[FFmpeg-cvslog] libavcodec/ppc/pixblockdsp.c : fix get_pixels_altivec() and diff_pixels_altivec() for POWER LE

[Rong Yan]

ffmpeg | branch: master | Rong Yan <rongyan236 at gmail.com> | Fri Oct 10 08:25:44 2014 +0000| [31dea05170e5c24328133261f689bd32bf950edd] | committer: Michael Niedermayer

libavcodec/ppc/pixblockdsp.c : fix get_pixels_altivec() and diff_pixels_altivec() for POWER LE

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

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=31dea05170e5c24328133261f689bd32bf950edd
---

 libavcodec/ppc/pixblockdsp.c |   95 ++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 95 insertions(+)

diff --git a/libavcodec/ppc/pixblockdsp.c b/libavcodec/ppc/pixblockdsp.c
index 4580c3d..9bbdf96 100644
--- a/libavcodec/ppc/pixblockdsp.c
+++ b/libavcodec/ppc/pixblockdsp.c
@@ -35,6 +35,34 @@
 
 #if HAVE_ALTIVEC
 
+#if HAVE_VSX
+static void get_pixels_altivec(int16_t *restrict block, const uint8_t *pixels,
+                               ptrdiff_t line_size)
+{
+    int i;
+    vector unsigned char perm =
+        (vector unsigned char) {0x00,0x10, 0x01,0x11,0x02,0x12,0x03,0x13,\
+            0x04,0x14,0x05,0x15,0x06,0x16,0x07,0x17};
+    const vector unsigned char zero =
+        (const vector unsigned char) vec_splat_u8(0);
+
+    for (i = 0; i < 8; i++) {
+        /* Read potentially unaligned pixels.
+         * We're reading 16 pixels, and actually only want 8,
+         * but we simply ignore the extras. */
+        vector unsigned char bytes = vec_vsx_ld(0, pixels);
+
+        // Convert the bytes into shorts.
+        //vector signed short shorts = (vector signed short) vec_perm(zero, bytes, perm);
+        vector signed short shorts = (vector signed short) vec_perm(bytes, zero, perm);
+
+        // Save the data to the block, we assume the block is 16-byte aligned.
+        vec_vsx_st(shorts, i * 16, (vector signed short *) block);
+
+        pixels += line_size;
+    }
+}
+#else
 static void get_pixels_altivec(int16_t *restrict block, const uint8_t *pixels,
                                ptrdiff_t line_size)
 {
@@ -62,6 +90,71 @@ static void get_pixels_altivec(int16_t *restrict block, const uint8_t *pixels,
     }
 }
 
+#endif /* HAVE_VSX */
+
+#if HAVE_VSX
+static void diff_pixels_altivec(int16_t *restrict block, const uint8_t *s1,
+                                const uint8_t *s2, int stride)
+{
+  int i;
+  const vector unsigned char zero =
+    (const vector unsigned char) vec_splat_u8(0);
+  vector signed short shorts1, shorts2;
+
+  for (i = 0; i < 4; i++) {
+    /* Read potentially unaligned pixels.
+     * We're reading 16 pixels, and actually only want 8,
+     * but we simply ignore the extras. */
+    vector unsigned char bytes = vec_vsx_ld(0,  s1);
+
+    // Convert the bytes into shorts.
+    shorts1 = (vector signed short) vec_mergeh(bytes, zero);
+
+    // Do the same for the second block of pixels.
+    bytes =vec_vsx_ld(0,  s2);
+
+    // Convert the bytes into shorts.
+    shorts2 = (vector signed short) vec_mergeh(bytes, zero);
+
+    // Do the subtraction.
+    shorts1 = vec_sub(shorts1, shorts2);
+
+    // Save the data to the block, we assume the block is 16-byte aligned.
+    vec_vsx_st(shorts1, 0, (vector signed short *) block);
+
+    s1    += stride;
+    s2    += stride;
+    block += 8;
+
+    /* The code below is a copy of the code above...
+     * This is a manual unroll. */
+
+    /* Read potentially unaligned pixels.
+     * We're reading 16 pixels, and actually only want 8,
+     * but we simply ignore the extras. */
+    bytes = vec_vsx_ld(0,  s1);
+
+    // Convert the bytes into shorts.
+    shorts1 = (vector signed short) vec_mergeh(bytes, zero);
+
+    // Do the same for the second block of pixels.
+    bytes = vec_vsx_ld(0,  s2);
+
+    // Convert the bytes into shorts.
+    shorts2 = (vector signed short) vec_mergeh(bytes, zero);
+
+    // Do the subtraction.
+    shorts1 = vec_sub(shorts1, shorts2);
+
+    // Save the data to the block, we assume the block is 16-byte aligned.
+    vec_vsx_st(shorts1, 0, (vector signed short *) block);
+
+    s1    += stride;
+    s2    += stride;
+    block += 8;
+  }
+}
+#else
 static void diff_pixels_altivec(int16_t *restrict block, const uint8_t *s1,
                                 const uint8_t *s2, int stride)
 {
@@ -134,6 +227,8 @@ static void diff_pixels_altivec(int16_t *restrict block, const uint8_t *s1,
     }
 }
 
+#endif /* HAVE_VSX */
+
 #endif /* HAVE_ALTIVEC */
 
 av_cold void ff_pixblockdsp_init_ppc(PixblockDSPContext *c,