ggml-cuda: flush legacy pool on OOM and retry (#22155)

* ggml-cuda: flush legacy pool on OOM and retry

Signed-off-by: 梁厚宏 <2695316095@qq.com>

* Address review comments: add explicit sync, update destructor, clean up MUSA macros

Signed-off-by: 梁厚宏 <2695316095@qq.com>

---------

Signed-off-by: 梁厚宏 <2695316095@qq.com>
This commit is contained in:
leonardHONG
2026-04-20 23:30:38 +02:00
committed by GitHub
parent 86f8daacfe
commit 97895129e5
3 changed files with 23 additions and 2 deletions
+1
View File
@@ -58,6 +58,7 @@
#define cudaDeviceProp hipDeviceProp_t
#define cudaDeviceSynchronize hipDeviceSynchronize
#define cudaError_t hipError_t
#define cudaErrorMemoryAllocation hipErrorOutOfMemory
#define cudaErrorPeerAccessAlreadyEnabled hipErrorPeerAccessAlreadyEnabled
#define cudaErrorPeerAccessNotEnabled hipErrorPeerAccessNotEnabled
#define cudaEventCreateWithFlags hipEventCreateWithFlags
+1
View File
@@ -42,6 +42,7 @@
#define cudaDeviceProp musaDeviceProp
#define cudaDeviceSynchronize musaDeviceSynchronize
#define cudaError_t musaError_t
#define cudaErrorMemoryAllocation musaErrorMemoryAllocation
#define cudaErrorPeerAccessAlreadyEnabled musaErrorPeerAccessAlreadyEnabled
#define cudaErrorPeerAccessNotEnabled musaErrorPeerAccessNotEnabled
#define cudaEventCreateWithFlags musaEventCreateWithFlags