[InstCombine] recognize missed i128 split optimization

llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp

@@ -3119,6 +3119,15 @@ static Value *matchOrConcat(Instruction &Or, InstCombiner::BuilderTy &Builder) {
       match(UpperSrc, m_BitReverse(m_Value(UpperBRev))))
     return ConcatIntrinsicCalls(Intrinsic::bitreverse, UpperBRev, LowerBRev);
 
+  Value *X;
+  if (match(LowerSrc, m_SExt(m_Value(X))) &&
+      match(UpperSrc,
+            m_SExt(m_AShr(
+                m_Specific(X),
+                m_SpecificInt(X->getType()->getScalarSizeInBits() - 1))))) {
+    return Builder.CreateSExt(X, Ty);
+  }
+
   return nullptr;
 }
 

llvm/test/Transforms/InstCombine/iX-ext-split.ll

@@ -0,0 +1,196 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt < %s -passes=instcombine -S | FileCheck %s
+
+; PR129363
+
+; ext split from i32 to i128
+define i128 @i128_ext_split(i32 noundef %x) {
+; CHECK-LABEL: define i128 @i128_ext_split(
+; CHECK-SAME: i32 noundef [[X:%.*]]) {
+; CHECK-NEXT:    [[XX:%.*]] = sext i32 [[X]] to i128
+; CHECK-NEXT:    ret i128 [[XX]]
+;
+  %coerce.sroa.0.0.extract.trunc = sext i32 %x to i64
+  %ashr = ashr i32 %x, 31
+  %coerce.sroa.2.0.extract.trunc = sext i32 %ashr to i64
+  %x.sroa.2.0.insert.ext.i = zext i64 %coerce.sroa.2.0.extract.trunc to i128
+  %x.sroa.2.0.insert.shift.i = shl nuw i128 %x.sroa.2.0.insert.ext.i, 64
+  %x.sroa.0.0.insert.ext.i = zext i64 %coerce.sroa.0.0.extract.trunc to i128
+  %x.sroa.0.0.insert.insert.i = or disjoint i128 %x.sroa.2.0.insert.shift.i, %x.sroa.0.0.insert.ext.i
+  ret i128 %x.sroa.0.0.insert.insert.i
+}
+
+; ext split from i32 to i128
+define void @i128_ext_split_store(i32 %x, ptr %out)  {
+; CHECK-LABEL: define void @i128_ext_split_store(
+; CHECK-SAME: i32 [[X:%.*]], ptr [[OUT:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[RES:%.*]] = sext i32 [[X]] to i128
+; CHECK-NEXT:    store i128 [[RES]], ptr [[OUT]], align 16
+; CHECK-NEXT:    ret void
+;
+entry:
+  %LowerSrc = sext i32 %x to i64
+  %lo = zext i64 %LowerSrc to i128
+
+  %sign = ashr i32 %x, 31
+  %UpperSrc = sext i32 %sign to i64
+  %widen = zext i64 %UpperSrc to i128
+  %hi = shl nuw i128 %widen, 64
+
+  %res = or disjoint i128 %hi, %lo
+  store i128 %res, ptr %out, align 16
+  ret void
+}
+
+; ext split from i16 to i64
+define void @i64_ext_split_store(i16 %x, ptr %out)  {
+; CHECK-LABEL: define void @i64_ext_split_store(
+; CHECK-SAME: i16 [[X:%.*]], ptr [[OUT:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[RES:%.*]] = sext i16 [[X]] to i64
+; CHECK-NEXT:    store i64 [[RES]], ptr [[OUT]], align 16
+; CHECK-NEXT:    ret void
+;
+entry:
+  %LowerSrc = sext i16 %x to i32
+  %lo = zext i32 %LowerSrc to i64
+
+  %sign = ashr i16 %x, 15
+  %UpperSrc = sext i16 %sign to i32
+  %widen = zext i32 %UpperSrc to i64
+  %hi = shl nuw i64 %widen, 32
+
+  %res = or disjoint i64 %hi, %lo
+  store i64 %res, ptr %out, align 16
+  ret void
+}
+
+; ext split from i16 to i128
+define void @i128_ext_split_store_i16(i16 %x, ptr %out)  {
+; CHECK-LABEL: define void @i128_ext_split_store_i16(
+; CHECK-SAME: i16 [[X:%.*]], ptr [[OUT:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[RES:%.*]] = sext i16 [[X]] to i128
+; CHECK-NEXT:    store i128 [[RES]], ptr [[OUT]], align 16
+; CHECK-NEXT:    ret void
+;
+entry:
+  %LowerSrc = sext i16 %x to i64
+  %lo = zext i64 %LowerSrc to i128
+
+  %sign = ashr i16 %x, 15
+  %UpperSrc = sext i16 %sign to i64
+  %widen = zext i64 %UpperSrc to i128
+  %hi = shl nuw i128 %widen, 64
+
+  %res = or disjoint i128 %hi, %lo
+  store i128 %res, ptr %out, align 16
+  ret void
+}
+
+; negative test - wrong constant value
+define i128 @i128_ext_split_neg1(i32 %x) {
+; CHECK-LABEL: define i128 @i128_ext_split_neg1(
+; CHECK-SAME: i32 [[X:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[LOWERSRC:%.*]] = sext i32 [[X]] to i64
+; CHECK-NEXT:    [[LO:%.*]] = zext i64 [[LOWERSRC]] to i128
+; CHECK-NEXT:    [[SIGN:%.*]] = ashr i32 [[X]], 31
+; CHECK-NEXT:    [[UPPERSRC:%.*]] = sext i32 [[SIGN]] to i64
+; CHECK-NEXT:    [[WIDEN:%.*]] = zext i64 [[UPPERSRC]] to i128
+; CHECK-NEXT:    [[HI:%.*]] = shl nuw i128 [[WIDEN]], 65
+; CHECK-NEXT:    [[RES:%.*]] = or disjoint i128 [[HI]], [[LO]]
+; CHECK-NEXT:    ret i128 [[RES]]
+;
+entry:
+  %LowerSrc = sext i32 %x to i64
+  %lo = zext i64 %LowerSrc to i128
+
+  %sign = ashr i32 %x, 31
+  %UpperSrc = sext i32 %sign to i64
+  %widen = zext i64 %UpperSrc to i128
+  %hi = shl nuw i128 %widen, 65
+
+  %res = or disjoint i128 %hi, %lo
+  ret i128 %res
+}
+
+; negative test - wrong shift value
+define i128 @i128_ext_split_neg2(i32 %x) {
+; CHECK-LABEL: define i128 @i128_ext_split_neg2(
+; CHECK-SAME: i32 [[X:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[LOWERSRC:%.*]] = sext i32 [[X]] to i64
+; CHECK-NEXT:    [[LO:%.*]] = zext i64 [[LOWERSRC]] to i128
+; CHECK-NEXT:    [[SIGN:%.*]] = ashr i32 [[X]], 3
+; CHECK-NEXT:    [[UPPERSRC:%.*]] = sext i32 [[SIGN]] to i64
+; CHECK-NEXT:    [[WIDEN:%.*]] = zext i64 [[UPPERSRC]] to i128
+; CHECK-NEXT:    [[HI:%.*]] = shl nuw i128 [[WIDEN]], 64
+; CHECK-NEXT:    [[RES:%.*]] = or disjoint i128 [[HI]], [[LO]]
+; CHECK-NEXT:    ret i128 [[RES]]
+;
+entry:
+  %LowerSrc = sext i32 %x to i64
+  %lo = zext i64 %LowerSrc to i128
+
+  %sign = ashr i32 %x, 3
+  %UpperSrc = sext i32 %sign to i64
+  %widen = zext i64 %UpperSrc to i128
+  %hi = shl nuw i128 %widen, 64
+
+  %res = or disjoint i128 %hi, %lo
+  ret i128 %res
+}
+
+; negative test - wrong ext instruction
+define i128 @i128_ext_split_neg3(i32 %x) {
+; CHECK-LABEL: define i128 @i128_ext_split_neg3(
+; CHECK-SAME: i32 [[X:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[LO:%.*]] = zext i32 [[X]] to i128
+; CHECK-NEXT:    [[SIGN:%.*]] = ashr i32 [[X]], 31
+; CHECK-NEXT:    [[UPPERSRC:%.*]] = sext i32 [[SIGN]] to i64
+; CHECK-NEXT:    [[WIDEN:%.*]] = zext i64 [[UPPERSRC]] to i128
+; CHECK-NEXT:    [[HI:%.*]] = shl nuw i128 [[WIDEN]], 64
+; CHECK-NEXT:    [[RES:%.*]] = or disjoint i128 [[HI]], [[LO]]
+; CHECK-NEXT:    ret i128 [[RES]]
+;
+entry:
+  %LowerSrc = zext i32 %x to i64
+  %lo = zext i64 %LowerSrc to i128
+
+  %sign = ashr i32 %x, 31
+  %UpperSrc = sext i32 %sign to i64
+  %widen = zext i64 %UpperSrc to i128
+  %hi = shl nuw i128 %widen, 64
+
+  %res = or disjoint i128 %hi, %lo
+  ret i128 %res
+}
+
+; negative test - wrong shift
+define i128 @i128_ext_split_neg4(i32 %x) {
+; CHECK-LABEL: define i128 @i128_ext_split_neg4(
+; CHECK-SAME: i32 [[X:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[LOWERSRC:%.*]] = sext i32 [[X]] to i64
+; CHECK-NEXT:    [[LO:%.*]] = zext i64 [[LOWERSRC]] to i128
+; CHECK-NEXT:    [[SIGN:%.*]] = lshr i32 [[X]], 31
+; CHECK-NEXT:    [[WIDEN:%.*]] = zext nneg i32 [[SIGN]] to i128
+; CHECK-NEXT:    [[HI:%.*]] = shl nuw nsw i128 [[WIDEN]], 64
+; CHECK-NEXT:    [[RES:%.*]] = or disjoint i128 [[HI]], [[LO]]
+; CHECK-NEXT:    ret i128 [[RES]]
+;
+entry:
+  %LowerSrc = sext i32 %x to i64
+  %lo = zext i64 %LowerSrc to i128
+
+  %sign = lshr i32 %x, 31
+  %UpperSrc = sext i32 %sign to i64
+  %widen = zext i64 %UpperSrc to i128
+  %hi = shl nuw i128 %widen, 64
+
+  %res = or disjoint i128 %hi, %lo
+  ret i128 %res
+}