6 files changed, 37 insertions, 14 deletions
diff --git a/src/gallium/auxiliary/tgsi/tgsi_dump.h b/src/gallium/auxiliary/tgsi/tgsi_dump.h
index 4cd27317b3..dd78b36100 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_dump.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_dump.h
@@ -28,6 +28,7 @@
 #ifndef TGSI_DUMP_H
 #define TGSI_DUMP_H
 
+#include "pipe/p_compiler.h"
 #include "pipe/p_shader_tokens.h"
 
 #if defined __cplusplus
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index 298f3d0a8b..0757f05dfa 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -3239,6 +3239,8 @@ exec_instruction(
 
          if (mach->CallStackTop == 0) {
             /* returning from main() */
+            mach->CondStackTop = 0;
+            mach->LoopStackTop = 0;
             *pc = -1;
             return;
          }
@@ -3767,6 +3769,9 @@ tgsi_exec_machine_run( struct tgsi_exec_machine *mach )
    }
 #endif
 
+   /* Strictly speaking, these assertions aren't really needed but they
+    * can potentially catch some bugs in the control flow code.
+    */
    assert(mach->CondStackTop == 0);
    assert(mach->LoopStackTop == 0);
    assert(mach->ContStackTop == 0);
diff --git a/src/gallium/auxiliary/tgsi/tgsi_info.h b/src/gallium/auxiliary/tgsi/tgsi_info.h
index 50248884fd..1992d11bbe 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_info.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_info.h
@@ -28,6 +28,7 @@
 #ifndef TGSI_INFO_H
 #define TGSI_INFO_H
 
+#include "pipe/p_compiler.h"
 #include "pipe/p_shader_tokens.h"
 
 #if defined __cplusplus
diff --git a/src/gallium/auxiliary/tgsi/tgsi_parse.c b/src/gallium/auxiliary/tgsi/tgsi_parse.c
index db9a342220..1891203abe 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_parse.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_parse.c
@@ -282,17 +282,6 @@ tgsi_parse_token(
 }
 
 
-unsigned
-tgsi_num_tokens(const struct tgsi_token *tokens)
-{
-   struct tgsi_parse_context ctx;
-   if (tgsi_parse_init(&ctx, tokens) == TGSI_PARSE_OK) {
-      unsigned len = (ctx.FullHeader.Header.HeaderSize +
-                      ctx.FullHeader.Header.BodySize);
-      return len;
-   }
-   return 0;
-}
 
 
 /**
@@ -319,3 +308,19 @@ tgsi_alloc_tokens(unsigned num_tokens)
    unsigned bytes = num_tokens * sizeof(struct tgsi_token);
    return (struct tgsi_token *) MALLOC(bytes);
 }
+
+
+void
+tgsi_dump_tokens(const struct tgsi_token *tokens)
+{
+   const unsigned *dwords = (const unsigned *)tokens;
+   int nr = tgsi_num_tokens(tokens);
+   int i;
+   
+   assert(sizeof(*tokens) == sizeof(unsigned));
+
+   debug_printf("const unsigned tokens[%d] = {\n", nr);
+   for (i = 0; i < nr; i++)
+      debug_printf("0x%08x,\n", dwords[i]);
+   debug_printf("};\n");
+}
diff --git a/src/gallium/auxiliary/tgsi/tgsi_parse.h b/src/gallium/auxiliary/tgsi/tgsi_parse.h
index 36de8807b4..d4df585176 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_parse.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_parse.h
@@ -28,6 +28,7 @@
 #ifndef TGSI_PARSE_H
 #define TGSI_PARSE_H
 
+#include "pipe/p_compiler.h"
 #include "pipe/p_shader_tokens.h"
 
 #if defined __cplusplus
@@ -132,8 +133,15 @@ void
 tgsi_parse_token(
    struct tgsi_parse_context *ctx );
 
-unsigned
-tgsi_num_tokens(const struct tgsi_token *tokens);
+static INLINE unsigned
+tgsi_num_tokens(const struct tgsi_token *tokens)
+{
+   struct tgsi_header header = *(const struct tgsi_header *) tokens;
+   return header.HeaderSize + header.BodySize;
+}
+
+void
+tgsi_dump_tokens(const struct tgsi_token *tokens);
 
 struct tgsi_token *
 tgsi_dup_tokens(const struct tgsi_token *tokens);
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sse2.h b/src/gallium/auxiliary/tgsi/tgsi_sse2.h
index d81ee3d00e..00aa8b84fe 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_sse2.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_sse2.h
@@ -32,9 +32,12 @@
 extern "C" {
 #endif
 
+#include "pipe/p_compiler.h"
+
+struct tgsi_exec_machine;
+struct tgsi_interp_coef;
 struct tgsi_token;
 struct x86_function;
-struct tgsi_interp_coef;
 
 unsigned
 tgsi_emit_sse2(