[llvm-branch-commits] [llvm] 76d1026 - [RISCV] Custom legalize bswap/bitreverse to GREVI with Zbp extension to enable them to combine with other GREVI instructions
Craig Topper via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Mon Nov 30 08:36:20 PST 2020
Author: Craig Topper
Date: 2020-11-30T08:30:40-08:00
New Revision: 76d1026b59bd04bd31645ab37dd82d1d89daa6d9
URL: https://github.com/llvm/llvm-project/commit/76d1026b59bd04bd31645ab37dd82d1d89daa6d9
DIFF: https://github.com/llvm/llvm-project/commit/76d1026b59bd04bd31645ab37dd82d1d89daa6d9.diff
LOG: [RISCV] Custom legalize bswap/bitreverse to GREVI with Zbp extension to enable them to combine with other GREVI instructions
This enables bswap/bitreverse to combine with other GREVI patterns or each other without needing to add more special cases to the DAG combine or new DAG combines.
I've also enabled the existing GREVI combine for GREVIW so that it can pick up the i32 bswap/bitreverse on RV64 after they've been type legalized to GREVIW.
Differential Revision: https://reviews.llvm.org/D92253
Added:
Modified:
llvm/lib/Target/RISCV/RISCVISelLowering.cpp
llvm/lib/Target/RISCV/RISCVInstrInfoB.td
llvm/test/CodeGen/RISCV/rv32Zbp.ll
llvm/test/CodeGen/RISCV/rv64Zbp.ll
Removed:
################################################################################
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 380b1a7d9c42..f1c25512000e 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -162,7 +162,8 @@ RISCVTargetLowering::RISCVTargetLowering(const TargetMachine &TM,
}
if (Subtarget.hasStdExtZbp()) {
- setOperationAction(ISD::BITREVERSE, XLenVT, Legal);
+ setOperationAction(ISD::BITREVERSE, XLenVT, Custom);
+ setOperationAction(ISD::BSWAP, XLenVT, Custom);
if (Subtarget.is64Bit()) {
setOperationAction(ISD::BITREVERSE, MVT::i32, Custom);
@@ -495,6 +496,20 @@ SDValue RISCVTargetLowering::LowerOperation(SDValue Op,
}
case ISD::INTRINSIC_WO_CHAIN:
return LowerINTRINSIC_WO_CHAIN(Op, DAG);
+ case ISD::BSWAP:
+ case ISD::BITREVERSE: {
+ // Convert BSWAP/BITREVERSE to GREVI to enable GREVI combinining.
+ assert(Subtarget.hasStdExtZbp() && "Unexpected custom legalisation");
+ MVT VT = Op.getSimpleValueType();
+ SDLoc DL(Op);
+ // Start with the maximum immediate value which is the bitwidth - 1.
+ unsigned Imm = VT.getSizeInBits() - 1;
+ // If this is BSWAP rather than BITREVERSE, clear the lower 3 bits.
+ if (Op.getOpcode() == ISD::BSWAP)
+ Imm &= ~0x7U;
+ return DAG.getNode(RISCVISD::GREVI, DL, VT, Op.getOperand(0),
+ DAG.getTargetConstant(Imm, DL, Subtarget.getXLenVT()));
+ }
}
}
@@ -1288,6 +1303,29 @@ static SDValue combineORToGORC(SDValue Op, SelectionDAG &DAG,
return SDValue();
}
+static SDValue combineGREVI(SDNode *N, SelectionDAG &DAG,
+ const RISCVSubtarget &Subtarget) {
+ // Combine (GREVI (GREVI x, C2), C1) -> (GREVI x, C1^C2) when C1^C2 is
+ // non-zero, and to x when it is. Any repeated GREVI stage undoes itself.
+ uint64_t ShAmt1 = N->getConstantOperandVal(1);
+ SDValue GREVSrc = N->getOperand(0);
+
+ if (GREVSrc->getOpcode() != N->getOpcode())
+ return SDValue();
+
+ uint64_t ShAmt2 = GREVSrc.getConstantOperandVal(1);
+ GREVSrc = GREVSrc->getOperand(0);
+
+ uint64_t CombinedShAmt = ShAmt1 ^ ShAmt2;
+ if (CombinedShAmt == 0)
+ return GREVSrc;
+
+ SDLoc DL(N);
+ return DAG.getNode(
+ N->getOpcode(), DL, N->getValueType(0), GREVSrc,
+ DAG.getTargetConstant(CombinedShAmt, DL, Subtarget.getXLenVT()));
+}
+
SDValue RISCVTargetLowering::PerformDAGCombine(SDNode *N,
DAGCombinerInfo &DCI) const {
SelectionDAG &DAG = DCI.DAG;
@@ -1383,6 +1421,11 @@ SDValue RISCVTargetLowering::PerformDAGCombine(SDNode *N,
DCI.AddToWorklist(N);
return SDValue(N, 0);
}
+
+ if (N->getOpcode() == RISCVISD::GREVIW)
+ if (SDValue V = combineGREVI(N, DCI.DAG, Subtarget))
+ return V;
+
break;
}
case RISCVISD::FMV_X_ANYEXTW_RV64: {
@@ -1415,23 +1458,8 @@ SDValue RISCVTargetLowering::PerformDAGCombine(SDNode *N,
return DAG.getNode(ISD::AND, DL, MVT::i64, NewFMV,
DAG.getConstant(~SignBit, DL, MVT::i64));
}
- case RISCVISD::GREVI: {
- // Combine (GREVI (GREVI x, C2), C1) -> (GREVI x, C1^C2) when C1^C2 is
- // non-zero, and to x when it is. Any repeated GREVI stage undoes itself.
- SDLoc DL(N);
- auto GREVSrc = N->getOperand(0);
- uint64_t ShAmt1 = N->getConstantOperandVal(1);
- if (GREVSrc->getOpcode() != RISCVISD::GREVI)
- break;
- uint64_t ShAmt2 = GREVSrc.getConstantOperandVal(1);
- GREVSrc = GREVSrc->getOperand(0);
- uint64_t CombinedShAmt = ShAmt1 ^ ShAmt2;
- if (CombinedShAmt == 0)
- return GREVSrc;
- return DAG.getNode(
- RISCVISD::GREVI, DL, N->getValueType(0), GREVSrc,
- DAG.getTargetConstant(CombinedShAmt, DL, Subtarget.getXLenVT()));
- }
+ case RISCVISD::GREVI:
+ return combineGREVI(N, DCI.DAG, Subtarget);
case ISD::OR:
if (auto GREV = combineORToGREV(SDValue(N, 0), DCI.DAG, Subtarget))
return GREV;
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoB.td b/llvm/lib/Target/RISCV/RISCVInstrInfoB.td
index 3926563f568b..8d9748d80ba3 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoB.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoB.td
@@ -730,17 +730,10 @@ def : Pat<(riscv_gorci GPR:$rs1, timm:$shamt), (GORCI GPR:$rs1, timm:$shamt)>;
} // Predicates = [HasStdExtZbp]
let Predicates = [HasStdExtZbp, IsRV32] in {
-def : Pat<(rotr (bswap GPR:$rs1), (i32 16)), (GREVI GPR:$rs1, 8)>;
-def : Pat<(rotl (bswap GPR:$rs1), (i32 16)), (GREVI GPR:$rs1, 8)>;
-def : Pat<(bswap GPR:$rs1), (GREVI GPR:$rs1, (i32 24))>;
-def : Pat<(bitreverse GPR:$rs1), (GREVI GPR:$rs1, (i32 31))>;
+def : Pat<(rotr (riscv_grevi GPR:$rs1, (i32 24)), (i32 16)), (GREVI GPR:$rs1, 8)>;
+def : Pat<(rotl (riscv_grevi GPR:$rs1, (i32 24)), (i32 16)), (GREVI GPR:$rs1, 8)>;
} // Predicates = [HasStdExtZbp, IsRV32]
-let Predicates = [HasStdExtZbp, IsRV64] in {
-def : Pat<(bswap GPR:$rs1), (GREVI GPR:$rs1, (i64 56))>;
-def : Pat<(bitreverse GPR:$rs1), (GREVI GPR:$rs1, (i64 63))>;
-} // Predicates = [HasStdExtZbp, IsRV64]
-
let Predicates = [HasStdExtZbt] in {
def : Pat<(or (and (not GPR:$rs2), GPR:$rs3), (and GPR:$rs2, GPR:$rs1)),
(CMIX GPR:$rs1, GPR:$rs2, GPR:$rs3)>;
diff --git a/llvm/test/CodeGen/RISCV/rv32Zbp.ll b/llvm/test/CodeGen/RISCV/rv32Zbp.ll
index 8442c8df8c0b..f1b8629f021d 100644
--- a/llvm/test/CodeGen/RISCV/rv32Zbp.ll
+++ b/llvm/test/CodeGen/RISCV/rv32Zbp.ll
@@ -1849,6 +1849,175 @@ define i32 @bswap_rotl_i32(i32 %a) {
ret i32 %2
}
+define i32 @bitreverse_bswap_i32(i32 %a) {
+; RV32I-LABEL: bitreverse_bswap_i32:
+; RV32I: # %bb.0:
+; RV32I-NEXT: srli a1, a0, 8
+; RV32I-NEXT: lui a2, 16
+; RV32I-NEXT: addi a2, a2, -256
+; RV32I-NEXT: and a1, a1, a2
+; RV32I-NEXT: srli a3, a0, 24
+; RV32I-NEXT: or a1, a1, a3
+; RV32I-NEXT: slli a3, a0, 8
+; RV32I-NEXT: lui a4, 4080
+; RV32I-NEXT: and a3, a3, a4
+; RV32I-NEXT: slli a0, a0, 24
+; RV32I-NEXT: or a0, a0, a3
+; RV32I-NEXT: or a0, a0, a1
+; RV32I-NEXT: lui a1, 61681
+; RV32I-NEXT: addi a1, a1, -241
+; RV32I-NEXT: and a1, a0, a1
+; RV32I-NEXT: slli a1, a1, 4
+; RV32I-NEXT: lui a3, 986895
+; RV32I-NEXT: addi a3, a3, 240
+; RV32I-NEXT: and a0, a0, a3
+; RV32I-NEXT: srli a0, a0, 4
+; RV32I-NEXT: or a0, a0, a1
+; RV32I-NEXT: lui a1, 209715
+; RV32I-NEXT: addi a1, a1, 819
+; RV32I-NEXT: and a1, a0, a1
+; RV32I-NEXT: slli a1, a1, 2
+; RV32I-NEXT: lui a3, 838861
+; RV32I-NEXT: addi a3, a3, -820
+; RV32I-NEXT: and a0, a0, a3
+; RV32I-NEXT: srli a0, a0, 2
+; RV32I-NEXT: or a0, a0, a1
+; RV32I-NEXT: lui a1, 349525
+; RV32I-NEXT: addi a1, a1, 1365
+; RV32I-NEXT: and a1, a0, a1
+; RV32I-NEXT: slli a1, a1, 1
+; RV32I-NEXT: lui a3, 699051
+; RV32I-NEXT: addi a3, a3, -1366
+; RV32I-NEXT: and a0, a0, a3
+; RV32I-NEXT: srli a0, a0, 1
+; RV32I-NEXT: or a0, a0, a1
+; RV32I-NEXT: srli a1, a0, 8
+; RV32I-NEXT: and a1, a1, a2
+; RV32I-NEXT: srli a2, a0, 24
+; RV32I-NEXT: or a1, a1, a2
+; RV32I-NEXT: slli a2, a0, 8
+; RV32I-NEXT: and a2, a2, a4
+; RV32I-NEXT: slli a0, a0, 24
+; RV32I-NEXT: or a0, a0, a2
+; RV32I-NEXT: or a0, a0, a1
+; RV32I-NEXT: ret
+;
+; RV32IB-LABEL: bitreverse_bswap_i32:
+; RV32IB: # %bb.0:
+; RV32IB-NEXT: rev.b a0, a0
+; RV32IB-NEXT: ret
+;
+; RV32IBP-LABEL: bitreverse_bswap_i32:
+; RV32IBP: # %bb.0:
+; RV32IBP-NEXT: rev.b a0, a0
+; RV32IBP-NEXT: ret
+ %1 = call i32 @llvm.bitreverse.i32(i32 %a)
+ %2 = call i32 @llvm.bswap.i32(i32 %1)
+ ret i32 %2
+}
+
+define i64 @bitreverse_bswap_i64(i64 %a) {
+; RV32I-LABEL: bitreverse_bswap_i64:
+; RV32I: # %bb.0:
+; RV32I-NEXT: srli a3, a1, 8
+; RV32I-NEXT: lui a2, 16
+; RV32I-NEXT: addi t1, a2, -256
+; RV32I-NEXT: and a3, a3, t1
+; RV32I-NEXT: srli a4, a1, 24
+; RV32I-NEXT: or a4, a3, a4
+; RV32I-NEXT: slli a5, a1, 8
+; RV32I-NEXT: lui a6, 4080
+; RV32I-NEXT: and a5, a5, a6
+; RV32I-NEXT: slli a1, a1, 24
+; RV32I-NEXT: or a1, a1, a5
+; RV32I-NEXT: or a1, a1, a4
+; RV32I-NEXT: lui a4, 61681
+; RV32I-NEXT: addi a7, a4, -241
+; RV32I-NEXT: and a5, a1, a7
+; RV32I-NEXT: slli a5, a5, 4
+; RV32I-NEXT: lui a3, 986895
+; RV32I-NEXT: addi t0, a3, 240
+; RV32I-NEXT: and a1, a1, t0
+; RV32I-NEXT: srli a1, a1, 4
+; RV32I-NEXT: or a1, a1, a5
+; RV32I-NEXT: lui a5, 209715
+; RV32I-NEXT: addi t2, a5, 819
+; RV32I-NEXT: and a4, a1, t2
+; RV32I-NEXT: slli a4, a4, 2
+; RV32I-NEXT: lui a3, 838861
+; RV32I-NEXT: addi t3, a3, -820
+; RV32I-NEXT: and a1, a1, t3
+; RV32I-NEXT: srli a1, a1, 2
+; RV32I-NEXT: or a1, a1, a4
+; RV32I-NEXT: lui a4, 349525
+; RV32I-NEXT: addi a4, a4, 1365
+; RV32I-NEXT: and a2, a1, a4
+; RV32I-NEXT: slli a2, a2, 1
+; RV32I-NEXT: lui a5, 699051
+; RV32I-NEXT: addi a5, a5, -1366
+; RV32I-NEXT: and a1, a1, a5
+; RV32I-NEXT: srli a1, a1, 1
+; RV32I-NEXT: or a1, a1, a2
+; RV32I-NEXT: srli a2, a0, 8
+; RV32I-NEXT: and a2, a2, t1
+; RV32I-NEXT: srli a3, a0, 24
+; RV32I-NEXT: or a2, a2, a3
+; RV32I-NEXT: slli a3, a0, 8
+; RV32I-NEXT: and a3, a3, a6
+; RV32I-NEXT: slli a0, a0, 24
+; RV32I-NEXT: or a0, a0, a3
+; RV32I-NEXT: or a0, a0, a2
+; RV32I-NEXT: and a2, a0, a7
+; RV32I-NEXT: slli a2, a2, 4
+; RV32I-NEXT: and a0, a0, t0
+; RV32I-NEXT: srli a0, a0, 4
+; RV32I-NEXT: or a0, a0, a2
+; RV32I-NEXT: and a2, a0, t2
+; RV32I-NEXT: slli a2, a2, 2
+; RV32I-NEXT: and a0, a0, t3
+; RV32I-NEXT: srli a0, a0, 2
+; RV32I-NEXT: or a0, a0, a2
+; RV32I-NEXT: and a2, a0, a4
+; RV32I-NEXT: slli a2, a2, 1
+; RV32I-NEXT: and a0, a0, a5
+; RV32I-NEXT: srli a0, a0, 1
+; RV32I-NEXT: or a0, a0, a2
+; RV32I-NEXT: srli a2, a0, 8
+; RV32I-NEXT: and a2, a2, t1
+; RV32I-NEXT: srli a3, a0, 24
+; RV32I-NEXT: or a2, a2, a3
+; RV32I-NEXT: slli a3, a0, 8
+; RV32I-NEXT: and a3, a3, a6
+; RV32I-NEXT: slli a0, a0, 24
+; RV32I-NEXT: or a0, a0, a3
+; RV32I-NEXT: or a0, a0, a2
+; RV32I-NEXT: srli a2, a1, 8
+; RV32I-NEXT: and a2, a2, t1
+; RV32I-NEXT: srli a3, a1, 24
+; RV32I-NEXT: or a2, a2, a3
+; RV32I-NEXT: slli a3, a1, 8
+; RV32I-NEXT: and a3, a3, a6
+; RV32I-NEXT: slli a1, a1, 24
+; RV32I-NEXT: or a1, a1, a3
+; RV32I-NEXT: or a1, a1, a2
+; RV32I-NEXT: ret
+;
+; RV32IB-LABEL: bitreverse_bswap_i64:
+; RV32IB: # %bb.0:
+; RV32IB-NEXT: rev.b a0, a0
+; RV32IB-NEXT: rev.b a1, a1
+; RV32IB-NEXT: ret
+;
+; RV32IBP-LABEL: bitreverse_bswap_i64:
+; RV32IBP: # %bb.0:
+; RV32IBP-NEXT: rev.b a0, a0
+; RV32IBP-NEXT: rev.b a1, a1
+; RV32IBP-NEXT: ret
+ %1 = call i64 @llvm.bitreverse.i64(i64 %a)
+ %2 = call i64 @llvm.bswap.i64(i64 %1)
+ ret i64 %2
+}
+
define i32 @shfl1_i32(i32 %a, i32 %b) nounwind {
; RV32I-LABEL: shfl1_i32:
; RV32I: # %bb.0:
diff --git a/llvm/test/CodeGen/RISCV/rv64Zbp.ll b/llvm/test/CodeGen/RISCV/rv64Zbp.ll
index b1f270e00b8f..367255d1c3f0 100644
--- a/llvm/test/CodeGen/RISCV/rv64Zbp.ll
+++ b/llvm/test/CodeGen/RISCV/rv64Zbp.ll
@@ -2377,6 +2377,262 @@ define i32 @bswap_rotl_i32(i32 %a) {
ret i32 %2
}
+define i32 @bitreverse_bswap_i32(i32 %a) {
+; RV64I-LABEL: bitreverse_bswap_i32:
+; RV64I: # %bb.0:
+; RV64I-NEXT: srli a1, a0, 24
+; RV64I-NEXT: lui a2, 4080
+; RV64I-NEXT: and a1, a1, a2
+; RV64I-NEXT: srli a2, a0, 8
+; RV64I-NEXT: addi a3, zero, 255
+; RV64I-NEXT: slli a4, a3, 24
+; RV64I-NEXT: and a2, a2, a4
+; RV64I-NEXT: or a1, a2, a1
+; RV64I-NEXT: srli a2, a0, 40
+; RV64I-NEXT: lui a4, 16
+; RV64I-NEXT: addiw a4, a4, -256
+; RV64I-NEXT: and a2, a2, a4
+; RV64I-NEXT: srli a4, a0, 56
+; RV64I-NEXT: or a2, a2, a4
+; RV64I-NEXT: or a4, a1, a2
+; RV64I-NEXT: slli a1, a0, 8
+; RV64I-NEXT: slli a2, a3, 32
+; RV64I-NEXT: and a2, a1, a2
+; RV64I-NEXT: slli a5, a0, 24
+; RV64I-NEXT: slli a6, a3, 40
+; RV64I-NEXT: and a5, a5, a6
+; RV64I-NEXT: or a5, a5, a2
+; RV64I-NEXT: slli a1, a0, 40
+; RV64I-NEXT: slli a2, a3, 48
+; RV64I-NEXT: and a1, a1, a2
+; RV64I-NEXT: slli a0, a0, 56
+; RV64I-NEXT: or a0, a0, a1
+; RV64I-NEXT: or a0, a0, a5
+; RV64I-NEXT: or a0, a0, a4
+; RV64I-NEXT: lui a1, 3855
+; RV64I-NEXT: addiw a1, a1, 241
+; RV64I-NEXT: slli a1, a1, 12
+; RV64I-NEXT: addi a1, a1, -241
+; RV64I-NEXT: slli a1, a1, 12
+; RV64I-NEXT: addi a1, a1, 241
+; RV64I-NEXT: slli a1, a1, 12
+; RV64I-NEXT: addi a1, a1, -241
+; RV64I-NEXT: and a1, a0, a1
+; RV64I-NEXT: slli a1, a1, 4
+; RV64I-NEXT: lui a3, 1044721
+; RV64I-NEXT: addiw a3, a3, -241
+; RV64I-NEXT: slli a3, a3, 12
+; RV64I-NEXT: addi a3, a3, 241
+; RV64I-NEXT: slli a3, a3, 12
+; RV64I-NEXT: addi a3, a3, -241
+; RV64I-NEXT: slli a3, a3, 12
+; RV64I-NEXT: addi a3, a3, 240
+; RV64I-NEXT: and a0, a0, a3
+; RV64I-NEXT: srli a0, a0, 4
+; RV64I-NEXT: or a0, a0, a1
+; RV64I-NEXT: lui a1, 13107
+; RV64I-NEXT: addiw a1, a1, 819
+; RV64I-NEXT: slli a1, a1, 12
+; RV64I-NEXT: addi a1, a1, 819
+; RV64I-NEXT: slli a1, a1, 12
+; RV64I-NEXT: addi a1, a1, 819
+; RV64I-NEXT: slli a1, a1, 12
+; RV64I-NEXT: addi a1, a1, 819
+; RV64I-NEXT: and a1, a0, a1
+; RV64I-NEXT: slli a1, a1, 2
+; RV64I-NEXT: lui a3, 1035469
+; RV64I-NEXT: addiw a3, a3, -819
+; RV64I-NEXT: slli a3, a3, 12
+; RV64I-NEXT: addi a3, a3, -819
+; RV64I-NEXT: slli a3, a3, 12
+; RV64I-NEXT: addi a3, a3, -819
+; RV64I-NEXT: slli a3, a3, 12
+; RV64I-NEXT: addi a3, a3, -820
+; RV64I-NEXT: and a0, a0, a3
+; RV64I-NEXT: srli a0, a0, 2
+; RV64I-NEXT: or a0, a0, a1
+; RV64I-NEXT: lui a1, 21845
+; RV64I-NEXT: addiw a1, a1, 1365
+; RV64I-NEXT: slli a1, a1, 12
+; RV64I-NEXT: addi a1, a1, 1365
+; RV64I-NEXT: slli a1, a1, 12
+; RV64I-NEXT: addi a1, a1, 1365
+; RV64I-NEXT: slli a1, a1, 12
+; RV64I-NEXT: addi a1, a1, 1365
+; RV64I-NEXT: and a1, a0, a1
+; RV64I-NEXT: slli a1, a1, 1
+; RV64I-NEXT: lui a3, 1026731
+; RV64I-NEXT: addiw a3, a3, -1365
+; RV64I-NEXT: slli a3, a3, 12
+; RV64I-NEXT: addi a3, a3, -1365
+; RV64I-NEXT: slli a3, a3, 12
+; RV64I-NEXT: addi a3, a3, -1365
+; RV64I-NEXT: slli a3, a3, 12
+; RV64I-NEXT: addi a3, a3, -1366
+; RV64I-NEXT: and a0, a0, a3
+; RV64I-NEXT: srli a0, a0, 1
+; RV64I-NEXT: or a0, a0, a1
+; RV64I-NEXT: slli a1, a0, 24
+; RV64I-NEXT: addi a3, zero, -1
+; RV64I-NEXT: slli a3, a3, 56
+; RV64I-NEXT: and a1, a1, a3
+; RV64I-NEXT: srli a3, a0, 8
+; RV64I-NEXT: and a3, a3, a6
+; RV64I-NEXT: srli a4, a0, 24
+; RV64I-NEXT: or a3, a3, a4
+; RV64I-NEXT: slli a0, a0, 8
+; RV64I-NEXT: and a0, a0, a2
+; RV64I-NEXT: or a0, a1, a0
+; RV64I-NEXT: or a0, a0, a3
+; RV64I-NEXT: srli a0, a0, 32
+; RV64I-NEXT: ret
+;
+; RV64IB-LABEL: bitreverse_bswap_i32:
+; RV64IB: # %bb.0:
+; RV64IB-NEXT: greviw a0, a0, 7
+; RV64IB-NEXT: ret
+;
+; RV64IBP-LABEL: bitreverse_bswap_i32:
+; RV64IBP: # %bb.0:
+; RV64IBP-NEXT: greviw a0, a0, 7
+; RV64IBP-NEXT: ret
+ %1 = call i32 @llvm.bitreverse.i32(i32 %a)
+ %2 = call i32 @llvm.bswap.i32(i32 %1)
+ ret i32 %2
+}
+
+define i64 @bitreverse_bswap_i64(i64 %a) {
+; RV64I-LABEL: bitreverse_bswap_i64:
+; RV64I: # %bb.0:
+; RV64I-NEXT: srli a1, a0, 24
+; RV64I-NEXT: lui a6, 4080
+; RV64I-NEXT: and a1, a1, a6
+; RV64I-NEXT: srli a3, a0, 8
+; RV64I-NEXT: addi a5, zero, 255
+; RV64I-NEXT: slli a7, a5, 24
+; RV64I-NEXT: and a3, a3, a7
+; RV64I-NEXT: or a3, a3, a1
+; RV64I-NEXT: srli a4, a0, 40
+; RV64I-NEXT: lui a1, 16
+; RV64I-NEXT: addiw a1, a1, -256
+; RV64I-NEXT: and a4, a4, a1
+; RV64I-NEXT: srli a2, a0, 56
+; RV64I-NEXT: or a2, a4, a2
+; RV64I-NEXT: or a2, a3, a2
+; RV64I-NEXT: slli a4, a0, 8
+; RV64I-NEXT: slli t0, a5, 32
+; RV64I-NEXT: and a3, a4, t0
+; RV64I-NEXT: slli a4, a0, 24
+; RV64I-NEXT: slli t1, a5, 40
+; RV64I-NEXT: and a4, a4, t1
+; RV64I-NEXT: or a3, a4, a3
+; RV64I-NEXT: slli a4, a0, 40
+; RV64I-NEXT: slli a5, a5, 48
+; RV64I-NEXT: and a4, a4, a5
+; RV64I-NEXT: slli a0, a0, 56
+; RV64I-NEXT: or a0, a0, a4
+; RV64I-NEXT: or a0, a0, a3
+; RV64I-NEXT: or a0, a0, a2
+; RV64I-NEXT: lui a2, 3855
+; RV64I-NEXT: addiw a2, a2, 241
+; RV64I-NEXT: slli a2, a2, 12
+; RV64I-NEXT: addi a2, a2, -241
+; RV64I-NEXT: slli a2, a2, 12
+; RV64I-NEXT: addi a2, a2, 241
+; RV64I-NEXT: slli a2, a2, 12
+; RV64I-NEXT: addi a2, a2, -241
+; RV64I-NEXT: and a2, a0, a2
+; RV64I-NEXT: slli a2, a2, 4
+; RV64I-NEXT: lui a3, 1044721
+; RV64I-NEXT: addiw a3, a3, -241
+; RV64I-NEXT: slli a3, a3, 12
+; RV64I-NEXT: addi a3, a3, 241
+; RV64I-NEXT: slli a3, a3, 12
+; RV64I-NEXT: addi a3, a3, -241
+; RV64I-NEXT: slli a3, a3, 12
+; RV64I-NEXT: addi a3, a3, 240
+; RV64I-NEXT: and a0, a0, a3
+; RV64I-NEXT: srli a0, a0, 4
+; RV64I-NEXT: or a0, a0, a2
+; RV64I-NEXT: lui a2, 13107
+; RV64I-NEXT: addiw a2, a2, 819
+; RV64I-NEXT: slli a2, a2, 12
+; RV64I-NEXT: addi a2, a2, 819
+; RV64I-NEXT: slli a2, a2, 12
+; RV64I-NEXT: addi a2, a2, 819
+; RV64I-NEXT: slli a2, a2, 12
+; RV64I-NEXT: addi a2, a2, 819
+; RV64I-NEXT: and a2, a0, a2
+; RV64I-NEXT: slli a2, a2, 2
+; RV64I-NEXT: lui a3, 1035469
+; RV64I-NEXT: addiw a3, a3, -819
+; RV64I-NEXT: slli a3, a3, 12
+; RV64I-NEXT: addi a3, a3, -819
+; RV64I-NEXT: slli a3, a3, 12
+; RV64I-NEXT: addi a3, a3, -819
+; RV64I-NEXT: slli a3, a3, 12
+; RV64I-NEXT: addi a3, a3, -820
+; RV64I-NEXT: and a0, a0, a3
+; RV64I-NEXT: srli a0, a0, 2
+; RV64I-NEXT: or a0, a0, a2
+; RV64I-NEXT: lui a2, 21845
+; RV64I-NEXT: addiw a2, a2, 1365
+; RV64I-NEXT: slli a2, a2, 12
+; RV64I-NEXT: addi a2, a2, 1365
+; RV64I-NEXT: slli a2, a2, 12
+; RV64I-NEXT: addi a2, a2, 1365
+; RV64I-NEXT: slli a2, a2, 12
+; RV64I-NEXT: addi a2, a2, 1365
+; RV64I-NEXT: and a2, a0, a2
+; RV64I-NEXT: slli a2, a2, 1
+; RV64I-NEXT: lui a3, 1026731
+; RV64I-NEXT: addiw a3, a3, -1365
+; RV64I-NEXT: slli a3, a3, 12
+; RV64I-NEXT: addi a3, a3, -1365
+; RV64I-NEXT: slli a3, a3, 12
+; RV64I-NEXT: addi a3, a3, -1365
+; RV64I-NEXT: slli a3, a3, 12
+; RV64I-NEXT: addi a3, a3, -1366
+; RV64I-NEXT: and a0, a0, a3
+; RV64I-NEXT: srli a0, a0, 1
+; RV64I-NEXT: or a0, a0, a2
+; RV64I-NEXT: srli a2, a0, 40
+; RV64I-NEXT: and a1, a2, a1
+; RV64I-NEXT: srli a2, a0, 56
+; RV64I-NEXT: or a1, a1, a2
+; RV64I-NEXT: srli a2, a0, 24
+; RV64I-NEXT: and a2, a2, a6
+; RV64I-NEXT: srli a3, a0, 8
+; RV64I-NEXT: and a3, a3, a7
+; RV64I-NEXT: or a2, a3, a2
+; RV64I-NEXT: or a1, a2, a1
+; RV64I-NEXT: slli a2, a0, 8
+; RV64I-NEXT: and a2, a2, t0
+; RV64I-NEXT: slli a3, a0, 24
+; RV64I-NEXT: and a3, a3, t1
+; RV64I-NEXT: or a2, a3, a2
+; RV64I-NEXT: slli a3, a0, 40
+; RV64I-NEXT: and a3, a3, a5
+; RV64I-NEXT: slli a0, a0, 56
+; RV64I-NEXT: or a0, a0, a3
+; RV64I-NEXT: or a0, a0, a2
+; RV64I-NEXT: or a0, a0, a1
+; RV64I-NEXT: ret
+;
+; RV64IB-LABEL: bitreverse_bswap_i64:
+; RV64IB: # %bb.0:
+; RV64IB-NEXT: rev.b a0, a0
+; RV64IB-NEXT: ret
+;
+; RV64IBP-LABEL: bitreverse_bswap_i64:
+; RV64IBP: # %bb.0:
+; RV64IBP-NEXT: rev.b a0, a0
+; RV64IBP-NEXT: ret
+ %1 = call i64 @llvm.bitreverse.i64(i64 %a)
+ %2 = call i64 @llvm.bswap.i64(i64 %1)
+ ret i64 %2
+}
+
; There's no [un]shfliw instruction as slliu.w occupies the encoding slot that
; would be occupied by shfliw.
More information about the llvm-branch-commits
mailing list