gallium/util: import the multithreaded job queue from amdgpu winsys (v2)

v2: rename the event to util_queue_fence

Reviewed-by: Nicolai Hähnle <[email protected]>

4 files changed, 18 insertions, 83 deletions

diff --git a/src/gallium/winsys/amdgpu/drm/amdgpu_cs.c b/src/gallium/winsys/amdgpu/drm/amdgpu_cs.c
index fefa5d6db53..4a7302ab2cb 100644
--- a/src/gallium/winsys/amdgpu/drm/amdgpu_cs.c
+++ b/src/gallium/winsys/amdgpu/drm/amdgpu_cs.c
@@ -605,7 +605,7 @@ amdgpu_cs_create(struct radeon_winsys_ctx *rwctx,
       return NULL;
    }
 
-   pipe_semaphore_init(&cs->flush_completed, 1);
+   util_queue_fence_init(&cs->flush_completed);
 
    cs->ctx = ctx;
    cs->flush_cs = flush;
@@ -872,8 +872,9 @@ static void amdgpu_add_fence_dependencies(struct amdgpu_cs *acs)
    }
 }
 
-void amdgpu_cs_submit_ib(struct amdgpu_cs *acs)
+void amdgpu_cs_submit_ib(void *job)
 {
+   struct amdgpu_cs *acs = (struct amdgpu_cs*)job;
    struct amdgpu_winsys *ws = acs->ctx->ws;
    struct amdgpu_cs_context *cs = acs->cst;
    int i, r;
@@ -957,14 +958,11 @@ cleanup:
 void amdgpu_cs_sync_flush(struct radeon_winsys_cs *rcs)
 {
    struct amdgpu_cs *cs = amdgpu_cs(rcs);
+   struct amdgpu_winsys *ws = cs->ctx->ws;
 
    /* Wait for any pending ioctl of this CS to complete. */
-   if (cs->ctx->ws->thread) {
-      /* wait and set the semaphore to "busy" */
-      pipe_semaphore_wait(&cs->flush_completed);
-      /* set the semaphore to "idle" */
-      pipe_semaphore_signal(&cs->flush_completed);
-   }
+   if (util_queue_is_initialized(&ws->cs_queue))
+      util_queue_job_wait(&cs->flush_completed);
 }
 
 DEBUG_GET_ONCE_BOOL_OPTION(noop, "RADEON_NOOP", FALSE)
@@ -1052,10 +1050,9 @@ static void amdgpu_cs_flush(struct radeon_winsys_cs *rcs,
       cs->cst = cur;
 
       /* Submit. */
-      if (ws->thread && (flags & RADEON_FLUSH_ASYNC)) {
-         /* Set the semaphore to "busy". */
-         pipe_semaphore_wait(&cs->flush_completed);
-         amdgpu_ws_queue_cs(ws, cs);
+      if ((flags & RADEON_FLUSH_ASYNC) &&
+          util_queue_is_initialized(&ws->cs_queue)) {
+         util_queue_add_job(&ws->cs_queue, cs, &cs->flush_completed);
       } else {
          amdgpu_cs_submit_ib(cs);
       }
@@ -1077,7 +1074,7 @@ static void amdgpu_cs_destroy(struct radeon_winsys_cs *rcs)
    struct amdgpu_cs *cs = amdgpu_cs(rcs);
 
    amdgpu_cs_sync_flush(rcs);
-   pipe_semaphore_destroy(&cs->flush_completed);
+   util_queue_fence_destroy(&cs->flush_completed);
    p_atomic_dec(&cs->ctx->ws->num_cs);
    pb_reference(&cs->main.big_ib_buffer, NULL);
    FREE(cs->main.base.prev);
diff --git a/src/gallium/winsys/amdgpu/drm/amdgpu_cs.h b/src/gallium/winsys/amdgpu/drm/amdgpu_cs.h
index cc1516cfbf2..354e403fa36 100644
--- a/src/gallium/winsys/amdgpu/drm/amdgpu_cs.h
+++ b/src/gallium/winsys/amdgpu/drm/amdgpu_cs.h
@@ -111,7 +111,7 @@ struct amdgpu_cs {
    void (*flush_cs)(void *ctx, unsigned flags, struct pipe_fence_handle **fence);
    void *flush_data;
 
-   pipe_semaphore flush_completed;
+   struct util_queue_fence flush_completed;
 };
 
 struct amdgpu_fence {
@@ -218,6 +218,6 @@ bool amdgpu_fence_wait(struct pipe_fence_handle *fence, uint64_t timeout,
                        bool absolute);
 void amdgpu_cs_sync_flush(struct radeon_winsys_cs *rcs);
 void amdgpu_cs_init_functions(struct amdgpu_winsys *ws);
-void amdgpu_cs_submit_ib(struct amdgpu_cs *cs);
+void amdgpu_cs_submit_ib(void *job);
 
 #endif
diff --git a/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.c b/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.c
index 701622184c8..7ef35298aa8 100644
--- a/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.c
+++ b/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.c
@@ -308,14 +308,9 @@ static void amdgpu_winsys_destroy(struct radeon_winsys *rws)
 {
    struct amdgpu_winsys *ws = (struct amdgpu_winsys*)rws;
 
-   if (ws->thread) {
-      ws->kill_thread = 1;
-      pipe_semaphore_signal(&ws->cs_queued);
-      pipe_thread_wait(ws->thread);
-   }
-   pipe_semaphore_destroy(&ws->cs_queue_has_space);
-   pipe_semaphore_destroy(&ws->cs_queued);
-   pipe_mutex_destroy(ws->cs_queue_lock);
+   if (util_queue_is_initialized(&ws->cs_queue))
+      util_queue_destroy(&ws->cs_queue);
+
    pipe_mutex_destroy(ws->bo_fence_lock);
    pb_cache_deinit(&ws->bo_cache);
    pipe_mutex_destroy(ws->global_bo_list_lock);
@@ -400,53 +395,7 @@ static int compare_dev(void *key1, void *key2)
    return key1 != key2;
 }
 
-void amdgpu_ws_queue_cs(struct amdgpu_winsys *ws, struct amdgpu_cs *cs)
-{
-   pipe_semaphore_wait(&ws->cs_queue_has_space);
-
-   pipe_mutex_lock(ws->cs_queue_lock);
-   assert(ws->num_enqueued_cs < ARRAY_SIZE(ws->cs_queue));
-   ws->cs_queue[ws->num_enqueued_cs++] = cs;
-   pipe_mutex_unlock(ws->cs_queue_lock);
-   pipe_semaphore_signal(&ws->cs_queued);
-}
-
-static PIPE_THREAD_ROUTINE(amdgpu_cs_thread_func, param)
-{
-   struct amdgpu_winsys *ws = (struct amdgpu_winsys *)param;
-   struct amdgpu_cs *cs;
-   unsigned i;
-
-   while (1) {
-      pipe_semaphore_wait(&ws->cs_queued);
-      if (ws->kill_thread)
-         break;
-
-      pipe_mutex_lock(ws->cs_queue_lock);
-      cs = ws->cs_queue[0];
-      for (i = 1; i < ws->num_enqueued_cs; i++)
-         ws->cs_queue[i - 1] = ws->cs_queue[i];
-      ws->cs_queue[--ws->num_enqueued_cs] = NULL;
-      pipe_mutex_unlock(ws->cs_queue_lock);
-
-      pipe_semaphore_signal(&ws->cs_queue_has_space);
-
-      if (cs) {
-         amdgpu_cs_submit_ib(cs);
-         pipe_semaphore_signal(&cs->flush_completed);
-      }
-   }
-   pipe_mutex_lock(ws->cs_queue_lock);
-   for (i = 0; i < ws->num_enqueued_cs; i++) {
-      pipe_semaphore_signal(&ws->cs_queue[i]->flush_completed);
-      ws->cs_queue[i] = NULL;
-   }
-   pipe_mutex_unlock(ws->cs_queue_lock);
-   return 0;
-}
-
 DEBUG_GET_ONCE_BOOL_OPTION(thread, "RADEON_THREAD", TRUE)
-static PIPE_THREAD_ROUTINE(amdgpu_cs_thread_func, param);
 
 static bool amdgpu_winsys_unref(struct radeon_winsys *rws)
 {
@@ -541,14 +490,10 @@ amdgpu_winsys_create(int fd, radeon_screen_create_t screen_create)
 
    LIST_INITHEAD(&ws->global_bo_list);
    pipe_mutex_init(ws->global_bo_list_lock);
-   pipe_mutex_init(ws->cs_queue_lock);
    pipe_mutex_init(ws->bo_fence_lock);
 
-   pipe_semaphore_init(&ws->cs_queue_has_space, ARRAY_SIZE(ws->cs_queue));
-   pipe_semaphore_init(&ws->cs_queued, 0);
-
    if (sysconf(_SC_NPROCESSORS_ONLN) > 1 && debug_get_option_thread())
-      ws->thread = pipe_thread_create(amdgpu_cs_thread_func, ws);
+      util_queue_init(&ws->cs_queue, amdgpu_cs_submit_ib);
 
    /* Create the screen at the end. The winsys must be initialized
     * completely.
diff --git a/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.h b/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.h
index d6734f7bfbc..b13a17e119c 100644
--- a/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.h
+++ b/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.h
@@ -35,7 +35,7 @@
 #include "pipebuffer/pb_cache.h"
 #include "gallium/drivers/radeon/radeon_winsys.h"
 #include "addrlib/addrinterface.h"
-#include "os/os_thread.h"
+#include "util/u_queue.h"
 #include <amdgpu.h>
 
 struct amdgpu_cs;
@@ -59,13 +59,7 @@ struct amdgpu_winsys {
    struct radeon_info info;
 
    /* multithreaded IB submission */
-   pipe_mutex cs_queue_lock;
-   pipe_semaphore cs_queue_has_space;
-   pipe_semaphore cs_queued;
-   pipe_thread thread;
-   int kill_thread;
-   int num_enqueued_cs;
-   struct amdgpu_cs *cs_queue[8];
+   struct util_queue cs_queue;
 
    struct amdgpu_gpu_info amdinfo;
    ADDR_HANDLE addrlib;
@@ -84,7 +78,6 @@ amdgpu_winsys(struct radeon_winsys *base)
    return (struct amdgpu_winsys*)base;
 }
 
-void amdgpu_ws_queue_cs(struct amdgpu_winsys *ws, struct amdgpu_cs *cs);
 void amdgpu_surface_init_functions(struct amdgpu_winsys *ws);
 ADDR_HANDLE amdgpu_addr_create(struct amdgpu_winsys *ws);