diff --git a/llvm/lib/Target/AArch64/AArch64InstructionSelector.cpp b/llvm/lib/Target/AArch64/AArch64InstructionSelector.cpp index 2a7f21819516..40fec6c26694 100644 --- a/llvm/lib/Target/AArch64/AArch64InstructionSelector.cpp +++ b/llvm/lib/Target/AArch64/AArch64InstructionSelector.cpp @@ -102,7 +102,7 @@ private: bool selectIntrinsicWithSideEffects(MachineInstr &I, MachineRegisterInfo &MRI) const; bool selectVectorICmp(MachineInstr &I, MachineRegisterInfo &MRI) const; - + bool selectIntrinsicTrunc(MachineInstr &I, MachineRegisterInfo &MRI) const; unsigned emitConstantPoolEntry(Constant *CPVal, MachineFunction &MF) const; MachineInstr *emitLoadFromConstantPool(Constant *CPVal, MachineIRBuilder &MIRBuilder) const; @@ -1847,6 +1847,8 @@ bool AArch64InstructionSelector::select(MachineInstr &I, return constrainSelectedInstRegOperands(*MovMI, TII, TRI, RBI); } } + case TargetOpcode::G_INTRINSIC_TRUNC: + return selectIntrinsicTrunc(I, MRI); case TargetOpcode::G_BUILD_VECTOR: return selectBuildVector(I, MRI); case TargetOpcode::G_MERGE_VALUES: @@ -1866,6 +1868,61 @@ bool AArch64InstructionSelector::select(MachineInstr &I, return false; } +bool AArch64InstructionSelector::selectIntrinsicTrunc( + MachineInstr &I, MachineRegisterInfo &MRI) const { + const LLT SrcTy = MRI.getType(I.getOperand(0).getReg()); + + // Select the correct opcode. + unsigned Opc = 0; + if (!SrcTy.isVector()) { + switch (SrcTy.getSizeInBits()) { + default: + case 16: + Opc = AArch64::FRINTZHr; + break; + case 32: + Opc = AArch64::FRINTZSr; + break; + case 64: + Opc = AArch64::FRINTZDr; + break; + } + } else { + unsigned NumElts = SrcTy.getNumElements(); + switch (SrcTy.getElementType().getSizeInBits()) { + default: + break; + case 16: + if (NumElts == 4) + Opc = AArch64::FRINTZv4f16; + else if (NumElts == 8) + Opc = AArch64::FRINTZv8f16; + break; + case 32: + if (NumElts == 2) + Opc = AArch64::FRINTZv2f32; + else if (NumElts == 4) + Opc = AArch64::FRINTZv4f32; + break; + case 64: + if (NumElts == 2) + Opc = AArch64::FRINTZv2f64; + break; + } + } + + if (!Opc) { + // Didn't get an opcode above, bail. + LLVM_DEBUG(dbgs() << "Unsupported type for G_INTRINSIC_TRUNC!\n"); + return false; + } + + // Legalization would have set us up perfectly for this; we just need to + // set the opcode and move on. + I.setDesc(TII.get(Opc)); + return constrainSelectedInstRegOperands(I, TII, TRI, RBI); +} + bool AArch64InstructionSelector::selectVectorICmp( MachineInstr &I, MachineRegisterInfo &MRI) const { unsigned DstReg = I.getOperand(0).getReg(); diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/select-intrinsic-trunc.mir b/llvm/test/CodeGen/AArch64/GlobalISel/select-intrinsic-trunc.mir new file mode 100644 index 000000000000..149264813155 --- /dev/null +++ b/llvm/test/CodeGen/AArch64/GlobalISel/select-intrinsic-trunc.mir @@ -0,0 +1,198 @@ +# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py +# RUN: llc %s -verify-machineinstrs -mtriple=aarch64-unknown-unknown -run-pass=instruction-select -mattr=+fullfp16 -o - | FileCheck %s + +... +--- +name: test_f64.intrinsic_trunc +alignment: 2 +legalized: true +regBankSelected: true +tracksRegLiveness: true +machineFunctionInfo: {} +body: | + bb.0: + liveins: $d0 + + ; CHECK-LABEL: name: test_f64.intrinsic_trunc + ; CHECK: liveins: $d0 + ; CHECK: [[COPY:%[0-9]+]]:fpr64 = COPY $d0 + ; CHECK: [[FRINTZDr:%[0-9]+]]:fpr64 = FRINTZDr [[COPY]] + ; CHECK: $d0 = COPY [[FRINTZDr]] + ; CHECK: RET_ReallyLR implicit $d0 + %0:fpr(s64) = COPY $d0 + %1:fpr(s64) = G_INTRINSIC_TRUNC %0 + $d0 = COPY %1(s64) + RET_ReallyLR implicit $d0 + +... +--- +name: test_f32.intrinsic_trunc +alignment: 2 +legalized: true +regBankSelected: true +tracksRegLiveness: true +machineFunctionInfo: {} +body: | + bb.0: + liveins: $s0 + + ; CHECK-LABEL: name: test_f32.intrinsic_trunc + ; CHECK: liveins: $s0 + ; CHECK: [[COPY:%[0-9]+]]:fpr32 = COPY $s0 + ; CHECK: [[FRINTZSr:%[0-9]+]]:fpr32 = FRINTZSr [[COPY]] + ; CHECK: $s0 = COPY [[FRINTZSr]] + ; CHECK: RET_ReallyLR implicit $s0 + %0:fpr(s32) = COPY $s0 + %1:fpr(s32) = G_INTRINSIC_TRUNC %0 + $s0 = COPY %1(s32) + RET_ReallyLR implicit $s0 + +... +--- +name: test_f16.intrinsic_trunc +alignment: 2 +legalized: true +regBankSelected: true +tracksRegLiveness: true +frameInfo: + maxCallFrameSize: 0 +machineFunctionInfo: {} +body: | + bb.0: + liveins: $h0 + + ; CHECK-LABEL: name: test_f16.intrinsic_trunc + ; CHECK: liveins: $h0 + ; CHECK: [[COPY:%[0-9]+]]:fpr16 = COPY $h0 + ; CHECK: [[FRINTZHr:%[0-9]+]]:fpr16 = FRINTZHr [[COPY]] + ; CHECK: $h0 = COPY [[FRINTZHr]] + ; CHECK: RET_ReallyLR implicit $h0 + %0:fpr(s16) = COPY $h0 + %1:fpr(s16) = G_INTRINSIC_TRUNC %0 + $h0 = COPY %1(s16) + RET_ReallyLR implicit $h0 + +... +--- +name: test_v4f16.intrinsic_trunc +alignment: 2 +legalized: true +regBankSelected: true +tracksRegLiveness: true +frameInfo: + maxCallFrameSize: 0 +machineFunctionInfo: {} +body: | + bb.0: + liveins: $d0 + + ; CHECK-LABEL: name: test_v4f16.intrinsic_trunc + ; CHECK: liveins: $d0 + ; CHECK: [[COPY:%[0-9]+]]:fpr64 = COPY $d0 + ; CHECK: [[FRINTZv4f16_:%[0-9]+]]:fpr64 = FRINTZv4f16 [[COPY]] + ; CHECK: $d0 = COPY [[FRINTZv4f16_]] + ; CHECK: RET_ReallyLR implicit $d0 + %0:fpr(<4 x s16>) = COPY $d0 + %1:fpr(<4 x s16>) = G_INTRINSIC_TRUNC %0 + $d0 = COPY %1(<4 x s16>) + RET_ReallyLR implicit $d0 + +... +--- +name: test_v8f16.intrinsic_trunc +alignment: 2 +legalized: true +regBankSelected: true +tracksRegLiveness: true +frameInfo: + maxCallFrameSize: 0 +machineFunctionInfo: {} +body: | + bb.0: + liveins: $q0 + + ; CHECK-LABEL: name: test_v8f16.intrinsic_trunc + ; CHECK: liveins: $q0 + ; CHECK: [[COPY:%[0-9]+]]:fpr128 = COPY $q0 + ; CHECK: [[FRINTZv8f16_:%[0-9]+]]:fpr128 = FRINTZv8f16 [[COPY]] + ; CHECK: $q0 = COPY [[FRINTZv8f16_]] + ; CHECK: RET_ReallyLR implicit $q0 + %0:fpr(<8 x s16>) = COPY $q0 + %1:fpr(<8 x s16>) = G_INTRINSIC_TRUNC %0 + $q0 = COPY %1(<8 x s16>) + RET_ReallyLR implicit $q0 + +... +--- +name: test_v2f32.intrinsic_trunc +alignment: 2 +legalized: true +regBankSelected: true +tracksRegLiveness: true +frameInfo: + maxCallFrameSize: 0 +machineFunctionInfo: {} +body: | + bb.0: + liveins: $d0 + + ; CHECK-LABEL: name: test_v2f32.intrinsic_trunc + ; CHECK: liveins: $d0 + ; CHECK: [[COPY:%[0-9]+]]:fpr64 = COPY $d0 + ; CHECK: [[FRINTZv2f32_:%[0-9]+]]:fpr64 = FRINTZv2f32 [[COPY]] + ; CHECK: $d0 = COPY [[FRINTZv2f32_]] + ; CHECK: RET_ReallyLR implicit $d0 + %0:fpr(<2 x s32>) = COPY $d0 + %1:fpr(<2 x s32>) = G_INTRINSIC_TRUNC %0 + $d0 = COPY %1(<2 x s32>) + RET_ReallyLR implicit $d0 + +... +--- +name: test_v4f32.intrinsic_trunc +alignment: 2 +legalized: true +regBankSelected: true +tracksRegLiveness: true +frameInfo: + maxCallFrameSize: 0 +machineFunctionInfo: {} +body: | + bb.0: + liveins: $q0 + + ; CHECK-LABEL: name: test_v4f32.intrinsic_trunc + ; CHECK: liveins: $q0 + ; CHECK: [[COPY:%[0-9]+]]:fpr128 = COPY $q0 + ; CHECK: [[FRINTZv4f32_:%[0-9]+]]:fpr128 = FRINTZv4f32 [[COPY]] + ; CHECK: $q0 = COPY [[FRINTZv4f32_]] + ; CHECK: RET_ReallyLR implicit $q0 + %0:fpr(<4 x s32>) = COPY $q0 + %1:fpr(<4 x s32>) = G_INTRINSIC_TRUNC %0 + $q0 = COPY %1(<4 x s32>) + RET_ReallyLR implicit $q0 + +... +--- +name: test_v2f64.intrinsic_trunc +alignment: 2 +legalized: true +regBankSelected: true +tracksRegLiveness: true +frameInfo: + maxCallFrameSize: 0 +machineFunctionInfo: {} +body: | + bb.0: + liveins: $q0 + + ; CHECK-LABEL: name: test_v2f64.intrinsic_trunc + ; CHECK: liveins: $q0 + ; CHECK: [[COPY:%[0-9]+]]:fpr128 = COPY $q0 + ; CHECK: [[FRINTZv2f64_:%[0-9]+]]:fpr128 = FRINTZv2f64 [[COPY]] + ; CHECK: $q0 = COPY [[FRINTZv2f64_]] + ; CHECK: RET_ReallyLR implicit $q0 + %0:fpr(<2 x s64>) = COPY $q0 + %1:fpr(<2 x s64>) = G_INTRINSIC_TRUNC %0 + $q0 = COPY %1(<2 x s64>) + RET_ReallyLR implicit $q0