[Power9] Add __float128 support for compare operations

Added handling for the select f128.

Differential Revision: https://reviews.llvm.org/D48294

llvm-svn: 336548

llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp

@@ -3617,9 +3617,12 @@ SDValue PPCDAGToDAGISel::SelectCC(SDValue LHS, SDValue RHS, ISD::CondCode CC,
     }
   } else if (LHS.getValueType() == MVT::f32) {
     Opc = PPC::FCMPUS;
-  } else {
+  } else if (LHS.getValueType() == MVT::f64) {
-    assert(LHS.getValueType() == MVT::f64 && "Unknown vt!");
     Opc = PPCSubTarget->hasVSX() ? PPC::XSCMPUDP : PPC::FCMPUD;
+  } else {
+    assert(LHS.getValueType() == MVT::f128 && "Unknown vt!");
+    assert(PPCSubTarget->hasVSX() && "__float128 requires VSX");
+    Opc = PPC::XSCMPUQP;
   }
   return SDValue(CurDAG->getMachineNode(Opc, dl, MVT::i32, LHS, RHS), 0);
 }
@@ -4564,6 +4567,8 @@ void PPCDAGToDAGISel::Select(SDNode *N) {
         SelectCCOp = PPC::SELECT_CC_VSFRC;
       else
         SelectCCOp = PPC::SELECT_CC_F8;
+    else if (N->getValueType(0) == MVT::f128)
+      SelectCCOp = PPC::SELECT_CC_F16;
     else if (PPCSubTarget->hasQPX() && N->getValueType(0) == MVT::v4f64)
       SelectCCOp = PPC::SELECT_CC_QFRC;
     else if (PPCSubTarget->hasQPX() && N->getValueType(0) == MVT::v4f32)

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

@@ -810,6 +810,13 @@ PPCTargetLowering::PPCTargetLowering(const PPCTargetMachine &TM,
         for (MVT FPT : MVT::fp_valuetypes())
           setLoadExtAction(ISD::EXTLOAD, MVT::f128, FPT, Expand);
         setOperationAction(ISD::FMA, MVT::f128, Legal);
+        setCondCodeAction(ISD::SETULT, MVT::f128, Expand);
+        setCondCodeAction(ISD::SETUGT, MVT::f128, Expand);
+        setCondCodeAction(ISD::SETUEQ, MVT::f128, Expand);
+        setCondCodeAction(ISD::SETOGE, MVT::f128, Expand);
+        setCondCodeAction(ISD::SETOLE, MVT::f128, Expand);
+        setCondCodeAction(ISD::SETONE, MVT::f128, Expand);
+        setOperationAction(ISD::SELECT, MVT::f128, Expand);
         setOperationAction(ISD::FP_ROUND, MVT::f64, Legal);
         setOperationAction(ISD::FP_ROUND, MVT::f32, Legal);
         setTruncStoreAction(MVT::f128, MVT::f64, Expand);
@@ -10206,6 +10213,7 @@ PPCTargetLowering::EmitInstrWithCustomInserter(MachineInstr &MI,
              MI.getOpcode() == PPC::SELECT_CC_I8 ||
              MI.getOpcode() == PPC::SELECT_CC_F4 ||
              MI.getOpcode() == PPC::SELECT_CC_F8 ||
+             MI.getOpcode() == PPC::SELECT_CC_F16 ||
              MI.getOpcode() == PPC::SELECT_CC_QFRC ||
              MI.getOpcode() == PPC::SELECT_CC_QSRC ||
              MI.getOpcode() == PPC::SELECT_CC_QBRC ||
@@ -10217,6 +10225,7 @@ PPCTargetLowering::EmitInstrWithCustomInserter(MachineInstr &MI,
              MI.getOpcode() == PPC::SELECT_I8 ||
              MI.getOpcode() == PPC::SELECT_F4 ||
              MI.getOpcode() == PPC::SELECT_F8 ||
+             MI.getOpcode() == PPC::SELECT_F16 ||
              MI.getOpcode() == PPC::SELECT_QFRC ||
              MI.getOpcode() == PPC::SELECT_QSRC ||
              MI.getOpcode() == PPC::SELECT_QBRC ||
@@ -10252,6 +10261,7 @@ PPCTargetLowering::EmitInstrWithCustomInserter(MachineInstr &MI,
 
     if (MI.getOpcode() == PPC::SELECT_I4 || MI.getOpcode() == PPC::SELECT_I8 ||
         MI.getOpcode() == PPC::SELECT_F4 || MI.getOpcode() == PPC::SELECT_F8 ||
+        MI.getOpcode() == PPC::SELECT_F16 ||
         MI.getOpcode() == PPC::SELECT_QFRC ||
         MI.getOpcode() == PPC::SELECT_QSRC ||
         MI.getOpcode() == PPC::SELECT_QBRC ||

llvm/lib/Target/PowerPC/PPCInstrInfo.td

@@ -1216,6 +1216,9 @@ let usesCustomInserter = 1,    // Expanded after instruction selection.
   def SELECT_CC_F8  : Pseudo<(outs f8rc:$dst), (ins crrc:$cond, f8rc:$T, f8rc:$F,
                               i32imm:$BROPC), "#SELECT_CC_F8",
                               []>;
+  def SELECT_CC_F16  : Pseudo<(outs vrrc:$dst), (ins crrc:$cond, vrrc:$T, vrrc:$F,
+                              i32imm:$BROPC), "#SELECT_CC_F16",
+                              []>;
   def SELECT_CC_VRRC: Pseudo<(outs vrrc:$dst), (ins crrc:$cond, vrrc:$T, vrrc:$F,
                               i32imm:$BROPC), "#SELECT_CC_VRRC",
                               []>;
@@ -1234,6 +1237,9 @@ let usesCustomInserter = 1,    // Expanded after instruction selection.
   def SELECT_F8  : Pseudo<(outs f8rc:$dst), (ins crbitrc:$cond,
                           f8rc:$T, f8rc:$F), "#SELECT_F8",
                           [(set f64:$dst, (select i1:$cond, f64:$T, f64:$F))]>;
+  def SELECT_F16  : Pseudo<(outs vrrc:$dst), (ins crbitrc:$cond,
+                          vrrc:$T, vrrc:$F), "#SELECT_F16",
+                          [(set f128:$dst, (select i1:$cond, f128:$T, f128:$F))]>;
   def SELECT_VRRC: Pseudo<(outs vrrc:$dst), (ins crbitrc:$cond,
                           vrrc:$T, vrrc:$F), "#SELECT_VRRC",
                           [(set v4i32:$dst,
@@ -3622,6 +3628,37 @@ defm : CRNotPat<(i1 (setcc f64:$s1, f64:$s2, SETNE)),
 defm : CRNotPat<(i1 (setcc f64:$s1, f64:$s2, SETO)),
                 (EXTRACT_SUBREG (FCMPUD $s1, $s2), sub_un)>;
 
+// SETCC for f128.
+def : Pat<(i1 (setcc f128:$s1, f128:$s2, SETOLT)),
+          (EXTRACT_SUBREG (XSCMPUQP $s1, $s2), sub_lt)>;
+def : Pat<(i1 (setcc f128:$s1, f128:$s2, SETLT)),
+          (EXTRACT_SUBREG (XSCMPUQP $s1, $s2), sub_lt)>;
+def : Pat<(i1 (setcc f128:$s1, f128:$s2, SETOGT)),
+          (EXTRACT_SUBREG (XSCMPUQP $s1, $s2), sub_gt)>;
+def : Pat<(i1 (setcc f128:$s1, f128:$s2, SETGT)),
+          (EXTRACT_SUBREG (XSCMPUQP $s1, $s2), sub_gt)>;
+def : Pat<(i1 (setcc f128:$s1, f128:$s2, SETOEQ)),
+          (EXTRACT_SUBREG (XSCMPUQP $s1, $s2), sub_eq)>;
+def : Pat<(i1 (setcc f128:$s1, f128:$s2, SETEQ)),
+          (EXTRACT_SUBREG (XSCMPUQP $s1, $s2), sub_eq)>;
+def : Pat<(i1 (setcc f128:$s1, f128:$s2, SETUO)),
+          (EXTRACT_SUBREG (XSCMPUQP $s1, $s2), sub_un)>;
+
+defm : CRNotPat<(i1 (setcc f128:$s1, f128:$s2, SETUGE)),
+                (EXTRACT_SUBREG (XSCMPUQP $s1, $s2), sub_lt)>;
+defm : CRNotPat<(i1 (setcc f128:$s1, f128:$s2, SETGE)),
+                (EXTRACT_SUBREG (XSCMPUQP $s1, $s2), sub_lt)>;
+defm : CRNotPat<(i1 (setcc f128:$s1, f128:$s2, SETULE)),
+                (EXTRACT_SUBREG (XSCMPUQP $s1, $s2), sub_gt)>;
+defm : CRNotPat<(i1 (setcc f128:$s1, f128:$s2, SETLE)),
+                (EXTRACT_SUBREG (XSCMPUQP $s1, $s2), sub_gt)>;
+defm : CRNotPat<(i1 (setcc f128:$s1, f128:$s2, SETUNE)),
+                (EXTRACT_SUBREG (XSCMPUQP $s1, $s2), sub_eq)>;
+defm : CRNotPat<(i1 (setcc f128:$s1, f128:$s2, SETNE)),
+                (EXTRACT_SUBREG (XSCMPUQP $s1, $s2), sub_eq)>;
+defm : CRNotPat<(i1 (setcc f128:$s1, f128:$s2, SETO)),
+                (EXTRACT_SUBREG (XSCMPUQP $s1, $s2), sub_un)>;
+
 // match select on i1 variables:
 def : Pat<(i1 (select i1:$cond, i1:$tval, i1:$fval)),
           (CROR (CRAND        $cond , $tval),
@@ -3746,6 +3783,27 @@ def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETUGT)),
 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETNE)),
           (SELECT_F8 (CRXOR $lhs, $rhs), $tval, $fval)>;
 
+def : Pat<(f128 (selectcc i1:$lhs, i1:$rhs, f128:$tval, f128:$fval, SETLT)),
+          (SELECT_F16 (CRANDC $lhs, $rhs), $tval, $fval)>;
+def : Pat<(f128 (selectcc i1:$lhs, i1:$rhs, f128:$tval, f128:$fval, SETULT)),
+          (SELECT_F16 (CRANDC $rhs, $lhs), $tval, $fval)>;
+def : Pat<(f128 (selectcc i1:$lhs, i1:$rhs, f128:$tval, f128:$fval, SETLE)),
+          (SELECT_F16 (CRORC  $lhs, $rhs), $tval, $fval)>;
+def : Pat<(f128 (selectcc i1:$lhs, i1:$rhs, f128:$tval, f128:$fval, SETULE)),
+          (SELECT_F16 (CRORC  $rhs, $lhs), $tval, $fval)>;
+def : Pat<(f128 (selectcc i1:$lhs, i1:$rhs, f128:$tval, f128:$fval, SETEQ)),
+          (SELECT_F16 (CREQV $lhs, $rhs), $tval, $fval)>;
+def : Pat<(f128 (selectcc i1:$lhs, i1:$rhs, f128:$tval, f128:$fval, SETGE)),
+         (SELECT_F16 (CRORC  $rhs, $lhs), $tval, $fval)>;
+def : Pat<(f128 (selectcc i1:$lhs, i1:$rhs, f128:$tval, f128:$fval, SETUGE)),
+          (SELECT_F16 (CRORC  $lhs, $rhs), $tval, $fval)>;
+def : Pat<(f128 (selectcc i1:$lhs, i1:$rhs, f128:$tval, f128:$fval, SETGT)),
+          (SELECT_F16 (CRANDC $rhs, $lhs), $tval, $fval)>;
+def : Pat<(f128 (selectcc i1:$lhs, i1:$rhs, f128:$tval, f128:$fval, SETUGT)),
+          (SELECT_F16 (CRANDC $lhs, $rhs), $tval, $fval)>;
+def : Pat<(f128 (selectcc i1:$lhs, i1:$rhs, f128:$tval, f128:$fval, SETNE)),
+          (SELECT_F16 (CRXOR $lhs, $rhs), $tval, $fval)>;
+
 def : Pat<(v4i32 (selectcc i1:$lhs, i1:$rhs, v4i32:$tval, v4i32:$fval, SETLT)),
           (SELECT_VRRC (CRANDC $lhs, $rhs), $tval, $fval)>;
 def : Pat<(v4i32 (selectcc i1:$lhs, i1:$rhs, v4i32:$tval, v4i32:$fval, SETULT)),

llvm/test/CodeGen/PowerPC/f128-compare.ll

@@ -0,0 +1,225 @@
+; RUN: llc -mcpu=pwr9 -mtriple=powerpc64le-unknown-unknown \
+; RUN:   -enable-ppc-quad-precision -verify-machineinstrs < %s | FileCheck %s
+
+@a_qp = common global fp128 0xL00000000000000000000000000000000, align 16
+@b_qp = common global fp128 0xL00000000000000000000000000000000, align 16
+
+; Function Attrs: noinline nounwind optnone
+define signext i32 @greater_qp() {
+entry:
+  %0 = load fp128, fp128* @a_qp, align 16
+  %1 = load fp128, fp128* @b_qp, align 16
+  %cmp = fcmp ogt fp128 %0, %1
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+; CHECK-LABEL: greater_qp
+; CHECK: xscmpuqp
+; CHECK: isel {{[0-9]+}}, {{[0-9]+}}, {{[0-9]+}}, 1
+; CHECK: blr
+}
+
+; Function Attrs: noinline nounwind optnone
+define signext i32 @less_qp() {
+entry:
+  %0 = load fp128, fp128* @a_qp, align 16
+  %1 = load fp128, fp128* @b_qp, align 16
+  %cmp = fcmp olt fp128 %0, %1
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+; CHECK-LABEL: less_qp
+; CHECK: xscmpuqp
+; CHECK: isel {{[0-9]+}}, {{[0-9]+}}, {{[0-9]+}}, 0
+; CHECK: blr
+}
+
+; Function Attrs: noinline nounwind optnone
+define signext i32 @greater_eq_qp() {
+entry:
+  %0 = load fp128, fp128* @a_qp, align 16
+  %1 = load fp128, fp128* @b_qp, align 16
+  %cmp = fcmp oge fp128 %0, %1
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+; CHECK-LABEL: greater_eq_qp
+; CHECK: xscmpuqp
+; CHECK: cror [[REG:[0-9]+]], {{[0-9]+}}, 0
+; CHECK: isel {{[0-9]+}}, {{[0-9]+}}, {{[0-9]+}}, [[REG]]
+; CHECK: blr
+}
+
+; Function Attrs: noinline nounwind optnone
+define signext i32 @less_eq_qp() {
+entry:
+  %0 = load fp128, fp128* @a_qp, align 16
+  %1 = load fp128, fp128* @b_qp, align 16
+  %cmp = fcmp ole fp128 %0, %1
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+; CHECK-LABEL: less_eq_qp
+; CHECK: xscmpuqp
+; CHECK: cror [[REG:[0-9]+]], {{[0-9]+}}, 1
+; CHECK: isel {{[0-9]+}}, {{[0-9]+}}, {{[0-9]+}}, [[REG]]
+; CHECK: blr
+}
+
+; Function Attrs: noinline nounwind optnone
+define signext i32 @equal_qp() {
+entry:
+  %0 = load fp128, fp128* @a_qp, align 16
+  %1 = load fp128, fp128* @b_qp, align 16
+  %cmp = fcmp oeq fp128 %0, %1
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+; CHECK-LABEL: equal_qp
+; CHECK: xscmpuqp
+; CHECK: isel {{[0-9]+}}, {{[0-9]+}}, {{[0-9]+}}, 2
+; CHECK: blr
+}
+
+; Function Attrs: noinline nounwind optnone
+define signext i32 @not_greater_qp() {
+entry:
+  %0 = load fp128, fp128* @a_qp, align 16
+  %1 = load fp128, fp128* @b_qp, align 16
+  %cmp = fcmp ogt fp128 %0, %1
+  %lnot = xor i1 %cmp, true
+  %lnot.ext = zext i1 %lnot to i32
+  ret i32 %lnot.ext
+; CHECK-LABEL: not_greater_qp
+; CHECK: xscmpuqp
+; CHECK: isel {{[0-9]+}}, {{[0-9]+}}, {{[0-9]+}}, 1
+; CHECK: blr
+}
+
+; Function Attrs: noinline nounwind optnone
+define signext i32 @not_less_qp() {
+entry:
+  %0 = load fp128, fp128* @a_qp, align 16
+  %1 = load fp128, fp128* @b_qp, align 16
+  %cmp = fcmp olt fp128 %0, %1
+  %lnot = xor i1 %cmp, true
+  %lnot.ext = zext i1 %lnot to i32
+  ret i32 %lnot.ext
+; CHECK-LABEL: not_less_qp
+; CHECK: xscmpuqp
+; CHECK: isel {{[0-9]+}}, {{[0-9]+}}, {{[0-9]+}}, 0
+; CHECK: blr
+}
+
+; Function Attrs: noinline nounwind optnone
+define signext i32 @not_greater_eq_qp() {
+entry:
+  %0 = load fp128, fp128* @a_qp, align 16
+  %1 = load fp128, fp128* @b_qp, align 16
+  %cmp = fcmp oge fp128 %0, %1
+  %lnot = xor i1 %cmp, true
+  %lnot.ext = zext i1 %lnot to i32
+  ret i32 %lnot.ext
+; CHECK-LABEL: not_greater_eq_qp
+; CHECK: xscmpuqp
+; CHECK: crnor [[REG:[0-9]+]], 0, {{[0-9]+}}
+; CHECK: isel {{[0-9]+}}, {{[0-9]+}}, {{[0-9]+}}, [[REG]]
+; CHECK: blr
+}
+
+; Function Attrs: noinline nounwind optnone
+define signext i32 @not_less_eq_qp() {
+entry:
+  %0 = load fp128, fp128* @a_qp, align 16
+  %1 = load fp128, fp128* @b_qp, align 16
+  %cmp = fcmp ole fp128 %0, %1
+  %lnot = xor i1 %cmp, true
+  %lnot.ext = zext i1 %lnot to i32
+  ret i32 %lnot.ext
+; CHECK-LABEL: not_less_eq_qp
+; CHECK: xscmpuqp
+; CHECK: crnor [[REG:[0-9]+]], 1, {{[0-9]+}}
+; CHECK: isel {{[0-9]+}}, {{[0-9]+}}, {{[0-9]+}}, [[REG]]
+; CHECK: blr
+}
+
+; Function Attrs: noinline nounwind optnone
+define signext i32 @not_equal_qp() {
+entry:
+  %0 = load fp128, fp128* @a_qp, align 16
+  %1 = load fp128, fp128* @b_qp, align 16
+  %cmp = fcmp une fp128 %0, %1
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+; CHECK-LABEL: not_equal_qp
+; CHECK: xscmpuqp
+; CHECK: isel {{[0-9]+}}, {{[0-9]+}}, {{[0-9]+}}, 2
+; CHECK: blr
+}
+
+; Function Attrs: norecurse nounwind readonly
+define fp128 @greater_sel_qp() {
+entry:
+  %0 = load fp128, fp128* @a_qp, align 16
+  %1 = load fp128, fp128* @b_qp, align 16
+  %cmp = fcmp ogt fp128 %0, %1
+  %cond = select i1 %cmp, fp128 %0, fp128 %1
+  ret fp128 %cond
+; CHECK-LABEL: greater_sel_qp
+; CHECK: xscmpuqp [[REG:[0-9]+]]
+; CHECK: bgtlr [[REG]]
+; CHECK: blr
+}
+
+; Function Attrs: noinline nounwind optnone
+define fp128 @less_sel_qp() {
+entry:
+  %0 = load fp128, fp128* @a_qp, align 16
+  %1 = load fp128, fp128* @b_qp, align 16
+  %cmp = fcmp olt fp128 %0, %1
+  %cond = select i1 %cmp, fp128 %0, fp128 %1
+  ret fp128 %cond
+; CHECK-LABEL: less_sel_qp
+; CHECK: xscmpuqp [[REG:[0-9]+]]
+; CHECK: bltlr [[REG]]
+; CHECK: blr
+}
+
+; Function Attrs: noinline nounwind optnone
+define fp128 @greater_eq_sel_qp() {
+entry:
+  %0 = load fp128, fp128* @a_qp, align 16
+  %1 = load fp128, fp128* @b_qp, align 16
+  %cmp = fcmp oge fp128 %0, %1
+  %cond = select i1 %cmp, fp128 %0, fp128 %1
+  ret fp128 %cond
+; CHECK-LABEL: greater_eq_sel_qp
+; CHECK: xscmpuqp
+; CHECK: crnor [[REG:[0-9]+]], {{[0-9]+}}, 0
+; CHECK: bclr {{[0-9]+}}, [[REG]]
+; CHECK: blr
+}
+
+; Function Attrs: noinline nounwind optnone
+define fp128 @less_eq_sel_qp() {
+entry:
+  %0 = load fp128, fp128* @a_qp, align 16
+  %1 = load fp128, fp128* @b_qp, align 16
+  %cmp = fcmp ole fp128 %0, %1
+  %cond = select i1 %cmp, fp128 %0, fp128 %1
+  ret fp128 %cond
+; CHECK-LABEL: less_eq_sel_qp
+; CHECK: xscmpuqp
+; CHECK: crnor [[REG:[0-9]+]], {{[0-9]+}}, 1
+; CHECK: bclr {{[0-9]+}}, [[REG]]
+; CHECK: blr
+}
+
+; Function Attrs: noinline nounwind optnone
+define fp128 @equal_sel_qp() {
+entry:
+  %0 = load fp128, fp128* @a_qp, align 16
+  %1 = load fp128, fp128* @b_qp, align 16
+  %cmp = fcmp oeq fp128 %0, %1
+  %cond = select i1 %cmp, fp128 %0, fp128 %1
+  ret fp128 %cond
+; CHECK-LABEL: equal_sel_qp
+; CHECK: xscmpuqp [[REG:[0-9]+]]
+; CHECK: beqlr [[REG]]
+; CHECK: blr
+}