diff options
| -rw-r--r-- | Eigen/src/Core/products/GeneralMatrixMatrix.h | 30 | ||||
| -rw-r--r-- | bench/bench_gemm.cpp | 2 |
2 files changed, 2 insertions, 30 deletions
diff --git a/Eigen/src/Core/products/GeneralMatrixMatrix.h b/Eigen/src/Core/products/GeneralMatrixMatrix.h index cf42855eb..b20a16cd0 100644 --- a/Eigen/src/Core/products/GeneralMatrixMatrix.h +++ b/Eigen/src/Core/products/GeneralMatrixMatrix.h @@ -96,12 +96,6 @@ static void run(int rows, int cols, int depth, Scalar* w = ei_aligned_stack_new(Scalar, sizeW); Scalar* blockB = (Scalar*)info[tid].blockB; - // if you have the GOTO blas library you can try our parallelization strategy - // using GOTO's optimized routines. - #ifdef USEGOTOROUTINES - void* u = alloca(4096+sizeW); - #endif - // For each horizontal panel of the rhs, and corresponding panel of the lhs... // (==GEMM_VAR1) for(int k=0; k<depth; k+=kc) @@ -110,28 +104,18 @@ static void run(int rows, int cols, int depth, // In order to reduce the chance that a thread has to wait for the other, // let's start by packing A'. - #ifndef USEGOTOROUTINES pack_lhs(blockA, &lhs(0,k), lhsStride, actual_kc, mc); - #else - sgemm_itcopy(actual_kc, mc, &lhs(0,k), lhsStride, blockA); - #endif - // Pack B_k to B' in parallel fashion: // each thread packs the sub block B_k,j to B'_j where j is the thread id. - // However, before copying to B'_j, we have to make sure that no other thread is still using it, // i.e., we test that info[tid].users equals 0. // Then, we set info[tid].users to the number of threads to mark that all other threads are going to use it. while(info[tid].users!=0) {} info[tid].users += threads; - #ifndef USEGOTOROUTINES pack_rhs(blockB+info[tid].rhs_start*kc, &rhs(k,info[tid].rhs_start), rhsStride, alpha, actual_kc, info[tid].rhs_length); - #else - sgemm_oncopy(actual_kc, info[tid].rhs_length, &rhs(k,info[tid].rhs_start), rhsStride, blockB+info[tid].rhs_start*kc); - #endif // Notify the other threads that the part B'_j is ready to go. info[tid].sync = k; @@ -147,12 +131,7 @@ static void run(int rows, int cols, int depth, if(shift>0) while(info[j].sync!=k) {} - #ifndef USEGOTOROUTINES gebp(res+info[j].rhs_start*resStride, resStride, blockA, blockB+info[j].rhs_start*kc, mc, actual_kc, info[j].rhs_length, -1,-1,0,0, w); - #else - sgemm_kernel(mc, info[j].rhs_length, actual_kc, alpha, blockA, blockB+info[j].rhs_start*kc, res+info[j].rhs_start*resStride, resStride); - #endif - } // Then keep going as usual with the remaining A' @@ -161,18 +140,10 @@ static void run(int rows, int cols, int depth, const int actual_mc = std::min(i+mc,rows)-i; // pack A_i,k to A' - #ifndef USEGOTOROUTINES pack_lhs(blockA, &lhs(i,k), lhsStride, actual_kc, actual_mc); - #else - sgemm_itcopy(actual_kc, actual_mc, &lhs(i,k), lhsStride, blockA); - #endif // C_i += A' * B' - #ifndef USEGOTOROUTINES gebp(res+i, resStride, blockA, blockB, actual_mc, actual_kc, cols, -1,-1,0,0, w); - #else - sgemm_kernel(actual_mc, cols, actual_kc, alpha, blockA, blockB, res+i, resStride); - #endif } // Release all the sub blocks B'_j of B' for the current thread, @@ -188,6 +159,7 @@ static void run(int rows, int cols, int depth, else #endif // EIGEN_HAS_OPENMP { + (void)info; // info is not used // this is the sequential version! Scalar* blockA = ei_aligned_stack_new(Scalar, kc*mc); std::size_t sizeB = kc*Blocking::PacketSize*Blocking::nr + kc*cols; diff --git a/bench/bench_gemm.cpp b/bench/bench_gemm.cpp index 653a880a8..3cb75c17a 100644 --- a/bench/bench_gemm.cpp +++ b/bench/bench_gemm.cpp @@ -3,7 +3,7 @@ // icpc bench_gemm.cpp -I .. -O3 -DNDEBUG -lrt -openmp && OMP_NUM_THREADS=2 ./a.out #include <Eigen/Core> - +#include <iostream> #include <bench/BenchTimer.h> using namespace std; |