[WebAssembly] Added 64-bit memory.grow/size/init/copy/fill

[aardappel]

Summary

This covers both the existing memory functions as well as the new bulk memory proposal.
Added new test files since changes where also required in the inputs.

[aardappel]

@tlively I have no idea if my changes in IntrinsicsWebAssembly.td make sense, they merely pass the tests ;) please advise.

[tlively]

llvm/include/llvm/IR/IntrinsicsWebAssembly.td

209

I'm surprised this doesn't break the bulk-memory-intrinsics.ll test since this makes the intrinsic polymorphic. It would be good to add the 64-bit versions of these intrinsics to that test as well, and if the polymorphism starts causing problems, it would probably be a good idea to replace the second llvm_anyint_ty here with a LLVMMatchType<3> so you only have to specify a single type in the intrinsic name.

Alternatively, you could just delete these intrinsics and the corresponding clang builtins. They are extremely useless because there is no way to know what segment ID to pass before linking. I'm surprised we don't have intrinsics and builtins for memory.copy and memory.fill, which could actually be useful.

Another test you might want to update is the bulk-memory-encodings.s MC test.

[dschuff]

llvm/include/llvm/IR/IntrinsicsWebAssembly.td

209

do these segment IDs get rewritten by the linker, i.e. could they be symbolic? Are they only synthesized by the linker now or do we generate them in the backend? (I forget and I think we went back and forth on that). Would it make sense to have e.g. in the assembly language but not in C/LLVM IR?

[tlively]

llvm/include/llvm/IR/IntrinsicsWebAssembly.td

209

No, they cannot currently be symbolic. Right now they have to be constant integers. Even if they could be symbolic, we would have to change the linker to give symbols referenced that way their own segments, rather than merging them with all the other symbols into one big segment. If we made those changes, I could see these instructions being very useful in low-level assembly programs, though. The Binaryen segment optimizations are already robust enough that we wouldn't have to do any work in Binaryen for this feature.

[dschuff]

llvm/include/llvm/IR/IntrinsicsWebAssembly.td

209

ok thanks. seems worth keeping in mind, but obviously not a high priority unless we get a use case or request.

[aardappel]

llvm/include/llvm/IR/IntrinsicsWebAssembly.td

209

Your suggestion of LLVMMatchType<3> gets me error GA8ED74C0: Parameter #4 has out of bounds matching number 3, and looking at the other definitions, it seems these refer to the previous list, not the current one?

[aardappel]

llvm/include/llvm/IR/IntrinsicsWebAssembly.td

209

Ok, tried actually fixing these intrinsics for wasm64, but ran into too much trouble, so taking @tlively up on his offer to remove them instead.

[aardappel]

Removed unused init/drop intrinsics instead of trying to make them work for wasm64.
Also testing bulk-memory-encoding.s

[tlively]

llvm/test/MC/WebAssembly/bulk-memory-encodings.s

10

Getting rid of the intrinsics and clang builtins is fine, but I think we should keep the instruction definitions and these tests around. They would still be useful for disassembling modules that contain these instructions in the linker-synthesized memory initialization function.

[aardappel]

Reinstated init/drop defs + test

[tlively]

LGTM, thanks!

clang/include/clang/Basic/BuiltinsWebAssembly.def

 #endif
 
 // Query the current memory size, and increase the current memory size.
 // Note that memory.size is not "c" (readnone) because it must be sequenced
 // with respect to memory.grow calls.
 BUILTIN(__builtin_wasm_memory_size, "zIi", "n")
 BUILTIN(__builtin_wasm_memory_grow, "zIiz", "n")
 
-// Bulk memory builtins
-TARGET_BUILTIN(__builtin_wasm_memory_init, "vIUiIUiv*UiUi", "", "bulk-memory")
-TARGET_BUILTIN(__builtin_wasm_data_drop, "vIUi", "", "bulk-memory")
-
 // Thread-local storage
 TARGET_BUILTIN(__builtin_wasm_tls_size, "z", "nc", "bulk-memory")
 TARGET_BUILTIN(__builtin_wasm_tls_align, "z", "nc", "bulk-memory")
 TARGET_BUILTIN(__builtin_wasm_tls_base, "v*", "nU", "bulk-memory")
 
 // Floating point min/max
 BUILTIN(__builtin_wasm_min_f32, "fff", "nc")
 BUILTIN(__builtin_wasm_max_f32, "fff", "nc")

clang/lib/CodeGen/CGBuiltin.cpp

case WebAssembly::BI__builtin_wasm_memory_grow: {
     llvm::Type *ResultType = ConvertType(E->getType());
     Value *Args[] = {
       EmitScalarExpr(E->getArg(0)),
       EmitScalarExpr(E->getArg(1))
     };
     Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_memory_grow, ResultType);
     return Builder.CreateCall(Callee, Args);
   }
-  case WebAssembly::BI__builtin_wasm_memory_init: {
-    llvm::APSInt SegConst;
-    if (!E->getArg(0)->isIntegerConstantExpr(SegConst, getContext()))
-      llvm_unreachable("Constant arg isn't actually constant?");
-    llvm::APSInt MemConst;
-    if (!E->getArg(1)->isIntegerConstantExpr(MemConst, getContext()))
-      llvm_unreachable("Constant arg isn't actually constant?");
-    if (!MemConst.isNullValue())
-      ErrorUnsupported(E, "non-zero memory index");
-    Value *Args[] = {llvm::ConstantInt::get(getLLVMContext(), SegConst),
-                     llvm::ConstantInt::get(getLLVMContext(), MemConst),
-                     EmitScalarExpr(E->getArg(2)), EmitScalarExpr(E->getArg(3)),
-                     EmitScalarExpr(E->getArg(4))};
-    Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_memory_init);
-    return Builder.CreateCall(Callee, Args);
-  }
-  case WebAssembly::BI__builtin_wasm_data_drop: {
-    llvm::APSInt SegConst;
-    if (!E->getArg(0)->isIntegerConstantExpr(SegConst, getContext()))
-      llvm_unreachable("Constant arg isn't actually constant?");
-    Value *Arg = llvm::ConstantInt::get(getLLVMContext(), SegConst);
-    Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_data_drop);
-    return Builder.CreateCall(Callee, {Arg});
-  }
   case WebAssembly::BI__builtin_wasm_tls_size: {
     llvm::Type *ResultType = ConvertType(E->getType());
     Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_tls_size, ResultType);
     return Builder.CreateCall(Callee);
   }
   case WebAssembly::BI__builtin_wasm_tls_align: {
     llvm::Type *ResultType = ConvertType(E->getType());
     Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_tls_align, ResultType);

clang/test/CodeGen/builtins-wasm.c

 }
 
 __SIZE_TYPE__ memory_grow(__SIZE_TYPE__ delta) {
   return __builtin_wasm_memory_grow(0, delta);
   // WEBASSEMBLY32: call i32 @llvm.wasm.memory.grow.i32(i32 0, i32 %{{.*}})
   // WEBASSEMBLY64: call i64 @llvm.wasm.memory.grow.i64(i32 0, i64 %{{.*}})
 }
 
-void memory_init(void *dest, int offset, int size) {
-  __builtin_wasm_memory_init(3, 0, dest, offset, size);
-  // WEBASSEMBLY32: call void @llvm.wasm.memory.init(i32 3, i32 0, i8* %{{.*}}, i32 %{{.*}}, i32 %{{.*}})
-  // WEBASSEMBLY64: call void @llvm.wasm.memory.init(i32 3, i32 0, i8* %{{.*}}, i32 %{{.*}}, i32 %{{.*}})
-}
-
-void data_drop() {
-  __builtin_wasm_data_drop(3);
-  // WEBASSEMBLY32: call void @llvm.wasm.data.drop(i32 3)
-  // WEBASSEMBLY64: call void @llvm.wasm.data.drop(i32 3)
-}
-
 __SIZE_TYPE__ tls_size() {
   return __builtin_wasm_tls_size();
   // WEBASSEMBLY32: call i32 @llvm.wasm.tls.size.i32()
   // WEBASSEMBLY64: call i64 @llvm.wasm.tls.size.i64()
 }
 
 __SIZE_TYPE__ tls_align() {
   return __builtin_wasm_tls_align();

llvm/include/llvm/IR/IntrinsicsWebAssembly.td

def int_wasm_trunc :
   Intrinsic<[llvm_anyvector_ty],
             [LLVMMatchType<0>],
             [IntrNoMem, IntrSpeculatable]>;
 def int_wasm_nearest :
   Intrinsic<[llvm_anyvector_ty],
             [LLVMMatchType<0>],
             [IntrNoMem, IntrSpeculatable]>;
 
 //===----------------------------------------------------------------------===//
-// Bulk memory intrinsics
-//===----------------------------------------------------------------------===//
-
-def int_wasm_memory_init :
-  Intrinsic<[],
-            [llvm_i32_ty, llvm_i32_ty, llvm_ptr_ty, llvm_i32_ty, llvm_i32_ty],
-            [IntrWriteMem, IntrInaccessibleMemOrArgMemOnly, WriteOnly<ArgIndex<2>>,
-             IntrHasSideEffects, ImmArg<ArgIndex<0>>, ImmArg<ArgIndex<1>>]>;
-def int_wasm_data_drop :
-  Intrinsic<[],
-            [llvm_i32_ty],
-            [IntrNoDuplicate, IntrHasSideEffects, ImmArg<ArgIndex<0>>]>;
-
-//===----------------------------------------------------------------------===//
 // Thread-local storage intrinsics
 //===----------------------------------------------------------------------===//
 
 def int_wasm_tls_size :
   Intrinsic<[llvm_anyint_ty],
             [],
             [IntrNoMem, IntrSpeculatable]>;
 

llvm/lib/Target/WebAssembly/WebAssemblyInstrBulkMemory.td

def wasm_memcpy : SDNode<"WebAssemblyISD::MEMORY_COPY", wasm_memcpy_t,
                          [SDNPHasChain, SDNPMayLoad, SDNPMayStore]>;
 
 def wasm_memset_t : SDTypeProfile<0, 4,
   [SDTCisInt<0>, SDTCisPtrTy<1>, SDTCisInt<2>, SDTCisInt<3>]
 >;
 def wasm_memset : SDNode<"WebAssemblyISD::MEMORY_FILL", wasm_memset_t,
                          [SDNPHasChain, SDNPMayStore]>;
 
+multiclass BulkMemoryOps<WebAssemblyRegClass rc, string B> {
+
 let mayStore = 1, hasSideEffects = 1 in
-defm MEMORY_INIT :
+defm MEMORY_INIT_A#B :
   BULK_I<(outs),
-         (ins i32imm_op:$seg, i32imm_op:$idx, I32:$dest,
-              I32:$offset, I32:$size),
+         (ins i32imm_op:$seg, i32imm_op:$idx, rc:$dest,
+              rc:$offset, rc:$size),
          (outs), (ins i32imm_op:$seg, i32imm_op:$idx),
-         [(int_wasm_memory_init (i32 timm:$seg), (i32 timm:$idx), I32:$dest,
-            I32:$offset, I32:$size
-          )],
+         [],
          "memory.init\t$seg, $idx, $dest, $offset, $size",
          "memory.init\t$seg, $idx", 0x08>;
 
 let hasSideEffects = 1 in
 defm DATA_DROP :
   BULK_I<(outs), (ins i32imm_op:$seg), (outs), (ins i32imm_op:$seg),
-         [(int_wasm_data_drop (i32 timm:$seg))],
+         [],
          "data.drop\t$seg", "data.drop\t$seg", 0x09>;
 
 let mayLoad = 1, mayStore = 1 in
-defm MEMORY_COPY :
+defm MEMORY_COPY_A#B :
   BULK_I<(outs), (ins i32imm_op:$src_idx, i32imm_op:$dst_idx,
-                      I32:$dst, I32:$src, I32:$len),
+                      rc:$dst, rc:$src, rc:$len),
          (outs), (ins i32imm_op:$src_idx, i32imm_op:$dst_idx),
          [(wasm_memcpy (i32 imm:$src_idx), (i32 imm:$dst_idx),
-           I32:$dst, I32:$src, I32:$len
+           rc:$dst, rc:$src, rc:$len
          )],
          "memory.copy\t$src_idx, $dst_idx, $dst, $src, $len",
          "memory.copy\t$src_idx, $dst_idx", 0x0a>;
 
 let mayStore = 1 in
-defm MEMORY_FILL :
-  BULK_I<(outs), (ins i32imm_op:$idx, I32:$dst, I32:$value, I32:$size),
+defm MEMORY_FILL_A#B :
+  BULK_I<(outs), (ins i32imm_op:$idx, rc:$dst, I32:$value, rc:$size),
          (outs), (ins i32imm_op:$idx),
-         [(wasm_memset (i32 imm:$idx), I32:$dst, I32:$value, I32:$size)],
+         [(wasm_memset (i32 imm:$idx), rc:$dst, I32:$value, rc:$size)],
          "memory.fill\t$idx, $dst, $value, $size",
          "memory.fill\t$idx", 0x0b>;
+}
+
+defm : BulkMemoryOps<I32, "32">;
+defm : BulkMemoryOps<I64, "64">;

llvm/lib/Target/WebAssembly/WebAssemblyInstrMemory.td

 defm : StorePatOffsetOnly<i64, truncstorei16, "STORE16_I64">;
 defm : StorePatOffsetOnly<i64, truncstorei32, "STORE32_I64">;
 defm : StorePatGlobalAddrOffOnly<i32, truncstorei8, "STORE8_I32">;
 defm : StorePatGlobalAddrOffOnly<i32, truncstorei16, "STORE16_I32">;
 defm : StorePatGlobalAddrOffOnly<i64, truncstorei8, "STORE8_I64">;
 defm : StorePatGlobalAddrOffOnly<i64, truncstorei16, "STORE16_I64">;
 defm : StorePatGlobalAddrOffOnly<i64, truncstorei32, "STORE32_I64">;
 
+multiclass MemoryOps<WebAssemblyRegClass rc, string B> {
 // Current memory size.
-defm MEMORY_SIZE_I32 : I<(outs I32:$dst), (ins i32imm:$flags),
+defm MEMORY_SIZE_A#B : I<(outs rc:$dst), (ins i32imm:$flags),
                          (outs), (ins i32imm:$flags),
-                         [(set I32:$dst,
+                         [(set rc:$dst,
                            (int_wasm_memory_size (i32 imm:$flags)))],
                          "memory.size\t$dst, $flags", "memory.size\t$flags",
                          0x3f>;
 
 // Grow memory.
-defm MEMORY_GROW_I32 : I<(outs I32:$dst), (ins i32imm:$flags, I32:$delta),
+defm MEMORY_GROW_A#B : I<(outs rc:$dst), (ins i32imm:$flags, rc:$delta),
                          (outs), (ins i32imm:$flags),
-                         [(set I32:$dst,
+                         [(set rc:$dst,
                            (int_wasm_memory_grow (i32 imm:$flags),
-                             I32:$delta))],
+                             rc:$delta))],
                          "memory.grow\t$dst, $flags, $delta",
                          "memory.grow\t$flags", 0x40>;
+}
+
+defm : MemoryOps<I32, "32">;
+defm : MemoryOps<I64, "64">;

llvm/lib/Target/WebAssembly/WebAssemblySelectionDAGInfo.cpp

 #define DEBUG_TYPE "wasm-selectiondag-info"
 
 WebAssemblySelectionDAGInfo::~WebAssemblySelectionDAGInfo() = default; // anchor
 
 SDValue WebAssemblySelectionDAGInfo::EmitTargetCodeForMemcpy(
     SelectionDAG &DAG, const SDLoc &DL, SDValue Chain, SDValue Dst, SDValue Src,
     SDValue Size, Align Alignment, bool IsVolatile, bool AlwaysInline,
     MachinePointerInfo DstPtrInfo, MachinePointerInfo SrcPtrInfo) const {
-  if (!DAG.getMachineFunction()
-           .getSubtarget<WebAssemblySubtarget>()
-           .hasBulkMemory())
+  auto &ST = DAG.getMachineFunction().getSubtarget<WebAssemblySubtarget>();
+  if (!ST.hasBulkMemory())
     return SDValue();
 
   SDValue MemIdx = DAG.getConstant(0, DL, MVT::i32);
+  auto LenMVT = ST.hasAddr64() ? MVT::i64 : MVT::i32;
   return DAG.getNode(WebAssemblyISD::MEMORY_COPY, DL, MVT::Other,
                      {Chain, MemIdx, MemIdx, Dst, Src,
-                      DAG.getZExtOrTrunc(Size, DL, MVT::i32)});
+                      DAG.getZExtOrTrunc(Size, DL, LenMVT)});
 }
 
 SDValue WebAssemblySelectionDAGInfo::EmitTargetCodeForMemmove(
     SelectionDAG &DAG, const SDLoc &DL, SDValue Chain, SDValue Op1, SDValue Op2,
     SDValue Op3, Align Alignment, bool IsVolatile,
     MachinePointerInfo DstPtrInfo, MachinePointerInfo SrcPtrInfo) const {
   return EmitTargetCodeForMemcpy(DAG, DL, Chain, Op1, Op2, Op3,
                                  Alignment, IsVolatile, false,
                                  DstPtrInfo, SrcPtrInfo);
 }
 
 SDValue WebAssemblySelectionDAGInfo::EmitTargetCodeForMemset(
     SelectionDAG &DAG, const SDLoc &DL, SDValue Chain, SDValue Dst, SDValue Val,
     SDValue Size, Align Alignment, bool IsVolatile,
     MachinePointerInfo DstPtrInfo) const {
-  if (!DAG.getMachineFunction()
-           .getSubtarget<WebAssemblySubtarget>()
-           .hasBulkMemory())
+  auto &ST = DAG.getMachineFunction().getSubtarget<WebAssemblySubtarget>();
+  if (!ST.hasBulkMemory())
     return SDValue();
 
   SDValue MemIdx = DAG.getConstant(0, DL, MVT::i32);
+  auto LenMVT = ST.hasAddr64() ? MVT::i64 : MVT::i32;
   // Only low byte matters for val argument, so anyext the i8
   return DAG.getNode(WebAssemblyISD::MEMORY_FILL, DL, MVT::Other, Chain, MemIdx,
                      Dst, DAG.getAnyExtOrTrunc(Val, DL, MVT::i32),
-                     DAG.getZExtOrTrunc(Size, DL, MVT::i32));
+                     DAG.getZExtOrTrunc(Size, DL, LenMVT));
 }

llvm/test/CodeGen/WebAssembly/bulk-memory-intrinsics.ll

-; RUN: llc < %s -asm-verbose=false -verify-machineinstrs -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -wasm-keep-registers -mattr=+bulk-memory | FileCheck %s
-
-; Test that bulk memory intrinsics lower correctly
-
-target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
-target triple = "wasm32-unknown-unknown"
-
-; CHECK-LABEL: memory_init:
-; CHECK-NEXT: .functype  memory_init (i32, i32, i32) -> ()
-; CHECK-NEXT: memory.init 3, 0, $0, $1, $2
-; CHECK-NEXT: return
-declare void @llvm.wasm.memory.init(i32, i32, i8*, i32, i32)
-define void @memory_init(i8* %dest, i32 %offset, i32 %size) {
-  call void @llvm.wasm.memory.init(
-    i32 3, i32 0, i8* %dest, i32 %offset, i32 %size
-  )
-  ret void
-}
-
-; CHECK-LABEL: data_drop:
-; CHECK-NEXT: .functype data_drop () -> ()
-; CHECK-NEXT: data.drop 3
-; CHECK-NEXT: return
-declare void @llvm.wasm.data.drop(i32)
-define void @data_drop() {
-  call void @llvm.wasm.data.drop(i32 3)
-  ret void
-}

llvm/test/CodeGen/WebAssembly/bulk-memory64.ll

+; RUN: llc < %s -asm-verbose=false -verify-machineinstrs -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -wasm-keep-registers -mattr=+bulk-memory | FileCheck %s --check-prefixes CHECK,BULK-MEM
+; RUN: llc < %s -asm-verbose=false -verify-machineinstrs -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -wasm-keep-registers -mattr=-bulk-memory | FileCheck %s --check-prefixes CHECK,NO-BULK-MEM
+
+; Test that basic bulk memory codegen works correctly
+
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm64-unknown-unknown"
+
+declare void @llvm.memcpy.p0i8.p0i8.i8(i8*, i8*, i8, i1)
+declare void @llvm.memcpy.p0i8.p0i8.i64(i8*, i8*, i64, i1)
+declare void @llvm.memcpy.p0i32.p0i32.i64(i32*, i32*, i64, i1)
+
+declare void @llvm.memmove.p0i8.p0i8.i8(i8*, i8*, i8, i1)
+declare void @llvm.memmove.p0i8.p0i8.i64(i8*, i8*, i64, i1)
+declare void @llvm.memmove.p0i32.p0i32.i64(i32*, i32*, i64, i1)
+
+declare void @llvm.memset.p0i8.i8(i8*, i8, i8, i1)
+declare void @llvm.memset.p0i8.i64(i8*, i8, i64, i1)
+declare void @llvm.memset.p0i32.i64(i32*, i8, i64, i1)
+
+; CHECK-LABEL: memcpy_i8:
+; NO-BULK-MEM-NOT: memory.copy
+; BULK-MEM-NEXT: .functype memcpy_i8 (i64, i64, i32) -> ()
+; BULK-MEM-NEXT: i64.extend_i32_u $push0=, $2
+; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $pop0
+; BULK-MEM-NEXT: return
+define void @memcpy_i8(i8* %dest, i8* %src, i8 zeroext %len) {
+  call void @llvm.memcpy.p0i8.p0i8.i8(i8* %dest, i8* %src, i8 %len, i1 0)
+  ret void
+}
+
+; CHECK-LABEL: memmove_i8:
+; NO-BULK-MEM-NOT: memory.copy
+; BULK-MEM-NEXT: .functype memmove_i8 (i64, i64, i32) -> ()
+; BULK-MEM-NEXT: i64.extend_i32_u $push0=, $2
+; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $pop0
+; BULK-MEM-NEXT: return
+define void @memmove_i8(i8* %dest, i8* %src, i8 zeroext %len) {
+  call void @llvm.memmove.p0i8.p0i8.i8(i8* %dest, i8* %src, i8 %len, i1 0)
+  ret void
+}
+
+; CHECK-LABEL: memset_i8:
+; NO-BULK-MEM-NOT: memory.fill
+; BULK-MEM-NEXT: .functype memset_i8 (i64, i32, i32) -> ()
+; BULK-MEM-NEXT: i64.extend_i32_u $push0=, $2
+; BULK-MEM-NEXT: memory.fill 0, $0, $1, $pop0
+; BULK-MEM-NEXT: return
+define void @memset_i8(i8* %dest, i8 %val, i8 zeroext %len) {
+  call void @llvm.memset.p0i8.i8(i8* %dest, i8 %val, i8 %len, i1 0)
+  ret void
+}
+
+; CHECK-LABEL: memcpy_i32:
+; NO-BULK-MEM-NOT: memory.copy
+; BULK-MEM-NEXT: .functype memcpy_i32 (i64, i64, i64) -> ()
+; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $2
+; BULK-MEM-NEXT: return
+define void @memcpy_i32(i32* %dest, i32* %src, i64 %len) {
+  call void @llvm.memcpy.p0i32.p0i32.i64(i32* %dest, i32* %src, i64 %len, i1 0)
+  ret void
+}
+
+; CHECK-LABEL: memmove_i32:
+; NO-BULK-MEM-NOT: memory.copy
+; BULK-MEM-NEXT: .functype memmove_i32 (i64, i64, i64) -> ()
+; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $2
+; BULK-MEM-NEXT: return
+define void @memmove_i32(i32* %dest, i32* %src, i64 %len) {
+  call void @llvm.memmove.p0i32.p0i32.i64(i32* %dest, i32* %src, i64 %len, i1 0)
+  ret void
+}
+
+; CHECK-LABEL: memset_i32:
+; NO-BULK-MEM-NOT: memory.fill
+; BULK-MEM-NEXT: .functype memset_i32 (i64, i32, i64) -> ()
+; BULK-MEM-NEXT: memory.fill 0, $0, $1, $2
+; BULK-MEM-NEXT: return
+define void @memset_i32(i32* %dest, i8 %val, i64 %len) {
+  call void @llvm.memset.p0i32.i64(i32* %dest, i8 %val, i64 %len, i1 0)
+  ret void
+}
+
+; CHECK-LABEL: memcpy_1:
+; CHECK-NEXT: .functype memcpy_1 (i64, i64) -> ()
+; CHECK-NEXT: i32.load8_u $push[[L0:[0-9]+]]=, 0($1)
+; CHECK-NEXT: i32.store8 0($0), $pop[[L0]]
+; CHECK-NEXT: return
+define void @memcpy_1(i8* %dest, i8* %src) {
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %dest, i8* %src, i64 1, i1 0)
+  ret void
+}
+
+; CHECK-LABEL: memmove_1:
+; CHECK-NEXT: .functype memmove_1 (i64, i64) -> ()
+; CHECK-NEXT: i32.load8_u $push[[L0:[0-9]+]]=, 0($1)
+; CHECK-NEXT: i32.store8 0($0), $pop[[L0]]
+; CHECK-NEXT: return
+define void @memmove_1(i8* %dest, i8* %src) {
+  call void @llvm.memmove.p0i8.p0i8.i64(i8* %dest, i8* %src, i64 1, i1 0)
+  ret void
+}
+
+; CHECK-LABEL: memset_1:
+; NO-BULK-MEM-NOT: memory.fill
+; BULK-MEM-NEXT: .functype memset_1 (i64, i32) -> ()
+; BULK-MEM-NEXT: i32.store8 0($0), $1
+; BULK-MEM-NEXT: return
+define void @memset_1(i8* %dest, i8 %val) {
+  call void @llvm.memset.p0i8.i64(i8* %dest, i8 %val, i64 1, i1 0)
+  ret void
+}
+
+; CHECK-LABEL: memcpy_1024:
+; NO-BULK-MEM-NOT: memory.copy
+; BULK-MEM-NEXT: .functype memcpy_1024 (i64, i64) -> ()
+; BULK-MEM-NEXT: i64.const $push[[L0:[0-9]+]]=, 1024
+; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $pop[[L0]]
+; BULK-MEM-NEXT: return
+define void @memcpy_1024(i8* %dest, i8* %src) {
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %dest, i8* %src, i64 1024, i1 0)
+  ret void
+}
+
+; CHECK-LABEL: memmove_1024:
+; NO-BULK-MEM-NOT: memory.copy
+; BULK-MEM-NEXT: .functype memmove_1024 (i64, i64) -> ()
+; BULK-MEM-NEXT: i64.const $push[[L0:[0-9]+]]=, 1024
+; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $pop[[L0]]
+; BULK-MEM-NEXT: return
+define void @memmove_1024(i8* %dest, i8* %src) {
+  call void @llvm.memmove.p0i8.p0i8.i64(i8* %dest, i8* %src, i64 1024, i1 0)
+  ret void
+}
+
+; CHECK-LABEL: memset_1024:
+; NO-BULK-MEM-NOT: memory.fill
+; BULK-MEM-NEXT: .functype memset_1024 (i64, i32) -> ()
+; BULK-MEM-NEXT: i64.const $push[[L0:[0-9]+]]=, 1024
+; BULK-MEM-NEXT: memory.fill 0, $0, $1, $pop[[L0]]
+; BULK-MEM-NEXT: return
+define void @memset_1024(i8* %dest, i8 %val) {
+  call void @llvm.memset.p0i8.i64(i8* %dest, i8 %val, i64 1024, i1 0)
+  ret void
+}
+
+; The following tests check that frame index elimination works for
+; bulk memory instructions. The stack pointer is bumped by 112 instead
+; of 100 because the stack pointer in WebAssembly is currently always
+; 16-byte aligned, even in leaf functions, although it is not written
+; back to the global in this case.
+
+; TODO: Change TransientStackAlignment to 1 to avoid this extra
+; arithmetic. This will require forcing the use of StackAlignment in
+; PrologEpilogEmitter.cpp when
+; WebAssemblyFrameLowering::needsSPWriteback would be true.
+
+; CHECK-LABEL: memcpy_alloca_src:
+; NO-BULK-MEM-NOT: memory.copy
+; BULK-MEM-NEXT: .functype memcpy_alloca_src (i64) -> ()
+; BULK-MEM-NEXT: global.get $push[[L0:[0-9]+]]=, __stack_pointer
+; BULK-MEM-NEXT: i64.const $push[[L1:[0-9]+]]=, 112
+; BULK-MEM-NEXT: i64.sub $push[[L2:[0-9]+]]=, $pop[[L0]], $pop[[L1]]
+; BULK-MEM-NEXT: i64.const $push[[L3:[0-9]+]]=, 12
+; BULK-MEM-NEXT: i64.add $push[[L4:[0-9]+]]=, $pop[[L2]], $pop[[L3]]
+; BULK-MEM-NEXT: i64.const $push[[L5:[0-9]+]]=, 100
+; BULK-MEM-NEXT: memory.copy 0, 0, $0, $pop[[L4]], $pop[[L5]]
+; BULK-MEM-NEXT: return
+define void @memcpy_alloca_src(i8* %dst) {
+  %a = alloca [100 x i8]
+  %p = bitcast [100 x i8]* %a to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %dst, i8* %p, i64 100, i1 false)
+  ret void
+}
+
+; CHECK-LABEL: memcpy_alloca_dst:
+; NO-BULK-MEM-NOT: memory.copy
+; BULK-MEM-NEXT: .functype memcpy_alloca_dst (i64) -> ()
+; BULK-MEM-NEXT: global.get $push[[L0:[0-9]+]]=, __stack_pointer
+; BULK-MEM-NEXT: i64.const $push[[L1:[0-9]+]]=, 112
+; BULK-MEM-NEXT: i64.sub $push[[L2:[0-9]+]]=, $pop[[L0]], $pop[[L1]]
+; BULK-MEM-NEXT: i64.const $push[[L3:[0-9]+]]=, 12
+; BULK-MEM-NEXT: i64.add $push[[L4:[0-9]+]]=, $pop[[L2]], $pop[[L3]]
+; BULK-MEM-NEXT: i64.const $push[[L5:[0-9]+]]=, 100
+; BULK-MEM-NEXT: memory.copy 0, 0, $pop[[L4]], $0, $pop[[L5]]
+; BULK-MEM-NEXT: return
+define void @memcpy_alloca_dst(i8* %src) {
+  %a = alloca [100 x i8]
+  %p = bitcast [100 x i8]* %a to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %p, i8* %src, i64 100, i1 false)
+  ret void
+}
+
+; CHECK-LABEL: memset_alloca:
+; NO-BULK-MEM-NOT: memory.fill
+; BULK-MEM-NEXT: .functype memset_alloca (i32) -> ()
+; BULK-MEM-NEXT: global.get $push[[L0:[0-9]+]]=, __stack_pointer
+; BULK-MEM-NEXT: i64.const $push[[L1:[0-9]+]]=, 112
+; BULK-MEM-NEXT: i64.sub $push[[L2:[0-9]+]]=, $pop[[L0]], $pop[[L1]]
+; BULK-MEM-NEXT: i64.const $push[[L3:[0-9]+]]=, 12
+; BULK-MEM-NEXT: i64.add $push[[L4:[0-9]+]]=, $pop[[L2]], $pop[[L3]]
+; BULK-MEM-NEXT: i64.const $push[[L5:[0-9]+]]=, 100
+; BULK-MEM-NEXT: memory.fill 0, $pop[[L4]], $0, $pop[[L5]]
+; BULK-MEM-NEXT: return
+define void @memset_alloca(i8 %val) {
+  %a = alloca [100 x i8]
+  %p = bitcast [100 x i8]* %a to i8*
+  call void @llvm.memset.p0i8.i64(i8* %p, i8 %val, i64 100, i1 false)
+  ret void
+}

llvm/test/CodeGen/WebAssembly/memory-addr64.ll

+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -wasm-keep-registers | FileCheck %s
+
+; Test that basic memory operations assemble as expected with 64-bit addresses.
+
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm64-unknown-unknown"
+
+declare i64 @llvm.wasm.memory.size.i64(i32) nounwind readonly
+declare i64 @llvm.wasm.memory.grow.i64(i32, i64) nounwind
+
+; CHECK-LABEL: memory_size:
+; CHECK-NEXT: .functype memory_size () -> (i64){{$}}
+; CHECK-NEXT: memory.size $push0=, 0{{$}}
+; CHECK-NEXT: return $pop0{{$}}
+define i64 @memory_size() {
+  %a = call i64 @llvm.wasm.memory.size.i64(i32 0)
+  ret i64 %a
+}
+
+; CHECK-LABEL: memory_grow:
+; CHECK-NEXT: .functype memory_grow (i64) -> (i64){{$}}
+; CHECK: memory.grow $push0=, 0, $0{{$}}
+; CHECK-NEXT: return $pop0{{$}}
+define i64 @memory_grow(i64 %n) {
+  %a = call i64 @llvm.wasm.memory.grow.i64(i32 0, i64 %n)
+  ret i64 %a
+}

llvm/test/MC/WebAssembly/bulk-memory-encodings.s

 # RUN: llvm-mc -show-encoding -triple=wasm32-unknown-unknown -mattr=+bulk-memory < %s | FileCheck %s
+# RUN: llvm-mc -show-encoding -triple=wasm64-unknown-unknown -mattr=+bulk-memory < %s | FileCheck %s
 
 main:
     .functype main () -> ()
 
     # CHECK: memory.init 3, 0 # encoding: [0xfc,0x08,0x03,0x00]
     memory.init 3, 0
 
     # CHECK: data.drop 3 # encoding: [0xfc,0x09,0x03]
     data.drop 3
 
     # CHECK: memory.copy 0, 0 # encoding: [0xfc,0x0a,0x00,0x00]
     memory.copy 0, 0
 
     # CHECK: memory.fill 0 # encoding: [0xfc,0x0b,0x00]
     memory.fill 0
 
     end_function