Source repositories:
https://github.com/openrisc/llvm-or1k
https://github.com/openrisc/clang-or1k
https://github.com/llvmpy/llvmpy (branch: llvm-3.4)

Known to work commit IDs:
LLVM (needs revert): b3a48efb2c05ed6cedc5395ae726c6a6573ef3ba
Clang (needs revert): 02d831c7e7dc1517abed9cc96abdfb937af954eb
llvmpy (head OK): 7af2f7140391d4f708adf2721e84f23c1b89e97a

Clang patch:
diff --cc lib/CodeGen/CGBuiltin.cpp
index 52e40db,d06c660..0000000
--- a/lib/CodeGen/CGBuiltin.cpp
+++ b/lib/CodeGen/CGBuiltin.cpp
@@@ -1760,191 -1766,352 +1760,6 @@@ static Value *EmitAArch64ScalarBuiltinE
    // argument that specifies the vector type, need to handle each case.
    switch (BuiltinID) {
    default: break;
--  // Scalar Add
--  case AArch64::BI__builtin_neon_vaddd_s64:
--    Int = Intrinsic::aarch64_neon_vaddds;
--    s = "vaddds"; break;
--  case AArch64::BI__builtin_neon_vaddd_u64:
--    Int = Intrinsic::aarch64_neon_vadddu;
--    s = "vadddu"; break;
--  // Scalar Sub
--  case AArch64::BI__builtin_neon_vsubd_s64:
--    Int = Intrinsic::aarch64_neon_vsubds;
--    s = "vsubds"; break;
--  case AArch64::BI__builtin_neon_vsubd_u64:
--    Int = Intrinsic::aarch64_neon_vsubdu;
--    s = "vsubdu"; break;
--  // Scalar Saturating Add
--  case AArch64::BI__builtin_neon_vqaddb_s8:
--  case AArch64::BI__builtin_neon_vqaddh_s16:
--  case AArch64::BI__builtin_neon_vqadds_s32:
--  case AArch64::BI__builtin_neon_vqaddd_s64:
--    Int = Intrinsic::aarch64_neon_vqadds;
--    s = "vqadds"; OverloadInt = true; break;
--  case AArch64::BI__builtin_neon_vqaddb_u8:
--  case AArch64::BI__builtin_neon_vqaddh_u16:
--  case AArch64::BI__builtin_neon_vqadds_u32:
--  case AArch64::BI__builtin_neon_vqaddd_u64:
--    Int = Intrinsic::aarch64_neon_vqaddu;
--    s = "vqaddu"; OverloadInt = true; break;
--  // Scalar Saturating Sub
--  case AArch64::BI__builtin_neon_vqsubb_s8:
--  case AArch64::BI__builtin_neon_vqsubh_s16:
--  case AArch64::BI__builtin_neon_vqsubs_s32:
--  case AArch64::BI__builtin_neon_vqsubd_s64:
--    Int = Intrinsic::aarch64_neon_vqsubs;
--    s = "vqsubs"; OverloadInt = true; break;
--  case AArch64::BI__builtin_neon_vqsubb_u8:
--  case AArch64::BI__builtin_neon_vqsubh_u16:
--  case AArch64::BI__builtin_neon_vqsubs_u32:
--  case AArch64::BI__builtin_neon_vqsubd_u64:
--    Int = Intrinsic::aarch64_neon_vqsubu;
--    s = "vqsubu"; OverloadInt = true; break;
--  // Scalar Shift Left
--  case AArch64::BI__builtin_neon_vshld_s64:
--    Int = Intrinsic::aarch64_neon_vshlds;
--    s = "vshlds"; break;
--  case AArch64::BI__builtin_neon_vshld_u64:
--    Int = Intrinsic::aarch64_neon_vshldu;
--    s = "vshldu"; break;
--  // Scalar Saturating Shift Left
--  case AArch64::BI__builtin_neon_vqshlb_s8:
--  case AArch64::BI__builtin_neon_vqshlh_s16:
--  case AArch64::BI__builtin_neon_vqshls_s32:
--  case AArch64::BI__builtin_neon_vqshld_s64:
--    Int = Intrinsic::aarch64_neon_vqshls;
--    s = "vqshls"; OverloadInt = true; break;
--  case AArch64::BI__builtin_neon_vqshlb_u8:
--  case AArch64::BI__builtin_neon_vqshlh_u16:
--  case AArch64::BI__builtin_neon_vqshls_u32:
--  case AArch64::BI__builtin_neon_vqshld_u64:
--    Int = Intrinsic::aarch64_neon_vqshlu;
--    s = "vqshlu"; OverloadInt = true; break;
--  // Scalar Rouding Shift Left
--  case AArch64::BI__builtin_neon_vrshld_s64:
--    Int = Intrinsic::aarch64_neon_vrshlds;
--    s = "vrshlds"; break;
--  case AArch64::BI__builtin_neon_vrshld_u64:
--    Int = Intrinsic::aarch64_neon_vrshldu;
--    s = "vrshldu"; break;
--  // Scalar Saturating Rouding Shift Left
--  case AArch64::BI__builtin_neon_vqrshlb_s8:
--  case AArch64::BI__builtin_neon_vqrshlh_s16:
--  case AArch64::BI__builtin_neon_vqrshls_s32:
--  case AArch64::BI__builtin_neon_vqrshld_s64:
--    Int = Intrinsic::aarch64_neon_vqrshls;
--    s = "vqrshls"; OverloadInt = true; break;
--  case AArch64::BI__builtin_neon_vqrshlb_u8:
--  case AArch64::BI__builtin_neon_vqrshlh_u16:
--  case AArch64::BI__builtin_neon_vqrshls_u32:
--  case AArch64::BI__builtin_neon_vqrshld_u64:
--    Int = Intrinsic::aarch64_neon_vqrshlu;
--    s = "vqrshlu"; OverloadInt = true; break;
--  // Scalar Reduce Pairwise Add
--  case AArch64::BI__builtin_neon_vpaddd_s64:
--    Int = Intrinsic::aarch64_neon_vpadd; s = "vpadd";
--    break;
--  case AArch64::BI__builtin_neon_vpadds_f32:
--    Int = Intrinsic::aarch64_neon_vpfadd; s = "vpfadd";
--    break;
--  case AArch64::BI__builtin_neon_vpaddd_f64:
--    Int = Intrinsic::aarch64_neon_vpfaddq; s = "vpfaddq";
--    break;
--  // Scalar Reduce Pairwise Floating Point Max
--  case AArch64::BI__builtin_neon_vpmaxs_f32:
--    Int = Intrinsic::aarch64_neon_vpmax; s = "vpmax";
--    break;
--  case AArch64::BI__builtin_neon_vpmaxqd_f64:
--    Int = Intrinsic::aarch64_neon_vpmaxq; s = "vpmaxq";
--    break;
--  // Scalar Reduce Pairwise Floating Point Min
--  case AArch64::BI__builtin_neon_vpmins_f32:
--    Int = Intrinsic::aarch64_neon_vpmin; s = "vpmin";
--    break;
--  case AArch64::BI__builtin_neon_vpminqd_f64:
--    Int = Intrinsic::aarch64_neon_vpminq; s = "vpminq";
--    break;
--  // Scalar Reduce Pairwise Floating Point Maxnm
--  case AArch64::BI__builtin_neon_vpmaxnms_f32:
--    Int = Intrinsic::aarch64_neon_vpfmaxnm; s = "vpfmaxnm";
--    break;
--  case AArch64::BI__builtin_neon_vpmaxnmqd_f64:
--    Int = Intrinsic::aarch64_neon_vpfmaxnmq; s = "vpfmaxnmq";
--    break;
--  // Scalar Reduce Pairwise Floating Point Minnm
--  case AArch64::BI__builtin_neon_vpminnms_f32:
--    Int = Intrinsic::aarch64_neon_vpfminnm; s = "vpfminnm";
--    break;
--  case AArch64::BI__builtin_neon_vpminnmqd_f64:
--    Int = Intrinsic::aarch64_neon_vpfminnmq; s = "vpfminnmq";
--    break;
--  // The followings are intrinsics with scalar results generated AcrossVec vectors
 -    #if 0
--  case AArch64::BI__builtin_neon_vaddlv_s8:
--  case AArch64::BI__builtin_neon_vaddlv_s16:
--  case AArch64::BI__builtin_neon_vaddlvq_s8:
--  case AArch64::BI__builtin_neon_vaddlvq_s16:
--  case AArch64::BI__builtin_neon_vaddlvq_s32:
--    Int = Intrinsic::aarch64_neon_saddlv;
--    AcrossVec = true; ExtendEle = true; s = "saddlv"; break;
--  case AArch64::BI__builtin_neon_vaddlv_u8:
--  case AArch64::BI__builtin_neon_vaddlv_u16:
--  case AArch64::BI__builtin_neon_vaddlvq_u8:
--  case AArch64::BI__builtin_neon_vaddlvq_u16:
--  case AArch64::BI__builtin_neon_vaddlvq_u32:
--    Int = Intrinsic::aarch64_neon_uaddlv;
--    AcrossVec = true; ExtendEle = true; s = "uaddlv"; break;
--  case AArch64::BI__builtin_neon_vmaxv_s8:
--  case AArch64::BI__builtin_neon_vmaxv_s16:
--  case AArch64::BI__builtin_neon_vmaxvq_s8:
--  case AArch64::BI__builtin_neon_vmaxvq_s16:
--  case AArch64::BI__builtin_neon_vmaxvq_s32:
--    Int = Intrinsic::aarch64_neon_smaxv;
--    AcrossVec = true; ExtendEle = false; s = "smaxv"; break;
--  case AArch64::BI__builtin_neon_vmaxv_u8:
--  case AArch64::BI__builtin_neon_vmaxv_u16:
--  case AArch64::BI__builtin_neon_vmaxvq_u8:
--  case AArch64::BI__builtin_neon_vmaxvq_u16:
--  case AArch64::BI__builtin_neon_vmaxvq_u32:
--    Int = Intrinsic::aarch64_neon_umaxv;
--    AcrossVec = true; ExtendEle = false; s = "umaxv"; break;
--  case AArch64::BI__builtin_neon_vminv_s8:
--  case AArch64::BI__builtin_neon_vminv_s16:
--  case AArch64::BI__builtin_neon_vminvq_s8:
--  case AArch64::BI__builtin_neon_vminvq_s16:
--  case AArch64::BI__builtin_neon_vminvq_s32:
--    Int = Intrinsic::aarch64_neon_sminv;
--    AcrossVec = true; ExtendEle = false; s = "sminv"; break;
--  case AArch64::BI__builtin_neon_vminv_u8:
--  case AArch64::BI__builtin_neon_vminv_u16:
--  case AArch64::BI__builtin_neon_vminvq_u8:
--  case AArch64::BI__builtin_neon_vminvq_u16:
--  case AArch64::BI__builtin_neon_vminvq_u32:
--    Int = Intrinsic::aarch64_neon_uminv;
--    AcrossVec = true; ExtendEle = false; s = "uminv"; break;
--  case AArch64::BI__builtin_neon_vaddv_s8:
--  case AArch64::BI__builtin_neon_vaddv_s16:
--  case AArch64::BI__builtin_neon_vaddvq_s8:
--  case AArch64::BI__builtin_neon_vaddvq_s16:
--  case AArch64::BI__builtin_neon_vaddvq_s32:
--  case AArch64::BI__builtin_neon_vaddv_u8:
--  case AArch64::BI__builtin_neon_vaddv_u16:
--  case AArch64::BI__builtin_neon_vaddvq_u8:
--  case AArch64::BI__builtin_neon_vaddvq_u16:
--  case AArch64::BI__builtin_neon_vaddvq_u32:
--    Int = Intrinsic::aarch64_neon_vaddv;
--    AcrossVec = true; ExtendEle = false; s = "vaddv"; break;      
--  case AArch64::BI__builtin_neon_vmaxvq_f32:
--    Int = Intrinsic::aarch64_neon_vmaxv;
--    AcrossVec = true; ExtendEle = false; s = "vmaxv"; break;
--  case AArch64::BI__builtin_neon_vminvq_f32:
--    Int = Intrinsic::aarch64_neon_vminv;
--    AcrossVec = true; ExtendEle = false; s = "vminv"; break;
--  case AArch64::BI__builtin_neon_vmaxnmvq_f32:
--    Int = Intrinsic::aarch64_neon_vmaxnmv;
--    AcrossVec = true; ExtendEle = false; s = "vmaxnmv"; break;
--  case AArch64::BI__builtin_neon_vminnmvq_f32:
--    Int = Intrinsic::aarch64_neon_vminnmv;
--    AcrossVec = true; ExtendEle = false; s = "vminnmv"; break;
 -  // Scalar Integer Saturating Doubling Multiply Half High
 -  case AArch64::BI__builtin_neon_vqdmulhh_s16:
 -  case AArch64::BI__builtin_neon_vqdmulhs_s32:
 -    Int = Intrinsic::arm_neon_vqdmulh;
 -    s = "vqdmulh"; OverloadInt = true; break;
 -  // Scalar Integer Saturating Rounding Doubling Multiply Half High
 -  case AArch64::BI__builtin_neon_vqrdmulhh_s16:
 -  case AArch64::BI__builtin_neon_vqrdmulhs_s32:
 -    Int = Intrinsic::arm_neon_vqrdmulh;
 -    s = "vqrdmulh"; OverloadInt = true; break;
 -  // Scalar Floating-point Multiply Extended
 -  case AArch64::BI__builtin_neon_vmulxs_f32:
 -  case AArch64::BI__builtin_neon_vmulxd_f64:
 -    Int = Intrinsic::aarch64_neon_vmulx;
 -    s = "vmulx"; OverloadInt = true; break;
 -  // Scalar Floating-point Reciprocal Step and
 -  case AArch64::BI__builtin_neon_vrecpss_f32:
 -  case AArch64::BI__builtin_neon_vrecpsd_f64:
 -    Int = Intrinsic::arm_neon_vrecps;
 -    s = "vrecps"; OverloadInt = true; break;
 -  // Scalar Floating-point Reciprocal Square Root Step
 -  case AArch64::BI__builtin_neon_vrsqrtss_f32:
 -  case AArch64::BI__builtin_neon_vrsqrtsd_f64:
 -    Int = Intrinsic::arm_neon_vrsqrts;
 -    s = "vrsqrts"; OverloadInt = true; break;
 -  // Scalar Signed Integer Convert To Floating-point
 -  case AArch64::BI__builtin_neon_vcvts_f32_s32:
 -    Int = Intrinsic::aarch64_neon_vcvtf32_s32,
 -    s = "vcvtf"; OverloadInt = false; break;
 -  case AArch64::BI__builtin_neon_vcvtd_f64_s64:
 -    Int = Intrinsic::aarch64_neon_vcvtf64_s64,
 -    s = "vcvtf"; OverloadInt = false; break;
 -  // Scalar Unsigned Integer Convert To Floating-point
 -  case AArch64::BI__builtin_neon_vcvts_f32_u32:
 -    Int = Intrinsic::aarch64_neon_vcvtf32_u32,
 -    s = "vcvtf"; OverloadInt = false; break;
 -  case AArch64::BI__builtin_neon_vcvtd_f64_u64:
 -    Int = Intrinsic::aarch64_neon_vcvtf64_u64,
 -    s = "vcvtf"; OverloadInt = false; break;
 -  // Scalar Floating-point Reciprocal Estimate
 -  case AArch64::BI__builtin_neon_vrecpes_f32:
 -  case AArch64::BI__builtin_neon_vrecped_f64:
 -    Int = Intrinsic::arm_neon_vrecpe;
 -    s = "vrecpe"; OverloadInt = true; break;
 -  // Scalar Floating-point Reciprocal Exponent
 -  case AArch64::BI__builtin_neon_vrecpxs_f32:
 -  case AArch64::BI__builtin_neon_vrecpxd_f64:
 -    Int = Intrinsic::aarch64_neon_vrecpx;
 -    s = "vrecpx"; OverloadInt = true; break;
 -  // Scalar Floating-point Reciprocal Square Root Estimate
 -  case AArch64::BI__builtin_neon_vrsqrtes_f32:
 -  case AArch64::BI__builtin_neon_vrsqrted_f64:
 -    Int = Intrinsic::arm_neon_vrsqrte;
 -    s = "vrsqrte"; OverloadInt = true; break;
 -  // Scalar Compare Equal
 -  case AArch64::BI__builtin_neon_vceqd_s64:
 -  case AArch64::BI__builtin_neon_vceqd_u64:
 -    Int = Intrinsic::aarch64_neon_vceq; s = "vceq";
 -    OverloadInt = false; break;
 -  // Scalar Compare Equal To Zero
 -  case AArch64::BI__builtin_neon_vceqzd_s64:
 -  case AArch64::BI__builtin_neon_vceqzd_u64:
 -    Int = Intrinsic::aarch64_neon_vceq; s = "vceq";
 -    // Add implicit zero operand.
 -    Ops.push_back(llvm::Constant::getNullValue(Ops[0]->getType()));
 -    OverloadInt = false; break;
 -  // Scalar Compare Greater Than or Equal
 -  case AArch64::BI__builtin_neon_vcged_s64:
 -    Int = Intrinsic::aarch64_neon_vcge; s = "vcge";
 -    OverloadInt = false; break;
 -  case AArch64::BI__builtin_neon_vcged_u64:
 -    Int = Intrinsic::aarch64_neon_vchs; s = "vcge";
 -    OverloadInt = false; break;
 -  // Scalar Compare Greater Than or Equal To Zero
 -  case AArch64::BI__builtin_neon_vcgezd_s64:
 -    Int = Intrinsic::aarch64_neon_vcge; s = "vcge";
 -    // Add implicit zero operand.
 -    Ops.push_back(llvm::Constant::getNullValue(Ops[0]->getType()));
 -    OverloadInt = false; break;
 -  // Scalar Compare Greater Than
 -  case AArch64::BI__builtin_neon_vcgtd_s64:
 -    Int = Intrinsic::aarch64_neon_vcgt; s = "vcgt";
 -    OverloadInt = false; break;
 -  case AArch64::BI__builtin_neon_vcgtd_u64:
 -    Int = Intrinsic::aarch64_neon_vchi; s = "vcgt";
 -    OverloadInt = false; break;
 -  // Scalar Compare Greater Than Zero
 -  case AArch64::BI__builtin_neon_vcgtzd_s64:
 -    Int = Intrinsic::aarch64_neon_vcgt; s = "vcgt";
 -    // Add implicit zero operand.
 -    Ops.push_back(llvm::Constant::getNullValue(Ops[0]->getType()));
 -    OverloadInt = false; break;
 -  // Scalar Compare Less Than or Equal
 -  case AArch64::BI__builtin_neon_vcled_s64:
 -    Int = Intrinsic::aarch64_neon_vcge; s = "vcge";
 -    OverloadInt = false; std::swap(Ops[0], Ops[1]); break;
 -  case AArch64::BI__builtin_neon_vcled_u64:
 -    Int = Intrinsic::aarch64_neon_vchs; s = "vchs";
 -    OverloadInt = false; std::swap(Ops[0], Ops[1]); break;
 -  // Scalar Compare Less Than or Equal To Zero
 -  case AArch64::BI__builtin_neon_vclezd_s64:
 -    Int = Intrinsic::aarch64_neon_vclez; s = "vcle";
 -    // Add implicit zero operand.
 -    Ops.push_back(llvm::Constant::getNullValue(Ops[0]->getType()));
 -    OverloadInt = false; break;
 -  // Scalar Compare Less Than
 -  case AArch64::BI__builtin_neon_vcltd_s64:
 -    Int = Intrinsic::aarch64_neon_vcgt; s = "vcgt";
 -    OverloadInt = false; std::swap(Ops[0], Ops[1]); break;
 -  case AArch64::BI__builtin_neon_vcltd_u64:
 -    Int = Intrinsic::aarch64_neon_vchi; s = "vchi";
 -    OverloadInt = false; std::swap(Ops[0], Ops[1]); break;
 -  // Scalar Compare Less Than Zero
 -  case AArch64::BI__builtin_neon_vcltzd_s64:
 -    Int = Intrinsic::aarch64_neon_vcltz; s = "vclt";
 -    // Add implicit zero operand.
 -    Ops.push_back(llvm::Constant::getNullValue(Ops[0]->getType()));
 -    OverloadInt = false; break;
 -  // Scalar Compare Bitwise Test Bits
 -  case AArch64::BI__builtin_neon_vtstd_s64:
 -  case AArch64::BI__builtin_neon_vtstd_u64:
 -    Int = Intrinsic::aarch64_neon_vtstd; s = "vtst";
 -    OverloadInt = false; break;
 -  // Scalar Absolute Value
 -  case AArch64::BI__builtin_neon_vabsd_s64:
 -    Int = Intrinsic::aarch64_neon_vabs;
 -    s = "vabs"; OverloadInt = false; break;
 -  // Scalar Signed Saturating Absolute Value
 -  case AArch64::BI__builtin_neon_vqabsb_s8:
 -  case AArch64::BI__builtin_neon_vqabsh_s16:
 -  case AArch64::BI__builtin_neon_vqabss_s32:
 -  case AArch64::BI__builtin_neon_vqabsd_s64:
 -    Int = Intrinsic::arm_neon_vqabs;
 -    s = "vqabs"; OverloadInt = true; break;
 -  // Scalar Negate
 -  case AArch64::BI__builtin_neon_vnegd_s64:
 -    Int = Intrinsic::aarch64_neon_vneg;
 -    s = "vneg"; OverloadInt = false; break;
 -  // Scalar Signed Saturating Negate
 -  case AArch64::BI__builtin_neon_vqnegb_s8:
 -  case AArch64::BI__builtin_neon_vqnegh_s16:
 -  case AArch64::BI__builtin_neon_vqnegs_s32:
 -  case AArch64::BI__builtin_neon_vqnegd_s64:
 -    Int = Intrinsic::arm_neon_vqneg;
 -    s = "vqneg"; OverloadInt = true; break;
 -  // Scalar Signed Saturating Accumulated of Unsigned Value
 -  case AArch64::BI__builtin_neon_vuqaddb_s8:
 -  case AArch64::BI__builtin_neon_vuqaddh_s16:
 -  case AArch64::BI__builtin_neon_vuqadds_s32:
 -  case AArch64::BI__builtin_neon_vuqaddd_s64:
 -    Int = Intrinsic::aarch64_neon_vuqadd;
 -    s = "vuqadd"; OverloadInt = true; break;
 -  // Scalar Unsigned Saturating Accumulated of Signed Value
 -  case AArch64::BI__builtin_neon_vsqaddb_u8:
 -  case AArch64::BI__builtin_neon_vsqaddh_u16:
 -  case AArch64::BI__builtin_neon_vsqadds_u32:
 -  case AArch64::BI__builtin_neon_vsqaddd_u64:
 -    Int = Intrinsic::aarch64_neon_vsqadd;
 -    s = "vsqadd"; OverloadInt = true; break;
 -  #endif
    }
  
    if (!Int)

LLVM patch:
diff --git a/lib/Target/OR1K/OR1KTargetMachine.cpp b/lib/Target/OR1K/OR1KTargetMachine.cpp
index 8f260f0..1771bce 100644
--- a/lib/Target/OR1K/OR1KTargetMachine.cpp
+++ b/lib/Target/OR1K/OR1KTargetMachine.cpp
@@ -41,6 +41,7 @@ OR1KTargetMachine(const Target &T, StringRef TT,
              "f64:32:32-v64:32:32-v128:32:32-a0:0:32-n32"),
   InstrInfo(), TLInfo(*this), TSInfo(*this),
   FrameLowering(Subtarget) {
+    initAsmInfo();
 }
 namespace {
 /// OR1K Code Generator Pass Configuration Options.

In llvmpy, edit core.py and replace line 177:

if llvm.version <= (3, 3):
        OPCODE_ICMP           = 45
        OPCODE_FCMP           = 46
...

with:
if True:
        OPCODE_ICMP           = 45
        OPCODE_FCMP           = 46
...