From e28b4a137b061f8ce755e20e0de8fdd635eab214 Mon Sep 17 00:00:00 2001 From: Rong Yan Date: Tue, 4 Nov 2014 06:27:58 +0000 Subject: [PATCH 2/3] libavcodec/ppc/hpeldsp_altivec.c : fix ff_put_pixels16_altivec() put_no_rnd_pixels16_xy2_altivec() put_no_rnd_pixels8_xy2_altivec() avg_pixels8_altivec() avg_pixels8_xy2_altivec() put_pixels16_xy2_altivec() put_pixels8_xy2_altivec() ff_avg_pixels16_altivec() for POWER LE --- libavcodec/ppc/hpeldsp_altivec.c | 147 ++++++++++++++++++++++++++++++--------- 1 file changed, 113 insertions(+), 34 deletions(-) diff --git a/libavcodec/ppc/hpeldsp_altivec.c b/libavcodec/ppc/hpeldsp_altivec.c index 79c2af8..5167a14 100644 --- a/libavcodec/ppc/hpeldsp_altivec.c +++ b/libavcodec/ppc/hpeldsp_altivec.c @@ -36,46 +36,22 @@ #if HAVE_ALTIVEC /* next one assumes that ((line_size % 16) == 0) */ -#if HAVE_VSX -void ff_put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) -{ - register vector unsigned char pixelsv1; - register vector unsigned char pixelsv1B; - register vector unsigned char pixelsv1C; - register vector unsigned char pixelsv1D; - - int i; - register ptrdiff_t line_size_2 = line_size << 1; - register ptrdiff_t line_size_3 = line_size + line_size_2; - register ptrdiff_t line_size_4 = line_size << 2; - -// hand-unrolling the loop by 4 gains about 15% -// mininum execution time goes from 74 to 60 cycles -// it's faster than -funroll-loops, but using -// -funroll-loops w/ this is bad - 74 cycles again. -// all this is on a 7450, tuning for the 7450 - for (i = 0; i < h; i += 4) { - pixelsv1 = vec_vsx_ld( 0, pixels); - pixelsv1B = vec_vsx_ld(line_size, pixels); - pixelsv1C = vec_vsx_ld(line_size_2, pixels); - pixelsv1D = vec_vsx_ld(line_size_3, pixels); - vec_vsx_st(pixelsv1, 0, (unsigned char*)block); - vec_vsx_st(pixelsv1B, line_size, (unsigned char*)block); - vec_vsx_st(pixelsv1C, line_size_2, (unsigned char*)block); - vec_st(pixelsv1D, line_size_3, (unsigned char*)block); - pixels+=line_size_4; - block +=line_size_4; - } -} -#else void ff_put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) { +#if HAVE_BIGENDIAN register vector unsigned char pixelsv1, pixelsv2; register vector unsigned char pixelsv1B, pixelsv2B; register vector unsigned char pixelsv1C, pixelsv2C; register vector unsigned char pixelsv1D, pixelsv2D; register vector unsigned char perm = vec_lvsl(0, pixels); +#else /* HAVE_BIGENDIAN */ + register vector unsigned char pixelsv1; + register vector unsigned char pixelsv1B; + register vector unsigned char pixelsv1C; + register vector unsigned char pixelsv1D; +#endif /* HAVE_BIGENDIAN */ + int i; register ptrdiff_t line_size_2 = line_size << 1; register ptrdiff_t line_size_3 = line_size + line_size_2; @@ -87,6 +63,7 @@ void ff_put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t li // -funroll-loops w/ this is bad - 74 cycles again. // all this is on a 7450, tuning for the 7450 for (i = 0; i < h; i += 4) { +#if HAVE_BIGENDIAN pixelsv1 = vec_ld( 0, pixels); pixelsv2 = vec_ld(15, pixels); pixelsv1B = vec_ld(line_size, pixels); @@ -103,26 +80,42 @@ void ff_put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t li line_size_2, (unsigned char*)block); vec_st(vec_perm(pixelsv1D, pixelsv2D, perm), line_size_3, (unsigned char*)block); +#else /* HAVE_BIGENDIAN */ + pixelsv1 = vec_vsx_ld( 0, pixels); + pixelsv1B = vec_vsx_ld(line_size, pixels); + pixelsv1C = vec_vsx_ld(line_size_2, pixels); + pixelsv1D = vec_vsx_ld(line_size_3, pixels); + vec_vsx_st(pixelsv1, 0, (unsigned char*)block); + vec_vsx_st(pixelsv1B, line_size, (unsigned char*)block); + vec_vsx_st(pixelsv1C, line_size_2, (unsigned char*)block); + vec_st(pixelsv1D, line_size_3, (unsigned char*)block); +#endif /* HAVE_BIGENDIAN */ pixels+=line_size_4; block +=line_size_4; } } -#endif /* HAVE_VSX */ - /* next one assumes that ((line_size % 16) == 0) */ #define op_avg(a,b) a = ( ((a)|(b)) - ((((a)^(b))&0xFEFEFEFEUL)>>1) ) void ff_avg_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) { register vector unsigned char pixelsv1, pixelsv2, pixelsv, blockv; +#if HAVE_BIGENDIAN register vector unsigned char perm = vec_lvsl(0, pixels); +#endif /* HAVE_BIGENDIAN */ int i; for (i = 0; i < h; i++) { +#if HAVE_BIGENDIAN pixelsv1 = vec_ld( 0, pixels); pixelsv2 = vec_ld(16,pixels); +#endif /* HAVE_BIGENDIAN */ blockv = vec_ld(0, block); +#if HAVE_BIGENDIAN pixelsv = vec_perm(pixelsv1, pixelsv2, perm); +#else /* HAVE_BIGENDIAN */ + pixelsv = vec_vsx_ld( 0, pixels); +#endif /* HAVE_BIGENDIAN */ blockv = vec_avg(blockv,pixelsv); vec_st(blockv, 0, (unsigned char*)block); pixels+=line_size; @@ -142,9 +135,13 @@ static void avg_pixels8_altivec(uint8_t * block, const uint8_t * pixels, ptrdiff int rightside = ((unsigned long)block & 0x0000000F); blockv = vec_ld(0, block); +#if HAVE_BIGENDIAN pixelsv1 = vec_ld( 0, pixels); pixelsv2 = vec_ld(16, pixels); pixelsv = vec_perm(pixelsv1, pixelsv2, vec_lvsl(0, pixels)); +#else /* HAVE_BIGENDIAN */ + pixelsv = vec_vsx_ld( 0, pixels); +#endif /* HAVE_BIGENDIAN */ if (rightside) { pixelsv = vec_perm(blockv, pixelsv, vcprm(0,1,s0,s1)); @@ -171,6 +168,7 @@ static void put_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdi register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0); register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2); +#if HAVE_BIGENDIAN temp1 = vec_ld(0, pixels); temp2 = vec_ld(16, pixels); pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels)); @@ -181,6 +179,12 @@ static void put_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdi } pixelsv1 = vec_mergeh(vczero, pixelsv1); pixelsv2 = vec_mergeh(vczero, pixelsv2); +#else /* HAVE_BIGENDIAN */ + pixelsv1 = vec_vsx_ld(0, pixels); + pixelsv2 = vec_vsx_ld(1, pixels); + pixelsv1 = vec_mergeh(pixelsv1, vczero); + pixelsv2 = vec_mergeh(pixelsv2, vczero); +#endif /* HAVE_BIGENDIAN */ pixelssum1 = vec_add((vector unsigned short)pixelsv1, (vector unsigned short)pixelsv2); pixelssum1 = vec_add(pixelssum1, vctwo); @@ -189,6 +193,7 @@ static void put_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdi int rightside = ((unsigned long)block & 0x0000000F); blockv = vec_ld(0, block); +#if HAVE_BIGENDIAN temp1 = vec_ld(line_size, pixels); temp2 = vec_ld(line_size + 16, pixels); pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels)); @@ -200,6 +205,12 @@ static void put_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdi pixelsv1 = vec_mergeh(vczero, pixelsv1); pixelsv2 = vec_mergeh(vczero, pixelsv2); +#else /* HAVE_BIGENDIAN */ + pixelsv1 = vec_vsx_ld(line_size, pixels); + pixelsv2 = vec_vsx_ld(line_size+1, pixels); + pixelsv1 = vec_mergeh(pixelsv1, vczero); + pixelsv2 = vec_mergeh(pixelsv2, vczero); +#endif /* HAVE_BIGENDIAN */ pixelssum2 = vec_add((vector unsigned short)pixelsv1, (vector unsigned short)pixelsv2); temp3 = vec_add(pixelssum1, pixelssum2); @@ -231,6 +242,7 @@ static void put_no_rnd_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1); register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2); +#if HAVE_BIGENDIAN temp1 = vec_ld(0, pixels); temp2 = vec_ld(16, pixels); pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels)); @@ -241,6 +253,12 @@ static void put_no_rnd_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels } pixelsv1 = vec_mergeh(vczero, pixelsv1); pixelsv2 = vec_mergeh(vczero, pixelsv2); +#else /* HAVE_BIGENDIAN */ + pixelsv1 = vec_vsx_ld(0, pixels); + pixelsv2 = vec_vsx_ld(1, pixels); + pixelsv1 = vec_mergeh(pixelsv1, vczero); + pixelsv2 = vec_mergeh(pixelsv2, vczero); +#endif /* HAVE_BIGENDIAN */ pixelssum1 = vec_add((vector unsigned short)pixelsv1, (vector unsigned short)pixelsv2); pixelssum1 = vec_add(pixelssum1, vcone); @@ -249,6 +267,7 @@ static void put_no_rnd_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels int rightside = ((unsigned long)block & 0x0000000F); blockv = vec_ld(0, block); +#if HAVE_BIGENDIAN temp1 = vec_ld(line_size, pixels); temp2 = vec_ld(line_size + 16, pixels); pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels)); @@ -260,6 +279,12 @@ static void put_no_rnd_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels pixelsv1 = vec_mergeh(vczero, pixelsv1); pixelsv2 = vec_mergeh(vczero, pixelsv2); +#else /* HAVE_BIGENDIAN */ + pixelsv1 = vec_vsx_ld(line_size, pixels); + pixelsv2 = vec_vsx_ld(line_size+1, pixels); + pixelsv1 = vec_mergeh(pixelsv1, vczero); + pixelsv2 = vec_mergeh(pixelsv2, vczero); +#endif /* HAVE_BIGENDIAN */ pixelssum2 = vec_add((vector unsigned short)pixelsv1, (vector unsigned short)pixelsv2); temp3 = vec_add(pixelssum1, pixelssum2); @@ -291,6 +316,7 @@ static void put_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, pt register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0); register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2); +#if HAVE_BIGENDIAN temp1 = vec_ld(0, pixels); temp2 = vec_ld(16, pixels); pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels)); @@ -303,6 +329,14 @@ static void put_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, pt pixelsv4 = vec_mergel(vczero, pixelsv2); pixelsv1 = vec_mergeh(vczero, pixelsv1); pixelsv2 = vec_mergeh(vczero, pixelsv2); +#else /* HAVE_BIGENDIAN */ + pixelsv1 = vec_vsx_ld(0, pixels); + pixelsv2 = vec_vsx_ld(1, pixels); + pixelsv3 = vec_mergel(pixelsv1, vczero); + pixelsv4 = vec_mergel(pixelsv2, vczero); + pixelsv1 = vec_mergeh(pixelsv1, vczero); + pixelsv2 = vec_mergeh(pixelsv2, vczero); +#endif /* HAVE_BIGENDIAN */ pixelssum3 = vec_add((vector unsigned short)pixelsv3, (vector unsigned short)pixelsv4); pixelssum3 = vec_add(pixelssum3, vctwo); @@ -313,6 +347,7 @@ static void put_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, pt for (i = 0; i < h ; i++) { blockv = vec_ld(0, block); +#if HAVE_BIGENDIAN temp1 = vec_ld(line_size, pixels); temp2 = vec_ld(line_size + 16, pixels); pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels)); @@ -326,6 +361,14 @@ static void put_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, pt pixelsv4 = vec_mergel(vczero, pixelsv2); pixelsv1 = vec_mergeh(vczero, pixelsv1); pixelsv2 = vec_mergeh(vczero, pixelsv2); +#else /* HAVE_BIGENDIAN */ + pixelsv1 = vec_vsx_ld(line_size, pixels); + pixelsv2 = vec_vsx_ld(line_size+1, pixels); + pixelsv3 = vec_mergel(pixelsv1,vczero); + pixelsv4 = vec_mergel(pixelsv2, vczero); + pixelsv1 = vec_mergeh(pixelsv1, vczero); + pixelsv2 = vec_mergeh(pixelsv2, vczero); +#endif /* HAVE_BIGENDIAN */ pixelssum4 = vec_add((vector unsigned short)pixelsv3, (vector unsigned short)pixelsv4); @@ -360,6 +403,7 @@ static void put_no_rnd_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pix register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1); register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2); +#if HAVE_BIGENDIAN temp1 = vec_ld(0, pixels); temp2 = vec_ld(16, pixels); pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels)); @@ -372,6 +416,14 @@ static void put_no_rnd_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pix pixelsv4 = vec_mergel(vczero, pixelsv2); pixelsv1 = vec_mergeh(vczero, pixelsv1); pixelsv2 = vec_mergeh(vczero, pixelsv2); +#else /* HAVE_BIGENDIAN */ + pixelsv1 = vec_vsx_ld(0, pixels); + pixelsv2 = vec_vsx_ld(1, pixels); + pixelsv3 = vec_mergel(pixelsv1, vczero); + pixelsv4 = vec_mergel(pixelsv2, vczero); + pixelsv1 = vec_mergeh(pixelsv1, vczero); + pixelsv2 = vec_mergeh(pixelsv2, vczero); +#endif /* HAVE_BIGENDIAN */ pixelssum3 = vec_add((vector unsigned short)pixelsv3, (vector unsigned short)pixelsv4); pixelssum3 = vec_add(pixelssum3, vcone); @@ -380,6 +432,7 @@ static void put_no_rnd_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pix pixelssum1 = vec_add(pixelssum1, vcone); for (i = 0; i < h ; i++) { +#if HAVE_BIGENDIAN blockv = vec_ld(0, block); temp1 = vec_ld(line_size, pixels); @@ -395,6 +448,14 @@ static void put_no_rnd_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pix pixelsv4 = vec_mergel(vczero, pixelsv2); pixelsv1 = vec_mergeh(vczero, pixelsv1); pixelsv2 = vec_mergeh(vczero, pixelsv2); +#else /* HAVE_BIGENDIAN */ + pixelsv1 = vec_vsx_ld(line_size, pixels); + pixelsv2 = vec_vsx_ld(line_size+1, pixels); + pixelsv3 = vec_mergel(pixelsv1, vczero); + pixelsv4 = vec_mergel(pixelsv2, vczero); + pixelsv1 = vec_mergeh(pixelsv1, vczero); + pixelsv2 = vec_mergeh(pixelsv2, vczero); +#endif /* HAVE_BIGENDIAN */ pixelssum4 = vec_add((vector unsigned short)pixelsv3, (vector unsigned short)pixelsv4); @@ -410,7 +471,11 @@ static void put_no_rnd_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pix blockv = vec_packsu(temp3, temp4); +#if HAVE_BIGENDIAN vec_st(blockv, 0, block); +#else /* HAVE_BIGENDIAN */ + vec_vsx_st(blockv, 0, block); +#endif /* HAVE_BIGENDIAN */ block += line_size; pixels += line_size; @@ -430,6 +495,7 @@ static void avg_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdi register const vector unsigned short vctwo = (const vector unsigned short) vec_splat_u16(2); +#if HAVE_BIGENDIAN temp1 = vec_ld(0, pixels); temp2 = vec_ld(16, pixels); pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels)); @@ -440,6 +506,12 @@ static void avg_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdi } pixelsv1 = vec_mergeh(vczero, pixelsv1); pixelsv2 = vec_mergeh(vczero, pixelsv2); +#else /* HAVE_BIGENDIAN */ + pixelsv1 = vec_vsx_ld(0, pixels); + pixelsv2 = vec_vsx_ld(1, pixels); + pixelsv1 = vec_mergeh(pixelsv1, vczero); + pixelsv2 = vec_mergeh(pixelsv2, vczero); +#endif /* HAVE_BIGENDIAN */ pixelssum1 = vec_add((vector unsigned short)pixelsv1, (vector unsigned short)pixelsv2); pixelssum1 = vec_add(pixelssum1, vctwo); @@ -448,6 +520,7 @@ static void avg_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdi int rightside = ((unsigned long)block & 0x0000000F); blockv = vec_ld(0, block); +#if HAVE_BIGENDIAN temp1 = vec_ld(line_size, pixels); temp2 = vec_ld(line_size + 16, pixels); pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels)); @@ -459,6 +532,12 @@ static void avg_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdi pixelsv1 = vec_mergeh(vczero, pixelsv1); pixelsv2 = vec_mergeh(vczero, pixelsv2); +#else /* HAVE_BIGENDIAN */ + pixelsv1 = vec_vsx_ld(line_size, pixels); + pixelsv2 = vec_vsx_ld(line_size+1, pixels); + pixelsv1 = vec_mergeh(pixelsv1, vczero); + pixelsv2 = vec_mergeh(pixelsv2, vczero); +#endif /* HAVE_BIGENDIAN */ pixelssum2 = vec_add((vector unsigned short)pixelsv1, (vector unsigned short)pixelsv2); temp3 = vec_add(pixelssum1, pixelssum2); -- 1.9.1