RepoMirrors/ffmpeg
1
0
Fork 0
mirror of https://git.ffmpeg.org/ffmpeg.git synced 2024-12-25 08:42:39 +00:00
Code Issues Projects Releases Wiki Activity
2fb2cd5c79
ffmpeg/libavcodec/riscv/rv40dsp_rvv.S

376 lines
11 KiB
ArmAsm
Raw Normal View History

lavc/rv40dsp: R-V V chroma_mc This is similar to h264, but here we use manual_avg instead of vaaddu because rv40's OP differs from h264. If we use vaaddu, rv40 would need to repeatedly switch between vxrm=0 and vxrm=2, and switching vxrm is very slow. C908: avg_chroma_mc4_c: 2330.0 avg_chroma_mc4_rvv_i32: 602.7 avg_chroma_mc8_c: 1211.0 avg_chroma_mc8_rvv_i32: 602.7 put_chroma_mc4_c: 1825.0 put_chroma_mc4_rvv_i32: 414.7 put_chroma_mc8_c: 932.0 put_chroma_mc8_rvv_i32: 414.7 Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>
2024-04-30 10:24:00 +00:00
/*
* Copyright (c) 2024 Institue of Software Chinese Academy of Sciences (ISCAS).
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "libavutil/riscv/asm.S"
.macro manual_avg dst src1 src2
vadd.vv \dst, \src1, \src2
vadd.vi \dst, \dst, 1
vsrl.vi \dst, \dst, 1
.endm
.macro do_chroma_mc type unroll
csrwi vxrm, 2
slli t2, a5, 3
mul t1, a5, a4
sh3add a5, a4, t2
slli a4, a4, 3
sub a5, t1, a5
sub a7, a4, t1
addi a6, a5, 64
sub t0, t2, t1
vsetvli t3, t6, e8, m1, ta, mu
beqz t1, 2f
blez a3, 8f
li t4, 0
li t2, 0
li t5, 1
addi a5, t3, 1
slli t3, a2, (1 + \unroll)
1: # if (xy != 0)
add a4, a1, t4
vsetvli zero, a5, e8, m1, ta, ma
.ifc \unroll,1
addi t2, t2, 4
.else
addi t2, t2, 2
.endif
vle8.v v10, (a4)
add a4, a4, a2
vslide1down.vx v11, v10, t5
vsetvli zero, t6, e8, m1, ta, ma
vwmulu.vx v8, v10, a6
vwmaccu.vx v8, a7, v11
vsetvli zero, a5, e8, m1, ta, ma
vle8.v v12, (a4)
vsetvli zero, t6, e8, m1, ta, ma
add a4, a4, a2
vwmaccu.vx v8, t0, v12
vsetvli zero, a5, e8, m1, ta, ma
vslide1down.vx v13, v12, t5
vsetvli zero, t6, e8, m1, ta, ma
vwmulu.vx v10, v12, a6
vwmaccu.vx v8, t1, v13
vwmaccu.vx v10, a7, v13
vsetvli zero, a5, e8, m1, ta, ma
vle8.v v14, (a4)
vsetvli zero, t6, e8, m1, ta, ma
add a4, a4, a2
vwmaccu.vx v10, t0, v14
vsetvli zero, a5, e8, m1, ta, ma
vslide1down.vx v15, v14, t5
vsetvli zero, t6, e8, m1, ta, ma
vwmulu.vx v12, v14, a6
vwmaccu.vx v10, t1, v15
vwmaccu.vx v12, a7, v15
vnclipu.wi v15, v8, 6
.ifc \type,avg
vle8.v v9, (a0)
manual_avg v15, v15, v9
.endif
vse8.v v15, (a0)
add a0, a0, a2
vnclipu.wi v8, v10, 6
.ifc \type,avg
vle8.v v9, (a0)
manual_avg v8, v8, v9
.endif
add t4, t4, t3
vse8.v v8, (a0)
add a0, a0, a2
.ifc \unroll,1
vsetvli zero, a5, e8, m1, ta, ma
vle8.v v14, (a4)
vsetvli zero, t6, e8, m1, ta, ma
add a4, a4, a2
vwmaccu.vx v12, t0, v14
vsetvli zero, a5, e8, m1, ta, ma
vslide1down.vx v15, v14, t5
vsetvli zero, t6, e8, m1, ta, ma
vwmulu.vx v16, v14, a6
vwmaccu.vx v12, t1, v15
vwmaccu.vx v16, a7, v15
vsetvli zero, a5, e8, m1, ta, ma
vle8.v v14, (a4)
vsetvli zero, t6, e8, m1, ta, ma
vwmaccu.vx v16, t0, v14
vsetvli zero, a5, e8, m1, ta, ma
vslide1down.vx v14, v14, t5
vsetvli zero, t6, e8, m1, ta, ma
vwmaccu.vx v16, t1, v14
vnclipu.wi v8, v12, 6
.ifc \type,avg
vle8.v v9, (a0)
manual_avg v8, v8, v9
.endif
vse8.v v8, (a0)
add a0, a0, a2
vnclipu.wi v8, v16, 6
.ifc \type,avg
vle8.v v9, (a0)
manual_avg v8, v8, v9
.endif
vse8.v v8, (a0)
add a0, a0, a2
.endif
blt t2, a3, 1b
j 8f
2:
bnez a4, 4f
beqz t2, 4f
blez a3, 8f
li a4, 0
li t1, 0
slli a7, a2, (1 + \unroll)
3: # if ((x8 - xy) == 0 && (y8 -xy) != 0)
add a5, a1, a4
vsetvli zero, zero, e8, m1, ta, ma
.ifc \unroll,1
addi t1, t1, 4
.else
addi t1, t1, 2
.endif
vle8.v v8, (a5)
add a5, a5, a2
add t2, a5, a2
vwmulu.vx v10, v8, a6
vle8.v v8, (a5)
vwmulu.vx v12, v8, a6
vle8.v v9, (t2)
add t2, t2, a2
add a5, t2, a2
vwmaccu.vx v10, t0, v8
add a4, a4, a7
vwmaccu.vx v12, t0, v9
vnclipu.wi v15, v10, 6
vwmulu.vx v10, v9, a6
vnclipu.wi v9, v12, 6
.ifc \type,avg
vle8.v v16, (a0)
manual_avg v15, v15, v16
.endif
vse8.v v15, (a0)
add a0, a0, a2
.ifc \type,avg
vle8.v v16, (a0)
manual_avg v9, v9, v16
.endif
vse8.v v9, (a0)
add a0, a0, a2
.ifc \unroll,1
vle8.v v8, (t2)
vle8.v v14, (a5)
vwmaccu.vx v10, t0, v8
vwmulu.vx v12, v8, a6
vnclipu.wi v8, v10, 6
vwmaccu.vx v12, t0, v14
.ifc \type,avg
vle8.v v16, (a0)
manual_avg v8, v8, v16
.endif
vse8.v v8, (a0)
add a0, a0, a2
vnclipu.wi v8, v12, 6
.ifc \type,avg
vle8.v v16, (a0)
manual_avg v8, v8, v16
.endif
vse8.v v8, (a0)
add a0, a0, a2
.endif
blt t1, a3, 3b
j 8f
4:
beqz a4, 6f
bnez t2, 6f
blez a3, 8f
li a4, 0
li t2, 0
addi t0, t3, 1
slli t1, a2, (1 + \unroll)
5: # if ((x8 - xy) != 0 && (y8 -xy) == 0)
add a5, a1, a4
vsetvli zero, t0, e8, m1, ta, ma
.ifc \unroll,1
addi t2, t2, 4
.else
addi t2, t2, 2
.endif
vle8.v v8, (a5)
add a5, a5, a2
vslide1down.vx v9, v8, t5
vsetvli zero, t6, e8, m1, ta, ma
vwmulu.vx v10, v8, a6
vwmaccu.vx v10, a7, v9
vsetvli zero, t0, e8, m1, ta, ma
vle8.v v8, (a5)
add a5, a5, a2
vslide1down.vx v9, v8, t5
vsetvli zero, t6, e8, m1, ta, ma
vwmulu.vx v12, v8, a6
vwmaccu.vx v12, a7, v9
vnclipu.wi v16, v10, 6
.ifc \type,avg
vle8.v v18, (a0)
manual_avg v16, v16, v18
.endif
vse8.v v16, (a0)
add a0, a0, a2
vnclipu.wi v10, v12, 6
.ifc \type,avg
vle8.v v18, (a0)
manual_avg v10, v10, v18
.endif
add a4, a4, t1
vse8.v v10, (a0)
add a0, a0, a2
.ifc \unroll,1
vsetvli zero, t0, e8, m1, ta, ma
vle8.v v8, (a5)
add a5, a5, a2
vslide1down.vx v9, v8, t5
vsetvli zero, t6, e8, m1, ta, ma
vwmulu.vx v14, v8, a6
vwmaccu.vx v14, a7, v9
vsetvli zero, t0, e8, m1, ta, ma
vle8.v v8, (a5)
vslide1down.vx v9, v8, t5
vsetvli zero, t6, e8, m1, ta, ma
vwmulu.vx v12, v8, a6
vnclipu.wi v8, v14, 6
vwmaccu.vx v12, a7, v9
.ifc \type,avg
vle8.v v18, (a0)
manual_avg v8, v8, v18
.endif
vse8.v v8, (a0)
add a0, a0, a2
vnclipu.wi v8, v12, 6
.ifc \type,avg
vle8.v v18, (a0)
manual_avg v8, v8, v18
.endif
vse8.v v8, (a0)
add a0, a0, a2
.endif
blt t2, a3, 5b
j 8f
6:
blez a3, 8f
li a4, 0
li t2, 0
slli a7, a2, (1 + \unroll)
7: # the final else, none of the above conditions are met
add t0, a1, a4
vsetvli zero, zero, e8, m1, ta, ma
add a5, a0, a4
add a4, a4, a7
.ifc \unroll,1
addi t2, t2, 4
.else
addi t2, t2, 2
.endif
vle8.v v8, (t0)
add t0, t0, a2
add t1, t0, a2
vwmulu.vx v10, v8, a6
vle8.v v8, (t0)
add t0, t1, a2
vnclipu.wi v13, v10, 6
vwmulu.vx v10, v8, a6
.ifc \type,avg
vle8.v v18, (a5)
manual_avg v13, v13, v18
.endif
vse8.v v13, (a5)
add a5, a5, a2
vnclipu.wi v8, v10, 6
.ifc \type,avg
vle8.v v18, (a5)
manual_avg v8, v8, v18
.endif
vse8.v v8, (a5)
add a5, a5, a2
.ifc \unroll,1
vle8.v v9, (t1)
vle8.v v12, (t0)
vwmulu.vx v10, v9, a6
vnclipu.wi v8, v10, 6
vwmulu.vx v10, v12, a6
.ifc \type,avg
vle8.v v18, (a5)
manual_avg v8, v8, v18
.endif
vse8.v v8, (a5)
add a5, a5, a2
vnclipu.wi v8, v10, 6
.ifc \type,avg
vle8.v v18, (a5)
manual_avg v8, v8, v18
.endif
vse8.v v8, (a5)
.endif
blt t2, a3, 7b
8:
ret
.endm
lavc/riscv: require B or zba explicitly
2024-07-22 17:23:50 +00:00
func ff_put_rv40_chroma_mc_rvv, zve32x, zba
lavc/rv40dsp: R-V V chroma_mc This is similar to h264, but here we use manual_avg instead of vaaddu because rv40's OP differs from h264. If we use vaaddu, rv40 would need to repeatedly switch between vxrm=0 and vxrm=2, and switching vxrm is very slow. C908: avg_chroma_mc4_c: 2330.0 avg_chroma_mc4_rvv_i32: 602.7 avg_chroma_mc8_c: 1211.0 avg_chroma_mc8_rvv_i32: 602.7 put_chroma_mc4_c: 1825.0 put_chroma_mc4_rvv_i32: 414.7 put_chroma_mc8_c: 932.0 put_chroma_mc8_rvv_i32: 414.7 Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>
2024-04-30 10:24:00 +00:00
11:
li a7, 3
blt a3, a7, 12f
do_chroma_mc put 1
12:
do_chroma_mc put 0
endfunc
lavc/riscv: require B or zba explicitly
2024-07-22 17:23:50 +00:00
func ff_avg_rv40_chroma_mc_rvv, zve32x, zba
lavc/rv40dsp: R-V V chroma_mc This is similar to h264, but here we use manual_avg instead of vaaddu because rv40's OP differs from h264. If we use vaaddu, rv40 would need to repeatedly switch between vxrm=0 and vxrm=2, and switching vxrm is very slow. C908: avg_chroma_mc4_c: 2330.0 avg_chroma_mc4_rvv_i32: 602.7 avg_chroma_mc8_c: 1211.0 avg_chroma_mc8_rvv_i32: 602.7 put_chroma_mc4_c: 1825.0 put_chroma_mc4_rvv_i32: 414.7 put_chroma_mc8_c: 932.0 put_chroma_mc8_rvv_i32: 414.7 Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>
2024-04-30 10:24:00 +00:00
21:
li a7, 3
blt a3, a7, 22f
do_chroma_mc avg 1
22:
do_chroma_mc avg 0
endfunc
func ff_put_rv40_chroma_mc8_rvv, zve32x
lavc/riscv: add forward-edge CFI landing pads
2024-07-22 19:17:40 +00:00
lpad 0
lavc/rv40dsp: R-V V chroma_mc This is similar to h264, but here we use manual_avg instead of vaaddu because rv40's OP differs from h264. If we use vaaddu, rv40 would need to repeatedly switch between vxrm=0 and vxrm=2, and switching vxrm is very slow. C908: avg_chroma_mc4_c: 2330.0 avg_chroma_mc4_rvv_i32: 602.7 avg_chroma_mc8_c: 1211.0 avg_chroma_mc8_rvv_i32: 602.7 put_chroma_mc4_c: 1825.0 put_chroma_mc4_rvv_i32: 414.7 put_chroma_mc8_c: 932.0 put_chroma_mc8_rvv_i32: 414.7 Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>
2024-04-30 10:24:00 +00:00
li t6, 8
j 11b
endfunc
func ff_put_rv40_chroma_mc4_rvv, zve32x
lavc/riscv: add forward-edge CFI landing pads
2024-07-22 19:17:40 +00:00
lpad 0
lavc/rv40dsp: R-V V chroma_mc This is similar to h264, but here we use manual_avg instead of vaaddu because rv40's OP differs from h264. If we use vaaddu, rv40 would need to repeatedly switch between vxrm=0 and vxrm=2, and switching vxrm is very slow. C908: avg_chroma_mc4_c: 2330.0 avg_chroma_mc4_rvv_i32: 602.7 avg_chroma_mc8_c: 1211.0 avg_chroma_mc8_rvv_i32: 602.7 put_chroma_mc4_c: 1825.0 put_chroma_mc4_rvv_i32: 414.7 put_chroma_mc8_c: 932.0 put_chroma_mc8_rvv_i32: 414.7 Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>
2024-04-30 10:24:00 +00:00
li t6, 4
j 11b
endfunc
func ff_avg_rv40_chroma_mc8_rvv, zve32x
lavc/riscv: add forward-edge CFI landing pads
2024-07-22 19:17:40 +00:00
lpad 0
lavc/rv40dsp: R-V V chroma_mc This is similar to h264, but here we use manual_avg instead of vaaddu because rv40's OP differs from h264. If we use vaaddu, rv40 would need to repeatedly switch between vxrm=0 and vxrm=2, and switching vxrm is very slow. C908: avg_chroma_mc4_c: 2330.0 avg_chroma_mc4_rvv_i32: 602.7 avg_chroma_mc8_c: 1211.0 avg_chroma_mc8_rvv_i32: 602.7 put_chroma_mc4_c: 1825.0 put_chroma_mc4_rvv_i32: 414.7 put_chroma_mc8_c: 932.0 put_chroma_mc8_rvv_i32: 414.7 Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>
2024-04-30 10:24:00 +00:00
li t6, 8
j 21b
endfunc
func ff_avg_rv40_chroma_mc4_rvv, zve32x
lavc/riscv: add forward-edge CFI landing pads
2024-07-22 19:17:40 +00:00
lpad 0
lavc/rv40dsp: R-V V chroma_mc This is similar to h264, but here we use manual_avg instead of vaaddu because rv40's OP differs from h264. If we use vaaddu, rv40 would need to repeatedly switch between vxrm=0 and vxrm=2, and switching vxrm is very slow. C908: avg_chroma_mc4_c: 2330.0 avg_chroma_mc4_rvv_i32: 602.7 avg_chroma_mc8_c: 1211.0 avg_chroma_mc8_rvv_i32: 602.7 put_chroma_mc4_c: 1825.0 put_chroma_mc4_rvv_i32: 414.7 put_chroma_mc8_c: 932.0 put_chroma_mc8_rvv_i32: 414.7 Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>
2024-04-30 10:24:00 +00:00
li t6, 4
j 21b
endfunc
Reference in New Issue Copy Permalink