1 #include "../../StdAfx.h"
\r
3 #include "parallel_yadif.h"
\r
7 #include <libavfilter/avfilter.h>
\r
10 #include <boost/thread/once.hpp>
\r
17 AVFilterBufferRef *cur;
\r
18 AVFilterBufferRef *next;
\r
19 AVFilterBufferRef *prev;
\r
20 AVFilterBufferRef *out;
\r
21 void (*filter_line)(uint8_t *dst,
\r
22 uint8_t *prev, uint8_t *cur, uint8_t *next,
\r
23 int w, int prefs, int mrefs, int parity, int mode);
\r
24 //const AVPixFmtDescriptor *csp;
\r
27 struct parallel_yadif_context
\r
45 parallel_yadif_context() : index(0){}
\r
48 void (*org_yadif_filter_line)(uint8_t *dst, uint8_t *prev, uint8_t *cur, uint8_t *next, int w, int prefs, int mrefs, int parity, int mode) = 0;
\r
50 void parallel_yadif_filter_line(parallel_yadif_context& ctx, uint8_t *dst, uint8_t *prev, uint8_t *cur, uint8_t *next, int w, int prefs, int mrefs, int parity, int mode)
\r
52 parallel_yadif_context::arg arg = {dst, prev, cur, next, w, prefs, mrefs, parity, mode};
\r
53 ctx.args[ctx.index++] = arg;
\r
57 tbb::parallel_for(tbb::blocked_range<size_t>(0, ctx.index), [=](const tbb::blocked_range<size_t>& r)
\r
59 for(auto n = r.begin(); n != r.end(); ++n)
\r
60 org_yadif_filter_line(ctx.args[n].dst, ctx.args[n].prev, ctx.args[n].cur, ctx.args[n].next, ctx.args[n].w, ctx.args[n].prefs, ctx.args[n].mrefs, ctx.args[n].parity, ctx.args[n].mode);
\r
66 #define RENAME(a) f ## a
\r
69 void RENAME(x)(uint8_t *dst, uint8_t *prev, uint8_t *cur, uint8_t *next, int w, int prefs, int mrefs, int parity, int mode) \
\r
71 static parallel_yadif_context ctx;\
\r
72 parallel_yadif_filter_line(ctx, dst, prev, cur, next, w, prefs, mrefs, parity, mode);\
\r
75 ff(0); ff(1); ff(2); ff(3); ff(4); ff(5); ff(6); ff(7); ff(8); ff(9); ff(10); ff(11); ff(12); ff(13); ff(14); ff(15); ff(16); ff(17);
\r
77 void (*fs[])(uint8_t *dst, uint8_t *prev, uint8_t *cur, uint8_t *next, int w, int prefs, int mrefs, int parity, int mode) =
\r
79 {f0, f1, f2, f3, f4, f5, f6, f7, f8, f9, f10, f11, f12, f13, f14, f15, f16, f17};
\r
83 tbb::concurrent_bounded_queue<decltype(org_yadif_filter_line)> parallel_line_func_pool;
\r
87 for(int n = 0; n < sizeof(fs)/sizeof(decltype(org_yadif_filter_line)); ++n)
\r
88 parallel_line_func_pool.push(fs[n]);
\r
91 void return_parallel_yadif(void* func)
\r
94 parallel_line_func_pool.push(reinterpret_cast<decltype(org_yadif_filter_line)>(func));
\r
97 std::shared_ptr<void> make_parallel_yadif(AVFilterContext* ctx)
\r
99 static boost::once_flag flag = BOOST_ONCE_INIT;
\r
100 boost::call_once(&init_pool, flag);
\r
102 YADIFContext* yadif = (YADIFContext*)ctx->priv;
\r
103 org_yadif_filter_line = yadif->filter_line; // Data race is not a problem.
\r
105 decltype(org_yadif_filter_line) func = nullptr;
\r
106 if(!parallel_line_func_pool.try_pop(func))
\r
107 CASPAR_LOG(warning) << "Not enough scalable-yadif instances. Running non-scalable";
\r
109 yadif->filter_line = func;
\r
111 return std::shared_ptr<void>(func, return_parallel_yadif);
\r