[DAGCombiner] try to convert pow(x, 1/3) to cbrt(x)

This is a follow-up suggested in D51630 and originally proposed as an IR transform in D49040.

Copying the motivational statement by @evandro from that patch:
"This transformation helps some benchmarks in SPEC CPU2000 and CPU2006, such as 188.ammp, 
447.dealII, 453.povray, and especially 300.twolf, as well as some proprietary benchmarks. 
Otherwise, no regressions on x86-64 or A64."

I'm proposing to add only the minimum support for a DAG node here. Since we don't have an 
LLVM IR intrinsic for cbrt, and there are no other DAG ways to create a FCBRT node yet, I 
don't think we need to worry about DAG builder, legalization, a strict variant, etc. We 
should be able to expand as needed when adding more functionality/transforms. For reference, 
these are transform suggestions currently listed in SimplifyLibCalls.cpp:

//   * cbrt(expN(X))  -> expN(x/3)
//   * cbrt(sqrt(x))  -> pow(x,1/6)
//   * cbrt(cbrt(x))  -> pow(x,1/9)

Also, given that we bail out on long double for now, there should not be any logical 
differences between platforms (unless there's some platform out there that has pow()
but not cbrt()).

Differential Revision: https://reviews.llvm.org/D51753

llvm-svn: 342348

3 files changed, 34 insertions, 1 deletions

diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 20e749ba569..4ef16cb7186 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -11571,6 +11571,34 @@ SDValue DAGCombiner::visitFPOW(SDNode *N) {
   if (!ExponentC)
     return SDValue();
 
+  // Try to convert x ** (1/3) into cube root.
+  // TODO: Handle the various flavors of long double.
+  // TODO: Since we're approximating, we don't need an exact 1/3 exponent.
+  //       Some range near 1/3 should be fine.
+  EVT VT = N->getValueType(0);
+  if ((VT == MVT::f32 && ExponentC->getValueAPF().isExactlyValue(1.0f/3.0f)) ||
+      (VT == MVT::f64 && ExponentC->getValueAPF().isExactlyValue(1.0/3.0))) {
+    // pow(-0.0, 1/3) = +0.0; cbrt(-0.0) = -0.0.
+    // pow(-inf, 1/3) = +inf; cbrt(-inf) = -inf.
+    // pow(-val, 1/3) =  nan; cbrt(-val) = -num.
+    // For regular numbers, rounding may cause the results to differ.
+    // Therefore, we require { nsz ninf nnan afn } for this transform.
+    // TODO: We could select out the special cases if we don't have nsz/ninf.
+    SDNodeFlags Flags = N->getFlags();
+    if (!Flags.hasNoSignedZeros() || !Flags.hasNoInfs() || !Flags.hasNoNaNs() ||
+        !Flags.hasApproximateFuncs())
+      return SDValue();
+
+    // Do not create a cbrt() libcall if the target does not have it, and do not
+    // turn a pow that has lowering support into a cbrt() libcall.
+    if (!DAG.getLibInfo().has(LibFunc_cbrt) ||
+        (!DAG.getTargetLoweringInfo().isOperationExpand(ISD::FPOW, VT) &&
+         DAG.getTargetLoweringInfo().isOperationExpand(ISD::FCBRT, VT)))
+      return SDValue();
+
+    return DAG.getNode(ISD::FCBRT, SDLoc(N), VT, N->getOperand(0), Flags);
+  }
+
   // Try to convert x ** (1/4) into square roots.
   // x ** (1/2) is canonicalized to sqrt, so we do not bother with that case.
   // TODO: This could be extended (using a target hook) to handle smaller
@@ -11587,7 +11615,6 @@ SDValue DAGCombiner::visitFPOW(SDNode *N) {
       return SDValue();
 
     // Don't double the number of libcalls. We are trying to inline fast code.
-    EVT VT = N->getValueType(0);
     if (!DAG.getTargetLoweringInfo().isOperationLegalOrCustom(ISD::FSQRT, VT))
       return SDValue();
 
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
index 13d412cd7de..b6bd8541ac7 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
@@ -4047,6 +4047,11 @@ void SelectionDAGLegalize::ConvertNodeToLibcall(SDNode *Node) {
                                       RTLIB::SQRT_F80, RTLIB::SQRT_F128,
                                       RTLIB::SQRT_PPCF128));
     break;
+  case ISD::FCBRT:
+    Results.push_back(ExpandFPLibCall(Node, RTLIB::CBRT_F32, RTLIB::CBRT_F64,
+                                      RTLIB::CBRT_F80, RTLIB::CBRT_F128,
+                                      RTLIB::CBRT_PPCF128));
+    break;
   case ISD::FSIN:
   case ISD::STRICT_FSIN:
     Results.push_back(ExpandFPLibCall(Node, RTLIB::SIN_F32, RTLIB::SIN_F64,
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
index 7289579502d..594a587e412 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
@@ -181,6 +181,7 @@ std::string SDNode::getOperationName(const SelectionDAG *G) const {
   case ISD::FNEG:                       return "fneg";
   case ISD::FSQRT:                      return "fsqrt";
   case ISD::STRICT_FSQRT:               return "strict_fsqrt";
+  case ISD::FCBRT:                      return "fcbrt";
   case ISD::FSIN:                       return "fsin";
   case ISD::STRICT_FSIN:                return "strict_fsin";
   case ISD::FCOS:                       return "fcos";