lavc/vp9dsp: R-V V mc bilin h v

C908 X60 vp9_avg_bilin_4h_8bpp_c : 5.5 4.7 vp9_avg_bilin_4h_8bpp_rvv_i32 : 1.7 1.5 vp9_avg_bilin_4v_8bpp_c : 5.5 4.7 vp9_avg_bilin_4v_8bpp_rvv_i32 : 1.5 1.2 vp9_avg_bilin_8h_8bpp_c : 20.0 17.7 vp9_avg_bilin_8h_8bpp_rvv_i32 : 3.0 2.7 vp9_avg_bilin_8v_8bpp_c : 20.7 18.7 vp9_avg_bilin_8v_8bpp_rvv_i32 : 3.0 2.7 vp9_avg_bilin_16h_8bpp_c : 78.2 69.7 vp9_avg_bilin_16h_8bpp_rvv_i32 : 7.0 6.2 vp9_avg_bilin_16v_8bpp_c : 98.5 73.2 vp9_avg_bilin_16v_8bpp_rvv_i32 : 7.0 6.0 vp9_avg_bilin_32h_8bpp_c : 325.5 275.5 vp9_avg_bilin_32h_8bpp_rvv_i32 : 23.0 20.5 vp9_avg_bilin_32v_8bpp_c : 342.2 290.0 vp9_avg_bilin_32v_8bpp_rvv_i32 : 21.7 19.5 vp9_avg_bilin_64h_8bpp_c : 1263.7 1095.7 vp9_avg_bilin_64h_8bpp_rvv_i32 : 91.2 81.2 vp9_avg_bilin_64v_8bpp_c : 1331.7 1155.2 vp9_avg_bilin_64v_8bpp_rvv_i32 : 91.2 81.0 vp9_put_bilin_4h_8bpp_c : 4.5 4.0 vp9_put_bilin_4h_8bpp_rvv_i32 : 1.0 1.0 vp9_put_bilin_4v_8bpp_c : 4.7 4.2 vp9_put_bilin_4v_8bpp_rvv_i32 : 1.0 1.0 vp9_put_bilin_8h_8bpp_c : 16.7 15.0 vp9_put_bilin_8h_8bpp_rvv_i32 : 2.2 2.0 vp9_put_bilin_8v_8bpp_c : 17.5 15.7 vp9_put_bilin_8v_8bpp_rvv_i32 : 2.2 2.0 vp9_put_bilin_16h_8bpp_c : 65.2 58.0 vp9_put_bilin_16h_8bpp_rvv_i32 : 6.0 5.5 vp9_put_bilin_16v_8bpp_c : 69.2 61.7 vp9_put_bilin_16v_8bpp_rvv_i32 : 5.7 5.2 vp9_put_bilin_32h_8bpp_c : 273.2 229.0 vp9_put_bilin_32h_8bpp_rvv_i32 : 19.7 17.7 vp9_put_bilin_32v_8bpp_c : 290.5 243.7 vp9_put_bilin_32v_8bpp_rvv_i32 : 18.7 16.7 vp9_put_bilin_64h_8bpp_c : 1040.5 910.5 vp9_put_bilin_64h_8bpp_rvv_i32 : 82.5 73.0 vp9_put_bilin_64v_8bpp_c : 1108.5 971.0 vp9_put_bilin_64v_8bpp_rvv_i32 : 82.2 73.2

sunyuechi committed 2 years ago

Commit 9edd2e71 parent 8f36c6f3 files · +132 −6Browse files

Unified · Split

3 files

libavcodec/riscv/vp9_mc_rvv.S +105 −0
libavcodec/riscv/vp9dsp.h +6 −6
libavcodec/riscv/vp9dsp_init.c +21 −0

libavcodec/riscv/vp9_mc_rvv.S+105−0

@@ -54,6 +54,111 @@ func ff_vp9_avg\len\()_rvv, zve32x · 53 unmodified lines
54	54	endfunc
55	55	.endm
56	56
	57	+.macro bilin_load dst, mn, type
	58	+.ifc \type,v
	59	+ add t4, a2, a3
	60	+.else
	61	+ addi t4, a2, 1
	62	+.endif
	63	+ vle8.v v4, (a2)
	64	+ vle8.v v0, (t4)
	65	+ vwmulu.vx v16, v0, \mn
	66	+ vwmaccsu.vx v16, t1, v4
	67	+ vwadd.wx v16, v16, t3
	68	+ vnsra.wi v16, v16, 4
	69	+ vadd.vv \dst, v16, v4
	70	+ add a2, a2, a3
	71	+.endm
	72	+
	73	+.macro bilin_h_v op, type, mn
	74	+func ff_\op\()_vp9_bilin_64\type\()_rvv, zve32x
	75	+ lpad 0
	76	+ vsetvlstatic8 64, t0, 64
	77	+.ifc \op,avg
	78	+ csrwi vxrm, 0
	79	+.endif
	80	+ li t3, 8
	81	+ neg t1, \mn
	82	+1:
	83	+ addi a4, a4, -1
	84	+ bilin_load v0, \mn, \type
	85	+.ifc \op,avg
	86	+ vle8.v v16, (a0)
	87	+ vaaddu.vv v0, v0, v16
	88	+.endif
	89	+ vse8.v v0, (a0)
	90	+ add a0, a0, a1
	91	+ bnez a4, 1b
	92	+ ret
	93	+
	94	+.Lbilin_\type\op:
	95	+.ifc \op,avg
	96	+ csrwi vxrm, 0
	97	+.endif
	98	+ li t4, 8
	99	+ neg t1, \mn
	100	+1:
	101	+ addi a4, a4, -2
	102	+ add t6, a0, a1
	103	+ add t0, a2, a3
	104	+ vle8.v v8, (a2)
	105	+ vle8.v v4, (t0)
	106	+.ifc \type,v
	107	+ add t2, t0, a3
	108	+ vwmulu.vx v16, v4, \mn
	109	+.else
	110	+ addi t3, a2, 1
	111	+ addi t2, t0, 1
	112	+ vle8.v v0, (t3)
	113	+ vwmulu.vx v16, v0, \mn
	114	+.endif
	115	+ vle8.v v12, (t2)
	116	+ vwmulu.vx v20, v12, \mn
	117	+ vwmaccsu.vx v16, t1, v8
	118	+ vwmaccsu.vx v20, t1, v4
	119	+ vwadd.wx v16, v16, t4
	120	+ vwadd.wx v20, v20, t4
	121	+ vnsra.wi v16, v16, 4
	122	+ vnsra.wi v20, v20, 4
	123	+ vadd.vv v0, v16, v8
	124	+ vadd.vv v12, v20, v4
	125	+.ifc \op,avg
	126	+ vle8.v v16, (a0)
	127	+ vle8.v v20, (t6)
	128	+ vaaddu.vv v0, v0, v16
	129	+ vaaddu.vv v12, v12, v20
	130	+.endif
	131	+ vse8.v v0, (a0)
	132	+ vse8.v v12, (t6)
	133	+ add a2, t0, a3
	134	+ add a0, t6, a1
	135	+ bnez a4, 1b
	136	+
	137	+ ret
	138	+endfunc
	139	+.endm
	140	+
57	141	.irp len, 64, 32, 16, 8, 4
58	142	copy_avg \len
59	143	.endr
	144	+
	145	+bilin_h_v put, h, a5
	146	+bilin_h_v avg, h, a5
	147	+bilin_h_v put, v, a6
	148	+bilin_h_v avg, v, a6
	149	+
	150	+.macro func_bilin_h_v len, op, type
	151	+func ff_\op\()_vp9_bilin_\len\()\type\()_rvv, zve32x
	152	+ lpad 0
	153	+ vsetvlstatic8 \len, t0, 64
	154	+ j .Lbilin_\type\()\op
	155	+endfunc
	156	+.endm
	157	+
	158	+.irp len, 32, 16, 8, 4
	159	+ .irp op, put, avg
	160	+ .irp type, h, v
	161	+ func_bilin_h_v \len, \op, \type
	162	+ .endr
	163	+ .endr
	164	+.endr

libavcodec/riscv/vp9dsp.h+6−6

@@ -107,27 +107,27 @@ void ff_avg_8tap_##type##_##SIZE##hv_rvv(uint8_t *dst, ptrdiff_t dststride, \ · 106 unmodified lines
107	107	int h, int mx, int my);
108	108
109	109	#define VP9_BILINEAR_RISCV_RVV_FUNC(SIZE) \
110		−void ff_put_bilin_##SIZE##h_rvv(uint8_t *dst, ptrdiff_t dststride, \
	110	+void ff_put_vp9_bilin_##SIZE##h_rvv(uint8_t *dst, ptrdiff_t dststride, \
111	111	const uint8_t *src, ptrdiff_t srcstride, \
112	112	int h, int mx, int my); \
113	113	\
114		−void ff_put_bilin_##SIZE##v_rvv(uint8_t *dst, ptrdiff_t dststride, \
	114	+void ff_put_vp9_bilin_##SIZE##v_rvv(uint8_t *dst, ptrdiff_t dststride, \
115	115	const uint8_t *src, ptrdiff_t srcstride, \
116	116	int h, int mx, int my); \
117	117	\
118		−void ff_put_bilin_##SIZE##hv_rvv(uint8_t *dst, ptrdiff_t dststride, \
	118	+void ff_put_vp9_bilin_##SIZE##hv_rvv(uint8_t *dst, ptrdiff_t dststride, \
119	119	const uint8_t *src, ptrdiff_t srcstride, \
120	120	int h, int mx, int my); \
121	121	\
122		−void ff_avg_bilin_##SIZE##h_rvv(uint8_t *dst, ptrdiff_t dststride, \
	122	+void ff_avg_vp9_bilin_##SIZE##h_rvv(uint8_t *dst, ptrdiff_t dststride, \
123	123	const uint8_t *src, ptrdiff_t srcstride, \
124	124	int h, int mx, int my); \
125	125	\
126		−void ff_avg_bilin_##SIZE##v_rvv(uint8_t *dst, ptrdiff_t dststride, \
	126	+void ff_avg_vp9_bilin_##SIZE##v_rvv(uint8_t *dst, ptrdiff_t dststride, \
127	127	const uint8_t *src, ptrdiff_t srcstride, \
128	128	int h, int mx, int my); \
129	129	\
130		−void ff_avg_bilin_##SIZE##hv_rvv(uint8_t *dst, ptrdiff_t dststride, \
	130	+void ff_avg_vp9_bilin_##SIZE##hv_rvv(uint8_t *dst, ptrdiff_t dststride, \
131	131	const uint8_t *src, ptrdiff_t srcstride, \
132	132	int h, int mx, int my);
133	133

libavcodec/riscv/vp9dsp_init.c+21−0

@@ -63,6 +63,27 @@ static av_cold void vp9dsp_mc_init_riscv(VP9DSPContext *dsp, int bpp) · 62 unmodified lines
63	63	init_fpel(3, 8);
64	64	init_fpel(4, 4);
65	65
	66	+ dsp->mc[0][FILTER_BILINEAR ][0][0][1] = ff_put_vp9_bilin_64v_rvv;
	67	+ dsp->mc[0][FILTER_BILINEAR ][0][1][0] = ff_put_vp9_bilin_64h_rvv;
	68	+ dsp->mc[0][FILTER_BILINEAR ][1][0][1] = ff_avg_vp9_bilin_64v_rvv;
	69	+ dsp->mc[0][FILTER_BILINEAR ][1][1][0] = ff_avg_vp9_bilin_64h_rvv;
	70	+ dsp->mc[1][FILTER_BILINEAR ][0][0][1] = ff_put_vp9_bilin_32v_rvv;
	71	+ dsp->mc[1][FILTER_BILINEAR ][0][1][0] = ff_put_vp9_bilin_32h_rvv;
	72	+ dsp->mc[1][FILTER_BILINEAR ][1][0][1] = ff_avg_vp9_bilin_32v_rvv;
	73	+ dsp->mc[1][FILTER_BILINEAR ][1][1][0] = ff_avg_vp9_bilin_32h_rvv;
	74	+ dsp->mc[2][FILTER_BILINEAR ][0][0][1] = ff_put_vp9_bilin_16v_rvv;
	75	+ dsp->mc[2][FILTER_BILINEAR ][0][1][0] = ff_put_vp9_bilin_16h_rvv;
	76	+ dsp->mc[2][FILTER_BILINEAR ][1][0][1] = ff_avg_vp9_bilin_16v_rvv;
	77	+ dsp->mc[2][FILTER_BILINEAR ][1][1][0] = ff_avg_vp9_bilin_16h_rvv;
	78	+ dsp->mc[3][FILTER_BILINEAR ][0][0][1] = ff_put_vp9_bilin_8v_rvv;
	79	+ dsp->mc[3][FILTER_BILINEAR ][0][1][0] = ff_put_vp9_bilin_8h_rvv;
	80	+ dsp->mc[3][FILTER_BILINEAR ][1][0][1] = ff_avg_vp9_bilin_8v_rvv;
	81	+ dsp->mc[3][FILTER_BILINEAR ][1][1][0] = ff_avg_vp9_bilin_8h_rvv;
	82	+ dsp->mc[4][FILTER_BILINEAR ][0][0][1] = ff_put_vp9_bilin_4v_rvv;
	83	+ dsp->mc[4][FILTER_BILINEAR ][0][1][0] = ff_put_vp9_bilin_4h_rvv;
	84	+ dsp->mc[4][FILTER_BILINEAR ][1][0][1] = ff_avg_vp9_bilin_4v_rvv;
	85	+ dsp->mc[4][FILTER_BILINEAR ][1][1][0] = ff_avg_vp9_bilin_4h_rvv;
	86	+
66	87	#undef init_fpel
67	88	}
68	89	#endif

lavc/vp9dsp: R-V V mc bilin h v

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code