lavc/me_cmp: R-V V vsse vsad

C908:
vsad_0_c: 936.0
vsad_0_rvv_i32: 236.2
vsad_1_c: 424.0
vsad_1_rvv_i32: 190.2
vsse_0_c: 877.0
vsse_0_rvv_i32: 204.2
vsse_1_c: 439.0
vsse_1_rvv_i32: 140.2

Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

libavcodec/riscv/me_cmp_init.c

@@ -46,6 +46,11 @@ int ff_sse8_rvv(MpegEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
 int ff_sse4_rvv(MpegEncContext *v, const uint8_t *pix1, const uint8_t *pix2,
                    ptrdiff_t stride, int h);
 
+int ff_vsse16_rvv(MpegEncContext *c, const uint8_t *s1, const uint8_t *s2, ptrdiff_t stride, int h);
+int ff_vsse8_rvv(MpegEncContext *c, const uint8_t *s1, const uint8_t *s2, ptrdiff_t stride, int h);
+int ff_vsad16_rvv(MpegEncContext *c, const uint8_t *s1, const uint8_t *s2, ptrdiff_t stride, int h);
+int ff_vsad8_rvv(MpegEncContext *c, const uint8_t *s1, const uint8_t *s2, ptrdiff_t stride, int h);
+
 av_cold void ff_me_cmp_init_riscv(MECmpContext *c, AVCodecContext *avctx)
 {
 #if HAVE_RVV
@@ -64,6 +69,11 @@ av_cold void ff_me_cmp_init_riscv(MECmpContext *c, AVCodecContext *avctx)
         c->sse[0] = ff_sse16_rvv;
         c->sse[1] = ff_sse8_rvv;
         c->sse[2] = ff_sse4_rvv;
+
+        c->vsse[0] = ff_vsse16_rvv;
+        c->vsse[1] = ff_vsse8_rvv;
+        c->vsad[0] = ff_vsad16_rvv;
+        c->vsad[1] = ff_vsad8_rvv;
     }
 #endif
 }

libavcodec/riscv/me_cmp_rvv.S

@@ -231,3 +231,101 @@ func ff_sse4_rvv, zve32x
         vmv.x.s         a0, v0
         ret
 endfunc
+
+.macro vabsaddu dst src tmp
+        vneg.v          \tmp, \src
+        vmax.vv         \tmp, \src, \tmp
+        vwaddu.wv       \dst, \dst, \tmp
+.endm
+
+.macro  vsad_vsse16 type
+        vsetivli        t0, 16, e32, m4, ta, ma
+        addi            a4, a4, -1
+        add             t1, a1, a3
+        add             t2, a2, a3
+        vmv.v.x         v24, zero
+        vmv.s.x         v0, zero
+1:
+        vsetvli         zero, zero, e8, m1, tu, ma
+        vle8.v          v4, (a1)
+        vle8.v          v8, (t1)
+        vle8.v          v12, (a2)
+        vle8.v          v16, (t2)
+        addi            a4, a4, -1
+        vwsubu.vv       v28, v4, v12
+        vwsubu.wv       v12, v28, v8
+        vwaddu.wv       v28, v12, v16
+        vsetvli         zero, zero, e16, m2, tu, ma
+
+.ifc \type,abs
+        vabsaddu        v24, v28, v12
+.endif
+.ifc \type,square
+        vwmacc.vv       v24, v28, v28
+.endif
+
+        add             a1, a1, a3
+        add             a2, a2, a3
+        add             t1, t1, a3
+        add             t2, t2, a3
+        bnez            a4, 1b
+
+        vsetvli         zero, zero, e32, m4, tu, ma
+        vredsum.vs      v0, v24, v0
+        vmv.x.s         a0, v0
+        ret
+.endm
+
+.macro  vsad_vsse8 type
+        vsetivli        t0, 8, e32, m2, ta, ma
+        addi            a4, a4, -1
+        add             t1, a1, a3
+        add             t2, a2, a3
+        vmv.v.x         v24, zero
+        vmv.s.x         v0, zero
+1:
+        vsetvli         zero, zero, e8, mf2, tu, ma
+        vle8.v          v4, (a1)
+        vle8.v          v8, (t1)
+        vle8.v          v12, (a2)
+        vle8.v          v16, (t2)
+        addi            a4, a4, -1
+        vwsubu.vv       v28, v4, v12
+        vwsubu.wv       v12, v28, v8
+        vwaddu.wv       v28, v12, v16
+        vsetvli         zero, zero, e16, m1, tu, ma
+
+.ifc \type,abs
+        vabsaddu        v24, v28, v12
+.endif
+.ifc \type,square
+        vwmacc.vv       v24, v28, v28
+.endif
+
+        add             a1, a1, a3
+        add             a2, a2, a3
+        add             t1, t1, a3
+        add             t2, t2, a3
+        bnez            a4, 1b
+
+        vsetvli         zero, zero, e32, m2, tu, ma
+        vredsum.vs      v0, v24, v0
+        vmv.x.s         a0, v0
+        ret
+.endm
+
+func ff_vsse16_rvv, zve32x
+        vsad_vsse16 square
+endfunc
+
+func ff_vsse8_rvv, zve32x
+        vsad_vsse8 square
+endfunc
+
+func ff_vsad16_rvv, zve32x
+        vsad_vsse16 abs
+endfunc
+
+func ff_vsad8_rvv, zve32x
+        vsad_vsse8 abs
+endfunc