https://llvm.org/bugs/show_bug.cgi?id=26815
Bug ID: 26815
Summary: Clang really shouldn't have ASM tests in it
Product: clang
Version: unspecified
Hardware: PC
OS: Linux
Status: NEW
Severity: normal
Priority: P
Component: -New Bugs
Assignee: unassignedclangb...@nondot.org
Reporter: renato.go...@linaro.org
CC: llvm-bugs@lists.llvm.org
Classification: Unclassified
After enough discussions in the mailing list, and what I believe is now the
consensus, we should move all ASM tests to LLVM and transform all Clang tests
into IR tests.
The few cases that people were arguing about:
1. NEON intrinsics need to lower to NEON instructions one-to-one.
That's not true. I love that the back-end combines vmul+vadd into vmla, and it
should still do it. Testing for a sequence of IR instructions instead (or
builtin) is not worse.
2. Inline ASM needs to be as is in asm output.
No it doesn't. There are a number of cases where we (and GAS) do slight
transformations on the asm and output a different code (ex. add r0, -1 -> sub
r0, 1). Furthermore, inline asm has a nice IR representation, which we can also
test.
3. Specific code-gen issues.
If C code gets translated into IR and then assembly, we can obviously always
find an IR that will produce the same assembly. To do that, simple -emit-llvm.
--
Having said that, the move itself will be *very* tedious, but it is necessary.
Given that most people were complaining about the ARM and AArch64 tests, I
think it's only fair that we, ARM folks, share the load into moving things.
I'll copy as many people as I can into this bug so we can coordinate our
efforts, but I'd like to have a much cleaner Clang test by 3.9.0.
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
