[llvm-bugs] [Bug 45447] New: Incorrect transformation: (undef u>> a) ^ -1 -> undef >> a, when a != 0
via llvm-bugs
llvm-bugs at lists.llvm.org
Mon Apr 6 07:05:52 PDT 2020
https://bugs.llvm.org/show_bug.cgi?id=45447
Bug ID: 45447
Summary: Incorrect transformation: (undef u>> a) ^ -1 ->
undef >> a, when a != 0
Product: libraries
Version: trunk
Hardware: All
OS: All
Status: NEW
Keywords: miscompilation
Severity: enhancement
Priority: P
Component: Scalar Optimizations
Assignee: unassignedbugs at nondot.org
Reporter: nunoplopes at sapo.pt
CC: juneyoung.lee at sf.snu.ac.kr, lebedev.ri at gmail.com,
llvm-bugs at lists.llvm.org, regehr at cs.utah.edu,
spatel+llvm at rotateright.com
Test: Transforms/InstCombine/vector-xor.ll
Summary: (undef u>> a) ^ -1 always leaves top bits as 1 when a > 0. However
'undef >> a' can leave top bits as 0 or 1, depending on the sign bit of undef.
define <4 x i32> @test_v4i32_not_lshr_nonnegative_const_undef(<4 x i32> %a0) {
%1 = lshr <4 x i32> { 3, 5, undef, 9 }, %a0
%2 = xor <4 x i32> { 4294967295, 4294967295, 4294967295, undef }, %1
ret <4 x i32> %2
}
=>
define <4 x i32> @test_v4i32_not_lshr_nonnegative_const_undef(<4 x i32> %a0) {
%1 = ashr <4 x i32> { 4294967292, 4294967290, undef, 4294967286 }, %a0
ret <4 x i32> %1
}
Transformation doesn't verify!
ERROR: Value mismatch
Example:
<4 x i32> %a0 = < #x00000000 (0), #xfffffffe (4294967294, -2), #x0000001f (31),
#xfffffffe (4294967294, -2) >
Source:
<4 x i32> %1 = < #x00000003 (3), poison, #x00000000 (0) [based on undef value],
poison >
<4 x i32> %2 = < #xfffffffc (4294967292, -4), poison, #xffffffff (4294967295,
-1), poison >
Target:
<4 x i32> %1 = < #xfffffffc (4294967292, -4), poison, #x00000000 (0), poison >
Source value: < #xfffffffc (4294967292, -4), poison, #xffffffff (4294967295,
-1), poison >
Target value: < #xfffffffc (4294967292, -4), poison, #x00000000 (0), poison >
https://web.ist.utl.pt/nuno.lopes/alive2/index.php?hash=2009353267698970&test=Transforms%2FInstCombine%2Fvector-xor.ll
--
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20200406/a67a9141/attachment-0001.html>
More information about the llvm-bugs
mailing list