On 17/10/2025 15:35, Thomas Schwinge wrote:
Hi Andrew!
On 2025-09-09T16:52:57+0000, Andrew Stubbs <[email protected]> wrote:
The previous definition had all the GFX11 register counts doubled to fix a bug
that was encountered in early testing. This seems to have been a
misunderstanding of the problem (which is no longer reproducible).
I can't comment on the historic aspects, but I can tell that since this
commit r16-3726-g7bc2e311688ac279f1abc2a47944e5b763f7ec89
"amdgcn: fix GFX10/GFX11 VGPR counts", '-march=gfx1100' testing is
completely broken; nothing but:
Memory access fault by GPU node-2 (Agent handle: [...]) on address (nil).
Reason: Page not present or supervisor privilege.
May I 'git push' my 'git revert', or should I keep that local, awaiting
your investigation?
It works for me!??????
I'm aware of PR121393, but that only affects the "for-<n>" testcases
(and was exposed by the vectorization improvements, not this).
Shouldn't the commit you pushed also have removed the following from
'gcc/config/gcn/gcn.cc:gcn_hsa_declare_function_name':
fprintf (file,
".kd\n"
[...]
" .vgpr_count: %i%s\n"
[...], next_free_vgpr,
(TARGET_WAVE64_COMPAT
? " ; wavefrontsize64 counts double on SIMD32"
: ""));
Good catch. It's just a comment though.
Andrew
