ganler opened a new issue #10528: URL: https://github.com/apache/tvm/issues/10528
Thanks for participating in the TVM community! We use https://discuss.tvm.ai for any general usage questions and discussions. The issue tracker is used for actionable items such as feature proposals discussion, roadmaps, and bug tracking. You are always welcomed to post on the forum first :smile_cat: Issues that are inactive for a period of time may get closed. We adopt this policy so that we won't lose track of actionable issues that may fall at the bottom of the pile. Feel free to reopen a new one if you feel there is an additional problem that needs attention when an old one gets closed. ### Expected behavior ```python """ def @main(%x: Tensor[(2, 2, 1, 1), float32]) -> Tensor[(2, 3, 4), float32] { %0 = nn.conv2d(%x, meta[relay.Constant][0] /* ty=Tensor[(1, 2, 3, 1), float32] */, strides=[2, 2], padding=[3, 3, 3, 3]) /* ty=Tensor[(2, 1, 3, 4), float32] */; squeeze(%0, axis=[1]) /* ty=Tensor[(2, 3, 4), float32] */ } """ ``` This simple conv2d + squeeze model should pass compilation but actually failed. ### Actual behavior ``` 5: tvm::relay::MixedModeMutator::VisitLeaf(tvm::RelayExpr const&) at /home/ganler/Documents/tvm/src/relay/ir/expr_functor.cc:81 4: tvm::relay::TempRealizer::DispatchVisitExpr(tvm::RelayExpr const&) at /home/ganler/Documents/tvm/src/relay/transforms/forward_rewrite.cc:46 3: tvm::relay::LayoutAlternatedExprNode<tvm::relay::alter_op_layout::AlterTransformMemorizer>::Realize() const at /home/ganler/Documents/tvm/src/relay/transforms/transform_layout.h:183 2: tvm::relay::TransformMemorizer::Transform(tvm::RelayExpr, tvm::tir::Layout const&, tvm::tir::Layout const&) at /home/ganler/Documents/tvm/src/relay/transforms/transform_layout.h:115 1: tvm::relay::TransformMemorizer::TransformHelper(tvm::RelayExpr, tvm::tir::Layout, tvm::tir::Layout) at /home/ganler/Documents/tvm/src/relay/transforms/transform_layout.h:157 0: tvm::tir::BijectiveLayout::BijectiveLayout(tvm::tir::Layout, tvm::tir::Layout) at /home/ganler/Documents/tvm/src/tir/ir/data_layout.cc:421 File "/home/ganler/Documents/tvm/src/tir/ir/data_layout.cc", line 422 TVMError: --------------------------------------------------------------- An error occurred during the execution of TVM. For more information, please see: https://tvm.apache.org/docs/errors.html --------------------------------------------------------------- Check failed: (GetStoreRule(&n->index_backward_rule, &n->shape_backward_rule, n->dst_layout, n->src_layout)) is false: NHW1c NHW ``` ### Environment Linux; Clang-14; TVM tag 8f6fa8f2c41406cb54d01647ba8731e4ceb8f4ab; ### Steps to reproduce Preferably a minimal script to cause the issue to occur. ```python import tvm from tvm import relay import numpy as np xshape = (2, 2, 1, 1) inp = np.random.uniform(size=xshape).astype(np.int64) x = relay.var("x", shape=xshape, dtype='float32') v1 = relay.nn.conv2d(x, weight=relay.const(value=np.random.random((1, 2, 3, 1))), strides=[2, 2], padding=[3, 3, 3, 3], kernel_size=[3, 1], channels=1) out = relay.squeeze(v1, axis=[1]) func = relay.Function([x], out) mod = tvm.IRModule.from_expr(func) with tvm.transform.PassContext(opt_level=4): relay.build_module.create_executor("graph", mod, tvm.cpu(), target='llvm').evaluate() ``` This seems to be related to https://github.com/apache/tvm/pull/9996 cc: @masahi @yangulei -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
