From 0e5d31b16b39f0b4d2e3307714af592dd65cf123 Mon Sep 17 00:00:00 2001 From: "Ronald S. Bultje" Date: Thu, 2 Jun 2011 20:17:20 -0700 Subject: swscale: split out unscaled altivec YUV converters in their own file. --- libswscale/Makefile | 3 +- libswscale/ppc/swscale_altivec_template.c | 155 ------------------------ libswscale/ppc/yuv2yuv_altivec.c | 191 ++++++++++++++++++++++++++++++ libswscale/swscale.c | 20 +--- libswscale/swscale_internal.h | 2 + 5 files changed, 200 insertions(+), 171 deletions(-) create mode 100644 libswscale/ppc/yuv2yuv_altivec.c diff --git a/libswscale/Makefile b/libswscale/Makefile index 7f8f721e45..8ecce99195 100644 --- a/libswscale/Makefile +++ b/libswscale/Makefile @@ -11,7 +11,8 @@ OBJS-$(ARCH_BFIN) += bfin/internal_bfin.o \ bfin/swscale_bfin.o \ bfin/yuv2rgb_bfin.o OBJS-$(CONFIG_MLIB) += mlib/yuv2rgb_mlib.o -OBJS-$(HAVE_ALTIVEC) += ppc/yuv2rgb_altivec.o +OBJS-$(HAVE_ALTIVEC) += ppc/yuv2rgb_altivec.o \ + ppc/yuv2yuv_altivec.o OBJS-$(HAVE_MMX) += x86/rgb2rgb.o \ x86/yuv2rgb_mmx.o OBJS-$(HAVE_VIS) += sparc/yuv2rgb_vis.o diff --git a/libswscale/ppc/swscale_altivec_template.c b/libswscale/ppc/swscale_altivec_template.c index 21e3b4eafb..5a821d0abf 100644 --- a/libswscale/ppc/swscale_altivec_template.c +++ b/libswscale/ppc/swscale_altivec_template.c @@ -390,158 +390,3 @@ static inline void hScale_altivec_real(int16_t *dst, int dstW, } } } - -static inline int yv12toyuy2_unscaled_altivec(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY, - int srcSliceH, uint8_t* dstParam[], int dstStride_a[]) -{ - uint8_t *dst=dstParam[0] + dstStride_a[0]*srcSliceY; - // yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]); - const uint8_t *ysrc = src[0]; - const uint8_t *usrc = src[1]; - const uint8_t *vsrc = src[2]; - const int width = c->srcW; - const int height = srcSliceH; - const int lumStride = srcStride[0]; - const int chromStride = srcStride[1]; - const int dstStride = dstStride_a[0]; - const vector unsigned char yperm = vec_lvsl(0, ysrc); - const int vertLumPerChroma = 2; - register unsigned int y; - - if (width&15) { - yv12toyuy2(ysrc, usrc, vsrc, dst, c->srcW, srcSliceH, lumStride, chromStride, dstStride); - return srcSliceH; - } - - /* This code assumes: - - 1) dst is 16 bytes-aligned - 2) dstStride is a multiple of 16 - 3) width is a multiple of 16 - 4) lum & chrom stride are multiples of 8 - */ - - for (y=0; y> 1; - vector unsigned char v_yA = vec_ld(i, ysrc); - vector unsigned char v_yB = vec_ld(i + 16, ysrc); - vector unsigned char v_yC = vec_ld(i + 32, ysrc); - vector unsigned char v_y1 = vec_perm(v_yA, v_yB, yperm); - vector unsigned char v_y2 = vec_perm(v_yB, v_yC, yperm); - vector unsigned char v_uA = vec_ld(j, usrc); - vector unsigned char v_uB = vec_ld(j + 16, usrc); - vector unsigned char v_u = vec_perm(v_uA, v_uB, vec_lvsl(j, usrc)); - vector unsigned char v_vA = vec_ld(j, vsrc); - vector unsigned char v_vB = vec_ld(j + 16, vsrc); - vector unsigned char v_v = vec_perm(v_vA, v_vB, vec_lvsl(j, vsrc)); - vector unsigned char v_uv_a = vec_mergeh(v_u, v_v); - vector unsigned char v_uv_b = vec_mergel(v_u, v_v); - vector unsigned char v_yuy2_0 = vec_mergeh(v_y1, v_uv_a); - vector unsigned char v_yuy2_1 = vec_mergel(v_y1, v_uv_a); - vector unsigned char v_yuy2_2 = vec_mergeh(v_y2, v_uv_b); - vector unsigned char v_yuy2_3 = vec_mergel(v_y2, v_uv_b); - vec_st(v_yuy2_0, (i << 1), dst); - vec_st(v_yuy2_1, (i << 1) + 16, dst); - vec_st(v_yuy2_2, (i << 1) + 32, dst); - vec_st(v_yuy2_3, (i << 1) + 48, dst); - } - if (i < width) { - const unsigned int j = i >> 1; - vector unsigned char v_y1 = vec_ld(i, ysrc); - vector unsigned char v_u = vec_ld(j, usrc); - vector unsigned char v_v = vec_ld(j, vsrc); - vector unsigned char v_uv_a = vec_mergeh(v_u, v_v); - vector unsigned char v_yuy2_0 = vec_mergeh(v_y1, v_uv_a); - vector unsigned char v_yuy2_1 = vec_mergel(v_y1, v_uv_a); - vec_st(v_yuy2_0, (i << 1), dst); - vec_st(v_yuy2_1, (i << 1) + 16, dst); - } - if ((y&(vertLumPerChroma-1)) == vertLumPerChroma-1) { - usrc += chromStride; - vsrc += chromStride; - } - ysrc += lumStride; - dst += dstStride; - } - - return srcSliceH; -} - -static inline int yv12touyvy_unscaled_altivec(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY, - int srcSliceH, uint8_t* dstParam[], int dstStride_a[]) -{ - uint8_t *dst=dstParam[0] + dstStride_a[0]*srcSliceY; - // yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]); - const uint8_t *ysrc = src[0]; - const uint8_t *usrc = src[1]; - const uint8_t *vsrc = src[2]; - const int width = c->srcW; - const int height = srcSliceH; - const int lumStride = srcStride[0]; - const int chromStride = srcStride[1]; - const int dstStride = dstStride_a[0]; - const int vertLumPerChroma = 2; - const vector unsigned char yperm = vec_lvsl(0, ysrc); - register unsigned int y; - - if (width&15) { - yv12touyvy(ysrc, usrc, vsrc, dst, c->srcW, srcSliceH, lumStride, chromStride, dstStride); - return srcSliceH; - } - - /* This code assumes: - - 1) dst is 16 bytes-aligned - 2) dstStride is a multiple of 16 - 3) width is a multiple of 16 - 4) lum & chrom stride are multiples of 8 - */ - - for (y=0; y> 1; - vector unsigned char v_yA = vec_ld(i, ysrc); - vector unsigned char v_yB = vec_ld(i + 16, ysrc); - vector unsigned char v_yC = vec_ld(i + 32, ysrc); - vector unsigned char v_y1 = vec_perm(v_yA, v_yB, yperm); - vector unsigned char v_y2 = vec_perm(v_yB, v_yC, yperm); - vector unsigned char v_uA = vec_ld(j, usrc); - vector unsigned char v_uB = vec_ld(j + 16, usrc); - vector unsigned char v_u = vec_perm(v_uA, v_uB, vec_lvsl(j, usrc)); - vector unsigned char v_vA = vec_ld(j, vsrc); - vector unsigned char v_vB = vec_ld(j + 16, vsrc); - vector unsigned char v_v = vec_perm(v_vA, v_vB, vec_lvsl(j, vsrc)); - vector unsigned char v_uv_a = vec_mergeh(v_u, v_v); - vector unsigned char v_uv_b = vec_mergel(v_u, v_v); - vector unsigned char v_uyvy_0 = vec_mergeh(v_uv_a, v_y1); - vector unsigned char v_uyvy_1 = vec_mergel(v_uv_a, v_y1); - vector unsigned char v_uyvy_2 = vec_mergeh(v_uv_b, v_y2); - vector unsigned char v_uyvy_3 = vec_mergel(v_uv_b, v_y2); - vec_st(v_uyvy_0, (i << 1), dst); - vec_st(v_uyvy_1, (i << 1) + 16, dst); - vec_st(v_uyvy_2, (i << 1) + 32, dst); - vec_st(v_uyvy_3, (i << 1) + 48, dst); - } - if (i < width) { - const unsigned int j = i >> 1; - vector unsigned char v_y1 = vec_ld(i, ysrc); - vector unsigned char v_u = vec_ld(j, usrc); - vector unsigned char v_v = vec_ld(j, vsrc); - vector unsigned char v_uv_a = vec_mergeh(v_u, v_v); - vector unsigned char v_uyvy_0 = vec_mergeh(v_uv_a, v_y1); - vector unsigned char v_uyvy_1 = vec_mergel(v_uv_a, v_y1); - vec_st(v_uyvy_0, (i << 1), dst); - vec_st(v_uyvy_1, (i << 1) + 16, dst); - } - if ((y&(vertLumPerChroma-1)) == vertLumPerChroma-1) { - usrc += chromStride; - vsrc += chromStride; - } - ysrc += lumStride; - dst += dstStride; - } - return srcSliceH; -} diff --git a/libswscale/ppc/yuv2yuv_altivec.c b/libswscale/ppc/yuv2yuv_altivec.c new file mode 100644 index 0000000000..4cd02ffe1d --- /dev/null +++ b/libswscale/ppc/yuv2yuv_altivec.c @@ -0,0 +1,191 @@ +/* + * AltiVec-enhanced yuv-to-yuv convertion routines. + * + * Copyright (C) 2004 Romain Dolbeau + * based on the equivalent C code in swscale.c + * + * 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 2.1 of the License, or (at your option) any later version. + * + * 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 Libav; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include +#include "config.h" +#include "libswscale/swscale.h" +#include "libswscale/swscale_internal.h" +#include "libavutil/cpu.h" + +static int yv12toyuy2_unscaled_altivec(SwsContext *c, const uint8_t* src[], + int srcStride[], int srcSliceY, + int srcSliceH, uint8_t* dstParam[], + int dstStride_a[]) +{ + uint8_t *dst=dstParam[0] + dstStride_a[0]*srcSliceY; + // yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]); + const uint8_t *ysrc = src[0]; + const uint8_t *usrc = src[1]; + const uint8_t *vsrc = src[2]; + const int width = c->srcW; + const int height = srcSliceH; + const int lumStride = srcStride[0]; + const int chromStride = srcStride[1]; + const int dstStride = dstStride_a[0]; + const vector unsigned char yperm = vec_lvsl(0, ysrc); + const int vertLumPerChroma = 2; + register unsigned int y; + + /* This code assumes: + + 1) dst is 16 bytes-aligned + 2) dstStride is a multiple of 16 + 3) width is a multiple of 16 + 4) lum & chrom stride are multiples of 8 + */ + + for (y=0; y> 1; + vector unsigned char v_yA = vec_ld(i, ysrc); + vector unsigned char v_yB = vec_ld(i + 16, ysrc); + vector unsigned char v_yC = vec_ld(i + 32, ysrc); + vector unsigned char v_y1 = vec_perm(v_yA, v_yB, yperm); + vector unsigned char v_y2 = vec_perm(v_yB, v_yC, yperm); + vector unsigned char v_uA = vec_ld(j, usrc); + vector unsigned char v_uB = vec_ld(j + 16, usrc); + vector unsigned char v_u = vec_perm(v_uA, v_uB, vec_lvsl(j, usrc)); + vector unsigned char v_vA = vec_ld(j, vsrc); + vector unsigned char v_vB = vec_ld(j + 16, vsrc); + vector unsigned char v_v = vec_perm(v_vA, v_vB, vec_lvsl(j, vsrc)); + vector unsigned char v_uv_a = vec_mergeh(v_u, v_v); + vector unsigned char v_uv_b = vec_mergel(v_u, v_v); + vector unsigned char v_yuy2_0 = vec_mergeh(v_y1, v_uv_a); + vector unsigned char v_yuy2_1 = vec_mergel(v_y1, v_uv_a); + vector unsigned char v_yuy2_2 = vec_mergeh(v_y2, v_uv_b); + vector unsigned char v_yuy2_3 = vec_mergel(v_y2, v_uv_b); + vec_st(v_yuy2_0, (i << 1), dst); + vec_st(v_yuy2_1, (i << 1) + 16, dst); + vec_st(v_yuy2_2, (i << 1) + 32, dst); + vec_st(v_yuy2_3, (i << 1) + 48, dst); + } + if (i < width) { + const unsigned int j = i >> 1; + vector unsigned char v_y1 = vec_ld(i, ysrc); + vector unsigned char v_u = vec_ld(j, usrc); + vector unsigned char v_v = vec_ld(j, vsrc); + vector unsigned char v_uv_a = vec_mergeh(v_u, v_v); + vector unsigned char v_yuy2_0 = vec_mergeh(v_y1, v_uv_a); + vector unsigned char v_yuy2_1 = vec_mergel(v_y1, v_uv_a); + vec_st(v_yuy2_0, (i << 1), dst); + vec_st(v_yuy2_1, (i << 1) + 16, dst); + } + if ((y&(vertLumPerChroma-1)) == vertLumPerChroma-1) { + usrc += chromStride; + vsrc += chromStride; + } + ysrc += lumStride; + dst += dstStride; + } + + return srcSliceH; +} + +static int yv12touyvy_unscaled_altivec(SwsContext *c, const uint8_t* src[], + int srcStride[], int srcSliceY, + int srcSliceH, uint8_t* dstParam[], + int dstStride_a[]) +{ + uint8_t *dst=dstParam[0] + dstStride_a[0]*srcSliceY; + // yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]); + const uint8_t *ysrc = src[0]; + const uint8_t *usrc = src[1]; + const uint8_t *vsrc = src[2]; + const int width = c->srcW; + const int height = srcSliceH; + const int lumStride = srcStride[0]; + const int chromStride = srcStride[1]; + const int dstStride = dstStride_a[0]; + const int vertLumPerChroma = 2; + const vector unsigned char yperm = vec_lvsl(0, ysrc); + register unsigned int y; + + /* This code assumes: + + 1) dst is 16 bytes-aligned + 2) dstStride is a multiple of 16 + 3) width is a multiple of 16 + 4) lum & chrom stride are multiples of 8 + */ + + for (y=0; y> 1; + vector unsigned char v_yA = vec_ld(i, ysrc); + vector unsigned char v_yB = vec_ld(i + 16, ysrc); + vector unsigned char v_yC = vec_ld(i + 32, ysrc); + vector unsigned char v_y1 = vec_perm(v_yA, v_yB, yperm); + vector unsigned char v_y2 = vec_perm(v_yB, v_yC, yperm); + vector unsigned char v_uA = vec_ld(j, usrc); + vector unsigned char v_uB = vec_ld(j + 16, usrc); + vector unsigned char v_u = vec_perm(v_uA, v_uB, vec_lvsl(j, usrc)); + vector unsigned char v_vA = vec_ld(j, vsrc); + vector unsigned char v_vB = vec_ld(j + 16, vsrc); + vector unsigned char v_v = vec_perm(v_vA, v_vB, vec_lvsl(j, vsrc)); + vector unsigned char v_uv_a = vec_mergeh(v_u, v_v); + vector unsigned char v_uv_b = vec_mergel(v_u, v_v); + vector unsigned char v_uyvy_0 = vec_mergeh(v_uv_a, v_y1); + vector unsigned char v_uyvy_1 = vec_mergel(v_uv_a, v_y1); + vector unsigned char v_uyvy_2 = vec_mergeh(v_uv_b, v_y2); + vector unsigned char v_uyvy_3 = vec_mergel(v_uv_b, v_y2); + vec_st(v_uyvy_0, (i << 1), dst); + vec_st(v_uyvy_1, (i << 1) + 16, dst); + vec_st(v_uyvy_2, (i << 1) + 32, dst); + vec_st(v_uyvy_3, (i << 1) + 48, dst); + } + if (i < width) { + const unsigned int j = i >> 1; + vector unsigned char v_y1 = vec_ld(i, ysrc); + vector unsigned char v_u = vec_ld(j, usrc); + vector unsigned char v_v = vec_ld(j, vsrc); + vector unsigned char v_uv_a = vec_mergeh(v_u, v_v); + vector unsigned char v_uyvy_0 = vec_mergeh(v_uv_a, v_y1); + vector unsigned char v_uyvy_1 = vec_mergel(v_uv_a, v_y1); + vec_st(v_uyvy_0, (i << 1), dst); + vec_st(v_uyvy_1, (i << 1) + 16, dst); + } + if ((y&(vertLumPerChroma-1)) == vertLumPerChroma-1) { + usrc += chromStride; + vsrc += chromStride; + } + ysrc += lumStride; + dst += dstStride; + } + return srcSliceH; +} + +void ff_swscale_get_unscaled_altivec(SwsContext *c) +{ + if ((av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC) && !(c->srcW & 15) && + !(c->flags & SWS_BITEXACT) && c->srcFormat == PIX_FMT_YUV420P) { + enum PixelFormat dstFormat = c->dstFormat; + + // unscaled YV12 -> packed YUV, we want speed + if (dstFormat == PIX_FMT_YUYV422) + c->swScale= yv12toyuy2_unscaled_altivec; + else if (dstFormat == PIX_FMT_UYVY422) + c->swScale= yv12touyvy_unscaled_altivec; + } +} diff --git a/libswscale/swscale.c b/libswscale/swscale.c index d6bdfdd3cc..5327948d3c 100644 --- a/libswscale/swscale.c +++ b/libswscale/swscale.c @@ -1857,18 +1857,6 @@ void ff_get_unscaled_swscale(SwsContext *c) if(srcFormat == PIX_FMT_UYVY422 && dstFormat == PIX_FMT_YUV422P) c->swScale= uyvyToYuv422Wrapper; -#if HAVE_ALTIVEC - if ((av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC) && - !(c->flags & SWS_BITEXACT) && - srcFormat == PIX_FMT_YUV420P) { - // unscaled YV12 -> packed YUV, we want speed - if (dstFormat == PIX_FMT_YUYV422) - c->swScale= yv12toyuy2_unscaled_altivec; - else if (dstFormat == PIX_FMT_UYVY422) - c->swScale= yv12touyvy_unscaled_altivec; - } -#endif - /* simple copy */ if ( srcFormat == dstFormat || (srcFormat == PIX_FMT_YUVA420P && dstFormat == PIX_FMT_YUV420P) @@ -1887,9 +1875,11 @@ void ff_get_unscaled_swscale(SwsContext *c) else /* Planar YUV or gray */ c->swScale= planarCopyWrapper; } -#if ARCH_BFIN - ff_bfin_get_unscaled_swscale (c); -#endif + + if (ARCH_BFIN) + ff_bfin_get_unscaled_swscale(c); + if (HAVE_ALTIVEC) + ff_swscale_get_unscaled_altivec(c); } static void reset_ptr(const uint8_t* src[], int format) diff --git a/libswscale/swscale_internal.h b/libswscale/swscale_internal.h index c17d55023d..09d122418f 100644 --- a/libswscale/swscale_internal.h +++ b/libswscale/swscale_internal.h @@ -479,6 +479,8 @@ extern const AVClass sws_context_class; */ void ff_get_unscaled_swscale(SwsContext *c); +void ff_swscale_get_unscaled_altivec(SwsContext *c); + /** * Returns function pointer to fastest main scaler path function depending * on architecture and available optimizations. -- cgit v1.2.3