{lib}[GCC/8.3.0 and gcccuda/2019b] pocl v1.4

[branfosj]

(created using eb --new-pr)

requires:

• Clang ({compiler}[GCCcore/8.3.0] Clang v9.0.1 #9589)
• Clang with CUDA ({compiler}[GCCcore/8.3.0] Clang v9.0.1 with CUDA 10.1.243 #9592)

Note that there is an issue to solve over CUDA dependencies - that I've detailed in #9592.

[branfosj]

The patch fixes the error:

/rds/bear-apps/devel/2019b/branfosj-eb-2/EL7/EL7-power9/software/GCCcore/8.3.0/include/c++/8.3.0/cmath:585:31: error: cannot convert int to __vector(4) __bool int in return
   { return __builtin_isinf(__x); }

https://bugs.freedesktop.org/show_bug.cgi?id=68504 talks about this bug. Other option is to replace -std=c++11 with -std=gnu++11.

[branfosj]

Test report by @branfosj
SUCCESS
Build succeeded for 1 out of 1 (1 easyconfigs in this PR)
bear-pg0211u03a.bear.cluster - Linux centos linux 7.7.1908, Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz, Python 2.7.5
See https://gist.github.com/9eff352ab58a8ca5a8c46f2a8787fd09 for a full test report.

[boegel]

Other option is to replace -std=c++11 with -std=gnu++11.

@branfosj You should be able to do that via toolchainopts:

toolchainopts = {'cstd': 'gnu++11'}

I think that's better than adding a patch file that we'll need to maintain...

[branfosj]

Other option is to replace -std=c++11 with -std=gnu++11.

@branfosj You should be able to do that via toolchainopts:

toolchainopts = {'cstd': 'gnu++11'}

I think that's better than adding a patch file that we'll need to maintain...

Sadly we are battling some hard-coding. By adding toolchainopts I can remove 1/3 of the patch, but the other two sections are still needed.

[branfosj]

Test report by @branfosj
SUCCESS
Build succeeded for 2 out of 2 (2 easyconfigs in this PR)
bear-pg0305u05a.bear.cluster - Linux RHEL 7.6, 8335-GTX, Python 2.7.5
See https://gist.github.com/e23b434c2ca716320259c39cd4488374 for a full test report.

[branfosj]

Test report by @branfosj
SUCCESS
Build succeeded for 2 out of 2 (2 easyconfigs in this PR)
bear-pg0212u15b.bear.cluster - Linux centos linux 7.7.1908, Intel(R) Xeon(R) CPU E5-2640 v4 @ 2.40GHz, Python 2.7.5
See https://gist.github.com/b6b8ac3ca2e2d4e3ba283da13feb2f4a for a full test report.

[branfosj]

OK, I've done some more testing. An alternative patch is:

diff -aur pocl-1.4.orig/include/CL/cl_platform.h pocl-1.4/include/CL/cl_platform.h
--- pocl-1.4.orig/include/CL/cl_platform.h      2020-01-03 09:55:56.986465000 +0000
+++ pocl-1.4/include/CL/cl_platform.h   2020-01-03 09:59:18.058173021 +0000
@@ -356,6 +356,8 @@
 /* Define basic vector types */
 #if defined( __VEC__ )
    #include <altivec.h>   /* may be omitted depending on compiler. AltiVec spec provides no way to detect whether the header is required. */
+   #undef bool
+   #undef vector
    typedef __vector unsigned char     __cl_uchar16;
    typedef __vector signed char       __cl_char16;
    typedef __vector unsigned short    __cl_ushort8;

Using this does also not involve setting toolchainopts. This seems a better, as it is a smaller patch. Is this preferable?

[branfosj]

Also, upstream bug report: pocl/pocl#800

[boegel]

OK, I've done some more testing. An alternative patch is:

diff -aur pocl-1.4.orig/include/CL/cl_platform.h pocl-1.4/include/CL/cl_platform.h
--- pocl-1.4.orig/include/CL/cl_platform.h      2020-01-03 09:55:56.986465000 +0000
+++ pocl-1.4/include/CL/cl_platform.h   2020-01-03 09:59:18.058173021 +0000
@@ -356,6 +356,8 @@
 /* Define basic vector types */
 #if defined( __VEC__ )
    #include <altivec.h>   /* may be omitted depending on compiler. AltiVec spec provides no way to detect whether the header is required. */
+   #undef bool
+   #undef vector
    typedef __vector unsigned char     __cl_uchar16;
    typedef __vector signed char       __cl_char16;
    typedef __vector unsigned short    __cl_ushort8;

Using this does also not involve setting toolchainopts. This seems a better, as it is a smaller patch. Is this preferable?

Yeah, let's go with that...

[boegel]

lgtm

[boegel]

Test report by @boegel
SUCCESS
Build succeeded for 2 out of 2 (2 easyconfigs in this PR)
node2587.golett.os - Linux centos linux 7.7.1908, Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz, Python 2.7.5
See https://gist.github.com/1024f76a0b45e80f02c920e691197113 for a full test report.

[branfosj]

Test report by @branfosj
SUCCESS
Build succeeded for 2 out of 2 (2 easyconfigs in this PR)
bear-pg0306u07a.bear.cluster - Linux RHEL 7.6, 8335-GTX, Python 2.7.5
See https://gist.github.com/77d191fc8b9caf549de1fac9015b052a for a full test report.

[branfosj]

Test report by @branfosj
SUCCESS
Build succeeded for 2 out of 2 (2 easyconfigs in this PR)
bear-pg0212u15a.bear.cluster - Linux centos linux 7.7.1908, Intel(R) Xeon(R) CPU E5-2640 v4 @ 2.40GHz, Python 2.7.5
See https://gist.github.com/1b2efb6596be4ca2753afe3a9c9a39e8 for a full test report.

[boegel]

Test report by @boegel
SUCCESS
Build succeeded for 2 out of 2 (2 easyconfigs in this PR)
generoso - Linux centos linux 7.6.1810, Intel(R) Xeon(R) CPU E5-2660 v4 @ 2.00GHz, Python 3.6.8
See https://gist.github.com/ebba79be0739fb3a334a24f85b466acd for a full test report.

[boegel]

Test report by @boegel
SUCCESS
Build succeeded for 2 out of 2 (2 easyconfigs in this PR)
node3159.skitty.os - Linux centos linux 7.7.1908, Intel(R) Xeon(R) Gold 6140 CPU @ 2.30GHz, Python 3.6.8
See https://gist.github.com/ac6ff4666cf0089cdcf48e441d2f8fed for a full test report.

[boegel]

Going in, thanks @branfosj!