closes #17

Author: planetis-m

src/computesim/core.nim

@@ -140,7 +140,6 @@ const
   InvalidId* = high(uint32) # Sentinel value for empty/invalid
 
 iterator threadsInGroup*(group: SubgroupThreadIDs): uint32 =
-  var idx: uint32 = 0
-  while group[idx] != InvalidId:
-    yield group[idx]
-    inc idx
+  let length = group[0]
+  for i in 1..length:
+    yield group[i]

src/computesim/debug.nim

@@ -38,7 +38,7 @@ proc formatThreadList(group: SubgroupThreadIDs): string =
   result.add("}")
 
 proc formatThreadValues(group: SubgroupThreadIDs, commands: SubgroupCommands): string =
-  let firstThreadId = group[0]
+  let firstThreadId = group[1]
   let opKind = commands[firstThreadId].kind
   case opKind
   of subgroupBroadcast, subgroupBroadcastFirst, subgroupAdd, subgroupMin, subgroupMax,

src/computesim/lockstep.nim

@@ -116,27 +116,25 @@ proc runThreads*(threads: SubgroupThreads; workGroup: WorkGroupContext,
           (threadStates[threadId] == atSubBarrier and canReconverge) or canPassBarrier):
         var found = false
         for groupIdx in 0..<numGroups:
-          let firstThreadId = threadGroups[groupIdx][0]
+          let firstThreadId = threadGroups[groupIdx][1] # First thread is at index 1
           if commands[firstThreadId].id == commands[threadId].id:
-            # Find first empty slot in group
-            for slot in 0..<SubgroupSize:
-              if threadGroups[groupIdx][slot] == InvalidId:
-                threadGroups[groupIdx][slot] = threadId
-                threadGroups[groupIdx][slot + 1] = InvalidId
-                break
+            let currentLen = threadGroups[groupIdx][0]
+            threadGroups[groupIdx][currentLen + 1] = threadId
+            threadGroups[groupIdx][0] = currentLen + 1
             found = true
             break
         if not found:
-          threadGroups[numGroups][0] = threadId
-          threadGroups[numGroups][1] = InvalidId
+          # Create new group
+          threadGroups[numGroups][0] = 1 # Length is 1
+          threadGroups[numGroups][1] = threadId
           inc numGroups
 
     template execSubgroupOp(op: untyped) =
       op(results, commands, threadGroups[groupIdx], firstThreadId, opId, showDebugOutput)
 
     # Process operation groups
     for groupIdx in 0..<numGroups:
-      let firstThreadId = threadGroups[groupIdx][0]
+      let firstThreadId = threadGroups[groupIdx][1]
       let opKind = commands[firstThreadId].kind
       let opId = commands[firstThreadId].id
       case opKind: