Use getIntrinsicID to test which one is actually the working one.

Author: mcourteaux

src/CodeGen_PTX_Dev.cpp

@@ -263,11 +263,11 @@ void CodeGen_PTX_Dev::visit(const Call *op) {
         internal_assert(fence_type_ptr) << "gpu_thread_barrier() parameter is not a constant integer.\n";
 
         llvm::Function *barrier;
-        if ((barrier = module->getFunction("llvm.nvvm.barrier.cta.sync.aligned.all"))) {
-            // LLVM 20 and above: https://github.com/llvm/llvm-project/pull/140615
+        if ((barrier = module->getFunction("llvm.nvvm.barrier.cta.sync.aligned.all")) && barrier->getIntrinsicID() != 0) {
+            // LLVM 20.1.6 and above: https://github.com/llvm/llvm-project/pull/140615
             builder->CreateCall(barrier, builder->getInt32(0));
-        } else if ((barrier = module->getFunction("llvm.nvvm.barrier0"))) {
-            // LLVM 19: Testing for llvm.nvvm.barrier0 can be removed once we drop support for LLVM 19
+        } else if ((barrier = module->getFunction("llvm.nvvm.barrier0")) && barrier->getIntrinsicID() != 0) {
+            // LLVM 21.1.5 and below: Testing for llvm.nvvm.barrier0 can be removed once we drop support for LLVM 20
             builder->CreateCall(barrier);
         } else {
             internal_error << "Could not find PTX barrier intrinsic llvm.nvvm.barrier0 nor llvm.nvvm.barrier.cta.sync.aligned.all\n";

src/runtime/ptx_dev.ll

@@ -1,5 +1,11 @@
-declare void @llvm.nvvm.barrier0()
-declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32)
+; The two forward declared intrinsics below refer to the same thing.
+; LLVM 20.1.6 introduced a new naming scheme for these intrinsics
+; We have to declare both, such that we can access them from the Module's
+; getFunction(), but one of those will map to an intrinsic, which we
+; will use to determine which intrinsic is supported by LLVM.
+declare void @llvm.nvvm.barrier0() ; LLVM <=20.1.5
+declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) ; LLVM >=20.1.6
+
 declare  i32 @llvm.nvvm.read.ptx.sreg.tid.x()
 declare  i32 @llvm.nvvm.read.ptx.sreg.ctaid.x()
 declare  i32 @llvm.nvvm.read.ptx.sreg.ntid.x()