[PATCH] D48264: [MCA][NFC] Add generic XOP resource tests

[Roman Lebedev via Phabricator via llvm-commits]

lebedev.ri added inline comments.


================
Comment at: test/tools/llvm-mca/X86/Generic/resources-xop.s:393
+vprotb $-9, %xmm0, %xmm3
+vprotb $-9, (%rax), %xmm3
+
----------------
RKSimon wrote:
> lebedev.ri wrote:
> > RKSimon wrote:
> > > Is it useful to test rotl/rotr here? Isn't this more of a MC test?
> > [[ https://support.amd.com/TechDocs/43479.pdf | AMD64 Architecture Programmer’s Manual Volume 6: 128-Bit and 256-Bit XOP and FMA4 Instructions ]], page 239:
> > ```
> > VPROTB Packed Rotate Bytes 
> > ...
> > The VPROTB instruction is an XOP instruction.
> > ```
> > So i think we want to test the form with immediate here?
> > Or is the question about testing multiple immediate?
> Testing different immediates seems over the top - we don't test every shuffle mask immediate etc. 
> 
> Is there any known differences in behaviour? The only thing I can recall is that TRUE/FALSE in VPCOM breaks dependencies - but that kind of thing needs to wait for bdver specific tests like we do for zero idioms etc. and isn't that useful in resource tests.
Hmm, you're right, this is over the top.
Let's keep only one immediate variant.


Repository:
  rL LLVM

https://reviews.llvm.org/D48264