[RISCV] Optimize vp.store with an all ones mask to avoid a vmset.

We can use riscv_vse intrinsic instead of riscv_vse_mask. The code here
is based on similar code for handling masked.scatter and vp.scatter.

Reviewed By: frasercrmck

Differential Revision: https://reviews.llvm.org/D110206

diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index af79b07..4cd1820 100644 (file)
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -4741,25 +4741,37 @@ SDValue RISCVTargetLowering::lowerMaskedStore(SDValue Op,
     Mask = MStore->getMask();
   }
 
+  bool IsUnmasked = ISD::isConstantSplatVectorAllOnes(Mask.getNode());
+
   MVT VT = Val.getSimpleValueType();
   MVT XLenVT = Subtarget.getXLenVT();
 
   MVT ContainerVT = VT;
   if (VT.isFixedLengthVector()) {
     ContainerVT = getContainerForFixedLengthVector(VT);
-    MVT MaskVT = MVT::getVectorVT(MVT::i1, ContainerVT.getVectorElementCount());
 
     Val = convertToScalableVector(ContainerVT, Val, DAG, Subtarget);
-    Mask = convertToScalableVector(MaskVT, Mask, DAG, Subtarget);
+    if (!IsUnmasked) {
+      MVT MaskVT =
+          MVT::getVectorVT(MVT::i1, ContainerVT.getVectorElementCount());
+      Mask = convertToScalableVector(MaskVT, Mask, DAG, Subtarget);
+    }
   }
 
   if (!VL)
     VL = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget).second;
 
-  SDValue IntID = DAG.getTargetConstant(Intrinsic::riscv_vse_mask, DL, XLenVT);
-  return DAG.getMemIntrinsicNode(
-      ISD::INTRINSIC_VOID, DL, DAG.getVTList(MVT::Other),
-      {Chain, IntID, Val, BasePtr, Mask, VL}, MemVT, MMO);
+  unsigned IntID =
+      IsUnmasked ? Intrinsic::riscv_vse : Intrinsic::riscv_vse_mask;
+  SmallVector<SDValue, 8> Ops{Chain, DAG.getTargetConstant(IntID, DL, XLenVT)};
+  Ops.push_back(Val);
+  Ops.push_back(BasePtr);
+  if (!IsUnmasked)
+    Ops.push_back(Mask);
+  Ops.push_back(VL);
+
+  return DAG.getMemIntrinsicNode(ISD::INTRINSIC_VOID, DL,
+                                 DAG.getVTList(MVT::Other), Ops, MemVT, MMO);
 }
 
 SDValue

diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vpstore.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vpstore.ll
index 2a9889c..28e2dc8 100644 (file)
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vpstore.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vpstore.ll
@@ -255,3 +255,15 @@ define void @vpstore_v8f64(<8 x double> %val, <8 x double>* %ptr, <8 x i1> %m, i
   call void @llvm.vp.store.v8f64(<8 x double> %val, <8 x double>* %ptr, <8 x i1> %m, i32 %evl)
   ret void
 }
+
+define void @vpstore_v2i8_allones_mask(<2 x i8> %val, <2 x i8>* %ptr, i32 zeroext %evl) {
+; CHECK-LABEL: vpstore_v2i8_allones_mask:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a1, e8, mf8, ta, mu
+; CHECK-NEXT:    vse8.v v8, (a0)
+; CHECK-NEXT:    ret
+  %a = insertelement <2 x i1> undef, i1 true, i32 0
+  %b = shufflevector <2 x i1> %a, <2 x i1> poison, <2 x i32> zeroinitializer
+  call void @llvm.vp.store.v2i8(<2 x i8> %val, <2 x i8>* %ptr, <2 x i1> %b, i32 %evl)
+  ret void
+}

diff --git a/llvm/test/CodeGen/RISCV/rvv/vpstore.ll b/llvm/test/CodeGen/RISCV/rvv/vpstore.ll
index de40242..e9fdb94 100644 (file)
--- a/llvm/test/CodeGen/RISCV/rvv/vpstore.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vpstore.ll
@@ -339,3 +339,15 @@ define void @vpstore_nxv8f64(<vscale x 8 x double> %val, <vscale x 8 x double>*
   call void @llvm.vp.store.nxv8f64(<vscale x 8 x double> %val, <vscale x 8 x double>* %ptr, <vscale x 8 x i1> %m, i32 %evl)
   ret void
 }
+
+define void @vpstore_nxv1i8_allones_mask(<vscale x 1 x i8> %val, <vscale x 1 x i8>* %ptr, i32 zeroext %evl) {
+; CHECK-LABEL: vpstore_nxv1i8_allones_mask:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a1, e8, mf8, ta, mu
+; CHECK-NEXT:    vse8.v v8, (a0)
+; CHECK-NEXT:    ret
+  %a = insertelement <vscale x 1 x i1> undef, i1 true, i32 0
+  %b = shufflevector <vscale x 1 x i1> %a, <vscale x 1 x i1> poison, <vscale x 1 x i32> zeroinitializer
+  call void @llvm.vp.store.nxv1i8(<vscale x 1 x i8> %val, <vscale x 1 x i8>* %ptr, <vscale x 1 x i1> %b, i32 %evl)
+  ret void
+}