Skip to content

Commit

Permalink
turn log to more informative comment
Browse files Browse the repository at this point in the history
  • Loading branch information
Masahiro Masuda committed Dec 5, 2020
1 parent 66d28ec commit 1937c1b
Showing 1 changed file with 4 additions and 2 deletions.
6 changes: 4 additions & 2 deletions src/auto_scheduler/search_task.cc
Original file line number Diff line number Diff line change
Expand Up @@ -70,8 +70,10 @@ HardwareParams HardwareParamsNode::GetDefaultHardwareParams(const Target& target
device_api->GetAttr(ctx, tvm::runtime::DeviceAttrKind::kMaxRegistersPerBlock, &ret);
int max_registers_per_block = ret;
if (max_registers_per_block == 0) {
LOG(INFO) << "Device API reported kMaxRegistersPerBlock being 0, using the value of "
"kMaxSharedMemoryPerBlock as max_registers_per_block.";
// On ROCm backend, the device API can report kMaxRegistersPerBlock being 0.
// Since CUDA returns a large value similar to the shared memory size for
// kMaxRegistersPerBlock, we also use the shared memory size as a substitute for the maxinum
// number of registers per block in this case.
max_registers_per_block = max_shared_memory_per_block;
}

Expand Down

0 comments on commit 1937c1b

Please sign in to comment.