vllm-project/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-12-06 06:53:12 +08:00
Code Issues Packages Projects Releases Wiki Activity

[ROCm][Bugfix] Fix compilation errors with fused_qknorm_rope_kernel.cu (#28682)

Browse Source 
Signed-off-by: Sage Moore <sage@neuralmagic.com>
This commit is contained in:
Sage Moore
2025-11-14 11:06:01 -08:00
committed by GitHub
parent fd4555089a
commit 8977ffb5e6
1 changed files with 10 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
csrc/fused_qknorm_rope_kernel.cu
View File

@@ -37,6 +37,16 @@
#ifdef USE_ROCM
#define FINAL_MASK 0xffffffffffffffffULL
#if defined(HIP_VERSION) && HIP_VERSION < 70000000
// On ROCm versions before 7.0, __syncwarp isn't defined. The below
// implementation is copy/pasted from the implementation in ROCm 7.0
__device__ inline void __syncwarp() {
__builtin_amdgcn_fence(__ATOMIC_RELEASE, "wavefront");
__builtin_amdgcn_wave_barrier();
__builtin_amdgcn_fence(__ATOMIC_ACQUIRE, "wavefront");
}
#endif
#else
#define FINAL_MASK 0xffffffff
#endif