[Mesa-dev] [PATCH] i965/fs: Fail the shader compile instead of asserting when we can't spill

[Francisco Jerez]

Jason Ekstrand <jason at jlekstrand.net> writes:

> Blorp doesn't handle spilling so we set allow_spilling to false in that
> case.  The blorp 16x MSAA resolve shader spills in 16-wide but not 8-wide.
> This commit makes it so that we fail the 16-wide compile and successfully
> fall back to 8-wide instead of just assert-failing when trying to compile
> the 16-wide shader.
>
> Signed-off-by: Jason Ekstrand <jason at jlekstrand.net>
> ---
>  src/mesa/drivers/dri/i965/brw_fs.cpp | 7 +++++--
>  1 file changed, 5 insertions(+), 2 deletions(-)
>
> diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp b/src/mesa/drivers/dri/i965/brw_fs.cpp
> index d0b55ae..73aa5d2 100644
> --- a/src/mesa/drivers/dri/i965/brw_fs.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_fs.cpp
> @@ -5906,6 +5906,11 @@ fs_visitor::allocate_registers(bool allow_spilling)
>     }
>  
>     if (!allocated_without_spills) {
> +      if (!allow_spilling) {
> +         assert(dispatch_width > 8);

Not sure it makes sense for the register allocator to set different
requirements based on the build dispatch width, wouldn't it make sense
to just fail() here consistently if allow_spilling is false regardless
of the dispatch width, and let blorp assert fail if none of the dispatch
widths it tries out compiles successfully?

> +         fail("Failure to register allocate and spilling is not allowed.");
> +      }
> +
>        /* We assume that any spilling is worse than just dropping back to
>         * SIMD8.  There's probably actually some intermediate point where
>         * SIMD16 with a couple of spills is still better.
> @@ -5930,8 +5935,6 @@ fs_visitor::allocate_registers(bool allow_spilling)
>        }
>     }
>  
> -   assert(last_scratch == 0 || allow_spilling);
> -
>     /* This must come after all optimization and register allocation, since
>      * it inserts dead code that happens to have side effects, and it does
>      * so based on the actual physical registers in use.
> -- 
> 2.5.0.400.gff86faf
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 212 bytes
Desc: not available
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20160908/ce98d46d/attachment.sig>