[Libclc-dev] [PATCH 01/14] native_log2: Switch to generic native intrinsic inc file

[Jan Vesely via Libclc-dev]

On Fri, 2017-11-10 at 08:04 -0600, Aaron Watry via Libclc-dev wrote:
> On Thu, 2017-11-09 at 16:05 -0500, Jan Vesely via Libclc-dev wrote:
> > On Thu, 2017-11-09 at 21:57 +0100, Jeroen Ketema wrote:
> > > > On 4 Nov 2017, at 00:07, Jan Vesely via Libclc-dev <libclc-dev at li
> > > > sts.llvm.org> wrote:
> > > > 
> > > > Signed-off-by: Jan Vesely <jan.vesely at rutgers.edu>
> > > > ---
> > > > generic/include/utils.h                             |  9
> > > > +++++++++
> > > > generic/lib/math/native_log2.cl                     |  8 ++------
> > > > .../{native_log2.inc => native_unary_intrinsic.inc} | 21
> > > > +++++++++++++++++++--
> > > > 3 files changed, 30 insertions(+), 8 deletions(-)
> > > > create mode 100644 generic/include/utils.h
> > > > rename generic/lib/math/{native_log2.inc =>
> > > > native_unary_intrinsic.inc} (65%)
> > > > 
> > > > diff --git a/generic/include/utils.h b/generic/include/utils.h
> > > 
> > > Does placing the file in generic/include mean that it will be
> > > copied
> > > upon installation? If so, why is that necessary?
> > 
> > no, only the include files in generic/include/clc are installed.
> > I've a cleanup to remove all but function declaration includes from
> > include/clc on my todo list, but it will take some time.
> 
> That was originally my question before I tested it...  That being said,
> if it's not being installed, are you just keeping it in the include/
> folder because it's used in sub-target overrides, or would it make
> sense to move it to generic/lib/ as a sibling of clcmacro.h?

I'd prefer if all private clc headers were in generic/include (maybe we
should come up with a better name/location).
Going forward I plan to remove all non-declaration files from
generic/include/clc
and provide either __clc_op and/or __clc_native_op (declared in
generic/include/) for at least math functions, thus target
implementation can be just wrappers and pick either one, depending on
what we know about precision of native instructions.

Jan

> 
> --Aaron
> 
> > 
> > > 
> > > > new file mode 100644
> > > > index 0000000..23f4d34
> > > > --- /dev/null
> > > > +++ b/generic/include/utils.h
> > > > @@ -0,0 +1,9 @@
> > > > +
> > > > +#ifndef __CLC_UTILS_H_
> > > > +#define __CLC_UTILS_H_
> > > > +
> > > > +#define __CLC_CONCAT(x, y) x ## y
> > > > +#define __CLC_STR(x) #x
> > > > +#define __CLC_XSTR(x) __CLC_STR(x)
> > > > +
> > > > +#endif
> > > > diff --git a/generic/lib/math/native_log2.cl
> > > > b/generic/lib/math/native_log2.cl
> > > > index 35ed18b..b610423 100644
> > > > --- a/generic/lib/math/native_log2.cl
> > > > +++ b/generic/lib/math/native_log2.cl
> > > > @@ -22,11 +22,7 @@
> > > > 
> > > > #include <clc/clc.h>
> > > > 
> > > > -#define __CLC_FUNCTION __clc_native_log2
> > > > -#define __CLC_INTRINSIC "llvm.log2"
> > > > -#undef cl_khr_fp64
> > > > -#include <clc/math/unary_intrin.inc>
> > > > -
> > > > -#define __CLC_BODY <native_log2.inc>
> > > > +#define __CLC_NATIVE_INTRINSIC log2
> > > > +#define __CLC_BODY <native_unary_intrinsic.inc>
> > > > #define __FLOAT_ONLY
> > > > #include <clc/math/gentype.inc>
> > > > diff --git a/generic/lib/math/native_log2.inc
> > > > b/generic/lib/math/native_unary_intrinsic.inc
> > > > similarity index 65%
> > > > rename from generic/lib/math/native_log2.inc
> > > > rename to generic/lib/math/native_unary_intrinsic.inc
> > > > index 0f6a509..048da8a 100644
> > > > --- a/generic/lib/math/native_log2.inc
> > > > +++ b/generic/lib/math/native_unary_intrinsic.inc
> > > > @@ -20,6 +20,23 @@
> > > >  * THE SOFTWARE.
> > > >  */
> > > > 
> > > > -_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE native_log2(__CLC_GENTYPE
> > > > val) {
> > > > -  return __clc_native_log2(val);
> > > > +#include <utils.h>
> > > > +
> > > > +#ifdef __CLC_SCALAR
> > > > +#define __CLC_FUNCTION __CLC_CONCAT(__clc_native,
> > > > __CLC_NATIVE_INTRINSIC)
> > > > +#define __CLC_INTRINSIC "llvm."
> > > > __CLC_XSTR(__CLC_NATIVE_INTRINSIC)
> > > > +
> > > > +#undef cl_khr_fp64
> > > > +#include <clc/math/unary_intrin.inc>
> > > > +
> > > > +#endif
> > > > +
> > > > +// For some reason this requires double redirection
> > > 
> > > This comment is not very helpful, and is for the same reason you
> > > need
> > > __CLC_XSTR. I would just get rid of the comment.
> > 
> > right, removed locally.
> > 
> > thanks,
> > Jan
> > 
> > > 
> > > Jeroen
> > > 
> > > > +#define __CLC_FUNCTION2(x) __CLC_CONCAT(native_, x)
> > > > +#define __CLC_FUNCTION __CLC_FUNCTION2(__CLC_NATIVE_INTRINSIC)
> > > > +
> > > > +_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE
> > > > __CLC_FUNCTION(__CLC_GENTYPE val) {
> > > > +  return __CLC_CONCAT(__clc_native,
> > > > __CLC_NATIVE_INTRINSIC)(val);
> > > > }
> > > > +#undef __CLC_FUNCTION2
> > > > +#undef __CLC_FUNCTION
> > > > -- 
> > > > 2.13.6
> > > > 
> > > > _______________________________________________
> > > > Libclc-dev mailing list
> > > > Libclc-dev at lists.llvm.org
> > > > http://lists.llvm.org/cgi-bin/mailman/listinfo/libclc-dev
> > > 
> > > 
> > 
> > _______________________________________________
> > Libclc-dev mailing list
> > Libclc-dev at lists.llvm.org
> > http://lists.llvm.org/cgi-bin/mailman/listinfo/libclc-dev
> 
> _______________________________________________
> Libclc-dev mailing list
> Libclc-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/libclc-dev
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: This is a digitally signed message part
URL: <http://lists.llvm.org/pipermail/libclc-dev/attachments/20171110/33a9d178/attachment.sig>