mozempk/ggml-org-llama-cpp-mirror
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-08-14 19:48:33 +02:00
Code Issues Packages Projects Releases Wiki Activity

ggml-hip : remove -funsafe-math-optimizations (#26696)

Browse Source 
It enables -fassociative-math, which reassociates FP reductions and can flip
greedy argmax on RDNA3.5 (e.g. MTP speculative decode diverging from the
non-speculative baseline). Drop it so HIP builds are IEEE-conformant.

Co-authored-by: Jim Wu <ywu@xilinx.com>
This commit is contained in:
Jim Wu
2026-08-13 08:38:02 +02:00
committed by GitHub
co-authored by Jim Wu
parent d86c7d62df
commit e79e4bf660
1 changed files with 0 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-hip/CMakeLists.txt
-3
View File
@@ -126,9 +126,6 @@ if (GGML_HIP_EXPORT_METRICS)
set(CMAKE_HIP_FLAGS "${CMAKE_HIP_FLAGS} -Rpass-analysis=kernel-resource-usage --save-temps")
endif()
# Fast math for HIP, like CUDA's -use_fast_math. Not -ffast-math: that implies -ffinite-math-only, which breaks ggml's INFINITY masking and produces NaNs.
set(CMAKE_HIP_FLAGS "${CMAKE_HIP_FLAGS} -funsafe-math-optimizations")
if (NOT GGML_CUDA_FA)
add_compile_definitions(GGML_CUDA_NO_FA)
endif()