Support 32-bit memories with 65536 pages

lib/api/src/sys/externals/memory.rs

@@ -131,7 +131,7 @@ impl Memory {
     pub fn data_size(&self) -> u64 {
         let definition = self.vm_memory.from.vmmemory();
         let def = unsafe { definition.as_ref() };
-        def.current_length.into()
+        def.current_length.try_into().unwrap()
     }
 
     /// Returns the size (in [`Pages`]) of the `Memory`.

lib/compiler-llvm/src/trampoline/wasm.rs

@@ -48,9 +48,10 @@ impl FuncTrampoline {
         let module = self.ctx.create_module("");
         let target_machine = &self.target_machine;
         let target_triple = target_machine.get_triple();
+        let target_data = target_machine.get_target_data();
         module.set_triple(&target_triple);
-        module.set_data_layout(&target_machine.get_target_data().get_data_layout());
-        let intrinsics = Intrinsics::declare(&module, &self.ctx);
+        module.set_data_layout(&target_data.get_data_layout());
+        let intrinsics = Intrinsics::declare(&module, &self.ctx, &target_data);
 
         let (callee_ty, callee_attrs) =
             self.abi
@@ -177,10 +178,11 @@ impl FuncTrampoline {
         let function = CompiledKind::DynamicFunctionTrampoline(ty.clone());
         let module = self.ctx.create_module("");
         let target_machine = &self.target_machine;
+        let target_data = target_machine.get_target_data();
         let target_triple = target_machine.get_triple();
         module.set_triple(&target_triple);
-        module.set_data_layout(&target_machine.get_target_data().get_data_layout());
-        let intrinsics = Intrinsics::declare(&module, &self.ctx);
+        module.set_data_layout(&target_data.get_data_layout());
+        let intrinsics = Intrinsics::declare(&module, &self.ctx, &target_data);
 
         let (trampoline_ty, trampoline_attrs) =
             self.abi

lib/compiler-llvm/src/translator/code.rs

@@ -83,16 +83,17 @@ impl FuncTranslator {
 
         let target_machine = &self.target_machine;
         let target_triple = target_machine.get_triple();
+        let target_data = target_machine.get_target_data();
         module.set_triple(&target_triple);
-        module.set_data_layout(&target_machine.get_target_data().get_data_layout());
+        module.set_data_layout(&target_data.get_data_layout());
         let wasm_fn_type = wasm_module
             .signatures
             .get(wasm_module.functions[func_index])
             .unwrap();
 
         // TODO: pointer width
         let offsets = VMOffsets::new(8, &wasm_module);
-        let intrinsics = Intrinsics::declare(&module, &self.ctx);
+        let intrinsics = Intrinsics::declare(&module, &self.ctx, &target_data);
         let (func_type, func_attrs) =
             self.abi
                 .func_type_to_llvm(&self.ctx, &intrinsics, Some(&offsets), wasm_fn_type)?;

lib/compiler-llvm/src/translator/intrinsics.rs

@@ -11,6 +11,7 @@ use inkwell::{
     builder::Builder,
     context::Context,
     module::{Linkage, Module},
+    targets::TargetData,
     types::{
         BasicMetadataTypeEnum, BasicType, BasicTypeEnum, FloatType, IntType, PointerType,
         StructType, VectorType, VoidType,
@@ -168,6 +169,7 @@ pub struct Intrinsics<'ctx> {
     pub i32_ty: IntType<'ctx>,
     pub i64_ty: IntType<'ctx>,
     pub i128_ty: IntType<'ctx>,
+    pub isize_ty: IntType<'ctx>,
     pub f32_ty: FloatType<'ctx>,
     pub f64_ty: FloatType<'ctx>,
 
@@ -185,6 +187,7 @@ pub struct Intrinsics<'ctx> {
     pub i32_ptr_ty: PointerType<'ctx>,
     pub i64_ptr_ty: PointerType<'ctx>,
     pub i128_ptr_ty: PointerType<'ctx>,
+    pub isize_ptr_ty: PointerType<'ctx>,
     pub f32_ptr_ty: PointerType<'ctx>,
     pub f64_ptr_ty: PointerType<'ctx>,
 
@@ -199,6 +202,7 @@ pub struct Intrinsics<'ctx> {
     pub i32_zero: IntValue<'ctx>,
     pub i64_zero: IntValue<'ctx>,
     pub i128_zero: IntValue<'ctx>,
+    pub isize_zero: IntValue<'ctx>,
     pub f32_zero: FloatValue<'ctx>,
     pub f64_zero: FloatValue<'ctx>,
     pub f32x4_zero: VectorValue<'ctx>,
@@ -260,7 +264,11 @@ pub struct Intrinsics<'ctx> {
 
 impl<'ctx> Intrinsics<'ctx> {
     /// Create an [`Intrinsics`] for the given [`Context`].
-    pub fn declare(module: &Module<'ctx>, context: &'ctx Context) -> Self {
+    pub fn declare(
+        module: &Module<'ctx>,
+        context: &'ctx Context,
+        target_data: &TargetData,
+    ) -> Self {
         let void_ty = context.void_type();
         let i1_ty = context.bool_type();
         let i2_ty = context.custom_width_int_type(2);
@@ -270,6 +278,7 @@ impl<'ctx> Intrinsics<'ctx> {
         let i32_ty = context.i32_type();
         let i64_ty = context.i64_type();
         let i128_ty = context.i128_type();
+        let isize_ty = context.ptr_sized_int_type(target_data, None);
         let f32_ty = context.f32_type();
         let f64_ty = context.f64_type();
 
@@ -289,6 +298,7 @@ impl<'ctx> Intrinsics<'ctx> {
         let i32_ptr_ty = i32_ty.ptr_type(AddressSpace::Generic);
         let i64_ptr_ty = i64_ty.ptr_type(AddressSpace::Generic);
         let i128_ptr_ty = i128_ty.ptr_type(AddressSpace::Generic);
+        let isize_ptr_ty = isize_ty.ptr_type(AddressSpace::Generic);
         let f32_ptr_ty = f32_ty.ptr_type(AddressSpace::Generic);
         let f64_ptr_ty = f64_ty.ptr_type(AddressSpace::Generic);
 
@@ -297,6 +307,7 @@ impl<'ctx> Intrinsics<'ctx> {
         let i32_zero = i32_ty.const_int(0, false);
         let i64_zero = i64_ty.const_int(0, false);
         let i128_zero = i128_ty.const_int(0, false);
+        let isize_zero = isize_ty.const_int(0, false);
         let f32_zero = f32_ty.const_float(0.0);
         let f64_zero = f64_ty.const_float(0.0);
         let f32x4_zero = f32x4_ty.const_zero();
@@ -703,6 +714,7 @@ impl<'ctx> Intrinsics<'ctx> {
             i32_ty,
             i64_ty,
             i128_ty,
+            isize_ty,
             f32_ty,
             f64_ty,
 
@@ -720,6 +732,7 @@ impl<'ctx> Intrinsics<'ctx> {
             i32_ptr_ty,
             i64_ptr_ty,
             i128_ptr_ty,
+            isize_ptr_ty,
             f32_ptr_ty,
             f64_ptr_ty,
 
@@ -734,6 +747,7 @@ impl<'ctx> Intrinsics<'ctx> {
             i32_zero,
             i64_zero,
             i128_zero,
+            isize_zero,
             f32_zero,
             f64_zero,
             f32x4_zero,
@@ -1001,9 +1015,8 @@ impl<'ctx> Intrinsics<'ctx> {
             vmfunction_import_body_element: 0,
             vmfunction_import_vmctx_element: 1,
 
-            // TODO: this i64 is actually a rust usize
             vmmemory_definition_ptr_ty: context
-                .struct_type(&[i8_ptr_ty_basic, i32_ty.into()], false)
+                .struct_type(&[i8_ptr_ty_basic, isize_ty.into()], false)
                 .ptr_type(AddressSpace::Generic),
             vmmemory_definition_base_element: 0,
             vmmemory_definition_current_length_element: 1,

lib/compiler-singlepass/src/codegen_x64.rs

@@ -1290,7 +1290,7 @@ impl<'a> FuncGen<'a> {
         // Load bound into temporary register, if needed.
         if need_check {
             self.assembler
-                .emit_mov(Size::S32, bound_loc, Location::GPR(tmp_bound));
+                .emit_mov(Size::S64, bound_loc, Location::GPR(tmp_bound));
 
             // Wasm -> Effective.
             // Assuming we never underflow - should always be true on Linux/macOS and Windows >=8,

lib/vm/src/instance/mod.rs

@@ -813,9 +813,9 @@ impl Instance {
         if src
             .checked_add(len)
             .map_or(true, |n| n as usize > data.len())
-            || dst
-                .checked_add(len)
-                .map_or(true, |m| m > memory.current_length)
+            || dst.checked_add(len).map_or(true, |m| {
+                usize::try_from(m).unwrap() > memory.current_length
+            })
         {
             return Err(Trap::lib(TrapCode::HeapAccessOutOfBounds));
         }

lib/vm/src/vmcontext.rs

@@ -345,7 +345,7 @@ pub struct VMMemoryDefinition {
     pub base: *mut u8,
 
     /// The current logical size of this linear memory in bytes.
-    pub current_length: u32,
+    pub current_length: usize,
 }
 
 /// # Safety
@@ -362,7 +362,7 @@ unsafe impl Sync for VMMemoryDefinition {}
 impl MemoryUsage for VMMemoryDefinition {
     fn size_of_val(&self, tracker: &mut dyn MemoryUsageTracker) -> usize {
         if tracker.track(self.base as *const _ as *const ()) {
-            POINTER_BYTE_SIZE * (self.current_length as usize)
+            POINTER_BYTE_SIZE * self.current_length
         } else {
             0
         }
@@ -384,10 +384,10 @@ impl VMMemoryDefinition {
         // https://webassembly.github.io/reference-types/core/exec/instructions.html#exec-memory-copy
         if src
             .checked_add(len)
-            .map_or(true, |n| n > self.current_length)
+            .map_or(true, |n| usize::try_from(n).unwrap() > self.current_length)
             || dst
                 .checked_add(len)
-                .map_or(true, |m| m > self.current_length)
+                .map_or(true, |m| usize::try_from(m).unwrap() > self.current_length)
         {
             return Err(Trap::lib(TrapCode::HeapAccessOutOfBounds));
         }
@@ -417,7 +417,7 @@ impl VMMemoryDefinition {
     pub(crate) unsafe fn memory_fill(&self, dst: u32, val: u32, len: u32) -> Result<(), Trap> {
         if dst
             .checked_add(len)
-            .map_or(true, |m| m > self.current_length)
+            .map_or(true, |m| usize::try_from(m).unwrap() > self.current_length)
         {
             return Err(Trap::lib(TrapCode::HeapAccessOutOfBounds));
         }
@@ -458,12 +458,6 @@ mod test_vmmemory_definition {
             offset_of!(VMMemoryDefinition, current_length),
             usize::from(offsets.vmmemory_definition_current_length())
         );
-        /* TODO: Assert that the size of `current_length` matches.
-        assert_eq!(
-            size_of::<VMMemoryDefinition::current_length>(),
-            usize::from(offsets.size_of_vmmemory_definition_current_length())
-        );
-        */
     }
 }
 

tests/ignores.txt

@@ -69,6 +69,9 @@ cranelift spec::simd::simd_i64x2_extmul_i32x4
 cranelift spec::simd::simd_i8x16_arith2
 cranelift spec::simd::simd_int_to_int_extend
 
+# Windows doesn't overcommit and fails to allocate 4GB of memory
+windows wast::wasmer::max_size_of_memory
+
 # Frontends
 
 ## WASI

tests/wast/wasmer/max_size_of_memory.wast

@@ -0,0 +1 @@
+(module (memory 65536))