[Clang][VE] Add vector load intrinsics

Add vector load intrinsic instructions for VE.

Reviewed By: simoll

Differential Revision: https://reviews.llvm.org/D121049

10 files changed, 369 insertions, 2 deletions

diff --git a/clang/include/clang/Basic/BuiltinsVE.def b/clang/include/clang/Basic/BuiltinsVE.def
new file mode 100644
index 000000000000..1cb525089381
--- /dev/null
+++ b/clang/include/clang/Basic/BuiltinsVE.def
@@ -0,0 +1,17 @@
+//===--- BuiltinsVE.def - VE Builtin function database ----------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the VE-specific builtin function database.  Users of
+// this file must define the BUILTIN macro to make use of this information.
+//
+//===----------------------------------------------------------------------===//
+
+// Use generated BUILTIN definitions
+#include "clang/Basic/BuiltinsVEVL.gen.def"
+
+#undef BUILTIN
diff --git a/clang/include/clang/Basic/BuiltinsVEVL.gen.def b/clang/include/clang/Basic/BuiltinsVEVL.gen.def
new file mode 100644
index 000000000000..d90e35e91745
--- /dev/null
+++ b/clang/include/clang/Basic/BuiltinsVEVL.gen.def
@@ -0,0 +1,32 @@
+BUILTIN(__builtin_ve_vl_vld_vssl, "V256dLUivC*Ui", "n")
+BUILTIN(__builtin_ve_vl_vld_vssvl, "V256dLUivC*V256dUi", "n")
+BUILTIN(__builtin_ve_vl_vldnc_vssl, "V256dLUivC*Ui", "n")
+BUILTIN(__builtin_ve_vl_vldnc_vssvl, "V256dLUivC*V256dUi", "n")
+BUILTIN(__builtin_ve_vl_vldu_vssl, "V256dLUivC*Ui", "n")
+BUILTIN(__builtin_ve_vl_vldu_vssvl, "V256dLUivC*V256dUi", "n")
+BUILTIN(__builtin_ve_vl_vldunc_vssl, "V256dLUivC*Ui", "n")
+BUILTIN(__builtin_ve_vl_vldunc_vssvl, "V256dLUivC*V256dUi", "n")
+BUILTIN(__builtin_ve_vl_vldlsx_vssl, "V256dLUivC*Ui", "n")
+BUILTIN(__builtin_ve_vl_vldlsx_vssvl, "V256dLUivC*V256dUi", "n")
+BUILTIN(__builtin_ve_vl_vldlsxnc_vssl, "V256dLUivC*Ui", "n")
+BUILTIN(__builtin_ve_vl_vldlsxnc_vssvl, "V256dLUivC*V256dUi", "n")
+BUILTIN(__builtin_ve_vl_vldlzx_vssl, "V256dLUivC*Ui", "n")
+BUILTIN(__builtin_ve_vl_vldlzx_vssvl, "V256dLUivC*V256dUi", "n")
+BUILTIN(__builtin_ve_vl_vldlzxnc_vssl, "V256dLUivC*Ui", "n")
+BUILTIN(__builtin_ve_vl_vldlzxnc_vssvl, "V256dLUivC*V256dUi", "n")
+BUILTIN(__builtin_ve_vl_vld2d_vssl, "V256dLUivC*Ui", "n")
+BUILTIN(__builtin_ve_vl_vld2d_vssvl, "V256dLUivC*V256dUi", "n")
+BUILTIN(__builtin_ve_vl_vld2dnc_vssl, "V256dLUivC*Ui", "n")
+BUILTIN(__builtin_ve_vl_vld2dnc_vssvl, "V256dLUivC*V256dUi", "n")
+BUILTIN(__builtin_ve_vl_vldu2d_vssl, "V256dLUivC*Ui", "n")
+BUILTIN(__builtin_ve_vl_vldu2d_vssvl, "V256dLUivC*V256dUi", "n")
+BUILTIN(__builtin_ve_vl_vldu2dnc_vssl, "V256dLUivC*Ui", "n")
+BUILTIN(__builtin_ve_vl_vldu2dnc_vssvl, "V256dLUivC*V256dUi", "n")
+BUILTIN(__builtin_ve_vl_vldl2dsx_vssl, "V256dLUivC*Ui", "n")
+BUILTIN(__builtin_ve_vl_vldl2dsx_vssvl, "V256dLUivC*V256dUi", "n")
+BUILTIN(__builtin_ve_vl_vldl2dsxnc_vssl, "V256dLUivC*Ui", "n")
+BUILTIN(__builtin_ve_vl_vldl2dsxnc_vssvl, "V256dLUivC*V256dUi", "n")
+BUILTIN(__builtin_ve_vl_vldl2dzx_vssl, "V256dLUivC*Ui", "n")
+BUILTIN(__builtin_ve_vl_vldl2dzx_vssvl, "V256dLUivC*V256dUi", "n")
+BUILTIN(__builtin_ve_vl_vldl2dzxnc_vssl, "V256dLUivC*Ui", "n")
+BUILTIN(__builtin_ve_vl_vldl2dzxnc_vssvl, "V256dLUivC*V256dUi", "n")
diff --git a/clang/include/clang/Basic/TargetBuiltins.h b/clang/include/clang/Basic/TargetBuiltins.h
index d4ea8e98b2e3..d8ad9858d8c8 100644
--- a/clang/include/clang/Basic/TargetBuiltins.h
+++ b/clang/include/clang/Basic/TargetBuiltins.h
@@ -121,7 +121,12 @@ namespace clang {
 
   /// VE builtins
   namespace VE {
-  enum { LastTIBuiltin = clang::Builtin::FirstTSBuiltin - 1, LastTSBuiltin };
+  enum {
+    LastTIBuiltin = clang::Builtin::FirstTSBuiltin - 1,
+#define BUILTIN(ID, TYPE, ATTRS) BI##ID,
+#include "clang/Basic/BuiltinsVE.def"
+    LastTSBuiltin
+  };
   }
 
   namespace RISCVVector {
diff --git a/clang/include/clang/module.modulemap b/clang/include/clang/module.modulemap
index 2b73cd5451b7..981f32aa25a1 100644
--- a/clang/include/clang/module.modulemap
+++ b/clang/include/clang/module.modulemap
@@ -50,6 +50,7 @@ module Clang_Basic {
   textual header "Basic/BuiltinsRISCVVector.def"
   textual header "Basic/BuiltinsSVE.def"
   textual header "Basic/BuiltinsSystemZ.def"
+  textual header "Basic/BuiltinsVE.def"
   textual header "Basic/BuiltinsWebAssembly.def"
   textual header "Basic/BuiltinsX86.def"
   textual header "Basic/BuiltinsX86_64.def"
diff --git a/clang/lib/Basic/Targets/VE.cpp b/clang/lib/Basic/Targets/VE.cpp
index 22223654e8ad..4d66c98edc92 100644
--- a/clang/lib/Basic/Targets/VE.cpp
+++ b/clang/lib/Basic/Targets/VE.cpp
@@ -18,6 +18,12 @@
 using namespace clang;
 using namespace clang::targets;
 
+const Builtin::Info VETargetInfo::BuiltinInfo[] = {
+#define BUILTIN(ID, TYPE, ATTRS)                                               \
+  {#ID, TYPE, ATTRS, nullptr, ALL_LANGUAGES, nullptr},
+#include "clang/Basic/BuiltinsVE.def"
+};
+
 void VETargetInfo::getTargetDefines(const LangOptions &Opts,
                                     MacroBuilder &Builder) const {
   Builder.defineMacro("_LP64", "1");
@@ -35,5 +41,6 @@ void VETargetInfo::getTargetDefines(const LangOptions &Opts,
 }
 
 ArrayRef<Builtin::Info> VETargetInfo::getTargetBuiltins() const {
-  return ArrayRef<Builtin::Info>();
+  return llvm::makeArrayRef(BuiltinInfo,
+                            clang::VE::LastTSBuiltin - Builtin::FirstTSBuiltin);
 }
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 26d7d041ae2a..a2699f5b3ea1 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -45,6 +45,7 @@
 #include "llvm/IR/IntrinsicsR600.h"
 #include "llvm/IR/IntrinsicsRISCV.h"
 #include "llvm/IR/IntrinsicsS390.h"
+#include "llvm/IR/IntrinsicsVE.h"
 #include "llvm/IR/IntrinsicsWebAssembly.h"
 #include "llvm/IR/IntrinsicsX86.h"
 #include "llvm/IR/MDBuilder.h"
diff --git a/clang/lib/Headers/CMakeLists.txt b/clang/lib/Headers/CMakeLists.txt
index 078988980c52..447b1400b58f 100644
--- a/clang/lib/Headers/CMakeLists.txt
+++ b/clang/lib/Headers/CMakeLists.txt
@@ -145,6 +145,8 @@ set(files
   xsaveoptintrin.h
   xsavesintrin.h
   xtestintrin.h
+  velintrin.h
+  velintrin_gen.h
   )
 
 set(cuda_wrapper_files
diff --git a/clang/lib/Headers/velintrin.h b/clang/lib/Headers/velintrin.h
new file mode 100644
index 000000000000..5988384b8786
--- /dev/null
+++ b/clang/lib/Headers/velintrin.h
@@ -0,0 +1,38 @@
+/*===---- velintrin.h - VEL intrinsics for VE ------------------------------===
+ *
+ * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+ * See https://llvm.org/LICENSE.txt for license information.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ *
+ *===-----------------------------------------------------------------------===
+ */
+#ifndef __VEL_INTRIN_H__
+#define __VEL_INTRIN_H__
+
+// Vector registers
+typedef double __vr __attribute__((__vector_size__(2048)));
+
+// TODO: Vector mask registers
+// Depend on https://reviews.llvm.org/D88905
+#if 0
+#if __STDC_VERSION__ >= 199901L
+// For C99
+typedef _Bool __vm    __attribute__((ext_vector_type(256)));
+typedef _Bool __vm256 __attribute__((ext_vector_type(256)));
+typedef _Bool __vm512 __attribute__((ext_vector_type(512)));
+#else
+#ifdef __cplusplus
+// For C++
+typedef bool __vm    __attribute__((ext_vector_type(256)));
+typedef bool __vm256 __attribute__((ext_vector_type(256)));
+typedef bool __vm512 __attribute__((ext_vector_type(512)));
+#else
+#error need C++ or C99 to use vector intrinsics for VE
+#endif
+#endif
+#endif
+
+// Use generated intrinsic name definitions
+#include <velintrin_gen.h>
+
+#endif
diff --git a/clang/lib/Headers/velintrin_gen.h b/clang/lib/Headers/velintrin_gen.h
new file mode 100644
index 000000000000..23d23573a7fe
--- /dev/null
+++ b/clang/lib/Headers/velintrin_gen.h
@@ -0,0 +1,32 @@
+#define _vel_vld_vssl __builtin_ve_vl_vld_vssl
+#define _vel_vld_vssvl __builtin_ve_vl_vld_vssvl
+#define _vel_vldnc_vssl __builtin_ve_vl_vldnc_vssl
+#define _vel_vldnc_vssvl __builtin_ve_vl_vldnc_vssvl
+#define _vel_vldu_vssl __builtin_ve_vl_vldu_vssl
+#define _vel_vldu_vssvl __builtin_ve_vl_vldu_vssvl
+#define _vel_vldunc_vssl __builtin_ve_vl_vldunc_vssl
+#define _vel_vldunc_vssvl __builtin_ve_vl_vldunc_vssvl
+#define _vel_vldlsx_vssl __builtin_ve_vl_vldlsx_vssl
+#define _vel_vldlsx_vssvl __builtin_ve_vl_vldlsx_vssvl
+#define _vel_vldlsxnc_vssl __builtin_ve_vl_vldlsxnc_vssl
+#define _vel_vldlsxnc_vssvl __builtin_ve_vl_vldlsxnc_vssvl
+#define _vel_vldlzx_vssl __builtin_ve_vl_vldlzx_vssl
+#define _vel_vldlzx_vssvl __builtin_ve_vl_vldlzx_vssvl
+#define _vel_vldlzxnc_vssl __builtin_ve_vl_vldlzxnc_vssl
+#define _vel_vldlzxnc_vssvl __builtin_ve_vl_vldlzxnc_vssvl
+#define _vel_vld2d_vssl __builtin_ve_vl_vld2d_vssl
+#define _vel_vld2d_vssvl __builtin_ve_vl_vld2d_vssvl
+#define _vel_vld2dnc_vssl __builtin_ve_vl_vld2dnc_vssl
+#define _vel_vld2dnc_vssvl __builtin_ve_vl_vld2dnc_vssvl
+#define _vel_vldu2d_vssl __builtin_ve_vl_vldu2d_vssl
+#define _vel_vldu2d_vssvl __builtin_ve_vl_vldu2d_vssvl
+#define _vel_vldu2dnc_vssl __builtin_ve_vl_vldu2dnc_vssl
+#define _vel_vldu2dnc_vssvl __builtin_ve_vl_vldu2dnc_vssvl
+#define _vel_vldl2dsx_vssl __builtin_ve_vl_vldl2dsx_vssl
+#define _vel_vldl2dsx_vssvl __builtin_ve_vl_vldl2dsx_vssvl
+#define _vel_vldl2dsxnc_vssl __builtin_ve_vl_vldl2dsxnc_vssl
+#define _vel_vldl2dsxnc_vssvl __builtin_ve_vl_vldl2dsxnc_vssvl
+#define _vel_vldl2dzx_vssl __builtin_ve_vl_vldl2dzx_vssl
+#define _vel_vldl2dzx_vssvl __builtin_ve_vl_vldl2dzx_vssvl
+#define _vel_vldl2dzxnc_vssl __builtin_ve_vl_vldl2dzxnc_vssl
+#define _vel_vldl2dzxnc_vssvl __builtin_ve_vl_vldl2dzxnc_vssvl
diff --git a/clang/test/CodeGen/VE/ve-velintrin.c b/clang/test/CodeGen/VE/ve-velintrin.c
new file mode 100644
index 000000000000..51f7c5d35912
--- /dev/null
+++ b/clang/test/CodeGen/VE/ve-velintrin.c
@@ -0,0 +1,232 @@
+// REQUIRES: ve-registered-target
+
+// RUN: %clang_cc1 -S -emit-llvm -triple ve-unknown-linux-gnu \
+// RUN:   -ffreestanding %s -o - | FileCheck %s
+
+#include <velintrin.h>
+
+__vr vr1;
+
+void __attribute__((noinline))
+test_vld_vssl(char* p, long idx) {
+  // CHECK-LABEL: @test_vld_vssl
+  // CHECK: call <256 x double> @llvm.ve.vl.vld.vssl(i64 %{{.*}}, i8* %{{.*}}, i32 256)
+  vr1 = _vel_vld_vssl(idx, p, 256);
+}
+
+void __attribute__((noinline))
+test_vld_vssvl(char* p, long idx) {
+  // CHECK-LABEL: @test_vld_vssvl
+  // CHECK: call <256 x double> @llvm.ve.vl.vld.vssvl(i64 %{{.*}}, i8* %{{.*}}, <256 x double> %{{.*}}, i32 256)
+  vr1 = _vel_vld_vssvl(idx, p, vr1, 256);
+}
+
+void __attribute__((noinline))
+test_vldnc_vssl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldnc_vssl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldnc.vssl(i64 %{{.*}}, i8* %{{.*}}, i32 256)
+  vr1 = _vel_vldnc_vssl(idx, p, 256);
+}
+
+void __attribute__((noinline))
+test_vldnc_vssvl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldnc_vssvl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldnc.vssvl(i64 %{{.*}}, i8* %{{.*}}, <256 x double> %{{.*}}, i32 256)
+  vr1 = _vel_vldnc_vssvl(idx, p, vr1, 256);
+}
+
+void __attribute__((noinline))
+test_vldu_vssl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldu_vssl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldu.vssl(i64 %{{.*}}, i8* %{{.*}}, i32 256)
+  vr1 = _vel_vldu_vssl(idx, p, 256);
+}
+
+void __attribute__((noinline))
+test_vldu_vssvl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldu_vssvl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldu.vssvl(i64 %{{.*}}, i8* %{{.*}}, <256 x double> %{{.*}}, i32 256)
+  vr1 = _vel_vldu_vssvl(idx, p, vr1, 256);
+}
+
+void __attribute__((noinline))
+test_vldunc_vssl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldunc_vssl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldunc.vssl(i64 %{{.*}}, i8* %{{.*}}, i32 256)
+  vr1 = _vel_vldunc_vssl(idx, p, 256);
+}
+
+void __attribute__((noinline))
+test_vldunc_vssvl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldunc_vssvl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldunc.vssvl(i64 %{{.*}}, i8* %{{.*}}, <256 x double> %{{.*}}, i32 256)
+  vr1 = _vel_vldunc_vssvl(idx, p, vr1, 256);
+}
+
+void __attribute__((noinline))
+test_vldlsx_vssl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldlsx_vssl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldlsx.vssl(i64 %{{.*}}, i8* %{{.*}}, i32 256)
+  vr1 = _vel_vldlsx_vssl(idx, p, 256);
+}
+
+void __attribute__((noinline))
+test_vldlsx_vssvl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldlsx_vssvl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldlsx.vssvl(i64 %{{.*}}, i8* %{{.*}}, <256 x double> %{{.*}}, i32 256)
+  vr1 = _vel_vldlsx_vssvl(idx, p, vr1, 256);
+}
+
+void __attribute__((noinline))
+test_vldlsxnc_vssl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldlsxnc_vssl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldlsxnc.vssl(i64 %{{.*}}, i8* %{{.*}}, i32 256)
+  vr1 = _vel_vldlsxnc_vssl(idx, p, 256);
+}
+
+void __attribute__((noinline))
+test_vldlsxnc_vssvl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldlsxnc_vssvl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldlsxnc.vssvl(i64 %{{.*}}, i8* %{{.*}}, <256 x double> %{{.*}}, i32 256)
+  vr1 = _vel_vldlsxnc_vssvl(idx, p, vr1, 256);
+}
+
+void __attribute__((noinline))
+test_vldlzx_vssl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldlzx_vssl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldlzx.vssl(i64 %{{.*}}, i8* %{{.*}}, i32 256)
+  vr1 = _vel_vldlzx_vssl(idx, p, 256);
+}
+
+void __attribute__((noinline))
+test_vldlzx_vssvl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldlzx_vssvl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldlzx.vssvl(i64 %{{.*}}, i8* %{{.*}}, <256 x double> %{{.*}}, i32 256)
+  vr1 = _vel_vldlzx_vssvl(idx, p, vr1, 256);
+}
+
+void __attribute__((noinline))
+test_vldlzxnc_vssl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldlzxnc_vssl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldlzxnc.vssl(i64 %{{.*}}, i8* %{{.*}}, i32 256)
+  vr1 = _vel_vldlzxnc_vssl(idx, p, 256);
+}
+
+void __attribute__((noinline))
+test_vldlzxnc_vssvl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldlzxnc_vssvl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldlzxnc.vssvl(i64 %{{.*}}, i8* %{{.*}}, <256 x double> %{{.*}}, i32 256)
+  vr1 = _vel_vldlzxnc_vssvl(idx, p, vr1, 256);
+}
+
+void __attribute__((noinline))
+test_vld2d_vssl(char* p, long idx) {
+  // CHECK-LABEL: @test_vld2d_vssl
+  // CHECK: call <256 x double> @llvm.ve.vl.vld2d.vssl(i64 %{{.*}}, i8* %{{.*}}, i32 256)
+  vr1 = _vel_vld2d_vssl(idx, p, 256);
+}
+
+void __attribute__((noinline))
+test_vld2d_vssvl(char* p, long idx) {
+  // CHECK-LABEL: @test_vld2d_vssvl
+  // CHECK: call <256 x double> @llvm.ve.vl.vld2d.vssvl(i64 %{{.*}}, i8* %{{.*}}, <256 x double> %{{.*}}, i32 256)
+  vr1 = _vel_vld2d_vssvl(idx, p, vr1, 256);
+}
+
+void __attribute__((noinline))
+test_vld2dnc_vssl(char* p, long idx) {
+  // CHECK-LABEL: @test_vld2dnc_vssl
+  // CHECK: call <256 x double> @llvm.ve.vl.vld2dnc.vssl(i64 %{{.*}}, i8* %{{.*}}, i32 256)
+  vr1 = _vel_vld2dnc_vssl(idx, p, 256);
+}
+
+void __attribute__((noinline))
+test_vld2dnc_vssvl(char* p, long idx) {
+  // CHECK-LABEL: @test_vld2dnc_vssvl
+  // CHECK: call <256 x double> @llvm.ve.vl.vld2dnc.vssvl(i64 %{{.*}}, i8* %{{.*}}, <256 x double> %{{.*}}, i32 256)
+  vr1 = _vel_vld2dnc_vssvl(idx, p, vr1, 256);
+}
+
+void __attribute__((noinline))
+test_vldu2d_vssl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldu2d_vssl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldu2d.vssl(i64 %{{.*}}, i8* %{{.*}}, i32 256)
+  vr1 = _vel_vldu2d_vssl(idx, p, 256);
+}
+
+void __attribute__((noinline))
+test_vldu2d_vssvl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldu2d_vssvl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldu2d.vssvl(i64 %{{.*}}, i8* %{{.*}}, <256 x double> %{{.*}}, i32 256)
+  vr1 = _vel_vldu2d_vssvl(idx, p, vr1, 256);
+}
+
+void __attribute__((noinline))
+test_vldu2dnc_vssl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldu2dnc_vssl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldu2dnc.vssl(i64 %{{.*}}, i8* %{{.*}}, i32 256)
+  vr1 = _vel_vldu2dnc_vssl(idx, p, 256);
+}
+
+void __attribute__((noinline))
+test_vldu2dnc_vssvl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldu2dnc_vssvl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldu2dnc.vssvl(i64 %{{.*}}, i8* %{{.*}}, <256 x double> %{{.*}}, i32 256)
+  vr1 = _vel_vldu2dnc_vssvl(idx, p, vr1, 256);
+}
+
+void __attribute__((noinline))
+test_vldl2dsx_vssl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldl2dsx_vssl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldl2dsx.vssl(i64 %{{.*}}, i8* %{{.*}}, i32 256)
+  vr1 = _vel_vldl2dsx_vssl(idx, p, 256);
+}
+
+void __attribute__((noinline))
+test_vldl2dsx_vssvl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldl2dsx_vssvl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldl2dsx.vssvl(i64 %{{.*}}, i8* %{{.*}}, <256 x double> %{{.*}}, i32 256)
+  vr1 = _vel_vldl2dsx_vssvl(idx, p, vr1, 256);
+}
+
+void __attribute__((noinline))
+test_vldl2dsxnc_vssl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldl2dsxnc_vssl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldl2dsxnc.vssl(i64 %{{.*}}, i8* %{{.*}}, i32 256)
+  vr1 = _vel_vldl2dsxnc_vssl(idx, p, 256);
+}
+
+void __attribute__((noinline))
+test_vldl2dsxnc_vssvl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldl2dsxnc_vssvl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldl2dsxnc.vssvl(i64 %{{.*}}, i8* %{{.*}}, <256 x double> %{{.*}}, i32 256)
+  vr1 = _vel_vldl2dsxnc_vssvl(idx, p, vr1, 256);
+}
+
+void __attribute__((noinline))
+test_vldl2dzx_vssl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldl2dzx_vssl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldl2dzx.vssl(i64 %{{.*}}, i8* %{{.*}}, i32 256)
+  vr1 = _vel_vldl2dzx_vssl(idx, p, 256);
+}
+
+void __attribute__((noinline))
+test_vldl2dzx_vssvl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldl2dzx_vssvl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldl2dzx.vssvl(i64 %{{.*}}, i8* %{{.*}}, <256 x double> %{{.*}}, i32 256)
+  vr1 = _vel_vldl2dzx_vssvl(idx, p, vr1, 256);
+}
+
+void __attribute__((noinline))
+test_vldl2dzxnc_vssl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldl2dzxnc_vssl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldl2dzxnc.vssl(i64 %{{.*}}, i8* %{{.*}}, i32 256)
+  vr1 = _vel_vldl2dzxnc_vssl(idx, p, 256);
+}
+
+void __attribute__((noinline))
+test_vldl2dzxnc_vssvl(char* p, long idx) {
+  // CHECK-LABEL: @test_vldl2dzxnc_vssvl
+  // CHECK: call <256 x double> @llvm.ve.vl.vldl2dzxnc.vssvl(i64 %{{.*}}, i8* %{{.*}}, <256 x double> %{{.*}}, i32 256)
+  vr1 = _vel_vldl2dzxnc_vssvl(idx, p, vr1, 256);
+}