[llvm] [LLVM][NVPTX] Add NVPTX codegen support for clusterlaunchcontrol instruction (PR #134568)

Artem Belevich via llvm-commits llvm-commits at lists.llvm.org
Mon Apr 7 14:53:50 PDT 2025


================
@@ -112,6 +112,20 @@ class NVPTXSubtarget : public NVPTXGenSubtargetInfo {
     return HasTcgen05 && PTXVersion >= 86;
   }
 
+  bool isBlackwellArchAccelerated() const {
+    auto HasRequiredSM = [&]() {
+      if (FullSmVersion == 1001)
+        return true;
+      if (FullSmVersion == 1011)
+        return true;
+      if (FullSmVersion == 1201)
+        return true;
+      return false;
+    };
+
+    return HasRequiredSM() && PTXVersion >= 86;
----------------
Artem-B wrote:

I'm not sure we need this custom predicate at the moment.
We already use `[hasSM<100>, hasArchAccelFeatures, hasPTX<86>]` in a lot of places.
We could shorten that in tablegen, if necessary, if it becomes too much of a burden, but it does not seem to be the case for now.

If/when some instructions eventually turn out to be blackwell-only, we'll sort it out *then*, when we have specific info.


https://github.com/llvm/llvm-project/pull/134568


More information about the llvm-commits mailing list