diff options
author | Tim Northover <tnorthover@apple.com> | 2014-02-10 16:20:29 +0000 |
---|---|---|
committer | Tim Northover <tnorthover@apple.com> | 2014-02-10 16:20:29 +0000 |
commit | 71313f88cb9ad3b30eaede19cbbb4ef1be1ff882 (patch) | |
tree | 0bbc0690df694846ca7c0d3fd00f357d8e118cd0 /lib/Target | |
parent | 5fcb5c92d3c8e659abd7a4c2b9a84be9955c525a (diff) | |
download | llvm-71313f88cb9ad3b30eaede19cbbb4ef1be1ff882.tar.gz llvm-71313f88cb9ad3b30eaede19cbbb4ef1be1ff882.tar.bz2 llvm-71313f88cb9ad3b30eaede19cbbb4ef1be1ff882.tar.xz |
ARM: use natural LLVM IR for vshll instructions
Similarly to the vshrn instructions, these are simple zext/sext + trunc
operations. Using normal LLVM IR should allow for better code, and more sharing
with the AArch64 backend.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201093 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'lib/Target')
-rw-r--r-- | lib/Target/ARM/ARMISelLowering.cpp | 19 | ||||
-rw-r--r-- | lib/Target/ARM/ARMISelLowering.h | 3 | ||||
-rw-r--r-- | lib/Target/ARM/ARMInstrNEON.td | 41 |
3 files changed, 27 insertions, 36 deletions
diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp index 73f1b8f747..414922a69a 100644 --- a/lib/Target/ARM/ARMISelLowering.cpp +++ b/lib/Target/ARM/ARMISelLowering.cpp @@ -1078,9 +1078,6 @@ const char *ARMTargetLowering::getTargetNodeName(unsigned Opcode) const { case ARMISD::VSHL: return "ARMISD::VSHL"; case ARMISD::VSHRs: return "ARMISD::VSHRs"; case ARMISD::VSHRu: return "ARMISD::VSHRu"; - case ARMISD::VSHLLs: return "ARMISD::VSHLLs"; - case ARMISD::VSHLLu: return "ARMISD::VSHLLu"; - case ARMISD::VSHLLi: return "ARMISD::VSHLLi"; case ARMISD::VRSHRs: return "ARMISD::VRSHRs"; case ARMISD::VRSHRu: return "ARMISD::VRSHRu"; case ARMISD::VRSHRN: return "ARMISD::VRSHRN"; @@ -9714,8 +9711,6 @@ static SDValue PerformIntrinsicCombine(SDNode *N, SelectionDAG &DAG) { // loads from a constant pool. case Intrinsic::arm_neon_vshifts: case Intrinsic::arm_neon_vshiftu: - case Intrinsic::arm_neon_vshiftls: - case Intrinsic::arm_neon_vshiftlu: case Intrinsic::arm_neon_vrshifts: case Intrinsic::arm_neon_vrshiftu: case Intrinsic::arm_neon_vrshiftn: @@ -9746,12 +9741,6 @@ static SDValue PerformIntrinsicCombine(SDNode *N, SelectionDAG &DAG) { } return SDValue(); - case Intrinsic::arm_neon_vshiftls: - case Intrinsic::arm_neon_vshiftlu: - if (isVShiftLImm(N->getOperand(2), VT, true, Cnt)) - break; - llvm_unreachable("invalid shift count for vshll intrinsic"); - case Intrinsic::arm_neon_vrshifts: case Intrinsic::arm_neon_vrshiftu: if (isVShiftRImm(N->getOperand(2), VT, false, true, Cnt)) @@ -9791,14 +9780,6 @@ static SDValue PerformIntrinsicCombine(SDNode *N, SelectionDAG &DAG) { case Intrinsic::arm_neon_vshiftu: // Opcode already set above. break; - case Intrinsic::arm_neon_vshiftls: - case Intrinsic::arm_neon_vshiftlu: - if (Cnt == VT.getVectorElementType().getSizeInBits()) - VShiftOpc = ARMISD::VSHLLi; - else - VShiftOpc = (IntNo == Intrinsic::arm_neon_vshiftls ? - ARMISD::VSHLLs : ARMISD::VSHLLu); - break; case Intrinsic::arm_neon_vrshifts: VShiftOpc = ARMISD::VRSHRs; break; case Intrinsic::arm_neon_vrshiftu: diff --git a/lib/Target/ARM/ARMISelLowering.h b/lib/Target/ARM/ARMISelLowering.h index fd54cbb647..bae23a69b5 100644 --- a/lib/Target/ARM/ARMISelLowering.h +++ b/lib/Target/ARM/ARMISelLowering.h @@ -113,9 +113,6 @@ namespace llvm { VSHL, // ...left VSHRs, // ...right (signed) VSHRu, // ...right (unsigned) - VSHLLs, // ...left long (signed) - VSHLLu, // ...left long (unsigned) - VSHLLi, // ...left long (with maximum shift count) // Vector rounding shift by immediate: VRSHRs, // ...right (signed) diff --git a/lib/Target/ARM/ARMInstrNEON.td b/lib/Target/ARM/ARMInstrNEON.td index a1e4159259..4bc82a2e43 100644 --- a/lib/Target/ARM/ARMInstrNEON.td +++ b/lib/Target/ARM/ARMInstrNEON.td @@ -466,9 +466,6 @@ def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>, def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>; def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>; def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>; -def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>; -def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>; -def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>; def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>; def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>; @@ -3038,12 +3035,12 @@ class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4, // Long shift by immediate. class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4, string OpcodeStr, string Dt, - ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode> + ValueType ResTy, ValueType OpTy, Operand ImmTy, + SDPatternOperator OpNode> : N2VImm<op24, op23, op11_8, op7, op6, op4, (outs QPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), N2RegVShLFrm, IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "", - [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm), - (i32 imm:$SIMM))))]>; + [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm), ImmTy:$SIMM)))]>; // Narrow shift by immediate. class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4, @@ -3942,7 +3939,8 @@ multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, // Neon Shift Long operations, // element sizes of 8, 16, 32 bits: multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, - bit op4, string OpcodeStr, string Dt, SDNode OpNode> { + bit op4, string OpcodeStr, string Dt, + SDPatternOperator OpNode> { def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, imm1_7, OpNode> { let Inst{21-19} = 0b001; // imm6 = 001xxx @@ -4947,24 +4945,39 @@ defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", "VSHRu", NEONvshru>; // VSHLL : Vector Shift Left Long -defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>; -defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>; +defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", + PatFrag<(ops node:$LHS, node:$RHS), (NEONvshl (sext node:$LHS), node:$RHS)>>; +defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", + PatFrag<(ops node:$LHS, node:$RHS), (NEONvshl (zext node:$LHS), node:$RHS)>>; // VSHLL : Vector Shift Left Long (with maximum shift count) class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7, bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy, - ValueType OpTy, Operand ImmTy, SDNode OpNode> + ValueType OpTy, Operand ImmTy> : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt, - ResTy, OpTy, ImmTy, OpNode> { + ResTy, OpTy, ImmTy, null_frag> { let Inst{21-16} = op21_16; let DecoderMethod = "DecodeVSHLMaxInstruction"; } def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8", - v8i16, v8i8, imm8, NEONvshlli>; + v8i16, v8i8, imm8>; def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16", - v4i32, v4i16, imm16, NEONvshlli>; + v4i32, v4i16, imm16>; def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32", - v2i64, v2i32, imm32, NEONvshlli>; + v2i64, v2i32, imm32>; + +def : Pat<(v8i16 (NEONvshl (zext (v8i8 DPR:$Rn)), (i32 8))), + (VSHLLi8 DPR:$Rn, 8)>; +def : Pat<(v4i32 (NEONvshl (zext (v4i16 DPR:$Rn)), (i32 16))), + (VSHLLi16 DPR:$Rn, 16)>; +def : Pat<(v2i64 (NEONvshl (zext (v2i32 DPR:$Rn)), (i32 32))), + (VSHLLi32 DPR:$Rn, 32)>; +def : Pat<(v8i16 (NEONvshl (sext (v8i8 DPR:$Rn)), (i32 8))), + (VSHLLi8 DPR:$Rn, 8)>; +def : Pat<(v4i32 (NEONvshl (sext (v4i16 DPR:$Rn)), (i32 16))), + (VSHLLi16 DPR:$Rn, 16)>; +def : Pat<(v2i64 (NEONvshl (sext (v2i32 DPR:$Rn)), (i32 32))), + (VSHLLi32 DPR:$Rn, 32)>; // VSHRN : Vector Shift Right and Narrow defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i", |