diff mbox series

[FFmpeg-devel,03/10] libavfilter/vulkan: Fix the way to use sem

Message ID 20210831014338.134086-3-wenbin.chen@intel.com
State New
Headers show
Series [FFmpeg-devel,01/10] libavfilter/vulkan: Fix problem when device have queue_count greater than 1
Related show

Checks

Context Check Description
andriy/make_x86 success Make finished
andriy/make_fate_x86 success Make fate finished
andriy/make_ppc success Make finished
andriy/make_fate_ppc success Make fate finished

Commit Message

Wenbin Chen Aug. 31, 2021, 1:43 a.m. UTC
From: "Chen,Wenbin" <wenbin.chen@intel.com>

We chould set waitSem and signalSem differently. Current ffmpeg-vulkan
uses the same sem to set waitSem and signalSem and it doesn't work on
latest intel-vulkan-driver. The commit:
a193060221c4df123e26a562949cae5df3e73cde on mesa causes this problem.
This commit add code to resets the signalSem. This will reset waitSem
too on current ffmpeg-vulkan. Now set waitSem and signalSem separetely.

Now the following command can run on the latest mesa on intel platform:
ffmpeg -v verbose -init_hw_device vulkan=vul:0,linear_images=1 -filter_hw_device vul
-i input1080p.264 -vf "hwupload=extra_hw_frames=16,scale_vulkan=1920:1080,
hwdownload,format=yuv420p" -f rawvideo output.yuv

Signed-off-by: Wenbin Chen <wenbin.chen@intel.com>
---
 libavfilter/vf_avgblur_vulkan.c   |  4 +--
 libavfilter/vf_chromaber_vulkan.c |  4 +--
 libavfilter/vf_overlay_vulkan.c   |  6 ++--
 libavfilter/vf_scale_vulkan.c     |  4 +--
 libavfilter/vulkan.c              | 55 +++++++++++++++++--------------
 libavfilter/vulkan.h              |  3 +-
 libavutil/hwcontext_vulkan.c      | 14 ++++----
 7 files changed, 50 insertions(+), 40 deletions(-)
diff mbox series

Patch

diff --git a/libavfilter/vf_avgblur_vulkan.c b/libavfilter/vf_avgblur_vulkan.c
index 5ae487fc8c..d2104c191e 100644
--- a/libavfilter/vf_avgblur_vulkan.c
+++ b/libavfilter/vf_avgblur_vulkan.c
@@ -304,8 +304,8 @@  static int process_frames(AVFilterContext *avctx, AVFrame *out_f, AVFrame *tmp_f
     vkCmdDispatch(cmd_buf, s->vkctx.output_width,
                   FFALIGN(s->vkctx.output_height, CGS)/CGS, 1);
 
-    ff_vk_add_exec_dep(avctx, s->exec, in_f, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT);
-    ff_vk_add_exec_dep(avctx, s->exec, out_f, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT);
+    ff_vk_add_exec_dep(avctx, s->exec, in_f, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, 1);
+    ff_vk_add_exec_dep(avctx, s->exec, out_f, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, 0);
 
     err = ff_vk_submit_exec_queue(avctx, s->exec);
     if (err)
diff --git a/libavfilter/vf_chromaber_vulkan.c b/libavfilter/vf_chromaber_vulkan.c
index 96fdd7bd9c..fe66a31cea 100644
--- a/libavfilter/vf_chromaber_vulkan.c
+++ b/libavfilter/vf_chromaber_vulkan.c
@@ -249,8 +249,8 @@  static int process_frames(AVFilterContext *avctx, AVFrame *out_f, AVFrame *in_f)
                   FFALIGN(s->vkctx.output_width,  CGROUPS[0])/CGROUPS[0],
                   FFALIGN(s->vkctx.output_height, CGROUPS[1])/CGROUPS[1], 1);
 
-    ff_vk_add_exec_dep(avctx, s->exec, in_f, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT);
-    ff_vk_add_exec_dep(avctx, s->exec, out_f, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT);
+    ff_vk_add_exec_dep(avctx, s->exec, in_f, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, 1);
+    ff_vk_add_exec_dep(avctx, s->exec, out_f, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, 0);
 
     err = ff_vk_submit_exec_queue(avctx, s->exec);
     if (err)
diff --git a/libavfilter/vf_overlay_vulkan.c b/libavfilter/vf_overlay_vulkan.c
index 1815709d82..2e5bef5be5 100644
--- a/libavfilter/vf_overlay_vulkan.c
+++ b/libavfilter/vf_overlay_vulkan.c
@@ -331,9 +331,9 @@  static int process_frames(AVFilterContext *avctx, AVFrame *out_f,
                   FFALIGN(s->vkctx.output_width,  CGROUPS[0])/CGROUPS[0],
                   FFALIGN(s->vkctx.output_height, CGROUPS[1])/CGROUPS[1], 1);
 
-    ff_vk_add_exec_dep(avctx, s->exec, main_f, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT);
-    ff_vk_add_exec_dep(avctx, s->exec, overlay_f, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT);
-    ff_vk_add_exec_dep(avctx, s->exec, out_f, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT);
+    ff_vk_add_exec_dep(avctx, s->exec, main_f, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, 1);
+    ff_vk_add_exec_dep(avctx, s->exec, overlay_f, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, 1);
+    ff_vk_add_exec_dep(avctx, s->exec, out_f, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, 0);
 
     err = ff_vk_submit_exec_queue(avctx, s->exec);
     if (err)
diff --git a/libavfilter/vf_scale_vulkan.c b/libavfilter/vf_scale_vulkan.c
index 4eb4fe5664..0d946e0416 100644
--- a/libavfilter/vf_scale_vulkan.c
+++ b/libavfilter/vf_scale_vulkan.c
@@ -377,8 +377,8 @@  static int process_frames(AVFilterContext *avctx, AVFrame *out_f, AVFrame *in_f)
                   FFALIGN(s->vkctx.output_width,  CGROUPS[0])/CGROUPS[0],
                   FFALIGN(s->vkctx.output_height, CGROUPS[1])/CGROUPS[1], 1);
 
-    ff_vk_add_exec_dep(avctx, s->exec, in_f, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT);
-    ff_vk_add_exec_dep(avctx, s->exec, out_f, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT);
+    ff_vk_add_exec_dep(avctx, s->exec, in_f, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, 1);
+    ff_vk_add_exec_dep(avctx, s->exec, out_f, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, 0);
 
     err = ff_vk_submit_exec_queue(avctx, s->exec);
     if (err)
diff --git a/libavfilter/vulkan.c b/libavfilter/vulkan.c
index e5b070b3e6..e8cbf66b2b 100644
--- a/libavfilter/vulkan.c
+++ b/libavfilter/vulkan.c
@@ -462,9 +462,10 @@  VkCommandBuffer ff_vk_get_exec_buf(AVFilterContext *avctx, FFVkExecContext *e)
 }
 
 int ff_vk_add_exec_dep(AVFilterContext *avctx, FFVkExecContext *e,
-                       AVFrame *frame, VkPipelineStageFlagBits in_wait_dst_flag)
+                       AVFrame *frame, VkPipelineStageFlagBits in_wait_dst_flag, int input_frame)
 {
     AVFrame **dst;
+    VkSemaphore *sem_temp;
     VulkanFilterContext *s = avctx->priv;
     AVVkFrame *f = (AVVkFrame *)frame->data[0];
     FFVkQueueCtx *q = &e->queues[s->cur_queue_idx];
@@ -472,33 +473,39 @@  int ff_vk_add_exec_dep(AVFilterContext *avctx, FFVkExecContext *e,
     int planes = av_pix_fmt_count_planes(fc->sw_format);
 
     for (int i = 0; i < planes; i++) {
-        e->sem_wait = av_fast_realloc(e->sem_wait, &e->sem_wait_alloc,
-                                      (e->sem_wait_cnt + 1)*sizeof(*e->sem_wait));
-        if (!e->sem_wait) {
-            ff_vk_discard_exec_deps(avctx, e);
-            return AVERROR(ENOMEM);
-        }
+        if (input_frame) {
+            sem_temp = av_fast_realloc(e->sem_wait, &e->sem_wait_alloc,
+                                        (e->sem_wait_cnt + 1)*sizeof(*e->sem_wait));
+            if (!sem_temp) {
+                ff_vk_discard_exec_deps(avctx, e);
+                return AVERROR(ENOMEM);
+            }
+            e->sem_wait = sem_temp;
 
-        e->sem_wait_dst = av_fast_realloc(e->sem_wait_dst, &e->sem_wait_dst_alloc,
-                                          (e->sem_wait_cnt + 1)*sizeof(*e->sem_wait_dst));
-        if (!e->sem_wait_dst) {
-            ff_vk_discard_exec_deps(avctx, e);
-            return AVERROR(ENOMEM);
-        }
+            sem_temp = av_fast_realloc(e->sem_wait_dst, &e->sem_wait_dst_alloc,
+                                            (e->sem_wait_cnt + 1)*sizeof(*e->sem_wait_dst));
+            if (!sem_temp) {
+                ff_vk_discard_exec_deps(avctx, e);
+                return AVERROR(ENOMEM);
+            }
+            e->sem_wait_dst = sem_temp;
 
-        e->sem_sig = av_fast_realloc(e->sem_sig, &e->sem_sig_alloc,
-                                     (e->sem_sig_cnt + 1)*sizeof(*e->sem_sig));
-        if (!e->sem_sig) {
-            ff_vk_discard_exec_deps(avctx, e);
-            return AVERROR(ENOMEM);
-        }
+            e->sem_wait[e->sem_wait_cnt] = f->sem[i];
+            e->sem_wait_dst[e->sem_wait_cnt] = in_wait_dst_flag;
+            e->sem_wait_cnt++;
+        } else {
 
-        e->sem_wait[e->sem_wait_cnt] = f->sem[i];
-        e->sem_wait_dst[e->sem_wait_cnt] = in_wait_dst_flag;
-        e->sem_wait_cnt++;
+            sem_temp = av_fast_realloc(e->sem_sig, &e->sem_sig_alloc,
+                                        (e->sem_sig_cnt + 1)*sizeof(*e->sem_sig));
+            if (!sem_temp) {
+                ff_vk_discard_exec_deps(avctx, e);
+                return AVERROR(ENOMEM);
+            }
+            e->sem_sig = sem_temp;
 
-        e->sem_sig[e->sem_sig_cnt] = f->sem[i];
-        e->sem_sig_cnt++;
+            e->sem_sig[e->sem_sig_cnt] = f->sem[i];
+            e->sem_sig_cnt++;
+        }
     }
 
     dst = av_fast_realloc(q->frame_deps, &q->frame_deps_alloc_size,
diff --git a/libavfilter/vulkan.h b/libavfilter/vulkan.h
index f9a4dc5839..2fdc0e1368 100644
--- a/libavfilter/vulkan.h
+++ b/libavfilter/vulkan.h
@@ -340,7 +340,8 @@  void ff_vk_discard_exec_deps(AVFilterContext *avctx, FFVkExecContext *e);
  * Must be called before submission.
  */
 int ff_vk_add_exec_dep(AVFilterContext *avctx, FFVkExecContext *e,
-                       AVFrame *frame, VkPipelineStageFlagBits in_wait_dst_flag);
+                       AVFrame *frame, VkPipelineStageFlagBits in_wait_dst_flag,
+                       int input_frame);
 
 /**
  * Submits a command buffer to the queue for execution.
diff --git a/libavutil/hwcontext_vulkan.c b/libavutil/hwcontext_vulkan.c
index 88db5b8b70..9a29267aed 100644
--- a/libavutil/hwcontext_vulkan.c
+++ b/libavutil/hwcontext_vulkan.c
@@ -1737,8 +1737,6 @@  static int prepare_frame(AVHWFramesContext *hwfc, VulkanExecCtx *ectx,
 
     VkSubmitInfo s_info = {
         .sType                = VK_STRUCTURE_TYPE_SUBMIT_INFO,
-        .pSignalSemaphores    = frame->sem,
-        .signalSemaphoreCount = planes,
     };
 
     VkPipelineStageFlagBits wait_st[AV_NUM_DATA_POINTERS];
@@ -1750,11 +1748,15 @@  static int prepare_frame(AVHWFramesContext *hwfc, VulkanExecCtx *ectx,
         new_layout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;
         new_access = VK_ACCESS_TRANSFER_WRITE_BIT;
         dst_qf     = VK_QUEUE_FAMILY_IGNORED;
+        s_info.pSignalSemaphores    = frame->sem;
+        s_info.signalSemaphoreCount = planes;
         break;
     case PREP_MODE_RO_SHADER:
         new_layout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL;
         new_access = VK_ACCESS_TRANSFER_READ_BIT;
         dst_qf     = VK_QUEUE_FAMILY_IGNORED;
+        s_info.pSignalSemaphores    = frame->sem;
+        s_info.signalSemaphoreCount = planes;
         break;
     case PREP_MODE_EXTERNAL_EXPORT:
         new_layout = VK_IMAGE_LAYOUT_GENERAL;
@@ -3226,11 +3228,11 @@  static int transfer_image_buf(AVHWFramesContext *hwfc, const AVFrame *f,
 
     VkSubmitInfo s_info = {
         .sType                = VK_STRUCTURE_TYPE_SUBMIT_INFO,
-        .pSignalSemaphores    = frame->sem,
-        .pWaitSemaphores      = frame->sem,
+        .pSignalSemaphores    = to_buf ? NULL: frame->sem,
+        .pWaitSemaphores      = to_buf ? frame->sem : NULL,
         .pWaitDstStageMask    = sem_wait_dst,
-        .signalSemaphoreCount = planes,
-        .waitSemaphoreCount   = planes,
+        .signalSemaphoreCount = to_buf ? 0 : planes,
+        .waitSemaphoreCount   = to_buf ? planes : 0,
     };
 
     if ((err = wait_start_exec_ctx(hwfc, ectx)))