make all phi kernels to 2(host/device) static libraries directly

[zhiqiu]

PR types

Others

PR changes

Others

Describe

make all phi kernels to 2(host/device) static libraries directly, it can reduce about 3min of total compilation time.

• before
  each kernel is compiled to a static library
  
• after
  all kernels are compiled to 2 static libraries (libphi_cpu.a and libphi_gpu.a)
  

why

Consider A kernel uses B in its implementation, if each kernel is compiled to a static library, then libA depends on libB, and thus A is compiled after B, which may result in low concurrency in parellel compilation.

For example, in the following timeline, the matrix_rank_tol_kernel is compiled after reduce_sum_kernel since it reuse reduce_sum and other kernels. It is the last phi kernel compiled and only one thread(process) is being executed at that time even if the command is make -j24.

after this pr, matrix_rank_tol_kernel can overlap with others.

[paddle-bot-old]

你的PR提交成功，感谢你对开源项目的贡献!
请关注后续CI自动化测试结果，详情请参考Paddle-CI手册。
Your PR has been submitted. Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.