Pierre Huchant
284827fd8b
julia: Update examples for CUDA.
|
пре 5 година | |
|---|---|---|
| .. | ||
| res | пре 5 година | |
| Makefile | пре 5 година | |
| README | пре 5 година | |
| cpu_mult.c | пре 5 година | |
| gpu_mult.cu | пре 5 година | |
| mult.c | пре 5 година | |
| mult.jl | пре 5 година | |
| mult.plot | пре 5 година | |
| mult_native.jl | пре 5 година | |
README
You first need to compile StarPU.jl
Now, type
> make
> make test
> gnuplot mult.plot
During the test, several versions are compared:
- mult.c: this is the original C+starpu code
- mult_native.jl: this is the native julia way of multiplying 2 matrices. There is a call to openBLAS
- mult_generatedc.jl: generates a C file, genc_matrix_mult.c. This file is compiled into a library, that is loaded and a function is executed.
- mult_calllib.jl: the env variable JULIA_TASK_LIB should point to a library containing the function matrix_mult. This function is called.
Examples of how to launch mult.jl are given in test rules.
A CUDA version is generated too. Untested so far (no nvidia gpu on my laptop!)
Now, type
> make
> make test
> gnuplot mult.plot
During the test, several versions are compared:
- mult.c: this is the original C+starpu code
- mult_native.jl: this is the native julia way of multiplying 2 matrices. There is a call to openBLAS
- mult_generatedc.jl: generates a C file, genc_matrix_mult.c. This file is compiled into a library, that is loaded and a function is executed.
- mult_calllib.jl: the env variable JULIA_TASK_LIB should point to a library containing the function matrix_mult. This function is called.
Examples of how to launch mult.jl are given in test rules.
A CUDA version is generated too. Untested so far (no nvidia gpu on my laptop!)