[Mesa-dev] [PATCH] gallium/radeon: unify and simplify checking for an empty gfx IB
Marek Olšák
maraeo at gmail.com
Thu Aug 25 19:20:52 UTC 2016
FYI, I've already committed this because our other team needs this bug fix.
Marek
On Thu, Aug 25, 2016 at 10:56 AM, Marek Olšák <maraeo at gmail.com> wrote:
> From: Marek Olšák <marek.olsak at amd.com>
>
> We can take advantage of the fact that multi_fence does the obvious thing
> with NULL fences.
>
> This fixes unflushed fences that can get stuck due to empty IBs.
> ---
> src/gallium/drivers/r600/r600_hw_context.c | 8 +-------
> src/gallium/drivers/radeon/r600_pipe_common.c | 29 ++++++++++++++++++---------
> src/gallium/drivers/radeonsi/si_hw_context.c | 13 +++---------
> 3 files changed, 23 insertions(+), 27 deletions(-)
>
> diff --git a/src/gallium/drivers/r600/r600_hw_context.c b/src/gallium/drivers/r600/r600_hw_context.c
> index 4447393..58ba09d 100644
> --- a/src/gallium/drivers/r600/r600_hw_context.c
> +++ b/src/gallium/drivers/r600/r600_hw_context.c
> @@ -248,28 +248,22 @@ void r600_flush_emit(struct r600_context *rctx)
> rctx->b.flags = 0;
> }
>
> void r600_context_gfx_flush(void *context, unsigned flags,
> struct pipe_fence_handle **fence)
> {
> struct r600_context *ctx = context;
> struct radeon_winsys_cs *cs = ctx->b.gfx.cs;
> struct radeon_winsys *ws = ctx->b.ws;
>
> - if (!radeon_emitted(cs, ctx->b.initial_gfx_cs_size) &&
> - (!fence || ctx->b.last_gfx_fence)) {
> - if (fence)
> - ws->fence_reference(fence, ctx->b.last_gfx_fence);
> - if (!(flags & RADEON_FLUSH_ASYNC))
> - ws->cs_sync_flush(cs);
> + if (!radeon_emitted(cs, ctx->b.initial_gfx_cs_size))
> return;
> - }
>
> r600_preflush_suspend_features(&ctx->b);
>
> /* flush the framebuffer cache */
> ctx->b.flags |= R600_CONTEXT_FLUSH_AND_INV |
> R600_CONTEXT_FLUSH_AND_INV_CB |
> R600_CONTEXT_FLUSH_AND_INV_DB |
> R600_CONTEXT_FLUSH_AND_INV_CB_META |
> R600_CONTEXT_FLUSH_AND_INV_DB_META |
> R600_CONTEXT_WAIT_3D_IDLE |
> diff --git a/src/gallium/drivers/radeon/r600_pipe_common.c b/src/gallium/drivers/radeon/r600_pipe_common.c
> index ab620eb..b1da22f 100644
> --- a/src/gallium/drivers/radeon/r600_pipe_common.c
> +++ b/src/gallium/drivers/radeon/r600_pipe_common.c
> @@ -258,54 +258,63 @@ void r600_postflush_resume_features(struct r600_common_context *ctx)
> if (!LIST_IS_EMPTY(&ctx->active_queries))
> r600_resume_queries(ctx);
> }
>
> static void r600_flush_from_st(struct pipe_context *ctx,
> struct pipe_fence_handle **fence,
> unsigned flags)
> {
> struct pipe_screen *screen = ctx->screen;
> struct r600_common_context *rctx = (struct r600_common_context *)ctx;
> + struct radeon_winsys *ws = rctx->ws;
> unsigned rflags = 0;
> struct pipe_fence_handle *gfx_fence = NULL;
> struct pipe_fence_handle *sdma_fence = NULL;
> bool deferred_fence = false;
>
> if (flags & PIPE_FLUSH_END_OF_FRAME)
> rflags |= RADEON_FLUSH_END_OF_FRAME;
> if (flags & PIPE_FLUSH_DEFERRED)
> rflags |= RADEON_FLUSH_ASYNC;
>
> if (rctx->dma.cs) {
> rctx->dma.flush(rctx, rflags, fence ? &sdma_fence : NULL);
> }
>
> - /* Instead of flushing, create a deferred fence. Constraints:
> - * - The state tracker must allow a deferred flush.
> - * - The state tracker must request a fence.
> - * Thread safety in fence_finish must be ensured by the state tracker.
> - */
> - if (flags & PIPE_FLUSH_DEFERRED && fence) {
> - gfx_fence = rctx->ws->cs_get_next_fence(rctx->gfx.cs);
> - deferred_fence = true;
> + if (!radeon_emitted(rctx->gfx.cs, rctx->initial_gfx_cs_size)) {
> + if (fence)
> + ws->fence_reference(&gfx_fence, rctx->last_gfx_fence);
> + if (!(rflags & RADEON_FLUSH_ASYNC))
> + ws->cs_sync_flush(rctx->gfx.cs);
> } else {
> - rctx->gfx.flush(rctx, rflags, fence ? &gfx_fence : NULL);
> + /* Instead of flushing, create a deferred fence. Constraints:
> + * - The state tracker must allow a deferred flush.
> + * - The state tracker must request a fence.
> + * Thread safety in fence_finish must be ensured by the state tracker.
> + */
> + if (flags & PIPE_FLUSH_DEFERRED && fence) {
> + gfx_fence = rctx->ws->cs_get_next_fence(rctx->gfx.cs);
> + deferred_fence = true;
> + } else {
> + rctx->gfx.flush(rctx, rflags, fence ? &gfx_fence : NULL);
> + }
> }
>
> /* Both engines can signal out of order, so we need to keep both fences. */
> - if (gfx_fence || sdma_fence) {
> + if (fence) {
> struct r600_multi_fence *multi_fence =
> CALLOC_STRUCT(r600_multi_fence);
> if (!multi_fence)
> return;
>
> multi_fence->reference.count = 1;
> + /* If both fences are NULL, fence_finish will always return true. */
> multi_fence->gfx = gfx_fence;
> multi_fence->sdma = sdma_fence;
>
> if (deferred_fence) {
> multi_fence->gfx_unflushed.ctx = rctx;
> multi_fence->gfx_unflushed.ib_index = rctx->num_gfx_cs_flushes;
> }
>
> screen->fence_reference(screen, fence, NULL);
> *fence = (struct pipe_fence_handle*)multi_fence;
> diff --git a/src/gallium/drivers/radeonsi/si_hw_context.c b/src/gallium/drivers/radeonsi/si_hw_context.c
> index 49d1a35..aeccb2d 100644
> --- a/src/gallium/drivers/radeonsi/si_hw_context.c
> +++ b/src/gallium/drivers/radeonsi/si_hw_context.c
> @@ -93,31 +93,24 @@ void si_need_cs_space(struct si_context *ctx)
> void si_context_gfx_flush(void *context, unsigned flags,
> struct pipe_fence_handle **fence)
> {
> struct si_context *ctx = context;
> struct radeon_winsys_cs *cs = ctx->b.gfx.cs;
> struct radeon_winsys *ws = ctx->b.ws;
>
> if (ctx->gfx_flush_in_progress)
> return;
>
> - ctx->gfx_flush_in_progress = true;
> -
> - if (!radeon_emitted(cs, ctx->b.initial_gfx_cs_size) &&
> - (!fence || ctx->b.last_gfx_fence)) {
> - if (fence)
> - ws->fence_reference(fence, ctx->b.last_gfx_fence);
> - if (!(flags & RADEON_FLUSH_ASYNC))
> - ws->cs_sync_flush(cs);
> - ctx->gfx_flush_in_progress = false;
> + if (!radeon_emitted(cs, ctx->b.initial_gfx_cs_size))
> return;
> - }
> +
> + ctx->gfx_flush_in_progress = true;
>
> r600_preflush_suspend_features(&ctx->b);
>
> ctx->b.flags |= SI_CONTEXT_CS_PARTIAL_FLUSH |
> SI_CONTEXT_PS_PARTIAL_FLUSH;
>
> /* DRM 3.1.0 doesn't flush TC for VI correctly. */
> if (ctx->b.chip_class == VI && ctx->b.screen->info.drm_minor <= 1)
> ctx->b.flags |= SI_CONTEXT_INV_GLOBAL_L2 |
> SI_CONTEXT_INV_VMEM_L1;
> --
> 2.7.4
>
More information about the mesa-dev
mailing list