Use __int128 in multiply9x9 if available

This reduces the time per number from around 19ns to less than 11ns if the compiler exposes this non-standard extension. Add test to make sure that the portable fallback path gives the same results.
root-project · Nov 23, 2020 · e37bfe3 · e37bfe3
1 parent 2856d17
commit e37bfe3
Show file tree

Hide file tree

Showing 5 changed files with 41 additions and 5 deletions.
diff --git a/math/mathcore/src/mulmod.h b/math/mathcore/src/mulmod.h
@@ -45,11 +45,20 @@ void multiply9x9(const uint64_t *in1, const uint64_t *in2, uint64_t *out)
          if (k < 0 || k >= 9)
             continue;
 
-         uint64_t upper1 = in1[j] >> 32;
-         uint64_t lower1 = static_cast<uint32_t>(in1[j]);
+         uint64_t fac1 = in1[j];
+         uint64_t fac2 = in2[k];
+#if defined(__SIZEOF_INT128__) && !defined(ROOT_NO_INT128)
+         unsigned __int128 prod = fac1;
+         prod = prod * fac2;
 
-         uint64_t upper2 = in2[k] >> 32;
-         uint64_t lower2 = static_cast<uint32_t>(in2[k]);
+         uint64_t upper = prod >> 64;
+         uint64_t lower = static_cast<uint64_t>(prod);
+#else
+         uint64_t upper1 = fac1 >> 32;
+         uint64_t lower1 = static_cast<uint32_t>(fac1);
+
+         uint64_t upper2 = fac2 >> 32;
+         uint64_t lower2 = static_cast<uint32_t>(fac2);
 
          // Multiply 32-bit parts, each product has a maximum value of
          // (2 ** 32 - 1) ** 2 = 2 ** 64 - 2 * 2 ** 32 + 1.
@@ -98,6 +107,7 @@ void multiply9x9(const uint64_t *in1, const uint64_t *in2, uint64_t *out)
          //     (2 ** 64 - 1) ** 2 = 2 ** 128 - 2 * 2 ** 64 + 1
          // with the upper bits matching the 2 ** 64 - 2 of the first case.
          upper += middle_upper;
+#endif
 
          // Add to current, remember carry.
          current += lower;

diff --git a/math/mathcore/test/CMakeLists.txt b/math/mathcore/test/CMakeLists.txt
@@ -79,7 +79,8 @@ ROOT_ADD_GTEST(GradientUnit testGradient.cxx LIBRARIES Core MathCore Hist )
 
 ROOT_ADD_GTEST(GradientFittingUnit testGradientFitting.cxx LIBRARIES Core MathCore Hist)
 
-ROOT_ADD_GTEST(MulmodUnit mulmod.cxx)
+ROOT_ADD_GTEST(MulmodUnitOpt mulmod_opt.cxx)
+ROOT_ADD_GTEST(MulmodUnitNoInt128 mulmod_noint128.cxx)
 
 ROOT_ADD_GTEST(RanluxppEngineTests RanluxppEngine.cxx
         LIBRARIES Core MathCore)

diff --git a/math/mathcore/test/mulmod.cxx → math/mathcore/test/mulmod.icc b/math/mathcore/test/mulmod.cxx → math/mathcore/test/mulmod.icc
diff --git a/math/mathcore/test/mulmod_noint128.cxx b/math/mathcore/test/mulmod_noint128.cxx
@@ -0,0 +1,13 @@
+// @(#)root/mathcore:$Id$
+// Author: Jonas Hahnfeld 11/2020
+
+/*************************************************************************
+ * Copyright (C) 1995-2020, Rene Brun and Fons Rademakers.               *
+ * All rights reserved.                                                  *
+ *                                                                       *
+ * For the licensing terms see $ROOTSYS/LICENSE.                         *
+ * For the list of contributors see $ROOTSYS/README/CREDITS.             *
+ *************************************************************************/
+
+#define ROOT_NO_INT128
+#include "mulmod.icc"
diff --git a/math/mathcore/test/mulmod_opt.cxx b/math/mathcore/test/mulmod_opt.cxx
@@ -0,0 +1,12 @@
+// @(#)root/mathcore:$Id$
+// Author: Jonas Hahnfeld 11/2020
+
+/*************************************************************************
+ * Copyright (C) 1995-2020, Rene Brun and Fons Rademakers.               *
+ * All rights reserved.                                                  *
+ *                                                                       *
+ * For the licensing terms see $ROOTSYS/LICENSE.                         *
+ * For the list of contributors see $ROOTSYS/README/CREDITS.             *
+ *************************************************************************/
+
+#include "mulmod.icc"