[PATCH] [AArch64] Fix a pattern match failure for movi with 1D result

[Jiangning Liu]

OK, please! Thank you!

-Jiangning
2013-12-7 上午11:33于 "Ana Pazos" <apazos at codeaurora.org>写道：

> Hi Jiangning,
>
>
>
> I suggest we do these changes to the test case:
>
>
>
> - Put the intrinsic test you created into the existing
> /test/CodeGen/AArch64/neon-aba-abd.ll instead of creating a new file.
>
>   That is where the sabd tests are.
>
>
>
> - Add a couple of test cases into test/CodeGen/AArch64/neon-mov.ll to
> cover movi 1D scenario without using the intrinsic as well.
>
>
>
> If you want I can do the changes and check in for you, if you are OOO.
>
>
>
> Thanks,
>
> Ana.
>
>
>
> Examples:
>
>
>
> +define <2 x i32> @movi1d_1() {
>
> +; CHECK: movi    d0, #0xffffffff0000
>
> +  ret <2 x i32> < i32  -65536, i32 65535>
>
> +}
>
> +
>
> +
>
> +declare <2 x i32> @test_movi1d(<2 x i32>, <2 x i32>)
>
> +define <2 x i32> @movi1d() {
>
> +; CHECK: movi     d1, #0xffffffff0000
>
> +  %1 = tail call <2 x i32> @test_movi1d(<2 x i32> <i32 -2147483648, i32
> 2147450880>, <2 x i32> <i32 -65536, i32 65535>)
>
> +  ret <2 x i32> %1
>
> +}
>
>
>
> Thanks,
>
> Ana.
>
>
>
> *From:* llvm-commits-bounces at cs.uiuc.edu [mailto:
> llvm-commits-bounces at cs.uiuc.edu] *On Behalf Of *Jiangning Liu
> *Sent:* Friday, December 06, 2013 5:07 AM
> *To:* llvm-commits at cs.uiuc.edu for LLVM
> *Subject:* [PATCH] [AArch64] Fix a pattern match failure for movi with 1D
> result
>
>
>
> Hi,
>
> Attached patch is to fix a pattern match failure for neon movi with scalar
> 1D result.
>
> diff --git a/lib/Target/AArch64/AArch64InstrNEON.td
> b/lib/Target/AArch64/AArch64InstrNEON.td
> old mode 100644
> new mode 100755
> index eb62c13..7a63935
> --- a/lib/Target/AArch64/AArch64InstrNEON.td
> +++ b/lib/Target/AArch64/AArch64InstrNEON.td
> @@ -1423,9 +1423,8 @@ let isReMaterializable = 1 in {
>  def MOVIdi : NeonI_1VModImm<0b0, 0b1,
>                             (outs FPR64:$Rd), (ins neon_uimm64_mask:$Imm),
>                             "movi\t $Rd, $Imm",
> -                           [(set (f64 FPR64:$Rd),
> -                              (f64 (bitconvert
> -                                (v1i64 (Neon_movi (timm:$Imm), (i32
> imm))))))],
> +                           [(set (v1i64 FPR64:$Rd),
> +                              (v1i64 (Neon_movi (timm:$Imm), (i32
> imm))))],
>                             NoItinerary> {
>    let cmode = 0b1110;
>  }
> diff --git a/test/CodeGen/AArch64/neon-movi.ll
> b/test/CodeGen/AArch64/neon-movi.ll
> new file mode 100644
> index 0000000..680ec47
> --- /dev/null
> +++ b/test/CodeGen/AArch64/neon-movi.ll
> @@ -0,0 +1,8 @@
> +; RUN: llc < %s -O2 -verify-machineinstrs -mtriple=aarch64-none-linux-gnu
> -mattr=+neon | FileCheck %s
> +
> +declare <2 x i32> @llvm.arm.neon.vabds.v2i32(<2 x i32>, <2 x i32>)
> +define <2 x i32> @test_movi_1d() {
> +; CHECK: movi     d1, #0xffffffff0000
> +  %1 = tail call <2 x i32> @llvm.arm.neon.vabds.v2i32(<2 x i32> <i32
> -2147483648, i32 2147450880>, <2 x i32> <i32 -65536, i32 65535>)
> +  ret <2 x i32> %1
> +}
>
>
> --
>
> Thanks,
>
> -Jiangning
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20131208/cdd91983/attachment.html>