Simplify [su]max(MAX, n) to MAX. This comes up in loop tripcount
computations in loops with multiple exits.
Adjust the testcase for PR4436 so that the relevant portion isn't
optimized away.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@74073 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/Analysis/ScalarEvolution.cpp b/lib/Analysis/ScalarEvolution.cpp
index 3bbeb95..6ac6efa 100644
--- a/lib/Analysis/ScalarEvolution.cpp
+++ b/lib/Analysis/ScalarEvolution.cpp
@@ -1690,10 +1690,14 @@
LHSC = cast<SCEVConstant>(Ops[0]);
}
- // If we are left with a constant -inf, strip it off.
+ // If we are left with a constant minimum-int, strip it off.
if (cast<SCEVConstant>(Ops[0])->getValue()->isMinValue(true)) {
Ops.erase(Ops.begin());
--Idx;
+ } else if (cast<SCEVConstant>(Ops[0])->getValue()->isMaxValue(true)) {
+ // If we have an smax with a constant maximum-int, it will always be
+ // maximum-int.
+ return Ops[0];
}
}
@@ -1777,10 +1781,14 @@
LHSC = cast<SCEVConstant>(Ops[0]);
}
- // If we are left with a constant zero, strip it off.
+ // If we are left with a constant minimum-int, strip it off.
if (cast<SCEVConstant>(Ops[0])->getValue()->isMinValue(false)) {
Ops.erase(Ops.begin());
--Idx;
+ } else if (cast<SCEVConstant>(Ops[0])->getValue()->isMaxValue(false)) {
+ // If we have an umax with a constant maximum-int, it will always be
+ // maximum-int.
+ return Ops[0];
}
}
diff --git a/test/Transforms/IndVarSimplify/loop_evaluate7.ll b/test/Transforms/IndVarSimplify/loop_evaluate7.ll
index bce65e0..6e31c55 100644
--- a/test/Transforms/IndVarSimplify/loop_evaluate7.ll
+++ b/test/Transforms/IndVarSimplify/loop_evaluate7.ll
@@ -4,7 +4,7 @@
target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:32:32"
target triple = "i386-pc-linux-gnu"
-define i8* @string_expandtabs() nounwind {
+define i8* @string_expandtabs(i32 %n, i8* %m) nounwind {
entry:
br i1 undef, label %bb33, label %bb1
@@ -22,11 +22,11 @@
bb20: ; preds = %bb19
%0 = load i32* undef, align 4 ; <i32> [#uses=1]
- %1 = sub i32 %0, undef ; <i32> [#uses=1]
+ %1 = sub i32 %0, %n ; <i32> [#uses=1]
br label %bb23
bb21: ; preds = %bb23
- %2 = icmp ult i8* %q.0, undef ; <i1> [#uses=1]
+ %2 = icmp ult i8* %q.0, %m ; <i1> [#uses=1]
br i1 %2, label %bb22, label %overflow2
bb22: ; preds = %bb21
diff --git a/test/Transforms/IndVarSimplify/loop_evaluate8.ll b/test/Transforms/IndVarSimplify/loop_evaluate8.ll
new file mode 100644
index 0000000..fa2f9e5
--- /dev/null
+++ b/test/Transforms/IndVarSimplify/loop_evaluate8.ll
@@ -0,0 +1,63 @@
+; RUN: llvm-as < %s | opt -indvars | llvm-dis | not grep select
+
+; This loop has backedge-taken-count zero. Indvars shouldn't expand any
+; instructions to compute a trip count.
+
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:32:32"
+target triple = "i386-pc-linux-gnu"
+
+define i8* @string_expandtabs() nounwind {
+entry:
+ br i1 undef, label %bb33, label %bb1
+
+bb1: ; preds = %entry
+ br i1 undef, label %overflow1, label %bb15
+
+bb15: ; preds = %bb1
+ br i1 undef, label %bb33, label %bb17
+
+bb17: ; preds = %bb15
+ br label %bb30
+
+bb19: ; preds = %bb30
+ br i1 undef, label %bb20, label %bb29
+
+bb20: ; preds = %bb19
+ %0 = load i32* undef, align 4 ; <i32> [#uses=1]
+ %1 = sub i32 %0, undef ; <i32> [#uses=1]
+ br label %bb23
+
+bb21: ; preds = %bb23
+ %2 = icmp ult i8* %q.0, undef ; <i1> [#uses=1]
+ br i1 %2, label %bb22, label %overflow2
+
+bb22: ; preds = %bb21
+ %3 = getelementptr i8* %q.0, i32 1 ; <i8*> [#uses=1]
+ br label %bb23
+
+bb23: ; preds = %bb22, %bb20
+ %i.2 = phi i32 [ %1, %bb20 ], [ %4, %bb22 ] ; <i32> [#uses=1]
+ %q.0 = phi i8* [ undef, %bb20 ], [ %3, %bb22 ] ; <i8*> [#uses=3]
+ %4 = add i32 %i.2, -1 ; <i32> [#uses=2]
+ %5 = icmp eq i32 %4, -1 ; <i1> [#uses=1]
+ br i1 %5, label %bb29, label %bb21
+
+bb29: ; preds = %bb23, %bb19
+ %q.1 = phi i8* [ undef, %bb19 ], [ %q.0, %bb23 ] ; <i8*> [#uses=0]
+ br label %bb30
+
+bb30: ; preds = %bb29, %bb17
+ br i1 undef, label %bb19, label %bb33
+
+overflow2: ; preds = %bb21
+ br i1 undef, label %bb32, label %overflow1
+
+bb32: ; preds = %overflow2
+ br label %overflow1
+
+overflow1: ; preds = %bb32, %overflow2, %bb1
+ ret i8* null
+
+bb33: ; preds = %bb30, %bb15, %entry
+ ret i8* undef
+}