[PATCH][AArch64] implement aarch64 neon instruction class AdvSIMD (shift)

[Jiangning Liu]

Hao,

The followings are missed in AArch64TargetLowering::getTargetNodeName,

  case AArch64ISD::NEON_DUPIMM:
    return "AArch64ISD::NEON_DUPIMM";
  case AArch64ISD::NEON_SLI:
    return "AArch64ISD::NEON_SLI";
  case AArch64ISD::NEON_SRI:
    return "AArch64ISD::NEON_SRI";
  case AArch64ISD::NEON_RSRAs:
    return "AArch64ISD::NEON_RSRAs";
  case AArch64ISD::NEON_RSRAu:
    return "AArch64ISD::NEON_RSRAu";
  case AArch64ISD::NEON_SRAs:
    return "AArch64ISD::NEON_SRAs";
  case AArch64ISD::NEON_SRAu:
    return "AArch64ISD::NEON_SRAu";

Thanks,
-Jiangning


2013/8/30 Hao Liu <Hao.Liu at arm.com>

> Hi,
>
> Attached are patches to implement aarch64 neon instruction class AdvSIMD
> (shift), which patches implemented 21 shift instructions and 4 convert
> instructions. Most of them are implemented like ARMv7, except :
> 1)  SHRN is implemented by IR (lshr/ashr, tuncate) instead of IR
> intrinsics.
> 2)  There are some special instructions added in AArch64: shift narrow
> high,
> which is implemented by combining shuffle vector and normal shift narrow
> instructions.
> 3) Simplify  vshll_high_n/vmovl_high ACLE functions by calling vget_high
> and
> vshll_n, instead of adding new __builtins.  In arm_neon.h, We can't write
> code like:
>                        (int16x8_t)vshll_n_s8(vget_high_s8(__a), __b),
>      as there is a local variable in Macro vshll_n_s8 also named "__a".
> Macro expanding will generate code like "__a = vget_high_s8(__a)", which
> will report an error when compiling ACLE functions.
>      So I make the generated code in arm_neon.h as following:
>                        int8x8_t __a1 = vget_high_s8(__a);
>                       (int16x8_t)vshll_n_s8(__a1, __b);
>
> There are some changed in Clang as new types need to be supported:
> 1)  Add a new naming type 'm' in arm_neon.td to generate "half width
> elements, same num elts", as there are some complicated  ACLE functions
> can't be generated, such as "int8x16_t vqshrn_high_n_s16(int8x8_t,
> int16x8_t, )".
> 2)  There is a minor bug about double type in ModType() NeonEmitter.cpp,
> which will result in wrong NeonTypeFlages. Fix it by adding:
>               +      if (type == 'd')
>               +        type = 'l';
> 3) Enable the range check for VCVT_N can check [1, 64] for floating-point
> 64-bit type in NeonEmitter.cpp.
>
>
> Ask for review, please.
>
> Thanks,
> -Hao
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
>


-- 
Thanks,
-Jiangning
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20130830/fa8c5c5d/attachment.html>