packages/rig/lib · raven · Tangled

forked from raven-ml/raven

raven packages rig lib

14 folders · 37 files

Thibaut Mattio · fix(rig): Define the fork's loss reason only where there is a fork · 2w ago

fix(rig): Block signals across calls below rig's core

fix(rig): Block signals across calls below rig's core

fix(rig): Give CUDA's free of page-locked memory no landing point

perf(rig): Call rig.disk's guarded entries without a closure

refactor(rig): Rearm signals through the runtime's public interface

docs(rig): State the limits of loss, host access and close

feat(rig): Name each path's GPUs and keep its refusal in Rig_gpus

fix(rig): Block signals across calls below rig's core

fix(rig): Block signals across calls below rig's core

fix(rig): Keep NV memory an RM object still names

fix(rig): Block signals across calls below rig's core

fix(rig): Run long file reads as blocking jobs of the host pool

feat(rig): Queue rig_program's moves at run-value offsets

fix(rig): Close the hosts a remote start opened when it raised

feat(rig): Let another PCI function reach a buffer's memory by DMA

feat(rig): Let another PCI function reach a buffer's memory by DMA

feat(rig): Let Buffer.free kill with a reason it is given

feat(rig): Let Buffer.free kill with a reason it is given

perf(rig): Commit staged legs as the copy queues them

perf(rig): Commit staged legs as the copy queues them

refactor(rig): Keep the device fields C reads first

fix(rig): Block signals across calls below rig's core

fix(rig): Block signals across calls below rig's core

perf(rig): Build only rig's own C without semantic interposition

fix(rig): Link host code at any section alignment

docs(rig): Name the io refusal and the host code rules where they live

refactor(rig): Resolve an image's function once into Image.func

refactor(rig): Resolve an image's function once into Image.func

feat(rig): Refuse a call from inside a lock as it starts

perf(rig): Give staging halves back in one hold of the lock

memory_device.ml

feat(rig): Let another PCI function reach a buffer's memory by DMA

memory_device.mli

fix(rig): Reach a memory device only from devices that map host memory

fix(rig): Block signals across calls below rig's core

fix(rig): Stop a lost device only at its close or reopen

feat(rig): Talk to the GPU ourselves

docs(rig): Point core's private interfaces at rig.mli's rules

perf(rig): Hand profiles their entries and events without closures

perf(rig): Hand profiles their entries and events without closures

fix(rig): Define the fork's loss reason only where there is a fork

fix(rig): Drain a stopped device's late collections at any call

rig_host_code.c

feat(rig): Give a room check the count of its submission's waits

fix(rig): Stop a lost device only at its close or reopen

perf(rig): Answer a live point inline in stamps walks

perf(rig): Answer a live point inline in stamps walks

feat(rig): Add rig_buffers_overlap to rig.h

feat(rig): Let another PCI function reach a buffer's memory by DMA

fix(rig): Block signals across calls below rig's core

fix(rig): Stop a lost device only at its close or reopen

perf(rig): Keep each device's fill run per byte

perf(rig): Find a cached allocation without boxing its lookup

docs(rig): Point core's private interfaces at rig.mli's rules