From 39506e3d6d703b070d861f88d4cde5689eea7ae2 Mon Sep 17 00:00:00 2001
From: Lynne <dev@lynne.ee>
Date: Sat, 4 Apr 2020 17:19:59 +0100
Subject: [PATCH] hwcontext_vulkan: only use one semaphore per image
The idea was to allow separate planes to be filtered independently, however,
in hindsight, literaly nothing uses separate per-plane semaphores and it
would only work when each plane is backed by separate device memory.
---
libavfilter/vulkan.c | 38 +++++-----
libavutil/hwcontext_vulkan.c | 131 ++++++++++++++++++-----------------
libavutil/hwcontext_vulkan.h | 4 +-
3 files changed, 86 insertions(+), 87 deletions(-)
@@ -390,32 +390,28 @@ int ff_vk_add_exec_dep(AVFilterContext *avctx, FFVkExecContext *e,
AVFrame *frame, VkPipelineStageFlagBits in_wait_dst_flag)
{
AVVkFrame *f = (AVVkFrame *)frame->data[0];
- AVHWFramesContext *fc = (AVHWFramesContext *)frame->hw_frames_ctx->data;
- 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)
- return AVERROR(ENOMEM);
+ 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)
+ return AVERROR(ENOMEM);
- 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)
- return AVERROR(ENOMEM);
+ 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)
+ return AVERROR(ENOMEM);
- 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)
- return AVERROR(ENOMEM);
+ 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)
+ 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++;
+ e->sem_wait[e->sem_wait_cnt] = f->sem;
+ e->sem_wait_dst[e->sem_wait_cnt] = in_wait_dst_flag;
+ e->sem_wait_cnt++;
- e->sem_sig[e->sem_sig_cnt] = f->sem[i];
- e->sem_sig_cnt++;
- }
+ e->sem_sig[e->sem_sig_cnt] = f->sem;
+ e->sem_sig_cnt++;
return 0;
}
@@ -81,7 +81,7 @@ typedef struct AVVkFrameInternal {
CUexternalMemory ext_mem[AV_NUM_DATA_POINTERS];
CUmipmappedArray cu_mma[AV_NUM_DATA_POINTERS];
CUarray cu_array[AV_NUM_DATA_POINTERS];
- CUexternalSemaphore cu_sem[AV_NUM_DATA_POINTERS];
+ CUexternalSemaphore cu_sem;
#endif
} AVVkFrameInternal;
@@ -1042,9 +1042,10 @@ static void vulkan_free_internal(AVVkFrameInternal *internal)
AVCUDADeviceContextInternal *cu_internal = cuda_dev->internal;
CudaFunctions *cu = cu_internal->cuda_dl;
+ if (internal->cu_sem)
+ CHECK_CU(cu->cuDestroyExternalSemaphore(internal->cu_sem));
+
for (int i = 0; i < planes; i++) {
- if (internal->cu_sem[i])
- CHECK_CU(cu->cuDestroyExternalSemaphore(internal->cu_sem[i]));
if (internal->cu_mma[i])
CHECK_CU(cu->cuMipmappedArrayDestroy(internal->cu_mma[i]));
if (internal->ext_mem[i])
@@ -1070,9 +1071,10 @@ static void vulkan_frame_free(void *opaque, uint8_t *data)
for (int i = 0; i < planes; i++) {
vkDestroyImage(hwctx->act_dev, f->img[i], hwctx->alloc);
vkFreeMemory(hwctx->act_dev, f->mem[i], hwctx->alloc);
- vkDestroySemaphore(hwctx->act_dev, f->sem[i], hwctx->alloc);
}
+ vkDestroySemaphore(hwctx->act_dev, f->sem, hwctx->alloc);
+
av_free(f);
}
@@ -1166,8 +1168,8 @@ static int prepare_frame(AVHWFramesContext *hwfc, VulkanExecCtx *ectx,
.commandBufferCount = 1,
.pCommandBuffers = &ectx->buf,
- .pSignalSemaphores = frame->sem,
- .signalSemaphoreCount = planes,
+ .pSignalSemaphores = &frame->sem,
+ .signalSemaphoreCount = 1,
};
switch (pmode) {
@@ -1288,19 +1290,19 @@ static int create_frame(AVHWFramesContext *hwfc, AVVkFrame **frame,
goto fail;
}
- /* Create semaphore */
- ret = vkCreateSemaphore(hwctx->act_dev, &sem_spawn,
- hwctx->alloc, &f->sem[i]);
- if (ret != VK_SUCCESS) {
- av_log(hwctx, AV_LOG_ERROR, "Failed to create semaphore: %s\n",
- vk_ret2str(ret));
- return AVERROR_EXTERNAL;
- }
-
f->layout[i] = image_create_info.initialLayout;
f->access[i] = 0x0;
}
+ /* Create semaphore */
+ ret = vkCreateSemaphore(hwctx->act_dev, &sem_spawn,
+ hwctx->alloc, &f->sem);
+ if (ret != VK_SUCCESS) {
+ av_log(hwctx, AV_LOG_ERROR, "Failed to create semaphore: %s\n",
+ vk_ret2str(ret));
+ return AVERROR_EXTERNAL;
+ }
+
f->flags = 0x0;
f->tiling = tiling;
@@ -1622,9 +1624,10 @@ static void vulkan_unmap_from(AVHWFramesContext *hwfc, HWMapDescriptor *hwmap)
for (int i = 0; i < planes; i++) {
vkDestroyImage(hwctx->act_dev, map->frame->img[i], hwctx->alloc);
vkFreeMemory(hwctx->act_dev, map->frame->mem[i], hwctx->alloc);
- vkDestroySemaphore(hwctx->act_dev, map->frame->sem[i], hwctx->alloc);
}
+ vkDestroySemaphore(hwctx->act_dev, map->frame->sem, hwctx->alloc);
+
av_freep(&map->frame);
}
@@ -1785,19 +1788,6 @@ static int vulkan_map_from_drm_frame_desc(AVHWFramesContext *hwfc, AVVkFrame **f
goto fail;
}
- ret = vkCreateSemaphore(hwctx->act_dev, &sem_spawn,
- hwctx->alloc, &f->sem[i]);
- if (ret != VK_SUCCESS) {
- av_log(hwctx, AV_LOG_ERROR, "Failed to create semaphore: %s\n",
- vk_ret2str(ret));
- return AVERROR_EXTERNAL;
- }
-
- /* We'd import a semaphore onto the one we created using
- * vkImportSemaphoreFdKHR but unfortunately neither DRM nor VAAPI
- * offer us anything we could import and sync with, so instead
- * just signal the semaphore we created. */
-
f->layout[i] = image_create_info.initialLayout;
f->access[i] = 0x0;
@@ -1818,6 +1808,19 @@ static int vulkan_map_from_drm_frame_desc(AVHWFramesContext *hwfc, AVVkFrame **f
}
}
+ ret = vkCreateSemaphore(hwctx->act_dev, &sem_spawn,
+ hwctx->alloc, &f->sem);
+ if (ret != VK_SUCCESS) {
+ av_log(hwctx, AV_LOG_ERROR, "Failed to create semaphore: %s\n",
+ vk_ret2str(ret));
+ return AVERROR_EXTERNAL;
+ }
+
+ /* We'd import a semaphore onto the one we created using
+ * vkImportSemaphoreFdKHR but unfortunately neither DRM nor VAAPI
+ * offer us anything we could import and sync with, so instead
+ * just signal the semaphore we created. */
+
/* Bind the allocated memory to the images */
ret = vkBindImageMemory2(hwctx->act_dev, bind_counts, bind_info);
if (ret != VK_SUCCESS) {
@@ -1838,12 +1841,11 @@ static int vulkan_map_from_drm_frame_desc(AVHWFramesContext *hwfc, AVVkFrame **f
return 0;
fail:
- for (int i = 0; i < desc->nb_layers; i++) {
+ for (int i = 0; i < desc->nb_layers; i++)
vkDestroyImage(hwctx->act_dev, f->img[i], hwctx->alloc);
- vkDestroySemaphore(hwctx->act_dev, f->sem[i], hwctx->alloc);
- }
for (int i = 0; i < desc->nb_objects; i++)
vkFreeMemory(hwctx->act_dev, f->mem[i], hwctx->alloc);
+ vkDestroySemaphore(hwctx->act_dev, f->sem, hwctx->alloc);
av_free(f);
@@ -1953,6 +1955,15 @@ static int vulkan_export_to_cuda(AVHWFramesContext *hwfc,
dst_int = dst_f->internal;
if (!dst_int || !dst_int->cuda_fc_ref) {
+ VkSemaphoreGetFdInfoKHR sem_export = {
+ .sType = VK_STRUCTURE_TYPE_SEMAPHORE_GET_FD_INFO_KHR,
+ .semaphore = dst_f->sem,
+ .handleType = VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT,
+ };
+ CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC ext_sem_desc = {
+ .type = CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD,
+ };
+
if (!dst_f->internal)
dst_f->internal = dst_int = av_mallocz(sizeof(*dst_f->internal));
@@ -1991,14 +2002,6 @@ static int vulkan_export_to_cuda(AVHWFramesContext *hwfc,
.memory = dst_f->mem[i],
.handleType = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT_KHR,
};
- VkSemaphoreGetFdInfoKHR sem_export = {
- .sType = VK_STRUCTURE_TYPE_SEMAPHORE_GET_FD_INFO_KHR,
- .semaphore = dst_f->sem[i],
- .handleType = VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT,
- };
- CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC ext_sem_desc = {
- .type = CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD,
- };
ret = pfn_vkGetMemoryFdKHR(hwctx->act_dev, &export_info,
&ext_desc.handle.fd);
@@ -2028,22 +2031,22 @@ static int vulkan_export_to_cuda(AVHWFramesContext *hwfc,
err = AVERROR_EXTERNAL;
goto fail;
}
+ }
- ret = pfn_vkGetSemaphoreFdKHR(hwctx->act_dev, &sem_export,
- &ext_sem_desc.handle.fd);
- if (ret != VK_SUCCESS) {
- av_log(ctx, AV_LOG_ERROR, "Failed to export semaphore: %s\n",
- vk_ret2str(ret));
- err = AVERROR_EXTERNAL;
- goto fail;
- }
+ ret = pfn_vkGetSemaphoreFdKHR(hwctx->act_dev, &sem_export,
+ &ext_sem_desc.handle.fd);
+ if (ret != VK_SUCCESS) {
+ av_log(ctx, AV_LOG_ERROR, "Failed to export semaphore: %s\n",
+ vk_ret2str(ret));
+ err = AVERROR_EXTERNAL;
+ goto fail;
+ }
- ret = CHECK_CU(cu->cuImportExternalSemaphore(&dst_int->cu_sem[i],
- &ext_sem_desc));
- if (ret < 0) {
- err = AVERROR_EXTERNAL;
- goto fail;
- }
+ ret = CHECK_CU(cu->cuImportExternalSemaphore(&dst_int->cu_sem,
+ &ext_sem_desc));
+ if (ret < 0) {
+ err = AVERROR_EXTERNAL;
+ goto fail;
}
}
@@ -2069,8 +2072,8 @@ static int vulkan_transfer_data_from_cuda(AVHWFramesContext *hwfc,
AVCUDADeviceContext *cuda_dev = cuda_cu->hwctx;
AVCUDADeviceContextInternal *cu_internal = cuda_dev->internal;
CudaFunctions *cu = cu_internal->cuda_dl;
- CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS s_w_par[AV_NUM_DATA_POINTERS] = { 0 };
- CUDA_EXTERNAL_SEMAPHORE_SIGNAL_PARAMS s_s_par[AV_NUM_DATA_POINTERS] = { 0 };
+ CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS s_w_par = { 0 };
+ CUDA_EXTERNAL_SEMAPHORE_SIGNAL_PARAMS s_s_par = { 0 };
ret = CHECK_CU(cu->cuCtxPushCurrent(cuda_dev->cuda_ctx));
if (ret < 0) {
@@ -2086,8 +2089,8 @@ static int vulkan_transfer_data_from_cuda(AVHWFramesContext *hwfc,
}
dst_int = dst_f->internal;
- ret = CHECK_CU(cu->cuWaitExternalSemaphoresAsync(dst_int->cu_sem, s_w_par,
- planes, cuda_dev->stream));
+ ret = CHECK_CU(cu->cuWaitExternalSemaphoresAsync(&dst_int->cu_sem, &s_w_par,
+ 1, cuda_dev->stream));
if (ret < 0) {
err = AVERROR_EXTERNAL;
goto fail;
@@ -2115,8 +2118,8 @@ static int vulkan_transfer_data_from_cuda(AVHWFramesContext *hwfc,
}
}
- ret = CHECK_CU(cu->cuSignalExternalSemaphoresAsync(dst_int->cu_sem, s_s_par,
- planes, cuda_dev->stream));
+ ret = CHECK_CU(cu->cuSignalExternalSemaphoresAsync(&dst_int->cu_sem, &s_s_par,
+ 1, cuda_dev->stream));
if (ret < 0) {
err = AVERROR_EXTERNAL;
goto fail;
@@ -2492,11 +2495,11 @@ static int transfer_image_buf(AVHWDeviceContext *ctx, AVVkFrame *frame,
.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO,
.commandBufferCount = 1,
.pCommandBuffers = &s->cmd.buf,
- .pSignalSemaphores = frame->sem,
- .pWaitSemaphores = frame->sem,
+ .pSignalSemaphores = &frame->sem,
+ .pWaitSemaphores = &frame->sem,
.pWaitDstStageMask = sem_wait_dst,
- .signalSemaphoreCount = planes,
- .waitSemaphoreCount = planes,
+ .signalSemaphoreCount = 1,
+ .waitSemaphoreCount = 1,
};
ret = vkBeginCommandBuffer(s->cmd.buf, &cmd_start);
@@ -133,10 +133,10 @@ typedef struct AVVkFrame {
VkImageLayout layout[AV_NUM_DATA_POINTERS];
/**
- * Per-image semaphores. Must not be freed manually. Must be waited on
+ * Per-frame semaphore. Must not be freed manually. Must be waited on
* and signalled at every queue submission.
*/
- VkSemaphore sem[AV_NUM_DATA_POINTERS];
+ VkSemaphore sem;
/**
* Internal data.
--
2.26.0