update llir/llvm to support LLVM 10.0

[mewmew]

This is a meta issue tracking the progress to support LLVM 10.0 in llir/llvm.

For reference, it will be very similar to the work done to support LLVM 9.0.

In particular, we need to:

• update the grammar to LLVM 10.0 (see update grammar to LLVM 9.0 #101 as reference)
  • llir/grammar
• update the test cases to LLVM 10.0 (see update test cases to LLVM 9.0 #105 as reference)
  • llir/testdata
• add new language constructs of LLVM 10.0 (see add new language constructs to llir/llvm/ir package for LLVM 9.0 #106 as reference)
  • llir/llvm/ir

Release candidate 2 of LLVM 10.0 was tagged on the 11th of February, and the final release is scheduled for the 26th of February.

cc: @dannypsnl would you care to take a look at this? I can help out if you get stuck anywhere. I tried to write in the issues (#101, #105 and #106 and associated PRs) how we did this last time :) Just let me know if anything feels unclear.

Cheers,
Robin

[mewmew]

Just take a look, seems like LLVM 10 haven't released yet.

Yeah, lets wait until 10.0 is released. I was just excited :)

[mewmew]

LLVM release candidate 5 is out. We are getting closer :)

[mewmew]

LLVM 10.0 has finally been released!

@dannypsnl, do you wanna take a look at updating llir/llvm? I'd be happy to support you and help if you get stuck :)

[dannypsnl]

LLVM 10.0 Changes Report

reference document

Changes to the LLVM IR

• Unnamed function arguments now get printed with their automatically generated name (e.g. “i32 %0”) in definitions.
  This may require front-ends to update their tests;
  if so there is a script utils/add_argument_names.py that correctly converted 80-90% of Clang tests.
  Some manual work will almost certainly still be needed.
• A new freeze instruction is added.
  The freeze instruction is used to stop IR-level propagation of undef and poison values.
  Currently its support is preliminary; a freeze-equivalent operation for SelDag/MIR needs to be added.

TODO:

• give parameter default name
• provide freeze instruction

freeze instruction

@@ -5788,8 +5815,21 @@ int LLParser::ParseInstruction(Instruction *&Inst, BasicBlock *BB,
   case lltok::kw_extractelement: return ParseExtractElement(Inst, PFS);
   case lltok::kw_insertelement:  return ParseInsertElement(Inst, PFS);
   case lltok::kw_shufflevector:  return ParseShuffleVector(Inst, PFS);
-  case lltok::kw_phi:            return ParsePHI(Inst, PFS);
+  case lltok::kw_phi: {
+    FastMathFlags FMF = EatFastMathFlagsIfPresent();
+    int Res = ParsePHI(Inst, PFS);
+    if (Res != 0)
+      return Res;
+    if (FMF.any()) {
+      if (!isa<FPMathOperator>(Inst))
+        return Error(Loc, "fast-math-flags specified for phi without "
+                          "floating-point scalar or vector return type");
+      Inst->setFastMathFlags(FMF);
+    }
+    return 0;
+  }
   case lltok::kw_landingpad:     return ParseLandingPad(Inst, PFS);
+  case lltok::kw_freeze:         return ParseFreeze(Inst, PFS);
   // Call.
   case lltok::kw_call:     return ParseCall(Inst, PFS, CallInst::TCK_None);
   case lltok::kw_tail:     return ParseCall(Inst, PFS, CallInst::TCK_Tail);

+/// ParseFreeze
+///   ::= 'freeze' Type Value
+bool LLParser::ParseFreeze(Instruction *&Inst, PerFunctionState &PFS) {
+  LocTy Loc;
+  Value *Op;
+  if (ParseTypeAndValue(Op, Loc, PFS))
+    return true;
+
+  Inst = new FreezeInst(Op);
+  return false;
+}

[mewmew]

From the "Parse errors" section of #132 (comment) a couple of test cases are listed for which we need to update the grammar to LLVM 10.0.

They include:

• FastMathFlag used in phi instructions
• !DIModule metadata

[mewmew]

@dannypsnl, is it only the freeze instruction left for LLVM 10.0? Let me know if you get stuck or want to bounce a few ideas on how to implement the last few things for 10.0.

[mewmew]

With #137 merged, all new language concepts of LLVM 10.0 have been added to llir/llvm. Thanks @dannypsnl! :)

What remains to add for LLVM 10.0 is only enum values.

From #132 (comment)

• New calling conventions:

@@ -585,6 +585,7 @@ lltok::Kind LLLexer::LexIdentifier() {
   KEYWORD(ccc);
   KEYWORD(fastcc);
   KEYWORD(coldcc);
+  KEYWORD(cfguard_checkcc);
   KEYWORD(x86_stdcallcc);
   KEYWORD(x86_fastcallcc);
   KEYWORD(x86_thiscallcc);
@@ -593,6 +594,7 @@ lltok::Kind LLLexer::LexIdentifier() {
   KEYWORD(arm_aapcscc);
   KEYWORD(arm_aapcs_vfpcc);
   KEYWORD(aarch64_vector_pcs);
+  KEYWORD(aarch64_sve_vector_pcs);
   KEYWORD(msp430_intrcc);
   KEYWORD(avr_intrcc);
   KEYWORD(avr_signalcc);
@@ -622,6 +624,7 @@ lltok::Kind LLLexer::LexIdentifier() {
   KEYWORD(amdgpu_ps);
   KEYWORD(amdgpu_cs);
   KEYWORD(amdgpu_kernel);
+  KEYWORD(tailcc);

• alwaysInline keyword. It is only used in module summaries (grammar: add module summary (introduced in LLVM 7.0) #43). We can skip this for now. Adding modules summaries is an old issue, and so far noone has needed this?

• nofree as parameter attribute (currently only used for FuncAttr in grammar).

[mewmew]

After merging #128 and llir/testdata#6, the only test case remaining for LLVM 10.0 integration is testdata/llvm/test/Bitcode/calling-conventions.3.2.ll.

--- FAIL: TestParseFile (40.55s)
    asm_test.go:516: unable to parse "../testdata/llvm/test/Bitcode/calling-conventions.3.2.ll" into AST; syntax error at line 29
        unable to parse "../testdata/llvm/test/Bitcode/calling-conventions.3.2.ll" into an AST

[mewmew]

@dannypsnl I think with #140 merged, we are now ready to close this issue as llir/llvm is now udpated to LLVM 10.0. Is there anything else you can think of that should be worked on? Otherwise, feel free to close this issue :)