webgpu continued #946
webgpu continued #946
Conversation
edevil
force-pushed
the
w4
branch
2 times, most recently
from
bbe4678 to
5ccafd6
Compare
edevil
force-pushed
the
w4
branch
9 times, most recently
from
edb3884 to
5b9583e
Compare
| : device_(d), async_(kj::refcounted<AsyncRunner>(d)) { | ||
| device_.SetLoggingCallback( | ||
| [](WGPULoggingType type, char const *message, void *userdata) { | ||
| KJ_LOG(INFO, "WebGPU logging", kj::str(type), message); |
There was a problem hiding this comment.
What's the purpose of the KJ_LOG statements? We do not generate INFO logs in production so if these are meant to be seen in production they need to use KJ_LOG(WARNING, ...
There was a problem hiding this comment.
This is purely for debugging purposes when someone is using --verbose. I don't yet have a place to send these.
Would it be possible/advisable to simulate console.log()s so that the script author could see these?
edevil
force-pushed
the
w4
branch
4 times, most recently
from
2308d3a to
278841b
Compare
Implemented more webgpu methods and objects.
Ensure `lost` promise is fulfilled when the callback is triggered or the device is destroyed.
Correctly handle case where there are issues finding a GPU.
If we store a local reference to the detach key it can happen that at detach time it won't match the key stored on the arraybuffer, therefore failing the detach operation.
|
Any additional comments @jasnell ? |
| ? buffer_.GetMappedRange(o, s) | ||
| : const_cast<void *>(buffer_.GetConstMappedRange(o, s)); | ||
| : const_cast<void*>(buffer_.GetConstMappedRange(o, s)); |
There was a problem hiding this comment.
What's the expected behavior here if offset and size are outside the allocated range? Will this return nullptr? For instance, if the range is only 10 and I specify an offset of 5 and a size of 10, will this return nullptr?
There was a problem hiding this comment.
Exactly, ptr will be nullptr and JSG_REQUIRE will bail out.
Use a kj::Array instead of a std::vector.
| } | ||
|
|
||
| kj::Maybe<GPUFeatureName> getFeatureName(wgpu::FeatureName& feature) { | ||
| switch (feature) { |
There was a problem hiding this comment.
You might consider implementing KJ_STRINGIFY for these enum structures. The implementation would be largely the same but it would allow you to do kj::str(feature), rather than kj::str(getFeatureName(feature)), etc. Search the code base for examples of where KJ_STRINGIFY has been implemented.
There was a problem hiding this comment.
I didn't use KJ_STRINGIFY because in this case I want to return an optional value. We may receive unknown features and I don't want to blow up but just ignore those features. KJ_STRINGIFY AFAICT, forces me to return some string.
Remove unique_ptr usage with V8Refs.
Implemented more webgpu methods and objects.