gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

GEMM: reverted AMD optimizations

Browse Source
This commit is contained in:
Philippe Tillet
2015-11-29 16:13:14 -05:00
parent b3c5251f91
commit f975ea7621
2 changed files with 664 additions and 663 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
bench/blas.cpp
View File

@@ -111,7 +111,7 @@ void bench(sc::numeric_type dtype, std::string operation)
std::cout << "#" << operation << " (" << metric[operation] << ")" << std::endl; std::cout << "#" << operation << " (" << metric[operation] << ")" << std::endl;
std::cout << "\"N\""; std::cout << "\"N\"";
std::cout << " \"ISAAC\""; std::cout << " \"ISAAC\"";
std::cout << " \"ISAAC (Best impl.)\""; // std::cout << " \"ISAAC (Best impl.)\"";
#ifdef BENCH_CLBLAS #ifdef BENCH_CLBLAS
std::cout << " \"clBLAS\""; std::cout << " \"clBLAS\"";
#endif #endif
@@ -314,7 +314,7 @@ void bench(sc::numeric_type dtype, std::string operation)
int_t lda = A.stride()[1], ldb = B.stride()[1], ldc = C.stride()[1]; int_t lda = A.stride()[1], ldb = B.stride()[1], ldc = C.stride()[1];
#endif #endif
BENCHMARK_ISAAC(C = sc::execution_handler(AT?(BT?dot(A.T,B.T):dot(A.T,B)):(BT?dot(A,B.T):dot(A,B)), sc::execution_options_type(0, &events), sc::dispatcher_options_type(false)), (double)2*M*N*K/t); BENCHMARK_ISAAC(C = sc::execution_handler(AT?(BT?dot(A.T,B.T):dot(A.T,B)):(BT?dot(A,B.T):dot(A,B)), sc::execution_options_type(0, &events), sc::dispatcher_options_type(false)), (double)2*M*N*K/t);
BENCHMARK_ISAAC(C = sc::execution_handler(AT?(BT?dot(A.T,B.T):dot(A.T,B)):(BT?dot(A,B.T):dot(A,B)), sc::execution_options_type(0, &events), sc::dispatcher_options_type(true)), (double)2*M*N*K/t); // BENCHMARK_ISAAC(C = sc::execution_handler(AT?(BT?dot(A.T,B.T):dot(A.T,B)):(BT?dot(A,B.T):dot(A,B)), sc::execution_options_type(0, &events), sc::dispatcher_options_type(true)), (double)2*M*N*K/t);
/* clblas */ /* clblas */
#ifdef BENCH_CLBLAS #ifdef BENCH_CLBLAS
if(C.context().backend()==sc::driver::OPENCL) if(C.context().backend()==sc::driver::OPENCL)

1323
lib/kernels/templates/gemm.cpp
View File

File diff suppressed because it is too large Load Diff