Update tips for wgpu 24.0.3.1

[NaitLee]

There are significant changes in the newest wgpu API (I have 24.0.3.1 because I need to manually download it). In new version without updating the guide even basic ones like hello triangle would be broken.

Most notable ones:

1. webgpu have its own "string" now

Old way setting a string no longer work:

CommandEncoderDescriptor encoderDesc = {};
encoderDesc.label = "My command encoder";

Now these fields need a WGPUStringView instance:

#define WSTR(s) WGPUStringView{ s, strlen(s) }

CommandEncoderDescriptor encoderDesc = {};
encoderDesc.label = WSTR("My command encoder");

I think this should be done in some c++ idiom :)

2. some struct names are changed

Like the ShaderModuleWGSLDescriptor has been renamed to WGPUShaderSourceWGSL. For somewhat "complex" structures like these chains, this may confuse people.

3. probably there are more

Sadly I progress very slowly on c++ and webgpu and may not help to write the guides directly, so updates would be glad :)

[eliemichel]

Yep, I know about this and I've started upgrading it in the hidden "next" section of the guide. The main reason not to fully upgrade yet is that emscripten did not change this yet: they will do all breaking changes all at once when the API becomes 1.0, so in the meantime I cannot have a guide that both follows the very last version of the native APIs and builds with regular emscripten at the same time.

One possibility is to use emdawnwebgpu, which is a fork of emscripten that better follows the API (and its late breaking changes), but in the context of such a guide it is too much setup to implement for something that is only temporary!

Note that your WSTR macro can be simplified into #define WSTR(s) WGPUStringView{ s, WGPU_STRLEN } as the special value WGPU_STRLEN means "the string is null terminated". ;) (I plan on using std::string_view as you can see here: https://eliemichel.github.io/LearnWebGPU/next/getting-started/adapter-and-device/the-adapter.html#displaying-wgpustringview)

Thanks for taking the time to report nonetheless, I'll keep this open for further readers!

[niranjanbalachandra]

Any updates on this topic ?

[eliemichel]

Yes, the "Next" rewrite keeps on going and just reached a point where it has more chapters marked as green than than the main one!

The problem of the Web build is now mostly fixed thanks to emdawn being officially released on a daily basis, and it is now included in WebGPU-distribution (it's even the default when building with emscripten now because that's the future official way, instead of -sUSE_WEBGPU).

So the main issue now is that wgpu-native did not upgrade its version of webgpu.h for a while (it is still using this 7 months old version). The "Hello triangle" of the "Next" section currently works with Dawn and emscripten (through emdawn, but you don't have to pay attention), but does not work with wgpu-native (even the last version, which upgraded wgpu-core but not webgpu.h).

Edit: The next section does not provide stepXXX-next branches in the accompagnying code repo, but CI script automatically create them to test them: https://github.com/eliemichel/LearnWebGPU/actions

1. Select a run of type "Tangle":

2. Download the tangled artifact:

3. Then pick the chapter of your choice:

It's not 100% clean because it is automatically generated by virtually reading the guide, but at least it builds and gives a good base. :)

[niranjanbalachandra]

Try to build Hello Triangle Next and see the following build error for win32 and emscripten :

build.log

Windows 10
cmake 3.29.2
Visual Studio 17 2022
emsdk-4.0.9

Using the following command:

cmake -B build-wgpu -DWEBGPU_BACKEND=WGPU -DWEBGPU_BUILD_FROM_SOURCE=OFF -DWEBGPU_LINK_TYPE=STATIC
cmake --build build-wgpu --config Release

call emcmake cmake -B build-emscripten
cmake --build build-emscripten --config Release

[eliemichel]

As I was mentioning, wgpu-native is late, so the Next version only works with Dawn for now. Use -DWEBGPU_BACKEND=DAWN. If you really want to use wgpu-native, you'll have to stick with the old version, but then use a version of wgpu-native that is in sync with the old version of the guide.

[niranjanbalachandra]

ok thanks, how about this call to build emscripten ?

call emcmake cmake -B build-emscripten
cmake --build build-emscripten --config Release

[eliemichel]

Oh sorry I hadn't noticed there was both outputs in your log file. Maybe try emcc --clear-ports to force emscripten to reinstall emdawn? It seems to use an outdated webgpu.h, probably the one that is in emscripten by default, instead of the port.

[niranjanbalachandra]

I tried emcc --clear-ports and it didn't work. I always delete build-emscripten before doing a build.

I though you can just take the Hello Triangle - Next code from repo and then do:

call emcmake cmake -B build-emscripten
cmake --build build-emscripten --config Release

and it should compile fine right ? Am I missing a step ? Have you tried that ?

Also what version of ninja/cmake/emsdk should I have ?

[eliemichel]

Okay I can indeed reproduce on a fresh build, it's probably just a matter of re-generating the webgpu.hpp used in the emdawn case to match the webgpu.h header, will do tomorrow.

[eliemichel]

Okey I've added a dedicated emdawn/ directory in this repo because the webgpu.h of emdawn is different from the one from dawn, I had forgot to push that change.

I've also updated WebGPU-distribution (added a new prerelease WebGPU-distribution-v0.3.0-gamma) to reflect this, as well as the Next version of the guide and all branches with "next" in the name in LearnWebGPU-Code. That should fix your issue, sorry for the trouble!

I at least manually cheked the step030-next (hello triangle).

[niranjanbalachandra]

That works thanks ... Just a couple more questions:

1. Are you planning on creating/updating ImGui guide ?
2. Any plans for web socket support ?
3. Any plans to support static linkage in dawn ? How can dawn be linked statically? WebGPU-distribution#34

[eliemichel]

Are you planning on creating/updating ImGui guide ?

I will for sure. Not there yet in the wave of upgrade, but step100-next2 is an example that uses a recent version of WebGPU and Dear ImGui (Dawn only for now)

Any plans for web socket support ?

Do you have anything in mind that relates the socket to WebGPU?

Any plans to support static linkage in dawn ?

Yes, waiting for this PR in Dawn to merge: https://dawn-review.googlesource.com/c/dawn/+/238874

[alexwennstrom]

Is it possible to update adapter and device setup to match the latest dawn implementation? Not sure if wgpu-native is in sync.
New setup could look something like this (C style, event more compact if Cpp):

WGPUAdapter wgpu_adapter = nullptr;
{
      WGPURequestAdapterOptions options{};
      options.compatibleSurface = wgpu_surface;
     
      WGPURequestAdapterCallbackInfo info{};
      info.mode = WGPUCallbackMode_WaitAnyOnly;
      info.callback = &RequestAdapterCallback;
      info.userdata1 = &wgpu_adapter;
    
      WGPUFuture future = wgpuInstanceRequestAdapter(wgpu_instance, &options, info);

      WGPUFutureWaitInfo wait{};
      wait.future = future;

      WGPUWaitStatus status = wgpuInstanceWaitAny(wgpu_instance, 1, &wait, 0);
      ASSERT(status == WGPUWaitStatus_Success);
  }