[clang] noconvergent does not affect calls to convergent functions

[ssahasra]

When placed on a function, the clang::noconvergent attribute ensures that the function is not assumed to be convergent. But the same attribute has no effect on function calls. A call is convergent if the callee is convergent. This is based on the fact that in LLVM, a call always inherits all the attributes of the callee. Only convergent is an attribute in LLVM IR, and there is no equivalent of clang::noconvergent.

[llvmbot]

@llvm/pr-subscribers-clang

Author: Sameer Sahasrabuddhe (ssahasra)

Changes

When placed on a function, the clang::noconvergent attribute ensures that the function is not assumed to be convergent. But the same attribute has no effect on function calls. A call is convergent if the callee is convergent. This is based on the fact that in LLVM, a call always inherits all the attributes of the callee. Only convergent is an attribute in LLVM IR, and there is no equivalent of clang::noconvergent.

---

Full diff: https://github.com/llvm/llvm-project/pull/132701.diff

1 Files Affected:

• (modified) clang/include/clang/Basic/AttrDocs.td (+10-10)

diff --git a/clang/include/clang/Basic/AttrDocs.td b/clang/include/clang/Basic/AttrDocs.td
index a8b588169725a..98d29838dce42 100644
--- a/clang/include/clang/Basic/AttrDocs.td
+++ b/clang/include/clang/Basic/AttrDocs.td
@@ -1658,17 +1658,16 @@ Sample usage:
 def NoConvergentDocs : Documentation {
   let Category = DocCatFunction;
   let Content = [{
-This attribute prevents a function from being treated as convergent, which
-means that optimizations can only move calls to that function to
-control-equivalent blocks. If a statement is marked as ``noconvergent`` and
-contains calls, it also prevents those calls from being treated as convergent.
-In other words, those calls are not restricted to only being moved to
-control-equivalent blocks.
+This attribute prevents a function from being treated as convergent; when a
+function is marked ``noconvergent``, calls to that function are not
+automatically assumed to be convergent, unless such calls are explicitly marked
+as ``convergent``. If a statement is marked as ``noconvergent``, any calls to
+inline ``asm`` in that statement are no longer treated as convergent.
 
 In languages following SPMD/SIMT programming model, e.g., CUDA/HIP, function
-declarations and calls are treated as convergent by default for correctness.
-This ``noconvergent`` attribute is helpful for developers to prevent them from
-being treated as convergent when it's safe.
+declarations and inline asm calls are treated as convergent by default for
+correctness. This ``noconvergent`` attribute is helpful for developers to
+prevent them from being treated as convergent when it's safe.
 
 .. code-block:: c
 
@@ -1677,7 +1676,8 @@ being treated as convergent when it's safe.
 
   __device__ int example(void) {
     float x;
-    [[clang::noconvergent]] x = bar(x);
+    [[clang::noconvergent]] x = bar(x); // no effect on convergence
+    [[clang::noconvergent]] { asm volatile ("nop"); } // the asm call is non-convergent
   }
 
   }];

[arsenm]

Only convergent is an attribute in LLVM IR, and there is no equivalent of clang::noconvergent.

There should be (and I still have a patch somewhere to add it)

[arsenm]

Maybe should have an example of an undefined usage