[ARM] Add support for the MVE long shift instructions

MVE adds the lsll, lsrl and asrl instructions, which perform a shift on a 64 bit value separated into two 32 bit registers.

The Expand64BitShift function is modified to accept ISD::SHL, ISD::SRL and ISD::SRA and convert it into the appropriate opcode in ARMISD. An SHL is converted into an lsll, an SRL is converted into an lsrl for the immediate form and a negation and lsll for the register form, and SRA is converted into an asrl.

test/CodeGen/ARM/shift_parts.ll is added to test the logic of emitting these instructions.

Differential Revision: https://reviews.llvm.org/D63430

llvm-svn: 364654

4 files changed, 85 insertions, 7 deletions

diff --git a/llvm/lib/Target/ARM/ARMISelLowering.cpp b/llvm/lib/Target/ARM/ARMISelLowering.cpp
index ae7e2b6e34d..d2ef680524a 100644
--- a/llvm/lib/Target/ARM/ARMISelLowering.cpp
+++ b/llvm/lib/Target/ARM/ARMISelLowering.cpp
@@ -932,6 +932,11 @@ ARMTargetLowering::ARMTargetLowering(const TargetMachine &TM,
   setOperationAction(ISD::SRA,       MVT::i64, Custom);
   setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::i64, Custom);
 
+  // MVE lowers 64 bit shifts to lsll and lsrl
+  // assuming that ISD::SRL and SRA of i64 are already marked custom
+  if (Subtarget->hasMVEIntegerOps())
+    setOperationAction(ISD::SHL, MVT::i64, Custom);
+
   // Expand to __aeabi_l{lsl,lsr,asr} calls for Thumb1.
   if (Subtarget->isThumb1Only()) {
     setOperationAction(ISD::SHL_PARTS, MVT::i32, Expand);
@@ -1411,6 +1416,10 @@ const char *ARMTargetLowering::getTargetNodeName(unsigned Opcode) const {
   case ARMISD::SSAT:          return "ARMISD::SSAT";
   case ARMISD::USAT:          return "ARMISD::USAT";
 
+  case ARMISD::ASRL:          return "ARMISD::ASRL";
+  case ARMISD::LSRL:          return "ARMISD::LSRL";
+  case ARMISD::LSLL:          return "ARMISD::LSLL";
+
   case ARMISD::SRL_FLAG:      return "ARMISD::SRL_FLAG";
   case ARMISD::SRA_FLAG:      return "ARMISD::SRA_FLAG";
   case ARMISD::RRX:           return "ARMISD::RRX";
@@ -5619,11 +5628,54 @@ static SDValue Expand64BitShift(SDNode *N, SelectionDAG &DAG,
   if (VT != MVT::i64)
     return SDValue();
 
-  assert((N->getOpcode() == ISD::SRL || N->getOpcode() == ISD::SRA) &&
+  assert((N->getOpcode() == ISD::SRL || N->getOpcode() == ISD::SRA ||
+          N->getOpcode() == ISD::SHL) &&
          "Unknown shift to lower!");
 
+  unsigned ShOpc = N->getOpcode();
+  if (ST->hasMVEIntegerOps()) {
+    SDValue ShAmt = N->getOperand(1);
+    unsigned ShPartsOpc = ARMISD::LSLL;
+    ConstantSDNode *Con = dyn_cast<ConstantSDNode>(ShAmt);
+
+    // If the shift amount is greater than 32 then do the default optimisation
+    if (Con && Con->getZExtValue() > 32)
+      return SDValue();
+
+    // Extract the lower 32 bits of the shift amount if it's an i64
+    if (ShAmt->getValueType(0) == MVT::i64)
+      ShAmt = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32, ShAmt,
+                          DAG.getConstant(0, dl, MVT::i32));
+
+    if (ShOpc == ISD::SRL) {
+      if (!Con)
+        // There is no t2LSRLr instruction so negate and perform an lsll if the
+        // shift amount is in a register, emulating a right shift.
+        ShAmt = DAG.getNode(ISD::SUB, dl, MVT::i32,
+                            DAG.getConstant(0, dl, MVT::i32), ShAmt);
+      else
+        // Else generate an lsrl on the immediate shift amount
+        ShPartsOpc = ARMISD::LSRL;
+    } else if (ShOpc == ISD::SRA)
+      ShPartsOpc = ARMISD::ASRL;
+
+    // Lower 32 bits of the destination/source
+    SDValue Lo = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32, N->getOperand(0),
+                             DAG.getConstant(0, dl, MVT::i32));
+    // Upper 32 bits of the destination/source
+    SDValue Hi = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32, N->getOperand(0),
+                             DAG.getConstant(1, dl, MVT::i32));
+
+    // Generate the shift operation as computed above
+    Lo = DAG.getNode(ShPartsOpc, dl, DAG.getVTList(MVT::i32, MVT::i32), Lo, Hi,
+                     ShAmt);
+    // The upper 32 bits come from the second return value of lsll
+    Hi = SDValue(Lo.getNode(), 1);
+    return DAG.getNode(ISD::BUILD_PAIR, dl, MVT::i64, Lo, Hi);
+  }
+
   // We only lower SRA, SRL of 1 here, all others use generic lowering.
-  if (!isOneConstant(N->getOperand(1)))
+  if (!isOneConstant(N->getOperand(1)) || N->getOpcode() == ISD::SHL)
     return SDValue();
 
   // If we are in thumb mode, we don't have RRX.
@@ -8291,6 +8343,7 @@ void ARMTargetLowering::ReplaceNodeResults(SDNode *N,
     break;
   case ISD::SRL:
   case ISD::SRA:
+  case ISD::SHL:
     Res = Expand64BitShift(N, DAG, Subtarget);
     break;
   case ISD::SREM:
diff --git a/llvm/lib/Target/ARM/ARMISelLowering.h b/llvm/lib/Target/ARM/ARMISelLowering.h
index 10a576caa67..1fb89c7eff4 100644
--- a/llvm/lib/Target/ARM/ARMISelLowering.h
+++ b/llvm/lib/Target/ARM/ARMISelLowering.h
@@ -76,6 +76,10 @@ class VectorType;
 
       PIC_ADD,      // Add with a PC operand and a PIC label.
 
+      ASRL,         // MVE long arithmetic shift right.
+      LSRL,         // MVE long shift right.
+      LSLL,         // MVE long shift left.
+
       CMP,          // ARM compare instructions.
       CMN,          // ARM CMN instructions.
       CMPZ,         // ARM compare that sets only Z flag.
diff --git a/llvm/lib/Target/ARM/ARMInstrInfo.td b/llvm/lib/Target/ARM/ARMInstrInfo.td
index 8557bccbdbe..3799acd855e 100644
--- a/llvm/lib/Target/ARM/ARMInstrInfo.td
+++ b/llvm/lib/Target/ARM/ARMInstrInfo.td
@@ -99,6 +99,13 @@ def SDT_LongMac  : SDTypeProfile<2, 4, [SDTCisVT<0, i32>,
                                         SDTCisSameAs<0, 4>,
                                         SDTCisSameAs<0, 5>]>;
 
+// ARMlsll, ARMlsrl, ARMasrl
+def SDT_ARMIntShiftParts : SDTypeProfile<2, 3, [SDTCisSameAs<0, 1>,
+                                              SDTCisSameAs<0, 2>,
+                                              SDTCisSameAs<0, 3>,
+                                              SDTCisInt<0>,
+                                              SDTCisInt<4>]>;
+
 def ARMSmlald        : SDNode<"ARMISD::SMLALD", SDT_LongMac>;
 def ARMSmlaldx       : SDNode<"ARMISD::SMLALDX", SDT_LongMac>;
 def ARMSmlsld        : SDNode<"ARMISD::SMLSLD", SDT_LongMac>;
@@ -171,6 +178,10 @@ def ARMcmpZ          : SDNode<"ARMISD::CMPZ", SDT_ARMCmp,
 
 def ARMpic_add       : SDNode<"ARMISD::PIC_ADD", SDT_ARMPICAdd>;
 
+def ARMasrl          : SDNode<"ARMISD::ASRL", SDT_ARMIntShiftParts, []>;
+def ARMlsrl          : SDNode<"ARMISD::LSRL", SDT_ARMIntShiftParts, []>;
+def ARMlsll          : SDNode<"ARMISD::LSLL", SDT_ARMIntShiftParts, []>;
+
 def ARMsrl_flag      : SDNode<"ARMISD::SRL_FLAG", SDTIntUnaryOp, [SDNPOutGlue]>;
 def ARMsra_flag      : SDNode<"ARMISD::SRA_FLAG", SDTIntUnaryOp, [SDNPOutGlue]>;
 def ARMrrx           : SDNode<"ARMISD::RRX"     , SDTIntUnaryOp, [SDNPInGlue ]>;
diff --git a/llvm/lib/Target/ARM/ARMInstrMVE.td b/llvm/lib/Target/ARM/ARMInstrMVE.td
index c92802d6d97..a4221c5014e 100644
--- a/llvm/lib/Target/ARM/ARMInstrMVE.td
+++ b/llvm/lib/Target/ARM/ARMInstrMVE.td
@@ -427,11 +427,21 @@ class MVE_ScalarShiftDRegReg<string iname, bit op5, bit op16,
   let DecoderMethod = "DecodeMVEOverlappingLongShift";
 }
 
-def MVE_ASRLr   : MVE_ScalarShiftDRegReg<"asrl",    0b1,  0b0>;
-def MVE_ASRLi   : MVE_ScalarShiftDRegImm<"asrl",    0b10, ?>;
-def MVE_LSLLr   : MVE_ScalarShiftDRegReg<"lsll",    0b0,  0b0>;
-def MVE_LSLLi   : MVE_ScalarShiftDRegImm<"lsll",    0b00, ?>;
-def MVE_LSRL    : MVE_ScalarShiftDRegImm<"lsrl",    0b01, ?>;
+def MVE_ASRLr   : MVE_ScalarShiftDRegReg<"asrl",    0b1,  0b0, [(set tGPREven:$RdaLo, tGPROdd:$RdaHi,
+                                        (ARMasrl tGPREven:$RdaLo_src,
+                                        tGPROdd:$RdaHi_src, rGPR:$Rm))]>;
+def MVE_ASRLi   : MVE_ScalarShiftDRegImm<"asrl",    0b10, ?, [(set tGPREven:$RdaLo, tGPROdd:$RdaHi,
+                                        (ARMasrl tGPREven:$RdaLo_src,
+                                        tGPROdd:$RdaHi_src, (i32 imm:$imm)))]>;
+def MVE_LSLLr   : MVE_ScalarShiftDRegReg<"lsll",    0b0,  0b0, [(set tGPREven:$RdaLo, tGPROdd:$RdaHi,
+                                        (ARMlsll tGPREven:$RdaLo_src,
+                                        tGPROdd:$RdaHi_src, rGPR:$Rm))]>;
+def MVE_LSLLi   : MVE_ScalarShiftDRegImm<"lsll",    0b00, ?, [(set tGPREven:$RdaLo, tGPROdd:$RdaHi,
+                                        (ARMlsll tGPREven:$RdaLo_src,
+                                        tGPROdd:$RdaHi_src, (i32 imm:$imm)))]>;
+def MVE_LSRL    : MVE_ScalarShiftDRegImm<"lsrl",    0b01, ?, [(set tGPREven:$RdaLo, tGPROdd:$RdaHi,
+                                        (ARMlsrl tGPREven:$RdaLo_src,
+                                        tGPROdd:$RdaHi_src, (i32 imm:$imm)))]>;
 
 def MVE_SQRSHRL : MVE_ScalarShiftDRegReg<"sqrshrl", 0b1,  0b1>;
 def MVE_SQSHLL  : MVE_ScalarShiftDRegImm<"sqshll",  0b11, 0b1>;