[llvm] r191286 - [mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics)

Aaron Ballman aaron at aaronballman.com
Mon Oct 7 10:45:49 PDT 2013


Another thing to note: I build Debug builds, but Chapuni's bot are
release builds.  If this is a miscompile from MSVC that needs to be
worked around, it's more likely to trigger in release builds.  Is that
what you are testing against?

~Aaron

On Mon, Oct 7, 2013 at 12:46 PM, Aaron Ballman <aaron at aaronballman.com> wrote:
> These tests are failing for the build bot, not for me.  So they've got
> all the default options for cmake.  As for the cl version, it's both
> MSVC 10 and MSVC 11 that are failing, but perhaps Takumi can give
> further information.
>
> In my tests, I'm on Win7 x64, building a 32-bit clang with MSVC 11
> without ninja and not seeing the failure.
>
> ~Aaron
>
> On Mon, Oct 7, 2013 at 12:43 PM, Daniel Sanders
> <Daniel.Sanders at imgtec.com> wrote:
>> I haven't run the full 'ninja check' yet, but those three tests pass in my MSVC build (using cmake and ninja) of r191286.
>>
>> Which compiler versions are you using? I'm using cl.exe version 16.00.30319.01. Also, have you set any non-default options in CMakeCache.txt?
>>
>>> -----Original Message-----
>>> From: Daniel Sanders
>>> Sent: 07 October 2013 14:27
>>> To: 'Aaron Ballman'
>>> Cc: llvm-commits
>>> Subject: RE: [llvm] r191286 - [mips][msa] Added support for matching
>>> comparisons from normal IR (i.e. not intrinsics)
>>>
>>> I rely on the buildbot at lab.llvm.org to inform me that my changes have
>>> broken builds. This failure has come from outside that buildbot so I
>>> wouldn't have known something had broken. Given that the lab.llvm.org
>>> builds were all ok, I don't think reverting this change (and all my subsequent
>>> changes) is a reasonable course of action.
>>>
>>> I'm setting up a MSVC host as we speak. Hopefully I should have some idea
>>> what's going on soon. I notice that Matt Arsenault has submitted a patch
>>> that might be related to the solution. His patch works around an MSVC
>>> miscompilation that is likely to affect
>>> CodeGen__Mips__msa__compare_float.ll.
>>>
>>> > -----Original Message-----
>>> > From: aaron.ballman at gmail.com [mailto:aaron.ballman at gmail.com] On
>>> > Behalf Of Aaron Ballman
>>> > Sent: 07 October 2013 13:52
>>> > To: Daniel Sanders
>>> > Cc: llvm-commits
>>> > Subject: Re: [llvm] r191286 - [mips][msa] Added support for matching
>>> > comparisons from normal IR (i.e. not intrinsics)
>>> >
>>> > On Mon, Oct 7, 2013 at 5:11 AM, Daniel Sanders
>>> > <Daniel.Sanders at imgtec.com> wrote:
>>> > > Hi,
>>> > >
>>> > > Thanks for the report. I'll look into this but it might take a while to fix it
>>> > since I'll have to set up a suitable development environment to reproduce
>>> > the problem.
>>> >
>>> > That's a bit worrisome since the build has been broken for two weeks
>>> > now.  Can you please revert your changes (and r191290) until you are
>>> > able to reproduce and resolve the problem?
>>> >
>>> > ~Aaron
>>> >
>>> > >
>>> > >> -----Original Message-----
>>> > >> From: aaron.ballman at gmail.com [mailto:aaron.ballman at gmail.com]
>>> On
>>> > >> Behalf Of Aaron Ballman
>>> > >> Sent: 06 October 2013 02:02
>>> > >> To: Daniel Sanders
>>> > >> Cc: llvm-commits
>>> > >> Subject: Re: [llvm] r191286 - [mips][msa] Added support for matching
>>> > >> comparisons from normal IR (i.e. not intrinsics)
>>> > >>
>>> > >> The MSVC builders have been broken since this commit. Specifically:
>>> > >>
>>> > >> LLVM :: CodeGen__Mips__msa__3rf_int_float.ll
>>> > >> LLVM :: CodeGen__Mips__msa__compare_float.ll
>>> > >>
>>> > >> http://bb.pgr.jp/builders/ninja-clang-i686-msc17-R/builds/5000
>>> > >>
>>> > >> r191290 caused LLVM :: CodeGen__Mips__msa__compare.ll to fail as
>>> well.
>>> > >>
>>> > >> This is failing on MSVC 10 and MSVC 11.
>>> > >>
>>> > >> ~Aaron
>>> > >>
>>> > >> On Tue, Sep 24, 2013 at 6:46 AM, Daniel Sanders
>>> > >> <daniel.sanders at imgtec.com> wrote:
>>> > >> > Author: dsanders
>>> > >> > Date: Tue Sep 24 05:46:19 2013
>>> > >> > New Revision: 191286
>>> > >> >
>>> > >> > URL: http://llvm.org/viewvc/llvm-project?rev=191286&view=rev
>>> > >> > Log:
>>> > >> > [mips][msa] Added support for matching comparisons from normal IR
>>> > (i.e.
>>> > >> not intrinsics)
>>> > >> >
>>> > >> > MIPS SelectionDAG changes:
>>> > >> > * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons
>>> > that
>>> > >> produce a bitmask.
>>> > >> >
>>> > >> > Added:
>>> > >> >     llvm/trunk/test/CodeGen/Mips/msa/compare.ll
>>> > >> >     llvm/trunk/test/CodeGen/Mips/msa/compare_float.ll
>>> > >> > Modified:
>>> > >> >     llvm/trunk/lib/Target/Mips/MipsISelLowering.cpp
>>> > >> >     llvm/trunk/lib/Target/Mips/MipsISelLowering.h
>>> > >> >     llvm/trunk/lib/Target/Mips/MipsMSAInstrInfo.td
>>> > >> >     llvm/trunk/lib/Target/Mips/MipsSEISelLowering.cpp
>>> > >> >
>>> > >> > Modified: llvm/trunk/lib/Target/Mips/MipsISelLowering.cpp
>>> > >> > URL: http://llvm.org/viewvc/llvm-
>>> > >>
>>> >
>>> project/llvm/trunk/lib/Target/Mips/MipsISelLowering.cpp?rev=191286&r1=
>>> > 1
>>> > >> 91285&r2=191286&view=diff
>>> > >> >
>>> > >> ==========================================================
>>> > >> ====================
>>> > >> > --- llvm/trunk/lib/Target/Mips/MipsISelLowering.cpp (original)
>>> > >> > +++ llvm/trunk/lib/Target/Mips/MipsISelLowering.cpp Tue Sep 24
>>> > 05:46:19
>>> > >> 2013
>>> > >> > @@ -212,6 +212,11 @@ const char *MipsTargetLowering::getTarge
>>> > >> >    case MipsISD::VANY_ZERO:         return "MipsISD::VANY_ZERO";
>>> > >> >    case MipsISD::VALL_NONZERO:      return "MipsISD::VALL_NONZERO";
>>> > >> >    case MipsISD::VANY_NONZERO:      return
>>> "MipsISD::VANY_NONZERO";
>>> > >> > +  case MipsISD::VCEQ:              return "MipsISD::VCEQ";
>>> > >> > +  case MipsISD::VCLE_S:            return "MipsISD::VCLE_S";
>>> > >> > +  case MipsISD::VCLE_U:            return "MipsISD::VCLE_U";
>>> > >> > +  case MipsISD::VCLT_S:            return "MipsISD::VCLT_S";
>>> > >> > +  case MipsISD::VCLT_U:            return "MipsISD::VCLT_U";
>>> > >> >    case MipsISD::VSPLAT:            return "MipsISD::VSPLAT";
>>> > >> >    case MipsISD::VSPLATD:           return "MipsISD::VSPLATD";
>>> > >> >    case MipsISD::VEXTRACT_SEXT_ELT: return
>>> > >> "MipsISD::VEXTRACT_SEXT_ELT";
>>> > >> >
>>> > >> > Modified: llvm/trunk/lib/Target/Mips/MipsISelLowering.h
>>> > >> > URL: http://llvm.org/viewvc/llvm-
>>> > >>
>>> >
>>> project/llvm/trunk/lib/Target/Mips/MipsISelLowering.h?rev=191286&r1=19
>>> > 1
>>> > >> 285&r2=191286&view=diff
>>> > >> >
>>> > >> ==========================================================
>>> > >> ====================
>>> > >> > --- llvm/trunk/lib/Target/Mips/MipsISelLowering.h (original)
>>> > >> > +++ llvm/trunk/lib/Target/Mips/MipsISelLowering.h Tue Sep 24
>>> 05:46:19
>>> > >> 2013
>>> > >> > @@ -153,11 +153,19 @@ namespace llvm {
>>> > >> >        SELECT_CC_DSP,
>>> > >> >
>>> > >> >        // Vector comparisons.
>>> > >> > +      // These take a vector and return a boolean.
>>> > >> >        VALL_ZERO,
>>> > >> >        VANY_ZERO,
>>> > >> >        VALL_NONZERO,
>>> > >> >        VANY_NONZERO,
>>> > >> >
>>> > >> > +      // These take a vector and return a vector bitmask.
>>> > >> > +      VCEQ,
>>> > >> > +      VCLE_S,
>>> > >> > +      VCLE_U,
>>> > >> > +      VCLT_S,
>>> > >> > +      VCLT_U,
>>> > >> > +
>>> > >> >        // Special case of BUILD_VECTOR where all elements are the same.
>>> > >> >        VSPLAT,
>>> > >> >        // Special case of VSPLAT where the result is v2i64, the operand is
>>> > >> >
>>> > >> > Modified: llvm/trunk/lib/Target/Mips/MipsMSAInstrInfo.td
>>> > >> > URL: http://llvm.org/viewvc/llvm-
>>> > >>
>>> >
>>> project/llvm/trunk/lib/Target/Mips/MipsMSAInstrInfo.td?rev=191286&r1=1
>>> > >> 91285&r2=191286&view=diff
>>> > >> >
>>> > >> ==========================================================
>>> > >> ====================
>>> > >> > --- llvm/trunk/lib/Target/Mips/MipsMSAInstrInfo.td (original)
>>> > >> > +++ llvm/trunk/lib/Target/Mips/MipsMSAInstrInfo.td Tue Sep 24
>>> > 05:46:19
>>> > >> 2013
>>> > >> > @@ -13,6 +13,14 @@
>>> > >> >
>>> > >> >  def SDT_MipsSplat : SDTypeProfile<1, 1, [SDTCisVec<0>,
>>> SDTCisInt<1>]>;
>>> > >> >  def SDT_MipsVecCond : SDTypeProfile<1, 1, [SDTCisInt<0>,
>>> > >> SDTCisVec<1>]>;
>>> > >> > +def SDT_VSetCC : SDTypeProfile<1, 3, [SDTCisInt<0>,
>>> > >> > +                                      SDTCisInt<1>,
>>> > >> > +                                      SDTCisSameAs<1, 2>,
>>> > >> > +                                      SDTCisVT<3, OtherVT>]>;
>>> > >> > +def SDT_VFSetCC : SDTypeProfile<1, 3, [SDTCisInt<0>,
>>> > >> > +                                       SDTCisFP<1>,
>>> > >> > +                                       SDTCisSameAs<1, 2>,
>>> > >> > +                                       SDTCisVT<3, OtherVT>]>;
>>> > >> >
>>> > >> >  def MipsVAllNonZero : SDNode<"MipsISD::VALL_NONZERO",
>>> > >> SDT_MipsVecCond>;
>>> > >> >  def MipsVAnyNonZero : SDNode<"MipsISD::VANY_NONZERO",
>>> > >> SDT_MipsVecCond>;
>>> > >> > @@ -23,6 +31,9 @@ def MipsVSplatD : SDNode<"MipsISD::VSPLA
>>> > >> >  def MipsVNOR : SDNode<"MipsISD::VNOR", SDTIntBinOp,
>>> > >> >                        [SDNPCommutative, SDNPAssociative]>;
>>> > >> >
>>> > >> > +def vsetcc : SDNode<"ISD::SETCC", SDT_VSetCC>;
>>> > >> > +def vfsetcc : SDNode<"ISD::SETCC", SDT_VFSetCC>;
>>> > >> > +
>>> > >> >  def MipsVExtractSExt : SDNode<"MipsISD::VEXTRACT_SEXT_ELT",
>>> > >> >      SDTypeProfile<1, 3, [SDTCisPtrTy<2>]>, []>;
>>> > >> >  def MipsVExtractZExt : SDNode<"MipsISD::VEXTRACT_ZEXT_ELT",
>>> > >> > @@ -50,6 +61,68 @@ def vinsert_v8i16 : PatFrag<(ops node:$v
>>> > >> >  def vinsert_v4i32 : PatFrag<(ops node:$vec, node:$val, node:$idx),
>>> > >> >      (v4i32 (vector_insert node:$vec, node:$val, node:$idx))>;
>>> > >> >
>>> > >> > +class vfsetcc_type<ValueType ResTy, ValueType OpTy, CondCode
>>> CC> :
>>> > >> > +  PatFrag<(ops node:$lhs, node:$rhs),
>>> > >> > +          (ResTy (vfsetcc (OpTy node:$lhs), (OpTy node:$rhs), CC))>;
>>> > >> > +
>>> > >> > +// ISD::SETFALSE cannot occur
>>> > >> > +def vfsetoeq_v4f32 : vfsetcc_type<v4i32, v4f32, SETOEQ>;
>>> > >> > +def vfsetoeq_v2f64 : vfsetcc_type<v2i64, v2f64, SETOEQ>;
>>> > >> > +def vfsetoge_v4f32 : vfsetcc_type<v4i32, v4f32, SETOGE>;
>>> > >> > +def vfsetoge_v2f64 : vfsetcc_type<v2i64, v2f64, SETOGE>;
>>> > >> > +def vfsetogt_v4f32 : vfsetcc_type<v4i32, v4f32, SETOGT>;
>>> > >> > +def vfsetogt_v2f64 : vfsetcc_type<v2i64, v2f64, SETOGT>;
>>> > >> > +def vfsetole_v4f32 : vfsetcc_type<v4i32, v4f32, SETOLE>;
>>> > >> > +def vfsetole_v2f64 : vfsetcc_type<v2i64, v2f64, SETOLE>;
>>> > >> > +def vfsetolt_v4f32 : vfsetcc_type<v4i32, v4f32, SETOLT>;
>>> > >> > +def vfsetolt_v2f64 : vfsetcc_type<v2i64, v2f64, SETOLT>;
>>> > >> > +def vfsetone_v4f32 : vfsetcc_type<v4i32, v4f32, SETONE>;
>>> > >> > +def vfsetone_v2f64 : vfsetcc_type<v2i64, v2f64, SETONE>;
>>> > >> > +def vfsetord_v4f32 : vfsetcc_type<v4i32, v4f32, SETO>;
>>> > >> > +def vfsetord_v2f64 : vfsetcc_type<v2i64, v2f64, SETO>;
>>> > >> > +def vfsetun_v4f32  : vfsetcc_type<v4i32, v4f32, SETUO>;
>>> > >> > +def vfsetun_v2f64  : vfsetcc_type<v2i64, v2f64, SETUO>;
>>> > >> > +def vfsetueq_v4f32 : vfsetcc_type<v4i32, v4f32, SETUEQ>;
>>> > >> > +def vfsetueq_v2f64 : vfsetcc_type<v2i64, v2f64, SETUEQ>;
>>> > >> > +def vfsetuge_v4f32 : vfsetcc_type<v4i32, v4f32, SETUGE>;
>>> > >> > +def vfsetuge_v2f64 : vfsetcc_type<v2i64, v2f64, SETUGE>;
>>> > >> > +def vfsetugt_v4f32 : vfsetcc_type<v4i32, v4f32, SETUGT>;
>>> > >> > +def vfsetugt_v2f64 : vfsetcc_type<v2i64, v2f64, SETUGT>;
>>> > >> > +def vfsetule_v4f32 : vfsetcc_type<v4i32, v4f32, SETULE>;
>>> > >> > +def vfsetule_v2f64 : vfsetcc_type<v2i64, v2f64, SETULE>;
>>> > >> > +def vfsetult_v4f32 : vfsetcc_type<v4i32, v4f32, SETULT>;
>>> > >> > +def vfsetult_v2f64 : vfsetcc_type<v2i64, v2f64, SETULT>;
>>> > >> > +def vfsetune_v4f32 : vfsetcc_type<v4i32, v4f32, SETUNE>;
>>> > >> > +def vfsetune_v2f64 : vfsetcc_type<v2i64, v2f64, SETUNE>;
>>> > >> > +// ISD::SETTRUE cannot occur
>>> > >> > +// ISD::SETFALSE2 cannot occur
>>> > >> > +// ISD::SETTRUE2 cannot occur
>>> > >> > +
>>> > >> > +class vsetcc_type<ValueType ResTy, CondCode CC> :
>>> > >> > +  PatFrag<(ops node:$lhs, node:$rhs),
>>> > >> > +          (ResTy (vsetcc node:$lhs, node:$rhs, CC))>;
>>> > >> > +
>>> > >> > +def vseteq_v16i8  : vsetcc_type<v16i8, SETEQ>;
>>> > >> > +def vseteq_v8i16  : vsetcc_type<v8i16, SETEQ>;
>>> > >> > +def vseteq_v4i32  : vsetcc_type<v4i32, SETEQ>;
>>> > >> > +def vseteq_v2i64  : vsetcc_type<v2i64, SETEQ>;
>>> > >> > +def vsetle_v16i8  : vsetcc_type<v16i8, SETLE>;
>>> > >> > +def vsetle_v8i16  : vsetcc_type<v8i16, SETLE>;
>>> > >> > +def vsetle_v4i32  : vsetcc_type<v4i32, SETLE>;
>>> > >> > +def vsetle_v2i64  : vsetcc_type<v2i64, SETLE>;
>>> > >> > +def vsetlt_v16i8  : vsetcc_type<v16i8, SETLT>;
>>> > >> > +def vsetlt_v8i16  : vsetcc_type<v8i16, SETLT>;
>>> > >> > +def vsetlt_v4i32  : vsetcc_type<v4i32, SETLT>;
>>> > >> > +def vsetlt_v2i64  : vsetcc_type<v2i64, SETLT>;
>>> > >> > +def vsetule_v16i8 : vsetcc_type<v16i8, SETULE>;
>>> > >> > +def vsetule_v8i16 : vsetcc_type<v8i16, SETULE>;
>>> > >> > +def vsetule_v4i32 : vsetcc_type<v4i32, SETULE>;
>>> > >> > +def vsetule_v2i64 : vsetcc_type<v2i64, SETULE>;
>>> > >> > +def vsetult_v16i8 : vsetcc_type<v16i8, SETULT>;
>>> > >> > +def vsetult_v8i16 : vsetcc_type<v8i16, SETULT>;
>>> > >> > +def vsetult_v4i32 : vsetcc_type<v4i32, SETULT>;
>>> > >> > +def vsetult_v2i64 : vsetcc_type<v2i64, SETULT>;
>>> > >> > +
>>> > >> >  def vsplati8  : PatFrag<(ops node:$in), (v16i8 (MipsVSplat (i32
>>> > node:$in)))>;
>>> > >> >  def vsplati16 : PatFrag<(ops node:$in), (v8i16 (MipsVSplat (i32
>>> > >> node:$in)))>;
>>> > >> >  def vsplati32 : PatFrag<(ops node:$in), (v4i32 (MipsVSplat (i32
>>> > >> node:$in)))>;
>>> > >> > @@ -909,12 +982,14 @@ class MSA_I5_X_DESC_BASE<string instr_as
>>> > >> >  }
>>> > >> >
>>> > >> >  class MSA_SI5_DESC_BASE<string instr_asm, SDPatternOperator
>>> > OpNode,
>>> > >> > -                       RegisterClass RCWD, RegisterClass RCWS = RCWD,
>>> > >> > +                       SDPatternOperator SplatNode, RegisterClass RCWD,
>>> > >> > +                       RegisterClass RCWS = RCWD,
>>> > >> >                         InstrItinClass itin = NoItinerary> {
>>> > >> >    dag OutOperandList = (outs RCWD:$wd);
>>> > >> >    dag InOperandList = (ins RCWS:$ws, simm5:$s5);
>>> > >> >    string AsmString = !strconcat(instr_asm, "\t$wd, $ws, $s5");
>>> > >> > -  list<dag> Pattern = [(set RCWD:$wd, (OpNode RCWS:$ws,
>>> > >> immSExt5:$s5))];
>>> > >> > +  list<dag> Pattern = [(set RCWD:$wd, (OpNode RCWS:$ws,
>>> > >> > +                                              (SplatNode immSExt5:$s5)))];
>>> > >> >    InstrItinClass Itinerary = itin;
>>> > >> >  }
>>> > >> >
>>> > >> > @@ -1228,19 +1303,23 @@ class BZ_D_DESC :
>>> > >> MSA_CBRANCH_DESC_BASE<
>>> > >> >
>>> > >> >  class BZ_V_DESC : MSA_CBRANCH_DESC_BASE<"bz.v", MSA128B>;
>>> > >> >
>>> > >> > -class CEQ_B_DESC : MSA_3R_DESC_BASE<"ceq.b", int_mips_ceq_b,
>>> > >> MSA128B>,
>>> > >> > +class CEQ_B_DESC : MSA_3R_DESC_BASE<"ceq.b", vseteq_v16i8,
>>> > >> MSA128B>,
>>> > >> >                     IsCommutable;
>>> > >> > -class CEQ_H_DESC : MSA_3R_DESC_BASE<"ceq.h", int_mips_ceq_h,
>>> > >> MSA128H>,
>>> > >> > +class CEQ_H_DESC : MSA_3R_DESC_BASE<"ceq.h", vseteq_v8i16,
>>> > >> MSA128H>,
>>> > >> >                     IsCommutable;
>>> > >> > -class CEQ_W_DESC : MSA_3R_DESC_BASE<"ceq.w", int_mips_ceq_w,
>>> > >> MSA128W>,
>>> > >> > +class CEQ_W_DESC : MSA_3R_DESC_BASE<"ceq.w", vseteq_v4i32,
>>> > >> MSA128W>,
>>> > >> >                     IsCommutable;
>>> > >> > -class CEQ_D_DESC : MSA_3R_DESC_BASE<"ceq.d", int_mips_ceq_d,
>>> > >> MSA128D>,
>>> > >> > +class CEQ_D_DESC : MSA_3R_DESC_BASE<"ceq.d", vseteq_v2i64,
>>> > >> MSA128D>,
>>> > >> >                     IsCommutable;
>>> > >> >
>>> > >> > -class CEQI_B_DESC : MSA_SI5_DESC_BASE<"ceqi.b", int_mips_ceqi_b,
>>> > >> MSA128B>;
>>> > >> > -class CEQI_H_DESC : MSA_SI5_DESC_BASE<"ceqi.h", int_mips_ceqi_h,
>>> > >> MSA128H>;
>>> > >> > -class CEQI_W_DESC : MSA_SI5_DESC_BASE<"ceqi.w",
>>> int_mips_ceqi_w,
>>> > >> MSA128W>;
>>> > >> > -class CEQI_D_DESC : MSA_SI5_DESC_BASE<"ceqi.d", int_mips_ceqi_d,
>>> > >> MSA128D>;
>>> > >> > +class CEQI_B_DESC : MSA_SI5_DESC_BASE<"ceqi.b", vseteq_v16i8,
>>> > >> vsplati8,
>>> > >> > +                                      MSA128B>;
>>> > >> > +class CEQI_H_DESC : MSA_SI5_DESC_BASE<"ceqi.h", vseteq_v8i16,
>>> > >> vsplati16,
>>> > >> > +                                      MSA128H>;
>>> > >> > +class CEQI_W_DESC : MSA_SI5_DESC_BASE<"ceqi.w", vseteq_v4i32,
>>> > >> vsplati32,
>>> > >> > +                                      MSA128W>;
>>> > >> > +class CEQI_D_DESC : MSA_SI5_DESC_BASE<"ceqi.d", vseteq_v2i64,
>>> > >> vsplati64,
>>> > >> > +                                      MSA128D>;
>>> > >> >
>>> > >> >  class CFCMSA_DESC {
>>> > >> >    dag OutOperandList = (outs GPR32:$rd);
>>> > >> > @@ -1250,61 +1329,61 @@ class CFCMSA_DESC {
>>> > >> >    bit hasSideEffects = 1;
>>> > >> >  }
>>> > >> >
>>> > >> > -class CLE_S_B_DESC : MSA_3R_DESC_BASE<"cle_s.b",
>>> int_mips_cle_s_b,
>>> > >> MSA128B>;
>>> > >> > -class CLE_S_H_DESC : MSA_3R_DESC_BASE<"cle_s.h",
>>> int_mips_cle_s_h,
>>> > >> MSA128H>;
>>> > >> > -class CLE_S_W_DESC : MSA_3R_DESC_BASE<"cle_s.w",
>>> > >> int_mips_cle_s_w, MSA128W>;
>>> > >> > -class CLE_S_D_DESC : MSA_3R_DESC_BASE<"cle_s.d",
>>> int_mips_cle_s_d,
>>> > >> MSA128D>;
>>> > >> > -
>>> > >> > -class CLE_U_B_DESC : MSA_3R_DESC_BASE<"cle_u.b",
>>> > int_mips_cle_u_b,
>>> > >> MSA128B>;
>>> > >> > -class CLE_U_H_DESC : MSA_3R_DESC_BASE<"cle_u.h",
>>> > int_mips_cle_u_h,
>>> > >> MSA128H>;
>>> > >> > -class CLE_U_W_DESC : MSA_3R_DESC_BASE<"cle_u.w",
>>> > >> int_mips_cle_u_w, MSA128W>;
>>> > >> > -class CLE_U_D_DESC : MSA_3R_DESC_BASE<"cle_u.d",
>>> > int_mips_cle_u_d,
>>> > >> MSA128D>;
>>> > >> > +class CLE_S_B_DESC : MSA_3R_DESC_BASE<"cle_s.b", vsetle_v16i8,
>>> > >> MSA128B>;
>>> > >> > +class CLE_S_H_DESC : MSA_3R_DESC_BASE<"cle_s.h", vsetle_v8i16,
>>> > >> MSA128H>;
>>> > >> > +class CLE_S_W_DESC : MSA_3R_DESC_BASE<"cle_s.w", vsetle_v4i32,
>>> > >> MSA128W>;
>>> > >> > +class CLE_S_D_DESC : MSA_3R_DESC_BASE<"cle_s.d", vsetle_v2i64,
>>> > >> MSA128D>;
>>> > >> > +
>>> > >> > +class CLE_U_B_DESC : MSA_3R_DESC_BASE<"cle_u.b", vsetule_v16i8,
>>> > >> MSA128B>;
>>> > >> > +class CLE_U_H_DESC : MSA_3R_DESC_BASE<"cle_u.h", vsetule_v8i16,
>>> > >> MSA128H>;
>>> > >> > +class CLE_U_W_DESC : MSA_3R_DESC_BASE<"cle_u.w", vsetule_v4i32,
>>> > >> MSA128W>;
>>> > >> > +class CLE_U_D_DESC : MSA_3R_DESC_BASE<"cle_u.d", vsetule_v2i64,
>>> > >> MSA128D>;
>>> > >> >
>>> > >> > -class CLEI_S_B_DESC : MSA_SI5_DESC_BASE<"clei_s.b",
>>> > >> int_mips_clei_s_b,
>>> > >> > +class CLEI_S_B_DESC : MSA_SI5_DESC_BASE<"clei_s.b", vsetle_v16i8,
>>> > >> vsplati8,
>>> > >> >                                          MSA128B>;
>>> > >> > -class CLEI_S_H_DESC : MSA_SI5_DESC_BASE<"clei_s.h",
>>> > >> int_mips_clei_s_h,
>>> > >> > +class CLEI_S_H_DESC : MSA_SI5_DESC_BASE<"clei_s.h", vsetle_v8i16,
>>> > >> vsplati16,
>>> > >> >                                          MSA128H>;
>>> > >> > -class CLEI_S_W_DESC : MSA_SI5_DESC_BASE<"clei_s.w",
>>> > >> int_mips_clei_s_w,
>>> > >> > +class CLEI_S_W_DESC : MSA_SI5_DESC_BASE<"clei_s.w", vsetle_v4i32,
>>> > >> vsplati32,
>>> > >> >                                          MSA128W>;
>>> > >> > -class CLEI_S_D_DESC : MSA_SI5_DESC_BASE<"clei_s.d",
>>> > >> int_mips_clei_s_d,
>>> > >> > +class CLEI_S_D_DESC : MSA_SI5_DESC_BASE<"clei_s.d", vsetle_v2i64,
>>> > >> vsplati64,
>>> > >> >                                          MSA128D>;
>>> > >> >
>>> > >> > -class CLEI_U_B_DESC : MSA_SI5_DESC_BASE<"clei_u.b",
>>> > >> int_mips_clei_u_b,
>>> > >> > -                                        MSA128B>;
>>> > >> > -class CLEI_U_H_DESC : MSA_SI5_DESC_BASE<"clei_u.h",
>>> > >> int_mips_clei_u_h,
>>> > >> > -                                        MSA128H>;
>>> > >> > -class CLEI_U_W_DESC : MSA_SI5_DESC_BASE<"clei_u.w",
>>> > >> int_mips_clei_u_w,
>>> > >> > -                                        MSA128W>;
>>> > >> > -class CLEI_U_D_DESC : MSA_SI5_DESC_BASE<"clei_u.d",
>>> > >> int_mips_clei_u_d,
>>> > >> > -                                        MSA128D>;
>>> > >> > +class CLEI_U_B_DESC : MSA_I5_DESC_BASE<"clei_u.b", vsetule_v16i8,
>>> > >> vsplati8,
>>> > >> > +                                       MSA128B>;
>>> > >> > +class CLEI_U_H_DESC : MSA_I5_DESC_BASE<"clei_u.h", vsetule_v8i16,
>>> > >> vsplati16,
>>> > >> > +                                       MSA128H>;
>>> > >> > +class CLEI_U_W_DESC : MSA_I5_DESC_BASE<"clei_u.w",
>>> vsetule_v4i32,
>>> > >> vsplati32,
>>> > >> > +                                       MSA128W>;
>>> > >> > +class CLEI_U_D_DESC : MSA_I5_DESC_BASE<"clei_u.d", vsetule_v2i64,
>>> > >> vsplati64,
>>> > >> > +                                       MSA128D>;
>>> > >> > +
>>> > >> > +class CLT_S_B_DESC : MSA_3R_DESC_BASE<"clt_s.b", vsetlt_v16i8,
>>> > >> MSA128B>;
>>> > >> > +class CLT_S_H_DESC : MSA_3R_DESC_BASE<"clt_s.h", vsetlt_v8i16,
>>> > >> MSA128H>;
>>> > >> > +class CLT_S_W_DESC : MSA_3R_DESC_BASE<"clt_s.w", vsetlt_v4i32,
>>> > >> MSA128W>;
>>> > >> > +class CLT_S_D_DESC : MSA_3R_DESC_BASE<"clt_s.d", vsetlt_v2i64,
>>> > >> MSA128D>;
>>> > >> > +
>>> > >> > +class CLT_U_B_DESC : MSA_3R_DESC_BASE<"clt_u.b", vsetult_v16i8,
>>> > >> MSA128B>;
>>> > >> > +class CLT_U_H_DESC : MSA_3R_DESC_BASE<"clt_u.h", vsetult_v8i16,
>>> > >> MSA128H>;
>>> > >> > +class CLT_U_W_DESC : MSA_3R_DESC_BASE<"clt_u.w", vsetult_v4i32,
>>> > >> MSA128W>;
>>> > >> > +class CLT_U_D_DESC : MSA_3R_DESC_BASE<"clt_u.d", vsetult_v2i64,
>>> > >> MSA128D>;
>>> > >> >
>>> > >> > -class CLT_S_B_DESC : MSA_3R_DESC_BASE<"clt_s.b",
>>> int_mips_clt_s_b,
>>> > >> MSA128B>;
>>> > >> > -class CLT_S_H_DESC : MSA_3R_DESC_BASE<"clt_s.h",
>>> int_mips_clt_s_h,
>>> > >> MSA128H>;
>>> > >> > -class CLT_S_W_DESC : MSA_3R_DESC_BASE<"clt_s.w",
>>> > int_mips_clt_s_w,
>>> > >> MSA128W>;
>>> > >> > -class CLT_S_D_DESC : MSA_3R_DESC_BASE<"clt_s.d",
>>> int_mips_clt_s_d,
>>> > >> MSA128D>;
>>> > >> > -
>>> > >> > -class CLT_U_B_DESC : MSA_3R_DESC_BASE<"clt_u.b",
>>> int_mips_clt_u_b,
>>> > >> MSA128B>;
>>> > >> > -class CLT_U_H_DESC : MSA_3R_DESC_BASE<"clt_u.h",
>>> int_mips_clt_u_h,
>>> > >> MSA128H>;
>>> > >> > -class CLT_U_W_DESC : MSA_3R_DESC_BASE<"clt_u.w",
>>> > >> int_mips_clt_u_w, MSA128W>;
>>> > >> > -class CLT_U_D_DESC : MSA_3R_DESC_BASE<"clt_u.d",
>>> int_mips_clt_u_d,
>>> > >> MSA128D>;
>>> > >> > -
>>> > >> > -class CLTI_S_B_DESC : MSA_SI5_DESC_BASE<"clti_s.b",
>>> > int_mips_clti_s_b,
>>> > >> > +class CLTI_S_B_DESC : MSA_SI5_DESC_BASE<"clti_s.b", vsetlt_v16i8,
>>> > >> vsplati8,
>>> > >> >                                          MSA128B>;
>>> > >> > -class CLTI_S_H_DESC : MSA_SI5_DESC_BASE<"clti_s.h",
>>> > int_mips_clti_s_h,
>>> > >> > +class CLTI_S_H_DESC : MSA_SI5_DESC_BASE<"clti_s.h", vsetlt_v8i16,
>>> > >> vsplati16,
>>> > >> >                                          MSA128H>;
>>> > >> > -class CLTI_S_W_DESC : MSA_SI5_DESC_BASE<"clti_s.w",
>>> > >> int_mips_clti_s_w,
>>> > >> > +class CLTI_S_W_DESC : MSA_SI5_DESC_BASE<"clti_s.w", vsetlt_v4i32,
>>> > >> vsplati32,
>>> > >> >                                          MSA128W>;
>>> > >> > -class CLTI_S_D_DESC : MSA_SI5_DESC_BASE<"clti_s.d",
>>> > int_mips_clti_s_d,
>>> > >> > +class CLTI_S_D_DESC : MSA_SI5_DESC_BASE<"clti_s.d", vsetlt_v2i64,
>>> > >> vsplati64,
>>> > >> >                                          MSA128D>;
>>> > >> >
>>> > >> > -class CLTI_U_B_DESC : MSA_SI5_DESC_BASE<"clti_u.b",
>>> > >> int_mips_clti_u_b,
>>> > >> > -                                        MSA128B>;
>>> > >> > -class CLTI_U_H_DESC : MSA_SI5_DESC_BASE<"clti_u.h",
>>> > >> int_mips_clti_u_h,
>>> > >> > -                                        MSA128H>;
>>> > >> > -class CLTI_U_W_DESC : MSA_SI5_DESC_BASE<"clti_u.w",
>>> > >> int_mips_clti_u_w,
>>> > >> > -                                        MSA128W>;
>>> > >> > -class CLTI_U_D_DESC : MSA_SI5_DESC_BASE<"clti_u.d",
>>> > >> int_mips_clti_u_d,
>>> > >> > -                                        MSA128D>;
>>> > >> > +class CLTI_U_B_DESC : MSA_I5_DESC_BASE<"clti_u.b", vsetult_v16i8,
>>> > >> vsplati8,
>>> > >> > +                                       MSA128B>;
>>> > >> > +class CLTI_U_H_DESC : MSA_I5_DESC_BASE<"clti_u.h", vsetult_v8i16,
>>> > >> vsplati16,
>>> > >> > +                                       MSA128H>;
>>> > >> > +class CLTI_U_W_DESC : MSA_I5_DESC_BASE<"clti_u.w", vsetult_v4i32,
>>> > >> vsplati32,
>>> > >> > +                                       MSA128W>;
>>> > >> > +class CLTI_U_D_DESC : MSA_I5_DESC_BASE<"clti_u.d", vsetult_v2i64,
>>> > >> vsplati64,
>>> > >> > +                                       MSA128D>;
>>> > >> >
>>> > >> >  class COPY_S_B_DESC : MSA_COPY_DESC_BASE<"copy_s.b",
>>> > >> vextract_sext_i8,  v16i8,
>>> > >> >                                           GPR32, MSA128B>;
>>> > >> > @@ -1394,9 +1473,9 @@ class FCAF_W_DESC :
>>> MSA_3RF_DESC_BASE<"f
>>> > >> >  class FCAF_D_DESC : MSA_3RF_DESC_BASE<"fcaf.d", int_mips_fcaf_d,
>>> > >> MSA128D>,
>>> > >> >                      IsCommutable;
>>> > >> >
>>> > >> > -class FCEQ_W_DESC : MSA_3RF_DESC_BASE<"fceq.w",
>>> > int_mips_fceq_w,
>>> > >> MSA128W>,
>>> > >> > +class FCEQ_W_DESC : MSA_3RF_DESC_BASE<"fceq.w",
>>> vfsetoeq_v4f32,
>>> > >> MSA128W>,
>>> > >> >                      IsCommutable;
>>> > >> > -class FCEQ_D_DESC : MSA_3RF_DESC_BASE<"fceq.d",
>>> int_mips_fceq_d,
>>> > >> MSA128D>,
>>> > >> > +class FCEQ_D_DESC : MSA_3RF_DESC_BASE<"fceq.d", vfsetoeq_v2f64,
>>> > >> MSA128D>,
>>> > >> >                      IsCommutable;
>>> > >> >
>>> > >> >  class FCLASS_W_DESC : MSA_2RF_DESC_BASE<"fclass.w",
>>> > >> int_mips_fclass_w,
>>> > >> > @@ -1404,45 +1483,45 @@ class FCLASS_W_DESC :
>>> > MSA_2RF_DESC_BASE<
>>> > >> >  class FCLASS_D_DESC : MSA_2RF_DESC_BASE<"fclass.d",
>>> > >> int_mips_fclass_d,
>>> > >> >                                          MSA128D>;
>>> > >> >
>>> > >> > -class FCLE_W_DESC : MSA_3RF_DESC_BASE<"fcle.w",
>>> int_mips_fcle_w,
>>> > >> MSA128W>;
>>> > >> > -class FCLE_D_DESC : MSA_3RF_DESC_BASE<"fcle.d", int_mips_fcle_d,
>>> > >> MSA128D>;
>>> > >> > +class FCLE_W_DESC : MSA_3RF_DESC_BASE<"fcle.w", vfsetole_v4f32,
>>> > >> MSA128W>;
>>> > >> > +class FCLE_D_DESC : MSA_3RF_DESC_BASE<"fcle.d", vfsetole_v2f64,
>>> > >> MSA128D>;
>>> > >> >
>>> > >> > -class FCLT_W_DESC : MSA_3RF_DESC_BASE<"fclt.w", int_mips_fclt_w,
>>> > >> MSA128W>;
>>> > >> > -class FCLT_D_DESC : MSA_3RF_DESC_BASE<"fclt.d", int_mips_fclt_d,
>>> > >> MSA128D>;
>>> > >> > +class FCLT_W_DESC : MSA_3RF_DESC_BASE<"fclt.w", vfsetolt_v4f32,
>>> > >> MSA128W>;
>>> > >> > +class FCLT_D_DESC : MSA_3RF_DESC_BASE<"fclt.d", vfsetolt_v2f64,
>>> > >> MSA128D>;
>>> > >> >
>>> > >> > -class FCNE_W_DESC : MSA_3RF_DESC_BASE<"fcne.w",
>>> > int_mips_fcne_w,
>>> > >> MSA128W>,
>>> > >> > +class FCNE_W_DESC : MSA_3RF_DESC_BASE<"fcne.w",
>>> vfsetone_v4f32,
>>> > >> MSA128W>,
>>> > >> >                      IsCommutable;
>>> > >> > -class FCNE_D_DESC : MSA_3RF_DESC_BASE<"fcne.d",
>>> int_mips_fcne_d,
>>> > >> MSA128D>,
>>> > >> > +class FCNE_D_DESC : MSA_3RF_DESC_BASE<"fcne.d", vfsetone_v2f64,
>>> > >> MSA128D>,
>>> > >> >                      IsCommutable;
>>> > >> >
>>> > >> > -class FCOR_W_DESC : MSA_3RF_DESC_BASE<"fcor.w",
>>> int_mips_fcor_w,
>>> > >> MSA128W>,
>>> > >> > +class FCOR_W_DESC : MSA_3RF_DESC_BASE<"fcor.w",
>>> vfsetord_v4f32,
>>> > >> MSA128W>,
>>> > >> >                      IsCommutable;
>>> > >> > -class FCOR_D_DESC : MSA_3RF_DESC_BASE<"fcor.d", int_mips_fcor_d,
>>> > >> MSA128D>,
>>> > >> > +class FCOR_D_DESC : MSA_3RF_DESC_BASE<"fcor.d", vfsetord_v2f64,
>>> > >> MSA128D>,
>>> > >> >                      IsCommutable;
>>> > >> >
>>> > >> > -class FCUEQ_W_DESC : MSA_3RF_DESC_BASE<"fcueq.w",
>>> > >> int_mips_fcueq_w, MSA128W>,
>>> > >> > +class FCUEQ_W_DESC : MSA_3RF_DESC_BASE<"fcueq.w",
>>> > >> vfsetueq_v4f32, MSA128W>,
>>> > >> >                       IsCommutable;
>>> > >> > -class FCUEQ_D_DESC : MSA_3RF_DESC_BASE<"fcueq.d",
>>> > >> int_mips_fcueq_d, MSA128D>,
>>> > >> > +class FCUEQ_D_DESC : MSA_3RF_DESC_BASE<"fcueq.d",
>>> > vfsetueq_v2f64,
>>> > >> MSA128D>,
>>> > >> >                       IsCommutable;
>>> > >> >
>>> > >> > -class FCULE_W_DESC : MSA_3RF_DESC_BASE<"fcule.w",
>>> > >> int_mips_fcule_w, MSA128W>,
>>> > >> > +class FCULE_W_DESC : MSA_3RF_DESC_BASE<"fcule.w",
>>> > vfsetule_v4f32,
>>> > >> MSA128W>,
>>> > >> >                       IsCommutable;
>>> > >> > -class FCULE_D_DESC : MSA_3RF_DESC_BASE<"fcule.d",
>>> > int_mips_fcule_d,
>>> > >> MSA128D>,
>>> > >> > +class FCULE_D_DESC : MSA_3RF_DESC_BASE<"fcule.d",
>>> vfsetule_v2f64,
>>> > >> MSA128D>,
>>> > >> >                       IsCommutable;
>>> > >> >
>>> > >> > -class FCULT_W_DESC : MSA_3RF_DESC_BASE<"fcult.w",
>>> > >> int_mips_fcult_w, MSA128W>,
>>> > >> > +class FCULT_W_DESC : MSA_3RF_DESC_BASE<"fcult.w",
>>> vfsetult_v4f32,
>>> > >> MSA128W>,
>>> > >> >                       IsCommutable;
>>> > >> > -class FCULT_D_DESC : MSA_3RF_DESC_BASE<"fcult.d",
>>> > int_mips_fcult_d,
>>> > >> MSA128D>,
>>> > >> > +class FCULT_D_DESC : MSA_3RF_DESC_BASE<"fcult.d", vfsetult_v2f64,
>>> > >> MSA128D>,
>>> > >> >                       IsCommutable;
>>> > >> >
>>> > >> > -class FCUN_W_DESC : MSA_3RF_DESC_BASE<"fcun.w",
>>> > int_mips_fcun_w,
>>> > >> MSA128W>,
>>> > >> > +class FCUN_W_DESC : MSA_3RF_DESC_BASE<"fcun.w", vfsetun_v4f32,
>>> > >> MSA128W>,
>>> > >> >                      IsCommutable;
>>> > >> > -class FCUN_D_DESC : MSA_3RF_DESC_BASE<"fcun.d",
>>> int_mips_fcun_d,
>>> > >> MSA128D>,
>>> > >> > +class FCUN_D_DESC : MSA_3RF_DESC_BASE<"fcun.d", vfsetun_v2f64,
>>> > >> MSA128D>,
>>> > >> >                      IsCommutable;
>>> > >> >
>>> > >> > -class FCUNE_W_DESC : MSA_3RF_DESC_BASE<"fcune.w",
>>> > >> int_mips_fcune_w, MSA128W>,
>>> > >> > +class FCUNE_W_DESC : MSA_3RF_DESC_BASE<"fcune.w",
>>> > >> vfsetune_v4f32, MSA128W>,
>>> > >> >                       IsCommutable;
>>> > >> > -class FCUNE_D_DESC : MSA_3RF_DESC_BASE<"fcune.d",
>>> > >> int_mips_fcune_d, MSA128D>,
>>> > >> > +class FCUNE_D_DESC : MSA_3RF_DESC_BASE<"fcune.d",
>>> > vfsetune_v2f64,
>>> > >> MSA128D>,
>>> > >> >                       IsCommutable;
>>> > >> >
>>> > >> >  class FDIV_W_DESC : MSA_3RF_DESC_BASE<"fdiv.w", fdiv,
>>> MSA128W>;
>>> > >> >
>>> > >> > Modified: llvm/trunk/lib/Target/Mips/MipsSEISelLowering.cpp
>>> > >> > URL: http://llvm.org/viewvc/llvm-
>>> > >>
>>> >
>>> project/llvm/trunk/lib/Target/Mips/MipsSEISelLowering.cpp?rev=191286&r
>>> > 1
>>> > >> =191285&r2=191286&view=diff
>>> > >> >
>>> > >> ==========================================================
>>> > >> ====================
>>> > >> > --- llvm/trunk/lib/Target/Mips/MipsSEISelLowering.cpp (original)
>>> > >> > +++ llvm/trunk/lib/Target/Mips/MipsSEISelLowering.cpp Tue Sep 24
>>> > >> 05:46:19 2013
>>> > >> > @@ -180,6 +180,13 @@ addMSAIntType(MVT::SimpleValueType Ty, c
>>> > >> >    setOperationAction(ISD::SUB, Ty, Legal);
>>> > >> >    setOperationAction(ISD::UDIV, Ty, Legal);
>>> > >> >    setOperationAction(ISD::XOR, Ty, Legal);
>>> > >> > +
>>> > >> > +  setOperationAction(ISD::SETCC, Ty, Legal);
>>> > >> > +  setCondCodeAction(ISD::SETNE, Ty, Expand);
>>> > >> > +  setCondCodeAction(ISD::SETGE, Ty, Expand);
>>> > >> > +  setCondCodeAction(ISD::SETGT, Ty, Expand);
>>> > >> > +  setCondCodeAction(ISD::SETUGE, Ty, Expand);
>>> > >> > +  setCondCodeAction(ISD::SETUGT, Ty, Expand);
>>> > >> >  }
>>> > >> >
>>> > >> >  // Enable MSA support for the given floating-point type and Register
>>> > class.
>>> > >> > @@ -204,6 +211,14 @@ addMSAFloatType(MVT::SimpleValueType Ty,
>>> > >> >      setOperationAction(ISD::FRINT, Ty, Legal);
>>> > >> >      setOperationAction(ISD::FSQRT, Ty, Legal);
>>> > >> >      setOperationAction(ISD::FSUB,  Ty, Legal);
>>> > >> > +
>>> > >> > +    setOperationAction(ISD::SETCC, Ty, Legal);
>>> > >> > +    setCondCodeAction(ISD::SETOGE, Ty, Expand);
>>> > >> > +    setCondCodeAction(ISD::SETOGT, Ty, Expand);
>>> > >> > +    setCondCodeAction(ISD::SETUGE, Ty, Expand);
>>> > >> > +    setCondCodeAction(ISD::SETUGT, Ty, Expand);
>>> > >> > +    setCondCodeAction(ISD::SETGE,  Ty, Expand);
>>> > >> > +    setCondCodeAction(ISD::SETGT,  Ty, Expand);
>>> > >> >    }
>>> > >> >  }
>>> > >> >
>>> > >> > @@ -1109,6 +1124,66 @@ SDValue MipsSETargetLowering::lowerINTRI
>>> > >> >      return lowerMSABranchIntr(Op, DAG, MipsISD::VALL_ZERO);
>>> > >> >    case Intrinsic::mips_bz_v:
>>> > >> >      return lowerMSABranchIntr(Op, DAG, MipsISD::VANY_ZERO);
>>> > >> > +  case Intrinsic::mips_ceq_b:
>>> > >> > +  case Intrinsic::mips_ceq_h:
>>> > >> > +  case Intrinsic::mips_ceq_w:
>>> > >> > +  case Intrinsic::mips_ceq_d:
>>> > >> > +    return DAG.getSetCC(SDLoc(Op), Op->getValueType(0), Op-
>>> > >> >getOperand(1),
>>> > >> > +                        Op->getOperand(2), ISD::SETEQ);
>>> > >> > +  case Intrinsic::mips_ceqi_b:
>>> > >> > +  case Intrinsic::mips_ceqi_h:
>>> > >> > +  case Intrinsic::mips_ceqi_w:
>>> > >> > +  case Intrinsic::mips_ceqi_d:
>>> > >> > +    return DAG.getSetCC(SDLoc(Op), Op->getValueType(0), Op-
>>> > >> >getOperand(1),
>>> > >> > +                        lowerMSASplatImm(Op, 2, DAG), ISD::SETEQ);
>>> > >> > +  case Intrinsic::mips_cle_s_b:
>>> > >> > +  case Intrinsic::mips_cle_s_h:
>>> > >> > +  case Intrinsic::mips_cle_s_w:
>>> > >> > +  case Intrinsic::mips_cle_s_d:
>>> > >> > +    return DAG.getSetCC(SDLoc(Op), Op->getValueType(0), Op-
>>> > >> >getOperand(1),
>>> > >> > +                        Op->getOperand(2), ISD::SETLE);
>>> > >> > +  case Intrinsic::mips_clei_s_b:
>>> > >> > +  case Intrinsic::mips_clei_s_h:
>>> > >> > +  case Intrinsic::mips_clei_s_w:
>>> > >> > +  case Intrinsic::mips_clei_s_d:
>>> > >> > +    return DAG.getSetCC(SDLoc(Op), Op->getValueType(0), Op-
>>> > >> >getOperand(1),
>>> > >> > +                        lowerMSASplatImm(Op, 2, DAG), ISD::SETLE);
>>> > >> > +  case Intrinsic::mips_cle_u_b:
>>> > >> > +  case Intrinsic::mips_cle_u_h:
>>> > >> > +  case Intrinsic::mips_cle_u_w:
>>> > >> > +  case Intrinsic::mips_cle_u_d:
>>> > >> > +    return DAG.getSetCC(SDLoc(Op), Op->getValueType(0), Op-
>>> > >> >getOperand(1),
>>> > >> > +                        Op->getOperand(2), ISD::SETULE);
>>> > >> > +  case Intrinsic::mips_clei_u_b:
>>> > >> > +  case Intrinsic::mips_clei_u_h:
>>> > >> > +  case Intrinsic::mips_clei_u_w:
>>> > >> > +  case Intrinsic::mips_clei_u_d:
>>> > >> > +    return DAG.getSetCC(SDLoc(Op), Op->getValueType(0), Op-
>>> > >> >getOperand(1),
>>> > >> > +                        lowerMSASplatImm(Op, 2, DAG), ISD::SETULE);
>>> > >> > +  case Intrinsic::mips_clt_s_b:
>>> > >> > +  case Intrinsic::mips_clt_s_h:
>>> > >> > +  case Intrinsic::mips_clt_s_w:
>>> > >> > +  case Intrinsic::mips_clt_s_d:
>>> > >> > +    return DAG.getSetCC(SDLoc(Op), Op->getValueType(0), Op-
>>> > >> >getOperand(1),
>>> > >> > +                        Op->getOperand(2), ISD::SETLT);
>>> > >> > +  case Intrinsic::mips_clti_s_b:
>>> > >> > +  case Intrinsic::mips_clti_s_h:
>>> > >> > +  case Intrinsic::mips_clti_s_w:
>>> > >> > +  case Intrinsic::mips_clti_s_d:
>>> > >> > +    return DAG.getSetCC(SDLoc(Op), Op->getValueType(0), Op-
>>> > >> >getOperand(1),
>>> > >> > +                        lowerMSASplatImm(Op, 2, DAG), ISD::SETLT);
>>> > >> > +  case Intrinsic::mips_clt_u_b:
>>> > >> > +  case Intrinsic::mips_clt_u_h:
>>> > >> > +  case Intrinsic::mips_clt_u_w:
>>> > >> > +  case Intrinsic::mips_clt_u_d:
>>> > >> > +    return DAG.getSetCC(SDLoc(Op), Op->getValueType(0), Op-
>>> > >> >getOperand(1),
>>> > >> > +                        Op->getOperand(2), ISD::SETULT);
>>> > >> > +  case Intrinsic::mips_clti_u_b:
>>> > >> > +  case Intrinsic::mips_clti_u_h:
>>> > >> > +  case Intrinsic::mips_clti_u_w:
>>> > >> > +  case Intrinsic::mips_clti_u_d:
>>> > >> > +    return DAG.getSetCC(SDLoc(Op), Op->getValueType(0), Op-
>>> > >> >getOperand(1),
>>> > >> > +                        lowerMSASplatImm(Op, 2, DAG), ISD::SETULT);
>>> > >> >    case Intrinsic::mips_copy_s_b:
>>> > >> >    case Intrinsic::mips_copy_s_h:
>>> > >> >    case Intrinsic::mips_copy_s_w:
>>> > >> > @@ -1130,6 +1205,47 @@ SDValue MipsSETargetLowering::lowerINTRI
>>> > >> >    case Intrinsic::mips_fadd_w:
>>> > >> >    case Intrinsic::mips_fadd_d:
>>> > >> >      return lowerMSABinaryIntr(Op, DAG, ISD::FADD);
>>> > >> > +  // Don't lower mips_fcaf_[wd] since LLVM folds SETFALSE condcodes
>>> > >> away
>>> > >> > +  case Intrinsic::mips_fceq_w:
>>> > >> > +  case Intrinsic::mips_fceq_d:
>>> > >> > +    return DAG.getSetCC(SDLoc(Op), Op->getValueType(0), Op-
>>> > >> >getOperand(1),
>>> > >> > +                        Op->getOperand(2), ISD::SETOEQ);
>>> > >> > +  case Intrinsic::mips_fcle_w:
>>> > >> > +  case Intrinsic::mips_fcle_d:
>>> > >> > +    return DAG.getSetCC(SDLoc(Op), Op->getValueType(0), Op-
>>> > >> >getOperand(1),
>>> > >> > +                        Op->getOperand(2), ISD::SETOLE);
>>> > >> > +  case Intrinsic::mips_fclt_w:
>>> > >> > +  case Intrinsic::mips_fclt_d:
>>> > >> > +    return DAG.getSetCC(SDLoc(Op), Op->getValueType(0), Op-
>>> > >> >getOperand(1),
>>> > >> > +                        Op->getOperand(2), ISD::SETOLT);
>>> > >> > +  case Intrinsic::mips_fcne_w:
>>> > >> > +  case Intrinsic::mips_fcne_d:
>>> > >> > +    return DAG.getSetCC(SDLoc(Op), Op->getValueType(0), Op-
>>> > >> >getOperand(1),
>>> > >> > +                        Op->getOperand(2), ISD::SETONE);
>>> > >> > +  case Intrinsic::mips_fcor_w:
>>> > >> > +  case Intrinsic::mips_fcor_d:
>>> > >> > +    return DAG.getSetCC(SDLoc(Op), Op->getValueType(0), Op-
>>> > >> >getOperand(1),
>>> > >> > +                        Op->getOperand(2), ISD::SETO);
>>> > >> > +  case Intrinsic::mips_fcueq_w:
>>> > >> > +  case Intrinsic::mips_fcueq_d:
>>> > >> > +    return DAG.getSetCC(SDLoc(Op), Op->getValueType(0), Op-
>>> > >> >getOperand(1),
>>> > >> > +                        Op->getOperand(2), ISD::SETUEQ);
>>> > >> > +  case Intrinsic::mips_fcule_w:
>>> > >> > +  case Intrinsic::mips_fcule_d:
>>> > >> > +    return DAG.getSetCC(SDLoc(Op), Op->getValueType(0), Op-
>>> > >> >getOperand(1),
>>> > >> > +                        Op->getOperand(2), ISD::SETULE);
>>> > >> > +  case Intrinsic::mips_fcult_w:
>>> > >> > +  case Intrinsic::mips_fcult_d:
>>> > >> > +    return DAG.getSetCC(SDLoc(Op), Op->getValueType(0), Op-
>>> > >> >getOperand(1),
>>> > >> > +                        Op->getOperand(2), ISD::SETULT);
>>> > >> > +  case Intrinsic::mips_fcun_w:
>>> > >> > +  case Intrinsic::mips_fcun_d:
>>> > >> > +    return DAG.getSetCC(SDLoc(Op), Op->getValueType(0), Op-
>>> > >> >getOperand(1),
>>> > >> > +                        Op->getOperand(2), ISD::SETUO);
>>> > >> > +  case Intrinsic::mips_fcune_w:
>>> > >> > +  case Intrinsic::mips_fcune_d:
>>> > >> > +    return DAG.getSetCC(SDLoc(Op), Op->getValueType(0), Op-
>>> > >> >getOperand(1),
>>> > >> > +                        Op->getOperand(2), ISD::SETUNE);
>>> > >> >    case Intrinsic::mips_fdiv_w:
>>> > >> >    case Intrinsic::mips_fdiv_d:
>>> > >> >      return lowerMSABinaryIntr(Op, DAG, ISD::FDIV);
>>> > >> >
>>> > >> > Added: llvm/trunk/test/CodeGen/Mips/msa/compare.ll
>>> > >> > URL: http://llvm.org/viewvc/llvm-
>>> > >>
>>> >
>>> project/llvm/trunk/test/CodeGen/Mips/msa/compare.ll?rev=191286&view
>>> > =
>>> > >> auto
>>> > >> >
>>> > >> ==========================================================
>>> > >> ====================
>>> > >> > --- llvm/trunk/test/CodeGen/Mips/msa/compare.ll (added)
>>> > >> > +++ llvm/trunk/test/CodeGen/Mips/msa/compare.ll Tue Sep 24
>>> > 05:46:19
>>> > >> 2013
>>> > >> > @@ -0,0 +1,641 @@
>>> > >> > +; RUN: llc -march=mips -mattr=+msa < %s | FileCheck %s
>>> > >> > +
>>> > >> > +define void @ceq_v16i8(<16 x i8>* %c, <16 x i8>* %a, <16 x i8>* %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: ceq_v16i8:
>>> > >> > +
>>> > >> > +  %1 = load <16 x i8>* %a
>>> > >> > +  ; CHECK-DAG: ld.b [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <16 x i8>* %b
>>> > >> > +  ; CHECK-DAG: ld.b [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = icmp eq <16 x i8> %1, %2
>>> > >> > +  %4 = sext <16 x i1> %3 to <16 x i8>
>>> > >> > +  ; CHECK-DAG: ceq.b [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <16 x i8> %4, <16 x i8>* %c
>>> > >> > +  ; CHECK-DAG: st.b [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size ceq_v16i8
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @ceq_v8i16(<8 x i16>* %c, <8 x i16>* %a, <8 x i16>* %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: ceq_v8i16:
>>> > >> > +
>>> > >> > +  %1 = load <8 x i16>* %a
>>> > >> > +  ; CHECK-DAG: ld.h [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <8 x i16>* %b
>>> > >> > +  ; CHECK-DAG: ld.h [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = icmp eq <8 x i16> %1, %2
>>> > >> > +  %4 = sext <8 x i1> %3 to <8 x i16>
>>> > >> > +  ; CHECK-DAG: ceq.h [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <8 x i16> %4, <8 x i16>* %c
>>> > >> > +  ; CHECK-DAG: st.h [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size ceq_v8i16
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @ceq_v4i32(<4 x i32>* %c, <4 x i32>* %a, <4 x i32>* %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: ceq_v4i32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x i32>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <4 x i32>* %b
>>> > >> > +  ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = icmp eq <4 x i32> %1, %2
>>> > >> > +  %4 = sext <4 x i1> %3 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: ceq.w [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <4 x i32> %4, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size ceq_v4i32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @ceq_v2i64(<2 x i64>* %c, <2 x i64>* %a, <2 x i64>* %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: ceq_v2i64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x i64>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <2 x i64>* %b
>>> > >> > +  ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = icmp eq <2 x i64> %1, %2
>>> > >> > +  %4 = sext <2 x i1> %3 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: ceq.d [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <2 x i64> %4, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size ceq_v2i64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @cle_s_v16i8(<16 x i8>* %c, <16 x i8>* %a, <16 x i8>* %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: cle_s_v16i8:
>>> > >> > +
>>> > >> > +  %1 = load <16 x i8>* %a
>>> > >> > +  ; CHECK-DAG: ld.b [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <16 x i8>* %b
>>> > >> > +  ; CHECK-DAG: ld.b [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = icmp sle <16 x i8> %1, %2
>>> > >> > +  %4 = sext <16 x i1> %3 to <16 x i8>
>>> > >> > +  ; CHECK-DAG: cle_s.b [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <16 x i8> %4, <16 x i8>* %c
>>> > >> > +  ; CHECK-DAG: st.b [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size cle_s_v16i8
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @cle_s_v8i16(<8 x i16>* %c, <8 x i16>* %a, <8 x i16>* %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: cle_s_v8i16:
>>> > >> > +
>>> > >> > +  %1 = load <8 x i16>* %a
>>> > >> > +  ; CHECK-DAG: ld.h [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <8 x i16>* %b
>>> > >> > +  ; CHECK-DAG: ld.h [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = icmp sle <8 x i16> %1, %2
>>> > >> > +  %4 = sext <8 x i1> %3 to <8 x i16>
>>> > >> > +  ; CHECK-DAG: cle_s.h [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <8 x i16> %4, <8 x i16>* %c
>>> > >> > +  ; CHECK-DAG: st.h [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size cle_s_v8i16
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @cle_s_v4i32(<4 x i32>* %c, <4 x i32>* %a, <4 x i32>* %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: cle_s_v4i32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x i32>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <4 x i32>* %b
>>> > >> > +  ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = icmp sle <4 x i32> %1, %2
>>> > >> > +  %4 = sext <4 x i1> %3 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: cle_s.w [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <4 x i32> %4, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size cle_s_v4i32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @cle_s_v2i64(<2 x i64>* %c, <2 x i64>* %a, <2 x i64>* %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: cle_s_v2i64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x i64>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <2 x i64>* %b
>>> > >> > +  ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = icmp sle <2 x i64> %1, %2
>>> > >> > +  %4 = sext <2 x i1> %3 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: cle_s.d [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <2 x i64> %4, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size cle_s_v2i64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @cle_u_v16i8(<16 x i8>* %c, <16 x i8>* %a, <16 x i8>* %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: cle_u_v16i8:
>>> > >> > +
>>> > >> > +  %1 = load <16 x i8>* %a
>>> > >> > +  ; CHECK-DAG: ld.b [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <16 x i8>* %b
>>> > >> > +  ; CHECK-DAG: ld.b [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = icmp ule <16 x i8> %1, %2
>>> > >> > +  %4 = sext <16 x i1> %3 to <16 x i8>
>>> > >> > +  ; CHECK-DAG: cle_u.b [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <16 x i8> %4, <16 x i8>* %c
>>> > >> > +  ; CHECK-DAG: st.b [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size cle_u_v16i8
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @cle_u_v8i16(<8 x i16>* %c, <8 x i16>* %a, <8 x i16>* %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: cle_u_v8i16:
>>> > >> > +
>>> > >> > +  %1 = load <8 x i16>* %a
>>> > >> > +  ; CHECK-DAG: ld.h [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <8 x i16>* %b
>>> > >> > +  ; CHECK-DAG: ld.h [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = icmp ule <8 x i16> %1, %2
>>> > >> > +  %4 = sext <8 x i1> %3 to <8 x i16>
>>> > >> > +  ; CHECK-DAG: cle_u.h [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <8 x i16> %4, <8 x i16>* %c
>>> > >> > +  ; CHECK-DAG: st.h [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size cle_u_v8i16
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @cle_u_v4i32(<4 x i32>* %c, <4 x i32>* %a, <4 x i32>* %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: cle_u_v4i32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x i32>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <4 x i32>* %b
>>> > >> > +  ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = icmp ule <4 x i32> %1, %2
>>> > >> > +  %4 = sext <4 x i1> %3 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: cle_u.w [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <4 x i32> %4, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size cle_u_v4i32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @cle_u_v2i64(<2 x i64>* %c, <2 x i64>* %a, <2 x i64>* %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: cle_u_v2i64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x i64>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <2 x i64>* %b
>>> > >> > +  ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = icmp ule <2 x i64> %1, %2
>>> > >> > +  %4 = sext <2 x i1> %3 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: cle_u.d [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <2 x i64> %4, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size cle_u_v2i64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clt_s_v16i8(<16 x i8>* %c, <16 x i8>* %a, <16 x i8>* %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: clt_s_v16i8:
>>> > >> > +
>>> > >> > +  %1 = load <16 x i8>* %a
>>> > >> > +  ; CHECK-DAG: ld.b [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <16 x i8>* %b
>>> > >> > +  ; CHECK-DAG: ld.b [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = icmp slt <16 x i8> %1, %2
>>> > >> > +  %4 = sext <16 x i1> %3 to <16 x i8>
>>> > >> > +  ; CHECK-DAG: clt_s.b [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <16 x i8> %4, <16 x i8>* %c
>>> > >> > +  ; CHECK-DAG: st.b [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clt_s_v16i8
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clt_s_v8i16(<8 x i16>* %c, <8 x i16>* %a, <8 x i16>* %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: clt_s_v8i16:
>>> > >> > +
>>> > >> > +  %1 = load <8 x i16>* %a
>>> > >> > +  ; CHECK-DAG: ld.h [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <8 x i16>* %b
>>> > >> > +  ; CHECK-DAG: ld.h [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = icmp slt <8 x i16> %1, %2
>>> > >> > +  %4 = sext <8 x i1> %3 to <8 x i16>
>>> > >> > +  ; CHECK-DAG: clt_s.h [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <8 x i16> %4, <8 x i16>* %c
>>> > >> > +  ; CHECK-DAG: st.h [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clt_s_v8i16
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clt_s_v4i32(<4 x i32>* %c, <4 x i32>* %a, <4 x i32>* %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: clt_s_v4i32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x i32>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <4 x i32>* %b
>>> > >> > +  ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = icmp slt <4 x i32> %1, %2
>>> > >> > +  %4 = sext <4 x i1> %3 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: clt_s.w [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <4 x i32> %4, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clt_s_v4i32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clt_s_v2i64(<2 x i64>* %c, <2 x i64>* %a, <2 x i64>* %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: clt_s_v2i64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x i64>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <2 x i64>* %b
>>> > >> > +  ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = icmp slt <2 x i64> %1, %2
>>> > >> > +  %4 = sext <2 x i1> %3 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: clt_s.d [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <2 x i64> %4, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clt_s_v2i64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clt_u_v16i8(<16 x i8>* %c, <16 x i8>* %a, <16 x i8>* %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: clt_u_v16i8:
>>> > >> > +
>>> > >> > +  %1 = load <16 x i8>* %a
>>> > >> > +  ; CHECK-DAG: ld.b [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <16 x i8>* %b
>>> > >> > +  ; CHECK-DAG: ld.b [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = icmp ult <16 x i8> %1, %2
>>> > >> > +  %4 = sext <16 x i1> %3 to <16 x i8>
>>> > >> > +  ; CHECK-DAG: clt_u.b [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <16 x i8> %4, <16 x i8>* %c
>>> > >> > +  ; CHECK-DAG: st.b [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clt_u_v16i8
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clt_u_v8i16(<8 x i16>* %c, <8 x i16>* %a, <8 x i16>* %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: clt_u_v8i16:
>>> > >> > +
>>> > >> > +  %1 = load <8 x i16>* %a
>>> > >> > +  ; CHECK-DAG: ld.h [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <8 x i16>* %b
>>> > >> > +  ; CHECK-DAG: ld.h [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = icmp ult <8 x i16> %1, %2
>>> > >> > +  %4 = sext <8 x i1> %3 to <8 x i16>
>>> > >> > +  ; CHECK-DAG: clt_u.h [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <8 x i16> %4, <8 x i16>* %c
>>> > >> > +  ; CHECK-DAG: st.h [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clt_u_v8i16
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clt_u_v4i32(<4 x i32>* %c, <4 x i32>* %a, <4 x i32>* %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: clt_u_v4i32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x i32>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <4 x i32>* %b
>>> > >> > +  ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = icmp ult <4 x i32> %1, %2
>>> > >> > +  %4 = sext <4 x i1> %3 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: clt_u.w [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <4 x i32> %4, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clt_u_v4i32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clt_u_v2i64(<2 x i64>* %c, <2 x i64>* %a, <2 x i64>* %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: clt_u_v2i64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x i64>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <2 x i64>* %b
>>> > >> > +  ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = icmp ult <2 x i64> %1, %2
>>> > >> > +  %4 = sext <2 x i1> %3 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: clt_u.d [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <2 x i64> %4, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clt_u_v2i64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @ceqi_v16i8(<16 x i8>* %c, <16 x i8>* %a) nounwind {
>>> > >> > +  ; CHECK: ceqi_v16i8:
>>> > >> > +
>>> > >> > +  %1 = load <16 x i8>* %a
>>> > >> > +  ; CHECK-DAG: ld.b [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = icmp eq <16 x i8> %1, <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1,
>>> > i8 1,
>>> > >> i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>
>>> > >> > +  %3 = sext <16 x i1> %2 to <16 x i8>
>>> > >> > +  ; CHECK-DAG: ceqi.b [[R3:\$w[0-9]+]], [[R1]], 1
>>> > >> > +  store <16 x i8> %3, <16 x i8>* %c
>>> > >> > +  ; CHECK-DAG: st.b [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size ceqi_v16i8
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @ceqi_v8i16(<8 x i16>* %c, <8 x i16>* %a) nounwind {
>>> > >> > +  ; CHECK: ceqi_v8i16:
>>> > >> > +
>>> > >> > +  %1 = load <8 x i16>* %a
>>> > >> > +  ; CHECK-DAG: ld.h [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = icmp eq <8 x i16> %1, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1,
>>> > i16 1>
>>> > >> > +  %3 = sext <8 x i1> %2 to <8 x i16>
>>> > >> > +  ; CHECK-DAG: ceqi.h [[R3:\$w[0-9]+]], [[R1]], 1
>>> > >> > +  store <8 x i16> %3, <8 x i16>* %c
>>> > >> > +  ; CHECK-DAG: st.h [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size ceqi_v8i16
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @ceqi_v4i32(<4 x i32>* %c, <4 x i32>* %a) nounwind {
>>> > >> > +  ; CHECK: ceqi_v4i32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x i32>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = icmp eq <4 x i32> %1, <i32 1, i32 1, i32 1, i32 1>
>>> > >> > +  %3 = sext <4 x i1> %2 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: ceqi.w [[R3:\$w[0-9]+]], [[R1]], 1
>>> > >> > +  store <4 x i32> %3, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size ceqi_v4i32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @ceqi_v2i64(<2 x i64>* %c, <2 x i64>* %a) nounwind {
>>> > >> > +  ; CHECK: ceqi_v2i64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x i64>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = icmp eq <2 x i64> %1, <i64 1, i64 1>
>>> > >> > +  %3 = sext <2 x i1> %2 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: ceqi.d [[R3:\$w[0-9]+]], [[R1]], 1
>>> > >> > +  store <2 x i64> %3, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size ceqi_v2i64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clei_s_v16i8(<16 x i8>* %c, <16 x i8>* %a) nounwind {
>>> > >> > +  ; CHECK: clei_s_v16i8:
>>> > >> > +
>>> > >> > +  %1 = load <16 x i8>* %a
>>> > >> > +  ; CHECK-DAG: ld.b [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = icmp sle <16 x i8> %1, <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8
>>> 1,
>>> > i8 1,
>>> > >> i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>
>>> > >> > +  %3 = sext <16 x i1> %2 to <16 x i8>
>>> > >> > +  ; CHECK-DAG: clei_s.b [[R3:\$w[0-9]+]], [[R1]], 1
>>> > >> > +  store <16 x i8> %3, <16 x i8>* %c
>>> > >> > +  ; CHECK-DAG: st.b [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clei_s_v16i8
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clei_s_v8i16(<8 x i16>* %c, <8 x i16>* %a) nounwind {
>>> > >> > +  ; CHECK: clei_s_v8i16:
>>> > >> > +
>>> > >> > +  %1 = load <8 x i16>* %a
>>> > >> > +  ; CHECK-DAG: ld.h [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = icmp sle <8 x i16> %1, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1,
>>> > i16
>>> > >> 1>
>>> > >> > +  %3 = sext <8 x i1> %2 to <8 x i16>
>>> > >> > +  ; CHECK-DAG: clei_s.h [[R3:\$w[0-9]+]], [[R1]], 1
>>> > >> > +  store <8 x i16> %3, <8 x i16>* %c
>>> > >> > +  ; CHECK-DAG: st.h [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clei_s_v8i16
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clei_s_v4i32(<4 x i32>* %c, <4 x i32>* %a) nounwind {
>>> > >> > +  ; CHECK: clei_s_v4i32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x i32>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = icmp sle <4 x i32> %1, <i32 1, i32 1, i32 1, i32 1>
>>> > >> > +  %3 = sext <4 x i1> %2 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: clei_s.w [[R3:\$w[0-9]+]], [[R1]], 1
>>> > >> > +  store <4 x i32> %3, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clei_s_v4i32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clei_s_v2i64(<2 x i64>* %c, <2 x i64>* %a) nounwind {
>>> > >> > +  ; CHECK: clei_s_v2i64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x i64>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = icmp sle <2 x i64> %1, <i64 1, i64 1>
>>> > >> > +  %3 = sext <2 x i1> %2 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: clei_s.d [[R3:\$w[0-9]+]], [[R1]], 1
>>> > >> > +  store <2 x i64> %3, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clei_s_v2i64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clei_u_v16i8(<16 x i8>* %c, <16 x i8>* %a) nounwind {
>>> > >> > +  ; CHECK: clei_u_v16i8:
>>> > >> > +
>>> > >> > +  %1 = load <16 x i8>* %a
>>> > >> > +  ; CHECK-DAG: ld.b [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = icmp ule <16 x i8> %1, <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8
>>> 1,
>>> > i8 1,
>>> > >> i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>
>>> > >> > +  %3 = sext <16 x i1> %2 to <16 x i8>
>>> > >> > +  ; CHECK-DAG: clei_u.b [[R3:\$w[0-9]+]], [[R1]], 1
>>> > >> > +  store <16 x i8> %3, <16 x i8>* %c
>>> > >> > +  ; CHECK-DAG: st.b [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clei_u_v16i8
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clei_u_v8i16(<8 x i16>* %c, <8 x i16>* %a) nounwind {
>>> > >> > +  ; CHECK: clei_u_v8i16:
>>> > >> > +
>>> > >> > +  %1 = load <8 x i16>* %a
>>> > >> > +  ; CHECK-DAG: ld.h [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = icmp ule <8 x i16> %1, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1,
>>> > i16
>>> > >> 1>
>>> > >> > +  %3 = sext <8 x i1> %2 to <8 x i16>
>>> > >> > +  ; CHECK-DAG: clei_u.h [[R3:\$w[0-9]+]], [[R1]], 1
>>> > >> > +  store <8 x i16> %3, <8 x i16>* %c
>>> > >> > +  ; CHECK-DAG: st.h [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clei_u_v8i16
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clei_u_v4i32(<4 x i32>* %c, <4 x i32>* %a) nounwind {
>>> > >> > +  ; CHECK: clei_u_v4i32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x i32>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = icmp ule <4 x i32> %1, <i32 1, i32 1, i32 1, i32 1>
>>> > >> > +  %3 = sext <4 x i1> %2 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: clei_u.w [[R3:\$w[0-9]+]], [[R1]], 1
>>> > >> > +  store <4 x i32> %3, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clei_u_v4i32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clei_u_v2i64(<2 x i64>* %c, <2 x i64>* %a) nounwind {
>>> > >> > +  ; CHECK: clei_u_v2i64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x i64>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = icmp ule <2 x i64> %1, <i64 1, i64 1>
>>> > >> > +  %3 = sext <2 x i1> %2 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: clei_u.d [[R3:\$w[0-9]+]], [[R1]], 1
>>> > >> > +  store <2 x i64> %3, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clei_u_v2i64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clti_s_v16i8(<16 x i8>* %c, <16 x i8>* %a) nounwind {
>>> > >> > +  ; CHECK: clti_s_v16i8:
>>> > >> > +
>>> > >> > +  %1 = load <16 x i8>* %a
>>> > >> > +  ; CHECK-DAG: ld.b [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = icmp slt <16 x i8> %1, <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1,
>>> > i8 1,
>>> > >> i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>
>>> > >> > +  %3 = sext <16 x i1> %2 to <16 x i8>
>>> > >> > +  ; CHECK-DAG: clti_s.b [[R3:\$w[0-9]+]], [[R1]], 1
>>> > >> > +  store <16 x i8> %3, <16 x i8>* %c
>>> > >> > +  ; CHECK-DAG: st.b [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clti_s_v16i8
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clti_s_v8i16(<8 x i16>* %c, <8 x i16>* %a) nounwind {
>>> > >> > +  ; CHECK: clti_s_v8i16:
>>> > >> > +
>>> > >> > +  %1 = load <8 x i16>* %a
>>> > >> > +  ; CHECK-DAG: ld.h [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = icmp slt <8 x i16> %1, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1,
>>> > i16 1>
>>> > >> > +  %3 = sext <8 x i1> %2 to <8 x i16>
>>> > >> > +  ; CHECK-DAG: clti_s.h [[R3:\$w[0-9]+]], [[R1]], 1
>>> > >> > +  store <8 x i16> %3, <8 x i16>* %c
>>> > >> > +  ; CHECK-DAG: st.h [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clti_s_v8i16
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clti_s_v4i32(<4 x i32>* %c, <4 x i32>* %a) nounwind {
>>> > >> > +  ; CHECK: clti_s_v4i32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x i32>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = icmp slt <4 x i32> %1, <i32 1, i32 1, i32 1, i32 1>
>>> > >> > +  %3 = sext <4 x i1> %2 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: clti_s.w [[R3:\$w[0-9]+]], [[R1]], 1
>>> > >> > +  store <4 x i32> %3, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clti_s_v4i32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clti_s_v2i64(<2 x i64>* %c, <2 x i64>* %a) nounwind {
>>> > >> > +  ; CHECK: clti_s_v2i64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x i64>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = icmp slt <2 x i64> %1, <i64 1, i64 1>
>>> > >> > +  %3 = sext <2 x i1> %2 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: clti_s.d [[R3:\$w[0-9]+]], [[R1]], 1
>>> > >> > +  store <2 x i64> %3, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clti_s_v2i64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clti_u_v16i8(<16 x i8>* %c, <16 x i8>* %a) nounwind {
>>> > >> > +  ; CHECK: clti_u_v16i8:
>>> > >> > +
>>> > >> > +  %1 = load <16 x i8>* %a
>>> > >> > +  ; CHECK-DAG: ld.b [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = icmp ult <16 x i8> %1, <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1,
>>> > i8 1,
>>> > >> i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>
>>> > >> > +  %3 = sext <16 x i1> %2 to <16 x i8>
>>> > >> > +  ; CHECK-DAG: clti_u.b [[R3:\$w[0-9]+]], [[R1]], 1
>>> > >> > +  store <16 x i8> %3, <16 x i8>* %c
>>> > >> > +  ; CHECK-DAG: st.b [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clti_u_v16i8
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clti_u_v8i16(<8 x i16>* %c, <8 x i16>* %a) nounwind {
>>> > >> > +  ; CHECK: clti_u_v8i16:
>>> > >> > +
>>> > >> > +  %1 = load <8 x i16>* %a
>>> > >> > +  ; CHECK-DAG: ld.h [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = icmp ult <8 x i16> %1, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1,
>>> > i16 1>
>>> > >> > +  %3 = sext <8 x i1> %2 to <8 x i16>
>>> > >> > +  ; CHECK-DAG: clti_u.h [[R3:\$w[0-9]+]], [[R1]], 1
>>> > >> > +  store <8 x i16> %3, <8 x i16>* %c
>>> > >> > +  ; CHECK-DAG: st.h [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clti_u_v8i16
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clti_u_v4i32(<4 x i32>* %c, <4 x i32>* %a) nounwind {
>>> > >> > +  ; CHECK: clti_u_v4i32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x i32>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = icmp ult <4 x i32> %1, <i32 1, i32 1, i32 1, i32 1>
>>> > >> > +  %3 = sext <4 x i1> %2 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: clti_u.w [[R3:\$w[0-9]+]], [[R1]], 1
>>> > >> > +  store <4 x i32> %3, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clti_u_v4i32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @clti_u_v2i64(<2 x i64>* %c, <2 x i64>* %a) nounwind {
>>> > >> > +  ; CHECK: clti_u_v2i64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x i64>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = icmp ult <2 x i64> %1, <i64 1, i64 1>
>>> > >> > +  %3 = sext <2 x i1> %2 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: clti_u.d [[R3:\$w[0-9]+]], [[R1]], 1
>>> > >> > +  store <2 x i64> %3, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size clti_u_v2i64
>>> > >> > +}
>>> > >> >
>>> > >> > Added: llvm/trunk/test/CodeGen/Mips/msa/compare_float.ll
>>> > >> > URL: http://llvm.org/viewvc/llvm-
>>> > >>
>>> >
>>> project/llvm/trunk/test/CodeGen/Mips/msa/compare_float.ll?rev=191286&
>>> > >> view=auto
>>> > >> >
>>> > >> ==========================================================
>>> > >> ====================
>>> > >> > --- llvm/trunk/test/CodeGen/Mips/msa/compare_float.ll (added)
>>> > >> > +++ llvm/trunk/test/CodeGen/Mips/msa/compare_float.ll Tue Sep 24
>>> > >> 05:46:19 2013
>>> > >> > @@ -0,0 +1,518 @@
>>> > >> > +; RUN: llc -march=mips -mattr=+msa < %s | FileCheck %s
>>> > >> > +
>>> > >> > +declare <4 x float> @llvm.mips.fmax.w(<4 x float>, <4 x float>)
>>> > nounwind
>>> > >> > +declare <2 x double> @llvm.mips.fmax.d(<2 x double>, <2 x double>)
>>> > >> nounwind
>>> > >> > +declare <4 x float> @llvm.mips.fmin.w(<4 x float>, <4 x float>)
>>> > nounwind
>>> > >> > +declare <2 x double> @llvm.mips.fmin.d(<2 x double>, <2 x double>)
>>> > >> nounwind
>>> > >> > +
>>> > >> > +define void @false_v4f32(<4 x i32>* %c, <4 x float>* %a, <4 x float>*
>>> > %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: false_v4f32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x float>* %a
>>> > >> > +  %2 = load <4 x float>* %b
>>> > >> > +  %3 = fcmp false <4 x float> %1, %2
>>> > >> > +  %4 = sext <4 x i1> %3 to <4 x i32>
>>> > >> > +  store <4 x i32> %4, <4 x i32>* %c
>>> > >> > +  ret void
>>> > >> > +
>>> > >> > +  ; (setcc $a, $b, SETFALSE) is always folded, so we won't get fcaf:
>>> > >> > +  ; CHECK-DAG: ldi.b [[R1:\$w[0-9]+]], 0
>>> > >> > +  ; CHECK-DAG: st.b [[R1]], 0($4)
>>> > >> > +  ; CHECK: .size false_v4f32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @false_v2f64(<2 x i64>* %c, <2 x double>* %a, <2 x
>>> > double>*
>>> > >> %b) nounwind {
>>> > >> > +  ; CHECK: false_v2f64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x double>* %a
>>> > >> > +  %2 = load <2 x double>* %b
>>> > >> > +  %3 = fcmp false <2 x double> %1, %2
>>> > >> > +  %4 = sext <2 x i1> %3 to <2 x i64>
>>> > >> > +  store <2 x i64> %4, <2 x i64>* %c
>>> > >> > +  ret void
>>> > >> > +
>>> > >> > +  ; FIXME: This code is correct, but poor. Ideally it would be similar to
>>> > >> > +  ;        the code in @false_v4f32
>>> > >> > +  ; CHECK-DAG: ldi.b [[R1:\$w[0-9]+]], 0
>>> > >> > +  ; CHECK-DAG: slli.d [[R3:\$w[0-9]+]], [[R1]], 63
>>> > >> > +  ; CHECK-DAG: srai.d [[R4:\$w[0-9]+]], [[R3]], 63
>>> > >> > +  ; CHECK-DAG: st.d [[R4]], 0($4)
>>> > >> > +  ; CHECK: .size false_v2f64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @oeq_v4f32(<4 x i32>* %c, <4 x float>* %a, <4 x float>*
>>> > %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: oeq_v4f32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x float>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <4 x float>* %b
>>> > >> > +  ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp oeq <4 x float> %1, %2
>>> > >> > +  %4 = sext <4 x i1> %3 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: fceq.w [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <4 x i32> %4, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size oeq_v4f32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @oeq_v2f64(<2 x i64>* %c, <2 x double>* %a, <2 x
>>> > double>*
>>> > >> %b) nounwind {
>>> > >> > +  ; CHECK: oeq_v2f64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x double>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <2 x double>* %b
>>> > >> > +  ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp oeq <2 x double> %1, %2
>>> > >> > +  %4 = sext <2 x i1> %3 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: fceq.d [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <2 x i64> %4, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size oeq_v2f64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @oge_v4f32(<4 x i32>* %c, <4 x float>* %a, <4 x float>*
>>> %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: oge_v4f32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x float>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <4 x float>* %b
>>> > >> > +  ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp oge <4 x float> %1, %2
>>> > >> > +  %4 = sext <4 x i1> %3 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: fcle.w [[R3:\$w[0-9]+]], [[R2]], [[R1]]
>>> > >> > +  store <4 x i32> %4, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size oge_v4f32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @oge_v2f64(<2 x i64>* %c, <2 x double>* %a, <2 x
>>> > double>*
>>> > >> %b) nounwind {
>>> > >> > +  ; CHECK: oge_v2f64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x double>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <2 x double>* %b
>>> > >> > +  ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp oge <2 x double> %1, %2
>>> > >> > +  %4 = sext <2 x i1> %3 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: fcle.d [[R3:\$w[0-9]+]], [[R2]], [[R1]]
>>> > >> > +  store <2 x i64> %4, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size oge_v2f64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @ogt_v4f32(<4 x i32>* %c, <4 x float>* %a, <4 x float>*
>>> %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: ogt_v4f32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x float>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <4 x float>* %b
>>> > >> > +  ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp ogt <4 x float> %1, %2
>>> > >> > +  %4 = sext <4 x i1> %3 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: fclt.w [[R3:\$w[0-9]+]], [[R2]], [[R1]]
>>> > >> > +  store <4 x i32> %4, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size ogt_v4f32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @ogt_v2f64(<2 x i64>* %c, <2 x double>* %a, <2 x
>>> double>*
>>> > >> %b) nounwind {
>>> > >> > +  ; CHECK: ogt_v2f64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x double>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <2 x double>* %b
>>> > >> > +  ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp ogt <2 x double> %1, %2
>>> > >> > +  %4 = sext <2 x i1> %3 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: fclt.d [[R3:\$w[0-9]+]], [[R2]], [[R1]]
>>> > >> > +  store <2 x i64> %4, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size ogt_v2f64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @ole_v4f32(<4 x i32>* %c, <4 x float>* %a, <4 x float>*
>>> %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: ole_v4f32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x float>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <4 x float>* %b
>>> > >> > +  ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp ole <4 x float> %1, %2
>>> > >> > +  %4 = sext <4 x i1> %3 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: fcle.w [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <4 x i32> %4, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size ole_v4f32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @ole_v2f64(<2 x i64>* %c, <2 x double>* %a, <2 x
>>> double>*
>>> > >> %b) nounwind {
>>> > >> > +  ; CHECK: ole_v2f64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x double>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <2 x double>* %b
>>> > >> > +  ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp ole <2 x double> %1, %2
>>> > >> > +  %4 = sext <2 x i1> %3 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: fcle.d [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <2 x i64> %4, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size ole_v2f64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @olt_v4f32(<4 x i32>* %c, <4 x float>* %a, <4 x float>*
>>> %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: olt_v4f32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x float>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <4 x float>* %b
>>> > >> > +  ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp olt <4 x float> %1, %2
>>> > >> > +  %4 = sext <4 x i1> %3 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: fclt.w [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <4 x i32> %4, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size olt_v4f32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @olt_v2f64(<2 x i64>* %c, <2 x double>* %a, <2 x
>>> double>*
>>> > >> %b) nounwind {
>>> > >> > +  ; CHECK: olt_v2f64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x double>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <2 x double>* %b
>>> > >> > +  ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp olt <2 x double> %1, %2
>>> > >> > +  %4 = sext <2 x i1> %3 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: fclt.d [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <2 x i64> %4, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size olt_v2f64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @one_v4f32(<4 x i32>* %c, <4 x float>* %a, <4 x float>*
>>> > %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: one_v4f32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x float>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <4 x float>* %b
>>> > >> > +  ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp one <4 x float> %1, %2
>>> > >> > +  %4 = sext <4 x i1> %3 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: fcne.w [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <4 x i32> %4, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size one_v4f32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @one_v2f64(<2 x i64>* %c, <2 x double>* %a, <2 x
>>> > double>*
>>> > >> %b) nounwind {
>>> > >> > +  ; CHECK: one_v2f64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x double>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <2 x double>* %b
>>> > >> > +  ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp one <2 x double> %1, %2
>>> > >> > +  %4 = sext <2 x i1> %3 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: fcne.d [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <2 x i64> %4, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size one_v2f64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @ord_v4f32(<4 x i32>* %c, <4 x float>* %a, <4 x float>*
>>> %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: ord_v4f32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x float>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <4 x float>* %b
>>> > >> > +  ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp ord <4 x float> %1, %2
>>> > >> > +  %4 = sext <4 x i1> %3 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: fcor.w [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <4 x i32> %4, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size ord_v4f32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @ord_v2f64(<2 x i64>* %c, <2 x double>* %a, <2 x
>>> > double>*
>>> > >> %b) nounwind {
>>> > >> > +  ; CHECK: ord_v2f64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x double>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <2 x double>* %b
>>> > >> > +  ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp ord <2 x double> %1, %2
>>> > >> > +  %4 = sext <2 x i1> %3 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: fcor.d [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <2 x i64> %4, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size ord_v2f64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @ueq_v4f32(<4 x i32>* %c, <4 x float>* %a, <4 x float>*
>>> > %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: ueq_v4f32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x float>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <4 x float>* %b
>>> > >> > +  ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp ueq <4 x float> %1, %2
>>> > >> > +  %4 = sext <4 x i1> %3 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: fcueq.w [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <4 x i32> %4, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size ueq_v4f32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @ueq_v2f64(<2 x i64>* %c, <2 x double>* %a, <2 x
>>> > double>*
>>> > >> %b) nounwind {
>>> > >> > +  ; CHECK: ueq_v2f64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x double>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <2 x double>* %b
>>> > >> > +  ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp ueq <2 x double> %1, %2
>>> > >> > +  %4 = sext <2 x i1> %3 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: fcueq.d [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <2 x i64> %4, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size ueq_v2f64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @uge_v4f32(<4 x i32>* %c, <4 x float>* %a, <4 x float>*
>>> %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: uge_v4f32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x float>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <4 x float>* %b
>>> > >> > +  ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp uge <4 x float> %1, %2
>>> > >> > +  %4 = sext <4 x i1> %3 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: fcule.w [[R3:\$w[0-9]+]], [[R2]], [[R1]]
>>> > >> > +  store <4 x i32> %4, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size uge_v4f32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @uge_v2f64(<2 x i64>* %c, <2 x double>* %a, <2 x
>>> > double>*
>>> > >> %b) nounwind {
>>> > >> > +  ; CHECK: uge_v2f64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x double>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <2 x double>* %b
>>> > >> > +  ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp uge <2 x double> %1, %2
>>> > >> > +  %4 = sext <2 x i1> %3 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: fcule.d [[R3:\$w[0-9]+]], [[R2]], [[R1]]
>>> > >> > +  store <2 x i64> %4, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size uge_v2f64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @ugt_v4f32(<4 x i32>* %c, <4 x float>* %a, <4 x float>*
>>> %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: ugt_v4f32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x float>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <4 x float>* %b
>>> > >> > +  ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp ugt <4 x float> %1, %2
>>> > >> > +  %4 = sext <4 x i1> %3 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: fcult.w [[R3:\$w[0-9]+]], [[R2]], [[R1]]
>>> > >> > +  store <4 x i32> %4, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size ugt_v4f32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @ugt_v2f64(<2 x i64>* %c, <2 x double>* %a, <2 x
>>> double>*
>>> > >> %b) nounwind {
>>> > >> > +  ; CHECK: ugt_v2f64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x double>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <2 x double>* %b
>>> > >> > +  ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp ugt <2 x double> %1, %2
>>> > >> > +  %4 = sext <2 x i1> %3 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: fcult.d [[R3:\$w[0-9]+]], [[R2]], [[R1]]
>>> > >> > +  store <2 x i64> %4, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size ugt_v2f64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @ule_v4f32(<4 x i32>* %c, <4 x float>* %a, <4 x float>*
>>> %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: ule_v4f32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x float>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <4 x float>* %b
>>> > >> > +  ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp ule <4 x float> %1, %2
>>> > >> > +  %4 = sext <4 x i1> %3 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: fcule.w [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <4 x i32> %4, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size ule_v4f32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @ule_v2f64(<2 x i64>* %c, <2 x double>* %a, <2 x
>>> double>*
>>> > >> %b) nounwind {
>>> > >> > +  ; CHECK: ule_v2f64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x double>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <2 x double>* %b
>>> > >> > +  ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp ule <2 x double> %1, %2
>>> > >> > +  %4 = sext <2 x i1> %3 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: fcule.d [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <2 x i64> %4, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size ule_v2f64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @ult_v4f32(<4 x i32>* %c, <4 x float>* %a, <4 x float>*
>>> %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: ult_v4f32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x float>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <4 x float>* %b
>>> > >> > +  ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp ult <4 x float> %1, %2
>>> > >> > +  %4 = sext <4 x i1> %3 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: fcult.w [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <4 x i32> %4, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size ult_v4f32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @ult_v2f64(<2 x i64>* %c, <2 x double>* %a, <2 x
>>> double>*
>>> > >> %b) nounwind {
>>> > >> > +  ; CHECK: ult_v2f64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x double>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <2 x double>* %b
>>> > >> > +  ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp ult <2 x double> %1, %2
>>> > >> > +  %4 = sext <2 x i1> %3 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: fcult.d [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <2 x i64> %4, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size ult_v2f64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @uno_v4f32(<4 x i32>* %c, <4 x float>* %a, <4 x float>*
>>> > %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: uno_v4f32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x float>* %a
>>> > >> > +  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <4 x float>* %b
>>> > >> > +  ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp uno <4 x float> %1, %2
>>> > >> > +  %4 = sext <4 x i1> %3 to <4 x i32>
>>> > >> > +  ; CHECK-DAG: fcun.w [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <4 x i32> %4, <4 x i32>* %c
>>> > >> > +  ; CHECK-DAG: st.w [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size uno_v4f32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @uno_v2f64(<2 x i64>* %c, <2 x double>* %a, <2 x
>>> > double>*
>>> > >> %b) nounwind {
>>> > >> > +  ; CHECK: uno_v2f64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x double>* %a
>>> > >> > +  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
>>> > >> > +  %2 = load <2 x double>* %b
>>> > >> > +  ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
>>> > >> > +  %3 = fcmp uno <2 x double> %1, %2
>>> > >> > +  %4 = sext <2 x i1> %3 to <2 x i64>
>>> > >> > +  ; CHECK-DAG: fcun.d [[R3:\$w[0-9]+]], [[R1]], [[R2]]
>>> > >> > +  store <2 x i64> %4, <2 x i64>* %c
>>> > >> > +  ; CHECK-DAG: st.d [[R3]], 0($4)
>>> > >> > +
>>> > >> > +  ret void
>>> > >> > +  ; CHECK: .size uno_v2f64
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @true_v4f32(<4 x i32>* %c, <4 x float>* %a, <4 x float>*
>>> > %b)
>>> > >> nounwind {
>>> > >> > +  ; CHECK: true_v4f32:
>>> > >> > +
>>> > >> > +  %1 = load <4 x float>* %a
>>> > >> > +  %2 = load <4 x float>* %b
>>> > >> > +  %3 = fcmp true <4 x float> %1, %2
>>> > >> > +  %4 = sext <4 x i1> %3 to <4 x i32>
>>> > >> > +  store <4 x i32> %4, <4 x i32>* %c
>>> > >> > +  ret void
>>> > >> > +
>>> > >> > +  ; (setcc $a, $b, SETTRUE) is always folded, so we won't get fcaf:
>>> > >> > +  ; CHECK-DAG: ldi.b [[R1:\$w[0-9]+]], -1
>>> > >> > +  ; CHECK-DAG: st.b [[R1]], 0($4)
>>> > >> > +  ; CHECK: .size true_v4f32
>>> > >> > +}
>>> > >> > +
>>> > >> > +define void @true_v2f64(<2 x i64>* %c, <2 x double>* %a, <2 x
>>> > double>*
>>> > >> %b) nounwind {
>>> > >> > +  ; CHECK: true_v2f64:
>>> > >> > +
>>> > >> > +  %1 = load <2 x double>* %a
>>> > >> > +  %2 = load <2 x double>* %b
>>> > >> > +  %3 = fcmp true <2 x double> %1, %2
>>> > >> > +  %4 = sext <2 x i1> %3 to <2 x i64>
>>> > >> > +  store <2 x i64> %4, <2 x i64>* %c
>>> > >> > +  ret void
>>> > >> > +
>>> > >> > +  ; FIXME: This code is correct, but poor. Ideally it would be similar to
>>> > >> > +  ;        the code in @true_v4f32
>>> > >> > +  ; CHECK-DAG: ldi.d [[R1:\$w[0-9]+]], 1
>>> > >> > +  ; CHECK-DAG: slli.d [[R3:\$w[0-9]+]], [[R1]], 63
>>> > >> > +  ; CHECK-DAG: srai.d [[R4:\$w[0-9]+]], [[R3]], 63
>>> > >> > +  ; CHECK-DAG: st.d [[R4]], 0($4)
>>> > >> > +  ; CHECK: .size true_v2f64
>>> > >> > +}
>>> > >> >
>>> > >> >
>>> > >> > _______________________________________________
>>> > >> > llvm-commits mailing list
>>> > >> > llvm-commits at cs.uiuc.edu
>>> > >> > http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits



More information about the llvm-commits mailing list