AI/rocm-systems
2
0
Forka 0
Codice Problemi Pull Request Actions Pacchetti Progetti Rilasci Wiki Attività

Revert "changes for hugepages backed host buffer for larger allocations (#1841)" (#1951)

Sfoglia il codice sorgente 
This reverts commit 65b69bf318.
This commit is contained in:
Nilesh M Negi
2025-10-02 23:43:09 -05:00
committato da GitHub
parent 5978d2f9ab
commit 342ec086e3
3 ha cambiato i file con 18 aggiunte e 87 eliminazioni
Scarica il file Patch Scarica il file Diff Expand all files Collapse all files
src/include/alloc.h
+12 -76
Vedi File
@@ -18,10 +18,6 @@
#include <stdlib.h>
#include <string.h>
#include "rccl_vars.h"
#include <unordered_map>
#include <mutex>
#define RCCL_HP_MIN_SIZE 2097152
#if CUDART_VERSION >= 11030
#include <cuda.h>
@@ -35,9 +31,6 @@ constexpr size_t ncclSizeOfT() { return sizeof(T); }
template<>
constexpr size_t ncclSizeOfT<void>() { return 1; }
extern std::unordered_map<void*, size_t> hugepageAllocs;
extern std::mutex hugepageAllocsMutex;
#if CUDART_VERSION >= 12020
static inline ncclResult_t ncclCuMemHostAlloc(void** ptr, CUmemGenericAllocationHandle *handlep, size_t size) {
@@ -112,100 +105,43 @@ static inline ncclResult_t ncclCuMemHostFree(void* ptr) {
}
#endif /* CUDART_VERSION >= 12020 */
template <typename T>
ncclResult_t ncclCudaHostCallocDebug(T** ptr, size_t nelem, const char *filefunc, int line, int hp_request=0 ) {
ncclResult_t ncclCudaHostCallocDebug(T** ptr, size_t nelem, const char *filefunc, int line) {
ncclResult_t result = ncclSuccess;
cudaStreamCaptureMode mode = cudaStreamCaptureModeRelaxed;
*ptr = nullptr;
size_t size = nelem * ncclSizeOfT<T>();
CUDACHECK(cudaThreadExchangeStreamCaptureMode(&mode));
int managed = 0;
int huge=0;
CUDACHECK(hipDeviceGetAttribute(&managed, hipDeviceAttributeDirectManagedMemAccessFromHost, 0));
if (nelem > 0) {
if (managed) {
#if defined(HIP_UNCACHED_MEMORY)
CUDACHECKGOTO(hipExtMallocWithFlags((void**)ptr, size, hipDeviceMallocUncached), result, finish);
CUDACHECKGOTO(hipExtMallocWithFlags((void**)ptr, nelem*ncclSizeOfT<T>(), hipDeviceMallocUncached), result, finish);
#else
CUDACHECKGOTO(hipExtMallocWithFlags((void**)ptr, size, hipDeviceMallocFinegrained), result, finish);
CUDACHECKGOTO(hipExtMallocWithFlags((void**)ptr, nelem*ncclSizeOfT<T>(), hipDeviceMallocFinegrained), result, finish);
#endif
} else {
if (hp_request) {
if (size < RCCL_HP_MIN_SIZE) {
WARN("small size : forcing back to hipHostMalloc");
} else
#if defined(HIP_HOST_UNCACHED_MEMORY)
CUDACHECKGOTO(hipHostMalloc(ptr, size, cudaHostAllocMapped | hipHostMallocUncached), result, finish);
CUDACHECKGOTO(hipHostMalloc(ptr, nelem*ncclSizeOfT<T>(), cudaHostAllocMapped | hipHostMallocUncached), result, finish);
#else
CUDACHECKGOTO(hipHostMalloc(ptr, size, cudaHostAllocMapped), result, finish);
CUDACHECKGOTO(hipHostMalloc(ptr, nelem*ncclSizeOfT<T>(), cudaHostAllocMapped), result, finish);
#endif
memset(*ptr, 0, size);
} else {
// Hugepage allocation via mmap
void* hostPtr = mmap(NULL, size, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB, -1, 0);
if (hostPtr == MAP_FAILED) {
WARN("Hugepage allocation failed. Falling back to hipHostMalloc");
#if defined(HIP_HOST_UNCACHED_MEMORY)
CUDACHECKGOTO(hipHostMalloc(ptr, size, cudaHostAllocMapped | hipHostMallocUncached), result, finish);
#else
CUDACHECKGOTO(hipHostMalloc(ptr, size, cudaHostAllocMapped), result, finish);
#endif
memset(*ptr, 0, size);
} else {
memset(hostPtr, 0, size);
CUDACHECKGOTO(hipHostRegister(hostPtr, size, hipHostRegisterMapped), result, finish);
void* devPtr = nullptr;
CUDACHECKGOTO(hipHostGetDevicePointer(&devPtr, hostPtr, 0), result, finish);
*ptr = reinterpret_cast<T*>(hostPtr);
INFO(NCCL_ALLOC, "Cuda Host Alloc Size done using hugepages");
huge=1;
std::lock_guard<std::mutex> lock(hugepageAllocsMutex);
hugepageAllocs[hostPtr] = size;
for (auto &kv : hugepageAllocs) INFO(NCCL_ALLOC, "updated Hugepage alloc ptr %p size %zu", kv.first, kv.second);
}
}
} else {
#if defined(HIP_HOST_UNCACHED_MEMORY)
CUDACHECKGOTO(hipHostMalloc(ptr, size, cudaHostAllocMapped | hipHostMallocUncached), result, finish);
#else
CUDACHECKGOTO(hipHostMalloc(ptr, size, cudaHostAllocMapped), result, finish);
#endif
memset(*ptr, 0, size);
}
}
memset(*ptr, 0, nelem*ncclSizeOfT<T>());
}
finish:
CUDACHECK(cudaThreadExchangeStreamCaptureMode(&mode));
if (*ptr == nullptr && nelem > 0) WARN("Failed to CUDA host alloc %ld bytes", size);
INFO(NCCL_ALLOC, "%s:%d Cuda Host Alloc Size %ld pointer %p hp_request %d managed %d hugepage_alloc %d", filefunc, line, size, *ptr, hp_request, managed, huge);
if (*ptr == nullptr && nelem > 0) WARN("Failed to CUDA host alloc %ld bytes", nelem*ncclSizeOfT<T>());
INFO(NCCL_ALLOC, "%s:%d Cuda Host Alloc Size %ld pointer %p", filefunc, line, nelem*ncclSizeOfT<T>(), *ptr);
return result;
}
static inline ncclResult_t ncclCudaHostFree(void* ptr, size_t alloc_size=0, int hp_request=0) {
if (hp_request) {
if (alloc_size > 0) {
std::lock_guard<std::mutex> lock(hugepageAllocsMutex);
// for (auto &kv : hugepageAllocs) INFO(NCCL_ALLOC, "Hugepage alloc ptr %p size %zu", kv.first, kv.second);
auto it = hugepageAllocs.find(ptr);
if (it != hugepageAllocs.end()) {
// INFO(NCCL_ALLOC, "%s:%d Cuda Host HugePage unmap size %ld pointer %p app_tracked_size %ld", __FILE__, __LINE__, it->second, ptr, alloc_size);
hipHostUnregister(ptr);
munmap(ptr, it->second);
hugepageAllocs.erase(it);
return ncclSuccess;
}
}
INFO(NCCL_ALLOC, "Cudafree being done to %p, size=%ld", ptr,alloc_size);
}
static inline ncclResult_t ncclCudaHostFree(void* ptr) {
CUDACHECK(cudaFreeHost(ptr));
return ncclSuccess;
}
#define ncclCudaHostCalloc(...) ncclCudaHostCallocDebug(__VA_ARGS__, __FILE__, __LINE__, 0)
#define ncclCudaHostCalloc(...) ncclCudaHostCallocDebug(__VA_ARGS__, __FILE__, __LINE__)
template <typename T>
ncclResult_t ncclCallocDebug(T** ptr, size_t nelem, const char *filefunc, int line) {
src/init.cc
-2
Vedi File
@@ -95,8 +95,6 @@ NCCL_PARAM(NvlsChannels, "NVLS_NCHANNELS", NCCL_CONFIG_UNDEF_INT);
struct allocationTracker allocTracker[MAX_ALLOC_TRACK_NGPU] = {};
static ncclResult_t commReclaim(ncclComm_t comm);
std::unordered_map<void*, size_t> hugepageAllocs;
std::mutex hugepageAllocsMutex;
#ifdef ENABLE_MSCCLPP
size_t std::hash<ncclUniqueId>::operator ()(const ncclUniqueId& uniqueId) const noexcept {
src/transport/net.cc
+6 -9
Vedi File
@@ -179,7 +179,6 @@ static ncclResult_t canConnect(int* ret, struct ncclComm* comm, struct ncclTopoG
NCCL_PARAM(NetSharedBuffers, "NET_SHARED_BUFFERS", -2);
NCCL_PARAM(NetSharedComms, "NET_SHARED_COMMS", 1);
RCCL_PARAM(NetHostBufferHugePageAlloc, "NET_HOST_BUFFER_HUGE_PAGE_ALLOC", 0);
#if defined(HIP_CONTIGUOUS_MEMORY)
RCCL_PARAM(NetContiguousMem, "NET_CONTIGUOUS_MEM", 0);
#endif
@@ -603,7 +602,7 @@ static ncclResult_t sharedNetBuffersInit(struct ncclProxyState* proxyState, int
}
}
if (!cuda && state->hostBuff == NULL) {
NCCLCHECK(ncclCudaHostCallocDebug(&state->hostBuff, state->size, __FILE__, __LINE__, rcclParamNetHostBufferHugePageAlloc()));
NCCLCHECK(ncclCudaHostCalloc(&state->hostBuff, state->size));
}
if (cpuPtr) *cpuPtr = cuda ? state->cudaBuff : state->hostBuff;
if (gpuPtr) *gpuPtr = (cpuPtr && sameProcess) ? *cpuPtr : NULL;
@@ -632,9 +631,7 @@ static ncclResult_t sharedNetBuffersDestroy(struct ncclProxyState* proxyState, i
}
NCCLCHECK(ncclCudaFree(state->cudaBuff));
}
if (state->hostBuff) {
NCCLCHECK(ncclCudaHostFree(state->hostBuff, (state->size)*(sizeof(int64_t)), rcclParamNetHostBufferHugePageAlloc()));
}
if (state->hostBuff) NCCLCHECK(ncclCudaHostFree(state->hostBuff));
}
if (peer->send.refcount || peer->recv.refcount) return ncclSuccess;
@@ -891,7 +888,7 @@ static ncclResult_t sendProxyConnect(struct ncclProxyConnection* connection, str
}
}
if (map->sameProcess) {
NCCLCHECK(ncclCudaHostCallocDebug(&map->mems[NCCL_NET_MAP_HOSTMEM].cpuPtr, map->mems[NCCL_NET_MAP_HOSTMEM].size, __FILE__, __LINE__, rcclParamNetHostBufferHugePageAlloc()));
NCCLCHECK(ncclCudaHostCalloc(&map->mems[NCCL_NET_MAP_HOSTMEM].cpuPtr, map->mems[NCCL_NET_MAP_HOSTMEM].size));
map->mems[NCCL_NET_MAP_HOSTMEM].gpuPtr = map->mems[NCCL_NET_MAP_HOSTMEM].cpuPtr;
} else {
NCCLCHECK(netCreateShm(proxyState, map->mems+NCCL_NET_MAP_HOSTMEM));
@@ -1093,7 +1090,7 @@ static ncclResult_t recvProxyConnect(struct ncclProxyConnection* connection, str
map->mems[NCCL_NET_MAP_DEVMEM].cpuPtr = map->mems[NCCL_NET_MAP_DEVMEM].gpuPtr;
}
}
NCCLCHECK(ncclCudaHostCallocDebug(&map->mems[NCCL_NET_MAP_HOSTMEM].cpuPtr, map->mems[NCCL_NET_MAP_HOSTMEM].size, __FILE__, __LINE__, rcclParamNetHostBufferHugePageAlloc()));
NCCLCHECK(ncclCudaHostCalloc(&map->mems[NCCL_NET_MAP_HOSTMEM].cpuPtr, map->mems[NCCL_NET_MAP_HOSTMEM].size));
map->mems[NCCL_NET_MAP_HOSTMEM].gpuPtr = map->mems[NCCL_NET_MAP_HOSTMEM].cpuPtr;
if (ncclGdrCopy && map->sameProcess) {
uint64_t *cpuPtr, *gpuPtr;
@@ -1168,7 +1165,7 @@ static ncclResult_t sendProxyFree(struct ncclProxyConnection* connection, struct
}
struct connectMapMem* mems = resources->map.mems;
if (resources->map.sameProcess) {
NCCLCHECK(ncclCudaHostFree(mems[NCCL_NET_MAP_HOSTMEM].cpuPtr, (mems[NCCL_NET_MAP_HOSTMEM].size)*(sizeof(int)), rcclParamNetHostBufferHugePageAlloc()));
NCCLCHECK(ncclCudaHostFree(mems[NCCL_NET_MAP_HOSTMEM].cpuPtr));
} else {
NCCLCHECK(ncclShmIpcClose(&mems[NCCL_NET_MAP_HOSTMEM].createDesc));
}
@@ -1212,7 +1209,7 @@ static ncclResult_t recvProxyFree(struct ncclProxyConnection* connection, struct
}
}
struct connectMapMem* mems = resources->map.mems;
NCCLCHECK(ncclCudaHostFree(mems[NCCL_NET_MAP_HOSTMEM].cpuPtr, (mems[NCCL_NET_MAP_HOSTMEM].size)*(sizeof(int)), rcclParamNetHostBufferHugePageAlloc()));
NCCLCHECK(ncclCudaHostFree(mems[NCCL_NET_MAP_HOSTMEM].cpuPtr));
NCCLCHECK(ncclCudaFree(mems[NCCL_NET_MAP_DEVMEM].cpuPtr));
if (!resources->map.sameProcess || ncclCuMemEnable()) {
// cuMem API support