aarch64/vvc: Add put_epel_hv

On Apple M1: put_chroma_hv_8_4x4_c: 1.7 ( 1.00x) put_chroma_hv_8_4x4_neon: 0.2 ( 7.67x) put_chroma_hv_8_8x8_c: 5.5 ( 1.00x) put_chroma_hv_8_8x8_neon: 0.5 (11.53x) put_chroma_hv_8_16x16_c: 18.5 ( 1.00x) put_chroma_hv_8_16x16_neon: 1.5 (12.53x) put_chroma_hv_8_32x32_c: 72.5 ( 1.00x) put_chroma_hv_8_32x32_neon: 4.7 (15.34x) put_chroma_hv_8_64x64_c: 274.0 ( 1.00x) put_chroma_hv_8_64x64_neon: 18.5 (14.83x) put_chroma_hv_8_128x128_c: 1058.7 ( 1.00x) put_chroma_hv_8_128x128_neon: 75.2 (14.07x) On Android Pixel 8 Pro: put_chroma_hv_8_4x4_c: 1.2 ( 1.00x) put_chroma_hv_8_4x4_neon: 0.0 ( 0.00x) put_chroma_hv_8_4x4_i8mm: 0.2 ( 5.00x) put_chroma_hv_8_8x8_c: 4.0 ( 1.00x) put_chroma_hv_8_8x8_neon: 0.5 ( 8.00x) put_chroma_hv_8_8x8_i8mm: 0.5 ( 8.00x) put_chroma_hv_8_16x16_c: 15.2 ( 1.00x) put_chroma_hv_8_16x16_neon: 2.5 ( 6.10x) put_chroma_hv_8_16x16_i8mm: 2.2 ( 6.78x) put_chroma_hv_8_32x32_c: 61.0 ( 1.00x) put_chroma_hv_8_32x32_neon: 9.8 ( 6.26x) put_chroma_hv_8_32x32_i8mm: 8.5 ( 7.18x) put_chroma_hv_8_64x64_c: 229.5 ( 1.00x) put_chroma_hv_8_64x64_neon: 38.5 ( 5.96x) put_chroma_hv_8_64x64_i8mm: 34.0 ( 6.75x) put_chroma_hv_8_128x128_c: 919.8 ( 1.00x) put_chroma_hv_8_128x128_neon: 154.5 ( 5.95x) put_chroma_hv_8_128x128_i8mm: 140.0 ( 6.57x)

Zhao Zhili committed 2 years ago

Commit 1be5a231 parent 0dcf2043 files · +147 −0Browse files

Unified · Split

3 files

libavcodec/aarch64/h26x/dsp.h +8 −0
libavcodec/aarch64/h26x/epel_neon.S +125 −0
libavcodec/aarch64/vvc/dsp_init.c +14 −0

libavcodec/aarch64/h26x/dsp.h+8−0

@@ -297,4 +297,12 @@ NEON8_FNPROTO_PARTIAL_6(qpel_hv, (int16_t *dst, · 296 unmodified lines
297	297	const uint8_t *src, ptrdiff_t srcstride, int height,
298	298	const int8_t hf, const int8_t vf, int width), _i8mm);
299	299
	300	+NEON8_FNPROTO_PARTIAL_6(epel_hv, (int16_t *dst,
	301	+ const uint8_t *src, ptrdiff_t srcstride, int height,
	302	+ const int8_t hf, const int8_t vf, int width),);
	303	+
	304	+NEON8_FNPROTO_PARTIAL_6(epel_hv, (int16_t *dst,
	305	+ const uint8_t *src, ptrdiff_t srcstride, int height,
	306	+ const int8_t hf, const int8_t vf, int width), _i8mm);
	307	+
300	308	#endif

libavcodec/aarch64/h26x/epel_neon.S+125−0

@@ -72,6 +72,11 @@ endconst · 71 unmodified lines
72	72	sxtl v0.8h, v0.8b
73	73	.endm
74	74
	75	+.macro vvc_load_epel_filterh freg
	76	+ ld1 {v0.8b}, [\freg]
	77	+ sxtl v0.8h, v0.8b
	78	+.endm
	79	+
75	80	.macro calc_epelh dst, src0, src1, src2, src3
76	81	smull \dst\().4s, \src0\().4h, v0.h[0]
77	82	smlal \dst\().4s, \src1\().4h, v0.h[1]
@@ -2299,10 +2304,16 @@ endfunc · 2221 unmodified lines
2299	2304	DISABLE_I8MM
2300	2305	#endif
2301	2306
	2307	+function vvc_put_epel_hv4_8_end_neon
	2308	+ vvc_load_epel_filterh x5
	2309	+ mov x10, #(VVC_MAX_PB_SIZE * 2)
	2310	+ b 0f
	2311	+endfunc
2302	2312
2303	2313	function hevc_put_hevc_epel_hv4_8_end_neon
2304	2314	load_epel_filterh x5, x4
2305	2315	mov x10, #(HEVC_MAX_PB_SIZE * 2)
	2316	+0:
2306	2317	ldr d16, [sp]
2307	2318	ldr d17, [sp, x10]
2308	2319	add sp, sp, x10, lsl #1
@@ -2339,9 +2350,16 @@ function hevc_put_hevc_epel_hv6_8_end_neon · 30 unmodified lines
2339	2350	2: ret
2340	2351	endfunc
2341	2352
	2353	+function vvc_put_epel_hv8_8_end_neon
	2354	+ vvc_load_epel_filterh x5
	2355	+ mov x10, #(VVC_MAX_PB_SIZE * 2)
	2356	+ b 0f
	2357	+endfunc
	2358	+
2342	2359	function hevc_put_hevc_epel_hv8_8_end_neon
2343	2360	load_epel_filterh x5, x4
2344	2361	mov x10, #(HEVC_MAX_PB_SIZE * 2)
	2362	+0:
2345	2363	ldr q16, [sp]
2346	2364	ldr q17, [sp, x10]
2347	2365	add sp, sp, x10, lsl #1
@@ -2379,9 +2397,16 @@ function hevc_put_hevc_epel_hv12_8_end_neon · 31 unmodified lines
2379	2397	2: ret
2380	2398	endfunc
2381	2399
	2400	+function vvc_put_epel_hv16_8_end_neon
	2401	+ vvc_load_epel_filterh x5
	2402	+ mov x10, #(VVC_MAX_PB_SIZE * 2)
	2403	+ b 0f
	2404	+endfunc
	2405	+
2382	2406	function hevc_put_hevc_epel_hv16_8_end_neon
2383	2407	load_epel_filterh x5, x4
2384	2408	mov x10, #(HEVC_MAX_PB_SIZE * 2)
	2409	+0:
2385	2410	ld1 {v16.8h, v17.8h}, [sp], x10
2386	2411	ld1 {v18.8h, v19.8h}, [sp], x10
2387	2412	ld1 {v20.8h, v21.8h}, [sp], x10
@@ -2437,6 +2462,21 @@ function ff_hevc_put_hevc_epel_hv4_8_\suffix, export=1 · 49 unmodified lines
2437	2462	b hevc_put_hevc_epel_hv4_8_end_neon
2438	2463	endfunc
2439	2464
	2465	+function ff_vvc_put_epel_hv4_8_\suffix, export=1
	2466	+ add w10, w3, #3
	2467	+ lsl x10, x10, #8
	2468	+ sub sp, sp, x10 // tmp_array
	2469	+ stp x5, x30, [sp, #-32]!
	2470	+ stp x0, x3, [sp, #16]
	2471	+ add x0, sp, #32
	2472	+ sub x1, x1, x2
	2473	+ add w3, w3, #3
	2474	+ bl X(ff_vvc_put_epel_h4_8_\suffix)
	2475	+ ldp x0, x3, [sp, #16]
	2476	+ ldp x5, x30, [sp], #32
	2477	+ b vvc_put_epel_hv4_8_end_neon
	2478	+endfunc
	2479	+
2440	2480	function ff_hevc_put_hevc_epel_hv6_8_\suffix, export=1
2441	2481	add w10, w3, #3
2442	2482	lsl x10, x10, #7
@@ -2467,6 +2507,21 @@ function ff_hevc_put_hevc_epel_hv8_8_\suffix, export=1 · 24 unmodified lines
2467	2507	b hevc_put_hevc_epel_hv8_8_end_neon
2468	2508	endfunc
2469	2509
	2510	+function ff_vvc_put_epel_hv8_8_\suffix, export=1
	2511	+ add w10, w3, #3
	2512	+ lsl x10, x10, #8
	2513	+ sub sp, sp, x10 // tmp_array
	2514	+ stp x5, x30, [sp, #-32]!
	2515	+ stp x0, x3, [sp, #16]
	2516	+ add x0, sp, #32
	2517	+ sub x1, x1, x2
	2518	+ add w3, w3, #3
	2519	+ bl X(ff_vvc_put_epel_h8_8_\suffix)
	2520	+ ldp x0, x3, [sp, #16]
	2521	+ ldp x5, x30, [sp], #32
	2522	+ b vvc_put_epel_hv8_8_end_neon
	2523	+endfunc
	2524	+
2470	2525	function ff_hevc_put_hevc_epel_hv12_8_\suffix, export=1
2471	2526	add w10, w3, #3
2472	2527	lsl x10, x10, #7
@@ -2497,6 +2552,21 @@ function ff_hevc_put_hevc_epel_hv16_8_\suffix, export=1 · 24 unmodified lines
2497	2552	b hevc_put_hevc_epel_hv16_8_end_neon
2498	2553	endfunc
2499	2554
	2555	+function ff_vvc_put_epel_hv16_8_\suffix, export=1
	2556	+ add w10, w3, #3
	2557	+ lsl x10, x10, #8
	2558	+ sub sp, sp, x10 // tmp_array
	2559	+ stp x5, x30, [sp, #-32]!
	2560	+ stp x0, x3, [sp, #16]
	2561	+ add x0, sp, #32
	2562	+ sub x1, x1, x2
	2563	+ add w3, w3, #3
	2564	+ bl X(ff_vvc_put_epel_h16_8_\suffix)
	2565	+ ldp x0, x3, [sp, #16]
	2566	+ ldp x5, x30, [sp], #32
	2567	+ b vvc_put_epel_hv16_8_end_neon
	2568	+endfunc
	2569	+
2500	2570	function ff_hevc_put_hevc_epel_hv24_8_\suffix, export=1
2501	2571	add w10, w3, #3
2502	2572	lsl x10, x10, #7
@@ -2530,6 +2600,24 @@ function ff_hevc_put_hevc_epel_hv32_8_\suffix, export=1 · 27 unmodified lines
2530	2600	ret
2531	2601	endfunc
2532	2602
	2603	+function ff_vvc_put_epel_hv32_8_\suffix, export=1
	2604	+ stp x4, x5, [sp, #-64]!
	2605	+ stp x2, x3, [sp, #16]
	2606	+ stp x0, x1, [sp, #32]
	2607	+ str x30, [sp, #48]
	2608	+ mov x6, #16
	2609	+ bl X(ff_vvc_put_epel_hv16_8_\suffix)
	2610	+ ldp x0, x1, [sp, #32]
	2611	+ ldp x2, x3, [sp, #16]
	2612	+ ldp x4, x5, [sp], #48
	2613	+ add x0, x0, #32
	2614	+ add x1, x1, #16
	2615	+ mov x6, #16
	2616	+ bl X(ff_vvc_put_epel_hv16_8_\suffix)
	2617	+ ldr x30, [sp], #16
	2618	+ ret
	2619	+endfunc
	2620	+
2533	2621	function ff_hevc_put_hevc_epel_hv48_8_\suffix, export=1
2534	2622	stp x4, x5, [sp, #-64]!
2535	2623	stp x2, x3, [sp, #16]
@@ -2579,6 +2667,43 @@ function ff_hevc_put_hevc_epel_hv64_8_\suffix, export=1 · 43 unmodified lines
2579	2667	ldr x30, [sp], #16
2580	2668	ret
2581	2669	endfunc
	2670	+
	2671	+function ff_vvc_put_epel_hv64_8_\suffix, export=1
	2672	+ stp x4, x5, [sp, #-64]!
	2673	+ stp x2, x3, [sp, #16]
	2674	+ stp x0, x1, [sp, #32]
	2675	+ str x30, [sp, #48]
	2676	+ mov x6, #32
	2677	+ bl X(ff_vvc_put_epel_hv32_8_\suffix)
	2678	+ ldp x0, x1, [sp, #32]
	2679	+ ldp x2, x3, [sp, #16]
	2680	+ ldp x4, x5, [sp], #48
	2681	+ add x0, x0, #64
	2682	+ add x1, x1, #32
	2683	+ mov x6, #32
	2684	+ bl X(ff_vvc_put_epel_hv32_8_\suffix)
	2685	+ ldr x30, [sp], #16
	2686	+ ret
	2687	+endfunc
	2688	+
	2689	+function ff_vvc_put_epel_hv128_8_\suffix, export=1
	2690	+ stp x4, x5, [sp, #-64]!
	2691	+ stp x2, x3, [sp, #16]
	2692	+ stp x0, x1, [sp, #32]
	2693	+ str x30, [sp, #48]
	2694	+ mov x6, #64
	2695	+ bl X(ff_vvc_put_epel_hv64_8_\suffix)
	2696	+ ldp x0, x1, [sp, #32]
	2697	+ ldp x2, x3, [sp, #16]
	2698	+ ldp x4, x5, [sp], #48
	2699	+ add x0, x0, #128
	2700	+ add x1, x1, #64
	2701	+ mov x6, #64
	2702	+ bl X(ff_vvc_put_epel_hv64_8_\suffix)
	2703	+ ldr x30, [sp], #16
	2704	+ ret
	2705	+endfunc
	2706	+
2582	2707	.endm
2583	2708
2584	2709	epel_hv neon

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

@@ -84,6 +84,13 @@ void ff_vvc_dsp_init_aarch64(VVCDSPContext *const c, const int bd) · 83 unmodified lines
84	84	c->inter.put[1][5][0][1] =
85	85	c->inter.put[1][6][0][1] = ff_vvc_put_epel_h32_8_neon;
86	86
	87	+ c->inter.put[1][1][1][1] = ff_vvc_put_epel_hv4_8_neon;
	88	+ c->inter.put[1][2][1][1] = ff_vvc_put_epel_hv8_8_neon;
	89	+ c->inter.put[1][3][1][1] = ff_vvc_put_epel_hv16_8_neon;
	90	+ c->inter.put[1][4][1][1] = ff_vvc_put_epel_hv32_8_neon;
	91	+ c->inter.put[1][5][1][1] = ff_vvc_put_epel_hv64_8_neon;
	92	+ c->inter.put[1][6][1][1] = ff_vvc_put_epel_hv128_8_neon;
	93	+
87	94	c->inter.put_uni[0][1][0][0] = ff_vvc_put_pel_uni_pixels4_8_neon;
88	95	c->inter.put_uni[0][2][0][0] = ff_vvc_put_pel_uni_pixels8_8_neon;
89	96	c->inter.put_uni[0][3][0][0] = ff_vvc_put_pel_uni_pixels16_8_neon;
@@ -134,6 +141,13 @@ void ff_vvc_dsp_init_aarch64(VVCDSPContext *const c, const int bd) · 44 unmodified lines
134	141	c->inter.put[1][4][0][1] = ff_vvc_put_epel_h32_8_neon_i8mm;
135	142	c->inter.put[1][5][0][1] = ff_vvc_put_epel_h64_8_neon_i8mm;
136	143	c->inter.put[1][6][0][1] = ff_vvc_put_epel_h128_8_neon_i8mm;
	144	+
	145	+ c->inter.put[1][1][1][1] = ff_vvc_put_epel_hv4_8_neon_i8mm;
	146	+ c->inter.put[1][2][1][1] = ff_vvc_put_epel_hv8_8_neon_i8mm;
	147	+ c->inter.put[1][3][1][1] = ff_vvc_put_epel_hv16_8_neon_i8mm;
	148	+ c->inter.put[1][4][1][1] = ff_vvc_put_epel_hv32_8_neon_i8mm;
	149	+ c->inter.put[1][5][1][1] = ff_vvc_put_epel_hv64_8_neon_i8mm;
	150	+ c->inter.put[1][6][1][1] = ff_vvc_put_epel_hv128_8_neon_i8mm;
137	151	}
138	152	} else if (bd == 10) {
139	153	c->alf.filter[LUMA] = alf_filter_luma_10_neon;

aarch64/vvc: Add put_epel_hv

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code