Update zon version

README.md

@@ -1,26 +1,28 @@
-# Minimal Zig WebGPU Compute Library
+# Minimal Zig WebGPU Compute & Render Library
 
-This is a minimal, self-contained Zig library designed to simplify running compute shaders using WebGPU.
-It abstracts away much of the boilerplate required for GPU device initialization, memory management, and pipeline execution.
+This is a minimal, self-contained Zig library designed to simplify running compute shaders and rendering pipelines using WebGPU. It abstracts away much of the boilerplate required for GPU device initialization, memory management, bind groups, and pipeline execution.
 
 ## Core Modules
 
-The library exports five primary components:
+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. By default, it enforces a 2 GB VRAM limit.
-* **`GpuArena` / `GpuAllocator`**: A memory management layer that tracks allocated VRAM bytes to prevent exceeding the device budget. The arena automatically destroys and releases all tracked WebGPU buffers when deinitialized.
-* **`GpuBuffer`**: Wraps native WebGPU buffers. It automatically aligns buffer sizes forward to a multiple of 4 bytes. It provides a `.load()` method for CPU-to-GPU data transfers (handling both aligned and unaligned lengths smoothly) 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. When running, it automatically splits the work into manageable chunks (up to 1 GB at a time) and dispatches workgroups of size 256.
+* **`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.
+* **`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.
 
-## Quick Start Example
+---
 
-Below is a complete, self-contained example demonstrating how to initialize the GPU, load data, run a compute shader, and read the results back to the CPU:
+## Example 1: Compute Pipeline
+
+Below is a complete example demonstrating how to initialize the GPU via the device allocator, manage VRAM using a GPU Arena, run a compute shader, and read the results back to the CPU:
 
 ```zig
 const std = @import("std");
 const gpu = @import("gpu");
 const GpuDevice = gpu.GpuDevice;
-const GpuArena = gpu.GpuArena;
+const GpuArenaAllocator = gpu.GpuArenaAllocator;
 const GpuBuffer = gpu.GpuBuffer;
 const GpuCompute = gpu.GpuCompute;
 
@@ -32,21 +34,23 @@ pub fn main(init: std.process.Init) !void {
     defer device.deinit();
 
     // 2. Create a GPU Arena to manage VRAM
-    var grena = GpuArena.init(allocator, device);
+    var grena = GpuArenaAllocator.init(allocator, device.gpuAllocator());
     defer grena.deinit();
-    const gloc = grena.gpuAllocator();
+    const glloc = grena.gpuAllocator();
 
     // 3. Load the WGSL compute pipeline
     const add_cp = try GpuCompute.init(
-        device,
+        glloc,
         @embedFile("shaders/add.wgsl"),
-        .{ .bindings = &.{
-            .{ .element_size = @sizeOf(f16) },
-            .{ .element_size = @sizeOf(f16) },
-            .{ .element_size = @sizeOf(f16) },
-        } },
+        .{
+            .label = "add",
+            .bindings = &.{
+                .{ .element_size = @sizeOf(f16) },
+                .{ .element_size = @sizeOf(f16) },
+                .{ .element_size = @sizeOf(f16) },
+            },
+        },
     );
-    defer add_cp.deinit();
 
     // 4. Setup CPU data
     const len: usize = 16;
@@ -62,19 +66,20 @@ 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(gloc, byte_size, .initMany(&.{ .Storage, .CopyDst, .CopySrc }));
-    const buf_b = try GpuBuffer.init(gloc, byte_size, .initMany(&.{ .Storage, .CopyDst, .CopySrc }));
-    const buf_out = try GpuBuffer.init(gloc, byte_size, .initMany(&.{ .Storage, .CopyDst, .CopySrc }));
+    const buf_a = try GpuBuffer.init(glloc, .{ .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_out = try GpuBuffer.init(glloc, .{ .label = "out", .size = byte_size, .usage = .initMany(&.{ .Storage, .CopyDst, .CopySrc }) });
 
-    // Note: The buffers are safely tied to the GpuArena which will automatically
+    // 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.
+    // This will also release pipelines, textures, ect. Everything using a GpuAllocator to init.
 
     // 6. Transfer data from CPU slices to GPU Buffers
     try buf_a.load(f16, data_a);
     try buf_b.load(f16, data_b);
 
     // 7. Dispatch the Compute
-    try add_cp.run(gloc, .{ buf_a, buf_b, buf_out });
+    try add_cp.run(glloc, .{ 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);
@@ -84,53 +89,145 @@ pub fn main(init: std.process.Init) !void {
 }
 ```
 
+---
+
+## Example 2: Rendering Pipeline (Offscreen to PPM Image)
+
+This example demonstrates how to initialize a rendering pipeline, allocate an output texture target, draw primitives via WebGPU,
+and pull the frame pixels back to the CPU to write a standard image file:
+
+```zig
+const std = @import("std");
+const gpu = @import("gpu");
+const GpuDevice = gpu.GpuDevice;
+const GpuArenaAllocator = gpu.GpuArenaAllocator;
+const GpuBuffer = gpu.GpuBuffer;
+const GpuRender = gpu.GpuRender;
+const GpuTexture = gpu.GpuTexture;
+const GpuTextureView = gpu.GpuTextureView;
+
+const width: u32 = 512;
+const height: u32 = 512;
+
+pub fn main(init: std.process.Init) !void {
+    const allocator = init.gpa;
+
+    // 1. Open GPU Device
+    const device = try GpuDevice.init(.{});
+    defer device.deinit();
+
+    // 2. Init VRAM Arena
+    var grena = GpuArenaAllocator.init(allocator, device.gpuAllocator());
+    defer grena.deinit();
+    const glloc = grena.gpuAllocator();
+
+    // 3. Load Render Pipeline
+    const circle_rp = try GpuRender.init(
+        glloc,
+        @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, .{
+        .format = .RGBA8Unorm,
+        .size = .{ .width = width, .height = height, .depthOrArrayLayers = 1 },
+        .usage = .initMany(&.{ .RenderAttachment, .CopySrc }),
+    });
+    defer texture.deinit();
+
+    // 5. Create a view from texture
+    const view = try GpuTextureView.init(glloc, texture, .{});
+    defer view.deinit();
+
+    // 6. Run the rendering pipeline
+    try circle_rp.draw(glloc, view, 4, .{});
+
+    // 7. Load Texture into GpuBuffer
+    const cpu_staging_cpu = try texture.buffCopy(glloc);
+    defer cpu_staging_cpu.deinit();
+
+    // 8. Read GpuBuffer to CPU
+    // This need to be free manually because CPU memory
+    const pixels = try cpu_staging_cpu.read(allocator, u8);
+    defer allocator.free(pixels);
+
+    // 9. Write a simple ppm image
+    try savePpm(init.io, "circle.ppm", width, height, pixels);
+}
+
+fn savePpm(io: std.Io, filename: []const u8, w: u32, h: u32, rgba_pixels: []const u8) !void {
+    const file = try std.Io.Dir.cwd().createFile(io, filename, .{});
+    defer file.close(io);
+
+    var buf: [255]u8 = undefined;
+    var writer = file.writer(io, &buf);
+
+    // PPM Header: P6 format means raw RGB bytes
+    try writer.interface.print("P6\n{d} {d}\n255\n", .{ w, h });
+
+    // Strip Alpha channel when writing out to standard RGB PPM format
+    var i: usize = 0;
+    while (i < rgba_pixels.len) : (i += 4) {
+        try writer.interface.writeAll(rgba_pixels[i .. i + 3]);
+    }
+}
+```
+
+---
+
+## Running Examples Locally
+
+If you have cloned the repository, you can run the included examples directly using the Zig build system:
+
+```bash
+# Run the rendering example (generates circle.ppm)
+zig build circle
+
+# Run the compute example
+zig build compute
+
+# Run the compute benchmark
+zig build bench_cp
+```
+
+---
+
 ## Dependencies
 
-* **`wgpu.h`**: The library relies on the WebGPU C API headers to bind to the native system graphics.
+* **`wgpu.h`**: The library relies on WebGPU C API headers to bind to the native system graphics.
 
 ## System Requirements
 
-Because this library binds to native system graphics APIs via `wgpu-native`,
-you must ensure the appropriate development headers and libraries are available on your system before compiling.
+Because this library binds to native system graphics APIs via `wgpu-native`, you must ensure the appropriate development headers and libraries are available on your system before compiling.
 
 ### Linux (Vulkan)
-You need the Vulkan development headers to compile the project, and a Vulkan-compatible driver to run it. 
 
-Depending on your distribution, install the following packages:
-* **Ubuntu / Debian:**
-  ```bash
-  sudo apt update
-  sudo apt install libvulkan-dev mesa-vulkan-drivers
-  ```
-* **Fedora / RHEL:**
-  ```bash
-  sudo dnf install vulkan-devel mesa-vulkan-drivers
-  ```
-* **Arch Linux:**
-  ```bash
-  sudo pacman -S vulkan-headers vulkan-icd-loader
-  ```
+* **Ubuntu / Debian:** `sudo apt update && sudo apt install libvulkan-dev mesa-vulkan-drivers`
+* **Fedora / RHEL:** `sudo dnf install vulkan-devel mesa-vulkan-drivers`
+* **Arch Linux:** `sudo pacman -S vulkan-headers vulkan-icd-loader`
 
 ### macOS (Metal)
-No extra installation is required.
-The build script automatically links against the standard Apple frameworks provided by the Xcode Command Line Tools
-(`Metal`, `QuartzCore`, `Foundation`, `CoreGraphics`).
+
+No extra installation required. Automatically links against `Metal`, `QuartzCore`, `Foundation`, and `CoreGraphics`.
 
 ### Windows (DirectX 12)
-No extra installation is required.
-The build script automatically links against the standard Windows SDK libraries (`d3d12`, `dxgi`, `user32`).
-Ensure you have the MSVC build tools installed.
+
+No extra installation required. Automatically links against `d3d12`, `dxgi`, and `user32`. Ensure you have MSVC build tools installed.
 
 ---
 
 ## Adding to your project
 
-To use this module in your own Zig project, add it to your `build.zig.zon`:
+Add it to your `build.zig.zon`:
+
 ```bash
-zig fetch --save git+https://git.bouvais.lu/adrien/zig-wgpu
+zig fetch --save git+https://git.bouvais.lu/adrien/zig-wgpu#ref=0.2.1
 ```
 
-Then, in your `build.zig`, import and add the module to your executable:
+Then, expose it in your `build.zig`:
+
 ```zig
 const zig_wgpu = b.dependency("zig-wgpu", .{
     .target = target,

build.zig

@@ -29,31 +29,33 @@ pub fn build(b: *std.Build) !void {
         mod.linkSystemLibrary("gcc_s", .{});
     }
 
-    var threaded: std.Io.Threaded = .init_single_threaded;
-    const io = threaded.io();
+    if (b.pkg_hash.len == 0) {
+        var threaded: std.Io.Threaded = .init_single_threaded;
+        const io = threaded.io();
 
-    var buf: [1024]u8 = undefined;
-    const exemples = try std.Io.Dir.cwd().openDir(io, "examples", .{ .access_sub_paths = false, .iterate = true });
-    var iter = exemples.iterate();
-    while (try iter.next(io)) |entry| {
-        if (entry.kind != .file) continue;
-        if (!std.mem.eql(u8, entry.name[entry.name.len - 4 ..], ".zig")) continue;
+        var buf: [1024]u8 = undefined;
+        const exemples = try std.Io.Dir.cwd().openDir(io, "examples", .{ .access_sub_paths = false, .iterate = true });
+        var iter = exemples.iterate();
+        while (try iter.next(io)) |entry| {
+            if (entry.kind != .file) continue;
+            if (!std.mem.eql(u8, entry.name[entry.name.len - 4 ..], ".zig")) continue;
 
-        const exe = b.addExecutable(.{
-            .name = entry.name[0 .. entry.name.len - 4],
-            .root_module = b.createModule(.{
-                .root_source_file = b.path(try std.fmt.bufPrint(&buf, "examples/{s}", .{entry.name})),
-                .target = target,
-                .optimize = optimize,
-                .imports = &.{},
-            }),
-        });
-        exe.root_module.addImport("gpu", mod);
+            const exe = b.addExecutable(.{
+                .name = entry.name[0 .. entry.name.len - 4],
+                .root_module = b.createModule(.{
+                    .root_source_file = b.path(try std.fmt.bufPrint(&buf, "examples/{s}", .{entry.name})),
+                    .target = target,
+                    .optimize = optimize,
+                    .imports = &.{},
+                }),
+            });
+            exe.root_module.addImport("gpu", mod);
 
-        b.installArtifact(exe);
+            b.installArtifact(exe);
 
-        const run_step = b.step(entry.name[0 .. entry.name.len - 4], try std.fmt.bufPrint(&buf, "Run {s} demo", .{entry.name}));
-        const run_cmd = b.addRunArtifact(exe);
-        run_step.dependOn(&run_cmd.step);
+            const run_step = b.step(entry.name[0 .. entry.name.len - 4], try std.fmt.bufPrint(&buf, "Run {s} demo", .{entry.name}));
+            const run_cmd = b.addRunArtifact(exe);
+            run_step.dependOn(&run_cmd.step);
+        }
     }
 }

build.zig.zon

@@ -1,6 +1,6 @@
 .{
     .name = .zig_wgpu,
-    .version = "0.1.0",
+    .version = "0.2.3",
     .fingerprint = 0x5d0e853acbc0c2c6,
     .minimum_zig_version = "0.16.0",
     .dependencies = .{},

examples/bench_cp.zig

@@ -6,26 +6,27 @@ const GpuAllocator = gpu.GpuAllocator;
 const GpuBuffer = gpu.GpuBuffer;
 const GpuCompute = gpu.GpuCompute;
 
+pub const std_options = std.Options{ .log_level = .info };
+
 /// Minimal implementation of a f16 Vector
 const Vec = struct {
     buf: GpuBuffer,
     len: usize,
 
-    // Changed: gloc is passed by value (const)
-    pub fn initZero(gloc: GpuAllocator, len: usize) !Vec {
+    // Changed: glloc is passed by value (const)
+    pub fn initZero(glloc: GpuAllocator, len: usize) !Vec {
         return .{
-            .buf = try GpuBuffer.init(
-                gloc,
-                len * @sizeOf(f16),
-                .initMany(&.{ .Storage, .CopyDst, .CopySrc }),
-            ),
+            .buf = try GpuBuffer.init(glloc, .{
+                .size = len * @sizeOf(f16),
+                .usage = .initMany(&.{ .Storage, .CopyDst, .CopySrc }),
+            }),
             .len = len,
         };
     }
 
-    // Changed: gloc is passed by value
-    pub fn initLoad(gloc: GpuAllocator, data: []const f16) !Vec {
-        var self = try initZero(gloc, data.len);
+    // Changed: glloc is passed by value
+    pub fn initLoad(glloc: GpuAllocator, data: []const f16) !Vec {
+        var self = try initZero(glloc, data.len);
         try self.load(data); // Direct access via the interface copy
         return self;
     }
@@ -39,18 +40,18 @@ const Vec = struct {
         try self.buf.load(f16, data);
     }
 
-    // Changed: gloc is passed by value instead of *GpuAllocator
-    pub fn run(self: Vec, gloc: GpuAllocator, other: Vec, process: GpuCompute) !Vec {
+    // Changed: glloc is passed by value instead of *GpuAllocator
+    pub fn run(self: Vec, glloc: GpuAllocator, other: Vec, process: GpuCompute) !Vec {
         std.debug.assert(self.len == other.len);
 
-        const result = try Vec.initZero(gloc, self.len);
+        const result = try Vec.initZero(glloc, self.len);
         errdefer result.deinit();
 
-        try process.run(gloc, .{ self.buf, other.buf, result.buf });
+        try process.run(glloc, .{ self.buf, other.buf, result.buf });
         return result;
     }
 
-    // Changed: gloc is passed by value instead of *GpuAllocator
+    // Changed: glloc is passed by value instead of *GpuAllocator
     pub fn read(self: Vec, alloc: std.mem.Allocator) ![]f16 {
         return self.buf.read(alloc, f16);
     }
@@ -62,9 +63,9 @@ pub fn main(init: std.process.Init) !void {
 
     var grena = GpuArenaAllocator.init(init.gpa, device.gpuAllocator());
     defer grena.deinit();
-    const gloc = grena.gpuAllocator();
+    const glloc = grena.gpuAllocator();
 
-    const add_pip = try GpuCompute.init(gloc, @embedFile("shaders/add.wgsl"), .{ .bindings = &.{
+    const add_pip = try GpuCompute.init(glloc, @embedFile("shaders/add.wgsl"), .{ .bindings = &.{
         .{ .element_size = @sizeOf(f16) },
         .{ .element_size = @sizeOf(f16) },
         .{ .element_size = @sizeOf(f16) },
@@ -119,9 +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(gloc, data_a);
+            const a = try Vec.initLoad(glloc, data_a);
             defer a.deinit();
-            const b = try Vec.initLoad(gloc, data_b);
+            const b = try Vec.initLoad(glloc, data_b);
             defer b.deinit();
 
             const alloc_duration = alloc_start.durationTo(std.Io.Clock.awake.now(init.io));
@@ -131,7 +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(gloc, b, add_pip);
+            const sum = try a.run(glloc, b, add_pip);
             defer sum.deinit();
 
             // All 3 buffers (a, b, sum) are currently resident in VRAM here.

examples/circle.zig

@@ -18,18 +18,20 @@ pub fn main(init: std.process.Init) !void {
     defer device.deinit();
 
     // 2. Init VRAM Arena
-    const gloc = device.gpuAllocator();
+    var grena = GpuArenaAllocator.init(allocator, device.gpuAllocator());
+    defer grena.deinit();
+    const glloc = grena.gpuAllocator();
 
     // 3. Load Render Pipeline
     const circle_rp = try GpuRender.init(
-        gloc,
+        glloc,
         @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(gloc, .{
+    const texture = try GpuTexture.init(glloc, .{
         .format = .RGBA8Unorm,
         .size = .{ .width = width, .height = height, .depthOrArrayLayers = 1 },
         .usage = .initMany(&.{ .RenderAttachment, .CopySrc }),
@@ -37,14 +39,14 @@ pub fn main(init: std.process.Init) !void {
     defer texture.deinit();
 
     // 5. Create a view from texture
-    const view = try GpuTextureView.init(gloc, texture, .{});
+    const view = try GpuTextureView.init(glloc, texture, .{});
     defer view.deinit();
 
     // 6. Run the rendering pipeline
-    try circle_rp.draw(gloc, view, 4, .{});
+    try circle_rp.draw(glloc, view, 4, .{});
 
     // 7. Load Texture into GpuBuffer
-    const cpu_staging_cpu = try texture.buffCopy(gloc);
+    const cpu_staging_cpu = try texture.buffCopy(glloc);
     defer cpu_staging_cpu.deinit();
 
     // 8. Read GpuBuffer to CPU

examples/compute.zig

@@ -15,17 +15,20 @@ 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 gloc = grena.gpuAllocator();
+    const glloc = grena.gpuAllocator();
 
     // 3. Load the WGSL compute pipeline
     const add_cp = try GpuCompute.init(
-        gloc,
+        glloc,
         @embedFile("shaders/add.wgsl"),
-        .{ .bindings = &.{
-            .{ .element_size = @sizeOf(f16) },
-            .{ .element_size = @sizeOf(f16) },
-            .{ .element_size = @sizeOf(f16) },
-        } },
+        .{
+            .label = "add",
+            .bindings = &.{
+                .{ .element_size = @sizeOf(f16) },
+                .{ .element_size = @sizeOf(f16) },
+                .{ .element_size = @sizeOf(f16) },
+            },
+        },
     );
 
     // 4. Setup CPU data
@@ -42,9 +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(gloc, byte_size, .initMany(&.{ .Storage, .CopyDst, .CopySrc }));
-    const buf_b = try GpuBuffer.init(gloc, byte_size, .initMany(&.{ .Storage, .CopyDst, .CopySrc }));
-    const buf_out = try GpuBuffer.init(gloc, byte_size, .initMany(&.{ .Storage, .CopyDst, .CopySrc }));
+    const buf_a = try GpuBuffer.init(glloc, .{ .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_out = try GpuBuffer.init(glloc, .{ .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.
@@ -55,7 +58,7 @@ pub fn main(init: std.process.Init) !void {
     try buf_b.load(f16, data_b);
 
     // 7. Dispatch the Compute
-    try add_cp.run(gloc, .{ buf_a, buf_b, buf_out });
+    try add_cp.run(glloc, .{ 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);

src/GpuArenaAllocator.zig

@@ -3,6 +3,7 @@ const GpuDevice = @import("GpuDevice.zig");
 const GpuAllocator = @import("GpuAllocator.zig");
 const GpuTextureFormat = @import("lib.zig").GpuTextureFormat;
 const c = @import("utils.zig").c;
+const viewStr = @import("utils.zig").viewStr;
 
 child_allocator: GpuAllocator, // I use Zig naming child_allocator, but that should be a parent for me. Likely something idk
 tracked_buffers: std.AutoHashMap(c.WGPUBuffer, c.WGPUBufferDescriptor),
@@ -24,30 +25,40 @@ pub fn init(cpu_allocator: std.mem.Allocator, child_allocator: GpuAllocator) @Th
 }
 
 pub fn deinit(self: *@This()) void {
+    std.log.debug("Freeing GpuArenaAllocator (Used VRAM: {}/{} MB)", .{
+        self.allocated_vram_bytes / 1024 / 1024,
+        self.child_allocator.device.def.vram_bytes_limit / 1024 / 1024,
+    });
+
     var it_buffer = self.tracked_buffers.keyIterator();
     while (it_buffer.next()) |buf_ptr|
-        self.child_allocator.freeBuffer(buf_ptr.*);
+        freeBuffer(self, buf_ptr.*);
     self.tracked_buffers.deinit();
 
     var it_tex = self.tracked_textures.keyIterator();
     while (it_tex.next()) |buf_ptr|
-        self.child_allocator.freeTexture(buf_ptr.*);
+        freeTexture(self, buf_ptr.*);
     self.tracked_textures.deinit();
 
     var it_view = self.tracked_views.keyIterator();
     while (it_view.next()) |buf_ptr|
-        self.child_allocator.freeTextureView(buf_ptr.*);
+        freeTextureView(self, buf_ptr.*);
     self.tracked_views.deinit();
 
     var it_render = self.tracked_renders.keyIterator();
     while (it_render.next()) |buf_ptr|
-        self.child_allocator.freeRenderPipeline(buf_ptr.*);
+        freeRenderPipeline(self, buf_ptr.*);
     self.tracked_renders.deinit();
 
     var it_compute = self.tracked_computes.keyIterator();
     while (it_compute.next()) |buf_ptr|
-        self.child_allocator.freeComputePipeline(buf_ptr.*);
+        freeComputePipeline(self, buf_ptr.*);
     self.tracked_computes.deinit();
+
+    std.log.debug("Freed GpuArenaAllocator (Used VRAM: {}/{} MB)", .{
+        self.allocated_vram_bytes / 1024 / 1024,
+        self.child_allocator.device.def.vram_bytes_limit / 1024 / 1024,
+    });
 }
 
 /// Returns the type-erased immutable interface wrapper
@@ -79,6 +90,14 @@ fn allocBuffer(ctx: *anyopaque, desc: c.WGPUBufferDescriptor) anyerror!c.WGPUBuf
     const raw = try self.child_allocator.allocBuffer(desc);
     self.tracked_buffers.putAssumeCapacity(raw, desc);
     self.allocated_vram_bytes += desc.size;
+
+    std.log.debug("Allocated Buffer '{s}': {d} B (Total VRAM: {}/{} MB)", .{
+        viewStr(desc.label),
+        desc.size,
+        self.allocated_vram_bytes / 1024 / 1024,
+        self.child_allocator.device.def.vram_bytes_limit / 1024 / 1024,
+    });
+
     return raw;
 }
 
@@ -87,6 +106,12 @@ fn freeBuffer(ctx: *anyopaque, raw: c.WGPUBuffer) void {
     if (self.tracked_buffers.fetchRemove(raw)) |kv| {
         self.child_allocator.freeBuffer(raw);
         self.allocated_vram_bytes -= kv.value.size;
+
+        std.log.debug("Freed Buffer '{s}' (Total VRAM: {}/{} MB)", .{
+            viewStr(kv.value.label),
+            self.allocated_vram_bytes / 1024 / 1024,
+            self.child_allocator.device.def.vram_bytes_limit / 1024 / 1024,
+        });
     }
 }
 
@@ -97,13 +122,21 @@ fn allocTexture(ctx: *anyopaque, desc: c.WGPUTextureDescriptor) anyerror!c.WGPUT
     const format: GpuTextureFormat = @enumFromInt(desc.format);
     const bytes_size = desc.size.width * desc.size.height * format.bytesPerPixel();
 
-    if (bytes_size + self.allocated_vram_bytes > self.child_allocator.device.config.vram_bytes_limit)
+    if (bytes_size + self.allocated_vram_bytes > self.child_allocator.device.def.vram_bytes_limit)
         return error.ExceedsVramBudget;
 
     const raw = try self.child_allocator.allocTexture(desc);
 
     self.tracked_textures.putAssumeCapacity(raw, desc);
     self.allocated_vram_bytes += bytes_size;
+
+    std.log.debug("Allocated Texture '{s}': {d} B (Total VRAM: {}/{} MB)", .{
+        viewStr(desc.label),
+        bytes_size,
+        self.allocated_vram_bytes / 1024 / 1024,
+        self.child_allocator.device.def.vram_bytes_limit / 1024 / 1024,
+    });
+
     return raw;
 }
 
@@ -117,6 +150,12 @@ fn freeTexture(ctx: *anyopaque, raw: c.WGPUTexture) void {
         const format: GpuTextureFormat = @enumFromInt(desc.format);
         const bytes_size = desc.size.width * desc.size.height * format.bytesPerPixel();
         self.allocated_vram_bytes -= bytes_size;
+
+        std.log.debug("Freed Texture '{s}' (Total VRAM: {}/{} MB)", .{
+            viewStr(desc.label),
+            self.allocated_vram_bytes / 1024 / 1024,
+            self.child_allocator.device.def.vram_bytes_limit / 1024 / 1024,
+        });
     }
 }
 
@@ -125,13 +164,17 @@ fn allocTextureView(ctx: *anyopaque, texture: c.WGPUTexture, desc: c.WGPUTexture
     try self.tracked_views.ensureTotalCapacity(self.tracked_views.count() + 1);
     const raw = try self.child_allocator.allocTextureView(texture, desc);
     self.tracked_views.putAssumeCapacity(raw, desc);
+    std.log.debug("Allocated Texture View '{s}'", .{viewStr(desc.label)});
     return raw;
 }
 
 fn freeTextureView(ctx: *anyopaque, raw: c.WGPUTextureView) void {
     const self: *@This() = @ptrCast(@alignCast(ctx));
-    if (self.tracked_views.remove(raw))
+    if (self.tracked_views.fetchRemove(raw)) |kv| {
         self.child_allocator.freeTextureView(raw);
+        const desc = kv.value;
+        std.log.debug("Freed Texture View '{s}'", .{viewStr(desc.label)});
+    }
 }
 
 fn allocRenderPipeline(ctx: *anyopaque, desc: c.WGPURenderPipelineDescriptor) anyerror!c.WGPURenderPipeline {
@@ -139,13 +182,17 @@ fn allocRenderPipeline(ctx: *anyopaque, desc: c.WGPURenderPipelineDescriptor) an
     try self.tracked_renders.ensureTotalCapacity(self.tracked_renders.count() + 1);
     const raw = try self.child_allocator.allocRenderPipeline(desc);
     self.tracked_renders.putAssumeCapacity(raw, desc);
+    std.log.debug("Allocated Render Pipeline '{s}'", .{viewStr(desc.label)});
     return raw;
 }
 
 fn freeRenderPipeline(ctx: *anyopaque, raw: c.WGPURenderPipeline) void {
     const self: *@This() = @ptrCast(@alignCast(ctx));
-    if (self.tracked_renders.remove(raw))
+    if (self.tracked_renders.fetchRemove(raw)) |kv| {
         self.child_allocator.freeRenderPipeline(raw);
+        const desc = kv.value;
+        std.log.debug("Freed Render Pipeline '{s}'", .{viewStr(desc.label)});
+    }
 }
 
 fn allocComputePipeline(ctx: *anyopaque, desc: c.WGPUComputePipelineDescriptor) anyerror!c.WGPUComputePipeline {
@@ -153,11 +200,15 @@ fn allocComputePipeline(ctx: *anyopaque, desc: c.WGPUComputePipelineDescriptor)
     try self.tracked_computes.ensureTotalCapacity(self.tracked_computes.count() + 1);
     const raw = try self.child_allocator.allocComputePipeline(desc);
     self.tracked_computes.putAssumeCapacity(raw, desc);
+    std.log.debug("Allocated Compute Pipeline '{s}'", .{viewStr(desc.label)});
     return raw;
 }
 
 fn freeComputePipeline(ctx: *anyopaque, raw: c.WGPUComputePipeline) void {
     const self: *@This() = @ptrCast(@alignCast(ctx));
-    if (self.tracked_computes.remove(raw))
+    if (self.tracked_computes.fetchRemove(raw)) |kv| {
         self.child_allocator.freeComputePipeline(raw);
+        const desc = kv.value;
+        std.log.debug("Freed Compute Pipeline '{s}'", .{viewStr(desc.label)});
+    }
 }

src/GpuBuffer.zig

@@ -1,11 +1,12 @@
 const std = @import("std");
 const c = @import("utils.zig").c;
 const GpuAllocator = @import("GpuAllocator.zig");
+const svOpt = @import("utils.zig").svOpt;
 
 raw: c.WGPUBuffer,
 size: u64,
 usage: c.WGPUBufferUsage,
-gloc: GpuAllocator,
+glloc: GpuAllocator,
 
 const BufferUsage = enum(u64) {
     None = 0x0000000000000000,
@@ -21,25 +22,35 @@ const BufferUsage = enum(u64) {
     QueryResolve = 0x0000000000000200,
 };
 
-pub fn init(gloc: GpuAllocator, size: u64, usage: std.EnumSet(BufferUsage)) !@This() {
+const GpuBufferDef = struct {
+    label: ?[]const u8 = null,
+    size: u64,
+    usage: std.EnumSet(BufferUsage),
+};
+
+pub fn init(glloc: GpuAllocator, def: GpuBufferDef) !@This() {
     var use: u64 = 0;
-    var iter = usage.iterator();
+    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, size, 4);
+    const aligned_size = std.mem.alignForward(u64, def.size, 4);
 
-    const raw_handle = try gloc.allocBuffer(.{ .size = aligned_size, .usage = use });
+    const raw_handle = try glloc.allocBuffer(.{
+        .size = aligned_size,
+        .usage = use,
+        .label = svOpt(def.label),
+    });
     return .{
         .raw = raw_handle,
         .size = aligned_size,
         .usage = use,
-        .gloc = gloc,
+        .glloc = glloc,
     };
 }
 
 pub fn deinit(self: @This()) void {
-    self.gloc.freeBuffer(self.raw);
+    self.glloc.freeBuffer(self.raw);
 }
 
 pub fn getConstMappedRange(self: @This(), offset: u64, size: u64) ?*const anyopaque {
@@ -70,20 +81,20 @@ pub fn load(
 
     if (bytes == self.size) {
         // Aligned path: direct download
-        c.wgpuQueueWriteBuffer(self.gloc.device.queue, self.raw, 0, data.ptr, self.size);
+        c.wgpuQueueWriteBuffer(self.glloc.device.queue, self.raw, 0, data.ptr, self.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.gloc.device.queue, self.raw, 0, data.ptr, aligned_part);
+            c.wgpuQueueWriteBuffer(self.glloc.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.gloc.device.queue, self.raw, aligned_part, &remainder_buf, 4);
+        c.wgpuQueueWriteBuffer(self.glloc.device.queue, self.raw, aligned_part, &remainder_buf, 4);
     }
 }
 
@@ -91,19 +102,19 @@ 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.gloc,
-        self.size,
-        .initMany(&.{ .MapRead, .CopyDst }),
-    );
+    const staging = try init(self.glloc, .{
+        .size = self.size,
+        .usage = .initMany(&.{ .MapRead, .CopyDst }),
+        .label = "staging_read_buffer",
+    });
     defer staging.deinit();
 
-    const enc = c.wgpuDeviceCreateCommandEncoder(self.gloc.device.device, null) orelse return error.Encoder;
+    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.gloc.device.queue, 1, &cmd);
+    c.wgpuQueueSubmit(self.glloc.device.queue, 1, &cmd);
 
     var mapped = false;
     staging.mapAsync(
@@ -112,7 +123,7 @@ pub fn read(self: @This(), alloc: std.mem.Allocator, T: type) ![]T {
         self.size,
         .{ .callback = onMapped, .userdata1 = &mapped },
     );
-    while (!mapped) self.gloc.device.poll();
+    while (!mapped) self.glloc.device.poll();
 
     const ptr: [*]const T = @ptrCast(@alignCast(
         staging.getConstMappedRange(0, self.size),

src/GpuCompute.zig

@@ -1,5 +1,6 @@
 const c = @import("utils.zig").c;
 const sv = @import("utils.zig").sv;
+const svOpt = @import("utils.zig").svOpt;
 const GpuAllocator = @import("GpuAllocator.zig");
 const GpuBuffer = @import("GpuBuffer.zig");
 const GpuDevice = @import("GpuDevice.zig");
@@ -11,6 +12,7 @@ pub const Binding = struct {
 };
 
 pub const ComputeDef = struct {
+    label: ?[]const u8 = null,
     bindings: []const Binding,
     workgroup_size: u32 = 256,
     max_workgroups: u32 = 65535,
@@ -20,37 +22,40 @@ pub const ComputeDef = struct {
 };
 
 pip: c.WGPUComputePipeline,
-gloc: GpuAllocator,
+glloc: GpuAllocator,
 def: ComputeDef,
 
-pub fn init(gloc: GpuAllocator, wgsl: []const u8, def: ComputeDef) !@This() {
+pub fn init(glloc: GpuAllocator, wgsl: []const u8, def: ComputeDef) !@This() {
     var wgsl_src = c.WGPUShaderSourceWGSL{
         .chain = .{ .sType = c.WGPUSType_ShaderSourceWGSL },
         .code = sv(wgsl),
     };
-    const shader = c.wgpuDeviceCreateShaderModule(gloc.device.device, &.{
+    const shader = c.wgpuDeviceCreateShaderModule(glloc.device.device, &.{
         .nextInChain = @ptrCast(&wgsl_src),
     }) orelse return error.Shader;
     defer c.wgpuShaderModuleRelease(shader);
 
-    const pip = try gloc.allocComputePipeline(.{ .compute = .{ .module = shader, .entryPoint = sv("main") } });
+    const pip = try glloc.allocComputePipeline(.{
+        .label = svOpt(def.label),
+        .compute = .{ .module = shader, .entryPoint = sv("main") },
+    });
 
     return .{
-        .gloc = gloc,
+        .glloc = glloc,
         .pip = pip,
         .def = def,
     };
 }
 
 pub fn deinit(self: @This()) void {
-    self.gloc.freeComputePipeline(self.pip);
+    self.glloc.freeComputePipeline(self.pip);
 }
 
 /// Execute the compute pass with arbitrary buffer bindings via a tuple.
-/// Example: `try proc.run(gloc, .{ buf_a, buf_b, buf_out });`
+/// Example: `try proc.run(glloc, .{ buf_a, buf_b, buf_out });`
 pub fn run(
     self: @This(),
-    gloc: GpuAllocator,
+    glloc: GpuAllocator,
     args: anytype,
 ) !void {
     const type_info = @typeInfo(@TypeOf(args));
@@ -108,12 +113,12 @@ pub fn run(
     defer if (info_buf) |b| b.deinit();
 
     if (self.def.append_info_buffer) {
-        info_buf = try GpuBuffer.init(
-            gloc,
-            @sizeOf(u32),
-            .initMany(&.{ .Uniform, .CopyDst }),
-        );
-        c.wgpuQueueWriteBuffer(gloc.device.queue, info_buf.?.raw, 0, &elements_count, @sizeOf(u32));
+        info_buf = try GpuBuffer.init(glloc, .{
+            .size = @sizeOf(u32),
+            .usage = .initMany(&.{ .Uniform, .CopyDst }),
+            .label = "compute_info_buffer",
+        });
+        c.wgpuQueueWriteBuffer(glloc.device.queue, info_buf.?.raw, 0, &elements_count, @sizeOf(u32));
 
         entries_buf[entry_count] = .{
             .binding = @intCast(entry_count),
@@ -125,11 +130,11 @@ pub fn run(
     }
 
     const entries = entries_buf[0..entry_count];
-    try submitPass(gloc, self.pip, entries, elements_count, self.def.workgroup_size, self.def.max_workgroups);
+    try submitPass(glloc, self.pip, entries, elements_count, self.def.workgroup_size, self.def.max_workgroups);
 }
 
 fn submitPass(
-    gloc: GpuAllocator,
+    glloc: GpuAllocator,
     pipeline: c.WGPUComputePipeline,
     entries: []const c.WGPUBindGroupEntry,
     n: usize,
@@ -141,14 +146,14 @@ fn submitPass(
     const bgl = c.wgpuComputePipelineGetBindGroupLayout(pipeline, 0);
     defer c.wgpuBindGroupLayoutRelease(bgl);
 
-    const bg = c.wgpuDeviceCreateBindGroup(gloc.device.device, &.{
+    const bg = c.wgpuDeviceCreateBindGroup(glloc.device.device, &.{
         .layout = bgl,
         .entries = entries.ptr,
         .entryCount = entries.len,
     }) orelse return error.BindGroup;
     defer c.wgpuBindGroupRelease(bg);
 
-    const enc = c.wgpuDeviceCreateCommandEncoder(gloc.device.device, null) orelse return error.Encoder;
+    const enc = c.wgpuDeviceCreateCommandEncoder(glloc.device.device, null) orelse return error.Encoder;
     const pass = c.wgpuCommandEncoderBeginComputePass(enc, null);
     c.wgpuComputePassEncoderSetPipeline(pass, pipeline);
     c.wgpuComputePassEncoderSetBindGroup(pass, 0, bg, 0, null);
@@ -163,7 +168,7 @@ fn submitPass(
     const cmd = c.wgpuCommandEncoderFinish(enc, null);
     defer c.wgpuCommandEncoderRelease(enc);
     defer c.wgpuCommandBufferRelease(cmd);
-    c.wgpuQueueSubmit(gloc.device.queue, 1, &cmd);
+    c.wgpuQueueSubmit(glloc.device.queue, 1, &cmd);
 }
 
 fn ceilDiv(n: usize, d: usize) usize {

src/GpuDevice.zig

@@ -1,20 +1,17 @@
 const std = @import("std");
 const c = @import("utils.zig").c;
 const sv = @import("utils.zig").sv;
+const svOpt = @import("utils.zig").svOpt;
 const GpuAllocator = @import("GpuAllocator.zig");
 const GpuTextureFormat = @import("lib.zig").GpuTextureFormat;
 
-// TODO: Make Allocator more zig like
-//  - GpuDevice can return a GpuAllocator that just allocate and nothing else
-//  - From this GpuAllocator, can create a GpuArena like std.heap.ArenaAllocator.init(allocator)
-//  - Rename GpuArenaAllocator too
-
 const Ctx = struct {
     adapter: c.WGPUAdapter = null,
     device: c.WGPUDevice = null,
 };
 
-const GpuDeviceConfig = struct {
+const GpuDeviceDef = struct {
+    label: ?[]const u8 = null,
     /// VRAM limit. Default 2 GB
     vram_bytes_limit: u64 = 2 * 1024 * 1024 * 1024,
     power_preference: enum(c_uint) {
@@ -31,9 +28,9 @@ device: c.WGPUDevice,
 queue: c.WGPUQueue,
 limits: c.WGPULimits,
 
-config: GpuDeviceConfig,
+def: GpuDeviceDef,
 
-pub fn init(config: GpuDeviceConfig) !@This() {
+pub fn init(def: GpuDeviceDef) !@This() {
     const instance = c.wgpuCreateInstance(
         &std.mem.zeroes(c.WGPUInstanceDescriptor),
     ) orelse return error.NoInstance;
@@ -42,13 +39,25 @@ pub fn init(config: GpuDeviceConfig) !@This() {
     var ctx = Ctx{};
     _ = c.wgpuInstanceRequestAdapter(
         instance,
-        &.{ .powerPreference = @intFromEnum(config.power_preference) },
+        &.{ .powerPreference = @intFromEnum(def.power_preference) },
         .{ .callback = onAdapter, .userdata1 = &ctx },
     );
     c.wgpuInstanceProcessEvents(instance);
     const adapter = ctx.adapter orelse return error.NoAdapter;
     errdefer c.wgpuAdapterRelease(adapter);
 
+    var adapter_info = std.mem.zeroes(c.WGPUAdapterInfo);
+    _ = c.wgpuAdapterGetInfo(adapter, &adapter_info);
+
+    std.log.info("=== WebGPU Device Initialized ===", .{});
+    if (adapter_info.device.length > 0 and adapter_info.device.data != null) {
+        std.log.info("  Device Name  : {s}", .{adapter_info.device.data[0..adapter_info.device.length]});
+    }
+    if (adapter_info.architecture.length > 0 and adapter_info.architecture.data != null) {
+        std.log.info("  Architecture : {s}", .{adapter_info.architecture.data[0..adapter_info.architecture.length]});
+    }
+    std.log.info("  Backend Type : {d}", .{adapter_info.backendType});
+
     var supported_features = std.mem.zeroes(c.WGPUSupportedFeatures);
     c.wgpuAdapterGetFeatures(adapter, &supported_features);
 
@@ -56,6 +65,11 @@ pub fn init(config: GpuDeviceConfig) !@This() {
     supported_limits.nextInChain = null;
     if (c.wgpuAdapterGetLimits(adapter, &supported_limits) != 1) return error.FailedToGetAdapterLimits;
 
+    std.log.info("  Max Buf Size : {d} MB", .{supported_limits.maxBufferSize / 1024 / 1024});
+    std.log.info("  Max Storage  : {d} MB", .{supported_limits.maxStorageBufferBindingSize / 1024 / 1024});
+    std.log.info("  Max Workgroup: X: {d}, Y: {d}, Z: {d}", .{ supported_limits.maxComputeWorkgroupSizeX, supported_limits.maxComputeWorkgroupSizeY, supported_limits.maxComputeWorkgroupSizeZ });
+    std.log.info("  VRAM Budget  : {d} MB", .{def.vram_bytes_limit / 1024 / 1024});
+
     var has_f16 = false;
     for (0..supported_features.featureCount) |i| {
         if (supported_features.features[i] == c.WGPUFeatureName_ShaderF16) {
@@ -63,6 +77,8 @@ pub fn init(config: GpuDeviceConfig) !@This() {
             break;
         }
     }
+    std.log.info("  Shader F16   : {}", .{has_f16});
+    std.log.info("=================================", .{});
 
     var feature_buf = [_]c.WGPUFeatureName{c.WGPUFeatureName_ShaderF16};
     const required_features: []const c.WGPUFeatureName =
@@ -70,7 +86,7 @@ pub fn init(config: GpuDeviceConfig) !@This() {
 
     const device_descriptor = c.WGPUDeviceDescriptor{
         .nextInChain = null,
-        .label = sv("TensorCompilerDevice"),
+        .label = svOpt(def.label),
         .requiredFeatureCount = required_features.len,
         .requiredFeatures = if (required_features.len > 0) required_features.ptr else null,
         .requiredLimits = &supported_limits,
@@ -89,7 +105,7 @@ pub fn init(config: GpuDeviceConfig) !@This() {
         .device = device,
         .queue = c.wgpuDeviceGetQueue(device),
         .limits = supported_limits,
-        .config = config,
+        .def = def,
     };
 }
 

src/GpuRender.zig

@@ -1,6 +1,7 @@
 const std = @import("std");
 const c = @import("utils.zig").c;
 const sv = @import("utils.zig").sv;
+const svOpt = @import("utils.zig").svOpt;
 const GpuAllocator = @import("GpuAllocator.zig");
 const GpuBuffer = @import("GpuBuffer.zig");
 const GpuDevice = @import("GpuDevice.zig");
@@ -12,6 +13,7 @@ pub const Binding = struct {
 };
 
 pub const GpuRenderDef = struct {
+    label: ?[]const u8 = null,
     bindings: []const Binding = &.{},
     /// The surface texture format we are rendering to (e.g., BGRA8Unorm)
     texture_format: GpuTextureFormat,
@@ -32,16 +34,16 @@ const GpuPrimitiveTopology = enum(c_uint) {
     Force32 = 0x7FFFFFFF,
 };
 
-gloc: GpuAllocator,
+glloc: GpuAllocator,
 pip: c.WGPURenderPipeline,
 def: GpuRenderDef,
 
-pub fn init(gloc: GpuAllocator, wgsl: []const u8, def: GpuRenderDef) !@This() {
+pub fn init(glloc: GpuAllocator, wgsl: []const u8, def: GpuRenderDef) !@This() {
     var wgsl_src = c.WGPUShaderSourceWGSL{
         .chain = .{ .sType = c.WGPUSType_ShaderSourceWGSL },
         .code = sv(wgsl),
     };
-    const shader = c.wgpuDeviceCreateShaderModule(gloc.device.device, &.{
+    const shader = c.wgpuDeviceCreateShaderModule(glloc.device.device, &.{
         .nextInChain = @ptrCast(&wgsl_src),
     }) orelse return error.Shader;
     defer c.wgpuShaderModuleRelease(shader);
@@ -67,7 +69,8 @@ pub fn init(gloc: GpuAllocator, wgsl: []const u8, def: GpuRenderDef) !@This() {
     };
 
     // 3. Compile the Complete Render Pipeline
-    const pip = try gloc.allocRenderPipeline(.{
+    const pip = try glloc.allocRenderPipeline(.{
+        .label = svOpt(def.label),
         .vertex = .{
             .module = shader,
             .entryPoint = sv(def.vertex_entry),
@@ -87,21 +90,21 @@ pub fn init(gloc: GpuAllocator, wgsl: []const u8, def: GpuRenderDef) !@This() {
     });
 
     return .{
-        .gloc = gloc,
+        .glloc = glloc,
         .pip = pip,
         .def = def,
     };
 }
 
 pub fn deinit(self: @This()) void {
-    self.gloc.freeRenderPipeline(self.pip);
+    self.glloc.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(),
-    gloc: GpuAllocator,
+    glloc: GpuAllocator,
     target_view: GpuTextureView,
     vertex_count: u32,
     args: anytype,
@@ -135,7 +138,7 @@ pub fn draw(
     const bgl = c.wgpuRenderPipelineGetBindGroupLayout(self.pip, 0);
     defer c.wgpuBindGroupLayoutRelease(bgl);
 
-    const bg = c.wgpuDeviceCreateBindGroup(gloc.device.device, &.{
+    const bg = c.wgpuDeviceCreateBindGroup(glloc.device.device, &.{
         .layout = bgl,
         .entries = entries.ptr,
         .entryCount = @intCast(entries.len),
@@ -143,7 +146,7 @@ pub fn draw(
     defer c.wgpuBindGroupRelease(bg);
 
     // Encode Render Command
-    const enc = c.wgpuDeviceCreateCommandEncoder(gloc.device.device, null) orelse return error.Encoder;
+    const enc = c.wgpuDeviceCreateCommandEncoder(glloc.device.device, null) orelse return error.Encoder;
     defer c.wgpuCommandEncoderRelease(enc);
 
     const color_attachment = c.WGPURenderPassColorAttachment{
@@ -177,5 +180,5 @@ pub fn draw(
     const cmd = c.wgpuCommandEncoderFinish(enc, null);
     defer c.wgpuCommandBufferRelease(cmd);
 
-    c.wgpuQueueSubmit(gloc.device.queue, 1, &cmd);
+    c.wgpuQueueSubmit(glloc.device.queue, 1, &cmd);
 }

src/GpuTexture.zig

@@ -1,26 +1,29 @@
 const std = @import("std");
 const c = @import("utils.zig").c;
+const svOpt = @import("utils.zig").svOpt;
 const GpuAllocator = @import("GpuAllocator.zig");
 const GpuBuffer = @import("GpuBuffer.zig");
 const GpuTextureFormat = @import("lib.zig").GpuTextureFormat;
 const GpuTextureUsage = @import("lib.zig").GpuTextureUsage;
 
 pub const GpuTextureDef = struct {
+    label: ?[]const u8 = null,
     size: c.WGPUExtent3D,
     usage: std.EnumSet(GpuTextureUsage),
     format: GpuTextureFormat,
 };
 
 raw: c.WGPUTexture,
-gloc: GpuAllocator,
+glloc: GpuAllocator,
 def: GpuTextureDef,
 
-pub fn init(gloc: GpuAllocator, def: GpuTextureDef) !@This() {
+pub fn init(glloc: GpuAllocator, def: GpuTextureDef) !@This() {
     var use: u64 = 0;
     var iter = def.usage.iterator();
     while (iter.next()) |flag| use |= @intFromEnum(flag);
 
     const desc = c.WGPUTextureDescriptor{
+        .label = svOpt(def.label),
         .usage = use,
         .dimension = c.WGPUTextureDimension_2D,
         .size = def.size,
@@ -28,13 +31,13 @@ pub fn init(gloc: GpuAllocator, def: GpuTextureDef) !@This() {
         .mipLevelCount = 1,
         .sampleCount = 1,
     };
-    const raw = try gloc.allocTexture(desc);
+    const raw = try glloc.allocTexture(desc);
 
-    return .{ .gloc = gloc, .raw = raw, .def = def };
+    return .{ .glloc = glloc, .raw = raw, .def = def };
 }
 
 pub fn deinit(self: @This()) void {
-    self.gloc.freeTexture(self.raw);
+    self.glloc.freeTexture(self.raw);
 }
 
 pub fn getConstMappedRange(self: @This(), offset: u64, size: u64) ?*const anyopaque {
@@ -50,10 +53,14 @@ pub fn bytesSizeRow(self: @This()) u32 {
 }
 
 /// Return a GpuBuffer containing a copy of the texture.
-pub fn buffCopy(self: @This(), gloc: GpuAllocator) !GpuBuffer {
-    const buf = try GpuBuffer.init(gloc, self.bytesSize(), .initMany(&.{ .CopyDst, .CopySrc }));
+pub fn buffCopy(self: @This(), glloc: GpuAllocator) !GpuBuffer {
+    const buf = try GpuBuffer.init(glloc, .{
+        .size = self.bytesSize(),
+        .usage = .initMany(&.{ .CopyDst, .CopySrc }),
+        .label = "texture_copy_buffer",
+    });
 
-    const enc = c.wgpuDeviceCreateCommandEncoder(gloc.device.device, null) orelse return error.Encoder;
+    const enc = c.wgpuDeviceCreateCommandEncoder(glloc.device.device, null) orelse return error.Encoder;
     defer c.wgpuCommandEncoderRelease(enc);
 
     const src_copy = c.WGPUTexelCopyTextureInfo{
@@ -75,7 +82,7 @@ pub fn buffCopy(self: @This(), gloc: GpuAllocator) !GpuBuffer {
 
     const cmd = c.wgpuCommandEncoderFinish(enc, null);
     defer c.wgpuCommandBufferRelease(cmd);
-    c.wgpuQueueSubmit(gloc.device.queue, 1, &cmd);
+    c.wgpuQueueSubmit(glloc.device.queue, 1, &cmd);
 
     return buf;
 }
@@ -102,42 +109,56 @@ pub fn load(
 ) !void {
     const bytes = data.len * @sizeOf(T);
 
-    if (bytes == self.size) {
-        // Aligned path: direct download
-        c.wgpuQueueWriteBuffer(self.gloc.device.queue, self.raw, 0, data.ptr, self.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.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.gloc.device.queue, self.raw, aligned_part, &remainder_buf, 4);
-    }
+    c.wgpuQueueWriteTexture(
+        self.glloc.device.queue,
+        &.{
+            .texture = self.raw,
+            .mipLevel = 0,
+            .origin = .{ .x = 0, .y = 0, .z = 0 },
+            .aspect = c.WGPUTextureAspect_All,
+        },
+        data.ptr,
+        bytes,
+        &.{
+            .offset = 0,
+            .bytesPerRow = self.bytesSizeRow(),
+            .rowsPerImage = self.def.size.height,
+        },
+        &self.def.size,
+    );
 }
 
 // GPU to CPU
 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.gloc,
-        self.size,
-        .initMany(&.{ .MapRead, .CopyDst }),
-    );
+    const staging = try init(self.glloc, .{
+        .size = self.size,
+        .usage = .initMany(&.{ .MapRead, .CopyDst }),
+        .label = "texture_read_staging",
+    });
     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 enc = c.wgpuDeviceCreateCommandEncoder(self.glloc.device.device, null) orelse return error.Encoder;
+    const src_copy = c.WGPUTexelCopyTextureInfo{
+        .texture = self.raw,
+        .mipLevel = 0,
+        .origin = .{ .x = 0, .y = 0, .z = 0 },
+        .aspect = c.WGPUTextureAspect_All,
+    };
+    const dst_copy = c.WGPUTexelCopyBufferInfo{
+        .buffer = staging.raw,
+        .layout = .{
+            .offset = 0,
+            .bytesPerRow = self.bytesSizeRow(),
+            .rowsPerImage = self.def.size.height,
+        },
+    };
+    c.wgpuCommandEncoderCopyTextureToBuffer(enc, &src_copy, &dst_copy, &self.def.size);
     const cmd = c.wgpuCommandEncoderFinish(enc, null);
     defer c.wgpuCommandEncoderRelease(enc);
     defer c.wgpuCommandBufferRelease(cmd);
-    c.wgpuQueueSubmit(self.gloc.device.queue, 1, &cmd);
+    c.wgpuQueueSubmit(self.glloc.device.queue, 1, &cmd);
 
     var mapped = false;
     staging.mapAsync(
@@ -146,7 +167,7 @@ pub fn read(self: @This(), alloc: std.mem.Allocator, T: type) ![]T {
         self.size,
         .{ .callback = onMapped, .userdata1 = &mapped },
     );
-    while (!mapped) self.gloc.device.poll();
+    while (!mapped) self.glloc.device.poll();
 
     const ptr: [*]const T = @ptrCast(@alignCast(
         staging.getConstMappedRange(0, self.size),

src/GpuTextureView.zig

@@ -1,32 +1,35 @@
 const std = @import("std");
 const c = @import("utils.zig").c;
+const svOpt = @import("utils.zig").svOpt;
 const GpuAllocator = @import("GpuAllocator.zig");
 const GpuTexture = @import("lib.zig").GpuTexture;
 const GpuTextureFormat = @import("lib.zig").GpuTextureFormat;
 const GpuTextureUsage = @import("lib.zig").GpuTextureUsage;
 
 pub const GpuViewDef = struct {
+    label: ?[]const u8 = null,
     usage: std.EnumSet(GpuTextureUsage) = .empty,
     format: GpuTextureFormat = .Undefined,
 };
 
 raw: c.WGPUTextureView,
-gloc: GpuAllocator,
+glloc: GpuAllocator,
 
-pub fn init(gloc: GpuAllocator, texture: GpuTexture, def: GpuViewDef) !@This() {
+pub fn init(glloc: 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 gloc.allocTextureView(texture.raw, .{
+    const raw = try glloc.allocTextureView(texture.raw, .{
+        .label = svOpt(def.label),
         .format = @intFromEnum(def.format),
         .usage = use,
         .mipLevelCount = 1,
         .arrayLayerCount = 1,
     });
-    return .{ .gloc = gloc, .raw = raw };
+    return .{ .glloc = glloc, .raw = raw };
 }
 
 pub fn deinit(self: @This()) void {
-    self.gloc.freeTextureView(self.raw);
+    self.glloc.freeTextureView(self.raw);
 }

src/utils.zig

@@ -3,3 +3,17 @@ pub const c = @cImport(@cInclude("wgpu.h"));
 pub fn sv(s: []const u8) c.WGPUStringView {
     return .{ .data = s.ptr, .length = s.len };
 }
+
+/// Allows safely passing an optional Zig string to a WebGPU string view.
+pub fn svOpt(s: ?[]const u8) c.WGPUStringView {
+    if (s) |str| return sv(str);
+    return .{ .data = null, .length = 0 };
+}
+
+/// Helper to print a WGPUStringView in your logs.
+pub fn viewStr(view: c.WGPUStringView) []const u8 {
+    if (view.data != null and view.length > 0) {
+        return view.data[0..view.length];
+    }
+    return "unnamed";
+}