[VE] Add vsum and vfsum intrinsic instructions

[kaz7]

Summary

Add vsum and vfsum intrinsic instructions and regression tests.

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

 let TargetPrefix = "ve" in def int_ve_vl_pvfmksnenan_Mvl : GCCBuiltin<"__builtin_ve_vl_pvfmksnenan_Mvl">, Intrinsic<[LLVMType<v512i1>], [LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_pvfmksnenan_MvMl : GCCBuiltin<"__builtin_ve_vl_pvfmksnenan_MvMl">, Intrinsic<[LLVMType<v512i1>], [LLVMType<v256f64>, LLVMType<v512i1>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_pvfmkseqnan_Mvl : GCCBuiltin<"__builtin_ve_vl_pvfmkseqnan_Mvl">, Intrinsic<[LLVMType<v512i1>], [LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_pvfmkseqnan_MvMl : GCCBuiltin<"__builtin_ve_vl_pvfmkseqnan_MvMl">, Intrinsic<[LLVMType<v512i1>], [LLVMType<v256f64>, LLVMType<v512i1>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_pvfmksgenan_Mvl : GCCBuiltin<"__builtin_ve_vl_pvfmksgenan_Mvl">, Intrinsic<[LLVMType<v512i1>], [LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_pvfmksgenan_MvMl : GCCBuiltin<"__builtin_ve_vl_pvfmksgenan_MvMl">, Intrinsic<[LLVMType<v512i1>], [LLVMType<v256f64>, LLVMType<v512i1>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_pvfmkslenan_Mvl : GCCBuiltin<"__builtin_ve_vl_pvfmkslenan_Mvl">, Intrinsic<[LLVMType<v512i1>], [LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_pvfmkslenan_MvMl : GCCBuiltin<"__builtin_ve_vl_pvfmkslenan_MvMl">, Intrinsic<[LLVMType<v512i1>], [LLVMType<v256f64>, LLVMType<v512i1>, LLVMType<i32>], [IntrNoMem]>;
+let TargetPrefix = "ve" in def int_ve_vl_vsumwsx_vvl : GCCBuiltin<"__builtin_ve_vl_vsumwsx_vvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
+let TargetPrefix = "ve" in def int_ve_vl_vsumwsx_vvml : GCCBuiltin<"__builtin_ve_vl_vsumwsx_vvml">, Intrinsic<[LLVMType<v256f64>], [LLVMType<v256f64>, LLVMType<v256i1>, LLVMType<i32>], [IntrNoMem]>;
+let TargetPrefix = "ve" in def int_ve_vl_vsumwzx_vvl : GCCBuiltin<"__builtin_ve_vl_vsumwzx_vvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
+let TargetPrefix = "ve" in def int_ve_vl_vsumwzx_vvml : GCCBuiltin<"__builtin_ve_vl_vsumwzx_vvml">, Intrinsic<[LLVMType<v256f64>], [LLVMType<v256f64>, LLVMType<v256i1>, LLVMType<i32>], [IntrNoMem]>;
+let TargetPrefix = "ve" in def int_ve_vl_vsuml_vvl : GCCBuiltin<"__builtin_ve_vl_vsuml_vvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
+let TargetPrefix = "ve" in def int_ve_vl_vsuml_vvml : GCCBuiltin<"__builtin_ve_vl_vsuml_vvml">, Intrinsic<[LLVMType<v256f64>], [LLVMType<v256f64>, LLVMType<v256i1>, LLVMType<i32>], [IntrNoMem]>;
+let TargetPrefix = "ve" in def int_ve_vl_vfsumd_vvl : GCCBuiltin<"__builtin_ve_vl_vfsumd_vvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
+let TargetPrefix = "ve" in def int_ve_vl_vfsumd_vvml : GCCBuiltin<"__builtin_ve_vl_vfsumd_vvml">, Intrinsic<[LLVMType<v256f64>], [LLVMType<v256f64>, LLVMType<v256i1>, LLVMType<i32>], [IntrNoMem]>;
+let TargetPrefix = "ve" in def int_ve_vl_vfsums_vvl : GCCBuiltin<"__builtin_ve_vl_vfsums_vvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
+let TargetPrefix = "ve" in def int_ve_vl_vfsums_vvml : GCCBuiltin<"__builtin_ve_vl_vfsums_vvml">, Intrinsic<[LLVMType<v256f64>], [LLVMType<v256f64>, LLVMType<v256i1>, LLVMType<i32>], [IntrNoMem]>;

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

 def : Pat<(int_ve_vl_pvfmksnenan_Mvl v256f64:$vz, i32:$vl), (VFMKSyvl CC_NENAN, v256f64:$vz, i32:$vl)>;
 def : Pat<(int_ve_vl_pvfmksnenan_MvMl v256f64:$vz, v512i1:$vm, i32:$vl), (VFMKSyvyl CC_NENAN, v256f64:$vz, v512i1:$vm, i32:$vl)>;
 def : Pat<(int_ve_vl_pvfmkseqnan_Mvl v256f64:$vz, i32:$vl), (VFMKSyvl CC_EQNAN, v256f64:$vz, i32:$vl)>;
 def : Pat<(int_ve_vl_pvfmkseqnan_MvMl v256f64:$vz, v512i1:$vm, i32:$vl), (VFMKSyvyl CC_EQNAN, v256f64:$vz, v512i1:$vm, i32:$vl)>;
 def : Pat<(int_ve_vl_pvfmksgenan_Mvl v256f64:$vz, i32:$vl), (VFMKSyvl CC_GENAN, v256f64:$vz, i32:$vl)>;
 def : Pat<(int_ve_vl_pvfmksgenan_MvMl v256f64:$vz, v512i1:$vm, i32:$vl), (VFMKSyvyl CC_GENAN, v256f64:$vz, v512i1:$vm, i32:$vl)>;
 def : Pat<(int_ve_vl_pvfmkslenan_Mvl v256f64:$vz, i32:$vl), (VFMKSyvl CC_LENAN, v256f64:$vz, i32:$vl)>;
 def : Pat<(int_ve_vl_pvfmkslenan_MvMl v256f64:$vz, v512i1:$vm, i32:$vl), (VFMKSyvyl CC_LENAN, v256f64:$vz, v512i1:$vm, i32:$vl)>;
+def : Pat<(int_ve_vl_vsumwsx_vvl v256f64:$vy, i32:$vl), (VSUMWSXvl v256f64:$vy, i32:$vl)>;
+def : Pat<(int_ve_vl_vsumwsx_vvml v256f64:$vy, v256i1:$vm, i32:$vl), (VSUMWSXvml v256f64:$vy, v256i1:$vm, i32:$vl)>;
+def : Pat<(int_ve_vl_vsumwzx_vvl v256f64:$vy, i32:$vl), (VSUMWZXvl v256f64:$vy, i32:$vl)>;
+def : Pat<(int_ve_vl_vsumwzx_vvml v256f64:$vy, v256i1:$vm, i32:$vl), (VSUMWZXvml v256f64:$vy, v256i1:$vm, i32:$vl)>;
+def : Pat<(int_ve_vl_vsuml_vvl v256f64:$vy, i32:$vl), (VSUMLvl v256f64:$vy, i32:$vl)>;
+def : Pat<(int_ve_vl_vsuml_vvml v256f64:$vy, v256i1:$vm, i32:$vl), (VSUMLvml v256f64:$vy, v256i1:$vm, i32:$vl)>;
+def : Pat<(int_ve_vl_vfsumd_vvl v256f64:$vy, i32:$vl), (VFSUMDvl v256f64:$vy, i32:$vl)>;
+def : Pat<(int_ve_vl_vfsumd_vvml v256f64:$vy, v256i1:$vm, i32:$vl), (VFSUMDvml v256f64:$vy, v256i1:$vm, i32:$vl)>;
+def : Pat<(int_ve_vl_vfsums_vvl v256f64:$vy, i32:$vl), (VFSUMSvl v256f64:$vy, i32:$vl)>;
+def : Pat<(int_ve_vl_vfsums_vvml v256f64:$vy, v256i1:$vm, i32:$vl), (VFSUMSvml v256f64:$vy, v256i1:$vm, i32:$vl)>;

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

+; RUN: llc < %s -mtriple=ve -mattr=+vpu | FileCheck %s
+
+;;; Test vector floating sum intrinsic instructions
+;;;
+;;; Note:
+;;;   We test VFSUM*vl and VFSUM*vml instructions.
+
+; Function Attrs: nounwind readnone
+define fastcc <256 x double> @vfsumd_vvl(<256 x double> %0) {
+; CHECK-LABEL: vfsumd_vvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s0, 256
+; CHECK-NEXT:    lvl %s0
+; CHECK-NEXT:    vfsum.d %v0, %v0
+; CHECK-NEXT:    b.l.t (, %s10)
+  %2 = tail call fast <256 x double> @llvm.ve.vl.vfsumd.vvl(<256 x double> %0, i32 256)
+  ret <256 x double> %2
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vfsumd.vvl(<256 x double>, i32)
+
+; Function Attrs: nounwind readnone
+define fastcc <256 x double> @vfsumd_vvml(<256 x double> %0, <256 x i1> %1) {
+; CHECK-LABEL: vfsumd_vvml:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s0, 256
+; CHECK-NEXT:    lvl %s0
+; CHECK-NEXT:    vfsum.d %v0, %v0, %vm1
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vfsumd.vvml(<256 x double> %0, <256 x i1> %1, i32 256)
+  ret <256 x double> %3
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vfsumd.vvml(<256 x double>, <256 x i1>, i32)
+
+; Function Attrs: nounwind readnone
+define fastcc <256 x double> @vfsums_vvl(<256 x double> %0) {
+; CHECK-LABEL: vfsums_vvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s0, 256
+; CHECK-NEXT:    lvl %s0
+; CHECK-NEXT:    vfsum.s %v0, %v0
+; CHECK-NEXT:    b.l.t (, %s10)
+  %2 = tail call fast <256 x double> @llvm.ve.vl.vfsums.vvl(<256 x double> %0, i32 256)
+  ret <256 x double> %2
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vfsums.vvl(<256 x double>, i32)
+
+; Function Attrs: nounwind readnone
+define fastcc <256 x double> @vfsums_vvml(<256 x double> %0, <256 x i1> %1) {
+; CHECK-LABEL: vfsums_vvml:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s0, 256
+; CHECK-NEXT:    lvl %s0
+; CHECK-NEXT:    vfsum.s %v0, %v0, %vm1
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vfsums.vvml(<256 x double> %0, <256 x i1> %1, i32 256)
+  ret <256 x double> %3
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vfsums.vvml(<256 x double>, <256 x i1>, i32)

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

+; RUN: llc < %s -mtriple=ve -mattr=+vpu | FileCheck %s
+
+;;; Test vector sum intrinsic instructions
+;;;
+;;; Note:
+;;;   We test VSUM*vl and VSUM*vml instructions.
+
+; Function Attrs: nounwind readnone
+define fastcc <256 x double> @vsumwsx_vvl(<256 x double> %0) {
+; CHECK-LABEL: vsumwsx_vvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s0, 256
+; CHECK-NEXT:    lvl %s0
+; CHECK-NEXT:    vsum.w.sx %v0, %v0
+; CHECK-NEXT:    b.l.t (, %s10)
+  %2 = tail call fast <256 x double> @llvm.ve.vl.vsumwsx.vvl(<256 x double> %0, i32 256)
+  ret <256 x double> %2
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vsumwsx.vvl(<256 x double>, i32)
+
+; Function Attrs: nounwind readnone
+define fastcc <256 x double> @vsumwsx_vvml(<256 x double> %0, <256 x i1> %1) {
+; CHECK-LABEL: vsumwsx_vvml:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s0, 256
+; CHECK-NEXT:    lvl %s0
+; CHECK-NEXT:    vsum.w.sx %v0, %v0, %vm1
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vsumwsx.vvml(<256 x double> %0, <256 x i1> %1, i32 256)
+  ret <256 x double> %3
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vsumwsx.vvml(<256 x double>, <256 x i1>, i32)
+
+; Function Attrs: nounwind readnone
+define fastcc <256 x double> @vsumwzx_vvl(<256 x double> %0) {
+; CHECK-LABEL: vsumwzx_vvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s0, 256
+; CHECK-NEXT:    lvl %s0
+; CHECK-NEXT:    vsum.w.zx %v0, %v0
+; CHECK-NEXT:    b.l.t (, %s10)
+  %2 = tail call fast <256 x double> @llvm.ve.vl.vsumwzx.vvl(<256 x double> %0, i32 256)
+  ret <256 x double> %2
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vsumwzx.vvl(<256 x double>, i32)
+
+; Function Attrs: nounwind readnone
+define fastcc <256 x double> @vsumwzx_vvml(<256 x double> %0, <256 x i1> %1) {
+; CHECK-LABEL: vsumwzx_vvml:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s0, 256
+; CHECK-NEXT:    lvl %s0
+; CHECK-NEXT:    vsum.w.zx %v0, %v0, %vm1
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vsumwzx.vvml(<256 x double> %0, <256 x i1> %1, i32 256)
+  ret <256 x double> %3
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vsumwzx.vvml(<256 x double>, <256 x i1>, i32)
+
+; Function Attrs: nounwind readnone
+define fastcc <256 x double> @vsuml_vvl(<256 x double> %0) {
+; CHECK-LABEL: vsuml_vvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s0, 256
+; CHECK-NEXT:    lvl %s0
+; CHECK-NEXT:    vsum.l %v0, %v0
+; CHECK-NEXT:    b.l.t (, %s10)
+  %2 = tail call fast <256 x double> @llvm.ve.vl.vsuml.vvl(<256 x double> %0, i32 256)
+  ret <256 x double> %2
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vsuml.vvl(<256 x double>, i32)
+
+; Function Attrs: nounwind readnone
+define fastcc <256 x double> @vsuml_vvml(<256 x double> %0, <256 x i1> %1) {
+; CHECK-LABEL: vsuml_vvml:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s0, 256
+; CHECK-NEXT:    lvl %s0
+; CHECK-NEXT:    vsum.l %v0, %v0, %vm1
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vsuml.vvml(<256 x double> %0, <256 x i1> %1, i32 256)
+  ret <256 x double> %3
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vsuml.vvml(<256 x double>, <256 x i1>, i32)