Our implementation in Pointer analysis assumes that make_range can create tensors of varying strides. For example, tt.make_range {end = 4096, start = 2048} : tensor <1024xi32> can create a tensor of size 1024 with stride == 2. However, in https://github.com/openai/triton/pull/2660, the new make_range's verifier states that stride must always be 1. We need to update our lit tests and visitOperandMakeRange's stride calculation to reflect this new requirement.
Our implementation in Pointer analysis assumes that
make_range
can create tensors of varying strides. For example,tt.make_range {end = 4096, start = 2048} : tensor <1024xi32>
can create a tensor of size1024
withstride == 2
. However, in https://github.com/openai/triton/pull/2660, the newmake_range
's verifier states that stride must always be 1. We need to update our lit tests andvisitOperandMakeRange
's stride calculation to reflect this new requirement.Related #53