diff options
| author | David Green <david.green@arm.com> | 2019-10-20 11:28:33 +0100 |
|---|---|---|
| committer | David Green <david.green@arm.com> | 2019-10-28 15:21:16 +0000 |
| commit | bf21f0d489fb461a8eeb4d6342d28ed2c6e4048d (patch) | |
| tree | 0de5001abd0788d689e5ba0c5979e6eeab619d59 | |
| parent | d157a9bc8ba1085cc4808c6941412322a7fd884e (diff) | |
| download | bcm5719-llvm-bf21f0d489fb461a8eeb4d6342d28ed2c6e4048d.tar.gz bcm5719-llvm-bf21f0d489fb461a8eeb4d6342d28ed2c6e4048d.zip | |
[InstCombine] Extra combine for uadd_sat
This is an extra fold for a canonical form of uadd_sat, as shown in
D68651. It essentially selects uadd from an add and a select.
Differential Revision: https://reviews.llvm.org/D69244
| -rw-r--r-- | llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp | 7 | ||||
| -rw-r--r-- | llvm/test/Transforms/InstCombine/saturating-add-sub.ll | 12 |
2 files changed, 11 insertions, 8 deletions
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp index 9fc871e49b3..b06d31a3fa2 100644 --- a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp @@ -781,6 +781,13 @@ static Value *canonicalizeSaturatedAdd(ICmpInst *Cmp, Value *TVal, Value *FVal, return Builder.CreateBinaryIntrinsic( Intrinsic::uadd_sat, BO->getOperand(0), BO->getOperand(1)); } + // The overflow may be detected via the add wrapping round. + if (match(Cmp0, m_c_Add(m_Specific(Cmp1), m_Value(Y))) && + match(FVal, m_c_Add(m_Specific(Cmp1), m_Specific(Y)))) { + // ((X + Y) u< X) ? -1 : (X + Y) --> uadd.sat(X, Y) + // ((X + Y) u< Y) ? -1 : (X + Y) --> uadd.sat(X, Y) + return Builder.CreateBinaryIntrinsic(Intrinsic::uadd_sat, Cmp1, Y); + } return nullptr; } diff --git a/llvm/test/Transforms/InstCombine/saturating-add-sub.ll b/llvm/test/Transforms/InstCombine/saturating-add-sub.ll index 06232070421..57ef7515e66 100644 --- a/llvm/test/Transforms/InstCombine/saturating-add-sub.ll +++ b/llvm/test/Transforms/InstCombine/saturating-add-sub.ll @@ -1486,10 +1486,8 @@ define i32 @uadd_sat_constant_commute(i32 %x) { define i32 @uadd_sat_canon(i32 %x, i32 %y) { ; CHECK-LABEL: @uadd_sat_canon( -; CHECK-NEXT: [[A:%.*]] = add i32 [[X:%.*]], [[Y:%.*]] -; CHECK-NEXT: [[C:%.*]] = icmp ult i32 [[A]], [[X]] -; CHECK-NEXT: [[R:%.*]] = select i1 [[C]], i32 -1, i32 [[A]] -; CHECK-NEXT: ret i32 [[R]] +; CHECK-NEXT: [[TMP1:%.*]] = call i32 @llvm.uadd.sat.i32(i32 [[X:%.*]], i32 [[Y:%.*]]) +; CHECK-NEXT: ret i32 [[TMP1]] ; %a = add i32 %x, %y %c = icmp ult i32 %a, %x @@ -1499,10 +1497,8 @@ define i32 @uadd_sat_canon(i32 %x, i32 %y) { define i32 @uadd_sat_canon_y(i32 %x, i32 %y) { ; CHECK-LABEL: @uadd_sat_canon_y( -; CHECK-NEXT: [[A:%.*]] = add i32 [[X:%.*]], [[Y:%.*]] -; CHECK-NEXT: [[C:%.*]] = icmp ult i32 [[A]], [[Y]] -; CHECK-NEXT: [[R:%.*]] = select i1 [[C]], i32 -1, i32 [[A]] -; CHECK-NEXT: ret i32 [[R]] +; CHECK-NEXT: [[TMP1:%.*]] = call i32 @llvm.uadd.sat.i32(i32 [[Y:%.*]], i32 [[X:%.*]]) +; CHECK-NEXT: ret i32 [[TMP1]] ; %a = add i32 %x, %y %c = icmp ult i32 %a, %y |
