[BUG][CuTeDSL] cute.print_tensor() won't print anything if the tensor is the result of a copy

[sfc-gh-lpaille]

Which component has the problem?

CuTe DSL

Bug Report

Describe the bug
cute.print_tensor() doesn't print anything if the tensor is the result of a copy.

Example output of the script:

(.leovenv) ubuntu@machine $ python cant_print_tensor_after_a_copy.py 
(.leovenv) ubuntu@machine $ 

Steps/Code to reproduce bug

import cutlass.cute as cute
import cutlass

@cute.kernel
def print_kernel():
    smem = cutlass.utils.SmemAllocator()
    s = smem.allocate_tensor(cute.Float16, cute.make_layout((8, 8), stride=(8, 1)), byte_alignment=16)
    r = cute.make_fragment(s.layout, s.element_type)
    cute.basic_copy(s, r) # if you comment this out, the tensor will print correctly
    cute.print_tensor(r)

@cute.jit
def print_launcher():
    print_kernel().launch(grid=(1, 1, 1),block=(1, 1, 1))

cutlass.cuda.initialize_cuda_context()
print_launcher()

Expected behavior
cute.print_tensor() should always print the tensor at runtime if the IR compiled correctly.

Environment details (please complete the following information):
using nvidia-cutlass-dsl==4.1.0 on ubuntu 22.04 arm64 on b200.

[anakinxc]

Hi @sfc-gh-lpaille,

Thanks for reporting this issue. Just verified this is already fixed in the latest codebase, please wait for next release and check again :)

[Leikoe]

Hi @sfc-gh-lpaille,

Thanks for reporting this issue. Just verified this is already fixed in the latest codebase, please wait for next release and check again :)

Hi!
Any when will the next release be approximately ?

[fengxie]

#2638 :)

[anakinxc]

Hi @sfc-gh-lpaille

Can you verify this is fixed in 4.2 release?

Thanks

[sfc-gh-lpaille]

yup! thank you ＼(≧▽≦)／