Added argument to interaction so that masking only occurs when needed.

src/hydro/Gadget2/hydro_iact.h

@@ -484,7 +484,7 @@ runner_iact_nonsym_2_vec_density(
     vector viy, vector viz, float *Vjx, float *Vjy, float *Vjz, float *Mj,
     vector *rhoSum, vector *rho_dhSum, vector *wcountSum, vector *wcount_dhSum,
     vector *div_vSum, vector *curlvxSum, vector *curlvySum, vector *curlvzSum,
-    mask_t mask, mask_t mask2) {
+    mask_t mask, mask_t mask2, short mask_cond) {
 
   vector r, ri, r2, xi, wi, wi_dx;
   vector mj;
@@ -615,24 +615,47 @@ runner_iact_nonsym_2_vec_density(
       curlvzSum->v, knlMask2, vec_mul(mj2.v, vec_mul(curlvrz2.v, wi_dx2.v)),
       curlvzSum->v);
 #else
-  rhoSum->v = vec_mask_add(rhoSum->v, vec_mul(mj.v, wi.v), mask);
-  rhoSum->v = vec_mask_add(rhoSum->v, vec_mul(mj2.v, wi2.v), mask2);
-  rho_dhSum->v = vec_mask_sub(rho_dhSum->v, vec_mul(mj.v, vec_fma(vec_set1(hydro_dimension), wi.v,
-                                                vec_mul(xi.v, wi_dx.v))), mask);
-  rho_dhSum->v = vec_mask_sub(rho_dhSum->v, vec_mul(mj2.v, vec_fma(vec_set1(hydro_dimension), wi2.v,
-                                                vec_mul(xi2.v, wi_dx2.v))), mask2);
-  wcountSum->v = vec_mask_add(wcountSum->v, wi.v, mask);
-  wcountSum->v = vec_mask_add(wcountSum->v, wi2.v, mask2);
-  wcount_dhSum->v = vec_mask_sub(wcount_dhSum->v, vec_mul(xi.v, wi_dx.v), mask);
-  wcount_dhSum->v = vec_mask_sub(wcount_dhSum->v, vec_mul(xi2.v, wi_dx2.v), mask2);
-  div_vSum->v = vec_mask_sub(div_vSum->v, vec_mul(mj.v, vec_mul(dvdr.v, wi_dx.v)), mask);
-  div_vSum->v = vec_mask_sub(div_vSum->v, vec_mul(mj2.v, vec_mul(dvdr2.v, wi_dx2.v)), mask2);
-  curlvxSum->v = vec_mask_add(curlvxSum->v,vec_mul(mj.v, vec_mul(curlvrx.v, wi_dx.v)), mask);
-  curlvxSum->v = vec_mask_add(curlvxSum->v,vec_mul(mj2.v, vec_mul(curlvrx2.v, wi_dx2.v)), mask2);
-  curlvySum->v = vec_mask_add(curlvySum->v,vec_mul(mj.v, vec_mul(curlvry.v, wi_dx.v)), mask);
-  curlvySum->v = vec_mask_add(curlvySum->v,vec_mul(mj2.v, vec_mul(curlvry2.v, wi_dx2.v)), mask2);
-  curlvzSum->v = vec_mask_add(curlvzSum->v,vec_mul(mj.v, vec_mul(curlvrz.v, wi_dx.v)), mask);
-  curlvzSum->v = vec_mask_add(curlvzSum->v,vec_mul(mj2.v, vec_mul(curlvrz2.v, wi_dx2.v)), mask2);
+  /* Mask only when needed. */
+  if(mask_cond) {
+    rhoSum->v = vec_mask_add(rhoSum->v, vec_mul(mj.v, wi.v), mask);
+    rhoSum->v = vec_mask_add(rhoSum->v, vec_mul(mj2.v, wi2.v), mask2);
+    rho_dhSum->v = vec_mask_sub(rho_dhSum->v, vec_mul(mj.v, vec_fma(vec_set1(hydro_dimension), wi.v,
+            vec_mul(xi.v, wi_dx.v))), mask);
+    rho_dhSum->v = vec_mask_sub(rho_dhSum->v, vec_mul(mj2.v, vec_fma(vec_set1(hydro_dimension), wi2.v,
+            vec_mul(xi2.v, wi_dx2.v))), mask2);
+    wcountSum->v = vec_mask_add(wcountSum->v, wi.v, mask);
+    wcountSum->v = vec_mask_add(wcountSum->v, wi2.v, mask2);
+    wcount_dhSum->v = vec_mask_sub(wcount_dhSum->v, vec_mul(xi.v, wi_dx.v), mask);
+    wcount_dhSum->v = vec_mask_sub(wcount_dhSum->v, vec_mul(xi2.v, wi_dx2.v), mask2);
+    div_vSum->v = vec_mask_sub(div_vSum->v, vec_mul(mj.v, vec_mul(dvdr.v, wi_dx.v)), mask);
+    div_vSum->v = vec_mask_sub(div_vSum->v, vec_mul(mj2.v, vec_mul(dvdr2.v, wi_dx2.v)), mask2);
+    curlvxSum->v = vec_mask_add(curlvxSum->v,vec_mul(mj.v, vec_mul(curlvrx.v, wi_dx.v)), mask);
+    curlvxSum->v = vec_mask_add(curlvxSum->v,vec_mul(mj2.v, vec_mul(curlvrx2.v, wi_dx2.v)), mask2);
+    curlvySum->v = vec_mask_add(curlvySum->v,vec_mul(mj.v, vec_mul(curlvry.v, wi_dx.v)), mask);
+    curlvySum->v = vec_mask_add(curlvySum->v,vec_mul(mj2.v, vec_mul(curlvry2.v, wi_dx2.v)), mask2);
+    curlvzSum->v = vec_mask_add(curlvzSum->v,vec_mul(mj.v, vec_mul(curlvrz.v, wi_dx.v)), mask);
+    curlvzSum->v = vec_mask_add(curlvzSum->v,vec_mul(mj2.v, vec_mul(curlvrz2.v, wi_dx2.v)), mask2);
+  }
+  else {
+    rhoSum->v += vec_mul(mj.v, wi.v);
+    rhoSum->v += vec_mul(mj2.v, wi2.v);
+    rho_dhSum->v -= vec_mul(mj.v, vec_fma(vec_set1(hydro_dimension), wi.v,
+            vec_mul(xi.v, wi_dx.v)));
+    rho_dhSum->v -= vec_mul(mj2.v, vec_fma(vec_set1(hydro_dimension), wi2.v,
+            vec_mul(xi2.v, wi_dx2.v)));
+    wcountSum->v += wi.v;
+    wcountSum->v += wi2.v;
+    wcount_dhSum->v -= vec_mul(xi.v, wi_dx.v);
+    wcount_dhSum->v -= vec_mul(xi2.v, wi_dx2.v);
+    div_vSum->v -= vec_mul(mj.v, vec_mul(dvdr.v, wi_dx.v));
+    div_vSum->v -= vec_mul(mj2.v, vec_mul(dvdr2.v, wi_dx2.v));
+    curlvxSum->v += vec_mul(mj.v, vec_mul(curlvrx.v, wi_dx.v));
+    curlvxSum->v += vec_mul(mj2.v, vec_mul(curlvrx2.v, wi_dx2.v));
+    curlvySum->v += vec_mul(mj.v, vec_mul(curlvry.v, wi_dx.v));
+    curlvySum->v += vec_mul(mj2.v, vec_mul(curlvry2.v, wi_dx2.v));
+    curlvzSum->v += vec_mul(mj.v, vec_mul(curlvrz.v, wi_dx.v));
+    curlvzSum->v += vec_mul(mj2.v, vec_mul(curlvrz2.v, wi_dx2.v));
+  }
 #endif
 }
 #endif

src/runner_doiact_vec.c

@@ -109,7 +109,7 @@ __attribute__((always_inline)) INLINE static void calcRemInteractions(
         v_hi_inv, v_vix, v_viy, v_viz, &int_cache->vxq[*icount_align],
         &int_cache->vyq[*icount_align], &int_cache->vzq[*icount_align],
         &int_cache->mq[*icount_align], rhoSum, rho_dhSum, wcountSum,
-        wcount_dhSum, div_vSum, curlvxSum, curlvySum, curlvzSum, int_mask, int_mask2);
+        wcount_dhSum, div_vSum, curlvxSum, curlvySum, curlvzSum, int_mask, int_mask2, 1);
   }
 }
 
@@ -231,7 +231,7 @@ __attribute__((always_inline)) INLINE static void storeInteractions(
           &int_cache->dzq[pjd], v_hi_inv, v_vix, v_viy, v_viz,
           &int_cache->vxq[pjd], &int_cache->vyq[pjd], &int_cache->vzq[pjd],
           &int_cache->mq[pjd], rhoSum, rho_dhSum, wcountSum, wcount_dhSum,
-          div_vSum, curlvxSum, curlvySum, curlvzSum, int_mask, int_mask2);
+          div_vSum, curlvxSum, curlvySum, curlvzSum, int_mask, int_mask2, 0);
     }
 
     /* Reset interaction count. */
@@ -777,7 +777,7 @@ __attribute__((always_inline)) INLINE void runner_doself1_density_vec(
           &int_cache.dzq[pjd], v_hi_inv, v_vix, v_viy, v_viz,
           &int_cache.vxq[pjd], &int_cache.vyq[pjd], &int_cache.vzq[pjd],
           &int_cache.mq[pjd], &rhoSum, &rho_dhSum, &wcountSum, &wcount_dhSum,
-          &div_vSum, &curlvxSum, &curlvySum, &curlvzSum, int_mask, int_mask2);
+          &div_vSum, &curlvxSum, &curlvySum, &curlvzSum, int_mask, int_mask2, 0);
     }
 
     /* Perform horizontal adds on vector sums and store result in particle pi.