GpuBuffer read no longer allocate an staging buffer itself. Instead need to manually do it and use new copy function to move data from a buffer to a mapable one

README.md

@@ -8,7 +8,7 @@ The library exports the following primary components:
 
 * **`GpuDevice`**: Initializes the WebGPU instance, adapter, device, and queue. It is configured to prioritize high performance and automatically requests the `ShaderF16` feature if the adapter supports it. It provides the base `GpuAllocator` for raw VRAM allocations.
 * **`GpuArenaAllocator`**: A memory management layer that wraps a base allocator to track and automatically destroy all allocated WebGPU buffers, textures, views, and pipelines when deinitialized.
-* **`GpuBuffer`**: Wraps native WebGPU buffers. It provides a `.load()` method for CPU-to-GPU data transfers and a `.read()` method to map GPU data back to the CPU.
+* **`GpuBuffer`**: Wraps native WebGPU buffers. It provides a `.load()` method for CPU-to-GPU data transfers and a `.read()` method that utilizes a staging buffer to map GPU data back to the CPU.
 * **`GpuCompute`**: Compiles WGSL source code into a compute pipeline and dispatches compute workgroups.
 * **`GpuRender` / `GpuTexture` / `GpuTextureView`**: Components used to initialize render pipelines, set up render attachments (textures), and bind render targets for offscreen drawing.
 
@@ -36,11 +36,11 @@ pub fn main(init: std.process.Init) !void {
     // 2. Create a GPU Arena to manage VRAM
     var grena = GpuArenaAllocator.init(allocator, device.gpuAllocator());
     defer grena.deinit();
-    const glloc = grena.gpuAllocator();
+    const gloc = grena.gpuAllocator();
 
     // 3. Load the WGSL compute pipeline
     const add_cp = try GpuCompute.init(
-        glloc,
+        gloc,
         @embedFile("shaders/add.wgsl"),
         .{
             .label = "add",
@@ -66,9 +66,9 @@ pub fn main(init: std.process.Init) !void {
 
     // 5. Initialize raw GPU Buffers
     const byte_size = len * @sizeOf(f16);
-    const buf_a = try GpuBuffer.init(glloc, .{ .label = "a", .size = byte_size, .usage = .initMany(&.{ .Storage, .CopyDst, .CopySrc }) });
+    const buf_a = try GpuBuffer.init(gloc, .{ .label = "a", .size = byte_size, .usage = .initMany(&.{ .Storage, .CopyDst, .CopySrc }) });
-    const buf_b = try GpuBuffer.init(glloc, .{ .label = "b", .size = byte_size, .usage = .initMany(&.{ .Storage, .CopyDst, .CopySrc }) });
+    const buf_b = try GpuBuffer.init(gloc, .{ .label = "b", .size = byte_size, .usage = .initMany(&.{ .Storage, .CopyDst, .CopySrc }) });
-    const buf_out = try GpuBuffer.init(glloc, .{ .label = "out", .size = byte_size, .usage = .initMany(&.{ .Storage, .CopyDst, .CopySrc }) });
+    const buf_out = try GpuBuffer.init(gloc, .{ .label = "out", .size = byte_size, .usage = .initMany(&.{ .Storage, .CopyDst, .CopySrc }) });
 
     // Note: Buffers are safely tied to the GpuArenaAllocator which will automatically
     // release them at the end. You can also manually call buf_x.deinit() if desired.
@@ -79,7 +79,7 @@ pub fn main(init: std.process.Init) !void {
     try buf_b.load(f16, data_b);
 
     // 7. Dispatch the Compute
-    try add_cp.run(glloc, .{ buf_a, buf_b, buf_out });
+    try add_cp.run(gloc, .{ buf_a, buf_b, buf_out });
 
     // 8. Map and copy the resulting buffer back to the CPU
     const out = try buf_out.read(allocator, f16);
@@ -119,18 +119,18 @@ pub fn main(init: std.process.Init) !void {
     // 2. Init VRAM Arena
     var grena = GpuArenaAllocator.init(allocator, device.gpuAllocator());
     defer grena.deinit();
-    const glloc = grena.gpuAllocator();
+    const gloc = grena.gpuAllocator();
 
     // 3. Load Render Pipeline
     const circle_rp = try GpuRender.init(
-        glloc,
+        gloc,
         @embedFile("shaders/circle.wgsl"),
         .{ .bindings = &.{}, .texture_format = .RGBA8Unorm, .topology = .TriangleStrip },
     );
     defer circle_rp.deinit();
 
     // 4. Create VRAM texture to render into
-    const texture = try GpuTexture.init(glloc, .{
+    const texture = try GpuTexture.init(gloc, .{
         .format = .RGBA8Unorm,
         .size = .{ .width = width, .height = height, .depthOrArrayLayers = 1 },
         .usage = .initMany(&.{ .RenderAttachment, .CopySrc }),
@@ -138,14 +138,14 @@ pub fn main(init: std.process.Init) !void {
     defer texture.deinit();
 
     // 5. Create a view from texture
-    const view = try GpuTextureView.init(glloc, texture, .{});
+    const view = try GpuTextureView.init(gloc, texture, .{});
     defer view.deinit();
 
     // 6. Run the rendering pipeline
-    try circle_rp.draw(glloc, view, 4, .{});
+    try circle_rp.draw(gloc, view, 4, .{});
 
     // 7. Load Texture into GpuBuffer
-    const cpu_staging_cpu = try texture.buffCopy(glloc);
+    const cpu_staging_cpu = try texture.buffCopy(gloc);
     defer cpu_staging_cpu.deinit();
 
     // 8. Read GpuBuffer to CPU
@@ -223,7 +223,7 @@ No extra installation required. Automatically links against `d3d12`, `dxgi`, and
 Add it to your `build.zig.zon`:
 
 ```bash
-zig fetch --save git+https://git.bouvais.lu/adrien/zig-wgpu#ref=0.2.1
+zig fetch --save git+[https://git.bouvais.lu/adrien/zig-wgpu#ref=0.2.0](https://git.bouvais.lu/adrien/zig-wgpu)
 ```
 
 Then, expose it in your `build.zig`:

build.zig.zon

@@ -1,6 +1,6 @@
 .{
     .name = .zig_wgpu,
-    .version = "0.2.3",
+    .version = "0.2.0",
     .fingerprint = 0x5d0e853acbc0c2c6,
     .minimum_zig_version = "0.16.0",
     .dependencies = .{},

examples/bench_cp.zig

@@ -13,10 +13,10 @@ const Vec = struct {
     buf: GpuBuffer,
     len: usize,
 
-    // Changed: glloc is passed by value (const)
+    // Changed: gloc is passed by value (const)
-    pub fn initZero(glloc: GpuAllocator, len: usize) !Vec {
+    pub fn initZero(gloc: GpuAllocator, len: usize) !Vec {
         return .{
-            .buf = try GpuBuffer.init(glloc, .{
+            .buf = try GpuBuffer.init(gloc, .{
                 .size = len * @sizeOf(f16),
                 .usage = .initMany(&.{ .Storage, .CopyDst, .CopySrc }),
             }),
@@ -24,9 +24,9 @@ const Vec = struct {
         };
     }
 
-    // Changed: glloc is passed by value
+    // Changed: gloc is passed by value
-    pub fn initLoad(glloc: GpuAllocator, data: []const f16) !Vec {
+    pub fn initLoad(gloc: GpuAllocator, data: []const f16) !Vec {
-        var self = try initZero(glloc, data.len);
+        var self = try initZero(gloc, data.len);
         try self.load(data); // Direct access via the interface copy
         return self;
     }
@@ -40,18 +40,18 @@ const Vec = struct {
         try self.buf.load(f16, data);
     }
 
-    // Changed: glloc is passed by value instead of *GpuAllocator
+    // Changed: gloc is passed by value instead of *GpuAllocator
-    pub fn run(self: Vec, glloc: GpuAllocator, other: Vec, process: GpuCompute) !Vec {
+    pub fn run(self: Vec, gloc: GpuAllocator, other: Vec, process: GpuCompute) !Vec {
         std.debug.assert(self.len == other.len);
 
-        const result = try Vec.initZero(glloc, self.len);
+        const result = try Vec.initZero(gloc, self.len);
         errdefer result.deinit();
 
-        try process.run(glloc, .{ self.buf, other.buf, result.buf });
+        try process.run(gloc, .{ self.buf, other.buf, result.buf });
         return result;
     }
 
-    // Changed: glloc is passed by value instead of *GpuAllocator
+    // Changed: gloc is passed by value instead of *GpuAllocator
     pub fn read(self: Vec, alloc: std.mem.Allocator) ![]f16 {
         return self.buf.read(alloc, f16);
     }
@@ -63,9 +63,9 @@ pub fn main(init: std.process.Init) !void {
 
     var grena = GpuArenaAllocator.init(init.gpa, device.gpuAllocator());
     defer grena.deinit();
-    const glloc = grena.gpuAllocator();
+    const gloc = grena.gpuAllocator();
 
-    const add_pip = try GpuCompute.init(glloc, @embedFile("shaders/add.wgsl"), .{ .bindings = &.{
+    const add_pip = try GpuCompute.init(gloc, @embedFile("shaders/add.wgsl"), .{ .bindings = &.{
         .{ .element_size = @sizeOf(f16) },
         .{ .element_size = @sizeOf(f16) },
         .{ .element_size = @sizeOf(f16) },
@@ -120,9 +120,9 @@ pub fn main(init: std.process.Init) !void {
             // --- 1. GPU ALLOCATION PHASE ---
             const alloc_start = std.Io.Clock.awake.now(init.io);
 
-            const a = try Vec.initLoad(glloc, data_a);
+            const a = try Vec.initLoad(gloc, data_a);
             defer a.deinit();
-            const b = try Vec.initLoad(glloc, data_b);
+            const b = try Vec.initLoad(gloc, data_b);
             defer b.deinit();
 
             const alloc_duration = alloc_start.durationTo(std.Io.Clock.awake.now(init.io));
@@ -132,7 +132,7 @@ pub fn main(init: std.process.Init) !void {
             // --- 2. COMPUTE PHASE ---
             const compute_start = std.Io.Clock.awake.now(init.io);
 
-            const sum = try a.run(glloc, b, add_pip);
+            const sum = try a.run(gloc, b, add_pip);
             defer sum.deinit();
 
             // All 3 buffers (a, b, sum) are currently resident in VRAM here.

examples/circle.zig

@@ -20,18 +20,18 @@ pub fn main(init: std.process.Init) !void {
     // 2. Init VRAM Arena
     var grena = GpuArenaAllocator.init(allocator, device.gpuAllocator());
     defer grena.deinit();
-    const glloc = grena.gpuAllocator();
+    const gloc = grena.gpuAllocator();
 
     // 3. Load Render Pipeline
     const circle_rp = try GpuRender.init(
-        glloc,
+        gloc,
         @embedFile("shaders/circle.wgsl"),
         .{ .bindings = &.{}, .texture_format = .RGBA8Unorm, .topology = .TriangleStrip },
     );
     defer circle_rp.deinit();
 
     // 4. Create VRAM texture to render into
-    const texture = try GpuTexture.init(glloc, .{
+    const texture = try GpuTexture.init(gloc, .{
         .format = .RGBA8Unorm,
         .size = .{ .width = width, .height = height, .depthOrArrayLayers = 1 },
         .usage = .initMany(&.{ .RenderAttachment, .CopySrc }),
@@ -39,14 +39,14 @@ pub fn main(init: std.process.Init) !void {
     defer texture.deinit();
 
     // 5. Create a view from texture
-    const view = try GpuTextureView.init(glloc, texture, .{});
+    const view = try GpuTextureView.init(gloc, texture, .{});
     defer view.deinit();
 
     // 6. Run the rendering pipeline
-    try circle_rp.draw(glloc, view, 4, .{});
+    try circle_rp.draw(gloc, view, 4, .{});
 
     // 7. Load Texture into GpuBuffer
-    const cpu_staging_cpu = try texture.buffCopy(glloc);
+    const cpu_staging_cpu = try texture.buffCopy(gloc);
     defer cpu_staging_cpu.deinit();
 
     // 8. Read GpuBuffer to CPU

examples/compute.zig

@@ -15,11 +15,11 @@ pub fn main(init: std.process.Init) !void {
     // 2. Create a GPU Arena to manage VRAM
     var grena = GpuArenaAllocator.init(allocator, device.gpuAllocator());
     defer grena.deinit();
-    const glloc = grena.gpuAllocator();
+    const gloc = grena.gpuAllocator();
 
     // 3. Load the WGSL compute pipeline
     const add_cp = try GpuCompute.init(
-        glloc,
+        gloc,
         @embedFile("shaders/add.wgsl"),
         .{
             .label = "add",
@@ -32,7 +32,7 @@ pub fn main(init: std.process.Init) !void {
     );
 
     // 4. Setup CPU data
-    const len: usize = 16;
+    const len: usize = 1024;
     const data_a = try allocator.alloc(f16, len);
     defer allocator.free(data_a);
     const data_b = try allocator.alloc(f16, len);
@@ -45,9 +45,9 @@ pub fn main(init: std.process.Init) !void {
 
     // 5. Initialize raw GPU Buffers
     const byte_size = len * @sizeOf(f16);
-    const buf_a = try GpuBuffer.init(glloc, .{ .label = "a", .size = byte_size, .usage = .initMany(&.{ .Storage, .CopyDst, .CopySrc }) });
+    const buf_a = try GpuBuffer.init(gloc, .{ .label = "a", .size = byte_size, .usage = .initMany(&.{ .Storage, .CopyDst, .CopySrc }) });
-    const buf_b = try GpuBuffer.init(glloc, .{ .label = "b", .size = byte_size, .usage = .initMany(&.{ .Storage, .CopyDst, .CopySrc }) });
+    const buf_b = try GpuBuffer.init(gloc, .{ .label = "b", .size = byte_size, .usage = .initMany(&.{ .Storage, .CopyDst, .CopySrc }) });
-    const buf_out = try GpuBuffer.init(glloc, .{ .label = "out", .size = byte_size, .usage = .initMany(&.{ .Storage, .CopyDst, .CopySrc }) });
+    const buf_out = try GpuBuffer.init(gloc, .{ .label = "out", .size = byte_size, .usage = .initMany(&.{ .Storage, .CopyDst, .CopySrc }) });
 
     // Note: Buffers are safely tied to the GpuArenaAllocator which will automatically
     // release them at the end. You can also manually call buf_x.deinit() if desired.
@@ -58,11 +58,18 @@ pub fn main(init: std.process.Init) !void {
     try buf_b.load(f16, data_b);
 
     // 7. Dispatch the Compute
-    try add_cp.run(glloc, .{ buf_a, buf_b, buf_out });
+    try add_cp.run(gloc, .{ buf_a, buf_b, buf_out });
 
     // 8. Map and copy the resulting buffer back to the CPU
-    const out = try buf_out.read(allocator, f16);
+    const staging = try GpuBuffer.init(gloc, .{
+        .size = byte_size,
+        .usage = .initMany(&.{ .MapRead, .CopyDst }),
+    });
+    defer staging.deinit();
+
+    try buf_out.copy(staging);
+    const out = try staging.read(allocator, f16);
     defer allocator.free(out);
 
-    std.debug.print("Result: {any}\n", .{out});
+    std.debug.print("Result: {any}\n", .{out[0..@min(6, len)]});
 }

src/GpuArenaAllocator.zig

@@ -87,6 +87,14 @@ pub fn gpuAllocator(self: *@This()) GpuAllocator {
 fn allocBuffer(ctx: *anyopaque, desc: c.WGPUBufferDescriptor) anyerror!c.WGPUBuffer {
     const self: *@This() = @ptrCast(@alignCast(ctx));
     try self.tracked_buffers.ensureTotalCapacity(self.tracked_buffers.count() + 1);
+
+    const bytes = desc.size;
+    if (bytes > self.child_allocator.device.limits.maxBufferSize)
+        return error.SingleBufferExceedsLimit;
+
+    if (bytes + self.allocated_vram_bytes > self.child_allocator.device.def.vram_bytes_limit)
+        return error.ExceedsVramBudget;
+
     const raw = try self.child_allocator.allocBuffer(desc);
     self.tracked_buffers.putAssumeCapacity(raw, desc);
     self.allocated_vram_bytes += desc.size;

src/GpuBuffer.zig

@@ -4,11 +4,10 @@ const GpuAllocator = @import("GpuAllocator.zig");
 const svOpt = @import("utils.zig").svOpt;
 
 raw: c.WGPUBuffer,
-size: u64,
+gloc: GpuAllocator,
-usage: c.WGPUBufferUsage,
+def: GpuBufferDef,
-glloc: GpuAllocator,
 
-const BufferUsage = enum(u64) {
+pub const GpuBufferUsage = enum(u64) {
     None = 0x0000000000000000,
     MapRead = 0x0000000000000001,
     MapWrite = 0x0000000000000002,
@@ -20,37 +19,40 @@ const BufferUsage = enum(u64) {
     Storage = 0x0000000000000080,
     Indirect = 0x0000000000000100,
     QueryResolve = 0x0000000000000200,
+
+    fn enumSetToWGPUBufferUsage(set: std.EnumSet(GpuBufferUsage)) c.WGPUBufferUsage {
+        var use: u64 = 0;
+        var iter = set.iterator();
+        while (iter.next()) |flag| use |= @intFromEnum(flag);
+        return use;
+    }
 };
 
-const GpuBufferDef = struct {
+pub const GpuBufferDef = struct {
     label: ?[]const u8 = null,
     size: u64,
-    usage: std.EnumSet(BufferUsage),
+    usage: std.EnumSet(GpuBufferUsage),
 };
 
-pub fn init(glloc: GpuAllocator, def: GpuBufferDef) !@This() {
+pub fn init(gloc: GpuAllocator, def: GpuBufferDef) !@This() {
-    var use: u64 = 0;
-    var iter = def.usage.iterator();
-    while (iter.next()) |flag| use |= @intFromEnum(flag);
 
     // Automatically align the buffer size forward to a multiple of 4 bytes under the hood
     const aligned_size = std.mem.alignForward(u64, def.size, 4);
 
-    const raw_handle = try glloc.allocBuffer(.{
+    const raw_handle = try gloc.allocBuffer(.{
         .size = aligned_size,
-        .usage = use,
+        .usage = GpuBufferUsage.enumSetToWGPUBufferUsage(def.usage),
         .label = svOpt(def.label),
     });
     return .{
         .raw = raw_handle,
-        .size = aligned_size,
+        .def = def,
-        .usage = use,
+        .gloc = gloc,
-        .glloc = glloc,
     };
 }
 
 pub fn deinit(self: @This()) void {
-    self.glloc.freeBuffer(self.raw);
+    self.gloc.freeBuffer(self.raw);
 }
 
 pub fn getConstMappedRange(self: @This(), offset: u64, size: u64) ?*const anyopaque {
@@ -79,57 +81,46 @@ pub fn load(
 ) !void {
     const bytes = data.len * @sizeOf(T);
 
-    if (bytes == self.size) {
+    if (bytes == self.def.size) {
         // Aligned path: direct download
-        c.wgpuQueueWriteBuffer(self.glloc.device.queue, self.raw, 0, data.ptr, self.size);
+        c.wgpuQueueWriteBuffer(self.gloc.device.queue, self.raw, 0, data.ptr, self.def.size);
     } else {
         // Unaligned path: Split the write into an aligned chunk and a padded remainder
         // to support arbitrary lengths without any allocations or large stack arrays.
         const aligned_part = (bytes / 4) * 4;
         if (aligned_part > 0) {
-            c.wgpuQueueWriteBuffer(self.glloc.device.queue, self.raw, 0, data.ptr, aligned_part);
+            c.wgpuQueueWriteBuffer(self.gloc.device.queue, self.raw, 0, data.ptr, aligned_part);
         }
 
         var remainder_buf: [4]u8 = .{ 0, 0, 0, 0 };
         const data_bytes = std.mem.sliceAsBytes(data);
         @memcpy(remainder_buf[0 .. bytes - aligned_part], data_bytes[aligned_part..bytes]);
 
-        c.wgpuQueueWriteBuffer(self.glloc.device.queue, self.raw, aligned_part, &remainder_buf, 4);
+        c.wgpuQueueWriteBuffer(self.gloc.device.queue, self.raw, aligned_part, &remainder_buf, 4);
     }
 }
 
 /// GPU to CPU
+/// Buffer must have MapRead usage or returns error.BufferNotMappable.
 pub fn read(self: @This(), alloc: std.mem.Allocator, T: type) ![]T {
-    const out = try alloc.alloc(T, @divExact(self.size, @sizeOf(T)));
+    if (!self.def.usage.contains(.MapRead)) return error.BufferNotMappable;
 
-    const staging = try init(self.glloc, .{
+    const out = try alloc.alloc(T, @divExact(self.def.size, @sizeOf(T)));
-        .size = self.size,
-        .usage = .initMany(&.{ .MapRead, .CopyDst }),
-        .label = "staging_read_buffer",
-    });
-    defer staging.deinit();
-
-    const enc = c.wgpuDeviceCreateCommandEncoder(self.glloc.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.glloc.device.queue, 1, &cmd);
 
     var mapped = false;
-    staging.mapAsync(
+    self.mapAsync(
         c.WGPUMapMode_Read,
         0,
-        self.size,
+        self.def.size,
         .{ .callback = onMapped, .userdata1 = &mapped },
     );
-    while (!mapped) self.glloc.device.poll();
+    while (!mapped) self.gloc.device.poll();
 
     const ptr: [*]const T = @ptrCast(@alignCast(
-        staging.getConstMappedRange(0, self.size),
+        self.getConstMappedRange(0, self.def.size),
     ));
     @memcpy(out[0..out.len], ptr[0..out.len]);
-    staging.unmap();
+    self.unmap();
 
     return out;
 }
@@ -143,3 +134,20 @@ fn onMapped(
     const flag: *bool = @ptrCast(@alignCast(userdata1.?));
     flag.* = (status == c.WGPUMapAsyncStatus_Success);
 }
+
+/// GPU to GPU. Both buffers must be same size, src needs CopySrc, dst needs CopyDst.
+pub fn copy(src: @This(), dst: @This()) !void {
+    if (src.def.size != dst.def.size) return error.SizeMismatch;
+
+    const copy_src: u64 = @intFromEnum(GpuBufferUsage.CopySrc);
+    const copy_dst: u64 = @intFromEnum(GpuBufferUsage.CopyDst);
+    if (@as(u64, GpuBufferUsage.enumSetToWGPUBufferUsage(src.def.usage)) & copy_src == 0) return error.SrcNotCopyable;
+    if (@as(u64, GpuBufferUsage.enumSetToWGPUBufferUsage(dst.def.usage)) & copy_dst == 0) return error.DstNotWritable;
+
+    const enc = c.wgpuDeviceCreateCommandEncoder(src.gloc.device.device, null) orelse return error.Encoder;
+    c.wgpuCommandEncoderCopyBufferToBuffer(enc, src.raw, 0, dst.raw, 0, src.def.size);
+    const cmd = c.wgpuCommandEncoderFinish(enc, null);
+    defer c.wgpuCommandEncoderRelease(enc);
+    defer c.wgpuCommandBufferRelease(cmd);
+    c.wgpuQueueSubmit(src.gloc.device.queue, 1, &cmd);
+}

src/GpuCompute.zig

@@ -22,40 +22,40 @@ pub const ComputeDef = struct {
 };
 
 pip: c.WGPUComputePipeline,
-glloc: GpuAllocator,
+gloc: GpuAllocator,
 def: ComputeDef,
 
-pub fn init(glloc: GpuAllocator, wgsl: []const u8, def: ComputeDef) !@This() {
+pub fn init(gloc: GpuAllocator, wgsl: []const u8, def: ComputeDef) !@This() {
     var wgsl_src = c.WGPUShaderSourceWGSL{
         .chain = .{ .sType = c.WGPUSType_ShaderSourceWGSL },
         .code = sv(wgsl),
     };
-    const shader = c.wgpuDeviceCreateShaderModule(glloc.device.device, &.{
+    const shader = c.wgpuDeviceCreateShaderModule(gloc.device.device, &.{
         .nextInChain = @ptrCast(&wgsl_src),
     }) orelse return error.Shader;
     defer c.wgpuShaderModuleRelease(shader);
 
-    const pip = try glloc.allocComputePipeline(.{
+    const pip = try gloc.allocComputePipeline(.{
         .label = svOpt(def.label),
         .compute = .{ .module = shader, .entryPoint = sv("main") },
     });
 
     return .{
-        .glloc = glloc,
+        .gloc = gloc,
         .pip = pip,
         .def = def,
     };
 }
 
 pub fn deinit(self: @This()) void {
-    self.glloc.freeComputePipeline(self.pip);
+    self.gloc.freeComputePipeline(self.pip);
 }
 
 /// Execute the compute pass with arbitrary buffer bindings via a tuple.
-/// Example: `try proc.run(glloc, .{ buf_a, buf_b, buf_out });`
+/// Example: `try proc.run(gloc, .{ buf_a, buf_b, buf_out });`
 pub fn run(
     self: @This(),
-    glloc: GpuAllocator,
+    gloc: GpuAllocator,
     args: anytype,
 ) !void {
     const type_info = @typeInfo(@TypeOf(args));
@@ -74,7 +74,7 @@ pub fn run(
             const buf = @field(args, field.name);
             const el_size = self.def.bindings[i].element_size;
             if (el_size > 0) {
-                elements_count = @intCast(buf.size / el_size);
+                elements_count = @intCast(buf.def.size / el_size);
             }
         }
     }
@@ -85,7 +85,7 @@ pub fn run(
         const el_size = self.def.bindings[i].element_size;
         if (el_size > 0) {
             const expected_min_bytes = @as(u64, elements_count) * el_size;
-            if (buf.size < expected_min_bytes)
+            if (buf.def.size < expected_min_bytes)
                 return error.BufferTooSmall;
         }
     }
@@ -103,7 +103,7 @@ pub fn run(
             .binding = @intCast(i),
             .buffer = buf.raw,
             .offset = 0,
-            .size = buf.size, // Size exposes the fully allocated length
+            .size = buf.def.size, // Size exposes the fully allocated length
         };
         entry_count += 1;
     }
@@ -113,12 +113,12 @@ pub fn run(
     defer if (info_buf) |b| b.deinit();
 
     if (self.def.append_info_buffer) {
-        info_buf = try GpuBuffer.init(glloc, .{
+        info_buf = try GpuBuffer.init(gloc, .{
             .size = @sizeOf(u32),
             .usage = .initMany(&.{ .Uniform, .CopyDst }),
             .label = "compute_info_buffer",
         });
-        c.wgpuQueueWriteBuffer(glloc.device.queue, info_buf.?.raw, 0, &elements_count, @sizeOf(u32));
+        c.wgpuQueueWriteBuffer(gloc.device.queue, info_buf.?.raw, 0, &elements_count, @sizeOf(u32));
 
         entries_buf[entry_count] = .{
             .binding = @intCast(entry_count),
@@ -130,11 +130,11 @@ pub fn run(
     }
 
     const entries = entries_buf[0..entry_count];
-    try submitPass(glloc, self.pip, entries, elements_count, self.def.workgroup_size, self.def.max_workgroups);
+    try submitPass(gloc, self.pip, entries, elements_count, self.def.workgroup_size, self.def.max_workgroups);
 }
 
 fn submitPass(
-    glloc: GpuAllocator,
+    gloc: GpuAllocator,
     pipeline: c.WGPUComputePipeline,
     entries: []const c.WGPUBindGroupEntry,
     n: usize,
@@ -146,14 +146,14 @@ fn submitPass(
     const bgl = c.wgpuComputePipelineGetBindGroupLayout(pipeline, 0);
     defer c.wgpuBindGroupLayoutRelease(bgl);
 
-    const bg = c.wgpuDeviceCreateBindGroup(glloc.device.device, &.{
+    const bg = c.wgpuDeviceCreateBindGroup(gloc.device.device, &.{
         .layout = bgl,
         .entries = entries.ptr,
         .entryCount = entries.len,
     }) orelse return error.BindGroup;
     defer c.wgpuBindGroupRelease(bg);
 
-    const enc = c.wgpuDeviceCreateCommandEncoder(glloc.device.device, null) orelse return error.Encoder;
+    const enc = c.wgpuDeviceCreateCommandEncoder(gloc.device.device, null) orelse return error.Encoder;
     const pass = c.wgpuCommandEncoderBeginComputePass(enc, null);
     c.wgpuComputePassEncoderSetPipeline(pass, pipeline);
     c.wgpuComputePassEncoderSetBindGroup(pass, 0, bg, 0, null);
@@ -168,7 +168,7 @@ fn submitPass(
     const cmd = c.wgpuCommandEncoderFinish(enc, null);
     defer c.wgpuCommandEncoderRelease(enc);
     defer c.wgpuCommandBufferRelease(cmd);
-    c.wgpuQueueSubmit(glloc.device.queue, 1, &cmd);
+    c.wgpuQueueSubmit(gloc.device.queue, 1, &cmd);
 }
 
 fn ceilDiv(n: usize, d: usize) usize {

src/GpuRender.zig

@@ -34,16 +34,16 @@ const GpuPrimitiveTopology = enum(c_uint) {
     Force32 = 0x7FFFFFFF,
 };
 
-glloc: GpuAllocator,
+gloc: GpuAllocator,
 pip: c.WGPURenderPipeline,
 def: GpuRenderDef,
 
-pub fn init(glloc: GpuAllocator, wgsl: []const u8, def: GpuRenderDef) !@This() {
+pub fn init(gloc: GpuAllocator, wgsl: []const u8, def: GpuRenderDef) !@This() {
     var wgsl_src = c.WGPUShaderSourceWGSL{
         .chain = .{ .sType = c.WGPUSType_ShaderSourceWGSL },
         .code = sv(wgsl),
     };
-    const shader = c.wgpuDeviceCreateShaderModule(glloc.device.device, &.{
+    const shader = c.wgpuDeviceCreateShaderModule(gloc.device.device, &.{
         .nextInChain = @ptrCast(&wgsl_src),
     }) orelse return error.Shader;
     defer c.wgpuShaderModuleRelease(shader);
@@ -69,7 +69,7 @@ pub fn init(glloc: GpuAllocator, wgsl: []const u8, def: GpuRenderDef) !@This() {
     };
 
     // 3. Compile the Complete Render Pipeline
-    const pip = try glloc.allocRenderPipeline(.{
+    const pip = try gloc.allocRenderPipeline(.{
         .label = svOpt(def.label),
         .vertex = .{
             .module = shader,
@@ -90,21 +90,21 @@ pub fn init(glloc: GpuAllocator, wgsl: []const u8, def: GpuRenderDef) !@This() {
     });
 
     return .{
-        .glloc = glloc,
+        .gloc = gloc,
         .pip = pip,
         .def = def,
     };
 }
 
 pub fn deinit(self: @This()) void {
-    self.glloc.freeRenderPipeline(self.pip);
+    self.gloc.freeRenderPipeline(self.pip);
 }
 
 /// Execute the render pass targeting a specific frame texture view.
 /// Passes bind groups via a tuple exactly like your original compute setup.
 pub fn draw(
     self: @This(),
-    glloc: GpuAllocator,
+    gloc: GpuAllocator,
     target_view: GpuTextureView,
     vertex_count: u32,
     args: anytype,
@@ -138,7 +138,7 @@ pub fn draw(
     const bgl = c.wgpuRenderPipelineGetBindGroupLayout(self.pip, 0);
     defer c.wgpuBindGroupLayoutRelease(bgl);
 
-    const bg = c.wgpuDeviceCreateBindGroup(glloc.device.device, &.{
+    const bg = c.wgpuDeviceCreateBindGroup(gloc.device.device, &.{
         .layout = bgl,
         .entries = entries.ptr,
         .entryCount = @intCast(entries.len),
@@ -146,7 +146,7 @@ pub fn draw(
     defer c.wgpuBindGroupRelease(bg);
 
     // Encode Render Command
-    const enc = c.wgpuDeviceCreateCommandEncoder(glloc.device.device, null) orelse return error.Encoder;
+    const enc = c.wgpuDeviceCreateCommandEncoder(gloc.device.device, null) orelse return error.Encoder;
     defer c.wgpuCommandEncoderRelease(enc);
 
     const color_attachment = c.WGPURenderPassColorAttachment{
@@ -180,5 +180,5 @@ pub fn draw(
     const cmd = c.wgpuCommandEncoderFinish(enc, null);
     defer c.wgpuCommandBufferRelease(cmd);
 
-    c.wgpuQueueSubmit(glloc.device.queue, 1, &cmd);
+    c.wgpuQueueSubmit(gloc.device.queue, 1, &cmd);
 }

src/GpuTexture.zig

@@ -14,10 +14,10 @@ pub const GpuTextureDef = struct {
 };
 
 raw: c.WGPUTexture,
-glloc: GpuAllocator,
+gloc: GpuAllocator,
 def: GpuTextureDef,
 
-pub fn init(glloc: GpuAllocator, def: GpuTextureDef) !@This() {
+pub fn init(gloc: GpuAllocator, def: GpuTextureDef) !@This() {
     var use: u64 = 0;
     var iter = def.usage.iterator();
     while (iter.next()) |flag| use |= @intFromEnum(flag);
@@ -31,13 +31,13 @@ pub fn init(glloc: GpuAllocator, def: GpuTextureDef) !@This() {
         .mipLevelCount = 1,
         .sampleCount = 1,
     };
-    const raw = try glloc.allocTexture(desc);
+    const raw = try gloc.allocTexture(desc);
 
-    return .{ .glloc = glloc, .raw = raw, .def = def };
+    return .{ .gloc = gloc, .raw = raw, .def = def };
 }
 
 pub fn deinit(self: @This()) void {
-    self.glloc.freeTexture(self.raw);
+    self.gloc.freeTexture(self.raw);
 }
 
 pub fn getConstMappedRange(self: @This(), offset: u64, size: u64) ?*const anyopaque {
@@ -53,14 +53,14 @@ pub fn bytesSizeRow(self: @This()) u32 {
 }
 
 /// Return a GpuBuffer containing a copy of the texture.
-pub fn buffCopy(self: @This(), glloc: GpuAllocator) !GpuBuffer {
+pub fn buffCopy(self: @This(), gloc: GpuAllocator) !GpuBuffer {
-    const buf = try GpuBuffer.init(glloc, .{
+    const buf = try GpuBuffer.init(gloc, .{
         .size = self.bytesSize(),
         .usage = .initMany(&.{ .CopyDst, .CopySrc }),
         .label = "texture_copy_buffer",
     });
 
-    const enc = c.wgpuDeviceCreateCommandEncoder(glloc.device.device, null) orelse return error.Encoder;
+    const enc = c.wgpuDeviceCreateCommandEncoder(gloc.device.device, null) orelse return error.Encoder;
     defer c.wgpuCommandEncoderRelease(enc);
 
     const src_copy = c.WGPUTexelCopyTextureInfo{
@@ -82,7 +82,7 @@ pub fn buffCopy(self: @This(), glloc: GpuAllocator) !GpuBuffer {
 
     const cmd = c.wgpuCommandEncoderFinish(enc, null);
     defer c.wgpuCommandBufferRelease(cmd);
-    c.wgpuQueueSubmit(glloc.device.queue, 1, &cmd);
+    c.wgpuQueueSubmit(gloc.device.queue, 1, &cmd);
 
     return buf;
 }
@@ -110,7 +110,7 @@ pub fn load(
     const bytes = data.len * @sizeOf(T);
 
     c.wgpuQueueWriteTexture(
-        self.glloc.device.queue,
+        self.gloc.device.queue,
         &.{
             .texture = self.raw,
             .mipLevel = 0,
@@ -132,14 +132,14 @@ pub fn load(
 pub fn read(self: @This(), alloc: std.mem.Allocator, T: type) ![]T {
     const out = try alloc.alloc(T, @divExact(self.size, @sizeOf(T)));
 
-    const staging = try init(self.glloc, .{
+    const staging = try init(self.gloc, .{
         .size = self.size,
         .usage = .initMany(&.{ .MapRead, .CopyDst }),
         .label = "texture_read_staging",
     });
     defer staging.deinit();
 
-    const enc = c.wgpuDeviceCreateCommandEncoder(self.glloc.device.device, null) orelse return error.Encoder;
+    const enc = c.wgpuDeviceCreateCommandEncoder(self.gloc.device.device, null) orelse return error.Encoder;
     const src_copy = c.WGPUTexelCopyTextureInfo{
         .texture = self.raw,
         .mipLevel = 0,
@@ -158,7 +158,7 @@ pub fn read(self: @This(), alloc: std.mem.Allocator, T: type) ![]T {
     const cmd = c.wgpuCommandEncoderFinish(enc, null);
     defer c.wgpuCommandEncoderRelease(enc);
     defer c.wgpuCommandBufferRelease(cmd);
-    c.wgpuQueueSubmit(self.glloc.device.queue, 1, &cmd);
+    c.wgpuQueueSubmit(self.gloc.device.queue, 1, &cmd);
 
     var mapped = false;
     staging.mapAsync(
@@ -167,7 +167,7 @@ pub fn read(self: @This(), alloc: std.mem.Allocator, T: type) ![]T {
         self.size,
         .{ .callback = onMapped, .userdata1 = &mapped },
     );
-    while (!mapped) self.glloc.device.poll();
+    while (!mapped) self.gloc.device.poll();
 
     const ptr: [*]const T = @ptrCast(@alignCast(
         staging.getConstMappedRange(0, self.size),

src/GpuTextureView.zig

@@ -13,23 +13,23 @@ pub const GpuViewDef = struct {
 };
 
 raw: c.WGPUTextureView,
-glloc: GpuAllocator,
+gloc: GpuAllocator,
 
-pub fn init(glloc: GpuAllocator, texture: GpuTexture, def: GpuViewDef) !@This() {
+pub fn init(gloc: GpuAllocator, texture: GpuTexture, def: GpuViewDef) !@This() {
     var use: u64 = 0;
     var iter = def.usage.iterator();
     while (iter.next()) |flag| use |= @intFromEnum(flag);
 
-    const raw = try glloc.allocTextureView(texture.raw, .{
+    const raw = try gloc.allocTextureView(texture.raw, .{
         .label = svOpt(def.label),
         .format = @intFromEnum(def.format),
         .usage = use,
         .mipLevelCount = 1,
         .arrayLayerCount = 1,
     });
-    return .{ .glloc = glloc, .raw = raw };
+    return .{ .gloc = gloc, .raw = raw };
 }
 
 pub fn deinit(self: @This()) void {
-    self.glloc.freeTextureView(self.raw);
+    self.gloc.freeTextureView(self.raw);
 }