static av_always_inline void h264_idct_dc_add_internal(uint8_t *dst, int16_t *block, int stride, int size)
{
vec_s16 dc16;
vec_u8 dcplus, dcminus, v0, v1, v2, v3, aligner;
vec_s32 v_dc32;
LOAD_ZERO;
DECLARE_ALIGNED(16, int, dc);
int i;
dc = (block[0] + 32) >> 6;
block[0] = 0;
v_dc32 = vec_lde(0, &dc);
dc16 = VEC_SPLAT16((vec_s16)v_dc32, 1);
if (size == 4)
dc16 = VEC_SLD16(dc16, zero_s16v, 8);
dcplus = vec_packsu(dc16, zero_s16v);
dcminus = vec_packsu(vec_sub(zero_s16v, dc16), zero_s16v);
aligner = vec_lvsr(0, dst);
#if !HAVE_BIGENDIAN
aligner = vec_perm(aligner, zero_u8v, vcswapc());
#endif
dcplus = vec_perm(dcplus, dcplus, aligner);
dcminus = vec_perm(dcminus, dcminus, aligner);
for (i = 0; i < size; i += 4) {
v0 = vec_ld(0, dst+0*stride);
v1 = vec_ld(0, dst+1*stride);
v2 = vec_ld(0, dst+2*stride);
v3 = vec_ld(0, dst+3*stride);
v0 = vec_adds(v0, dcplus);
v1 = vec_adds(v1, dcplus);
v2 = vec_adds(v2, dcplus);
v3 = vec_adds(v3, dcplus);
v0 = vec_subs(v0, dcminus);
v1 = vec_subs(v1, dcminus);
v2 = vec_subs(v2, dcminus);
v3 = vec_subs(v3, dcminus);
vec_st(v0, 0, dst+0*stride);
vec_st(v1, 0, dst+1*stride);
vec_st(v2, 0, dst+2*stride);
vec_st(v3, 0, dst+3*stride);
dst += 4*stride;
}
}
SIMD_INLINE void BgrToGray(const Loader<align> & bgr, Storer<align> & gray)
{
v128_u8 _bgr[3];
_bgr[0] = Load<align, first>(bgr);
_bgr[1] = Load<align, false>(bgr);
_bgr[2] = Load<align, false>(bgr);
const v128_u16 lo = vec_packsu(
BgraToGray32(vec_perm(_bgr[0], _bgr[1], K8_PERM_0)),
BgraToGray32(vec_perm(_bgr[0], _bgr[1], K8_PERM_1)));
const v128_u16 hi = vec_packsu(
BgraToGray32(vec_perm(_bgr[1], _bgr[2], K8_PERM_2)),
BgraToGray32(vec_perm(_bgr[1], _bgr[2], K8_PERM_3)));
Store<align, first>(gray, vec_packsu(lo, hi));
}
static void yuv2plane1_8_vsx(const int16_t *src, uint8_t *dest, int dstW,
const uint8_t *dither, int offset)
{
const int dst_u = -(uintptr_t)dest & 15;
int i, j;
LOCAL_ALIGNED(16, int16_t, val, [16]);
const vector uint16_t shifts = (vector uint16_t) {7, 7, 7, 7, 7, 7, 7, 7};
vector int16_t vi, vileft, ditherleft, ditherright;
vector uint8_t vd;
for (j = 0; j < 16; j++) {
val[j] = dither[(dst_u + offset + j) & 7];
}
ditherleft = vec_ld(0, val);
ditherright = vec_ld(0, &val[8]);
yuv2plane1_8_u(src, dest, dst_u, dither, offset, 0);
for (i = dst_u; i < dstW - 15; i += 16) {
vi = vec_vsx_ld(0, &src[i]);
vi = vec_adds(ditherleft, vi);
vileft = vec_sra(vi, shifts);
vi = vec_vsx_ld(0, &src[i + 8]);
vi = vec_adds(ditherright, vi);
vi = vec_sra(vi, shifts);
vd = vec_packsu(vileft, vi);
vec_st(vd, 0, &dest[i]);
}
yuv2plane1_8_u(src, dest, dstW, dither, offset, i);
}
static void yuv2plane1_16_vsx(const int32_t *src, uint16_t *dest, int dstW,
int big_endian, int output_bits)
{
const int dst_u = -(uintptr_t)dest & 7;
const int shift = 3;
const int add = (1 << (shift - 1));
const vector uint32_t vadd = (vector uint32_t) {add, add, add, add};
const vector uint16_t vswap = (vector uint16_t) vec_splat_u16(big_endian ? 8 : 0);
const vector uint32_t vshift = (vector uint32_t) vec_splat_u32(shift);
vector uint32_t v, v2;
vector uint16_t vd;
int i;
yuv2plane1_16_u(src, dest, dst_u, big_endian, output_bits, 0);
for (i = dst_u; i < dstW - 7; i += 8) {
v = vec_vsx_ld(0, (const uint32_t *) &src[i]);
v = vec_add(v, vadd);
v = vec_sr(v, vshift);
v2 = vec_vsx_ld(0, (const uint32_t *) &src[i + 4]);
v2 = vec_add(v2, vadd);
v2 = vec_sr(v2, vshift);
vd = vec_packsu(v, v2);
vd = vec_rl(vd, vswap);
vec_st(vd, 0, &dest[i]);
}
yuv2plane1_16_u(src, dest, dstW, big_endian, output_bits, i);
}
static av_always_inline
void put_vp8_epel_h_altivec_core(uint8_t *dst, ptrdiff_t dst_stride,
uint8_t *src, ptrdiff_t src_stride,
int h, int mx, int w, int is6tap)
{
LOAD_H_SUBPEL_FILTER(mx-1);
vec_u8 align_vec0, align_vec8, permh0, permh8, filt;
vec_u8 perm_6tap0, perm_6tap8, perml0, perml8;
vec_u8 a, b, pixh, pixl, outer;
vec_s16 f16h, f16l;
vec_s32 filth, filtl;
vec_u8 perm_inner6 = { 1,2,3,4, 2,3,4,5, 3,4,5,6, 4,5,6,7 };
vec_u8 perm_inner4 = { 0,1,2,3, 1,2,3,4, 2,3,4,5, 3,4,5,6 };
vec_u8 perm_inner = is6tap ? perm_inner6 : perm_inner4;
vec_u8 perm_outer = { 4,9, 0,5, 5,10, 1,6, 6,11, 2,7, 7,12, 3,8 };
vec_s32 c64 = vec_sl(vec_splat_s32(1), vec_splat_u32(6));
vec_u16 c7 = vec_splat_u16(7);
align_vec0 = vec_lvsl( -is6tap-1, src);
align_vec8 = vec_lvsl(8-is6tap-1, src);
permh0 = vec_perm(align_vec0, align_vec0, perm_inner);
permh8 = vec_perm(align_vec8, align_vec8, perm_inner);
perm_inner = vec_add(perm_inner, vec_splat_u8(4));
perml0 = vec_perm(align_vec0, align_vec0, perm_inner);
perml8 = vec_perm(align_vec8, align_vec8, perm_inner);
perm_6tap0 = vec_perm(align_vec0, align_vec0, perm_outer);
perm_6tap8 = vec_perm(align_vec8, align_vec8, perm_outer);
while (h --> 0) {
FILTER_H(f16h, 0);
if (w == 16) {
FILTER_H(f16l, 8);
filt = vec_packsu(f16h, f16l);
vec_st(filt, 0, dst);
} else {
filt = vec_packsu(f16h, f16h);
vec_ste((vec_u32)filt, 0, (uint32_t*)dst);
if (w == 8)
vec_ste((vec_u32)filt, 4, (uint32_t*)dst);
}
src += src_stride;
dst += dst_stride;
}
}
/* next one assumes that ((line_size % 8) == 0) */
static void put_no_rnd_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
{
register int i;
register vector unsigned char pixelsv1, pixelsv2, pixelsavg;
register vector unsigned char blockv, temp1, temp2;
register vector unsigned short pixelssum1, pixelssum2, temp3;
register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1);
register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
temp1 = vec_ld(0, pixels);
temp2 = vec_ld(16, pixels);
pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels));
if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) {
pixelsv2 = temp2;
} else {
pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels));
}
pixelsv1 = vec_mergeh(vczero, pixelsv1);
pixelsv2 = vec_mergeh(vczero, pixelsv2);
pixelssum1 = vec_add((vector unsigned short)pixelsv1,
(vector unsigned short)pixelsv2);
pixelssum1 = vec_add(pixelssum1, vcone);
for (i = 0; i < h ; i++) {
int rightside = ((unsigned long)block & 0x0000000F);
blockv = vec_ld(0, block);
temp1 = vec_ld(line_size, pixels);
temp2 = vec_ld(line_size + 16, pixels);
pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) {
pixelsv2 = temp2;
} else {
pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
}
pixelsv1 = vec_mergeh(vczero, pixelsv1);
pixelsv2 = vec_mergeh(vczero, pixelsv2);
pixelssum2 = vec_add((vector unsigned short)pixelsv1,
(vector unsigned short)pixelsv2);
temp3 = vec_add(pixelssum1, pixelssum2);
temp3 = vec_sra(temp3, vctwo);
pixelssum1 = vec_add(pixelssum2, vcone);
pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);
if (rightside) {
blockv = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
} else {
blockv = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
}
vec_st(blockv, 0, block);
block += line_size;
pixels += line_size;
}
}
static av_always_inline void h264_idct_dc_add_internal(uint8_t *dst, int16_t *block, int stride, int size)
{
vec_s16 dc16;
vec_u8 dcplus, dcminus, v0, v1, v2, v3, aligner;
LOAD_ZERO;
DECLARE_ALIGNED(16, int, dc);
int i;
dc = (block[0] + 32) >> 6;
block[0] = 0;
dc16 = vec_splat((vec_s16) vec_lde(0, &dc), 1);
if (size == 4)
dc16 = vec_sld(dc16, zero_s16v, 8);
dcplus = vec_packsu(dc16, zero_s16v);
dcminus = vec_packsu(vec_sub(zero_s16v, dc16), zero_s16v);
aligner = vec_lvsr(0, dst);
dcplus = vec_perm(dcplus, dcplus, aligner);
dcminus = vec_perm(dcminus, dcminus, aligner);
for (i = 0; i < size; i += 4) {
v0 = vec_ld(0, dst+0*stride);
v1 = vec_ld(0, dst+1*stride);
v2 = vec_ld(0, dst+2*stride);
v3 = vec_ld(0, dst+3*stride);
v0 = vec_adds(v0, dcplus);
v1 = vec_adds(v1, dcplus);
v2 = vec_adds(v2, dcplus);
v3 = vec_adds(v3, dcplus);
v0 = vec_subs(v0, dcminus);
v1 = vec_subs(v1, dcminus);
v2 = vec_subs(v2, dcminus);
v3 = vec_subs(v3, dcminus);
vec_st(v0, 0, dst+0*stride);
vec_st(v1, 0, dst+1*stride);
vec_st(v2, 0, dst+2*stride);
vec_st(v3, 0, dst+3*stride);
dst += 4*stride;
}
}
static void yuv2planeX_16_altivec(const int16_t *filter, int filterSize,
const int16_t **src, uint8_t *dest,
const uint8_t *dither, int offset, int x)
{
register int i, j;
DECLARE_ALIGNED(16, int, val)[16];
vector signed int vo1, vo2, vo3, vo4;
vector unsigned short vs1, vs2;
vector unsigned char vf;
vector unsigned int altivec_vectorShiftInt19 =
vec_add(vec_splat_u32(10), vec_splat_u32(9));
for (i = 0; i < 16; i++)
val[i] = dither[(x + i + offset) & 7] << 12;
vo1 = vec_ld(0, val);
vo2 = vec_ld(16, val);
vo3 = vec_ld(32, val);
vo4 = vec_ld(48, val);
for (j = 0; j < filterSize; j++) {
vector signed short l1, vLumFilter = vec_ld(j << 1, filter);
vector unsigned char perm, perm0 = vec_lvsl(j << 1, filter);
vLumFilter = vec_perm(vLumFilter, vLumFilter, perm0);
vLumFilter = vec_splat(vLumFilter, 0); // lumFilter[j] is loaded 8 times in vLumFilter
perm = vec_lvsl(x << 1, src[j]);
l1 = vec_ld(x << 1, src[j]);
yuv2planeX_8(vo1, vo2, l1, src[j], x, perm, vLumFilter);
yuv2planeX_8(vo3, vo4, l1, src[j], x + 8, perm, vLumFilter);
}
vo1 = vec_sra(vo1, altivec_vectorShiftInt19);
vo2 = vec_sra(vo2, altivec_vectorShiftInt19);
vo3 = vec_sra(vo3, altivec_vectorShiftInt19);
vo4 = vec_sra(vo4, altivec_vectorShiftInt19);
vs1 = vec_packsu(vo1, vo2);
vs2 = vec_packsu(vo3, vo4);
vf = vec_packsu(vs1, vs2);
vec_st(vf, 0, dest);
}
static void yuv2planeX_16_altivec(const int16_t *filter, int filterSize,
const int16_t **src, uint8_t *dest,
const uint8_t *dither, int offset, int x)
{
register int i, j;
LOCAL_ALIGNED(16, int, val, [16]);
vector signed int vo1, vo2, vo3, vo4;
vector unsigned short vs1, vs2;
vector unsigned char vf;
vector unsigned int altivec_vectorShiftInt19 =
vec_add(vec_splat_u32(10), vec_splat_u32(9));
for (i = 0; i < 16; i++)
val[i] = dither[(x + i + offset) & 7] << 12;
vo1 = vec_ld(0, val);
vo2 = vec_ld(16, val);
vo3 = vec_ld(32, val);
vo4 = vec_ld(48, val);
for (j = 0; j < filterSize; j++) {
unsigned int joffset=j<<1;
unsigned int xoffset=x<<1;
vector unsigned char perm;
vector signed short l1,vLumFilter;
LOAD_FILTER(vLumFilter,filter);
vLumFilter = vec_splat(vLumFilter, 0);
LOAD_L1(l1,src[j],perm);
yuv2planeX_8(vo1, vo2, l1, src[j], x, perm, vLumFilter);
yuv2planeX_8(vo3, vo4, l1, src[j], x + 8, perm, vLumFilter);
}
vo1 = vec_sra(vo1, altivec_vectorShiftInt19);
vo2 = vec_sra(vo2, altivec_vectorShiftInt19);
vo3 = vec_sra(vo3, altivec_vectorShiftInt19);
vo4 = vec_sra(vo4, altivec_vectorShiftInt19);
vs1 = vec_packsu(vo1, vo2);
vs2 = vec_packsu(vo3, vo4);
vf = vec_packsu(vs1, vs2);
VEC_ST(vf, 0, dest);
}
static inline void
altivec_packIntArrayToCharArray(int *val, uint8_t* dest, int dstW)
{
register int i;
vector unsigned int altivec_vectorShiftInt19 =
vec_add(vec_splat_u32(10), vec_splat_u32(9));
if ((uintptr_t)dest % 16) {
/* badly aligned store, we force store alignment */
/* and will handle load misalignment on val w/ vec_perm */
vector unsigned char perm1;
vector signed int v1;
for (i = 0 ; (i < dstW) &&
(((uintptr_t)dest + i) % 16) ; i++) {
int t = val[i] >> 19;
dest[i] = (t < 0) ? 0 : ((t > 255) ? 255 : t);
}
perm1 = vec_lvsl(i << 2, val);
v1 = vec_ld(i << 2, val);
for ( ; i < (dstW - 15); i+=16) {
int offset = i << 2;
vector signed int v2 = vec_ld(offset + 16, val);
vector signed int v3 = vec_ld(offset + 32, val);
vector signed int v4 = vec_ld(offset + 48, val);
vector signed int v5 = vec_ld(offset + 64, val);
vector signed int v12 = vec_perm(v1, v2, perm1);
vector signed int v23 = vec_perm(v2, v3, perm1);
vector signed int v34 = vec_perm(v3, v4, perm1);
vector signed int v45 = vec_perm(v4, v5, perm1);
vector signed int vA = vec_sra(v12, altivec_vectorShiftInt19);
vector signed int vB = vec_sra(v23, altivec_vectorShiftInt19);
vector signed int vC = vec_sra(v34, altivec_vectorShiftInt19);
vector signed int vD = vec_sra(v45, altivec_vectorShiftInt19);
vector unsigned short vs1 = vec_packsu(vA, vB);
vector unsigned short vs2 = vec_packsu(vC, vD);
vector unsigned char vf = vec_packsu(vs1, vs2);
vec_st(vf, i, dest);
v1 = v5;
}
} else { // dest is properly aligned, great
void pix_diff :: processRGBA_Altivec(imageStruct &image, imageStruct &right)
{
int datasize = image.xsize * image.ysize / 4;
vector signed short hiImage, loImage, hiRight, loRight;
vector unsigned char zero = vec_splat_u8(0);
vector unsigned char *inData = (vector unsigned char *)image.data;
vector unsigned char *rightData = (vector unsigned char *)right.data;
#ifndef PPC970
UInt32 prefetchSize = GetPrefetchConstant( 16, 1, 256 );
vec_dst( inData, prefetchSize, 0 );
vec_dst( rightData, prefetchSize, 1 );
vec_dst( inData+256, prefetchSize, 2 );
vec_dst( rightData+256, prefetchSize, 3 );
#endif
do {
#ifndef PPC970
vec_dst( inData, prefetchSize, 0 );
vec_dst( rightData, prefetchSize, 1 );
vec_dst( inData+256, prefetchSize, 2 );
vec_dst( rightData+256, prefetchSize, 3 );
#endif
hiImage = (vector signed short)vec_mergeh(zero,inData[0]);
loImage = (vector signed short)vec_mergel(zero,inData[0]);
hiRight = (vector signed short)vec_mergeh(zero,rightData[0]);
loRight = (vector signed short)vec_mergel(zero,rightData[0]);
hiImage = vec_subs(hiImage,hiRight);
loImage = vec_subs(loImage,loRight);
hiImage = vec_abs(hiImage);
loImage = vec_abs(loImage);
inData[0] = vec_packsu(hiImage,loImage);
inData++;
rightData++;
}
while (--datasize);
#ifndef PPC970
vec_dss( 0 );
vec_dss( 1 );
vec_dss( 2 );
vec_dss( 3 );
#endif
}
static void predict_16x16_p_altivec( uint8_t *src )
{
int16_t a, b, c, i;
int H = 0;
int V = 0;
int16_t i00;
for( i = 1; i <= 8; i++ )
{
H += i * ( src[7+i - FDEC_STRIDE ] - src[7-i - FDEC_STRIDE ] );
V += i * ( src[(7+i)*FDEC_STRIDE -1] - src[(7-i)*FDEC_STRIDE -1] );
}
a = 16 * ( src[15*FDEC_STRIDE -1] + src[15 - FDEC_STRIDE] );
b = ( 5 * H + 32 ) >> 6;
c = ( 5 * V + 32 ) >> 6;
i00 = a - b * 7 - c * 7 + 16;
vect_sshort_u i00_u, b_u, c_u;
i00_u.s[0] = i00;
b_u.s[0] = b;
c_u.s[0] = c;
vec_u16_t val5_v = vec_splat_u16(5);
vec_s16_t i00_v, b_v, c_v;
i00_v = vec_splat(i00_u.v, 0);
b_v = vec_splat(b_u.v, 0);
c_v = vec_splat(c_u.v, 0);
vec_s16_t induc_v = (vec_s16_t) CV(0, 1, 2, 3, 4, 5, 6, 7);
vec_s16_t b8_v = vec_sl(b_v, vec_splat_u16(3));
vec_s32_t mule_b_v = vec_mule(induc_v, b_v);
vec_s32_t mulo_b_v = vec_mulo(induc_v, b_v);
vec_s16_t mul_b_induc0_v = vec_pack(vec_mergeh(mule_b_v, mulo_b_v), vec_mergel(mule_b_v, mulo_b_v));
vec_s16_t add_i0_b_0v = vec_adds(i00_v, mul_b_induc0_v);
vec_s16_t add_i0_b_8v = vec_adds(b8_v, add_i0_b_0v);
int y;
for( y = 0; y < 16; y++ )
{
vec_s16_t shift_0_v = vec_sra(add_i0_b_0v, val5_v);
vec_s16_t shift_8_v = vec_sra(add_i0_b_8v, val5_v);
vec_u8_t com_sat_v = vec_packsu(shift_0_v, shift_8_v);
vec_st( com_sat_v, 0, &src[0]);
src += FDEC_STRIDE;
i00 += c;
add_i0_b_0v = vec_adds(add_i0_b_0v, c_v);
add_i0_b_8v = vec_adds(add_i0_b_8v, c_v);
}
}
static void predict_8x8c_p_altivec( uint8_t *src )
{
int H = 0, V = 0;
for( int i = 0; i < 4; i++ )
{
H += ( i + 1 ) * ( src[4+i - FDEC_STRIDE] - src[2 - i -FDEC_STRIDE] );
V += ( i + 1 ) * ( src[-1 +(i+4)*FDEC_STRIDE] - src[-1+(2-i)*FDEC_STRIDE] );
}
int a = 16 * ( src[-1+7*FDEC_STRIDE] + src[7 - FDEC_STRIDE] );
int b = ( 17 * H + 16 ) >> 5;
int c = ( 17 * V + 16 ) >> 5;
int i00 = a -3*b -3*c + 16;
vec_s16_u i00_u, b_u, c_u;
i00_u.s[0] = i00;
b_u.s[0] = b;
c_u.s[0] = c;
vec_u16_t val5_v = vec_splat_u16(5);
vec_s16_t i00_v, b_v, c_v;
i00_v = vec_splat(i00_u.v, 0);
b_v = vec_splat(b_u.v, 0);
c_v = vec_splat(c_u.v, 0);
vec_s16_t induc_v = (vec_s16_t) CV(0, 1, 2, 3, 4, 5, 6, 7);
vec_s16_t add_i0_b_0v = vec_mladd(induc_v, b_v, i00_v);
PREP_STORE8;
for( int i = 0; i < 8; ++i )
{
vec_s16_t shift_0_v = vec_sra(add_i0_b_0v, val5_v);
vec_u8_t com_sat_v = vec_packsu(shift_0_v, shift_0_v);
VEC_STORE8(com_sat_v, &src[0]);
src += FDEC_STRIDE;
add_i0_b_0v = vec_adds(add_i0_b_0v, c_v);
}
}
static void predict_16x16_p_altivec( uint8_t *src )
{
int H = 0, V = 0;
for( int i = 1; i <= 8; i++ )
{
H += i * ( src[7+i - FDEC_STRIDE ] - src[7-i - FDEC_STRIDE ] );
V += i * ( src[(7+i)*FDEC_STRIDE -1] - src[(7-i)*FDEC_STRIDE -1] );
}
int a = 16 * ( src[15*FDEC_STRIDE -1] + src[15 - FDEC_STRIDE] );
int b = ( 5 * H + 32 ) >> 6;
int c = ( 5 * V + 32 ) >> 6;
int i00 = a - b * 7 - c * 7 + 16;
vec_s16_u i00_u, b_u, c_u;
i00_u.s[0] = i00;
b_u.s[0] = b;
c_u.s[0] = c;
vec_u16_t val5_v = vec_splat_u16(5);
vec_s16_t i00_v, b_v, c_v;
i00_v = vec_splat(i00_u.v, 0);
b_v = vec_splat(b_u.v, 0);
c_v = vec_splat(c_u.v, 0);
vec_s16_t induc_v = (vec_s16_t) CV(0, 1, 2, 3, 4, 5, 6, 7);
vec_s16_t b8_v = vec_sl(b_v, vec_splat_u16(3));
vec_s16_t add_i0_b_0v = vec_mladd(induc_v, b_v, i00_v);
vec_s16_t add_i0_b_8v = vec_adds(b8_v, add_i0_b_0v);
for( int y = 0; y < 16; y++ )
{
vec_s16_t shift_0_v = vec_sra(add_i0_b_0v, val5_v);
vec_s16_t shift_8_v = vec_sra(add_i0_b_8v, val5_v);
vec_u8_t com_sat_v = vec_packsu(shift_0_v, shift_8_v);
vec_st( com_sat_v, 0, &src[0]);
src += FDEC_STRIDE;
add_i0_b_0v = vec_adds(add_i0_b_0v, c_v);
add_i0_b_8v = vec_adds(add_i0_b_8v, c_v);
}
}
static force_inline vector unsigned int
pix_multiply (vector unsigned int p, vector unsigned int a)
{
vector unsigned short hi, lo, mod;
/* unpack to short */
hi = (vector unsigned short)
vec_mergeh ((vector unsigned char)AVV (0),
(vector unsigned char)p);
mod = (vector unsigned short)
vec_mergeh ((vector unsigned char)AVV (0),
(vector unsigned char)a);
hi = vec_mladd (hi, mod, (vector unsigned short)
AVV (0x0080, 0x0080, 0x0080, 0x0080,
0x0080, 0x0080, 0x0080, 0x0080));
hi = vec_adds (hi, vec_sr (hi, vec_splat_u16 (8)));
hi = vec_sr (hi, vec_splat_u16 (8));
/* unpack to short */
lo = (vector unsigned short)
vec_mergel ((vector unsigned char)AVV (0),
(vector unsigned char)p);
mod = (vector unsigned short)
vec_mergel ((vector unsigned char)AVV (0),
(vector unsigned char)a);
lo = vec_mladd (lo, mod, (vector unsigned short)
AVV (0x0080, 0x0080, 0x0080, 0x0080,
0x0080, 0x0080, 0x0080, 0x0080));
lo = vec_adds (lo, vec_sr (lo, vec_splat_u16 (8)));
lo = vec_sr (lo, vec_splat_u16 (8));
return (vector unsigned int)vec_packsu (hi, lo);
}
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;
}
}
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 ProjectDlightTexture_altivec( void ) {
int i, l;
vec_t origin0, origin1, origin2;
float texCoords0, texCoords1;
vector float floatColorVec0, floatColorVec1;
vector float modulateVec, colorVec, zero;
vector short colorShort;
vector signed int colorInt;
vector unsigned char floatColorVecPerm, modulatePerm, colorChar;
vector unsigned char vSel = VECCONST_UINT8(0x00, 0x00, 0x00, 0xff,
0x00, 0x00, 0x00, 0xff,
0x00, 0x00, 0x00, 0xff,
0x00, 0x00, 0x00, 0xff);
float *texCoords;
byte *colors;
byte clipBits[SHADER_MAX_VERTEXES];
float texCoordsArray[SHADER_MAX_VERTEXES][2];
byte colorArray[SHADER_MAX_VERTEXES][4];
unsigned hitIndexes[SHADER_MAX_INDEXES];
int numIndexes;
float scale;
float radius;
vec3_t floatColor;
float modulate = 0.0f;
if ( !backEnd.refdef.num_dlights ) {
return;
}
// There has to be a better way to do this so that floatColor
// and/or modulate are already 16-byte aligned.
floatColorVecPerm = vec_lvsl(0,(float *)floatColor);
modulatePerm = vec_lvsl(0,(float *)&modulate);
modulatePerm = (vector unsigned char)vec_splat((vector unsigned int)modulatePerm,0);
zero = (vector float)vec_splat_s8(0);
for ( l = 0 ; l < backEnd.refdef.num_dlights ; l++ ) {
dlight_t *dl;
if ( !( tess.dlightBits & ( 1 << l ) ) ) {
continue; // this surface definately doesn't have any of this light
}
texCoords = texCoordsArray[0];
colors = colorArray[0];
dl = &backEnd.refdef.dlights[l];
origin0 = dl->transformed[0];
origin1 = dl->transformed[1];
origin2 = dl->transformed[2];
radius = dl->radius;
scale = 1.0f / radius;
if(r_greyscale->integer)
{
float luminance;
luminance = LUMA(dl->color[0], dl->color[1], dl->color[2]) * 255.0f;
floatColor[0] = floatColor[1] = floatColor[2] = luminance;
}
else if(r_greyscale->value)
{
float luminance;
luminance = LUMA(dl->color[0], dl->color[1], dl->color[2]) * 255.0f;
floatColor[0] = LERP(dl->color[0] * 255.0f, luminance, r_greyscale->value);
floatColor[1] = LERP(dl->color[1] * 255.0f, luminance, r_greyscale->value);
floatColor[2] = LERP(dl->color[2] * 255.0f, luminance, r_greyscale->value);
}
else
{
floatColor[0] = dl->color[0] * 255.0f;
floatColor[1] = dl->color[1] * 255.0f;
floatColor[2] = dl->color[2] * 255.0f;
}
floatColorVec0 = vec_ld(0, floatColor);
floatColorVec1 = vec_ld(11, floatColor);
floatColorVec0 = vec_perm(floatColorVec0,floatColorVec0,floatColorVecPerm);
for ( i = 0 ; i < tess.numVertexes ; i++, texCoords += 2, colors += 4 ) {
int clip = 0;
vec_t dist0, dist1, dist2;
dist0 = origin0 - tess.xyz[i][0];
dist1 = origin1 - tess.xyz[i][1];
dist2 = origin2 - tess.xyz[i][2];
backEnd.pc.c_dlightVertexes++;
texCoords0 = 0.5f + dist0 * scale;
texCoords1 = 0.5f + dist1 * scale;
if( !r_dlightBacks->integer &&
// dist . tess.normal[i]
( dist0 * tess.normal[i][0] +
dist1 * tess.normal[i][1] +
dist2 * tess.normal[i][2] ) < 0.0f ) {
clip = 63;
} else {
if ( texCoords0 < 0.0f ) {
clip |= 1;
} else if ( texCoords0 > 1.0f ) {
clip |= 2;
}
if ( texCoords1 < 0.0f ) {
clip |= 4;
} else if ( texCoords1 > 1.0f ) {
clip |= 8;
}
texCoords[0] = texCoords0;
texCoords[1] = texCoords1;
// modulate the strength based on the height and color
if ( dist2 > radius ) {
clip |= 16;
modulate = 0.0f;
} else if ( dist2 < -radius ) {
clip |= 32;
modulate = 0.0f;
} else {
dist2 = Q_fabs(dist2);
if ( dist2 < radius * 0.5f ) {
modulate = 1.0f;
} else {
modulate = 2.0f * (radius - dist2) * scale;
}
}
}
clipBits[i] = clip;
modulateVec = vec_ld(0,(float *)&modulate);
modulateVec = vec_perm(modulateVec,modulateVec,modulatePerm);
colorVec = vec_madd(floatColorVec0,modulateVec,zero);
colorInt = vec_cts(colorVec,0); // RGBx
colorShort = vec_pack(colorInt,colorInt); // RGBxRGBx
colorChar = vec_packsu(colorShort,colorShort); // RGBxRGBxRGBxRGBx
colorChar = vec_sel(colorChar,vSel,vSel); // RGBARGBARGBARGBA replace alpha with 255
vec_ste((vector unsigned int)colorChar,0,(unsigned int *)colors); // store color
}
// build a list of triangles that need light
numIndexes = 0;
for ( i = 0 ; i < tess.numIndexes ; i += 3 ) {
int a, b, c;
a = tess.indexes[i];
b = tess.indexes[i+1];
c = tess.indexes[i+2];
if ( clipBits[a] & clipBits[b] & clipBits[c] ) {
continue; // not lighted
}
hitIndexes[numIndexes] = a;
hitIndexes[numIndexes+1] = b;
hitIndexes[numIndexes+2] = c;
numIndexes += 3;
}
if ( !numIndexes ) {
continue;
}
qglEnableClientState( GL_TEXTURE_COORD_ARRAY );
qglTexCoordPointer( 2, GL_FLOAT, 0, texCoordsArray[0] );
qglEnableClientState( GL_COLOR_ARRAY );
qglColorPointer( 4, GL_UNSIGNED_BYTE, 0, colorArray );
GL_Bind( tr.dlightImage );
// include GLS_DEPTHFUNC_EQUAL so alpha tested surfaces don't add light
// where they aren't rendered
if ( dl->additive ) {
GL_State( GLS_SRCBLEND_ONE | GLS_DSTBLEND_ONE | GLS_DEPTHFUNC_EQUAL );
}
else {
GL_State( GLS_SRCBLEND_DST_COLOR | GLS_DSTBLEND_ONE | GLS_DEPTHFUNC_EQUAL );
}
R_DrawElements( numIndexes, hitIndexes );
backEnd.pc.c_totalIndexes += numIndexes;
backEnd.pc.c_dlightIndexes += numIndexes;
}
}
void pix_background :: processYUVAltivec(imageStruct &image)
{
register int h,w,i,j,width;
int pixsize = image.xsize * image.ysize * image.csize;
h = image.ysize;
w = image.xsize/8;
width = image.xsize/8;
//check to see if the buffer isn't 16byte aligned (highly unlikely)
if (image.ysize*image.xsize % 16 != 0){
error("image not properly aligned for Altivec - try something SD or HD maybe?");
return;
}
union{
unsigned short s[8];
vector unsigned short v;
}shortBuffer;
if(m_savedImage.xsize!=image.xsize ||
m_savedImage.ysize!=image.ysize ||
m_savedImage.format!=image.format)m_reset=1;
m_savedImage.xsize=image.xsize;
m_savedImage.ysize=image.ysize;
m_savedImage.setCsizeByFormat(image.format);
m_savedImage.reallocate();
if (m_reset){
memcpy(m_savedImage.data,image.data,pixsize);
m_reset = 0;
}
register vector unsigned short UVres1, Yres1, UVres2, Yres2;//interleave;
register vector unsigned short hiImage, loImage;
register vector unsigned short Yrange, UVrange, Yblank,UVblank,blank;
register vector bool short Ymasklo,Ymaskhi, UVmaskhi;
register vector unsigned short Yhi,Ylo,UVhi,UVlo;
register vector unsigned char one = vec_splat_u8(1);
register vector unsigned short sone = vec_splat_u16(1);
register vector unsigned int Uhi, Ulo, Vhi, Vlo,Ures,Vres;
register vector bool int Umasklo, Umaskhi, Vmaskhi, Vmasklo;
vector unsigned char *inData = (vector unsigned char*) image.data;
vector unsigned char *rightData = (vector unsigned char*) m_savedImage.data;
shortBuffer.s[0] = m_Yrange;
Yrange = shortBuffer.v;
Yrange = vec_splat(Yrange,0);
shortBuffer.s[0] = 128;
shortBuffer.s[1] = 0;
shortBuffer.s[2] = 128;
shortBuffer.s[3] = 0;
shortBuffer.s[4] = 128;
shortBuffer.s[5] = 0;
shortBuffer.s[6] = 128;
shortBuffer.s[7] = 0;
blank = shortBuffer.v;
shortBuffer.s[0] = 0;
Yblank = shortBuffer.v;
Yblank = vec_splat(Yblank,0);
shortBuffer.s[0] = 128;
UVblank = shortBuffer.v;
UVblank = vec_splat(UVblank,0);
shortBuffer.s[0] = m_Urange;
shortBuffer.s[1] = m_Vrange;
shortBuffer.s[2] = m_Urange;
shortBuffer.s[3] = m_Vrange;
shortBuffer.s[4] = m_Urange;
shortBuffer.s[5] = m_Vrange;
shortBuffer.s[6] = m_Urange;
shortBuffer.s[7] = m_Vrange;
UVrange = shortBuffer.v;
//setup the cache prefetch -- A MUST!!!
UInt32 prefetchSize = GetPrefetchConstant( 16, 1, 256 );
#ifndef PPC970
vec_dst( inData, prefetchSize, 0 );
vec_dst( rightData, prefetchSize, 1 );
vec_dst( inData+32, prefetchSize, 2 );
vec_dst( rightData+32, prefetchSize, 3 );
#endif //PPC970
for ( i=0; i<h; i++){
for (j=0; j<w; j++)
{
#ifndef PPC970
//this function is probably memory bound on most G4's -- what else is new?
vec_dst( inData, prefetchSize, 0 );
vec_dst( rightData, prefetchSize, 1 );
vec_dst( inData+32, prefetchSize, 2 );
vec_dst( rightData+32, prefetchSize, 3 );
#endif
//separate the U and V from Y
UVres1 = (vector unsigned short)vec_mule(one,inData[0]);
UVres2 = (vector unsigned short)vec_mule(one,rightData[0]);
//vec_mulo Y * 1 to short vector Y Y Y Y shorts
Yres1 = (vector unsigned short)vec_mulo(one,inData[0]);
Yres2 = (vector unsigned short)vec_mulo(one,rightData[0]);
Yhi = vec_adds(Yres2,Yrange);
Ylo = vec_subs(Yres2,Yrange);
//go to ints for comparison
UVhi = vec_adds(UVres2,UVrange);
UVlo = vec_subs(UVres2,UVrange);
Uhi = vec_mule(sone,UVhi);
Ulo = vec_mule(sone,UVlo);
Vhi = vec_mulo(sone,UVhi);
Vlo = vec_mulo(sone,UVlo);
Ures = vec_mule(sone,UVres1);
Vres = vec_mulo(sone,UVres1);
Umasklo = vec_cmpgt(Ures,Ulo);
Umaskhi = vec_cmplt(Ures,Uhi);
Vmasklo = vec_cmpgt(Vres,Vlo);
Vmaskhi = vec_cmplt(Vres,Vhi);
Umaskhi = vec_and(Umaskhi,Umasklo);
Vmaskhi = vec_and(Vmaskhi,Vmasklo);
Umasklo = vec_and(Umaskhi,Vmaskhi);
Vmasklo = vec_and(Umaskhi,Vmaskhi);
hiImage = (vector unsigned short)vec_mergeh(Umasklo,Vmasklo);
loImage = (vector unsigned short)vec_mergel(Umasklo,Vmasklo);
//pack it back down to bool short
UVmaskhi = (vector bool short)vec_packsu(hiImage,loImage);
Ymasklo = vec_cmpgt(Yres1,Ylo);
Ymaskhi = vec_cmplt(Yres1,Yhi);
Ymaskhi = vec_and(Ymaskhi,Ymasklo);
Ymaskhi = vec_and(Ymaskhi,UVmaskhi);
UVmaskhi = vec_and(Ymaskhi,UVmaskhi);
//bitwise comparison and move using the result of the comparison as a mask
Yres1 = vec_sel(Yres1,Yblank,Ymaskhi);
//UVres1 = vec_sel(UVres1,UVres2,UVmaskhi);
UVres1 = vec_sel(UVres1,UVblank,UVmaskhi);
//merge the Y and UV back together
hiImage = vec_mergeh(UVres1,Yres1);
loImage = vec_mergel(UVres1,Yres1);
//pack it back down to unsigned char to store
inData[0] = vec_packsu(hiImage,loImage);
inData++;
rightData++;
}
#ifndef PPC970
vec_dss(0);
vec_dss(1);
vec_dss(2);
vec_dss(3);
#endif
}
}
/* this code assume that stride % 16 == 0 */
void put_no_rnd_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride, int h, int x, int y) {
signed int ABCD[4] __attribute__((aligned(16))) =
{((8 - x) * (8 - y)),
((x) * (8 - y)),
((8 - x) * (y)),
((x) * (y))};
register int i;
vector unsigned char fperm;
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 v28ss = vec_sub(vec_sl(vec_splat_s16(1),vec_splat_u16(5)),vec_splat_s16(4));
const vector unsigned short 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;
vector unsigned char vsrcAuc, vsrcBuc, vsrcperm0, vsrcperm1;
vector unsigned char vsrc0uc, vsrc1uc;
vector signed short vsrc0ssH, vsrc1ssH;
vector unsigned char vsrcCuc, vsrc2uc, vsrc3uc;
vector signed short vsrc2ssH, vsrc3ssH, psum;
vector unsigned char vdst, ppsum, fsum;
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);
}
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 = (vector signed short)vec_mergeh((vector unsigned char)vzero,
(vector unsigned char)vsrc0uc);
vsrc1ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero,
(vector unsigned char)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 = (vector signed short)vec_mergeh((vector unsigned char)vzero,
(vector unsigned char)vsrc2uc);
vsrc3ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero,
(vector unsigned char)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(v28ss, psum);
psum = vec_sra(psum, v6us);
vdst = vec_ld(0, dst);
ppsum = (vector unsigned char)vec_packsu(psum, psum);
fsum = vec_perm(vdst, ppsum, fperm);
vec_st(fsum, 0, dst);
vsrc0ssH = vsrc2ssH;
vsrc1ssH = vsrc3ssH;
dst += stride;
src += stride;
}
} else {
vector unsigned char 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 = (vector signed short)vec_mergeh((vector unsigned char)vzero,
(vector unsigned char)vsrc2uc);
vsrc3ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero,
(vector unsigned char)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(v28ss, psum);
psum = vec_sr(psum, v6us);
vdst = vec_ld(0, dst);
ppsum = (vector unsigned char)vec_pack(psum, psum);
fsum = vec_perm(vdst, ppsum, fperm);
vec_st(fsum, 0, dst);
vsrc0ssH = vsrc2ssH;
vsrc1ssH = vsrc3ssH;
dst += stride;
src += stride;
}
}
}
/* next one assumes that ((line_size % 16) == 0) */
static void put_no_rnd_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, ptrdiff_t line_size, int h)
{
register int i;
register vector unsigned char pixelsv1, pixelsv2, pixelsv3, pixelsv4;
register vector unsigned char blockv, temp1, temp2;
register vector unsigned short temp3, temp4,
pixelssum1, pixelssum2, pixelssum3, pixelssum4;
register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1);
register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
temp1 = vec_ld(0, pixels);
temp2 = vec_ld(16, pixels);
pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels));
if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) {
pixelsv2 = temp2;
} else {
pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels));
}
pixelsv3 = vec_mergel(vczero, pixelsv1);
pixelsv4 = vec_mergel(vczero, pixelsv2);
pixelsv1 = vec_mergeh(vczero, pixelsv1);
pixelsv2 = vec_mergeh(vczero, pixelsv2);
pixelssum3 = vec_add((vector unsigned short)pixelsv3,
(vector unsigned short)pixelsv4);
pixelssum3 = vec_add(pixelssum3, vcone);
pixelssum1 = vec_add((vector unsigned short)pixelsv1,
(vector unsigned short)pixelsv2);
pixelssum1 = vec_add(pixelssum1, vcone);
for (i = 0; i < h ; i++) {
blockv = vec_ld(0, block);
temp1 = vec_ld(line_size, pixels);
temp2 = vec_ld(line_size + 16, pixels);
pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) {
pixelsv2 = temp2;
} else {
pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
}
pixelsv3 = vec_mergel(vczero, pixelsv1);
pixelsv4 = vec_mergel(vczero, pixelsv2);
pixelsv1 = vec_mergeh(vczero, pixelsv1);
pixelsv2 = vec_mergeh(vczero, pixelsv2);
pixelssum4 = vec_add((vector unsigned short)pixelsv3,
(vector unsigned short)pixelsv4);
pixelssum2 = vec_add((vector unsigned short)pixelsv1,
(vector unsigned short)pixelsv2);
temp4 = vec_add(pixelssum3, pixelssum4);
temp4 = vec_sra(temp4, vctwo);
temp3 = vec_add(pixelssum1, pixelssum2);
temp3 = vec_sra(temp3, vctwo);
pixelssum3 = vec_add(pixelssum4, vcone);
pixelssum1 = vec_add(pixelssum2, vcone);
blockv = vec_packsu(temp3, temp4);
vec_st(blockv, 0, block);
block += line_size;
pixels += line_size;
}
}
int main ()
{
vector float fa = {1.0, 2.0, 3.0, -4.0};
vector float fb = {-2.0, -3.0, -4.0, -5.0};
vector float fc = vec_cpsgn (fa, fb);
vector long long la = {5L, 14L};
vector long long lb = {3L, 86L};
vector long long lc = vec_and (la, lb);
vector bool long long ld = {0, -1};
vector long long le = vec_and (la, ld);
vector long long lf = vec_and (ld, lb);
vector unsigned long long ua = {5L, 14L};
vector unsigned long long ub = {3L, 86L};
vector unsigned long long uc = vec_and (ua, ub);
vector bool long long ud = {0, -1};
vector unsigned long long ue = vec_and (ua, ud);
vector unsigned long long uf = vec_and (ud, ub);
vector long long lg = vec_andc (la, lb);
vector long long lh = vec_andc (la, ld);
vector long long li = vec_andc (ld, lb);
vector unsigned long long ug = vec_andc (ua, ub);
vector unsigned long long uh = vec_andc (ua, ud);
vector unsigned long long ui = vec_andc (ud, ub);
vector double da = {1.0, -4.0};
vector double db = {-2.0, 5.0};
vector double dc = vec_cpsgn (da, db);
vector long long lj = vec_mergeh (la, lb);
vector long long lk = vec_mergeh (la, ld);
vector long long ll = vec_mergeh (ld, la);
vector unsigned long long uj = vec_mergeh (ua, ub);
vector unsigned long long uk = vec_mergeh (ua, ud);
vector unsigned long long ul = vec_mergeh (ud, ua);
vector long long lm = vec_mergel (la, lb);
vector long long ln = vec_mergel (la, ld);
vector long long lo = vec_mergel (ld, la);
vector unsigned long long um = vec_mergel (ua, ub);
vector unsigned long long un = vec_mergel (ua, ud);
vector unsigned long long uo = vec_mergel (ud, ua);
vector long long lp = vec_nor (la, lb);
vector long long lq = vec_nor (la, ld);
vector long long lr = vec_nor (ld, la);
vector unsigned long long up = vec_nor (ua, ub);
vector unsigned long long uq = vec_nor (ua, ud);
vector unsigned long long ur = vec_nor (ud, ua);
vector long long ls = vec_or (la, lb);
vector long long lt = vec_or (la, ld);
vector long long lu = vec_or (ld, la);
vector unsigned long long us = vec_or (ua, ub);
vector unsigned long long ut = vec_or (ua, ud);
vector unsigned long long uu = vec_or (ud, ua);
vector unsigned char ca = {0,4,8,1,5,9,2,6,10,3,7,11,15,12,14,13};
vector long long lv = vec_perm (la, lb, ca);
vector unsigned long long uv = vec_perm (ua, ub, ca);
vector long long lw = vec_sel (la, lb, lc);
vector long long lx = vec_sel (la, lb, uc);
vector long long ly = vec_sel (la, lb, ld);
vector unsigned long long uw = vec_sel (ua, ub, lc);
vector unsigned long long ux = vec_sel (ua, ub, uc);
vector unsigned long long uy = vec_sel (ua, ub, ld);
vector long long lz = vec_xor (la, lb);
vector long long l0 = vec_xor (la, ld);
vector long long l1 = vec_xor (ld, la);
vector unsigned long long uz = vec_xor (ua, ub);
vector unsigned long long u0 = vec_xor (ua, ud);
vector unsigned long long u1 = vec_xor (ud, ua);
int ia = vec_all_eq (ua, ub);
int ib = vec_all_ge (ua, ub);
int ic = vec_all_gt (ua, ub);
int id = vec_all_le (ua, ub);
int ie = vec_all_lt (ua, ub);
int ig = vec_all_ne (ua, ub);
int ih = vec_any_eq (ua, ub);
int ii = vec_any_ge (ua, ub);
int ij = vec_any_gt (ua, ub);
int ik = vec_any_le (ua, ub);
int il = vec_any_lt (ua, ub);
int im = vec_any_ne (ua, ub);
vector int sia = {9, 16, 25, 36};
vector int sib = {-8, -27, -64, -125};
vector int sic = vec_mergee (sia, sib);
vector int sid = vec_mergeo (sia, sib);
vector unsigned int uia = {9, 16, 25, 36};
vector unsigned int uib = {8, 27, 64, 125};
vector unsigned int uic = vec_mergee (uia, uib);
vector unsigned int uid = vec_mergeo (uia, uib);
vector bool int bia = {0, -1, -1, 0};
vector bool int bib = {-1, -1, 0, -1};
vector bool int bic = vec_mergee (bia, bib);
vector bool int bid = vec_mergeo (bia, bib);
vector unsigned int uie = vec_packsu (ua, ub);
vector long long l2 = vec_cntlz (la);
vector unsigned long long u2 = vec_cntlz (ua);
vector int sie = vec_cntlz (sia);
vector unsigned int uif = vec_cntlz (uia);
vector short ssa = {20, -40, -60, 80, 100, -120, -140, 160};
vector short ssb = vec_cntlz (ssa);
vector unsigned short usa = {81, 72, 63, 54, 45, 36, 27, 18};
vector unsigned short usb = vec_cntlz (usa);
vector signed char sca = {-4, 3, -9, 15, -31, 31, 0, 0,
1, 117, -36, 99, 98, 97, 96, 95};
vector signed char scb = vec_cntlz (sca);
vector unsigned char cb = vec_cntlz (ca);
vector double dd = vec_xl (0, &y);
vec_xst (dd, 0, &z);
vector double de = vec_round (dd);
vector double df = vec_splat (de, 0);
vector double dg = vec_splat (de, 1);
vector long long l3 = vec_splat (l2, 0);
vector long long l4 = vec_splat (l2, 1);
vector unsigned long long u3 = vec_splat (u2, 0);
vector unsigned long long u4 = vec_splat (u2, 1);
vector bool long long l5 = vec_splat (ld, 0);
vector bool long long l6 = vec_splat (ld, 1);
vector long long l7 = vec_div (l3, l4);
vector unsigned long long u5 = vec_div (u3, u4);
vector long long l8 = vec_mul (l3, l4);
vector unsigned long long u6 = vec_mul (u3, u4);
vector double dh = vec_ctf (la, -2);
vector double di = vec_ctf (ua, 2);
vector long long l9 = vec_cts (dh, -2);
vector unsigned long long u7 = vec_ctu (di, 2);
return 0;
}
/* 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);
}
/* more optimized version - unrolled and load-hoisted */
void pix_offset :: processYUVAltivec(imageStruct &image)
{
register int h,w,width,height;
width = image.xsize/16; //for altivec
height = image.ysize;
//format is U Y V Y
// start of working altivec function
union {
short elements[8];
vector signed short v;
} transferBuffer;
register vector signed short c, hi, lo;
register vector signed short hi1, lo1;
register vector signed short loadhi, loadhi1, loadlo, loadlo1;
register vector unsigned char zero = vec_splat_u8(0);
register vector unsigned char *inData = (vector unsigned char*) image.data;
//Write the pixel (pair) to the transfer buffer
//transferBuffer.i = (U << 24) | (Y << 16) | (V << 8 ) | Y;
transferBuffer.elements[0] = U;
transferBuffer.elements[1] = Y;
transferBuffer.elements[2] = V;
transferBuffer.elements[3] = Y;
transferBuffer.elements[4] = U;
transferBuffer.elements[5] = Y;
transferBuffer.elements[6] = V;
transferBuffer.elements[7] = Y;
//Load it into the vector unit
c = transferBuffer.v;
#ifndef PPC970
UInt32 prefetchSize = GetPrefetchConstant( 16, 1,
256 );
vec_dst( inData, prefetchSize, 0 );
vec_dst( inData+16, prefetchSize, 1 );
vec_dst( inData+32, prefetchSize, 2 );
vec_dst( inData+64, prefetchSize, 3 );
#endif
//expand the UInt8's to short's
loadhi = (vector signed short) vec_mergeh( zero, inData[0] );
loadlo = (vector signed short) vec_mergel( zero, inData[0] );
loadhi1 = (vector signed short) vec_mergeh( zero, inData[1] );
loadlo1 = (vector signed short) vec_mergel( zero, inData[1] );
\
for ( h=0; h<height; h++) {
for (w=0; w<width; w++) {
#ifndef PPC970
vec_dst( inData, prefetchSize, 0 );
vec_dst( inData+16, prefetchSize, 1 );
vec_dst( inData+32, prefetchSize, 2 );
vec_dst( inData+64, prefetchSize, 3 );
#endif
//add the constant to it
hi = vec_add( loadhi, c );
lo = vec_add( loadlo, c );
hi1 = vec_add( loadhi1, c );
lo1 = vec_add( loadlo1, c );
//expand the UInt8's to short's
loadhi = (vector signed short) vec_mergeh( zero, inData[2] );
loadlo = (vector signed short) vec_mergel( zero, inData[2] );
loadhi1 = (vector signed short) vec_mergeh( zero, inData[3] );
loadlo1 = (vector signed short) vec_mergel( zero, inData[3] );
//pack the result back down, with saturation
inData[0] = vec_packsu( hi, lo );
inData++;
inData[0] = vec_packsu( hi1, lo1 );
inData++;
}
}
//
// finish the last iteration after the loop
//
hi = vec_add( loadhi, c );
lo = vec_add( loadlo, c );
hi1 = vec_add( loadhi1, c );
lo1 = vec_add( loadlo1, c );
//pack the result back down, with saturation
inData[0] = vec_packsu( hi, lo );
inData++;
inData[0] = vec_packsu( hi1, lo1 );
inData++;
#ifndef PPC970
vec_dss( 0 );
vec_dss( 1 );
vec_dss( 2 );
vec_dss( 3 ); //end of working altivec function
#endif
}
/* 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 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 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_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;
}
}
/* 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);
}
void pix_add :: processYUV_Altivec(imageStruct &image, imageStruct &right)
{
int h,w,width;
width = image.xsize/8;
//format is U Y V Y
union
{
//unsigned int i;
short elements[8];
//vector signed char v;
vector signed short v;
}shortBuffer;
union
{
//unsigned int i;
unsigned char elements[16];
//vector signed char v;
vector unsigned char v;
}charBuffer;
//vector unsigned char c;
register vector signed short d, hiImage, loImage, YRight, UVRight, YImage, UVImage, UVTemp, YTemp;
// vector unsigned char zero = vec_splat_u8(0);
register vector unsigned char c,one;
// vector signed short zshort = vec_splat_s16(0);
vector unsigned char *inData = (vector unsigned char*) image.data;
vector unsigned char *rightData = (vector unsigned char*) right.data;
//Write the pixel (pair) to the transfer buffer
charBuffer.elements[0] = 2;
charBuffer.elements[1] = 1;
charBuffer.elements[2] = 2;
charBuffer.elements[3] = 1;
charBuffer.elements[4] = 2;
charBuffer.elements[5] = 1;
charBuffer.elements[6] = 2;
charBuffer.elements[7] = 1;
charBuffer.elements[8] = 2;
charBuffer.elements[9] = 1;
charBuffer.elements[10] = 2;
charBuffer.elements[11] = 1;
charBuffer.elements[12] = 2;
charBuffer.elements[13] = 1;
charBuffer.elements[14] = 2;
charBuffer.elements[15] = 1;
//Load it into the vector unit
c = charBuffer.v;
one = vec_splat_u8( 1 );
shortBuffer.elements[0] = 255;
//Load it into the vector unit
d = shortBuffer.v;
d = static_cast<vector signed short>(vec_splat(static_cast<vector signed short>(d),0));
#ifndef PPC970
UInt32 prefetchSize = GetPrefetchConstant( 16, 1, 256 );
vec_dst( inData, prefetchSize, 0 );
vec_dst( rightData, prefetchSize, 1 );
#endif
for ( h=0; h<image.ysize; h++){
for (w=0; w<width; w++)
{
#ifndef PPC970
vec_dst( inData, prefetchSize, 0 );
vec_dst( rightData, prefetchSize, 1 );
#endif
//interleaved U Y V Y chars
//vec_mule UV * 2 to short vector U V U V shorts
UVImage = static_cast<vector signed short>(vec_mule(one,inData[0]));
UVRight = static_cast<vector signed short>(vec_mule(c,rightData[0]));
//vec_mulo Y * 1 to short vector Y Y Y Y shorts
YImage = static_cast<vector signed short>(vec_mulo(c,inData[0]));
YRight = static_cast<vector signed short>(vec_mulo(c,rightData[0]));
//vel_subs UV - 255
UVRight = static_cast<vector signed short>(vec_subs(UVRight, d));
//vec_adds UV
UVTemp = vec_adds(UVImage,UVRight);
//vec_adds Y
YTemp = vec_adds(YImage,YRight);
hiImage = vec_mergeh(UVTemp,YTemp);
loImage = vec_mergel(UVTemp,YTemp);
//vec_mergel + vec_mergeh Y and UV
inData[0] = vec_packsu(hiImage, loImage);
inData++;
rightData++;
}
#ifndef PPC970
vec_dss( 0 );
vec_dss( 1 );
#endif
} /*end of working altivec function */
}