[llvm] [NVPTX] Make nvptx mma instructions convergent. (PR #96521)

[Quentin Colombet via llvm-commits]

================
@@ -0,0 +1,32 @@
+; RUN: llc < %s -march=nvptx64 -mcpu=sm_80 -mattr=+ptx81 | FileCheck %s
+
+declare noundef i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1
+
+declare noundef i32 @llvm.nvvm.read.ptx.sreg.ntid.x() #1
+
+declare noundef i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1
+
+declare { float, float, float, float } @llvm.nvvm.mma.m16n8k4.row.col.tf32(i32, i32, i32, float, float, float, float) #2
+
+declare noundef i32 @llvm.nvvm.read.ptx.sreg.laneid() #1
+
+; COM: llvm.nvvm.mma should not sink to the next block and gets reordered to be after laneid check.
+; CHECK-LABEL: no_reorder_mma_and_laneid_check
+define dso_local void @no_reorder_mma_and_laneid_check(ptr %0, ptr %1, i64 %2) #0 {
----------------
qcolombet wrote:

Please get rid of the implicit variables (run `opt -passes=instnamer` on your input IR and update the file)

https://github.com/llvm/llvm-project/pull/96521