[FFmpeg-cvslog] avcodec/ppc/h264qpel: POWER LE support for put_pixels16_l2_altivec() and avg_pixels16_l2_altivec()

[Rong Yan]

ffmpeg | branch: master | Rong Yan <rongyan236 at gmail.com> | Fri Nov 28 06:53:35 2014 +0000| [bd67d0ead1451c8ea9b2ef40629360c4a9d9c99c] | committer: Michael Niedermayer

avcodec/ppc/h264qpel: POWER LE support for put_pixels16_l2_altivec() and avg_pixels16_l2_altivec()

add put_unligned_store() and avg_unligned_store() macros

Signed-off-by: Michael Niedermayer <michaelni at gmx.at>

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=bd67d0ead1451c8ea9b2ef40629360c4a9d9c99c
---

 libavcodec/ppc/h264qpel.c |  109 +++++++++++++++++++++------------------------
 1 file changed, 51 insertions(+), 58 deletions(-)

diff --git a/libavcodec/ppc/h264qpel.c b/libavcodec/ppc/h264qpel.c
index 4a01f17..575f504 100644
--- a/libavcodec/ppc/h264qpel.c
+++ b/libavcodec/ppc/h264qpel.c
@@ -191,86 +191,79 @@ static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## CODETYPE(uint8_t *dst, cons
     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
 }\
 
+#if HAVE_BIGENDIAN
+#define put_unligned_store(s, dest) {    \
+    tmp1 = vec_ld(0, dest);              \
+    mask = vec_lvsl(0, dest);            \
+    tmp2 = vec_ld(15, dest);             \
+    edges = vec_perm(tmp2, tmp1, mask);  \
+    align = vec_lvsr(0, dest);           \
+    tmp2 = vec_perm(s, edges, align);    \
+    tmp1 = vec_perm(edges, s, align);    \
+    vec_st(tmp2, 15, dest);              \
+    vec_st(tmp1, 0 , dest);              \
+ }
+#else
+#define put_unligned_store(s, dest) vec_vsx_st(s, 0, dest);
+#endif /* HAVE_BIGENDIAN */
+
 static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
                                     const uint8_t * src2, int dst_stride,
                                     int src_stride1, int h)
 {
     int i;
-    vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align;
-
+    vec_u8 a, b, d, mask_;
+#if HAVE_BIGENDIAN
+    vec_u8 tmp1, tmp2, mask, edges, align;
     mask_ = vec_lvsl(0, src2);
+#endif
 
     for (i = 0; i < h; i++) {
-
-        tmp1 = vec_ld(i * src_stride1, src1);
-        mask = vec_lvsl(i * src_stride1, src1);
-        tmp2 = vec_ld(i * src_stride1 + 15, src1);
-
-        a = vec_perm(tmp1, tmp2, mask);
-
-        tmp1 = vec_ld(i * 16, src2);
-        tmp2 = vec_ld(i * 16 + 15, src2);
-
-        b = vec_perm(tmp1, tmp2, mask_);
-
-        tmp1 = vec_ld(0, dst);
-        mask = vec_lvsl(0, dst);
-        tmp2 = vec_ld(15, dst);
-
+        a = unaligned_load(i * src_stride1, src1);
+        b = load_with_perm_vec(i * 16, src2, mask_);
         d = vec_avg(a, b);
-
-        edges = vec_perm(tmp2, tmp1, mask);
-
-        align = vec_lvsr(0, dst);
-
-        tmp2 = vec_perm(d, edges, align);
-        tmp1 = vec_perm(edges, d, align);
-
-        vec_st(tmp2, 15, dst);
-        vec_st(tmp1, 0 , dst);
-
+        put_unligned_store(d, dst);
         dst += dst_stride;
     }
 }
 
+#if HAVE_BIGENDIAN
+#define avg_unligned_store(s, dest){            \
+    tmp1 = vec_ld(0, dest);                     \
+    mask = vec_lvsl(0, dest);                   \
+    tmp2 = vec_ld(15, dest);                    \
+    a = vec_avg(vec_perm(tmp1, tmp2, mask), s); \
+    edges = vec_perm(tmp2, tmp1, mask);         \
+    align = vec_lvsr(0, dest);                  \
+    tmp2 = vec_perm(a, edges, align);           \
+    tmp1 = vec_perm(edges, a, align);           \
+    vec_st(tmp2, 15, dest);                     \
+    vec_st(tmp1, 0 , dest);                     \
+ }
+#else
+#define avg_unligned_store(s, dest){            \
+    a = vec_avg(vec_vsx_ld(0, dst), s);         \
+    vec_vsx_st(a, 0, dst);                      \
+ }
+#endif /* HAVE_BIGENDIAN */
+
 static inline void avg_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
                                     const uint8_t * src2, int dst_stride,
                                     int src_stride1, int h)
 {
     int i;
-    vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align;
+    vec_u8 a, b, d, mask_;
 
+#if HAVE_BIGENDIAN
+    vec_u8 tmp1, tmp2, mask, edges, align;
     mask_ = vec_lvsl(0, src2);
+#endif
 
     for (i = 0; i < h; i++) {
-
-        tmp1 = vec_ld(i * src_stride1, src1);
-        mask = vec_lvsl(i * src_stride1, src1);
-        tmp2 = vec_ld(i * src_stride1 + 15, src1);
-
-        a = vec_perm(tmp1, tmp2, mask);
-
-        tmp1 = vec_ld(i * 16, src2);
-        tmp2 = vec_ld(i * 16 + 15, src2);
-
-        b = vec_perm(tmp1, tmp2, mask_);
-
-        tmp1 = vec_ld(0, dst);
-        mask = vec_lvsl(0, dst);
-        tmp2 = vec_ld(15, dst);
-
-        d = vec_avg(vec_perm(tmp1, tmp2, mask), vec_avg(a, b));
-
-        edges = vec_perm(tmp2, tmp1, mask);
-
-        align = vec_lvsr(0, dst);
-
-        tmp2 = vec_perm(d, edges, align);
-        tmp1 = vec_perm(edges, d, align);
-
-        vec_st(tmp2, 15, dst);
-        vec_st(tmp1, 0 , dst);
-
+        a = unaligned_load(i * src_stride1, src1);
+        b = load_with_perm_vec(i * 16, src2, mask_);
+        d = vec_avg(a, b);
+        avg_unligned_store(d, dst);
         dst += dst_stride;
     }
 }