izackp/Apollo
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Allow cuda kernel to run in parallell

Browse Source
This commit is contained in:
loki-47-6F-64
2021-09-27 17:58:35 +02:00
parent 847d7b6980
commit 4177b02064
4 changed files with 95 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
sunshine/platform/linux/cuda.h
View File

@@ -1,16 +1,17 @@
#if !defined(SUNSHINE_PLATFORM_CUDA_H) && defined(SUNSHINE_BUILD_CUDA)
#define SUNSHINE_PLATFORM_CUDA_H
#include <vector>
#include <memory>
#include <string>
#include <vector>
namespace platf {
class hwdevice_t;
class img_t;
}
class hwdevice_t;
class img_t;
} // namespace platf
namespace cuda {
namespace nvfbc {
std::vector<std::string> display_names();
}
@@ -21,8 +22,10 @@ int init();
typedef struct cudaArray *cudaArray_t;
#if !defined(__CUDACC__)
typedef struct CUstream_st *cudaStream_t;
typedef unsigned long long cudaTextureObject_t;
#else /* defined(__CUDACC__) */
typedef __location__(device_builtin) struct CUstream_st *cudaStream_t;
typedef __location__(device_builtin) unsigned long long cudaTextureObject_t;
#endif /* !defined(__CUDACC__) */
@@ -33,7 +36,15 @@ public:
void operator()(void *ptr);
};
using ptr_t = std::unique_ptr<void, freeCudaPtr_t>;
class freeCudaStream_t {
public:
void operator()(cudaStream_t ptr);
};
using ptr_t = std::unique_ptr<void, freeCudaPtr_t>;
using stream_t = std::unique_ptr<CUstream_st, freeCudaStream_t>;
stream_t make_stream(int flags = 0);
struct viewport_t {
int width, height;
@@ -75,8 +86,8 @@ public:
static std::unique_ptr<sws_t> make(int in_width, int in_height, int out_width, int out_height, int pitch);
// Converts loaded image into a CUDevicePtr
int convert(std::uint8_t *Y, std::uint8_t *UV, std::uint32_t pitchY, std::uint32_t pitchUV, cudaTextureObject_t texture);
int convert(std::uint8_t *Y, std::uint8_t *UV, std::uint32_t pitchY, std::uint32_t pitchUV, cudaTextureObject_t texture, const viewport_t &viewport);
int convert(std::uint8_t *Y, std::uint8_t *UV, std::uint32_t pitchY, std::uint32_t pitchUV, cudaTextureObject_t texture, stream_t::pointer stream);
int convert(std::uint8_t *Y, std::uint8_t *UV, std::uint32_t pitchY, std::uint32_t pitchUV, cudaTextureObject_t texture, stream_t::pointer stream, const viewport_t &viewport);
void set_colorspace(std::uint32_t colorspace, std::uint32_t color_range);