[llvm-branch-commits] [llvm] [AMDGPU] Add `.amdgpu.info` section for per-function metadata (PR #192384)

[Shilei Tian via llvm-branch-commits]

================
@@ -537,6 +538,149 @@ void AMDGPUAsmPrinter::validateMCResourceInfo(Function &F) {
   }
 }
 
+static void appendTypeEncoding(std::string &Enc, Type *Ty,
+                               const DataLayout &DL) {
+  if (Ty->isVoidTy()) {
+    Enc += 'v';
+    return;
+  }
+  unsigned Bits = DL.getTypeSizeInBits(Ty);
+  if (Bits <= 32)
+    Enc += 'i';
+  else if (Bits <= 64)
+    Enc += 'l';
+  else
+    Enc.append(divideCeil(Bits, 32), 'i');
+}
+
+static std::string computeTypeId(const FunctionType *FTy,
----------------
shiltian wrote:

Yes. This encoding scheme is sufficient for the purpose of checking potential callees. We had some internal discussion, and then decided to use ABI type for the encoding.

The linker simply compares the indirect call type ID with function's type ID, and if they match, the function can be a potential callee of an indirect call.

https://github.com/llvm/llvm-project/pull/192384