From 6611bf548497924eb6bfa544d9aa260c393994a3 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?R=C3=A9mi=20Denis-Courmont?= Date: Sun, 1 Sep 2024 18:58:27 +0300 Subject: [PATCH] lavc/h264dsp: optimise R-V V biweight for shorter heights T-Head C908: h264_biweight2_8_c: 313.7 ( 1.00x) h264_biweight2_8_rvv_i32: before 239.5 ( 1.23x) h264_biweight2_8_rvv_i32: after 72.7 ( 4.31x) h264_biweight4_8_c: 582.0 ( 1.00x) h264_biweight4_8_rvv_i32: before 471.0 ( 1.16x) h264_biweight4_8_rvv_i32: after 91.5 ( 6.36x) h264_biweight8_8_c: 1110.0 ( 1.00x) h264_biweight8_8_rvv_i32: before 943.3 ( 1.10x) h264_biweight8_8_rvv_i64: after 147.0 ( 7.55x) SpacemiT X60: h264_biweight2_8_c: 311.4 ( 1.00x) h264_biweight2_8_rvv_i32: before 363.1 ( 0.83x) h264_biweight2_8_rvv_i32: after 103.1 ( 3.02x) h264_biweight4_8_c: 571.9 ( 1.00x) h264_biweight4_8_rvv_i32: before 717.4 ( 0.78x) h264_biweight4_8_rvv_i32: after 71.8 ( 7.96x) h264_biweight8_8_c: 1103.1 ( 1.00x) h264_biweight8_8_rvv_i32: before 1415.2 ( 0.76x) h264_biweight8_8_rvv_i64: ater 92.8 (11.88x) --- libavcodec/riscv/h264dsp_init.c | 23 +++++++---- libavcodec/riscv/h264dsp_rvv.S | 69 ++++++++++++--------------------- 2 files changed, 40 insertions(+), 52 deletions(-) diff --git a/libavcodec/riscv/h264dsp_init.c b/libavcodec/riscv/h264dsp_init.c index 643673d1d6..30dd272d6e 100644 --- a/libavcodec/riscv/h264dsp_init.c +++ b/libavcodec/riscv/h264dsp_init.c @@ -97,23 +97,30 @@ av_cold void ff_h264dsp_init_riscv(H264DSPContext *dsp, const int bit_depth, const bool zvl128b = ff_rv_vlen_least(128); if (bit_depth == 8) { - if (zvl128b && (flags & AV_CPU_FLAG_RVB)) - dsp->weight_h264_pixels_tab[0] = - ff_h264_weight_funcs_8_rvv[0].weight; - if (flags & AV_CPU_FLAG_RVV_I64) + if (zvl128b) { + if (flags & AV_CPU_FLAG_RVB) + dsp->weight_h264_pixels_tab[0] = + ff_h264_weight_funcs_8_rvv[0].weight; + dsp->biweight_h264_pixels_tab[0] = + ff_h264_weight_funcs_8_rvv[0].biweight; + } + if (flags & AV_CPU_FLAG_RVV_I64) { dsp->weight_h264_pixels_tab[1] = ff_h264_weight_funcs_8_rvv[1].weight; + dsp->biweight_h264_pixels_tab[1] = + ff_h264_weight_funcs_8_rvv[1].biweight; + } dsp->weight_h264_pixels_tab[2] = ff_h264_weight_funcs_8_rvv[2].weight; + dsp->biweight_h264_pixels_tab[2] = + ff_h264_weight_funcs_8_rvv[2].biweight; dsp->weight_h264_pixels_tab[3] = ff_h264_weight_funcs_8_rvv[3].weight; + dsp->biweight_h264_pixels_tab[3] = + ff_h264_weight_funcs_8_rvv[3].biweight; } if (bit_depth == 8 && zvl128b) { - for (int i = 0; i < 4; i++) - dsp->biweight_h264_pixels_tab[i] = - ff_h264_weight_funcs_8_rvv[i].biweight; - dsp->h264_v_loop_filter_luma = ff_h264_v_loop_filter_luma_8_rvv; dsp->h264_h_loop_filter_luma = ff_h264_h_loop_filter_luma_8_rvv; dsp->h264_h_loop_filter_luma_mbaff = diff --git a/libavcodec/riscv/h264dsp_rvv.S b/libavcodec/riscv/h264dsp_rvv.S index 464c880d4b..079896b17b 100644 --- a/libavcodec/riscv/h264dsp_rvv.S +++ b/libavcodec/riscv/h264dsp_rvv.S @@ -115,68 +115,49 @@ func ff_h264_weight_pixels\w\()_\depth\()_rvv, zve64x ret .endif endfunc -.endm - .variant_cc ff_h264_biweight_pixels_8_rvv -func ff_h264_biweight_pixels_8_rvv, zve32x +func ff_h264_biweight_pixels\w\()_\depth\()_rvv, zve64x + lpad 0 + li t6, \w + .ifb \b + j ff_h264_biweight_pixels_simple_8_rvv + .else csrwi vxrm, 2 addi a7, a7, 1 ori a7, a7, 1 sll a7, a7, a4 addi a4, a4, 1 1: - mv t0, a0 - mv t1, a1 - mv t5, t6 -2: - vsetvli t2, a3, e16, m8, ta, ma - vlsseg2e8.v v0, (t0), a2 - vlsseg2e8.v v8, (t1), a2 - addi t5, t5, -2 - vmv.v.x v16, a7 - vmv.v.x v24, a7 - vsetvli zero, zero, e8, m4, ta, ma - vwmaccsu.vx v16, a5, v0 - vwmaccsu.vx v24, a5, v4 - vwmaccsu.vx v16, a6, v8 - vwmaccsu.vx v24, a6, v12 - vsetvli zero, zero, e16, m8, ta, ma + vsetvli t1, a3, e\b, m2, ta, ma + vlse\b\().v v8, (a0), a2 + sub a3, a3, t1 + vlse\b\().v v12, (a1), a2 + mul t2, t1, a2 + vsetvli t0, zero, e16, m4, ta, ma + vmv.v.x v16, a7 + vsetvli zero, zero, e8, m2, ta, ma + vwmaccsu.vx v16, a5, v8 + add a1, a1, t2 + vwmaccsu.vx v16, a6, v12 + vsetvli zero, zero, e16, m4, ta, ma vmax.vx v16, v16, zero - vmax.vx v24, v24, zero - vsetvli zero, zero, e8, m4, ta, ma - vnclipu.wx v0, v16, a4 - vnclipu.wx v4, v24, a4 - vssseg2e8.v v0, (t0), a2 - addi t0, t0, 2 - addi t1, t1, 2 - bnez t5, 2b - - mul t3, a2, t2 - sub a3, a3, t2 - add a0, a0, t3 - add a1, a1, t3 + vsetvli zero, zero, e8, m2, ta, ma + vnclipu.wx v8, v16, a4 + vsetvli zero, t1, e\b, m2, ta, ma + vsse\b\().v v8, (a0), a2 + add a0, a0, t2 + .endif bnez a3, 1b ret endfunc +.endm h264_weight 8, 2, 16 h264_weight 8, 4, 32 h264_weight 8, 8, 64 h264_weight 8, 16 -.irp w, 16, 8, 4, 2 -func ff_h264_biweight_pixels\w\()_8_rvv, zve32x - lpad 0 - li t6, \w - .if \w == 16 - j ff_h264_biweight_pixels_simple_8_rvv - .else - j ff_h264_biweight_pixels_8_rvv - .endif -endfunc -.endr - .global ff_h264_weight_funcs_8_rvv .hidden ff_h264_weight_funcs_8_rvv const ff_h264_weight_funcs_8_rvv