/* An accurate vector division routine using the reciprocal estimate and
* two Newton-Raphson iterations
*/
static inline vector float vec_div(vector float A, vector float B)
{
vector float y0;
vector float y1;
vector float y2;
vector float Q;
vector float R;
vector float one = (vector float) (1.0f);
vector float zero = (vector float) (-0.0f);
vector float mone = (vector float) (-1.0f);
y0 = vec_re(B); // approximate 1/B
// y1 = y0*(-(y0*B - 1.0))+y0 i.e. y0+y0*(1.0 - y0*B)
y1 = vec_madd(y0,vec_nmsub(y0, B, one),y0);
// REPEAT the Newton-Raphson to get the required 24 bits
y2 = vec_madd(y1, vec_nmsub(y1, B, one),y1);
// y2 = y1*(-(y1*B - 1.0f))+y1 i.e. y1+y1*(1.0f - y1*B)
// y2 is now the correctly rounded reciprocal, and the manual considers this
// OK for use in computing the remainder: Q = A*y2, R = A - B*Q
Q = vec_madd(A,y2,zero); // -0.0 IEEE
R = vec_nmsub(B,Q,A); // -(B*Q-A) == (A-B*Q)
// final rouding adjustment
return(vec_madd(R, y2, Q));
}
static void vector_fmul_window_altivec(float *dst, const float *src0, const float *src1, const float *win, int len)
{
vector float zero, t0, t1, s0, s1, wi, wj;
const vector unsigned char reverse = vcprm(3,2,1,0);
int i,j;
dst += len;
win += len;
src0+= len;
zero = (vector float)vec_splat_u32(0);
for(i=-len*4, j=len*4-16; i<0; i+=16, j-=16) {
s0 = vec_ld(i, src0);
s1 = vec_ld(j, src1);
wi = vec_ld(i, win);
wj = vec_ld(j, win);
s1 = vec_perm(s1, s1, reverse);
wj = vec_perm(wj, wj, reverse);
t0 = vec_madd(s0, wj, zero);
t0 = vec_nmsub(s1, wi, t0);
t1 = vec_madd(s0, wi, zero);
t1 = vec_madd(s1, wj, t1);
t1 = vec_perm(t1, t1, reverse);
vec_st(t0, i, dst);
vec_st(t1, j, dst);
}
}
vector float
f(vector float a, vector float b, vector float c)
{
vector float q = vec_expte(a);
vector float r = vec_vsubfp(c, q);
vector float s = vec_re(b);
vector float t = vec_nmsub(s, c, r);
return t;
}
void
nb_kernel310_ppc_altivec (int * p_nri,
int iinr[],
int jindex[],
int jjnr[],
int shift[],
float shiftvec[],
float fshift[],
int gid[],
float pos[],
float faction[],
float charge[],
float * p_facel,
float * p_krf,
float * p_crf,
float Vc[],
int type[],
int * p_ntype,
float vdwparam[],
float Vvdw[],
float * p_tabscale,
float VFtab[],
float invsqrta[],
float dvda[],
float * p_gbtabscale,
float GBtab[],
int * p_nthreads,
int * count,
void * mtx,
int * outeriter,
int * inneriter,
float * work)
{
vector float ix,iy,iz,shvec;
vector float vfacel,tsc,fs,fs2,nul;
vector float dx,dy,dz;
vector float Vvdwtot,vctot,qq,iq,c6,c12,VVc,FFc;
vector float fix,fiy,fiz;
vector float tmp1,tmp2,tmp3,tmp4;
vector float rinv,r,rinvsq,rsq,rinvsix,Vvdw6,Vvdw12;
int n,k,ii,is3,ii3,ntiA,nj0,nj1;
int jnra,jnrb,jnrc,jnrd;
int j3a,j3b,j3c,j3d;
int nri, ntype, nouter, ninner;
int tja,tjb,tjc,tjd;
#ifdef GMX_THREADS
int nn0, nn1;
#endif
nouter = 0;
ninner = 0;
nri = *p_nri;
ntype = *p_ntype;
nul=vec_zero();
vfacel=load_float_and_splat(p_facel);
tsc=load_float_and_splat(p_tabscale);
#ifdef GMX_THREADS
nthreads = *p_nthreads;
do {
gmx_thread_mutex_lock((gmx_thread_mutex_t *)mtx);
nn0 = *count;
nn1 = nn0+(nri-nn0)/(2*nthreads)+3;
*count = nn1;
gmx_thread_mutex_unlock((gmx_thread_mutex_t *)mtx);
if(nn1>nri) nn1=nri;
for(n=nn0; (n<nn1); n++) {
#if 0
} /* maintain correct indentation even with conditional left braces */
#endif
#else /* without gmx_threads */
for(n=0;n<nri;n++) {
#endif
is3 = 3*shift[n];
shvec = load_xyz(shiftvec+is3);
ii = iinr[n];
ii3 = 3*ii;
ix = load_xyz(pos+ii3);
Vvdwtot = nul;
vctot = nul;
fix = nul;
fiy = nul;
fiz = nul;
ix = vec_add(ix,shvec);
nj0 = jindex[n];
nj1 = jindex[n+1];
splat_xyz_to_vectors(ix,&ix,&iy,&iz);
ntiA = 2*ntype*type[ii];
iq = vec_madd(load_float_and_splat(charge+ii),vfacel,nul);
for(k=nj0; k<(nj1-3); k+=4) {
jnra = jjnr[k];
jnrb = jjnr[k+1];
jnrc = jjnr[k+2];
jnrd = jjnr[k+3];
j3a = 3*jnra;
j3b = 3*jnrb;
j3c = 3*jnrc;
j3d = 3*jnrd;
transpose_4_to_3(load_xyz(pos+j3a),
load_xyz(pos+j3b),
load_xyz(pos+j3c),
load_xyz(pos+j3d),&dx,&dy,&dz);
dx = vec_sub(ix,dx);
dy = vec_sub(iy,dy);
dz = vec_sub(iz,dz);
rsq = vec_madd(dx,dx,nul);
rsq = vec_madd(dy,dy,rsq);
rsq = vec_madd(dz,dz,rsq);
rinv = do_invsqrt(rsq);
rinvsq = vec_madd(rinv,rinv,nul);
r = vec_madd(rinv,rsq,nul);
rinvsix = vec_madd(rinvsq,rinvsq,nul);
rinvsix = vec_madd(rinvsix,rinvsq,nul);
tja = ntiA+2*type[jnra];
tjb = ntiA+2*type[jnrb];
tjc = ntiA+2*type[jnrc];
tjd = ntiA+2*type[jnrd];
qq = vec_madd(load_4_float(charge+jnra,charge+jnrb,
charge+jnrc,charge+jnrd),iq,nul);
load_4_pair(vdwparam+tja,vdwparam+tjb,vdwparam+tjc,vdwparam+tjd,&c6,&c12);
do_4_ctable_coul(VFtab,vec_madd(r,tsc,nul),&VVc,&FFc);
fs2 = vec_madd(qq,FFc,nul); /* fijC */
vctot = vec_madd(qq,VVc,vctot);
Vvdw6 = vec_madd(c6,rinvsix,nul);
Vvdw12 = vec_madd(c12,vec_madd(rinvsix,rinvsix,nul),
nul);
fs = vec_madd(vec_twelve(),Vvdw12,nul);
fs = vec_nmsub(vec_six(),Vvdw6,fs);
fs = vec_madd(fs,rinv,nul);
Vvdwtot = vec_add(Vvdwtot,Vvdw12);
fs = vec_nmsub(fs2,tsc,fs);
fs = vec_madd(fs,rinv,nul);
Vvdwtot = vec_sub(Vvdwtot,Vvdw6);
fix = vec_madd(fs,dx,fix); /* +=fx */
fiy = vec_madd(fs,dy,fiy); /* +=fy */
fiz = vec_madd(fs,dz,fiz); /* +=fz */
dx = vec_nmsub(dx,fs,nul); /* -fx */
dy = vec_nmsub(dy,fs,nul); /* -fy */
dz = vec_nmsub(dz,fs,nul); /* -fz */
transpose_3_to_4(dx,dy,dz,&tmp1,&tmp2,&tmp3,&tmp4);
add_xyz_to_mem(faction+j3a,tmp1);
add_xyz_to_mem(faction+j3b,tmp2);
add_xyz_to_mem(faction+j3c,tmp3);
add_xyz_to_mem(faction+j3d,tmp4);
}
if(k<(nj1-1)) {
jnra = jjnr[k];
jnrb = jjnr[k+1];
j3a = 3*jnra;
j3b = 3*jnrb;
transpose_2_to_3(load_xyz(pos+j3a),
load_xyz(pos+j3b),&dx,&dy,&dz);
dx = vec_sub(ix,dx);
dy = vec_sub(iy,dy);
dz = vec_sub(iz,dz);
rsq = vec_madd(dx,dx,nul);
rsq = vec_madd(dy,dy,rsq);
rsq = vec_madd(dz,dz,rsq);
zero_highest_2_elements_in_vector(&rsq);
rinv = do_invsqrt(rsq);
zero_highest_2_elements_in_vector(&rinv);
rinvsq = vec_madd(rinv,rinv,nul);
r = vec_madd(rinv,rsq,nul);
rinvsix = vec_madd(rinvsq,rinvsq,nul);
rinvsix = vec_madd(rinvsix,rinvsq,nul);
tja = ntiA+2*type[jnra];
tjb = ntiA+2*type[jnrb];
qq = vec_madd(load_2_float(charge+jnra,charge+jnrb),iq,nul);
load_2_pair(vdwparam+tja,vdwparam+tjb,&c6,&c12);
do_2_ctable_coul(VFtab,vec_madd(r,tsc,nul),&VVc,&FFc);
fs2 = vec_madd(qq,FFc,nul); /* fijC */
vctot = vec_madd(qq,VVc,vctot);
Vvdw6 = vec_madd(c6,rinvsix,nul);
Vvdw12 = vec_madd(c12,vec_madd(rinvsix,rinvsix,nul),
nul);
fs = vec_madd(vec_twelve(),Vvdw12,nul);
fs = vec_nmsub(vec_six(),Vvdw6,fs);
Vvdwtot = vec_add(Vvdwtot,Vvdw12);
fs = vec_madd(fs,rinv,nul);
fs = vec_nmsub(fs2,tsc,fs);
fs = vec_madd(fs,rinv,nul);
Vvdwtot = vec_sub(Vvdwtot,Vvdw6);
fix = vec_madd(fs,dx,fix); /* +=fx */
fiy = vec_madd(fs,dy,fiy); /* +=fy */
fiz = vec_madd(fs,dz,fiz); /* +=fz */
dx = vec_nmsub(dx,fs,nul); /* -fx */
dy = vec_nmsub(dy,fs,nul); /* -fy */
dz = vec_nmsub(dz,fs,nul); /* -fz */
transpose_3_to_2(dx,dy,dz,&tmp1,&tmp2);
add_xyz_to_mem(faction+j3a,tmp1);
add_xyz_to_mem(faction+j3b,tmp2);
k += 2;
}
if((nj1-nj0) & 0x1) {
jnra = jjnr[k];
j3a = 3*jnra;
transpose_1_to_3(load_xyz(pos+j3a),&dx,&dy,&dz);
dx = vec_sub(ix,dx);
dy = vec_sub(iy,dy);
dz = vec_sub(iz,dz);
rsq = vec_madd(dx,dx,nul);
rsq = vec_madd(dy,dy,rsq);
rsq = vec_madd(dz,dz,rsq);
zero_highest_3_elements_in_vector(&rsq);
rinv = do_invsqrt(rsq);
zero_highest_3_elements_in_vector(&rinv);
rinvsq = vec_madd(rinv,rinv,nul);
r = vec_madd(rinv,rsq,nul);
rinvsix = vec_madd(rinvsq,rinvsq,nul);
rinvsix = vec_madd(rinvsix,rinvsq,nul);
tja = ntiA+2*type[jnra];
qq = vec_madd(load_1_float(charge+jnra),iq,nul);
load_1_pair(vdwparam+tja,&c6,&c12);
do_1_ctable_coul(VFtab,vec_madd(r,tsc,nul),&VVc,&FFc);
fs2 = vec_madd(qq,FFc,nul); /* fijC */
vctot = vec_madd(qq,VVc,vctot);
Vvdw6 = vec_madd(c6,rinvsix,nul);
Vvdw12 = vec_madd(c12,vec_madd(rinvsix,rinvsix,nul),
nul);
fs = vec_madd(vec_twelve(),Vvdw12,nul);
fs = vec_nmsub(vec_six(),Vvdw6,fs);
fs = vec_madd(fs,rinv,nul);
Vvdwtot = vec_add(Vvdwtot,Vvdw12);
fs = vec_nmsub(fs2,tsc,fs);
fs = vec_madd(fs,rinv,nul);
Vvdwtot = vec_sub(Vvdwtot,Vvdw6);
fix = vec_madd(fs,dx,fix); /* +=fx */
fiy = vec_madd(fs,dy,fiy); /* +=fy */
fiz = vec_madd(fs,dz,fiz); /* +=fz */
dx = vec_nmsub(dx,fs,nul); /* -fx */
dy = vec_nmsub(dy,fs,nul); /* -fy */
dz = vec_nmsub(dz,fs,nul); /* -fz */
transpose_3_to_1(dx,dy,dz,&tmp1);
add_xyz_to_mem(faction+j3a,tmp1);
}
/* update outer data */
transpose_3_to_4(fix,fiy,fiz,&tmp1,&tmp2,&tmp3,&tmp4);
tmp1 = vec_add(tmp1,tmp3);
tmp2 = vec_add(tmp2,tmp4);
tmp1 = vec_add(tmp1,tmp2);
add_xyz_to_mem(faction+ii3,tmp1);
add_xyz_to_mem(fshift+is3,tmp1);
add_vector_to_float(Vc+gid[n],vctot);
add_vector_to_float(Vvdw+gid[n],Vvdwtot);
ninner += nj1 - nj0;
}
#ifdef GMX_THREADS
nouter += nn1 - nn0;
} while (nn1<nri);
#else
nouter = nri;
#endif
*outeriter = nouter;
*inneriter = ninner;
}
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>
}
void
nb_kernel010nf_ppc_altivec(int * p_nri,
int iinr[],
int jindex[],
int jjnr[],
int shift[],
float shiftvec[],
float fshift[],
int gid[],
float pos[],
float faction[],
float charge[],
float * p_facel,
float * p_krf,
float * p_crf,
float Vc[],
int type[],
int * p_ntype,
float vdwparam[],
float Vvdw[],
float * p_tabscale,
float VFtab[],
float invsqrta[],
float dvda[],
float * p_gbtabscale,
float GBtab[],
int * p_nthreads,
int * count,
void * mtx,
int * outeriter,
int * inneriter,
float * work)
{
vector float ix,iy,iz,shvec;
vector float nul;
vector float dx,dy,dz;
vector float Vvdwtot,c6,c12;
vector float rinvsq,rsq,rinvsix;
int n,k,ii,is3,ii3,nj0,nj1;
int jnra,jnrb,jnrc,jnrd;
int j3a,j3b,j3c,j3d;
int nri, ntype, nouter, ninner;
int ntiA,tja,tjb,tjc,tjd;
#ifdef GMX_THREAD_SHM_FDECOMP
int nn0, nn1;
#endif
nouter = 0;
ninner = 0;
nri = *p_nri;
ntype = *p_ntype;
nul=vec_zero();
#ifdef GMX_THREAD_SHM_FDECOMP
nthreads = *p_nthreads;
do {
tMPI_Thread_mutex_lock((tMPI_Thread_mutex_t *)mtx);
nn0 = *count;
nn1 = nn0+(nri-nn0)/(2*nthreads)+3;
*count = nn1;
tMPI_Thread_mutex_unlock((tMPI_Thread_mutex_t *)mtx);
if(nn1>nri) nn1=nri;
for(n=nn0; (n<nn1); n++) {
#if 0
} /* maintain correct indentation even with conditional left braces */
#endif
#else /* without tMPI_Threads */
for(n=0;n<nri;n++) {
#endif
is3 = 3*shift[n];
shvec = load_xyz(shiftvec+is3);
ii = iinr[n];
ii3 = 3*ii;
ix = load_xyz(pos+ii3);
Vvdwtot = nul;
ix = vec_add(ix,shvec);
nj0 = jindex[n];
nj1 = jindex[n+1];
splat_xyz_to_vectors(ix,&ix,&iy,&iz);
ntiA = 2*ntype*type[ii];
for(k=nj0; k<(nj1-3); k+=4) {
jnra = jjnr[k];
jnrb = jjnr[k+1];
jnrc = jjnr[k+2];
jnrd = jjnr[k+3];
j3a = 3*jnra;
j3b = 3*jnrb;
j3c = 3*jnrc;
j3d = 3*jnrd;
transpose_4_to_3(load_xyz(pos+j3a),
load_xyz(pos+j3b),
load_xyz(pos+j3c),
load_xyz(pos+j3d),&dx,&dy,&dz);
dx = vec_sub(ix,dx);
dy = vec_sub(iy,dy);
dz = vec_sub(iz,dz);
rsq = vec_madd(dx,dx,nul);
rsq = vec_madd(dy,dy,rsq);
rsq = vec_madd(dz,dz,rsq);
rinvsq = do_recip(rsq);
rinvsix = vec_madd(rinvsq,rinvsq,nul);
rinvsix = vec_madd(rinvsix,rinvsq,nul);
tja = ntiA+2*type[jnra];
tjb = ntiA+2*type[jnrb];
tjc = ntiA+2*type[jnrc];
tjd = ntiA+2*type[jnrd];
load_4_pair(vdwparam+tja,vdwparam+tjb,vdwparam+tjc,vdwparam+tjd,&c6,&c12);
Vvdwtot = vec_nmsub(c6,rinvsix,Vvdwtot);
Vvdwtot = vec_madd(c12,
vec_madd(rinvsix,rinvsix,nul),
Vvdwtot);
}
if(k<(nj1-1)) {
jnra = jjnr[k];
jnrb = jjnr[k+1];
j3a = 3*jnra;
j3b = 3*jnrb;
transpose_2_to_3(load_xyz(pos+j3a),
load_xyz(pos+j3b),&dx,&dy,&dz);
dx = vec_sub(ix,dx);
dy = vec_sub(iy,dy);
dz = vec_sub(iz,dz);
rsq = vec_madd(dx,dx,nul);
rsq = vec_madd(dy,dy,rsq);
rsq = vec_madd(dz,dz,rsq);
rinvsq = do_recip(rsq);
zero_highest_2_elements_in_vector(&rinvsq);
rinvsix = vec_madd(rinvsq,rinvsq,nul);
rinvsix = vec_madd(rinvsix,rinvsq,nul);
tja = ntiA+2*type[jnra];
tjb = ntiA+2*type[jnrb];
load_2_pair(vdwparam+tja,vdwparam+tjb,&c6,&c12);
Vvdwtot = vec_nmsub(c6,rinvsix,Vvdwtot);
Vvdwtot = vec_madd(c12,
vec_madd(rinvsix,rinvsix,nul),
Vvdwtot);
k += 2;
}
if((nj1-nj0) & 0x1) {
jnra = jjnr[k];
j3a = 3*jnra;
transpose_1_to_3(load_xyz(pos+j3a),&dx,&dy,&dz);
dx = vec_sub(ix,dx);
dy = vec_sub(iy,dy);
dz = vec_sub(iz,dz);
rsq = vec_madd(dx,dx,nul);
rsq = vec_madd(dy,dy,rsq);
rsq = vec_madd(dz,dz,rsq);
rinvsq = do_recip(rsq);
zero_highest_3_elements_in_vector(&rinvsq);
rinvsix = vec_madd(rinvsq,rinvsq,nul);
rinvsix = vec_madd(rinvsix,rinvsq,nul);
tja = ntiA+2*type[jnra];
load_1_pair(vdwparam+tja,&c6,&c12);
Vvdwtot = vec_nmsub(c6,rinvsix,Vvdwtot);
Vvdwtot = vec_madd(c12,
vec_madd(rinvsix,rinvsix,nul),
Vvdwtot);
}
/* update outer data */
add_vector_to_float(Vvdw+gid[n],Vvdwtot);
ninner += nj1 - nj0;
}
#ifdef GMX_THREAD_SHM_FDECOMP
nouter += nn1 - nn0;
} while (nn1<nri);
#else
nouter = nri;
#endif
*outeriter = nouter;
*inneriter = ninner;
}
void
nb_kernel133_ppc_altivec (int * p_nri,
int iinr[],
int jindex[],
int jjnr[],
int shift[],
float shiftvec[],
float fshift[],
int gid[],
float pos[],
float faction[],
float charge[],
float * p_facel,
float * p_krf,
float * p_crf,
float Vc[],
int type[],
int * p_ntype,
float vdwparam[],
float Vvdw[],
float * p_tabscale,
float VFtab[],
float invsqrta[],
float dvda[],
float * p_gbtabscale,
float GBtab[],
int * p_nthreads,
int * count,
void * mtx,
int * outeriter,
int * inneriter,
float * work)
{
vector float iOx,iOy,iOz,iH1x,iH1y,iH1z,iH2x,iH2y,iH2z,iMx,iMy,iMz;
vector float dOx,dOy,dOz,dH1x,dH1y,dH1z,dH2x,dH2y,dH2z,dMx,dMy,dMz;
vector float vfacel,vcoulM,vcoulH1,vcoulH2,nul;
vector float Vvdwtot,c6,c12,VVd,VVr,FFd,FFr,tsc,r;
vector float fsO,fsH1,fsH2,fsM;
vector float vctot,qqM,qqH,iqM,iqH,jq;
vector float fiOx,fiOy,fiOz,fiH1x,fiH1y,fiH1z,fiH2x,fiH2y,fiH2z;
vector float fiMx,fiMy,fiMz;
vector float tmp1,tmp2,tmp3,tmp4;
vector float rinvH1,rinvH2,rinvM,rinvO;
vector float rinvsqH1,rinvsqH2,rinvsqM,rsqO,rsqH1,rsqH2,rsqM;
int n,k,ii,is3,ii3,ntiA,nj0,nj1;
int jnra,jnrb,jnrc,jnrd;
int j3a,j3b,j3c,j3d;
int nri, ntype, nouter, ninner;
int tja,tjb,tjc,tjd;
#ifdef GMX_THREADS
int nn0, nn1;
#endif
nouter = 0;
ninner = 0;
nri = *p_nri;
ntype = *p_ntype;
nul=vec_zero();
tsc=load_float_and_splat(p_tabscale);
vfacel=load_float_and_splat(p_facel);
ii = iinr[0];
iqH = vec_madd(load_float_and_splat(charge+ii+1),vfacel,nul);
iqM = vec_madd(load_float_and_splat(charge+ii+3),vfacel,nul);
ntiA = 2*ntype*type[ii];
#ifdef GMX_THREADS
nthreads = *p_nthreads;
do {
gmx_thread_mutex_lock((gmx_thread_mutex_t *)mtx);
nn0 = *count;
nn1 = nn0+(nri-nn0)/(2*nthreads)+3;
*count = nn1;
gmx_thread_mutex_unlock((gmx_thread_mutex_t *)mtx);
if(nn1>nri) nn1=nri;
for(n=nn0; (n<nn1); n++) {
#if 0
} /* maintain correct indentation even with conditional left braces */
#endif
#else /* without gmx_threads */
for(n=0;n<nri;n++) {
#endif
is3 = 3*shift[n];
ii = iinr[n];
ii3 = 3*ii;
load_1_4atoms_shift_and_splat(pos+ii3,shiftvec+is3,&iOx,&iOy,&iOz,
&iH1x,&iH1y,&iH1z,&iH2x,&iH2y,&iH2z,
&iMx,&iMy,&iMz);
vctot = nul;
Vvdwtot = nul;
fiOx = nul;
fiOy = nul;
fiOz = nul;
fiH1x = nul;
fiH1y = nul;
fiH1z = nul;
fiH2x = nul;
fiH2y = nul;
fiH2z = nul;
fiMx = nul;
fiMy = nul;
fiMz = nul;
nj0 = jindex[n];
nj1 = jindex[n+1];
for(k=nj0; k<(nj1-3); k+=4)
{
jnra = jjnr[k];
jnrb = jjnr[k+1];
jnrc = jjnr[k+2];
jnrd = jjnr[k+3];
j3a = 3*jnra;
j3b = 3*jnrb;
j3c = 3*jnrc;
j3d = 3*jnrd;
transpose_4_to_3(load_xyz(pos+j3a),
load_xyz(pos+j3b),
load_xyz(pos+j3c),
load_xyz(pos+j3d),&dMx,&dMy,&dMz);
dOx = vec_sub(iOx,dMx);
dOy = vec_sub(iOy,dMy);
dOz = vec_sub(iOz,dMz);
dH1x = vec_sub(iH1x,dMx);
dH1y = vec_sub(iH1y,dMy);
dH1z = vec_sub(iH1z,dMz);
dH2x = vec_sub(iH2x,dMx);
dH2y = vec_sub(iH2y,dMy);
dH2z = vec_sub(iH2z,dMz);
dMx = vec_sub(iMx,dMx);
dMy = vec_sub(iMy,dMy);
dMz = vec_sub(iMz,dMz);
rsqO = vec_madd(dOx,dOx,nul);
rsqH1 = vec_madd(dH1x,dH1x,nul);
rsqH2 = vec_madd(dH2x,dH2x,nul);
rsqM = vec_madd(dMx,dMx,nul);
rsqO = vec_madd(dOy,dOy,rsqO);
rsqH1 = vec_madd(dH1y,dH1y,rsqH1);
rsqH2 = vec_madd(dH2y,dH2y,rsqH2);
rsqM = vec_madd(dMy,dMy,rsqM);
rsqO = vec_madd(dOz,dOz,rsqO);
rsqH1 = vec_madd(dH1z,dH1z,rsqH1);
rsqH2 = vec_madd(dH2z,dH2z,rsqH2);
rsqM = vec_madd(dMz,dMz,rsqM);
rinvO = do_invsqrt(rsqO);
do_3_invsqrt(rsqM,rsqH1,rsqH2,&rinvM,&rinvH1,&rinvH2);
rinvsqM = vec_madd(rinvM,rinvM,nul);
rinvsqH1 = vec_madd(rinvH1,rinvH1,nul);
rinvsqH2 = vec_madd(rinvH2,rinvH2,nul);
r = vec_madd(rsqO,rinvO,nul);
tja = ntiA+2*type[jnra];
tjb = ntiA+2*type[jnrb];
tjc = ntiA+2*type[jnrc];
tjd = ntiA+2*type[jnrd];
/* load 4 j charges and multiply by iq */
jq=load_4_float(charge+jnra,charge+jnrb,charge+jnrc,
charge+jnrd);
load_4_pair(vdwparam+tja,vdwparam+tjb,vdwparam+tjc,vdwparam+tjd,&c6,&c12);
do_4_ljtable_lj(VFtab,vec_madd(r,tsc,nul),&VVd,&FFd,&VVr,&FFr);
Vvdwtot = vec_madd(c6,VVd,Vvdwtot);
fsO = vec_nmsub(c6,FFd,nul);
Vvdwtot = vec_madd(c12,VVr,Vvdwtot);
fsO = vec_nmsub(c12,FFr,fsO);
fsO = vec_madd(fsO,tsc,nul);
fsO = vec_madd(fsO,rinvO,nul);
qqM = vec_madd(iqM,jq,nul);
qqH = vec_madd(iqH,jq,nul);
vcoulM = vec_madd(qqM,rinvM,nul);
vcoulH1 = vec_madd(qqH,rinvH1,nul);
vcoulH2 = vec_madd(qqH,rinvH2,nul);
fsH1 = vec_madd(vcoulH1,rinvsqH1,nul);
fsH2 = vec_madd(vcoulH2,rinvsqH2,nul);
fsM = vec_madd(vcoulM,rinvsqM,nul);
vctot = vec_add(vctot,vcoulM);
vcoulH1 = vec_add(vcoulH1,vcoulH2);
vctot = vec_add(vctot,vcoulH1);
fiOx = vec_madd(fsO,dOx,fiOx); /* +=fx */
dOx = vec_nmsub(fsO,dOx,nul); /* -fx */
fiOy = vec_madd(fsO,dOy,fiOy); /* +=fy */
dOy = vec_nmsub(fsO,dOy,nul); /* -fy */
fiOz = vec_madd(fsO,dOz,fiOz); /* +=fz */
dOz = vec_nmsub(fsO,dOz,nul); /* -fz */
fiH1x = vec_madd(fsH1,dH1x,fiH1x); /* +=fx */
dOx = vec_nmsub(fsH1,dH1x,dOx); /* -fx */
fiH1y = vec_madd(fsH1,dH1y,fiH1y); /* +=fy */
dOy = vec_nmsub(fsH1,dH1y,dOy); /* -fy */
fiH1z = vec_madd(fsH1,dH1z,fiH1z); /* +=fz */
dOz = vec_nmsub(fsH1,dH1z,dOz); /* -fz */
fiH2x = vec_madd(fsH2,dH2x,fiH2x); /* +=fx */
dOx = vec_nmsub(fsH2,dH2x,dOx); /* -fx */
fiH2y = vec_madd(fsH2,dH2y,fiH2y); /* +=fy */
dOy = vec_nmsub(fsH2,dH2y,dOy); /* -fy */
fiH2z = vec_madd(fsH2,dH2z,fiH2z); /* +=fz */
dOz = vec_nmsub(fsH2,dH2z,dOz); /* -fz */
fiMx = vec_madd(fsM,dMx,fiMx); /* +=fx */
dOx = vec_nmsub(fsM,dMx,dOx); /* -fx */
fiMy = vec_madd(fsM,dMy,fiMy); /* +=fy */
dOy = vec_nmsub(fsM,dMy,dOy); /* -fy */
fiMz = vec_madd(fsM,dMz,fiMz); /* +=fz */
dOz = vec_nmsub(fsM,dMz,dOz); /* -fz */
transpose_3_to_4(dOx,dOy,dOz,&tmp1,&tmp2,&tmp3,&tmp4);
add_xyz_to_mem(faction+j3a,tmp1);
add_xyz_to_mem(faction+j3b,tmp2);
add_xyz_to_mem(faction+j3c,tmp3);
add_xyz_to_mem(faction+j3d,tmp4);
}
if(k<(nj1-2))
{
jnra = jjnr[k];
jnrb = jjnr[k+1];
jnrc = jjnr[k+2];
j3a = 3*jnra;
j3b = 3*jnrb;
j3c = 3*jnrc;
transpose_4_to_3(load_xyz(pos+j3a),
load_xyz(pos+j3b),
load_xyz(pos+j3c),nul,&dMx,&dMy,&dMz);
dOx = vec_sub(iOx,dMx);
dOy = vec_sub(iOy,dMy);
dOz = vec_sub(iOz,dMz);
dH1x = vec_sub(iH1x,dMx);
dH1y = vec_sub(iH1y,dMy);
dH1z = vec_sub(iH1z,dMz);
dH2x = vec_sub(iH2x,dMx);
dH2y = vec_sub(iH2y,dMy);
dH2z = vec_sub(iH2z,dMz);
dMx = vec_sub(iMx,dMx);
dMy = vec_sub(iMy,dMy);
dMz = vec_sub(iMz,dMz);
rsqO = vec_madd(dOx,dOx,nul);
rsqH1 = vec_madd(dH1x,dH1x,nul);
rsqH2 = vec_madd(dH2x,dH2x,nul);
rsqM = vec_madd(dMx,dMx,nul);
rsqO = vec_madd(dOy,dOy,rsqO);
rsqH1 = vec_madd(dH1y,dH1y,rsqH1);
rsqH2 = vec_madd(dH2y,dH2y,rsqH2);
rsqM = vec_madd(dMy,dMy,rsqM);
rsqO = vec_madd(dOz,dOz,rsqO);
rsqH1 = vec_madd(dH1z,dH1z,rsqH1);
rsqH2 = vec_madd(dH2z,dH2z,rsqH2);
rsqM = vec_madd(dMz,dMz,rsqM);
rinvO = do_invsqrt(rsqO);
do_3_invsqrt(rsqM,rsqH1,rsqH2,&rinvM,&rinvH1,&rinvH2);
zero_highest_element_in_vector(&rinvO);
zero_highest_element_in_vector(&rsqO);
zero_highest_element_in_3_vectors(&rinvM,&rinvH1,&rinvH2);
rinvsqM = vec_madd(rinvM,rinvM,nul);
rinvsqH1 = vec_madd(rinvH1,rinvH1,nul);
rinvsqH2 = vec_madd(rinvH2,rinvH2,nul);
r = vec_madd(rsqO,rinvO,nul);
jq=load_3_float(charge+jnra,charge+jnrb,charge+jnrc);
tja = ntiA+2*type[jnra];
tjb = ntiA+2*type[jnrb];
tjc = ntiA+2*type[jnrc];
/* load 3 j charges and multiply by iq */
load_3_pair(vdwparam+tja,vdwparam+tjb,vdwparam+tjc,&c6,&c12);
do_3_ljtable_lj(VFtab,vec_madd(r,tsc,nul),&VVd,&FFd,&VVr,&FFr);
Vvdwtot = vec_madd(c6,VVd,Vvdwtot);
fsO = vec_nmsub(c6,FFd,nul);
Vvdwtot = vec_madd(c12,VVr,Vvdwtot);
fsO = vec_nmsub(c12,FFr,fsO);
fsO = vec_madd(fsO,tsc,nul);
fsO = vec_madd(fsO,rinvO,nul);
qqM = vec_madd(iqM,jq,nul);
qqH = vec_madd(iqH,jq,nul);
vcoulM = vec_madd(qqM,rinvM,nul);
vcoulH1 = vec_madd(qqH,rinvH1,nul);
vcoulH2 = vec_madd(qqH,rinvH2,nul);
fsH1 = vec_madd(vcoulH1,rinvsqH1,nul);
fsH2 = vec_madd(vcoulH2,rinvsqH2,nul);
fsM = vec_madd(vcoulM,rinvsqM,nul);
vctot = vec_add(vctot,vcoulM);
vcoulH1 = vec_add(vcoulH1,vcoulH2);
vctot = vec_add(vctot,vcoulH1);
fiOx = vec_madd(fsO,dOx,fiOx); /* +=fx */
dOx = vec_nmsub(fsO,dOx,nul); /* -fx */
fiOy = vec_madd(fsO,dOy,fiOy); /* +=fy */
dOy = vec_nmsub(fsO,dOy,nul); /* -fy */
fiOz = vec_madd(fsO,dOz,fiOz); /* +=fz */
dOz = vec_nmsub(fsO,dOz,nul); /* -fz */
fiH1x = vec_madd(fsH1,dH1x,fiH1x); /* +=fx */
dOx = vec_nmsub(fsH1,dH1x,dOx); /* -fx */
fiH1y = vec_madd(fsH1,dH1y,fiH1y); /* +=fy */
dOy = vec_nmsub(fsH1,dH1y,dOy); /* -fy */
fiH1z = vec_madd(fsH1,dH1z,fiH1z); /* +=fz */
dOz = vec_nmsub(fsH1,dH1z,dOz); /* -fz */
fiH2x = vec_madd(fsH2,dH2x,fiH2x); /* +=fx */
dOx = vec_nmsub(fsH2,dH2x,dOx); /* -fx */
fiH2y = vec_madd(fsH2,dH2y,fiH2y); /* +=fy */
dOy = vec_nmsub(fsH2,dH2y,dOy); /* -fy */
fiH2z = vec_madd(fsH2,dH2z,fiH2z); /* +=fz */
dOz = vec_nmsub(fsH2,dH2z,dOz); /* -fz */
fiMx = vec_madd(fsM,dMx,fiMx); /* +=fx */
dOx = vec_nmsub(fsM,dMx,dOx); /* -fx */
fiMy = vec_madd(fsM,dMy,fiMy); /* +=fy */
dOy = vec_nmsub(fsM,dMy,dOy); /* -fy */
fiMz = vec_madd(fsM,dMz,fiMz); /* +=fz */
dOz = vec_nmsub(fsM,dMz,dOz); /* -fz */
transpose_4_to_3(dOx,dOy,dOz,nul,&tmp1,&tmp2,&tmp3);
add_xyz_to_mem(faction+j3a,tmp1);
add_xyz_to_mem(faction+j3b,tmp2);
add_xyz_to_mem(faction+j3c,tmp3);
}
else if(k<(nj1-1))
{
jnra = jjnr[k];
jnrb = jjnr[k+1];
j3a = 3*jnra;
j3b = 3*jnrb;
transpose_2_to_3(load_xyz(pos+j3a),
load_xyz(pos+j3b),&dMx,&dMy,&dMz);
dOx = vec_sub(iOx,dMx);
dOy = vec_sub(iOy,dMy);
dOz = vec_sub(iOz,dMz);
dH1x = vec_sub(iH1x,dMx);
dH1y = vec_sub(iH1y,dMy);
dH1z = vec_sub(iH1z,dMz);
dH2x = vec_sub(iH2x,dMx);
dH2y = vec_sub(iH2y,dMy);
dH2z = vec_sub(iH2z,dMz);
dMx = vec_sub(iMx,dMx);
dMy = vec_sub(iMy,dMy);
dMz = vec_sub(iMz,dMz);
rsqO = vec_madd(dOx,dOx,nul);
rsqH1 = vec_madd(dH1x,dH1x,nul);
rsqH2 = vec_madd(dH2x,dH2x,nul);
rsqM = vec_madd(dMx,dMx,nul);
rsqO = vec_madd(dOy,dOy,rsqO);
rsqH1 = vec_madd(dH1y,dH1y,rsqH1);
rsqH2 = vec_madd(dH2y,dH2y,rsqH2);
rsqM = vec_madd(dMy,dMy,rsqM);
rsqO = vec_madd(dOz,dOz,rsqO);
rsqH1 = vec_madd(dH1z,dH1z,rsqH1);
rsqH2 = vec_madd(dH2z,dH2z,rsqH2);
rsqM = vec_madd(dMz,dMz,rsqM);
rinvO = do_invsqrt(rsqO);
do_3_invsqrt(rsqM,rsqH1,rsqH2,&rinvM,&rinvH1,&rinvH2);
zero_highest_2_elements_in_vector(&rinvO);
zero_highest_2_elements_in_vector(&rsqO);
zero_highest_2_elements_in_3_vectors(&rinvM,&rinvH1,&rinvH2);
rinvsqM = vec_madd(rinvM,rinvM,nul);
rinvsqH1 = vec_madd(rinvH1,rinvH1,nul);
rinvsqH2 = vec_madd(rinvH2,rinvH2,nul);
r = vec_madd(rsqO,rinvO,nul);
jq=load_2_float(charge+jnra,charge+jnrb);
tja = ntiA+2*type[jnra];
tjb = ntiA+2*type[jnrb];
/* load 2 j charges and multiply by iq */
load_2_pair(vdwparam+tja,vdwparam+tjb,&c6,&c12);
do_2_ljtable_lj(VFtab,vec_madd(r,tsc,nul),&VVd,&FFd,&VVr,&FFr);
Vvdwtot = vec_madd(c6,VVd,Vvdwtot);
fsO = vec_nmsub(c6,FFd,nul);
Vvdwtot = vec_madd(c12,VVr,Vvdwtot);
fsO = vec_nmsub(c12,FFr,fsO);
fsO = vec_madd(fsO,tsc,nul);
fsO = vec_madd(fsO,rinvO,nul);
qqM = vec_madd(iqM,jq,nul);
qqH = vec_madd(iqH,jq,nul);
vcoulM = vec_madd(qqM,rinvM,nul);
vcoulH1 = vec_madd(qqH,rinvH1,nul);
vcoulH2 = vec_madd(qqH,rinvH2,nul);
fsH1 = vec_madd(vcoulH1,rinvsqH1,nul);
fsH2 = vec_madd(vcoulH2,rinvsqH2,nul);
fsM = vec_madd(vcoulM,rinvsqM,nul);
vctot = vec_add(vctot,vcoulM);
vcoulH1 = vec_add(vcoulH1,vcoulH2);
vctot = vec_add(vctot,vcoulH1);
fiOx = vec_madd(fsO,dOx,fiOx); /* +=fx */
dOx = vec_nmsub(fsO,dOx,nul); /* -fx */
fiOy = vec_madd(fsO,dOy,fiOy); /* +=fy */
dOy = vec_nmsub(fsO,dOy,nul); /* -fy */
fiOz = vec_madd(fsO,dOz,fiOz); /* +=fz */
dOz = vec_nmsub(fsO,dOz,nul); /* -fz */
fiH1x = vec_madd(fsH1,dH1x,fiH1x); /* +=fx */
dOx = vec_nmsub(fsH1,dH1x,dOx); /* -fx */
fiH1y = vec_madd(fsH1,dH1y,fiH1y); /* +=fy */
dOy = vec_nmsub(fsH1,dH1y,dOy); /* -fy */
fiH1z = vec_madd(fsH1,dH1z,fiH1z); /* +=fz */
dOz = vec_nmsub(fsH1,dH1z,dOz); /* -fz */
fiH2x = vec_madd(fsH2,dH2x,fiH2x); /* +=fx */
dOx = vec_nmsub(fsH2,dH2x,dOx); /* -fx */
fiH2y = vec_madd(fsH2,dH2y,fiH2y); /* +=fy */
dOy = vec_nmsub(fsH2,dH2y,dOy); /* -fy */
fiH2z = vec_madd(fsH2,dH2z,fiH2z); /* +=fz */
dOz = vec_nmsub(fsH2,dH2z,dOz); /* -fz */
fiMx = vec_madd(fsM,dMx,fiMx); /* +=fx */
dOx = vec_nmsub(fsM,dMx,dOx); /* -fx */
fiMy = vec_madd(fsM,dMy,fiMy); /* +=fy */
dOy = vec_nmsub(fsM,dMy,dOy); /* -fy */
fiMz = vec_madd(fsM,dMz,fiMz); /* +=fz */
dOz = vec_nmsub(fsM,dMz,dOz); /* -fz */
transpose_4_to_2(dOx,dOy,dOz,nul,&tmp1,&tmp2);
add_xyz_to_mem(faction+j3a,tmp1);
add_xyz_to_mem(faction+j3b,tmp2);
}
else if(k<nj1)
{
jnra = jjnr[k];
j3a = 3*jnra;
transpose_1_to_3(load_xyz(pos+j3a),
&dMx,&dMy,&dMz);
dOx = vec_sub(iOx,dMx);
dOy = vec_sub(iOy,dMy);
dOz = vec_sub(iOz,dMz);
dH1x = vec_sub(iH1x,dMx);
dH1y = vec_sub(iH1y,dMy);
dH1z = vec_sub(iH1z,dMz);
dH2x = vec_sub(iH2x,dMx);
dH2y = vec_sub(iH2y,dMy);
dH2z = vec_sub(iH2z,dMz);
dMx = vec_sub(iMx,dMx);
dMy = vec_sub(iMy,dMy);
dMz = vec_sub(iMz,dMz);
rsqO = vec_madd(dOx,dOx,nul);
rsqH1 = vec_madd(dH1x,dH1x,nul);
rsqH2 = vec_madd(dH2x,dH2x,nul);
rsqM = vec_madd(dMx,dMx,nul);
rsqO = vec_madd(dOy,dOy,rsqO);
rsqH1 = vec_madd(dH1y,dH1y,rsqH1);
rsqH2 = vec_madd(dH2y,dH2y,rsqH2);
rsqM = vec_madd(dMy,dMy,rsqM);
rsqO = vec_madd(dOz,dOz,rsqO);
rsqH1 = vec_madd(dH1z,dH1z,rsqH1);
rsqH2 = vec_madd(dH2z,dH2z,rsqH2);
rsqM = vec_madd(dMz,dMz,rsqM);
rinvO = do_invsqrt(rsqO);
do_3_invsqrt(rsqM,rsqH1,rsqH2,&rinvM,&rinvH1,&rinvH2);
zero_highest_3_elements_in_vector(&rinvO);
zero_highest_3_elements_in_vector(&rsqO);
zero_highest_3_elements_in_3_vectors(&rinvM,&rinvH1,&rinvH2);
rinvsqM = vec_madd(rinvM,rinvM,nul);
rinvsqH1 = vec_madd(rinvH1,rinvH1,nul);
rinvsqH2 = vec_madd(rinvH2,rinvH2,nul);
r = vec_madd(rsqO,rinvO,nul);
tja = ntiA+2*type[jnra];
/* load 1 j charges and multiply by iq */
jq=load_1_float(charge+jnra);
load_1_pair(vdwparam+tja,&c6,&c12);
do_1_ljtable_lj(VFtab,vec_madd(r,tsc,nul),&VVd,&FFd,&VVr,&FFr);
Vvdwtot = vec_madd(c6,VVd,Vvdwtot);
fsO = vec_nmsub(c6,FFd,nul);
Vvdwtot = vec_madd(c12,VVr,Vvdwtot);
fsO = vec_nmsub(c12,FFr,fsO);
fsO = vec_madd(fsO,tsc,nul);
fsO = vec_madd(fsO,rinvO,nul);
qqM = vec_madd(iqM,jq,nul);
qqH = vec_madd(iqH,jq,nul);
vcoulM = vec_madd(qqM,rinvM,nul);
vcoulH1 = vec_madd(qqH,rinvH1,nul);
vcoulH2 = vec_madd(qqH,rinvH2,nul);
fsH1 = vec_madd(vcoulH1,rinvsqH1,nul);
fsH2 = vec_madd(vcoulH2,rinvsqH2,nul);
fsM = vec_madd(vcoulM,rinvsqM,nul);
vctot = vec_add(vctot,vcoulM);
vcoulH1 = vec_add(vcoulH1,vcoulH2);
vctot = vec_add(vctot,vcoulH1);
fiOx = vec_madd(fsO,dOx,fiOx); /* +=fx */
dOx = vec_nmsub(fsO,dOx,nul); /* -fx */
fiOy = vec_madd(fsO,dOy,fiOy); /* +=fy */
dOy = vec_nmsub(fsO,dOy,nul); /* -fy */
fiOz = vec_madd(fsO,dOz,fiOz); /* +=fz */
dOz = vec_nmsub(fsO,dOz,nul); /* -fz */
fiH1x = vec_madd(fsH1,dH1x,fiH1x); /* +=fx */
dOx = vec_nmsub(fsH1,dH1x,dOx); /* -fx */
fiH1y = vec_madd(fsH1,dH1y,fiH1y); /* +=fy */
dOy = vec_nmsub(fsH1,dH1y,dOy); /* -fy */
fiH1z = vec_madd(fsH1,dH1z,fiH1z); /* +=fz */
dOz = vec_nmsub(fsH1,dH1z,dOz); /* -fz */
fiH2x = vec_madd(fsH2,dH2x,fiH2x); /* +=fx */
dOx = vec_nmsub(fsH2,dH2x,dOx); /* -fx */
fiH2y = vec_madd(fsH2,dH2y,fiH2y); /* +=fy */
dOy = vec_nmsub(fsH2,dH2y,dOy); /* -fy */
fiH2z = vec_madd(fsH2,dH2z,fiH2z); /* +=fz */
dOz = vec_nmsub(fsH2,dH2z,dOz); /* -fz */
fiMx = vec_madd(fsM,dMx,fiMx); /* +=fx */
dOx = vec_nmsub(fsM,dMx,dOx); /* -fx */
fiMy = vec_madd(fsM,dMy,fiMy); /* +=fy */
dOy = vec_nmsub(fsM,dMy,dOy); /* -fy */
fiMz = vec_madd(fsM,dMz,fiMz); /* +=fz */
dOz = vec_nmsub(fsM,dMz,dOz); /* -fz */
transpose_4_to_1(dOx,dOy,dOz,nul,&tmp1);
add_xyz_to_mem(faction+j3a,tmp1);
}
/* update outer data */
update_i_4atoms_forces(faction+ii3,fshift+is3,
fiOx,fiOy,fiOz,fiH1x,fiH1y,fiH1z,
fiH2x,fiH2y,fiH2z,fiMx,fiMy,fiMz);
add_vector_to_float(Vc+gid[n],vctot);
add_vector_to_float(Vvdw+gid[n],Vvdwtot);
ninner += nj1 - nj0;
}
#ifdef GMX_THREADS
nouter += nn1 - nn0;
} while (nn1<nri);
#else
nouter = nri;
#endif
*outeriter = nouter;
*inneriter = ninner;
}
float vsincos2f(float x) {
// Load x into an aligned float array
float __attribute__((aligned(16))) xa[4];
xa[0] = x;
// We want to calculate these:
// nom = 166320.0 * x - 22260.0 * POW3(x) + 551.0 * POW5(x);
// denom = 166320.0 + 5460.0 * POW2(x) + 75.0 * POW4(x);
// res = nom/denom;
//
// We first setup our constants:
// vc1 = | a1 | a3 | b0 | b2 |
// vc2 = | 0.0 | a5 | 0.0 | 0.0 |
vector float vc1 = { 166320.0, -22260, 166320.0, 5460.0 },
vc2 = { 0.0, 551.0, 0.0, 75.0 };
vector float vx = vec_ld(0, xa);
vector float vres, vdenom, vest1,
vx2, vx02, vx13, vx24,
v0 = (vector float)vec_splat_u32(0),
v1 = vec_ctf(vec_splat_u32(1),0);
// Load x into a vector and splat it all over
vx = vec_splat(vx, 0);
// get the vector with all elements: x^2
vx2 = vec_madd(vx, vx, v0);
// We need a vector with | 1.0 | x^2 | 1.0 | x^2 |
vx02 = vec_mergeh(v1, vx2);
// Multiply with x -> | x | x^3 | x | x^3 |
vx13 = vec_madd(vx, vx02, v0);
// Now shift left and combine with vx02 -> | x | x^3 | 1.0 | x^2 |
vx13 = vec_sld(vx13, vx02, 8);
// Again with x^2 -> | x^3 | x^5 | x^2 | x^4 |
vx24 = vec_madd(vx13, vx2, v0);
// Multiply with the coefficients vectors:
// First with vc1 -> | a1*x | a3*x^3 | b0*1.0 | b2*x^2 |
vres = vec_madd(vx13, vc1, v0);
// Now with vc2 (and add previous result) -> | a1*x + 0*x^3 | a3*x^3 + a5*x^5 | b0*1.0 + 0.0*x^2 | b2*x^2 + b4*x^4 |
vres = vec_madd(vx24, vc2, vres);
// Shift left by 4 and add the vectors -> | nom | .. | denom | .. |
vres = vec_add(vres, vec_sld(vres, vres, 4));
// Now splat denom (we don't have to splat nom, we'll just take the first element after the division.
vdenom = vec_splat(vres, 2);
vest1 = vec_re(vdenom);
//1st round of Newton-Raphson refinement
vdenom = vec_madd( vest1, vec_nmsub( vest1, vdenom, v1 ), vest1 );
// 2nd round of Newton-Raphson refinement
// vdenom = vec_madd( vest2, vec_nmsub( vest2, vdenom, v1 ), vest2 );
vres = vec_madd(vres, vdenom, v0);
vec_st(vres, 0, xa);
//printf("vres = %2.7f %2.7f %2.7f %2.7f\n", xa[0], xa[1], xa[2], xa[3]);
/* float nom, denom, res;
nom = 166320.0 * x - 22260.0 * POW3(x) + 551.0 * POW5(x);
denom = 166320.0 + 5460.0 * POW2(x) + 75.0 * POW4(x);
printf("nom = %2.7f, denom = %2.7f\n", nom, denom);
res = nom/denom;
printf("res = %2.7f\n", res);*/
printf("res = %2.7f\n", xa[0]);
return xa[0];
}
void test1() {
// CHECK-LABEL: define void @test1
res_vd = vec_add(vd, vd);
// CHECK: 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>
res_vd = vec_and(vd, vbll);
// CHECK: and <2 x i64>
// CHECK: 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>
dummy();
// CHECK: 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>
dummy();
// CHECK: 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>
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()
res_vd = vec_ceil(vd);
// CHECK: 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]*}})
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]*}})
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]*}})
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]*}})
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]*}})
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]*}})
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]*}})
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]*}})
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]*}})
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]*}})
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]*}})
/* vec_div */
res_vf = vec_div(vf, vf);
// CHECK: @llvm.ppc.vsx.xvdivsp
res_vd = vec_div(vd, vd);
// CHECK: @llvm.ppc.vsx.xvdivdp
/* vec_max */
res_vf = vec_max(vf, vf);
// CHECK: @llvm.ppc.vsx.xvmaxsp
res_vd = vec_max(vd, vd);
// CHECK: @llvm.ppc.vsx.xvmaxdp
res_vf = vec_vmaxfp(vf, vf);
// CHECK: @llvm.ppc.vsx.xvmaxsp
/* vec_min */
res_vf = vec_min(vf, vf);
// CHECK: @llvm.ppc.vsx.xvminsp
res_vd = vec_min(vd, vd);
// CHECK: @llvm.ppc.vsx.xvmindp
res_vf = vec_vminfp(vf, vf);
// CHECK: @llvm.ppc.vsx.xvminsp
res_d = __builtin_vsx_xsmaxdp(d, d);
// CHECK: @llvm.ppc.vsx.xsmaxdp
res_d = __builtin_vsx_xsmindp(d, d);
// CHECK: @llvm.ppc.vsx.xsmindp
/* vec_perm */
res_vsll = vec_perm(vsll, vsll, vuc);
// CHECK: @llvm.ppc.altivec.vperm
res_vull = vec_perm(vull, vull, vuc);
// CHECK: @llvm.ppc.altivec.vperm
res_vd = vec_perm(vd, vd, vuc);
// CHECK: @llvm.ppc.altivec.vperm
res_vsll = vec_vperm(vsll, vsll, vuc);
// CHECK: @llvm.ppc.altivec.vperm
res_vull = vec_vperm(vull, vull, vuc);
// CHECK: @llvm.ppc.altivec.vperm
res_vd = vec_vperm(vd, vd, vuc);
// CHECK: @llvm.ppc.altivec.vperm
/* vec_vsx_ld */
res_vsi = vec_vsx_ld(0, &vsi);
// CHECK: @llvm.ppc.vsx.lxvw4x
res_vui = vec_vsx_ld(0, &vui);
// CHECK: @llvm.ppc.vsx.lxvw4x
res_vf = vec_vsx_ld (0, &vf);
// CHECK: @llvm.ppc.vsx.lxvw4x
res_vsll = vec_vsx_ld(0, &vsll);
// CHECK: @llvm.ppc.vsx.lxvd2x
res_vull = vec_vsx_ld(0, &vull);
// CHECK: @llvm.ppc.vsx.lxvd2x
res_vd = vec_vsx_ld(0, &vd);
// CHECK: @llvm.ppc.vsx.lxvd2x
/* vec_vsx_st */
vec_vsx_st(vsi, 0, &res_vsi);
// CHECK: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vui, 0, &res_vui);
// CHECK: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vf, 0, &res_vf);
// CHECK: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vsll, 0, &res_vsll);
// CHECK: @llvm.ppc.vsx.stxvd2x
vec_vsx_st(vull, 0, &res_vull);
// CHECK: @llvm.ppc.vsx.stxvd2x
vec_vsx_st(vd, 0, &res_vd);
// CHECK: @llvm.ppc.vsx.stxvd2x
/* vec_and */
res_vsll = vec_and(vsll, vsll);
// CHECK: and <2 x i64>
res_vsll = vec_and(vbll, vsll);
// CHECK: and <2 x i64>
res_vsll = vec_and(vsll, vbll);
// CHECK: and <2 x i64>
res_vull = vec_and(vull, vull);
// CHECK: and <2 x i64>
res_vull = vec_and(vbll, vull);
// CHECK: and <2 x i64>
res_vull = vec_and(vull, vbll);
// CHECK: and <2 x i64>
res_vbll = vec_and(vbll, vbll);
// CHECK: and <2 x i64>
/* vec_vand */
res_vsll = vec_vand(vsll, vsll);
// CHECK: and <2 x i64>
res_vsll = vec_vand(vbll, vsll);
// CHECK: and <2 x i64>
res_vsll = vec_vand(vsll, vbll);
// CHECK: and <2 x i64>
res_vull = vec_vand(vull, vull);
// CHECK: and <2 x i64>
res_vull = vec_vand(vbll, vull);
// CHECK: and <2 x i64>
res_vull = vec_vand(vull, vbll);
// CHECK: and <2 x i64>
res_vbll = vec_vand(vbll, vbll);
// CHECK: and <2 x i64>
/* vec_andc */
res_vsll = vec_andc(vsll, vsll);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
res_vsll = vec_andc(vbll, vsll);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
res_vsll = vec_andc(vsll, vbll);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
res_vull = vec_andc(vull, vull);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
res_vull = vec_andc(vbll, vull);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
res_vull = vec_andc(vull, vbll);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
res_vbll = vec_andc(vbll, vbll);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
res_vf = vec_floor(vf);
// CHECK: 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]+}})
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]+}})
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]+}})
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>
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>
res_vf = vec_mul(vf, vf);
// CHECK: fmul <4 x float> %{{[0-9]+}}, %{{[0-9]+}}
res_vd = vec_mul(vd, vd);
// CHECK: 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]+}})
res_vd = vec_nearbyint(vd);
// CHECK: 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]]
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]]
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]+}}
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]]
/* vec_nor */
res_vsll = vec_nor(vsll, vsll);
// CHECK: or <2 x i64>
// CHECK: xor <2 x i64>
res_vull = vec_nor(vull, vull);
// CHECK: or <2 x i64>
// CHECK: xor <2 x i64>
res_vull = vec_nor(vbll, vbll);
// CHECK: or <2 x i64>
// CHECK: 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>
/* vec_or */
res_vsll = vec_or(vsll, vsll);
// CHECK: or <2 x i64>
res_vsll = vec_or(vbll, vsll);
// CHECK: or <2 x i64>
res_vsll = vec_or(vsll, vbll);
// CHECK: or <2 x i64>
res_vull = vec_or(vull, vull);
// CHECK: or <2 x i64>
res_vull = vec_or(vbll, vull);
// CHECK: or <2 x i64>
res_vull = vec_or(vull, vbll);
// CHECK: or <2 x i64>
res_vbll = vec_or(vbll, vbll);
// CHECK: 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]+}}
res_vf = vec_rint(vf);
// CHECK: 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]+}})
res_vf = vec_rsqrte(vf);
// CHECK: 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]+}})
dummy();
// CHECK: 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>
dummy();
// CHECK: 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>
res_vf = vec_sqrt(vf);
// CHECK: 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]+}})
res_vd = vec_sub(vd, vd);
// CHECK: 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]+}})
res_vd = vec_trunc(vd);
// CHECK: 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>
res_vsll = vec_vor(vbll, vsll);
// CHECK: or <2 x i64>
res_vsll = vec_vor(vsll, vbll);
// CHECK: or <2 x i64>
res_vull = vec_vor(vull, vull);
// CHECK: or <2 x i64>
res_vull = vec_vor(vbll, vull);
// CHECK: or <2 x i64>
res_vull = vec_vor(vull, vbll);
// CHECK: or <2 x i64>
res_vbll = vec_vor(vbll, vbll);
// CHECK: or <2 x i64>
/* vec_xor */
res_vsll = vec_xor(vsll, vsll);
// CHECK: xor <2 x i64>
res_vsll = vec_xor(vbll, vsll);
// CHECK: xor <2 x i64>
res_vsll = vec_xor(vsll, vbll);
// CHECK: xor <2 x i64>
res_vull = vec_xor(vull, vull);
// CHECK: xor <2 x i64>
res_vull = vec_xor(vbll, vull);
// CHECK: xor <2 x i64>
res_vull = vec_xor(vull, vbll);
// CHECK: xor <2 x i64>
res_vbll = vec_xor(vbll, vbll);
// CHECK: xor <2 x i64>
dummy();
// CHECK: 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>
dummy();
// CHECK: 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>
dummy();
// CHECK: 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>
/* vec_vxor */
res_vsll = vec_vxor(vsll, vsll);
// CHECK: xor <2 x i64>
res_vsll = vec_vxor(vbll, vsll);
// CHECK: xor <2 x i64>
res_vsll = vec_vxor(vsll, vbll);
// CHECK: xor <2 x i64>
res_vull = vec_vxor(vull, vull);
// CHECK: xor <2 x i64>
res_vull = vec_vxor(vbll, vull);
// CHECK: xor <2 x i64>
res_vull = vec_vxor(vull, vbll);
// CHECK: xor <2 x i64>
res_vbll = vec_vxor(vbll, vbll);
// CHECK: xor <2 x i64>
}
vector float
f(vector float a, vector float b, vector float c)
{
return vec_nmsub(a, vec_re(b), vec_nmsub(b, c, vec_expte(a)));
}
/* a vectorized version of the Voigt function using Altivec / VMX instructions */
void my_voigt(const float *damping, const float *frequency_offset, float *voigt_value, int N)
{
// coefficients of the rational approximation formula
// to the complementary error function
const vector float A0 = (vector float) (122.607931777104326f);
const vector float A1 = (vector float) (214.382388694706425f);
const vector float A2 = (vector float) (181.928533092181549f);
const vector float A3 = (vector float) (93.155580458138441f);
const vector float A4 = (vector float) (30.180142196210589f);
const vector float A5 = (vector float) (5.912626209773153f);
const vector float A6 = (vector float) (0.564189583562615f);
const vector float B0 = (vector float) (122.60793177387535f);
const vector float B1 = (vector float) (352.730625110963558f);
const vector float B2 = (vector float) (457.334478783897737f);
const vector float B3 = (vector float) (348.703917719495792f);
const vector float B4 = (vector float) (170.354001821091472f);
const vector float B5 = (vector float) (53.992906912940207f);
const vector float B6 = (vector float) (10.479857114260399f);
vector float ivsigno;
vector float V;
vector float Z1_real;
vector float Z1_imag;
vector float Z2_real;
vector float Z2_imag;
vector float Z3_real;
vector float Z3_imag;
vector float Z4_real;
vector float Z4_imag;
vector float Z5_real;
vector float Z5_imag;
vector float Z6_real;
vector float Z6_imag;
vector float ZZ1_real;
vector float ZZ1_imag;
vector float ZZ2_real;
vector float ZZ2_imag;
vector float ZZ3_real;
vector float ZZ3_imag;
vector float ZZ4_real;
vector float ZZ4_imag;
vector float ZZ5_real;
vector float ZZ5_imag;
vector float ZZ6_real;
vector float ZZ6_imag;
vector float ZZ7_real;
vector float ZZ7_imag;
vector float division_factor;
vector float ZZZ_real;
vector bool int mask;
const vector float one = (vector float) (1.0f);
const vector float zero = (vector float) (-0.0f);
const vector float mone = (vector float) (-1.0f);
vector float damp;
vector float offs;
for(int i=0; i<N; i+=4){
damp = vec_ld(0,(float *) &damping[i]);
offs = vec_ld(0,(float *) &frequency_offset[i]);
mask = vec_cmplt(offs,zero);
ivsigno = vec_sel(mone, one, mask);
//ivsigno = (vector float) (1.0f);
V = vec_madd(ivsigno, offs, zero);
Z1_real = vec_madd(A6, damp, A5);
Z1_imag = vec_nmsub(A6, V, zero);
Z2_real = vec_add(vec_madd(Z1_real,damp,zero),vec_madd(Z1_imag,V,A4));
Z2_imag = vec_add(vec_nmsub(Z1_real,V,zero),vec_madd(Z1_imag,damp,zero));
Z3_real = vec_add(vec_madd(Z2_real,damp,zero),vec_madd(Z2_imag,V,A3));
Z3_imag = vec_add(vec_nmsub(Z2_real,V,zero),vec_madd(Z2_imag,damp,zero));
Z4_real = vec_add(vec_madd(Z3_real,damp,zero),vec_madd(Z3_imag,V,A2));
Z4_imag = vec_add(vec_nmsub(Z3_real,V,zero),vec_madd(Z3_imag,damp,zero));
Z5_real = vec_add(vec_madd(Z4_real,damp,zero),vec_madd(Z4_imag,V,A1));
Z5_imag = vec_add(vec_nmsub(Z4_real,V,zero),vec_madd(Z4_imag,damp,zero));
Z6_real = vec_add(vec_madd(Z5_real,damp,zero),vec_madd(Z5_imag,V,A0));
Z6_imag = vec_add(vec_nmsub(Z5_real,V,zero),vec_madd(Z5_imag,damp,zero));
ZZ1_real = vec_add(damp,B6);
ZZ1_imag = vec_madd(mone,V,zero);
ZZ2_real = vec_add(vec_madd(ZZ1_real,damp,zero),vec_madd(ZZ1_imag,V,B5));
ZZ2_imag = vec_add(vec_nmsub(ZZ1_real,V,zero),vec_madd(ZZ1_imag,damp,zero));
ZZ3_real = vec_add(vec_madd(ZZ2_real,damp,zero),vec_madd(ZZ2_imag,V,B4));
ZZ3_imag = vec_add(vec_nmsub(ZZ2_real,V,zero),vec_madd(ZZ2_imag,damp,zero));
ZZ4_real = vec_add(vec_madd(ZZ3_real,damp,zero),vec_madd(ZZ3_imag,V,B3));
ZZ4_imag = vec_add(vec_nmsub(ZZ3_real,V,zero),vec_madd(ZZ3_imag,damp,zero));
ZZ5_real = vec_add(vec_madd(ZZ4_real,damp,zero),vec_madd(ZZ4_imag,V,B2));
ZZ5_imag = vec_add(vec_nmsub(ZZ4_real,V,zero),vec_madd(ZZ4_imag,damp,zero));
ZZ6_real = vec_add(vec_madd(ZZ5_real,damp,zero),vec_madd(ZZ5_imag,V,B1));
ZZ6_imag = vec_add(vec_nmsub(ZZ5_real,V,zero),vec_madd(ZZ5_imag,damp,zero));
ZZ7_real = vec_add(vec_madd(ZZ6_real,damp,zero),vec_madd(ZZ6_imag,V,B0));
ZZ7_imag = vec_add(vec_nmsub(ZZ6_real,V,zero),vec_madd(ZZ6_imag,damp,zero));
division_factor = vec_div(one,vec_madd(ZZ7_real,ZZ7_real,vec_madd(ZZ7_imag,ZZ7_imag,zero)));
ZZZ_real = vec_madd(vec_madd(Z6_real,ZZ7_real,vec_madd(Z6_imag,ZZ7_imag,zero)),division_factor,zero);
vec_st(ZZZ_real,0,(float *)&voigt_value[i]);
}
}
void
nb_kernel400_ppc_altivec (int * p_nri,
int iinr[],
int jindex[],
int jjnr[],
int shift[],
float shiftvec[],
float fshift[],
int gid[],
float pos[],
float faction[],
float charge[],
float * p_facel,
float * p_krf,
float * p_crf,
float Vc[],
int type[],
int * p_ntype,
float vdwparam[],
float Vvdw[],
float * p_tabscale,
float VFtab[],
float invsqrta[],
float dvda[],
float * p_gbtabscale,
float GBtab[],
int * p_nthreads,
int * count,
void * mtx,
int * outeriter,
int * inneriter,
float * work)
{
vector float ix,iy,iz,shvec;
vector float vfacel,fs,nul;
vector float dx,dy,dz;
vector float vctot,qq,iq;
vector float fix,fiy,fiz;
vector float tmp1,tmp2,tmp3,tmp4;
vector float rinv,r,rsq,VVc,FFc;
vector float isai,isaj,isaprod,gbtsc,dvdasum,dvdaj,dvdatmp,gbscale,half;
int n,k,ii,is3,ii3,nj0,nj1;
int jnra,jnrb,jnrc,jnrd;
int j3a,j3b,j3c,j3d;
int nri, ntype, nouter, ninner;
#ifdef GMX_THREADS
int nn0, nn1;
#endif
nouter = 0;
ninner = 0;
nri = *p_nri;
ntype = *p_ntype;
nul=vec_zero();
half=vec_half();
vfacel=load_float_and_splat(p_facel);
gbtsc=load_float_and_splat(p_gbtabscale);
#ifdef GMX_THREADS
nthreads = *p_nthreads;
do {
gmx_thread_mutex_lock((gmx_thread_mutex_t *)mtx);
nn0 = *count;
nn1 = nn0+(nri-nn0)/(2*nthreads)+3;
*count = nn1;
gmx_thread_mutex_unlock((gmx_thread_mutex_t *)mtx);
if(nn1>nri) nn1=nri;
for(n=nn0; (n<nn1); n++) {
#if 0
} /* maintain correct indentation even with conditional left braces */
#endif
#else /* without gmx_threads */
for(n=0;n<nri;n++) {
#endif
is3 = 3*shift[n];
shvec = load_xyz(shiftvec+is3);
ii = iinr[n];
ii3 = 3*ii;
ix = load_xyz(pos+ii3);
vctot = nul;
dvdasum = nul;
fix = nul;
fiy = nul;
fiz = nul;
ix = vec_add(ix,shvec);
nj0 = jindex[n];
nj1 = jindex[n+1];
splat_xyz_to_vectors(ix,&ix,&iy,&iz);
iq = vec_madd(load_float_and_splat(charge+ii),vfacel,nul);
isai = load_float_and_splat(invsqrta+ii);
for(k=nj0; k<(nj1-3); k+=4) {
jnra = jjnr[k];
jnrb = jjnr[k+1];
jnrc = jjnr[k+2];
jnrd = jjnr[k+3];
j3a = 3*jnra;
j3b = 3*jnrb;
j3c = 3*jnrc;
j3d = 3*jnrd;
transpose_4_to_3(load_xyz(pos+j3a),
load_xyz(pos+j3b),
load_xyz(pos+j3c),
load_xyz(pos+j3d),&dx,&dy,&dz);
dx = vec_sub(ix,dx);
dy = vec_sub(iy,dy);
dz = vec_sub(iz,dz);
rsq = vec_madd(dx,dx,nul);
rsq = vec_madd(dy,dy,rsq);
rsq = vec_madd(dz,dz,rsq);
rinv = do_invsqrt(rsq);
r = vec_madd(rinv,rsq,nul);
/* load 1/sqrt(a2) and multiply with 1/sqrt(a1) */
isaj = load_4_float(invsqrta+jnra,invsqrta+jnrb,
invsqrta+jnrc,invsqrta+jnrd);
isaprod = vec_madd(isai,isaj,nul);
/* load 4 j charges and multiply by iq and 1/sqrt(a1*a2) */
qq = vec_madd(load_4_float(charge+jnra,charge+jnrb,
charge+jnrc,charge+jnrd),iq,nul);
qq = vec_madd(isaprod,qq,nul);
gbscale = vec_madd(isaprod,gbtsc,nul);
do_4_ctable_coul(GBtab,vec_madd(r,gbscale,nul),&VVc,&FFc);
dvdaj = load_4_float(dvda+jnra,dvda+jnrb,
dvda+jnrc,dvda+jnrd);
fs = vec_madd(qq,FFc,nul);
fs = vec_madd(fs,gbscale,nul);
vctot = vec_madd(qq,VVc,vctot);
dvdatmp = vec_madd(fs,r,nul);
dvdatmp = vec_madd(qq,VVc,dvdatmp);
dvdasum = vec_sub(dvdasum,dvdatmp);
dvdaj = vec_sub(dvdaj,dvdatmp);
store_4_float(dvdaj,dvda+jnra,dvda+jnrb,dvda+jnrc,dvda+jnrd);
fs = vec_nmsub(fs,rinv,nul);
fix = vec_madd(fs,dx,fix); /* +=fx */
fiy = vec_madd(fs,dy,fiy); /* +=fy */
fiz = vec_madd(fs,dz,fiz); /* +=fz */
dx = vec_nmsub(dx,fs,nul); /* -fx */
dy = vec_nmsub(dy,fs,nul); /* -fy */
dz = vec_nmsub(dz,fs,nul); /* -fz */
transpose_3_to_4(dx,dy,dz,&tmp1,&tmp2,&tmp3,&tmp4);
add_xyz_to_mem(faction+j3a,tmp1);
add_xyz_to_mem(faction+j3b,tmp2);
add_xyz_to_mem(faction+j3c,tmp3);
add_xyz_to_mem(faction+j3d,tmp4);
}
if(k<(nj1-1)) {
jnra = jjnr[k];
jnrb = jjnr[k+1];
j3a = 3*jnra;
j3b = 3*jnrb;
transpose_2_to_3(load_xyz(pos+j3a),
load_xyz(pos+j3b),&dx,&dy,&dz);
dx = vec_sub(ix,dx);
dy = vec_sub(iy,dy);
dz = vec_sub(iz,dz);
rsq = vec_madd(dx,dx,nul);
rsq = vec_madd(dy,dy,rsq);
rsq = vec_madd(dz,dz,rsq);
zero_highest_2_elements_in_vector(&rsq);
rinv = do_invsqrt(rsq);
zero_highest_2_elements_in_vector(&rinv);
r = vec_madd(rinv,rsq,nul);
/* load 1/sqrt(a2) and multiply with 1/sqrt(a1) */
isaj = load_2_float(invsqrta+jnra,invsqrta+jnrb);
isaprod = vec_madd(isai,isaj,nul);
/* load 2 j charges and multiply by iq and 1/sqrt(a1*a2) */
qq = vec_madd(load_2_float(charge+jnra,charge+jnrb),iq,nul);
qq = vec_madd(isaprod,qq,nul);
gbscale = vec_madd(isaprod,gbtsc,nul);
do_2_ctable_coul(GBtab,vec_madd(r,gbscale,nul),&VVc,&FFc);
dvdaj = load_2_float(dvda+jnra,dvda+jnrb);
fs = vec_madd(qq,FFc,nul);
fs = vec_madd(fs,gbscale,nul);
vctot = vec_madd(qq,VVc,vctot);
dvdatmp = vec_madd(fs,r,nul);
dvdatmp = vec_madd(qq,VVc,dvdatmp);
dvdasum = vec_sub(dvdasum,dvdatmp);
dvdaj = vec_sub(dvdaj,dvdatmp);
store_2_float(dvdaj,dvda+jnra,dvda+jnrb);
fs = vec_nmsub(fs,rinv,nul);
fix = vec_madd(fs,dx,fix); /* +=fx */
fiy = vec_madd(fs,dy,fiy); /* +=fy */
fiz = vec_madd(fs,dz,fiz); /* +=fz */
dx = vec_nmsub(dx,fs,nul); /* -fx */
dy = vec_nmsub(dy,fs,nul); /* -fy */
dz = vec_nmsub(dz,fs,nul); /* -fz */
transpose_3_to_2(dx,dy,dz,&tmp1,&tmp2);
add_xyz_to_mem(faction+j3a,tmp1);
add_xyz_to_mem(faction+j3b,tmp2);
k += 2;
}
if((nj1-nj0) & 0x1) {
jnra = jjnr[k];
j3a = 3*jnra;
transpose_1_to_3(load_xyz(pos+j3a),&dx,&dy,&dz);
dx = vec_sub(ix,dx);
dy = vec_sub(iy,dy);
dz = vec_sub(iz,dz);
rsq = vec_madd(dx,dx,nul);
rsq = vec_madd(dy,dy,rsq);
rsq = vec_madd(dz,dz,rsq);
zero_highest_3_elements_in_vector(&rsq);
rinv = do_invsqrt(rsq);
zero_highest_3_elements_in_vector(&rinv);
r = vec_madd(rinv,rsq,nul);
/* load 1/sqrt(a2) and multiply with 1/sqrt(a1) */
isaj = load_1_float(invsqrta+jnra);
isaprod = vec_madd(isai,isaj,nul);
/* load 1 j charge and multiply by iq and 1/sqrt(a1*a2) */
qq = vec_madd(load_1_float(charge+jnra),iq,nul);
qq = vec_madd(isaprod,qq,nul);
gbscale = vec_madd(isaprod,gbtsc,nul);
do_1_ctable_coul(GBtab,vec_madd(r,gbscale,nul),&VVc,&FFc);
dvdaj = load_1_float(dvda+jnra);
fs = vec_madd(qq,FFc,nul);
fs = vec_madd(fs,gbscale,nul);
vctot = vec_madd(qq,VVc,vctot);
dvdatmp = vec_madd(fs,r,nul);
dvdatmp = vec_madd(qq,VVc,dvdatmp);
dvdasum = vec_sub(dvdasum,dvdatmp);
dvdaj = vec_sub(dvdaj,dvdatmp);
store_1_float(dvdaj,dvda+jnra);
fs = vec_nmsub(fs,rinv,nul);
fix = vec_madd(fs,dx,fix); /* +=fx */
fiy = vec_madd(fs,dy,fiy); /* +=fy */
fiz = vec_madd(fs,dz,fiz); /* +=fz */
dx = vec_nmsub(dx,fs,nul); /* -fx */
dy = vec_nmsub(dy,fs,nul); /* -fy */
dz = vec_nmsub(dz,fs,nul); /* -fz */
transpose_3_to_1(dx,dy,dz,&tmp1);
add_xyz_to_mem(faction+j3a,tmp1);
}
/* update outer data */
transpose_3_to_4(fix,fiy,fiz,&tmp1,&tmp2,&tmp3,&tmp4);
tmp1 = vec_add(tmp1,tmp3);
tmp2 = vec_add(tmp2,tmp4);
tmp1 = vec_add(tmp1,tmp2);
add_xyz_to_mem(faction+ii3,tmp1);
add_xyz_to_mem(fshift+is3,tmp1);
add_vector_to_float(Vc+gid[n],vctot);
add_vector_to_float(dvda+ii,dvdasum);
ninner += nj1 - nj0;
}
#ifdef GMX_THREADS
nouter += nn1 - nn0;
} while (nn1<nri);
#else
nouter = nri;
#endif
*outeriter = nouter;
*inneriter = ninner;
}
