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

Fix handling of NAN and INF arguments

This commit is contained in:
Martin Kroeker 2024-01-07 15:29:42 +01:00 committed by GitHub
parent e48627c999
commit def4996170
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 32 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

61
kernel/x86_64/zscal.c
View File

@ -39,7 +39,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#endif
#include "common.h"
#include <float.h>
#if defined (SKYLAKEX) || defined (COOPERLAKE) || defined (SAPPHIRERAPIDS)
#include "zscal_microk_skylakex-2.c"
@ -222,12 +222,10 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i,
if ( da_r == 0.0 )
{
BLASLONG n1 = n & -2;
if ( da_i == 0.0 )
{
while(j < n1)
{
@ -253,7 +251,6 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i,
}
else
{
while(j < n1)
{
@ -356,49 +353,59 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i,
alpha[0] = da_r;
alpha[1] = da_i;
if ( da_r == 0.0 )
if ( da_i == 0 )
zscal_kernel_8_zero(n1 , alpha , x);
else
zscal_kernel_8_zero_r(n1 , alpha , x);
// zscal_kernel_8_zero_r(n1 , alpha , x);
zscal_kernel_8(n1 , alpha , x);
else
if ( da_i == 0 )
if ( da_i == 0 && da_r == da_r)
zscal_kernel_8_zero_i(n1 , alpha , x);
else
zscal_kernel_8(n1 , alpha , x);
}
i = n1 << 1;
j = n1;
}
if ( da_r == 0.0 )
if ( da_r == 0.0 || da_r != da_r )
{
if ( da_i == 0.0 )
{
FLOAT res=0.0;
if (da_r != da_r) res= da_r;
while(j < n)
{
x[i]=0.0;
x[i+1]=0.0;
x[i]=res;
x[i+1]=res;
i += 2 ;
j++;
}
}
else
else if (da_r < -FLT_MAX || da_r > FLT_MAX) {
while(j < n)
{
x[i]=SNAN;
x[i+1]=da_r;
i += 2 ;
j++;
}
} else
{
while(j < n)
{
temp0 = -da_i * x[i+1];
if (x[i] < -FLT_MAX || x[i] > FLT_MAX)
temp0 = SNAN;
x[i+1] = da_i * x[i];
x[i] = temp0;
if ( x[i] == x[i]) //preserve NaN
x[i] = temp0;
i += 2 ;
j++;
@ -409,12 +416,10 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i,
}
else
{
if ( da_i == 0.0 )
if (da_i == 0.0)
{
while(j < n)
{
while(j < n)
{
temp0 = da_r * x[i];
x[i+1] = da_r * x[i+1];
@ -422,15 +427,13 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i,
i += 2 ;
j++;
}
}
}
else
{
while(j < n)
{
temp0 = da_r * x[i] - da_i * x[i+1];
x[i+1] = da_r * x[i+1] + da_i * x[i];
x[i] = temp0;
@ -445,5 +448,3 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i,
return(0);
}

1
kernel/x86_64/zscal_sse2.S
View File

@ -82,6 +82,7 @@
pxor %xmm15, %xmm15
comisd %xmm0, %xmm15
jne .L100
jp .L100
comisd %xmm1, %xmm15
jne .L100