static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
const uint8_t * src2, int dst_stride,
int src_stride1, int h)
{
int i;
vec_u8 a, b, d, mask1, mask2;
#if HAVE_BIGENDIAN
vec_u8 tmp1, tmp2, mask, edges, align;
mask1 = vec_lvsl(0, src1);
mask2 = vec_lvsl(0, src2);
mask = vec_lvsl(0, dst);
align = vec_lvsr(0, dst);
#endif
for (i = 0; i < h; i++) {
a = load_with_perm_vec(i * src_stride1, src1, mask1);
b = load_with_perm_vec(i * 16, src2, mask2);
d = vec_avg(a, b);
put_unaligned_store_with_mask_align(d, dst, mask, align);
dst += dst_stride;
}
}
static inline void avg_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
const uint8_t * src2, int dst_stride,
int src_stride1, int h)
{
int i;
vec_u8 a, b, d, mask_;
#if HAVE_BIGENDIAN
vec_u8 tmp1, tmp2, mask, edges, align;
mask_ = vec_lvsl(0, src2);
#endif
for (i = 0; i < h; i++) {
a = unaligned_load(i * src_stride1, src1);
b = load_with_perm_vec(i * 16, src2, mask_);
d = vec_avg(a, b);
avg_unligned_store(d, dst);
dst += dst_stride;
}
}
static av_always_inline
void put_vp8_epel_v_altivec_core(uint8_t *dst, ptrdiff_t dst_stride,
uint8_t *src, ptrdiff_t src_stride,
int h, int my, int w, int is6tap)
{
LOAD_V_SUBPEL_FILTER(my-1);
vec_u8 s0, s1, s2, s3, s4, s5, filt, align_vech, perm_vec, align_vecl;
vec_s16 s0f, s1f, s2f, s3f, s4f, s5f, f16h, f16l;
vec_s16 c64 = vec_sl(vec_splat_s16(1), vec_splat_u16(6));
vec_u16 c7 = vec_splat_u16(7);
// we want pixels 0-7 to be in the even positions and 8-15 in the odd,
// so combine this permute with the alignment permute vector
align_vech = vec_lvsl(0, src);
align_vecl = vec_sld(align_vech, align_vech, 8);
if (w ==16)
perm_vec = vec_mergeh(align_vech, align_vecl);
else
perm_vec = vec_mergeh(align_vech, align_vech);
if (is6tap)
s0 = load_with_perm_vec(-2*src_stride, src, perm_vec);
s1 = load_with_perm_vec(-1*src_stride, src, perm_vec);
s2 = load_with_perm_vec( 0*src_stride, src, perm_vec);
s3 = load_with_perm_vec( 1*src_stride, src, perm_vec);
if (is6tap)
s4 = load_with_perm_vec( 2*src_stride, src, perm_vec);
src += (2+is6tap)*src_stride;
while (h --> 0) {
if (is6tap)
s5 = load_with_perm_vec(0, src, perm_vec);
else
s4 = load_with_perm_vec(0, src, perm_vec);
FILTER_V(f16h, vec_mule);
if (w == 16) {
FILTER_V(f16l, vec_mulo);
filt = vec_packsu(f16h, f16l);
vec_st(filt, 0, dst);
} else {
filt = vec_packsu(f16h, f16h);
if (w == 4)
filt = (vec_u8)vec_splat((vec_u32)filt, 0);
else
vec_ste((vec_u32)filt, 4, (uint32_t*)dst);
vec_ste((vec_u32)filt, 0, (uint32_t*)dst);
}
if (is6tap)
s0 = s1;
s1 = s2;
s2 = s3;
s3 = s4;
if (is6tap)
s4 = s5;
dst += dst_stride;
src += src_stride;
}
}
static void PREFIX_h264_qpel16_v_lowpass_altivec(uint8_t *dst,
const uint8_t *src,
int dstStride, int srcStride)
{
register int i;
LOAD_ZERO;
vec_u8 perm;
#if HAVE_BIGENDIAN
perm = vec_lvsl(0, src);
#endif
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));
const uint8_t *srcbis = src - (srcStride * 2);
const vec_u8 srcM2 = load_with_perm_vec(0, srcbis, perm);
srcbis += srcStride;
const vec_u8 srcM1 = load_with_perm_vec(0, srcbis, perm);
srcbis += srcStride;
const vec_u8 srcP0 = load_with_perm_vec(0, srcbis, perm);
srcbis += srcStride;
const vec_u8 srcP1 = load_with_perm_vec(0, srcbis, perm);
srcbis += srcStride;
const vec_u8 srcP2 = load_with_perm_vec(0, srcbis, perm);
srcbis += srcStride;
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 sum, fsum, srcP3;
for (i = 0 ; i < 16 ; i++) {
srcP3 = load_with_perm_vec(0, srcbis, perm);
srcbis += srcStride;
srcP3ssA = (vec_s16) VEC_MERGEH(zero_u8v, srcP3);
srcP3ssB = (vec_s16) VEC_MERGEL(zero_u8v, srcP3);
sum1A = vec_adds(srcP0ssA, srcP1ssA);
sum1B = vec_adds(srcP0ssB, srcP1ssB);
sum2A = vec_adds(srcM1ssA, srcP2ssA);
sum2B = vec_adds(srcM1ssB, srcP2ssB);
sum3A = vec_adds(srcM2ssA, srcP3ssA);
sum3B = vec_adds(srcM2ssB, srcP3ssB);
srcM2ssA = srcM1ssA;
srcM2ssB = srcM1ssB;
srcM1ssA = srcP0ssA;
srcM1ssB = srcP0ssB;
srcP0ssA = srcP1ssA;
srcP0ssB = srcP1ssB;
srcP1ssA = srcP2ssA;
srcP1ssB = srcP2ssB;
srcP2ssA = srcP3ssA;
srcP2ssB = srcP3ssB;
pp1A = vec_mladd(sum1A, v20ss, v16ss);
pp1B = vec_mladd(sum1B, v20ss, v16ss);
pp2A = vec_mladd(sum2A, v5ss, zero_s16v);
pp2B = vec_mladd(sum2B, v5ss, zero_s16v);
pp3A = vec_add(sum3A, pp1A);
pp3B = vec_add(sum3B, pp1B);
psumA = vec_sub(pp3A, pp2A);
psumB = vec_sub(pp3B, pp2B);
sumA = vec_sra(psumA, v5us);
sumB = vec_sra(psumB, v5us);
sum = vec_packsu(sumA, sumB);
ASSERT_ALIGNED(dst);
OP_U8_ALTIVEC(fsum, sum, vec_ld(0, dst));
vec_st(fsum, 0, dst);
dst += dstStride;
}
}
