[clang] [llvm] [AMDGPU] Add async variants of tensor load/store LDS intrinsics. (PR #200775)

[Shilei Tian via cfe-commits]

================
@@ -2182,9 +2182,10 @@ class VIMAGE_TENSOR_Pseudo<string opName, bit _UpTo2D = 0> :
   let hasSideEffects = 0;
 
   bit UpTo2D = _UpTo2D;
-  let InOperandList = !if(UpTo2D, (ins SReg_128_XNULL:$vaddr0, SReg_256_XNULL:$vaddr1, R128A16:$r128, CPol:$cpol),
+  let InOperandList = !if(UpTo2D, (ins SReg_128_XNULL:$vaddr0, SReg_256_XNULL:$vaddr1,
+                                       R128A16:$r128, CPol:$cpol, i1imm:$IsAsync),
                                       (ins SReg_128_XNULL:$vaddr0, SReg_256_XNULL:$vaddr1, SReg_128_XNULL:$vaddr2,
-                                       SReg_128_XNULL:$vaddr3, R128A16:$r128, CPol:$cpol));
+                                       SReg_128_XNULL:$vaddr3, R128A16:$r128, CPol:$cpol, i1imm:$IsAsync));
----------------
shiltian wrote:

why do we need an operand to indicate it is async instead of opcode?

https://github.com/llvm/llvm-project/pull/200775