[MC] Introduce ability to add target specific AsmStreamer

[tltao]

Currently the AsmStreamer class MCAsmStreamer.cpp only outputs GNU assembly syntax. However, we wish to add additional AsmStreamer classes for specific targets that can output alternative assembly syntaxes (the immediate use case is HLASM syntax for z/OS).

This is the first patch to add this ability by utilizing the TargetRegistry functions.

[llvmbot]

@llvm/pr-subscribers-mc

Author: None (tltao)

Changes

Currently the AsmStreamer class MCAsmStreamer.cpp only outputs GNU assembly syntax. However, we wish to add additional AsmStreamer classes for specific targets that can output alternative assembly syntaxes (the immediate use case is HLASM syntax for z/OS).

This is the first patch to add this ability by utilizing the TargetRegistry functions.

---

Full diff: https://github.com/llvm/llvm-project/pull/107415.diff

2 Files Affected:

• (modified) llvm/include/llvm/MC/TargetRegistry.h (+12)
• (modified) llvm/lib/MC/TargetRegistry.cpp (+8-2)

diff --git a/llvm/include/llvm/MC/TargetRegistry.h b/llvm/include/llvm/MC/TargetRegistry.h
index 42d510c17bce39..1850139ba81491 100644
--- a/llvm/include/llvm/MC/TargetRegistry.h
+++ b/llvm/include/llvm/MC/TargetRegistry.h
@@ -208,6 +208,10 @@ class Target {
   using AsmTargetStreamerCtorTy =
       MCTargetStreamer *(*)(MCStreamer &S, formatted_raw_ostream &OS,
                             MCInstPrinter *InstPrint);
+  using AsmStreamerCtorTy =
+      MCStreamer *(*)(MCContext &Ctx, std::unique_ptr<formatted_raw_ostream> OS,
+                      MCInstPrinter *IP, std::unique_ptr<MCCodeEmitter> CE,
+                      std::unique_ptr<MCAsmBackend> TAB);
   using ObjectTargetStreamerCtorTy = MCTargetStreamer *(*)(
       MCStreamer &S, const MCSubtargetInfo &STI);
   using MCRelocationInfoCtorTy = MCRelocationInfo *(*)(const Triple &TT,
@@ -316,6 +320,10 @@ class Target {
   /// registered (default = nullptr).
   AsmTargetStreamerCtorTy AsmTargetStreamerCtorFn = nullptr;
 
+  /// Construction function for this target's AsmStreamer, if
+  /// registered (default = nullptr).
+  AsmStreamerCtorTy AsmStreamerCtorFn = nullptr;
+
   /// Construction function for this target's obj TargetStreamer, if
   /// registered (default = nullptr).
   ObjectTargetStreamerCtorTy ObjectTargetStreamerCtorFn = nullptr;
@@ -927,6 +935,10 @@ struct TargetRegistry {
     T.NullTargetStreamerCtorFn = Fn;
   }
 
+  static void RegisterAsmStreamer(Target &T, Target::AsmStreamerCtorTy Fn) {
+    T.AsmStreamerCtorFn = Fn;
+  }
+
   static void RegisterAsmTargetStreamer(Target &T,
                                         Target::AsmTargetStreamerCtorTy Fn) {
     T.AsmTargetStreamerCtorFn = Fn;
diff --git a/llvm/lib/MC/TargetRegistry.cpp b/llvm/lib/MC/TargetRegistry.cpp
index 3be6f1d4634990..0d343a2c4e4d5f 100644
--- a/llvm/lib/MC/TargetRegistry.cpp
+++ b/llvm/lib/MC/TargetRegistry.cpp
@@ -93,8 +93,14 @@ MCStreamer *Target::createAsmStreamer(MCContext &Ctx,
                                       std::unique_ptr<MCCodeEmitter> CE,
                                       std::unique_ptr<MCAsmBackend> TAB) const {
   formatted_raw_ostream &OSRef = *OS;
-  MCStreamer *S = llvm::createAsmStreamer(Ctx, std::move(OS), IP,
-                                          std::move(CE), std::move(TAB));
+  MCStreamer *S;
+  if (AsmStreamerCtorFn)
+    S = AsmStreamerCtorFn(Ctx, std::move(OS), IP, std::move(CE),
+                          std::move(TAB));
+  else
+    S = llvm::createAsmStreamer(Ctx, std::move(OS), IP, std::move(CE),
+                                std::move(TAB));
+
   createAsmTargetStreamer(*S, OSRef, IP);
   return S;
 }

[tltao]

@redstar @MaskRay @abhina-sree @zibi2 @Ariel-Burton @perry-ca

[abhina-sree]

LGTM, assuming testcases that traverse this code path will be added in the future PRs

[tltao]

LGTM, assuming testcases that traverse this code path will be added in the future PRs

Yeah I didn't include any test case because the behaviour in this PR is identical to the current behaviour, and I don't see a way to test anything else with the existing changes.

[tltao]

@iliya-diyachkov @lhames FYI

Also added use case in #108433

[tltao]

Closing this in favour of #130535.