lavc/hevc: R-V V put_pixels(pow2)

k230 banana_f3 put_hevc_pel_pixels4_8_c: 61.6 ( 1.00x) 69.5 ( 1.00x) put_hevc_pel_pixels4_8_rvv_i32: 24.6 ( 2.50x) 28.0 ( 2.48x) put_hevc_pel_pixels8_8_c: 209.8 ( 1.00x) 215.5 ( 1.00x) put_hevc_pel_pixels8_8_rvv_i32: 52.6 ( 3.99x) 38.2 ( 5.64x) put_hevc_pel_pixels16_8_c: 839.4 ( 1.00x) 830.0 ( 1.00x) put_hevc_pel_pixels16_8_rvv_i32: 126.6 ( 6.63x) 90.5 ( 9.17x) put_hevc_pel_pixels32_8_c: 3246.6 ( 1.00x) 3246.7 ( 1.00x) put_hevc_pel_pixels32_8_rvv_i32: 311.6 (10.42x) 257.0 (12.63x) put_hevc_pel_pixels64_8_c: 12894.6 ( 1.00x) 12892.7 ( 1.00x) put_hevc_pel_pixels64_8_rvv_i32: 1135.8 (11.35x) 778.0 (16.57x)

sunyuechi committed 1 year ago

Commit b3f74401 parent dad062c5 files · +96 −0Browse files

Unified · Split

5 files

libavcodec/hevc/dsp.c +2 −0
libavcodec/hevc/dsp.h +1 −0
libavcodec/riscv/Makefile +2 −0
libavcodec/riscv/h26x/h2656_inter_rvv.S +24 −0
libavcodec/riscv/hevcdsp_init.c +67 −0

libavcodec/hevc/dsp.c+2−0

@@ -265,6 +265,8 @@ int i = 0; · 264 unmodified lines
265	265	ff_hevc_dsp_init_arm(hevcdsp, bit_depth);
266	266	#elif ARCH_PPC
267	267	ff_hevc_dsp_init_ppc(hevcdsp, bit_depth);
	268	+#elif ARCH_RISCV
	269	+ ff_hevc_dsp_init_riscv(hevcdsp, bit_depth);
268	270	#elif ARCH_WASM
269	271	ff_hevc_dsp_init_wasm(hevcdsp, bit_depth);
270	272	#elif ARCH_X86

libavcodec/hevc/dsp.h+1−0

@@ -133,6 +133,7 @@ extern const int8_t ff_hevc_qpel_filters[4][16]; · 132 unmodified lines
133	133	void ff_hevc_dsp_init_aarch64(HEVCDSPContext *c, const int bit_depth);
134	134	void ff_hevc_dsp_init_arm(HEVCDSPContext *c, const int bit_depth);
135	135	void ff_hevc_dsp_init_ppc(HEVCDSPContext *c, const int bit_depth);
	136	+void ff_hevc_dsp_init_riscv(HEVCDSPContext *c, const int bit_depth);
136	137	void ff_hevc_dsp_init_wasm(HEVCDSPContext *c, const int bit_depth);
137	138	void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth);
138	139	void ff_hevc_dsp_init_mips(HEVCDSPContext *c, const int bit_depth);

libavcodec/riscv/Makefile+2−0

@@ -35,6 +35,8 @@ RVV-OBJS-$(CONFIG_H264DSP) += riscv/h264addpx_rvv.o riscv/h264dsp_rvv.o \ · 34 unmodified lines
35	35	riscv/h264idct_rvv.o
36	36	OBJS-$(CONFIG_H264QPEL) += riscv/h264qpel_init.o
37	37	RVV-OBJS-$(CONFIG_H264QPEL) += riscv/h264qpel_rvv.o
	38	+OBJS-$(CONFIG_HEVC_DECODER) += riscv/hevcdsp_init.o
	39	+RVV-OBJS-$(CONFIG_HEVC_DECODER) += riscv/h26x/h2656_inter_rvv.o
38	40	OBJS-$(CONFIG_HUFFYUV_DECODER) += riscv/huffyuvdsp_init.o
39	41	RVV-OBJS-$(CONFIG_HUFFYUV_DECODER) += riscv/huffyuvdsp_rvv.o
40	42	OBJS-$(CONFIG_IDCTDSP) += riscv/idctdsp_init.o

libavcodec/riscv/h26x/h2656_inter_rvv.S+24−0

@@ -0,0 +1,24 @@
	1	+/*
	2	+ * Copyright (c) 2024 Institue of Software Chinese Academy of Sciences (ISCAS).
	3	+ *
	4	+ * This file is part of FFmpeg.
	5	+ *
	6	+ * FFmpeg is free software; you can redistribute it and/or
	7	+ * modify it under the terms of the GNU Lesser General Public
	8	+ * License as published by the Free Software Foundation; either
	9	+ * version 2.1 of the License, or (at your option) any later version.
	10	+ *
	11	+ * FFmpeg is distributed in the hope that it will be useful,
	12	+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
	13	+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	14	+ * Lesser General Public License for more details.
	15	+ *
	16	+ * You should have received a copy of the GNU Lesser General Public
	17	+ * License along with FFmpeg; if not, write to the Free Software
	18	+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
	19	+ */
	20	+
	21	+#include "libavcodec/riscv/h26x/asm.S"
	22	+
	23	+func_put_pixels 256, 64, h2656
	24	+func_put_pixels 128, 64, h2656

libavcodec/riscv/hevcdsp_init.c+67−0

@@ -0,0 +1,67 @@
	1	+/*
	2	+ * Copyright (c) 2024 Institue of Software Chinese Academy of Sciences (ISCAS).
	3	+ *
	4	+ * This file is part of FFmpeg.
	5	+ *
	6	+ * FFmpeg is free software; you can redistribute it and/or
	7	+ * modify it under the terms of the GNU Lesser General Public
	8	+ * License as published by the Free Software Foundation; either
	9	+ * version 2.1 of the License, or (at your option) any later version.
	10	+ *
	11	+ * FFmpeg is distributed in the hope that it will be useful,
	12	+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
	13	+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	14	+ * Lesser General Public License for more details.
	15	+ *
	16	+ * You should have received a copy of the GNU Lesser General Public
	17	+ * License along with FFmpeg; if not, write to the Free Software
	18	+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
	19	+ */
	20	+
	21	+#include "config.h"
	22	+
	23	+#include "libavutil/attributes.h"
	24	+#include "libavutil/cpu.h"
	25	+#include "libavutil/riscv/cpu.h"
	26	+
	27	+#include "libavcodec/hevc/dsp.h"
	28	+#include "libavcodec/riscv/h26x/h2656dsp.h"
	29	+
	30	+#define RVV_FNASSIGN(member, v, h, fn, ext) \
	31	+ member[1][v][h] = ff_h2656_put_pixels_##8_##ext; \
	32	+ member[3][v][h] = ff_h2656_put_pixels_##8_##ext; \
	33	+ member[5][v][h] = ff_h2656_put_pixels_##8_##ext; \
	34	+ member[7][v][h] = ff_h2656_put_pixels_##8_##ext; \
	35	+ member[9][v][h] = ff_h2656_put_pixels_##8_##ext;
	36	+
	37	+void ff_hevc_dsp_init_riscv(HEVCDSPContext *c, const int bit_depth)
	38	+{
	39	+#if HAVE_RVV
	40	+ const int flags = av_get_cpu_flags();
	41	+ int vlenb;
	42	+
	43	+ if (!(flags & AV_CPU_FLAG_RVV_I32) \|\| !(flags & AV_CPU_FLAG_RVB))
	44	+ return;
	45	+
	46	+ vlenb = ff_get_rv_vlenb();
	47	+ if (vlenb >= 32) {
	48	+ switch (bit_depth) {
	49	+ case 8:
	50	+ RVV_FNASSIGN(c->put_hevc_qpel, 0, 0, pel_pixels, rvv_256);
	51	+ RVV_FNASSIGN(c->put_hevc_epel, 0, 0, pel_pixels, rvv_256);
	52	+ break;
	53	+ default:
	54	+ break;
	55	+ }
	56	+ } else if (vlenb >= 16) {
	57	+ switch (bit_depth) {
	58	+ case 8:
	59	+ RVV_FNASSIGN(c->put_hevc_qpel, 0, 0, pel_pixels, rvv_128);
	60	+ RVV_FNASSIGN(c->put_hevc_epel, 0, 0, pel_pixels, rvv_128);
	61	+ break;
	62	+ default:
	63	+ break;
	64	+ }
	65	+ }
	66	+#endif
	67	+}

lavc/hevc: R-V V put_pixels(pow2)

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code