Hi Thomas,
are you really telling me, that gfortran's coarray test library is compiled for
offloading to GPU (or other SIMD processors)? Because that's what NVPTX is used
for most, right? In my opinion that makes no sense, because coarrays in Fortran
are used for SISD style accesses. Although the new style now extracts "kernels"
for the access (which are tiny; and therefore should not perform on any GPU),
I'd rather expect the caf_single library to be not compiled for
NVPTX-offloading. Are there different opinions?
Sorry, for disturbing the NVPTX build. I wasn't aware, that it was done for
caf_*.
Regards,
Andre
On Thu, 27 Feb 2025 21:53:42 +0100
Thomas Schwinge <tschwi...@baylibre.com> wrote:
> As of recent commit 8bf0ee8d62b8a08e808344d31354ab713157e15d
> "Fortran: Add transfer_between_remotes [PR107635]", we've got 'alloca' usage
> in 'libgfortran/caf/single.c:_gfortran_caf_transfer_between_remotes', and
> the libgfortran target library fails to build for legacy configurations where
> PTX 'alloca' is not available:
>
> ../../../../source-gcc/libgfortran/caf/single.c: In function
> ‘_gfortran_caf_transfer_between_remotes’:
> ../../../../source-gcc/libgfortran/caf/single.c:675:23: sorry, unimplemented:
> dynamic stack allocation not supported 675 | transfer_desc =
> __builtin_alloca (desc_size); |
> ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
> ../../../../source-gcc/libgfortran/caf/single.c:680:20: sorry, unimplemented:
> dynamic stack allocation not supported 680 | transfer_ptr =
> __builtin_alloca (*opt_dst_charlen * src_size); |
> ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ make[6]: *** [Makefile:4675:
> caf/single.lo] Error 1
>
> With '-mfake-ptx-alloca', libgfortran again succeeds to build, and compared
> to before, we've got only a small number of regressions due to nvptx 'ld'
> complaining about 'unresolved symbol __GCC_nvptx__PTX_alloca_not_supported':
>
> [-PASS:-]{+FAIL:+} gfortran.dg/coarray/codimension_2.f90 -fcoarray=lib
> -O2 -lcaf_single (test for excess errors)
>
> [-PASS:-]{+FAIL:+} gfortran.dg/coarray/event_4.f08 -fcoarray=lib -O2
> -lcaf_single (test for excess errors) [-PASS:-]{+UNRESOLVED:+}
> gfortran.dg/coarray/event_4.f08 -fcoarray=lib -O2 -lcaf_single [-execution
> test-]{+compilation failed to produce executable+}
>
> [-PASS:-]{+FAIL:+} gfortran.dg/coarray/fail_image_2.f08 -fcoarray=lib
> -O2 -lcaf_single (test for excess errors) [-PASS:-]{+UNRESOLVED:+}
> gfortran.dg/coarray/fail_image_2.f08 -fcoarray=lib -O2 -lcaf_single
> [-execution test-]{+compilation failed to produce executable+}
>
> [-PASS:-]{+FAIL:+} gfortran.dg/coarray/proc_pointer_assign_1.f90
> -fcoarray=lib -O2 -lcaf_single (test for excess errors)
> [-PASS:-]{+UNRESOLVED:+} gfortran.dg/coarray/proc_pointer_assign_1.f90
> -fcoarray=lib -O2 -lcaf_single [-execution test-]{+compilation failed to
> produce executable+}
>
> [-PASS:-]{+FAIL:+} gfortran.dg/coarray_43.f90 -O (test for excess
> errors)
>
> That's acceptable for such legacy PTX configurations.
>
> PR target/107635
> libgfortran/
> * config/t-nvptx: New.
> * configure.host [nvptx] (tmake_file): Add it.
> ---
> libgfortran/config/t-nvptx | 2 ++
> libgfortran/configure.host | 4 ++++
> 2 files changed, 6 insertions(+)
> create mode 100644 libgfortran/config/t-nvptx
>
> diff --git a/libgfortran/config/t-nvptx b/libgfortran/config/t-nvptx
> new file mode 100644
> index 00000000000..88bd6eee982
> --- /dev/null
> +++ b/libgfortran/config/t-nvptx
> @@ -0,0 +1,2 @@
> +# Re 'alloca' usage in '../caf/single.c':
> +AM_CFLAGS += -mfake-ptx-alloca
> diff --git a/libgfortran/configure.host b/libgfortran/configure.host
> index 291188d19c2..9abd40f511a 100644
> --- a/libgfortran/configure.host
> +++ b/libgfortran/configure.host
> @@ -91,6 +91,10 @@ case "${target}" in
> tmake_file="t-aix"
> ;;
>
> + nvptx-*-none)
> + tmake_file="$tmake_file t-nvptx"
> + ;;
> +
> *)
> ;;
>
--
Andre Vehreschild * Email: vehre ad gmx dot de
