[llvm] 1725f28 - DAG: Add new control for ISD::FMAD formation

[Matt Arsenault via llvm-commits]

Author: Matt Arsenault
Date: 2019-10-31T07:51:38-07:00
New Revision: 1725f2884175ca618d29b06e35f5c6ebd618053d

URL: https://github.com/llvm/llvm-project/commit/1725f2884175ca618d29b06e35f5c6ebd618053d
DIFF: https://github.com/llvm/llvm-project/commit/1725f2884175ca618d29b06e35f5c6ebd618053d.diff

LOG: DAG: Add new control for ISD::FMAD formation

For AMDGPU this depends on whether denormals are enabled in the
default FP mode for the function. Currently this is treated as a
subtarget feature, so FMAD is selectively legal based on that. I want
to move this out of the subtarget features so this can be controlled
with a denormal mode attribute. Additionally, this will allow folding
based on a future ftz fast math flag.

Added: 
    

Modified: 
    llvm/include/llvm/CodeGen/TargetLowering.h
    llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
    llvm/lib/Target/AMDGPU/SIISelLowering.cpp
    llvm/lib/Target/AMDGPU/SIISelLowering.h

Removed: 
    


################################################################################
diff  --git a/llvm/include/llvm/CodeGen/TargetLowering.h b/llvm/include/llvm/CodeGen/TargetLowering.h
index b9bf19474ca9..3e9bef0254e6 100644
--- a/llvm/include/llvm/CodeGen/TargetLowering.h
+++ b/llvm/include/llvm/CodeGen/TargetLowering.h
@@ -2541,6 +2541,14 @@ class TargetLoweringBase {
     return false;
   }
 
+  /// Returns true if the FADD or FSUB node passed could legally be combined with
+  /// an fmul to form an ISD::FMAD.
+  virtual bool isFMADLegalForFAddFSub(const SelectionDAG &DAG,
+                                      const SDNode *N) const {
+    assert(N->getOpcode() == ISD::FADD || N->getOpcode() == ISD::FSUB);
+    return isOperationLegal(ISD::FMAD, N->getValueType(0));
+  }
+
   /// Return true if it's profitable to narrow operations of type VT1 to
   /// VT2. e.g. on x86, it's profitable to narrow from i32 to i8 but not from
   /// i32 to i16.

diff  --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 5003c8f05364..e9fb560e8586 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -11330,7 +11330,7 @@ SDValue DAGCombiner::visitFADDForFMACombine(SDNode *N) {
   const TargetOptions &Options = DAG.getTarget().Options;
 
   // Floating-point multiply-add with intermediate rounding.
-  bool HasFMAD = (LegalOperations && TLI.isOperationLegal(ISD::FMAD, VT));
+  bool HasFMAD = (LegalOperations && TLI.isFMADLegalForFAddFSub(DAG, N));
 
   // Floating-point multiply-add without intermediate rounding.
   bool HasFMA =
@@ -11541,7 +11541,7 @@ SDValue DAGCombiner::visitFSUBForFMACombine(SDNode *N) {
 
   const TargetOptions &Options = DAG.getTarget().Options;
   // Floating-point multiply-add with intermediate rounding.
-  bool HasFMAD = (LegalOperations && TLI.isOperationLegal(ISD::FMAD, VT));
+  bool HasFMAD = (LegalOperations && TLI.isFMADLegalForFAddFSub(DAG, N));
 
   // Floating-point multiply-add without intermediate rounding.
   bool HasFMA =

diff  --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index ea38a1e6171f..b1018e8538ed 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -3939,6 +3939,19 @@ bool SITargetLowering::isFMAFasterThanFMulAndFAdd(EVT VT) const {
   return false;
 }
 
+bool SITargetLowering::isFMADLegalForFAddFSub(const SelectionDAG &DAG,
+                                              const SDNode *N) const {
+  // TODO: Check future ftz flag
+  // v_mad_f32/v_mac_f32 do not support denormals.
+  EVT VT = N->getValueType(0);
+  if (VT == MVT::f32)
+    return !Subtarget->hasFP32Denormals();
+  if (VT == MVT::f16)
+    return !Subtarget->hasFP16Denormals() && Subtarget->hasMadF16();
+
+  return false;
+}
+
 //===----------------------------------------------------------------------===//
 // Custom DAG Lowering Operations
 //===----------------------------------------------------------------------===//

diff  --git a/llvm/lib/Target/AMDGPU/SIISelLowering.h b/llvm/lib/Target/AMDGPU/SIISelLowering.h
index b36574f8ff60..c3605a3f1582 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.h
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.h
@@ -349,6 +349,9 @@ class SITargetLowering final : public AMDGPUTargetLowering {
                          EVT VT) const override;
   MVT getScalarShiftAmountTy(const DataLayout &, EVT) const override;
   bool isFMAFasterThanFMulAndFAdd(EVT VT) const override;
+  bool isFMADLegalForFAddFSub(const SelectionDAG &DAG,
+                              const SDNode *N) const override;
+
   SDValue splitUnaryVectorOp(SDValue Op, SelectionDAG &DAG) const;
   SDValue splitBinaryVectorOp(SDValue Op, SelectionDAG &DAG) const;
   SDValue splitTernaryVectorOp(SDValue Op, SelectionDAG &DAG) const;