From db2c5213377ecde3e24d2f4a27cb790c158a7ef9 Mon Sep 17 00:00:00 2001 From: Rong Yan Date: Thu, 27 Nov 2014 05:51:01 +0000 Subject: [PATCH 3/5] libavcodec/ppc/h264chroma_template.c :fix PREFIX_h264_chroma_mc8_altivec() PREFIX_no_rnd_vc1_chroma_mc8_altivec() CHROMA_MC8_ALTIVEC_CORE_SIMPLE() CHROMA_MC8_ALTIVEC_CORE() add marcos GET_VSRC1() GET_VSRC() for POWER LE --- libavcodec/ppc/h264chroma_template.c | 265 ++++++++++++++--------------------- 1 file changed, 109 insertions(+), 156 deletions(-) diff --git a/libavcodec/ppc/h264chroma_template.c b/libavcodec/ppc/h264chroma_template.c index 7436e11..590aeec 100644 --- a/libavcodec/ppc/h264chroma_template.c +++ b/libavcodec/ppc/h264chroma_template.c @@ -19,57 +19,97 @@ */ #include "libavutil/mem.h" +#include "libavutil/ppc/types_altivec.h" +#include "libavutil/ppc/util_altivec.h" /* this code assume that stride % 16 == 0 */ -#define CHROMA_MC8_ALTIVEC_CORE(BIAS1, BIAS2) \ - vsrc2ssH = (vec_s16)vec_mergeh(zero_u8v,(vec_u8)vsrc2uc);\ - vsrc3ssH = (vec_s16)vec_mergeh(zero_u8v,(vec_u8)vsrc3uc);\ -\ +#define CHROMA_MC8_ALTIVEC_CORE(BIAS1, BIAS2) {\ + vsrc2ssH = (vec_s16)VEC_MERGEH(zero_u8v,(vec_u8)vsrc2uc);\ + vsrc3ssH = (vec_s16)VEC_MERGEH(zero_u8v,(vec_u8)vsrc3uc);\ + \ psum = vec_mladd(vA, vsrc0ssH, BIAS1);\ psum = vec_mladd(vB, vsrc1ssH, psum);\ psum = vec_mladd(vC, vsrc2ssH, psum);\ psum = vec_mladd(vD, vsrc3ssH, psum);\ psum = BIAS2(psum);\ psum = vec_sr(psum, v6us);\ -\ + \ vdst = vec_ld(0, dst);\ ppsum = (vec_u8)vec_pack(psum, psum);\ vfdst = vec_perm(vdst, ppsum, fperm);\ -\ + \ OP_U8_ALTIVEC(fsum, vfdst, vdst);\ -\ + \ vec_st(fsum, 0, dst);\ -\ + \ vsrc0ssH = vsrc2ssH;\ vsrc1ssH = vsrc3ssH;\ -\ + \ dst += stride;\ - src += stride; - -#define CHROMA_MC8_ALTIVEC_CORE_SIMPLE \ -\ - vsrc0ssH = (vec_s16)vec_mergeh(zero_u8v,(vec_u8)vsrc0uc);\ - vsrc1ssH = (vec_s16)vec_mergeh(zero_u8v,(vec_u8)vsrc1uc);\ -\ + src += stride;\ +} +#define CHROMA_MC8_ALTIVEC_CORE_SIMPLE {\ + \ + vsrc0ssH = (vec_s16)VEC_MERGEH(zero_u8v,(vec_u8)vsrc0uc);\ + vsrc1ssH = (vec_s16)VEC_MERGEH(zero_u8v,(vec_u8)vsrc1uc);\ + \ psum = vec_mladd(vA, vsrc0ssH, v32ss);\ psum = vec_mladd(vE, vsrc1ssH, psum);\ psum = vec_sr(psum, v6us);\ -\ + \ vdst = vec_ld(0, dst);\ ppsum = (vec_u8)vec_pack(psum, psum);\ vfdst = vec_perm(vdst, ppsum, fperm);\ -\ + \ OP_U8_ALTIVEC(fsum, vfdst, vdst);\ -\ + \ vec_st(fsum, 0, dst);\ -\ + \ dst += stride;\ - src += stride; + src += stride;\ +} #define noop(a) a #define add28(a) vec_add(v28ss, a) +#if HAVE_BIGENDIAN +#define GET_VSRC1(vs0, off, b, perm0, s){\ + vec_u8 vsrcCuc, vsrcDuc;\ + vsrcCuc = vec_ld(off, s);\ + if (loadSecond){\ + vsrcDuc = vec_ld(off + b, s);\ + } else\ + vsrcDuc = vsrcCuc;\ + \ + vs0 = vec_perm(vsrcCuc, vsrcDuc, perm0);\ +} +#define GET_VSRC(vs0, vs1, off, b, perm0, perm1, s){\ + vec_u8 vsrcCuc, vsrcDuc;\ + vsrcCuc = vec_ld(off, s);\ + if (loadSecond){\ + vsrcDuc = vec_ld(off + b, s);\ + } else\ + vsrcDuc = vsrcCuc;\ + \ + vs0 = vec_perm(vsrcCuc, vsrcDuc, perm0);\ + if (reallyBadAlign){\ + vs1 = vsrcDuc;\ + } else\ + vs1 = vec_perm(vsrcCuc, vsrcDuc, perm1);\ +} +#else /* HAVE_BIGENDIAN */ + +#define GET_VSRC1(vs0, off, b, perm0, s){\ + vs0 = vec_vsx_ld(off, s);\ +} +#define GET_VSRC(vs0, vs1, off, b, perm0, perm1, s){\ + vs0 = vec_vsx_ld(off, s);\ + vs1 = vec_vsx_ld(off + 1, s);\ +} +#endif /* HAVE_BIGENDIAN */ + + #ifdef PREFIX_h264_chroma_mc8_altivec static void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride, int h, int x, int y) { @@ -78,25 +118,29 @@ static void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, (( x) * (8 - y)), ((8 - x) * ( y)), (( x) * ( y))}; - register int i; vec_u8 fperm; - const vec_s32 vABCD = vec_ld(0, ABCD); - const vec_s16 vA = vec_splat((vec_s16)vABCD, 1); - const vec_s16 vB = vec_splat((vec_s16)vABCD, 3); - const vec_s16 vC = vec_splat((vec_s16)vABCD, 5); - const vec_s16 vD = vec_splat((vec_s16)vABCD, 7); LOAD_ZERO; + const vec_s32 vABCD = vec_ld(0, ABCD); + const vec_s16 vA = VEC_SPLAT16(vABCD, 1); + const vec_s16 vB = VEC_SPLAT16(vABCD, 3); + const vec_s16 vC = VEC_SPLAT16(vABCD, 5); + const vec_s16 vD = VEC_SPLAT16(vABCD, 7); const vec_s16 v32ss = vec_sl(vec_splat_s16(1),vec_splat_u16(5)); const vec_u16 v6us = vec_splat_u16(6); - register int loadSecond = (((unsigned long)src) % 16) <= 7 ? 0 : 1; - register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0; - - vec_u8 vsrcAuc, av_uninit(vsrcBuc), vsrcperm0, vsrcperm1; + vec_u8 vsrcperm0, vsrcperm1; vec_u8 vsrc0uc, vsrc1uc; vec_s16 vsrc0ssH, vsrc1ssH; - vec_u8 vsrcCuc, vsrc2uc, vsrc3uc; + vec_u8 vsrc2uc, vsrc3uc; vec_s16 vsrc2ssH, vsrc3ssH, psum; vec_u8 vdst, ppsum, vfdst, fsum; + register int i; +#if HAVE_BIGENDIAN + register int loadSecond = (((unsigned long)src) % 16) <= 7 ? 0 : 1; + register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0; + + vsrcperm0 = vec_lvsl(0, src); + vsrcperm1 = vec_lvsl(1, src); +#endif if (((unsigned long)dst) % 16 == 0) { fperm = (vec_u8){0x10, 0x11, 0x12, 0x13, @@ -110,89 +154,28 @@ static void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, 0x1C, 0x1D, 0x1E, 0x1F}; } - vsrcAuc = vec_ld(0, src); - - if (loadSecond) - vsrcBuc = vec_ld(16, src); - vsrcperm0 = vec_lvsl(0, src); - vsrcperm1 = vec_lvsl(1, src); + GET_VSRC(vsrc0uc, vsrc1uc, 0, 16, vsrcperm0, vsrcperm1, src); - vsrc0uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm0); - if (reallyBadAlign) - vsrc1uc = vsrcBuc; - else - vsrc1uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm1); - - vsrc0ssH = (vec_s16)vec_mergeh(zero_u8v,(vec_u8)vsrc0uc); - vsrc1ssH = (vec_s16)vec_mergeh(zero_u8v,(vec_u8)vsrc1uc); + vsrc0ssH = (vec_s16)VEC_MERGEH(zero_u8v,(vec_u8)vsrc0uc); + vsrc1ssH = (vec_s16)VEC_MERGEH(zero_u8v,(vec_u8)vsrc1uc); if (ABCD[3]) { - if (!loadSecond) {// -> !reallyBadAlign - for (i = 0 ; i < h ; i++) { - vsrcCuc = vec_ld(stride + 0, src); - vsrc2uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0); - vsrc3uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm1); - - CHROMA_MC8_ALTIVEC_CORE(v32ss, noop) - } - } else { - vec_u8 vsrcDuc; - for (i = 0 ; i < h ; i++) { - vsrcCuc = vec_ld(stride + 0, src); - vsrcDuc = vec_ld(stride + 16, src); - vsrc2uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0); - if (reallyBadAlign) - vsrc3uc = vsrcDuc; - else - vsrc3uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm1); - - CHROMA_MC8_ALTIVEC_CORE(v32ss, noop) - } + for (i = 0 ; i < h ; i++) { + GET_VSRC(vsrc2uc, vsrc3uc, stride, 16, vsrcperm0, vsrcperm1, src); + CHROMA_MC8_ALTIVEC_CORE(v32ss, noop); } } else { const vec_s16 vE = vec_add(vB, vC); if (ABCD[2]) { // x == 0 B == 0 - if (!loadSecond) {// -> !reallyBadAlign - for (i = 0 ; i < h ; i++) { - vsrcCuc = vec_ld(stride + 0, src); - vsrc1uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0); - CHROMA_MC8_ALTIVEC_CORE_SIMPLE - - vsrc0uc = vsrc1uc; - } - } else { - vec_u8 vsrcDuc; - for (i = 0 ; i < h ; i++) { - vsrcCuc = vec_ld(stride + 0, src); - vsrcDuc = vec_ld(stride + 15, src); - vsrc1uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0); - CHROMA_MC8_ALTIVEC_CORE_SIMPLE - - vsrc0uc = vsrc1uc; - } + for (i = 0 ; i < h ; i++) { + GET_VSRC1(vsrc1uc, stride, 15, vsrcperm0, src); + CHROMA_MC8_ALTIVEC_CORE_SIMPLE; + vsrc0uc = vsrc1uc; } } else { // y == 0 C == 0 - if (!loadSecond) {// -> !reallyBadAlign - for (i = 0 ; i < h ; i++) { - vsrcCuc = vec_ld(0, src); - vsrc0uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0); - vsrc1uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm1); - - CHROMA_MC8_ALTIVEC_CORE_SIMPLE - } - } else { - vec_u8 vsrcDuc; - for (i = 0 ; i < h ; i++) { - vsrcCuc = vec_ld(0, src); - vsrcDuc = vec_ld(15, src); - vsrc0uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0); - if (reallyBadAlign) - vsrc1uc = vsrcDuc; - else - vsrc1uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm1); - - CHROMA_MC8_ALTIVEC_CORE_SIMPLE - } + for (i = 0 ; i < h ; i++) { + GET_VSRC(vsrc0uc, vsrc1uc, 0, 15, vsrcperm0, vsrcperm1, src); + CHROMA_MC8_ALTIVEC_CORE_SIMPLE; } } } @@ -207,25 +190,28 @@ static void PREFIX_no_rnd_vc1_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, i (( x) * (8 - y)), ((8 - x) * ( y)), (( x) * ( y))}; - register int i; vec_u8 fperm; - const vec_s32 vABCD = vec_ld(0, ABCD); - const vec_s16 vA = vec_splat((vec_s16)vABCD, 1); - const vec_s16 vB = vec_splat((vec_s16)vABCD, 3); - const vec_s16 vC = vec_splat((vec_s16)vABCD, 5); - const vec_s16 vD = vec_splat((vec_s16)vABCD, 7); LOAD_ZERO; + const vec_s32 vABCD = vec_ld(0, ABCD); + const vec_s16 vA = VEC_SPLAT16(vABCD, 1); + const vec_s16 vB = VEC_SPLAT16(vABCD, 3); + const vec_s16 vC = VEC_SPLAT16(vABCD, 5); + const vec_s16 vD = VEC_SPLAT16(vABCD, 7); const vec_s16 v28ss = vec_sub(vec_sl(vec_splat_s16(1),vec_splat_u16(5)),vec_splat_s16(4)); const vec_u16 v6us = vec_splat_u16(6); - register int loadSecond = (((unsigned long)src) % 16) <= 7 ? 0 : 1; - register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0; - - vec_u8 vsrcAuc, av_uninit(vsrcBuc), vsrcperm0, vsrcperm1; + vec_u8 vsrcperm0, vsrcperm1; vec_u8 vsrc0uc, vsrc1uc; vec_s16 vsrc0ssH, vsrc1ssH; - vec_u8 vsrcCuc, vsrc2uc, vsrc3uc; + vec_u8 vsrc2uc, vsrc3uc; vec_s16 vsrc2ssH, vsrc3ssH, psum; vec_u8 vdst, ppsum, vfdst, fsum; + register int i; +#if HAVE_BIGENDIAN + register int loadSecond = (((unsigned long)src) % 16) <= 7 ? 0 : 1; + register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0; + vsrcperm0 = vec_lvsl(0, src); + vsrcperm1 = vec_lvsl(1, src); +#endif if (((unsigned long)dst) % 16 == 0) { fperm = (vec_u8){0x10, 0x11, 0x12, 0x13, @@ -239,47 +225,14 @@ static void PREFIX_no_rnd_vc1_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, i 0x1C, 0x1D, 0x1E, 0x1F}; } - vsrcAuc = vec_ld(0, src); - - if (loadSecond) - vsrcBuc = vec_ld(16, src); - vsrcperm0 = vec_lvsl(0, src); - vsrcperm1 = vec_lvsl(1, src); - - vsrc0uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm0); - if (reallyBadAlign) - vsrc1uc = vsrcBuc; - else - vsrc1uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm1); - - vsrc0ssH = (vec_s16)vec_mergeh(zero_u8v, (vec_u8)vsrc0uc); - vsrc1ssH = (vec_s16)vec_mergeh(zero_u8v, (vec_u8)vsrc1uc); - - if (!loadSecond) {// -> !reallyBadAlign - for (i = 0 ; i < h ; i++) { - - - vsrcCuc = vec_ld(stride + 0, src); - - vsrc2uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0); - vsrc3uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm1); - - CHROMA_MC8_ALTIVEC_CORE(vec_splat_s16(0), add28) - } - } else { - vec_u8 vsrcDuc; - for (i = 0 ; i < h ; i++) { - vsrcCuc = vec_ld(stride + 0, src); - vsrcDuc = vec_ld(stride + 16, src); + GET_VSRC(vsrc0uc, vsrc1uc, 0, 16, vsrcperm0, vsrcperm1, src); - vsrc2uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0); - if (reallyBadAlign) - vsrc3uc = vsrcDuc; - else - vsrc3uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm1); + vsrc0ssH = (vec_s16)VEC_MERGEH(zero_u8v, (vec_u8)vsrc0uc); + vsrc1ssH = (vec_s16)VEC_MERGEH(zero_u8v, (vec_u8)vsrc1uc); - CHROMA_MC8_ALTIVEC_CORE(vec_splat_s16(0), add28) - } + for (i = 0 ; i < h ; i++) { + GET_VSRC(vsrc2uc, vsrc3uc, stride, 16, vsrcperm0, vsrcperm1, src); + CHROMA_MC8_ALTIVEC_CORE(vec_splat_s16(0), add28); } } #endif -- 1.9.1