[OpenMP] Move synchronization into __tgt_async_info

The AsyncInfo should be passed everywhere and it should offer a way to
ensure synchronization, given a libomptarget Device.

This replaces D96431.

Reviewed By: tianshilei1992

Differential Revision: https://reviews.llvm.org/D96438
This commit is contained in:
Johannes Doerfert 2021-02-10 11:06:00 -06:00
parent 942728763b
commit a2fc0d34db
3 changed files with 67 additions and 43 deletions

View File

@ -119,6 +119,8 @@ struct __tgt_target_table {
*EntriesEnd; // End of the table with all the entries (non inclusive)
};
// clang-format on
/// This struct contains information exchanged between different asynchronous
/// operations for device-dependent optimization and potential synchronization
struct __tgt_async_info {
@ -128,6 +130,29 @@ struct __tgt_async_info {
void *Queue = nullptr;
};
struct DeviceTy;
/// The libomptarget wrapper around a __tgt_async_info object directly
/// associated with a libomptarget layer device. RAII semantics to avoid
/// mistakes.
class AsyncInfoTy {
__tgt_async_info AsyncInfo;
DeviceTy &Device;
public:
AsyncInfoTy(DeviceTy &Device) : Device(Device) {}
~AsyncInfoTy() { synchronize(); }
/// Implicit conversion to the __tgt_async_info which is used in the
/// plugin interface.
operator __tgt_async_info *() { return &AsyncInfo; }
/// Synchronize all pending actions.
///
/// \returns OFFLOAD_FAIL or OFFLOAD_SUCCESS appropriately.
int synchronize();
};
/// This struct is a record of non-contiguous information
struct __tgt_target_non_contig {
uint64_t Offset;
@ -135,8 +160,6 @@ struct __tgt_target_non_contig {
uint64_t Stride;
};
// clang-format on
#ifdef __cplusplus
extern "C" {
#endif

View File

@ -19,6 +19,18 @@
#include <cassert>
#include <vector>
int AsyncInfoTy::synchronize() {
int Result = OFFLOAD_SUCCESS;
if (AsyncInfo.Queue) {
// If we have a queue we need to synchronize it now.
Result = Device.synchronize(&AsyncInfo);
assert(AsyncInfo.Queue == nullptr &&
"The device plugin should have nulled the queue to indicate there "
"are no outstanding actions!");
}
return Result;
}
/* All begin addresses for partially mapped structs must be 8-aligned in order
* to ensure proper alignment of members. E.g.
*
@ -248,7 +260,7 @@ int targetDataMapper(ident_t *loc, DeviceTy &Device, void *arg_base, void *arg,
MapperArgsBase.data(), MapperArgs.data(),
MapperArgSizes.data(), MapperArgTypes.data(),
MapperArgNames.data(), /*arg_mappers*/ nullptr,
/*__tgt_async_info*/ nullptr);
/* AsyncInfoTy */ nullptr);
return rc;
}
@ -257,7 +269,7 @@ int targetDataMapper(ident_t *loc, DeviceTy &Device, void *arg_base, void *arg,
int targetDataBegin(ident_t *loc, DeviceTy &Device, int32_t arg_num,
void **args_base, void **args, int64_t *arg_sizes,
int64_t *arg_types, map_var_info_t *arg_names,
void **arg_mappers, __tgt_async_info *AsyncInfo) {
void **arg_mappers, AsyncInfoTy *AsyncInfo) {
// process each input.
for (int32_t i = 0; i < arg_num; ++i) {
// Ignore private variables and arrays - there is no mapping for them.
@ -404,7 +416,7 @@ int targetDataBegin(ident_t *loc, DeviceTy &Device, int32_t arg_num,
DP("Moving %" PRId64 " bytes (hst:" DPxMOD ") -> (tgt:" DPxMOD ")\n",
data_size, DPxPTR(HstPtrBegin), DPxPTR(TgtPtrBegin));
int rt =
Device.submitData(TgtPtrBegin, HstPtrBegin, data_size, AsyncInfo);
Device.submitData(TgtPtrBegin, HstPtrBegin, data_size, *AsyncInfo);
if (rt != OFFLOAD_SUCCESS) {
REPORT("Copying data to device failed.\n");
return OFFLOAD_FAIL;
@ -418,7 +430,7 @@ int targetDataBegin(ident_t *loc, DeviceTy &Device, int32_t arg_num,
uint64_t Delta = (uint64_t)HstPtrBegin - (uint64_t)HstPtrBase;
void *TgtPtrBase = (void *)((uint64_t)TgtPtrBegin - Delta);
int rt = Device.submitData(PointerTgtPtrBegin, &TgtPtrBase,
sizeof(void *), AsyncInfo);
sizeof(void *), *AsyncInfo);
if (rt != OFFLOAD_SUCCESS) {
REPORT("Copying data to device failed.\n");
return OFFLOAD_FAIL;
@ -452,24 +464,13 @@ struct DeallocTgtPtrInfo {
: HstPtrBegin(HstPtr), DataSize(Size), ForceDelete(ForceDelete),
HasCloseModifier(HasCloseModifier) {}
};
/// Synchronize device
static int syncDevice(DeviceTy &Device, __tgt_async_info *AsyncInfo) {
assert(AsyncInfo && AsyncInfo->Queue && "Invalid AsyncInfo");
if (Device.synchronize(AsyncInfo) != OFFLOAD_SUCCESS) {
REPORT("Failed to synchronize device.\n");
return OFFLOAD_FAIL;
}
return OFFLOAD_SUCCESS;
}
} // namespace
/// Internal function to undo the mapping and retrieve the data from the device.
int targetDataEnd(ident_t *loc, DeviceTy &Device, int32_t ArgNum,
void **ArgBases, void **Args, int64_t *ArgSizes,
int64_t *ArgTypes, map_var_info_t *ArgNames,
void **ArgMappers, __tgt_async_info *AsyncInfo) {
void **ArgMappers, AsyncInfoTy *AsyncInfo) {
int Ret;
std::vector<DeallocTgtPtrInfo> DeallocTgtPtrs;
// process each input.
@ -584,7 +585,7 @@ int targetDataEnd(ident_t *loc, DeviceTy &Device, int32_t ArgNum,
DP("Moving %" PRId64 " bytes (tgt:" DPxMOD ") -> (hst:" DPxMOD ")\n",
DataSize, DPxPTR(TgtPtrBegin), DPxPTR(HstPtrBegin));
Ret = Device.retrieveData(HstPtrBegin, TgtPtrBegin, DataSize,
AsyncInfo);
*AsyncInfo);
if (Ret != OFFLOAD_SUCCESS) {
REPORT("Copying data from device failed.\n");
return OFFLOAD_FAIL;
@ -642,8 +643,8 @@ int targetDataEnd(ident_t *loc, DeviceTy &Device, int32_t ArgNum,
// nullptr, there is no data transfer happened because once there is,
// AsyncInfo->Queue will not be nullptr, so again, we don't need to
// synchronize.
if (AsyncInfo && AsyncInfo->Queue) {
Ret = syncDevice(Device, AsyncInfo);
if (AsyncInfo) {
Ret = AsyncInfo->synchronize();
if (Ret != OFFLOAD_SUCCESS)
return OFFLOAD_FAIL;
}
@ -798,7 +799,7 @@ static int getNonContigMergedDimension(__tgt_target_non_contig *NonContig,
int targetDataUpdate(ident_t *loc, DeviceTy &Device, int32_t ArgNum,
void **ArgsBase, void **Args, int64_t *ArgSizes,
int64_t *ArgTypes, map_var_info_t *ArgNames,
void **ArgMappers, __tgt_async_info *AsyncInfo) {
void **ArgMappers, AsyncInfoTy *AsyncInfo) {
// process each input.
for (int32_t I = 0; I < ArgNum; ++I) {
if ((ArgTypes[I] & OMP_TGT_MAPTYPE_LITERAL) ||
@ -948,8 +949,8 @@ class PrivateArgumentManagerTy {
/// A reference to the \p DeviceTy object
DeviceTy &Device;
/// A pointer to a \p __tgt_async_info object
__tgt_async_info *AsyncInfo;
/// A pointer to a \p AsyncInfoTy object
AsyncInfoTy *AsyncInfo;
// TODO: What would be the best value here? Should we make it configurable?
// If the size is larger than this threshold, we will allocate and transfer it
@ -958,7 +959,7 @@ class PrivateArgumentManagerTy {
public:
/// Constructor
PrivateArgumentManagerTy(DeviceTy &Dev, __tgt_async_info *AsyncInfo)
PrivateArgumentManagerTy(DeviceTy &Dev, AsyncInfoTy *AsyncInfo)
: Device(Dev), AsyncInfo(AsyncInfo) {}
/// Add a private argument
@ -985,7 +986,7 @@ public:
#endif
// If first-private, copy data from host
if (IsFirstPrivate) {
int Ret = Device.submitData(TgtPtr, HstPtr, ArgSize, AsyncInfo);
int Ret = Device.submitData(TgtPtr, HstPtr, ArgSize, *AsyncInfo);
if (Ret != OFFLOAD_SUCCESS) {
DP("Copying data to device failed, failed.\n");
return OFFLOAD_FAIL;
@ -1041,7 +1042,7 @@ public:
FirstPrivateArgSize, DPxPTR(TgtPtr));
// Transfer data to target device
int Ret = Device.submitData(TgtPtr, FirstPrivateArgBuffer.data(),
FirstPrivateArgSize, AsyncInfo);
FirstPrivateArgSize, *AsyncInfo);
if (Ret != OFFLOAD_SUCCESS) {
DP("Failed to submit data of private arguments.\n");
return OFFLOAD_FAIL;
@ -1089,7 +1090,7 @@ static int processDataBefore(ident_t *loc, int64_t DeviceId, void *HostPtr,
std::vector<void *> &TgtArgs,
std::vector<ptrdiff_t> &TgtOffsets,
PrivateArgumentManagerTy &PrivateArgumentManager,
__tgt_async_info *AsyncInfo) {
AsyncInfoTy *AsyncInfo) {
TIMESCOPE_WITH_NAME_AND_IDENT("mappingBeforeTargetRegion", loc);
DeviceTy &Device = PM->Devices[DeviceId];
int Ret = targetDataBegin(loc, Device, ArgNum, ArgBases, Args, ArgSizes,
@ -1140,7 +1141,7 @@ static int processDataBefore(ident_t *loc, int64_t DeviceId, void *HostPtr,
DP("Update lambda reference (" DPxMOD ") -> [" DPxMOD "]\n",
DPxPTR(PointerTgtPtrBegin), DPxPTR(TgtPtrBegin));
Ret = Device.submitData(TgtPtrBegin, &PointerTgtPtrBegin,
sizeof(void *), AsyncInfo);
sizeof(void *), *AsyncInfo);
if (Ret != OFFLOAD_SUCCESS) {
REPORT("Copying data to device failed.\n");
return OFFLOAD_FAIL;
@ -1210,7 +1211,7 @@ static int processDataAfter(ident_t *loc, int64_t DeviceId, void *HostPtr,
int64_t *ArgSizes, int64_t *ArgTypes,
map_var_info_t *ArgNames, void **ArgMappers,
PrivateArgumentManagerTy &PrivateArgumentManager,
__tgt_async_info *AsyncInfo) {
AsyncInfoTy *AsyncInfo) {
TIMESCOPE_WITH_NAME_AND_IDENT("mappingAfterTargetRegion", loc);
DeviceTy &Device = PM->Devices[DeviceId];
@ -1242,8 +1243,7 @@ static int processDataAfter(ident_t *loc, int64_t DeviceId, void *HostPtr,
int target(ident_t *loc, DeviceTy &Device, void *HostPtr, int32_t ArgNum,
void **ArgBases, void **Args, int64_t *ArgSizes, int64_t *ArgTypes,
map_var_info_t *ArgNames, void **ArgMappers, int32_t TeamNum,
int32_t ThreadLimit, int IsTeamConstruct,
__tgt_async_info *AsyncInfo) {
int32_t ThreadLimit, int IsTeamConstruct, AsyncInfoTy *AsyncInfo) {
int32_t DeviceId = Device.DeviceID;
TableMap *TM = getTableMap(HostPtr);
@ -1266,7 +1266,7 @@ int target(ident_t *loc, DeviceTy &Device, void *HostPtr, int32_t ArgNum,
// TODO: This will go away as soon as we consequently pass in async info
// objects (as references).
__tgt_async_info InternalAsyncInfo;
AsyncInfoTy InternalAsyncInfo(Device);
if (!AsyncInfo)
AsyncInfo = &InternalAsyncInfo;
@ -1301,10 +1301,10 @@ int target(ident_t *loc, DeviceTy &Device, void *HostPtr, int32_t ArgNum,
if (IsTeamConstruct)
Ret = Device.runTeamRegion(TgtEntryPtr, &TgtArgs[0], &TgtOffsets[0],
TgtArgs.size(), TeamNum, ThreadLimit,
LoopTripCount, AsyncInfo);
LoopTripCount, *AsyncInfo);
else
Ret = Device.runRegion(TgtEntryPtr, &TgtArgs[0], &TgtOffsets[0],
TgtArgs.size(), AsyncInfo);
TgtArgs.size(), *AsyncInfo);
}
if (Ret != OFFLOAD_SUCCESS) {
@ -1322,11 +1322,13 @@ int target(ident_t *loc, DeviceTy &Device, void *HostPtr, int32_t ArgNum,
REPORT("Failed to process data after launching the kernel.\n");
return OFFLOAD_FAIL;
}
} else if (AsyncInfo->Queue) {
} else {
// TODO: We should not synchronize here but on the outer level once we pass
// in a reference AsyncInfo object.
// If ArgNum is zero, but AsyncInfo.Queue is valid, then the kernel doesn't
// hava any argument, and the device supports async operations, so we need a
// sync at this point.
return syncDevice(Device, AsyncInfo);
return AsyncInfo->synchronize();
}
return OFFLOAD_SUCCESS;

View File

@ -23,23 +23,23 @@
extern int targetDataBegin(ident_t *loc, DeviceTy &Device, int32_t arg_num,
void **args_base, void **args, int64_t *arg_sizes,
int64_t *arg_types, map_var_info_t *arg_names,
void **arg_mappers, __tgt_async_info *AsyncInfo);
void **arg_mappers, AsyncInfoTy *AsyncInfo);
extern int targetDataEnd(ident_t *loc, DeviceTy &Device, int32_t ArgNum,
void **ArgBases, void **Args, int64_t *ArgSizes,
int64_t *ArgTypes, map_var_info_t *arg_names,
void **ArgMappers, __tgt_async_info *AsyncInfo);
void **ArgMappers, AsyncInfoTy *AsyncInfo);
extern int targetDataUpdate(ident_t *loc, DeviceTy &Device, int32_t arg_num,
void **args_base, void **args, int64_t *arg_sizes,
int64_t *arg_types, map_var_info_t *arg_names,
void **arg_mappers, __tgt_async_info *AsyncInfo);
void **arg_mappers, AsyncInfoTy *AsyncInfo);
extern int target(ident_t *loc, DeviceTy &Device, void *HostPtr, int32_t ArgNum,
void **ArgBases, void **Args, int64_t *ArgSizes,
int64_t *ArgTypes, map_var_info_t *arg_names,
void **ArgMappers, int32_t TeamNum, int32_t ThreadLimit,
int IsTeamConstruct, __tgt_async_info *AsyncInfo);
int IsTeamConstruct, AsyncInfoTy *AsyncInfo);
extern int CheckDeviceAndCtors(int64_t device_id);
@ -76,8 +76,7 @@ typedef void (*MapperFuncPtrTy)(void *, void *, void *, int64_t, int64_t,
// targetDataEnd and targetDataUpdate).
typedef int (*TargetDataFuncPtrTy)(ident_t *, DeviceTy &, int32_t, void **,
void **, int64_t *, int64_t *,
map_var_info_t *, void **,
__tgt_async_info *);
map_var_info_t *, void **, AsyncInfoTy *);
// Implemented in libomp, they are called from within __tgt_* functions.
#ifdef __cplusplus