[Mesa-dev] [PATCH] radv: disable the auto-waitcnt-before-barrier LLVM option

[Bas Nieuwenhuizen]

Reviewed-by: Bas Nieuwenhuizen <bas at basnieuwenhuizen.nl>

On Wed, Aug 15, 2018 at 3:09 PM, Samuel Pitoiset
<samuel.pitoiset at gmail.com> wrote:
> This option allows us to remove additional s_waitcnt instructions
> because s_barrier internally does s_waitcnt 0.
>
> Though, apparently there is a problem with LDS accesses that
> causes rendering issues with FFXV and DXVK. Disable this
> optimization for now (RadeonSI still uses it).
>
> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107460
> CC: 18.2 <mesa-stable at lists.freedesktop.org>
> Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
> ---
>  src/amd/common/ac_llvm_util.c          | 3 ++-
>  src/amd/common/ac_llvm_util.h          | 1 +
>  src/gallium/drivers/radeonsi/si_pipe.c | 1 +
>  3 files changed, 4 insertions(+), 1 deletion(-)
>
> diff --git a/src/amd/common/ac_llvm_util.c b/src/amd/common/ac_llvm_util.c
> index 10e1ca99d4..008a2cfa4e 100644
> --- a/src/amd/common/ac_llvm_util.c
> +++ b/src/amd/common/ac_llvm_util.c
> @@ -149,7 +149,8 @@ static LLVMTargetMachineRef ac_create_target_machine(enum radeon_family family,
>         char features[256];
>         const char *triple = (tm_options & AC_TM_SUPPORTS_SPILL) ? "amdgcn-mesa-mesa3d" : "amdgcn--";
>         LLVMTargetRef target = ac_get_llvm_target(triple);
> -       bool barrier_does_waitcnt = family != CHIP_VEGA20;
> +       bool barrier_does_waitcnt = tm_options & AC_TM_AUTO_WAITCNT_BEFORE_BARRIER &&

Can you please do parentheses around the and? I know they're not
strictly necessary but I'd like not to have to
think about precedence too much.

> +                                   family != CHIP_VEGA20;
>
>         snprintf(features, sizeof(features),
>                  "+DumpCode,+vgpr-spilling,-fp32-denormals,+fp64-denormals%s%s%s%s%s",
> diff --git a/src/amd/common/ac_llvm_util.h b/src/amd/common/ac_llvm_util.h
> index eaf5f21876..e252bed3bb 100644
> --- a/src/amd/common/ac_llvm_util.h
> +++ b/src/amd/common/ac_llvm_util.h
> @@ -65,6 +65,7 @@ enum ac_target_machine_options {
>         AC_TM_CHECK_IR = (1 << 5),
>         AC_TM_ENABLE_GLOBAL_ISEL = (1 << 6),
>         AC_TM_CREATE_LOW_OPT = (1 << 7),
> +       AC_TM_AUTO_WAITCNT_BEFORE_BARRIER = (1 << 8),
>  };
>
>  enum ac_float_mode {
> diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
> index 13fcf1f3ae..4327a3f749 100644
> --- a/src/gallium/drivers/radeonsi/si_pipe.c
> +++ b/src/gallium/drivers/radeonsi/si_pipe.c
> @@ -115,6 +115,7 @@ static void si_init_compiler(struct si_screen *sscreen,
>                                        sscreen->info.chip_class <= VI;
>
>         enum ac_target_machine_options tm_options =
> +               AC_TM_AUTO_WAITCNT_BEFORE_BARRIER |
>                 (sscreen->debug_flags & DBG(SI_SCHED) ? AC_TM_SISCHED : 0) |
>                 (sscreen->debug_flags & DBG(GISEL) ? AC_TM_ENABLE_GLOBAL_ISEL : 0) |
>                 (sscreen->info.chip_class >= GFX9 ? AC_TM_FORCE_ENABLE_XNACK : 0) |
> --
> 2.18.0
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev