Merge pull request dacelib#44 from abgandar/master

branch predictors and storage qualifiers
zhjsun · Oct 22, 2022 · df6901c · df6901c
2 parents 2bda904 + 7a56659
commit df6901c
Show file tree

Hide file tree

Showing 5 changed files with 14 additions and 14 deletions.
diff --git a/Tutorials/Tutorial2/1Basics-Ex.cpp b/Tutorials/Tutorial2/1Basics-Ex.cpp
@@ -70,5 +70,5 @@ int main( void )
     cout << "Exercise 1.2.2: Breaking bad\n"
          << ex1_2_2( 0.0, 0.0 ) << endl
          << ex1_2_2( x, y ) << endl;
-    } catch( exception &ex ) { cout << ex.what( ) << endl; }
+    } catch( exception const& ex ) { cout << ex.what( ) << endl; }
 }
diff --git a/core/daceaux.c b/core/daceaux.c
@@ -334,7 +334,7 @@ void daceDecode(const unsigned int jc, unsigned int jj[])
    \param[in] cc C array of nmmax monomials
    \param[in] inc Pointer to DA object to pack the monomials into
 */
-void dacePack(double cc[], DACEDA *inc)
+void dacePack(double cc[restrict], DACEDA *restrict inc)
 {
     monomial *ipoc; unsigned int ilmc, illc;
 
@@ -350,7 +350,7 @@ void dacePack(double cc[], DACEDA *inc)
         {
             for(unsigned int i = 0; i < DACECom.nmmax; i++)
             {
-                if(!(fabs(cc[i]) <= DACECom_t.eps))  //  && (DACECom.ieo[i] <= DACECom_t.nocut) is avoided for performance
+                if(LIKELY(!(fabs(cc[i]) <= DACECom_t.eps)))  //  && (DACECom.ieo[i] <= DACECom_t.nocut) is avoided for performance
                 {
                     ic->ii = i;
                     ic->cc = cc[i];
@@ -363,9 +363,9 @@ void dacePack(double cc[], DACEDA *inc)
         {
             for(unsigned int i = 0; i < DACECom.nmmax; i++)
             {
-                if(!(fabs(cc[i]) <= DACECom_t.eps) && DACECom.ieo[i] <= DACECom_t.nocut)     // here we remove also cut orders to save as much space as possible
+                if(LIKELY(!(fabs(cc[i]) <= DACECom_t.eps) && DACECom.ieo[i] <= DACECom_t.nocut))     // here we remove also cut orders to save as much space as possible
                 {
-                    if(ic >= ipoc+ilmc)
+                    if(UNLIKELY(ic >= ipoc+ilmc))
                     {
                         daceSetError(__func__, DACE_ERROR, 21);
                         for(unsigned int j = i; j < DACECom.nmmax; j++) cc[j] = 0.0;
@@ -384,9 +384,9 @@ void dacePack(double cc[], DACEDA *inc)
     {
         for(unsigned int i = 0; i < DACECom.nmmax; i++)
         {
-            if((DACECom.ifi[i] != 0) && !(fabs(cc[i]) <= DACECom_t.eps) && (DACECom.ieo[i] <= DACECom_t.nocut))
+            if(LIKELY((DACECom.ifi[i] != 0) && !(fabs(cc[i]) <= DACECom_t.eps) && (DACECom.ieo[i] <= DACECom_t.nocut)))
             {
-                if(ic >= ipoc+ilmc)
+                if(UNLIKELY(ic >= ipoc+ilmc))
                 {
                     daceSetError(__func__, DACE_ERROR, 21);
                     for(unsigned int j = 0; j < DACECom.nmmax; j++) cc[j] = 0.0;

diff --git a/core/dacemath.c b/core/dacemath.c
@@ -126,7 +126,7 @@ void daceMultiply(const DACEDA *ina, const DACEDA *inb, DACEDA *inc)
 			ipbeg[i] = emb + daceCountMonomials(i - 1, DACECom.nvmax);
     }
 #else
-    static DACE_THREAD_LOCAL double *cc = NULL;
+    static DACE_THREAD_LOCAL double *restrict cc = NULL;
     static DACE_THREAD_LOCAL extended_monomial *emb = NULL;
     static DACE_THREAD_LOCAL extended_monomial **ipbeg = NULL;
     static DACE_THREAD_LOCAL extended_monomial **ipend = NULL;
@@ -171,18 +171,18 @@ void daceMultiply(const DACEDA *ina, const DACEDA *inb, DACEDA *inc)
     for(unsigned int i = 0; i <= DACECom_t.nocut; i++) ipend[i] = ipbeg[i];
 
     // sort vector b by order
-    for(monomial *ib = ipob; ib < ipob+illb; ib++)
+    for(const monomial *ib = ipob; ib < ipob+illb; ib++)
     {
         const unsigned int noib = DACECom.ieo[ib->ii];
-        if(noib > DACECom_t.nocut) continue;
+        if(UNLIKELY(noib > DACECom_t.nocut)) continue;
         ipend[noib]->i1 = DACECom.ie1[ib->ii];
         ipend[noib]->i2 = DACECom.ie2[ib->ii];
         ipend[noib]->cc = ib->cc;
         ipend[noib]++;
     }
 
     // perform actual multiplication
-    for(monomial *ia = ipoa; ia < ipoa+illa; ia++)
+    for(const monomial *ia = ipoa; ia < ipoa+illa; ia++)
     {
         const unsigned int i1ia = DACECom.ie1[ia->ii];
         const unsigned int i2ia = DACECom.ie2[ia->ii];
@@ -191,7 +191,7 @@ void daceMultiply(const DACEDA *ina, const DACEDA *inb, DACEDA *inc)
         //#pragma omp parallel for
         for(int noib = DACECom_t.nocut-DACECom.ieo[ia->ii]; noib >= 0; noib--)
         {
-            for(extended_monomial *ib = ipbeg[noib]; ib < ipend[noib]; ib++)
+            for(const extended_monomial *ib = ipbeg[noib]; ib < ipend[noib]; ib++)
             {
                 const unsigned int ic = DACECom.ia1[i1ia+ib->i1] + DACECom.ia2[i2ia+ib->i2];
                 cc[ic] += ccia*ib->cc;

diff --git a/core/include/dace/daceaux.h b/core/include/dace/daceaux.h
@@ -143,7 +143,7 @@ unsigned int daceNextOrderedMonomial(unsigned int p[], const unsigned int no, co
 // internal routines
 void daceInitializeThread0();
 void daceSetError(const char *c, const unsigned int ix, const unsigned int iyy);
-void dacePack(double cc[], DACEDA *inc);
+void dacePack(double cc[restrict], DACEDA *restrict inc);
 void daceMultiplicativeInverse0(const DACEDA *ina, DACEDA *inc, const double a0);
 int BesselWrapper(const double x, const int n0, const int n1, const int type, double *bz);
 int ModifiedBesselWrapper(const double x, const int n0, const int n1, const int type, double *bz);

diff --git a/interfaces/cxx/DA.cpp b/interfaces/cxx/DA.cpp
@@ -66,7 +66,7 @@ void DA::init(const unsigned int ord, const unsigned int nvar) {
  */
     try {
         checkVersion();
-    } catch(DACEException ex) {
+    } catch(DACEException const& ex) {
         std::cerr << ex << std::endl;
         std::terminate(); }
     daceInitialize(ord,nvar);