Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix indexing in ReduceScatter in hierarchical collective #30

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Fix indexing in ReduceScatter in hierarchical collective #30

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
b134c5d
Fix indexing in ReduceScatter in hierarchical collective
dsidler Dec 11, 2024
1b55f95
Merge branch 'main' into dev/dasidler/fix-pcie-hierarchical
dsidler Dec 11, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 5 additions & 3 deletions examples/mscclang/allreduce_a100_pcie_hierarchical.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,14 +12,16 @@ def allpairs_reduce_scatter(gpuIds, size, offset):
if gpuIds[r1] != gpuIds[r2]:
index = offset + r2 * size
c = chunk(gpuIds[r1], Buffer.input, index, size=size)
c.copy(gpuIds[r2], 'scratch', sendtb=gpuIds[r2], recvtb=gpuIds[r1])
c.copy(gpuIds[r2], 'scratch', index=r1*size, sendtb=gpuIds[r2], recvtb=gpuIds[r1])

# Each rank performs a local reduction on the nth chunk
# Utilize 8 threadblocks for this reduction for better parallelism
for r in range(ngpus):
for index in range(0, size * (ngpus-1)):
for index in range(0, size):
c = chunk(gpuIds[r], Buffer.input, offset + r*size + (index % size))
c.reduce(chunk(gpuIds[r], 'scratch', index), sendtb=(index % size))
for r2 in range(ngpus):
if gpuIds[r] != gpuIds[r2]:
c.reduce(chunk(gpuIds[r], 'scratch', (r2 * size) + index), sendtb=(index % size))


def allpairs_all_gather(gpuIds, size, offset):
Expand Down
Loading