r300g: offload the CS ioctl to another thread

This is a multi-threading optimization which hides the kernel overhead
behind a thread. It improves performance in CPU-limited apps by 2-15%.
Of course you must have at least 2 cores for it to make any difference.

It can be disabled with:

export RADEON_THREAD=0

1 files changed, 8 insertions, 0 deletions

diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_bo.c b/src/gallium/winsys/radeon/drm/radeon_drm_bo.c
index f3c4002883..afb8131acb 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_bo.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_bo.c
@@ -116,6 +116,10 @@ static void radeon_bo_wait(struct r300_winsys_bo *_buf)
     struct radeon_bo *bo = get_radeon_bo(pb_buffer(_buf));
     struct drm_radeon_gem_wait_idle args = {};
 
+    while (p_atomic_read(&bo->num_active_ioctls)) {
+        sched_yield();
+    }
+
     args.handle = bo->handle;
     while (drmCommandWriteRead(bo->rws->fd, DRM_RADEON_GEM_WAIT_IDLE,
                                &args, sizeof(args)) == -EBUSY);
@@ -126,6 +130,10 @@ static boolean radeon_bo_is_busy(struct r300_winsys_bo *_buf)
     struct radeon_bo *bo = get_radeon_bo(pb_buffer(_buf));
     struct drm_radeon_gem_busy args = {};
 
+    if (p_atomic_read(&bo->num_active_ioctls)) {
+        return TRUE;
+    }
+
     args.handle = bo->handle;
     return drmCommandWriteRead(bo->rws->fd, DRM_RADEON_GEM_BUSY,
                                &args, sizeof(args)) != 0;