topic Re: DO CONCURRENT offload to Intel MAX GPU on DevCloud not compiling in Intel® Fortran Compiler

DO CONCURRENT offload to Intel MAX GPU on DevCloud not compiling

caplanr — Tue, 22 Nov 2022 01:13:08 GMT

Hi,

I am on devcloud testing a Fortran DO CONCCURENT code.

I have sucessfully compiled it and ran it multi-threaded on the Xeon CPU nodes using "-qopenmp".

I am now trying to compile it for use with the MAX GPU.

I am using the following flags:

-O3 -fp-model precise -heap-arrays -fiopenmp -fopenmp-target-do-concurrent -fopenmp-targets=spir64

I get the following:

INTEL-DEVCLOUD: ~/diffuse_experimental $ make
ifx -c -O3 -fp-model precise -heap-arrays -I/home/u178585/ext_deps/deps/hdf5/include -fiopenmp -fopenmp-target-do-concurrent -fopenmp-targets=spir64 number_types.f
ifx -c -O3 -fp-model precise -heap-arrays -I/home/u178585/ext_deps/deps/hdf5/include -fiopenmp -fopenmp-target-do-concurrent -fopenmp-targets=spir64 zm_parse_modules.f
ifx -c -O3 -fp-model precise -heap-arrays -I/home/u178585/ext_deps/deps/hdf5/include -fiopenmp -fopenmp-target-do-concurrent -fopenmp-targets=spir64 zm_parse.f
ifx -c -O3 -fp-model precise -heap-arrays -I/home/u178585/ext_deps/deps/hdf5/include -fiopenmp -fopenmp-target-do-concurrent -fopenmp-targets=spir64 zm_sds_modules.f
ifx -c -O3 -fp-model precise -heap-arrays -I/home/u178585/ext_deps/deps/hdf5/include -fiopenmp -fopenmp-target-do-concurrent -fopenmp-targets=spir64 zm_sds.f
ifx -o diffuse_experimental_gpu.o -c -O3 -fp-model precise -heap-arrays -I/home/u178585/ext_deps/deps/hdf5/include -fiopenmp -fopenmp-target-do-concurrent -fopenmp-targets=spir64 diffuse_experimental.f
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x146924a]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x11501a2]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x11b6082]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x11b61b0]
/lib/x86_64-linux-gnu/libc.so.6(+0x3ef20)[0x7f92d2c7df20]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x2a9ecaf]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x2a9c8b2]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x2a999d1]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x222312d]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x1544355]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x221e2ad]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x1d0f68f]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x2231f2d]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x1d0b1ec]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x2022e7d]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x1d0d78c]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x202322d]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x1d0c5d3]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x202303d]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x1542f8e]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x201fffe]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x2220fed]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x1542f8e]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x10fcf51]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0x1259aec]
/lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0xe7)[0x7f92d2c60b97]
/glob/development-tools/versions/oneapi/2022.3.1/oneapi/compiler/2022.2.1/linux/bin-llvm/xfortcom[0xeeba29]

/home/u178585/tmp/ifxqXVsvi.bc: error #5633: **Internal compiler error: segmentation violation signal raised** Please report this error along with the circumstances in which it occurred in a Software Problem Report. Note: File and line given may not be explicit cause of this error.
compilation aborted for /home/u178585/tmp/ifxqXVsvi.bc (code 3)
Makefile:43: recipe for target 'diffuse_experimental_gpu.o' failed
make: *** [diffuse_experimental_gpu.o] Error 3

Any ideas on what I am doing wrong?

Or is this a compiler bug?

The code is attached.

Note that this code was used in our publication:

https://link.springer.com/chapter/10.1007/978-3-030-97759-7_1

which has all the codes and build scripts included.

- Ron Caplan

Re:DO CONCURRENT offload to Intel MAX GPU on DevCloud not compiling

Barbara_P_Intel — Tue, 22 Nov 2022 17:05:08 GMT

Thanks for reporting this. I'll investigate. There's an awful lot of module files required to duplicate the ICE (Internal Compiler Error).

Re:DO CONCURRENT offload to Intel MAX GPU on DevCloud not compiling

Barbara_P_Intel — Tue, 22 Nov 2022 17:33:48 GMT

@caplanr Can you please upload just the files needed to reproduce this ICE? Small reproducers are better than the 15GB file you point to (unless I'm looking in the wrong place.) A gzip'd tar file would be awesome.

Re: Re:DO CONCURRENT offload to Intel MAX GPU on DevCloud not compiling

caplanr — Tue, 22 Nov 2022 19:47:50 GMT

Sure! I will try to get that to you later today.

Do you have access to my DevCloud account/folders? That might be easiest?

The code is setup in:

/home/u178585/diffuse_experimental/

- Ron

Re:DO CONCURRENT offload to Intel MAX GPU on DevCloud not compiling

Barbara_P_Intel — Tue, 22 Nov 2022 21:03:14 GMT

Thank you. Can you create the tar ball? I don't have ready access to DevCloud to grab it.

Re: DO CONCURRENT offload to Intel MAX GPU on DevCloud not compiling

caplanr — Tue, 22 Nov 2022 22:11:07 GMT

Hi,

Here is the tarball.

The code relies on the HDF5 library.

I have included the tar file for the version of HDF5 I am using.

It does not seem to compile with IFX and ICX, so I compiled it with IFORT and ICC/ICPC and that worked.

That shoudl not be a probelm as the code compiled and ran fine on the XEON CPU node.

You will need to modify the Makefile to point to wherever the HDF5 is installed to.

- Ron

Re: DO CONCURRENT offload to Intel MAX GPU on DevCloud not compiling

Barbara_P_Intel — Wed, 23 Nov 2022 16:51:21 GMT

Thank you! I appreciate this. I'll take a closer look.

Re:DO CONCURRENT offload to Intel MAX GPU on DevCloud not compiling

Barbara_P_Intel — Wed, 23 Nov 2022 20:52:15 GMT

Good News! This ICE is resolved in the next release! This version of ifx will be part of oneAPI 2023.0 that is planned for release around the end of the year. DevCloud should have it available within a day or two of the announcement.

Re: Re:DO CONCURRENT offload to Intel MAX GPU on DevCloud not compiling

caplanr — Wed, 23 Nov 2022 20:54:27 GMT

Hi,

Ok cool!

I will give it a try after New Years

Thanks!

-Ron

Re:DO CONCURRENT offload to Intel MAX GPU on DevCloud not compiling

Barbara_P_Intel — Wed, 23 Nov 2022 20:58:37 GMT

This is the version I used.

$ ifx --version

ifx (IFORT) 2023.0.0 20221113

Re:DO CONCURRENT offload to Intel MAX GPU on DevCloud not compiling

Barbara_P_Intel — Tue, 03 Jan 2023 18:16:15 GMT

I just confirmed the that Internal Compiler Error is gone in the release of ifx that was made available in December. Please try it!

Re: Re:DO CONCURRENT offload to Intel MAX GPU on DevCloud not compiling

caplanr — Tue, 03 Jan 2023 21:00:20 GMT

Hi,

The compiler works and does not seg fault anymore.

Thanks!

- Ron