diff mbox series

[FFmpeg-devel] hwcontext_vulkan: add the VK_IMAGE_USAGE_SAMPLED_BIT bit to new images

Message ID Ng0j3W6--3-9@lynne.ee
State New
Headers show
Series [FFmpeg-devel] hwcontext_vulkan: add the VK_IMAGE_USAGE_SAMPLED_BIT bit to new images | expand

Checks

Context Check Description
yinshiyou/make_loongarch64 success Make finished
yinshiyou/make_fate_loongarch64 success Make fate finished

Commit Message

Lynne Oct. 5, 2023, 9:26 p.m. UTC
This fixes filtering on Nvidia.

Patch attached.

Comments

Philip Langdale Oct. 5, 2023, 9:27 p.m. UTC | #1
On Thu, 5 Oct 2023 23:26:38 +0200 (CEST)
Lynne <dev@lynne.ee> wrote:

> This fixes filtering on Nvidia.
> 
> Patch attached.
> 

Tested locally. Filtering works and vulkan validation passes.

Ship it!

--phil
Lynne Oct. 5, 2023, 10:03 p.m. UTC | #2
Oct 5, 2023, 23:28 by ffmpeg-devel@ffmpeg.org:

> On Thu, 5 Oct 2023 23:26:38 +0200 (CEST)
> Lynne <dev@lynne.ee> wrote:
>
>> This fixes filtering on Nvidia.
>>
>> Patch attached.
>>
>
> Tested locally. Filtering works and vulkan validation passes.
>
> Ship it!
>

Thanks, tested everywhere I could think of. Works, so pushed.
Now Nvidia users can enjoy filtering!
diff mbox series

Patch

From 2ea0d50049bae792477c53780b3cfd0e3863d914 Mon Sep 17 00:00:00 2001
From: Lynne <dev@lynne.ee>
Date: Thu, 5 Oct 2023 20:25:33 +0200
Subject: [PATCH] hwcontext_vulkan: properly support STORAGE usage for
 mutliplane images

Fixes multiplane support on Nvidia.

Also, remove the ENCODE usage, even if the driver signals it as supported.
Currently, it's not used, and when it is used, it'll be gated behind
two extension checks.
---
 libavfilter/vulkan_filter.c  | 24 +++++++++++++++---------
 libavutil/hwcontext_vulkan.c | 36 ++++++++++++++++++++++++------------
 2 files changed, 39 insertions(+), 21 deletions(-)

diff --git a/libavfilter/vulkan_filter.c b/libavfilter/vulkan_filter.c
index b4d8f952b5..f3f40465be 100644
--- a/libavfilter/vulkan_filter.c
+++ b/libavfilter/vulkan_filter.c
@@ -42,18 +42,23 @@  int ff_vk_filter_init_context(AVFilterContext *avctx, FFVulkanContext *s,
         vk_frames = frames_ctx->hwctx;
         vk_dev = device_ctx->hwctx;
 
-        /* Basic format validation */
+        /* Width and height mismatch */
         if (width != frames_ctx->width ||
-            height != frames_ctx->height ||
-            sw_format != frames_ctx->sw_format ||
-            (vk_frames->tiling != VK_IMAGE_TILING_LINEAR &&
-             vk_frames->tiling != VK_IMAGE_TILING_OPTIMAL) ||
-            !(vk_frames->usage & VK_IMAGE_USAGE_SAMPLED_BIT)) {
+            height != frames_ctx->height)
+            goto skip;
+
+        /* Format mismatch */
+        if (sw_format != frames_ctx->sw_format)
             goto skip;
-        }
 
-        if (vk_frames->usage & VK_IMAGE_USAGE_STORAGE_BIT)
-            goto accept;
+        /* Unusual tiling mismatch. Don't let linear through either. */
+        if (vk_frames->tiling != VK_IMAGE_TILING_OPTIMAL)
+            goto skip;
+
+        /* Usage mismatch */
+        if ((vk_frames->usage & (VK_IMAGE_USAGE_SAMPLED_BIT | VK_IMAGE_USAGE_STORAGE_BIT)) !=
+                                (VK_IMAGE_USAGE_SAMPLED_BIT | VK_IMAGE_USAGE_STORAGE_BIT))
+            goto skip;
 
         s->extensions = ff_vk_extensions_to_mask(vk_dev->enabled_dev_extensions,
                                                  vk_dev->nb_enabled_dev_extensions);
@@ -110,6 +115,7 @@  accept:
         vk_frames = frames_ctx->hwctx;
         vk_frames->tiling = VK_IMAGE_TILING_OPTIMAL;
         vk_frames->usage  = VK_IMAGE_USAGE_SAMPLED_BIT |
+                            VK_IMAGE_USAGE_STORAGE_BIT |
                             VK_IMAGE_USAGE_TRANSFER_SRC_BIT |
                             VK_IMAGE_USAGE_TRANSFER_DST_BIT;
 
diff --git a/libavutil/hwcontext_vulkan.c b/libavutil/hwcontext_vulkan.c
index c676f4fc57..b955ec483e 100644
--- a/libavutil/hwcontext_vulkan.c
+++ b/libavutil/hwcontext_vulkan.c
@@ -282,9 +282,11 @@  FN_MAP_TO(VkImageUsageFlags, usage, VkFormatFeatureFlagBits2, feats)
 
 static int vkfmt_from_pixfmt2(AVHWDeviceContext *dev_ctx, enum AVPixelFormat p,
                               VkImageTiling tiling,
-                              VkFormat fmts[AV_NUM_DATA_POINTERS],
-                              int *nb_images, VkImageAspectFlags *aspect,
-                              VkImageUsageFlags *supported_usage, int disable_multiplane)
+                              VkFormat fmts[AV_NUM_DATA_POINTERS], /* Output format list */
+                              int *nb_images,                      /* Output number of images */
+                              VkImageAspectFlags *aspect,          /* Output aspect */
+                              VkImageUsageFlags *supported_usage,  /* Output supported usage */
+                              int disable_multiplane, int need_storage)
 {
     AVVulkanDeviceContext *hwctx = dev_ctx->hwctx;
     VulkanDevicePriv *priv = dev_ctx->internal->priv;
@@ -301,6 +303,7 @@  static int vkfmt_from_pixfmt2(AVHWDeviceContext *dev_ctx, enum AVPixelFormat p,
             };
             VkFormatFeatureFlagBits2 feats_primary, feats_secondary;
             int basics_primary = 0, basics_secondary = 0;
+            int storage_primary = 0, storage_secondary = 0;
 
             vk->GetPhysicalDeviceFormatProperties2(hwctx->phys_dev,
                                                    vk_formats_list[i].vkf,
@@ -310,6 +313,7 @@  static int vkfmt_from_pixfmt2(AVHWDeviceContext *dev_ctx, enum AVPixelFormat p,
                              prop.formatProperties.linearTilingFeatures :
                              prop.formatProperties.optimalTilingFeatures;
             basics_primary = (feats_primary & basic_flags) == basic_flags;
+            storage_primary = !!(feats_primary & VK_FORMAT_FEATURE_2_STORAGE_IMAGE_BIT);
 
             if (vk_formats_list[i].vkf != vk_formats_list[i].fallback[0]) {
                 vk->GetPhysicalDeviceFormatProperties2(hwctx->phys_dev,
@@ -319,11 +323,15 @@  static int vkfmt_from_pixfmt2(AVHWDeviceContext *dev_ctx, enum AVPixelFormat p,
                                   prop.formatProperties.linearTilingFeatures :
                                   prop.formatProperties.optimalTilingFeatures;
                 basics_secondary = (feats_secondary & basic_flags) == basic_flags;
+                storage_secondary = !!(feats_secondary & VK_FORMAT_FEATURE_2_STORAGE_IMAGE_BIT);
             } else {
                 basics_secondary = basics_primary;
+                storage_secondary = storage_primary;
             }
 
-            if (basics_primary && !(disable_multiplane && vk_formats_list[i].vk_planes > 1)) {
+            if (basics_primary &&
+                !(disable_multiplane && vk_formats_list[i].vk_planes > 1) &&
+                (!need_storage || (need_storage && (storage_primary | storage_secondary)))) {
                 if (fmts)
                     fmts[0] = vk_formats_list[i].vkf;
                 if (nb_images)
@@ -331,9 +339,12 @@  static int vkfmt_from_pixfmt2(AVHWDeviceContext *dev_ctx, enum AVPixelFormat p,
                 if (aspect)
                     *aspect = vk_formats_list[i].aspect;
                 if (supported_usage)
-                    *supported_usage = map_feats_to_usage(feats_primary);
+                    *supported_usage = map_feats_to_usage(feats_primary) |
+                                       ((need_storage && (storage_primary | storage_secondary)) ?
+                                        VK_IMAGE_USAGE_STORAGE_BIT : 0);
                 return 0;
-            } else if (basics_secondary) {
+            } else if (basics_secondary &&
+                       (!need_storage || (need_storage && storage_secondary))) {
                 if (fmts) {
                     for (int j = 0; j < vk_formats_list[i].nb_images_fallback; j++)
                         fmts[j] = vk_formats_list[i].fallback[j];
@@ -1640,7 +1651,7 @@  static int vulkan_frames_get_constraints(AVHWDeviceContext *ctx,
         count += vkfmt_from_pixfmt2(ctx, vk_formats_list[i].pixfmt,
                                     p->use_linear_images ? VK_IMAGE_TILING_LINEAR :
                                                            VK_IMAGE_TILING_OPTIMAL,
-                                    NULL, NULL, NULL, NULL, 0) >= 0;
+                                    NULL, NULL, NULL, NULL, 0, 0) >= 0;
     }
 
 #if CONFIG_CUDA
@@ -1658,7 +1669,7 @@  static int vulkan_frames_get_constraints(AVHWDeviceContext *ctx,
         if (vkfmt_from_pixfmt2(ctx, vk_formats_list[i].pixfmt,
                                p->use_linear_images ? VK_IMAGE_TILING_LINEAR :
                                                       VK_IMAGE_TILING_OPTIMAL,
-                               NULL, NULL, NULL, NULL, 0) >= 0) {
+                               NULL, NULL, NULL, NULL, 0, 0) >= 0) {
             constraints->valid_sw_formats[count++] = vk_formats_list[i].pixfmt;
         }
     }
@@ -2294,7 +2305,8 @@  static int vulkan_frames_init(AVHWFramesContext *hwfc)
         /* Check if the sw_format itself is supported */
         err = vkfmt_from_pixfmt2(hwfc->device_ctx, hwfc->sw_format,
                                  hwctx->tiling, NULL,
-                                 NULL, NULL, &supported_usage, 0);
+                                 NULL, NULL, &supported_usage, 0,
+                                 hwctx->usage & VK_IMAGE_USAGE_STORAGE_BIT);
         if (err < 0) {
             av_log(hwfc, AV_LOG_ERROR, "Unsupported sw format: %s!\n",
                    av_get_pix_fmt_name(hwfc->sw_format));
@@ -2304,7 +2316,8 @@  static int vulkan_frames_init(AVHWFramesContext *hwfc)
         err = vkfmt_from_pixfmt2(hwfc->device_ctx, hwfc->sw_format,
                                  hwctx->tiling, hwctx->format, NULL,
                                  NULL, &supported_usage,
-                                 disable_multiplane);
+                                 disable_multiplane,
+                                 hwctx->usage & VK_IMAGE_USAGE_STORAGE_BIT);
         if (err < 0)
             return err;
     }
@@ -2314,8 +2327,7 @@  static int vulkan_frames_init(AVHWFramesContext *hwfc)
         hwctx->usage = supported_usage & (VK_BUFFER_USAGE_TRANSFER_DST_BIT |
                                           VK_BUFFER_USAGE_TRANSFER_SRC_BIT |
                                           VK_IMAGE_USAGE_STORAGE_BIT       |
-                                          VK_IMAGE_USAGE_SAMPLED_BIT       |
-                                          VK_IMAGE_USAGE_VIDEO_ENCODE_SRC_BIT_KHR);
+                                          VK_IMAGE_USAGE_SAMPLED_BIT);
     }
 
     /* Image creation flags.
-- 
2.42.0