aarch64/vvc: Add put_qpel_hv

With Apple M1 (no i8mm): put_luma_hv_8_4x4_c: 2.2 ( 1.00x) put_luma_hv_8_4x4_neon: 0.8 ( 3.00x) put_luma_hv_8_8x8_c: 7.0 ( 1.00x) put_luma_hv_8_8x8_neon: 0.8 ( 9.33x) put_luma_hv_8_16x16_c: 22.8 ( 1.00x) put_luma_hv_8_16x16_neon: 2.5 ( 9.10x) put_luma_hv_8_32x32_c: 84.8 ( 1.00x) put_luma_hv_8_32x32_neon: 9.5 ( 8.92x) put_luma_hv_8_64x64_c: 333.0 ( 1.00x) put_luma_hv_8_64x64_neon: 35.5 ( 9.38x) put_luma_hv_8_128x128_c: 1294.5 ( 1.00x) put_luma_hv_8_128x128_neon: 137.8 ( 9.40x) With Pixel 8 Pro: put_luma_hv_8_4x4_c: 5.0 ( 1.00x) put_luma_hv_8_4x4_neon: 0.8 ( 6.67x) put_luma_hv_8_4x4_i8mm: 0.2 (20.00x) put_luma_hv_8_8x8_c: 13.2 ( 1.00x) put_luma_hv_8_8x8_neon: 1.2 (10.60x) put_luma_hv_8_8x8_i8mm: 1.2 (10.60x) put_luma_hv_8_16x16_c: 44.2 ( 1.00x) put_luma_hv_8_16x16_neon: 4.5 ( 9.83x) put_luma_hv_8_16x16_i8mm: 4.2 (10.41x) put_luma_hv_8_32x32_c: 160.8 ( 1.00x) put_luma_hv_8_32x32_neon: 17.5 ( 9.19x) put_luma_hv_8_32x32_i8mm: 16.0 (10.05x) put_luma_hv_8_64x64_c: 611.2 ( 1.00x) put_luma_hv_8_64x64_neon: 68.0 ( 8.99x) put_luma_hv_8_64x64_i8mm: 62.2 ( 9.82x) put_luma_hv_8_128x128_c: 2384.8 ( 1.00x) put_luma_hv_8_128x128_neon: 268.8 ( 8.87x) put_luma_hv_8_128x128_i8mm: 245.8 ( 9.70x)

Zhao Zhili committed 2 years ago

Commit 5ac69251 parent a0b52af3 files · +162 −0Browse files

Unified · Split

3 files

libavcodec/aarch64/h26x/dsp.h +8 −0
libavcodec/aarch64/h26x/qpel_neon.S +140 −0
libavcodec/aarch64/vvc/dsp_init.c +14 −0

libavcodec/aarch64/h26x/dsp.h+8−0

@@ -282,4 +282,12 @@ void ff_vvc_put_qpel_v8_8_neon(int16_t dst, const uint8_t _src, · 281 unmodified lines
282	ptrdiff_t _srcstride, int height,	282	ptrdiff_t _srcstride, int height,
283	const int8_t hf, const int8_t vf, int width);	283	const int8_t hf, const int8_t vf, int width);
284		284
		285	NEON8_FNPROTO_PARTIAL_6(qpel_hv, (int16_t *dst,
		286	const uint8_t *src, ptrdiff_t srcstride, int height,
		287	const int8_t hf, const int8_t vf, int width),);
		288
		289	NEON8_FNPROTO_PARTIAL_6(qpel_hv, (int16_t *dst,
		290	const uint8_t *src, ptrdiff_t srcstride, int height,
		291	const int8_t hf, const int8_t vf, int width), _i8mm);
		292
285	#endif	293	#endif

libavcodec/aarch64/h26x/qpel_neon.S+140−0

@@ -4140,9 +4140,15 @@ endfunc · 4139 unmodified lines
4140	DISABLE_I8MM	4140	DISABLE_I8MM
4141	#endif	4141	#endif
4142		4142
		4143	function vvc_put_qpel_hv4_8_end_neon
		4144	vvc_load_qpel_filterh x5
		4145	mov x7, #(VVC_MAX_PB_SIZE * 2)
		4146	b 1f
		4147	endfunc
4143		4148
4144	function hevc_put_hevc_qpel_hv4_8_end_neon	4149	function hevc_put_hevc_qpel_hv4_8_end_neon
4145	load_qpel_filterh x5, x4	4150	load_qpel_filterh x5, x4
		4151	1:
4146	ldr d16, [sp]	4152	ldr d16, [sp]
4147	ldr d17, [sp, x7]	4153	ldr d17, [sp, x7]
4148	add sp, sp, x7, lsl #1	4154	add sp, sp, x7, lsl #1
@@ -4194,9 +4200,16 @@ function hevc_put_hevc_qpel_hv6_8_end_neon · 45 unmodified lines
4194	ret	4200	ret
4195	endfunc	4201	endfunc
4196		4202
		4203	function vvc_put_qpel_hv8_8_end_neon
		4204	vvc_load_qpel_filterh x5
		4205	mov x7, #(VVC_MAX_PB_SIZE * 2)
		4206	b 1f
		4207	endfunc
		4208
4197	function hevc_put_hevc_qpel_hv8_8_end_neon	4209	function hevc_put_hevc_qpel_hv8_8_end_neon
4198	mov x7, #128	4210	mov x7, #128
4199	load_qpel_filterh x5, x4	4211	load_qpel_filterh x5, x4
		4212	1:
4200	ldr q16, [sp]	4213	ldr q16, [sp]
4201	ldr q17, [sp, x7]	4214	ldr q17, [sp, x7]
4202	add sp, sp, x7, lsl #1	4215	add sp, sp, x7, lsl #1
@@ -4247,9 +4260,16 @@ function hevc_put_hevc_qpel_hv12_8_end_neon · 44 unmodified lines
4247	ret	4260	ret
4248	endfunc	4261	endfunc
4249		4262
		4263	function vvc_put_qpel_hv16_8_end_neon
		4264	vvc_load_qpel_filterh x5
		4265	mov x7, #(VVC_MAX_PB_SIZE * 2)
		4266	b 1f
		4267	endfunc
		4268
4250	function hevc_put_hevc_qpel_hv16_8_end_neon	4269	function hevc_put_hevc_qpel_hv16_8_end_neon
4251	mov x7, #128	4270	mov x7, #128
4252	load_qpel_filterh x5, x4	4271	load_qpel_filterh x5, x4
		4272	1:
4253	ld1 {v16.8h, v17.8h}, [sp], x7	4273	ld1 {v16.8h, v17.8h}, [sp], x7
4254	ld1 {v18.8h, v19.8h}, [sp], x7	4274	ld1 {v18.8h, v19.8h}, [sp], x7
4255	ld1 {v20.8h, v21.8h}, [sp], x7	4275	ld1 {v20.8h, v21.8h}, [sp], x7
@@ -4272,6 +4292,12 @@ function hevc_put_hevc_qpel_hv16_8_end_neon · 16 unmodified lines
4272	ret	4292	ret
4273	endfunc	4293	endfunc
4274		4294
		4295	function vvc_put_qpel_hv32_8_end_neon
		4296	vvc_load_qpel_filterh x5
		4297	mov x7, #(VVC_MAX_PB_SIZE * 2)
		4298	b 0f
		4299	endfunc
		4300
4275	function hevc_put_hevc_qpel_hv32_8_end_neon	4301	function hevc_put_hevc_qpel_hv32_8_end_neon
4276	mov x7, #128	4302	mov x7, #128
4277	load_qpel_filterh x5, x4	4303	load_qpel_filterh x5, x4
@@ -4325,6 +4351,25 @@ function ff_hevc_put_hevc_qpel_hv4_8_\suffix, export=1 · 47 unmodified lines
4325	b hevc_put_hevc_qpel_hv4_8_end_neon	4351	b hevc_put_hevc_qpel_hv4_8_end_neon
4326	endfunc	4352	endfunc
4327		4353
		4354	function ff_vvc_put_qpel_hv4_8_\suffix, export=1
		4355	add w10, w3, #8
		4356	lsl x10, x10, #8
		4357	mov x14, sp
		4358	sub sp, sp, x10 // tmp_array
		4359	stp x5, x30, [sp, #-48]!
		4360	stp x0, x3, [sp, #16]
		4361	str x14, [sp, #32]
		4362	add x0, sp, #48
		4363	sub x1, x1, x2, lsl #1
		4364	add x3, x3, #7
		4365	sub x1, x1, x2
		4366	bl X(ff_vvc_put_qpel_h4_8_\suffix)
		4367	ldr x14, [sp, #32]
		4368	ldp x0, x3, [sp, #16]
		4369	ldp x5, x30, [sp], #48
		4370	b vvc_put_qpel_hv4_8_end_neon
		4371	endfunc
		4372
4328	function ff_hevc_put_hevc_qpel_hv6_8_\suffix, export=1	4373	function ff_hevc_put_hevc_qpel_hv6_8_\suffix, export=1
4329	add w10, w3, #8	4374	add w10, w3, #8
4330	mov x7, #128	4375	mov x7, #128
@@ -4364,6 +4409,25 @@ function ff_hevc_put_hevc_qpel_hv8_8_\suffix, export=1 · 33 unmodified lines
4364	b hevc_put_hevc_qpel_hv8_8_end_neon	4409	b hevc_put_hevc_qpel_hv8_8_end_neon
4365	endfunc	4410	endfunc
4366		4411
		4412	function ff_vvc_put_qpel_hv8_8_\suffix, export=1
		4413	add w10, w3, #8
		4414	lsl x10, x10, #8
		4415	sub x1, x1, x2, lsl #1
		4416	mov x14, sp
		4417	sub sp, sp, x10 // tmp_array
		4418	stp x5, x30, [sp, #-48]!
		4419	stp x0, x3, [sp, #16]
		4420	str x14, [sp, #32]
		4421	add x0, sp, #48
		4422	add x3, x3, #7
		4423	sub x1, x1, x2
		4424	bl X(ff_vvc_put_qpel_h8_8_\suffix)
		4425	ldr x14, [sp, #32]
		4426	ldp x0, x3, [sp, #16]
		4427	ldp x5, x30, [sp], #48
		4428	b vvc_put_qpel_hv8_8_end_neon
		4429	endfunc
		4430
4367	function ff_hevc_put_hevc_qpel_hv12_8_\suffix, export=1	4431	function ff_hevc_put_hevc_qpel_hv12_8_\suffix, export=1
4368	add w10, w3, #8	4432	add w10, w3, #8
4369	lsl x10, x10, #7	4433	lsl x10, x10, #7
@@ -4403,6 +4467,25 @@ function ff_hevc_put_hevc_qpel_hv16_8_\suffix, export=1 · 33 unmodified lines
4403	b hevc_put_hevc_qpel_hv16_8_end_neon	4467	b hevc_put_hevc_qpel_hv16_8_end_neon
4404	endfunc	4468	endfunc
4405		4469
		4470	function ff_vvc_put_qpel_hv16_8_\suffix, export=1
		4471	add w10, w3, #8
		4472	lsl x10, x10, #8
		4473	sub x1, x1, x2, lsl #1
		4474	mov x14, sp
		4475	sub sp, sp, x10 // tmp_array
		4476	stp x5, x30, [sp, #-48]!
		4477	stp x0, x3, [sp, #16]
		4478	str x14, [sp, #32]
		4479	add x3, x3, #7
		4480	add x0, sp, #48
		4481	sub x1, x1, x2
		4482	bl X(ff_vvc_put_qpel_h16_8_\suffix)
		4483	ldr x14, [sp, #32]
		4484	ldp x0, x3, [sp, #16]
		4485	ldp x5, x30, [sp], #48
		4486	b vvc_put_qpel_hv16_8_end_neon
		4487	endfunc
		4488
4406	function ff_hevc_put_hevc_qpel_hv24_8_\suffix, export=1	4489	function ff_hevc_put_hevc_qpel_hv24_8_\suffix, export=1
4407	stp x4, x5, [sp, #-64]!	4490	stp x4, x5, [sp, #-64]!
4408	stp x2, x3, [sp, #16]	4491	stp x2, x3, [sp, #16]
@@ -4439,6 +4522,26 @@ function ff_hevc_put_hevc_qpel_hv32_8_\suffix, export=1 · 30 unmodified lines
4439	b hevc_put_hevc_qpel_hv32_8_end_neon	4522	b hevc_put_hevc_qpel_hv32_8_end_neon
4440	endfunc	4523	endfunc
4441		4524
		4525	function ff_vvc_put_qpel_hv32_8_\suffix, export=1
		4526	add w10, w3, #8
		4527	sub x1, x1, x2, lsl #1
		4528	lsl x10, x10, #8
		4529	sub x1, x1, x2
		4530	mov x14, sp
		4531	sub sp, sp, x10 // tmp_array
		4532	stp x5, x30, [sp, #-48]!
		4533	stp x0, x3, [sp, #16]
		4534	str x14, [sp, #32]
		4535	add x3, x3, #7
		4536	add x0, sp, #48
		4537	mov w6, #32
		4538	bl X(ff_vvc_put_qpel_h32_8_\suffix)
		4539	ldr x14, [sp, #32]
		4540	ldp x0, x3, [sp, #16]
		4541	ldp x5, x30, [sp], #48
		4542	b vvc_put_qpel_hv32_8_end_neon
		4543	endfunc
		4544
4442	function ff_hevc_put_hevc_qpel_hv48_8_\suffix, export=1	4545	function ff_hevc_put_hevc_qpel_hv48_8_\suffix, export=1
4443	stp x4, x5, [sp, #-64]!	4546	stp x4, x5, [sp, #-64]!
4444	stp x2, x3, [sp, #16]	4547	stp x2, x3, [sp, #16]
@@ -4472,6 +4575,43 @@ function ff_hevc_put_hevc_qpel_hv64_8_\suffix, export=1 · 27 unmodified lines
4472	ldr x30, [sp], #16	4575	ldr x30, [sp], #16
4473	ret	4576	ret
4474	endfunc	4577	endfunc
		4578
		4579	function ff_vvc_put_qpel_hv64_8_\suffix, export=1
		4580	stp x4, x5, [sp, #-64]!
		4581	stp x2, x3, [sp, #16]
		4582	stp x0, x1, [sp, #32]
		4583	str x30, [sp, #48]
		4584	mov x6, #32
		4585	bl X(ff_vvc_put_qpel_hv32_8_\suffix)
		4586	ldp x0, x1, [sp, #32]
		4587	ldp x2, x3, [sp, #16]
		4588	ldp x4, x5, [sp], #48
		4589	add x1, x1, #32
		4590	add x0, x0, #64
		4591	mov x6, #32
		4592	bl X(ff_vvc_put_qpel_hv32_8_\suffix)
		4593	ldr x30, [sp], #16
		4594	ret
		4595	endfunc
		4596
		4597	function ff_vvc_put_qpel_hv128_8_\suffix, export=1
		4598	stp x4, x5, [sp, #-64]!
		4599	stp x2, x3, [sp, #16]
		4600	stp x0, x1, [sp, #32]
		4601	str x30, [sp, #48]
		4602	mov x6, #64
		4603	bl X(ff_vvc_put_qpel_hv64_8_\suffix)
		4604	ldp x0, x1, [sp, #32]
		4605	ldp x2, x3, [sp, #16]
		4606	ldp x4, x5, [sp], #48
		4607	add x1, x1, #64
		4608	add x0, x0, #128
		4609	mov x6, #64
		4610	bl X(ff_vvc_put_qpel_hv64_8_\suffix)
		4611	ldr x30, [sp], #16
		4612	ret
		4613	endfunc
		4614
4475	.endm	4615	.endm
4476		4616
4477	qpel_hv neon	4617	qpel_hv neon

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

@@ -67,6 +67,13 @@ void ff_vvc_dsp_init_aarch64(VVCDSPContext *const c, const int bd) · 66 unmodified lines
67	c->inter.put[0][5][1][0] =	67	c->inter.put[0][5][1][0] =
68	c->inter.put[0][6][1][0] = ff_vvc_put_qpel_v8_8_neon;	68	c->inter.put[0][6][1][0] = ff_vvc_put_qpel_v8_8_neon;
69		69
		70	c->inter.put[0][1][1][1] = ff_vvc_put_qpel_hv4_8_neon;
		71	c->inter.put[0][2][1][1] = ff_vvc_put_qpel_hv8_8_neon;
		72	c->inter.put[0][3][1][1] = ff_vvc_put_qpel_hv16_8_neon;
		73	c->inter.put[0][4][1][1] = ff_vvc_put_qpel_hv32_8_neon;
		74	c->inter.put[0][5][1][1] = ff_vvc_put_qpel_hv64_8_neon;
		75	c->inter.put[0][6][1][1] = ff_vvc_put_qpel_hv128_8_neon;
		76
70	c->inter.put_uni[0][1][0][0] = ff_vvc_put_pel_uni_pixels4_8_neon;	77	c->inter.put_uni[0][1][0][0] = ff_vvc_put_pel_uni_pixels4_8_neon;
71	c->inter.put_uni[0][2][0][0] = ff_vvc_put_pel_uni_pixels8_8_neon;	78	c->inter.put_uni[0][2][0][0] = ff_vvc_put_pel_uni_pixels8_8_neon;
72	c->inter.put_uni[0][3][0][0] = ff_vvc_put_pel_uni_pixels16_8_neon;	79	c->inter.put_uni[0][3][0][0] = ff_vvc_put_pel_uni_pixels16_8_neon;
@@ -103,6 +110,13 @@ void ff_vvc_dsp_init_aarch64(VVCDSPContext *const c, const int bd) · 30 unmodified lines
103	c->inter.put[0][4][0][1] = ff_vvc_put_qpel_h32_8_neon_i8mm;	110	c->inter.put[0][4][0][1] = ff_vvc_put_qpel_h32_8_neon_i8mm;
104	c->inter.put[0][5][0][1] = ff_vvc_put_qpel_h64_8_neon_i8mm;	111	c->inter.put[0][5][0][1] = ff_vvc_put_qpel_h64_8_neon_i8mm;
105	c->inter.put[0][6][0][1] = ff_vvc_put_qpel_h128_8_neon_i8mm;	112	c->inter.put[0][6][0][1] = ff_vvc_put_qpel_h128_8_neon_i8mm;
		113
		114	c->inter.put[0][1][1][1] = ff_vvc_put_qpel_hv4_8_neon_i8mm;
		115	c->inter.put[0][2][1][1] = ff_vvc_put_qpel_hv8_8_neon_i8mm;
		116	c->inter.put[0][3][1][1] = ff_vvc_put_qpel_hv16_8_neon_i8mm;
		117	c->inter.put[0][4][1][1] = ff_vvc_put_qpel_hv32_8_neon_i8mm;
		118	c->inter.put[0][5][1][1] = ff_vvc_put_qpel_hv64_8_neon_i8mm;
		119	c->inter.put[0][6][1][1] = ff_vvc_put_qpel_hv128_8_neon_i8mm;
106	}	120	}
107	} else if (bd == 10) {	121	} else if (bd == 10) {
108	c->alf.filter[LUMA] = alf_filter_luma_10_neon;	122	c->alf.filter[LUMA] = alf_filter_luma_10_neon;

aarch64/vvc: Add put_qpel_hv

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code