// We need each element to be the same type of value, and check that each
// element has a single use.
- if (all_of(drop_begin(Item), [Item](InstLane IL) {
- Value *FrontV = Item.front().first->get();
- if (!IL.first)
- return true;
- Value *V = IL.first->get();
- if (auto *I = dyn_cast<Instruction>(V); I && !I->hasOneUse())
- return false;
- if (V->getValueID() != FrontV->getValueID())
- return false;
- if (auto *CI = dyn_cast<CmpInst>(V))
- if (CI->getPredicate() != cast<CmpInst>(FrontV)->getPredicate())
- return false;
- if (auto *CI = dyn_cast<CastInst>(V))
- if (CI->getSrcTy() != cast<CastInst>(FrontV)->getSrcTy())
- return false;
- if (auto *SI = dyn_cast<SelectInst>(V))
- if (!isa<VectorType>(SI->getOperand(0)->getType()) ||
- SI->getOperand(0)->getType() !=
- cast<SelectInst>(FrontV)->getOperand(0)->getType())
- return false;
- if (isa<CallInst>(V) && !isa<IntrinsicInst>(V))
- return false;
- auto *II = dyn_cast<IntrinsicInst>(V);
- return !II || (isa<IntrinsicInst>(FrontV) &&
- II->getIntrinsicID() ==
- cast<IntrinsicInst>(FrontV)->getIntrinsicID());
- })) {
+ auto CheckLaneIsEquivalentToFirst = [Item](InstLane IL) {
+ Value *FrontV = Item.front().first->get();
+ if (!IL.first)
+ return true;
+ Value *V = IL.first->get();
+ if (auto *I = dyn_cast<Instruction>(V); I && !I->hasOneUse())
+ return false;
+ if (V->getValueID() != FrontV->getValueID())
+ return false;
+ if (auto *CI = dyn_cast<CmpInst>(V))
+ if (CI->getPredicate() != cast<CmpInst>(FrontV)->getPredicate())
+ return false;
+ if (auto *CI = dyn_cast<CastInst>(V))
+ if (CI->getSrcTy() != cast<CastInst>(FrontV)->getSrcTy())
+ return false;
+ if (auto *SI = dyn_cast<SelectInst>(V))
+ if (!isa<VectorType>(SI->getOperand(0)->getType()) ||
+ SI->getOperand(0)->getType() !=
+ cast<SelectInst>(FrontV)->getOperand(0)->getType())
+ return false;
+ if (isa<CallInst>(V) && !isa<IntrinsicInst>(V))
+ return false;
+ auto *II = dyn_cast<IntrinsicInst>(V);
+ return !II || (isa<IntrinsicInst>(FrontV) &&
+ II->getIntrinsicID() ==
+ cast<IntrinsicInst>(FrontV)->getIntrinsicID() &&
+ !II->hasOperandBundles());
+ };
+ if (all_of(drop_begin(Item), CheckLaneIsEquivalentToFirst)) {
// Check the operator is one that we support.
if (isa<BinaryOperator, CmpInst>(FrontU)) {
// We exclude div/rem in case they hit UB from poison lanes.
Worklist.push_back(generateInstLaneVectorFromOperand(Item, 2));
continue;
} else if (auto *II = dyn_cast<IntrinsicInst>(FrontU);
- II && isTriviallyVectorizable(II->getIntrinsicID())) {
+ II && isTriviallyVectorizable(II->getIntrinsicID()) &&
+ !II->hasOperandBundles()) {
for (unsigned Op = 0, E = II->getNumOperands() - 1; Op < E; Op++) {
if (isVectorIntrinsicWithScalarOpAtArg(II->getIntrinsicID(), Op)) {
if (!all_of(drop_begin(Item), [Item, Op](InstLane &IL) {
ret <2 x float> %4
}
+define <16 x i64> @operandbundles(<4 x i64> %a, <4 x i64> %b, <4 x i64> %c) {
+; CHECK-LABEL: @operandbundles(
+; CHECK-NEXT: [[CALL:%.*]] = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> [[A:%.*]], <4 x i64> [[B:%.*]], <4 x i64> [[C:%.*]]) [ "jl_roots"(ptr addrspace(10) null, ptr addrspace(10) null) ]
+; CHECK-NEXT: [[SHUFFLEVECTOR:%.*]] = shufflevector <4 x i64> [[CALL]], <4 x i64> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT: [[SHUFFLEVECTOR1:%.*]] = shufflevector <16 x i64> [[SHUFFLEVECTOR]], <16 x i64> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT: ret <16 x i64> [[SHUFFLEVECTOR1]]
+;
+ %call = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a, <4 x i64> %b, <4 x i64> %c) [ "jl_roots"(ptr addrspace(10) null, ptr addrspace(10) null) ]
+ %shufflevector = shufflevector <4 x i64> %call, <4 x i64> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+ %shufflevector1 = shufflevector <16 x i64> %shufflevector, <16 x i64> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+ ret <16 x i64> %shufflevector1
+}
+
+define <8 x i8> @operandbundles_first(<8 x i8> %a) {
+; CHECK-LABEL: @operandbundles_first(
+; CHECK-NEXT: [[AB:%.*]] = shufflevector <8 x i8> [[A:%.*]], <8 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT: [[AT:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
+; CHECK-NEXT: [[ABT:%.*]] = call <4 x i8> @llvm.abs.v4i8(<4 x i8> [[AT]], i1 false) [ "jl_roots"(ptr addrspace(10) null, ptr addrspace(10) null) ]
+; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x i8> [[AT]], <4 x i8> [[AB]], <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT: [[R:%.*]] = call <8 x i8> @llvm.abs.v8i8(<8 x i8> [[TMP1]], i1 false)
+; CHECK-NEXT: ret <8 x i8> [[R]]
+;
+ %ab = shufflevector <8 x i8> %a, <8 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+ %at = shufflevector <8 x i8> %a, <8 x i8> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
+ %abt = call <4 x i8> @llvm.abs.v4i8(<4 x i8> %at, i1 false) [ "jl_roots"(ptr addrspace(10) null, ptr addrspace(10) null) ]
+ %abb = call <4 x i8> @llvm.abs.v4i8(<4 x i8> %ab, i1 false)
+ %r = shufflevector <4 x i8> %abt, <4 x i8> %abb, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+ ret <8 x i8> %r
+}
+
+define <8 x i8> @operandbundles_second(<8 x i8> %a) {
+; CHECK-LABEL: @operandbundles_second(
+; CHECK-NEXT: [[AB:%.*]] = shufflevector <8 x i8> [[A:%.*]], <8 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT: [[AT:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
+; CHECK-NEXT: [[ABB:%.*]] = call <4 x i8> @llvm.abs.v4i8(<4 x i8> [[AB]], i1 false) [ "jl_roots"(ptr addrspace(10) null, ptr addrspace(10) null) ]
+; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x i8> [[AT]], <4 x i8> [[AB]], <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT: [[R:%.*]] = call <8 x i8> @llvm.abs.v8i8(<8 x i8> [[TMP1]], i1 false)
+; CHECK-NEXT: ret <8 x i8> [[R]]
+;
+ %ab = shufflevector <8 x i8> %a, <8 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+ %at = shufflevector <8 x i8> %a, <8 x i8> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
+ %abt = call <4 x i8> @llvm.abs.v4i8(<4 x i8> %at, i1 false)
+ %abb = call <4 x i8> @llvm.abs.v4i8(<4 x i8> %ab, i1 false) [ "jl_roots"(ptr addrspace(10) null, ptr addrspace(10) null) ]
+ %r = shufflevector <4 x i8> %abt, <4 x i8> %abb, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+ ret <8 x i8> %r
+}
+
+declare <4 x i64> @llvm.fshl.v4i64(<4 x i64>, <4 x i64>, <4 x i64>)
declare void @use(<4 x i8>)