zink: move queue submit thread to screen

this needs to be a screen-based queue in order to serialize properly
with multiple contexts

Acked-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11437>

5 files changed, 12 insertions, 13 deletions

diff --git a/src/gallium/drivers/zink/zink_batch.c b/src/gallium/drivers/zink/zink_batch.c
index 3fb4cc6eabd..a775b3dfb02 100644
--- a/src/gallium/drivers/zink/zink_batch.c
+++ b/src/gallium/drivers/zink/zink_batch.c
@@ -300,9 +300,6 @@ zink_start_batch(struct zink_context *ctx, struct zink_batch *batch)
    if (ctx->last_fence) {
       struct zink_batch_state *last_state = zink_batch_state(ctx->last_fence);
       batch->last_batch_usage = &last_state->usage;
-   } else {
-      if (zink_screen(ctx->base.screen)->threaded)
-         util_queue_init(&batch->flush_queue, "zfq", 8, 1, UTIL_QUEUE_INIT_RESIZE_IF_FULL, NULL);
    }
    if (!ctx->queries_disabled)
       zink_resume_queries(ctx, batch);
@@ -546,9 +543,9 @@ zink_end_batch(struct zink_context *ctx, struct zink_batch *batch)
    if (screen->device_lost)
       return;
 
-   if (util_queue_is_initialized(&batch->flush_queue)) {
+   if (screen->threaded) {
       batch->state->queue = screen->thread_queue;
-      util_queue_add_job(&batch->flush_queue, batch->state, &batch->state->flush_completed,
+      util_queue_add_job(&screen->flush_queue, batch->state, &batch->state->flush_completed,
                          submit_queue, post_submit, 0);
    } else {
       batch->state->queue = screen->queue;
diff --git a/src/gallium/drivers/zink/zink_batch.h b/src/gallium/drivers/zink/zink_batch.h
index 04827bfb7e5..bf04599aadf 100644
--- a/src/gallium/drivers/zink/zink_batch.h
+++ b/src/gallium/drivers/zink/zink_batch.h
@@ -101,7 +101,6 @@ struct zink_batch {
    struct zink_batch_state *state;
 
    struct zink_batch_usage *last_batch_usage;
-   struct util_queue flush_queue; //TODO: move to wsi
 
    bool has_work;
    bool in_rp; //renderpass is currently active
diff --git a/src/gallium/drivers/zink/zink_context.c b/src/gallium/drivers/zink/zink_context.c
index ca912f1f566..31933128976 100644
--- a/src/gallium/drivers/zink/zink_context.c
+++ b/src/gallium/drivers/zink/zink_context.c
@@ -90,9 +90,6 @@ zink_context_destroy(struct pipe_context *pctx)
    zink_surface_reference(screen, (struct zink_surface**)&ctx->dummy_surface, NULL);
    zink_buffer_view_reference(screen, &ctx->dummy_bufferview, NULL);
 
-   if (ctx->tc)
-      util_queue_destroy(&ctx->batch.flush_queue);
-
    simple_mtx_destroy(&ctx->batch_mtx);
    zink_clear_batch_state(ctx, ctx->batch.state);
    zink_batch_state_reference(screen, &ctx->batch.state, NULL);
@@ -187,7 +184,7 @@ zink_set_context_param(struct pipe_context *pctx, enum pipe_context_param param,
 
    switch (param) {
    case PIPE_CONTEXT_PARAM_PIN_THREADS_TO_L3_CACHE:
-      util_set_thread_affinity(ctx->batch.flush_queue.threads[0],
+      util_set_thread_affinity(zink_screen(ctx->base.screen)->flush_queue.threads[0],
                                util_get_cpu_caps()->L3_affinity_mask[value],
                                NULL, util_get_cpu_caps()->num_cpu_mask_bits);
       break;
@@ -1688,7 +1685,7 @@ zink_end_render_pass(struct zink_context *ctx, struct zink_batch *batch)
 static void
 sync_flush(struct zink_context *ctx, struct zink_batch_state *bs)
 {
-   if (util_queue_is_initialized(&ctx->batch.flush_queue))
+   if (zink_screen(ctx->base.screen)->threaded)
       util_queue_fence_wait(&bs->flush_completed);
 }
 
@@ -2560,7 +2557,7 @@ zink_check_batch_completion(struct zink_context *ctx, uint32_t batch_id)
    }
    simple_mtx_unlock(&ctx->batch_mtx);
    assert(fence);
-   if (util_queue_is_initialized(&ctx->batch.flush_queue) &&
+   if (zink_screen(ctx->base.screen)->threaded &&
        !util_queue_fence_is_signalled(&zink_batch_state(fence)->flush_completed))
       return false;
    return zink_vkfence_wait(zink_screen(ctx->base.screen), fence, 0);
@@ -3177,7 +3174,7 @@ zink_resource_commit(struct pipe_context *pctx, struct pipe_resource *pres, unsi
    mem_bind.memoryOffset = box->x;
    mem_bind.flags = 0;
    sparse_bind.pBinds = &mem_bind;
-   VkQueue queue = util_queue_is_initialized(&ctx->batch.flush_queue) ? screen->thread_queue : screen->queue;
+   VkQueue queue = screen->threaded ? screen->thread_queue : screen->queue;
 
    VkResult ret = vkQueueBindSparse(queue, 1, &sparse, VK_NULL_HANDLE);
    if (!zink_screen_handle_vkresult(screen, ret)) {
diff --git a/src/gallium/drivers/zink/zink_screen.c b/src/gallium/drivers/zink/zink_screen.c
index b3694cf22b7..1f8b03ff381 100644
--- a/src/gallium/drivers/zink/zink_screen.c
+++ b/src/gallium/drivers/zink/zink_screen.c
@@ -1020,6 +1020,8 @@ zink_destroy_screen(struct pipe_screen *pscreen)
    if (screen->prev_sem)
       vkDestroySemaphore(screen->dev, screen->prev_sem, NULL);
 
+   if (screen->threaded)
+      util_queue_destroy(&screen->flush_queue);
 
    vkDestroyDevice(screen->dev, NULL);
    vkDestroyInstance(screen->instance, NULL);
@@ -1596,6 +1598,8 @@ zink_internal_create_screen(const struct pipe_screen_config *config)
 
    util_cpu_detect();
    screen->threaded = util_get_cpu_caps()->nr_cpus > 1 && debug_get_bool_option("GALLIUM_THREAD", util_get_cpu_caps()->nr_cpus > 1);
+   if (screen->threaded)
+      util_queue_init(&screen->flush_queue, "zfq", 8, 1, UTIL_QUEUE_INIT_RESIZE_IF_FULL, NULL);
 
    zink_debug = debug_get_option_zink_debug();
    screen->descriptor_mode = debug_get_option_zink_descriptor_mode();
diff --git a/src/gallium/drivers/zink/zink_screen.h b/src/gallium/drivers/zink/zink_screen.h
index 193380c8312..0d861d5122c 100644
--- a/src/gallium/drivers/zink/zink_screen.h
+++ b/src/gallium/drivers/zink/zink_screen.h
@@ -35,6 +35,7 @@
 #include "util/disk_cache.h"
 #include "util/log.h"
 #include "util/simple_mtx.h"
+#include "util/u_queue.h"
 #include "util/u_live_shader_cache.h"
 
 #include <vulkan/vulkan.h>
@@ -72,6 +73,7 @@ struct zink_screen {
    uint32_t last_finished; //this is racy but ultimately doesn't matter
    VkSemaphore sem;
    VkSemaphore prev_sem;
+   struct util_queue flush_queue;
 
    bool device_lost;
    struct sw_winsys *winsys;