[AMDGPU] Fix DAGTypeLegalizer::SplitInteger for shift amount type

DAGTypeLegalizer::SplitInteger uses default pointer size as shift amount constant type,
which causes less performant ISA in amdgcn---amdgiz target since the default pointer
type is i64 whereas the desired shift amount type is i32.

This patch fixes that by using TLI.getScalarShiftAmountTy in DAGTypeLegalizer::SplitInteger.

The X86 change is necessary since splitting i512 requires shifting amount of 256, which
cannot be held by i8.

Differential Revision: https://reviews.llvm.org/D40148

llvm-svn: 318727

2 files changed, 14 insertions, 5 deletions

diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.cpp
index 85154ffb14d..699081e2a5c 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.cpp
@@ -1172,9 +1172,11 @@ void DAGTypeLegalizer::SplitInteger(SDValue Op,
   assert(LoVT.getSizeInBits() + HiVT.getSizeInBits() ==
          Op.getValueSizeInBits() && "Invalid integer splitting!");
   Lo = DAG.getNode(ISD::TRUNCATE, dl, LoVT, Op);
-  Hi = DAG.getNode(ISD::SRL, dl, Op.getValueType(), Op,
-                   DAG.getConstant(LoVT.getSizeInBits(), dl,
-                                   TLI.getPointerTy(DAG.getDataLayout())));
+  Hi =
+      DAG.getNode(ISD::SRL, dl, Op.getValueType(), Op,
+                  DAG.getConstant(LoVT.getSizeInBits(), dl,
+                                  TLI.getScalarShiftAmountTy(
+                                      DAG.getDataLayout(), Op.getValueType())));
   Hi = DAG.getNode(ISD::TRUNCATE, dl, HiVT, Hi);
 }
 
diff --git a/llvm/lib/Target/X86/X86ISelLowering.h b/llvm/lib/Target/X86/X86ISelLowering.h
index fa3fedf987c..426fb347af6 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.h
+++ b/llvm/lib/Target/X86/X86ISelLowering.h
@@ -18,6 +18,7 @@
 #include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/SelectionDAG.h"
 #include "llvm/CodeGen/TargetLowering.h"
+#include "llvm/Support/MathExtras.h"
 #include "llvm/Target/TargetOptions.h"
 
 namespace llvm {
@@ -664,8 +665,14 @@ namespace llvm {
     void markLibCallAttributes(MachineFunction *MF, unsigned CC,
                                ArgListTy &Args) const override;
 
-    MVT getScalarShiftAmountTy(const DataLayout &, EVT) const override {
-      return MVT::i8;
+    // For i512, DAGTypeLegalizer::SplitInteger needs a shift amount 256,
+    // which cannot be held by i8, therefore use i16 instead. In all the
+    // other situations i8 is sufficient.
+    MVT getScalarShiftAmountTy(const DataLayout &, EVT VT) const override {
+      MVT T = VT.getSizeInBits() >= 512 ? MVT::i16 : MVT::i8;
+      assert((VT.getSizeInBits() + 1) / 2 < (1U << T.getSizeInBits()) &&
+             "Scalar shift amount type too small");
+      return T;
     }
 
     const MCExpr *