[wasm] Change signature of memory_copy_wrapper

All other simple C functions take a pointer to a stack slot which
contains the actual parameters, whereas the memory_copy_wrapper takes
three parameters. This makes the code generation from Liftoff more
difficult. This CL changes the signature of memory_copy_wrapper to match
the signature of other simple C functions.

As MemoryCopy and MemoryInit are already implemented with C calls, this
change should not make a big difference in terms of performance. Simpler
and smaller Liftoff code may have more effect on performance. If this
assumption turns out wrong, we can change it in the future.

R=clemensb@chromium.org

Bug: v8:10281
Change-Id: I39e0ea00fcb22b4e84e612fe58eb4642856b72c9
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2078576
Commit-Queue: Andreas Haas <ahaas@chromium.org>
Reviewed-by: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/master@{#66607}

src/compiler/wasm-compiler.cc

@@ -4956,10 +4956,31 @@ Node* WasmGraphBuilder::MemoryInit(uint32_t data_segment_index, Node* dst,
 
   Node* function = graph()->NewNode(mcgraph()->common()->ExternalConstant(
       ExternalReference::wasm_memory_copy()));
-  MachineType sig_types[] = {MachineType::Pointer(), MachineType::Pointer(),
-                             MachineType::Uint32()};
-  MachineSignature sig(0, 3, sig_types);
-  return SetEffect(BuildCCall(&sig, function, dst, src, size));
+
+  int stack_slot_bytes = kSystemPointerSize * 2 + sizeof(uint32_t);
+  Node* stack_slot =
+      graph()->NewNode(mcgraph()->machine()->StackSlot(stack_slot_bytes));
+
+  SetEffect(graph()->NewNode(
+      mcgraph()->machine()->Store(StoreRepresentation(
+          MachineType::PointerRepresentation(), kNoWriteBarrier)),
+      stack_slot, mcgraph()->Int32Constant(0), dst, effect(), control()));
+
+  SetEffect(graph()->NewNode(
+      mcgraph()->machine()->Store(StoreRepresentation(
+          MachineType::PointerRepresentation(), kNoWriteBarrier)),
+      stack_slot, mcgraph()->Int32Constant(kSystemPointerSize), src, effect(),
+      control()));
+
+  SetEffect(graph()->NewNode(
+      mcgraph()->machine()->Store(
+          StoreRepresentation(MachineRepresentation::kWord32, kNoWriteBarrier)),
+      stack_slot, mcgraph()->Int32Constant(kSystemPointerSize * 2), size,
+      effect(), control()));
+
+  MachineType sig_types[] = {MachineType::Pointer()};
+  MachineSignature sig(0, 1, sig_types);
+  return SetEffect(BuildCCall(&sig, function, stack_slot));
 }
 
 Node* WasmGraphBuilder::DataDrop(uint32_t data_segment_index,
@@ -4986,10 +5007,31 @@ Node* WasmGraphBuilder::MemoryCopy(Node* dst, Node* src, Node* size,
 
   Node* function = graph()->NewNode(mcgraph()->common()->ExternalConstant(
       ExternalReference::wasm_memory_copy()));
-  MachineType sig_types[] = {MachineType::Pointer(), MachineType::Pointer(),
-                             MachineType::Uint32()};
-  MachineSignature sig(0, 3, sig_types);
-  return SetEffect(BuildCCall(&sig, function, dst, src, size));
+
+  int stack_slot_bytes = kSystemPointerSize * 2 + sizeof(uint32_t);
+  Node* stack_slot =
+      graph()->NewNode(mcgraph()->machine()->StackSlot(stack_slot_bytes));
+
+  SetEffect(graph()->NewNode(
+      mcgraph()->machine()->Store(StoreRepresentation(
+          MachineType::PointerRepresentation(), kNoWriteBarrier)),
+      stack_slot, mcgraph()->Int32Constant(0), dst, effect(), control()));
+
+  SetEffect(graph()->NewNode(
+      mcgraph()->machine()->Store(StoreRepresentation(
+          MachineType::PointerRepresentation(), kNoWriteBarrier)),
+      stack_slot, mcgraph()->Int32Constant(kSystemPointerSize), src, effect(),
+      control()));
+
+  SetEffect(graph()->NewNode(
+      mcgraph()->machine()->Store(
+          StoreRepresentation(MachineRepresentation::kWord32, kNoWriteBarrier)),
+      stack_slot, mcgraph()->Int32Constant(kSystemPointerSize * 2), size,
+      effect(), control()));
+
+  MachineType sig_types[] = {MachineType::Pointer()};
+  MachineSignature sig(0, 1, sig_types);
+  return SetEffect(BuildCCall(&sig, function, stack_slot));
 }
 
 Node* WasmGraphBuilder::MemoryFill(Node* dst, Node* value, Node* size,

src/wasm/module-instantiate.cc

@@ -673,7 +673,7 @@ void InstanceBuilder::LoadDataSegments(Handle<WasmInstanceObject> instance) {
           reinterpret_cast<Address>(instance->memory_start()) + dest_offset;
       Address src_addr = reinterpret_cast<Address>(wire_bytes.begin()) +
                          segment.source.offset();
-      memory_copy_wrapper(dest_addr, src_addr, size);
+      memory_copy(dest_addr, src_addr, size);
     } else {
       DCHECK(segment.active);
       // Segments of size == 0 are just nops.

src/wasm/wasm-external-refs.cc

@@ -331,7 +331,7 @@ void float64_pow_wrapper(Address data) {
 // reset the thread-in-wasm flag before calling this function. However,
 // as this is only a problem with Asan on Windows, we did not consider
 // it worth the overhead.
-DISABLE_ASAN void memory_copy_wrapper(Address dst, Address src, uint32_t size) {
+DISABLE_ASAN void memory_copy(Address dst, Address src, uint32_t size) {
   // Use explicit forward and backward copy to match the required semantics for
   // the memory.copy instruction. It is assumed that the caller of this
   // function has already performed bounds checks, so {src + size} and
@@ -352,6 +352,14 @@ DISABLE_ASAN void memory_copy_wrapper(Address dst, Address src, uint32_t size) {
   }
 }
 
+void memory_copy_wrapper(Address data) {
+  Address dst = ReadUnalignedValue<Address>(data);
+  Address src = ReadUnalignedValue<Address>(data + sizeof(dst));
+  uint32_t size =
+      ReadUnalignedValue<uint32_t>(data + sizeof(dst) + sizeof(src));
+  memory_copy(dst, src, size);
+}
+
 // Asan on Windows triggers exceptions in this function that confuse the
 // WebAssembly trap handler, so Asan is disabled. See the comment on
 // memory_copy_wrapper above for more info.

src/wasm/wasm-external-refs.h

@@ -71,7 +71,9 @@ V8_EXPORT_PRIVATE void word64_ror_wrapper(Address data);
 
 V8_EXPORT_PRIVATE void float64_pow_wrapper(Address data);
 
-void memory_copy_wrapper(Address dst, Address src, uint32_t size);
+void memory_copy_wrapper(Address data);
+
+void memory_copy(Address dst, Address src, uint32_t size);
 
 void memory_fill_wrapper(Address dst, uint32_t value, uint32_t size);
 

src/wasm/wasm-interpreter.cc

@@ -1824,7 +1824,7 @@ class ThreadImpl {
         Address src_addr =
             instance_object_->data_segment_starts()[imm.data_segment_index] +
             src;
-        memory_copy_wrapper(dst_addr, src_addr, size);
+        memory_copy(dst_addr, src_addr, size);
         return true;
       }
       case kExprDataDrop: {
@@ -1850,7 +1850,7 @@ class ThreadImpl {
           return false;
         }
 
-        memory_copy_wrapper(dst_addr, src_addr, size);
+        memory_copy(dst_addr, src_addr, size);
         return true;
       }
       case kExprMemoryFill: {