[VE] Add vbrd intrinsic instructions

[kaz7]

Summary

Add vbrd intrinsic instructions and a regression test.

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

 let TargetPrefix = "ve" in def int_ve_vl_lsv_vvss : GCCBuiltin<"__builtin_ve_vl_lsv_vvss">, Intrinsic<[LLVMType<v256f64>], [LLVMType<v256f64>, LLVMType<i32>, LLVMType<i64>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_lvsl_svs : GCCBuiltin<"__builtin_ve_vl_lvsl_svs">, Intrinsic<[LLVMType<i64>], [LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_lvsd_svs : GCCBuiltin<"__builtin_ve_vl_lvsd_svs">, Intrinsic<[LLVMType<f64>], [LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_lvss_svs : GCCBuiltin<"__builtin_ve_vl_lvss_svs">, Intrinsic<[LLVMType<f32>], [LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_lvm_mmss : GCCBuiltin<"__builtin_ve_vl_lvm_mmss">, Intrinsic<[LLVMType<v256i1>], [LLVMType<v256i1>, LLVMType<i64>, LLVMType<i64>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_lvm_MMss : GCCBuiltin<"__builtin_ve_vl_lvm_MMss">, Intrinsic<[LLVMType<v512i1>], [LLVMType<v512i1>, LLVMType<i64>, LLVMType<i64>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_svm_sms : GCCBuiltin<"__builtin_ve_vl_svm_sms">, Intrinsic<[LLVMType<i64>], [LLVMType<v256i1>, LLVMType<i64>], [IntrNoMem]>;
 let TargetPrefix = "ve" in def int_ve_vl_svm_sMs : GCCBuiltin<"__builtin_ve_vl_svm_sMs">, Intrinsic<[LLVMType<i64>], [LLVMType<v512i1>, LLVMType<i64>], [IntrNoMem]>;
+let TargetPrefix = "ve" in def int_ve_vl_vbrdd_vsl : GCCBuiltin<"__builtin_ve_vl_vbrdd_vsl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<f64>, LLVMType<i32>], [IntrNoMem]>;
+let TargetPrefix = "ve" in def int_ve_vl_vbrdd_vsvl : GCCBuiltin<"__builtin_ve_vl_vbrdd_vsvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<f64>, LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
+let TargetPrefix = "ve" in def int_ve_vl_vbrdd_vsmvl : GCCBuiltin<"__builtin_ve_vl_vbrdd_vsmvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<f64>, LLVMType<v256i1>, LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
+let TargetPrefix = "ve" in def int_ve_vl_vbrdl_vsl : GCCBuiltin<"__builtin_ve_vl_vbrdl_vsl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<i64>, LLVMType<i32>], [IntrNoMem]>;
+let TargetPrefix = "ve" in def int_ve_vl_vbrdl_vsvl : GCCBuiltin<"__builtin_ve_vl_vbrdl_vsvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<i64>, LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
+let TargetPrefix = "ve" in def int_ve_vl_vbrdl_vsmvl : GCCBuiltin<"__builtin_ve_vl_vbrdl_vsmvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<i64>, LLVMType<v256i1>, LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
+let TargetPrefix = "ve" in def int_ve_vl_vbrds_vsl : GCCBuiltin<"__builtin_ve_vl_vbrds_vsl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<f32>, LLVMType<i32>], [IntrNoMem]>;
+let TargetPrefix = "ve" in def int_ve_vl_vbrds_vsvl : GCCBuiltin<"__builtin_ve_vl_vbrds_vsvl">, Intrinsic<[LLVMType<v256f64>], [LLVMType<f32>, LLVMType<v256f64>, LLVMType<i32>], [IntrNoMem]>;
+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]>;

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

 def : Pat<(int_ve_vl_pfchv_ssl i64:$sy, i64:$sz, i32:$vl), (PFCHVrrl i64:$sy, i64:$sz, i32:$vl)>;
 def : Pat<(int_ve_vl_pfchv_ssl simm7:$I, i64:$sz, i32:$vl), (PFCHVirl (LO7 $I), i64:$sz, i32:$vl)>;
 def : Pat<(int_ve_vl_pfchvnc_ssl i64:$sy, i64:$sz, i32:$vl), (PFCHVNCrrl i64:$sy, i64:$sz, i32:$vl)>;
 def : Pat<(int_ve_vl_pfchvnc_ssl simm7:$I, i64:$sz, i32:$vl), (PFCHVNCirl (LO7 $I), i64:$sz, i32:$vl)>;
 def : Pat<(int_ve_vl_lvm_mmss v256i1:$ptm, uimm6:$N, i64:$sz), (LVMir_m (ULO7 $N), i64:$sz, v256i1:$ptm)>;
 def : Pat<(int_ve_vl_lvm_MMss v512i1:$ptm, uimm6:$N, i64:$sz), (LVMyir_y (ULO7 $N), i64:$sz, v512i1:$ptm)>;
 def : Pat<(int_ve_vl_svm_sms v256i1:$vmz, uimm6:$N), (SVMmi v256i1:$vmz, (ULO7 $N))>;
 def : Pat<(int_ve_vl_svm_sMs v512i1:$vmz, uimm6:$N), (SVMyi v512i1:$vmz, (ULO7 $N))>;
+def : Pat<(int_ve_vl_vbrdd_vsl f64:$sy, i32:$vl), (VBRDrl f64:$sy, i32:$vl)>;
+def : Pat<(int_ve_vl_vbrdd_vsvl f64:$sy, v256f64:$pt, i32:$vl), (VBRDrl_v f64:$sy, i32:$vl, v256f64:$pt)>;
+def : Pat<(int_ve_vl_vbrdd_vsmvl f64:$sy, v256i1:$vm, v256f64:$pt, i32:$vl), (VBRDrml_v f64:$sy, v256i1:$vm, i32:$vl, v256f64:$pt)>;
+def : Pat<(int_ve_vl_vbrdl_vsl i64:$sy, i32:$vl), (VBRDrl i64:$sy, i32:$vl)>;
+def : Pat<(int_ve_vl_vbrdl_vsvl i64:$sy, v256f64:$pt, i32:$vl), (VBRDrl_v i64:$sy, i32:$vl, v256f64:$pt)>;
+def : Pat<(int_ve_vl_vbrdl_vsmvl i64:$sy, v256i1:$vm, v256f64:$pt, i32:$vl), (VBRDrml_v i64:$sy, v256i1:$vm, i32:$vl, v256f64:$pt)>;
+def : Pat<(int_ve_vl_vbrdl_vsl simm7:$I, i32:$vl), (VBRDil (LO7 $I), i32:$vl)>;
+def : Pat<(int_ve_vl_vbrdl_vsvl simm7:$I, v256f64:$pt, i32:$vl), (VBRDil_v (LO7 $I), i32:$vl, v256f64:$pt)>;
+def : Pat<(int_ve_vl_vbrdl_vsmvl simm7:$I, v256i1:$vm, v256f64:$pt, i32:$vl), (VBRDiml_v (LO7 $I), v256i1:$vm, i32:$vl, v256f64:$pt)>;
+def : Pat<(int_ve_vl_vbrds_vsl f32:$sy, i32:$vl), (VBRDUrl f32:$sy, i32:$vl)>;
+def : Pat<(int_ve_vl_vbrds_vsvl f32:$sy, v256f64:$pt, i32:$vl), (VBRDUrl_v f32:$sy, i32:$vl, v256f64:$pt)>;
+def : Pat<(int_ve_vl_vbrds_vsmvl f32:$sy, v256i1:$vm, v256f64:$pt, i32:$vl), (VBRDUrml_v f32:$sy, v256i1:$vm, i32:$vl, v256f64:$pt)>;
+def : Pat<(int_ve_vl_vbrdw_vsl i32:$sy, i32:$vl), (VBRDLrl i32:$sy, i32:$vl)>;
+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)>;

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

+; RUN: llc < %s -mtriple=ve -mattr=+vpu | FileCheck %s
+
+;;; Test vector broadcast intrinsic instructions
+;;;
+;;; Note:
+;;;   We test VLD*rrl, VLD*irl, VLD*rrl_v, and VLD*irl_v instructions.
+
+; Function Attrs: nounwind
+define void @vbrdd_vsl(double %0, i8* %1) {
+; CHECK-LABEL: vbrdd_vsl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s2, 256
+; CHECK-NEXT:    lvl %s2
+; CHECK-NEXT:    vbrd %v0, %s0
+; CHECK-NEXT:    #APP
+; CHECK-NEXT:    vst %v0, 8, %s1
+; CHECK-NEXT:    #NO_APP
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vbrdd.vsl(double %0, i32 256)
+  tail call void asm sideeffect "vst ${0:v}, 8, $1", "v,r"(<256 x double> %3, i8* %1)
+  ret void
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vbrdd.vsl(double, i32)
+
+; Function Attrs: nounwind
+define void @vbrdd_vsvl(double %0, i8* %1) {
+; CHECK-LABEL: vbrdd_vsvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s2, 256
+; CHECK-NEXT:    lvl %s2
+; CHECK-NEXT:    vld %v0, 8, %s1
+; CHECK-NEXT:    vbrd %v0, %s0
+; CHECK-NEXT:    vst %v0, 8, %s1
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vld.vssl(i64 8, i8* %1, i32 256)
+  %4 = tail call fast <256 x double> @llvm.ve.vl.vbrdd.vsvl(double %0, <256 x double> %3, i32 256)
+  tail call void @llvm.ve.vl.vst.vssl(<256 x double> %4, i64 8, i8* %1, 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.vbrdd.vsvl(double, <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 @vbrdd_vsmvl(double %0, i8* %1) {
+; CHECK-LABEL: vbrdd_vsmvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s2, 256
+; CHECK-NEXT:    lvl %s2
+; CHECK-NEXT:    vld %v0, 8, %s1
+; CHECK-NEXT:    lea.sl %s3, 1138753536
+; CHECK-NEXT:    fcmp.d %s4, %s0, %s3
+; CHECK-NEXT:    fsub.d %s3, %s0, %s3
+; CHECK-NEXT:    cvt.l.d.rz %s3, %s3
+; CHECK-NEXT:    xor %s3, %s3, (1)1
+; CHECK-NEXT:    cvt.l.d.rz %s5, %s0
+; CHECK-NEXT:    cmov.d.lt %s3, %s5, %s4
+; CHECK-NEXT:    lvm %vm1, 3, %s3
+; CHECK-NEXT:    vbrd %v0, %s0, %vm1
+; CHECK-NEXT:    vst %v0, 8, %s1
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vld.vssl(i64 8, i8* %1, i32 256)
+  %4 = fptoui double %0 to i64
+  %5 = tail call <256 x i1> @llvm.ve.vl.lvm.mmss(<256 x i1> undef, i64 3, i64 %4)
+  %6 = tail call fast <256 x double> @llvm.ve.vl.vbrdd.vsmvl(double %0, <256 x i1> %5, <256 x double> %3, i32 256)
+  tail call void @llvm.ve.vl.vst.vssl(<256 x double> %6, i64 8, i8* %1, i32 256)
+  ret void
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x i1> @llvm.ve.vl.lvm.mmss(<256 x i1>, i64, i64)
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vbrdd.vsmvl(double, <256 x i1>, <256 x double>, i32)
+
+; Function Attrs: nounwind
+define void @vbrdl_vsl(i64 %0, i8* %1) {
+; CHECK-LABEL: vbrdl_vsl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s2, 256
+; CHECK-NEXT:    lvl %s2
+; CHECK-NEXT:    vbrd %v0, %s0
+; CHECK-NEXT:    #APP
+; CHECK-NEXT:    vst %v0, 8, %s1
+; CHECK-NEXT:    #NO_APP
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vbrdl.vsl(i64 %0, i32 256)
+  tail call void asm sideeffect "vst ${0:v}, 8, $1", "v,r"(<256 x double> %3, i8* %1)
+  ret void
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vbrdl.vsl(i64, i32)
+
+; Function Attrs: nounwind
+define void @vbrdl_vsvl(i64 %0, i8* %1) {
+; CHECK-LABEL: vbrdl_vsvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s2, 256
+; CHECK-NEXT:    lvl %s2
+; CHECK-NEXT:    vld %v0, 8, %s1
+; CHECK-NEXT:    vbrd %v0, %s0
+; CHECK-NEXT:    vst %v0, 8, %s1
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vld.vssl(i64 8, i8* %1, i32 256)
+  %4 = tail call fast <256 x double> @llvm.ve.vl.vbrdl.vsvl(i64 %0, <256 x double> %3, i32 256)
+  tail call void @llvm.ve.vl.vst.vssl(<256 x double> %4, i64 8, i8* %1, i32 256)
+  ret void
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vbrdl.vsvl(i64, <256 x double>, i32)
+
+; Function Attrs: nounwind
+define void @vbrdl_vsmvl(i64 %0, i8* %1) {
+; CHECK-LABEL: vbrdl_vsmvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s2, 256
+; CHECK-NEXT:    lvl %s2
+; CHECK-NEXT:    vld %v0, 8, %s1
+; CHECK-NEXT:    lvm %vm1, 3, %s0
+; CHECK-NEXT:    vbrd %v0, %s0, %vm1
+; CHECK-NEXT:    vst %v0, 8, %s1
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vld.vssl(i64 8, i8* %1, i32 256)
+  %4 = tail call <256 x i1> @llvm.ve.vl.lvm.mmss(<256 x i1> undef, i64 3, i64 %0)
+  %5 = tail call fast <256 x double> @llvm.ve.vl.vbrdl.vsmvl(i64 %0, <256 x i1> %4, <256 x double> %3, i32 256)
+  tail call void @llvm.ve.vl.vst.vssl(<256 x double> %5, i64 8, i8* %1, i32 256)
+  ret void
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vbrdl.vsmvl(i64, <256 x i1>, <256 x double>, i32)
+
+; Function Attrs: nounwind
+define void @vbrdl_imm_vsl(i64 %0, i8* %1) {
+; CHECK-LABEL: vbrdl_imm_vsl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s0, 256
+; CHECK-NEXT:    lvl %s0
+; CHECK-NEXT:    vbrd %v0, 31
+; CHECK-NEXT:    #APP
+; CHECK-NEXT:    vst %v0, 8, %s1
+; CHECK-NEXT:    #NO_APP
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vbrdl.vsl(i64 31, i32 256)
+  tail call void asm sideeffect "vst ${0:v}, 8, $1", "v,r"(<256 x double> %3, i8* %1)
+  ret void
+}
+
+; Function Attrs: nounwind
+define void @vbrdl_imm_vsvl(i64 %0, i8* %1) {
+; CHECK-LABEL: vbrdl_imm_vsvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s0, 256
+; CHECK-NEXT:    lvl %s0
+; CHECK-NEXT:    vld %v0, 8, %s1
+; CHECK-NEXT:    vbrd %v0, 31
+; CHECK-NEXT:    vst %v0, 8, %s1
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vld.vssl(i64 8, i8* %1, i32 256)
+  %4 = tail call fast <256 x double> @llvm.ve.vl.vbrdl.vsvl(i64 31, <256 x double> %3, i32 256)
+  tail call void @llvm.ve.vl.vst.vssl(<256 x double> %4, i64 8, i8* %1, i32 256)
+  ret void
+}
+
+; Function Attrs: nounwind
+define void @vbrdl_imm_vsmvl(i64 %0, i8* %1) {
+; CHECK-LABEL: vbrdl_imm_vsmvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s2, 256
+; CHECK-NEXT:    lvl %s2
+; CHECK-NEXT:    vld %v0, 8, %s1
+; CHECK-NEXT:    lvm %vm1, 3, %s0
+; CHECK-NEXT:    vbrd %v0, 31, %vm1
+; CHECK-NEXT:    vst %v0, 8, %s1
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vld.vssl(i64 8, i8* %1, i32 256)
+  %4 = tail call <256 x i1> @llvm.ve.vl.lvm.mmss(<256 x i1> undef, i64 3, i64 %0)
+  %5 = tail call fast <256 x double> @llvm.ve.vl.vbrdl.vsmvl(i64 31, <256 x i1> %4, <256 x double> %3, i32 256)
+  tail call void @llvm.ve.vl.vst.vssl(<256 x double> %5, i64 8, i8* %1, i32 256)
+  ret void
+}
+
+; Function Attrs: nounwind
+define void @vbrds_vsl(float %0, i8* %1) {
+; CHECK-LABEL: vbrds_vsl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s2, 256
+; CHECK-NEXT:    lvl %s2
+; CHECK-NEXT:    vbrdu %v0, %s0
+; CHECK-NEXT:    #APP
+; CHECK-NEXT:    vst %v0, 8, %s1
+; CHECK-NEXT:    #NO_APP
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vbrds.vsl(float %0, i32 256)
+  tail call void asm sideeffect "vst ${0:v}, 8, $1", "v,r"(<256 x double> %3, i8* %1)
+  ret void
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vbrds.vsl(float, i32)
+
+; Function Attrs: nounwind
+define void @vbrds_vsvl(float %0, i8* %1) {
+; CHECK-LABEL: vbrds_vsvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s2, 256
+; CHECK-NEXT:    lvl %s2
+; CHECK-NEXT:    vld %v0, 8, %s1
+; CHECK-NEXT:    vbrdu %v0, %s0
+; CHECK-NEXT:    vst %v0, 8, %s1
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vld.vssl(i64 8, i8* %1, i32 256)
+  %4 = tail call fast <256 x double> @llvm.ve.vl.vbrds.vsvl(float %0, <256 x double> %3, i32 256)
+  tail call void @llvm.ve.vl.vst.vssl(<256 x double> %4, i64 8, i8* %1, i32 256)
+  ret void
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vbrds.vsvl(float, <256 x double>, i32)
+
+; Function Attrs: nounwind
+define void @vbrds_vsmvl(float %0, i8* %1) {
+; CHECK-LABEL: vbrds_vsmvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s2, 256
+; CHECK-NEXT:    lvl %s2
+; CHECK-NEXT:    vld %v0, 8, %s1
+; CHECK-NEXT:    lea.sl %s3, 1593835520
+; CHECK-NEXT:    fcmp.s %s4, %s0, %s3
+; CHECK-NEXT:    fsub.s %s3, %s0, %s3
+; CHECK-NEXT:    cvt.d.s %s3, %s3
+; CHECK-NEXT:    cvt.l.d.rz %s3, %s3
+; CHECK-NEXT:    xor %s3, %s3, (1)1
+; CHECK-NEXT:    cvt.d.s %s5, %s0
+; CHECK-NEXT:    cvt.l.d.rz %s5, %s5
+; CHECK-NEXT:    cmov.s.lt %s3, %s5, %s4
+; CHECK-NEXT:    lvm %vm1, 3, %s3
+; CHECK-NEXT:    vbrdu %v0, %s0, %vm1
+; CHECK-NEXT:    vst %v0, 8, %s1
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vld.vssl(i64 8, i8* %1, i32 256)
+  %4 = fptoui float %0 to i64
+  %5 = tail call <256 x i1> @llvm.ve.vl.lvm.mmss(<256 x i1> undef, i64 3, i64 %4)
+  %6 = tail call fast <256 x double> @llvm.ve.vl.vbrds.vsmvl(float %0, <256 x i1> %5, <256 x double> %3, i32 256)
+  tail call void @llvm.ve.vl.vst.vssl(<256 x double> %6, i64 8, i8* %1, i32 256)
+  ret void
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vbrds.vsmvl(float, <256 x i1>, <256 x double>, i32)
+
+; Function Attrs: nounwind
+define void @vbrdw_vsl(i32 signext %0, i8* %1) {
+; CHECK-LABEL: vbrdw_vsl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    adds.w.sx %s0, %s0, (0)1
+; CHECK-NEXT:    lea %s2, 256
+; CHECK-NEXT:    lvl %s2
+; CHECK-NEXT:    vbrdl %v0, %s0
+; CHECK-NEXT:    #APP
+; CHECK-NEXT:    vst %v0, 8, %s1
+; CHECK-NEXT:    #NO_APP
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vbrdw.vsl(i32 %0, i32 256)
+  tail call void asm sideeffect "vst ${0:v}, 8, $1", "v,r"(<256 x double> %3, i8* %1)
+  ret void
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vbrdw.vsl(i32, i32)
+
+; Function Attrs: nounwind
+define void @vbrdw_vsvl(i32 signext %0, i8* %1) {
+; CHECK-LABEL: vbrdw_vsvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s2, 256
+; CHECK-NEXT:    lvl %s2
+; CHECK-NEXT:    vld %v0, 8, %s1
+; CHECK-NEXT:    adds.w.sx %s0, %s0, (0)1
+; CHECK-NEXT:    vbrdl %v0, %s0
+; CHECK-NEXT:    vst %v0, 8, %s1
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vld.vssl(i64 8, i8* %1, i32 256)
+  %4 = tail call fast <256 x double> @llvm.ve.vl.vbrdw.vsvl(i32 %0, <256 x double> %3, i32 256)
+  tail call void @llvm.ve.vl.vst.vssl(<256 x double> %4, i64 8, i8* %1, i32 256)
+  ret void
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vbrdw.vsvl(i32, <256 x double>, i32)
+
+; Function Attrs: nounwind
+define void @vbrdw_vsmvl(i32 signext %0, i8* %1) {
+; CHECK-LABEL: vbrdw_vsmvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s2, 256
+; CHECK-NEXT:    lvl %s2
+; CHECK-NEXT:    vld %v0, 8, %s1
+; CHECK-NEXT:    adds.w.sx %s3, %s0, (0)1
+; CHECK-NEXT:    lvm %vm1, 3, %s0
+; CHECK-NEXT:    vbrdl %v0, %s3, %vm1
+; CHECK-NEXT:    vst %v0, 8, %s1
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vld.vssl(i64 8, i8* %1, i32 256)
+  %4 = sext i32 %0 to i64
+  %5 = tail call <256 x i1> @llvm.ve.vl.lvm.mmss(<256 x i1> undef, i64 3, i64 %4)
+  %6 = tail call fast <256 x double> @llvm.ve.vl.vbrdw.vsmvl(i32 %0, <256 x i1> %5, <256 x double> %3, i32 256)
+  tail call void @llvm.ve.vl.vst.vssl(<256 x double> %6, i64 8, i8* %1, i32 256)
+  ret void
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.vbrdw.vsmvl(i32, <256 x i1>, <256 x double>, i32)
+
+; Function Attrs: nounwind
+define void @vbrdw_imm_vsl(i32 signext %0, i8* %1) {
+; CHECK-LABEL: vbrdw_imm_vsl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s0, 256
+; CHECK-NEXT:    lvl %s0
+; CHECK-NEXT:    vbrdl %v0, 31
+; CHECK-NEXT:    #APP
+; CHECK-NEXT:    vst %v0, 8, %s1
+; CHECK-NEXT:    #NO_APP
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vbrdw.vsl(i32 31, i32 256)
+  tail call void asm sideeffect "vst ${0:v}, 8, $1", "v,r"(<256 x double> %3, i8* %1)
+  ret void
+}
+
+; Function Attrs: nounwind
+define void @vbrdw_imm_vsvl(i32 signext %0, i8* %1) {
+; CHECK-LABEL: vbrdw_imm_vsvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s0, 256
+; CHECK-NEXT:    lvl %s0
+; CHECK-NEXT:    vld %v0, 8, %s1
+; CHECK-NEXT:    vbrdl %v0, 31
+; CHECK-NEXT:    vst %v0, 8, %s1
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vld.vssl(i64 8, i8* %1, i32 256)
+  %4 = tail call fast <256 x double> @llvm.ve.vl.vbrdw.vsvl(i32 31, <256 x double> %3, i32 256)
+  tail call void @llvm.ve.vl.vst.vssl(<256 x double> %4, i64 8, i8* %1, i32 256)
+  ret void
+}
+
+; Function Attrs: nounwind
+define void @vbrdw_imm_vsmvl(i32 signext %0, i8* %1) {
+; CHECK-LABEL: vbrdw_imm_vsmvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s2, 256
+; CHECK-NEXT:    lvl %s2
+; CHECK-NEXT:    vld %v0, 8, %s1
+; CHECK-NEXT:    lvm %vm1, 3, %s0
+; CHECK-NEXT:    vbrdl %v0, 31, %vm1
+; CHECK-NEXT:    vst %v0, 8, %s1
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vld.vssl(i64 8, i8* %1, i32 256)
+  %4 = sext i32 %0 to i64
+  %5 = tail call <256 x i1> @llvm.ve.vl.lvm.mmss(<256 x i1> undef, i64 3, i64 %4)
+  %6 = tail call fast <256 x double> @llvm.ve.vl.vbrdw.vsmvl(i32 31, <256 x i1> %5, <256 x double> %3, i32 256)
+  tail call void @llvm.ve.vl.vst.vssl(<256 x double> %6, i64 8, i8* %1, i32 256)
+  ret void
+}
+
+; Function Attrs: nounwind
+define void @pvbrd_vsl(i64 %0, i8* %1) {
+; CHECK-LABEL: pvbrd_vsl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s2, 256
+; CHECK-NEXT:    lvl %s2
+; CHECK-NEXT:    pvbrd %v0, %s0
+; CHECK-NEXT:    #APP
+; CHECK-NEXT:    vst %v0, 8, %s1
+; CHECK-NEXT:    #NO_APP
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.pvbrd.vsl(i64 %0, i32 256)
+  tail call void asm sideeffect "vst ${0:v}, 8, $1", "v,r"(<256 x double> %3, i8* %1)
+  ret void
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.pvbrd.vsl(i64, i32)
+
+; Function Attrs: nounwind
+define void @pvbrd_vsvl(i64 %0, i8* %1) {
+; CHECK-LABEL: pvbrd_vsvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s2, 256
+; CHECK-NEXT:    lvl %s2
+; CHECK-NEXT:    vld %v0, 8, %s1
+; CHECK-NEXT:    pvbrd %v0, %s0
+; CHECK-NEXT:    vst %v0, 8, %s1
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vld.vssl(i64 8, i8* %1, i32 256)
+  %4 = tail call fast <256 x double> @llvm.ve.vl.pvbrd.vsvl(i64 %0, <256 x double> %3, i32 256)
+  tail call void @llvm.ve.vl.vst.vssl(<256 x double> %4, i64 8, i8* %1, i32 256)
+  ret void
+}
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.pvbrd.vsvl(i64, <256 x double>, i32)
+
+; Function Attrs: nounwind
+define void @pvbrd_vsMvl(i64 %0, i8* %1) {
+; CHECK-LABEL: pvbrd_vsMvl:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lea %s2, 256
+; CHECK-NEXT:    lvl %s2
+; CHECK-NEXT:    vld %v0, 8, %s1
+; CHECK-NEXT:    lvm %vm3, 1, %s0
+; CHECK-NEXT:    lvm %vm2, 2, %s0
+; CHECK-NEXT:    pvbrd %v0, %s0, %vm2
+; CHECK-NEXT:    vst %v0, 8, %s1
+; CHECK-NEXT:    b.l.t (, %s10)
+  %3 = tail call fast <256 x double> @llvm.ve.vl.vld.vssl(i64 8, i8* %1, i32 256)
+  %4 = tail call <512 x i1> @llvm.ve.vl.lvm.MMss(<512 x i1> undef, i64 1, i64 %0)
+  %5 = tail call <512 x i1> @llvm.ve.vl.lvm.MMss(<512 x i1> %4, i64 6, i64 %0)
+  %6 = tail call fast <256 x double> @llvm.ve.vl.pvbrd.vsMvl(i64 %0, <512 x i1> %5, <256 x double> %3, i32 256)
+  tail call void @llvm.ve.vl.vst.vssl(<256 x double> %6, i64 8, i8* %1, i32 256)
+  ret void
+}
+
+; Function Attrs: nounwind readnone
+declare <512 x i1> @llvm.ve.vl.lvm.MMss(<512 x i1>, i64, i64)
+
+; Function Attrs: nounwind readnone
+declare <256 x double> @llvm.ve.vl.pvbrd.vsMvl(i64, <512 x i1>, <256 x double>, i32)