diff options
-rw-r--r-- | libavcodec/mips/Makefile | 3 | ||||
-rw-r--r-- | libavcodec/mips/hevc_macros_msa.h | 30 | ||||
-rw-r--r-- | libavcodec/mips/hevc_mc_bi_msa.c | 4462 | ||||
-rw-r--r-- | libavcodec/mips/hevcdsp_init_mips.c | 68 | ||||
-rw-r--r-- | libavcodec/mips/hevcdsp_mips.h | 80 | ||||
-rw-r--r-- | libavutil/mips/generic_macros_msa.h | 150 |
6 files changed, 4792 insertions, 1 deletions
diff --git a/libavcodec/mips/Makefile b/libavcodec/mips/Makefile index 9786175ad4..610ff1ae46 100644 --- a/libavcodec/mips/Makefile +++ b/libavcodec/mips/Makefile @@ -21,6 +21,7 @@ MIPSFPU-OBJS-$(CONFIG_AAC_ENCODER) += mips/iirfilter_mips.o OBJS-$(CONFIG_HEVC_DECODER) += mips/hevcdsp_init_mips.o OBJS-$(CONFIG_H264DSP) += mips/h264dsp_init_mips.o MSA-OBJS-$(CONFIG_HEVC_DECODER) += mips/hevcdsp_msa.o \ - mips/hevc_mc_uni_msa.o + mips/hevc_mc_uni_msa.o \ + mips/hevc_mc_bi_msa.o MSA-OBJS-$(CONFIG_H264DSP) += mips/h264dsp_msa.o LOONGSON3-OBJS-$(CONFIG_H264DSP) += mips/h264dsp_mmi.o diff --git a/libavcodec/mips/hevc_macros_msa.h b/libavcodec/mips/hevc_macros_msa.h index 7a89f93a9c..b06c5ad9b9 100644 --- a/libavcodec/mips/hevc_macros_msa.h +++ b/libavcodec/mips/hevc_macros_msa.h @@ -37,6 +37,27 @@ out = (v4i32) __msa_pckev_b((v16i8) tmp1_m, (v16i8) tmp0_m); \ } +#define HEVC_PCK_SW_SB8(in0, in1, in2, in3, in4, in5, in6, in7, out0, out1) \ +{ \ + v8i16 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \ + \ + PCKEV_H4_SH(in0, in1, in2, in3, in4, in5, in6, in7, \ + tmp0_m, tmp1_m, tmp2_m, tmp3_m); \ + PCKEV_B2_SW(tmp1_m, tmp0_m, tmp3_m, tmp2_m, out0, out1); \ +} + +#define HEVC_PCK_SW_SB12(in0, in1, in2, in3, in4, in5, in6, in7, \ + in8, in9, in10, in11, out0, out1, out2) \ +{ \ + v8i16 tmp0_m, tmp1_m, tmp2_m, tmp3_m, tmp4_m, tmp5_m; \ + \ + PCKEV_H4_SH(in0, in1, in2, in3, in4, in5, in6, in7, \ + tmp0_m, tmp1_m, tmp2_m, tmp3_m); \ + PCKEV_H2_SH(in8, in9, in10, in11, tmp4_m, tmp5_m); \ + PCKEV_B2_SW(tmp1_m, tmp0_m, tmp3_m, tmp2_m, out0, out1); \ + out2 = (v4i32) __msa_pckev_b((v16i8) tmp5_m, (v16i8) tmp4_m); \ +} + #define HEVC_FILT_8TAP(in0, in1, in2, in3, \ filt0, filt1, filt2, filt3) \ ( { \ @@ -48,4 +69,13 @@ out_m; \ } ) +#define HEVC_FILT_4TAP(in0, in1, filt0, filt1) \ +( { \ + v4i32 out_m; \ + \ + out_m = __msa_dotp_s_w(in0, (v8i16) filt0); \ + out_m = __msa_dpadd_s_w(out_m, in1, (v8i16) filt1); \ + out_m; \ +} ) + #endif /* AVCODEC_MIPS_HEVC_MACROS_MSA_H */ diff --git a/libavcodec/mips/hevc_mc_bi_msa.c b/libavcodec/mips/hevc_mc_bi_msa.c new file mode 100644 index 0000000000..0709b40ccf --- /dev/null +++ b/libavcodec/mips/hevc_mc_bi_msa.c @@ -0,0 +1,4462 @@ +/* + * Copyright (c) 2015 Manojkumar Bhosale (Manojkumar.Bhosale@imgtec.com) + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "libavutil/mips/generic_macros_msa.h" +#include "libavcodec/mips/hevcdsp_mips.h" +#include "libavcodec/mips/hevc_macros_msa.h" + +#define HEVC_BI_RND_CLIP2(in0, in1, vec0, vec1, rnd_val, out0, out1) \ +{ \ + ADDS_SH2_SH(vec0, in0, vec1, in1, out0, out1); \ + SRARI_H2_SH(out0, out1, rnd_val); \ + CLIP_SH2_0_255(out0, out1); \ +} + +#define HEVC_BI_RND_CLIP4(in0, in1, in2, in3, \ + vec0, vec1, vec2, vec3, rnd_val, \ + out0, out1, out2, out3) \ +{ \ + HEVC_BI_RND_CLIP2(in0, in1, vec0, vec1, rnd_val, out0, out1); \ + HEVC_BI_RND_CLIP2(in2, in3, vec2, vec3, rnd_val, out2, out3); \ +} + +static void hevc_bi_copy_4w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + int32_t height) +{ + v16i8 zero = { 0 }; + + if (2 == height) { + v16i8 src0, src1; + v8i16 dst0, in0, in1; + + LD_SB2(src0_ptr, src_stride, src0, src1); + LD_SH2(src1_ptr, src2_stride, in0, in1); + + src0 = (v16i8) __msa_ilvr_w((v4i32) src1, (v4i32) src0); + in0 = (v8i16) __msa_ilvr_d((v2i64) in1, (v2i64) in0); + + dst0 = (v8i16) __msa_ilvr_b(zero, src0); + dst0 <<= 6; + dst0 += in0; + dst0 = __msa_srari_h(dst0, 7); + dst0 = CLIP_SH_0_255(dst0); + + dst0 = (v8i16) __msa_pckev_b((v16i8) dst0, (v16i8) dst0); + ST4x2_UB(dst0, dst, dst_stride); + } else if (4 == height) { + v16i8 src0, src1, src2, src3; + v8i16 dst0, dst1; + v8i16 in0, in1, in2, in3; + + LD_SB4(src0_ptr, src_stride, src0, src1, src2, src3); + LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3); + ILVR_W2_SB(src1, src0, src3, src2, src0, src1); + ILVR_D2_SH(in1, in0, in3, in2, in0, in1); + ILVR_B2_SH(zero, src0, zero, src1, dst0, dst1); + + dst0 <<= 6; + dst1 <<= 6; + HEVC_BI_RND_CLIP2(in0, in1, dst0, dst1, 7, dst0, dst1); + + dst0 = (v8i16) __msa_pckev_b((v16i8) dst1, (v16i8) dst0); + ST4x4_UB(dst0, dst0, 0, 1, 2, 3, dst, dst_stride); + } else if (0 == height % 8) { + uint32_t loop_cnt; + v16i8 src0, src1, src2, src3, src4, src5, src6, src7; + v8i16 dst0, dst1, dst2, dst3; + v8i16 in0, in1, in2, in3, in4, in5, in6, in7; + + for (loop_cnt = (height >> 3); loop_cnt--;) { + LD_SB8(src0_ptr, src_stride, + src0, src1, src2, src3, src4, src5, src6, src7); + src0_ptr += (8 * src_stride); + + LD_SH8(src1_ptr, src2_stride, + in0, in1, in2, in3, in4, in5, in6, in7); + src1_ptr += (8 * src2_stride); + + ILVR_D2_SH(in1, in0, in3, in2, in0, in1); + ILVR_D2_SH(in5, in4, in7, in6, in2, in3); + + ILVR_W4_SB(src1, src0, src3, src2, src5, src4, src7, src6, + src0, src1, src2, src3); + ILVR_B4_SH(zero, src0, zero, src1, zero, src2, zero, src3, + dst0, dst1, dst2, dst3); + + SLLI_4V(dst0, dst1, dst2, dst3, 6); + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0, dst1, dst2, dst3, 7, + dst0, dst1, dst2, dst3); + + PCKEV_B2_SH(dst1, dst0, dst3, dst2, dst0, dst1); + ST4x8_UB(dst0, dst1, dst, dst_stride); + dst += (8 * dst_stride); + } + } +} + +static void hevc_bi_copy_6w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + int32_t height) +{ + uint32_t loop_cnt; + v16i8 zero = { 0 }; + v16i8 src0, src1, src2, src3, src4, src5, src6, src7; + v8i16 in0, in1, in2, in3, in4, in5, in6, in7; + v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7; + + for (loop_cnt = (height >> 3); loop_cnt--;) { + LD_SB8(src0_ptr, src_stride, + src0, src1, src2, src3, src4, src5, src6, src7); + src0_ptr += (8 * src_stride); + LD_SH8(src1_ptr, src2_stride, in0, in1, in2, in3, in4, in5, in6, in7); + src1_ptr += (8 * src2_stride); + ILVR_B4_SH(zero, src0, zero, src1, zero, src2, zero, src3, + dst0, dst1, dst2, dst3); + ILVR_B4_SH(zero, src4, zero, src5, zero, src6, zero, src7, + dst4, dst5, dst6, dst7); + + SLLI_4V(dst0, dst1, dst2, dst3, 6); + SLLI_4V(dst4, dst5, dst6, dst7, 6); + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3); + + PCKEV_B2_SH(dst1, dst0, dst3, dst2, dst0, dst1); + ST6x4_UB(dst0, dst1, dst, dst_stride); + dst += (4 * dst_stride); + + HEVC_BI_RND_CLIP4(in4, in5, in6, in7, + dst4, dst5, dst6, dst7, 7, dst4, dst5, dst6, dst7); + + PCKEV_B2_SH(dst5, dst4, dst7, dst6, dst4, dst5); + ST6x4_UB(dst4, dst5, dst, dst_stride); + dst += (4 * dst_stride); + } +} + +static void hevc_bi_copy_8w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + int32_t height) +{ + v16i8 zero = { 0 }; + + if (2 == height) { + v16i8 src0, src1; + v8i16 in0, in1; + v8i16 dst0, dst1; + + LD_SB2(src0_ptr, src_stride, src0, src1); + LD_SH2(src1_ptr, src2_stride, in0, in1); + ILVR_B2_SH(zero, src0, zero, src1, dst0, dst1); + + dst0 <<= 6; + dst1 <<= 6; + HEVC_BI_RND_CLIP2(in0, in1, dst0, dst1, 7, dst0, dst1); + + dst0 = (v8i16) __msa_pckev_b((v16i8) dst1, (v16i8) dst0); + ST8x2_UB(dst0, dst, dst_stride); + } else if (4 == height) { + v16i8 src0, src1, src2, src3; + v8i16 in0, in1, in2, in3; + v8i16 dst0, dst1, dst2, dst3; + + LD_SB4(src0_ptr, src_stride, src0, src1, src2, src3); + LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3); + ILVR_B4_SH(zero, src0, zero, src1, zero, src2, zero, src3, + dst0, dst1, dst2, dst3); + + SLLI_4V(dst0, dst1, dst2, dst3, 6); + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3); + + PCKEV_B2_SH(dst1, dst0, dst3, dst2, dst0, dst1); + ST8x4_UB(dst0, dst1, dst, dst_stride); + } else if (6 == height) { + v16i8 src0, src1, src2, src3, src4, src5; + v8i16 in0, in1, in2, in3, in4, in5; + v8i16 dst0, dst1, dst2, dst3, dst4, dst5; + + LD_SB6(src0_ptr, src_stride, src0, src1, src2, src3, src4, src5); + LD_SH6(src1_ptr, src2_stride, in0, in1, in2, in3, in4, in5); + ILVR_B4_SH(zero, src0, zero, src1, zero, src2, zero, src3, + dst0, dst1, dst2, dst3); + ILVR_B2_SH(zero, src4, zero, src5, dst4, dst5); + + SLLI_4V(dst0, dst1, dst2, dst3, 6); + dst4 <<= 6; + dst5 <<= 6; + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3); + HEVC_BI_RND_CLIP2(in4, in5, dst4, dst5, 7, dst4, dst5); + + PCKEV_B2_SH(dst1, dst0, dst3, dst2, dst0, dst1); + dst2 = (v8i16) __msa_pckev_b((v16i8) dst5, (v16i8) dst4); + ST8x4_UB(dst0, dst1, dst, dst_stride); + dst += (4 * dst_stride); + ST8x2_UB(dst2, dst, dst_stride); + } else if (0 == height % 8) { + v16i8 src0, src1, src2, src3; + v8i16 in0, in1, in2, in3; + v8i16 dst0, dst1, dst2, dst3; + uint32_t loop_cnt; + + for (loop_cnt = (height >> 3); loop_cnt--;) { + LD_SB4(src0_ptr, src_stride, src0, src1, src2, src3); + src0_ptr += (4 * src_stride); + LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3); + src1_ptr += (4 * src2_stride); + ILVR_B4_SH(zero, src0, zero, src1, zero, src2, zero, src3, + dst0, dst1, dst2, dst3); + + SLLI_4V(dst0, dst1, dst2, dst3, 6); + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0, dst1, dst2, dst3, 7, + dst0, dst1, dst2, dst3); + + PCKEV_B2_SH(dst1, dst0, dst3, dst2, dst0, dst1); + ST8x4_UB(dst0, dst1, dst, dst_stride); + dst += (4 * dst_stride); + + LD_SB4(src0_ptr, src_stride, src0, src1, src2, src3); + src0_ptr += (4 * src_stride); + LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3); + src1_ptr += (4 * src2_stride); + ILVR_B4_SH(zero, src0, zero, src1, zero, src2, zero, src3, + dst0, dst1, dst2, dst3); + + SLLI_4V(dst0, dst1, dst2, dst3, 6); + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0, dst1, dst2, dst3, 7, + dst0, dst1, dst2, dst3); + + PCKEV_B2_SH(dst1, dst0, dst3, dst2, dst0, dst1); + ST8x4_UB(dst0, dst1, dst, dst_stride); + dst += (4 * dst_stride); + } + } +} + +static void hevc_bi_copy_12w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + int32_t height) +{ + uint32_t loop_cnt; + v16i8 src0, src1, src2, src3; + v8i16 in0, in1, in2, in3, in4, in5, in6, in7; + v8i16 dst0, dst1, dst2, dst3, dst4, dst5; + v16i8 zero = { 0 }; + + for (loop_cnt = (16 >> 2); loop_cnt--;) { + LD_SB4(src0_ptr, src_stride, src0, src1, src2, src3); + src0_ptr += (4 * src_stride); + + LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3); + LD_SH4(src1_ptr + 8, src2_stride, in4, in5, in6, in7); + src1_ptr += (4 * src2_stride); + ILVR_D2_SH(in5, in4, in7, in6, in4, in5); + ILVR_B4_SH(zero, src0, zero, src1, zero, src2, zero, src3, + dst0, dst1, dst2, dst3); + + SLLI_4V(dst0, dst1, dst2, dst3, 6); + ILVL_W2_SB(src1, src0, src3, src2, src0, src1); + ILVR_B2_SH(zero, src0, zero, src1, dst4, dst5); + dst4 <<= 6; + dst5 <<= 6; + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3); + HEVC_BI_RND_CLIP2(in4, in5, dst4, dst5, 7, dst4, dst5); + + PCKEV_B2_SH(dst1, dst0, dst3, dst2, dst0, dst1); + dst2 = (v8i16) __msa_pckev_b((v16i8) dst5, (v16i8) dst4); + ST12x4_UB(dst0, dst1, dst2, dst, dst_stride); + dst += (4 * dst_stride); + } +} + +static void hevc_bi_copy_16multx4mult_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + int32_t height, + int32_t width) +{ + uint32_t loop_cnt; + uint32_t cnt; + uint8_t *src0_ptr_tmp; + int16_t *src1_ptr_tmp; + uint8_t *dst_tmp; + v16i8 zero = { 0 }; + + for (cnt = (width >> 4); cnt--;) { + src0_ptr_tmp = src0_ptr; + src1_ptr_tmp = src1_ptr; + dst_tmp = dst; + + for (loop_cnt = (height >> 2); loop_cnt--;) { + v16i8 src0, src1, src2, src3; + v8i16 in0, in1, in2, in3, in4, in5, in6, in7; + v8i16 dst0_r, dst1_r, dst2_r, dst3_r; + v8i16 dst0_l, dst1_l, dst2_l, dst3_l; + + LD_SB4(src0_ptr_tmp, src_stride, src0, src1, src2, src3); + src0_ptr_tmp += (4 * src_stride); + LD_SH4(src1_ptr_tmp, src2_stride, in0, in1, in2, in3); + LD_SH4(src1_ptr_tmp + 8, src2_stride, in4, in5, in6, in7); + src1_ptr_tmp += (4 * src2_stride); + + ILVR_B4_SH(zero, src0, zero, src1, zero, src2, zero, src3, + dst0_r, dst1_r, dst2_r, dst3_r); + ILVL_B4_SH(zero, src0, zero, src1, zero, src2, zero, src3, + dst0_l, dst1_l, dst2_l, dst3_l); + + SLLI_4V(dst0_r, dst1_r, dst2_r, dst3_r, 6); + SLLI_4V(dst0_l, dst1_l, dst2_l, dst3_l, 6); + HEVC_BI_RND_CLIP4(in0, in1, in4, in5, + dst0_r, dst1_r, dst0_l, dst1_l, 7, + dst0_r, dst1_r, dst0_l, dst1_l); + + PCKEV_B2_SH(dst0_l, dst0_r, dst1_l, dst1_r, dst0_r, dst1_r); + ST_SH2(dst0_r, dst1_r, dst_tmp, dst_stride); + dst_tmp += (2 * dst_stride); + + HEVC_BI_RND_CLIP4(in2, in3, in6, in7, + dst2_r, dst3_r, dst2_l, dst3_l, 7, + dst2_r, dst3_r, dst2_l, dst3_l); + + PCKEV_B2_SH(dst2_l, dst2_r, dst3_l, dst3_r, dst2_r, dst3_r); + ST_SH2(dst2_r, dst3_r, dst_tmp, dst_stride); + dst_tmp += (2 * dst_stride); + } + + src0_ptr += 16; + src1_ptr += 16; + dst += 16; + } +} + +static void hevc_bi_copy_16w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + int32_t height) +{ + hevc_bi_copy_16multx4mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, height, 16); +} + +static void hevc_bi_copy_24w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + int32_t height) +{ + hevc_bi_copy_16multx4mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, height, 16); + + hevc_bi_copy_8w_msa(src0_ptr + 16, src_stride, src1_ptr + 16, src2_stride, + dst + 16, dst_stride, height); +} + +static void hevc_bi_copy_32w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + int32_t height) +{ + hevc_bi_copy_16multx4mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, height, 32); +} + +static void hevc_bi_copy_48w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + int32_t height) +{ + hevc_bi_copy_16multx4mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, height, 48); +} + +static void hevc_bi_copy_64w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + int32_t height) +{ + hevc_bi_copy_16multx4mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, height, 64); +} + +static void hevc_hz_bi_8t_4w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + uint32_t loop_cnt; + v8i16 filt0, filt1, filt2, filt3; + v16i8 src0, src1, src2, src3, src4, src5, src6, src7; + v16i8 mask1, mask2, mask3; + v16i8 vec0, vec1, vec2, vec3; + v8i16 dst0, dst1, dst2, dst3; + v8i16 in0, in1, in2, in3, in4, in5, in6, in7; + v8i16 filter_vec, const_vec; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20 }; + + src0_ptr -= 3; + + /* rearranging filter */ + filter_vec = LD_SH(filter); + SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3); + + mask1 = mask0 + 2; + mask2 = mask0 + 4; + mask3 = mask0 + 6; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + for (loop_cnt = (height >> 3); loop_cnt--;) { + LD_SB8(src0_ptr, src_stride, src0, src1, src2, src3, + src4, src5, src6, src7); + src0_ptr += (8 * src_stride); + LD_SH8(src1_ptr, src2_stride, in0, in1, in2, in3, in4, in5, in6, in7); + src1_ptr += (8 * src2_stride); + + ILVR_D2_SH(in1, in0, in3, in2, in0, in1); + ILVR_D2_SH(in5, in4, in7, in6, in2, in3); + XORI_B8_128_SB(src0, src1, src2, src3, src4, src5, src6, src7); + + VSHF_B4_SB(src0, src1, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst0 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst0, dst0, dst0, dst0); + VSHF_B4_SB(src2, src3, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst1 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst1, dst1, dst1, dst1); + VSHF_B4_SB(src4, src5, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst2 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst2, dst2, dst2, dst2); + VSHF_B4_SB(src6, src7, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst3 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst3, dst3, dst3, dst3); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3); + + PCKEV_B2_SH(dst1, dst0, dst3, dst2, dst0, dst1); + ST4x8_UB(dst0, dst1, dst, dst_stride); + dst += (8 * dst_stride); + } +} + +static void hevc_hz_bi_8t_8w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + uint32_t loop_cnt; + v8i16 filt0, filt1, filt2, filt3; + v16i8 src0, src1, src2, src3; + v16i8 mask1, mask2, mask3; + v16i8 vec0, vec1, vec2, vec3; + v8i16 dst0, dst1, dst2, dst3; + v8i16 in0, in1, in2, in3; + v8i16 filter_vec, const_vec; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + + src0_ptr -= 3; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3); + + mask1 = mask0 + 2; + mask2 = mask0 + 4; + mask3 = mask0 + 6; + + for (loop_cnt = (height >> 2); loop_cnt--;) { + LD_SB4(src0_ptr, src_stride, src0, src1, src2, src3); + src0_ptr += (4 * src_stride); + LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3); + src1_ptr += (4 * src2_stride); + XORI_B4_128_SB(src0, src1, src2, src3); + + VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst0 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst0, dst0, dst0, dst0); + VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst1 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst1, dst1, dst1, dst1); + VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst2 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst2, dst2, dst2, dst2); + VSHF_B4_SB(src3, src3, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst3 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst3, dst3, dst3, dst3); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3); + + PCKEV_B2_SH(dst1, dst0, dst3, dst2, dst0, dst1); + ST8x4_UB(dst0, dst1, dst, dst_stride); + dst += (4 * dst_stride); + } +} + +static void hevc_hz_bi_8t_12w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + hevc_hz_bi_8t_8w_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter, height); + hevc_hz_bi_8t_4w_msa(src0_ptr + 8, src_stride, src1_ptr + 8, src2_stride, + dst + 8, dst_stride, filter, height); +} + +static void hevc_hz_bi_8t_16w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + uint32_t loop_cnt; + v16i8 src0, src1, src2, src3; + v8i16 filt0, filt1, filt2, filt3; + v16i8 mask1, mask2, mask3; + v16i8 vec0, vec1, vec2, vec3; + v8i16 dst0, dst1, dst2, dst3; + v8i16 in0, in1, in2, in3; + v8i16 filter_vec, const_vec; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + + src0_ptr -= 3; + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3); + + mask1 = mask0 + 2; + mask2 = mask0 + 4; + mask3 = mask0 + 6; + + for (loop_cnt = (height >> 1); loop_cnt--;) { + LD_SB2(src0_ptr, 8, src0, src1); + src0_ptr += src_stride; + LD_SB2(src0_ptr, 8, src2, src3); + src0_ptr += src_stride; + LD_SH2(src1_ptr, 8, in0, in1); + src1_ptr += src2_stride; + LD_SH2(src1_ptr, 8, in2, in3); + src1_ptr += src2_stride; + XORI_B4_128_SB(src0, src1, src2, src3); + + VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst0 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst0, dst0, dst0, dst0); + VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst1 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst1, dst1, dst1, dst1); + VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst2 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst2, dst2, dst2, dst2); + VSHF_B4_SB(src3, src3, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst3 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst3, dst3, dst3, dst3); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3); + + PCKEV_B2_SH(dst1, dst0, dst3, dst2, dst0, dst1); + ST_SH2(dst0, dst1, dst, dst_stride); + dst += (2 * dst_stride); + } +} + +static void hevc_hz_bi_8t_24w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + uint32_t loop_cnt; + uint64_t dst_val0; + v16i8 src0, src1, tmp0, tmp1; + v8i16 filt0, filt1, filt2, filt3; + v16i8 mask1, mask2, mask3, mask4, mask5, mask6, mask7; + v16i8 vec0, vec1, vec2, vec3; + v8i16 dst0, dst1, dst2; + v8i16 in0, in1, in2; + v8i16 filter_vec, const_vec; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + + src0_ptr = src0_ptr - 3; + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3); + + mask1 = mask0 + 2; + mask2 = mask0 + 4; + mask3 = mask0 + 6; + mask4 = mask0 + 8; + mask5 = mask0 + 10; + mask6 = mask0 + 12; + mask7 = mask0 + 14; + + for (loop_cnt = height; loop_cnt--;) { + LD_SB2(src0_ptr, 16, src0, src1); + src0_ptr += src_stride; + LD_SH2(src1_ptr, 8, in0, in1); + in2 = LD_SH(src1_ptr + 16); + src1_ptr += src2_stride; + XORI_B2_128_SB(src0, src1); + + VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst0 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst0, dst0, dst0, dst0); + VSHF_B4_SB(src0, src1, mask4, mask5, mask6, mask7, + vec0, vec1, vec2, vec3); + dst1 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst1, dst1, dst1, dst1); + VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst2 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst2, dst2, dst2, dst2); + + HEVC_BI_RND_CLIP2(in0, in1, dst0, dst1, 7, dst0, dst1); + dst2 = __msa_adds_s_h(dst2, in2); + dst2 = __msa_srari_h(dst2, 7); + dst2 = CLIP_SH_0_255(dst2); + + PCKEV_B2_SB(dst1, dst0, dst2, dst2, tmp0, tmp1); + dst_val0 = __msa_copy_u_d((v2i64) tmp1, 0); + ST_SB(tmp0, dst); + SD(dst_val0, dst + 16); + dst += dst_stride; + } +} + +static void hevc_hz_bi_8t_32w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + uint32_t loop_cnt; + v16i8 src0, src1, src2, tmp0, tmp1; + v8i16 filt0, filt1, filt2, filt3; + v16i8 mask1, mask2, mask3, mask4, mask5, mask6, mask7; + v16i8 vec0, vec1, vec2, vec3; + v8i16 dst0, dst1, dst2, dst3; + v8i16 in0, in1, in2, in3; + v8i16 filter_vec, const_vec; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + + src0_ptr -= 3; + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3); + + mask1 = mask0 + 2; + mask2 = mask0 + 4; + mask3 = mask0 + 6; + mask4 = mask0 + 8; + mask5 = mask0 + 10; + mask6 = mask0 + 12; + mask7 = mask0 + 14; + + for (loop_cnt = height; loop_cnt--;) { + LD_SB2(src0_ptr, 16, src0, src1); + src2 = LD_SB(src0_ptr + 24); + src0_ptr += src_stride; + LD_SH4(src1_ptr, 8, in0, in1, in2, in3); + src1_ptr += src2_stride; + XORI_B3_128_SB(src0, src1, src2); + + VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst0 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst0, dst0, dst0, dst0); + VSHF_B4_SB(src0, src1, mask4, mask5, mask6, mask7, + vec0, vec1, vec2, vec3); + dst1 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst1, dst1, dst1, dst1); + VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst2 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst2, dst2, dst2, dst2); + VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst3 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst3, dst3, dst3, dst3); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3); + + PCKEV_B2_SB(dst1, dst0, dst3, dst2, tmp0, tmp1); + ST_SB2(tmp0, tmp1, dst, 16); + dst += dst_stride; + } +} + +static void hevc_hz_bi_8t_48w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + uint32_t loop_cnt; + v16i8 src0, src1, src2, src3; + v16i8 tmp0, tmp1, tmp2; + v8i16 filt0, filt1, filt2, filt3; + v16i8 mask1, mask2, mask3, mask4, mask5, mask6, mask7; + v16i8 vec0, vec1, vec2, vec3; + v8i16 dst0, dst1, dst2, dst3, dst4, dst5; + v8i16 in0, in1, in2, in3, in4, in5; + v8i16 filter_vec, const_vec; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + + src0_ptr -= 3; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3); + + mask1 = mask0 + 2; + mask2 = mask0 + 4; + mask3 = mask0 + 6; + mask4 = mask0 + 8; + mask5 = mask0 + 10; + mask6 = mask0 + 12; + mask7 = mask0 + 14; + + for (loop_cnt = height; loop_cnt--;) { + LD_SB2(src0_ptr, 16, src0, src1); + XORI_B2_128_SB(src0, src1); + LD_SH2(src1_ptr, 8, in0, in1); + + VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst0 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst0, dst0, dst0, dst0); + VSHF_B4_SB(src0, src1, mask4, mask5, mask6, mask7, + vec0, vec1, vec2, vec3); + dst1 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst1, dst1, dst1, dst1); + + HEVC_BI_RND_CLIP2(in0, in1, dst0, dst1, 7, dst0, dst1); + + tmp0 = __msa_pckev_b((v16i8) dst1, (v16i8) dst0); + ST_SB(tmp0, dst); + + LD_SB2(src0_ptr + 32, 8, src2, src3); + XORI_B2_128_SB(src2, src3); + src0_ptr += src_stride; + + LD_SH2(src1_ptr + 16, 8, in2, in3); + + VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst2 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst2, dst2, dst2, dst2); + VSHF_B4_SB(src1, src2, mask4, mask5, mask6, mask7, + vec0, vec1, vec2, vec3); + dst3 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst3, dst3, dst3, dst3); + + HEVC_BI_RND_CLIP2(in2, in3, dst2, dst3, 7, dst2, dst3); + + tmp1 = __msa_pckev_b((v16i8) dst3, (v16i8) dst2); + ST_SB(tmp1, dst + 16); + + LD_SH2(src1_ptr + 32, 8, in4, in5); + src1_ptr += src2_stride; + + VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst4 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst4, dst4, dst4, dst4); + VSHF_B4_SB(src3, src3, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst5 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst5, dst5, dst5, dst5); + + HEVC_BI_RND_CLIP2(in4, in5, dst4, dst5, 7, dst4, dst5); + + tmp2 = __msa_pckev_b((v16i8) dst5, (v16i8) dst4); + ST_SB(tmp2, dst + 32); + dst += dst_stride; + } +} + +static void hevc_hz_bi_8t_64w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + uint8_t *src0_ptr_tmp; + uint8_t *dst_tmp; + int16_t *src1_ptr_tmp; + uint32_t loop_cnt; + uint32_t cnt; + v16i8 src0, src1, src2, tmp0, tmp1; + v8i16 filt0, filt1, filt2, filt3; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask1, mask2, mask3, mask4, mask5, mask6, mask7; + v16i8 vec0, vec1, vec2, vec3; + v8i16 dst0, dst1, dst2, dst3; + v8i16 in0, in1, in2, in3; + v8i16 filter_vec, const_vec; + + src0_ptr -= 3; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3); + + mask1 = mask0 + 2; + mask2 = mask0 + 4; + mask3 = mask0 + 6; + mask4 = mask0 + 8; + mask5 = mask0 + 10; + mask6 = mask0 + 12; + mask7 = mask0 + 14; + + for (loop_cnt = height; loop_cnt--;) { + src0_ptr_tmp = src0_ptr; + dst_tmp = dst; + src1_ptr_tmp = src1_ptr; + + for (cnt = 2; cnt--;) { + LD_SB2(src0_ptr_tmp, 16, src0, src1); + src2 = LD_SB(src0_ptr_tmp + 24); + src0_ptr_tmp += 32; + LD_SH4(src1_ptr_tmp, 8, in0, in1, in2, in3); + src1_ptr_tmp += 32; + XORI_B3_128_SB(src0, src1, src2); + + VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst0 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst0, dst0, dst0, dst0); + VSHF_B4_SB(src0, src1, mask4, mask5, mask6, mask7, + vec0, vec1, vec2, vec3); + dst1 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst1, dst1, dst1, dst1); + VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst2 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst2, dst2, dst2, dst2); + VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst3 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst3, dst3, dst3, dst3); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0, dst1, dst2, dst3, 7, + dst0, dst1, dst2, dst3); + + PCKEV_B2_SB(dst1, dst0, dst3, dst2, tmp0, tmp1); + ST_SB2(tmp0, tmp1, dst_tmp, 16); + dst_tmp += 32; + } + + src1_ptr += src2_stride; + src0_ptr += src_stride; + dst += dst_stride; + } +} + +static void hevc_vt_bi_8t_4w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + int32_t loop_cnt; + v16i8 src0, src1, src2, src3, src4, src5; + v16i8 src6, src7, src8, src9, src10; + v8i16 in0, in1, in2, in3, in4, in5, in6, in7; + v16i8 src11, src12, src13, src14; + v16i8 src10_r, src32_r, src54_r, src76_r, src98_r; + v16i8 src21_r, src43_r, src65_r, src87_r, src109_r; + v16i8 src1110_r, src1211_r, src1312_r, src1413_r; + v16i8 src2110, src4332, src6554, src8776, src10998; + v16i8 src12111110, src14131312; + v8i16 dst10, dst32, dst54, dst76; + v8i16 filt0, filt1, filt2, filt3; + v8i16 filter_vec, const_vec; + + src0_ptr -= (3 * src_stride); + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3); + + LD_SB7(src0_ptr, src_stride, src0, src1, src2, src3, src4, src5, src6); + src0_ptr += (7 * src_stride); + ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, + src10_r, src32_r, src54_r, src21_r); + ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r); + ILVR_D3_SB(src21_r, src10_r, src43_r, src32_r, src65_r, src54_r, + src2110, src4332, src6554); + XORI_B3_128_SB(src2110, src4332, src6554); + + for (loop_cnt = (height >> 3); loop_cnt--;) { + LD_SB8(src0_ptr, src_stride, + src7, src8, src9, src10, src11, src12, src13, src14); + src0_ptr += (8 * src_stride); + LD_SH8(src1_ptr, src2_stride, in0, in1, in2, in3, in4, in5, in6, in7); + src1_ptr += (8 * src2_stride); + + ILVR_D2_SH(in1, in0, in3, in2, in0, in1); + ILVR_D2_SH(in5, in4, in7, in6, in2, in3); + ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, + src76_r, src87_r, src98_r, src109_r); + ILVR_B4_SB(src11, src10, src12, src11, src13, src12, src14, src13, + src1110_r, src1211_r, src1312_r, src1413_r); + ILVR_D4_SB(src87_r, src76_r, src109_r, src98_r, src1211_r, src1110_r, + src1413_r, src1312_r, + src8776, src10998, src12111110, src14131312); + XORI_B4_128_SB(src8776, src10998, src12111110, src14131312); + + dst10 = const_vec; + DPADD_SB4_SH(src2110, src4332, src6554, src8776, + filt0, filt1, filt2, filt3, dst10, dst10, dst10, dst10); + dst32 = const_vec; + DPADD_SB4_SH(src4332, src6554, src8776, src10998, + filt0, filt1, filt2, filt3, dst32, dst32, dst32, dst32); + dst54 = const_vec; + DPADD_SB4_SH(src6554, src8776, src10998, src12111110, + filt0, filt1, filt2, filt3, dst54, dst54, dst54, dst54); + dst76 = const_vec; + DPADD_SB4_SH(src8776, src10998, src12111110, src14131312, + filt0, filt1, filt2, filt3, dst76, dst76, dst76, dst76); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst10, dst32, dst54, dst76, 7, + dst10, dst32, dst54, dst76); + + PCKEV_B2_SH(dst32, dst10, dst76, dst54, dst10, dst54); + ST4x8_UB(dst10, dst54, dst, dst_stride); + dst += (8 * dst_stride); + + src2110 = src10998; + src4332 = src12111110; + src6554 = src14131312; + src6 = src14; + } +} + +static void hevc_vt_bi_8t_8w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + int32_t loop_cnt; + v16i8 src0, src1, src2, src3, src4, src5; + v16i8 src6, src7, src8, src9, src10; + v8i16 in0, in1, in2, in3; + v16i8 src10_r, src32_r, src54_r, src76_r, src98_r; + v16i8 src21_r, src43_r, src65_r, src87_r, src109_r; + v8i16 dst0_r, dst1_r, dst2_r, dst3_r; + v8i16 filt0, filt1, filt2, filt3; + v8i16 filter_vec, const_vec; + + src0_ptr -= (3 * src_stride); + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3); + + LD_SB7(src0_ptr, src_stride, src0, src1, src2, src3, src4, src5, src6); + src0_ptr += (7 * src_stride); + XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6); + ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, + src10_r, src32_r, src54_r, src21_r); + ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r); + + for (loop_cnt = (height >> 2); loop_cnt--;) { + LD_SB4(src0_ptr, src_stride, src7, src8, src9, src10); + src0_ptr += (4 * src_stride); + LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3); + src1_ptr += (4 * src2_stride); + XORI_B4_128_SB(src7, src8, src9, src10); + ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, + src76_r, src87_r, src98_r, src109_r); + + dst0_r = const_vec; + DPADD_SB4_SH(src10_r, src32_r, src54_r, src76_r, + filt0, filt1, filt2, filt3, + dst0_r, dst0_r, dst0_r, dst0_r); + dst1_r = const_vec; + DPADD_SB4_SH(src21_r, src43_r, src65_r, src87_r, + filt0, filt1, filt2, filt3, + dst1_r, dst1_r, dst1_r, dst1_r); + dst2_r = const_vec; + DPADD_SB4_SH(src32_r, src54_r, src76_r, src98_r, + filt0, filt1, filt2, filt3, + dst2_r, dst2_r, dst2_r, dst2_r); + dst3_r = const_vec; + DPADD_SB4_SH(src43_r, src65_r, src87_r, src109_r, + filt0, filt1, filt2, filt3, + dst3_r, dst3_r, dst3_r, dst3_r); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0_r, dst1_r, dst2_r, dst3_r, 7, + dst0_r, dst1_r, dst2_r, dst3_r); + + PCKEV_B2_SH(dst1_r, dst0_r, dst3_r, dst2_r, dst0_r, dst1_r); + ST8x4_UB(dst0_r, dst1_r, dst, dst_stride); + dst += (4 * dst_stride); + + src10_r = src54_r; + src32_r = src76_r; + src54_r = src98_r; + src21_r = src65_r; + src43_r = src87_r; + src65_r = src109_r; + + src6 = src10; + } +} + +static void hevc_vt_bi_8t_12w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + int32_t loop_cnt; + v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10; + v8i16 in0, in1, in2, in3, in4, in5, in6, in7; + v16i8 src10_r, src32_r, src54_r, src76_r, src98_r; + v16i8 src21_r, src43_r, src65_r, src87_r, src109_r; + v8i16 dst0_r, dst1_r, dst2_r, dst3_r; + v16i8 src10_l, src32_l, src54_l, src76_l, src98_l; + v16i8 src21_l, src43_l, src65_l, src87_l, src109_l; + v16i8 src2110, src4332, src6554, src8776, src10998; + v8i16 dst0_l, dst1_l; + v8i16 filt0, filt1, filt2, filt3; + v8i16 filter_vec, const_vec; + + src0_ptr -= (3 * src_stride); + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3); + + LD_SB7(src0_ptr, src_stride, src0, src1, src2, src3, src4, src5, src6); + src0_ptr += (7 * src_stride); + XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6); + + ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, + src10_r, src32_r, src54_r, src21_r); + ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r); + ILVL_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, + src10_l, src32_l, src54_l, src21_l); + ILVL_B2_SB(src4, src3, src6, src5, src43_l, src65_l); + ILVR_D3_SB(src21_l, src10_l, src43_l, src32_l, src65_l, src54_l, + src2110, src4332, src6554); + + for (loop_cnt = (height >> 2); loop_cnt--;) { + LD_SB4(src0_ptr, src_stride, src7, src8, src9, src10); + src0_ptr += (4 * src_stride); + LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3); + LD_SH4((src1_ptr + 8), src2_stride, in4, in5, in6, in7); + src1_ptr += (4 * src2_stride); + + ILVR_D2_SH(in5, in4, in7, in6, in4, in5); + XORI_B4_128_SB(src7, src8, src9, src10); + ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, + src76_r, src87_r, src98_r, src109_r); + ILVL_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, + src76_l, src87_l, src98_l, src109_l); + ILVR_D2_SB(src87_l, src76_l, src109_l, src98_l, src8776, src10998); + + dst0_r = const_vec; + DPADD_SB4_SH(src10_r, src32_r, src54_r, src76_r, + filt0, filt1, filt2, filt3, + dst0_r, dst0_r, dst0_r, dst0_r); + dst1_r = const_vec; + DPADD_SB4_SH(src21_r, src43_r, src65_r, src87_r, + filt0, filt1, filt2, filt3, + dst1_r, dst1_r, dst1_r, dst1_r); + dst2_r = const_vec; + DPADD_SB4_SH(src32_r, src54_r, src76_r, src98_r, + filt0, filt1, filt2, filt3, + dst2_r, dst2_r, dst2_r, dst2_r); + dst3_r = const_vec; + DPADD_SB4_SH(src43_r, src65_r, src87_r, src109_r, + filt0, filt1, filt2, filt3, + dst3_r, dst3_r, dst3_r, dst3_r); + dst0_l = const_vec; + DPADD_SB4_SH(src2110, src4332, src6554, src8776, + filt0, filt1, filt2, filt3, + dst0_l, dst0_l, dst0_l, dst0_l); + dst1_l = const_vec; + DPADD_SB4_SH(src4332, src6554, src8776, src10998, + filt0, filt1, filt2, filt3, + dst1_l, dst1_l, dst1_l, dst1_l); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0_r, dst1_r, dst2_r, dst3_r, 7, + dst0_r, dst1_r, dst2_r, dst3_r); + HEVC_BI_RND_CLIP2(in4, in5, dst0_l, dst1_l, 7, dst0_l, dst1_l); + + + PCKEV_B2_SH(dst1_r, dst0_r, dst3_r, dst2_r, dst0_r, dst1_r); + dst0_l = (v8i16) __msa_pckev_b((v16i8) dst1_l, (v16i8) dst0_l); + ST12x4_UB(dst0_r, dst1_r, dst0_l, dst, dst_stride); + dst += (4 * dst_stride); + + src10_r = src54_r; + src32_r = src76_r; + src54_r = src98_r; + src21_r = src65_r; + src43_r = src87_r; + src65_r = src109_r; + src2110 = src6554; + src4332 = src8776; + src6554 = src10998; + src6 = src10; + } +} + +static void hevc_vt_bi_8t_16multx2mult_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height, int32_t width) +{ + uint8_t *src0_ptr_tmp; + int16_t *src1_ptr_tmp; + uint8_t *dst_tmp; + uint32_t loop_cnt; + uint32_t cnt; + v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8; + v8i16 in0, in1, in2, in3; + v16i8 src10_r, src32_r, src54_r, src76_r; + v16i8 src21_r, src43_r, src65_r, src87_r; + v8i16 dst0_r, dst1_r; + v16i8 src10_l, src32_l, src54_l, src76_l; + v16i8 src21_l, src43_l, src65_l, src87_l; + v8i16 dst0_l, dst1_l; + v8i16 filt0, filt1, filt2, filt3; + v8i16 filter_vec, const_vec; + + src0_ptr -= (3 * src_stride); + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3); + + for (cnt = (width >> 4); cnt--;) { + src0_ptr_tmp = src0_ptr; + src1_ptr_tmp = src1_ptr; + dst_tmp = dst; + + LD_SB7(src0_ptr_tmp, src_stride, + src0, src1, src2, src3, src4, src5, src6); + src0_ptr_tmp += (7 * src_stride); + XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6); + + ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, + src10_r, src32_r, src54_r, src21_r); + ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r); + ILVL_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, + src10_l, src32_l, src54_l, src21_l); + ILVL_B2_SB(src4, src3, src6, src5, src43_l, src65_l); + + for (loop_cnt = (height >> 1); loop_cnt--;) { + LD_SB2(src0_ptr_tmp, src_stride, src7, src8); + src0_ptr_tmp += (2 * src_stride); + LD_SH2(src1_ptr_tmp, src2_stride, in0, in1); + LD_SH2((src1_ptr_tmp + 8), src2_stride, in2, in3); + src1_ptr_tmp += (2 * src2_stride); + XORI_B2_128_SB(src7, src8); + + ILVR_B2_SB(src7, src6, src8, src7, src76_r, src87_r); + ILVL_B2_SB(src7, src6, src8, src7, src76_l, src87_l); + + dst0_r = const_vec; + DPADD_SB4_SH(src10_r, src32_r, src54_r, src76_r, + filt0, filt1, filt2, filt3, + dst0_r, dst0_r, dst0_r, dst0_r); + dst1_r = const_vec; + DPADD_SB4_SH(src21_r, src43_r, src65_r, src87_r, + filt0, filt1, filt2, filt3, + dst1_r, dst1_r, dst1_r, dst1_r); + dst0_l = const_vec; + DPADD_SB4_SH(src10_l, src32_l, src54_l, src76_l, + filt0, filt1, filt2, filt3, + dst0_l, dst0_l, dst0_l, dst0_l); + dst1_l = const_vec; + DPADD_SB4_SH(src21_l, src43_l, src65_l, src87_l, + filt0, filt1, filt2, filt3, + dst1_l, dst1_l, dst1_l, dst1_l); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0_r, dst1_r, dst0_l, dst1_l, 7, + dst0_r, dst1_r, dst0_l, dst1_l); + + PCKEV_B2_SH(dst0_l, dst0_r, dst1_l, dst1_r, dst0_r, dst1_r); + ST_SH2(dst0_r, dst1_r, dst_tmp, dst_stride); + dst_tmp += (2 * dst_stride); + + src10_r = src32_r; + src32_r = src54_r; + src54_r = src76_r; + src21_r = src43_r; + src43_r = src65_r; + src65_r = src87_r; + src10_l = src32_l; + src32_l = src54_l; + src54_l = src76_l; + src21_l = src43_l; + src43_l = src65_l; + src65_l = src87_l; + src6 = src8; + } + + src0_ptr += 16; + src1_ptr += 16; + dst += 16; + } +} + +static void hevc_vt_bi_8t_16w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + hevc_vt_bi_8t_16multx2mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter, height, 16); +} + +static void hevc_vt_bi_8t_24w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + hevc_vt_bi_8t_16multx2mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter, height, 16); + hevc_vt_bi_8t_8w_msa(src0_ptr + 16, src_stride, src1_ptr + 16, src2_stride, + dst + 16, dst_stride, filter, height); +} + +static void hevc_vt_bi_8t_32w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + hevc_vt_bi_8t_16multx2mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter, height, 32); +} + +static void hevc_vt_bi_8t_48w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + hevc_vt_bi_8t_16multx2mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter, height, 48); +} + +static void hevc_vt_bi_8t_64w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + hevc_vt_bi_8t_16multx2mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter, height, 64); +} + +static void hevc_hv_bi_8t_4w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter_x, + const int8_t *filter_y, + int32_t height) +{ + uint32_t loop_cnt; + v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8; + v8i16 in0, in1; + v8i16 filt0, filt1, filt2, filt3; + v4i32 filt_h0, filt_h1, filt_h2, filt_h3; + v16i8 mask1, mask2, mask3; + v8i16 filter_vec, const_vec; + v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7; + v16i8 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15; + v8i16 dst30, dst41, dst52, dst63, dst66, dst87; + v4i32 dst0_r, dst1_r, in0_r, in0_l; + v8i16 dst10_r, dst32_r, dst54_r, dst76_r; + v8i16 dst21_r, dst43_r, dst65_r, dst87_r; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20 }; + v8u16 mask4 = { 0, 4, 1, 5, 2, 6, 3, 7 }; + + src0_ptr -= ((3 * src_stride) + 3); + filter_vec = LD_SH(filter_x); + SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3); + + filter_vec = LD_SH(filter_y); + vec0 = __msa_clti_s_b((v16i8) filter_vec, 0); + filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec); + + SPLATI_W4_SW(filter_vec, filt_h0, filt_h1, filt_h2, filt_h3); + + mask1 = mask0 + 2; + mask2 = mask0 + 4; + mask3 = mask0 + 6; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + LD_SB7(src0_ptr, src_stride, src0, src1, src2, src3, src4, src5, src6); + src0_ptr += (7 * src_stride); + XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6); + + /* row 0 row 1 row 2 row 3 */ + VSHF_B4_SB(src0, src3, mask0, mask1, mask2, mask3, vec0, vec1, vec2, vec3); + VSHF_B4_SB(src1, src4, mask0, mask1, mask2, mask3, vec4, vec5, vec6, vec7); + VSHF_B4_SB(src2, src5, mask0, mask1, mask2, mask3, + vec8, vec9, vec10, vec11); + VSHF_B4_SB(src3, src6, mask0, mask1, mask2, mask3, + vec12, vec13, vec14, vec15); + + dst30 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst30, dst30, dst30, dst30); + dst41 = const_vec; + DPADD_SB4_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2, filt3, + dst41, dst41, dst41, dst41); + dst52 = const_vec; + DPADD_SB4_SH(vec8, vec9, vec10, vec11, filt0, filt1, filt2, filt3, + dst52, dst52, dst52, dst52); + dst63 = const_vec; + DPADD_SB4_SH(vec12, vec13, vec14, vec15, filt0, filt1, filt2, filt3, + dst63, dst63, dst63, dst63); + + ILVR_H3_SH(dst41, dst30, dst52, dst41, dst63, dst52, + dst10_r, dst21_r, dst32_r); + dst43_r = __msa_ilvl_h(dst41, dst30); + dst54_r = __msa_ilvl_h(dst52, dst41); + dst65_r = __msa_ilvl_h(dst63, dst52); + dst66 = (v8i16) __msa_splati_d((v2i64) dst63, 1); + + for (loop_cnt = height >> 1; loop_cnt--;) { + LD_SB2(src0_ptr, src_stride, src7, src8); + src0_ptr += (2 * src_stride); + LD_SH2(src1_ptr, src2_stride, in0, in1); + src1_ptr += (2 * src2_stride); + + in0 = (v8i16) __msa_ilvr_d((v2i64) in1, (v2i64) in0); + XORI_B2_128_SB(src7, src8); + + VSHF_B4_SB(src7, src8, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst87 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst87, dst87, dst87, dst87); + dst76_r = __msa_ilvr_h(dst87, dst66); + dst0_r = HEVC_FILT_8TAP(dst10_r, dst32_r, dst54_r, dst76_r, + filt_h0, filt_h1, filt_h2, filt_h3); + dst87_r = __msa_vshf_h((v8i16) mask4, dst87, dst87); + dst1_r = HEVC_FILT_8TAP(dst21_r, dst43_r, dst65_r, dst87_r, + filt_h0, filt_h1, filt_h2, filt_h3); + + dst0_r >>= 6; + dst1_r >>= 6; + UNPCK_SH_SW(in0, in0_r, in0_l); + dst0_r = __msa_adds_s_w(dst0_r, in0_r); + dst1_r = __msa_adds_s_w(dst1_r, in0_l); + SRARI_W2_SW(dst0_r, dst1_r, 7); + dst0_r = CLIP_SW_0_255(dst0_r); + dst1_r = CLIP_SW_0_255(dst1_r); + + HEVC_PCK_SW_SB2(dst1_r, dst0_r, dst0_r); + ST4x2_UB(dst0_r, dst, dst_stride); + dst += (2 * dst_stride); + + dst10_r = dst32_r; + dst32_r = dst54_r; + dst54_r = dst76_r; + dst21_r = dst43_r; + dst43_r = dst65_r; + dst65_r = dst87_r; + dst66 = (v8i16) __msa_splati_d((v2i64) dst87, 1); + } +} + +static void hevc_hv_bi_8t_8multx2mult_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter_x, + const int8_t *filter_y, + int32_t height, int32_t width) +{ + uint32_t loop_cnt; + uint32_t cnt; + uint8_t *src0_ptr_tmp; + int16_t *src1_ptr_tmp; + uint8_t *dst_tmp; + v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8; + v8i16 in0, in1; + v4i32 in0_r, in0_l, in1_r, in1_l; + v8i16 filt0, filt1, filt2, filt3; + v4i32 filt_h0, filt_h1, filt_h2, filt_h3; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask1, mask2, mask3; + v8i16 filter_vec, const_vec; + v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7; + v16i8 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15; + v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8; + v4i32 dst0_r, dst0_l, dst1_r, dst1_l; + v8i16 dst10_r, dst32_r, dst54_r, dst76_r; + v8i16 dst10_l, dst32_l, dst54_l, dst76_l; + v8i16 dst21_r, dst43_r, dst65_r, dst87_r; + v8i16 dst21_l, dst43_l, dst65_l, dst87_l; + + src0_ptr -= ((3 * src_stride) + 3); + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter_x); + SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3); + + filter_vec = LD_SH(filter_y); + vec0 = __msa_clti_s_b((v16i8) filter_vec, 0); + filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec); + SPLATI_W4_SW(filter_vec, filt_h0, filt_h1, filt_h2, filt_h3); + + mask1 = mask0 + 2; + mask2 = mask0 + 4; + mask3 = mask0 + 6; + + for (cnt = width >> 3; cnt--;) { + src0_ptr_tmp = src0_ptr; + dst_tmp = dst; + src1_ptr_tmp = src1_ptr; + + LD_SB7(src0_ptr_tmp, src_stride, + src0, src1, src2, src3, src4, src5, src6); + src0_ptr_tmp += (7 * src_stride); + XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6); + + /* row 0 row 1 row 2 row 3 */ + VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3, + vec4, vec5, vec6, vec7); + VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3, + vec8, vec9, vec10, vec11); + VSHF_B4_SB(src3, src3, mask0, mask1, mask2, mask3, + vec12, vec13, vec14, vec15); + dst0 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst0, dst0, dst0, dst0); + dst1 = const_vec; + DPADD_SB4_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2, filt3, + dst1, dst1, dst1, dst1); + dst2 = const_vec; + DPADD_SB4_SH(vec8, vec9, vec10, vec11, filt0, filt1, filt2, filt3, + dst2, dst2, dst2, dst2); + dst3 = const_vec; + DPADD_SB4_SH(vec12, vec13, vec14, vec15, filt0, filt1, filt2, filt3, + dst3, dst3, dst3, dst3); + + VSHF_B4_SB(src4, src4, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + VSHF_B4_SB(src5, src5, mask0, mask1, mask2, mask3, + vec4, vec5, vec6, vec7); + VSHF_B4_SB(src6, src6, mask0, mask1, mask2, mask3, + vec8, vec9, vec10, vec11); + dst4 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst4, dst4, dst4, dst4); + dst5 = const_vec; + DPADD_SB4_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2, filt3, + dst5, dst5, dst5, dst5); + dst6 = const_vec; + DPADD_SB4_SH(vec8, vec9, vec10, vec11, filt0, filt1, filt2, filt3, + dst6, dst6, dst6, dst6); + + ILVR_H4_SH(dst1, dst0, dst3, dst2, dst5, dst4, dst2, dst1, + dst10_r, dst32_r, dst54_r, dst21_r); + ILVR_H2_SH(dst4, dst3, dst6, dst5, dst43_r, dst65_r); + ILVL_H4_SH(dst1, dst0, dst3, dst2, dst5, dst4, dst2, dst1, + dst10_l, dst32_l, dst54_l, dst21_l); + ILVL_H2_SH(dst4, dst3, dst6, dst5, dst43_l, dst65_l); + + for (loop_cnt = height >> 1; loop_cnt--;) { + /* row 7 */ + LD_SB2(src0_ptr_tmp, src_stride, src7, src8); + XORI_B2_128_SB(src7, src8); + src0_ptr_tmp += 2 * src_stride; + + LD_SH2(src1_ptr_tmp, src2_stride, in0, in1); + src1_ptr_tmp += (2 * src2_stride); + + VSHF_B4_SB(src7, src7, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst7 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst7, dst7, dst7, dst7); + + ILVRL_H2_SH(dst7, dst6, dst76_r, dst76_l); + dst0_r = HEVC_FILT_8TAP(dst10_r, dst32_r, dst54_r, dst76_r, + filt_h0, filt_h1, filt_h2, filt_h3); + dst0_l = HEVC_FILT_8TAP(dst10_l, dst32_l, dst54_l, dst76_l, + filt_h0, filt_h1, filt_h2, filt_h3); + dst0_r >>= 6; + dst0_l >>= 6; + + VSHF_B4_SB(src8, src8, mask0, mask1, mask2, mask3, + vec0, vec1, vec2, vec3); + dst8 = const_vec; + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, + dst8, dst8, dst8, dst8); + + ILVRL_H2_SH(dst8, dst7, dst87_r, dst87_l); + dst1_r = HEVC_FILT_8TAP(dst21_r, dst43_r, dst65_r, dst87_r, + filt_h0, filt_h1, filt_h2, filt_h3); + dst1_l = HEVC_FILT_8TAP(dst21_l, dst43_l, dst65_l, dst87_l, + filt_h0, filt_h1, filt_h2, filt_h3); + dst1_r >>= 6; + dst1_l >>= 6; + + UNPCK_SH_SW(in0, in0_r, in0_l); + UNPCK_SH_SW(in1, in1_r, in1_l); + in0_r = __msa_adds_s_w(in0_r, dst0_r); + in0_l = __msa_adds_s_w(in0_l, dst0_l); + in1_r = __msa_adds_s_w(in1_r, dst1_r); + in1_l = __msa_adds_s_w(in1_l, dst1_l); + SRARI_W4_SW(in0_r, in0_l, in1_r, in1_l, 7); + in0_r = CLIP_SW_0_255(in0_r); + in0_l = CLIP_SW_0_255(in0_l); + in1_r = CLIP_SW_0_255(in1_r); + in1_l = CLIP_SW_0_255(in1_l); + + HEVC_PCK_SW_SB4(in0_l, in0_r, in1_l, in1_r, dst0_r); + ST8x2_UB(dst0_r, dst_tmp, dst_stride); + dst_tmp += (2 * dst_stride); + + dst10_r = dst32_r; + dst32_r = dst54_r; + dst54_r = dst76_r; + dst10_l = dst32_l; + dst32_l = dst54_l; + dst54_l = dst76_l; + dst21_r = dst43_r; + dst43_r = dst65_r; + dst65_r = dst87_r; + dst21_l = dst43_l; + dst43_l = dst65_l; + dst65_l = dst87_l; + dst6 = dst8; + } + + src0_ptr += 8; + dst += 8; + src1_ptr += 8; + } +} + +static void hevc_hv_bi_8t_8w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter_x, + const int8_t *filter_y, + int32_t height) +{ + hevc_hv_bi_8t_8multx2mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter_x, filter_y, + height, 8); +} + +static void hevc_hv_bi_8t_12w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter_x, + const int8_t *filter_y, + int32_t height) +{ + hevc_hv_bi_8t_8multx2mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter_x, filter_y, + height, 8); + + hevc_hv_bi_8t_4w_msa(src0_ptr + 8, src_stride, src1_ptr + 8, src2_stride, + dst + 8, dst_stride, filter_x, filter_y, height); +} + +static void hevc_hv_bi_8t_16w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter_x, + const int8_t *filter_y, + int32_t height) +{ + hevc_hv_bi_8t_8multx2mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter_x, filter_y, + height, 16); +} + +static void hevc_hv_bi_8t_24w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter_x, + const int8_t *filter_y, + int32_t height) +{ + hevc_hv_bi_8t_8multx2mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter_x, filter_y, + height, 24); +} + +static void hevc_hv_bi_8t_32w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter_x, + const int8_t *filter_y, + int32_t height) +{ + hevc_hv_bi_8t_8multx2mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter_x, filter_y, + height, 32); +} + +static void hevc_hv_bi_8t_48w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter_x, + const int8_t *filter_y, + int32_t height) +{ + hevc_hv_bi_8t_8multx2mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter_x, filter_y, + height, 48); +} + +static void hevc_hv_bi_8t_64w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter_x, + const int8_t *filter_y, + int32_t height) +{ + hevc_hv_bi_8t_8multx2mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter_x, filter_y, + height, 64); +} + +static void hevc_hz_bi_4t_4x2_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + v8i16 filt0, filt1; + v16i8 src0, src1, dst0, vec0, vec1; + v8i16 in0, in1; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20 }; + v16i8 mask1; + v8i16 tmp0; + v8i16 filter_vec, const_vec; + + src0_ptr -= 1; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + mask1 = mask0 + 2; + + LD_SB2(src0_ptr, src_stride, src0, src1); + LD_SH2(src1_ptr, src2_stride, in0, in1); + in0 = (v8i16) __msa_ilvr_d((v2i64) in1, (v2i64) in0); + XORI_B2_128_SB(src0, src1); + VSHF_B2_SB(src0, src1, src0, src1, mask0, mask1, vec0, vec1); + tmp0 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, tmp0, tmp0); + + tmp0 = __msa_adds_s_h(tmp0, in0); + tmp0 = __msa_srari_h(tmp0, 7); + tmp0 = CLIP_SH_0_255(tmp0); + dst0 = __msa_pckev_b((v16i8) tmp0, (v16i8) tmp0); + + ST4x2_UB(dst0, dst, dst_stride); +} + +static void hevc_hz_bi_4t_4x4_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + v8i16 filt0, filt1; + v16i8 src0, src1, src2, src3, dst0, vec0, vec1; + v8i16 in0, in1, in2, in3; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20 }; + v16i8 mask1; + v8i16 tmp0, tmp1; + v8i16 filter_vec, const_vec; + + src0_ptr -= 1; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + mask1 = mask0 + 2; + + LD_SB4(src0_ptr, src_stride, src0, src1, src2, src3); + LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3); + + ILVR_D2_SH(in1, in0, in3, in2, in0, in1); + XORI_B4_128_SB(src0, src1, src2, src3); + + VSHF_B2_SB(src0, src1, src0, src1, mask0, mask1, vec0, vec1); + tmp0 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, tmp0, tmp0); + VSHF_B2_SB(src2, src3, src2, src3, mask0, mask1, vec0, vec1); + tmp1 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, tmp1, tmp1); + HEVC_BI_RND_CLIP2(in0, in1, tmp0, tmp1, 7, tmp0, tmp1); + dst0 = __msa_pckev_b((v16i8) tmp1, (v16i8) tmp0); + + ST4x4_UB(dst0, dst0, 0, 1, 2, 3, dst, dst_stride); +} + +static void hevc_hz_bi_4t_4x8multiple_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + uint32_t loop_cnt; + v8i16 filt0, filt1; + v16i8 src0, src1, src2, src3, src4, src5, src6, src7; + v16i8 dst0, dst1; + v8i16 in0, in1, in2, in3, in4, in5, in6, in7; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20 }; + v16i8 mask1, vec0, vec1; + v8i16 tmp0, tmp1, tmp2, tmp3; + v8i16 filter_vec, const_vec; + + src0_ptr -= 1; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + mask1 = mask0 + 2; + + for (loop_cnt = (height >> 3); loop_cnt--;) { + LD_SB8(src0_ptr, src_stride, + src0, src1, src2, src3, src4, src5, src6, src7); + src0_ptr += (8 * src_stride); + LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3); + src1_ptr += (4 * src2_stride); + LD_SH4(src1_ptr, src2_stride, in4, in5, in6, in7); + src1_ptr += (4 * src2_stride); + ILVR_D2_SH(in1, in0, in3, in2, in0, in1); + ILVR_D2_SH(in5, in4, in7, in6, in2, in3); + XORI_B8_128_SB(src0, src1, src2, src3, src4, src5, src6, src7); + + VSHF_B2_SB(src0, src1, src0, src1, mask0, mask1, vec0, vec1); + tmp0 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, tmp0, tmp0); + VSHF_B2_SB(src2, src3, src2, src3, mask0, mask1, vec0, vec1); + tmp1 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, tmp1, tmp1); + VSHF_B2_SB(src4, src5, src4, src5, mask0, mask1, vec0, vec1); + tmp2 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, tmp2, tmp2); + VSHF_B2_SB(src6, src7, src6, src7, mask0, mask1, vec0, vec1); + tmp3 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, tmp3, tmp3); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + tmp0, tmp1, tmp2, tmp3, 7, tmp0, tmp1, tmp2, tmp3); + + PCKEV_B2_SB(tmp1, tmp0, tmp3, tmp2, dst0, dst1); + ST4x8_UB(dst0, dst1, dst, dst_stride); + dst += (8 * dst_stride); + } +} + +static void hevc_hz_bi_4t_4w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + if (2 == height) { + hevc_hz_bi_4t_4x2_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter, height); + } else if (4 == height) { + hevc_hz_bi_4t_4x4_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter, height); + } else if (8 == height || 16 == height) { + hevc_hz_bi_4t_4x8multiple_msa(src0_ptr, src_stride, + src1_ptr, src2_stride, + dst, dst_stride, filter, height); + } +} + +static void hevc_hz_bi_4t_6w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + uint32_t loop_cnt; + v8i16 filt0, filt1; + v16i8 src0, src1, src2, src3; + v8i16 in0, in1, in2, in3; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask1; + v16i8 vec0, vec1; + v8i16 dst0, dst1, dst2, dst3; + v8i16 filter_vec, const_vec; + + src0_ptr -= 1; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + mask1 = mask0 + 2; + + for (loop_cnt = (height >> 2); loop_cnt--;) { + LD_SB4(src0_ptr, src_stride, src0, src1, src2, src3); + src0_ptr += (4 * src_stride); + LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3); + src1_ptr += (4 * src2_stride); + XORI_B4_128_SB(src0, src1, src2, src3); + + VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1); + dst0 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0); + VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec0, vec1); + dst1 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1); + VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec0, vec1); + dst2 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2); + VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1); + dst3 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3); + + PCKEV_B2_SH(dst1, dst0, dst3, dst2, dst0, dst1); + ST6x4_UB(dst0, dst1, dst, dst_stride); + dst += (4 * dst_stride); + } +} + +static void hevc_hz_bi_4t_8x2_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + v8i16 filt0, filt1; + v16i8 src0, src1; + v8i16 in0, in1; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask1, vec0, vec1; + v8i16 dst0, dst1; + v8i16 filter_vec, const_vec; + + src0_ptr -= 1; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + mask1 = mask0 + 2; + + LD_SB2(src0_ptr, src_stride, src0, src1); + LD_SH2(src1_ptr, src2_stride, in0, in1); + XORI_B2_128_SB(src0, src1); + + VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1); + dst0 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0); + VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec0, vec1); + dst1 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1); + HEVC_BI_RND_CLIP2(in0, in1, dst0, dst1, 7, dst0, dst1); + + dst0 = (v8i16) __msa_pckev_b((v16i8) dst1, (v16i8) dst0); + ST8x2_UB(dst0, dst, dst_stride); +} + +static void hevc_hz_bi_4t_8x6_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + v8i16 filt0, filt1; + v16i8 src0, src1, src2, src3, src4, src5; + v8i16 in0, in1, in2, in3, in4, in5; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask1; + v16i8 vec0, vec1; + v8i16 dst0, dst1, dst2, dst3, dst4, dst5; + v8i16 filter_vec, const_vec; + + src0_ptr -= 1; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + mask1 = mask0 + 2; + + LD_SB6(src0_ptr, src_stride, src0, src1, src2, src3, src4, src5); + LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3); + src1_ptr += (4 * src2_stride); + LD_SH2(src1_ptr, src2_stride, in4, in5); + XORI_B6_128_SB(src0, src1, src2, src3, src4, src5); + + VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1); + dst0 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0); + VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec0, vec1); + dst1 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1); + VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec0, vec1); + dst2 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2); + VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1); + dst3 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3); + VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1); + dst4 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4); + VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec0, vec1); + dst5 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst5, dst5); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3); + HEVC_BI_RND_CLIP2(in4, in5, dst4, dst5, 7, dst4, dst5); + + PCKEV_B2_SH(dst1, dst0, dst3, dst2, dst0, dst1); + dst2 = (v8i16) __msa_pckev_b((v16i8) dst5, (v16i8) dst4); + ST8x4_UB(dst0, dst1, dst, dst_stride); + dst += (4 * dst_stride); + ST8x2_UB(dst2, dst, dst_stride); +} + +static void hevc_hz_bi_4t_8x4multiple_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + uint32_t loop_cnt; + v8i16 filt0, filt1; + v16i8 src0, src1, src2, src3; + v8i16 in0, in1, in2, in3; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask1; + v16i8 vec0, vec1; + v8i16 dst0, dst1, dst2, dst3; + v8i16 filter_vec, const_vec; + + src0_ptr -= 1; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + mask1 = mask0 + 2; + + for (loop_cnt = (height >> 2); loop_cnt--;) { + LD_SB4(src0_ptr, src_stride, src0, src1, src2, src3); + src0_ptr += (4 * src_stride); + LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3); + src1_ptr += (4 * src2_stride); + XORI_B4_128_SB(src0, src1, src2, src3); + + VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1); + dst0 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0); + VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec0, vec1); + dst1 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1); + VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec0, vec1); + dst2 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2); + VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1); + dst3 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3); + + PCKEV_B2_SH(dst1, dst0, dst3, dst2, dst0, dst1); + ST8x4_UB(dst0, dst1, dst, dst_stride); + dst += (4 * dst_stride); + } +} + +static void hevc_hz_bi_4t_8w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + if (2 == height) { + hevc_hz_bi_4t_8x2_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter, height); + } else if (6 == height) { + hevc_hz_bi_4t_8x6_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter, height); + } else if (0 == (height % 4)) { + hevc_hz_bi_4t_8x4multiple_msa(src0_ptr, src_stride, + src1_ptr, src2_stride, + dst, dst_stride, filter, height); + } +} + +static void hevc_hz_bi_4t_12w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + uint32_t loop_cnt; + v8i16 filt0, filt1; + v16i8 src0, src1, src2, src3; + v8i16 in0, in1, in2, in3, in4, in5, in6, in7; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask2 = { + 8, 9, 9, 10, 10, 11, 11, 12, 24, 25, 25, 26, 26, 27, 27, 28 + }; + v16i8 mask1, mask3; + v16i8 vec0, vec1; + v8i16 dst0, dst1, dst2, dst3, dst4, dst5; + v8i16 filter_vec, const_vec; + + src0_ptr -= 1; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + mask1 = mask0 + 2; + mask3 = mask2 + 2; + + for (loop_cnt = (height >> 2); loop_cnt--;) { + LD_SB4(src0_ptr, src_stride, src0, src1, src2, src3); + src0_ptr += (4 * src_stride); + LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3); + LD_SH4(src1_ptr + 8, src2_stride, in4, in5, in6, in7); + src1_ptr += (4 * src2_stride); + + ILVR_D2_SH(in5, in4, in7, in6, in4, in5); + XORI_B4_128_SB(src0, src1, src2, src3); + + VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1); + dst0 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0); + VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec0, vec1); + dst1 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1); + VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec0, vec1); + dst2 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2); + VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1); + dst3 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3); + VSHF_B2_SB(src0, src1, src0, src1, mask2, mask3, vec0, vec1); + dst4 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4); + VSHF_B2_SB(src2, src3, src2, src3, mask2, mask3, vec0, vec1); + dst5 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst5, dst5); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3); + HEVC_BI_RND_CLIP2(in4, in5, dst4, dst5, 7, dst4, dst5); + + PCKEV_B2_SH(dst1, dst0, dst3, dst2, dst0, dst1); + dst2 = (v8i16) __msa_pckev_b((v16i8) dst5, (v16i8) dst4); + ST12x4_UB(dst0, dst1, dst2, dst, dst_stride); + dst += (4 * dst_stride); + } +} + +static void hevc_hz_bi_4t_16w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + uint32_t loop_cnt; + v16i8 src0, src1, src2, src3, src4, src5, src6, src7; + v8i16 in0, in1, in2, in3, in4, in5, in6, in7; + v8i16 filt0, filt1; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask1; + v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7; + v16i8 vec0, vec1; + v8i16 filter_vec, const_vec; + + src0_ptr -= 1; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + mask1 = mask0 + 2; + + for (loop_cnt = (height >> 2); loop_cnt--;) { + LD_SB4(src0_ptr, src_stride, src0, src2, src4, src6); + LD_SB4(src0_ptr + 8, src_stride, src1, src3, src5, src7); + src0_ptr += (4 * src_stride); + LD_SH4(src1_ptr, src2_stride, in0, in2, in4, in6); + LD_SH4(src1_ptr + 8, src2_stride, in1, in3, in5, in7); + src1_ptr += (4 * src2_stride); + XORI_B8_128_SB(src0, src1, src2, src3, src4, src5, src6, src7); + + VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1); + dst0 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0); + VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec0, vec1); + dst1 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1); + VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec0, vec1); + dst2 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2); + VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1); + dst3 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3); + VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1); + dst4 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4); + VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec0, vec1); + dst5 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst5, dst5); + VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec0, vec1); + dst6 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst6, dst6); + VSHF_B2_SB(src7, src7, src7, src7, mask0, mask1, vec0, vec1); + dst7 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst7, dst7); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3); + HEVC_BI_RND_CLIP4(in4, in5, in6, in7, + dst4, dst5, dst6, dst7, 7, dst4, dst5, dst6, dst7); + + PCKEV_B4_SH(dst1, dst0, dst3, dst2, + dst5, dst4, dst7, dst6, dst0, dst1, dst2, dst3); + ST_SH4(dst0, dst1, dst2, dst3, dst, dst_stride); + dst += (4 * dst_stride); + } +} + +static void hevc_hz_bi_4t_24w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + int16_t *src1_ptr_tmp; + uint8_t *dst_tmp; + uint32_t loop_cnt; + v16i8 src0, src1, src2, src3, src4, src5, src6, src7; + v8i16 in0, in1, in2, in3, in4, in5, in6, in7; + v8i16 filt0, filt1; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask1, mask2, mask3; + v16i8 vec0, vec1; + v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7; + v8i16 filter_vec, const_vec; + + src0_ptr -= 1; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + mask1 = mask0 + 2; + mask2 = mask0 + 8; + mask3 = mask0 + 10; + + dst_tmp = dst + 16; + src1_ptr_tmp = src1_ptr + 16; + + for (loop_cnt = (height >> 2); loop_cnt--;) { + LD_SB4(src0_ptr, src_stride, src0, src2, src4, src6); + LD_SB4(src0_ptr + 16, src_stride, src1, src3, src5, src7); + src0_ptr += (4 * src_stride); + LD_SH4(src1_ptr, src2_stride, in0, in2, in4, in6); + LD_SH4(src1_ptr + 8, src2_stride, in1, in3, in5, in7); + src1_ptr += (4 * src2_stride); + XORI_B8_128_SB(src0, src1, src2, src3, src4, src5, src6, src7); + + VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1); + dst0 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0); + VSHF_B2_SB(src0, src1, src0, src1, mask2, mask3, vec0, vec1); + dst1 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1); + VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec0, vec1); + dst2 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2); + VSHF_B2_SB(src2, src3, src2, src3, mask2, mask3, vec0, vec1); + dst3 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3); + VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1); + dst4 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4); + VSHF_B2_SB(src4, src5, src4, src5, mask2, mask3, vec0, vec1); + dst5 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst5, dst5); + VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec0, vec1); + dst6 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst6, dst6); + VSHF_B2_SB(src6, src7, src6, src7, mask2, mask3, vec0, vec1); + dst7 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst7, dst7); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3); + HEVC_BI_RND_CLIP4(in4, in5, in6, in7, + dst4, dst5, dst6, dst7, 7, dst4, dst5, dst6, dst7); + + PCKEV_B4_SH(dst1, dst0, dst3, dst2, + dst5, dst4, dst7, dst6, dst0, dst1, dst2, dst3); + ST_SH4(dst0, dst1, dst2, dst3, dst, dst_stride); + dst += (4 * dst_stride); + + LD_SH4(src1_ptr_tmp, src2_stride, in0, in1, in2, in3); + src1_ptr_tmp += (4 * src2_stride); + + VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec0, vec1); + dst0 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0); + VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1); + dst1 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1); + VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec0, vec1); + dst2 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2); + VSHF_B2_SB(src7, src7, src7, src7, mask0, mask1, vec0, vec1); + dst3 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3); + + PCKEV_B2_SH(dst1, dst0, dst3, dst2, dst0, dst1); + ST8x4_UB(dst0, dst1, dst_tmp, dst_stride); + dst_tmp += (4 * dst_stride); + } +} + +static void hevc_hz_bi_4t_32w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + uint32_t loop_cnt; + v16i8 src0, src1, src2; + v8i16 in0, in1, in2, in3; + v8i16 filt0, filt1; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask1, mask2, mask3; + v8i16 dst0, dst1, dst2, dst3; + v16i8 vec0, vec1; + v8i16 filter_vec, const_vec; + + src0_ptr -= 1; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + mask1 = mask0 + 2; + mask2 = mask0 + 8; + mask3 = mask0 + 10; + + for (loop_cnt = (height >> 1); loop_cnt--;) { + LD_SB2(src0_ptr, 16, src0, src1); + src2 = LD_SB(src0_ptr + 24); + src0_ptr += src_stride; + LD_SH4(src1_ptr, 8, in0, in1, in2, in3); + src1_ptr += src2_stride; + XORI_B3_128_SB(src0, src1, src2); + + VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1); + dst0 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0); + VSHF_B2_SB(src0, src1, src0, src1, mask2, mask3, vec0, vec1); + dst1 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1); + VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec0, vec1); + dst2 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2); + VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec0, vec1); + dst3 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3); + + PCKEV_B2_SH(dst1, dst0, dst3, dst2, dst0, dst1); + ST_SH2(dst0, dst1, dst, 16); + dst += dst_stride; + + LD_SB2(src0_ptr, 16, src0, src1); + src2 = LD_SB(src0_ptr + 24); + src0_ptr += src_stride; + LD_SH4(src1_ptr, 8, in0, in1, in2, in3); + src1_ptr += src2_stride; + XORI_B3_128_SB(src0, src1, src2); + + VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1); + dst0 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0); + VSHF_B2_SB(src0, src1, src0, src1, mask2, mask3, vec0, vec1); + dst1 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1); + VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec0, vec1); + dst2 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2); + VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec0, vec1); + dst3 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0, dst1, dst2, dst3, 7, dst0, dst1, dst2, dst3); + + PCKEV_B2_SH(dst1, dst0, dst3, dst2, dst0, dst1); + ST_SH2(dst0, dst1, dst, 16); + dst += dst_stride; + } +} + +static void hevc_vt_bi_4t_4x2_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + v16i8 src0, src1, src2, src3, src4; + v8i16 in0, in1; + v16i8 src10_r, src32_r, src21_r, src43_r, src2110, src4332; + v8i16 dst10; + v8i16 filt0, filt1; + v8i16 filter_vec, const_vec; + + src0_ptr -= src_stride; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + LD_SB3(src0_ptr, src_stride, src0, src1, src2); + src0_ptr += (3 * src_stride); + + ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r); + src2110 = (v16i8) __msa_ilvr_d((v2i64) src21_r, (v2i64) src10_r); + src2110 = (v16i8) __msa_xori_b((v16u8) src2110, 128); + + LD_SB2(src0_ptr, src_stride, src3, src4); + LD_SH2(src1_ptr, src2_stride, in0, in1); + in0 = (v8i16) __msa_ilvr_d((v2i64) in1, (v2i64) in0); + ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r); + src4332 = (v16i8) __msa_ilvr_d((v2i64) src43_r, (v2i64) src32_r); + src4332 = (v16i8) __msa_xori_b((v16u8) src4332, 128); + + dst10 = const_vec; + DPADD_SB2_SH(src2110, src4332, filt0, filt1, dst10, dst10); + dst10 = __msa_adds_s_h(dst10, in0); + dst10 = __msa_srari_h(dst10, 7); + dst10 = CLIP_SH_0_255(dst10); + + dst10 = (v8i16) __msa_pckev_b((v16i8) dst10, (v16i8) dst10); + ST4x2_UB(dst10, dst, dst_stride); +} + +static void hevc_vt_bi_4t_4x4_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + v16i8 src0, src1, src2, src3, src4, src5, src6; + v8i16 in0, in1, in2, in3; + v16i8 src10_r, src32_r, src54_r, src21_r, src43_r, src65_r; + v16i8 src2110, src4332, src6554; + v8i16 dst10, dst32; + v8i16 filt0, filt1; + v8i16 filter_vec, const_vec; + + src0_ptr -= src_stride; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + LD_SB3(src0_ptr, src_stride, src0, src1, src2); + src0_ptr += (3 * src_stride); + ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r); + src2110 = (v16i8) __msa_ilvr_d((v2i64) src21_r, (v2i64) src10_r); + src2110 = (v16i8) __msa_xori_b((v16u8) src2110, 128); + + LD_SB4(src0_ptr, src_stride, src3, src4, src5, src6); + LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3); + ILVR_D2_SH(in1, in0, in3, in2, in0, in1); + ILVR_B4_SB(src3, src2, src4, src3, src5, src4, src6, src5, + src32_r, src43_r, src54_r, src65_r); + ILVR_D2_SB(src43_r, src32_r, src65_r, src54_r, src4332, src6554); + XORI_B2_128_SB(src4332, src6554); + + dst10 = const_vec; + DPADD_SB2_SH(src2110, src4332, filt0, filt1, dst10, dst10); + dst32 = const_vec; + DPADD_SB2_SH(src4332, src6554, filt0, filt1, dst32, dst32); + HEVC_BI_RND_CLIP2(in0, in1, dst10, dst32, 7, dst10, dst32); + + dst10 = (v8i16) __msa_pckev_b((v16i8) dst32, (v16i8) dst10); + ST4x4_UB(dst10, dst10, 0, 1, 2, 3, dst, dst_stride); +} + +static void hevc_vt_bi_4t_4x8multiple_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + int32_t loop_cnt; + v16i8 src0, src1, src2, src3, src4, src5; + v8i16 in0, in1, in2, in3, in4, in5, in6, in7; + v16i8 src6, src7, src8, src9; + v16i8 src10_r, src32_r, src54_r, src76_r, src98_r; + v16i8 src21_r, src43_r, src65_r, src87_r, src109_r; + v16i8 src2110, src4332, src6554, src8776; + v8i16 dst10, dst32, dst54, dst76; + v8i16 filt0, filt1; + v8i16 filter_vec, const_vec; + + src0_ptr -= src_stride; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + LD_SB3(src0_ptr, src_stride, src0, src1, src2); + src0_ptr += (3 * src_stride); + ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r); + src2110 = (v16i8) __msa_ilvr_d((v2i64) src21_r, (v2i64) src10_r); + src2110 = (v16i8) __msa_xori_b((v16u8) src2110, 128); + + for (loop_cnt = (height >> 3); loop_cnt--;) { + LD_SB6(src0_ptr, src_stride, src3, src4, src5, src6, src7, src8); + src0_ptr += (6 * src_stride); + LD_SH8(src1_ptr, src2_stride, in0, in1, in2, in3, in4, in5, in6, in7); + src1_ptr += (8 * src2_stride); + ILVR_D2_SH(in1, in0, in3, in2, in0, in1); + ILVR_D2_SH(in5, in4, in7, in6, in2, in3); + ILVR_B4_SB(src3, src2, src4, src3, src5, src4, src6, src5, + src32_r, src43_r, src54_r, src65_r); + ILVR_B2_SB(src7, src6, src8, src7, src76_r, src87_r); + ILVR_D3_SB(src43_r, src32_r, src65_r, src54_r, src87_r, src76_r, + src4332, src6554, src8776); + XORI_B3_128_SB(src4332, src6554, src8776); + + dst10 = const_vec; + DPADD_SB2_SH(src2110, src4332, filt0, filt1, dst10, dst10); + dst32 = const_vec; + DPADD_SB2_SH(src4332, src6554, filt0, filt1, dst32, dst32); + dst54 = const_vec; + DPADD_SB2_SH(src6554, src8776, filt0, filt1, dst54, dst54); + + LD_SB2(src0_ptr, src_stride, src9, src2); + src0_ptr += (2 * src_stride); + ILVR_B2_SB(src9, src8, src2, src9, src98_r, src109_r); + src2110 = (v16i8) __msa_ilvr_d((v2i64) src109_r, (v2i64) src98_r); + src2110 = (v16i8) __msa_xori_b((v16u8) src2110, 128); + dst76 = const_vec; + DPADD_SB2_SH(src8776, src2110, filt0, filt1, dst76, dst76); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst10, dst32, dst54, dst76, 7, + dst10, dst32, dst54, dst76); + + PCKEV_B2_SH(dst32, dst10, dst76, dst54, dst10, dst54); + ST4x8_UB(dst10, dst54, dst, dst_stride); + dst += (8 * dst_stride); + } +} + +static void hevc_vt_bi_4t_4w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + if (2 == height) { + hevc_vt_bi_4t_4x2_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter, height); + } else if (4 == height) { + hevc_vt_bi_4t_4x4_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter, height); + } else { + hevc_vt_bi_4t_4x8multiple_msa(src0_ptr, src_stride, + src1_ptr, src2_stride, + dst, dst_stride, filter, height); + } +} + +static void hevc_vt_bi_4t_6w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + int32_t loop_cnt; + v16i8 src0, src1, src2, src3, src4, src5; + v8i16 in0, in1, in2, in3; + v16i8 src10_r, src32_r, src21_r, src43_r; + v8i16 dst0_r, dst1_r, dst2_r, dst3_r; + v8i16 filt0, filt1; + v8i16 filter_vec, const_vec; + + src0_ptr -= src_stride; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + LD_SB3(src0_ptr, src_stride, src0, src1, src2); + src0_ptr += (3 * src_stride); + XORI_B3_128_SB(src0, src1, src2); + ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r); + + for (loop_cnt = (height >> 2); loop_cnt--;) { + LD_SB2(src0_ptr, src_stride, src3, src4); + src0_ptr += (2 * src_stride); + LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3); + src1_ptr += (4 * src2_stride); + XORI_B2_128_SB(src3, src4); + ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r); + + dst0_r = const_vec; + DPADD_SB2_SH(src10_r, src32_r, filt0, filt1, dst0_r, dst0_r); + dst1_r = const_vec; + DPADD_SB2_SH(src21_r, src43_r, filt0, filt1, dst1_r, dst1_r); + + LD_SB2(src0_ptr, src_stride, src5, src2); + src0_ptr += (2 * src_stride); + XORI_B2_128_SB(src5, src2); + ILVR_B2_SB(src5, src4, src2, src5, src10_r, src21_r); + + dst2_r = const_vec; + DPADD_SB2_SH(src32_r, src10_r, filt0, filt1, dst2_r, dst2_r); + dst3_r = const_vec; + DPADD_SB2_SH(src43_r, src21_r, filt0, filt1, dst3_r, dst3_r); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0_r, dst1_r, dst2_r, dst3_r, 7, + dst0_r, dst1_r, dst2_r, dst3_r); + + PCKEV_B2_SH(dst1_r, dst0_r, dst3_r, dst2_r, dst0_r, dst1_r); + ST6x4_UB(dst0_r, dst1_r, dst, dst_stride); + dst += (4 * dst_stride); + } +} + +static void hevc_vt_bi_4t_8x2_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + v16i8 src0, src1, src2, src3, src4; + v8i16 in0, in1, dst0_r, dst1_r; + v16i8 src10_r, src32_r, src21_r, src43_r; + v8i16 filt0, filt1; + v8i16 filter_vec, const_vec; + + src0_ptr -= src_stride; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + LD_SB3(src0_ptr, src_stride, src0, src1, src2); + src0_ptr += (3 * src_stride); + XORI_B3_128_SB(src0, src1, src2); + ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r); + + LD_SB2(src0_ptr, src_stride, src3, src4); + LD_SH2(src1_ptr, src2_stride, in0, in1); + XORI_B2_128_SB(src3, src4); + ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r); + + dst0_r = const_vec; + DPADD_SB2_SH(src10_r, src32_r, filt0, filt1, dst0_r, dst0_r); + dst1_r = const_vec; + DPADD_SB2_SH(src21_r, src43_r, filt0, filt1, dst1_r, dst1_r); + + HEVC_BI_RND_CLIP2(in0, in1, dst0_r, dst1_r, 7, dst0_r, dst1_r); + dst0_r = (v8i16) __msa_pckev_b((v16i8) dst1_r, (v16i8) dst0_r); + + ST8x2_UB(dst0_r, dst, dst_stride); +} + +static void hevc_vt_bi_4t_8x6_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8; + v8i16 in0, in1, in2, in3, in4, in5; + v16i8 src10_r, src32_r, src54_r, src76_r; + v16i8 src21_r, src43_r, src65_r, src87_r; + v8i16 dst0_r, dst1_r, dst2_r, dst3_r, dst4_r, dst5_r; + v8i16 filt0, filt1; + v8i16 filter_vec, const_vec; + + src0_ptr -= src_stride; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + LD_SB3(src0_ptr, src_stride, src0, src1, src2); + src0_ptr += (3 * src_stride); + XORI_B3_128_SB(src0, src1, src2); + ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r); + + LD_SB6(src0_ptr, src_stride, src3, src4, src5, src6, src7, src8); + LD_SH6(src1_ptr, src2_stride, in0, in1, in2, in3, in4, in5); + XORI_B6_128_SB(src3, src4, src5, src6, src7, src8); + ILVR_B4_SB(src3, src2, src4, src3, src5, src4, src6, src5, + src32_r, src43_r, src54_r, src65_r); + ILVR_B2_SB(src7, src6, src8, src7, src76_r, src87_r); + + dst0_r = const_vec; + DPADD_SB2_SH(src10_r, src32_r, filt0, filt1, dst0_r, dst0_r); + dst1_r = const_vec; + DPADD_SB2_SH(src21_r, src43_r, filt0, filt1, dst1_r, dst1_r); + dst2_r = const_vec; + DPADD_SB2_SH(src32_r, src54_r, filt0, filt1, dst2_r, dst2_r); + dst3_r = const_vec; + DPADD_SB2_SH(src43_r, src65_r, filt0, filt1, dst3_r, dst3_r); + dst4_r = const_vec; + DPADD_SB2_SH(src54_r, src76_r, filt0, filt1, dst4_r, dst4_r); + dst5_r = const_vec; + DPADD_SB2_SH(src65_r, src87_r, filt0, filt1, dst5_r, dst5_r); + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0_r, dst1_r, dst2_r, dst3_r, 7, + dst0_r, dst1_r, dst2_r, dst3_r); + HEVC_BI_RND_CLIP2(in4, in5, dst4_r, dst5_r, 7, dst4_r, dst5_r); + + PCKEV_B2_SH(dst1_r, dst0_r, dst3_r, dst2_r, dst0_r, dst1_r); + dst2_r = (v8i16) __msa_pckev_b((v16i8) dst5_r, (v16i8) dst4_r); + ST8x4_UB(dst0_r, dst1_r, dst, dst_stride); + dst += (4 * dst_stride); + ST8x2_UB(dst2_r, dst, dst_stride); +} + +static void hevc_vt_bi_4t_8x4multiple_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + int32_t loop_cnt; + v16i8 src0, src1, src2, src3, src4, src5; + v8i16 in0, in1, in2, in3; + v16i8 src10_r, src32_r, src21_r, src43_r; + v8i16 dst0_r, dst1_r, dst2_r, dst3_r; + v8i16 filt0, filt1; + v8i16 filter_vec, const_vec; + + src0_ptr -= src_stride; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + LD_SB3(src0_ptr, src_stride, src0, src1, src2); + src0_ptr += (3 * src_stride); + XORI_B3_128_SB(src0, src1, src2); + ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r); + + for (loop_cnt = (height >> 2); loop_cnt--;) { + LD_SB2(src0_ptr, src_stride, src3, src4); + src0_ptr += (2 * src_stride); + LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3); + src1_ptr += (4 * src2_stride); + XORI_B2_128_SB(src3, src4); + ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r); + + dst0_r = const_vec; + DPADD_SB2_SH(src10_r, src32_r, filt0, filt1, dst0_r, dst0_r); + dst1_r = const_vec; + DPADD_SB2_SH(src21_r, src43_r, filt0, filt1, dst1_r, dst1_r); + + LD_SB2(src0_ptr, src_stride, src5, src2); + src0_ptr += (2 * src_stride); + XORI_B2_128_SB(src5, src2); + ILVR_B2_SB(src5, src4, src2, src5, src10_r, src21_r); + + dst2_r = const_vec; + DPADD_SB2_SH(src32_r, src10_r, filt0, filt1, dst2_r, dst2_r); + dst3_r = const_vec; + DPADD_SB2_SH(src43_r, src21_r, filt0, filt1, dst3_r, dst3_r); + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0_r, dst1_r, dst2_r, dst3_r, 7, + dst0_r, dst1_r, dst2_r, dst3_r); + + PCKEV_B2_SH(dst1_r, dst0_r, dst3_r, dst2_r, dst0_r, dst1_r); + ST8x4_UB(dst0_r, dst1_r, dst, dst_stride); + dst += (4 * dst_stride); + } +} + +static void hevc_vt_bi_4t_8w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + if (2 == height) { + hevc_vt_bi_4t_8x2_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter, height); + } else if (6 == height) { + hevc_vt_bi_4t_8x6_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter, height); + } else { + hevc_vt_bi_4t_8x4multiple_msa(src0_ptr, src_stride, + src1_ptr, src2_stride, + dst, dst_stride, filter, height); + } +} + +static void hevc_vt_bi_4t_12w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + int32_t loop_cnt; + v16i8 src0, src1, src2, src3, src4, src5; + v8i16 in0, in1, in2, in3, in4, in5, in6, in7; + v16i8 src10_r, src32_r, src21_r, src43_r; + v8i16 dst0_r, dst1_r, dst2_r, dst3_r; + v16i8 src10_l, src32_l, src54_l, src21_l, src43_l, src65_l; + v16i8 src2110, src4332; + v8i16 dst0_l, dst1_l, filt0, filt1; + v8i16 filter_vec, const_vec; + + src0_ptr -= (1 * src_stride); + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + LD_SB3(src0_ptr, src_stride, src0, src1, src2); + src0_ptr += (3 * src_stride); + XORI_B3_128_SB(src0, src1, src2); + ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r); + ILVL_B2_SB(src1, src0, src2, src1, src10_l, src21_l); + src2110 = (v16i8) __msa_ilvr_d((v2i64) src21_l, (v2i64) src10_l); + + for (loop_cnt = (height >> 2); loop_cnt--;) { + LD_SB2(src0_ptr, src_stride, src3, src4); + src0_ptr += (2 * src_stride); + LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3); + LD_SH4((src1_ptr + 8), src2_stride, in4, in5, in6, in7); + src1_ptr += (4 * src2_stride); + ILVR_D2_SH(in5, in4, in7, in6, in4, in5); + XORI_B2_128_SB(src3, src4); + + ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r); + ILVL_B2_SB(src3, src2, src4, src3, src32_l, src43_l); + src4332 = (v16i8) __msa_ilvr_d((v2i64) src43_l, (v2i64) src32_l); + + dst0_r = const_vec; + DPADD_SB2_SH(src10_r, src32_r, filt0, filt1, dst0_r, dst0_r); + dst1_r = const_vec; + DPADD_SB2_SH(src21_r, src43_r, filt0, filt1, dst1_r, dst1_r); + dst0_l = const_vec; + DPADD_SB2_SH(src2110, src4332, filt0, filt1, dst0_l, dst0_l); + + LD_SB2(src0_ptr, src_stride, src5, src2); + src0_ptr += (2 * src_stride); + XORI_B2_128_SB(src5, src2); + + ILVR_B2_SB(src5, src4, src2, src5, src10_r, src21_r); + ILVL_B2_SB(src5, src4, src2, src5, src54_l, src65_l); + src2110 = (v16i8) __msa_ilvr_d((v2i64) src65_l, (v2i64) src54_l); + + dst2_r = const_vec; + DPADD_SB2_SH(src32_r, src10_r, filt0, filt1, dst2_r, dst2_r); + dst3_r = const_vec; + DPADD_SB2_SH(src43_r, src21_r, filt0, filt1, dst3_r, dst3_r); + dst1_l = const_vec; + DPADD_SB2_SH(src4332, src2110, filt0, filt1, dst1_l, dst1_l); + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0_r, dst1_r, dst2_r, dst3_r, 7, + dst0_r, dst1_r, dst2_r, dst3_r); + HEVC_BI_RND_CLIP2(in4, in5, dst0_l, dst1_l, 7, dst0_l, dst1_l); + + PCKEV_B2_SH(dst1_r, dst0_r, dst3_r, dst2_r, dst0_r, dst1_r); + dst0_l = (v8i16) __msa_pckev_b((v16i8) dst1_l, (v16i8) dst0_l); + ST12x4_UB(dst0_r, dst1_r, dst0_l, dst, dst_stride); + dst += (4 * dst_stride); + } +} + +static void hevc_vt_bi_4t_16w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + int32_t loop_cnt; + v16i8 src0, src1, src2, src3, src4, src5; + v8i16 in0, in1, in2, in3; + v16i8 src10_r, src32_r, src21_r, src43_r; + v16i8 src10_l, src32_l, src21_l, src43_l; + v8i16 dst0_r, dst1_r, dst0_l, dst1_l; + v8i16 filt0, filt1; + v8i16 filter_vec, const_vec; + + src0_ptr -= src_stride; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + LD_SB3(src0_ptr, src_stride, src0, src1, src2); + src0_ptr += (3 * src_stride); + XORI_B3_128_SB(src0, src1, src2); + ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r); + ILVL_B2_SB(src1, src0, src2, src1, src10_l, src21_l); + + for (loop_cnt = (height >> 2); loop_cnt--;) { + LD_SB2(src0_ptr, src_stride, src3, src4); + src0_ptr += (2 * src_stride); + LD_SH2(src1_ptr, src2_stride, in0, in1); + LD_SH2((src1_ptr + 8), src2_stride, in2, in3); + src1_ptr += (2 * src2_stride); + XORI_B2_128_SB(src3, src4); + ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r); + ILVL_B2_SB(src3, src2, src4, src3, src32_l, src43_l); + + dst0_r = const_vec; + DPADD_SB2_SH(src10_r, src32_r, filt0, filt1, dst0_r, dst0_r); + dst1_r = const_vec; + DPADD_SB2_SH(src21_r, src43_r, filt0, filt1, dst1_r, dst1_r); + dst0_l = const_vec; + DPADD_SB2_SH(src10_l, src32_l, filt0, filt1, dst0_l, dst0_l); + dst1_l = const_vec; + DPADD_SB2_SH(src21_l, src43_l, filt0, filt1, dst1_l, dst1_l); + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0_r, dst1_r, dst0_l, dst1_l, 7, + dst0_r, dst1_r, dst0_l, dst1_l); + + PCKEV_B2_SH(dst0_l, dst0_r, dst1_l, dst1_r, dst0_r, dst1_r); + ST_SH2(dst0_r, dst1_r, dst, dst_stride); + dst += (2 * dst_stride); + + LD_SB2(src0_ptr, src_stride, src5, src2); + src0_ptr += (2 * src_stride); + LD_SH2(src1_ptr, src2_stride, in0, in1); + LD_SH2((src1_ptr + 8), src2_stride, in2, in3); + src1_ptr += (2 * src2_stride); + XORI_B2_128_SB(src5, src2); + ILVR_B2_SB(src5, src4, src2, src5, src10_r, src21_r); + ILVL_B2_SB(src5, src4, src2, src5, src10_l, src21_l); + + dst0_r = const_vec; + DPADD_SB2_SH(src32_r, src10_r, filt0, filt1, dst0_r, dst0_r); + dst0_l = const_vec; + DPADD_SB2_SH(src32_l, src10_l, filt0, filt1, dst0_l, dst0_l); + dst1_r = const_vec; + DPADD_SB2_SH(src43_r, src21_r, filt0, filt1, dst1_r, dst1_r); + dst1_l = const_vec; + DPADD_SB2_SH(src43_l, src21_l, filt0, filt1, dst1_l, dst1_l); + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0_r, dst1_r, dst0_l, dst1_l, 7, + dst0_r, dst1_r, dst0_l, dst1_l); + + PCKEV_B2_SH(dst0_l, dst0_r, dst1_l, dst1_r, dst0_r, dst1_r); + ST_SH2(dst0_r, dst1_r, dst, dst_stride); + dst += (2 * dst_stride); + } +} + +static void hevc_vt_bi_4t_24w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + uint32_t loop_cnt; + v16i8 src0, src1, src2, src3, src4, src5; + v16i8 src6, src7, src8, src9, src10, src11; + v8i16 in0, in1, in2, in3, in4, in5; + v16i8 src10_r, src32_r, src76_r, src98_r; + v16i8 src21_r, src43_r, src87_r, src109_r; + v16i8 src10_l, src32_l, src21_l, src43_l; + v8i16 dst0_r, dst1_r, dst2_r, dst3_r; + v8i16 dst0_l, dst1_l; + v8i16 filt0, filt1; + v8i16 filter_vec, const_vec; + + src0_ptr -= src_stride; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + /* 16width */ + LD_SB3(src0_ptr, src_stride, src0, src1, src2); + XORI_B3_128_SB(src0, src1, src2); + ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r); + ILVL_B2_SB(src1, src0, src2, src1, src10_l, src21_l); + /* 8width */ + LD_SB3(src0_ptr + 16, src_stride, src6, src7, src8); + src0_ptr += (3 * src_stride); + XORI_B3_128_SB(src6, src7, src8); + ILVR_B2_SB(src7, src6, src8, src7, src76_r, src87_r); + + for (loop_cnt = (height >> 2); loop_cnt--;) { + /* 16width */ + LD_SB2(src0_ptr, src_stride, src3, src4); + LD_SH2(src1_ptr, src2_stride, in0, in1); + LD_SH2((src1_ptr + 8), src2_stride, in2, in3); + LD_SH2((src1_ptr + 16), src2_stride, in4, in5); + src1_ptr += (2 * src2_stride); + XORI_B2_128_SB(src3, src4); + ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r); + ILVL_B2_SB(src3, src2, src4, src3, src32_l, src43_l); + /* 8width */ + LD_SB2(src0_ptr + 16, src_stride, src9, src10); + src0_ptr += (2 * src_stride); + XORI_B2_128_SB(src9, src10); + ILVR_B2_SB(src9, src8, src10, src9, src98_r, src109_r); + /* 16width */ + dst0_r = const_vec; + DPADD_SB2_SH(src10_r, src32_r, filt0, filt1, dst0_r, dst0_r); + dst0_l = const_vec; + DPADD_SB2_SH(src10_l, src32_l, filt0, filt1, dst0_l, dst0_l); + dst1_r = const_vec; + DPADD_SB2_SH(src21_r, src43_r, filt0, filt1, dst1_r, dst1_r); + dst1_l = const_vec; + DPADD_SB2_SH(src21_l, src43_l, filt0, filt1, dst1_l, dst1_l); + /* 8width */ + dst2_r = const_vec; + DPADD_SB2_SH(src76_r, src98_r, filt0, filt1, dst2_r, dst2_r); + dst3_r = const_vec; + DPADD_SB2_SH(src87_r, src109_r, filt0, filt1, dst3_r, dst3_r); + /* 16width */ + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0_r, dst1_r, dst0_l, dst1_l, 7, + dst0_r, dst1_r, dst0_l, dst1_l); + + HEVC_BI_RND_CLIP2(in4, in5, dst2_r, dst3_r, 7, dst2_r, dst3_r); + + PCKEV_B2_SH(dst0_l, dst0_r, dst1_l, dst1_r, dst0_r, dst1_r); + dst2_r = (v8i16) __msa_pckev_b((v16i8) dst3_r, (v16i8) dst2_r); + ST_SH2(dst0_r, dst1_r, dst, dst_stride); + ST8x2_UB(dst2_r, dst + 16, dst_stride); + dst += (2 * dst_stride); + + /* 16width */ + LD_SB2(src0_ptr, src_stride, src5, src2); + LD_SH2(src1_ptr, src2_stride, in0, in1); + LD_SH2((src1_ptr + 8), src2_stride, in2, in3); + LD_SH2((src1_ptr + 16), src2_stride, in4, in5); + src1_ptr += (2 * src2_stride); + XORI_B2_128_SB(src5, src2); + ILVR_B2_SB(src5, src4, src2, src5, src10_r, src21_r); + ILVL_B2_SB(src5, src4, src2, src5, src10_l, src21_l); + /* 8width */ + LD_SB2(src0_ptr + 16, src_stride, src11, src8); + src0_ptr += (2 * src_stride); + XORI_B2_128_SB(src11, src8); + ILVR_B2_SB(src11, src10, src8, src11, src76_r, src87_r); + /* 16width */ + dst0_r = const_vec; + DPADD_SB2_SH(src32_r, src10_r, filt0, filt1, dst0_r, dst0_r); + dst0_l = const_vec; + DPADD_SB2_SH(src32_l, src10_l, filt0, filt1, dst0_l, dst0_l); + dst1_r = const_vec; + DPADD_SB2_SH(src43_r, src21_r, filt0, filt1, dst1_r, dst1_r); + dst1_l = const_vec; + DPADD_SB2_SH(src43_l, src21_l, filt0, filt1, dst1_l, dst1_l); + /* 8width */ + dst2_r = const_vec; + DPADD_SB2_SH(src98_r, src76_r, filt0, filt1, dst2_r, dst2_r); + dst3_r = const_vec; + DPADD_SB2_SH(src109_r, src87_r, filt0, filt1, dst3_r, dst3_r); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0_r, dst1_r, dst0_l, dst1_l, 7, + dst0_r, dst1_r, dst0_l, dst1_l); + HEVC_BI_RND_CLIP2(in4, in5, dst2_r, dst3_r, 7, dst2_r, dst3_r); + + PCKEV_B2_SH(dst0_l, dst0_r, dst1_l, dst1_r, dst0_r, dst1_r); + dst2_r = (v8i16) __msa_pckev_b((v16i8) dst3_r, (v16i8) dst2_r); + ST_SH2(dst0_r, dst1_r, dst, dst_stride); + ST8x2_UB(dst2_r, dst + 16, dst_stride); + dst += (2 * dst_stride); + } +} + +static void hevc_vt_bi_4t_32w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) +{ + uint32_t loop_cnt; + uint8_t *dst_tmp = dst + 16; + v16i8 src0, src1, src2, src3, src4, src6, src7, src8, src9, src10; + v8i16 in0, in1, in2, in3, in4, in5, in6, in7; + v16i8 src10_r, src32_r, src76_r, src98_r; + v16i8 src21_r, src43_r, src87_r, src109_r; + v8i16 dst0_r, dst1_r, dst2_r, dst3_r; + v16i8 src10_l, src32_l, src76_l, src98_l; + v16i8 src21_l, src43_l, src87_l, src109_l; + v8i16 dst0_l, dst1_l, dst2_l, dst3_l; + v8i16 filt0, filt1; + v8i16 filter_vec, const_vec; + + src0_ptr -= src_stride; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + /* 16width */ + LD_SB3(src0_ptr, src_stride, src0, src1, src2); + XORI_B3_128_SB(src0, src1, src2); + ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r); + ILVL_B2_SB(src1, src0, src2, src1, src10_l, src21_l); + + /* next 16width */ + LD_SB3(src0_ptr + 16, src_stride, src6, src7, src8); + src0_ptr += (3 * src_stride); + XORI_B3_128_SB(src6, src7, src8); + ILVR_B2_SB(src7, src6, src8, src7, src76_r, src87_r); + ILVL_B2_SB(src7, src6, src8, src7, src76_l, src87_l); + + for (loop_cnt = (height >> 1); loop_cnt--;) { + /* 16width */ + LD_SB2(src0_ptr, src_stride, src3, src4); + LD_SH2(src1_ptr, src2_stride, in0, in1); + LD_SH2((src1_ptr + 8), src2_stride, in2, in3); + LD_SH2((src1_ptr + 16), src2_stride, in4, in5); + LD_SH2((src1_ptr + 24), src2_stride, in6, in7); + src1_ptr += (2 * src2_stride); + XORI_B2_128_SB(src3, src4); + ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r); + ILVL_B2_SB(src3, src2, src4, src3, src32_l, src43_l); + /* 16width */ + dst0_r = const_vec; + DPADD_SB2_SH(src10_r, src32_r, filt0, filt1, dst0_r, dst0_r); + dst0_l = const_vec; + DPADD_SB2_SH(src10_l, src32_l, filt0, filt1, dst0_l, dst0_l); + dst1_r = const_vec; + DPADD_SB2_SH(src21_r, src43_r, filt0, filt1, dst1_r, dst1_r); + dst1_l = const_vec; + DPADD_SB2_SH(src21_l, src43_l, filt0, filt1, dst1_l, dst1_l); + /* 16width */ + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + dst0_r, dst1_r, dst0_l, dst1_l, 7, + dst0_r, dst1_r, dst0_l, dst1_l); + + src10_r = src32_r; + src21_r = src43_r; + src10_l = src32_l; + src21_l = src43_l; + src2 = src4; + + PCKEV_B2_SH(dst0_l, dst0_r, dst1_l, dst1_r, dst0_r, dst1_r); + ST_SH2(dst0_r, dst1_r, dst, dst_stride); + dst += (2 * dst_stride); + + /* next 16width */ + LD_SB2(src0_ptr + 16, src_stride, src9, src10); + src0_ptr += (2 * src_stride); + XORI_B2_128_SB(src9, src10); + ILVR_B2_SB(src9, src8, src10, src9, src98_r, src109_r); + ILVL_B2_SB(src9, src8, src10, src9, src98_l, src109_l); + /* next 16width */ + dst2_r = const_vec; + DPADD_SB2_SH(src76_r, src98_r, filt0, filt1, dst2_r, dst2_r); + dst2_l = const_vec; + DPADD_SB2_SH(src76_l, src98_l, filt0, filt1, dst2_l, dst2_l); + dst3_r = const_vec; + DPADD_SB2_SH(src87_r, src109_r, filt0, filt1, dst3_r, dst3_r); + dst3_l = const_vec; + DPADD_SB2_SH(src87_l, src109_l, filt0, filt1, dst3_l, dst3_l); + /* next 16width */ + HEVC_BI_RND_CLIP4(in4, in5, in6, in7, + dst2_r, dst3_r, dst2_l, dst3_l, 7, + dst2_r, dst3_r, dst2_l, dst3_l); + + PCKEV_B2_SH(dst2_l, dst2_r, dst3_l, dst3_r, dst2_r, dst3_r); + ST_SH2(dst2_r, dst3_r, dst_tmp, dst_stride); + dst_tmp += (2 * dst_stride); + + src76_r = src98_r; + src87_r = src109_r; + src76_l = src98_l; + src87_l = src109_l; + src8 = src10; + } +} + +static void hevc_hv_bi_4t_4x2_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter_x, + const int8_t *filter_y, + int32_t height) +{ + v8i16 in0, in1; + v16i8 src0, src1, src2, src3, src4; + v8i16 filt0, filt1; + v4i32 filt_h0, filt_h1; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask1; + v8i16 filter_vec, const_vec; + v16i8 vec0, vec1, vec2, vec3, vec4, vec5; + v8i16 dst0, dst1, dst2, dst3, dst4; + v4i32 dst0_r, dst1_r; + v8i16 dst10_r, dst32_r, dst21_r, dst43_r; + + src0_ptr -= (src_stride + 1); + + filter_vec = LD_SH(filter_x); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + filter_vec = LD_SH(filter_y); + vec0 = __msa_clti_s_b((v16i8) filter_vec, 0); + filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec); + + SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1); + + mask1 = mask0 + 2; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + LD_SB3(src0_ptr, src_stride, src0, src1, src2); + src0_ptr += (3 * src_stride); + XORI_B3_128_SB(src0, src1, src2); + + VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1); + VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3); + VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5); + dst0 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0); + dst1 = const_vec; + DPADD_SB2_SH(vec2, vec3, filt0, filt1, dst1, dst1); + dst2 = const_vec; + DPADD_SB2_SH(vec4, vec5, filt0, filt1, dst2, dst2); + ILVR_H2_SH(dst1, dst0, dst2, dst1, dst10_r, dst21_r); + + LD_SB2(src0_ptr, src_stride, src3, src4); + LD_SH2(src1_ptr, src2_stride, in0, in1); + in0 = (v8i16) __msa_ilvr_d((v2i64) in1, (v2i64) in0); + XORI_B2_128_SB(src3, src4); + /* row 3 */ + VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1); + dst3 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3); + dst32_r = __msa_ilvr_h(dst3, dst2); + dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1); + dst0_r >>= 6; + /* row 4 */ + VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1); + dst4 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4); + dst43_r = __msa_ilvr_h(dst4, dst3); + dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1); + dst1_r >>= 6; + dst0_r = (v4i32) __msa_pckev_h((v8i16) dst1_r, (v8i16) dst0_r); + dst0_r = (v4i32) __msa_adds_s_h((v8i16) dst0_r, in0); + dst0_r = (v4i32) __msa_srari_h((v8i16) dst0_r, 7); + dst0_r = (v4i32) CLIP_SH_0_255(dst0_r); + + dst0_r = (v4i32) __msa_pckev_b((v16i8) dst0_r, (v16i8) dst0_r); + ST4x2_UB(dst0_r, dst, dst_stride); +} + +static void hevc_hv_bi_4t_4x4_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter_x, + const int8_t *filter_y, + int32_t height) +{ + v8i16 in0, in1, in2, in3; + v16i8 src0, src1, src2, src3, src4, src5, src6; + v8i16 filt0, filt1; + v4i32 filt_h0, filt_h1; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask1; + v8i16 filter_vec, const_vec; + v16i8 vec0, vec1, vec2, vec3, vec4, vec5; + v8i16 dst0, dst1, dst2, dst3, dst4, dst5; + v8i16 dst0_r, dst1_r; + v4i32 tmp0, tmp1, tmp2, tmp3; + v8i16 dst10_r, dst32_r, dst21_r, dst43_r; + + src0_ptr -= (src_stride + 1); + + filter_vec = LD_SH(filter_x); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + filter_vec = LD_SH(filter_y); + vec0 = __msa_clti_s_b((v16i8) filter_vec, 0); + filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec); + + SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1); + + mask1 = mask0 + 2; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + LD_SB3(src0_ptr, src_stride, src0, src1, src2); + src0_ptr += (3 * src_stride); + XORI_B3_128_SB(src0, src1, src2); + + VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1); + VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3); + VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5); + dst0 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0); + dst1 = const_vec; + DPADD_SB2_SH(vec2, vec3, filt0, filt1, dst1, dst1); + dst2 = const_vec; + DPADD_SB2_SH(vec4, vec5, filt0, filt1, dst2, dst2); + ILVR_H2_SH(dst1, dst0, dst2, dst1, dst10_r, dst21_r); + + LD_SB4(src0_ptr, src_stride, src3, src4, src5, src6); + LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3); + ILVR_D2_SH(in1, in0, in3, in2, in0, in1); + XORI_B4_128_SB(src3, src4, src5, src6); + /* row 3 */ + VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1); + dst3 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3); + dst32_r = __msa_ilvr_h(dst3, dst2); + tmp0 = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1); + tmp0 >>= 6; + /* row 4 */ + VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1); + dst4 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4); + dst43_r = __msa_ilvr_h(dst4, dst3); + tmp1 = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1); + tmp1 >>= 6; + /* row 5 */ + VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec0, vec1); + dst5 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst5, dst5); + dst10_r = __msa_ilvr_h(dst5, dst4); + tmp2 = HEVC_FILT_4TAP(dst32_r, dst10_r, filt_h0, filt_h1); + tmp2 >>= 6; + /* row 6 */ + VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec0, vec1); + dst2 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2); + dst21_r = __msa_ilvr_h(dst2, dst5); + tmp3 = HEVC_FILT_4TAP(dst43_r, dst21_r, filt_h0, filt_h1); + tmp3 >>= 6; + PCKEV_H2_SH(tmp1, tmp0, tmp3, tmp2, dst0_r, dst1_r); + HEVC_BI_RND_CLIP2(in0, in1, dst0_r, dst1_r, 7, dst0_r, dst1_r); + + dst0_r = (v8i16) __msa_pckev_b((v16i8) dst1_r, (v16i8) dst0_r); + ST4x4_UB(dst0_r, dst0_r, 0, 1, 2, 3, dst, dst_stride); + dst += (4 * dst_stride); +} + +static void hevc_hv_bi_4t_4multx8mult_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter_x, + const int8_t *filter_y, + int32_t height) +{ + uint32_t loop_cnt; + v8i16 in0, in1, in2, in3, in4, in5, in6, in7; + v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10; + v8i16 filt0, filt1; + v4i32 filt_h0, filt_h1; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask1; + v8i16 filter_vec, const_vec; + v16i8 vec0, vec1, vec2, vec3, vec4, vec5; + v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8, dst9; + v8i16 tmp0, tmp1, tmp2, tmp3; + v8i16 dst10_r, dst32_r, dst54_r, dst76_r; + v8i16 dst21_r, dst43_r, dst65_r, dst87_r; + v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst4_r, dst5_r, dst6_r, dst7_r; + + src0_ptr -= (src_stride + 1); + + filter_vec = LD_SH(filter_x); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + filter_vec = LD_SH(filter_y); + vec0 = __msa_clti_s_b((v16i8) filter_vec, 0); + filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec); + + SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1); + + mask1 = mask0 + 2; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + LD_SB3(src0_ptr, src_stride, src0, src1, src2); + src0_ptr += (3 * src_stride); + XORI_B3_128_SB(src0, src1, src2); + + VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1); + VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3); + VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5); + dst0 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0); + dst1 = const_vec; + DPADD_SB2_SH(vec2, vec3, filt0, filt1, dst1, dst1); + dst2 = const_vec; + DPADD_SB2_SH(vec4, vec5, filt0, filt1, dst2, dst2); + ILVR_H2_SH(dst1, dst0, dst2, dst1, dst10_r, dst21_r); + + for (loop_cnt = height >> 3; loop_cnt--;) { + LD_SB8(src0_ptr, src_stride, + src3, src4, src5, src6, src7, src8, src9, src10); + src0_ptr += (8 * src_stride); + LD_SH8(src1_ptr, src2_stride, in0, in1, in2, in3, in4, in5, in6, in7); + src1_ptr += (8 * src2_stride); + ILVR_D2_SH(in1, in0, in3, in2, in0, in1); + ILVR_D2_SH(in5, in4, in7, in6, in2, in3); + XORI_B8_128_SB(src3, src4, src5, src6, src7, src8, src9, src10); + /* row 3 */ + VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1); + dst3 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3); + dst32_r = __msa_ilvr_h(dst3, dst2); + dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1); + dst0_r >>= 6; + /* row 4 */ + VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1); + dst4 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4); + dst43_r = __msa_ilvr_h(dst4, dst3); + dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1); + dst1_r >>= 6; + /* row 5 */ + VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec0, vec1); + dst5 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst5, dst5); + dst54_r = __msa_ilvr_h(dst5, dst4); + dst2_r = HEVC_FILT_4TAP(dst32_r, dst54_r, filt_h0, filt_h1); + dst2_r >>= 6; + /* row 6 */ + VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec0, vec1); + dst6 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst6, dst6); + dst65_r = __msa_ilvr_h(dst6, dst5); + dst3_r = HEVC_FILT_4TAP(dst43_r, dst65_r, filt_h0, filt_h1); + dst3_r >>= 6; + /* row 7 */ + VSHF_B2_SB(src7, src7, src7, src7, mask0, mask1, vec0, vec1); + dst7 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst7, dst7); + dst76_r = __msa_ilvr_h(dst7, dst6); + dst4_r = HEVC_FILT_4TAP(dst54_r, dst76_r, filt_h0, filt_h1); + dst4_r >>= 6; + /* row 8 */ + VSHF_B2_SB(src8, src8, src8, src8, mask0, mask1, vec0, vec1); + dst8 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst8, dst8); + dst87_r = __msa_ilvr_h(dst8, dst7); + dst5_r = HEVC_FILT_4TAP(dst65_r, dst87_r, filt_h0, filt_h1); + dst5_r >>= 6; + /* row 9 */ + VSHF_B2_SB(src9, src9, src9, src9, mask0, mask1, vec0, vec1); + dst9 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst9, dst9); + dst10_r = __msa_ilvr_h(dst9, dst8); + dst6_r = HEVC_FILT_4TAP(dst76_r, dst10_r, filt_h0, filt_h1); + dst6_r >>= 6; + /* row 10 */ + VSHF_B2_SB(src10, src10, src10, src10, mask0, mask1, vec0, vec1); + dst2 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2); + dst21_r = __msa_ilvr_h(dst2, dst9); + dst7_r = HEVC_FILT_4TAP(dst87_r, dst21_r, filt_h0, filt_h1); + dst7_r >>= 6; + PCKEV_H4_SH(dst1_r, dst0_r, dst3_r, dst2_r, + dst5_r, dst4_r, dst7_r, dst6_r, tmp0, tmp1, tmp2, tmp3); + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + tmp0, tmp1, tmp2, tmp3, 7, tmp0, tmp1, tmp2, tmp3); + + PCKEV_B2_SH(tmp1, tmp0, tmp3, tmp2, tmp0, tmp1); + ST4x8_UB(tmp0, tmp1, dst, dst_stride); + dst += (8 * dst_stride); + } +} + +static void hevc_hv_bi_4t_4w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter_x, + const int8_t *filter_y, + int32_t height) +{ + if (2 == height) { + hevc_hv_bi_4t_4x2_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter_x, filter_y, height); + } else if (4 == height) { + hevc_hv_bi_4t_4x4_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter_x, filter_y, height); + } else if (0 == (height % 8)) { + hevc_hv_bi_4t_4multx8mult_msa(src0_ptr, src_stride, + src1_ptr, src2_stride, + dst, dst_stride, + filter_x, filter_y, height); + } +} + +static void hevc_hv_bi_4t_6w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter_x, + const int8_t *filter_y, + int32_t height) +{ + uint32_t loop_cnt; + v16i8 src0, src1, src2, src3, src4, src5, src6; + v8i16 in0, in1, in2, in3; + v8i16 filt0, filt1; + v4i32 filt_h0, filt_h1; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask1; + v8i16 filter_vec, const_vec; + v16i8 vec0, vec1, vec2, vec3, vec4, vec5; + v8i16 dst0, dst1, dst2, dst3, dst4, dst5; + v4i32 dst0_r, dst0_l, dst1_r, dst1_l, dst2_r, dst2_l, dst3_r, dst3_l; + v8i16 tmp0, tmp1, tmp2, tmp3; + v8i16 dst10_r, dst32_r, dst21_r, dst43_r; + v8i16 dst10_l, dst32_l, dst21_l, dst43_l; + + src0_ptr -= (src_stride + 1); + + filter_vec = LD_SH(filter_x); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + filter_vec = LD_SH(filter_y); + vec0 = __msa_clti_s_b((v16i8) filter_vec, 0); + filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec); + + SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1); + + mask1 = mask0 + 2; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + LD_SB3(src0_ptr, src_stride, src0, src1, src2); + src0_ptr += (3 * src_stride); + XORI_B3_128_SB(src0, src1, src2); + + VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1); + VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3); + VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5); + dst0 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0); + dst1 = const_vec; + DPADD_SB2_SH(vec2, vec3, filt0, filt1, dst1, dst1); + dst2 = const_vec; + DPADD_SB2_SH(vec4, vec5, filt0, filt1, dst2, dst2); + + ILVRL_H2_SH(dst1, dst0, dst10_r, dst10_l); + ILVRL_H2_SH(dst2, dst1, dst21_r, dst21_l); + + for (loop_cnt = height >> 2; loop_cnt--;) { + LD_SB4(src0_ptr, src_stride, src3, src4, src5, src6); + src0_ptr += (4 * src_stride); + LD_SH4(src1_ptr, src2_stride, in0, in1, in2, in3); + src1_ptr += (4 * src2_stride); + XORI_B4_128_SB(src3, src4, src5, src6); + + VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1); + dst3 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3); + + ILVRL_H2_SH(dst3, dst2, dst32_r, dst32_l); + dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1); + dst0_l = HEVC_FILT_4TAP(dst10_l, dst32_l, filt_h0, filt_h1); + dst0_r >>= 6; + dst0_l >>= 6; + + VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1); + dst4 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4); + + ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l); + dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1); + dst1_l = HEVC_FILT_4TAP(dst21_l, dst43_l, filt_h0, filt_h1); + dst1_r >>= 6; + dst1_l >>= 6; + + VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec0, vec1); + dst5 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst5, dst5); + + ILVRL_H2_SH(dst5, dst4, dst10_r, dst10_l); + dst2_r = HEVC_FILT_4TAP(dst32_r, dst10_r, filt_h0, filt_h1); + dst2_l = HEVC_FILT_4TAP(dst32_l, dst10_l, filt_h0, filt_h1); + dst2_r >>= 6; + dst2_l >>= 6; + + VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec0, vec1); + dst2 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2); + + ILVRL_H2_SH(dst2, dst5, dst21_r, dst21_l); + dst3_r = HEVC_FILT_4TAP(dst43_r, dst21_r, filt_h0, filt_h1); + dst3_l = HEVC_FILT_4TAP(dst43_l, dst21_l, filt_h0, filt_h1); + dst3_r >>= 6; + dst3_l >>= 6; + PCKEV_H4_SH(dst0_l, dst0_r, dst1_l, dst1_r, + dst2_l, dst2_r, dst3_l, dst3_r, tmp0, tmp1, tmp2, tmp3); + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + tmp0, tmp1, tmp2, tmp3, 7, tmp0, tmp1, tmp2, tmp3); + + PCKEV_B2_SW(tmp1, tmp0, tmp3, tmp2, dst0_r, dst1_r); + ST6x4_UB(dst0_r, dst1_r, dst, dst_stride); + dst += (4 * dst_stride); + } +} + +static void hevc_hv_bi_4t_8x2_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter_x, + const int8_t *filter_y, + int32_t height) +{ + v16i8 src0, src1, src2, src3, src4; + v8i16 filt0, filt1; + v4i32 filt_h0, filt_h1; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask1; + v8i16 filter_vec, const_vec; + v16i8 vec0, vec1, vec2, vec3, vec4, vec5; + v8i16 dst0, dst1, dst2, dst3, dst4; + v4i32 dst0_r, dst0_l, dst1_r, dst1_l; + v8i16 dst10_r, dst32_r, dst21_r, dst43_r; + v8i16 dst10_l, dst32_l, dst21_l, dst43_l; + v8i16 tmp0, tmp1; + v8i16 in0, in1; + + src0_ptr -= (src_stride + 1); + + filter_vec = LD_SH(filter_x); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + filter_vec = LD_SH(filter_y); + vec0 = __msa_clti_s_b((v16i8) filter_vec, 0); + filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec); + + SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1); + + mask1 = mask0 + 2; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + LD_SB3(src0_ptr, src_stride, src0, src1, src2); + src0_ptr += (3 * src_stride); + XORI_B3_128_SB(src0, src1, src2); + + VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1); + VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3); + VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5); + dst0 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0); + dst1 = const_vec; + DPADD_SB2_SH(vec2, vec3, filt0, filt1, dst1, dst1); + dst2 = const_vec; + DPADD_SB2_SH(vec4, vec5, filt0, filt1, dst2, dst2); + + ILVRL_H2_SH(dst1, dst0, dst10_r, dst10_l); + ILVRL_H2_SH(dst2, dst1, dst21_r, dst21_l); + + LD_SB2(src0_ptr, src_stride, src3, src4); + LD_SH2(src1_ptr, src2_stride, in0, in1); + XORI_B2_128_SB(src3, src4); + + VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1); + dst3 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3); + + ILVRL_H2_SH(dst3, dst2, dst32_r, dst32_l); + dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1); + dst0_l = HEVC_FILT_4TAP(dst10_l, dst32_l, filt_h0, filt_h1); + dst0_r >>= 6; + dst0_l >>= 6; + + VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1); + dst4 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4); + + ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l); + dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1); + dst1_l = HEVC_FILT_4TAP(dst21_l, dst43_l, filt_h0, filt_h1); + dst1_r >>= 6; + dst1_l >>= 6; + + PCKEV_H2_SH(dst0_l, dst0_r, dst1_l, dst1_r, tmp0, tmp1); + HEVC_BI_RND_CLIP2(in0, in1, tmp0, tmp1, 7, tmp0, tmp1); + + dst0_r = (v4i32) __msa_pckev_b((v16i8) tmp1, (v16i8) tmp0); + ST8x2_UB(dst0_r, dst, dst_stride); +} + +static void hevc_hv_bi_4t_8x6_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter_x, + const int8_t *filter_y, + int32_t height) +{ + v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8; + v8i16 in0, in1, in2, in3, in4, in5; + v8i16 filt0, filt1; + v4i32 filt_h0, filt_h1; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask1; + v8i16 filter_vec, const_vec; + v16i8 vec0, vec1, vec2, vec3, vec4, vec5; + v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5; + v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8; + v4i32 dst0_r, dst0_l, dst1_r, dst1_l, dst2_r, dst2_l, dst3_r, dst3_l; + v4i32 dst4_r, dst4_l, dst5_r, dst5_l; + v8i16 dst10_r, dst32_r, dst10_l, dst32_l; + v8i16 dst21_r, dst43_r, dst21_l, dst43_l; + v8i16 dst54_r, dst54_l, dst65_r, dst65_l; + v8i16 dst76_r, dst76_l, dst87_r, dst87_l; + + src0_ptr -= (src_stride + 1); + + filter_vec = LD_SH(filter_x); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + filter_vec = LD_SH(filter_y); + vec0 = __msa_clti_s_b((v16i8) filter_vec, 0); + filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec); + + SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1); + + mask1 = mask0 + 2; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + LD_SB3(src0_ptr, src_stride, src0, src1, src2); + src0_ptr += (3 * src_stride); + XORI_B3_128_SB(src0, src1, src2); + VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1); + VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3); + VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5); + dst0 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0); + dst1 = const_vec; + DPADD_SB2_SH(vec2, vec3, filt0, filt1, dst1, dst1); + dst2 = const_vec; + DPADD_SB2_SH(vec4, vec5, filt0, filt1, dst2, dst2); + + ILVRL_H2_SH(dst1, dst0, dst10_r, dst10_l); + ILVRL_H2_SH(dst2, dst1, dst21_r, dst21_l); + + LD_SB2(src0_ptr, src_stride, src3, src4); + src0_ptr += (2 * src_stride); + XORI_B2_128_SB(src3, src4); + LD_SH6(src1_ptr, src2_stride, in0, in1, in2, in3, in4, in5); + VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1); + dst3 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3); + + ILVRL_H2_SH(dst3, dst2, dst32_r, dst32_l); + dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1); + dst0_l = HEVC_FILT_4TAP(dst10_l, dst32_l, filt_h0, filt_h1); + dst0_r >>= 6; + dst0_l >>= 6; + tmp0 = __msa_pckev_h((v8i16) dst0_l, (v8i16) dst0_r); + + VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1); + dst4 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4); + + ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l); + dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1); + dst1_l = HEVC_FILT_4TAP(dst21_l, dst43_l, filt_h0, filt_h1); + dst1_r >>= 6; + dst1_l >>= 6; + tmp1 = __msa_pckev_h((v8i16) dst1_l, (v8i16) dst1_r); + + LD_SB2(src0_ptr, src_stride, src5, src6); + src0_ptr += (2 * src_stride); + XORI_B2_128_SB(src5, src6); + /* row 5 */ + VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec0, vec1); + dst5 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst5, dst5); + + ILVRL_H2_SH(dst5, dst4, dst54_r, dst54_l); + dst2_r = HEVC_FILT_4TAP(dst32_r, dst54_r, filt_h0, filt_h1); + dst2_l = HEVC_FILT_4TAP(dst32_l, dst54_l, filt_h0, filt_h1); + dst2_r >>= 6; + dst2_l >>= 6; + tmp2 = __msa_pckev_h((v8i16) dst2_l, (v8i16) dst2_r); + + /* row 6 */ + VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec0, vec1); + dst6 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst6, dst6); + + ILVRL_H2_SH(dst6, dst5, dst65_r, dst65_l); + dst3_r = HEVC_FILT_4TAP(dst43_r, dst65_r, filt_h0, filt_h1); + dst3_l = HEVC_FILT_4TAP(dst43_l, dst65_l, filt_h0, filt_h1); + dst3_r >>= 6; + dst3_l >>= 6; + tmp3 = __msa_pckev_h((v8i16) dst3_l, (v8i16) dst3_r); + + LD_SB2(src0_ptr, src_stride, src7, src8); + XORI_B2_128_SB(src7, src8); + /* row 7 */ + VSHF_B2_SB(src7, src7, src7, src7, mask0, mask1, vec0, vec1); + dst7 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst7, dst7); + + ILVRL_H2_SH(dst7, dst6, dst76_r, dst76_l); + dst4_r = HEVC_FILT_4TAP(dst54_r, dst76_r, filt_h0, filt_h1); + dst4_l = HEVC_FILT_4TAP(dst54_l, dst76_l, filt_h0, filt_h1); + + dst4_r >>= 6; + dst4_l >>= 6; + tmp4 = __msa_pckev_h((v8i16) dst4_l, (v8i16) dst4_r); + /* row 8 */ + VSHF_B2_SB(src8, src8, src8, src8, mask0, mask1, vec0, vec1); + dst8 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst8, dst8); + + ILVRL_H2_SH(dst8, dst7, dst87_r, dst87_l); + dst5_r = HEVC_FILT_4TAP(dst65_r, dst87_r, filt_h0, filt_h1); + dst5_l = HEVC_FILT_4TAP(dst65_l, dst87_l, filt_h0, filt_h1); + dst5_r >>= 6; + dst5_l >>= 6; + tmp5 = __msa_pckev_h((v8i16) dst5_l, (v8i16) dst5_r); + + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + tmp0, tmp1, tmp2, tmp3, 7, tmp0, tmp1, tmp2, tmp3); + HEVC_BI_RND_CLIP2(in4, in5, tmp4, tmp5, 7, tmp4, tmp5); + + PCKEV_B2_SW(tmp1, tmp0, tmp3, tmp2, dst0_r, dst1_r); + dst2_r = (v4i32) __msa_pckev_b((v16i8) tmp5, (v16i8) tmp4); + ST8x4_UB(dst0_r, dst1_r, dst, dst_stride); + dst += (4 * dst_stride); + ST8x2_UB(dst2_r, dst, dst_stride); +} + +static void hevc_hv_bi_4t_8multx4mult_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter_x, + const int8_t *filter_y, + int32_t height, + int32_t width) +{ + uint32_t loop_cnt, cnt; + uint8_t *src0_ptr_tmp; + int16_t *src1_ptr_tmp; + uint8_t *dst_tmp; + v16i8 src0, src1, src2, src3, src4, src5, src6; + v8i16 in0, in1, in2, in3; + v8i16 filt0, filt1; + v4i32 filt_h0, filt_h1; + v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask1; + v8i16 filter_vec, const_vec; + v16i8 vec0, vec1, vec2, vec3, vec4, vec5; + v8i16 dst0, dst1, dst2, dst3, dst4, dst5; + v4i32 dst0_r, dst0_l, dst1_r, dst1_l, dst2_r, dst2_l, dst3_r, dst3_l; + v8i16 tmp0, tmp1, tmp2, tmp3; + v8i16 dst10_r, dst32_r, dst21_r, dst43_r; + v8i16 dst10_l, dst32_l, dst21_l, dst43_l; + + src0_ptr -= (src_stride + 1); + + filter_vec = LD_SH(filter_x); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); + + filter_vec = LD_SH(filter_y); + vec0 = __msa_clti_s_b((v16i8) filter_vec, 0); + filter_vec = (v8i16) __msa_ilvr_b(vec0, (v16i8) filter_vec); + + SPLATI_W2_SW(filter_vec, 0, filt_h0, filt_h1); + + mask1 = mask0 + 2; + + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + for (cnt = width >> 3; cnt--;) { + src0_ptr_tmp = src0_ptr; + dst_tmp = dst; + src1_ptr_tmp = src1_ptr; + + LD_SB3(src0_ptr_tmp, src_stride, src0, src1, src2); + src0_ptr_tmp += (3 * src_stride); + XORI_B3_128_SB(src0, src1, src2); + + VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1); + VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3); + VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5); + dst0 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0); + dst1 = const_vec; + DPADD_SB2_SH(vec2, vec3, filt0, filt1, dst1, dst1); + dst2 = const_vec; + DPADD_SB2_SH(vec4, vec5, filt0, filt1, dst2, dst2); + + ILVRL_H2_SH(dst1, dst0, dst10_r, dst10_l); + ILVRL_H2_SH(dst2, dst1, dst21_r, dst21_l); + + for (loop_cnt = height >> 2; loop_cnt--;) { + LD_SB4(src0_ptr_tmp, src_stride, src3, src4, src5, src6); + src0_ptr_tmp += (4 * src_stride); + LD_SH4(src1_ptr_tmp, src2_stride, in0, in1, in2, in3); + src1_ptr_tmp += (4 * src2_stride); + XORI_B4_128_SB(src3, src4, src5, src6); + + VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1); + dst3 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3); + + ILVRL_H2_SH(dst3, dst2, dst32_r, dst32_l); + dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1); + dst0_l = HEVC_FILT_4TAP(dst10_l, dst32_l, filt_h0, filt_h1); + dst0_r >>= 6; + dst0_l >>= 6; + + VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1); + dst4 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst4, dst4); + + ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l); + dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1); + dst1_l = HEVC_FILT_4TAP(dst21_l, dst43_l, filt_h0, filt_h1); + dst1_r >>= 6; + dst1_l >>= 6; + + VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec0, vec1); + dst5 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst5, dst5); + + ILVRL_H2_SH(dst5, dst4, dst10_r, dst10_l); + dst2_r = HEVC_FILT_4TAP(dst32_r, dst10_r, filt_h0, filt_h1); + dst2_l = HEVC_FILT_4TAP(dst32_l, dst10_l, filt_h0, filt_h1); + dst2_r >>= 6; + dst2_l >>= 6; + + VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec0, vec1); + dst2 = const_vec; + DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2); + + ILVRL_H2_SH(dst2, dst5, dst21_r, dst21_l); + dst3_r = HEVC_FILT_4TAP(dst43_r, dst21_r, filt_h0, filt_h1); + dst3_l = HEVC_FILT_4TAP(dst43_l, dst21_l, filt_h0, filt_h1); + dst3_r >>= 6; + dst3_l >>= 6; + + PCKEV_H4_SH(dst0_l, dst0_r, dst1_l, dst1_r, + dst2_l, dst2_r, dst3_l, dst3_r, tmp0, tmp1, tmp2, tmp3); + HEVC_BI_RND_CLIP4(in0, in1, in2, in3, + tmp0, tmp1, tmp2, tmp3, 7, + tmp0, tmp1, tmp2, tmp3); + + PCKEV_B2_SW(tmp1, tmp0, tmp3, tmp2, dst0_r, dst1_r); + ST8x4_UB(dst0_r, dst1_r, dst_tmp, dst_stride); + dst_tmp += (4 * dst_stride); + } + + src0_ptr += 8; + dst += 8; + src1_ptr += 8; + } +} + +static void hevc_hv_bi_4t_8w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter_x, + const int8_t *filter_y, + int32_t height) +{ + if (2 == height) { + hevc_hv_bi_4t_8x2_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter_x, filter_y, height); + } else if (6 == height) { + hevc_hv_bi_4t_8x6_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter_x, filter_y, height); + } else { + hevc_hv_bi_4t_8multx4mult_msa(src0_ptr, src_stride, + src1_ptr, src2_stride, + dst, dst_stride, + filter_x, filter_y, height, 8); + } +} + +static void hevc_hv_bi_4t_12w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter_x, + const int8_t *filter_y, + int32_t height) +{ + hevc_hv_bi_4t_8multx4mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter_x, filter_y, + height, 8); + hevc_hv_bi_4t_4w_msa(src0_ptr + 8, src_stride, src1_ptr + 8, src2_stride, + dst + 8, dst_stride, filter_x, filter_y, height); +} + +static void hevc_hv_bi_4t_16w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter_x, + const int8_t *filter_y, + int32_t height) +{ + hevc_hv_bi_4t_8multx4mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter_x, filter_y, + height, 16); +} + +static void hevc_hv_bi_4t_24w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter_x, + const int8_t *filter_y, + int32_t height) +{ + hevc_hv_bi_4t_8multx4mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter_x, filter_y, + height, 24); +} + +static void hevc_hv_bi_4t_32w_msa(uint8_t *src0_ptr, + int32_t src_stride, + int16_t *src1_ptr, + int32_t src2_stride, + uint8_t *dst, + int32_t dst_stride, + const int8_t *filter_x, + const const int8_t *filter_y, + int32_t height) +{ + hevc_hv_bi_4t_8multx4mult_msa(src0_ptr, src_stride, src1_ptr, src2_stride, + dst, dst_stride, filter_x, filter_y, + height, 32); +} + +#define BI_MC_COPY(WIDTH) \ +void ff_hevc_put_hevc_bi_pel_pixels##WIDTH##_8_msa(uint8_t *dst, \ + ptrdiff_t dst_stride, \ + uint8_t *src, \ + ptrdiff_t src_stride, \ + int16_t *src_16bit, \ + int height, \ + intptr_t mx, \ + intptr_t my, \ + int width) \ +{ \ + hevc_bi_copy_##WIDTH##w_msa(src, src_stride, src_16bit, MAX_PB_SIZE, \ + dst, dst_stride, height); \ +} + +BI_MC_COPY(4); +BI_MC_COPY(6); +BI_MC_COPY(8); +BI_MC_COPY(12); +BI_MC_COPY(16); +BI_MC_COPY(24); +BI_MC_COPY(32); +BI_MC_COPY(48); +BI_MC_COPY(64); + +#undef BI_MC_COPY + +#define BI_MC(PEL, DIR, WIDTH, TAP, DIR1, FILT_DIR) \ +void ff_hevc_put_hevc_bi_##PEL##_##DIR####WIDTH##_8_msa(uint8_t *dst, \ + ptrdiff_t dst_stride, \ + uint8_t *src, \ + ptrdiff_t src_stride, \ + int16_t *src_16bit, \ + int height, \ + intptr_t mx, \ + intptr_t my, \ + int width) \ +{ \ + const int8_t *filter = ff_hevc_##PEL##_filters[FILT_DIR - 1]; \ + \ + hevc_##DIR1##_bi_##TAP##t_##WIDTH##w_msa(src, src_stride, src_16bit, \ + MAX_PB_SIZE, dst, dst_stride, \ + filter, height); \ +} + +BI_MC(qpel, h, 4, 8, hz, mx); +BI_MC(qpel, h, 8, 8, hz, mx); +BI_MC(qpel, h, 12, 8, hz, mx); +BI_MC(qpel, h, 16, 8, hz, mx); +BI_MC(qpel, h, 24, 8, hz, mx); +BI_MC(qpel, h, 32, 8, hz, mx); +BI_MC(qpel, h, 48, 8, hz, mx); +BI_MC(qpel, h, 64, 8, hz, mx); + +BI_MC(qpel, v, 4, 8, vt, my); +BI_MC(qpel, v, 8, 8, vt, my); +BI_MC(qpel, v, 12, 8, vt, my); +BI_MC(qpel, v, 16, 8, vt, my); +BI_MC(qpel, v, 24, 8, vt, my); +BI_MC(qpel, v, 32, 8, vt, my); +BI_MC(qpel, v, 48, 8, vt, my); +BI_MC(qpel, v, 64, 8, vt, my); + +BI_MC(epel, h, 4, 4, hz, mx); +BI_MC(epel, h, 8, 4, hz, mx); +BI_MC(epel, h, 6, 4, hz, mx); +BI_MC(epel, h, 12, 4, hz, mx); +BI_MC(epel, h, 16, 4, hz, mx); +BI_MC(epel, h, 24, 4, hz, mx); +BI_MC(epel, h, 32, 4, hz, mx); + +BI_MC(epel, v, 4, 4, vt, my); +BI_MC(epel, v, 8, 4, vt, my); +BI_MC(epel, v, 6, 4, vt, my); +BI_MC(epel, v, 12, 4, vt, my); +BI_MC(epel, v, 16, 4, vt, my); +BI_MC(epel, v, 24, 4, vt, my); +BI_MC(epel, v, 32, 4, vt, my); + +#undef BI_MC + +#define BI_MC_HV(PEL, DIR, WIDTH, TAP, DIR1) \ +void ff_hevc_put_hevc_bi_##PEL##_##DIR####WIDTH##_8_msa(uint8_t *dst, \ + ptrdiff_t dst_stride, \ + uint8_t *src, \ + ptrdiff_t src_stride, \ + int16_t *src_16bit, \ + int height, \ + intptr_t mx, \ + intptr_t my, \ + int width) \ +{ \ + const int8_t *filter_x = ff_hevc_##PEL##_filters[mx - 1]; \ + const int8_t *filter_y = ff_hevc_##PEL##_filters[my - 1]; \ + \ + hevc_##DIR1##_bi_##TAP##t_##WIDTH##w_msa(src, src_stride, src_16bit, \ + MAX_PB_SIZE, dst, dst_stride, \ + filter_x, filter_y, \ + height); \ +} + +BI_MC_HV(qpel, hv, 4, 8, hv); +BI_MC_HV(qpel, hv, 8, 8, hv); +BI_MC_HV(qpel, hv, 12, 8, hv); +BI_MC_HV(qpel, hv, 16, 8, hv); +BI_MC_HV(qpel, hv, 24, 8, hv); +BI_MC_HV(qpel, hv, 32, 8, hv); +BI_MC_HV(qpel, hv, 48, 8, hv); +BI_MC_HV(qpel, hv, 64, 8, hv); + +BI_MC_HV(epel, hv, 4, 4, hv); +BI_MC_HV(epel, hv, 8, 4, hv); +BI_MC_HV(epel, hv, 6, 4, hv); +BI_MC_HV(epel, hv, 12, 4, hv); +BI_MC_HV(epel, hv, 16, 4, hv); +BI_MC_HV(epel, hv, 24, 4, hv); +BI_MC_HV(epel, hv, 32, 4, hv); + +#undef BI_MC_HV diff --git a/libavcodec/mips/hevcdsp_init_mips.c b/libavcodec/mips/hevcdsp_init_mips.c index d2e3c60546..1d8ed9a868 100644 --- a/libavcodec/mips/hevcdsp_init_mips.c +++ b/libavcodec/mips/hevcdsp_init_mips.c @@ -96,6 +96,74 @@ static av_cold void hevc_dsp_init_msa(HEVCDSPContext *c, c->put_hevc_qpel_uni[7][1][1] = ff_hevc_put_hevc_uni_qpel_hv32_8_msa; c->put_hevc_qpel_uni[8][1][1] = ff_hevc_put_hevc_uni_qpel_hv48_8_msa; c->put_hevc_qpel_uni[9][1][1] = ff_hevc_put_hevc_uni_qpel_hv64_8_msa; + + c->put_hevc_qpel_bi[1][0][0] = ff_hevc_put_hevc_bi_pel_pixels4_8_msa; + c->put_hevc_qpel_bi[3][0][0] = ff_hevc_put_hevc_bi_pel_pixels8_8_msa; + c->put_hevc_qpel_bi[4][0][0] = ff_hevc_put_hevc_bi_pel_pixels12_8_msa; + c->put_hevc_qpel_bi[5][0][0] = ff_hevc_put_hevc_bi_pel_pixels16_8_msa; + c->put_hevc_qpel_bi[6][0][0] = ff_hevc_put_hevc_bi_pel_pixels24_8_msa; + c->put_hevc_qpel_bi[7][0][0] = ff_hevc_put_hevc_bi_pel_pixels32_8_msa; + c->put_hevc_qpel_bi[8][0][0] = ff_hevc_put_hevc_bi_pel_pixels48_8_msa; + c->put_hevc_qpel_bi[9][0][0] = ff_hevc_put_hevc_bi_pel_pixels64_8_msa; + + c->put_hevc_qpel_bi[1][0][1] = ff_hevc_put_hevc_bi_qpel_h4_8_msa; + c->put_hevc_qpel_bi[3][0][1] = ff_hevc_put_hevc_bi_qpel_h8_8_msa; + c->put_hevc_qpel_bi[4][0][1] = ff_hevc_put_hevc_bi_qpel_h12_8_msa; + c->put_hevc_qpel_bi[5][0][1] = ff_hevc_put_hevc_bi_qpel_h16_8_msa; + c->put_hevc_qpel_bi[6][0][1] = ff_hevc_put_hevc_bi_qpel_h24_8_msa; + c->put_hevc_qpel_bi[7][0][1] = ff_hevc_put_hevc_bi_qpel_h32_8_msa; + c->put_hevc_qpel_bi[8][0][1] = ff_hevc_put_hevc_bi_qpel_h48_8_msa; + c->put_hevc_qpel_bi[9][0][1] = ff_hevc_put_hevc_bi_qpel_h64_8_msa; + + c->put_hevc_qpel_bi[1][1][0] = ff_hevc_put_hevc_bi_qpel_v4_8_msa; + c->put_hevc_qpel_bi[3][1][0] = ff_hevc_put_hevc_bi_qpel_v8_8_msa; + c->put_hevc_qpel_bi[4][1][0] = ff_hevc_put_hevc_bi_qpel_v12_8_msa; + c->put_hevc_qpel_bi[5][1][0] = ff_hevc_put_hevc_bi_qpel_v16_8_msa; + c->put_hevc_qpel_bi[6][1][0] = ff_hevc_put_hevc_bi_qpel_v24_8_msa; + c->put_hevc_qpel_bi[7][1][0] = ff_hevc_put_hevc_bi_qpel_v32_8_msa; + c->put_hevc_qpel_bi[8][1][0] = ff_hevc_put_hevc_bi_qpel_v48_8_msa; + c->put_hevc_qpel_bi[9][1][0] = ff_hevc_put_hevc_bi_qpel_v64_8_msa; + + c->put_hevc_qpel_bi[1][1][1] = ff_hevc_put_hevc_bi_qpel_hv4_8_msa; + c->put_hevc_qpel_bi[3][1][1] = ff_hevc_put_hevc_bi_qpel_hv8_8_msa; + c->put_hevc_qpel_bi[4][1][1] = ff_hevc_put_hevc_bi_qpel_hv12_8_msa; + c->put_hevc_qpel_bi[5][1][1] = ff_hevc_put_hevc_bi_qpel_hv16_8_msa; + c->put_hevc_qpel_bi[6][1][1] = ff_hevc_put_hevc_bi_qpel_hv24_8_msa; + c->put_hevc_qpel_bi[7][1][1] = ff_hevc_put_hevc_bi_qpel_hv32_8_msa; + c->put_hevc_qpel_bi[8][1][1] = ff_hevc_put_hevc_bi_qpel_hv48_8_msa; + c->put_hevc_qpel_bi[9][1][1] = ff_hevc_put_hevc_bi_qpel_hv64_8_msa; + + c->put_hevc_epel_bi[1][0][0] = ff_hevc_put_hevc_bi_pel_pixels4_8_msa; + c->put_hevc_epel_bi[2][0][0] = ff_hevc_put_hevc_bi_pel_pixels6_8_msa; + c->put_hevc_epel_bi[3][0][0] = ff_hevc_put_hevc_bi_pel_pixels8_8_msa; + c->put_hevc_epel_bi[4][0][0] = ff_hevc_put_hevc_bi_pel_pixels12_8_msa; + c->put_hevc_epel_bi[5][0][0] = ff_hevc_put_hevc_bi_pel_pixels16_8_msa; + c->put_hevc_epel_bi[6][0][0] = ff_hevc_put_hevc_bi_pel_pixels24_8_msa; + c->put_hevc_epel_bi[7][0][0] = ff_hevc_put_hevc_bi_pel_pixels32_8_msa; + + c->put_hevc_epel_bi[1][0][1] = ff_hevc_put_hevc_bi_epel_h4_8_msa; + c->put_hevc_epel_bi[2][0][1] = ff_hevc_put_hevc_bi_epel_h6_8_msa; + c->put_hevc_epel_bi[3][0][1] = ff_hevc_put_hevc_bi_epel_h8_8_msa; + c->put_hevc_epel_bi[4][0][1] = ff_hevc_put_hevc_bi_epel_h12_8_msa; + c->put_hevc_epel_bi[5][0][1] = ff_hevc_put_hevc_bi_epel_h16_8_msa; + c->put_hevc_epel_bi[6][0][1] = ff_hevc_put_hevc_bi_epel_h24_8_msa; + c->put_hevc_epel_bi[7][0][1] = ff_hevc_put_hevc_bi_epel_h32_8_msa; + + c->put_hevc_epel_bi[1][1][0] = ff_hevc_put_hevc_bi_epel_v4_8_msa; + c->put_hevc_epel_bi[2][1][0] = ff_hevc_put_hevc_bi_epel_v6_8_msa; + c->put_hevc_epel_bi[3][1][0] = ff_hevc_put_hevc_bi_epel_v8_8_msa; + c->put_hevc_epel_bi[4][1][0] = ff_hevc_put_hevc_bi_epel_v12_8_msa; + c->put_hevc_epel_bi[5][1][0] = ff_hevc_put_hevc_bi_epel_v16_8_msa; + c->put_hevc_epel_bi[6][1][0] = ff_hevc_put_hevc_bi_epel_v24_8_msa; + c->put_hevc_epel_bi[7][1][0] = ff_hevc_put_hevc_bi_epel_v32_8_msa; + + c->put_hevc_epel_bi[1][1][1] = ff_hevc_put_hevc_bi_epel_hv4_8_msa; + c->put_hevc_epel_bi[2][1][1] = ff_hevc_put_hevc_bi_epel_hv6_8_msa; + c->put_hevc_epel_bi[3][1][1] = ff_hevc_put_hevc_bi_epel_hv8_8_msa; + c->put_hevc_epel_bi[4][1][1] = ff_hevc_put_hevc_bi_epel_hv12_8_msa; + c->put_hevc_epel_bi[5][1][1] = ff_hevc_put_hevc_bi_epel_hv16_8_msa; + c->put_hevc_epel_bi[6][1][1] = ff_hevc_put_hevc_bi_epel_hv24_8_msa; + c->put_hevc_epel_bi[7][1][1] = ff_hevc_put_hevc_bi_epel_hv32_8_msa; } } #endif // #if HAVE_MSA diff --git a/libavcodec/mips/hevcdsp_mips.h b/libavcodec/mips/hevcdsp_mips.h index a8c8848b30..d4f43e453e 100644 --- a/libavcodec/mips/hevcdsp_mips.h +++ b/libavcodec/mips/hevcdsp_mips.h @@ -116,3 +116,83 @@ UNI_MC(qpel, hv, 48); UNI_MC(qpel, hv, 64); #undef UNI_MC + +#define BI_MC(PEL, DIR, WIDTH) \ +void ff_hevc_put_hevc_bi_##PEL##_##DIR####WIDTH##_8_msa(uint8_t *dst, \ + ptrdiff_t dst_stride, \ + uint8_t *src, \ + ptrdiff_t src_stride, \ + int16_t *src_16bit, \ + int height, \ + intptr_t mx, \ + intptr_t my, \ + int width) + +BI_MC(pel, pixels, 4); +BI_MC(pel, pixels, 6); +BI_MC(pel, pixels, 8); +BI_MC(pel, pixels, 12); +BI_MC(pel, pixels, 16); +BI_MC(pel, pixels, 24); +BI_MC(pel, pixels, 32); +BI_MC(pel, pixels, 48); +BI_MC(pel, pixels, 64); + +BI_MC(qpel, h, 4); +BI_MC(qpel, h, 8); +BI_MC(qpel, h, 12); +BI_MC(qpel, h, 16); +BI_MC(qpel, h, 24); +BI_MC(qpel, h, 32); +BI_MC(qpel, h, 48); +BI_MC(qpel, h, 64); + +BI_MC(qpel, v, 4); +BI_MC(qpel, v, 8); +BI_MC(qpel, v, 12); +BI_MC(qpel, v, 16); +BI_MC(qpel, v, 24); +BI_MC(qpel, v, 32); +BI_MC(qpel, v, 48); +BI_MC(qpel, v, 64); + +BI_MC(qpel, hv, 4); +BI_MC(qpel, hv, 8); +BI_MC(qpel, hv, 12); +BI_MC(qpel, hv, 16); +BI_MC(qpel, hv, 24); +BI_MC(qpel, hv, 32); +BI_MC(qpel, hv, 48); +BI_MC(qpel, hv, 64); + +BI_MC(epel, h, 4); +BI_MC(epel, h, 6); +BI_MC(epel, h, 8); +BI_MC(epel, h, 12); +BI_MC(epel, h, 16); +BI_MC(epel, h, 24); +BI_MC(epel, h, 32); +BI_MC(epel, h, 48); +BI_MC(epel, h, 64); + +BI_MC(epel, v, 4); +BI_MC(epel, v, 6); +BI_MC(epel, v, 8); +BI_MC(epel, v, 12); +BI_MC(epel, v, 16); +BI_MC(epel, v, 24); +BI_MC(epel, v, 32); +BI_MC(epel, v, 48); +BI_MC(epel, v, 64); + +BI_MC(epel, hv, 4); +BI_MC(epel, hv, 6); +BI_MC(epel, hv, 8); +BI_MC(epel, hv, 12); +BI_MC(epel, hv, 16); +BI_MC(epel, hv, 24); +BI_MC(epel, hv, 32); +BI_MC(epel, hv, 48); +BI_MC(epel, hv, 64); + +#undef BI_MC diff --git a/libavutil/mips/generic_macros_msa.h b/libavutil/mips/generic_macros_msa.h index e2c9de2bc8..c4c549d771 100644 --- a/libavutil/mips/generic_macros_msa.h +++ b/libavutil/mips/generic_macros_msa.h @@ -29,18 +29,23 @@ #define LD_SB(...) LD_B(v16i8, __VA_ARGS__) #define LD_H(RTYPE, psrc) *((RTYPE *)(psrc)) +#define LD_UH(...) LD_H(v8u16, __VA_ARGS__) #define LD_SH(...) LD_H(v8i16, __VA_ARGS__) #define LD_W(RTYPE, psrc) *((RTYPE *)(psrc)) +#define LD_UW(...) LD_W(v4u32, __VA_ARGS__) #define LD_SW(...) LD_W(v4i32, __VA_ARGS__) #define ST_B(RTYPE, in, pdst) *((RTYPE *)(pdst)) = (in) #define ST_UB(...) ST_B(v16u8, __VA_ARGS__) +#define ST_SB(...) ST_B(v16i8, __VA_ARGS__) #define ST_H(RTYPE, in, pdst) *((RTYPE *)(pdst)) = (in) +#define ST_UH(...) ST_H(v8u16, __VA_ARGS__) #define ST_SH(...) ST_H(v8i16, __VA_ARGS__) #define ST_W(RTYPE, in, pdst) *((RTYPE *)(pdst)) = (in) +#define ST_UW(...) ST_W(v4u32, __VA_ARGS__) #define ST_SW(...) ST_W(v4i32, __VA_ARGS__) #if (__mips_isa_rev >= 6) @@ -328,6 +333,46 @@ #define LD_UB8(...) LD_B8(v16u8, __VA_ARGS__) #define LD_SB8(...) LD_B8(v16i8, __VA_ARGS__) +/* Description : Load vectors with 8 halfword elements with stride + Arguments : Inputs - psrc (source pointer to load from) + - stride + Outputs - out0, out1 + Details : Loads 8 halfword elements in 'out0' from (psrc) + Loads 8 halfword elements in 'out1' from (psrc + stride) +*/ +#define LD_H2(RTYPE, psrc, stride, out0, out1) \ +{ \ + out0 = LD_H(RTYPE, (psrc)); \ + out1 = LD_H(RTYPE, (psrc) + (stride)); \ +} +#define LD_UH2(...) LD_H2(v8u16, __VA_ARGS__) +#define LD_SH2(...) LD_H2(v8i16, __VA_ARGS__) + +#define LD_H4(RTYPE, psrc, stride, out0, out1, out2, out3) \ +{ \ + LD_H2(RTYPE, (psrc), stride, out0, out1); \ + LD_H2(RTYPE, (psrc) + 2 * stride, stride, out2, out3); \ +} +#define LD_UH4(...) LD_H4(v8u16, __VA_ARGS__) +#define LD_SH4(...) LD_H4(v8i16, __VA_ARGS__) + +#define LD_H6(RTYPE, psrc, stride, out0, out1, out2, out3, out4, out5) \ +{ \ + LD_H4(RTYPE, (psrc), stride, out0, out1, out2, out3); \ + LD_H2(RTYPE, (psrc) + 4 * stride, stride, out4, out5); \ +} +#define LD_UH6(...) LD_H6(v8u16, __VA_ARGS__) +#define LD_SH6(...) LD_H6(v8i16, __VA_ARGS__) + +#define LD_H8(RTYPE, psrc, stride, \ + out0, out1, out2, out3, out4, out5, out6, out7) \ +{ \ + LD_H4(RTYPE, (psrc), stride, out0, out1, out2, out3); \ + LD_H4(RTYPE, (psrc) + 4 * stride, stride, out4, out5, out6, out7); \ +} +#define LD_UH8(...) LD_H8(v8u16, __VA_ARGS__) +#define LD_SH8(...) LD_H8(v8i16, __VA_ARGS__) + /* Description : Store vectors of 16 byte elements with stride Arguments : Inputs - in0, in1, stride Outputs - pdst (destination pointer to store to) @@ -478,6 +523,55 @@ \ SW4(out0_m, out1_m, out2_m, out3_m, pblk_4x4_m, stride); \ } +#define ST4x8_UB(in0, in1, pdst, stride) \ +{ \ + uint8_t *pblk_4x8 = (uint8_t *) (pdst); \ + \ + ST4x4_UB(in0, in0, 0, 1, 2, 3, pblk_4x8, stride); \ + ST4x4_UB(in1, in1, 0, 1, 2, 3, pblk_4x8 + 4 * stride, stride); \ +} + +/* Description : Store as 6x4 byte block to destination memory from input + vectors + Arguments : Inputs - in0, in1, pdst, stride + Return Type - unsigned byte + Details : Index 0 word element from input vector 'in0' is copied and + stored on first line followed by index 2 halfword element + Index 2 word element from input vector 'in0' is copied and + stored on second line followed by index 2 halfword element + Index 0 word element from input vector 'in1' is copied and + stored on third line followed by index 2 halfword element + Index 2 word element from input vector 'in1' is copied and + stored on fourth line followed by index 2 halfword element +*/ +#define ST6x4_UB(in0, in1, pdst, stride) \ +{ \ + uint32_t out0_m, out1_m, out2_m, out3_m; \ + uint16_t out4_m, out5_m, out6_m, out7_m; \ + uint8_t *pblk_6x4_m = (uint8_t *) (pdst); \ + \ + out0_m = __msa_copy_u_w((v4i32) in0, 0); \ + out1_m = __msa_copy_u_w((v4i32) in0, 2); \ + out2_m = __msa_copy_u_w((v4i32) in1, 0); \ + out3_m = __msa_copy_u_w((v4i32) in1, 2); \ + \ + out4_m = __msa_copy_u_h((v8i16) in0, 2); \ + out5_m = __msa_copy_u_h((v8i16) in0, 6); \ + out6_m = __msa_copy_u_h((v8i16) in1, 2); \ + out7_m = __msa_copy_u_h((v8i16) in1, 6); \ + \ + SW(out0_m, pblk_6x4_m); \ + SH(out4_m, (pblk_6x4_m + 4)); \ + pblk_6x4_m += stride; \ + SW(out1_m, pblk_6x4_m); \ + SH(out5_m, (pblk_6x4_m + 4)); \ + pblk_6x4_m += stride; \ + SW(out2_m, pblk_6x4_m); \ + SH(out6_m, (pblk_6x4_m + 4)); \ + pblk_6x4_m += stride; \ + SW(out3_m, pblk_6x4_m); \ + SH(out7_m, (pblk_6x4_m + 4)); \ +} /* Description : Store as 8x2 byte block to destination memory from input vector Arguments : Inputs - in, pdst, stride @@ -529,6 +623,15 @@ ST8x4_UB(in0, in1, pblk_8x8_m, stride); \ ST8x4_UB(in2, in3, pblk_8x8_m + 4 * stride, stride); \ } +#define ST12x4_UB(in0, in1, in2, pdst, stride) \ +{ \ + uint8_t *pblk_12x4_m = (uint8_t *) (pdst); \ + \ + /* left 8x4 */ \ + ST8x4_UB(in0, in1, pblk_12x4_m, stride); \ + /* right 4x4 */ \ + ST4x4_UB(in2, in2, 0, 1, 2, 3, pblk_12x4_m + 8, stride); \ +} /* Description : Store as 12x8 byte block to destination memory from input vectors @@ -1246,6 +1349,8 @@ } #define PCKEV_B4_SB(...) PCKEV_B4(v16i8, __VA_ARGS__) #define PCKEV_B4_UB(...) PCKEV_B4(v16u8, __VA_ARGS__) +#define PCKEV_B4_SH(...) PCKEV_B4(v8i16, __VA_ARGS__) +#define PCKEV_B4_SW(...) PCKEV_B4(v4i32, __VA_ARGS__) /* Description : Pack even halfword elements of vector pairs Arguments : Inputs - in0, in1, in2, in3 @@ -1317,6 +1422,13 @@ } #define XORI_B5_128_SB(...) XORI_B5_128(v16i8, __VA_ARGS__) +#define XORI_B6_128(RTYPE, in0, in1, in2, in3, in4, in5) \ +{ \ + XORI_B4_128(RTYPE, in0, in1, in2, in3); \ + XORI_B2_128(RTYPE, in4, in5); \ +} +#define XORI_B6_128_SB(...) XORI_B6_128(v16i8, __VA_ARGS__) + #define XORI_B7_128(RTYPE, in0, in1, in2, in3, in4, in5, in6) \ { \ XORI_B4_128(RTYPE, in0, in1, in2, in3); \ @@ -1442,6 +1554,25 @@ } #define SRAR_H4_UH(...) SRAR_H4(v8u16, __VA_ARGS__) #define SRAR_H4_SH(...) SRAR_H4(v8i16, __VA_ARGS__) + +/* Description : Shift right arithmetic rounded (immediate) + Arguments : Inputs - in0, in1, in2, in3, shift + Outputs - in0, in1, in2, in3 (in place) + Return Type - as per RTYPE + Details : Each element of vector 'in0' is shifted right arithmetic by + value in 'shift'. + The last discarded bit is added to shifted value for rounding + and the result is in place written to 'in0' + Similar for other pairs +*/ +#define SRARI_H2(RTYPE, in0, in1, shift) \ +{ \ + in0 = (RTYPE) __msa_srari_h((v8i16) in0, shift); \ + in1 = (RTYPE) __msa_srari_h((v8i16) in1, shift); \ +} +#define SRARI_H2_UH(...) SRARI_H2(v8u16, __VA_ARGS__) +#define SRARI_H2_SH(...) SRARI_H2(v8i16, __VA_ARGS__) + /* Description : Shift right arithmetic rounded (immediate) Arguments : Inputs - in0, in1, shift Outputs - in0, in1 (in place) @@ -1499,6 +1630,25 @@ ILVRL_B2_SH(zero_m, in, out0, out1); \ } +/* Description : Sign extend halfword elements from input vector and return + result in pair of vectors + Arguments : Inputs - in (1 input halfword vector) + Outputs - out0, out1 (sign extended 2 word vectors) + Return Type - signed word + Details : Sign bit of halfword elements from input vector 'in' is + extracted and interleaved right with same vector 'in0' to + generate 4 signed word elements in 'out0' + Then interleaved left with same vector 'in0' to + generate 4 signed word elements in 'out1' +*/ +#define UNPCK_SH_SW(in, out0, out1) \ +{ \ + v8i16 tmp_m; \ + \ + tmp_m = __msa_clti_s_h((v8i16) in, 0); \ + ILVRL_H2_SW(tmp_m, in, out0, out1); \ +} + /* Description : Transposes input 4x4 byte block Arguments : Inputs - in0, in1, in2, in3 (input 4x4 byte block) Outputs - out0, out1, out2, out3 (output 4x4 byte block) |