[llvm] [AMDGPU] Add intrinsic and MI for image_bvh_dual_intersect_ray (PR #130038)

[Jay Foad via llvm-commits]

================
@@ -3217,10 +3217,13 @@ void AMDGPURegisterBankInfo::applyMappingImpl(
     applyMappingImage(B, MI, OpdMapper, RSrcIntrin->RsrcArg);
     return;
   }
-  case AMDGPU::G_AMDGPU_BVH_INTERSECT_RAY: {
-    unsigned N = MI.getNumExplicitOperands() - 2;
+  case AMDGPU::G_AMDGPU_BVH_INTERSECT_RAY:
+  case AMDGPU::G_AMDGPU_BVH_DUAL_INTERSECT_RAY: {
+    bool IsDual = MI.getOpcode() == AMDGPU::G_AMDGPU_BVH_DUAL_INTERSECT_RAY;
+    unsigned NumMods = !IsDual ? 1 : 0; // Has A16 modifier
+    unsigned LastRegOpIdx = (MI.getNumExplicitOperands() - 1) - NumMods;
----------------
jayfoad wrote:

```suggestion
    unsigned NumMods = IsDual ? 0 : 1; // Has A16 modifier
    unsigned LastRegOpIdx = MI.getNumExplicitOperands() - 1 - NumMods;
```

https://github.com/llvm/llvm-project/pull/130038