radv: optimize creating signaled syncobj with amdgpu_cs_create_syncobj2()

This creates a syncobj and sets it as signaled with one ioctl
instead of two.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5921>

diff --git a/src/amd/vulkan/radv_device.c b/src/amd/vulkan/radv_device.c
index 5a1a67288d3d9e0474fb4016a18e7b6cc30916d1..00f87ebdf52a31874f0c7c9278e59709b5720068 100644 (file)
--- a/src/amd/vulkan/radv_device.c
+++ b/src/amd/vulkan/radv_device.c
@@ -5204,14 +5204,16 @@ VkResult radv_CreateFence(
        fence->fence_wsi = NULL;
        fence->temp_syncobj = 0;
        if (device->always_use_syncobj || handleTypes) {
-               int ret = device->ws->create_syncobj(device->ws, &fence->syncobj);
+               bool create_signaled = false;
+               if (pCreateInfo->flags & VK_FENCE_CREATE_SIGNALED_BIT)
+                       create_signaled = true;
+
+               int ret = device->ws->create_syncobj(device->ws, create_signaled,
+                                                    &fence->syncobj);
                if (ret) {
                        radv_destroy_fence(device, pAllocator, fence);
                        return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
                }
-               if (pCreateInfo->flags & VK_FENCE_CREATE_SIGNALED_BIT) {
-                       device->ws->signal_syncobj(device->ws, fence->syncobj);
-               }
                fence->fence = NULL;
        } else {
                fence->fence = device->ws->create_fence();
@@ -5539,7 +5541,7 @@ radv_timeline_add_point_locked(struct radv_device *device,
 
        if (list_is_empty(&timeline->free_points)) {
                ret = malloc(sizeof(struct radv_timeline_point));
-               device->ws->create_syncobj(device->ws, &ret->syncobj);
+               device->ws->create_syncobj(device->ws, false, &ret->syncobj);
        } else {
                ret = list_first_entry(&timeline->free_points, struct radv_timeline_point, list);
                list_del(&ret->list);
@@ -5680,7 +5682,8 @@ VkResult radv_CreateSemaphore(
                sem->permanent.kind = RADV_SEMAPHORE_TIMELINE;
        } else if (device->always_use_syncobj || handleTypes) {
                assert (device->physical_device->rad_info.has_syncobj);
-               int ret = device->ws->create_syncobj(device->ws, &sem->permanent.syncobj);
+               int ret = device->ws->create_syncobj(device->ws, false,
+                                                    &sem->permanent.syncobj);
                if (ret) {
                        radv_destroy_semaphore(device, pAllocator, sem);
                        return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
@@ -7062,23 +7065,26 @@ static VkResult radv_import_sync_fd(struct radv_device *device,
         * leave a syncobj in an undetermined state in the fence. */
        uint32_t syncobj_handle =  *syncobj;
        if (!syncobj_handle) {
-               int ret = device->ws->create_syncobj(device->ws, &syncobj_handle);
+               bool create_signaled = fd == -1 ? true : false;
+
+               int ret = device->ws->create_syncobj(device->ws, create_signaled,
+                                                    &syncobj_handle);
                if (ret) {
                        return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
                }
+       } else {
+               if (fd == -1)
+                       device->ws->signal_syncobj(device->ws, syncobj_handle);
        }
 
-       if (fd == -1) {
-               device->ws->signal_syncobj(device->ws, syncobj_handle);
-       } else {
+       if (fd != -1) {
                int ret = device->ws->import_syncobj_from_sync_file(device->ws, syncobj_handle, fd);
-       if (ret != 0)
-               return vk_error(device->instance, VK_ERROR_INVALID_EXTERNAL_HANDLE);
+               if (ret)
+                       return vk_error(device->instance, VK_ERROR_INVALID_EXTERNAL_HANDLE);
+               close(fd);
        }
 
        *syncobj = syncobj_handle;
-       if (fd != -1)
-               close(fd);
 
        return VK_SUCCESS;
 }

diff --git a/src/amd/vulkan/radv_radeon_winsys.h b/src/amd/vulkan/radv_radeon_winsys.h
index 802dbdb1783417914eb377fa9228bac46a58fef4..2cf68dc8edb75ef5bcaefdaa701de910140b71f8 100644 (file)
--- a/src/amd/vulkan/radv_radeon_winsys.h
+++ b/src/amd/vulkan/radv_radeon_winsys.h
@@ -320,7 +320,8 @@ struct radeon_winsys {
        void (*destroy_sem)(struct radeon_winsys_sem *sem);
 
        /* new shareable sync objects */
-       int (*create_syncobj)(struct radeon_winsys *ws, uint32_t *handle);
+       int (*create_syncobj)(struct radeon_winsys *ws, bool create_signaled,
+                             uint32_t *handle);
        void (*destroy_syncobj)(struct radeon_winsys *ws, uint32_t handle);
 
        void (*reset_syncobj)(struct radeon_winsys *ws, uint32_t handle);

diff --git a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c
index 74ba56ccb46f3669d04f1f0e3e6e027889e27aa3..daebdf9cc5fb3a8d46430e1e6a2056b6674040b6 100644 (file)
--- a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c
+++ b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c
@@ -1600,10 +1600,16 @@ error_out:
 }
 
 static int radv_amdgpu_create_syncobj(struct radeon_winsys *_ws,
+                                     bool create_signaled,
                                      uint32_t *handle)
 {
        struct radv_amdgpu_winsys *ws = radv_amdgpu_winsys(_ws);
-       return amdgpu_cs_create_syncobj(ws->dev, handle);
+       uint32_t flags = 0;
+
+       if (create_signaled)
+               flags |= DRM_SYNCOBJ_CREATE_SIGNALED;
+
+       return amdgpu_cs_create_syncobj2(ws->dev, flags, handle);
 }
 
 static void radv_amdgpu_destroy_syncobj(struct radeon_winsys *_ws,