static void vector_fmul_reverse_altivec(float *dst, const float *src0,
const float *src1, int len)
{
int i;
vector float d, s0, s1, h0, l0,
s2, s3, zero = (vector float)vec_splat_u32(0);
src1 += len-4;
for(i=0; i<len-7; i+=8) {
s1 = vec_ld(0, src1-i); // [a,b,c,d]
s0 = vec_ld(0, src0+i);
l0 = vec_mergel(s1, s1); // [c,c,d,d]
s3 = vec_ld(-16, src1-i);
h0 = vec_mergeh(s1, s1); // [a,a,b,b]
s2 = vec_ld(16, src0+i);
s1 = vec_mergeh(vec_mergel(l0,h0), // [d,b,d,b]
vec_mergeh(l0,h0)); // [c,a,c,a]
// [d,c,b,a]
l0 = vec_mergel(s3, s3);
d = vec_madd(s0, s1, zero);
h0 = vec_mergeh(s3, s3);
vec_st(d, 0, dst+i);
s3 = vec_mergeh(vec_mergel(l0,h0),
vec_mergeh(l0,h0));
d = vec_madd(s2, s3, zero);
vec_st(d, 16, dst+i);
}
}
static void test()
{
/* Input vectors. */
vector long vla = {-2,-1};
vector long vlb = {0,1};
vector double vda = {-2.0,-1.0};
vector double vdb = {0.0,1.0};
/* Result vectors. */
vector long vlh, vll;
vector double vdh, vdl;
/* Expected result vectors. */
#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
vector long vlrh = {1,-1};
vector long vlrl = {0,-2};
vector double vdrh = {1.0,-1.0};
vector double vdrl = {0.0,-2.0};
#else
vector long vlrh = {-2,0};
vector long vlrl = {-1,1};
vector double vdrh = {-2.0,0.0};
vector double vdrl = {-1.0,1.0};
#endif
vlh = vec_mergeh (vla, vlb);
vll = vec_mergel (vla, vlb);
vdh = vec_mergeh (vda, vdb);
vdl = vec_mergel (vda, vdb);
check (vec_long_eq (vlh, vlrh), "vlh");
check (vec_long_eq (vll, vlrl), "vll");
check (vec_double_eq (vdh, vdrh), "vdh" );
check (vec_double_eq (vdl, vdrl), "vdl" );
}
static void
float_to_int16_interleave_altivec(int16_t *dst, const float **src,
long len, int channels)
{
int i;
vector signed short d0, d1, d2, c0, c1, t0, t1;
vector unsigned char align;
if(channels == 1)
float_to_int16_altivec(dst, src[0], len);
else if (channels == 2)
{
if(((long)dst) & 15)
for(i = 0; i < len - 7; i += 8)
{
d0 = vec_ld(0, dst + i);
t0 = float_to_int16_one_altivec(src[0] + i);
d1 = vec_ld(31, dst + i);
t1 = float_to_int16_one_altivec(src[1] + i);
c0 = vec_mergeh(t0, t1);
c1 = vec_mergel(t0, t1);
d2 = vec_perm(d1, d0, vec_lvsl(0, dst + i));
align = vec_lvsr(0, dst + i);
d0 = vec_perm(d2, c0, align);
d1 = vec_perm(c0, c1, align);
vec_st(d0, 0, dst + i);
d0 = vec_perm(c1, d2, align);
vec_st(d1, 15, dst + i);
vec_st(d0, 31, dst + i);
dst += 8;
}
else
for(i = 0; i < len - 7; i += 8)
{
t0 = float_to_int16_one_altivec(src[0] + i);
t1 = float_to_int16_one_altivec(src[1] + i);
d0 = vec_mergeh(t0, t1);
d1 = vec_mergel(t0, t1);
vec_st(d0, 0, dst + i);
vec_st(d1, 16, dst + i);
dst += 8;
}
}
else
{
DECLARE_ALIGNED(16, int16_t, tmp)[len];
int c, j;
for (c = 0; c < channels; c++)
{
float_to_int16_altivec(tmp, src[c], len);
for (i = 0, j = c; i < len; i++, j += channels)
{
dst[j] = tmp[i];
}
}
}
}
inline v_float32x4 v_reduce_sum4(const v_float32x4& a, const v_float32x4& b,
const v_float32x4& c, const v_float32x4& d)
{
vec_float4 ac = vec_add(vec_mergel(a.val, c.val), vec_mergeh(a.val, c.val));
ac = vec_add(ac, vec_sld(ac, ac, 8));
vec_float4 bd = vec_add(vec_mergel(b.val, d.val), vec_mergeh(b.val, d.val));
bd = vec_add(bd, vec_sld(bd, bd, 8));
return v_float32x4(vec_mergeh(ac, bd));
}
static void test()
{
/* Input vectors. */
vector long long vla = {-2,-1};
vector long long vlb = {0,1};
vector double vda = {-2.0,-1.0};
vector double vdb = {0.0,1.0};
vector unsigned int vuia = {0,1,2,3};
vector unsigned int vuib = {4,5,6,7};
vector signed int vsia = {-4,-3,-2,-1};
vector signed int vsib = {0,1,2,3};
vector float vfa = {-4.0,-3.0,-2.0,-1.0};
vector float vfb = {0.0,1.0,2.0,3.0};
/* Result vectors. */
vector long long vlh, vll;
vector double vdh, vdl;
vector unsigned int vuih, vuil;
vector signed int vsih, vsil;
vector float vfh, vfl;
/* Expected result vectors. */
vector long long vlrh = {-2,0};
vector long long vlrl = {-1,1};
vector double vdrh = {-2.0,0.0};
vector double vdrl = {-1.0,1.0};
vector unsigned int vuirh = {0,4,1,5};
vector unsigned int vuirl = {2,6,3,7};
vector signed int vsirh = {-4,0,-3,1};
vector signed int vsirl = {-2,2,-1,3};
vector float vfrh = {-4.0,0.0,-3.0,1.0};
vector float vfrl = {-2.0,2.0,-1.0,3.0};
vlh = vec_mergeh (vla, vlb);
vll = vec_mergel (vla, vlb);
vdh = vec_mergeh (vda, vdb);
vdl = vec_mergel (vda, vdb);
vuih = vec_mergeh (vuia, vuib);
vuil = vec_mergel (vuia, vuib);
vsih = vec_mergeh (vsia, vsib);
vsil = vec_mergel (vsia, vsib);
vfh = vec_mergeh (vfa, vfb );
vfl = vec_mergel (vfa, vfb );
check (vec_long_long_eq (vlh, vlrh), "vlh");
check (vec_long_long_eq (vll, vlrl), "vll");
check (vec_double_eq (vdh, vdrh), "vdh" );
check (vec_double_eq (vdl, vdrl), "vdl" );
check (vec_all_eq (vuih, vuirh), "vuih");
check (vec_all_eq (vuil, vuirl), "vuil");
check (vec_all_eq (vsih, vsirh), "vsih");
check (vec_all_eq (vsil, vsirl), "vsil");
check (vec_all_eq (vfh, vfrh), "vfh");
check (vec_all_eq (vfl, vfrl), "vfl");
}
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
}
void foo (vector bool long long *vblr,
vector double *vdr, vector unsigned long long *vullz,
vector double *vdz, vector bool char *vbcz,
vector signed char *vscz, vector unsigned char *vucz,
vector bool int *vbiz, vector int *viz,
vector unsigned int *vuiz, vector signed long long int *vslliz,
vector bool short int *vbsiz, vector signed short int *vssiz,
vector unsigned short int *vusiz, vector float *vfz)
{
*vblr++ = vec_andc (vbla, vblb);
*vdr++ = vec_double (vslla);
*vdr++ = vec_double (vulla);
*vblr++ = vec_mergeh (vbla, vblb);
*vblr++ = vec_mergel (vbla, vblb);
*vblr++ = vec_nor (vbla, vblb);
*vblr++ = vec_or (vbla, vblb);
*vblr++ = vec_sel (vbla, vblb, vblc);
*vblr++ = vec_sel (vbla, vblb, vullc);
*vblr++ = vec_xor (vbla, vblb);
*vullz++ = vec_sel (vulla, vullb, vbllc);
*vullz++ = vec_sel (vulla, vullb, vullc);
*vdz++ = vec_sel(vda, vdb, vullc);
*vbcz++ = vec_sel (vbca, vbcb, vbcc);
*vbcz++ = vec_sel (vbca, vbcb, vucc);
*vbcz++ = vec_xor (vbca, vbcb);
*vscz++ = vec_sel (vsca, vscb, vbcc);
*vscz++ = vec_sel (vsca, vscb, vucc);
*vucz++ = vec_sel (vuca, vucb, vbcc);
*vucz++ = vec_sel (vuca, vucb, vucc);
*vbiz++ = vec_sel (vbia, vbib, vbic);
*vbiz++ = vec_sel (vbia, vbib, vuic);
*vbiz++ = vec_xor (vbia, vbib);
*viz++ = vec_sel (vsia, vsib, vbic);
*viz++ = vec_sel (vsia, vsib, vuic);
*vuiz++ = vec_sel (vuia, vuib, vbic);
*vuiz++ = vec_sel (vuia, vuib, vuic);
*vslliz++ = vec_sel(vslla, vsllb, vbllc);
*vslliz++ = vec_sel(vslla, vsllb, vullc);
*vssiz++ = vec_sel(vssia, vssib, vbsic);
*vssiz++ = vec_sel(vssia, vssib, vusic);
*vusiz++ = vec_sel(vusia, vusib, vbsic);
*vusiz++ = vec_sel(vusia, vusib, vusic);
*vbsiz++ = vec_sel (vbsia, vbsib, vbsic);
*vbsiz++ = vec_sel (vbsia, vbsib, vusic);
*vbsiz++ = vec_xor (vbsia, vbsib);
*vdz++ = vec_sel (vda, vdb, vbllc);
*vfz++ = vec_sel (vfa, vfb, vbic);
*vfz++ = vec_sel (vfa, vfb, vuic);
*vfz++ = vec_xor (vfa, vfb);
}
static void float_to_int16_interleave_altivec(int16_t *dst, const float **src,
long len, int channels)
{
int i;
vector signed short d0, d1, d2, c0, c1, t0, t1;
vector unsigned char align;
if (channels == 1)
float_to_int16_altivec(dst, src[0], len);
else {
if (channels == 2) {
if (((long)dst) & 15) {
for (i = 0; i < len - 7; i += 8) {
d0 = vec_ld(0, dst + i);
t0 = float_to_int16_one_altivec(src[0] + i);
d1 = vec_ld(31, dst + i);
t1 = float_to_int16_one_altivec(src[1] + i);
c0 = vec_mergeh(t0, t1);
c1 = vec_mergel(t0, t1);
d2 = vec_perm(d1, d0, vec_lvsl(0, dst + i));
align = vec_lvsr(0, dst + i);
d0 = vec_perm(d2, c0, align);
d1 = vec_perm(c0, c1, align);
vec_st(d0, 0, dst + i);
d0 = vec_perm(c1, d2, align);
vec_st(d1, 15, dst + i);
vec_st(d0, 31, dst + i);
dst += 8;
}
} else {
for (i = 0; i < len - 7; i += 8) {
t0 = float_to_int16_one_altivec(src[0] + i);
t1 = float_to_int16_one_altivec(src[1] + i);
d0 = vec_mergeh(t0, t1);
d1 = vec_mergel(t0, t1);
vec_st(d0, 0, dst + i);
vec_st(d1, 16, dst + i);
dst += 8;
}
}
} else {
for (i = 0; i < channels; i++)
float_to_int16_stride_altivec(dst + i, src[i], len, channels);
}
}
}
static void OSX_AudioIOProc16Bit_Altivec(SInt16 *myInBuffer, float *myOutBuffer)
{
register UInt32 i;
float f = SOUND_BUFFER_SCALE_16BIT;
const vector float gain = vec_load_ps1(&f); // multiplier
const vector float mix = vec_setzero();
if (gBufferMono2Stereo)
{
int j=0;
// TEST: OK
for (i=0;i<SOUND_BUFFER_SIZE;i+=8, j+=16)
{
vector short int v0 = vec_ld(0, myInBuffer + i); // Load 8 shorts
vector float v1 = vec_ctf((vector signed int)vec_unpackh(v0), 0); // convert to float
vector float v2 = vec_ctf((vector signed int)vec_unpackl(v0), 0); // convert to float
vector float v3 = vec_madd(v1, gain, mix); // scale
vector float v4 = vec_madd(v2, gain, mix); // scale
vector float v5 = vec_mergel(v3, v3); // v3(0,0,1,1);
vector float v6 = vec_mergeh(v3, v3); // v3(2,2,3,3);
vector float v7 = vec_mergel(v4, v4); // v4(0,0,1,1);
vector float v8 = vec_mergeh(v4, v4); // v4(2,2,3,3);
vec_st(v5, 0, myOutBuffer + j); // Store 4 floats
vec_st(v6, 0, myOutBuffer + 4 + j); // Store 4 floats
vec_st(v7, 0, myOutBuffer + 8 + j); // Store 4 floats
vec_st(v8, 0, myOutBuffer + 12 + j); // Store 4 floats
}
}
else
{
// TEST: OK
for (i=0;i<SOUND_BUFFER_SIZE;i+=8)
{
vector short int v0 = vec_ld(0, myInBuffer + i); // Load 8 shorts
vector float v1 = vec_ctf((vector signed int)vec_unpackh(v0), 0); // convert to float
vector float v2 = vec_ctf((vector signed int)vec_unpackl(v0), 0); // convert to float
vector float v3 = vec_madd(v1, gain, mix); // scale
vector float v4 = vec_madd(v2, gain, mix); // scale
vec_st(v3, 0, myOutBuffer + i); // Store 4 floats
vec_st(v4, 0, myOutBuffer + 4 + i); // Store 4 floats
}
}
}
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);
}
void v_store_interleave_f32(float *ptr, vector float a, vector float b, vector float c)
{
vector float hbc = vec_mergeh(b, c);
static const vector unsigned char ahbc = {0, 1, 2, 3, 16, 17, 18, 19, 20, 21, 22, 23, 4, 5, 6, 7};
vec_xst(vec_perm(a, hbc, ahbc), 0, ptr);
vector float lab = vec_mergel(a, b);
vec_xst(vec_sld(lab, hbc, 8), 16, ptr);
static const vector unsigned char clab = {8, 9, 10, 11, 24, 25, 26, 27, 28, 29, 30, 31, 12, 13, 14, 15};
vec_xst(vec_perm(c, lab, clab), 32, ptr);
}
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);
}
}
void foo (vector bool long long *vblr,
vector double *vdr)
{
*vblr++ = vec_andc (vbla, vblb);
*vdr++ = vec_double (vsla);
*vdr++ = vec_double (vula);
*vblr++ = vec_mergeh (vbla, vblb);
*vblr++ = vec_mergel (vbla, vblb);
*vblr++ = vec_nor (vbla, vblb);
*vblr++ = vec_or (vbla, vblb);
*vblr++ = vec_sel (vbla, vblb, vblc);
*vblr++ = vec_sel (vbla, vblb, vulc);
*vblr++ = vec_xor (vbla, vblb);
}
/* Place the content of the array of structures
in vectors x_vec, y_vec, z_vec, and t_vec */
int main(int argc, char **argv) {
vector float x_vec, y_vec, z_vec,
t_vec, hold[4], tmp[4];
/* Load structures into vectors */
hold[0] = vec_ld(0, (float*)p_motion);
hold[1] = vec_ld(0, (float*)&p_motion[1]);
hold[2] = vec_ld(0, (float*)&p_motion[2]);
hold[3] = vec_ld(0, (float*)&p_motion[3]);
/* Perform first step of the swizzle */
tmp[0] = vec_mergeh(hold[0], hold[2]);
tmp[1] = vec_mergeh(hold[1], hold[3]);
tmp[2] = vec_mergel(hold[0], hold[2]);
tmp[3] = vec_mergel(hold[1], hold[3]);
/* Perform second step of the swizzle */
x_vec = vec_mergeh(tmp[0], tmp[1]);
y_vec = vec_mergel(tmp[0], tmp[1]);
z_vec = vec_mergeh(tmp[2], tmp[3]);
t_vec = vec_mergel(tmp[2], tmp[3]);
return 0;
}
static int a52_resample_STEREO_to_2_altivec(float * _f, int16_t * s16){
#if 0
int i;
int32_t * f = (int32_t *) _f;
for (i = 0; i < 256; i++) {
s16[2*i] = convert (f[i]);
s16[2*i+1] = convert (f[i+256]);
}
return 2*256;
#else
int i = 0;
int32_t * f = (int32_t *) _f;
register vector signed int f0, f4, f256, f260;
register vector signed short reven, rodd, r0, r1;
for (i = 0; i < 256; i+= 8) {
f0 = vec_ld(0, f);
f4 = vec_ld(16, f);
f256 = vec_ld(1024, f);
f260 = vec_ld(1040, f);
reven = convert16_altivec(f0, f4);
rodd = convert16_altivec(f256, f260);
r0 = vec_mergeh(reven, rodd);
r1 = vec_mergel(reven, rodd);
// FIXME can be merged to spare some I/O
unaligned_store(r0, 0, s16);
unaligned_store(r1, 16, s16);
f += 8;
s16 += 16;
}
return(2*256);
#endif
}
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;
}
}
tmp1 = vec_perm(table[6], table[7], tmpIndex);
stmp1 = vec_perm(slope_cos[6], slope_cos[7], tmpIndex);
select = (vector unsigned short)vec_cmpgt(PerIndex, (((vector unsigned char){95,95,95,95,95,95,95,95,95,95,95,95,95,95,95,95})) );
tmp3 = vec_sel(tmp0, tmp1, select);
stmp3 = vec_sel(stmp0, stmp1, select);
select = (vector unsigned short)vec_cmpgt(PerIndex, (((vector unsigned char){63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63})) );
table1 = vec_sel(tmp2, tmp3, select);
slope1 = vec_sel(stmp2, stmp3, select);
L_tmp0 = vec_sra(vec_mule(slope0, offset0), (((vector unsigned int){12,12,12,12})) );
L_tmp1 = vec_sra(vec_mulo(slope0, offset0), (((vector unsigned int){12,12,12,12})) );
L_tmp2 = vec_sra(vec_mule(slope1, offset1), (((vector unsigned int){12,12,12,12})) );
L_tmp3 = vec_sra(vec_mulo(slope1, offset1), (((vector unsigned int){12,12,12,12})) );
tmp0 = vec_packs(L_tmp0, L_tmp2);
tmp1 = vec_packs(L_tmp1, L_tmp3);
tmp2 = vec_mergeh(tmp0, tmp1);
tmp3 = vec_mergel(tmp0, tmp1);
lspq[0] = vec_adds(table0, tmp2);
lspq[1] = vec_adds(table1, tmp3);
return;
}
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
}
}
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 */
}
void ff_fdct_altivec(int16_t *block)
{
vector signed short *bp;
vector float *cp;
vector float b00, b10, b20, b30, b40, b50, b60, b70;
vector float b01, b11, b21, b31, b41, b51, b61, b71;
vector float mzero, cnst, cnsts0, cnsts1, cnsts2;
vector float x0, x1, x2, x3, x4, x5, x6, x7, x8;
/* setup constants {{{ */
/* mzero = -0.0 */
mzero = ((vector float)vec_splat_u32(-1));
mzero = ((vector float)vec_sl(vu32(mzero), vu32(mzero)));
cp = fdctconsts;
cnsts0 = vec_ld(0, cp); cp++;
cnsts1 = vec_ld(0, cp); cp++;
cnsts2 = vec_ld(0, cp);
/* }}} */
/* 8x8 matrix transpose (vector short[8]) {{{ */
#define MERGE_S16(hl,a,b) vec_merge##hl(vs16(a), vs16(b))
bp = (vector signed short*)block;
b00 = ((vector float)vec_ld(0, bp));
b40 = ((vector float)vec_ld(16*4, bp));
b01 = ((vector float)MERGE_S16(h, b00, b40));
b11 = ((vector float)MERGE_S16(l, b00, b40));
bp++;
b10 = ((vector float)vec_ld(0, bp));
b50 = ((vector float)vec_ld(16*4, bp));
b21 = ((vector float)MERGE_S16(h, b10, b50));
b31 = ((vector float)MERGE_S16(l, b10, b50));
bp++;
b20 = ((vector float)vec_ld(0, bp));
b60 = ((vector float)vec_ld(16*4, bp));
b41 = ((vector float)MERGE_S16(h, b20, b60));
b51 = ((vector float)MERGE_S16(l, b20, b60));
bp++;
b30 = ((vector float)vec_ld(0, bp));
b70 = ((vector float)vec_ld(16*4, bp));
b61 = ((vector float)MERGE_S16(h, b30, b70));
b71 = ((vector float)MERGE_S16(l, b30, b70));
x0 = ((vector float)MERGE_S16(h, b01, b41));
x1 = ((vector float)MERGE_S16(l, b01, b41));
x2 = ((vector float)MERGE_S16(h, b11, b51));
x3 = ((vector float)MERGE_S16(l, b11, b51));
x4 = ((vector float)MERGE_S16(h, b21, b61));
x5 = ((vector float)MERGE_S16(l, b21, b61));
x6 = ((vector float)MERGE_S16(h, b31, b71));
x7 = ((vector float)MERGE_S16(l, b31, b71));
b00 = ((vector float)MERGE_S16(h, x0, x4));
b10 = ((vector float)MERGE_S16(l, x0, x4));
b20 = ((vector float)MERGE_S16(h, x1, x5));
b30 = ((vector float)MERGE_S16(l, x1, x5));
b40 = ((vector float)MERGE_S16(h, x2, x6));
b50 = ((vector float)MERGE_S16(l, x2, x6));
b60 = ((vector float)MERGE_S16(h, x3, x7));
b70 = ((vector float)MERGE_S16(l, x3, x7));
#undef MERGE_S16
/* }}} */
/* Some of the initial calculations can be done as vector short before
* conversion to vector float. The following code section takes advantage
* of this.
*/
/* fdct rows {{{ */
x0 = ((vector float)vec_add(vs16(b00), vs16(b70)));
x7 = ((vector float)vec_sub(vs16(b00), vs16(b70)));
x1 = ((vector float)vec_add(vs16(b10), vs16(b60)));
x6 = ((vector float)vec_sub(vs16(b10), vs16(b60)));
x2 = ((vector float)vec_add(vs16(b20), vs16(b50)));
x5 = ((vector float)vec_sub(vs16(b20), vs16(b50)));
x3 = ((vector float)vec_add(vs16(b30), vs16(b40)));
x4 = ((vector float)vec_sub(vs16(b30), vs16(b40)));
b70 = ((vector float)vec_add(vs16(x0), vs16(x3)));
b10 = ((vector float)vec_add(vs16(x1), vs16(x2)));
b00 = ((vector float)vec_add(vs16(b70), vs16(b10)));
b40 = ((vector float)vec_sub(vs16(b70), vs16(b10)));
#define CTF0(n) \
b##n##1 = ((vector float)vec_unpackl(vs16(b##n##0))); \
b##n##0 = ((vector float)vec_unpackh(vs16(b##n##0))); \
b##n##1 = vec_ctf(vs32(b##n##1), 0); \
b##n##0 = vec_ctf(vs32(b##n##0), 0);
CTF0(0);
CTF0(4);
b20 = ((vector float)vec_sub(vs16(x0), vs16(x3)));
b60 = ((vector float)vec_sub(vs16(x1), vs16(x2)));
CTF0(2);
CTF0(6);
#undef CTF0
x0 = vec_add(b60, b20);
x1 = vec_add(b61, b21);
cnst = LD_W2;
x0 = vec_madd(cnst, x0, mzero);
x1 = vec_madd(cnst, x1, mzero);
cnst = LD_W1;
b20 = vec_madd(cnst, b20, x0);
b21 = vec_madd(cnst, b21, x1);
cnst = LD_W0;
b60 = vec_madd(cnst, b60, x0);
b61 = vec_madd(cnst, b61, x1);
#define CTFX(x,b) \
b##0 = ((vector float)vec_unpackh(vs16(x))); \
b##1 = ((vector float)vec_unpackl(vs16(x))); \
b##0 = vec_ctf(vs32(b##0), 0); \
b##1 = vec_ctf(vs32(b##1), 0); \
CTFX(x4, b7);
CTFX(x5, b5);
CTFX(x6, b3);
CTFX(x7, b1);
#undef CTFX
x0 = vec_add(b70, b10);
x1 = vec_add(b50, b30);
x2 = vec_add(b70, b30);
x3 = vec_add(b50, b10);
x8 = vec_add(x2, x3);
cnst = LD_W3;
x8 = vec_madd(cnst, x8, mzero);
cnst = LD_W8;
x0 = vec_madd(cnst, x0, mzero);
cnst = LD_W9;
x1 = vec_madd(cnst, x1, mzero);
cnst = LD_WA;
x2 = vec_madd(cnst, x2, x8);
cnst = LD_WB;
x3 = vec_madd(cnst, x3, x8);
cnst = LD_W4;
b70 = vec_madd(cnst, b70, x0);
cnst = LD_W5;
b50 = vec_madd(cnst, b50, x1);
cnst = LD_W6;
b30 = vec_madd(cnst, b30, x1);
cnst = LD_W7;
b10 = vec_madd(cnst, b10, x0);
b70 = vec_add(b70, x2);
b50 = vec_add(b50, x3);
b30 = vec_add(b30, x2);
b10 = vec_add(b10, x3);
x0 = vec_add(b71, b11);
x1 = vec_add(b51, b31);
x2 = vec_add(b71, b31);
x3 = vec_add(b51, b11);
x8 = vec_add(x2, x3);
cnst = LD_W3;
x8 = vec_madd(cnst, x8, mzero);
cnst = LD_W8;
x0 = vec_madd(cnst, x0, mzero);
cnst = LD_W9;
x1 = vec_madd(cnst, x1, mzero);
cnst = LD_WA;
x2 = vec_madd(cnst, x2, x8);
cnst = LD_WB;
x3 = vec_madd(cnst, x3, x8);
cnst = LD_W4;
b71 = vec_madd(cnst, b71, x0);
cnst = LD_W5;
b51 = vec_madd(cnst, b51, x1);
cnst = LD_W6;
b31 = vec_madd(cnst, b31, x1);
cnst = LD_W7;
b11 = vec_madd(cnst, b11, x0);
b71 = vec_add(b71, x2);
b51 = vec_add(b51, x3);
b31 = vec_add(b31, x2);
b11 = vec_add(b11, x3);
/* }}} */
/* 8x8 matrix transpose (vector float[8][2]) {{{ */
x0 = vec_mergel(b00, b20);
x1 = vec_mergeh(b00, b20);
x2 = vec_mergel(b10, b30);
x3 = vec_mergeh(b10, b30);
b00 = vec_mergeh(x1, x3);
b10 = vec_mergel(x1, x3);
b20 = vec_mergeh(x0, x2);
b30 = vec_mergel(x0, x2);
x4 = vec_mergel(b41, b61);
x5 = vec_mergeh(b41, b61);
x6 = vec_mergel(b51, b71);
x7 = vec_mergeh(b51, b71);
b41 = vec_mergeh(x5, x7);
b51 = vec_mergel(x5, x7);
b61 = vec_mergeh(x4, x6);
b71 = vec_mergel(x4, x6);
x0 = vec_mergel(b01, b21);
x1 = vec_mergeh(b01, b21);
x2 = vec_mergel(b11, b31);
x3 = vec_mergeh(b11, b31);
x4 = vec_mergel(b40, b60);
x5 = vec_mergeh(b40, b60);
x6 = vec_mergel(b50, b70);
x7 = vec_mergeh(b50, b70);
b40 = vec_mergeh(x1, x3);
b50 = vec_mergel(x1, x3);
b60 = vec_mergeh(x0, x2);
b70 = vec_mergel(x0, x2);
b01 = vec_mergeh(x5, x7);
b11 = vec_mergel(x5, x7);
b21 = vec_mergeh(x4, x6);
b31 = vec_mergel(x4, x6);
/* }}} */
FDCTCOL(b00, b10, b20, b30, b40, b50, b60, b70);
FDCTCOL(b01, b11, b21, b31, b41, b51, b61, b71);
/* round, convert back to short {{{ */
#define CTS(n) \
b##n##0 = vec_round(b##n##0); \
b##n##1 = vec_round(b##n##1); \
b##n##0 = ((vector float)vec_cts(b##n##0, 0)); \
b##n##1 = ((vector float)vec_cts(b##n##1, 0)); \
b##n##0 = ((vector float)vec_pack(vs32(b##n##0), vs32(b##n##1))); \
vec_st(vs16(b##n##0), 0, bp);
bp = (vector signed short*)block;
CTS(0); bp++;
CTS(1); bp++;
CTS(2); bp++;
CTS(3); bp++;
CTS(4); bp++;
CTS(5); bp++;
CTS(6); bp++;
CTS(7);
#undef CTS
/* }}} */
}
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;
}
void test1() {
// CHECK-LABEL: define void @test1
// CHECK-LE-LABEL: define void @test1
res_vf = vec_abs(vf);
// CHECK: call <4 x float> @llvm.fabs.v4f32(<4 x float> %{{[0-9]*}})
// CHECK-LE: call <4 x float> @llvm.fabs.v4f32(<4 x float> %{{[0-9]*}})
dummy();
// CHECK: call void @dummy()
// CHECK-LE: call void @dummy()
res_vd = vec_add(vd, vd);
// CHECK: fadd <2 x double>
// CHECK-LE: fadd <2 x double>
res_vd = vec_and(vbll, vd);
// CHECK: and <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
// CHECK-LE: and <2 x i64>
// CHECK-LE: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
res_vd = vec_and(vd, vbll);
// CHECK: and <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
// CHECK-LE: and <2 x i64>
// CHECK-LE: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
res_vd = vec_and(vd, vd);
// CHECK: and <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
// CHECK-LE: and <2 x i64>
// CHECK-LE: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
dummy();
// CHECK: call void @dummy()
// CHECK-LE: call void @dummy()
res_vd = vec_andc(vbll, vd);
// CHECK: bitcast <2 x double> %{{[0-9]*}} to <2 x i64>
// CHECK: xor <2 x i64> %{{[0-9]*}}, <i64 -1, i64 -1>
// CHECK: and <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
// CHECK-LE: bitcast <2 x double> %{{[0-9]*}} to <2 x i64>
// CHECK-LE: xor <2 x i64> %{{[0-9]*}}, <i64 -1, i64 -1>
// CHECK-LE: and <2 x i64>
// CHECK-LE: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
dummy();
// CHECK: call void @dummy()
// CHECK-LE: call void @dummy()
res_vd = vec_andc(vd, vbll);
// CHECK: bitcast <2 x double> %{{[0-9]*}} to <2 x i64>
// CHECK: xor <2 x i64> %{{[0-9]*}}, <i64 -1, i64 -1>
// CHECK: and <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
// CHECK-LE: bitcast <2 x double> %{{[0-9]*}} to <2 x i64>
// CHECK-LE: xor <2 x i64> %{{[0-9]*}}, <i64 -1, i64 -1>
// CHECK-LE: and <2 x i64>
// CHECK-LE: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
dummy();
// CHECK: call void @dummy()
res_vd = vec_andc(vd, vd);
// CHECK: bitcast <2 x double> %{{[0-9]*}} to <2 x i64>
// CHECK: xor <2 x i64> %{{[0-9]*}}, <i64 -1, i64 -1>
// CHECK: and <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
dummy();
// CHECK: call void @dummy()
// CHECK-LE: call void @dummy()
res_vd = vec_ceil(vd);
// CHECK: call <2 x double> @llvm.ceil.v2f64(<2 x double> %{{[0-9]*}})
// CHECK-LE: call <2 x double> @llvm.ceil.v2f64(<2 x double> %{{[0-9]*}})
res_vf = vec_ceil(vf);
// CHECK: call <4 x float> @llvm.ceil.v4f32(<4 x float> %{{[0-9]*}})
// CHECK-LE: call <4 x float> @llvm.ceil.v4f32(<4 x float> %{{[0-9]*}})
res_vbll = vec_cmpeq(vd, vd);
// CHECK: call <2 x i64> @llvm.ppc.vsx.xvcmpeqdp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
// CHECK-LE: call <2 x i64> @llvm.ppc.vsx.xvcmpeqdp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
res_vbi = vec_cmpeq(vf, vf);
// CHECK: call <4 x i32> @llvm.ppc.vsx.xvcmpeqsp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
// CHECK-LE: call <4 x i32> @llvm.ppc.vsx.xvcmpeqsp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
res_vbll = vec_cmpge(vd, vd);
// CHECK: call <2 x i64> @llvm.ppc.vsx.xvcmpgedp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
// CHECK-LE: call <2 x i64> @llvm.ppc.vsx.xvcmpgedp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
res_vbi = vec_cmpge(vf, vf);
// CHECK: call <4 x i32> @llvm.ppc.vsx.xvcmpgesp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
// CHECK-LE: call <4 x i32> @llvm.ppc.vsx.xvcmpgesp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
res_vbll = vec_cmpgt(vd, vd);
// CHECK: call <2 x i64> @llvm.ppc.vsx.xvcmpgtdp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
// CHECK-LE: call <2 x i64> @llvm.ppc.vsx.xvcmpgtdp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
res_vbi = vec_cmpgt(vf, vf);
// CHECK: call <4 x i32> @llvm.ppc.vsx.xvcmpgtsp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
// CHECK-LE: call <4 x i32> @llvm.ppc.vsx.xvcmpgtsp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
res_vbll = vec_cmple(vd, vd);
// CHECK: call <2 x i64> @llvm.ppc.vsx.xvcmpgedp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
// CHECK-LE: call <2 x i64> @llvm.ppc.vsx.xvcmpgedp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
res_vbi = vec_cmple(vf, vf);
// CHECK: call <4 x i32> @llvm.ppc.vsx.xvcmpgesp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
// CHECK-LE: call <4 x i32> @llvm.ppc.vsx.xvcmpgesp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
res_vbll = vec_cmplt(vd, vd);
// CHECK: call <2 x i64> @llvm.ppc.vsx.xvcmpgtdp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
// CHECK-LE: call <2 x i64> @llvm.ppc.vsx.xvcmpgtdp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
res_vbi = vec_cmplt(vf, vf);
// CHECK: call <4 x i32> @llvm.ppc.vsx.xvcmpgtsp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
// CHECK-LE: call <4 x i32> @llvm.ppc.vsx.xvcmpgtsp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
/* vec_cpsgn */
res_vf = vec_cpsgn(vf, vf);
// CHECK: call <4 x float> @llvm.copysign.v4f32(<4 x float> %{{.+}}, <4 x float> %{{.+}})
// CHECK-LE: call <4 x float> @llvm.copysign.v4f32(<4 x float> %{{.+}}, <4 x float> %{{.+}})
res_vd = vec_cpsgn(vd, vd);
// CHECK: call <2 x double> @llvm.copysign.v2f64(<2 x double> %{{.+}}, <2 x double> %{{.+}})
// CHECK-LE: call <2 x double> @llvm.copysign.v2f64(<2 x double> %{{.+}}, <2 x double> %{{.+}})
/* vec_div */
res_vsll = vec_div(vsll, vsll);
// CHECK: sdiv <2 x i64>
// CHECK-LE: sdiv <2 x i64>
res_vull = vec_div(vull, vull);
// CHECK: udiv <2 x i64>
// CHECK-LE: udiv <2 x i64>
res_vf = vec_div(vf, vf);
// CHECK: fdiv <4 x float>
// CHECK-LE: fdiv <4 x float>
res_vd = vec_div(vd, vd);
// CHECK: fdiv <2 x double>
// CHECK-LE: fdiv <2 x double>
/* vec_max */
res_vf = vec_max(vf, vf);
// CHECK: @llvm.ppc.vsx.xvmaxsp
// CHECK-LE: @llvm.ppc.vsx.xvmaxsp
res_vd = vec_max(vd, vd);
// CHECK: @llvm.ppc.vsx.xvmaxdp
// CHECK-LE: @llvm.ppc.vsx.xvmaxdp
res_vf = vec_vmaxfp(vf, vf);
// CHECK: @llvm.ppc.vsx.xvmaxsp
// CHECK-LE: @llvm.ppc.vsx.xvmaxsp
/* vec_min */
res_vf = vec_min(vf, vf);
// CHECK: @llvm.ppc.vsx.xvminsp
// CHECK-LE: @llvm.ppc.vsx.xvminsp
res_vd = vec_min(vd, vd);
// CHECK: @llvm.ppc.vsx.xvmindp
// CHECK-LE: @llvm.ppc.vsx.xvmindp
res_vf = vec_vminfp(vf, vf);
// CHECK: @llvm.ppc.vsx.xvminsp
// CHECK-LE: @llvm.ppc.vsx.xvminsp
res_d = __builtin_vsx_xsmaxdp(d, d);
// CHECK: @llvm.ppc.vsx.xsmaxdp
// CHECK-LE: @llvm.ppc.vsx.xsmaxdp
res_d = __builtin_vsx_xsmindp(d, d);
// CHECK: @llvm.ppc.vsx.xsmindp
// CHECK-LE: @llvm.ppc.vsx.xsmindp
/* vec_perm */
res_vsll = vec_perm(vsll, vsll, vuc);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vull = vec_perm(vull, vull, vuc);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vbll = vec_perm(vbll, vbll, vuc);
// CHECK: [[T1:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK: [[T2:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> [[T1]], <4 x i32> [[T2]], <16 x i8>
// CHECK-LE: xor <16 x i8>
// CHECK-LE: [[T1:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK-LE: [[T2:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK-LE: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> [[T1]], <4 x i32> [[T2]], <16 x i8>
res_vf = vec_round(vf);
// CHECK: call <4 x float> @llvm.round.v4f32(<4 x float>
// CHECK-LE: call <4 x float> @llvm.round.v4f32(<4 x float>
res_vd = vec_round(vd);
// CHECK: call <2 x double> @llvm.round.v2f64(<2 x double>
// CHECK-LE: call <2 x double> @llvm.round.v2f64(<2 x double>
res_vd = vec_perm(vd, vd, vuc);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vd = vec_splat(vd, 1);
// CHECK: [[T1:%.+]] = bitcast <2 x double> {{.+}} to <4 x i32>
// CHECK: [[T2:%.+]] = bitcast <2 x double> {{.+}} to <4 x i32>
// CHECK: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> [[T1]], <4 x i32> [[T2]], <16 x i8>
// CHECK-LE: xor <16 x i8>
// CHECK-LE: [[T1:%.+]] = bitcast <2 x double> {{.+}} to <4 x i32>
// CHECK-LE: [[T2:%.+]] = bitcast <2 x double> {{.+}} to <4 x i32>
// CHECK-LE: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> [[T1]], <4 x i32> [[T2]], <16 x i8>
res_vbll = vec_splat(vbll, 1);
// CHECK: [[T1:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK: [[T2:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> [[T1]], <4 x i32> [[T2]], <16 x i8>
// CHECK-LE: xor <16 x i8>
// CHECK-LE: [[T1:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK-LE: [[T2:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK-LE: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> [[T1]], <4 x i32> [[T2]], <16 x i8>
res_vsll = vec_splat(vsll, 1);
// CHECK: [[T1:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK: [[T2:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> [[T1]], <4 x i32> [[T2]], <16 x i8>
// CHECK-LE: xor <16 x i8>
// CHECK-LE: [[T1:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK-LE: [[T2:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK-LE: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> [[T1]], <4 x i32> [[T2]], <16 x i8>
res_vull = vec_splat(vull, 1);
// CHECK: [[T1:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK: [[T2:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> [[T1]], <4 x i32> [[T2]], <16 x i8>
// CHECK-LE: xor <16 x i8>
// CHECK-LE: [[T1:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK-LE: [[T2:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK-LE: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> [[T1]], <4 x i32> [[T2]], <16 x i8>
res_vsi = vec_pack(vsll, vsll);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vui = vec_pack(vull, vull);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vbi = vec_pack(vbll, vbll);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vsll = vec_vperm(vsll, vsll, vuc);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vull = vec_vperm(vull, vull, vuc);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vd = vec_vperm(vd, vd, vuc);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
/* vec_vsx_ld */
res_vsi = vec_vsx_ld(0, &vsi);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
res_vui = vec_vsx_ld(0, &vui);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
res_vf = vec_vsx_ld (0, &vf);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
res_vsll = vec_vsx_ld(0, &vsll);
// CHECK: @llvm.ppc.vsx.lxvd2x
// CHECK-LE: @llvm.ppc.vsx.lxvd2x
res_vull = vec_vsx_ld(0, &vull);
// CHECK: @llvm.ppc.vsx.lxvd2x
// CHECK-LE: @llvm.ppc.vsx.lxvd2x
res_vd = vec_vsx_ld(0, &vd);
// CHECK: @llvm.ppc.vsx.lxvd2x
// CHECK-LE: @llvm.ppc.vsx.lxvd2x
res_vull = vec_vsx_ld(0, &vull);
// CHECK: @llvm.ppc.vsx.lxvd2x
// CHECK-LE: @llvm.ppc.vsx.lxvd2x
res_vd = vec_vsx_ld(0, &vd);
// CHECK: @llvm.ppc.vsx.lxvd2x
// CHECK-LE: @llvm.ppc.vsx.lxvd2x
res_vss = vec_vsx_ld(0, &vss);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
res_vss = vec_vsx_ld(0, &ss);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
res_vus = vec_vsx_ld(0, &vus);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
res_vus = vec_vsx_ld(0, &us);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
res_vbc = vec_vsx_ld(0, &vbc);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
res_vsc = vec_vsx_ld(0, &vsc);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
res_vuc = vec_vsx_ld(0, &vuc);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
res_vsc = vec_vsx_ld(0, &sc);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
res_vuc = vec_vsx_ld(0, &uc);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
/* vec_vsx_st */
vec_vsx_st(vsi, 0, &res_vsi);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vsi, 0, &res_si);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vui, 0, &res_vui);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vui, 0, &res_ui);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vf, 0, &res_vf);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vsll, 0, &res_vsll);
// CHECK: @llvm.ppc.vsx.stxvd2x
// CHECK-LE: @llvm.ppc.vsx.stxvd2x
vec_vsx_st(vull, 0, &res_vull);
// CHECK: @llvm.ppc.vsx.stxvd2x
// CHECK-LE: @llvm.ppc.vsx.stxvd2x
vec_vsx_st(vd, 0, &res_vd);
// CHECK: @llvm.ppc.vsx.stxvd2x
// CHECK-LE: @llvm.ppc.vsx.stxvd2x
vec_vsx_st(vss, 0, &res_vss);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vss, 0, &res_ss);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vus, 0, &res_vus);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vus, 0, &res_us);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vsc, 0, &res_vsc);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vsc, 0, &res_sc);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vuc, 0, &res_vuc);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vuc, 0, &res_uc);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vbc, 0, &res_vbc);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vbc, 0, &res_sc);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vbc, 0, &res_uc);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
/* vec_and */
res_vsll = vec_and(vsll, vsll);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vsll = vec_and(vbll, vsll);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vsll = vec_and(vsll, vbll);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vull = vec_and(vull, vull);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vull = vec_and(vbll, vull);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vull = vec_and(vull, vbll);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vbll = vec_and(vbll, vbll);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
/* vec_vand */
res_vsll = vec_vand(vsll, vsll);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vsll = vec_vand(vbll, vsll);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vsll = vec_vand(vsll, vbll);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vull = vec_vand(vull, vull);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vull = vec_vand(vbll, vull);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vull = vec_vand(vull, vbll);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vbll = vec_vand(vbll, vbll);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
/* vec_andc */
res_vsll = vec_andc(vsll, vsll);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
// CHECK-LE: xor <2 x i64>
// CHECK-LE: and <2 x i64>
res_vsll = vec_andc(vbll, vsll);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
// CHECK-LE: xor <2 x i64>
// CHECK-LE: and <2 x i64>
res_vsll = vec_andc(vsll, vbll);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
// CHECK-LE: xor <2 x i64>
// CHECK-LE: and <2 x i64>
res_vull = vec_andc(vull, vull);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
// CHECK-LE: xor <2 x i64>
// CHECK-LE: and <2 x i64>
res_vull = vec_andc(vbll, vull);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
// CHECK-LE: xor <2 x i64>
// CHECK-LE: and <2 x i64>
res_vull = vec_andc(vull, vbll);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
// CHECK-LE: xor <2 x i64>
// CHECK-LE: and <2 x i64>
res_vbll = vec_andc(vbll, vbll);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
// CHECK-LE: xor <2 x i64>
// CHECK-LE: and <2 x i64>
res_vf = vec_floor(vf);
// CHECK: call <4 x float> @llvm.floor.v4f32(<4 x float> %{{[0-9]+}})
// CHECK-LE: call <4 x float> @llvm.floor.v4f32(<4 x float> %{{[0-9]+}})
res_vd = vec_floor(vd);
// CHECK: call <2 x double> @llvm.floor.v2f64(<2 x double> %{{[0-9]+}})
// CHECK-LE: call <2 x double> @llvm.floor.v2f64(<2 x double> %{{[0-9]+}})
res_vf = vec_madd(vf, vf, vf);
// CHECK: call <4 x float> @llvm.fma.v4f32(<4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}})
// CHECK-LE: call <4 x float> @llvm.fma.v4f32(<4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}})
res_vd = vec_madd(vd, vd, vd);
// CHECK: call <2 x double> @llvm.fma.v2f64(<2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}})
// CHECK-LE: call <2 x double> @llvm.fma.v2f64(<2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}})
/* vec_mergeh */
res_vsll = vec_mergeh(vsll, vsll);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vsll = vec_mergeh(vsll, vbll);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vsll = vec_mergeh(vbll, vsll);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vull = vec_mergeh(vull, vull);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vull = vec_mergeh(vull, vbll);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vull = vec_mergeh(vbll, vull);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
/* vec_mergel */
res_vsll = vec_mergel(vsll, vsll);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vsll = vec_mergel(vsll, vbll);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vsll = vec_mergel(vbll, vsll);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vull = vec_mergel(vull, vull);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vull = vec_mergel(vull, vbll);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vull = vec_mergel(vbll, vull);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
/* vec_msub */
res_vf = vec_msub(vf, vf, vf);
// CHECK: fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %{{[0-9]+}}
// CHECK-NEXT: call <4 x float> @llvm.fma.v4f32(<4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}}, <4 x float>
// CHECK-LE: fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %{{[0-9]+}}
// CHECK-LE-NEXT: call <4 x float> @llvm.fma.v4f32(<4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}}, <4 x float>
res_vd = vec_msub(vd, vd, vd);
// CHECK: fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %{{[0-9]+}}
// CHECK-NEXT: call <2 x double> @llvm.fma.v2f64(<2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}}, <2 x double>
// CHECK-LE: fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %{{[0-9]+}}
// CHECK-LE-NEXT: call <2 x double> @llvm.fma.v2f64(<2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}}, <2 x double>
res_vsll = vec_mul(vsll, vsll);
// CHECK: mul <2 x i64>
// CHECK-LE: mul <2 x i64>
res_vull = vec_mul(vull, vull);
// CHECK: mul <2 x i64>
// CHECK-LE: mul <2 x i64>
res_vf = vec_mul(vf, vf);
// CHECK: fmul <4 x float> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-LE: fmul <4 x float> %{{[0-9]+}}, %{{[0-9]+}}
res_vd = vec_mul(vd, vd);
// CHECK: fmul <2 x double> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-LE: fmul <2 x double> %{{[0-9]+}}, %{{[0-9]+}}
res_vf = vec_nearbyint(vf);
// CHECK: call <4 x float> @llvm.round.v4f32(<4 x float> %{{[0-9]+}})
// CHECK-LE: call <4 x float> @llvm.round.v4f32(<4 x float> %{{[0-9]+}})
res_vd = vec_nearbyint(vd);
// CHECK: call <2 x double> @llvm.round.v2f64(<2 x double> %{{[0-9]+}})
// CHECK-LE: call <2 x double> @llvm.round.v2f64(<2 x double> %{{[0-9]+}})
res_vf = vec_nmadd(vf, vf, vf);
// CHECK: [[FM:[0-9]+]] = call <4 x float> @llvm.fma.v4f32(<4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}})
// CHECK-NEXT: fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %[[FM]]
// CHECK-LE: [[FM:[0-9]+]] = call <4 x float> @llvm.fma.v4f32(<4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}})
// CHECK-LE-NEXT: fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %[[FM]]
res_vd = vec_nmadd(vd, vd, vd);
// CHECK: [[FM:[0-9]+]] = call <2 x double> @llvm.fma.v2f64(<2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}})
// CHECK-NEXT: fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %[[FM]]
// CHECK-LE: [[FM:[0-9]+]] = call <2 x double> @llvm.fma.v2f64(<2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}})
// CHECK-LE-NEXT: fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %[[FM]]
res_vf = vec_nmsub(vf, vf, vf);
// CHECK: fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %{{[0-9]+}}
// CHECK-NEXT: call <4 x float> @llvm.fma.v4f32(<4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}}, <4 x float>
// CHECK: fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %{{[0-9]+}}
// CHECK-LE: fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %{{[0-9]+}}
// CHECK-LE-NEXT: call <4 x float> @llvm.fma.v4f32(<4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}}, <4 x float>
// CHECK-LE: fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %{{[0-9]+}}
res_vd = vec_nmsub(vd, vd, vd);
// CHECK: fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %{{[0-9]+}}
// CHECK-NEXT: [[FM:[0-9]+]] = call <2 x double> @llvm.fma.v2f64(<2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}}, <2 x double>
// CHECK-NEXT: fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %[[FM]]
// CHECK-LE: fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %{{[0-9]+}}
// CHECK-LE-NEXT: [[FM:[0-9]+]] = call <2 x double> @llvm.fma.v2f64(<2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}}, <2 x double>
// CHECK-LE-NEXT: fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %[[FM]]
/* vec_nor */
res_vsll = vec_nor(vsll, vsll);
// CHECK: or <2 x i64>
// CHECK: xor <2 x i64>
// CHECK-LE: or <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vull = vec_nor(vull, vull);
// CHECK: or <2 x i64>
// CHECK: xor <2 x i64>
// CHECK-LE: or <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vull = vec_nor(vbll, vbll);
// CHECK: or <2 x i64>
// CHECK: xor <2 x i64>
// CHECK-LE: or <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vd = vec_nor(vd, vd);
// CHECK: bitcast <2 x double> %{{[0-9]+}} to <2 x i64>
// CHECK: [[OR:%.+]] = or <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-NEXT: xor <2 x i64> [[OR]], <i64 -1, i64 -1>
// CHECK-LE: bitcast <2 x double> %{{[0-9]+}} to <2 x i64>
// CHECK-LE: [[OR:%.+]] = or <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-LE-NEXT: xor <2 x i64> [[OR]], <i64 -1, i64 -1>
/* vec_or */
res_vsll = vec_or(vsll, vsll);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vsll = vec_or(vbll, vsll);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vsll = vec_or(vsll, vbll);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vull = vec_or(vull, vull);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vull = vec_or(vbll, vull);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vull = vec_or(vull, vbll);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vbll = vec_or(vbll, vbll);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vd = vec_or(vd, vd);
// CHECK: bitcast <2 x double> %{{[0-9]+}} to <2 x i64>
// CHECK: or <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-LE: bitcast <2 x double> %{{[0-9]+}} to <2 x i64>
// CHECK-LE: or <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
res_vd = vec_or(vbll, vd);
// CHECK: [[T1:%.+]] = bitcast <2 x double> %{{[0-9]+}} to <2 x i64>
// CHECK: [[T2:%.+]] = or <2 x i64> %{{[0-9]+}}, [[T1]]
// CHECK: bitcast <2 x i64> [[T2]] to <2 x double>
// CHECK-LE: [[T1:%.+]] = bitcast <2 x double> %{{[0-9]+}} to <2 x i64>
// CHECK-LE: [[T2:%.+]] = or <2 x i64> %{{[0-9]+}}, [[T1]]
// CHECK-LE: bitcast <2 x i64> [[T2]] to <2 x double>
res_vd = vec_or(vd, vbll);
// CHECK: [[T1:%.+]] = bitcast <2 x double> %{{[0-9]+}} to <2 x i64>
// CHECK: [[T2:%.+]] = or <2 x i64> [[T1]], %{{[0-9]+}}
// CHECK: bitcast <2 x i64> [[T2]] to <2 x double>
// CHECK-LE: [[T1:%.+]] = bitcast <2 x double> %{{[0-9]+}} to <2 x i64>
// CHECK-LE: [[T2:%.+]] = or <2 x i64> [[T1]], %{{[0-9]+}}
// CHECK-LE: bitcast <2 x i64> [[T2]] to <2 x double>
res_vf = vec_re(vf);
// CHECK: call <4 x float> @llvm.ppc.vsx.xvresp(<4 x float>
// CHECK-LE: call <4 x float> @llvm.ppc.vsx.xvresp(<4 x float>
res_vd = vec_re(vd);
// CHECK: call <2 x double> @llvm.ppc.vsx.xvredp(<2 x double>
// CHECK-LE: call <2 x double> @llvm.ppc.vsx.xvredp(<2 x double>
res_vf = vec_rint(vf);
// CHECK: call <4 x float> @llvm.nearbyint.v4f32(<4 x float> %{{[0-9]+}})
// CHECK-LE: call <4 x float> @llvm.nearbyint.v4f32(<4 x float> %{{[0-9]+}})
res_vd = vec_rint(vd);
// CHECK: call <2 x double> @llvm.nearbyint.v2f64(<2 x double> %{{[0-9]+}})
// CHECK-LE: call <2 x double> @llvm.nearbyint.v2f64(<2 x double> %{{[0-9]+}})
res_vf = vec_rsqrte(vf);
// CHECK: call <4 x float> @llvm.ppc.vsx.xvrsqrtesp(<4 x float> %{{[0-9]+}})
// CHECK-LE: call <4 x float> @llvm.ppc.vsx.xvrsqrtesp(<4 x float> %{{[0-9]+}})
res_vd = vec_rsqrte(vd);
// CHECK: call <2 x double> @llvm.ppc.vsx.xvrsqrtedp(<2 x double> %{{[0-9]+}})
// CHECK-LE: call <2 x double> @llvm.ppc.vsx.xvrsqrtedp(<2 x double> %{{[0-9]+}})
dummy();
// CHECK: call void @dummy()
// CHECK-LE: call void @dummy()
res_vf = vec_sel(vd, vd, vbll);
// CHECK: xor <2 x i64> %{{[0-9]+}}, <i64 -1, i64 -1>
// CHECK: and <2 x i64> %{{[0-9]+}},
// CHECK: and <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK: or <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]+}} to <2 x double>
// CHECK-LE: xor <2 x i64> %{{[0-9]+}}, <i64 -1, i64 -1>
// CHECK-LE: and <2 x i64> %{{[0-9]+}},
// CHECK-LE: and <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-LE: or <2 x i64>
// CHECK-LE: bitcast <2 x i64> %{{[0-9]+}} to <2 x double>
dummy();
// CHECK: call void @dummy()
// CHECK-LE: call void @dummy()
res_vd = vec_sel(vd, vd, vull);
// CHECK: xor <2 x i64> %{{[0-9]+}}, <i64 -1, i64 -1>
// CHECK: and <2 x i64> %{{[0-9]+}},
// CHECK: and <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK: or <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]+}} to <2 x double>
// CHECK-LE: xor <2 x i64> %{{[0-9]+}}, <i64 -1, i64 -1>
// CHECK-LE: and <2 x i64> %{{[0-9]+}},
// CHECK-LE: and <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-LE: or <2 x i64>
// CHECK-LE: bitcast <2 x i64> %{{[0-9]+}} to <2 x double>
res_vf = vec_sqrt(vf);
// CHECK: call <4 x float> @llvm.sqrt.v4f32(<4 x float> %{{[0-9]+}})
// CHECK-LE: call <4 x float> @llvm.sqrt.v4f32(<4 x float> %{{[0-9]+}})
res_vd = vec_sqrt(vd);
// CHECK: call <2 x double> @llvm.sqrt.v2f64(<2 x double> %{{[0-9]+}})
// CHECK-LE: call <2 x double> @llvm.sqrt.v2f64(<2 x double> %{{[0-9]+}})
res_vd = vec_sub(vd, vd);
// CHECK: fsub <2 x double> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-LE: fsub <2 x double> %{{[0-9]+}}, %{{[0-9]+}}
res_vf = vec_trunc(vf);
// CHECK: call <4 x float> @llvm.trunc.v4f32(<4 x float> %{{[0-9]+}})
// CHECK-LE: call <4 x float> @llvm.trunc.v4f32(<4 x float> %{{[0-9]+}})
res_vd = vec_trunc(vd);
// CHECK: call <2 x double> @llvm.trunc.v2f64(<2 x double> %{{[0-9]+}})
// CHECK-LE: call <2 x double> @llvm.trunc.v2f64(<2 x double> %{{[0-9]+}})
/* vec_vor */
res_vsll = vec_vor(vsll, vsll);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vsll = vec_vor(vbll, vsll);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vsll = vec_vor(vsll, vbll);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vull = vec_vor(vull, vull);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vull = vec_vor(vbll, vull);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vull = vec_vor(vull, vbll);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vbll = vec_vor(vbll, vbll);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
/* vec_xor */
res_vsll = vec_xor(vsll, vsll);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vsll = vec_xor(vbll, vsll);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vsll = vec_xor(vsll, vbll);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vull = vec_xor(vull, vull);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vull = vec_xor(vbll, vull);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vull = vec_xor(vull, vbll);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vbll = vec_xor(vbll, vbll);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
dummy();
// CHECK: call void @dummy()
// CHECK-LE: call void @dummy()
res_vd = vec_xor(vd, vd);
// CHECK: [[X1:%.+]] = xor <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK: bitcast <2 x i64> [[X1]] to <2 x double>
// CHECK-LE: [[X1:%.+]] = xor <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-LE: bitcast <2 x i64> [[X1]] to <2 x double>
dummy();
// CHECK: call void @dummy()
// CHECK-LE: call void @dummy()
res_vd = vec_xor(vd, vbll);
// CHECK: [[X1:%.+]] = xor <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK: bitcast <2 x i64> [[X1]] to <2 x double>
// CHECK-LE: [[X1:%.+]] = xor <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-LE: bitcast <2 x i64> [[X1]] to <2 x double>
dummy();
// CHECK: call void @dummy()
// CHECK-LE: call void @dummy()
res_vd = vec_xor(vbll, vd);
// CHECK: [[X1:%.+]] = xor <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK: bitcast <2 x i64> [[X1]] to <2 x double>
// CHECK-LE: [[X1:%.+]] = xor <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-LE: bitcast <2 x i64> [[X1]] to <2 x double>
/* vec_vxor */
res_vsll = vec_vxor(vsll, vsll);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vsll = vec_vxor(vbll, vsll);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vsll = vec_vxor(vsll, vbll);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vull = vec_vxor(vull, vull);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vull = vec_vxor(vbll, vull);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vull = vec_vxor(vull, vbll);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vbll = vec_vxor(vbll, vbll);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vsll = vec_cts(vd, 0);
// CHECK: fmul <2 x double>
// CHECK: fptosi <2 x double> %{{.*}} to <2 x i64>
// CHECK-LE: fmul <2 x double>
// CHECK-LE: fptosi <2 x double> %{{.*}} to <2 x i64>
res_vsll = vec_cts(vd, 31);
// CHECK: fmul <2 x double>
// CHECK: fptosi <2 x double> %{{.*}} to <2 x i64>
// CHECK-LE: fmul <2 x double>
// CHECK-LE: fptosi <2 x double> %{{.*}} to <2 x i64>
res_vsll = vec_ctu(vd, 0);
// CHECK: fmul <2 x double>
// CHECK: fptoui <2 x double> %{{.*}} to <2 x i64>
// CHECK-LE: fmul <2 x double>
// CHECK-LE: fptoui <2 x double> %{{.*}} to <2 x i64>
res_vsll = vec_ctu(vd, 31);
// CHECK: fmul <2 x double>
// CHECK: fptoui <2 x double> %{{.*}} to <2 x i64>
// CHECK-LE: fmul <2 x double>
// CHECK-LE: fptoui <2 x double> %{{.*}} to <2 x i64>
res_vd = vec_ctf(vsll, 0);
// CHECK: sitofp <2 x i64> %{{.*}} to <2 x double>
// CHECK: fmul <2 x double>
// CHECK-LE: sitofp <2 x i64> %{{.*}} to <2 x double>
// CHECK-LE: fmul <2 x double>
res_vd = vec_ctf(vsll, 31);
// CHECK: sitofp <2 x i64> %{{.*}} to <2 x double>
// CHECK: fmul <2 x double>
// CHECK-LE: sitofp <2 x i64> %{{.*}} to <2 x double>
// CHECK-LE: fmul <2 x double>
res_vd = vec_ctf(vull, 0);
// CHECK: uitofp <2 x i64> %{{.*}} to <2 x double>
// CHECK: fmul <2 x double>
// CHECK-LE: uitofp <2 x i64> %{{.*}} to <2 x double>
// CHECK-LE: fmul <2 x double>
res_vd = vec_ctf(vull, 31);
// CHECK: uitofp <2 x i64> %{{.*}} to <2 x double>
// CHECK: fmul <2 x double>
// CHECK-LE: uitofp <2 x i64> %{{.*}} to <2 x double>
// CHECK-LE: fmul <2 x double>
}
/* 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
}
static int dct_quantize_altivec(MpegEncContext* s,
DCTELEM* data, int n,
int qscale, int* overflow)
{
int lastNonZero;
vector float row0, row1, row2, row3, row4, row5, row6, row7;
vector float alt0, alt1, alt2, alt3, alt4, alt5, alt6, alt7;
const vector float zero = (const vector float)FOUROF(0.);
// used after quantize step
int oldBaseValue = 0;
// Load the data into the row/alt vectors
{
vector signed short data0, data1, data2, data3, data4, data5, data6, data7;
data0 = vec_ld(0, data);
data1 = vec_ld(16, data);
data2 = vec_ld(32, data);
data3 = vec_ld(48, data);
data4 = vec_ld(64, data);
data5 = vec_ld(80, data);
data6 = vec_ld(96, data);
data7 = vec_ld(112, data);
// Transpose the data before we start
TRANSPOSE8(data0, data1, data2, data3, data4, data5, data6, data7);
// load the data into floating point vectors. We load
// the high half of each row into the main row vectors
// and the low half into the alt vectors.
row0 = vec_ctf(vec_unpackh(data0), 0);
alt0 = vec_ctf(vec_unpackl(data0), 0);
row1 = vec_ctf(vec_unpackh(data1), 0);
alt1 = vec_ctf(vec_unpackl(data1), 0);
row2 = vec_ctf(vec_unpackh(data2), 0);
alt2 = vec_ctf(vec_unpackl(data2), 0);
row3 = vec_ctf(vec_unpackh(data3), 0);
alt3 = vec_ctf(vec_unpackl(data3), 0);
row4 = vec_ctf(vec_unpackh(data4), 0);
alt4 = vec_ctf(vec_unpackl(data4), 0);
row5 = vec_ctf(vec_unpackh(data5), 0);
alt5 = vec_ctf(vec_unpackl(data5), 0);
row6 = vec_ctf(vec_unpackh(data6), 0);
alt6 = vec_ctf(vec_unpackl(data6), 0);
row7 = vec_ctf(vec_unpackh(data7), 0);
alt7 = vec_ctf(vec_unpackl(data7), 0);
}
// The following block could exist as a separate an altivec dct
// function. However, if we put it inline, the DCT data can remain
// in the vector local variables, as floats, which we'll use during the
// quantize step...
{
const vector float vec_0_298631336 = (vector float)FOUROF(0.298631336f);
const vector float vec_0_390180644 = (vector float)FOUROF(-0.390180644f);
const vector float vec_0_541196100 = (vector float)FOUROF(0.541196100f);
const vector float vec_0_765366865 = (vector float)FOUROF(0.765366865f);
const vector float vec_0_899976223 = (vector float)FOUROF(-0.899976223f);
const vector float vec_1_175875602 = (vector float)FOUROF(1.175875602f);
const vector float vec_1_501321110 = (vector float)FOUROF(1.501321110f);
const vector float vec_1_847759065 = (vector float)FOUROF(-1.847759065f);
const vector float vec_1_961570560 = (vector float)FOUROF(-1.961570560f);
const vector float vec_2_053119869 = (vector float)FOUROF(2.053119869f);
const vector float vec_2_562915447 = (vector float)FOUROF(-2.562915447f);
const vector float vec_3_072711026 = (vector float)FOUROF(3.072711026f);
int whichPass, whichHalf;
for(whichPass = 1; whichPass<=2; whichPass++) {
for(whichHalf = 1; whichHalf<=2; whichHalf++) {
vector float tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
vector float tmp10, tmp11, tmp12, tmp13;
vector float z1, z2, z3, z4, z5;
tmp0 = vec_add(row0, row7); // tmp0 = dataptr[0] + dataptr[7];
tmp7 = vec_sub(row0, row7); // tmp7 = dataptr[0] - dataptr[7];
tmp3 = vec_add(row3, row4); // tmp3 = dataptr[3] + dataptr[4];
tmp4 = vec_sub(row3, row4); // tmp4 = dataptr[3] - dataptr[4];
tmp1 = vec_add(row1, row6); // tmp1 = dataptr[1] + dataptr[6];
tmp6 = vec_sub(row1, row6); // tmp6 = dataptr[1] - dataptr[6];
tmp2 = vec_add(row2, row5); // tmp2 = dataptr[2] + dataptr[5];
tmp5 = vec_sub(row2, row5); // tmp5 = dataptr[2] - dataptr[5];
tmp10 = vec_add(tmp0, tmp3); // tmp10 = tmp0 + tmp3;
tmp13 = vec_sub(tmp0, tmp3); // tmp13 = tmp0 - tmp3;
tmp11 = vec_add(tmp1, tmp2); // tmp11 = tmp1 + tmp2;
tmp12 = vec_sub(tmp1, tmp2); // tmp12 = tmp1 - tmp2;
// dataptr[0] = (DCTELEM) ((tmp10 + tmp11) << PASS1_BITS);
row0 = vec_add(tmp10, tmp11);
// dataptr[4] = (DCTELEM) ((tmp10 - tmp11) << PASS1_BITS);
row4 = vec_sub(tmp10, tmp11);
// z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100);
z1 = vec_madd(vec_add(tmp12, tmp13), vec_0_541196100, (vector float)zero);
// dataptr[2] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp13, FIX_0_765366865),
// CONST_BITS-PASS1_BITS);
row2 = vec_madd(tmp13, vec_0_765366865, z1);
// dataptr[6] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp12, - FIX_1_847759065),
// CONST_BITS-PASS1_BITS);
row6 = vec_madd(tmp12, vec_1_847759065, z1);
z1 = vec_add(tmp4, tmp7); // z1 = tmp4 + tmp7;
z2 = vec_add(tmp5, tmp6); // z2 = tmp5 + tmp6;
z3 = vec_add(tmp4, tmp6); // z3 = tmp4 + tmp6;
z4 = vec_add(tmp5, tmp7); // z4 = tmp5 + tmp7;
// z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
z5 = vec_madd(vec_add(z3, z4), vec_1_175875602, (vector float)zero);
// z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
z3 = vec_madd(z3, vec_1_961570560, z5);
// z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
z4 = vec_madd(z4, vec_0_390180644, z5);
// The following adds are rolled into the multiplies above
// z3 = vec_add(z3, z5); // z3 += z5;
// z4 = vec_add(z4, z5); // z4 += z5;
// z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
// Wow! It's actually more efficient to roll this multiply
// into the adds below, even thought the multiply gets done twice!
// z2 = vec_madd(z2, vec_2_562915447, (vector float)zero);
// z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
// Same with this one...
// z1 = vec_madd(z1, vec_0_899976223, (vector float)zero);
// tmp4 = MULTIPLY(tmp4, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
// dataptr[7] = (DCTELEM) DESCALE(tmp4 + z1 + z3, CONST_BITS-PASS1_BITS);
row7 = vec_madd(tmp4, vec_0_298631336, vec_madd(z1, vec_0_899976223, z3));
// tmp5 = MULTIPLY(tmp5, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
// dataptr[5] = (DCTELEM) DESCALE(tmp5 + z2 + z4, CONST_BITS-PASS1_BITS);
row5 = vec_madd(tmp5, vec_2_053119869, vec_madd(z2, vec_2_562915447, z4));
// tmp6 = MULTIPLY(tmp6, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
// dataptr[3] = (DCTELEM) DESCALE(tmp6 + z2 + z3, CONST_BITS-PASS1_BITS);
row3 = vec_madd(tmp6, vec_3_072711026, vec_madd(z2, vec_2_562915447, z3));
// tmp7 = MULTIPLY(tmp7, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
// dataptr[1] = (DCTELEM) DESCALE(tmp7 + z1 + z4, CONST_BITS-PASS1_BITS);
row1 = vec_madd(z1, vec_0_899976223, vec_madd(tmp7, vec_1_501321110, z4));
// Swap the row values with the alts. If this is the first half,
// this sets up the low values to be acted on in the second half.
// If this is the second half, it puts the high values back in
// the row values where they are expected to be when we're done.
SWAP(row0, alt0);
SWAP(row1, alt1);
SWAP(row2, alt2);
SWAP(row3, alt3);
SWAP(row4, alt4);
SWAP(row5, alt5);
SWAP(row6, alt6);
SWAP(row7, alt7);
}
if (whichPass == 1) {
// transpose the data for the second pass
// First, block transpose the upper right with lower left.
SWAP(row4, alt0);
SWAP(row5, alt1);
SWAP(row6, alt2);
SWAP(row7, alt3);
// Now, transpose each block of four
TRANSPOSE4(row0, row1, row2, row3);
TRANSPOSE4(row4, row5, row6, row7);
TRANSPOSE4(alt0, alt1, alt2, alt3);
TRANSPOSE4(alt4, alt5, alt6, alt7);
}
}
}
// perform the quantize step, using the floating point data
// still in the row/alt registers
{
const int* biasAddr;
const vector signed int* qmat;
vector float bias, negBias;
if (s->mb_intra) {
vector signed int baseVector;
// We must cache element 0 in the intra case
// (it needs special handling).
baseVector = vec_cts(vec_splat(row0, 0), 0);
vec_ste(baseVector, 0, &oldBaseValue);
qmat = (vector signed int*)s->q_intra_matrix[qscale];
biasAddr = &(s->intra_quant_bias);
} else {
qmat = (vector signed int*)s->q_inter_matrix[qscale];
biasAddr = &(s->inter_quant_bias);
}
// Load the bias vector (We add 0.5 to the bias so that we're
// rounding when we convert to int, instead of flooring.)
{
vector signed int biasInt;
const vector float negOneFloat = (vector float)FOUROF(-1.0f);
LOAD4(biasInt, biasAddr);
bias = vec_ctf(biasInt, QUANT_BIAS_SHIFT);
negBias = vec_madd(bias, negOneFloat, zero);
}
{
vector float q0, q1, q2, q3, q4, q5, q6, q7;
q0 = vec_ctf(qmat[0], QMAT_SHIFT);
q1 = vec_ctf(qmat[2], QMAT_SHIFT);
q2 = vec_ctf(qmat[4], QMAT_SHIFT);
q3 = vec_ctf(qmat[6], QMAT_SHIFT);
q4 = vec_ctf(qmat[8], QMAT_SHIFT);
q5 = vec_ctf(qmat[10], QMAT_SHIFT);
q6 = vec_ctf(qmat[12], QMAT_SHIFT);
q7 = vec_ctf(qmat[14], QMAT_SHIFT);
row0 = vec_sel(vec_madd(row0, q0, negBias), vec_madd(row0, q0, bias),
vec_cmpgt(row0, zero));
row1 = vec_sel(vec_madd(row1, q1, negBias), vec_madd(row1, q1, bias),
vec_cmpgt(row1, zero));
row2 = vec_sel(vec_madd(row2, q2, negBias), vec_madd(row2, q2, bias),
vec_cmpgt(row2, zero));
row3 = vec_sel(vec_madd(row3, q3, negBias), vec_madd(row3, q3, bias),
vec_cmpgt(row3, zero));
row4 = vec_sel(vec_madd(row4, q4, negBias), vec_madd(row4, q4, bias),
vec_cmpgt(row4, zero));
row5 = vec_sel(vec_madd(row5, q5, negBias), vec_madd(row5, q5, bias),
vec_cmpgt(row5, zero));
row6 = vec_sel(vec_madd(row6, q6, negBias), vec_madd(row6, q6, bias),
vec_cmpgt(row6, zero));
row7 = vec_sel(vec_madd(row7, q7, negBias), vec_madd(row7, q7, bias),
vec_cmpgt(row7, zero));
q0 = vec_ctf(qmat[1], QMAT_SHIFT);
q1 = vec_ctf(qmat[3], QMAT_SHIFT);
q2 = vec_ctf(qmat[5], QMAT_SHIFT);
q3 = vec_ctf(qmat[7], QMAT_SHIFT);
q4 = vec_ctf(qmat[9], QMAT_SHIFT);
q5 = vec_ctf(qmat[11], QMAT_SHIFT);
q6 = vec_ctf(qmat[13], QMAT_SHIFT);
q7 = vec_ctf(qmat[15], QMAT_SHIFT);
alt0 = vec_sel(vec_madd(alt0, q0, negBias), vec_madd(alt0, q0, bias),
vec_cmpgt(alt0, zero));
alt1 = vec_sel(vec_madd(alt1, q1, negBias), vec_madd(alt1, q1, bias),
vec_cmpgt(alt1, zero));
alt2 = vec_sel(vec_madd(alt2, q2, negBias), vec_madd(alt2, q2, bias),
vec_cmpgt(alt2, zero));
alt3 = vec_sel(vec_madd(alt3, q3, negBias), vec_madd(alt3, q3, bias),
vec_cmpgt(alt3, zero));
alt4 = vec_sel(vec_madd(alt4, q4, negBias), vec_madd(alt4, q4, bias),
vec_cmpgt(alt4, zero));
alt5 = vec_sel(vec_madd(alt5, q5, negBias), vec_madd(alt5, q5, bias),
vec_cmpgt(alt5, zero));
alt6 = vec_sel(vec_madd(alt6, q6, negBias), vec_madd(alt6, q6, bias),
vec_cmpgt(alt6, zero));
alt7 = vec_sel(vec_madd(alt7, q7, negBias), vec_madd(alt7, q7, bias),
vec_cmpgt(alt7, zero));
}
}
// Store the data back into the original block
{
vector signed short data0, data1, data2, data3, data4, data5, data6, data7;
data0 = vec_pack(vec_cts(row0, 0), vec_cts(alt0, 0));
data1 = vec_pack(vec_cts(row1, 0), vec_cts(alt1, 0));
data2 = vec_pack(vec_cts(row2, 0), vec_cts(alt2, 0));
data3 = vec_pack(vec_cts(row3, 0), vec_cts(alt3, 0));
data4 = vec_pack(vec_cts(row4, 0), vec_cts(alt4, 0));
data5 = vec_pack(vec_cts(row5, 0), vec_cts(alt5, 0));
data6 = vec_pack(vec_cts(row6, 0), vec_cts(alt6, 0));
data7 = vec_pack(vec_cts(row7, 0), vec_cts(alt7, 0));
{
// Clamp for overflow
vector signed int max_q_int, min_q_int;
vector signed short max_q, min_q;
LOAD4(max_q_int, &(s->max_qcoeff));
LOAD4(min_q_int, &(s->min_qcoeff));
max_q = vec_pack(max_q_int, max_q_int);
min_q = vec_pack(min_q_int, min_q_int);
data0 = vec_max(vec_min(data0, max_q), min_q);
data1 = vec_max(vec_min(data1, max_q), min_q);
data2 = vec_max(vec_min(data2, max_q), min_q);
data4 = vec_max(vec_min(data4, max_q), min_q);
data5 = vec_max(vec_min(data5, max_q), min_q);
data6 = vec_max(vec_min(data6, max_q), min_q);
data7 = vec_max(vec_min(data7, max_q), min_q);
}
{
vector bool char zero_01, zero_23, zero_45, zero_67;
vector signed char scanIndexes_01, scanIndexes_23, scanIndexes_45, scanIndexes_67;
vector signed char negOne = vec_splat_s8(-1);
vector signed char* scanPtr =
(vector signed char*)(s->intra_scantable.inverse);
signed char lastNonZeroChar;
// Determine the largest non-zero index.
zero_01 = vec_pack(vec_cmpeq(data0, (vector signed short)zero),
vec_cmpeq(data1, (vector signed short)zero));
zero_23 = vec_pack(vec_cmpeq(data2, (vector signed short)zero),
vec_cmpeq(data3, (vector signed short)zero));
zero_45 = vec_pack(vec_cmpeq(data4, (vector signed short)zero),
vec_cmpeq(data5, (vector signed short)zero));
zero_67 = vec_pack(vec_cmpeq(data6, (vector signed short)zero),
vec_cmpeq(data7, (vector signed short)zero));
// 64 biggest values
scanIndexes_01 = vec_sel(scanPtr[0], negOne, zero_01);
scanIndexes_23 = vec_sel(scanPtr[1], negOne, zero_23);
scanIndexes_45 = vec_sel(scanPtr[2], negOne, zero_45);
scanIndexes_67 = vec_sel(scanPtr[3], negOne, zero_67);
// 32 largest values
scanIndexes_01 = vec_max(scanIndexes_01, scanIndexes_23);
scanIndexes_45 = vec_max(scanIndexes_45, scanIndexes_67);
// 16 largest values
scanIndexes_01 = vec_max(scanIndexes_01, scanIndexes_45);
// 8 largest values
scanIndexes_01 = vec_max(vec_mergeh(scanIndexes_01, negOne),
vec_mergel(scanIndexes_01, negOne));
// 4 largest values
scanIndexes_01 = vec_max(vec_mergeh(scanIndexes_01, negOne),
vec_mergel(scanIndexes_01, negOne));
// 2 largest values
scanIndexes_01 = vec_max(vec_mergeh(scanIndexes_01, negOne),
vec_mergel(scanIndexes_01, negOne));
// largest value
scanIndexes_01 = vec_max(vec_mergeh(scanIndexes_01, negOne),
vec_mergel(scanIndexes_01, negOne));
scanIndexes_01 = vec_splat(scanIndexes_01, 0);
vec_ste(scanIndexes_01, 0, &lastNonZeroChar);
lastNonZero = lastNonZeroChar;
// While the data is still in vectors we check for the transpose IDCT permute
// and handle it using the vector unit if we can. This is the permute used
// by the altivec idct, so it is common when using the altivec dct.
if ((lastNonZero > 0) && (s->dsp.idct_permutation_type == FF_TRANSPOSE_IDCT_PERM)) {
TRANSPOSE8(data0, data1, data2, data3, data4, data5, data6, data7);
}
vec_st(data0, 0, data);
vec_st(data1, 16, data);
vec_st(data2, 32, data);
vec_st(data3, 48, data);
vec_st(data4, 64, data);
vec_st(data5, 80, data);
vec_st(data6, 96, data);
vec_st(data7, 112, data);
}
}
// special handling of block[0]
if (s->mb_intra) {
if (!s->h263_aic) {
if (n < 4)
oldBaseValue /= s->y_dc_scale;
else
oldBaseValue /= s->c_dc_scale;
}
// Divide by 8, rounding the result
data[0] = (oldBaseValue + 4) >> 3;
}
// We handled the transpose permutation above and we don't
// need to permute the "no" permutation case.
if ((lastNonZero > 0) &&
(s->dsp.idct_permutation_type != FF_TRANSPOSE_IDCT_PERM) &&
(s->dsp.idct_permutation_type != FF_NO_IDCT_PERM)) {
ff_block_permute(data, s->dsp.idct_permutation,
s->intra_scantable.scantable, lastNonZero);
}
return lastNonZero;
}
/* 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);
}
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;
}
}
/* 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;
}
}
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 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);
}
