static void PREFIX_h264_qpel16_h_lowpass_altivec(uint8_t * dst, uint8_t * src, int dstStride, int srcStride) {
register int i;
LOAD_ZERO;
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 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 sum, fsum;
for (i = 0 ; i < 16 ; i ++) {
vec_u8 srcR1 = vec_ld(-2, src);
vec_u8 srcR2 = vec_ld(14, src);
switch (align) {
default: {
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
srcP1 = vec_perm(srcR1, srcR2, permP1);
srcP2 = vec_perm(srcR1, srcR2, permP2);
srcP3 = vec_perm(srcR1, srcR2, permP3);
} break;
case 11: {
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
srcP1 = vec_perm(srcR1, srcR2, permP1);
srcP2 = vec_perm(srcR1, srcR2, permP2);
srcP3 = srcR2;
} break;
case 12: {
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);
srcP1 = vec_perm(srcR1, srcR2, permP1);
srcP2 = srcR2;
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 13: {
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);
srcP1 = srcR2;
srcP2 = vec_perm(srcR2, srcR3, permP2);
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 14: {
vec_u8 srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = srcR2;
srcP1 = vec_perm(srcR2, srcR3, permP1);
srcP2 = vec_perm(srcR2, srcR3, permP2);
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 15: {
vec_u8 srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = srcR2;
srcP0 = vec_perm(srcR2, srcR3, permP0);
srcP1 = vec_perm(srcR2, srcR3, permP1);
srcP2 = vec_perm(srcR2, srcR3, permP2);
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
}
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) 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) 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);
sum2A = vec_adds(srcM1A, srcP2A);
sum2B = vec_adds(srcM1B, srcP2B);
sum3A = vec_adds(srcM2A, srcP3A);
sum3B = vec_adds(srcM2B, srcP3B);
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);
src += srcStride;
dst += dstStride;
}
}
static void PREFIX_h264_qpel16_v_lowpass_altivec(uint8_t * dst, uint8_t * src, int dstStride, int srcStride) {
register int i;
LOAD_ZERO;
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 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 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 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 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 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 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, srcP3a, srcP3b, srcP3;
for (i = 0 ; i < 16 ; i++) {
srcP3a = vec_ld(0, srcbis += srcStride);
srcP3b = vec_ld(16, srcbis);
srcP3 = vec_perm(srcP3a, srcP3b, perm);
srcP3ssA = (vec_s16) vec_mergeh(zero_u8v, srcP3);
srcP3ssB = (vec_s16) vec_mergel(zero_u8v, srcP3);
//srcbis += srcStride;
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;
}
}
static void PREFIX_h264_qpel16_hv_lowpass_altivec(uint8_t * dst, int16_t * tmp, uint8_t * src, int dstStride, int tmpStride, int srcStride) {
register int i;
LOAD_ZERO;
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 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 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 tmpM1ssA, tmpM1ssB, tmpM2ssA, tmpM2ssB,
tmpP0ssA, tmpP0ssB, tmpP1ssA, tmpP1ssB,
tmpP2ssA, tmpP2ssB;
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 fsum, sumv, sum;
vec_s16 ssume, ssumo;
src -= (2 * srcStride);
for (i = 0 ; i < 21 ; i ++) {
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: {
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
srcP1 = vec_perm(srcR1, srcR2, permP1);
srcP2 = vec_perm(srcR1, srcR2, permP2);
srcP3 = vec_perm(srcR1, srcR2, permP3);
} break;
case 11: {
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
srcP1 = vec_perm(srcR1, srcR2, permP1);
srcP2 = vec_perm(srcR1, srcR2, permP2);
srcP3 = srcR2;
} break;
case 12: {
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);
srcP1 = vec_perm(srcR1, srcR2, permP1);
srcP2 = srcR2;
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 13: {
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);
srcP1 = srcR2;
srcP2 = vec_perm(srcR2, srcR3, permP2);
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 14: {
vec_u8 srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = srcR2;
srcP1 = vec_perm(srcR2, srcR3, permP1);
srcP2 = vec_perm(srcR2, srcR3, permP2);
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 15: {
vec_u8 srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = srcR2;
srcP0 = vec_perm(srcR2, srcR3, permP0);
srcP1 = vec_perm(srcR2, srcR3, permP1);
srcP2 = vec_perm(srcR2, srcR3, permP2);
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
}
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) 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) 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);
sum2A = vec_adds(srcM1A, srcP2A);
sum2B = vec_adds(srcM1B, srcP2B);
sum3A = vec_adds(srcM2A, srcP3A);
sum3B = vec_adds(srcM2B, srcP3B);
pp1A = vec_mladd(sum1A, v20ss, sum3A);
pp1B = vec_mladd(sum1B, v20ss, sum3B);
pp2A = vec_mladd(sum2A, v5ss, zero_s16v);
pp2B = vec_mladd(sum2B, v5ss, zero_s16v);
psumA = vec_sub(pp1A, pp2A);
psumB = vec_sub(pp1B, pp2B);
vec_st(psumA, 0, tmp);
vec_st(psumB, 16, tmp);
src += srcStride;
tmp += tmpStride; /* int16_t*, and stride is 16, so it's OK here */
}
tmpM2ssA = vec_ld(0, tmpbis);
tmpM2ssB = vec_ld(16, tmpbis);
tmpbis += tmpStride;
tmpM1ssA = vec_ld(0, tmpbis);
tmpM1ssB = vec_ld(16, tmpbis);
tmpbis += tmpStride;
tmpP0ssA = vec_ld(0, tmpbis);
tmpP0ssB = vec_ld(16, tmpbis);
tmpbis += tmpStride;
tmpP1ssA = vec_ld(0, tmpbis);
tmpP1ssB = vec_ld(16, tmpbis);
tmpbis += tmpStride;
tmpP2ssA = vec_ld(0, tmpbis);
tmpP2ssB = vec_ld(16, tmpbis);
tmpbis += tmpStride;
for (i = 0 ; i < 16 ; i++) {
const vec_s16 tmpP3ssA = vec_ld(0, tmpbis);
const vec_s16 tmpP3ssB = vec_ld(16, tmpbis);
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;
tmpM2ssA = tmpM1ssA;
tmpM2ssB = tmpM1ssB;
tmpM1ssA = tmpP0ssA;
tmpM1ssB = tmpP0ssB;
tmpP0ssA = tmpP1ssA;
tmpP0ssB = tmpP1ssB;
tmpP1ssA = tmpP2ssA;
tmpP1ssB = tmpP2ssB;
tmpP2ssA = tmpP3ssA;
tmpP2ssB = tmpP3ssB;
pp1Ae = vec_mule(sum1A, v20ss);
pp1Ao = vec_mulo(sum1A, v20ss);
pp1Be = vec_mule(sum1B, v20ss);
pp1Bo = vec_mulo(sum1B, v20ss);
pp2Ae = vec_mule(sum2A, v5ss);
pp2Ao = vec_mulo(sum2A, v5ss);
pp2Be = vec_mule(sum2B, v5ss);
pp2Bo = vec_mulo(sum2B, v5ss);
pp3Ae = vec_sra((vec_s32)sum3A, v16ui);
pp3Ao = vec_mulo(sum3A, v1ss);
pp3Be = vec_sra((vec_s32)sum3B, v16ui);
pp3Bo = vec_mulo(sum3B, v1ss);
pp1cAe = vec_add(pp1Ae, v512si);
pp1cAo = vec_add(pp1Ao, v512si);
pp1cBe = vec_add(pp1Be, v512si);
pp1cBo = vec_add(pp1Bo, v512si);
pp32Ae = vec_sub(pp3Ae, pp2Ae);
pp32Ao = vec_sub(pp3Ao, pp2Ao);
pp32Be = vec_sub(pp3Be, pp2Be);
pp32Bo = vec_sub(pp3Bo, pp2Bo);
sumAe = vec_add(pp1cAe, pp32Ae);
sumAo = vec_add(pp1cAo, pp32Ao);
sumBe = vec_add(pp1cBe, pp32Be);
sumBo = vec_add(pp1cBo, pp32Bo);
ssumAe = vec_sra(sumAe, v10ui);
ssumAo = vec_sra(sumAo, v10ui);
ssumBe = vec_sra(sumBe, v10ui);
ssumBo = vec_sra(sumBo, v10ui);
ssume = vec_packs(ssumAe, ssumBe);
ssumo = vec_packs(ssumAo, ssumBo);
sumv = vec_packsu(ssume, ssumo);
sum = vec_perm(sumv, sumv, mperm);
ASSERT_ALIGNED(dst);
OP_U8_ALTIVEC(fsum, sum, vec_ld(0, dst));
vec_st(fsum, 0, dst);
dst += dstStride;
}
}
/* this code assume stride % 16 == 0 *and* tmp is properly aligned */
static void PREFIX_h264_qpel16_hv_lowpass_altivec(uint8_t * dst, int16_t * tmp, uint8_t * src, int dstStride, int tmpStride, int srcStride) {
POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_hv_lowpass_num, 1);
POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_hv_lowpass_num, 1);
register int i;
const vector signed int vzero = vec_splat_s32(0);
const vector unsigned char permM2 = vec_lvsl(-2, src);
const vector unsigned char permM1 = vec_lvsl(-1, src);
const vector unsigned char permP0 = vec_lvsl(+0, src);
const vector unsigned char permP1 = vec_lvsl(+1, src);
const vector unsigned char permP2 = vec_lvsl(+2, src);
const vector unsigned char permP3 = vec_lvsl(+3, src);
const vector signed short v20ss = (const vector signed short)AVV(20);
const vector unsigned int v10ui = vec_splat_u32(10);
const vector signed short v5ss = vec_splat_s16(5);
const vector signed short v1ss = vec_splat_s16(1);
const vector signed int v512si = (const vector signed int)AVV(512);
const vector unsigned int v16ui = (const vector unsigned int)AVV(16);
register int align = ((((unsigned long)src) - 2) % 16);
src -= (2 * srcStride);
for (i = 0 ; i < 21 ; i ++) {
vector unsigned char srcM2, srcM1, srcP0, srcP1, srcP2, srcP3;
vector unsigned char srcR1 = vec_ld(-2, src);
vector unsigned char srcR2 = vec_ld(14, src);
switch (align) {
default: {
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
srcP1 = vec_perm(srcR1, srcR2, permP1);
srcP2 = vec_perm(srcR1, srcR2, permP2);
srcP3 = vec_perm(srcR1, srcR2, permP3);
} break;
case 11: {
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
srcP1 = vec_perm(srcR1, srcR2, permP1);
srcP2 = vec_perm(srcR1, srcR2, permP2);
srcP3 = srcR2;
} break;
case 12: {
vector unsigned char srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
srcP1 = vec_perm(srcR1, srcR2, permP1);
srcP2 = srcR2;
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 13: {
vector unsigned char srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
srcP1 = srcR2;
srcP2 = vec_perm(srcR2, srcR3, permP2);
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 14: {
vector unsigned char srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = srcR2;
srcP1 = vec_perm(srcR2, srcR3, permP1);
srcP2 = vec_perm(srcR2, srcR3, permP2);
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 15: {
vector unsigned char srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = srcR2;
srcP0 = vec_perm(srcR2, srcR3, permP0);
srcP1 = vec_perm(srcR2, srcR3, permP1);
srcP2 = vec_perm(srcR2, srcR3, permP2);
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
}
const vector signed short srcP0A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP0);
const vector signed short srcP0B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP0);
const vector signed short srcP1A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP1);
const vector signed short srcP1B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP1);
const vector signed short srcP2A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP2);
const vector signed short srcP2B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP2);
const vector signed short srcP3A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP3);
const vector signed short srcP3B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP3);
const vector signed short srcM1A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM1);
const vector signed short srcM1B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM1);
const vector signed short srcM2A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM2);
const vector signed short srcM2B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM2);
const vector signed short sum1A = vec_adds(srcP0A, srcP1A);
const vector signed short sum1B = vec_adds(srcP0B, srcP1B);
const vector signed short sum2A = vec_adds(srcM1A, srcP2A);
const vector signed short sum2B = vec_adds(srcM1B, srcP2B);
const vector signed short sum3A = vec_adds(srcM2A, srcP3A);
const vector signed short sum3B = vec_adds(srcM2B, srcP3B);
const vector signed short pp1A = vec_mladd(sum1A, v20ss, sum3A);
const vector signed short pp1B = vec_mladd(sum1B, v20ss, sum3B);
const vector signed short pp2A = vec_mladd(sum2A, v5ss, (vector signed short)vzero);
const vector signed short pp2B = vec_mladd(sum2B, v5ss, (vector signed short)vzero);
const vector signed short psumA = vec_sub(pp1A, pp2A);
const vector signed short psumB = vec_sub(pp1B, pp2B);
vec_st(psumA, 0, tmp);
vec_st(psumB, 16, tmp);
src += srcStride;
tmp += tmpStride; /* int16_t*, and stride is 16, so it's OK here */
}
const vector unsigned char dstperm = vec_lvsr(0, dst);
const vector unsigned char neg1 = (const vector unsigned char)vec_splat_s8(-1);
const vector unsigned char dstmask = vec_perm((const vector unsigned char)vzero, neg1, dstperm);
const vector unsigned char mperm = (const vector unsigned char)
AVV(0x00, 0x08, 0x01, 0x09, 0x02, 0x0A, 0x03, 0x0B,
0x04, 0x0C, 0x05, 0x0D, 0x06, 0x0E, 0x07, 0x0F);
int16_t *tmpbis = tmp - (tmpStride * 21);
vector signed short tmpM2ssA = vec_ld(0, tmpbis);
vector signed short tmpM2ssB = vec_ld(16, tmpbis);
tmpbis += tmpStride;
vector signed short tmpM1ssA = vec_ld(0, tmpbis);
vector signed short tmpM1ssB = vec_ld(16, tmpbis);
tmpbis += tmpStride;
vector signed short tmpP0ssA = vec_ld(0, tmpbis);
vector signed short tmpP0ssB = vec_ld(16, tmpbis);
tmpbis += tmpStride;
vector signed short tmpP1ssA = vec_ld(0, tmpbis);
vector signed short tmpP1ssB = vec_ld(16, tmpbis);
tmpbis += tmpStride;
vector signed short tmpP2ssA = vec_ld(0, tmpbis);
vector signed short tmpP2ssB = vec_ld(16, tmpbis);
tmpbis += tmpStride;
for (i = 0 ; i < 16 ; i++) {
const vector signed short tmpP3ssA = vec_ld(0, tmpbis);
const vector signed short tmpP3ssB = vec_ld(16, tmpbis);
tmpbis += tmpStride;
const vector signed short sum1A = vec_adds(tmpP0ssA, tmpP1ssA);
const vector signed short sum1B = vec_adds(tmpP0ssB, tmpP1ssB);
const vector signed short sum2A = vec_adds(tmpM1ssA, tmpP2ssA);
const vector signed short sum2B = vec_adds(tmpM1ssB, tmpP2ssB);
const vector signed short sum3A = vec_adds(tmpM2ssA, tmpP3ssA);
const vector signed short sum3B = vec_adds(tmpM2ssB, tmpP3ssB);
tmpM2ssA = tmpM1ssA;
tmpM2ssB = tmpM1ssB;
tmpM1ssA = tmpP0ssA;
tmpM1ssB = tmpP0ssB;
tmpP0ssA = tmpP1ssA;
tmpP0ssB = tmpP1ssB;
tmpP1ssA = tmpP2ssA;
tmpP1ssB = tmpP2ssB;
tmpP2ssA = tmpP3ssA;
tmpP2ssB = tmpP3ssB;
const vector signed int pp1Ae = vec_mule(sum1A, v20ss);
const vector signed int pp1Ao = vec_mulo(sum1A, v20ss);
const vector signed int pp1Be = vec_mule(sum1B, v20ss);
const vector signed int pp1Bo = vec_mulo(sum1B, v20ss);
const vector signed int pp2Ae = vec_mule(sum2A, v5ss);
const vector signed int pp2Ao = vec_mulo(sum2A, v5ss);
const vector signed int pp2Be = vec_mule(sum2B, v5ss);
const vector signed int pp2Bo = vec_mulo(sum2B, v5ss);
const vector signed int pp3Ae = vec_sra((vector signed int)sum3A, v16ui);
const vector signed int pp3Ao = vec_mulo(sum3A, v1ss);
const vector signed int pp3Be = vec_sra((vector signed int)sum3B, v16ui);
const vector signed int pp3Bo = vec_mulo(sum3B, v1ss);
const vector signed int pp1cAe = vec_add(pp1Ae, v512si);
const vector signed int pp1cAo = vec_add(pp1Ao, v512si);
const vector signed int pp1cBe = vec_add(pp1Be, v512si);
const vector signed int pp1cBo = vec_add(pp1Bo, v512si);
const vector signed int pp32Ae = vec_sub(pp3Ae, pp2Ae);
const vector signed int pp32Ao = vec_sub(pp3Ao, pp2Ao);
const vector signed int pp32Be = vec_sub(pp3Be, pp2Be);
const vector signed int pp32Bo = vec_sub(pp3Bo, pp2Bo);
const vector signed int sumAe = vec_add(pp1cAe, pp32Ae);
const vector signed int sumAo = vec_add(pp1cAo, pp32Ao);
const vector signed int sumBe = vec_add(pp1cBe, pp32Be);
const vector signed int sumBo = vec_add(pp1cBo, pp32Bo);
const vector signed int ssumAe = vec_sra(sumAe, v10ui);
const vector signed int ssumAo = vec_sra(sumAo, v10ui);
const vector signed int ssumBe = vec_sra(sumBe, v10ui);
const vector signed int ssumBo = vec_sra(sumBo, v10ui);
const vector signed short ssume = vec_packs(ssumAe, ssumBe);
const vector signed short ssumo = vec_packs(ssumAo, ssumBo);
const vector unsigned char sumv = vec_packsu(ssume, ssumo);
const vector unsigned char sum = vec_perm(sumv, sumv, mperm);
const vector unsigned char dst1 = vec_ld(0, dst);
const vector unsigned char dst2 = vec_ld(16, dst);
const vector unsigned char vdst = vec_perm(dst1, dst2, vec_lvsl(0, dst));
vector unsigned char fsum;
OP_U8_ALTIVEC(fsum, sum, vdst);
const vector unsigned char rsum = vec_perm(fsum, fsum, dstperm);
const vector unsigned char fdst1 = vec_sel(dst1, rsum, dstmask);
const vector unsigned char fdst2 = vec_sel(rsum, dst2, dstmask);
vec_st(fdst1, 0, dst);
vec_st(fdst2, 16, dst);
dst += dstStride;
}
POWERPC_PERF_STOP_COUNT(PREFIX_h264_qpel16_hv_lowpass_num, 1);
}
/* this code assume that stride % 16 == 0 */
void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride, int h, int x, int y) {
POWERPC_PERF_DECLARE(PREFIX_h264_chroma_mc8_num, 1);
DECLARE_ALIGNED_16(signed int, ABCD[4]) =
{((8 - x) * (8 - y)),
((x) * (8 - y)),
((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);
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);
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;
POWERPC_PERF_START_COUNT(PREFIX_h264_chroma_mc8_num, 1);
if (((unsigned long)dst) % 16 == 0) {
fperm = (vec_u8_t)AVV(0x10, 0x11, 0x12, 0x13,
0x14, 0x15, 0x16, 0x17,
0x08, 0x09, 0x0A, 0x0B,
0x0C, 0x0D, 0x0E, 0x0F);
} else {
fperm = (vec_u8_t)AVV(0x00, 0x01, 0x02, 0x03,
0x04, 0x05, 0x06, 0x07,
0x18, 0x19, 0x1A, 0x1B,
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_t)vec_mergeh(zero_u8v,(vec_u8_t)vsrc0uc);
vsrc1ssH = (vec_s16_t)vec_mergeh(zero_u8v,(vec_u8_t)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);
vsrc2ssH = (vec_s16_t)vec_mergeh(zero_u8v,(vec_u8_t)vsrc2uc);
vsrc3ssH = (vec_s16_t)vec_mergeh(zero_u8v,(vec_u8_t)vsrc3uc);
psum = vec_mladd(vA, vsrc0ssH, vec_splat_s16(0));
psum = vec_mladd(vB, vsrc1ssH, psum);
psum = vec_mladd(vC, vsrc2ssH, psum);
psum = vec_mladd(vD, vsrc3ssH, psum);
psum = vec_add(v32ss, psum);
psum = vec_sra(psum, v6us);
vdst = vec_ld(0, dst);
ppsum = (vec_u8_t)vec_packsu(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;
}
} else {
vec_u8_t 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);
vsrc2ssH = (vec_s16_t)vec_mergeh(zero_u8v,(vec_u8_t)vsrc2uc);
vsrc3ssH = (vec_s16_t)vec_mergeh(zero_u8v,(vec_u8_t)vsrc3uc);
psum = vec_mladd(vA, vsrc0ssH, vec_splat_s16(0));
psum = vec_mladd(vB, vsrc1ssH, psum);
psum = vec_mladd(vC, vsrc2ssH, psum);
psum = vec_mladd(vD, vsrc3ssH, psum);
psum = vec_add(v32ss, psum);
psum = vec_sr(psum, v6us);
vdst = vec_ld(0, dst);
ppsum = (vec_u8_t)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;
}
}
POWERPC_PERF_STOP_COUNT(PREFIX_h264_chroma_mc8_num, 1);
}
/* this code assume stride % 16 == 0 */
static void PREFIX_h264_qpel16_v_lowpass_altivec(uint8_t * dst, uint8_t * src, int dstStride, int srcStride) {
POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_v_lowpass_num, 1);
POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_v_lowpass_num, 1);
register int i;
const vector signed int vzero = vec_splat_s32(0);
const vector unsigned char perm = vec_lvsl(0, src);
const vector signed short v20ss = (const vector signed short)AVV(20);
const vector unsigned short v5us = vec_splat_u16(5);
const vector signed short v5ss = vec_splat_s16(5);
const vector signed short v16ss = (const vector signed short)AVV(16);
const vector unsigned char dstperm = vec_lvsr(0, dst);
const vector unsigned char neg1 = (const vector unsigned char)vec_splat_s8(-1);
const vector unsigned char dstmask = vec_perm((const vector unsigned char)vzero, neg1, dstperm);
uint8_t *srcbis = src - (srcStride * 2);
const vector unsigned char srcM2a = vec_ld(0, srcbis);
const vector unsigned char srcM2b = vec_ld(16, srcbis);
const vector unsigned char srcM2 = vec_perm(srcM2a, srcM2b, perm);
srcbis += srcStride;
const vector unsigned char srcM1a = vec_ld(0, srcbis);
const vector unsigned char srcM1b = vec_ld(16, srcbis);
const vector unsigned char srcM1 = vec_perm(srcM1a, srcM1b, perm);
srcbis += srcStride;
const vector unsigned char srcP0a = vec_ld(0, srcbis);
const vector unsigned char srcP0b = vec_ld(16, srcbis);
const vector unsigned char srcP0 = vec_perm(srcP0a, srcP0b, perm);
srcbis += srcStride;
const vector unsigned char srcP1a = vec_ld(0, srcbis);
const vector unsigned char srcP1b = vec_ld(16, srcbis);
const vector unsigned char srcP1 = vec_perm(srcP1a, srcP1b, perm);
srcbis += srcStride;
const vector unsigned char srcP2a = vec_ld(0, srcbis);
const vector unsigned char srcP2b = vec_ld(16, srcbis);
const vector unsigned char srcP2 = vec_perm(srcP2a, srcP2b, perm);
srcbis += srcStride;
vector signed short srcM2ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM2);
vector signed short srcM2ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM2);
vector signed short srcM1ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM1);
vector signed short srcM1ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM1);
vector signed short srcP0ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP0);
vector signed short srcP0ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP0);
vector signed short srcP1ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP1);
vector signed short srcP1ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP1);
vector signed short srcP2ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP2);
vector signed short srcP2ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP2);
for (i = 0 ; i < 16 ; i++) {
const vector unsigned char srcP3a = vec_ld(0, srcbis);
const vector unsigned char srcP3b = vec_ld(16, srcbis);
const vector unsigned char srcP3 = vec_perm(srcP3a, srcP3b, perm);
const vector signed short srcP3ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP3);
const vector signed short srcP3ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP3);
srcbis += srcStride;
const vector signed short sum1A = vec_adds(srcP0ssA, srcP1ssA);
const vector signed short sum1B = vec_adds(srcP0ssB, srcP1ssB);
const vector signed short sum2A = vec_adds(srcM1ssA, srcP2ssA);
const vector signed short sum2B = vec_adds(srcM1ssB, srcP2ssB);
const vector signed short sum3A = vec_adds(srcM2ssA, srcP3ssA);
const vector signed short 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;
const vector signed short pp1A = vec_mladd(sum1A, v20ss, v16ss);
const vector signed short pp1B = vec_mladd(sum1B, v20ss, v16ss);
const vector signed short pp2A = vec_mladd(sum2A, v5ss, (vector signed short)vzero);
const vector signed short pp2B = vec_mladd(sum2B, v5ss, (vector signed short)vzero);
const vector signed short pp3A = vec_add(sum3A, pp1A);
const vector signed short pp3B = vec_add(sum3B, pp1B);
const vector signed short psumA = vec_sub(pp3A, pp2A);
const vector signed short psumB = vec_sub(pp3B, pp2B);
const vector signed short sumA = vec_sra(psumA, v5us);
const vector signed short sumB = vec_sra(psumB, v5us);
const vector unsigned char sum = vec_packsu(sumA, sumB);
const vector unsigned char dst1 = vec_ld(0, dst);
const vector unsigned char dst2 = vec_ld(16, dst);
const vector unsigned char vdst = vec_perm(dst1, dst2, vec_lvsl(0, dst));
vector unsigned char fsum;
OP_U8_ALTIVEC(fsum, sum, vdst);
const vector unsigned char rsum = vec_perm(fsum, fsum, dstperm);
const vector unsigned char fdst1 = vec_sel(dst1, rsum, dstmask);
const vector unsigned char fdst2 = vec_sel(rsum, dst2, dstmask);
vec_st(fdst1, 0, dst);
vec_st(fdst2, 16, dst);
dst += dstStride;
}
POWERPC_PERF_STOP_COUNT(PREFIX_h264_qpel16_v_lowpass_num, 1);
}
/* this code assume that stride % 16 == 0 */
void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride, int h, int x, int y) {
POWERPC_PERF_DECLARE(PREFIX_h264_chroma_mc8_num, 1);
POWERPC_PERF_START_COUNT(PREFIX_h264_chroma_mc8_num, 1);
signed int ABCD[4] __attribute__((aligned(16)));
register int i;
ABCD[0] = ((8 - x) * (8 - y));
ABCD[1] = ((x) * (8 - y));
ABCD[2] = ((8 - x) * (y));
ABCD[3] = ((x) * (y));
const vector signed int vABCD = vec_ld(0, ABCD);
const vector signed short vA = vec_splat((vector signed short)vABCD, 1);
const vector signed short vB = vec_splat((vector signed short)vABCD, 3);
const vector signed short vC = vec_splat((vector signed short)vABCD, 5);
const vector signed short vD = vec_splat((vector signed short)vABCD, 7);
const vector signed int vzero = vec_splat_s32(0);
const vector signed short v32ss = (const vector signed short)AVV(32);
const vector unsigned short v6us = vec_splat_u16(6);
vector unsigned char fperm;
if (((unsigned long)dst) % 16 == 0) {
fperm = (vector unsigned char)AVV(0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17,
0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E, 0x0F);
} else {
fperm = (vector unsigned char)AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F);
}
register int loadSecond = (((unsigned long)src) % 16) <= 7 ? 0 : 1;
register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0;
vector unsigned char vsrcAuc;
vector unsigned char vsrcBuc;
vector unsigned char vsrcperm0;
vector unsigned char vsrcperm1;
vsrcAuc = vec_ld(0, src);
if (loadSecond)
vsrcBuc = vec_ld(16, src);
vsrcperm0 = vec_lvsl(0, src);
vsrcperm1 = vec_lvsl(1, src);
vector unsigned char vsrc0uc;
vector unsigned char vsrc1uc;
vsrc0uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm0);
if (reallyBadAlign)
vsrc1uc = vsrcBuc;
else
vsrc1uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm1);
vector signed short vsrc0ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero, (vector unsigned char)vsrc0uc);
vector signed short vsrc1ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero, (vector unsigned char)vsrc1uc);
if (!loadSecond) {// -> !reallyBadAlign
for (i = 0 ; i < h ; i++) {
vector unsigned char vsrcCuc;
vsrcCuc = vec_ld(stride + 0, src);
vector unsigned char vsrc2uc;
vector unsigned char vsrc3uc;
vsrc2uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0);
vsrc3uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm1);
vector signed short vsrc2ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero, (vector unsigned char)vsrc2uc);
vector signed short vsrc3ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero, (vector unsigned char)vsrc3uc);
vector signed short psum;
psum = vec_mladd(vA, vsrc0ssH, vec_splat_s16(0));
psum = vec_mladd(vB, vsrc1ssH, psum);
psum = vec_mladd(vC, vsrc2ssH, psum);
psum = vec_mladd(vD, vsrc3ssH, psum);
psum = vec_add(v32ss, psum);
psum = vec_sra(psum, v6us);
vector unsigned char vdst = vec_ld(0, dst);
vector unsigned char ppsum = (vector unsigned char)vec_packsu(psum, psum);
vector unsigned char vfdst = vec_perm(vdst, ppsum, fperm);
vector unsigned char fsum;
OP_U8_ALTIVEC(fsum, vfdst, vdst);
vec_st(fsum, 0, dst);
vsrc0ssH = vsrc2ssH;
vsrc1ssH = vsrc3ssH;
dst += stride;
src += stride;
}
} else {
for (i = 0 ; i < h ; i++) {
vector unsigned char vsrcCuc;
vector unsigned char vsrcDuc;
vsrcCuc = vec_ld(stride + 0, src);
vsrcDuc = vec_ld(stride + 16, src);
vector unsigned char vsrc2uc;
vector unsigned char vsrc3uc;
vsrc2uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0);
if (reallyBadAlign)
vsrc3uc = vsrcDuc;
else
vsrc3uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm1);
vector signed short vsrc2ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero, (vector unsigned char)vsrc2uc);
vector signed short vsrc3ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero, (vector unsigned char)vsrc3uc);
vector signed short psum;
psum = vec_mladd(vA, vsrc0ssH, vec_splat_s16(0));
psum = vec_mladd(vB, vsrc1ssH, psum);
psum = vec_mladd(vC, vsrc2ssH, psum);
psum = vec_mladd(vD, vsrc3ssH, psum);
psum = vec_add(v32ss, psum);
psum = vec_sr(psum, v6us);
vector unsigned char vdst = vec_ld(0, dst);
vector unsigned char ppsum = (vector unsigned char)vec_pack(psum, psum);
vector unsigned char vfdst = vec_perm(vdst, ppsum, fperm);
vector unsigned char fsum;
OP_U8_ALTIVEC(fsum, vfdst, vdst);
vec_st(fsum, 0, dst);
vsrc0ssH = vsrc2ssH;
vsrc1ssH = vsrc3ssH;
dst += stride;
src += stride;
}
}
POWERPC_PERF_STOP_COUNT(PREFIX_h264_chroma_mc8_num, 1);
}
/* this code assume stride % 16 == 0 */
static void PREFIX_h264_qpel16_h_lowpass_altivec(uint8_t * dst, uint8_t * src, int dstStride, int srcStride) {
POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_h_lowpass_num, 1);
POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_h_lowpass_num, 1);
register int i;
const vector signed int vzero = vec_splat_s32(0);
const vector unsigned char permM2 = vec_lvsl(-2, src);
const vector unsigned char permM1 = vec_lvsl(-1, src);
const vector unsigned char permP0 = vec_lvsl(+0, src);
const vector unsigned char permP1 = vec_lvsl(+1, src);
const vector unsigned char permP2 = vec_lvsl(+2, src);
const vector unsigned char permP3 = vec_lvsl(+3, src);
const vector signed short v20ss = (const vector signed short)AVV(20);
const vector unsigned short v5us = vec_splat_u16(5);
const vector signed short v5ss = vec_splat_s16(5);
const vector signed short v16ss = (const vector signed short)AVV(16);
const vector unsigned char dstperm = vec_lvsr(0, dst);
const vector unsigned char neg1 = (const vector unsigned char)vec_splat_s8(-1);
const vector unsigned char dstmask = vec_perm((const vector unsigned char)vzero, neg1, dstperm);
register int align = ((((unsigned long)src) - 2) % 16);
for (i = 0 ; i < 16 ; i ++) {
vector unsigned char srcM2, srcM1, srcP0, srcP1, srcP2, srcP3;
vector unsigned char srcR1 = vec_ld(-2, src);
vector unsigned char srcR2 = vec_ld(14, src);
switch (align) {
default: {
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
srcP1 = vec_perm(srcR1, srcR2, permP1);
srcP2 = vec_perm(srcR1, srcR2, permP2);
srcP3 = vec_perm(srcR1, srcR2, permP3);
} break;
case 11: {
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
srcP1 = vec_perm(srcR1, srcR2, permP1);
srcP2 = vec_perm(srcR1, srcR2, permP2);
srcP3 = srcR2;
} break;
case 12: {
vector unsigned char srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
srcP1 = vec_perm(srcR1, srcR2, permP1);
srcP2 = srcR2;
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 13: {
vector unsigned char srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
srcP1 = srcR2;
srcP2 = vec_perm(srcR2, srcR3, permP2);
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 14: {
vector unsigned char srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = srcR2;
srcP1 = vec_perm(srcR2, srcR3, permP1);
srcP2 = vec_perm(srcR2, srcR3, permP2);
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 15: {
vector unsigned char srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = srcR2;
srcP0 = vec_perm(srcR2, srcR3, permP0);
srcP1 = vec_perm(srcR2, srcR3, permP1);
srcP2 = vec_perm(srcR2, srcR3, permP2);
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
}
const vector signed short srcP0A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP0);
const vector signed short srcP0B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP0);
const vector signed short srcP1A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP1);
const vector signed short srcP1B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP1);
const vector signed short srcP2A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP2);
const vector signed short srcP2B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP2);
const vector signed short srcP3A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP3);
const vector signed short srcP3B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP3);
const vector signed short srcM1A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM1);
const vector signed short srcM1B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM1);
const vector signed short srcM2A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM2);
const vector signed short srcM2B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM2);
const vector signed short sum1A = vec_adds(srcP0A, srcP1A);
const vector signed short sum1B = vec_adds(srcP0B, srcP1B);
const vector signed short sum2A = vec_adds(srcM1A, srcP2A);
const vector signed short sum2B = vec_adds(srcM1B, srcP2B);
const vector signed short sum3A = vec_adds(srcM2A, srcP3A);
const vector signed short sum3B = vec_adds(srcM2B, srcP3B);
const vector signed short pp1A = vec_mladd(sum1A, v20ss, v16ss);
const vector signed short pp1B = vec_mladd(sum1B, v20ss, v16ss);
const vector signed short pp2A = vec_mladd(sum2A, v5ss, (vector signed short)vzero);
const vector signed short pp2B = vec_mladd(sum2B, v5ss, (vector signed short)vzero);
const vector signed short pp3A = vec_add(sum3A, pp1A);
const vector signed short pp3B = vec_add(sum3B, pp1B);
const vector signed short psumA = vec_sub(pp3A, pp2A);
const vector signed short psumB = vec_sub(pp3B, pp2B);
const vector signed short sumA = vec_sra(psumA, v5us);
const vector signed short sumB = vec_sra(psumB, v5us);
const vector unsigned char sum = vec_packsu(sumA, sumB);
const vector unsigned char dst1 = vec_ld(0, dst);
const vector unsigned char dst2 = vec_ld(16, dst);
const vector unsigned char vdst = vec_perm(dst1, dst2, vec_lvsl(0, dst));
vector unsigned char fsum;
OP_U8_ALTIVEC(fsum, sum, vdst);
const vector unsigned char rsum = vec_perm(fsum, fsum, dstperm);
const vector unsigned char fdst1 = vec_sel(dst1, rsum, dstmask);
const vector unsigned char fdst2 = vec_sel(rsum, dst2, dstmask);
vec_st(fdst1, 0, dst);
vec_st(fdst2, 16, dst);
src += srcStride;
dst += dstStride;
}
POWERPC_PERF_STOP_COUNT(PREFIX_h264_qpel16_h_lowpass_num, 1);
}
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;
}
}
static void PREFIX_h264_qpel16_h_lowpass_altivec(uint8_t *dst,
const uint8_t *src,
int dstStride, int srcStride)
{
register int i;
LOAD_ZERO;
vec_u8 permM2, permM1, permP0, permP1, permP2, permP3;
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 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 sum, fsum;
#if HAVE_BIGENDIAN
permM2 = vec_lvsl(-2, src);
permM1 = vec_lvsl(-1, src);
permP0 = vec_lvsl(+0, src);
permP1 = vec_lvsl(+1, src);
permP2 = vec_lvsl(+2, src);
permP3 = vec_lvsl(+3, src);
#endif /* HAVE_BIGENDIAN */
for (i = 0 ; i < 16 ; i ++) {
load_alignment(src, align, permM2, permM1, permP0, permP1, permP2, permP3);
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) 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) 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);
sum2A = vec_adds(srcM1A, srcP2A);
sum2B = vec_adds(srcM1B, srcP2B);
sum3A = vec_adds(srcM2A, srcP3A);
sum3B = vec_adds(srcM2B, srcP3B);
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);
src += srcStride;
dst += dstStride;
}
}