Slight code change to avoid generation of partial store

Small code change to void clearing of member t before setting it which
resulted in suboptimal code generation and a performance penalty.

Author: fda0

IGC/BiFModule/Languages/OpenCL/Raytracing/IBiF_intel_rt_production.cl

@@ -74,10 +74,17 @@ intel_ray_query_t intel_ray_query_init(
     MemRay_setRayFlags(&rtStack->ray[bvh_level],    ray.flags);
     MemRay_setRayMask(&rtStack->ray[bvh_level],     ray.mask);
 
-    MemHit_clear(&rtStack->hit[COMMITTED], /*done=*/0, /*valid=*/0);
-    MemHit_clear(&rtStack->hit[POTENTIAL], /*done=*/1, /*valid=*/1);
+    rtStack->hit[COMMITTED].u = rtStack->hit[COMMITTED].v = 0.0f;
     rtStack->hit[COMMITTED].t = INFINITY;
+    rtStack->hit[COMMITTED].data0 = 0;
+    MemHit_setValid(&rtStack->hit[COMMITTED], 0);
+    MemHit_setDone(&rtStack->hit[COMMITTED], 0);
+
+    rtStack->hit[POTENTIAL].u = rtStack->hit[POTENTIAL].v = 0.0f;
     rtStack->hit[POTENTIAL].t = INFINITY;
+    rtStack->hit[POTENTIAL].data0 = 0;
+    MemHit_setValid(&rtStack->hit[POTENTIAL], 1);
+    MemHit_setDone(&rtStack->hit[POTENTIAL], 1);
 
     intel_ray_query_t rayquery = __builtin_IB_intel_init_ray_query(
         NULL,

IGC/BiFModule/Languages/OpenCL/Raytracing/IBiF_intel_rt_struct_defs.cl

@@ -149,14 +149,6 @@ inline ulong MemHit_getHitGroupRecPtr1(MemHit* memhit) { return getBits64(memhit
 inline global void* MemHit_getPrimLeafPtr(MemHit* memhit)     { return to_global((void*)((ulong)MemHit_getPrimLeafAddr(memhit) * 64)); }
 inline global void* MemHit_getInstanceLeafPtr(MemHit* memhit) { return to_global((void*)((ulong)MemHit_getInstLeafAddr(memhit) * 64)); }
 
-inline void MemHit_clear(MemHit* memhit, bool _done, bool _valid)
-{
-    memhit->t = memhit->u = memhit->v = 0.0f;
-    memhit->data0 = 0;
-    MemHit_setValid(memhit, _valid);
-    MemHit_setDone(memhit, _done);
-}
-
 
 // === --------------------------------------------------------------------===
 // === RTStack