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);
}
}
void audio_convert_s16_to_float_altivec(float *out,
const int16_t *in, size_t samples, float gain)
{
const vector float gain_vec = vec_splats(gain);
const vector float zero_vec = vec_splats(0.0f);
// Unaligned loads/store is a bit expensive, so we optimize for the good path (very likely).
if (((uintptr_t)out & 15) + ((uintptr_t)in & 15) == 0)
{
size_t i;
for (i = 0; i + 8 <= samples; i += 8, in += 8, out += 8)
{
vector signed short input = vec_ld(0, in);
vector signed int hi = vec_unpackh(input);
vector signed int lo = vec_unpackl(input);
vector float out_hi = vec_madd(vec_ctf(hi, 15), gain_vec, zero_vec);
vector float out_lo = vec_madd(vec_ctf(lo, 15), gain_vec, zero_vec);
vec_st(out_hi, 0, out);
vec_st(out_lo, 16, out);
}
audio_convert_s16_to_float_C(out, in, samples - i, gain);
}
else
audio_convert_s16_to_float_C(out, in, samples, gain);
}
/* Using FChoose() here would mean allocating tmp space for 2M-1 paths;
* instead we use the fact that E(i) is itself the necessary normalization
* factor, and implement FChoose's algorithm here for an on-the-fly
* calculation.
*/
static inline int
select_e(ESL_RANDOMNESS *rng, const P7_OPROFILE *om, const P7_OMX *ox, int i, int *ret_k)
{
int Q = p7O_NQF(ox->M);
double sum = 0.0;
double roll = esl_random(rng);
double norm = 1.0 / ox->xmx[i*p7X_NXCELLS+p7X_E]; /* all M, D already scaled exactly the same */
vector float xEv = esl_vmx_set_float(norm);
vector float zerov = (vector float) vec_splat_u32(0);
union { vector float v; float p[4]; } u;
int q,r;
while (1) {
for (q = 0; q < Q; q++)
{
u.v = vec_madd(ox->dpf[i][q*3 + p7X_M], xEv, zerov);
for (r = 0; r < 4; r++) {
sum += u.p[r];
if (roll < sum) { *ret_k = r*Q + q + 1; return p7T_M;}
}
u.v = vec_madd(ox->dpf[i][q*3 + p7X_D], xEv, zerov);
for (r = 0; r < 4; r++) {
sum += u.p[r];
if (roll < sum) { *ret_k = r*Q + q + 1; return p7T_D;}
}
}
ESL_DASSERT1(sum > 0.99);
}
/*UNREACHED*/
ESL_EXCEPTION(-1, "unreached code was reached. universe collapses.");
}
/* D(i,k) is reached from M(i, k-1) or D(i,k-1). */
static inline int
select_d(ESL_RANDOMNESS *rng, const P7_OPROFILE *om, const P7_OMX *ox, int i, int k)
{
int Q = p7O_NQF(ox->M);
int q = (k-1) % Q; /* (q,r) is position of the current DP cell D(i,k) */
int r = (k-1) / Q;
vector float zerov;
vector float mpv, dpv;
vector float tmdv, tddv;
union { vector float v; float p[4]; } u;
float path[2];
int state[2] = { p7T_M, p7T_D };
zerov = (vector float) vec_splat_u32(0);
if (q > 0) {
mpv = ox->dpf[i][(q-1)*3 + p7X_M];
dpv = ox->dpf[i][(q-1)*3 + p7X_D];
tmdv = om->tfv[7*(q-1) + p7O_MD];
tddv = om->tfv[7*Q + (q-1)];
} else {
mpv = vec_sld(zerov, ox->dpf[i][(Q-1)*3 + p7X_M], 12);
dpv = vec_sld(zerov, ox->dpf[i][(Q-1)*3 + p7X_D], 12);
tmdv = vec_sld(zerov, om->tfv[7*(Q-1) + p7O_MD], 12);
tddv = vec_sld(zerov, om->tfv[8*Q-1], 12);
}
u.v = vec_madd(mpv, tmdv, zerov); path[0] = u.p[r];
u.v = vec_madd(dpv, tddv, zerov); path[1] = u.p[r];
esl_vec_FNorm(path, 2);
return state[esl_rnd_FChoose(rng, path, 2)];
}
/* 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));
}
/* M(i,k) is reached from B(i-1), M(i-1,k-1), D(i-1,k-1), or I(i-1,k-1). */
static inline int
select_m(ESL_RANDOMNESS *rng, const P7_OPROFILE *om, const P7_OMX *ox, int i, int k)
{
int Q = p7O_NQF(ox->M);
int q = (k-1) % Q; /* (q,r) is position of the current DP cell M(i,k) */
int r = (k-1) / Q;
vector float *tp = om->tfv + 7*q; /* *tp now at start of transitions to cur cell M(i,k) */
vector float xBv;
vector float zerov;
vector float mpv, dpv, ipv;
union { vector float v; float p[4]; } u;
float path[4];
int state[4] = { p7T_B, p7T_M, p7T_I, p7T_D };
xBv = esl_vmx_set_float(ox->xmx[(i-1)*p7X_NXCELLS+p7X_B]);
zerov = (vector float) vec_splat_u32(0);
if (q > 0) {
mpv = ox->dpf[i-1][(q-1)*3 + p7X_M];
dpv = ox->dpf[i-1][(q-1)*3 + p7X_D];
ipv = ox->dpf[i-1][(q-1)*3 + p7X_I];
} else {
mpv = vec_sld(zerov, ox->dpf[i-1][(Q-1)*3 + p7X_M], 12);
dpv = vec_sld(zerov, ox->dpf[i-1][(Q-1)*3 + p7X_D], 12);
ipv = vec_sld(zerov, ox->dpf[i-1][(Q-1)*3 + p7X_I], 12);
}
u.v = vec_madd(xBv, *tp, zerov); tp++; path[0] = u.p[r];
u.v = vec_madd(mpv, *tp, zerov); tp++; path[1] = u.p[r];
u.v = vec_madd(ipv, *tp, zerov); tp++; path[2] = u.p[r];
u.v = vec_madd(dpv, *tp, zerov); path[3] = u.p[r];
esl_vec_FNorm(path, 4);
return state[esl_rnd_FChoose(rng, path, 4)];
}
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);
}
}
inline v_float32x4 v_matmuladd(const v_float32x4& v, const v_float32x4& m0,
const v_float32x4& m1, const v_float32x4& m2,
const v_float32x4& a)
{
const vec_float4 v0 = vec_splat(v.val, 0);
const vec_float4 v1 = vec_splat(v.val, 1);
const vec_float4 v2 = vec_splat(v.val, 2);
return v_float32x4(vec_madd(v0, m0.val, vec_madd(v1, m1.val, vec_madd(v2, m2.val, a.val))));
}
inline v_float32x4 v_matmul(const v_float32x4& v, const v_float32x4& m0,
const v_float32x4& m1, const v_float32x4& m2,
const v_float32x4& m3)
{
const vec_float4 v0 = vec_splat(v.val, 0);
const vec_float4 v1 = vec_splat(v.val, 1);
const vec_float4 v2 = vec_splat(v.val, 2);
VSX_UNUSED(const vec_float4) v3 = vec_splat(v.val, 3);
return v_float32x4(vec_madd(v0, m0.val, vec_madd(v1, m1.val, vec_madd(v2, m2.val, vec_mul(v3, m3.val)))));
}
static void vector_fmul_altivec(float *dst, const float *src, int len)
{
int i;
vector float d0, d1, s, zero = (vector float)vec_splat_u32(0);
for(i=0; i<len-7; i+=8) {
d0 = vec_ld(0, dst+i);
s = vec_ld(0, src+i);
d1 = vec_ld(16, dst+i);
d0 = vec_madd(d0, s, zero);
d1 = vec_madd(d1, vec_ld(16,src+i), zero);
vec_st(d0, 0, dst+i);
vec_st(d1, 16, dst+i);
}
}
// exectue vectorized structured inner dot product
void dotprod_crcf_execute(dotprod_crcf _q,
float complex * _x,
float complex * _r)
{
int al; // input data alignment
vector float *ar,*d;
vector float s0,s1,s2,s3;
union { vector float v; float w[4];} s;
unsigned int nblocks;
ar = (vector float*)( (int)_x & ~15);
al = ((int)_x & 15)/sizeof(float);
d = (vector float*)_q->h[al];
// number of blocks doubles because of complex type
nblocks = (2*_q->n + al - 1)/4 + 1;
// split into four vectors each with four 32-bit
// partial sums. Effectively each loop iteration
// operates on 16 input samples at a time.
s0 = s1 = s2 = s3 = (vector float)(0);
while (nblocks >= 4) {
s0 = vec_madd(ar[nblocks-1],d[nblocks-1],s0);
s1 = vec_madd(ar[nblocks-2],d[nblocks-2],s1);
s2 = vec_madd(ar[nblocks-3],d[nblocks-3],s2);
s3 = vec_madd(ar[nblocks-4],d[nblocks-4],s3);
nblocks -= 4;
}
// fold the resulting partial sums into vector s0
s0 = vec_add(s0,s1); // s0 = s0+s1
s2 = vec_add(s2,s3); // s2 = s2+s3
s0 = vec_add(s0,s2); // s0 = s0+s2
// finish partial summing operations
while (nblocks-- > 0)
s0 = vec_madd(ar[nblocks],d[nblocks],s0);
// move the result into the union s (effetively,
// this loads the four 32-bit values in s0 into
// the array w).
s.v = vec_add(s0,(vector float)(0));
// sum the resulting array
//*_r = s.w[0] + s.w[1] + s.w[2] + s.w[3];
*_r = (s.w[0] + s.w[2]) + (s.w[1] + s.w[3]) * _Complex_I;
}
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
}
}
}
__SIMDi _SIMD_madd_epi32(__SIMDi a, __SIMDi b, __SIMDi c)
{
#ifdef USE_SSE
return _SIMD_add_epi32(_SIMD_mul_epi32(a,b),c);
#elif defined USE_AVX
return _m256_madd_ps(a,b);
#elif defined USE_IBM
return vec_madd(a,b);
#endif
}
__SIMDd _SIMD_madd_pd(__SIMDd a, __SIMDd b, __SIMDd c)
{
#ifdef USE_SSE
return _mm_add_pd(_mm_mul_pd(a,b),c);
#elif defined USE_AVX
return _m256_madd_ps(a,b);
#elif defined USE_IBM
return vec_madd(a,b);
#endif
}
/* I(i,k) is reached from M(i-1, k) or I(i-1,k). */
static inline int
select_i(ESL_RANDOMNESS *rng, const P7_OPROFILE *om, const P7_OMX *ox, int i, int k)
{
int Q = p7O_NQF(ox->M);
int q = (k-1) % Q; /* (q,r) is position of the current DP cell D(i,k) */
int r = (k-1) / Q;
vector float zerov;
vector float mpv = ox->dpf[i-1][q*3 + p7X_M];
vector float ipv = ox->dpf[i-1][q*3 + p7X_I];
vector float *tp = om->tfv + 7*q + p7O_MI;
union { vector float v; float p[4]; } u;
float path[2];
int state[2] = { p7T_M, p7T_I };
zerov = (vector float) vec_splat_u32(0);
u.v = vec_madd(mpv, *tp, zerov); tp++; path[0] = u.p[r];
u.v = vec_madd(ipv, *tp, zerov); path[1] = u.p[r];
esl_vec_FNorm(path, 2);
return state[esl_rnd_FChoose(rng, path, 2)];
}
static vector float appbones (State *s,
struct skin *skin,
vector float x,
vector float y,
vector float z,
vector float nx,
vector float ny,
vector float nz,
vector float *np)
{
int j;
int num_bones;
int bone_index;
struct abone *b;
vector float vz = (vector float) vec_splat_u32 (0);
vector float v, w, n;
vector unsigned char S = {0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,4<<3};
v = n = vz;
w = vec_ld (0, skin->weights);
num_bones = skin->boneinfo & 3;
bone_index = skin->boneinfo >> 2;
for (j = 0; j < num_bones; ++j) {
vector float t0, t1, t2, t3, t4, t5, r0, r1, r2, r3, vw;
b = &s->abones[bone_index & 0x3ff];
bone_index >>= 10;
vw = vec_splat (w, 0);
w = vec_slo (w, S);
r0 = vec_ld ( 0, b->cm);
r1 = vec_ld (16, b->cm);
r2 = vec_ld (32, b->cm);
r3 = vec_ld (48, b->cm);
t0 = vec_madd (r0, x, r3);
t1 = vec_madd (r1, y, t0);
t2 = vec_madd (r2, z, t1);
v = vec_madd (t2, vw, v);
t3 = vec_madd (r0, nx, vz);
t4 = vec_madd (r1, ny, t3);
t5 = vec_madd (r2, nz, t4);
n = vec_madd (t5, vw, n);
}
*np = n;
return v;
}
static void int32_to_float_fmul_scalar_altivec(float *dst, const int *src, float mul, int len)
{
union {
vector float v;
float s[4];
} mul_u;
int i;
vector float src1, src2, dst1, dst2, mul_v, zero;
zero = (vector float)vec_splat_u32(0);
mul_u.s[0] = mul;
mul_v = vec_splat(mul_u.v, 0);
for(i=0; i<len; i+=8) {
src1 = vec_ctf(vec_ld(0, src+i), 0);
src2 = vec_ctf(vec_ld(16, src+i), 0);
dst1 = vec_madd(src1, mul_v, zero);
dst2 = vec_madd(src2, mul_v, zero);
vec_st(dst1, 0, dst+i);
vec_st(dst2, 16, dst+i);
}
}
void assign_add_mul_r_32(spinor32 * const R, spinor32 * const S, const float c, const int N) {
#ifdef TM_USE_OMP
#pragma omp parallel
{
#endif
vector4double x0, x1, x2, x3, x4, x5, y0, y1, y2, y3, y4, y5;
vector4double z0, z1, z2, z3, z4, z5, k;
float *s, *r;
float ALIGN32 _c;
_c = c;
__prefetch_by_load(S);
__prefetch_by_load(R);
k = vec_splats((double)_c);
__alignx(16, s);
__alignx(16, r);
__alignx(16, S);
__alignx(16, R);
#ifdef TM_USE_OMP
#pragma omp for
#else
#pragma unroll(2)
#endif
for(int i = 0; i < N; i++) {
s=(float*)((spinor32 *) S + i);
r=(float*)((spinor32 *) R + i);
__prefetch_by_load(S + i + 1);
__prefetch_by_stream(1, R + i + 1);
x0 = vec_ld(0, r);
x1 = vec_ld(0, r+4);
x2 = vec_ld(0, r+8);
x3 = vec_ld(0, r+12);
x4 = vec_ld(0, r+16);
x5 = vec_ld(0, r+20);
y0 = vec_ld(0, s);
y1 = vec_ld(0, s+4);
y2 = vec_ld(0, s+8);
y3 = vec_ld(0, s+12);
y4 = vec_ld(0, s+16);
y5 = vec_ld(0, s+20);
z0 = vec_madd(k, y0, x0);
z1 = vec_madd(k, y1, x1);
z2 = vec_madd(k, y2, x2);
z3 = vec_madd(k, y3, x3);
z4 = vec_madd(k, y4, x4);
z5 = vec_madd(k, y5, x5);
vec_st(z0, 0, r);
vec_st(z1, 0, r+4);
vec_st(z2, 0, r+8);
vec_st(z3, 0, r+12);
vec_st(z4, 0, r+16);
vec_st(z5, 0, r+20);
}
#ifdef TM_USE_OMP
} /* OpenMP closing brace */
#endif
return;
}
__vector float func(__vector float vx)
{
__vector float vy;
__vector float va = (__vector float) { 5.0f, 5.0f, 5.0f, 5.0f };
__vector float vb = (__vector float) { -16.0f, -16.0f, -16.0f, -16.0f };
__vector float vc = (__vector float) { -36.0f, -36.0f, -36.0f, -36.0f };
__vector float vd = (__vector float) { 64.0f, 64.0f, 64.0f, 64.0f };
__vector float ve = (__vector float) { 192.0f, 192.0f, 192.0f, 192.0f };
vy = vec_madd(va, vx, vb);
vy = vec_madd(vy, vx, vc);
vy = vec_madd(vy, vx, vd);
vy = vec_madd(vy, vx, ve);
return vy;
}
float calc_integral(float start, float end, float delta)
{
int i;
float *sum;
__vector float vx = (__vector float) { start+delta*0, start+delta*1,
start+delta*2, start+delta*3 };
__vector float vsum = (__vector float) { 0.0f, 0.0f, 0.0f, 0.0f };
__vector float vdelta = (__vector float) { delta, delta, delta, delta };
__vector float vstep = (__vector float) { 4.0f, 4.0f, 4.0f, 4.0f };
for (i = 0; i < (end-start)/delta; i += 4) {
vsum = vec_madd(func(vx), vdelta, vsum);
vx = vec_madd(vdelta, vstep, vx);
}
sum = (float *) &vsum;
return (sum[0] + sum[1] + sum[2] + sum[3]);
}
int main(int argc, char **argv)
{
float start = 0.0f;
float end = 4.0f;
float delta = 0.00001f;
float result;
printf("start = %f, end = %f\n", start, end);
result = calc_integral(start, end, delta);
printf("result = %f\n", result);
return 0;
}
void CDSPToolsOSX::Mul32(tfloat32* pDest, tfloat32 Src)
{
#ifdef _Mac_PowerPC
vector float vSrc = (vector float)(Src, Src, Src, Src);
vector float vDest = vec_ldl(0, pDest);
vDest = vec_madd(vDest, vSrc, (vector float)(0, 0, 0, 0));
vec_st(vDest, 0, pDest);
vDest = vec_ldl(4 * 4, pDest);
vDest = vec_madd(vDest, vSrc, (vector float)(0, 0, 0, 0));
vec_st(vDest, 4 * 4, pDest);
vDest = vec_ldl(8 * 4, pDest);
vDest = vec_madd(vDest, vSrc, (vector float)(0, 0, 0, 0));
vec_st(vDest, 8 * 4, pDest);
vDest = vec_ldl(12 * 4, pDest);
vDest = vec_madd(vDest, vSrc, (vector float)(0, 0, 0, 0));
vec_st(vDest, 12 * 4, pDest);
vDest = vec_ldl(16 * 4, pDest);
vDest = vec_madd(vDest, vSrc, (vector float)(0, 0, 0, 0));
vec_st(vDest, 16 * 4, pDest);
vDest = vec_ldl(20 * 4, pDest);
vDest = vec_madd(vDest, vSrc, (vector float)(0, 0, 0, 0));
vec_st(vDest, 20 * 4, pDest);
vDest = vec_ldl(24 * 4, pDest);
vDest = vec_madd(vDest, vSrc, (vector float)(0, 0, 0, 0));
vec_st(vDest, 24 * 4, pDest);
vDest = vec_ldl(28 * 4, pDest);
vDest = vec_madd(vDest, vSrc, (vector float)(0, 0, 0, 0));
vec_st(vDest, 28 * 4, pDest);
#else // _Mac_PowerPC
CDSPTools::Mul32(pDest, Src);
#endif // _Mac_PowerPC
}
/* *************************************************************************
NAME: test_madd
USAGE:
test_madd();
returns: void
DESCRIPTION:
see how the combination multiply/add operation works
REFERENCES:
Ian Ollmann's Altivec Tutorial
LIMITATIONS:
GLOBAL VARIABLES:
accessed: none
modified: none
FUNCTIONS CALLED:
fprintf
vec_madd - multiply two float vectors and add to the sum a float value
all in one operation
REVISION HISTORY:
STR Description of Revision Author
27-Feb-11 initial coding kaj
************************************************************************* */
void test_madd(void)
{
vector float floatVector1 =
{ -17.777777, 0.0, 37.777777, 100.0};
vector float addVector =
{ 32.0, 32.0, 32.0, 32.0};
vector float coeffVector =
{ 1.8, 1.8, 1.8, 1.8};
vector float fahrenheitVector;
short printfloat[FLOAT_ARRAYSIZE] __attribute__ ((aligned (16)));
/* print vectors performing madd on */
fprintf(stderr,"-----------------------------------------------------------\n\n");
printVecFloats("vec_madd input vector 1", floatVector1,FLOAT_ARRAYSIZE);
printVecFloats("vec_madd input vector to add", addVector,FLOAT_ARRAYSIZE);
printVecFloats("vec_madd coeffvector to multiply", coeffVector,FLOAT_ARRAYSIZE);
/* calculate */
fahrenheitVector = vec_madd(floatVector1,coeffVector,addVector);
printVecFloats("vec_madd vector (Input*9/5+32)", fahrenheitVector,FLOAT_ARRAYSIZE);
} /* test_madd */
static void vector_fmul_add_altivec(float *dst, const float *src0,
const float *src1, const float *src2,
int len)
{
int i;
vector float d, s0, s1, s2, t0, t1, edges;
vector unsigned char align = vec_lvsr(0,dst),
mask = vec_lvsl(0, dst);
for (i=0; i<len-3; i+=4) {
t0 = vec_ld(0, dst+i);
t1 = vec_ld(15, dst+i);
s0 = vec_ld(0, src0+i);
s1 = vec_ld(0, src1+i);
s2 = vec_ld(0, src2+i);
edges = vec_perm(t1 ,t0, mask);
d = vec_madd(s0,s1,s2);
t1 = vec_perm(d, edges, align);
t0 = vec_perm(edges, d, align);
vec_st(t1, 15, dst+i);
vec_st(t0, 0, dst+i);
}
}
int main(int argc, char **argv)
{
int i;
__vector float *vin = (__vector float *) in;
__vector float *vout = (__vector float *) out;
__vector float vin_negative;
__vector unsigned int vpat;
__vector float vzero = (__vector float) { 0.0f, 0.0f, 0.0f, 0.0f };
__vector float vminus = (__vector float) { -1.0f, -1.0f, -1.0f, -1.0f };
for (i = 0; i < SIZE/4; i++) {
vpat = vec_cmpgt(vin[i], vzero);
vin_negative = vec_madd(vin[i], vminus, vzero);
vout[i] = vec_sel(vin_negative, vin[i], vpat);
}
for (i = 0; i < SIZE; i++) {
printf("out[%02d]=%0.0f\n", i, out[i]);
}
return 0;
}
static void ProjectDlightTexture_altivec( void ) {
int i, l;
vec_t origin0, origin1, origin2;
float texCoords0, texCoords1;
vector float floatColorVec0, floatColorVec1;
vector float modulateVec, colorVec, zero;
vector short colorShort;
vector signed int colorInt;
vector unsigned char floatColorVecPerm, modulatePerm, colorChar;
vector unsigned char vSel = VECCONST_UINT8(0x00, 0x00, 0x00, 0xff,
0x00, 0x00, 0x00, 0xff,
0x00, 0x00, 0x00, 0xff,
0x00, 0x00, 0x00, 0xff);
float *texCoords;
byte *colors;
byte clipBits[SHADER_MAX_VERTEXES];
float texCoordsArray[SHADER_MAX_VERTEXES][2];
byte colorArray[SHADER_MAX_VERTEXES][4];
unsigned hitIndexes[SHADER_MAX_INDEXES];
int numIndexes;
float scale;
float radius;
vec3_t floatColor;
float modulate = 0.0f;
if ( !backEnd.refdef.num_dlights ) {
return;
}
// There has to be a better way to do this so that floatColor
// and/or modulate are already 16-byte aligned.
floatColorVecPerm = vec_lvsl(0,(float *)floatColor);
modulatePerm = vec_lvsl(0,(float *)&modulate);
modulatePerm = (vector unsigned char)vec_splat((vector unsigned int)modulatePerm,0);
zero = (vector float)vec_splat_s8(0);
for ( l = 0 ; l < backEnd.refdef.num_dlights ; l++ ) {
dlight_t *dl;
if ( !( tess.dlightBits & ( 1 << l ) ) ) {
continue; // this surface definately doesn't have any of this light
}
texCoords = texCoordsArray[0];
colors = colorArray[0];
dl = &backEnd.refdef.dlights[l];
origin0 = dl->transformed[0];
origin1 = dl->transformed[1];
origin2 = dl->transformed[2];
radius = dl->radius;
scale = 1.0f / radius;
if(r_greyscale->integer)
{
float luminance;
luminance = LUMA(dl->color[0], dl->color[1], dl->color[2]) * 255.0f;
floatColor[0] = floatColor[1] = floatColor[2] = luminance;
}
else if(r_greyscale->value)
{
float luminance;
luminance = LUMA(dl->color[0], dl->color[1], dl->color[2]) * 255.0f;
floatColor[0] = LERP(dl->color[0] * 255.0f, luminance, r_greyscale->value);
floatColor[1] = LERP(dl->color[1] * 255.0f, luminance, r_greyscale->value);
floatColor[2] = LERP(dl->color[2] * 255.0f, luminance, r_greyscale->value);
}
else
{
floatColor[0] = dl->color[0] * 255.0f;
floatColor[1] = dl->color[1] * 255.0f;
floatColor[2] = dl->color[2] * 255.0f;
}
floatColorVec0 = vec_ld(0, floatColor);
floatColorVec1 = vec_ld(11, floatColor);
floatColorVec0 = vec_perm(floatColorVec0,floatColorVec0,floatColorVecPerm);
for ( i = 0 ; i < tess.numVertexes ; i++, texCoords += 2, colors += 4 ) {
int clip = 0;
vec_t dist0, dist1, dist2;
dist0 = origin0 - tess.xyz[i][0];
dist1 = origin1 - tess.xyz[i][1];
dist2 = origin2 - tess.xyz[i][2];
backEnd.pc.c_dlightVertexes++;
texCoords0 = 0.5f + dist0 * scale;
texCoords1 = 0.5f + dist1 * scale;
if( !r_dlightBacks->integer &&
// dist . tess.normal[i]
( dist0 * tess.normal[i][0] +
dist1 * tess.normal[i][1] +
dist2 * tess.normal[i][2] ) < 0.0f ) {
clip = 63;
} else {
if ( texCoords0 < 0.0f ) {
clip |= 1;
} else if ( texCoords0 > 1.0f ) {
clip |= 2;
}
if ( texCoords1 < 0.0f ) {
clip |= 4;
} else if ( texCoords1 > 1.0f ) {
clip |= 8;
}
texCoords[0] = texCoords0;
texCoords[1] = texCoords1;
// modulate the strength based on the height and color
if ( dist2 > radius ) {
clip |= 16;
modulate = 0.0f;
} else if ( dist2 < -radius ) {
clip |= 32;
modulate = 0.0f;
} else {
dist2 = Q_fabs(dist2);
if ( dist2 < radius * 0.5f ) {
modulate = 1.0f;
} else {
modulate = 2.0f * (radius - dist2) * scale;
}
}
}
clipBits[i] = clip;
modulateVec = vec_ld(0,(float *)&modulate);
modulateVec = vec_perm(modulateVec,modulateVec,modulatePerm);
colorVec = vec_madd(floatColorVec0,modulateVec,zero);
colorInt = vec_cts(colorVec,0); // RGBx
colorShort = vec_pack(colorInt,colorInt); // RGBxRGBx
colorChar = vec_packsu(colorShort,colorShort); // RGBxRGBxRGBxRGBx
colorChar = vec_sel(colorChar,vSel,vSel); // RGBARGBARGBARGBA replace alpha with 255
vec_ste((vector unsigned int)colorChar,0,(unsigned int *)colors); // store color
}
// build a list of triangles that need light
numIndexes = 0;
for ( i = 0 ; i < tess.numIndexes ; i += 3 ) {
int a, b, c;
a = tess.indexes[i];
b = tess.indexes[i+1];
c = tess.indexes[i+2];
if ( clipBits[a] & clipBits[b] & clipBits[c] ) {
continue; // not lighted
}
hitIndexes[numIndexes] = a;
hitIndexes[numIndexes+1] = b;
hitIndexes[numIndexes+2] = c;
numIndexes += 3;
}
if ( !numIndexes ) {
continue;
}
qglEnableClientState( GL_TEXTURE_COORD_ARRAY );
qglTexCoordPointer( 2, GL_FLOAT, 0, texCoordsArray[0] );
qglEnableClientState( GL_COLOR_ARRAY );
qglColorPointer( 4, GL_UNSIGNED_BYTE, 0, colorArray );
GL_Bind( tr.dlightImage );
// include GLS_DEPTHFUNC_EQUAL so alpha tested surfaces don't add light
// where they aren't rendered
if ( dl->additive ) {
GL_State( GLS_SRCBLEND_ONE | GLS_DSTBLEND_ONE | GLS_DEPTHFUNC_EQUAL );
}
else {
GL_State( GLS_SRCBLEND_DST_COLOR | GLS_DSTBLEND_ONE | GLS_DEPTHFUNC_EQUAL );
}
R_DrawElements( numIndexes, hitIndexes );
backEnd.pc.c_totalIndexes += numIndexes;
backEnd.pc.c_dlightIndexes += numIndexes;
}
}
/* two dimensional discrete cosine transform */
void ff_fdct_altivec(int16_t *block)
{
vector signed short *bp;
const vector float *cp = fdctconsts;
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)));
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_FMERGEL(b00, b20);
x1 = VEC_FMERGEH(b00, b20);
x2 = VEC_FMERGEL(b10, b30);
x3 = VEC_FMERGEH(b10, b30);
b00 = VEC_FMERGEH(x1, x3);
b10 = VEC_FMERGEL(x1, x3);
b20 = VEC_FMERGEH(x0, x2);
b30 = VEC_FMERGEL(x0, x2);
x4 = VEC_FMERGEL(b41, b61);
x5 = VEC_FMERGEH(b41, b61);
x6 = VEC_FMERGEL(b51, b71);
x7 = VEC_FMERGEH(b51, b71);
b41 = VEC_FMERGEH(x5, x7);
b51 = VEC_FMERGEL(x5, x7);
b61 = VEC_FMERGEH(x4, x6);
b71 = VEC_FMERGEL(x4, x6);
x0 = VEC_FMERGEL(b01, b21);
x1 = VEC_FMERGEH(b01, b21);
x2 = VEC_FMERGEL(b11, b31);
x3 = VEC_FMERGEH(b11, b31);
x4 = VEC_FMERGEL(b40, b60);
x5 = VEC_FMERGEH(b40, b60);
x6 = VEC_FMERGEL(b50, b70);
x7 = VEC_FMERGEH(b50, b70);
b40 = VEC_FMERGEH(x1, x3);
b50 = VEC_FMERGEL(x1, x3);
b60 = VEC_FMERGEH(x0, x2);
b70 = VEC_FMERGEL(x0, x2);
b01 = VEC_FMERGEH(x5, x7);
b11 = VEC_FMERGEL(x5, x7);
b21 = VEC_FMERGEH(x4, x6);
b31 = VEC_FMERGEL(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
/* }}} */
}
void cfft2(unsigned int n,float x[][2],float y[][2],float w[][2], float sign)
{
/*
altivec version of cfft2 from Petersen and Arbenz book, "Intro.
to Parallel Computing", Oxford Univ. Press, 2003, Section 3.6
wpp 14. Dec. 2003
*/
int jb,jc,jd,jw,k,k2,k4,lj,m,j,mj,mj2,pass,tgle;
float rp,up,wr[4] __attribute((aligned(16)));
float wu[4] __attribute((aligned(16)));
float *a,*b,*c,*d;
const vector float vminus = (vector float) {
-0.,0.,-0.,0.
};
const vector float vzero = (vector float) {
0.,0.,0.,0.
};
const vector unsigned char pv3201 =
(vector unsigned char) {
4,5,6,7,0,1,2,3,12,13,14,15,8,9,10,11
};
vector float V0,V1,V2,V3,V4,V5,V6,V7;
vector float V8,V9,V10,V11,V12,V13,V14,V15;
if(n<=1) {
y[0][0] = x[0][0];
y[0][1] = x[0][1];
return;
}
m = (int) (log((float) n)/log(1.99));
mj = 1;
mj2 = 2;
lj = n/2;
/* first pass thru data: x -> y */
for(j=0; j<lj; j++) {
jb = n/2+j;
jc = j*mj2;
jd = jc + 1;
rp = w[j][0];
up = w[j][1];
if(sign<0.0) up = -up;
y[jd][0] = rp*(x[j][0] - x[jb][0]) - up*(x[j][1] - x[jb][1]);
y[jd][1] = up*(x[j][0] - x[jb][0]) + rp*(x[j][1] - x[jb][1]);
y[jc][0] = x[j][0] + x[jb][0];
y[jc][1] = x[j][1] + x[jb][1];
}
if(n==2) return;
/* next pass is mj = 2 */
mj = 2;
mj2 = 4;
lj = n/4;
a = (float *)&y[0][0];
b = (float *)&y[n/2][0];
c = (float *)&x[0][0];
d = (float *)&x[mj][0];
if(n==4) {
c = (float *)&y[0][0];
d = (float *)&y[mj][0];
}
for(j=0; j<lj; j++) {
jw = j*mj;
jc = j*mj2;
jd = 2*jc;
rp = w[jw][0];
up = w[jw][1];
if(sign<0.0) up = -up;
wr[0] = rp;
wr[1] = rp;
wr[2] = rp;
wr[3] = rp;
wu[0] = up;
wu[1] = up;
wu[2] = up;
wu[3] = up;
V6 = vec_ld(0,wr);
V7 = vec_ld(0,wu);
V7 = vec_xor(V7,vminus);
V0 = vec_ld(0,(vector float *) (a+jc));
V1 = vec_ld(0,(vector float *) (b+jc));
V2 = vec_add(V0,V1); /* a + b */
vec_st(V2,0,(vector float *) (c+jd)); /* store c */
V3 = vec_sub(V0,V1); /* a - b */
V4 = vec_perm(V3,V3,pv3201);
V0 = vec_madd(V6,V3,vzero);
V1 = vec_madd(V7,V4,vzero);
V2 = vec_add(V0,V1); /* w*(a - b) */
vec_st(V2,0,(vector float*) (d+jd)); /* store d */
}
if(n==4) return;
mj *= 2;
mj2 = 2*mj;
lj = n/mj2;
tgle = 0;
for(pass=2; pass<m-1; pass++) {
if(tgle) {
a = (float *)&y[0][0];
b = (float *)&y[n/2][0];
c = (float *)&x[0][0];
d = (float *)&x[mj][0];
tgle = 0;
} else {
a = (float *)&x[0][0];
b = (float *)&x[n/2][0];
c = (float *)&y[0][0];
d = (float *)&y[mj][0];
tgle = 1;
}
for(j=0; j<lj; j++) {
jw = j*mj;
jc = j*mj2;
jd = 2*jc;
rp = w[jw][0];
up = w[jw][1];
if(sign<0.0) up = -up;
wr[0] = rp;
wr[1] = rp;
wr[2] = rp;
wr[3] = rp;
wu[0] = up;
wu[1] = up;
wu[2] = up;
wu[3] = up;
V6 = vec_ld(0,wr);
V7 = vec_ld(0,wu);
V7 = vec_xor(V7,vminus);
for(k=0; k<mj; k+=4) {
k2 = 2*k;
k4 = k2+4;
V0 = vec_ld(0,(vector float *) (a+jc+k2));
V1 = vec_ld(0,(vector float *) (b+jc+k2));
V2 = vec_add(V0,V1); /* a + b */
vec_st(V2,0,(vector float*) (c+jd+k2)); /* store c */
V3 = vec_sub(V0,V1); /* a - b */
V4 = vec_perm(V3,V3,pv3201);
V0 = vec_madd(V6,V3,vzero);
V1 = vec_madd(V7,V4,vzero);
V2 = vec_add(V0,V1); /* w*(a - b) */
vec_st(V2,0,(vector float *) (d+jd+k2)); /* store d */
V8 = vec_ld(0,(vector float *) (a+jc+k4));
V9 = vec_ld(0,(vector float *) (b+jc+k4));
V10 = vec_add(V8,V9); /* a + b */
vec_st(V10,0,(vector float *) (c+jd+k4)); /* store c */
V11 = vec_sub(V8,V9); /* a - b */
V12 = vec_perm(V11,V11,pv3201);
V8 = vec_madd(V6,V11,vzero);
V9 = vec_madd(V7,V12,vzero);
V10 = vec_add(V8,V9); /* w*(a - b) */
vec_st(V10,0,(vector float *) (d+jd+k4)); /* store d */
}
}
mj *= 2;
mj2 = 2*mj;
lj = n/mj2;
}
/* last pass thru data: in-place if previous in y */
c = (float *)&y[0][0];
d = (float *)&y[n/2][0];
if(tgle) {
a = (float *)&y[0][0];
b = (float *)&y[n/2][0];
} else {
a = (float *)&x[0][0];
b = (float *)&x[n/2][0];
}
for(k=0; k<(n/2); k+=4) {
k2 = 2*k;
k4 = k2+4;
V0 = vec_ld(0,(vector float *) (a+k2));
V1 = vec_ld(0,(vector float *) (b+k2));
V2 = vec_add(V0,V1); /* a + b */
vec_st(V2,0,(vector float*) (c+k2)); /* store c */
V3 = vec_sub(V0,V1); /* a - b */
vec_st(V3,0,(vector float *) (d+k2)); /* store d */
V4 = vec_ld(0,(vector float *) (a+k4));
V5 = vec_ld(0,(vector float *) (b+k4));
V6 = vec_add(V4,V5); /* a + b */
vec_st(V6,0,(vector float *) (c+k4)); /* store c */
V7 = vec_sub(V4,V5); /* a - b */
vec_st(V7,0,(vector float *) (d+k4)); /* store d */
}
}
// LLVM LOCAL begin
// Implementations of sin() and cos() may vary slightly in the accuracy of
// their results, typically only in the least significant bit. Round to make
// the results consistent across platforms.
typedef union {
double d;
unsigned long long ll;
} dbl_ll_union;
static double LLVMsin(double d) {
dbl_ll_union u;
u.d = sin(d);
u.ll = (u.ll + 1) & ~1ULL;
return u.d;
}
static double LLVMcos(double d) {
dbl_ll_union u;
u.d = cos(d);
u.ll = (u.ll + 1) & ~1ULL;
return u.d;
}
// LLVM LOCAL end
void cffti(int n, float w[][2])
{
/* initialization routine for cfft2: computes
cos(twopi*k),sin(twopi*k) for k=0..n/2-1
- the "twiddle factors" for a binary radix FFT */
int i,n2;
float aw,arg,pi;
pi = 3.141592653589793;
n2 = n/2;
aw = 2.0*pi/((float)n);
for(i=0; i<n2; i++) {
arg = aw*((float)i);
w[i][0] = LLVMcos(arg);
w[i][1] = LLVMsin(arg);
}
}
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;
}
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);
vkrf=load_float_and_splat(p_krf);
vcrf=load_float_and_splat(p_crf);
ii = iinr[0];
iqO = vec_madd(load_float_and_splat(charge+ii),vfacel,nul);
iqH = vec_madd(load_float_and_splat(charge+ii+1),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 */
