[clang] [llvm] AMDGPU: Move enqueued block handling into clang (PR #128519)

[Matt Arsenault via cfe-commits]

================
@@ -614,6 +614,20 @@ void AMDGPUTargetCodeGenInfo::setCUDAKernelCallingConvention(
       FT, FT->getExtInfo().withCallingConv(CC_OpenCLKernel));
 }
 
+/// Return IR struct type for rtinfo struct in rocm-device-libs used for device
+/// enqueue.
+///
+/// ptr addrspace(1) kernel_object, i32 private_segment_size,
----------------
arsenm wrote:

No, this isn't in the implicit arguments. It could be constant, but I think I had difficulty getting externally_initialized + constant to work as expected.

I also think in principle the special pass isn't necessary anymore. I think something about visibility was how it ended up this way, and might require a runtime change to fully delete it 

https://github.com/llvm/llvm-project/pull/128519