Use Ray Query Return value in Compute Ray Tracing Extension

 Modified intel_get_hit_candidate and intel_is_traversal_done functions.

Author: ustachow

IGC/BiFModule/Headers/bif_flag_controls.h

@@ -38,5 +38,6 @@ BIF_FLAG_CONTROL(int, MaxHWThreadIDPerSubDevice)
 BIF_FLAG_CONTROL(bool, UseAssumeInGetGlobalId)
 BIF_FLAG_CONTROL(int, JointMatrixLoadStoreOpt)
 BIF_FLAG_CONTROL(bool, UseOOBChecks)
+BIF_FLAG_CONTROL(bool, IsRayQueryReturnOptimizationEnabled)
 BIF_FLAG_CONTROL(bool, UseBindlessImage)
 #endif // __BIF_FLAG_CONTROL_H__

IGC/BiFModule/Languages/OpenCL/Raytracing/IBiF_intel_rt_production.cl

@@ -375,6 +375,23 @@ intel_float4x3 intel_get_hit_object_to_world(
 intel_candidate_type_t
 intel_get_hit_candidate(intel_ray_query_t rayquery, intel_hit_type_t hit_type)
 {
+    if (BIF_FLAG_CTRL_GET(IsRayQueryReturnOptimizationEnabled))
+    {
+        rtfence_t fence = __builtin_IB_intel_query_rt_fence(rayquery);
+        uintptr_t fenceInt = (uintptr_t)fence;
+
+        intel_candidate_type_t committedStatus = (fenceInt >> 1) & 0x3;
+        intel_candidate_type_t candidateType = (fenceInt>> 3) & 0x1;
+
+        if (hit_type == intel_hit_type_committed_hit)
+        {
+            return committedStatus - 1;
+        }
+        else if (hit_type == intel_hit_type_potential_hit)
+        {
+            return candidateType;
+        }
+    }
     return MemHit_getLeafType(get_query_hit(rayquery, hit_type)) == NODE_TYPE_QUAD
                ? intel_candidate_type_triangle
                : intel_candidate_type_procedural;
@@ -446,13 +463,33 @@ int intel_get_ray_mask(intel_ray_query_t rayquery, uint bvh_level)
 // a procedural leaf or a non-opaque triangle leaf, and requires shader processing.
 bool intel_is_traversal_done(intel_ray_query_t rayquery)
 {
+    if (BIF_FLAG_CTRL_GET(IsRayQueryReturnOptimizationEnabled))
+    {
+        rtfence_t fence = __builtin_IB_intel_query_rt_fence(rayquery);
+        uintptr_t fenceInt = (uintptr_t)fence;
+
+        bool proceedFurther = (fenceInt & 0x1) != 0;
+
+
+        return !proceedFurther;
+    }
     bool isTraversalDone = MemHit_getDone(get_query_hit(rayquery, intel_hit_type_potential_hit));
     return isTraversalDone;
 }
 
 // if traversal is done one can test for the presence of a committed hit to either invoke miss or closest hit shader
 bool intel_has_committed_hit(intel_ray_query_t rayquery)
 {
+    if (BIF_FLAG_CTRL_GET(IsRayQueryReturnOptimizationEnabled))
+    {
+        rtfence_t fence = __builtin_IB_intel_query_rt_fence(rayquery);
+        uintptr_t fenceInt = (uintptr_t)fence;
+
+        intel_candidate_type_t committedStatus = (fenceInt >> 1) & 0x3;
+
+        return committedStatus != 0;
+
+    }
     return MemHit_getValid(get_query_hit(rayquery, intel_hit_type_committed_hit));
 }
 

IGC/Compiler/Builtins/BIFFlagCtrl/BIFFlagCtrlResolution.cpp

@@ -97,6 +97,7 @@ void BIFFlagCtrlResolution::FillFlagCtrl() {
   }
 
   BIF_FLAG_CTRL_SET(UseOOBChecks, PtrCGC->platform.needsOutOfBoundsBuiltinChecks());
+  BIF_FLAG_CTRL_SET(IsRayQueryReturnOptimizationEnabled, PtrCGC->platform.isRayQueryReturnOptimizationEnabled());
 
   // NOTE: No need to check for UseLegacyBindlessMode,
   //       as it's unrelated to images.

IGC/VectorCompiler/lib/GenXOpts/CMTrans/GenXBIFFlagCtrlResolution.cpp

@@ -94,6 +94,7 @@ void GenXBIFFlagCtrlResolution::FillFlagCtrl() {
   BIF_FLAG_CTRL_SET(UseNativeFP64GlobalAtomicAdd, false);
   BIF_FLAG_CTRL_SET(MaxHWThreadIDPerSubDevice, 1);
   BIF_FLAG_CTRL_SET(UseOOBChecks, false);
+  BIF_FLAG_CTRL_SET(IsRayQueryReturnOptimizationEnabled, false);
   BIF_FLAG_CTRL_SET(UseBindlessImage, false);
 }