[clang] [llvm] [BPF] Add load-acquire and store-release instructions under -mcpu=v4 (PR #108636)

[via cfe-commits]

================
@@ -622,6 +665,47 @@ let Predicates = [BPFHasLdsx] in {
 
 def LDD : LOADi64<BPF_DW, BPF_MEM, "u64", load>;
 
+class LOAD_ACQUIRE<BPFWidthModifer SizeOp, string OpcodeStr, RegisterClass RegTp>
+    : TYPE_LD_ST<BPF_ATOMIC.Value, SizeOp.Value,
+                 (outs RegTp:$dst),
+                 (ins MEMri:$addr),
+                 "$dst = load_acquire(("#OpcodeStr#" *)($addr))",
+                 []> {
+  bits<4> dst;
+  bits<20> addr;
+
+  let Inst{51-48} = dst;
+  let Inst{55-52} = addr{19-16}; // base reg
+  let Inst{47-32} = addr{15-0}; // offset
+  let Inst{7-4} = BPF_LOAD_ACQ.Value;
----------------
eddyz87 wrote:

I'm curious what would happen with BPF compiled as big endian? Instruction format reference says that `imm` field uses host byte order. It looks like we have a pass that fixes endianness in a generic way, but could you please extend `acquire-release.ll` to check what happens when `-march=bpfeb`?

https://github.com/llvm/llvm-project/pull/108636