--- portab.h	2002/03/16 14:15:10	1.2
+++ portab.h	2002/03/30 00:58:20	1.12
@@ -31,6 +31,17 @@
 
 #define EMMS() __asm {emms}
 
+#define CACHE_LINE  16
+
+#if _MSC_VER <= 1200
+#define DECLARE_ALIGNED_MATRIX(name,sizex,sizey,type,alignment) \
+	type name##_storage[(sizex)*(sizey)+(alignment)-1]; \
+	type * name = (type *) (((int32_t) name##_storage+(alignment - 1)) & ~((int32_t)(alignment)-1))
+#else
+#define DECLARE_ALIGNED_MATRIX(name,sizex,sizey,type,alignment) \
+	__declspec(align(alignment)) type name[(sizex)*(sizey)]
+#endif
+
 // needed for bitstream.h
 #define BSWAP(a) __asm mov eax,a __asm bswap eax __asm mov a, eax
 
@@ -52,35 +63,79 @@
 
 #elif defined(LINUX) || defined(DJGPP)
 
+
+#ifdef _DEBUG
+
 #include <stdio.h>
-#define DEBUG_WHERE		stdout
-#define DEBUG(S)        fprintf(DEBUG_WHERE, "%s\n", (S));
-#define DEBUG1(S,I)     fprintf(DEBUG_WHERE, "%s %i\n", (S), (I))
-#define DEBUG2(S,A,B)   fprintf(DEBUG_WHERE, "%s%i=%i\n", (S), (A), (B))
-#define DEBUG3(S,A,B,C) fprintf(DEBUG_WHERE, "%s %i %x %x\n", (S), (A), (B), (C))
+#define DEBUG_WHERE               stdout
+#define DEBUG(S)                  fprintf(DEBUG_WHERE, "%s\n", (S));
+#define DEBUG1(S,I)               fprintf(DEBUG_WHERE, "%s %i\n", (S), (I))
+#define DEBUG2(S,A,B)             fprintf(DEBUG_WHERE, "%s%i=%i\n", (S), (A), (B))
+#define DEBUG3(S,A,B,C)           fprintf(DEBUG_WHERE, "%s %i %x %x\n", (S), (A), (B), (C))
 #define DEBUG8(S,A,B,C,D,E,F,G,H)
+#else
+#define DEBUG(S)
+#define DEBUG1(S,I)
+#define DEBUG2(X,A,B)
+#define DEBUG3(X,A,B,C)
+#define DEBUG8(X,A,B,C,D,E,F,G,H)
+#endif
+
+#define CACHE_LINE  16
 
 #if defined(LINUX)
 
 #include <stdint.h>
 
+#define DECLARE_ALIGNED_MATRIX(name,sizex,sizey,type,alignment) \
+	type name##_storage[(sizex)*(sizey)+(alignment)-1]; \
+	type * name = (type *) (((int32_t) name##_storage+(alignment - 1)) & ~((int32_t)(alignment)-1))
+
 #else
 
-#define int8_t char
-#define uint8_t unsigned char
-#define int16_t short
+#define #define #define DECLARE_ALIGNED_MATRIX(name,sizex,sizey,type,alignment) \
+	__attribute__ ((__aligned__(CACHE_LINE))) type name[(sizex)*(sizey)]
+
+#define int8_t   char
+#define uint8_t  unsigned char
+#define int16_t  short
 #define uint16_t unsigned short
-#define int32_t int
+#define int32_t  int
 #define uint32_t unsigned int
-#define int64_t long long
+#define int64_t  long long
 #define uint64_t unsigned long long
 
 #endif
 
-#define EMMS() __asm__("emms\n\t")
 
 // needed for bitstream.h
-#define BSWAP(a) __asm__ ( "bswapl %0\n" : "=r" (a) : "0" (a) )
+#ifdef ARCH_PPC
+	#define BSWAP(a) __asm__ ( "lwbrx %0,0,%1; eieio" : "=r" (a) : \
+		"r" (&(a)), "m" (a));
+	#define EMMS()
+
+	static __inline unsigned long get_tbl(void) {
+		unsigned long tbl;
+		asm volatile("mftb %0" : "=r" (tbl));
+		return tbl;
+	}
+	static __inline unsigned long get_tbu(void) {
+		unsigned long tbl;
+		asm volatile("mftbu %0" : "=r" (tbl));
+		return tbl;
+	}
+	static __inline int64_t read_counter() {
+		unsigned long tb, tu;
+		do {
+			tb = get_tbl();
+			tu = get_tbu();
+		} while(tb != get_tbl());
+		return (((int64_t)tu) << 32) | (int64_t)tb;
+	}
+#else
+	#define BSWAP(a) __asm__ ( "bswapl %0\n" : "=r" (a) : "0" (a) )
+	#define EMMS() __asm__("emms\n\t")
+
 
 // needed for timer.c
 static __inline int64_t read_counter() {
@@ -94,6 +149,8 @@
     return ts;
 }
 
+#endif
+
 #else // OTHER OS
 
 #define DEBUG(S)
@@ -116,6 +173,9 @@
 	return 0;
 }
 
+#define CACHE_LINE  16
+#define CACHE_ALIGN
+
 #endif
 
 #endif // _PORTAB_H_