On 2/12/19 5:05 PM, Jeremie Courreges-Anglas wrote:
On Tue, Feb 12 2019, Brian Callahan <[email protected]> wrote:
On 2/1/19 4:35 PM, Brian Callahan wrote:
Hi ports --
Every time we update LLVM, flang needs to be updated and it takes me
some time to get it done.
Regarding this, do you want some time for the next major upgrades?
I wasn't able to coordinate with the LLVM update this time around and I
don't think flang should hold up LLVM updates so I'm fine with flang
playing catch-up.
It's done now. The patch is attached.
Some notes:
1. Haven't turned arm64 back on. I expect it to work but would like to
compile/run test it first instead of blindly turning it on.
arm64 bulk builds take ~4 days, I don't think phessler and arm64 users
will mind.
I tried it myself and things are still broken on arm64 so I'll have to
talk to the flang people...
2. flang-7 wants to use -lLLVM and build and link to its own separate
libclang.so. I don't want to do either of those things to avoid
potential conflicts, so I've disabled building libclang.so and
ham-fisted a replacement for -lLLVM-7 in a post-configure target,
forcing it to use the LLVM static libs. This creates a very huge
(>80MB) binary. But I think this is fine since the only archs that
flang works on are amd64 and arm64 anyway.
3. Everything is good with the NIST Fortran 77 test suite.
Tests appreciated.
~Brian
No Fortran people? I will just commit it then...
I would not know what to test! If only flang was a possible choice in
fortran.port.mk... :)
I use this: http://www.fortran-2000.com/ArnaudRecipes/fcvs21_f95.html
Specifically, the modified version that conforms to F95.
But yes, thanks for reminding me that we might want to add flang to
fortran.port.mk
~Brian
