Merge commit 'origin/gallium-0.1' into gallium-0.2

Conflicts:

	src/mesa/shader/slang/slang_link.c

30 files changed, 336 insertions, 73 deletions

diff --git a/src/gallium/auxiliary/cso_cache/cso_context.c b/src/gallium/auxiliary/cso_cache/cso_context.c
index f22ba40824..68508f24de 100644
--- a/src/gallium/auxiliary/cso_cache/cso_context.c
+++ b/src/gallium/auxiliary/cso_cache/cso_context.c
@@ -80,6 +80,10 @@ struct cso_context {
 };
 
 
+static void
+free_framebuffer_state(struct pipe_framebuffer_state *fb);
+
+
 static boolean delete_blend_state(struct cso_context *ctx, void *state)
 {
    struct cso_blend *cso = (struct cso_blend *)state;
@@ -252,6 +256,9 @@ void cso_release_all( struct cso_context *ctx )
       pipe_texture_reference(&ctx->textures_saved[i], NULL);
    }
 
+   free_framebuffer_state(&ctx->fb);
+   free_framebuffer_state(&ctx->fb_saved);
+
    if (ctx->cache) {
       cso_cache_delete( ctx->cache );
       ctx->cache = NULL;
@@ -765,12 +772,42 @@ void cso_restore_vertex_shader(struct cso_context *ctx)
 }
 
 
+/**
+ * Copy framebuffer state from src to dst with refcounting of surfaces.
+ */
+static void
+copy_framebuffer_state(struct pipe_framebuffer_state *dst,
+                       const struct pipe_framebuffer_state *src)
+{
+   uint i;
+
+   dst->width = src->width;
+   dst->height = src->height;
+   dst->num_cbufs = src->num_cbufs;
+   for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
+      pipe_surface_reference(&dst->cbufs[i], src->cbufs[i]);
+   }
+   pipe_surface_reference(&dst->zsbuf, src->zsbuf);
+}
+
+
+static void
+free_framebuffer_state(struct pipe_framebuffer_state *fb)
+{
+   uint i;
+
+   for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
+      pipe_surface_reference(&fb->cbufs[i], NULL);
+   }
+   pipe_surface_reference(&fb->zsbuf, NULL);
+}
+
 
 enum pipe_error cso_set_framebuffer(struct cso_context *ctx,
                                     const struct pipe_framebuffer_state *fb)
 {
    if (memcmp(&ctx->fb, fb, sizeof(*fb)) != 0) {
-      ctx->fb = *fb;
+      copy_framebuffer_state(&ctx->fb, fb);
       ctx->pipe->set_framebuffer_state(ctx->pipe, fb);
    }
    return PIPE_OK;
@@ -778,14 +815,15 @@ enum pipe_error cso_set_framebuffer(struct cso_context *ctx,
 
 void cso_save_framebuffer(struct cso_context *ctx)
 {
-   ctx->fb_saved = ctx->fb;
+   copy_framebuffer_state(&ctx->fb_saved, &ctx->fb);
 }
 
 void cso_restore_framebuffer(struct cso_context *ctx)
 {
    if (memcmp(&ctx->fb, &ctx->fb_saved, sizeof(ctx->fb))) {
-      ctx->fb = ctx->fb_saved;
+      copy_framebuffer_state(&ctx->fb, &ctx->fb_saved);
       ctx->pipe->set_framebuffer_state(ctx->pipe, &ctx->fb);
+      free_framebuffer_state(&ctx->fb_saved);
    }
 }
 
diff --git a/src/gallium/auxiliary/draw/draw_pipe_wide_point.c b/src/gallium/auxiliary/draw/draw_pipe_wide_point.c
index 4f1326053d..e1af9e56a2 100644
--- a/src/gallium/auxiliary/draw/draw_pipe_wide_point.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_wide_point.c
@@ -222,8 +222,8 @@ static void widepoint_first_point( struct draw_stage *stage,
       /* find fragment shader PointCoord/Fog input */
       wide->point_coord_fs_input = 0; /* XXX fix this! */
 
-      /* setup extra vp output */
-      draw->extra_vp_outputs.semantic_name = TGSI_SEMANTIC_FOG;
+      /* setup extra vp output (point coord implemented as a texcoord) */
+      draw->extra_vp_outputs.semantic_name = TGSI_SEMANTIC_GENERIC;
       draw->extra_vp_outputs.semantic_index = 0;
       draw->extra_vp_outputs.slot = draw->vs.num_vs_outputs;
    }
diff --git a/src/gallium/auxiliary/pipebuffer/pb_buffer_malloc.c b/src/gallium/auxiliary/pipebuffer/pb_buffer_malloc.c
index 20fc87b39d..1bf22a2ec0 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_buffer_malloc.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_buffer_malloc.c
@@ -129,7 +129,7 @@ pb_malloc_buffer_create(size_t size,
 
 
 static struct pb_buffer *
-pb_malloc_buffer_create_buffer(struct pb_manager *mgr, 
+pb_malloc_bufmgr_create_buffer(struct pb_manager *mgr, 
                                size_t size,
                                const struct pb_desc *desc) 
 {
@@ -138,6 +138,13 @@ pb_malloc_buffer_create_buffer(struct pb_manager *mgr,
 
 
 static void
+pb_malloc_bufmgr_flush(struct pb_manager *mgr) 
+{
+   /* No-op */
+}
+
+
+static void
 pb_malloc_bufmgr_destroy(struct pb_manager *mgr) 
 {
    /* No-op */
@@ -146,8 +153,9 @@ pb_malloc_bufmgr_destroy(struct pb_manager *mgr)
 
 static struct pb_manager 
 pb_malloc_bufmgr = {
-   pb_malloc_buffer_create_buffer,
-   pb_malloc_bufmgr_destroy
+   pb_malloc_bufmgr_destroy,
+   pb_malloc_bufmgr_create_buffer,
+   pb_malloc_bufmgr_flush
 };
 
 
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h b/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h
index 32867029ee..cafbee045a 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h
@@ -69,13 +69,22 @@ struct pipe_winsys;
  */
 struct pb_manager
 {
+   void
+   (*destroy)( struct pb_manager *mgr );
+
    struct pb_buffer *
    (*create_buffer)( struct pb_manager *mgr, 
 	             size_t size,
 	             const struct pb_desc *desc);
 
+   /**
+    * Flush all temporary-held buffers.
+    * 
+    * Used mostly to aid debugging memory issues or to clean up resources when 
+    * the drivers are long lived.
+    */
    void
-   (*destroy)( struct pb_manager *mgr );
+   (*flush)( struct pb_manager *mgr );
 };
 
 
@@ -153,9 +162,6 @@ struct pb_manager *
 pb_cache_manager_create(struct pb_manager *provider, 
                      	unsigned usecs); 
 
-void
-pb_cache_flush(struct pb_manager *mgr);
-
 
 /** 
  * Fenced buffer manager.
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_alt.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_alt.c
index 2afaeafa1a..c956924cc7 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_alt.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_alt.c
@@ -76,6 +76,21 @@ pb_alt_manager_create_buffer(struct pb_manager *_mgr,
 
 
 static void
+pb_alt_manager_flush(struct pb_manager *_mgr)
+{
+   struct pb_alt_manager *mgr = pb_alt_manager(_mgr);
+   
+   assert(mgr->provider1->flush);
+   if(mgr->provider1->flush)
+      mgr->provider1->flush(mgr->provider1);
+   
+   assert(mgr->provider2->flush);
+   if(mgr->provider2->flush)
+      mgr->provider2->flush(mgr->provider2);
+}
+
+
+static void
 pb_alt_manager_destroy(struct pb_manager *mgr)
 {
    FREE(mgr);
@@ -97,6 +112,7 @@ pb_alt_manager_create(struct pb_manager *provider1,
 
    mgr->base.destroy = pb_alt_manager_destroy;
    mgr->base.create_buffer = pb_alt_manager_create_buffer;
+   mgr->base.flush = pb_alt_manager_flush;
    mgr->provider1 = provider1;
    mgr->provider2 = provider2;
       
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_cache.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_cache.c
index 1ec422fb19..8f118874ec 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_cache.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_cache.c
@@ -306,8 +306,8 @@ pb_cache_manager_create_buffer(struct pb_manager *_mgr,
 }
 
 
-void
-pb_cache_flush(struct pb_manager *_mgr)
+static void
+pb_cache_manager_flush(struct pb_manager *_mgr)
 {
    struct pb_cache_manager *mgr = pb_cache_manager(_mgr);
    struct list_head *curr, *next;
@@ -323,13 +323,17 @@ pb_cache_flush(struct pb_manager *_mgr)
       next = curr->next;
    }
    pipe_mutex_unlock(mgr->mutex);
+   
+   assert(mgr->provider->flush);
+   if(mgr->provider->flush)
+      mgr->provider->flush(mgr->provider);
 }
 
 
 static void
 pb_cache_manager_destroy(struct pb_manager *mgr)
 {
-   pb_cache_flush(mgr);
+   pb_cache_manager_flush(mgr);
    FREE(mgr);
 }
 
@@ -349,6 +353,7 @@ pb_cache_manager_create(struct pb_manager *provider,
 
    mgr->base.destroy = pb_cache_manager_destroy;
    mgr->base.create_buffer = pb_cache_manager_create_buffer;
+   mgr->base.flush = pb_cache_manager_flush;
    mgr->provider = provider;
    mgr->usecs = usecs;
    LIST_INITHEAD(&mgr->delayed);
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_debug.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_debug.c
index 5f1ed3e5a8..1675e6e182 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_debug.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_debug.c
@@ -314,6 +314,16 @@ pb_debug_manager_create_buffer(struct pb_manager *_mgr,
 
 
 static void
+pb_debug_manager_flush(struct pb_manager *_mgr)
+{
+   struct pb_debug_manager *mgr = pb_debug_manager(_mgr);
+   assert(mgr->provider->flush);
+   if(mgr->provider->flush)
+      mgr->provider->flush(mgr->provider);
+}
+
+
+static void
 pb_debug_manager_destroy(struct pb_manager *_mgr)
 {
    struct pb_debug_manager *mgr = pb_debug_manager(_mgr);
@@ -336,6 +346,7 @@ pb_debug_manager_create(struct pb_manager *provider, size_t band_size)
 
    mgr->base.destroy = pb_debug_manager_destroy;
    mgr->base.create_buffer = pb_debug_manager_create_buffer;
+   mgr->base.flush = pb_debug_manager_flush;
    mgr->provider = provider;
    mgr->band_size = band_size;
       
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_fenced.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_fenced.c
index 8fc63ce648..633ee70a75 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_fenced.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_fenced.c
@@ -95,6 +95,19 @@ fenced_bufmgr_create_buffer(struct pb_manager *mgr,
 
 
 static void
+fenced_bufmgr_flush(struct pb_manager *mgr)
+{
+   struct fenced_pb_manager *fenced_mgr = fenced_pb_manager(mgr);
+
+   fenced_buffer_list_check_free(fenced_mgr->fenced_list, TRUE);
+
+   assert(fenced_mgr->provider->flush);
+   if(fenced_mgr->provider->flush)
+      fenced_mgr->provider->flush(fenced_mgr->provider);
+}
+
+
+static void
 fenced_bufmgr_destroy(struct pb_manager *mgr)
 {
    struct fenced_pb_manager *fenced_mgr = fenced_pb_manager(mgr);
@@ -123,6 +136,7 @@ fenced_bufmgr_create(struct pb_manager *provider,
 
    fenced_mgr->base.destroy = fenced_bufmgr_destroy;
    fenced_mgr->base.create_buffer = fenced_bufmgr_create_buffer;
+   fenced_mgr->base.flush = fenced_bufmgr_flush;
 
    fenced_mgr->provider = provider;
    fenced_mgr->fenced_list = fenced_buffer_list_create(winsys);
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_mm.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_mm.c
index e8c7f8e1f8..fe80ca30ee 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_mm.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_mm.c
@@ -200,6 +200,13 @@ mm_bufmgr_create_buffer(struct pb_manager *mgr,
 
 
 static void
+mm_bufmgr_flush(struct pb_manager *mgr)
+{
+   /* No-op */
+}
+
+
+static void
 mm_bufmgr_destroy(struct pb_manager *mgr)
 {
    struct mm_pb_manager *mm = mm_pb_manager(mgr);
@@ -230,8 +237,9 @@ mm_bufmgr_create_from_buffer(struct pb_buffer *buffer,
    if (!mm)
       return NULL;
 
-   mm->base.create_buffer = mm_bufmgr_create_buffer;
    mm->base.destroy = mm_bufmgr_destroy;
+   mm->base.create_buffer = mm_bufmgr_create_buffer;
+   mm->base.flush = mm_bufmgr_flush;
 
    mm->size = size;
    mm->align2 = align2; /* 64-byte alignment */
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_pool.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_pool.c
index 3ef72c5bbb..61ac291ed7 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_pool.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_pool.c
@@ -203,6 +203,13 @@ pool_bufmgr_create_buffer(struct pb_manager *mgr,
 
 
 static void
+pool_bufmgr_flush(struct pb_manager *mgr)
+{
+   /* No-op */
+}
+
+
+static void
 pool_bufmgr_destroy(struct pb_manager *mgr)
 {
    struct pool_pb_manager *pool = pool_pb_manager(mgr);
@@ -238,6 +245,7 @@ pool_bufmgr_create(struct pb_manager *provider,
 
    pool->base.destroy = pool_bufmgr_destroy;
    pool->base.create_buffer = pool_bufmgr_create_buffer;
+   pool->base.flush = pool_bufmgr_flush;
 
    LIST_INITHEAD(&pool->free);
 
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_slab.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_slab.c
index 8698c4cff6..2a80154920 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_slab.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_slab.c
@@ -407,6 +407,17 @@ pb_slab_manager_create_buffer(struct pb_manager *_mgr,
 
 
 static void
+pb_slab_manager_flush(struct pb_manager *_mgr)
+{
+   struct pb_slab_manager *mgr = pb_slab_manager(_mgr);
+
+   assert(mgr->provider->flush);
+   if(mgr->provider->flush)
+      mgr->provider->flush(mgr->provider);
+}
+
+
+static void
 pb_slab_manager_destroy(struct pb_manager *_mgr)
 {
    struct pb_slab_manager *mgr = pb_slab_manager(_mgr);
@@ -430,6 +441,7 @@ pb_slab_manager_create(struct pb_manager *provider,
 
    mgr->base.destroy = pb_slab_manager_destroy;
    mgr->base.create_buffer = pb_slab_manager_create_buffer;
+   mgr->base.flush = pb_slab_manager_flush;
 
    mgr->provider = provider;
    mgr->bufSize = bufSize;
@@ -466,6 +478,19 @@ pb_slab_range_manager_create_buffer(struct pb_manager *_mgr,
 
 
 static void
+pb_slab_range_manager_flush(struct pb_manager *_mgr)
+{
+   struct pb_slab_range_manager *mgr = pb_slab_range_manager(_mgr);
+
+   /* Individual slabs don't hold any temporary buffers so no need to call them */
+   
+   assert(mgr->provider->flush);
+   if(mgr->provider->flush)
+      mgr->provider->flush(mgr->provider);
+}
+
+
+static void
 pb_slab_range_manager_destroy(struct pb_manager *_mgr)
 {
    struct pb_slab_range_manager *mgr = pb_slab_range_manager(_mgr);
@@ -499,6 +524,7 @@ pb_slab_range_manager_create(struct pb_manager *provider,
 
    mgr->base.destroy = pb_slab_range_manager_destroy;
    mgr->base.create_buffer = pb_slab_range_manager_create_buffer;
+   mgr->base.flush = pb_slab_range_manager_flush;
 
    mgr->provider = provider;
    mgr->minBufSize = minBufSize;
diff --git a/src/gallium/auxiliary/tgsi/Makefile b/src/gallium/auxiliary/tgsi/Makefile
index 806a2bd4c5..c7155a9316 100644
--- a/src/gallium/auxiliary/tgsi/Makefile
+++ b/src/gallium/auxiliary/tgsi/Makefile
@@ -4,6 +4,7 @@ include $(TOP)/configs/current
 LIBNAME = tgsi
 
 C_SOURCES = \
+	tgsi_sanity.c \
 	tgsi_build.c \
 	tgsi_dump.c \
 	tgsi_exec.c \
@@ -12,6 +13,7 @@ C_SOURCES = \
 	tgsi_parse.c \
 	tgsi_scan.c \
 	tgsi_sse2.c \
+	tgsi_text.c \
 	tgsi_transform.c \
 	tgsi_util.c
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_dump.c b/src/gallium/auxiliary/tgsi/tgsi_dump.c
index afc8ffa553..3177f54952 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_dump.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_dump.c
@@ -68,6 +68,7 @@ dump_enum(
 #define CHR(C)          ctx->printf( ctx, "%c", C )
 #define UIX(I)          ctx->printf( ctx, "0x%x", I )
 #define UID(I)          ctx->printf( ctx, "%u", I )
+#define INSTID(I)          ctx->printf( ctx, "% 3u", I )
 #define SID(I)          ctx->printf( ctx, "%d", I )
 #define FLT(F)          ctx->printf( ctx, "%10.4f", F )
 #define ENM(E,ENUMS)    dump_enum( ctx, E, ENUMS, sizeof( ENUMS ) / sizeof( *ENUMS ) )
@@ -315,8 +316,8 @@ iter_instruction(
    uint i;
    boolean first_reg = TRUE;
 
-   UID( instno );
-   CHR( ':' );
+   INSTID( instno );
+   TXT( ": " );
    TXT( tgsi_get_opcode_info( inst->Instruction.Opcode )->mnemonic );
 
    switch (inst->Instruction.Saturate) {
diff --git a/src/gallium/auxiliary/tgsi/tgsi_info.c b/src/gallium/auxiliary/tgsi/tgsi_info.c
index a4899cd4c2..68c7a6b7f5 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_info.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_info.c
@@ -69,7 +69,7 @@ static const struct tgsi_opcode_info opcode_info[TGSI_OPCODE_LAST] =
    { 1, 1, 0, 0, "COS" },
    { 1, 1, 0, 0, "DDX" },
    { 1, 1, 0, 0, "DDY" },
-   { 0, 1, 0, 0, "KILP" },
+   { 0, 0, 0, 0, "KILP" },
    { 1, 1, 0, 0, "PK2H" },
    { 1, 1, 0, 0, "PK2US" },
    { 1, 1, 0, 0, "PK4B" },
@@ -146,7 +146,7 @@ static const struct tgsi_opcode_info opcode_info[TGSI_OPCODE_LAST] =
    { 0, 1, 0, 0, "CALLNZ" },
    { 0, 1, 0, 0, "IFC" },
    { 0, 1, 0, 0, "BREAKC" },
-   { 0, 0, 0, 0, "KIL" },
+   { 0, 1, 0, 0, "KIL" },
    { 0, 0, 0, 0, "END" },
    { 1, 1, 0, 0, "SWZ" }
 };
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sanity.c b/src/gallium/auxiliary/tgsi/tgsi_sanity.c
index c659027296..11659247c0 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_sanity.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_sanity.c
@@ -152,6 +152,12 @@ check_register_usage(
 {
    if (!check_file_name( ctx, file ))
       return FALSE;
+
+   if (index < 0 || index > MAX_REGISTERS) {
+      report_error( ctx, "%s[%i]: Invalid index %s", file_names[file], index, name );
+      return FALSE;
+   }
+
    if (indirect_access) {
       if (!is_any_register_declared( ctx, file ))
          report_error( ctx, "%s: Undeclared %s register", file_names[file], name );
@@ -174,12 +180,10 @@ iter_instruction(
    const struct tgsi_opcode_info *info;
    uint i;
 
-   /* There must be no other instructions after END.
-    */
-   if (ctx->index_of_END != ~0) {
-      report_error( ctx, "Unexpected instruction after END" );
-   }
-   else if (inst->Instruction.Opcode == TGSI_OPCODE_END) {
+   if (inst->Instruction.Opcode == TGSI_OPCODE_END) {
+      if (ctx->index_of_END != ~0) {
+         report_error( ctx, "Too many END instructions" );
+      }
       ctx->index_of_END = ctx->num_instructions;
    }
 
@@ -301,10 +305,10 @@ epilog(
    struct sanity_check_ctx *ctx = (struct sanity_check_ctx *) iter;
    uint file;
 
-   /* There must be an END instruction at the end.
+   /* There must be an END instruction somewhere.
     */
-   if (ctx->index_of_END == ~0 || ctx->index_of_END != ctx->num_instructions - 1) {
-      report_error( ctx, "Expected END at end of instruction sequence" );
+   if (ctx->index_of_END == ~0) {
+      report_error( ctx, "Missing END instruction" );
    }
 
    /* Check if all declared registers were used.
diff --git a/src/gallium/auxiliary/util/p_debug_mem.c b/src/gallium/auxiliary/util/p_debug_mem.c
index ed18c6540e..9511479cbb 100644
--- a/src/gallium/auxiliary/util/p_debug_mem.c
+++ b/src/gallium/auxiliary/util/p_debug_mem.c
@@ -122,8 +122,12 @@ debug_malloc(const char *file, unsigned line, const char *function,
    struct debug_memory_footer *ftr;
    
    hdr = real_malloc(sizeof(*hdr) + size + sizeof(*ftr));
-   if(!hdr)
+   if(!hdr) {
+      debug_printf("%s:%u:%s: out of memory when trying to allocate %lu bytes\n",
+                   file, line, function,
+                   (long unsigned)size);
       return NULL;
+   }
  
    hdr->no = last_no++;
    hdr->file = file;
@@ -219,8 +223,12 @@ debug_realloc(const char *file, unsigned line, const char *function,
 
    /* alloc new */
    new_hdr = real_malloc(sizeof(*new_hdr) + new_size + sizeof(*new_ftr));
-   if(!new_hdr)
+   if(!new_hdr) {
+      debug_printf("%s:%u:%s: out of memory when trying to allocate %lu bytes\n",
+                   file, line, function,
+                   (long unsigned)new_size);
       return NULL;
+   }
    new_hdr->no = old_hdr->no;
    new_hdr->file = old_hdr->file;
    new_hdr->line = old_hdr->line;
@@ -261,8 +269,19 @@ debug_memory_end(unsigned long start_no)
    for (; entry != &list; entry = entry->prev) {
       struct debug_memory_header *hdr;
       void *ptr;
+      struct debug_memory_footer *ftr;
+
       hdr = LIST_ENTRY(struct debug_memory_header, entry, head);
       ptr = data_from_header(hdr);
+      ftr = footer_from_header(hdr);
+
+      if(hdr->magic != DEBUG_MEMORY_MAGIC) {
+         debug_printf("%s:%u:%s: bad or corrupted memory %p\n",
+                      hdr->file, hdr->line, hdr->function,
+                      ptr);
+         debug_assert(0);
+      }
+
       if((start_no <= hdr->no && hdr->no < last_no) ||
 	 (last_no < start_no && (hdr->no < last_no || start_no <= hdr->no))) {
 	 debug_printf("%s:%u:%s: %u bytes at %p not freed\n",
@@ -270,7 +289,15 @@ debug_memory_end(unsigned long start_no)
 		      hdr->size, ptr);
 	 total_size += hdr->size;
       }
+
+      if(ftr->magic != DEBUG_MEMORY_MAGIC) {
+         debug_printf("%s:%u:%s: buffer overflow %p\n",
+                      hdr->file, hdr->line, hdr->function,
+                      ptr);
+         debug_assert(0);
+      }
    }
+
    if(total_size) {
       debug_printf("Total of %u KB of system memory apparently leaked\n",
 		   (total_size + 1023)/1024);
diff --git a/src/gallium/auxiliary/util/u_math.h b/src/gallium/auxiliary/util/u_math.h
index 0b10622ee7..196aeb28fa 100644
--- a/src/gallium/auxiliary/util/u_math.h
+++ b/src/gallium/auxiliary/util/u_math.h
@@ -40,6 +40,7 @@
 
 
 #include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
 
 
 #ifdef __cplusplus
diff --git a/src/gallium/drivers/trace/README b/src/gallium/drivers/trace/README
index e7a2f12b02..f0e1cd596d 100644
--- a/src/gallium/drivers/trace/README
+++ b/src/gallium/drivers/trace/README
@@ -10,7 +10,7 @@ This directory contains a Gallium3D pipe driver which traces all incoming calls.
 
 To build, invoke scons on the top dir as
  
- scons statetrackers=mesa drivers=softpipe,i915simple,trace winsys=xlib
+ scons statetrackers=mesa drivers=softpipe,i965simple,trace winsys=xlib
 
 
 = Usage =
diff --git a/src/mesa/main/mtypes.h b/src/mesa/main/mtypes.h
index 4cb895e486..8e4f6a2e66 100644
--- a/src/mesa/main/mtypes.h
+++ b/src/mesa/main/mtypes.h
@@ -1933,7 +1933,10 @@ struct gl_fragment_program
 {
    struct gl_program Base;   /**< base class */
    GLenum FogOption;
-   GLboolean UsesKill;
+   GLboolean UsesKill;          /**< shader uses KIL instruction */
+   GLboolean UsesPointCoord;    /**< shader uses gl_PointCoord */
+   GLboolean UsesFrontFacing;   /**< shader used gl_FrontFacing */
+   GLboolean UsesFogFragCoord;  /**< shader used gl_FogFragCoord */
 };
 
 
diff --git a/src/mesa/shader/shader_api.c b/src/mesa/shader/shader_api.c
index 0d65af9738..decdec53ed 100644
--- a/src/mesa/shader/shader_api.c
+++ b/src/mesa/shader/shader_api.c
@@ -1112,7 +1112,8 @@ get_matrix_dims(GLenum type, GLint *rows, GLint *cols)
 
 /**
  * Determine the number of rows and columns occupied by a uniform
- * according to its datatype.
+ * according to its datatype.  For non-matrix types (such as GL_FLOAT_VEC4),
+ * the number of rows = 1 and cols = number of elements in the vector.
  */
 static void
 get_uniform_rows_cols(const struct gl_program_parameter *p,
@@ -1121,11 +1122,17 @@ get_uniform_rows_cols(const struct gl_program_parameter *p,
    get_matrix_dims(p->DataType, rows, cols);
    if (*rows == 0 && *cols == 0) {
       /* not a matrix type, probably a float or vector */
-      *rows = p->Size / 4 + 1;
-      if (p->Size % 4 == 0)
-         *cols = 4;
-      else
-         *cols = p->Size % 4;
+      if (p->Size <= 4) {
+         *rows = 1;
+         *cols = p->Size;
+      }
+      else {
+         *rows = p->Size / 4 + 1;
+         if (p->Size % 4 == 0)
+            *cols = 4;
+         else
+            *cols = p->Size % 4;
+      }
    }
 }
 
diff --git a/src/mesa/shader/slang/slang_link.c b/src/mesa/shader/slang/slang_link.c
index a6390846b2..d884be2a75 100644
--- a/src/mesa/shader/slang/slang_link.c
+++ b/src/mesa/shader/slang/slang_link.c
@@ -42,6 +42,24 @@
 #include "slang_link.h"
 
 
+/** cast wrapper */
+static struct gl_vertex_program *
+vertex_program(struct gl_program *prog)
+{
+   assert(prog->Target == GL_VERTEX_PROGRAM_ARB);
+   return (struct gl_vertex_program *) prog;
+}
+
+
+/** cast wrapper */
+static struct gl_fragment_program *
+fragment_program(struct gl_program *prog)
+{
+   assert(prog->Target == GL_FRAGMENT_PROGRAM_ARB);
+   return (struct gl_fragment_program *) prog;
+}
+
+
 /**
  * Record a linking error.
  */
@@ -363,6 +381,7 @@ static void
 _slang_update_inputs_outputs(struct gl_program *prog)
 {
    GLuint i, j;
+   GLuint maxAddrReg = 0;
 
    prog->InputsRead = 0x0;
    prog->OutputsWritten = 0x0;
@@ -373,30 +392,33 @@ _slang_update_inputs_outputs(struct gl_program *prog)
       for (j = 0; j < numSrc; j++) {
          if (inst->SrcReg[j].File == PROGRAM_INPUT) {
             prog->InputsRead |= 1 << inst->SrcReg[j].Index;
+            if (prog->Target == GL_FRAGMENT_PROGRAM_ARB &&
+                inst->SrcReg[j].Index == FRAG_ATTRIB_FOGC) {
+               /* The fragment shader FOGC input is used for fog,
+                * front-facing and sprite/point coord.
+                */
+               struct gl_fragment_program *fp = fragment_program(prog);
+               const GLint swz = GET_SWZ(inst->SrcReg[j].Swizzle, 0);
+               if (swz == SWIZZLE_X)
+                  fp->UsesFogFragCoord = GL_TRUE;
+               else if (swz == SWIZZLE_Y)
+                  fp->UsesFrontFacing = GL_TRUE;
+               else if (swz == SWIZZLE_Z || swz == SWIZZLE_W)
+                  fp->UsesPointCoord = GL_TRUE;
+            }
+         }
+         else if (inst->SrcReg[j].File == PROGRAM_ADDRESS) {
+            maxAddrReg = MAX2(maxAddrReg, inst->SrcReg[j].Index + 1);
          }
       }
       if (inst->DstReg.File == PROGRAM_OUTPUT) {
          prog->OutputsWritten |= 1 << inst->DstReg.Index;
       }
+      else if (inst->DstReg.File == PROGRAM_ADDRESS) {
+         maxAddrReg = MAX2(maxAddrReg, inst->DstReg.Index + 1);
+      }
    }
-}
-
-
-/** cast wrapper */
-static struct gl_vertex_program *
-vertex_program(struct gl_program *prog)
-{
-   assert(prog->Target == GL_VERTEX_PROGRAM_ARB);
-   return (struct gl_vertex_program *) prog;
-}
-
-
-/** cast wrapper */
-static struct gl_fragment_program *
-fragment_program(struct gl_program *prog)
-{
-   assert(prog->Target == GL_FRAGMENT_PROGRAM_ARB);
-   return (struct gl_fragment_program *) prog;
+   prog->NumAddressRegs = maxAddrReg;
 }
 
 
diff --git a/src/mesa/state_tracker/st_atom_rasterizer.c b/src/mesa/state_tracker/st_atom_rasterizer.c
index e286dc5116..fc47896c24 100644
--- a/src/mesa/state_tracker/st_atom_rasterizer.c
+++ b/src/mesa/state_tracker/st_atom_rasterizer.c
@@ -254,7 +254,7 @@ static void update_raster_state( struct st_context *st )
    raster->line_stipple_factor = ctx->Line.StippleFactor - 1;
 
    /* _NEW_MULTISAMPLE */
-   if (ctx->Multisample._Enabled)
+   if (ctx->Multisample._Enabled || st->force_msaa)
       raster->multisample = 1;
 
    /* _NEW_SCISSOR */
diff --git a/src/mesa/state_tracker/st_cb_fbo.c b/src/mesa/state_tracker/st_cb_fbo.c
index 8406bf247f..00076f61e0 100644
--- a/src/mesa/state_tracker/st_cb_fbo.c
+++ b/src/mesa/state_tracker/st_cb_fbo.c
@@ -170,6 +170,7 @@ st_renderbuffer_alloc_storage(GLcontext * ctx, struct gl_renderbuffer *rb,
                                                   0, 0, 0,
                                                   surface_usage );
 
+   assert(strb->surface->texture);
    assert(strb->surface->buffer);
    assert(strb->surface->format);
    assert(strb->surface->block.size);
diff --git a/src/mesa/state_tracker/st_cb_texture.c b/src/mesa/state_tracker/st_cb_texture.c
index a3e8fc992d..2e1ad93942 100644
--- a/src/mesa/state_tracker/st_cb_texture.c
+++ b/src/mesa/state_tracker/st_cb_texture.c
@@ -455,6 +455,11 @@ st_TexImage(GLcontext * ctx,
       _mesa_align_free(texImage->Data);
    }
 
+   if (width == 0 || height == 0 || depth == 0) {
+      /* stop after freeing old image */
+      return;
+   }
+
    /* If this is the only mipmap level in the texture, could call
     * bmBufferData with NULL data to free the old block and avoid
     * waiting on any outstanding fences.
@@ -1048,7 +1053,8 @@ st_copy_texsubimage(GLcontext *ctx,
    GLboolean use_fallback = GL_TRUE;
    GLboolean matching_base_formats;
 
-   st_flush(ctx->st, PIPE_FLUSH_RENDER_CACHE, NULL);
+   /* any rendering in progress must complete before we grab the fb image */
+   st_finish(ctx->st);
 
    /* determine if copying depth or color data */
    if (texBaseFormat == GL_DEPTH_COMPONENT) {
diff --git a/src/mesa/state_tracker/st_context.c b/src/mesa/state_tracker/st_context.c
index 08d4db7f7f..cca808d328 100644
--- a/src/mesa/state_tracker/st_context.c
+++ b/src/mesa/state_tracker/st_context.c
@@ -88,6 +88,19 @@ void st_invalidate_state(GLcontext * ctx, GLuint new_state)
 }
 
 
+/**
+ * Check for multisample env var override.
+ */
+int
+st_get_msaa(void)
+{
+   const char *msaa = _mesa_getenv("__GL_FSAA_MODE");
+   if (msaa)
+      return atoi(msaa);
+   return 0;
+}
+
+
 static struct st_context *
 st_create_context_priv( GLcontext *ctx, struct pipe_context *pipe )
 {
@@ -141,6 +154,8 @@ st_create_context_priv( GLcontext *ctx, struct pipe_context *pipe )
 
    st->pixel_xfer.cache = _mesa_new_program_cache();
 
+   st->force_msaa = st_get_msaa();
+
    /* GL limits and extensions */
    st_init_limits(st);
    st_init_extensions(st);
@@ -188,8 +203,6 @@ static void st_destroy_context_priv( struct st_context *st )
    st_destroy_drawtex(st);
 #endif
 
-   _vbo_DestroyContext(st->ctx);
-
    for (i = 0; i < Elements(st->state.sampler_texture); i++) {
       pipe_texture_reference(&st->state.sampler_texture[i], NULL);
    }
@@ -223,6 +236,8 @@ void st_destroy_context( struct st_context *st )
 
    _mesa_delete_program_cache(st->ctx, st->pixel_xfer.cache);
 
+   _vbo_DestroyContext(st->ctx);
+
    _mesa_free_context_data(ctx);
 
    st_destroy_context_priv(st);
diff --git a/src/mesa/state_tracker/st_context.h b/src/mesa/state_tracker/st_context.h
index 4314d9af5c..1d1aca3111 100644
--- a/src/mesa/state_tracker/st_context.h
+++ b/src/mesa/state_tracker/st_context.h
@@ -181,6 +181,8 @@ struct st_context
    struct blit_state *blit;
 
    struct cso_context *cso_context;
+
+   int force_msaa;
 };
 
 
@@ -238,4 +240,8 @@ st_fb_orientation(const struct gl_framebuffer *fb)
 }
 
 
+extern int
+st_get_msaa(void);
+
+
 #endif
diff --git a/src/mesa/state_tracker/st_framebuffer.c b/src/mesa/state_tracker/st_framebuffer.c
index 0f4a03fa48..ec8928f200 100644
--- a/src/mesa/state_tracker/st_framebuffer.c
+++ b/src/mesa/state_tracker/st_framebuffer.c
@@ -51,13 +51,12 @@ st_create_framebuffer( const __GLcontextModes *visual,
 {
    struct st_framebuffer *stfb = CALLOC_STRUCT(st_framebuffer);
    if (stfb) {
-      int samples = 0;
-      const char *msaa_override = _mesa_getenv("__GL_FSAA_MODE");
+      int samples = st_get_msaa();
+
+      if (visual->sampleBuffers)
+         samples = visual->samples;
+
       _mesa_initialize_framebuffer(&stfb->Base, visual);
-      if (visual->sampleBuffers) samples = visual->samples;
-      if (msaa_override) {
-         samples = _mesa_atoi(msaa_override);
-      }
 
       {
          /* fake frontbuffer */
diff --git a/src/mesa/state_tracker/st_mesa_to_tgsi.c b/src/mesa/state_tracker/st_mesa_to_tgsi.c
index 5ec9fddd7f..b9807bb807 100644
--- a/src/mesa/state_tracker/st_mesa_to_tgsi.c
+++ b/src/mesa/state_tracker/st_mesa_to_tgsi.c
@@ -35,10 +35,13 @@
 #include "tgsi/tgsi_parse.h"
 #include "tgsi/tgsi_build.h"
 #include "tgsi/tgsi_util.h"
+#include "tgsi/tgsi_dump.h"
+#include "tgsi/tgsi_sanity.h"
 #include "st_mesa_to_tgsi.h"
 #include "shader/prog_instruction.h"
 #include "shader/prog_parameter.h"
-
+#include "shader/prog_print.h"
+#include "pipe/p_debug.h"
 
 /*
  * Map mesa register file to TGSI register file.
@@ -239,6 +242,15 @@ compile_instruction(
          immediateMapping,
          indirectAccess );
 
+      /**
+       * This not at all the correct solution.
+       * FIXME: Roll this up in the above map functions
+       */
+      if (fullsrc->SrcRegister.File == TGSI_FILE_IMMEDIATE && fullsrc->SrcRegister.Index == ~0) {
+         fullsrc->SrcRegister.File = TGSI_FILE_CONSTANT;
+         fullsrc->SrcRegister.Index = inst->SrcReg[i].Index;
+      }
+
       /* swizzle (ext swizzle also depends on negation) */
       {
          GLuint swz[4];
@@ -980,5 +992,17 @@ tgsi_translate_mesa_program(
          maxTokens - ti );
    }
 
+#if DEBUG
+   if(!tgsi_sanity_check(tokens)) {
+      debug_printf("Due to sanity check failure(s) above the following shader program is invalid:\n");
+      debug_printf("\nOriginal program:\n%s", program->String);
+      debug_printf("\nMesa program:\n");
+      _mesa_print_program(program);
+      debug_printf("\nTGSI program:\n");
+      tgsi_dump(tokens, 0);
+      assert(0);
+   }
+#endif
+
    return ti;
 }
diff --git a/src/mesa/state_tracker/st_program.c b/src/mesa/state_tracker/st_program.c
index 936a6e32ea..b2abf0286e 100644
--- a/src/mesa/state_tracker/st_program.c
+++ b/src/mesa/state_tracker/st_program.c
@@ -409,7 +409,10 @@ st_translate_fragment_program(struct st_context *st,
             interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
             break;
          case FRAG_ATTRIB_FOGC:
-            stfp->input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
+            if (stfp->Base.UsesPointCoord)
+               stfp->input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
+            else
+               stfp->input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
             stfp->input_semantic_index[slot] = 0;
             interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
             break;
diff --git a/src/mesa/vbo/vbo_context.c b/src/mesa/vbo/vbo_context.c
index 691bf5b450..bbf745b0c6 100644
--- a/src/mesa/vbo/vbo_context.c
+++ b/src/mesa/vbo/vbo_context.c
@@ -244,12 +244,14 @@ void _vbo_DestroyContext( GLcontext *ctx )
       ctx->aelt_context = NULL;
    }
 
-   vbo_exec_destroy(ctx);
+   if (vbo_context(ctx)) {
+      vbo_exec_destroy(ctx);
 #if FEATURE_dlist
-   vbo_save_destroy(ctx);
+      vbo_save_destroy(ctx);
 #endif
-   FREE(vbo_context(ctx));
-   ctx->swtnl_im = NULL;
+      FREE(vbo_context(ctx));
+      ctx->swtnl_im = NULL;
+   }
 }