[llvm] [NVPTX] restrict `cvta.param` use to kernels only. (PR #112278)

Akshay Deodhar via llvm-commits llvm-commits at lists.llvm.org
Fri Oct 18 10:21:46 PDT 2024


================
@@ -2,6 +2,72 @@
 ; RUN: opt < %s -S -nvptx-lower-args --mtriple nvptx64-nvidia-cuda -mcpu=sm_70 -mattr=+ptx77 | FileCheck %s --check-prefixes OPT
 ; RUN: llc < %s --mtriple nvptx64-nvidia-cuda -mcpu=sm_70 -mattr=+ptx77 | FileCheck %s --check-prefixes PTX
 
+%struct.uint4 = type { i32, i32, i32, i32 }
+
+ at gi = dso_local addrspace(1) externally_initialized global %struct.uint4 { i32 50462976, i32 117835012, i32 185207048, i32 252579084 }, align 16
+
+; Function Attrs: mustprogress nofree noinline norecurse nosync nounwind willreturn memory(read, inaccessiblemem: none)
+; Regular functions mus still make a copy. `cvta.param` does not always work there.
----------------
akshayrdeodhar wrote:

nit: s/mus/must, but shouldn't matter.

https://github.com/llvm/llvm-project/pull/112278


More information about the llvm-commits mailing list