NeonEmitter: clean up prototype modifiers

[t.p.northover]

Summary

What does the modifier 'j' do on a vector type in arm_neon.td? I don't know; I don't think I ever knew; and I wouldn't be surprised if no-one knows.

So when implementing new intrinsics everyone just has to keep the magic table of modifiers open in a side-panel. Additionally, we're rapidly running out of letters that we can add when intrinsics with new and strange prototypes come along (I'm looking at you, dot product!). So unless we adopt the mathematicians' trick of starting on Greek and Hebrew alphabets once the Latin letters run out (is TableGen UTF-8 safe?), a solution is needed.

So, the main point of this patch is to allow multiple modifiers per type, and strip the list of modifiers down to something much more orthogonal, and hopefully intuitive. Some things I had to make a choice on to write the patch, and might warrant debate:

• Using '1' for scalar is inconsistent with '2, '3', '4', but '0' seems weird to me.
• Parens to group is a different system from the typespec it's next to (where modifiers precede a canonical base type).
• Switching default from 'd' to '.' might be egregious, but I prefer it and we're rewriting the whole file anyway.

I've included a script which I used to automatically port our .td files, and should be equally useful for any downstream users with changes to these files (I suggest running it on the last downstream version before this change hit).

[efriedma]

It looks like this patch contains a few other changes, besides the changes to the prototypes. In particular, the change to CGBuiltin.cpp, and there are a few new lines in the .td files that don't correspond to anything in the old versions. Is that accidental, or is it part of cleaning up the prototypes, somehow?

[t.p.northover]

It looks like this patch contains a few other changes, besides the changes to the prototypes. In particular, the change to CGBuiltin.cpp, and there are a few new lines in the .td files that don't correspond to anything in the old versions. Is that accidental, or is it part of cleaning up the prototypes, somehow?

The extra .td lines are because just those 3 intrinsics used a fixed-width modifier ("give me half, no matter the input") with multiple sizes of input so there's no way to represent that in the new scheme and they need to be split up. Notice the integer ones are already split up because there was no corrresponding "give me int32_t" modifier. That change is actually already a separate NFC commit in my local repository and I'd commit it like that so that the script worked cleanly.

The CGBuiltin change follows from dropping the heuristic hasFloatingProtoModifier when deciding what type to pass to CGBuiltin for the intrinsics. This affected vmulx and the vcvt intrinsics. In vcvt's case I eventually decided to support them by moving to an explicit '!' modifier and special-casing conversion because they make good use of having signedness on the type they're given. I didn't revisit vmulx after that change, but I'd be inclined to leave it as it is; I kind of think it's unlikely someone implementing that now would make use of the ! modifier, which seems like a pretty rare requirement.

There are two other things that I think are pretty straightforward, but do clutter this patch so I'll split them out: removing the special behaviour of 'a' (it can be implemented in .td at a net -ve lines); and changing Type to use an enum instead of a series of bools. I'll upload new diffs and update this one.

The other

[t.p.northover]

Separated off the two features I mentioned as D69715 and D69716.

[t.p.northover]

Ping.

[efriedma]

LGTM with a couple nits.

clang/include/clang/Basic/arm_neon_incl.td

203

'd' is gone.

clang/utils/convert_arm_neon.py

1

Are you going to commit this script? If you are, probably makes sense to include some sort of date, so it's clear which change you're talking about, and when it makes sense to remove it from the tree.

[t.p.northover]

Thanks. Pushed it with those suggestions:

To github.com:llvm/llvm-project.git

c34478f5f6c7..3f91705ca54b  master -> master

[efriedma]

It looks like this broke vcreate_u16 and friends. From http://lab.llvm.org:8011/builders/aosp-O3-polly-before-vectorizer-unprofitable/builds/1224/steps/build-aosp/logs/stdio :

external/skia/src/opts/SkBitmapProcState_filter_neon.h:53:42: error: C-style cast from scalar 'int' to vector 'uint16x4_t' (vector of 4 'uint16_t' values) of different size
    vres = vshrn_n_u16(vcombine_u16(tmp, vcreate_u16(0)), 8); // shift down result by 8
           ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~
llvm.inst/lib/clang/10.0.0/include/arm_neon.h:4149:11: note: expanded from macro 'vcreate_u16'
  __ret = (uint16x4_t)(__p0); \
          ^
llvm.inst/lib/clang/10.0.0/include/arm_neon.h:24249:21: note: expanded from macro 'vshrn_n_u16'
  uint16x8_t __s0 = __p0; \
                    ^~~~

[t.p.northover]

Sorry about the delay investigating this, your e-mail bypassed my inbox for some reason and I only noticed the issue when Hans reverted the change this afternoon. I'm looking into it now.

clang/include/clang/Basic/arm_fp16.td

 //===----------------------------------------------------------------------===//
 
 include "arm_neon_incl.td"
 
 // ARMv8.2-A FP16 intrinsics.
 let ArchGuard = "defined(__ARM_FEATURE_FP16_SCALAR_ARITHMETIC) && defined(__aarch64__)" in {
 
   // Negate
-  def VNEGSH          : SInst<"vneg", "ss", "Sh">;
+  def VNEGSH          : SInst<"vneg", "11", "Sh">;
 
   // Reciprocal/Sqrt
-  def SCALAR_FRECPSH  : IInst<"vrecps", "sss", "Sh">;
-  def FSQRTSH         : SInst<"vsqrt", "ss", "Sh">;
-  def SCALAR_FRSQRTSH : IInst<"vrsqrts", "sss", "Sh">;
+  def SCALAR_FRECPSH  : IInst<"vrecps", "111", "Sh">;
+  def FSQRTSH         : SInst<"vsqrt", "11", "Sh">;
+  def SCALAR_FRSQRTSH : IInst<"vrsqrts", "111", "Sh">;
 
   // Reciprocal Estimate
-  def SCALAR_FRECPEH  : IInst<"vrecpe", "ss", "Sh">;
+  def SCALAR_FRECPEH  : IInst<"vrecpe", "11", "Sh">;
 
   // Reciprocal Exponent
-  def SCALAR_FRECPXH  : IInst<"vrecpx", "ss", "Sh">;
+  def SCALAR_FRECPXH  : IInst<"vrecpx", "11", "Sh">;
 
   // Reciprocal Square Root Estimate
-  def SCALAR_FRSQRTEH : IInst<"vrsqrte", "ss", "Sh">;
+  def SCALAR_FRSQRTEH : IInst<"vrsqrte", "11", "Sh">;
 
   // Rounding
-  def FRINTZ_S64H     : SInst<"vrnd", "ss", "Sh">;
-  def FRINTA_S64H     : SInst<"vrnda", "ss", "Sh">;
-  def FRINTI_S64H     : SInst<"vrndi", "ss", "Sh">;
-  def FRINTM_S64H     : SInst<"vrndm", "ss", "Sh">;
-  def FRINTN_S64H     : SInst<"vrndn", "ss", "Sh">;
-  def FRINTP_S64H     : SInst<"vrndp", "ss", "Sh">;
-  def FRINTX_S64H     : SInst<"vrndx", "ss", "Sh">;
+  def FRINTZ_S64H     : SInst<"vrnd", "11", "Sh">;
+  def FRINTA_S64H     : SInst<"vrnda", "11", "Sh">;
+  def FRINTI_S64H     : SInst<"vrndi", "11", "Sh">;
+  def FRINTM_S64H     : SInst<"vrndm", "11", "Sh">;
+  def FRINTN_S64H     : SInst<"vrndn", "11", "Sh">;
+  def FRINTP_S64H     : SInst<"vrndp", "11", "Sh">;
+  def FRINTX_S64H     : SInst<"vrndx", "11", "Sh">;
 
   // Conversion
-  def SCALAR_SCVTFSH  : SInst<"vcvth_f16", "Ys", "silUsUiUl">;
-  def SCALAR_FCVTZSH  : SInst<"vcvt_s16", "$s", "Sh">;
-  def SCALAR_FCVTZSH1 : SInst<"vcvt_s32", "Is", "Sh">;
-  def SCALAR_FCVTZSH2 : SInst<"vcvt_s64", "Ls", "Sh">;
-  def SCALAR_FCVTZUH  : SInst<"vcvt_u16", "bs", "Sh">;
-  def SCALAR_FCVTZUH1 : SInst<"vcvt_u32", "Us", "Sh">;
-  def SCALAR_FCVTZUH2 : SInst<"vcvt_u64", "Os", "Sh">;
-  def SCALAR_FCVTASH  : SInst<"vcvta_s16", "$s", "Sh">;
-  def SCALAR_FCVTASH1 : SInst<"vcvta_s32", "Is", "Sh">;
-  def SCALAR_FCVTASH2 : SInst<"vcvta_s64", "Ls", "Sh">;
-  def SCALAR_FCVTAUH  : SInst<"vcvta_u16", "bs", "Sh">;
-  def SCALAR_FCVTAUH1 : SInst<"vcvta_u32", "Us", "Sh">;
-  def SCALAR_FCVTAUH2 : SInst<"vcvta_u64", "Os", "Sh">;
-  def SCALAR_FCVTMSH  : SInst<"vcvtm_s16", "$s", "Sh">;
-  def SCALAR_FCVTMSH1 : SInst<"vcvtm_s32", "Is", "Sh">;
-  def SCALAR_FCVTMSH2 : SInst<"vcvtm_s64", "Ls", "Sh">;
-  def SCALAR_FCVTMUH  : SInst<"vcvtm_u16", "bs", "Sh">;
-  def SCALAR_FCVTMUH1 : SInst<"vcvtm_u32", "Us", "Sh">;
-  def SCALAR_FCVTMUH2 : SInst<"vcvtm_u64", "Os", "Sh">;
-  def SCALAR_FCVTNSH  : SInst<"vcvtn_s16", "$s", "Sh">;
-  def SCALAR_FCVTNSH1 : SInst<"vcvtn_s32", "Is", "Sh">;
-  def SCALAR_FCVTNSH2 : SInst<"vcvtn_s64", "Ls", "Sh">;
-  def SCALAR_FCVTNUH  : SInst<"vcvtn_u16", "bs", "Sh">;
-  def SCALAR_FCVTNUH1 : SInst<"vcvtn_u32", "Us", "Sh">;
-  def SCALAR_FCVTNUH2 : SInst<"vcvtn_u64", "Os", "Sh">;
-  def SCALAR_FCVTPSH  : SInst<"vcvtp_s16", "$s", "Sh">;
-  def SCALAR_FCVTPSH1 : SInst<"vcvtp_s32", "Is", "Sh">;
-  def SCALAR_FCVTPSH2 : SInst<"vcvtp_s64", "Ls", "Sh">;
-  def SCALAR_FCVTPUH  : SInst<"vcvtp_u16", "bs", "Sh">;
-  def SCALAR_FCVTPUH1 : SInst<"vcvtp_u32", "Us", "Sh">;
-  def SCALAR_FCVTPUH2 : SInst<"vcvtp_u64", "Os", "Sh">;
+  def SCALAR_SCVTFSH  : SInst<"vcvth_f16", "(1F)1", "sUs">;
+  def SCALAR_SCVTFSH1 : SInst<"vcvth_f16", "(1F<)1", "iUi">;
+  def SCALAR_SCVTFSH2 : SInst<"vcvth_f16", "(1F<<)1", "lUl">;
+  def SCALAR_FCVTZSH  : SInst<"vcvt_s16", "(1S)1", "Sh">;
+  def SCALAR_FCVTZSH1 : SInst<"vcvt_s32", "(1S>)1", "Sh">;
+  def SCALAR_FCVTZSH2 : SInst<"vcvt_s64", "(1S>>)1", "Sh">;
+  def SCALAR_FCVTZUH  : SInst<"vcvt_u16", "(1U)1", "Sh">;
+  def SCALAR_FCVTZUH1 : SInst<"vcvt_u32", "(1U>)1", "Sh">;
+  def SCALAR_FCVTZUH2 : SInst<"vcvt_u64", "(1U>>)1", "Sh">;
+  def SCALAR_FCVTASH  : SInst<"vcvta_s16", "(1S)1", "Sh">;
+  def SCALAR_FCVTASH1 : SInst<"vcvta_s32", "(1S>)1", "Sh">;
+  def SCALAR_FCVTASH2 : SInst<"vcvta_s64", "(1S>>)1", "Sh">;
+  def SCALAR_FCVTAUH  : SInst<"vcvta_u16", "(1U)1", "Sh">;
+  def SCALAR_FCVTAUH1 : SInst<"vcvta_u32", "(1U>)1", "Sh">;
+  def SCALAR_FCVTAUH2 : SInst<"vcvta_u64", "(1U>>)1", "Sh">;
+  def SCALAR_FCVTMSH  : SInst<"vcvtm_s16", "(1S)1", "Sh">;
+  def SCALAR_FCVTMSH1 : SInst<"vcvtm_s32", "(1S>)1", "Sh">;
+  def SCALAR_FCVTMSH2 : SInst<"vcvtm_s64", "(1S>>)1", "Sh">;
+  def SCALAR_FCVTMUH  : SInst<"vcvtm_u16", "(1U)1", "Sh">;
+  def SCALAR_FCVTMUH1 : SInst<"vcvtm_u32", "(1U>)1", "Sh">;
+  def SCALAR_FCVTMUH2 : SInst<"vcvtm_u64", "(1U>>)1", "Sh">;
+  def SCALAR_FCVTNSH  : SInst<"vcvtn_s16", "(1S)1", "Sh">;
+  def SCALAR_FCVTNSH1 : SInst<"vcvtn_s32", "(1S>)1", "Sh">;
+  def SCALAR_FCVTNSH2 : SInst<"vcvtn_s64", "(1S>>)1", "Sh">;
+  def SCALAR_FCVTNUH  : SInst<"vcvtn_u16", "(1U)1", "Sh">;
+  def SCALAR_FCVTNUH1 : SInst<"vcvtn_u32", "(1U>)1", "Sh">;
+  def SCALAR_FCVTNUH2 : SInst<"vcvtn_u64", "(1U>>)1", "Sh">;
+  def SCALAR_FCVTPSH  : SInst<"vcvtp_s16", "(1S)1", "Sh">;
+  def SCALAR_FCVTPSH1 : SInst<"vcvtp_s32", "(1S>)1", "Sh">;
+  def SCALAR_FCVTPSH2 : SInst<"vcvtp_s64", "(1S>>)1", "Sh">;
+  def SCALAR_FCVTPUH  : SInst<"vcvtp_u16", "(1U)1", "Sh">;
+  def SCALAR_FCVTPUH1 : SInst<"vcvtp_u32", "(1U>)1", "Sh">;
+  def SCALAR_FCVTPUH2 : SInst<"vcvtp_u64", "(1U>>)1", "Sh">;
   let isVCVT_N = 1 in {
-    def SCALAR_SCVTFSHO : SInst<"vcvth_n_f16", "Ysi", "silUsUiUl">;
-    def SCALAR_FCVTZSHO : SInst<"vcvt_n_s16", "$si", "Sh">;
-    def SCALAR_FCVTZSH1O: SInst<"vcvt_n_s32", "Isi", "Sh">;
-    def SCALAR_FCVTZSH2O: SInst<"vcvt_n_s64", "Lsi", "Sh">;
-    def SCALAR_FCVTZUHO : SInst<"vcvt_n_u16", "bsi", "Sh">;
-    def SCALAR_FCVTZUH1O: SInst<"vcvt_n_u32", "Usi", "Sh">;
-    def SCALAR_FCVTZUH2O: SInst<"vcvt_n_u64", "Osi", "Sh">;
+    def SCALAR_SCVTFSHO : SInst<"vcvth_n_f16", "(1F)1I", "sUs">;
+    def SCALAR_SCVTFSH1O: SInst<"vcvth_n_f16", "(1F<)1I", "iUi">;
+    def SCALAR_SCVTFSH2O: SInst<"vcvth_n_f16", "(1F<<)1I", "lUl">;
+    def SCALAR_FCVTZSHO : SInst<"vcvt_n_s16", "(1S)1I", "Sh">;
+    def SCALAR_FCVTZSH1O: SInst<"vcvt_n_s32", "(1S>)1I", "Sh">;
+    def SCALAR_FCVTZSH2O: SInst<"vcvt_n_s64", "(1S>>)1I", "Sh">;
+    def SCALAR_FCVTZUHO : SInst<"vcvt_n_u16", "(1U)1I", "Sh">;
+    def SCALAR_FCVTZUH1O: SInst<"vcvt_n_u32", "(1U>)1I", "Sh">;
+    def SCALAR_FCVTZUH2O: SInst<"vcvt_n_u64", "(1U>>)1I", "Sh">;
   }
   // Comparison
-  def SCALAR_CMEQRH   : SInst<"vceq", "bss", "Sh">;
-  def SCALAR_CMEQZH   : SInst<"vceqz", "bs", "Sh">;
-  def SCALAR_CMGERH   : SInst<"vcge", "bss", "Sh">;
-  def SCALAR_CMGEZH   : SInst<"vcgez", "bs", "Sh">;
-  def SCALAR_CMGTRH   : SInst<"vcgt", "bss", "Sh">;
-  def SCALAR_CMGTZH   : SInst<"vcgtz", "bs", "Sh">;
-  def SCALAR_CMLERH   : SInst<"vcle", "bss", "Sh">;
-  def SCALAR_CMLEZH   : SInst<"vclez", "bs", "Sh">;
-  def SCALAR_CMLTH    : SInst<"vclt", "bss", "Sh">;
-  def SCALAR_CMLTZH   : SInst<"vcltz", "bs", "Sh">;
+  def SCALAR_CMEQRH   : SInst<"vceq", "(1U)11", "Sh">;
+  def SCALAR_CMEQZH   : SInst<"vceqz", "(1U)1", "Sh">;
+  def SCALAR_CMGERH   : SInst<"vcge", "(1U)11", "Sh">;
+  def SCALAR_CMGEZH   : SInst<"vcgez", "(1U)1", "Sh">;
+  def SCALAR_CMGTRH   : SInst<"vcgt", "(1U)11", "Sh">;
+  def SCALAR_CMGTZH   : SInst<"vcgtz", "(1U)1", "Sh">;
+  def SCALAR_CMLERH   : SInst<"vcle", "(1U)11", "Sh">;
+  def SCALAR_CMLEZH   : SInst<"vclez", "(1U)1", "Sh">;
+  def SCALAR_CMLTH    : SInst<"vclt", "(1U)11", "Sh">;
+  def SCALAR_CMLTZH   : SInst<"vcltz", "(1U)1", "Sh">;
 
   // Absolute Compare Mask Greater Than Or Equal
-  def SCALAR_FACGEH   : IInst<"vcage", "bss", "Sh">;
-  def SCALAR_FACLEH   : IInst<"vcale", "bss", "Sh">;
+  def SCALAR_FACGEH   : IInst<"vcage", "(1U)11", "Sh">;
+  def SCALAR_FACLEH   : IInst<"vcale", "(1U)11", "Sh">;
 
   // Absolute Compare Mask Greater Than
-  def SCALAR_FACGT    : IInst<"vcagt", "bss", "Sh">;
-  def SCALAR_FACLT    : IInst<"vcalt", "bss", "Sh">;
+  def SCALAR_FACGT    : IInst<"vcagt", "(1U)11", "Sh">;
+  def SCALAR_FACLT    : IInst<"vcalt", "(1U)11", "Sh">;
 
   // Scalar Absolute Value
-  def SCALAR_ABSH     : SInst<"vabs", "ss", "Sh">;
+  def SCALAR_ABSH     : SInst<"vabs", "11", "Sh">;
 
   // Scalar Absolute Difference
-  def SCALAR_ABDH: IInst<"vabd", "sss", "Sh">;
+  def SCALAR_ABDH: IInst<"vabd", "111", "Sh">;
 
   // Add/Sub
-  def VADDSH          : SInst<"vadd", "sss", "Sh">;
-  def VSUBHS          : SInst<"vsub", "sss", "Sh">;
+  def VADDSH          : SInst<"vadd", "111", "Sh">;
+  def VSUBHS          : SInst<"vsub", "111", "Sh">;
 
   // Max/Min
-  def VMAXHS          : SInst<"vmax", "sss", "Sh">;
-  def VMINHS          : SInst<"vmin", "sss", "Sh">;
-  def FMAXNMHS        : SInst<"vmaxnm", "sss", "Sh">;
-  def FMINNMHS        : SInst<"vminnm", "sss", "Sh">;
+  def VMAXHS          : SInst<"vmax", "111", "Sh">;
+  def VMINHS          : SInst<"vmin", "111", "Sh">;
+  def FMAXNMHS        : SInst<"vmaxnm", "111", "Sh">;
+  def FMINNMHS        : SInst<"vminnm", "111", "Sh">;
 
   // Multiplication/Division
-  def VMULHS          : SInst<"vmul", "sss", "Sh">;
-  def MULXHS          : SInst<"vmulx", "sss", "Sh">;
-  def FDIVHS          : SInst<"vdiv", "sss",  "Sh">;
+  def VMULHS          : SInst<"vmul", "111", "Sh">;
+  def MULXHS          : SInst<"vmulx", "111", "Sh">;
+  def FDIVHS          : SInst<"vdiv", "111",  "Sh">;
 
   // Vector fused multiply-add operations
-  def VFMAHS          : SInst<"vfma", "ssss", "Sh">;
-  def VFMSHS          : SInst<"vfms", "ssss", "Sh">;
+  def VFMAHS          : SInst<"vfma", "1111", "Sh">;
+  def VFMSHS          : SInst<"vfms", "1111", "Sh">;
 }

clang/include/clang/Basic/arm_neon.td

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",
+def VADD    : IOpInst<"vadd", "...",
                       "csilfUcUsUiUlQcQsQiQlQfQUcQUsQUiQUl", OP_ADD>;
-def VADDL   : SOpInst<"vaddl", "wdd", "csiUcUsUi", OP_ADDL>;
-def VADDW   : SOpInst<"vaddw", "wwd", "csiUcUsUi", OP_ADDW>;
-def VHADD   : SInst<"vhadd", "ddd", "csiUcUsUiQcQsQiQUcQUsQUi">;
-def VRHADD  : SInst<"vrhadd", "ddd", "csiUcUsUiQcQsQiQUcQUsQUi">;
-def VQADD   : SInst<"vqadd", "ddd", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
-def VADDHN  : IInst<"vaddhn", "hkk", "silUsUiUl">;
-def VRADDHN : IInst<"vraddhn", "hkk", "silUsUiUl">;
+def VADDL   : SOpInst<"vaddl", "(>Q)..", "csiUcUsUi", OP_ADDL>;
+def VADDW   : SOpInst<"vaddw", "(>Q)(>Q).", "csiUcUsUi", OP_ADDW>;
+def VHADD   : SInst<"vhadd", "...", "csiUcUsUiQcQsQiQUcQUsQUi">;
+def VRHADD  : SInst<"vrhadd", "...", "csiUcUsUiQcQsQiQUcQUsQUi">;
+def VQADD   : SInst<"vqadd", "...", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
+def VADDHN  : IInst<"vaddhn", "<QQ", "silUsUiUl">;
+def VRADDHN : IInst<"vraddhn", "<QQ", "silUsUiUl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.2 Multiplication
-def VMUL     : IOpInst<"vmul", "ddd", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_MUL>;
-def VMULP    : SInst<"vmul", "ddd", "PcQPc">;
-def VMLA     : IOpInst<"vmla", "dddd", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_MLA>;
-def VMLAL    : SOpInst<"vmlal", "wwdd", "csiUcUsUi", OP_MLAL>;
-def VMLS     : IOpInst<"vmls", "dddd", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_MLS>;
-def VMLSL    : SOpInst<"vmlsl", "wwdd", "csiUcUsUi", OP_MLSL>;
-def VQDMULH  : SInst<"vqdmulh", "ddd", "siQsQi">;
-def VQRDMULH : SInst<"vqrdmulh", "ddd", "siQsQi">;
+def VMUL     : IOpInst<"vmul", "...", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_MUL>;
+def VMULP    : SInst<"vmul", "...", "PcQPc">;
+def VMLA     : IOpInst<"vmla", "....", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_MLA>;
+def VMLAL    : SOpInst<"vmlal", "(>Q)(>Q)..", "csiUcUsUi", OP_MLAL>;
+def VMLS     : IOpInst<"vmls", "....", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_MLS>;
+def VMLSL    : SOpInst<"vmlsl", "(>Q)(>Q)..", "csiUcUsUi", OP_MLSL>;
+def VQDMULH  : SInst<"vqdmulh", "...", "siQsQi">;
+def VQRDMULH : SInst<"vqrdmulh", "...", "siQsQi">;
 
 let ArchGuard = "defined(__ARM_FEATURE_QRDMX)" in {
-def VQRDMLAH : SOpInst<"vqrdmlah", "dddd", "siQsQi", OP_QRDMLAH>;
-def VQRDMLSH : SOpInst<"vqrdmlsh", "dddd", "siQsQi", OP_QRDMLSH>;
+def VQRDMLAH : SOpInst<"vqrdmlah", "....", "siQsQi", OP_QRDMLAH>;
+def VQRDMLSH : SOpInst<"vqrdmlsh", "....", "siQsQi", OP_QRDMLSH>;
 }
 
-def VQDMLAL  : SInst<"vqdmlal", "wwdd", "si">;
-def VQDMLSL  : SInst<"vqdmlsl", "wwdd", "si">;
-def VMULL    : SInst<"vmull", "wdd", "csiUcUsUiPc">;
-def VQDMULL  : SInst<"vqdmull", "wdd", "si">;
+def VQDMLAL  : SInst<"vqdmlal", "(>Q)(>Q)..", "si">;
+def VQDMLSL  : SInst<"vqdmlsl", "(>Q)(>Q)..", "si">;
+def VMULL    : SInst<"vmull", "(>Q)..", "csiUcUsUiPc">;
+def VQDMULL  : SInst<"vqdmull", "(>Q)..", "si">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.3 Subtraction
-def VSUB    : IOpInst<"vsub", "ddd",
+def VSUB    : IOpInst<"vsub", "...",
                       "csilfUcUsUiUlQcQsQiQlQfQUcQUsQUiQUl", OP_SUB>;
-def VSUBL   : SOpInst<"vsubl", "wdd", "csiUcUsUi", OP_SUBL>;
-def VSUBW   : SOpInst<"vsubw", "wwd", "csiUcUsUi", OP_SUBW>;
-def VQSUB   : SInst<"vqsub", "ddd", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
-def VHSUB   : SInst<"vhsub", "ddd", "csiUcUsUiQcQsQiQUcQUsQUi">;
-def VSUBHN  : IInst<"vsubhn", "hkk", "silUsUiUl">;
-def VRSUBHN : IInst<"vrsubhn", "hkk", "silUsUiUl">;
+def VSUBL   : SOpInst<"vsubl", "(>Q)..", "csiUcUsUi", OP_SUBL>;
+def VSUBW   : SOpInst<"vsubw", "(>Q)(>Q).", "csiUcUsUi", OP_SUBW>;
+def VQSUB   : SInst<"vqsub", "...", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
+def VHSUB   : SInst<"vhsub", "...", "csiUcUsUiQcQsQiQUcQUsQUi">;
+def VSUBHN  : IInst<"vsubhn", "<QQ", "silUsUiUl">;
+def VRSUBHN : IInst<"vrsubhn", "<QQ", "silUsUiUl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.4 Comparison
-def VCEQ  : IOpInst<"vceq", "udd", "csifUcUsUiPcQcQsQiQfQUcQUsQUiQPc", OP_EQ>;
-def VCGE  : SOpInst<"vcge", "udd", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_GE>;
+def VCEQ  : IOpInst<"vceq", "U..", "csifUcUsUiPcQcQsQiQfQUcQUsQUiQPc", OP_EQ>;
+def VCGE  : SOpInst<"vcge", "U..", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_GE>;
 let InstName = "vcge" in
-def VCLE  : SOpInst<"vcle", "udd", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_LE>;
-def VCGT  : SOpInst<"vcgt", "udd", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_GT>;
+def VCLE  : SOpInst<"vcle", "U..", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_LE>;
+def VCGT  : SOpInst<"vcgt", "U..", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_GT>;
 let InstName = "vcgt" in
-def VCLT  : SOpInst<"vclt", "udd", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_LT>;
+def VCLT  : SOpInst<"vclt", "U..", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_LT>;
 let InstName = "vacge" in {
-def VCAGE : IInst<"vcage", "udd", "fQf">;
-def VCALE : IInst<"vcale", "udd", "fQf">;
+def VCAGE : IInst<"vcage", "U..", "fQf">;
+def VCALE : IInst<"vcale", "U..", "fQf">;
 }
 let InstName = "vacgt" in {
-def VCAGT : IInst<"vcagt", "udd", "fQf">;
-def VCALT : IInst<"vcalt", "udd", "fQf">;
+def VCAGT : IInst<"vcagt", "U..", "fQf">;
+def VCALT : IInst<"vcalt", "U..", "fQf">;
 }
-def VTST  : WInst<"vtst", "udd", "csiUcUsUiPcPsQcQsQiQUcQUsQUiQPcQPs">;
+def VTST  : WInst<"vtst", "U..", "csiUcUsUiPcPsQcQsQiQUcQUsQUiQPcQPs">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.5 Absolute Difference
-def VABD  : SInst<"vabd", "ddd",  "csiUcUsUifQcQsQiQUcQUsQUiQf">;
-def VABDL : SOpInst<"vabdl", "wdd",  "csiUcUsUi", OP_ABDL>;
-def VABA  : SOpInst<"vaba", "dddd", "csiUcUsUiQcQsQiQUcQUsQUi", OP_ABA>;
-def VABAL : SOpInst<"vabal", "wwdd", "csiUcUsUi", OP_ABAL>;
+def VABD  : SInst<"vabd", "...",  "csiUcUsUifQcQsQiQUcQUsQUiQf">;
+def VABDL : SOpInst<"vabdl", "(>Q)..",  "csiUcUsUi", OP_ABDL>;
+def VABA  : SOpInst<"vaba", "....", "csiUcUsUiQcQsQiQUcQUsQUi", OP_ABA>;
+def VABAL : SOpInst<"vabal", "(>Q)(>Q)..", "csiUcUsUi", OP_ABAL>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.6 Max/Min
-def VMAX : SInst<"vmax", "ddd", "csiUcUsUifQcQsQiQUcQUsQUiQf">;
-def VMIN : SInst<"vmin", "ddd", "csiUcUsUifQcQsQiQUcQUsQUiQf">;
+def VMAX : SInst<"vmax", "...", "csiUcUsUifQcQsQiQUcQUsQUiQf">;
+def VMIN : SInst<"vmin", "...", "csiUcUsUifQcQsQiQUcQUsQUiQf">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.7 Pairwise Addition
-def VPADD  : IInst<"vpadd", "ddd", "csiUcUsUif">;
-def VPADDL : SInst<"vpaddl", "nd",  "csiUcUsUiQcQsQiQUcQUsQUi">;
-def VPADAL : SInst<"vpadal", "nnd", "csiUcUsUiQcQsQiQUcQUsQUi">;
+def VPADD  : IInst<"vpadd", "...", "csiUcUsUif">;
+def VPADDL : SInst<"vpaddl", ">.",  "csiUcUsUiQcQsQiQUcQUsQUi">;
+def VPADAL : SInst<"vpadal", ">>.", "csiUcUsUiQcQsQiQUcQUsQUi">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.8-9 Folding Max/Min
-def VPMAX : SInst<"vpmax", "ddd", "csiUcUsUif">;
-def VPMIN : SInst<"vpmin", "ddd", "csiUcUsUif">;
+def VPMAX : SInst<"vpmax", "...", "csiUcUsUif">;
+def VPMIN : SInst<"vpmin", "...", "csiUcUsUif">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.10 Reciprocal/Sqrt
-def VRECPS  : IInst<"vrecps", "ddd", "fQf">;
-def VRSQRTS : IInst<"vrsqrts", "ddd", "fQf">;
+def VRECPS  : IInst<"vrecps", "...", "fQf">;
+def VRSQRTS : IInst<"vrsqrts", "...", "fQf">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.11 Shifts by signed variable
-def VSHL   : SInst<"vshl", "ddx", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
-def VQSHL  : SInst<"vqshl", "ddx", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
-def VRSHL  : SInst<"vrshl", "ddx", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
-def VQRSHL : SInst<"vqrshl", "ddx", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
+def VSHL   : SInst<"vshl", "..S", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
+def VQSHL  : SInst<"vqshl", "..S", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
+def VRSHL  : SInst<"vrshl", "..S", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
+def VQRSHL : SInst<"vqrshl", "..S", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.12 Shifts by constant
 let isShift = 1 in {
-def VSHR_N     : SInst<"vshr_n", "ddi", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
-def VSHL_N     : IInst<"vshl_n", "ddi", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
-def VRSHR_N    : SInst<"vrshr_n", "ddi", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
-def VSRA_N     : SInst<"vsra_n", "dddi", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
-def VRSRA_N    : SInst<"vrsra_n", "dddi", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
-def VQSHL_N    : SInst<"vqshl_n", "ddi", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
-def VQSHLU_N   : SInst<"vqshlu_n", "udi", "csilQcQsQiQl">;
-def VSHRN_N    : IInst<"vshrn_n", "hki", "silUsUiUl">;
-def VQSHRUN_N  : SInst<"vqshrun_n", "eki", "sil">;
-def VQRSHRUN_N : SInst<"vqrshrun_n", "eki", "sil">;
-def VQSHRN_N   : SInst<"vqshrn_n", "hki", "silUsUiUl">;
-def VRSHRN_N   : IInst<"vrshrn_n", "hki", "silUsUiUl">;
-def VQRSHRN_N  : SInst<"vqrshrn_n", "hki", "silUsUiUl">;
-def VSHLL_N    : SInst<"vshll_n", "wdi", "csiUcUsUi">;
+def VSHR_N     : SInst<"vshr_n", "..I", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
+def VSHL_N     : IInst<"vshl_n", "..I", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
+def VRSHR_N    : SInst<"vrshr_n", "..I", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
+def VSRA_N     : SInst<"vsra_n", "...I", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
+def VRSRA_N    : SInst<"vrsra_n", "...I", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
+def VQSHL_N    : SInst<"vqshl_n", "..I", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
+def VQSHLU_N   : SInst<"vqshlu_n", "U.I", "csilQcQsQiQl">;
+def VSHRN_N    : IInst<"vshrn_n", "<QI", "silUsUiUl">;
+def VQSHRUN_N  : SInst<"vqshrun_n", "(<U)QI", "sil">;
+def VQRSHRUN_N : SInst<"vqrshrun_n", "(<U)QI", "sil">;
+def VQSHRN_N   : SInst<"vqshrn_n", "<QI", "silUsUiUl">;
+def VRSHRN_N   : IInst<"vrshrn_n", "<QI", "silUsUiUl">;
+def VQRSHRN_N  : SInst<"vqrshrn_n", "<QI", "silUsUiUl">;
+def VSHLL_N    : SInst<"vshll_n", "(>Q).I", "csiUcUsUi">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.13 Shifts with insert
-def VSRI_N : WInst<"vsri_n", "dddi",
+def VSRI_N : WInst<"vsri_n", "...I",
                    "csilUcUsUiUlPcPsQcQsQiQlQUcQUsQUiQUlQPcQPs">;
-def VSLI_N : WInst<"vsli_n", "dddi",
+def VSLI_N : WInst<"vsli_n", "...I",
                    "csilUcUsUiUlPcPsQcQsQiQlQUcQUsQUiQUlQPcQPs">;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.14 Loads and stores of a single vector
-def VLD1      : WInst<"vld1", "dc",
+def VLD1      : WInst<"vld1", ".(c*!)",
                       "QUcQUsQUiQUlQcQsQiQlQfQPcQPsUcUsUiUlcsilfPcPs">;
-def VLD1_X2   : WInst<"vld1_x2", "2c",
+def VLD1_X2   : WInst<"vld1_x2", "2(c*!)",
                       "cfilsUcUiUlUsQcQfQiQlQsQUcQUiQUlQUsPcPsQPcQPs">;
-def VLD1_X3   : WInst<"vld1_x3", "3c",
+def VLD1_X3   : WInst<"vld1_x3", "3(c*!)",
                       "cfilsUcUiUlUsQcQfQiQlQsQUcQUiQUlQUsPcPsQPcQPs">;
-def VLD1_X4   : WInst<"vld1_x4", "4c",
+def VLD1_X4   : WInst<"vld1_x4", "4(c*!)",
                       "cfilsUcUiUlUsQcQfQiQlQsQUcQUiQUlQUsPcPsQPcQPs">;
-def VLD1_LANE : WInst<"vld1_lane", "dcdi",
+def VLD1_LANE : WInst<"vld1_lane", ".(c*!).I",
                       "QUcQUsQUiQUlQcQsQiQlQfQPcQPsUcUsUiUlcsilfPcPs">;
-def VLD1_DUP  : WInst<"vld1_dup", "dc",
+def VLD1_DUP  : WInst<"vld1_dup", ".(c*!)",
                       "QUcQUsQUiQUlQcQsQiQlQfQPcQPsUcUsUiUlcsilfPcPs">;
-def VST1      : WInst<"vst1", "vpd",
+def VST1      : WInst<"vst1", "v*(.!)",
                       "QUcQUsQUiQUlQcQsQiQlQfQPcQPsUcUsUiUlcsilfPcPs">;
-def VST1_X2   : WInst<"vst1_x2", "vp2",
+def VST1_X2   : WInst<"vst1_x2", "v*(2!)",
                       "cfilsUcUiUlUsQcQfQiQlQsQUcQUiQUlQUsPcPsQPcQPs">;
-def VST1_X3   : WInst<"vst1_x3", "vp3",
+def VST1_X3   : WInst<"vst1_x3", "v*(3!)",
                       "cfilsUcUiUlUsQcQfQiQlQsQUcQUiQUlQUsPcPsQPcQPs">;
-def VST1_X4   : WInst<"vst1_x4", "vp4",
+def VST1_X4   : WInst<"vst1_x4", "v*(4!)",
                       "cfilsUcUiUlUsQcQfQiQlQsQUcQUiQUlQUsPcPsQPcQPs">;
-def VST1_LANE : WInst<"vst1_lane", "vpdi",
+def VST1_LANE : WInst<"vst1_lane", "v*(.!)I",
                       "QUcQUsQUiQUlQcQsQiQlQfQPcQPsUcUsUiUlcsilfPcPs">;
 let ArchGuard = "(__ARM_FP & 2)" in {
-def VLD1_F16      : WInst<"vld1", "dc", "hQh">;
-def VLD1_X2_F16   : WInst<"vld1_x2", "2c", "hQh">;
-def VLD1_X3_F16   : WInst<"vld1_x3", "3c", "hQh">;
-def VLD1_X4_F16   : WInst<"vld1_x4", "4c", "hQh">;
-def VLD1_LANE_F16 : WInst<"vld1_lane", "dcdi", "hQh">;
-def VLD1_DUP_F16  : WInst<"vld1_dup", "dc", "hQh">;
-def VST1_F16      : WInst<"vst1", "vpd", "hQh">;
-def VST1_X2_F16   : WInst<"vst1_x2", "vp2", "hQh">;
-def VST1_X3_F16   : WInst<"vst1_x3", "vp3", "hQh">;
-def VST1_X4_F16   : WInst<"vst1_x4", "vp4", "hQh">;
-def VST1_LANE_F16 : WInst<"vst1_lane", "vpdi", "hQh">;
+def VLD1_F16      : WInst<"vld1", ".(c*!)", "hQh">;
+def VLD1_X2_F16   : WInst<"vld1_x2", "2(c*!)", "hQh">;
+def VLD1_X3_F16   : WInst<"vld1_x3", "3(c*!)", "hQh">;
+def VLD1_X4_F16   : WInst<"vld1_x4", "4(c*!)", "hQh">;
+def VLD1_LANE_F16 : WInst<"vld1_lane", ".(c*!).I", "hQh">;
+def VLD1_DUP_F16  : WInst<"vld1_dup", ".(c*!)", "hQh">;
+def VST1_F16      : WInst<"vst1", "v*(.!)", "hQh">;
+def VST1_X2_F16   : WInst<"vst1_x2", "v*(2!)", "hQh">;
+def VST1_X3_F16   : WInst<"vst1_x3", "v*(3!)", "hQh">;
+def VST1_X4_F16   : WInst<"vst1_x4", "v*(4!)", "hQh">;
+def VST1_LANE_F16 : WInst<"vst1_lane", "v*(.!)I", "hQh">;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.15 Loads and stores of an N-element structure
-def VLD2 : WInst<"vld2", "2c", "QUcQUsQUiQcQsQiQfQPcQPsUcUsUiUlcsilfPcPs">;
-def VLD3 : WInst<"vld3", "3c", "QUcQUsQUiQcQsQiQfQPcQPsUcUsUiUlcsilfPcPs">;
-def VLD4 : WInst<"vld4", "4c", "QUcQUsQUiQcQsQiQfQPcQPsUcUsUiUlcsilfPcPs">;
-def VLD2_DUP  : WInst<"vld2_dup", "2c",
+def VLD2 : WInst<"vld2", "2(c*!)", "QUcQUsQUiQcQsQiQfQPcQPsUcUsUiUlcsilfPcPs">;
+def VLD3 : WInst<"vld3", "3(c*!)", "QUcQUsQUiQcQsQiQfQPcQPsUcUsUiUlcsilfPcPs">;
+def VLD4 : WInst<"vld4", "4(c*!)", "QUcQUsQUiQcQsQiQfQPcQPsUcUsUiUlcsilfPcPs">;
+def VLD2_DUP  : WInst<"vld2_dup", "2(c*!)",
                       "UcUsUiUlcsilfPcPsQcQfQiQlQsQPcQPsQUcQUiQUlQUs">;
-def VLD3_DUP  : WInst<"vld3_dup", "3c",
+def VLD3_DUP  : WInst<"vld3_dup", "3(c*!)",
                       "UcUsUiUlcsilfPcPsQcQfQiQlQsQPcQPsQUcQUiQUlQUs">;
-def VLD4_DUP  : WInst<"vld4_dup", "4c",
+def VLD4_DUP  : WInst<"vld4_dup", "4(c*!)",
                       "UcUsUiUlcsilfPcPsQcQfQiQlQsQPcQPsQUcQUiQUlQUs">;
-def VLD2_LANE : WInst<"vld2_lane", "2c2i", "QUsQUiQsQiQfQPsUcUsUicsifPcPs">;
-def VLD3_LANE : WInst<"vld3_lane", "3c3i", "QUsQUiQsQiQfQPsUcUsUicsifPcPs">;
-def VLD4_LANE : WInst<"vld4_lane", "4c4i", "QUsQUiQsQiQfQPsUcUsUicsifPcPs">;
-def VST2 : WInst<"vst2", "vp2", "QUcQUsQUiQcQsQiQfQPcQPsUcUsUiUlcsilfPcPs">;
-def VST3 : WInst<"vst3", "vp3", "QUcQUsQUiQcQsQiQfQPcQPsUcUsUiUlcsilfPcPs">;
-def VST4 : WInst<"vst4", "vp4", "QUcQUsQUiQcQsQiQfQPcQPsUcUsUiUlcsilfPcPs">;
-def VST2_LANE : WInst<"vst2_lane", "vp2i", "QUsQUiQsQiQfQPsUcUsUicsifPcPs">;
-def VST3_LANE : WInst<"vst3_lane", "vp3i", "QUsQUiQsQiQfQPsUcUsUicsifPcPs">;
-def VST4_LANE : WInst<"vst4_lane", "vp4i", "QUsQUiQsQiQfQPsUcUsUicsifPcPs">;
+def VLD2_LANE : WInst<"vld2_lane", "2(c*!)2I", "QUsQUiQsQiQfQPsUcUsUicsifPcPs">;
+def VLD3_LANE : WInst<"vld3_lane", "3(c*!)3I", "QUsQUiQsQiQfQPsUcUsUicsifPcPs">;
+def VLD4_LANE : WInst<"vld4_lane", "4(c*!)4I", "QUsQUiQsQiQfQPsUcUsUicsifPcPs">;
+def VST2 : WInst<"vst2", "v*(2!)", "QUcQUsQUiQcQsQiQfQPcQPsUcUsUiUlcsilfPcPs">;
+def VST3 : WInst<"vst3", "v*(3!)", "QUcQUsQUiQcQsQiQfQPcQPsUcUsUiUlcsilfPcPs">;
+def VST4 : WInst<"vst4", "v*(4!)", "QUcQUsQUiQcQsQiQfQPcQPsUcUsUiUlcsilfPcPs">;
+def VST2_LANE : WInst<"vst2_lane", "v*(2!)I", "QUsQUiQsQiQfQPsUcUsUicsifPcPs">;
+def VST3_LANE : WInst<"vst3_lane", "v*(3!)I", "QUsQUiQsQiQfQPsUcUsUicsifPcPs">;
+def VST4_LANE : WInst<"vst4_lane", "v*(4!)I", "QUsQUiQsQiQfQPsUcUsUicsifPcPs">;
 let ArchGuard = "(__ARM_FP & 2)" in {
-def VLD2_F16      : WInst<"vld2", "2c", "hQh">;
-def VLD3_F16      : WInst<"vld3", "3c", "hQh">;
-def VLD4_F16      : WInst<"vld4", "4c", "hQh">;
-def VLD2_DUP_F16  : WInst<"vld2_dup", "2c", "hQh">;
-def VLD3_DUP_F16  : WInst<"vld3_dup", "3c", "hQh">;
-def VLD4_DUP_F16  : WInst<"vld4_dup", "4c", "hQh">;
-def VLD2_LANE_F16 : WInst<"vld2_lane", "2c2i", "hQh">;
-def VLD3_LANE_F16 : WInst<"vld3_lane", "3c3i", "hQh">;
-def VLD4_LANE_F16 : WInst<"vld4_lane", "4c4i", "hQh">;
-def VST2_F16      : WInst<"vst2", "vp2", "hQh">;
-def VST3_F16      : WInst<"vst3", "vp3", "hQh">;
-def VST4_F16      : WInst<"vst4", "vp4", "hQh">;
-def VST2_LANE_F16 : WInst<"vst2_lane", "vp2i", "hQh">;
-def VST3_LANE_F16 : WInst<"vst3_lane", "vp3i", "hQh">;
-def VST4_LANE_F16 : WInst<"vst4_lane", "vp4i", "hQh">;
+def VLD2_F16      : WInst<"vld2", "2(c*!)", "hQh">;
+def VLD3_F16      : WInst<"vld3", "3(c*!)", "hQh">;
+def VLD4_F16      : WInst<"vld4", "4(c*!)", "hQh">;
+def VLD2_DUP_F16  : WInst<"vld2_dup", "2(c*!)", "hQh">;
+def VLD3_DUP_F16  : WInst<"vld3_dup", "3(c*!)", "hQh">;
+def VLD4_DUP_F16  : WInst<"vld4_dup", "4(c*!)", "hQh">;
+def VLD2_LANE_F16 : WInst<"vld2_lane", "2(c*!)2I", "hQh">;
+def VLD3_LANE_F16 : WInst<"vld3_lane", "3(c*!)3I", "hQh">;
+def VLD4_LANE_F16 : WInst<"vld4_lane", "4(c*!)4I", "hQh">;
+def VST2_F16      : WInst<"vst2", "v*(2!)", "hQh">;
+def VST3_F16      : WInst<"vst3", "v*(3!)", "hQh">;
+def VST4_F16      : WInst<"vst4", "v*(4!)", "hQh">;
+def VST2_LANE_F16 : WInst<"vst2_lane", "v*(2!)I", "hQh">;
+def VST3_LANE_F16 : WInst<"vst3_lane", "v*(3!)I", "hQh">;
+def VST4_LANE_F16 : WInst<"vst4_lane", "v*(4!)I", "hQh">;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.16 Extract lanes from a vector
 let InstName = "vmov" in
-def VGET_LANE : IInst<"vget_lane", "sdi",
+def VGET_LANE : IInst<"vget_lane", "1.I",
                       "UcUsUicsiPcPsfQUcQUsQUiQcQsQiQPcQPsQflUlQlQUl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.17 Set lanes within a vector
 let InstName = "vmov" in
-def VSET_LANE : IInst<"vset_lane", "dsdi",
+def VSET_LANE : IInst<"vset_lane", ".1.I",
                       "UcUsUicsiPcPsfQUcQUsQUiQcQsQiQPcQPsQflUlQlQUl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.18 Initialize a vector from bit pattern
-def VCREATE : NoTestOpInst<"vcreate", "dl", "csihfUcUsUiUlPcPsl", OP_CAST> {
+def VCREATE : NoTestOpInst<"vcreate", ".(IU>)", "csihfUcUsUiUlPcPsl", OP_CAST> {
   let BigEndianSafe = 1;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.19 Set all lanes to same value
 let InstName = "vmov" in {
-def VDUP_N   : WOpInst<"vdup_n", "ds",
+def VDUP_N   : WOpInst<"vdup_n", ".1",
                        "UcUsUicsiPcPshfQUcQUsQUiQcQsQiQPcQPsQhQflUlQlQUl",
                        OP_DUP>;
-def VMOV_N   : WOpInst<"vmov_n", "ds",
+def VMOV_N   : WOpInst<"vmov_n", ".1",
                        "UcUsUicsiPcPshfQUcQUsQUiQcQsQiQPcQPsQhQflUlQlQUl",
                        OP_DUP>;
 }
 let InstName = "" in
-def VDUP_LANE: WOpInst<"vdup_lane", "dgi",
+def VDUP_LANE: WOpInst<"vdup_lane", ".qI",
                        "UcUsUicsiPcPsfQUcQUsQUiQcQsQiQPcQPsQflUlQlQUl",
                        OP_DUP_LN>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.20 Combining vectors
-def VCOMBINE : NoTestOpInst<"vcombine", "kdd", "csilhfUcUsUiUlPcPs", OP_CONC>;
+def VCOMBINE : NoTestOpInst<"vcombine", "Q..", "csilhfUcUsUiUlPcPs", OP_CONC>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.21 Splitting vectors
 // Note that the ARM NEON Reference 2.0 mistakenly document the vget_high_f16()
 // and vget_low_f16() intrinsics as AArch64-only. We (and GCC) support all
 // versions of these intrinsics in both AArch32 and AArch64 architectures. See
 // D45668 for more details.
 let InstName = "vmov" in {
-def VGET_HIGH : NoTestOpInst<"vget_high", "dk", "csilhfUcUsUiUlPcPs", OP_HI>;
-def VGET_LOW  : NoTestOpInst<"vget_low", "dk", "csilhfUcUsUiUlPcPs", OP_LO>;
+def VGET_HIGH : NoTestOpInst<"vget_high", ".Q", "csilhfUcUsUiUlPcPs", OP_HI>;
+def VGET_LOW  : NoTestOpInst<"vget_low", ".Q", "csilhfUcUsUiUlPcPs", OP_LO>;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.22 Converting vectors
 
 let ArchGuard = "(__ARM_FP & 2)" in {
-  def VCVT_F16_F32 : SInst<"vcvt_f16_f32", "md", "Hf">;
-  def VCVT_F32_F16 : SInst<"vcvt_f32_f16", "wd", "h">;
+  def VCVT_F16_F32 : SInst<"vcvt_f16_f32", "(<q)(.!)", "Hf">;
+  def VCVT_F32_F16 : SInst<"vcvt_f32_f16", "(>Q)(.!)", "h">;
 }
 
-def VCVT_S32     : SInst<"vcvt_s32", "xd",  "fQf">;
-def VCVT_U32     : SInst<"vcvt_u32", "ud",  "fQf">;
-def VCVT_F32     : SInst<"vcvt_f32", "fd",  "iUiQiQUi">;
+def VCVT_S32     : SInst<"vcvt_s32", "S.",  "fQf">;
+def VCVT_U32     : SInst<"vcvt_u32", "U.",  "fQf">;
+def VCVT_F32     : SInst<"vcvt_f32", "F(.!)",  "iUiQiQUi">;
 let isVCVT_N = 1 in {
-def VCVT_N_S32   : SInst<"vcvt_n_s32", "xdi", "fQf">;
-def VCVT_N_U32   : SInst<"vcvt_n_u32", "udi", "fQf">;
-def VCVT_N_F32   : SInst<"vcvt_n_f32", "fdi", "iUiQiQUi">;
+def VCVT_N_S32   : SInst<"vcvt_n_s32", "S.I", "fQf">;
+def VCVT_N_U32   : SInst<"vcvt_n_u32", "U.I", "fQf">;
+def VCVT_N_F32   : SInst<"vcvt_n_f32", "F(.!)I", "iUiQiQUi">;
 }
 
-def VMOVN        : IInst<"vmovn", "hk",  "silUsUiUl">;
-def VMOVL        : SInst<"vmovl", "wd",  "csiUcUsUi">;
-def VQMOVN       : SInst<"vqmovn", "hk",  "silUsUiUl">;
-def VQMOVUN      : SInst<"vqmovun", "ek",  "sil">;
+def VMOVN        : IInst<"vmovn", "<Q",  "silUsUiUl">;
+def VMOVL        : SInst<"vmovl", "(>Q).",  "csiUcUsUi">;
+def VQMOVN       : SInst<"vqmovn", "<Q",  "silUsUiUl">;
+def VQMOVUN      : SInst<"vqmovun", "(<U)Q",  "sil">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.23-24 Table lookup, Extended table lookup
 let InstName = "vtbl" in {
-def VTBL1 : WInst<"vtbl1", "ddt",  "UccPc">;
-def VTBL2 : WInst<"vtbl2", "d2t",  "UccPc">;
-def VTBL3 : WInst<"vtbl3", "d3t",  "UccPc">;
-def VTBL4 : WInst<"vtbl4", "d4t",  "UccPc">;
+def VTBL1 : WInst<"vtbl1", "..p",  "UccPc">;
+def VTBL2 : WInst<"vtbl2", ".2p",  "UccPc">;
+def VTBL3 : WInst<"vtbl3", ".3p",  "UccPc">;
+def VTBL4 : WInst<"vtbl4", ".4p",  "UccPc">;
 }
 let InstName = "vtbx" in {
-def VTBX1 : WInst<"vtbx1", "dddt", "UccPc">;
-def VTBX2 : WInst<"vtbx2", "dd2t", "UccPc">;
-def VTBX3 : WInst<"vtbx3", "dd3t", "UccPc">;
-def VTBX4 : WInst<"vtbx4", "dd4t", "UccPc">;
+def VTBX1 : WInst<"vtbx1", "...p", "UccPc">;
+def VTBX2 : WInst<"vtbx2", "..2p", "UccPc">;
+def VTBX3 : WInst<"vtbx3", "..3p", "UccPc">;
+def VTBX4 : WInst<"vtbx4", "..4p", "UccPc">;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.25 Operations with a scalar value
-def VMLA_LANE     : IOpInst<"vmla_lane", "dddgi",
+def VMLA_LANE     : IOpInst<"vmla_lane", "...qI",
                             "siUsUifQsQiQUsQUiQf", OP_MLA_LN>;
-def VMLAL_LANE    : SOpInst<"vmlal_lane", "wwddi", "siUsUi", OP_MLAL_LN>;
-def VQDMLAL_LANE  : SOpInst<"vqdmlal_lane", "wwddi", "si", OP_QDMLAL_LN>;
-def VMLS_LANE     : IOpInst<"vmls_lane", "dddgi",
+def VMLAL_LANE    : SOpInst<"vmlal_lane", "(>Q)(>Q)..I", "siUsUi", OP_MLAL_LN>;
+def VQDMLAL_LANE  : SOpInst<"vqdmlal_lane", "(>Q)(>Q)..I", "si", OP_QDMLAL_LN>;
+def VMLS_LANE     : IOpInst<"vmls_lane", "...qI",
                             "siUsUifQsQiQUsQUiQf", OP_MLS_LN>;
-def VMLSL_LANE    : SOpInst<"vmlsl_lane", "wwddi", "siUsUi", OP_MLSL_LN>;
-def VQDMLSL_LANE  : SOpInst<"vqdmlsl_lane", "wwddi", "si", OP_QDMLSL_LN>;
-def VMUL_N        : IOpInst<"vmul_n", "dds", "sifUsUiQsQiQfQUsQUi", OP_MUL_N>;
-def VMUL_LANE     : IOpInst<"vmul_lane", "ddgi",
+def VMLSL_LANE    : SOpInst<"vmlsl_lane", "(>Q)(>Q)..I", "siUsUi", OP_MLSL_LN>;
+def VQDMLSL_LANE  : SOpInst<"vqdmlsl_lane", "(>Q)(>Q)..I", "si", OP_QDMLSL_LN>;
+def VMUL_N        : IOpInst<"vmul_n", "..1", "sifUsUiQsQiQfQUsQUi", OP_MUL_N>;
+def VMUL_LANE     : IOpInst<"vmul_lane", "..qI",
                             "sifUsUiQsQiQfQUsQUi", OP_MUL_LN>;
-def VMULL_N       : SOpInst<"vmull_n", "wds", "siUsUi", OP_MULL_N>;
-def VMULL_LANE    : SOpInst<"vmull_lane", "wddi", "siUsUi", OP_MULL_LN>;
-def VQDMULL_N     : SOpInst<"vqdmull_n", "wds", "si", OP_QDMULL_N>;
-def VQDMULL_LANE  : SOpInst<"vqdmull_lane", "wddi", "si", OP_QDMULL_LN>;
-def VQDMULH_N     : SOpInst<"vqdmulh_n", "dds", "siQsQi", OP_QDMULH_N>;
-def VQDMULH_LANE  : SOpInst<"vqdmulh_lane", "ddgi", "siQsQi", OP_QDMULH_LN>;
-def VQRDMULH_N    : SOpInst<"vqrdmulh_n", "dds", "siQsQi", OP_QRDMULH_N>;
-def VQRDMULH_LANE : SOpInst<"vqrdmulh_lane", "ddgi", "siQsQi", OP_QRDMULH_LN>;
+def VMULL_N       : SOpInst<"vmull_n", "(>Q).1", "siUsUi", OP_MULL_N>;
+def VMULL_LANE    : SOpInst<"vmull_lane", "(>Q)..I", "siUsUi", OP_MULL_LN>;
+def VQDMULL_N     : SOpInst<"vqdmull_n", "(>Q).1", "si", OP_QDMULL_N>;
+def VQDMULL_LANE  : SOpInst<"vqdmull_lane", "(>Q)..I", "si", OP_QDMULL_LN>;
+def VQDMULH_N     : SOpInst<"vqdmulh_n", "..1", "siQsQi", OP_QDMULH_N>;
+def VQDMULH_LANE  : SOpInst<"vqdmulh_lane", "..qI", "siQsQi", OP_QDMULH_LN>;
+def VQRDMULH_N    : SOpInst<"vqrdmulh_n", "..1", "siQsQi", OP_QRDMULH_N>;
+def VQRDMULH_LANE : SOpInst<"vqrdmulh_lane", "..qI", "siQsQi", OP_QRDMULH_LN>;
 
 let ArchGuard = "defined(__ARM_FEATURE_QRDMX)" in {
-def VQRDMLAH_LANE : SOpInst<"vqrdmlah_lane", "dddgi", "siQsQi", OP_QRDMLAH_LN>;
-def VQRDMLSH_LANE : SOpInst<"vqrdmlsh_lane", "dddgi", "siQsQi", OP_QRDMLSH_LN>;
+def VQRDMLAH_LANE : SOpInst<"vqrdmlah_lane", "...qI", "siQsQi", OP_QRDMLAH_LN>;
+def VQRDMLSH_LANE : SOpInst<"vqrdmlsh_lane", "...qI", "siQsQi", OP_QRDMLSH_LN>;
 }
 
-def VMLA_N        : IOpInst<"vmla_n", "ddds", "siUsUifQsQiQUsQUiQf", OP_MLA_N>;
-def VMLAL_N       : SOpInst<"vmlal_n", "wwds", "siUsUi", OP_MLAL_N>;
-def VQDMLAL_N     : SOpInst<"vqdmlal_n", "wwds", "si", OP_QDMLAL_N>;
-def VMLS_N        : IOpInst<"vmls_n", "ddds", "siUsUifQsQiQUsQUiQf", OP_MLS_N>;
-def VMLSL_N       : SOpInst<"vmlsl_n", "wwds", "siUsUi", OP_MLSL_N>;
-def VQDMLSL_N     : SOpInst<"vqdmlsl_n", "wwds", "si", OP_QDMLSL_N>;
+def VMLA_N        : IOpInst<"vmla_n", "...1", "siUsUifQsQiQUsQUiQf", OP_MLA_N>;
+def VMLAL_N       : SOpInst<"vmlal_n", "(>Q)(>Q).1", "siUsUi", OP_MLAL_N>;
+def VQDMLAL_N     : SOpInst<"vqdmlal_n", "(>Q)(>Q).1", "si", OP_QDMLAL_N>;
+def VMLS_N        : IOpInst<"vmls_n", "...1", "siUsUifQsQiQUsQUiQf", OP_MLS_N>;
+def VMLSL_N       : SOpInst<"vmlsl_n", "(>Q)(>Q).1", "siUsUi", OP_MLSL_N>;
+def VQDMLSL_N     : SOpInst<"vqdmlsl_n", "(>Q)(>Q).1", "si", OP_QDMLSL_N>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.26 Vector Extract
-def VEXT : WInst<"vext", "dddi",
+def VEXT : WInst<"vext", "...I",
                  "cUcPcsUsPsiUilUlfQcQUcQPcQsQUsQPsQiQUiQlQUlQf">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.27 Reverse vector elements
-def VREV64 : WOpInst<"vrev64", "dd", "csiUcUsUiPcPsfQcQsQiQUcQUsQUiQPcQPsQf",
+def VREV64 : WOpInst<"vrev64", "..", "csiUcUsUiPcPsfQcQsQiQUcQUsQUiQPcQPsQf",
                   OP_REV64>;
-def VREV32 : WOpInst<"vrev32", "dd", "csUcUsPcPsQcQsQUcQUsQPcQPs", OP_REV32>;
-def VREV16 : WOpInst<"vrev16", "dd", "cUcPcQcQUcQPc", OP_REV16>;
+def VREV32 : WOpInst<"vrev32", "..", "csUcUsPcPsQcQsQUcQUsQPcQPs", OP_REV32>;
+def VREV16 : WOpInst<"vrev16", "..", "cUcPcQcQUcQPc", OP_REV16>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.28 Other single operand arithmetic
-def VABS    : SInst<"vabs", "dd", "csifQcQsQiQf">;
-def VQABS   : SInst<"vqabs", "dd", "csiQcQsQi">;
-def VNEG    : SOpInst<"vneg", "dd", "csifQcQsQiQf", OP_NEG>;
-def VQNEG   : SInst<"vqneg", "dd", "csiQcQsQi">;
-def VCLS    : SInst<"vcls", "dd", "csiQcQsQi">;
-def VCLZ    : IInst<"vclz", "dd", "csiUcUsUiQcQsQiQUcQUsQUi">;
-def VCNT    : WInst<"vcnt", "dd", "UccPcQUcQcQPc">;
-def VRECPE  : SInst<"vrecpe", "dd", "fUiQfQUi">;
-def VRSQRTE : SInst<"vrsqrte", "dd", "fUiQfQUi">;
+def VABS    : SInst<"vabs", "..", "csifQcQsQiQf">;
+def VQABS   : SInst<"vqabs", "..", "csiQcQsQi">;
+def VNEG    : SOpInst<"vneg", "..", "csifQcQsQiQf", OP_NEG>;
+def VQNEG   : SInst<"vqneg", "..", "csiQcQsQi">;
+def VCLS    : SInst<"vcls", "..", "csiQcQsQi">;
+def VCLZ    : IInst<"vclz", "..", "csiUcUsUiQcQsQiQUcQUsQUi">;
+def VCNT    : WInst<"vcnt", "..", "UccPcQUcQcQPc">;
+def VRECPE  : SInst<"vrecpe", "..", "fUiQfQUi">;
+def VRSQRTE : SInst<"vrsqrte", "..", "fUiQfQUi">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.29 Logical operations
-def VMVN : LOpInst<"vmvn", "dd", "csiUcUsUiPcQcQsQiQUcQUsQUiQPc", OP_NOT>;
-def VAND : LOpInst<"vand", "ddd", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl", OP_AND>;
-def VORR : LOpInst<"vorr", "ddd", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl", OP_OR>;
-def VEOR : LOpInst<"veor", "ddd", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl", OP_XOR>;
-def VBIC : LOpInst<"vbic", "ddd", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl", OP_ANDN>;
-def VORN : LOpInst<"vorn", "ddd", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl", OP_ORN>;
+def VMVN : LOpInst<"vmvn", "..", "csiUcUsUiPcQcQsQiQUcQUsQUiQPc", OP_NOT>;
+def VAND : LOpInst<"vand", "...", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl", OP_AND>;
+def VORR : LOpInst<"vorr", "...", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl", OP_OR>;
+def VEOR : LOpInst<"veor", "...", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl", OP_XOR>;
+def VBIC : LOpInst<"vbic", "...", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl", OP_ANDN>;
+def VORN : LOpInst<"vorn", "...", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl", OP_ORN>;
 let isHiddenLInst = 1 in
-def VBSL : SInst<"vbsl", "dudd",
+def VBSL : SInst<"vbsl", ".U..",
                 "csilUcUsUiUlfPcPsQcQsQiQlQUcQUsQUiQUlQfQPcQPs">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.30 Transposition operations
-def VTRN : WInst<"vtrn", "2dd", "csiUcUsUifPcPsQcQsQiQUcQUsQUiQfQPcQPs">;
-def VZIP : WInst<"vzip", "2dd", "csiUcUsUifPcPsQcQsQiQUcQUsQUiQfQPcQPs">;
-def VUZP : WInst<"vuzp", "2dd", "csiUcUsUifPcPsQcQsQiQUcQUsQUiQfQPcQPs">;
+def VTRN : WInst<"vtrn", "2..", "csiUcUsUifPcPsQcQsQiQUcQUsQUiQfQPcQPs">;
+def VZIP : WInst<"vzip", "2..", "csiUcUsUifPcPsQcQsQiQUcQUsQUiQfQPcQPs">;
+def VUZP : WInst<"vuzp", "2..", "csiUcUsUifPcPsQcQsQiQUcQUsQUiQfQPcQPs">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.31 Vector reinterpret cast operations
 def VREINTERPRET
-  : NoTestOpInst<"vreinterpret", "dd",
+  : NoTestOpInst<"vreinterpret", "..",
          "csilUcUsUiUlhfPcPsQcQsQiQlQUcQUsQUiQUlQhQfQPcQPs", OP_REINT> {
   let CartesianProductOfTypes = 1;
   let ArchGuard = "!defined(__aarch64__)";
   let BigEndianSafe = 1;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Vector fused multiply-add operations
 
 let ArchGuard = "defined(__ARM_FEATURE_FMA)" in {
-  def VFMA : SInst<"vfma", "dddd", "fQf">;
-  def VFMS : SOpInst<"vfms", "dddd", "fQf", OP_FMLS>;
-  def FMLA_N_F32 : SOpInst<"vfma_n", "ddds", "fQf", OP_FMLA_N>;
+  def VFMA : SInst<"vfma", "....", "fQf">;
+  def VFMS : SOpInst<"vfms", "....", "fQf", OP_FMLS>;
+  def FMLA_N_F32 : SOpInst<"vfma_n", "...1", "fQf", OP_FMLA_N>;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // fp16 vector operations
-def SCALAR_HALF_GET_LANE : IOpInst<"vget_lane", "sdi", "h", OP_SCALAR_HALF_GET_LN>;
-def SCALAR_HALF_SET_LANE : IOpInst<"vset_lane", "dsdi", "h", OP_SCALAR_HALF_SET_LN>;
-def SCALAR_HALF_GET_LANEQ : IOpInst<"vget_lane", "sdi", "Qh", OP_SCALAR_HALF_GET_LNQ>;
-def SCALAR_HALF_SET_LANEQ : IOpInst<"vset_lane", "dsdi", "Qh", OP_SCALAR_HALF_SET_LNQ>;
+def SCALAR_HALF_GET_LANE : IOpInst<"vget_lane", "1.I", "h", OP_SCALAR_HALF_GET_LN>;
+def SCALAR_HALF_SET_LANE : IOpInst<"vset_lane", ".1.I", "h", OP_SCALAR_HALF_SET_LN>;
+def SCALAR_HALF_GET_LANEQ : IOpInst<"vget_lane", "1.I", "Qh", OP_SCALAR_HALF_GET_LNQ>;
+def SCALAR_HALF_SET_LANEQ : IOpInst<"vset_lane", ".1.I", "Qh", OP_SCALAR_HALF_SET_LNQ>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // AArch64 Intrinsics
 
 let ArchGuard = "defined(__aarch64__)" in {
 
 ////////////////////////////////////////////////////////////////////////////////
 // Load/Store
-def LD1 : WInst<"vld1", "dc", "dQdPlQPl">;
-def LD2 : WInst<"vld2", "2c", "QUlQldQdPlQPl">;
-def LD3 : WInst<"vld3", "3c", "QUlQldQdPlQPl">;
-def LD4 : WInst<"vld4", "4c", "QUlQldQdPlQPl">;
-def ST1 : WInst<"vst1", "vpd", "dQdPlQPl">;
-def ST2 : WInst<"vst2", "vp2", "QUlQldQdPlQPl">;
-def ST3 : WInst<"vst3", "vp3", "QUlQldQdPlQPl">;
-def ST4 : WInst<"vst4", "vp4", "QUlQldQdPlQPl">;
+def LD1 : WInst<"vld1", ".(c*!)", "dQdPlQPl">;
+def LD2 : WInst<"vld2", "2(c*!)", "QUlQldQdPlQPl">;
+def LD3 : WInst<"vld3", "3(c*!)", "QUlQldQdPlQPl">;
+def LD4 : WInst<"vld4", "4(c*!)", "QUlQldQdPlQPl">;
+def ST1 : WInst<"vst1", "v*(.!)", "dQdPlQPl">;
+def ST2 : WInst<"vst2", "v*(2!)", "QUlQldQdPlQPl">;
+def ST3 : WInst<"vst3", "v*(3!)", "QUlQldQdPlQPl">;
+def ST4 : WInst<"vst4", "v*(4!)", "QUlQldQdPlQPl">;
 
-def LD1_X2 : WInst<"vld1_x2", "2c",
+def LD1_X2 : WInst<"vld1_x2", "2(c*!)",
                    "dQdPlQPl">;
-def LD1_X3 : WInst<"vld1_x3", "3c",
+def LD1_X3 : WInst<"vld1_x3", "3(c*!)",
                    "dQdPlQPl">;
-def LD1_X4 : WInst<"vld1_x4", "4c",
+def LD1_X4 : WInst<"vld1_x4", "4(c*!)",
                    "dQdPlQPl">;
 
-def ST1_X2 : WInst<"vst1_x2", "vp2", "dQdPlQPl">;
-def ST1_X3 : WInst<"vst1_x3", "vp3", "dQdPlQPl">;
-def ST1_X4 : WInst<"vst1_x4", "vp4", "dQdPlQPl">;
-
-def LD1_LANE : WInst<"vld1_lane", "dcdi", "dQdPlQPl">;
-def LD2_LANE : WInst<"vld2_lane", "2c2i", "lUlQcQUcQPcQlQUldQdPlQPl">;
-def LD3_LANE : WInst<"vld3_lane", "3c3i", "lUlQcQUcQPcQlQUldQdPlQPl">;
-def LD4_LANE : WInst<"vld4_lane", "4c4i", "lUlQcQUcQPcQlQUldQdPlQPl">;
-def ST1_LANE : WInst<"vst1_lane", "vpdi", "dQdPlQPl">;
-def ST2_LANE : WInst<"vst2_lane", "vp2i", "lUlQcQUcQPcQlQUldQdPlQPl">;
-def ST3_LANE : WInst<"vst3_lane", "vp3i", "lUlQcQUcQPcQlQUldQdPlQPl">;
-def ST4_LANE : WInst<"vst4_lane", "vp4i", "lUlQcQUcQPcQlQUldQdPlQPl">;
-
-def LD1_DUP  : WInst<"vld1_dup", "dc", "dQdPlQPl">;
-def LD2_DUP  : WInst<"vld2_dup", "2c", "dQdPlQPl">;
-def LD3_DUP  : WInst<"vld3_dup", "3c", "dQdPlQPl">;
-def LD4_DUP  : WInst<"vld4_dup", "4c", "dQdPlQPl">;
+def ST1_X2 : WInst<"vst1_x2", "v*(2!)", "dQdPlQPl">;
+def ST1_X3 : WInst<"vst1_x3", "v*(3!)", "dQdPlQPl">;
+def ST1_X4 : WInst<"vst1_x4", "v*(4!)", "dQdPlQPl">;
+
+def LD1_LANE : WInst<"vld1_lane", ".(c*!).I", "dQdPlQPl">;
+def LD2_LANE : WInst<"vld2_lane", "2(c*!)2I", "lUlQcQUcQPcQlQUldQdPlQPl">;
+def LD3_LANE : WInst<"vld3_lane", "3(c*!)3I", "lUlQcQUcQPcQlQUldQdPlQPl">;
+def LD4_LANE : WInst<"vld4_lane", "4(c*!)4I", "lUlQcQUcQPcQlQUldQdPlQPl">;
+def ST1_LANE : WInst<"vst1_lane", "v*(.!)I", "dQdPlQPl">;
+def ST2_LANE : WInst<"vst2_lane", "v*(2!)I", "lUlQcQUcQPcQlQUldQdPlQPl">;
+def ST3_LANE : WInst<"vst3_lane", "v*(3!)I", "lUlQcQUcQPcQlQUldQdPlQPl">;
+def ST4_LANE : WInst<"vst4_lane", "v*(4!)I", "lUlQcQUcQPcQlQUldQdPlQPl">;
+
+def LD1_DUP  : WInst<"vld1_dup", ".(c*!)", "dQdPlQPl">;
+def LD2_DUP  : WInst<"vld2_dup", "2(c*!)", "dQdPlQPl">;
+def LD3_DUP  : WInst<"vld3_dup", "3(c*!)", "dQdPlQPl">;
+def LD4_DUP  : WInst<"vld4_dup", "4(c*!)", "dQdPlQPl">;
 
-def VLDRQ : WInst<"vldrq", "sc", "Pk">;
-def VSTRQ : WInst<"vstrq", "vps", "Pk">;
+def VLDRQ : WInst<"vldrq", "1(c*!)", "Pk">;
+def VSTRQ : WInst<"vstrq", "v*(1!)", "Pk">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Addition
-def ADD : IOpInst<"vadd", "ddd", "dQd", OP_ADD>;
+def ADD : IOpInst<"vadd", "...", "dQd", OP_ADD>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Subtraction
-def SUB : IOpInst<"vsub", "ddd", "dQd", OP_SUB>;
+def SUB : IOpInst<"vsub", "...", "dQd", OP_SUB>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Multiplication
-def MUL     : IOpInst<"vmul", "ddd", "dQd", OP_MUL>;
-def MLA     : IOpInst<"vmla", "dddd", "dQd", OP_MLA>;
-def MLS     : IOpInst<"vmls", "dddd", "dQd", OP_MLS>;
+def MUL     : IOpInst<"vmul", "...", "dQd", OP_MUL>;
+def MLA     : IOpInst<"vmla", "....", "dQd", OP_MLA>;
+def MLS     : IOpInst<"vmls", "....", "dQd", OP_MLS>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Multiplication Extended
-def MULX : SInst<"vmulx", "ddd", "fdQfQd">;
+def MULX : SInst<"vmulx", "...", "fdQfQd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Division
-def FDIV : IOpInst<"vdiv", "ddd",  "fdQfQd", OP_DIV>;
+def FDIV : IOpInst<"vdiv", "...",  "fdQfQd", OP_DIV>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Vector fused multiply-add operations
-def FMLA : SInst<"vfma", "dddd", "dQd">;
-def FMLS : SOpInst<"vfms", "dddd", "dQd", OP_FMLS>;
+def FMLA : SInst<"vfma", "....", "dQd">;
+def FMLS : SOpInst<"vfms", "....", "dQd", OP_FMLS>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // MUL, MLA, MLS, FMA, FMS definitions with scalar argument
-def VMUL_N_A64 : IOpInst<"vmul_n", "dds", "Qd", OP_MUL_N>;
+def VMUL_N_A64 : IOpInst<"vmul_n", "..1", "Qd", OP_MUL_N>;
 
-def FMLA_N : SOpInst<"vfma_n", "ddds", "dQd", OP_FMLA_N>;
-def FMLS_N : SOpInst<"vfms_n", "ddds", "fdQfQd", OP_FMLS_N>;
+def FMLA_N : SOpInst<"vfma_n", "...1", "dQd", OP_FMLA_N>;
+def FMLS_N : SOpInst<"vfms_n", "...1", "fdQfQd", OP_FMLS_N>;
 
-def MLA_N : SOpInst<"vmla_n", "ddds", "Qd", OP_MLA_N>;
-def MLS_N : SOpInst<"vmls_n", "ddds", "Qd", OP_MLS_N>;
+def MLA_N : SOpInst<"vmla_n", "...1", "Qd", OP_MLA_N>;
+def MLS_N : SOpInst<"vmls_n", "...1", "Qd", OP_MLS_N>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Logical operations
-def BSL : SInst<"vbsl", "dudd", "dPlQdQPl">;
+def BSL : SInst<"vbsl", ".U..", "dPlQdQPl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Absolute Difference
-def ABD  : SInst<"vabd", "ddd",  "dQd">;
+def ABD  : SInst<"vabd", "...",  "dQd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // saturating absolute/negate
-def ABS    : SInst<"vabs", "dd", "dQdlQl">;
-def QABS   : SInst<"vqabs", "dd", "lQl">;
-def NEG    : SOpInst<"vneg", "dd", "dlQdQl", OP_NEG>;
-def QNEG   : SInst<"vqneg", "dd", "lQl">;
+def ABS    : SInst<"vabs", "..", "dQdlQl">;
+def QABS   : SInst<"vqabs", "..", "lQl">;
+def NEG    : SOpInst<"vneg", "..", "dlQdQl", OP_NEG>;
+def QNEG   : SInst<"vqneg", "..", "lQl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Signed Saturating Accumulated of Unsigned Value
-def SUQADD : SInst<"vuqadd", "ddu", "csilQcQsQiQl">;
+def SUQADD : SInst<"vuqadd", "..U", "csilQcQsQiQl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Unsigned Saturating Accumulated of Signed Value
-def USQADD : SInst<"vsqadd", "ddx", "UcUsUiUlQUcQUsQUiQUl">;
+def USQADD : SInst<"vsqadd", "..S", "UcUsUiUlQUcQUsQUiQUl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Reciprocal/Sqrt
-def FRECPS  : IInst<"vrecps", "ddd", "dQd">;
-def FRSQRTS : IInst<"vrsqrts", "ddd", "dQd">;
-def FRECPE  : SInst<"vrecpe", "dd", "dQd">;
-def FRSQRTE : SInst<"vrsqrte", "dd", "dQd">;
-def FSQRT   : SInst<"vsqrt", "dd", "fdQfQd">;
+def FRECPS  : IInst<"vrecps", "...", "dQd">;
+def FRSQRTS : IInst<"vrsqrts", "...", "dQd">;
+def FRECPE  : SInst<"vrecpe", "..", "dQd">;
+def FRSQRTE : SInst<"vrsqrte", "..", "dQd">;
+def FSQRT   : SInst<"vsqrt", "..", "fdQfQd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // bitwise reverse
-def RBIT : IInst<"vrbit", "dd", "cUcPcQcQUcQPc">;
+def RBIT : IInst<"vrbit", "..", "cUcPcQcQUcQPc">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Integer extract and narrow to high
-def XTN2 : SOpInst<"vmovn_high", "qhk", "silUsUiUl", OP_XTN>;
+def XTN2 : SOpInst<"vmovn_high", "(<Q)<Q", "silUsUiUl", OP_XTN>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Signed integer saturating extract and unsigned narrow to high
-def SQXTUN2 : SOpInst<"vqmovun_high", "emd", "HsHiHl", OP_SQXTUN>;
+def SQXTUN2 : SOpInst<"vqmovun_high", "(<U)(<q).", "HsHiHl", OP_SQXTUN>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Integer saturating extract and narrow to high
-def QXTN2 : SOpInst<"vqmovn_high", "qhk", "silUsUiUl", OP_QXTN>;
+def QXTN2 : SOpInst<"vqmovn_high", "(<Q)<Q", "silUsUiUl", OP_QXTN>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Converting vectors
 
-def VCVT_F32_F64 : SInst<"vcvt_f32_f64", "md", "Qd">;
-def VCVT_F64_F32 : SInst<"vcvt_f64_f32", "wd", "f">;
+def VCVT_F32_F64 : SInst<"vcvt_f32_f64", "(<q).", "Qd">;
+def VCVT_F64_F32 : SInst<"vcvt_f64_f32", "(>Q).", "f">;
 
-def VCVT_S64 : SInst<"vcvt_s64", "xd",  "dQd">;
-def VCVT_U64 : SInst<"vcvt_u64", "ud",  "dQd">;
-def VCVT_F64 : SInst<"vcvt_f64", "Fd",  "lUlQlQUl">;
-
-def VCVT_HIGH_F16_F32 : SOpInst<"vcvt_high_f16", "hmj", "Hf", OP_VCVT_NA_HI_F16>;
-def VCVT_HIGH_F32_F16 : SOpInst<"vcvt_high_f32", "wk", "h", OP_VCVT_EX_HI_F32>;
-def VCVT_HIGH_F32_F64 : SOpInst<"vcvt_high_f32", "qfj", "d", OP_VCVT_NA_HI_F32>;
-def VCVT_HIGH_F64_F32 : SOpInst<"vcvt_high_f64", "wj", "f", OP_VCVT_EX_HI_F64>;
+def VCVT_S64 : SInst<"vcvt_s64", "S.",  "dQd">;
+def VCVT_U64 : SInst<"vcvt_u64", "U.",  "dQd">;
+def VCVT_F64 : SInst<"vcvt_f64", "F(.!)",  "lUlQlQUl">;
+
+def VCVT_HIGH_F16_F32 : SOpInst<"vcvt_high_f16", "<(<q!)Q", "Hf", OP_VCVT_NA_HI_F16>;
+def VCVT_HIGH_F32_F16 : SOpInst<"vcvt_high_f32", "(>Q)(Q!)", "h", OP_VCVT_EX_HI_F32>;
+def VCVT_HIGH_F32_F64 : SOpInst<"vcvt_high_f32", "(<Q)(F<!)Q", "d", OP_VCVT_NA_HI_F32>;
+def VCVT_HIGH_F64_F32 : SOpInst<"vcvt_high_f64", "(>Q)(Q!)", "f", OP_VCVT_EX_HI_F64>;
 
-def VCVTX_F32_F64      : SInst<"vcvtx_f32", "fj",  "d">;
-def VCVTX_HIGH_F32_F64 : SOpInst<"vcvtx_high_f32", "qfj", "d", OP_VCVTX_HI>;
+def VCVTX_F32_F64      : SInst<"vcvtx_f32", "(F<)(Q!)",  "d">;
+def VCVTX_HIGH_F32_F64 : SOpInst<"vcvtx_high_f32", "(<Q)(F<!)Q", "d", OP_VCVTX_HI>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Comparison
-def FCAGE : IInst<"vcage", "udd", "dQd">;
-def FCAGT : IInst<"vcagt", "udd", "dQd">;
-def FCALE : IInst<"vcale", "udd", "dQd">;
-def FCALT : IInst<"vcalt", "udd", "dQd">;
-def CMTST  : WInst<"vtst", "udd", "lUlPlQlQUlQPl">;
-def CFMEQ  : SOpInst<"vceq", "udd", "lUldQdQlQUlPlQPl", OP_EQ>;
-def CFMGE  : SOpInst<"vcge", "udd", "lUldQdQlQUl", OP_GE>;
-def CFMLE  : SOpInst<"vcle", "udd", "lUldQdQlQUl", OP_LE>;
-def CFMGT  : SOpInst<"vcgt", "udd", "lUldQdQlQUl", OP_GT>;
-def CFMLT  : SOpInst<"vclt", "udd", "lUldQdQlQUl", OP_LT>;
+def FCAGE : IInst<"vcage", "U..", "dQd">;
+def FCAGT : IInst<"vcagt", "U..", "dQd">;
+def FCALE : IInst<"vcale", "U..", "dQd">;
+def FCALT : IInst<"vcalt", "U..", "dQd">;
+def CMTST  : WInst<"vtst", "U..", "lUlPlQlQUlQPl">;
+def CFMEQ  : SOpInst<"vceq", "U..", "lUldQdQlQUlPlQPl", OP_EQ>;
+def CFMGE  : SOpInst<"vcge", "U..", "lUldQdQlQUl", OP_GE>;
+def CFMLE  : SOpInst<"vcle", "U..", "lUldQdQlQUl", OP_LE>;
+def CFMGT  : SOpInst<"vcgt", "U..", "lUldQdQlQUl", OP_GT>;
+def CFMLT  : SOpInst<"vclt", "U..", "lUldQdQlQUl", OP_LT>;
 
-def CMEQ  : SInst<"vceqz", "ud",
+def CMEQ  : SInst<"vceqz", "U.",
                   "csilfUcUsUiUlPcPsPlQcQsQiQlQfQUcQUsQUiQUlQPcQPsdQdQPl">;
-def CMGE  : SInst<"vcgez", "ud", "csilfdQcQsQiQlQfQd">;
-def CMLE  : SInst<"vclez", "ud", "csilfdQcQsQiQlQfQd">;
-def CMGT  : SInst<"vcgtz", "ud", "csilfdQcQsQiQlQfQd">;
-def CMLT  : SInst<"vcltz", "ud", "csilfdQcQsQiQlQfQd">;
+def CMGE  : SInst<"vcgez", "U.", "csilfdQcQsQiQlQfQd">;
+def CMLE  : SInst<"vclez", "U.", "csilfdQcQsQiQlQfQd">;
+def CMGT  : SInst<"vcgtz", "U.", "csilfdQcQsQiQlQfQd">;
+def CMLT  : SInst<"vcltz", "U.", "csilfdQcQsQiQlQfQd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Max/Min Integer
-def MAX : SInst<"vmax", "ddd", "dQd">;
-def MIN : SInst<"vmin", "ddd", "dQd">;
+def MAX : SInst<"vmax", "...", "dQd">;
+def MIN : SInst<"vmin", "...", "dQd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Pairwise Max/Min
-def MAXP : SInst<"vpmax", "ddd", "QcQsQiQUcQUsQUiQfQd">;
-def MINP : SInst<"vpmin", "ddd", "QcQsQiQUcQUsQUiQfQd">;
+def MAXP : SInst<"vpmax", "...", "QcQsQiQUcQUsQUiQfQd">;
+def MINP : SInst<"vpmin", "...", "QcQsQiQUcQUsQUiQfQd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Pairwise MaxNum/MinNum Floating Point
-def FMAXNMP : SInst<"vpmaxnm", "ddd", "fQfQd">;
-def FMINNMP : SInst<"vpminnm", "ddd", "fQfQd">;
+def FMAXNMP : SInst<"vpmaxnm", "...", "fQfQd">;
+def FMINNMP : SInst<"vpminnm", "...", "fQfQd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Pairwise Addition
-def ADDP  : IInst<"vpadd", "ddd", "QcQsQiQlQUcQUsQUiQUlQfQd">;
+def ADDP  : IInst<"vpadd", "...", "QcQsQiQlQUcQUsQUiQUlQfQd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Shifts by constant
 let isShift = 1 in {
 // Left shift long high
-def SHLL_HIGH_N    : SOpInst<"vshll_high_n", "ndi", "HcHsHiHUcHUsHUi",
+def SHLL_HIGH_N    : SOpInst<"vshll_high_n", ">.I", "HcHsHiHUcHUsHUi",
                              OP_LONG_HI>;
 
 ////////////////////////////////////////////////////////////////////////////////
-def SRI_N : WInst<"vsri_n", "dddi", "PlQPl">;
-def SLI_N : WInst<"vsli_n", "dddi", "PlQPl">;
+def SRI_N : WInst<"vsri_n", "...I", "PlQPl">;
+def SLI_N : WInst<"vsli_n", "...I", "PlQPl">;
 
 // Right shift narrow high
-def SHRN_HIGH_N    : IOpInst<"vshrn_high_n", "hmdi",
+def SHRN_HIGH_N    : IOpInst<"vshrn_high_n", "<(<q).I",
                              "HsHiHlHUsHUiHUl", OP_NARROW_HI>;
-def QSHRUN_HIGH_N  : SOpInst<"vqshrun_high_n", "hmdi",
+def QSHRUN_HIGH_N  : SOpInst<"vqshrun_high_n", "<(<q).I",
                              "HsHiHl", OP_NARROW_HI>;
-def RSHRN_HIGH_N   : IOpInst<"vrshrn_high_n", "hmdi",
+def RSHRN_HIGH_N   : IOpInst<"vrshrn_high_n", "<(<q).I",
                              "HsHiHlHUsHUiHUl", OP_NARROW_HI>;
-def QRSHRUN_HIGH_N : SOpInst<"vqrshrun_high_n", "hmdi",
+def QRSHRUN_HIGH_N : SOpInst<"vqrshrun_high_n", "<(<q).I",
                              "HsHiHl", OP_NARROW_HI>;
-def QSHRN_HIGH_N   : SOpInst<"vqshrn_high_n", "hmdi",
+def QSHRN_HIGH_N   : SOpInst<"vqshrn_high_n", "<(<q).I",
                              "HsHiHlHUsHUiHUl", OP_NARROW_HI>;
-def QRSHRN_HIGH_N  : SOpInst<"vqrshrn_high_n", "hmdi",
+def QRSHRN_HIGH_N  : SOpInst<"vqrshrn_high_n", "<(<q).I",
                              "HsHiHlHUsHUiHUl", OP_NARROW_HI>;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Converting vectors
-def VMOVL_HIGH   : SOpInst<"vmovl_high", "nd", "HcHsHiHUcHUsHUi", OP_MOVL_HI>;
+def VMOVL_HIGH   : SOpInst<"vmovl_high", ">.", "HcHsHiHUcHUsHUi", OP_MOVL_HI>;
 
 let isVCVT_N = 1 in {
-def CVTF_N_F64   : SInst<"vcvt_n_f64", "Fdi", "lUlQlQUl">;
-def FCVTZS_N_S64 : SInst<"vcvt_n_s64", "xdi", "dQd">;
-def FCVTZS_N_U64 : SInst<"vcvt_n_u64", "udi", "dQd">;
+def CVTF_N_F64   : SInst<"vcvt_n_f64", "F(.!)I", "lUlQlQUl">;
+def FCVTZS_N_S64 : SInst<"vcvt_n_s64", "S.I", "dQd">;
+def FCVTZS_N_U64 : SInst<"vcvt_n_u64", "U.I", "dQd">;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // 3VDiff class using high 64-bit in operands
-def VADDL_HIGH   : SOpInst<"vaddl_high", "wkk", "csiUcUsUi", OP_ADDLHi>;
-def VADDW_HIGH   : SOpInst<"vaddw_high", "wwk", "csiUcUsUi", OP_ADDWHi>;
-def VSUBL_HIGH   : SOpInst<"vsubl_high", "wkk", "csiUcUsUi", OP_SUBLHi>;
-def VSUBW_HIGH   : SOpInst<"vsubw_high", "wwk", "csiUcUsUi", OP_SUBWHi>;
-
-def VABDL_HIGH   : SOpInst<"vabdl_high", "wkk",  "csiUcUsUi", OP_ABDLHi>;
-def VABAL_HIGH   : SOpInst<"vabal_high", "wwkk", "csiUcUsUi", OP_ABALHi>;
-
-def VMULL_HIGH   : SOpInst<"vmull_high", "wkk", "csiUcUsUiPc", OP_MULLHi>;
-def VMULL_HIGH_N : SOpInst<"vmull_high_n", "wks", "siUsUi", OP_MULLHi_N>;
-def VMLAL_HIGH   : SOpInst<"vmlal_high", "wwkk", "csiUcUsUi", OP_MLALHi>;
-def VMLAL_HIGH_N : SOpInst<"vmlal_high_n", "wwks", "siUsUi", OP_MLALHi_N>;
-def VMLSL_HIGH   : SOpInst<"vmlsl_high", "wwkk", "csiUcUsUi", OP_MLSLHi>;
-def VMLSL_HIGH_N : SOpInst<"vmlsl_high_n", "wwks", "siUsUi", OP_MLSLHi_N>;
-
-def VADDHN_HIGH  : SOpInst<"vaddhn_high", "qhkk", "silUsUiUl", OP_ADDHNHi>;
-def VRADDHN_HIGH : SOpInst<"vraddhn_high", "qhkk", "silUsUiUl", OP_RADDHNHi>;
-def VSUBHN_HIGH  : SOpInst<"vsubhn_high", "qhkk", "silUsUiUl", OP_SUBHNHi>;
-def VRSUBHN_HIGH : SOpInst<"vrsubhn_high", "qhkk", "silUsUiUl", OP_RSUBHNHi>;
-
-def VQDMULL_HIGH : SOpInst<"vqdmull_high", "wkk", "si", OP_QDMULLHi>;
-def VQDMULL_HIGH_N : SOpInst<"vqdmull_high_n", "wks", "si", OP_QDMULLHi_N>;
-def VQDMLAL_HIGH : SOpInst<"vqdmlal_high", "wwkk", "si", OP_QDMLALHi>;
-def VQDMLAL_HIGH_N : SOpInst<"vqdmlal_high_n", "wwks", "si", OP_QDMLALHi_N>;
-def VQDMLSL_HIGH : SOpInst<"vqdmlsl_high", "wwkk", "si", OP_QDMLSLHi>;
-def VQDMLSL_HIGH_N : SOpInst<"vqdmlsl_high_n", "wwks", "si", OP_QDMLSLHi_N>;
-def VMULL_P64    : SInst<"vmull", "rss", "Pl">;
-def VMULL_HIGH_P64 : SOpInst<"vmull_high", "rdd", "HPl", OP_MULLHi_P64>;
+def VADDL_HIGH   : SOpInst<"vaddl_high", "(>Q)QQ", "csiUcUsUi", OP_ADDLHi>;
+def VADDW_HIGH   : SOpInst<"vaddw_high", "(>Q)(>Q)Q", "csiUcUsUi", OP_ADDWHi>;
+def VSUBL_HIGH   : SOpInst<"vsubl_high", "(>Q)QQ", "csiUcUsUi", OP_SUBLHi>;
+def VSUBW_HIGH   : SOpInst<"vsubw_high", "(>Q)(>Q)Q", "csiUcUsUi", OP_SUBWHi>;
+
+def VABDL_HIGH   : SOpInst<"vabdl_high", "(>Q)QQ",  "csiUcUsUi", OP_ABDLHi>;
+def VABAL_HIGH   : SOpInst<"vabal_high", "(>Q)(>Q)QQ", "csiUcUsUi", OP_ABALHi>;
+
+def VMULL_HIGH   : SOpInst<"vmull_high", "(>Q)QQ", "csiUcUsUiPc", OP_MULLHi>;
+def VMULL_HIGH_N : SOpInst<"vmull_high_n", "(>Q)Q1", "siUsUi", OP_MULLHi_N>;
+def VMLAL_HIGH   : SOpInst<"vmlal_high", "(>Q)(>Q)QQ", "csiUcUsUi", OP_MLALHi>;
+def VMLAL_HIGH_N : SOpInst<"vmlal_high_n", "(>Q)(>Q)Q1", "siUsUi", OP_MLALHi_N>;
+def VMLSL_HIGH   : SOpInst<"vmlsl_high", "(>Q)(>Q)QQ", "csiUcUsUi", OP_MLSLHi>;
+def VMLSL_HIGH_N : SOpInst<"vmlsl_high_n", "(>Q)(>Q)Q1", "siUsUi", OP_MLSLHi_N>;
+
+def VADDHN_HIGH  : SOpInst<"vaddhn_high", "(<Q)<QQ", "silUsUiUl", OP_ADDHNHi>;
+def VRADDHN_HIGH : SOpInst<"vraddhn_high", "(<Q)<QQ", "silUsUiUl", OP_RADDHNHi>;
+def VSUBHN_HIGH  : SOpInst<"vsubhn_high", "(<Q)<QQ", "silUsUiUl", OP_SUBHNHi>;
+def VRSUBHN_HIGH : SOpInst<"vrsubhn_high", "(<Q)<QQ", "silUsUiUl", OP_RSUBHNHi>;
+
+def VQDMULL_HIGH : SOpInst<"vqdmull_high", "(>Q)QQ", "si", OP_QDMULLHi>;
+def VQDMULL_HIGH_N : SOpInst<"vqdmull_high_n", "(>Q)Q1", "si", OP_QDMULLHi_N>;
+def VQDMLAL_HIGH : SOpInst<"vqdmlal_high", "(>Q)(>Q)QQ", "si", OP_QDMLALHi>;
+def VQDMLAL_HIGH_N : SOpInst<"vqdmlal_high_n", "(>Q)(>Q)Q1", "si", OP_QDMLALHi_N>;
+def VQDMLSL_HIGH : SOpInst<"vqdmlsl_high", "(>Q)(>Q)QQ", "si", OP_QDMLSLHi>;
+def VQDMLSL_HIGH_N : SOpInst<"vqdmlsl_high_n", "(>Q)(>Q)Q1", "si", OP_QDMLSLHi_N>;
+def VMULL_P64    : SInst<"vmull", "(1>)11", "Pl">;
+def VMULL_HIGH_P64 : SOpInst<"vmull_high", "(1>)..", "HPl", OP_MULLHi_P64>;
 
 
 ////////////////////////////////////////////////////////////////////////////////
 // Extract or insert element from vector
-def GET_LANE : IInst<"vget_lane", "sdi", "dQdPlQPl">;
-def SET_LANE : IInst<"vset_lane", "dsdi", "dQdPlQPl">;
-def COPY_LANE : IOpInst<"vcopy_lane", "ddidi",
+def GET_LANE : IInst<"vget_lane", "1.I", "dQdPlQPl">;
+def SET_LANE : IInst<"vset_lane", ".1.I", "dQdPlQPl">;
+def COPY_LANE : IOpInst<"vcopy_lane", "..I.I",
                         "csilUcUsUiUlPcPsPlfd", OP_COPY_LN>;
-def COPYQ_LANE : IOpInst<"vcopy_lane", "ddigi",
+def COPYQ_LANE : IOpInst<"vcopy_lane", "..IqI",
                         "QcQsQiQlQUcQUsQUiQUlQPcQPsQfQdQPl", OP_COPY_LN>;
-def COPY_LANEQ : IOpInst<"vcopy_laneq", "ddiki",
+def COPY_LANEQ : IOpInst<"vcopy_laneq", "..IQI",
                      "csilPcPsPlUcUsUiUlfd", OP_COPY_LN>;
-def COPYQ_LANEQ : IOpInst<"vcopy_laneq", "ddidi",
+def COPYQ_LANEQ : IOpInst<"vcopy_laneq", "..I.I",
                      "QcQsQiQlQUcQUsQUiQUlQPcQPsQfQdQPl", OP_COPY_LN>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Set all lanes to same value
-def VDUP_LANE1: WOpInst<"vdup_lane", "dgi", "hdQhQdPlQPl", OP_DUP_LN>;
-def VDUP_LANE2: WOpInst<"vdup_laneq", "dji",
+def VDUP_LANE1: WOpInst<"vdup_lane", ".qI", "hdQhQdPlQPl", OP_DUP_LN>;
+def VDUP_LANE2: WOpInst<"vdup_laneq", ".QI",
                   "csilUcUsUiUlPcPshfdQcQsQiQlQPcQPsQUcQUsQUiQUlQhQfQdPlQPl",
                         OP_DUP_LN>;
-def DUP_N   : WOpInst<"vdup_n", "ds", "dQdPlQPl", OP_DUP>;
-def MOV_N   : WOpInst<"vmov_n", "ds", "dQdPlQPl", OP_DUP>;
+def DUP_N   : WOpInst<"vdup_n", ".1", "dQdPlQPl", OP_DUP>;
+def MOV_N   : WOpInst<"vmov_n", ".1", "dQdPlQPl", OP_DUP>;
 
 ////////////////////////////////////////////////////////////////////////////////
-def COMBINE : NoTestOpInst<"vcombine", "kdd", "dPl", OP_CONC>;
+def COMBINE : NoTestOpInst<"vcombine", "Q..", "dPl", OP_CONC>;
 
 ////////////////////////////////////////////////////////////////////////////////
 //Initialize a vector from bit pattern
-def CREATE : NoTestOpInst<"vcreate", "dl", "dPl", OP_CAST> {
+def CREATE : NoTestOpInst<"vcreate", ".(IU>)", "dPl", OP_CAST> {
   let BigEndianSafe = 1;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 
-def VMLA_LANEQ   : IOpInst<"vmla_laneq", "dddji",
+def VMLA_LANEQ   : IOpInst<"vmla_laneq", "...QI",
                            "siUsUifQsQiQUsQUiQf", OP_MLA_LN>;
-def VMLS_LANEQ   : IOpInst<"vmls_laneq", "dddji",
+def VMLS_LANEQ   : IOpInst<"vmls_laneq", "...QI",
                            "siUsUifQsQiQUsQUiQf", OP_MLS_LN>;
 
-def VFMA_LANE    : IInst<"vfma_lane", "dddgi", "fdQfQd">;
-def VFMA_LANEQ   : IInst<"vfma_laneq", "dddji", "fdQfQd"> {
+def VFMA_LANE    : IInst<"vfma_lane", "...qI", "fdQfQd">;
+def VFMA_LANEQ   : IInst<"vfma_laneq", "...QI", "fdQfQd"> {
   let isLaneQ = 1;
 }
-def VFMS_LANE    : IOpInst<"vfms_lane", "dddgi", "fdQfQd", OP_FMS_LN>;
-def VFMS_LANEQ   : IOpInst<"vfms_laneq", "dddji", "fdQfQd", OP_FMS_LNQ>;
+def VFMS_LANE    : IOpInst<"vfms_lane", "...qI", "fdQfQd", OP_FMS_LN>;
+def VFMS_LANEQ   : IOpInst<"vfms_laneq", "...QI", "fdQfQd", OP_FMS_LNQ>;
 
-def VMLAL_LANEQ  : SOpInst<"vmlal_laneq", "wwdki", "siUsUi", OP_MLAL_LN>;
-def VMLAL_HIGH_LANE   : SOpInst<"vmlal_high_lane", "wwkdi", "siUsUi",
+def VMLAL_LANEQ  : SOpInst<"vmlal_laneq", "(>Q)(>Q).QI", "siUsUi", OP_MLAL_LN>;
+def VMLAL_HIGH_LANE   : SOpInst<"vmlal_high_lane", "(>Q)(>Q)Q.I", "siUsUi",
                                 OP_MLALHi_LN>;
-def VMLAL_HIGH_LANEQ  : SOpInst<"vmlal_high_laneq", "wwkki", "siUsUi",
+def VMLAL_HIGH_LANEQ  : SOpInst<"vmlal_high_laneq", "(>Q)(>Q)QQI", "siUsUi",
                                 OP_MLALHi_LN>;
-def VMLSL_LANEQ  : SOpInst<"vmlsl_laneq", "wwdki", "siUsUi", OP_MLSL_LN>;
-def VMLSL_HIGH_LANE   : SOpInst<"vmlsl_high_lane", "wwkdi", "siUsUi",
+def VMLSL_LANEQ  : SOpInst<"vmlsl_laneq", "(>Q)(>Q).QI", "siUsUi", OP_MLSL_LN>;
+def VMLSL_HIGH_LANE   : SOpInst<"vmlsl_high_lane", "(>Q)(>Q)Q.I", "siUsUi",
                                 OP_MLSLHi_LN>;
-def VMLSL_HIGH_LANEQ  : SOpInst<"vmlsl_high_laneq", "wwkki", "siUsUi",
+def VMLSL_HIGH_LANEQ  : SOpInst<"vmlsl_high_laneq", "(>Q)(>Q)QQI", "siUsUi",
                                 OP_MLSLHi_LN>;
 
-def VQDMLAL_LANEQ  : SOpInst<"vqdmlal_laneq", "wwdki", "si", OP_QDMLAL_LN>;
-def VQDMLAL_HIGH_LANE   : SOpInst<"vqdmlal_high_lane", "wwkdi", "si",
+def VQDMLAL_LANEQ  : SOpInst<"vqdmlal_laneq", "(>Q)(>Q).QI", "si", OP_QDMLAL_LN>;
+def VQDMLAL_HIGH_LANE   : SOpInst<"vqdmlal_high_lane", "(>Q)(>Q)Q.I", "si",
                                 OP_QDMLALHi_LN>;
-def VQDMLAL_HIGH_LANEQ  : SOpInst<"vqdmlal_high_laneq", "wwkki", "si",
+def VQDMLAL_HIGH_LANEQ  : SOpInst<"vqdmlal_high_laneq", "(>Q)(>Q)QQI", "si",
                                 OP_QDMLALHi_LN>;
-def VQDMLSL_LANEQ  : SOpInst<"vqdmlsl_laneq", "wwdki", "si", OP_QDMLSL_LN>;
-def VQDMLSL_HIGH_LANE   : SOpInst<"vqdmlsl_high_lane", "wwkdi", "si",
+def VQDMLSL_LANEQ  : SOpInst<"vqdmlsl_laneq", "(>Q)(>Q).QI", "si", OP_QDMLSL_LN>;
+def VQDMLSL_HIGH_LANE   : SOpInst<"vqdmlsl_high_lane", "(>Q)(>Q)Q.I", "si",
                                 OP_QDMLSLHi_LN>;
-def VQDMLSL_HIGH_LANEQ  : SOpInst<"vqdmlsl_high_laneq", "wwkki", "si",
+def VQDMLSL_HIGH_LANEQ  : SOpInst<"vqdmlsl_high_laneq", "(>Q)(>Q)QQI", "si",
                                 OP_QDMLSLHi_LN>;
 
 // Newly add double parameter for vmul_lane in aarch64
 // Note: d type is handled by SCALAR_VMUL_LANE
-def VMUL_LANE_A64 : IOpInst<"vmul_lane", "ddgi", "Qd", OP_MUL_LN>;
+def VMUL_LANE_A64 : IOpInst<"vmul_lane", "..qI", "Qd", OP_MUL_LN>;
 
 // Note: d type is handled by SCALAR_VMUL_LANEQ
-def VMUL_LANEQ   : IOpInst<"vmul_laneq", "ddji",
+def VMUL_LANEQ   : IOpInst<"vmul_laneq", "..QI",
                            "sifUsUiQsQiQUsQUiQfQd", OP_MUL_LN>;
-def VMULL_LANEQ  : SOpInst<"vmull_laneq", "wdki", "siUsUi", OP_MULL_LN>;
-def VMULL_HIGH_LANE   : SOpInst<"vmull_high_lane", "wkdi", "siUsUi",
+def VMULL_LANEQ  : SOpInst<"vmull_laneq", "(>Q).QI", "siUsUi", OP_MULL_LN>;
+def VMULL_HIGH_LANE   : SOpInst<"vmull_high_lane", "(>Q)Q.I", "siUsUi",
                                 OP_MULLHi_LN>;
-def VMULL_HIGH_LANEQ  : SOpInst<"vmull_high_laneq", "wkki", "siUsUi",
+def VMULL_HIGH_LANEQ  : SOpInst<"vmull_high_laneq", "(>Q)QQI", "siUsUi",
                                 OP_MULLHi_LN>;
 
-def VQDMULL_LANEQ  : SOpInst<"vqdmull_laneq", "wdki", "si", OP_QDMULL_LN>;
-def VQDMULL_HIGH_LANE   : SOpInst<"vqdmull_high_lane", "wkdi", "si",
+def VQDMULL_LANEQ  : SOpInst<"vqdmull_laneq", "(>Q).QI", "si", OP_QDMULL_LN>;
+def VQDMULL_HIGH_LANE   : SOpInst<"vqdmull_high_lane", "(>Q)Q.I", "si",
                                   OP_QDMULLHi_LN>;
-def VQDMULL_HIGH_LANEQ  : SOpInst<"vqdmull_high_laneq", "wkki", "si",
+def VQDMULL_HIGH_LANEQ  : SOpInst<"vqdmull_high_laneq", "(>Q)QQI", "si",
                                   OP_QDMULLHi_LN>;
 
-def VQDMULH_LANEQ  : SOpInst<"vqdmulh_laneq", "ddji", "siQsQi", OP_QDMULH_LN>;
-def VQRDMULH_LANEQ : SOpInst<"vqrdmulh_laneq", "ddji", "siQsQi", OP_QRDMULH_LN>;
+def VQDMULH_LANEQ  : SOpInst<"vqdmulh_laneq", "..QI", "siQsQi", OP_QDMULH_LN>;
+def VQRDMULH_LANEQ : SOpInst<"vqrdmulh_laneq", "..QI", "siQsQi", OP_QRDMULH_LN>;
 
 let ArchGuard = "defined(__ARM_FEATURE_QRDMX) && defined(__aarch64__)" in {
-def VQRDMLAH_LANEQ : SOpInst<"vqrdmlah_laneq", "dddji", "siQsQi", OP_QRDMLAH_LN>;
-def VQRDMLSH_LANEQ : SOpInst<"vqrdmlsh_laneq", "dddji", "siQsQi", OP_QRDMLSH_LN>;
+def VQRDMLAH_LANEQ : SOpInst<"vqrdmlah_laneq", "...QI", "siQsQi", OP_QRDMLAH_LN>;
+def VQRDMLSH_LANEQ : SOpInst<"vqrdmlsh_laneq", "...QI", "siQsQi", OP_QRDMLSH_LN>;
 }
 
 // Note: d type implemented by SCALAR_VMULX_LANE
-def VMULX_LANE : IOpInst<"vmulx_lane", "ddgi", "fQfQd", OP_MULX_LN>;
+def VMULX_LANE : IOpInst<"vmulx_lane", "..qI", "fQfQd", OP_MULX_LN>;
 // Note: d type is implemented by SCALAR_VMULX_LANEQ
-def VMULX_LANEQ : IOpInst<"vmulx_laneq", "ddji", "fQfQd", OP_MULX_LN>;
+def VMULX_LANEQ : IOpInst<"vmulx_laneq", "..QI", "fQfQd", OP_MULX_LN>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Across vectors class
-def VADDLV  : SInst<"vaddlv", "rd", "csiUcUsUiQcQsQiQUcQUsQUi">;
-def VMAXV   : SInst<"vmaxv", "sd", "csifUcUsUiQcQsQiQUcQUsQUiQfQd">;
-def VMINV   : SInst<"vminv", "sd", "csifUcUsUiQcQsQiQUcQUsQUiQfQd">;
-def VADDV   : SInst<"vaddv", "sd", "csifUcUsUiQcQsQiQUcQUsQUiQfQdQlQUl">;
-def FMAXNMV : SInst<"vmaxnmv", "sd", "fQfQd">;
-def FMINNMV : SInst<"vminnmv", "sd", "fQfQd">;
+def VADDLV  : SInst<"vaddlv", "(1>).", "csiUcUsUiQcQsQiQUcQUsQUi">;
+def VMAXV   : SInst<"vmaxv", "1.", "csifUcUsUiQcQsQiQUcQUsQUiQfQd">;
+def VMINV   : SInst<"vminv", "1.", "csifUcUsUiQcQsQiQUcQUsQUiQfQd">;
+def VADDV   : SInst<"vaddv", "1.", "csifUcUsUiQcQsQiQUcQUsQUiQfQdQlQUl">;
+def FMAXNMV : SInst<"vmaxnmv", "1.", "fQfQd">;
+def FMINNMV : SInst<"vminnmv", "1.", "fQfQd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Newly added Vector Extract for f64
-def VEXT_A64 : WInst<"vext", "dddi", "dQdPlQPl">;
+def VEXT_A64 : WInst<"vext", "...I", "dQdPlQPl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Crypto
 let ArchGuard = "__ARM_ARCH >= 8 && defined(__ARM_FEATURE_CRYPTO)" in {
-def AESE : SInst<"vaese", "ddd", "QUc">;
-def AESD : SInst<"vaesd", "ddd", "QUc">;
-def AESMC : SInst<"vaesmc", "dd", "QUc">;
-def AESIMC : SInst<"vaesimc", "dd", "QUc">;
-
-def SHA1H : SInst<"vsha1h", "ss", "Ui">;
-def SHA1SU1 : SInst<"vsha1su1", "ddd", "QUi">;
-def SHA256SU0 : SInst<"vsha256su0", "ddd", "QUi">;
-
-def SHA1C : SInst<"vsha1c", "ddsd", "QUi">;
-def SHA1P : SInst<"vsha1p", "ddsd", "QUi">;
-def SHA1M : SInst<"vsha1m", "ddsd", "QUi">;
-def SHA1SU0 : SInst<"vsha1su0", "dddd", "QUi">;
-def SHA256H : SInst<"vsha256h", "dddd", "QUi">;
-def SHA256H2 : SInst<"vsha256h2", "dddd", "QUi">;
-def SHA256SU1 : SInst<"vsha256su1", "dddd", "QUi">;
+def AESE : SInst<"vaese", "...", "QUc">;
+def AESD : SInst<"vaesd", "...", "QUc">;
+def AESMC : SInst<"vaesmc", "..", "QUc">;
+def AESIMC : SInst<"vaesimc", "..", "QUc">;
+
+def SHA1H : SInst<"vsha1h", "11", "Ui">;
+def SHA1SU1 : SInst<"vsha1su1", "...", "QUi">;
+def SHA256SU0 : SInst<"vsha256su0", "...", "QUi">;
+
+def SHA1C : SInst<"vsha1c", "..1.", "QUi">;
+def SHA1P : SInst<"vsha1p", "..1.", "QUi">;
+def SHA1M : SInst<"vsha1m", "..1.", "QUi">;
+def SHA1SU0 : SInst<"vsha1su0", "....", "QUi">;
+def SHA256H : SInst<"vsha256h", "....", "QUi">;
+def SHA256H2 : SInst<"vsha256h2", "....", "QUi">;
+def SHA256SU1 : SInst<"vsha256su1", "....", "QUi">;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Float -> Int conversions with explicit rounding mode
 
 let ArchGuard = "__ARM_ARCH >= 8" in {
-def FCVTNS_S32 : SInst<"vcvtn_s32", "xd", "fQf">;
-def FCVTNU_S32 : SInst<"vcvtn_u32", "ud", "fQf">;
-def FCVTPS_S32 : SInst<"vcvtp_s32", "xd", "fQf">;
-def FCVTPU_S32 : SInst<"vcvtp_u32", "ud", "fQf">;
-def FCVTMS_S32 : SInst<"vcvtm_s32", "xd", "fQf">;
-def FCVTMU_S32 : SInst<"vcvtm_u32", "ud", "fQf">;
-def FCVTAS_S32 : SInst<"vcvta_s32", "xd", "fQf">;
-def FCVTAU_S32 : SInst<"vcvta_u32", "ud", "fQf">;
+def FCVTNS_S32 : SInst<"vcvtn_s32", "S.", "fQf">;
+def FCVTNU_S32 : SInst<"vcvtn_u32", "U.", "fQf">;
+def FCVTPS_S32 : SInst<"vcvtp_s32", "S.", "fQf">;
+def FCVTPU_S32 : SInst<"vcvtp_u32", "U.", "fQf">;
+def FCVTMS_S32 : SInst<"vcvtm_s32", "S.", "fQf">;
+def FCVTMU_S32 : SInst<"vcvtm_u32", "U.", "fQf">;
+def FCVTAS_S32 : SInst<"vcvta_s32", "S.", "fQf">;
+def FCVTAU_S32 : SInst<"vcvta_u32", "U.", "fQf">;
 }
 
 let ArchGuard = "__ARM_ARCH >= 8 && defined(__aarch64__)" in {
-def FCVTNS_S64 : SInst<"vcvtn_s64", "xd", "dQd">;
-def FCVTNU_S64 : SInst<"vcvtn_u64", "ud", "dQd">;
-def FCVTPS_S64 : SInst<"vcvtp_s64", "xd", "dQd">;
-def FCVTPU_S64 : SInst<"vcvtp_u64", "ud", "dQd">;
-def FCVTMS_S64 : SInst<"vcvtm_s64", "xd", "dQd">;
-def FCVTMU_S64 : SInst<"vcvtm_u64", "ud", "dQd">;
-def FCVTAS_S64 : SInst<"vcvta_s64", "xd", "dQd">;
-def FCVTAU_S64 : SInst<"vcvta_u64", "ud", "dQd">;
+def FCVTNS_S64 : SInst<"vcvtn_s64", "S.", "dQd">;
+def FCVTNU_S64 : SInst<"vcvtn_u64", "U.", "dQd">;
+def FCVTPS_S64 : SInst<"vcvtp_s64", "S.", "dQd">;
+def FCVTPU_S64 : SInst<"vcvtp_u64", "U.", "dQd">;
+def FCVTMS_S64 : SInst<"vcvtm_s64", "S.", "dQd">;
+def FCVTMU_S64 : SInst<"vcvtm_u64", "U.", "dQd">;
+def FCVTAS_S64 : SInst<"vcvta_s64", "S.", "dQd">;
+def FCVTAU_S64 : SInst<"vcvta_u64", "U.", "dQd">;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Round to Integral
 
 let ArchGuard = "__ARM_ARCH >= 8 && defined(__ARM_FEATURE_DIRECTED_ROUNDING)" in {
-def FRINTN_S32 : SInst<"vrndn", "dd", "fQf">;
-def FRINTA_S32 : SInst<"vrnda", "dd", "fQf">;
-def FRINTP_S32 : SInst<"vrndp", "dd", "fQf">;
-def FRINTM_S32 : SInst<"vrndm", "dd", "fQf">;
-def FRINTX_S32 : SInst<"vrndx", "dd", "fQf">;
-def FRINTZ_S32 : SInst<"vrnd", "dd", "fQf">;
-def FRINTI_S32 : SInst<"vrndi", "dd", "fQf">;
+def FRINTN_S32 : SInst<"vrndn", "..", "fQf">;
+def FRINTA_S32 : SInst<"vrnda", "..", "fQf">;
+def FRINTP_S32 : SInst<"vrndp", "..", "fQf">;
+def FRINTM_S32 : SInst<"vrndm", "..", "fQf">;
+def FRINTX_S32 : SInst<"vrndx", "..", "fQf">;
+def FRINTZ_S32 : SInst<"vrnd", "..", "fQf">;
+def FRINTI_S32 : SInst<"vrndi", "..", "fQf">;
 }
 
 let ArchGuard = "__ARM_ARCH >= 8 && defined(__aarch64__) && defined(__ARM_FEATURE_DIRECTED_ROUNDING)" in {
-def FRINTN_S64 : SInst<"vrndn", "dd", "dQd">;
-def FRINTA_S64 : SInst<"vrnda", "dd", "dQd">;
-def FRINTP_S64 : SInst<"vrndp", "dd", "dQd">;
-def FRINTM_S64 : SInst<"vrndm", "dd", "dQd">;
-def FRINTX_S64 : SInst<"vrndx", "dd", "dQd">;
-def FRINTZ_S64 : SInst<"vrnd", "dd", "dQd">;
-def FRINTI_S64 : SInst<"vrndi", "dd", "dQd">;
+def FRINTN_S64 : SInst<"vrndn", "..", "dQd">;
+def FRINTA_S64 : SInst<"vrnda", "..", "dQd">;
+def FRINTP_S64 : SInst<"vrndp", "..", "dQd">;
+def FRINTM_S64 : SInst<"vrndm", "..", "dQd">;
+def FRINTX_S64 : SInst<"vrndx", "..", "dQd">;
+def FRINTZ_S64 : SInst<"vrnd", "..", "dQd">;
+def FRINTI_S64 : SInst<"vrndi", "..", "dQd">;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // MaxNum/MinNum Floating Point
 
 let ArchGuard = "__ARM_ARCH >= 8 && defined(__ARM_FEATURE_NUMERIC_MAXMIN)" in {
-def FMAXNM_S32 : SInst<"vmaxnm", "ddd", "fQf">;
-def FMINNM_S32 : SInst<"vminnm", "ddd", "fQf">;
+def FMAXNM_S32 : SInst<"vmaxnm", "...", "fQf">;
+def FMINNM_S32 : SInst<"vminnm", "...", "fQf">;
 }
 
 let ArchGuard = "__ARM_ARCH >= 8 && defined(__aarch64__) && defined(__ARM_FEATURE_NUMERIC_MAXMIN)" in {
-def FMAXNM_S64 : SInst<"vmaxnm", "ddd", "dQd">;
-def FMINNM_S64 : SInst<"vminnm", "ddd", "dQd">;
+def FMAXNM_S64 : SInst<"vmaxnm", "...", "dQd">;
+def FMINNM_S64 : SInst<"vminnm", "...", "dQd">;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Permutation
-def VTRN1 : SOpInst<"vtrn1", "ddd",
+def VTRN1 : SOpInst<"vtrn1", "...",
                     "csiUcUsUifPcPsQcQsQiQlQUcQUsQUiQUlQfQdQPcQPsQPl", OP_TRN1>;
-def VZIP1 : SOpInst<"vzip1", "ddd",
+def VZIP1 : SOpInst<"vzip1", "...",
                     "csiUcUsUifPcPsQcQsQiQlQUcQUsQUiQUlQfQdQPcQPsQPl", OP_ZIP1>;
-def VUZP1 : SOpInst<"vuzp1", "ddd",
+def VUZP1 : SOpInst<"vuzp1", "...",
                     "csiUcUsUifPcPsQcQsQiQlQUcQUsQUiQUlQfQdQPcQPsQPl", OP_UZP1>;
-def VTRN2 : SOpInst<"vtrn2", "ddd",
+def VTRN2 : SOpInst<"vtrn2", "...",
                     "csiUcUsUifPcPsQcQsQiQlQUcQUsQUiQUlQfQdQPcQPsQPl", OP_TRN2>;
-def VZIP2 : SOpInst<"vzip2", "ddd",
+def VZIP2 : SOpInst<"vzip2", "...",
                     "csiUcUsUifPcPsQcQsQiQlQUcQUsQUiQUlQfQdQPcQPsQPl", OP_ZIP2>;
-def VUZP2 : SOpInst<"vuzp2", "ddd",
+def VUZP2 : SOpInst<"vuzp2", "...",
                     "csiUcUsUifPcPsQcQsQiQlQUcQUsQUiQUlQfQdQPcQPsQPl", OP_UZP2>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Table lookup
 let InstName = "vtbl" in {
-def VQTBL1_A64 : WInst<"vqtbl1", "dju",  "UccPcQUcQcQPc">;
-def VQTBL2_A64 : WInst<"vqtbl2", "dBu",  "UccPcQUcQcQPc">;
-def VQTBL3_A64 : WInst<"vqtbl3", "dCu",  "UccPcQUcQcQPc">;
-def VQTBL4_A64 : WInst<"vqtbl4", "dDu",  "UccPcQUcQcQPc">;
+def VQTBL1_A64 : WInst<"vqtbl1", ".QU",  "UccPcQUcQcQPc">;
+def VQTBL2_A64 : WInst<"vqtbl2", ".(2Q)U",  "UccPcQUcQcQPc">;
+def VQTBL3_A64 : WInst<"vqtbl3", ".(3Q)U",  "UccPcQUcQcQPc">;
+def VQTBL4_A64 : WInst<"vqtbl4", ".(4Q)U",  "UccPcQUcQcQPc">;
 }
 let InstName = "vtbx" in {
-def VQTBX1_A64 : WInst<"vqtbx1", "ddju", "UccPcQUcQcQPc">;
-def VQTBX2_A64 : WInst<"vqtbx2", "ddBu", "UccPcQUcQcQPc">;
-def VQTBX3_A64 : WInst<"vqtbx3", "ddCu", "UccPcQUcQcQPc">;
-def VQTBX4_A64 : WInst<"vqtbx4", "ddDu", "UccPcQUcQcQPc">;
+def VQTBX1_A64 : WInst<"vqtbx1", "..QU", "UccPcQUcQcQPc">;
+def VQTBX2_A64 : WInst<"vqtbx2", "..(2Q)U", "UccPcQUcQcQPc">;
+def VQTBX3_A64 : WInst<"vqtbx3", "..(3Q)U", "UccPcQUcQcQPc">;
+def VQTBX4_A64 : WInst<"vqtbx4", "..(4Q)U", "UccPcQUcQcQPc">;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Vector reinterpret cast operations
 
 // NeonEmitter implicitly takes the cartesian product of the type string with
 // itself during generation so, unlike all other intrinsics, this one should
 // include *all* types, not just additional ones.
 def VVREINTERPRET
-  : NoTestOpInst<"vreinterpret", "dd",
+  : NoTestOpInst<"vreinterpret", "..",
        "csilUcUsUiUlhfdPcPsPlQcQsQiQlQUcQUsQUiQUlQhQfQdQPcQPsQPlQPk", OP_REINT> {
   let CartesianProductOfTypes = 1;
   let BigEndianSafe = 1;
   let ArchGuard = "__ARM_ARCH >= 8 && defined(__aarch64__)";
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Intrinsics
 // Scalar Arithmetic
 
 // Scalar Addition
-def SCALAR_ADD : SInst<"vadd", "sss",  "SlSUl">;
+def SCALAR_ADD : SInst<"vadd", "111",  "SlSUl">;
 // Scalar  Saturating Add
-def SCALAR_QADD   : SInst<"vqadd", "sss", "ScSsSiSlSUcSUsSUiSUl">;
+def SCALAR_QADD   : SInst<"vqadd", "111", "ScSsSiSlSUcSUsSUiSUl">;
 
 // Scalar Subtraction
-def SCALAR_SUB : SInst<"vsub", "sss",  "SlSUl">;
+def SCALAR_SUB : SInst<"vsub", "111",  "SlSUl">;
 // Scalar  Saturating Sub
-def SCALAR_QSUB   : SInst<"vqsub", "sss", "ScSsSiSlSUcSUsSUiSUl">;
+def SCALAR_QSUB   : SInst<"vqsub", "111", "ScSsSiSlSUcSUsSUiSUl">;
 
 let InstName = "vmov" in {
-def VGET_HIGH_A64 : NoTestOpInst<"vget_high", "dk", "dPl", OP_HI>;
-def VGET_LOW_A64  : NoTestOpInst<"vget_low", "dk", "dPl", OP_LO>;
+def VGET_HIGH_A64 : NoTestOpInst<"vget_high", ".Q", "dPl", OP_HI>;
+def VGET_LOW_A64  : NoTestOpInst<"vget_low", ".Q", "dPl", OP_LO>;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Shift
 // Scalar Shift Left
-def SCALAR_SHL: SInst<"vshl", "sss", "SlSUl">;
+def SCALAR_SHL: SInst<"vshl", "111", "SlSUl">;
 // Scalar Saturating Shift Left
-def SCALAR_QSHL: SInst<"vqshl", "sss", "ScSsSiSlSUcSUsSUiSUl">;
+def SCALAR_QSHL: SInst<"vqshl", "111", "ScSsSiSlSUcSUsSUiSUl">;
 // Scalar Saturating Rounding Shift Left
-def SCALAR_QRSHL: SInst<"vqrshl", "sss", "ScSsSiSlSUcSUsSUiSUl">;
+def SCALAR_QRSHL: SInst<"vqrshl", "111", "ScSsSiSlSUcSUsSUiSUl">;
 // Scalar Shift Rounding Left
-def SCALAR_RSHL: SInst<"vrshl", "sss", "SlSUl">;
+def SCALAR_RSHL: SInst<"vrshl", "111", "SlSUl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Shift (Immediate)
 let isScalarShift = 1 in {
 // Signed/Unsigned Shift Right (Immediate)
-def SCALAR_SSHR_N: SInst<"vshr_n", "ssi", "SlSUl">;
+def SCALAR_SSHR_N: SInst<"vshr_n", "11I", "SlSUl">;
 // Signed/Unsigned Rounding Shift Right (Immediate)
-def SCALAR_SRSHR_N: SInst<"vrshr_n", "ssi", "SlSUl">;
+def SCALAR_SRSHR_N: SInst<"vrshr_n", "11I", "SlSUl">;
 
 // Signed/Unsigned Shift Right and Accumulate (Immediate)
-def SCALAR_SSRA_N: SInst<"vsra_n", "sssi", "SlSUl">;
+def SCALAR_SSRA_N: SInst<"vsra_n", "111I", "SlSUl">;
 // Signed/Unsigned Rounding Shift Right and Accumulate (Immediate)
-def SCALAR_SRSRA_N: SInst<"vrsra_n", "sssi", "SlSUl">;
+def SCALAR_SRSRA_N: SInst<"vrsra_n", "111I", "SlSUl">;
 
 // Shift Left (Immediate)
-def SCALAR_SHL_N: SInst<"vshl_n", "ssi", "SlSUl">;
+def SCALAR_SHL_N: SInst<"vshl_n", "11I", "SlSUl">;
 // Signed/Unsigned Saturating Shift Left (Immediate)
-def SCALAR_SQSHL_N: SInst<"vqshl_n", "ssi", "ScSsSiSlSUcSUsSUiSUl">;
+def SCALAR_SQSHL_N: SInst<"vqshl_n", "11I", "ScSsSiSlSUcSUsSUiSUl">;
 // Signed Saturating Shift Left Unsigned (Immediate)
-def SCALAR_SQSHLU_N: SInst<"vqshlu_n", "ssi", "ScSsSiSl">;
+def SCALAR_SQSHLU_N: SInst<"vqshlu_n", "11I", "ScSsSiSl">;
 
 // Shift Right And Insert (Immediate)
-def SCALAR_SRI_N: SInst<"vsri_n", "sssi", "SlSUl">;
+def SCALAR_SRI_N: SInst<"vsri_n", "111I", "SlSUl">;
 // Shift Left And Insert (Immediate)
-def SCALAR_SLI_N: SInst<"vsli_n", "sssi", "SlSUl">;
+def SCALAR_SLI_N: SInst<"vsli_n", "111I", "SlSUl">;
 
 let isScalarNarrowShift = 1 in {
   // Signed/Unsigned Saturating Shift Right Narrow (Immediate)
-  def SCALAR_SQSHRN_N: SInst<"vqshrn_n", "zsi", "SsSiSlSUsSUiSUl">;
+  def SCALAR_SQSHRN_N: SInst<"vqshrn_n", "(1<)1I", "SsSiSlSUsSUiSUl">;
   // Signed/Unsigned Saturating Rounded Shift Right Narrow (Immediate)
-  def SCALAR_SQRSHRN_N: SInst<"vqrshrn_n", "zsi", "SsSiSlSUsSUiSUl">;
+  def SCALAR_SQRSHRN_N: SInst<"vqrshrn_n", "(1<)1I", "SsSiSlSUsSUiSUl">;
   // Signed Saturating Shift Right Unsigned Narrow (Immediate)
-  def SCALAR_SQSHRUN_N: SInst<"vqshrun_n", "zsi", "SsSiSl">;
+  def SCALAR_SQSHRUN_N: SInst<"vqshrun_n", "(1<)1I", "SsSiSl">;
   // Signed Saturating Rounded Shift Right Unsigned Narrow (Immediate)
-  def SCALAR_SQRSHRUN_N: SInst<"vqrshrun_n", "zsi", "SsSiSl">;
+  def SCALAR_SQRSHRUN_N: SInst<"vqrshrun_n", "(1<)1I", "SsSiSl">;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Signed/Unsigned Fixed-point Convert To Floating-Point (Immediate)
-def SCALAR_SCVTF_N_F32: SInst<"vcvt_n_f32", "ysi", "SiSUi">;
-def SCALAR_SCVTF_N_F64: SInst<"vcvt_n_f64", "osi", "SlSUl">;
+def SCALAR_SCVTF_N_F32: SInst<"vcvt_n_f32", "(1F)(1!)I", "SiSUi">;
+def SCALAR_SCVTF_N_F64: SInst<"vcvt_n_f64", "(1F)(1!)I", "SlSUl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Floating-point Convert To Signed/Unsigned Fixed-point (Immediate)
-def SCALAR_FCVTZS_N_S32 : SInst<"vcvt_n_s32", "$si", "Sf">;
-def SCALAR_FCVTZU_N_U32 : SInst<"vcvt_n_u32", "bsi", "Sf">;
-def SCALAR_FCVTZS_N_S64 : SInst<"vcvt_n_s64", "$si", "Sd">;
-def SCALAR_FCVTZU_N_U64 : SInst<"vcvt_n_u64", "bsi", "Sd">;
+def SCALAR_FCVTZS_N_S32 : SInst<"vcvt_n_s32", "(1S)1I", "Sf">;
+def SCALAR_FCVTZU_N_U32 : SInst<"vcvt_n_u32", "(1U)1I", "Sf">;
+def SCALAR_FCVTZS_N_S64 : SInst<"vcvt_n_s64", "(1S)1I", "Sd">;
+def SCALAR_FCVTZU_N_U64 : SInst<"vcvt_n_u64", "(1U)1I", "Sd">;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Floating-point Round to Integral
 let ArchGuard = "__ARM_ARCH >= 8 && defined(__ARM_FEATURE_DIRECTED_ROUNDING)" in {
-def SCALAR_FRINTN_S32 : SInst<"vrndn", "ss", "Sf">;
+def SCALAR_FRINTN_S32 : SInst<"vrndn", "11", "Sf">;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Reduce Pairwise Addition (Scalar and Floating Point)
-def SCALAR_ADDP  : SInst<"vpadd", "sd", "SfSHlSHdSHUl">;
+def SCALAR_ADDP  : SInst<"vpadd", "1.", "SfSHlSHdSHUl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Reduce Floating Point Pairwise Max/Min
-def SCALAR_FMAXP : SInst<"vpmax", "sd", "SfSQd">;
+def SCALAR_FMAXP : SInst<"vpmax", "1.", "SfSQd">;
 
-def SCALAR_FMINP : SInst<"vpmin", "sd", "SfSQd">;
+def SCALAR_FMINP : SInst<"vpmin", "1.", "SfSQd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Reduce Floating Point Pairwise maxNum/minNum
-def SCALAR_FMAXNMP : SInst<"vpmaxnm", "sd", "SfSQd">;
-def SCALAR_FMINNMP : SInst<"vpminnm", "sd", "SfSQd">;
+def SCALAR_FMAXNMP : SInst<"vpmaxnm", "1.", "SfSQd">;
+def SCALAR_FMINNMP : SInst<"vpminnm", "1.", "SfSQd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Integer Saturating Doubling Multiply Half High
-def SCALAR_SQDMULH : SInst<"vqdmulh", "sss", "SsSi">;
+def SCALAR_SQDMULH : SInst<"vqdmulh", "111", "SsSi">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Integer Saturating Rounding Doubling Multiply Half High
-def SCALAR_SQRDMULH : SInst<"vqrdmulh", "sss", "SsSi">;
+def SCALAR_SQRDMULH : SInst<"vqrdmulh", "111", "SsSi">;
 
 let ArchGuard = "defined(__ARM_FEATURE_QRDMX) && defined(__aarch64__)" in {
 ////////////////////////////////////////////////////////////////////////////////
 // Signed Saturating Rounding Doubling Multiply Accumulate Returning High Half
-def SCALAR_SQRDMLAH : SOpInst<"vqrdmlah", "ssss", "SsSi", OP_QRDMLAH>;
+def SCALAR_SQRDMLAH : SOpInst<"vqrdmlah", "1111", "SsSi", OP_QRDMLAH>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Signed Saturating Rounding Doubling Multiply Subtract Returning High Half
-def SCALAR_SQRDMLSH : SOpInst<"vqrdmlsh", "ssss", "SsSi", OP_QRDMLSH>;
+def SCALAR_SQRDMLSH : SOpInst<"vqrdmlsh", "1111", "SsSi", OP_QRDMLSH>;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Floating-point Multiply Extended
-def SCALAR_FMULX : IInst<"vmulx", "sss", "SfSd">;
+def SCALAR_FMULX : IInst<"vmulx", "111", "SfSd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Floating-point Reciprocal Step
-def SCALAR_FRECPS : IInst<"vrecps", "sss", "SfSd">;
+def SCALAR_FRECPS : IInst<"vrecps", "111", "SfSd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Floating-point Reciprocal Square Root Step
-def SCALAR_FRSQRTS : IInst<"vrsqrts", "sss", "SfSd">;
+def SCALAR_FRSQRTS : IInst<"vrsqrts", "111", "SfSd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Signed Integer Convert To Floating-point
-def SCALAR_SCVTFS : SInst<"vcvt_f32", "ys", "Si">;
-def SCALAR_SCVTFD : SInst<"vcvt_f64", "os", "Sl">;
+def SCALAR_SCVTFS : SInst<"vcvt_f32", "(1F)(1!)", "Si">;
+def SCALAR_SCVTFD : SInst<"vcvt_f64", "(1F)(1!)", "Sl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Unsigned Integer Convert To Floating-point
-def SCALAR_UCVTFS : SInst<"vcvt_f32", "ys", "SUi">;
-def SCALAR_UCVTFD : SInst<"vcvt_f64", "os", "SUl">;
+def SCALAR_UCVTFS : SInst<"vcvt_f32", "(1F)(1!)", "SUi">;
+def SCALAR_UCVTFD : SInst<"vcvt_f64", "(1F)(1!)", "SUl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Floating-point Converts
-def SCALAR_FCVTXN  : IInst<"vcvtx_f32", "ys", "Sd">;
-def SCALAR_FCVTNSS : SInst<"vcvtn_s32", "$s", "Sf">;
-def SCALAR_FCVTNUS : SInst<"vcvtn_u32", "bs", "Sf">;
-def SCALAR_FCVTNSD : SInst<"vcvtn_s64", "$s", "Sd">;
-def SCALAR_FCVTNUD : SInst<"vcvtn_u64", "bs", "Sd">;
-def SCALAR_FCVTMSS : SInst<"vcvtm_s32", "$s", "Sf">;
-def SCALAR_FCVTMUS : SInst<"vcvtm_u32", "bs", "Sf">;
-def SCALAR_FCVTMSD : SInst<"vcvtm_s64", "$s", "Sd">;
-def SCALAR_FCVTMUD : SInst<"vcvtm_u64", "bs", "Sd">;
-def SCALAR_FCVTASS : SInst<"vcvta_s32", "$s", "Sf">;
-def SCALAR_FCVTAUS : SInst<"vcvta_u32", "bs", "Sf">;
-def SCALAR_FCVTASD : SInst<"vcvta_s64", "$s", "Sd">;
-def SCALAR_FCVTAUD : SInst<"vcvta_u64", "bs", "Sd">;
-def SCALAR_FCVTPSS : SInst<"vcvtp_s32", "$s", "Sf">;
-def SCALAR_FCVTPUS : SInst<"vcvtp_u32", "bs", "Sf">;
-def SCALAR_FCVTPSD : SInst<"vcvtp_s64", "$s", "Sd">;
-def SCALAR_FCVTPUD : SInst<"vcvtp_u64", "bs", "Sd">;
-def SCALAR_FCVTZSS : SInst<"vcvt_s32", "$s", "Sf">;
-def SCALAR_FCVTZUS : SInst<"vcvt_u32", "bs", "Sf">;
-def SCALAR_FCVTZSD : SInst<"vcvt_s64", "$s", "Sd">;
-def SCALAR_FCVTZUD : SInst<"vcvt_u64", "bs", "Sd">;
+def SCALAR_FCVTXN  : IInst<"vcvtx_f32", "(1F<)(1!)", "Sd">;
+def SCALAR_FCVTNSS : SInst<"vcvtn_s32", "(1S)1", "Sf">;
+def SCALAR_FCVTNUS : SInst<"vcvtn_u32", "(1U)1", "Sf">;
+def SCALAR_FCVTNSD : SInst<"vcvtn_s64", "(1S)1", "Sd">;
+def SCALAR_FCVTNUD : SInst<"vcvtn_u64", "(1U)1", "Sd">;
+def SCALAR_FCVTMSS : SInst<"vcvtm_s32", "(1S)1", "Sf">;
+def SCALAR_FCVTMUS : SInst<"vcvtm_u32", "(1U)1", "Sf">;
+def SCALAR_FCVTMSD : SInst<"vcvtm_s64", "(1S)1", "Sd">;
+def SCALAR_FCVTMUD : SInst<"vcvtm_u64", "(1U)1", "Sd">;
+def SCALAR_FCVTASS : SInst<"vcvta_s32", "(1S)1", "Sf">;
+def SCALAR_FCVTAUS : SInst<"vcvta_u32", "(1U)1", "Sf">;
+def SCALAR_FCVTASD : SInst<"vcvta_s64", "(1S)1", "Sd">;
+def SCALAR_FCVTAUD : SInst<"vcvta_u64", "(1U)1", "Sd">;
+def SCALAR_FCVTPSS : SInst<"vcvtp_s32", "(1S)1", "Sf">;
+def SCALAR_FCVTPUS : SInst<"vcvtp_u32", "(1U)1", "Sf">;
+def SCALAR_FCVTPSD : SInst<"vcvtp_s64", "(1S)1", "Sd">;
+def SCALAR_FCVTPUD : SInst<"vcvtp_u64", "(1U)1", "Sd">;
+def SCALAR_FCVTZSS : SInst<"vcvt_s32", "(1S)1", "Sf">;
+def SCALAR_FCVTZUS : SInst<"vcvt_u32", "(1U)1", "Sf">;
+def SCALAR_FCVTZSD : SInst<"vcvt_s64", "(1S)1", "Sd">;
+def SCALAR_FCVTZUD : SInst<"vcvt_u64", "(1U)1", "Sd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Floating-point Reciprocal Estimate
-def SCALAR_FRECPE : IInst<"vrecpe", "ss", "SfSd">;
+def SCALAR_FRECPE : IInst<"vrecpe", "11", "SfSd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Floating-point Reciprocal Exponent
-def SCALAR_FRECPX : IInst<"vrecpx", "ss", "SfSd">;
+def SCALAR_FRECPX : IInst<"vrecpx", "11", "SfSd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Floating-point Reciprocal Square Root Estimate
-def SCALAR_FRSQRTE : IInst<"vrsqrte", "ss", "SfSd">;
+def SCALAR_FRSQRTE : IInst<"vrsqrte", "11", "SfSd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Integer Comparison
-def SCALAR_CMEQ : SInst<"vceq", "sss", "SlSUl">;
-def SCALAR_CMEQZ : SInst<"vceqz", "ss", "SlSUl">;
-def SCALAR_CMGE : SInst<"vcge", "sss", "Sl">;
-def SCALAR_CMGEZ : SInst<"vcgez", "ss", "Sl">;
-def SCALAR_CMHS : SInst<"vcge", "sss", "SUl">;
-def SCALAR_CMLE : SInst<"vcle", "sss", "SlSUl">;
-def SCALAR_CMLEZ : SInst<"vclez", "ss", "Sl">;
-def SCALAR_CMLT : SInst<"vclt", "sss", "SlSUl">;
-def SCALAR_CMLTZ : SInst<"vcltz", "ss", "Sl">;
-def SCALAR_CMGT : SInst<"vcgt", "sss", "Sl">;
-def SCALAR_CMGTZ : SInst<"vcgtz", "ss", "Sl">;
-def SCALAR_CMHI : SInst<"vcgt", "sss", "SUl">;
-def SCALAR_CMTST : SInst<"vtst", "sss", "SlSUl">;
+def SCALAR_CMEQ : SInst<"vceq", "111", "SlSUl">;
+def SCALAR_CMEQZ : SInst<"vceqz", "11", "SlSUl">;
+def SCALAR_CMGE : SInst<"vcge", "111", "Sl">;
+def SCALAR_CMGEZ : SInst<"vcgez", "11", "Sl">;
+def SCALAR_CMHS : SInst<"vcge", "111", "SUl">;
+def SCALAR_CMLE : SInst<"vcle", "111", "SlSUl">;
+def SCALAR_CMLEZ : SInst<"vclez", "11", "Sl">;
+def SCALAR_CMLT : SInst<"vclt", "111", "SlSUl">;
+def SCALAR_CMLTZ : SInst<"vcltz", "11", "Sl">;
+def SCALAR_CMGT : SInst<"vcgt", "111", "Sl">;
+def SCALAR_CMGTZ : SInst<"vcgtz", "11", "Sl">;
+def SCALAR_CMHI : SInst<"vcgt", "111", "SUl">;
+def SCALAR_CMTST : SInst<"vtst", "111", "SlSUl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Floating-point Comparison
-def SCALAR_FCMEQ : IInst<"vceq", "bss", "SfSd">;
-def SCALAR_FCMEQZ : IInst<"vceqz", "bs", "SfSd">;
-def SCALAR_FCMGE : IInst<"vcge", "bss", "SfSd">;
-def SCALAR_FCMGEZ : IInst<"vcgez", "bs", "SfSd">;
-def SCALAR_FCMGT : IInst<"vcgt", "bss", "SfSd">;
-def SCALAR_FCMGTZ : IInst<"vcgtz", "bs", "SfSd">;
-def SCALAR_FCMLE : IInst<"vcle", "bss", "SfSd">;
-def SCALAR_FCMLEZ : IInst<"vclez", "bs", "SfSd">;
-def SCALAR_FCMLT : IInst<"vclt", "bss", "SfSd">;
-def SCALAR_FCMLTZ : IInst<"vcltz", "bs", "SfSd">;
+def SCALAR_FCMEQ : IInst<"vceq", "(1U)11", "SfSd">;
+def SCALAR_FCMEQZ : IInst<"vceqz", "(1U)1", "SfSd">;
+def SCALAR_FCMGE : IInst<"vcge", "(1U)11", "SfSd">;
+def SCALAR_FCMGEZ : IInst<"vcgez", "(1U)1", "SfSd">;
+def SCALAR_FCMGT : IInst<"vcgt", "(1U)11", "SfSd">;
+def SCALAR_FCMGTZ : IInst<"vcgtz", "(1U)1", "SfSd">;
+def SCALAR_FCMLE : IInst<"vcle", "(1U)11", "SfSd">;
+def SCALAR_FCMLEZ : IInst<"vclez", "(1U)1", "SfSd">;
+def SCALAR_FCMLT : IInst<"vclt", "(1U)11", "SfSd">;
+def SCALAR_FCMLTZ : IInst<"vcltz", "(1U)1", "SfSd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Floating-point Absolute Compare Mask Greater Than Or Equal
-def SCALAR_FACGE : IInst<"vcage", "bss", "SfSd">;
-def SCALAR_FACLE : IInst<"vcale", "bss", "SfSd">;
+def SCALAR_FACGE : IInst<"vcage", "(1U)11", "SfSd">;
+def SCALAR_FACLE : IInst<"vcale", "(1U)11", "SfSd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Floating-point Absolute Compare Mask Greater Than
-def SCALAR_FACGT : IInst<"vcagt", "bss", "SfSd">;
-def SCALAR_FACLT : IInst<"vcalt", "bss", "SfSd">;
+def SCALAR_FACGT : IInst<"vcagt", "(1U)11", "SfSd">;
+def SCALAR_FACLT : IInst<"vcalt", "(1U)11", "SfSd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Absolute Value
-def SCALAR_ABS : SInst<"vabs", "ss", "Sl">;
+def SCALAR_ABS : SInst<"vabs", "11", "Sl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Absolute Difference
-def SCALAR_ABD : IInst<"vabd", "sss", "SfSd">;
+def SCALAR_ABD : IInst<"vabd", "111", "SfSd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Signed Saturating Absolute Value
-def SCALAR_SQABS : SInst<"vqabs", "ss", "ScSsSiSl">;
+def SCALAR_SQABS : SInst<"vqabs", "11", "ScSsSiSl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Negate
-def SCALAR_NEG : SInst<"vneg", "ss", "Sl">;
+def SCALAR_NEG : SInst<"vneg", "11", "Sl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Signed Saturating Negate
-def SCALAR_SQNEG : SInst<"vqneg", "ss", "ScSsSiSl">;
+def SCALAR_SQNEG : SInst<"vqneg", "11", "ScSsSiSl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Signed Saturating Accumulated of Unsigned Value
-def SCALAR_SUQADD : SInst<"vuqadd", "ssb", "ScSsSiSl">;
+def SCALAR_SUQADD : SInst<"vuqadd", "11(1U)", "ScSsSiSl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Unsigned Saturating Accumulated of Signed Value
-def SCALAR_USQADD : SInst<"vsqadd", "ss$", "SUcSUsSUiSUl">;
+def SCALAR_USQADD : SInst<"vsqadd", "11(1S)", "SUcSUsSUiSUl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Signed Saturating Doubling Multiply-Add Long
-def SCALAR_SQDMLAL : SInst<"vqdmlal", "rrss", "SsSi">;
+def SCALAR_SQDMLAL : SInst<"vqdmlal", "(1>)(1>)11", "SsSi">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Signed Saturating Doubling Multiply-Subtract Long
-def SCALAR_SQDMLSL : SInst<"vqdmlsl", "rrss", "SsSi">;
+def SCALAR_SQDMLSL : SInst<"vqdmlsl", "(1>)(1>)11", "SsSi">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Signed Saturating Doubling Multiply Long
-def SCALAR_SQDMULL : SInst<"vqdmull", "rss", "SsSi">;
+def SCALAR_SQDMULL : SInst<"vqdmull", "(1>)11", "SsSi">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Signed Saturating Extract Unsigned Narrow
-def SCALAR_SQXTUN : SInst<"vqmovun", "zs", "SsSiSl">;
+def SCALAR_SQXTUN : SInst<"vqmovun", "(1<)1", "SsSiSl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Signed Saturating Extract Narrow
-def SCALAR_SQXTN : SInst<"vqmovn", "zs", "SsSiSl">;
+def SCALAR_SQXTN : SInst<"vqmovn", "(1<)1", "SsSiSl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Unsigned Saturating Extract Narrow
-def SCALAR_UQXTN : SInst<"vqmovn", "zs", "SUsSUiSUl">;
+def SCALAR_UQXTN : SInst<"vqmovn", "(1<)1", "SUsSUiSUl">;
 
 // Scalar Floating Point  multiply (scalar, by element)
-def SCALAR_FMUL_LANE : IOpInst<"vmul_lane", "ssdi", "SfSd", OP_SCALAR_MUL_LN>;
-def SCALAR_FMUL_LANEQ : IOpInst<"vmul_laneq", "ssji", "SfSd", OP_SCALAR_MUL_LN>;
+def SCALAR_FMUL_LANE : IOpInst<"vmul_lane", "11.I", "SfSd", OP_SCALAR_MUL_LN>;
+def SCALAR_FMUL_LANEQ : IOpInst<"vmul_laneq", "11QI", "SfSd", OP_SCALAR_MUL_LN>;
 
 // Scalar Floating Point  multiply extended (scalar, by element)
-def SCALAR_FMULX_LANE : IOpInst<"vmulx_lane", "ssdi", "SfSd", OP_SCALAR_MULX_LN>;
-def SCALAR_FMULX_LANEQ : IOpInst<"vmulx_laneq", "ssji", "SfSd", OP_SCALAR_MULX_LN>;
+def SCALAR_FMULX_LANE : IOpInst<"vmulx_lane", "11.I", "SfSd", OP_SCALAR_MULX_LN>;
+def SCALAR_FMULX_LANEQ : IOpInst<"vmulx_laneq", "11QI", "SfSd", OP_SCALAR_MULX_LN>;
 
-def SCALAR_VMUL_N : IInst<"vmul_n", "dds", "d">;
+def SCALAR_VMUL_N : IInst<"vmul_n", "..1", "d">;
 
 // VMUL_LANE_A64 d type implemented using scalar mul lane
-def SCALAR_VMUL_LANE : IInst<"vmul_lane", "ddgi", "d">;
+def SCALAR_VMUL_LANE : IInst<"vmul_lane", "..qI", "d">;
 
 // VMUL_LANEQ d type implemented using scalar mul lane
-def SCALAR_VMUL_LANEQ   : IInst<"vmul_laneq", "ddji", "d"> {
+def SCALAR_VMUL_LANEQ   : IInst<"vmul_laneq", "..QI", "d"> {
   let isLaneQ = 1;
 }
 
 // VMULX_LANE d type implemented using scalar vmulx_lane
-def SCALAR_VMULX_LANE : IOpInst<"vmulx_lane", "ddgi", "d", OP_SCALAR_VMULX_LN>;
+def SCALAR_VMULX_LANE : IOpInst<"vmulx_lane", "..qI", "d", OP_SCALAR_VMULX_LN>;
 
 // VMULX_LANEQ d type implemented using scalar vmulx_laneq
-def SCALAR_VMULX_LANEQ : IOpInst<"vmulx_laneq", "ddji", "d", OP_SCALAR_VMULX_LNQ>;
+def SCALAR_VMULX_LANEQ : IOpInst<"vmulx_laneq", "..QI", "d", OP_SCALAR_VMULX_LNQ>;
 
 // Scalar Floating Point fused multiply-add (scalar, by element)
-def SCALAR_FMLA_LANE : IInst<"vfma_lane", "sssdi", "SfSd">;
-def SCALAR_FMLA_LANEQ : IInst<"vfma_laneq", "sssji", "SfSd">;
+def SCALAR_FMLA_LANE : IInst<"vfma_lane", "111.I", "SfSd">;
+def SCALAR_FMLA_LANEQ : IInst<"vfma_laneq", "111QI", "SfSd">;
 
 // Scalar Floating Point fused multiply-subtract (scalar, by element)
-def SCALAR_FMLS_LANE : IOpInst<"vfms_lane", "sssdi", "SfSd", OP_FMS_LN>;
-def SCALAR_FMLS_LANEQ : IOpInst<"vfms_laneq", "sssji", "SfSd", OP_FMS_LNQ>;
+def SCALAR_FMLS_LANE : IOpInst<"vfms_lane", "111.I", "SfSd", OP_FMS_LN>;
+def SCALAR_FMLS_LANEQ : IOpInst<"vfms_laneq", "111QI", "SfSd", OP_FMS_LNQ>;
 
 // Signed Saturating Doubling Multiply Long (scalar by element)
-def SCALAR_SQDMULL_LANE : SOpInst<"vqdmull_lane", "rsdi", "SsSi", OP_SCALAR_QDMULL_LN>;
-def SCALAR_SQDMULL_LANEQ : SOpInst<"vqdmull_laneq", "rsji", "SsSi", OP_SCALAR_QDMULL_LN>;
+def SCALAR_SQDMULL_LANE : SOpInst<"vqdmull_lane", "(1>)1.I", "SsSi", OP_SCALAR_QDMULL_LN>;
+def SCALAR_SQDMULL_LANEQ : SOpInst<"vqdmull_laneq", "(1>)1QI", "SsSi", OP_SCALAR_QDMULL_LN>;
 
 // Signed Saturating Doubling Multiply-Add Long (scalar by element)
-def SCALAR_SQDMLAL_LANE : SInst<"vqdmlal_lane", "rrsdi", "SsSi">;
-def SCALAR_SQDMLAL_LANEQ : SInst<"vqdmlal_laneq", "rrsji", "SsSi">;
+def SCALAR_SQDMLAL_LANE : SInst<"vqdmlal_lane", "(1>)(1>)1.I", "SsSi">;
+def SCALAR_SQDMLAL_LANEQ : SInst<"vqdmlal_laneq", "(1>)(1>)1QI", "SsSi">;
 
 // Signed Saturating Doubling Multiply-Subtract Long (scalar by element)
-def SCALAR_SQDMLS_LANE : SInst<"vqdmlsl_lane", "rrsdi", "SsSi">;
-def SCALAR_SQDMLS_LANEQ : SInst<"vqdmlsl_laneq", "rrsji", "SsSi">;
+def SCALAR_SQDMLS_LANE : SInst<"vqdmlsl_lane", "(1>)(1>)1.I", "SsSi">;
+def SCALAR_SQDMLS_LANEQ : SInst<"vqdmlsl_laneq", "(1>)(1>)1QI", "SsSi">;
 
 // Scalar Integer Saturating Doubling Multiply Half High (scalar by element)
-def SCALAR_SQDMULH_LANE : SOpInst<"vqdmulh_lane", "ssdi", "SsSi", OP_SCALAR_QDMULH_LN>;
-def SCALAR_SQDMULH_LANEQ : SOpInst<"vqdmulh_laneq", "ssji", "SsSi", OP_SCALAR_QDMULH_LN>;
+def SCALAR_SQDMULH_LANE : SOpInst<"vqdmulh_lane", "11.I", "SsSi", OP_SCALAR_QDMULH_LN>;
+def SCALAR_SQDMULH_LANEQ : SOpInst<"vqdmulh_laneq", "11QI", "SsSi", OP_SCALAR_QDMULH_LN>;
 
 // Scalar Integer Saturating Rounding Doubling Multiply Half High
-def SCALAR_SQRDMULH_LANE : SOpInst<"vqrdmulh_lane", "ssdi", "SsSi", OP_SCALAR_QRDMULH_LN>;
-def SCALAR_SQRDMULH_LANEQ : SOpInst<"vqrdmulh_laneq", "ssji", "SsSi", OP_SCALAR_QRDMULH_LN>;
+def SCALAR_SQRDMULH_LANE : SOpInst<"vqrdmulh_lane", "11.I", "SsSi", OP_SCALAR_QRDMULH_LN>;
+def SCALAR_SQRDMULH_LANEQ : SOpInst<"vqrdmulh_laneq", "11QI", "SsSi", OP_SCALAR_QRDMULH_LN>;
 
 let ArchGuard = "defined(__ARM_FEATURE_QRDMX) && defined(__aarch64__)" in {
 // Signed Saturating Rounding Doubling Multiply Accumulate Returning High Half
-def SCALAR_SQRDMLAH_LANE : SOpInst<"vqrdmlah_lane", "sssdi", "SsSi", OP_SCALAR_QRDMLAH_LN>;
-def SCALAR_SQRDMLAH_LANEQ : SOpInst<"vqrdmlah_laneq", "sssji", "SsSi", OP_SCALAR_QRDMLAH_LN>;
+def SCALAR_SQRDMLAH_LANE : SOpInst<"vqrdmlah_lane", "111.I", "SsSi", OP_SCALAR_QRDMLAH_LN>;
+def SCALAR_SQRDMLAH_LANEQ : SOpInst<"vqrdmlah_laneq", "111QI", "SsSi", OP_SCALAR_QRDMLAH_LN>;
 
 // Signed Saturating Rounding Doubling Multiply Subtract Returning High Half
-def SCALAR_SQRDMLSH_LANE : SOpInst<"vqrdmlsh_lane", "sssdi", "SsSi", OP_SCALAR_QRDMLSH_LN>;
-def SCALAR_SQRDMLSH_LANEQ : SOpInst<"vqrdmlsh_laneq", "sssji", "SsSi", OP_SCALAR_QRDMLSH_LN>;
+def SCALAR_SQRDMLSH_LANE : SOpInst<"vqrdmlsh_lane", "111.I", "SsSi", OP_SCALAR_QRDMLSH_LN>;
+def SCALAR_SQRDMLSH_LANEQ : SOpInst<"vqrdmlsh_laneq", "111QI", "SsSi", OP_SCALAR_QRDMLSH_LN>;
 }
 
-def SCALAR_VDUP_LANE : IInst<"vdup_lane", "sdi", "ScSsSiSlSfSdSUcSUsSUiSUlSPcSPs">;
-def SCALAR_VDUP_LANEQ : IInst<"vdup_laneq", "sji", "ScSsSiSlSfSdSUcSUsSUiSUlSPcSPs">;
+def SCALAR_VDUP_LANE : IInst<"vdup_lane", "1.I", "ScSsSiSlSfSdSUcSUsSUiSUlSPcSPs">;
+def SCALAR_VDUP_LANEQ : IInst<"vdup_laneq", "1QI", "ScSsSiSlSfSdSUcSUsSUiSUlSPcSPs">;
 }
 
 // ARMv8.2-A FP16 vector intrinsics for A32/A64.
 let ArchGuard = "defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC)" in {
 
   // ARMv8.2-A FP16 one-operand vector intrinsics.
 
   // Comparison
-  def CMEQH    : SInst<"vceqz", "ud", "hQh">;
-  def CMGEH    : SInst<"vcgez", "ud", "hQh">;
-  def CMGTH    : SInst<"vcgtz", "ud", "hQh">;
-  def CMLEH    : SInst<"vclez", "ud", "hQh">;
-  def CMLTH    : SInst<"vcltz", "ud", "hQh">;
+  def CMEQH    : SInst<"vceqz", "U.", "hQh">;
+  def CMGEH    : SInst<"vcgez", "U.", "hQh">;
+  def CMGTH    : SInst<"vcgtz", "U.", "hQh">;
+  def CMLEH    : SInst<"vclez", "U.", "hQh">;
+  def CMLTH    : SInst<"vcltz", "U.", "hQh">;
 
   // Vector conversion
-  def VCVT_F16     : SInst<"vcvt_f16", "Hd",  "sUsQsQUs">;
-  def VCVT_S16     : SInst<"vcvt_s16", "xd",  "hQh">;
-  def VCVT_U16     : SInst<"vcvt_u16", "ud",  "hQh">;
-  def VCVTA_S16    : SInst<"vcvta_s16", "xd", "hQh">;
-  def VCVTA_U16    : SInst<"vcvta_u16", "ud", "hQh">;
-  def VCVTM_S16    : SInst<"vcvtm_s16", "xd", "hQh">;
-  def VCVTM_U16    : SInst<"vcvtm_u16", "ud", "hQh">;
-  def VCVTN_S16    : SInst<"vcvtn_s16", "xd", "hQh">;
-  def VCVTN_U16    : SInst<"vcvtn_u16", "ud", "hQh">;
-  def VCVTP_S16    : SInst<"vcvtp_s16", "xd", "hQh">;
-  def VCVTP_U16    : SInst<"vcvtp_u16", "ud", "hQh">;
+  def VCVT_F16     : SInst<"vcvt_f16", "F(.!)",  "sUsQsQUs">;
+  def VCVT_S16     : SInst<"vcvt_s16", "S.",  "hQh">;
+  def VCVT_U16     : SInst<"vcvt_u16", "U.",  "hQh">;
+  def VCVTA_S16    : SInst<"vcvta_s16", "S.", "hQh">;
+  def VCVTA_U16    : SInst<"vcvta_u16", "U.", "hQh">;
+  def VCVTM_S16    : SInst<"vcvtm_s16", "S.", "hQh">;
+  def VCVTM_U16    : SInst<"vcvtm_u16", "U.", "hQh">;
+  def VCVTN_S16    : SInst<"vcvtn_s16", "S.", "hQh">;
+  def VCVTN_U16    : SInst<"vcvtn_u16", "U.", "hQh">;
+  def VCVTP_S16    : SInst<"vcvtp_s16", "S.", "hQh">;
+  def VCVTP_U16    : SInst<"vcvtp_u16", "U.", "hQh">;
 
   // Vector rounding
   let ArchGuard = "__ARM_ARCH >= 8 && defined(__ARM_FEATURE_DIRECTED_ROUNDING) && defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC)" in {
-    def FRINTZH      : SInst<"vrnd",  "dd", "hQh">;
-    def FRINTNH      : SInst<"vrndn", "dd", "hQh">;
-    def FRINTAH      : SInst<"vrnda", "dd", "hQh">;
-    def FRINTPH      : SInst<"vrndp", "dd", "hQh">;
-    def FRINTMH      : SInst<"vrndm", "dd", "hQh">;
-    def FRINTXH      : SInst<"vrndx", "dd", "hQh">;
+    def FRINTZH      : SInst<"vrnd",  "..", "hQh">;
+    def FRINTNH      : SInst<"vrndn", "..", "hQh">;
+    def FRINTAH      : SInst<"vrnda", "..", "hQh">;
+    def FRINTPH      : SInst<"vrndp", "..", "hQh">;
+    def FRINTMH      : SInst<"vrndm", "..", "hQh">;
+    def FRINTXH      : SInst<"vrndx", "..", "hQh">;
   }
 
   // Misc.
-  def VABSH        : SInst<"vabs", "dd", "hQh">;
-  def VNEGH        : SOpInst<"vneg", "dd", "hQh", OP_NEG>;
-  def VRECPEH      : SInst<"vrecpe", "dd", "hQh">;
-  def FRSQRTEH     : SInst<"vrsqrte", "dd", "hQh">;
+  def VABSH        : SInst<"vabs", "..", "hQh">;
+  def VNEGH        : SOpInst<"vneg", "..", "hQh", OP_NEG>;
+  def VRECPEH      : SInst<"vrecpe", "..", "hQh">;
+  def FRSQRTEH     : SInst<"vrsqrte", "..", "hQh">;
 
   // ARMv8.2-A FP16 two-operands vector intrinsics.
 
   // Misc.
-  def VADDH        : SOpInst<"vadd", "ddd", "hQh", OP_ADD>;
-  def VABDH        : SInst<"vabd", "ddd",  "hQh">;
-  def VSUBH         : SOpInst<"vsub", "ddd", "hQh", OP_SUB>;
+  def VADDH        : SOpInst<"vadd", "...", "hQh", OP_ADD>;
+  def VABDH        : SInst<"vabd", "...",  "hQh">;
+  def VSUBH         : SOpInst<"vsub", "...", "hQh", OP_SUB>;
 
   // Comparison
   let InstName = "vacge" in {
-    def VCAGEH     : SInst<"vcage", "udd", "hQh">;
-    def VCALEH     : SInst<"vcale", "udd", "hQh">;
+    def VCAGEH     : SInst<"vcage", "U..", "hQh">;
+    def VCALEH     : SInst<"vcale", "U..", "hQh">;
   }
   let InstName = "vacgt" in {
-    def VCAGTH     : SInst<"vcagt", "udd", "hQh">;
-    def VCALTH     : SInst<"vcalt", "udd", "hQh">;
+    def VCAGTH     : SInst<"vcagt", "U..", "hQh">;
+    def VCALTH     : SInst<"vcalt", "U..", "hQh">;
   }
-  def VCEQH        : SOpInst<"vceq", "udd", "hQh", OP_EQ>;
-  def VCGEH        : SOpInst<"vcge", "udd", "hQh", OP_GE>;
-  def VCGTH        : SOpInst<"vcgt", "udd", "hQh", OP_GT>;
+  def VCEQH        : SOpInst<"vceq", "U..", "hQh", OP_EQ>;
+  def VCGEH        : SOpInst<"vcge", "U..", "hQh", OP_GE>;
+  def VCGTH        : SOpInst<"vcgt", "U..", "hQh", OP_GT>;
   let InstName = "vcge" in
-    def VCLEH      : SOpInst<"vcle", "udd", "hQh", OP_LE>;
+    def VCLEH      : SOpInst<"vcle", "U..", "hQh", OP_LE>;
   let InstName = "vcgt" in
-    def VCLTH      : SOpInst<"vclt", "udd", "hQh", OP_LT>;
+    def VCLTH      : SOpInst<"vclt", "U..", "hQh", OP_LT>;
 
   // Vector conversion
   let isVCVT_N = 1 in {
-    def VCVT_N_F16 : SInst<"vcvt_n_f16", "Hdi", "sUsQsQUs">;
-    def VCVT_N_S16 : SInst<"vcvt_n_s16", "xdi", "hQh">;
-    def VCVT_N_U16 : SInst<"vcvt_n_u16", "udi", "hQh">;
+    def VCVT_N_F16 : SInst<"vcvt_n_f16", "F(.!)I", "sUsQsQUs">;
+    def VCVT_N_S16 : SInst<"vcvt_n_s16", "S.I", "hQh">;
+    def VCVT_N_U16 : SInst<"vcvt_n_u16", "U.I", "hQh">;
   }
 
   // Max/Min
-  def VMAXH         : SInst<"vmax", "ddd", "hQh">;
-  def VMINH         : SInst<"vmin", "ddd", "hQh">;
+  def VMAXH         : SInst<"vmax", "...", "hQh">;
+  def VMINH         : SInst<"vmin", "...", "hQh">;
   let ArchGuard = "__ARM_ARCH >= 8 && defined(__ARM_FEATURE_NUMERIC_MAXMIN) && defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC)" in {
-    def FMAXNMH       : SInst<"vmaxnm", "ddd", "hQh">;
-    def FMINNMH       : SInst<"vminnm", "ddd", "hQh">;
+    def FMAXNMH       : SInst<"vmaxnm", "...", "hQh">;
+    def FMINNMH       : SInst<"vminnm", "...", "hQh">;
   }
 
   // Multiplication/Division
-  def VMULH         : SOpInst<"vmul", "ddd", "hQh", OP_MUL>;
+  def VMULH         : SOpInst<"vmul", "...", "hQh", OP_MUL>;
 
   // Pairwise addition
-  def VPADDH        : SInst<"vpadd", "ddd", "h">;
+  def VPADDH        : SInst<"vpadd", "...", "h">;
 
   // Pairwise Max/Min
-  def VPMAXH        : SInst<"vpmax", "ddd", "h">;
-  def VPMINH        : SInst<"vpmin", "ddd", "h">;
+  def VPMAXH        : SInst<"vpmax", "...", "h">;
+  def VPMINH        : SInst<"vpmin", "...", "h">;
 
   // Reciprocal/Sqrt
-  def VRECPSH       : SInst<"vrecps", "ddd", "hQh">;
-  def VRSQRTSH      : SInst<"vrsqrts", "ddd", "hQh">;
+  def VRECPSH       : SInst<"vrecps", "...", "hQh">;
+  def VRSQRTSH      : SInst<"vrsqrts", "...", "hQh">;
 
   // ARMv8.2-A FP16 three-operands vector intrinsics.
 
   // Vector fused multiply-add operations
-  def VFMAH        : SInst<"vfma", "dddd", "hQh">;
-  def VFMSH        : SOpInst<"vfms", "dddd", "hQh", OP_FMLS>;
+  def VFMAH        : SInst<"vfma", "....", "hQh">;
+  def VFMSH        : SOpInst<"vfms", "....", "hQh", OP_FMLS>;
 
   // ARMv8.2-A FP16 lane vector intrinsics.
 
   // Mul lane
-  def VMUL_LANEH    : IOpInst<"vmul_lane", "ddgi", "hQh", OP_MUL_LN>;
-  def VMUL_NH       : IOpInst<"vmul_n", "dds", "hQh", OP_MUL_N>;
+  def VMUL_LANEH    : IOpInst<"vmul_lane", "..qI", "hQh", OP_MUL_LN>;
+  def VMUL_NH       : IOpInst<"vmul_n", "..1", "hQh", OP_MUL_N>;
 
   // Data processing intrinsics - section 5
 
   // Logical operations
   let isHiddenLInst = 1 in
-  def VBSLH    : SInst<"vbsl", "dudd", "hQh">;
+  def VBSLH    : SInst<"vbsl", ".U..", "hQh">;
 
   // Transposition operations
-  def VZIPH    : WInst<"vzip", "2dd", "hQh">;
-  def VUZPH    : WInst<"vuzp", "2dd", "hQh">;
-  def VTRNH    : WInst<"vtrn", "2dd", "hQh">;
+  def VZIPH    : WInst<"vzip", "2..", "hQh">;
+  def VUZPH    : WInst<"vuzp", "2..", "hQh">;
+  def VTRNH    : WInst<"vtrn", "2..", "hQh">;
 
 
   let ArchGuard = "!defined(__aarch64__)" in {
     // Set all lanes to same value.
     // Already implemented prior to ARMv8.2-A.
-    def VMOV_NH  : WOpInst<"vmov_n", "ds", "hQh", OP_DUP>;
-    def VDUP_NH  : WOpInst<"vdup_n", "ds", "hQh", OP_DUP>;
-    def VDUP_LANE1H : WOpInst<"vdup_lane", "dgi", "hQh", OP_DUP_LN>;
+    def VMOV_NH  : WOpInst<"vmov_n", ".1", "hQh", OP_DUP>;
+    def VDUP_NH  : WOpInst<"vdup_n", ".1", "hQh", OP_DUP>;
+    def VDUP_LANE1H : WOpInst<"vdup_lane", ".qI", "hQh", OP_DUP_LN>;
   }
 
   // Vector Extract
-  def VEXTH      : WInst<"vext", "dddi", "hQh">;
+  def VEXTH      : WInst<"vext", "...I", "hQh">;
 
   // Reverse vector elements
-  def VREV64H    : WOpInst<"vrev64", "dd", "hQh", OP_REV64>;
+  def VREV64H    : WOpInst<"vrev64", "..", "hQh", OP_REV64>;
 }
 
 // ARMv8.2-A FP16 vector intrinsics for A64 only.
 let ArchGuard = "defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) && defined(__aarch64__)" in {
 
   // Vector rounding
-  def FRINTIH      : SInst<"vrndi", "dd", "hQh">;
+  def FRINTIH      : SInst<"vrndi", "..", "hQh">;
 
   // Misc.
-  def FSQRTH       : SInst<"vsqrt", "dd", "hQh">;
+  def FSQRTH       : SInst<"vsqrt", "..", "hQh">;
 
   // Multiplication/Division
-  def MULXH         : SInst<"vmulx", "ddd", "hQh">;
-  def FDIVH         : IOpInst<"vdiv", "ddd",  "hQh", OP_DIV>;
+  def MULXH         : SInst<"vmulx", "...", "hQh">;
+  def FDIVH         : IOpInst<"vdiv", "...",  "hQh", OP_DIV>;
 
   // Pairwise addition
-  def VPADDH1       : SInst<"vpadd", "ddd", "Qh">;
+  def VPADDH1       : SInst<"vpadd", "...", "Qh">;
 
   // Pairwise Max/Min
-  def VPMAXH1       : SInst<"vpmax", "ddd", "Qh">;
-  def VPMINH1       : SInst<"vpmin", "ddd", "Qh">;
+  def VPMAXH1       : SInst<"vpmax", "...", "Qh">;
+  def VPMINH1       : SInst<"vpmin", "...", "Qh">;
 
   // Pairwise MaxNum/MinNum
-  def FMAXNMPH      : SInst<"vpmaxnm", "ddd", "hQh">;
-  def FMINNMPH      : SInst<"vpminnm", "ddd", "hQh">;
+  def FMAXNMPH      : SInst<"vpmaxnm", "...", "hQh">;
+  def FMINNMPH      : SInst<"vpminnm", "...", "hQh">;
 
   // ARMv8.2-A FP16 lane vector intrinsics.
 
   // FMA lane
-  def VFMA_LANEH   : IInst<"vfma_lane", "dddgi", "hQh">;
-  def VFMA_LANEQH  : IInst<"vfma_laneq", "dddji", "hQh">;
+  def VFMA_LANEH   : IInst<"vfma_lane", "...qI", "hQh">;
+  def VFMA_LANEQH  : IInst<"vfma_laneq", "...QI", "hQh">;
 
   // FMA lane with scalar argument
-  def FMLA_NH      : SOpInst<"vfma_n", "ddds", "hQh", OP_FMLA_N>;
+  def FMLA_NH      : SOpInst<"vfma_n", "...1", "hQh", OP_FMLA_N>;
   // Scalar floating point fused multiply-add (scalar, by element)
-  def SCALAR_FMLA_LANEH  : IInst<"vfma_lane", "sssdi", "Sh">;
-  def SCALAR_FMLA_LANEQH : IInst<"vfma_laneq", "sssji", "Sh">;
+  def SCALAR_FMLA_LANEH  : IInst<"vfma_lane", "111.I", "Sh">;
+  def SCALAR_FMLA_LANEQH : IInst<"vfma_laneq", "111QI", "Sh">;
 
   // FMS lane
-  def VFMS_LANEH   : IOpInst<"vfms_lane", "dddgi", "hQh", OP_FMS_LN>;
-  def VFMS_LANEQH  : IOpInst<"vfms_laneq", "dddji", "hQh", OP_FMS_LNQ>;
+  def VFMS_LANEH   : IOpInst<"vfms_lane", "...qI", "hQh", OP_FMS_LN>;
+  def VFMS_LANEQH  : IOpInst<"vfms_laneq", "...QI", "hQh", OP_FMS_LNQ>;
   // FMS lane with scalar argument
-  def FMLS_NH      : SOpInst<"vfms_n", "ddds", "hQh", OP_FMLS_N>;
+  def FMLS_NH      : SOpInst<"vfms_n", "...1", "hQh", OP_FMLS_N>;
   // Scalar floating foint fused multiply-subtract (scalar, by element)
-  def SCALAR_FMLS_LANEH  : IOpInst<"vfms_lane", "sssdi", "Sh", OP_FMS_LN>;
-  def SCALAR_FMLS_LANEQH : IOpInst<"vfms_laneq", "sssji", "Sh", OP_FMS_LNQ>;
+  def SCALAR_FMLS_LANEH  : IOpInst<"vfms_lane", "111.I", "Sh", OP_FMS_LN>;
+  def SCALAR_FMLS_LANEQH : IOpInst<"vfms_laneq", "111QI", "Sh", OP_FMS_LNQ>;
 
   // Mul lane
-  def VMUL_LANEQH   : IOpInst<"vmul_laneq", "ddji", "hQh", OP_MUL_LN>;
+  def VMUL_LANEQH   : IOpInst<"vmul_laneq", "..QI", "hQh", OP_MUL_LN>;
   // Scalar floating point  multiply (scalar, by element)
-  def SCALAR_FMUL_LANEH  : IOpInst<"vmul_lane", "ssdi", "Sh", OP_SCALAR_MUL_LN>;
-  def SCALAR_FMUL_LANEQH : IOpInst<"vmul_laneq", "ssji", "Sh", OP_SCALAR_MUL_LN>;
+  def SCALAR_FMUL_LANEH  : IOpInst<"vmul_lane", "11.I", "Sh", OP_SCALAR_MUL_LN>;
+  def SCALAR_FMUL_LANEQH : IOpInst<"vmul_laneq", "11QI", "Sh", OP_SCALAR_MUL_LN>;
 
   // Mulx lane
-  def VMULX_LANEH   : IOpInst<"vmulx_lane", "ddgi", "hQh", OP_MULX_LN>;
-  def VMULX_LANEQH  : IOpInst<"vmulx_laneq", "ddji", "hQh", OP_MULX_LN>;
-  def VMULX_NH      : IOpInst<"vmulx_n", "dds", "hQh", OP_MULX_N>;
+  def VMULX_LANEH   : IOpInst<"vmulx_lane", "..qI", "hQh", OP_MULX_LN>;
+  def VMULX_LANEQH  : IOpInst<"vmulx_laneq", "..QI", "hQh", OP_MULX_LN>;
+  def VMULX_NH      : IOpInst<"vmulx_n", "..1", "hQh", OP_MULX_N>;
   // Scalar floating point  mulx (scalar, by element)
-  def SCALAR_FMULX_LANEH : IInst<"vmulx_lane", "ssdi", "Sh">;
-  def SCALAR_FMULX_LANEQH : IInst<"vmulx_laneq", "ssji", "Sh">;
+  def SCALAR_FMULX_LANEH : IInst<"vmulx_lane", "11.I", "Sh">;
+  def SCALAR_FMULX_LANEQH : IInst<"vmulx_laneq", "11QI", "Sh">;
 
   // ARMv8.2-A FP16 reduction vector intrinsics.
-  def VMAXVH   : SInst<"vmaxv", "sd", "hQh">;
-  def VMINVH   : SInst<"vminv", "sd", "hQh">;
-  def FMAXNMVH : SInst<"vmaxnmv", "sd", "hQh">;
-  def FMINNMVH : SInst<"vminnmv", "sd", "hQh">;
+  def VMAXVH   : SInst<"vmaxv", "1.", "hQh">;
+  def VMINVH   : SInst<"vminv", "1.", "hQh">;
+  def FMAXNMVH : SInst<"vmaxnmv", "1.", "hQh">;
+  def FMINNMVH : SInst<"vminnmv", "1.", "hQh">;
 
   // Permutation
-  def VTRN1H     : SOpInst<"vtrn1", "ddd", "hQh", OP_TRN1>;
-  def VZIP1H     : SOpInst<"vzip1", "ddd", "hQh", OP_ZIP1>;
-  def VUZP1H     : SOpInst<"vuzp1", "ddd", "hQh", OP_UZP1>;
-  def VTRN2H     : SOpInst<"vtrn2", "ddd", "hQh", OP_TRN2>;
-  def VZIP2H     : SOpInst<"vzip2", "ddd", "hQh", OP_ZIP2>;
-  def VUZP2H     : SOpInst<"vuzp2", "ddd", "hQh", OP_UZP2>;
+  def VTRN1H     : SOpInst<"vtrn1", "...", "hQh", OP_TRN1>;
+  def VZIP1H     : SOpInst<"vzip1", "...", "hQh", OP_ZIP1>;
+  def VUZP1H     : SOpInst<"vuzp1", "...", "hQh", OP_UZP1>;
+  def VTRN2H     : SOpInst<"vtrn2", "...", "hQh", OP_TRN2>;
+  def VZIP2H     : SOpInst<"vzip2", "...", "hQh", OP_ZIP2>;
+  def VUZP2H     : SOpInst<"vuzp2", "...", "hQh", OP_UZP2>;
 
-  def SCALAR_VDUP_LANEH  : IInst<"vdup_lane", "sdi", "Sh">;
-  def SCALAR_VDUP_LANEQH : IInst<"vdup_laneq", "sji", "Sh">;
+  def SCALAR_VDUP_LANEH  : IInst<"vdup_lane", "1.I", "Sh">;
+  def SCALAR_VDUP_LANEQH : IInst<"vdup_laneq", "1QI", "Sh">;
 }
 
 // v8.2-A dot product instructions.
 let ArchGuard = "defined(__ARM_FEATURE_DOTPROD)" in {
-  def DOT : SInst<"vdot", "dd88", "iQiUiQUi">;
-  def DOT_LANE : SOpInst<"vdot_lane", "dd87i", "iUiQiQUi", OP_DOT_LN>;
+  def DOT : SInst<"vdot", "..(<<)(<<)", "iQiUiQUi">;
+  def DOT_LANE : SOpInst<"vdot_lane", "..(<<)(<<q)I", "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>;
+  def UDOT_LANEQ : SOpInst<"vdot_laneq", "..(<<)(<<Q)I", "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", "hQh">;
-  def VFMLSL_LOW  : SInst<"vfmlsl_low",  "ffHH", "hQh">;
-  def VFMLAL_HIGH : SInst<"vfmlal_high", "ffHH", "hQh">;
-  def VFMLSL_HIGH : SInst<"vfmlsl_high", "ffHH", "hQh">;
-
-  def VFMLAL_LANE_LOW  : SOpInst<"vfmlal_lane_low",  "ffH0i", "hQh", OP_FMLAL_LN>;
-  def VFMLSL_LANE_LOW  : SOpInst<"vfmlsl_lane_low",  "ffH0i", "hQh", OP_FMLSL_LN>;
-  def VFMLAL_LANE_HIGH : SOpInst<"vfmlal_lane_high", "ffH0i", "hQh", OP_FMLAL_LN_Hi>;
-  def VFMLSL_LANE_HIGH : SOpInst<"vfmlsl_lane_high", "ffH0i", "hQh", OP_FMLSL_LN_Hi>;
-
-  def VFMLAL_LANEQ_LOW  : SOpInst<"vfmlal_laneq_low",  "ffH1i", "hQh", OP_FMLAL_LN>;
-  def VFMLSL_LANEQ_LOW  : SOpInst<"vfmlsl_laneq_low",  "ffH1i", "hQh", OP_FMLSL_LN>;
-  def VFMLAL_LANEQ_HIGH : SOpInst<"vfmlal_laneq_high", "ffH1i", "hQh", OP_FMLAL_LN_Hi>;
-  def VFMLSL_LANEQ_HIGH : SOpInst<"vfmlsl_laneq_high", "ffH1i", "hQh", OP_FMLSL_LN_Hi>;
+  def VFMLAL_LOW  : SInst<"vfmlal_low",  "(F>)(F>)FF", "hQh">;
+  def VFMLSL_LOW  : SInst<"vfmlsl_low",  "(F>)(F>)FF", "hQh">;
+  def VFMLAL_HIGH : SInst<"vfmlal_high", "(F>)(F>)FF", "hQh">;
+  def VFMLSL_HIGH : SInst<"vfmlsl_high", "(F>)(F>)FF", "hQh">;
+
+  def VFMLAL_LANE_LOW  : SOpInst<"vfmlal_lane_low",  "(F>)(F>)F(Fq)I", "hQh", OP_FMLAL_LN>;
+  def VFMLSL_LANE_LOW  : SOpInst<"vfmlsl_lane_low",  "(F>)(F>)F(Fq)I", "hQh", OP_FMLSL_LN>;
+  def VFMLAL_LANE_HIGH : SOpInst<"vfmlal_lane_high", "(F>)(F>)F(Fq)I", "hQh", OP_FMLAL_LN_Hi>;
+  def VFMLSL_LANE_HIGH : SOpInst<"vfmlsl_lane_high", "(F>)(F>)F(Fq)I", "hQh", OP_FMLSL_LN_Hi>;
+
+  def VFMLAL_LANEQ_LOW  : SOpInst<"vfmlal_laneq_low",  "(F>)(F>)F(FQ)I", "hQh", OP_FMLAL_LN>;
+  def VFMLSL_LANEQ_LOW  : SOpInst<"vfmlsl_laneq_low",  "(F>)(F>)F(FQ)I", "hQh", OP_FMLSL_LN>;
+  def VFMLAL_LANEQ_HIGH : SOpInst<"vfmlal_laneq_high", "(F>)(F>)F(FQ)I", "hQh", OP_FMLAL_LN_Hi>;
+  def VFMLSL_LANEQ_HIGH : SOpInst<"vfmlsl_laneq_high", "(F>)(F>)F(FQ)I", "hQh", OP_FMLSL_LN_Hi>;
 }

clang/include/clang/Basic/arm_neon_incl.td

 //
 // The string given to an Inst class is a sequence of typespecs. The intrinsic
 // is instantiated for every typespec in the sequence. For example "sdQsQd".
 //
 // The prototype is a string that defines the return type of the intrinsic
 // and the type of each argument. The return type and every argument gets a
 // "modifier" that can change in some way the "base type" of the intrinsic.
 //
 // The modifier 'd' means "default" and does not modify the base type in any
 // way. The available modifiers are given below.
 //
 // Typespecs
 // ---------
 // c: char
 // s: short
 // i: int
 // l: long

 // Q: 128b
 // H: 128b without mangling 'q'
 // P: polynomial
 //
 // Prototype modifiers
 // -------------------
 // prototype: return (arg, arg, ...)
 //
-// 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
-// m: half width elements, same num elts
-// i: constant int
-// l: constant uint64
-// s: scalar of element type
-// z: scalar of half width element type, signed
-// r: scalar of double width element type, signed
-// b: scalar of unsigned integer/long type (int/float args)
-// $: scalar of signed integer/long type (int/float args)
-// y: scalar of float
-// o: scalar of double
-// k: default elt width, double num elts
-// 2,3,4: array of default vectors
-// B,C,D: array of default elts, force 'Q' size modifier.
-// p: pointer type
-// c: const pointer type
-// 7: vector of 8-bit elements, ignore 'Q' size modifier
-// 8: vector of 8-bit elements, same width as default type
-// 9: vector of 8-bit elements, force 'Q' size modifier
+// Each type modifier is either a single character, or a group surrounded by
+// parentheses.
+//
+// .: default
+// v: change to void category.
+// S: change to signed integer category.
+// U: change to unsigned integer category.
+// F: change to floating category.
+// P: change to polynomial category.
+// p: change polynomial to equivalent integer category. Otherwise nop.
+//
+// >: double element width (vector size unchanged).
+// <: half element width (vector size unchanged).
+//
+// 1: change to scalar.
+// 2: change to struct of two vectors.
+// 3: change to struct of three vectors.
+// 4: change to struct of four vectors.
+//
+// *: make a pointer argument.
+// c: make a constant argument (for pointers).
+//
+// Q: force 128-bit width.
+// q: force 64-bit width.
+//
+// I: make 32-bit signed scalar immediate
+// !: make this the key type passed to CGBuiltin.cpp in a polymorphic call.
+
 
 // Every intrinsic subclasses Inst.
 class Inst <string n, string p, string t, Operation o> {
   string Name = n;
   string Prototype = p;
   string Types = t;
   string ArchGuard = "";
 

clang/lib/CodeGen/CGBuiltin.cpp

Value *CodeGenFunction::EmitCommonNeonBuiltinExpr(
   case NEON::BI__builtin_neon_vcvtmq_s32_v:
   case NEON::BI__builtin_neon_vcvtmq_s64_v:
   case NEON::BI__builtin_neon_vcvtmq_u16_v:
   case NEON::BI__builtin_neon_vcvtmq_u32_v:
   case NEON::BI__builtin_neon_vcvtmq_u64_v: {
     llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) };
     return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, NameHint);
   }
+  case NEON::BI__builtin_neon_vcvtx_f32_v: {
+    llvm::Type *Tys[2] = { VTy->getTruncatedElementVectorType(VTy), Ty};
+    return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, NameHint);
+
+  }
   case NEON::BI__builtin_neon_vext_v:
   case NEON::BI__builtin_neon_vextq_v: {
     int CV = cast<ConstantInt>(Ops[2])->getSExtValue();
     SmallVector<uint32_t, 16> Indices;
     for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i)
       Indices.push_back(i+CV);
 
     Ops[0] = Builder.CreateBitCast(Ops[0], Ty);

clang/test/CodeGen/aarch64-neon-intrinsics.c

 // CHECK-LABEL: @test_vqrshld_u64(
 // CHECK:   [[VQRSHLD_U64_I:%.*]] = call i64 @llvm.aarch64.neon.uqrshl.i64(i64 %a, i64 %b)
 // CHECK:   ret i64 [[VQRSHLD_U64_I]]
 uint64_t test_vqrshld_u64(uint64_t a, uint64_t b) {
   return vqrshld_u64(a, b);
 }
 
 // CHECK-LABEL: @test_vpaddd_s64(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i64> %a to <16 x i8>
 // CHECK:   [[VPADDD_S64_I:%.*]] = call i64 @llvm.aarch64.neon.uaddv.i64.v2i64(<2 x i64> %a)
 // CHECK:   ret i64 [[VPADDD_S64_I]]
 int64_t test_vpaddd_s64(int64x2_t a) {
   return vpaddd_s64(a);
 }
 
 // CHECK-LABEL: @test_vpadds_f32(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x float> %a to <8 x i8>
 // CHECK:   [[LANE0_I:%.*]] = extractelement <2 x float> %a, i64 0
 // CHECK:   [[LANE1_I:%.*]] = extractelement <2 x float> %a, i64 1
 // CHECK:   [[VPADDD_I:%.*]] = fadd float [[LANE0_I]], [[LANE1_I]]
 // CHECK:   ret float [[VPADDD_I]]
 float32_t test_vpadds_f32(float32x2_t a) {
   return vpadds_f32(a);
 }
 
 // CHECK-LABEL: @test_vpaddd_f64(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x double> %a to <16 x i8>
 // CHECK:   [[LANE0_I:%.*]] = extractelement <2 x double> %a, i64 0
 // CHECK:   [[LANE1_I:%.*]] = extractelement <2 x double> %a, i64 1
 // CHECK:   [[VPADDD_I:%.*]] = fadd double [[LANE0_I]], [[LANE1_I]]
 // CHECK:   ret double [[VPADDD_I]]
 float64_t test_vpaddd_f64(float64x2_t a) {
   return vpaddd_f64(a);
 }
 
 // CHECK-LABEL: @test_vpmaxnms_f32(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x float> %a to <8 x i8>
 // CHECK:   [[VPMAXNMS_F32_I:%.*]] = call float @llvm.aarch64.neon.fmaxnmv.f32.v2f32(<2 x float> %a)
 // CHECK:   ret float [[VPMAXNMS_F32_I]]
 float32_t test_vpmaxnms_f32(float32x2_t a) {
   return vpmaxnms_f32(a);
 }
 
 // CHECK-LABEL: @test_vpmaxnmqd_f64(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x double> %a to <16 x i8>
 // CHECK:   [[VPMAXNMQD_F64_I:%.*]] = call double @llvm.aarch64.neon.fmaxnmv.f64.v2f64(<2 x double> %a)
 // CHECK:   ret double [[VPMAXNMQD_F64_I]]
 float64_t test_vpmaxnmqd_f64(float64x2_t a) {
   return vpmaxnmqd_f64(a);
 }
 
 // CHECK-LABEL: @test_vpmaxs_f32(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x float> %a to <8 x i8>
 // CHECK:   [[VPMAXS_F32_I:%.*]] = call float @llvm.aarch64.neon.fmaxv.f32.v2f32(<2 x float> %a)
 // CHECK:   ret float [[VPMAXS_F32_I]]
 float32_t test_vpmaxs_f32(float32x2_t a) {
   return vpmaxs_f32(a);
 }
 
 // CHECK-LABEL: @test_vpmaxqd_f64(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x double> %a to <16 x i8>
 // CHECK:   [[VPMAXQD_F64_I:%.*]] = call double @llvm.aarch64.neon.fmaxv.f64.v2f64(<2 x double> %a)
 // CHECK:   ret double [[VPMAXQD_F64_I]]
 float64_t test_vpmaxqd_f64(float64x2_t a) {
   return vpmaxqd_f64(a);
 }
 
 // CHECK-LABEL: @test_vpminnms_f32(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x float> %a to <8 x i8>
 // CHECK:   [[VPMINNMS_F32_I:%.*]] = call float @llvm.aarch64.neon.fminnmv.f32.v2f32(<2 x float> %a)
 // CHECK:   ret float [[VPMINNMS_F32_I]]
 float32_t test_vpminnms_f32(float32x2_t a) {
   return vpminnms_f32(a);
 }
 
 // CHECK-LABEL: @test_vpminnmqd_f64(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x double> %a to <16 x i8>
 // CHECK:   [[VPMINNMQD_F64_I:%.*]] = call double @llvm.aarch64.neon.fminnmv.f64.v2f64(<2 x double> %a)
 // CHECK:   ret double [[VPMINNMQD_F64_I]]
 float64_t test_vpminnmqd_f64(float64x2_t a) {
   return vpminnmqd_f64(a);
 }
 
 // CHECK-LABEL: @test_vpmins_f32(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x float> %a to <8 x i8>
 // CHECK:   [[VPMINS_F32_I:%.*]] = call float @llvm.aarch64.neon.fminv.f32.v2f32(<2 x float> %a)
 // CHECK:   ret float [[VPMINS_F32_I]]
 float32_t test_vpmins_f32(float32x2_t a) {
   return vpmins_f32(a);
 }
 
 // CHECK-LABEL: @test_vpminqd_f64(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x double> %a to <16 x i8>
 // CHECK:   [[VPMINQD_F64_I:%.*]] = call double @llvm.aarch64.neon.fminv.f64.v2f64(<2 x double> %a)
 // CHECK:   ret double [[VPMINQD_F64_I]]
 float64_t test_vpminqd_f64(float64x2_t a) {
   return vpminqd_f64(a);
 }
 
 // CHECK-LABEL: @test_vqdmulhh_s16(
 // CHECK:   [[TMP0:%.*]] = insertelement <4 x i16> undef, i16 %a, i64 0

 // CHECK-LABEL: @test_vneg_s64(
 // CHECK:   [[SUB_I:%.*]] = sub <1 x i64> zeroinitializer, %a
 // CHECK:   ret <1 x i64> [[SUB_I]]
 int64x1_t test_vneg_s64(int64x1_t a) {
   return vneg_s64(a);
 }
 
 // CHECK-LABEL: @test_vaddv_f32(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x float> %a to <8 x i8>
 // CHECK:   [[VADDV_F32_I:%.*]] = call float @llvm.aarch64.neon.faddv.f32.v2f32(<2 x float> %a)
 // CHECK:   ret float [[VADDV_F32_I]]
 float32_t test_vaddv_f32(float32x2_t a) {
   return vaddv_f32(a);
 }
 
 // CHECK-LABEL: @test_vaddvq_f32(
-// CHECK:   [[TMP0:%.*]] = bitcast <4 x float> %a to <16 x i8>
 // CHECK:   [[VADDVQ_F32_I:%.*]] = call float @llvm.aarch64.neon.faddv.f32.v4f32(<4 x float> %a)
 // CHECK:   ret float [[VADDVQ_F32_I]]
 float32_t test_vaddvq_f32(float32x4_t a) {
   return vaddvq_f32(a);
 }
 
 // CHECK-LABEL: @test_vaddvq_f64(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x double> %a to <16 x i8>
 // CHECK:   [[VADDVQ_F64_I:%.*]] = call double @llvm.aarch64.neon.faddv.f64.v2f64(<2 x double> %a)
 // CHECK:   ret double [[VADDVQ_F64_I]]
 float64_t test_vaddvq_f64(float64x2_t a) {
   return vaddvq_f64(a);
 }
 
 // CHECK-LABEL: @test_vmaxv_f32(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x float> %a to <8 x i8>
 // CHECK:   [[VMAXV_F32_I:%.*]] = call float @llvm.aarch64.neon.fmaxv.f32.v2f32(<2 x float> %a)
 // CHECK:   ret float [[VMAXV_F32_I]]
 float32_t test_vmaxv_f32(float32x2_t a) {
   return vmaxv_f32(a);
 }
 
 // CHECK-LABEL: @test_vmaxvq_f64(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x double> %a to <16 x i8>
 // CHECK:   [[VMAXVQ_F64_I:%.*]] = call double @llvm.aarch64.neon.fmaxv.f64.v2f64(<2 x double> %a)
 // CHECK:   ret double [[VMAXVQ_F64_I]]
 float64_t test_vmaxvq_f64(float64x2_t a) {
   return vmaxvq_f64(a);
 }
 
 // CHECK-LABEL: @test_vminv_f32(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x float> %a to <8 x i8>
 // CHECK:   [[VMINV_F32_I:%.*]] = call float @llvm.aarch64.neon.fminv.f32.v2f32(<2 x float> %a)
 // CHECK:   ret float [[VMINV_F32_I]]
 float32_t test_vminv_f32(float32x2_t a) {
   return vminv_f32(a);
 }
 
 // CHECK-LABEL: @test_vminvq_f64(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x double> %a to <16 x i8>
 // CHECK:   [[VMINVQ_F64_I:%.*]] = call double @llvm.aarch64.neon.fminv.f64.v2f64(<2 x double> %a)
 // CHECK:   ret double [[VMINVQ_F64_I]]
 float64_t test_vminvq_f64(float64x2_t a) {
   return vminvq_f64(a);
 }
 
 // CHECK-LABEL: @test_vmaxnmvq_f64(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x double> %a to <16 x i8>
 // CHECK:   [[VMAXNMVQ_F64_I:%.*]] = call double @llvm.aarch64.neon.fmaxnmv.f64.v2f64(<2 x double> %a)
 // CHECK:   ret double [[VMAXNMVQ_F64_I]]
 float64_t test_vmaxnmvq_f64(float64x2_t a) {
   return vmaxnmvq_f64(a);
 }
 
 // CHECK-LABEL: @test_vmaxnmv_f32(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x float> %a to <8 x i8>
 // CHECK:   [[VMAXNMV_F32_I:%.*]] = call float @llvm.aarch64.neon.fmaxnmv.f32.v2f32(<2 x float> %a)
 // CHECK:   ret float [[VMAXNMV_F32_I]]
 float32_t test_vmaxnmv_f32(float32x2_t a) {
   return vmaxnmv_f32(a);
 }
 
 // CHECK-LABEL: @test_vminnmvq_f64(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x double> %a to <16 x i8>
 // CHECK:   [[VMINNMVQ_F64_I:%.*]] = call double @llvm.aarch64.neon.fminnmv.f64.v2f64(<2 x double> %a)
 // CHECK:   ret double [[VMINNMVQ_F64_I]]
 float64_t test_vminnmvq_f64(float64x2_t a) {
   return vminnmvq_f64(a);
 }
 
 // CHECK-LABEL: @test_vminnmv_f32(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x float> %a to <8 x i8>
 // CHECK:   [[VMINNMV_F32_I:%.*]] = call float @llvm.aarch64.neon.fminnmv.f32.v2f32(<2 x float> %a)
 // CHECK:   ret float [[VMINNMV_F32_I]]
 float32_t test_vminnmv_f32(float32x2_t a) {
   return vminnmv_f32(a);
 }
 
 // CHECK-LABEL: @test_vpaddq_s64(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i64> %a to <16 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <2 x i64> %b to <16 x i8>
 // CHECK:   [[VPADDQ_V2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.addp.v2i64(<2 x i64> %a, <2 x i64> %b)
 // CHECK:   [[VPADDQ_V3_I:%.*]] = bitcast <2 x i64> [[VPADDQ_V2_I]] to <16 x i8>
 // CHECK:   ret <2 x i64> [[VPADDQ_V2_I]]
 int64x2_t test_vpaddq_s64(int64x2_t a, int64x2_t b) {
   return vpaddq_s64(a, b);
 }
 
 // CHECK-LABEL: @test_vpaddq_u64(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i64> %a to <16 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <2 x i64> %b to <16 x i8>
 // CHECK:   [[VPADDQ_V2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.addp.v2i64(<2 x i64> %a, <2 x i64> %b)
 // CHECK:   [[VPADDQ_V3_I:%.*]] = bitcast <2 x i64> [[VPADDQ_V2_I]] to <16 x i8>
 // CHECK:   ret <2 x i64> [[VPADDQ_V2_I]]
 uint64x2_t test_vpaddq_u64(uint64x2_t a, uint64x2_t b) {
   return vpaddq_u64(a, b);
 }
 
 // CHECK-LABEL: @test_vpaddd_u64(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i64> %a to <16 x i8>
 // CHECK:   [[VPADDD_U64_I:%.*]] = call i64 @llvm.aarch64.neon.uaddv.i64.v2i64(<2 x i64> %a)
 // CHECK:   ret i64 [[VPADDD_U64_I]]
 uint64_t test_vpaddd_u64(uint64x2_t a) {
   return vpaddd_u64(a);
 }
 
 // CHECK-LABEL: @test_vaddvq_s64(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i64> %a to <16 x i8>
 // CHECK:   [[VADDVQ_S64_I:%.*]] = call i64 @llvm.aarch64.neon.saddv.i64.v2i64(<2 x i64> %a)
 // CHECK:   ret i64 [[VADDVQ_S64_I]]
 int64_t test_vaddvq_s64(int64x2_t a) {
   return vaddvq_s64(a);
 }
 
 // CHECK-LABEL: @test_vaddvq_u64(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i64> %a to <16 x i8>
 // CHECK:   [[VADDVQ_U64_I:%.*]] = call i64 @llvm.aarch64.neon.uaddv.i64.v2i64(<2 x i64> %a)
 // CHECK:   ret i64 [[VADDVQ_U64_I]]
 uint64_t test_vaddvq_u64(uint64x2_t a) {
   return vaddvq_u64(a);
 }
 
 // CHECK-LABEL: @test_vadd_f64(
 // CHECK:   [[ADD_I:%.*]] = fadd <1 x double> %a, %b

 // CHECK:   [[VRSQRTS_V2_I:%.*]] = call <1 x double> @llvm.aarch64.neon.frsqrts.v1f64(<1 x double> %a, <1 x double> %b)
 // CHECK:   [[VRSQRTS_V3_I:%.*]] = bitcast <1 x double> [[VRSQRTS_V2_I]] to <8 x i8>
 // CHECK:   ret <1 x double> [[VRSQRTS_V2_I]]
 float64x1_t test_vrsqrts_f64(float64x1_t a, float64x1_t b) {
   return vrsqrts_f64(a, b);
 }
 
 // CHECK-LABEL: @test_vminv_s32(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> %a to <8 x i8>
 // CHECK:   [[VMINV_S32_I:%.*]] = call i32 @llvm.aarch64.neon.sminv.i32.v2i32(<2 x i32> %a)
 // CHECK:   ret i32 [[VMINV_S32_I]]
 int32_t test_vminv_s32(int32x2_t a) {
   return vminv_s32(a);
 }
 
 // CHECK-LABEL: @test_vminv_u32(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> %a to <8 x i8>
 // CHECK:   [[VMINV_U32_I:%.*]] = call i32 @llvm.aarch64.neon.uminv.i32.v2i32(<2 x i32> %a)
 // CHECK:   ret i32 [[VMINV_U32_I]]
 uint32_t test_vminv_u32(uint32x2_t a) {
   return vminv_u32(a);
 }
 
 // CHECK-LABEL: @test_vmaxv_s32(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> %a to <8 x i8>
 // CHECK:   [[VMAXV_S32_I:%.*]] = call i32 @llvm.aarch64.neon.smaxv.i32.v2i32(<2 x i32> %a)
 // CHECK:   ret i32 [[VMAXV_S32_I]]
 int32_t test_vmaxv_s32(int32x2_t a) {
   return vmaxv_s32(a);
 }
 
 // CHECK-LABEL: @test_vmaxv_u32(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> %a to <8 x i8>
 // CHECK:   [[VMAXV_U32_I:%.*]] = call i32 @llvm.aarch64.neon.umaxv.i32.v2i32(<2 x i32> %a)
 // CHECK:   ret i32 [[VMAXV_U32_I]]
 uint32_t test_vmaxv_u32(uint32x2_t a) {
   return vmaxv_u32(a);
 }
 
 // CHECK-LABEL: @test_vaddv_s32(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> %a to <8 x i8>
 // CHECK:   [[VADDV_S32_I:%.*]] = call i32 @llvm.aarch64.neon.saddv.i32.v2i32(<2 x i32> %a)
 // CHECK:   ret i32 [[VADDV_S32_I]]
 int32_t test_vaddv_s32(int32x2_t a) {
   return vaddv_s32(a);
 }
 
 // CHECK-LABEL: @test_vaddv_u32(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> %a to <8 x i8>
 // CHECK:   [[VADDV_U32_I:%.*]] = call i32 @llvm.aarch64.neon.uaddv.i32.v2i32(<2 x i32> %a)
 // CHECK:   ret i32 [[VADDV_U32_I]]
 uint32_t test_vaddv_u32(uint32x2_t a) {
   return vaddv_u32(a);
 }
 
 // CHECK-LABEL: @test_vaddlv_s32(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> %a to <8 x i8>
 // CHECK:   [[VADDLV_S32_I:%.*]] = call i64 @llvm.aarch64.neon.saddlv.i64.v2i32(<2 x i32> %a)
 // CHECK:   ret i64 [[VADDLV_S32_I]]
 int64_t test_vaddlv_s32(int32x2_t a) {
   return vaddlv_s32(a);
 }
 
 // CHECK-LABEL: @test_vaddlv_u32(
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> %a to <8 x i8>
 // CHECK:   [[VADDLV_U32_I:%.*]] = call i64 @llvm.aarch64.neon.uaddlv.i64.v2i32(<2 x i32> %a)
 // CHECK:   ret i64 [[VADDLV_U32_I]]
 uint64_t test_vaddlv_u32(uint32x2_t a) {
   return vaddlv_u32(a);
 }

clang/test/CodeGen/aarch64-neon-scalar-x-indexed-elem.c

 // CHECK:   [[VQDMLXL:%.*]] = call i64 @llvm.aarch64.neon.sqdmulls.scalar(i32 %b, i32 [[LANE]])
 // CHECK:   [[VQDMLXL1:%.*]] = call i64 @llvm.aarch64.neon.sqsub.i64(i64 %a, i64 [[VQDMLXL]])
 // CHECK:   ret i64 [[VQDMLXL1]]
 int64_t test_vqdmlsls_laneq_s32(int64_t a, int32_t b, int32x4_t c) {
   return vqdmlsls_laneq_s32(a, b, c, 3);
 }
 
 // CHECK-LABEL: define <1 x double> @test_vmulx_lane_f64_0() #0 {
-// CHECK:   [[TMP0:%.*]] = bitcast i64 4599917171378402754 to <1 x double>
-// CHECK:   [[TMP1:%.*]] = bitcast i64 4606655882138939123 to <1 x double>
-// CHECK:   [[TMP2:%.*]] = bitcast <1 x double> [[TMP0]] to <8 x i8>
-// CHECK:   [[TMP3:%.*]] = bitcast <8 x i8> [[TMP2]] to <1 x double>
-// CHECK:   [[VGET_LANE:%.*]] = extractelement <1 x double> [[TMP3]], i32 0
-// CHECK:   [[TMP4:%.*]] = bitcast <1 x double> [[TMP1]] to <8 x i8>
-// CHECK:   [[TMP5:%.*]] = bitcast <8 x i8> [[TMP4]] to <1 x double>
-// CHECK:   [[VGET_LANE7:%.*]] = extractelement <1 x double> [[TMP5]], i32 0
-// CHECK:   [[VMULXD_F64_I:%.*]] = call double @llvm.aarch64.neon.fmulx.f64(double [[VGET_LANE]], double [[VGET_LANE7]])
-// CHECK:   [[TMP6:%.*]] = bitcast <1 x double> [[TMP0]] to <8 x i8>
-// CHECK:   [[TMP7:%.*]] = bitcast <8 x i8> [[TMP6]] to <1 x double>
-// CHECK:   [[VSET_LANE:%.*]] = insertelement <1 x double> [[TMP7]], double [[VMULXD_F64_I]], i32 0
+// CHECK: [[TMP0:%.*]] = bitcast <1 x double> <double 0x3FD6304BC43AB5C2> to <8 x i8>
+// CHECK: [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <1 x double>
+// CHECK: [[VGET_LANE:%.*]] = extractelement <1 x double> [[TMP1]], i32 0
+// CHECK: [[TMP2:%.*]] = bitcast <1 x double> <double 0x3FEE211E215AEEF3> to <8 x i8>
+// CHECK: [[TMP3:%.*]] = bitcast <8 x i8> [[TMP2]] to <1 x double>
+// CHECK: [[VGET_LANE7:%.*]] = extractelement <1 x double> [[TMP3]], i32 0
+// CHECK: [[VMULXD:%.*]] = call double @llvm.aarch64.neon.fmulx.f64(double [[VGET_LANE]], double [[VGET_LANE7]]) #4
+// CHECK: [[TMP4:%.*]] = bitcast <1 x double> <double 0x3FD6304BC43AB5C2> to <8 x i8>
+// CHECK: [[TMP5:%.*]] = bitcast <8 x i8> [[TMP4]] to <1 x double>
+// CHECK: [[VSET_LANE:%.*]] = insertelement <1 x double> [[TMP5]], double [[VMULXD]], i32 0
 // CHECK:   ret <1 x double> [[VSET_LANE]]
 float64x1_t test_vmulx_lane_f64_0() {
       float64x1_t arg1;
       float64x1_t arg2;
       float64x1_t result;
       float64_t sarg1, sarg2, sres;
       arg1 = vcreate_f64(UINT64_C(0x3fd6304bc43ab5c2));
       arg2 = vcreate_f64(UINT64_C(0x3fee211e215aeef3));
       result = vmulx_lane_f64(arg1, arg2, 0);
       return result;
 }
 
 // CHECK-LABEL: define <1 x double> @test_vmulx_laneq_f64_2() #1 {
-// CHECK:   [[TMP0:%.*]] = bitcast i64 4599917171378402754 to <1 x double>
-// CHECK:   [[TMP1:%.*]] = bitcast i64 4606655882138939123 to <1 x double>
-// CHECK:   [[SHUFFLE_I:%.*]] = shufflevector <1 x double> [[TMP0]], <1 x double> [[TMP1]], <2 x i32> <i32 0, i32 1>
-// CHECK:   [[TMP2:%.*]] = bitcast <1 x double> [[TMP0]] to <8 x i8>
-// CHECK:   [[TMP3:%.*]] = bitcast <8 x i8> [[TMP2]] to <1 x double>
-// CHECK:   [[VGET_LANE:%.*]] = extractelement <1 x double> [[TMP3]], i32 0
-// CHECK:   [[TMP4:%.*]] = bitcast <2 x double> [[SHUFFLE_I]] to <16 x i8>
-// CHECK:   [[TMP5:%.*]] = bitcast <16 x i8> [[TMP4]] to <2 x double>
-// CHECK:   [[VGETQ_LANE:%.*]] = extractelement <2 x double> [[TMP5]], i32 1
-// CHECK:   [[VMULXD_F64_I:%.*]] = call double @llvm.aarch64.neon.fmulx.f64(double [[VGET_LANE]], double [[VGETQ_LANE]])
-// CHECK:   [[TMP6:%.*]] = bitcast <1 x double> [[TMP0]] to <8 x i8>
-// CHECK:   [[TMP7:%.*]] = bitcast <8 x i8> [[TMP6]] to <1 x double>
-// CHECK:   [[VSET_LANE:%.*]] = insertelement <1 x double> [[TMP7]], double [[VMULXD_F64_I]], i32 0
-// CHECK:   ret <1 x double> [[VSET_LANE]]
+// CHECK: [[SHUFFLE:%.*]] = shufflevector <1 x double> <double 0x3FD6304BC43AB5C2>, <1 x double> <double 0x3FEE211E215AEEF3>, <2 x i32> <i32 0, i32 1>
+// CHECK: [[TMP0:%.*]] = bitcast <1 x double> <double 0x3FD6304BC43AB5C2> to <8 x i8>
+// CHECK: [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <1 x double>
+// CHECK: [[VGET_LANE:%.*]] = extractelement <1 x double> [[TMP1]], i32 0
+// CHECK: [[TMP2:%.*]] = bitcast <2 x double> [[SHUFFLE]] to <16 x i8>
+// CHECK: [[TMP3:%.*]] = bitcast <16 x i8> [[TMP2]] to <2 x double>
+// CHECK: [[VGETQ_LANE:%.*]] = extractelement <2 x double> [[TMP3]], i32 1
+// CHECK: [[VMULXD:%.*]] = call double @llvm.aarch64.neon.fmulx.f64(double [[VGET_LANE]], double [[VGETQ_LANE]]) #4
+// CHECK: [[TMP4:%.*]] = bitcast <1 x double> <double 0x3FD6304BC43AB5C2> to <8 x i8>
+// CHECK: [[TMP5:%.*]] = bitcast <8 x i8> [[TMP4]] to <1 x double>
+// CHECK: [[VSET_LANE:%.*]] = insertelement <1 x double> [[TMP5]], double [[VMULXD]], i32 0
+// CHECK: ret <1 x double> [[VSET_LANE]]
 float64x1_t test_vmulx_laneq_f64_2() {
       float64x1_t arg1;
       float64x1_t arg2;
       float64x2_t arg3;
       float64x1_t result;
       float64_t sarg1, sarg2, sres;
       arg1 = vcreate_f64(UINT64_C(0x3fd6304bc43ab5c2));
       arg2 = vcreate_f64(UINT64_C(0x3fee211e215aeef3));
       arg3 = vcombine_f64(arg1, arg2);
       result = vmulx_laneq_f64(arg1, arg3, 1);
       return result;
 }
 
 // CHECK: attributes #0 ={{.*}}"min-legal-vector-width"="64"
 // CHECK: attributes #1 ={{.*}}"min-legal-vector-width"="128"

clang/utils/TableGen/NeonEmitter.cpp

private:
   unsigned Bitwidth, ElementBitwidth, NumVectors;
 
 public:
   Type()
       : Kind(Void), Immediate(false), Constant(false),
         Pointer(false), ScalarForMangling(false), NoManglingQ(false),
         Bitwidth(0), ElementBitwidth(0), NumVectors(0) {}
 
-  Type(TypeSpec TS, char CharMod)
+  Type(TypeSpec TS, StringRef CharMods)
       : TS(std::move(TS)), Kind(Void), Immediate(false),
         Constant(false), Pointer(false), ScalarForMangling(false),
         NoManglingQ(false), Bitwidth(0), ElementBitwidth(0), NumVectors(0) {
-    applyModifier(CharMod);
+    applyModifiers(CharMods);
   }
 
   /// Returns a type representing "void".
   static Type getVoid() { return Type(); }
 
   bool operator==(const Type &Other) const { return str() == Other.str(); }
   bool operator!=(const Type &Other) const { return !operator==(Other); }
 
   //
   // Query functions
   //
   bool isScalarForMangling() const { return ScalarForMangling; }
   bool noManglingQ() const { return NoManglingQ; }
 
   bool isPointer() const { return Pointer; }
+  bool isValue() const { return !isVoid() && !isPointer(); }
+  bool isScalar() const { return isValue() && NumVectors == 0; }
+  bool isVector() const { return isValue() && NumVectors > 0; }
+  bool isConstPointer() const { return Constant; }
   bool isFloating() const { return Kind == Float; }
   bool isInteger() const { return Kind == SInt || Kind == UInt; }
   bool isPoly() const { return Kind == Poly; }
   bool isSigned() const { return Kind == SInt; }
   bool isImmediate() const { return Immediate; }
-  bool isScalar() const { return NumVectors == 0; }
-  bool isVector() const { return NumVectors > 0; }
   bool isFloat() const { return isFloating() && ElementBitwidth == 32; }
   bool isDouble() const { return isFloating() && ElementBitwidth == 64; }
   bool isHalf() const { return isFloating() && ElementBitwidth == 16; }
   bool isChar() const { return ElementBitwidth == 8; }
   bool isShort() const { return isInteger() && ElementBitwidth == 16; }
   bool isInt() const { return isInteger() && ElementBitwidth == 32; }
   bool isLong() const { return isInteger() && ElementBitwidth == 64; }
   bool isVoid() const { return Kind == Void; }
   unsigned getNumElements() const { return Bitwidth / ElementBitwidth; }
   unsigned getSizeInBits() const { return Bitwidth; }
   unsigned getElementSizeInBits() const { return ElementBitwidth; }
   unsigned getNumVectors() const { return NumVectors; }
 
   //
   // Mutator functions
   //
   void makeUnsigned() {
-    assert(isInteger() && "not a potentially signed type");
+    assert(!isVoid() && "not a potentially signed type");
     Kind = UInt;
   }
   void makeSigned() {
-    assert(isInteger() && "not a potentially signed type");
+    assert(!isVoid() && "not a potentially signed type");
     Kind = SInt;
   }
 
   void makeInteger(unsigned ElemWidth, bool Sign) {
     assert(!isVoid() && "converting void to int probably not useful");
     Kind = Sign ? SInt : UInt;
     Immediate = false;
     ElementBitwidth = ElemWidth;
public:
   static Type fromTypedefName(StringRef Name);
 
 private:
   /// Creates the type based on the typespec string in TS.
   /// Sets "Quad" to true if the "Q" or "H" modifiers were
   /// seen. This is needed by applyModifier as some modifiers
   /// only take effect if the type size was changed by "Q" or "H".
   void applyTypespec(bool &Quad);
-  /// Applies a prototype modifiers to the type.
-  void applyModifier(char Mod);
+  /// Applies prototype modifiers to the type.
+  void applyModifiers(StringRef Mods);
 };
 
 //===----------------------------------------------------------------------===//
 // Variable
 //===----------------------------------------------------------------------===//
 
 /// A variable is a simple class that just has a type and a name.
 class Variable {

 
 /// The main grunt class. This represents an instantiation of an intrinsic with
 /// a particular typespec and prototype.
 class Intrinsic {
   friend class DagEmitter;
 
   /// The Record this intrinsic was created from.
   Record *R;
-  /// The unmangled name and prototype.
-  std::string Name, Proto;
+  /// The unmangled name.
+  std::string Name;
   /// The input and output typespecs. InTS == OutTS except when
   /// CartesianProductOfTypes is 1 - this is the case for vreinterpret.
   TypeSpec OutTS, InTS;
   /// The base class kind. Most intrinsics use ClassS, which has full type
   /// info for integers (s32/u32). Some use ClassI, which doesn't care about
   /// signedness (i32), while some (ClassB) have no type at all, only a width
   /// (32).
   ClassKind CK;
   /// The list of DAGs for the body. May be empty, in which case we should
   /// emit a builtin call.
   ListInit *Body;
   /// The architectural #ifdef guard.
   std::string Guard;
   /// Set if the Unavailable bit is 1. This means we don't generate a body,
   /// just an "unavailable" attribute on a declaration.
   bool IsUnavailable;
   /// Is this intrinsic safe for big-endian? or does it need its arguments
   /// reversing?
   bool BigEndianSafe;
 
   /// The types of return value [0] and parameters [1..].
   std::vector<Type> Types;
+  /// The index of the key type passed to CGBuiltin.cpp for polymorphic calls.
+  int PolymorphicKeyType;
   /// The local variables defined.
   std::map<std::string, Variable> Variables;
   /// NeededEarly - set if any other intrinsic depends on this intrinsic.
   bool NeededEarly;
   /// UseMacro - set if we should implement using a macro or unset for a
   ///            function.
   bool UseMacro;
   /// The set of intrinsics that this intrinsic uses/requires.
bool isBigEndianSafe() const {
     }
     return true;
   }
 
 public:
   Intrinsic(Record *R, StringRef Name, StringRef Proto, TypeSpec OutTS,
             TypeSpec InTS, ClassKind CK, ListInit *Body, NeonEmitter &Emitter,
             StringRef Guard, bool IsUnavailable, bool BigEndianSafe)
-      : R(R), Name(Name.str()), Proto(Proto.str()), OutTS(OutTS), InTS(InTS),
-        CK(CK), Body(Body), Guard(Guard.str()), IsUnavailable(IsUnavailable),
-        BigEndianSafe(BigEndianSafe), NeededEarly(false), UseMacro(false),
-        BaseType(OutTS, 'd'), InBaseType(InTS, 'd'), Emitter(Emitter) {
+      : R(R), Name(Name.str()), OutTS(OutTS), InTS(InTS), CK(CK), Body(Body),
+        Guard(Guard.str()), IsUnavailable(IsUnavailable),
+        BigEndianSafe(BigEndianSafe), PolymorphicKeyType(0), NeededEarly(false),
+        UseMacro(false), BaseType(OutTS, "."), InBaseType(InTS, "."),
+        Emitter(Emitter) {
+    // Modify the TypeSpec per-argument to get a concrete Type, and create
+    // known variables for each.
+    // Types[0] is the return value.
+    unsigned Pos = 0;
+    Types.emplace_back(OutTS, getNextModifiers(Proto, Pos));
+    StringRef Mods = getNextModifiers(Proto, Pos);
+    while (!Mods.empty()) {
+      Types.emplace_back(InTS, Mods);
+      if (Mods.find("!") != StringRef::npos)
+        PolymorphicKeyType = Types.size() - 1;
+
+      Mods = getNextModifiers(Proto, Pos);
+    }
+
+    for (auto Type : Types) {
       // If this builtin takes an immediate argument, we need to #define it rather
       // than use a standard declaration, so that SemaChecking can range check
       // the immediate passed by the user.
-    if (Proto.find('i') != std::string::npos)
-      UseMacro = true;
 
       // Pointer arguments need to use macros to avoid hiding aligned attributes
       // from the pointer type.
-    if (Proto.find('p') != std::string::npos ||
-        Proto.find('c') != std::string::npos)
-      UseMacro = true;
 
       // It is not permitted to pass or return an __fp16 by value, so intrinsics
       // taking a scalar float16_t must be implemented as macros.
-    if (OutTS.find('h') != std::string::npos &&
-        Proto.find('s') != std::string::npos)
+      if (Type.isImmediate() || Type.isPointer() ||
+          (Type.isScalar() && Type.isHalf()))
         UseMacro = true;
-
-    // Modify the TypeSpec per-argument to get a concrete Type, and create
-    // known variables for each.
-    // Types[0] is the return value.
-    Types.emplace_back(OutTS, Proto[0]);
-    for (unsigned I = 1; I < Proto.size(); ++I)
-      Types.emplace_back(InTS, Proto[I]);
+    }
   }
 
   /// Get the Record that this intrinsic is based off.
   Record *getRecord() const { return R; }
   /// Get the set of Intrinsics that this intrinsic calls.
   /// this is the set of immediate dependencies, NOT the
   /// transitive closure.
   const std::set<Intrinsic *> &getDependencies() const { return Dependencies; }
   /// Get the architectural guard string (#ifdef).
   std::string getGuard() const { return Guard; }
   /// Get the non-mangled name.
   std::string getName() const { return Name; }
 
   /// Return true if the intrinsic takes an immediate operand.
   bool hasImmediate() const {
-    return Proto.find('i') != std::string::npos;
+    return std::any_of(Types.begin(), Types.end(),
+                       [](const Type &T) { return T.isImmediate(); });
   }
 
   /// Return the parameter index of the immediate operand.
   unsigned getImmediateIdx() const {
-    assert(hasImmediate());
-    unsigned Idx = Proto.find('i');
-    assert(Idx > 0 && "Can't return an immediate!");
+    for (unsigned Idx = 0; Idx < Types.size(); ++Idx)
+      if (Types[Idx].isImmediate())
         return Idx - 1;
+    llvm_unreachable("Intrinsic has no immediate");
   }
 
-  unsigned getNumParams() const { return Proto.size() - 1; }
+
+  unsigned getNumParams() const { return Types.size() - 1; }
   Type getReturnType() const { return Types[0]; }
   Type getParamType(unsigned I) const { return Types[I + 1]; }
   Type getBaseType() const { return BaseType; }
-  /// Return the raw prototype string.
-  std::string getProto() const { return Proto; }
+  Type getPolymorphicKeyType() const { return Types[PolymorphicKeyType]; }
 
   /// Return true if the prototype has a scalar argument.
   bool protoHasScalar() const;
 
   /// Return the index that parameter PIndex will sit at
   /// in a generated function call. This is often just PIndex,
   /// but may not be as things such as multiple-vector operands
   /// and sret parameters need to be taken into accont.
public:
 
   /// Generate the intrinsic, returning code.
   std::string generate();
   /// Perform type checking and populate the dependency graph, but
   /// don't generate code yet.
   void indexBody();
 
 private:
+  StringRef getNextModifiers(StringRef Proto, unsigned &Pos) const;
+
   std::string mangleName(std::string Name, ClassKind CK) const;
 
   void initVariables();
   std::string replaceParamsIn(std::string S);
 
   void emitBodyAsBuiltinCall();
 
   void generateImpl(bool ReverseArguments,
std::string Type::str() const {
   return S;
 }
 
 std::string Type::builtin_str() const {
   std::string S;
   if (isVoid())
     return "v";
 
-  if (Pointer)
+  if (isPointer()) {
     // All pointers are void pointers.
-    S += "v";
-  else if (isInteger())
+    S = "v";
+    if (isConstPointer())
+      S += "C";
+    S += "*";
+    return S;
+  } else if (isInteger())
     switch (ElementBitwidth) {
     case 8: S += "c"; break;
     case 16: S += "s"; break;
     case 32: S += "i"; break;
     case 64: S += "Wi"; break;
     case 128: S += "LLLi"; break;
     default: llvm_unreachable("Unhandled case!");
     }
   else
     switch (ElementBitwidth) {
     case 16: S += "h"; break;
     case 32: S += "f"; break;
     case 64: S += "d"; break;
     default: llvm_unreachable("Unhandled case!");
     }
 
-  if (isChar() && !Pointer)
+  if (isChar())
     // Make chars explicitly signed.
     S = "S" + S;
-  else if (!isPointer() && isInteger() && !isSigned())
+  else if (isInteger() && !isSigned())
     S = "U" + S;
 
   // Constant indices are "int", but have the "constant expression" modifier.
   if (isImmediate()) {
     assert(isInteger() && isSigned());
     S = "I" + S;
   }
 
-  if (isScalar()) {
-    if (Constant) S += "C";
-    if (Pointer) S += "*";
+  if (isScalar())
     return S;
-  }
 
   std::string Ret;
   for (unsigned I = 0; I < NumVectors; ++I)
     Ret += "V" + utostr(getNumElements()) + S;
 
   return Ret;
 }
 
default:
       llvm_unreachable("Unhandled type code!");
     }
   }
   assert(ElementBitwidth != ~0U && "Bad element bitwidth!");
 
   Bitwidth = Quad ? 128 : 64;
 }
 
-void Type::applyModifier(char Mod) {
+void Type::applyModifiers(StringRef Mods) {
   bool AppliedQuad = false;
   applyTypespec(AppliedQuad);
 
+  for (char Mod : Mods) {
     switch (Mod) {
+    case '.':
+      break;
     case 'v':
       Kind = Void;
       break;
-  case 't':
-    if (isPoly())
-      Kind = UInt;
-    break;
-  case 'b':
-    Kind = UInt;
-    NumVectors = 0;
-    Bitwidth = ElementBitwidth;
-    break;
-  case '$':
-    Kind = SInt;
-    NumVectors = 0;
-    Bitwidth = ElementBitwidth;
-    break;
-  case 'u':
-    Kind = UInt;
-    break;
-  case 'x':
-    assert(!isPoly() && "'u' can't be used with poly types!");
-    Kind = SInt;
-    break;
-  case 'o':
-    Bitwidth = ElementBitwidth = 64;
-    NumVectors = 0;
-    Kind = Float;
-    break;
-  case 'y':
-    Bitwidth = ElementBitwidth = 32;
-    NumVectors = 0;
-    Kind = Float;
-    break;
-  case 'Y':
-    Bitwidth = ElementBitwidth = 16;
-    NumVectors = 0;
-    Kind = Float;
-    break;
-  case 'I':
-    Bitwidth = ElementBitwidth = 32;
-    NumVectors = 0;
-    Kind = SInt;
-    break;
-  case 'L':
-    Bitwidth = ElementBitwidth = 64;
-    NumVectors = 0;
+    case 'S':
       Kind = SInt;
       break;
     case 'U':
-    Bitwidth = ElementBitwidth = 32;
-    NumVectors = 0;
       Kind = UInt;
       break;
-  case 'O':
-    Bitwidth = ElementBitwidth = 64;
-    NumVectors = 0;
-    Kind = UInt;
-    break;
-  case 'f':
-    Kind = Float;
-    ElementBitwidth = 32;
-    break;
     case 'F':
       Kind = Float;
-    ElementBitwidth = 64;
-    break;
-  case 'H':
-    Kind = Float;
-    ElementBitwidth = 16;
-    break;
-  case '0':
-    Kind = Float;
-    if (AppliedQuad)
-      Bitwidth /= 2;
-    ElementBitwidth = 16;
       break;
-  case '1':
-    Kind = Float;
-    if (!AppliedQuad)
-      Bitwidth *= 2;
-    ElementBitwidth = 16;
-    break;
-  case 'g':
-    if (AppliedQuad)
-      Bitwidth /= 2;
-    break;
-  case 'j':
-    if (!AppliedQuad)
-      Bitwidth *= 2;
-    break;
-  case 'w':
-    ElementBitwidth *= 2;
-    Bitwidth *= 2;
+    case 'P':
+      Kind = Poly;
       break;
-  case 'n':
+    case '>':
+      assert(ElementBitwidth < 128);
       ElementBitwidth *= 2;
       break;
-  case 'i':
-    Kind = SInt;
-    ElementBitwidth = Bitwidth = 32;
-    NumVectors = 0;
-    Immediate = true;
-    break;
-  case 'l':
-    Kind = UInt;
-    ElementBitwidth = Bitwidth = 64;
-    NumVectors = 0;
-    Immediate = true;
-    break;
-  case 'z':
+    case '<':
+      assert(ElementBitwidth > 8);
       ElementBitwidth /= 2;
-    Bitwidth = ElementBitwidth;
-    NumVectors = 0;
-    break;
-  case 'r':
-    ElementBitwidth *= 2;
-    Bitwidth = ElementBitwidth;
-    NumVectors = 0;
-    break;
-  case 's':
-    Bitwidth = ElementBitwidth;
-    NumVectors = 0;
-    break;
-  case 'k':
-    Bitwidth *= 2;
       break;
-  case 'c':
-    Constant = true;
-    LLVM_FALLTHROUGH;
-  case 'p':
-    Pointer = true;
-    Bitwidth = ElementBitwidth;
+    case '1':
       NumVectors = 0;
       break;
-  case 'h':
-    ElementBitwidth /= 2;
-    break;
-  case 'q':
-    ElementBitwidth /= 2;
-    Bitwidth *= 2;
-    break;
-  case 'e':
-    ElementBitwidth /= 2;
-    Kind = UInt;
-    break;
-  case 'm':
-    ElementBitwidth /= 2;
-    Bitwidth /= 2;
-    break;
-  case 'd':
-    break;
     case '2':
       NumVectors = 2;
       break;
     case '3':
       NumVectors = 3;
       break;
     case '4':
       NumVectors = 4;
       break;
-  case 'B':
-    NumVectors = 2;
-    if (!AppliedQuad)
-      Bitwidth *= 2;
+    case '*':
+      Pointer = true;
       break;
-  case 'C':
-    NumVectors = 3;
-    if (!AppliedQuad)
-      Bitwidth *= 2;
+    case 'c':
+      Constant = true;
       break;
-  case 'D':
-    NumVectors = 4;
-    if (!AppliedQuad)
-      Bitwidth *= 2;
+    case 'Q':
+      Bitwidth = 128;
       break;
-  case '7':
-    if (AppliedQuad)
-      Bitwidth /= 2;
-    ElementBitwidth = 8;
+    case 'q':
+      Bitwidth = 64;
       break;
-  case '8':
-    ElementBitwidth = 8;
+    case 'I':
+      Kind = SInt;
+      ElementBitwidth = Bitwidth = 32;
+      NumVectors = 0;
+      Immediate = true;
       break;
-  case '9':
-    if (!AppliedQuad)
-      Bitwidth *= 2;
-    ElementBitwidth = 8;
+    case 'p':
+      if (isPoly())
+        Kind = UInt;
+      break;
+    case '!':
+      // Key type, handled elsewhere.
       break;
     default:
       llvm_unreachable("Unhandled character!");
     }
   }
+}
 
 //===----------------------------------------------------------------------===//
 // Intrinsic implementation
 //===----------------------------------------------------------------------===//
 
+StringRef Intrinsic::getNextModifiers(StringRef Proto, unsigned &Pos) const {
+  if (Proto.size() == Pos)
+    return StringRef();
+  else if (Proto[Pos] != '(')
+    return Proto.substr(Pos++, 1);
+
+  size_t Start = Pos + 1;
+  size_t End = Proto.find(')', Start);
+  assert_with_loc(End != StringRef::npos, "unmatched modifier group paren");
+  Pos = End + 1;
+  return Proto.slice(Start, End);
+}
+
 std::string Intrinsic::getInstTypeCode(Type T, ClassKind CK) const {
   char typeCode = '\0';
   bool printNumber = true;
 
   if (CK == ClassB)
     return "";
 
   if (T.isPoly())
std::string Intrinsic::getInstTypeCode(Type T, ClassKind CK) const {
   if (typeCode != '\0')
     S.push_back(typeCode);
   if (printNumber)
     S += utostr(T.getElementSizeInBits());
 
   return S;
 }
 
-static bool isFloatingPointProtoModifier(char Mod) {
-  return Mod == 'F' || Mod == 'f' || Mod == 'H' || Mod == 'Y' || Mod == 'I';
-}
-
 std::string Intrinsic::getBuiltinTypeStr() {
   ClassKind LocalCK = getClassKind(true);
   std::string S;
 
   Type RetT = getReturnType();
   if ((LocalCK == ClassI || LocalCK == ClassW) && RetT.isScalar() &&
-      !RetT.isFloating() && !RetT.isVoid())
+      !RetT.isFloating())
     RetT.makeInteger(RetT.getElementSizeInBits(), false);
 
   // Since the return value must be one type, return a vector type of the
   // appropriate width which we will bitcast.  An exception is made for
   // returning structs of 2, 3, or 4 vectors which are returned in a sret-like
   // fashion, storing them to a pointer arg.
   if (RetT.getNumVectors() > 1) {
     S += "vv*"; // void result with void* first argument
   } else {
     if (RetT.isPoly())
       RetT.makeInteger(RetT.getElementSizeInBits(), false);
     if (!RetT.isScalar() && RetT.isInteger() && !RetT.isSigned())
       RetT.makeSigned();
 
-    bool ForcedVectorFloatingType = isFloatingPointProtoModifier(Proto[0]);
-    if (LocalCK == ClassB && !RetT.isVoid() && !RetT.isScalar() &&
-        !ForcedVectorFloatingType)
+    if (LocalCK == ClassB && RetT.isValue() && !RetT.isScalar())
       // Cast to vector of 8-bit elements.
       RetT.makeInteger(8, true);
 
     S += RetT.builtin_str();
   }
 
   for (unsigned I = 0; I < getNumParams(); ++I) {
     Type T = getParamType(I);
     if (T.isPoly())
       T.makeInteger(T.getElementSizeInBits(), false);
 
-    bool ForcedFloatingType = isFloatingPointProtoModifier(Proto[I + 1]);
-    if (LocalCK == ClassB && !T.isScalar() && !ForcedFloatingType)
+    if (LocalCK == ClassB && !T.isScalar())
       T.makeInteger(8, true);
     // Halves always get converted to 8-bit elements.
     if (T.isHalf() && T.isVector() && !T.isScalarForMangling())
       T.makeInteger(8, true);
 
     if (LocalCK == ClassI && T.isInteger())
       T.makeSigned();
 
std::string Intrinsic::replaceParamsIn(std::string S) {
   return S;
 }
 
 void Intrinsic::initVariables() {
   Variables.clear();
 
   // Modify the TypeSpec per-argument to get a concrete Type, and create
   // known variables for each.
-  for (unsigned I = 1; I < Proto.size(); ++I) {
+  for (unsigned I = 1; I < Types.size(); ++I) {
     char NameC = '0' + (I - 1);
     std::string Name = "p";
     Name.push_back(NameC);
 
     Variables[Name] = Variable(Types[I], Name + VariablePostfix);
   }
   RetVar = Variable(Types[0], "ret" + VariablePostfix);
 }
void Intrinsic::emitShadowedArgs() {
   // Macro arguments are not type-checked like inline function arguments,
   // so assign them to local temporaries to get the right type checking.
   if (!UseMacro)
     return;
 
   for (unsigned I = 0; I < getNumParams(); ++I) {
     // Do not create a temporary for an immediate argument.
     // That would defeat the whole point of using a macro!
-    if (hasImmediate() && Proto[I+1] == 'i')
+    if (getParamType(I).isImmediate())
       continue;
     // Do not create a temporary for pointer arguments. The input
     // pointer may have an alignment hint.
     if (getParamType(I).isPointer())
       continue;
 
     std::string Name = "p" + utostr(I);
 
     assert(Variables.find(Name) != Variables.end());
     Variable &V = Variables[Name];
 
     std::string NewName = "s" + utostr(I);
     Variable V2(V.getType(), NewName + VariablePostfix);
 
     OS << "  " << V2.getType().str() << " " << V2.getName() << " = "
        << V.getName() << ";";
     emitNewLine();
 
     V = V2;
   }
 }
 
 bool Intrinsic::protoHasScalar() const {
-  return (Proto.find('s') != std::string::npos ||
-          Proto.find('z') != std::string::npos ||
-          Proto.find('r') != std::string::npos ||
-          Proto.find('b') != std::string::npos ||
-          Proto.find('$') != std::string::npos ||
-          Proto.find('y') != std::string::npos ||
-          Proto.find('o') != std::string::npos);
+  return std::any_of(Types.begin(), Types.end(), [](const Type &T) {
+    return T.isScalar() && !T.isImmediate();
+  });
 }
 
 void Intrinsic::emitBodyAsBuiltinCall() {
   std::string S;
 
   // If this builtin returns a struct 2, 3, or 4 vectors, pass it as an implicit
   // sret-like argument.
   bool SRet = getReturnType().getNumVectors() >= 2;
if (CastToType.isVector()) {
       Arg = "(" + CastToType.str() + ")" + Arg;
     }
 
     S += Arg + ", ";
   }
 
   // Extra constant integer to hold type class enum for this function, e.g. s8
   if (getClassKind(true) == ClassB) {
-    Type ThisTy = getReturnType();
-    if (Proto[0] == 'v' || isFloatingPointProtoModifier(Proto[0]))
-      ThisTy = getParamType(0);
-    if (ThisTy.isPointer())
-      ThisTy = getParamType(1);
-
-    S += utostr(ThisTy.getNeonEnum());
+    S += utostr(getPolymorphicKeyType().getNeonEnum());
   } else {
     // Remove extraneous ", ".
     S.pop_back();
     S.pop_back();
   }
   S += ");";
 
   std::string RetExpr;
void NeonEmitter::createIntrinsic(Record *R,
 
   ClassKind CK = ClassNone;
   if (R->getSuperClasses().size() >= 2)
     CK = ClassMap[R->getSuperClasses()[1].first];
 
   std::vector<std::pair<TypeSpec, TypeSpec>> NewTypeSpecs;
   for (auto TS : TypeSpecs) {
     if (CartesianProductOfTypes) {
-      Type DefaultT(TS, 'd');
+      Type DefaultT(TS, ".");
       for (auto SrcTS : TypeSpecs) {
-        Type DefaultSrcT(SrcTS, 'd');
+        Type DefaultSrcT(SrcTS, ".");
         if (TS == SrcTS ||
             DefaultSrcT.getSizeInBits() != DefaultT.getSizeInBits())
           continue;
         NewTypeSpecs.push_back(std::make_pair(TS, SrcTS));
       }
     } else {
       NewTypeSpecs.push_back(std::make_pair(TS, TS));
     }
for (auto *Def : Defs) {
     if (Def->hasBody())
       continue;
     // Functions which have a scalar argument cannot be overloaded, no need to
     // check them if we are emitting the type checking code.
     if (Def->protoHasScalar())
       continue;
 
     uint64_t Mask = 0ULL;
-    Type Ty = Def->getReturnType();
-    if (Def->getProto()[0] == 'v' ||
-        isFloatingPointProtoModifier(Def->getProto()[0]))
-      Ty = Def->getParamType(0);
-    if (Ty.isPointer())
-      Ty = Def->getParamType(1);
-
-    Mask |= 1ULL << Ty.getNeonEnum();
+    Mask |= 1ULL << Def->getPolymorphicKeyType().getNeonEnum();
 
     // Check if the function has a pointer or const pointer argument.
-    std::string Proto = Def->getProto();
     int PtrArgNum = -1;
     bool HasConstPtr = false;
     for (unsigned I = 0; I < Def->getNumParams(); ++I) {
-      char ArgType = Proto[I + 1];
-      if (ArgType == 'c') {
-        HasConstPtr = true;
+      const auto &Type = Def->getParamType(I);
+      if (Type.isPointer()) {
         PtrArgNum = I;
-        break;
-      }
-      if (ArgType == 'p') {
-        PtrArgNum = I;
-        break;
+        HasConstPtr = Type.isConstPointer();
       }
     }
+
     // For sret builtins, adjust the pointer argument index.
     if (PtrArgNum >= 0 && Def->getReturnType().getNumVectors() > 1)
       PtrArgNum += 1;
 
     std::string Name = Def->getName();
     // Omit type checking for the pointer arguments of vld1_lane, vld1_dup,
     // and vst1_lane intrinsics.  Using a pointer to the vector element
     // type with one of those operations causes codegen to select an aligned
void NeonEmitter::run(raw_ostream &OS) {
   std::string TypedefTypes(
       "cQcsQsiQilQlUcQUcUsQUsUiQUiUlQUlhQhfQfdQdPcQPcPsQPsPlQPl");
   std::vector<TypeSpec> TDTypeVec = TypeSpec::fromTypeSpecs(TypedefTypes);
 
   // Emit vector typedefs.
   bool InIfdef = false;
   for (auto &TS : TDTypeVec) {
     bool IsA64 = false;
-    Type T(TS, 'd');
+    Type T(TS, ".");
     if (T.isDouble() || (T.isPoly() && T.getElementSizeInBits() == 64))
       IsA64 = true;
 
     if (InIfdef && !IsA64) {
       OS << "#endif\n";
       InIfdef = false;
     }
     if (!InIfdef && IsA64) {
if (InIfdef)
     OS << "#endif\n";
   OS << "\n";
 
   // Emit struct typedefs.
   InIfdef = false;
   for (unsigned NumMembers = 2; NumMembers <= 4; ++NumMembers) {
     for (auto &TS : TDTypeVec) {
       bool IsA64 = false;
-      Type T(TS, 'd');
+      Type T(TS, ".");
       if (T.isDouble() || (T.isPoly() && T.getElementSizeInBits() == 64))
         IsA64 = true;
 
       if (InIfdef && !IsA64) {
         OS << "#endif\n";
         InIfdef = false;
       }
       if (!InIfdef && IsA64) {
         OS << "#ifdef __aarch64__\n";
         InIfdef = true;
       }
 
-      char M = '2' + (NumMembers - 2);
-      Type VT(TS, M);
+      const char Mods[] = { static_cast<char>('2' + (NumMembers - 2)), 0};
+      Type VT(TS, Mods);
       OS << "typedef struct " << VT.str() << " {\n";
       OS << "  " << T.str() << " val";
       OS << "[" << NumMembers << "]";
       OS << ";\n} ";
       OS << VT.str() << ";\n";
       OS << "\n";
     }
   }

clang/utils/convert_arm_neon.py

+#!/usr/bin/env python3
+
+# This script is pipe based and converts an arm_neon.td (or arm_fp16.td) file
+# using the old single-char type modifiers to an equivalent new-style form where
+# each modifier is orthogonal and they can be composed.
+#
+# It was used to directly generate the .td files on master, so if you have any
+# local additions I would suggest implementing any modifiers here, and running
+# it over your entire pre-merge .td files rather than trying to resolve any
+# conflicts manually.
+import re, sys
+MOD_MAP = {
+    'v': 'v',
+    'x': 'S',
+    'u': 'U',
+    'd': '.',
+    'g': 'q',
+    'j': 'Q',
+    'w': '>Q',
+    'n': '>',
+    'h': '<',
+    'q': '<Q',
+    'e': '<U',
+    'm': '<q',
+    'i': 'I',
+    'l': 'IU>',
+    's': '1',
+    'z': '1<',
+    'r': '1>',
+    'b': '1U',
+    '$': '1S',
+    'k': 'Q',
+    '2': '2',
+    '3': '3',
+    '4': '4',
+    'B': '2Q',
+    'C': '3Q',
+    'D': '4Q',
+    'p': '*',
+    'c': 'c*',
+    '7': '<<q',
+    '8': '<<',
+    '9': '<<Q',
+    't': 'p'
+    }
+
+
+def typespec_elt_size(typespec):
+    if 'c' in typespec:
+        return 8
+    elif 's' in typespec or 'h' in typespec:
+        return 16
+    elif 'i' in typespec or 'f' in typespec:
+        return 32
+    elif 'l' in typespec or 'd' in typespec:
+        return 64
+    elif 'k' in typespec:
+        return 128
+
+def get_resize(cur, desired):
+    res = ''
+    while cur < desired:
+        res += '>'
+        cur *= 2
+    while cur > desired:
+        res += '<'
+        cur /= 2
+    return res
+
+
+def remap_protocol(proto, typespec, name):
+    key_type = 0
+
+    # Conversions like to see the integer type so they know signedness.
+    if 'vcvt' in name and '_f' in name and name != 'vcvt_f32_f64' and name != 'vcvt_f64_f32':
+        key_type = 1
+    default_width = typespec_elt_size(typespec)
+    inconsistent_width = False
+    for elt in typespec:
+        new_width = typespec_elt_size(elt)
+        if new_width and new_width != default_width:
+            inconsistent_width = True
+
+    res = ''
+    for i, c in enumerate(proto):
+        # void and pointers make for bad discriminators in CGBuiltin.cpp.
+        if c in 'vcp':
+                key_type += 1
+
+        if c in MOD_MAP:
+            cur_mod = MOD_MAP[c]
+        elif inconsistent_width:
+            # Otherwise it's a fixed output width modifier.
+            sys.stderr.write(f'warning: {name} uses fixed output size but has inconsistent input widths: {proto} {typespec}\n')
+
+        if c == 'Y':
+            # y: scalar of half float
+            resize = get_resize(default_width, 16)
+            cur_mod = f'1F{resize}'
+        elif c == 'y':
+            # y: scalar of float
+            resize = get_resize(default_width, 32)
+            cur_mod = f'1F{resize}'
+        elif c == 'o':
+            # o: scalar of double
+            resize = get_resize(default_width, 64)
+            cur_mod = f'1F{resize}'
+        elif c == 'I':
+            # I: scalar of 32-bit signed
+            resize = get_resize(default_width, 32)
+            cur_mod = f'1S{resize}'
+        elif c == 'L':
+            # L: scalar of 64-bit signed
+            resize = get_resize(default_width, 64)
+            cur_mod = f'1S{resize}'
+        elif c == 'U':
+            # I: scalar of 32-bit unsigned
+            resize = get_resize(default_width, 32)
+            cur_mod = f'1U{resize}'
+        elif c == 'O':
+            # O: scalar of 64-bit unsigned
+            resize = get_resize(default_width, 64)
+            cur_mod = f'1U{resize}'
+        elif c == 'f':
+            # f: float (int args)
+            resize = get_resize(default_width, 32)
+            cur_mod = f'F{resize}'
+        elif c == 'F':
+            # F: double (int args)
+            resize = get_resize(default_width, 64)
+            cur_mod = f'F{resize}'
+        elif c == 'H':
+            # H: half (int args)
+            resize = get_resize(default_width, 16)
+            cur_mod = f'F{resize}'
+        elif c == '0':
+            # 0: half (int args), ignore 'Q' size modifier.
+            resize = get_resize(default_width, 16)
+            cur_mod = f'Fq{resize}'
+        elif c == '1':
+            # 1: half (int args), force 'Q' size modifier.
+            resize = get_resize(default_width, 16)
+            cur_mod = f'FQ{resize}'
+
+        if len(cur_mod) == 0:
+            raise Exception(f'WTF: {c} in {name}')
+
+        if key_type != 0 and key_type == i:
+            cur_mod += '!'
+
+        if len(cur_mod) == 1:
+            res += cur_mod
+        else:
+            res += '(' + cur_mod + ')'
+
+    return res
+
+def replace_insts(m):
+    start, end = m.span('proto')
+    start -= m.start()
+    end -= m.start()
+    new_proto = remap_protocol(m['proto'], m['kinds'], m['name'])
+    return m.group()[:start] + new_proto + m.group()[end:]
+
+INST = re.compile(r'Inst<"(?P<name>.*?)",\s*"(?P<proto>.*?)",\s*"(?P<kinds>.*?)"')
+
+new_td = INST.sub(replace_insts, sys.stdin.read())
+sys.stdout.write(new_td)