AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Added Functional Tests for CSV Tuner Plugin (#1968)

Browse Source 
* Add functional tests for CSV Tuner Plugin

* Updated directory structure

* Updated and renamed directories

* Updated csv conf files

* Updated readme

* Updated readme

* Updated readme

[ROCm/rccl commit: c8da880dc7]
This commit is contained in:
Kapil S. Pawar
2025-11-11 10:11:19 -06:00
committed by GitHub
parent 0d3fba9a22
commit c4d7680749
17 changed files with 2824 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/test/ext-plugins/assets/csv_confs/incorrect_values_config.conf
+61
View File
@@ -0,0 +1,61 @@
# Test configuration with invalid/incorrect values to test plugin robustness - 8B-128M
# Invalid collective type (should default to allreduce)
wrongtype,8,65536,ring,simple,-1,-1,-1,-1,-1
# Invalid algorithm (should default to ring)
allreduce,65537,16777216,badring,simple,-1,-1,-1,-1,-1
# Invalid protocol (should default to simple)
allreduce,16777217,67108864,tree,badproto,-1,-1,-1,-1,-1
# Invalid numbers (letters instead of numbers)
allreduce,abc,def,ring,simple,xyz,-1,-1,-1,-1
# Wrong field count (should be ignored completely)
allreduce,8,128M,ring
# Mixed valid and invalid
allreduce,67108865,134217728,ring,simple,2,-1,-1,-1,-1
# Broadcast configurations with invalid/incorrect values - 8B-128M
# Invalid algorithm (should default to ring for broadcast)
broadcast,8,65536,badring,simple,-1,-1,-1,-1,-1
# Invalid protocol (should default to simple)
broadcast,65537,16777216,ring,badproto,-1,-1,-1,-1,-1
# Invalid numbers (letters instead of numbers)
broadcast,abc,def,ring,simple,xyz,-1,-1,-1,-1
# Wrong field count (should be ignored completely)
broadcast,8,128M,ring
# Mixed valid and invalid
broadcast,16777217,134217728,ring,simple,4,-1,-1,-1,-1
# AllGather configurations with invalid/incorrect values - 8B-128M
# Invalid algorithm (should default to ring for allgather)
allgather,8,65536,badring,simple,-1,-1,-1,-1,-1
# Invalid protocol (should default to simple)
allgather,65537,16777216,ring,badproto,-1,-1,-1,-1,-1
# Invalid numbers (letters instead of numbers)
allgather,abc,def,ring,simple,xyz,-1,-1,-1,-1
# Wrong field count (should be ignored completely)
allgather,8,128M,ring
# Mixed valid and invalid
allgather,16777217,134217728,ring,simple,4,-1,-1,-1,-1
# Reduce configurations with invalid/incorrect values - 8B-128M
# Invalid algorithm (should default to ring for reduce)
reduce,8,65536,badring,simple,-1,-1,-1,-1,-1
# Invalid protocol (should default to simple)
reduce,65537,16777216,ring,badproto,-1,-1,-1,-1,-1
# Invalid numbers (letters instead of numbers)
reduce,abc,def,ring,simple,xyz,-1,-1,-1,-1
# Wrong field count (should be ignored completely)
reduce,8,128M,ring
# Mixed valid and invalid
reduce,16777217,134217728,ring,simple,4,-1,-1,-1,-1
# ReduceScatter configurations with invalid/incorrect values - 8B-128M
# Invalid algorithm (should default to ring for reducescatter)
reducescatter,8,65536,badring,simple,-1,-1,-1,-1,-1
# Invalid protocol (should default to simple)
reducescatter,65537,16777216,ring,badproto,-1,-1,-1,-1,-1
# Invalid numbers (letters instead of numbers)
reducescatter,abc,def,ring,simple,xyz,-1,-1,-1,-1
# Wrong field count (should be ignored completely)
reducescatter,8,128M,ring
# Mixed valid and invalid
reducescatter,16777217,134217728,ring,simple,4,-1,-1,-1,-1
projects/rccl/test/ext-plugins/assets/csv_confs/multinode_config.conf
+104
View File
@@ -0,0 +1,104 @@
# AllReduce configurations for multi-node setups - 8B-128M
# 2 nodes, 16 ranks total - 8B-128M
allreduce,8,65536,tree,ll,4,2,16,-1,-1 # Small: tree (8B to 64K)
allreduce,65537,16777216,ring,ll128,6,2,16,-1,-1 # Medium: ring (64K to 16M)
allreduce,16777217,134217728,ring,simple,8,2,16,-1,-1 # Large: ring/simple (16M to 128M)
# 3 nodes, 24 ranks total - 8B-128M
allreduce,8,65536,tree,ll,4,2,24,-1,-1 # Small: tree (8B to 64K)
allreduce,65537,16777216,ring,ll128,6,2,24,-1,-1 # Medium: ring (64K to 16M)
allreduce,16777217,134217728,ring,simple,8,2,24,-1,-1 # Large: ring/simple (16M to 128M)
# 4 nodes, 32 ranks total - 8B-128M
allreduce,8,65536,tree,ll,4,2,32,-1,-1 # Small: tree (8B to 64K)
allreduce,65537,16777216,ring,ll128,6,2,32,-1,-1 # Medium: ring (64K to 16M)
allreduce,16777217,134217728,ring,simple,8,2,32,-1,-1 # Large: ring/simple (16M to 128M)
# 5 nodes, 40 ranks total - 8B-128M
allreduce,8,65536,tree,ll,4,2,40,-1,-1 # Small: tree (8B to 64K)
allreduce,65537,16777216,ring,ll128,6,2,40,-1,-1 # Medium: ring (64K to 16M)
allreduce,16777217,134217728,ring,simple,8,2,40,-1,-1 # Large: ring/simple (16M to 128M)
# Broadcast configurations for multi-node setups - 8B-128M
# 2 nodes, 16 ranks total
broadcast,8,65536,ring,ll,4,2,16,-1,-1 # Small: ring (8B to 64K)
broadcast,65537,16777216,ring,ll128,6,2,16,-1,-1 # Medium: ring (64K to 16M)
broadcast,16777217,134217728,ring,simple,8,2,16,-1,-1 # Large: ring/simple (16M to 128M)
# 3 nodes, 24 ranks total
broadcast,8,65536,ring,ll,4,2,24,-1,-1 # Small: ring (8B to 64K)
broadcast,65537,16777216,ring,ll128,6,2,24,-1,-1 # Medium: ring (64K to 16M)
broadcast,16777217,134217728,ring,simple,8,2,24,-1,-1 # Large: ring/simple (16M to 128M)
# 4 nodes, 32 ranks total
broadcast,8,65536,ring,ll,4,2,32,-1,-1 # Small: ring (8B to 64K)
broadcast,65537,16777216,ring,ll128,6,2,32,-1,-1 # Medium: ring (64K to 16M)
broadcast,16777217,134217728,ring,simple,8,2,32,-1,-1 # Large: ring/simple (16M to 128M)
# 5 nodes, 40 ranks total
broadcast,8,65536,ring,ll,4,2,40,-1,-1 # Small: ring (8B to 64K)
broadcast,65537,16777216,ring,ll128,6,2,40,-1,-1 # Medium: ring (64K to 16M)
broadcast,16777217,134217728,ring,simple,8,2,40,-1,-1 # Large: ring/simple (16M to 128M)
# AllGather configurations for multi-node setups - 8B-128M
# 2 nodes, 16 ranks total
allgather,8,65536,ring,ll,4,2,16,-1,-1 # Small: ring (8B to 64K)
allgather,65537,16777216,ring,ll128,6,2,16,-1,-1 # Medium: ring (64K to 16M)
allgather,16777217,134217728,ring,simple,8,2,16,-1,-1 # Large: ring/simple (16M to 128M)
# 3 nodes, 24 ranks total
allgather,8,65536,ring,ll,4,2,24,-1,-1 # Small: ring (8B to 64K)
allgather,65537,16777216,ring,ll128,6,2,24,-1,-1 # Medium: ring (64K to 16M)
allgather,16777217,134217728,ring,simple,8,2,24,-1,-1 # Large: ring/simple (16M to 128M)
# 4 nodes, 32 ranks total
allgather,8,65536,ring,ll,4,2,32,-1,-1 # Small: ring (8B to 64K)
allgather,65537,16777216,ring,ll128,6,2,32,-1,-1 # Medium: ring (64K to 16M)
allgather,16777217,134217728,ring,simple,8,2,32,-1,-1 # Large: ring/simple (16M to 128M)
# 5 nodes, 40 ranks total
allgather,8,65536,ring,ll,4,2,40,-1,-1 # Small: ring (8B to 64K)
allgather,65537,16777216,ring,ll128,6,2,40,-1,-1 # Medium: ring (64K to 16M)
allgather,16777217,134217728,ring,simple,8,2,40,-1,-1 # Large: ring/simple (16M to 128M)
# Reduce configurations for multi-node setups - 8B-128M
# 2 nodes, 16 ranks total
reduce,8,65536,ring,ll,4,2,16,-1,-1 # Small: ring (8B to 64K)
reduce,65537,16777216,ring,ll128,6,2,16,-1,-1 # Medium: ring (64K to 16M)
reduce,16777217,134217728,ring,simple,8,2,16,-1,-1 # Large: ring/simple (16M to 128M)
# 3 nodes, 24 ranks total
reduce,8,65536,ring,ll,4,2,24,-1,-1 # Small: ring (8B to 64K)
reduce,65537,16777216,ring,ll128,6,2,24,-1,-1 # Medium: ring (64K to 16M)
reduce,16777217,134217728,ring,simple,8,2,24,-1,-1 # Large: ring/simple (16M to 128M)
# 4 nodes, 32 ranks total
reduce,8,65536,ring,ll,4,2,32,-1,-1 # Small: ring (8B to 64K)
reduce,65537,16777216,ring,ll128,6,2,32,-1,-1 # Medium: ring (64K to 16M)
reduce,16777217,134217728,ring,simple,8,2,32,-1,-1 # Large: ring/simple (16M to 128M)
# 5 nodes, 40 ranks total
reduce,8,65536,ring,ll,4,2,40,-1,-1 # Small: ring (8B to 64K)
reduce,65537,16777216,ring,ll128,6,2,40,-1,-1 # Medium: ring (64K to 16M)
reduce,16777217,134217728,ring,simple,8,2,40,-1,-1 # Large: ring/simple (16M to 128M)
# ReduceScatter configurations for multi-node setups - 8B-128M
# 2 nodes, 16 ranks total
reducescatter,8,65536,ring,ll,4,2,16,-1,-1 # Small: ring (8B to 64K)
reducescatter,65537,16777216,ring,ll128,6,2,16,-1,-1 # Medium: ring (64K to 16M)
reducescatter,16777217,134217728,ring,simple,8,2,16,-1,-1 # Large: ring/simple (16M to 128M)
# 3 nodes, 24 ranks total
reducescatter,8,65536,ring,ll,4,2,24,-1,-1 # Small: ring (8B to 64K)
reducescatter,65537,16777216,ring,ll128,6,2,24,-1,-1 # Medium: ring (64K to 16M)
reducescatter,16777217,134217728,ring,simple,8,2,24,-1,-1 # Large: ring/simple (16M to 128M)
# 4 nodes, 32 ranks total
reducescatter,8,65536,ring,ll,4,2,32,-1,-1 # Small: ring (8B to 64K)
reducescatter,65537,16777216,ring,ll128,6,2,32,-1,-1 # Medium: ring (64K to 16M)
reducescatter,16777217,134217728,ring,simple,8,2,32,-1,-1 # Large: ring/simple (16M to 128M)
# 5 nodes, 40 ranks total
reducescatter,8,65536,ring,ll,4,2,40,-1,-1 # Small: ring (8B to 64K)
reducescatter,65537,16777216,ring,ll128,6,2,40,-1,-1 # Medium: ring (64K to 16M)
reducescatter,16777217,134217728,ring,simple,8,2,40,-1,-1 # Large: ring/simple (16M to 128M)
projects/rccl/test/ext-plugins/assets/csv_confs/no_matching_config.conf
+44
View File
@@ -0,0 +1,44 @@
# Invalid configuration that should NOT match the test environment
# Wrong collective type (broadcast instead of allreduce for allreduce tests) - 8B-128M
broadcast,0,7,ring,simple,-1,-1,-1,-1,-1 # Size too small (broadcast test starts at 8B)
broadcast,134217729,268435456,ring,simple,-1,-1,-1,-1,-1 # Size too large (broadcast test ends at 128M)
# Wrong collective type (allreduce instead of broadcast for broadcast tests) - 8B-128M
allreduce,0,7,tree,simple,-1,-1,-1,-1,-1 # Size too small (doesn't include 8B+)
allreduce,134217729,268435456,tree,simple,-1,-1,-1,-1,-1 # Size too large (doesn't include 8B-128M)
# Wrong collective type (allgather instead of other tests) - 8B-128M
allgather,0,7,ring,simple,-1,-1,-1,-1,-1 # Size too small (allgather test starts at 8B)
allgather,134217729,268435456,ring,simple,-1,-1,-1,-1,-1 # Size too large (allgather test ends at 128M)
# Wrong collective type (reduce instead of other tests) - 8B-128M
reduce,0,7,ring,simple,-1,-1,-1,-1,-1 # Size too small (reduce test starts at 8B)
reduce,134217729,268435456,ring,simple,-1,-1,-1,-1,-1 # Size too large (reduce test ends at 128M)
# Wrong collective type (reducescatter instead of other tests) - 8B-128M
reducescatter,0,7,ring,simple,-1,-1,-1,-1,-1 # Size too small (reducescatter test starts at 8B)
reducescatter,134217729,268435456,ring,simple,-1,-1,-1,-1,-1 # Size too large (reducescatter test ends at 128M)
# Wrong size ranges (tests use 8 bytes to 128M, these are way outside)
allreduce,2000000000,4000000000,ring,simple,-1,-1,-1,-1,-1
allreduce,5000000000,6000000000,tree,ll128,-1,-1,-1,-1,-1
broadcast,2000000000,4000000000,ring,simple,-1,-1,-1,-1,-1
broadcast,5000000000,6000000000,ring,ll128,-1,-1,-1,-1,-1
allgather,2000000000,4000000000,ring,simple,-1,-1,-1,-1,-1
allgather,5000000000,6000000000,ring,ll128,-1,-1,-1,-1,-1
reduce,2000000000,4000000000,ring,simple,-1,-1,-1,-1,-1
reduce,5000000000,6000000000,ring,ll128,-1,-1,-1,-1,-1
reducescatter,2000000000,4000000000,ring,simple,-1,-1,-1,-1,-1
reducescatter,5000000000,6000000000,ring,ll128,-1,-1,-1,-1,-1
# Wrong node/rank counts (broadcast tests use 1 node, 4/8 ranks - these specify different, size outside 8B-128M)
broadcast,134217729,268435456,ring,simple,-1,4,16,-1,-1 # 4 nodes, 16 ranks, size too large
broadcast,268435457,536870912,ring,simple,-1,2,4,-1,-1 # 2 nodes, 4 ranks, size too large
# Wrong node/rank counts (allreduce tests use 1 node, 4/8 ranks - these specify different, size outside 8B-128M)
allreduce,134217729,268435456,ring,simple,-1,8,16,-1,-1 # 8 nodes, 16 ranks, size too large
allreduce,268435457,536870912,tree,simple,-1,2,8,-1,-1 # 2 nodes, 8 ranks, size too large
# Wrong node/rank counts (allgather tests use 1 node, 8 ranks - these specify different, size outside 8B-128M)
allgather,134217729,268435456,ring,simple,-1,4,16,-1,-1 # 4 nodes, 16 ranks, size too large
allgather,268435457,536870912,ring,simple,-1,2,4,-1,-1 # 2 nodes, 4 ranks, size too large
# Wrong node/rank counts (reduce tests use 1 node, 8 ranks - these specify different, size outside 8B-128M)
reduce,134217729,268435456,ring,simple,-1,4,16,-1,-1 # 4 nodes, 16 ranks, size too large
reduce,268435457,536870912,ring,simple,-1,2,4,-1,-1 # 2 nodes, 4 ranks, size too large
# Wrong node/rank counts (reducescatter tests use 1 node, 8 ranks - these specify different, size outside 8B-128M)
reducescatter,134217729,268435456,ring,simple,-1,4,16,-1,-1 # 4 nodes, 16 ranks, size too large
reducescatter,268435457,536870912,ring,simple,-1,2,4,-1,-1 # 2 nodes, 4 ranks, size too large
projects/rccl/test/ext-plugins/assets/csv_confs/singlenode_config.conf
+39
View File
@@ -0,0 +1,39 @@
# Single-node allreduce configuration for 8 ranks, 1 node - use working algorithm/protocol combinations - 8B-128M
# Small allreduce: tree/ll (8 bytes to 64K)
allreduce,8,65536,tree,ll,2,1,8,-1,-1
# Medium allreduce: ring/ll128 (64K to 16M)
allreduce,65537,16777216,ring,ll128,4,1,8,-1,-1
# Large allreduce: ring/simple (16M to 128M)
allreduce,16777217,134217728,ring,simple,6,1,8,-1,-1
# Single-node broadcast configuration for 8 ranks, 1 node - use ring algorithm (supported for broadcast) - 8B-128M
# Small broadcast: ring/ll (8 bytes to 64K)
broadcast,8,65536,ring,ll,2,1,8,-1,-1
# Medium broadcast: ring/ll128 (64K to 16M)
broadcast,65537,16777216,ring,ll128,4,1,8,-1,-1
# Large broadcast: ring/simple (16M to 128M)
broadcast,16777217,134217728,ring,simple,6,1,8,-1,-1
# Single-node allgather configuration for 8 ranks, 1 node - use ring algorithm (supported for allgather) - 8B-128M
# Small allgather: ring/ll (8 bytes to 64K)
allgather,8,65536,ring,ll,2,1,8,-1,-1
# Medium allgather: ring/ll128 (64K to 16M)
allgather,65537,16777216,ring,ll128,4,1,8,-1,-1
# Large allgather: ring/simple (16M to 128M)
allgather,16777217,134217728,ring,simple,6,1,8,-1,-1
# Single-node reduce configuration for 8 ranks, 1 node - use ring algorithm (supported for reduce) - 8B-128M
# Small reduce: ring/ll (8 bytes to 64K)
reduce,8,65536,ring,ll,2,1,8,-1,-1
# Medium reduce: ring/ll128 (64K to 16M)
reduce,65537,16777216,ring,ll128,4,1,8,-1,-1
# Large reduce: ring/simple (16M to 128M)
reduce,16777217,134217728,ring,simple,6,1,8,-1,-1
# Single-node reducescatter configuration for 8 ranks, 1 node - use ring algorithm (supported for reducescatter) - 8B-128M
# Small reducescatter: ring/ll (8 bytes to 64K)
reducescatter,8,65536,ring,ll,2,1,8,-1,-1
# Medium reducescatter: ring/ll128 (64K to 16M)
reducescatter,65537,16777216,ring,ll128,4,1,8,-1,-1
# Large reducescatter: ring/simple (16M to 128M)
reducescatter,16777217,134217728,ring,simple,6,1,8,-1,-1
projects/rccl/test/ext-plugins/assets/csv_confs/unsupported_algo_proto_config.conf
+40
View File
@@ -0,0 +1,40 @@
# AllReduce configurations with unsupported algorithm/protocol combinations - 8B-128M
# These have valid size ranges and collective types but unsupported algorithms
allreduce,8,65536,collnet_direct,simple,2,-1,-1,-1,-1
allreduce,65537,8388608,collnet_chain,ll,4,-1,-1,-1,-1
allreduce,8388609,33554432,nvls,simple,-1,-1,-1,-1,-1
allreduce,33554433,67108864,nvls_tree,ll128,-1,-1,-1,-1,-1
allreduce,67108865,100663296,pat,simple,-1,-1,-1,-1,-1
allreduce,100663297,134217728,collnet_direct,ll128,-1,-1,-1,-1,-1
# Broadcast configurations with unsupported algorithm/protocol combinations - 8B-128M
broadcast,8,65536,collnet_direct,simple,2,-1,-1,-1,-1
broadcast,65537,8388608,collnet_chain,ll,4,-1,-1,-1,-1
broadcast,8388609,33554432,nvls,simple,-1,-1,-1,-1,-1
broadcast,33554433,67108864,nvls_tree,ll128,-1,-1,-1,-1,-1
broadcast,67108865,100663296,pat,simple,-1,-1,-1,-1,-1
broadcast,100663297,134217728,collnet_direct,ll128,-1,-1,-1,-1,-1
# AllGather configurations with unsupported algorithm/protocol combinations - 8B-128M
allgather,8,65536,collnet_direct,simple,2,-1,-1,-1,-1
allgather,65537,8388608,collnet_chain,ll,4,-1,-1,-1,-1
allgather,8388609,33554432,nvls,simple,-1,-1,-1,-1,-1
allgather,33554433,67108864,nvls_tree,ll128,-1,-1,-1,-1,-1
allgather,67108865,100663296,pat,simple,-1,-1,-1,-1,-1
allgather,100663297,134217728,collnet_direct,ll128,-1,-1,-1,-1,-1
# Reduce configurations with unsupported algorithm/protocol combinations - 8B-128M
reduce,8,65536,collnet_direct,simple,2,-1,-1,-1,-1
reduce,65537,8388608,collnet_chain,ll,4,-1,-1,-1,-1
reduce,8388609,33554432,nvls,simple,-1,-1,-1,-1,-1
reduce,33554433,67108864,nvls_tree,ll128,-1,-1,-1,-1,-1
reduce,67108865,100663296,pat,simple,-1,-1,-1,-1,-1
reduce,100663297,134217728,collnet_direct,ll128,-1,-1,-1,-1,-1
# ReduceScatter configurations with unsupported algorithm/protocol combinations - 8B-128M
reducescatter,8,65536,collnet_direct,simple,2,-1,-1,-1,-1
reducescatter,65537,8388608,collnet_chain,ll,4,-1,-1,-1,-1
reducescatter,8388609,33554432,nvls,simple,-1,-1,-1,-1,-1
reducescatter,33554433,67108864,nvls_tree,ll128,-1,-1,-1,-1,-1
reducescatter,67108865,100663296,pat,simple,-1,-1,-1,-1,-1
reducescatter,100663297,134217728,collnet_direct,ll128,-1,-1,-1,-1,-1
projects/rccl/test/ext-plugins/assets/csv_confs/valid_config_with_wildcards.conf
+39
View File
@@ -0,0 +1,39 @@
# AllReduce configurations with supported algo/proto combinations - 8B-128M
# Small allreduce: tree/ll (8 bytes to 64K)
allreduce,8,65536,tree,ll,2,-1,-1,-1,-1
# Medium allreduce: ring/ll128 (64K to 16M)
allreduce,65537,16777216,ring,ll128,4,-1,-1,-1,-1
# Large allreduce: ring/simple (16M to 128M)
allreduce,16777217,134217728,ring,simple,-1,-1,-1,-1,-1
# Broadcast configurations with supported algo/proto combinations - 8B-128M
# Small broadcast: ring/ll (8 bytes to 64K)
broadcast,8,65536,ring,ll,2,-1,-1,-1,-1
# Medium broadcast: ring/ll128 (64K to 16M)
broadcast,65537,16777216,ring,ll128,4,-1,-1,-1,-1
# Large broadcast: ring/simple (16M to 128M)
broadcast,16777217,134217728,ring,simple,-1,-1,-1,-1,-1
# AllGather configurations with supported algo/proto combinations - 8B-128M
# Small allgather: ring/ll (8 bytes to 64K)
allgather,8,65536,ring,ll,2,-1,-1,-1,-1
# Medium allgather: ring/ll128 (64K to 16M)
allgather,65537,16777216,ring,ll128,4,-1,-1,-1,-1
# Large allgather: ring/simple (16M to 128M)
allgather,16777217,134217728,ring,simple,-1,-1,-1,-1,-1
# Reduce configurations with supported algo/proto combinations - 8B-128M
# Small reduce: ring/ll (8 bytes to 64K)
reduce,8,65536,ring,ll,2,-1,-1,-1,-1
# Medium reduce: ring/ll128 (64K to 16M)
reduce,65537,16777216,ring,ll128,4,-1,-1,-1,-1
# Large reduce: ring/simple (16M to 128M)
reduce,16777217,134217728,ring,simple,-1,-1,-1,-1,-1
# ReduceScatter configurations with supported algo/proto combinations - 8B-128M
# Small reducescatter: ring/ll (8 bytes to 64K)
reducescatter,8,65536,ring,ll,2,-1,-1,-1,-1
# Medium reducescatter: ring/ll128 (64K to 16M)
reducescatter,65537,16777216,ring,ll128,4,-1,-1,-1,-1
# Large reducescatter: ring/simple (16M to 128M)
reducescatter,16777217,134217728,ring,simple,-1,-1,-1,-1,-1
projects/rccl/test/ext-plugins/assets/csv_confs/valid_config_without_wildcards.conf
+40
View File
@@ -0,0 +1,40 @@
# AllReduce configurations with specific nodes/ranks values (no wildcards for topology) - 8B-128M
# But use wildcards for pipeOps/regBuff since they vary at runtime
# Small allreduce: tree/ll, 2 channels, 1 node, 4 ranks, any pipeOps, any regBuff (8 bytes to 64K)
allreduce,8,65536,tree,ll,2,1,4,-1,-1
# Medium allreduce: ring/ll128, 4 channels, 1 node, 4 ranks, any pipeOps, any regBuff (64K to 16M)
allreduce,65537,16777216,ring,ll128,4,1,4,-1,-1
# Large allreduce: ring/simple, 6 channels, 1 node, 4 ranks, any pipeOps, any regBuff (16M to 128M)
allreduce,16777217,134217728,ring,simple,6,1,4,-1,-1
# Broadcast configurations with specific nodes/ranks values (no wildcards for topology) - 8B-128M
# Small broadcast: ring/ll, 2 channels, 1 node, 4 ranks, any pipeOps, any regBuff (8 bytes to 64K)
broadcast,8,65536,ring,ll,2,1,4,-1,-1
# Medium broadcast: ring/ll128, 4 channels, 1 node, 4 ranks, any pipeOps, any regBuff (64K to 16M)
broadcast,65537,16777216,ring,ll128,4,1,4,-1,-1
# Large broadcast: ring/simple, 6 channels, 1 node, 4 ranks, any pipeOps, any regBuff (16M to 128M)
broadcast,16777217,134217728,ring,simple,6,1,4,-1,-1
# AllGather configurations with specific nodes/ranks values (no wildcards for topology) - 8B-128M
# Small allgather: ring/ll, 2 channels, 1 node, 4 ranks, any pipeOps, any regBuff (8 bytes to 64K)
allgather,8,65536,ring,ll,2,1,4,-1,-1
# Medium allgather: ring/ll128, 4 channels, 1 node, 4 ranks, any pipeOps, any regBuff (64K to 16M)
allgather,65537,16777216,ring,ll128,4,1,4,-1,-1
# Large allgather: ring/simple, 6 channels, 1 node, 4 ranks, any pipeOps, any regBuff (16M to 128M)
allgather,16777217,134217728,ring,simple,6,1,4,-1,-1
# Reduce configurations with specific nodes/ranks values (no wildcards for topology) - 8B-128M
# Small reduce: ring/ll, 2 channels, 1 node, 4 ranks, any pipeOps, any regBuff (8 bytes to 64K)
reduce,8,65536,ring,ll,2,1,4,-1,-1
# Medium reduce: ring/ll128, 4 channels, 1 node, 4 ranks, any pipeOps, any regBuff (64K to 16M)
reduce,65537,16777216,ring,ll128,4,1,4,-1,-1
# Large reduce: ring/simple, 6 channels, 1 node, 4 ranks, any pipeOps, any regBuff (16M to 128M)
reduce,16777217,134217728,ring,simple,6,1,4,-1,-1
# ReduceScatter configurations with specific nodes/ranks values (no wildcards for topology) - 8B-128M
# Small reducescatter: ring/ll, 2 channels, 1 node, 4 ranks, any pipeOps, any regBuff (8 bytes to 64K)
reducescatter,8,65536,ring,ll,2,1,4,-1,-1
# Medium reducescatter: ring/ll128, 4 channels, 1 node, 4 ranks, any pipeOps, any regBuff (64K to 16M)
reducescatter,65537,16777216,ring,ll128,4,1,4,-1,-1
# Large reducescatter: ring/simple, 6 channels, 1 node, 4 ranks, any pipeOps, any regBuff (16M to 128M)
reducescatter,16777217,134217728,ring,simple,6,1,4,-1,-1