From bc2e640595a40114f82851476b7b4c6e34d52732 Mon Sep 17 00:00:00 2001
From: Gergely Futo <gergely.futo@hightec-rt.com>
Date: Sat, 17 May 2025 09:37:43 +0200
Subject: [PATCH] [RISCV] Implement
 RISCVTargetLowering::getRoundingControlRegisters

By adding FRM/FFLAGS as implicit defs, ReadFRM is not optimized out.
---
 llvm/lib/Target/RISCV/RISCVISelLowering.cpp |  8 ++++
 llvm/lib/Target/RISCV/RISCVISelLowering.h   |  2 +
 llvm/test/CodeGen/RISCV/fpenv.ll            | 44 +++++++++++++--------
 3 files changed, 38 insertions(+), 16 deletions(-)

diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 8c9ca63751a9b..1158499718737 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -24509,3 +24509,11 @@ RISCVTargetLowering::emitDynamicProbedAlloc(MachineInstr &MI,
   MF.getInfo<RISCVMachineFunctionInfo>()->setDynamicAllocation();
   return ExitMBB->begin()->getParent();
 }
+
+ArrayRef<MCPhysReg> RISCVTargetLowering::getRoundingControlRegisters() const {
+  if (Subtarget.hasStdExtFOrZfinx()) {
+    static const MCPhysReg RCRegs[] = {RISCV::FRM, RISCV::FFLAGS};
+    return RCRegs;
+  }
+  return {};
+}
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.h b/llvm/lib/Target/RISCV/RISCVISelLowering.h
index c865da80bcdf6..fc8d8b8ce1b56 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.h
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.h
@@ -450,6 +450,8 @@ class RISCVTargetLowering : public TargetLowering {
   MachineBasicBlock *emitDynamicProbedAlloc(MachineInstr &MI,
                                             MachineBasicBlock *MBB) const;
 
+  ArrayRef<MCPhysReg> getRoundingControlRegisters() const override;
+
 private:
   void analyzeInputArgs(MachineFunction &MF, CCState &CCInfo,
                         const SmallVectorImpl<ISD::InputArg> &Ins, bool IsRet,
diff --git a/llvm/test/CodeGen/RISCV/fpenv.ll b/llvm/test/CodeGen/RISCV/fpenv.ll
index d9da14bc6799d..11e104a290e1b 100644
--- a/llvm/test/CodeGen/RISCV/fpenv.ll
+++ b/llvm/test/CodeGen/RISCV/fpenv.ll
@@ -40,16 +40,22 @@ define i1 @test_get_rounding_sideeffect() #0 {
 ; RV32IF-NEXT:    frrm a0
 ; RV32IF-NEXT:    lui a1, 66
 ; RV32IF-NEXT:    slli a0, a0, 2
-; RV32IF-NEXT:    addi a1, a1, 769
-; RV32IF-NEXT:    srl s0, a1, a0
+; RV32IF-NEXT:    addi s0, a1, 769
+; RV32IF-NEXT:    srl a0, s0, a0
+; RV32IF-NEXT:    andi a0, a0, 7
+; RV32IF-NEXT:    beqz a0, .LBB1_2
+; RV32IF-NEXT:  # %bb.1:
 ; RV32IF-NEXT:    li a0, 0
-; RV32IF-NEXT:    andi s0, s0, 7
-; RV32IF-NEXT:    bnez s0, .LBB1_2
-; RV32IF-NEXT:  # %bb.1: # %if.end
+; RV32IF-NEXT:    j .LBB1_3
+; RV32IF-NEXT:  .LBB1_2: # %if.end
 ; RV32IF-NEXT:    call fesetround
-; RV32IF-NEXT:    addi s0, s0, -1
-; RV32IF-NEXT:    seqz a0, s0
-; RV32IF-NEXT:  .LBB1_2: # %return
+; RV32IF-NEXT:    frrm a0
+; RV32IF-NEXT:    slli a0, a0, 2
+; RV32IF-NEXT:    srl a0, s0, a0
+; RV32IF-NEXT:    andi a0, a0, 7
+; RV32IF-NEXT:    addi a0, a0, -1
+; RV32IF-NEXT:    seqz a0, a0
+; RV32IF-NEXT:  .LBB1_3: # %return
 ; RV32IF-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
 ; RV32IF-NEXT:    .cfi_restore ra
@@ -71,16 +77,22 @@ define i1 @test_get_rounding_sideeffect() #0 {
 ; RV64IF-NEXT:    frrm a0
 ; RV64IF-NEXT:    lui a1, 66
 ; RV64IF-NEXT:    slli a0, a0, 2
-; RV64IF-NEXT:    addiw a1, a1, 769
-; RV64IF-NEXT:    srl s0, a1, a0
+; RV64IF-NEXT:    addiw s0, a1, 769
+; RV64IF-NEXT:    srl a0, s0, a0
+; RV64IF-NEXT:    andi a0, a0, 7
+; RV64IF-NEXT:    beqz a0, .LBB1_2
+; RV64IF-NEXT:  # %bb.1:
 ; RV64IF-NEXT:    li a0, 0
-; RV64IF-NEXT:    andi s0, s0, 7
-; RV64IF-NEXT:    bnez s0, .LBB1_2
-; RV64IF-NEXT:  # %bb.1: # %if.end
+; RV64IF-NEXT:    j .LBB1_3
+; RV64IF-NEXT:  .LBB1_2: # %if.end
 ; RV64IF-NEXT:    call fesetround
-; RV64IF-NEXT:    addi s0, s0, -1
-; RV64IF-NEXT:    seqz a0, s0
-; RV64IF-NEXT:  .LBB1_2: # %return
+; RV64IF-NEXT:    frrm a0
+; RV64IF-NEXT:    slli a0, a0, 2
+; RV64IF-NEXT:    srl a0, s0, a0
+; RV64IF-NEXT:    andi a0, a0, 7
+; RV64IF-NEXT:    addi a0, a0, -1
+; RV64IF-NEXT:    seqz a0, a0
+; RV64IF-NEXT:  .LBB1_3: # %return
 ; RV64IF-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64IF-NEXT:    ld s0, 0(sp) # 8-byte Folded Reload
 ; RV64IF-NEXT:    .cfi_restore ra