static const struct {
enum AVPixelFormat pixfmt;
- const VkFormat vkfmts[3];
+ const VkFormat vkfmts[4];
} vk_pixfmt_map[] = {
{ AV_PIX_FMT_GRAY8, { VK_FORMAT_R8_UNORM } },
{ AV_PIX_FMT_GRAY16, { VK_FORMAT_R16_UNORM } },
{ AV_PIX_FMT_GRAYF32, { VK_FORMAT_R32_SFLOAT } },
{ AV_PIX_FMT_NV12, { VK_FORMAT_R8_UNORM, VK_FORMAT_R8G8_UNORM } },
+ { AV_PIX_FMT_NV21, { VK_FORMAT_R8_UNORM, VK_FORMAT_R8G8_UNORM } },
{ AV_PIX_FMT_P010, { VK_FORMAT_R16_UNORM, VK_FORMAT_R16G16_UNORM } },
{ AV_PIX_FMT_P016, { VK_FORMAT_R16_UNORM, VK_FORMAT_R16G16_UNORM } },
- { AV_PIX_FMT_YUV420P, { VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM } },
- { AV_PIX_FMT_YUV422P, { VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM } },
- { AV_PIX_FMT_YUV444P, { VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM } },
+ { AV_PIX_FMT_NV16, { VK_FORMAT_R8_UNORM, VK_FORMAT_R8G8_UNORM } },
+ { AV_PIX_FMT_NV24, { VK_FORMAT_R8_UNORM, VK_FORMAT_R8G8_UNORM } },
+ { AV_PIX_FMT_NV42, { VK_FORMAT_R8_UNORM, VK_FORMAT_R8G8_UNORM } },
+
+ { AV_PIX_FMT_YUV420P, { VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM } },
+ { AV_PIX_FMT_YUV420P10, { VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM } },
+ { AV_PIX_FMT_YUV420P12, { VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM } },
{ AV_PIX_FMT_YUV420P16, { VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM } },
+
+ { AV_PIX_FMT_YUV422P, { VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM } },
+ { AV_PIX_FMT_YUV422P10, { VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM } },
+ { AV_PIX_FMT_YUV422P12, { VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM } },
{ AV_PIX_FMT_YUV422P16, { VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM } },
+
+ { AV_PIX_FMT_YUV444P, { VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM } },
+ { AV_PIX_FMT_YUV444P10, { VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM } },
+ { AV_PIX_FMT_YUV444P12, { VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM } },
{ AV_PIX_FMT_YUV444P16, { VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM } },
- { AV_PIX_FMT_ABGR, { VK_FORMAT_A8B8G8R8_UNORM_PACK32 } },
+ { AV_PIX_FMT_YUVA420P, { VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM } },
+ { AV_PIX_FMT_YUVA420P10, { VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM } },
+ /* There is no AV_PIX_FMT_YUVA420P12 */
+ { AV_PIX_FMT_YUVA420P16, { VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM } },
+
+ { AV_PIX_FMT_YUVA422P, { VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM } },
+ { AV_PIX_FMT_YUVA422P10, { VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM } },
+ { AV_PIX_FMT_YUVA422P12, { VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM } },
+ { AV_PIX_FMT_YUVA422P16, { VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM } },
+
+ { AV_PIX_FMT_YUVA444P, { VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM } },
+ { AV_PIX_FMT_YUVA444P10, { VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM } },
+ { AV_PIX_FMT_YUVA444P12, { VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM } },
+ { AV_PIX_FMT_YUVA444P16, { VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM } },
+
{ AV_PIX_FMT_BGRA, { VK_FORMAT_B8G8R8A8_UNORM } },
{ AV_PIX_FMT_RGBA, { VK_FORMAT_R8G8B8A8_UNORM } },
{ AV_PIX_FMT_RGB24, { VK_FORMAT_R8G8B8_UNORM } },
{ AV_PIX_FMT_BGR24, { VK_FORMAT_B8G8R8_UNORM } },
{ AV_PIX_FMT_RGB48, { VK_FORMAT_R16G16B16_UNORM } },
{ AV_PIX_FMT_RGBA64, { VK_FORMAT_R16G16B16A16_UNORM } },
+ { AV_PIX_FMT_RGBA64, { VK_FORMAT_R16G16B16A16_UNORM } },
{ AV_PIX_FMT_RGB565, { VK_FORMAT_R5G6B5_UNORM_PACK16 } },
{ AV_PIX_FMT_BGR565, { VK_FORMAT_B5G6R5_UNORM_PACK16 } },
{ AV_PIX_FMT_BGR0, { VK_FORMAT_B8G8R8A8_UNORM } },
- { AV_PIX_FMT_0BGR, { VK_FORMAT_A8B8G8R8_UNORM_PACK32 } },
{ AV_PIX_FMT_RGB0, { VK_FORMAT_R8G8B8A8_UNORM } },
+ /* Lower priority as there's an endianess-dependent overlap between these
+ * and rgba/bgr0, and PACK32 formats are more limited */
+ { AV_PIX_FMT_BGR32, { VK_FORMAT_A8B8G8R8_UNORM_PACK32 } },
+ { AV_PIX_FMT_0BGR32, { VK_FORMAT_A8B8G8R8_UNORM_PACK32 } },
+
+ { AV_PIX_FMT_X2RGB10, { VK_FORMAT_A2R10G10B10_UNORM_PACK32 } },
+
+ { AV_PIX_FMT_GBRAP, { VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM, VK_FORMAT_R8_UNORM } },
+ { AV_PIX_FMT_GBRAP16, { VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM, VK_FORMAT_R16_UNORM } },
{ AV_PIX_FMT_GBRPF32, { VK_FORMAT_R32_SFLOAT, VK_FORMAT_R32_SFLOAT, VK_FORMAT_R32_SFLOAT } },
+ { AV_PIX_FMT_GBRAPF32, { VK_FORMAT_R32_SFLOAT, VK_FORMAT_R32_SFLOAT, VK_FORMAT_R32_SFLOAT, VK_FORMAT_R32_SFLOAT } },
};
const VkFormat *av_vkfmt_from_pixfmt(enum AVPixelFormat p)
EXT_EXTERNAL_FD_MEMORY = 1ULL << 2, /* VK_KHR_external_memory_fd */
EXT_EXTERNAL_FD_SEM = 1ULL << 3, /* VK_KHR_external_semaphore_fd */
EXT_EXTERNAL_HOST_MEMORY = 1ULL << 4, /* VK_EXT_external_memory_host */
+ EXT_PUSH_DESCRIPTORS = 1ULL << 5, /* VK_KHR_push_descriptor */
+ EXT_HOST_QUERY_RESET = 1ULL << 6, /* VK_EXT_host_query_reset */
EXT_NO_FLAG = 1ULL << 63,
};
{ VK_EXT_IMAGE_DRM_FORMAT_MODIFIER_EXTENSION_NAME, EXT_DRM_MODIFIER_FLAGS, },
{ VK_KHR_EXTERNAL_SEMAPHORE_FD_EXTENSION_NAME, EXT_EXTERNAL_FD_SEM, },
{ VK_EXT_EXTERNAL_MEMORY_HOST_EXTENSION_NAME, EXT_EXTERNAL_HOST_MEMORY, },
+ { VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME, EXT_PUSH_DESCRIPTORS, },
+ { VK_EXT_HOST_QUERY_RESET_EXTENSION_NAME, EXT_HOST_QUERY_RESET, },
};
/* Converts return values to strings */
vkGetPhysicalDeviceFeatures(hwctx->phys_dev, &dev_features);
#define COPY_FEATURE(DST, NAME) (DST).features.NAME = dev_features.NAME;
COPY_FEATURE(hwctx->device_features, shaderImageGatherExtended)
+ COPY_FEATURE(hwctx->device_features, shaderStorageImageReadWithoutFormat)
+ COPY_FEATURE(hwctx->device_features, shaderStorageImageWriteWithoutFormat)
COPY_FEATURE(hwctx->device_features, fragmentStoresAndAtomics)
COPY_FEATURE(hwctx->device_features, vertexPipelineStoresAndAtomics)
COPY_FEATURE(hwctx->device_features, shaderInt64)
return submit_exec_ctx(hwfc, ectx, &s_info, 0);
}
+static inline void get_plane_wh(int *w, int *h, enum AVPixelFormat format,
+ int frame_w, int frame_h, int plane)
+{
+ const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
+
+ /* Currently always true unless gray + alpha support is added */
+ if (!plane || (plane == 3) || desc->flags & AV_PIX_FMT_FLAG_RGB ||
+ !(desc->flags & AV_PIX_FMT_FLAG_PLANAR)) {
+ *w = frame_w;
+ *h = frame_h;
+ return;
+ }
+
+ *w = AV_CEIL_RSHIFT(frame_w, desc->log2_chroma_w);
+ *h = AV_CEIL_RSHIFT(frame_h, desc->log2_chroma_h);
+}
+
static int create_frame(AVHWFramesContext *hwfc, AVVkFrame **frame,
VkImageTiling tiling, VkImageUsageFlagBits usage,
void *create_pnext)
/* Create the images */
for (int i = 0; i < planes; i++) {
- const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
- int w = hwfc->width;
- int h = hwfc->height;
- const int p_w = i > 0 ? AV_CEIL_RSHIFT(w, desc->log2_chroma_w) : w;
- const int p_h = i > 0 ? AV_CEIL_RSHIFT(h, desc->log2_chroma_h) : h;
-
- VkImageCreateInfo image_create_info = {
+ VkImageCreateInfo create_info = {
.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
.pNext = create_pnext,
.imageType = VK_IMAGE_TYPE_2D,
.format = img_fmts[i],
- .extent.width = p_w,
- .extent.height = p_h,
.extent.depth = 1,
.mipLevels = 1,
.arrayLayers = 1,
VK_SHARING_MODE_EXCLUSIVE,
};
- ret = vkCreateImage(hwctx->act_dev, &image_create_info,
+ get_plane_wh(&create_info.extent.width, &create_info.extent.height,
+ format, hwfc->width, hwfc->height, i);
+
+ ret = vkCreateImage(hwctx->act_dev, &create_info,
hwctx->alloc, &f->img[i]);
if (ret != VK_SUCCESS) {
av_log(ctx, AV_LOG_ERROR, "Image creation failure: %s\n",
return AVERROR_EXTERNAL;
}
- f->layout[i] = image_create_info.initialLayout;
+ f->layout[i] = create_info.initialLayout;
f->access[i] = 0x0;
}
}
}
-static AVBufferRef *vulkan_pool_alloc(void *opaque, int size)
+static AVBufferRef *vulkan_pool_alloc(void *opaque, size_t size)
{
int err;
AVVkFrame *f;
}
static int vulkan_map_from_drm_frame_desc(AVHWFramesContext *hwfc, AVVkFrame **frame,
- AVDRMFrameDescriptor *desc)
+ const AVFrame *src)
{
int err = 0;
VkResult ret;
VulkanDevicePriv *p = ctx->internal->priv;
VulkanFramesPriv *fp = hwfc->internal->priv;
AVVulkanFramesContext *frames_hwctx = hwfc->hwctx;
- const AVPixFmtDescriptor *fmt_desc = av_pix_fmt_desc_get(hwfc->sw_format);
+ const AVDRMFrameDescriptor *desc = (AVDRMFrameDescriptor *)src->data[0];
const int has_modifiers = !!(p->extensions & EXT_DRM_MODIFIER_FLAGS);
VkSubresourceLayout plane_data[AV_NUM_DATA_POINTERS] = { 0 };
VkBindImageMemoryInfo bind_info[AV_NUM_DATA_POINTERS] = { 0 };
.sType = VK_STRUCTURE_TYPE_SEMAPHORE_CREATE_INFO,
};
- const int p_w = i > 0 ? AV_CEIL_RSHIFT(hwfc->width, fmt_desc->log2_chroma_w) : hwfc->width;
- const int p_h = i > 0 ? AV_CEIL_RSHIFT(hwfc->height, fmt_desc->log2_chroma_h) : hwfc->height;
-
- VkImageCreateInfo image_create_info = {
+ VkImageCreateInfo create_info = {
.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
.pNext = &einfo,
.imageType = VK_IMAGE_TYPE_2D,
.format = drm_to_vulkan_fmt(desc->layers[i].format),
- .extent.width = p_w,
- .extent.height = p_h,
.extent.depth = 1,
.mipLevels = 1,
.arrayLayers = 1,
VK_SHARING_MODE_EXCLUSIVE,
};
+ get_plane_wh(&create_info.extent.width, &create_info.extent.height,
+ hwfc->sw_format, src->width, src->height, i);
+
for (int j = 0; j < planes; j++) {
plane_data[j].offset = desc->layers[i].planes[j].offset;
plane_data[j].rowPitch = desc->layers[i].planes[j].pitch;
}
/* Create image */
- ret = vkCreateImage(hwctx->act_dev, &image_create_info,
+ ret = vkCreateImage(hwctx->act_dev, &create_info,
hwctx->alloc, &f->img[i]);
if (ret != VK_SUCCESS) {
av_log(ctx, AV_LOG_ERROR, "Image creation failure: %s\n",
* 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->layout[i] = create_info.initialLayout;
f->access[i] = 0x0;
}
AVVkFrame *f;
VulkanMapping *map = NULL;
- err = vulkan_map_from_drm_frame_desc(hwfc, &f,
- (AVDRMFrameDescriptor *)src->data[0]);
- if (err)
+ if ((err = vulkan_map_from_drm_frame_desc(hwfc, &f, src)))
return err;
/* The unmapping function will free this */
CUDA_EXTERNAL_MEMORY_MIPMAPPED_ARRAY_DESC tex_desc = {
.offset = 0,
.arrayDesc = {
- .Width = i > 0 ? AV_CEIL_RSHIFT(hwfc->width, desc->log2_chroma_w)
- : hwfc->width,
- .Height = i > 0 ? AV_CEIL_RSHIFT(hwfc->height, desc->log2_chroma_h)
- : hwfc->height,
.Depth = 0,
.Format = cufmt,
.NumChannels = 1 + ((planes == 2) && i),
.type = CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD,
};
+ int p_w, p_h;
+ get_plane_wh(&p_w, &p_h, hwfc->sw_format, hwfc->width, hwfc->height, i);
+
+ tex_desc.arrayDesc.Width = p_w;
+ tex_desc.arrayDesc.Height = p_h;
+
ret = pfn_vkGetMemoryFdKHR(hwctx->act_dev, &export_info,
&ext_desc.handle.fd);
if (ret != VK_SUCCESS) {
.dstMemoryType = CU_MEMORYTYPE_ARRAY,
.dstArray = dst_int->cu_array[i],
- .WidthInBytes = (i > 0 ? AV_CEIL_RSHIFT(hwfc->width, desc->log2_chroma_w)
- : hwfc->width) * desc->comp[i].step,
- .Height = i > 0 ? AV_CEIL_RSHIFT(hwfc->height, desc->log2_chroma_h)
- : hwfc->height,
};
+ int p_w, p_h;
+ get_plane_wh(&p_w, &p_h, hwfc->sw_format, hwfc->width, hwfc->height, i);
+
+ cpy.WidthInBytes = p_w * desc->comp[i].step;
+ cpy.Height = p_h;
+
ret = CHECK_CU(cu->cuMemcpy2DAsync(&cpy, cuda_dev->stream));
if (ret < 0) {
err = AVERROR_EXTERNAL;
/* Schedule a copy for each plane */
for (int i = 0; i < planes; i++) {
ImageBuffer *vkbuf = (ImageBuffer *)bufs[i]->data;
- const int p_w = i > 0 ? AV_CEIL_RSHIFT(w, desc->log2_chroma_w) : w;
- const int p_h = i > 0 ? AV_CEIL_RSHIFT(h, desc->log2_chroma_h) : h;
VkBufferImageCopy buf_reg = {
.bufferOffset = buf_offsets[i],
- /* Buffer stride isn't in bytes, it's in samples, the implementation
- * uses the image's VkFormat to know how many bytes per sample
- * the buffer has. So we have to convert by dividing. Stupid.
- * Won't work with YUVA or other planar formats with alpha. */
.bufferRowLength = buf_stride[i] / desc->comp[i].step,
- .bufferImageHeight = p_h,
.imageSubresource.layerCount = 1,
.imageSubresource.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
.imageOffset = { 0, 0, 0, },
- .imageExtent = { p_w, p_h, 1, },
};
+ int p_w, p_h;
+ get_plane_wh(&p_w, &p_h, pix_fmt, w, h, i);
+
+ buf_reg.bufferImageHeight = p_h;
+ buf_reg.imageExtent = (VkExtent3D){ p_w, p_h, 1, };
+
if (to_buf)
vkCmdCopyImageToBuffer(cmd_buf, frame->img[i], frame->layout[i],
vkbuf->buf, 1, &buf_reg);
AVBufferRef *bufs[AV_NUM_DATA_POINTERS] = { 0 };
size_t buf_offsets[AV_NUM_DATA_POINTERS] = { 0 };
+ int p_w, p_h;
const int planes = av_pix_fmt_count_planes(swf->format);
- int log2_chroma = av_pix_fmt_desc_get(swf->format)->log2_chroma_h;
int host_mapped[AV_NUM_DATA_POINTERS] = { 0 };
const int map_host = !!(p->extensions & EXT_EXTERNAL_HOST_MEMORY);
/* Create buffers */
for (int i = 0; i < planes; i++) {
size_t req_size;
- int h = swf->height;
- int p_height = i > 0 ? AV_CEIL_RSHIFT(h, log2_chroma) : h;
VkExternalMemoryBufferCreateInfo create_desc = {
.sType = VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_BUFFER_CREATE_INFO,
.sType = VK_STRUCTURE_TYPE_MEMORY_HOST_POINTER_PROPERTIES_EXT,
};
+ get_plane_wh(&p_w, &p_h, swf->format, swf->width, swf->height, i);
+
tmp.linesize[i] = FFABS(swf->linesize[i]);
/* Do not map images with a negative stride */
/* We have to compensate for the few extra bytes of padding we
* completely ignore at the start */
- req_size = FFALIGN(offs + tmp.linesize[i] * p_height,
+ req_size = FFALIGN(offs + tmp.linesize[i] * p_h,
p->hprops.minImportedHostPointerAlignment);
ret = pfn_vkGetMemoryHostPointerPropertiesEXT(hwctx->act_dev,
}
if (!host_mapped[i])
- req_size = get_req_buffer_size(p, &tmp.linesize[i], p_height);
+ req_size = get_req_buffer_size(p, &tmp.linesize[i], p_h);
err = create_buf(dev_ctx, &bufs[i],
from ? VK_BUFFER_USAGE_TRANSFER_DST_BIT :
goto end;
for (int i = 0; i < planes; i++) {
- int h = swf->height;
- int p_height = i > 0 ? AV_CEIL_RSHIFT(h, log2_chroma) : h;
-
if (host_mapped[i])
continue;
+ get_plane_wh(&p_w, &p_h, swf->format, swf->width, swf->height, i);
+
av_image_copy_plane(tmp.data[i], tmp.linesize[i],
(const uint8_t *)swf->data[i], swf->linesize[i],
FFMIN(tmp.linesize[i], FFABS(swf->linesize[i])),
- p_height);
+ p_h);
}
if ((err = unmap_buffers(dev_ctx, bufs, planes, 1)))
goto end;
for (int i = 0; i < planes; i++) {
- int h = swf->height;
- int p_height = i > 0 ? AV_CEIL_RSHIFT(h, log2_chroma) : h;
-
if (host_mapped[i])
continue;
+ get_plane_wh(&p_w, &p_h, swf->format, swf->width, swf->height, i);
+
av_image_copy_plane(swf->data[i], swf->linesize[i],
(const uint8_t *)tmp.data[i], tmp.linesize[i],
FFMIN(tmp.linesize[i], FFABS(swf->linesize[i])),
- p_height);
+ p_h);
}
if ((err = unmap_buffers(dev_ctx, bufs, planes, 1)))
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 };
ret = CHECK_CU(cu->cuCtxPushCurrent(cuda_dev->cuda_ctx));
if (ret < 0)
dst_int = dst_f->internal;
+ ret = CHECK_CU(cu->cuWaitExternalSemaphoresAsync(dst_int->cu_sem, s_w_par,
+ planes, cuda_dev->stream));
+ if (ret < 0) {
+ err = AVERROR_EXTERNAL;
+ goto fail;
+ }
+
for (int i = 0; i < planes; i++) {
CUDA_MEMCPY2D cpy = {
.dstMemoryType = CU_MEMORYTYPE_DEVICE,
.srcMemoryType = CU_MEMORYTYPE_ARRAY,
.srcArray = dst_int->cu_array[i],
- .WidthInBytes = (i > 0 ? AV_CEIL_RSHIFT(hwfc->width, desc->log2_chroma_w)
- : hwfc->width) * desc->comp[i].step,
- .Height = i > 0 ? AV_CEIL_RSHIFT(hwfc->height, desc->log2_chroma_h)
- : hwfc->height,
};
+ int w, h;
+ get_plane_wh(&w, &h, hwfc->sw_format, hwfc->width, hwfc->height, i);
+
+ cpy.WidthInBytes = w * desc->comp[i].step;
+ cpy.Height = h;
+
ret = CHECK_CU(cu->cuMemcpy2DAsync(&cpy, cuda_dev->stream));
if (ret < 0) {
err = AVERROR_EXTERNAL;
}
}
+ ret = CHECK_CU(cu->cuSignalExternalSemaphoresAsync(dst_int->cu_sem, s_s_par,
+ planes, cuda_dev->stream));
+ if (ret < 0) {
+ err = AVERROR_EXTERNAL;
+ goto fail;
+ }
+
CHECK_CU(cu->cuCtxPopCurrent(&dummy));
av_log(hwfc, AV_LOG_VERBOSE, "Transfered Vulkan image to CUDA!\n");
projects / ffmpeg / blobdiff