From f814169da15b8e0e8234bba2f5300338addebab7 Mon Sep 17 00:00:00 2001 From: Rong Yan Date: Thu, 16 Oct 2014 01:42:23 +0000 Subject: [PATCH] libswscale/ppc/swscale_altivec.c : fix hScale_altivec_real() yuv2planeX_16_altivec() yuv2planeX_8() for POWER LE --- libswscale/ppc/swscale_altivec.c | 169 +++++++++++++++++++++++++++++++++++++++ 1 file changed, 169 insertions(+) diff --git a/libswscale/ppc/swscale_altivec.c b/libswscale/ppc/swscale_altivec.c index 86f40ab..98249cf 100644 --- a/libswscale/ppc/swscale_altivec.c +++ b/libswscale/ppc/swscale_altivec.c @@ -33,6 +33,16 @@ #if HAVE_ALTIVEC #define vzero vec_splat_s32(0) +#if !HAVE_BIGENDIAN +#define yuv2planeX_8(d1, d2, l1, src, x, filter) do { \ + vector signed int i1 = vec_mule(filter, l1); \ + vector signed int i2 = vec_mulo(filter, l1); \ + vector signed int vf1 = vec_mergel(i2, i1); \ + vector signed int vf2 = vec_mergeh(i2, i1); \ + d1 = vec_add(d1, vf1); \ + d2 = vec_add(d2, vf2); \ + } while (0) +#else #define yuv2planeX_8(d1, d2, l1, src, x, perm, filter) do { \ vector signed short l2 = vec_ld(((x) << 1) + 16, src); \ vector signed short ls = vec_perm(l1, l2, perm); \ @@ -44,11 +54,49 @@ d2 = vec_add(d2, vf2); \ l1 = l2; \ } while (0) +#endif static void yuv2planeX_16_altivec(const int16_t *filter, int filterSize, const int16_t **src, uint8_t *dest, const uint8_t *dither, int offset, int x) { +#if !HAVE_BIGENDIAN + register int i, j; + DECLARE_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++) { + vector signed short l1, l2, vLumFilter = vec_vsx_ld(j << 1, filter); + vLumFilter = vec_splat(vLumFilter, 0); // lumFilter[j] is loaded 8 times in vLumFilter + + l1 = vec_vsx_ld(x << 1, src[j]); + l2 = vec_vsx_ld(((x) << 1) + 16, src[j]); + + yuv2planeX_8(vo1, vo2, l1, src[j], x, vLumFilter); + yuv2planeX_8(vo3, vo4, l2, src[j], x + 8, 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_vsx_st(vf, 0, dest); +#else /* else of #if !HAVE_BIGENDIAN */ register int i, j; DECLARE_ALIGNED(16, int, val)[16]; vector signed int vo1, vo2, vo3, vo4; @@ -86,6 +134,7 @@ static void yuv2planeX_16_altivec(const int16_t *filter, int filterSize, vs2 = vec_packsu(vo3, vo4); vf = vec_packsu(vs1, vs2); vec_st(vf, 0, dest); +#endif /* end of #if !HAVE_BIGENDIAN */ } static inline void yuv2planeX_u(const int16_t *filter, int filterSize, @@ -122,6 +171,125 @@ static void hScale_altivec_real(SwsContext *c, int16_t *dst, int dstW, const uint8_t *src, const int16_t *filter, const int32_t *filterPos, int filterSize) { +#if !HAVE_BIGENDIAN + register int i; + DECLARE_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 = vec_vsx_ld(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( src_vF, (vector unsigned char)vzero)); + // now put our elements in the even slots + src_v = vec_mergeh(src_v, (vector signed short)vzero); + filter_v = vec_vsx_ld(i << 3, filter); + filter_v = vec_mergeh(filter_v, (vector signed short)vzero); + + val_vEven = vec_mule(src_v, filter_v); + val_s = vec_sums(val_vEven, vzero); + vec_vsx_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 = vec_vsx_ld(srcPos, src); + vector signed short src_v, filter_v; + vector signed int val_v, val_s; + + src_v = // vec_unpackh sign-extends... + (vector signed short)(vec_mergeh(src_vF, (vector unsigned char)vzero)); + filter_v = vec_vsx_ld(i << 4, filter); + // the 4 above is 3 (filterSize == 8) + 1 (sizeof(short) == 2) + + val_v = vec_msums(src_v, filter_v, (vector signed int)vzero); + val_s = vec_sums(val_v, vzero); + vec_vsx_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 = vec_vsx_ld(srcPos, src); + + vector signed short src_vA = // vec_unpackh sign-extends... + (vector signed short)(vec_mergeh(src_vF, (vector unsigned char)vzero)); + vector signed short src_vB = // vec_unpackh sign-extends... + (vector signed short)(vec_mergel(src_vF, (vector unsigned char)vzero)); + + vector signed short filter_v0 = vec_vsx_ld(i << 5, filter); + vector signed short filter_v1 = vec_vsx_ld((i << 5) + 16, filter); + // the 5 above are 4 (filterSize == 16) + 1 (sizeof(short) == 2) + + 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_vsx_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; + register int srcPos = filterPos[i]; + + vector signed int val_s, val_v = (vector signed int)vzero; + + for (j = 0; j < filterSize - 15; j +=16) { + vector unsigned char src_vF = vec_vsx_ld(srcPos + j, src); + + vector signed short src_vA = // vec_unpackh sign-extends... + (vector signed short)(vec_mergeh(src_vF, (vector unsigned char)vzero)); + vector signed short src_vB = // vec_unpackh sign-extends... + (vector signed short)(vec_mergel(src_vF, (vector unsigned char)vzero)); + vector signed short filter_v0 = vec_vsx_ld((i * 2 * filterSize) + (j * 2), filter); + vector signed short filter_v1 = vec_vsx_ld((i * 2 * filterSize) + (j * 2) + 16, filter); + + vector signed int val_acc = vec_msums(src_vA, filter_v0, val_v); + val_v = vec_msums(src_vB, filter_v1, val_acc); + } + + if (j < filterSize - 7) { + // loading src_v0 is useless, it's already done above + vector unsigned char src_vF = vec_vsx_ld(srcPos + j, src); + vector signed short src_v, filter_v; + + src_v = // vec_unpackh sign-extends... + (vector signed short)(vec_mergeh(src_vF, (vector unsigned char)vzero)); + filter_v = vec_vsx_ld((i * 2 * filterSize) + (j * 2), filter); + val_v = vec_msums(src_v, filter_v, val_v); + } + val_s = vec_sums(val_v, vzero); + vec_vsx_st(val_s, 0, tempo); + dst[i] = FFMIN(tempo[3] >> 7, (1 << 15) - 1); + } + } + +#else /* else of #if !HAVE_BIGENDIAN */ register int i; DECLARE_ALIGNED(16, int, tempo)[4]; @@ -284,6 +452,7 @@ static void hScale_altivec_real(SwsContext *c, int16_t *dst, int dstW, dst[i] = FFMIN(tempo[3] >> 7, (1 << 15) - 1); } } +#endif /* end of #if !HAVE_BIGENDIAN */ } #endif /* HAVE_ALTIVEC */ -- 1.9.1