A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 8, 08:45 UTC
small adjustment to wvSplitKrc (#34410)
Signed-off-by: Hashem Hashemi <hashem.hashemi@amd.com>
Hashem Hashemi committed
Feb 12, 2026 at 12:26 UTC
fac4e96940d9f2ac8dde8fc864b4c76cbdfd0e2d
1 file changed
+1
-2
csrc/rocm/skinny_gemms.cu
+1
-2
@@ -1568,8 +1568,7 @@ __global__ void __launch_bounds__(WvPrGrp* THRDS)
1568
{
1569
#endif
1570
unsigned int kOff = k + (thrd * A_CHUNK);
1571
- unsigned int kOffcp =
1572
- k_str + kOff; // min__(K - A_CHUNK, k_str + kOff);
1571
+ unsigned int kOffcp = min__(K - A_CHUNK, k_str + kOff);
1572
for (unsigned int n = 0; n < N; n += CHUNKK * sprdN) {
1573
__builtin_amdgcn_global_load_lds(
1574
(int*)(&A[min__(