Prepare support for AltiVec multiply, divide, and sqrt.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24700 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 19 insertions, 2 deletions

diff --git a/lib/Target/PowerPC/PPCISelLowering.h b/lib/Target/PowerPC/PPCISelLowering.h
index 392e735..4cce315 100644
--- a/lib/Target/PowerPC/PPCISelLowering.h
+++ b/lib/Target/PowerPC/PPCISelLowering.h
@@ -39,6 +39,10 @@ namespace llvm {
       /// of that FP value.
       FCTIDZ, FCTIWZ,
       
+      // VMADDFP, VNMSUBFP - The VMADDFP and VNMSUBFP instructions, taking
+      // three v4f32 operands and producing a v4f32 result.
+      VMADDFP, VNMSUBFP,
+      
       /// Hi/Lo - These represent the high and low 16-bit parts of a global
       /// address respectively.  These nodes have two operands, the first of
       /// which must be a TargetGlobalAddress, and the second of which must be a
diff --git a/lib/Target/PowerPC/PPCInstrInfo.td b/lib/Target/PowerPC/PPCInstrInfo.td
index 62b36b8..3ea0920 100644
--- a/lib/Target/PowerPC/PPCInstrInfo.td
+++ b/lib/Target/PowerPC/PPCInstrInfo.td
@@ -27,8 +27,10 @@ def PPCfsel   : SDNode<"PPCISD::FSEL",
    SDTypeProfile<1, 3, [SDTCisSameAs<0, 2>, SDTCisSameAs<0, 3>, 
                         SDTCisFP<0>, SDTCisVT<1, f64>]>, []>;
 
-def PPChi     : SDNode<"PPCISD::Hi", SDTIntBinOp, []>;
-def PPClo     : SDNode<"PPCISD::Lo", SDTIntBinOp, []>;
+def PPChi       : SDNode<"PPCISD::Hi", SDTIntBinOp, []>;
+def PPClo       : SDNode<"PPCISD::Lo", SDTIntBinOp, []>;
+def PPCvmaddfp  : SDNode<"PPCISD::VMADDFP", SDTFPTernaryOp, []>;
+def PPCvnmsubfp : SDNode<"PPCISD::VNMSUBFP", SDTFPTernaryOp, []>;
 
 // These nodes represent the 32-bit PPC shifts that operate on 6-bit shift
 // amounts.  These nodes are generated by the multi-precision shift code.
@@ -949,6 +951,14 @@ def : Pat<(add GPRC:$in, (PPChi tglobaladdr:$g, 0)),
 def : Pat<(add GPRC:$in, (PPChi tconstpool:$g, 0)),
           (ADDIS GPRC:$in, tconstpool:$g)>;
 
+// Fused multiply add and multiply sub for packed float.  These are represented
+// separately from the real instructions above, for operations that must have
+// the additional precision, such as Newton-Rhapson (used by divide, sqrt)
+def : Pat<(PPCvmaddfp VRRC:$A, VRRC:$B, VRRC:$C),
+          (VMADDFP VRRC:$A, VRRC:$B, VRRC:$C)>;
+def : Pat<(PPCvnmsubfp VRRC:$A, VRRC:$B, VRRC:$C),
+          (VNMSUBFP VRRC:$A, VRRC:$B, VRRC:$C)>;
+
 // Standard shifts.  These are represented separately from the real shifts above
 // so that we can distinguish between shifts that allow 5-bit and 6-bit shift
 // amounts.
diff --git a/lib/Target/TargetSelectionDAG.td b/lib/Target/TargetSelectionDAG.td
index 83f5593..31d8e43 100644
--- a/lib/Target/TargetSelectionDAG.td
+++ b/lib/Target/TargetSelectionDAG.td
@@ -82,6 +82,9 @@ def SDTIntShiftOp : SDTypeProfile<1, 2, [   // shl, sra, srl
 def SDTFPBinOp : SDTypeProfile<1, 2, [      // fadd, fmul, etc.
   SDTCisSameAs<0, 1>, SDTCisSameAs<0, 2>, SDTCisFP<0>
 ]>;
+def SDTFPTernaryOp : SDTypeProfile<1, 3, [      // fmadd, fnmsub, etc.
+  SDTCisSameAs<0, 1>, SDTCisSameAs<0, 2>, SDTCisSameAs<0, 3>, SDTCisFP<0>
+]>;
 def SDTIntUnaryOp : SDTypeProfile<1, 1, [   // ctlz
   SDTCisSameAs<0, 1>, SDTCisInt<0>
 ]>;