[AArch64] Implement FP16FML intrinsics

[bryanpkc]

Summary

Generate the FP16FML intrinsics into arm_neon.h (AArch64 only for now).
Add two new type modifiers to NeonEmitter to handle the new prototypes.
Define __ARM_FEATURE_FP16FML when +fp16fml is enabled and guard the
intrinsics with the macro in arm_neon.h.

Based on a patch by Gao Yiling.

[t.p.northover]

I think this is reasonable.

[bryanpkc]

In D53633#1274621, @t.p.northover wrote:

I think this is reasonable.

Thanks Tim. Could you also review D53632, which is the LLVM part of this implementation?

[ab]

cfe/trunk/test/CodeGen/aarch64-neon-fp16fml.c

12

Hey folks, I'm curious: where does the "_u32" suffix come from? Should it be _f16?

Also, are there any new ACLE/intrinsic list documents? As far as I can tell there hasn't been any release since IHI0073B/IHI0053D.

[SjoerdMeijer]

cfe/trunk/test/CodeGen/aarch64-neon-fp16fml.c

12

Also, are there any new ACLE/intrinsic list documents? As far as I can tell there hasn't been any release since IHI0073B/IHI0053D.

I've checked, and an updated ACLE that includes these FP16FML intrinsics is coming soon.

where does the "_u32" suffix come from? Should it be _f16?

Good question. It could probably be _f32 or _f16, but _u32 doesn't seem to make much sense. Looks like the spec says _u32, and that's also what GCC has implemented. I think we want to update the spec and fix the name before the updated spec is available. Will chase this, and let you know once I know more.

[SjoerdMeijer]

cfe/trunk/test/CodeGen/aarch64-neon-fp16fml.c

12

An update on this: we should change this to _f32 (because the first suffixes were refering to the ouput type). The ACLE will be updated accordingly, and also GCC will change its current implementation (from _u32 to _f32). Many thanks for raising this issue.
Is there a volunteer to prepare a patch? Or do you have one already? :-) I could look at it, but that will be towards the end of next week.

[ab]

cfe/trunk/test/CodeGen/aarch64-neon-fp16fml.c

12

I've checked, and an updated ACLE that includes these FP16FML intrinsics is coming soon.

Great, thanks!

An update on this: we should change this to _f32 (because the first suffixes were refering to the ouput type).

Hmm, I was thinking _f16 based on the vmlal intrinsics: they seem to be named after the multiplication type rather than that of the accumulator/output.

Either way seems fine to me though, I'll defer to you folks.

The ACLE will be updated accordingly, and also GCC will change its current implementation (from _u32 to _f32). Many thanks for raising this issue.

Is there a volunteer to prepare a patch? Or do you have one already? :-) I could look at it, but that will be towards the end of next week.

Sure: D58306 (with _f16 though, let me know what you think of vmlal)

Thanks for checking!

[SjoerdMeijer]

FYI: a new ACLE version has been published, please find it here: https://developer.arm.com/architectures/system-architectures/software-standards/acle

The "Neon Intrinsics" section contains these new intrinsics.

cfe/trunk/include/clang/Basic/arm_neon.td

 
 def OP_DOT_LN
     : Op<(call "vdot", $p0, $p1,
           (bitcast $p1, (splat(bitcast "uint32x2_t", $p2), $p3)))>;
 def OP_DOT_LNQ
     : Op<(call "vdot", $p0, $p1,
           (bitcast $p1, (splat(bitcast "uint32x4_t", $p2), $p3)))>;
 
+def OP_FMLAL_LN     : Op<(call "vfmlal_low", $p0, $p1,
+                           (dup_typed $p1, (call "vget_lane", $p2, $p3)))>;
+def OP_FMLSL_LN     : Op<(call "vfmlsl_low", $p0, $p1,
+                           (dup_typed $p1, (call "vget_lane", $p2, $p3)))>;
+def OP_FMLAL_LN_Hi  : Op<(call "vfmlal_high", $p0, $p1,
+                           (dup_typed $p1, (call "vget_lane", $p2, $p3)))>;
+def OP_FMLSL_LN_Hi  : Op<(call "vfmlsl_high", $p0, $p1,
+                           (dup_typed $p1, (call "vget_lane", $p2, $p3)))>;
+
 //===----------------------------------------------------------------------===//
 // Instructions
 //===----------------------------------------------------------------------===//
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.1 Addition
 def VADD    : IOpInst<"vadd", "ddd",
                       "csilfUcUsUiUlQcQsQiQlQfQUcQUsQUiQUl", OP_ADD>;

 let ArchGuard = "defined(__ARM_FEATURE_DOTPROD)" in {
   def DOT : SInst<"vdot", "dd88", "iQiUiQUi">;
   def DOT_LANE : SOpInst<"vdot_lane", "dd87i", "iUiQiQUi", OP_DOT_LN>;
 }
 let ArchGuard = "defined(__ARM_FEATURE_DOTPROD) && defined(__aarch64__)" in {
   // Variants indexing into a 128-bit vector are A64 only.
   def UDOT_LANEQ : SOpInst<"vdot_laneq", "dd89i", "iUiQiQUi", OP_DOT_LNQ>;
 }
+
+// v8.2-A FP16 fused multiply-add long instructions.
+let ArchGuard = "defined(__ARM_FEATURE_FP16FML) && defined(__aarch64__)" in {
+  def VFMLAL_LOW  : SInst<"vfmlal_low", "ffHH", "UiQUi">;
+  def VFMLSL_LOW  : SInst<"vfmlsl_low", "ffHH", "UiQUi">;
+  def VFMLAL_HIGH : SInst<"vfmlal_high", "ffHH", "UiQUi">;
+  def VFMLSL_HIGH : SInst<"vfmlsl_high", "ffHH", "UiQUi">;
+
+  def VFMLAL_LANE_LOW  : SOpInst<"vfmlal_lane_low", "ffH0i", "UiQUi", OP_FMLAL_LN>;
+  def VFMLSL_LANE_LOW  : SOpInst<"vfmlsl_lane_low", "ffH0i", "UiQUi", OP_FMLSL_LN>;
+  def VFMLAL_LANE_HIGH : SOpInst<"vfmlal_lane_high", "ffH0i", "UiQUi", OP_FMLAL_LN_Hi>;
+  def VFMLSL_LANE_HIGH : SOpInst<"vfmlsl_lane_high", "ffH0i", "UiQUi", OP_FMLSL_LN_Hi>;
+
+  def VFMLAL_LANEQ_LOW  : SOpInst<"vfmlal_laneq_low", "ffH1i", "UiQUi", OP_FMLAL_LN>;
+  def VFMLSL_LANEQ_LOW  : SOpInst<"vfmlsl_laneq_low", "ffH1i", "UiQUi", OP_FMLSL_LN>;
+  def VFMLAL_LANEQ_HIGH : SOpInst<"vfmlal_laneq_high", "ffH1i", "UiQUi", OP_FMLAL_LN_Hi>;
+  def VFMLSL_LANEQ_HIGH : SOpInst<"vfmlsl_laneq_high", "ffH1i", "UiQUi", OP_FMLSL_LN_Hi>;
+}

cfe/trunk/include/clang/Basic/arm_neon_incl.td

 //           The VAL argument is saved to a temporary so it can be used
 //           as an l-value.
 def bitcast;
 // dup - Take a scalar argument and create a vector by duplicating it into
 //       all lanes. The type of the vector is the base type of the intrinsic.
 // example: (dup $p1) -> "(uint32x2_t) {__p1, __p1}" (assuming the base type
 //          is uint32x2_t).
 def dup;
+// dup_typed - Take a vector and a scalar argument, and create a new vector of
+//             the same type by duplicating the scalar value into all lanes.
+// example: (dup_typed $p1, $p2) -> "(float16x4_t) {__p2, __p2, __p2, __p2}"
+//          (assuming __p1 is float16x4_t, and __p2 is a compatible scalar).
+def dup_typed;
 // splat - Take a vector and a lane index, and return a vector of the same type
 //         containing repeated instances of the source vector at the lane index.
 // example: (splat $p0, $p1) ->
 //            "__builtin_shufflevector(__p0, __p0, __p1, __p1, __p1, __p1)"
 //          (assuming __p0 has four elements).
 def splat;
 // save_temp - Create a temporary (local) variable. The variable takes a name
 //             based on the zero'th parameter and can be referenced using

 //
 // v: void
 // t: best-fit integer (int/poly args)
 // x: signed integer   (int/float args)
 // u: unsigned integer (int/float args)
 // f: float (int args)
 // F: double (int args)
 // H: half (int args)
+// 0: half (int args), ignore 'Q' size modifier.
+// 1: half (int args), force 'Q' size modifier.
 // d: default
 // g: default, ignore 'Q' size modifier.
 // j: default, force 'Q' size modifier.
 // w: double width elements, same num elts
 // n: double width elements, half num elts
 // h: half width elements, double num elts
 // q: half width elements, quad num elts
 // e: half width elements, double num elts, unsigned

cfe/trunk/lib/Basic/Targets/AArch64.h

class LLVM_LIBRARY_VISIBILITY AArch64TargetInfo : public TargetInfo {
   enum FPUModeEnum { FPUMode, NeonMode = (1 << 0), SveMode = (1 << 1) };
 
   unsigned FPU;
   unsigned CRC;
   unsigned Crypto;
   unsigned Unaligned;
   unsigned HasFullFP16;
   unsigned HasDotProd;
+  unsigned HasFP16FML;
   llvm::AArch64::ArchKind ArchKind;
 
   static const Builtin::Info BuiltinInfo[];
 
   std::string ABI;
 
 public:
   AArch64TargetInfo(const llvm::Triple &Triple, const TargetOptions &Opts);

cfe/trunk/lib/Basic/Targets/AArch64.cpp

void AArch64TargetInfo::getTargetDefines(const LangOptions &Opts,
   if ((FPU & NeonMode) && HasFullFP16)
     Builder.defineMacro("__ARM_FEATURE_FP16_VECTOR_ARITHMETIC", "1");
   if (HasFullFP16)
    Builder.defineMacro("__ARM_FEATURE_FP16_SCALAR_ARITHMETIC", "1");
 
   if (HasDotProd)
     Builder.defineMacro("__ARM_FEATURE_DOTPROD", "1");
 
+  if ((FPU & NeonMode) && HasFP16FML)
+    Builder.defineMacro("__ARM_FEATURE_FP16FML", "1");
+
   switch (ArchKind) {
   default:
     break;
   case llvm::AArch64::ArchKind::ARMV8_1A:
     getTargetDefinesARMV81A(Opts, Builder);
     break;
   case llvm::AArch64::ArchKind::ARMV8_2A:
     getTargetDefinesARMV82A(Opts, Builder);

 bool AArch64TargetInfo::handleTargetFeatures(std::vector<std::string> &Features,
                                              DiagnosticsEngine &Diags) {
   FPU = FPUMode;
   CRC = 0;
   Crypto = 0;
   Unaligned = 1;
   HasFullFP16 = 0;
   HasDotProd = 0;
+  HasFP16FML = 0;
   ArchKind = llvm::AArch64::ArchKind::ARMV8A;
 
   for (const auto &Feature : Features) {
     if (Feature == "+neon")
       FPU |= NeonMode;
     if (Feature == "+sve")
       FPU |= SveMode;
     if (Feature == "+crc")
       CRC = 1;
     if (Feature == "+crypto")
       Crypto = 1;
     if (Feature == "+strict-align")
       Unaligned = 0;
     if (Feature == "+v8.1a")
       ArchKind = llvm::AArch64::ArchKind::ARMV8_1A;
     if (Feature == "+v8.2a")
       ArchKind = llvm::AArch64::ArchKind::ARMV8_2A;
     if (Feature == "+fullfp16")
       HasFullFP16 = 1;
     if (Feature == "+dotprod")
       HasDotProd = 1;
+    if (Feature == "+fp16fml")
+      HasFP16FML = 1;
   }
 
   setDataLayout();
 
   return true;
 }
 
 TargetInfo::CallingConvCheckResult

cfe/trunk/lib/CodeGen/CGBuiltin.cpp

static const NeonIntrinsicInfo AArch64SIMDIntrinsicMap[] = {
   NEONMAP1(vcvtq_n_u64_v, aarch64_neon_vcvtfp2fxu, 0),
   NEONMAP1(vcvtx_f32_v, aarch64_neon_fcvtxn, AddRetType | Add1ArgType),
   NEONMAP2(vdot_v, aarch64_neon_udot, aarch64_neon_sdot, 0),
   NEONMAP2(vdotq_v, aarch64_neon_udot, aarch64_neon_sdot, 0),
   NEONMAP0(vext_v),
   NEONMAP0(vextq_v),
   NEONMAP0(vfma_v),
   NEONMAP0(vfmaq_v),
+  NEONMAP1(vfmlal_high_v, aarch64_neon_fmlal2, 0),
+  NEONMAP1(vfmlal_low_v, aarch64_neon_fmlal, 0),
+  NEONMAP1(vfmlalq_high_v, aarch64_neon_fmlal2, 0),
+  NEONMAP1(vfmlalq_low_v, aarch64_neon_fmlal, 0),
+  NEONMAP1(vfmlsl_high_v, aarch64_neon_fmlsl2, 0),
+  NEONMAP1(vfmlsl_low_v, aarch64_neon_fmlsl, 0),
+  NEONMAP1(vfmlslq_high_v, aarch64_neon_fmlsl2, 0),
+  NEONMAP1(vfmlslq_low_v, aarch64_neon_fmlsl, 0),
   NEONMAP2(vhadd_v, aarch64_neon_uhadd, aarch64_neon_shadd, Add1ArgType | UnsignedAlts),
   NEONMAP2(vhaddq_v, aarch64_neon_uhadd, aarch64_neon_shadd, Add1ArgType | UnsignedAlts),
   NEONMAP2(vhsub_v, aarch64_neon_uhsub, aarch64_neon_shsub, Add1ArgType | UnsignedAlts),
   NEONMAP2(vhsubq_v, aarch64_neon_uhsub, aarch64_neon_shsub, Add1ArgType | UnsignedAlts),
   NEONMAP1(vld1_x2_v, aarch64_neon_ld1x2, 0),
   NEONMAP1(vld1_x3_v, aarch64_neon_ld1x3, 0),
   NEONMAP1(vld1_x4_v, aarch64_neon_ld1x4, 0),
   NEONMAP1(vld1q_x2_v, aarch64_neon_ld1x2, 0),
Value *CodeGenFunction::EmitCommonNeonBuiltinExpr(
   case NEON::BI__builtin_neon_vdot_v:
   case NEON::BI__builtin_neon_vdotq_v: {
     llvm::Type *InputTy =
         llvm::VectorType::get(Int8Ty, Ty->getPrimitiveSizeInBits() / 8);
     llvm::Type *Tys[2] = { Ty, InputTy };
     Int = Usgn ? LLVMIntrinsic : AltLLVMIntrinsic;
     return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vdot");
   }
+  case NEON::BI__builtin_neon_vfmlal_low_v:
+  case NEON::BI__builtin_neon_vfmlalq_low_v: {
+    llvm::Type *InputTy =
+        llvm::VectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16);
+    llvm::Type *Tys[2] = { Ty, InputTy };
+    return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlal_low");
+  }
+  case NEON::BI__builtin_neon_vfmlsl_low_v:
+  case NEON::BI__builtin_neon_vfmlslq_low_v: {
+    llvm::Type *InputTy =
+        llvm::VectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16);
+    llvm::Type *Tys[2] = { Ty, InputTy };
+    return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlsl_low");
+  }
+  case NEON::BI__builtin_neon_vfmlal_high_v:
+  case NEON::BI__builtin_neon_vfmlalq_high_v: {
+    llvm::Type *InputTy =
+           llvm::VectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16);
+    llvm::Type *Tys[2] = { Ty, InputTy };
+    return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlal_high");
+  }
+  case NEON::BI__builtin_neon_vfmlsl_high_v:
+  case NEON::BI__builtin_neon_vfmlslq_high_v: {
+    llvm::Type *InputTy =
+           llvm::VectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16);
+    llvm::Type *Tys[2] = { Ty, InputTy };
+    return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlsl_high");
+  }
   }
 
   assert(Int && "Expected valid intrinsic number");
 
   // Determine the type(s) of this overloaded AArch64 intrinsic.
   Function *F = LookupNeonLLVMIntrinsic(Int, Modifier, Ty, E);
 
   Value *Result = EmitNeonCall(F, Ops, NameHint);

cfe/trunk/test/CodeGen/aarch64-neon-fp16fml.c

+// RUN: %clang_cc1 -triple arm64-none-linux-gnu -target-feature +v8.2a -target-feature +neon -target-feature +fp16fml \
+// RUN: -fallow-half-arguments-and-returns -disable-O0-optnone -emit-llvm -o - %s | opt -S -instcombine | FileCheck %s
+
+// REQUIRES: aarch64-registered-target
+
+// Test AArch64 Armv8.2-A FP16 Fused Multiply-Add Long intrinsics
+
+#include <arm_neon.h>
+
+// Vector form
+
+float32x2_t test_vfmlal_low_u32(float32x2_t a, float16x4_t b, float16x4_t c) {
+// CHECK-LABEL: define <2 x float> @test_vfmlal_low_u32(<2 x float> %a, <4 x half> %b, <4 x half> %c)
+// CHECK: [[RESULT:%.*]] = call <2 x float> @llvm.aarch64.neon.fmlal.v2f32.v4f16(<2 x float> %a, <4 x half> %b, <4 x half> %c)
+// CHECK: ret <2 x float> [[RESULT]]
+  return vfmlal_low_u32(a, b, c);
+}
+
+float32x2_t test_vfmlsl_low_u32(float32x2_t a, float16x4_t b, float16x4_t c) {
+// CHECK-LABEL: define <2 x float> @test_vfmlsl_low_u32(<2 x float> %a, <4 x half> %b, <4 x half> %c)
+// CHECK: [[RESULT:%.*]] = call <2 x float> @llvm.aarch64.neon.fmlsl.v2f32.v4f16(<2 x float> %a, <4 x half> %b, <4 x half> %c)
+// CHECK: ret <2 x float> [[RESULT]]
+  return vfmlsl_low_u32(a, b, c);
+}
+
+float32x2_t test_vfmlal_high_u32(float32x2_t a, float16x4_t b, float16x4_t c) {
+// CHECK-LABEL: define <2 x float> @test_vfmlal_high_u32(<2 x float> %a, <4 x half> %b, <4 x half> %c)
+// CHECK: [[RESULT:%.*]] = call <2 x float> @llvm.aarch64.neon.fmlal2.v2f32.v4f16(<2 x float> %a, <4 x half> %b, <4 x half> %c)
+// CHECK: ret <2 x float> [[RESULT]]
+  return vfmlal_high_u32(a, b, c);
+}
+
+float32x2_t test_vfmlsl_high_u32(float32x2_t a, float16x4_t b, float16x4_t c) {
+// CHECK-LABEL: define <2 x float> @test_vfmlsl_high_u32(<2 x float> %a, <4 x half> %b, <4 x half> %c)
+// CHECK: [[RESULT:%.*]] = call <2 x float> @llvm.aarch64.neon.fmlsl2.v2f32.v4f16(<2 x float> %a, <4 x half> %b, <4 x half> %c)
+// CHECK: ret <2 x float> [[RESULT]]
+  return vfmlsl_high_u32(a, b, c);
+}
+
+float32x4_t test_vfmlalq_low_u32(float32x4_t a, float16x8_t b, float16x8_t c) {
+// CHECK-LABEL: define <4 x float> @test_vfmlalq_low_u32(<4 x float> %a, <8 x half> %b, <8 x half> %c)
+// CHECK: [[RESULT:%.*]] = call <4 x float> @llvm.aarch64.neon.fmlal.v4f32.v8f16(<4 x float> %a, <8 x half> %b, <8 x half> %c)
+// CHECK: ret <4 x float> [[RESULT]]
+  return vfmlalq_low_u32(a, b, c);
+}
+
+float32x4_t test_vfmlslq_low_u32(float32x4_t a, float16x8_t b, float16x8_t c) {
+// CHECK-LABEL: define <4 x float> @test_vfmlslq_low_u32(<4 x float> %a, <8 x half> %b, <8 x half> %c)
+// CHECK: [[RESULT:%.*]] = call <4 x float> @llvm.aarch64.neon.fmlsl.v4f32.v8f16(<4 x float> %a, <8 x half> %b, <8 x half> %c)
+// CHECK: ret <4 x float> [[RESULT]]
+  return vfmlslq_low_u32(a, b, c);
+}
+
+float32x4_t test_vfmlalq_high_u32(float32x4_t a, float16x8_t b, float16x8_t c) {
+// CHECK-LABEL: define <4 x float> @test_vfmlalq_high_u32(<4 x float> %a, <8 x half> %b, <8 x half> %c)
+// CHECK: [[RESULT:%.*]] = call <4 x float> @llvm.aarch64.neon.fmlal2.v4f32.v8f16(<4 x float> %a, <8 x half> %b, <8 x half> %c)
+// CHECK: ret <4 x float> [[RESULT]]
+  return vfmlalq_high_u32(a, b, c);
+}
+
+float32x4_t test_vfmlslq_high_u32(float32x4_t a, float16x8_t b, float16x8_t c) {
+// CHECK-LABEL: define <4 x float> @test_vfmlslq_high_u32(<4 x float> %a, <8 x half> %b, <8 x half> %c)
+// CHECK: [[RESULT:%.*]] = call <4 x float> @llvm.aarch64.neon.fmlsl2.v4f32.v8f16(<4 x float> %a, <8 x half> %b, <8 x half> %c)
+// CHECK: ret <4 x float> [[RESULT]]
+  return vfmlslq_high_u32(a, b, c);
+}
+
+// Indexed form
+
+float32x2_t test_vfmlal_lane_low_u32(float32x2_t a, float16x4_t b, float16x4_t c) {
+// CHECK-LABEL: define <2 x float> @test_vfmlal_lane_low_u32(<2 x float> %a, <4 x half> %b, <4 x half> %c)
+// CHECK: [[SHUFFLE:%.*]] = shufflevector <4 x half> %c, <4 x half> undef, <4 x i32> zeroinitializer
+// CHECK: [[RESULT:%.*]] = call <2 x float> @llvm.aarch64.neon.fmlal.v2f32.v4f16(<2 x float> %a, <4 x half> %b, <4 x half> [[SHUFFLE]])
+// CHECK: ret <2 x float> [[RESULT]]
+  return vfmlal_lane_low_u32(a, b, c, 0);
+}
+
+float32x2_t test_vfmlal_lane_high_u32(float32x2_t a, float16x4_t b, float16x4_t c) {
+// CHECK-LABEL: define <2 x float> @test_vfmlal_lane_high_u32(<2 x float> %a, <4 x half> %b, <4 x half> %c)
+// CHECK: [[SHUFFLE:%.*]] = shufflevector <4 x half> %c, <4 x half> undef, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK: [[RESULT:%.*]] = call <2 x float> @llvm.aarch64.neon.fmlal2.v2f32.v4f16(<2 x float> %a, <4 x half> %b, <4 x half> [[SHUFFLE]])
+// CHECK: ret <2 x float> [[RESULT]]
+  return vfmlal_lane_high_u32(a, b, c, 1);
+}
+
+float32x4_t test_vfmlalq_lane_low_u32(float32x4_t a, float16x8_t b, float16x4_t c) {
+// CHECK-LABEL: define <4 x float> @test_vfmlalq_lane_low_u32(<4 x float> %a, <8 x half> %b, <4 x half> %c)
+// CHECK: [[SHUFFLE:%.*]] = shufflevector <4 x half> %c, <4 x half> undef, <8 x i32> <i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2>
+// CHECK: [[RESULT:%.*]] = call <4 x float> @llvm.aarch64.neon.fmlal.v4f32.v8f16(<4 x float> %a, <8 x half> %b, <8 x half> [[SHUFFLE]])
+// CHECK: ret <4 x float> [[RESULT]]
+  return vfmlalq_lane_low_u32(a, b, c, 2);
+}
+
+float32x4_t test_vfmlalq_lane_high_u32(float32x4_t a, float16x8_t b, float16x4_t c) {
+// CHECK-LABEL: define <4 x float> @test_vfmlalq_lane_high_u32(<4 x float> %a, <8 x half> %b, <4 x half> %c)
+// CHECK: [[SHUFFLE:%.*]] = shufflevector <4 x half> %c, <4 x half> undef, <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK: [[RESULT:%.*]] = call <4 x float> @llvm.aarch64.neon.fmlal2.v4f32.v8f16(<4 x float> %a, <8 x half> %b, <8 x half> [[SHUFFLE]])
+// CHECK: ret <4 x float> [[RESULT]]
+  return vfmlalq_lane_high_u32(a, b, c, 3);
+}
+
+float32x2_t test_vfmlal_laneq_low_u32(float32x2_t a, float16x4_t b, float16x8_t c) {
+// CHECK-LABEL: define <2 x float> @test_vfmlal_laneq_low_u32(<2 x float> %a, <4 x half> %b, <8 x half> %c)
+// CHECK: [[SHUFFLE:%.*]] = shufflevector <8 x half> %c, <8 x half> undef, <4 x i32> <i32 4, i32 4, i32 4, i32 4>
+// CHECK: [[RESULT:%.*]] = call <2 x float> @llvm.aarch64.neon.fmlal.v2f32.v4f16(<2 x float> %a, <4 x half> %b, <4 x half> [[SHUFFLE]])
+// CHECK: ret <2 x float> [[RESULT]]
+  return vfmlal_laneq_low_u32(a, b, c, 4);
+}
+
+float32x2_t test_vfmlal_laneq_high_u32(float32x2_t a, float16x4_t b, float16x8_t c) {
+// CHECK-LABEL: define <2 x float> @test_vfmlal_laneq_high_u32(<2 x float> %a, <4 x half> %b, <8 x half> %c)
+// CHECK: [[SHUFFLE:%.*]] = shufflevector <8 x half> %c, <8 x half> undef, <4 x i32> <i32 5, i32 5, i32 5, i32 5>
+// CHECK: [[RESULT:%.*]] = call <2 x float> @llvm.aarch64.neon.fmlal2.v2f32.v4f16(<2 x float> %a, <4 x half> %b, <4 x half> [[SHUFFLE]])
+// CHECK: ret <2 x float> [[RESULT]]
+  return vfmlal_laneq_high_u32(a, b, c, 5);
+}
+
+float32x4_t test_vfmlalq_laneq_low_u32(float32x4_t a, float16x8_t b, float16x8_t c) {
+// CHECK-LABEL: define <4 x float> @test_vfmlalq_laneq_low_u32(<4 x float> %a, <8 x half> %b, <8 x half> %c)
+// CHECK: [[SHUFFLE:%.*]] = shufflevector <8 x half> %c, <8 x half> undef, <8 x i32> <i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6>
+// CHECK: [[RESULT:%.*]] = call <4 x float> @llvm.aarch64.neon.fmlal.v4f32.v8f16(<4 x float> %a, <8 x half> %b, <8 x half> [[SHUFFLE]])
+// CHECK: ret <4 x float> [[RESULT]]
+  return vfmlalq_laneq_low_u32(a, b, c, 6);
+}
+
+float32x4_t test_vfmlalq_laneq_high_u32(float32x4_t a, float16x8_t b, float16x8_t c) {
+// CHECK-LABEL: define <4 x float> @test_vfmlalq_laneq_high_u32(<4 x float> %a, <8 x half> %b, <8 x half> %c)
+// CHECK: [[SHUFFLE:%.*]] = shufflevector <8 x half> %c, <8 x half> undef, <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+// CHECK: [[RESULT:%.*]] = call <4 x float> @llvm.aarch64.neon.fmlal2.v4f32.v8f16(<4 x float> %a, <8 x half> %b, <8 x half> [[SHUFFLE]])
+// CHECK: ret <4 x float> [[RESULT]]
+  return vfmlalq_laneq_high_u32(a, b, c, 7);
+}
+
+float32x2_t test_vfmlsl_lane_low_u32(float32x2_t a, float16x4_t b, float16x4_t c) {
+// CHECK-LABEL: define <2 x float> @test_vfmlsl_lane_low_u32(<2 x float> %a, <4 x half> %b, <4 x half> %c)
+// CHECK: [[SHUFFLE:%.*]] = shufflevector <4 x half> %c, <4 x half> undef, <4 x i32> zeroinitializer
+// CHECK: [[RESULT:%.*]] = call <2 x float> @llvm.aarch64.neon.fmlsl.v2f32.v4f16(<2 x float> %a, <4 x half> %b, <4 x half> [[SHUFFLE]])
+// CHECK: ret <2 x float> [[RESULT]]
+  return vfmlsl_lane_low_u32(a, b, c, 0);
+}
+
+float32x2_t test_vfmlsl_lane_high_u32(float32x2_t a, float16x4_t b, float16x4_t c) {
+// CHECK-LABEL: define <2 x float> @test_vfmlsl_lane_high_u32(<2 x float> %a, <4 x half> %b, <4 x half> %c)
+// CHECK: [[SHUFFLE:%.*]] = shufflevector <4 x half> %c, <4 x half> undef, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK: [[RESULT:%.*]] = call <2 x float> @llvm.aarch64.neon.fmlsl2.v2f32.v4f16(<2 x float> %a, <4 x half> %b, <4 x half> [[SHUFFLE]])
+// CHECK: ret <2 x float> [[RESULT]]
+  return vfmlsl_lane_high_u32(a, b, c, 1);
+}
+
+float32x4_t test_vfmlslq_lane_low_u32(float32x4_t a, float16x8_t b, float16x4_t c) {
+// CHECK-LABEL: define <4 x float> @test_vfmlslq_lane_low_u32(<4 x float> %a, <8 x half> %b, <4 x half> %c)
+// CHECK: [[SHUFFLE:%.*]] = shufflevector <4 x half> %c, <4 x half> undef, <8 x i32> <i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2>
+// CHECK: [[RESULT:%.*]] = call <4 x float> @llvm.aarch64.neon.fmlsl.v4f32.v8f16(<4 x float> %a, <8 x half> %b, <8 x half> [[SHUFFLE]])
+// CHECK: ret <4 x float> [[RESULT]]
+  return vfmlslq_lane_low_u32(a, b, c, 2);
+}
+
+float32x4_t test_vfmlslq_lane_high_u32(float32x4_t a, float16x8_t b, float16x4_t c) {
+// CHECK-LABEL: define <4 x float> @test_vfmlslq_lane_high_u32(<4 x float> %a, <8 x half> %b, <4 x half> %c)
+// CHECK: [[SHUFFLE:%.*]] = shufflevector <4 x half> %c, <4 x half> undef, <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK: [[RESULT:%.*]] = call <4 x float> @llvm.aarch64.neon.fmlsl2.v4f32.v8f16(<4 x float> %a, <8 x half> %b, <8 x half> [[SHUFFLE]])
+// CHECK: ret <4 x float> [[RESULT]]
+  return vfmlslq_lane_high_u32(a, b, c, 3);
+}
+
+float32x2_t test_vfmlsl_laneq_low_u32(float32x2_t a, float16x4_t b, float16x8_t c) {
+// CHECK-LABEL: define <2 x float> @test_vfmlsl_laneq_low_u32(<2 x float> %a, <4 x half> %b, <8 x half> %c)
+// CHECK: [[SHUFFLE:%.*]] = shufflevector <8 x half> %c, <8 x half> undef, <4 x i32> <i32 4, i32 4, i32 4, i32 4>
+// CHECK: [[RESULT:%.*]] = call <2 x float> @llvm.aarch64.neon.fmlsl.v2f32.v4f16(<2 x float> %a, <4 x half> %b, <4 x half> [[SHUFFLE]])
+// CHECK: ret <2 x float> [[RESULT]]
+  return vfmlsl_laneq_low_u32(a, b, c, 4);
+}
+
+float32x2_t test_vfmlsl_laneq_high_u32(float32x2_t a, float16x4_t b, float16x8_t c) {
+// CHECK-LABEL: define <2 x float> @test_vfmlsl_laneq_high_u32(<2 x float> %a, <4 x half> %b, <8 x half> %c)
+// CHECK: [[SHUFFLE:%.*]] = shufflevector <8 x half> %c, <8 x half> undef, <4 x i32> <i32 5, i32 5, i32 5, i32 5>
+// CHECK: [[RESULT:%.*]] = call <2 x float> @llvm.aarch64.neon.fmlsl2.v2f32.v4f16(<2 x float> %a, <4 x half> %b, <4 x half> [[SHUFFLE]])
+// CHECK: ret <2 x float> [[RESULT]]
+  return vfmlsl_laneq_high_u32(a, b, c, 5);
+}
+
+float32x4_t test_vfmlslq_laneq_low_u32(float32x4_t a, float16x8_t b, float16x8_t c) {
+// CHECK-LABEL: define <4 x float> @test_vfmlslq_laneq_low_u32(<4 x float> %a, <8 x half> %b, <8 x half> %c)
+// CHECK: [[SHUFFLE:%.*]] = shufflevector <8 x half> %c, <8 x half> undef, <8 x i32> <i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6>
+// CHECK: [[RESULT:%.*]] = call <4 x float> @llvm.aarch64.neon.fmlsl.v4f32.v8f16(<4 x float> %a, <8 x half> %b, <8 x half> [[SHUFFLE]])
+// CHECK: ret <4 x float> [[RESULT]]
+  return vfmlslq_laneq_low_u32(a, b, c, 6);
+}
+
+float32x4_t test_vfmlslq_laneq_high_u32(float32x4_t a, float16x8_t b, float16x8_t c) {
+// CHECK-LABEL: define <4 x float> @test_vfmlslq_laneq_high_u32(<4 x float> %a, <8 x half> %b, <8 x half> %c)
+// CHECK: [[SHUFFLE:%.*]] = shufflevector <8 x half> %c, <8 x half> undef, <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+// CHECK: [[RESULT:%.*]] = call <4 x float> @llvm.aarch64.neon.fmlsl2.v4f32.v8f16(<4 x float> %a, <8 x half> %b, <8 x half> [[SHUFFLE]])
+// CHECK: ret <4 x float> [[RESULT]]
+  return vfmlslq_laneq_high_u32(a, b, c, 7);
+}

cfe/trunk/test/Preprocessor/aarch64-target-features.c

 // RUN: %clang -target aarch64 -mtune=cyclone -### -c %s 2>&1 | FileCheck -check-prefix=CHECK-MTUNE-CYCLONE %s
 
 // RUN: %clang -target aarch64-none-linux-gnu -march=armv8-a+sve -x c -E -dM %s -o - | FileCheck --check-prefix=CHECK-SVE %s
 // CHECK-SVE: __ARM_FEATURE_SVE 1
 
 // RUN: %clang -target aarch64-none-linux-gnu -march=armv8.2a+dotprod -x c -E -dM %s -o - | FileCheck --check-prefix=CHECK-DOTPROD %s
 // CHECK-DOTPROD: __ARM_FEATURE_DOTPROD 1
 
-// RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.2-a+nofp16fml+fp16 -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-VECTOR-SCALAR %s
-// RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.2-a+nofp16+fp16fml -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-VECTOR-SCALAR %s
-// RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.2-a+fp16+nofp16fml -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-VECTOR-SCALAR %s
-// RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8-a+fp16fml -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-VECTOR-SCALAR %s
-// RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8-a+fp16 -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-VECTOR-SCALAR %s
-// RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.4-a+nofp16fml+fp16 -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-VECTOR-SCALAR %s
-// RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.4-a+nofp16+fp16fml -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-VECTOR-SCALAR %s
-// RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.4-a+fp16+nofp16fml -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-VECTOR-SCALAR %s
-// RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.4-a+fp16fml -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-VECTOR-SCALAR %s
-// RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.4-a+fp16 -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-VECTOR-SCALAR %s
+// On ARMv8.2-A and above, +fp16fml implies +fp16.
+// On ARMv8.4-A and above, +fp16 implies +fp16fml.
+// RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.2-a+nofp16fml+fp16 -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-NOFML --check-prefix=CHECK-FULLFP16-VECTOR-SCALAR %s
+// RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.2-a+nofp16+fp16fml -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-FML --check-prefix=CHECK-FULLFP16-VECTOR-SCALAR %s
+// RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.2-a+fp16+nofp16fml -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-NOFML --check-prefix=CHECK-FULLFP16-VECTOR-SCALAR %s
+// RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8-a+fp16fml -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-FML --check-prefix=CHECK-FULLFP16-VECTOR-SCALAR %s
+// RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8-a+fp16 -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-NOFML --check-prefix=CHECK-FULLFP16-VECTOR-SCALAR %s
+// RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.4-a+nofp16fml+fp16 -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-FML --check-prefix=CHECK-FULLFP16-VECTOR-SCALAR %s
+// RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.4-a+nofp16+fp16fml -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-FML --check-prefix=CHECK-FULLFP16-VECTOR-SCALAR %s
+// RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.4-a+fp16+nofp16fml -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-NOFML --check-prefix=CHECK-FULLFP16-VECTOR-SCALAR %s
+// RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.4-a+fp16fml -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-FML --check-prefix=CHECK-FULLFP16-VECTOR-SCALAR %s
+// RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.4-a+fp16 -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-FML --check-prefix=CHECK-FULLFP16-VECTOR-SCALAR %s
+// CHECK-FULLFP16-FML:           #define __ARM_FEATURE_FP16FML 1
+// CHECK-FULLFP16-NOFML-NOT:     #define __ARM_FEATURE_FP16FML 1
 // CHECK-FULLFP16-VECTOR-SCALAR: #define __ARM_FEATURE_FP16_SCALAR_ARITHMETIC 1
 // CHECK-FULLFP16-VECTOR-SCALAR: #define __ARM_FEATURE_FP16_VECTOR_ARITHMETIC 1
 // CHECK-FULLFP16-VECTOR-SCALAR: #define __ARM_FP 0xE
 // CHECK-FULLFP16-VECTOR-SCALAR: #define __ARM_FP16_FORMAT_IEEE 1
 
 // +fp16fml+nosimd doesn't make sense as the fp16fml instructions all require SIMD.
 // However, as +fp16fml implies +fp16 there is a set of defines that we would expect.
 // RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8-a+fp16fml+nosimd -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-SCALAR %s
 // RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8-a+fp16+nosimd -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-SCALAR %s
 // RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.4-a+fp16fml+nosimd -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-SCALAR %s
 // RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.4-a+fp16+nosimd -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-SCALAR %s
+// CHECK-FULLFP16-SCALAR-NOT:   #define __ARM_FEATURE_FP16FML 1
 // CHECK-FULLFP16-SCALAR:       #define __ARM_FEATURE_FP16_SCALAR_ARITHMETIC 1
 // CHECK-FULLFP16-SCALAR-NOT:   #define __ARM_FEATURE_FP16_VECTOR_ARITHMETIC 1
 // CHECK-FULLFP16-SCALAR:       #define __ARM_FP 0xE
 // CHECK-FULLFP16-SCALAR:       #define __ARM_FP16_FORMAT_IEEE 1
 
 // RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.2-a -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-NOFML-VECTOR-SCALAR %s
 // RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.2-a+nofp16 -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-NOFML-VECTOR-SCALAR %s
 // RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.2-a+nofp16fml -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-NOFML-VECTOR-SCALAR %s
 // RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.2-a+fp16fml+nofp16 -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-NOFML-VECTOR-SCALAR %s
 // RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.4-a -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-NOFML-VECTOR-SCALAR %s
 // RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.4-a+nofp16 -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-NOFML-VECTOR-SCALAR %s
 // RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.4-a+nofp16fml -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-NOFML-VECTOR-SCALAR %s
 // RUN: %clang -target aarch64-none-linux-gnueabi -march=armv8.4-a+fp16fml+nofp16 -x c -E -dM %s -o - | FileCheck -match-full-lines --check-prefix=CHECK-FULLFP16-NOFML-VECTOR-SCALAR %s
+// CHECK-FULLFP16-NOFML-VECTOR-SCALAR-NOT: #define __ARM_FEATURE_FP16FML 1
 // CHECK-FULLFP16-NOFML-VECTOR-SCALAR-NOT: #define __ARM_FEATURE_FP16_SCALAR_ARITHMETIC 1
 // CHECK-FULLFP16-NOFML-VECTOR-SCALAR-NOT: #define __ARM_FEATURE_FP16_VECTOR_ARITHMETIC 1
-// CHECK-FULLFP16-NOFML-VECTOR-SCALAR: #define __ARM_FP 0xE
-// CHECK-FULLFP16-NOFML-VECTOR-SCALAR: #define __ARM_FP16_FORMAT_IEEE 1
+// CHECK-FULLFP16-NOFML-VECTOR-SCALAR:     #define __ARM_FP 0xE
+// CHECK-FULLFP16-NOFML-VECTOR-SCALAR:     #define __ARM_FP16_FORMAT_IEEE 1
 
 // ================== Check whether -mtune accepts mixed-case features.
 // RUN: %clang -target aarch64 -mtune=CYCLONE -### -c %s 2>&1 | FileCheck -check-prefix=CHECK-MTUNE-CYCLONE %s
 // CHECK-MTUNE-CYCLONE: "-cc1"{{.*}} "-triple" "aarch64{{.*}}" "-target-feature" "+neon" "-target-feature" "+zcm" "-target-feature" "+zcz"
 
 // RUN: %clang -target aarch64 -mcpu=cyclone -### -c %s 2>&1 | FileCheck -check-prefix=CHECK-MCPU-CYCLONE %s
 // RUN: %clang -target aarch64 -mcpu=cortex-a35 -### -c %s 2>&1 | FileCheck -check-prefix=CHECK-MCPU-A35 %s
 // RUN: %clang -target aarch64 -mcpu=cortex-a53 -### -c %s 2>&1 | FileCheck -check-prefix=CHECK-MCPU-A53 %s

cfe/trunk/utils/TableGen/NeonEmitter.cpp

private:
   public:
     DagEmitter(Intrinsic &Intr, StringRef CallPrefix) :
       Intr(Intr), CallPrefix(CallPrefix) {
     }
     std::pair<Type, std::string> emitDagArg(Init *Arg, std::string ArgName);
     std::pair<Type, std::string> emitDagSaveTemp(DagInit *DI);
     std::pair<Type, std::string> emitDagSplat(DagInit *DI);
     std::pair<Type, std::string> emitDagDup(DagInit *DI);
+    std::pair<Type, std::string> emitDagDupTyped(DagInit *DI);
     std::pair<Type, std::string> emitDagShuffle(DagInit *DI);
     std::pair<Type, std::string> emitDagCast(DagInit *DI, bool IsBitCast);
     std::pair<Type, std::string> emitDagCall(DagInit *DI);
     std::pair<Type, std::string> emitDagNameReplace(DagInit *DI);
     std::pair<Type, std::string> emitDagLiteral(DagInit *DI);
     std::pair<Type, std::string> emitDagOp(DagInit *DI);
     std::pair<Type, std::string> emitDag(DagInit *DI);
   };
void Type::applyModifier(char Mod) {
   case 'F':
     Float = true;
     ElementBitwidth = 64;
     break;
   case 'H':
     Float = true;
     ElementBitwidth = 16;
     break;
+  case '0':
+    Float = true;
+    if (AppliedQuad)
+      Bitwidth /= 2;
+    ElementBitwidth = 16;
+    break;
+  case '1':
+    Float = true;
+    if (!AppliedQuad)
+      Bitwidth *= 2;
+    ElementBitwidth = 16;
+    break;
   case 'g':
     if (AppliedQuad)
       Bitwidth /= 2;
     break;
   case 'j':
     if (!AppliedQuad)
       Bitwidth *= 2;
     break;
std::pair<Type, std::string> Intrinsic::DagEmitter::emitDag(DagInit *DI) {
   std::string Op = DefI->getAsString();
 
   if (Op == "cast" || Op == "bitcast")
     return emitDagCast(DI, Op == "bitcast");
   if (Op == "shuffle")
     return emitDagShuffle(DI);
   if (Op == "dup")
     return emitDagDup(DI);
+  if (Op == "dup_typed")
+    return emitDagDupTyped(DI);
   if (Op == "splat")
     return emitDagSplat(DI);
   if (Op == "save_temp")
     return emitDagSaveTemp(DI);
   if (Op == "op")
     return emitDagOp(DI);
   if (Op == "call")
     return emitDagCall(DI);
if (I != 0)
       S += ", ";
     S += A.second;
   }
   S += "}";
 
   return std::make_pair(T, S);
 }
 
+std::pair<Type, std::string> Intrinsic::DagEmitter::emitDagDupTyped(DagInit *DI) {
+  assert_with_loc(DI->getNumArgs() == 2, "dup_typed() expects two arguments");
+  std::pair<Type, std::string> A = emitDagArg(DI->getArg(0),
+                                              DI->getArgNameStr(0));
+  std::pair<Type, std::string> B = emitDagArg(DI->getArg(1),
+                                              DI->getArgNameStr(1));
+  assert_with_loc(B.first.isScalar(),
+                  "dup_typed() requires a scalar as the second argument");
+
+  Type T = A.first;
+  assert_with_loc(T.isVector(), "dup_typed() used but target type is scalar!");
+  std::string S = "(" + T.str() + ") {";
+  for (unsigned I = 0; I < T.getNumElements(); ++I) {
+    if (I != 0)
+      S += ", ";
+    S += B.second;
+  }
+  S += "}";
+
+  return std::make_pair(T, S);
+}
+
 std::pair<Type, std::string> Intrinsic::DagEmitter::emitDagSplat(DagInit *DI) {
   assert_with_loc(DI->getNumArgs() == 2, "splat() expects two arguments");
   std::pair<Type, std::string> A = emitDagArg(DI->getArg(0),
                                               DI->getArgNameStr(0));
   std::pair<Type, std::string> B = emitDagArg(DI->getArg(1),
                                               DI->getArgNameStr(1));
 
   assert_with_loc(B.first.isScalar(),