aarch64/vvc: Add put_qpel_h_* and put_qpel_uni_h_*

Just share hevc implementation. checkasm --test=vvc_mc --benchmark: put_luma_h_8_4x4_c: 0.2 ( 1.00x) put_luma_h_8_4x4_neon: 0.2 ( 1.00x) put_luma_h_8_8x8_c: 1.0 ( 1.00x) put_luma_h_8_8x8_neon: 0.2 ( 4.33x) put_luma_h_8_16x16_c: 3.2 ( 1.00x) put_luma_h_8_16x16_neon: 1.2 ( 2.63x) put_luma_h_8_32x32_c: 13.7 ( 1.00x) put_luma_h_8_32x32_neon: 4.0 ( 3.45x) put_luma_h_8_64x64_c: 48.2 ( 1.00x) put_luma_h_8_64x64_neon: 15.7 ( 3.07x) put_luma_h_8_128x128_c: 203.5 ( 1.00x) put_luma_h_8_128x128_neon: 62.0 ( 3.28x) put_uni_h_luma_8_4x4_c: 0.2 ( 1.00x) put_uni_h_luma_8_4x4_neon: 0.2 ( 1.00x) put_uni_h_luma_8_8x8_c: 1.5 ( 1.00x) put_uni_h_luma_8_8x8_neon: 0.2 ( 6.56x) put_uni_h_luma_8_16x16_c: 5.7 ( 1.00x) put_uni_h_luma_8_16x16_neon: 1.2 ( 4.67x) put_uni_h_luma_8_32x32_c: 24.0 ( 1.00x) put_uni_h_luma_8_32x32_neon: 4.7 ( 5.07x) put_uni_h_luma_8_64x64_c: 90.0 ( 1.00x) put_uni_h_luma_8_64x64_neon: 17.0 ( 5.30x) put_uni_h_luma_8_128x128_c: 357.7 ( 1.00x) put_uni_h_luma_8_128x128_neon: 67.5 ( 5.30x)

Zhao Zhili committed 2 years ago

Commit 20f2bf51 parent 46f07ce4 files · +171 −59Browse files

Showing the changes to libavcodec (4 of 4 files). Show all

Unified · Split

4 files

libavcodec/aarch64/h26x/dsp.h +13 −0
libavcodec/aarch64/h26x/qpel_neon.S +143 −59
libavcodec/aarch64/vvc/Makefile +1 −0
libavcodec/aarch64/vvc/dsp_init.c +14 −0

libavcodec/aarch64/h26x/dsp.h+13−0

@@ -235,4 +235,17 @@ NEON8_FNPROTO(qpel_bi_hv, (uint8_t *dst, ptrdiff_t dststride, · 234 unmodified lines
235	235	const uint8_t src, ptrdiff_t srcstride, const int16_t src2,
236	236	int height, intptr_t mx, intptr_t my, int width), _i8mm);
237	237
	238	+#undef NEON8_FNPROTO_PARTIAL_4
	239	+#define NEON8_FNPROTO_PARTIAL_4(fn, args, ext) \
	240	+ void ff_vvc_put_##fn##_h4_8_neon##ext args; \
	241	+ void ff_vvc_put_##fn##_h8_8_neon##ext args; \
	242	+ void ff_vvc_put_##fn##_h16_8_neon##ext args; \
	243	+ void ff_vvc_put_##fn##_h32_8_neon##ext args;
	244	+
	245	+NEON8_FNPROTO_PARTIAL_4(qpel, (int16_t dst, const uint8_t _src, ptrdiff_t _srcstride, int height,
	246	+ const int8_t hf, const int8_t vf, int width),)
	247	+
	248	+NEON8_FNPROTO_PARTIAL_4(qpel_uni, (uint8_t _dst, ptrdiff_t _dststride, const uint8_t _src,
	249	+ ptrdiff_t _srcstride, int height, const int8_t hf, const int8_t vf, int width),)
	250	+
238	251	#endif

libavcodec/aarch64/h26x/qpel_neon.S+143−59

@@ -21,7 +21,8 @@ · 20 unmodified lines
21	21	*/
22	22
23	23	#include "libavutil/aarch64/asm.S"
24		−#define MAX_PB_SIZE 64
	24	+#define HEVC_MAX_PB_SIZE 64
	25	+#define VVC_MAX_PB_SIZE 128
25	26
26	27	const qpel_filters, align=4
27	28	.byte 0, 0, 0, 0, 0, 0, 0, 0
@@ -44,6 +45,11 @@ endconst · 16 unmodified lines
44	45	sxtl v0.8h, v0.8b
45	46	.endm
46	47
	48	+.macro vvc_load_filter m
	49	+ ld1 {v0.8b}, [\m]
	50	+ sxtl v0.8h, v0.8b
	51	+.endm
	52	+
47	53	.macro load_qpel_filterb freg, xreg
48	54	movrel \xreg, qpel_filters_abs
49	55	add \xreg, \xreg, \freg, lsl #3
@@ -212,22 +218,40 @@ function ff_hevc_put_hevc_h4_8_neon, export=0 · 162 unmodified lines
212	218	endfunc
213	219	.endif
214	220
	221	+.ifnc \type, qpel_bi
	222	+function ff_vvc_put_\type\()_h4_8_neon, export=1
	223	+ vvc_load_filter mx
	224	+ sub src, src, #3
	225	+ mov mx, x30
	226	+.ifc \type, qpel
	227	+ mov dststride, #(VVC_MAX_PB_SIZE << 1)
	228	+ lsl x13, srcstride, #1 // srcstridel
	229	+ mov x14, #(VVC_MAX_PB_SIZE << 2)
	230	+.else
	231	+ lsl x14, dststride, #1 // dststridel
	232	+ lsl x13, srcstride, #1 // srcstridel
	233	+.endif
	234	+ b 1f
	235	+endfunc
	236	+.endif // !qpel_bi
	237	+
215	238	function ff_hevc_put_hevc_\type\()_h4_8_neon, export=1
216	239	load_filter mx
217	240	.ifc \type, qpel_bi
218		− mov x16, #(MAX_PB_SIZE << 2) // src2bstridel
219		− add x15, x4, #(MAX_PB_SIZE << 1) // src2b
	241	+ mov x16, #(HEVC_MAX_PB_SIZE << 2) // src2bstridel
	242	+ add x15, x4, #(HEVC_MAX_PB_SIZE << 1) // src2b
220	243	.endif
221	244	sub src, src, #3
222	245	mov mx, x30
223	246	.ifc \type, qpel
224		− mov dststride, #(MAX_PB_SIZE << 1)
	247	+ mov dststride, #(HEVC_MAX_PB_SIZE << 1)
225	248	lsl x13, srcstride, #1 // srcstridel
226		− mov x14, #(MAX_PB_SIZE << 2)
	249	+ mov x14, #(HEVC_MAX_PB_SIZE << 2)
227	250	.else
228	251	lsl x14, dststride, #1 // dststridel
229	252	lsl x13, srcstride, #1 // srcstridel
230	253	.endif
	254	+1:
231	255	add x10, dst, dststride // dstb
232	256	add x12, src, srcstride // srcb
233	257	0: ld1 {v16.8b, v17.8b}, [src], x13
@@ -283,15 +307,15 @@ endfunc · 49 unmodified lines
283	307	function ff_hevc_put_hevc_\type\()_h6_8_neon, export=1
284	308	load_filter mx
285	309	.ifc \type, qpel_bi
286		− mov x16, #(MAX_PB_SIZE << 2) // src2bstridel
287		− add x15, x4, #(MAX_PB_SIZE << 1) // src2b
	310	+ mov x16, #(HEVC_MAX_PB_SIZE << 2) // src2bstridel
	311	+ add x15, x4, #(HEVC_MAX_PB_SIZE << 1) // src2b
288	312	.endif
289	313	sub src, src, #3
290	314	mov mx, x30
291	315	.ifc \type, qpel
292		− mov dststride, #(MAX_PB_SIZE << 1)
	316	+ mov dststride, #(HEVC_MAX_PB_SIZE << 1)
293	317	lsl x13, srcstride, #1 // srcstridel
294		− mov x14, #((MAX_PB_SIZE << 2) - 8)
	318	+ mov x14, #((HEVC_MAX_PB_SIZE << 2) - 8)
295	319	.else
296	320	lsl x14, dststride, #1 // dststridel
297	321	lsl x13, srcstride, #1 // srcstridel
@@ -333,22 +357,40 @@ function ff_hevc_put_hevc_\type\()_h6_8_neon, export=1 · 35 unmodified lines
333	357	ret mx
334	358	endfunc
335	359
	360	+.ifnc \type, qpel_bi
	361	+function ff_vvc_put_\type\()_h8_8_neon, export=1
	362	+ vvc_load_filter mx
	363	+ sub src, src, #3
	364	+ mov mx, x30
	365	+.ifc \type, qpel
	366	+ mov dststride, #(VVC_MAX_PB_SIZE << 1)
	367	+ lsl x13, srcstride, #1 // srcstridel
	368	+ mov x14, #(VVC_MAX_PB_SIZE << 2)
	369	+.else
	370	+ lsl x14, dststride, #1 // dststridel
	371	+ lsl x13, srcstride, #1 // srcstridel
	372	+.endif
	373	+ b 1f
	374	+endfunc
	375	+.endif // !qpel_bi
	376	+
336	377	function ff_hevc_put_hevc_\type\()_h8_8_neon, export=1
337	378	load_filter mx
338	379	.ifc \type, qpel_bi
339		− mov x16, #(MAX_PB_SIZE << 2) // src2bstridel
340		− add x15, x4, #(MAX_PB_SIZE << 1) // src2b
	380	+ mov x16, #(HEVC_MAX_PB_SIZE << 2) // src2bstridel
	381	+ add x15, x4, #(HEVC_MAX_PB_SIZE << 1) // src2b
341	382	.endif
342	383	sub src, src, #3
343	384	mov mx, x30
344	385	.ifc \type, qpel
345		− mov dststride, #(MAX_PB_SIZE << 1)
	386	+ mov dststride, #(HEVC_MAX_PB_SIZE << 1)
346	387	lsl x13, srcstride, #1 // srcstridel
347		− mov x14, #(MAX_PB_SIZE << 2)
	388	+ mov x14, #(HEVC_MAX_PB_SIZE << 2)
348	389	.else
349	390	lsl x14, dststride, #1 // dststridel
350	391	lsl x13, srcstride, #1 // srcstridel
351	392	.endif
	393	+1:
352	394	add x10, dst, dststride // dstb
353	395	add x12, src, srcstride // srcb
354	396	0: ld1 {v16.8b, v17.8b}, [src], x13
@@ -415,16 +457,16 @@ function ff_hevc_put_hevc_\type\()_h12_8_neon, export=1 · 60 unmodified lines
415	457	sxtw height, heightw
416	458	.ifc \type, qpel_bi
417	459	ldrh w8, [sp] // width
418		− mov x16, #(MAX_PB_SIZE << 2) // src2bstridel
419		− lsl x17, height, #7 // src2b reset (height * (MAX_PB_SIZE << 1))
420		− add x15, x4, #(MAX_PB_SIZE << 1) // src2b
	460	+ mov x16, #(HEVC_MAX_PB_SIZE << 2) // src2bstridel
	461	+ lsl x17, height, #7 // src2b reset (height * (HEVC_MAX_PB_SIZE << 1))
	462	+ add x15, x4, #(HEVC_MAX_PB_SIZE << 1) // src2b
421	463	.endif
422	464	sub src, src, #3
423	465	mov mx, x30
424	466	.ifc \type, qpel
425		− mov dststride, #(MAX_PB_SIZE << 1)
	467	+ mov dststride, #(HEVC_MAX_PB_SIZE << 1)
426	468	lsl x13, srcstride, #1 // srcstridel
427		− mov x14, #((MAX_PB_SIZE << 2) - 16)
	469	+ mov x14, #((HEVC_MAX_PB_SIZE << 2) - 16)
428	470	.else
429	471	lsl x14, dststride, #1 // dststridel
430	472	lsl x13, srcstride, #1 // srcstridel
@@ -497,25 +539,45 @@ function ff_hevc_put_hevc_\type\()_h12_8_neon, export=1 · 66 unmodified lines
497	539	ret mx
498	540	endfunc
499	541
	542	+.ifnc \type, qpel_bi
	543	+function ff_vvc_put_\type\()_h16_8_neon, export=1
	544	+ vvc_load_filter mx
	545	+ sxtw height, heightw
	546	+ mov mx, x30
	547	+ sub src, src, #3
	548	+ mov mx, x30
	549	+.ifc \type, qpel
	550	+ mov dststride, #(VVC_MAX_PB_SIZE << 1)
	551	+ lsl x13, srcstride, #1 // srcstridel
	552	+ mov x14, #(VVC_MAX_PB_SIZE << 2)
	553	+.else
	554	+ lsl x14, dststride, #1 // dststridel
	555	+ lsl x13, srcstride, #1 // srcstridel
	556	+.endif
	557	+ b 0f
	558	+endfunc
	559	+.endif // !qpel_bi
	560	+
500	561	function ff_hevc_put_hevc_\type\()_h16_8_neon, export=1
501	562	load_filter mx
502	563	sxtw height, heightw
503	564	mov mx, x30
504	565	.ifc \type, qpel_bi
505	566	ldrh w8, [sp] // width
506		− mov x16, #(MAX_PB_SIZE << 2) // src2bstridel
507		− add x15, x4, #(MAX_PB_SIZE << 1) // src2b
	567	+ mov x16, #(HEVC_MAX_PB_SIZE << 2) // src2bstridel
	568	+ add x15, x4, #(HEVC_MAX_PB_SIZE << 1) // src2b
508	569	.endif
509	570	sub src, src, #3
510	571	mov mx, x30
511	572	.ifc \type, qpel
512		− mov dststride, #(MAX_PB_SIZE << 1)
	573	+ mov dststride, #(HEVC_MAX_PB_SIZE << 1)
513	574	lsl x13, srcstride, #1 // srcstridel
514		− mov x14, #(MAX_PB_SIZE << 2)
	575	+ mov x14, #(HEVC_MAX_PB_SIZE << 2)
515	576	.else
516	577	lsl x14, dststride, #1 // dststridel
517	578	lsl x13, srcstride, #1 // srcstridel
518	579	.endif
	580	+0:
519	581	add x10, dst, dststride // dstb
520	582	add x12, src, srcstride // srcb
521	583
@@ -555,29 +617,51 @@ function ff_hevc_put_hevc_\type\()_h16_8_neon, export=1 · 33 unmodified lines
555	617	ret mx
556	618	endfunc
557	619
	620	+.ifnc \type, qpel_bi
	621	+function ff_vvc_put_\type\()_h32_8_neon, export=1
	622	+ vvc_load_filter mx
	623	+ sxtw height, heightw
	624	+ mov mx, x30
	625	+ sub src, src, #3
	626	+ mov mx, x30
	627	+.ifc \type, qpel
	628	+ mov dststride, #(VVC_MAX_PB_SIZE << 1)
	629	+ lsl x13, srcstride, #1 // srcstridel
	630	+ mov x14, #(VVC_MAX_PB_SIZE << 2)
	631	+ sub x14, x14, width, uxtw #1
	632	+.else
	633	+ lsl x14, dststride, #1 // dststridel
	634	+ lsl x13, srcstride, #1 // srcstridel
	635	+ sub x14, x14, width, uxtw
	636	+.endif
	637	+ b 1f
	638	+endfunc
	639	+.endif // !qpel_bi
	640	+
558	641	function ff_hevc_put_hevc_\type\()_h32_8_neon, export=1
559	642	load_filter mx
560	643	sxtw height, heightw
561	644	mov mx, x30
562	645	.ifc \type, qpel_bi
563	646	ldrh w8, [sp] // width
564		− mov x16, #(MAX_PB_SIZE << 2) // src2bstridel
	647	+ mov x16, #(HEVC_MAX_PB_SIZE << 2) // src2bstridel
565	648	lsl x17, x5, #7 // src2b reset
566		− add x15, x4, #(MAX_PB_SIZE << 1) // src2b
	649	+ add x15, x4, #(HEVC_MAX_PB_SIZE << 1) // src2b
567	650	sub x16, x16, width, uxtw #1
568	651	.endif
569	652	sub src, src, #3
570	653	mov mx, x30
571	654	.ifc \type, qpel
572		− mov dststride, #(MAX_PB_SIZE << 1)
	655	+ mov dststride, #(HEVC_MAX_PB_SIZE << 1)
573	656	lsl x13, srcstride, #1 // srcstridel
574		− mov x14, #(MAX_PB_SIZE << 2)
	657	+ mov x14, #(HEVC_MAX_PB_SIZE << 2)
575	658	sub x14, x14, width, uxtw #1
576	659	.else
577	660	lsl x14, dststride, #1 // dststridel
578	661	lsl x13, srcstride, #1 // srcstridel
579	662	sub x14, x14, width, uxtw
580	663	.endif
	664	+1:
581	665	sub x13, x13, width, uxtw
582	666	sub x13, x13, #8
583	667	add x10, dst, dststride // dstb
@@ -651,7 +735,7 @@ put_hevc qpel_bi · 67 unmodified lines
651	735	function ff_hevc_put_hevc_qpel_v4_8_neon, export=1
652	736	load_qpel_filterb x5, x4
653	737	sub x1, x1, x2, lsl #1
654		− mov x9, #(MAX_PB_SIZE * 2)
	738	+ mov x9, #(HEVC_MAX_PB_SIZE * 2)
655	739	sub x1, x1, x2
656	740	ldr s16, [x1]
657	741	ldr s17, [x1, x2]
@@ -680,7 +764,7 @@ endfunc · 22 unmodified lines
680	764	function ff_hevc_put_hevc_qpel_v6_8_neon, export=1
681	765	load_qpel_filterb x5, x4
682	766	sub x1, x1, x2, lsl #1
683		− mov x9, #(MAX_PB_SIZE * 2 - 8)
	767	+ mov x9, #(HEVC_MAX_PB_SIZE * 2 - 8)
684	768	sub x1, x1, x2
685	769	ldr d16, [x1]
686	770	ldr d17, [x1, x2]
@@ -709,7 +793,7 @@ endfunc · 22 unmodified lines
709	793	function ff_hevc_put_hevc_qpel_v8_8_neon, export=1
710	794	load_qpel_filterb x5, x4
711	795	sub x1, x1, x2, lsl #1
712		− mov x9, #(MAX_PB_SIZE * 2)
	796	+ mov x9, #(HEVC_MAX_PB_SIZE * 2)
713	797	sub x1, x1, x2
714	798	ldr d16, [x1]
715	799	ldr d17, [x1, x2]
@@ -737,7 +821,7 @@ endfunc · 21 unmodified lines
737	821	function ff_hevc_put_hevc_qpel_v12_8_neon, export=1
738	822	load_qpel_filterb x5, x4
739	823	sub x1, x1, x2, lsl #1
740		− mov x9, #(MAX_PB_SIZE * 2 - 16)
	824	+ mov x9, #(HEVC_MAX_PB_SIZE * 2 - 16)
741	825	sub x1, x1, x2
742	826	ldr q16, [x1]
743	827	ldr q17, [x1, x2]
@@ -768,7 +852,7 @@ endfunc · 24 unmodified lines
768	852	function ff_hevc_put_hevc_qpel_v16_8_neon, export=1
769	853	load_qpel_filterb x5, x4
770	854	sub x1, x1, x2, lsl #1
771		− mov x9, #(MAX_PB_SIZE * 2)
	855	+ mov x9, #(HEVC_MAX_PB_SIZE * 2)
772	856	sub x1, x1, x2
773	857	ldr q16, [x1]
774	858	ldr q17, [x1, x2]
@@ -802,7 +886,7 @@ function ff_hevc_put_hevc_qpel_v24_8_neon, export=1 · 27 unmodified lines
802	886	load_qpel_filterb x5, x4
803	887	sub x1, x1, x2, lsl #1
804	888	sub x1, x1, x2
805		− mov x9, #(MAX_PB_SIZE * 2)
	889	+ mov x9, #(HEVC_MAX_PB_SIZE * 2)
806	890	ld1 {v16.16b, v17.16b}, [x1], x2
807	891	ld1 {v18.16b, v19.16b}, [x1], x2
808	892	ld1 {v20.16b, v21.16b}, [x1], x2
@@ -833,7 +917,7 @@ function ff_hevc_put_hevc_qpel_v32_8_neon, export=1 · 24 unmodified lines
833	917	st1 {v8.8b-v11.8b}, [sp]
834	918	load_qpel_filterb x5, x4
835	919	sub x1, x1, x2, lsl #1
836		− mov x9, #(MAX_PB_SIZE * 2)
	920	+ mov x9, #(HEVC_MAX_PB_SIZE * 2)
837	921	sub x1, x1, x2
838	922	ld1 {v16.16b, v17.16b}, [x1], x2
839	923	ld1 {v18.16b, v19.16b}, [x1], x2
@@ -883,7 +967,7 @@ function ff_hevc_put_hevc_qpel_v64_8_neon, export=1 · 43 unmodified lines
883	967	load_qpel_filterb x5, x4
884	968	sub x1, x1, x2, lsl #1
885	969	sub x1, x1, x2
886		− mov x9, #(MAX_PB_SIZE * 2)
	970	+ mov x9, #(HEVC_MAX_PB_SIZE * 2)
887	971	0: mov x8, x1 // src
888	972	ld1 {v16.16b, v17.16b}, [x8], x2
889	973	mov w11, w3 // height
@@ -921,7 +1005,7 @@ function ff_hevc_put_hevc_qpel_bi_v4_8_neon, export=1 · 31 unmodified lines
921	1005	load_qpel_filterb x7, x6
922	1006	sub x2, x2, x3, lsl #1
923	1007	sub x2, x2, x3
924		− mov x12, #(MAX_PB_SIZE * 2)
	1008	+ mov x12, #(HEVC_MAX_PB_SIZE * 2)
925	1009	ld1 {v16.s}[0], [x2], x3
926	1010	ld1 {v17.s}[0], [x2], x3
927	1011	ld1 {v18.s}[0], [x2], x3
@@ -951,7 +1035,7 @@ function ff_hevc_put_hevc_qpel_bi_v6_8_neon, export=1 · 23 unmodified lines
951	1035	ld1 {v16.8b}, [x2], x3
952	1036	sub x1, x1, #4
953	1037	ld1 {v17.8b}, [x2], x3
954		− mov x12, #(MAX_PB_SIZE * 2)
	1038	+ mov x12, #(HEVC_MAX_PB_SIZE * 2)
955	1039	ld1 {v18.8b}, [x2], x3
956	1040	ld1 {v19.8b}, [x2], x3
957	1041	ld1 {v20.8b}, [x2], x3
@@ -977,7 +1061,7 @@ function ff_hevc_put_hevc_qpel_bi_v8_8_neon, export=1 · 19 unmodified lines
977	1061	load_qpel_filterb x7, x6
978	1062	sub x2, x2, x3, lsl #1
979	1063	sub x2, x2, x3
980		− mov x12, #(MAX_PB_SIZE * 2)
	1064	+ mov x12, #(HEVC_MAX_PB_SIZE * 2)
981	1065	ld1 {v16.8b}, [x2], x3
982	1066	ld1 {v17.8b}, [x2], x3
983	1067	ld1 {v18.8b}, [x2], x3
@@ -1006,7 +1090,7 @@ function ff_hevc_put_hevc_qpel_bi_v12_8_neon, export=1 · 22 unmodified lines
1006	1090	sub x2, x2, x3
1007	1091	sub x1, x1, #8
1008	1092	ld1 {v16.16b}, [x2], x3
1009		− mov x12, #(MAX_PB_SIZE * 2)
	1093	+ mov x12, #(HEVC_MAX_PB_SIZE * 2)
1010	1094	ld1 {v17.16b}, [x2], x3
1011	1095	ld1 {v18.16b}, [x2], x3
1012	1096	ld1 {v19.16b}, [x2], x3
@@ -1037,7 +1121,7 @@ function ff_hevc_put_hevc_qpel_bi_v16_8_neon, export=1 · 24 unmodified lines
1037	1121	load_qpel_filterb x7, x6
1038	1122	sub x2, x2, x3, lsl #1
1039	1123	sub x2, x2, x3
1040		− mov x12, #(MAX_PB_SIZE * 2)
	1124	+ mov x12, #(HEVC_MAX_PB_SIZE * 2)
1041	1125	ld1 {v16.16b}, [x2], x3
1042	1126	ld1 {v17.16b}, [x2], x3
1043	1127	ld1 {v18.16b}, [x2], x3
@@ -1092,7 +1176,7 @@ function ff_hevc_put_hevc_qpel_bi_v32_8_neon, export=1 · 48 unmodified lines
1092	1176	sub x2, x2, x3
1093	1177	load_qpel_filterb x7, x6
1094	1178	ldr w6, [sp, #64]
1095		− mov x12, #(MAX_PB_SIZE * 2)
	1179	+ mov x12, #(HEVC_MAX_PB_SIZE * 2)
1096	1180	0: mov x8, x2 // src
1097	1181	ld1 {v16.16b, v17.16b}, [x8], x3
1098	1182	mov w11, w5 // height
@@ -2147,7 +2231,7 @@ function ff_hevc_put_hevc_qpel_uni_w_v64_8_neon, export=1 · 1048 unmodified lines
2147	2231	endfunc
2148	2232
2149	2233	function hevc_put_hevc_qpel_uni_hv4_8_end_neon
2150		− mov x9, #(MAX_PB_SIZE * 2)
	2234	+ mov x9, #(HEVC_MAX_PB_SIZE * 2)
2151	2235	load_qpel_filterh x6, x5
2152	2236	ldr d16, [sp]
2153	2237	ldr d17, [sp, x9]
@@ -2174,7 +2258,7 @@ function hevc_put_hevc_qpel_uni_hv4_8_end_neon · 20 unmodified lines
2174	2258	endfunc
2175	2259
2176	2260	function hevc_put_hevc_qpel_uni_hv6_8_end_neon
2177		− mov x9, #(MAX_PB_SIZE * 2)
	2261	+ mov x9, #(HEVC_MAX_PB_SIZE * 2)
2178	2262	load_qpel_filterh x6, x5
2179	2263	sub x1, x1, #4
2180	2264	ldr q16, [sp]
@@ -2204,7 +2288,7 @@ function hevc_put_hevc_qpel_uni_hv6_8_end_neon · 23 unmodified lines
2204	2288	endfunc
2205	2289
2206	2290	function hevc_put_hevc_qpel_uni_hv8_8_end_neon
2207		− mov x9, #(MAX_PB_SIZE * 2)
	2291	+ mov x9, #(HEVC_MAX_PB_SIZE * 2)
2208	2292	load_qpel_filterh x6, x5
2209	2293	ldr q16, [sp]
2210	2294	ldr q17, [sp, x9]
@@ -2232,7 +2316,7 @@ function hevc_put_hevc_qpel_uni_hv8_8_end_neon · 21 unmodified lines
2232	2316	endfunc
2233	2317
2234	2318	function hevc_put_hevc_qpel_uni_hv12_8_end_neon
2235		− mov x9, #(MAX_PB_SIZE * 2)
	2319	+ mov x9, #(HEVC_MAX_PB_SIZE * 2)
2236	2320	load_qpel_filterh x6, x5
2237	2321	sub x1, x1, #8
2238	2322	ld1 {v16.8h, v17.8h}, [sp], x9
@@ -2260,7 +2344,7 @@ function hevc_put_hevc_qpel_uni_hv12_8_end_neon · 21 unmodified lines
2260	2344	endfunc
2261	2345
2262	2346	function hevc_put_hevc_qpel_uni_hv16_8_end_neon
2263		− mov x9, #(MAX_PB_SIZE * 2)
	2347	+ mov x9, #(HEVC_MAX_PB_SIZE * 2)
2264	2348	load_qpel_filterh x6, x5
2265	2349	sub w12, w9, w7, lsl #1
2266	2350	0: mov x8, sp // src
@@ -3355,7 +3439,7 @@ endfunc · 1088 unmodified lines
3355	3439
3356	3440	function ff_hevc_put_hevc_qpel_h4_8_neon_i8mm, export=1
3357	3441	QPEL_H_HEADER
3358		− mov x10, #MAX_PB_SIZE * 2
	3442	+ mov x10, #HEVC_MAX_PB_SIZE * 2
3359	3443	1:
3360	3444	ld1 {v0.16b}, [x1], x2
3361	3445	ext v1.16b, v0.16b, v0.16b, #1
@@ -3378,7 +3462,7 @@ endfunc · 16 unmodified lines
3378	3462
3379	3463	function ff_hevc_put_hevc_qpel_h6_8_neon_i8mm, export=1
3380	3464	QPEL_H_HEADER
3381		− mov x10, #MAX_PB_SIZE * 2
	3465	+ mov x10, #HEVC_MAX_PB_SIZE * 2
3382	3466	add x15, x0, #8
3383	3467	1:
3384	3468	ld1 {v0.16b}, [x1], x2
@@ -3411,7 +3495,7 @@ endfunc · 26 unmodified lines
3411	3495
3412	3496	function ff_hevc_put_hevc_qpel_h8_8_neon_i8mm, export=1
3413	3497	QPEL_H_HEADER
3414		− mov x10, #MAX_PB_SIZE * 2
	3498	+ mov x10, #HEVC_MAX_PB_SIZE * 2
3415	3499	1:
3416	3500	ld1 {v0.16b}, [x1], x2
3417	3501	ext v1.16b, v0.16b, v0.16b, #1
@@ -3457,7 +3541,7 @@ endfunc · 39 unmodified lines
3457	3541
3458	3542	function ff_hevc_put_hevc_qpel_h12_8_neon_i8mm, export=1
3459	3543	QPEL_H_HEADER
3460		− mov x10, #MAX_PB_SIZE * 2
	3544	+ mov x10, #HEVC_MAX_PB_SIZE * 2
3461	3545	add x15, x0, #16
3462	3546	1:
3463	3547	ld1 {v16.16b, v17.16b}, [x1], x2
@@ -3495,7 +3579,7 @@ endfunc · 31 unmodified lines
3495	3579
3496	3580	function ff_hevc_put_hevc_qpel_h16_8_neon_i8mm, export=1
3497	3581	QPEL_H_HEADER
3498		− mov x10, #MAX_PB_SIZE * 2
	3582	+ mov x10, #HEVC_MAX_PB_SIZE * 2
3499	3583	1:
3500	3584	ld1 {v16.16b, v17.16b}, [x1], x2
3501	3585	ext v1.16b, v16.16b, v17.16b, #1
@@ -3533,7 +3617,7 @@ endfunc · 31 unmodified lines
3533	3617
3534	3618	function ff_hevc_put_hevc_qpel_h24_8_neon_i8mm, export=1
3535	3619	QPEL_H_HEADER
3536		− mov x10, #MAX_PB_SIZE * 2
	3620	+ mov x10, #HEVC_MAX_PB_SIZE * 2
3537	3621	add x15, x0, #32
3538	3622	1:
3539	3623	ld1 {v16.16b, v17.16b}, [x1], x2
@@ -3585,7 +3669,7 @@ endfunc · 45 unmodified lines
3585	3669
3586	3670	function ff_hevc_put_hevc_qpel_h32_8_neon_i8mm, export=1
3587	3671	QPEL_H_HEADER
3588		− mov x10, #MAX_PB_SIZE * 2
	3672	+ mov x10, #HEVC_MAX_PB_SIZE * 2
3589	3673	add x15, x0, #32
3590	3674	1:
3591	3675	ld1 {v16.16b, v17.16b, v18.16b}, [x1], x2
@@ -3642,7 +3726,7 @@ endfunc · 50 unmodified lines
3642	3726
3643	3727	function ff_hevc_put_hevc_qpel_h48_8_neon_i8mm, export=1
3644	3728	QPEL_H_HEADER
3645		− mov x10, #MAX_PB_SIZE * 2 - 64
	3729	+ mov x10, #HEVC_MAX_PB_SIZE * 2 - 64
3646	3730	1:
3647	3731	ld1 {v16.16b, v17.16b, v18.16b, v19.16b}, [x1], x2
3648	3732	ext v1.16b, v16.16b, v17.16b, #1
@@ -4173,7 +4257,7 @@ DISABLE_I8MM · 524 unmodified lines
4173	4257	stp x24, x25, [sp, #48]
4174	4258	stp x26, x27, [sp, #64]
4175	4259	mov x19, sp
4176		− mov x11, #(MAX_PB_SIZE(MAX_PB_SIZE+8)2)
	4260	+ mov x11, #(HEVC_MAX_PB_SIZE(HEVC_MAX_PB_SIZE+8)2)
4177	4261	sub sp, sp, x11
4178	4262	mov x20, x0
4179	4263	mov x21, x1
@@ -4204,7 +4288,7 @@ DISABLE_I8MM · 24 unmodified lines
4204	4288	add x9, x9, x23, lsl #3
4205	4289	ld1 {v0.8b}, [x9]
4206	4290	sxtl v0.8h, v0.8b
4207		− mov x10, #(MAX_PB_SIZE * 2)
	4291	+ mov x10, #(HEVC_MAX_PB_SIZE * 2)
4208	4292	dup v28.4s, w24
4209	4293	dup v29.4s, w25
4210	4294	dup v30.4s, w26
@@ -4591,7 +4675,7 @@ endfunc · 380 unmodified lines
4591	4675	qpel_uni_w_hv neon
4592	4676
4593	4677	function hevc_put_hevc_qpel_bi_hv4_8_end_neon
4594		− mov x9, #(MAX_PB_SIZE * 2)
	4678	+ mov x9, #(HEVC_MAX_PB_SIZE * 2)
4595	4679	load_qpel_filterh x7, x6
4596	4680	ld1 {v16.4h}, [sp], x9
4597	4681	ld1 {v17.4h}, [sp], x9
@@ -4617,7 +4701,7 @@ function hevc_put_hevc_qpel_bi_hv4_8_end_neon · 19 unmodified lines
4617	4701	endfunc
4618	4702
4619	4703	function hevc_put_hevc_qpel_bi_hv6_8_end_neon
4620		− mov x9, #(MAX_PB_SIZE * 2)
	4704	+ mov x9, #(HEVC_MAX_PB_SIZE * 2)
4621	4705	load_qpel_filterh x7, x6
4622	4706	sub x1, x1, #4
4623	4707	ld1 {v16.8h}, [sp], x9
@@ -4648,7 +4732,7 @@ function hevc_put_hevc_qpel_bi_hv6_8_end_neon · 24 unmodified lines
4648	4732	endfunc
4649	4733
4650	4734	function hevc_put_hevc_qpel_bi_hv8_8_end_neon
4651		− mov x9, #(MAX_PB_SIZE * 2)
	4735	+ mov x9, #(HEVC_MAX_PB_SIZE * 2)
4652	4736	load_qpel_filterh x7, x6
4653	4737	ld1 {v16.8h}, [sp], x9
4654	4738	ld1 {v17.8h}, [sp], x9
@@ -4678,7 +4762,7 @@ endfunc · 23 unmodified lines
4678	4762
4679	4763	function hevc_put_hevc_qpel_bi_hv16_8_end_neon
4680	4764	load_qpel_filterh x7, x8
4681		− mov x9, #(MAX_PB_SIZE * 2)
	4765	+ mov x9, #(HEVC_MAX_PB_SIZE * 2)
4682	4766	mov x10, x6
4683	4767	0: mov x8, sp // src
4684	4768	ld1 {v16.8h, v17.8h}, [x8], x9

libavcodec/aarch64/vvc/Makefile+1−0

@@ -3,4 +3,5 @@ clean:: · 2 unmodified lines
3	3
4	4	OBJS-$(CONFIG_VVC_DECODER) += aarch64/vvc/dsp_init.o
5	5	NEON-OBJS-$(CONFIG_VVC_DECODER) += aarch64/vvc/alf.o \
	6	+ aarch64/h26x/qpel_neon.o \
6	7	aarch64/h26x/sao_neon.o

libavcodec/aarch64/vvc/dsp_init.c+14−0

@@ -46,6 +46,20 @@ void ff_vvc_dsp_init_aarch64(VVCDSPContext *const c, const int bd) · 45 unmodified lines
46	46	return;
47	47
48	48	if (bd == 8) {
	49	+ c->inter.put[0][1][0][1] = ff_vvc_put_qpel_h4_8_neon;
	50	+ c->inter.put[0][2][0][1] = ff_vvc_put_qpel_h8_8_neon;
	51	+ c->inter.put[0][3][0][1] = ff_vvc_put_qpel_h16_8_neon;
	52	+ c->inter.put[0][4][0][1] =
	53	+ c->inter.put[0][5][0][1] =
	54	+ c->inter.put[0][6][0][1] = ff_vvc_put_qpel_h32_8_neon;
	55	+
	56	+ c->inter.put_uni[0][1][0][1] = ff_vvc_put_qpel_uni_h4_8_neon;
	57	+ c->inter.put_uni[0][2][0][1] = ff_vvc_put_qpel_uni_h8_8_neon;
	58	+ c->inter.put_uni[0][3][0][1] = ff_vvc_put_qpel_uni_h16_8_neon;
	59	+ c->inter.put_uni[0][4][0][1] =
	60	+ c->inter.put_uni[0][5][0][1] =
	61	+ c->inter.put_uni[0][6][0][1] = ff_vvc_put_qpel_uni_h32_8_neon;
	62	+
49	63	for (int i = 0; i < FF_ARRAY_ELEMS(c->sao.band_filter); i++)
50	64	c->sao.band_filter[i] = ff_h26x_sao_band_filter_8x8_8_neon;
51	65	c->sao.edge_filter[0] = ff_vvc_sao_edge_filter_8x8_8_neon;

aarch64/vvc: Add put_qpel_h_* and put_qpel_uni_h_*

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code