[Bug] Failed when using te.gradient for `roi_align`

[hgt312]

Meet this when updating tvm version

E             [bt] (8) 9   libtvm.dylib                        0x0000000163f79895 tvm::arith::IRMutatorWithAnalyzer::VisitExpr_(tvm::tir::ReduceNode const*) + 245
E             [bt] (7) 8   libtvm.dylib                        0x0000000163e322ec tvm::arith::Analyzer::Bind(tvm::tir::Var const&, tvm::Range const&, bool) + 396
E             [bt] (6) 7   libtvm.dylib                        0x0000000163f31987 tvm::arith::IntSetAnalyzer::Bind(tvm::tir::Var const&, tvm::Range const&, bool) + 55
E             [bt] (5) 6   libtvm.dylib                        0x0000000163f319fd tvm::arith::IntSetAnalyzer::Impl::Bind(tvm::tir::Var const&, tvm::Range const&, bool) + 109
E             [bt] (4) 5   libtvm.dylib                        0x0000000163f31861 tvm::arith::IntSetAnalyzer::Impl::Update(tvm::tir::Var const&, tvm::arith::IntSet const&, bool) + 1537
E             [bt] (3) 4   libtvm.dylib                        0x0000000163e323f5 tvm::runtime::detail::LogFatal::~LogFatal() + 21
E             [bt] (2) 3   libtvm.dylib                        0x0000000163e3452d tvm::runtime::detail::LogFatal::~LogFatal() + 29
E             [bt] (1) 2   libtvm.dylib                        0x0000000163e345d5 tvm::runtime::detail::LogFatal::Entry::Finalize() + 149
E             [bt] (0) 1   libtvm.dylib                        0x0000000167c2f76e tvm::runtime::Backtrace() + 30
E             File "../3rdparty/tvm/src/arith/int_set.cc", line 589
E             File "../src/impl/op.cc", line 178
E           TVMError: Cannot find a valid dispatch for op raf.op.roi_align_dx:
E           [TVM] Failed to JIT: raf_op_tvm_roi_align_dx_6: TVMError:
E           ---------------------------------------------------------------
E           An error occurred during the execution of TVM.
E           For more information, please see: https://tvm.apache.org/docs/errors.html
E           ---------------------------------------------------------------
E             Check failed: (ExprDeepEqual()(old_info.max(), info.max())) is false: Trying to update var 'rh' with a different maximum value: original=(int32(tir.ceil((max((placeholder[i, 4] - placeholder[i, 2]), 1f)/7f))) - 1), new=(int32(tir.ceil((max((placeholder[i, 4] - placeholder[i, 2]), 1f)*0.142857f))) - 1)

Seems that the only difference is x/7 vs x*(1/7), can this be solved by using simplify before equal?

@Lunderberg

[hgt312]

I have tried to use ProveEqual instead of ExprDeepEqual() in src/arith/int_set.cc, it works