Additional unit test failures on setting execution engine to "naive"

[juliusshufan]

With naïve engine (by export MXNET_ENGINE_TYPE = NaiveEngine)
Compared to the tests results using the default engine, there are additional UT failures, they are:

CASE NAME	Error Log
test_quantization_mkldnn	AssertionError: Items are not equal: Error 6065500000000000068157440.000000 exceeds tolerance rtol=0.000010, atol=0.000000.  Location of maximum error:(1, 12, 8, 6), a=121310.000000, b=0.000000
test_subgraph:test_pos_single_conv	AssertionError: Items are not equal: Error 16186.005859 exceeds tolerance rtol=0.001000, atol=0.001000.  Location of maximum error:(3, 42, 0, 2), a=-16.600754, b=0.027311
test_subgraph:test_pos_conv_relu	AssertionError: Items are not equal: Error 15091.447266 exceeds tolerance rtol=0.001000, atol=0.001000.  Location of maximum error:(2, 59, 3, 1), a=15.091448, b=0.000000
test_subgraph:test_pos_conv_bn	AssertionError: Items are not equal: Error 83243.062500 exceeds tolerance rtol=0.001000, atol=0.001000.  Location of maximum error:(3, 53, 4, 7), a=152.526474, b=-0.842422
test_subgraph:test_pos_conv_add	AssertionError: Items are not equal: Error 35976.011719 exceeds tolerance rtol=0.001000, atol=0.001000.  Location of maximum error:(3, 15, 7, 3), a=-41.533928, b=-0.150311
test_subgraph:test_pos_conv_add2	AssertionError: Items are not equal: Error 23006.433594 exceeds tolerance rtol=0.001000, atol=0.001000.  Location of maximum error:(0, 42, 6, 3), a=-23.886930, b=-0.036677
test_subgraph:test_pos_conv_bn_relu	ssertionError: Items are not equal: Error 180822.453125 exceeds tolerance rtol=0.001000, atol=0.001000.  Location of maximum error:(0, 40, 3, 1), a=180.822464, b=0.000000
test_subgraph:test_pos_conv_bn_sum_relu	AssertionError: Items are not equal: Error 167951.921875 exceeds tolerance rtol=0.001000, atol=0.001000.  Location of maximum error:(0, 58, 4, 0), a=167.951935, b=0.000000
test_exc_handling:test_exc_imperative	mxnet.base.MXNetError: [10:45:55] src/operator/random/./sample_op.h:400: Check failed: param.scale > 0 (-1 vs. 0) : scale parameter in gaussian has to be positive
test_exc_handling:test_exc_gluon	mxnet.base.MXNetError: [10:46:04] src/operator/random/./sample_op.h:400: Check failed: param.scale > 0 (-10 vs. 0) : scale parameter in gaussian has to be positive
test_exc_handling:test_exc_post_fail	UnboundLocalError: local variable 'b' referenced before assignment
test_mkldnn:test_mkldnn_ndarray_slice	TypeError: 'float' object is not subscriptable

@TaoLv @pengzhao-intel @eric-haibin-lin

[mxnet-label-bot]

Hey, this is the MXNet Label Bot.
Thank you for submitting the issue! I will try and suggest some labels so that the appropriate MXNet community members can help resolve it.
Here are my recommended labels: Test

[vdantu]

@mxnet-label-bot add [test]

[pengzhao-intel]

@eric-haibin-lin @szha the MKLDNN related issue will be fixed in #15089.
Since people are using the naive engine for better performance now.
Does anyone in your side look into the engine?