Merge commit '65d5d5865845f057cc6530a8d0f34db952d9009c'

* commit '65d5d5865845f057cc6530a8d0f34db952d9009c':
  dsputil: Move SVQ1 encoding specific bits into svq1enc

Conflicts:
	libavcodec/x86/Makefile

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

libavcodec/dsputil.c

@@ -2216,16 +2216,6 @@ static int vsse16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2,
     return score;
 }
 
-static int ssd_int8_vs_int16_c(const int8_t *pix1, const int16_t *pix2,
-                               int size)
-{
-    int score = 0, i;
-
-    for (i = 0; i < size; i++)
-        score += (pix1[i] - pix2[i]) * (pix1[i] - pix2[i]);
-    return score;
-}
-
 #define WRAPPER8_16_SQ(name8, name16)                                   \
 static int name16(MpegEncContext *s, uint8_t *dst, uint8_t *src,        \
                   int stride, int h)                                    \
@@ -2626,8 +2616,6 @@ av_cold void ff_dsputil_init(DSPContext *c, AVCodecContext *avctx)
     ff_dsputil_init_dwt(c);
 #endif
 
-    c->ssd_int8_vs_int16 = ssd_int8_vs_int16_c;
-
     c->bswap_buf   = bswap_buf;
     c->bswap16_buf = bswap16_buf;
 

libavcodec/dsputil.h

@@ -177,9 +177,6 @@ typedef struct DSPContext {
     me_cmp_func ildct_cmp[6]; // only width 16 used
     me_cmp_func frame_skip_cmp[6]; // only width 8 used
 
-    int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
-                             int size);
-
     qpel_mc_func put_qpel_pixels_tab[2][16];
     qpel_mc_func avg_qpel_pixels_tab[2][16];
     qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];

libavcodec/ppc/Makefile

@@ -12,6 +12,7 @@ OBJS-$(CONFIG_MPEGVIDEO)               += ppc/mpegvideo_altivec.o
 OBJS-$(CONFIG_VIDEODSP)                += ppc/videodsp_ppc.o
 OBJS-$(CONFIG_VP3DSP)                  += ppc/vp3dsp_altivec.o
 
+OBJS-$(CONFIG_SVQ1_ENCODER)            += ppc/svq1enc_altivec.o
 OBJS-$(CONFIG_VC1_DECODER)             += ppc/vc1dsp_altivec.o
 OBJS-$(CONFIG_VORBIS_DECODER)          += ppc/vorbisdsp_altivec.o
 OBJS-$(CONFIG_VP7_DECODER)             += ppc/vp8dsp_altivec.o

libavcodec/ppc/int_altivec.c

@@ -34,48 +34,6 @@
 #include "libavcodec/dsputil.h"
 #include "dsputil_altivec.h"
 
-static int ssd_int8_vs_int16_altivec(const int8_t *pix1, const int16_t *pix2,
-                                     int size)
-{
-    int i, size16 = size >> 4;
-    vector signed char vpix1;
-    vector signed short vpix2, vdiff, vpix1l, vpix1h;
-    union {
-        vector signed int vscore;
-        int32_t score[4];
-    } u = { .vscore = vec_splat_s32(0) };
-
-// XXX lazy way, fix it later
-
-    while (size16) {
-        // score += (pix1[i] - pix2[i]) * (pix1[i] - pix2[i]);
-        // load pix1 and the first batch of pix2
-
-        vpix1 = vec_unaligned_load(pix1);
-        vpix2 = vec_unaligned_load(pix2);
-        pix2 += 8;
-        // unpack
-        vpix1h = vec_unpackh(vpix1);
-        vdiff  = vec_sub(vpix1h, vpix2);
-        vpix1l = vec_unpackl(vpix1);
-        // load another batch from pix2
-        vpix2    = vec_unaligned_load(pix2);
-        u.vscore = vec_msum(vdiff, vdiff, u.vscore);
-        vdiff    = vec_sub(vpix1l, vpix2);
-        u.vscore = vec_msum(vdiff, vdiff, u.vscore);
-        pix1    += 16;
-        pix2    += 8;
-        size16--;
-    }
-    u.vscore = vec_sums(u.vscore, vec_splat_s32(0));
-
-    size %= 16;
-    for (i = 0; i < size; i++)
-        u.score[3] += (pix1[i] - pix2[i]) * (pix1[i] - pix2[i]);
-
-    return u.score[3];
-}
-
 static int32_t scalarproduct_int16_altivec(const int16_t *v1, const int16_t *v2,
                                            int order)
 {
@@ -140,8 +98,6 @@ static int32_t scalarproduct_and_madd_int16_altivec(int16_t *v1,
 
 av_cold void ff_int_init_altivec(DSPContext *c, AVCodecContext *avctx)
 {
-    c->ssd_int8_vs_int16 = ssd_int8_vs_int16_altivec;
-
     c->scalarproduct_int16 = scalarproduct_int16_altivec;
 
     c->scalarproduct_and_madd_int16 = scalarproduct_and_madd_int16_altivec;

libavcodec/ppc/svq1enc_altivec.c

@@ -0,0 +1,80 @@
+/*
+ * Copyright (c) 2007 Luca Barbato <lu_zero@gentoo.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 <stdint.h>
+
+#include "config.h"
+#if HAVE_ALTIVEC_H
+#include <altivec.h>
+#endif
+
+#include "libavutil/attributes.h"
+#include "libavutil/ppc/types_altivec.h"
+#include "libavutil/ppc/util_altivec.h"
+#include "libavcodec/svq1enc.h"
+
+#if HAVE_ALTIVEC
+static int ssd_int8_vs_int16_altivec(const int8_t *pix1, const int16_t *pix2,
+                                     int size)
+{
+    int i, size16 = size >> 4;
+    vector signed char vpix1;
+    vector signed short vpix2, vdiff, vpix1l, vpix1h;
+    union {
+        vector signed int vscore;
+        int32_t score[4];
+    } u = { .vscore = vec_splat_s32(0) };
+
+    while (size16) {
+        // score += (pix1[i] - pix2[i]) * (pix1[i] - pix2[i]);
+        // load pix1 and the first batch of pix2
+
+        vpix1 = vec_unaligned_load(pix1);
+        vpix2 = vec_unaligned_load(pix2);
+        pix2 += 8;
+        // unpack
+        vpix1h = vec_unpackh(vpix1);
+        vdiff  = vec_sub(vpix1h, vpix2);
+        vpix1l = vec_unpackl(vpix1);
+        // load another batch from pix2
+        vpix2    = vec_unaligned_load(pix2);
+        u.vscore = vec_msum(vdiff, vdiff, u.vscore);
+        vdiff    = vec_sub(vpix1l, vpix2);
+        u.vscore = vec_msum(vdiff, vdiff, u.vscore);
+        pix1    += 16;
+        pix2    += 8;
+        size16--;
+    }
+    u.vscore = vec_sums(u.vscore, vec_splat_s32(0));
+
+    size %= 16;
+    for (i = 0; i < size; i++)
+        u.score[3] += (pix1[i] - pix2[i]) * (pix1[i] - pix2[i]);
+
+    return u.score[3];
+}
+#endif /* HAVE_ALTIVEC */
+
+av_cold void ff_svq1enc_init_ppc(SVQ1EncContext *c)
+{
+#if HAVE_ALTIVEC
+    c->ssd_int8_vs_int16 = ssd_int8_vs_int16_altivec;
+#endif /* HAVE_ALTIVEC */
+}

libavcodec/svq1enc.c

@@ -34,48 +34,11 @@
 #include "internal.h"
 #include "mpegutils.h"
 #include "svq1.h"
+#include "svq1enc.h"
 #include "svq1enc_cb.h"
 #include "libavutil/avassert.h"
 
 
-typedef struct SVQ1EncContext {
-    /* FIXME: Needed for motion estimation, should not be used for anything
-     * else, the idea is to make the motion estimation eventually independent
-     * of MpegEncContext, so this will be removed then. */
-    MpegEncContext m;
-    AVCodecContext *avctx;
-    DSPContext dsp;
-    HpelDSPContext hdsp;
-    AVFrame *current_picture;
-    AVFrame *last_picture;
-    PutBitContext pb;
-    GetBitContext gb;
-
-    /* why ooh why this sick breadth first order,
-     * everything is slower and more complex */
-    PutBitContext reorder_pb[6];
-
-    int frame_width;
-    int frame_height;
-
-    /* Y plane block dimensions */
-    int y_block_width;
-    int y_block_height;
-
-    /* U & V plane (C planes) block dimensions */
-    int c_block_width;
-    int c_block_height;
-
-    uint16_t *mb_type;
-    uint32_t *dummy;
-    int16_t (*motion_val8[3])[2];
-    int16_t (*motion_val16[3])[2];
-
-    int64_t rd_total;
-
-    uint8_t *scratchbuf;
-} SVQ1EncContext;
-
 static void svq1_write_header(SVQ1EncContext *s, int frame_type)
 {
     int i;
@@ -113,6 +76,16 @@ static void svq1_write_header(SVQ1EncContext *s, int frame_type)
 #define QUALITY_THRESHOLD    100
 #define THRESHOLD_MULTIPLIER 0.6
 
+static int ssd_int8_vs_int16_c(const int8_t *pix1, const int16_t *pix2,
+                               int size)
+{
+    int score = 0, i;
+
+    for (i = 0; i < size; i++)
+        score += (pix1[i] - pix2[i]) * (pix1[i] - pix2[i]);
+    return score;
+}
+
 static int encode_block(SVQ1EncContext *s, uint8_t *src, uint8_t *ref,
                         uint8_t *decoded, int stride, int level,
                         int threshold, int lambda, int intra)
@@ -174,7 +147,7 @@ static int encode_block(SVQ1EncContext *s, uint8_t *src, uint8_t *ref,
                 int sqr, diff, score;
 
                 vector = codebook + stage * size * 16 + i * size;
-                sqr    = s->dsp.ssd_int8_vs_int16(vector, block[stage], size);
+                sqr    = s->ssd_int8_vs_int16(vector, block[stage], size);
                 diff   = block_sum[stage] - sum;
                 score  = sqr - (diff * (int64_t)diff >> (level + 3)); // FIXME: 64bit slooow
                 if (score < best_vector_score) {
@@ -580,6 +553,13 @@ static av_cold int svq1_encode_init(AVCodecContext *avctx)
                                         s->y_block_height * sizeof(int16_t));
     s->dummy               = av_mallocz((s->y_block_width + 1) *
                                         s->y_block_height * sizeof(int32_t));
+    s->ssd_int8_vs_int16   = ssd_int8_vs_int16_c;
+
+    if (ARCH_PPC)
+        ff_svq1enc_init_ppc(s);
+    if (ARCH_X86)
+        ff_svq1enc_init_x86(s);
+
     ff_h263_encode_init(&s->m); // mv_penalty
 
     return 0;

libavcodec/svq1enc.h

@@ -0,0 +1,78 @@
+/*
+ * SVQ1 encoder
+ *
+ * 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
+ */
+
+#ifndef AVCODEC_SVQ1ENC_H
+#define AVCODEC_SVQ1ENC_H
+
+#include <stdint.h>
+
+#include "libavutil/frame.h"
+#include "avcodec.h"
+#include "dsputil.h"
+#include "get_bits.h"
+#include "hpeldsp.h"
+#include "mpegvideo.h"
+#include "put_bits.h"
+
+typedef struct SVQ1EncContext {
+    /* FIXME: Needed for motion estimation, should not be used for anything
+     * else, the idea is to make the motion estimation eventually independent
+     * of MpegEncContext, so this will be removed then. */
+    MpegEncContext m;
+    AVCodecContext *avctx;
+    DSPContext dsp;
+    HpelDSPContext hdsp;
+    AVFrame *current_picture;
+    AVFrame *last_picture;
+    PutBitContext pb;
+    GetBitContext gb;
+
+    /* why ooh why this sick breadth first order,
+     * everything is slower and more complex */
+    PutBitContext reorder_pb[6];
+
+    int frame_width;
+    int frame_height;
+
+    /* Y plane block dimensions */
+    int y_block_width;
+    int y_block_height;
+
+    /* U & V plane (C planes) block dimensions */
+    int c_block_width;
+    int c_block_height;
+
+    uint16_t *mb_type;
+    uint32_t *dummy;
+    int16_t (*motion_val8[3])[2];
+    int16_t (*motion_val16[3])[2];
+
+    int64_t rd_total;
+
+    uint8_t *scratchbuf;
+
+    int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
+                             int size);
+} SVQ1EncContext;
+
+void ff_svq1enc_init_ppc(SVQ1EncContext *c);
+void ff_svq1enc_init_x86(SVQ1EncContext *c);
+
+#endif /* AVCODEC_SVQ1ENC_H */

libavcodec/x86/Makefile

@@ -59,6 +59,7 @@ MMX-OBJS-$(CONFIG_HUFFYUVDSP)          += x86/huffyuvdsp_mmx.o
 
 MMX-OBJS-$(CONFIG_SNOW_DECODER)        += x86/snowdsp.o
 MMX-OBJS-$(CONFIG_SNOW_ENCODER)        += x86/snowdsp.o
+MMX-OBJS-$(CONFIG_SVQ1_ENCODER)        += x86/svq1enc_mmx.o
 MMX-OBJS-$(CONFIG_VC1_DECODER)         += x86/vc1dsp_mmx.o
 
 YASM-OBJS                              += x86/deinterlace.o             \

libavcodec/x86/dsputilenc_mmx.c

@@ -703,40 +703,6 @@ static int vsad16_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
 #undef SUM
 
 
-static int ssd_int8_vs_int16_mmx(const int8_t *pix1, const int16_t *pix2,
-                                 int size)
-{
-    int sum;
-    x86_reg i = size;
-
-    __asm__ volatile (
-        "pxor %%mm4, %%mm4 \n"
-        "1: \n"
-        "sub $8, %0 \n"
-        "movq (%2, %0), %%mm2 \n"
-        "movq (%3, %0, 2), %%mm0 \n"
-        "movq 8(%3, %0, 2), %%mm1 \n"
-        "punpckhbw %%mm2, %%mm3 \n"
-        "punpcklbw %%mm2, %%mm2 \n"
-        "psraw $8, %%mm3 \n"
-        "psraw $8, %%mm2 \n"
-        "psubw %%mm3, %%mm1 \n"
-        "psubw %%mm2, %%mm0 \n"
-        "pmaddwd %%mm1, %%mm1 \n"
-        "pmaddwd %%mm0, %%mm0 \n"
-        "paddd %%mm1, %%mm4 \n"
-        "paddd %%mm0, %%mm4 \n"
-        "jg 1b \n"
-        "movq %%mm4, %%mm3 \n"
-        "psrlq $32, %%mm3 \n"
-        "paddd %%mm3, %%mm4 \n"
-        "movd %%mm4, %1 \n"
-        : "+r" (i), "=r" (sum)
-        : "r" (pix1), "r" (pix2));
-
-    return sum;
-}
-
 #define PHADDD(a, t)                            \
     "movq  " #a ", " #t "               \n\t"   \
     "psrlq    $32, " #a "               \n\t"   \
@@ -854,8 +820,6 @@ av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx,
             c->try_8x8basis = try_8x8basis_mmx;
         }
         c->add_8x8basis = add_8x8basis_mmx;
-
-        c->ssd_int8_vs_int16 = ssd_int8_vs_int16_mmx;
     }
 
     if (INLINE_AMD3DNOW(cpu_flags)) {

libavcodec/x86/svq1enc_mmx.c

@@ -0,0 +1,73 @@
+/*
+ * 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 "config.h"
+#include "libavutil/attributes.h"
+#include "libavutil/cpu.h"
+#include "libavutil/x86/asm.h"
+#include "libavutil/x86/cpu.h"
+#include "libavcodec/svq1enc.h"
+
+#if HAVE_INLINE_ASM
+
+static int ssd_int8_vs_int16_mmx(const int8_t *pix1, const int16_t *pix2,
+                                 int size)
+{
+    int sum;
+    x86_reg i = size;
+
+    __asm__ volatile (
+        "pxor %%mm4, %%mm4 \n"
+        "1: \n"
+        "sub $8, %0 \n"
+        "movq (%2, %0), %%mm2 \n"
+        "movq (%3, %0, 2), %%mm0 \n"
+        "movq 8(%3, %0, 2), %%mm1 \n"
+        "punpckhbw %%mm2, %%mm3 \n"
+        "punpcklbw %%mm2, %%mm2 \n"
+        "psraw $8, %%mm3 \n"
+        "psraw $8, %%mm2 \n"
+        "psubw %%mm3, %%mm1 \n"
+        "psubw %%mm2, %%mm0 \n"
+        "pmaddwd %%mm1, %%mm1 \n"
+        "pmaddwd %%mm0, %%mm0 \n"
+        "paddd %%mm1, %%mm4 \n"
+        "paddd %%mm0, %%mm4 \n"
+        "jg 1b \n"
+        "movq %%mm4, %%mm3 \n"
+        "psrlq $32, %%mm3 \n"
+        "paddd %%mm3, %%mm4 \n"
+        "movd %%mm4, %1 \n"
+        : "+r" (i), "=r" (sum)
+        : "r" (pix1), "r" (pix2));
+
+    return sum;
+}
+
+#endif /* HAVE_INLINE_ASM */
+
+av_cold void ff_svq1enc_init_x86(SVQ1EncContext *c)
+{
+#if HAVE_INLINE_ASM
+    int cpu_flags = av_get_cpu_flags();
+
+    if (INLINE_MMX(cpu_flags)) {
+        c->ssd_int8_vs_int16 = ssd_int8_vs_int16_mmx;
+    }
+#endif /* HAVE_INLINE_ASM */
+}