[ROCM] Failed topi unittests for rocm backend #552
Description
[Update 10/23] All failing test cases were resolved.
[Update 10/20] Currently 3 tests in topi unittests are failing.
See the summary here.
test_topi_conv2d_nchw.py
3 out of 12 test cases fail.
case verify_conv2d_nchw(1, 3, 224, 64, 7, 3, 2)
$ python test_topi_conv2d_nchw.py
AssertionError:
Not equal to tolerance rtol=1e-05, atol=0
(mismatch 99.98687454346238%)
x: array([[[[ 17.740227, 25.133751, 23.486025, ..., 25.639071, 24.148636,
21.255533],
[ 24.602934, 33.843094, 35.173992, ..., 39.136459, 36.724586,...
y: array([[[[ 16.973913, 25.437023, 24.094303, ..., 27.416625, 25.55851 ,
22.340623],
[ 25.350509, 35.822084, 36.533454, ..., 38.809231, 36.673988,...
case verify_conv2d_nchw(1, 128, 28, 128, 3, 1, 1)
$ python test_topi_conv2d_nchw.py
Use memoize topi.tests.test_topi_conv2d.verify_con2d_nchw.get_ref_data.pkl(5, (1, 128, 28, 28), 'float32', 1, 1, (128, 128, 3, 3))
testing rocm backend
Memory access fault by GPU node-1 on address 0x1000. Reason: Page not present or supervisor privilege.
Aborted (core dumped)
case verify_conv2d_nchw(1, 256, 14, 256, 3, 1, 1)
$ python test_topi_conv2d_nchw.py
Use memoize topi.tests.test_topi_conv2d.verify_con2d_nchw.get_ref_data.pkl(5, (1, 256, 14, 14), 'float32', 1, 1, (256, 256, 3, 3))
testing rocm backend
Memory access fault by GPU node-1 on address 0x5fc4e1000. Reason: Page not present or supervisor privilege.
Aborted (core dumped)
test_topi_depthwise_conv2d.py
2 out of 15 test cases fail.
case depthwise_conv2d_with_workload_nhwc(1, 728, 32, 1, 3, 1, "SAME")
$ python test_topi_depthwise_conv2d.py
AssertionError:
Not equal to tolerance rtol=1e-05, atol=0
(mismatch 29.669793097527474%)
x: array([[[[ 0.79067 , 0.833293, 1.011379, ..., 0.855761, 0.685508,
1.101951],
[ 1.64011 , 1.610993, 1.381752, ..., 1.928582, 1.182396,...
y: array([[[[ 0.79067 , 0.833293, 1.011379, ..., 0.855761, 0.685508,
1.101951],
[ 1.64011 , 1.610993, 1.381752, ..., 1.928582, 1.182396,...
case depthwise_conv2d_with_workload_nhwc(1, 728, 32, 1, 3, 1, "VALID")
$ python test_topi_depthwise_conv2d.py
AssertionError:
Not equal to tolerance rtol=1e-05, atol=0
(mismatch 27.69184981684981%)
x: array([[[[ 2.178229, 1.748752, 2.032793, ..., 2.637433, 1.739626,
1.027414],
[ 1.45854 , 1.976049, 1.973471, ..., 1.987515, 1.736224,...
y: array([[[[ 2.178229, 1.748752, 2.032793, ..., 2.637433, 1.739626,
1.027414],
[ 1.45854 , 1.976049, 1.973471, ..., 1.987515, 1.736224,...
test_topi_pooling.py
1 out of 8 test cases fails.
case verify_pool(1, 256, 32, 2, 2, [0, 0], 'avg')
$ python test_topi_pooling.py
AssertionError:
Not equal to tolerance rtol=1e-05, atol=0
(mismatch 50.0%)
x: array([[[[ 0.383371, 0.234747, 0.431097, ..., 0.498689, 0.369492,
0.783951],
[ 0.551467, 0.405494, 0.410852, ..., 0.414632, 0.581095,...
y: array([[[[ 0.383371, 0.234747, 0.431097, ..., 0.498689, 0.369492,
0.783951],
[ 0.551467, 0.405494, 0.410852, ..., 0.414632, 0.581095,...