Fix the minor bugs in Transformer.

Find some fp16 bugs during decoding NMT system. We decide to roll back some codes to the last version. These codes need to be reviewed.

1. Update the installing methods in manual.
2. Remove the traditional installing way of Windows in manual.
3. Fix the bugs in CMake.

Find some bugs during training NMT system. We decide to roll back some codes to the last version. These codes need to be reviewed.

Support to check whether to compile half precision parts in CUDA codes.

Fix the __half bugs in SetData.

1. Fix the bugs in backward process.
2. Support the float16 class.
3. Fix bugs.
4. Clean the codes.
5. Remove the makefile.

Merge with the branch of xuchen (NOT update the float16, this needs code review) and fix the bugs in Gather function.
1. Support Reciprocal fucntion.
2. Fix the safe delete bugs in XDevice.
3. Support new API to convert the data type of tensor.
4. Support to show the memory usage of buffer memory.
5. Fix minor errors.

Fix minor errors.

Fix the bugs in compiling NiuTensor with OpenBLAS (the version is not 0.3.10).

Fix minor errors.

Highlight to users about installing MKL, OpenBLAS and CUDA by themselves.

1. Support to generate the dynamic link library on GPU.
2. Support to compile the project on the MKL and OpenBLAS.
3. Update the manual to support the new CMake file.
4. Fix minor errors.

Fix the bug in destroy the stream of XDevice.

1. Support to generate the dynamic link library on GPU.
2. Support to compile the project on the MKL and OpenBLAS.

1. Try to fix the bugs in destroy the stream of XDevice (Uncheck).
2. Fix the bug of memory leak in ReduceSumAll function.
2. Adjust the directory structure.
3. Fix the minor errors.

Fix the bugs in Sum and Sub when MKL or OpenBlas is used.

1. Update the codes of machine translation sample. The current version is the same with NiuTrans.NMT.
2. Update the XList class.
3. Bugs fix.

1. Offer macro to set whether use the half precision in cuda codes.
2. Update the manuals.

1. Fix the memory leak bugs in XList.
2. Replace the half with the unsigned short data type.

Fix the bugs in ReduceMax and its related functions.

Support fp16 data type for more operations and fix the minor errors.  (Don't use this! It's an incomplete version)

Support fp16 data type for more operations and fix the minor errors.  (Don't use this! It's an incomplete version)

Update the minimum version of CUDA from 9.0 to 9.2.

Support fp16 data type for more operations and fix the minor errors.  (Don't use this! It's an incomplete version)

Update the version number which is synchronized with the github version.

1. Support KB size for memory pool.
2. Clean the codes: 1000000 -> 1024 * 1024.

1. Fix the mistakes in manual.
2. Fix the minor errors in Stack and LogSoftmax functions.

1. Fix the mistakes in manual.
2. Fix the minor errors in Normalize and ScaleAndShift functions.

1. Fix the mistakes in manual.
2. Fix the bug in the link process of ConvertDataType function.

1. Fix the mistakes in manual. By the way, I have to say there are so many mistakes in the manual. I'm shocked it has been checked a lot of times, but why they are still be there. No one care about that? Really???
2. Clean the codes.

1. Offer CMakeLists.txt to compile the NiuTensor.
2. Update the manuals for CMake.

1. Call the Scale or Shift functions when the parameter shift is 0.0F or scale is 1.0F.
2. Add the unit test for ScaleAndShift for the cases which parameter shift is 0.0F or scale is 1.0F.

Use the _CheckData without tolerance in some unit test to speedup the test process.