[Libclc-dev] vload problems (was: Re: [PATCH 1/2] amdgcn: Implement {read_, write_, }mem_fence builtin)

[Jan Vesely via Libclc-dev]

On Sat, 2017-08-12 at 13:18 -0500, Aaron Watry wrote:
> On Fri, Aug 11, 2017 at 8:59 PM, Jan Vesely via Libclc-dev
> <libclc-dev at lists.llvm.org> wrote:
> > Signed-off-by: Jan Vesely <jan.vesely at rutgers.edu>
> > ---
> > 
> > __builtin_amdgcn_s_waitcnt path is compile tested only. I currently
> > don't have machine with GCN hw and LLVM > 4
> 
> Tested on GCN 1.0 (PITCAIRN) with LLVM 6.0 svn (from sometime last week).
> 
> I ran the local-memory/global-memory piglit tests, and the conformance
> basic/test_basic vload*, vstore*, and barrier tests. No change in
> pass-rate.
> 
> The only ones that fail are vload_private with char/uchar/short/ushort
> data types, but those failed before your series.

do you remember if those failures are new in LLVM 6? I just posted new
vload/vstore piglit tests, and those ran OK* on carrizo/iceland system
using LLVM 5.

*mostly. vload_half is missing from libclc to those test failed.
Unlike Turks, which fails ~40% of them (LLVM 6).

regards,
Jan

[SNIP]
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: This is a digitally signed message part
URL: <http://lists.llvm.org/pipermail/libclc-dev/attachments/20170816/d8a38eb9/attachment.sig>