msccl: enable basic collective trace (#959)

To avoid increasing number of kernels, colltrace is only enabled with
RCCL_MSCCL_FORCE_FULLOPS=1
This commit is contained in:
Wenkai Du
2023-11-08 20:14:28 -08:00
committed by GitHub
parent 8e0258a73d
commit 5a800e00cd
5 changed files with 69 additions and 12 deletions
+1
View File
@@ -36,6 +36,7 @@ struct mscclSchedulerParam {
int nRanks;
bool scheduled;
mscclAlgoHandle_t handle;
uint64_t opCount;
};
typedef struct {
+1 -1
View File
@@ -239,7 +239,7 @@ struct mscclWork {
int nChunksPerLoop;
bool hasReduce;
bool redOpArgIsPtr;
uint32_t pad[1];
uint32_t fnIndex;
};
static_assert(sizeof(struct mscclWork) % 16 == 0, "mscclWork needs to be 16B aligned");