[llvm-bugs] [Bug 31517] New: The LoadCombine pass wrongly reorder load and store
via llvm-bugs
llvm-bugs at lists.llvm.org
Tue Jan 3 07:39:56 PST 2017
https://llvm.org/bugs/show_bug.cgi?id=31517
Bug ID: 31517
Summary: The LoadCombine pass wrongly reorder load and store
Product: libraries
Version: trunk
Hardware: PC
OS: Linux
Status: NEW
Severity: normal
Priority: P
Component: Scalar Optimizations
Assignee: unassignedbugs at nondot.org
Reporter: karl-johan.karlsson at ericsson.com
CC: llvm-bugs at lists.llvm.org
Classification: Unclassified
Created attachment 17800
--> https://llvm.org/bugs/attachment.cgi?id=17800&action=edit
opt -S -load-combine -adce -o - loadcombine.ll | FileCheck loadcombine.ll
Input code:
define i16 @bazinga() {
%_tmp9 = getelementptr %rec11, %rec11* @str, i16 0, i32 1
%_tmp10 = load i16, i16* %_tmp9
%_tmp12 = getelementptr %rec11, %rec11* @str, i16 0, i32 0
store i16 %_tmp10, i16* %_tmp12
%_tmp13 = getelementptr %rec11, %rec11* @str, i16 0, i32 0
%_tmp14 = load i16, i16* %_tmp13
%_tmp15 = icmp eq i16 %_tmp14, 3
%_tmp16 = select i1 %_tmp15, i16 1, i16 0
br label %bb1
bb1:
ret i16 %_tmp16
}
The load combiner can not reorder the three memory operations above (load,
store, load) as the first load will load the value from the second location in
the struct, the store will write the loaded value in the first location in the
struct and the last load will load a value from the second location in the
struct.
Output code:
define i16 @bazinga() {
%_tmp14.combined = load i32, i32* bitcast (%rec11* @str to i32*)
%combine.extract.shift = lshr i32 %_tmp14.combined, 16
%combine.extract.trunc1 = trunc i32 %combine.extract.shift to i16
%_tmp12 = getelementptr %rec11, %rec11* @str, i16 0, i32 0
store i16 %combine.extract.trunc1, i16* %_tmp12
%combine.extract.trunc = trunc i32 %_tmp14.combined to i16
%_tmp15 = icmp eq i16 %combine.extract.trunc, 3
%_tmp16 = select i1 %_tmp15, i16 1, i16 0
br label %bb1
bb1:
ret i16 %_tmp16
}
The load combine pass have combined the two loads and thereby reordered the
instructions as the store is done afterwards. Result is that the input value to
the icmp do not get the updated value written by the store.
--
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/20170103/f81ccd7d/attachment.html>
More information about the llvm-bugs
mailing list