[FFmpeg-devel] [PATCH v2 2/3] hwcontext_vulkan: minimize queue allocation on NVIDIA

[Lynne]

On NVIDIA, there's a global maximum limit of approximately 112 queues,
which means it takes ONLY 7 total programs using the maximum amount of
queues to cause the driver to error out/*segfault* during initialization.

Also, each queue takes about 30ms to allocate, which quickly adds up.

This reduces the queues allocate to the minimum that we would be happy
with. Its not worth limiting decode/encode queues as they're generally
not a lot, and do help.
---
 libavutil/hwcontext_vulkan.c | 24 +++++++++++++++++++++++-
 1 file changed, 23 insertions(+), 1 deletion(-)

diff --git a/libavutil/hwcontext_vulkan.c b/libavutil/hwcontext_vulkan.c
index dc9d49ccb4..243d68dc44 100644
--- a/libavutil/hwcontext_vulkan.c
+++ b/libavutil/hwcontext_vulkan.c
@@ -1418,6 +1418,13 @@ static int setup_queue_families(AVHWDeviceContext *ctx, VkDeviceCreateInfo *cd)
     VulkanDevicePriv *p = ctx->hwctx;
     AVVulkanDeviceContext *hwctx = &p->p;
     FFVulkanFunctions *vk = &p->vkctx.vkfn;
+    VkPhysicalDeviceDriverProperties dprops = {
+        .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DRIVER_PROPERTIES,
+    };
+    VkPhysicalDeviceProperties2 props2 = {
+        .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2,
+        .pNext = &dprops,
+    };
 
     VkQueueFamilyProperties2 *qf = NULL;
     VkQueueFamilyVideoPropertiesKHR *qf_vid = NULL;
@@ -1471,7 +1478,14 @@ static int setup_queue_families(AVHWDeviceContext *ctx, VkDeviceCreateInfo *cd)
 
     hwctx->nb_qf = 0;
 
-    /* Pick each queue family to use */
+    /* NVIDIA's proprietary drivers have stupid limits, where each queue
+     * you allocate takes tens of milliseconds, and the more queues you
+     * allocate, the less you'll have left before initializing a device
+     * simply fails (112 seems to be the max). GLOBALLY.
+     * Detect this, and minimize using queues as much as possible. */
+    vk->GetPhysicalDeviceProperties2(hwctx->phys_dev, &props2);
+
+    /* Pick each queue family to use. */
 #define PICK_QF(type, vid_op)                                            \
     do {                                                                 \
         uint32_t i;                                                      \
@@ -1495,6 +1509,14 @@ static int setup_queue_families(AVHWDeviceContext *ctx, VkDeviceCreateInfo *cd)
         if (i == hwctx->nb_qf) {                                         \
             hwctx->qf[i].idx = idx;                                      \
             hwctx->qf[i].num = qf[idx].queueFamilyProperties.queueCount; \
+            if (dprops.driverID == VK_DRIVER_ID_NVIDIA_PROPRIETARY) {    \
+                if (type == VK_QUEUE_GRAPHICS_BIT)                       \
+                    hwctx->qf[i].num = FFMIN(hwctx->qf[i].num, 1);       \
+                if (type == VK_QUEUE_COMPUTE_BIT)                        \
+                    hwctx->qf[i].num = FFMIN(hwctx->qf[i].num, 4);       \
+                if (type == VK_QUEUE_TRANSFER_BIT)                       \
+                    hwctx->qf[i].num = FFMIN(hwctx->qf[i].num, 2);       \
+            }                                                            \
             hwctx->qf[i].flags = type;                                   \
             hwctx->qf[i].video_caps = vid_op;                            \
             hwctx->nb_qf++;                                              \
-- 
2.49.0.395.g12beb8f557c