https://github.com/wangpc-pp updated https://github.com/llvm/llvm-project/pull/84455
>From 35d0ea085b43a67c092e6263e6ec9d34e66e1453 Mon Sep 17 00:00:00 2001 From: Wang Pengcheng <wangpengcheng...@bytedance.com> Date: Tue, 12 Mar 2024 17:31:47 +0800 Subject: [PATCH 1/2] Reduce copies Created using spr 1.3.4 --- llvm/lib/Target/RISCV/RISCVInstrInfo.cpp | 89 +++++----- llvm/test/CodeGen/RISCV/rvv/vmv-copy.mir | 30 +--- llvm/test/CodeGen/RISCV/rvv/zvlsseg-copy.mir | 175 +++++++------------ 3 files changed, 106 insertions(+), 188 deletions(-) diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp b/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp index 7895e87702c711..9fe5666d6a81f4 100644 --- a/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp +++ b/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp @@ -302,58 +302,38 @@ void RISCVInstrInfo::copyPhysRegVector(MachineBasicBlock &MBB, RISCVII::VLMUL LMul, unsigned NF) const { const TargetRegisterInfo *TRI = STI.getRegisterInfo(); - int I = 0, End = NF, Incr = 1; unsigned SrcEncoding = TRI->getEncodingValue(SrcReg); unsigned DstEncoding = TRI->getEncodingValue(DstReg); unsigned LMulVal; bool Fractional; std::tie(LMulVal, Fractional) = RISCVVType::decodeVLMUL(LMul); assert(!Fractional && "It is impossible be fractional lmul here."); - if (forwardCopyWillClobberTuple(DstEncoding, SrcEncoding, NF * LMulVal)) { - I = NF - 1; - End = -1; - Incr = -1; - } + unsigned NumRegs = NF * LMulVal; + bool ReversedCopy = + forwardCopyWillClobberTuple(DstEncoding, SrcEncoding, NumRegs); - for (; I != End; I += Incr) { + unsigned I = 0; + while (I != NumRegs) { auto GetCopyInfo = - [](RISCVII::VLMUL LMul,unsigned NF) -> std::tuple<unsigned, unsigned, unsigned, unsigned> { - unsigned Opc; - unsigned SubRegIdx; - unsigned VVOpc, VIOpc; - switch (LMul) { - default: - llvm_unreachable("Impossible LMUL for vector register copy."); - case RISCVII::LMUL_1: - Opc = RISCV::VMV1R_V; - SubRegIdx = RISCV::sub_vrm1_0; - VVOpc = RISCV::PseudoVMV_V_V_M1; - VIOpc = RISCV::PseudoVMV_V_I_M1; - break; - case RISCVII::LMUL_2: - Opc = RISCV::VMV2R_V; - SubRegIdx = RISCV::sub_vrm2_0; - VVOpc = RISCV::PseudoVMV_V_V_M2; - VIOpc = RISCV::PseudoVMV_V_I_M2; - break; - case RISCVII::LMUL_4: - Opc = RISCV::VMV4R_V; - SubRegIdx = RISCV::sub_vrm4_0; - VVOpc = RISCV::PseudoVMV_V_V_M4; - VIOpc = RISCV::PseudoVMV_V_I_M4; - break; - case RISCVII::LMUL_8: - assert(NF == 1); - Opc = RISCV::VMV8R_V; - SubRegIdx = RISCV::sub_vrm1_0; // There is no sub_vrm8_0. - VVOpc = RISCV::PseudoVMV_V_V_M8; - VIOpc = RISCV::PseudoVMV_V_I_M8; - break; - } - return {SubRegIdx, Opc, VVOpc, VIOpc}; + [&](unsigned SrcReg, + unsigned DstReg) -> std::tuple<int, const TargetRegisterClass &, + unsigned, unsigned, unsigned> { + unsigned SrcEncoding = TRI->getEncodingValue(SrcReg); + unsigned DstEncoding = TRI->getEncodingValue(DstReg); + if (!(SrcEncoding & 0b111) && !(DstEncoding & 0b111) && I + 8 <= NumRegs) + return {8, RISCV::VRM8RegClass, RISCV::VMV8R_V, RISCV::PseudoVMV_V_V_M8, + RISCV::PseudoVMV_V_I_M8}; + if (!(SrcEncoding & 0b11) && !(DstEncoding & 0b11) && I + 4 <= NumRegs) + return {4, RISCV::VRM4RegClass, RISCV::VMV4R_V, RISCV::PseudoVMV_V_V_M4, + RISCV::PseudoVMV_V_I_M4}; + if (!(SrcEncoding & 0b1) && !(DstEncoding & 0b1) && I + 2 <= NumRegs) + return {2, RISCV::VRM2RegClass, RISCV::VMV2R_V, RISCV::PseudoVMV_V_V_M2, + RISCV::PseudoVMV_V_I_M2}; + return {1, RISCV::VRRegClass, RISCV::VMV1R_V, RISCV::PseudoVMV_V_V_M1, + RISCV::PseudoVMV_V_I_M1}; }; - auto [SubRegIdx, Opc, VVOpc, VIOpc] = GetCopyInfo(LMul, NF); + auto [NumCopied, RegClass, Opc, VVOpc, VIOpc] = GetCopyInfo(SrcReg, DstReg); MachineBasicBlock::const_iterator DefMBBI; if (isConvertibleToVMV_V_V(STI, MBB, MBBI, DefMBBI, LMul)) { @@ -364,6 +344,20 @@ void RISCVInstrInfo::copyPhysRegVector(MachineBasicBlock &MBB, } } + for (MCPhysReg Reg : RegClass.getRegisters()) { + if (TRI->getEncodingValue(Reg) == TRI->getEncodingValue(SrcReg)) { + SrcReg = Reg; + break; + } + } + + for (MCPhysReg Reg : RegClass.getRegisters()) { + if (TRI->getEncodingValue(Reg) == TRI->getEncodingValue(DstReg)) { + DstReg = Reg; + break; + } + } + auto EmitCopy = [&](MCRegister SrcReg, MCRegister DstReg, unsigned Opcode) { auto MIB = BuildMI(MBB, MBBI, DL, get(Opcode), DstReg); bool UseVMV_V_I = RISCV::getRVVMCOpcode(Opcode) == RISCV::VMV_V_I; @@ -385,13 +379,10 @@ void RISCVInstrInfo::copyPhysRegVector(MachineBasicBlock &MBB, } }; - if (NF == 1) { - EmitCopy(SrcReg, DstReg, Opc); - return; - } - - EmitCopy(TRI->getSubReg(SrcReg, SubRegIdx + I), - TRI->getSubReg(DstReg, SubRegIdx + I), Opc); + EmitCopy(SrcReg, DstReg, Opc); + SrcReg = SrcReg.id() + (ReversedCopy ? -NumCopied : NumCopied); + DstReg = DstReg.id() + (ReversedCopy ? -NumCopied : NumCopied); + I += NumCopied; } } diff --git a/llvm/test/CodeGen/RISCV/rvv/vmv-copy.mir b/llvm/test/CodeGen/RISCV/rvv/vmv-copy.mir index 4fa29e174602d0..dd569129db4d72 100644 --- a/llvm/test/CodeGen/RISCV/rvv/vmv-copy.mir +++ b/llvm/test/CodeGen/RISCV/rvv/vmv-copy.mir @@ -8,7 +8,6 @@ tracksRegLiveness: true body: | bb.0: liveins: $x14, $x16 - ; 82 = e32,m4 ; CHECK-LABEL: name: copy_different_lmul ; CHECK: liveins: $x14, $x16 ; CHECK-NEXT: {{ $}} @@ -25,7 +24,6 @@ tracksRegLiveness: true body: | bb.0: liveins: $x14, $x16 - ; 82 = e32,m4 ; CHECK-LABEL: name: copy_convert_to_vmv_v_v ; CHECK: liveins: $x14, $x16 ; CHECK-NEXT: {{ $}} @@ -42,7 +40,6 @@ tracksRegLiveness: true body: | bb.0: liveins: $x14 - ; 82 = e32,m4 ; CHECK-LABEL: name: copy_convert_to_vmv_v_i ; CHECK: liveins: $x14 ; CHECK-NEXT: {{ $}} @@ -59,7 +56,6 @@ tracksRegLiveness: true body: | bb.0: liveins: $x14, $x16 - ; 82 = e32,m4 ; CHECK-LABEL: name: copy_from_whole_load_store ; CHECK: liveins: $x14, $x16 ; CHECK-NEXT: {{ $}} @@ -76,7 +72,6 @@ tracksRegLiveness: true body: | bb.0: liveins: $x14, $x16 - ; 82 = e32,m4 ; CHECK-LABEL: name: copy_with_vleff ; CHECK: liveins: $x14, $x16 ; CHECK-NEXT: {{ $}} @@ -95,8 +90,6 @@ tracksRegLiveness: true body: | bb.0: liveins: $x14, $x16, $x17, $x18 - ; 82 = e32,m4 - ; 73 = e16,m2 ; CHECK-LABEL: name: copy_with_vsetvl_x0_x0_1 ; CHECK: liveins: $x14, $x16, $x17, $x18 ; CHECK-NEXT: {{ $}} @@ -121,8 +114,6 @@ tracksRegLiveness: true body: | bb.0: liveins: $x14, $x16, $x17, $x18 - ; 82 = e32,m4 - ; 73 = e16,m2 ; CHECK-LABEL: name: copy_with_vsetvl_x0_x0_2 ; CHECK: liveins: $x14, $x16, $x17, $x18 ; CHECK-NEXT: {{ $}} @@ -147,8 +138,6 @@ tracksRegLiveness: true body: | bb.0: liveins: $x14, $x16, $x17, $x18 - ; 82 = e32,m4 - ; 73 = e16,m2 ; CHECK-LABEL: name: copy_with_vsetvl_x0_x0_3 ; CHECK: liveins: $x14, $x16, $x17, $x18 ; CHECK-NEXT: {{ $}} @@ -169,7 +158,6 @@ tracksRegLiveness: true body: | bb.0: liveins: $x16, $x17 - ; 73 = e16,m2 ; CHECK-LABEL: name: copy_subregister ; CHECK: liveins: $x16, $x17 ; CHECK-NEXT: {{ $}} @@ -191,8 +179,6 @@ tracksRegLiveness: true body: | bb.0: liveins: $x14, $x16 - ; 82 = e32,m4 - ; 74 = e16,m4 ; CHECK-LABEL: name: copy_with_different_vlmax ; CHECK: liveins: $x14, $x16 ; CHECK-NEXT: {{ $}} @@ -231,7 +217,6 @@ tracksRegLiveness: true body: | bb.0: liveins: $x14, $x16 - ; 80 = e32,m1 ; CHECK-LABEL: name: copy_zvlsseg_reg ; CHECK: liveins: $x14, $x16 ; CHECK-NEXT: {{ $}} @@ -248,14 +233,12 @@ tracksRegLiveness: true body: | bb.0: liveins: $x14, $x16 - ; 80 = e32,m1 ; CHECK-LABEL: name: copy_zvlsseg_reg_2 ; CHECK: liveins: $x14, $x16 ; CHECK-NEXT: {{ $}} ; CHECK-NEXT: $x15 = PseudoVSETVLI $x14, 80 /* e32, m1, ta, mu */, implicit-def $vl, implicit-def $vtype ; CHECK-NEXT: $v8_v9 = PseudoVLSEG2E32_V_M1 undef $v8_v9, killed $x16, $noreg, 5 /* e32 */, 0 /* tu, mu */, implicit $vl, implicit $vtype - ; CHECK-NEXT: $v10 = PseudoVMV_V_V_M1 undef $v10, $v8, $noreg, 5 /* e32 */, 0 /* tu, mu */, implicit $vl, implicit $vtype - ; CHECK-NEXT: $v11 = PseudoVMV_V_V_M1 undef $v11, $v9, $noreg, 5 /* e32 */, 0 /* tu, mu */, implicit $vl, implicit $vtype + ; CHECK-NEXT: $v10m2 = PseudoVMV_V_V_M2 undef $v10m2, $v8m2, $noreg, 5 /* e32 */, 0 /* tu, mu */, implicit $vl, implicit $vtype $x15 = PseudoVSETVLI $x14, 80, implicit-def $vl, implicit-def $vtype $v8_v9 = PseudoVLSEG2E32_V_M1 undef $v8_v9, killed $x16, $noreg, 5, 0, implicit $vl, implicit $vtype $v10_v11 = COPY $v8_v9 @@ -266,7 +249,6 @@ tracksRegLiveness: true body: | bb.0: liveins: $x14, $x16 - ; 87 = e32,mf2 ; CHECK-LABEL: name: copy_fractional_lmul ; CHECK: liveins: $x14, $x16 ; CHECK-NEXT: {{ $}} @@ -283,7 +265,6 @@ tracksRegLiveness: true body: | bb.0: liveins: $x12, $x14, $x16 - ; 80 = e32,m1 ; CHECK-LABEL: name: copy_implicit_def ; CHECK: liveins: $x12, $x14, $x16 ; CHECK-NEXT: {{ $}} @@ -291,14 +272,7 @@ body: | ; CHECK-NEXT: $v8_v9_v10_v11_v12_v13_v14_v15 = PseudoVLSEG8E32_V_M1 undef $v8_v9_v10_v11_v12_v13_v14_v15, killed $x12, $noreg, 5 /* e32 */, 0 /* tu, mu */, implicit $vl, implicit $vtype ; CHECK-NEXT: $x0 = PseudoVSETIVLI 10, 80 /* e32, m1, ta, mu */, implicit-def $vl, implicit-def $vtype ; CHECK-NEXT: $v15 = PseudoVLE32_V_M1 undef $v15, killed $x16, $noreg, 5 /* e32 */, 0 /* tu, mu */, implicit $vl, implicit $vtype, implicit killed $v8_v9_v10_v11_v12_v13_v14_v15, implicit-def $v8_v9_v10_v11_v12_v13_v14_v15 - ; CHECK-NEXT: $v24 = VMV1R_V killed $v8 - ; CHECK-NEXT: $v25 = VMV1R_V killed $v9 - ; CHECK-NEXT: $v26 = VMV1R_V killed $v10 - ; CHECK-NEXT: $v27 = VMV1R_V killed $v11 - ; CHECK-NEXT: $v28 = VMV1R_V killed $v12 - ; CHECK-NEXT: $v29 = VMV1R_V killed $v13 - ; CHECK-NEXT: $v30 = VMV1R_V killed $v14 - ; CHECK-NEXT: $v31 = VMV1R_V killed $v15 + ; CHECK-NEXT: $v24m8 = VMV8R_V killed $v8m8 $x0 = PseudoVSETVLI $x14, 80, implicit-def $vl, implicit-def $vtype $v8_v9_v10_v11_v12_v13_v14_v15 = PseudoVLSEG8E32_V_M1 undef $v8_v9_v10_v11_v12_v13_v14_v15, killed $x12, $noreg, 5, 0, implicit $vl, implicit $vtype $x0 = PseudoVSETIVLI 10, 80, implicit-def $vl, implicit-def $vtype diff --git a/llvm/test/CodeGen/RISCV/rvv/zvlsseg-copy.mir b/llvm/test/CodeGen/RISCV/rvv/zvlsseg-copy.mir index 85bb54471ed3c8..afd8b27a3b90b9 100644 --- a/llvm/test/CodeGen/RISCV/rvv/zvlsseg-copy.mir +++ b/llvm/test/CodeGen/RISCV/rvv/zvlsseg-copy.mir @@ -7,30 +7,24 @@ name: copy_zvlsseg_N2 body: | bb.0: ; CHECK-LABEL: name: copy_zvlsseg_N2 - ; CHECK: $v2 = VMV1R_V $v4 - ; CHECK-NEXT: $v3 = VMV1R_V $v5 + ; CHECK: $v2m2 = VMV2R_V $v4m2 ; CHECK-NEXT: $v3 = VMV1R_V $v4 ; CHECK-NEXT: $v4 = VMV1R_V $v5 - ; CHECK-NEXT: $v6 = VMV1R_V $v5 ; CHECK-NEXT: $v5 = VMV1R_V $v4 - ; CHECK-NEXT: $v6 = VMV1R_V $v4 - ; CHECK-NEXT: $v7 = VMV1R_V $v5 - ; CHECK-NEXT: $v0m2 = VMV2R_V $v4m2 - ; CHECK-NEXT: $v2m2 = VMV2R_V $v6m2 + ; CHECK-NEXT: $v4 = VMV1R_V $v3 + ; CHECK-NEXT: $v6m2 = VMV2R_V $v4m2 + ; CHECK-NEXT: $v0m4 = VMV4R_V $v4m4 ; CHECK-NEXT: $v2m2 = VMV2R_V $v4m2 ; CHECK-NEXT: $v4m2 = VMV2R_V $v6m2 - ; CHECK-NEXT: $v8m2 = VMV2R_V $v6m2 ; CHECK-NEXT: $v6m2 = VMV2R_V $v4m2 - ; CHECK-NEXT: $v8m2 = VMV2R_V $v4m2 - ; CHECK-NEXT: $v10m2 = VMV2R_V $v6m2 - ; CHECK-NEXT: $v0m4 = VMV4R_V $v8m4 - ; CHECK-NEXT: $v4m4 = VMV4R_V $v12m4 + ; CHECK-NEXT: $v4m2 = VMV2R_V $v0m2 + ; CHECK-NEXT: $v8m4 = VMV4R_V $v4m4 + ; CHECK-NEXT: $v0m8 = VMV8R_V $v8m8 ; CHECK-NEXT: $v4m4 = VMV4R_V $v8m4 ; CHECK-NEXT: $v8m4 = VMV4R_V $v12m4 - ; CHECK-NEXT: $v16m4 = VMV4R_V $v12m4 ; CHECK-NEXT: $v12m4 = VMV4R_V $v8m4 - ; CHECK-NEXT: $v16m4 = VMV4R_V $v8m4 - ; CHECK-NEXT: $v20m4 = VMV4R_V $v12m4 + ; CHECK-NEXT: $v8m4 = VMV4R_V $v4m4 + ; CHECK-NEXT: $v16m8 = VMV8R_V $v8m8 $v2_v3 = COPY $v4_v5 $v3_v4 = COPY $v4_v5 $v5_v6 = COPY $v4_v5 @@ -55,29 +49,23 @@ body: | ; CHECK-NEXT: $v3 = VMV1R_V $v6 ; CHECK-NEXT: $v4 = VMV1R_V $v7 ; CHECK-NEXT: $v3 = VMV1R_V $v5 - ; CHECK-NEXT: $v4 = VMV1R_V $v6 - ; CHECK-NEXT: $v5 = VMV1R_V $v7 + ; CHECK-NEXT: $v4m2 = VMV2R_V $v6m2 ; CHECK-NEXT: $v4 = VMV1R_V $v5 ; CHECK-NEXT: $v5 = VMV1R_V $v6 ; CHECK-NEXT: $v6 = VMV1R_V $v7 - ; CHECK-NEXT: $v9 = VMV1R_V $v7 - ; CHECK-NEXT: $v8 = VMV1R_V $v6 ; CHECK-NEXT: $v7 = VMV1R_V $v5 + ; CHECK-NEXT: $v6m2 = VMV2R_V $v4m2 ; CHECK-NEXT: $v9 = VMV1R_V $v5 - ; CHECK-NEXT: $v10 = VMV1R_V $v6 - ; CHECK-NEXT: $v11 = VMV1R_V $v7 + ; CHECK-NEXT: $v10m2 = VMV2R_V $v6m2 ; CHECK-NEXT: $v0m2 = VMV2R_V $v6m2 - ; CHECK-NEXT: $v2m2 = VMV2R_V $v8m2 - ; CHECK-NEXT: $v4m2 = VMV2R_V $v10m2 + ; CHECK-NEXT: $v0m4 = VMV4R_V $v8m4 ; CHECK-NEXT: $v2m2 = VMV2R_V $v6m2 - ; CHECK-NEXT: $v4m2 = VMV2R_V $v8m2 - ; CHECK-NEXT: $v6m2 = VMV2R_V $v10m2 - ; CHECK-NEXT: $v14m2 = VMV2R_V $v10m2 - ; CHECK-NEXT: $v12m2 = VMV2R_V $v8m2 + ; CHECK-NEXT: $v4m4 = VMV4R_V $v8m4 ; CHECK-NEXT: $v10m2 = VMV2R_V $v6m2 + ; CHECK-NEXT: $v8m4 = VMV4R_V $v4m4 ; CHECK-NEXT: $v12m2 = VMV2R_V $v6m2 ; CHECK-NEXT: $v14m2 = VMV2R_V $v8m2 - ; CHECK-NEXT: $v16m2 = VMV2R_V $v10m2 + ; CHECK-NEXT: $v16m2 = VMV2R_V $v8m2 $v2_v3_v4 = COPY $v5_v6_v7 $v3_v4_v5 = COPY $v5_v6_v7 $v4_v5_v6 = COPY $v5_v6_v7 @@ -94,37 +82,30 @@ name: copy_zvlsseg_N4 body: | bb.0: ; CHECK-LABEL: name: copy_zvlsseg_N4 - ; CHECK: $v6 = VMV1R_V $v10 - ; CHECK-NEXT: $v7 = VMV1R_V $v11 - ; CHECK-NEXT: $v8 = VMV1R_V $v12 - ; CHECK-NEXT: $v9 = VMV1R_V $v13 + ; CHECK: $v6m2 = VMV2R_V $v10m2 + ; CHECK-NEXT: $v8m2 = VMV2R_V $v12m2 ; CHECK-NEXT: $v7 = VMV1R_V $v10 ; CHECK-NEXT: $v8 = VMV1R_V $v11 ; CHECK-NEXT: $v9 = VMV1R_V $v12 ; CHECK-NEXT: $v10 = VMV1R_V $v13 - ; CHECK-NEXT: $v16 = VMV1R_V $v13 - ; CHECK-NEXT: $v15 = VMV1R_V $v12 - ; CHECK-NEXT: $v14 = VMV1R_V $v11 ; CHECK-NEXT: $v13 = VMV1R_V $v10 - ; CHECK-NEXT: $v14 = VMV1R_V $v10 - ; CHECK-NEXT: $v15 = VMV1R_V $v11 - ; CHECK-NEXT: $v16 = VMV1R_V $v12 - ; CHECK-NEXT: $v17 = VMV1R_V $v13 + ; CHECK-NEXT: $v12 = VMV1R_V $v9 + ; CHECK-NEXT: $v11 = VMV1R_V $v8 + ; CHECK-NEXT: $v10 = VMV1R_V $v7 + ; CHECK-NEXT: $v14m2 = VMV2R_V $v10m2 + ; CHECK-NEXT: $v16m2 = VMV2R_V $v12m2 ; CHECK-NEXT: $v2m2 = VMV2R_V $v10m2 - ; CHECK-NEXT: $v4m2 = VMV2R_V $v12m2 - ; CHECK-NEXT: $v6m2 = VMV2R_V $v14m2 + ; CHECK-NEXT: $v4m4 = VMV4R_V $v12m4 ; CHECK-NEXT: $v8m2 = VMV2R_V $v16m2 ; CHECK-NEXT: $v4m2 = VMV2R_V $v10m2 ; CHECK-NEXT: $v6m2 = VMV2R_V $v12m2 ; CHECK-NEXT: $v8m2 = VMV2R_V $v14m2 - ; CHECK-NEXT: $v10m2 = VMV2R_V $v16m2 - ; CHECK-NEXT: $v22m2 = VMV2R_V $v16m2 - ; CHECK-NEXT: $v20m2 = VMV2R_V $v14m2 - ; CHECK-NEXT: $v18m2 = VMV2R_V $v12m2 + ; CHECK-NEXT: $v8m2 = VMV2R_V $v16m2 ; CHECK-NEXT: $v16m2 = VMV2R_V $v10m2 + ; CHECK-NEXT: $v12m4 = VMV4R_V $v8m4 + ; CHECK-NEXT: $v8m2 = VMV2R_V $v4m2 ; CHECK-NEXT: $v18m2 = VMV2R_V $v10m2 - ; CHECK-NEXT: $v20m2 = VMV2R_V $v12m2 - ; CHECK-NEXT: $v22m2 = VMV2R_V $v14m2 + ; CHECK-NEXT: $v20m4 = VMV4R_V $v12m4 ; CHECK-NEXT: $v24m2 = VMV2R_V $v16m2 $v6_v7_v8_v9 = COPY $v10_v11_v12_v13 $v7_v8_v9_v10 = COPY $v10_v11_v12_v13 @@ -146,16 +127,12 @@ body: | ; CHECK-NEXT: $v7 = VMV1R_V $v12 ; CHECK-NEXT: $v8 = VMV1R_V $v13 ; CHECK-NEXT: $v9 = VMV1R_V $v14 - ; CHECK-NEXT: $v6 = VMV1R_V $v10 - ; CHECK-NEXT: $v7 = VMV1R_V $v11 - ; CHECK-NEXT: $v8 = VMV1R_V $v12 - ; CHECK-NEXT: $v9 = VMV1R_V $v13 - ; CHECK-NEXT: $v10 = VMV1R_V $v14 - ; CHECK-NEXT: $v18 = VMV1R_V $v14 - ; CHECK-NEXT: $v17 = VMV1R_V $v13 - ; CHECK-NEXT: $v16 = VMV1R_V $v12 - ; CHECK-NEXT: $v15 = VMV1R_V $v11 - ; CHECK-NEXT: $v14 = VMV1R_V $v10 + ; CHECK-NEXT: $v6m2 = VMV2R_V $v10m2 + ; CHECK-NEXT: $v8m2 = VMV2R_V $v12m2 + ; CHECK-NEXT: $v8 = VMV1R_V $v14 + ; CHECK-NEXT: $v14m2 = VMV2R_V $v10m2 + ; CHECK-NEXT: $v12m2 = VMV2R_V $v8m2 + ; CHECK-NEXT: $v8 = VMV1R_V $v4 ; CHECK-NEXT: $v15 = VMV1R_V $v10 ; CHECK-NEXT: $v16 = VMV1R_V $v11 ; CHECK-NEXT: $v17 = VMV1R_V $v12 @@ -171,30 +148,23 @@ name: copy_zvlsseg_N6 body: | bb.0: ; CHECK-LABEL: name: copy_zvlsseg_N6 - ; CHECK: $v4 = VMV1R_V $v10 - ; CHECK-NEXT: $v5 = VMV1R_V $v11 - ; CHECK-NEXT: $v6 = VMV1R_V $v12 - ; CHECK-NEXT: $v7 = VMV1R_V $v13 - ; CHECK-NEXT: $v8 = VMV1R_V $v14 - ; CHECK-NEXT: $v9 = VMV1R_V $v15 + ; CHECK: $v4m2 = VMV2R_V $v10m2 + ; CHECK-NEXT: $v6m2 = VMV2R_V $v12m2 + ; CHECK-NEXT: $v8m2 = VMV2R_V $v14m2 ; CHECK-NEXT: $v5 = VMV1R_V $v10 ; CHECK-NEXT: $v6 = VMV1R_V $v11 ; CHECK-NEXT: $v7 = VMV1R_V $v12 ; CHECK-NEXT: $v8 = VMV1R_V $v13 ; CHECK-NEXT: $v9 = VMV1R_V $v14 ; CHECK-NEXT: $v10 = VMV1R_V $v15 - ; CHECK-NEXT: $v20 = VMV1R_V $v15 - ; CHECK-NEXT: $v19 = VMV1R_V $v14 - ; CHECK-NEXT: $v18 = VMV1R_V $v13 - ; CHECK-NEXT: $v17 = VMV1R_V $v12 - ; CHECK-NEXT: $v16 = VMV1R_V $v11 ; CHECK-NEXT: $v15 = VMV1R_V $v10 - ; CHECK-NEXT: $v16 = VMV1R_V $v10 - ; CHECK-NEXT: $v17 = VMV1R_V $v11 - ; CHECK-NEXT: $v18 = VMV1R_V $v12 - ; CHECK-NEXT: $v19 = VMV1R_V $v13 - ; CHECK-NEXT: $v20 = VMV1R_V $v14 - ; CHECK-NEXT: $v21 = VMV1R_V $v15 + ; CHECK-NEXT: $v14 = VMV1R_V $v9 + ; CHECK-NEXT: $v13 = VMV1R_V $v8 + ; CHECK-NEXT: $v12 = VMV1R_V $v7 + ; CHECK-NEXT: $v11 = VMV1R_V $v6 + ; CHECK-NEXT: $v10 = VMV1R_V $v5 + ; CHECK-NEXT: $v16m2 = VMV2R_V $v10m2 + ; CHECK-NEXT: $v16m4 = VMV4R_V $v12m4 $v4_v5_v6_v7_v8_v9 = COPY $v10_v11_v12_v13_v14_v15 $v5_v6_v7_v8_v9_v10 = COPY $v10_v11_v12_v13_v14_v15 $v15_v16_v17_v18_v19_v20 = COPY $v10_v11_v12_v13_v14_v15 @@ -212,20 +182,13 @@ body: | ; CHECK-NEXT: $v7 = VMV1R_V $v14 ; CHECK-NEXT: $v8 = VMV1R_V $v15 ; CHECK-NEXT: $v9 = VMV1R_V $v16 - ; CHECK-NEXT: $v4 = VMV1R_V $v10 - ; CHECK-NEXT: $v5 = VMV1R_V $v11 - ; CHECK-NEXT: $v6 = VMV1R_V $v12 - ; CHECK-NEXT: $v7 = VMV1R_V $v13 - ; CHECK-NEXT: $v8 = VMV1R_V $v14 - ; CHECK-NEXT: $v9 = VMV1R_V $v15 - ; CHECK-NEXT: $v10 = VMV1R_V $v16 - ; CHECK-NEXT: $v22 = VMV1R_V $v16 - ; CHECK-NEXT: $v21 = VMV1R_V $v15 - ; CHECK-NEXT: $v20 = VMV1R_V $v14 - ; CHECK-NEXT: $v19 = VMV1R_V $v13 - ; CHECK-NEXT: $v18 = VMV1R_V $v12 - ; CHECK-NEXT: $v17 = VMV1R_V $v11 - ; CHECK-NEXT: $v16 = VMV1R_V $v10 + ; CHECK-NEXT: $v4m2 = VMV2R_V $v10m2 + ; CHECK-NEXT: $v6m2 = VMV2R_V $v12m2 + ; CHECK-NEXT: $v8m2 = VMV2R_V $v14m2 + ; CHECK-NEXT: $v8 = VMV1R_V $v16 + ; CHECK-NEXT: $v16m2 = VMV2R_V $v10m2 + ; CHECK-NEXT: $v12m4 = VMV4R_V $v8m4 + ; CHECK-NEXT: $v8 = VMV1R_V $v4 ; CHECK-NEXT: $v17 = VMV1R_V $v10 ; CHECK-NEXT: $v18 = VMV1R_V $v11 ; CHECK-NEXT: $v19 = VMV1R_V $v12 @@ -243,14 +206,9 @@ name: copy_zvlsseg_N8 body: | bb.0: ; CHECK-LABEL: name: copy_zvlsseg_N8 - ; CHECK: $v2 = VMV1R_V $v10 - ; CHECK-NEXT: $v3 = VMV1R_V $v11 - ; CHECK-NEXT: $v4 = VMV1R_V $v12 - ; CHECK-NEXT: $v5 = VMV1R_V $v13 - ; CHECK-NEXT: $v6 = VMV1R_V $v14 - ; CHECK-NEXT: $v7 = VMV1R_V $v15 - ; CHECK-NEXT: $v8 = VMV1R_V $v16 - ; CHECK-NEXT: $v9 = VMV1R_V $v17 + ; CHECK: $v2m2 = VMV2R_V $v10m2 + ; CHECK-NEXT: $v4m4 = VMV4R_V $v12m4 + ; CHECK-NEXT: $v8m2 = VMV2R_V $v16m2 ; CHECK-NEXT: $v3 = VMV1R_V $v10 ; CHECK-NEXT: $v4 = VMV1R_V $v11 ; CHECK-NEXT: $v5 = VMV1R_V $v12 @@ -259,22 +217,17 @@ body: | ; CHECK-NEXT: $v8 = VMV1R_V $v15 ; CHECK-NEXT: $v9 = VMV1R_V $v16 ; CHECK-NEXT: $v10 = VMV1R_V $v17 - ; CHECK-NEXT: $v24 = VMV1R_V $v17 - ; CHECK-NEXT: $v23 = VMV1R_V $v16 - ; CHECK-NEXT: $v22 = VMV1R_V $v15 - ; CHECK-NEXT: $v21 = VMV1R_V $v14 - ; CHECK-NEXT: $v20 = VMV1R_V $v13 - ; CHECK-NEXT: $v19 = VMV1R_V $v12 - ; CHECK-NEXT: $v18 = VMV1R_V $v11 ; CHECK-NEXT: $v17 = VMV1R_V $v10 - ; CHECK-NEXT: $v18 = VMV1R_V $v10 - ; CHECK-NEXT: $v19 = VMV1R_V $v11 - ; CHECK-NEXT: $v20 = VMV1R_V $v12 - ; CHECK-NEXT: $v21 = VMV1R_V $v13 - ; CHECK-NEXT: $v22 = VMV1R_V $v14 - ; CHECK-NEXT: $v23 = VMV1R_V $v15 - ; CHECK-NEXT: $v24 = VMV1R_V $v16 - ; CHECK-NEXT: $v25 = VMV1R_V $v17 + ; CHECK-NEXT: $v16 = VMV1R_V $v9 + ; CHECK-NEXT: $v15 = VMV1R_V $v8 + ; CHECK-NEXT: $v14 = VMV1R_V $v7 + ; CHECK-NEXT: $v13 = VMV1R_V $v6 + ; CHECK-NEXT: $v12 = VMV1R_V $v5 + ; CHECK-NEXT: $v11 = VMV1R_V $v4 + ; CHECK-NEXT: $v10 = VMV1R_V $v3 + ; CHECK-NEXT: $v18m2 = VMV2R_V $v10m2 + ; CHECK-NEXT: $v20m4 = VMV4R_V $v12m4 + ; CHECK-NEXT: $v24m2 = VMV2R_V $v16m2 $v2_v3_v4_v5_v6_v7_v8_v9 = COPY $v10_v11_v12_v13_v14_v15_v16_v17 $v3_v4_v5_v6_v7_v8_v9_v10 = COPY $v10_v11_v12_v13_v14_v15_v16_v17 $v17_v18_v19_v20_v21_v22_v23_v24 = COPY $v10_v11_v12_v13_v14_v15_v16_v17 >From de09e8b8d26c835e551879c058cda3f8130ae053 Mon Sep 17 00:00:00 2001 From: Wang Pengcheng <wangpengcheng...@bytedance.com> Date: Thu, 14 Mar 2024 16:45:26 +0800 Subject: [PATCH 2/2] Address comments and fix wrong LMUL passed to isConvertibleToVMV_V_V Created using spr 1.3.4 --- llvm/lib/Target/RISCV/RISCVInstrInfo.cpp | 104 +++++++++++------------ llvm/test/CodeGen/RISCV/rvv/vmv-copy.mir | 12 ++- 2 files changed, 58 insertions(+), 58 deletions(-) diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp b/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp index 9fe5666d6a81f4..3dd0d0c456dd5f 100644 --- a/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp +++ b/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp @@ -313,73 +313,65 @@ void RISCVInstrInfo::copyPhysRegVector(MachineBasicBlock &MBB, forwardCopyWillClobberTuple(DstEncoding, SrcEncoding, NumRegs); unsigned I = 0; + auto GetCopyInfo = [&](MCRegister SrcReg, MCRegister DstReg) + -> std::tuple<RISCVII::VLMUL, const TargetRegisterClass &, unsigned, + unsigned, unsigned> { + unsigned SrcEncoding = TRI->getEncodingValue(SrcReg); + unsigned DstEncoding = TRI->getEncodingValue(DstReg); + if (!(SrcEncoding & 0b111) && !(DstEncoding & 0b111) && I + 8 <= NumRegs) + return {RISCVII::LMUL_8, RISCV::VRM8RegClass, RISCV::VMV8R_V, + RISCV::PseudoVMV_V_V_M8, RISCV::PseudoVMV_V_I_M8}; + if (!(SrcEncoding & 0b11) && !(DstEncoding & 0b11) && I + 4 <= NumRegs) + return {RISCVII::LMUL_4, RISCV::VRM4RegClass, RISCV::VMV4R_V, + RISCV::PseudoVMV_V_V_M4, RISCV::PseudoVMV_V_I_M4}; + if (!(SrcEncoding & 0b1) && !(DstEncoding & 0b1) && I + 2 <= NumRegs) + return {RISCVII::LMUL_2, RISCV::VRM2RegClass, RISCV::VMV2R_V, + RISCV::PseudoVMV_V_V_M2, RISCV::PseudoVMV_V_I_M2}; + return {RISCVII::LMUL_1, RISCV::VRRegClass, RISCV::VMV1R_V, + RISCV::PseudoVMV_V_V_M1, RISCV::PseudoVMV_V_I_M1}; + }; while (I != NumRegs) { - auto GetCopyInfo = - [&](unsigned SrcReg, - unsigned DstReg) -> std::tuple<int, const TargetRegisterClass &, - unsigned, unsigned, unsigned> { - unsigned SrcEncoding = TRI->getEncodingValue(SrcReg); - unsigned DstEncoding = TRI->getEncodingValue(DstReg); - if (!(SrcEncoding & 0b111) && !(DstEncoding & 0b111) && I + 8 <= NumRegs) - return {8, RISCV::VRM8RegClass, RISCV::VMV8R_V, RISCV::PseudoVMV_V_V_M8, - RISCV::PseudoVMV_V_I_M8}; - if (!(SrcEncoding & 0b11) && !(DstEncoding & 0b11) && I + 4 <= NumRegs) - return {4, RISCV::VRM4RegClass, RISCV::VMV4R_V, RISCV::PseudoVMV_V_V_M4, - RISCV::PseudoVMV_V_I_M4}; - if (!(SrcEncoding & 0b1) && !(DstEncoding & 0b1) && I + 2 <= NumRegs) - return {2, RISCV::VRM2RegClass, RISCV::VMV2R_V, RISCV::PseudoVMV_V_V_M2, - RISCV::PseudoVMV_V_I_M2}; - return {1, RISCV::VRRegClass, RISCV::VMV1R_V, RISCV::PseudoVMV_V_V_M1, - RISCV::PseudoVMV_V_I_M1}; - }; - - auto [NumCopied, RegClass, Opc, VVOpc, VIOpc] = GetCopyInfo(SrcReg, DstReg); + auto [LMul, RegClass, Opc, VVOpc, VIOpc] = GetCopyInfo(SrcReg, DstReg); + unsigned NumCopied = 1 << LMul; MachineBasicBlock::const_iterator DefMBBI; if (isConvertibleToVMV_V_V(STI, MBB, MBBI, DefMBBI, LMul)) { Opc = VVOpc; - - if (DefMBBI->getOpcode() == VIOpc) { + if (DefMBBI->getOpcode() == VIOpc) Opc = VIOpc; - } } - for (MCPhysReg Reg : RegClass.getRegisters()) { - if (TRI->getEncodingValue(Reg) == TRI->getEncodingValue(SrcReg)) { - SrcReg = Reg; - break; - } - } + ArrayRef<MCPhysReg> Regs = RegClass.getRegisters(); + const auto *FoundSrcReg = llvm::find_if(Regs, [&](MCPhysReg Reg) { + return TRI->getEncodingValue(Reg) == TRI->getEncodingValue(SrcReg); + }); + assert(FoundSrcReg != Regs.end()); + SrcReg = *FoundSrcReg; - for (MCPhysReg Reg : RegClass.getRegisters()) { - if (TRI->getEncodingValue(Reg) == TRI->getEncodingValue(DstReg)) { - DstReg = Reg; - break; - } + const auto *FoundDstReg = llvm::find_if(Regs, [&](MCPhysReg Reg) { + return TRI->getEncodingValue(Reg) == TRI->getEncodingValue(DstReg); + }); + assert(FoundDstReg != Regs.end()); + DstReg = *FoundDstReg; + + auto MIB = BuildMI(MBB, MBBI, DL, get(Opc), DstReg); + bool UseVMV_V_I = RISCV::getRVVMCOpcode(Opc) == RISCV::VMV_V_I; + bool UseVMV = UseVMV_V_I || RISCV::getRVVMCOpcode(Opc) == RISCV::VMV_V_V; + if (UseVMV) + MIB.addReg(DstReg, RegState::Undef); + if (UseVMV_V_I) + MIB = MIB.add(DefMBBI->getOperand(2)); + else + MIB = MIB.addReg(SrcReg, getKillRegState(KillSrc)); + if (UseVMV) { + const MCInstrDesc &Desc = DefMBBI->getDesc(); + MIB.add(DefMBBI->getOperand(RISCVII::getVLOpNum(Desc))); // AVL + MIB.add(DefMBBI->getOperand(RISCVII::getSEWOpNum(Desc))); // SEW + MIB.addImm(0); // tu, mu + MIB.addReg(RISCV::VL, RegState::Implicit); + MIB.addReg(RISCV::VTYPE, RegState::Implicit); } - auto EmitCopy = [&](MCRegister SrcReg, MCRegister DstReg, unsigned Opcode) { - auto MIB = BuildMI(MBB, MBBI, DL, get(Opcode), DstReg); - bool UseVMV_V_I = RISCV::getRVVMCOpcode(Opcode) == RISCV::VMV_V_I; - bool UseVMV = - UseVMV_V_I || RISCV::getRVVMCOpcode(Opcode) == RISCV::VMV_V_V; - if (UseVMV) - MIB.addReg(DstReg, RegState::Undef); - if (UseVMV_V_I) - MIB = MIB.add(DefMBBI->getOperand(2)); - else - MIB = MIB.addReg(SrcReg, getKillRegState(KillSrc)); - if (UseVMV) { - const MCInstrDesc &Desc = DefMBBI->getDesc(); - MIB.add(DefMBBI->getOperand(RISCVII::getVLOpNum(Desc))); // AVL - MIB.add(DefMBBI->getOperand(RISCVII::getSEWOpNum(Desc))); // SEW - MIB.addImm(0); // tu, mu - MIB.addReg(RISCV::VL, RegState::Implicit); - MIB.addReg(RISCV::VTYPE, RegState::Implicit); - } - }; - - EmitCopy(SrcReg, DstReg, Opc); SrcReg = SrcReg.id() + (ReversedCopy ? -NumCopied : NumCopied); DstReg = DstReg.id() + (ReversedCopy ? -NumCopied : NumCopied); I += NumCopied; diff --git a/llvm/test/CodeGen/RISCV/rvv/vmv-copy.mir b/llvm/test/CodeGen/RISCV/rvv/vmv-copy.mir index dd569129db4d72..6449dd0bfc67c1 100644 --- a/llvm/test/CodeGen/RISCV/rvv/vmv-copy.mir +++ b/llvm/test/CodeGen/RISCV/rvv/vmv-copy.mir @@ -238,7 +238,8 @@ body: | ; CHECK-NEXT: {{ $}} ; CHECK-NEXT: $x15 = PseudoVSETVLI $x14, 80 /* e32, m1, ta, mu */, implicit-def $vl, implicit-def $vtype ; CHECK-NEXT: $v8_v9 = PseudoVLSEG2E32_V_M1 undef $v8_v9, killed $x16, $noreg, 5 /* e32 */, 0 /* tu, mu */, implicit $vl, implicit $vtype - ; CHECK-NEXT: $v10m2 = PseudoVMV_V_V_M2 undef $v10m2, $v8m2, $noreg, 5 /* e32 */, 0 /* tu, mu */, implicit $vl, implicit $vtype + ; CHECK-NEXT: $v10 = PseudoVMV_V_V_M1 undef $v10, $v8, $noreg, 5 /* e32 */, 0 /* tu, mu */, implicit $vl, implicit $vtype + ; CHECK-NEXT: $v11 = PseudoVMV_V_V_M1 undef $v11, $v9, $noreg, 5 /* e32 */, 0 /* tu, mu */, implicit $vl, implicit $vtype $x15 = PseudoVSETVLI $x14, 80, implicit-def $vl, implicit-def $vtype $v8_v9 = PseudoVLSEG2E32_V_M1 undef $v8_v9, killed $x16, $noreg, 5, 0, implicit $vl, implicit $vtype $v10_v11 = COPY $v8_v9 @@ -272,7 +273,14 @@ body: | ; CHECK-NEXT: $v8_v9_v10_v11_v12_v13_v14_v15 = PseudoVLSEG8E32_V_M1 undef $v8_v9_v10_v11_v12_v13_v14_v15, killed $x12, $noreg, 5 /* e32 */, 0 /* tu, mu */, implicit $vl, implicit $vtype ; CHECK-NEXT: $x0 = PseudoVSETIVLI 10, 80 /* e32, m1, ta, mu */, implicit-def $vl, implicit-def $vtype ; CHECK-NEXT: $v15 = PseudoVLE32_V_M1 undef $v15, killed $x16, $noreg, 5 /* e32 */, 0 /* tu, mu */, implicit $vl, implicit $vtype, implicit killed $v8_v9_v10_v11_v12_v13_v14_v15, implicit-def $v8_v9_v10_v11_v12_v13_v14_v15 - ; CHECK-NEXT: $v24m8 = VMV8R_V killed $v8m8 + ; CHECK-NEXT: $v24 = VMV1R_V killed $v8 + ; CHECK-NEXT: $v25 = VMV1R_V killed $v9 + ; CHECK-NEXT: $v26 = VMV1R_V killed $v10 + ; CHECK-NEXT: $v27 = VMV1R_V killed $v11 + ; CHECK-NEXT: $v28 = VMV1R_V killed $v12 + ; CHECK-NEXT: $v29 = VMV1R_V killed $v13 + ; CHECK-NEXT: $v30 = VMV1R_V killed $v14 + ; CHECK-NEXT: $v31 = VMV1R_V killed $v15 $x0 = PseudoVSETVLI $x14, 80, implicit-def $vl, implicit-def $vtype $v8_v9_v10_v11_v12_v13_v14_v15 = PseudoVLSEG8E32_V_M1 undef $v8_v9_v10_v11_v12_v13_v14_v15, killed $x12, $noreg, 5, 0, implicit $vl, implicit $vtype $x0 = PseudoVSETIVLI 10, 80, implicit-def $vl, implicit-def $vtype _______________________________________________ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits