[PATCH] D46250: [AArch64][SVE] Asm: Support for LD1RQ load-and-replicate quad-word vector instructions.

[Sander de Smalen via Phabricator via llvm-commits]

sdesmalen added inline comments.


================
Comment at: test/MC/AArch64/SVE/ld1rqb-diagnostics.s:44
+
+ld1rqb z0.h, p0/z, [x0]
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid operand
----------------
SjoerdMeijer wrote:
> fhahn wrote:
> > sdesmalen wrote:
> > > SjoerdMeijer wrote:
> > > > Nit, and feel free to ignore: this "invalid operand" is a really useless error message because it doesn't tell you much (not related/cause by this patch of course). But at least the culprit should be indicated with the carrot symbol, something like this: 
> > > > 
> > > >   ld1rqb z0.h, p0/z, [x0]
> > > >          ^
> > > > 
> > > > Perhaps nice to match that too because it tells where the error is?
> > > Good suggestion, we can now use the DiagnosticPredicate (https://reviews.llvm.org/D45879) to improve these diagnostics. I'll do that in a separate patch though.
> > > 
> > > Checking for the caret would be a nice addition, but since FileCheck doesn't check white-space coding this explicitly in the test using a regular expression becomes a bit cumbersome. I noticed this is also not done in any of the other diagnostics files (although I reckon this is not an argument not to do it), but I think the coverage without caret is sufficient if the rest of the instruction is valid.
> > > Checking for the caret would be a nice addition, but since FileCheck doesn't check white-space coding this explicitly in the test using a regular  expression becomes a bit cumbersome. I noticed this is also not done in any of the other diagnostics files (although I reckon this is not an argument not to do it), but I think the coverage without caret is sufficient if the rest of the instruction is valid.
> > 
> > Would  `-strict-whitespace` do the trick?
> True about the whitespaces. There are plenty of tests however that checks that check for "^". For example, the first file I opened: test/MC/AArch64/arm64-diags.s. That would be a bit more meaningful with adding -strict-whitespace indeed, which is used by other files. Anyway, it was just a suggestion, and if you're planning in fixing the diagnostic then that's fine of course.
Ah, I didn't know about --strict-whitespace! I agree that would be nice to have, but I'll leave for a later patch as well. Thanks for the suggestions.


https://reviews.llvm.org/D46250