[VE] Add vcp and vex intrinsic instructions

[kaz7]

Summary

Add vcp and vex intrinsic instructions and regression tests.

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

 let TargetPrefix = "ve" in def int_ve_vl_vmrg_vsvml : GCCBuiltin<"__builtin_ve_vl_vmrg_vsvml">, Intrinsic<[LLVMType<v256f64>], [LLVMType<i64>, LLVMType<v256f64>, LLVMType<v256i1>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_vmrg_vsvmvl : GCCBuiltin<"__builtin_ve_vl_vmrg_vsvmvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<i64>, LLVMType<v256f64>, LLVMType<v256i1>, LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_vmrgw_vvvMl : GCCBuiltin<"__builtin_ve_vl_vmrgw_vvvMl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<v256f64>, LLVMType<v256f64>, LLVMType<v512i1>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_vmrgw_vvvMvl : GCCBuiltin<"__builtin_ve_vl_vmrgw_vvvMvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<v256f64>, LLVMType<v256f64>, LLVMType<v512i1>, LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_vmrgw_vsvMl : GCCBuiltin<"__builtin_ve_vl_vmrgw_vsvMl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<i32>, LLVMType<v256f64>, LLVMType<v512i1>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_vmrgw_vsvMvl : GCCBuiltin<"__builtin_ve_vl_vmrgw_vsvMvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<i32>, LLVMType<v256f64>, LLVMType<v512i1>, LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_vshf_vvvsl : GCCBuiltin<"__builtin_ve_vl_vshf_vvvsl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<v256f64>, LLVMType<v256f64>, LLVMType<i64>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_vshf_vvvsvl : GCCBuiltin<"__builtin_ve_vl_vshf_vvvsvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<v256f64>, LLVMType<v256f64>, LLVMType<i64>, LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
+let TargetPrefix = "ve" in def int_ve_vl_vcp_vvmvl : GCCBuiltin<"__builtin_ve_vl_vcp_vvmvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<v256f64>, LLVMType<v256i1>, LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
+let TargetPrefix = "ve" in def int_ve_vl_vex_vvmvl : GCCBuiltin<"__builtin_ve_vl_vex_vvmvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<v256f64>, LLVMType<v256i1>, LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;

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

 def : Pat<(int_ve_vl_vmrg_vsvml simm7:$I, v256f64:$vz, v256i1:$vm, i32:$vl), (VMRGivml (LO7 $I), v256f64:$vz, v256i1:$vm, i32:$vl)>;
 def : Pat<(int_ve_vl_vmrg_vsvmvl simm7:$I, v256f64:$vz, v256i1:$vm, v256f64:$pt, i32:$vl), (VMRGivml_v (LO7 $I), v256f64:$vz, v256i1:$vm, i32:$vl, v256f64:$pt)>;
 def : Pat<(int_ve_vl_vmrgw_vvvMl v256f64:$vy, v256f64:$vz, v512i1:$vm, i32:$vl), (VMRGWvvml v256f64:$vy, v256f64:$vz, v512i1:$vm, i32:$vl)>;
 def : Pat<(int_ve_vl_vmrgw_vvvMvl v256f64:$vy, v256f64:$vz, v512i1:$vm, v256f64:$pt, i32:$vl), (VMRGWvvml_v v256f64:$vy, v256f64:$vz, v512i1:$vm, i32:$vl, v256f64:$pt)>;
 def : Pat<(int_ve_vl_vshf_vvvsl v256f64:$vy, v256f64:$vz, i64:$sy, i32:$vl), (VSHFvvrl v256f64:$vy, v256f64:$vz, i64:$sy, i32:$vl)>;
 def : Pat<(int_ve_vl_vshf_vvvsvl v256f64:$vy, v256f64:$vz, i64:$sy, v256f64:$pt, i32:$vl), (VSHFvvrl_v v256f64:$vy, v256f64:$vz, i64:$sy, i32:$vl, v256f64:$pt)>;
 def : Pat<(int_ve_vl_vshf_vvvsl v256f64:$vy, v256f64:$vz, uimm6:$N, i32:$vl), (VSHFvvil v256f64:$vy, v256f64:$vz, (ULO7 $N), i32:$vl)>;
 def : Pat<(int_ve_vl_vshf_vvvsvl v256f64:$vy, v256f64:$vz, uimm6:$N, v256f64:$pt, i32:$vl), (VSHFvvil_v v256f64:$vy, v256f64:$vz, (ULO7 $N), i32:$vl, v256f64:$pt)>;
+def : Pat<(int_ve_vl_vcp_vvmvl v256f64:$vz, v256i1:$vm, v256f64:$pt, i32:$vl), (VCPvml_v v256f64:$vz, v256i1:$vm, i32:$vl, v256f64:$pt)>;
+def : Pat<(int_ve_vl_vex_vvmvl v256f64:$vz, v256i1:$vm, v256f64:$pt, i32:$vl), (VEXvml_v v256f64:$vz, v256i1:$vm, i32:$vl, v256f64:$pt)>;

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

+; RUN: llc < %s -mtriple=ve -mattr=+vpu | FileCheck %s
+
+;;; Test vector compress intrinsic instructions
+;;;
+;;; Note:
+;;;   We test VCP*vml_v instruction.
+
+; Function Attrs: nounwind readnone
+define fastcc <256 x double> @vcp_vvmvl(<256 x double> %0, <256 x i1> %1, <256 x double> %2) {
+; CHECK-LABEL: vcp_vvmvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s0, 128
+; CHECK-NEXT:    lvl %s0
+; CHECK-NEXT:    vcp %v1, %v0, %vm1
+; CHECK-NEXT:    lea %s16, 256
+; CHECK-NEXT:    lvl %s16
+; CHECK-NEXT:    vor %v0, (0)1, %v1
+; CHECK-NEXT:    b.l.t (, %s10)
+  %4 = tail call fast <256 x double> @llvm.ve.vl.vcp.vvmvl(<256 x double> %0, <256 x i1> %1, <256 x double> %2, i32 128)
+  ret <256 x double> %4
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vcp.vvmvl(<256 x double>, <256 x i1>, <256 x double>, i32)

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

+; RUN: llc < %s -mtriple=ve -mattr=+vpu | FileCheck %s
+
+;;; Test vector expand intrinsic instructions
+;;;
+;;; Note:
+;;;   We test VEX*vml_v instruction.
+
+; Function Attrs: nounwind readnone
+define fastcc <256 x double> @vex_vvmvl(<256 x double> %0, <256 x i1> %1, <256 x double> %2) {
+; CHECK-LABEL: vex_vvmvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s0, 128
+; CHECK-NEXT:    lvl %s0
+; CHECK-NEXT:    vex %v1, %v0, %vm1
+; CHECK-NEXT:    lea %s16, 256
+; CHECK-NEXT:    lvl %s16
+; CHECK-NEXT:    vor %v0, (0)1, %v1
+; CHECK-NEXT:    b.l.t (, %s10)
+  %4 = tail call fast <256 x double> @llvm.ve.vl.vex.vvmvl(<256 x double> %0, <256 x i1> %1, <256 x double> %2, i32 128)
+  ret <256 x double> %4
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vex.vvmvl(<256 x double>, <256 x i1>, <256 x double>, i32)