Add performance monitoring for LLT

bench/perf_monitoring/gemm/changesets.txt

@@ -31,7 +31,6 @@ before-evaluators
 6845:7333ed40c6ef   # change prefetching in gebp
 #6856:b5be5e10eb7f   # merge index conversion
 6893:c3a64aba7c70   # clean blocking size computation
-6898:6fb31ebe6492   # rotating kernel for ARM
 6899:877facace746   # rotating kernel for ARM only
 #6904:c250623ae9fa   # result_of
 6921:915f1b1fc158   # fix prefetching change for ARM
@@ -50,7 +49,7 @@ before-evaluators
 7098:b6f1db9cf9ec   # Bug 992: don't select a 3p GEMM path with non-vectorizable scalar types, this hits unsupported paths in symm/triangular products code
 7591:09a8e2186610   # 3.3-alpha1
 7650:b0f3c8f43025   # help clang inlining
-8744:74b789ada92a   # Improved the matrix multiplication blocking in the case where mr is not a power of 2 (e.g on Haswell CPUs)
+#8744:74b789ada92a   # Improved the matrix multiplication blocking in the case where mr is not a power of 2 (e.g on Haswell CPUs)
 8789:efcb912e4356   # Made the index type a template parameter to evaluateProductBlockingSizes. Use numext::mini and numext::maxi instead of std::min/std::max to compute blocking sizes
 8972:81d53c711775   # Don't optimize the processing of the last rows of a matrix matrix product in cases that violate the assumptions made by the optimized code path
 8985:d935df21a082   # Remove the rotating kernel.
@@ -65,5 +64,5 @@ before-evaluators
 9942:b1d3eba60130   # Operators += and -= do not resize!
 9943:79bb9887afd4   # Ease compiler job to generate clean and efficient code in mat*vec
 9946:2213991340ea   # Complete rewrite of column-major-matrix * vector product to deliver higher performance of modern CPU.
-9953:21acc0e8d782   # Improve performance of row-major-dense-matrix * vector products for recent CPUs.
+9955:630471c3298c   # Improve performance of row-major-dense-matrix * vector products for recent CPUs. (this is the next changeset fixing a typo)
 

bench/perf_monitoring/gemm/gemm.cpp

@@ -1,17 +1,4 @@
-#include <iostream>
+#include "gemm_common.h"
-#include <fstream>
-#include <vector>
-#include <Eigen/Core>
-#include "../../BenchTimer.h"
-using namespace Eigen;
-
-#ifndef SCALAR
-#error SCALAR must be defined
-#endif
-
-typedef SCALAR Scalar;
-
-typedef Matrix<Scalar,Dynamic,Dynamic> Mat;
 
 EIGEN_DONT_INLINE
 void gemm(const Mat &A, const Mat &B, Mat &C)
@@ -19,52 +6,7 @@ void gemm(const Mat &A, const Mat &B, Mat &C)
   C.noalias() += A * B;
 }
 
-EIGEN_DONT_INLINE
-double bench(long m, long n, long k)
-{
-  Mat A(m,k);
-  Mat B(k,n);
-  Mat C(m,n);
-  A.setRandom();
-  B.setRandom();
-  C.setZero();
-  
-  BenchTimer t;
-  
-  double up = 1e8*4/sizeof(Scalar);
-  double tm0 = 4, tm1 = 10;
-  if(NumTraits<Scalar>::IsComplex)
-  {
-    up /= 4;
-    tm0 = 2;
-    tm1 = 4;
-  }
-  
-  double flops = 2. * m * n * k;
-  long rep = std::max(1., std::min(100., up/flops) );
-  long tries = std::max(tm0, std::min(tm1, up/flops) );
-  
-  BENCH(t, tries, rep, gemm(A,B,C));
-  
-  return 1e-9 * rep * flops / t.best();
-}
-
 int main(int argc, char **argv)
 {
-  std::vector<double> results;
+  return main_gemm(argc, argv, gemm);
-  
-  std::string filename = std::string("gemm_settings.txt");
-  if(argc>1)
-    filename = std::string(argv[1]);
-  std::ifstream settings(filename);
-  long m, n, k;
-  while(settings >> m >> n >> k)
-  {
-    //std::cerr << "  Testing " << m << " " << n << " " << k << std::endl;
-    results.push_back( bench(m, n, k) );
-  }
-  
-  std::cout << RowVectorXd::Map(results.data(), results.size());
-  
-  return 0;
 }

bench/perf_monitoring/gemm/gemm_common.h

@@ -0,0 +1,66 @@
+#include <iostream>
+#include <fstream>
+#include <vector>
+#include <Eigen/Core>
+#include "../../BenchTimer.h"
+using namespace Eigen;
+
+#ifndef SCALAR
+#error SCALAR must be defined
+#endif
+
+typedef SCALAR Scalar;
+
+typedef Matrix<Scalar,Dynamic,Dynamic> Mat;
+
+template<typename Func>
+EIGEN_DONT_INLINE
+double bench(long m, long n, long k, const Func& f)
+{
+  Mat A(m,k);
+  Mat B(k,n);
+  Mat C(m,n);
+  A.setRandom();
+  B.setRandom();
+  C.setZero();
+  
+  BenchTimer t;
+  
+  double up = 1e8*4/sizeof(Scalar);
+  double tm0 = 4, tm1 = 10;
+  if(NumTraits<Scalar>::IsComplex)
+  {
+    up /= 4;
+    tm0 = 2;
+    tm1 = 4;
+  }
+  
+  double flops = 2. * m * n * k;
+  long rep = std::max(1., std::min(100., up/flops) );
+  long tries = std::max(tm0, std::min(tm1, up/flops) );
+  
+  BENCH(t, tries, rep, f(A,B,C));
+  
+  return 1e-9 * rep * flops / t.best();
+}
+
+template<typename Func>
+int main_gemm(int argc, char **argv, const Func& f)
+{
+  std::vector<double> results;
+  
+  std::string filename = std::string("gemm_settings.txt");
+  if(argc>1)
+    filename = std::string(argv[1]);
+  std::ifstream settings(filename);
+  long m, n, k;
+  while(settings >> m >> n >> k)
+  {
+    //std::cerr << "  Testing " << m << " " << n << " " << k << std::endl;
+    results.push_back( bench(m, n, k, f) );
+  }
+  
+  std::cout << RowVectorXd::Map(results.data(), results.size());
+  
+  return 0;
+}

bench/perf_monitoring/gemm/gemm_square_settings.txt

@@ -0,0 +1,11 @@
+8 8 8
+9 9 9
+12 12 12
+15 15 15
+16 16 16
+24 24 24
+102 102 102
+239 239 239
+240 240 240
+2400 2400 2400
+2463 2463 2463

bench/perf_monitoring/gemm/gemv_common.h

@@ -47,14 +47,14 @@ double bench(long m, long n, Func &f)
 }
 
 template<typename Func>
-int main_gemv(int argc, char **argv, Func& f, const std::string &setting_filename)
+int main_gemv(int argc, char **argv, Func& f)
 {
   std::vector<double> results;
 
   std::string filename = std::string("gemv_settings.txt");
   if(argc>1)
     filename = std::string(argv[1]);
-  std::ifstream settings(setting_filename);
+  std::ifstream settings(filename);
   long m, n;
   while(settings >> m >> n)
   {

bench/perf_monitoring/gemm/gemv_square_settings.txt

@@ -0,0 +1,13 @@
+8 8
+9 9
+12 12
+15 15
+16 16
+24 24
+53 53
+74 74
+102 102
+239 239
+240 240
+2400 2400
+2463 2463

bench/perf_monitoring/gemm/llt.cpp

@@ -0,0 +1,15 @@
+#include "gemm_common.h"
+#include <Eigen/Cholesky>
+
+EIGEN_DONT_INLINE
+void llt(const Mat &A, const Mat &B, Mat &C)
+{
+  C = A;
+  C.diagonal().array() += 1000;
+  Eigen::internal::llt_inplace<Mat::Scalar, Lower>::blocked(C);
+}
+
+int main(int argc, char **argv)
+{
+  return main_gemm(argc, argv, llt);
+}

bench/perf_monitoring/gemm/runall.sh

@@ -19,4 +19,5 @@
 ./run.sh trmv_lo gemv_square_settings.txt $*
 ./run.sh trmv_upt gemv_square_settings.txt $*
 ./run.sh trmv_lot gemv_square_settings.txt $*
+./run.sh llt gemm_square_settings.txt $*
 

bench/perf_monitoring/gemm/trmv_lot.cpp

@@ -1,7 +1,7 @@
 #include "gemv_common.h"
 
 EIGEN_DONT_INLINE
-void gemv(const Mat &A, Vec &B, const Vec &C)
+void trmv(const Mat &A, Vec &B, const Vec &C)
 {
   B.noalias() += A.transpose().triangularView<Lower>() * C;
 }