going into extremes to reduce more bytes, 32bytes saved

Author: planetis-m

src/computesim.nim

@@ -140,8 +140,7 @@ proc subgroupProc[A, B, C](wg: WorkGroupContext; numActiveThreads: uint32; barri
         globalOffsetX + x,
         globalOffsetY + y,
         globalOffsetZ + z
-      ),
-      gl_SubgroupInvocationID: threadId
+      )
     )
     # Update coordinates
     inc x

src/computesim/core.nim

@@ -96,7 +96,6 @@ type
   ThreadContext* = object
     gl_GlobalInvocationID*: UVec3
     gl_LocalInvocationID*: UVec3
-    gl_SubgroupInvocationID*: uint32
 
   BarrierHandle* = object
     x: ptr Barrier
@@ -129,8 +128,8 @@ proc wait*(m: BarrierHandle) {.inline.} =
   wait(m.x[])
 
 type
-  ThreadClosure* = iterator (iterArg: SubgroupResult,
-                             wg: WorkGroupContext, thread: ThreadContext): SubgroupCommand
+  ThreadClosure* = iterator (iterArg: SubgroupResult, wg: WorkGroupContext,
+                             thread: ThreadContext, threadId: uint32): SubgroupCommand
   SubgroupResults* = array[SubgroupSize, SubgroupResult]
   SubgroupCommands* = array[SubgroupSize, SubgroupCommand]
   SubgroupThreadIDs* = array[SubgroupSize, uint32]

src/computesim/lockstep.nim

@@ -62,7 +62,7 @@ proc runThreads*(threads: SubgroupThreads; workGroup: WorkGroupContext,
           threadStates[threadId] == running or canReconverge or canPassBarrier:
         madeProgress = true
         {.cast(gcsafe).}:
-          commands[threadId] = threads[threadId](results[threadId], workGroup, threadContexts[threadId])
+          commands[threadId] = threads[threadId](results[threadId], workGroup, threadContexts[threadId], threadId)
         if finished(threads[threadId]):
           threadStates[threadId] = finished
         elif commands[threadId].kind == barrier:

src/computesim/transform.nim

@@ -153,8 +153,7 @@ proc generateWorkGroupTemplates(wgSym: NimNode): NimNode =
 proc generateThreadTemplates(threadSym: NimNode): NimNode =
   generateTemplates(threadSym, [
     "gl_GlobalInvocationID",
-    "gl_LocalInvocationID",
-    "gl_SubgroupInvocationID"
+    "gl_LocalInvocationID"
   ])
 
 proc isDiscard(n: NimNode, op: SubgroupOp): bool =
@@ -273,16 +272,18 @@ macro computeShader*(prc: untyped): untyped =
   # Create symbols for both contexts
   let wgSym = genSym(nskParam, "wg")
   let threadSym = genSym(nskParam, "thread")
+  let tidSym = genSym(nskParam, "threadId")
   # Generate template declarations for both contexts
   let wgTemplates = generateWorkGroupTemplates(wgSym)
   let threadTemplates = generateThreadTemplates(threadSym)
 
   result = quote do:
     proc `procName`(): ThreadClosure =
       iterator (`iterArg`: SubgroupResult, `wgSym`: WorkGroupContext,
-                `threadSym`: ThreadContext): SubgroupCommand =
-        `wgTemplates`
+                `threadSym`: ThreadContext, `tidSym`: uint32): SubgroupCommand =
+        template gl_SubgroupInvocationID(): uint32 {.used.} = `tidSym`
         `threadTemplates`
+        `wgTemplates`
         `traversedBody`
 
   # Now inject the parameters and pragmas from original proc