set outputs_safe to 0 as it's possible for the code generation

to slip over the allocated memory for the vb.

pull in sse fixes from gallium-0.1

1 files changed, 37 insertions, 26 deletions

diff --git a/src/mesa/x86/rtasm/x86sse.h b/src/mesa/x86/rtasm/x86sse.h
index 42b09937bca..c2aa416492e 100644
--- a/src/mesa/x86/rtasm/x86sse.h
+++ b/src/mesa/x86/rtasm/x86sse.h
@@ -2,26 +2,25 @@
 #ifndef _X86SSE_H_
 #define _X86SSE_H_
 
-#if defined(USE_X86_ASM) || defined(SLANG_X86)
-
-#include "glheader.h"
+#if defined(__i386__) || defined(__386__)
 
 /* It is up to the caller to ensure that instructions issued are
  * suitable for the host cpu.  There are no checks made in this module
  * for mmx/sse/sse2 support on the cpu.
  */
 struct x86_reg {
-   GLuint file:3;
-   GLuint idx:3;
-   GLuint mod:2;		/* mod_REG if this is just a register */
-   GLint  disp:24;		/* only +/- 23bits of offset - should be enough... */
+   unsigned file:3;
+   unsigned idx:3;
+   unsigned mod:2;		/* mod_REG if this is just a register */
+   int      disp:24;		/* only +/- 23bits of offset - should be enough... */
 };
 
 struct x86_function {
-   GLubyte *store;
-   GLubyte *csr;
-   GLuint stack_offset;
-   GLint need_emms;
+   unsigned size;
+   unsigned char *store;
+   unsigned char *csr;
+   unsigned stack_offset;
+   int need_emms;
    const char *fn;
 };
 
@@ -80,7 +79,8 @@ enum sse_cc {
  */
 
 
-GLboolean x86_init_func( struct x86_function *p, GLuint code_size );
+void x86_init_func( struct x86_function *p );
+void x86_init_func_size( struct x86_function *p, unsigned code_size );
 void x86_release_func( struct x86_function *p );
 void (*x86_get_func( struct x86_function *p ))( void );
 
@@ -92,7 +92,7 @@ struct x86_reg x86_make_reg( enum x86_reg_file file,
 			     enum x86_reg_name idx );
 
 struct x86_reg x86_make_disp( struct x86_reg reg,
-			      GLint disp );
+			      int disp );
 
 struct x86_reg x86_deref( struct x86_reg reg );
 
@@ -101,31 +101,32 @@ struct x86_reg x86_get_base_reg( struct x86_reg reg );
 
 /* Labels, jumps and fixup:
  */
-GLubyte *x86_get_label( struct x86_function *p );
+unsigned char *x86_get_label( struct x86_function *p );
 
 void x86_jcc( struct x86_function *p,
 	      enum x86_cc cc,
-	      GLubyte *label );
+	      unsigned char *label );
 
-GLubyte *x86_jcc_forward( struct x86_function *p,
+unsigned char *x86_jcc_forward( struct x86_function *p,
 			  enum x86_cc cc );
 
-GLubyte *x86_jmp_forward( struct x86_function *p);
+unsigned char *x86_jmp_forward( struct x86_function *p);
 
-GLubyte *x86_call_forward( struct x86_function *p);
+unsigned char *x86_call_forward( struct x86_function *p);
 
 void x86_fixup_fwd_jump( struct x86_function *p,
-			 GLubyte *fixup );
+			 unsigned char *fixup );
 
-void x86_jmp( struct x86_function *p, GLubyte *label );
+void x86_jmp( struct x86_function *p, unsigned char *label );
 
-void x86_call( struct x86_function *p, GLubyte *label );
+/* void x86_call( struct x86_function *p, void (*label)() ); */
+void x86_call( struct x86_function *p, struct x86_reg reg);
 
 /* michal:
  * Temporary. As I need immediate operands, and dont want to mess with the codegen,
  * I load the immediate into general purpose register and use it.
  */
-void x86_mov_reg_imm( struct x86_function *p, struct x86_reg dst, GLint imm );
+void x86_mov_reg_imm( struct x86_function *p, struct x86_reg dst, int imm );
 
 
 /* Macro for sse_shufps() and sse2_pshufd():
@@ -141,19 +142,24 @@ void mmx_packssdw( struct x86_function *p, struct x86_reg dst, struct x86_reg sr
 void mmx_packuswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 
 void sse2_cvtps2dq( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_cvttps2dq( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void sse2_movd( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void sse2_packssdw( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void sse2_packsswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void sse2_packuswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
-void sse2_pshufd( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0, GLubyte shuf );
+void sse2_pshufd( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0,
+                  unsigned char shuf );
+void sse2_rcpps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void sse2_rcpss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 
 void sse_addps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void sse_addss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void sse_cvtps2pi( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void sse_divss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_andnps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void sse_andps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
-void sse_cmpps( struct x86_function *p, struct x86_reg dst, struct x86_reg src, GLubyte cc );
+void sse_cmpps( struct x86_function *p, struct x86_reg dst, struct x86_reg src,
+                unsigned char cc );
 void sse_maxps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void sse_maxss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void sse_minps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
@@ -166,9 +172,14 @@ void sse_movss( struct x86_function *p, struct x86_reg dst, struct x86_reg src )
 void sse_movups( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void sse_mulps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void sse_mulss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_orps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_xorps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void sse_subps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_rsqrtps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void sse_rsqrtss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
-void sse_shufps( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0, GLubyte shuf );
+void sse_shufps( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0,
+                 unsigned char shuf );
+void sse_pmovmskb( struct x86_function *p, struct x86_reg dest, struct x86_reg src );
 
 void x86_add( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void x86_and( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
@@ -239,7 +250,7 @@ void x87_fucom( struct x86_function *p, struct x86_reg arg );
  * account any push/pop activity.  Note - doesn't track explict
  * manipulation of ESP by other instructions.
  */
-struct x86_reg x86_fn_arg( struct x86_function *p, GLuint arg );
+struct x86_reg x86_fn_arg( struct x86_function *p, unsigned arg );
 
 #endif
 #endif