# HG changeset patch
# User mellum
# Date 1025658584 0
# Node ID fb670ca9f8eb862e5698824f4c7de3ba2a912cd9
# Parent  ba67fefada472d7161a352c8e0dcd1b089752e68
Use updated motion compensation routines.

diff -r ba67fefada47 -r fb670ca9f8eb alpha/dsputil_alpha.c
--- a/alpha/dsputil_alpha.c	Tue Jul 02 02:22:46 2002 +0000
+++ b/alpha/dsputil_alpha.c	Wed Jul 03 01:09:44 2002 +0000
@@ -105,132 +105,137 @@
 }
 #endif
 
-/* Average 8 unsigned bytes in parallel: (b1 + b2) >> 1
-   Since the immediate result could be greater than 255, we do the
-   shift first. The result is too low by one if the bytes were both
-   odd, so we need to add (l1 & l2) & BYTE_VEC(0x01).  */
-static inline UINT64 avg2_no_rnd(UINT64 l1, UINT64 l2)
+static inline uint64_t avg2_no_rnd(uint64_t a, uint64_t b)
 {
-    UINT64 correction = (l1 & l2) & BYTE_VEC(0x01);
-    l1 = (l1 & ~BYTE_VEC(0x01)) >> 1;
-    l2 = (l2 & ~BYTE_VEC(0x01)) >> 1;
-    return l1 + l2 + correction;
+    return (a & b) + (((a ^ b) & BYTE_VEC(0xfe)) >> 1);
+}
+
+static inline uint64_t avg2(uint64_t a, uint64_t b)
+{
+    return (a | b) - (((a ^ b) & BYTE_VEC(0xfe)) >> 1);    
 }
 
-/* Average 8 bytes with rounding: (b1 + b2 + 1) >> 1
-   The '1' only has an effect when one byte is even and the other odd,
-   i. e. we also need to add (l1 ^ l2) & BYTE_VEC(0x01).
-   Incidentally, that is equivalent to (l1 | l2) & BYTE_VEC(0x01).  */
-static inline UINT64 avg2(UINT64 l1, UINT64 l2)
+static inline uint64_t avg4(uint64_t l1, uint64_t l2, uint64_t l3, uint64_t l4)
 {
-    UINT64 correction = (l1 | l2) & BYTE_VEC(0x01);
-    l1 = (l1 & ~BYTE_VEC(0x01)) >> 1;
-    l2 = (l2 & ~BYTE_VEC(0x01)) >> 1;
-    return l1 + l2 + correction;
+    uint64_t r1 = ((l1 & ~BYTE_VEC(0x03)) >> 2)
+		+ ((l2 & ~BYTE_VEC(0x03)) >> 2)
+		+ ((l3 & ~BYTE_VEC(0x03)) >> 2)
+		+ ((l4 & ~BYTE_VEC(0x03)) >> 2);
+    uint64_t r2 = ((  (l1 & BYTE_VEC(0x03))
+		    + (l2 & BYTE_VEC(0x03))
+		    + (l3 & BYTE_VEC(0x03))
+		    + (l4 & BYTE_VEC(0x03))
+		    + BYTE_VEC(0x02)) >> 2) & BYTE_VEC(0x03);
+    return r1 + r2;
 }
 
-static inline UINT64 avg4(UINT64 l1, UINT64 l2, UINT64 l3, UINT64 l4)
+static inline uint64_t avg4_no_rnd(uint64_t l1, uint64_t l2,
+				   uint64_t l3, uint64_t l4)
 {
-    UINT64 r1 = ((l1 & ~BYTE_VEC(0x03)) >> 2)
-	      + ((l2 & ~BYTE_VEC(0x03)) >> 2)
-	      + ((l3 & ~BYTE_VEC(0x03)) >> 2)
-	      + ((l4 & ~BYTE_VEC(0x03)) >> 2);
-    UINT64 r2 = ((  (l1 & BYTE_VEC(0x03))
-		  + (l2 & BYTE_VEC(0x03))
-		  + (l3 & BYTE_VEC(0x03))
-		  + (l4 & BYTE_VEC(0x03))
-		  + BYTE_VEC(0x02)) >> 2) & BYTE_VEC(0x03);
+    uint64_t r1 = ((l1 & ~BYTE_VEC(0x03)) >> 2)
+		+ ((l2 & ~BYTE_VEC(0x03)) >> 2)
+		+ ((l3 & ~BYTE_VEC(0x03)) >> 2)
+		+ ((l4 & ~BYTE_VEC(0x03)) >> 2);
+    uint64_t r2 = ((  (l1 & BYTE_VEC(0x03))
+		    + (l2 & BYTE_VEC(0x03))
+		    + (l3 & BYTE_VEC(0x03))
+		    + (l4 & BYTE_VEC(0x03))
+		    + BYTE_VEC(0x01)) >> 2) & BYTE_VEC(0x03);
     return r1 + r2;
 }
 
-static inline UINT64 avg4_no_rnd(UINT64 l1, UINT64 l2, UINT64 l3, UINT64 l4)
-{
-    UINT64 r1 = ((l1 & ~BYTE_VEC(0x03)) >> 2)
-	      + ((l2 & ~BYTE_VEC(0x03)) >> 2)
-	      + ((l3 & ~BYTE_VEC(0x03)) >> 2)
-	      + ((l4 & ~BYTE_VEC(0x03)) >> 2);
-    UINT64 r2 = (( (l1 & BYTE_VEC(0x03))
-		 + (l2 & BYTE_VEC(0x03))
-		 + (l3 & BYTE_VEC(0x03))
-		 + (l4 & BYTE_VEC(0x03))
-		 + BYTE_VEC(0x01)) >> 2) & BYTE_VEC(0x03);
-    return r1 + r2;
+#define OP(LOAD, STORE, INCR)			\
+    do {					\
+	STORE(LOAD(pixels), block);		\
+	pixels += line_size;			\
+	block += INCR;				\
+    } while (--h)
+
+#define OP_X2(LOAD, STORE, INCR)				\
+    do {							\
+	uint64_t pix1, pix2;					\
+								\
+	pix1 = LOAD(pixels);					\
+	pix2 = pix1 >> 8 | ((uint64_t) pixels[8] << 56);	\
+	STORE(AVG2(pix1, pix2), block);				\
+	pixels += line_size;					\
+	block += INCR;						\
+    } while (--h)
+
+#define OP_Y2(LOAD, STORE, INCR)		\
+    do {					\
+	uint64_t pix = LOAD(pixels);		\
+	do {					\
+	    uint64_t next_pix;			\
+						\
+	    pixels += line_size;		\
+	    next_pix = LOAD(pixels);		\
+	    STORE(AVG2(pix, next_pix), block);	\
+	    block += INCR;			\
+	    pix = next_pix;			\
+	} while (--h);				\
+    } while (0)
+
+#define OP_XY2(LOAD, STORE, INCR)					\
+    do {								\
+	uint64_t pix1 = LOAD(pixels);					\
+	uint64_t pix2 = pix1 >> 8 | ((uint64_t) pixels[8] << 56);	\
+									\
+	do {								\
+	    uint64_t next_pix1, next_pix2;				\
+									\
+	    pixels += line_size;					\
+	    next_pix1 = LOAD(pixels);					\
+	    next_pix2 = next_pix1 >> 8 | ((uint64_t) pixels[8] << 56);	\
+									\
+	    STORE(AVG4(pix1, pix2, next_pix1, next_pix2), block);	\
+									\
+	    block += INCR;						\
+	    pix1 = next_pix1;						\
+	    pix2 = next_pix2;						\
+	} while (--h);							\
+    } while (0)
+
+#define MAKE_OP(BTYPE, OPNAME, SUFF, OPKIND, STORE, INCR)		\
+static void OPNAME ## _pixels ## SUFF ## _axp(BTYPE *block,		\
+					      const uint8_t *pixels,	\
+					      int line_size, int h)	\
+{									\
+    if ((size_t) pixels & 0x7) {					\
+	OPKIND(uldq, STORE, INCR);					\
+    } else {								\
+	OPKIND(ldq, STORE, INCR);					\
+    }									\
 }
 
-#define PIXOPNAME(suffix) put ## suffix
-#define BTYPE UINT8
+#define PIXOP(BTYPE, OPNAME, STORE, INCR)		\
+    MAKE_OP(BTYPE, OPNAME, ,	 OP,	 STORE, INCR);	\
+    MAKE_OP(BTYPE, OPNAME, _x2,	 OP_X2,	 STORE, INCR);	\
+    MAKE_OP(BTYPE, OPNAME, _y2,	 OP_Y2,	 STORE, INCR);	\
+    MAKE_OP(BTYPE, OPNAME, _xy2, OP_XY2, STORE, INCR);
+
+/* Rounding primitives.  */
 #define AVG2 avg2
 #define AVG4 avg4
 #define STORE(l, b) stq(l, b)
-#include "pixops.h"
-#undef PIXOPNAME
-#undef BTYPE
+PIXOP(uint8_t, put, STORE, line_size);
+
+#undef STORE
+#define STORE(l, b) stq(AVG2(l, ldq(b)), b);
+PIXOP(uint8_t, avg, STORE, line_size);
+
+/* Not rounding primitives.  */
 #undef AVG2
 #undef AVG4
 #undef STORE
-
-#define PIXOPNAME(suffix) put_no_rnd ## suffix
-#define BTYPE UINT8
 #define AVG2 avg2_no_rnd
 #define AVG4 avg4_no_rnd
 #define STORE(l, b) stq(l, b)
-#include "pixops.h"
-#undef PIXOPNAME
-#undef BTYPE
-#undef AVG2
-#undef AVG4
-#undef STORE
-
-/* The following functions are untested.  */
-#if 0
+PIXOP(uint8_t, put_no_rnd, STORE, line_size);
 
-#define PIXOPNAME(suffix) avg ## suffix
-#define BTYPE UINT8
-#define AVG2 avg2
-#define AVG4 avg4
-#define STORE(l, b) stq(AVG2(l, ldq(b)), b);
-#include "pixops.h"
-#undef PIXOPNAME
-#undef BTYPE
-#undef AVG2
-#undef AVG4
 #undef STORE
-
-#define PIXOPNAME(suffix) avg_no_rnd ## suffix
-#define BTYPE UINT8
-#define AVG2 avg2_no_rnd
-#define AVG4 avg4_no_rnd
 #define STORE(l, b) stq(AVG2(l, ldq(b)), b);
-#include "pixops.h"
-#undef PIXOPNAME
-#undef BTYPE
-#undef AVG2
-#undef AVG4
-#undef STORE
-
-#define PIXOPNAME(suffix) sub ## suffix
-#define BTYPE DCTELEM
-#define AVG2 avg2
-#define AVG4 avg4
-#define STORE(l, block) do {		\
-    UINT64 xxx = l;			\
-    (block)[0] -= (xxx >>  0) & 0xff;	\
-    (block)[1] -= (xxx >>  8) & 0xff;	\
-    (block)[2] -= (xxx >> 16) & 0xff;	\
-    (block)[3] -= (xxx >> 24) & 0xff;	\
-    (block)[4] -= (xxx >> 32) & 0xff;	\
-    (block)[5] -= (xxx >> 40) & 0xff;	\
-    (block)[6] -= (xxx >> 48) & 0xff;	\
-    (block)[7] -= (xxx >> 56) & 0xff;	\
-} while (0)
-#include "pixops.h"
-#undef PIXOPNAME
-#undef BTYPE
-#undef AVG2
-#undef AVG4
-#undef STORE
-
-#endif
+PIXOP(uint8_t, avg_no_rnd, STORE, line_size);
 
 void dsputil_init_alpha(void)
 {
@@ -244,6 +249,16 @@
     put_no_rnd_pixels_tab[2] = put_no_rnd_pixels_y2_axp;
     put_no_rnd_pixels_tab[3] = put_no_rnd_pixels_xy2_axp;
 
+    avg_pixels_tab[0] = avg_pixels_axp;
+    avg_pixels_tab[1] = avg_pixels_x2_axp;
+    avg_pixels_tab[2] = avg_pixels_y2_axp;
+    avg_pixels_tab[3] = avg_pixels_xy2_axp;
+
+    avg_no_rnd_pixels_tab[0] = avg_no_rnd_pixels_axp;
+    avg_no_rnd_pixels_tab[1] = avg_no_rnd_pixels_x2_axp;
+    avg_no_rnd_pixels_tab[2] = avg_no_rnd_pixels_y2_axp;
+    avg_no_rnd_pixels_tab[3] = avg_no_rnd_pixels_xy2_axp;
+
     /* amask clears all bits that correspond to present features.  */
     if (amask(AMASK_MVI) == 0) {
         put_pixels_clamped = put_pixels_clamped_mvi_asm;
diff -r ba67fefada47 -r fb670ca9f8eb alpha/pixops.h
--- a/alpha/pixops.h	Tue Jul 02 02:22:46 2002 +0000
+++ /dev/null	Thu Jan 01 00:00:00 1970 +0000
@@ -1,135 +0,0 @@
-/*
- * Alpha optimized DSP utils
- * Copyright (c) 2002 Falk Hueffner <falk@debian.org>
- *
- * This library is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation; either
- * version 2 of the License, or (at your option) any later version.
- *
- * This library is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * Lesser General Public License for more details.
- *
- * You should have received a copy of the GNU Lesser General Public
- * License along with this library; if not, write to the Free Software
- * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
- */
-
-/* This file is intended to be #included with proper definitions of
- * PIXOPNAME, BTYPE, AVG2, AVG4 and STORE.  */
-
-static void PIXOPNAME(_pixels_axp)(BTYPE *block, const UINT8 *pixels,
-				   int line_size, int h)
-{
-    if ((size_t) pixels & 0x7) {
-	do {
-	    STORE(uldq(pixels), block);
-	    pixels += line_size;
-	    block  += line_size;
-	} while (--h);
-    } else {
-	do {
-	    STORE(ldq(pixels), block);
-	    pixels += line_size;
-	    block  += line_size;
-	} while (--h);
-    }
-}
-
-static void PIXOPNAME(_pixels_x2_axp)(BTYPE *block, const UINT8 *pixels,
-				      int line_size, int h)
-{
-    if ((size_t) pixels & 0x7) {
-	do {
-	    UINT64 pix1, pix2;
-
-	    pix1 = uldq(pixels);
-	    pix2 = pix1 >> 8 | ((UINT64) pixels[8] << 56);
-	    STORE(AVG2(pix1, pix2), block);
-	    pixels += line_size;
-	    block += line_size;
-	} while (--h);
-    } else {
-	do {
-	    UINT64 pix1, pix2;
-
-	    pix1 = ldq(pixels);
-	    pix2 = pix1 >> 8 | ((UINT64) pixels[8] << 56);
-	    STORE(AVG2(pix1, pix2), block);
-	    pixels += line_size;
-	    block += line_size;
-	} while (--h);
-    }
-}
-
-static void PIXOPNAME(_pixels_y2_axp)(BTYPE *block, const UINT8 *pixels,
-				      int line_size, int h)
-{
-    if ((size_t) pixels & 0x7) {
-	UINT64 pix = uldq(pixels);
-	do {
-	    UINT64 next_pix;
-
-	    pixels += line_size;
-	    next_pix = uldq(pixels);
-	    STORE(AVG2(pix, next_pix), block);
-	    block += line_size;
-	    pix = next_pix;
-	} while (--h);
-    } else {
-	UINT64 pix = ldq(pixels);
-	do {
-	    UINT64 next_pix;
-
-	    pixels += line_size;
-	    next_pix = ldq(pixels);
-	    STORE(AVG2(pix, next_pix), block);
-	    block += line_size;
-	    pix = next_pix;
-	} while (--h);
-    }
-}
-
-/* This could be further sped up by recycling AVG4 intermediate
-  results from the previous loop pass.  */
-static void PIXOPNAME(_pixels_xy2_axp)(BTYPE *block, const UINT8 *pixels,
-				       int line_size, int h)
-{
-    if ((size_t) pixels & 0x7) {
-	UINT64 pix1 = uldq(pixels);
-	UINT64 pix2 = pix1 >> 8 | ((UINT64) pixels[8] << 56);
-
-	do {
-	    UINT64 next_pix1, next_pix2;
-
-	    pixels += line_size;
-	    next_pix1 = uldq(pixels);
-	    next_pix2 = next_pix1 >> 8 | ((UINT64) pixels[8] << 56);
-
-	    STORE(AVG4(pix1, pix2, next_pix1, next_pix2), block);
-
-	    block += line_size;
-	    pix1 = next_pix1;
-	    pix2 = next_pix2;
-	} while (--h);
-    } else {
-	UINT64 pix1 = ldq(pixels);
-	UINT64 pix2 = pix1 >> 8 | ((UINT64) pixels[8] << 56);
-
-	do {
-	    UINT64 next_pix1, next_pix2;
-
-	    pixels += line_size;
-	    next_pix1 = ldq(pixels);
-	    next_pix2 = next_pix1 >> 8 | ((UINT64) pixels[8] << 56);
-
-	    STORE(AVG4(pix1, pix2, next_pix1, next_pix2), block);
-
-	    block += line_size;
-	    pix1 = next_pix1;
-	    pix2 = next_pix2;
-	} while (--h);
-    }
-}