[RISCV] Support __builtin_nontemporal_load/store by MachineMemOperand

[BeMg]

Summary

In this commit, we support nontemporal feature by checking the MachineMemOperand from
load/store MI.

There are two conditions that need to be met.

1. The Load/Store and NTLH is immediate subsequent.
2. Maintain the correct code size for relaxation.

For condition 1, it creates the RISCVInsertNTLHInstsPass and executes it in addPreEmitPass2.
This MachineFunction Pass does the similar thing as RISCVExpandPseudoPass
but it only expands the instruction with non-temporal flag (check through MachineMemOperand)

For condition 2, it checks the all instructions with non-temporal flag
(also through MachineMemOperand) and calculate in getInstSizeInBytes.

This way could support both scalar type and vector type (fixed-length).

[reames]

The way other targets appear to be approach this is to modify ISEL. Can you explain why you believe the late hint insertion pass is the right approach here? The alternative would be to have a family of non-temporal pseudos which expanded into a hint + a memory op. I don't have a strong opinion, just curious for the justification.

llvm/test/CodeGen/RISCV/nontemporal-c.ll

5 ↗

(On Diff #494988)

These need updated to use modern "ptr" syntax.

llvm/test/CodeGen/RISCV/nontemporal-vector.ll

42 ↗

(On Diff #494988)

Having scalar tests in a vector file is a bit odd...

llvm/test/CodeGen/RISCV/nontemporal.ll

466

I couple of general issues with your tests:

• It's not really clear what the separation between the test files is. You appear to repeat tests multiple times. Did you maybe want to use multiple check lines in a single file?
• You don't appear to have any test coverage of scalable vectors.
• You need to update syntax to use modern "ptr".
• You mix volatile and non-volatile without discussion on why (that I saw on a quick skim). In general, you should not be using volatile you're actually testing volatile. (i.e. don't use it just to "stabilize a test")

[craig.topper]

In D143361#4150627, @reames wrote:

The way other targets appear to be approach this is to modify ISEL. Can you explain why you believe the late hint insertion pass is the right approach here? The alternative would be to have a family of non-temporal pseudos which expanded into a hint + a memory op. I don't have a strong opinion, just curious for the justification.

I think this was to reduce the number of vector pseudos required. There 2088 vector load pseudos today if you consider unit stride, strided, indexed(with different eews for index). Then there are segment versions of those with different seqment sizes. Then there are masked, unmasked, and tail undisturbed versions. For full generality we would need to apply non-temporal hint to all of them. Though __builtin_nontemporal_load doesn't need all of those variations we could have target specific intrinsics that do.

[BeMg]

Reorganizing the test case

1. Use ptr instead of type*
2. Remove volatile
3. Merge scalar/C-ext/fixed-length into one testcase file
4. Add scalable vector testcase (put in nontemporal-scalable.ll because the scalable type can't be compiled without the V extension)

[craig.topper]

llvm/lib/Target/RISCV/RISCVInsertNTLHInsts.cpp

57

Move the declaration and assignment of Changed to line 64, right before the loop.

71

This should be hasStdExtCOrZca()

llvm/lib/Target/RISCV/RISCVInstrInfo.cpp

1251

COrZca

llvm/lib/Target/RISCV/RISCVInstrInfoZihintntl.td

1

This line exceeds 80 characters

14

Doesn't the Pseudo class already imply isCodeGenOnly = 1?

[BeMg]

1. Move Changed
2. hasStdExtC -> hasStdExtCOrZca
3. Update format
4. Remove isCodeGenOnly

[BeMg]

rebase and ping

[craig.topper]

llvm/lib/Target/RISCV/RISCVInsertNTLHInsts.cpp

58

const auto &ST = MF.getSubtarget<RISCVSubtarget>();
TII = ST.getInstrInfo();

[BeMg]

Update TII

[craig.topper]

LGTM

llvm/lib/Target/RISCV/CMakeLists.txt

 add_llvm_target(RISCVCodeGen
   RISCVAsmPrinter.cpp
   RISCVCodeGenPrepare.cpp
   RISCVMakeCompressible.cpp
   RISCVExpandAtomicPseudoInsts.cpp
   RISCVExpandPseudoInsts.cpp
   RISCVFrameLowering.cpp
   RISCVGatherScatterLowering.cpp
+  RISCVInsertNTLHInsts.cpp
   RISCVInsertVSETVLI.cpp
   RISCVInstrInfo.cpp
   RISCVISelDAGToDAG.cpp
   RISCVISelLowering.cpp
   RISCVMachineFunctionInfo.cpp
   RISCVMacroFusion.cpp
   RISCVMCInstLower.cpp
   RISCVMergeBaseOffset.cpp

llvm/lib/Target/RISCV/RISCV.h

 void initializeRISCVExpandPseudoPass(PassRegistry &);
 
 FunctionPass *createRISCVPreRAExpandPseudoPass();
 void initializeRISCVPreRAExpandPseudoPass(PassRegistry &);
 
 FunctionPass *createRISCVExpandAtomicPseudoPass();
 void initializeRISCVExpandAtomicPseudoPass(PassRegistry &);
 
+FunctionPass *createRISCVInsertNTLHInstsPass();
+void initializeRISCVInsertNTLHInstsPass(PassRegistry &);
+
 FunctionPass *createRISCVInsertVSETVLIPass();
 void initializeRISCVInsertVSETVLIPass(PassRegistry &);
 
 FunctionPass *createRISCVRedundantCopyEliminationPass();
 void initializeRISCVRedundantCopyEliminationPass(PassRegistry &);
 
 FunctionPass *createRISCVInitUndefPass();
 void initializeRISCVInitUndefPass(PassRegistry &);

llvm/lib/Target/RISCV/RISCVInsertNTLHInsts.cpp

+//===-- RISCVInsertNTLHInsts.cpp - Insert NTLH extension instrution -------===//
+//
+// 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 implements a function pass that inserts non-temporal hint
+// instructions where needed.
+//
+// It checks the MachineMemOperand of all MachineInstr.
+// If the instruction has a MachineMemOperand and isNontemporal is true,
+// then ntlh instruction is inserted before it.
+//
+//===----------------------------------------------------------------------===//
+
+#include "RISCV.h"
+#include "RISCVInstrInfo.h"
+#include "RISCVTargetMachine.h"
+
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+
+using namespace llvm;
+
+#define RISCV_INSERT_NTLH_INSTS_NAME "RISCV insert NTLH instruction pass"
+
+namespace {
+
+class RISCVInsertNTLHInsts : public MachineFunctionPass {
+public:
+  const RISCVInstrInfo *TII;
+  static char ID;
+
+  RISCVInsertNTLHInsts() : MachineFunctionPass(ID) {
+    initializeRISCVInsertNTLHInstsPass(*PassRegistry::getPassRegistry());
+  }
+
+  bool runOnMachineFunction(MachineFunction &MF) override;
+
+  void getAnalysisUsage(AnalysisUsage &AU) const override {
+    AU.setPreservesCFG();
+    MachineFunctionPass::getAnalysisUsage(AU);
+  }
+
+  StringRef getPassName() const override {
+    return RISCV_INSERT_NTLH_INSTS_NAME;
+  }
+};
+
+} // end of anonymous namespace
+
+char RISCVInsertNTLHInsts::ID = 0;
+
+bool RISCVInsertNTLHInsts::runOnMachineFunction(MachineFunction &MF) {
+  bool Changed = false;
+  TII = static_cast<const RISCVInstrInfo *>(MF.getSubtarget().getInstrInfo());
+  const auto &ST = MF.getSubtarget<RISCVSubtarget>();
+
+  if (!ST.hasStdExtZihintntl())
+    return false;
+
+  for (auto &MBB : MF) {
+    for (auto &MBBI : MBB) {
+      if (MBBI.memoperands_empty())
+        continue;
+      MachineMemOperand *MMO = *(MBBI.memoperands_begin());
+      if (MMO->isNonTemporal()) {
+        DebugLoc DL = MBBI.getDebugLoc();
+        if (ST.hasStdExtC() && ST.enableRVCHintInstrs())
+          BuildMI(MBB, MBBI, DL, TII->get(RISCV::PseudoCNTLALL));
+        else
+          BuildMI(MBB, MBBI, DL, TII->get(RISCV::PseudoNTLALL));
+        Changed = true;
+      }
+    }
+  }
+
+  return Changed;
+}
+
+INITIALIZE_PASS(RISCVInsertNTLHInsts, "riscv-insert-ntlh-insts",
+                RISCV_INSERT_NTLH_INSTS_NAME, false, false)
+
+namespace llvm {
+
+FunctionPass *createRISCVInsertNTLHInstsPass() {
+  return new RISCVInsertNTLHInsts();
+}
+
+} // end of namespace llvm

llvm/lib/Target/RISCV/RISCVInstrInfo.cpp

unsigned RISCVInstrInfo::getInstSizeInBytes(const MachineInstr &MI) const {
   if (Opcode == TargetOpcode::INLINEASM ||
       Opcode == TargetOpcode::INLINEASM_BR) {
     const MachineFunction &MF = *MI.getParent()->getParent();
     const auto &TM = static_cast<const RISCVTargetMachine &>(MF.getTarget());
     return getInlineAsmLength(MI.getOperand(0).getSymbolName(),
                               *TM.getMCAsmInfo());
   }
 
+  if (!MI.memoperands_empty()) {
+    MachineMemOperand *MMO = *(MI.memoperands_begin());
+    const MachineFunction &MF = *MI.getParent()->getParent();
+    const auto &ST = MF.getSubtarget<RISCVSubtarget>();
+    if (ST.hasStdExtZihintntl() && MMO->isNonTemporal()) {
+      if (ST.hasStdExtC() && ST.enableRVCHintInstrs()) {
+        if (isCompressibleInst(MI, STI))
+          return 4; // c.ntl.all + c.load/c.store
+        return 6;   // c.ntl.all + load/store
+      }
+      return 8; // ntl.all + load/store
+    }
+  }
+
   if (MI.getParent() && MI.getParent()->getParent()) {
     if (isCompressibleInst(MI, STI))
       return 2;
   }
   return get(Opcode).getSize();
 }
 
 bool RISCVInstrInfo::isAsCheapAsAMove(const MachineInstr &MI) const {

llvm/lib/Target/RISCV/RISCVInstrInfo.td

 include "RISCVInstrInfoZicbo.td"
 
 //===----------------------------------------------------------------------===//
 // Vendor extensions
 //===----------------------------------------------------------------------===//
 
 include "RISCVInstrInfoXVentana.td"
 include "RISCVInstrInfoXTHead.td"
+include "RISCVInstrInfoZihintntl.td"

llvm/lib/Target/RISCV/RISCVInstrInfoZihintntl.td

+//RISCVInstrInfoZihintntl.td - RISC-V 'Zihintntl' instructions -*- tablegen -*-//
+//
+// 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 describes the RISC-V instructions from Non-Temporal Locality Hints 
+/// extension document (zihintntl).
+///
+//===----------------------------------------------------------------------===//
+
+let hasSideEffects = 0, mayLoad = 0, mayStore = 0, Size = 4, isCodeGenOnly = 1 in {
+  def PseudoNTLALL :  Pseudo<(outs), (ins), [], "ntl.all">, 
+                             PseudoInstExpansion<(ADD X0, X0, X5)>;
+}
+
+let hasSideEffects = 0, mayLoad = 0, mayStore = 0, Size = 2, isCodeGenOnly = 1 in {
+  def PseudoCNTLALL :  Pseudo<(outs), (ins), [], "c.ntl.all">,
+                              PseudoInstExpansion<(C_ADD_HINT X0, X0, X5)>;
+}

llvm/lib/Target/RISCV/RISCVTargetMachine.cpp

extern "C" LLVM_EXTERNAL_VISIBILITY void LLVMInitializeRISCVTarget() {
   initializeRISCVMakeCompressibleOptPass(*PR);
   initializeRISCVGatherScatterLoweringPass(*PR);
   initializeRISCVCodeGenPreparePass(*PR);
   initializeRISCVMergeBaseOffsetOptPass(*PR);
   initializeRISCVSExtWRemovalPass(*PR);
   initializeRISCVStripWSuffixPass(*PR);
   initializeRISCVPreRAExpandPseudoPass(*PR);
   initializeRISCVExpandPseudoPass(*PR);
+  initializeRISCVInsertNTLHInstsPass(*PR);
   initializeRISCVInsertVSETVLIPass(*PR);
   initializeRISCVDAGToDAGISelPass(*PR);
   initializeRISCVInitUndefPass(*PR);
 }
 
 static StringRef computeDataLayout(const Triple &TT) {
   if (TT.isArch64Bit())
     return "e-m:e-p:64:64-i64:64-i128:128-n32:64-S128";

 void RISCVPassConfig::addPreEmitPass2() {
   addPass(createRISCVExpandPseudoPass());
 
   // Do the copy propagation after expanding pseudos because we may produce some
   // MVs when expanding.
   if (TM->getOptLevel() >= CodeGenOpt::Default && EnableRISCVCopyPropagation)
     addPass(createMachineCopyPropagationPass(true));
 
+  addPass(createRISCVInsertNTLHInstsPass());
   // Schedule the expansion of AMOs at the last possible moment, avoiding the
   // possibility for other passes to break the requirements for forward
   // progress in the LR/SC block.
   addPass(createRISCVExpandAtomicPseudoPass());
 }
 
 void RISCVPassConfig::addMachineSSAOptimization() {
   TargetPassConfig::addMachineSSAOptimization();

llvm/test/CodeGen/RISCV/O0-pipeline.ll

 ; CHECK-NEXT:       Contiguously Lay Out Funclets
 ; CHECK-NEXT:       StackMap Liveness Analysis
 ; CHECK-NEXT:       Live DEBUG_VALUE analysis
 ; CHECK-NEXT:       Machine Sanitizer Binary Metadata
 ; CHECK-NEXT:       Lazy Machine Block Frequency Analysis
 ; CHECK-NEXT:       Machine Optimization Remark Emitter
 ; CHECK-NEXT:       Stack Frame Layout Analysis
 ; CHECK-NEXT:       RISCV pseudo instruction expansion pass
+; CHECK-NEXT:       RISCV insert NTLH instruction pass
 ; CHECK-NEXT:       RISCV atomic pseudo instruction expansion pass
 ; CHECK-NEXT:       Lazy Machine Block Frequency Analysis
 ; CHECK-NEXT:       Machine Optimization Remark Emitter
 ; CHECK-NEXT:       RISCV Assembly Printer
 ; CHECK-NEXT:       Free MachineFunction

llvm/test/CodeGen/RISCV/O3-pipeline.ll

 ; CHECK-NEXT:       Machine Sanitizer Binary Metadata
 ; CHECK-NEXT:     Machine Outliner
 ; CHECK-NEXT:     FunctionPass Manager
 ; CHECK-NEXT:       Lazy Machine Block Frequency Analysis
 ; CHECK-NEXT:       Machine Optimization Remark Emitter
 ; CHECK-NEXT:       Stack Frame Layout Analysis
 ; CHECK-NEXT:       RISCV pseudo instruction expansion pass
 ; CHECK-NEXT:       Machine Copy Propagation Pass
+; CHECK-NEXT:       RISCV insert NTLH instruction pass
 ; CHECK-NEXT:       RISCV atomic pseudo instruction expansion pass
 ; CHECK-NEXT:       Lazy Machine Block Frequency Analysis
 ; CHECK-NEXT:       Machine Optimization Remark Emitter
 ; CHECK-NEXT:       RISCV Assembly Printer
 ; CHECK-NEXT:       Free MachineFunction

llvm/test/CodeGen/RISCV/nontemporal-scalable.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -mtriple=riscv64 -mattr=+experimental-zihintntl,+f,+d,+zfh,+v < %s | FileCheck %s -check-prefix=CHECK-RV64V
+; RUN: llc -mtriple=riscv32 -mattr=+experimental-zihintntl,+f,+d,+zfh,+v < %s | FileCheck %s -check-prefix=CHECK-RV32V
+
+define <vscale x 2 x i64> @test_nontemporal_load_nxv2i64(ptr %p) {
+; CHECK-RV64V-LABEL: test_nontemporal_load_nxv2i64:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    vl2re64.v v8, (a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_load_nxv2i64:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    vl2re64.v v8, (a0)
+; CHECK-RV32V-NEXT:    ret
+  %1 = load <vscale x 2 x i64>, ptr %p, !nontemporal !0
+  ret <vscale x 2 x i64> %1
+}
+
+define <vscale x 4 x i32> @test_nontemporal_load_nxv4i32(ptr %p) {
+; CHECK-RV64V-LABEL: test_nontemporal_load_nxv4i32:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    vl2re32.v v8, (a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_load_nxv4i32:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    vl2re32.v v8, (a0)
+; CHECK-RV32V-NEXT:    ret
+  %1 = load <vscale x 4 x i32>, ptr %p, !nontemporal !0
+  ret <vscale x 4 x i32> %1
+}
+
+define <vscale x 8 x i16> @test_nontemporal_load_nxv8i16(ptr %p) {
+; CHECK-RV64V-LABEL: test_nontemporal_load_nxv8i16:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    vl2re16.v v8, (a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_load_nxv8i16:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    vl2re16.v v8, (a0)
+; CHECK-RV32V-NEXT:    ret
+  %1 = load <vscale x 8 x i16>, ptr %p, !nontemporal !0
+  ret <vscale x 8 x i16> %1
+}
+
+define <vscale x 16 x i8> @test_nontemporal_load_nxv16i8(ptr %p) {
+; CHECK-RV64V-LABEL: test_nontemporal_load_nxv16i8:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    vl2r.v v8, (a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_load_nxv16i8:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    vl2r.v v8, (a0)
+; CHECK-RV32V-NEXT:    ret
+  %1 = load <vscale x 16 x i8>, ptr %p, !nontemporal !0
+  ret <vscale x 16 x i8> %1
+}
+
+define void @test_nontemporal_store_nxv2i64(ptr %p, <vscale x 2 x i64> %v) {
+; CHECK-RV64V-LABEL: test_nontemporal_store_nxv2i64:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    vs2r.v v8, (a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_store_nxv2i64:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    vs2r.v v8, (a0)
+; CHECK-RV32V-NEXT:    ret
+  store <vscale x 2 x i64> %v, ptr %p, !nontemporal !0
+  ret void
+}
+
+define void @test_nontemporal_store_nxv4i32(ptr %p, <vscale x 4 x i32> %v) {
+; CHECK-RV64V-LABEL: test_nontemporal_store_nxv4i32:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    vs2r.v v8, (a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_store_nxv4i32:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    vs2r.v v8, (a0)
+; CHECK-RV32V-NEXT:    ret
+  store <vscale x 4 x i32> %v, ptr %p, !nontemporal !0
+  ret void
+}
+
+define void @test_nontemporal_store_nxv8i16(ptr %p, <vscale x 8 x i16> %v) {
+; CHECK-RV64V-LABEL: test_nontemporal_store_nxv8i16:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    vs2r.v v8, (a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_store_nxv8i16:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    vs2r.v v8, (a0)
+; CHECK-RV32V-NEXT:    ret
+  store <vscale x 8 x i16> %v, ptr %p, !nontemporal !0
+  ret void
+}
+
+define void @test_nontemporal_store_nxv16i8(ptr %p, <vscale x 16 x i8> %v) {
+; CHECK-RV64V-LABEL: test_nontemporal_store_nxv16i8:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    vs2r.v v8, (a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_store_nxv16i8:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    vs2r.v v8, (a0)
+; CHECK-RV32V-NEXT:    ret
+  store <vscale x 16 x i8> %v, ptr %p, !nontemporal !0
+  ret void
+}
+
+!0 = !{i32 1}

llvm/test/CodeGen/RISCV/nontemporal.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv64 -mattr=+experimental-zihintntl,+f,+d,+zfh < %s | FileCheck %s -check-prefix=CHECK-RV64
+; RUN: llc -mtriple=riscv32 -mattr=+experimental-zihintntl,+f,+d,+zfh < %s | FileCheck %s -check-prefix=CHECK-RV32
+; RUN: llc -mtriple=riscv64 -mattr=+experimental-zihintntl,+f,+d,+zfh,+c < %s | FileCheck %s -check-prefix=CHECK-RV64C
+; RUN: llc -mtriple=riscv32 -mattr=+experimental-zihintntl,+f,+d,+zfh,+c < %s | FileCheck %s -check-prefix=CHECK-RV32C
+; RUN: llc -mtriple=riscv64 -mattr=+experimental-zihintntl,+f,+d,+zfh,+v < %s | FileCheck %s -check-prefix=CHECK-RV64V
+; RUN: llc -mtriple=riscv32 -mattr=+experimental-zihintntl,+f,+d,+zfh,+v < %s | FileCheck %s -check-prefix=CHECK-RV32V
+
+define i64 @test_nontemporal_load_i64(ptr %p) {
+; CHECK-RV64-LABEL: test_nontemporal_load_i64:
+; CHECK-RV64:       # %bb.0:
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    ld a0, 0(a0)
+; CHECK-RV64-NEXT:    ret
+;
+; CHECK-RV32-LABEL: test_nontemporal_load_i64:
+; CHECK-RV32:       # %bb.0:
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    lw a2, 0(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    lw a1, 4(a0)
+; CHECK-RV32-NEXT:    mv a0, a2
+; CHECK-RV32-NEXT:    ret
+;
+; CHECK-RV64C-LABEL: test_nontemporal_load_i64:
+; CHECK-RV64C:       # %bb.0:
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    ld a0, 0(a0)
+; CHECK-RV64C-NEXT:    ret
+;
+; CHECK-RV32C-LABEL: test_nontemporal_load_i64:
+; CHECK-RV32C:       # %bb.0:
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    lw a2, 0(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    lw a1, 4(a0)
+; CHECK-RV32C-NEXT:    mv a0, a2
+; CHECK-RV32C-NEXT:    ret
+;
+; CHECK-RV64V-LABEL: test_nontemporal_load_i64:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    ld a0, 0(a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_load_i64:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    lw a2, 0(a0)
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    lw a1, 4(a0)
+; CHECK-RV32V-NEXT:    mv a0, a2
+; CHECK-RV32V-NEXT:    ret
+
+  %1 = load i64, ptr %p, !nontemporal !0
+  ret i64 %1
+}
+
+define i32 @test_nontemporal_load_i32(ptr %p) {
+; CHECK-RV64-LABEL: test_nontemporal_load_i32:
+; CHECK-RV64:       # %bb.0:
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    lw a0, 0(a0)
+; CHECK-RV64-NEXT:    ret
+;
+; CHECK-RV32-LABEL: test_nontemporal_load_i32:
+; CHECK-RV32:       # %bb.0:
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    lw a0, 0(a0)
+; CHECK-RV32-NEXT:    ret
+;
+; CHECK-RV64C-LABEL: test_nontemporal_load_i32:
+; CHECK-RV64C:       # %bb.0:
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    lw a0, 0(a0)
+; CHECK-RV64C-NEXT:    ret
+;
+; CHECK-RV32C-LABEL: test_nontemporal_load_i32:
+; CHECK-RV32C:       # %bb.0:
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    lw a0, 0(a0)
+; CHECK-RV32C-NEXT:    ret
+;
+; CHECK-RV64V-LABEL: test_nontemporal_load_i32:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    lw a0, 0(a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_load_i32:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    lw a0, 0(a0)
+; CHECK-RV32V-NEXT:    ret
+
+  %1 = load i32, ptr %p, !nontemporal !0
+  ret i32 %1
+}
+
+define i16 @test_nontemporal_load_i16(ptr %p) {
+; CHECK-RV64-LABEL: test_nontemporal_load_i16:
+; CHECK-RV64:       # %bb.0:
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    lh a0, 0(a0)
+; CHECK-RV64-NEXT:    ret
+;
+; CHECK-RV32-LABEL: test_nontemporal_load_i16:
+; CHECK-RV32:       # %bb.0:
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    lh a0, 0(a0)
+; CHECK-RV32-NEXT:    ret
+;
+; CHECK-RV64C-LABEL: test_nontemporal_load_i16:
+; CHECK-RV64C:       # %bb.0:
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    lh a0, 0(a0)
+; CHECK-RV64C-NEXT:    ret
+;
+; CHECK-RV32C-LABEL: test_nontemporal_load_i16:
+; CHECK-RV32C:       # %bb.0:
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    lh a0, 0(a0)
+; CHECK-RV32C-NEXT:    ret
+;
+; CHECK-RV64V-LABEL: test_nontemporal_load_i16:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    lh a0, 0(a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_load_i16:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    lh a0, 0(a0)
+; CHECK-RV32V-NEXT:    ret
+
+  %1 = load i16, ptr %p, !nontemporal !0
+  ret i16 %1
+}
+
+define i8 @test_nontemporal_load_i8(ptr %p) {
+; CHECK-RV64-LABEL: test_nontemporal_load_i8:
+; CHECK-RV64:       # %bb.0:
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    lb a0, 0(a0)
+; CHECK-RV64-NEXT:    ret
+;
+; CHECK-RV32-LABEL: test_nontemporal_load_i8:
+; CHECK-RV32:       # %bb.0:
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    lb a0, 0(a0)
+; CHECK-RV32-NEXT:    ret
+;
+; CHECK-RV64C-LABEL: test_nontemporal_load_i8:
+; CHECK-RV64C:       # %bb.0:
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    lb a0, 0(a0)
+; CHECK-RV64C-NEXT:    ret
+;
+; CHECK-RV32C-LABEL: test_nontemporal_load_i8:
+; CHECK-RV32C:       # %bb.0:
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    lb a0, 0(a0)
+; CHECK-RV32C-NEXT:    ret
+;
+; CHECK-RV64V-LABEL: test_nontemporal_load_i8:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    lb a0, 0(a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_load_i8:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    lb a0, 0(a0)
+; CHECK-RV32V-NEXT:    ret
+
+  %1 = load i8, ptr %p, !nontemporal !0
+  ret i8 %1
+}
+
+define half @test_nontemporal_load_half(ptr %p) nounwind {
+; CHECK-RV64-LABEL: test_nontemporal_load_half:
+; CHECK-RV64:       # %bb.0:
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    flh ft0, 0(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    flh ft1, 6(a0)
+; CHECK-RV64-NEXT:    fadd.h fa0, ft0, ft1
+; CHECK-RV64-NEXT:    ret
+;
+; CHECK-RV32-LABEL: test_nontemporal_load_half:
+; CHECK-RV32:       # %bb.0:
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    flh ft0, 0(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    flh ft1, 6(a0)
+; CHECK-RV32-NEXT:    fadd.h fa0, ft0, ft1
+; CHECK-RV32-NEXT:    ret
+;
+; CHECK-RV64C-LABEL: test_nontemporal_load_half:
+; CHECK-RV64C:       # %bb.0:
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    flh ft0, 0(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    flh ft1, 6(a0)
+; CHECK-RV64C-NEXT:    fadd.h fa0, ft0, ft1
+; CHECK-RV64C-NEXT:    ret
+;
+; CHECK-RV32C-LABEL: test_nontemporal_load_half:
+; CHECK-RV32C:       # %bb.0:
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    flh ft0, 0(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    flh ft1, 6(a0)
+; CHECK-RV32C-NEXT:    fadd.h fa0, ft0, ft1
+; CHECK-RV32C-NEXT:    ret
+;
+; CHECK-RV64V-LABEL: test_nontemporal_load_half:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    flh ft0, 0(a0)
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    flh ft1, 6(a0)
+; CHECK-RV64V-NEXT:    fadd.h fa0, ft0, ft1
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_load_half:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    flh ft0, 0(a0)
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    flh ft1, 6(a0)
+; CHECK-RV32V-NEXT:    fadd.h fa0, ft0, ft1
+; CHECK-RV32V-NEXT:    ret
+
+  %1 = load half, ptr %p, !nontemporal !0
+  %2 = getelementptr half, ptr %p, i32 3
+  %3 = load half, ptr %2, !nontemporal !0
+  %4 = fadd half %1, %3
+  ret half %4
+}
+
+define float @test_nontemporal_load_float(ptr %p) {
+; CHECK-RV64-LABEL: test_nontemporal_load_float:
+; CHECK-RV64:       # %bb.0:
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    flw fa0, 0(a0)
+; CHECK-RV64-NEXT:    ret
+;
+; CHECK-RV32-LABEL: test_nontemporal_load_float:
+; CHECK-RV32:       # %bb.0:
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    flw fa0, 0(a0)
+; CHECK-RV32-NEXT:    ret
+;
+; CHECK-RV64C-LABEL: test_nontemporal_load_float:
+; CHECK-RV64C:       # %bb.0:
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    flw fa0, 0(a0)
+; CHECK-RV64C-NEXT:    ret
+;
+; CHECK-RV32C-LABEL: test_nontemporal_load_float:
+; CHECK-RV32C:       # %bb.0:
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    flw fa0, 0(a0)
+; CHECK-RV32C-NEXT:    ret
+;
+; CHECK-RV64V-LABEL: test_nontemporal_load_float:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    flw fa0, 0(a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_load_float:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    flw fa0, 0(a0)
+; CHECK-RV32V-NEXT:    ret
+
+  %1 = load float, ptr %p, !nontemporal !0
+  ret float %1
+}
+
+define double @test_nontemporal_load_double(ptr %p) {
+; CHECK-RV64-LABEL: test_nontemporal_load_double:
+; CHECK-RV64:       # %bb.0:
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    fld fa0, 0(a0)
+; CHECK-RV64-NEXT:    ret
+;
+; CHECK-RV32-LABEL: test_nontemporal_load_double:
+; CHECK-RV32:       # %bb.0:
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    fld fa0, 0(a0)
+; CHECK-RV32-NEXT:    ret
+;
+; CHECK-RV64C-LABEL: test_nontemporal_load_double:
+; CHECK-RV64C:       # %bb.0:
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    fld fa0, 0(a0)
+; CHECK-RV64C-NEXT:    ret
+;
+; CHECK-RV32C-LABEL: test_nontemporal_load_double:
+; CHECK-RV32C:       # %bb.0:
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    fld fa0, 0(a0)
+; CHECK-RV32C-NEXT:    ret
+;
+; CHECK-RV64V-LABEL: test_nontemporal_load_double:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    fld fa0, 0(a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_load_double:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    fld fa0, 0(a0)
+; CHECK-RV32V-NEXT:    ret
+
+  %1 = load double, ptr %p, !nontemporal !0
+  ret double %1
+}
+
+define <16 x i8> @test_nontemporal_load_v16i8(ptr %p) {
+; CHECK-RV64-LABEL: test_nontemporal_load_v16i8:
+; CHECK-RV64:       # %bb.0:
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    ld a2, 8(a1)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    ld a1, 0(a1)
+; CHECK-RV64-NEXT:    sd a2, 8(a0)
+; CHECK-RV64-NEXT:    sd a1, 0(a0)
+; CHECK-RV64-NEXT:    ret
+;
+; CHECK-RV32-LABEL: test_nontemporal_load_v16i8:
+; CHECK-RV32:       # %bb.0:
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    lw a2, 12(a1)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    lw a3, 8(a1)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    lw a4, 4(a1)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    lw a1, 0(a1)
+; CHECK-RV32-NEXT:    sw a2, 12(a0)
+; CHECK-RV32-NEXT:    sw a3, 8(a0)
+; CHECK-RV32-NEXT:    sw a4, 4(a0)
+; CHECK-RV32-NEXT:    sw a1, 0(a0)
+; CHECK-RV32-NEXT:    ret
+;
+; CHECK-RV64C-LABEL: test_nontemporal_load_v16i8:
+; CHECK-RV64C:       # %bb.0:
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    ld a2, 8(a1)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    ld a1, 0(a1)
+; CHECK-RV64C-NEXT:    sd a2, 8(a0)
+; CHECK-RV64C-NEXT:    sd a1, 0(a0)
+; CHECK-RV64C-NEXT:    ret
+;
+; CHECK-RV32C-LABEL: test_nontemporal_load_v16i8:
+; CHECK-RV32C:       # %bb.0:
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    lw a2, 12(a1)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    lw a3, 8(a1)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    lw a4, 4(a1)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    lw a1, 0(a1)
+; CHECK-RV32C-NEXT:    sw a2, 12(a0)
+; CHECK-RV32C-NEXT:    sw a3, 8(a0)
+; CHECK-RV32C-NEXT:    sw a4, 4(a0)
+; CHECK-RV32C-NEXT:    sw a1, 0(a0)
+; CHECK-RV32C-NEXT:    ret
+;
+; CHECK-RV64V-LABEL: test_nontemporal_load_v16i8:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    vsetivli zero, 16, e8, m1, ta, ma
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    vle8.v v8, (a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_load_v16i8:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    vsetivli zero, 16, e8, m1, ta, ma
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    vle8.v v8, (a0)
+; CHECK-RV32V-NEXT:    ret
+
+  %1 = load <16 x i8>, ptr %p, !nontemporal !0
+  ret <16 x i8> %1
+}
+
+define <8 x i16> @test_nontemporal_load_v8i16(ptr %p) {
+; CHECK-RV64-LABEL: test_nontemporal_load_v8i16:
+; CHECK-RV64:       # %bb.0:
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    ld a2, 8(a1)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    ld a1, 0(a1)
+; CHECK-RV64-NEXT:    sd a2, 8(a0)
+; CHECK-RV64-NEXT:    sd a1, 0(a0)
+; CHECK-RV64-NEXT:    ret
+;
+; CHECK-RV32-LABEL: test_nontemporal_load_v8i16:
+; CHECK-RV32:       # %bb.0:
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    lw a2, 12(a1)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    lw a3, 8(a1)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    lw a4, 4(a1)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    lw a1, 0(a1)
+; CHECK-RV32-NEXT:    sw a2, 12(a0)
+; CHECK-RV32-NEXT:    sw a3, 8(a0)
+; CHECK-RV32-NEXT:    sw a4, 4(a0)
+; CHECK-RV32-NEXT:    sw a1, 0(a0)
+; CHECK-RV32-NEXT:    ret
+;
+; CHECK-RV64C-LABEL: test_nontemporal_load_v8i16:
+; CHECK-RV64C:       # %bb.0:
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    ld a2, 8(a1)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    ld a1, 0(a1)
+; CHECK-RV64C-NEXT:    sd a2, 8(a0)
+; CHECK-RV64C-NEXT:    sd a1, 0(a0)
+; CHECK-RV64C-NEXT:    ret
+;
+; CHECK-RV32C-LABEL: test_nontemporal_load_v8i16:
+; CHECK-RV32C:       # %bb.0:
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    lw a2, 12(a1)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    lw a3, 8(a1)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    lw a4, 4(a1)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    lw a1, 0(a1)
+; CHECK-RV32C-NEXT:    sw a2, 12(a0)
+; CHECK-RV32C-NEXT:    sw a3, 8(a0)
+; CHECK-RV32C-NEXT:    sw a4, 4(a0)
+; CHECK-RV32C-NEXT:    sw a1, 0(a0)
+; CHECK-RV32C-NEXT:    ret
+;
+; CHECK-RV64V-LABEL: test_nontemporal_load_v8i16:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    vsetivli zero, 8, e16, m1, ta, ma
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    vle16.v v8, (a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_load_v8i16:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    vsetivli zero, 8, e16, m1, ta, ma
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    vle16.v v8, (a0)
+; CHECK-RV32V-NEXT:    ret
+
+  %1 = load <8 x i16>, ptr %p, !nontemporal !0
+  ret <8 x i16> %1
+}
+
+define <4 x i32> @test_nontemporal_load_v4i32(ptr %p) {
+; CHECK-RV64-LABEL: test_nontemporal_load_v4i32:
+; CHECK-RV64:       # %bb.0:
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    ld a2, 8(a1)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    ld a1, 0(a1)
+; CHECK-RV64-NEXT:    sd a2, 8(a0)
+; CHECK-RV64-NEXT:    sd a1, 0(a0)
+; CHECK-RV64-NEXT:    ret
+;
+; CHECK-RV32-LABEL: test_nontemporal_load_v4i32:
+; CHECK-RV32:       # %bb.0:
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    lw a2, 12(a1)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    lw a3, 8(a1)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    lw a4, 4(a1)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    lw a1, 0(a1)
+; CHECK-RV32-NEXT:    sw a2, 12(a0)
+; CHECK-RV32-NEXT:    sw a3, 8(a0)
+; CHECK-RV32-NEXT:    sw a4, 4(a0)
+; CHECK-RV32-NEXT:    sw a1, 0(a0)
+; CHECK-RV32-NEXT:    ret
+;
+; CHECK-RV64C-LABEL: test_nontemporal_load_v4i32:
+; CHECK-RV64C:       # %bb.0:
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    ld a2, 8(a1)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    ld a1, 0(a1)
+; CHECK-RV64C-NEXT:    sd a2, 8(a0)
+; CHECK-RV64C-NEXT:    sd a1, 0(a0)
+; CHECK-RV64C-NEXT:    ret
+;
+; CHECK-RV32C-LABEL: test_nontemporal_load_v4i32:
+; CHECK-RV32C:       # %bb.0:
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    lw a2, 12(a1)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    lw a3, 8(a1)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    lw a4, 4(a1)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    lw a1, 0(a1)
+; CHECK-RV32C-NEXT:    sw a2, 12(a0)
+; CHECK-RV32C-NEXT:    sw a3, 8(a0)
+; CHECK-RV32C-NEXT:    sw a4, 4(a0)
+; CHECK-RV32C-NEXT:    sw a1, 0(a0)
+; CHECK-RV32C-NEXT:    ret
+;
+; CHECK-RV64V-LABEL: test_nontemporal_load_v4i32:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    vsetivli zero, 4, e32, m1, ta, ma
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    vle32.v v8, (a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_load_v4i32:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    vsetivli zero, 4, e32, m1, ta, ma
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    vle32.v v8, (a0)
+; CHECK-RV32V-NEXT:    ret
+
+  %1 = load <4 x i32>, ptr %p, !nontemporal !0
+  ret <4 x i32> %1
+}
+
+define <2 x i64> @test_nontemporal_load_v2i64(ptr %p) {
+; CHECK-RV64-LABEL: test_nontemporal_load_v2i64:
+; CHECK-RV64:       # %bb.0:
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    ld a2, 0(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    ld a1, 8(a0)
+; CHECK-RV64-NEXT:    mv a0, a2
+; CHECK-RV64-NEXT:    ret
+;
+; CHECK-RV32-LABEL: test_nontemporal_load_v2i64:
+; CHECK-RV32:       # %bb.0:
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    lw a2, 12(a1)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    lw a3, 8(a1)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    lw a4, 4(a1)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    lw a1, 0(a1)
+; CHECK-RV32-NEXT:    sw a2, 12(a0)
+; CHECK-RV32-NEXT:    sw a3, 8(a0)
+; CHECK-RV32-NEXT:    sw a4, 4(a0)
+; CHECK-RV32-NEXT:    sw a1, 0(a0)
+; CHECK-RV32-NEXT:    ret
+;
+; CHECK-RV64C-LABEL: test_nontemporal_load_v2i64:
+; CHECK-RV64C:       # %bb.0:
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    ld a2, 0(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    ld a1, 8(a0)
+; CHECK-RV64C-NEXT:    mv a0, a2
+; CHECK-RV64C-NEXT:    ret
+;
+; CHECK-RV32C-LABEL: test_nontemporal_load_v2i64:
+; CHECK-RV32C:       # %bb.0:
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    lw a2, 12(a1)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    lw a3, 8(a1)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    lw a4, 4(a1)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    lw a1, 0(a1)
+; CHECK-RV32C-NEXT:    sw a2, 12(a0)
+; CHECK-RV32C-NEXT:    sw a3, 8(a0)
+; CHECK-RV32C-NEXT:    sw a4, 4(a0)
+; CHECK-RV32C-NEXT:    sw a1, 0(a0)
+; CHECK-RV32C-NEXT:    ret
+;
+; CHECK-RV64V-LABEL: test_nontemporal_load_v2i64:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    vsetivli zero, 2, e64, m1, ta, ma
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    vle64.v v8, (a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_load_v2i64:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    vsetivli zero, 2, e64, m1, ta, ma
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    vle64.v v8, (a0)
+; CHECK-RV32V-NEXT:    ret
+
+  %1 = load <2 x i64>, ptr %p, !nontemporal !0
+  ret <2 x i64> %1
+}
+
+define void @test_nontemporal_store_i64(ptr %p, i64 %v) {
+; CHECK-RV64-LABEL: test_nontemporal_store_i64:
+; CHECK-RV64:       # %bb.0:
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sd a1, 0(a0)
+; CHECK-RV64-NEXT:    ret
+;
+; CHECK-RV32-LABEL: test_nontemporal_store_i64:
+; CHECK-RV32:       # %bb.0:
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sw a2, 4(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sw a1, 0(a0)
+; CHECK-RV32-NEXT:    ret
+;
+; CHECK-RV64C-LABEL: test_nontemporal_store_i64:
+; CHECK-RV64C:       # %bb.0:
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sd a1, 0(a0)
+; CHECK-RV64C-NEXT:    ret
+;
+; CHECK-RV32C-LABEL: test_nontemporal_store_i64:
+; CHECK-RV32C:       # %bb.0:
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sw a2, 4(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sw a1, 0(a0)
+; CHECK-RV32C-NEXT:    ret
+;
+; CHECK-RV64V-LABEL: test_nontemporal_store_i64:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    sd a1, 0(a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_store_i64:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    sw a2, 4(a0)
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    sw a1, 0(a0)
+; CHECK-RV32V-NEXT:    ret
+
+  store i64 %v, ptr %p, !nontemporal !0
+  ret void
+}
+
+define void @test_nontemporal_store_i32(ptr %p, i32 %v) {
+; CHECK-RV64-LABEL: test_nontemporal_store_i32:
+; CHECK-RV64:       # %bb.0:
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sw a1, 0(a0)
+; CHECK-RV64-NEXT:    ret
+;
+; CHECK-RV32-LABEL: test_nontemporal_store_i32:
+; CHECK-RV32:       # %bb.0:
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sw a1, 0(a0)
+; CHECK-RV32-NEXT:    ret
+;
+; CHECK-RV64C-LABEL: test_nontemporal_store_i32:
+; CHECK-RV64C:       # %bb.0:
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sw a1, 0(a0)
+; CHECK-RV64C-NEXT:    ret
+;
+; CHECK-RV32C-LABEL: test_nontemporal_store_i32:
+; CHECK-RV32C:       # %bb.0:
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sw a1, 0(a0)
+; CHECK-RV32C-NEXT:    ret
+;
+; CHECK-RV64V-LABEL: test_nontemporal_store_i32:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    sw a1, 0(a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_store_i32:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    sw a1, 0(a0)
+; CHECK-RV32V-NEXT:    ret
+
+  store i32 %v, ptr %p, !nontemporal !0
+  ret void
+}
+
+define void @test_nontemporal_store_i16(ptr %p, i16 %v) {
+; CHECK-RV64-LABEL: test_nontemporal_store_i16:
+; CHECK-RV64:       # %bb.0:
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sh a1, 0(a0)
+; CHECK-RV64-NEXT:    ret
+;
+; CHECK-RV32-LABEL: test_nontemporal_store_i16:
+; CHECK-RV32:       # %bb.0:
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sh a1, 0(a0)
+; CHECK-RV32-NEXT:    ret
+;
+; CHECK-RV64C-LABEL: test_nontemporal_store_i16:
+; CHECK-RV64C:       # %bb.0:
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sh a1, 0(a0)
+; CHECK-RV64C-NEXT:    ret
+;
+; CHECK-RV32C-LABEL: test_nontemporal_store_i16:
+; CHECK-RV32C:       # %bb.0:
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sh a1, 0(a0)
+; CHECK-RV32C-NEXT:    ret
+;
+; CHECK-RV64V-LABEL: test_nontemporal_store_i16:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    sh a1, 0(a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_store_i16:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    sh a1, 0(a0)
+; CHECK-RV32V-NEXT:    ret
+
+  store i16 %v, ptr %p, !nontemporal !0
+  ret void
+}
+
+define void @test_nontemporal_store_i8(ptr %p, i8 %v) {
+; CHECK-RV64-LABEL: test_nontemporal_store_i8:
+; CHECK-RV64:       # %bb.0:
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sb a1, 0(a0)
+; CHECK-RV64-NEXT:    ret
+;
+; CHECK-RV32-LABEL: test_nontemporal_store_i8:
+; CHECK-RV32:       # %bb.0:
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sb a1, 0(a0)
+; CHECK-RV32-NEXT:    ret
+;
+; CHECK-RV64C-LABEL: test_nontemporal_store_i8:
+; CHECK-RV64C:       # %bb.0:
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sb a1, 0(a0)
+; CHECK-RV64C-NEXT:    ret
+;
+; CHECK-RV32C-LABEL: test_nontemporal_store_i8:
+; CHECK-RV32C:       # %bb.0:
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sb a1, 0(a0)
+; CHECK-RV32C-NEXT:    ret
+;
+; CHECK-RV64V-LABEL: test_nontemporal_store_i8:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    sb a1, 0(a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_store_i8:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    sb a1, 0(a0)
+; CHECK-RV32V-NEXT:    ret
+
+  store i8 %v, ptr %p, !nontemporal !0
+  ret void
+}
+
+define void @test_nontemporal_store_half(ptr %p, half %v) {
+; CHECK-RV64-LABEL: test_nontemporal_store_half:
+; CHECK-RV64:       # %bb.0:
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    fsh fa0, 0(a0)
+; CHECK-RV64-NEXT:    ret
+;
+; CHECK-RV32-LABEL: test_nontemporal_store_half:
+; CHECK-RV32:       # %bb.0:
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    fsh fa0, 0(a0)
+; CHECK-RV32-NEXT:    ret
+;
+; CHECK-RV64C-LABEL: test_nontemporal_store_half:
+; CHECK-RV64C:       # %bb.0:
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    fsh fa0, 0(a0)
+; CHECK-RV64C-NEXT:    ret
+;
+; CHECK-RV32C-LABEL: test_nontemporal_store_half:
+; CHECK-RV32C:       # %bb.0:
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    fsh fa0, 0(a0)
+; CHECK-RV32C-NEXT:    ret
+;
+; CHECK-RV64V-LABEL: test_nontemporal_store_half:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    fsh fa0, 0(a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_store_half:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    fsh fa0, 0(a0)
+; CHECK-RV32V-NEXT:    ret
+
+  store half %v, ptr %p, !nontemporal !0
+  ret void
+}
+
+define void @test_nontemporal_store_float(ptr %p, float %v) {
+; CHECK-RV64-LABEL: test_nontemporal_store_float:
+; CHECK-RV64:       # %bb.0:
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    fsw fa0, 0(a0)
+; CHECK-RV64-NEXT:    ret
+;
+; CHECK-RV32-LABEL: test_nontemporal_store_float:
+; CHECK-RV32:       # %bb.0:
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    fsw fa0, 0(a0)
+; CHECK-RV32-NEXT:    ret
+;
+; CHECK-RV64C-LABEL: test_nontemporal_store_float:
+; CHECK-RV64C:       # %bb.0:
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    fsw fa0, 0(a0)
+; CHECK-RV64C-NEXT:    ret
+;
+; CHECK-RV32C-LABEL: test_nontemporal_store_float:
+; CHECK-RV32C:       # %bb.0:
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    fsw fa0, 0(a0)
+; CHECK-RV32C-NEXT:    ret
+;
+; CHECK-RV64V-LABEL: test_nontemporal_store_float:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    fsw fa0, 0(a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_store_float:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    fsw fa0, 0(a0)
+; CHECK-RV32V-NEXT:    ret
+
+  store float %v, ptr %p, !nontemporal !0
+  ret void
+}
+
+define void @test_nontemporal_store_double(ptr %p, double %v) {
+; CHECK-RV64-LABEL: test_nontemporal_store_double:
+; CHECK-RV64:       # %bb.0:
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    fsd fa0, 0(a0)
+; CHECK-RV64-NEXT:    ret
+;
+; CHECK-RV32-LABEL: test_nontemporal_store_double:
+; CHECK-RV32:       # %bb.0:
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    fsd fa0, 0(a0)
+; CHECK-RV32-NEXT:    ret
+;
+; CHECK-RV64C-LABEL: test_nontemporal_store_double:
+; CHECK-RV64C:       # %bb.0:
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    fsd fa0, 0(a0)
+; CHECK-RV64C-NEXT:    ret
+;
+; CHECK-RV32C-LABEL: test_nontemporal_store_double:
+; CHECK-RV32C:       # %bb.0:
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    fsd fa0, 0(a0)
+; CHECK-RV32C-NEXT:    ret
+;
+; CHECK-RV64V-LABEL: test_nontemporal_store_double:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    fsd fa0, 0(a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_store_double:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    fsd fa0, 0(a0)
+; CHECK-RV32V-NEXT:    ret
+
+  store double %v, ptr %p, !nontemporal !0
+  ret void
+}
+
+define void @test_nontemporal_store_v16i8(ptr %p, <16 x i8> %v) {
+; CHECK-RV64-LABEL: test_nontemporal_store_v16i8:
+; CHECK-RV64:       # %bb.0:
+; CHECK-RV64-NEXT:    addi sp, sp, -16
+; CHECK-RV64-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-RV64-NEXT:    sd s0, 8(sp) # 8-byte Folded Spill
+; CHECK-RV64-NEXT:    sd s1, 0(sp) # 8-byte Folded Spill
+; CHECK-RV64-NEXT:    .cfi_offset s0, -8
+; CHECK-RV64-NEXT:    .cfi_offset s1, -16
+; CHECK-RV64-NEXT:    lb a2, 0(a1)
+; CHECK-RV64-NEXT:    lb a3, 8(a1)
+; CHECK-RV64-NEXT:    lb a4, 16(a1)
+; CHECK-RV64-NEXT:    lb a5, 24(a1)
+; CHECK-RV64-NEXT:    lb a6, 32(a1)
+; CHECK-RV64-NEXT:    lb a7, 40(a1)
+; CHECK-RV64-NEXT:    lb t0, 48(a1)
+; CHECK-RV64-NEXT:    lb t1, 56(a1)
+; CHECK-RV64-NEXT:    lb t2, 64(a1)
+; CHECK-RV64-NEXT:    lb t3, 72(a1)
+; CHECK-RV64-NEXT:    lb t4, 80(a1)
+; CHECK-RV64-NEXT:    lb t5, 88(a1)
+; CHECK-RV64-NEXT:    lb t6, 120(a1)
+; CHECK-RV64-NEXT:    lb s0, 112(a1)
+; CHECK-RV64-NEXT:    lb s1, 104(a1)
+; CHECK-RV64-NEXT:    lb a1, 96(a1)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sb t6, 15(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sb s0, 14(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sb s1, 13(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sb a1, 12(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sb t5, 11(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sb t4, 10(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sb t3, 9(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sb t2, 8(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sb t1, 7(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sb t0, 6(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sb a7, 5(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sb a6, 4(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sb a5, 3(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sb a4, 2(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sb a3, 1(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sb a2, 0(a0)
+; CHECK-RV64-NEXT:    ld s0, 8(sp) # 8-byte Folded Reload
+; CHECK-RV64-NEXT:    ld s1, 0(sp) # 8-byte Folded Reload
+; CHECK-RV64-NEXT:    addi sp, sp, 16
+; CHECK-RV64-NEXT:    ret
+;
+; CHECK-RV32-LABEL: test_nontemporal_store_v16i8:
+; CHECK-RV32:       # %bb.0:
+; CHECK-RV32-NEXT:    addi sp, sp, -16
+; CHECK-RV32-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-RV32-NEXT:    sw s0, 12(sp) # 4-byte Folded Spill
+; CHECK-RV32-NEXT:    sw s1, 8(sp) # 4-byte Folded Spill
+; CHECK-RV32-NEXT:    .cfi_offset s0, -4
+; CHECK-RV32-NEXT:    .cfi_offset s1, -8
+; CHECK-RV32-NEXT:    lb a2, 0(a1)
+; CHECK-RV32-NEXT:    lb a3, 4(a1)
+; CHECK-RV32-NEXT:    lb a4, 8(a1)
+; CHECK-RV32-NEXT:    lb a5, 12(a1)
+; CHECK-RV32-NEXT:    lb a6, 16(a1)
+; CHECK-RV32-NEXT:    lb a7, 20(a1)
+; CHECK-RV32-NEXT:    lb t0, 24(a1)
+; CHECK-RV32-NEXT:    lb t1, 28(a1)
+; CHECK-RV32-NEXT:    lb t2, 32(a1)
+; CHECK-RV32-NEXT:    lb t3, 36(a1)
+; CHECK-RV32-NEXT:    lb t4, 40(a1)
+; CHECK-RV32-NEXT:    lb t5, 44(a1)
+; CHECK-RV32-NEXT:    lb t6, 60(a1)
+; CHECK-RV32-NEXT:    lb s0, 56(a1)
+; CHECK-RV32-NEXT:    lb s1, 52(a1)
+; CHECK-RV32-NEXT:    lb a1, 48(a1)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sb t6, 15(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sb s0, 14(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sb s1, 13(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sb a1, 12(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sb t5, 11(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sb t4, 10(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sb t3, 9(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sb t2, 8(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sb t1, 7(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sb t0, 6(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sb a7, 5(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sb a6, 4(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sb a5, 3(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sb a4, 2(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sb a3, 1(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sb a2, 0(a0)
+; CHECK-RV32-NEXT:    lw s0, 12(sp) # 4-byte Folded Reload
+; CHECK-RV32-NEXT:    lw s1, 8(sp) # 4-byte Folded Reload
+; CHECK-RV32-NEXT:    addi sp, sp, 16
+; CHECK-RV32-NEXT:    ret
+;
+; CHECK-RV64C-LABEL: test_nontemporal_store_v16i8:
+; CHECK-RV64C:       # %bb.0:
+; CHECK-RV64C-NEXT:    addi sp, sp, -16
+; CHECK-RV64C-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-RV64C-NEXT:    sd s0, 8(sp) # 8-byte Folded Spill
+; CHECK-RV64C-NEXT:    sd s1, 0(sp) # 8-byte Folded Spill
+; CHECK-RV64C-NEXT:    .cfi_offset s0, -8
+; CHECK-RV64C-NEXT:    .cfi_offset s1, -16
+; CHECK-RV64C-NEXT:    lb a6, 0(a1)
+; CHECK-RV64C-NEXT:    lb a7, 8(a1)
+; CHECK-RV64C-NEXT:    lb t0, 16(a1)
+; CHECK-RV64C-NEXT:    lb t1, 24(a1)
+; CHECK-RV64C-NEXT:    lb t2, 32(a1)
+; CHECK-RV64C-NEXT:    lb t3, 40(a1)
+; CHECK-RV64C-NEXT:    lb t4, 48(a1)
+; CHECK-RV64C-NEXT:    lb t5, 56(a1)
+; CHECK-RV64C-NEXT:    lb t6, 64(a1)
+; CHECK-RV64C-NEXT:    lb a3, 72(a1)
+; CHECK-RV64C-NEXT:    lb a4, 80(a1)
+; CHECK-RV64C-NEXT:    lb a5, 88(a1)
+; CHECK-RV64C-NEXT:    lb a2, 120(a1)
+; CHECK-RV64C-NEXT:    lb s0, 112(a1)
+; CHECK-RV64C-NEXT:    lb s1, 104(a1)
+; CHECK-RV64C-NEXT:    lb a1, 96(a1)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sb a2, 15(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sb s0, 14(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sb s1, 13(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sb a1, 12(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sb a5, 11(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sb a4, 10(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sb a3, 9(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sb t6, 8(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sb t5, 7(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sb t4, 6(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sb t3, 5(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sb t2, 4(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sb t1, 3(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sb t0, 2(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sb a7, 1(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sb a6, 0(a0)
+; CHECK-RV64C-NEXT:    ld s0, 8(sp) # 8-byte Folded Reload
+; CHECK-RV64C-NEXT:    ld s1, 0(sp) # 8-byte Folded Reload
+; CHECK-RV64C-NEXT:    addi sp, sp, 16
+; CHECK-RV64C-NEXT:    ret
+;
+; CHECK-RV32C-LABEL: test_nontemporal_store_v16i8:
+; CHECK-RV32C:       # %bb.0:
+; CHECK-RV32C-NEXT:    addi sp, sp, -16
+; CHECK-RV32C-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-RV32C-NEXT:    sw s0, 12(sp) # 4-byte Folded Spill
+; CHECK-RV32C-NEXT:    sw s1, 8(sp) # 4-byte Folded Spill
+; CHECK-RV32C-NEXT:    .cfi_offset s0, -4
+; CHECK-RV32C-NEXT:    .cfi_offset s1, -8
+; CHECK-RV32C-NEXT:    lb a6, 0(a1)
+; CHECK-RV32C-NEXT:    lb a7, 4(a1)
+; CHECK-RV32C-NEXT:    lb t0, 8(a1)
+; CHECK-RV32C-NEXT:    lb t1, 12(a1)
+; CHECK-RV32C-NEXT:    lb t2, 16(a1)
+; CHECK-RV32C-NEXT:    lb t3, 20(a1)
+; CHECK-RV32C-NEXT:    lb t4, 24(a1)
+; CHECK-RV32C-NEXT:    lb t5, 28(a1)
+; CHECK-RV32C-NEXT:    lb t6, 32(a1)
+; CHECK-RV32C-NEXT:    lb a3, 36(a1)
+; CHECK-RV32C-NEXT:    lb a4, 40(a1)
+; CHECK-RV32C-NEXT:    lb a5, 44(a1)
+; CHECK-RV32C-NEXT:    lb a2, 60(a1)
+; CHECK-RV32C-NEXT:    lb s0, 56(a1)
+; CHECK-RV32C-NEXT:    lb s1, 52(a1)
+; CHECK-RV32C-NEXT:    lb a1, 48(a1)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sb a2, 15(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sb s0, 14(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sb s1, 13(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sb a1, 12(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sb a5, 11(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sb a4, 10(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sb a3, 9(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sb t6, 8(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sb t5, 7(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sb t4, 6(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sb t3, 5(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sb t2, 4(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sb t1, 3(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sb t0, 2(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sb a7, 1(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sb a6, 0(a0)
+; CHECK-RV32C-NEXT:    lw s0, 12(sp) # 4-byte Folded Reload
+; CHECK-RV32C-NEXT:    lw s1, 8(sp) # 4-byte Folded Reload
+; CHECK-RV32C-NEXT:    addi sp, sp, 16
+; CHECK-RV32C-NEXT:    ret
+;
+; CHECK-RV64V-LABEL: test_nontemporal_store_v16i8:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    vsetivli zero, 16, e8, m1, ta, ma
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    vse8.v v8, (a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_store_v16i8:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    vsetivli zero, 16, e8, m1, ta, ma
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    vse8.v v8, (a0)
+; CHECK-RV32V-NEXT:    ret
+  store <16 x i8> %v, ptr %p, !nontemporal !0
+  ret void
+}
+
+define void @test_nontemporal_store_v8i16(ptr %p, <8 x i16> %v) {
+; CHECK-RV64-LABEL: test_nontemporal_store_v8i16:
+; CHECK-RV64:       # %bb.0:
+; CHECK-RV64-NEXT:    lh a2, 0(a1)
+; CHECK-RV64-NEXT:    lh a3, 8(a1)
+; CHECK-RV64-NEXT:    lh a4, 16(a1)
+; CHECK-RV64-NEXT:    lh a5, 24(a1)
+; CHECK-RV64-NEXT:    lh a6, 56(a1)
+; CHECK-RV64-NEXT:    lh a7, 48(a1)
+; CHECK-RV64-NEXT:    lh t0, 40(a1)
+; CHECK-RV64-NEXT:    lh a1, 32(a1)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sh a6, 14(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sh a7, 12(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sh t0, 10(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sh a1, 8(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sh a5, 6(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sh a4, 4(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sh a3, 2(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sh a2, 0(a0)
+; CHECK-RV64-NEXT:    ret
+;
+; CHECK-RV32-LABEL: test_nontemporal_store_v8i16:
+; CHECK-RV32:       # %bb.0:
+; CHECK-RV32-NEXT:    lh a2, 0(a1)
+; CHECK-RV32-NEXT:    lh a3, 4(a1)
+; CHECK-RV32-NEXT:    lh a4, 8(a1)
+; CHECK-RV32-NEXT:    lh a5, 12(a1)
+; CHECK-RV32-NEXT:    lh a6, 28(a1)
+; CHECK-RV32-NEXT:    lh a7, 24(a1)
+; CHECK-RV32-NEXT:    lh t0, 20(a1)
+; CHECK-RV32-NEXT:    lh a1, 16(a1)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sh a6, 14(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sh a7, 12(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sh t0, 10(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sh a1, 8(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sh a5, 6(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sh a4, 4(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sh a3, 2(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sh a2, 0(a0)
+; CHECK-RV32-NEXT:    ret
+;
+; CHECK-RV64C-LABEL: test_nontemporal_store_v8i16:
+; CHECK-RV64C:       # %bb.0:
+; CHECK-RV64C-NEXT:    lh a6, 0(a1)
+; CHECK-RV64C-NEXT:    lh a7, 8(a1)
+; CHECK-RV64C-NEXT:    lh t0, 16(a1)
+; CHECK-RV64C-NEXT:    lh a5, 24(a1)
+; CHECK-RV64C-NEXT:    lh a2, 56(a1)
+; CHECK-RV64C-NEXT:    lh a3, 48(a1)
+; CHECK-RV64C-NEXT:    lh a4, 40(a1)
+; CHECK-RV64C-NEXT:    lh a1, 32(a1)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sh a2, 14(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sh a3, 12(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sh a4, 10(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sh a1, 8(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sh a5, 6(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sh t0, 4(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sh a7, 2(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sh a6, 0(a0)
+; CHECK-RV64C-NEXT:    ret
+;
+; CHECK-RV32C-LABEL: test_nontemporal_store_v8i16:
+; CHECK-RV32C:       # %bb.0:
+; CHECK-RV32C-NEXT:    lh a6, 0(a1)
+; CHECK-RV32C-NEXT:    lh a7, 4(a1)
+; CHECK-RV32C-NEXT:    lh t0, 8(a1)
+; CHECK-RV32C-NEXT:    lh a5, 12(a1)
+; CHECK-RV32C-NEXT:    lh a2, 28(a1)
+; CHECK-RV32C-NEXT:    lh a3, 24(a1)
+; CHECK-RV32C-NEXT:    lh a4, 20(a1)
+; CHECK-RV32C-NEXT:    lh a1, 16(a1)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sh a2, 14(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sh a3, 12(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sh a4, 10(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sh a1, 8(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sh a5, 6(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sh t0, 4(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sh a7, 2(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sh a6, 0(a0)
+; CHECK-RV32C-NEXT:    ret
+;
+; CHECK-RV64V-LABEL: test_nontemporal_store_v8i16:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    vsetivli zero, 8, e16, m1, ta, ma
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    vse16.v v8, (a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_store_v8i16:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    vsetivli zero, 8, e16, m1, ta, ma
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    vse16.v v8, (a0)
+; CHECK-RV32V-NEXT:    ret
+  store <8 x i16> %v, ptr %p, !nontemporal !0
+  ret void
+}
+
+define void @test_nontemporal_store_v4i32(ptr %p, <4 x i32> %v) {
+; CHECK-RV64-LABEL: test_nontemporal_store_v4i32:
+; CHECK-RV64:       # %bb.0:
+; CHECK-RV64-NEXT:    lw a2, 24(a1)
+; CHECK-RV64-NEXT:    lw a3, 16(a1)
+; CHECK-RV64-NEXT:    lw a4, 8(a1)
+; CHECK-RV64-NEXT:    lw a1, 0(a1)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sw a2, 12(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sw a3, 8(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sw a4, 4(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sw a1, 0(a0)
+; CHECK-RV64-NEXT:    ret
+;
+; CHECK-RV32-LABEL: test_nontemporal_store_v4i32:
+; CHECK-RV32:       # %bb.0:
+; CHECK-RV32-NEXT:    lw a2, 12(a1)
+; CHECK-RV32-NEXT:    lw a3, 8(a1)
+; CHECK-RV32-NEXT:    lw a4, 4(a1)
+; CHECK-RV32-NEXT:    lw a1, 0(a1)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sw a2, 12(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sw a3, 8(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sw a4, 4(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sw a1, 0(a0)
+; CHECK-RV32-NEXT:    ret
+;
+; CHECK-RV64C-LABEL: test_nontemporal_store_v4i32:
+; CHECK-RV64C:       # %bb.0:
+; CHECK-RV64C-NEXT:    lw a2, 24(a1)
+; CHECK-RV64C-NEXT:    lw a3, 16(a1)
+; CHECK-RV64C-NEXT:    lw a4, 8(a1)
+; CHECK-RV64C-NEXT:    lw a1, 0(a1)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sw a2, 12(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sw a3, 8(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sw a4, 4(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sw a1, 0(a0)
+; CHECK-RV64C-NEXT:    ret
+;
+; CHECK-RV32C-LABEL: test_nontemporal_store_v4i32:
+; CHECK-RV32C:       # %bb.0:
+; CHECK-RV32C-NEXT:    lw a2, 12(a1)
+; CHECK-RV32C-NEXT:    lw a3, 8(a1)
+; CHECK-RV32C-NEXT:    lw a4, 4(a1)
+; CHECK-RV32C-NEXT:    lw a1, 0(a1)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sw a2, 12(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sw a3, 8(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sw a4, 4(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sw a1, 0(a0)
+; CHECK-RV32C-NEXT:    ret
+;
+; CHECK-RV64V-LABEL: test_nontemporal_store_v4i32:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    vsetivli zero, 4, e32, m1, ta, ma
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    vse32.v v8, (a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_store_v4i32:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    vsetivli zero, 4, e32, m1, ta, ma
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    vse32.v v8, (a0)
+; CHECK-RV32V-NEXT:    ret
+  store <4 x i32> %v, ptr %p, !nontemporal !0
+  ret void
+}
+
+define void @test_nontemporal_store_v2i64(ptr %p, <2 x i64> %v) {
+; CHECK-RV64-LABEL: test_nontemporal_store_v2i64:
+; CHECK-RV64:       # %bb.0:
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sd a2, 8(a0)
+; CHECK-RV64-NEXT:    ntl.all
+; CHECK-RV64-NEXT:    sd a1, 0(a0)
+; CHECK-RV64-NEXT:    ret
+;
+; CHECK-RV32-LABEL: test_nontemporal_store_v2i64:
+; CHECK-RV32:       # %bb.0:
+; CHECK-RV32-NEXT:    lw a2, 12(a1)
+; CHECK-RV32-NEXT:    lw a3, 8(a1)
+; CHECK-RV32-NEXT:    lw a4, 4(a1)
+; CHECK-RV32-NEXT:    lw a1, 0(a1)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sw a2, 12(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sw a3, 8(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sw a4, 4(a0)
+; CHECK-RV32-NEXT:    ntl.all
+; CHECK-RV32-NEXT:    sw a1, 0(a0)
+; CHECK-RV32-NEXT:    ret
+;
+; CHECK-RV64C-LABEL: test_nontemporal_store_v2i64:
+; CHECK-RV64C:       # %bb.0:
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sd a2, 8(a0)
+; CHECK-RV64C-NEXT:    c.ntl.all
+; CHECK-RV64C-NEXT:    sd a1, 0(a0)
+; CHECK-RV64C-NEXT:    ret
+;
+; CHECK-RV32C-LABEL: test_nontemporal_store_v2i64:
+; CHECK-RV32C:       # %bb.0:
+; CHECK-RV32C-NEXT:    lw a2, 12(a1)
+; CHECK-RV32C-NEXT:    lw a3, 8(a1)
+; CHECK-RV32C-NEXT:    lw a4, 4(a1)
+; CHECK-RV32C-NEXT:    lw a1, 0(a1)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sw a2, 12(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sw a3, 8(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sw a4, 4(a0)
+; CHECK-RV32C-NEXT:    c.ntl.all
+; CHECK-RV32C-NEXT:    sw a1, 0(a0)
+; CHECK-RV32C-NEXT:    ret
+;
+; CHECK-RV64V-LABEL: test_nontemporal_store_v2i64:
+; CHECK-RV64V:       # %bb.0:
+; CHECK-RV64V-NEXT:    vsetivli zero, 2, e64, m1, ta, ma
+; CHECK-RV64V-NEXT:    ntl.all
+; CHECK-RV64V-NEXT:    vse64.v v8, (a0)
+; CHECK-RV64V-NEXT:    ret
+;
+; CHECK-RV32V-LABEL: test_nontemporal_store_v2i64:
+; CHECK-RV32V:       # %bb.0:
+; CHECK-RV32V-NEXT:    vsetivli zero, 2, e64, m1, ta, ma
+; CHECK-RV32V-NEXT:    ntl.all
+; CHECK-RV32V-NEXT:    vse64.v v8, (a0)
+; CHECK-RV32V-NEXT:    ret
+  store <2 x i64> %v, ptr %p, !nontemporal !0
+  ret void
+}
+
+!0 = !{i32 1}