[Patch] [AArch64] Implement allowsUnalignedMemoryAccesses()

[Jiangning Liu]

Hi zhaoshi,

> I believe "ld1/st1 + rev" or "ldr" on big-endian should be controlled by
pattern matching and addressed by a separate patch... Can you point me to
the review thread you mentioned?

http://permalink.gmane.org/gmane.comp.compilers.llvm.cvs/180464

> I looked into ARM64 backend. All vector types are promoted to v2i64/v2i32
and the pattern matcher emits ldr/str instructions.

Yes, and this is probably the way we should follow for non-strict alignment
mode, i.e. generate ldr/str. So it may affect your performance finally due
to instrction change for little endian. For strict mode, we should have to
use ld1+rev for big endian to gurantee the same semantic.

> My assumption is Linux by default is configured to support unaligned
accesses. Do you mean I should add -aarch64-no-strict-align to the command
line in my test?

Yes, I think you should add more RUNs in your test to cover all opts you
are going to add.

> Can you elaborate a little bit about "the ld1/st1 CHECK should be
guaranteed for both little-endian and big-endian"?

Since that big-endian patch has not been upstreamed yet, at present you can
only use ld1/st1 to make correct test case for both little endian and big
endian. I expect your test case should be updated to use ldr/str by the big
endian patch, if your patch will be upstreamed first.

Thanks,
-Jiangning
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140410/f67d3afc/attachment.html>