[llvm] [AMDGPU] Use vector memory types for 64/128-bit cooperative atomics (PR #229501)

Matt Arsenault via llvm-commits llvm-commits at lists.llvm.org
Tue Oct 6 10:53:05 PDT 2026


================
@@ -0,0 +1,286 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -global-isel=0 -mtriple=amdgpu12.50-amd-amdhsa < %s | FileCheck -check-prefixes=GFX1250,GFX1250-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgpu12.50-amd-amdhsa < %s | FileCheck -check-prefixes=GFX1250,GFX1250-GISEL %s
+
+; 64-bit and 128-bit cooperative atomics on global and flat pointers, including
+; uses of the loaded value that query known bits and sign bits, in the same
+; block and across blocks.
+
----------------
arsenm wrote:

Should these cases go in the existing cooperative atomics tests? I suspect there's some overlap already? Are these expected to work with 16-bit element vectors?

https://github.com/llvm/llvm-project/pull/229501


More information about the llvm-commits mailing list