avfilter/scale_cuda: add support for rgb32/bgr32 conversions

As we are introducing two new formats and supporting conversions between them, and also with the existing 0rgb32/0bgr32 formats, we get a combinatorial explosion of kernels. I introduced a few new macros to keep the things mostly managable. The conversions are all simple, following existing patterns, with four specific exceptions. When converting from 0rgb32/0bgr32 to rgb32/bgr32, we need to ensure the alpha value is set to 1. In all other cases, it can just be passed through, either to be used or ignored.

Philip Langdale committed 3 years ago

Commit 3c07c241 parent f42df833 files · +123 −32Browse files

Showing the changes to libavfilter (3 of 3 files). Show all

Unified · Split

3 files

libavfilter/version.h +1 −1
libavfilter/vf_scale_cuda.c +2 −0
libavfilter/vf_scale_cuda.cu +120 −31

libavfilter/version.h+1−1

@@ -32,7 +32,7 @@ · 31 unmodified lines
32	#include "version_major.h"	32	#include "version_major.h"
33		33
34	#define LIBAVFILTER_VERSION_MINOR 8	34	#define LIBAVFILTER_VERSION_MINOR 8
35	#define LIBAVFILTER_VERSION_MICRO 101	35	#define LIBAVFILTER_VERSION_MICRO 102
36		36
37		37
38	#define LIBAVFILTER_VERSION_INT AV_VERSION_INT(LIBAVFILTER_VERSION_MAJOR, \	38	#define LIBAVFILTER_VERSION_INT AV_VERSION_INT(LIBAVFILTER_VERSION_MAJOR, \

libavfilter/vf_scale_cuda.c+2−0

@@ -51,6 +51,8 @@ static const enum AVPixelFormat supported_formats[] = { · 50 unmodified lines
51	AV_PIX_FMT_YUV444P16,	51	AV_PIX_FMT_YUV444P16,
52	AV_PIX_FMT_0RGB32,	52	AV_PIX_FMT_0RGB32,
53	AV_PIX_FMT_0BGR32,	53	AV_PIX_FMT_0BGR32,
		54	AV_PIX_FMT_RGB32,
		55	AV_PIX_FMT_BGR32,
54	};	56	};
55		57
56	#define DIV_UP(a, b) ( ((a) + (b) - 1) / (b) )	58	#define DIV_UP(a, b) ( ((a) + (b) - 1) / (b) )

libavfilter/vf_scale_cuda.cu+120−31

@@ -853,9 +853,67 @@ struct Convert_yuv444p16le_yuv444p16le · 852 unmodified lines
853	}	853	}
854	};	854	};
855		855
856	// bgr0->X	856	#define DEF_CONVERT_IDENTITY(fmt1, fmt2)\
857		857	\
858	struct Convert_bgr0_bgr0	858	struct Convert_##fmt1##_##fmt2 \
		859	{ \
		860	static const int in_bit_depth = 8; \
		861	typedef uchar4 in_T; \
		862	typedef uchar in_T_uv; \
		863	typedef uchar4 out_T; \
		864	typedef uchar out_T_uv; \
		865	\
		866	DEF_F(Convert, out_T) \
		867	{ \
		868	DEFAULT_DST(0) = SUB_F(y, 0); \
		869	} \
		870	\
		871	DEF_F(Convert_uv, out_T_uv) \
		872	{ \
		873	} \
		874	}; \
		875
		876	#define DEF_CONVERT_REORDER(fmt1, fmt2) \
		877	\
		878	struct Convert_##fmt1##_##fmt2 \
		879	{ \
		880	static const int in_bit_depth = 8; \
		881	typedef uchar4 in_T; \
		882	typedef uchar in_T_uv; \
		883	typedef uchar4 out_T; \
		884	typedef uchar out_T_uv; \
		885	\
		886	DEF_F(Convert, out_T) \
		887	{ \
		888	uchar4 res = SUB_F(y, 0); \
		889	DEFAULT_DST(0) = make_uchar4( \
		890	res.z, \
		891	res.y, \
		892	res.x, \
		893	res.w \
		894	); \
		895	} \
		896	\
		897	DEF_F(Convert_uv, out_T_uv) \
		898	{ \
		899	} \
		900	}; \
		901
		902	#define DEF_CONVERT_RGB(fmt1, fmt2) \
		903	\
		904	DEF_CONVERT_IDENTITY(fmt1, fmt1) \
		905	DEF_CONVERT_REORDER (fmt1, fmt2) \
		906	DEF_CONVERT_REORDER (fmt2, fmt1) \
		907	DEF_CONVERT_IDENTITY(fmt2, fmt2)
		908
		909	DEF_CONVERT_RGB(rgb0, bgr0)
		910	DEF_CONVERT_RGB(rgba, bgra)
		911	DEF_CONVERT_IDENTITY(rgba, rgb0)
		912	DEF_CONVERT_IDENTITY(bgra, bgr0)
		913	DEF_CONVERT_REORDER(rgba, bgr0)
		914	DEF_CONVERT_REORDER(bgra, rgb0)
		915
		916	struct Convert_bgr0_bgra
859	{	917	{
860	static const int in_bit_depth = 8;	918	static const int in_bit_depth = 8;
861	typedef uchar4 in_T;	919	typedef uchar4 in_T;
@@ -865,7 +923,13 @@ struct Convert_bgr0_bgr0 · 3 unmodified lines
865		923
866	DEF_F(Convert, out_T)	924	DEF_F(Convert, out_T)
867	{	925	{
868	DEFAULT_DST(0) = SUB_F(y, 0);	926	uchar4 res = SUB_F(y, 0);
		927	DEFAULT_DST(0) = make_uchar4(
		928	res.x,
		929	res.y,
		930	res.z,
		931	1
		932	);
869	}	933	}
870		934
871	DEF_F(Convert_uv, out_T_uv)	935	DEF_F(Convert_uv, out_T_uv)
@@ -873,7 +937,7 @@ struct Convert_bgr0_bgr0 · 1 unmodified line
873	}	937	}
874	};	938	};
875		939
876	struct Convert_bgr0_rgb0	940	struct Convert_bgr0_rgba
877	{	941	{
878	static const int in_bit_depth = 8;	942	static const int in_bit_depth = 8;
879	typedef uchar4 in_T;	943	typedef uchar4 in_T;
@@ -888,7 +952,7 @@ struct Convert_bgr0_rgb0 · 8 unmodified lines
888	res.z,	952	res.z,
889	res.y,	953	res.y,
890	res.x,	954	res.x,
891	res.w	955	1
892	);	956	);
893	}	957	}
894		958
@@ -897,9 +961,7 @@ struct Convert_bgr0_rgb0 · 2 unmodified lines
897	}	961	}
898	};	962	};
899		963
900	// rgb0->X	964	struct Convert_rgb0_bgra
901
902	struct Convert_rgb0_bgr0
903	{	965	{
904	static const int in_bit_depth = 8;	966	static const int in_bit_depth = 8;
905	typedef uchar4 in_T;	967	typedef uchar4 in_T;
@@ -914,7 +976,7 @@ struct Convert_rgb0_bgr0 · 8 unmodified lines
914	res.z,	976	res.z,
915	res.y,	977	res.y,
916	res.x,	978	res.x,
917	res.w	979	1
918	);	980	);
919	}	981	}
920		982
@@ -923,7 +985,7 @@ struct Convert_rgb0_bgr0 · 2 unmodified lines
923	}	985	}
924	};	986	};
925		987
926	struct Convert_rgb0_rgb0	988	struct Convert_rgb0_rgba
927	{	989	{
928	static const int in_bit_depth = 8;	990	static const int in_bit_depth = 8;
929	typedef uchar4 in_T;	991	typedef uchar4 in_T;
@@ -933,7 +995,13 @@ struct Convert_rgb0_rgb0 · 3 unmodified lines
933		995
934	DEF_F(Convert, out_T)	996	DEF_F(Convert, out_T)
935	{	997	{
936	DEFAULT_DST(0) = SUB_F(y, 0);	998	uchar4 res = SUB_F(y, 0);
		999	DEFAULT_DST(0) = make_uchar4(
		1000	res.x,
		1001	res.y,
		1002	res.z,
		1003	1
		1004	);
937	}	1005	}
938		1006
939	DEF_F(Convert_uv, out_T_uv)	1007	DEF_F(Convert_uv, out_T_uv)
@@ -1117,6 +1185,12 @@ extern "C" { · 177 unmodified lines
1117	NEAREST_KERNEL_RAW(p016le_ ## C) \	1185	NEAREST_KERNEL_RAW(p016le_ ## C) \
1118	NEAREST_KERNEL_RAW(yuv444p16le_ ## C)	1186	NEAREST_KERNEL_RAW(yuv444p16le_ ## C)
1119		1187
		1188	#define NEAREST_KERNELS_RGB(C) \
		1189	NEAREST_KERNEL_RAW(rgb0_ ## C) \
		1190	NEAREST_KERNEL_RAW(bgr0_ ## C) \
		1191	NEAREST_KERNEL_RAW(rgba_ ## C) \
		1192	NEAREST_KERNEL_RAW(bgra_ ## C) \
		1193
1120	NEAREST_KERNELS(yuv420p)	1194	NEAREST_KERNELS(yuv420p)
1121	NEAREST_KERNELS(nv12)	1195	NEAREST_KERNELS(nv12)
1122	NEAREST_KERNELS(yuv444p)	1196	NEAREST_KERNELS(yuv444p)
@@ -1124,11 +1198,10 @@ NEAREST_KERNELS(p010le) · 1 unmodified line
1124	NEAREST_KERNELS(p016le)	1198	NEAREST_KERNELS(p016le)
1125	NEAREST_KERNELS(yuv444p16le)	1199	NEAREST_KERNELS(yuv444p16le)
1126		1200
1127	NEAREST_KERNEL_RAW(bgr0_bgr0)	1201	NEAREST_KERNELS_RGB(rgb0)
1128	NEAREST_KERNEL_RAW(rgb0_rgb0)	1202	NEAREST_KERNELS_RGB(bgr0)
1129	NEAREST_KERNEL_RAW(bgr0_rgb0)	1203	NEAREST_KERNELS_RGB(rgba)
1130	NEAREST_KERNEL_RAW(rgb0_bgr0)	1204	NEAREST_KERNELS_RGB(bgra)
1131
1132		1205
1133	#define BILINEAR_KERNEL(C, S) \	1206	#define BILINEAR_KERNEL(C, S) \
1134	__global__ void Subsample_Bilinear_##C##S( \	1207	__global__ void Subsample_Bilinear_##C##S( \
@@ -1152,6 +1225,12 @@ NEAREST_KERNEL_RAW(rgb0_bgr0) · 17 unmodified lines
1152	BILINEAR_KERNEL_RAW(p016le_ ## C) \	1225	BILINEAR_KERNEL_RAW(p016le_ ## C) \
1153	BILINEAR_KERNEL_RAW(yuv444p16le_ ## C)	1226	BILINEAR_KERNEL_RAW(yuv444p16le_ ## C)
1154		1227
		1228	#define BILINEAR_KERNELS_RGB(C) \
		1229	BILINEAR_KERNEL_RAW(rgb0_ ## C) \
		1230	BILINEAR_KERNEL_RAW(bgr0_ ## C) \
		1231	BILINEAR_KERNEL_RAW(rgba_ ## C) \
		1232	BILINEAR_KERNEL_RAW(bgra_ ## C)
		1233
1155	BILINEAR_KERNELS(yuv420p)	1234	BILINEAR_KERNELS(yuv420p)
1156	BILINEAR_KERNELS(nv12)	1235	BILINEAR_KERNELS(nv12)
1157	BILINEAR_KERNELS(yuv444p)	1236	BILINEAR_KERNELS(yuv444p)
@@ -1159,10 +1238,10 @@ BILINEAR_KERNELS(p010le) · 1 unmodified line
1159	BILINEAR_KERNELS(p016le)	1238	BILINEAR_KERNELS(p016le)
1160	BILINEAR_KERNELS(yuv444p16le)	1239	BILINEAR_KERNELS(yuv444p16le)
1161		1240
1162	BILINEAR_KERNEL_RAW(bgr0_bgr0)	1241	BILINEAR_KERNELS_RGB(rgb0)
1163	BILINEAR_KERNEL_RAW(rgb0_rgb0)	1242	BILINEAR_KERNELS_RGB(bgr0)
1164	BILINEAR_KERNEL_RAW(bgr0_rgb0)	1243	BILINEAR_KERNELS_RGB(rgba)
1165	BILINEAR_KERNEL_RAW(rgb0_bgr0)	1244	BILINEAR_KERNELS_RGB(bgra)
1166		1245
1167	#define BICUBIC_KERNEL(C, S) \	1246	#define BICUBIC_KERNEL(C, S) \
1168	__global__ void Subsample_Bicubic_##C##S( \	1247	__global__ void Subsample_Bicubic_##C##S( \
@@ -1186,6 +1265,12 @@ BILINEAR_KERNEL_RAW(rgb0_bgr0) · 17 unmodified lines
1186	BICUBIC_KERNEL_RAW(p016le_ ## C) \	1265	BICUBIC_KERNEL_RAW(p016le_ ## C) \
1187	BICUBIC_KERNEL_RAW(yuv444p16le_ ## C)	1266	BICUBIC_KERNEL_RAW(yuv444p16le_ ## C)
1188		1267
		1268	#define BICUBIC_KERNELS_RGB(C) \
		1269	BICUBIC_KERNEL_RAW(rgb0_ ## C) \
		1270	BICUBIC_KERNEL_RAW(bgr0_ ## C) \
		1271	BICUBIC_KERNEL_RAW(rgba_ ## C) \
		1272	BICUBIC_KERNEL_RAW(bgra_ ## C)
		1273
1189	BICUBIC_KERNELS(yuv420p)	1274	BICUBIC_KERNELS(yuv420p)
1190	BICUBIC_KERNELS(nv12)	1275	BICUBIC_KERNELS(nv12)
1191	BICUBIC_KERNELS(yuv444p)	1276	BICUBIC_KERNELS(yuv444p)
@@ -1193,11 +1278,10 @@ BICUBIC_KERNELS(p010le) · 1 unmodified line
1193	BICUBIC_KERNELS(p016le)	1278	BICUBIC_KERNELS(p016le)
1194	BICUBIC_KERNELS(yuv444p16le)	1279	BICUBIC_KERNELS(yuv444p16le)
1195		1280
1196	BICUBIC_KERNEL_RAW(bgr0_bgr0)	1281	BICUBIC_KERNELS_RGB(rgb0)
1197	BICUBIC_KERNEL_RAW(rgb0_rgb0)	1282	BICUBIC_KERNELS_RGB(bgr0)
1198	BICUBIC_KERNEL_RAW(bgr0_rgb0)	1283	BICUBIC_KERNELS_RGB(rgba)
1199	BICUBIC_KERNEL_RAW(rgb0_bgr0)	1284	BICUBIC_KERNELS_RGB(bgra)
1200
1201		1285
1202	#define LANCZOS_KERNEL(C, S) \	1286	#define LANCZOS_KERNEL(C, S) \
1203	__global__ void Subsample_Lanczos_##C##S( \	1287	__global__ void Subsample_Lanczos_##C##S( \
@@ -1221,6 +1305,12 @@ BICUBIC_KERNEL_RAW(rgb0_bgr0) · 17 unmodified lines
1221	LANCZOS_KERNEL_RAW(p016le_ ## C) \	1305	LANCZOS_KERNEL_RAW(p016le_ ## C) \
1222	LANCZOS_KERNEL_RAW(yuv444p16le_ ## C)	1306	LANCZOS_KERNEL_RAW(yuv444p16le_ ## C)
1223		1307
		1308	#define LANCZOS_KERNELS_RGB(C) \
		1309	LANCZOS_KERNEL_RAW(rgb0_ ## C) \
		1310	LANCZOS_KERNEL_RAW(bgr0_ ## C) \
		1311	LANCZOS_KERNEL_RAW(rgba_ ## C) \
		1312	LANCZOS_KERNEL_RAW(bgra_ ## C)
		1313
1224	LANCZOS_KERNELS(yuv420p)	1314	LANCZOS_KERNELS(yuv420p)
1225	LANCZOS_KERNELS(nv12)	1315	LANCZOS_KERNELS(nv12)
1226	LANCZOS_KERNELS(yuv444p)	1316	LANCZOS_KERNELS(yuv444p)
@@ -1228,9 +1318,8 @@ LANCZOS_KERNELS(p010le) · 1 unmodified line
1228	LANCZOS_KERNELS(p016le)	1318	LANCZOS_KERNELS(p016le)
1229	LANCZOS_KERNELS(yuv444p16le)	1319	LANCZOS_KERNELS(yuv444p16le)
1230		1320
1231	LANCZOS_KERNEL_RAW(bgr0_bgr0)	1321	LANCZOS_KERNELS_RGB(rgb0)
1232	LANCZOS_KERNEL_RAW(rgb0_rgb0)	1322	LANCZOS_KERNELS_RGB(bgr0)
1233	LANCZOS_KERNEL_RAW(bgr0_rgb0)	1323	LANCZOS_KERNELS_RGB(rgba)
1234	LANCZOS_KERNEL_RAW(rgb0_bgr0)	1324	LANCZOS_KERNELS_RGB(bgra)
1235
1236	}	1325	}

avfilter/scale_cuda: add support for rgb32/bgr32 conversions

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code