* Alpha optimized DSP utils
* Copyright (c) 2002 Falk Hueffner <falk@debian.org>
*
- * This library is free software; you can redistribute it and/or
+ * This file is part of Libav.
+ *
+ * Libav 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 of the License, or (at your option) any later version.
+ * version 2.1 of the License, or (at your option) any later version.
*
- * This library is distributed in the hope that it will be useful,
+ * Libav 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 this library; if not, write to the Free Software
+ * License along with Libav; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
+#include "libavcodec/dsputil.h"
+#include "libavcodec/mpegvideo.h"
#include "asm.h"
-#include "../dsputil.h"
-#include "../mpegvideo.h"
-static void dct_unquantize_h263_intra_axp(MpegEncContext *s, DCTELEM *block,
- int n, int qscale)
+static void dct_unquantize_h263_axp(DCTELEM *block, int n_coeffs,
+ uint64_t qscale, uint64_t qadd)
{
- 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 */
+ uint64_t qmul = qscale << 1;
+ uint64_t correction = WORD_VEC(qmul * 255 >> 8);
+ int i;
- 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
+ qadd = WORD_VEC(qadd);
for(i = 0; i <= n_coeffs; block += 4, i += 4) {
- uint64_t levels, negmask, zeros, add;
+ uint64_t levels, negmask, zeros, add, sub;
levels = ldq(block);
if (levels == 0)
levels *= qmul;
levels -= correction & (negmask << 16);
- /* Negate qadd for negative levels. */
- add = qadd ^ negmask;
- add += WORD_VEC(0x0001) & negmask;
+ add = qadd & ~negmask;
+ sub = qadd & negmask;
/* Set qadd to 0 for levels == 0. */
add = zap(add, zeros);
-
levels += add;
+ levels -= sub;
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,
+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;
-
- 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
+ int n_coeffs;
+ uint64_t qadd;
+ DCTELEM block0 = block[0];
- zeros = cmpbge(0, levels);
- zeros &= zeros >> 1;
- /* zeros |= zeros << 1 is not needed since qadd <= 255, so
- zapping the lower byte suffices. */
+ if (!s->h263_aic) {
+ if (n < 4)
+ block0 *= s->y_dc_scale;
+ else
+ block0 *= s->c_dc_scale;
+ qadd = (qscale - 1) | 1;
+ } else {
+ qadd = 0;
+ }
- levels *= qmul;
- levels -= correction & (negmask << 16);
+ if(s->ac_pred)
+ n_coeffs = 63;
+ else
+ n_coeffs = s->inter_scantable.raster_end[s->block_last_index[n]];
- /* Negate qadd for negative levels. */
- add = qadd ^ negmask;
- add += WORD_VEC(0x0001) & negmask;
- /* Set qadd to 0 for levels == 0. */
- add = zap(add, zeros);
+ dct_unquantize_h263_axp(block, n_coeffs, qscale, qadd);
- levels += add;
+ block[0] = block0;
+}
- stq(levels, block);
- }
+static void dct_unquantize_h263_inter_axp(MpegEncContext *s, DCTELEM *block,
+ int n, int qscale)
+{
+ int n_coeffs = s->inter_scantable.raster_end[s->block_last_index[n]];
+ dct_unquantize_h263_axp(block, n_coeffs, qscale, (qscale - 1) | 1);
}
void MPV_common_init_axp(MpegEncContext *s)