_mrf to either 0 or 1 in order to fit in m0..m15.
1813 +    */
1814 +   int base_mrf = 1;
1815     int nr = base_mrf;
1816     int reg_width = c->dispatch_width / 8;
1817
1818 diff --git a/src/mesa/drivers/dri/i965/brw_tex_layout.c b/src/mesa/drivers/dri/i965/brw_tex_layout.c
1819 index 7a1b91f..8bf1d3d 100644
1820 --- a/src/mesa/drivers/dri/i965/brw_tex_layout.c
1821 +++ b/src/mesa/drivers/dri/i965/brw_tex_layout.c
1822 @@ -115,6 +115,8 @@ brw_miptree_layout(struct intel_context *intel, struct intel_mipmap_tree *mt)
1823                intel_miptree_set_image_offset(mt, level, q, x, y);
1824                x += pack_x_pitch;
1825             }
1826 +            if (x > mt->total_width)
1827 +               mt->total_width = x;
1828
1829             x = 0;
1830             y += pack_y_pitch;
1831 @@ -135,10 +137,9 @@ brw_miptree_layout(struct intel_context *intel, struct intel_mipmap_tree *mt)
1832                pack_x_nr <<= 1;
1833             }
1834          } else {
1835 +            pack_x_nr <<= 1;
1836             if (pack_x_pitch > 4) {
1837                pack_x_pitch >>= 1;
1838 -              pack_x_nr <<= 1;
1839 -              assert(pack_x_pitch * pack_x_nr <= mt->total_width);
1840             }
1841
1842             if (pack_y_pitch > 2) {
1843 diff --git a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
1844 index 51d3a46..97ae489 100644
1845 --- a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
1846 +++ b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
1847 @@ -916,12 +916,48 @@ brw_update_renderbuffer_surface(struct brw_context *brw,
1848     struct gl_context *ctx = &intel->ctx;
1849     struct intel_renderbuffer *irb = intel_renderbuffer(rb);
1850     struct intel_mipmap_tree *mt = irb->mt;
1851 -   struct intel_region *region = irb->mt->region;
1852 +   struct intel_region *region;
1853     uint32_t *surf;
1854     uint32_t tile_x, tile_y;
1855     uint32_t format = 0;
1856     gl_format rb_format = intel_rb_format(irb);
1857
1858 +   if (irb->tex_image && !brw->has_surface_tile_offset) {
1859 +      intel_renderbuffer_tile_offsets(irb, &tile_x, &tile_y);
1860 +
1861 +      if (tile_x != 0 || tile_y != 0) {
1862 +        /* Original gen4 hardware couldn't draw to a non-tile-aligned
1863 +         * destination in a miptree unless you actually setup your renderbuffer
1864 +         * as a miptree and used the fragile lod/array_index/etc. controls to
1865 +         * select the image.  So, instead, we just make a new single-level
1866 +         * miptree and render into that.
1867 +         */
1868 +        struct intel_context *intel = intel_context(ctx);
1869 +        struct intel_texture_image *intel_image =
1870 +           intel_texture_image(irb->tex_image);
1871 +        struct intel_mipmap_tree *new_mt;
1872 +        int width, height, depth;
1873 +
1874 +        intel_miptree_get_dimensions_for_image(irb->tex_image, &width, &height, &depth);
1875 +
1876 +        new_mt = intel_miptree_create(intel, irb->tex_image->TexObject->Target,
1877 +                                      intel_image->base.Base.TexFormat,
1878 +                                      intel_image->base.Base.Level,
1879 +                                      intel_image->base.Base.Level,
1880 +                                      width, height, depth,
1881 +                                      true);
1882 +
1883 +        intel_miptree_copy_teximage(intel, intel_image, new_mt);
1884 +        intel_miptree_reference(&irb->mt, intel_image->mt);
1885 +        intel_renderbuffer_set_draw_offset(irb);
1886 +        intel_miptree_release(&new_mt);
1887 +
1888 +        mt = irb->mt;
1889 +      }
1890 +   }
1891 +
1892 +   region = irb->mt->region;
1893 +
1894     surf = brw_state_batch(brw, AUB_TRACE_SURFACE_STATE,
1895                           6 * 4, 32, &brw->bind.surf_offset[unit]);
1896
1897 diff --git a/src/mesa/drivers/dri/i965/gen6_sampler_state.c b/src/mesa/drivers/dri/i965/gen6_sampler_state.c
1898 index 15cae0a..a9a9df5 100644
1899 --- a/src/mesa/drivers/dri/i965/gen6_sampler_state.c
1900 +++ b/src/mesa/drivers/dri/i965/gen6_sampler_state.c
1901 @@ -41,7 +41,7 @@ upload_sampler_state_pointers(struct brw_context *brw)
1902              GS_SAMPLER_STATE_CHANGE |
1903              PS_SAMPLER_STATE_CHANGE |
1904              (4 - 2));
1905 -   OUT_BATCH(0); /* VS */
1906 +   OUT_BATCH(brw->sampler.offset); /* VS */
1907     OUT_BATCH(0); /* GS */
1908     OUT_BATCH(brw->sampler.offset);
1909     ADVANCE_BATCH();
1910 diff --git a/src/mesa/drivers/dri/intel/intel_context.c b/src/mesa/drivers/dri/intel/intel_context.c
1911 index d3c0d70..9cdd804 100644
1912 --- a/src/mesa/drivers/dri/intel/intel_context.c
1913 +++ b/src/mesa/drivers/dri/intel/intel_context.c
1914 @@ -1225,6 +1225,10 @@ intel_process_dri2_buffer_with_separate_stencil(struct intel_context *intel,
1915     if (!rb)
1916        return;
1917
1918 +   /* Check if we failed to allocate the depth miptree earlier. */
1919 +   if (buffer->attachment == __DRI_BUFFER_HIZ && rb->mt == NULL)
1920 +     return;
1921 +
1922     /* If the renderbuffer's and DRIbuffer's regions match, then continue. */
1923     if ((buffer->attachment != __DRI_BUFFER_HIZ &&
1924         rb->mt &&
1925 @@ -1266,6 +1270,7 @@ intel_process_dri2_buffer_with_separate_stencil(struct intel_context *intel,
1926      * due to failure to allocate new storage.
1927      */
1928     if (buffer->attachment == __DRI_BUFFER_HIZ) {
1929 +      assert(rb->mt);
1930        intel_miptree_release(&rb->mt->hiz_mt);
1931     } else {
1932        intel_miptree_release(&rb->mt);
1933 @@ -1291,6 +1296,7 @@ intel_process_dri2_buffer_with_separate_stencil(struct intel_context *intel,
1934
1935     /* Associate buffer with new storage. */
1936     if (buffer->attachment == __DRI_BUFFER_HIZ) {
1937 +      assert(rb->mt);
1938        rb->mt->hiz_mt = mt;
1939     } else {
1940        rb->mt = mt;
1941 diff --git a/src/mesa/drivers/dri/intel/intel_fbo.c b/src/mesa/drivers/dri/intel/intel_fbo.c
1942 index 185602a..c5097c3 100644
1943 --- a/src/mesa/drivers/dri/intel/intel_fbo.c
1944 +++ b/src/mesa/drivers/dri/intel/intel_fbo.c
1945 @@ -553,22 +553,6 @@ intel_renderbuffer_tile_offsets(struct intel_renderbuffer *irb,
1946     }
1947  }
1948
1949 -#ifndef I915
1950 -static bool
1951 -need_tile_offset_workaround(struct brw_context *brw,
1952 -                           struct intel_renderbuffer *irb)
1953 -{
1954 -   uint32_t tile_x, tile_y;
1955 -
1956 -   if (brw->has_surface_tile_offset)
1957 -      return false;
1958 -
1959 -   intel_renderbuffer_tile_offsets(irb, &tile_x, &tile_y);
1960 -
1961 -   return tile_x != 0 || tile_y != 0;
1962 -}
1963 -#endif
1964 -
1965  /**
1966   * Called by glFramebufferTexture[123]DEXT() (and other places) to
1967   * prepare for rendering into texture memory.  This might be called
1968 @@ -626,42 +610,13 @@ intel_render_texture(struct gl_context * ctx,
1969         return;
1970     }
1971
1972 +   irb->tex_image = image;
1973 +
1974     DBG("Begin render %s texture tex=%u w=%d h=%d refcount=%d\n",
1975         _mesa_get_format_name(image->TexFormat),
1976         att->Texture->Name, image->Width, image->Height,
1977         irb->Base.Base.RefCount);
1978
1979 -   intel_image->used_as_render_target = true;
1980 -
1981 -#ifndef I915
1982 -   if (need_tile_offset_workaround(brw_context(ctx), irb)) {
1983 -      /* Original gen4 hardware couldn't draw to a non-tile-aligned
1984 -       * destination in a miptree unless you actually setup your
1985 -       * renderbuffer as a miptree and used the fragile
1986 -       * lod/array_index/etc. controls to select the image.  So,
1987 -       * instead, we just make a new single-level miptree and render
1988 -       * into that.
1989 -       */
1990 -      struct intel_context *intel = intel_context(ctx);
1991 -      struct intel_mipmap_tree *new_mt;
1992 -      int width, height, depth;
1993 -
1994 -      intel_miptree_get_dimensions_for_image(image, &width, &height, &depth);
1995 -
1996 -      new_mt = intel_miptree_create(intel, image->TexObject->Target,
1997 -                                   intel_image->base.Base.TexFormat,
1998 -                                   intel_image->base.Base.Level,
1999 -                                   intel_image->base.Base.Level,
2000 -                                    width, height, depth,
2001 -                                   true);
2002 -
2003 -      intel_miptree_copy_teximage(intel, intel_image, new_mt);
2004 -      intel_renderbuffer_set_draw_offset(irb);
2005 -
2006 -      intel_miptree_reference(&irb->mt, intel_image->mt);
2007 -      intel_miptree_release(&new_mt);
2008 -   }
2009 -#endif
2010     /* update drawing region, etc */
2011     intel_draw_buffer(ctx);
2012  }
2013 @@ -678,14 +633,13 @@ intel_finish_render_texture(struct gl_context * ctx,
2014     struct gl_texture_object *tex_obj = att->Texture;
2015     struct gl_texture_image *image =
2016        tex_obj->Image[att->CubeMapFace][att->TextureLevel];
2017 -   struct intel_texture_image *intel_image = intel_texture_image(image);
2018 +   struct intel_renderbuffer *irb = intel_renderbuffer(att->Renderbuffer);
2019
2020     DBG("Finish render %s texture tex=%u\n",
2021         _mesa_get_format_name(image->TexFormat), att->Texture->Name);
2022
2023 -   /* Flag that this image may now be validated into the object's miptree. */
2024 -   if (intel_image)
2025 -      intel_image->used_as_render_target = false;
2026 +   if (irb)
2027 +      irb->tex_image = NULL;
2028
2029     /* Since we've (probably) rendered to the texture and will (likely) use
2030      * it in the texture domain later on in this batchbuffer, flush the
2031 diff --git a/src/mesa/drivers/dri/intel/intel_fbo.h b/src/mesa/drivers/dri/intel/intel_fbo.h
2032 index a2c1b1a..724f141 100644
2033 --- a/src/mesa/drivers/dri/intel/intel_fbo.h
2034 +++ b/src/mesa/drivers/dri/intel/intel_fbo.h
2035 @@ -47,6 +47,9 @@ struct intel_renderbuffer
2036     struct intel_mipmap_tree *mt; /**< The renderbuffer storage. */
2037     drm_intel_bo *map_bo;
2038
2039 +   /* Current texture image this renderbuffer is attached to. */
2040 +   struct gl_texture_image *tex_image;
2041 +
2042     /**
2043      * \name Miptree view
2044      * \{
2045 diff --git a/src/mesa/drivers/dri/intel/intel_tex_obj.h b/src/mesa/drivers/dri/intel/intel_tex_obj.h
2046 index 8b278ba..d1a5f05 100644
2047 --- a/src/mesa/drivers/dri/intel/intel_tex_obj.h
2048 +++ b/src/mesa/drivers/dri/intel/intel_tex_obj.h
2049 @@ -65,7 +65,6 @@ struct intel_texture_image
2050      * Else there is no image data.
2051      */
2052     struct intel_mipmap_tree *mt;
2053 -   bool used_as_render_target;
2054  };
2055
2056  static INLINE struct intel_texture_object *
2057 diff --git a/src/mesa/drivers/dri/intel/intel_tex_validate.c b/src/mesa/drivers/dri/intel/intel_tex_validate.c
2058 index b96f2a4..a63068b 100644
2059 --- a/src/mesa/drivers/dri/intel/intel_tex_validate.c
2060 +++ b/src/mesa/drivers/dri/intel/intel_tex_validate.c
2061 @@ -97,14 +97,8 @@ intel_finalize_mipmap_tree(struct intel_context *intel, GLuint unit)
2062          /* skip too small size mipmap */
2063          if (intelImage == NULL)
2064                  break;
2065 -        /* Need to import images in main memory or held in other trees.
2066 -         * If it's a render target, then its data isn't needed to be in
2067 -         * the object tree (otherwise we'd be FBO incomplete), and we need
2068 -         * to keep track of the image's MT as needing to be pulled in still,
2069 -         * or we'll lose the rendering that's done to it.
2070 -          */
2071 -         if (intelObj->mt != intelImage->mt &&
2072 -            !intelImage->used_as_render_target) {
2073 +
2074 +         if (intelObj->mt != intelImage->mt) {
2075              intel_miptree_copy_teximage(intel, intelImage, intelObj->mt);
2076           }
2077        }
2078 diff --git a/src/mesa/drivers/windows/gdi/wmesa.c b/src/mesa/drivers/windows/gdi/wmesa.c
2079 index 40aa56e..93da05f 100644
2080 --- a/src/mesa/drivers/windows/gdi/wmesa.c
2081 +++ b/src/mesa/drivers/windows/gdi/wmesa.c
2082 @@ -243,39 +243,9 @@ static void wmesa_flush(struct gl_context *ctx)
2083  /*****                   CLEAR Functions                          *****/
2084  /**********************************************************************/
2085
2086 -/* If we do not implement these, Mesa clears the buffers via the pixel
2087 - * span writing interface, which is very slow for a clear operation.
2088 - */
2089 -
2090 -/*
2091 - * Set the color used to clear the color buffer.
2092 - */
2093 -static void clear_color(struct gl_context *ctx,
2094 -                        const union gl_color_union color)
2095 -{
2096 -    WMesaContext pwc = wmesa_context(ctx);
2097 -    GLubyte col[3];
2098 -
2099 -    UNCLAMPED_FLOAT_TO_UBYTE(col[0], color.f[0]);
2100 -    UNCLAMPED_FLOAT_TO_UBYTE(col[1], color.f[1]);
2101 -    UNCLAMPED_FLOAT_TO_UBYTE(col[2], color.f[2]);
2102 -    pwc->clearColorRef = RGB(col[0], col[1], col[2]);
2103 -    DeleteObject(pwc->clearPen);
2104 -    DeleteObject(pwc->clearBrush);
2105 -    pwc->clearPen = CreatePen(PS_SOLID, 1, pwc->clearColorRef);
2106 -    pwc->clearBrush = CreateSolidBrush(pwc->clearColorRef);
2107 -}
2108 -
2109 -
2110  /*
2111 - * Clear the specified region of the color buffer using the clear color
2112 - * or index as specified by one of the two functions above.
2113 - *
2114 - * This procedure clears either the front and/or the back COLOR buffers.
2115 - * Only the "left" buffer is cleared since we are not stereo.
2116 - * Clearing of the other non-color buffers is left to the swrast.
2117 + * Clear the color/depth/stencil buffers.
2118   */
2119 -
2120  static void clear(struct gl_context *ctx, GLbitfield mask)
2121  {
2122  #define FLIP(Y)  (ctx->DrawBuffer->Height - (Y) - 1)
2123 @@ -298,6 +268,20 @@ static void clear(struct gl_context *ctx, GLbitfield mask)
2124         return;
2125      }
2126
2127 +    if (mask & BUFFER_BITS_COLOR) {
2128 +       /* setup the clearing color */
2129 +       const union gl_color_union color = ctx->Color.ClearColor;
2130 +       GLubyte col[3];
2131 +       UNCLAMPED_FLOAT_TO_UBYTE(col[0], color.f[0]);
2132 +       UNCLAMPED_FLOAT_TO_UBYTE(col[1], color.f[1]);
2133 +       UNCLAMPED_FLOAT_TO_UBYTE(col[2], color.f[2]);
2134 +       pwc->clearColorRef = RGB(col[0], col[1], col[2]);
2135 +       DeleteObject(pwc->clearPen);
2136 +       DeleteObject(pwc->clearBrush);
2137 +       pwc->clearPen = CreatePen(PS_SOLID, 1, pwc->clearColorRef);
2138 +       pwc->clearBrush = CreateSolidBrush(pwc->clearColorRef);
2139 +    }
2140 +
2141      /* Back buffer */
2142      if (mask & BUFFER_BIT_BACK_LEFT) {
2143
2144 @@ -940,54 +924,6 @@ wmesa_renderbuffer_storage(struct gl_context *ctx,
2145
2146
2147  /**
2148 - * Plug in the Get/PutRow/Values functions for a renderbuffer depending
2149 - * on if we're drawing to the front or back color buffer.
2150 - */
2151 -static void
2152 -wmesa_set_renderbuffer_funcs(struct gl_renderbuffer *rb, int pixelformat,
2153 -                             int cColorBits, int double_buffer)
2154 -{
2155 -    if (double_buffer) {
2156 -        /* back buffer */
2157 -       /* Picking the correct span functions is important because
2158 -        * the DIB was allocated with the indicated depth. */
2159 -       switch(pixelformat) {
2160 -       case PF_5R6G5B:
2161 -           rb->PutRow = write_rgba_span_16;
2162 -           rb->PutValues = write_rgba_pixels_16;
2163 -           rb->GetRow = read_rgba_span_16;
2164 -           rb->GetValues = read_rgba_pixels_16;
2165 -           break;
2166 -       case PF_8R8G8B:
2167 -               if (cColorBits == 24)
2168 -               {
2169 -                   rb->PutRow = write_rgba_span_24;
2170 -                   rb->PutValues = write_rgba_pixels_24;
2171 -                   rb->GetRow = read_rgba_span_24;
2172 -                   rb->GetValues = read_rgba_pixels_24;
2173 -               }
2174 -               else
2175 -               {
2176 -                    rb->PutRow = write_rgba_span_32;
2177 -                    rb->PutValues = write_rgba_pixels_32;
2178 -                    rb->GetRow = read_rgba_span_32;
2179 -                    rb->GetValues = read_rgba_pixels_32;
2180 -               }
2181 -           break;
2182 -       default:
2183 -           break;
2184 -       }
2185 -    }
2186 -    else {
2187 -        /* front buffer (actual Windows window) */
2188 -       rb->PutRow = write_rgba_span_front;
2189 -       rb->PutValues = write_rgba_pixels_front;
2190 -       rb->GetRow = read_rgba_span_front;
2191 -       rb->GetValues = read_rgba_pixels_front;
2192 -    }
2193 -}
2194 -
2195 -/**
2196   * Called by ctx->Driver.ResizeBuffers()
2197   * Resize the front/back colorbuffers to match the latest window size.
2198   */
2199 @@ -1143,7 +1079,6 @@ WMesaContext WMesaCreateContext(HDC hDC,
2200      functions.GetBufferSize = wmesa_get_buffer_size;
2201      functions.Flush = wmesa_flush;
2202      functions.Clear = clear;
2203 -    functions.ClearColor = clear_color;
2204      functions.ResizeBuffers = wmesa_resize_buffers;
2205      functions.Viewport = wmesa_viewport;
2206
2207 @@ -1275,11 +1210,9 @@ void WMesaMakeCurrent(WMesaContext c, HDC hdc)
2208          if (visual->doubleBufferMode == 1) {
2209              rb = wmesa_new_renderbuffer();
2210              _mesa_add_renderbuffer(&pwfb->Base, BUFFER_BACK_LEFT, rb);
2211 -            wmesa_set_renderbuffer_funcs(rb, pwfb->pixelformat, pwfb->cColorBits, 1);
2212         }
2213          rb = wmesa_new_renderbuffer();
2214          _mesa_add_renderbuffer(&pwfb->Base, BUFFER_FRONT_LEFT, rb);
2215 -        wmesa_set_renderbuffer_funcs(rb, pwfb->pixelformat, pwfb->cColorBits, 0);
2216
2217         /* Let Mesa own the Depth, Stencil, and Accum buffers */
2218          _swrast_add_soft_renderbuffers(&pwfb->Base,
2219 diff --git a/src/mesa/main/readpix.c b/src/mesa/main/readpix.c
2220 index 5b3c246..f3a0d10 100644
2221 --- a/src/mesa/main/readpix.c
2222 +++ b/src/mesa/main/readpix.c
2223 @@ -196,6 +196,11 @@ read_stencil_pixels( struct gl_context *ctx,
2224     ctx->Driver.UnmapRenderbuffer(ctx, rb);
2225  }
2226
2227 +
2228 +/**
2229 + * Try to do glReadPixels of RGBA data using a simple memcpy or swizzle.
2230 + * \return GL_TRUE if successful, GL_FALSE otherwise (use the slow path)
2231 + */
2232  static GLboolean
2233  fast_read_rgba_pixels_memcpy( struct gl_context *ctx,
2234                               GLint x, GLint y,
2235 @@ -208,8 +213,20 @@ fast_read_rgba_pixels_memcpy( struct gl_context *ctx,
2236     struct gl_renderbuffer *rb = ctx->ReadBuffer->_ColorReadBuffer;
2237     GLubyte *dst, *map;
2238     int dstStride, stride, j, texelBytes;
2239 +   GLboolean swizzle_rb = GL_FALSE, copy_xrgb = GL_FALSE;
2240
2241 -   if (!_mesa_format_matches_format_and_type(rb->Format, format, type))
2242 +   /* XXX we could check for other swizzle/special cases here as needed */
2243 +   if (rb->Format == MESA_FORMAT_RGBA8888_REV &&
2244 +       format == GL_BGRA &&
2245 +       type == GL_UNSIGNED_INT_8_8_8_8_REV) {
2246 +      swizzle_rb = GL_TRUE;
2247 +   }
2248 +   else if (rb->Format == MESA_FORMAT_XRGB8888 &&
2249 +            format == GL_BGRA &&
2250 +            type == GL_UNSIGNED_INT_8_8_8_8_REV) {
2251 +      copy_xrgb = GL_TRUE;
2252 +   }
2253 +   else if (!_mesa_format_matches_format_and_type(rb->Format, format, type))
2254        return GL_FALSE;
2255
2256     /* check for things we can't handle here */
2257 @@ -240,10 +257,39 @@ fast_read_rgba_pixels_memcpy( struct gl_context *ctx,
2258     }
2259
2260     texelBytes = _mesa_get_format_bytes(rb->Format);
2261 -   for (j = 0; j < height; j++) {
2262 -      memcpy(dst, map, width * texelBytes);
2263 -      dst += dstStride;
2264 -      map += stride;
2265 +
2266 +   if (swizzle_rb) {
2267 +      /* swap R/B */
2268 +      for (j = 0; j < height; j++) {
2269 +         int i;
2270 +         for (i = 0; i < width; i++) {
2271 +            GLuint *dst4 = (GLuint *) dst, *map4 = (GLuint *) map;
2272 +            GLuint pixel = map4[i];
2273 +            dst4[i] = (pixel & 0xff00ff00)
2274 +                   | ((pixel & 0x00ff0000) >> 16)
2275 +                   | ((pixel & 0x000000ff) << 16);
2276 +         }
2277 +         dst += dstStride;
2278 +         map += stride;
2279 +      }
2280 +   } else if (copy_xrgb) {
2281 +      /* convert xrgb -> argb */
2282 +      for (j = 0; j < height; j++) {
2283 +         GLuint *dst4 = (GLuint *) dst, *map4 = (GLuint *) map;
2284 +         int i;
2285 +         for (i = 0; i < width; i++) {
2286 +            dst4[i] = map4[i] | 0xff000000;  /* set A=0xff */
2287 +         }
2288 +         dst += dstStride;
2289 +         map += stride;
2290 +      }
2291 +   } else {
2292 +      /* just memcpy */
2293 +      for (j = 0; j < height; j++) {
2294 +         memcpy(dst, map, width * texelBytes);
2295 +         dst += dstStride;
2296 +         map += stride;
2297 +      }
2298     }
2299
2300     ctx->Driver.UnmapRenderbuffer(ctx, rb);