falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
Browse Source

Cleanup _t types in libavcodec/ppc 

Originally committed as revision 16357 to svn://svn.ffmpeg.org/ffmpeg/trunk
tags/v0.5
Luca Barbato 16 years ago
parent
7f9b3266c9
commit
a6b4448cdf
5 changed files with 330 additions and 338 deletions
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. +146
    -146
      libavcodec/ppc/h264_altivec.c
  2. +142
    -142
      libavcodec/ppc/h264_template_altivec.c
  3. +21
    -29
      libavcodec/ppc/idct_altivec.c
  4. +8
    -8
      libavcodec/ppc/int_altivec.c
  5. +13
    -13
      libavcodec/ppc/types_altivec.h

+ 146
- 146
libavcodec/ppc/h264_altivec.c View File

@@ -189,32 +189,32 @@ void put_no_rnd_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride
((8 - x) * (y)),
((x) * (y))};
register int i;
vec_u8_t fperm;
const vec_s32_t vABCD = vec_ld(0, ABCD);
const vec_s16_t vA = vec_splat((vec_s16_t)vABCD, 1);
const vec_s16_t vB = vec_splat((vec_s16_t)vABCD, 3);
const vec_s16_t vC = vec_splat((vec_s16_t)vABCD, 5);
const vec_s16_t vD = vec_splat((vec_s16_t)vABCD, 7);
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_s16_t v28ss = vec_sub(vec_sl(vec_splat_s16(1),vec_splat_u16(5)),vec_splat_s16(4));
const vec_u16_t v6us = vec_splat_u16(6);
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_t vsrcAuc, vsrcBuc, vsrcperm0, vsrcperm1;
vec_u8_t vsrc0uc, vsrc1uc;
vec_s16_t vsrc0ssH, vsrc1ssH;
vec_u8_t vsrcCuc, vsrc2uc, vsrc3uc;
vec_s16_t vsrc2ssH, vsrc3ssH, psum;
vec_u8_t vdst, ppsum, fsum;
vec_u8 vsrcAuc, vsrcBuc, vsrcperm0, vsrcperm1;
vec_u8 vsrc0uc, vsrc1uc;
vec_s16 vsrc0ssH, vsrc1ssH;
vec_u8 vsrcCuc, vsrc2uc, vsrc3uc;
vec_s16 vsrc2ssH, vsrc3ssH, psum;
vec_u8 vdst, ppsum, fsum;

if (((unsigned long)dst) % 16 == 0) {
fperm = (vec_u8_t){0x10, 0x11, 0x12, 0x13,
fperm = (vec_u8){0x10, 0x11, 0x12, 0x13,
0x14, 0x15, 0x16, 0x17,
0x08, 0x09, 0x0A, 0x0B,
0x0C, 0x0D, 0x0E, 0x0F};
} else {
fperm = (vec_u8_t){0x00, 0x01, 0x02, 0x03,
fperm = (vec_u8){0x00, 0x01, 0x02, 0x03,
0x04, 0x05, 0x06, 0x07,
0x18, 0x19, 0x1A, 0x1B,
0x1C, 0x1D, 0x1E, 0x1F};
@@ -233,8 +233,8 @@ void put_no_rnd_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride
else
vsrc1uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm1);

vsrc0ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc0uc);
vsrc1ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc1uc);
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++) {
@@ -245,8 +245,8 @@ void put_no_rnd_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride
vsrc2uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0);
vsrc3uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm1);

vsrc2ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc2uc);
vsrc3ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc3uc);
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, vec_splat_s16(0));
psum = vec_mladd(vB, vsrc1ssH, psum);
@@ -256,7 +256,7 @@ void put_no_rnd_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride
psum = vec_sra(psum, v6us);

vdst = vec_ld(0, dst);
ppsum = (vec_u8_t)vec_packsu(psum, psum);
ppsum = (vec_u8)vec_packsu(psum, psum);
fsum = vec_perm(vdst, ppsum, fperm);

vec_st(fsum, 0, dst);
@@ -268,7 +268,7 @@ void put_no_rnd_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride
src += stride;
}
} else {
vec_u8_t vsrcDuc;
vec_u8 vsrcDuc;
for (i = 0 ; i < h ; i++) {
vsrcCuc = vec_ld(stride + 0, src);
vsrcDuc = vec_ld(stride + 16, src);
@@ -279,8 +279,8 @@ void put_no_rnd_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride
else
vsrc3uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm1);

vsrc2ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc2uc);
vsrc3ssH = (vec_s16_t)vec_mergeh(zero_u8v, (vec_u8_t)vsrc3uc);
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, vec_splat_s16(0));
psum = vec_mladd(vB, vsrc1ssH, psum);
@@ -290,7 +290,7 @@ void put_no_rnd_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride
psum = vec_sr(psum, v6us);

vdst = vec_ld(0, dst);
ppsum = (vec_u8_t)vec_pack(psum, psum);
ppsum = (vec_u8)vec_pack(psum, psum);
fsum = vec_perm(vdst, ppsum, fperm);

vec_st(fsum, 0, dst);
@@ -309,7 +309,7 @@ static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
int src_stride1, int h)
{
int i;
vec_u8_t a, b, d, tmp1, tmp2, mask, mask_, edges, align;
vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align;

mask_ = vec_lvsl(0, src2);

@@ -351,7 +351,7 @@ static inline void avg_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
int src_stride1, int h)
{
int i;
vec_u8_t a, b, d, tmp1, tmp2, mask, mask_, edges, align;
vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align;

mask_ = vec_lvsl(0, src2);

@@ -432,23 +432,23 @@ H264_MC(avg_, 16, altivec)
#define VEC_LOAD_U8_ADD_S16_STORE_U8(va) \
vdst_orig = vec_ld(0, dst); \
vdst = vec_perm(vdst_orig, zero_u8v, vdst_mask); \
vdst_ss = (vec_s16_t) vec_mergeh(zero_u8v, vdst); \
vdst_ss = (vec_s16) vec_mergeh(zero_u8v, vdst); \
va = vec_add(va, vdst_ss); \
va_u8 = vec_packsu(va, zero_s16v); \
va_u32 = vec_splat((vec_u32_t)va_u8, 0); \
va_u32 = vec_splat((vec_u32)va_u8, 0); \
vec_ste(va_u32, element, (uint32_t*)dst);

static void ff_h264_idct_add_altivec(uint8_t *dst, DCTELEM *block, int stride)
{
vec_s16_t va0, va1, va2, va3;
vec_s16_t vz0, vz1, vz2, vz3;
vec_s16_t vtmp0, vtmp1, vtmp2, vtmp3;
vec_u8_t va_u8;
vec_u32_t va_u32;
vec_s16_t vdst_ss;
const vec_u16_t v6us = vec_splat_u16(6);
vec_u8_t vdst, vdst_orig;
vec_u8_t vdst_mask = vec_lvsl(0, dst);
vec_s16 va0, va1, va2, va3;
vec_s16 vz0, vz1, vz2, vz3;
vec_s16 vtmp0, vtmp1, vtmp2, vtmp3;
vec_u8 va_u8;
vec_u32 va_u32;
vec_s16 vdst_ss;
const vec_u16 v6us = vec_splat_u16(6);
vec_u8 vdst, vdst_orig;
vec_u8 vdst_mask = vec_lvsl(0, dst);
int element = ((unsigned long)dst & 0xf) >> 2;
LOAD_ZERO;

@@ -479,40 +479,40 @@ static void ff_h264_idct_add_altivec(uint8_t *dst, DCTELEM *block, int stride)

#define IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7, d0, d1, d2, d3, d4, d5, d6, d7) {\
/* a0 = SRC(0) + SRC(4); */ \
vec_s16_t a0v = vec_add(s0, s4); \
vec_s16 a0v = vec_add(s0, s4); \
/* a2 = SRC(0) - SRC(4); */ \
vec_s16_t a2v = vec_sub(s0, s4); \
vec_s16 a2v = vec_sub(s0, s4); \
/* a4 = (SRC(2)>>1) - SRC(6); */ \
vec_s16_t a4v = vec_sub(vec_sra(s2, onev), s6); \
vec_s16 a4v = vec_sub(vec_sra(s2, onev), s6); \
/* a6 = (SRC(6)>>1) + SRC(2); */ \
vec_s16_t a6v = vec_add(vec_sra(s6, onev), s2); \
vec_s16 a6v = vec_add(vec_sra(s6, onev), s2); \
/* b0 = a0 + a6; */ \
vec_s16_t b0v = vec_add(a0v, a6v); \
vec_s16 b0v = vec_add(a0v, a6v); \
/* b2 = a2 + a4; */ \
vec_s16_t b2v = vec_add(a2v, a4v); \
vec_s16 b2v = vec_add(a2v, a4v); \
/* b4 = a2 - a4; */ \
vec_s16_t b4v = vec_sub(a2v, a4v); \
vec_s16 b4v = vec_sub(a2v, a4v); \
/* b6 = a0 - a6; */ \
vec_s16_t b6v = vec_sub(a0v, a6v); \
vec_s16 b6v = vec_sub(a0v, a6v); \
/* a1 = SRC(5) - SRC(3) - SRC(7) - (SRC(7)>>1); */ \
/* a1 = (SRC(5)-SRC(3)) - (SRC(7) + (SRC(7)>>1)); */ \
vec_s16_t a1v = vec_sub( vec_sub(s5, s3), vec_add(s7, vec_sra(s7, onev)) ); \
vec_s16 a1v = vec_sub( vec_sub(s5, s3), vec_add(s7, vec_sra(s7, onev)) ); \
/* a3 = SRC(7) + SRC(1) - SRC(3) - (SRC(3)>>1); */ \
/* a3 = (SRC(7)+SRC(1)) - (SRC(3) + (SRC(3)>>1)); */ \
vec_s16_t a3v = vec_sub( vec_add(s7, s1), vec_add(s3, vec_sra(s3, onev)) );\
vec_s16 a3v = vec_sub( vec_add(s7, s1), vec_add(s3, vec_sra(s3, onev)) );\
/* a5 = SRC(7) - SRC(1) + SRC(5) + (SRC(5)>>1); */ \
/* a5 = (SRC(7)-SRC(1)) + SRC(5) + (SRC(5)>>1); */ \
vec_s16_t a5v = vec_add( vec_sub(s7, s1), vec_add(s5, vec_sra(s5, onev)) );\
vec_s16 a5v = vec_add( vec_sub(s7, s1), vec_add(s5, vec_sra(s5, onev)) );\
/* a7 = SRC(5)+SRC(3) + SRC(1) + (SRC(1)>>1); */ \
vec_s16_t a7v = vec_add( vec_add(s5, s3), vec_add(s1, vec_sra(s1, onev)) );\
vec_s16 a7v = vec_add( vec_add(s5, s3), vec_add(s1, vec_sra(s1, onev)) );\
/* b1 = (a7>>2) + a1; */ \
vec_s16_t b1v = vec_add( vec_sra(a7v, twov), a1v); \
vec_s16 b1v = vec_add( vec_sra(a7v, twov), a1v); \
/* b3 = a3 + (a5>>2); */ \
vec_s16_t b3v = vec_add(a3v, vec_sra(a5v, twov)); \
vec_s16 b3v = vec_add(a3v, vec_sra(a5v, twov)); \
/* b5 = (a3>>2) - a5; */ \
vec_s16_t b5v = vec_sub( vec_sra(a3v, twov), a5v); \
vec_s16 b5v = vec_sub( vec_sra(a3v, twov), a5v); \
/* b7 = a7 - (a1>>2); */ \
vec_s16_t b7v = vec_sub( a7v, vec_sra(a1v, twov)); \
vec_s16 b7v = vec_sub( a7v, vec_sra(a1v, twov)); \
/* DST(0, b0 + b7); */ \
d0 = vec_add(b0v, b7v); \
/* DST(1, b2 + b5); */ \
@@ -533,17 +533,17 @@ static void ff_h264_idct_add_altivec(uint8_t *dst, DCTELEM *block, int stride)

#define ALTIVEC_STORE_SUM_CLIP(dest, idctv, perm_ldv, perm_stv, sel) { \
/* unaligned load */ \
vec_u8_t hv = vec_ld( 0, dest ); \
vec_u8_t lv = vec_ld( 7, dest ); \
vec_u8_t dstv = vec_perm( hv, lv, (vec_u8_t)perm_ldv ); \
vec_s16_t idct_sh6 = vec_sra(idctv, sixv); \
vec_u16_t dst16 = (vec_u16_t)vec_mergeh(zero_u8v, dstv); \
vec_s16_t idstsum = vec_adds(idct_sh6, (vec_s16_t)dst16); \
vec_u8_t idstsum8 = vec_packsu(zero_s16v, idstsum); \
vec_u8_t edgehv; \
vec_u8 hv = vec_ld( 0, dest ); \
vec_u8 lv = vec_ld( 7, dest ); \
vec_u8 dstv = vec_perm( hv, lv, (vec_u8)perm_ldv ); \
vec_s16 idct_sh6 = vec_sra(idctv, sixv); \
vec_u16 dst16 = (vec_u16)vec_mergeh(zero_u8v, dstv); \
vec_s16 idstsum = vec_adds(idct_sh6, (vec_s16)dst16); \
vec_u8 idstsum8 = vec_packsu(zero_s16v, idstsum); \
vec_u8 edgehv; \
/* unaligned store */ \
vec_u8_t bodyv = vec_perm( idstsum8, idstsum8, perm_stv );\
vec_u8_t edgelv = vec_perm( sel, zero_u8v, perm_stv ); \
vec_u8 bodyv = vec_perm( idstsum8, idstsum8, perm_stv );\
vec_u8 edgelv = vec_perm( sel, zero_u8v, perm_stv ); \
lv = vec_sel( lv, bodyv, edgelv ); \
vec_st( lv, 7, dest ); \
hv = vec_ld( 0, dest ); \
@@ -553,18 +553,18 @@ static void ff_h264_idct_add_altivec(uint8_t *dst, DCTELEM *block, int stride)
}

void ff_h264_idct8_add_altivec( uint8_t *dst, DCTELEM *dct, int stride ) {
vec_s16_t s0, s1, s2, s3, s4, s5, s6, s7;
vec_s16_t d0, d1, d2, d3, d4, d5, d6, d7;
vec_s16_t idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7;
vec_s16 s0, s1, s2, s3, s4, s5, s6, s7;
vec_s16 d0, d1, d2, d3, d4, d5, d6, d7;
vec_s16 idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7;

vec_u8_t perm_ldv = vec_lvsl(0, dst);
vec_u8_t perm_stv = vec_lvsr(8, dst);
vec_u8 perm_ldv = vec_lvsl(0, dst);
vec_u8 perm_stv = vec_lvsr(8, dst);

const vec_u16_t onev = vec_splat_u16(1);
const vec_u16_t twov = vec_splat_u16(2);
const vec_u16_t sixv = vec_splat_u16(6);
const vec_u16 onev = vec_splat_u16(1);
const vec_u16 twov = vec_splat_u16(2);
const vec_u16 sixv = vec_splat_u16(6);

const vec_u8_t sel = (vec_u8_t) {0,0,0,0,0,0,0,0,-1,-1,-1,-1,-1,-1,-1,-1};
const vec_u8 sel = (vec_u8) {0,0,0,0,0,0,0,0,-1,-1,-1,-1,-1,-1,-1,-1};
LOAD_ZERO;

dct[0] += 32; // rounding for the >>6 at the end
@@ -621,10 +621,10 @@ static void ff_h264_idct8_add4_altivec(uint8_t *dst, const int *block_offset, DC
}

#define transpose4x16(r0, r1, r2, r3) { \
register vec_u8_t r4; \
register vec_u8_t r5; \
register vec_u8_t r6; \
register vec_u8_t r7; \
register vec_u8 r4; \
register vec_u8 r5; \
register vec_u8 r6; \
register vec_u8 r7; \
\
r4 = vec_mergeh(r0, r2); /*0, 2 set 0*/ \
r5 = vec_mergel(r0, r2); /*0, 2 set 1*/ \
@@ -638,8 +638,8 @@ static void ff_h264_idct8_add4_altivec(uint8_t *dst, const int *block_offset, DC
}

static inline void write16x4(uint8_t *dst, int dst_stride,
register vec_u8_t r0, register vec_u8_t r1,
register vec_u8_t r2, register vec_u8_t r3) {
register vec_u8 r0, register vec_u8 r1,
register vec_u8 r2, register vec_u8 r3) {
DECLARE_ALIGNED_16(unsigned char, result[64]);
uint32_t *src_int = (uint32_t *)result, *dst_int = (uint32_t *)dst;
int int_dst_stride = dst_stride/4;
@@ -671,16 +671,16 @@ static inline void write16x4(uint8_t *dst, int dst_stride,
\todo FIXME: see if we can't spare some vec_lvsl() by them factorizing
out of unaligned_load() */
#define readAndTranspose16x6(src, src_stride, r8, r9, r10, r11, r12, r13) {\
register vec_u8_t r0 = unaligned_load(0, src); \
register vec_u8_t r1 = unaligned_load( src_stride, src); \
register vec_u8_t r2 = unaligned_load(2* src_stride, src); \
register vec_u8_t r3 = unaligned_load(3* src_stride, src); \
register vec_u8_t r4 = unaligned_load(4* src_stride, src); \
register vec_u8_t r5 = unaligned_load(5* src_stride, src); \
register vec_u8_t r6 = unaligned_load(6* src_stride, src); \
register vec_u8_t r7 = unaligned_load(7* src_stride, src); \
register vec_u8_t r14 = unaligned_load(14*src_stride, src); \
register vec_u8_t r15 = unaligned_load(15*src_stride, src); \
register vec_u8 r0 = unaligned_load(0, src); \
register vec_u8 r1 = unaligned_load( src_stride, src); \
register vec_u8 r2 = unaligned_load(2* src_stride, src); \
register vec_u8 r3 = unaligned_load(3* src_stride, src); \
register vec_u8 r4 = unaligned_load(4* src_stride, src); \
register vec_u8 r5 = unaligned_load(5* src_stride, src); \
register vec_u8 r6 = unaligned_load(6* src_stride, src); \
register vec_u8 r7 = unaligned_load(7* src_stride, src); \
register vec_u8 r14 = unaligned_load(14*src_stride, src); \
register vec_u8 r15 = unaligned_load(15*src_stride, src); \
\
r8 = unaligned_load( 8*src_stride, src); \
r9 = unaligned_load( 9*src_stride, src); \
@@ -730,26 +730,26 @@ static inline void write16x4(uint8_t *dst, int dst_stride,
}

// out: o = |x-y| < a
static inline vec_u8_t diff_lt_altivec ( register vec_u8_t x,
register vec_u8_t y,
register vec_u8_t a) {
register vec_u8_t diff = vec_subs(x, y);
register vec_u8_t diffneg = vec_subs(y, x);
register vec_u8_t o = vec_or(diff, diffneg); /* |x-y| */
o = (vec_u8_t)vec_cmplt(o, a);
static inline vec_u8 diff_lt_altivec ( register vec_u8 x,
register vec_u8 y,
register vec_u8 a) {
register vec_u8 diff = vec_subs(x, y);
register vec_u8 diffneg = vec_subs(y, x);
register vec_u8 o = vec_or(diff, diffneg); /* |x-y| */
o = (vec_u8)vec_cmplt(o, a);
return o;
}

static inline vec_u8_t h264_deblock_mask ( register vec_u8_t p0,
register vec_u8_t p1,
register vec_u8_t q0,
register vec_u8_t q1,
register vec_u8_t alpha,
register vec_u8_t beta) {
static inline vec_u8 h264_deblock_mask ( register vec_u8 p0,
register vec_u8 p1,
register vec_u8 q0,
register vec_u8 q1,
register vec_u8 alpha,
register vec_u8 beta) {

register vec_u8_t mask;
register vec_u8_t tempmask;
register vec_u8 mask;
register vec_u8 tempmask;

mask = diff_lt_altivec(p0, q0, alpha);
tempmask = diff_lt_altivec(p1, p0, beta);
@@ -761,19 +761,19 @@ static inline vec_u8_t h264_deblock_mask ( register vec_u8_t p0,
}

// out: newp1 = clip((p2 + ((p0 + q0 + 1) >> 1)) >> 1, p1-tc0, p1+tc0)
static inline vec_u8_t h264_deblock_q1(register vec_u8_t p0,
register vec_u8_t p1,
register vec_u8_t p2,
register vec_u8_t q0,
register vec_u8_t tc0) {
register vec_u8_t average = vec_avg(p0, q0);
register vec_u8_t temp;
register vec_u8_t uncliped;
register vec_u8_t ones;
register vec_u8_t max;
register vec_u8_t min;
register vec_u8_t newp1;
static inline vec_u8 h264_deblock_q1(register vec_u8 p0,
register vec_u8 p1,
register vec_u8 p2,
register vec_u8 q0,
register vec_u8 tc0) {
register vec_u8 average = vec_avg(p0, q0);
register vec_u8 temp;
register vec_u8 uncliped;
register vec_u8 ones;
register vec_u8 max;
register vec_u8 min;
register vec_u8 newp1;

temp = vec_xor(average, p2);
average = vec_avg(average, p2); /*avg(p2, avg(p0, q0)) */
@@ -789,16 +789,16 @@ static inline vec_u8_t h264_deblock_q1(register vec_u8_t p0,

#define h264_deblock_p0_q0(p0, p1, q0, q1, tc0masked) { \
\
const vec_u8_t A0v = vec_sl(vec_splat_u8(10), vec_splat_u8(4)); \
const vec_u8 A0v = vec_sl(vec_splat_u8(10), vec_splat_u8(4)); \
\
register vec_u8_t pq0bit = vec_xor(p0,q0); \
register vec_u8_t q1minus; \
register vec_u8_t p0minus; \
register vec_u8_t stage1; \
register vec_u8_t stage2; \
register vec_u8_t vec160; \
register vec_u8_t delta; \
register vec_u8_t deltaneg; \
register vec_u8 pq0bit = vec_xor(p0,q0); \
register vec_u8 q1minus; \
register vec_u8 p0minus; \
register vec_u8 stage1; \
register vec_u8 stage2; \
register vec_u8 vec160; \
register vec_u8 delta; \
register vec_u8 deltaneg; \
\
q1minus = vec_nor(q1, q1); /* 255 - q1 */ \
stage1 = vec_avg(p1, q1minus); /* (p1 - q1 + 256)>>1 */ \
@@ -821,16 +821,16 @@ static inline vec_u8_t h264_deblock_q1(register vec_u8_t p0,

#define h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0) { \
DECLARE_ALIGNED_16(unsigned char, temp[16]); \
register vec_u8_t alphavec; \
register vec_u8_t betavec; \
register vec_u8_t mask; \
register vec_u8_t p1mask; \
register vec_u8_t q1mask; \
register vec_u8 alphavec; \
register vec_u8 betavec; \
register vec_u8 mask; \
register vec_u8 p1mask; \
register vec_u8 q1mask; \
register vector signed char tc0vec; \
register vec_u8_t finaltc0; \
register vec_u8_t tc0masked; \
register vec_u8_t newp1; \
register vec_u8_t newq1; \
register vec_u8 finaltc0; \
register vec_u8 tc0masked; \
register vec_u8 newp1; \
register vec_u8 newq1; \
\
temp[0] = alpha; \
temp[1] = beta; \
@@ -844,18 +844,18 @@ static inline vec_u8_t h264_deblock_q1(register vec_u8_t p0,
tc0vec = vec_mergeh(tc0vec, tc0vec); \
tc0vec = vec_mergeh(tc0vec, tc0vec); \
mask = vec_and(mask, vec_cmpgt(tc0vec, vec_splat_s8(-1))); /* if tc0[i] >= 0 */ \
finaltc0 = vec_and((vec_u8_t)tc0vec, mask); /* tc = tc0 */ \
finaltc0 = vec_and((vec_u8)tc0vec, mask); /* tc = tc0 */ \
\
p1mask = diff_lt_altivec(p2, p0, betavec); \
p1mask = vec_and(p1mask, mask); /* if ( |p2 - p0| < beta) */ \
tc0masked = vec_and(p1mask, (vec_u8_t)tc0vec); \
tc0masked = vec_and(p1mask, (vec_u8)tc0vec); \
finaltc0 = vec_sub(finaltc0, p1mask); /* tc++ */ \
newp1 = h264_deblock_q1(p0, p1, p2, q0, tc0masked); \
/*end if*/ \
\
q1mask = diff_lt_altivec(q2, q0, betavec); \
q1mask = vec_and(q1mask, mask); /* if ( |q2 - q0| < beta ) */\
tc0masked = vec_and(q1mask, (vec_u8_t)tc0vec); \
tc0masked = vec_and(q1mask, (vec_u8)tc0vec); \
finaltc0 = vec_sub(finaltc0, q1mask); /* tc++ */ \
newq1 = h264_deblock_q1(p0, q1, q2, q0, tc0masked); \
/*end if*/ \
@@ -868,12 +868,12 @@ static inline vec_u8_t h264_deblock_q1(register vec_u8_t p0,
static void h264_v_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {

if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) >= 0) {
register vec_u8_t p2 = vec_ld(-3*stride, pix);
register vec_u8_t p1 = vec_ld(-2*stride, pix);
register vec_u8_t p0 = vec_ld(-1*stride, pix);
register vec_u8_t q0 = vec_ld(0, pix);
register vec_u8_t q1 = vec_ld(stride, pix);
register vec_u8_t q2 = vec_ld(2*stride, pix);
register vec_u8 p2 = vec_ld(-3*stride, pix);
register vec_u8 p1 = vec_ld(-2*stride, pix);
register vec_u8 p0 = vec_ld(-1*stride, pix);
register vec_u8 q0 = vec_ld(0, pix);
register vec_u8 q1 = vec_ld(stride, pix);
register vec_u8 q2 = vec_ld(2*stride, pix);
h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0);
vec_st(p1, -2*stride, pix);
vec_st(p0, -1*stride, pix);
@@ -884,7 +884,7 @@ static void h264_v_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha,

static void h264_h_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {

register vec_u8_t line0, line1, line2, line3, line4, line5;
register vec_u8 line0, line1, line2, line3, line4, line5;
if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) < 0)
return;
readAndTranspose16x6(pix-3, stride, line0, line1, line2, line3, line4, line5);


+ 142
- 142
libavcodec/ppc/h264_template_altivec.c View File

@@ -28,8 +28,8 @@
/* this code assume that stride % 16 == 0 */

#define CHROMA_MC8_ALTIVEC_CORE \
vsrc2ssH = (vec_s16_t)vec_mergeh(zero_u8v,(vec_u8_t)vsrc2uc);\
vsrc3ssH = (vec_s16_t)vec_mergeh(zero_u8v,(vec_u8_t)vsrc3uc);\
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, v32ss);\
psum = vec_mladd(vB, vsrc1ssH, psum);\
@@ -38,7 +38,7 @@
psum = vec_sr(psum, v6us);\
\
vdst = vec_ld(0, dst);\
ppsum = (vec_u8_t)vec_pack(psum, psum);\
ppsum = (vec_u8)vec_pack(psum, psum);\
vfdst = vec_perm(vdst, ppsum, fperm);\
\
OP_U8_ALTIVEC(fsum, vfdst, vdst);\
@@ -53,15 +53,15 @@

#define CHROMA_MC8_ALTIVEC_CORE_SIMPLE \
\
vsrc0ssH = (vec_s16_t)vec_mergeh(zero_u8v,(vec_u8_t)vsrc0uc);\
vsrc1ssH = (vec_s16_t)vec_mergeh(zero_u8v,(vec_u8_t)vsrc1uc);\
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_t)vec_pack(psum, psum);\
ppsum = (vec_u8)vec_pack(psum, psum);\
vfdst = vec_perm(vdst, ppsum, fperm);\
\
OP_U8_ALTIVEC(fsum, vfdst, vdst);\
@@ -80,34 +80,34 @@ void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src,
((8 - x) * ( y)),
(( x) * ( y))};
register int i;
vec_u8_t fperm;
const vec_s32_t vABCD = vec_ld(0, ABCD);
const vec_s16_t vA = vec_splat((vec_s16_t)vABCD, 1);
const vec_s16_t vB = vec_splat((vec_s16_t)vABCD, 3);
const vec_s16_t vC = vec_splat((vec_s16_t)vABCD, 5);
const vec_s16_t vD = vec_splat((vec_s16_t)vABCD, 7);
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_s16_t v32ss = vec_sl(vec_splat_s16(1),vec_splat_u16(5));
const vec_u16_t v6us = vec_splat_u16(6);
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_t vsrcAuc, vsrcBuc, vsrcperm0, vsrcperm1;
vec_u8_t vsrc0uc, vsrc1uc;
vec_s16_t vsrc0ssH, vsrc1ssH;
vec_u8_t vsrcCuc, vsrc2uc, vsrc3uc;
vec_s16_t vsrc2ssH, vsrc3ssH, psum;
vec_u8_t vdst, ppsum, vfdst, fsum;
vec_u8 vsrcAuc, vsrcBuc, vsrcperm0, vsrcperm1;
vec_u8 vsrc0uc, vsrc1uc;
vec_s16 vsrc0ssH, vsrc1ssH;
vec_u8 vsrcCuc, vsrc2uc, vsrc3uc;
vec_s16 vsrc2ssH, vsrc3ssH, psum;
vec_u8 vdst, ppsum, vfdst, fsum;

POWERPC_PERF_START_COUNT(PREFIX_h264_chroma_mc8_num, 1);

if (((unsigned long)dst) % 16 == 0) {
fperm = (vec_u8_t){0x10, 0x11, 0x12, 0x13,
fperm = (vec_u8){0x10, 0x11, 0x12, 0x13,
0x14, 0x15, 0x16, 0x17,
0x08, 0x09, 0x0A, 0x0B,
0x0C, 0x0D, 0x0E, 0x0F};
} else {
fperm = (vec_u8_t){0x00, 0x01, 0x02, 0x03,
fperm = (vec_u8){0x00, 0x01, 0x02, 0x03,
0x04, 0x05, 0x06, 0x07,
0x18, 0x19, 0x1A, 0x1B,
0x1C, 0x1D, 0x1E, 0x1F};
@@ -126,8 +126,8 @@ void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src,
else
vsrc1uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm1);

vsrc0ssH = (vec_s16_t)vec_mergeh(zero_u8v,(vec_u8_t)vsrc0uc);
vsrc1ssH = (vec_s16_t)vec_mergeh(zero_u8v,(vec_u8_t)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
@@ -139,7 +139,7 @@ void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src,
CHROMA_MC8_ALTIVEC_CORE
}
} else {
vec_u8_t vsrcDuc;
vec_u8 vsrcDuc;
for (i = 0 ; i < h ; i++) {
vsrcCuc = vec_ld(stride + 0, src);
vsrcDuc = vec_ld(stride + 16, src);
@@ -153,7 +153,7 @@ void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src,
}
}
} else {
const vec_s16_t vE = vec_add(vB, vC);
const vec_s16 vE = vec_add(vB, vC);
if (ABCD[2]) { // x == 0 B == 0
if (!loadSecond) {// -> !reallyBadAlign
for (i = 0 ; i < h ; i++) {
@@ -164,7 +164,7 @@ void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src,
vsrc0uc = vsrc1uc;
}
} else {
vec_u8_t vsrcDuc;
vec_u8 vsrcDuc;
for (i = 0 ; i < h ; i++) {
vsrcCuc = vec_ld(stride + 0, src);
vsrcDuc = vec_ld(stride + 15, src);
@@ -184,7 +184,7 @@ void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src,
CHROMA_MC8_ALTIVEC_CORE_SIMPLE
}
} else {
vec_u8_t vsrcDuc;
vec_u8 vsrcDuc;
for (i = 0 ; i < h ; i++) {
vsrcCuc = vec_ld(0, src);
vsrcDuc = vec_ld(15, src);
@@ -210,35 +210,35 @@ static void PREFIX_h264_qpel16_h_lowpass_altivec(uint8_t * dst, uint8_t * src, i
register int i;

LOAD_ZERO;
const vec_u8_t permM2 = vec_lvsl(-2, src);
const vec_u8_t permM1 = vec_lvsl(-1, src);
const vec_u8_t permP0 = vec_lvsl(+0, src);
const vec_u8_t permP1 = vec_lvsl(+1, src);
const vec_u8_t permP2 = vec_lvsl(+2, src);
const vec_u8_t permP3 = vec_lvsl(+3, src);
const vec_s16_t v5ss = vec_splat_s16(5);
const vec_u16_t v5us = vec_splat_u16(5);
const vec_s16_t v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2));
const vec_s16_t v16ss = vec_sl(vec_splat_s16(1),vec_splat_u16(4));
vec_u8_t srcM2, srcM1, srcP0, srcP1, srcP2, srcP3;
const vec_u8 permM2 = vec_lvsl(-2, src);
const vec_u8 permM1 = vec_lvsl(-1, src);
const vec_u8 permP0 = vec_lvsl(+0, src);
const vec_u8 permP1 = vec_lvsl(+1, src);
const vec_u8 permP2 = vec_lvsl(+2, src);
const vec_u8 permP3 = vec_lvsl(+3, src);
const vec_s16 v5ss = vec_splat_s16(5);
const vec_u16 v5us = vec_splat_u16(5);
const vec_s16 v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2));
const vec_s16 v16ss = vec_sl(vec_splat_s16(1),vec_splat_u16(4));
vec_u8 srcM2, srcM1, srcP0, srcP1, srcP2, srcP3;

register int align = ((((unsigned long)src) - 2) % 16);

vec_s16_t srcP0A, srcP0B, srcP1A, srcP1B,
vec_s16 srcP0A, srcP0B, srcP1A, srcP1B,
srcP2A, srcP2B, srcP3A, srcP3B,
srcM1A, srcM1B, srcM2A, srcM2B,
sum1A, sum1B, sum2A, sum2B, sum3A, sum3B,
pp1A, pp1B, pp2A, pp2B, pp3A, pp3B,
psumA, psumB, sumA, sumB;

vec_u8_t sum, vdst, fsum;
vec_u8 sum, vdst, fsum;

POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_h_lowpass_num, 1);

for (i = 0 ; i < 16 ; i ++) {
vec_u8_t srcR1 = vec_ld(-2, src);
vec_u8_t srcR2 = vec_ld(14, src);
vec_u8 srcR1 = vec_ld(-2, src);
vec_u8 srcR2 = vec_ld(14, src);

switch (align) {
default: {
@@ -258,7 +258,7 @@ static void PREFIX_h264_qpel16_h_lowpass_altivec(uint8_t * dst, uint8_t * src, i
srcP3 = srcR2;
} break;
case 12: {
vec_u8_t srcR3 = vec_ld(30, src);
vec_u8 srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
@@ -267,7 +267,7 @@ static void PREFIX_h264_qpel16_h_lowpass_altivec(uint8_t * dst, uint8_t * src, i
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 13: {
vec_u8_t srcR3 = vec_ld(30, src);
vec_u8 srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
@@ -276,7 +276,7 @@ static void PREFIX_h264_qpel16_h_lowpass_altivec(uint8_t * dst, uint8_t * src, i
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 14: {
vec_u8_t srcR3 = vec_ld(30, src);
vec_u8 srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = srcR2;
@@ -285,7 +285,7 @@ static void PREFIX_h264_qpel16_h_lowpass_altivec(uint8_t * dst, uint8_t * src, i
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 15: {
vec_u8_t srcR3 = vec_ld(30, src);
vec_u8 srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = srcR2;
srcP0 = vec_perm(srcR2, srcR3, permP0);
@@ -295,20 +295,20 @@ static void PREFIX_h264_qpel16_h_lowpass_altivec(uint8_t * dst, uint8_t * src, i
} break;
}

srcP0A = (vec_s16_t) vec_mergeh(zero_u8v, srcP0);
srcP0B = (vec_s16_t) vec_mergel(zero_u8v, srcP0);
srcP1A = (vec_s16_t) vec_mergeh(zero_u8v, srcP1);
srcP1B = (vec_s16_t) vec_mergel(zero_u8v, srcP1);
srcP0A = (vec_s16) vec_mergeh(zero_u8v, srcP0);
srcP0B = (vec_s16) vec_mergel(zero_u8v, srcP0);
srcP1A = (vec_s16) vec_mergeh(zero_u8v, srcP1);
srcP1B = (vec_s16) vec_mergel(zero_u8v, srcP1);

srcP2A = (vec_s16_t) vec_mergeh(zero_u8v, srcP2);
srcP2B = (vec_s16_t) vec_mergel(zero_u8v, srcP2);
srcP3A = (vec_s16_t) vec_mergeh(zero_u8v, srcP3);
srcP3B = (vec_s16_t) vec_mergel(zero_u8v, srcP3);
srcP2A = (vec_s16) vec_mergeh(zero_u8v, srcP2);
srcP2B = (vec_s16) vec_mergel(zero_u8v, srcP2);
srcP3A = (vec_s16) vec_mergeh(zero_u8v, srcP3);
srcP3B = (vec_s16) vec_mergel(zero_u8v, srcP3);

srcM1A = (vec_s16_t) vec_mergeh(zero_u8v, srcM1);
srcM1B = (vec_s16_t) vec_mergel(zero_u8v, srcM1);
srcM2A = (vec_s16_t) vec_mergeh(zero_u8v, srcM2);
srcM2B = (vec_s16_t) vec_mergel(zero_u8v, srcM2);
srcM1A = (vec_s16) vec_mergeh(zero_u8v, srcM1);
srcM1B = (vec_s16) vec_mergel(zero_u8v, srcM1);
srcM2A = (vec_s16) vec_mergeh(zero_u8v, srcM2);
srcM2B = (vec_s16) vec_mergel(zero_u8v, srcM2);

sum1A = vec_adds(srcP0A, srcP1A);
sum1B = vec_adds(srcP0B, srcP1B);
@@ -354,52 +354,52 @@ static void PREFIX_h264_qpel16_v_lowpass_altivec(uint8_t * dst, uint8_t * src, i
register int i;

LOAD_ZERO;
const vec_u8_t perm = vec_lvsl(0, src);
const vec_s16_t v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2));
const vec_u16_t v5us = vec_splat_u16(5);
const vec_s16_t v5ss = vec_splat_s16(5);
const vec_s16_t v16ss = vec_sl(vec_splat_s16(1),vec_splat_u16(4));
const vec_u8 perm = vec_lvsl(0, src);
const vec_s16 v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2));
const vec_u16 v5us = vec_splat_u16(5);
const vec_s16 v5ss = vec_splat_s16(5);
const vec_s16 v16ss = vec_sl(vec_splat_s16(1),vec_splat_u16(4));

uint8_t *srcbis = src - (srcStride * 2);

const vec_u8_t srcM2a = vec_ld(0, srcbis);
const vec_u8_t srcM2b = vec_ld(16, srcbis);
const vec_u8_t srcM2 = vec_perm(srcM2a, srcM2b, perm);
const vec_u8 srcM2a = vec_ld(0, srcbis);
const vec_u8 srcM2b = vec_ld(16, srcbis);
const vec_u8 srcM2 = vec_perm(srcM2a, srcM2b, perm);
//srcbis += srcStride;
const vec_u8_t srcM1a = vec_ld(0, srcbis += srcStride);
const vec_u8_t srcM1b = vec_ld(16, srcbis);
const vec_u8_t srcM1 = vec_perm(srcM1a, srcM1b, perm);
const vec_u8 srcM1a = vec_ld(0, srcbis += srcStride);
const vec_u8 srcM1b = vec_ld(16, srcbis);
const vec_u8 srcM1 = vec_perm(srcM1a, srcM1b, perm);
//srcbis += srcStride;
const vec_u8_t srcP0a = vec_ld(0, srcbis += srcStride);
const vec_u8_t srcP0b = vec_ld(16, srcbis);
const vec_u8_t srcP0 = vec_perm(srcP0a, srcP0b, perm);
const vec_u8 srcP0a = vec_ld(0, srcbis += srcStride);
const vec_u8 srcP0b = vec_ld(16, srcbis);
const vec_u8 srcP0 = vec_perm(srcP0a, srcP0b, perm);
//srcbis += srcStride;
const vec_u8_t srcP1a = vec_ld(0, srcbis += srcStride);
const vec_u8_t srcP1b = vec_ld(16, srcbis);
const vec_u8_t srcP1 = vec_perm(srcP1a, srcP1b, perm);
const vec_u8 srcP1a = vec_ld(0, srcbis += srcStride);
const vec_u8 srcP1b = vec_ld(16, srcbis);
const vec_u8 srcP1 = vec_perm(srcP1a, srcP1b, perm);
//srcbis += srcStride;
const vec_u8_t srcP2a = vec_ld(0, srcbis += srcStride);
const vec_u8_t srcP2b = vec_ld(16, srcbis);
const vec_u8_t srcP2 = vec_perm(srcP2a, srcP2b, perm);
const vec_u8 srcP2a = vec_ld(0, srcbis += srcStride);
const vec_u8 srcP2b = vec_ld(16, srcbis);
const vec_u8 srcP2 = vec_perm(srcP2a, srcP2b, perm);
//srcbis += srcStride;

vec_s16_t srcM2ssA = (vec_s16_t) vec_mergeh(zero_u8v, srcM2);
vec_s16_t srcM2ssB = (vec_s16_t) vec_mergel(zero_u8v, srcM2);
vec_s16_t srcM1ssA = (vec_s16_t) vec_mergeh(zero_u8v, srcM1);
vec_s16_t srcM1ssB = (vec_s16_t) vec_mergel(zero_u8v, srcM1);
vec_s16_t srcP0ssA = (vec_s16_t) vec_mergeh(zero_u8v, srcP0);
vec_s16_t srcP0ssB = (vec_s16_t) vec_mergel(zero_u8v, srcP0);
vec_s16_t srcP1ssA = (vec_s16_t) vec_mergeh(zero_u8v, srcP1);
vec_s16_t srcP1ssB = (vec_s16_t) vec_mergel(zero_u8v, srcP1);
vec_s16_t srcP2ssA = (vec_s16_t) vec_mergeh(zero_u8v, srcP2);
vec_s16_t srcP2ssB = (vec_s16_t) vec_mergel(zero_u8v, srcP2);
vec_s16_t pp1A, pp1B, pp2A, pp2B, pp3A, pp3B,
vec_s16 srcM2ssA = (vec_s16) vec_mergeh(zero_u8v, srcM2);
vec_s16 srcM2ssB = (vec_s16) vec_mergel(zero_u8v, srcM2);
vec_s16 srcM1ssA = (vec_s16) vec_mergeh(zero_u8v, srcM1);
vec_s16 srcM1ssB = (vec_s16) vec_mergel(zero_u8v, srcM1);
vec_s16 srcP0ssA = (vec_s16) vec_mergeh(zero_u8v, srcP0);
vec_s16 srcP0ssB = (vec_s16) vec_mergel(zero_u8v, srcP0);
vec_s16 srcP1ssA = (vec_s16) vec_mergeh(zero_u8v, srcP1);
vec_s16 srcP1ssB = (vec_s16) vec_mergel(zero_u8v, srcP1);
vec_s16 srcP2ssA = (vec_s16) vec_mergeh(zero_u8v, srcP2);
vec_s16 srcP2ssB = (vec_s16) vec_mergel(zero_u8v, srcP2);
vec_s16 pp1A, pp1B, pp2A, pp2B, pp3A, pp3B,
psumA, psumB, sumA, sumB,
srcP3ssA, srcP3ssB,
sum1A, sum1B, sum2A, sum2B, sum3A, sum3B;

vec_u8_t sum, vdst, fsum, srcP3a, srcP3b, srcP3;
vec_u8 sum, vdst, fsum, srcP3a, srcP3b, srcP3;

POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_v_lowpass_num, 1);

@@ -407,8 +407,8 @@ static void PREFIX_h264_qpel16_v_lowpass_altivec(uint8_t * dst, uint8_t * src, i
srcP3a = vec_ld(0, srcbis += srcStride);
srcP3b = vec_ld(16, srcbis);
srcP3 = vec_perm(srcP3a, srcP3b, perm);
srcP3ssA = (vec_s16_t) vec_mergeh(zero_u8v, srcP3);
srcP3ssB = (vec_s16_t) vec_mergel(zero_u8v, srcP3);
srcP3ssA = (vec_s16) vec_mergeh(zero_u8v, srcP3);
srcP3ssB = (vec_s16) vec_mergel(zero_u8v, srcP3);
//srcbis += srcStride;

sum1A = vec_adds(srcP0ssA, srcP1ssA);
@@ -463,49 +463,49 @@ static void PREFIX_h264_qpel16_hv_lowpass_altivec(uint8_t * dst, int16_t * tmp,
POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_hv_lowpass_num, 1);
register int i;
LOAD_ZERO;
const vec_u8_t permM2 = vec_lvsl(-2, src);
const vec_u8_t permM1 = vec_lvsl(-1, src);
const vec_u8_t permP0 = vec_lvsl(+0, src);
const vec_u8_t permP1 = vec_lvsl(+1, src);
const vec_u8_t permP2 = vec_lvsl(+2, src);
const vec_u8_t permP3 = vec_lvsl(+3, src);
const vec_s16_t v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2));
const vec_u32_t v10ui = vec_splat_u32(10);
const vec_s16_t v5ss = vec_splat_s16(5);
const vec_s16_t v1ss = vec_splat_s16(1);
const vec_s32_t v512si = vec_sl(vec_splat_s32(1),vec_splat_u32(9));
const vec_u32_t v16ui = vec_sl(vec_splat_u32(1),vec_splat_u32(4));
const vec_u8 permM2 = vec_lvsl(-2, src);
const vec_u8 permM1 = vec_lvsl(-1, src);
const vec_u8 permP0 = vec_lvsl(+0, src);
const vec_u8 permP1 = vec_lvsl(+1, src);
const vec_u8 permP2 = vec_lvsl(+2, src);
const vec_u8 permP3 = vec_lvsl(+3, src);
const vec_s16 v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2));
const vec_u32 v10ui = vec_splat_u32(10);
const vec_s16 v5ss = vec_splat_s16(5);
const vec_s16 v1ss = vec_splat_s16(1);
const vec_s32 v512si = vec_sl(vec_splat_s32(1),vec_splat_u32(9));
const vec_u32 v16ui = vec_sl(vec_splat_u32(1),vec_splat_u32(4));

register int align = ((((unsigned long)src) - 2) % 16);

vec_s16_t srcP0A, srcP0B, srcP1A, srcP1B,
vec_s16 srcP0A, srcP0B, srcP1A, srcP1B,
srcP2A, srcP2B, srcP3A, srcP3B,
srcM1A, srcM1B, srcM2A, srcM2B,
sum1A, sum1B, sum2A, sum2B, sum3A, sum3B,
pp1A, pp1B, pp2A, pp2B, psumA, psumB;

const vec_u8_t mperm = (const vec_u8_t)
const vec_u8 mperm = (const vec_u8)
{0x00, 0x08, 0x01, 0x09, 0x02, 0x0A, 0x03, 0x0B,
0x04, 0x0C, 0x05, 0x0D, 0x06, 0x0E, 0x07, 0x0F};
int16_t *tmpbis = tmp;

vec_s16_t tmpM1ssA, tmpM1ssB, tmpM2ssA, tmpM2ssB,
vec_s16 tmpM1ssA, tmpM1ssB, tmpM2ssA, tmpM2ssB,
tmpP0ssA, tmpP0ssB, tmpP1ssA, tmpP1ssB,
tmpP2ssA, tmpP2ssB;

vec_s32_t pp1Ae, pp1Ao, pp1Be, pp1Bo, pp2Ae, pp2Ao, pp2Be, pp2Bo,
vec_s32 pp1Ae, pp1Ao, pp1Be, pp1Bo, pp2Ae, pp2Ao, pp2Be, pp2Bo,
pp3Ae, pp3Ao, pp3Be, pp3Bo, pp1cAe, pp1cAo, pp1cBe, pp1cBo,
pp32Ae, pp32Ao, pp32Be, pp32Bo, sumAe, sumAo, sumBe, sumBo,
ssumAe, ssumAo, ssumBe, ssumBo;
vec_u8_t fsum, sumv, sum, vdst;
vec_s16_t ssume, ssumo;
vec_u8 fsum, sumv, sum, vdst;
vec_s16 ssume, ssumo;

POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_hv_lowpass_num, 1);
src -= (2 * srcStride);
for (i = 0 ; i < 21 ; i ++) {
vec_u8_t srcM2, srcM1, srcP0, srcP1, srcP2, srcP3;
vec_u8_t srcR1 = vec_ld(-2, src);
vec_u8_t srcR2 = vec_ld(14, src);
vec_u8 srcM2, srcM1, srcP0, srcP1, srcP2, srcP3;
vec_u8 srcR1 = vec_ld(-2, src);
vec_u8 srcR2 = vec_ld(14, src);

switch (align) {
default: {
@@ -525,7 +525,7 @@ static void PREFIX_h264_qpel16_hv_lowpass_altivec(uint8_t * dst, int16_t * tmp,
srcP3 = srcR2;
} break;
case 12: {
vec_u8_t srcR3 = vec_ld(30, src);
vec_u8 srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
@@ -534,7 +534,7 @@ static void PREFIX_h264_qpel16_hv_lowpass_altivec(uint8_t * dst, int16_t * tmp,
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 13: {
vec_u8_t srcR3 = vec_ld(30, src);
vec_u8 srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
@@ -543,7 +543,7 @@ static void PREFIX_h264_qpel16_hv_lowpass_altivec(uint8_t * dst, int16_t * tmp,
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 14: {
vec_u8_t srcR3 = vec_ld(30, src);
vec_u8 srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = srcR2;
@@ -552,7 +552,7 @@ static void PREFIX_h264_qpel16_hv_lowpass_altivec(uint8_t * dst, int16_t * tmp,
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 15: {
vec_u8_t srcR3 = vec_ld(30, src);
vec_u8 srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = srcR2;
srcP0 = vec_perm(srcR2, srcR3, permP0);
@@ -562,20 +562,20 @@ static void PREFIX_h264_qpel16_hv_lowpass_altivec(uint8_t * dst, int16_t * tmp,
} break;
}

srcP0A = (vec_s16_t) vec_mergeh(zero_u8v, srcP0);
srcP0B = (vec_s16_t) vec_mergel(zero_u8v, srcP0);
srcP1A = (vec_s16_t) vec_mergeh(zero_u8v, srcP1);
srcP1B = (vec_s16_t) vec_mergel(zero_u8v, srcP1);
srcP0A = (vec_s16) vec_mergeh(zero_u8v, srcP0);
srcP0B = (vec_s16) vec_mergel(zero_u8v, srcP0);
srcP1A = (vec_s16) vec_mergeh(zero_u8v, srcP1);
srcP1B = (vec_s16) vec_mergel(zero_u8v, srcP1);

srcP2A = (vec_s16_t) vec_mergeh(zero_u8v, srcP2);
srcP2B = (vec_s16_t) vec_mergel(zero_u8v, srcP2);
srcP3A = (vec_s16_t) vec_mergeh(zero_u8v, srcP3);
srcP3B = (vec_s16_t) vec_mergel(zero_u8v, srcP3);
srcP2A = (vec_s16) vec_mergeh(zero_u8v, srcP2);
srcP2B = (vec_s16) vec_mergel(zero_u8v, srcP2);
srcP3A = (vec_s16) vec_mergeh(zero_u8v, srcP3);
srcP3B = (vec_s16) vec_mergel(zero_u8v, srcP3);

srcM1A = (vec_s16_t) vec_mergeh(zero_u8v, srcM1);
srcM1B = (vec_s16_t) vec_mergel(zero_u8v, srcM1);
srcM2A = (vec_s16_t) vec_mergeh(zero_u8v, srcM2);
srcM2B = (vec_s16_t) vec_mergel(zero_u8v, srcM2);
srcM1A = (vec_s16) vec_mergeh(zero_u8v, srcM1);
srcM1B = (vec_s16) vec_mergel(zero_u8v, srcM1);
srcM2A = (vec_s16) vec_mergeh(zero_u8v, srcM2);
srcM2B = (vec_s16) vec_mergel(zero_u8v, srcM2);

sum1A = vec_adds(srcP0A, srcP1A);
sum1B = vec_adds(srcP0B, srcP1B);
@@ -617,15 +617,15 @@ static void PREFIX_h264_qpel16_hv_lowpass_altivec(uint8_t * dst, int16_t * tmp,
tmpbis += tmpStride;

for (i = 0 ; i < 16 ; i++) {
const vec_s16_t tmpP3ssA = vec_ld(0, tmpbis);
const vec_s16_t tmpP3ssB = vec_ld(16, tmpbis);
const vec_s16 tmpP3ssA = vec_ld(0, tmpbis);
const vec_s16 tmpP3ssB = vec_ld(16, tmpbis);

const vec_s16_t sum1A = vec_adds(tmpP0ssA, tmpP1ssA);
const vec_s16_t sum1B = vec_adds(tmpP0ssB, tmpP1ssB);
const vec_s16_t sum2A = vec_adds(tmpM1ssA, tmpP2ssA);
const vec_s16_t sum2B = vec_adds(tmpM1ssB, tmpP2ssB);
const vec_s16_t sum3A = vec_adds(tmpM2ssA, tmpP3ssA);
const vec_s16_t sum3B = vec_adds(tmpM2ssB, tmpP3ssB);
const vec_s16 sum1A = vec_adds(tmpP0ssA, tmpP1ssA);
const vec_s16 sum1B = vec_adds(tmpP0ssB, tmpP1ssB);
const vec_s16 sum2A = vec_adds(tmpM1ssA, tmpP2ssA);
const vec_s16 sum2B = vec_adds(tmpM1ssB, tmpP2ssB);
const vec_s16 sum3A = vec_adds(tmpM2ssA, tmpP3ssA);
const vec_s16 sum3B = vec_adds(tmpM2ssB, tmpP3ssB);

tmpbis += tmpStride;

@@ -650,9 +650,9 @@ static void PREFIX_h264_qpel16_hv_lowpass_altivec(uint8_t * dst, int16_t * tmp,
pp2Be = vec_mule(sum2B, v5ss);
pp2Bo = vec_mulo(sum2B, v5ss);

pp3Ae = vec_sra((vec_s32_t)sum3A, v16ui);
pp3Ae = vec_sra((vec_s32)sum3A, v16ui);
pp3Ao = vec_mulo(sum3A, v1ss);
pp3Be = vec_sra((vec_s32_t)sum3B, v16ui);
pp3Be = vec_sra((vec_s32)sum3B, v16ui);
pp3Bo = vec_mulo(sum3B, v1ss);

pp1cAe = vec_add(pp1Ae, v512si);


+ 21
- 29
libavcodec/ppc/idct_altivec.c View File

@@ -40,17 +40,9 @@
#include "libavcodec/dsputil.h"

#include "gcc_fixes.h"
#include "types_altivec.h"
#include "dsputil_ppc.h"

#define vector_s16_t vector signed short
#define const_vector_s16_t const vector signed short
#define vector_u16_t vector unsigned short
#define vector_s8_t vector signed char
#define vector_u8_t vector unsigned char
#define vector_s32_t vector signed int
#define vector_u32_t vector unsigned int

#define IDCT_HALF \
/* 1st stage */ \
t1 = vec_mradds (a1, vx7, vx1 ); \
@@ -88,11 +80,11 @@


#define IDCT \
vector_s16_t vx0, vx1, vx2, vx3, vx4, vx5, vx6, vx7; \
vector_s16_t vy0, vy1, vy2, vy3, vy4, vy5, vy6, vy7; \
vector_s16_t a0, a1, a2, ma2, c4, mc4, zero, bias; \
vector_s16_t t0, t1, t2, t3, t4, t5, t6, t7, t8; \
vector_u16_t shift; \
vec_s16 vx0, vx1, vx2, vx3, vx4, vx5, vx6, vx7; \
vec_s16 vy0, vy1, vy2, vy3, vy4, vy5, vy6, vy7; \
vec_s16 a0, a1, a2, ma2, c4, mc4, zero, bias; \
vec_s16 t0, t1, t2, t3, t4, t5, t6, t7, t8; \
vec_u16 shift; \
\
c4 = vec_splat (constants[0], 0); \
a0 = vec_splat (constants[0], 1); \
@@ -100,7 +92,7 @@
a2 = vec_splat (constants[0], 3); \
mc4 = vec_splat (constants[0], 4); \
ma2 = vec_splat (constants[0], 5); \
bias = (vector_s16_t)vec_splat ((vector_s32_t)constants[0], 3); \
bias = (vec_s16)vec_splat ((vec_s32)constants[0], 3); \
\
zero = vec_splat_s16 (0); \
shift = vec_splat_u16 (4); \
@@ -156,7 +148,7 @@
vx7 = vec_sra (vy7, shift);


static const_vector_s16_t constants[5] = {
static const vec_s16 constants[5] = {
{23170, 13573, 6518, 21895, -23170, -21895, 32, 31},
{16384, 22725, 21407, 19266, 16384, 19266, 21407, 22725},
{22725, 31521, 29692, 26722, 22725, 26722, 29692, 31521},
@@ -164,10 +156,10 @@ static const_vector_s16_t constants[5] = {
{19266, 26722, 25172, 22654, 19266, 22654, 25172, 26722}
};

void idct_put_altivec(uint8_t* dest, int stride, vector_s16_t* block)
void idct_put_altivec(uint8_t* dest, int stride, vec_s16* block)
{
POWERPC_PERF_DECLARE(altivec_idct_put_num, 1);
vector_u8_t tmp;
vec_u8 tmp;

#ifdef CONFIG_POWERPC_PERF
POWERPC_PERF_START_COUNT(altivec_idct_put_num, 1);
@@ -176,8 +168,8 @@ POWERPC_PERF_START_COUNT(altivec_idct_put_num, 1);

#define COPY(dest,src) \
tmp = vec_packsu (src, src); \
vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); \
vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest);
vec_ste ((vec_u32)tmp, 0, (unsigned int *)dest); \
vec_ste ((vec_u32)tmp, 4, (unsigned int *)dest);

COPY (dest, vx0) dest += stride;
COPY (dest, vx1) dest += stride;
@@ -191,14 +183,14 @@ POWERPC_PERF_START_COUNT(altivec_idct_put_num, 1);
POWERPC_PERF_STOP_COUNT(altivec_idct_put_num, 1);
}

void idct_add_altivec(uint8_t* dest, int stride, vector_s16_t* block)
void idct_add_altivec(uint8_t* dest, int stride, vec_s16* block)
{
POWERPC_PERF_DECLARE(altivec_idct_add_num, 1);
vector_u8_t tmp;
vector_s16_t tmp2, tmp3;
vector_u8_t perm0;
vector_u8_t perm1;
vector_u8_t p0, p1, p;
vec_u8 tmp;
vec_s16 tmp2, tmp3;
vec_u8 perm0;
vec_u8 perm1;
vec_u8 p0, p1, p;

#ifdef CONFIG_POWERPC_PERF
POWERPC_PERF_START_COUNT(altivec_idct_add_num, 1);
@@ -215,11 +207,11 @@ POWERPC_PERF_START_COUNT(altivec_idct_add_num, 1);
#define ADD(dest,src,perm) \
/* *(uint64_t *)&tmp = *(uint64_t *)dest; */ \
tmp = vec_ld (0, dest); \
tmp2 = (vector_s16_t)vec_perm (tmp, (vector_u8_t)zero, perm); \
tmp2 = (vec_s16)vec_perm (tmp, (vec_u8)zero, perm); \
tmp3 = vec_adds (tmp2, src); \
tmp = vec_packsu (tmp3, tmp3); \
vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); \
vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest);
vec_ste ((vec_u32)tmp, 0, (unsigned int *)dest); \
vec_ste ((vec_u32)tmp, 4, (unsigned int *)dest);

ADD (dest, vx0, perm0) dest += stride;
ADD (dest, vx1, perm1) dest += stride;


+ 8
- 8
libavcodec/ppc/int_altivec.c View File

@@ -79,10 +79,10 @@ static int ssd_int8_vs_int16_altivec(const int8_t *pix1, const int16_t *pix2,
static void add_int16_altivec(int16_t * v1, int16_t * v2, int order)
{
int i;
register vec_s16_t vec, *pv;
register vec_s16 vec, *pv;

for(i = 0; i < order; i += 8){
pv = (vec_s16_t*)v2;
pv = (vec_s16*)v2;
vec = vec_perm(pv[0], pv[1], vec_lvsl(0, v2));
vec_st(vec_add(vec_ld(0, v1), vec), 0, v1);
v1 += 8;
@@ -93,10 +93,10 @@ static void add_int16_altivec(int16_t * v1, int16_t * v2, int order)
static void sub_int16_altivec(int16_t * v1, int16_t * v2, int order)
{
int i;
register vec_s16_t vec, *pv;
register vec_s16 vec, *pv;

for(i = 0; i < order; i += 8){
pv = (vec_s16_t*)v2;
pv = (vec_s16*)v2;
vec = vec_perm(pv[0], pv[1], vec_lvsl(0, v2));
vec_st(vec_sub(vec_ld(0, v1), vec), 0, v1);
v1 += 8;
@@ -108,9 +108,9 @@ static int32_t scalarproduct_int16_altivec(int16_t * v1, int16_t * v2, int order
{
int i;
LOAD_ZERO;
register vec_s16_t vec1, *pv;
register vec_s32_t res = vec_splat_s32(0), t;
register vec_u32_t shifts;
register vec_s16 vec1, *pv;
register vec_s32 res = vec_splat_s32(0), t;
register vec_u32 shifts;
DECLARE_ALIGNED_16(int32_t, ires);

shifts = zero_u32v;
@@ -121,7 +121,7 @@ static int32_t scalarproduct_int16_altivec(int16_t * v1, int16_t * v2, int order
if(shift & 0x01) shifts = vec_add(shifts, vec_splat_u32(0x01));

for(i = 0; i < order; i += 8){
pv = (vec_s16_t*)v1;
pv = (vec_s16*)v1;
vec1 = vec_perm(pv[0], pv[1], vec_lvsl(0, v1));
t = vec_msum(vec1, vec_ld(0, v2), zero_s32v);
t = vec_sr(t, shifts);


+ 13
- 13
libavcodec/ppc/types_altivec.h View File

@@ -24,23 +24,23 @@
/***********************************************************************
* Vector types
**********************************************************************/
#define vec_u8_t vector unsigned char
#define vec_s8_t vector signed char
#define vec_u16_t vector unsigned short
#define vec_s16_t vector signed short
#define vec_u32_t vector unsigned int
#define vec_s32_t vector signed int
#define vec_u8 vector unsigned char
#define vec_s8 vector signed char
#define vec_u16 vector unsigned short
#define vec_s16 vector signed short
#define vec_u32 vector unsigned int
#define vec_s32 vector signed int

/***********************************************************************
* Null vector
**********************************************************************/
#define LOAD_ZERO const vec_u8_t zerov = vec_splat_u8( 0 )
#define LOAD_ZERO const vec_u8 zerov = vec_splat_u8( 0 )

#define zero_u8v (vec_u8_t) zerov
#define zero_s8v (vec_s8_t) zerov
#define zero_u16v (vec_u16_t) zerov
#define zero_s16v (vec_s16_t) zerov
#define zero_u32v (vec_u32_t) zerov
#define zero_s32v (vec_s32_t) zerov
#define zero_u8v (vec_u8) zerov
#define zero_s8v (vec_s8) zerov
#define zero_u16v (vec_u16) zerov
#define zero_s16v (vec_s16) zerov
#define zero_u32v (vec_u32) zerov
#define zero_s32v (vec_s32) zerov

#endif /* AVCODEC_PPC_TYPES_ALTIVEC_H */

Write Preview
Loading…
Cancel
Save