Julian Samaroo
Julian Samaroo
With the new timespan logging from #263, `@log_finish` is not guaranteed to be hit if the function it's being called in throws an error. We should implement a convenience macro...
When allocating memory on the device with `AMDGPU.malloc`, it's possible to free those allocations with `AMDGPU.Device.free`. However, this doesn't match Julia's allocate-and-GC-later model; to that end, `malloc`'d data is automatically...
Right now, `@rocprintf` assumes that all lanes are executing it in lockstep, and that all lanes in the workgroup will execute it. We can't reliably make this assumption in all...
As seen in issues like #305, Pkg will often select mismatched artifact versions, and won't always properly match with the version of LLVM being actively used (which is also not...
This makes it a bit easier to parse raw log records, and provides additional utilities for enabling, configuring, fetching, and rendering logs. In particular, we add a new GraphViz.jl-powered DAG...
Adds a `spawn_streaming` task queue to transform tasks into continuously-executing equivalents that automatically take from inputs streams/channels and put their result to an output stream/channel. Useful for processing tons of...
This was started during JuliaCon with @pszufe but I never finished it - it's a naive Graphs.jl directed graph implementation that uses Dagger to partition the graph. @gdalle Todo: -...
Dagger has previously been a purely-functional model, expecting that new values are calculated by applying a function to a set of existing values, without allowing for arguments to be mutated...
As Dagger is a very complicated set of interacting components and APIs, it would be very useful to be able to track Dagger's performance, scalability, and latency over time to...
This PR brings `Dagger.@spawn` closer to `Threads.@spawn` by allowing an alternative parsing mode when `begin ... end` blocks are encountered; specifically, such blocks are treated as a function without any...