Skip to content

GitLab

You need to sign in or sign up before continuing.
Unverified Commit 1618700c authored by Daniel Hiltgen's avatar Daniel Hiltgen Committed by GitHub
Browse files

Workaround buggy P2P ROCm copy on windows (#7466) 

This enables the workaround code only for windows which should help windows users with muliple AMD GPUs
parent b111aa5a
Hide whitespace changes
Inline Side-by-side
Showing with 6 additions and 0 deletions
llama/make/Makefile.rocm
View file @ 1618700c
...@@ -92,6 +92,12 @@ GPU_COMPILER_CUFLAGS = \ ...@@ -92,6 +92,12 @@ GPU_COMPILER_CUFLAGS = \
-Wno-unused-result \ -Wno-unused-result \
-I. -I.
# Workaround buggy P2P copy on some windows multi-GPU setups
# This workaround breaks linux systems with small system RAM, so only enable on windows
ifeq ($(OS),windows)
GPU_COMPILER_CUFLAGS += -DGGML_CUDA_NO_PEER_COPY=1
endif
include make/gpu.make include make/gpu.make
# Adjust the rules from gpu.make to handle the ROCm dependencies properly # Adjust the rules from gpu.make to handle the ROCm dependencies properly
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment