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

make NAN handling depend on DUMMY2 parameter

This commit is contained in:
Martin Kroeker 2024-07-21 13:42:47 +02:00 committed by GitHub
parent dfbc2348a8
commit 73f8866ffb
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 71 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

34
kernel/power/dscal.c
View File

@ -73,6 +73,15 @@ static void dscal_kernel_8_zero (BLASLONG n, FLOAT *x)
for( i=0; i<n; i+=8 ) for( i=0; i<n; i+=8 )
{ {
x[0] = alpha;
x[1] = alpha;
x[2] = alpha;
x[3] = alpha;
x[4] = alpha;
x[5] = alpha;
x[6] = alpha;
x[7] = alpha;
#if 0
if(isfinite(x[0])) if(isfinite(x[0]))
x[0] = alpha; x[0] = alpha;
else else
@ -106,7 +115,8 @@ static void dscal_kernel_8_zero (BLASLONG n, FLOAT *x)
else else
x[7] = NAN; x[7] = NAN;
x+=8; x+=8;
} #endif
}
} }
@ -130,6 +140,11 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
if ( n >= 16 ) if ( n >= 16 )
{ {
BLASLONG align = ((32 - ((uintptr_t)x & (uintptr_t)0x1F)) >> 3) & 0x3; BLASLONG align = ((32 - ((uintptr_t)x & (uintptr_t)0x1F)) >> 3) & 0x3;
if (dummy2 == 0)
for (j = 0; j < align; j++) {
x [j] = 0.0;
}
else
for (j = 0; j < align; j++) { for (j = 0; j < align; j++) {
if (isfinite(x[j])) if (isfinite(x[j]))
x[j] = 0.0; x[j] = 0.0;
@ -151,7 +166,13 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
j=n1; j=n1;
} }
#endif #endif
if (dummy2 == 0)
while(j < n)
{
x[j]=0.0;
j++;
}
else
while(j < n) while(j < n)
{ {
if (!isfinite(x[j])) if (!isfinite(x[j]))
@ -202,7 +223,14 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
if ( da == 0.0 ) if ( da == 0.0 )
{ {
if (dummy2 == 0)
while(j < n)
{
x[i]=0.0;
i += inc_x;
j++;
}
else
while(j < n) while(j < n)
{ {
if (!isfinite(x[i])) if (!isfinite(x[i]))

44
kernel/power/sscal.c
View File

@ -74,7 +74,24 @@ static void sscal_kernel_16_zero( BLASLONG n, FLOAT *x )
for( i=0; i<n; i+=8 ) for( i=0; i<n; i+=8 )
{ {
if (isfinite(x[0])) x[0] = alpha;
x[1] = alpha;
x[2] = alpha;
x[3] = alpha;
x[4] = alpha;
x[5] = alpha;
x[6] = alpha;
x[7] = alpha;
x[8] = alpha;
x[9] = alpha;
x[10] = alpha;
x[11] = alpha;
x[12] = alpha;
x[13] = alpha;
x[14] = alpha;
x[15] = alpha;
#if 0
if (isfinite(x[0]))
x[0] = alpha; x[0] = alpha;
else else
x[0] = NAN; x[0] = NAN;
@ -107,7 +124,8 @@ static void sscal_kernel_16_zero( BLASLONG n, FLOAT *x )
else else
x[7] = NAN; x[7] = NAN;
x+=8; x+=8;
} #endif
}
} }
@ -132,6 +150,11 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
if ( n >= 32 ) if ( n >= 32 )
{ {
BLASLONG align = ((32 - ((uintptr_t)x & (uintptr_t)0x1F)) >> 2) & 0x7; BLASLONG align = ((32 - ((uintptr_t)x & (uintptr_t)0x1F)) >> 2) & 0x7;
if (dummy2 == 0)
for (j = 0; j < align; j++){
x[j] = 0.0;
}
else
for (j = 0; j < align; j++) { for (j = 0; j < align; j++) {
if (isfinite(x[j])) if (isfinite(x[j]))
x[j] = 0.0; x[j] = 0.0;
@ -153,7 +176,13 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
j=n1; j=n1;
} }
#endif #endif
if (dummy2 == 0)
while(j < n)
{
x[j] = 0.0;
j++;
}
else
while(j < n) while(j < n)
{ {
if (isfinite(x[j])) if (isfinite(x[j]))
@ -204,7 +233,14 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
if ( da == 0.0 ) if ( da == 0.0 )
{ {
if (dummy2 == 0)
while(j < n)
{
x[i]=0.0;
i += inc_x;
j++;
}
else
while(j < n) while(j < n)
{ {
if (isfinite(x[i])) if (isfinite(x[i]))