floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

x86: Fixed numpy CI failure when the target is ZEN.

This commit is contained in:
gxw 2024-07-10 15:11:12 +08:00
parent e1eef56e05
commit f3cebb3ca3
3 changed files with 279 additions and 296 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
interface/scal.c
View File

@ -85,7 +85,7 @@ void CNAME(blasint n, FLOAT alpha, FLOAT *x, blasint incx){
if (nthreads == 1) { if (nthreads == 1) {
#endif #endif
SCAL_K(n, 0, 0, alpha, x, incx, NULL, 0, NULL, 0); SCAL_K(n, 0, 0, alpha, x, incx, NULL, 0, NULL, 1);
#ifdef SMP #ifdef SMP
} else { } else {
@ -102,7 +102,7 @@ void CNAME(blasint n, FLOAT alpha, FLOAT *x, blasint incx){
#else #else
&alpha, &alpha,
#endif #endif
x, incx, NULL, 0, NULL, 0, (int (*)(void))SCAL_K, nthreads); x, incx, NULL, 0, NULL, 1, (int (*)(void))SCAL_K, nthreads);
} }
#endif #endif

142
kernel/x86_64/dscal.c
View File

@ -150,50 +150,17 @@ static void dscal_kernel_inc_8(BLASLONG n, FLOAT *alpha, FLOAT *x, BLASLONG inc_
"%xmm12", "%xmm13", "%xmm14", "%xmm15", "%xmm12", "%xmm13", "%xmm14", "%xmm15",
"memory" "memory"
); );
} }
int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2) int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2)
{ {
BLASLONG i=0,j=0; BLASLONG i = 0, j = 0;
// Resolved issue 4728 when the caller is dscal
if (dummy2 == 1 && da == 0.0)
{
if ( inc_x != 1 ) if ( inc_x != 1 )
{ {
if ( da == 0.0 )
{
BLASLONG n1 = n & -2;
while(j < n1)
{
if (isinf(x[i])||isnan(x[i]))
x[i]=NAN;
else x[i]=0.0;
if (isinf(x[i+inc_x])||isnan(x[i+inc_x]))
x[i+inc_x]=NAN;
else x[i+inc_x]=0.0;
i += 2*inc_x ;
j+=2;
}
while(j < n)
{
if (isinf(x[i])||isnan(x[i]))
x[i]=NAN;
else x[i]=0.0;
i += inc_x ;
j++;
}
}
else
{
BLASLONG n1 = n & -8; BLASLONG n1 = n & -8;
if ( n1 > 0 ) if ( n1 > 0 )
{ {
@ -201,55 +168,78 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
i = n1 * inc_x; i = n1 * inc_x;
j = n1; j = n1;
} }
while(j < n) while(j < n)
{ {
x[i] *= da; x[i] *= da;
i += inc_x ; i += inc_x ;
j++; j++;
} }
}
return(0);
}
BLASLONG n1 = n & -8;
if ( n1 > 0 )
{
// if ( da == 0.0 )
// dscal_kernel_8_zero(n1 , &da , x);
// else
dscal_kernel_8(n1 , &da , x);
}
if ( da == 0.0 )
{
for ( i=n1 ; i<n; i++ )
{
if(isinf(x[i])||isnan(x[i]))
x[i]=NAN;
else x[i] = 0.0;
}
}
else if (isinf(da)){
for ( i=n1 ; i<n; i++)
if (x[i]==0.) x[i]=NAN;
else x[i] *=da;
} }
else else
{ {
BLASLONG n1 = n & -8;
for ( i=n1 ; i<n; i++ ) if ( n1 > 0)
dscal_kernel_8(n1 , &da , x);
for ( i = n1 ; i < n; i++ )
x[i] *= da;
}
}
else
{ {
if(isinf(x[i])) if ( inc_x != 1 )
x[i]=NAN; {
else x[i] *= da; if( da == 0.0)
{
BLASLONG n1 = n & -2;
while(j < n1)
{
x[i] = 0.0;
x[i+inc_x] = 0.0;
i += 2 * inc_x ;
j += 2;
}
while(j < n)
{
x[i] = 0.0;
i += inc_x ;
j++;
}
}
else
{
BLASLONG n1 = n & -8;
if ( n1 > 0 )
{
dscal_kernel_inc_8(n1, &da, x, inc_x);
i = n1 * inc_x;
j = n1;
}
while(j < n)
{
x[i] *= da;
i += inc_x ;
j++;
}
}
}
else
{
if ( da == 0.0 )
{
BLASLONG n1 = n & -8;
if ( n1 > 0)
dscal_kernel_8_zero(n1, &da, x);
for ( i = n1 ; i < n; i++ )
x[i] = 0.0;
}
else
{
BLASLONG n1 = n & -8;
if ( n1 > 0)
dscal_kernel_8(n1 , &da , x);
for ( i = n1 ; i < n; i++ )
x[i] *= da;
}
} }
} }
return(0);
} }

139
kernel/x86_64/sscal.c
View File

@ -112,41 +112,13 @@ static void sscal_kernel_inc_8(BLASLONG n, FLOAT *alpha, FLOAT *x, BLASLONG inc_
int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2) int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *dummy, BLASLONG dummy2)
{ {
BLASLONG i=0,j=0; BLASLONG i = 0, j = 0;
// Resolved issue 4728 when the caller is sscal
if (dummy2 == 1 && da == 0.0)
{
if ( inc_x != 1 ) if ( inc_x != 1 )
{ {
if ( da == 0.0 )
{
BLASLONG n1 = n & -2;
while(j < n1)
{
if (isinf(x[i])||isnan(x[i]))
x[i]=NAN;
else x[i]=0.0;
if (isinf(x[i+inc_x])||isnan(x[i+inc_x]))
x[i+inc_x]=NAN;
else x[i+inc_x]=0.0;
i += 2*inc_x ;
j+=2;
}
while(j < n)
{
if (isinf(x[i])||isnan(x[i]))
x[i]=NAN;
else x[i]=0.0;
i += inc_x ;
j++;
}
}
else
{
#if 1
BLASLONG n1 = n & -8; BLASLONG n1 = n & -8;
if ( n1 > 0 ) if ( n1 > 0 )
{ {
@ -154,57 +126,78 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
i = n1 * inc_x; i = n1 * inc_x;
j = n1; j = n1;
} }
#endif
while(j < n) while(j < n)
{ {
x[i] *= da; x[i] *= da;
i += inc_x ; i += inc_x ;
j++; j++;
}
}
return(0);
}
BLASLONG n1 = n & -16;
if ( n1 > 0 )
{
//if ( da == 0.0 )
// sscal_kernel_16_zero(n1 , &da , x);
//else
sscal_kernel_16(n1 , &da , x);
}
if ( da == 0.0 )
{
for ( i=n1 ; i<n; i++ )
{
if (isinf(x[i])||isnan(x[i]))
x[i]=NAN;
else x[i]=0.0;
}
}
else if ( isinf(da) )
{
for ( i=n1 ; i<n; i++ )
{
if (x[i] == 0.0)
x[i]=NAN;
else x[i] *= da;
} }
} }
else else
{ {
BLASLONG n1 = n & -16;
for ( i=n1 ; i<n; i++ ) if ( n1 > 0)
sscal_kernel_16(n1 , &da , x);
for ( i = n1 ; i < n; i++ )
x[i] *= da;
}
}
else
{ {
if (isinf(x[i])) if ( inc_x != 1 )
x[i]=NAN; {
else x[i] *= da; if( da == 0.0)
{
BLASLONG n1 = n & -2;
while(j < n1)
{
x[i] = 0.0;
x[i+inc_x] = 0.0;
i += 2 * inc_x ;
j += 2;
}
while(j < n)
{
x[i] = 0.0;
i += inc_x ;
j++;
}
}
else
{
BLASLONG n1 = n & -8;
if ( n1 > 0 )
{
sscal_kernel_inc_8(n1, &da, x, inc_x);
i = n1 * inc_x;
j = n1;
}
while(j < n)
{
x[i] *= da;
i += inc_x ;
j++;
}
}
}
else
{
if ( da == 0.0 )
{
BLASLONG n1 = n & -16;
if ( n1 > 0)
sscal_kernel_16_zero(n1, &da, x);
for ( i = n1 ; i < n; i++ )
x[i] = 0.0;
}
else
{
BLASLONG n1 = n & -16;
if ( n1 > 0)
sscal_kernel_16(n1 , &da , x);
for ( i = n1 ; i < n; i++ )
x[i] *= da;
}
} }
} }
return(0);
} }