This resolves some rendering issues in source games. See https://bugs.freedesktop.org/show_bug.cgi?id=64323
Signed-off-by: Ilia Mirkin <imir...@alum.mit.edu> Cc: "9.2 10.0" <mesa-sta...@lists.freedesktop.org> --- Doing a nouveau_bo_wait works as well, but I got a slightly higher framerate from glretrace doing it this way. I tried to get an actual source game running, but was unsuccessful... (something about a missing filesystem_steam.so which indeed was absent) This is clearly not optimal, but neither is having broken source games. The other workaround simply went the other path which would have to do a wait and a double-copy (vram -> pushbuf, pushbuf xfer from gart), which is worse than just the wait. src/gallium/drivers/nouveau/nouveau_buffer.c | 3 +++ src/gallium/drivers/nouveau/nv50/nv50_vbo.c | 9 +++++++++ 2 files changed, 12 insertions(+) diff --git a/src/gallium/drivers/nouveau/nouveau_buffer.c b/src/gallium/drivers/nouveau/nouveau_buffer.c index 3e04049..95905a8 100644 --- a/src/gallium/drivers/nouveau/nouveau_buffer.c +++ b/src/gallium/drivers/nouveau/nouveau_buffer.c @@ -205,6 +205,9 @@ nouveau_transfer_write(struct nouveau_context *nv, struct nouveau_transfer *tx, base, size / 4, (const uint32_t *)data); else nv->push_data(nv, buf->bo, buf->offset + base, buf->domain, size, data); + + nouveau_fence_ref(nv->screen->fence.current, &buf->fence); + nouveau_fence_ref(nv->screen->fence.current, &buf->fence_wr); } diff --git a/src/gallium/drivers/nouveau/nv50/nv50_vbo.c b/src/gallium/drivers/nouveau/nv50/nv50_vbo.c index c6162b5..947c67d 100644 --- a/src/gallium/drivers/nouveau/nv50/nv50_vbo.c +++ b/src/gallium/drivers/nouveau/nv50/nv50_vbo.c @@ -597,6 +597,15 @@ nv50_draw_elements(struct nv50_context *nv50, boolean shorten, assert(nouveau_resource_mapped_by_gpu(nv50->idxbuf.buffer)); + /* This shouldn't have to be here. The going theory is that the buffer + * is being filled in by PGRAPH, and it's not done yet by the time it + * gets submitted to PFIFO, which in turn starts immediately prefetching + * the not-yet-written data. Ideally this wait would only happen on + * pushbuf submit, but it's probably not a big performance difference. + */ + if (buf->fence_wr && !nouveau_fence_signalled(buf->fence_wr)) + nouveau_fence_wait(buf->fence_wr); + while (instance_count--) { BEGIN_NV04(push, NV50_3D(VERTEX_BEGIN_GL), 1); PUSH_DATA (push, prim); -- 1.8.3.2 _______________________________________________ Nouveau mailing list Nouveau@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/nouveau