[vlc-commits] i420_yuy2: clobber MM and XMM registers correctly

[Rémi Denis-Courmont]

vlc | branch: master | Rémi Denis-Courmont <remi at remlab.net> | Sat Oct 13 16:41:14 2012 +0300| [01924381fd15196a68f29d4cc279fb7a42f6f556] | committer: Rémi Denis-Courmont

i420_yuy2: clobber MM and XMM registers correctly

> http://git.videolan.org/gitweb.cgi/vlc.git/?a=commit;h=01924381fd15196a68f29d4cc279fb7a42f6f556
---

 modules/video_chroma/i420_yuy2.c |   14 +++++++++-----
 modules/video_chroma/i420_yuy2.h |   12 ++++++++----
 2 files changed, 17 insertions(+), 9 deletions(-)

diff --git a/modules/video_chroma/i420_yuy2.c b/modules/video_chroma/i420_yuy2.c
index 59969fb..a157e94 100644
--- a/modules/video_chroma/i420_yuy2.c
+++ b/modules/video_chroma/i420_yuy2.c
@@ -45,12 +45,16 @@
 
 #if defined (MODULE_NAME_IS_i420_yuy2)
 #    define DEST_FOURCC "YUY2,YUNV,YVYU,UYVY,UYNV,Y422,IUYV,cyuv,Y211"
+#    define VLC_TARGET
 #elif defined (MODULE_NAME_IS_i420_yuy2_mmx)
 #    define DEST_FOURCC "YUY2,YUNV,YVYU,UYVY,UYNV,Y422,IUYV,cyuv"
+#    define VLC_TARGET VLC_MMX
 #elif defined (MODULE_NAME_IS_i420_yuy2_sse2)
 #    define DEST_FOURCC "YUY2,YUNV,YVYU,UYVY,UYNV,Y422,IUYV,cyuv"
+#    define VLC_TARGET VLC_SSE
 #elif defined (MODULE_NAME_IS_i420_yuy2_altivec)
 #    define DEST_FOURCC "YUY2,YUNV,YVYU,UYVY,UYNV,Y422"
+#    define VLC_TARGET
 #endif
 
 /*****************************************************************************
@@ -184,12 +188,12 @@ static inline unsigned long long read_cycles(void)
 
 /* Following functions are local */
 
-VIDEO_FILTER_WRAPPER( I420_YUY2 )
-VIDEO_FILTER_WRAPPER( I420_YVYU )
-VIDEO_FILTER_WRAPPER( I420_UYVY )
+VLC_TARGET VIDEO_FILTER_WRAPPER( I420_YUY2 )
+VLC_TARGET VIDEO_FILTER_WRAPPER( I420_YVYU )
+VLC_TARGET VIDEO_FILTER_WRAPPER( I420_UYVY )
 #if !defined (MODULE_NAME_IS_i420_yuy2_altivec)
-VIDEO_FILTER_WRAPPER( I420_IUYV )
-VIDEO_FILTER_WRAPPER( I420_cyuv )
+VLC_TARGET VIDEO_FILTER_WRAPPER( I420_IUYV )
+VLC_TARGET VIDEO_FILTER_WRAPPER( I420_cyuv )
 #endif
 #if defined (MODULE_NAME_IS_i420_yuy2)
 VIDEO_FILTER_WRAPPER( I420_Y211 )
diff --git a/modules/video_chroma/i420_yuy2.h b/modules/video_chroma/i420_yuy2.h
index e699442..573c30f 100644
--- a/modules/video_chroma/i420_yuy2.h
+++ b/modules/video_chroma/i420_yuy2.h
@@ -39,12 +39,14 @@ movq       (%3), %%mm3  # Load 8 Y            Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0     \n\
 " \
         :                                   \
         : "r" (p_u), "r" (p_v),             \
-          "r" (p_y1), "r" (p_y2) );         \
+          "r" (p_y1), "r" (p_y2)            \
+        : "mm0", "mm1", "mm2", "mm3");      \
     __asm__ __volatile__(                   \
         ".p2align 3 \n\t"                   \
         MMX_INSTRUCTIONS                    \
         :                                   \
-        : "r" (p_line1), "r" (p_line2) );   \
+        : "r" (p_line1), "r" (p_line2)      \
+        : "mm0", "mm1", "mm2", "mm3");      \
         p_line1 += 16; p_line2 += 16;       \
         p_y1 += 8; p_y2 += 8;               \
         p_u += 4; p_v += 4;                 \
@@ -199,13 +201,15 @@ movq        (%0), %%xmm1  # Load 8 Cb         u7 u6 u5 u4 u3 u2 u1 u0     \n\
 movq        (%1), %%xmm2  # Load 8 Cr         v7 06 v5 v4 v3 v2 v1 v0     \n\
 " \
         :                               \
-        : "r" (p_u),  "r" (p_v) );      \
+        : "r" (p_u),  "r" (p_v)         \
+        : "xmm1", "xmm2");              \
     __asm__ __volatile__(               \
         ".p2align 3 \n\t"               \
         SSE2_INSTRUCTIONS               \
         :                               \
         : "r" (p_line1), "r" (p_line2), \
-          "r" (p_y1),  "r" (p_y2) );    \
+          "r" (p_y1),  "r" (p_y2)       \
+        : "xmm0", "xmm1", "xmm2", "xmm3", "xmm4"); \
         p_line1 += 32; p_line2 += 32;   \
         p_y1 += 16; p_y2 += 16;         \
         p_u += 8; p_v += 8;             \