cxx122 opened a new issue, #12378:
URL: https://github.com/apache/tvm/issues/12378
```
TENSOR_0 = te.compute([2,2], lambda pcg,wcv:te.max_value("int8"), name
="TENSOR_0")
TENSOR_1 = te.placeholder([2,2], dtype="int64", name="TENSOR_1")
TENSOR_2 = te.compute([2,2], lambda
zcv,tcu:TENSOR_0[zcv,tcu]*TENSOR_0[zcv,tcu]*TENSOR_1[zcv,tcu], name ="TENSOR_2")
```
The tir program before compute_inline:
```
@main = primfn(TENSOR_0_1: handle, TENSOR_1_1: handle, TENSOR_2_1: handle)
-> ()
attr = {"from_legacy_te_schedule": True, "global_symbol": "main",
"tir.noalias": True}
buffers = {TENSOR_0: Buffer(TENSOR_0_2: Pointer(int8), int8, [4], []),
TENSOR_1: Buffer(TENSOR_1_2: Pointer(int64), int64, [4], []),
TENSOR_2: Buffer(TENSOR_2_2: Pointer(int64), int64, [4], [])}
buffer_map = {TENSOR_0_1: TENSOR_0, TENSOR_1_1: TENSOR_1, TENSOR_2_1:
TENSOR_2}
preflattened_buffer_map = {TENSOR_0_1: TENSOR_0_3: Buffer(TENSOR_0_2,
int8, [2, 2], []), TENSOR_1_1: TENSOR_1_3: Buffer(TENSOR_1_2, int64, [2, 2],
[]), TENSOR_2_1: TENSOR_2_3: Buffer(TENSOR_2_2, int64, [2, 2], [])} {
for (pcg: int32, 0, 2) {
for (wcv: int32, 0, 2) {
TENSOR_0[((pcg*2) + wcv)] = 127i8
}
}
for (zcv: int32, 0, 2) {
for (tcu: int32, 0, 2) {
let cse_var_1: int32 = ((zcv*2) + tcu)
TENSOR_2[cse_var_1] = (cast(int64,
(TENSOR_0[cse_var_1]*TENSOR_0[cse_var_1]))*TENSOR_1[cse_var_1])
}
}
}
```
The tir program after compute_inline:
```
@main = primfn(TENSOR_0_1: handle, TENSOR_1_1: handle, TENSOR_2_1: handle)
-> ()
attr = {"from_legacy_te_schedule": True, "global_symbol": "main",
"tir.noalias": True}
buffers = {TENSOR_0: Buffer(TENSOR_0_2: Pointer(int8), int8, [4], []),
TENSOR_1: Buffer(TENSOR_1_2: Pointer(int64), int64, [4], []),
TENSOR_2: Buffer(TENSOR_2_2: Pointer(int64), int64, [4], [])}
buffer_map = {TENSOR_0_1: TENSOR_0, TENSOR_1_1: TENSOR_1, TENSOR_2_1:
TENSOR_2}
preflattened_buffer_map = {TENSOR_0_1: TENSOR_0_3: Buffer(TENSOR_0_2,
int8, [2, 2], []), TENSOR_1_1: TENSOR_1_3: Buffer(TENSOR_1_2, int64, [2, 2],
[]), TENSOR_2_1: TENSOR_2_3: Buffer(TENSOR_2_2, int64, [2, 2], [])} {
for (zcv: int32, 0, 2) {
for (tcu: int32, 0, 2) {
let cse_var_1: int32 = ((zcv*2) + tcu)
TENSOR_2[cse_var_1] = (TENSOR_1[cse_var_1]*16129i64)
}
}
}
```
### Actual behavior
```
Traceback (most recent call last):
File
"/Scuzer/src/bugs/bug16/IncorrectResult__3ce1533f-dbeb-42c9-bb3a-5fe018abe989/Incorrect_bug.py",
line 46, in <module>
tvm.testing.assert_allclose(pre_list[2].numpy(),
after_list[2].numpy(),rtol=1e-5)
File "/Scuzer/tvm_cov_patch/tvm/python/tvm/testing/utils.py", line 114, in
assert_allclose
np.testing.assert_allclose(actual, desired, rtol=rtol, atol=atol,
verbose=True)
File
"/root/miniconda3/envs/py38/lib/python3.8/site-packages/numpy/testing/_private/utils.py",
line 1527, in assert_allclose
assert_array_compare(compare, actual, desired, err_msg=str(err_msg),
File
"/root/miniconda3/envs/py38/lib/python3.8/site-packages/numpy/testing/_private/utils.py",
line 840, in assert_array_compare
raise AssertionError(msg)
AssertionError:
Not equal to tolerance rtol=1e-05, atol=1e-07
Mismatched elements: 3 / 4 (75%)
Max absolute difference: 64512
Max relative difference: 0.999938
x: array([[0, 1],
[3, 4]])
y: array([[ 0, 16129],
[48387, 64516]])
```
### 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
TENSOR_0 = te.compute([2,2], lambda pcg,wcv:te.max_value("int8"), name
="TENSOR_0")
TENSOR_1 = te.placeholder([2,2], dtype="int64", name="TENSOR_1")
TENSOR_2 = te.compute([2,2], lambda
zcv,tcu:TENSOR_0[zcv,tcu]*TENSOR_0[zcv,tcu]*TENSOR_1[zcv,tcu], name ="TENSOR_2")
s = te.create_schedule(TENSOR_2.op)
tensor_list = [TENSOR_0,TENSOR_1,TENSOR_2]
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
s[TENSOR_0].compute_inline()
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[2].numpy(),
after_list[2].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]
