[LTO][Legacy] Add new C APIs to query undefined symbols in assembly (#145413)

Add new APIs to legacy LTO C API to surface undefined symbols that
parsed from assembly. This information is needed by thin LTO to figure
out the symbols not to dead strip, while such information is
automatically forwarded in full LTO already. Linker needs to fetch the
information and adds the undefs from assembly to MustPreserveSymbols
list just like treating undefs from a non-LTO object file.

Resolves: #29340

Author: cachemeifyoucan

llvm/include/llvm-c/lto.h

@@ -46,7 +46,7 @@ typedef bool lto_bool_t;
  * @{
  */
 
-#define LTO_API_VERSION 29
+#define LTO_API_VERSION 30
 
 /**
  * \since prior to LTO_API_VERSION=3
@@ -286,6 +286,21 @@ lto_module_get_symbol_name(lto_module_t mod, unsigned int index);
 extern lto_symbol_attributes
 lto_module_get_symbol_attribute(lto_module_t mod, unsigned int index);
 
+/**
+ * Returns the number of asm undefined symbols in the object module.
+ *
+ * \since prior to LTO_API_VERSION=30
+ */
+extern unsigned int lto_module_get_num_asm_undef_symbols(lto_module_t mod);
+
+/**
+ * Returns the name of the ith asm undefined symbol in the object module.
+ *
+ * \since prior to LTO_API_VERSION=30
+ */
+extern const char *lto_module_get_asm_undef_symbol_name(lto_module_t mod,
+                                                        unsigned int index);
+
 /**
  * Returns the module's linker options.
  *

llvm/include/llvm/LTO/legacy/LTOModule.h

@@ -143,6 +143,14 @@ struct LTOModule {
     return StringRef();
   }
 
+  uint32_t getAsmUndefSymbolCount() { return _asm_undefines.size(); }
+
+  StringRef getAsmUndefSymbolName(uint32_t index) {
+    if (index < _asm_undefines.size())
+      return _asm_undefines[index];
+    return StringRef();
+  }
+
   const GlobalValue *getSymbolGV(uint32_t index) {
     if (index < _symbols.size())
       return _symbols[index].symbol;

llvm/test/LTO/AArch64/module-asm.ll

@@ -0,0 +1,21 @@
+; RUN: llvm-as %s -o %t.o
+; RUN: llvm-lto %t.o --list-symbols-only | FileCheck %s
+
+; CHECK: ___foo    { function defined hidden }
+; CHECK: ___bar    { function defined default }
+; CHECK: _foo    { data defined default }
+; CHECK: ___foo    { asm extern }
+
+target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-n32:64-S128-Fn32"
+target triple = "arm64-apple-macosx12.0.0"
+
+module asm ".globl _foo"
+module asm "_foo = ___foo"
+
+define hidden i32 @__foo() {
+  ret i32 0
+}
+
+define i32 @__bar() {
+  ret i32 0
+}

llvm/tools/llvm-lto/llvm-lto.cpp

@@ -475,6 +475,8 @@ static void testLTOModule(const TargetOptions &Options) {
         printLTOSymbolAttributes(Module->getSymbolAttributes(I));
         outs() << "\n";
       }
+      for (int I = 0, E = Module->getAsmUndefSymbolCount(); I != E; ++I)
+        outs() << Module->getAsmUndefSymbolName(I) << "    { asm extern }\n";
     }
     if (QueryHasCtorDtor)
       outs() << Filename

llvm/tools/lto/lto.cpp

@@ -322,6 +322,15 @@ lto_symbol_attributes lto_module_get_symbol_attribute(lto_module_t mod,
   return unwrap(mod)->getSymbolAttributes(index);
 }
 
+unsigned int lto_module_get_num_asm_undef_symbols(lto_module_t mod) {
+  return unwrap(mod)->getAsmUndefSymbolCount();
+}
+
+const char *lto_module_get_asm_undef_symbol_name(lto_module_t mod,
+                                                 unsigned int index) {
+  return unwrap(mod)->getAsmUndefSymbolName(index).data();
+}
+
 const char* lto_module_get_linkeropts(lto_module_t mod) {
   return unwrap(mod)->getLinkerOpts().data();
 }

llvm/tools/lto/lto.exports

@@ -14,6 +14,8 @@ lto_module_get_macho_cputype
 lto_module_get_num_symbols
 lto_module_get_symbol_attribute
 lto_module_get_symbol_name
+lto_module_get_num_asm_undef_symbols
+lto_module_get_asm_undef_symbol_name
 lto_module_get_target_triple
 lto_module_set_target_triple
 lto_module_is_object_file