https://llvm.org/bugs/show_bug.cgi?id=24498
Bug ID: 24498
Summary: Worse diagnostics for NEON instructions
Product: libraries
Version: trunk
Hardware: PC
OS: All
Status: NEW
Severity: normal
Priority: P
Component: Backend: AArch64
Assignee: unassignedb...@nondot.org
Reporter: petr.pa...@arm.com
CC: llvm-bugs@lists.llvm.org
Classification: Unclassified
Commit r244955 [1] introduced worse error reporting in some cases.
New behaviour (using r245428):
$ echo "fabs.4s v0, v0" | llvm-mc -triple=aarch64 -mattr=-neon -show-encoding
.text
<stdin>:1:1: error: too few operands for instruction
fabs.4s v0, v0
^
Previous behaviour (using r245428 but reverting r244955):
$ echo "fabs.4s v0, v0" | llvm-mc -triple=aarch64 -mattr=-neon -show-encoding
.text
<stdin>:1:6: error: invalid type suffix for instruction
fabs.4s v0, v0
^
What happens is that the short-form NEON matching gives
`MatchResult = Match_MissingFeature' and `ErrorInfo = 0x8' (identifier of the
missing NEON feature). Then the long-form NEON matching gives
`MatchResult = Match_InvalidOperand' and `ErrorInfo = 1' (number of an invalid
operand). However, the new code in r244955 incorrectly mixes these two together
and creates `Match_InvalidOperand' and `ErrorInfo = 0x8'.
I think LLVM should report the 'instruction requires: ...' error for this
example. This could be achieved by moving the newly introduced rewrite code
before the `switch (MatchResult) {...}' code and fixing the rewrite so it also
uses `MatchResult' from the short-form matching.
[1]
http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20150810/293460.html
--
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
