GPU: Add missing static_assert to guarantee correct unroll factor, will crash otherwise (apparently forgotten by AMD)

davidrohr · davidrohr · commit e966e71fcd20 · 2025-04-18T07:25:35.000+02:00
diff --git a/GPU/GPUTracking/SectorTracker/GPUTPCNeighboursFinder.cxx b/GPU/GPUTracking/SectorTracker/GPUTPCNeighboursFinder.cxx
@@ -76,6 +76,7 @@ GPUdii() void GPUTPCNeighboursFinder::Thread<0>(int32_t /*nBlocks*/, int32_t nTh
   }
 
   static constexpr uint32_t UNROLL_GLOBAL = 4;
+  static_assert(GPUCA_MAXN % UNROLL_GLOBAL == 0);
   static constexpr uint32_t MAX_SHARED = GPUCA_PAR_NEIGHBOURS_FINDER_MAX_NNEIGHUP;
   static constexpr uint32_t MAX_GLOBAL = (MAX_SHARED < GPUCA_MAXN) ? (((GPUCA_MAXN - MAX_SHARED - 1) / UNROLL_GLOBAL + 1) * UNROLL_GLOBAL) : 0;
   static constexpr uint32_t MAX_TOTAL = MAX_SHARED + MAX_GLOBAL;

Original file line number	Diff line number	Diff line change
`@@ -76,6 +76,7 @@ GPUdii() void GPUTPCNeighboursFinder::Thread<0>(int32_t /nBlocks/, int32_t nTh`
`76`	`76`	`}`
`77`	`77`
`78`	`78`	`static constexpr uint32_t UNROLL_GLOBAL = 4;`
	`79`	`+ static_assert(GPUCA_MAXN % UNROLL_GLOBAL == 0);`
`79`	`80`	`static constexpr uint32_t MAX_SHARED = GPUCA_PAR_NEIGHBOURS_FINDER_MAX_NNEIGHUP;`
`80`	`81`	`static constexpr uint32_t MAX_GLOBAL = (MAX_SHARED < GPUCA_MAXN) ? (((GPUCA_MAXN - MAX_SHARED - 1) / UNROLL_GLOBAL + 1) * UNROLL_GLOBAL) : 0;`
`81`	`82`	`static constexpr uint32_t MAX_TOTAL = MAX_SHARED + MAX_GLOBAL;`