Adds openmp offload for XSBench, RSBench, miniFE #4
Conversation
config.yaml
Outdated
| } | ||
| copy: [ 'openmp-threading/XSBench' ] | ||
| bin: 'XSBench' | ||
| run: 'env OMP_NUM_THREADS=1 OMP_PROC_BIND=true ./XSBench' | ||
| input: '-t 1 -k 1 -s small' | ||
| measure: | ||
| input: '-t 1 -m event -s small' |
There was a problem hiding this comment.
Is this input compatible with the other configurations (omp, seq)?
config.yaml
Outdated
| } | ||
| copy: [ 'openmp-threading/rsbench' ] | ||
| bin: 'rsbench' | ||
| run: 'env OMP_NUM_THREADS=1 OMP_PROC_BIND=true ./rsbench' | ||
| input: '-t 1 -s small' | ||
| measure: | ||
| input: '-s small -m event -t 1' |
There was a problem hiding this comment.
Is this input compatible with the other configurations (seq, omp)?
config.yaml
Outdated
| @@ -183,6 +190,14 @@ miniFE: | |||
| -DMINIFE_CSR_MATRIX -DMINIFE_RESTRICT=__restrict__ | |||
| -DMPICH_IGNORE_CXX_SEEK -DMINIFE_INFO=1 -DMINIFE_KERNELS=0"' | |||
| ], | |||
| omp-offload: [ 'cd openmp45/src; | |||
| make -j CC=clang CXX=clang++ | |||
| CFLAGS="-v -O3 -fopenmp -fopenmp-targets=nvptx64-nvidia-cuda | |||
| print_CLI_error(); | ||
| } | ||
| - else | ||
| + else if( strcmp(arg, "-t") == 0 ) |
There was a problem hiding this comment.
Is this to enable multi-threaded target creation? Is an option on the number of threads correct? Shouldn't be 1 for correct program execution?
| + } | ||
| // n_gridpoints (-g) | ||
| - if( strcmp(arg, "-g") == 0 ) | ||
| + else if( strcmp(arg, "-g") == 0 ) |
There was a problem hiding this comment.
Change to if, otherwise -t precludes -g
| char * arg = argv[i]; | ||
|
|
||
| + // This will be ignored | ||
| + if( strcmp(arg, "-t") == 0 ) |
There was a problem hiding this comment.
Is an option on the number of threads correct? Shouldn't be 1 for correct program execution?
config.yaml
Outdated
| @@ -106,20 +106,24 @@ miniAMR: | |||
| clean: [ 'git clean -fx' ] | |||
|
|
|||
| XSBench: | |||
| fetch: 'git clone https://github.com/ANL-CESAR/XSBench.git' | |||
| fetch: 'git clone https://github.com/ANL-CESAR/XSBench.git; | |||
There was a problem hiding this comment.
Please add the command to clone/checkout the specific commit that this patch works with. Also for RSBench fetch. We should do it for every repo that needs a patch.
|
Possibly add: miniQMC |
- Move offload applications to another configuration file. - Add Clang and IBM compilation toolchain for applications. - Extends Harness to parse information out of the nvlink info - The parsed data are stored under the results/results-'application name'.yaml file - Add miniQMC application
koparasy
force-pushed
the
features/openmp-offload
branch
from
d4d3ca2
to
17bcb73
Compare
This patch adds compiling configurations for XSBench, RSBench, and miniFE using the OpenMP offload programming model.