Workaround the HW bug for rayquery check/release

Workaround the HW bug for rayquery check/release

Author: jaladreips

IGC/AdaptorCommon/RayTracing/DynamicRayManagementPass.cpp

@@ -358,8 +358,19 @@ bool DynamicRayManagementPass::TryProceedBasedApproach(Function& F)
             // following single entry multiple exits loop model, we insert one check and multiple releases
             checkBBs.insert(start);
 
-            for (auto* exitBB : exitBlocks)
-                releaseBBs.insert(m_PDT->findNearestCommonDominator(end, exitBB));
+            if (m_CGCtx->platform.allowDivergentControlFlowRayQueryCheckRelease())
+            {
+                for (auto* exitBB : exitBlocks)
+                    releaseBBs.insert(m_PDT->findNearestCommonDominator(end, exitBB));
+            }
+            else
+            {
+                for (auto* exitBB : exitBlocks)
+                    end = m_PDT->findNearestCommonDominator(end, exitBB);
+
+                releaseBBs.insert(end);
+            }
+
         }
 
         llvm::erase_if(
@@ -385,7 +396,7 @@ bool DynamicRayManagementPass::TryProceedBasedApproach(Function& F)
     auto* init_guard = IRB.CreateStore(IRB.getFalse(), guard);
     guardStoresAndLoads.push_back(init_guard);
 
-    SmallVector<Instruction*> CheckReleaseIntrinsics;
+    SmallVector<CallInst*> CheckReleaseIntrinsics;
 
     for (auto* checkBB : checkBBs)
     {
@@ -439,6 +450,7 @@ bool DynamicRayManagementPass::TryProceedBasedApproach(Function& F)
 
     SimplifyQuery SQ(F.getParent()->getDataLayout());
 
+    SmallVector<Instruction*> toErase;
     for (auto* I : CheckReleaseIntrinsics)
     {
         Value* flag = I->getOperand(0);
@@ -458,13 +470,42 @@ bool DynamicRayManagementPass::TryProceedBasedApproach(Function& F)
         if (auto* CI = dyn_cast_or_null<ConstantInt>(flag))
         {
             if (CI->isZero())
-                I->eraseFromParent();
+                toErase.push_back(I);
 
             if (CI->isOne())
                 I->setOperand(0, IRB.getTrue());
         }
+        else
+        {
+            // if we encounter a nonconstant predicate and the platform can't handle divergent rayquery check/release
+            // we undo as much as we can and bail from approach entirely
+            if (!m_CGCtx->platform.allowDivergentControlFlowRayQueryCheckRelease())
+            {
+                llvm::for_each(
+                    CheckReleaseIntrinsics,
+                    [&](auto* I) {
+                        I->eraseFromParent();
+                    }
+                );
+
+                return false;
+            }
+        }
+
+        // prevent LLVM from merging the calls
+        if (!m_CGCtx->platform.allowDivergentControlFlowRayQueryCheckRelease())
+        {
+            I->addFnAttr(llvm::Attribute::NoMerge);
+        }
     }
 
+    llvm::for_each(
+        toErase,
+        [&](auto* I) {
+            I->eraseFromParent();
+        }
+    );
+
     return true;
 
 #endif // LLVM_VERSION_MAJOR >= 10

IGC/Compiler/CISACodeGen/Platform.hpp

@@ -1743,5 +1743,10 @@ bool preferLSCCache() const
 }
 
 
+bool allowDivergentControlFlowRayQueryCheckRelease() const
+{
+    return false;
+}
+
 };
 }//namespace IGC

IGC/common/igc_flags.h

@@ -914,7 +914,7 @@ DECLARE_IGC_GROUP("Raytracing Options")
     DECLARE_IGC_REGKEY(bool, DisableRayQueryDynamicRayManagementMechanismForExternalFunctionsCalls, false, "Disable dynamic ray management mechanism for shaders with external functions calls", true)
     DECLARE_IGC_REGKEY(bool, DisableRayQueryDynamicRayManagementMechanismForBarriers, false, "Disable dynamic ray management mechanism for shaders with barriers", true)
     DECLARE_IGC_REGKEY(bool, EnableOuterLoopHoistingForRayQueryDynamicRayManagementMechanism, false, "Disable dynamic ray management mechanism for shaders with barriers", true)
-    DECLARE_IGC_REGKEY(bool, DisableProceedBasedApproachForRayQueryDynamicRayManagementMechanism, true, "Disables proceed based approach for dynamic ray management mechanism", true)
+    DECLARE_IGC_REGKEY(bool, DisableProceedBasedApproachForRayQueryDynamicRayManagementMechanism, false, "Disables proceed based approach for dynamic ray management mechanism", true)
     DECLARE_IGC_REGKEY(bool, DisableWideTraceRay, false, "Disable SIMD16 style message payloads for send.rta", true)
     DECLARE_IGC_REGKEY(bool, ForceRTCheckInstanceLeafPtr, true, "Check MemHit::valid before loading GeometryIndex, PrimitiveIndex, etc.", true)
     DECLARE_IGC_REGKEY(DWORD, RTInValidDefaultIndex, 0xFFFFFFFF, "If MemHit::valid is false, the default value to return for some intrinsics like GeometryIndex or PrimitiveIndex etc.", true)