<html>
<head>
<base href="https://bugs.llvm.org/">
</head>
<body><table border="1" cellspacing="0" cellpadding="8">
<tr>
<th>Bug ID</th>
<td><a class="bz_bug_link
bz_status_NEW "
title="NEW - Missed fold for switch between two values"
href="https://bugs.llvm.org/show_bug.cgi?id=51785">51785</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>Missed fold for switch between two values
</td>
</tr>
<tr>
<th>Product</th>
<td>libraries
</td>
</tr>
<tr>
<th>Version</th>
<td>trunk
</td>
</tr>
<tr>
<th>Hardware</th>
<td>PC
</td>
</tr>
<tr>
<th>OS</th>
<td>Windows NT
</td>
</tr>
<tr>
<th>Status</th>
<td>NEW
</td>
</tr>
<tr>
<th>Severity</th>
<td>enhancement
</td>
</tr>
<tr>
<th>Priority</th>
<td>P
</td>
</tr>
<tr>
<th>Component</th>
<td>Scalar Optimizations
</td>
</tr>
<tr>
<th>Assignee</th>
<td>unassignedbugs@nondot.org
</td>
</tr>
<tr>
<th>Reporter</th>
<td>llvm@rifkin.dev
</td>
</tr>
<tr>
<th>CC</th>
<td>llvm-bugs@lists.llvm.org
</td>
</tr></table>
<p>
<div>
<pre>LLVM fails to optimize
int foo(int x) {
switch (x) {
case 0: return 0;
case 1: return 1;
}
__builtin_unreachable();
}
into
int foo(int x) {
return x;
}
<a href="https://godbolt.org/z/M9YzaWexo">https://godbolt.org/z/M9YzaWexo</a>
GCC misses this too.
Both LLVM and GCC fold well when another case is added, e.g. case 2: return 2;:
int bar(int x) {
switch (x) {
case 0: return 0;
case 1: return 1;
case 2: return 2;
}
__builtin_unreachable();
}
*** IR Dump After SROA on _Z3fooi ***
; Function Attrs: mustprogress nounwind uwtable
define dso_local i32 @_Z3fooi(i32 %0) #0 {
%2 = icmp ult i32 %0, 1
br i1 %2, label %3, label %4
3: ; preds = %1
br label %5
4: ; preds = %1
br label %5
5: ; preds = %4, %3
%6 = phi i32 [ 0, %3 ], [ 1, %4 ]
ret i32 %6
}
*** IR Dump After SimplifyCFGPass on _Z3fooi ***
; Function Attrs: mustprogress nounwind uwtable
define dso_local i32 @_Z3fooi(i32 %0) local_unnamed_addr #0 {
%2 = icmp eq i32 %0, 0
%3 = select i1 %2, i32 0, i32 1
ret i32 %3
}
After the SimplifyCFGPass here, the information required regarding undefined
paths to perform the fold on foo is lost and the transformation cannot be made
correctly later.
For bar, the optimization is actually done in the SimplifyCFGPass:
*** IR Dump After InstCombinePass on _Z3fooi ***
; Function Attrs: mustprogress nofree norecurse nosync nounwind readnone
uwtable willreturn
define dso_local i32 @_Z3fooi(i32 %0) local_unnamed_addr #0 {
switch i32 %0, label %4 [
i32 0, label %5
i32 1, label %2
i32 2, label %3
]
2: ; preds = %1
br label %5
3: ; preds = %1
br label %5
4: ; preds = %1
unreachable
5: ; preds = %1, %3, %2
%6 = phi i32 [ 2, %3 ], [ 1, %2 ], [ 0, %1 ]
ret i32 %6
}
*** IR Dump After SimplifyCFGPass on _Z3fooi ***
; Function Attrs: mustprogress nofree norecurse nosync nounwind readnone
uwtable willreturn
define dso_local i32 @_Z3fooi(i32 %0) local_unnamed_addr #0 {
ret i32 %0
}</pre>
</div>
</p>
<hr>
<span>You are receiving this mail because:</span>
<ul>
<li>You are on the CC list for the bug.</li>
</ul>
</body>
</html>