C908: pix_abs_0_2_c: 904.0 pix_abs_0_2_rvv_i32: 172.2 pix_abs_1_2_c: 460.0 pix_abs_1_2_rvv_i32: 168.2
Unified · Split
| @@ -34,6 +34,10 @@ int ff_pix_abs16_x2_rvv(MpegEncContext *v, const uint8_t *pix1, const uint8_t *p · 33 unmodified lines | ||
| 34 | 34 | ptrdiff_t stride, int h); |
| 35 | 35 | int ff_pix_abs8_x2_rvv(MpegEncContext *v, const uint8_t *pix1, const uint8_t *pix2, |
| 36 | 36 | ptrdiff_t stride, int h); |
| 37 | +int ff_pix_abs16_y2_rvv(MpegEncContext *v, const uint8_t *pix1, const uint8_t *pix2, | |
| 38 | + ptrdiff_t stride, int h); | |
| 39 | +int ff_pix_abs8_y2_rvv(MpegEncContext *v, const uint8_t *pix1, const uint8_t *pix2, | |
| 40 | + ptrdiff_t stride, int h); | |
| 37 | 41 | |
| 38 | 42 | av_cold void ff_me_cmp_init_riscv(MECmpContext *c, AVCodecContext *avctx) |
| 39 | 43 | { |
| @@ -47,6 +51,8 @@ av_cold void ff_me_cmp_init_riscv(MECmpContext *c, AVCodecContext *avctx) · 7 unmodified lines | ||
| 47 | 51 | c->sad[1] = ff_pix_abs8_rvv; |
| 48 | 52 | c->pix_abs[0][1] = ff_pix_abs16_x2_rvv; |
| 49 | 53 | c->pix_abs[1][1] = ff_pix_abs8_x2_rvv; |
| 54 | + c->pix_abs[0][2] = ff_pix_abs16_y2_rvv; | |
| 55 | + c->pix_abs[1][2] = ff_pix_abs8_y2_rvv; | |
| 50 | 56 | } |
| 51 | 57 | #endif |
| 52 | 58 | } |
| @@ -115,3 +115,53 @@ func ff_pix_abs8_x2_rvv, zve32x · 114 unmodified lines | ||
| 115 | 115 | |
| 116 | 116 | pix_abs_ret |
| 117 | 117 | endfunc |
| 118 | + | |
| 119 | +func ff_pix_abs16_y2_rvv, zve32x | |
| 120 | + csrwi vxrm, 0 | |
| 121 | + vsetivli zero, 1, e32, m1, ta, ma | |
| 122 | + add t1, a2, a3 | |
| 123 | + vmv.s.x v0, zero | |
| 124 | +1: | |
| 125 | + vsetivli zero, 16, e8, m1, tu, ma | |
| 126 | + vle8.v v12, (a2) | |
| 127 | + vle8.v v24, (t1) | |
| 128 | + addi a4, a4, -1 | |
| 129 | + vle8.v v4, (a1) | |
| 130 | + vaaddu.vv v12, v12, v24 | |
| 131 | + vwsubu.vv v16, v4, v12 | |
| 132 | + vwsubu.vv v20, v12, v4 | |
| 133 | + add a1, a1, a3 | |
| 134 | + vsetvli zero, zero, e16, m2, tu, ma | |
| 135 | + add a2, a2, a3 | |
| 136 | + vmax.vv v16, v16, v20 | |
| 137 | + add t1, t1, a3 | |
| 138 | + vwredsum.vs v0, v16, v0 | |
| 139 | + bnez a4, 1b | |
| 140 | + | |
| 141 | + pix_abs_ret | |
| 142 | +endfunc | |
| 143 | + | |
| 144 | +func ff_pix_abs8_y2_rvv, zve32x | |
| 145 | + csrwi vxrm, 0 | |
| 146 | + vsetivli zero, 1, e32, m1, ta, ma | |
| 147 | + add t1, a2, a3 | |
| 148 | + vmv.s.x v0, zero | |
| 149 | +1: | |
| 150 | + vsetivli zero, 8, e8, mf2, tu, ma | |
| 151 | + vle8.v v12, (a2) | |
| 152 | + vle8.v v24, (t1) | |
| 153 | + addi a4, a4, -1 | |
| 154 | + vle8.v v4, (a1) | |
| 155 | + vaaddu.vv v12, v12, v24 | |
| 156 | + vwsubu.vv v16, v4, v12 | |
| 157 | + vwsubu.vv v20, v12, v4 | |
| 158 | + add a1, a1, a3 | |
| 159 | + vsetvli zero, zero, e16, m1, tu, ma | |
| 160 | + add a2, a2, a3 | |
| 161 | + vmax.vv v16, v16, v20 | |
| 162 | + add t1, t1, a3 | |
| 163 | + vwredsum.vs v0, v16, v0 | |
| 164 | + bnez a4, 1b | |
| 165 | + | |
| 166 | + pix_abs_ret | |
| 167 | +endfunc | |