Moved read stuff to GpuBuffer directly

src/GpuBuffer.zig

@@ -22,22 +22,16 @@ const BufferUsage = enum(u64) {
 };
 
 /// Allocates the underlying WebGPU handle and registers it to the parent GpuAllocator
-pub fn init(gloc: GpuAllocator, T: type, len: usize, usage: std.EnumSet(BufferUsage)) !@This() {
+pub fn init(gloc: GpuAllocator, size: u64, usage: std.EnumSet(BufferUsage)) !@This() {
-    switch (@typeInfo(T)) {
-        .int, .float => {},
-        else => @compileError("GpuBuffer can only use int and float type"),
-    }
-
     var use: u64 = 0;
     var iter = usage.iterator();
     while (iter.next()) |flag| use |= @intFromEnum(flag);
 
-    const bytes = @sizeOf(T) * len;
+    const raw_handle = try gloc.allocBuffer(size, use);
-    const raw_handle = try gloc.allocBuffer(bytes, use);
 
     return .{
         .raw = raw_handle,
-        .size = bytes,
+        .size = size,
         .usage = use,
         .gloc = gloc,
     };
@@ -48,6 +42,11 @@ pub fn deinit(self: @This()) void {
     self.gloc.freeBuffer(self.raw, self.size);
 }
 
+/// Native getConstMappedRange wrapper
+pub fn getConstMappedRange(self: @This(), offset: u64, size: u64) ?*const anyopaque {
+    return c.wgpuBufferGetConstMappedRange(self.raw, offset, size);
+}
+
 /// Native mapAsync wrapper
 pub fn mapAsync(
     self: @This(),
@@ -59,11 +58,6 @@ pub fn mapAsync(
     _ = c.wgpuBufferMapAsync(self.raw, mode, offset, size, callback_info);
 }
 
-/// Native getConstMappedRange wrapper
-pub fn getConstMappedRange(self: @This(), offset: u64, size: u64) ?*const anyopaque {
-    return c.wgpuBufferGetConstMappedRange(self.raw, offset, size);
-}
-
 /// Native unmap wrapper
 pub fn unmap(self: @This()) void {
     c.wgpuBufferUnmap(self.raw);
@@ -76,3 +70,48 @@ pub fn load(
 ) !void {
     c.wgpuQueueWriteBuffer(self.gloc.device.queue, self.raw, 0, data.ptr, self.size);
 }
+
+pub fn read(self: @This(), alloc: std.mem.Allocator, T: type) ![]f16 {
+    const out = try alloc.alloc(T, @divExact(self.size, @sizeOf(T)));
+
+    const staging = try init(
+        self.gloc,
+        self.size,
+        .initMany(&.{ .MapRead, .CopyDst }),
+    );
+    defer staging.deinit();
+
+    const enc = c.wgpuDeviceCreateCommandEncoder(self.gloc.device.device, null) orelse return error.Encoder;
+    c.wgpuCommandEncoderCopyBufferToBuffer(enc, self.raw, 0, staging.raw, 0, self.size);
+    const cmd = c.wgpuCommandEncoderFinish(enc, null);
+    defer c.wgpuCommandEncoderRelease(enc);
+    defer c.wgpuCommandBufferRelease(cmd);
+    c.wgpuQueueSubmit(self.gloc.device.queue, 1, &cmd);
+
+    var mapped = false;
+    staging.mapAsync(
+        c.WGPUMapMode_Read,
+        0,
+        self.size,
+        .{ .callback = onMapped, .userdata1 = &mapped },
+    );
+    while (!mapped) self.gloc.device.poll();
+
+    const ptr: [*]const T = @ptrCast(@alignCast(
+        staging.getConstMappedRange(0, self.size),
+    ));
+    @memcpy(out[0..out.len], ptr[0..out.len]);
+    staging.unmap();
+
+    return out;
+}
+
+fn onMapped(
+    status: c.WGPUMapAsyncStatus,
+    _: c.WGPUStringView,
+    userdata1: ?*anyopaque,
+    _: ?*anyopaque,
+) callconv(.c) void {
+    const flag: *bool = @ptrCast(@alignCast(userdata1.?));
+    flag.* = (status == c.WGPUMapAsyncStatus_Success);
+}

src/Vec.zig

@@ -15,8 +15,7 @@ pub fn initZero(gloc: GpuAllocator, len: usize) !Vec {
     return .{
         .buf = try GpuBuffer.init(
             gloc,
-            f16,
+            len * @sizeOf(f16),
-            len,
             .initMany(&.{ .Storage, .CopyDst, .CopySrc }),
         ),
         .len = len,
@@ -59,41 +58,8 @@ pub fn run(self: Vec, gloc: GpuAllocator, other: Vec, pip: GpuPipeline) !Vec {
 }
 
 // Changed: gloc is passed by value instead of *GpuAllocator
-pub fn read(self: Vec, gloc: GpuAllocator, alloc: std.mem.Allocator) ![]f16 {
+pub fn read(self: Vec, alloc: std.mem.Allocator) ![]f16 {
-    const out = try alloc.alloc(f16, self.len);
+    return self.buf.read(alloc, f16);
-    const bytes = self.byteSize();
-
-    const staging = try GpuBuffer.init(
-        gloc,
-        f16,
-        self.len,
-        .initMany(&.{ .MapRead, .CopyDst }),
-    );
-    defer staging.deinit();
-
-    const enc = c.wgpuDeviceCreateCommandEncoder(gloc.device.device, null) orelse return error.Encoder;
-    c.wgpuCommandEncoderCopyBufferToBuffer(enc, self.buf.raw, 0, staging.raw, 0, bytes);
-    const cmd = c.wgpuCommandEncoderFinish(enc, null);
-    defer c.wgpuCommandEncoderRelease(enc);
-    defer c.wgpuCommandBufferRelease(cmd);
-    c.wgpuQueueSubmit(gloc.device.queue, 1, &cmd);
-
-    var mapped = false;
-    staging.mapAsync(
-        c.WGPUMapMode_Read,
-        0,
-        bytes,
-        .{ .callback = onMapped, .userdata1 = &mapped },
-    );
-    while (!mapped) gloc.device.poll();
-
-    const ptr: [*]const f16 = @ptrCast(@alignCast(
-        staging.getConstMappedRange(0, bytes),
-    ));
-    @memcpy(out[0..self.len], ptr[0..self.len]);
-    staging.unmap();
-
-    return out;
 }
 
 fn onMapped(
@@ -124,8 +90,7 @@ fn dispatch2in1out(
 
         const info_buf = try GpuBuffer.init(
             gloc,
-            u32,
+            @sizeOf(u32),
-            1,
             .initMany(&.{ .Uniform, .CopyDst }),
         );
         defer info_buf.deinit();

src/example.zig

@@ -41,7 +41,7 @@ pub fn main(init: std.process.Init) !void {
 
     std.debug.print("Bytes used: {d} (3 * {d})\n", .{ grena.allocated_vram_bytes, a.byteSize() });
 
-    const out = try sum.read(gloc, allocator);
+    const out = try sum.read(allocator);
     defer allocator.free(out);
 
     std.debug.print("{any}\n", .{out});