Merge commit 'fab9df63a3156ffe1f9490aafaea41e03ef60ddf'

* commit 'fab9df63a3156ffe1f9490aafaea41e03ef60ddf':
  dsputil: Split off global motion compensation bits into a separate context

Conflicts:
	libavcodec/dsputil.c
	libavcodec/dsputil.h
	libavcodec/ppc/dsputil_altivec.h
	libavcodec/x86/dsputil_init.c
	libavcodec/x86/dsputil_mmx.c
	libavcodec/x86/dsputil_x86.h

Merged-by: Michael Niedermayer <michaelni@gmx.at>

1 files changed, 133 insertions, 0 deletions

diff --git a/libavcodec/ppc/mpegvideodsp.c b/libavcodec/ppc/mpegvideodsp.c
new file mode 100644
index 0000000000..7696954335
--- /dev/null
+++ b/libavcodec/ppc/mpegvideodsp.c
@@ -0,0 +1,133 @@
+/*
+ * GMC (Global Motion Compensation), AltiVec-enabled
+ *
+ * Copyright (c) 2003 Romain Dolbeau <romain@dolbeau.org>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg 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 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg 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 FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/mem.h"
+#include "libavutil/ppc/types_altivec.h"
+#include "libavutil/ppc/util_altivec.h"
+#include "libavcodec/mpegvideodsp.h"
+
+#if HAVE_ALTIVEC
+/* AltiVec-enhanced gmc1. ATM this code assumes stride is a multiple of 8
+ * to preserve proper dst alignment. */
+static void gmc1_altivec(uint8_t *dst /* align 8 */, uint8_t *src /* align1 */,
+                         int stride, int h, int x16, int y16, int rounder)
+{
+    int i;
+    const DECLARE_ALIGNED(16, unsigned short, rounder_a) = rounder;
+    const DECLARE_ALIGNED(16, unsigned short, ABCD)[8] = {
+        (16 - x16) * (16 - y16), /* A */
+             (x16) * (16 - y16), /* B */
+        (16 - x16) * (y16),      /* C */
+             (x16) * (y16),      /* D */
+        0, 0, 0, 0               /* padding */
+    };
+    register const vector unsigned char vczero =
+        (const vector unsigned char) vec_splat_u8(0);
+    register const vector unsigned short vcsr8 =
+        (const vector unsigned short) vec_splat_u16(8);
+    register vector unsigned char dstv, dstv2, srcvB, srcvC, srcvD;
+    register vector unsigned short tempB, tempC, tempD;
+    unsigned long dst_odd        = (unsigned long) dst & 0x0000000F;
+    unsigned long src_really_odd = (unsigned long) src & 0x0000000F;
+    register vector unsigned short tempA =
+        vec_ld(0, (const unsigned short *) ABCD);
+    register vector unsigned short Av = vec_splat(tempA, 0);
+    register vector unsigned short Bv = vec_splat(tempA, 1);
+    register vector unsigned short Cv = vec_splat(tempA, 2);
+    register vector unsigned short Dv = vec_splat(tempA, 3);
+    register vector unsigned short rounderV =
+        vec_splat((vec_u16) vec_lde(0, &rounder_a), 0);
+
+    /* we'll be able to pick-up our 9 char elements at src from those
+     * 32 bytes we load the first batch here, as inside the loop we can
+     * reuse 'src + stride' from one iteration as the 'src' of the next. */
+    register vector unsigned char src_0 = vec_ld(0, src);
+    register vector unsigned char src_1 = vec_ld(16, src);
+    register vector unsigned char srcvA = vec_perm(src_0, src_1,
+                                                   vec_lvsl(0, src));
+
+    if (src_really_odd != 0x0000000F)
+        /* If (src & 0xF) == 0xF, then (src + 1) is properly aligned
+         * on the second vector. */
+        srcvB = vec_perm(src_0, src_1, vec_lvsl(1, src));
+    else
+        srcvB = src_1;
+    srcvA = vec_mergeh(vczero, srcvA);
+    srcvB = vec_mergeh(vczero, srcvB);
+
+    for (i = 0; i < h; i++) {
+        dst_odd        =   (unsigned long) dst            & 0x0000000F;
+        src_really_odd = (((unsigned long) src) + stride) & 0x0000000F;
+
+        dstv = vec_ld(0, dst);
+
+        /* We'll be able to pick-up our 9 char elements at src + stride from
+         * those 32 bytes then reuse the resulting 2 vectors srvcC and srcvD
+         * as the next srcvA and srcvB. */
+        src_0 = vec_ld(stride +  0, src);
+        src_1 = vec_ld(stride + 16, src);
+        srcvC = vec_perm(src_0, src_1, vec_lvsl(stride + 0, src));
+
+        if (src_really_odd != 0x0000000F)
+            /* If (src & 0xF) == 0xF, then (src + 1) is properly aligned
+             * on the second vector. */
+            srcvD = vec_perm(src_0, src_1, vec_lvsl(stride + 1, src));
+        else
+            srcvD = src_1;
+
+        srcvC = vec_mergeh(vczero, srcvC);
+        srcvD = vec_mergeh(vczero, srcvD);
+
+        /* OK, now we (finally) do the math :-)
+         * Those four instructions replace 32 int muls & 32 int adds.
+         * Isn't AltiVec nice? */
+        tempA = vec_mladd((vector unsigned short) srcvA, Av, rounderV);
+        tempB = vec_mladd((vector unsigned short) srcvB, Bv, tempA);
+        tempC = vec_mladd((vector unsigned short) srcvC, Cv, tempB);
+        tempD = vec_mladd((vector unsigned short) srcvD, Dv, tempC);
+
+        srcvA = srcvC;
+        srcvB = srcvD;
+
+        tempD = vec_sr(tempD, vcsr8);
+
+        dstv2 = vec_pack(tempD, (vector unsigned short) vczero);
+
+        if (dst_odd)
+            dstv2 = vec_perm(dstv, dstv2, vcprm(0, 1, s0, s1));
+        else
+            dstv2 = vec_perm(dstv, dstv2, vcprm(s0, s1, 2, 3));
+
+        vec_st(dstv2, 0, dst);
+
+        dst += stride;
+        src += stride;
+    }
+}
+#endif /* HAVE_ALTIVEC */
+
+av_cold void ff_mpegvideodsp_init_ppc(MpegVideoDSPContext *c)
+{
+#if HAVE_ALTIVEC
+    c->gmc1 = gmc1_altivec;
+#endif /* HAVE_ALTIVEC */
+}