diff --git a/src/software/SoftBuffer.zig b/src/software/SoftBuffer.zig
index 34b1cf4..bc3ab54 100644
--- a/src/software/SoftBuffer.zig
+++ b/src/software/SoftBuffer.zig
@@ -80,35 +80,35 @@ pub inline fn mapTo(self: *const Self, comptime T: type) VkError!T {
     return self.mapToWithAddedOffset(T, 0);
 }
 
-pub inline fn mapAsSlice(self: *const Self, comptime T: type, size: usize) VkError![]T {
+pub inline fn mapAsSlice(self: *const Self, comptime T: type, size: vk.DeviceSize) VkError![]T {
     return self.mapAsSliceWithAddedOffset(T, 0, size);
 }
 
-pub inline fn mapAsWithAddedOffset(self: *const Self, comptime T: type, offset: usize) VkError!*T {
+pub inline fn mapAsWithAddedOffset(self: *const Self, comptime T: type, offset: vk.DeviceSize) VkError!*T {
     return self.mapAsWithOffset(T, self.interface.offset + offset);
 }
 
-pub inline fn mapToWithAddedOffset(self: *const Self, comptime T: type, offset: usize) VkError!T {
+pub inline fn mapToWithAddedOffset(self: *const Self, comptime T: type, offset: vk.DeviceSize) VkError!T {
     return self.mapToWithOffset(T, self.interface.offset + offset);
 }
 
-pub inline fn mapAsSliceWithAddedOffset(self: *const Self, comptime T: type, offset: usize, size: usize) VkError![]T {
+pub inline fn mapAsSliceWithAddedOffset(self: *const Self, comptime T: type, offset: vk.DeviceSize, size: vk.DeviceSize) VkError![]T {
     return self.mapAsSliceWithOffset(T, self.interface.offset + offset, size);
 }
 
-pub fn mapAsWithOffset(self: *const Self, comptime T: type, offset: usize) VkError!*T {
+pub fn mapAsWithOffset(self: *const Self, comptime T: type, offset: vk.DeviceSize) VkError!*T {
     const memory = if (self.interface.memory) |memory| memory else return VkError.InvalidDeviceMemoryDrv;
     const map = try memory.map(offset, @sizeOf(T));
     return @alignCast(std.mem.bytesAsValue(T, map));
 }
 
-pub fn mapToWithOffset(self: *const Self, comptime T: type, offset: usize) VkError!T {
+pub fn mapToWithOffset(self: *const Self, comptime T: type, offset: vk.DeviceSize) VkError!T {
     const memory = if (self.interface.memory) |memory| memory else return VkError.InvalidDeviceMemoryDrv;
     const map = try memory.map(offset, @sizeOf(T));
     return std.mem.bytesToValue(T, map);
 }
 
-pub fn mapAsSliceWithOffset(self: *const Self, comptime T: type, offset: usize, size: usize) VkError![]T {
+pub fn mapAsSliceWithOffset(self: *const Self, comptime T: type, offset: vk.DeviceSize, size: vk.DeviceSize) VkError![]T {
     const memory = if (self.interface.memory) |memory| memory else return VkError.InvalidDeviceMemoryDrv;
     const map = try memory.map(offset, size);
     return @alignCast(std.mem.bytesAsSlice(T, map));
diff --git a/src/software/SoftCommandBuffer.zig b/src/software/SoftCommandBuffer.zig
index ce979b7..dbf980e 100644
--- a/src/software/SoftCommandBuffer.zig
+++ b/src/software/SoftCommandBuffer.zig
@@ -128,6 +128,7 @@ pub fn create(device: *base.Device, allocator: std.mem.Allocator, info: *const v
 
     self.* = .{
         .interface = interface,
+        // SAFETY: the command allocator is initialized after the object is constructed
         .command_allocator = undefined,
         .commands = .empty,
     };
diff --git a/src/software/SoftDeviceMemory.zig b/src/software/SoftDeviceMemory.zig
index db000ce..217ddbf 100644
--- a/src/software/SoftDeviceMemory.zig
+++ b/src/software/SoftDeviceMemory.zig
@@ -25,9 +25,10 @@ pub fn create(device: *SoftDevice, allocator: std.mem.Allocator, size: vk.Device
         .invalidateRange = invalidateRange,
     };
 
+    const allocation_size = std.math.cast(usize, size) orelse return VkError.OutOfDeviceMemory;
     self.* = .{
         .interface = interface,
-        .data = device.interface.device_allocator.allocator().alloc(u8, size) catch return VkError.OutOfDeviceMemory,
+        .data = device.interface.device_allocator.allocator().alloc(u8, allocation_size) catch return VkError.OutOfDeviceMemory,
     };
     return self;
 }
@@ -54,10 +55,18 @@ pub fn invalidateRange(interface: *Interface, offset: vk.DeviceSize, size: vk.De
 
 pub fn map(interface: *Interface, offset: vk.DeviceSize, size: vk.DeviceSize) VkError![]u8 {
     const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
-    if (offset >= self.data.len or (size != vk.WHOLE_SIZE and offset + size > self.data.len)) {
+    const map_offset = std.math.cast(usize, offset) orelse return VkError.MemoryMapFailed;
+    if (map_offset >= self.data.len) {
         return VkError.MemoryMapFailed;
     }
-    return if (size == vk.WHOLE_SIZE) self.data[offset..] else self.data[offset..(offset + size)];
+    const map_size = if (size == vk.WHOLE_SIZE)
+        self.data.len - map_offset
+    else
+        std.math.cast(usize, size) orelse return VkError.MemoryMapFailed;
+    if (map_size > self.data.len - map_offset) {
+        return VkError.MemoryMapFailed;
+    }
+    return self.data[map_offset..][0..map_size];
 }
 
 pub fn unmap(_: *Interface) void {
diff --git a/src/software/SoftImage.zig b/src/software/SoftImage.zig
index 881926b..b5e2b24 100644
--- a/src/software/SoftImage.zig
+++ b/src/software/SoftImage.zig
@@ -569,35 +569,35 @@ pub inline fn mapTo(self: *const Self, comptime T: type) VkError!T {
     return self.mapToWithAddedOffset(T, 0);
 }
 
-pub inline fn mapAsSlice(self: *const Self, comptime T: type, size: usize) VkError![]T {
+pub inline fn mapAsSlice(self: *const Self, comptime T: type, size: vk.DeviceSize) VkError![]T {
     return self.mapAsSliceWithAddedOffset(T, 0, size);
 }
 
-pub inline fn mapAsWithAddedOffset(self: *const Self, comptime T: type, offset: usize) VkError!*T {
+pub inline fn mapAsWithAddedOffset(self: *const Self, comptime T: type, offset: vk.DeviceSize) VkError!*T {
     return self.mapAsWithOffset(T, self.interface.memory_offset + offset);
 }
 
-pub inline fn mapToWithAddedOffset(self: *const Self, comptime T: type, offset: usize) VkError!T {
+pub inline fn mapToWithAddedOffset(self: *const Self, comptime T: type, offset: vk.DeviceSize) VkError!T {
     return self.mapToWithOffset(T, self.interface.memory_offset + offset);
 }
 
-pub inline fn mapAsSliceWithAddedOffset(self: *const Self, comptime T: type, offset: usize, size: usize) VkError![]T {
+pub inline fn mapAsSliceWithAddedOffset(self: *const Self, comptime T: type, offset: vk.DeviceSize, size: vk.DeviceSize) VkError![]T {
     return self.mapAsSliceWithOffset(T, self.interface.memory_offset + offset, size);
 }
 
-pub fn mapAsWithOffset(self: *const Self, comptime T: type, offset: usize) VkError!*T {
+pub fn mapAsWithOffset(self: *const Self, comptime T: type, offset: vk.DeviceSize) VkError!*T {
     const memory = if (self.interface.memory) |memory| memory else return VkError.InvalidDeviceMemoryDrv;
     const map = try memory.map(offset, @sizeOf(T));
     return @alignCast(std.mem.bytesAsValue(T, map));
 }
 
-pub fn mapToWithOffset(self: *const Self, comptime T: type, offset: usize) VkError!T {
+pub fn mapToWithOffset(self: *const Self, comptime T: type, offset: vk.DeviceSize) VkError!T {
     const memory = if (self.interface.memory) |memory| memory else return VkError.InvalidDeviceMemoryDrv;
     const map = try memory.map(offset, @sizeOf(T));
     return std.mem.bytesToValue(T, map);
 }
 
-pub fn mapAsSliceWithOffset(self: *const Self, comptime T: type, offset: usize, size: usize) VkError![]T {
+pub fn mapAsSliceWithOffset(self: *const Self, comptime T: type, offset: vk.DeviceSize, size: vk.DeviceSize) VkError![]T {
     const memory = if (self.interface.memory) |memory| memory else return VkError.InvalidDeviceMemoryDrv;
     const map = try memory.map(offset, size);
     return @alignCast(std.mem.bytesAsSlice(T, map));
diff --git a/src/software/device/Renderer.zig b/src/software/device/Renderer.zig
index e2ecd5f..dcfd143 100644
--- a/src/software/device/Renderer.zig
+++ b/src/software/device/Renderer.zig
@@ -23,6 +23,7 @@ const F32x4 = zm.F32x4;
 const Self = @This();
 
 const @"1GiB" = 1_073_741_824;
+const draw_allocator_limit: usize = @intCast(@min(@as(u64, 4) * @"1GiB", @as(u64, std.math.maxInt(usize))));
 
 pub const VertexBuffer = struct {
     buffer: *const SoftBuffer,
@@ -196,12 +197,12 @@ pub fn resetInputAttachmentSnapshots(self: *Self) void {
 }
 
 pub fn draw(self: *Self, vertex_count: usize, instance_count: usize, first_vertex: usize, first_instance: usize) VkError!void {
-    var bounded_allocator: BoundedAllocator = .init(self.device.interface.device_allocator.allocator(), 4 * @"1GiB");
+    var bounded_allocator: BoundedAllocator = .init(self.device.interface.device_allocator.allocator(), draw_allocator_limit);
     try self.drawCall(&bounded_allocator, vertex_count, instance_count, first_vertex, first_instance, null, null);
 }
 
 pub fn drawIndexed(self: *Self, index_count: usize, instance_count: usize, first_index: usize, first_instance: usize, vertex_offset: i32) VkError!void {
-    var bounded_allocator: BoundedAllocator = .init(self.device.interface.device_allocator.allocator(), 4 * @"1GiB");
+    var bounded_allocator: BoundedAllocator = .init(self.device.interface.device_allocator.allocator(), draw_allocator_limit);
     const allocator = bounded_allocator.allocator();
 
     const indexed_draw = try self.readIndexBuffer(allocator, index_count, first_index, vertex_offset);
diff --git a/src/software/device/blitter.zig b/src/software/device/blitter.zig
index 33b6362..c9c7a08 100644
--- a/src/software/device/blitter.zig
+++ b/src/software/device/blitter.zig
@@ -794,7 +794,12 @@ pub fn readFloat4(map: []const u8, src_format: vk.Format) F32x4 {
             c[3] = normalizedI16(std.mem.bytesToValue(u16, map[6..]));
         },
 
-        .r16g16b16a16_sfloat => c = std.mem.bytesToValue(@Vector(4, f16), map),
+        .r16g16b16a16_sfloat => {
+            c[0] = std.mem.bytesToValue(f16, map[0..]);
+            c[1] = std.mem.bytesToValue(f16, map[2..]);
+            c[2] = std.mem.bytesToValue(f16, map[4..]);
+            c[3] = std.mem.bytesToValue(f16, map[6..]);
+        },
 
         .r32g32b32a32_sfloat => c = std.mem.bytesToValue(F32x4, map),
 
diff --git a/src/vulkan/NonDispatchable.zig b/src/vulkan/NonDispatchable.zig
index b993840..14f9205 100644
--- a/src/vulkan/NonDispatchable.zig
+++ b/src/vulkan/NonDispatchable.zig
@@ -44,7 +44,8 @@ pub fn NonDispatchable(comptime T: type) type {
             if (handle == 0) {
                 return VkError.InvalidHandleDrv;
             }
-            const non_dispatchable: *Self = @ptrFromInt(handle);
+            const address = std.math.cast(usize, handle) orelse return VkError.InvalidHandleDrv;
+            const non_dispatchable: *Self = @ptrFromInt(address);
             if (non_dispatchable.object_type != T.ObjectType) {
                 return VkError.InvalidHandleDrv;
             }
diff --git a/src/vulkan/QueryPool.zig b/src/vulkan/QueryPool.zig
index 189384c..3dbdf5e 100644
--- a/src/vulkan/QueryPool.zig
+++ b/src/vulkan/QueryPool.zig
@@ -14,8 +14,51 @@ queries: []Query,
 
 vtable: *const VTable,
 
+/// 32 bits platforms only supportss atomics up to 32 bits,
+/// so serialize 64-bit query access with a 32-bit lock.
+const LockedU64 = struct {
+    mutex: std.atomic.Value(u32) = std.atomic.Value(u32).init(0),
+    raw: u64,
+
+    fn init(value: u64) LockedU64 {
+        return .{ .raw = value };
+    }
+
+    fn lock(self: *LockedU64) void {
+        while (self.mutex.cmpxchgWeak(0, 1, .acquire, .monotonic) != null) {
+            std.atomic.spinLoopHint();
+        }
+    }
+
+    fn unlock(self: *LockedU64) void {
+        self.mutex.store(0, .release);
+    }
+
+    fn load(self: *LockedU64, comptime _: std.builtin.AtomicOrder) u64 {
+        self.lock();
+        defer self.unlock();
+        return self.raw;
+    }
+
+    fn store(self: *LockedU64, value: u64, comptime _: std.builtin.AtomicOrder) void {
+        self.lock();
+        defer self.unlock();
+        self.raw = value;
+    }
+
+    fn fetchAdd(self: *LockedU64, value: u64, comptime _: std.builtin.AtomicOrder) u64 {
+        self.lock();
+        defer self.unlock();
+        const previous = self.raw;
+        self.raw +%= value;
+        return previous;
+    }
+};
+
+const QueryValue = if (@bitSizeOf(usize) >= 64) std.atomic.Value(u64) else LockedU64;
+
 const Query = struct {
-    value: std.atomic.Value(u64) = std.atomic.Value(u64).init(0),
+    value: QueryValue = QueryValue.init(0),
     available: bool = false,
     active: bool = false,
 };
diff --git a/src/vulkan/lib_vulkan.zig b/src/vulkan/lib_vulkan.zig
index be96754..13386ed 100644
--- a/src/vulkan/lib_vulkan.zig
+++ b/src/vulkan/lib_vulkan.zig
@@ -1860,7 +1860,8 @@ pub export fn apeCmdBindIndexBuffer(p_cmd: vk.CommandBuffer, p_buffer: vk.Buffer
 
     const cmd = Dispatchable(CommandBuffer).fromHandleObject(p_cmd) catch |err| return errorLogger(err);
     const buffer = NonDispatchable(Buffer).fromHandleObject(p_buffer) catch |err| return errorLogger(err);
-    cmd.bindIndexBuffer(buffer, offset, index_type) catch |err| return errorLogger(err);
+    const host_offset = std.math.cast(usize, offset) orelse return errorLogger(VkError.ValidationFailed);
+    cmd.bindIndexBuffer(buffer, host_offset, index_type) catch |err| return errorLogger(err);
 }
 
 pub export fn apeCmdBindPipeline(p_cmd: vk.CommandBuffer, bind_point: vk.PipelineBindPoint, p_pipeline: vk.Pipeline) callconv(vk.vulkan_call_conv) void {
@@ -1879,7 +1880,8 @@ pub export fn apeCmdBindVertexBuffers(p_cmd: vk.CommandBuffer, first: u32, count
     const cmd = Dispatchable(CommandBuffer).fromHandleObject(p_cmd) catch |err| return errorLogger(err);
     for (p_buffers, offsets, 0..count) |p_buffer, offset, i| {
         const buffer = NonDispatchable(Buffer).fromHandleObject(p_buffer) catch |err| return errorLogger(err);
-        cmd.bindVertexBuffer(first + i, buffer, offset) catch |err| return errorLogger(err);
+        const host_offset = std.math.cast(usize, offset) orelse return errorLogger(VkError.ValidationFailed);
+        cmd.bindVertexBuffer(first + i, buffer, host_offset) catch |err| return errorLogger(err);
     }
 }
 
@@ -2035,7 +2037,8 @@ pub export fn apeCmdDrawIndexedIndirect(p_cmd: vk.CommandBuffer, p_buffer: vk.Bu
 
     const cmd = Dispatchable(CommandBuffer).fromHandleObject(p_cmd) catch |err| return errorLogger(err);
     const buffer = NonDispatchable(Buffer).fromHandleObject(p_buffer) catch |err| return errorLogger(err);
-    cmd.drawIndexedIndirect(buffer, offset, count, stride) catch |err| return errorLogger(err);
+    const host_offset = std.math.cast(usize, offset) orelse return errorLogger(VkError.ValidationFailed);
+    cmd.drawIndexedIndirect(buffer, host_offset, count, stride) catch |err| return errorLogger(err);
 }
 
 pub export fn apeCmdDrawIndirect(p_cmd: vk.CommandBuffer, p_buffer: vk.Buffer, offset: vk.DeviceSize, count: u32, stride: u32) callconv(vk.vulkan_call_conv) void {
@@ -2044,7 +2047,8 @@ pub export fn apeCmdDrawIndirect(p_cmd: vk.CommandBuffer, p_buffer: vk.Buffer, o
 
     const cmd = Dispatchable(CommandBuffer).fromHandleObject(p_cmd) catch |err| return errorLogger(err);
     const buffer = NonDispatchable(Buffer).fromHandleObject(p_buffer) catch |err| return errorLogger(err);
-    cmd.drawIndirect(buffer, offset, count, stride) catch |err| return errorLogger(err);
+    const host_offset = std.math.cast(usize, offset) orelse return errorLogger(VkError.ValidationFailed);
+    cmd.drawIndirect(buffer, host_offset, count, stride) catch |err| return errorLogger(err);
 }
 
 pub export fn apeCmdEndQuery(p_cmd: vk.CommandBuffer, p_pool: vk.QueryPool, query: u32) callconv(vk.vulkan_call_conv) void {
@@ -2265,8 +2269,9 @@ pub export fn apeCmdUpdateBuffer(p_cmd: vk.CommandBuffer, p_buffer: vk.Buffer, o
 
     const cmd = Dispatchable(CommandBuffer).fromHandleObject(p_cmd) catch |err| return errorLogger(err);
     const buffer = NonDispatchable(Buffer).fromHandleObject(p_buffer) catch |err| return errorLogger(err);
+    const data_size = std.math.cast(usize, size) orelse return errorLogger(VkError.ValidationFailed);
     const data_bytes: [*]const u8 = @ptrCast(data);
-    cmd.updateBuffer(buffer, offset, data_bytes[0..size]) catch |err| return errorLogger(err);
+    cmd.updateBuffer(buffer, offset, data_bytes[0..data_size]) catch |err| return errorLogger(err);
 }
 
 pub export fn apeCmdWaitEvents(