https://github.com/heiher updated https://github.com/llvm/llvm-project/pull/233760
>From fafcd65ec6ef34192bd0e34639331fa68acc783b Mon Sep 17 00:00:00 2001 From: WANG Rui <[email protected]> Date: Tue, 20 Oct 2026 15:42:12 +0800 Subject: [PATCH] [LoongArch] Lower FP_TO_SINT_SAT to [X]VFTINTRZ/FTINTRZ The ftintrz.{w,l}.{s,d} and [x]vftintrz.{w.s,l.d} instructions saturate out-of-range inputs to the minimum/maximum integer and convert NaN to 0 (when the invalid operation exception is not trapped), which matches the semantics of llvm.fptosi.sat when the saturation width is the native 32/64 bits. This was verified on LA664 for all scalar, LSX and LASX variants and matches QEMU's LoongArch model. Use them instead of the generic expansion, which emits three compares and selects per element and fully scalarizes vectors. Also override shouldConvertFpToSat so DAGCombiner only folds min/max clamps and NaN-guarded selects into FP_TO_SINT_SAT when that lowers to a single ftintrz, avoiding the generic expansion for mismatched vector element widths and for i64 results without the D extension. This is what Rust emits for `f32 as i32`. rustc-perf svg-render-1 on LA664: instructions -6.4%, cycles -4.0%. --- .../LoongArch/LoongArchISelLowering.cpp | 58 ++ .../Target/LoongArch/LoongArchISelLowering.h | 3 + .../LoongArch/LoongArchLASXInstrInfo.td | 2 + .../Target/LoongArch/LoongArchLSXInstrInfo.td | 2 + llvm/test/CodeGen/LoongArch/fptosi-sat.ll | 183 +----- .../test/CodeGen/LoongArch/lasx/fptosi-sat.ll | 565 +----------------- llvm/test/CodeGen/LoongArch/lsx/fptosi-sat.ll | 317 +--------- 7 files changed, 107 insertions(+), 1023 deletions(-) diff --git a/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp b/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp index 1d01e76a5ebd2d..700b1a07b718c6 100644 --- a/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp +++ b/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp @@ -111,6 +111,8 @@ LoongArchTargetLowering::LoongArchTargetLowering(const TargetMachine &TM, setOperationAction(ISD::SRA_PARTS, GRLenVT, Custom); setOperationAction(ISD::SRL_PARTS, GRLenVT, Custom); setOperationAction(ISD::FP_TO_SINT, GRLenVT, Custom); + if (Subtarget.hasBasicF()) + setOperationAction(ISD::FP_TO_SINT_SAT, GRLenVT, Custom); setOperationAction(ISD::ROTL, GRLenVT, Expand); setOperationAction(ISD::CTPOP, GRLenVT, Expand); @@ -364,6 +366,7 @@ LoongArchTargetLowering::LoongArchTargetLowering(const TargetMachine &TM, for (MVT VT : {MVT::v4i32, MVT::v2i64}) { setOperationAction({ISD::SINT_TO_FP, ISD::UINT_TO_FP}, VT, Legal); setOperationAction({ISD::FP_TO_SINT, ISD::FP_TO_UINT}, VT, Legal); + setOperationAction(ISD::FP_TO_SINT_SAT, VT, Custom); } setOperationAction(ISD::UINT_TO_FP, GRLenVT, Custom); for (MVT VT : {MVT::v4f32, MVT::v2f64}) { @@ -464,6 +467,8 @@ LoongArchTargetLowering::LoongArchTargetLowering(const TargetMachine &TM, setOperationAction(ISD::BITREVERSE, VT, Custom); for (MVT VT : {MVT::v16i16, MVT::v8i32, MVT::v4i64}) setOperationAction(ISD::BSWAP, VT, Legal); + for (MVT VT : {MVT::v8i32, MVT::v4i64}) + setOperationAction(ISD::FP_TO_SINT_SAT, VT, Custom); for (MVT VT : {MVT::v8i32, MVT::v4i32, MVT::v4i64}) { setOperationAction({ISD::FP_TO_SINT, ISD::FP_TO_UINT}, VT, Legal); setOperationAction(ISD::SINT_TO_FP, VT, Legal); @@ -619,6 +624,8 @@ SDValue LoongArchTargetLowering::LowerOperation(SDValue Op, return lowerFP_TO_SINT(Op, DAG); case ISD::FP_TO_UINT: return lowerFP_TO_UINT(Op, DAG); + case ISD::FP_TO_SINT_SAT: + return lowerFP_TO_SINT_SAT(Op, DAG); case ISD::BITCAST: return lowerBITCAST(Op, DAG); case ISD::UINT_TO_FP: @@ -4381,6 +4388,43 @@ SDValue LoongArchTargetLowering::lowerFP_TO_SINT(SDValue Op, return DAG.getNode(ISD::BITCAST, DL, Op.getValueType(), Trunc); } +// [X]VFTINTRZ and FTINTRZ saturate out-of-range values and convert NaN to 0, +// which is exactly the semantics of FP_TO_SINT_SAT with a native saturation +// width. +SDValue LoongArchTargetLowering::lowerFP_TO_SINT_SAT(SDValue Op, + SelectionDAG &DAG) const { + SDLoc DL(Op); + SDValue Src = Op.getOperand(0); + EVT VT = Op.getValueType(); + EVT SrcVT = Src.getValueType(); + unsigned SatWidth = cast<VTSDNode>(Op.getOperand(1))->getVT().getSizeInBits(); + + // Vectors are selected by patterns. + if (VT.isVector()) { + if (SatWidth == VT.getScalarSizeInBits() && + SatWidth == SrcVT.getScalarSizeInBits()) + return Op; + return SDValue(); + } + + if (SrcVT != MVT::f32 && SrcVT != MVT::f64) + return SDValue(); + + if (SatWidth == 32) { + SDValue Ftint = DAG.getNode(LoongArchISD::FTINT, DL, MVT::f32, Src); + if (Subtarget.is64Bit()) + return DAG.getNode(LoongArchISD::MOVFR2GR_S_LA64, DL, VT, Ftint); + return DAG.getBitcast(VT, Ftint); + } + + if (SatWidth == 64 && Subtarget.is64Bit() && Subtarget.hasBasicD()) { + SDValue Ftint = DAG.getNode(LoongArchISD::FTINT, DL, MVT::f64, Src); + return DAG.getBitcast(VT, Ftint); + } + + return SDValue(); +} + SDValue LoongArchTargetLowering::lowerFP_TO_UINT(SDValue Op, SelectionDAG &DAG) const { if (!Subtarget.hasExtLSX()) @@ -11018,6 +11062,20 @@ bool LoongArchTargetLowering::isFPImmVLDILegal(const APFloat &Imm, return false; } +// Only form FP_TO_SINT_SAT from clamp/NaN-guard idioms when it will be lowered +// to a single [X]VFTINTRZ/FTINTRZ, otherwise the generic expansion is worse +// than the original sequence. +bool LoongArchTargetLowering::shouldConvertFpToSat(unsigned Op, EVT FPVT, + EVT VT) const { + if (!isOperationLegalOrCustom(Op, VT)) + return false; + if (VT.isVector()) + return FPVT.getScalarSizeInBits() == VT.getScalarSizeInBits(); + if (FPVT == MVT::f32) + return VT.getSizeInBits() == 32 || Subtarget.hasBasicD(); + return FPVT == MVT::f64 && Subtarget.hasBasicD(); +} + bool LoongArchTargetLowering::isFPImmLegal(const APFloat &Imm, EVT VT, bool ForCodeSize) const { // TODO: Maybe need more checks here after vector extension is supported. diff --git a/llvm/lib/Target/LoongArch/LoongArchISelLowering.h b/llvm/lib/Target/LoongArch/LoongArchISelLowering.h index 7c3ae7d6aac14a..ee66fb191c2429 100644 --- a/llvm/lib/Target/LoongArch/LoongArchISelLowering.h +++ b/llvm/lib/Target/LoongArch/LoongArchISelLowering.h @@ -229,6 +229,7 @@ class LoongArchTargetLowering : public TargetLowering { SDValue lowerEH_DWARF_CFA(SDValue Op, SelectionDAG &DAG) const; SDValue lowerFP_TO_SINT(SDValue Op, SelectionDAG &DAG) const; SDValue lowerFP_TO_UINT(SDValue Op, SelectionDAG &DAG) const; + SDValue lowerFP_TO_SINT_SAT(SDValue Op, SelectionDAG &DAG) const; SDValue lowerBITCAST(SDValue Op, SelectionDAG &DAG) const; SDValue lowerUINT_TO_FP(SDValue Op, SelectionDAG &DAG) const; SDValue lowerSINT_TO_FP(SDValue Op, SelectionDAG &DAG) const; @@ -269,6 +270,8 @@ class LoongArchTargetLowering : public TargetLowering { bool isFPImmLegal(const APFloat &Imm, EVT VT, bool ForCodeSize) const override; + bool shouldConvertFpToSat(unsigned Op, EVT FPVT, EVT VT) const override; + bool shouldInsertFencesForAtomic(const Instruction *I) const override; ConstraintType getConstraintType(StringRef Constraint) const override; diff --git a/llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td b/llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td index fcd134770332de..690fe4afff212a 100644 --- a/llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td +++ b/llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td @@ -2193,6 +2193,8 @@ def : Pat<(v8f32 (loongarch_vffint_s_l (v4i64 LASX256:$xj), (v4i64 LASX256:$xk)) // XVFTINTRZ_{W_S/L_D} def : Pat<(v8i32 (fp_to_sint v8f32:$vj)), (XVFTINTRZ_W_S v8f32:$vj)>; def : Pat<(v4i64 (fp_to_sint v4f64:$vj)), (XVFTINTRZ_L_D v4f64:$vj)>; +def : Pat<(v8i32 (fp_to_sint_sat v8f32:$vj, i32)), (XVFTINTRZ_W_S v8f32:$vj)>; +def : Pat<(v4i64 (fp_to_sint_sat v4f64:$vj, i64)), (XVFTINTRZ_L_D v4f64:$vj)>; def : Pat<(v4i64 (fp_to_sint v4f32:$vj)), (XVFTINTRZL_L_S (XVPERMI_D (SUBREG_TO_REG v4f32:$vj, sub_128), 216))>; diff --git a/llvm/lib/Target/LoongArch/LoongArchLSXInstrInfo.td b/llvm/lib/Target/LoongArch/LoongArchLSXInstrInfo.td index e1489f9e36d69b..d6eb050de12776 100644 --- a/llvm/lib/Target/LoongArch/LoongArchLSXInstrInfo.td +++ b/llvm/lib/Target/LoongArch/LoongArchLSXInstrInfo.td @@ -2412,6 +2412,8 @@ def : Pat<(v4f32 (loongarch_vffint_s_l (v2i64 LSX128:$vj), (v2i64 LSX128:$vk))), // VFTINTRZ_{W_S/L_D} def : Pat<(v4i32 (fp_to_sint v4f32:$vj)), (VFTINTRZ_W_S v4f32:$vj)>; def : Pat<(v2i64 (fp_to_sint v2f64:$vj)), (VFTINTRZ_L_D v2f64:$vj)>; +def : Pat<(v4i32 (fp_to_sint_sat v4f32:$vj, i32)), (VFTINTRZ_W_S v4f32:$vj)>; +def : Pat<(v2i64 (fp_to_sint_sat v2f64:$vj, i64)), (VFTINTRZ_L_D v2f64:$vj)>; // VFTINTRZ_{W_SU/L_DU} def : Pat<(v4i32 (fp_to_uint v4f32:$vj)), (VFTINTRZ_WU_S v4f32:$vj)>; diff --git a/llvm/test/CodeGen/LoongArch/fptosi-sat.ll b/llvm/test/CodeGen/LoongArch/fptosi-sat.ll index 96c563a48ef000..73889aed522c2f 100644 --- a/llvm/test/CodeGen/LoongArch/fptosi-sat.ll +++ b/llvm/test/CodeGen/LoongArch/fptosi-sat.ll @@ -141,110 +141,26 @@ entry: define i32 @fptosi_sat_f32_to_i32(float %a) nounwind { ; LA32F-LABEL: fptosi_sat_f32_to_i32: ; LA32F: # %bb.0: # %entry -; LA32F-NEXT: lu12i.w $a0, -200704 -; LA32F-NEXT: movgr2fr.w $fa1, $a0 -; LA32F-NEXT: fcmp.cult.s $fcc0, $fa0, $fa1 -; LA32F-NEXT: ftintrz.w.s $fa1, $fa0 -; LA32F-NEXT: movfr2gr.s $a0, $fa1 -; LA32F-NEXT: movcf2gr $a1, $fcc0 -; LA32F-NEXT: masknez $a0, $a0, $a1 -; LA32F-NEXT: lu12i.w $a2, -524288 -; LA32F-NEXT: maskeqz $a1, $a2, $a1 -; LA32F-NEXT: lu12i.w $a2, 323583 -; LA32F-NEXT: ori $a2, $a2, 4095 -; LA32F-NEXT: movgr2fr.w $fa1, $a2 -; LA32F-NEXT: fcmp.clt.s $fcc0, $fa1, $fa0 -; LA32F-NEXT: or $a0, $a1, $a0 -; LA32F-NEXT: movcf2gr $a1, $fcc0 -; LA32F-NEXT: masknez $a0, $a0, $a1 -; LA32F-NEXT: lu12i.w $a2, 524287 -; LA32F-NEXT: ori $a2, $a2, 4095 -; LA32F-NEXT: maskeqz $a1, $a2, $a1 -; LA32F-NEXT: fcmp.cun.s $fcc0, $fa0, $fa0 -; LA32F-NEXT: or $a0, $a1, $a0 -; LA32F-NEXT: movcf2gr $a1, $fcc0 -; LA32F-NEXT: masknez $a0, $a0, $a1 +; LA32F-NEXT: ftintrz.w.s $fa0, $fa0 +; LA32F-NEXT: movfr2gr.s $a0, $fa0 ; LA32F-NEXT: ret ; ; LA32D-LABEL: fptosi_sat_f32_to_i32: ; LA32D: # %bb.0: # %entry -; LA32D-NEXT: lu12i.w $a0, -200704 -; LA32D-NEXT: movgr2fr.w $fa1, $a0 -; LA32D-NEXT: fcmp.cult.s $fcc0, $fa0, $fa1 -; LA32D-NEXT: ftintrz.w.s $fa1, $fa0 -; LA32D-NEXT: movfr2gr.s $a0, $fa1 -; LA32D-NEXT: movcf2gr $a1, $fcc0 -; LA32D-NEXT: masknez $a0, $a0, $a1 -; LA32D-NEXT: lu12i.w $a2, -524288 -; LA32D-NEXT: maskeqz $a1, $a2, $a1 -; LA32D-NEXT: lu12i.w $a2, 323583 -; LA32D-NEXT: ori $a2, $a2, 4095 -; LA32D-NEXT: movgr2fr.w $fa1, $a2 -; LA32D-NEXT: fcmp.clt.s $fcc0, $fa1, $fa0 -; LA32D-NEXT: or $a0, $a1, $a0 -; LA32D-NEXT: movcf2gr $a1, $fcc0 -; LA32D-NEXT: masknez $a0, $a0, $a1 -; LA32D-NEXT: lu12i.w $a2, 524287 -; LA32D-NEXT: ori $a2, $a2, 4095 -; LA32D-NEXT: maskeqz $a1, $a2, $a1 -; LA32D-NEXT: fcmp.cun.s $fcc0, $fa0, $fa0 -; LA32D-NEXT: or $a0, $a1, $a0 -; LA32D-NEXT: movcf2gr $a1, $fcc0 -; LA32D-NEXT: masknez $a0, $a0, $a1 +; LA32D-NEXT: ftintrz.w.s $fa0, $fa0 +; LA32D-NEXT: movfr2gr.s $a0, $fa0 ; LA32D-NEXT: ret ; ; LA64F-LABEL: fptosi_sat_f32_to_i32: ; LA64F: # %bb.0: # %entry -; LA64F-NEXT: lu12i.w $a0, -200704 -; LA64F-NEXT: movgr2fr.w $fa1, $a0 -; LA64F-NEXT: fcmp.cult.s $fcc0, $fa0, $fa1 -; LA64F-NEXT: ftintrz.w.s $fa1, $fa0 -; LA64F-NEXT: movfr2gr.s $a0, $fa1 -; LA64F-NEXT: movcf2gr $a1, $fcc0 -; LA64F-NEXT: masknez $a0, $a0, $a1 -; LA64F-NEXT: lu12i.w $a2, -524288 -; LA64F-NEXT: maskeqz $a1, $a2, $a1 -; LA64F-NEXT: lu12i.w $a2, 323583 -; LA64F-NEXT: ori $a2, $a2, 4095 -; LA64F-NEXT: movgr2fr.w $fa1, $a2 -; LA64F-NEXT: fcmp.clt.s $fcc0, $fa1, $fa0 -; LA64F-NEXT: or $a0, $a1, $a0 -; LA64F-NEXT: movcf2gr $a1, $fcc0 -; LA64F-NEXT: masknez $a0, $a0, $a1 -; LA64F-NEXT: lu12i.w $a2, 524287 -; LA64F-NEXT: ori $a2, $a2, 4095 -; LA64F-NEXT: maskeqz $a1, $a2, $a1 -; LA64F-NEXT: fcmp.cun.s $fcc0, $fa0, $fa0 -; LA64F-NEXT: or $a0, $a1, $a0 -; LA64F-NEXT: movcf2gr $a1, $fcc0 -; LA64F-NEXT: masknez $a0, $a0, $a1 +; LA64F-NEXT: ftintrz.w.s $fa0, $fa0 +; LA64F-NEXT: movfr2gr.s $a0, $fa0 ; LA64F-NEXT: ret ; ; LA64D-LABEL: fptosi_sat_f32_to_i32: ; LA64D: # %bb.0: # %entry -; LA64D-NEXT: lu12i.w $a0, -200704 -; LA64D-NEXT: movgr2fr.w $fa1, $a0 -; LA64D-NEXT: fcmp.cult.s $fcc0, $fa0, $fa1 -; LA64D-NEXT: ftintrz.l.s $fa1, $fa0 -; LA64D-NEXT: movfr2gr.d $a0, $fa1 -; LA64D-NEXT: movcf2gr $a1, $fcc0 -; LA64D-NEXT: masknez $a0, $a0, $a1 -; LA64D-NEXT: lu12i.w $a2, -524288 -; LA64D-NEXT: maskeqz $a1, $a2, $a1 -; LA64D-NEXT: lu12i.w $a2, 323583 -; LA64D-NEXT: ori $a2, $a2, 4095 -; LA64D-NEXT: movgr2fr.w $fa1, $a2 -; LA64D-NEXT: fcmp.clt.s $fcc0, $fa1, $fa0 -; LA64D-NEXT: or $a0, $a1, $a0 -; LA64D-NEXT: movcf2gr $a1, $fcc0 -; LA64D-NEXT: masknez $a0, $a0, $a1 -; LA64D-NEXT: lu12i.w $a2, 524287 -; LA64D-NEXT: ori $a2, $a2, 4095 -; LA64D-NEXT: maskeqz $a1, $a2, $a1 -; LA64D-NEXT: fcmp.cun.s $fcc0, $fa0, $fa0 -; LA64D-NEXT: or $a0, $a1, $a0 -; LA64D-NEXT: movcf2gr $a1, $fcc0 -; LA64D-NEXT: masknez $a0, $a0, $a1 +; LA64D-NEXT: ftintrz.w.s $fa0, $fa0 +; LA64D-NEXT: movfr2gr.s $a0, $fa0 ; LA64D-NEXT: ret entry: %r = call i32 @llvm.fptosi.sat.i32.f32(float %a) @@ -377,29 +293,8 @@ define i64 @fptosi_sat_f32_to_i64(float %a) nounwind { ; ; LA64D-LABEL: fptosi_sat_f32_to_i64: ; LA64D: # %bb.0: # %entry -; LA64D-NEXT: lu12i.w $a0, -135168 -; LA64D-NEXT: movgr2fr.w $fa1, $a0 -; LA64D-NEXT: fcmp.cult.s $fcc0, $fa0, $fa1 -; LA64D-NEXT: ftintrz.l.s $fa1, $fa0 -; LA64D-NEXT: movfr2gr.d $a0, $fa1 -; LA64D-NEXT: movcf2gr $a1, $fcc0 -; LA64D-NEXT: masknez $a0, $a0, $a1 -; LA64D-NEXT: lu52i.d $a2, $zero, -2048 -; LA64D-NEXT: maskeqz $a1, $a2, $a1 -; LA64D-NEXT: lu12i.w $a2, 389119 -; LA64D-NEXT: ori $a2, $a2, 4095 -; LA64D-NEXT: movgr2fr.w $fa1, $a2 -; LA64D-NEXT: fcmp.clt.s $fcc0, $fa1, $fa0 -; LA64D-NEXT: or $a0, $a1, $a0 -; LA64D-NEXT: movcf2gr $a1, $fcc0 -; LA64D-NEXT: masknez $a0, $a0, $a1 -; LA64D-NEXT: addi.w $a2, $zero, -1 -; LA64D-NEXT: lu52i.d $a2, $a2, 2047 -; LA64D-NEXT: maskeqz $a1, $a2, $a1 -; LA64D-NEXT: fcmp.cun.s $fcc0, $fa0, $fa0 -; LA64D-NEXT: or $a0, $a1, $a0 -; LA64D-NEXT: movcf2gr $a1, $fcc0 -; LA64D-NEXT: masknez $a0, $a0, $a1 +; LA64D-NEXT: ftintrz.l.s $fa0, $fa0 +; LA64D-NEXT: movfr2gr.d $a0, $fa0 ; LA64D-NEXT: ret entry: %r = call i64 @llvm.fptosi.sat.i64.f32(float %a) @@ -700,18 +595,8 @@ define i32 @fptosi_sat_f64_to_i32(double %a) nounwind { ; ; LA32D-LABEL: fptosi_sat_f64_to_i32: ; LA32D: # %bb.0: # %entry -; LA32D-NEXT: movgr2fr.w $fa1, $zero -; LA32D-NEXT: pcalau12i $a0, %pc_hi20(.LCPI6_0) -; LA32D-NEXT: fld.d $fa2, $a0, %pc_lo12(.LCPI6_0) -; LA32D-NEXT: lu12i.w $a0, -254464 -; LA32D-NEXT: movgr2frh.w $fa1, $a0 -; LA32D-NEXT: fmax.d $fa1, $fa0, $fa1 -; LA32D-NEXT: fmin.d $fa1, $fa1, $fa2 -; LA32D-NEXT: ftintrz.w.d $fa1, $fa1 -; LA32D-NEXT: fcmp.cun.d $fcc0, $fa0, $fa0 -; LA32D-NEXT: movfr2gr.s $a0, $fa1 -; LA32D-NEXT: movcf2gr $a1, $fcc0 -; LA32D-NEXT: masknez $a0, $a0, $a1 +; LA32D-NEXT: ftintrz.w.d $fa0, $fa0 +; LA32D-NEXT: movfr2gr.s $a0, $fa0 ; LA32D-NEXT: ret ; ; LA64F-LABEL: fptosi_sat_f64_to_i32: @@ -757,18 +642,8 @@ define i32 @fptosi_sat_f64_to_i32(double %a) nounwind { ; ; LA64D-LABEL: fptosi_sat_f64_to_i32: ; LA64D: # %bb.0: # %entry -; LA64D-NEXT: lu52i.d $a0, $zero, -994 -; LA64D-NEXT: movgr2fr.d $fa1, $a0 -; LA64D-NEXT: fmax.d $fa1, $fa0, $fa1 -; LA64D-NEXT: lu12i.w $a0, -1024 -; LA64D-NEXT: lu52i.d $a0, $a0, 1053 -; LA64D-NEXT: movgr2fr.d $fa2, $a0 -; LA64D-NEXT: fmin.d $fa1, $fa1, $fa2 -; LA64D-NEXT: ftintrz.l.d $fa1, $fa1 -; LA64D-NEXT: fcmp.cun.d $fcc0, $fa0, $fa0 -; LA64D-NEXT: movfr2gr.d $a0, $fa1 -; LA64D-NEXT: movcf2gr $a1, $fcc0 -; LA64D-NEXT: masknez $a0, $a0, $a1 +; LA64D-NEXT: ftintrz.w.d $fa0, $fa0 +; LA64D-NEXT: movfr2gr.s $a0, $fa0 ; LA64D-NEXT: ret entry: %r = call i32 @llvm.fptosi.sat.i32.f64(double %a) @@ -928,28 +803,8 @@ define i64 @fptosi_sat_f64_to_i64(double %a) nounwind { ; ; LA64D-LABEL: fptosi_sat_f64_to_i64: ; LA64D: # %bb.0: # %entry -; LA64D-NEXT: lu52i.d $a0, $zero, -962 -; LA64D-NEXT: movgr2fr.d $fa1, $a0 -; LA64D-NEXT: fcmp.cult.d $fcc0, $fa0, $fa1 -; LA64D-NEXT: ftintrz.l.d $fa1, $fa0 -; LA64D-NEXT: movfr2gr.d $a0, $fa1 -; LA64D-NEXT: movcf2gr $a1, $fcc0 -; LA64D-NEXT: masknez $a0, $a0, $a1 -; LA64D-NEXT: lu52i.d $a2, $zero, -2048 -; LA64D-NEXT: maskeqz $a1, $a2, $a1 -; LA64D-NEXT: addi.w $a2, $zero, -1 -; LA64D-NEXT: lu52i.d $a3, $a2, 1085 -; LA64D-NEXT: movgr2fr.d $fa1, $a3 -; LA64D-NEXT: fcmp.clt.d $fcc0, $fa1, $fa0 -; LA64D-NEXT: or $a0, $a1, $a0 -; LA64D-NEXT: movcf2gr $a1, $fcc0 -; LA64D-NEXT: masknez $a0, $a0, $a1 -; LA64D-NEXT: lu52i.d $a2, $a2, 2047 -; LA64D-NEXT: maskeqz $a1, $a2, $a1 -; LA64D-NEXT: fcmp.cun.d $fcc0, $fa0, $fa0 -; LA64D-NEXT: or $a0, $a1, $a0 -; LA64D-NEXT: movcf2gr $a1, $fcc0 -; LA64D-NEXT: masknez $a0, $a0, $a1 +; LA64D-NEXT: ftintrz.l.d $fa0, $fa0 +; LA64D-NEXT: movfr2gr.d $a0, $fa0 ; LA64D-NEXT: ret entry: %r = call i64 @llvm.fptosi.sat.i64.f64(double %a) @@ -1002,11 +857,8 @@ define i64 @fptosi_nanguard_f32_to_i64(float %a) nounwind { ; ; LA64D-LABEL: fptosi_nanguard_f32_to_i64: ; LA64D: # %bb.0: # %entry -; LA64D-NEXT: fcmp.cun.s $fcc0, $fa0, $fa0 ; LA64D-NEXT: ftintrz.l.s $fa0, $fa0 ; LA64D-NEXT: movfr2gr.d $a0, $fa0 -; LA64D-NEXT: movcf2gr $a1, $fcc0 -; LA64D-NEXT: masknez $a0, $a0, $a1 ; LA64D-NEXT: ret entry: %c = fcmp uno float %a, 0.0 @@ -1081,11 +933,8 @@ define i64 @fptosi_nanguard_f64_to_i64(double %a) nounwind { ; ; LA64D-LABEL: fptosi_nanguard_f64_to_i64: ; LA64D: # %bb.0: # %entry -; LA64D-NEXT: fcmp.cun.d $fcc0, $fa0, $fa0 ; LA64D-NEXT: ftintrz.l.d $fa0, $fa0 ; LA64D-NEXT: movfr2gr.d $a0, $fa0 -; LA64D-NEXT: movcf2gr $a1, $fcc0 -; LA64D-NEXT: masknez $a0, $a0, $a1 ; LA64D-NEXT: ret entry: %c = fcmp uno double %a, 0.0 diff --git a/llvm/test/CodeGen/LoongArch/lasx/fptosi-sat.ll b/llvm/test/CodeGen/LoongArch/lasx/fptosi-sat.ll index 204ec3a9d4e10c..d5266e4c07981d 100644 --- a/llvm/test/CodeGen/LoongArch/lasx/fptosi-sat.ll +++ b/llvm/test/CodeGen/LoongArch/lasx/fptosi-sat.ll @@ -1,309 +1,14 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc --mtriple=loongarch32 --mattr=+32s,+lasx < %s | FileCheck %s --check-prefixes=CHECK,LA32 -; RUN: llc --mtriple=loongarch64 --mattr=+lasx < %s | FileCheck %s --check-prefixes=CHECK,LA64 +; RUN: llc --mtriple=loongarch32 --mattr=+32s,+lasx < %s | FileCheck %s +; RUN: llc --mtriple=loongarch64 --mattr=+lasx < %s | FileCheck %s define void @fptosi_sat_v8f32_to_v8i32(ptr %res, ptr %a) nounwind { -; LA32-LABEL: fptosi_sat_v8f32_to_v8i32: -; LA32: # %bb.0: # %entry -; LA32-NEXT: xvld $xr0, $a1, 0 -; LA32-NEXT: xvpickve.w $xr3, $xr0, 5 -; LA32-NEXT: lu12i.w $a1, -200704 -; LA32-NEXT: movgr2fr.w $fa1, $a1 -; LA32-NEXT: fcmp.cult.s $fcc0, $fa3, $fa1 -; LA32-NEXT: lu12i.w $a1, -524288 -; LA32-NEXT: movcf2gr $a2, $fcc0 -; LA32-NEXT: maskeqz $a3, $a1, $a2 -; LA32-NEXT: ftintrz.w.s $fa2, $fa3 -; LA32-NEXT: movfr2gr.s $a4, $fa2 -; LA32-NEXT: masknez $a2, $a4, $a2 -; LA32-NEXT: lu12i.w $a4, 323583 -; LA32-NEXT: ori $a4, $a4, 4095 -; LA32-NEXT: movgr2fr.w $fa2, $a4 -; LA32-NEXT: fcmp.clt.s $fcc0, $fa2, $fa3 -; LA32-NEXT: or $a2, $a3, $a2 -; LA32-NEXT: movcf2gr $a3, $fcc0 -; LA32-NEXT: masknez $a4, $a2, $a3 -; LA32-NEXT: lu12i.w $a2, 524287 -; LA32-NEXT: ori $a2, $a2, 4095 -; LA32-NEXT: maskeqz $a3, $a2, $a3 -; LA32-NEXT: fcmp.cun.s $fcc0, $fa3, $fa3 -; LA32-NEXT: or $a3, $a3, $a4 -; LA32-NEXT: movcf2gr $a4, $fcc0 -; LA32-NEXT: xvpickve.w $xr3, $xr0, 4 -; LA32-NEXT: fcmp.cult.s $fcc0, $fa3, $fa1 -; LA32-NEXT: masknez $a3, $a3, $a4 -; LA32-NEXT: movcf2gr $a4, $fcc0 -; LA32-NEXT: maskeqz $a5, $a1, $a4 -; LA32-NEXT: ftintrz.w.s $fa4, $fa3 -; LA32-NEXT: movfr2gr.s $a6, $fa4 -; LA32-NEXT: masknez $a4, $a6, $a4 -; LA32-NEXT: fcmp.clt.s $fcc0, $fa2, $fa3 -; LA32-NEXT: or $a4, $a5, $a4 -; LA32-NEXT: movcf2gr $a5, $fcc0 -; LA32-NEXT: masknez $a4, $a4, $a5 -; LA32-NEXT: maskeqz $a5, $a2, $a5 -; LA32-NEXT: fcmp.cun.s $fcc0, $fa3, $fa3 -; LA32-NEXT: or $a4, $a5, $a4 -; LA32-NEXT: movcf2gr $a5, $fcc0 -; LA32-NEXT: masknez $a4, $a4, $a5 -; LA32-NEXT: vinsgr2vr.w $vr3, $a4, 0 -; LA32-NEXT: xvpickve.w $xr4, $xr0, 6 -; LA32-NEXT: fcmp.cult.s $fcc0, $fa4, $fa1 -; LA32-NEXT: vinsgr2vr.w $vr3, $a3, 1 -; LA32-NEXT: movcf2gr $a3, $fcc0 -; LA32-NEXT: maskeqz $a4, $a1, $a3 -; LA32-NEXT: ftintrz.w.s $fa5, $fa4 -; LA32-NEXT: movfr2gr.s $a5, $fa5 -; LA32-NEXT: masknez $a3, $a5, $a3 -; LA32-NEXT: fcmp.clt.s $fcc0, $fa2, $fa4 -; LA32-NEXT: or $a3, $a4, $a3 -; LA32-NEXT: movcf2gr $a4, $fcc0 -; LA32-NEXT: masknez $a3, $a3, $a4 -; LA32-NEXT: maskeqz $a4, $a2, $a4 -; LA32-NEXT: fcmp.cun.s $fcc0, $fa4, $fa4 -; LA32-NEXT: or $a3, $a4, $a3 -; LA32-NEXT: movcf2gr $a4, $fcc0 -; LA32-NEXT: masknez $a3, $a3, $a4 -; LA32-NEXT: xvpickve.w $xr4, $xr0, 7 -; LA32-NEXT: fcmp.cult.s $fcc0, $fa4, $fa1 -; LA32-NEXT: vinsgr2vr.w $vr3, $a3, 2 -; LA32-NEXT: movcf2gr $a3, $fcc0 -; LA32-NEXT: maskeqz $a4, $a1, $a3 -; LA32-NEXT: ftintrz.w.s $fa5, $fa4 -; LA32-NEXT: movfr2gr.s $a5, $fa5 -; LA32-NEXT: masknez $a3, $a5, $a3 -; LA32-NEXT: fcmp.clt.s $fcc0, $fa2, $fa4 -; LA32-NEXT: or $a3, $a4, $a3 -; LA32-NEXT: movcf2gr $a4, $fcc0 -; LA32-NEXT: masknez $a3, $a3, $a4 -; LA32-NEXT: maskeqz $a4, $a2, $a4 -; LA32-NEXT: fcmp.cun.s $fcc0, $fa4, $fa4 -; LA32-NEXT: or $a3, $a4, $a3 -; LA32-NEXT: movcf2gr $a4, $fcc0 -; LA32-NEXT: masknez $a3, $a3, $a4 -; LA32-NEXT: xvpickve.w $xr4, $xr0, 1 -; LA32-NEXT: fcmp.cult.s $fcc0, $fa4, $fa1 -; LA32-NEXT: vinsgr2vr.w $vr3, $a3, 3 -; LA32-NEXT: movcf2gr $a3, $fcc0 -; LA32-NEXT: maskeqz $a4, $a1, $a3 -; LA32-NEXT: ftintrz.w.s $fa5, $fa4 -; LA32-NEXT: movfr2gr.s $a5, $fa5 -; LA32-NEXT: masknez $a3, $a5, $a3 -; LA32-NEXT: fcmp.clt.s $fcc0, $fa2, $fa4 -; LA32-NEXT: or $a3, $a4, $a3 -; LA32-NEXT: movcf2gr $a4, $fcc0 -; LA32-NEXT: masknez $a3, $a3, $a4 -; LA32-NEXT: maskeqz $a4, $a2, $a4 -; LA32-NEXT: fcmp.cun.s $fcc0, $fa4, $fa4 -; LA32-NEXT: or $a3, $a4, $a3 -; LA32-NEXT: movcf2gr $a4, $fcc0 -; LA32-NEXT: xvpickve.w $xr4, $xr0, 0 -; LA32-NEXT: fcmp.cult.s $fcc0, $fa4, $fa1 -; LA32-NEXT: masknez $a3, $a3, $a4 -; LA32-NEXT: movcf2gr $a4, $fcc0 -; LA32-NEXT: maskeqz $a5, $a1, $a4 -; LA32-NEXT: ftintrz.w.s $fa5, $fa4 -; LA32-NEXT: movfr2gr.s $a6, $fa5 -; LA32-NEXT: masknez $a4, $a6, $a4 -; LA32-NEXT: fcmp.clt.s $fcc0, $fa2, $fa4 -; LA32-NEXT: or $a4, $a5, $a4 -; LA32-NEXT: movcf2gr $a5, $fcc0 -; LA32-NEXT: masknez $a4, $a4, $a5 -; LA32-NEXT: maskeqz $a5, $a2, $a5 -; LA32-NEXT: fcmp.cun.s $fcc0, $fa4, $fa4 -; LA32-NEXT: or $a4, $a5, $a4 -; LA32-NEXT: movcf2gr $a5, $fcc0 -; LA32-NEXT: masknez $a4, $a4, $a5 -; LA32-NEXT: vinsgr2vr.w $vr4, $a4, 0 -; LA32-NEXT: xvpickve.w $xr5, $xr0, 2 -; LA32-NEXT: fcmp.cult.s $fcc0, $fa5, $fa1 -; LA32-NEXT: vinsgr2vr.w $vr4, $a3, 1 -; LA32-NEXT: movcf2gr $a3, $fcc0 -; LA32-NEXT: maskeqz $a4, $a1, $a3 -; LA32-NEXT: ftintrz.w.s $fa6, $fa5 -; LA32-NEXT: movfr2gr.s $a5, $fa6 -; LA32-NEXT: masknez $a3, $a5, $a3 -; LA32-NEXT: fcmp.clt.s $fcc0, $fa2, $fa5 -; LA32-NEXT: or $a3, $a4, $a3 -; LA32-NEXT: movcf2gr $a4, $fcc0 -; LA32-NEXT: masknez $a3, $a3, $a4 -; LA32-NEXT: maskeqz $a4, $a2, $a4 -; LA32-NEXT: fcmp.cun.s $fcc0, $fa5, $fa5 -; LA32-NEXT: or $a3, $a4, $a3 -; LA32-NEXT: movcf2gr $a4, $fcc0 -; LA32-NEXT: masknez $a3, $a3, $a4 -; LA32-NEXT: xvpickve.w $xr0, $xr0, 3 -; LA32-NEXT: fcmp.cult.s $fcc0, $fa0, $fa1 -; LA32-NEXT: vinsgr2vr.w $vr4, $a3, 2 -; LA32-NEXT: movcf2gr $a3, $fcc0 -; LA32-NEXT: maskeqz $a1, $a1, $a3 -; LA32-NEXT: ftintrz.w.s $fa1, $fa0 -; LA32-NEXT: movfr2gr.s $a4, $fa1 -; LA32-NEXT: masknez $a3, $a4, $a3 -; LA32-NEXT: fcmp.clt.s $fcc0, $fa2, $fa0 -; LA32-NEXT: or $a1, $a1, $a3 -; LA32-NEXT: movcf2gr $a3, $fcc0 -; LA32-NEXT: masknez $a1, $a1, $a3 -; LA32-NEXT: maskeqz $a2, $a2, $a3 -; LA32-NEXT: fcmp.cun.s $fcc0, $fa0, $fa0 -; LA32-NEXT: or $a1, $a2, $a1 -; LA32-NEXT: movcf2gr $a2, $fcc0 -; LA32-NEXT: masknez $a1, $a1, $a2 -; LA32-NEXT: vinsgr2vr.w $vr4, $a1, 3 -; LA32-NEXT: xvpermi.q $xr4, $xr3, 2 -; LA32-NEXT: xvst $xr4, $a0, 0 -; LA32-NEXT: ret -; -; LA64-LABEL: fptosi_sat_v8f32_to_v8i32: -; LA64: # %bb.0: # %entry -; LA64-NEXT: xvld $xr0, $a1, 0 -; LA64-NEXT: xvpickve.w $xr3, $xr0, 5 -; LA64-NEXT: lu12i.w $a1, -200704 -; LA64-NEXT: movgr2fr.w $fa1, $a1 -; LA64-NEXT: fcmp.cult.s $fcc0, $fa3, $fa1 -; LA64-NEXT: lu12i.w $a1, -524288 -; LA64-NEXT: movcf2gr $a2, $fcc0 -; LA64-NEXT: maskeqz $a3, $a1, $a2 -; LA64-NEXT: ftintrz.l.s $fa2, $fa3 -; LA64-NEXT: movfr2gr.d $a4, $fa2 -; LA64-NEXT: masknez $a2, $a4, $a2 -; LA64-NEXT: lu12i.w $a4, 323583 -; LA64-NEXT: ori $a4, $a4, 4095 -; LA64-NEXT: movgr2fr.w $fa2, $a4 -; LA64-NEXT: fcmp.clt.s $fcc0, $fa2, $fa3 -; LA64-NEXT: or $a2, $a3, $a2 -; LA64-NEXT: movcf2gr $a3, $fcc0 -; LA64-NEXT: masknez $a4, $a2, $a3 -; LA64-NEXT: lu12i.w $a2, 524287 -; LA64-NEXT: ori $a2, $a2, 4095 -; LA64-NEXT: maskeqz $a3, $a2, $a3 -; LA64-NEXT: fcmp.cun.s $fcc0, $fa3, $fa3 -; LA64-NEXT: or $a3, $a3, $a4 -; LA64-NEXT: movcf2gr $a4, $fcc0 -; LA64-NEXT: xvpickve.w $xr3, $xr0, 4 -; LA64-NEXT: fcmp.cult.s $fcc0, $fa3, $fa1 -; LA64-NEXT: masknez $a3, $a3, $a4 -; LA64-NEXT: movcf2gr $a4, $fcc0 -; LA64-NEXT: maskeqz $a5, $a1, $a4 -; LA64-NEXT: ftintrz.l.s $fa4, $fa3 -; LA64-NEXT: movfr2gr.d $a6, $fa4 -; LA64-NEXT: masknez $a4, $a6, $a4 -; LA64-NEXT: fcmp.clt.s $fcc0, $fa2, $fa3 -; LA64-NEXT: or $a4, $a5, $a4 -; LA64-NEXT: movcf2gr $a5, $fcc0 -; LA64-NEXT: masknez $a4, $a4, $a5 -; LA64-NEXT: maskeqz $a5, $a2, $a5 -; LA64-NEXT: fcmp.cun.s $fcc0, $fa3, $fa3 -; LA64-NEXT: or $a4, $a5, $a4 -; LA64-NEXT: movcf2gr $a5, $fcc0 -; LA64-NEXT: masknez $a4, $a4, $a5 -; LA64-NEXT: vinsgr2vr.w $vr3, $a4, 0 -; LA64-NEXT: xvpickve.w $xr4, $xr0, 6 -; LA64-NEXT: fcmp.cult.s $fcc0, $fa4, $fa1 -; LA64-NEXT: vinsgr2vr.w $vr3, $a3, 1 -; LA64-NEXT: movcf2gr $a3, $fcc0 -; LA64-NEXT: maskeqz $a4, $a1, $a3 -; LA64-NEXT: ftintrz.l.s $fa5, $fa4 -; LA64-NEXT: movfr2gr.d $a5, $fa5 -; LA64-NEXT: masknez $a3, $a5, $a3 -; LA64-NEXT: fcmp.clt.s $fcc0, $fa2, $fa4 -; LA64-NEXT: or $a3, $a4, $a3 -; LA64-NEXT: movcf2gr $a4, $fcc0 -; LA64-NEXT: masknez $a3, $a3, $a4 -; LA64-NEXT: maskeqz $a4, $a2, $a4 -; LA64-NEXT: fcmp.cun.s $fcc0, $fa4, $fa4 -; LA64-NEXT: or $a3, $a4, $a3 -; LA64-NEXT: movcf2gr $a4, $fcc0 -; LA64-NEXT: masknez $a3, $a3, $a4 -; LA64-NEXT: xvpickve.w $xr4, $xr0, 7 -; LA64-NEXT: fcmp.cult.s $fcc0, $fa4, $fa1 -; LA64-NEXT: vinsgr2vr.w $vr3, $a3, 2 -; LA64-NEXT: movcf2gr $a3, $fcc0 -; LA64-NEXT: maskeqz $a4, $a1, $a3 -; LA64-NEXT: ftintrz.l.s $fa5, $fa4 -; LA64-NEXT: movfr2gr.d $a5, $fa5 -; LA64-NEXT: masknez $a3, $a5, $a3 -; LA64-NEXT: fcmp.clt.s $fcc0, $fa2, $fa4 -; LA64-NEXT: or $a3, $a4, $a3 -; LA64-NEXT: movcf2gr $a4, $fcc0 -; LA64-NEXT: masknez $a3, $a3, $a4 -; LA64-NEXT: maskeqz $a4, $a2, $a4 -; LA64-NEXT: fcmp.cun.s $fcc0, $fa4, $fa4 -; LA64-NEXT: or $a3, $a4, $a3 -; LA64-NEXT: movcf2gr $a4, $fcc0 -; LA64-NEXT: masknez $a3, $a3, $a4 -; LA64-NEXT: xvpickve.w $xr4, $xr0, 1 -; LA64-NEXT: fcmp.cult.s $fcc0, $fa4, $fa1 -; LA64-NEXT: vinsgr2vr.w $vr3, $a3, 3 -; LA64-NEXT: movcf2gr $a3, $fcc0 -; LA64-NEXT: maskeqz $a4, $a1, $a3 -; LA64-NEXT: ftintrz.l.s $fa5, $fa4 -; LA64-NEXT: movfr2gr.d $a5, $fa5 -; LA64-NEXT: masknez $a3, $a5, $a3 -; LA64-NEXT: fcmp.clt.s $fcc0, $fa2, $fa4 -; LA64-NEXT: or $a3, $a4, $a3 -; LA64-NEXT: movcf2gr $a4, $fcc0 -; LA64-NEXT: masknez $a3, $a3, $a4 -; LA64-NEXT: maskeqz $a4, $a2, $a4 -; LA64-NEXT: fcmp.cun.s $fcc0, $fa4, $fa4 -; LA64-NEXT: or $a3, $a4, $a3 -; LA64-NEXT: movcf2gr $a4, $fcc0 -; LA64-NEXT: xvpickve.w $xr4, $xr0, 0 -; LA64-NEXT: fcmp.cult.s $fcc0, $fa4, $fa1 -; LA64-NEXT: masknez $a3, $a3, $a4 -; LA64-NEXT: movcf2gr $a4, $fcc0 -; LA64-NEXT: maskeqz $a5, $a1, $a4 -; LA64-NEXT: ftintrz.l.s $fa5, $fa4 -; LA64-NEXT: movfr2gr.d $a6, $fa5 -; LA64-NEXT: masknez $a4, $a6, $a4 -; LA64-NEXT: fcmp.clt.s $fcc0, $fa2, $fa4 -; LA64-NEXT: or $a4, $a5, $a4 -; LA64-NEXT: movcf2gr $a5, $fcc0 -; LA64-NEXT: masknez $a4, $a4, $a5 -; LA64-NEXT: maskeqz $a5, $a2, $a5 -; LA64-NEXT: fcmp.cun.s $fcc0, $fa4, $fa4 -; LA64-NEXT: or $a4, $a5, $a4 -; LA64-NEXT: movcf2gr $a5, $fcc0 -; LA64-NEXT: masknez $a4, $a4, $a5 -; LA64-NEXT: vinsgr2vr.w $vr4, $a4, 0 -; LA64-NEXT: xvpickve.w $xr5, $xr0, 2 -; LA64-NEXT: fcmp.cult.s $fcc0, $fa5, $fa1 -; LA64-NEXT: vinsgr2vr.w $vr4, $a3, 1 -; LA64-NEXT: movcf2gr $a3, $fcc0 -; LA64-NEXT: maskeqz $a4, $a1, $a3 -; LA64-NEXT: ftintrz.l.s $fa6, $fa5 -; LA64-NEXT: movfr2gr.d $a5, $fa6 -; LA64-NEXT: masknez $a3, $a5, $a3 -; LA64-NEXT: fcmp.clt.s $fcc0, $fa2, $fa5 -; LA64-NEXT: or $a3, $a4, $a3 -; LA64-NEXT: movcf2gr $a4, $fcc0 -; LA64-NEXT: masknez $a3, $a3, $a4 -; LA64-NEXT: maskeqz $a4, $a2, $a4 -; LA64-NEXT: fcmp.cun.s $fcc0, $fa5, $fa5 -; LA64-NEXT: or $a3, $a4, $a3 -; LA64-NEXT: movcf2gr $a4, $fcc0 -; LA64-NEXT: masknez $a3, $a3, $a4 -; LA64-NEXT: xvpickve.w $xr0, $xr0, 3 -; LA64-NEXT: fcmp.cult.s $fcc0, $fa0, $fa1 -; LA64-NEXT: vinsgr2vr.w $vr4, $a3, 2 -; LA64-NEXT: movcf2gr $a3, $fcc0 -; LA64-NEXT: maskeqz $a1, $a1, $a3 -; LA64-NEXT: ftintrz.l.s $fa1, $fa0 -; LA64-NEXT: movfr2gr.d $a4, $fa1 -; LA64-NEXT: masknez $a3, $a4, $a3 -; LA64-NEXT: fcmp.clt.s $fcc0, $fa2, $fa0 -; LA64-NEXT: or $a1, $a1, $a3 -; LA64-NEXT: movcf2gr $a3, $fcc0 -; LA64-NEXT: masknez $a1, $a1, $a3 -; LA64-NEXT: maskeqz $a2, $a2, $a3 -; LA64-NEXT: fcmp.cun.s $fcc0, $fa0, $fa0 -; LA64-NEXT: or $a1, $a2, $a1 -; LA64-NEXT: movcf2gr $a2, $fcc0 -; LA64-NEXT: masknez $a1, $a1, $a2 -; LA64-NEXT: vinsgr2vr.w $vr4, $a1, 3 -; LA64-NEXT: xvpermi.q $xr4, $xr3, 2 -; LA64-NEXT: xvst $xr4, $a0, 0 -; LA64-NEXT: ret +; CHECK-LABEL: fptosi_sat_v8f32_to_v8i32: +; CHECK: # %bb.0: # %entry +; CHECK-NEXT: xvld $xr0, $a1, 0 +; CHECK-NEXT: xvftintrz.w.s $xr0, $xr0 +; CHECK-NEXT: xvst $xr0, $a0, 0 +; CHECK-NEXT: ret entry: %v = load <8 x float>, ptr %a %r = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> %v) @@ -312,254 +17,12 @@ entry: } define void @fptosi_sat_v4f64_to_v4i64(ptr %res, ptr %a) nounwind { -; LA32-LABEL: fptosi_sat_v4f64_to_v4i64: -; LA32: # %bb.0: # %entry -; LA32-NEXT: addi.w $sp, $sp, -160 -; LA32-NEXT: st.w $ra, $sp, 156 # 4-byte Folded Spill -; LA32-NEXT: st.w $fp, $sp, 152 # 4-byte Folded Spill -; LA32-NEXT: st.w $s0, $sp, 148 # 4-byte Folded Spill -; LA32-NEXT: st.w $s1, $sp, 144 # 4-byte Folded Spill -; LA32-NEXT: st.w $s2, $sp, 140 # 4-byte Folded Spill -; LA32-NEXT: fst.d $fs0, $sp, 128 # 8-byte Folded Spill -; LA32-NEXT: fst.d $fs1, $sp, 120 # 8-byte Folded Spill -; LA32-NEXT: xvld $xr0, $a1, 0 -; LA32-NEXT: xvst $xr0, $sp, 80 # 32-byte Folded Spill -; LA32-NEXT: move $fp, $a0 -; LA32-NEXT: xvpickve.d $xr0, $xr0, 2 -; LA32-NEXT: xvst $xr0, $sp, 48 # 32-byte Folded Spill -; LA32-NEXT: movgr2fr.w $fs0, $zero -; LA32-NEXT: lu12i.w $a0, -246272 -; LA32-NEXT: movgr2frh.w $fs0, $a0 -; LA32-NEXT: fcmp.cult.d $fcc0, $fa0, $fs0 -; LA32-NEXT: movcf2gr $a0, $fcc0 -; LA32-NEXT: st.w $a0, $sp, 16 -; LA32-NEXT: # kill: def $f0_64 killed $f0_64 killed $xr0 -; LA32-NEXT: bl __fixdfdi -; LA32-NEXT: ld.w $a2, $sp, 16 -; LA32-NEXT: movgr2cf $fcc0, $a2 -; LA32-NEXT: movcf2gr $a2, $fcc0 -; LA32-NEXT: pcalau12i $a3, %pc_hi20(.LCPI1_0) -; LA32-NEXT: fld.d $fs1, $a3, %pc_lo12(.LCPI1_0) -; LA32-NEXT: masknez $a1, $a1, $a2 -; LA32-NEXT: lu12i.w $s0, -524288 -; LA32-NEXT: maskeqz $a3, $s0, $a2 -; LA32-NEXT: xvld $xr0, $sp, 48 # 32-byte Folded Reload -; LA32-NEXT: fcmp.clt.d $fcc0, $fs1, $fa0 -; LA32-NEXT: or $a1, $a3, $a1 -; LA32-NEXT: movcf2gr $a3, $fcc0 -; LA32-NEXT: masknez $a1, $a1, $a3 -; LA32-NEXT: lu12i.w $a4, 524287 -; LA32-NEXT: ori $s1, $a4, 4095 -; LA32-NEXT: maskeqz $a4, $s1, $a3 -; LA32-NEXT: fcmp.cun.d $fcc0, $fa0, $fa0 -; LA32-NEXT: or $a1, $a4, $a1 -; LA32-NEXT: movcf2gr $a4, $fcc0 -; LA32-NEXT: masknez $a1, $a1, $a4 -; LA32-NEXT: masknez $a0, $a0, $a2 -; LA32-NEXT: masknez $a0, $a0, $a3 -; LA32-NEXT: addi.w $s2, $zero, -1 -; LA32-NEXT: maskeqz $a2, $s2, $a3 -; LA32-NEXT: or $a0, $a2, $a0 -; LA32-NEXT: masknez $a0, $a0, $a4 -; LA32-NEXT: vinsgr2vr.w $vr0, $a0, 0 -; LA32-NEXT: vinsgr2vr.w $vr0, $a1, 1 -; LA32-NEXT: xvst $xr0, $sp, 48 # 32-byte Folded Spill -; LA32-NEXT: xvld $xr0, $sp, 80 # 32-byte Folded Reload -; LA32-NEXT: xvpickve.d $xr0, $xr0, 3 -; LA32-NEXT: xvst $xr0, $sp, 16 # 32-byte Folded Spill -; LA32-NEXT: fcmp.clt.d $fcc0, $fs1, $fa0 -; LA32-NEXT: movcf2gr $a0, $fcc0 -; LA32-NEXT: st.w $a0, $sp, 12 -; LA32-NEXT: fcmp.cult.d $fcc0, $fa0, $fs0 -; LA32-NEXT: movcf2gr $a0, $fcc0 -; LA32-NEXT: st.w $a0, $sp, 8 -; LA32-NEXT: # kill: def $f0_64 killed $f0_64 killed $xr0 -; LA32-NEXT: bl __fixdfdi -; LA32-NEXT: ld.w $a2, $sp, 8 -; LA32-NEXT: movgr2cf $fcc0, $a2 -; LA32-NEXT: movcf2gr $a2, $fcc0 -; LA32-NEXT: masknez $a0, $a0, $a2 -; LA32-NEXT: ld.w $a3, $sp, 12 -; LA32-NEXT: movgr2cf $fcc0, $a3 -; LA32-NEXT: movcf2gr $a3, $fcc0 -; LA32-NEXT: masknez $a0, $a0, $a3 -; LA32-NEXT: maskeqz $a4, $s2, $a3 -; LA32-NEXT: xvld $xr0, $sp, 16 # 32-byte Folded Reload -; LA32-NEXT: fcmp.cun.d $fcc0, $fa0, $fa0 -; LA32-NEXT: or $a0, $a4, $a0 -; LA32-NEXT: movcf2gr $a4, $fcc0 -; LA32-NEXT: masknez $a0, $a0, $a4 -; LA32-NEXT: xvld $xr0, $sp, 48 # 32-byte Folded Reload -; LA32-NEXT: vinsgr2vr.w $vr0, $a0, 2 -; LA32-NEXT: masknez $a0, $a1, $a2 -; LA32-NEXT: maskeqz $a1, $s0, $a2 -; LA32-NEXT: or $a0, $a1, $a0 -; LA32-NEXT: masknez $a0, $a0, $a3 -; LA32-NEXT: maskeqz $a1, $s1, $a3 -; LA32-NEXT: or $a0, $a1, $a0 -; LA32-NEXT: masknez $a0, $a0, $a4 -; LA32-NEXT: vinsgr2vr.w $vr0, $a0, 3 -; LA32-NEXT: xvst $xr0, $sp, 48 # 32-byte Folded Spill -; LA32-NEXT: xvld $xr0, $sp, 80 # 32-byte Folded Reload -; LA32-NEXT: xvpickve.d $xr0, $xr0, 0 -; LA32-NEXT: xvst $xr0, $sp, 16 # 32-byte Folded Spill -; LA32-NEXT: fcmp.cult.d $fcc0, $fa0, $fs0 -; LA32-NEXT: movcf2gr $a0, $fcc0 -; LA32-NEXT: st.w $a0, $sp, 12 -; LA32-NEXT: # kill: def $f0_64 killed $f0_64 killed $xr0 -; LA32-NEXT: bl __fixdfdi -; LA32-NEXT: ld.w $a2, $sp, 12 -; LA32-NEXT: movgr2cf $fcc0, $a2 -; LA32-NEXT: movcf2gr $a2, $fcc0 -; LA32-NEXT: masknez $a1, $a1, $a2 -; LA32-NEXT: maskeqz $a3, $s0, $a2 -; LA32-NEXT: xvld $xr0, $sp, 16 # 32-byte Folded Reload -; LA32-NEXT: fcmp.clt.d $fcc0, $fs1, $fa0 -; LA32-NEXT: or $a1, $a3, $a1 -; LA32-NEXT: movcf2gr $a3, $fcc0 -; LA32-NEXT: masknez $a1, $a1, $a3 -; LA32-NEXT: maskeqz $a4, $s1, $a3 -; LA32-NEXT: fcmp.cun.d $fcc0, $fa0, $fa0 -; LA32-NEXT: or $a1, $a4, $a1 -; LA32-NEXT: movcf2gr $a4, $fcc0 -; LA32-NEXT: masknez $a1, $a1, $a4 -; LA32-NEXT: masknez $a0, $a0, $a2 -; LA32-NEXT: masknez $a0, $a0, $a3 -; LA32-NEXT: maskeqz $a2, $s2, $a3 -; LA32-NEXT: or $a0, $a2, $a0 -; LA32-NEXT: masknez $a0, $a0, $a4 -; LA32-NEXT: vinsgr2vr.w $vr0, $a0, 0 -; LA32-NEXT: vinsgr2vr.w $vr0, $a1, 1 -; LA32-NEXT: xvst $xr0, $sp, 16 # 32-byte Folded Spill -; LA32-NEXT: xvld $xr0, $sp, 80 # 32-byte Folded Reload -; LA32-NEXT: xvpickve.d $xr0, $xr0, 1 -; LA32-NEXT: xvst $xr0, $sp, 80 # 32-byte Folded Spill -; LA32-NEXT: fcmp.clt.d $fcc0, $fs1, $fa0 -; LA32-NEXT: movcf2gr $a0, $fcc0 -; LA32-NEXT: st.w $a0, $sp, 12 -; LA32-NEXT: fcmp.cult.d $fcc0, $fa0, $fs0 -; LA32-NEXT: movcf2gr $a0, $fcc0 -; LA32-NEXT: st.w $a0, $sp, 8 -; LA32-NEXT: # kill: def $f0_64 killed $f0_64 killed $xr0 -; LA32-NEXT: bl __fixdfdi -; LA32-NEXT: ld.w $a2, $sp, 8 -; LA32-NEXT: movgr2cf $fcc0, $a2 -; LA32-NEXT: movcf2gr $a2, $fcc0 -; LA32-NEXT: masknez $a0, $a0, $a2 -; LA32-NEXT: ld.w $a3, $sp, 12 -; LA32-NEXT: movgr2cf $fcc0, $a3 -; LA32-NEXT: movcf2gr $a3, $fcc0 -; LA32-NEXT: masknez $a0, $a0, $a3 -; LA32-NEXT: maskeqz $a4, $s2, $a3 -; LA32-NEXT: xvld $xr0, $sp, 80 # 32-byte Folded Reload -; LA32-NEXT: fcmp.cun.d $fcc0, $fa0, $fa0 -; LA32-NEXT: or $a0, $a4, $a0 -; LA32-NEXT: movcf2gr $a4, $fcc0 -; LA32-NEXT: masknez $a0, $a0, $a4 -; LA32-NEXT: xvld $xr1, $sp, 16 # 32-byte Folded Reload -; LA32-NEXT: vinsgr2vr.w $vr1, $a0, 2 -; LA32-NEXT: masknez $a0, $a1, $a2 -; LA32-NEXT: maskeqz $a1, $s0, $a2 -; LA32-NEXT: or $a0, $a1, $a0 -; LA32-NEXT: masknez $a0, $a0, $a3 -; LA32-NEXT: maskeqz $a1, $s1, $a3 -; LA32-NEXT: or $a0, $a1, $a0 -; LA32-NEXT: masknez $a0, $a0, $a4 -; LA32-NEXT: vinsgr2vr.w $vr1, $a0, 3 -; LA32-NEXT: xvld $xr0, $sp, 48 # 32-byte Folded Reload -; LA32-NEXT: xvpermi.q $xr1, $xr0, 2 -; LA32-NEXT: xvst $xr1, $fp, 0 -; LA32-NEXT: fld.d $fs1, $sp, 120 # 8-byte Folded Reload -; LA32-NEXT: fld.d $fs0, $sp, 128 # 8-byte Folded Reload -; LA32-NEXT: ld.w $s2, $sp, 140 # 4-byte Folded Reload -; LA32-NEXT: ld.w $s1, $sp, 144 # 4-byte Folded Reload -; LA32-NEXT: ld.w $s0, $sp, 148 # 4-byte Folded Reload -; LA32-NEXT: ld.w $fp, $sp, 152 # 4-byte Folded Reload -; LA32-NEXT: ld.w $ra, $sp, 156 # 4-byte Folded Reload -; LA32-NEXT: addi.w $sp, $sp, 160 -; LA32-NEXT: ret -; -; LA64-LABEL: fptosi_sat_v4f64_to_v4i64: -; LA64: # %bb.0: # %entry -; LA64-NEXT: xvld $xr0, $a1, 0 -; LA64-NEXT: xvpickve.d $xr3, $xr0, 3 -; LA64-NEXT: lu52i.d $a1, $zero, -962 -; LA64-NEXT: movgr2fr.d $fa1, $a1 -; LA64-NEXT: fcmp.cult.d $fcc0, $fa3, $fa1 -; LA64-NEXT: lu52i.d $a1, $zero, -2048 -; LA64-NEXT: movcf2gr $a2, $fcc0 -; LA64-NEXT: maskeqz $a3, $a1, $a2 -; LA64-NEXT: ftintrz.l.d $fa2, $fa3 -; LA64-NEXT: movfr2gr.d $a4, $fa2 -; LA64-NEXT: masknez $a2, $a4, $a2 -; LA64-NEXT: addi.w $a4, $zero, -1 -; LA64-NEXT: lu52i.d $a5, $a4, 1085 -; LA64-NEXT: movgr2fr.d $fa2, $a5 -; LA64-NEXT: fcmp.clt.d $fcc0, $fa2, $fa3 -; LA64-NEXT: or $a2, $a3, $a2 -; LA64-NEXT: movcf2gr $a3, $fcc0 -; LA64-NEXT: masknez $a5, $a2, $a3 -; LA64-NEXT: lu52i.d $a2, $a4, 2047 -; LA64-NEXT: maskeqz $a3, $a2, $a3 -; LA64-NEXT: fcmp.cun.d $fcc0, $fa3, $fa3 -; LA64-NEXT: or $a3, $a3, $a5 -; LA64-NEXT: movcf2gr $a4, $fcc0 -; LA64-NEXT: xvpickve.d $xr3, $xr0, 2 -; LA64-NEXT: fcmp.cult.d $fcc0, $fa3, $fa1 -; LA64-NEXT: masknez $a3, $a3, $a4 -; LA64-NEXT: movcf2gr $a4, $fcc0 -; LA64-NEXT: maskeqz $a5, $a1, $a4 -; LA64-NEXT: ftintrz.l.d $fa4, $fa3 -; LA64-NEXT: movfr2gr.d $a6, $fa4 -; LA64-NEXT: masknez $a4, $a6, $a4 -; LA64-NEXT: fcmp.clt.d $fcc0, $fa2, $fa3 -; LA64-NEXT: or $a4, $a5, $a4 -; LA64-NEXT: movcf2gr $a5, $fcc0 -; LA64-NEXT: masknez $a4, $a4, $a5 -; LA64-NEXT: maskeqz $a5, $a2, $a5 -; LA64-NEXT: fcmp.cun.d $fcc0, $fa3, $fa3 -; LA64-NEXT: or $a4, $a5, $a4 -; LA64-NEXT: movcf2gr $a5, $fcc0 -; LA64-NEXT: masknez $a4, $a4, $a5 -; LA64-NEXT: vinsgr2vr.d $vr3, $a4, 0 -; LA64-NEXT: xvpickve.d $xr4, $xr0, 1 -; LA64-NEXT: fcmp.cult.d $fcc0, $fa4, $fa1 -; LA64-NEXT: vinsgr2vr.d $vr3, $a3, 1 -; LA64-NEXT: movcf2gr $a3, $fcc0 -; LA64-NEXT: maskeqz $a4, $a1, $a3 -; LA64-NEXT: ftintrz.l.d $fa5, $fa4 -; LA64-NEXT: movfr2gr.d $a5, $fa5 -; LA64-NEXT: masknez $a3, $a5, $a3 -; LA64-NEXT: fcmp.clt.d $fcc0, $fa2, $fa4 -; LA64-NEXT: or $a3, $a4, $a3 -; LA64-NEXT: movcf2gr $a4, $fcc0 -; LA64-NEXT: masknez $a3, $a3, $a4 -; LA64-NEXT: maskeqz $a4, $a2, $a4 -; LA64-NEXT: fcmp.cun.d $fcc0, $fa4, $fa4 -; LA64-NEXT: or $a3, $a4, $a3 -; LA64-NEXT: movcf2gr $a4, $fcc0 -; LA64-NEXT: xvpickve.d $xr0, $xr0, 0 -; LA64-NEXT: fcmp.cult.d $fcc0, $fa0, $fa1 -; LA64-NEXT: masknez $a3, $a3, $a4 -; LA64-NEXT: movcf2gr $a4, $fcc0 -; LA64-NEXT: maskeqz $a1, $a1, $a4 -; LA64-NEXT: ftintrz.l.d $fa1, $fa0 -; LA64-NEXT: movfr2gr.d $a5, $fa1 -; LA64-NEXT: masknez $a4, $a5, $a4 -; LA64-NEXT: fcmp.clt.d $fcc0, $fa2, $fa0 -; LA64-NEXT: or $a1, $a1, $a4 -; LA64-NEXT: movcf2gr $a4, $fcc0 -; LA64-NEXT: masknez $a1, $a1, $a4 -; LA64-NEXT: maskeqz $a2, $a2, $a4 -; LA64-NEXT: fcmp.cun.d $fcc0, $fa0, $fa0 -; LA64-NEXT: or $a1, $a2, $a1 -; LA64-NEXT: movcf2gr $a2, $fcc0 -; LA64-NEXT: masknez $a1, $a1, $a2 -; LA64-NEXT: vinsgr2vr.d $vr0, $a1, 0 -; LA64-NEXT: vinsgr2vr.d $vr0, $a3, 1 -; LA64-NEXT: xvpermi.q $xr0, $xr3, 2 -; LA64-NEXT: xvst $xr0, $a0, 0 -; LA64-NEXT: ret +; CHECK-LABEL: fptosi_sat_v4f64_to_v4i64: +; CHECK: # %bb.0: # %entry +; CHECK-NEXT: xvld $xr0, $a1, 0 +; CHECK-NEXT: xvftintrz.l.d $xr0, $xr0 +; CHECK-NEXT: xvst $xr0, $a0, 0 +; CHECK-NEXT: ret entry: %v = load <4 x double>, ptr %a %r = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> %v) diff --git a/llvm/test/CodeGen/LoongArch/lsx/fptosi-sat.ll b/llvm/test/CodeGen/LoongArch/lsx/fptosi-sat.ll index d92465766a0025..712ded72aaf5a7 100644 --- a/llvm/test/CodeGen/LoongArch/lsx/fptosi-sat.ll +++ b/llvm/test/CodeGen/LoongArch/lsx/fptosi-sat.ll @@ -3,169 +3,12 @@ ; RUN: llc --mtriple=loongarch64 --mattr=+lsx < %s | FileCheck %s --check-prefixes=CHECK,LA64 define void @fptosi_sat_v4f32_to_v4i32(ptr %res, ptr %a) nounwind { -; LA32-LABEL: fptosi_sat_v4f32_to_v4i32: -; LA32: # %bb.0: # %entry -; LA32-NEXT: vld $vr0, $a1, 0 -; LA32-NEXT: vreplvei.w $vr3, $vr0, 1 -; LA32-NEXT: lu12i.w $a1, -200704 -; LA32-NEXT: movgr2fr.w $fa1, $a1 -; LA32-NEXT: fcmp.cult.s $fcc0, $fa3, $fa1 -; LA32-NEXT: lu12i.w $a1, -524288 -; LA32-NEXT: movcf2gr $a2, $fcc0 -; LA32-NEXT: maskeqz $a3, $a1, $a2 -; LA32-NEXT: ftintrz.w.s $fa2, $fa3 -; LA32-NEXT: movfr2gr.s $a4, $fa2 -; LA32-NEXT: masknez $a2, $a4, $a2 -; LA32-NEXT: lu12i.w $a4, 323583 -; LA32-NEXT: ori $a4, $a4, 4095 -; LA32-NEXT: movgr2fr.w $fa2, $a4 -; LA32-NEXT: fcmp.clt.s $fcc0, $fa2, $fa3 -; LA32-NEXT: or $a2, $a3, $a2 -; LA32-NEXT: movcf2gr $a3, $fcc0 -; LA32-NEXT: masknez $a4, $a2, $a3 -; LA32-NEXT: lu12i.w $a2, 524287 -; LA32-NEXT: ori $a2, $a2, 4095 -; LA32-NEXT: maskeqz $a3, $a2, $a3 -; LA32-NEXT: fcmp.cun.s $fcc0, $fa3, $fa3 -; LA32-NEXT: or $a3, $a3, $a4 -; LA32-NEXT: movcf2gr $a4, $fcc0 -; LA32-NEXT: vreplvei.w $vr3, $vr0, 0 -; LA32-NEXT: fcmp.cult.s $fcc0, $fa3, $fa1 -; LA32-NEXT: masknez $a3, $a3, $a4 -; LA32-NEXT: movcf2gr $a4, $fcc0 -; LA32-NEXT: maskeqz $a5, $a1, $a4 -; LA32-NEXT: ftintrz.w.s $fa4, $fa3 -; LA32-NEXT: movfr2gr.s $a6, $fa4 -; LA32-NEXT: masknez $a4, $a6, $a4 -; LA32-NEXT: fcmp.clt.s $fcc0, $fa2, $fa3 -; LA32-NEXT: or $a4, $a5, $a4 -; LA32-NEXT: movcf2gr $a5, $fcc0 -; LA32-NEXT: masknez $a4, $a4, $a5 -; LA32-NEXT: maskeqz $a5, $a2, $a5 -; LA32-NEXT: fcmp.cun.s $fcc0, $fa3, $fa3 -; LA32-NEXT: or $a4, $a5, $a4 -; LA32-NEXT: movcf2gr $a5, $fcc0 -; LA32-NEXT: masknez $a4, $a4, $a5 -; LA32-NEXT: vinsgr2vr.w $vr3, $a4, 0 -; LA32-NEXT: vreplvei.w $vr4, $vr0, 2 -; LA32-NEXT: fcmp.cult.s $fcc0, $fa4, $fa1 -; LA32-NEXT: vinsgr2vr.w $vr3, $a3, 1 -; LA32-NEXT: movcf2gr $a3, $fcc0 -; LA32-NEXT: maskeqz $a4, $a1, $a3 -; LA32-NEXT: ftintrz.w.s $fa5, $fa4 -; LA32-NEXT: movfr2gr.s $a5, $fa5 -; LA32-NEXT: masknez $a3, $a5, $a3 -; LA32-NEXT: fcmp.clt.s $fcc0, $fa2, $fa4 -; LA32-NEXT: or $a3, $a4, $a3 -; LA32-NEXT: movcf2gr $a4, $fcc0 -; LA32-NEXT: masknez $a3, $a3, $a4 -; LA32-NEXT: maskeqz $a4, $a2, $a4 -; LA32-NEXT: fcmp.cun.s $fcc0, $fa4, $fa4 -; LA32-NEXT: or $a3, $a4, $a3 -; LA32-NEXT: movcf2gr $a4, $fcc0 -; LA32-NEXT: masknez $a3, $a3, $a4 -; LA32-NEXT: vreplvei.w $vr0, $vr0, 3 -; LA32-NEXT: fcmp.cult.s $fcc0, $fa0, $fa1 -; LA32-NEXT: vinsgr2vr.w $vr3, $a3, 2 -; LA32-NEXT: movcf2gr $a3, $fcc0 -; LA32-NEXT: maskeqz $a1, $a1, $a3 -; LA32-NEXT: ftintrz.w.s $fa1, $fa0 -; LA32-NEXT: movfr2gr.s $a4, $fa1 -; LA32-NEXT: masknez $a3, $a4, $a3 -; LA32-NEXT: fcmp.clt.s $fcc0, $fa2, $fa0 -; LA32-NEXT: or $a1, $a1, $a3 -; LA32-NEXT: movcf2gr $a3, $fcc0 -; LA32-NEXT: masknez $a1, $a1, $a3 -; LA32-NEXT: maskeqz $a2, $a2, $a3 -; LA32-NEXT: fcmp.cun.s $fcc0, $fa0, $fa0 -; LA32-NEXT: or $a1, $a2, $a1 -; LA32-NEXT: movcf2gr $a2, $fcc0 -; LA32-NEXT: masknez $a1, $a1, $a2 -; LA32-NEXT: vinsgr2vr.w $vr3, $a1, 3 -; LA32-NEXT: vst $vr3, $a0, 0 -; LA32-NEXT: ret -; -; LA64-LABEL: fptosi_sat_v4f32_to_v4i32: -; LA64: # %bb.0: # %entry -; LA64-NEXT: vld $vr0, $a1, 0 -; LA64-NEXT: vreplvei.w $vr3, $vr0, 1 -; LA64-NEXT: lu12i.w $a1, -200704 -; LA64-NEXT: movgr2fr.w $fa1, $a1 -; LA64-NEXT: fcmp.cult.s $fcc0, $fa3, $fa1 -; LA64-NEXT: lu12i.w $a1, -524288 -; LA64-NEXT: movcf2gr $a2, $fcc0 -; LA64-NEXT: maskeqz $a3, $a1, $a2 -; LA64-NEXT: ftintrz.l.s $fa2, $fa3 -; LA64-NEXT: movfr2gr.d $a4, $fa2 -; LA64-NEXT: masknez $a2, $a4, $a2 -; LA64-NEXT: lu12i.w $a4, 323583 -; LA64-NEXT: ori $a4, $a4, 4095 -; LA64-NEXT: movgr2fr.w $fa2, $a4 -; LA64-NEXT: fcmp.clt.s $fcc0, $fa2, $fa3 -; LA64-NEXT: or $a2, $a3, $a2 -; LA64-NEXT: movcf2gr $a3, $fcc0 -; LA64-NEXT: masknez $a4, $a2, $a3 -; LA64-NEXT: lu12i.w $a2, 524287 -; LA64-NEXT: ori $a2, $a2, 4095 -; LA64-NEXT: maskeqz $a3, $a2, $a3 -; LA64-NEXT: fcmp.cun.s $fcc0, $fa3, $fa3 -; LA64-NEXT: or $a3, $a3, $a4 -; LA64-NEXT: movcf2gr $a4, $fcc0 -; LA64-NEXT: vreplvei.w $vr3, $vr0, 0 -; LA64-NEXT: fcmp.cult.s $fcc0, $fa3, $fa1 -; LA64-NEXT: masknez $a3, $a3, $a4 -; LA64-NEXT: movcf2gr $a4, $fcc0 -; LA64-NEXT: maskeqz $a5, $a1, $a4 -; LA64-NEXT: ftintrz.l.s $fa4, $fa3 -; LA64-NEXT: movfr2gr.d $a6, $fa4 -; LA64-NEXT: masknez $a4, $a6, $a4 -; LA64-NEXT: fcmp.clt.s $fcc0, $fa2, $fa3 -; LA64-NEXT: or $a4, $a5, $a4 -; LA64-NEXT: movcf2gr $a5, $fcc0 -; LA64-NEXT: masknez $a4, $a4, $a5 -; LA64-NEXT: maskeqz $a5, $a2, $a5 -; LA64-NEXT: fcmp.cun.s $fcc0, $fa3, $fa3 -; LA64-NEXT: or $a4, $a5, $a4 -; LA64-NEXT: movcf2gr $a5, $fcc0 -; LA64-NEXT: masknez $a4, $a4, $a5 -; LA64-NEXT: vinsgr2vr.w $vr3, $a4, 0 -; LA64-NEXT: vreplvei.w $vr4, $vr0, 2 -; LA64-NEXT: fcmp.cult.s $fcc0, $fa4, $fa1 -; LA64-NEXT: vinsgr2vr.w $vr3, $a3, 1 -; LA64-NEXT: movcf2gr $a3, $fcc0 -; LA64-NEXT: maskeqz $a4, $a1, $a3 -; LA64-NEXT: ftintrz.l.s $fa5, $fa4 -; LA64-NEXT: movfr2gr.d $a5, $fa5 -; LA64-NEXT: masknez $a3, $a5, $a3 -; LA64-NEXT: fcmp.clt.s $fcc0, $fa2, $fa4 -; LA64-NEXT: or $a3, $a4, $a3 -; LA64-NEXT: movcf2gr $a4, $fcc0 -; LA64-NEXT: masknez $a3, $a3, $a4 -; LA64-NEXT: maskeqz $a4, $a2, $a4 -; LA64-NEXT: fcmp.cun.s $fcc0, $fa4, $fa4 -; LA64-NEXT: or $a3, $a4, $a3 -; LA64-NEXT: movcf2gr $a4, $fcc0 -; LA64-NEXT: masknez $a3, $a3, $a4 -; LA64-NEXT: vreplvei.w $vr0, $vr0, 3 -; LA64-NEXT: fcmp.cult.s $fcc0, $fa0, $fa1 -; LA64-NEXT: vinsgr2vr.w $vr3, $a3, 2 -; LA64-NEXT: movcf2gr $a3, $fcc0 -; LA64-NEXT: maskeqz $a1, $a1, $a3 -; LA64-NEXT: ftintrz.l.s $fa1, $fa0 -; LA64-NEXT: movfr2gr.d $a4, $fa1 -; LA64-NEXT: masknez $a3, $a4, $a3 -; LA64-NEXT: fcmp.clt.s $fcc0, $fa2, $fa0 -; LA64-NEXT: or $a1, $a1, $a3 -; LA64-NEXT: movcf2gr $a3, $fcc0 -; LA64-NEXT: masknez $a1, $a1, $a3 -; LA64-NEXT: maskeqz $a2, $a2, $a3 -; LA64-NEXT: fcmp.cun.s $fcc0, $fa0, $fa0 -; LA64-NEXT: or $a1, $a2, $a1 -; LA64-NEXT: movcf2gr $a2, $fcc0 -; LA64-NEXT: masknez $a1, $a1, $a2 -; LA64-NEXT: vinsgr2vr.w $vr3, $a1, 3 -; LA64-NEXT: vst $vr3, $a0, 0 -; LA64-NEXT: ret +; CHECK-LABEL: fptosi_sat_v4f32_to_v4i32: +; CHECK: # %bb.0: # %entry +; CHECK-NEXT: vld $vr0, $a1, 0 +; CHECK-NEXT: vftintrz.w.s $vr0, $vr0 +; CHECK-NEXT: vst $vr0, $a0, 0 +; CHECK-NEXT: ret entry: %v = load <4 x float>, ptr %a %r = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> %v) @@ -174,148 +17,12 @@ entry: } define void @fptosi_sat_v2f64_to_v2i64(ptr %res, ptr %a) nounwind { -; LA32-LABEL: fptosi_sat_v2f64_to_v2i64: -; LA32: # %bb.0: # %entry -; LA32-NEXT: addi.w $sp, $sp, -80 -; LA32-NEXT: st.w $ra, $sp, 76 # 4-byte Folded Spill -; LA32-NEXT: st.w $fp, $sp, 72 # 4-byte Folded Spill -; LA32-NEXT: st.w $s0, $sp, 68 # 4-byte Folded Spill -; LA32-NEXT: st.w $s1, $sp, 64 # 4-byte Folded Spill -; LA32-NEXT: st.w $s2, $sp, 60 # 4-byte Folded Spill -; LA32-NEXT: fst.d $fs0, $sp, 48 # 8-byte Folded Spill -; LA32-NEXT: vld $vr0, $a1, 0 -; LA32-NEXT: vst $vr0, $sp, 32 # 16-byte Folded Spill -; LA32-NEXT: move $fp, $a0 -; LA32-NEXT: vreplvei.d $vr0, $vr0, 0 -; LA32-NEXT: vst $vr0, $sp, 16 # 16-byte Folded Spill -; LA32-NEXT: movgr2fr.w $fs0, $zero -; LA32-NEXT: lu12i.w $a0, -246272 -; LA32-NEXT: movgr2frh.w $fs0, $a0 -; LA32-NEXT: fcmp.cult.d $fcc0, $fa0, $fs0 -; LA32-NEXT: movcf2gr $a0, $fcc0 -; LA32-NEXT: st.w $a0, $sp, 12 -; LA32-NEXT: # kill: def $f0_64 killed $f0_64 killed $vr0 -; LA32-NEXT: bl __fixdfdi -; LA32-NEXT: ld.w $a2, $sp, 12 -; LA32-NEXT: movgr2cf $fcc0, $a2 -; LA32-NEXT: movcf2gr $a2, $fcc0 -; LA32-NEXT: pcalau12i $a3, %pc_hi20(.LCPI1_0) -; LA32-NEXT: fld.d $fa2, $a3, %pc_lo12(.LCPI1_0) -; LA32-NEXT: masknez $a1, $a1, $a2 -; LA32-NEXT: lu12i.w $s0, -524288 -; LA32-NEXT: maskeqz $a3, $s0, $a2 -; LA32-NEXT: vld $vr1, $sp, 16 # 16-byte Folded Reload -; LA32-NEXT: fcmp.clt.d $fcc0, $fa2, $fa1 -; LA32-NEXT: or $a1, $a3, $a1 -; LA32-NEXT: movcf2gr $a3, $fcc0 -; LA32-NEXT: masknez $a1, $a1, $a3 -; LA32-NEXT: lu12i.w $a4, 524287 -; LA32-NEXT: ori $s1, $a4, 4095 -; LA32-NEXT: maskeqz $a4, $s1, $a3 -; LA32-NEXT: fcmp.cun.d $fcc0, $fa1, $fa1 -; LA32-NEXT: or $a1, $a4, $a1 -; LA32-NEXT: movcf2gr $a4, $fcc0 -; LA32-NEXT: masknez $a1, $a1, $a4 -; LA32-NEXT: masknez $a0, $a0, $a2 -; LA32-NEXT: masknez $a0, $a0, $a3 -; LA32-NEXT: addi.w $s2, $zero, -1 -; LA32-NEXT: maskeqz $a2, $s2, $a3 -; LA32-NEXT: or $a0, $a2, $a0 -; LA32-NEXT: masknez $a0, $a0, $a4 -; LA32-NEXT: vinsgr2vr.w $vr1, $a0, 0 -; LA32-NEXT: vinsgr2vr.w $vr1, $a1, 1 -; LA32-NEXT: vst $vr1, $sp, 16 # 16-byte Folded Spill -; LA32-NEXT: vld $vr1, $sp, 32 # 16-byte Folded Reload -; LA32-NEXT: vreplvei.d $vr0, $vr1, 1 -; LA32-NEXT: vst $vr0, $sp, 32 # 16-byte Folded Spill -; LA32-NEXT: fcmp.clt.d $fcc0, $fa2, $fa0 -; LA32-NEXT: movcf2gr $a0, $fcc0 -; LA32-NEXT: st.w $a0, $sp, 12 -; LA32-NEXT: fcmp.cult.d $fcc0, $fa0, $fs0 -; LA32-NEXT: movcf2gr $a0, $fcc0 -; LA32-NEXT: st.w $a0, $sp, 8 -; LA32-NEXT: # kill: def $f0_64 killed $f0_64 killed $vr0 -; LA32-NEXT: bl __fixdfdi -; LA32-NEXT: ld.w $a2, $sp, 8 -; LA32-NEXT: movgr2cf $fcc0, $a2 -; LA32-NEXT: movcf2gr $a2, $fcc0 -; LA32-NEXT: masknez $a0, $a0, $a2 -; LA32-NEXT: ld.w $a3, $sp, 12 -; LA32-NEXT: movgr2cf $fcc0, $a3 -; LA32-NEXT: movcf2gr $a3, $fcc0 -; LA32-NEXT: masknez $a0, $a0, $a3 -; LA32-NEXT: maskeqz $a4, $s2, $a3 -; LA32-NEXT: vld $vr0, $sp, 32 # 16-byte Folded Reload -; LA32-NEXT: fcmp.cun.d $fcc0, $fa0, $fa0 -; LA32-NEXT: or $a0, $a4, $a0 -; LA32-NEXT: movcf2gr $a4, $fcc0 -; LA32-NEXT: masknez $a0, $a0, $a4 -; LA32-NEXT: vld $vr0, $sp, 16 # 16-byte Folded Reload -; LA32-NEXT: vinsgr2vr.w $vr0, $a0, 2 -; LA32-NEXT: masknez $a0, $a1, $a2 -; LA32-NEXT: maskeqz $a1, $s0, $a2 -; LA32-NEXT: or $a0, $a1, $a0 -; LA32-NEXT: masknez $a0, $a0, $a3 -; LA32-NEXT: maskeqz $a1, $s1, $a3 -; LA32-NEXT: or $a0, $a1, $a0 -; LA32-NEXT: masknez $a0, $a0, $a4 -; LA32-NEXT: vinsgr2vr.w $vr0, $a0, 3 -; LA32-NEXT: vst $vr0, $fp, 0 -; LA32-NEXT: fld.d $fs0, $sp, 48 # 8-byte Folded Reload -; LA32-NEXT: ld.w $s2, $sp, 60 # 4-byte Folded Reload -; LA32-NEXT: ld.w $s1, $sp, 64 # 4-byte Folded Reload -; LA32-NEXT: ld.w $s0, $sp, 68 # 4-byte Folded Reload -; LA32-NEXT: ld.w $fp, $sp, 72 # 4-byte Folded Reload -; LA32-NEXT: ld.w $ra, $sp, 76 # 4-byte Folded Reload -; LA32-NEXT: addi.w $sp, $sp, 80 -; LA32-NEXT: ret -; -; LA64-LABEL: fptosi_sat_v2f64_to_v2i64: -; LA64: # %bb.0: # %entry -; LA64-NEXT: vld $vr0, $a1, 0 -; LA64-NEXT: vreplvei.d $vr1, $vr0, 1 -; LA64-NEXT: lu52i.d $a1, $zero, -962 -; LA64-NEXT: movgr2fr.d $fa2, $a1 -; LA64-NEXT: fcmp.cult.d $fcc0, $fa1, $fa2 -; LA64-NEXT: lu52i.d $a1, $zero, -2048 -; LA64-NEXT: movcf2gr $a2, $fcc0 -; LA64-NEXT: maskeqz $a3, $a1, $a2 -; LA64-NEXT: ftintrz.l.d $fa3, $fa1 -; LA64-NEXT: movfr2gr.d $a4, $fa3 -; LA64-NEXT: masknez $a2, $a4, $a2 -; LA64-NEXT: addi.w $a4, $zero, -1 -; LA64-NEXT: lu52i.d $a5, $a4, 1085 -; LA64-NEXT: movgr2fr.d $fa3, $a5 -; LA64-NEXT: fcmp.clt.d $fcc0, $fa3, $fa1 -; LA64-NEXT: or $a2, $a3, $a2 -; LA64-NEXT: movcf2gr $a3, $fcc0 -; LA64-NEXT: masknez $a2, $a2, $a3 -; LA64-NEXT: lu52i.d $a4, $a4, 2047 -; LA64-NEXT: maskeqz $a3, $a4, $a3 -; LA64-NEXT: fcmp.cun.d $fcc0, $fa1, $fa1 -; LA64-NEXT: or $a2, $a3, $a2 -; LA64-NEXT: movcf2gr $a3, $fcc0 -; LA64-NEXT: vreplvei.d $vr0, $vr0, 0 -; LA64-NEXT: fcmp.cult.d $fcc0, $fa0, $fa2 -; LA64-NEXT: masknez $a2, $a2, $a3 -; LA64-NEXT: movcf2gr $a3, $fcc0 -; LA64-NEXT: maskeqz $a1, $a1, $a3 -; LA64-NEXT: ftintrz.l.d $fa1, $fa0 -; LA64-NEXT: movfr2gr.d $a5, $fa1 -; LA64-NEXT: masknez $a3, $a5, $a3 -; LA64-NEXT: fcmp.clt.d $fcc0, $fa3, $fa0 -; LA64-NEXT: or $a1, $a1, $a3 -; LA64-NEXT: movcf2gr $a3, $fcc0 -; LA64-NEXT: masknez $a1, $a1, $a3 -; LA64-NEXT: maskeqz $a3, $a4, $a3 -; LA64-NEXT: fcmp.cun.d $fcc0, $fa0, $fa0 -; LA64-NEXT: or $a1, $a3, $a1 -; LA64-NEXT: movcf2gr $a3, $fcc0 -; LA64-NEXT: masknez $a1, $a1, $a3 -; LA64-NEXT: vinsgr2vr.d $vr0, $a1, 0 -; LA64-NEXT: vinsgr2vr.d $vr0, $a2, 1 -; LA64-NEXT: vst $vr0, $a0, 0 -; LA64-NEXT: ret +; CHECK-LABEL: fptosi_sat_v2f64_to_v2i64: +; CHECK: # %bb.0: # %entry +; CHECK-NEXT: vld $vr0, $a1, 0 +; CHECK-NEXT: vftintrz.l.d $vr0, $vr0 +; CHECK-NEXT: vst $vr0, $a0, 0 +; CHECK-NEXT: ret entry: %v = load <2 x double>, ptr %a %r = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> %v) _______________________________________________ llvm-branch-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
