swscale/ppc: Move VSX-using code to its own file

Passes fate on LE (with "lavc/jrevdct: Avoid an aliasing violation" applied).
Signed-off-by: Lauri Kasanen <cand@gmx.com>
Tested-by: Michael Kostylev on BE
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

libswscale/ppc/Makefile

 OBJS += ppc/swscale_altivec.o                                           \
         ppc/yuv2rgb_altivec.o                                           \
         ppc/yuv2yuv_altivec.o                                           \
+        ppc/swscale_vsx.o

libswscale/ppc/swscale_ppc_template.c

+/*
+ * AltiVec-enhanced yuv2yuvX
+ *
+ * Copyright (C) 2004 Romain Dolbeau <romain@dolbeau.org>
+ * based on the equivalent C code in swscale.c
+ *
+ * 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
+ */
+
+static void FUNC(yuv2planeX_16)(const int16_t *filter, int filterSize,
+                                  const int16_t **src, uint8_t *dest,
+                                  const uint8_t *dither, int offset, int x)
+{
+    register int i, j;
+    LOCAL_ALIGNED(16, int, val, [16]);
+    vector signed int vo1, vo2, vo3, vo4;
+    vector unsigned short vs1, vs2;
+    vector unsigned char vf;
+    vector unsigned int altivec_vectorShiftInt19 =
+        vec_add(vec_splat_u32(10), vec_splat_u32(9));
+
+    for (i = 0; i < 16; i++)
+        val[i] = dither[(x + i + offset) & 7] << 12;
+
+    vo1 = vec_ld(0,  val);
+    vo2 = vec_ld(16, val);
+    vo3 = vec_ld(32, val);
+    vo4 = vec_ld(48, val);
+
+    for (j = 0; j < filterSize; j++) {
+        unsigned int joffset=j<<1;
+        unsigned int xoffset=x<<1;
+        vector unsigned char perm;
+        vector signed short l1,vLumFilter;
+        LOAD_FILTER(vLumFilter,filter);
+        vLumFilter = vec_splat(vLumFilter, 0);
+        LOAD_L1(l1,src[j],perm);
+        yuv2planeX_8(vo1, vo2, l1, src[j], x,     perm, vLumFilter);
+        yuv2planeX_8(vo3, vo4, l1, src[j], x + 8, perm, vLumFilter);
+    }
+
+    vo1 = vec_sra(vo1, altivec_vectorShiftInt19);
+    vo2 = vec_sra(vo2, altivec_vectorShiftInt19);
+    vo3 = vec_sra(vo3, altivec_vectorShiftInt19);
+    vo4 = vec_sra(vo4, altivec_vectorShiftInt19);
+    vs1 = vec_packsu(vo1, vo2);
+    vs2 = vec_packsu(vo3, vo4);
+    vf  = vec_packsu(vs1, vs2);
+    VEC_ST(vf, 0, dest);
+}
+
+
+static inline void yuv2planeX_u(const int16_t *filter, int filterSize,
+                                const int16_t **src, uint8_t *dest, int dstW,
+                                const uint8_t *dither, int offset, int x)
+{
+    int i, j;
+
+    for (i = x; i < dstW; i++) {
+        int t = dither[(i + offset) & 7] << 12;
+        for (j = 0; j < filterSize; j++)
+            t += src[j][i] * filter[j];
+        dest[i] = av_clip_uint8(t >> 19);
+    }
+}
+
+static void FUNC(yuv2planeX)(const int16_t *filter, int filterSize,
+                               const int16_t **src, uint8_t *dest, int dstW,
+                               const uint8_t *dither, int offset)
+{
+    int dst_u = -(uintptr_t)dest & 15;
+    int i;
+
+    yuv2planeX_u(filter, filterSize, src, dest, dst_u, dither, offset, 0);
+
+    for (i = dst_u; i < dstW - 15; i += 16)
+        FUNC(yuv2planeX_16)(filter, filterSize, src, dest + i, dither,
+                              offset, i);
+
+    yuv2planeX_u(filter, filterSize, src, dest, dstW, dither, offset, i);
+}
+
+static void FUNC(hScale_real)(SwsContext *c, int16_t *dst, int dstW,
+                                const uint8_t *src, const int16_t *filter,
+                                const int32_t *filterPos, int filterSize)
+{
+    register int i;
+    LOCAL_ALIGNED(16, int, tempo, [4]);
+
+    if (filterSize % 4) {
+        for (i = 0; i < dstW; i++) {
+            register int j;
+            register int srcPos = filterPos[i];
+            register int val    = 0;
+            for (j = 0; j < filterSize; j++)
+                val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
+            dst[i] = FFMIN(val >> 7, (1 << 15) - 1);
+        }
+    } else
+        switch (filterSize) {
+        case 4:
+            for (i = 0; i < dstW; i++) {
+                register int srcPos = filterPos[i];
+
+                vector unsigned char src_vF = unaligned_load(srcPos, src);
+                vector signed short src_v, filter_v;
+                vector signed int val_vEven, val_s;
+                src_v = // vec_unpackh sign-extends...
+                        (vector signed short)(VEC_MERGEH((vector unsigned char)vzero, src_vF));
+                // now put our elements in the even slots
+                src_v = vec_mergeh(src_v, (vector signed short)vzero);
+                GET_VF4(i, filter_v, filter);
+                val_vEven = vec_mule(src_v, filter_v);
+                val_s     = vec_sums(val_vEven, vzero);
+                vec_st(val_s, 0, tempo);
+                dst[i] = FFMIN(tempo[3] >> 7, (1 << 15) - 1);
+            }
+        break;
+        case 8:
+            for (i = 0; i < dstW; i++) {
+                register int srcPos = filterPos[i];
+                vector unsigned char src_vF, src_v0, src_v1;
+                vector unsigned char permS;
+                vector signed short src_v, filter_v;
+                vector signed int val_v, val_s;
+                FIRST_LOAD(src_v0, srcPos, src, permS);
+                LOAD_SRCV8(srcPos, 0, src, permS, src_v0, src_v1, src_vF);
+                src_v = // vec_unpackh sign-extends...
+                        (vector signed short)(VEC_MERGEH((vector unsigned char)vzero, src_vF));
+                filter_v = vec_ld(i << 4, filter);
+                val_v = vec_msums(src_v, filter_v, (vector signed int)vzero);
+                val_s = vec_sums(val_v, vzero);
+                vec_st(val_s, 0, tempo);
+                dst[i] = FFMIN(tempo[3] >> 7, (1 << 15) - 1);
+            }
+        break;
+
+        case 16:
+            for (i = 0; i < dstW; i++) {
+                register int srcPos = filterPos[i];
+
+                vector unsigned char src_vF = unaligned_load(srcPos, src);
+                vector signed short src_vA = // vec_unpackh sign-extends...
+                                             (vector signed short)(VEC_MERGEH((vector unsigned char)vzero, src_vF));
+                vector signed short src_vB = // vec_unpackh sign-extends...
+                                             (vector signed short)(VEC_MERGEL((vector unsigned char)vzero, src_vF));
+                vector signed short filter_v0 = vec_ld(i << 5, filter);
+                vector signed short filter_v1 = vec_ld((i << 5) + 16, filter);
+
+                vector signed int val_acc = vec_msums(src_vA, filter_v0, (vector signed int)vzero);
+                vector signed int val_v   = vec_msums(src_vB, filter_v1, val_acc);
+
+                vector signed int val_s = vec_sums(val_v, vzero);
+
+                VEC_ST(val_s, 0, tempo);
+                dst[i] = FFMIN(tempo[3] >> 7, (1 << 15) - 1);
+            }
+        break;
+
+        default:
+            for (i = 0; i < dstW; i++) {
+                register int j, offset = i * 2 * filterSize;
+                register int srcPos = filterPos[i];
+
+                vector signed int val_s, val_v = (vector signed int)vzero;
+                vector signed short filter_v0R;
+                vector unsigned char permF, src_v0, permS;
+                FIRST_LOAD(filter_v0R, offset, filter, permF);
+                FIRST_LOAD(src_v0, srcPos, src, permS);
+
+                for (j = 0; j < filterSize - 15; j += 16) {
+                    vector unsigned char src_v1, src_vF;
+                    vector signed short filter_v1R, filter_v2R, filter_v0, filter_v1;
+                    LOAD_SRCV(srcPos, j, src, permS, src_v0, src_v1, src_vF);
+                    vector signed short src_vA = // vec_unpackh sign-extends...
+                                                 (vector signed short)(VEC_MERGEH((vector unsigned char)vzero, src_vF));
+                    vector signed short src_vB = // vec_unpackh sign-extends...
+                                                 (vector signed short)(VEC_MERGEL((vector unsigned char)vzero, src_vF));
+                    GET_VFD(i, j, filter, filter_v0R, filter_v1R, permF, filter_v0, 0);
+                    GET_VFD(i, j, filter, filter_v1R, filter_v2R, permF, filter_v1, 16);
+
+                    vector signed int val_acc = vec_msums(src_vA, filter_v0, val_v);
+                    val_v = vec_msums(src_vB, filter_v1, val_acc);
+                    UPDATE_PTR(filter_v2R, filter_v0R, src_v1, src_v0);
+                }
+
+                if (j < filterSize - 7) {
+                    // loading src_v0 is useless, it's already done above
+                    vector unsigned char src_v1, src_vF;
+                    vector signed short src_v, filter_v1R, filter_v;
+                    LOAD_SRCV8(srcPos, j, src, permS, src_v0, src_v1, src_vF);
+                    src_v = // vec_unpackh sign-extends...
+                            (vector signed short)(VEC_MERGEH((vector unsigned char)vzero, src_vF));
+                    GET_VFD(i, j, filter, filter_v0R, filter_v1R, permF, filter_v, 0);
+                    val_v = vec_msums(src_v, filter_v, val_v);
+                }
+                val_s = vec_sums(val_v, vzero);
+
+                VEC_ST(val_s, 0, tempo);
+                dst[i] = FFMIN(tempo[3] >> 7, (1 << 15) - 1);
+            }
+        }
+}

libswscale/ppc/swscale_vsx.c

+/*
+ * AltiVec-enhanced yuv2yuvX
+ *
+ * Copyright (C) 2004 Romain Dolbeau <romain@dolbeau.org>
+ * based on the equivalent C code in swscale.c
+ *
+ * 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 <inttypes.h>
+
+#include "config.h"
+#include "libswscale/swscale.h"
+#include "libswscale/swscale_internal.h"
+#include "libavutil/attributes.h"
+#include "libavutil/cpu.h"
+#include "yuv2rgb_altivec.h"
+#include "libavutil/ppc/util_altivec.h"
+
+#if HAVE_VSX
+#define vzero vec_splat_s32(0)
+
+#if !HAVE_BIGENDIAN
+#define  GET_LS(a,b,c,s) {\
+        ls  = a;\
+        a = vec_vsx_ld(((b) << 1)  + 16, s);\
+    }
+
+#define yuv2planeX_8(d1, d2, l1, src, x, perm, filter) do {\
+        vector signed short ls;\
+        GET_LS(l1, x, perm, src);\
+        vector signed int   i1  = vec_mule(filter, ls);\
+        vector signed int   i2  = vec_mulo(filter, ls);\
+        vector signed int   vf1, vf2;\
+        vf1 = vec_mergeh(i1, i2);\
+        vf2 = vec_mergel(i1, i2);\
+        d1 = vec_add(d1, vf1);\
+        d2 = vec_add(d2, vf2);\
+    } while (0)
+
+#define LOAD_FILTER(vf,f) {\
+        vf = vec_vsx_ld(joffset, f);\
+}
+#define LOAD_L1(ll1,s,p){\
+        ll1  = vec_vsx_ld(xoffset, s);\
+}
+
+// The 3 above is 2 (filterSize == 4) + 1 (sizeof(short) == 2).
+
+// The neat trick: We only care for half the elements,
+// high or low depending on (i<<3)%16 (it's 0 or 8 here),
+// and we're going to use vec_mule, so we choose
+// carefully how to "unpack" the elements into the even slots.
+#define GET_VF4(a, vf, f) {\
+    vf = (vector signed short)vec_vsx_ld(a << 3, f);\
+    vf = vec_mergeh(vf, (vector signed short)vzero);\
+}
+#define FIRST_LOAD(sv, pos, s, per) {}
+#define UPDATE_PTR(s0, d0, s1, d1) {}
+#define LOAD_SRCV(pos, a, s, per, v0, v1, vf) {\
+    vf = vec_vsx_ld(pos + a, s);\
+}
+#define LOAD_SRCV8(pos, a, s, per, v0, v1, vf) LOAD_SRCV(pos, a, s, per, v0, v1, vf)
+#define GET_VFD(a, b, f, vf0, vf1, per, vf, off) {\
+    vf  = vec_vsx_ld((a * 2 * filterSize) + (b * 2) + off, f);\
+}
+
+#define FUNC(name) name ## _vsx
+#include "swscale_ppc_template.c"
+#undef FUNC
+
+#endif /* !HAVE_BIGENDIAN */
+
+static void yuv2plane1_8_u(const int16_t *src, uint8_t *dest, int dstW,
+                           const uint8_t *dither, int offset, int start)
+{
+    int i;
+    for (i = start; i < dstW; i++) {
+        int val = (src[i] + dither[(i + offset) & 7]) >> 7;
+        dest[i] = av_clip_uint8(val);
+    }
+}
+
+static void yuv2plane1_8_vsx(const int16_t *src, uint8_t *dest, int dstW,
+                           const uint8_t *dither, int offset)
+{
+    const int dst_u = -(uintptr_t)dest & 15;
+    int i, j;
+    LOCAL_ALIGNED(16, int16_t, val, [16]);
+    const vector uint16_t shifts = (vector uint16_t) {7, 7, 7, 7, 7, 7, 7, 7};
+    vector int16_t vi, vileft, ditherleft, ditherright;
+    vector uint8_t vd;
+
+    for (j = 0; j < 16; j++) {
+        val[j] = dither[(dst_u + offset + j) & 7];
+    }
+
+    ditherleft = vec_ld(0, val);
+    ditherright = vec_ld(0, &val[8]);
+
+    yuv2plane1_8_u(src, dest, dst_u, dither, offset, 0);
+
+    for (i = dst_u; i < dstW - 15; i += 16) {
+
+        vi = vec_vsx_ld(0, &src[i]);
+        vi = vec_adds(ditherleft, vi);
+        vileft = vec_sra(vi, shifts);
+
+        vi = vec_vsx_ld(0, &src[i + 8]);
+        vi = vec_adds(ditherright, vi);
+        vi = vec_sra(vi, shifts);
+
+        vd = vec_packsu(vileft, vi);
+        vec_st(vd, 0, &dest[i]);
+    }
+
+    yuv2plane1_8_u(src, dest, dstW, dither, offset, i);
+}
+
+#endif /* HAVE_VSX */
+
+av_cold void ff_sws_init_swscale_vsx(SwsContext *c)
+{
+#if HAVE_VSX
+    enum AVPixelFormat dstFormat = c->dstFormat;
+
+    if (!(av_get_cpu_flags() & AV_CPU_FLAG_VSX))
+        return;
+
+#if !HAVE_BIGENDIAN
+    if (c->srcBpc == 8 && c->dstBpc <= 14) {
+        c->hyScale = c->hcScale = hScale_real_vsx;
+    }
+    if (!is16BPS(dstFormat) && !isNBPS(dstFormat) &&
+        dstFormat != AV_PIX_FMT_NV12 && dstFormat != AV_PIX_FMT_NV21 &&
+        dstFormat != AV_PIX_FMT_GRAYF32BE && dstFormat != AV_PIX_FMT_GRAYF32LE &&
+        !c->needAlpha) {
+        c->yuv2planeX = yuv2planeX_vsx;
+    }
+#endif
+
+    if (!(c->flags & (SWS_BITEXACT | SWS_FULL_CHR_H_INT)) && !c->needAlpha) {
+        switch (c->dstBpc) {
+        case 8:
+            c->yuv2plane1 = yuv2plane1_8_vsx;
+            break;
+        }
+    }
+#endif /* HAVE_VSX */
+}

libswscale/swscale_internal.h

@@ -868,6 +868,7 @@ void ff_sws_init_output_funcs(SwsContext *c,
                               yuv2packedX_fn *yuv2packedX,
                               yuv2anyX_fn *yuv2anyX);
 void ff_sws_init_swscale_ppc(SwsContext *c);
+void ff_sws_init_swscale_vsx(SwsContext *c);
 void ff_sws_init_swscale_x86(SwsContext *c);
 void ff_sws_init_swscale_aarch64(SwsContext *c);
 void ff_sws_init_swscale_arm(SwsContext *c);