[CSKY] Add atomic expand pass to support atomic operation with libcall

For now, just support atomic operations by libcall. Further, should investigate atomic implementation in CSKY target and codegen with atomic and fence related instructions.
llvm · Apr 6, 2022 · ec2de74 · ec2de74
1 parent 586182a
commit ec2de74
Show file tree

Hide file tree

Showing 7 changed files with 8,878 additions and 3 deletions.
diff --git a/llvm/lib/Target/CSKY/CSKYISelLowering.cpp b/llvm/lib/Target/CSKY/CSKYISelLowering.cpp
@@ -104,9 +104,7 @@ CSKYTargetLowering::CSKYTargetLowering(const TargetMachine &TM,
     setOperationAction(ISD::UDIV, MVT::i32, Expand);
   }
 
-  if (!Subtarget.has3r2E3r3()) {
-    setOperationAction(ISD::ATOMIC_FENCE, MVT::Other, Expand);
-  }
+  setOperationAction(ISD::ATOMIC_FENCE, MVT::Other, Expand);
 
   // Float
 

diff --git a/llvm/lib/Target/CSKY/CSKYTargetMachine.cpp b/llvm/lib/Target/CSKY/CSKYTargetMachine.cpp
@@ -96,6 +96,7 @@ class CSKYPassConfig : public TargetPassConfig {
     return getTM<CSKYTargetMachine>();
   }
 
+  void addIRPasses() override;
   bool addInstSelector() override;
   void addPreEmitPass() override;
 };
@@ -106,6 +107,11 @@ TargetPassConfig *CSKYTargetMachine::createPassConfig(PassManagerBase &PM) {
   return new CSKYPassConfig(*this, PM);
 }
 
+void CSKYPassConfig::addIRPasses() {
+  addPass(createAtomicExpandPass());
+  TargetPassConfig::addIRPasses();
+}
+
 bool CSKYPassConfig::addInstSelector() {
   addPass(createCSKYISelDag(getCSKYTargetMachine()));
 

diff --git a/llvm/test/CodeGen/CSKY/atomic-cmpxchg-flag.ll b/llvm/test/CodeGen/CSKY/atomic-cmpxchg-flag.ll
@@ -0,0 +1,32 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=csky -verify-machineinstrs -csky-no-aliases -mattr=+2e3 < %s \
+; RUN:   | FileCheck -check-prefix=CSKY %s
+
+define i1 @cmpxchg_i32_seq_cst_seq_cst(i32* %ptr, i32 signext %cmp,
+                                       i32 signext %val) nounwind {
+; CSKY-LABEL: cmpxchg_i32_seq_cst_seq_cst:
+; CSKY:       # %bb.0: # %entry
+; CSKY-NEXT:    subi16 sp, sp, 4
+; CSKY-NEXT:    st32.w lr, (sp, 0) # 4-byte Folded Spill
+; CSKY-NEXT:    subi16 sp, sp, 8
+; CSKY-NEXT:    st16.w a1, (sp, 4)
+; CSKY-NEXT:    movi16 a1, 5
+; CSKY-NEXT:    st16.w a1, (sp, 0)
+; CSKY-NEXT:    addi16 a1, sp, 4
+; CSKY-NEXT:    movi16 a3, 5
+; CSKY-NEXT:    jsri32 [.LCPI0_0]
+; CSKY-NEXT:    addi16 sp, sp, 8
+; CSKY-NEXT:    ld32.w lr, (sp, 0) # 4-byte Folded Reload
+; CSKY-NEXT:    addi16 sp, sp, 4
+; CSKY-NEXT:    rts16
+; CSKY-NEXT:    .p2align 1
+; CSKY-NEXT:  # %bb.1:
+; CSKY-NEXT:    .p2align 2
+; CSKY-NEXT:  .LCPI0_0:
+; CSKY-NEXT:    .long __atomic_compare_exchange_4
+;
+entry:
+  %0 = cmpxchg i32* %ptr, i32 %cmp, i32 %val seq_cst seq_cst
+  %1 = extractvalue { i32, i1 } %0, 1
+  ret i1 %1
+}