[llvm] [LoongArch] Custom scalar UINT_TO_FP and FP_TO_UINT with LSX instructions (PR #200901)
via llvm-commits
llvm-commits at lists.llvm.org
Tue Jun 2 06:34:03 PDT 2026
https://github.com/lrzlin updated https://github.com/llvm/llvm-project/pull/200901
>From 41ea2dec8ab69d25049200eb86d0173bd7a93754 Mon Sep 17 00:00:00 2001
From: Lin Runze <linrunze at loongson.cn>
Date: Fri, 29 May 2026 18:11:23 +0800
Subject: [PATCH] [LoongArch] Custom scalar UINT_TO_FP and FP_TO_UINT with LSX
instructions
---
.../LoongArch/LoongArchISelLowering.cpp | 51 +++++++++++++++++--
.../Target/LoongArch/LoongArchISelLowering.h | 1 +
.../ir-instruction/double-convert.ll | 33 +++---------
.../LoongArch/ir-instruction/float-convert.ll | 32 +++---------
4 files changed, 63 insertions(+), 54 deletions(-)
diff --git a/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp b/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
index 561064b3e1090..ebd31a9c2ed21 100644
--- a/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
+++ b/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
@@ -163,7 +163,7 @@ LoongArchTargetLowering::LoongArchTargetLowering(const TargetMachine &TM,
setOperationAction({ISD::SMUL_LOHI, ISD::UMUL_LOHI}, GRLenVT, Expand);
setOperationAction(ISD::FP_TO_UINT, GRLenVT, Custom);
- setOperationAction(ISD::UINT_TO_FP, GRLenVT, Expand);
+ setOperationAction(ISD::UINT_TO_FP, GRLenVT, Custom);
// Set operations for LA64 only.
@@ -594,6 +594,8 @@ SDValue LoongArchTargetLowering::LowerOperation(SDValue Op,
return lowerConstantPool(Op, DAG);
case ISD::FP_TO_SINT:
return lowerFP_TO_SINT(Op, DAG);
+ case ISD::FP_TO_UINT:
+ return lowerFP_TO_UINT(Op, DAG);
case ISD::BITCAST:
return lowerBITCAST(Op, DAG);
case ISD::UINT_TO_FP:
@@ -4097,11 +4099,33 @@ SDValue LoongArchTargetLowering::lowerVASTART(SDValue Op,
SDValue LoongArchTargetLowering::lowerUINT_TO_FP(SDValue Op,
SelectionDAG &DAG) const {
+ SDLoc DL(Op);
+ SDValue Op0 = Op.getOperand(0);
+ EVT VT = Op.getValueType();
+ EVT Op0VT = Op0.getValueType();
+
+ if ((DAG.SignBitIsZero(Op0) || Op->getFlags().hasNonNeg()) &&
+ !isOperationLegal(ISD::UINT_TO_FP, Op0VT) &&
+ isOperationLegal(ISD::SINT_TO_FP, Op0VT))
+ return DAG.getNode(ISD::SINT_TO_FP, DL, VT, Op0);
+
+ if (Subtarget.hasExtLSX() && Op0VT == MVT::i64 &&
+ (VT == MVT::f32 || VT == MVT::f64)) {
+ Op0 = DAG.getNode(ISD::SCALAR_TO_VECTOR, DL, MVT::v2i64, Op0);
+ SDValue Conv = DAG.getNode(ISD::UINT_TO_FP, DL, MVT::v2f64, Op0);
+ Conv = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, MVT::f64, Conv,
+ DAG.getIntPtrConstant(0, DL));
+ if (VT == MVT::f32)
+ Conv = DAG.getFPExtendOrRound(Conv, DL, VT);
+ return Conv;
+ }
+
+ if (!Subtarget.is64Bit() || !Subtarget.hasBasicF() || Subtarget.hasBasicD())
+ return SDValue();
+
assert(Subtarget.is64Bit() && Subtarget.hasBasicF() &&
!Subtarget.hasBasicD() && "unexpected target features");
- SDLoc DL(Op);
- SDValue Op0 = Op.getOperand(0);
if (Op0->getOpcode() == ISD::AND) {
auto *C = dyn_cast<ConstantSDNode>(Op0.getOperand(1));
if (C && C->getZExtValue() < UINT64_C(0xFFFFFFFF))
@@ -4195,6 +4219,27 @@ SDValue LoongArchTargetLowering::lowerFP_TO_SINT(SDValue Op,
return DAG.getNode(ISD::BITCAST, DL, Op.getValueType(), Trunc);
}
+SDValue LoongArchTargetLowering::lowerFP_TO_UINT(SDValue Op,
+ SelectionDAG &DAG) const {
+ if (!Subtarget.hasExtLSX())
+ return SDValue();
+
+ SDLoc DL(Op);
+ SDValue Src = Op.getOperand(0);
+ EVT VT = Op.getValueType();
+ EVT SrcVT = Src.getValueType();
+ if (VT == MVT::i64 && (SrcVT == MVT::f32 || SrcVT == MVT::f64)) {
+ if (SrcVT == MVT::f32)
+ Src = DAG.getNode(ISD::FP_EXTEND, DL, MVT::f64, Src);
+ Src = DAG.getNode(ISD::SCALAR_TO_VECTOR, DL, MVT::v2f64, Src);
+ SDValue Conv = DAG.getNode(ISD::FP_TO_UINT, DL, MVT::v2i64, Src);
+ return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, VT, Conv,
+ DAG.getIntPtrConstant(0, DL));
+ }
+
+ return SDValue();
+}
+
static SDValue getTargetNode(GlobalAddressSDNode *N, SDLoc DL, EVT Ty,
SelectionDAG &DAG, unsigned Flags) {
return DAG.getTargetGlobalAddress(N->getGlobal(), DL, Ty, 0, Flags);
diff --git a/llvm/lib/Target/LoongArch/LoongArchISelLowering.h b/llvm/lib/Target/LoongArch/LoongArchISelLowering.h
index 189ecbe4820d2..2f2eda1e2c7d9 100644
--- a/llvm/lib/Target/LoongArch/LoongArchISelLowering.h
+++ b/llvm/lib/Target/LoongArch/LoongArchISelLowering.h
@@ -226,6 +226,7 @@ class LoongArchTargetLowering : public TargetLowering {
SDValue lowerConstantPool(SDValue Op, SelectionDAG &DAG) const;
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 lowerBITCAST(SDValue Op, SelectionDAG &DAG) const;
SDValue lowerUINT_TO_FP(SDValue Op, SelectionDAG &DAG) const;
SDValue lowerSINT_TO_FP(SDValue Op, SelectionDAG &DAG) const;
diff --git a/llvm/test/CodeGen/LoongArch/ir-instruction/double-convert.ll b/llvm/test/CodeGen/LoongArch/ir-instruction/double-convert.ll
index ef211139afdf5..ef6b8fdb86ce7 100644
--- a/llvm/test/CodeGen/LoongArch/ir-instruction/double-convert.ll
+++ b/llvm/test/CodeGen/LoongArch/ir-instruction/double-convert.ll
@@ -174,20 +174,9 @@ define i64 @convert_double_to_u64(double %a) nounwind {
;
; LA64-LABEL: convert_double_to_u64:
; LA64: # %bb.0:
-; LA64-NEXT: lu52i.d $a0, $zero, 1086
-; LA64-NEXT: movgr2fr.d $fa1, $a0
-; LA64-NEXT: fcmp.clt.d $fcc0, $fa0, $fa1
-; LA64-NEXT: fsub.d $fa1, $fa0, $fa1
-; LA64-NEXT: ftintrz.l.d $fa1, $fa1
-; LA64-NEXT: movfr2gr.d $a0, $fa1
-; LA64-NEXT: lu52i.d $a1, $zero, -2048
-; LA64-NEXT: xor $a0, $a0, $a1
-; LA64-NEXT: movcf2gr $a1, $fcc0
-; LA64-NEXT: masknez $a0, $a0, $a1
-; LA64-NEXT: ftintrz.l.d $fa0, $fa0
-; LA64-NEXT: movfr2gr.d $a2, $fa0
-; LA64-NEXT: maskeqz $a1, $a2, $a1
-; LA64-NEXT: or $a0, $a1, $a0
+; LA64-NEXT: # kill: def $f0_64 killed $f0_64 def $vr0
+; LA64-NEXT: vftintrz.lu.d $vr0, $vr0
+; LA64-NEXT: vpickve2gr.d $a0, $vr0, 0
; LA64-NEXT: ret
%1 = fptoui double %a to i64
ret i64 %1
@@ -261,18 +250,10 @@ define double @convert_u64_to_double(i64 %a) nounwind {
;
; LA64-LABEL: convert_u64_to_double:
; LA64: # %bb.0:
-; LA64-NEXT: srli.d $a1, $a0, 32
-; LA64-NEXT: lu52i.d $a2, $zero, 1107
-; LA64-NEXT: or $a1, $a1, $a2
-; LA64-NEXT: movgr2fr.d $fa0, $a1
-; LA64-NEXT: lu12i.w $a1, 256
-; LA64-NEXT: lu52i.d $a1, $a1, 1107
-; LA64-NEXT: movgr2fr.d $fa1, $a1
-; LA64-NEXT: fsub.d $fa0, $fa0, $fa1
-; LA64-NEXT: lu12i.w $a1, 275200
-; LA64-NEXT: bstrins.d $a0, $a1, 63, 32
-; LA64-NEXT: movgr2fr.d $fa1, $a0
-; LA64-NEXT: fadd.d $fa0, $fa1, $fa0
+; LA64-NEXT: vinsgr2vr.d $vr0, $a0, 0
+; LA64-NEXT: vffint.d.lu $vr0, $vr0
+; LA64-NEXT: vreplvei.d $vr0, $vr0, 0
+; LA64-NEXT: # kill: def $f0_64 killed $f0_64 killed $vr0
; LA64-NEXT: ret
%1 = uitofp i64 %a to double
ret double %1
diff --git a/llvm/test/CodeGen/LoongArch/ir-instruction/float-convert.ll b/llvm/test/CodeGen/LoongArch/ir-instruction/float-convert.ll
index 8328bb02cf8b5..c2e7a95d7a762 100644
--- a/llvm/test/CodeGen/LoongArch/ir-instruction/float-convert.ll
+++ b/llvm/test/CodeGen/LoongArch/ir-instruction/float-convert.ll
@@ -283,20 +283,9 @@ define i64 @convert_float_to_u64(float %a) nounwind {
;
; LA64D-LABEL: convert_float_to_u64:
; LA64D: # %bb.0:
-; LA64D-NEXT: lu12i.w $a0, 389120
-; LA64D-NEXT: movgr2fr.w $fa1, $a0
-; LA64D-NEXT: fcmp.clt.s $fcc0, $fa0, $fa1
-; LA64D-NEXT: fsub.s $fa1, $fa0, $fa1
-; LA64D-NEXT: ftintrz.l.s $fa1, $fa1
-; LA64D-NEXT: movfr2gr.d $a0, $fa1
-; LA64D-NEXT: lu52i.d $a1, $zero, -2048
-; LA64D-NEXT: xor $a0, $a0, $a1
-; LA64D-NEXT: movcf2gr $a1, $fcc0
-; LA64D-NEXT: masknez $a0, $a0, $a1
-; LA64D-NEXT: ftintrz.l.s $fa0, $fa0
-; LA64D-NEXT: movfr2gr.d $a2, $fa0
-; LA64D-NEXT: maskeqz $a1, $a2, $a1
-; LA64D-NEXT: or $a0, $a1, $a0
+; LA64D-NEXT: fcvt.d.s $fa0, $fa0
+; LA64D-NEXT: vftintrz.lu.d $vr0, $vr0
+; LA64D-NEXT: vpickve2gr.d $a0, $vr0, 0
; LA64D-NEXT: ret
%1 = fptoui float %a to i64
ret i64 %1
@@ -562,17 +551,10 @@ define float @convert_u64_to_float(i64 %a) nounwind {
;
; LA64D-LABEL: convert_u64_to_float:
; LA64D: # %bb.0:
-; LA64D-NEXT: srli.d $a1, $a0, 1
-; LA64D-NEXT: andi $a2, $a0, 1
-; LA64D-NEXT: or $a1, $a2, $a1
-; LA64D-NEXT: movgr2fr.d $fa0, $a1
-; LA64D-NEXT: ffint.s.l $fa0, $fa0
-; LA64D-NEXT: fadd.s $fa0, $fa0, $fa0
-; LA64D-NEXT: slti $a1, $a0, 0
-; LA64D-NEXT: movgr2fr.d $fa1, $a0
-; LA64D-NEXT: ffint.s.l $fa1, $fa1
-; LA64D-NEXT: movgr2cf $fcc0, $a1
-; LA64D-NEXT: fsel $fa0, $fa1, $fa0, $fcc0
+; LA64D-NEXT: vinsgr2vr.d $vr0, $a0, 0
+; LA64D-NEXT: vffint.d.lu $vr0, $vr0
+; LA64D-NEXT: vreplvei.d $vr0, $vr0, 0
+; LA64D-NEXT: fcvt.s.d $fa0, $fa0
; LA64D-NEXT: ret
%1 = uitofp i64 %a to float
ret float %1
More information about the llvm-commits
mailing list