Created by: luotao1

最新的develop分支, 在GPU环境和MKL库下，跑test_matrixCompare，在Matrix.unary处挂了：

12: Test command: /home/luotao/Paddle/build/paddle/math/tests/test_matrixCompare
12: Test timeout computed to be: 9.99988e+06
12: WARNING: Logging before InitGoogleLogging() is written to STDERR
12: I1107 17:20:58.170979 12290 hl_dso_loader.cc:51] Try to find cuda library: libcudart.so from default system path.
12: I1107 17:20:58.172441 12290 Util.cpp:155] commandline: /home/luotao/Paddle/build/paddle/math/tests/test_matrixCompare 
12: I1107 17:20:58.292949 12290 hl_dso_loader.cc:51] Try to find cuda library: libcublas.so from default system path.
12: I1107 17:20:58.435550 12290 hl_dso_loader.cc:51] Try to find cuda library: libcurand.so from default system path.
12: I1107 17:20:58.448356 12290 hl_dso_loader.cc:51] Try to find cuda library: libcudnn.so from default system path.
12: I1107 17:20:59.391186 12290 Util.cpp:130] Calling runInitFunctions
12: I1107 17:20:59.391809 12290 Util.cpp:143] Call runInitFunctions done.
12: [==========] Running 24 tests from 3 test cases.
12: [----------] Global test environment set-up.
12: [----------] 21 tests from Matrix
12: [ RUN      ] Matrix.projection
12: [       OK ] Matrix.projection (812 ms)
12: [ RUN      ] Matrix.maxSequence
12: [       OK ] Matrix.maxSequence (240 ms)
12: [ RUN      ] Matrix.unary
12: /home/luotao/Paddle/paddle/math/tests/test_matrixCompare.cpp:90: Failure
12: Value of: 0
12: Expected: count
12: Which is: 157
12: There are 157 different element.
12: [  FAILED  ] Matrix.unary (2254 ms)
12: [ RUN      ] Matrix.softmax
12: [       OK ] Matrix.softmax (125 ms)
...

具体是testMatrixInverse中657行没过：

outputCheck->copyFrom(*gpuI);
MatrixCheckErr(*cpuI, *outputCheck);

打印了一下cpuI和outputCheck的值，发现小数点后三位开始不同：

12: I1107 18:05:41.903053   400 test_matrixCompare.cpp:86] 4.37313,4.37211
12: I1107 18:05:41.903162   400 test_matrixCompare.cpp:86] -4.36522,-4.36408
12: I1107 18:05:41.903209   400 test_matrixCompare.cpp:86] -3.97099,-3.96996
12: I1107 18:05:41.903218   400 test_matrixCompare.cpp:86] -3.99913,-3.99811
12: I1107 18:05:41.903224   400 test_matrixCompare.cpp:86] 3.8104,3.80938
12: I1107 18:05:41.903231   400 test_matrixCompare.cpp:86] 5.36973,5.36836
12: I1107 18:05:41.903239   400 test_matrixCompare.cpp:86] -4.05777,-4.05671
12: I1107 18:05:41.903245   400 test_matrixCompare.cpp:86] 3.68704,3.68603
12: I1107 18:05:41.903288   400 test_matrixCompare.cpp:86] -4.9984,-4.9972
12: I1107 18:05:41.903296   400 test_matrixCompare.cpp:86] -4.67913,-4.678
12: I1107 18:05:41.903303   400 test_matrixCompare.cpp:86] 4.2865,4.28546
12: I1107 18:05:41.903309   400 test_matrixCompare.cpp:86] 4.11046,4.10944
12: I1107 18:05:41.903316   400 test_matrixCompare.cpp:86] 4.84822,4.84704
...