[RISCV] Lower llvm.clear_cache to __riscv_flush_icache for glibc targets (#93481)

This change is a preliminary step to support trampolines on RISC-V. Trampolines are used by flang to implement obtaining the address of an internal program (i.e., a nested function in Fortran parlance).

In this change we lower `llvm.clear_cache` intrinsic on glibc targets to
`__riscv_flush_icache` which is what GCC is currently doing for Linux targets.

Author: rofirrim

llvm/include/llvm/IR/RuntimeLibcalls.def

@@ -623,6 +623,7 @@ HANDLE_LIBCALL(RETURN_ADDRESS, nullptr)
 
 // Clear cache
 HANDLE_LIBCALL(CLEAR_CACHE, "__clear_cache")
+HANDLE_LIBCALL(RISCV_FLUSH_ICACHE, "__riscv_flush_icache")
 
 HANDLE_LIBCALL(UNKNOWN_LIBCALL, nullptr)
 

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

@@ -662,6 +662,11 @@ RISCVTargetLowering::RISCVTargetLowering(const TargetMachine &TM,
 
   setBooleanContents(ZeroOrOneBooleanContent);
 
+  if (getTargetMachine().getTargetTriple().isOSLinux()) {
+    // Custom lowering of llvm.clear_cache.
+    setOperationAction(ISD::CLEAR_CACHE, MVT::Other, Custom);
+  }
+
   if (Subtarget.hasVInstructions()) {
     setBooleanVectorContents(ZeroOrOneBooleanContent);
 
@@ -7142,9 +7147,29 @@ SDValue RISCVTargetLowering::LowerOperation(SDValue Op,
     return lowerVPSpliceExperimental(Op, DAG);
   case ISD::EXPERIMENTAL_VP_REVERSE:
     return lowerVPReverseExperimental(Op, DAG);
+  case ISD::CLEAR_CACHE: {
+    assert(getTargetMachine().getTargetTriple().isOSLinux() &&
+           "llvm.clear_cache only needs custom lower on Linux targets");
+    SDLoc DL(Op);
+    SDValue Flags = DAG.getConstant(0, DL, Subtarget.getXLenVT());
+    return emitFlushICache(DAG, Op.getOperand(0), Op.getOperand(1),
+                           Op.getOperand(2), Flags, DL);
+  }
   }
 }
 
+SDValue RISCVTargetLowering::emitFlushICache(SelectionDAG &DAG, SDValue InChain,
+                                             SDValue Start, SDValue End,
+                                             SDValue Flags, SDLoc DL) const {
+  MakeLibCallOptions CallOptions;
+  std::pair<SDValue, SDValue> CallResult =
+      makeLibCall(DAG, RTLIB::RISCV_FLUSH_ICACHE, MVT::isVoid,
+                  {Start, End, Flags}, CallOptions, DL, InChain);
+
+  // This function returns void so only the out chain matters.
+  return CallResult.second;
+}
+
 static SDValue getTargetNode(GlobalAddressSDNode *N, const SDLoc &DL, EVT Ty,
                              SelectionDAG &DAG, unsigned Flags) {
   return DAG.getTargetGlobalAddress(N->getGlobal(), DL, Ty, 0, Flags);

llvm/lib/Target/RISCV/RISCVISelLowering.h

@@ -1037,6 +1037,9 @@ class RISCVTargetLowering : public TargetLowering {
                                          const APInt &AndMask) const override;
 
   unsigned getMinimumJumpTableEntries() const override;
+
+  SDValue emitFlushICache(SelectionDAG &DAG, SDValue InChain, SDValue Start,
+                          SDValue End, SDValue Flags, SDLoc DL) const;
 };
 
 /// As per the spec, the rules for passing vector arguments are as follows:

llvm/test/CodeGen/RISCV/clear-cache.ll

@@ -0,0 +1,51 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=riscv32 < %s | FileCheck --check-prefix=RV32 %s
+; RUN: llc -mtriple=riscv64 < %s | FileCheck --check-prefix=RV64 %s
+; RUN: llc -mtriple=riscv32-unknown-linux-gnu < %s | FileCheck --check-prefix=RV32-LINUX %s
+; RUN: llc -mtriple=riscv64-unknown-linux-gnu < %s | FileCheck --check-prefix=RV64-LINUX %s
+; RUN: llc -mtriple=riscv32-unknown-linux-musl < %s | FileCheck --check-prefix=RV32-LINUX %s
+; RUN: llc -mtriple=riscv64-unknown-linux-musl < %s | FileCheck --check-prefix=RV64-LINUX %s
+
+declare void @llvm.clear_cache(ptr, ptr)
+
+define void @foo(ptr %a, ptr %b) nounwind {
+; RV32-LABEL: foo:
+; RV32:       # %bb.0:
+; RV32-NEXT:    addi sp, sp, -16
+; RV32-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32-NEXT:    call __clear_cache
+; RV32-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32-NEXT:    addi sp, sp, 16
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: foo:
+; RV64:       # %bb.0:
+; RV64-NEXT:    addi sp, sp, -16
+; RV64-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64-NEXT:    call __clear_cache
+; RV64-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64-NEXT:    addi sp, sp, 16
+; RV64-NEXT:    ret
+;
+; RV32-LINUX-LABEL: foo:
+; RV32-LINUX:       # %bb.0:
+; RV32-LINUX-NEXT:    addi sp, sp, -16
+; RV32-LINUX-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32-LINUX-NEXT:    li a2, 0
+; RV32-LINUX-NEXT:    call __riscv_flush_icache
+; RV32-LINUX-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32-LINUX-NEXT:    addi sp, sp, 16
+; RV32-LINUX-NEXT:    ret
+;
+; RV64-LINUX-LABEL: foo:
+; RV64-LINUX:       # %bb.0:
+; RV64-LINUX-NEXT:    addi sp, sp, -16
+; RV64-LINUX-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64-LINUX-NEXT:    li a2, 0
+; RV64-LINUX-NEXT:    call __riscv_flush_icache
+; RV64-LINUX-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64-LINUX-NEXT:    addi sp, sp, 16
+; RV64-LINUX-NEXT:    ret
+  call void @llvm.clear_cache(ptr %a, ptr %b)
+  ret void
+}