jpeg2000_rct_int_c: 2592.2 jpeg2000_rct_int_rvv_i32: 1154.2
Unified · Split
| @@ -24,13 +24,17 @@ · 23 unmodified lines | |||
| 24 | #include "libavcodec/jpeg2000dsp.h" | 24 | #include "libavcodec/jpeg2000dsp.h" |
| 25 | 25 | ||
| 26 | void ff_ict_float_rvv(void *src0, void *src1, void *src2, int csize); | 26 | void ff_ict_float_rvv(void *src0, void *src1, void *src2, int csize); |
| 27 | void ff_rct_int_rvv(void *src0, void *src1, void *src2, int csize); | ||
| 27 | 28 | ||
| 28 | av_cold void ff_jpeg2000dsp_init_riscv(Jpeg2000DSPContext *c) | 29 | av_cold void ff_jpeg2000dsp_init_riscv(Jpeg2000DSPContext *c) |
| 29 | { | 30 | { |
| 30 | #if HAVE_RVV | 31 | #if HAVE_RVV |
| 31 | int flags = av_get_cpu_flags(); | 32 | int flags = av_get_cpu_flags(); |
| 32 | 33 | ||
| 33 | if ((flags & AV_CPU_FLAG_RVV_F32) && (flags & AV_CPU_FLAG_RVB_ADDR)) | 34 | if ((flags & AV_CPU_FLAG_RVV_I32) && (flags & AV_CPU_FLAG_RVB_ADDR)) { |
| 34 | c->mct_decode[FF_DWT97] = ff_ict_float_rvv; | 35 | if (flags & AV_CPU_FLAG_RVV_F32) |
| 36 | c->mct_decode[FF_DWT97] = ff_ict_float_rvv; | ||
| 37 | c->mct_decode[FF_DWT53] = ff_rct_int_rvv; | ||
| 38 | } | ||
| 35 | #endif | 39 | #endif |
| 36 | } | 40 | } |
| @@ -47,3 +47,26 @@ func ff_ict_float_rvv, zve32f · 46 unmodified lines | |||
| 47 | 47 | ||
| 48 | ret | 48 | ret |
| 49 | endfunc | 49 | endfunc |
| 50 | |||
| 51 | func ff_rct_int_rvv, zve32x | ||
| 52 | 1: | ||
| 53 | vsetvli t0, a3, e32, m8, ta, ma | ||
| 54 | vle32.v v16, (a1) | ||
| 55 | sub a3, a3, t0 | ||
| 56 | vle32.v v24, (a2) | ||
| 57 | vle32.v v8, (a0) | ||
| 58 | vadd.vv v0, v16, v24 | ||
| 59 | vsra.vi v0, v0, 2 | ||
| 60 | vsub.vv v0, v8, v0 | ||
| 61 | vadd.vv v8, v0, v24 | ||
| 62 | vadd.vv v24, v0, v16 | ||
| 63 | vse32.v v8, (a0) | ||
| 64 | sh2add a0, t0, a0 | ||
| 65 | vse32.v v0, (a1) | ||
| 66 | sh2add a1, t0, a1 | ||
| 67 | vse32.v v24, (a2) | ||
| 68 | sh2add a2, t0, a2 | ||
| 69 | bnez a3, 1b | ||
| 70 | |||
| 71 | ret | ||
| 72 | endfunc | ||