As an attempt to fix errors in Flang regression tests on RISCV64 platform, RISCV64 target was added, and subsequent tests were provided.
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
Do you have a buildbot that test this target for Flang?
Your patch has clang-format issues.
I have test it under qemu-user locally with good results, and am actively trying to compile it under risc-v unmatched board. The compilation is extremely lengthy and I can post the result once it has been finished.
Make sure your source tree is up to date. Target.cpp specifically changed in the area the last chunk is making changes to.
flang/lib/Optimizer/CodeGen/Target.cpp | ||
---|---|---|
357 | I am not sure about this. Can you please take a look at https://godbolt.org/z/7PYsv8PjM? It seems that the arguments are passed as two separate values. |
flang/lib/Optimizer/CodeGen/Target.cpp | ||
---|---|---|
357 | The riscv-cc says
so I was a bit unsure about whether or not this procedure(reducing struct of two values into two separate value) should be done here. |
Failed Tests (1):
flang-OldUnit :: Evaluate/real.test
FAIL: flang-OldUnit :: Evaluate/real.test (1044 of 1652)
- TEST 'flang-OldUnit :: Evaluate/real.test' FAILED ****
.../llvm-project/flang/unittests/Evaluate/real.cpp:504: FAIL: FlagsToBits(prod.flags) == 0x18, not 0x10
0 0x800001 * 0xbf7ffffe
The only test failed was not due to this patch and it is caused by risc-v not flagging for underflow.
Issues and a fix is created for this error.
https://github.com/llvm/llvm-project/issues/59132
https://reviews.llvm.org/D138503
I am not sure about this. Can you please take a look at https://godbolt.org/z/7PYsv8PjM?
It seems that the arguments are passed as two separate values.