[llvm] [WIP][AMDGPU] Split `isInlinableLiteral16` into three and call the specific version if possible (PR #81345)

[Shilei Tian via llvm-commits]

================
@@ -105,18 +105,11 @@ class AMDGPUDAGToDAGISel : public SelectionDAGISel {
 
 private:
   std::pair<SDValue, SDValue> foldFrameIndex(SDValue N) const;
-  bool isInlineImmediate(const SDNode *N) const;
-
-  bool isInlineImmediate16(int64_t Imm) const {
-    return AMDGPU::isInlinableLiteral16(Imm, Subtarget->hasInv2PiInlineImm());
-  }
 
-  bool isInlineImmediate32(int64_t Imm) const {
-    return AMDGPU::isInlinableLiteral32(Imm, Subtarget->hasInv2PiInlineImm());
-  }
+  bool isInlineImmediate(const SDNode *N) const;
 
-  bool isInlineImmediate64(int64_t Imm) const {
-    return AMDGPU::isInlinableLiteral64(Imm, Subtarget->hasInv2PiInlineImm());
+  bool isInlineImmediate(const APInt &Imm) const {
----------------
shiltian wrote:

For integer value, `APInt` has its size so we can call the corresponding version.

https://github.com/llvm/llvm-project/pull/81345