RepoMirrors
/
mpv
mirror of https://github.com/mpv-player/mpv
1
0
Fork 0
Code Issues Releases Wiki Activity
mpv/video/out/opengl/hwdec_cuda.c

350 lines
9.7 KiB
C
Raw Normal View History

hwdec/opengl: Add support for CUDA and cuvid/NvDecode Nvidia's "NvDecode" API (up until recently called "cuvid" is a cross platform, but nvidia proprietary API that exposes their hardware video decoding capabilities. It is analogous to their DXVA or VDPAU support on Windows or Linux but without using platform specific API calls. As a rule, you'd rather use DXVA or VDPAU as these are more mature and well supported APIs, but on Linux, VDPAU is falling behind the hardware capabilities, and there's no sign that nvidia are making the investments to update it. Most concretely, this means that there is no VP8/9 or HEVC Main10 support in VDPAU. On the other hand, NvDecode does export vp8/9 and partial support for HEVC Main10 (more on that below). ffmpeg already has support in the form of the "cuvid" family of decoders. Due to the design of the API, it is best exposed as a full decoder rather than an hwaccel. As such, there are decoders like h264_cuvid, hevc_cuvid, etc. These decoders support two output paths today - in both cases, NV12 frames are returned, either in CUDA device memory or regular system memory. In the case of the system memory path, the decoders can be used as-is in mpv today with a command line like: mpv --vd=lavc:h264_cuvid foobar.mp4 Doing this will take advantage of hardware decoding, but the cost of the memcpy to system memory adds up, especially for high resolution video (4K etc). To avoid that, we need an hwdec that takes advantage of CUDA's OpenGL interop to copy from device memory into OpenGL textures. That is what this change implements. The process is relatively simple as only basic device context aquisition needs to be done by us - the CUDA buffer pool is managed by the decoder - thankfully. The hwdec looks a bit like the vdpau interop one - the hwdec maintains a single set of plane textures and each output frame is repeatedly mapped into these textures to pass on. The frames are always in NV12 format, at least until 10bit output supports emerges. The only slightly interesting part of the copying process is that CUDA works by associating PBOs, so we need to define these for each of the textures. TODO Items: * I need to add a download_image function for screenshots. This would do the same copy to system memory that the decoder's system memory output does. * There are items to investigate on the ffmpeg side. There appears to be a problem with timestamps for some content. Final note: I mentioned HEVC Main10. While there is no 10bit output support, NvDecode can return dithered 8bit NV12 so you can take advantage of the hardware acceleration. This particular mode requires compiling ffmpeg with a modified header (or possibly the CUDA 8 RC) and is not upstream in ffmpeg yet. Usage: You will need to specify vo=opengl and hwdec=cuda. Note that hwdec=auto will probably not work as it will try to use vdpau first. mpv --hwdec=cuda --vo=opengl foobar.mp4 If you want to use filters that require frames in system memory, just use the decoder directly without the hwdec, as documented above.
2016-09-04 22:23:55 +00:00
/*
* Copyright (c) 2016 Philip Langdale <philipl@overt.org>
*
* This file is part of mpv.
*
* mpv is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* mpv is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with mpv. If not, see <http://www.gnu.org/licenses/>.
*/
/*
* This hwdec implements an optimized output path using CUDA->OpenGL
* interop for frame data that is stored in CUDA device memory.
* Although it is not explicit in the code here, the only practical way
* to get data in this form is from the 'cuvid' decoder (aka NvDecode).
*
* For now, cuvid/NvDecode will always return images in NV12 format, even
* when decoding 10bit streams (there is some hardware dithering going on).
*/
#include <libavutil/hwcontext.h>
vo_opengl: hwdec_cuda: Use dynamic loading for cuda functions This change applies the pattern used in ffmpeg to dynamically load cuda, to avoid requiring the CUDA SDK at build time.
2016-10-08 23:51:15 +00:00
#include "cuda_dynamic.h"
hwdec_cuda: Implement download_image for screenshots Data has to be copied to system memory for screenshots.
2016-09-10 20:50:38 +00:00
#include "video/mp_image_pool.h"
hwdec/opengl: Add support for CUDA and cuvid/NvDecode Nvidia's "NvDecode" API (up until recently called "cuvid" is a cross platform, but nvidia proprietary API that exposes their hardware video decoding capabilities. It is analogous to their DXVA or VDPAU support on Windows or Linux but without using platform specific API calls. As a rule, you'd rather use DXVA or VDPAU as these are more mature and well supported APIs, but on Linux, VDPAU is falling behind the hardware capabilities, and there's no sign that nvidia are making the investments to update it. Most concretely, this means that there is no VP8/9 or HEVC Main10 support in VDPAU. On the other hand, NvDecode does export vp8/9 and partial support for HEVC Main10 (more on that below). ffmpeg already has support in the form of the "cuvid" family of decoders. Due to the design of the API, it is best exposed as a full decoder rather than an hwaccel. As such, there are decoders like h264_cuvid, hevc_cuvid, etc. These decoders support two output paths today - in both cases, NV12 frames are returned, either in CUDA device memory or regular system memory. In the case of the system memory path, the decoders can be used as-is in mpv today with a command line like: mpv --vd=lavc:h264_cuvid foobar.mp4 Doing this will take advantage of hardware decoding, but the cost of the memcpy to system memory adds up, especially for high resolution video (4K etc). To avoid that, we need an hwdec that takes advantage of CUDA's OpenGL interop to copy from device memory into OpenGL textures. That is what this change implements. The process is relatively simple as only basic device context aquisition needs to be done by us - the CUDA buffer pool is managed by the decoder - thankfully. The hwdec looks a bit like the vdpau interop one - the hwdec maintains a single set of plane textures and each output frame is repeatedly mapped into these textures to pass on. The frames are always in NV12 format, at least until 10bit output supports emerges. The only slightly interesting part of the copying process is that CUDA works by associating PBOs, so we need to define these for each of the textures. TODO Items: * I need to add a download_image function for screenshots. This would do the same copy to system memory that the decoder's system memory output does. * There are items to investigate on the ffmpeg side. There appears to be a problem with timestamps for some content. Final note: I mentioned HEVC Main10. While there is no 10bit output support, NvDecode can return dithered 8bit NV12 so you can take advantage of the hardware acceleration. This particular mode requires compiling ffmpeg with a modified header (or possibly the CUDA 8 RC) and is not upstream in ffmpeg yet. Usage: You will need to specify vo=opengl and hwdec=cuda. Note that hwdec=auto will probably not work as it will try to use vdpau first. mpv --hwdec=cuda --vo=opengl foobar.mp4 If you want to use filters that require frames in system memory, just use the decoder directly without the hwdec, as documented above.
2016-09-04 22:23:55 +00:00
#include "hwdec.h"
#include "video.h"
vo_opengl: hwdec_cuda: Use dynamic loading for cuda functions This change applies the pattern used in ffmpeg to dynamically load cuda, to avoid requiring the CUDA SDK at build time.
2016-10-08 23:51:15 +00:00
#include <libavutil/hwcontext_cuda.h>
hwdec/opengl: Add support for CUDA and cuvid/NvDecode Nvidia's "NvDecode" API (up until recently called "cuvid" is a cross platform, but nvidia proprietary API that exposes their hardware video decoding capabilities. It is analogous to their DXVA or VDPAU support on Windows or Linux but without using platform specific API calls. As a rule, you'd rather use DXVA or VDPAU as these are more mature and well supported APIs, but on Linux, VDPAU is falling behind the hardware capabilities, and there's no sign that nvidia are making the investments to update it. Most concretely, this means that there is no VP8/9 or HEVC Main10 support in VDPAU. On the other hand, NvDecode does export vp8/9 and partial support for HEVC Main10 (more on that below). ffmpeg already has support in the form of the "cuvid" family of decoders. Due to the design of the API, it is best exposed as a full decoder rather than an hwaccel. As such, there are decoders like h264_cuvid, hevc_cuvid, etc. These decoders support two output paths today - in both cases, NV12 frames are returned, either in CUDA device memory or regular system memory. In the case of the system memory path, the decoders can be used as-is in mpv today with a command line like: mpv --vd=lavc:h264_cuvid foobar.mp4 Doing this will take advantage of hardware decoding, but the cost of the memcpy to system memory adds up, especially for high resolution video (4K etc). To avoid that, we need an hwdec that takes advantage of CUDA's OpenGL interop to copy from device memory into OpenGL textures. That is what this change implements. The process is relatively simple as only basic device context aquisition needs to be done by us - the CUDA buffer pool is managed by the decoder - thankfully. The hwdec looks a bit like the vdpau interop one - the hwdec maintains a single set of plane textures and each output frame is repeatedly mapped into these textures to pass on. The frames are always in NV12 format, at least until 10bit output supports emerges. The only slightly interesting part of the copying process is that CUDA works by associating PBOs, so we need to define these for each of the textures. TODO Items: * I need to add a download_image function for screenshots. This would do the same copy to system memory that the decoder's system memory output does. * There are items to investigate on the ffmpeg side. There appears to be a problem with timestamps for some content. Final note: I mentioned HEVC Main10. While there is no 10bit output support, NvDecode can return dithered 8bit NV12 so you can take advantage of the hardware acceleration. This particular mode requires compiling ffmpeg with a modified header (or possibly the CUDA 8 RC) and is not upstream in ffmpeg yet. Usage: You will need to specify vo=opengl and hwdec=cuda. Note that hwdec=auto will probably not work as it will try to use vdpau first. mpv --hwdec=cuda --vo=opengl foobar.mp4 If you want to use filters that require frames in system memory, just use the decoder directly without the hwdec, as documented above.
2016-09-04 22:23:55 +00:00
struct priv {
struct mp_hwdec_ctx hwctx;
struct mp_image layout;
GLuint gl_textures[2];
hwdec_cuda: Use the non-deprecated CUDA-GL interop API The nvidia examples use the old (as in CUDA 3.x) interop API which is deprecated, and I think not even functional on recent versions of CUDA for windows. As I was following the examples, I used this old API. So, let's update to the new API, and hopefully, it'll start working on windows too.
2016-09-10 03:07:43 +00:00
CUgraphicsResource cu_res[2];
vo_opengl: hwdec_cuda: directly map GL textures and skip using PBOs The documentation around this stuff is poor, but I found an nvidia sample that demonstrates how to use the interop API most efficiently. (https://github.com/nvpro-samples/gl_cuda_interop_pingpong_st) Key lessons are: 1) you can register the texture itself and have cuda write to it, thereby skipping an additional copy through the PBO. 2) You don't have to be mapped when you do the copy - once you get a mapped pointer, it remains valid. Magic! This lets us throw out the PBOs as well as much of the explicit alignment and stride handling. CPU usage is slightly (~3%) lower for 4K content in one test case, so it makes a detectable difference, and presumably saves memory.
2016-09-23 04:39:40 +00:00
CUarray cu_array[2];
vo_opengl: hwdec_cuda: Support P016 output surfaces The latest 375.xx nvidia drivers add support for P016 output surfaces. In combination with an ffmpeg change to return those surfaces, we can display them. The bulk of the work is related to knowing which format you're dealing with at the right time. Once you know, it's straight forward.
2016-11-19 21:57:23 +00:00
int sample_width;
hwdec/opengl: Add support for CUDA and cuvid/NvDecode Nvidia's "NvDecode" API (up until recently called "cuvid" is a cross platform, but nvidia proprietary API that exposes their hardware video decoding capabilities. It is analogous to their DXVA or VDPAU support on Windows or Linux but without using platform specific API calls. As a rule, you'd rather use DXVA or VDPAU as these are more mature and well supported APIs, but on Linux, VDPAU is falling behind the hardware capabilities, and there's no sign that nvidia are making the investments to update it. Most concretely, this means that there is no VP8/9 or HEVC Main10 support in VDPAU. On the other hand, NvDecode does export vp8/9 and partial support for HEVC Main10 (more on that below). ffmpeg already has support in the form of the "cuvid" family of decoders. Due to the design of the API, it is best exposed as a full decoder rather than an hwaccel. As such, there are decoders like h264_cuvid, hevc_cuvid, etc. These decoders support two output paths today - in both cases, NV12 frames are returned, either in CUDA device memory or regular system memory. In the case of the system memory path, the decoders can be used as-is in mpv today with a command line like: mpv --vd=lavc:h264_cuvid foobar.mp4 Doing this will take advantage of hardware decoding, but the cost of the memcpy to system memory adds up, especially for high resolution video (4K etc). To avoid that, we need an hwdec that takes advantage of CUDA's OpenGL interop to copy from device memory into OpenGL textures. That is what this change implements. The process is relatively simple as only basic device context aquisition needs to be done by us - the CUDA buffer pool is managed by the decoder - thankfully. The hwdec looks a bit like the vdpau interop one - the hwdec maintains a single set of plane textures and each output frame is repeatedly mapped into these textures to pass on. The frames are always in NV12 format, at least until 10bit output supports emerges. The only slightly interesting part of the copying process is that CUDA works by associating PBOs, so we need to define these for each of the textures. TODO Items: * I need to add a download_image function for screenshots. This would do the same copy to system memory that the decoder's system memory output does. * There are items to investigate on the ffmpeg side. There appears to be a problem with timestamps for some content. Final note: I mentioned HEVC Main10. While there is no 10bit output support, NvDecode can return dithered 8bit NV12 so you can take advantage of the hardware acceleration. This particular mode requires compiling ffmpeg with a modified header (or possibly the CUDA 8 RC) and is not upstream in ffmpeg yet. Usage: You will need to specify vo=opengl and hwdec=cuda. Note that hwdec=auto will probably not work as it will try to use vdpau first. mpv --hwdec=cuda --vo=opengl foobar.mp4 If you want to use filters that require frames in system memory, just use the decoder directly without the hwdec, as documented above.
2016-09-04 22:23:55 +00:00
CUcontext cuda_ctx;
};
static int check_cu(struct gl_hwdec *hw, CUresult err, const char *func)
{
const char *err_name;
const char *err_string;
MP_TRACE(hw, "Calling %s\n", func);
if (err == CUDA_SUCCESS)
return 0;
cuGetErrorName(err, &err_name);
cuGetErrorString(err, &err_string);
MP_ERR(hw, "%s failed", func);
if (err_name && err_string)
MP_ERR(hw, " -> %s: %s", err_name, err_string);
MP_ERR(hw, "\n");
return -1;
}
#define CHECK_CU(x) check_cu(hw, (x), #x)
hwdec_cuda: Implement download_image for screenshots Data has to be copied to system memory for screenshots.
2016-09-10 20:50:38 +00:00
static struct mp_image *cuda_download_image(struct mp_hwdec_ctx *ctx,
struct mp_image *hw_image,
struct mp_image_pool *swpool)
{
CUcontext cuda_ctx = ctx->ctx;
CUcontext dummy;
CUresult err, eerr;
if (hw_image->imgfmt != IMGFMT_CUDA)
return NULL;
vo_opengl: hwdec_cuda: Support P016 output surfaces The latest 375.xx nvidia drivers add support for P016 output surfaces. In combination with an ffmpeg change to return those surfaces, we can display them. The bulk of the work is related to knowing which format you're dealing with at the right time. Once you know, it's straight forward.
2016-11-19 21:57:23 +00:00
int sample_width;
switch (hw_image->params.hw_subfmt) {
case IMGFMT_NV12:
sample_width = 1;
break;
case IMGFMT_P010:
case IMGFMT_P016:
sample_width = 2;
break;
default:
return NULL;
}
struct mp_image *out = mp_image_pool_get(swpool,
hw_image->params.hw_subfmt,
hwdec_cuda: Implement download_image for screenshots Data has to be copied to system memory for screenshots.
2016-09-10 20:50:38 +00:00
hw_image->w, hw_image->h);
if (!out)
return NULL;
err = cuCtxPushCurrent(cuda_ctx);
if (err != CUDA_SUCCESS)
goto error;
mp_image_set_size(out, hw_image->w, hw_image->h);
mp_image_copy_attributes(out, hw_image);
for (int n = 0; n < 2; n++) {
CUDA_MEMCPY2D cpy = {
.srcMemoryType = CU_MEMORYTYPE_DEVICE,
.dstMemoryType = CU_MEMORYTYPE_HOST,
.srcDevice = (CUdeviceptr)hw_image->planes[n],
.dstHost = out->planes[n],
.srcPitch = hw_image->stride[n],
.dstPitch = out->stride[n],
vo_opengl: hwdec_cuda: Support P016 output surfaces The latest 375.xx nvidia drivers add support for P016 output surfaces. In combination with an ffmpeg change to return those surfaces, we can display them. The bulk of the work is related to knowing which format you're dealing with at the right time. Once you know, it's straight forward.
2016-11-19 21:57:23 +00:00
.WidthInBytes = mp_image_plane_w(out, n) *
(n + 1) * sample_width,
hwdec_cuda: Implement download_image for screenshots Data has to be copied to system memory for screenshots.
2016-09-10 20:50:38 +00:00
.Height = mp_image_plane_h(out, n),
};
err = cuMemcpy2D(&cpy);
if (err != CUDA_SUCCESS) {
goto error;
}
}
error:
eerr = cuCtxPopCurrent(&dummy);
if (eerr != CUDA_SUCCESS || err != CUDA_SUCCESS) {
talloc_free(out);
return NULL;
}
return out;
}
hwdec/opengl: Add support for CUDA and cuvid/NvDecode Nvidia's "NvDecode" API (up until recently called "cuvid" is a cross platform, but nvidia proprietary API that exposes their hardware video decoding capabilities. It is analogous to their DXVA or VDPAU support on Windows or Linux but without using platform specific API calls. As a rule, you'd rather use DXVA or VDPAU as these are more mature and well supported APIs, but on Linux, VDPAU is falling behind the hardware capabilities, and there's no sign that nvidia are making the investments to update it. Most concretely, this means that there is no VP8/9 or HEVC Main10 support in VDPAU. On the other hand, NvDecode does export vp8/9 and partial support for HEVC Main10 (more on that below). ffmpeg already has support in the form of the "cuvid" family of decoders. Due to the design of the API, it is best exposed as a full decoder rather than an hwaccel. As such, there are decoders like h264_cuvid, hevc_cuvid, etc. These decoders support two output paths today - in both cases, NV12 frames are returned, either in CUDA device memory or regular system memory. In the case of the system memory path, the decoders can be used as-is in mpv today with a command line like: mpv --vd=lavc:h264_cuvid foobar.mp4 Doing this will take advantage of hardware decoding, but the cost of the memcpy to system memory adds up, especially for high resolution video (4K etc). To avoid that, we need an hwdec that takes advantage of CUDA's OpenGL interop to copy from device memory into OpenGL textures. That is what this change implements. The process is relatively simple as only basic device context aquisition needs to be done by us - the CUDA buffer pool is managed by the decoder - thankfully. The hwdec looks a bit like the vdpau interop one - the hwdec maintains a single set of plane textures and each output frame is repeatedly mapped into these textures to pass on. The frames are always in NV12 format, at least until 10bit output supports emerges. The only slightly interesting part of the copying process is that CUDA works by associating PBOs, so we need to define these for each of the textures. TODO Items: * I need to add a download_image function for screenshots. This would do the same copy to system memory that the decoder's system memory output does. * There are items to investigate on the ffmpeg side. There appears to be a problem with timestamps for some content. Final note: I mentioned HEVC Main10. While there is no 10bit output support, NvDecode can return dithered 8bit NV12 so you can take advantage of the hardware acceleration. This particular mode requires compiling ffmpeg with a modified header (or possibly the CUDA 8 RC) and is not upstream in ffmpeg yet. Usage: You will need to specify vo=opengl and hwdec=cuda. Note that hwdec=auto will probably not work as it will try to use vdpau first. mpv --hwdec=cuda --vo=opengl foobar.mp4 If you want to use filters that require frames in system memory, just use the decoder directly without the hwdec, as documented above.
2016-09-04 22:23:55 +00:00
static int cuda_create(struct gl_hwdec *hw)
{
CUdevice device;
CUcontext cuda_ctx = NULL;
CUcontext dummy;
vo_opengl: hwdec_cuda: get the cuda device from the GL context Obviously, in the vast majority of cases, there's only one device in the system, but doing this means we're more likely to get a usable device in the multi-device case. cuda would support decoding on one device and displaying on another but the peer memory handling is not transparent and I have no way to test it so I can't really write it.
2016-09-23 04:52:23 +00:00
unsigned int device_count;
hwdec/opengl: Add support for CUDA and cuvid/NvDecode Nvidia's "NvDecode" API (up until recently called "cuvid" is a cross platform, but nvidia proprietary API that exposes their hardware video decoding capabilities. It is analogous to their DXVA or VDPAU support on Windows or Linux but without using platform specific API calls. As a rule, you'd rather use DXVA or VDPAU as these are more mature and well supported APIs, but on Linux, VDPAU is falling behind the hardware capabilities, and there's no sign that nvidia are making the investments to update it. Most concretely, this means that there is no VP8/9 or HEVC Main10 support in VDPAU. On the other hand, NvDecode does export vp8/9 and partial support for HEVC Main10 (more on that below). ffmpeg already has support in the form of the "cuvid" family of decoders. Due to the design of the API, it is best exposed as a full decoder rather than an hwaccel. As such, there are decoders like h264_cuvid, hevc_cuvid, etc. These decoders support two output paths today - in both cases, NV12 frames are returned, either in CUDA device memory or regular system memory. In the case of the system memory path, the decoders can be used as-is in mpv today with a command line like: mpv --vd=lavc:h264_cuvid foobar.mp4 Doing this will take advantage of hardware decoding, but the cost of the memcpy to system memory adds up, especially for high resolution video (4K etc). To avoid that, we need an hwdec that takes advantage of CUDA's OpenGL interop to copy from device memory into OpenGL textures. That is what this change implements. The process is relatively simple as only basic device context aquisition needs to be done by us - the CUDA buffer pool is managed by the decoder - thankfully. The hwdec looks a bit like the vdpau interop one - the hwdec maintains a single set of plane textures and each output frame is repeatedly mapped into these textures to pass on. The frames are always in NV12 format, at least until 10bit output supports emerges. The only slightly interesting part of the copying process is that CUDA works by associating PBOs, so we need to define these for each of the textures. TODO Items: * I need to add a download_image function for screenshots. This would do the same copy to system memory that the decoder's system memory output does. * There are items to investigate on the ffmpeg side. There appears to be a problem with timestamps for some content. Final note: I mentioned HEVC Main10. While there is no 10bit output support, NvDecode can return dithered 8bit NV12 so you can take advantage of the hardware acceleration. This particular mode requires compiling ffmpeg with a modified header (or possibly the CUDA 8 RC) and is not upstream in ffmpeg yet. Usage: You will need to specify vo=opengl and hwdec=cuda. Note that hwdec=auto will probably not work as it will try to use vdpau first. mpv --hwdec=cuda --vo=opengl foobar.mp4 If you want to use filters that require frames in system memory, just use the decoder directly without the hwdec, as documented above.
2016-09-04 22:23:55 +00:00
int ret = 0, eret = 0;
if (hw->gl->version < 210 && hw->gl->es < 300) {
MP_ERR(hw, "need OpenGL >= 2.1 or OpenGL-ES >= 3.0\n");
return -1;
}
struct priv *p = talloc_zero(hw, struct priv);
hw->priv = p;
vo_opengl: hwdec_cuda: Use dynamic loading for cuda functions This change applies the pattern used in ffmpeg to dynamically load cuda, to avoid requiring the CUDA SDK at build time.
2016-10-08 23:51:15 +00:00
bool loaded = cuda_load();
if (!loaded) {
MP_ERR(hw, "Failed to load CUDA symbols\n");
vo_opengl: hwdec_cuda: fix crash when trying to use hwdec=cuda if cuda SDK is not present If CUDA SDK wasn't installed, mpv crashed immediately with the message "Failed to load CUDA symbols"
2016-11-24 12:11:44 +00:00
return -1;
vo_opengl: hwdec_cuda: Use dynamic loading for cuda functions This change applies the pattern used in ffmpeg to dynamically load cuda, to avoid requiring the CUDA SDK at build time.
2016-10-08 23:51:15 +00:00
}
hwdec/opengl: Add support for CUDA and cuvid/NvDecode Nvidia's "NvDecode" API (up until recently called "cuvid" is a cross platform, but nvidia proprietary API that exposes their hardware video decoding capabilities. It is analogous to their DXVA or VDPAU support on Windows or Linux but without using platform specific API calls. As a rule, you'd rather use DXVA or VDPAU as these are more mature and well supported APIs, but on Linux, VDPAU is falling behind the hardware capabilities, and there's no sign that nvidia are making the investments to update it. Most concretely, this means that there is no VP8/9 or HEVC Main10 support in VDPAU. On the other hand, NvDecode does export vp8/9 and partial support for HEVC Main10 (more on that below). ffmpeg already has support in the form of the "cuvid" family of decoders. Due to the design of the API, it is best exposed as a full decoder rather than an hwaccel. As such, there are decoders like h264_cuvid, hevc_cuvid, etc. These decoders support two output paths today - in both cases, NV12 frames are returned, either in CUDA device memory or regular system memory. In the case of the system memory path, the decoders can be used as-is in mpv today with a command line like: mpv --vd=lavc:h264_cuvid foobar.mp4 Doing this will take advantage of hardware decoding, but the cost of the memcpy to system memory adds up, especially for high resolution video (4K etc). To avoid that, we need an hwdec that takes advantage of CUDA's OpenGL interop to copy from device memory into OpenGL textures. That is what this change implements. The process is relatively simple as only basic device context aquisition needs to be done by us - the CUDA buffer pool is managed by the decoder - thankfully. The hwdec looks a bit like the vdpau interop one - the hwdec maintains a single set of plane textures and each output frame is repeatedly mapped into these textures to pass on. The frames are always in NV12 format, at least until 10bit output supports emerges. The only slightly interesting part of the copying process is that CUDA works by associating PBOs, so we need to define these for each of the textures. TODO Items: * I need to add a download_image function for screenshots. This would do the same copy to system memory that the decoder's system memory output does. * There are items to investigate on the ffmpeg side. There appears to be a problem with timestamps for some content. Final note: I mentioned HEVC Main10. While there is no 10bit output support, NvDecode can return dithered 8bit NV12 so you can take advantage of the hardware acceleration. This particular mode requires compiling ffmpeg with a modified header (or possibly the CUDA 8 RC) and is not upstream in ffmpeg yet. Usage: You will need to specify vo=opengl and hwdec=cuda. Note that hwdec=auto will probably not work as it will try to use vdpau first. mpv --hwdec=cuda --vo=opengl foobar.mp4 If you want to use filters that require frames in system memory, just use the decoder directly without the hwdec, as documented above.
2016-09-04 22:23:55 +00:00
ret = CHECK_CU(cuInit(0));
if (ret < 0)
goto error;
vo_opengl: hwdec_cuda: get the cuda device from the GL context Obviously, in the vast majority of cases, there's only one device in the system, but doing this means we're more likely to get a usable device in the multi-device case. cuda would support decoding on one device and displaying on another but the peer memory handling is not transparent and I have no way to test it so I can't really write it.
2016-09-23 04:52:23 +00:00
ret = CHECK_CU(cuGLGetDevices(&device_count, &device, 1,
CU_GL_DEVICE_LIST_ALL));
hwdec/opengl: Add support for CUDA and cuvid/NvDecode Nvidia's "NvDecode" API (up until recently called "cuvid" is a cross platform, but nvidia proprietary API that exposes their hardware video decoding capabilities. It is analogous to their DXVA or VDPAU support on Windows or Linux but without using platform specific API calls. As a rule, you'd rather use DXVA or VDPAU as these are more mature and well supported APIs, but on Linux, VDPAU is falling behind the hardware capabilities, and there's no sign that nvidia are making the investments to update it. Most concretely, this means that there is no VP8/9 or HEVC Main10 support in VDPAU. On the other hand, NvDecode does export vp8/9 and partial support for HEVC Main10 (more on that below). ffmpeg already has support in the form of the "cuvid" family of decoders. Due to the design of the API, it is best exposed as a full decoder rather than an hwaccel. As such, there are decoders like h264_cuvid, hevc_cuvid, etc. These decoders support two output paths today - in both cases, NV12 frames are returned, either in CUDA device memory or regular system memory. In the case of the system memory path, the decoders can be used as-is in mpv today with a command line like: mpv --vd=lavc:h264_cuvid foobar.mp4 Doing this will take advantage of hardware decoding, but the cost of the memcpy to system memory adds up, especially for high resolution video (4K etc). To avoid that, we need an hwdec that takes advantage of CUDA's OpenGL interop to copy from device memory into OpenGL textures. That is what this change implements. The process is relatively simple as only basic device context aquisition needs to be done by us - the CUDA buffer pool is managed by the decoder - thankfully. The hwdec looks a bit like the vdpau interop one - the hwdec maintains a single set of plane textures and each output frame is repeatedly mapped into these textures to pass on. The frames are always in NV12 format, at least until 10bit output supports emerges. The only slightly interesting part of the copying process is that CUDA works by associating PBOs, so we need to define these for each of the textures. TODO Items: * I need to add a download_image function for screenshots. This would do the same copy to system memory that the decoder's system memory output does. * There are items to investigate on the ffmpeg side. There appears to be a problem with timestamps for some content. Final note: I mentioned HEVC Main10. While there is no 10bit output support, NvDecode can return dithered 8bit NV12 so you can take advantage of the hardware acceleration. This particular mode requires compiling ffmpeg with a modified header (or possibly the CUDA 8 RC) and is not upstream in ffmpeg yet. Usage: You will need to specify vo=opengl and hwdec=cuda. Note that hwdec=auto will probably not work as it will try to use vdpau first. mpv --hwdec=cuda --vo=opengl foobar.mp4 If you want to use filters that require frames in system memory, just use the decoder directly without the hwdec, as documented above.
2016-09-04 22:23:55 +00:00
if (ret < 0)
goto error;
ret = CHECK_CU(cuCtxCreate(&cuda_ctx, CU_CTX_SCHED_BLOCKING_SYNC, device));
if (ret < 0)
goto error;
p->cuda_ctx = cuda_ctx;
p->hwctx = (struct mp_hwdec_ctx) {
.type = HWDEC_CUDA,
.ctx = cuda_ctx,
hwdec_cuda: Implement download_image for screenshots Data has to be copied to system memory for screenshots.
2016-09-10 20:50:38 +00:00
.download_image = cuda_download_image,
hwdec/opengl: Add support for CUDA and cuvid/NvDecode Nvidia's "NvDecode" API (up until recently called "cuvid" is a cross platform, but nvidia proprietary API that exposes their hardware video decoding capabilities. It is analogous to their DXVA or VDPAU support on Windows or Linux but without using platform specific API calls. As a rule, you'd rather use DXVA or VDPAU as these are more mature and well supported APIs, but on Linux, VDPAU is falling behind the hardware capabilities, and there's no sign that nvidia are making the investments to update it. Most concretely, this means that there is no VP8/9 or HEVC Main10 support in VDPAU. On the other hand, NvDecode does export vp8/9 and partial support for HEVC Main10 (more on that below). ffmpeg already has support in the form of the "cuvid" family of decoders. Due to the design of the API, it is best exposed as a full decoder rather than an hwaccel. As such, there are decoders like h264_cuvid, hevc_cuvid, etc. These decoders support two output paths today - in both cases, NV12 frames are returned, either in CUDA device memory or regular system memory. In the case of the system memory path, the decoders can be used as-is in mpv today with a command line like: mpv --vd=lavc:h264_cuvid foobar.mp4 Doing this will take advantage of hardware decoding, but the cost of the memcpy to system memory adds up, especially for high resolution video (4K etc). To avoid that, we need an hwdec that takes advantage of CUDA's OpenGL interop to copy from device memory into OpenGL textures. That is what this change implements. The process is relatively simple as only basic device context aquisition needs to be done by us - the CUDA buffer pool is managed by the decoder - thankfully. The hwdec looks a bit like the vdpau interop one - the hwdec maintains a single set of plane textures and each output frame is repeatedly mapped into these textures to pass on. The frames are always in NV12 format, at least until 10bit output supports emerges. The only slightly interesting part of the copying process is that CUDA works by associating PBOs, so we need to define these for each of the textures. TODO Items: * I need to add a download_image function for screenshots. This would do the same copy to system memory that the decoder's system memory output does. * There are items to investigate on the ffmpeg side. There appears to be a problem with timestamps for some content. Final note: I mentioned HEVC Main10. While there is no 10bit output support, NvDecode can return dithered 8bit NV12 so you can take advantage of the hardware acceleration. This particular mode requires compiling ffmpeg with a modified header (or possibly the CUDA 8 RC) and is not upstream in ffmpeg yet. Usage: You will need to specify vo=opengl and hwdec=cuda. Note that hwdec=auto will probably not work as it will try to use vdpau first. mpv --hwdec=cuda --vo=opengl foobar.mp4 If you want to use filters that require frames in system memory, just use the decoder directly without the hwdec, as documented above.
2016-09-04 22:23:55 +00:00
};
p->hwctx.driver_name = hw->driver->name;
hwdec_devices_add(hw->devs, &p->hwctx);
error:
eret = CHECK_CU(cuCtxPopCurrent(&dummy));
if (eret < 0)
return eret;
return ret;
}
static int reinit(struct gl_hwdec *hw, struct mp_image_params *params)
{
struct priv *p = hw->priv;
GL *gl = hw->gl;
CUcontext dummy;
int ret = 0, eret = 0;
assert(params->imgfmt == hw->driver->imgfmt);
vo_opengl: hwdec_cuda: Support P016 output surfaces The latest 375.xx nvidia drivers add support for P016 output surfaces. In combination with an ffmpeg change to return those surfaces, we can display them. The bulk of the work is related to knowing which format you're dealing with at the right time. Once you know, it's straight forward.
2016-11-19 21:57:23 +00:00
params->imgfmt = params->hw_subfmt;
hwdec/opengl: Add support for CUDA and cuvid/NvDecode Nvidia's "NvDecode" API (up until recently called "cuvid" is a cross platform, but nvidia proprietary API that exposes their hardware video decoding capabilities. It is analogous to their DXVA or VDPAU support on Windows or Linux but without using platform specific API calls. As a rule, you'd rather use DXVA or VDPAU as these are more mature and well supported APIs, but on Linux, VDPAU is falling behind the hardware capabilities, and there's no sign that nvidia are making the investments to update it. Most concretely, this means that there is no VP8/9 or HEVC Main10 support in VDPAU. On the other hand, NvDecode does export vp8/9 and partial support for HEVC Main10 (more on that below). ffmpeg already has support in the form of the "cuvid" family of decoders. Due to the design of the API, it is best exposed as a full decoder rather than an hwaccel. As such, there are decoders like h264_cuvid, hevc_cuvid, etc. These decoders support two output paths today - in both cases, NV12 frames are returned, either in CUDA device memory or regular system memory. In the case of the system memory path, the decoders can be used as-is in mpv today with a command line like: mpv --vd=lavc:h264_cuvid foobar.mp4 Doing this will take advantage of hardware decoding, but the cost of the memcpy to system memory adds up, especially for high resolution video (4K etc). To avoid that, we need an hwdec that takes advantage of CUDA's OpenGL interop to copy from device memory into OpenGL textures. That is what this change implements. The process is relatively simple as only basic device context aquisition needs to be done by us - the CUDA buffer pool is managed by the decoder - thankfully. The hwdec looks a bit like the vdpau interop one - the hwdec maintains a single set of plane textures and each output frame is repeatedly mapped into these textures to pass on. The frames are always in NV12 format, at least until 10bit output supports emerges. The only slightly interesting part of the copying process is that CUDA works by associating PBOs, so we need to define these for each of the textures. TODO Items: * I need to add a download_image function for screenshots. This would do the same copy to system memory that the decoder's system memory output does. * There are items to investigate on the ffmpeg side. There appears to be a problem with timestamps for some content. Final note: I mentioned HEVC Main10. While there is no 10bit output support, NvDecode can return dithered 8bit NV12 so you can take advantage of the hardware acceleration. This particular mode requires compiling ffmpeg with a modified header (or possibly the CUDA 8 RC) and is not upstream in ffmpeg yet. Usage: You will need to specify vo=opengl and hwdec=cuda. Note that hwdec=auto will probably not work as it will try to use vdpau first. mpv --hwdec=cuda --vo=opengl foobar.mp4 If you want to use filters that require frames in system memory, just use the decoder directly without the hwdec, as documented above.
2016-09-04 22:23:55 +00:00
params->hw_subfmt = 0;
mp_image_set_params(&p->layout, params);
vo_opengl: hwdec_cuda: Support P016 output surfaces The latest 375.xx nvidia drivers add support for P016 output surfaces. In combination with an ffmpeg change to return those surfaces, we can display them. The bulk of the work is related to knowing which format you're dealing with at the right time. Once you know, it's straight forward.
2016-11-19 21:57:23 +00:00
GLint luma_format, chroma_format;
GLenum type;
switch (params->imgfmt) {
case IMGFMT_NV12:
luma_format = GL_R8;
chroma_format = GL_RG8;
type = GL_UNSIGNED_BYTE;
p->sample_width = 1;
break;
case IMGFMT_P010:
case IMGFMT_P016:
luma_format = GL_R16;
chroma_format = GL_RG16;
type = GL_UNSIGNED_SHORT;
p->sample_width = 2;
break;
default:
MP_ERR(hw, "Unsupported format: %s\n", mp_imgfmt_to_name(params->imgfmt));
return -1;
}
hwdec/opengl: Add support for CUDA and cuvid/NvDecode Nvidia's "NvDecode" API (up until recently called "cuvid" is a cross platform, but nvidia proprietary API that exposes their hardware video decoding capabilities. It is analogous to their DXVA or VDPAU support on Windows or Linux but without using platform specific API calls. As a rule, you'd rather use DXVA or VDPAU as these are more mature and well supported APIs, but on Linux, VDPAU is falling behind the hardware capabilities, and there's no sign that nvidia are making the investments to update it. Most concretely, this means that there is no VP8/9 or HEVC Main10 support in VDPAU. On the other hand, NvDecode does export vp8/9 and partial support for HEVC Main10 (more on that below). ffmpeg already has support in the form of the "cuvid" family of decoders. Due to the design of the API, it is best exposed as a full decoder rather than an hwaccel. As such, there are decoders like h264_cuvid, hevc_cuvid, etc. These decoders support two output paths today - in both cases, NV12 frames are returned, either in CUDA device memory or regular system memory. In the case of the system memory path, the decoders can be used as-is in mpv today with a command line like: mpv --vd=lavc:h264_cuvid foobar.mp4 Doing this will take advantage of hardware decoding, but the cost of the memcpy to system memory adds up, especially for high resolution video (4K etc). To avoid that, we need an hwdec that takes advantage of CUDA's OpenGL interop to copy from device memory into OpenGL textures. That is what this change implements. The process is relatively simple as only basic device context aquisition needs to be done by us - the CUDA buffer pool is managed by the decoder - thankfully. The hwdec looks a bit like the vdpau interop one - the hwdec maintains a single set of plane textures and each output frame is repeatedly mapped into these textures to pass on. The frames are always in NV12 format, at least until 10bit output supports emerges. The only slightly interesting part of the copying process is that CUDA works by associating PBOs, so we need to define these for each of the textures. TODO Items: * I need to add a download_image function for screenshots. This would do the same copy to system memory that the decoder's system memory output does. * There are items to investigate on the ffmpeg side. There appears to be a problem with timestamps for some content. Final note: I mentioned HEVC Main10. While there is no 10bit output support, NvDecode can return dithered 8bit NV12 so you can take advantage of the hardware acceleration. This particular mode requires compiling ffmpeg with a modified header (or possibly the CUDA 8 RC) and is not upstream in ffmpeg yet. Usage: You will need to specify vo=opengl and hwdec=cuda. Note that hwdec=auto will probably not work as it will try to use vdpau first. mpv --hwdec=cuda --vo=opengl foobar.mp4 If you want to use filters that require frames in system memory, just use the decoder directly without the hwdec, as documented above.
2016-09-04 22:23:55 +00:00
ret = CHECK_CU(cuCtxPushCurrent(p->cuda_ctx));
if (ret < 0)
return ret;
gl->GenTextures(2, p->gl_textures);
for (int n = 0; n < 2; n++) {
gl->BindTexture(GL_TEXTURE_2D, p->gl_textures[n]);
GLenum filter = GL_NEAREST;
gl->TexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, filter);
gl->TexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, filter);
gl->TexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
gl->TexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
vo_opengl: hwdec_cuda: Support P016 output surfaces The latest 375.xx nvidia drivers add support for P016 output surfaces. In combination with an ffmpeg change to return those surfaces, we can display them. The bulk of the work is related to knowing which format you're dealing with at the right time. Once you know, it's straight forward.
2016-11-19 21:57:23 +00:00
gl->TexImage2D(GL_TEXTURE_2D, 0, n == 0 ? luma_format : chroma_format,
hwdec/opengl: Add support for CUDA and cuvid/NvDecode Nvidia's "NvDecode" API (up until recently called "cuvid" is a cross platform, but nvidia proprietary API that exposes their hardware video decoding capabilities. It is analogous to their DXVA or VDPAU support on Windows or Linux but without using platform specific API calls. As a rule, you'd rather use DXVA or VDPAU as these are more mature and well supported APIs, but on Linux, VDPAU is falling behind the hardware capabilities, and there's no sign that nvidia are making the investments to update it. Most concretely, this means that there is no VP8/9 or HEVC Main10 support in VDPAU. On the other hand, NvDecode does export vp8/9 and partial support for HEVC Main10 (more on that below). ffmpeg already has support in the form of the "cuvid" family of decoders. Due to the design of the API, it is best exposed as a full decoder rather than an hwaccel. As such, there are decoders like h264_cuvid, hevc_cuvid, etc. These decoders support two output paths today - in both cases, NV12 frames are returned, either in CUDA device memory or regular system memory. In the case of the system memory path, the decoders can be used as-is in mpv today with a command line like: mpv --vd=lavc:h264_cuvid foobar.mp4 Doing this will take advantage of hardware decoding, but the cost of the memcpy to system memory adds up, especially for high resolution video (4K etc). To avoid that, we need an hwdec that takes advantage of CUDA's OpenGL interop to copy from device memory into OpenGL textures. That is what this change implements. The process is relatively simple as only basic device context aquisition needs to be done by us - the CUDA buffer pool is managed by the decoder - thankfully. The hwdec looks a bit like the vdpau interop one - the hwdec maintains a single set of plane textures and each output frame is repeatedly mapped into these textures to pass on. The frames are always in NV12 format, at least until 10bit output supports emerges. The only slightly interesting part of the copying process is that CUDA works by associating PBOs, so we need to define these for each of the textures. TODO Items: * I need to add a download_image function for screenshots. This would do the same copy to system memory that the decoder's system memory output does. * There are items to investigate on the ffmpeg side. There appears to be a problem with timestamps for some content. Final note: I mentioned HEVC Main10. While there is no 10bit output support, NvDecode can return dithered 8bit NV12 so you can take advantage of the hardware acceleration. This particular mode requires compiling ffmpeg with a modified header (or possibly the CUDA 8 RC) and is not upstream in ffmpeg yet. Usage: You will need to specify vo=opengl and hwdec=cuda. Note that hwdec=auto will probably not work as it will try to use vdpau first. mpv --hwdec=cuda --vo=opengl foobar.mp4 If you want to use filters that require frames in system memory, just use the decoder directly without the hwdec, as documented above.
2016-09-04 22:23:55 +00:00
mp_image_plane_w(&p->layout, n),
mp_image_plane_h(&p->layout, n),
vo_opengl: hwdec_cuda: Support P016 output surfaces The latest 375.xx nvidia drivers add support for P016 output surfaces. In combination with an ffmpeg change to return those surfaces, we can display them. The bulk of the work is related to knowing which format you're dealing with at the right time. Once you know, it's straight forward.
2016-11-19 21:57:23 +00:00
0, n == 0 ? GL_RED : GL_RG, type, NULL);
vo_opengl: hwdec_cuda: directly map GL textures and skip using PBOs The documentation around this stuff is poor, but I found an nvidia sample that demonstrates how to use the interop API most efficiently. (https://github.com/nvpro-samples/gl_cuda_interop_pingpong_st) Key lessons are: 1) you can register the texture itself and have cuda write to it, thereby skipping an additional copy through the PBO. 2) You don't have to be mapped when you do the copy - once you get a mapped pointer, it remains valid. Magic! This lets us throw out the PBOs as well as much of the explicit alignment and stride handling. CPU usage is slightly (~3%) lower for 4K content in one test case, so it makes a detectable difference, and presumably saves memory.
2016-09-23 04:39:40 +00:00
gl->BindTexture(GL_TEXTURE_2D, 0);
hwdec/opengl: Add support for CUDA and cuvid/NvDecode Nvidia's "NvDecode" API (up until recently called "cuvid" is a cross platform, but nvidia proprietary API that exposes their hardware video decoding capabilities. It is analogous to their DXVA or VDPAU support on Windows or Linux but without using platform specific API calls. As a rule, you'd rather use DXVA or VDPAU as these are more mature and well supported APIs, but on Linux, VDPAU is falling behind the hardware capabilities, and there's no sign that nvidia are making the investments to update it. Most concretely, this means that there is no VP8/9 or HEVC Main10 support in VDPAU. On the other hand, NvDecode does export vp8/9 and partial support for HEVC Main10 (more on that below). ffmpeg already has support in the form of the "cuvid" family of decoders. Due to the design of the API, it is best exposed as a full decoder rather than an hwaccel. As such, there are decoders like h264_cuvid, hevc_cuvid, etc. These decoders support two output paths today - in both cases, NV12 frames are returned, either in CUDA device memory or regular system memory. In the case of the system memory path, the decoders can be used as-is in mpv today with a command line like: mpv --vd=lavc:h264_cuvid foobar.mp4 Doing this will take advantage of hardware decoding, but the cost of the memcpy to system memory adds up, especially for high resolution video (4K etc). To avoid that, we need an hwdec that takes advantage of CUDA's OpenGL interop to copy from device memory into OpenGL textures. That is what this change implements. The process is relatively simple as only basic device context aquisition needs to be done by us - the CUDA buffer pool is managed by the decoder - thankfully. The hwdec looks a bit like the vdpau interop one - the hwdec maintains a single set of plane textures and each output frame is repeatedly mapped into these textures to pass on. The frames are always in NV12 format, at least until 10bit output supports emerges. The only slightly interesting part of the copying process is that CUDA works by associating PBOs, so we need to define these for each of the textures. TODO Items: * I need to add a download_image function for screenshots. This would do the same copy to system memory that the decoder's system memory output does. * There are items to investigate on the ffmpeg side. There appears to be a problem with timestamps for some content. Final note: I mentioned HEVC Main10. While there is no 10bit output support, NvDecode can return dithered 8bit NV12 so you can take advantage of the hardware acceleration. This particular mode requires compiling ffmpeg with a modified header (or possibly the CUDA 8 RC) and is not upstream in ffmpeg yet. Usage: You will need to specify vo=opengl and hwdec=cuda. Note that hwdec=auto will probably not work as it will try to use vdpau first. mpv --hwdec=cuda --vo=opengl foobar.mp4 If you want to use filters that require frames in system memory, just use the decoder directly without the hwdec, as documented above.
2016-09-04 22:23:55 +00:00
vo_opengl: hwdec_cuda: directly map GL textures and skip using PBOs The documentation around this stuff is poor, but I found an nvidia sample that demonstrates how to use the interop API most efficiently. (https://github.com/nvpro-samples/gl_cuda_interop_pingpong_st) Key lessons are: 1) you can register the texture itself and have cuda write to it, thereby skipping an additional copy through the PBO. 2) You don't have to be mapped when you do the copy - once you get a mapped pointer, it remains valid. Magic! This lets us throw out the PBOs as well as much of the explicit alignment and stride handling. CPU usage is slightly (~3%) lower for 4K content in one test case, so it makes a detectable difference, and presumably saves memory.
2016-09-23 04:39:40 +00:00
ret = CHECK_CU(cuGraphicsGLRegisterImage(&p->cu_res[n], p->gl_textures[n],
GL_TEXTURE_2D,
CU_GRAPHICS_REGISTER_FLAGS_WRITE_DISCARD));
if (ret < 0)
goto error;
ret = CHECK_CU(cuGraphicsMapResources(1, &p->cu_res[n], 0));
if (ret < 0)
goto error;
ret = CHECK_CU(cuGraphicsSubResourceGetMappedArray(&p->cu_array[n], p->cu_res[n],
0, 0));
hwdec/opengl: Add support for CUDA and cuvid/NvDecode Nvidia's "NvDecode" API (up until recently called "cuvid" is a cross platform, but nvidia proprietary API that exposes their hardware video decoding capabilities. It is analogous to their DXVA or VDPAU support on Windows or Linux but without using platform specific API calls. As a rule, you'd rather use DXVA or VDPAU as these are more mature and well supported APIs, but on Linux, VDPAU is falling behind the hardware capabilities, and there's no sign that nvidia are making the investments to update it. Most concretely, this means that there is no VP8/9 or HEVC Main10 support in VDPAU. On the other hand, NvDecode does export vp8/9 and partial support for HEVC Main10 (more on that below). ffmpeg already has support in the form of the "cuvid" family of decoders. Due to the design of the API, it is best exposed as a full decoder rather than an hwaccel. As such, there are decoders like h264_cuvid, hevc_cuvid, etc. These decoders support two output paths today - in both cases, NV12 frames are returned, either in CUDA device memory or regular system memory. In the case of the system memory path, the decoders can be used as-is in mpv today with a command line like: mpv --vd=lavc:h264_cuvid foobar.mp4 Doing this will take advantage of hardware decoding, but the cost of the memcpy to system memory adds up, especially for high resolution video (4K etc). To avoid that, we need an hwdec that takes advantage of CUDA's OpenGL interop to copy from device memory into OpenGL textures. That is what this change implements. The process is relatively simple as only basic device context aquisition needs to be done by us - the CUDA buffer pool is managed by the decoder - thankfully. The hwdec looks a bit like the vdpau interop one - the hwdec maintains a single set of plane textures and each output frame is repeatedly mapped into these textures to pass on. The frames are always in NV12 format, at least until 10bit output supports emerges. The only slightly interesting part of the copying process is that CUDA works by associating PBOs, so we need to define these for each of the textures. TODO Items: * I need to add a download_image function for screenshots. This would do the same copy to system memory that the decoder's system memory output does. * There are items to investigate on the ffmpeg side. There appears to be a problem with timestamps for some content. Final note: I mentioned HEVC Main10. While there is no 10bit output support, NvDecode can return dithered 8bit NV12 so you can take advantage of the hardware acceleration. This particular mode requires compiling ffmpeg with a modified header (or possibly the CUDA 8 RC) and is not upstream in ffmpeg yet. Usage: You will need to specify vo=opengl and hwdec=cuda. Note that hwdec=auto will probably not work as it will try to use vdpau first. mpv --hwdec=cuda --vo=opengl foobar.mp4 If you want to use filters that require frames in system memory, just use the decoder directly without the hwdec, as documented above.
2016-09-04 22:23:55 +00:00
if (ret < 0)
goto error;
hwdec_cuda: Use the non-deprecated CUDA-GL interop API The nvidia examples use the old (as in CUDA 3.x) interop API which is deprecated, and I think not even functional on recent versions of CUDA for windows. As I was following the examples, I used this old API. So, let's update to the new API, and hopefully, it'll start working on windows too.
2016-09-10 03:07:43 +00:00
vo_opengl: hwdec_cuda: directly map GL textures and skip using PBOs The documentation around this stuff is poor, but I found an nvidia sample that demonstrates how to use the interop API most efficiently. (https://github.com/nvpro-samples/gl_cuda_interop_pingpong_st) Key lessons are: 1) you can register the texture itself and have cuda write to it, thereby skipping an additional copy through the PBO. 2) You don't have to be mapped when you do the copy - once you get a mapped pointer, it remains valid. Magic! This lets us throw out the PBOs as well as much of the explicit alignment and stride handling. CPU usage is slightly (~3%) lower for 4K content in one test case, so it makes a detectable difference, and presumably saves memory.
2016-09-23 04:39:40 +00:00
ret = CHECK_CU(cuGraphicsUnmapResources(1, &p->cu_res[n], 0));
if (ret < 0)
goto error;
hwdec/opengl: Add support for CUDA and cuvid/NvDecode Nvidia's "NvDecode" API (up until recently called "cuvid" is a cross platform, but nvidia proprietary API that exposes their hardware video decoding capabilities. It is analogous to their DXVA or VDPAU support on Windows or Linux but without using platform specific API calls. As a rule, you'd rather use DXVA or VDPAU as these are more mature and well supported APIs, but on Linux, VDPAU is falling behind the hardware capabilities, and there's no sign that nvidia are making the investments to update it. Most concretely, this means that there is no VP8/9 or HEVC Main10 support in VDPAU. On the other hand, NvDecode does export vp8/9 and partial support for HEVC Main10 (more on that below). ffmpeg already has support in the form of the "cuvid" family of decoders. Due to the design of the API, it is best exposed as a full decoder rather than an hwaccel. As such, there are decoders like h264_cuvid, hevc_cuvid, etc. These decoders support two output paths today - in both cases, NV12 frames are returned, either in CUDA device memory or regular system memory. In the case of the system memory path, the decoders can be used as-is in mpv today with a command line like: mpv --vd=lavc:h264_cuvid foobar.mp4 Doing this will take advantage of hardware decoding, but the cost of the memcpy to system memory adds up, especially for high resolution video (4K etc). To avoid that, we need an hwdec that takes advantage of CUDA's OpenGL interop to copy from device memory into OpenGL textures. That is what this change implements. The process is relatively simple as only basic device context aquisition needs to be done by us - the CUDA buffer pool is managed by the decoder - thankfully. The hwdec looks a bit like the vdpau interop one - the hwdec maintains a single set of plane textures and each output frame is repeatedly mapped into these textures to pass on. The frames are always in NV12 format, at least until 10bit output supports emerges. The only slightly interesting part of the copying process is that CUDA works by associating PBOs, so we need to define these for each of the textures. TODO Items: * I need to add a download_image function for screenshots. This would do the same copy to system memory that the decoder's system memory output does. * There are items to investigate on the ffmpeg side. There appears to be a problem with timestamps for some content. Final note: I mentioned HEVC Main10. While there is no 10bit output support, NvDecode can return dithered 8bit NV12 so you can take advantage of the hardware acceleration. This particular mode requires compiling ffmpeg with a modified header (or possibly the CUDA 8 RC) and is not upstream in ffmpeg yet. Usage: You will need to specify vo=opengl and hwdec=cuda. Note that hwdec=auto will probably not work as it will try to use vdpau first. mpv --hwdec=cuda --vo=opengl foobar.mp4 If you want to use filters that require frames in system memory, just use the decoder directly without the hwdec, as documented above.
2016-09-04 22:23:55 +00:00
}
error:
eret = CHECK_CU(cuCtxPopCurrent(&dummy));
if (eret < 0)
return eret;
return ret;
}
static void destroy(struct gl_hwdec *hw)
{
struct priv *p = hw->priv;
GL *gl = hw->gl;
CUcontext dummy;
// Don't bail if any CUDA calls fail. This is all best effort.
CHECK_CU(cuCtxPushCurrent(p->cuda_ctx));
for (int n = 0; n < 2; n++) {
hwdec_cuda: Use the non-deprecated CUDA-GL interop API The nvidia examples use the old (as in CUDA 3.x) interop API which is deprecated, and I think not even functional on recent versions of CUDA for windows. As I was following the examples, I used this old API. So, let's update to the new API, and hopefully, it'll start working on windows too.
2016-09-10 03:07:43 +00:00
if (p->cu_res[n] > 0)
CHECK_CU(cuGraphicsUnregisterResource(p->cu_res[n]));
hwdec/opengl: Add support for CUDA and cuvid/NvDecode Nvidia's "NvDecode" API (up until recently called "cuvid" is a cross platform, but nvidia proprietary API that exposes their hardware video decoding capabilities. It is analogous to their DXVA or VDPAU support on Windows or Linux but without using platform specific API calls. As a rule, you'd rather use DXVA or VDPAU as these are more mature and well supported APIs, but on Linux, VDPAU is falling behind the hardware capabilities, and there's no sign that nvidia are making the investments to update it. Most concretely, this means that there is no VP8/9 or HEVC Main10 support in VDPAU. On the other hand, NvDecode does export vp8/9 and partial support for HEVC Main10 (more on that below). ffmpeg already has support in the form of the "cuvid" family of decoders. Due to the design of the API, it is best exposed as a full decoder rather than an hwaccel. As such, there are decoders like h264_cuvid, hevc_cuvid, etc. These decoders support two output paths today - in both cases, NV12 frames are returned, either in CUDA device memory or regular system memory. In the case of the system memory path, the decoders can be used as-is in mpv today with a command line like: mpv --vd=lavc:h264_cuvid foobar.mp4 Doing this will take advantage of hardware decoding, but the cost of the memcpy to system memory adds up, especially for high resolution video (4K etc). To avoid that, we need an hwdec that takes advantage of CUDA's OpenGL interop to copy from device memory into OpenGL textures. That is what this change implements. The process is relatively simple as only basic device context aquisition needs to be done by us - the CUDA buffer pool is managed by the decoder - thankfully. The hwdec looks a bit like the vdpau interop one - the hwdec maintains a single set of plane textures and each output frame is repeatedly mapped into these textures to pass on. The frames are always in NV12 format, at least until 10bit output supports emerges. The only slightly interesting part of the copying process is that CUDA works by associating PBOs, so we need to define these for each of the textures. TODO Items: * I need to add a download_image function for screenshots. This would do the same copy to system memory that the decoder's system memory output does. * There are items to investigate on the ffmpeg side. There appears to be a problem with timestamps for some content. Final note: I mentioned HEVC Main10. While there is no 10bit output support, NvDecode can return dithered 8bit NV12 so you can take advantage of the hardware acceleration. This particular mode requires compiling ffmpeg with a modified header (or possibly the CUDA 8 RC) and is not upstream in ffmpeg yet. Usage: You will need to specify vo=opengl and hwdec=cuda. Note that hwdec=auto will probably not work as it will try to use vdpau first. mpv --hwdec=cuda --vo=opengl foobar.mp4 If you want to use filters that require frames in system memory, just use the decoder directly without the hwdec, as documented above.
2016-09-04 22:23:55 +00:00
}
CHECK_CU(cuCtxPopCurrent(&dummy));
vo_opengl: hwdec_cuda: Use dynamic loading for cuda functions This change applies the pattern used in ffmpeg to dynamically load cuda, to avoid requiring the CUDA SDK at build time.
2016-10-08 23:51:15 +00:00
CHECK_CU(cuCtxDestroy(p->cuda_ctx));
hwdec/opengl: Add support for CUDA and cuvid/NvDecode Nvidia's "NvDecode" API (up until recently called "cuvid" is a cross platform, but nvidia proprietary API that exposes their hardware video decoding capabilities. It is analogous to their DXVA or VDPAU support on Windows or Linux but without using platform specific API calls. As a rule, you'd rather use DXVA or VDPAU as these are more mature and well supported APIs, but on Linux, VDPAU is falling behind the hardware capabilities, and there's no sign that nvidia are making the investments to update it. Most concretely, this means that there is no VP8/9 or HEVC Main10 support in VDPAU. On the other hand, NvDecode does export vp8/9 and partial support for HEVC Main10 (more on that below). ffmpeg already has support in the form of the "cuvid" family of decoders. Due to the design of the API, it is best exposed as a full decoder rather than an hwaccel. As such, there are decoders like h264_cuvid, hevc_cuvid, etc. These decoders support two output paths today - in both cases, NV12 frames are returned, either in CUDA device memory or regular system memory. In the case of the system memory path, the decoders can be used as-is in mpv today with a command line like: mpv --vd=lavc:h264_cuvid foobar.mp4 Doing this will take advantage of hardware decoding, but the cost of the memcpy to system memory adds up, especially for high resolution video (4K etc). To avoid that, we need an hwdec that takes advantage of CUDA's OpenGL interop to copy from device memory into OpenGL textures. That is what this change implements. The process is relatively simple as only basic device context aquisition needs to be done by us - the CUDA buffer pool is managed by the decoder - thankfully. The hwdec looks a bit like the vdpau interop one - the hwdec maintains a single set of plane textures and each output frame is repeatedly mapped into these textures to pass on. The frames are always in NV12 format, at least until 10bit output supports emerges. The only slightly interesting part of the copying process is that CUDA works by associating PBOs, so we need to define these for each of the textures. TODO Items: * I need to add a download_image function for screenshots. This would do the same copy to system memory that the decoder's system memory output does. * There are items to investigate on the ffmpeg side. There appears to be a problem with timestamps for some content. Final note: I mentioned HEVC Main10. While there is no 10bit output support, NvDecode can return dithered 8bit NV12 so you can take advantage of the hardware acceleration. This particular mode requires compiling ffmpeg with a modified header (or possibly the CUDA 8 RC) and is not upstream in ffmpeg yet. Usage: You will need to specify vo=opengl and hwdec=cuda. Note that hwdec=auto will probably not work as it will try to use vdpau first. mpv --hwdec=cuda --vo=opengl foobar.mp4 If you want to use filters that require frames in system memory, just use the decoder directly without the hwdec, as documented above.
2016-09-04 22:23:55 +00:00
gl->DeleteTextures(2, p->gl_textures);
hwdec_devices_remove(hw->devs, &p->hwctx);
}
static int map_frame(struct gl_hwdec *hw, struct mp_image *hw_image,
struct gl_hwdec_frame *out_frame)
{
struct priv *p = hw->priv;
CUcontext dummy;
int ret = 0, eret = 0;
ret = CHECK_CU(cuCtxPushCurrent(p->cuda_ctx));
if (ret < 0)
return ret;
*out_frame = (struct gl_hwdec_frame) { 0, };
for (int n = 0; n < 2; n++) {
vo_opengl: hwdec_cuda: directly map GL textures and skip using PBOs The documentation around this stuff is poor, but I found an nvidia sample that demonstrates how to use the interop API most efficiently. (https://github.com/nvpro-samples/gl_cuda_interop_pingpong_st) Key lessons are: 1) you can register the texture itself and have cuda write to it, thereby skipping an additional copy through the PBO. 2) You don't have to be mapped when you do the copy - once you get a mapped pointer, it remains valid. Magic! This lets us throw out the PBOs as well as much of the explicit alignment and stride handling. CPU usage is slightly (~3%) lower for 4K content in one test case, so it makes a detectable difference, and presumably saves memory.
2016-09-23 04:39:40 +00:00
// widthInBytes must account for the chroma plane
vo_opengl: hwdec_cuda: Support P016 output surfaces The latest 375.xx nvidia drivers add support for P016 output surfaces. In combination with an ffmpeg change to return those surfaces, we can display them. The bulk of the work is related to knowing which format you're dealing with at the right time. Once you know, it's straight forward.
2016-11-19 21:57:23 +00:00
// elements being two samples wide.
hwdec/opengl: Add support for CUDA and cuvid/NvDecode Nvidia's "NvDecode" API (up until recently called "cuvid" is a cross platform, but nvidia proprietary API that exposes their hardware video decoding capabilities. It is analogous to their DXVA or VDPAU support on Windows or Linux but without using platform specific API calls. As a rule, you'd rather use DXVA or VDPAU as these are more mature and well supported APIs, but on Linux, VDPAU is falling behind the hardware capabilities, and there's no sign that nvidia are making the investments to update it. Most concretely, this means that there is no VP8/9 or HEVC Main10 support in VDPAU. On the other hand, NvDecode does export vp8/9 and partial support for HEVC Main10 (more on that below). ffmpeg already has support in the form of the "cuvid" family of decoders. Due to the design of the API, it is best exposed as a full decoder rather than an hwaccel. As such, there are decoders like h264_cuvid, hevc_cuvid, etc. These decoders support two output paths today - in both cases, NV12 frames are returned, either in CUDA device memory or regular system memory. In the case of the system memory path, the decoders can be used as-is in mpv today with a command line like: mpv --vd=lavc:h264_cuvid foobar.mp4 Doing this will take advantage of hardware decoding, but the cost of the memcpy to system memory adds up, especially for high resolution video (4K etc). To avoid that, we need an hwdec that takes advantage of CUDA's OpenGL interop to copy from device memory into OpenGL textures. That is what this change implements. The process is relatively simple as only basic device context aquisition needs to be done by us - the CUDA buffer pool is managed by the decoder - thankfully. The hwdec looks a bit like the vdpau interop one - the hwdec maintains a single set of plane textures and each output frame is repeatedly mapped into these textures to pass on. The frames are always in NV12 format, at least until 10bit output supports emerges. The only slightly interesting part of the copying process is that CUDA works by associating PBOs, so we need to define these for each of the textures. TODO Items: * I need to add a download_image function for screenshots. This would do the same copy to system memory that the decoder's system memory output does. * There are items to investigate on the ffmpeg side. There appears to be a problem with timestamps for some content. Final note: I mentioned HEVC Main10. While there is no 10bit output support, NvDecode can return dithered 8bit NV12 so you can take advantage of the hardware acceleration. This particular mode requires compiling ffmpeg with a modified header (or possibly the CUDA 8 RC) and is not upstream in ffmpeg yet. Usage: You will need to specify vo=opengl and hwdec=cuda. Note that hwdec=auto will probably not work as it will try to use vdpau first. mpv --hwdec=cuda --vo=opengl foobar.mp4 If you want to use filters that require frames in system memory, just use the decoder directly without the hwdec, as documented above.
2016-09-04 22:23:55 +00:00
CUDA_MEMCPY2D cpy = {
.srcMemoryType = CU_MEMORYTYPE_DEVICE,
vo_opengl: hwdec_cuda: directly map GL textures and skip using PBOs The documentation around this stuff is poor, but I found an nvidia sample that demonstrates how to use the interop API most efficiently. (https://github.com/nvpro-samples/gl_cuda_interop_pingpong_st) Key lessons are: 1) you can register the texture itself and have cuda write to it, thereby skipping an additional copy through the PBO. 2) You don't have to be mapped when you do the copy - once you get a mapped pointer, it remains valid. Magic! This lets us throw out the PBOs as well as much of the explicit alignment and stride handling. CPU usage is slightly (~3%) lower for 4K content in one test case, so it makes a detectable difference, and presumably saves memory.
2016-09-23 04:39:40 +00:00
.dstMemoryType = CU_MEMORYTYPE_ARRAY,
hwdec/opengl: Add support for CUDA and cuvid/NvDecode Nvidia's "NvDecode" API (up until recently called "cuvid" is a cross platform, but nvidia proprietary API that exposes their hardware video decoding capabilities. It is analogous to their DXVA or VDPAU support on Windows or Linux but without using platform specific API calls. As a rule, you'd rather use DXVA or VDPAU as these are more mature and well supported APIs, but on Linux, VDPAU is falling behind the hardware capabilities, and there's no sign that nvidia are making the investments to update it. Most concretely, this means that there is no VP8/9 or HEVC Main10 support in VDPAU. On the other hand, NvDecode does export vp8/9 and partial support for HEVC Main10 (more on that below). ffmpeg already has support in the form of the "cuvid" family of decoders. Due to the design of the API, it is best exposed as a full decoder rather than an hwaccel. As such, there are decoders like h264_cuvid, hevc_cuvid, etc. These decoders support two output paths today - in both cases, NV12 frames are returned, either in CUDA device memory or regular system memory. In the case of the system memory path, the decoders can be used as-is in mpv today with a command line like: mpv --vd=lavc:h264_cuvid foobar.mp4 Doing this will take advantage of hardware decoding, but the cost of the memcpy to system memory adds up, especially for high resolution video (4K etc). To avoid that, we need an hwdec that takes advantage of CUDA's OpenGL interop to copy from device memory into OpenGL textures. That is what this change implements. The process is relatively simple as only basic device context aquisition needs to be done by us - the CUDA buffer pool is managed by the decoder - thankfully. The hwdec looks a bit like the vdpau interop one - the hwdec maintains a single set of plane textures and each output frame is repeatedly mapped into these textures to pass on. The frames are always in NV12 format, at least until 10bit output supports emerges. The only slightly interesting part of the copying process is that CUDA works by associating PBOs, so we need to define these for each of the textures. TODO Items: * I need to add a download_image function for screenshots. This would do the same copy to system memory that the decoder's system memory output does. * There are items to investigate on the ffmpeg side. There appears to be a problem with timestamps for some content. Final note: I mentioned HEVC Main10. While there is no 10bit output support, NvDecode can return dithered 8bit NV12 so you can take advantage of the hardware acceleration. This particular mode requires compiling ffmpeg with a modified header (or possibly the CUDA 8 RC) and is not upstream in ffmpeg yet. Usage: You will need to specify vo=opengl and hwdec=cuda. Note that hwdec=auto will probably not work as it will try to use vdpau first. mpv --hwdec=cuda --vo=opengl foobar.mp4 If you want to use filters that require frames in system memory, just use the decoder directly without the hwdec, as documented above.
2016-09-04 22:23:55 +00:00
.srcDevice = (CUdeviceptr)hw_image->planes[n],
.srcPitch = hw_image->stride[n],
.srcY = 0,
vo_opengl: hwdec_cuda: directly map GL textures and skip using PBOs The documentation around this stuff is poor, but I found an nvidia sample that demonstrates how to use the interop API most efficiently. (https://github.com/nvpro-samples/gl_cuda_interop_pingpong_st) Key lessons are: 1) you can register the texture itself and have cuda write to it, thereby skipping an additional copy through the PBO. 2) You don't have to be mapped when you do the copy - once you get a mapped pointer, it remains valid. Magic! This lets us throw out the PBOs as well as much of the explicit alignment and stride handling. CPU usage is slightly (~3%) lower for 4K content in one test case, so it makes a detectable difference, and presumably saves memory.
2016-09-23 04:39:40 +00:00
.dstArray = p->cu_array[n],
vo_opengl: hwdec_cuda: Support P016 output surfaces The latest 375.xx nvidia drivers add support for P016 output surfaces. In combination with an ffmpeg change to return those surfaces, we can display them. The bulk of the work is related to knowing which format you're dealing with at the right time. Once you know, it's straight forward.
2016-11-19 21:57:23 +00:00
.WidthInBytes = mp_image_plane_w(&p->layout, n) *
(n + 1) * p->sample_width,
hwdec/opengl: Add support for CUDA and cuvid/NvDecode Nvidia's "NvDecode" API (up until recently called "cuvid" is a cross platform, but nvidia proprietary API that exposes their hardware video decoding capabilities. It is analogous to their DXVA or VDPAU support on Windows or Linux but without using platform specific API calls. As a rule, you'd rather use DXVA or VDPAU as these are more mature and well supported APIs, but on Linux, VDPAU is falling behind the hardware capabilities, and there's no sign that nvidia are making the investments to update it. Most concretely, this means that there is no VP8/9 or HEVC Main10 support in VDPAU. On the other hand, NvDecode does export vp8/9 and partial support for HEVC Main10 (more on that below). ffmpeg already has support in the form of the "cuvid" family of decoders. Due to the design of the API, it is best exposed as a full decoder rather than an hwaccel. As such, there are decoders like h264_cuvid, hevc_cuvid, etc. These decoders support two output paths today - in both cases, NV12 frames are returned, either in CUDA device memory or regular system memory. In the case of the system memory path, the decoders can be used as-is in mpv today with a command line like: mpv --vd=lavc:h264_cuvid foobar.mp4 Doing this will take advantage of hardware decoding, but the cost of the memcpy to system memory adds up, especially for high resolution video (4K etc). To avoid that, we need an hwdec that takes advantage of CUDA's OpenGL interop to copy from device memory into OpenGL textures. That is what this change implements. The process is relatively simple as only basic device context aquisition needs to be done by us - the CUDA buffer pool is managed by the decoder - thankfully. The hwdec looks a bit like the vdpau interop one - the hwdec maintains a single set of plane textures and each output frame is repeatedly mapped into these textures to pass on. The frames are always in NV12 format, at least until 10bit output supports emerges. The only slightly interesting part of the copying process is that CUDA works by associating PBOs, so we need to define these for each of the textures. TODO Items: * I need to add a download_image function for screenshots. This would do the same copy to system memory that the decoder's system memory output does. * There are items to investigate on the ffmpeg side. There appears to be a problem with timestamps for some content. Final note: I mentioned HEVC Main10. While there is no 10bit output support, NvDecode can return dithered 8bit NV12 so you can take advantage of the hardware acceleration. This particular mode requires compiling ffmpeg with a modified header (or possibly the CUDA 8 RC) and is not upstream in ffmpeg yet. Usage: You will need to specify vo=opengl and hwdec=cuda. Note that hwdec=auto will probably not work as it will try to use vdpau first. mpv --hwdec=cuda --vo=opengl foobar.mp4 If you want to use filters that require frames in system memory, just use the decoder directly without the hwdec, as documented above.
2016-09-04 22:23:55 +00:00
.Height = mp_image_plane_h(&p->layout, n),
};
ret = CHECK_CU(cuMemcpy2D(&cpy));
if (ret < 0)
goto error;
out_frame->planes[n] = (struct gl_hwdec_plane){
.gl_texture = p->gl_textures[n],
.gl_target = GL_TEXTURE_2D,
.tex_w = mp_image_plane_w(&p->layout, n),
.tex_h = mp_image_plane_h(&p->layout, n),
};
}
hwdec_cuda: Use the non-deprecated CUDA-GL interop API The nvidia examples use the old (as in CUDA 3.x) interop API which is deprecated, and I think not even functional on recent versions of CUDA for windows. As I was following the examples, I used this old API. So, let's update to the new API, and hopefully, it'll start working on windows too.
2016-09-10 03:07:43 +00:00
hwdec/opengl: Add support for CUDA and cuvid/NvDecode Nvidia's "NvDecode" API (up until recently called "cuvid" is a cross platform, but nvidia proprietary API that exposes their hardware video decoding capabilities. It is analogous to their DXVA or VDPAU support on Windows or Linux but without using platform specific API calls. As a rule, you'd rather use DXVA or VDPAU as these are more mature and well supported APIs, but on Linux, VDPAU is falling behind the hardware capabilities, and there's no sign that nvidia are making the investments to update it. Most concretely, this means that there is no VP8/9 or HEVC Main10 support in VDPAU. On the other hand, NvDecode does export vp8/9 and partial support for HEVC Main10 (more on that below). ffmpeg already has support in the form of the "cuvid" family of decoders. Due to the design of the API, it is best exposed as a full decoder rather than an hwaccel. As such, there are decoders like h264_cuvid, hevc_cuvid, etc. These decoders support two output paths today - in both cases, NV12 frames are returned, either in CUDA device memory or regular system memory. In the case of the system memory path, the decoders can be used as-is in mpv today with a command line like: mpv --vd=lavc:h264_cuvid foobar.mp4 Doing this will take advantage of hardware decoding, but the cost of the memcpy to system memory adds up, especially for high resolution video (4K etc). To avoid that, we need an hwdec that takes advantage of CUDA's OpenGL interop to copy from device memory into OpenGL textures. That is what this change implements. The process is relatively simple as only basic device context aquisition needs to be done by us - the CUDA buffer pool is managed by the decoder - thankfully. The hwdec looks a bit like the vdpau interop one - the hwdec maintains a single set of plane textures and each output frame is repeatedly mapped into these textures to pass on. The frames are always in NV12 format, at least until 10bit output supports emerges. The only slightly interesting part of the copying process is that CUDA works by associating PBOs, so we need to define these for each of the textures. TODO Items: * I need to add a download_image function for screenshots. This would do the same copy to system memory that the decoder's system memory output does. * There are items to investigate on the ffmpeg side. There appears to be a problem with timestamps for some content. Final note: I mentioned HEVC Main10. While there is no 10bit output support, NvDecode can return dithered 8bit NV12 so you can take advantage of the hardware acceleration. This particular mode requires compiling ffmpeg with a modified header (or possibly the CUDA 8 RC) and is not upstream in ffmpeg yet. Usage: You will need to specify vo=opengl and hwdec=cuda. Note that hwdec=auto will probably not work as it will try to use vdpau first. mpv --hwdec=cuda --vo=opengl foobar.mp4 If you want to use filters that require frames in system memory, just use the decoder directly without the hwdec, as documented above.
2016-09-04 22:23:55 +00:00
error:
eret = CHECK_CU(cuCtxPopCurrent(&dummy));
if (eret < 0)
return eret;
return ret;
}
const struct gl_hwdec_driver gl_hwdec_cuda = {
.name = "cuda",
.api = HWDEC_CUDA,
.imgfmt = IMGFMT_CUDA,
.create = cuda_create,
.reinit = reinit,
.map_frame = map_frame,
.destroy = destroy,
};