From 2e79813a8ec7998ae4e86688e929a5ca9a131ae7 Mon Sep 17 00:00:00 2001 From: Kaustubh Raste Date: Mon, 4 Sep 2017 18:01:32 +0530 Subject: [PATCH] avcodec/mips: Improve vp9 lpf msa functions Updated VP9_LPF_FILTER4_4W macro to process on 8 bit data. Replaced VP9_LPF_FILTER4_8W with VP9_LPF_FILTER4_4W. Signed-off-by: Kaustubh Raste Reviewed-by: Manojkumar Bhosale Signed-off-by: Michael Niedermayer --- libavcodec/mips/vp9_lpf_msa.c | 94 ++++++----------------------------- 1 file changed, 14 insertions(+), 80 deletions(-) diff --git a/libavcodec/mips/vp9_lpf_msa.c b/libavcodec/mips/vp9_lpf_msa.c index eef8afc482..c82a9e98bc 100644 --- a/libavcodec/mips/vp9_lpf_msa.c +++ b/libavcodec/mips/vp9_lpf_msa.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2015 Shivraj Patil (Shivraj.Patil@imgtec.com) + * Copyright (c) 2015 - 2017 Shivraj Patil (Shivraj.Patil@imgtec.com) * * This file is part of FFmpeg. * @@ -22,63 +22,12 @@ #include "libavutil/mips/generic_macros_msa.h" #include "vp9dsp_mips.h" -#define VP9_LPF_FILTER4_8W(p1_in, p0_in, q0_in, q1_in, mask_in, hev_in, \ - p1_out, p0_out, q0_out, q1_out) \ -{ \ - v16i8 p1_m, p0_m, q0_m, q1_m, q0_sub_p0, filt_sign; \ - v16i8 filt, filt1, filt2, cnst4b, cnst3b; \ - v8i16 q0_sub_p0_r, filt_r, cnst3h; \ - \ - p1_m = (v16i8) __msa_xori_b(p1_in, 0x80); \ - p0_m = (v16i8) __msa_xori_b(p0_in, 0x80); \ - q0_m = (v16i8) __msa_xori_b(q0_in, 0x80); \ - q1_m = (v16i8) __msa_xori_b(q1_in, 0x80); \ - \ - filt = __msa_subs_s_b(p1_m, q1_m); \ - filt = filt & (v16i8) hev_in; \ - q0_sub_p0 = q0_m - p0_m; \ - filt_sign = __msa_clti_s_b(filt, 0); \ - \ - cnst3h = __msa_ldi_h(3); \ - q0_sub_p0_r = (v8i16) __msa_ilvr_b(q0_sub_p0, q0_sub_p0); \ - q0_sub_p0_r = __msa_dotp_s_h((v16i8) q0_sub_p0_r, (v16i8) cnst3h); \ - filt_r = (v8i16) __msa_ilvr_b(filt_sign, filt); \ - filt_r += q0_sub_p0_r; \ - filt_r = __msa_sat_s_h(filt_r, 7); \ - \ - /* combine left and right part */ \ - filt = __msa_pckev_b((v16i8) filt_r, (v16i8) filt_r); \ - \ - filt = filt & (v16i8) mask_in; \ - cnst4b = __msa_ldi_b(4); \ - filt1 = __msa_adds_s_b(filt, cnst4b); \ - filt1 >>= 3; \ - \ - cnst3b = __msa_ldi_b(3); \ - filt2 = __msa_adds_s_b(filt, cnst3b); \ - filt2 >>= 3; \ - \ - q0_m = __msa_subs_s_b(q0_m, filt1); \ - q0_out = __msa_xori_b((v16u8) q0_m, 0x80); \ - p0_m = __msa_adds_s_b(p0_m, filt2); \ - p0_out = __msa_xori_b((v16u8) p0_m, 0x80); \ - \ - filt = __msa_srari_b(filt1, 1); \ - hev_in = __msa_xori_b((v16u8) hev_in, 0xff); \ - filt = filt & (v16i8) hev_in; \ - \ - q1_m = __msa_subs_s_b(q1_m, filt); \ - q1_out = __msa_xori_b((v16u8) q1_m, 0x80); \ - p1_m = __msa_adds_s_b(p1_m, filt); \ - p1_out = __msa_xori_b((v16u8) p1_m, 0x80); \ -} - #define VP9_LPF_FILTER4_4W(p1_in, p0_in, q0_in, q1_in, mask_in, hev_in, \ p1_out, p0_out, q0_out, q1_out) \ { \ - v16i8 p1_m, p0_m, q0_m, q1_m, q0_sub_p0, filt_sign; \ - v16i8 filt, filt1, filt2, cnst4b, cnst3b; \ - v8i16 q0_sub_p0_r, q0_sub_p0_l, filt_l, filt_r, cnst3h; \ + v16i8 p1_m, p0_m, q0_m, q1_m, q0_sub_p0, filt, filt1, filt2; \ + const v16i8 cnst4b = __msa_ldi_b(4); \ + const v16i8 cnst3b = __msa_ldi_b(3); \ \ p1_m = (v16i8) __msa_xori_b(p1_in, 0x80); \ p0_m = (v16i8) __msa_xori_b(p0_in, 0x80); \ @@ -89,30 +38,15 @@ \ filt = filt & (v16i8) hev_in; \ \ - q0_sub_p0 = q0_m - p0_m; \ - filt_sign = __msa_clti_s_b(filt, 0); \ - \ - cnst3h = __msa_ldi_h(3); \ - q0_sub_p0_r = (v8i16) __msa_ilvr_b(q0_sub_p0, q0_sub_p0); \ - q0_sub_p0_r = __msa_dotp_s_h((v16i8) q0_sub_p0_r, (v16i8) cnst3h); \ - filt_r = (v8i16) __msa_ilvr_b(filt_sign, filt); \ - filt_r += q0_sub_p0_r; \ - filt_r = __msa_sat_s_h(filt_r, 7); \ - \ - q0_sub_p0_l = (v8i16) __msa_ilvl_b(q0_sub_p0, q0_sub_p0); \ - q0_sub_p0_l = __msa_dotp_s_h((v16i8) q0_sub_p0_l, (v16i8) cnst3h); \ - filt_l = (v8i16) __msa_ilvl_b(filt_sign, filt); \ - filt_l += q0_sub_p0_l; \ - filt_l = __msa_sat_s_h(filt_l, 7); \ - \ - filt = __msa_pckev_b((v16i8) filt_l, (v16i8) filt_r); \ + q0_sub_p0 = __msa_subs_s_b(q0_m, p0_m); \ + filt = __msa_adds_s_b(filt, q0_sub_p0); \ + filt = __msa_adds_s_b(filt, q0_sub_p0); \ + filt = __msa_adds_s_b(filt, q0_sub_p0); \ filt = filt & (v16i8) mask_in; \ \ - cnst4b = __msa_ldi_b(4); \ filt1 = __msa_adds_s_b(filt, cnst4b); \ filt1 >>= 3; \ \ - cnst3b = __msa_ldi_b(3); \ filt2 = __msa_adds_s_b(filt, cnst3b); \ filt2 >>= 3; \ \ @@ -277,7 +211,7 @@ void ff_loop_filter_v_4_8_msa(uint8_t *src, ptrdiff_t pitch, LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev, mask, flat); - VP9_LPF_FILTER4_8W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, + VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out); p1_d = __msa_copy_u_d((v2i64) p1_out, 0); @@ -342,7 +276,7 @@ void ff_loop_filter_v_8_8_msa(uint8_t *src, ptrdiff_t pitch, LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev, mask, flat); VP9_FLAT4(p3, p2, p0, q0, q2, q3, flat); - VP9_LPF_FILTER4_8W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, + VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out); flat = (v16u8) __msa_ilvr_d((v2i64) zero, (v2i64) flat); @@ -1065,7 +999,7 @@ void ff_loop_filter_v_16_8_msa(uint8_t *src, ptrdiff_t pitch, LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev, mask, flat); VP9_FLAT4(p3, p2, p0, q0, q2, q3, flat); - VP9_LPF_FILTER4_8W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, + VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out); flat = (v16u8) __msa_ilvr_d((v2i64) zero, (v2i64) flat); @@ -1280,7 +1214,7 @@ void ff_loop_filter_h_4_8_msa(uint8_t *src, ptrdiff_t pitch, p3, p2, p1, p0, q0, q1, q2, q3); LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev, mask, flat); - VP9_LPF_FILTER4_8W(p1, p0, q0, q1, mask, hev, p1, p0, q0, q1); + VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1, p0, q0, q1); ILVR_B2_SH(p0, p1, q1, q0, vec0, vec1); ILVRL_H2_SH(vec1, vec0, vec2, vec3); @@ -1367,7 +1301,7 @@ void ff_loop_filter_h_8_8_msa(uint8_t *src, ptrdiff_t pitch, /* flat4 */ VP9_FLAT4(p3, p2, p0, q0, q2, q3, flat); /* filter4 */ - VP9_LPF_FILTER4_8W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, + VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out); flat = (v16u8) __msa_ilvr_d((v2i64) zero, (v2i64) flat); @@ -1868,7 +1802,7 @@ static int32_t vp9_vt_lpf_t4_and_t8_8w(uint8_t *src, uint8_t *filter48, /* flat4 */ VP9_FLAT4(p3, p2, p0, q0, q2, q3, flat); /* filter4 */ - VP9_LPF_FILTER4_8W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, + VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out); flat = (v16u8) __msa_ilvr_d((v2i64) zero, (v2i64) flat);