Attempt to fix overloaded bar (again)

preda · Dec 23, 2024 · 002f686 · 002f686
1 parent e52572e
commit 002f686
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 4 deletions.
diff --git a/src/cl/base.cl b/src/cl/base.cl
@@ -172,7 +172,7 @@ T2 U2(T a, T b) { return (T2) (a, b); }
 #endif
 #endif
 
-OVERLOAD void bar() {
+void OVERLOAD bar(void) {
   // barrier(CLK_LOCAL_MEM_FENCE) is correct, but it turns out that on some GPUs
   // (in particular on Radeon VII and Radeon PRO VII) barrier(0) works as well and is faster.
   // So allow selecting the faster path when it works with -use FAST_BARRIER
@@ -183,7 +183,7 @@ OVERLOAD void bar() {
 #endif
 }
 
-OVERLOAD void bar(u32 WG) { if (WG > WAVEFRONT) { bar(); } }
+void OVERLOAD bar(u32 WG) { if (WG > WAVEFRONT) { bar(); } }
 
 // A half-barrier is only needed when half-a-workgroup needs a barrier.
 // This is used e.g. by the double-wide tailSquare, where LDS is split between the halves.

diff --git a/src/cl/math.cl b/src/cl/math.cl
@@ -5,8 +5,8 @@
 #include "base.cl"
 
 // a * (b + 1) == a * b + a
-OVERLOAD T  fancyMul(T a, T b)   { return fma(a, b, a); }
-OVERLOAD T2 fancyMul(T2 a, T2 b) { return U2(fancyMul(a.x, b.x), fancyMul(a.y, b.y)); }
+T OVERLOAD fancyMul(T a, T b)   { return fma(a, b, a); }
+T2 OVERLOAD fancyMul(T2 a, T2 b) { return U2(fancyMul(a.x, b.x), fancyMul(a.y, b.y)); }
 
 T2 cmul(T2 a, T2 b) {
 #if 1