Start working on a branch where FFmpeg is not copied, patched and carved to be built with automake but instead imported inline and built using its own build system. This is an import of a slightly modified FFmpeg current tree. xine-lib builds, install and run fine with it, but there are of course plenty of things that needs to be fixed before it can even be considered for a 1.2.x series. Work will continue in the next days of course.

CVS patchset: 8397
CVS date: 2006/12/02 01:19:48

1 files changed, 147 insertions, 0 deletions

diff --git a/contrib/ffmpeg/libavcodec/alpha/mpegvideo_alpha.c b/contrib/ffmpeg/libavcodec/alpha/mpegvideo_alpha.c
new file mode 100644
index 000000000..8ad264b06
--- /dev/null
+++ b/contrib/ffmpeg/libavcodec/alpha/mpegvideo_alpha.c
@@ -0,0 +1,147 @@
+/*
+ * Alpha optimized DSP utils
+ * Copyright (c) 2002 Falk Hueffner <falk@debian.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 "asm.h"
+#include "../dsputil.h"
+#include "../mpegvideo.h"
+
+static void dct_unquantize_h263_intra_axp(MpegEncContext *s, DCTELEM *block,
+                                    int n, int qscale)
+{
+    int i, n_coeffs;
+    uint64_t qmul, qadd;
+    uint64_t correction;
+    DCTELEM *orig_block = block;
+    DCTELEM block0;             /* might not be used uninitialized */
+
+    qadd = WORD_VEC((qscale - 1) | 1);
+    qmul = qscale << 1;
+    /* This mask kills spill from negative subwords to the next subword.  */
+    correction = WORD_VEC((qmul - 1) + 1); /* multiplication / addition */
+
+    if (!s->h263_aic) {
+        if (n < 4)
+            block0 = block[0] * s->y_dc_scale;
+        else
+            block0 = block[0] * s->c_dc_scale;
+    } else {
+        qadd = 0;
+    }
+    n_coeffs = 63; // does not always use zigzag table
+
+    for(i = 0; i <= n_coeffs; block += 4, i += 4) {
+        uint64_t levels, negmask, zeros, add;
+
+        levels = ldq(block);
+        if (levels == 0)
+            continue;
+
+#ifdef __alpha_max__
+        /* I don't think the speed difference justifies runtime
+           detection.  */
+        negmask = maxsw4(levels, -1); /* negative -> ffff (-1) */
+        negmask = minsw4(negmask, 0); /* positive -> 0000 (0) */
+#else
+        negmask = cmpbge(WORD_VEC(0x7fff), levels);
+        negmask &= (negmask >> 1) | (1 << 7);
+        negmask = zap(-1, negmask);
+#endif
+
+        zeros = cmpbge(0, levels);
+        zeros &= zeros >> 1;
+        /* zeros |= zeros << 1 is not needed since qadd <= 255, so
+           zapping the lower byte suffices.  */
+
+        levels *= qmul;
+        levels -= correction & (negmask << 16);
+
+        /* Negate qadd for negative levels.  */
+        add = qadd ^ negmask;
+        add += WORD_VEC(0x0001) & negmask;
+        /* Set qadd to 0 for levels == 0.  */
+        add = zap(add, zeros);
+
+        levels += add;
+
+        stq(levels, block);
+    }
+
+    if (s->mb_intra && !s->h263_aic)
+        orig_block[0] = block0;
+}
+
+static void dct_unquantize_h263_inter_axp(MpegEncContext *s, DCTELEM *block,
+                                    int n, int qscale)
+{
+    int i, n_coeffs;
+    uint64_t qmul, qadd;
+    uint64_t correction;
+
+    qadd = WORD_VEC((qscale - 1) | 1);
+    qmul = qscale << 1;
+    /* This mask kills spill from negative subwords to the next subword.  */
+    correction = WORD_VEC((qmul - 1) + 1); /* multiplication / addition */
+
+    n_coeffs = s->intra_scantable.raster_end[s->block_last_index[n]];
+
+    for(i = 0; i <= n_coeffs; block += 4, i += 4) {
+        uint64_t levels, negmask, zeros, add;
+
+        levels = ldq(block);
+        if (levels == 0)
+            continue;
+
+#ifdef __alpha_max__
+        /* I don't think the speed difference justifies runtime
+           detection.  */
+        negmask = maxsw4(levels, -1); /* negative -> ffff (-1) */
+        negmask = minsw4(negmask, 0); /* positive -> 0000 (0) */
+#else
+        negmask = cmpbge(WORD_VEC(0x7fff), levels);
+        negmask &= (negmask >> 1) | (1 << 7);
+        negmask = zap(-1, negmask);
+#endif
+
+        zeros = cmpbge(0, levels);
+        zeros &= zeros >> 1;
+        /* zeros |= zeros << 1 is not needed since qadd <= 255, so
+           zapping the lower byte suffices.  */
+
+        levels *= qmul;
+        levels -= correction & (negmask << 16);
+
+        /* Negate qadd for negative levels.  */
+        add = qadd ^ negmask;
+        add += WORD_VEC(0x0001) & negmask;
+        /* Set qadd to 0 for levels == 0.  */
+        add = zap(add, zeros);
+
+        levels += add;
+
+        stq(levels, block);
+    }
+}
+
+void MPV_common_init_axp(MpegEncContext *s)
+{
+    s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_axp;
+    s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_axp;
+}