[Apache TVM Discuss] [Questions] InferBound error

[Swift.Sun via Apache TVM Discuss]

log:
Check failed: (found_attach || stage_attach.size() == 0) is false:

Invalid Schedule, cannot find the producer

compute(data_pad,

body=[tir.if_then_else(((((i1 >= 1) && (i1 < 2)) && (i2 >= 1)) && (i2 < 513)), 
data[i0, (i1 - 1), (i2 - 1), i3], 0f)],

axis=[iter_var(i0, range(min=0, ext=1)), iter_var(i1, range(min=0, ext=514)), 
iter_var(i2, range(min=0, ext=1025)), iter_var(i3, range(min=0, ext=512))], 
reduce_axis=[], tag=injective,pad, attrs={})

along the loop nest specified by compute_at of consumer

compute(input_tile, body=[data_pad[floordiv(p, 128), eps, ((floormod(p, 128)*4) 
+ nu), ci]],

axis=[iter_var(eps, range(min=0, ext=6)), iter_var(nu, range(min=0, ext=6)), 
iter_var(p, range(min=0, ext=128)), iter_var(ci, range(min=0, ext=512))], 
reduce_axis=[], tag=, attrs={})

stage_attach.size():2 found_attach:false

InferBound fails on the state:

Placeholder: data, kernel, bias

data_pack auto_unroll: 64

parallel p.0@ci.0@ (0,8192)

  for eps (None)

    for nu (None)

      for p (None)

        vectorize ci (None)

          input_tile = ...

  for p.1 (0,8)

    for i0 (None)

      for i1 (None)

        for i2 (None)

          vectorize i3 (None)

            data_pad = ...

    unroll eps (0,6)

      unroll nu (0,6)

        unroll r_a (0,6)

          unroll r_b (0,6)

            data_pack = ...

parallel eps.0@nu.0@co.0@ci.0@ (0,192)

  kernel_pack.local auto_unroll: 64

  for eps_c.0 (None)

    for nu_c.0 (None)

      for co_c.0 (None)

        for ci_c.0 (None)

          for eps_c.1 (None)

            for nu_c.1 (None)

              for co_c.1 (None)

                for ci_c.1 (None)

                  for r_kh.0 (None)

                    for r_kw.0 (None)

                      for eps_c.2 (None)

                        for nu_c.2 (None)

                          for co_c.2 (None)

                            for ci_c.2 (None)

                              for r_kh.1 (None)

                                for r_kw.1 (None)

                                  for eps_c.3 (None)

                                    for nu_c.3 (None)

                                      for co_c.3 (None)

                                        vectorize ci_c.3 (None)

                                          kernel_pack.local = ...

  for eps.1 (0,3)

    for nu.1 (0,2)

      for co.1 (0,16)

        for ci.1 (0,32)

          kernel_pack = ...

bgemm auto_unroll: 512

parallel eps.0@nu.0@p.0@co.0@eps.1@ (0,768)

  for p.1 (0,2)

    for co.1 (0,2)

      for ci.0 (0,128)

        for nu.2 (0,6)

          for p.2 (0,8)

            for ci.1 (0,4)

              bgemm = ...

parallel p.0@co.0@p.1@co.1@ (0,4096)

  unroll vh (0,4)

    unroll vw (0,4)

      unroll r_a (0,6)

        unroll r_b (0,6)

          inverse = ...

parallel i0@i1@i2@i3@ (0,16384)

  for n (None)

    for h (None)

      vectorize w@co@ (None)

        conv2d_winograd = ...

  compute = ...

with: [10:47:00] /srv/workspace/sunjiwei/code/tvm/src/te/schedule/bound.cc:177:





---
[Visit Topic](https://discuss.tvm.apache.org/t/inferbound-error/10394/1) 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/b45692e92c250e13e02abf500d71360374cdf164051d6800af2a586ceddc9870).