[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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@318727 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/SelectionDAG/LegalizeTypes.cpp b/lib/CodeGen/SelectionDAG/LegalizeTypes.cpp
index 85154ffb14d05256d0a955662f9bb5fd7df5b182..699081e2a5cf70e5f2a22664ebfd9815e24a4f74 100644 (file)
--- a/lib/CodeGen/SelectionDAG/LegalizeTypes.cpp
+++ b/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/lib/Target/X86/X86ISelLowering.h b/lib/Target/X86/X86ISelLowering.h
index fa3fedf987ca6a97ac3495d23af22ef89f046acb..426fb347af6a27f6cdd3c8064febd6000817a666 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.h
+++ b/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 *

diff --git a/test/CodeGen/AMDGPU/insert_vector_elt.v2i16.ll b/test/CodeGen/AMDGPU/insert_vector_elt.v2i16.ll
index 403dae3ff8843add52dfdb998472de7a75f597a7..de973eb0ba7e4b58af00c804cb09df117d12b05f 100644 (file)
--- a/test/CodeGen/AMDGPU/insert_vector_elt.v2i16.ll
+++ b/test/CodeGen/AMDGPU/insert_vector_elt.v2i16.ll
@@ -1,6 +1,6 @@
-; RUN: llc -verify-machineinstrs -march=amdgcn -mcpu=gfx900 -enable-amdgpu-aa=0 -mattr=+flat-for-global,-fp64-fp16-denormals < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=GFX9 -check-prefix=GFX89 %s
-; RUN: llc -verify-machineinstrs -march=amdgcn -mcpu=fiji -enable-amdgpu-aa=0 -mattr=+flat-for-global < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=CIVI -check-prefix=VI -check-prefix=GFX89 %s
-; RUN: llc -verify-machineinstrs -march=amdgcn -mcpu=hawaii -enable-amdgpu-aa=0 -mattr=+flat-for-global < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=CIVI -check-prefix=CI %s
+; RUN: llc -verify-machineinstrs -march=amdgcn -mtriple=amdgcn---amdgiz -mcpu=gfx900 -enable-amdgpu-aa=0 -mattr=+flat-for-global,-fp64-fp16-denormals < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 -check-prefix=GFX89 %s
+; RUN: llc -verify-machineinstrs -march=amdgcn -mtriple=amdgcn---amdgiz -mcpu=fiji -enable-amdgpu-aa=0 -mattr=+flat-for-global < %s | FileCheck -check-prefix=GCN -check-prefix=CIVI -check-prefix=VI -check-prefix=GFX89 %s
+; RUN: llc -verify-machineinstrs -march=amdgcn -mtriple=amdgcn---amdgiz -mcpu=hawaii -enable-amdgpu-aa=0 -mattr=+flat-for-global < %s | FileCheck -check-prefix=GCN -check-prefix=CIVI -check-prefix=CI %s
 
 ; GCN-LABEL: {{^}}s_insertelement_v2i16_0:
 ; GCN: s_load_dword [[VEC:s[0-9]+]]