[VE] Add vmv intrinsic instructions

[kaz7]

Summary

Add vmv intrinsic instructions and regression tests.

llvm/include/llvm/IR/IntrinsicsVEVL.gen.td

 let TargetPrefix = "ve" in def int_ve_vl_vbrds_vsmvl : GCCBuiltin<"__builtin_ve_vl_vbrds_vsmvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<f32>, LLVMType<v256i1>, LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_vbrdw_vsl : GCCBuiltin<"__builtin_ve_vl_vbrdw_vsl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<i32>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_vbrdw_vsvl : GCCBuiltin<"__builtin_ve_vl_vbrdw_vsvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<i32>, LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_vbrdw_vsmvl : GCCBuiltin<"__builtin_ve_vl_vbrdw_vsmvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<i32>, LLVMType<v256i1>, LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_pvbrd_vsl : GCCBuiltin<"__builtin_ve_vl_pvbrd_vsl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<i64>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_pvbrd_vsvl : GCCBuiltin<"__builtin_ve_vl_pvbrd_vsvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<i64>, LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_pvbrd_vsMvl : GCCBuiltin<"__builtin_ve_vl_pvbrd_vsMvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<i64>, LLVMType<v512i1>, LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_vmv_vsvl : GCCBuiltin<"__builtin_ve_vl_vmv_vsvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<i32>, LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
+let TargetPrefix = "ve" in def int_ve_vl_vmv_vsvvl : GCCBuiltin<"__builtin_ve_vl_vmv_vsvvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<i32>, LLVMType<v256f64>, LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
+let TargetPrefix = "ve" in def int_ve_vl_vmv_vsvmvl : GCCBuiltin<"__builtin_ve_vl_vmv_vsvmvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<i32>, LLVMType<v256f64>, LLVMType<v256i1>, LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;

llvm/lib/Target/VE/VEInstrIntrinsicVL.gen.td

 def : Pat<(int_ve_vl_vbrdw_vsvl i32:$sy, v256f64:$pt, i32:$vl), (VBRDLrl_v i32:$sy, i32:$vl, v256f64:$pt)>;
 def : Pat<(int_ve_vl_vbrdw_vsmvl i32:$sy, v256i1:$vm, v256f64:$pt, i32:$vl), (VBRDLrml_v i32:$sy, v256i1:$vm, i32:$vl, v256f64:$pt)>;
 def : Pat<(int_ve_vl_vbrdw_vsl simm7:$I, i32:$vl), (VBRDLil (LO7 $I), i32:$vl)>;
 def : Pat<(int_ve_vl_vbrdw_vsvl simm7:$I, v256f64:$pt, i32:$vl), (VBRDLil_v (LO7 $I), i32:$vl, v256f64:$pt)>;
 def : Pat<(int_ve_vl_vbrdw_vsmvl simm7:$I, v256i1:$vm, v256f64:$pt, i32:$vl), (VBRDLiml_v (LO7 $I), v256i1:$vm, i32:$vl, v256f64:$pt)>;
 def : Pat<(int_ve_vl_pvbrd_vsl i64:$sy, i32:$vl), (PVBRDrl i64:$sy, i32:$vl)>;
 def : Pat<(int_ve_vl_pvbrd_vsvl i64:$sy, v256f64:$pt, i32:$vl), (PVBRDrl_v i64:$sy, i32:$vl, v256f64:$pt)>;
 def : Pat<(int_ve_vl_pvbrd_vsMvl i64:$sy, v512i1:$vm, v256f64:$pt, i32:$vl), (PVBRDrml_v i64:$sy, v512i1:$vm, i32:$vl, v256f64:$pt)>;
+def : Pat<(int_ve_vl_vmv_vsvl uimm7:$N, v256f64:$vz, i32:$vl), (VMVivl (ULO7 $N), v256f64:$vz, i32:$vl)>;
+def : Pat<(int_ve_vl_vmv_vsvvl uimm7:$N, v256f64:$vz, v256f64:$pt, i32:$vl), (VMVivl_v (ULO7 $N), v256f64:$vz, i32:$vl, v256f64:$pt)>;
+def : Pat<(int_ve_vl_vmv_vsvmvl uimm7:$N, v256f64:$vz, v256i1:$vm, v256f64:$pt, i32:$vl), (VMVivml_v (ULO7 $N), v256f64:$vz, v256i1:$vm, i32:$vl, v256f64:$pt)>;

llvm/test/CodeGen/VE/VELIntrinsics/vmv.ll

+; RUN: llc < %s -mtriple=ve -mattr=+vpu | FileCheck %s
+
+;;; Test vector move intrinsic instructions
+;;;
+;;; Note:
+;;;   We test VMVivl and VMVivl_v, and VMVivml_v instructions.
+
+; Function Attrs: nounwind
+define void @vmv_vsvl(i8* %0, i64 %1, i32 signext %2) {
+; CHECK-LABEL: vmv_vsvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s1, 256
+; CHECK-NEXT:    lvl %s1
+; CHECK-NEXT:    vld %v0, 8, %s0
+; CHECK-NEXT:    vmv %v0, 31, %v0
+; CHECK-NEXT:    vst %v0, 8, %s0
+; CHECK-NEXT:    b.l.t (, %s10)
+  %4 = tail call fast <256 x double> @llvm.ve.vl.vld.vssl(i64 8, i8* %0, i32 256)
+  %5 = tail call fast <256 x double> @llvm.ve.vl.vmv.vsvl(i32 31, <256 x double> %4, i32 256)
+  tail call void @llvm.ve.vl.vst.vssl(<256 x double> %5, i64 8, i8* %0, i32 256)
+  ret void
+}
+
+; Function Attrs: nounwind readonly
+declare <256 x double> @llvm.ve.vl.vld.vssl(i64, i8*, i32)
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vmv.vsvl(i32, <256 x double>, i32)
+
+; Function Attrs: nounwind writeonly
+declare void @llvm.ve.vl.vst.vssl(<256 x double>, i64, i8*, i32)
+
+; Function Attrs: nounwind
+define void @vmv_vsvvl(i8* %0, i32 signext %1) {
+; CHECK-LABEL: vmv_vsvvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s1, 256
+; CHECK-NEXT:    lvl %s1
+; CHECK-NEXT:    vld %v0, 8, %s0
+; CHECK-NEXT:    lea %s2, 128
+; CHECK-NEXT:    lvl %s2
+; CHECK-NEXT:    vmv %v0, 31, %v0
+; CHECK-NEXT:    lvl %s1
+; CHECK-NEXT:    vst %v0, 8, %s0
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vld.vssl(i64 8, i8* %0, i32 256)
+  %4 = tail call fast <256 x double> @llvm.ve.vl.vmv.vsvvl(i32 31, <256 x double> %3, <256 x double> %3, i32 128)
+  tail call void @llvm.ve.vl.vst.vssl(<256 x double> %4, i64 8, i8* %0, i32 256)
+  ret void
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vmv.vsvvl(i32, <256 x double>, <256 x double>, i32)
+
+; Function Attrs: nounwind
+define void @vmv_vsvmvl(i8* %0, i32 signext %1) {
+; CHECK-LABEL: vmv_vsvmvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s1, 256
+; CHECK-NEXT:    lvl %s1
+; CHECK-NEXT:    vld %v0, 8, %s0
+; CHECK-NEXT:    lea %s2, 128
+; CHECK-NEXT:    lvl %s2
+; CHECK-NEXT:    vmv %v0, 31, %v0, %vm1
+; CHECK-NEXT:    lvl %s1
+; CHECK-NEXT:    vst %v0, 8, %s0
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vld.vssl(i64 8, i8* %0, i32 256)
+  %4 = tail call fast <256 x double> @llvm.ve.vl.vmv.vsvmvl(i32 31, <256 x double> %3, <256 x i1> undef, <256 x double> %3, i32 128)
+  tail call void @llvm.ve.vl.vst.vssl(<256 x double> %4, i64 8, i8* %0, i32 256)
+  ret void
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vmv.vsvmvl(i32, <256 x double>, <256 x i1>, <256 x double>, i32)