R600: Implement 64bit SHL

v2: Use c++ style comment

Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211157 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Target/R600/R600ISelLowering.cpp

@@ -157,6 +157,10 @@ R600TargetLowering::R600TargetLowering(TargetMachine &TM) :
   setOperationAction(ISD::UDIV, MVT::i64, Custom);
   setOperationAction(ISD::UREM, MVT::i64, Custom);
 
+  // We don't have 64-bit shifts. Thus we need either SHX i64 or SHX_PARTS i32
+  //  to be Legal/Custom in order to avoid library calls.
+  setOperationAction(ISD::SHL_PARTS, MVT::i32, Custom);
+
   setOperationAction(ISD::GlobalAddress, MVT::i32, Custom);
 
   setBooleanContents(ZeroOrNegativeOneBooleanContent);
@@ -552,6 +556,7 @@ SDValue R600TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const
   default: return AMDGPUTargetLowering::LowerOperation(Op, DAG);
   case ISD::EXTRACT_VECTOR_ELT: return LowerEXTRACT_VECTOR_ELT(Op, DAG);
   case ISD::INSERT_VECTOR_ELT: return LowerINSERT_VECTOR_ELT(Op, DAG);
+  case ISD::SHL_PARTS: return LowerSHLParts(Op, DAG);
   case ISD::FCOS:
   case ISD::FSIN: return LowerTrig(Op, DAG);
   case ISD::SELECT_CC: return LowerSELECT_CC(Op, DAG);
@@ -905,6 +910,42 @@ SDValue R600TargetLowering::LowerTrig(SDValue Op, SelectionDAG &DAG) const {
       DAG.getConstantFP(3.14159265359, MVT::f32));
 }
 
+SDValue R600TargetLowering::LowerSHLParts(SDValue Op, SelectionDAG &DAG) const {
+  SDLoc DL(Op);
+  EVT VT = Op.getValueType();
+
+  SDValue Lo = Op.getOperand(0);
+  SDValue Hi = Op.getOperand(1);
+  SDValue Shift = Op.getOperand(2);
+  SDValue Zero = DAG.getConstant(0, VT);
+  SDValue One  = DAG.getConstant(1, VT);
+
+  SDValue Width  = DAG.getConstant(VT.getSizeInBits(), VT);
+  SDValue Width1 = DAG.getConstant(VT.getSizeInBits() - 1, VT);
+  SDValue BigShift  = DAG.getNode(ISD::SUB, DL, VT, Shift, Width);
+  SDValue CompShift = DAG.getNode(ISD::SUB, DL, VT, Width1, Shift);
+
+  // The dance around Width1 is necessary for 0 special case.
+  // Without it the CompShift might be 32, producing incorrect results in
+  // Overflow. So we do the shift in two steps, the alternative is to
+  // add a conditional to filter the special case.
+
+  SDValue Overflow = DAG.getNode(ISD::SRL, DL, VT, Lo, CompShift);
+  Overflow = DAG.getNode(ISD::SRL, DL, VT, Overflow, One);
+
+  SDValue HiSmall = DAG.getNode(ISD::SHL, DL, VT, Hi, Shift);
+  HiSmall = DAG.getNode(ISD::OR, DL, VT, HiSmall, Overflow);
+  SDValue LoSmall = DAG.getNode(ISD::SHL, DL, VT, Lo, Shift);
+
+  SDValue HiBig = DAG.getNode(ISD::SHL, DL, VT, Lo, BigShift);
+  SDValue LoBig = Zero;
+
+  Hi = DAG.getSelectCC(DL, Shift, Width, HiSmall, HiBig, ISD::SETULT);
+  Lo = DAG.getSelectCC(DL, Shift, Width, LoSmall, LoBig, ISD::SETULT);
+
+  return DAG.getNode(ISD::MERGE_VALUES, DL, DAG.getVTList(VT,VT), Lo, Hi);
+}
+
 SDValue R600TargetLowering::LowerFPTOUINT(SDValue Op, SelectionDAG &DAG) const {
   return DAG.getNode(
       ISD::SETCC,

lib/Target/R600/R600ISelLowering.h

@@ -60,6 +60,7 @@ private:
   SDValue LowerFPTOUINT(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerLOAD(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerTrig(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSHLParts(SDValue Op, SelectionDAG &DAG) const;
 
   SDValue stackPtrToRegIndex(SDValue Ptr, unsigned StackWidth,
                                           SelectionDAG &DAG) const;

test/CodeGen/R600/shl.ll

@@ -39,5 +39,118 @@ define void @shl_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in
   ret void
 }
 
-; XXX: Add SI test for i64 shl once i64 stores and i64 function arguments are
+;EG-CHECK: @shl_i64
-; supported.
+;EG-CHECK: SUB_INT {{\*? *}}[[COMPSH:T[0-9]+\.[XYZW]]], {{literal.[xy]}}, [[SHIFT:T[0-9]+\.[XYZW]]]
+;EG-CHECK: LSHR {{\* *}}[[TEMP:T[0-9]+\.[XYZW]]], [[OPLO:T[0-9]+\.[XYZW]]], {{[[COMPSH]]|PV.[XYZW]}}
+;EG-CHECK: LSHR {{\*? *}}[[OVERF:T[0-9]+\.[XYZW]]], {{[[TEMP]]|PV.[XYZW]}}, 1
+;EG_CHECK-DAG: ADD_INT {{\*? *}}[[BIGSH:T[0-9]+\.[XYZW]]], [[SHIFT]], literal
+;EG-CHECK-DAG: LSHL {{\*? *}}[[HISMTMP:T[0-9]+\.[XYZW]]], [[OPHI:T[0-9]+\.[XYZW]]], [[SHIFT]]
+;EG-CHECK-DAG: OR_INT {{\*? *}}[[HISM:T[0-9]+\.[XYZW]]], {{[[HISMTMP]]|PV.[XYZW]}}, {{[[OVERF]]|PV.[XYZW]}}
+;EG-CHECK-DAG: LSHL {{\*? *}}[[LOSM:T[0-9]+\.[XYZW]]], [[OPLO]], {{PS|[[SHIFT]]}}
+;EG-CHECK-DAG: SETGT_UINT {{\*? *}}[[RESC:T[0-9]+\.[XYZW]]], [[SHIFT]], literal
+;EG-CHECK-DAG: CNDE_INT {{\*? *}}[[RESLO:T[0-9]+\.[XYZW]]], {{T[0-9]+\.[XYZW]}}
+;EG-CHECK-DAG: CNDE_INT {{\*? *}}[[RESHI:T[0-9]+\.[XYZW]]], {{T[0-9]+\.[XYZW], .*}}, 0.0
+
+;SI-CHECK: @shl_i64
+;SI-CHECK: V_LSHL_B64 {{v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\], v[0-9]+}}
+
+define void @shl_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %in) {
+  %b_ptr = getelementptr i64 addrspace(1)* %in, i64 1
+  %a = load i64 addrspace(1) * %in
+  %b = load i64 addrspace(1) * %b_ptr
+  %result = shl i64 %a, %b
+  store i64 %result, i64 addrspace(1)* %out
+  ret void
+}
+
+;EG-CHECK: @shl_v2i64
+;EG-CHECK-DAG: SUB_INT {{\*? *}}[[COMPSHA:T[0-9]+\.[XYZW]]], {{literal.[xy]}}, [[SHA:T[0-9]+\.[XYZW]]]
+;EG-CHECK-DAG: SUB_INT {{\*? *}}[[COMPSHB:T[0-9]+\.[XYZW]]], {{literal.[xy]}}, [[SHB:T[0-9]+\.[XYZW]]]
+;EG-CHECK-DAG: LSHR {{\*? *}}[[COMPSHA]]
+;EG-CHECK-DAG: LSHR {{\*? *}}[[COMPSHB]]
+;EG-CHECK-DAG: LSHR {{.*}}, 1
+;EG-CHECK-DAG: LSHR {{.*}}, 1
+;EG-CHECK-DAG: ADD_INT  {{\*? *}}[[BIGSHA:T[0-9]+\.[XYZW]]]{{.*}}, literal
+;EG-CHECK-DAG: ADD_INT  {{\*? *}}[[BIGSHB:T[0-9]+\.[XYZW]]]{{.*}}, literal
+;EG-CHECK-DAG: LSHL {{.*}}, [[SHA]]
+;EG-CHECK-DAG: LSHL {{.*}}, [[SHB]]
+;EG-CHECK-DAG: LSHL {{.*}}, [[SHA]]
+;EG-CHECK-DAG: LSHL {{.*}}, [[SHB]]
+;EG-CHECK-DAG: LSHL
+;EG-CHECK-DAG: LSHL
+;EG-CHECK-DAG: SETGT_UINT {{\*? *T[0-9]\.[XYZW]}}, [[SHA]], literal
+;EG-CHECK-DAG: SETGT_UINT {{\*? *T[0-9]\.[XYZW]}}, [[SHB]], literal
+;EG-CHECK-DAG: CNDE_INT {{.*}}, 0.0
+;EG-CHECK-DAG: CNDE_INT {{.*}}, 0.0
+;EG-CHECK-DAG: CNDE_INT
+;EG-CHECK-DAG: CNDE_INT
+
+;SI-CHECK: @shl_v2i64
+;SI-CHECK: V_LSHL_B64 {{v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\], v[0-9]+}}
+;SI-CHECK: V_LSHL_B64 {{v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\], v[0-9]+}}
+
+define void @shl_v2i64(<2 x i64> addrspace(1)* %out, <2 x i64> addrspace(1)* %in) {
+  %b_ptr = getelementptr <2 x i64> addrspace(1)* %in, i64 1
+  %a = load <2 x i64> addrspace(1) * %in
+  %b = load <2 x i64> addrspace(1) * %b_ptr
+  %result = shl <2 x i64> %a, %b
+  store <2 x i64> %result, <2 x i64> addrspace(1)* %out
+  ret void
+}
+
+;EG-CHECK: @shl_v4i64
+;EG-CHECK-DAG: SUB_INT {{\*? *}}[[COMPSHA:T[0-9]+\.[XYZW]]], {{literal.[xy]}}, [[SHA:T[0-9]+\.[XYZW]]]
+;EG-CHECK-DAG: SUB_INT {{\*? *}}[[COMPSHB:T[0-9]+\.[XYZW]]], {{literal.[xy]}}, [[SHB:T[0-9]+\.[XYZW]]]
+;EG-CHECK-DAG: SUB_INT {{\*? *}}[[COMPSHC:T[0-9]+\.[XYZW]]], {{literal.[xy]}}, [[SHC:T[0-9]+\.[XYZW]]]
+;EG-CHECK-DAG: SUB_INT {{\*? *}}[[COMPSHD:T[0-9]+\.[XYZW]]], {{literal.[xy]}}, [[SHD:T[0-9]+\.[XYZW]]]
+;EG-CHECK-DAG: LSHR {{\*? *}}[[COMPSHA]]
+;EG-CHECK-DAG: LSHR {{\*? *}}[[COMPSHB]]
+;EG-CHECK-DAG: LSHR {{\*? *}}[[COMPSHC]]
+;EG-CHECK-DAG: LSHR {{\*? *}}[[COMPSHD]]
+;EG-CHECK-DAG: LSHR {{.*}}, 1
+;EG-CHECK-DAG: LSHR {{.*}}, 1
+;EG-CHECK-DAG: LSHR {{.*}}, 1
+;EG-CHECK-DAG: LSHR {{.*}}, 1
+;EG-CHECK-DAG: ADD_INT  {{\*? *}}[[BIGSHA:T[0-9]+\.[XYZW]]]{{.*}}, literal
+;EG-CHECK-DAG: ADD_INT  {{\*? *}}[[BIGSHB:T[0-9]+\.[XYZW]]]{{.*}}, literal
+;EG-CHECK-DAG: ADD_INT  {{\*? *}}[[BIGSHC:T[0-9]+\.[XYZW]]]{{.*}}, literal
+;EG-CHECK-DAG: ADD_INT  {{\*? *}}[[BIGSHD:T[0-9]+\.[XYZW]]]{{.*}}, literal
+;EG-CHECK-DAG: LSHL {{.*}}, [[SHA]]
+;EG-CHECK-DAG: LSHL {{.*}}, [[SHB]]
+;EG-CHECK-DAG: LSHL {{.*}}, [[SHC]]
+;EG-CHECK-DAG: LSHL {{.*}}, [[SHD]]
+;EG-CHECK-DAG: LSHL {{.*}}, [[SHA]]
+;EG-CHECK-DAG: LSHL {{.*}}, [[SHB]]
+;EG-CHECK-DAG: LSHL {{.*}}, [[SHC]]
+;EG-CHECK-DAG: LSHL {{.*}}, [[SHD]]
+;EG-CHECK-DAG: LSHL
+;EG-CHECK-DAG: LSHL
+;EG-CHECK-DAG: LSHL
+;EG-CHECK-DAG: LSHL
+;EG-CHECK-DAG: SETGT_UINT {{\*? *T[0-9]\.[XYZW]}}, [[SHA]], literal
+;EG-CHECK-DAG: SETGT_UINT {{\*? *T[0-9]\.[XYZW]}}, [[SHB]], literal
+;EG-CHECK-DAG: SETGT_UINT {{\*? *T[0-9]\.[XYZW]}}, [[SHC]], literal
+;EG-CHECK-DAG: SETGT_UINT {{\*? *T[0-9]\.[XYZW]}}, [[SHD]], literal
+;EG-CHECK-DAG: CNDE_INT {{.*}}, 0.0
+;EG-CHECK-DAG: CNDE_INT {{.*}}, 0.0
+;EG-CHECK-DAG: CNDE_INT {{.*}}, 0.0
+;EG-CHECK-DAG: CNDE_INT {{.*}}, 0.0
+;EG-CHECK-DAG: CNDE_INT
+;EG-CHECK-DAG: CNDE_INT
+;EG-CHECK-DAG: CNDE_INT
+;EG-CHECK-DAG: CNDE_INT
+
+;SI-CHECK: @shl_v4i64
+;SI-CHECK: V_LSHL_B64 {{v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\], v[0-9]+}}
+;SI-CHECK: V_LSHL_B64 {{v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\], v[0-9]+}}
+;SI-CHECK: V_LSHL_B64 {{v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\], v[0-9]+}}
+;SI-CHECK: V_LSHL_B64 {{v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\], v[0-9]+}}
+
+define void @shl_v4i64(<4 x i64> addrspace(1)* %out, <4 x i64> addrspace(1)* %in) {
+  %b_ptr = getelementptr <4 x i64> addrspace(1)* %in, i64 1
+  %a = load <4 x i64> addrspace(1) * %in
+  %b = load <4 x i64> addrspace(1) * %b_ptr
+  %result = shl <4 x i64> %a, %b
+  store <4 x i64> %result, <4 x i64> addrspace(1)* %out
+  ret void
+}