#include "hwcontext.h"
#include "hwcontext_internal.h"
#include "hwcontext_cuda_internal.h"
+#if CONFIG_VULKAN
+#include "hwcontext_vulkan.h"
+#endif
+#include "cuda_check.h"
#include "mem.h"
#include "pixdesc.h"
#include "pixfmt.h"
-
-#define CUDA_FRAME_ALIGNMENT 256
+#include "imgutils.h"
typedef struct CUDAFramesContext {
int shift_width, shift_height;
+ int tex_alignment;
} CUDAFramesContext;
static const enum AVPixelFormat supported_formats[] = {
AV_PIX_FMT_NV12,
AV_PIX_FMT_YUV420P,
+ AV_PIX_FMT_YUVA420P,
AV_PIX_FMT_YUV444P,
AV_PIX_FMT_P010,
AV_PIX_FMT_P016,
AV_PIX_FMT_YUV444P16,
+ AV_PIX_FMT_0RGB32,
+ AV_PIX_FMT_0BGR32,
+#if CONFIG_VULKAN
+ AV_PIX_FMT_VULKAN,
+#endif
};
+#define CHECK_CU(x) FF_CUDA_CHECK_DL(device_ctx, cu, x)
+
static int cuda_frames_get_constraints(AVHWDeviceContext *ctx,
const void *hwconfig,
AVHWFramesConstraints *constraints)
static void cuda_buffer_free(void *opaque, uint8_t *data)
{
- AVHWFramesContext *ctx = opaque;
- AVCUDADeviceContext *hwctx = ctx->device_ctx->hwctx;
- CudaFunctions *cu = hwctx->internal->cuda_dl;
+ AVHWFramesContext *ctx = opaque;
+ AVHWDeviceContext *device_ctx = ctx->device_ctx;
+ AVCUDADeviceContext *hwctx = device_ctx->hwctx;
+ CudaFunctions *cu = hwctx->internal->cuda_dl;
CUcontext dummy;
- cu->cuCtxPushCurrent(hwctx->cuda_ctx);
+ CHECK_CU(cu->cuCtxPushCurrent(hwctx->cuda_ctx));
- cu->cuMemFree((CUdeviceptr)data);
+ CHECK_CU(cu->cuMemFree((CUdeviceptr)data));
- cu->cuCtxPopCurrent(&dummy);
+ CHECK_CU(cu->cuCtxPopCurrent(&dummy));
}
-static AVBufferRef *cuda_pool_alloc(void *opaque, int size)
+static AVBufferRef *cuda_pool_alloc(void *opaque, size_t size)
{
- AVHWFramesContext *ctx = opaque;
- AVCUDADeviceContext *hwctx = ctx->device_ctx->hwctx;
- CudaFunctions *cu = hwctx->internal->cuda_dl;
+ AVHWFramesContext *ctx = opaque;
+ AVHWDeviceContext *device_ctx = ctx->device_ctx;
+ AVCUDADeviceContext *hwctx = device_ctx->hwctx;
+ CudaFunctions *cu = hwctx->internal->cuda_dl;
AVBufferRef *ret = NULL;
CUcontext dummy = NULL;
CUdeviceptr data;
- CUresult err;
+ int err;
- err = cu->cuCtxPushCurrent(hwctx->cuda_ctx);
- if (err != CUDA_SUCCESS) {
- av_log(ctx, AV_LOG_ERROR, "Error setting current CUDA context\n");
+ err = CHECK_CU(cu->cuCtxPushCurrent(hwctx->cuda_ctx));
+ if (err < 0)
return NULL;
- }
- err = cu->cuMemAlloc(&data, size);
- if (err != CUDA_SUCCESS)
+ err = CHECK_CU(cu->cuMemAlloc(&data, size));
+ if (err < 0)
goto fail;
ret = av_buffer_create((uint8_t*)data, size, cuda_buffer_free, ctx, 0);
if (!ret) {
- cu->cuMemFree(data);
+ CHECK_CU(cu->cuMemFree(data));
goto fail;
}
fail:
- cu->cuCtxPopCurrent(&dummy);
+ CHECK_CU(cu->cuCtxPopCurrent(&dummy));
return ret;
}
static int cuda_frames_init(AVHWFramesContext *ctx)
{
- CUDAFramesContext *priv = ctx->internal->priv;
- int aligned_width = FFALIGN(ctx->width, CUDA_FRAME_ALIGNMENT);
- int i;
+ AVHWDeviceContext *device_ctx = ctx->device_ctx;
+ AVCUDADeviceContext *hwctx = device_ctx->hwctx;
+ CUDAFramesContext *priv = ctx->internal->priv;
+ CudaFunctions *cu = hwctx->internal->cuda_dl;
+ int err, i;
for (i = 0; i < FF_ARRAY_ELEMS(supported_formats); i++) {
if (ctx->sw_format == supported_formats[i])
return AVERROR(ENOSYS);
}
+ err = CHECK_CU(cu->cuDeviceGetAttribute(&priv->tex_alignment,
+ 14 /* CU_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT */,
+ hwctx->internal->cuda_device));
+ if (err < 0)
+ return err;
+
+ av_log(ctx, AV_LOG_DEBUG, "CUDA texture alignment: %d\n", priv->tex_alignment);
+
+ // YUV420P is a special case.
+ // Since nvenc expects the U/V planes to have half the linesize of the Y plane
+ // alignment has to be doubled to ensure the U/V planes still end up aligned.
+ if (ctx->sw_format == AV_PIX_FMT_YUV420P)
+ priv->tex_alignment *= 2;
+
av_pix_fmt_get_chroma_sub_sample(ctx->sw_format, &priv->shift_width, &priv->shift_height);
if (!ctx->pool) {
- int size;
-
- switch (ctx->sw_format) {
- case AV_PIX_FMT_NV12:
- case AV_PIX_FMT_YUV420P:
- size = aligned_width * ctx->height * 3 / 2;
- break;
- case AV_PIX_FMT_YUV444P:
- case AV_PIX_FMT_P010:
- case AV_PIX_FMT_P016:
- size = aligned_width * ctx->height * 3;
- break;
- case AV_PIX_FMT_YUV444P16:
- size = aligned_width * ctx->height * 6;
- break;
- default:
- av_log(ctx, AV_LOG_ERROR, "BUG: Pixel format missing from size calculation.");
- return AVERROR_BUG;
- }
+ int size = av_image_get_buffer_size(ctx->sw_format, ctx->width, ctx->height, priv->tex_alignment);
+ if (size < 0)
+ return size;
ctx->internal->pool_internal = av_buffer_pool_init2(size, ctx, cuda_pool_alloc, NULL);
if (!ctx->internal->pool_internal)
static int cuda_get_buffer(AVHWFramesContext *ctx, AVFrame *frame)
{
- int aligned_width;
- int width_in_bytes = ctx->width;
-
- if (ctx->sw_format == AV_PIX_FMT_P010 ||
- ctx->sw_format == AV_PIX_FMT_P016 ||
- ctx->sw_format == AV_PIX_FMT_YUV444P16) {
- width_in_bytes *= 2;
- }
- aligned_width = FFALIGN(width_in_bytes, CUDA_FRAME_ALIGNMENT);
+ CUDAFramesContext *priv = ctx->internal->priv;
+ int res;
frame->buf[0] = av_buffer_pool_get(ctx->pool);
if (!frame->buf[0])
return AVERROR(ENOMEM);
- switch (ctx->sw_format) {
- case AV_PIX_FMT_NV12:
- case AV_PIX_FMT_P010:
- case AV_PIX_FMT_P016:
- frame->data[0] = frame->buf[0]->data;
- frame->data[1] = frame->data[0] + aligned_width * ctx->height;
- frame->linesize[0] = aligned_width;
- frame->linesize[1] = aligned_width;
- break;
- case AV_PIX_FMT_YUV420P:
- frame->data[0] = frame->buf[0]->data;
- frame->data[2] = frame->data[0] + aligned_width * ctx->height;
- frame->data[1] = frame->data[2] + aligned_width * ctx->height / 4;
- frame->linesize[0] = aligned_width;
- frame->linesize[1] = aligned_width / 2;
- frame->linesize[2] = aligned_width / 2;
- break;
- case AV_PIX_FMT_YUV444P:
- case AV_PIX_FMT_YUV444P16:
- frame->data[0] = frame->buf[0]->data;
- frame->data[1] = frame->data[0] + aligned_width * ctx->height;
- frame->data[2] = frame->data[1] + aligned_width * ctx->height;
- frame->linesize[0] = aligned_width;
- frame->linesize[1] = aligned_width;
- frame->linesize[2] = aligned_width;
- break;
- default:
- av_frame_unref(frame);
- return AVERROR_BUG;
+ res = av_image_fill_arrays(frame->data, frame->linesize, frame->buf[0]->data,
+ ctx->sw_format, ctx->width, ctx->height, priv->tex_alignment);
+ if (res < 0)
+ return res;
+
+ // YUV420P is a special case.
+ // Nvenc expects the U/V planes in swapped order from how ffmpeg expects them, also chroma is half-aligned
+ if (ctx->sw_format == AV_PIX_FMT_YUV420P) {
+ frame->linesize[1] = frame->linesize[2] = frame->linesize[0] / 2;
+ frame->data[2] = frame->data[1];
+ frame->data[1] = frame->data[2] + frame->linesize[2] * (ctx->height / 2);
}
frame->format = AV_PIX_FMT_CUDA;
return 0;
}
-static int cuda_transfer_data_from(AVHWFramesContext *ctx, AVFrame *dst,
- const AVFrame *src)
+static int cuda_transfer_data(AVHWFramesContext *ctx, AVFrame *dst,
+ const AVFrame *src)
{
- CUDAFramesContext *priv = ctx->internal->priv;
- AVCUDADeviceContext *device_hwctx = ctx->device_ctx->hwctx;
- CudaFunctions *cu = device_hwctx->internal->cuda_dl;
+ CUDAFramesContext *priv = ctx->internal->priv;
+ AVHWDeviceContext *device_ctx = ctx->device_ctx;
+ AVCUDADeviceContext *hwctx = device_ctx->hwctx;
+ CudaFunctions *cu = hwctx->internal->cuda_dl;
CUcontext dummy;
- CUresult err;
- int i;
+ int i, ret;
+
+ if ((src->hw_frames_ctx && ((AVHWFramesContext*)src->hw_frames_ctx->data)->format != AV_PIX_FMT_CUDA) ||
+ (dst->hw_frames_ctx && ((AVHWFramesContext*)dst->hw_frames_ctx->data)->format != AV_PIX_FMT_CUDA))
+ return AVERROR(ENOSYS);
- err = cu->cuCtxPushCurrent(device_hwctx->cuda_ctx);
- if (err != CUDA_SUCCESS)
- return AVERROR_UNKNOWN;
+ ret = CHECK_CU(cu->cuCtxPushCurrent(hwctx->cuda_ctx));
+ if (ret < 0)
+ return ret;
for (i = 0; i < FF_ARRAY_ELEMS(src->data) && src->data[i]; i++) {
CUDA_MEMCPY2D cpy = {
- .srcMemoryType = CU_MEMORYTYPE_DEVICE,
- .dstMemoryType = CU_MEMORYTYPE_HOST,
- .srcDevice = (CUdeviceptr)src->data[i],
- .dstHost = dst->data[i],
.srcPitch = src->linesize[i],
.dstPitch = dst->linesize[i],
.WidthInBytes = FFMIN(src->linesize[i], dst->linesize[i]),
- .Height = src->height >> (i ? priv->shift_height : 0),
+ .Height = src->height >> ((i == 0 || i == 3) ? 0 : priv->shift_height),
};
- err = cu->cuMemcpy2D(&cpy);
- if (err != CUDA_SUCCESS) {
- av_log(ctx, AV_LOG_ERROR, "Error transferring the data from the CUDA frame\n");
- return AVERROR_UNKNOWN;
+ if (src->hw_frames_ctx) {
+ cpy.srcMemoryType = CU_MEMORYTYPE_DEVICE;
+ cpy.srcDevice = (CUdeviceptr)src->data[i];
+ } else {
+ cpy.srcMemoryType = CU_MEMORYTYPE_HOST;
+ cpy.srcHost = src->data[i];
}
- }
-
- cu->cuCtxPopCurrent(&dummy);
- return 0;
-}
-
-static int cuda_transfer_data_to(AVHWFramesContext *ctx, AVFrame *dst,
- const AVFrame *src)
-{
- CUDAFramesContext *priv = ctx->internal->priv;
- AVCUDADeviceContext *device_hwctx = ctx->device_ctx->hwctx;
- CudaFunctions *cu = device_hwctx->internal->cuda_dl;
-
- CUcontext dummy;
- CUresult err;
- int i;
-
- err = cu->cuCtxPushCurrent(device_hwctx->cuda_ctx);
- if (err != CUDA_SUCCESS)
- return AVERROR_UNKNOWN;
+ if (dst->hw_frames_ctx) {
+ cpy.dstMemoryType = CU_MEMORYTYPE_DEVICE;
+ cpy.dstDevice = (CUdeviceptr)dst->data[i];
+ } else {
+ cpy.dstMemoryType = CU_MEMORYTYPE_HOST;
+ cpy.dstHost = dst->data[i];
+ }
- for (i = 0; i < FF_ARRAY_ELEMS(src->data) && src->data[i]; i++) {
- CUDA_MEMCPY2D cpy = {
- .srcMemoryType = CU_MEMORYTYPE_HOST,
- .dstMemoryType = CU_MEMORYTYPE_DEVICE,
- .srcHost = src->data[i],
- .dstDevice = (CUdeviceptr)dst->data[i],
- .srcPitch = src->linesize[i],
- .dstPitch = dst->linesize[i],
- .WidthInBytes = FFMIN(src->linesize[i], dst->linesize[i]),
- .Height = src->height >> (i ? priv->shift_height : 0),
- };
+ ret = CHECK_CU(cu->cuMemcpy2DAsync(&cpy, hwctx->stream));
+ if (ret < 0)
+ goto exit;
+ }
- err = cu->cuMemcpy2D(&cpy);
- if (err != CUDA_SUCCESS) {
- av_log(ctx, AV_LOG_ERROR, "Error transferring the data from the CUDA frame\n");
- return AVERROR_UNKNOWN;
- }
+ if (!dst->hw_frames_ctx) {
+ ret = CHECK_CU(cu->cuStreamSynchronize(hwctx->stream));
+ if (ret < 0)
+ goto exit;
}
- cu->cuCtxPopCurrent(&dummy);
+exit:
+ CHECK_CU(cu->cuCtxPopCurrent(&dummy));
return 0;
}
-static void cuda_device_uninit(AVHWDeviceContext *ctx)
+static void cuda_device_uninit(AVHWDeviceContext *device_ctx)
{
- AVCUDADeviceContext *hwctx = ctx->hwctx;
+ AVCUDADeviceContext *hwctx = device_ctx->hwctx;
if (hwctx->internal) {
+ CudaFunctions *cu = hwctx->internal->cuda_dl;
+
if (hwctx->internal->is_allocated && hwctx->cuda_ctx) {
- hwctx->internal->cuda_dl->cuCtxDestroy(hwctx->cuda_ctx);
+ if (hwctx->internal->flags & AV_CUDA_USE_PRIMARY_CONTEXT)
+ CHECK_CU(cu->cuDevicePrimaryCtxRelease(hwctx->internal->cuda_device));
+ else
+ CHECK_CU(cu->cuCtxDestroy(hwctx->cuda_ctx));
+
hwctx->cuda_ctx = NULL;
}
+
cuda_free_functions(&hwctx->internal->cuda_dl);
}
return ret;
}
-static int cuda_device_create(AVHWDeviceContext *ctx, const char *device,
+static int cuda_context_init(AVHWDeviceContext *device_ctx, int flags) {
+ AVCUDADeviceContext *hwctx = device_ctx->hwctx;
+ CudaFunctions *cu;
+ CUcontext dummy;
+ int ret, dev_active = 0;
+ unsigned int dev_flags = 0;
+
+ const unsigned int desired_flags = CU_CTX_SCHED_BLOCKING_SYNC;
+
+ cu = hwctx->internal->cuda_dl;
+
+ hwctx->internal->flags = flags;
+
+ if (flags & AV_CUDA_USE_PRIMARY_CONTEXT) {
+ ret = CHECK_CU(cu->cuDevicePrimaryCtxGetState(hwctx->internal->cuda_device,
+ &dev_flags, &dev_active));
+ if (ret < 0)
+ return ret;
+
+ if (dev_active && dev_flags != desired_flags) {
+ av_log(device_ctx, AV_LOG_ERROR, "Primary context already active with incompatible flags.\n");
+ return AVERROR(ENOTSUP);
+ } else if (dev_flags != desired_flags) {
+ ret = CHECK_CU(cu->cuDevicePrimaryCtxSetFlags(hwctx->internal->cuda_device,
+ desired_flags));
+ if (ret < 0)
+ return ret;
+ }
+
+ ret = CHECK_CU(cu->cuDevicePrimaryCtxRetain(&hwctx->cuda_ctx,
+ hwctx->internal->cuda_device));
+ if (ret < 0)
+ return ret;
+ } else {
+ ret = CHECK_CU(cu->cuCtxCreate(&hwctx->cuda_ctx, desired_flags,
+ hwctx->internal->cuda_device));
+ if (ret < 0)
+ return ret;
+
+ CHECK_CU(cu->cuCtxPopCurrent(&dummy));
+ }
+
+ hwctx->internal->is_allocated = 1;
+
+ // Setting stream to NULL will make functions automatically use the default CUstream
+ hwctx->stream = NULL;
+
+ return 0;
+}
+
+static int cuda_device_create(AVHWDeviceContext *device_ctx,
+ const char *device,
AVDictionary *opts, int flags)
{
- AVCUDADeviceContext *hwctx = ctx->hwctx;
+ AVCUDADeviceContext *hwctx = device_ctx->hwctx;
CudaFunctions *cu;
- CUdevice cu_device;
- CUcontext dummy;
- CUresult err;
- int device_idx = 0;
+ int ret, device_idx = 0;
if (device)
device_idx = strtol(device, NULL, 0);
- if (cuda_device_init(ctx) < 0)
+ if (cuda_device_init(device_ctx) < 0)
goto error;
cu = hwctx->internal->cuda_dl;
- err = cu->cuInit(0);
- if (err != CUDA_SUCCESS) {
- av_log(ctx, AV_LOG_ERROR, "Could not initialize the CUDA driver API\n");
+ ret = CHECK_CU(cu->cuInit(0));
+ if (ret < 0)
+ goto error;
+
+ ret = CHECK_CU(cu->cuDeviceGet(&hwctx->internal->cuda_device, device_idx));
+ if (ret < 0)
goto error;
+
+ ret = cuda_context_init(device_ctx, flags);
+ if (ret < 0)
+ goto error;
+
+ return 0;
+
+error:
+ cuda_device_uninit(device_ctx);
+ return AVERROR_UNKNOWN;
+}
+
+static int cuda_device_derive(AVHWDeviceContext *device_ctx,
+ AVHWDeviceContext *src_ctx, AVDictionary *opts,
+ int flags) {
+ AVCUDADeviceContext *hwctx = device_ctx->hwctx;
+ CudaFunctions *cu;
+ const char *src_uuid = NULL;
+ int ret, i, device_count;
+
+#if CONFIG_VULKAN
+ VkPhysicalDeviceIDProperties vk_idp = {
+ .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ID_PROPERTIES,
+ };
+#endif
+
+ switch (src_ctx->type) {
+#if CONFIG_VULKAN
+ case AV_HWDEVICE_TYPE_VULKAN: {
+ AVVulkanDeviceContext *vkctx = src_ctx->hwctx;
+ VkPhysicalDeviceProperties2 vk_dev_props = {
+ .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2,
+ .pNext = &vk_idp,
+ };
+ vkGetPhysicalDeviceProperties2(vkctx->phys_dev, &vk_dev_props);
+ src_uuid = vk_idp.deviceUUID;
+ break;
+ }
+#endif
+ default:
+ return AVERROR(ENOSYS);
}
- err = cu->cuDeviceGet(&cu_device, device_idx);
- if (err != CUDA_SUCCESS) {
- av_log(ctx, AV_LOG_ERROR, "Could not get the device number %d\n", device_idx);
+ if (!src_uuid) {
+ av_log(device_ctx, AV_LOG_ERROR,
+ "Failed to get UUID of source device.\n");
goto error;
}
- err = cu->cuCtxCreate(&hwctx->cuda_ctx, CU_CTX_SCHED_BLOCKING_SYNC, cu_device);
- if (err != CUDA_SUCCESS) {
- av_log(ctx, AV_LOG_ERROR, "Error creating a CUDA context\n");
+ if (cuda_device_init(device_ctx) < 0)
goto error;
+
+ cu = hwctx->internal->cuda_dl;
+
+ ret = CHECK_CU(cu->cuInit(0));
+ if (ret < 0)
+ goto error;
+
+ ret = CHECK_CU(cu->cuDeviceGetCount(&device_count));
+ if (ret < 0)
+ goto error;
+
+ hwctx->internal->cuda_device = -1;
+ for (i = 0; i < device_count; i++) {
+ CUdevice dev;
+ CUuuid uuid;
+
+ ret = CHECK_CU(cu->cuDeviceGet(&dev, i));
+ if (ret < 0)
+ goto error;
+
+ ret = CHECK_CU(cu->cuDeviceGetUuid(&uuid, dev));
+ if (ret < 0)
+ goto error;
+
+ if (memcmp(src_uuid, uuid.bytes, sizeof (uuid.bytes)) == 0) {
+ hwctx->internal->cuda_device = dev;
+ break;
+ }
}
- cu->cuCtxPopCurrent(&dummy);
+ if (hwctx->internal->cuda_device == -1) {
+ av_log(device_ctx, AV_LOG_ERROR, "Could not derive CUDA device.\n");
+ goto error;
+ }
- hwctx->internal->is_allocated = 1;
+ ret = cuda_context_init(device_ctx, flags);
+ if (ret < 0)
+ goto error;
return 0;
error:
- cuda_device_uninit(ctx);
+ cuda_device_uninit(device_ctx);
return AVERROR_UNKNOWN;
}
.frames_priv_size = sizeof(CUDAFramesContext),
.device_create = cuda_device_create,
+ .device_derive = cuda_device_derive,
.device_init = cuda_device_init,
.device_uninit = cuda_device_uninit,
.frames_get_constraints = cuda_frames_get_constraints,
.frames_init = cuda_frames_init,
.frames_get_buffer = cuda_get_buffer,
.transfer_get_formats = cuda_transfer_get_formats,
- .transfer_data_to = cuda_transfer_data_to,
- .transfer_data_from = cuda_transfer_data_from,
+ .transfer_data_to = cuda_transfer_data,
+ .transfer_data_from = cuda_transfer_data,
.pix_fmts = (const enum AVPixelFormat[]){ AV_PIX_FMT_CUDA, AV_PIX_FMT_NONE },
};