Git Inbox Mirror of the ffmpeg-devel mailing list - see https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
 help / color / mirror / Atom feed
From: Lynne <dev@lynne.ee>
To: FFmpeg development discussions and patches <ffmpeg-devel@ffmpeg.org>
Subject: [FFmpeg-devel] [PATCH 2/3] nlmeans_vulkan: reduce dispatches by parallelizing the planes
Date: Sat, 7 Oct 2023 17:04:04 +0200 (CEST)
Message-ID: <Ng9eg6t--7-9@lynne.ee> (raw)
In-Reply-To: <Ng9eVFH--3-9@lynne.ee-Ng9eZVQ----9>

[-- Attachment #1: Type: text/plain, Size: 17 bytes --]

Patch attached.


[-- Attachment #2: 0002-nlmeans_vulkan-reduce-dispatches-by-parallelizing-th.patch --]
[-- Type: text/x-diff, Size: 4331 bytes --]

From 863a9977192abf00d131d7b0ed88569210fe0d0c Mon Sep 17 00:00:00 2001
From: Lynne <dev@lynne.ee>
Date: Sat, 16 Sep 2023 01:04:18 +0200
Subject: [PATCH 2/3] nlmeans_vulkan: reduce dispatches by parallelizing the
 planes

---
 libavfilter/vf_nlmeans_vulkan.c | 33 +++++++++++++++++----------------
 1 file changed, 17 insertions(+), 16 deletions(-)

diff --git a/libavfilter/vf_nlmeans_vulkan.c b/libavfilter/vf_nlmeans_vulkan.c
index 5b623eb7a6..9741dd67ac 100644
--- a/libavfilter/vf_nlmeans_vulkan.c
+++ b/libavfilter/vf_nlmeans_vulkan.c
@@ -538,28 +538,29 @@ static av_cold int init_denoise_pipeline(FFVulkanContext *vkctx, FFVkExecPool *e
     GLSLC(0, {                                                                );
     GLSLC(1,     ivec2 size;                                                  );
     GLSLC(1,     const ivec2 pos = ivec2(gl_GlobalInvocationID.xy);           );
+    GLSLC(1,     const uint plane = uint(gl_WorkGroupID.z);                   );
     GLSLC(0,                                                                  );
     GLSLC(1,     float w_sum;                                                 );
     GLSLC(1,     float sum;                                                   );
     GLSLC(1,     vec4 src;                                                    );
     GLSLC(1,     vec4 r;                                                      );
     GLSLC(0,                                                                  );
-
-    for (int i = 0; i < planes; i++) {
-        GLSLF(1, src = texture(input_img[%i], pos);                         ,i);
-        for (int c = 0; c < desc->nb_components; c++) {
-            if (desc->comp[c].plane == i) {
-                int off = desc->comp[c].offset / (FFALIGN(desc->comp[c].depth, 8)/8);
-                GLSLF(1, w_sum = weights_%i[pos.y*ws_stride[%i] + pos.x];               ,c, c);
-                GLSLF(1, sum = sums_%i[pos.y*ws_stride[%i] + pos.x];                    ,c, c);
-                GLSLF(1, r[%i] = (sum + src[%i]*255) / (1.0 + w_sum) / 255;         ,off, off);
-                GLSLC(0,                                                                     );
-            }
-        }
-        GLSLF(1, imageStore(output_img[%i], pos, r);                        ,i);
-        GLSLC(0,                                                              );
+    GLSLC(1,     size = imageSize(output_img[plane]);                         );
+    GLSLC(1,     if (!IS_WITHIN(pos, size))                                   );
+    GLSLC(2,         return;                                                  );
+    GLSLC(0,                                                                  );
+    GLSLC(1,     src = texture(input_img[plane], pos);                        );
+    GLSLC(0,                                                                  );
+    for (int c = 0; c < desc->nb_components; c++) {
+        int off = desc->comp[c].offset / (FFALIGN(desc->comp[c].depth, 8)/8);
+        GLSLF(1, if (plane == %i) {                                              ,desc->comp[c].plane);
+        GLSLF(2,     w_sum = weights_%i[pos.y*ws_stride[%i] + pos.x];                           ,c, c);
+        GLSLF(2,     sum = sums_%i[pos.y*ws_stride[%i] + pos.x];                                ,c, c);
+        GLSLF(2,     r[%i] = (sum + src[%i]*255) / (1.0 + w_sum) / 255;                     ,off, off);
+        GLSLC(1, }                                                                                   );
+        GLSLC(0,                                                                                     );
     }
-
+    GLSLC(1, imageStore(output_img[plane], pos, r);                           );
     GLSLC(0, }                                                                );
 
     RET(spv->compile_shader(spv, vkctx, shd, &spv_data, &spv_len, "main", &spv_opaque));
@@ -716,7 +717,7 @@ static int denoise_pass(NLMeansVulkanContext *s, FFVkExecContext *exec,
     vk->CmdDispatch(exec->buf,
                     FFALIGN(vkctx->output_width,  s->pl_denoise.wg_size[0])/s->pl_denoise.wg_size[0],
                     FFALIGN(vkctx->output_height, s->pl_denoise.wg_size[1])/s->pl_denoise.wg_size[1],
-                    1);
+                    av_pix_fmt_count_planes(s->vkctx.output_format));
 
     return 0;
 }
-- 
2.42.0


[-- Attachment #3: Type: text/plain, Size: 251 bytes --]

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

  parent reply	other threads:[~2023-10-07 15:04 UTC|newest]

Thread overview: 4+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-10-07 15:03 [FFmpeg-devel] [PATCH 1/3] nlmeans_vulkan: fix width/height for chroma plane weights calculation Lynne
     [not found] ` <Ng9eVFH--3-9@lynne.ee-Ng9eZVQ----9>
2023-10-07 15:04   ` Lynne [this message]
2023-10-07 15:07   ` [FFmpeg-devel] [PATCH 3/3] nlmeans_vulkan: parallelize workgroup invocations Lynne
     [not found]   ` <Ng9fYdg--3-9@lynne.ee-Ng9faxr----9>
2023-10-11  3:34     ` Lynne

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=Ng9eg6t--7-9@lynne.ee \
    --to=dev@lynne.ee \
    --cc=ffmpeg-devel@ffmpeg.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

Git Inbox Mirror of the ffmpeg-devel mailing list - see https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

This inbox may be cloned and mirrored by anyone:

	git clone --mirror https://master.gitmailbox.com/ffmpegdev/0 ffmpegdev/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 ffmpegdev ffmpegdev/ https://master.gitmailbox.com/ffmpegdev \
		ffmpegdev@gitmailbox.com
	public-inbox-index ffmpegdev

Example config snippet for mirrors.


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git