Skip to content

Commit 92376c3

Browse files
shiltianshiltian
authored
[Offload][OMPX] Add the runtime support for multi-dim grid and block (llvm#118042)
1 parent 2f4eac6 commit 92376c3

File tree

13 files changed

+168
-95
lines changed

13 files changed

+168
-95
lines changed

‎offload/plugins-nextgen/amdgpu/src/rtl.cpp‎

Lines changed: 38 additions & 31 deletions
Original file line numberDiff line numberDiff line change
@@ -559,15 +559,15 @@ struct AMDGPUKernelTy : public GenericKernelTy {
559559
}
560560

561561
/// Launch the AMDGPU kernel function.
562-
Error launchImpl(GenericDeviceTy &GenericDevice, uint32_t NumThreads,
563-
uint64_t NumBlocks, KernelArgsTy &KernelArgs,
562+
Error launchImpl(GenericDeviceTy &GenericDevice, uint32_t NumThreads[3],
563+
uint32_t NumBlocks[3], KernelArgsTy &KernelArgs,
564564
KernelLaunchParamsTy LaunchParams,
565565
AsyncInfoWrapperTy &AsyncInfoWrapper) const override;
566566

567567
/// Print more elaborate kernel launch info for AMDGPU
568568
Error printLaunchInfoDetails(GenericDeviceTy &GenericDevice,
569-
KernelArgsTy &KernelArgs, uint32_t NumThreads,
570-
uint64_t NumBlocks) const override;
569+
KernelArgsTy &KernelArgs, uint32_t NumThreads[3],
570+
uint32_t NumBlocks[3]) const override;
571571

572572
/// Get group and private segment kernel size.
573573
uint32_t getGroupSize() const { return GroupSize; }
@@ -719,7 +719,7 @@ struct AMDGPUQueueTy {
719719
/// Push a kernel launch to the queue. The kernel launch requires an output
720720
/// signal and can define an optional input signal (nullptr if none).
721721
Error pushKernelLaunch(const AMDGPUKernelTy &Kernel, void *KernelArgs,
722-
uint32_t NumThreads, uint64_t NumBlocks,
722+
uint32_t NumThreads[3], uint32_t NumBlocks[3],
723723
uint32_t GroupSize, uint64_t StackSize,
724724
AMDGPUSignalTy *OutputSignal,
725725
AMDGPUSignalTy *InputSignal) {
@@ -746,14 +746,18 @@ struct AMDGPUQueueTy {
746746
assert(Packet && "Invalid packet");
747747

748748
// The first 32 bits of the packet are written after the other fields
749-
uint16_t Setup = UINT16_C(1) << HSA_KERNEL_DISPATCH_PACKET_SETUP_DIMENSIONS;
750-
Packet->workgroup_size_x = NumThreads;
751-
Packet->workgroup_size_y = 1;
752-
Packet->workgroup_size_z = 1;
749+
uint16_t Dims = NumBlocks[2] * NumThreads[2] > 1
750+
? 3
751+
: 1 + (NumBlocks[1] * NumThreads[1] != 1);
752+
uint16_t Setup = UINT16_C(Dims)
753+
<< HSA_KERNEL_DISPATCH_PACKET_SETUP_DIMENSIONS;
754+
Packet->workgroup_size_x = NumThreads[0];
755+
Packet->workgroup_size_y = NumThreads[1];
756+
Packet->workgroup_size_z = NumThreads[2];
753757
Packet->reserved0 = 0;
754-
Packet->grid_size_x = NumBlocks * NumThreads;
755-
Packet->grid_size_y = 1;
756-
Packet->grid_size_z = 1;
758+
Packet->grid_size_x = NumBlocks[0] * NumThreads[0];
759+
Packet->grid_size_y = NumBlocks[1] * NumThreads[1];
760+
Packet->grid_size_z = NumBlocks[2] * NumThreads[2];
757761
Packet->private_segment_size =
758762
Kernel.usesDynamicStack() ? StackSize : Kernel.getPrivateSize();
759763
Packet->group_segment_size = GroupSize;
@@ -1240,7 +1244,7 @@ struct AMDGPUStreamTy {
12401244
/// the kernel finalizes. Once the kernel is finished, the stream will release
12411245
/// the kernel args buffer to the specified memory manager.
12421246
Error pushKernelLaunch(const AMDGPUKernelTy &Kernel, void *KernelArgs,
1243-
uint32_t NumThreads, uint64_t NumBlocks,
1247+
uint32_t NumThreads[3], uint32_t NumBlocks[3],
12441248
uint32_t GroupSize, uint64_t StackSize,
12451249
AMDGPUMemoryManagerTy &MemoryManager) {
12461250
if (Queue == nullptr)
@@ -2827,10 +2831,10 @@ struct AMDGPUDeviceTy : public GenericDeviceTy, AMDGenericDeviceTy {
28272831
AsyncInfoWrapperTy AsyncInfoWrapper(*this, nullptr);
28282832

28292833
KernelArgsTy KernelArgs = {};
2830-
if (auto Err =
2831-
AMDGPUKernel.launchImpl(*this, /*NumThread=*/1u,
2832-
/*NumBlocks=*/1ul, KernelArgs,
2833-
KernelLaunchParamsTy{}, AsyncInfoWrapper))
2834+
uint32_t NumBlocksAndThreads[3] = {1u, 1u, 1u};
2835+
if (auto Err = AMDGPUKernel.launchImpl(
2836+
*this, NumBlocksAndThreads, NumBlocksAndThreads, KernelArgs,
2837+
KernelLaunchParamsTy{}, AsyncInfoWrapper))
28342838
return Err;
28352839

28362840
Error Err = Plugin::success();
@@ -3328,7 +3332,7 @@ struct AMDGPUPluginTy final : public GenericPluginTy {
33283332
};
33293333

33303334
Error AMDGPUKernelTy::launchImpl(GenericDeviceTy &GenericDevice,
3331-
uint32_t NumThreads, uint64_t NumBlocks,
3335+
uint32_t NumThreads[3], uint32_t NumBlocks[3],
33323336
KernelArgsTy &KernelArgs,
33333337
KernelLaunchParamsTy LaunchParams,
33343338
AsyncInfoWrapperTy &AsyncInfoWrapper) const {
@@ -3385,13 +3389,15 @@ Error AMDGPUKernelTy::launchImpl(GenericDeviceTy &GenericDevice,
33853389
// Only COV5 implicitargs needs to be set. COV4 implicitargs are not used.
33863390
if (ImplArgs &&
33873391
getImplicitArgsSize() == sizeof(hsa_utils::AMDGPUImplicitArgsTy)) {
3388-
ImplArgs->BlockCountX = NumBlocks;
3389-
ImplArgs->BlockCountY = 1;
3390-
ImplArgs->BlockCountZ = 1;
3391-
ImplArgs->GroupSizeX = NumThreads;
3392-
ImplArgs->GroupSizeY = 1;
3393-
ImplArgs->GroupSizeZ = 1;
3394-
ImplArgs->GridDims = 1;
3392+
ImplArgs->BlockCountX = NumBlocks[0];
3393+
ImplArgs->BlockCountY = NumBlocks[1];
3394+
ImplArgs->BlockCountZ = NumBlocks[2];
3395+
ImplArgs->GroupSizeX = NumThreads[0];
3396+
ImplArgs->GroupSizeY = NumThreads[1];
3397+
ImplArgs->GroupSizeZ = NumThreads[2];
3398+
ImplArgs->GridDims = NumBlocks[2] * NumThreads[2] > 1
3399+
? 3
3400+
: 1 + (NumBlocks[1] * NumThreads[1] != 1);
33953401
ImplArgs->DynamicLdsSize = KernelArgs.DynCGroupMem;
33963402
}
33973403

@@ -3402,8 +3408,8 @@ Error AMDGPUKernelTy::launchImpl(GenericDeviceTy &GenericDevice,
34023408

34033409
Error AMDGPUKernelTy::printLaunchInfoDetails(GenericDeviceTy &GenericDevice,
34043410
KernelArgsTy &KernelArgs,
3405-
uint32_t NumThreads,
3406-
uint64_t NumBlocks) const {
3411+
uint32_t NumThreads[3],
3412+
uint32_t NumBlocks[3]) const {
34073413
// Only do all this when the output is requested
34083414
if (!(getInfoLevel() & OMP_INFOTYPE_PLUGIN_KERNEL))
34093415
return Plugin::success();
@@ -3440,12 +3446,13 @@ Error AMDGPUKernelTy::printLaunchInfoDetails(GenericDeviceTy &GenericDevice,
34403446
// S/VGPR Spill Count: how many S/VGPRs are spilled by the kernel
34413447
// Tripcount: loop tripcount for the kernel
34423448
INFO(OMP_INFOTYPE_PLUGIN_KERNEL, GenericDevice.getDeviceId(),
3443-
"#Args: %d Teams x Thrds: %4lux%4u (MaxFlatWorkGroupSize: %u) LDS "
3449+
"#Args: %d Teams x Thrds: %4ux%4u (MaxFlatWorkGroupSize: %u) LDS "
34443450
"Usage: %uB #SGPRs/VGPRs: %u/%u #SGPR/VGPR Spills: %u/%u Tripcount: "
34453451
"%lu\n",
3446-
ArgNum, NumGroups, ThreadsPerGroup, MaxFlatWorkgroupSize,
3447-
GroupSegmentSize, SGPRCount, VGPRCount, SGPRSpillCount, VGPRSpillCount,
3448-
LoopTripCount);
3452+
ArgNum, NumGroups[0] * NumGroups[1] * NumGroups[2],
3453+
ThreadsPerGroup[0] * ThreadsPerGroup[1] * ThreadsPerGroup[2],
3454+
MaxFlatWorkgroupSize, GroupSegmentSize, SGPRCount, VGPRCount,
3455+
SGPRSpillCount, VGPRSpillCount, LoopTripCount);
34493456

34503457
return Plugin::success();
34513458
}

‎offload/plugins-nextgen/common/include/PluginInterface.h‎

Lines changed: 8 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -269,8 +269,9 @@ struct GenericKernelTy {
269269
Error launch(GenericDeviceTy &GenericDevice, void **ArgPtrs,
270270
ptrdiff_t *ArgOffsets, KernelArgsTy &KernelArgs,
271271
AsyncInfoWrapperTy &AsyncInfoWrapper) const;
272-
virtual Error launchImpl(GenericDeviceTy &GenericDevice, uint32_t NumThreads,
273-
uint64_t NumBlocks, KernelArgsTy &KernelArgs,
272+
virtual Error launchImpl(GenericDeviceTy &GenericDevice,
273+
uint32_t NumThreads[3], uint32_t NumBlocks[3],
274+
KernelArgsTy &KernelArgs,
274275
KernelLaunchParamsTy LaunchParams,
275276
AsyncInfoWrapperTy &AsyncInfoWrapper) const = 0;
276277

@@ -320,15 +321,15 @@ struct GenericKernelTy {
320321

321322
/// Prints generic kernel launch information.
322323
Error printLaunchInfo(GenericDeviceTy &GenericDevice,
323-
KernelArgsTy &KernelArgs, uint32_t NumThreads,
324-
uint64_t NumBlocks) const;
324+
KernelArgsTy &KernelArgs, uint32_t NumThreads[3],
325+
uint32_t NumBlocks[3]) const;
325326

326327
/// Prints plugin-specific kernel launch information after generic kernel
327328
/// launch information
328329
virtual Error printLaunchInfoDetails(GenericDeviceTy &GenericDevice,
329330
KernelArgsTy &KernelArgs,
330-
uint32_t NumThreads,
331-
uint64_t NumBlocks) const;
331+
uint32_t NumThreads[3],
332+
uint32_t NumBlocks[3]) const;
332333

333334
private:
334335
/// Prepare the arguments before launching the kernel.
@@ -347,7 +348,7 @@ struct GenericKernelTy {
347348
/// The number of threads \p NumThreads can be adjusted by this method.
348349
/// \p IsNumThreadsFromUser is true is \p NumThreads is defined by user via
349350
/// thread_limit clause.
350-
uint64_t getNumBlocks(GenericDeviceTy &GenericDevice,
351+
uint32_t getNumBlocks(GenericDeviceTy &GenericDevice,
351352
uint32_t BlockLimitClause[3], uint64_t LoopTripCount,
352353
uint32_t &NumThreads, bool IsNumThreadsFromUser) const;
353354

‎offload/plugins-nextgen/common/src/PluginInterface.cpp‎

Lines changed: 27 additions & 21 deletions
Original file line numberDiff line numberDiff line change
@@ -526,20 +526,21 @@ GenericKernelTy::getKernelLaunchEnvironment(
526526

527527
Error GenericKernelTy::printLaunchInfo(GenericDeviceTy &GenericDevice,
528528
KernelArgsTy &KernelArgs,
529-
uint32_t NumThreads,
530-
uint64_t NumBlocks) const {
529+
uint32_t NumThreads[3],
530+
uint32_t NumBlocks[3]) const {
531531
INFO(OMP_INFOTYPE_PLUGIN_KERNEL, GenericDevice.getDeviceId(),
532-
"Launching kernel %s with %" PRIu64
533-
" blocks and %d threads in %s mode\n",
534-
getName(), NumBlocks, NumThreads, getExecutionModeName());
532+
"Launching kernel %s with [%u,%u,%u] blocks and [%u,%u,%u] threads in "
533+
"%s mode\n",
534+
getName(), NumBlocks[0], NumBlocks[1], NumBlocks[2], NumThreads[0],
535+
NumThreads[1], NumThreads[2], getExecutionModeName());
535536
return printLaunchInfoDetails(GenericDevice, KernelArgs, NumThreads,
536537
NumBlocks);
537538
}
538539

539540
Error GenericKernelTy::printLaunchInfoDetails(GenericDeviceTy &GenericDevice,
540541
KernelArgsTy &KernelArgs,
541-
uint32_t NumThreads,
542-
uint64_t NumBlocks) const {
542+
uint32_t NumThreads[3],
543+
uint32_t NumBlocks[3]) const {
543544
return Plugin::success();
544545
}
545546

@@ -566,10 +567,16 @@ Error GenericKernelTy::launch(GenericDeviceTy &GenericDevice, void **ArgPtrs,
566567
Args, Ptrs, *KernelLaunchEnvOrErr);
567568
}
568569

569-
uint32_t NumThreads = getNumThreads(GenericDevice, KernelArgs.ThreadLimit);
570-
uint64_t NumBlocks =
571-
getNumBlocks(GenericDevice, KernelArgs.NumTeams, KernelArgs.Tripcount,
572-
NumThreads, KernelArgs.ThreadLimit[0] > 0);
570+
uint32_t NumThreads[3] = {KernelArgs.ThreadLimit[0],
571+
KernelArgs.ThreadLimit[1],
572+
KernelArgs.ThreadLimit[2]};
573+
uint32_t NumBlocks[3] = {KernelArgs.NumTeams[0], KernelArgs.NumTeams[1],
574+
KernelArgs.NumTeams[2]};
575+
if (!IsBareKernel) {
576+
NumThreads[0] = getNumThreads(GenericDevice, NumThreads);
577+
NumBlocks[0] = getNumBlocks(GenericDevice, NumBlocks, KernelArgs.Tripcount,
578+
NumThreads[0], KernelArgs.ThreadLimit[0] > 0);
579+
}
573580

574581
// Record the kernel description after we modified the argument count and num
575582
// blocks/threads.
@@ -578,7 +585,8 @@ Error GenericKernelTy::launch(GenericDeviceTy &GenericDevice, void **ArgPtrs,
578585
RecordReplay.saveImage(getName(), getImage());
579586
RecordReplay.saveKernelInput(getName(), getImage());
580587
RecordReplay.saveKernelDescr(getName(), LaunchParams, KernelArgs.NumArgs,
581-
NumBlocks, NumThreads, KernelArgs.Tripcount);
588+
NumBlocks[0], NumThreads[0],
589+
KernelArgs.Tripcount);
582590
}
583591

584592
if (auto Err =
@@ -618,11 +626,10 @@ KernelLaunchParamsTy GenericKernelTy::prepareArgs(
618626

619627
uint32_t GenericKernelTy::getNumThreads(GenericDeviceTy &GenericDevice,
620628
uint32_t ThreadLimitClause[3]) const {
621-
assert(ThreadLimitClause[1] == 0 && ThreadLimitClause[2] == 0 &&
622-
"Multi dimensional launch not supported yet.");
629+
assert(!IsBareKernel && "bare kernel should not call this function");
623630

624-
if (IsBareKernel && ThreadLimitClause[0] > 0)
625-
return ThreadLimitClause[0];
631+
assert(ThreadLimitClause[1] == 1 && ThreadLimitClause[2] == 1 &&
632+
"Multi dimensional launch not supported yet.");
626633

627634
if (ThreadLimitClause[0] > 0 && isGenericMode())
628635
ThreadLimitClause[0] += GenericDevice.getWarpSize();
@@ -632,16 +639,15 @@ uint32_t GenericKernelTy::getNumThreads(GenericDeviceTy &GenericDevice,
632639
: PreferredNumThreads);
633640
}
634641

635-
uint64_t GenericKernelTy::getNumBlocks(GenericDeviceTy &GenericDevice,
642+
uint32_t GenericKernelTy::getNumBlocks(GenericDeviceTy &GenericDevice,
636643
uint32_t NumTeamsClause[3],
637644
uint64_t LoopTripCount,
638645
uint32_t &NumThreads,
639646
bool IsNumThreadsFromUser) const {
640-
assert(NumTeamsClause[1] == 0 && NumTeamsClause[2] == 0 &&
641-
"Multi dimensional launch not supported yet.");
647+
assert(!IsBareKernel && "bare kernel should not call this function");
642648

643-
if (IsBareKernel && NumTeamsClause[0] > 0)
644-
return NumTeamsClause[0];
649+
assert(NumTeamsClause[1] == 1 && NumTeamsClause[2] == 1 &&
650+
"Multi dimensional launch not supported yet.");
645651

646652
if (NumTeamsClause[0] > 0) {
647653
// TODO: We need to honor any value and consequently allow more than the

‎offload/plugins-nextgen/cuda/src/rtl.cpp‎

Lines changed: 9 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -149,8 +149,8 @@ struct CUDAKernelTy : public GenericKernelTy {
149149
}
150150

151151
/// Launch the CUDA kernel function.
152-
Error launchImpl(GenericDeviceTy &GenericDevice, uint32_t NumThreads,
153-
uint64_t NumBlocks, KernelArgsTy &KernelArgs,
152+
Error launchImpl(GenericDeviceTy &GenericDevice, uint32_t NumThreads[3],
153+
uint32_t NumBlocks[3], KernelArgsTy &KernelArgs,
154154
KernelLaunchParamsTy LaunchParams,
155155
AsyncInfoWrapperTy &AsyncInfoWrapper) const override;
156156

@@ -1228,10 +1228,10 @@ struct CUDADeviceTy : public GenericDeviceTy {
12281228
AsyncInfoWrapperTy AsyncInfoWrapper(*this, nullptr);
12291229

12301230
KernelArgsTy KernelArgs = {};
1231-
if (auto Err =
1232-
CUDAKernel.launchImpl(*this, /*NumThread=*/1u,
1233-
/*NumBlocks=*/1ul, KernelArgs,
1234-
KernelLaunchParamsTy{}, AsyncInfoWrapper))
1231+
uint32_t NumBlocksAndThreads[3] = {1u, 1u, 1u};
1232+
if (auto Err = CUDAKernel.launchImpl(
1233+
*this, NumBlocksAndThreads, NumBlocksAndThreads, KernelArgs,
1234+
KernelLaunchParamsTy{}, AsyncInfoWrapper))
12351235
return Err;
12361236

12371237
Error Err = Plugin::success();
@@ -1274,7 +1274,7 @@ struct CUDADeviceTy : public GenericDeviceTy {
12741274
};
12751275

12761276
Error CUDAKernelTy::launchImpl(GenericDeviceTy &GenericDevice,
1277-
uint32_t NumThreads, uint64_t NumBlocks,
1277+
uint32_t NumThreads[3], uint32_t NumBlocks[3],
12781278
KernelArgsTy &KernelArgs,
12791279
KernelLaunchParamsTy LaunchParams,
12801280
AsyncInfoWrapperTy &AsyncInfoWrapper) const {
@@ -1292,9 +1292,8 @@ Error CUDAKernelTy::launchImpl(GenericDeviceTy &GenericDevice,
12921292
reinterpret_cast<void *>(&LaunchParams.Size),
12931293
CU_LAUNCH_PARAM_END};
12941294

1295-
CUresult Res = cuLaunchKernel(Func, NumBlocks, /*gridDimY=*/1,
1296-
/*gridDimZ=*/1, NumThreads,
1297-
/*blockDimY=*/1, /*blockDimZ=*/1,
1295+
CUresult Res = cuLaunchKernel(Func, NumBlocks[0], NumBlocks[1], NumBlocks[2],
1296+
NumThreads[0], NumThreads[1], NumThreads[2],
12981297
MaxDynCGroupMem, Stream, nullptr, Config);
12991298
return Plugin::check(Res, "Error in cuLaunchKernel for '%s': %s", getName());
13001299
}

‎offload/plugins-nextgen/host/src/rtl.cpp‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -89,8 +89,8 @@ struct GenELF64KernelTy : public GenericKernelTy {
8989
}
9090

9191
/// Launch the kernel using the libffi.
92-
Error launchImpl(GenericDeviceTy &GenericDevice, uint32_t NumThreads,
93-
uint64_t NumBlocks, KernelArgsTy &KernelArgs,
92+
Error launchImpl(GenericDeviceTy &GenericDevice, uint32_t NumThreads[3],
93+
uint32_t NumBlocks[3], KernelArgsTy &KernelArgs,
9494
KernelLaunchParamsTy LaunchParams,
9595
AsyncInfoWrapperTy &AsyncInfoWrapper) const override {
9696
// Create a vector of ffi_types, one per argument.

‎offload/src/interface.cpp‎

Lines changed: 15 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -284,14 +284,25 @@ static KernelArgsTy *upgradeKernelArgs(KernelArgsTy *KernelArgs,
284284
LocalKernelArgs.Flags = KernelArgs->Flags;
285285
LocalKernelArgs.DynCGroupMem = 0;
286286
LocalKernelArgs.NumTeams[0] = NumTeams;
287-
LocalKernelArgs.NumTeams[1] = 0;
288-
LocalKernelArgs.NumTeams[2] = 0;
287+
LocalKernelArgs.NumTeams[1] = 1;
288+
LocalKernelArgs.NumTeams[2] = 1;
289289
LocalKernelArgs.ThreadLimit[0] = ThreadLimit;
290-
LocalKernelArgs.ThreadLimit[1] = 0;
291-
LocalKernelArgs.ThreadLimit[2] = 0;
290+
LocalKernelArgs.ThreadLimit[1] = 1;
291+
LocalKernelArgs.ThreadLimit[2] = 1;
292292
return &LocalKernelArgs;
293293
}
294294

295+
// FIXME: This is a WA to "calibrate" the bad work done in the front end.
296+
// Delete this ugly code after the front end emits proper values.
297+
auto CorrectMultiDim = [](uint32_t(&Val)[3]) {
298+
if (Val[1] == 0)
299+
Val[1] = 1;
300+
if (Val[2] == 0)
301+
Val[2] = 1;
302+
};
303+
CorrectMultiDim(KernelArgs->ThreadLimit);
304+
CorrectMultiDim(KernelArgs->NumTeams);
305+
295306
return KernelArgs;
296307
}
297308

@@ -320,12 +331,6 @@ static inline int targetKernel(ident_t *Loc, int64_t DeviceId, int32_t NumTeams,
320331
KernelArgs =
321332
upgradeKernelArgs(KernelArgs, LocalKernelArgs, NumTeams, ThreadLimit);
322333

323-
assert(KernelArgs->NumTeams[0] == static_cast<uint32_t>(NumTeams) &&
324-
!KernelArgs->NumTeams[1] && !KernelArgs->NumTeams[2] &&
325-
"OpenMP interface should not use multiple dimensions");
326-
assert(KernelArgs->ThreadLimit[0] == static_cast<uint32_t>(ThreadLimit) &&
327-
!KernelArgs->ThreadLimit[1] && !KernelArgs->ThreadLimit[2] &&
328-
"OpenMP interface should not use multiple dimensions");
329334
TIMESCOPE_WITH_DETAILS_AND_IDENT(
330335
"Runtime: target exe",
331336
"NumTeams=" + std::to_string(NumTeams) +

‎offload/src/omptarget.cpp‎

Lines changed: 0 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1451,8 +1451,6 @@ int target(ident_t *Loc, DeviceTy &Device, void *HostPtr,
14511451
Loc);
14521452

14531453
#ifdef OMPT_SUPPORT
1454-
assert(KernelArgs.NumTeams[1] == 0 && KernelArgs.NumTeams[2] == 0 &&
1455-
"Multi dimensional launch not supported yet.");
14561454
/// RAII to establish tool anchors before and after kernel launch
14571455
int32_t NumTeams = KernelArgs.NumTeams[0];
14581456
// No need to guard this with OMPT_IF_BUILT

0 commit comments

Comments
 (0)