I've reduced this to an MRE repo (erichdongubler-mre/wgpu-matCx3-padding-clobber) that runs natively on the wgpu crate, converting …:array_of_matCx3:columns=2;use_struct=false to Rust code. I recommend using the following
RUST_LOG=debug cargo run
Expected vs observed behavior
Examining the MRE's output, it seems we're failing these tests because we clobber the padding in the firstmatCx3 element, but not the others. Weird!
Extra materials
-
Platform
This is confirmed to be an issue on MacOS. Firefox's CI reports issues on DX12, too, but I haven't investigated that platform thoroughly or reproduced locally (despite non-exhaustive attempts with the MRE above).
Description
When running some
webgpu:shader,execution,padding:*
tests in Firefox, I noticed that the following were failing:webgpu:shader,execution,padding:matCx3:*
webgpu:shader,execution,padding:array_of_matCx3:*
Repro steps
I've reduced this to an MRE repo (
erichdongubler-mre/wgpu-matCx3-padding-clobber
) that runs natively on thewgpu
crate, converting…:array_of_matCx3:columns=2;use_struct=false
to Rust code. I recommend using the followingExpected vs observed behavior
Examining the MRE's output, it seems we're failing these tests because we clobber the padding in the first
matCx3
element, but not the others. Weird!Extra materials
-
Platform
This is confirmed to be an issue on MacOS. Firefox's CI reports issues on DX12, too, but I haven't investigated that platform thoroughly or reproduced locally (despite non-exhaustive attempts with the MRE above).