diff options
| -rw-r--r-- | Eigen/src/Core/util/Memory.h | 4 | ||||
| -rw-r--r-- | bench/bench_gemm.cpp | 14 |
2 files changed, 11 insertions, 7 deletions
diff --git a/Eigen/src/Core/util/Memory.h b/Eigen/src/Core/util/Memory.h index 1b1ef6712..810bbfacd 100644 --- a/Eigen/src/Core/util/Memory.h +++ b/Eigen/src/Core/util/Memory.h @@ -601,7 +601,7 @@ public: /** \internal * \returns the size in Bytes of the L1 data cache */ -inline std::ptrdiff_t ei_fetchL1CacheSize() +inline std::ptrdiff_t ei_queryL1CacheSize() { int abcd[4]; @@ -644,7 +644,7 @@ inline std::ptrdiff_t ei_fetchL1CacheSize() /** \internal * \returns the size in Bytes of the L2 or L3 cache if this later is present */ -inline std::ptrdiff_t ei_fetchTopLevelCacheSize() +inline std::ptrdiff_t ei_queryTopLevelCacheSize() { int abcd[4]; EIGEN_CPUID(abcd,0x80000006); diff --git a/bench/bench_gemm.cpp b/bench/bench_gemm.cpp index 7b12736da..ee34e6ddc 100644 --- a/bench/bench_gemm.cpp +++ b/bench/bench_gemm.cpp @@ -57,6 +57,7 @@ void blas_gemm(const MatrixXd& a, const MatrixXd& b, MatrixXd& c) #endif +template<typename M> void gemm(const M& a, const M& b, M& c) { c.noalias() += a * b; @@ -64,6 +65,9 @@ void gemm(const M& a, const M& b, M& c) int main(int argc, char ** argv) { + std::cout << "L1 cache size = " << ei_queryL1CacheSize()/1024 << " KB\n"; + std::cout << "L2/L3 cache size = " << ei_queryTopLevelCacheSize()/1024 << " KB\n"; + int rep = 1; // number of repetitions per try int tries = 5; // number of tries, we keep the best @@ -90,11 +94,6 @@ int main(int argc, char ** argv) if(cache_size>0) setCpuCacheSizes(cache_size,32*cache_size); - std::cout << "Matrix size = " << s << "\n"; - std::ptrdiff_t cm, cn, ck; - getBlockingSizes<Scalar>(ck, cm, cn); - std::cout << "blocking size = " << cm << " x " << ck << "\n"; - int m = s; int n = s; int p = s; @@ -102,6 +101,11 @@ int main(int argc, char ** argv) M b(n,p); b.setRandom(); M c(m,p); c.setOnes(); + std::cout << "Matrix sizes = " << m << "x" << p << " * " << p << "x" << n << "\n"; + std::ptrdiff_t cm, cn, ck; + getBlockingSizes<Scalar>(ck, cm, cn); + std::cout << "blocking size = " << cm << " x " << ck << "\n"; + M r = c; // check the parallel product is correct |