Bug 99850

Created attachment 129693 [details]
Carrizo run with HEAD~

Dug up the original thread that introduced this patch.  Might prove useful in debugging it.

https://lists.freedesktop.org/archives/mesa-dev/2016-May/116152.html

Bas, any ideas?

Alex commented that mesa could use this addition to the kmd to selectively enable/disable this feature on APUs

commit 921e71c548b313f704745dcc609ffb0206da1afe                  
Author:     Junwei Zhang <Jerry.Zhang@amd.com>                   
AuthorDate: Fri Feb 17 11:05:49 2017 +0800                       
Commit:     Junwei Zhang <Jerry.Zhang@amd.com>                   
CommitDate: Tue Feb 21 10:22:30 2017 +0800                       

    drm/amdgpu: export gfx config double offchip LDS buffers (v3)

    v2: move the config struct to drm_amdgpu_info_device
    v3: move the config feature to amdgpu_gca_config

    Change-Id: I9651ab3e06a2d3c891b4c52a5da5b4a45d5809e5
    Signed-off-by: Junwei Zhang <Jerry.Zhang@amd.com>            
    Reviewed-by: Christian König <christian.koenig@amd.com>

(In reply to Tom St Denis from comment #4)
> Alex commented that mesa could use this addition to the kmd to selectively
> enable/disable this feature on APUs

Not to selectively enable, but I think it might be related.  We may need some adjustments in the UMD to deal with the hw differences.

You can try this diff:

diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c
index f615aa8..750cdd6 100644
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -2312,7 +2312,9 @@ static bool si_update_spi_tmpring_size(struct si_context *sctx)
 
 static void si_init_tess_factor_ring(struct si_context *sctx)
 {
-       bool double_offchip_buffers = sctx->b.chip_class >= CIK;
+       bool double_offchip_buffers = sctx->b.chip_class >= CIK &&
+                                     sctx->b.family != CHIP_CARRIZO &&
+                                     sctx->b.family != CHIP_STONEY;
        unsigned max_offchip_buffers_per_se = double_offchip_buffers ? 128 : 64;
        unsigned max_offchip_buffers = max_offchip_buffers_per_se *
                                       sctx->screen->b.info.max_se;

I'll try this momentarily (rig is configured for polaris10).

Probably cleaner though to check the flag from the KMD though in the long run right?

With that diff applied unigine-heaven works with tess enabled on the tip of mesa{master} as of right now.

Confirmed 1759 "pass"'es with tessellation tests in piglit vs 1724 passes with HEAD~.

Yipee.

Bug reports should only be resolved once the fix has landed in Git, preferably with a reference to the commit fixing it.

BTW, always specify explicit Git commit hashes instead of "HEAD". HEAD refers to the currently checked out commit, it has no meaning outside of your local repository.

Thanks Michel, I was a bit quick in closing the bug.

Note the bug description mentions the offending commit hash so instead of pasting that everywhere I called that HEAD and the good commit HEAD~.  Conventions I suppose... :-)

Should be fixed by 35915af6c9ab4bdc0f1f8584ca346602405bd7e4.