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

X86_64: Fixed bug in zscal 

Fixed handling of NAN and INF arguments when
inc is greater than 1.
This commit is contained in:
gxw 2024-01-31 11:20:25 +08:00
parent bb043a021f
commit 969601a1dc
1 changed files with 50 additions and 41 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

91
kernel/x86_64/zscal.c
View File

@ -69,16 +69,16 @@ static void zscal_kernel_8( BLASLONG n, FLOAT *alpha , FLOAT *x )
for( i=0; i<n; i+=4 )
{
t0 = da_r *x[0] - da_i *x[1];
t1 = da_r *x[2] - da_i *x[3];
t2 = da_r *x[4] - da_i *x[5];
t3 = da_r *x[6] - da_i *x[7];
t0 = da_r *x[0] - da_i *x[1];
t1 = da_r *x[2] - da_i *x[3];
t2 = da_r *x[4] - da_i *x[5];
t3 = da_r *x[6] - da_i *x[7];
x[1] = da_r * x[1] + da_i * x[0];
x[3] = da_r * x[3] + da_i * x[2];
x[5] = da_r * x[5] + da_i * x[4];
x[7] = da_r * x[7] + da_i * x[6];
x[0] = t0;
x[2] = t1;
x[4] = t2;
@ -99,16 +99,16 @@ static void zscal_kernel_8_zero_r( BLASLONG n, FLOAT *alpha , FLOAT *x )
for( i=0; i<n; i+=4 )
{
t0 = - da_i *x[1];
t1 = - da_i *x[3];
t2 = - da_i *x[5];
t3 = - da_i *x[7];
t0 = - da_i *x[1];
t1 = - da_i *x[3];
t2 = - da_i *x[5];
t3 = - da_i *x[7];
x[1] = da_i * x[0];
x[3] = da_i * x[2];
x[5] = da_i * x[4];
x[7] = da_i * x[6];
x[0] = t0;
x[2] = t1;
x[4] = t2;
@ -129,16 +129,16 @@ static void zscal_kernel_8_zero_i( BLASLONG n, FLOAT *alpha , FLOAT *x )
for( i=0; i<n; i+=4 )
{
t0 = da_r *x[0];
t1 = da_r *x[2];
t2 = da_r *x[4];
t3 = da_r *x[6];
t0 = da_r *x[0];
t1 = da_r *x[2];
t2 = da_r *x[4];
t3 = da_r *x[6];
x[1] = da_r * x[1];
x[3] = da_r * x[3];
x[5] = da_r * x[5];
x[7] = da_r * x[7];
x[0] = t0;
x[2] = t1;
x[4] = t2;
@ -157,14 +157,14 @@ static void zscal_kernel_8_zero( BLASLONG n, FLOAT *alpha , FLOAT *x )
BLASLONG i;
for( i=0; i<n; i+=4 )
{
x[0] = 0.0;
x[1] = 0.0;
x[2] = 0.0;
x[3] = 0.0;
x[4] = 0.0;
x[5] = 0.0;
x[6] = 0.0;
x[7] = 0.0;
x[0] = 0.0;
x[1] = 0.0;
x[2] = 0.0;
x[3] = 0.0;
x[4] = 0.0;
x[5] = 0.0;
x[6] = 0.0;
x[7] = 0.0;
x+=8;
}
@ -186,10 +186,10 @@ static void zscal_kernel_inc_8(BLASLONG n, FLOAT *alpha, FLOAT *x, BLASLONG inc_
for ( i=0; i<n; i+=4 )
{
t0 = da_r * x[0] - da_i *x[1];
t1 = da_r * x[inc_x] - da_i *x[inc_x + 1];
t2 = da_r * x[inc_x2] - da_i *x[inc_x2 + 1];
t3 = da_r * x[inc_x3] - da_i *x[inc_x3 + 1];
t0 = da_r * x[0] - da_i *x[1];
t1 = da_r * x[inc_x] - da_i *x[inc_x + 1];
t2 = da_r * x[inc_x2] - da_i *x[inc_x2 + 1];
t3 = da_r * x[inc_x3] - da_i *x[inc_x3 + 1];
x[1] = da_i * x[0] + da_r * x[1];
x[inc_x +1] = da_i * x[inc_x] + da_r * x[inc_x +1];
@ -228,7 +228,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i,
{
while(j < n1)
{
x[i]=0.0;
x[i+1]=0.0;
x[i+inc_x]=0.0;
@ -240,7 +240,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i,
while(j < n)
{
x[i]=0.0;
x[i+1]=0.0;
i += inc_x ;
@ -253,11 +253,17 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i,
{
while(j < n1)
{
temp0 = -da_i * x[i+1];
if (isnan(x[i]) || isinf(x[i]))
temp0 = NAN;
else
temp0 = -da_i * x[i+1];
x[i+1] = da_i * x[i];
x[i] = temp0;
temp1 = -da_i * x[i+1+inc_x];
if (isnan(x[i+inc_x]) || isinf(x[i+inc_x]))
temp1 = NAN;
else
temp1 = -da_i * x[i+1+inc_x];
x[i+1+inc_x] = da_i * x[i+inc_x];
x[i+inc_x] = temp1;
i += 2*inc_x ;
@ -267,8 +273,11 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i,
while(j < n)
{
temp0 = -da_i * x[i+1];
if (isnan(x[i]) || isinf(x[i]))
temp0 = NAN;
else
temp0 = -da_i * x[i+1];
x[i+1] = da_i * x[i];
x[i] = temp0;
i += inc_x ;
@ -291,7 +300,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i,
while(j < n1)
{
temp0 = da_r * x[i];
x[i+1] = da_r * x[i+1];
x[i] = temp0;
@ -305,7 +314,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i,
while(j < n)
{
temp0 = da_r * x[i];
x[i+1] = da_r * x[i+1];
x[i] = temp0;
@ -368,7 +377,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i,
}
i = n1 << 1;
j = n1;
if ( da_r == 0.0 || da_r != da_r )
{
if ( da_i == 0.0 )
@ -385,7 +394,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i,
}
}
else if (da_r < -FLT_MAX || da_r > FLT_MAX) {
else if (da_r < -FLT_MAX || da_r > FLT_MAX) {
while(j < n)
{
x[i]= NAN;
@ -404,7 +413,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i,
if (x[i] < -FLT_MAX || x[i] > FLT_MAX)
temp0 = NAN;
x[i+1] = da_i * x[i];
if ( x[i] == x[i]) //preserve NaN
if ( x[i] == x[i]) //preserve NaN
x[i] = temp0;
i += 2 ;
j++;
@ -420,7 +429,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i,
{
while(j < n)
{
temp0 = da_r * x[i];
x[i+1] = da_r * x[i+1];
x[i] = temp0;
@ -442,7 +451,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r, FLOAT da_i,
}
}
}
}