Re: math/R: Flang is broken for the foreseeable future; switch arm64 to gfortran
On Thursday, August 13, 2020 10:59 AM, Peter Hessler wrote: > I've slipped this into the currently running aarch64 bulk build, it > fails with: > > ... > checking size of int... configure: error: in > `/usr/obj/ports/R-4.0.2/build-aarch64': > configure: error: cannot compute sizeof (int) > > -peter > Thanks Peter. That gives me more motivation to beat Nvidia/ARM to the punch and do the upgrade myself. ~Brian
math/R: Flang is broken for the foreseeable future; switch arm64 to gfortran
Hi Ingo -- When we upgraded to llvm-10.0.0, I had to mark flang as BROKEN [0] because upstream has not yet ported what is now called Classic Flang to llvm-10, as most efforts have shifted to the new Flang, which has been upstreamed into mainline LLVM. There are plans to upgrade Classic Flang to llvm-10 but nothing as of yet has manifested, and there is no timetable for such an upgrade. As a result, R will no longer build on arm64, since it uses flang for its Fortran code. The solution is, at least for now, to switch arm64 to use gfortran, which the attached diff does. I don't have an arm64 available to test, so I am hoping someone can volunteer to test this out and make sure it builds. OK? ~Brian [0] https://marc.info/?l=openbsd-ports-cvs=159661468621760=2 Index: Makefile === RCS file: /cvs/ports/math/R/Makefile,v retrieving revision 1.118 diff -u -p -r1.118 Makefile --- Makefile 23 Jun 2020 18:54:24 - 1.118 +++ Makefile 13 Aug 2020 14:34:00 - @@ -2,6 +2,7 @@ COMMENT= powerful math/statistics/graphics language DISTNAME= R-4.0.2 +REVISION= 0 SO_VERSION= 37.0 .for _lib in R Rblas Rlapack @@ -29,11 +30,7 @@ MASTER_SITES= https://cran.r-project.org MODULES= fortran \ x11/tk -.if ${MACHINE_ARCH:Maarch64} -MODFORTRAN_COMPILER = flang -.else MODFORTRAN_COMPILER = gfortran -.endif BUILD_DEPENDS= ${MODFORTRAN_BUILD_DEPENDS} \ archivers/gtar \
Re: math/R: Flang is broken for the foreseeable future; switch arm64 to gfortran
I've slipped this into the currently running aarch64 bulk build, it fails with: ... checking size of int... configure: error: in `/usr/obj/ports/R-4.0.2/build-aarch64': configure: error: cannot compute sizeof (int) -peter On 2020 Aug 13 (Thu) at 14:47:48 + (+), Brian Callahan wrote: :Hi Ingo -- : :When we upgraded to llvm-10.0.0, I had to mark flang as BROKEN [0] :because upstream has not yet ported what is now called Classic Flang :to llvm-10, as most efforts have shifted to the new Flang, which has :been upstreamed into mainline LLVM. There are plans to upgrade Classic :Flang to llvm-10 but nothing as of yet has manifested, and there is no :timetable for such an upgrade. : :As a result, R will no longer build on arm64, since it uses flang for :its Fortran code. The solution is, at least for now, to switch arm64 :to use gfortran, which the attached diff does. I don't have an arm64 :available to test, so I am hoping someone can volunteer to test this :out and make sure it builds. : :OK? : :~Brian : :[0] https://marc.info/?l=openbsd-ports-cvs=159661468621760=2 :Index: Makefile :=== :RCS file: /cvs/ports/math/R/Makefile,v :retrieving revision 1.118 :diff -u -p -r1.118 Makefile :--- Makefile 23 Jun 2020 18:54:24 - 1.118 :+++ Makefile 13 Aug 2020 14:34:00 - :@@ -2,6 +2,7 @@ : : COMMENT= powerful math/statistics/graphics language : DISTNAME= R-4.0.2 :+REVISION= 0 : : SO_VERSION= 37.0 : .for _lib in R Rblas Rlapack :@@ -29,11 +30,7 @@ MASTER_SITES= https://cran.r-project.org : MODULES= fortran \ : x11/tk : :-.if ${MACHINE_ARCH:Maarch64} :-MODFORTRAN_COMPILER = flang :-.else : MODFORTRAN_COMPILER = gfortran :-.endif : : BUILD_DEPENDS=${MODFORTRAN_BUILD_DEPENDS} \ : archivers/gtar \