cxx122 opened a new issue, #12380:
URL: https://github.com/apache/tvm/issues/12380
```
TENSOR_0 = te.placeholder([3,3,3], dtype="float16", name="TENSOR_0")
TENSOR_1 = te.compute([3,3,3,3], lambda
dck,zch,vci,wcs:TENSOR_0[wcs,vci,dck]*te.sqrt(TENSOR_0[wcs,wcs,zch]), name
="TENSOR_1")
```
```
TENSOR_0 = te.placeholder([3,3,3], dtype="float16", name="TENSOR_0")
TENSOR_1 = te.compute([3,3,3,3], lambda
dck,zch,vci,wcs:TENSOR_0[wcs,vci,dck]*te.cos(TENSOR_0[wcs,wcs,zch]), name
="TENSOR_1")
```
When changing the shape num of the tensor higher or equal to 3, it will
cause an inconsistency after using fuse.
### Actual behavior
cos:
```
AssertionError:
Not equal to tolerance rtol=1e-05, atol=1e-07
Mismatched elements: 12 / 81 (14.8%)
Max absolute difference: 0.007812
Max relative difference: 0.000825
x: array([[[[ 3.8 , 5.203 , 2.787 ],
[ 5.234 , 10.79 , 2.96 ],
[ 1.436 , 9.68 , 0.3608]],...
y: array([[[[ 3.8 , 5.203 , 2.787 ],
[ 5.234 , 10.79 , 2.96 ],
[ 1.436 , 9.68 , 0.3608]],...
```
sqrt:
```
AssertionError:
Not equal to tolerance rtol=1e-05, atol=1e-07
Mismatched elements: 16 / 81 (19.8%)
Max absolute difference: 0.007812
Max relative difference: 0.0009217
x: array([[[[6.320e+00, 1.156e+00, 9.766e+00],
[4.926e+00, 1.188e+00, 2.744e-01],
[2.333e-02, 4.898e+00, 1.105e+01]],...
y: array([[[[6.3203e+00, 1.1562e+00, 9.7656e+00],
[4.9258e+00, 1.1885e+00, 2.7441e-01],
[2.3331e-02, 4.8984e+00, 1.1047e+01]],...
```
### Environment
Operating System: Ubuntu 18.04, TVM version: tag0.9.0 [d361585]
### Steps to reproduce
```
import os
import numpy as np
import tvm
from tvm import te, auto_scheduler, topi
import tvm.testing
# Your TE program
TENSOR_0 = te.placeholder([3,3,3], dtype="float16", name="TENSOR_0")
TENSOR_1 = te.compute([3,3,3,3], lambda
dck,zch,vci,wcs:TENSOR_0[wcs,vci,dck]*te.sqrt(TENSOR_0[wcs,wcs,zch]), name
="TENSOR_1")
s = te.create_schedule(TENSOR_1.op)
tensor_list = [TENSOR_0,TENSOR_1]
dev = tvm.cpu(0)
pre_list = []
after_list = []
for tensor in tensor_list:
shape = [x.value if 'value' in dir(x) and isinstance(x.value, int) else
1 for x in tensor.shape]
params = (5*np.random.uniform(size=shape)).astype(tensor.dtype)
pre_list.append(tvm.nd.array(params.copy(), dev))
after_list.append(tvm.nd.array(params.copy(), dev))
pre_mod = tvm.lower(s, tensor_list, simple_mode=True)
with tvm.transform.PassContext(opt_level=4):
f = tvm.build(pre_mod)
f(*pre_list)
# Schedule
TENSOR_1_dck, TENSOR_1_zch, TENSOR_1_vci, TENSOR_1_wcs =
tuple(TENSOR_1.op.axis) + tuple(TENSOR_1.op.reduce_axis)
TENSOR_1_dck_zch_fused_vci_fused = s[TENSOR_1].fuse(TENSOR_1_dck,
TENSOR_1_zch, TENSOR_1_vci)
now_mod = tvm.lower(s, tensor_list, simple_mode=True)
with tvm.transform.PassContext(opt_level=4):
f = tvm.build(now_mod)
f(*after_list)
tvm.testing.assert_allclose(pre_list[1].numpy(),
after_list[1].numpy(),rtol=1e-5)
```
--
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]
