Hi, why is there an issue with the ABI? I previously just thought that s0 shouldn't be used here.
Rémi Denis-Courmont <r...@remlab.net> 于2024年12月1日周日 00:19写道： > Le keskiviikkona 20. marraskuuta 2024, 3.26.52 EET u...@foxmail.com a > écrit : > > From: sunyuechi <sunyue...@iscas.ac.cn> > > > > --- > > libavcodec/riscv/rv40dsp_rvv.S | 113 ++++++++++++++++++++++----------- > > 1 file changed, 75 insertions(+), 38 deletions(-) > > > > diff --git a/libavcodec/riscv/rv40dsp_rvv.S > b/libavcodec/riscv/rv40dsp_rvv.S > > index ca431eb8ab..634cdd2b34 100644 > > --- a/libavcodec/riscv/rv40dsp_rvv.S > > +++ b/libavcodec/riscv/rv40dsp_rvv.S > > @@ -20,15 +20,28 @@ > > > > #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 > > +const rv40_bias > > + .byte 0, 16, 32, 16 > > + .byte 32, 28, 32, 28 > > + .byte 0, 32, 16, 32 > > + .byte 32, 28, 32, 28 > > +endconst > > > > .macro do_chroma_mc type unroll > > - csrwi vxrm, 2 > > + csrwi vxrm, 0 > > +#if __riscv_xlen == 64 > > + addi sp, sp, -8 > > + sd s2, (sp) > > +#else > > + addi sp, sp, -4 > > + sw s2, (sp) > > +#endif > > Still not ABI compliant. And exclude rv128 if you can't find a way not to > spill. > > > + lla t4, rv40_bias > > + srli t5, a4, 1 > > + sh1add t4, a5, t4 > > + add t5, t4, t5 > > slli t2, a5, 3 > > + lb s2, (t5) > > mul t1, a5, a4 > > sh3add a5, a4, t2 > > slli a4, a4, 3 > > @@ -80,17 +93,19 @@ > > vwmulu.vx v12, v14, a6 > > vwmaccu.vx v10, t1, v15 > > vwmaccu.vx v12, a7, v15 > > - vnclipu.wi v15, v8, 6 > > + vwaddu.wx v20, v8, s2 > > + vnsrl.wi v15, v20, 6 > > .ifc \type,avg > > vle8.v v9, (a0) > > - manual_avg v15, v15, v9 > > + vaaddu.vv v15, v15, v9 > > .endif > > vse8.v v15, (a0) > > add a0, a0, a2 > > - vnclipu.wi v8, v10, 6 > > + vwaddu.wx v20, v10, s2 > > + vnsrl.wi v8, v20, 6 > > .ifc \type,avg > > vle8.v v9, (a0) > > - manual_avg v8, v8, v9 > > + vaaddu.vv v8, v8, v9 > > .endif > > add t4, t4, t3 > > vse8.v v8, (a0) > > @@ -115,17 +130,19 @@ > > vslide1down.vx v14, v14, t5 > > vsetvli zero, t6, e8, m1, ta, ma > > vwmaccu.vx v16, t1, v14 > > - vnclipu.wi v8, v12, 6 > > + vwaddu.wx v20, v12, s2 > > + vnsrl.wi v8, v20, 6 > > .ifc \type,avg > > vle8.v v9, (a0) > > - manual_avg v8, v8, v9 > > + vaaddu.vv v8, v8, v9 > > .endif > > vse8.v v8, (a0) > > add a0, a0, a2 > > - vnclipu.wi v8, v16, 6 > > + vwaddu.wx v20, v16, s2 > > + vnsrl.wi v8, v20, 6 > > .ifc \type,avg > > vle8.v v9, (a0) > > - manual_avg v8, v8, v9 > > + vaaddu.vv v8, v8, v9 > > .endif > > vse8.v v8, (a0) > > add a0, a0, a2 > > @@ -159,18 +176,20 @@ > > vwmaccu.vx v10, t0, v8 > > add a4, a4, a7 > > vwmaccu.vx v12, t0, v9 > > - vnclipu.wi v15, v10, 6 > > + vwaddu.wx v20, v10, s2 > > + vnsrl.wi v15, v20, 6 > > vwmulu.vx v10, v9, a6 > > - vnclipu.wi v9, v12, 6 > > + vwaddu.wx v20, v12, s2 > > + vnsrl.wi v9, v20, 6 > > .ifc \type,avg > > vle8.v v16, (a0) > > - manual_avg v15, v15, v16 > > + vaaddu.vv v15, v15, v16 > > .endif > > vse8.v v15, (a0) > > add a0, a0, a2 > > .ifc \type,avg > > vle8.v v16, (a0) > > - manual_avg v9, v9, v16 > > + vaaddu.vv v9, v9, v16 > > .endif > > vse8.v v9, (a0) > > add a0, a0, a2 > > @@ -179,18 +198,20 @@ > > vle8.v v14, (a5) > > vwmaccu.vx v10, t0, v8 > > vwmulu.vx v12, v8, a6 > > - vnclipu.wi v8, v10, 6 > > + vwaddu.wx v20, v10, s2 > > + vnsrl.wi v8, v20, 6 > > vwmaccu.vx v12, t0, v14 > > .ifc \type,avg > > vle8.v v16, (a0) > > - manual_avg v8, v8, v16 > > + vaaddu.vv v8, v8, v16 > > .endif > > vse8.v v8, (a0) > > add a0, a0, a2 > > - vnclipu.wi v8, v12, 6 > > + vwaddu.wx v20, v12, s2 > > + vnsrl.wi v8, v20, 6 > > .ifc \type,avg > > vle8.v v16, (a0) > > - manual_avg v8, v8, v16 > > + vaaddu.vv v8, v8, v16 > > .endif > > vse8.v v8, (a0) > > add a0, a0, a2 > > @@ -226,17 +247,19 @@ > > vsetvli zero, t6, e8, m1, ta, ma > > vwmulu.vx v12, v8, a6 > > vwmaccu.vx v12, a7, v9 > > - vnclipu.wi v16, v10, 6 > > + vwaddu.wx v20, v10, s2 > > + vnsrl.wi v16, v20, 6 > > .ifc \type,avg > > vle8.v v18, (a0) > > - manual_avg v16, v16, v18 > > + vaaddu.vv v16, v16, v18 > > .endif > > vse8.v v16, (a0) > > add a0, a0, a2 > > - vnclipu.wi v10, v12, 6 > > + vwaddu.wx v20, v12, s2 > > + vnsrl.wi v10, v20, 6 > > .ifc \type,avg > > vle8.v v18, (a0) > > - manual_avg v10, v10, v18 > > + vaaddu.vv v10, v10, v18 > > .endif > > add a4, a4, t1 > > vse8.v v10, (a0) > > @@ -254,18 +277,20 @@ > > vslide1down.vx v9, v8, t5 > > vsetvli zero, t6, e8, m1, ta, ma > > vwmulu.vx v12, v8, a6 > > - vnclipu.wi v8, v14, 6 > > + vwaddu.wx v20, v14, s2 > > + vnsrl.wi v8, v20, 6 > > vwmaccu.vx v12, a7, v9 > > .ifc \type,avg > > vle8.v v18, (a0) > > - manual_avg v8, v8, v18 > > + vaaddu.vv v8, v8, v18 > > .endif > > vse8.v v8, (a0) > > add a0, a0, a2 > > - vnclipu.wi v8, v12, 6 > > + vwaddu.wx v20, v12, s2 > > + vnsrl.wi v8, v20, 6 > > .ifc \type,avg > > vle8.v v18, (a0) > > - manual_avg v8, v8, v18 > > + vaaddu.vv v8, v8, v18 > > .endif > > vse8.v v8, (a0) > > add a0, a0, a2 > > @@ -293,18 +318,20 @@ > > vwmulu.vx v10, v8, a6 > > vle8.v v8, (t0) > > add t0, t1, a2 > > - vnclipu.wi v13, v10, 6 > > + vwaddu.wx v20, v10, s2 > > + vnsrl.wi v13, v20, 6 > > vwmulu.vx v10, v8, a6 > > .ifc \type,avg > > vle8.v v18, (a5) > > - manual_avg v13, v13, v18 > > + vaaddu.vv v13, v13, v18 > > .endif > > vse8.v v13, (a5) > > add a5, a5, a2 > > - vnclipu.wi v8, v10, 6 > > + vwaddu.wx v20, v10, s2 > > + vnsrl.wi v8, v20, 6 > > .ifc \type,avg > > vle8.v v18, (a5) > > - manual_avg v8, v8, v18 > > + vaaddu.vv v8, v8, v18 > > .endif > > vse8.v v8, (a5) > > add a5, a5, a2 > > @@ -312,23 +339,33 @@ > > vle8.v v9, (t1) > > vle8.v v12, (t0) > > vwmulu.vx v10, v9, a6 > > - vnclipu.wi v8, v10, 6 > > + vwaddu.wx v20, v10, s2 > > + vnsrl.wi v8, v20, 6 > > vwmulu.vx v10, v12, a6 > > .ifc \type,avg > > vle8.v v18, (a5) > > - manual_avg v8, v8, v18 > > + vaaddu.vv v8, v8, v18 > > .endif > > vse8.v v8, (a5) > > add a5, a5, a2 > > - vnclipu.wi v8, v10, 6 > > + vwaddu.wx v20, v10, s2 > > + vnsrl.wi v8, v20, 6 > > .ifc \type,avg > > vle8.v v18, (a5) > > - manual_avg v8, v8, v18 > > + vaaddu.vv v8, v8, v18 > > .endif > > vse8.v v8, (a5) > > .endif > > blt t2, a3, 7b > > 8: > > +#if __riscv_xlen == 64 > > + ld s2, (sp) > > + addi sp, sp, 8 > > +#else > > + lw s2, (sp) > > + addi sp, sp, 4 > > +#endif > > + > > ret > > .endm > > > -- > Rémi Denis-Courmont > http://www.remlab.net/ > > > > _______________________________________________ > ffmpeg-devel mailing list > ffmpeg-devel@ffmpeg.org > https://ffmpeg.org/mailman/listinfo/ffmpeg-devel > > To unsubscribe, visit link above, or email > ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe". > _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
