[Beignet] [PATCH 2/2] utest_generator.py: add OCL_STRICT_CONFORMANCE enviroment condition.

Zhigang Gong zhigang.gong at linux.intel.com
Mon Jun 23 02:01:52 PDT 2014


Yi,

I already fixed the issue in this patch and pushed the modified version.
Thanks for the patch.

On Mon, Jun 23, 2014 at 10:51:33AM +0800, Zhigang Gong wrote:
> On Mon, Jun 23, 2014 at 08:56:33AM +0800, Yi Sun wrote:
> > For auto-generated math cases, when OCL_STRICT_CONFORMANCE is set, the expected diff increases to 0.001
> > 
> > Signed-off-by: Yi Sun <yi.sun at intel.com>
> > 
> > diff --git a/utests/utest_generator.py b/utests/utest_generator.py
> > index 94900d9..37473d7 100644
> > --- a/utests/utest_generator.py
> > +++ b/utests/utest_generator.py
> > @@ -21,6 +21,8 @@ def udebug(ulpSize,returnType):
> >    text='''
> >      static const char* INFORNAN;
> >      static %s ULPSIZE;
> > +
> > +    const char* env_strict = getenv("OCL_STRICT_CONFORMANCE");
> >      
> >      if (isinf(cpu_data[index])){
> >        INFORNAN="INF";
> > @@ -32,6 +34,9 @@ def udebug(ulpSize,returnType):
> >        ULPSIZE=cl_%s(cpu_data[index]) * %s;
> >      }
> >  
> > +    if ( strcmp(env_strict, "0") == 0)
> > +      ULPSIZE = 0.01;
> Set ULPSIZE to a fixed value is not correct. You still need to consider the target result.
> If the target result is 1e20, then 0.01 will be an impossible accuracy diff even with full
> double precision support.
> 
> > +
> >  #if udebug 
> >      if (isinf(cpu_data[index])){ 
> >        if (isinf(gpu_data[index]))
> > @@ -128,6 +133,7 @@ which can print more values and information to assist debuging the issue.
> >  #include <stdio.h>
> >  #include <math.h>
> >  #include <algorithm>
> > +#include <string.h>
> >  
> >  #define udebug 0
> >  #define FLT_MAX 0x1.fffffep127f
> > -- 
> > 1.8.5.3
> > 
> > _______________________________________________
> > Beignet mailing list
> > Beignet at lists.freedesktop.org
> > http://lists.freedesktop.org/mailman/listinfo/beignet
> _______________________________________________
> Beignet mailing list
> Beignet at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/beignet


More information about the Beignet mailing list