/* * Copyright (c) 2026, Alliance for Open Media. All rights reserved. * * This source code is subject to the terms of the BSD 2 Clause License and * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License * was not distributed with this source code in the LICENSE file, you can * obtain it at www.aomedia.org/license/software. If the Alliance for Open * Media Patent License 1.0 was not distributed with this source code in the * PATENTS file, you can obtain it at www.aomedia.org/license/patent. */ #include #include #include "config/aom_config.h" #include "config/av1_rtcd.h" #include "aom_dsp/riscv/mem_rvv.h" #include "av1/common/convolve.h" #include "av1/common/restoration.h" static inline vuint16m1_t highbd_wiener_convolve5_8_2d_h_rvv( const vint16m1_t s0, const vint16m1_t s1, const vint16m1_t s2, const vint16m1_t s3, const vint16m1_t s4, const vint16m1_t xf1, const vint16m1_t xf2, const vint16m1_t xf3, const vint32m2_t round_vec, int horiz_shift, uint16_t im_max_val, size_t vl) { vint16m1_t s04 = __riscv_vadd_vv_i16m1(s0, s4, vl); vint16m1_t s13 = __riscv_vadd_vv_i16m1(s1, s3, vl); vint32m2_t sum = round_vec; sum = __riscv_vwmacc_vv_i32m2(sum, xf1, s04, vl); sum = __riscv_vwmacc_vv_i32m2(sum, xf2, s13, vl); sum = __riscv_vwmacc_vv_i32m2(sum, xf3, s2, vl); #if __riscv_v_intrinsic >= 12000 vuint16m1_t res = __riscv_vnclipu_wx_u16m1( __riscv_vreinterpret_v_i32m2_u32m2(__riscv_vmax_vx_i32m2(sum, 0, vl)), (size_t)horiz_shift, __RISCV_VXRM_RNU, vl); #else vuint16m1_t res = __riscv_vnclipu_wx_u16m1( __riscv_vreinterpret_v_i32m2_u32m2(__riscv_vmax_vx_i32m2(sum, 0, vl)), (size_t)horiz_shift, vl); #endif return __riscv_vminu_vx_u16m1(res, im_max_val, vl); } static inline vuint16m1_t highbd_wiener_convolve7_8_2d_h_rvv( const vint16m1_t s0, const vint16m1_t s1, const vint16m1_t s2, const vint16m1_t s3, const vint16m1_t s4, const vint16m1_t s5, const vint16m1_t s6, const vint16m1_t xf0, const vint16m1_t xf1, const vint16m1_t xf2, const vint16m1_t xf3, const vint32m2_t round_vec, int horiz_shift, uint16_t im_max_val, size_t vl) { vint16m1_t s06 = __riscv_vadd_vv_i16m1(s0, s6, vl); vint16m1_t s15 = __riscv_vadd_vv_i16m1(s1, s5, vl); vint16m1_t s24 = __riscv_vadd_vv_i16m1(s2, s4, vl); vint32m2_t sum = round_vec; sum = __riscv_vwmacc_vv_i32m2(sum, xf0, s06, vl); sum = __riscv_vwmacc_vv_i32m2(sum, xf1, s15, vl); sum = __riscv_vwmacc_vv_i32m2(sum, xf2, s24, vl); sum = __riscv_vwmacc_vv_i32m2(sum, xf3, s3, vl); #if __riscv_v_intrinsic >= 12000 vuint16m1_t res = __riscv_vnclipu_wx_u16m1( __riscv_vreinterpret_v_i32m2_u32m2(__riscv_vmax_vx_i32m2(sum, 0, vl)), (size_t)horiz_shift, __RISCV_VXRM_RNU, vl); #else vuint16m1_t res = __riscv_vnclipu_wx_u16m1( __riscv_vreinterpret_v_i32m2_u32m2(__riscv_vmax_vx_i32m2(sum, 0, vl)), (size_t)horiz_shift, vl); #endif return __riscv_vminu_vx_u16m1(res, im_max_val, vl); } static inline vuint16m1_t highbd_wiener_convolve5_8_2d_v_rvv( const vint16m1_t s0, const vint16m1_t s1, const vint16m1_t s2, const vint16m1_t s3, const vint16m1_t s4, const vint16m1_t yf1, const vint16m1_t yf2, const vint16m1_t yf3, const vint32m2_t round_vec, int vert_shift, uint16_t res_max_val, size_t vl) { vint32m2_t sum = round_vec; sum = __riscv_vwmacc_vv_i32m2(sum, yf1, s0, vl); sum = __riscv_vwmacc_vv_i32m2(sum, yf1, s4, vl); sum = __riscv_vwmacc_vv_i32m2(sum, yf2, s1, vl); sum = __riscv_vwmacc_vv_i32m2(sum, yf2, s3, vl); sum = __riscv_vwmacc_vv_i32m2(sum, yf3, s2, vl); #if __riscv_v_intrinsic >= 12000 vuint16m1_t res = __riscv_vnclipu_wx_u16m1( __riscv_vreinterpret_v_i32m2_u32m2(__riscv_vmax_vx_i32m2(sum, 0, vl)), (size_t)vert_shift, __RISCV_VXRM_RNU, vl); #else vuint16m1_t res = __riscv_vnclipu_wx_u16m1( __riscv_vreinterpret_v_i32m2_u32m2(__riscv_vmax_vx_i32m2(sum, 0, vl)), (size_t)vert_shift, vl); #endif return __riscv_vminu_vx_u16m1(res, res_max_val, vl); } static inline vuint16m1_t highbd_wiener_convolve7_8_2d_v_rvv( const vint16m1_t s0, const vint16m1_t s1, const vint16m1_t s2, const vint16m1_t s3, const vint16m1_t s4, const vint16m1_t s5, const vint16m1_t s6, const vint16m1_t yf0, const vint16m1_t yf1, const vint16m1_t yf2, const vint16m1_t yf3, const vint32m2_t round_vec, int vert_shift, uint16_t res_max_val, size_t vl) { vint32m2_t sum = round_vec; sum = __riscv_vwmacc_vv_i32m2(sum, yf0, s0, vl); sum = __riscv_vwmacc_vv_i32m2(sum, yf0, s6, vl); sum = __riscv_vwmacc_vv_i32m2(sum, yf1, s1, vl); sum = __riscv_vwmacc_vv_i32m2(sum, yf1, s5, vl); sum = __riscv_vwmacc_vv_i32m2(sum, yf2, s2, vl); sum = __riscv_vwmacc_vv_i32m2(sum, yf2, s4, vl); sum = __riscv_vwmacc_vv_i32m2(sum, yf3, s3, vl); #if __riscv_v_intrinsic >= 12000 vuint16m1_t res = __riscv_vnclipu_wx_u16m1( __riscv_vreinterpret_v_i32m2_u32m2(__riscv_vmax_vx_i32m2(sum, 0, vl)), (size_t)vert_shift, __RISCV_VXRM_RNU, vl); #else vuint16m1_t res = __riscv_vnclipu_wx_u16m1( __riscv_vreinterpret_v_i32m2_u32m2(__riscv_vmax_vx_i32m2(sum, 0, vl)), (size_t)vert_shift, vl); #endif return __riscv_vminu_vx_u16m1(res, res_max_val, vl); } static void highbd_convolve_add_src_horiz_5tap_rvv( const uint16_t *src_ptr, ptrdiff_t src_stride, uint16_t *dst_ptr, ptrdiff_t dst_stride, int w, int h, const int16_t *x_filter, int32_t round_val, int horiz_shift, uint16_t im_max_val) { const size_t vl_max = __riscv_vsetvlmax_e16m1(); const vint16m1_t xf1 = __riscv_vmv_v_x_i16m1(x_filter[1], vl_max); const vint16m1_t xf2 = __riscv_vmv_v_x_i16m1(x_filter[2], vl_max); const vint16m1_t xf3 = __riscv_vmv_v_x_i16m1(x_filter[3], vl_max); const vint32m2_t round_vec = __riscv_vmv_v_x_i32m2(round_val, vl_max); int row = 0; for (; row <= h - 2; row += 2) { const int16_t *sa = (const int16_t *)src_ptr; const int16_t *sb = (const int16_t *)(src_ptr + src_stride); uint16_t *da = dst_ptr; uint16_t *db = dst_ptr + dst_stride; int width = w; while (width > 0) { const size_t vl = __riscv_vsetvl_e16m1((size_t)width); vint16m1_t a0, a1, a2, a3, a4; load_s16_8x5(sa, 1, &a0, &a1, &a2, &a3, &a4, vl); vint16m1_t b0, b1, b2, b3, b4; load_s16_8x5(sb, 1, &b0, &b1, &b2, &b3, &b4, vl); const vuint16m1_t ra = highbd_wiener_convolve5_8_2d_h_rvv( a0, a1, a2, a3, a4, xf1, xf2, xf3, round_vec, horiz_shift, im_max_val, vl); const vuint16m1_t rb = highbd_wiener_convolve5_8_2d_h_rvv( b0, b1, b2, b3, b4, xf1, xf2, xf3, round_vec, horiz_shift, im_max_val, vl); __riscv_vse16_v_u16m1(da, ra, vl); __riscv_vse16_v_u16m1(db, rb, vl); sa += vl; sb += vl; da += vl; db += vl; width -= (int)vl; } src_ptr += 2 * src_stride; dst_ptr += 2 * dst_stride; } for (; row < h; row++) { const int16_t *s = (const int16_t *)src_ptr; uint16_t *d = dst_ptr; int width = w; while (width > 0) { const size_t vl = __riscv_vsetvl_e16m1((size_t)width); vint16m1_t s0, s1, s2, s3, s4; load_s16_8x5(s, 1, &s0, &s1, &s2, &s3, &s4, vl); const vuint16m1_t d0 = highbd_wiener_convolve5_8_2d_h_rvv( s0, s1, s2, s3, s4, xf1, xf2, xf3, round_vec, horiz_shift, im_max_val, vl); __riscv_vse16_v_u16m1(d, d0, vl); s += vl; d += vl; width -= (int)vl; } src_ptr += src_stride; dst_ptr += dst_stride; } } static void highbd_convolve_add_src_horiz_7tap_rvv( const uint16_t *src_ptr, ptrdiff_t src_stride, uint16_t *dst_ptr, ptrdiff_t dst_stride, int w, int h, const int16_t *x_filter, int32_t round_val, int horiz_shift, uint16_t im_max_val) { const size_t vl_max = __riscv_vsetvlmax_e16m1(); const vint16m1_t xf0 = __riscv_vmv_v_x_i16m1(x_filter[0], vl_max); const vint16m1_t xf1 = __riscv_vmv_v_x_i16m1(x_filter[1], vl_max); const vint16m1_t xf2 = __riscv_vmv_v_x_i16m1(x_filter[2], vl_max); const vint16m1_t xf3 = __riscv_vmv_v_x_i16m1(x_filter[3], vl_max); const vint32m2_t round_vec = __riscv_vmv_v_x_i32m2(round_val, vl_max); int row = 0; for (; row <= h - 2; row += 2) { const int16_t *sa = (const int16_t *)src_ptr; const int16_t *sb = (const int16_t *)(src_ptr + src_stride); uint16_t *da = dst_ptr; uint16_t *db = dst_ptr + dst_stride; int width = w; while (width > 0) { const size_t vl = __riscv_vsetvl_e16m1((size_t)width); vint16m1_t a0, a1, a2, a3, a4, a5, a6; load_s16_8x7(sa, 1, &a0, &a1, &a2, &a3, &a4, &a5, &a6, vl); vint16m1_t b0, b1, b2, b3, b4, b5, b6; load_s16_8x7(sb, 1, &b0, &b1, &b2, &b3, &b4, &b5, &b6, vl); const vuint16m1_t ra = highbd_wiener_convolve7_8_2d_h_rvv( a0, a1, a2, a3, a4, a5, a6, xf0, xf1, xf2, xf3, round_vec, horiz_shift, im_max_val, vl); const vuint16m1_t rb = highbd_wiener_convolve7_8_2d_h_rvv( b0, b1, b2, b3, b4, b5, b6, xf0, xf1, xf2, xf3, round_vec, horiz_shift, im_max_val, vl); __riscv_vse16_v_u16m1(da, ra, vl); __riscv_vse16_v_u16m1(db, rb, vl); sa += vl; sb += vl; da += vl; db += vl; width -= (int)vl; } src_ptr += 2 * src_stride; dst_ptr += 2 * dst_stride; } for (; row < h; row++) { const int16_t *s = (const int16_t *)src_ptr; uint16_t *d = dst_ptr; int width = w; while (width > 0) { const size_t vl = __riscv_vsetvl_e16m1((size_t)width); vint16m1_t s0, s1, s2, s3, s4, s5, s6; load_s16_8x7(s, 1, &s0, &s1, &s2, &s3, &s4, &s5, &s6, vl); const vuint16m1_t d0 = highbd_wiener_convolve7_8_2d_h_rvv( s0, s1, s2, s3, s4, s5, s6, xf0, xf1, xf2, xf3, round_vec, horiz_shift, im_max_val, vl); __riscv_vse16_v_u16m1(d, d0, vl); s += vl; d += vl; width -= (int)vl; } src_ptr += src_stride; dst_ptr += dst_stride; } } static void highbd_convolve_add_src_vert_5tap_rvv( const uint16_t *src_ptr, ptrdiff_t src_stride, uint16_t *dst_ptr, ptrdiff_t dst_stride, int w, int h, const int16_t *y_filter, int32_t round_val, int vert_shift, uint16_t res_max_val) { const size_t vl_max = __riscv_vsetvlmax_e16m1(); const vint16m1_t yf1 = __riscv_vmv_v_x_i16m1(y_filter[1], vl_max); const vint16m1_t yf2 = __riscv_vmv_v_x_i16m1(y_filter[2], vl_max); const vint16m1_t yf3 = __riscv_vmv_v_x_i16m1(y_filter[3], vl_max); const vint32m2_t round_vec = __riscv_vmv_v_x_i32m2(round_val, vl_max); int width = w; const int16_t *s_base = (const int16_t *)src_ptr; uint16_t *d_base = dst_ptr; while (width > 0) { const size_t vl = __riscv_vsetvl_e16m1((size_t)width); const int16_t *s = s_base; uint16_t *d = d_base; int height = h; while (height > 3) { vint16m1_t s0, s1, s2, s3, s4, s5, s6, s7; load_s16_8x8(s, (int)src_stride, &s0, &s1, &s2, &s3, &s4, &s5, &s6, &s7, vl); vuint16m1_t d0 = highbd_wiener_convolve5_8_2d_v_rvv( s0, s1, s2, s3, s4, yf1, yf2, yf3, round_vec, vert_shift, res_max_val, vl); vuint16m1_t d1 = highbd_wiener_convolve5_8_2d_v_rvv( s1, s2, s3, s4, s5, yf1, yf2, yf3, round_vec, vert_shift, res_max_val, vl); vuint16m1_t d2 = highbd_wiener_convolve5_8_2d_v_rvv( s2, s3, s4, s5, s6, yf1, yf2, yf3, round_vec, vert_shift, res_max_val, vl); vuint16m1_t d3 = highbd_wiener_convolve5_8_2d_v_rvv( s3, s4, s5, s6, s7, yf1, yf2, yf3, round_vec, vert_shift, res_max_val, vl); store_u16_8x4(d, (int)dst_stride, d0, d1, d2, d3, vl); s += 4 * src_stride; d += 4 * dst_stride; height -= 4; } while (height-- != 0) { vint16m1_t s0, s1, s2, s3, s4; load_s16_8x5(s, (int)src_stride, &s0, &s1, &s2, &s3, &s4, vl); vuint16m1_t d0 = highbd_wiener_convolve5_8_2d_v_rvv( s0, s1, s2, s3, s4, yf1, yf2, yf3, round_vec, vert_shift, res_max_val, vl); __riscv_vse16_v_u16m1(d, d0, vl); s += src_stride; d += dst_stride; } s_base += vl; d_base += vl; width -= (int)vl; } } static void highbd_convolve_add_src_vert_7tap_rvv( const uint16_t *src_ptr, ptrdiff_t src_stride, uint16_t *dst_ptr, ptrdiff_t dst_stride, int w, int h, const int16_t *y_filter, int32_t round_val, int vert_shift, uint16_t res_max_val) { const size_t vl_max = __riscv_vsetvlmax_e16m1(); const vint16m1_t yf0 = __riscv_vmv_v_x_i16m1(y_filter[0], vl_max); const vint16m1_t yf1 = __riscv_vmv_v_x_i16m1(y_filter[1], vl_max); const vint16m1_t yf2 = __riscv_vmv_v_x_i16m1(y_filter[2], vl_max); const vint16m1_t yf3 = __riscv_vmv_v_x_i16m1(y_filter[3], vl_max); const vint32m2_t round_vec = __riscv_vmv_v_x_i32m2(round_val, vl_max); int width = w; const int16_t *s_base = (const int16_t *)src_ptr; uint16_t *d_base = dst_ptr; while (width > 0) { const size_t vl = __riscv_vsetvl_e16m1((size_t)width); const int16_t *s = s_base; uint16_t *d = d_base; int height = h; while (height > 3) { vint16m1_t s0, s1, s2, s3, s4, s5, s6, s7, s8, s9; load_s16_8x10(s, (int)src_stride, &s0, &s1, &s2, &s3, &s4, &s5, &s6, &s7, &s8, &s9, vl); vuint16m1_t d0 = highbd_wiener_convolve7_8_2d_v_rvv( s0, s1, s2, s3, s4, s5, s6, yf0, yf1, yf2, yf3, round_vec, vert_shift, res_max_val, vl); vuint16m1_t d1 = highbd_wiener_convolve7_8_2d_v_rvv( s1, s2, s3, s4, s5, s6, s7, yf0, yf1, yf2, yf3, round_vec, vert_shift, res_max_val, vl); vuint16m1_t d2 = highbd_wiener_convolve7_8_2d_v_rvv( s2, s3, s4, s5, s6, s7, s8, yf0, yf1, yf2, yf3, round_vec, vert_shift, res_max_val, vl); vuint16m1_t d3 = highbd_wiener_convolve7_8_2d_v_rvv( s3, s4, s5, s6, s7, s8, s9, yf0, yf1, yf2, yf3, round_vec, vert_shift, res_max_val, vl); store_u16_8x4(d, (int)dst_stride, d0, d1, d2, d3, vl); s += 4 * src_stride; d += 4 * dst_stride; height -= 4; } while (height-- != 0) { vint16m1_t s0, s1, s2, s3, s4, s5, s6; load_s16_8x7(s, (int)src_stride, &s0, &s1, &s2, &s3, &s4, &s5, &s6, vl); vuint16m1_t d0 = highbd_wiener_convolve7_8_2d_v_rvv( s0, s1, s2, s3, s4, s5, s6, yf0, yf1, yf2, yf3, round_vec, vert_shift, res_max_val, vl); __riscv_vse16_v_u16m1(d, d0, vl); s += src_stride; d += dst_stride; } s_base += vl; d_base += vl; width -= (int)vl; } } static inline int highbd_get_wiener_filter_taps(const int16_t *filter) { assert(filter[7] == 0); if (filter[0] == 0 && filter[6] == 0) return WIENER_WIN_REDUCED; return WIENER_WIN; } void av1_highbd_wiener_convolve_add_src_rvv( const uint8_t *src8, ptrdiff_t src_stride, uint8_t *dst8, ptrdiff_t dst_stride, const int16_t *x_filter, int x_step_q4, const int16_t *y_filter, int y_step_q4, int w, int h, const WienerConvolveParams *conv_params, int bd) { (void)x_step_q4; (void)y_step_q4; assert(w % 8 == 0); assert(w <= MAX_SB_SIZE && h <= MAX_SB_SIZE); assert(x_step_q4 == 16 && y_step_q4 == 16); assert(x_filter[7] == 0 && y_filter[7] == 0); DECLARE_ALIGNED(16, uint16_t, im_block[(MAX_SB_SIZE + WIENER_WIN - 1) * MAX_SB_SIZE]); const int x_filter_taps = highbd_get_wiener_filter_taps(x_filter); const int y_filter_taps = highbd_get_wiener_filter_taps(y_filter); const int horiz_shift = conv_params->round_0; const int vert_shift = conv_params->round_1; const int im_stride = MAX_SB_SIZE; const int im_h = h + y_filter_taps - 1; const int horiz_offset = x_filter_taps / 2; const int vert_offset = (y_filter_taps / 2) * (int)src_stride; const int extraprec_clamp_limit = WIENER_CLAMP_LIMIT(conv_params->round_0, bd); const uint16_t im_max_val = (uint16_t)(extraprec_clamp_limit - 1); const int32_t horiz_round_val = 1 << (bd + FILTER_BITS - 1); const uint16_t res_max_val = (uint16_t)((1 << bd) - 1); const int32_t vert_round_val = -(1 << (bd + conv_params->round_1 - 1)); uint16_t *src = CONVERT_TO_SHORTPTR(src8); uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); // Filter taps are symmetric; add 128 to center tap (index 3) for rounding // bias, matching the NEON implementation. int16_t xf[4] = { x_filter[0], x_filter[1], x_filter[2], (int16_t)(x_filter[3] + 128) }; int16_t yf[4] = { y_filter[0], y_filter[1], y_filter[2], (int16_t)(y_filter[3] + 128) }; const uint16_t *hsrc = src - horiz_offset - vert_offset; if (x_filter_taps == WIENER_WIN_REDUCED) { highbd_convolve_add_src_horiz_5tap_rvv( hsrc, src_stride, im_block, im_stride, w, im_h, xf, horiz_round_val, horiz_shift, im_max_val); } else { highbd_convolve_add_src_horiz_7tap_rvv( hsrc, src_stride, im_block, im_stride, w, im_h, xf, horiz_round_val, horiz_shift, im_max_val); } if (y_filter_taps == WIENER_WIN_REDUCED) { highbd_convolve_add_src_vert_5tap_rvv(im_block, im_stride, dst, dst_stride, w, h, yf, vert_round_val, vert_shift, res_max_val); } else { highbd_convolve_add_src_vert_7tap_rvv(im_block, im_stride, dst, dst_stride, w, h, yf, vert_round_val, vert_shift, res_max_val); } }