ppc: dsputil: do unaligned block accesses correctly

To load unaligned vector data in the usual way, explicit vec_ld()
should be used rather than dereferencing a pointer to a vector type.
When the VSX extension is enabled, gcc may compile vector pointer
dereferences using the VSX lxvw4x instruction instead of the lvx
instruction typically used with Altivec/VMX.  As the behaviour of
these instructions with unaligned addresses differs, it is important
that only lvx is used here.

Signed-off-by: Mans Rullgard <mans@mansr.com>

1 files changed, 12 insertions, 12 deletions

diff --git a/libavcodec/ppc/int_altivec.c b/libavcodec/ppc/int_altivec.c
index f81b478449..b94b636dec 100644
--- a/libavcodec/ppc/int_altivec.c
+++ b/libavcodec/ppc/int_altivec.c
@@ -114,31 +114,31 @@ static int32_t scalarproduct_and_madd_int16_altivec(int16_t *v1, const int16_t *
 {
     LOAD_ZERO;
     vec_s16 *pv1 = (vec_s16*)v1;
-    vec_s16 *pv2 = (vec_s16*)v2;
-    vec_s16 *pv3 = (vec_s16*)v3;
     register vec_s16 muls = {mul,mul,mul,mul,mul,mul,mul,mul};
-    register vec_s16 t0, t1, i0, i1;
-    register vec_s16 i2 = pv2[0], i3 = pv3[0];
+    register vec_s16 t0, t1, i0, i1, i4;
+    register vec_s16 i2 = vec_ld(0, v2), i3 = vec_ld(0, v3);
     register vec_s32 res = zero_s32v;
     register vec_u8 align = vec_lvsl(0, v2);
     int32_t ires;
     order >>= 4;
     do {
-        t0 = vec_perm(i2, pv2[1], align);
-        i2 = pv2[2];
-        t1 = vec_perm(pv2[1], i2, align);
+        i1 = vec_ld(16, v2);
+        t0 = vec_perm(i2, i1, align);
+        i2 = vec_ld(32, v2);
+        t1 = vec_perm(i1, i2, align);
         i0 = pv1[0];
         i1 = pv1[1];
         res = vec_msum(t0, i0, res);
         res = vec_msum(t1, i1, res);
-        t0 = vec_perm(i3, pv3[1], align);
-        i3 = pv3[2];
-        t1 = vec_perm(pv3[1], i3, align);
+        i4 = vec_ld(16, v3);
+        t0 = vec_perm(i3, i4, align);
+        i3 = vec_ld(32, v3);
+        t1 = vec_perm(i4, i3, align);
         pv1[0] = vec_mladd(t0, muls, i0);
         pv1[1] = vec_mladd(t1, muls, i1);
         pv1 += 2;
-        pv2 += 2;
-        pv3 += 2;
+        v2  += 8;
+        v3  += 8;
     } while(--order);
     res = vec_splat(vec_sums(res, zero_s32v), 3);
     vec_ste(res, 0, &ires);