[llvm-commits] [PATCH] Use vld1/vst1 for unaligned load/store

[David Tweed]

Just a note to say that this commit is causing 3 tests in the regression
suite to fail on ARM/Linux

 

test/CodeGen/ARM/twoaddrinstr.ll

test/CodeGen/ARM/reg_sequence.ll

test/CodeGen/ARM/vbsl-constant.ll

 

I'm haven't yet had time to look into them in any detail and see if it's a
test that's no longer valid or it's actually producing wrong code. (It all
seems to come down to 64 not 32 bit instructions, which the changelog
suggests is intended and the test is wrong:

 

/home/buildslave/buildslave/runtests/llvm/test/CodeGen/ARM/twoaddrinstr.ll:7
:10: error: expected string not found in input

; CHECK: vld1.32

         ^

<stdin>:15:10: note: scanning from here

PR13378: @ @PR13378

         ^

<stdin>:18:2: note: possible intended match here

vld1.64 {d0, d1}, [r0]

 

)

 

Regards,

David Tweed

 

From: llvm-commits-bounces at cs.uiuc.edu
[mailto:llvm-commits-bounces at cs.uiuc.edu] On Behalf Of Evan Cheng
Sent: 18 September 2012 02:48
To: David Peixotto
Cc: llvm-commits at cs.uiuc.edu
Subject: Re: [llvm-commits] [PATCH] Use vld1/vst1 for unaligned load/store

 

r164089. Thanks.

 

Evan

 

On Sep 17, 2012, at 11:11 AM, Evan Cheng <evan.cheng at apple.com> wrote:





Hi David,

 

Thanks for working on this. This is a big omission that I was planning to
look at. It's good you got to it first. Some comments though:

 

 bool ARMTargetLowering::allowsUnalignedMemoryAccesses(EVT VT) const {


-  if (!Subtarget->allowsUnalignedMem())


-    return false;


+  // The AllowsUnaliged flag models the SCTLR.A setting in ARM cpus


+  bool AllowsUnaligned = Subtarget->allowsUnalignedMem();


 


   switch (VT.getSimpleVT().SimpleTy) {


   default:


@@ -9034,10 +9034,15 @@ bool
ARMTargetLowering::allowsUnalignedMemoryAccesses(EVT VT) const {

   case MVT::i8:


   case MVT::i16:


   case MVT::i32:


-    return true;


+    // Unaligned access can use (for example) LRDB, LRDH, LDR


+    return AllowsUnaligned;


   case MVT::f64:


-    return Subtarget->hasNEON();


-  // FIXME: VLD1 etc with standard alignment is legal.


+  case MVT::v2f64:


+    // For any little-endian targets with neon, we can support unaligned
ld/st


+    // of D and Q (e.g. {D0,D1}) registers by using vld1.i8/vst1.i8.


+    // A big-endian target may also explictly support unaligned accesses


+    return Subtarget->hasNEON() &&


+           (getTargetData()->isLittleEndian() || AllowsUnaligned);


   }


 } 

 

This part is not quite right:

+    return Subtarget->hasNEON() &&


+           (getTargetData()->isLittleEndian() || AllowsUnaligned);   

 

vld1 / vst1 requires alignment of element size. If not, then it's a fault
unless SCTLR.A is 1.  This should not require true for all little endian
cpus with NEON. It should still be controlled by the subtarget feature.

 

I'll fix up your patch and commit it for you. Thanks.

 

Evan

 

 

 

On Sep 13, 2012, at 5:53 PM, David Peixotto <dpeixott at codeaurora.org> wrote:





This patch is the result of a discussion of unaligned vector loads/store on
llvmdev:
<http://lists.cs.uiuc.edu/pipermail/llvmdev/2012-September/053082.html>
http://lists.cs.uiuc.edu/pipermail/llvmdev/2012-September/053082.html.

 

The vld1 and vst1 variants in armv7 neon only require memory

alignment to the element size of the vector. Because of this

property, we can use a vld1.8 and vst1.8 to load/store f64 and v2f64

vectors to unaligned addresses on little-endian targets. This should

be faster than the target-independent codegen lowering that does an

aligned load/store to the stack and unaligned load/store of each

element of the vector.

 

This patch includes two changes:

  1. Add new patterns for selecting vld1/vst1 for byte and half-word

     aligned vector stores for v2f64 vectors.

  2. Allow unaligned load/store using vld1/vst1 for little-endian

     arm targets that support NEON.  The vld1/vst1 instructions will

     be used to load/store f64 and v2f64 types aligned along byte

     and half-word memory accesses.

 

-- Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted
by The Linux Foundation

 

 

<0001-Use-vld1-vst1-for-unaligned-load-store.patch>_________________________
______________________
llvm-commits mailing list
 <mailto:llvm-commits at cs.uiuc.edu> llvm-commits at cs.uiuc.edu
 <http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits>
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits

 

_______________________________________________
llvm-commits mailing list
llvm-commits at cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits

 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20120920/d3a7d387/attachment.html>