[RISCV][SelectionDAG] Support VP_ADD/VP_MUL/VP_SUB mask operations

[Jimerlife]

Summary

I abandoned D124080. In SelectionDAG::getNode, use VP_XOR or VP_AND replace VP_ADD/VP_MUL/VP_SUB when operand type is mask.

[craig.topper]

This should be done in SelectionDAG::getNode not SelectionDAGBuilder so it happens any time a mask VP_ADD/VP_SUB/VP_MUL is created.

llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

7586 ↗

(On Diff #424087)

VP_MUL should be replaced with VP_AND.

0 * 0 -> 0
0 * 1 -> 0
1 * 0 -> 0
1 * 1 -> 1

The same as the truth table for AND.

[Jimerlife]

Address comments.
1.Move CheckVPMaskOp to SelectionDAG::getNode.

2. use VP_AND replace VP_MUL

[Jimerlife]

rebase main

[Jimerlife]

In D124144#3463736, @craig.topper wrote:

This should be done in SelectionDAG::getNode not SelectionDAGBuilder so it happens any time a mask VP_ADD/VP_SUB/VP_MUL is created.

I have done it in SelectionDAG::getNode. Thank for your advice.

[craig.topper]

llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

8838

Why not make this part of the switch? You don't need the VTList. This function only receives one VT.

[Jimerlife]

address comment

[craig.topper]

llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

8840

I meant use the existing switch at line 8796.

[Jimerlife]

address comment

[craig.topper]

LGTM

llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

case ISD::SELECT_CC:
     assert(Ops[2].getValueType() == VT &&
            "select_cc node must be of same type as true and false value!");
     break;
   case ISD::BR_CC:
     assert(NumOps == 5 && "BR_CC takes 5 operands!");
     assert(Ops[2].getValueType() == Ops[3].getValueType() &&
            "LHS/RHS of comparison should match types!");
     break;
+  case ISD::VP_ADD:
+  case ISD::VP_SUB:
+    // If it is VP_ADD/VP_SUB mask operation then turn it to VP_XOR
+    if (VT.isVector() && VT.getVectorElementType() == MVT::i1)
+      Opcode = ISD::VP_XOR;
+    break;
+  case ISD::VP_MUL:
+    // If it is VP_MUL mask operation then turn it to VP_AND
+    if (VT.isVector() && VT.getVectorElementType() == MVT::i1)
+      Opcode = ISD::VP_AND;
+    break;
   }
 
   // Memoize nodes.
   SDNode *N;
   SDVTList VTs = getVTList(VT);
 
   if (VT != MVT::Glue) {
     FoldingSetNodeID ID;
     AddNodeIDNode(ID, Opcode, VTs, Ops);
     void *IP = nullptr;
 
     if (SDNode *E = FindNodeOrInsertPos(ID, DL, IP))
       return SDValue(E, 0);
 
     N = newSDNode<SDNode>(Opcode, DL.getIROrder(), DL.getDebugLoc(), VTs);
     createOperands(N, Ops);
 

llvm/test/CodeGen/RISCV/fixed-vectors-vadd-vp-mask.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -mattr=+v -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s \
+; RUN:   | FileCheck %s --check-prefixes=CHECK
+; RUN: llc -mtriple=riscv64 -mattr=+v -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s \
+; RUN:   | FileCheck %s --check-prefixes=CHECK
+
+declare <2 x i1> @llvm.vp.add.v2i1(<2 x i1>, <2 x i1>, <2 x i1>, i32)
+
+define <2 x i1> @vadd_vv_v2i1(<2 x i1> %va, <2 x i1> %b, <2 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vadd_vv_v2i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, mf8, ta, mu
+; CHECK-NEXT:    vmxor.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <2 x i1> @llvm.vp.add.v2i1(<2 x i1> %va, <2 x i1> %b, <2 x i1> %m, i32 %evl)
+  ret <2 x i1> %v
+}
+
+declare <4 x i1> @llvm.vp.add.v4i1(<4 x i1>, <4 x i1>, <4 x i1>, i32)
+
+define <4 x i1> @vadd_vv_v4i1(<4 x i1> %va, <4 x i1> %b, <4 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vadd_vv_v4i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, mf4, ta, mu
+; CHECK-NEXT:    vmxor.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <4 x i1> @llvm.vp.add.v4i1(<4 x i1> %va, <4 x i1> %b, <4 x i1> %m, i32 %evl)
+  ret <4 x i1> %v
+}
+
+declare <8 x i1> @llvm.vp.add.v8i1(<8 x i1>, <8 x i1>, <8 x i1>, i32)
+
+define <8 x i1> @vadd_vv_v8i1(<8 x i1> %va, <8 x i1> %b, <8 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vadd_vv_v8i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, mf2, ta, mu
+; CHECK-NEXT:    vmxor.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <8 x i1> @llvm.vp.add.v8i1(<8 x i1> %va, <8 x i1> %b, <8 x i1> %m, i32 %evl)
+  ret <8 x i1> %v
+}
+
+declare <16 x i1> @llvm.vp.add.v16i1(<16 x i1>, <16 x i1>, <16 x i1>, i32)
+
+define <16 x i1> @vadd_vv_v16i1(<16 x i1> %va, <16 x i1> %b, <16 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vadd_vv_v16i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, m1, ta, mu
+; CHECK-NEXT:    vmxor.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <16 x i1> @llvm.vp.add.v16i1(<16 x i1> %va, <16 x i1> %b, <16 x i1> %m, i32 %evl)
+  ret <16 x i1> %v
+}
+
+declare <32 x i1> @llvm.vp.add.v32i1(<32 x i1>, <32 x i1>, <32 x i1>, i32)
+
+define <32 x i1> @vadd_vv_v32i1(<32 x i1> %va, <32 x i1> %b, <32 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vadd_vv_v32i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, m2, ta, mu
+; CHECK-NEXT:    vmxor.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <32 x i1> @llvm.vp.add.v32i1(<32 x i1> %va, <32 x i1> %b, <32 x i1> %m, i32 %evl)
+  ret <32 x i1> %v
+}
+
+declare <64 x i1> @llvm.vp.add.v64i1(<64 x i1>, <64 x i1>, <64 x i1>, i32)
+
+define <64 x i1> @vadd_vv_v64i1(<64 x i1> %va, <64 x i1> %b, <64 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vadd_vv_v64i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, m4, ta, mu
+; CHECK-NEXT:    vmxor.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <64 x i1> @llvm.vp.add.v64i1(<64 x i1> %va, <64 x i1> %b, <64 x i1> %m, i32 %evl)
+  ret <64 x i1> %v
+}

llvm/test/CodeGen/RISCV/fixed-vectors-vmul-vp-mask.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -mattr=+v -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s \
+; RUN:   | FileCheck %s --check-prefixes=CHECK
+; RUN: llc -mtriple=riscv64 -mattr=+v -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s \
+; RUN:   | FileCheck %s --check-prefixes=CHECK
+
+declare <2 x i1> @llvm.vp.mul.v2i1(<2 x i1>, <2 x i1>, <2 x i1>, i32)
+
+define <2 x i1> @vmul_vv_v2i1(<2 x i1> %va, <2 x i1> %b, <2 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vmul_vv_v2i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, mf8, ta, mu
+; CHECK-NEXT:    vmand.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <2 x i1> @llvm.vp.mul.v2i1(<2 x i1> %va, <2 x i1> %b, <2 x i1> %m, i32 %evl)
+  ret <2 x i1> %v
+}
+
+declare <4 x i1> @llvm.vp.mul.v4i1(<4 x i1>, <4 x i1>, <4 x i1>, i32)
+
+define <4 x i1> @vmul_vv_v4i1(<4 x i1> %va, <4 x i1> %b, <4 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vmul_vv_v4i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, mf4, ta, mu
+; CHECK-NEXT:    vmand.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <4 x i1> @llvm.vp.mul.v4i1(<4 x i1> %va, <4 x i1> %b, <4 x i1> %m, i32 %evl)
+  ret <4 x i1> %v
+}
+
+declare <8 x i1> @llvm.vp.mul.v8i1(<8 x i1>, <8 x i1>, <8 x i1>, i32)
+
+define <8 x i1> @vmul_vv_v8i1(<8 x i1> %va, <8 x i1> %b, <8 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vmul_vv_v8i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, mf2, ta, mu
+; CHECK-NEXT:    vmand.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <8 x i1> @llvm.vp.mul.v8i1(<8 x i1> %va, <8 x i1> %b, <8 x i1> %m, i32 %evl)
+  ret <8 x i1> %v
+}
+
+declare <16 x i1> @llvm.vp.mul.v16i1(<16 x i1>, <16 x i1>, <16 x i1>, i32)
+
+define <16 x i1> @vmul_vv_v16i1(<16 x i1> %va, <16 x i1> %b, <16 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vmul_vv_v16i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, m1, ta, mu
+; CHECK-NEXT:    vmand.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <16 x i1> @llvm.vp.mul.v16i1(<16 x i1> %va, <16 x i1> %b, <16 x i1> %m, i32 %evl)
+  ret <16 x i1> %v
+}
+
+declare <32 x i1> @llvm.vp.mul.v32i1(<32 x i1>, <32 x i1>, <32 x i1>, i32)
+
+define <32 x i1> @vmul_vv_v32i1(<32 x i1> %va, <32 x i1> %b, <32 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vmul_vv_v32i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, m2, ta, mu
+; CHECK-NEXT:    vmand.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <32 x i1> @llvm.vp.mul.v32i1(<32 x i1> %va, <32 x i1> %b, <32 x i1> %m, i32 %evl)
+  ret <32 x i1> %v
+}
+
+declare <64 x i1> @llvm.vp.mul.v64i1(<64 x i1>, <64 x i1>, <64 x i1>, i32)
+
+define <64 x i1> @vmul_vv_v64i1(<64 x i1> %va, <64 x i1> %b, <64 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vmul_vv_v64i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, m4, ta, mu
+; CHECK-NEXT:    vmand.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <64 x i1> @llvm.vp.mul.v64i1(<64 x i1> %va, <64 x i1> %b, <64 x i1> %m, i32 %evl)
+  ret <64 x i1> %v
+}

llvm/test/CodeGen/RISCV/fixed-vectors-vsub-vp-mask.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -mattr=+v -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s \
+; RUN:   | FileCheck %s --check-prefixes=CHECK
+; RUN: llc -mtriple=riscv64 -mattr=+v -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s \
+; RUN:   | FileCheck %s --check-prefixes=CHECK
+
+declare <2 x i1> @llvm.vp.sub.v2i1(<2 x i1>, <2 x i1>, <2 x i1>, i32)
+
+define <2 x i1> @vsub_vv_v2i1(<2 x i1> %va, <2 x i1> %b, <2 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vsub_vv_v2i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, mf8, ta, mu
+; CHECK-NEXT:    vmxor.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <2 x i1> @llvm.vp.sub.v2i1(<2 x i1> %va, <2 x i1> %b, <2 x i1> %m, i32 %evl)
+  ret <2 x i1> %v
+}
+
+declare <4 x i1> @llvm.vp.sub.v4i1(<4 x i1>, <4 x i1>, <4 x i1>, i32)
+
+define <4 x i1> @vsub_vv_v4i1(<4 x i1> %va, <4 x i1> %b, <4 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vsub_vv_v4i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, mf4, ta, mu
+; CHECK-NEXT:    vmxor.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <4 x i1> @llvm.vp.sub.v4i1(<4 x i1> %va, <4 x i1> %b, <4 x i1> %m, i32 %evl)
+  ret <4 x i1> %v
+}
+
+declare <8 x i1> @llvm.vp.sub.v8i1(<8 x i1>, <8 x i1>, <8 x i1>, i32)
+
+define <8 x i1> @vsub_vv_v8i1(<8 x i1> %va, <8 x i1> %b, <8 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vsub_vv_v8i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, mf2, ta, mu
+; CHECK-NEXT:    vmxor.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <8 x i1> @llvm.vp.sub.v8i1(<8 x i1> %va, <8 x i1> %b, <8 x i1> %m, i32 %evl)
+  ret <8 x i1> %v
+}
+
+declare <16 x i1> @llvm.vp.sub.v16i1(<16 x i1>, <16 x i1>, <16 x i1>, i32)
+
+define <16 x i1> @vsub_vv_v16i1(<16 x i1> %va, <16 x i1> %b, <16 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vsub_vv_v16i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, m1, ta, mu
+; CHECK-NEXT:    vmxor.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <16 x i1> @llvm.vp.sub.v16i1(<16 x i1> %va, <16 x i1> %b, <16 x i1> %m, i32 %evl)
+  ret <16 x i1> %v
+}
+
+declare <32 x i1> @llvm.vp.sub.v32i1(<32 x i1>, <32 x i1>, <32 x i1>, i32)
+
+define <32 x i1> @vsub_vv_v32i1(<32 x i1> %va, <32 x i1> %b, <32 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vsub_vv_v32i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, m2, ta, mu
+; CHECK-NEXT:    vmxor.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <32 x i1> @llvm.vp.sub.v32i1(<32 x i1> %va, <32 x i1> %b, <32 x i1> %m, i32 %evl)
+  ret <32 x i1> %v
+}
+
+declare <64 x i1> @llvm.vp.sub.v64i1(<64 x i1>, <64 x i1>, <64 x i1>, i32)
+
+define <64 x i1> @vsub_vv_v64i1(<64 x i1> %va, <64 x i1> %b, <64 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vsub_vv_v64i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, m4, ta, mu
+; CHECK-NEXT:    vmxor.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <64 x i1> @llvm.vp.sub.v64i1(<64 x i1> %va, <64 x i1> %b, <64 x i1> %m, i32 %evl)
+  ret <64 x i1> %v
+}

llvm/test/CodeGen/RISCV/vadd-vp-mask.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -mattr=+v -verify-machineinstrs < %s \
+; RUN:   | FileCheck %s --check-prefixes=CHECK
+; RUN: llc -mtriple=riscv64 -mattr=+v -verify-machineinstrs < %s \
+; RUN:   | FileCheck %s --check-prefixes=CHECK
+
+
+declare <vscale x 2 x i1> @llvm.vp.add.nxv2i1(<vscale x 2 x i1>, <vscale x 2 x i1>, <vscale x 2 x i1>, i32)
+
+define <vscale x 2 x i1> @vadd_vv_nxv2i1(<vscale x 2 x i1> %va, <vscale x 2 x i1> %b, <vscale x 2 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vadd_vv_nxv2i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, mf4, ta, mu
+; CHECK-NEXT:    vmxor.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <vscale x 2 x i1> @llvm.vp.add.nxv2i1(<vscale x 2 x i1> %va, <vscale x 2 x i1> %b, <vscale x 2 x i1> %m, i32 %evl)
+  ret <vscale x 2 x i1> %v
+}
+
+declare <vscale x 4 x i1> @llvm.vp.add.nxv4i1(<vscale x 4 x i1>, <vscale x 4 x i1>, <vscale x 4 x i1>, i32)
+
+define <vscale x 4 x i1> @vadd_vv_nxv4i1(<vscale x 4 x i1> %va, <vscale x 4 x i1> %b, <vscale x 4 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vadd_vv_nxv4i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, mf2, ta, mu
+; CHECK-NEXT:    vmxor.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <vscale x 4 x i1> @llvm.vp.add.nxv4i1(<vscale x 4 x i1> %va, <vscale x 4 x i1> %b, <vscale x 4 x i1> %m, i32 %evl)
+  ret <vscale x 4 x i1> %v
+}
+
+declare <vscale x 8 x i1> @llvm.vp.add.nxv8i1(<vscale x 8 x i1>, <vscale x 8 x i1>, <vscale x 8 x i1>, i32)
+
+define <vscale x 8 x i1> @vadd_vv_nxv8i1(<vscale x 8 x i1> %va, <vscale x 8 x i1> %b, <vscale x 8 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vadd_vv_nxv8i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, m1, ta, mu
+; CHECK-NEXT:    vmxor.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <vscale x 8 x i1> @llvm.vp.add.nxv8i1(<vscale x 8 x i1> %va, <vscale x 8 x i1> %b, <vscale x 8 x i1> %m, i32 %evl)
+  ret <vscale x 8 x i1> %v
+}
+
+declare <vscale x 16 x i1> @llvm.vp.add.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>, <vscale x 16 x i1>, i32)
+
+define <vscale x 16 x i1> @vadd_vv_nxv16i1(<vscale x 16 x i1> %va, <vscale x 16 x i1> %b, <vscale x 16 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vadd_vv_nxv16i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, m2, ta, mu
+; CHECK-NEXT:    vmxor.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <vscale x 16 x i1> @llvm.vp.add.nxv16i1(<vscale x 16 x i1> %va, <vscale x 16 x i1> %b, <vscale x 16 x i1> %m, i32 %evl)
+  ret <vscale x 16 x i1> %v
+}
+
+declare <vscale x 32 x i1> @llvm.vp.add.nxv32i1(<vscale x 32 x i1>, <vscale x 32 x i1>, <vscale x 32 x i1>, i32)
+
+define <vscale x 32 x i1> @vadd_vv_nxv32i1(<vscale x 32 x i1> %va, <vscale x 32 x i1> %b, <vscale x 32 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vadd_vv_nxv32i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, m4, ta, mu
+; CHECK-NEXT:    vmxor.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <vscale x 32 x i1> @llvm.vp.add.nxv32i1(<vscale x 32 x i1> %va, <vscale x 32 x i1> %b, <vscale x 32 x i1> %m, i32 %evl)
+  ret <vscale x 32 x i1> %v
+}

llvm/test/CodeGen/RISCV/vmul-vp-mask.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -mattr=+v -verify-machineinstrs < %s \
+; RUN:   | FileCheck %s --check-prefixes=CHECK
+; RUN: llc -mtriple=riscv64 -mattr=+v -verify-machineinstrs < %s \
+; RUN:   | FileCheck %s --check-prefixes=CHECK
+
+
+declare <vscale x 2 x i1> @llvm.vp.mul.nxv2i1(<vscale x 2 x i1>, <vscale x 2 x i1>, <vscale x 2 x i1>, i32)
+
+define <vscale x 2 x i1> @vmul_vv_nxv2i1(<vscale x 2 x i1> %va, <vscale x 2 x i1> %b, <vscale x 2 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vmul_vv_nxv2i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, mf4, ta, mu
+; CHECK-NEXT:    vmand.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <vscale x 2 x i1> @llvm.vp.mul.nxv2i1(<vscale x 2 x i1> %va, <vscale x 2 x i1> %b, <vscale x 2 x i1> %m, i32 %evl)
+  ret <vscale x 2 x i1> %v
+}
+
+declare <vscale x 4 x i1> @llvm.vp.mul.nxv4i1(<vscale x 4 x i1>, <vscale x 4 x i1>, <vscale x 4 x i1>, i32)
+
+define <vscale x 4 x i1> @vmul_vv_nxv4i1(<vscale x 4 x i1> %va, <vscale x 4 x i1> %b, <vscale x 4 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vmul_vv_nxv4i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, mf2, ta, mu
+; CHECK-NEXT:    vmand.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <vscale x 4 x i1> @llvm.vp.mul.nxv4i1(<vscale x 4 x i1> %va, <vscale x 4 x i1> %b, <vscale x 4 x i1> %m, i32 %evl)
+  ret <vscale x 4 x i1> %v
+}
+
+declare <vscale x 8 x i1> @llvm.vp.mul.nxv8i1(<vscale x 8 x i1>, <vscale x 8 x i1>, <vscale x 8 x i1>, i32)
+
+define <vscale x 8 x i1> @vmul_vv_nxv8i1(<vscale x 8 x i1> %va, <vscale x 8 x i1> %b, <vscale x 8 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vmul_vv_nxv8i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, m1, ta, mu
+; CHECK-NEXT:    vmand.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <vscale x 8 x i1> @llvm.vp.mul.nxv8i1(<vscale x 8 x i1> %va, <vscale x 8 x i1> %b, <vscale x 8 x i1> %m, i32 %evl)
+  ret <vscale x 8 x i1> %v
+}
+
+declare <vscale x 16 x i1> @llvm.vp.mul.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>, <vscale x 16 x i1>, i32)
+
+define <vscale x 16 x i1> @vmul_vv_nxv16i1(<vscale x 16 x i1> %va, <vscale x 16 x i1> %b, <vscale x 16 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vmul_vv_nxv16i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, m2, ta, mu
+; CHECK-NEXT:    vmand.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <vscale x 16 x i1> @llvm.vp.mul.nxv16i1(<vscale x 16 x i1> %va, <vscale x 16 x i1> %b, <vscale x 16 x i1> %m, i32 %evl)
+  ret <vscale x 16 x i1> %v
+}
+
+declare <vscale x 32 x i1> @llvm.vp.mul.nxv32i1(<vscale x 32 x i1>, <vscale x 32 x i1>, <vscale x 32 x i1>, i32)
+
+define <vscale x 32 x i1> @vmul_vv_nxv32i1(<vscale x 32 x i1> %va, <vscale x 32 x i1> %b, <vscale x 32 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vmul_vv_nxv32i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, m4, ta, mu
+; CHECK-NEXT:    vmand.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <vscale x 32 x i1> @llvm.vp.mul.nxv32i1(<vscale x 32 x i1> %va, <vscale x 32 x i1> %b, <vscale x 32 x i1> %m, i32 %evl)
+  ret <vscale x 32 x i1> %v
+}

llvm/test/CodeGen/RISCV/vsub-vp-mask.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -mattr=+v -verify-machineinstrs < %s \
+; RUN:   | FileCheck %s --check-prefixes=CHECK
+; RUN: llc -mtriple=riscv64 -mattr=+v -verify-machineinstrs < %s \
+; RUN:   | FileCheck %s --check-prefixes=CHECK
+
+
+declare <vscale x 2 x i1> @llvm.vp.sub.nxv2i1(<vscale x 2 x i1>, <vscale x 2 x i1>, <vscale x 2 x i1>, i32)
+
+define <vscale x 2 x i1> @vsub_vv_nxv2i1(<vscale x 2 x i1> %va, <vscale x 2 x i1> %b, <vscale x 2 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vsub_vv_nxv2i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, mf4, ta, mu
+; CHECK-NEXT:    vmxor.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <vscale x 2 x i1> @llvm.vp.sub.nxv2i1(<vscale x 2 x i1> %va, <vscale x 2 x i1> %b, <vscale x 2 x i1> %m, i32 %evl)
+  ret <vscale x 2 x i1> %v
+}
+
+declare <vscale x 4 x i1> @llvm.vp.sub.nxv4i1(<vscale x 4 x i1>, <vscale x 4 x i1>, <vscale x 4 x i1>, i32)
+
+define <vscale x 4 x i1> @vsub_vv_nxv4i1(<vscale x 4 x i1> %va, <vscale x 4 x i1> %b, <vscale x 4 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vsub_vv_nxv4i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, mf2, ta, mu
+; CHECK-NEXT:    vmxor.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <vscale x 4 x i1> @llvm.vp.sub.nxv4i1(<vscale x 4 x i1> %va, <vscale x 4 x i1> %b, <vscale x 4 x i1> %m, i32 %evl)
+  ret <vscale x 4 x i1> %v
+}
+
+declare <vscale x 8 x i1> @llvm.vp.sub.nxv8i1(<vscale x 8 x i1>, <vscale x 8 x i1>, <vscale x 8 x i1>, i32)
+
+define <vscale x 8 x i1> @vsub_vv_nxv8i1(<vscale x 8 x i1> %va, <vscale x 8 x i1> %b, <vscale x 8 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vsub_vv_nxv8i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, m1, ta, mu
+; CHECK-NEXT:    vmxor.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <vscale x 8 x i1> @llvm.vp.sub.nxv8i1(<vscale x 8 x i1> %va, <vscale x 8 x i1> %b, <vscale x 8 x i1> %m, i32 %evl)
+  ret <vscale x 8 x i1> %v
+}
+
+declare <vscale x 16 x i1> @llvm.vp.sub.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>, <vscale x 16 x i1>, i32)
+
+define <vscale x 16 x i1> @vsub_vv_nxv16i1(<vscale x 16 x i1> %va, <vscale x 16 x i1> %b, <vscale x 16 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vsub_vv_nxv16i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, m2, ta, mu
+; CHECK-NEXT:    vmxor.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <vscale x 16 x i1> @llvm.vp.sub.nxv16i1(<vscale x 16 x i1> %va, <vscale x 16 x i1> %b, <vscale x 16 x i1> %m, i32 %evl)
+  ret <vscale x 16 x i1> %v
+}
+
+declare <vscale x 32 x i1> @llvm.vp.sub.nxv32i1(<vscale x 32 x i1>, <vscale x 32 x i1>, <vscale x 32 x i1>, i32)
+
+define <vscale x 32 x i1> @vsub_vv_nxv32i1(<vscale x 32 x i1> %va, <vscale x 32 x i1> %b, <vscale x 32 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vsub_vv_nxv32i1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e8, m4, ta, mu
+; CHECK-NEXT:    vmxor.mm v0, v0, v8
+; CHECK-NEXT:    ret
+  %v = call <vscale x 32 x i1> @llvm.vp.sub.nxv32i1(<vscale x 32 x i1> %va, <vscale x 32 x i1> %b, <vscale x 32 x i1> %m, i32 %evl)
+  ret <vscale x 32 x i1> %v
+}