Hi:
I am trying to compile a small Tensorflow graph with NMS operator. It works fine with LLVM target, but it fails with CUDA target.
The output log:
Traceback (most recent call last):
File "debug_nms.py", line 65, in <module>
nms_lab()
File "debug_nms.py", line 50, in nms_lab
exe = relay.vm.compile(mod, target=target, params=params)
File "/root/Codes/tvm_in_mac/python/tvm/relay/backend/vm.py", line 69, in compile
compiler.lower(mod, target, target_host)
File "/root/Codes/tvm_in_mac/python/tvm/relay/backend/vm.py", line 135, in lower
self._lower(mod, target, target_host)
File "/root/Codes/tvm_in_mac/python/tvm/_ffi/_ctypes/packed_func.py", line 225, in __call__
raise get_last_ffi_error()
tvm._ffi.base.TVMError: Traceback (most recent call last):
[bt] (8) /root/Codes/tvm_in_mac/build/libtvm.so(tvm::relay::vm::VMFunctionCompiler::EmitInvokeTVMOp(tvm::relay::Function const&, tvm::RelayExpr const&, tvm::RelayExpr const&)+0x91d) [0x7f60fa807d9d]
[bt] (7) /root/Codes/tvm_in_mac/build/libtvm.so(tvm::relay::CompileEngineImpl::Lower(tvm::relay::CCacheKey const&)+0x20) [0x7f60fa7b5b60]
[bt] (6) /root/Codes/tvm_in_mac/build/libtvm.so(tvm::relay::CompileEngineImpl::LowerInternal(tvm::relay::CCacheKey const&)+0x75a) [0x7f60fa7b4c4a]
[bt] (5) /root/Codes/tvm_in_mac/build/libtvm.so(tvm::relay::ScheduleGetter::Create(tvm::relay::Function const&)+0x94b) [0x7f60fa7b399b]
[bt] (4) /root/Codes/tvm_in_mac/build/libtvm.so(tvm::relay::backend::MemoizedExprTranslator<tvm::runtime::Array<tvm::te::Tensor, void> >::VisitExpr(tvm::RelayExpr const&)+0xa6) [0x7f60fa7b7b66]
[bt] (3) /root/Codes/tvm_in_mac/build/libtvm.so(tvm::relay::ExprFunctor<tvm::runtime::Array<tvm::te::Tensor, void> (tvm::RelayExpr const&)>::VisitExpr(tvm::RelayExpr const&)+0x89) [0x7f60fa7b78f9]
[bt] (2) /root/Codes/tvm_in_mac/build/libtvm.so(tvm::relay::ExprFunctor<tvm::runtime::Array<tvm::te::Tensor, void> (tvm::RelayExpr const&)>::InitVTable()::{lambda(tvm::runtime::ObjectRef const&, tvm::relay::ExprFunctor<tvm::runtime::Array<tvm::te::Tensor, void> (tvm::RelayExpr const&)>*)#6}::_FUN(tvm::runtime::ObjectRef const&, tvm::relay::ExprFunctor<tvm::runtime::Array<tvm::te::Tensor, void> (tvm::RelayExpr const&)>*)+0x27) [0x7f60fa7aa107]
[bt] (1) /root/Codes/tvm_in_mac/build/libtvm.so(tvm::relay::ScheduleGetter::VisitExpr_(tvm::relay::CallNode const*)+0x534) [0x7f60fa7b2084]
[bt] (0) /root/Codes/tvm_in_mac/build/libtvm.so(+0x15bfa6b) [0x7f60fa924a6b]
File "/root/Codes/tvm_in_mac/python/tvm/_ffi/_ctypes/packed_func.py", line 78, in cfun
rv = local_pyfunc(*pyargs)
File "/root/Codes/tvm_in_mac/python/tvm/relay/backend/compile_engine.py", line 263, in lower_call
op, call.attrs, inputs, ret_type, target)
File "/root/Codes/tvm_in_mac/python/tvm/relay/backend/compile_engine.py", line 199, in select_implementation
outs = impl.compute(attrs, inputs, out_type)
File "/root/Codes/tvm_in_mac/python/tvm/relay/op/op.py", line 89, in compute
return _OpImplementationCompute(self, attrs, inputs, out_type)
File "/root/Codes/tvm_in_mac/python/tvm/_ffi/_ctypes/packed_func.py", line 225, in __call__
raise get_last_ffi_error()
[bt] (4) /root/Codes/tvm_in_mac/build/libtvm.so(TVMFuncCall+0x61) [0x7f60fa928651]
[bt] (3) /root/Codes/tvm_in_mac/build/libtvm.so(+0x15009cd) [0x7f60fa8659cd]
[bt] (2) /root/Codes/tvm_in_mac/build/libtvm.so(tvm::relay::OpImplementation::Compute(tvm::Attrs const&, tvm::runtime::Array<tvm::te::Tensor, void> const&, tvm::Type const&)+0xbc) [0x7f60fa8657ac]
[bt] (1) /root/Codes/tvm_in_mac/build/libtvm.so(tvm::runtime::Array<tvm::te::Tensor, void> tvm::runtime::TVMPODValue_::AsObjectRef<tvm::runtime::Array<tvm::te::Tensor, void> >() const+0x3af) [0x7f60fa1290df]
[bt] (0) /root/Codes/tvm_in_mac/build/libtvm.so(dmlc::LogMessageFatal::~LogMessageFatal()+0x67) [0x7f60f9f5c907]
File "/root/Codes/tvm_in_mac/include/tvm/runtime/packed_func.h", line 1402
TVMError: Check failed: ObjectTypeChecker<TObjectRef>: :Check(ptr): Expect Array[Tensor] but get Tensor
The test script
import os
import tvm
from tvm import te
from tvm import relay
bx_shape = (20, 4)
score_shape = (20,)
iou_threshold = 0.5
score_threshold = 0.6
out_size = 10
dtype = "float32"
boxes = np.random.uniform(0, 10, size=bx_shape).astype(dtype)
scores = np.random.uniform(size=score_shape).astype(dtype)
def get_nms_relay_ir_module():
nms_graph = tf.Graph()
with nms_graph.as_default():
in_data_1 = tf.placeholder(dtype, boxes.shape, name="in_data_1")
in_data_2 = tf.placeholder(dtype, scores.shape, name="in_data_2")
nms_result = tf.image.non_max_suppression(
boxes=in_data_1, scores=in_data_2,
max_output_size=out_size, iou_threshold=iou_threshold,
score_threshold=score_threshold, name="nms")
nms_graph_def = nms_graph.as_graph_def()
layout = "NCHW"
out_names = ['nms/NonMaxSuppressionV3:0']
input_data = [boxes, scores]
input_node = ['in_data_1', 'in_data_2']
shape_dict = {e: i.shape for e, i in zip(input_node, input_data)}
mod, params = relay.frontend.from_tensorflow(nms_graph_def,
layout=layout,
shape=shape_dict,
outputs=out_names)
return mod, params
def nms_lab():
use_gpu = True
if use_gpu:
target = "cuda"
context = tvm.gpu()
else:
target = "llvm"
context = tvm.cpu()
mod, params = get_nms_relay_ir_module()
exe = relay.vm.compile(mod, target=target, params=params)
code, lib = exe.save()
des_exec = tvm.runtime.vm.Executable.load_exec(code, lib)
des_vm = tvm.runtime.vm.VirtualMachine(des_exec)
des_vm.init(context)
args = (scores, boxes)
print('\n****** Start the running **********')
ret = des_vm.run(*args)
print('ret type: ', type(ret))
print('ret: ', ret)
if __name__ == '__main__':
nms_lab()