[llvm] r326811 - [AArch64] define isExtractSubvectorCheap

[David Green via llvm-commits]

Brilliant, thanks for the quick fix!

Dave


From: Sebastian Pop <sebpop at gmail.com>
Sent: 07 March 2018 22:10
To: David Green; llvm-commits
Subject: Re: [llvm] r326811 - [AArch64] define isExtractSubvectorCheap

Fixed in https://reviews.llvm.org/rL326949

On Wed, Mar 7, 2018 at 3:51 PM, Sebastian Pop <sebpop at gmail.com<mailto:sebpop at gmail.com>> wrote:
The error is due to a missing def-pat for (i64 0)

ISEL: Starting pattern match
  Initial Opcode index to 85068
  Match failed at index 85076
LLVM ERROR: Cannot select: t47: v8i16 = insert_subvector undef:v8i16, t43, Constant:i64<0>

The code generated from llvm/lib/Target/AArch64/AArch64InstrInfo.td

def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (i32 0)),
          (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;

is in ninja/lib/Target/AArch64/AArch64GenDAGISel.inc
At the location of the error it is:
/* 85076*/    OPC_CheckChild2Type, MVT::i32,

And it failed to match the type of operand 2.

adding another def-pat like this fixes it:

def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (i64 0)),
          (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;

I will submit a patch.

Sebastian

On Wed, Mar 7, 2018 at 2:50 PM, Sebastian Pop <sebpop at gmail.com<mailto:sebpop at gmail.com>> wrote:
I'm looking at the reduced test.
Thanks for the heads'up.

On Wed, Mar 7, 2018 at 11:11 AM, David Green <David.Green at arm.com<mailto:David.Green at arm.com>> wrote:
Hello

I believe this is the commit causing this code to fail to compile. It's reduced from some randomised neon code, so sorry if it's a bit ugly. Running with llc claims:

error in backend: Cannot select: t47: v8i16 = insert_subvector undef:v8i16, t43,
      Constant:i64<0>


target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
target triple = "aarch64--linux-gnu"

@d = common dso_local local_unnamed_addr global <4 x i16> zeroinitializer, align 8

define <8 x i16> @c(i32 %e) {
entry:
  %0 = load <4 x i16>, <4 x i16>* @d, align 8
  %vminv = tail call i32 @llvm.aarch64.neon.uminv.i32.v4i16(<4 x i16> %0)
  %1 = trunc i32 %vminv to i16
  %vecinit3 = insertelement <4 x i16> <i16 undef, i16 undef, i16 0, i16 0>, i16 %1, i32 1
  %call = tail call <8 x i16> @c(i32 0) #3
  %vgetq_lane = extractelement <8 x i16> %call, i32 0
  %vset_lane = insertelement <4 x i16> %vecinit3, i16 %vgetq_lane, i32 0
  %call4 = tail call i32 bitcast (i32 (...)* @k to i32 (<4 x i16>)*)(<4 x i16> %vset_lane) #3
  ret <8 x i16> undef
}

declare i32 @llvm.aarch64.neon.uminv.i32.v4i16(<4 x i16>)
declare i32 @k(...)


Any ideas? I'm surprised this isExtractSubvectorCheap would be causing this.
Cheers
Dave
IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.