Default to rv64gc on riscv64 #4390
Conversation
|
I have no opinion/expertise on this, but we shouldn't be switching back and forth between defaults every release. @zyedidia @andreas-schwab Please add to this PR a changelog/release notes entry about this. Possibly related: llvm/llvm-project#61991 |
|
No strong opinion from my side either, but I think the default should account for all features the D language mandates, e.g., double-precision floating point support etc. |
|
Fedora's riscv64 appears to be built using --with-arch=rv64gc --with-abi=lp64d (taken from gcc configure invocation) so from my point of view, this PR seems like a step in the right direction. |
|
As long as we have options to change the enabled extensions and abi (which we do as far as I know) then I think this default is reasonable and consistent with other toolchains. Ideally I think we would have the |
|
I tried to amend this as |
|
Made a silly typo, retesting. |
| @@ -120,6 +127,10 @@ const char *getABI(const llvm::Triple &triple) { | |||
| case llvm::Triple::ppc64le: | |||
| return "elfv2"; | |||
| case llvm::Triple::riscv64: | |||
| if (hasFeature("d")) | |||
| return "lp64d"; | |||
There was a problem hiding this comment.
So a -mattr=-d would get me a lp64d ABI; is that intended? Or should it be narrowed to a +d/+f only?
An entry in CHANGELOG.md would be nice too. :)
Also select the correct ABI by default matching the enabled features (double, float or no floating point). Fixes #4375
|
Thx Andreas! |
Fixes #4375