[Mesa-dev] [PATCH 1/8] radv: don't flush src stages when dstStageMask == BOTTOM_OF_PIPE

[Fredrik Höglund]

On Wednesday 27 June 2018, Bas Nieuwenhuizen wrote:
> Don't we still need this when having layout transitions?

I think the answer is probably yes, though I haven't noticed any
regressions with this patch.

I'll post an updated version that takes transitions into account.

> On Tue, Jun 26, 2018 at 11:49 PM, Fredrik Höglund <fredrik at kde.org> wrote:
> > The Vulkan specification says:
> >
> >    "An execution dependency with only VK_PIPELINE_STAGE_BOTTOM_OF_-
> >     PIPE_BIT in the destination stage mask [...] does not delay
> >     processing of subsequent commands."
> >
> > Cc: <mesa-stable at lists.freedesktop.org>
> > Signed-off-by: Fredrik Höglund <fredrik at kde.org>
> > ---
> >  src/amd/vulkan/radv_cmd_buffer.c | 3 ++-
> >  src/amd/vulkan/radv_pass.c       | 6 ++++--
> >  2 files changed, 6 insertions(+), 3 deletions(-)
> >
> > diff --git a/src/amd/vulkan/radv_cmd_buffer.c b/src/amd/vulkan/radv_cmd_buffer.c
> > index 110a9a960a9..5bfcba28d83 100644
> > --- a/src/amd/vulkan/radv_cmd_buffer.c
> > +++ b/src/amd/vulkan/radv_cmd_buffer.c
> > @@ -4197,7 +4197,8 @@ void radv_CmdPipelineBarrier(
> >                                                         image);
> >         }
> >
> > -       radv_stage_flush(cmd_buffer, srcStageMask);
> > +       if (destStageMask != VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT)
> > +               radv_stage_flush(cmd_buffer, srcStageMask);
> >         cmd_buffer->state.flush_bits |= src_flush_bits;
> >
> >         for (uint32_t i = 0; i < imageMemoryBarrierCount; i++) {
> > diff --git a/src/amd/vulkan/radv_pass.c b/src/amd/vulkan/radv_pass.c
> > index 15fee444cdc..7a0dca09496 100644
> > --- a/src/amd/vulkan/radv_pass.c
> > +++ b/src/amd/vulkan/radv_pass.c
> > @@ -174,11 +174,13 @@ VkResult radv_CreateRenderPass(
> >         for (unsigned i = 0; i < pCreateInfo->dependencyCount; ++i) {
> >                 uint32_t dst = pCreateInfo->pDependencies[i].dstSubpass;
> >                 if (dst == VK_SUBPASS_EXTERNAL) {
> > -                       pass->end_barrier.src_stage_mask = pCreateInfo->pDependencies[i].srcStageMask;
> > +                       if (pCreateInfo->pDependencies[i].dstStageMask != VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT)
> > +                               pass->end_barrier.src_stage_mask = pCreateInfo->pDependencies[i].srcStageMask;
> >                         pass->end_barrier.src_access_mask = pCreateInfo->pDependencies[i].srcAccessMask;
> >                         pass->end_barrier.dst_access_mask = pCreateInfo->pDependencies[i].dstAccessMask;
> >                 } else {
> > -                       pass->subpasses[dst].start_barrier.src_stage_mask = pCreateInfo->pDependencies[i].srcStageMask;
> > +                       if (pCreateInfo->pDependencies[i].dstStageMask != VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT)
> > +                               pass->subpasses[dst].start_barrier.src_stage_mask = pCreateInfo->pDependencies[i].srcStageMask;
> >                         pass->subpasses[dst].start_barrier.src_access_mask = pCreateInfo->pDependencies[i].srcAccessMask;
> >                         pass->subpasses[dst].start_barrier.dst_access_mask = pCreateInfo->pDependencies[i].dstAccessMask;
> >                 }
> > --
> > 2.17.0
> >
> > _______________________________________________
> > mesa-dev mailing list
> > mesa-dev at lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>