HeCBench: matrixT-hip benchmark fails due to missing cooperative groups support

[kballeda]

I find that co-operative groups support related compile error for matrixT-hip benchmark. Do we have a support for co-operative groups in chipStar, also I ran the same benchmark on AMD Mi250 and it was successful.

Benchmark: matrixT-hip in HeCBench Compile Error:

HeCBench/src/matrixT-hip$ make
hipcc  -std=c++14 -Wall -O3 -c main.cu -o main.o
main.cu:33:16: error: expected namespace name
namespace cg = cooperative_groups;
               ^
main.cu:81:3: error: use of undeclared identifier 'cg'
  cg::thread_block cta = cg::this_thread_block();

AMD Hip Flow on AMD Mi250:

HeCBench/src/matrixT-hip$ make run
./main 16384 16384 200

Matrix size: 16384x16384 (1024x1024 tiles), tile size: 16x16, block size: 16x16

Average kernel (simple copy       ) execution time: 2225.056641 (us)
Average kernel (shared memory copy) execution time: 2222.464111 (us)
Average kernel (naive             ) execution time: 10578.705078 (us)
Average kernel (coalesced         ) execution time: 10464.847656 (us)
Average kernel (optimized         ) execution time: 10529.517578 (us)
Average kernel (coarse-grained    ) execution time: 10540.002930 (us)
Average kernel (fine-grained      ) execution time: 2320.846191 (us)
Average kernel (diagonal          ) execution time: 4871.502441 (us)
PASS

[pjaaskel]

No support for co-operative groups yet.