Avoid double-storing the done bit in new raytracing

After initializating the ray, we don't have to store the done bit again. So let's just skip that part.
Also fixes a small bug with proceed result not being inserted in a correct block

Author: jaladreips

IGC/AdaptorCommon/RayTracing/AutoGenRTStackAccessPrivateOS.h

@@ -2541,7 +2541,7 @@ void _createTraceRayInlinePrologue_Xe(Value* arg_0, Value* arg_1, Value* arg_2,
   auto* V_37 = CreateInBoundsGEP(_igc_struct_RTStackFormat__RTStack(*Ctx.getModule()), arg_0, { getInt64(0), getInt32(0), getInt32(3) });
   CreateStore(getInt32(0), V_37);
   auto* V_38 = CreateInBoundsGEP(_igc_struct_RTStackFormat__RTStack(*Ctx.getModule()), arg_0, { getInt64(0), getInt32(1), getInt32(3) });
-  CreateStore(getInt32(0), V_38);
+  CreateStore(getInt32(268435456), V_38);
   CreateBr(_JoinBB);
   V_21->addIncoming(V_20, BB_9);
   V_21->addIncoming(arg_3, BB_8);
@@ -2594,7 +2594,7 @@ void _createTraceRayInlinePrologue_Xe3(Value* arg_0, Value* arg_1, Value* arg_2,
   auto* V_33 = CreateInBoundsGEP(_igc_struct_RTStackFormat__RTStack_8(*Ctx.getModule()), arg_0, { getInt64(0), getInt32(0), getInt32(3) });
   CreateStore(getInt32(0), V_33);
   auto* V_34 = CreateInBoundsGEP(_igc_struct_RTStackFormat__RTStack_8(*Ctx.getModule()), arg_0, { getInt64(0), getInt32(1), getInt32(3) });
-  CreateStore(getInt32(0), V_34);
+  CreateStore(getInt32(268435456), V_34);
   CreateBr(_JoinBB);
   V_18->addIncoming(V_17, BB_9);
   V_18->addIncoming(arg_3, BB_8);

IGC/AdaptorCommon/RayTracing/NewTraceRayInlineLoweringPass.cpp

@@ -414,11 +414,41 @@ void InlineRaytracing::LowerIntrinsics(Function &F) {
       auto *result =
           IRB.CreatePHI(IRB.getInt1Ty(), 2, VALUE_NAME("ProceedResult"));
       result->addIncoming(IRB.getFalse(), result->getParent());
-      auto [proceedBB, _] = IRB.createTriangleFlow(
+      auto [proceedBB, abortBB] = IRB.createTriangleFlow(
           doNotAbort, result, VALUE_NAME("NotAbortedProceedBB"),
           VALUE_NAME("PostProceedBB"));
 
-      IRB.SetInsertPoint(proceedBB->getTerminator());
+      auto *entryBB = proceedBB->getUniquePredecessor();
+      entryBB->getTerminator()->eraseFromParent();
+
+      // clang-format off
+      // there are 4 cases here:
+      // 1. the ray was just initialized:
+      //    enter the traversal block
+      // 2. we are mid traversal and app did not commit any hit since last
+      // proceed
+      //    set the done bit to 1
+      //    enter the traversal block
+      // 3. we are mid traversal and app has committed a hit since last proceed
+      //    set the done bit to 1
+      //    set the valid bit to 1
+      // 4. we are done with traversal
+      //    skip the traversal block
+      // clang-format on
+
+      // Create a block to handle 2 and 3
+      auto *setDoneBB = BasicBlock::Create(
+          *m_pCGCtx->getLLVMContext(), VALUE_NAME("setDoneBB"), &F, proceedBB);
+
+      IRB.SetInsertPoint(entryBB);
+      auto *switchI = IRB.CreateSwitch(traceRayCtrl, setDoneBB, 2);
+      switchI->addCase(IRB.getInt32(TRACE_RAY_DONE), abortBB);
+      switchI->addCase(IRB.getInt32(TRACE_RAY_INITIAL), proceedBB);
+
+      // add unreachable to the new block so we can split it
+      IRB.SetInsertPoint(setDoneBB);
+      auto *IP = IRB.CreateUnreachable();
+      IRB.SetInsertPoint(IP);
 
       {
         // make sure the done bit is set to 0
@@ -429,21 +459,26 @@ void InlineRaytracing::LowerIntrinsics(Function &F) {
         auto *cond =
             IRB.CreateICmpEQ(data.CommittedDataLocation,
                              IRB.getInt32(CommittedDataLocation::PotentialHit));
-        auto *IP = &*IRB.GetInsertPoint();
 
         Instruction *ifTerm, *elseTerm;
 
         SplitBlockAndInsertIfThenElse(cond, IP, &ifTerm, &elseTerm);
         IRB.SetInsertPoint(ifTerm);
         IRB.setDoneBit(getStackPtr(IRB, rqObject), false);
         IRB.setHitValid(getStackPtr(IRB, rqObject), false);
+        IRB.CreateBr(proceedBB);
+        ifTerm->eraseFromParent();
 
         IRB.SetInsertPoint(elseTerm);
         IRB.setDoneBit(getStackPtr(IRB, rqObject), false);
+        IRB.CreateBr(proceedBB);
+        elseTerm->eraseFromParent();
 
         IRB.SetInsertPoint(IP);
       }
 
+      IRB.SetInsertPoint(proceedBB->getFirstNonPHI());
+
       EmitPreTraceRayFence(IRB, rqObject);
 
 
@@ -490,7 +525,7 @@ void InlineRaytracing::LowerIntrinsics(Function &F) {
                 1);
 
         auto *notDone = IRB.CreateICmpEQ(proceedFurther, IRB.getInt32(1));
-        result->addIncoming(notDone, proceedBB);
+        result->addIncoming(notDone, IRB.GetInsertBlock());
 
         data.TraceRayCtrl =
             IRB.CreateSelect(notDone, IRB.getInt32(TRACE_RAY_CONTINUE),
@@ -501,7 +536,7 @@ void InlineRaytracing::LowerIntrinsics(Function &F) {
         data.CommittedDataLocation = IRB.getInt32(CommittedHit);
       } else {
         auto *notDone = IRB.isDoneBitNotSet(getStackPtr(IRB, rqObject), false);
-        result->addIncoming(notDone, proceedBB);
+        result->addIncoming(notDone, IRB.GetInsertBlock());
 
         data.TraceRayCtrl =
             IRB.CreateSelect(notDone, IRB.getInt32(TRACE_RAY_CONTINUE),