You are viewing a plain text version of this content. The canonical link for it is here.
Posted to discuss-archive@tvm.apache.org by Hao Luo via Apache TVM Discuss <no...@discuss.tvm.ai> on 2020/10/27 02:45:41 UTC
[Apache TVM Discuss] [Questions] Relay cannot compile while_loop
I want to test the usage of relay while_loop and write the following simple example
```
x = relay.var("x", shape=(10, 20))
i = relay.var("i", shape=tuple(), dtype="int32")
def myfun(x, i):
z = relay.add(x, relay.const(1, "float32"))
j = relay.add(i, relay.const(1, "int32"))
return z, j
def cond(x, i):
return relay.less(i, relay.const(10, "int32"))
myloop = while_loop(cond, [x,i], myfun)
z = myloop(x, relay.const(0, "int32"))
result = relay.Function([x], relay.TupleGetItem(z, 0))
print(result.astext())
zz = result
module = tvm.IRModule.from_expr(zz)
print(module["main"])
with relay.build_config(opt_level=1):
ex = relay.create_executor("debug", mod=module, ctx=tvm.cpu(), target="llvm") #"debug"
re = ex.evaluate()(np.zeros((10,20), dtype='float32'))
print(re)
graph, mod, param = relay.build(module,target="llvm")
```
and the Relay IR seems to be correctly and "debug" interpreter run passed.
This is the Relay IR
```
fn (%x: Tensor[(10, 20), float32]) {
%3 = (
let %while_loop = fn (%x1: Tensor[(10, 20), float32], %i: int32) {
%0 = less(%i, 10);
if (%0) {
%1 = add(%x1, 1f);
%2 = add(%i, 1);
%while_loop(%1, %2)
} else {
(%x1, %i)
}
};
%while_loop
);
%4 = %3(%x, 0);
%4.0
}
```
However, When I try to build the module, I get an error message like this:
```
Traceback (most recent call last):
File "jacobi.py", line 53, in <module>
test_fuse_simple()
File "jacobi.py", line 44, in test_fuse_simple
graph, mod, param = relay.build(module,target="llvm")
File "tvm-latest/lib/python3.7/site-packages/tvm-0.8.dev0-py3.7-linux-x86_64.egg/tvm/relay/build_module.py", line 260, in build
graph_json, mod, params = bld_mod.build(mod, target, target_host, params)
File "tvm-latest/lib/python3.7/site-packages/tvm-0.8.dev0-py3.7-linux-x86_64.egg/tvm/relay/build_module.py", line 127, in build
self._build(mod, target, target_host)
File "tvm-latest/lib/python3.7/site-packages/tvm-0.8.dev0-py3.7-linux-x86_64.egg/tvm/_ffi/_ctypes/packed_func.py", line 237, in __call__
raise get_last_ffi_error()
tvm._ffi.base.TVMError: Traceback (most recent call last):
[bt] (8) .local/tvm/lib/libtvm.so(std::_Function_handler<void (tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*), tvm::relay::backend::GraphRuntimeCodegenModule::GetFunction(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, tvm::runtime::ObjectPtr<tvm::runtime::Object> const&)::{lambda(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*)#2}>::_M_invoke(std::_Any_data const&, tvm::runtime::TVMArgs&&, tvm::runtime::TVMRetValue*&&)+0x6e) [0x7faa7a736a4e]
[bt] (7) .local/tvm/lib/libtvm.so(tvm::relay::backend::GraphRuntimeCodegen::Codegen(tvm::relay::Function)+0x49d) [0x7faa7a734e5d]
[bt] (6) .local/tvm/lib/libtvm.so(tvm::relay::backend::MemoizedExprTranslator<std::vector<tvm::relay::backend::GraphNodeRef, std::allocator<tvm::relay::backend::GraphNodeRef> > >::VisitExpr(tvm::RelayExpr const&)+0x16a) [0x7faa7a7316da]
[bt] (5) .local/tvm/lib/libtvm.so(tvm::relay::ExprFunctor<std::vector<tvm::relay::backend::GraphNodeRef, std::allocator<tvm::relay::backend::GraphNodeRef> > (tvm::RelayExpr const&)>::InitVTable()::{lambda(tvm::runtime::ObjectRef const&, tvm::relay::ExprFunctor<std::vector<tvm::relay::backend::GraphNodeRef, std::allocator<tvm::relay::backend::GraphNodeRef> > (tvm::RelayExpr const&)>*)#10}::_FUN(tvm::runtime::ObjectRef const&, tvm::relay::ExprFunctor<std::vector<tvm::relay::backend::GraphNodeRef, std::allocator<tvm::relay::backend::GraphNodeRef> > (tvm::RelayExpr const&)>*)+0x14) [0x7faa7a71ba14]
[bt] (4) .local/tvm/lib/libtvm.so(tvm::relay::backend::GraphRuntimeCodegen::VisitExpr_(tvm::relay::TupleGetItemNode const*)+0x1d) [0x7faa7a71c19d]
[bt] (3) .local/tvm/lib/libtvm.so(tvm::relay::backend::MemoizedExprTranslator<std::vector<tvm::relay::backend::GraphNodeRef, std::allocator<tvm::relay::backend::GraphNodeRef> > >::VisitExpr(tvm::RelayExpr const&)+0x16a) [0x7faa7a7316da]
[bt] (2) .local/tvm/lib/libtvm.so(tvm::relay::ExprFunctor<std::vector<tvm::relay::backend::GraphNodeRef, std::allocator<tvm::relay::backend::GraphNodeRef> > (tvm::RelayExpr const&)>::InitVTable()::{lambda(tvm::runtime::ObjectRef const&, tvm::relay::ExprFunctor<std::vector<tvm::relay::backend::GraphNodeRef, std::allocator<tvm::relay::backend::GraphNodeRef> > (tvm::RelayExpr const&)>*)#6}::_FUN(tvm::runtime::ObjectRef const&, tvm::relay::ExprFunctor<std::vector<tvm::relay::backend::GraphNodeRef, std::allocator<tvm::relay::backend::GraphNodeRef> > (tvm::RelayExpr const&)>*)+0x14) [0x7faa7a71b9d4]
[bt] (1) .local/tvm/lib/libtvm.so(tvm::relay::backend::GraphRuntimeCodegen::VisitExpr_(tvm::relay::CallNode const*)+0x1fb) [0x7faa7a72d24b]
[bt] (0) .local/tvm/lib/libtvm.so(+0x17c64d4) [0x7faa7a71b4d4]
File "download-file/tvm/src/relay/backend/graph_runtime_codegen.cc", line 355
TVMError: TVM runtime does not support calls to relay.Let
```
Does TVM currently support build of while_loops?
---
[Visit Topic](https://discuss.tvm.apache.org/t/relay-cannot-compile-while-loop/8294/1) to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click here](https://discuss.tvm.apache.org/email/unsubscribe/7c62bdc85747465d2ad8e32d088f72068b9b258d8c8af6e4f8e5362db5fee3e2).
[Apache TVM Discuss] [Questions] Relay cannot compile while_loop
Posted by Hao Luo via Apache TVM Discuss <no...@discuss.tvm.ai>.
Thank you very much. I have understand it
---
[Visit Topic](https://discuss.tvm.apache.org/t/relay-cannot-compile-while-loop/8294/3) to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click here](https://discuss.tvm.apache.org/email/unsubscribe/eb03e8789ecb80dd965a08ed89c96250202a07d54eb533d8ace417a8f3b79677).
[Apache TVM Discuss] [Questions] Relay cannot compile while_loop
Posted by masahi via Apache TVM Discuss <no...@discuss.tvm.ai>.
You cannot use `relay.build(...)` to build a model with control flow. For that, you need to use VM.
See for example,
https://github.com/apache/incubator-tvm/blob/efe3a79aacd934ea5ffb13170230bf199a473e72/tests/python/frontend/pytorch/test_forward.py#L1914
---
[Visit Topic](https://discuss.tvm.apache.org/t/relay-cannot-compile-while-loop/8294/2) to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click here](https://discuss.tvm.apache.org/email/unsubscribe/a5129c8bbbd00ca24a4f68bfde8f3e20373427328a8447ed7a2abe43cbeafa45).