[flang-commits] [clang] [flang] [llvm] [clang-tools-extra] [lldb] [IndVars] Add check of loop invariant for trunc instructions (PR #71072)
Markos Horro via flang-commits
flang-commits at lists.llvm.org
Sun Nov 5 04:56:57 PST 2023
================
@@ -0,0 +1,28 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -passes=indvars -S | FileCheck %s
+
+declare void @foo(i16 noundef)
+
+; Function Attrs: mustprogress noreturn uwtable
+define void @bar(i64 noundef %ptr) {
+; CHECK-LABEL: @bar(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[TMP0:%.*]] = trunc i64 [[PTR:%.*]] to i4
+; CHECK-NEXT: [[TMP1:%.*]] = zext i4 [[TMP0]] to i16
+; CHECK-NEXT: br label [[WHILE_BODY:%.*]]
+; CHECK: while.body:
+; CHECK-NEXT: tail call void @foo(i16 noundef signext [[TMP1]])
+; CHECK-NEXT: br label [[WHILE_BODY]]
+;
+entry:
+ br label %while.body
+
+while.body: ; preds = %entry, %while.body
+ %0 = phi i64 [ %ptr, %entry ], [ %add.ptr, %while.body ]
+ %1 = trunc i64 %0 to i16
+ %and = and i16 %1, 15 ; loop invariant
+ tail call void @foo(i16 noundef signext %and)
----------------
markoshorro wrote:
What I've also observed from the regression is that we have one test failing: `test_11` in `Transforms/IndVarSimplify/X86/eliminate-trunc.ll`.
In concrete, in the function the statement failing is:
```ll
define void @test_11() {
...
bb5: ; preds = %bb5, %bb5
%_tmp24 = icmp slt i16 %_tmp15, 0
br i1 %_tmp24, label %bb5, label %bb5
...
}
```
which instead of the expected (as it is currently in the test):
```ll
; CHECK: bb5:
; CHECK-NEXT: [[_TMP24:%.*]] = icmp slt i16 poison, 0
; CHECK-NEXT: br i1 [[_TMP24]], label [[BB5:%.*]], label [[BB5]]
```
It gets simplified to:
```ll
bb5: ; preds = %bb5, %bb5
br i1 poison, label %bb5, label %bb5
```
I'm assuming this simplification is correct, but I don't understand why this is happening with this change.
https://github.com/llvm/llvm-project/pull/71072
More information about the flang-commits
mailing list