[VE] Dynamic stack allocation

[simoll]

Summary

This patch implements dynamic stack allocation for the VE target. Changes:

• compiler-rt: __ve_grow_stack to request stack allocation on the VE.
• VE: base pointer support, dynamic stack allocation.

[arsenm]

llvm/lib/Target/VE/VEInstrInfo.cpp

555

getSubtarget<VESubtarget> and remove the cast. You also added Subtarget as a member, so you don't need the get anyway

563

getSubtarget<VESubtarget>() and use the VEFrameLowering?

[simoll]

• @arsenm 's comments
• rebased
• fixed commit attribution (to @kaz7 )

[simoll]

• rebased
• formatting

[simoll]

ping. Any more comments for this?

[arsenm]

llvm/include/llvm/IR/CallingConv.h

244–247

Seems weird to me that this needs its own CC, or that it would have llvm in the name. This should be split into a separate patch, and needs some bitcode encoding tests

[simoll]

Will try to get rid of the custom CC. Thanks for having another look at this!

[simoll]

• use standard C CC for grow_stack builtins.
• added test for aligned stack allocation.
• renamed builtins to __builtin_grow_stack[_align].
• rebased.

[arsenm]

compiler-rt/lib/builtins/ve/grow_stack_align.S

17 ↗

(On Diff #262581)

Calling this builtin seems a bit wrong to me. builtin is usually a clang/C/frontend wrapper around an IR intrinsic. __ve_grow_stack_align? If it's already called this in some ABI spec then it doesn't matter

[simoll]

• Renamed to __ve_grow_stack . The sequence has no specific name in the documentation, we can simply use a more palatable name.
• Minor cleanup.
• Rebased.

[simoll]

• made lint a little happier with camelCaseFuncNames.
• rebased.

[kaz7]

In D79084#2024595, @simoll wrote:

• use standard C CC for grow_stack builtins.

@simoll @arsenm

Sorry for late reply. I originally want to use it's own calling convention to allow caller to not spill registers since many program may call this function to grow their stack. Is it not acceptable in general? Or it's OK if there is a reason?

It's OK to upstreaming C CC as a first step. I would like to know what we should do to improve the performance of alloca or dynamic stack allocation in C++. Thanks!

[arsenm]

In D79084#2042936, @kaz7 wrote:

In D79084#2024595, @simoll wrote:

• use standard C CC for grow_stack builtins.

@simoll @arsenm

Sorry for late reply. I originally want to use it's own calling convention to allow caller to not spill registers since many program may call this function to grow their stack. Is it not acceptable in general? Or it's OK if there is a reason?

It's OK to upstreaming C CC as a first step. I would like to know what we should do to improve the performance of alloca or dynamic stack allocation in C++. Thanks!

Would this just be the same as the existing PreserveMost or PreserveAll conventions?

[kaz7]

@arsenm Yes! I didn't notice those calling conventions. We can use them as our target depend special calling conventions. Thank you for informing us.

[simoll]

• Use 'PreserveAll' CC for ve_grow_stack.
• Rebased.

[kaz7]

verifyLeafProcRegUse function is not used anymore.

llvm/lib/Target/VE/VEFrameLowering.cpp

354

Remove this function from upstreaming. This is obsoleted function.

[simoll]

remove verifyLeafProcRegUse

[simoll]

• NFC, Lint issues.
• rebased.

[simoll]

Any feedback for this one?

[arsenm]

LGTM with nits

llvm/lib/Target/VE/VEFrameLowering.cpp

337

Else after return

346

No else after return

[simoll]

• Fixed else after return
• Rebased

Thanks!

compiler-rt/cmake/Modules/CompilerRTUtils.cmake

macro(detect_target_arch)
   check_symbol_exists(__powerpc64__ "" __PPC64)
   check_symbol_exists(__powerpc64le__ "" __PPC64LE)
   check_symbol_exists(__riscv "" __RISCV)
   check_symbol_exists(__s390x__ "" __S390X)
   check_symbol_exists(__sparc "" __SPARC)
   check_symbol_exists(__sparcv9 "" __SPARCV9)
   check_symbol_exists(__wasm32__ "" __WEBASSEMBLY32)
   check_symbol_exists(__wasm64__ "" __WEBASSEMBLY64)
+  check_symbol_exists(__ve__ "" __VE)
   if(__ARM)
     add_default_target_arch(arm)
   elseif(__AARCH64)
     add_default_target_arch(aarch64)
   elseif(__X86_64)
     add_default_target_arch(x86_64)
   elseif(__I386)
     add_default_target_arch(i386)
macro(detect_target_arch)
   elseif(__SPARCV9)
     add_default_target_arch(sparcv9)
   elseif(__SPARC)
     add_default_target_arch(sparc)
   elseif(__WEBASSEMBLY32)
     add_default_target_arch(wasm32)
   elseif(__WEBASSEMBLY64)
     add_default_target_arch(wasm64)
+  elseif(__VE)
+    add_default_target_arch(ve)
   endif()
 endmacro()
 
 macro(load_llvm_config)
   if (NOT LLVM_CONFIG_PATH)
     find_program(LLVM_CONFIG_PATH "llvm-config"
                  DOC "Path to llvm-config binary")
     if (NOT LLVM_CONFIG_PATH)

compiler-rt/cmake/base-config-ix.cmake

elseif(NOT APPLE) # Supported archs for Apple platforms are generated later
     elseif("${COMPILER_RT_DEFAULT_TARGET_ARCH}" MATCHES "riscv32")
       test_target_arch(riscv32 "" "")
     elseif("${COMPILER_RT_DEFAULT_TARGET_ARCH}" MATCHES "riscv64")
       test_target_arch(riscv64 "" "")
     elseif("${COMPILER_RT_DEFAULT_TARGET_ARCH}" MATCHES "wasm32")
       test_target_arch(wasm32 "" "--target=wasm32-unknown-unknown")
     elseif("${COMPILER_RT_DEFAULT_TARGET_ARCH}" MATCHES "wasm64")
       test_target_arch(wasm64 "" "--target=wasm64-unknown-unknown")
+    elseif("${COMPILER_RT_DEFAULT_TARGET_ARCH}" MATCHES "ve")
+      test_target_arch(ve "__ve__" "--target=ve-unknown-none")
     endif()
     set(COMPILER_RT_OS_SUFFIX "")
   endif()
 endmacro()

compiler-rt/cmake/builtin-config-ix.cmake

 set(MIPS64 mips64 mips64el)
 set(PPC64 powerpc64 powerpc64le)
 set(RISCV32 riscv32)
 set(RISCV64 riscv64)
 set(SPARC sparc)
 set(SPARCV9 sparcv9)
 set(WASM32 wasm32)
 set(WASM64 wasm64)
+set(VE ve)
 
 if(APPLE)
   set(ARM64 arm64 arm64e)
   set(ARM32 armv7 armv7k armv7s)
   set(X86_64 x86_64 x86_64h)
 endif()
 
-set(ALL_BUILTIN_SUPPORTED_ARCH ${X86} ${X86_64} ${ARM32} ${ARM64}
-    ${HEXAGON} ${MIPS32} ${MIPS64} ${PPC64} ${RISCV32} ${RISCV64} ${SPARC} ${SPARCV9} ${WASM32} ${WASM64})
+set(ALL_BUILTIN_SUPPORTED_ARCH
+  ${X86} ${X86_64} ${ARM32} ${ARM64}
+  ${HEXAGON} ${MIPS32} ${MIPS64} ${PPC64}
+  ${RISCV32} ${RISCV64} ${SPARC} ${SPARCV9}
+  ${WASM32} ${WASM64} ${VE})
 
 include(CompilerRTUtils)
 include(CompilerRTDarwinUtils)
 
 if(APPLE)
 
   find_darwin_sdk_dir(DARWIN_osx_SYSROOT macosx)
   find_darwin_sdk_dir(DARWIN_iossim_SYSROOT iphonesimulator)

compiler-rt/lib/builtins/CMakeLists.txt

set(wasm32_SOURCES
   ${GENERIC_TF_SOURCES}
   ${GENERIC_SOURCES}
 )
 set(wasm64_SOURCES
   ${GENERIC_TF_SOURCES}
   ${GENERIC_SOURCES}
 )
 
+set(ve_SOURCES
+  ve/llvm_grow_stack.S
+  ve/llvm_grow_stack_align.S
+  ${GENERIC_TF_SOURCES}
+  ${GENERIC_SOURCES})
+
 add_custom_target(builtins)
 set_target_properties(builtins PROPERTIES FOLDER "Compiler-RT Misc")
 
 if (APPLE)
   add_subdirectory(Darwin-excludes)
   add_subdirectory(macho_embedded)
   darwin_add_builtin_libraries(${BUILTIN_SUPPORTED_OS})
 else ()

compiler-rt/lib/builtins/ve/llvm_grow_stack.S

+// 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
+
+#include "../assembly.h"
+
+// grow_stack routine
+// This routine is VE specific
+// https://www.nec.com/en/global/prod/hpc/aurora/document/VE-ABI_v1.1.pdf
+
+// destroy %s62 and %s63 only
+
+#ifdef __ve__
+
+.text
+.p2align        4
+DEFINE_COMPILERRT_FUNCTION(__llvm_grow_stack)
+        subu.l          %sp, %sp, %s0           # sp -= alloca size
+        and             %sp, -16, %sp           # align sp
+        brge.l.t        %sp, %sl, 1f
+        ld              %s63, 0x18(,%tp)        # load param area
+        lea             %s62, 0x13b             # syscall # of grow
+        shm.l           %s62, 0x0(%s63)         # stored at addr:0
+        shm.l           %sl, 0x8(%s63)          # old limit at addr:8
+        shm.l           %sp, 0x10(%s63)         # new limit at addr:16
+        monc
+1:
+        b.l             (,%lr)
+END_COMPILERRT_FUNCTION(__llvm_grow_stack)
+
+#endif // __ve__

compiler-rt/lib/builtins/ve/llvm_grow_stack_align.S

+// 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
+
+#include "../assembly.h"
+
+// grow_stack routine
+// This routine is VE specific
+// https://www.nec.com/en/global/prod/hpc/aurora/document/VE-ABI_v1.1.pdf
+
+// destroy %s62 and %s63 only
+
+#ifdef __ve__
+
+.text
+.p2align        4
+DEFINE_COMPILERRT_FUNCTION(__llvm_grow_stack_align)
+        subu.l          %sp, %sp, %s0           # sp -= alloca size
+        and             %sp, %sp, %s1           # align sp
+        brge.l.t        %sp, %sl, 1f
+        ld              %s63, 0x18(,%tp)        # load param area
+        lea             %s62, 0x13b             # syscall # of grow
+        shm.l           %s62, 0x0(%s63)         # stored at addr:0
+        shm.l           %sl, 0x8(%s63)          # old limit at addr:8
+        shm.l           %sp, 0x10(%s63)         # new limit at addr:16
+        monc
+1:
+        b.l             (,%lr)
+END_COMPILERRT_FUNCTION(__llvm_grow_stack_align)
+
+#endif // __ve__

llvm/include/llvm/IR/CallingConv.h

enum {
     /// Calling convention between AArch64 SVE functions
     AArch64_SVE_VectorCall = 98,
 
     /// Calling convention for emscripten __invoke_* functions. The first
     /// argument is required to be the function ptr being indirectly called.
     /// The remainder matches the regular calling convention.
     WASM_EmscriptenInvoke = 99,
 
+    /// Calling convention used for NEC SX-Aurora VE llvm_grow_stack intrinsic
+    /// function.
+    VE_LLVM_GROW_STACK = 100,
+
     /// The highest possible calling convention ID. Must be some 2^k - 1.
     MaxID = 1023
   };
 
 } // end namespace CallingConv
 
 } // end namespace llvm
 
 #endif // LLVM_IR_CALLINGCONV_H

llvm/lib/Target/VE/VECallingConv.td

def RetCC_VE : CallingConv<[
   // long long/double --> generic 64 bit registers
   CCIfType<[i64, f64],
            CCAssignToReg<[SX0, SX1, SX2, SX3, SX4, SX5, SX6, SX7]>>,
 ]>;
 
 // Callee-saved registers
 def CSR : CalleeSavedRegs<(add (sequence "SX%u", 18, 33))>;
 def CSR_NoRegs : CalleeSavedRegs<(add)>;
+
+// llvm_grow_stack destroys s62 and s63
+def CSR_llvm_grow_stack : CalleeSavedRegs<(add (sequence "SX%u", 0, 61))>;

llvm/lib/Target/VE/VEFrameLowering.h

 public:
   explicit VEFrameLowering(const VESubtarget &ST);
 
   /// emitProlog/emitEpilog - These methods insert prolog and epilog code into
   /// the function.
   void emitPrologue(MachineFunction &MF, MachineBasicBlock &MBB) const override;
   void emitEpilogue(MachineFunction &MF, MachineBasicBlock &MBB) const override;
   void emitPrologueInsns(MachineFunction &MF, MachineBasicBlock &MBB,
-                         MachineBasicBlock::iterator MBBI, int NumBytes,
+                         MachineBasicBlock::iterator MBBI, uint64_t NumBytes,
                          bool RequireFPUpdate) const;
   void emitEpilogueInsns(MachineFunction &MF, MachineBasicBlock &MBB,
-                         MachineBasicBlock::iterator MBBI, int NumBytes,
+                         MachineBasicBlock::iterator MBBI, uint64_t NumBytes,
                          bool RequireFPUpdate) const;
 
   MachineBasicBlock::iterator
   eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
                                 MachineBasicBlock::iterator I) const override;
 
-  bool hasReservedCallFrame(const MachineFunction &MF) const override;
+  bool hasBP(const MachineFunction &MF) const;
   bool hasFP(const MachineFunction &MF) const override;
+  // VE reserves argument space always for call sites in the function
+  // immediately on entry of the current function.
+  bool hasReservedCallFrame(const MachineFunction &MF) const override {
+    return true;
+  }
   void determineCalleeSaves(MachineFunction &MF, BitVector &SavedRegs,
                             RegScavenger *RS = nullptr) const override;
 
   int getFrameIndexReference(const MachineFunction &MF, int FI,
                              Register &FrameReg) const override;
 
   const SpillSlot *
   getCalleeSavedSpillSlots(unsigned &NumEntries) const override {
     static const SpillSlot Offsets[] = {
         {VE::SX17, 40},  {VE::SX18, 48},  {VE::SX19, 56},  {VE::SX20, 64},
         {VE::SX21, 72},  {VE::SX22, 80},  {VE::SX23, 88},  {VE::SX24, 96},
         {VE::SX25, 104}, {VE::SX26, 112}, {VE::SX27, 120}, {VE::SX28, 128},
         {VE::SX29, 136}, {VE::SX30, 144}, {VE::SX31, 152}, {VE::SX32, 160},
         {VE::SX33, 168}};
     NumEntries = array_lengthof(Offsets);
     return Offsets;
   }
 
-  /// targetHandlesStackFrameRounding - Returns true if the target is
-  /// responsible for rounding up the stack frame (probably at emitPrologue
-  /// time).
-  bool targetHandlesStackFrameRounding() const override { return true; }
+protected:
+  const VESubtarget &STI;
 
 private:
   // Returns true if MF is a leaf procedure.
   bool isLeafProc(MachineFunction &MF) const;
 
   // Emits code for adjusting SP in function prologue/epilogue.
   void emitSPAdjustment(MachineFunction &MF, MachineBasicBlock &MBB,
-                        MachineBasicBlock::iterator MBBI, int NumBytes) const;
+                        MachineBasicBlock::iterator MBBI, int64_t NumBytes,
+                        MaybeAlign MayAlign = MaybeAlign()) const;
 
   // Emits code for extending SP in function prologue/epilogue.
   void emitSPExtend(MachineFunction &MF, MachineBasicBlock &MBB,
-                    MachineBasicBlock::iterator MBBI, int NumBytes) const;
+                    MachineBasicBlock::iterator MBBI) const;
 };
 
 } // namespace llvm
 
 #endif

llvm/lib/Target/VE/VEFrameLowering.cpp

 #include "llvm/Support/CommandLine.h"
 #include "llvm/Target/TargetOptions.h"
 #include "llvm/Support/MathExtras.h"
 
 using namespace llvm;
 
 VEFrameLowering::VEFrameLowering(const VESubtarget &ST)
     : TargetFrameLowering(TargetFrameLowering::StackGrowsDown, Align(16), 0,
-                          Align(16)) {}
+                          Align(16)),
+      STI(ST) {}
 
 void VEFrameLowering::emitPrologueInsns(MachineFunction &MF,
                                         MachineBasicBlock &MBB,
                                         MachineBasicBlock::iterator MBBI,
-                                        int NumBytes,
+                                        uint64_t NumBytes,
                                         bool RequireFPUpdate) const {
 
   DebugLoc dl;
   const VEInstrInfo &TII =
       *static_cast<const VEInstrInfo *>(MF.getSubtarget().getInstrInfo());
   // Insert following codes here as prologue
   //
   //    st %fp, 0(,%sp)
   //    st %lr, 8(,%sp)
   //    st %got, 24(,%sp)
   //    st %plt, 32(,%sp)
+  //    st %s17, 40(,%sp) iff this function is using s17 as BP
   //    or %fp, 0, %sp
 
   BuildMI(MBB, MBBI, dl, TII.get(VE::STrii))
       .addReg(VE::SX11)
       .addImm(0)
       .addImm(0)
       .addReg(VE::SX9);
   BuildMI(MBB, MBBI, dl, TII.get(VE::STrii))
       .addReg(VE::SX11)
       .addImm(0)
       .addImm(8)
       .addReg(VE::SX10);
   BuildMI(MBB, MBBI, dl, TII.get(VE::STrii))
       .addReg(VE::SX11)
       .addImm(0)
       .addImm(24)
       .addReg(VE::SX15);
   BuildMI(MBB, MBBI, dl, TII.get(VE::STrii))
       .addReg(VE::SX11)
       .addImm(0)
       .addImm(32)
       .addReg(VE::SX16);
+  if (hasBP(MF))
+    BuildMI(MBB, MBBI, dl, TII.get(VE::STrii))
+        .addReg(VE::SX11)
+        .addImm(0)
+        .addImm(40)
+        .addReg(VE::SX17);
   BuildMI(MBB, MBBI, dl, TII.get(VE::ORri), VE::SX9)
       .addReg(VE::SX11)
       .addImm(0);
 }
 
 void VEFrameLowering::emitEpilogueInsns(MachineFunction &MF,
                                         MachineBasicBlock &MBB,
                                         MachineBasicBlock::iterator MBBI,
-                                        int NumBytes,
+                                        uint64_t NumBytes,
                                         bool RequireFPUpdate) const {
 
   DebugLoc dl;
   const VEInstrInfo &TII =
       *static_cast<const VEInstrInfo *>(MF.getSubtarget().getInstrInfo());
   // Insert following codes here as epilogue
   //
   //    or %sp, 0, %fp
+  //    ld %s17, 40(,%sp) iff this function is using s17 as BP
   //    ld %got, 32(,%sp)
   //    ld %plt, 24(,%sp)
   //    ld %lr, 8(,%sp)
   //    ld %fp, 0(,%sp)
 
   BuildMI(MBB, MBBI, dl, TII.get(VE::ORri), VE::SX11)
       .addReg(VE::SX9)
       .addImm(0);
+  if (hasBP(MF))
+    BuildMI(MBB, MBBI, dl, TII.get(VE::LDrii), VE::SX17)
+        .addReg(VE::SX11)
+        .addImm(0)
+        .addImm(40);
   BuildMI(MBB, MBBI, dl, TII.get(VE::LDrii), VE::SX16)
       .addReg(VE::SX11)
       .addImm(0)
       .addImm(32);
   BuildMI(MBB, MBBI, dl, TII.get(VE::LDrii), VE::SX15)
       .addReg(VE::SX11)
       .addImm(0)
       .addImm(24);
   BuildMI(MBB, MBBI, dl, TII.get(VE::LDrii), VE::SX10)
       .addReg(VE::SX11)
       .addImm(0)
       .addImm(8);
   BuildMI(MBB, MBBI, dl, TII.get(VE::LDrii), VE::SX9)
       .addReg(VE::SX11)
       .addImm(0)
       .addImm(0);
 }
 
 void VEFrameLowering::emitSPAdjustment(MachineFunction &MF,
                                        MachineBasicBlock &MBB,
                                        MachineBasicBlock::iterator MBBI,
-                                       int NumBytes) const {
+                                       int64_t NumBytes,
+                                       MaybeAlign MaybeAlign) const {
   DebugLoc dl;
   const VEInstrInfo &TII =
       *static_cast<const VEInstrInfo *>(MF.getSubtarget().getInstrInfo());
 
   if (NumBytes >= -64 && NumBytes < 63) {
     BuildMI(MBB, MBBI, dl, TII.get(VE::ADDSLri), VE::SX11)
         .addReg(VE::SX11)
         .addImm(NumBytes);
BuildMI(MBB, MBBI, dl, TII.get(VE::LEAzii), VE::SX13)
       .addImm(Lo_32(NumBytes));
   BuildMI(MBB, MBBI, dl, TII.get(VE::ANDrm), VE::SX13)
       .addReg(VE::SX13)
       .addImm(M0(32));
   BuildMI(MBB, MBBI, dl, TII.get(VE::LEASLrri), VE::SX11)
       .addReg(VE::SX11)
       .addReg(VE::SX13)
       .addImm(Hi_32(NumBytes));
+
+  if (MaybeAlign) {
+    // and %sp, %sp, Align-1
+    BuildMI(MBB, MBBI, dl, TII.get(VE::ANDrm), VE::SX11)
+        .addReg(VE::SX11)
+        .addImm(M1(64 - Log2_64(MaybeAlign.valueOrOne().value())));
+  }
 }
 
 void VEFrameLowering::emitSPExtend(MachineFunction &MF, MachineBasicBlock &MBB,
-                                   MachineBasicBlock::iterator MBBI,
-                                   int NumBytes) const {
+                                   MachineBasicBlock::iterator MBBI) const {
   DebugLoc dl;
   const VEInstrInfo &TII =
       *static_cast<const VEInstrInfo *>(MF.getSubtarget().getInstrInfo());
 
   // Emit following codes.  It is not possible to insert multiple
   // BasicBlocks in PEI pass, so we emit two pseudo instructions here.
   //
   //   EXTEND_STACK                     // pseudo instrcution
void VEFrameLowering::emitSPExtend(MachineFunction &MF, MachineBasicBlock &MBB,
   BuildMI(MBB, MBBI, dl, TII.get(VE::EXTEND_STACK));
   BuildMI(MBB, MBBI, dl, TII.get(VE::EXTEND_STACK_GUARD));
 }
 
 void VEFrameLowering::emitPrologue(MachineFunction &MF,
                                    MachineBasicBlock &MBB) const {
   assert(&MF.front() == &MBB && "Shrink-wrapping not yet supported");
   MachineFrameInfo &MFI = MF.getFrameInfo();
-  const VESubtarget &Subtarget = MF.getSubtarget<VESubtarget>();
-  const VEInstrInfo &TII =
-      *static_cast<const VEInstrInfo *>(Subtarget.getInstrInfo());
-  const VERegisterInfo &RegInfo =
-      *static_cast<const VERegisterInfo *>(Subtarget.getRegisterInfo());
+  const VEInstrInfo &TII = *STI.getInstrInfo();
+  const VERegisterInfo &RegInfo = *STI.getRegisterInfo();
   MachineBasicBlock::iterator MBBI = MBB.begin();
   // Debug location must be unknown since the first debug location is used
   // to determine the end of the prologue.
   DebugLoc dl;
   bool NeedsStackRealignment = RegInfo.needsStackRealignment(MF);
 
   // FIXME: unfortunately, returning false from canRealignStack
   // actually just causes needsStackRealignment to return false,
   // rather than reporting an error, as would be sensible. This is
   // poor, but fixing that bogosity is going to be a large project.
   // For now, just see if it's lied, and report an error here.
   if (!NeedsStackRealignment && MFI.getMaxAlign() > getStackAlign())
     report_fatal_error("Function \"" + Twine(MF.getName()) +
                        "\" required "
                        "stack re-alignment, but LLVM couldn't handle it "
                        "(probably because it has a dynamic alloca).");
 
   // Get the number of bytes to allocate from the FrameInfo
-  int NumBytes = (int)MFI.getStackSize();
-  // The VE ABI requires a reserved 176-byte area in the user's stack, starting
-  // at %sp + 16. This is for the callee Register Save Area (RSA).
-  //
-  // We therefore need to add that offset to the total stack size
-  // after all the stack objects are placed by
-  // PrologEpilogInserter calculateFrameObjectOffsets. However, since the stack
-  // needs to be aligned *after* the extra size is added, we need to disable
-  // calculateFrameObjectOffsets's built-in stack alignment, by having
-  // targetHandlesStackFrameRounding return true.
-
-  // Add the extra call frame stack size, if needed. (This is the same
-  // code as in PrologEpilogInserter, but also gets disabled by
-  // targetHandlesStackFrameRounding)
-  if (MFI.adjustsStack() && hasReservedCallFrame(MF))
-    NumBytes += MFI.getMaxCallFrameSize();
-
-  // Adds the VE subtarget-specific spill area to the stack
-  // size. Also ensures target-required alignment.
-  NumBytes = Subtarget.getAdjustedFrameSize(NumBytes);
+  uint64_t NumBytes = MFI.getStackSize();
+
+  // The VE ABI requires a reserved 176 bytes area at the top
+  // of stack as described in VESubtarget.cpp.  So, we adjust it here.
+  NumBytes = STI.getAdjustedFrameSize(NumBytes);
 
   // Finally, ensure that the size is sufficiently aligned for the
   // data on the stack.
-  NumBytes = alignTo(NumBytes, MFI.getMaxAlign().value());
+  NumBytes = alignTo(NumBytes, MFI.getMaxAlign());
 
   // Update stack size with corrected value.
   MFI.setStackSize(NumBytes);
 
   // Emit Prologue instructions to save %lr
   emitPrologueInsns(MF, MBB, MBBI, NumBytes, true);
 
   // Emit stack adjust instructions
-  emitSPAdjustment(MF, MBB, MBBI, -NumBytes);
+  MaybeAlign RuntimeAlign =
+      NeedsStackRealignment ? MaybeAlign(MFI.getMaxAlign()) : None;
+  emitSPAdjustment(MF, MBB, MBBI, -(int64_t)NumBytes, RuntimeAlign);
+
+  if (hasBP(MF)) {
+    // Copy SP to BP.
+    BuildMI(MBB, MBBI, dl, TII.get(VE::ORri), VE::SX17)
+        .addReg(VE::SX11)
+        .addImm(0);
+  }
 
   // Emit stack extend instructions
-  emitSPExtend(MF, MBB, MBBI, -NumBytes);
+  emitSPExtend(MF, MBB, MBBI);
 
-  unsigned regFP = RegInfo.getDwarfRegNum(VE::SX9, true);
+  Register RegFP = RegInfo.getDwarfRegNum(VE::SX9, true);
 
   // Emit ".cfi_def_cfa_register 30".
   unsigned CFIIndex =
-      MF.addFrameInst(MCCFIInstruction::createDefCfaRegister(nullptr, regFP));
+      MF.addFrameInst(MCCFIInstruction::createDefCfaRegister(nullptr, RegFP));
   BuildMI(MBB, MBBI, dl, TII.get(TargetOpcode::CFI_INSTRUCTION))
       .addCFIIndex(CFIIndex);
 
   // Emit ".cfi_window_save".
   CFIIndex = MF.addFrameInst(MCCFIInstruction::createWindowSave(nullptr));
   BuildMI(MBB, MBBI, dl, TII.get(TargetOpcode::CFI_INSTRUCTION))
       .addCFIIndex(CFIIndex);
 }
 
 MachineBasicBlock::iterator VEFrameLowering::eliminateCallFramePseudoInstr(
     MachineFunction &MF, MachineBasicBlock &MBB,
     MachineBasicBlock::iterator I) const {
   if (!hasReservedCallFrame(MF)) {
     MachineInstr &MI = *I;
-    int Size = MI.getOperand(0).getImm();
+    int64_t Size = MI.getOperand(0).getImm();
     if (MI.getOpcode() == VE::ADJCALLSTACKDOWN)
       Size = -Size;
 
     if (Size)
       emitSPAdjustment(MF, MBB, I, Size);
   }
   return MBB.erase(I);
 }
 
 void VEFrameLowering::emitEpilogue(MachineFunction &MF,
                                    MachineBasicBlock &MBB) const {
   MachineBasicBlock::iterator MBBI = MBB.getLastNonDebugInstr();
   DebugLoc dl = MBBI->getDebugLoc();
   MachineFrameInfo &MFI = MF.getFrameInfo();
 
-  int NumBytes = (int)MFI.getStackSize();
+  uint64_t NumBytes = MFI.getStackSize();
 
   // Emit Epilogue instructions to restore %lr
   emitEpilogueInsns(MF, MBB, MBBI, NumBytes, true);
 }
 
-bool VEFrameLowering::hasReservedCallFrame(const MachineFunction &MF) const {
-  // Reserve call frame if there are no variable sized objects on the stack.
-  return !MF.getFrameInfo().hasVarSizedObjects();
-}
-
 // hasFP - Return true if the specified function should have a dedicated frame
-// pointer register.  This is true if the function has variable sized allocas or
-// if frame pointer elimination is disabled.
+// pointer register.  This is true if the function has variable sized allocas
+// or if frame pointer elimination is disabled.  For the case of VE, we don't
+// implement FP eliminator yet, but we returns false from this function to
+// not refer fp from generated code.
 bool VEFrameLowering::hasFP(const MachineFunction &MF) const {
   const TargetRegisterInfo *RegInfo = MF.getSubtarget().getRegisterInfo();
 
   const MachineFrameInfo &MFI = MF.getFrameInfo();
   return MF.getTarget().Options.DisableFramePointerElim(MF) ||
          RegInfo->needsStackRealignment(MF) || MFI.hasVarSizedObjects() ||
          MFI.isFrameAddressTaken();
 }
 
+bool VEFrameLowering::hasBP(const MachineFunction &MF) const {
+  const MachineFrameInfo &MFI = MF.getFrameInfo();
+  const TargetRegisterInfo *TRI = STI.getRegisterInfo();
+
+  return MFI.hasVarSizedObjects() && TRI->needsStackRealignment(MF);
+}
+
 int VEFrameLowering::getFrameIndexReference(const MachineFunction &MF, int FI,
                                             Register &FrameReg) const {
-  const VESubtarget &Subtarget = MF.getSubtarget<VESubtarget>();
   const MachineFrameInfo &MFI = MF.getFrameInfo();
-  const VERegisterInfo *RegInfo = Subtarget.getRegisterInfo();
+  const VERegisterInfo *RegInfo = STI.getRegisterInfo();
   const VEMachineFunctionInfo *FuncInfo = MF.getInfo<VEMachineFunctionInfo>();
   bool isFixed = MFI.isFixedObjectIndex(FI);
 
-  // Addressable stack objects are accessed using neg. offsets from
-  // %fp, or positive offsets from %sp.
-  bool UseFP = true;
-
-  // VE uses FP-based references in general, even when "hasFP" is
-  // false. That function is rather a misnomer, because %fp is
-  // actually always available, unless isLeafProc.
+  int64_t FrameOffset = MF.getFrameInfo().getObjectOffset(FI);
+
   if (FuncInfo->isLeafProc()) {
     // If there's a leaf proc, all offsets need to be %sp-based,
     // because we haven't caused %fp to actually point to our frame.
-    UseFP = false;
-  } else if (isFixed) {
-    // Otherwise, argument access should always use %fp.
-    UseFP = true;
-  } else if (RegInfo->needsStackRealignment(MF)) {
+    FrameReg = VE::SX11; // %sp
+    return FrameOffset + MF.getFrameInfo().getStackSize();
+  } else if (RegInfo->needsStackRealignment(MF) && !isFixed) {
     // If there is dynamic stack realignment, all local object
-    // references need to be via %sp, to take account of the
-    // re-alignment.
-    UseFP = false;
-  }
-
-  int64_t FrameOffset = MF.getFrameInfo().getObjectOffset(FI);
-
-  if (UseFP) {
+    // references need to be via %sp or %s17 (bp), to take account
+    // of the re-alignment.
+    if (hasBP(MF))
+      FrameReg = VE::SX17; // %bp
+    else
+      FrameReg = VE::SX11; // %sp
+    return FrameOffset + MF.getFrameInfo().getStackSize();
+  } else {
+    // Finally, default to using %fp.
     FrameReg = RegInfo->getFrameRegister(MF);
     return FrameOffset;
   }
+}
 
-  FrameReg = VE::SX11; // %sp
-  return FrameOffset + MF.getFrameInfo().getStackSize();
+static bool LLVM_ATTRIBUTE_UNUSED
+verifyLeafProcRegUse(MachineRegisterInfo *MRI) {
+
+  // If any of parameter registers are used, this is not leaf function.
+  for (unsigned reg = VE::SX0; reg <= VE::SX7; ++reg)
+    if (MRI->isPhysRegUsed(reg))
+      return false;
+
+  // If any of callee-saved registers are used, this is not leaf function.
+  for (unsigned reg = VE::SX18; reg <= VE::SX33; ++reg)
+    if (MRI->isPhysRegUsed(reg))
+      return false;
+
+  return true;
 }
 
 bool VEFrameLowering::isLeafProc(MachineFunction &MF) const {
 
   MachineRegisterInfo &MRI = MF.getRegInfo();
   MachineFrameInfo &MFI = MF.getFrameInfo();
 
   return !MFI.hasCalls()                 // No calls

llvm/lib/Target/VE/VEISelLowering.h

 
 namespace VEISD {
 enum NodeType : unsigned {
   FIRST_NUMBER = ISD::BUILTIN_OP_END,
 
   Hi,
   Lo, // Hi/Lo operations, typically on a global address.
 
-  GETFUNPLT,  // load function address through %plt insturction
-  GETTLSADDR, // load address for TLS access
+  GETFUNPLT,   // load function address through %plt insturction
+  GETTLSADDR,  // load address for TLS access
+  GETSTACKTOP, // retrieve address of stack top (first address of
+               // locals and temporaries)
 
   CALL,            // A call instruction.
   RET_FLAG,        // Return with a flag operand.
   GLOBAL_BASE_REG, // Global base reg for PIC.
 };
 }
 
 class VETargetLowering : public TargetLowering {
public:
   SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const override;
 
   SDValue LowerVASTART(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerVAARG(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerBlockAddress(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerGlobalAddress(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerToTLSGeneralDynamicModel(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerDYNAMIC_STACKALLOC(SDValue Op, SelectionDAG &DAG) const;
   /// } Custom Lower
 
   SDValue withTargetFlags(SDValue Op, unsigned TF, SelectionDAG &DAG) const;
   SDValue makeHiLoPair(SDValue Op, unsigned HiTF, unsigned LoTF,
                        SelectionDAG &DAG) const;
   SDValue makeAddress(SDValue Op, SelectionDAG &DAG) const;
 
   bool isFPImmLegal(const APFloat &Imm, EVT VT,

llvm/lib/Target/VE/VEISelLowering.cpp

VETargetLowering::VETargetLowering(const TargetMachine &TM,
   setOperationAction(ISD::VASTART, MVT::Other, Custom);
   // VAARG needs to be lowered to access with 8 bytes alignment.
   setOperationAction(ISD::VAARG, MVT::Other, Custom);
   // Use the default implementation.
   setOperationAction(ISD::VACOPY, MVT::Other, Expand);
   setOperationAction(ISD::VAEND, MVT::Other, Expand);
   /// } VAARG handling
 
+  /// Stack {
+  setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Custom);
+  setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i64, Custom);
+  /// } Stack
+
   /// Int Ops {
   for (MVT IntVT : {MVT::i32, MVT::i64}) {
     // VE has no REM or DIVREM operations.
     setOperationAction(ISD::UREM, IntVT, Expand);
     setOperationAction(ISD::SREM, IntVT, Expand);
     setOperationAction(ISD::SDIVREM, IntVT, Expand);
     setOperationAction(ISD::UDIVREM, IntVT, Expand);
 
#define TARGET_NODE_CASE(NAME)                                                 \
   case VEISD::NAME:                                                            \
     return "VEISD::" #NAME;
   switch ((VEISD::NodeType)Opcode) {
   case VEISD::FIRST_NUMBER:
     break;
     TARGET_NODE_CASE(Lo)
     TARGET_NODE_CASE(Hi)
     TARGET_NODE_CASE(GETFUNPLT)
+    TARGET_NODE_CASE(GETSTACKTOP)
     TARGET_NODE_CASE(GETTLSADDR)
     TARGET_NODE_CASE(CALL)
     TARGET_NODE_CASE(RET_FLAG)
     TARGET_NODE_CASE(GLOBAL_BASE_REG)
   }
 #undef TARGET_NODE_CASE
   return nullptr;
 }
SDValue VETargetLowering::LowerVAARG(SDValue Op, SelectionDAG &DAG) const {
   InChain = DAG.getStore(Chain, DL, NextPtr, VAListPtr, MachinePointerInfo(SV));
 
   // Load the actual argument out of the pointer VAList.
   // We can't count on greater alignment than the word size.
   return DAG.getLoad(VT, DL, InChain, VAList, MachinePointerInfo(),
                      std::min(PtrVT.getSizeInBits(), VT.getSizeInBits()) / 8);
 }
 
+SDValue VETargetLowering::LowerDYNAMIC_STACKALLOC(SDValue Op,
+                                                  SelectionDAG &DAG) const {
+  // Generate following code.
+  //   (void)__llvm_grow_stack(size);
+  //   ret = GETSTACKTOP;        // pseudo instruction
+  SDLoc dl(Op);
+
+  SDValue Size = Op.getOperand(1); // Legalize the size.
+  EVT VT = Size->getValueType(0);
+
+  // Prepare arguments
+  TargetLowering::ArgListTy Args;
+  TargetLowering::ArgListEntry Entry;
+  Entry.Node = Size;
+  Entry.Ty = Entry.Node.getValueType().getTypeForEVT(*DAG.getContext());
+  Args.push_back(Entry);
+  Type *RetTy = Type::getVoidTy(*DAG.getContext());
+
+  EVT PtrVT = getPointerTy(DAG.getDataLayout());
+  SDValue Callee = DAG.getTargetExternalSymbol("__llvm_grow_stack", PtrVT, 0);
+
+  TargetLowering::CallLoweringInfo CLI(DAG);
+  CLI.setDebugLoc(dl)
+      .setChain(DAG.getEntryNode())
+      .setCallee(CallingConv::VE_LLVM_GROW_STACK, RetTy, Callee,
+                 std::move(Args))
+      .setDiscardResult(true);
+  std::pair<SDValue, SDValue> pair = LowerCallTo(CLI);
+  SDValue Chain = pair.second;
+  SDValue Value = DAG.getNode(VEISD::GETSTACKTOP, dl, VT, Chain);
+  SDValue Ops[2] = {Value, Chain};
+  return DAG.getMergeValues(Ops, dl);
+}
+
 SDValue VETargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
   switch (Op.getOpcode()) {
   default:
     llvm_unreachable("Should not custom lower this!");
   case ISD::BlockAddress:
     return LowerBlockAddress(Op, DAG);
+  case ISD::DYNAMIC_STACKALLOC:
+    return LowerDYNAMIC_STACKALLOC(Op, DAG);
   case ISD::GlobalAddress:
     return LowerGlobalAddress(Op, DAG);
   case ISD::GlobalTLSAddress:
     return LowerGlobalTLSAddress(Op, DAG);
   case ISD::VASTART:
     return LowerVASTART(Op, DAG);
   case ISD::VAARG:
     return LowerVAARG(Op, DAG);
   }
 }
 /// } Custom Lower

llvm/lib/Target/VE/VEInstrInfo.h

public:
   /// } Stack Spill & Reload
 
   Register getGlobalBaseReg(MachineFunction *MF) const;
 
   // Lower pseudo instructions after register allocation.
   bool expandPostRAPseudo(MachineInstr &MI) const override;
 
   bool expandExtendStackPseudo(MachineInstr &MI) const;
+  bool expandGetStackTopPseudo(MachineInstr &MI) const;
 };
 
 } // namespace llvm
 
 #endif

llvm/lib/Target/VE/VEInstrInfo.cpp

 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineMemOperand.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/TargetRegistry.h"
 
-#define DEBUG_TYPE "ve"
+#define DEBUG_TYPE "ve-instr-info"
 
 using namespace llvm;
 
 #define GET_INSTRINFO_CTOR_DTOR
 #include "VEGenInstrInfo.inc"
 
 // Pin the vtable to this file.
 void VEInstrInfo::anchor() {}
bool VEInstrInfo::expandPostRAPseudo(MachineInstr &MI) const {
   switch (MI.getOpcode()) {
   case VE::EXTEND_STACK: {
     return expandExtendStackPseudo(MI);
   }
   case VE::EXTEND_STACK_GUARD: {
     MI.eraseFromParent(); // The pseudo instruction is gone now.
     return true;
   }
+  case VE::GETSTACKTOP: {
+    return expandGetStackTopPseudo(MI);
+  }
   }
   return false;
 }
 
 bool VEInstrInfo::expandExtendStackPseudo(MachineInstr &MI) const {
   MachineBasicBlock &MBB = *MI.getParent();
   MachineFunction &MF = *MBB.getParent();
-  const VEInstrInfo &TII =
-      *static_cast<const VEInstrInfo *>(MF.getSubtarget().getInstrInfo());
+  const VESubtarget &STI = MF.getSubtarget<VESubtarget>();
+  const VEInstrInfo &TII = *STI.getInstrInfo();
   DebugLoc dl = MBB.findDebugLoc(MI);
 
   // Create following instructions and multiple basic blocks.
   //
   // thisBB:
   //   brge.l.t %sp, %sl, sinkBB
   // syscallBB:
   //   ld      %s61, 0x18(, %tp)        // load param area
bool VEInstrInfo::expandExtendStackPseudo(MachineInstr &MI) const {
 
   BuildMI(BB, dl, TII.get(VE::ORri), VE::SX0)
       .addReg(VE::SX62)
       .addImm(0);
 
   MI.eraseFromParent(); // The pseudo instruction is gone now.
   return true;
 }
+
+bool VEInstrInfo::expandGetStackTopPseudo(MachineInstr &MI) const {
+  MachineBasicBlock *MBB = MI.getParent();
+  MachineFunction &MF = *MBB->getParent();
+  const VESubtarget &STI = MF.getSubtarget<VESubtarget>();
+  const VEInstrInfo &TII = *STI.getInstrInfo();
+  DebugLoc dl = MBB->findDebugLoc(MI);
+
+  // Create following instruction
+  //
+  //   dst = %sp + target specific frame + the size of parameter area
+
+  const MachineFrameInfo &MFI = MF.getFrameInfo();
+  const VEFrameLowering &TFL = *STI.getFrameLowering();
+
+  // The VE ABI requires a reserved 176 bytes area at the top
+  // of stack as described in VESubtarget.cpp.  So, we adjust it here.
+  unsigned NumBytes = STI.getAdjustedFrameSize(0);
+
+  // Also adds the size of parameter area.
+  if (MFI.adjustsStack() && TFL.hasReservedCallFrame(MF))
+    NumBytes += MFI.getMaxCallFrameSize();
+
+  BuildMI(*MBB, MI, dl, TII.get(VE::LEArii))
+      .addDef(MI.getOperand(0).getReg())
+      .addReg(VE::SX11)
+      .addImm(0)
+      .addImm(NumBytes);
+
+  MI.eraseFromParent(); // The pseudo instruction is gone now.
+  return true;
+}

llvm/lib/Target/VE/VEInstrInfo.td

 // GETFUNPLT for PIC
 def GetFunPLT : SDNode<"VEISD::GETFUNPLT", SDTIntUnaryOp>;
 
 // GETTLSADDR for TLS
 def GetTLSAddr : SDNode<"VEISD::GETTLSADDR", SDT_SPCall,
                         [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
                          SDNPVariadic]>;
 
+// GETSTACKTOP
+def GetStackTop : SDNode<"VEISD::GETSTACKTOP", SDTNone,
+                        [SDNPHasChain, SDNPSideEffect]>;
 
 
 //===----------------------------------------------------------------------===//
 // VE Flag Conditions
 //===----------------------------------------------------------------------===//
 
 // Note that these values must be kept in sync with the CCOp::CondCode enum
 // values.

 def EXTEND_STACK : Pseudo<(outs), (ins),
                           "# EXTEND STACK",
                           []>;
 let  hasSideEffects = 0 in
 def EXTEND_STACK_GUARD : Pseudo<(outs), (ins),
                                 "# EXTEND STACK GUARD",
                                 []>;
 
+// Dynamic stack allocation yields a __llvm_grow_stack for VE targets.
+// These calls are needed to probe the stack when allocating more over
+// %s8 (%sl - stack limit).
+
+let Uses = [SX11], hasSideEffects = 1 in
+def GETSTACKTOP : Pseudo<(outs I64:$dst), (ins),
+                         "# GET STACK TOP",
+                         [(set iPTR:$dst, (GetStackTop))]>;
 // SETCC pattern matches
 //
 //   CMP  %tmp, lhs, rhs     ; compare lhs and rhs
 //   or   %res, 0, (0)1      ; initialize by 0
 //   CMOV %res, (63)0, %tmp  ; set 1 if %tmp is true
 
 def : Pat<(i32 (setcc i64:$LHS, i64:$RHS, CCSIOp:$cond)),
           (EXTRACT_SUBREG

llvm/lib/Target/VE/VERegisterInfo.cpp

 
 const MCPhysReg *
 VERegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const {
   return CSR_SaveList;
 }
 
 const uint32_t *VERegisterInfo::getCallPreservedMask(const MachineFunction &MF,
                                                      CallingConv::ID CC) const {
+  switch (CC) {
+  case CallingConv::VE_LLVM_GROW_STACK:
+    return CSR_llvm_grow_stack_RegMask;
+  default:
     return CSR_RegMask;
   }
+}
 
 const uint32_t *VERegisterInfo::getNoPreservedMask() const {
   return CSR_NoRegs_RegMask;
 }
 
 BitVector VERegisterInfo::getReservedRegs(const MachineFunction &MF) const {
   BitVector Reserved(getNumRegs());
 

llvm/lib/Target/VE/VESubtarget.h

class VESubtarget : public VEGenSubtargetInfo {
   SelectionDAGTargetInfo TSInfo;
   VEFrameLowering FrameLowering;
 
 public:
   VESubtarget(const Triple &TT, const std::string &CPU, const std::string &FS,
               const TargetMachine &TM);
 
   const VEInstrInfo *getInstrInfo() const override { return &InstrInfo; }
-  const TargetFrameLowering *getFrameLowering() const override {
+  const VEFrameLowering *getFrameLowering() const override {
     return &FrameLowering;
   }
   const VERegisterInfo *getRegisterInfo() const override {
     return &InstrInfo.getRegisterInfo();
   }
   const VETargetLowering *getTargetLowering() const override { return &TLInfo; }
   const SelectionDAGTargetInfo *getSelectionDAGInfo() const override {
     return &TSInfo;

llvm/test/CodeGen/VE/alloca.ll

+; RUN: llc < %s -mtriple=ve-unknown-unknown | FileCheck %s
+
+declare void @bar(i8*, i64)
+
+; Function Attrs: nounwind
+define void @test(i64 %n) {
+; CHECK-LABEL: test:
+; CHECK:       .LBB{{[0-9]+}}_2:
+; CHECK-NEXT:    or %s1, 0, %s0
+; CHECK-NEXT:    lea %s0, 15(, %s0)
+; CHECK-NEXT:    and %s0, -16, %s0
+; CHECK-NEXT:    lea %s2, __llvm_grow_stack@lo
+; CHECK-NEXT:    and %s2, %s2, (32)0
+; CHECK-NEXT:    lea.sl %s12, __llvm_grow_stack@hi(, %s2)
+; CHECK-NEXT:    bsic %s10, (, %s12)
+; CHECK-NEXT:    lea %s0, 240(, %s11)
+; CHECK-NEXT:    lea %s2, bar@lo
+; CHECK-NEXT:    and %s2, %s2, (32)0
+; CHECK-NEXT:    lea.sl %s12, bar@hi(, %s2)
+; CHECK-NEXT:    bsic %s10, (, %s12)
+; CHECK-NEXT:    or %s11, 0, %s9
+  %dyna = alloca i8, i64 %n, align 8
+  call void @bar(i8* %dyna, i64 %n)
+  ret void
+}