[llvm] Intrinsic: introduce minimumnum and maximumnum for IR and SelectionDAG (PR #96649)
YunQiang Su via llvm-commits
llvm-commits at lists.llvm.org
Thu Jul 25 00:19:25 PDT 2024
https://github.com/wzssyqa updated https://github.com/llvm/llvm-project/pull/96649
>From 8209ec7b16a75eb57c8757a99a6f5b530e043bf0 Mon Sep 17 00:00:00 2001
From: YunQiang Su <syq at gcc.gnu.org>
Date: Tue, 25 Jun 2024 22:23:41 +0800
Subject: [PATCH 1/8] Intrinsic: introduce minimumnum and maximumnum for IR and
SelectionDAG
C23 introduced new functions fminimum_num and fmaximum_num,
and they follow the minimumNumber and maximumNumber of IEEE754-2019.
Let's introduce new intrinsics to support them.
This patch introduces support only support for scalar values.
The support of
vector (vp, vp.reduce, vector.reduce),
experimental.constrained
will be added in future patches.
With this patch, MIPSr6 and LoongArch can work out of box.
Other architectures will fallback to libc call. The support of them
will be added in future patches.
Background
Currently we have fminnum/fmaxnum, which have different
behavior on different platform for NUM vs sNaN:
1) Fallback to fmin(3)/fmax(3): return qNaN.
2) ARM64/ARM32+Neon: same as libc.
3) MIPSr6/LoongArch/RISC-V: return NUM.
And the fix of fminnum/fmaxnum to follow minNUM/maxNUM of IEEE754-2008
will submit as separated patches.
---
llvm/docs/LangRef.rst | 186 ++++++++++++++++++
.../llvm/Analysis/TargetLibraryInfo.def | 33 ++++
llvm/include/llvm/CodeGen/BasicTTIImpl.h | 6 +
llvm/include/llvm/CodeGen/ISDOpcodes.h | 5 +
llvm/include/llvm/CodeGen/TargetLowering.h | 7 +
llvm/include/llvm/IR/IRBuilder.h | 12 ++
llvm/include/llvm/IR/IntrinsicInst.h | 2 +
llvm/include/llvm/IR/Intrinsics.td | 8 +
llvm/include/llvm/IR/RuntimeLibcalls.def | 10 +
.../include/llvm/Target/TargetSelectionDAG.td | 4 +
llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp | 5 +-
llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp | 18 ++
.../SelectionDAG/LegalizeFloatTypes.cpp | 46 +++++
llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h | 4 +
.../lib/CodeGen/SelectionDAG/SelectionDAG.cpp | 8 +-
.../SelectionDAG/SelectionDAGBuilder.cpp | 24 +++
.../SelectionDAG/SelectionDAGDumper.cpp | 2 +
.../CodeGen/SelectionDAG/TargetLowering.cpp | 87 ++++++++
llvm/lib/CodeGen/TargetLoweringBase.cpp | 1 +
.../LoongArch/fp-maximumnum-minimumnum.ll | 154 +++++++++++++++
.../CodeGen/Mips/fp-maximumnum-minimumnum.ll | 100 ++++++++++
.../tools/llvm-tli-checker/ps4-tli-check.yaml | 14 +-
.../Analysis/TargetLibraryInfoTest.cpp | 8 +-
23 files changed, 737 insertions(+), 7 deletions(-)
create mode 100644 llvm/test/CodeGen/LoongArch/fp-maximumnum-minimumnum.ll
create mode 100644 llvm/test/CodeGen/Mips/fp-maximumnum-minimumnum.ll
diff --git a/llvm/docs/LangRef.rst b/llvm/docs/LangRef.rst
index cd86156ec816f..61a08f924955e 100644
--- a/llvm/docs/LangRef.rst
+++ b/llvm/docs/LangRef.rst
@@ -16101,6 +16101,100 @@ The returned value is completely identical to the input except for the sign bit;
in particular, if the input is a NaN, then the quiet/signaling bit and payload
are perfectly preserved.
+.. _i_fminmax_family:
+
+'``llvm.min.*``' Intrinsics Comparation
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Standard:
+"""""""""
+
+IEEE754 and ISO C define some min/max operations, and they have some differences
+on working with qNaN/sNaN and +0.0/-0.0. Here is the list:
+
+.. list-table::
+ :header-rows: 2
+
+ * - ``ISO C``
+ - fmin/fmax
+ - fmininum/fmaximum
+ - fminimum_num/fmaximum_num
+
+ * - ``IEEE754``
+ - minNum/maxNum (2008)
+ - minimum/maximum (2019)
+ - minimumNumber/maximumNumber (2019)
+
+ * - ``+0.0 vs -0.0``
+ - either one
+ - +0.0 > -0.0
+ - +0.0 > -0.0
+
+ * - ``NUM vs sNaN``
+ - qNaN, invalid exception
+ - qNaN, invalid exception
+ - NUM, invalid exception
+
+ * - ``qNaN vs sNaN``
+ - qNaN, invalid exception
+ - qNaN, invalid exception
+ - qNaN, invalid exception
+
+ * - ``NUM vs qNaN``
+ - NUM, no exception
+ - qNaN, no exception
+ - NUM, no exception
+
+LLVM Implementation:
+""""""""""""""""""""
+
+LLVM implements all ISO C flavors as listed in this table.
+Only basic intrinsics list here. The constrained version
+ones may have different behaivor on exception.
+
+Since some architectures implement minNum/maxNum with +0.0>-0.0,
+so we define internal ISD::MINNUM_IEEE and ISD::MAXNUM_IEEE.
+They will be helpful to implement minimumnum/maximumnum.
+
+.. list-table::
+ :header-rows: 1
+ :widths: 16 28 28 28
+
+ * - Operation
+ - minnum/maxnum
+ - minimum/maximum
+ - minimumnum/maximumnum
+
+ * - ``NUM vs qNaN``
+ - NUM, no exception
+ - qNaN, no exception
+ - NUM, no exception
+
+ * - ``NUM vs sNaN``
+ - qNaN, invalid exception
+ - qNaN, invalid exception
+ - NUM, invalid exception
+
+ * - ``qNaN vs sNaN``
+ - qNaN, invalid exception
+ - qNaN, invalid exception
+ - qNaN, invalid exception
+
+ * - ``sNaN vs sNaN``
+ - qNaN, invalid exception
+ - qNaN, invalid exception
+ - qNaN, invalid exception
+
+ * - ``+0.0 vs -0.0``
+ - either one
+ - +0.0(max)/-0.0(min)
+ - +0.0(max)/-0.0(min)
+
+ * - ``NUM vs NUM``
+ - larger(max)/smaller(min)
+ - larger(max)/smaller(min)
+ - larger(max)/smaller(min)
+
.. _i_minnum:
'``llvm.minnum.*``' Intrinsic
@@ -16282,6 +16376,98 @@ of the two arguments. -0.0 is considered to be less than +0.0 for this
intrinsic. Note that these are the semantics specified in the draft of
IEEE 754-2019.
+.. _i_minimumnum:
+
+'``llvm.minimumnum.*``' Intrinsic
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Syntax:
+"""""""
+
+This is an overloaded intrinsic. You can use ``llvm.minimumnum`` on any
+floating-point or vector of floating-point type. Not all targets support
+all types however.
+
+::
+
+ declare float @llvm.minimumnum.f32(float %Val0, float %Val1)
+ declare double @llvm.minimumnum.f64(double %Val0, double %Val1)
+ declare x86_fp80 @llvm.minimumnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
+ declare fp128 @llvm.minimumnum.f128(fp128 %Val0, fp128 %Val1)
+ declare ppc_fp128 @llvm.minimumnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
+
+Overview:
+"""""""""
+
+The '``llvm.minimumnum.*``' intrinsics return the minimum of the two
+arguments, not propagating NaNs and treating -0.0 as less than +0.0.
+
+
+Arguments:
+""""""""""
+
+The arguments and return value are floating-point numbers of the same
+type.
+
+Semantics:
+""""""""""
+If both operands are NaNs (including sNaN), returns qNaN. If one operand
+is NaN (including sNaN) and another operand is a number, return the number.
+Otherwise returns the lesser of the two arguments. -0.0 is considered to
+be less than +0.0 for this intrinsic.
+
+Note that these are the semantics of minimumNumber specified in IEEE 754-2019.
+
+It has some different with '``llvm.minnum.*``':
+1)'``llvm.minnum.*``' will return qNaN if either operand is sNaN.
+2)'``llvm.minnum*``' may return either one if we compare +0.0 vs -0.0.
+
+.. _i_maximumnum:
+
+'``llvm.maximumnum.*``' Intrinsic
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Syntax:
+"""""""
+
+This is an overloaded intrinsic. You can use ``llvm.maximumnum`` on any
+floating-point or vector of floating-point type. Not all targets support
+all types however.
+
+::
+
+ declare float @llvm.maximumnum.f32(float %Val0, float %Val1)
+ declare double @llvm.maximumnum.f64(double %Val0, double %Val1)
+ declare x86_fp80 @llvm.maximumnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
+ declare fp128 @llvm.maximumnum.f128(fp128 %Val0, fp128 %Val1)
+ declare ppc_fp128 @llvm.maximumnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
+
+Overview:
+"""""""""
+
+The '``llvm.maximumnum.*``' intrinsics return the maximum of the two
+arguments, not propagating NaNs and treating -0.0 as less than +0.0.
+
+
+Arguments:
+""""""""""
+
+The arguments and return value are floating-point numbers of the same
+type.
+
+Semantics:
+""""""""""
+If both operands are NaNs (including sNaN), returns qNaN. If one operand
+is NaN (including sNaN) and another operand is a number, return the number.
+Otherwise returns the greater of the two arguments. -0.0 is considered to
+be less than +0.0 for this intrinsic.
+
+Note that these are the semantics of maximumNumber specified in IEEE 754-2019.
+
+It has some different with '``llvm.maxnum.*``':
+1)'``llvm.maxnum.*``' will return qNaN if either operand is sNaN.
+2)'``llvm.maxnum*``' may return either one if we compare +0.0 vs -0.0.
+
.. _int_copysign:
'``llvm.copysign.*``' Intrinsic
diff --git a/llvm/include/llvm/Analysis/TargetLibraryInfo.def b/llvm/include/llvm/Analysis/TargetLibraryInfo.def
index 623cdb4b6e0b7..80df8eedf7be9 100644
--- a/llvm/include/llvm/Analysis/TargetLibraryInfo.def
+++ b/llvm/include/llvm/Analysis/TargetLibraryInfo.def
@@ -1347,6 +1347,39 @@ TLI_DEFINE_ENUM_INTERNAL(fminl)
TLI_DEFINE_STRING_INTERNAL("fminl")
TLI_DEFINE_SIG_INTERNAL(Floating, Same, Same)
+// Calls to fmaximum_num and fminimum_num library functions expand to the llvm.maximumnum and
+// llvm.minimumnum intrinsics with the correct parameter types for the arguments
+// (all types must match).
+/// double fmaximum_num(double x, double y);
+TLI_DEFINE_ENUM_INTERNAL(fmaximum_num)
+TLI_DEFINE_STRING_INTERNAL("fmaximum_num")
+TLI_DEFINE_SIG_INTERNAL(Floating, Same, Same)
+
+/// float fmaximum_numf(float x, float y);
+TLI_DEFINE_ENUM_INTERNAL(fmaximum_numf)
+TLI_DEFINE_STRING_INTERNAL("fmaximum_numf")
+TLI_DEFINE_SIG_INTERNAL(Floating, Same, Same)
+
+/// long double fmaximum_numl(long double x, long double y);
+TLI_DEFINE_ENUM_INTERNAL(fmaximum_numl)
+TLI_DEFINE_STRING_INTERNAL("fmaximum_numl")
+TLI_DEFINE_SIG_INTERNAL(Floating, Same, Same)
+
+/// double fminimum_num(double x, double y);
+TLI_DEFINE_ENUM_INTERNAL(fminimum_num)
+TLI_DEFINE_STRING_INTERNAL("fminimum_num")
+TLI_DEFINE_SIG_INTERNAL(Floating, Same, Same)
+
+/// float fminimum_numf(float x, float y);
+TLI_DEFINE_ENUM_INTERNAL(fminimum_numf)
+TLI_DEFINE_STRING_INTERNAL("fminimum_numf")
+TLI_DEFINE_SIG_INTERNAL(Floating, Same, Same)
+
+/// long double fminimum_numl(long double x, long double y);
+TLI_DEFINE_ENUM_INTERNAL(fminimum_numl)
+TLI_DEFINE_STRING_INTERNAL("fminimum_numl")
+TLI_DEFINE_SIG_INTERNAL(Floating, Same, Same)
+
/// double fmod(double x, double y);
TLI_DEFINE_ENUM_INTERNAL(fmod)
TLI_DEFINE_STRING_INTERNAL("fmod")
diff --git a/llvm/include/llvm/CodeGen/BasicTTIImpl.h b/llvm/include/llvm/CodeGen/BasicTTIImpl.h
index 5b9cc5dfeeadb..fe580e908ed16 100644
--- a/llvm/include/llvm/CodeGen/BasicTTIImpl.h
+++ b/llvm/include/llvm/CodeGen/BasicTTIImpl.h
@@ -2033,6 +2033,12 @@ class BasicTTIImplBase : public TargetTransformInfoImplCRTPBase<T> {
case Intrinsic::maximum:
ISD = ISD::FMAXIMUM;
break;
+ case Intrinsic::minimumnum:
+ ISD = ISD::FMINIMUMNUM;
+ break;
+ case Intrinsic::maximumnum:
+ ISD = ISD::FMAXIMUMNUM;
+ break;
case Intrinsic::copysign:
ISD = ISD::FCOPYSIGN;
break;
diff --git a/llvm/include/llvm/CodeGen/ISDOpcodes.h b/llvm/include/llvm/CodeGen/ISDOpcodes.h
index 5b657fb171296..be2c3e8a0249a 100644
--- a/llvm/include/llvm/CodeGen/ISDOpcodes.h
+++ b/llvm/include/llvm/CodeGen/ISDOpcodes.h
@@ -1027,6 +1027,11 @@ enum NodeType {
FMINIMUM,
FMAXIMUM,
+ /// FMINIMUMNUM/FMAXIMUMNUM - minimumnum/maximumnum that is same with
+ /// FMINNUM_IEEE and FMAXNUM_IEEE besides if either operand is sNaN.
+ FMINIMUMNUM,
+ FMAXIMUMNUM,
+
/// FSINCOS - Compute both fsin and fcos as a single operation.
FSINCOS,
diff --git a/llvm/include/llvm/CodeGen/TargetLowering.h b/llvm/include/llvm/CodeGen/TargetLowering.h
index 9d9886f4920a2..1e08bd86af372 100644
--- a/llvm/include/llvm/CodeGen/TargetLowering.h
+++ b/llvm/include/llvm/CodeGen/TargetLowering.h
@@ -2907,6 +2907,8 @@ class TargetLoweringBase {
case ISD::FMAXNUM_IEEE:
case ISD::FMINIMUM:
case ISD::FMAXIMUM:
+ case ISD::FMINIMUMNUM:
+ case ISD::FMAXIMUMNUM:
case ISD::AVGFLOORS:
case ISD::AVGFLOORU:
case ISD::AVGCEILS:
@@ -5139,6 +5141,8 @@ class TargetLowering : public TargetLoweringBase {
/// through to the default expansion/soften to libcall, we might introduce a
/// link-time dependency on libm into a file that originally did not have one.
SDValue createSelectForFMINNUM_FMAXNUM(SDNode *Node, SelectionDAG &DAG) const;
+ SDValue createSelectForFMINIMUMNUM_FMAXIMUMNUM(SDNode *Node,
+ SelectionDAG &DAG) const;
/// Return a reciprocal estimate value for the input operand.
/// \p Enabled is a ReciprocalEstimate enum with value either 'Unspecified' or
@@ -5270,6 +5274,9 @@ class TargetLowering : public TargetLoweringBase {
/// Expand fminimum/fmaximum into multiple comparison with selects.
SDValue expandFMINIMUM_FMAXIMUM(SDNode *N, SelectionDAG &DAG) const;
+ /// Expand fminimumnum/fmaximumnum into multiple comparison with selects.
+ SDValue expandFMINIMUMNUM_FMAXIMUMNUM(SDNode *N, SelectionDAG &DAG) const;
+
/// Expand FP_TO_[US]INT_SAT into FP_TO_[US]INT and selects or min/max.
/// \param N Node to expand
/// \returns The expansion result
diff --git a/llvm/include/llvm/IR/IRBuilder.h b/llvm/include/llvm/IR/IRBuilder.h
index 31a1fef321995..ef6e8a614526a 100644
--- a/llvm/include/llvm/IR/IRBuilder.h
+++ b/llvm/include/llvm/IR/IRBuilder.h
@@ -1015,6 +1015,18 @@ class IRBuilderBase {
return CreateBinaryIntrinsic(Intrinsic::maximum, LHS, RHS, nullptr, Name);
}
+ /// Create call to the minimumnum intrinsic.
+ Value *CreateMinimumNum(Value *LHS, Value *RHS, const Twine &Name = "") {
+ return CreateBinaryIntrinsic(Intrinsic::minimumnum, LHS, RHS, nullptr,
+ Name);
+ }
+
+ /// Create call to the maximum intrinsic.
+ Value *CreateMaximumNum(Value *LHS, Value *RHS, const Twine &Name = "") {
+ return CreateBinaryIntrinsic(Intrinsic::maximumnum, LHS, RHS, nullptr,
+ Name);
+ }
+
/// Create call to the copysign intrinsic.
Value *CreateCopySign(Value *LHS, Value *RHS,
Instruction *FMFSource = nullptr,
diff --git a/llvm/include/llvm/IR/IntrinsicInst.h b/llvm/include/llvm/IR/IntrinsicInst.h
index fe3f92da400f8..8c268c400dbf3 100644
--- a/llvm/include/llvm/IR/IntrinsicInst.h
+++ b/llvm/include/llvm/IR/IntrinsicInst.h
@@ -76,6 +76,8 @@ class IntrinsicInst : public CallInst {
case Intrinsic::minnum:
case Intrinsic::maximum:
case Intrinsic::minimum:
+ case Intrinsic::maximumnum:
+ case Intrinsic::minimumnum:
case Intrinsic::smax:
case Intrinsic::smin:
case Intrinsic::umax:
diff --git a/llvm/include/llvm/IR/Intrinsics.td b/llvm/include/llvm/IR/Intrinsics.td
index b4e758136b39f..0841273fd2e1e 100644
--- a/llvm/include/llvm/IR/Intrinsics.td
+++ b/llvm/include/llvm/IR/Intrinsics.td
@@ -1085,6 +1085,14 @@ def int_maximum : DefaultAttrsIntrinsic<[llvm_anyfloat_ty],
[LLVMMatchType<0>, LLVMMatchType<0>],
[IntrNoMem, IntrSpeculatable, IntrWillReturn, Commutative]
>;
+def int_minimumnum : DefaultAttrsIntrinsic<[llvm_anyfloat_ty],
+ [LLVMMatchType<0>, LLVMMatchType<0>],
+ [IntrNoMem, IntrSpeculatable, IntrWillReturn, Commutative]
+>;
+def int_maximumnum : DefaultAttrsIntrinsic<[llvm_anyfloat_ty],
+ [LLVMMatchType<0>, LLVMMatchType<0>],
+ [IntrNoMem, IntrSpeculatable, IntrWillReturn, Commutative]
+>;
// Internal interface for object size checking
def int_objectsize : DefaultAttrsIntrinsic<[llvm_anyint_ty],
diff --git a/llvm/include/llvm/IR/RuntimeLibcalls.def b/llvm/include/llvm/IR/RuntimeLibcalls.def
index 89aaf6d1ad83f..c3d5ef9f4e4f8 100644
--- a/llvm/include/llvm/IR/RuntimeLibcalls.def
+++ b/llvm/include/llvm/IR/RuntimeLibcalls.def
@@ -299,6 +299,16 @@ HANDLE_LIBCALL(FMAX_F64, "fmax")
HANDLE_LIBCALL(FMAX_F80, "fmaxl")
HANDLE_LIBCALL(FMAX_F128, "fmaxl")
HANDLE_LIBCALL(FMAX_PPCF128, "fmaxl")
+HANDLE_LIBCALL(FMINIMUMNUM_F32, "fminimum_numf")
+HANDLE_LIBCALL(FMINIMUMNUM_F64, "fminimum_num")
+HANDLE_LIBCALL(FMINIMUMNUM_F80, "fminimum_numl")
+HANDLE_LIBCALL(FMINIMUMNUM_F128, "fminmum_numl")
+HANDLE_LIBCALL(FMINIMUMNUM_PPCF128, "fminimum_numl")
+HANDLE_LIBCALL(FMAXIMUMNUM_F32, "fmaximum_numf")
+HANDLE_LIBCALL(FMAXIMUMNUM_F64, "fmaximum_num")
+HANDLE_LIBCALL(FMAXIMUMNUM_F80, "fmaximum_numl")
+HANDLE_LIBCALL(FMAXIMUMNUM_F128, "fmaxmum_numl")
+HANDLE_LIBCALL(FMAXIMUMNUM_PPCF128, "fmaximum_numl")
HANDLE_LIBCALL(LROUND_F32, "lroundf")
HANDLE_LIBCALL(LROUND_F64, "lround")
HANDLE_LIBCALL(LROUND_F80, "lroundl")
diff --git a/llvm/include/llvm/Target/TargetSelectionDAG.td b/llvm/include/llvm/Target/TargetSelectionDAG.td
index 46044aab79a83..284172e4fd7ea 100644
--- a/llvm/include/llvm/Target/TargetSelectionDAG.td
+++ b/llvm/include/llvm/Target/TargetSelectionDAG.td
@@ -514,6 +514,10 @@ def fminimum : SDNode<"ISD::FMINIMUM" , SDTFPBinOp,
[SDNPCommutative, SDNPAssociative]>;
def fmaximum : SDNode<"ISD::FMAXIMUM" , SDTFPBinOp,
[SDNPCommutative, SDNPAssociative]>;
+def fminimumnum : SDNode<"ISD::FMINIMUMNUM" , SDTFPBinOp,
+ [SDNPCommutative, SDNPAssociative]>;
+def fmaximumnum : SDNode<"ISD::FMAXIMUMNUM" , SDTFPBinOp,
+ [SDNPCommutative, SDNPAssociative]>;
def fgetsign : SDNode<"ISD::FGETSIGN" , SDTFPToIntOp>;
def fcanonicalize : SDNode<"ISD::FCANONICALIZE", SDTFPUnaryOp>;
def fneg : SDNode<"ISD::FNEG" , SDTFPUnaryOp>;
diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 060e66175d965..90b8922e65748 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -1935,7 +1935,9 @@ SDValue DAGCombiner::visit(SDNode *N) {
case ISD::FMINNUM:
case ISD::FMAXNUM:
case ISD::FMINIMUM:
- case ISD::FMAXIMUM: return visitFMinMax(N);
+ case ISD::FMAXIMUM:
+ case ISD::FMINIMUMNUM:
+ case ISD::FMAXIMUMNUM: return visitFMinMax(N);
case ISD::FCEIL: return visitFCEIL(N);
case ISD::FTRUNC: return visitFTRUNC(N);
case ISD::FFREXP: return visitFFREXP(N);
@@ -6085,6 +6087,7 @@ static bool arebothOperandsNotNan(SDValue Operand1, SDValue Operand2,
return DAG.isKnownNeverNaN(Operand2) && DAG.isKnownNeverNaN(Operand1);
}
+// FIXME: use FMINIMUMNUM if possible, such as for RISC-V.
static unsigned getMinMaxOpcodeForFP(SDValue Operand1, SDValue Operand2,
ISD::CondCode CC, unsigned OrAndOpcode,
SelectionDAG &DAG,
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
index 3a39f6a4d2b4a..7f34c98a91e24 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
@@ -3650,6 +3650,12 @@ bool SelectionDAGLegalize::ExpandNode(SDNode *Node) {
Results.push_back(Expanded);
break;
}
+ case ISD::FMINIMUMNUM:
+ case ISD::FMAXIMUMNUM: {
+ if (SDValue Expanded = TLI.expandFMINIMUMNUM_FMAXIMUMNUM(Node, DAG))
+ Results.push_back(Expanded);
+ break;
+ }
case ISD::FSIN:
case ISD::FCOS: {
EVT VT = Node->getValueType(0);
@@ -4519,6 +4525,16 @@ void SelectionDAGLegalize::ConvertNodeToLibcall(SDNode *Node) {
RTLIB::FMAX_F80, RTLIB::FMAX_F128,
RTLIB::FMAX_PPCF128, Results);
break;
+ case ISD::FMINIMUMNUM:
+ ExpandFPLibCall(Node, RTLIB::FMINIMUMNUM_F32, RTLIB::FMINIMUMNUM_F64,
+ RTLIB::FMINIMUMNUM_F80, RTLIB::FMINIMUMNUM_F128,
+ RTLIB::FMINIMUMNUM_PPCF128, Results);
+ break;
+ case ISD::FMAXIMUMNUM:
+ ExpandFPLibCall(Node, RTLIB::FMAXIMUMNUM_F32, RTLIB::FMAXIMUMNUM_F64,
+ RTLIB::FMAXIMUMNUM_F80, RTLIB::FMAXIMUMNUM_F128,
+ RTLIB::FMAXIMUMNUM_PPCF128, Results);
+ break;
case ISD::FSQRT:
case ISD::STRICT_FSQRT:
ExpandFPLibCall(Node, RTLIB::SQRT_F32, RTLIB::SQRT_F64,
@@ -5444,6 +5460,8 @@ void SelectionDAGLegalize::PromoteNode(SDNode *Node) {
case ISD::FMAXNUM:
case ISD::FMINIMUM:
case ISD::FMAXIMUM:
+ case ISD::FMINIMUMNUM:
+ case ISD::FMAXIMUMNUM:
case ISD::FPOW:
Tmp1 = DAG.getNode(ISD::FP_EXTEND, dl, NVT, Node->getOperand(0));
Tmp2 = DAG.getNode(ISD::FP_EXTEND, dl, NVT, Node->getOperand(1));
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
index 41fcc9afe4e90..5a3c70ec4cdab 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
@@ -74,6 +74,8 @@ void DAGTypeLegalizer::SoftenFloatResult(SDNode *N, unsigned ResNo) {
case ISD::FMINNUM: R = SoftenFloatRes_FMINNUM(N); break;
case ISD::STRICT_FMAXNUM:
case ISD::FMAXNUM: R = SoftenFloatRes_FMAXNUM(N); break;
+ case ISD::FMINIMUMNUM: R = SoftenFloatRes_FMINIMUMNUM(N); break;
+ case ISD::FMAXIMUMNUM: R = SoftenFloatRes_FMAXIMUMNUM(N); break;
case ISD::STRICT_FADD:
case ISD::FADD: R = SoftenFloatRes_FADD(N); break;
case ISD::STRICT_FACOS:
@@ -323,6 +325,24 @@ SDValue DAGTypeLegalizer::SoftenFloatRes_FMAXNUM(SDNode *N) {
RTLIB::FMAX_PPCF128));
}
+SDValue DAGTypeLegalizer::SoftenFloatRes_FMINIMUMNUM(SDNode *N) {
+ if (SDValue SelCC = TLI.createSelectForFMINIMUMNUM_FMAXIMUMNUM(N, DAG))
+ return SoftenFloatRes_SELECT_CC(SelCC.getNode());
+ return SoftenFloatRes_Binary(
+ N, GetFPLibCall(N->getValueType(0), RTLIB::FMINIMUMNUM_F32,
+ RTLIB::FMINIMUMNUM_F64, RTLIB::FMINIMUMNUM_F80,
+ RTLIB::FMINIMUMNUM_F128, RTLIB::FMINIMUMNUM_PPCF128));
+}
+
+SDValue DAGTypeLegalizer::SoftenFloatRes_FMAXIMUMNUM(SDNode *N) {
+ if (SDValue SelCC = TLI.createSelectForFMINIMUMNUM_FMAXIMUMNUM(N, DAG))
+ return SoftenFloatRes_SELECT_CC(SelCC.getNode());
+ return SoftenFloatRes_Binary(
+ N, GetFPLibCall(N->getValueType(0), RTLIB::FMAXIMUMNUM_F32,
+ RTLIB::FMAXIMUMNUM_F64, RTLIB::FMAXIMUMNUM_F80,
+ RTLIB::FMAXIMUMNUM_F128, RTLIB::FMAXIMUMNUM_PPCF128));
+}
+
SDValue DAGTypeLegalizer::SoftenFloatRes_FADD(SDNode *N) {
return SoftenFloatRes_Binary(N, GetFPLibCall(N->getValueType(0),
RTLIB::ADD_F32,
@@ -1404,6 +1424,8 @@ void DAGTypeLegalizer::ExpandFloatResult(SDNode *N, unsigned ResNo) {
case ISD::FMINNUM: ExpandFloatRes_FMINNUM(N, Lo, Hi); break;
case ISD::STRICT_FMAXNUM:
case ISD::FMAXNUM: ExpandFloatRes_FMAXNUM(N, Lo, Hi); break;
+ case ISD::FMINIMUMNUM: ExpandFloatRes_FMINIMUMNUM(N, Lo, Hi); break;
+ case ISD::FMAXIMUMNUM: ExpandFloatRes_FMAXIMUMNUM(N, Lo, Hi); break;
case ISD::STRICT_FADD:
case ISD::FADD: ExpandFloatRes_FADD(N, Lo, Hi); break;
case ISD::STRICT_FACOS:
@@ -1561,6 +1583,26 @@ void DAGTypeLegalizer::ExpandFloatRes_FMAXNUM(SDNode *N, SDValue &Lo,
RTLIB::FMAX_PPCF128), Lo, Hi);
}
+void DAGTypeLegalizer::ExpandFloatRes_FMINIMUMNUM(SDNode *N, SDValue &Lo,
+ SDValue &Hi) {
+ ExpandFloatRes_Binary(
+ N,
+ GetFPLibCall(N->getValueType(0), RTLIB::FMINIMUMNUM_F32,
+ RTLIB::FMINIMUMNUM_F64, RTLIB::FMINIMUMNUM_F80,
+ RTLIB::FMINIMUMNUM_F128, RTLIB::FMINIMUMNUM_PPCF128),
+ Lo, Hi);
+}
+
+void DAGTypeLegalizer::ExpandFloatRes_FMAXIMUMNUM(SDNode *N, SDValue &Lo,
+ SDValue &Hi) {
+ ExpandFloatRes_Binary(
+ N,
+ GetFPLibCall(N->getValueType(0), RTLIB::FMAXIMUMNUM_F32,
+ RTLIB::FMAXIMUMNUM_F64, RTLIB::FMAXIMUMNUM_F80,
+ RTLIB::FMAXIMUMNUM_F128, RTLIB::FMAXIMUMNUM_PPCF128),
+ Lo, Hi);
+}
+
void DAGTypeLegalizer::ExpandFloatRes_FADD(SDNode *N, SDValue &Lo,
SDValue &Hi) {
ExpandFloatRes_Binary(N, GetFPLibCall(N->getValueType(0),
@@ -2627,6 +2669,8 @@ void DAGTypeLegalizer::PromoteFloatResult(SDNode *N, unsigned ResNo) {
case ISD::FDIV:
case ISD::FMAXIMUM:
case ISD::FMINIMUM:
+ case ISD::FMAXIMUMNUM:
+ case ISD::FMINIMUMNUM:
case ISD::FMAXNUM:
case ISD::FMINNUM:
case ISD::FMAXNUM_IEEE:
@@ -3069,6 +3113,8 @@ void DAGTypeLegalizer::SoftPromoteHalfResult(SDNode *N, unsigned ResNo) {
case ISD::FDIV:
case ISD::FMAXIMUM:
case ISD::FMINIMUM:
+ case ISD::FMAXIMUMNUM:
+ case ISD::FMINIMUMNUM:
case ISD::FMAXNUM:
case ISD::FMINNUM:
case ISD::FMUL:
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
index d4e61c8588901..2125572208d1e 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
@@ -566,6 +566,8 @@ class LLVM_LIBRARY_VISIBILITY DAGTypeLegalizer {
SDValue SoftenFloatRes_FATAN(SDNode *N);
SDValue SoftenFloatRes_FMINNUM(SDNode *N);
SDValue SoftenFloatRes_FMAXNUM(SDNode *N);
+ SDValue SoftenFloatRes_FMINIMUMNUM(SDNode *N);
+ SDValue SoftenFloatRes_FMAXIMUMNUM(SDNode *N);
SDValue SoftenFloatRes_FADD(SDNode *N);
SDValue SoftenFloatRes_FCBRT(SDNode *N);
SDValue SoftenFloatRes_FCEIL(SDNode *N);
@@ -658,6 +660,8 @@ class LLVM_LIBRARY_VISIBILITY DAGTypeLegalizer {
void ExpandFloatRes_FATAN (SDNode *N, SDValue &Lo, SDValue &Hi);
void ExpandFloatRes_FMINNUM (SDNode *N, SDValue &Lo, SDValue &Hi);
void ExpandFloatRes_FMAXNUM (SDNode *N, SDValue &Lo, SDValue &Hi);
+ void ExpandFloatRes_FMINIMUMNUM(SDNode *N, SDValue &Lo, SDValue &Hi);
+ void ExpandFloatRes_FMAXIMUMNUM(SDNode *N, SDValue &Lo, SDValue &Hi);
void ExpandFloatRes_FADD (SDNode *N, SDValue &Lo, SDValue &Hi);
void ExpandFloatRes_FCBRT (SDNode *N, SDValue &Lo, SDValue &Hi);
void ExpandFloatRes_FCEIL (SDNode *N, SDValue &Lo, SDValue &Hi);
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index bbc44a4716405..a9a91e153f5dd 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -5463,7 +5463,9 @@ bool SelectionDAG::isKnownNeverNaN(SDValue Op, bool SNaN, unsigned Depth) const
return false;
}
case ISD::FMINNUM:
- case ISD::FMAXNUM: {
+ case ISD::FMAXNUM:
+ case ISD::FMINIMUMNUM:
+ case ISD::FMAXIMUMNUM: {
// Only one needs to be known not-nan, since it will be returned if the
// other ends up being one.
return isKnownNeverNaN(Op.getOperand(0), SNaN, Depth + 1) ||
@@ -6803,6 +6805,10 @@ SDValue SelectionDAG::foldConstantFPMath(unsigned Opcode, const SDLoc &DL,
return getConstantFP(minimum(C1, C2), DL, VT);
case ISD::FMAXIMUM:
return getConstantFP(maximum(C1, C2), DL, VT);
+ case ISD::FMINIMUMNUM:
+ return getConstantFP(minimumnum(C1, C2), DL, VT);
+ case ISD::FMAXIMUMNUM:
+ return getConstantFP(maximumnum(C1, C2), DL, VT);
default: break;
}
}
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
index 1791f1b503379..2eb7c726e85bf 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -6881,6 +6881,18 @@ void SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I,
getValue(I.getArgOperand(0)),
getValue(I.getArgOperand(1)), Flags));
return;
+ case Intrinsic::minimumnum:
+ setValue(&I, DAG.getNode(ISD::FMINIMUMNUM, sdl,
+ getValue(I.getArgOperand(0)).getValueType(),
+ getValue(I.getArgOperand(0)),
+ getValue(I.getArgOperand(1)), Flags));
+ return;
+ case Intrinsic::maximumnum:
+ setValue(&I, DAG.getNode(ISD::FMAXIMUMNUM, sdl,
+ getValue(I.getArgOperand(0)).getValueType(),
+ getValue(I.getArgOperand(0)),
+ getValue(I.getArgOperand(1)), Flags));
+ return;
case Intrinsic::copysign:
setValue(&I, DAG.getNode(ISD::FCOPYSIGN, sdl,
getValue(I.getArgOperand(0)).getValueType(),
@@ -9255,6 +9267,18 @@ void SelectionDAGBuilder::visitCall(const CallInst &I) {
if (visitBinaryFloatCall(I, ISD::FMAXNUM))
return;
break;
+ case LibFunc_fminimum_num:
+ case LibFunc_fminimum_numf:
+ case LibFunc_fminimum_numl:
+ if (visitBinaryFloatCall(I, ISD::FMINIMUMNUM))
+ return;
+ break;
+ case LibFunc_fmaximum_num:
+ case LibFunc_fmaximum_numf:
+ case LibFunc_fmaximum_numl:
+ if (visitBinaryFloatCall(I, ISD::FMAXIMUMNUM))
+ return;
+ break;
case LibFunc_sin:
case LibFunc_sinf:
case LibFunc_sinl:
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
index 16fc52caebb75..209c23eab1dda 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
@@ -203,6 +203,8 @@ std::string SDNode::getOperationName(const SelectionDAG *G) const {
case ISD::STRICT_FMINIMUM: return "strict_fminimum";
case ISD::FMAXIMUM: return "fmaximum";
case ISD::STRICT_FMAXIMUM: return "strict_fmaximum";
+ case ISD::FMINIMUMNUM: return "fminimumnum";
+ case ISD::FMAXIMUMNUM: return "fmaximumnum";
case ISD::FNEG: return "fneg";
case ISD::FSQRT: return "fsqrt";
case ISD::STRICT_FSQRT: return "strict_fsqrt";
diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 140c97ccd90ba..b97554a4d0dc4 100644
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -8531,6 +8531,93 @@ SDValue TargetLowering::expandFMINIMUM_FMAXIMUM(SDNode *N,
return MinMax;
}
+SDValue TargetLowering::createSelectForFMINIMUMNUM_FMAXIMUMNUM(
+ SDNode *Node, SelectionDAG &DAG) const {
+ unsigned Opcode = Node->getOpcode();
+ assert((Opcode == ISD::FMINIMUMNUM || Opcode == ISD::FMAXIMUMNUM) &&
+ "Wrong opcode");
+
+ if (Node->getFlags().hasNoNaNs()) {
+ ISD::CondCode Pred = Opcode == ISD::FMINIMUMNUM ? ISD::SETLT : ISD::SETGT;
+ SDValue Op1 = Node->getOperand(0);
+ SDValue Op2 = Node->getOperand(1);
+ SDValue SelCC = DAG.getSelectCC(SDLoc(Node), Op1, Op2, Op1, Op2, Pred);
+ // Copy FMF flags, but always set the no-signed-zeros flag
+ // as this is implied by the FMINNUM/FMAXNUM semantics.
+ SDNodeFlags Flags = Node->getFlags();
+ Flags.setNoSignedZeros(true);
+ SelCC->setFlags(Flags);
+ return SelCC;
+ }
+
+ return SDValue();
+}
+
+SDValue TargetLowering::expandFMINIMUMNUM_FMAXIMUMNUM(SDNode *Node,
+ SelectionDAG &DAG) const {
+ SDLoc dl(Node);
+ unsigned NewOp = Node->getOpcode() == ISD::FMINIMUMNUM ? ISD::FMINNUM_IEEE
+ : ISD::FMAXNUM_IEEE;
+ EVT VT = Node->getValueType(0);
+
+ if (VT.isScalableVector())
+ report_fatal_error(
+ "Expanding fminimumnum/fmaximumnum for scalable vectors is undefined.");
+
+ if (isOperationLegalOrCustom(NewOp, VT)) {
+ SDValue Quiet0 = Node->getOperand(0);
+ SDValue Quiet1 = Node->getOperand(1);
+
+ if (!Node->getFlags().hasNoNaNs()) {
+ // Insert canonicalizes if it's possible we need to quiet to get correct
+ // sNaN behavior.
+ if (!DAG.isKnownNeverSNaN(Quiet0)) {
+ Quiet0 =
+ DAG.getNode(ISD::FCANONICALIZE, dl, VT, Quiet0, Node->getFlags());
+ }
+ if (!DAG.isKnownNeverSNaN(Quiet1)) {
+ Quiet1 =
+ DAG.getNode(ISD::FCANONICALIZE, dl, VT, Quiet1, Node->getFlags());
+ }
+ }
+
+ return DAG.getNode(NewOp, dl, VT, Quiet0, Quiet1, Node->getFlags());
+ }
+
+ // FMINIMUM/FMAXIMUM always return NaN if either operand is NaN.
+ // It has same behavior about +0.0 vs -0.0.
+ if (Node->getFlags().hasNoNaNs() ||
+ (DAG.isKnownNeverNaN(Node->getOperand(0)) &&
+ DAG.isKnownNeverNaN(Node->getOperand(1)))) {
+ unsigned IEEE2019Op =
+ Node->getOpcode() == ISD::FMINIMUMNUM ? ISD::FMINIMUM : ISD::FMAXIMUM;
+ if (isOperationLegalOrCustom(IEEE2019Op, VT))
+ return DAG.getNode(IEEE2019Op, dl, VT, Node->getOperand(0),
+ Node->getOperand(1), Node->getFlags());
+ }
+
+ // FMINNUM/FMAXIMUM returns qNaN if either operand is sNaN, and it may return
+ // either one for +0.0 vs -0.0.
+ // FIXME: maybe we need hasNoSNaNs().
+ if ((Node->getFlags().hasNoNaNs() ||
+ (DAG.isKnownNeverSNaN(Node->getOperand(0)) &&
+ DAG.isKnownNeverSNaN(Node->getOperand(1)))) &&
+ (Node->getFlags().hasNoSignedZeros() ||
+ DAG.isKnownNeverZeroFloat(Node->getOperand(0)) ||
+ DAG.isKnownNeverZeroFloat(Node->getOperand(1)))) {
+ unsigned IEEE2008Op =
+ Node->getOpcode() == ISD::FMINIMUMNUM ? ISD::FMINNUM : ISD::FMAXNUM;
+ if (isOperationLegalOrCustom(IEEE2008Op, VT))
+ return DAG.getNode(IEEE2008Op, dl, VT, Node->getOperand(0),
+ Node->getOperand(1), Node->getFlags());
+ }
+
+ if (SDValue SelCC = createSelectForFMINIMUMNUM_FMAXIMUMNUM(Node, DAG))
+ return SelCC;
+
+ return SDValue();
+}
+
/// Returns a true value if if this FPClassTest can be performed with an ordered
/// fcmp to 0, and a false value if it's an unordered fcmp to 0. Returns
/// std::nullopt if it cannot be performed as a compare with 0.
diff --git a/llvm/lib/CodeGen/TargetLoweringBase.cpp b/llvm/lib/CodeGen/TargetLoweringBase.cpp
index bc5fc9659469d..b553c12a6cbd6 100644
--- a/llvm/lib/CodeGen/TargetLoweringBase.cpp
+++ b/llvm/lib/CodeGen/TargetLoweringBase.cpp
@@ -713,6 +713,7 @@ void TargetLoweringBase::initActions() {
ISD::FMINNUM, ISD::FMAXNUM,
ISD::FMINNUM_IEEE, ISD::FMAXNUM_IEEE,
ISD::FMINIMUM, ISD::FMAXIMUM,
+ ISD::FMINIMUMNUM, ISD::FMAXIMUMNUM,
ISD::FMAD, ISD::SMIN,
ISD::SMAX, ISD::UMIN,
ISD::UMAX, ISD::ABS,
diff --git a/llvm/test/CodeGen/LoongArch/fp-maximumnum-minimumnum.ll b/llvm/test/CodeGen/LoongArch/fp-maximumnum-minimumnum.ll
new file mode 100644
index 0000000000000..5f86efaa45e99
--- /dev/null
+++ b/llvm/test/CodeGen/LoongArch/fp-maximumnum-minimumnum.ll
@@ -0,0 +1,154 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc --mtriple=loongarch32 --mattr=+f,-d < %s | FileCheck %s --check-prefix=LA32F
+; RUN: llc --mtriple=loongarch32 --mattr=+d < %s | FileCheck %s --check-prefix=LA32D
+; RUN: llc --mtriple=loongarch64 --mattr=+f,-d < %s | FileCheck %s --check-prefix=LA64F
+; RUN: llc --mtriple=loongarch64 --mattr=+d < %s | FileCheck %s --check-prefix=LA64D
+
+declare float @llvm.maximumnum.f32(float, float)
+declare double @llvm.maximumnum.f64(double, double)
+declare float @llvm.minimumnum.f32(float, float)
+declare double @llvm.minimumnum.f64(double, double)
+
+define float @maximumnum_float(float %x, float %y) {
+; LA32F-LABEL: maximumnum_float:
+; LA32F: # %bb.0:
+; LA32F-NEXT: fmax.s $fa1, $fa1, $fa1
+; LA32F-NEXT: fmax.s $fa0, $fa0, $fa0
+; LA32F-NEXT: fmax.s $fa0, $fa0, $fa1
+; LA32F-NEXT: ret
+;
+; LA32D-LABEL: maximumnum_float:
+; LA32D: # %bb.0:
+; LA32D-NEXT: fmax.s $fa1, $fa1, $fa1
+; LA32D-NEXT: fmax.s $fa0, $fa0, $fa0
+; LA32D-NEXT: fmax.s $fa0, $fa0, $fa1
+; LA32D-NEXT: ret
+;
+; LA64F-LABEL: maximumnum_float:
+; LA64F: # %bb.0:
+; LA64F-NEXT: fmax.s $fa1, $fa1, $fa1
+; LA64F-NEXT: fmax.s $fa0, $fa0, $fa0
+; LA64F-NEXT: fmax.s $fa0, $fa0, $fa1
+; LA64F-NEXT: ret
+;
+; LA64D-LABEL: maximumnum_float:
+; LA64D: # %bb.0:
+; LA64D-NEXT: fmax.s $fa1, $fa1, $fa1
+; LA64D-NEXT: fmax.s $fa0, $fa0, $fa0
+; LA64D-NEXT: fmax.s $fa0, $fa0, $fa1
+; LA64D-NEXT: ret
+ %z = call float @llvm.maximumnum.f32(float %x, float %y)
+ ret float %z
+}
+
+define double @maximumnum_double(double %x, double %y) {
+; LA32F-LABEL: maximumnum_double:
+; LA32F: # %bb.0:
+; LA32F-NEXT: addi.w $sp, $sp, -16
+; LA32F-NEXT: .cfi_def_cfa_offset 16
+; LA32F-NEXT: st.w $ra, $sp, 12 # 4-byte Folded Spill
+; LA32F-NEXT: .cfi_offset 1, -4
+; LA32F-NEXT: bl %plt(fmaximum_num)
+; LA32F-NEXT: ld.w $ra, $sp, 12 # 4-byte Folded Reload
+; LA32F-NEXT: addi.w $sp, $sp, 16
+; LA32F-NEXT: ret
+;
+; LA32D-LABEL: maximumnum_double:
+; LA32D: # %bb.0:
+; LA32D-NEXT: fmax.d $fa1, $fa1, $fa1
+; LA32D-NEXT: fmax.d $fa0, $fa0, $fa0
+; LA32D-NEXT: fmax.d $fa0, $fa0, $fa1
+; LA32D-NEXT: ret
+;
+; LA64F-LABEL: maximumnum_double:
+; LA64F: # %bb.0:
+; LA64F-NEXT: addi.d $sp, $sp, -16
+; LA64F-NEXT: .cfi_def_cfa_offset 16
+; LA64F-NEXT: st.d $ra, $sp, 8 # 8-byte Folded Spill
+; LA64F-NEXT: .cfi_offset 1, -8
+; LA64F-NEXT: bl %plt(fmaximum_num)
+; LA64F-NEXT: ld.d $ra, $sp, 8 # 8-byte Folded Reload
+; LA64F-NEXT: addi.d $sp, $sp, 16
+; LA64F-NEXT: ret
+;
+; LA64D-LABEL: maximumnum_double:
+; LA64D: # %bb.0:
+; LA64D-NEXT: fmax.d $fa1, $fa1, $fa1
+; LA64D-NEXT: fmax.d $fa0, $fa0, $fa0
+; LA64D-NEXT: fmax.d $fa0, $fa0, $fa1
+; LA64D-NEXT: ret
+ %z = call double @llvm.maximumnum.f64(double %x, double %y)
+ ret double %z
+}
+
+define float @minimumnum_float(float %x, float %y) {
+; LA32F-LABEL: minimumnum_float:
+; LA32F: # %bb.0:
+; LA32F-NEXT: fmax.s $fa1, $fa1, $fa1
+; LA32F-NEXT: fmax.s $fa0, $fa0, $fa0
+; LA32F-NEXT: fmin.s $fa0, $fa0, $fa1
+; LA32F-NEXT: ret
+;
+; LA32D-LABEL: minimumnum_float:
+; LA32D: # %bb.0:
+; LA32D-NEXT: fmax.s $fa1, $fa1, $fa1
+; LA32D-NEXT: fmax.s $fa0, $fa0, $fa0
+; LA32D-NEXT: fmin.s $fa0, $fa0, $fa1
+; LA32D-NEXT: ret
+;
+; LA64F-LABEL: minimumnum_float:
+; LA64F: # %bb.0:
+; LA64F-NEXT: fmax.s $fa1, $fa1, $fa1
+; LA64F-NEXT: fmax.s $fa0, $fa0, $fa0
+; LA64F-NEXT: fmin.s $fa0, $fa0, $fa1
+; LA64F-NEXT: ret
+;
+; LA64D-LABEL: minimumnum_float:
+; LA64D: # %bb.0:
+; LA64D-NEXT: fmax.s $fa1, $fa1, $fa1
+; LA64D-NEXT: fmax.s $fa0, $fa0, $fa0
+; LA64D-NEXT: fmin.s $fa0, $fa0, $fa1
+; LA64D-NEXT: ret
+ %z = call float @llvm.minimumnum.f32(float %x, float %y)
+ ret float %z
+}
+
+define double @minimumnum_double(double %x, double %y) {
+; LA32F-LABEL: minimumnum_double:
+; LA32F: # %bb.0:
+; LA32F-NEXT: addi.w $sp, $sp, -16
+; LA32F-NEXT: .cfi_def_cfa_offset 16
+; LA32F-NEXT: st.w $ra, $sp, 12 # 4-byte Folded Spill
+; LA32F-NEXT: .cfi_offset 1, -4
+; LA32F-NEXT: bl %plt(fminimum_num)
+; LA32F-NEXT: ld.w $ra, $sp, 12 # 4-byte Folded Reload
+; LA32F-NEXT: addi.w $sp, $sp, 16
+; LA32F-NEXT: ret
+;
+; LA32D-LABEL: minimumnum_double:
+; LA32D: # %bb.0:
+; LA32D-NEXT: fmax.d $fa1, $fa1, $fa1
+; LA32D-NEXT: fmax.d $fa0, $fa0, $fa0
+; LA32D-NEXT: fmin.d $fa0, $fa0, $fa1
+; LA32D-NEXT: ret
+;
+; LA64F-LABEL: minimumnum_double:
+; LA64F: # %bb.0:
+; LA64F-NEXT: addi.d $sp, $sp, -16
+; LA64F-NEXT: .cfi_def_cfa_offset 16
+; LA64F-NEXT: st.d $ra, $sp, 8 # 8-byte Folded Spill
+; LA64F-NEXT: .cfi_offset 1, -8
+; LA64F-NEXT: bl %plt(fminimum_num)
+; LA64F-NEXT: ld.d $ra, $sp, 8 # 8-byte Folded Reload
+; LA64F-NEXT: addi.d $sp, $sp, 16
+; LA64F-NEXT: ret
+;
+; LA64D-LABEL: minimumnum_double:
+; LA64D: # %bb.0:
+; LA64D-NEXT: fmax.d $fa1, $fa1, $fa1
+; LA64D-NEXT: fmax.d $fa0, $fa0, $fa0
+; LA64D-NEXT: fmin.d $fa0, $fa0, $fa1
+; LA64D-NEXT: ret
+ %z = call double @llvm.minimumnum.f64(double %x, double %y)
+ ret double %z
+}
diff --git a/llvm/test/CodeGen/Mips/fp-maximumnum-minimumnum.ll b/llvm/test/CodeGen/Mips/fp-maximumnum-minimumnum.ll
new file mode 100644
index 0000000000000..8ee4070e53434
--- /dev/null
+++ b/llvm/test/CodeGen/Mips/fp-maximumnum-minimumnum.ll
@@ -0,0 +1,100 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc --mtriple=mipsisa32r6 < %s | FileCheck %s --check-prefix=MIPS32R6
+; RUN: llc --mtriple=mips < %s | FileCheck %s --check-prefix=MIPS32
+
+declare float @llvm.maximumnum.f32(float, float)
+declare double @llvm.maximumnum.f64(double, double)
+declare float @llvm.minimumnum.f32(float, float)
+declare double @llvm.minimumnum.f64(double, double)
+
+define float @maximumnum_float(float %x, float %y) {
+; MIPS32R6-LABEL: maximumnum_float:
+; MIPS32R6: # %bb.0:
+; MIPS32R6-NEXT: min.s $f0, $f14, $f14
+; MIPS32R6-NEXT: min.s $f1, $f12, $f12
+; MIPS32R6-NEXT: jr $ra
+; MIPS32R6-NEXT: max.s $f0, $f1, $f0
+;
+; MIPS32-LABEL: maximumnum_float:
+; MIPS32: # %bb.0:
+; MIPS32-NEXT: addiu $sp, $sp, -24
+; MIPS32-NEXT: .cfi_def_cfa_offset 24
+; MIPS32-NEXT: sw $ra, 20($sp) # 4-byte Folded Spill
+; MIPS32-NEXT: .cfi_offset 31, -4
+; MIPS32-NEXT: jal fmaximum_numf
+; MIPS32-NEXT: nop
+; MIPS32-NEXT: lw $ra, 20($sp) # 4-byte Folded Reload
+; MIPS32-NEXT: jr $ra
+; MIPS32-NEXT: addiu $sp, $sp, 24
+ %z = call float @llvm.maximumnum.f32(float %x, float %y)
+ ret float %z
+}
+
+define double @maximumnum_double(double %x, double %y) {
+; MIPS32R6-LABEL: maximumnum_double:
+; MIPS32R6: # %bb.0:
+; MIPS32R6-NEXT: min.d $f0, $f14, $f14
+; MIPS32R6-NEXT: min.d $f1, $f12, $f12
+; MIPS32R6-NEXT: jr $ra
+; MIPS32R6-NEXT: max.d $f0, $f1, $f0
+;
+; MIPS32-LABEL: maximumnum_double:
+; MIPS32: # %bb.0:
+; MIPS32-NEXT: addiu $sp, $sp, -24
+; MIPS32-NEXT: .cfi_def_cfa_offset 24
+; MIPS32-NEXT: sw $ra, 20($sp) # 4-byte Folded Spill
+; MIPS32-NEXT: .cfi_offset 31, -4
+; MIPS32-NEXT: jal fmaximum_num
+; MIPS32-NEXT: nop
+; MIPS32-NEXT: lw $ra, 20($sp) # 4-byte Folded Reload
+; MIPS32-NEXT: jr $ra
+; MIPS32-NEXT: addiu $sp, $sp, 24
+ %z = call double @llvm.maximumnum.f64(double %x, double %y)
+ ret double %z
+}
+
+define float @minimumnum_float(float %x, float %y) {
+; MIPS32R6-LABEL: minimumnum_float:
+; MIPS32R6: # %bb.0:
+; MIPS32R6-NEXT: min.s $f0, $f14, $f14
+; MIPS32R6-NEXT: min.s $f1, $f12, $f12
+; MIPS32R6-NEXT: jr $ra
+; MIPS32R6-NEXT: min.s $f0, $f1, $f0
+;
+; MIPS32-LABEL: minimumnum_float:
+; MIPS32: # %bb.0:
+; MIPS32-NEXT: addiu $sp, $sp, -24
+; MIPS32-NEXT: .cfi_def_cfa_offset 24
+; MIPS32-NEXT: sw $ra, 20($sp) # 4-byte Folded Spill
+; MIPS32-NEXT: .cfi_offset 31, -4
+; MIPS32-NEXT: jal fminimum_numf
+; MIPS32-NEXT: nop
+; MIPS32-NEXT: lw $ra, 20($sp) # 4-byte Folded Reload
+; MIPS32-NEXT: jr $ra
+; MIPS32-NEXT: addiu $sp, $sp, 24
+ %z = call float @llvm.minimumnum.f32(float %x, float %y)
+ ret float %z
+}
+
+define double @minimumnum_double(double %x, double %y) {
+; MIPS32R6-LABEL: minimumnum_double:
+; MIPS32R6: # %bb.0:
+; MIPS32R6-NEXT: min.d $f0, $f14, $f14
+; MIPS32R6-NEXT: min.d $f1, $f12, $f12
+; MIPS32R6-NEXT: jr $ra
+; MIPS32R6-NEXT: min.d $f0, $f1, $f0
+;
+; MIPS32-LABEL: minimumnum_double:
+; MIPS32: # %bb.0:
+; MIPS32-NEXT: addiu $sp, $sp, -24
+; MIPS32-NEXT: .cfi_def_cfa_offset 24
+; MIPS32-NEXT: sw $ra, 20($sp) # 4-byte Folded Spill
+; MIPS32-NEXT: .cfi_offset 31, -4
+; MIPS32-NEXT: jal fminimum_num
+; MIPS32-NEXT: nop
+; MIPS32-NEXT: lw $ra, 20($sp) # 4-byte Folded Reload
+; MIPS32-NEXT: jr $ra
+; MIPS32-NEXT: addiu $sp, $sp, 24
+ %z = call double @llvm.minimumnum.f64(double %x, double %y)
+ ret double %z
+}
diff --git a/llvm/test/tools/llvm-tli-checker/ps4-tli-check.yaml b/llvm/test/tools/llvm-tli-checker/ps4-tli-check.yaml
index 5840e024e9786..f672a51c1fd00 100644
--- a/llvm/test/tools/llvm-tli-checker/ps4-tli-check.yaml
+++ b/llvm/test/tools/llvm-tli-checker/ps4-tli-check.yaml
@@ -32,14 +32,20 @@
# RUN: FileCheck %s --check-prefix=AVAIL --input-file %t3.txt
# RUN: FileCheck %s --check-prefix=UNAVAIL --input-file %t3.txt
#
-# CHECK: << Total TLI yes SDK no: 8
+# CHECK: << Total TLI yes SDK no: 14
# CHECK: >> Total TLI no SDK yes: 0
# CHECK: == Total TLI yes SDK yes: 242
#
# WRONG_DETAIL: << TLI yes SDK no : '_ZdaPv' aka operator delete[](void*)
# WRONG_DETAIL: >> TLI no SDK yes: '_ZdaPvj' aka operator delete[](void*, unsigned int)
# WRONG_DETAIL-COUNT-8: << TLI yes SDK no : {{.*}}__hot_cold_t
-# WRONG_SUMMARY: << Total TLI yes SDK no: 9{{$}}
+# WRONG_DETAIL: << TLI yes SDK no : 'fmaximum_num'
+# WRONG_DETAIL: << TLI yes SDK no : 'fmaximum_numf'
+# WRONG_DETAIL: << TLI yes SDK no : 'fmaximum_numl'
+# WRONG_DETAIL: << TLI yes SDK no : 'fminimum_num'
+# WRONG_DETAIL: << TLI yes SDK no : 'fminimum_numf'
+# WRONG_DETAIL: << TLI yes SDK no : 'fminimum_numl'
+# WRONG_SUMMARY: << Total TLI yes SDK no: 15{{$}}
# WRONG_SUMMARY: >> Total TLI no SDK yes: 1{{$}}
# WRONG_SUMMARY: == Total TLI yes SDK yes: 241
#
@@ -47,8 +53,8 @@
## the exact count first; the two directives should add up to that.
## Yes, this means additions to TLI will fail this test, but the argument
## to -COUNT can't be an expression.
-# AVAIL: TLI knows 483 symbols, 250 available
-# AVAIL-COUNT-250: {{^}} available
+# AVAIL: TLI knows 486 symbols, 253 available
+# AVAIL-COUNT-253: {{^}} available
# AVAIL-NOT: {{^}} available
# UNAVAIL-COUNT-233: not available
# UNAVAIL-NOT: not available
diff --git a/llvm/unittests/Analysis/TargetLibraryInfoTest.cpp b/llvm/unittests/Analysis/TargetLibraryInfoTest.cpp
index 1447291844dd2..4c66a48ef117d 100644
--- a/llvm/unittests/Analysis/TargetLibraryInfoTest.cpp
+++ b/llvm/unittests/Analysis/TargetLibraryInfoTest.cpp
@@ -188,6 +188,12 @@ TEST_F(TargetLibraryInfoTest, ValidProto) {
"declare double @fmin(double, double)\n"
"declare float @fminf(float, float)\n"
"declare x86_fp80 @fminl(x86_fp80, x86_fp80)\n"
+ "declare double @fmaximum_num(double, double)\n"
+ "declare float @fmaximum_numf(float, float)\n"
+ "declare x86_fp80 @fmaximum_numl(x86_fp80, x86_fp80)\n"
+ "declare double @fminimum_num(double, double)\n"
+ "declare float @fminimum_numf(float, float)\n"
+ "declare x86_fp80 @fminimum_numl(x86_fp80, x86_fp80)\n"
"declare double @fmod(double, double)\n"
"declare float @fmodf(float, float)\n"
"declare x86_fp80 @fmodl(x86_fp80, x86_fp80)\n"
@@ -663,4 +669,4 @@ class TLITestAarch64 : public ::testing::Test {
TEST_F(TLITestAarch64, TestFrem) {
EXPECT_EQ(getScalarName(Instruction::FRem, Type::getDoubleTy(Ctx)), "fmod");
EXPECT_EQ(getScalarName(Instruction::FRem, Type::getFloatTy(Ctx)), "fmodf");
-}
\ No newline at end of file
+}
>From 4b4d97b71167f8ed1750d96091084d115f233586 Mon Sep 17 00:00:00 2001
From: YunQiang Su <syq at gcc.gnu.org>
Date: Thu, 4 Jul 2024 10:44:23 +0800
Subject: [PATCH 2/8] Remove MINNUM_IEEE in LangRef
---
llvm/docs/LangRef.rst | 4 ----
1 file changed, 4 deletions(-)
diff --git a/llvm/docs/LangRef.rst b/llvm/docs/LangRef.rst
index 61a08f924955e..c2ab4f301f1ac 100644
--- a/llvm/docs/LangRef.rst
+++ b/llvm/docs/LangRef.rst
@@ -16152,10 +16152,6 @@ LLVM implements all ISO C flavors as listed in this table.
Only basic intrinsics list here. The constrained version
ones may have different behaivor on exception.
-Since some architectures implement minNum/maxNum with +0.0>-0.0,
-so we define internal ISD::MINNUM_IEEE and ISD::MAXNUM_IEEE.
-They will be helpful to implement minimumnum/maximumnum.
-
.. list-table::
:header-rows: 1
:widths: 16 28 28 28
>From 9727acb44e8bc06b20760201b81ccdb8a052f949 Mon Sep 17 00:00:00 2001
From: YunQiang Su <syq at gcc.gnu.org>
Date: Thu, 4 Jul 2024 10:44:43 +0800
Subject: [PATCH 3/8] fix code style
---
llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
index 209c23eab1dda..8bcdfa39fc82c 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
@@ -203,8 +203,8 @@ std::string SDNode::getOperationName(const SelectionDAG *G) const {
case ISD::STRICT_FMINIMUM: return "strict_fminimum";
case ISD::FMAXIMUM: return "fmaximum";
case ISD::STRICT_FMAXIMUM: return "strict_fmaximum";
- case ISD::FMINIMUMNUM: return "fminimumnum";
- case ISD::FMAXIMUMNUM: return "fmaximumnum";
+ case ISD::FMINIMUMNUM: return "fminimumnum";
+ case ISD::FMAXIMUMNUM: return "fmaximumnum";
case ISD::FNEG: return "fneg";
case ISD::FSQRT: return "fsqrt";
case ISD::STRICT_FSQRT: return "strict_fsqrt";
>From fc39e30defc233eab04fd9693935ec1acc1c7c47 Mon Sep 17 00:00:00 2001
From: YunQiang Su <syq at gcc.gnu.org>
Date: Thu, 4 Jul 2024 15:00:57 +0800
Subject: [PATCH 4/8] remove createSelectForFMINIMUMNUM_FMAXIMUMNUM
---
llvm/include/llvm/CodeGen/TargetLowering.h | 2 --
.../SelectionDAG/LegalizeFloatTypes.cpp | 4 ---
.../CodeGen/SelectionDAG/TargetLowering.cpp | 28 ++-----------------
3 files changed, 2 insertions(+), 32 deletions(-)
diff --git a/llvm/include/llvm/CodeGen/TargetLowering.h b/llvm/include/llvm/CodeGen/TargetLowering.h
index 1e08bd86af372..140a47ad471c1 100644
--- a/llvm/include/llvm/CodeGen/TargetLowering.h
+++ b/llvm/include/llvm/CodeGen/TargetLowering.h
@@ -5141,8 +5141,6 @@ class TargetLowering : public TargetLoweringBase {
/// through to the default expansion/soften to libcall, we might introduce a
/// link-time dependency on libm into a file that originally did not have one.
SDValue createSelectForFMINNUM_FMAXNUM(SDNode *Node, SelectionDAG &DAG) const;
- SDValue createSelectForFMINIMUMNUM_FMAXIMUMNUM(SDNode *Node,
- SelectionDAG &DAG) const;
/// Return a reciprocal estimate value for the input operand.
/// \p Enabled is a ReciprocalEstimate enum with value either 'Unspecified' or
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
index 5a3c70ec4cdab..9109b60c70dcc 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
@@ -326,8 +326,6 @@ SDValue DAGTypeLegalizer::SoftenFloatRes_FMAXNUM(SDNode *N) {
}
SDValue DAGTypeLegalizer::SoftenFloatRes_FMINIMUMNUM(SDNode *N) {
- if (SDValue SelCC = TLI.createSelectForFMINIMUMNUM_FMAXIMUMNUM(N, DAG))
- return SoftenFloatRes_SELECT_CC(SelCC.getNode());
return SoftenFloatRes_Binary(
N, GetFPLibCall(N->getValueType(0), RTLIB::FMINIMUMNUM_F32,
RTLIB::FMINIMUMNUM_F64, RTLIB::FMINIMUMNUM_F80,
@@ -335,8 +333,6 @@ SDValue DAGTypeLegalizer::SoftenFloatRes_FMINIMUMNUM(SDNode *N) {
}
SDValue DAGTypeLegalizer::SoftenFloatRes_FMAXIMUMNUM(SDNode *N) {
- if (SDValue SelCC = TLI.createSelectForFMINIMUMNUM_FMAXIMUMNUM(N, DAG))
- return SoftenFloatRes_SELECT_CC(SelCC.getNode());
return SoftenFloatRes_Binary(
N, GetFPLibCall(N->getValueType(0), RTLIB::FMAXIMUMNUM_F32,
RTLIB::FMAXIMUMNUM_F64, RTLIB::FMAXIMUMNUM_F80,
diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index b97554a4d0dc4..06b6f774bb856 100644
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -8531,28 +8531,6 @@ SDValue TargetLowering::expandFMINIMUM_FMAXIMUM(SDNode *N,
return MinMax;
}
-SDValue TargetLowering::createSelectForFMINIMUMNUM_FMAXIMUMNUM(
- SDNode *Node, SelectionDAG &DAG) const {
- unsigned Opcode = Node->getOpcode();
- assert((Opcode == ISD::FMINIMUMNUM || Opcode == ISD::FMAXIMUMNUM) &&
- "Wrong opcode");
-
- if (Node->getFlags().hasNoNaNs()) {
- ISD::CondCode Pred = Opcode == ISD::FMINIMUMNUM ? ISD::SETLT : ISD::SETGT;
- SDValue Op1 = Node->getOperand(0);
- SDValue Op2 = Node->getOperand(1);
- SDValue SelCC = DAG.getSelectCC(SDLoc(Node), Op1, Op2, Op1, Op2, Pred);
- // Copy FMF flags, but always set the no-signed-zeros flag
- // as this is implied by the FMINNUM/FMAXNUM semantics.
- SDNodeFlags Flags = Node->getFlags();
- Flags.setNoSignedZeros(true);
- SelCC->setFlags(Flags);
- return SelCC;
- }
-
- return SDValue();
-}
-
SDValue TargetLowering::expandFMINIMUMNUM_FMAXIMUMNUM(SDNode *Node,
SelectionDAG &DAG) const {
SDLoc dl(Node);
@@ -8596,7 +8574,7 @@ SDValue TargetLowering::expandFMINIMUMNUM_FMAXIMUMNUM(SDNode *Node,
Node->getOperand(1), Node->getFlags());
}
- // FMINNUM/FMAXIMUM returns qNaN if either operand is sNaN, and it may return
+ // FMINNUM/FMAXMUM returns qNaN if either operand is sNaN, and it may return
// either one for +0.0 vs -0.0.
// FIXME: maybe we need hasNoSNaNs().
if ((Node->getFlags().hasNoNaNs() ||
@@ -8612,9 +8590,7 @@ SDValue TargetLowering::expandFMINIMUMNUM_FMAXIMUMNUM(SDNode *Node,
Node->getOperand(1), Node->getFlags());
}
- if (SDValue SelCC = createSelectForFMINIMUMNUM_FMAXIMUMNUM(Node, DAG))
- return SelCC;
-
+ // FIXME: we may use setCC/getSELECT to optimize it instead of fallback to libcall.
return SDValue();
}
>From fa0ecf3381a31287e6058d3f0a83da451c3a7591 Mon Sep 17 00:00:00 2001
From: YunQiang Su <syq at gcc.gnu.org>
Date: Thu, 4 Jul 2024 16:49:20 +0800
Subject: [PATCH 5/8] fix code style
---
llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 06b6f774bb856..5c79c1c512480 100644
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -8590,7 +8590,8 @@ SDValue TargetLowering::expandFMINIMUMNUM_FMAXIMUMNUM(SDNode *Node,
Node->getOperand(1), Node->getFlags());
}
- // FIXME: we may use setCC/getSELECT to optimize it instead of fallback to libcall.
+ // FIXME: we may use setCC/getSELECT to optimize it instead of fallback to
+ // libcall.
return SDValue();
}
>From 875223c97e37af1b985454a783eb451bb90ebeb3 Mon Sep 17 00:00:00 2001
From: YunQiang Su <syq at gcc.gnu.org>
Date: Wed, 24 Jul 2024 23:34:52 +0800
Subject: [PATCH 6/8] merge ps4-tli-check.yaml
---
llvm/test/tools/llvm-tli-checker/ps4-tli-check.yaml | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/llvm/test/tools/llvm-tli-checker/ps4-tli-check.yaml b/llvm/test/tools/llvm-tli-checker/ps4-tli-check.yaml
index f672a51c1fd00..cea667b19747d 100644
--- a/llvm/test/tools/llvm-tli-checker/ps4-tli-check.yaml
+++ b/llvm/test/tools/llvm-tli-checker/ps4-tli-check.yaml
@@ -53,8 +53,8 @@
## the exact count first; the two directives should add up to that.
## Yes, this means additions to TLI will fail this test, but the argument
## to -COUNT can't be an expression.
-# AVAIL: TLI knows 486 symbols, 253 available
-# AVAIL-COUNT-253: {{^}} available
+# AVAIL: TLI knows 489 symbols, 256 available
+# AVAIL-COUNT-256: {{^}} available
# AVAIL-NOT: {{^}} available
# UNAVAIL-COUNT-233: not available
# UNAVAIL-NOT: not available
>From a4e20c0ec0c4acd91bce4c9574dc4331e16265b6 Mon Sep 17 00:00:00 2001
From: YunQiang Su <syq at gcc.gnu.org>
Date: Thu, 25 Jul 2024 08:49:15 +0800
Subject: [PATCH 7/8] improve LangRef.rst
---
llvm/docs/LangRef.rst | 10 +++++-----
1 file changed, 5 insertions(+), 5 deletions(-)
diff --git a/llvm/docs/LangRef.rst b/llvm/docs/LangRef.rst
index c2ab4f301f1ac..51870bff2aef1 100644
--- a/llvm/docs/LangRef.rst
+++ b/llvm/docs/LangRef.rst
@@ -16148,9 +16148,9 @@ on working with qNaN/sNaN and +0.0/-0.0. Here is the list:
LLVM Implementation:
""""""""""""""""""""
-LLVM implements all ISO C flavors as listed in this table.
-Only basic intrinsics list here. The constrained version
-ones may have different behaivor on exception.
+LLVM implements all ISO C flavors as listed in this table, except in the
+default floating-point environment exceptions are ignored. The constrained
+versions of the intrinsics respect the exception behavior.
.. list-table::
:header-rows: 1
@@ -16414,7 +16414,7 @@ be less than +0.0 for this intrinsic.
Note that these are the semantics of minimumNumber specified in IEEE 754-2019.
-It has some different with '``llvm.minnum.*``':
+It has some differences with '``llvm.minnum.*``':
1)'``llvm.minnum.*``' will return qNaN if either operand is sNaN.
2)'``llvm.minnum*``' may return either one if we compare +0.0 vs -0.0.
@@ -16460,7 +16460,7 @@ be less than +0.0 for this intrinsic.
Note that these are the semantics of maximumNumber specified in IEEE 754-2019.
-It has some different with '``llvm.maxnum.*``':
+It has some differences with '``llvm.maxnum.*``':
1)'``llvm.maxnum.*``' will return qNaN if either operand is sNaN.
2)'``llvm.maxnum*``' may return either one if we compare +0.0 vs -0.0.
>From 3c884f0451c09e3452ab6d029fb4ffffa12aff57 Mon Sep 17 00:00:00 2001
From: YunQiang Su <syq at gcc.gnu.org>
Date: Thu, 25 Jul 2024 15:15:35 +0800
Subject: [PATCH 8/8] Use SetCC/SelectCC as final fallback
It seems exposing a bug of RISC-V backend:
```
define dso_local float @f(float noundef %a, float noundef %b) local_unnamed_addr #0 {
entry:
%0 = tail call float @llvm.minimumnum.f32(float %a, float %b)
ret float %0
}
```
```
./bin/clang --target=riscv64-linux-gnu -S yy.ll -O2 -o -
li a0, 1
bnez a0, .LBB0_2
```
---
.../CodeGen/SelectionDAG/TargetLowering.cpp | 96 ++++++++++++-------
1 file changed, 60 insertions(+), 36 deletions(-)
diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 5c79c1c512480..3a5bdf7a4029c 100644
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -8533,66 +8533,90 @@ SDValue TargetLowering::expandFMINIMUM_FMAXIMUM(SDNode *N,
SDValue TargetLowering::expandFMINIMUMNUM_FMAXIMUMNUM(SDNode *Node,
SelectionDAG &DAG) const {
- SDLoc dl(Node);
- unsigned NewOp = Node->getOpcode() == ISD::FMINIMUMNUM ? ISD::FMINNUM_IEEE
- : ISD::FMAXNUM_IEEE;
+ SDLoc DL(Node);
+ SDValue LHS = Node->getOperand(0);
+ SDValue RHS = Node->getOperand(1);
+ unsigned Opc = Node->getOpcode();
EVT VT = Node->getValueType(0);
+ EVT CCVT = getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(), VT);
+ bool IsMax = Opc == ISD::FMAXIMUMNUM;
+ SDNodeFlags Flags = Node->getFlags();
- if (VT.isScalableVector())
- report_fatal_error(
- "Expanding fminimumnum/fmaximumnum for scalable vectors is undefined.");
+ unsigned NewOp =
+ Opc == ISD::FMINIMUMNUM ? ISD::FMINNUM_IEEE : ISD::FMAXNUM_IEEE;
if (isOperationLegalOrCustom(NewOp, VT)) {
- SDValue Quiet0 = Node->getOperand(0);
- SDValue Quiet1 = Node->getOperand(1);
-
- if (!Node->getFlags().hasNoNaNs()) {
+ if (!Flags.hasNoNaNs()) {
// Insert canonicalizes if it's possible we need to quiet to get correct
// sNaN behavior.
- if (!DAG.isKnownNeverSNaN(Quiet0)) {
- Quiet0 =
- DAG.getNode(ISD::FCANONICALIZE, dl, VT, Quiet0, Node->getFlags());
+ if (!DAG.isKnownNeverSNaN(LHS)) {
+ LHS = DAG.getNode(ISD::FCANONICALIZE, DL, VT, LHS, Node->getFlags());
}
- if (!DAG.isKnownNeverSNaN(Quiet1)) {
- Quiet1 =
- DAG.getNode(ISD::FCANONICALIZE, dl, VT, Quiet1, Node->getFlags());
+ if (!DAG.isKnownNeverSNaN(RHS)) {
+ RHS = DAG.getNode(ISD::FCANONICALIZE, DL, VT, RHS, Node->getFlags());
}
}
- return DAG.getNode(NewOp, dl, VT, Quiet0, Quiet1, Node->getFlags());
+ return DAG.getNode(NewOp, DL, VT, LHS, RHS, Flags);
}
// FMINIMUM/FMAXIMUM always return NaN if either operand is NaN.
// It has same behavior about +0.0 vs -0.0.
- if (Node->getFlags().hasNoNaNs() ||
- (DAG.isKnownNeverNaN(Node->getOperand(0)) &&
- DAG.isKnownNeverNaN(Node->getOperand(1)))) {
+ if (Flags.hasNoNaNs() ||
+ (DAG.isKnownNeverNaN(LHS) && DAG.isKnownNeverNaN(RHS))) {
unsigned IEEE2019Op =
- Node->getOpcode() == ISD::FMINIMUMNUM ? ISD::FMINIMUM : ISD::FMAXIMUM;
+ Opc == ISD::FMINIMUMNUM ? ISD::FMINIMUM : ISD::FMAXIMUM;
if (isOperationLegalOrCustom(IEEE2019Op, VT))
- return DAG.getNode(IEEE2019Op, dl, VT, Node->getOperand(0),
- Node->getOperand(1), Node->getFlags());
+ return DAG.getNode(IEEE2019Op, DL, VT, LHS, RHS, Flags);
}
// FMINNUM/FMAXMUM returns qNaN if either operand is sNaN, and it may return
// either one for +0.0 vs -0.0.
// FIXME: maybe we need hasNoSNaNs().
- if ((Node->getFlags().hasNoNaNs() ||
- (DAG.isKnownNeverSNaN(Node->getOperand(0)) &&
- DAG.isKnownNeverSNaN(Node->getOperand(1)))) &&
- (Node->getFlags().hasNoSignedZeros() ||
- DAG.isKnownNeverZeroFloat(Node->getOperand(0)) ||
- DAG.isKnownNeverZeroFloat(Node->getOperand(1)))) {
- unsigned IEEE2008Op =
- Node->getOpcode() == ISD::FMINIMUMNUM ? ISD::FMINNUM : ISD::FMAXNUM;
+ if ((Flags.hasNoNaNs() ||
+ (DAG.isKnownNeverSNaN(LHS) && DAG.isKnownNeverSNaN(RHS))) &&
+ (Flags.hasNoSignedZeros() || DAG.isKnownNeverZeroFloat(LHS) ||
+ DAG.isKnownNeverZeroFloat(RHS))) {
+ unsigned IEEE2008Op = Opc == ISD::FMINIMUMNUM ? ISD::FMINNUM : ISD::FMAXNUM;
if (isOperationLegalOrCustom(IEEE2008Op, VT))
- return DAG.getNode(IEEE2008Op, dl, VT, Node->getOperand(0),
- Node->getOperand(1), Node->getFlags());
+ return DAG.getNode(IEEE2008Op, DL, VT, LHS, RHS, Flags);
}
- // FIXME: we may use setCC/getSELECT to optimize it instead of fallback to
- // libcall.
- return SDValue();
+ SDValue MinMax;
+ // If one operand is NaN, let's move another value to it.
+ // So that if only one operand is NaN, we can have both two operands are
+ // non-NaN now.
+ SDValue LHSCompareLHS = DAG.getSetCC(DL, CCVT, LHS, LHS, ISD::SETEQ);
+ LHS = DAG.getSelect(DL, VT, LHSCompareLHS, LHS, RHS, Flags);
+ SDValue RHSCompareRHS = DAG.getSetCC(DL, CCVT, RHS, RHS, ISD::SETEQ);
+ RHS = DAG.getSelect(DL, VT, RHSCompareRHS, RHS, LHS, Flags);
+
+ SDValue Compare =
+ DAG.getSetCC(DL, CCVT, LHS, RHS, IsMax ? ISD::SETGT : ISD::SETLT);
+ MinMax = DAG.getSelect(DL, VT, Compare, LHS, RHS, Flags);
+
+ // If MinMax is NaN, let's quiet it with MinMax = MinMax + 0.
+ SDValue MinMaxCompareMinMax =
+ DAG.getSetCC(DL, CCVT, MinMax, MinMax, ISD::SETEQ);
+ SDValue MinMaxQuiet = DAG.getNode(ISD::FADD, DL, VT, MinMax,
+ DAG.getConstantFP(0.0, DL, VT), Flags);
+ MinMax =
+ DAG.getSelect(DL, VT, MinMaxCompareMinMax, MinMax, MinMaxQuiet, Flags);
+
+ // Let's process +0.0 vs -0.0.
+ SDValue TestZero =
+ DAG.getTargetConstant(IsMax ? fcPosZero : fcNegZero, DL, MVT::i32);
+ SDValue IsZero = DAG.getSetCC(DL, CCVT, MinMax,
+ DAG.getConstantFP(0.0, DL, VT), ISD::SETEQ);
+ SDValue LCmp = DAG.getSelect(
+ DL, VT, DAG.getNode(ISD::IS_FPCLASS, DL, CCVT, LHS, TestZero), LHS,
+ MinMax, Flags);
+ SDValue RCmp = DAG.getSelect(
+ DL, VT, DAG.getNode(ISD::IS_FPCLASS, DL, CCVT, RHS, TestZero), RHS, LCmp,
+ Flags);
+ MinMax = DAG.getSelect(DL, VT, IsZero, RCmp, MinMax, Flags);
+
+ return MinMax;
}
/// Returns a true value if if this FPClassTest can be performed with an ordered
More information about the llvm-commits
mailing list