diff --git a/test/convolve_test.cc b/test/convolve_test.cc index 6f1a294ae..a42bdc378 100644 --- a/test/convolve_test.cc +++ b/test/convolve_test.cc @@ -1815,8 +1815,7 @@ INSTANTIATE_TEST_CASE_P(DSPR2, ConvolveTest, ::testing::Values( make_tuple(64, 64, &convolve8_dspr2))); #endif -#if 0 // HAVE_MSA -// TODO(parag): enable when function hooks are added +#if HAVE_MSA const ConvolveFunctions convolve8_msa( vp9_convolve_copy_msa, vp9_convolve_avg_msa, vp9_convolve8_horiz_msa, vp9_convolve8_avg_horiz_c, diff --git a/vp9/common/mips/msa/vp9_convolve8_horiz_msa.c b/vp9/common/mips/msa/vp9_convolve8_horiz_msa.c index e2247435e..f175bf9b6 100644 --- a/vp9/common/mips/msa/vp9_convolve8_horiz_msa.c +++ b/vp9/common/mips/msa/vp9_convolve8_horiz_msa.c @@ -14,37 +14,29 @@ static void common_hz_8t_4x4_msa(const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride, int8_t *filter) { - v16i8 filt0, filt1, filt2, filt3; - v16i8 src0, src1, src2, src3; - v16u8 mask0, mask1, mask2, mask3; + v16u8 mask0, mask1, mask2, mask3, out; + v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3; v8i16 filt, out0, out1; - mask0 = LOAD_UB(&mc_filt_mask_arr[16]); - + mask0 = LD_UB(&mc_filt_mask_arr[16]); src -= 3; /* rearranging filter */ - filt = LOAD_SH(filter); - filt0 = (v16i8)__msa_splati_h(filt, 0); - filt1 = (v16i8)__msa_splati_h(filt, 1); - filt2 = (v16i8)__msa_splati_h(filt, 2); - filt3 = (v16i8)__msa_splati_h(filt, 3); + filt = LD_SH(filter); + SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3); mask1 = mask0 + 2; mask2 = mask0 + 4; mask3 = mask0 + 6; - LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); - - XORI_B_4VECS_SB(src0, src1, src2, src3, src0, src1, src2, src3, 128); - + LD_SB4(src, src_stride, src0, src1, src2, src3); + XORI_B4_128_SB(src0, src1, src2, src3); HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3, filt0, filt1, filt2, filt3, out0, out1); - - out0 = SRARI_SATURATE_SIGNED_H(out0, FILTER_BITS, 7); - out1 = SRARI_SATURATE_SIGNED_H(out1, FILTER_BITS, 7); - - PCKEV_2B_XORI128_STORE_4_BYTES_4(out0, out1, dst, dst_stride); + SRARI_H2_SH(out0, out1, FILTER_BITS); + SAT_SH2_SH(out0, out1, 7); + out = PCKEV_XORI128_UB(out0, out1); + ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride); } static void common_hz_8t_4x8_msa(const uint8_t *src, int32_t src_stride, @@ -52,47 +44,36 @@ static void common_hz_8t_4x8_msa(const uint8_t *src, int32_t src_stride, int8_t *filter) { v16i8 filt0, filt1, filt2, filt3; v16i8 src0, src1, src2, src3; - v16u8 mask0, mask1, mask2, mask3; + v16u8 mask0, mask1, mask2, mask3, out; v8i16 filt, out0, out1, out2, out3; - mask0 = LOAD_UB(&mc_filt_mask_arr[16]); - + mask0 = LD_UB(&mc_filt_mask_arr[16]); src -= 3; /* rearranging filter */ - filt = LOAD_SH(filter); - filt0 = (v16i8)__msa_splati_h(filt, 0); - filt1 = (v16i8)__msa_splati_h(filt, 1); - filt2 = (v16i8)__msa_splati_h(filt, 2); - filt3 = (v16i8)__msa_splati_h(filt, 3); + filt = LD_SH(filter); + SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3); mask1 = mask0 + 2; mask2 = mask0 + 4; mask3 = mask0 + 6; - LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); + LD_SB4(src, src_stride, src0, src1, src2, src3); + XORI_B4_128_SB(src0, src1, src2, src3); src += (4 * src_stride); - - XORI_B_4VECS_SB(src0, src1, src2, src3, src0, src1, src2, src3, 128); - HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3, filt0, filt1, filt2, filt3, out0, out1); - - LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); - - XORI_B_4VECS_SB(src0, src1, src2, src3, src0, src1, src2, src3, 128); - + LD_SB4(src, src_stride, src0, src1, src2, src3); + XORI_B4_128_SB(src0, src1, src2, src3); HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3, filt0, filt1, filt2, filt3, out2, out3); - - out0 = SRARI_SATURATE_SIGNED_H(out0, FILTER_BITS, 7); - out1 = SRARI_SATURATE_SIGNED_H(out1, FILTER_BITS, 7); - out2 = SRARI_SATURATE_SIGNED_H(out2, FILTER_BITS, 7); - out3 = SRARI_SATURATE_SIGNED_H(out3, FILTER_BITS, 7); - - PCKEV_2B_XORI128_STORE_4_BYTES_4(out0, out1, dst, dst_stride); + SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS); + SAT_SH4_SH(out0, out1, out2, out3, 7); + out = PCKEV_XORI128_UB(out0, out1); + ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride); dst += (4 * dst_stride); - PCKEV_2B_XORI128_STORE_4_BYTES_4(out2, out3, dst, dst_stride); + out = PCKEV_XORI128_UB(out2, out3); + ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride); } static void common_hz_8t_4w_msa(const uint8_t *src, int32_t src_stride, @@ -108,82 +89,64 @@ static void common_hz_8t_4w_msa(const uint8_t *src, int32_t src_stride, static void common_hz_8t_8x4_msa(const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride, int8_t *filter) { - v16i8 filt0, filt1, filt2, filt3; - v16i8 src0, src1, src2, src3; - v16u8 mask0, mask1, mask2, mask3; + v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3; + v16u8 mask0, mask1, mask2, mask3, tmp0, tmp1; v8i16 filt, out0, out1, out2, out3; - mask0 = LOAD_UB(&mc_filt_mask_arr[0]); - + mask0 = LD_UB(&mc_filt_mask_arr[0]); src -= 3; /* rearranging filter */ - filt = LOAD_SH(filter); - filt0 = (v16i8)__msa_splati_h(filt, 0); - filt1 = (v16i8)__msa_splati_h(filt, 1); - filt2 = (v16i8)__msa_splati_h(filt, 2); - filt3 = (v16i8)__msa_splati_h(filt, 3); + filt = LD_SH(filter); + SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3); mask1 = mask0 + 2; mask2 = mask0 + 4; mask3 = mask0 + 6; - LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); - - XORI_B_4VECS_SB(src0, src1, src2, src3, src0, src1, src2, src3, 128); - + LD_SB4(src, src_stride, src0, src1, src2, src3); + XORI_B4_128_SB(src0, src1, src2, src3); HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3, filt0, filt1, filt2, filt3, out0, out1, out2, out3); - - out0 = SRARI_SATURATE_SIGNED_H(out0, FILTER_BITS, 7); - out1 = SRARI_SATURATE_SIGNED_H(out1, FILTER_BITS, 7); - out2 = SRARI_SATURATE_SIGNED_H(out2, FILTER_BITS, 7); - out3 = SRARI_SATURATE_SIGNED_H(out3, FILTER_BITS, 7); - - PCKEV_B_4_XORI128_STORE_8_BYTES_4(out0, out1, out2, out3, dst, dst_stride); + SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS); + SAT_SH4_SH(out0, out1, out2, out3, 7); + tmp0 = PCKEV_XORI128_UB(out0, out1); + tmp1 = PCKEV_XORI128_UB(out2, out3); + ST8x4_UB(tmp0, tmp1, dst, dst_stride); } static void common_hz_8t_8x8mult_msa(const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride, int8_t *filter, int32_t height) { uint32_t loop_cnt; - v16i8 filt0, filt1, filt2, filt3; - v16i8 src0, src1, src2, src3; - v16u8 mask0, mask1, mask2, mask3; + v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3; + v16u8 mask0, mask1, mask2, mask3, tmp0, tmp1; v8i16 filt, out0, out1, out2, out3; - mask0 = LOAD_UB(&mc_filt_mask_arr[0]); - + mask0 = LD_UB(&mc_filt_mask_arr[0]); src -= 3; /* rearranging filter */ - filt = LOAD_SH(filter); - filt0 = (v16i8)__msa_splati_h(filt, 0); - filt1 = (v16i8)__msa_splati_h(filt, 1); - filt2 = (v16i8)__msa_splati_h(filt, 2); - filt3 = (v16i8)__msa_splati_h(filt, 3); + filt = LD_SH(filter); + SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3); mask1 = mask0 + 2; mask2 = mask0 + 4; mask3 = mask0 + 6; for (loop_cnt = (height >> 2); loop_cnt--;) { - LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); + LD_SB4(src, src_stride, src0, src1, src2, src3); + XORI_B4_128_SB(src0, src1, src2, src3); src += (4 * src_stride); - - XORI_B_4VECS_SB(src0, src1, src2, src3, src0, src1, src2, src3, 128); - HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3, filt0, filt1, filt2, filt3, out0, out1, out2, out3); - - out0 = SRARI_SATURATE_SIGNED_H(out0, FILTER_BITS, 7); - out1 = SRARI_SATURATE_SIGNED_H(out1, FILTER_BITS, 7); - out2 = SRARI_SATURATE_SIGNED_H(out2, FILTER_BITS, 7); - out3 = SRARI_SATURATE_SIGNED_H(out3, FILTER_BITS, 7); - - PCKEV_B_4_XORI128_STORE_8_BYTES_4(out0, out1, out2, out3, dst, dst_stride); + SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS); + SAT_SH4_SH(out0, out1, out2, out3, 7); + tmp0 = PCKEV_XORI128_UB(out0, out1); + tmp1 = PCKEV_XORI128_UB(out2, out3); + ST8x4_UB(tmp0, tmp1, dst, dst_stride); dst += (4 * dst_stride); } } @@ -202,48 +165,36 @@ static void common_hz_8t_16w_msa(const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride, int8_t *filter, int32_t height) { uint32_t loop_cnt; - v16i8 src0, src1, src2, src3; - v16i8 filt0, filt1, filt2, filt3; - v16u8 mask0, mask1, mask2, mask3; + v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3; + v16u8 mask0, mask1, mask2, mask3, out; v8i16 filt, out0, out1, out2, out3; - mask0 = LOAD_UB(&mc_filt_mask_arr[0]); - + mask0 = LD_UB(&mc_filt_mask_arr[0]); src -= 3; /* rearranging filter */ - filt = LOAD_SH(filter); - filt0 = (v16i8)__msa_splati_h(filt, 0); - filt1 = (v16i8)__msa_splati_h(filt, 1); - filt2 = (v16i8)__msa_splati_h(filt, 2); - filt3 = (v16i8)__msa_splati_h(filt, 3); + filt = LD_SH(filter); + SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3); mask1 = mask0 + 2; mask2 = mask0 + 4; mask3 = mask0 + 6; for (loop_cnt = (height >> 1); loop_cnt--;) { - src0 = LOAD_SB(src); - src1 = LOAD_SB(src + 8); - src += src_stride; - src2 = LOAD_SB(src); - src3 = LOAD_SB(src + 8); - src += src_stride; - - XORI_B_4VECS_SB(src0, src1, src2, src3, src0, src1, src2, src3, 128); - + LD_SB2(src, src_stride, src0, src2); + LD_SB2(src + 8, src_stride, src1, src3); + XORI_B4_128_SB(src0, src1, src2, src3); + src += (2 * src_stride); HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3, filt0, filt1, filt2, filt3, out0, out1, out2, out3); - - out0 = SRARI_SATURATE_SIGNED_H(out0, FILTER_BITS, 7); - out1 = SRARI_SATURATE_SIGNED_H(out1, FILTER_BITS, 7); - out2 = SRARI_SATURATE_SIGNED_H(out2, FILTER_BITS, 7); - out3 = SRARI_SATURATE_SIGNED_H(out3, FILTER_BITS, 7); - - PCKEV_B_XORI128_STORE_VEC(out1, out0, dst); + SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS); + SAT_SH4_SH(out0, out1, out2, out3, 7); + out = PCKEV_XORI128_UB(out0, out1); + ST_UB(out, dst); dst += dst_stride; - PCKEV_B_XORI128_STORE_VEC(out3, out2, dst); + out = PCKEV_XORI128_UB(out2, out3); + ST_UB(out, dst); dst += dst_stride; } } @@ -252,68 +203,56 @@ static void common_hz_8t_32w_msa(const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride, int8_t *filter, int32_t height) { uint32_t loop_cnt; - v16i8 src0, src1, src2, src3; - v16i8 filt0, filt1, filt2, filt3; - v16u8 mask0, mask1, mask2, mask3; + v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3; + v16u8 mask0, mask1, mask2, mask3, out; v8i16 filt, out0, out1, out2, out3; - mask0 = LOAD_UB(&mc_filt_mask_arr[0]); - + mask0 = LD_UB(&mc_filt_mask_arr[0]); src -= 3; /* rearranging filter */ - filt = LOAD_SH(filter); - filt0 = (v16i8)__msa_splati_h(filt, 0); - filt1 = (v16i8)__msa_splati_h(filt, 1); - filt2 = (v16i8)__msa_splati_h(filt, 2); - filt3 = (v16i8)__msa_splati_h(filt, 3); + filt = LD_SH(filter); + SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3); mask1 = mask0 + 2; mask2 = mask0 + 4; mask3 = mask0 + 6; for (loop_cnt = (height >> 1); loop_cnt--;) { - src0 = LOAD_SB(src); - src2 = LOAD_SB(src + 16); - src3 = LOAD_SB(src + 24); - src1 = __msa_sld_b((v16i8)src2, (v16i8)src0, 8); + src0 = LD_SB(src); + src2 = LD_SB(src + 16); + src3 = LD_SB(src + 24); + src1 = __msa_sldi_b(src2, src0, 8); src += src_stride; - - XORI_B_4VECS_SB(src0, src1, src2, src3, src0, src1, src2, src3, 128); - + XORI_B4_128_SB(src0, src1, src2, src3); HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3, filt0, filt1, filt2, filt3, out0, out1, out2, out3); + SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS); + SAT_SH4_SH(out0, out1, out2, out3, 7); - out0 = SRARI_SATURATE_SIGNED_H(out0, FILTER_BITS, 7); - out1 = SRARI_SATURATE_SIGNED_H(out1, FILTER_BITS, 7); - out2 = SRARI_SATURATE_SIGNED_H(out2, FILTER_BITS, 7); - out3 = SRARI_SATURATE_SIGNED_H(out3, FILTER_BITS, 7); + src0 = LD_SB(src); + src2 = LD_SB(src + 16); + src3 = LD_SB(src + 24); + src1 = __msa_sldi_b(src2, src0, 8); + src += src_stride; - src0 = LOAD_SB(src); - src2 = LOAD_SB(src + 16); - src3 = LOAD_SB(src + 24); - src1 = __msa_sld_b((v16i8)src2, (v16i8)src0, 8); - - PCKEV_B_XORI128_STORE_VEC(out1, out0, dst); - PCKEV_B_XORI128_STORE_VEC(out3, out2, (dst + 16)); + out = PCKEV_XORI128_UB(out0, out1); + ST_UB(out, dst); + out = PCKEV_XORI128_UB(out2, out3); + ST_UB(out, dst + 16); dst += dst_stride; - XORI_B_4VECS_SB(src0, src1, src2, src3, src0, src1, src2, src3, 128); - + XORI_B4_128_SB(src0, src1, src2, src3); HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3, filt0, filt1, filt2, filt3, out0, out1, out2, out3); - - out0 = SRARI_SATURATE_SIGNED_H(out0, FILTER_BITS, 7); - out1 = SRARI_SATURATE_SIGNED_H(out1, FILTER_BITS, 7); - out2 = SRARI_SATURATE_SIGNED_H(out2, FILTER_BITS, 7); - out3 = SRARI_SATURATE_SIGNED_H(out3, FILTER_BITS, 7); - - PCKEV_B_XORI128_STORE_VEC(out1, out0, dst); - PCKEV_B_XORI128_STORE_VEC(out3, out2, (dst + 16)); - - src += src_stride; + SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS); + SAT_SH4_SH(out0, out1, out2, out3, 7); + out = PCKEV_XORI128_UB(out0, out1); + ST_UB(out, dst); + out = PCKEV_XORI128_UB(out2, out3); + ST_UB(out, dst + 16); dst += dst_stride; } } @@ -321,50 +260,55 @@ static void common_hz_8t_32w_msa(const uint8_t *src, int32_t src_stride, static void common_hz_8t_64w_msa(const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride, int8_t *filter, int32_t height) { - uint32_t loop_cnt, cnt; - v16i8 src0, src1, src2, src3; - v16i8 filt0, filt1, filt2, filt3; - v16u8 mask0, mask1, mask2, mask3; + int32_t loop_cnt; + v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3; + v16u8 mask0, mask1, mask2, mask3, out; v8i16 filt, out0, out1, out2, out3; - mask0 = LOAD_UB(&mc_filt_mask_arr[0]); - + mask0 = LD_UB(&mc_filt_mask_arr[0]); src -= 3; /* rearranging filter */ - filt = LOAD_SH(filter); - filt0 = (v16i8)__msa_splati_h(filt, 0); - filt1 = (v16i8)__msa_splati_h(filt, 1); - filt2 = (v16i8)__msa_splati_h(filt, 2); - filt3 = (v16i8)__msa_splati_h(filt, 3); + filt = LD_SH(filter); + SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3); mask1 = mask0 + 2; mask2 = mask0 + 4; mask3 = mask0 + 6; for (loop_cnt = height; loop_cnt--;) { - for (cnt = 0; cnt < 2; ++cnt) { - src0 = LOAD_SB(&src[cnt << 5]); - src2 = LOAD_SB(&src[16 + (cnt << 5)]); - src3 = LOAD_SB(&src[24 + (cnt << 5)]); - src1 = __msa_sld_b((v16i8)src2, (v16i8)src0, 8); + src0 = LD_SB(src); + src2 = LD_SB(src + 16); + src3 = LD_SB(src + 24); + src1 = __msa_sldi_b(src2, src0, 8); - XORI_B_4VECS_SB(src0, src1, src2, src3, src0, src1, src2, src3, 128); - - HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, - mask3, filt0, filt1, filt2, filt3, out0, out1, - out2, out3); - - out0 = SRARI_SATURATE_SIGNED_H(out0, FILTER_BITS, 7); - out1 = SRARI_SATURATE_SIGNED_H(out1, FILTER_BITS, 7); - out2 = SRARI_SATURATE_SIGNED_H(out2, FILTER_BITS, 7); - out3 = SRARI_SATURATE_SIGNED_H(out3, FILTER_BITS, 7); - - PCKEV_B_XORI128_STORE_VEC(out1, out0, &dst[cnt << 5]); - PCKEV_B_XORI128_STORE_VEC(out3, out2, &dst[16 + (cnt << 5)]); - } + XORI_B4_128_SB(src0, src1, src2, src3); + HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, + mask3, filt0, filt1, filt2, filt3, out0, out1, + out2, out3); + SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS); + SAT_SH4_SH(out0, out1, out2, out3, 7); + out = PCKEV_XORI128_UB(out0, out1); + ST_UB(out, dst); + out = PCKEV_XORI128_UB(out2, out3); + ST_UB(out, dst + 16); + src0 = LD_SB(src + 32); + src2 = LD_SB(src + 48); + src3 = LD_SB(src + 56); + src1 = __msa_sldi_b(src2, src0, 8); src += src_stride; + + XORI_B4_128_SB(src0, src1, src2, src3); + HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, + mask3, filt0, filt1, filt2, filt3, out0, out1, + out2, out3); + SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS); + SAT_SH4_SH(out0, out1, out2, out3, 7); + out = PCKEV_XORI128_UB(out0, out1); + ST_UB(out, dst + 32); + out = PCKEV_XORI128_UB(out2, out3); + ST_UB(out, dst + 48); dst += dst_stride; } } @@ -372,124 +316,55 @@ static void common_hz_8t_64w_msa(const uint8_t *src, int32_t src_stride, static void common_hz_2t_4x4_msa(const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride, int8_t *filter) { - uint32_t out0, out1, out2, out3; v16i8 src0, src1, src2, src3, mask; - v16u8 vec0, vec1, filt0; - v16i8 res0, res1; + v16u8 filt0, vec0, vec1, res0, res1; v8u16 vec2, vec3, filt, const255; - mask = LOAD_SB(&mc_filt_mask_arr[16]); + mask = LD_SB(&mc_filt_mask_arr[16]); /* rearranging filter */ - filt = LOAD_UH(filter); - filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0); + filt = LD_UH(filter); + filt0 = (v16u8) __msa_splati_h((v8i16) filt, 0); - const255 = (v8u16)__msa_ldi_h(255); + const255 = (v8u16) __msa_ldi_h(255); - LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); - - vec0 = (v16u8)__msa_vshf_b(mask, src1, src0); - vec1 = (v16u8)__msa_vshf_b(mask, src3, src2); - - vec2 = __msa_dotp_u_h(vec0, filt0); - vec3 = __msa_dotp_u_h(vec1, filt0); - - vec2 = (v8u16)__msa_srari_h((v8i16)vec2, FILTER_BITS); - vec3 = (v8u16)__msa_srari_h((v8i16)vec3, FILTER_BITS); - - vec2 = __msa_min_u_h(vec2, const255); - vec3 = __msa_min_u_h(vec3, const255); - - res0 = __msa_pckev_b((v16i8)vec2, (v16i8)vec2); - res1 = __msa_pckev_b((v16i8)vec3, (v16i8)vec3); - - out0 = __msa_copy_u_w((v4i32)res0, 0); - out1 = __msa_copy_u_w((v4i32)res0, 1); - out2 = __msa_copy_u_w((v4i32)res1, 0); - out3 = __msa_copy_u_w((v4i32)res1, 1); - - STORE_WORD(dst, out0); - dst += dst_stride; - STORE_WORD(dst, out1); - dst += dst_stride; - STORE_WORD(dst, out2); - dst += dst_stride; - STORE_WORD(dst, out3); + LD_SB4(src, src_stride, src0, src1, src2, src3); + VSHF_B2_UB(src0, src1, src2, src3, mask, mask, vec0, vec1); + DOTP_UB2_UH(vec0, vec1, filt0, filt0, vec2, vec3); + SRARI_H2_UH(vec2, vec3, FILTER_BITS); + MIN_UH2_UH(vec2, vec3, const255); + PCKEV_B2_UB(vec2, vec2, vec3, vec3, res0, res1); + ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride); } static void common_hz_2t_4x8_msa(const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride, int8_t *filter) { - uint32_t out0, out1, out2, out3; - v16u8 filt0; + v16u8 vec0, vec1, vec2, vec3, filt0; v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask; - v16u8 vec0, vec1, vec2, vec3; - v8u16 vec4, vec5, vec6, vec7; v16i8 res0, res1, res2, res3; - v8u16 filt, const255; + v8u16 vec4, vec5, vec6, vec7, filt, const255; - mask = LOAD_SB(&mc_filt_mask_arr[16]); + mask = LD_SB(&mc_filt_mask_arr[16]); /* rearranging filter */ - filt = LOAD_UH(filter); - filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0); + filt = LD_UH(filter); + filt0 = (v16u8) __msa_splati_h((v8i16) filt, 0); - const255 = (v8u16)__msa_ldi_h(255); + const255 = (v8u16) __msa_ldi_h(255); - LOAD_8VECS_SB(src, src_stride, - src0, src1, src2, src3, src4, src5, src6, src7); - - vec0 = (v16u8)__msa_vshf_b(mask, src1, src0); - vec1 = (v16u8)__msa_vshf_b(mask, src3, src2); - vec2 = (v16u8)__msa_vshf_b(mask, src5, src4); - vec3 = (v16u8)__msa_vshf_b(mask, src7, src6); - - vec4 = __msa_dotp_u_h(vec0, filt0); - vec5 = __msa_dotp_u_h(vec1, filt0); - vec6 = __msa_dotp_u_h(vec2, filt0); - vec7 = __msa_dotp_u_h(vec3, filt0); - - vec4 = (v8u16)__msa_srari_h((v8i16)vec4, FILTER_BITS); - vec5 = (v8u16)__msa_srari_h((v8i16)vec5, FILTER_BITS); - vec6 = (v8u16)__msa_srari_h((v8i16)vec6, FILTER_BITS); - vec7 = (v8u16)__msa_srari_h((v8i16)vec7, FILTER_BITS); - - vec4 = __msa_min_u_h(vec4, const255); - vec5 = __msa_min_u_h(vec5, const255); - vec6 = __msa_min_u_h(vec6, const255); - vec7 = __msa_min_u_h(vec7, const255); - - res0 = __msa_pckev_b((v16i8)vec4, (v16i8)vec4); - res1 = __msa_pckev_b((v16i8)vec5, (v16i8)vec5); - res2 = __msa_pckev_b((v16i8)vec6, (v16i8)vec6); - res3 = __msa_pckev_b((v16i8)vec7, (v16i8)vec7); - - out0 = __msa_copy_u_w((v4i32)res0, 0); - out1 = __msa_copy_u_w((v4i32)res0, 1); - out2 = __msa_copy_u_w((v4i32)res1, 0); - out3 = __msa_copy_u_w((v4i32)res1, 1); - - STORE_WORD(dst, out0); - dst += dst_stride; - STORE_WORD(dst, out1); - dst += dst_stride; - STORE_WORD(dst, out2); - dst += dst_stride; - STORE_WORD(dst, out3); - dst += dst_stride; - - out0 = __msa_copy_u_w((v4i32)res2, 0); - out1 = __msa_copy_u_w((v4i32)res2, 1); - out2 = __msa_copy_u_w((v4i32)res3, 0); - out3 = __msa_copy_u_w((v4i32)res3, 1); - - STORE_WORD(dst, out0); - dst += dst_stride; - STORE_WORD(dst, out1); - dst += dst_stride; - STORE_WORD(dst, out2); - dst += dst_stride; - STORE_WORD(dst, out3); + LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7); + VSHF_B2_UB(src0, src1, src2, src3, mask, mask, vec0, vec1); + VSHF_B2_UB(src4, src5, src6, src7, mask, mask, vec2, vec3); + DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec4, vec5, + vec6, vec7); + SRARI_H4_UH(vec4, vec5, vec6, vec7, FILTER_BITS); + MIN_UH4_UH(vec4, vec5, vec6, vec7, const255); + PCKEV_B4_SB(vec4, vec4, vec5, vec5, vec6, vec6, vec7, vec7, res0, res1, + res2, res3); + ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride); + dst += (4 * dst_stride); + ST4x4_UB(res2, res3, 0, 1, 0, 1, dst, dst_stride); } static void common_hz_2t_4w_msa(const uint8_t *src, int32_t src_stride, @@ -507,149 +382,93 @@ static void common_hz_2t_8x4_msa(const uint8_t *src, int32_t src_stride, int8_t *filter) { v16u8 filt0; v16i8 src0, src1, src2, src3, mask; - v8u16 vec0, vec1, vec2, vec3; - v8u16 out0, out1, out2, out3; - v8u16 const255, filt; + v8u16 vec0, vec1, vec2, vec3, const255, filt; - mask = LOAD_SB(&mc_filt_mask_arr[0]); + mask = LD_SB(&mc_filt_mask_arr[0]); /* rearranging filter */ - filt = LOAD_UH(filter); - filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0); + filt = LD_UH(filter); + filt0 = (v16u8) __msa_splati_h((v8i16) filt, 0); - const255 = (v8u16)__msa_ldi_h(255); + const255 = (v8u16) __msa_ldi_h(255); - LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); - - vec0 = (v8u16)__msa_vshf_b(mask, src0, src0); - vec1 = (v8u16)__msa_vshf_b(mask, src1, src1); - vec2 = (v8u16)__msa_vshf_b(mask, src2, src2); - vec3 = (v8u16)__msa_vshf_b(mask, src3, src3); - - vec0 = __msa_dotp_u_h((v16u8)vec0, filt0); - vec1 = __msa_dotp_u_h((v16u8)vec1, filt0); - vec2 = __msa_dotp_u_h((v16u8)vec2, filt0); - vec3 = __msa_dotp_u_h((v16u8)vec3, filt0); - - SRARI_H_4VECS_UH(vec0, vec1, vec2, vec3, vec0, vec1, vec2, vec3, FILTER_BITS); - - out0 = __msa_min_u_h(vec0, const255); - out1 = __msa_min_u_h(vec1, const255); - out2 = __msa_min_u_h(vec2, const255); - out3 = __msa_min_u_h(vec3, const255); - - PCKEV_B_STORE_8_BYTES_4(out0, out1, out2, out3, dst, dst_stride); + LD_SB4(src, src_stride, src0, src1, src2, src3); + VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1); + VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3); + DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1, + vec2, vec3); + SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS); + MIN_UH4_UH(vec0, vec1, vec2, vec3, const255); + PCKEV_B2_SB(vec1, vec0, vec3, vec2, src0, src1); + ST8x4_UB(src0, src1, dst, dst_stride); } static void common_hz_2t_8x8mult_msa(const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride, int8_t *filter, int32_t height) { v16u8 filt0; - v16i8 src0, src1, src2, src3, mask; - v8u16 vec0, vec1, vec2, vec3; - v8u16 filt, const255; + v16i8 src0, src1, src2, src3, mask, out0, out1; + v8u16 vec0, vec1, vec2, vec3, filt, const255; - mask = LOAD_SB(&mc_filt_mask_arr[0]); + mask = LD_SB(&mc_filt_mask_arr[0]); /* rearranging filter */ - filt = LOAD_UH(filter); - filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0); + filt = LD_UH(filter); + filt0 = (v16u8) __msa_splati_h((v8i16) filt, 0); - const255 = (v8u16)__msa_ldi_h(255); + const255 = (v8u16) __msa_ldi_h(255); - LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); + LD_SB4(src, src_stride, src0, src1, src2, src3); src += (4 * src_stride); - vec0 = (v8u16)__msa_vshf_b(mask, src0, src0); - vec1 = (v8u16)__msa_vshf_b(mask, src1, src1); - vec2 = (v8u16)__msa_vshf_b(mask, src2, src2); - vec3 = (v8u16)__msa_vshf_b(mask, src3, src3); + VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1); + VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3); + DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1, + vec2, vec3); + SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS); + MIN_UH4_UH(vec0, vec1, vec2, vec3, const255); - vec0 = __msa_dotp_u_h((v16u8)vec0, filt0); - vec1 = __msa_dotp_u_h((v16u8)vec1, filt0); - vec2 = __msa_dotp_u_h((v16u8)vec2, filt0); - vec3 = __msa_dotp_u_h((v16u8)vec3, filt0); - - SRARI_H_4VECS_UH(vec0, vec1, vec2, vec3, vec0, vec1, vec2, vec3, FILTER_BITS); - - vec0 = __msa_min_u_h(vec0, const255); - vec1 = __msa_min_u_h(vec1, const255); - vec2 = __msa_min_u_h(vec2, const255); - vec3 = __msa_min_u_h(vec3, const255); - - LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); + LD_SB4(src, src_stride, src0, src1, src2, src3); src += (4 * src_stride); - PCKEV_B_STORE_8_BYTES_4(vec0, vec1, vec2, vec3, dst, dst_stride); + PCKEV_B2_SB(vec1, vec0, vec3, vec2, out0, out1); + ST8x4_UB(out0, out1, dst, dst_stride); dst += (4 * dst_stride); - vec0 = (v8u16)__msa_vshf_b(mask, src0, src0); - vec1 = (v8u16)__msa_vshf_b(mask, src1, src1); - vec2 = (v8u16)__msa_vshf_b(mask, src2, src2); - vec3 = (v8u16)__msa_vshf_b(mask, src3, src3); - - vec0 = __msa_dotp_u_h((v16u8)vec0, filt0); - vec1 = __msa_dotp_u_h((v16u8)vec1, filt0); - vec2 = __msa_dotp_u_h((v16u8)vec2, filt0); - vec3 = __msa_dotp_u_h((v16u8)vec3, filt0); - - SRARI_H_4VECS_UH(vec0, vec1, vec2, vec3, vec0, vec1, vec2, vec3, FILTER_BITS); - - vec0 = __msa_min_u_h(vec0, const255); - vec1 = __msa_min_u_h(vec1, const255); - vec2 = __msa_min_u_h(vec2, const255); - vec3 = __msa_min_u_h(vec3, const255); - - PCKEV_B_STORE_8_BYTES_4(vec0, vec1, vec2, vec3, dst, dst_stride); + VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1); + VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3); + DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1, + vec2, vec3); + SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS); + MIN_UH4_UH(vec0, vec1, vec2, vec3, const255); + PCKEV_B2_SB(vec1, vec0, vec3, vec2, out0, out1); + ST8x4_UB(out0, out1, dst, dst_stride); dst += (4 * dst_stride); if (16 == height) { - LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); + LD_SB4(src, src_stride, src0, src1, src2, src3); src += (4 * src_stride); - vec0 = (v8u16)__msa_vshf_b(mask, src0, src0); - vec1 = (v8u16)__msa_vshf_b(mask, src1, src1); - vec2 = (v8u16)__msa_vshf_b(mask, src2, src2); - vec3 = (v8u16)__msa_vshf_b(mask, src3, src3); - - vec0 = __msa_dotp_u_h((v16u8)vec0, filt0); - vec1 = __msa_dotp_u_h((v16u8)vec1, filt0); - vec2 = __msa_dotp_u_h((v16u8)vec2, filt0); - vec3 = __msa_dotp_u_h((v16u8)vec3, filt0); - - SRARI_H_4VECS_UH(vec0, vec1, vec2, vec3, - vec0, vec1, vec2, vec3, FILTER_BITS); - - vec0 = __msa_min_u_h(vec0, const255); - vec1 = __msa_min_u_h(vec1, const255); - vec2 = __msa_min_u_h(vec2, const255); - vec3 = __msa_min_u_h(vec3, const255); - - LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); + VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1); + VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3); + DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1, + vec2, vec3); + SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS); + MIN_UH4_UH(vec0, vec1, vec2, vec3, const255); + LD_SB4(src, src_stride, src0, src1, src2, src3); src += (4 * src_stride); - PCKEV_B_STORE_8_BYTES_4(vec0, vec1, vec2, vec3, dst, dst_stride); - dst += (4 * dst_stride); + PCKEV_B2_SB(vec1, vec0, vec3, vec2, out0, out1); + ST8x4_UB(out0, out1, dst, dst_stride); - vec0 = (v8u16)__msa_vshf_b(mask, src0, src0); - vec1 = (v8u16)__msa_vshf_b(mask, src1, src1); - vec2 = (v8u16)__msa_vshf_b(mask, src2, src2); - vec3 = (v8u16)__msa_vshf_b(mask, src3, src3); - - vec0 = __msa_dotp_u_h((v16u8)vec0, filt0); - vec1 = __msa_dotp_u_h((v16u8)vec1, filt0); - vec2 = __msa_dotp_u_h((v16u8)vec2, filt0); - vec3 = __msa_dotp_u_h((v16u8)vec3, filt0); - - SRARI_H_4VECS_UH(vec0, vec1, vec2, vec3, - vec0, vec1, vec2, vec3, FILTER_BITS); - - vec0 = __msa_min_u_h(vec0, const255); - vec1 = __msa_min_u_h(vec1, const255); - vec2 = __msa_min_u_h(vec2, const255); - vec3 = __msa_min_u_h(vec3, const255); - - PCKEV_B_STORE_8_BYTES_4(vec0, vec1, vec2, vec3, dst, dst_stride); + VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1); + VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3); + DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1, + vec2, vec3); + SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS); + MIN_UH4_UH(vec0, vec1, vec2, vec3, const255); + PCKEV_B2_SB(vec1, vec0, vec3, vec2, out0, out1); + ST8x4_UB(out0, out1, dst + 4 * dst_stride, dst_stride); } } @@ -668,136 +487,68 @@ static void common_hz_2t_16w_msa(const uint8_t *src, int32_t src_stride, int8_t *filter, int32_t height) { uint32_t loop_cnt; v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask; - v16u8 filt0; - v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7; - v8u16 out0, out1, out2, out3, out4, out5, out6, out7; - v8u16 filt, const255; + v16u8 filt0, vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7; + v8u16 out0, out1, out2, out3, out4, out5, out6, out7, filt, const255; - mask = LOAD_SB(&mc_filt_mask_arr[0]); + mask = LD_SB(&mc_filt_mask_arr[0]); loop_cnt = (height >> 2) - 1; /* rearranging filter */ - filt = LOAD_UH(filter); - filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0); + filt = LD_UH(filter); + filt0 = (v16u8) __msa_splati_h((v8i16) filt, 0); - const255 = (v8u16)__msa_ldi_h(255); + const255 = (v8u16) __msa_ldi_h(255); - src0 = LOAD_SB(src); - src1 = LOAD_SB(src + 8); - src += src_stride; - src2 = LOAD_SB(src); - src3 = LOAD_SB(src + 8); - src += src_stride; - src4 = LOAD_SB(src); - src5 = LOAD_SB(src + 8); - src += src_stride; - src6 = LOAD_SB(src); - src7 = LOAD_SB(src + 8); - src += src_stride; + LD_SB4(src, src_stride, src0, src2, src4, src6); + LD_SB4(src + 8, src_stride, src1, src3, src5, src7); + src += (4 * src_stride); - vec0 = (v16u8)__msa_vshf_b(mask, src0, src0); - vec1 = (v16u8)__msa_vshf_b(mask, src1, src1); - vec2 = (v16u8)__msa_vshf_b(mask, src2, src2); - vec3 = (v16u8)__msa_vshf_b(mask, src3, src3); - vec4 = (v16u8)__msa_vshf_b(mask, src4, src4); - vec5 = (v16u8)__msa_vshf_b(mask, src5, src5); - vec6 = (v16u8)__msa_vshf_b(mask, src6, src6); - vec7 = (v16u8)__msa_vshf_b(mask, src7, src7); - - out0 = __msa_dotp_u_h(vec0, filt0); - out1 = __msa_dotp_u_h(vec1, filt0); - out2 = __msa_dotp_u_h(vec2, filt0); - out3 = __msa_dotp_u_h(vec3, filt0); - out4 = __msa_dotp_u_h(vec4, filt0); - out5 = __msa_dotp_u_h(vec5, filt0); - out6 = __msa_dotp_u_h(vec6, filt0); - out7 = __msa_dotp_u_h(vec7, filt0); - - out0 = (v8u16)__msa_srari_h((v8i16)out0, FILTER_BITS); - out1 = (v8u16)__msa_srari_h((v8i16)out1, FILTER_BITS); - out2 = (v8u16)__msa_srari_h((v8i16)out2, FILTER_BITS); - out3 = (v8u16)__msa_srari_h((v8i16)out3, FILTER_BITS); - out4 = (v8u16)__msa_srari_h((v8i16)out4, FILTER_BITS); - out5 = (v8u16)__msa_srari_h((v8i16)out5, FILTER_BITS); - out6 = (v8u16)__msa_srari_h((v8i16)out6, FILTER_BITS); - out7 = (v8u16)__msa_srari_h((v8i16)out7, FILTER_BITS); - - out0 = __msa_min_u_h(out0, const255); - out1 = __msa_min_u_h(out1, const255); - out2 = __msa_min_u_h(out2, const255); - out3 = __msa_min_u_h(out3, const255); - out4 = __msa_min_u_h(out4, const255); - out5 = __msa_min_u_h(out5, const255); - out6 = __msa_min_u_h(out6, const255); - out7 = __msa_min_u_h(out7, const255); - - PCKEV_B_STORE_VEC(out1, out0, dst); + VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1); + VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3); + VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5); + VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7); + DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, out0, out1, + out2, out3); + DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, out4, out5, + out6, out7); + SRARI_H4_UH(out0, out1, out2, out3, FILTER_BITS); + SRARI_H4_UH(out4, out5, out6, out7, FILTER_BITS); + MIN_UH4_UH(out0, out1, out2, out3, const255); + MIN_UH4_UH(out4, out5, out6, out7, const255); + PCKEV_ST_SB(out0, out1, dst); dst += dst_stride; - PCKEV_B_STORE_VEC(out3, out2, dst); + PCKEV_ST_SB(out2, out3, dst); dst += dst_stride; - PCKEV_B_STORE_VEC(out5, out4, dst); + PCKEV_ST_SB(out4, out5, dst); dst += dst_stride; - PCKEV_B_STORE_VEC(out7, out6, dst); + PCKEV_ST_SB(out6, out7, dst); dst += dst_stride; for (; loop_cnt--;) { - src0 = LOAD_SB(src); - src1 = LOAD_SB(src + 8); - src += src_stride; - src2 = LOAD_SB(src); - src3 = LOAD_SB(src + 8); - src += src_stride; - src4 = LOAD_SB(src); - src5 = LOAD_SB(src + 8); - src += src_stride; - src6 = LOAD_SB(src); - src7 = LOAD_SB(src + 8); - src += src_stride; + LD_SB4(src, src_stride, src0, src2, src4, src6); + LD_SB4(src + 8, src_stride, src1, src3, src5, src7); + src += (4 * src_stride); - vec0 = (v16u8)__msa_vshf_b(mask, src0, src0); - vec1 = (v16u8)__msa_vshf_b(mask, src1, src1); - vec2 = (v16u8)__msa_vshf_b(mask, src2, src2); - vec3 = (v16u8)__msa_vshf_b(mask, src3, src3); - vec4 = (v16u8)__msa_vshf_b(mask, src4, src4); - vec5 = (v16u8)__msa_vshf_b(mask, src5, src5); - vec6 = (v16u8)__msa_vshf_b(mask, src6, src6); - vec7 = (v16u8)__msa_vshf_b(mask, src7, src7); - - out0 = __msa_dotp_u_h(vec0, filt0); - out1 = __msa_dotp_u_h(vec1, filt0); - out2 = __msa_dotp_u_h(vec2, filt0); - out3 = __msa_dotp_u_h(vec3, filt0); - out4 = __msa_dotp_u_h(vec4, filt0); - out5 = __msa_dotp_u_h(vec5, filt0); - out6 = __msa_dotp_u_h(vec6, filt0); - out7 = __msa_dotp_u_h(vec7, filt0); - - out0 = (v8u16)__msa_srari_h((v8i16)out0, FILTER_BITS); - out1 = (v8u16)__msa_srari_h((v8i16)out1, FILTER_BITS); - out2 = (v8u16)__msa_srari_h((v8i16)out2, FILTER_BITS); - out3 = (v8u16)__msa_srari_h((v8i16)out3, FILTER_BITS); - out4 = (v8u16)__msa_srari_h((v8i16)out4, FILTER_BITS); - out5 = (v8u16)__msa_srari_h((v8i16)out5, FILTER_BITS); - out6 = (v8u16)__msa_srari_h((v8i16)out6, FILTER_BITS); - out7 = (v8u16)__msa_srari_h((v8i16)out7, FILTER_BITS); - - out0 = __msa_min_u_h(out0, const255); - out1 = __msa_min_u_h(out1, const255); - out2 = __msa_min_u_h(out2, const255); - out3 = __msa_min_u_h(out3, const255); - out4 = __msa_min_u_h(out4, const255); - out5 = __msa_min_u_h(out5, const255); - out6 = __msa_min_u_h(out6, const255); - out7 = __msa_min_u_h(out7, const255); - - PCKEV_B_STORE_VEC(out1, out0, dst); + VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1); + VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3); + VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5); + VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7); + DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, out0, out1, + out2, out3); + DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, out4, out5, + out6, out7); + SRARI_H4_UH(out0, out1, out2, out3, FILTER_BITS); + SRARI_H4_UH(out4, out5, out6, out7, FILTER_BITS); + MIN_UH4_UH(out0, out1, out2, out3, const255); + MIN_UH4_UH(out4, out5, out6, out7, const255); + PCKEV_ST_SB(out0, out1, dst); dst += dst_stride; - PCKEV_B_STORE_VEC(out3, out2, dst); + PCKEV_ST_SB(out2, out3, dst); dst += dst_stride; - PCKEV_B_STORE_VEC(out5, out4, dst); + PCKEV_ST_SB(out4, out5, dst); dst += dst_stride; - PCKEV_B_STORE_VEC(out7, out6, dst); + PCKEV_ST_SB(out6, out7, dst); dst += dst_stride; } } @@ -807,72 +558,46 @@ static void common_hz_2t_32w_msa(const uint8_t *src, int32_t src_stride, int8_t *filter, int32_t height) { uint32_t loop_cnt; v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask; - v16u8 filt0; - v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7; - v8u16 out0, out1, out2, out3, out4, out5, out6, out7; - v8u16 filt, const255; + v16u8 filt0, vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7; + v8u16 out0, out1, out2, out3, out4, out5, out6, out7, filt, const255; - mask = LOAD_SB(&mc_filt_mask_arr[0]); + mask = LD_SB(&mc_filt_mask_arr[0]); /* rearranging filter */ - filt = LOAD_UH(filter); - filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0); + filt = LD_UH(filter); + filt0 = (v16u8) __msa_splati_h((v8i16) filt, 0); - const255 = (v8u16)__msa_ldi_h(255); + const255 = (v8u16) __msa_ldi_h(255); for (loop_cnt = height >> 1; loop_cnt--;) { - src0 = LOAD_SB(src); - src2 = LOAD_SB(src + 16); - src3 = LOAD_SB(src + 24); - src1 = __msa_sld_b(src2, src0, 8); + src0 = LD_SB(src); + src2 = LD_SB(src + 16); + src3 = LD_SB(src + 24); + src1 = __msa_sldi_b(src2, src0, 8); src += src_stride; - src4 = LOAD_SB(src); - src6 = LOAD_SB(src + 16); - src7 = LOAD_SB(src + 24); - src5 = __msa_sld_b(src6, src4, 8); + src4 = LD_SB(src); + src6 = LD_SB(src + 16); + src7 = LD_SB(src + 24); + src5 = __msa_sldi_b(src6, src4, 8); src += src_stride; - vec0 = (v16u8)__msa_vshf_b(mask, src0, src0); - vec1 = (v16u8)__msa_vshf_b(mask, src1, src1); - vec2 = (v16u8)__msa_vshf_b(mask, src2, src2); - vec3 = (v16u8)__msa_vshf_b(mask, src3, src3); - vec4 = (v16u8)__msa_vshf_b(mask, src4, src4); - vec5 = (v16u8)__msa_vshf_b(mask, src5, src5); - vec6 = (v16u8)__msa_vshf_b(mask, src6, src6); - vec7 = (v16u8)__msa_vshf_b(mask, src7, src7); - - out0 = __msa_dotp_u_h(vec0, filt0); - out1 = __msa_dotp_u_h(vec1, filt0); - out2 = __msa_dotp_u_h(vec2, filt0); - out3 = __msa_dotp_u_h(vec3, filt0); - out4 = __msa_dotp_u_h(vec4, filt0); - out5 = __msa_dotp_u_h(vec5, filt0); - out6 = __msa_dotp_u_h(vec6, filt0); - out7 = __msa_dotp_u_h(vec7, filt0); - - out0 = (v8u16)__msa_srari_h((v8i16)out0, FILTER_BITS); - out1 = (v8u16)__msa_srari_h((v8i16)out1, FILTER_BITS); - out2 = (v8u16)__msa_srari_h((v8i16)out2, FILTER_BITS); - out3 = (v8u16)__msa_srari_h((v8i16)out3, FILTER_BITS); - out4 = (v8u16)__msa_srari_h((v8i16)out4, FILTER_BITS); - out5 = (v8u16)__msa_srari_h((v8i16)out5, FILTER_BITS); - out6 = (v8u16)__msa_srari_h((v8i16)out6, FILTER_BITS); - out7 = (v8u16)__msa_srari_h((v8i16)out7, FILTER_BITS); - - out0 = __msa_min_u_h(out0, const255); - out1 = __msa_min_u_h(out1, const255); - out2 = __msa_min_u_h(out2, const255); - out3 = __msa_min_u_h(out3, const255); - out4 = __msa_min_u_h(out4, const255); - out5 = __msa_min_u_h(out5, const255); - out6 = __msa_min_u_h(out6, const255); - out7 = __msa_min_u_h(out7, const255); - - PCKEV_B_STORE_VEC(out1, out0, dst); - PCKEV_B_STORE_VEC(out3, out2, dst + 16); + VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1); + VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3); + VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5); + VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7); + DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, out0, out1, + out2, out3); + DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, out4, out5, + out6, out7); + SRARI_H4_UH(out0, out1, out2, out3, FILTER_BITS); + SRARI_H4_UH(out4, out5, out6, out7, FILTER_BITS); + MIN_UH4_UH(out0, out1, out2, out3, const255); + MIN_UH4_UH(out4, out5, out6, out7, const255); + PCKEV_ST_SB(out0, out1, dst); + PCKEV_ST_SB(out2, out3, dst + 16); dst += dst_stride; - PCKEV_B_STORE_VEC(out5, out4, dst); - PCKEV_B_STORE_VEC(out7, out6, dst + 16); + PCKEV_ST_SB(out4, out5, dst); + PCKEV_ST_SB(out6, out7, dst + 16); dst += dst_stride; } } @@ -882,70 +607,42 @@ static void common_hz_2t_64w_msa(const uint8_t *src, int32_t src_stride, int8_t *filter, int32_t height) { uint32_t loop_cnt; v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask; - v16u8 filt0; - v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7; - v8u16 out0, out1, out2, out3, out4, out5, out6, out7; - v8u16 filt, const255; + v16u8 filt0, vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7; + v8u16 out0, out1, out2, out3, out4, out5, out6, out7, filt, const255; - mask = LOAD_SB(&mc_filt_mask_arr[0]); + mask = LD_SB(&mc_filt_mask_arr[0]); /* rearranging filter */ - filt = LOAD_UH(filter); - filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0); + filt = LD_UH(filter); + filt0 = (v16u8) __msa_splati_h((v8i16) filt, 0); - const255 = (v8u16)__msa_ldi_h(255); + const255 = (v8u16) __msa_ldi_h(255); for (loop_cnt = height; loop_cnt--;) { - src0 = LOAD_SB(src); - src2 = LOAD_SB(src + 16); - src4 = LOAD_SB(src + 32); - src6 = LOAD_SB(src + 48); - src7 = LOAD_SB(src + 56); - src1 = __msa_sld_b(src2, src0, 8); - src3 = __msa_sld_b(src4, src2, 8); - src5 = __msa_sld_b(src6, src4, 8); + src0 = LD_SB(src); + src2 = LD_SB(src + 16); + src4 = LD_SB(src + 32); + src6 = LD_SB(src + 48); + src7 = LD_SB(src + 56); + SLDI_B3_SB(src2, src4, src6, src0, src2, src4, src1, src3, src5, 8); src += src_stride; - vec0 = (v16u8)__msa_vshf_b(mask, src0, src0); - vec1 = (v16u8)__msa_vshf_b(mask, src1, src1); - vec2 = (v16u8)__msa_vshf_b(mask, src2, src2); - vec3 = (v16u8)__msa_vshf_b(mask, src3, src3); - vec4 = (v16u8)__msa_vshf_b(mask, src4, src4); - vec5 = (v16u8)__msa_vshf_b(mask, src5, src5); - vec6 = (v16u8)__msa_vshf_b(mask, src6, src6); - vec7 = (v16u8)__msa_vshf_b(mask, src7, src7); - - out0 = __msa_dotp_u_h(vec0, filt0); - out1 = __msa_dotp_u_h(vec1, filt0); - out2 = __msa_dotp_u_h(vec2, filt0); - out3 = __msa_dotp_u_h(vec3, filt0); - out4 = __msa_dotp_u_h(vec4, filt0); - out5 = __msa_dotp_u_h(vec5, filt0); - out6 = __msa_dotp_u_h(vec6, filt0); - out7 = __msa_dotp_u_h(vec7, filt0); - - out0 = (v8u16)__msa_srari_h((v8i16)out0, FILTER_BITS); - out1 = (v8u16)__msa_srari_h((v8i16)out1, FILTER_BITS); - out2 = (v8u16)__msa_srari_h((v8i16)out2, FILTER_BITS); - out3 = (v8u16)__msa_srari_h((v8i16)out3, FILTER_BITS); - out4 = (v8u16)__msa_srari_h((v8i16)out4, FILTER_BITS); - out5 = (v8u16)__msa_srari_h((v8i16)out5, FILTER_BITS); - out6 = (v8u16)__msa_srari_h((v8i16)out6, FILTER_BITS); - out7 = (v8u16)__msa_srari_h((v8i16)out7, FILTER_BITS); - - out0 = __msa_min_u_h(out0, const255); - out1 = __msa_min_u_h(out1, const255); - out2 = __msa_min_u_h(out2, const255); - out3 = __msa_min_u_h(out3, const255); - out4 = __msa_min_u_h(out4, const255); - out5 = __msa_min_u_h(out5, const255); - out6 = __msa_min_u_h(out6, const255); - out7 = __msa_min_u_h(out7, const255); - - PCKEV_B_STORE_VEC(out1, out0, dst); - PCKEV_B_STORE_VEC(out3, out2, dst + 16); - PCKEV_B_STORE_VEC(out5, out4, dst + 32); - PCKEV_B_STORE_VEC(out7, out6, dst + 48); + VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1); + VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3); + VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5); + VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7); + DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, out0, out1, + out2, out3); + DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, out4, out5, + out6, out7); + SRARI_H4_UH(out0, out1, out2, out3, FILTER_BITS); + SRARI_H4_UH(out4, out5, out6, out7, FILTER_BITS); + MIN_UH4_UH(out0, out1, out2, out3, const255); + MIN_UH4_UH(out4, out5, out6, out7, const255); + PCKEV_ST_SB(out0, out1, dst); + PCKEV_ST_SB(out2, out3, dst + 16); + PCKEV_ST_SB(out4, out5, dst + 32); + PCKEV_ST_SB(out6, out7, dst + 48); dst += dst_stride; } } diff --git a/vp9/common/mips/msa/vp9_convolve8_msa.c b/vp9/common/mips/msa/vp9_convolve8_msa.c index d0c374648..b1279d97c 100644 --- a/vp9/common/mips/msa/vp9_convolve8_msa.c +++ b/vp9/common/mips/msa/vp9_convolve8_msa.c @@ -26,93 +26,68 @@ static void common_hv_8ht_8vt_4w_msa(const uint8_t *src, int32_t src_stride, int32_t height) { uint32_t loop_cnt; v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10; - v16i8 filt_horiz0, filt_horiz1, filt_horiz2, filt_horiz3; - v16u8 mask0, mask1, mask2, mask3; - v8i16 filt_horiz; - v8i16 horiz_out0, horiz_out1, horiz_out2, horiz_out3, horiz_out4; - v8i16 horiz_out5, horiz_out6, horiz_out7, horiz_out8, horiz_out9; - v8i16 tmp0, tmp1, out0, out1, out2, out3, out4; - v8i16 filt, filt_vert0, filt_vert1, filt_vert2, filt_vert3; - - mask0 = LOAD_UB(&mc_filt_mask_arr[16]); + v16i8 filt_hz0, filt_hz1, filt_hz2, filt_hz3; + v16u8 mask0, mask1, mask2, mask3, out; + v8i16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, hz_out5, hz_out6; + v8i16 hz_out7, hz_out8, hz_out9, tmp0, tmp1, out0, out1, out2, out3, out4; + v8i16 filt, filt_vt0, filt_vt1, filt_vt2, filt_vt3; + mask0 = LD_UB(&mc_filt_mask_arr[16]); src -= (3 + 3 * src_stride); /* rearranging filter */ - filt_horiz = LOAD_SH(filter_horiz); - filt_horiz0 = (v16i8)__msa_splati_h(filt_horiz, 0); - filt_horiz1 = (v16i8)__msa_splati_h(filt_horiz, 1); - filt_horiz2 = (v16i8)__msa_splati_h(filt_horiz, 2); - filt_horiz3 = (v16i8)__msa_splati_h(filt_horiz, 3); + filt = LD_SH(filter_horiz); + SPLATI_H4_SB(filt, 0, 1, 2, 3, filt_hz0, filt_hz1, filt_hz2, filt_hz3); mask1 = mask0 + 2; mask2 = mask0 + 4; mask3 = mask0 + 6; - LOAD_7VECS_SB(src, src_stride, src0, src1, src2, src3, src4, src5, src6); + LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6); + XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6); src += (7 * src_stride); - XORI_B_7VECS_SB(src0, src1, src2, src3, src4, src5, src6, - src0, src1, src2, src3, src4, src5, src6, 128); + hz_out0 = HORIZ_8TAP_FILT(src0, src1, mask0, mask1, mask2, mask3, filt_hz0, + filt_hz1, filt_hz2, filt_hz3); + hz_out2 = HORIZ_8TAP_FILT(src2, src3, mask0, mask1, mask2, mask3, filt_hz0, + filt_hz1, filt_hz2, filt_hz3); + hz_out4 = HORIZ_8TAP_FILT(src4, src5, mask0, mask1, mask2, mask3, filt_hz0, + filt_hz1, filt_hz2, filt_hz3); + hz_out5 = HORIZ_8TAP_FILT(src5, src6, mask0, mask1, mask2, mask3, filt_hz0, + filt_hz1, filt_hz2, filt_hz3); + SLDI_B2_SH(hz_out2, hz_out4, hz_out0, hz_out2, hz_out1, hz_out3, 8); - horiz_out0 = HORIZ_8TAP_FILT_2VECS(src0, src1, mask0, mask1, mask2, mask3, - filt_horiz0, filt_horiz1, filt_horiz2, - filt_horiz3); - horiz_out2 = HORIZ_8TAP_FILT_2VECS(src2, src3, mask0, mask1, mask2, mask3, - filt_horiz0, filt_horiz1, filt_horiz2, - filt_horiz3); - horiz_out4 = HORIZ_8TAP_FILT_2VECS(src4, src5, mask0, mask1, mask2, mask3, - filt_horiz0, filt_horiz1, filt_horiz2, - filt_horiz3); - horiz_out5 = HORIZ_8TAP_FILT_2VECS(src5, src6, mask0, mask1, mask2, mask3, - filt_horiz0, filt_horiz1, filt_horiz2, - filt_horiz3); - horiz_out1 = (v8i16)__msa_sldi_b((v16i8)horiz_out2, (v16i8)horiz_out0, 8); - horiz_out3 = (v8i16)__msa_sldi_b((v16i8)horiz_out4, (v16i8)horiz_out2, 8); + filt = LD_SH(filter_vert); + SPLATI_H4_SH(filt, 0, 1, 2, 3, filt_vt0, filt_vt1, filt_vt2, filt_vt3); - filt = LOAD_SH(filter_vert); - filt_vert0 = __msa_splati_h(filt, 0); - filt_vert1 = __msa_splati_h(filt, 1); - filt_vert2 = __msa_splati_h(filt, 2); - filt_vert3 = __msa_splati_h(filt, 3); - - out0 = (v8i16)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); - out1 = (v8i16)__msa_ilvev_b((v16i8)horiz_out3, (v16i8)horiz_out2); - out2 = (v8i16)__msa_ilvev_b((v16i8)horiz_out5, (v16i8)horiz_out4); + ILVEV_B2_SH(hz_out0, hz_out1, hz_out2, hz_out3, out0, out1); + out2 = (v8i16)__msa_ilvev_b((v16i8)hz_out5, (v16i8)hz_out4); for (loop_cnt = (height >> 2); loop_cnt--;) { - LOAD_4VECS_SB(src, src_stride, src7, src8, src9, src10); + LD_SB4(src, src_stride, src7, src8, src9, src10); + XORI_B4_128_SB(src7, src8, src9, src10); src += (4 * src_stride); - XORI_B_4VECS_SB(src7, src8, src9, src10, src7, src8, src9, src10, 128); + hz_out7 = HORIZ_8TAP_FILT(src7, src8, mask0, mask1, mask2, mask3, + filt_hz0, filt_hz1, filt_hz2, filt_hz3); + hz_out6 = (v8i16)__msa_sldi_b((v16i8)hz_out7, (v16i8)hz_out5, 8); + out3 = (v8i16)__msa_ilvev_b((v16i8)hz_out7, (v16i8)hz_out6); + tmp0 = FILT_8TAP_DPADD_S_H(out0, out1, out2, out3, filt_vt0, filt_vt1, + filt_vt2, filt_vt3); - horiz_out7 = HORIZ_8TAP_FILT_2VECS(src7, src8, mask0, mask1, mask2, mask3, - filt_horiz0, filt_horiz1, filt_horiz2, - filt_horiz3); - horiz_out6 = (v8i16)__msa_sldi_b((v16i8)horiz_out7, (v16i8)horiz_out5, 8); - - out3 = (v8i16)__msa_ilvev_b((v16i8)horiz_out7, (v16i8)horiz_out6); - - tmp0 = FILT_8TAP_DPADD_S_H(out0, out1, out2, out3, filt_vert0, filt_vert1, - filt_vert2, filt_vert3); - - horiz_out9 = HORIZ_8TAP_FILT_2VECS(src9, src10, mask0, mask1, mask2, mask3, - filt_horiz0, filt_horiz1, filt_horiz2, - filt_horiz3); - horiz_out8 = (v8i16)__msa_sldi_b((v16i8)horiz_out9, (v16i8)horiz_out7, 8); - - out4 = (v8i16)__msa_ilvev_b((v16i8)horiz_out9, (v16i8)horiz_out8); - - tmp1 = FILT_8TAP_DPADD_S_H(out1, out2, out3, out4, filt_vert0, filt_vert1, - filt_vert2, filt_vert3); - tmp0 = SRARI_SATURATE_SIGNED_H(tmp0, FILTER_BITS, 7); - tmp1 = SRARI_SATURATE_SIGNED_H(tmp1, FILTER_BITS, 7); - - PCKEV_2B_XORI128_STORE_4_BYTES_4(tmp0, tmp1, dst, dst_stride); + hz_out9 = HORIZ_8TAP_FILT(src9, src10, mask0, mask1, mask2, mask3, + filt_hz0, filt_hz1, filt_hz2, filt_hz3); + hz_out8 = (v8i16)__msa_sldi_b((v16i8)hz_out9, (v16i8)hz_out7, 8); + out4 = (v8i16)__msa_ilvev_b((v16i8)hz_out9, (v16i8)hz_out8); + tmp1 = FILT_8TAP_DPADD_S_H(out1, out2, out3, out4, filt_vt0, filt_vt1, + filt_vt2, filt_vt3); + SRARI_H2_SH(tmp0, tmp1, FILTER_BITS); + SAT_SH2_SH(tmp0, tmp1, 7); + out = PCKEV_XORI128_UB(tmp0, tmp1); + ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride); dst += (4 * dst_stride); - horiz_out5 = horiz_out9; - + hz_out5 = hz_out9; out0 = out2; out1 = out3; out2 = out4; @@ -125,108 +100,87 @@ static void common_hv_8ht_8vt_8w_msa(const uint8_t *src, int32_t src_stride, int32_t height) { uint32_t loop_cnt; v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10; - v16i8 filt_horiz0, filt_horiz1, filt_horiz2, filt_horiz3; - v8i16 filt_horiz, filt, filt_vert0, filt_vert1, filt_vert2, filt_vert3; - v16u8 mask0, mask1, mask2, mask3; - v8i16 horiz_out0, horiz_out1, horiz_out2, horiz_out3; - v8i16 horiz_out4, horiz_out5, horiz_out6, horiz_out7; - v8i16 horiz_out8, horiz_out9, horiz_out10; + v16i8 filt_hz0, filt_hz1, filt_hz2, filt_hz3; + v16u8 mask0, mask1, mask2, mask3, vec0, vec1; + v8i16 filt, filt_vt0, filt_vt1, filt_vt2, filt_vt3; + v8i16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, hz_out5, hz_out6; + v8i16 hz_out7, hz_out8, hz_out9, hz_out10, tmp0, tmp1, tmp2, tmp3; v8i16 out0, out1, out2, out3, out4, out5, out6, out7, out8, out9; - v8i16 tmp0, tmp1, tmp2, tmp3; - - mask0 = LOAD_UB(&mc_filt_mask_arr[0]); + mask0 = LD_UB(&mc_filt_mask_arr[0]); src -= (3 + 3 * src_stride); /* rearranging filter */ - filt_horiz = LOAD_SH(filter_horiz); - filt_horiz0 = (v16i8)__msa_splati_h(filt_horiz, 0); - filt_horiz1 = (v16i8)__msa_splati_h(filt_horiz, 1); - filt_horiz2 = (v16i8)__msa_splati_h(filt_horiz, 2); - filt_horiz3 = (v16i8)__msa_splati_h(filt_horiz, 3); + filt = LD_SH(filter_horiz); + SPLATI_H4_SB(filt, 0, 1, 2, 3, filt_hz0, filt_hz1, filt_hz2, filt_hz3); mask1 = mask0 + 2; mask2 = mask0 + 4; mask3 = mask0 + 6; - LOAD_7VECS_SB(src, src_stride, src0, src1, src2, src3, src4, src5, src6); + LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6); src += (7 * src_stride); - XORI_B_7VECS_SB(src0, src1, src2, src3, src4, src5, src6, - src0, src1, src2, src3, src4, src5, src6, 128); + XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6); + hz_out0 = HORIZ_8TAP_FILT(src0, src0, mask0, mask1, mask2, mask3, filt_hz0, + filt_hz1, filt_hz2, filt_hz3); + hz_out1 = HORIZ_8TAP_FILT(src1, src1, mask0, mask1, mask2, mask3, filt_hz0, + filt_hz1, filt_hz2, filt_hz3); + hz_out2 = HORIZ_8TAP_FILT(src2, src2, mask0, mask1, mask2, mask3, filt_hz0, + filt_hz1, filt_hz2, filt_hz3); + hz_out3 = HORIZ_8TAP_FILT(src3, src3, mask0, mask1, mask2, mask3, filt_hz0, + filt_hz1, filt_hz2, filt_hz3); + hz_out4 = HORIZ_8TAP_FILT(src4, src4, mask0, mask1, mask2, mask3, filt_hz0, + filt_hz1, filt_hz2, filt_hz3); + hz_out5 = HORIZ_8TAP_FILT(src5, src5, mask0, mask1, mask2, mask3, filt_hz0, + filt_hz1, filt_hz2, filt_hz3); + hz_out6 = HORIZ_8TAP_FILT(src6, src6, mask0, mask1, mask2, mask3, filt_hz0, + filt_hz1, filt_hz2, filt_hz3); - horiz_out0 = HORIZ_8TAP_FILT(src0, mask0, mask1, mask2, mask3, filt_horiz0, - filt_horiz1, filt_horiz2, filt_horiz3); - horiz_out1 = HORIZ_8TAP_FILT(src1, mask0, mask1, mask2, mask3, filt_horiz0, - filt_horiz1, filt_horiz2, filt_horiz3); - horiz_out2 = HORIZ_8TAP_FILT(src2, mask0, mask1, mask2, mask3, filt_horiz0, - filt_horiz1, filt_horiz2, filt_horiz3); - horiz_out3 = HORIZ_8TAP_FILT(src3, mask0, mask1, mask2, mask3, filt_horiz0, - filt_horiz1, filt_horiz2, filt_horiz3); - horiz_out4 = HORIZ_8TAP_FILT(src4, mask0, mask1, mask2, mask3, filt_horiz0, - filt_horiz1, filt_horiz2, filt_horiz3); - horiz_out5 = HORIZ_8TAP_FILT(src5, mask0, mask1, mask2, mask3, filt_horiz0, - filt_horiz1, filt_horiz2, filt_horiz3); - horiz_out6 = HORIZ_8TAP_FILT(src6, mask0, mask1, mask2, mask3, filt_horiz0, - filt_horiz1, filt_horiz2, filt_horiz3); + filt = LD_SH(filter_vert); + SPLATI_H4_SH(filt, 0, 1, 2, 3, filt_vt0, filt_vt1, filt_vt2, filt_vt3); - filt = LOAD_SH(filter_vert); - filt_vert0 = __msa_splati_h(filt, 0); - filt_vert1 = __msa_splati_h(filt, 1); - filt_vert2 = __msa_splati_h(filt, 2); - filt_vert3 = __msa_splati_h(filt, 3); - - out0 = (v8i16)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); - out1 = (v8i16)__msa_ilvev_b((v16i8)horiz_out3, (v16i8)horiz_out2); - out2 = (v8i16)__msa_ilvev_b((v16i8)horiz_out5, (v16i8)horiz_out4); - out4 = (v8i16)__msa_ilvev_b((v16i8)horiz_out2, (v16i8)horiz_out1); - out5 = (v8i16)__msa_ilvev_b((v16i8)horiz_out4, (v16i8)horiz_out3); - out6 = (v8i16)__msa_ilvev_b((v16i8)horiz_out6, (v16i8)horiz_out5); + ILVEV_B2_SH(hz_out0, hz_out1, hz_out2, hz_out3, out0, out1); + ILVEV_B2_SH(hz_out4, hz_out5, hz_out1, hz_out2, out2, out4); + ILVEV_B2_SH(hz_out3, hz_out4, hz_out5, hz_out6, out5, out6); for (loop_cnt = (height >> 2); loop_cnt--;) { - LOAD_4VECS_SB(src, src_stride, src7, src8, src9, src10); + LD_SB4(src, src_stride, src7, src8, src9, src10); src += (4 * src_stride); - XORI_B_4VECS_SB(src7, src8, src9, src10, src7, src8, src9, src10, 128); + XORI_B4_128_SB(src7, src8, src9, src10); - horiz_out7 = HORIZ_8TAP_FILT(src7, mask0, mask1, mask2, mask3, filt_horiz0, - filt_horiz1, filt_horiz2, filt_horiz3); + hz_out7 = HORIZ_8TAP_FILT(src7, src7, mask0, mask1, mask2, mask3, + filt_hz0, filt_hz1, filt_hz2, filt_hz3); + out3 = (v8i16)__msa_ilvev_b((v16i8)hz_out7, (v16i8)hz_out6); + tmp0 = FILT_8TAP_DPADD_S_H(out0, out1, out2, out3, filt_vt0, filt_vt1, + filt_vt2, filt_vt3); - out3 = (v8i16)__msa_ilvev_b((v16i8)horiz_out7, (v16i8)horiz_out6); - tmp0 = FILT_8TAP_DPADD_S_H(out0, out1, out2, out3, filt_vert0, filt_vert1, - filt_vert2, filt_vert3); - tmp0 = SRARI_SATURATE_SIGNED_H(tmp0, FILTER_BITS, 7); + hz_out8 = HORIZ_8TAP_FILT(src8, src8, mask0, mask1, mask2, mask3, + filt_hz0, filt_hz1, filt_hz2, filt_hz3); + out7 = (v8i16)__msa_ilvev_b((v16i8)hz_out8, (v16i8)hz_out7); + tmp1 = FILT_8TAP_DPADD_S_H(out4, out5, out6, out7, filt_vt0, filt_vt1, + filt_vt2, filt_vt3); - horiz_out8 = HORIZ_8TAP_FILT(src8, mask0, mask1, mask2, mask3, filt_horiz0, - filt_horiz1, filt_horiz2, filt_horiz3); + hz_out9 = HORIZ_8TAP_FILT(src9, src9, mask0, mask1, mask2, mask3, + filt_hz0, filt_hz1, filt_hz2, filt_hz3); + out8 = (v8i16)__msa_ilvev_b((v16i8)hz_out9, (v16i8)hz_out8); + tmp2 = FILT_8TAP_DPADD_S_H(out1, out2, out3, out8, filt_vt0, filt_vt1, + filt_vt2, filt_vt3); - out7 = (v8i16)__msa_ilvev_b((v16i8)horiz_out8, (v16i8)horiz_out7); - tmp1 = FILT_8TAP_DPADD_S_H(out4, out5, out6, out7, filt_vert0, filt_vert1, - filt_vert2, filt_vert3); - tmp1 = SRARI_SATURATE_SIGNED_H(tmp1, FILTER_BITS, 7); - - horiz_out9 = HORIZ_8TAP_FILT(src9, mask0, mask1, mask2, mask3, filt_horiz0, - filt_horiz1, filt_horiz2, filt_horiz3); - - out8 = (v8i16)__msa_ilvev_b((v16i8)horiz_out9, (v16i8)horiz_out8); - tmp2 = FILT_8TAP_DPADD_S_H(out1, out2, out3, out8, filt_vert0, filt_vert1, - filt_vert2, filt_vert3); - tmp2 = SRARI_SATURATE_SIGNED_H(tmp2, FILTER_BITS, 7); - - horiz_out10 = HORIZ_8TAP_FILT(src10, mask0, mask1, mask2, mask3, - filt_horiz0, filt_horiz1, filt_horiz2, - filt_horiz3); - - out9 = (v8i16)__msa_ilvev_b((v16i8)horiz_out10, (v16i8)horiz_out9); - tmp3 = FILT_8TAP_DPADD_S_H(out5, out6, out7, out9, filt_vert0, filt_vert1, - filt_vert2, filt_vert3); - tmp3 = SRARI_SATURATE_SIGNED_H(tmp3, FILTER_BITS, 7); - - PCKEV_B_4_XORI128_STORE_8_BYTES_4(tmp0, tmp1, tmp2, tmp3, dst, dst_stride); + hz_out10 = HORIZ_8TAP_FILT(src10, src10, mask0, mask1, mask2, mask3, + filt_hz0, filt_hz1, filt_hz2, filt_hz3); + out9 = (v8i16)__msa_ilvev_b((v16i8)hz_out10, (v16i8)hz_out9); + tmp3 = FILT_8TAP_DPADD_S_H(out5, out6, out7, out9, filt_vt0, filt_vt1, + filt_vt2, filt_vt3); + SRARI_H4_SH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS); + SAT_SH4_SH(tmp0, tmp1, tmp2, tmp3, 7); + vec0 = PCKEV_XORI128_UB(tmp0, tmp1); + vec1 = PCKEV_XORI128_UB(tmp2, tmp3); + ST8x4_UB(vec0, vec1, dst, dst_stride); dst += (4 * dst_stride); - horiz_out6 = horiz_out10; - + hz_out6 = hz_out10; out0 = out2; out1 = out3; out2 = out8; @@ -279,175 +233,89 @@ static void common_hv_2ht_2vt_4x4_msa(const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride, int8_t *filter_horiz, int8_t *filter_vert) { - uint32_t out0, out1, out2, out3; v16i8 src0, src1, src2, src3, src4, mask; - v16u8 res0, res1, horiz_vec; - v16u8 filt_vert, filt_horiz, vec0, vec1; - v8u16 filt, tmp0, tmp1; - v8u16 horiz_out0, horiz_out1, horiz_out2, horiz_out3, horiz_out4; + v16u8 filt_vt, filt_hz, vec0, vec1, res0, res1; + v8u16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, filt, tmp0, tmp1; - mask = LOAD_SB(&mc_filt_mask_arr[16]); + mask = LD_SB(&mc_filt_mask_arr[16]); /* rearranging filter */ - filt = LOAD_UH(filter_horiz); - filt_horiz = (v16u8)__msa_splati_h((v8i16)filt, 0); + filt = LD_UH(filter_horiz); + filt_hz = (v16u8)__msa_splati_h((v8i16)filt, 0); - filt = LOAD_UH(filter_vert); - filt_vert = (v16u8)__msa_splati_h((v8i16)filt, 0); + filt = LD_UH(filter_vert); + filt_vt = (v16u8)__msa_splati_h((v8i16)filt, 0); - LOAD_5VECS_SB(src, src_stride, src0, src1, src2, src3, src4); + LD_SB5(src, src_stride, src0, src1, src2, src3, src4); + hz_out0 = HORIZ_2TAP_FILT_UH(src0, src1, mask, filt_hz, FILTER_BITS); + hz_out2 = HORIZ_2TAP_FILT_UH(src2, src3, mask, filt_hz, FILTER_BITS); + hz_out4 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS); + hz_out1 = (v8u16)__msa_sldi_b((v16i8)hz_out2, (v16i8)hz_out0, 8); + hz_out3 = (v8u16)__msa_pckod_d((v2i64)hz_out4, (v2i64)hz_out2); - horiz_vec = (v16u8)__msa_vshf_b(mask, src1, src0); - horiz_out0 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_out0, FILTER_BITS, 7); - - horiz_vec = (v16u8)__msa_vshf_b(mask, src3, src2); - horiz_out2 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out2 = SRARI_SATURATE_UNSIGNED_H(horiz_out2, FILTER_BITS, 7); - - horiz_vec = (v16u8)__msa_vshf_b(mask, src4, src4); - horiz_out4 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out4 = SRARI_SATURATE_UNSIGNED_H(horiz_out4, FILTER_BITS, 7); - - horiz_out1 = (v8u16)__msa_sldi_b((v16i8)horiz_out2, (v16i8)horiz_out0, 8); - horiz_out3 = (v8u16)__msa_pckod_d((v2i64)horiz_out4, (v2i64)horiz_out2); - - vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); - vec1 = (v16u8)__msa_ilvev_b((v16i8)horiz_out3, (v16i8)horiz_out2); - - tmp0 = __msa_dotp_u_h(vec0, filt_vert); - tmp1 = __msa_dotp_u_h(vec1, filt_vert); - tmp0 = SRARI_SATURATE_UNSIGNED_H(tmp0, FILTER_BITS, 7); - tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); - - res0 = (v16u8)__msa_pckev_b((v16i8)tmp0, (v16i8)tmp0); - res1 = (v16u8)__msa_pckev_b((v16i8)tmp1, (v16i8)tmp1); - - out0 = __msa_copy_u_w((v4i32)res0, 0); - out1 = __msa_copy_u_w((v4i32)res0, 1); - out2 = __msa_copy_u_w((v4i32)res1, 0); - out3 = __msa_copy_u_w((v4i32)res1, 1); - - STORE_WORD(dst, out0); - dst += dst_stride; - STORE_WORD(dst, out1); - dst += dst_stride; - STORE_WORD(dst, out2); - dst += dst_stride; - STORE_WORD(dst, out3); + ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1); + DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp0, tmp1); + SRARI_H2_UH(tmp0, tmp1, FILTER_BITS); + SAT_UH2_UH(tmp0, tmp1, 7); + PCKEV_B2_UB(tmp0, tmp0, tmp1, tmp1, res0, res1); + ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride); } static void common_hv_2ht_2vt_4x8_msa(const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride, int8_t *filter_horiz, int8_t *filter_vert) { - uint32_t out0, out1, out2, out3; v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, mask; - v16u8 filt_horiz, filt_vert, horiz_vec; - v16u8 vec0, vec1, vec2, vec3; - v8u16 horiz_out0, horiz_out1, horiz_out2, horiz_out3; - v8u16 vec4, vec5, vec6, vec7, filt; - v8u16 horiz_out4, horiz_out5, horiz_out6, horiz_out7, horiz_out8; v16i8 res0, res1, res2, res3; + v16u8 filt_hz, filt_vt, vec0, vec1, vec2, vec3; + v8u16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, hz_out5, hz_out6; + v8u16 hz_out7, hz_out8, vec4, vec5, vec6, vec7, filt; - mask = LOAD_SB(&mc_filt_mask_arr[16]); + mask = LD_SB(&mc_filt_mask_arr[16]); /* rearranging filter */ - filt = LOAD_UH(filter_horiz); - filt_horiz = (v16u8)__msa_splati_h((v8i16)filt, 0); + filt = LD_UH(filter_horiz); + filt_hz = (v16u8)__msa_splati_h((v8i16)filt, 0); - filt = LOAD_UH(filter_vert); - filt_vert = (v16u8)__msa_splati_h((v8i16)filt, 0); + filt = LD_UH(filter_vert); + filt_vt = (v16u8)__msa_splati_h((v8i16)filt, 0); - LOAD_8VECS_SB(src, src_stride, - src0, src1, src2, src3, src4, src5, src6, src7); + LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7); src += (8 * src_stride); - src8 = LOAD_SB(src); + src8 = LD_SB(src); - horiz_vec = (v16u8)__msa_vshf_b(mask, src1, src0); - horiz_out0 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_out0, FILTER_BITS, 7); + hz_out0 = HORIZ_2TAP_FILT_UH(src0, src1, mask, filt_hz, FILTER_BITS); + hz_out2 = HORIZ_2TAP_FILT_UH(src2, src3, mask, filt_hz, FILTER_BITS); + hz_out4 = HORIZ_2TAP_FILT_UH(src4, src5, mask, filt_hz, FILTER_BITS); + hz_out6 = HORIZ_2TAP_FILT_UH(src6, src7, mask, filt_hz, FILTER_BITS); + hz_out8 = HORIZ_2TAP_FILT_UH(src8, src8, mask, filt_hz, FILTER_BITS); + SLDI_B3_UH(hz_out2, hz_out4, hz_out6, hz_out0, hz_out2, hz_out4, hz_out1, + hz_out3, hz_out5, 8); + hz_out7 = (v8u16)__msa_pckod_d((v2i64)hz_out8, (v2i64)hz_out6); - horiz_vec = (v16u8)__msa_vshf_b(mask, src3, src2); - horiz_out2 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out2 = SRARI_SATURATE_UNSIGNED_H(horiz_out2, FILTER_BITS, 7); - - horiz_vec = (v16u8)__msa_vshf_b(mask, src5, src4); - horiz_out4 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out4 = SRARI_SATURATE_UNSIGNED_H(horiz_out4, FILTER_BITS, 7); - - horiz_vec = (v16u8)__msa_vshf_b(mask, src7, src6); - horiz_out6 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out6 = SRARI_SATURATE_UNSIGNED_H(horiz_out6, FILTER_BITS, 7); - - horiz_vec = (v16u8)__msa_vshf_b(mask, src8, src8); - horiz_out8 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out8 = SRARI_SATURATE_UNSIGNED_H(horiz_out8, FILTER_BITS, 7); - - horiz_out1 = (v8u16)__msa_sldi_b((v16i8)horiz_out2, (v16i8)horiz_out0, 8); - horiz_out3 = (v8u16)__msa_sldi_b((v16i8)horiz_out4, (v16i8)horiz_out2, 8); - horiz_out5 = (v8u16)__msa_sldi_b((v16i8)horiz_out6, (v16i8)horiz_out4, 8); - horiz_out7 = (v8u16)__msa_pckod_d((v2i64)horiz_out8, (v2i64)horiz_out6); - - vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); - vec1 = (v16u8)__msa_ilvev_b((v16i8)horiz_out3, (v16i8)horiz_out2); - vec2 = (v16u8)__msa_ilvev_b((v16i8)horiz_out5, (v16i8)horiz_out4); - vec3 = (v16u8)__msa_ilvev_b((v16i8)horiz_out7, (v16i8)horiz_out6); - - vec4 = __msa_dotp_u_h(vec0, filt_vert); - vec5 = __msa_dotp_u_h(vec1, filt_vert); - vec6 = __msa_dotp_u_h(vec2, filt_vert); - vec7 = __msa_dotp_u_h(vec3, filt_vert); - - vec4 = SRARI_SATURATE_UNSIGNED_H(vec4, FILTER_BITS, 7); - vec5 = SRARI_SATURATE_UNSIGNED_H(vec5, FILTER_BITS, 7); - vec6 = SRARI_SATURATE_UNSIGNED_H(vec6, FILTER_BITS, 7); - vec7 = SRARI_SATURATE_UNSIGNED_H(vec7, FILTER_BITS, 7); - - res0 = __msa_pckev_b((v16i8)vec4, (v16i8)vec4); - res1 = __msa_pckev_b((v16i8)vec5, (v16i8)vec5); - res2 = __msa_pckev_b((v16i8)vec6, (v16i8)vec6); - res3 = __msa_pckev_b((v16i8)vec7, (v16i8)vec7); - - out0 = __msa_copy_u_w((v4i32)res0, 0); - out1 = __msa_copy_u_w((v4i32)res0, 1); - out2 = __msa_copy_u_w((v4i32)res1, 0); - out3 = __msa_copy_u_w((v4i32)res1, 1); - - STORE_WORD(dst, out0); - dst += dst_stride; - STORE_WORD(dst, out1); - dst += dst_stride; - STORE_WORD(dst, out2); - dst += dst_stride; - STORE_WORD(dst, out3); - dst += dst_stride; - - out0 = __msa_copy_u_w((v4i32)res2, 0); - out1 = __msa_copy_u_w((v4i32)res2, 1); - out2 = __msa_copy_u_w((v4i32)res3, 0); - out3 = __msa_copy_u_w((v4i32)res3, 1); - - STORE_WORD(dst, out0); - dst += dst_stride; - STORE_WORD(dst, out1); - dst += dst_stride; - STORE_WORD(dst, out2); - dst += dst_stride; - STORE_WORD(dst, out3); + ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1); + ILVEV_B2_UB(hz_out4, hz_out5, hz_out6, hz_out7, vec2, vec3); + DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt_vt, filt_vt, filt_vt, filt_vt, + vec4, vec5, vec6, vec7); + SRARI_H4_UH(vec4, vec5, vec6, vec7, FILTER_BITS); + SAT_UH4_UH(vec4, vec5, vec6, vec7, 7); + PCKEV_B4_SB(vec4, vec4, vec5, vec5, vec6, vec6, vec7, vec7, res0, res1, + res2, res3); + ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride); + dst += (4 * dst_stride); + ST4x4_UB(res2, res3, 0, 1, 0, 1, dst, dst_stride); } static void common_hv_2ht_2vt_4w_msa(const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride, - int8_t *filter_horiz, - int8_t *filter_vert, + int8_t *filter_horiz, int8_t *filter_vert, int32_t height) { if (4 == height) { - common_hv_2ht_2vt_4x4_msa(src, src_stride, dst, dst_stride, - filter_horiz, filter_vert); + common_hv_2ht_2vt_4x4_msa(src, src_stride, dst, dst_stride, filter_horiz, + filter_vert); } else if (8 == height) { - common_hv_2ht_2vt_4x8_msa(src, src_stride, dst, dst_stride, - filter_horiz, filter_vert); + common_hv_2ht_2vt_4x8_msa(src, src_stride, dst, dst_stride, filter_horiz, + filter_vert); } } @@ -455,63 +323,43 @@ static void common_hv_2ht_2vt_8x4_msa(const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride, int8_t *filter_horiz, int8_t *filter_vert) { - v16i8 src0, src1, src2, src3, src4, mask; - v16u8 filt_horiz, filt_vert, horiz_vec; - v16u8 vec0, vec1, vec2, vec3; - v8u16 horiz_out0, horiz_out1; - v8u16 tmp0, tmp1, tmp2, tmp3; + v16i8 src0, src1, src2, src3, src4, mask, out0, out1; + v16u8 filt_hz, filt_vt, vec0, vec1, vec2, vec3; + v8u16 hz_out0, hz_out1, tmp0, tmp1, tmp2, tmp3; v8i16 filt; - mask = LOAD_SB(&mc_filt_mask_arr[0]); + mask = LD_SB(&mc_filt_mask_arr[0]); /* rearranging filter */ - filt = LOAD_SH(filter_horiz); - filt_horiz = (v16u8)__msa_splati_h(filt, 0); + filt = LD_SH(filter_horiz); + filt_hz = (v16u8)__msa_splati_h(filt, 0); - filt = LOAD_SH(filter_vert); - filt_vert = (v16u8)__msa_splati_h(filt, 0); + filt = LD_SH(filter_vert); + filt_vt = (v16u8)__msa_splati_h(filt, 0); - LOAD_5VECS_SB(src, src_stride, src0, src1, src2, src3, src4); - src += (5 * src_stride); + LD_SB5(src, src_stride, src0, src1, src2, src3, src4); - horiz_vec = (v16u8)__msa_vshf_b(mask, src0, src0); - horiz_out0 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_out0, FILTER_BITS, 7); + hz_out0 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS); + hz_out1 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS); + vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0); + tmp0 = __msa_dotp_u_h(vec0, filt_vt); - horiz_vec = (v16u8)__msa_vshf_b(mask, src1, src1); - horiz_out1 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out1 = SRARI_SATURATE_UNSIGNED_H(horiz_out1, FILTER_BITS, 7); + hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz, FILTER_BITS); + vec1 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1); + tmp1 = __msa_dotp_u_h(vec1, filt_vt); - vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); - tmp0 = __msa_dotp_u_h(vec0, filt_vert); + hz_out1 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz, FILTER_BITS); + vec2 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0); + tmp2 = __msa_dotp_u_h(vec2, filt_vt); - horiz_vec = (v16u8)__msa_vshf_b(mask, src2, src2); - horiz_out0 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_out0, FILTER_BITS, 7); + hz_out0 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS); + vec3 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1); + tmp3 = __msa_dotp_u_h(vec3, filt_vt); - vec1 = (v16u8)__msa_ilvev_b((v16i8)horiz_out0, (v16i8)horiz_out1); - tmp1 = __msa_dotp_u_h(vec1, filt_vert); - - horiz_vec = (v16u8)__msa_vshf_b(mask, src3, src3); - horiz_out1 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out1 = SRARI_SATURATE_UNSIGNED_H(horiz_out1, FILTER_BITS, 7); - - vec2 = (v16u8)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); - tmp2 = __msa_dotp_u_h(vec2, filt_vert); - - horiz_vec = (v16u8)__msa_vshf_b(mask, src4, src4); - horiz_out0 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_out0, FILTER_BITS, 7); - - vec3 = (v16u8)__msa_ilvev_b((v16i8)horiz_out0, (v16i8)horiz_out1); - tmp3 = __msa_dotp_u_h(vec3, filt_vert); - - tmp0 = SRARI_SATURATE_UNSIGNED_H(tmp0, FILTER_BITS, 7); - tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); - tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); - tmp3 = SRARI_SATURATE_UNSIGNED_H(tmp3, FILTER_BITS, 7); - - PCKEV_B_STORE_8_BYTES_4(tmp0, tmp1, tmp2, tmp3, dst, dst_stride); + SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS); + SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7); + PCKEV_B2_SB(tmp1, tmp0, tmp3, tmp2, out0, out1); + ST8x4_UB(out0, out1, dst, dst_stride); } static void common_hv_2ht_2vt_8x8mult_msa(const uint8_t *src, @@ -522,106 +370,76 @@ static void common_hv_2ht_2vt_8x8mult_msa(const uint8_t *src, int8_t *filter_vert, int32_t height) { uint32_t loop_cnt; - v16i8 src0, src1, src2, src3, src4, mask; - v16u8 filt_horiz, filt_vert, vec0, horiz_vec; - v8u16 horiz_out0, horiz_out1; - v8u16 tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8; + v16i8 src0, src1, src2, src3, src4, mask, out0, out1; + v16u8 filt_hz, filt_vt, vec0; + v8u16 hz_out0, hz_out1, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8; v8i16 filt; - mask = LOAD_SB(&mc_filt_mask_arr[0]); + mask = LD_SB(&mc_filt_mask_arr[0]); /* rearranging filter */ - filt = LOAD_SH(filter_horiz); - filt_horiz = (v16u8)__msa_splati_h(filt, 0); + filt = LD_SH(filter_horiz); + filt_hz = (v16u8)__msa_splati_h(filt, 0); - filt = LOAD_SH(filter_vert); - filt_vert = (v16u8)__msa_splati_h(filt, 0); + filt = LD_SH(filter_vert); + filt_vt = (v16u8)__msa_splati_h(filt, 0); - src0 = LOAD_SB(src); + src0 = LD_SB(src); src += src_stride; - horiz_vec = (v16u8)__msa_vshf_b(mask, src0, src0); - horiz_out0 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_out0, FILTER_BITS, 7); + hz_out0 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS); for (loop_cnt = (height >> 3); loop_cnt--;) { - LOAD_4VECS_SB(src, src_stride, src1, src2, src3, src4); + LD_SB4(src, src_stride, src1, src2, src3, src4); src += (4 * src_stride); - horiz_vec = (v16u8)__msa_vshf_b(mask, src1, src1); - horiz_out1 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out1 = SRARI_SATURATE_UNSIGNED_H(horiz_out1, FILTER_BITS, 7); + hz_out1 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS); + vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0); + tmp1 = __msa_dotp_u_h(vec0, filt_vt); - vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); - tmp1 = __msa_dotp_u_h(vec0, filt_vert); + hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz, FILTER_BITS); + vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1); + tmp2 = __msa_dotp_u_h(vec0, filt_vt); - horiz_vec = (v16u8)__msa_vshf_b(mask, src2, src2); - horiz_out0 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_out0, FILTER_BITS, 7); + SRARI_H2_UH(tmp1, tmp2, FILTER_BITS); + SAT_UH2_UH(tmp1, tmp2, 7); - vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out0, (v16i8)horiz_out1); - tmp2 = (v8u16)__msa_dotp_u_h(vec0, filt_vert); + hz_out1 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz, FILTER_BITS); + vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0); + tmp3 = __msa_dotp_u_h(vec0, filt_vt); - tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); - tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); - - horiz_vec = (v16u8)__msa_vshf_b(mask, src3, src3); - horiz_out1 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out1 = SRARI_SATURATE_UNSIGNED_H(horiz_out1, FILTER_BITS, 7); - - vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); - tmp3 = __msa_dotp_u_h(vec0, filt_vert); - - horiz_vec = (v16u8)__msa_vshf_b(mask, src4, src4); - horiz_out0 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_out0, FILTER_BITS, 7); - - LOAD_4VECS_SB(src, src_stride, src1, src2, src3, src4); + hz_out0 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS); + LD_SB4(src, src_stride, src1, src2, src3, src4); src += (4 * src_stride); + vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1); + tmp4 = __msa_dotp_u_h(vec0, filt_vt); - vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out0, (v16i8)horiz_out1); - tmp4 = __msa_dotp_u_h(vec0, filt_vert); - - tmp3 = SRARI_SATURATE_UNSIGNED_H(tmp3, FILTER_BITS, 7); - tmp4 = SRARI_SATURATE_UNSIGNED_H(tmp4, FILTER_BITS, 7); - - PCKEV_B_STORE_8_BYTES_4(tmp1, tmp2, tmp3, tmp4, dst, dst_stride); + SRARI_H2_UH(tmp3, tmp4, FILTER_BITS); + SAT_UH2_UH(tmp3, tmp4, 7); + PCKEV_B2_SB(tmp2, tmp1, tmp4, tmp3, out0, out1); + ST8x4_UB(out0, out1, dst, dst_stride); dst += (4 * dst_stride); - horiz_vec = (v16u8)__msa_vshf_b(mask, src1, src1); - horiz_out1 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out1 = SRARI_SATURATE_UNSIGNED_H(horiz_out1, FILTER_BITS, 7); + hz_out1 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS); + vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0); + tmp5 = __msa_dotp_u_h(vec0, filt_vt); - vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); - tmp5 = __msa_dotp_u_h(vec0, filt_vert); + hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz, FILTER_BITS); + vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1); + tmp6 = __msa_dotp_u_h(vec0, filt_vt); - horiz_vec = (v16u8)__msa_vshf_b(mask, src2, src2); - horiz_out0 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_out0, FILTER_BITS, 7); + hz_out1 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz, FILTER_BITS); + vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0); + tmp7 = __msa_dotp_u_h(vec0, filt_vt); - vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out0, (v16i8)horiz_out1); - tmp6 = __msa_dotp_u_h(vec0, filt_vert); + hz_out0 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS); + vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1); + tmp8 = __msa_dotp_u_h(vec0, filt_vt); - horiz_vec = (v16u8)__msa_vshf_b(mask, src3, src3); - horiz_out1 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out1 = SRARI_SATURATE_UNSIGNED_H(horiz_out1, FILTER_BITS, 7); - - vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); - tmp7 = __msa_dotp_u_h(vec0, filt_vert); - - horiz_vec = (v16u8)__msa_vshf_b(mask, src4, src4); - horiz_out0 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_out0, FILTER_BITS, 7); - - vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out0, (v16i8)horiz_out1); - tmp8 = __msa_dotp_u_h(vec0, filt_vert); - - tmp5 = SRARI_SATURATE_UNSIGNED_H(tmp5, FILTER_BITS, 7); - tmp6 = SRARI_SATURATE_UNSIGNED_H(tmp6, FILTER_BITS, 7); - tmp7 = SRARI_SATURATE_UNSIGNED_H(tmp7, FILTER_BITS, 7); - tmp8 = SRARI_SATURATE_UNSIGNED_H(tmp8, FILTER_BITS, 7); - - PCKEV_B_STORE_8_BYTES_4(tmp5, tmp6, tmp7, tmp8, dst, dst_stride); + SRARI_H4_UH(tmp5, tmp6, tmp7, tmp8, FILTER_BITS); + SAT_UH4_UH(tmp5, tmp6, tmp7, tmp8, 7); + PCKEV_B2_SB(tmp6, tmp5, tmp8, tmp7, out0, out1); + ST8x4_UB(out0, out1, dst, dst_stride); dst += (4 * dst_stride); } } @@ -645,108 +463,64 @@ static void common_hv_2ht_2vt_16w_msa(const uint8_t *src, int32_t src_stride, int32_t height) { uint32_t loop_cnt; v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask; - v16u8 filt_horiz, filt_vert, vec0, horiz_vec; - v8u16 horiz_vec0, horiz_vec1, tmp1, tmp2; - v8u16 horiz_out0, horiz_out1, horiz_out2, horiz_out3; + v16u8 filt_hz, filt_vt, vec0, vec1; + v8u16 tmp1, tmp2, hz_out0, hz_out1, hz_out2, hz_out3; v8i16 filt; - mask = LOAD_SB(&mc_filt_mask_arr[0]); + mask = LD_SB(&mc_filt_mask_arr[0]); /* rearranging filter */ - filt = LOAD_SH(filter_horiz); - filt_horiz = (v16u8)__msa_splati_h(filt, 0); + filt = LD_SH(filter_horiz); + filt_hz = (v16u8)__msa_splati_h(filt, 0); - filt = LOAD_SH(filter_vert); - filt_vert = (v16u8)__msa_splati_h(filt, 0); - - src0 = LOAD_SB(src); - src1 = LOAD_SB(src + 8); - - horiz_vec = (v16u8)__msa_vshf_b(mask, src0, src0); - horiz_vec0 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_vec0, FILTER_BITS, 7); - - horiz_vec = (v16u8)__msa_vshf_b(mask, src1, src1); - horiz_vec1 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out2 = SRARI_SATURATE_UNSIGNED_H(horiz_vec1, FILTER_BITS, 7); + filt = LD_SH(filter_vert); + filt_vt = (v16u8)__msa_splati_h(filt, 0); + LD_SB2(src, 8, src0, src1); src += src_stride; + hz_out0 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS); + hz_out2 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS); + for (loop_cnt = (height >> 2); loop_cnt--;) { - LOAD_4VECS_SB(src, src_stride, src0, src2, src4, src6); - LOAD_4VECS_SB(src + 8, src_stride, src1, src3, src5, src7); + LD_SB4(src, src_stride, src0, src2, src4, src6); + LD_SB4(src + 8, src_stride, src1, src3, src5, src7); src += (4 * src_stride); - horiz_vec = (v16u8)__msa_vshf_b(mask, src0, src0); - horiz_vec0 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out1 = SRARI_SATURATE_UNSIGNED_H(horiz_vec0, FILTER_BITS, 7); - - horiz_vec = (v16u8)__msa_vshf_b(mask, src1, src1); - horiz_vec1 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out3 = SRARI_SATURATE_UNSIGNED_H(horiz_vec1, FILTER_BITS, 7); - - vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); - tmp1 = __msa_dotp_u_h(vec0, filt_vert); - vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out3, (v16i8)horiz_out2); - tmp2 = __msa_dotp_u_h(vec0, filt_vert); - tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); - tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); - - PCKEV_B_STORE_VEC(tmp2, tmp1, dst); + hz_out1 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS); + hz_out3 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS); + ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1); + DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp1, tmp2); + SRARI_H2_UH(tmp1, tmp2, FILTER_BITS); + SAT_UH2_UH(tmp1, tmp2, 7); + PCKEV_ST_SB(tmp1, tmp2, dst); dst += dst_stride; - horiz_vec = (v16u8)__msa_vshf_b(mask, src2, src2); - horiz_vec0 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_vec0, FILTER_BITS, 7); - - horiz_vec = (v16u8)__msa_vshf_b(mask, src3, src3); - horiz_vec1 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out2 = SRARI_SATURATE_UNSIGNED_H(horiz_vec1, FILTER_BITS, 7); - - vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out0, (v16i8)horiz_out1); - tmp1 = __msa_dotp_u_h(vec0, filt_vert); - vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out2, (v16i8)horiz_out3); - tmp2 = __msa_dotp_u_h(vec0, filt_vert); - tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); - tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); - - PCKEV_B_STORE_VEC(tmp2, tmp1, dst); + hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz, FILTER_BITS); + hz_out2 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz, FILTER_BITS); + ILVEV_B2_UB(hz_out1, hz_out0, hz_out3, hz_out2, vec0, vec1); + DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp1, tmp2); + SRARI_H2_UH(tmp1, tmp2, FILTER_BITS); + SAT_UH2_UH(tmp1, tmp2, 7); + PCKEV_ST_SB(tmp1, tmp2, dst); dst += dst_stride; - horiz_vec = (v16u8)__msa_vshf_b(mask, src4, src4); - horiz_vec0 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out1 = SRARI_SATURATE_UNSIGNED_H(horiz_vec0, FILTER_BITS, 7); - - horiz_vec = (v16u8)__msa_vshf_b(mask, src5, src5); - horiz_vec1 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out3 = SRARI_SATURATE_UNSIGNED_H(horiz_vec1, FILTER_BITS, 7); - - vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); - tmp1 = __msa_dotp_u_h(vec0, filt_vert); - vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out3, (v16i8)horiz_out2); - tmp2 = __msa_dotp_u_h(vec0, filt_vert); - tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); - tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); - - PCKEV_B_STORE_VEC(tmp2, tmp1, dst); + hz_out1 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS); + hz_out3 = HORIZ_2TAP_FILT_UH(src5, src5, mask, filt_hz, FILTER_BITS); + ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1); + DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp1, tmp2); + SRARI_H2_UH(tmp1, tmp2, FILTER_BITS); + SAT_UH2_UH(tmp1, tmp2, 7); + PCKEV_ST_SB(tmp1, tmp2, dst); dst += dst_stride; - horiz_vec = (v16u8)__msa_vshf_b(mask, src6, src6); - horiz_vec0 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_vec0, FILTER_BITS, 7); - - horiz_vec = (v16u8)__msa_vshf_b(mask, src7, src7); - horiz_vec1 = __msa_dotp_u_h(horiz_vec, filt_horiz); - horiz_out2 = SRARI_SATURATE_UNSIGNED_H(horiz_vec1, FILTER_BITS, 7); - - vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out0, (v16i8)horiz_out1); - tmp1 = __msa_dotp_u_h(vec0, filt_vert); - vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out2, (v16i8)horiz_out3); - tmp2 = __msa_dotp_u_h(vec0, filt_vert); - tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); - tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); - - PCKEV_B_STORE_VEC(tmp2, tmp1, dst); + hz_out0 = HORIZ_2TAP_FILT_UH(src6, src6, mask, filt_hz, FILTER_BITS); + hz_out2 = HORIZ_2TAP_FILT_UH(src7, src7, mask, filt_hz, FILTER_BITS); + ILVEV_B2_UB(hz_out1, hz_out0, hz_out3, hz_out2, vec0, vec1); + DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp1, tmp2); + SRARI_H2_UH(tmp1, tmp2, FILTER_BITS); + SAT_UH2_UH(tmp1, tmp2, 7); + PCKEV_ST_SB(tmp1, tmp2, dst); dst += dst_stride; } } diff --git a/vp9/common/mips/msa/vp9_convolve8_vert_msa.c b/vp9/common/mips/msa/vp9_convolve8_vert_msa.c index 6b71ec1c0..e9ec2507a 100644 --- a/vp9/common/mips/msa/vp9_convolve8_vert_msa.c +++ b/vp9/common/mips/msa/vp9_convolve8_vert_msa.c @@ -16,58 +16,48 @@ static void common_vt_8t_4w_msa(const uint8_t *src, int32_t src_stride, int8_t *filter, int32_t height) { uint32_t loop_cnt; v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10; - 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, src10998; + v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r; + v16i8 src65_r, src87_r, src109_r, src2110, src4332, src6554, src8776; + v16i8 src10998, filt0, filt1, filt2, filt3; + v16u8 out; v8i16 filt, out10, out32; - v16i8 filt0, filt1, filt2, filt3; src -= (3 * src_stride); - filt = LOAD_SH(filter); - filt0 = (v16i8)__msa_splati_h(filt, 0); - filt1 = (v16i8)__msa_splati_h(filt, 1); - filt2 = (v16i8)__msa_splati_h(filt, 2); - filt3 = (v16i8)__msa_splati_h(filt, 3); + filt = LD_SH(filter); + SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3); - LOAD_7VECS_SB(src, src_stride, src0, src1, src2, src3, src4, src5, src6); + LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6); src += (7 * src_stride); - ILVR_B_6VECS_SB(src0, src2, src4, src1, src3, src5, - src1, src3, src5, src2, src4, src6, - src10_r, src32_r, src54_r, src21_r, src43_r, src65_r); - - ILVR_D_3VECS_SB(src2110, src21_r, src10_r, src4332, src43_r, src32_r, - src6554, src65_r, src54_r); - - XORI_B_3VECS_SB(src2110, src4332, src6554, src2110, src4332, src6554, 128); + 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 >> 2); loop_cnt--;) { - LOAD_4VECS_SB(src, src_stride, src7, src8, src9, src10); + LD_SB4(src, src_stride, src7, src8, src9, src10); src += (4 * src_stride); - ILVR_B_4VECS_SB(src6, src7, src8, src9, src7, src8, src9, src10, - src76_r, src87_r, src98_r, src109_r); - - ILVR_D_2VECS_SB(src8776, src87_r, src76_r, src10998, src109_r, src98_r); - - XORI_B_2VECS_SB(src8776, src10998, src8776, src10998, 128); - - out10 = FILT_8TAP_DPADD_S_H(src2110, src4332, src6554, src8776, - filt0, filt1, filt2, filt3); - out32 = FILT_8TAP_DPADD_S_H(src4332, src6554, src8776, src10998, - filt0, filt1, filt2, filt3); - - out10 = SRARI_SATURATE_SIGNED_H(out10, FILTER_BITS, 7); - out32 = SRARI_SATURATE_SIGNED_H(out32, FILTER_BITS, 7); - - PCKEV_2B_XORI128_STORE_4_BYTES_4(out10, out32, dst, dst_stride); + ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r, + src87_r, src98_r, src109_r); + ILVR_D2_SB(src87_r, src76_r, src109_r, src98_r, src8776, src10998); + XORI_B2_128_SB(src8776, src10998); + out10 = FILT_8TAP_DPADD_S_H(src2110, src4332, src6554, src8776, filt0, + filt1, filt2, filt3); + out32 = FILT_8TAP_DPADD_S_H(src4332, src6554, src8776, src10998, filt0, + filt1, filt2, filt3); + SRARI_H2_SH(out10, out32, FILTER_BITS); + SAT_SH2_SH(out10, out32, 7); + out = PCKEV_XORI128_UB(out10, out32); + ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride); dst += (4 * dst_stride); src2110 = src6554; src4332 = src8776; src6554 = src10998; - src6 = src10; } } @@ -77,54 +67,43 @@ static void common_vt_8t_8w_msa(const uint8_t *src, int32_t src_stride, int8_t *filter, int32_t height) { uint32_t loop_cnt; v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10; - v16i8 src10_r, src32_r, src54_r, src76_r, src98_r; - v16i8 src21_r, src43_r, src65_r, src87_r, src109_r; - v16i8 filt0, filt1, filt2, filt3; + v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r; + v16i8 src65_r, src87_r, src109_r, filt0, filt1, filt2, filt3; + v16u8 tmp0, tmp1; v8i16 filt, out0_r, out1_r, out2_r, out3_r; src -= (3 * src_stride); - filt = LOAD_SH(filter); - filt0 = (v16i8)__msa_splati_h(filt, 0); - filt1 = (v16i8)__msa_splati_h(filt, 1); - filt2 = (v16i8)__msa_splati_h(filt, 2); - filt3 = (v16i8)__msa_splati_h(filt, 3); + filt = LD_SH(filter); + SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3); - LOAD_7VECS_SB(src, src_stride, src0, src1, src2, src3, src4, src5, src6); + LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6); + XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6); src += (7 * src_stride); - - XORI_B_7VECS_SB(src0, src1, src2, src3, src4, src5, src6, - src0, src1, src2, src3, src4, src5, src6, 128); - - ILVR_B_6VECS_SB(src0, src2, src4, src1, src3, src5, - src1, src3, src5, src2, src4, src6, - src10_r, src32_r, src54_r, src21_r, src43_r, src65_r); + 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--;) { - LOAD_4VECS_SB(src, src_stride, src7, src8, src9, src10); + LD_SB4(src, src_stride, src7, src8, src9, src10); + XORI_B4_128_SB(src7, src8, src9, src10); src += (4 * src_stride); - XORI_B_4VECS_SB(src7, src8, src9, src10, src7, src8, src9, src10, 128); - - ILVR_B_4VECS_SB(src6, src7, src8, src9, src7, src8, src9, src10, - src76_r, src87_r, src98_r, src109_r); - - out0_r = FILT_8TAP_DPADD_S_H(src10_r, src32_r, src54_r, src76_r, - filt0, filt1, filt2, filt3); - out1_r = FILT_8TAP_DPADD_S_H(src21_r, src43_r, src65_r, src87_r, - filt0, filt1, filt2, filt3); - out2_r = FILT_8TAP_DPADD_S_H(src32_r, src54_r, src76_r, src98_r, - filt0, filt1, filt2, filt3); - out3_r = FILT_8TAP_DPADD_S_H(src43_r, src65_r, src87_r, src109_r, - filt0, filt1, filt2, filt3); - - out0_r = SRARI_SATURATE_SIGNED_H(out0_r, FILTER_BITS, 7); - out1_r = SRARI_SATURATE_SIGNED_H(out1_r, FILTER_BITS, 7); - out2_r = SRARI_SATURATE_SIGNED_H(out2_r, FILTER_BITS, 7); - out3_r = SRARI_SATURATE_SIGNED_H(out3_r, FILTER_BITS, 7); - - PCKEV_B_4_XORI128_STORE_8_BYTES_4(out0_r, out1_r, out2_r, out3_r, - dst, dst_stride); + ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r, + src87_r, src98_r, src109_r); + out0_r = FILT_8TAP_DPADD_S_H(src10_r, src32_r, src54_r, src76_r, filt0, + filt1, filt2, filt3); + out1_r = FILT_8TAP_DPADD_S_H(src21_r, src43_r, src65_r, src87_r, filt0, + filt1, filt2, filt3); + out2_r = FILT_8TAP_DPADD_S_H(src32_r, src54_r, src76_r, src98_r, filt0, + filt1, filt2, filt3); + out3_r = FILT_8TAP_DPADD_S_H(src43_r, src65_r, src87_r, src109_r, filt0, + filt1, filt2, filt3); + SRARI_H4_SH(out0_r, out1_r, out2_r, out3_r, FILTER_BITS); + SAT_SH4_SH(out0_r, out1_r, out2_r, out3_r, 7); + tmp0 = PCKEV_XORI128_UB(out0_r, out1_r); + tmp1 = PCKEV_XORI128_UB(out2_r, out3_r); + ST8x4_UB(tmp0, tmp1, dst, dst_stride); dst += (4 * dst_stride); src10_r = src54_r; @@ -133,7 +112,84 @@ static void common_vt_8t_8w_msa(const uint8_t *src, int32_t src_stride, src21_r = src65_r; src43_r = src87_r; src65_r = src109_r; + src6 = src10; + } +} +static void common_vt_8t_16w_msa(const uint8_t *src, int32_t src_stride, + uint8_t *dst, int32_t dst_stride, + int8_t *filter, int32_t height) { + uint32_t loop_cnt; + v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10; + v16i8 filt0, filt1, filt2, filt3; + v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r; + v16i8 src65_r, src87_r, src109_r, src10_l, src32_l, src54_l, src76_l; + v16i8 src98_l, src21_l, src43_l, src65_l, src87_l, src109_l; + v16u8 tmp0, tmp1, tmp2, tmp3; + v8i16 filt, out0_r, out1_r, out2_r, out3_r, out0_l, out1_l, out2_l, out3_l; + + src -= (3 * src_stride); + + filt = LD_SH(filter); + SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3); + + LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6); + XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6); + src += (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); + 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 >> 2); loop_cnt--;) { + LD_SB4(src, src_stride, src7, src8, src9, src10); + XORI_B4_128_SB(src7, src8, src9, src10); + src += (4 * src_stride); + + 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); + out0_r = FILT_8TAP_DPADD_S_H(src10_r, src32_r, src54_r, src76_r, filt0, + filt1, filt2, filt3); + out1_r = FILT_8TAP_DPADD_S_H(src21_r, src43_r, src65_r, src87_r, filt0, + filt1, filt2, filt3); + out2_r = FILT_8TAP_DPADD_S_H(src32_r, src54_r, src76_r, src98_r, filt0, + filt1, filt2, filt3); + out3_r = FILT_8TAP_DPADD_S_H(src43_r, src65_r, src87_r, src109_r, filt0, + filt1, filt2, filt3); + out0_l = FILT_8TAP_DPADD_S_H(src10_l, src32_l, src54_l, src76_l, filt0, + filt1, filt2, filt3); + out1_l = FILT_8TAP_DPADD_S_H(src21_l, src43_l, src65_l, src87_l, filt0, + filt1, filt2, filt3); + out2_l = FILT_8TAP_DPADD_S_H(src32_l, src54_l, src76_l, src98_l, filt0, + filt1, filt2, filt3); + out3_l = FILT_8TAP_DPADD_S_H(src43_l, src65_l, src87_l, src109_l, filt0, + filt1, filt2, filt3); + SRARI_H4_SH(out0_r, out1_r, out2_r, out3_r, FILTER_BITS); + SRARI_H4_SH(out0_l, out1_l, out2_l, out3_l, FILTER_BITS); + SAT_SH4_SH(out0_r, out1_r, out2_r, out3_r, 7); + SAT_SH4_SH(out0_l, out1_l, out2_l, out3_l, 7); + PCKEV_B4_UB(out0_l, out0_r, out1_l, out1_r, out2_l, out2_r, out3_l, out3_r, + tmp0, tmp1, tmp2, tmp3); + XORI_B4_128_UB(tmp0, tmp1, tmp2, tmp3); + ST_UB4(tmp0, tmp1, tmp2, tmp3, 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; + src10_l = src54_l; + src32_l = src76_l; + src54_l = src98_l; + src21_l = src65_l; + src43_l = src87_l; + src65_l = src109_l; src6 = src10; } } @@ -147,89 +203,63 @@ static void common_vt_8t_16w_mult_msa(const uint8_t *src, int32_t src_stride, uint32_t loop_cnt, cnt; v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10; v16i8 filt0, filt1, filt2, filt3; - v16i8 src10_r, src32_r, src54_r, src76_r, src98_r; - v16i8 src21_r, src43_r, src65_r, src87_r, src109_r; - v16i8 src10_l, src32_l, src54_l, src76_l, src98_l; - v16i8 src21_l, src43_l, src65_l, src87_l, src109_l; - v8i16 out0_r, out1_r, out2_r, out3_r, out0_l, out1_l, out2_l, out3_l; - v8i16 filt; + v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r; + v16i8 src65_r, src87_r, src109_r, src10_l, src32_l, src54_l, src76_l; + v16i8 src98_l, src21_l, src43_l, src65_l, src87_l, src109_l; v16u8 tmp0, tmp1, tmp2, tmp3; + v8i16 filt, out0_r, out1_r, out2_r, out3_r, out0_l, out1_l, out2_l, out3_l; src -= (3 * src_stride); - filt = LOAD_SH(filter); - filt0 = (v16i8)__msa_splati_h(filt, 0); - filt1 = (v16i8)__msa_splati_h(filt, 1); - filt2 = (v16i8)__msa_splati_h(filt, 2); - filt3 = (v16i8)__msa_splati_h(filt, 3); + filt = LD_SH(filter); + SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3); for (cnt = (width >> 4); cnt--;) { src_tmp = src; dst_tmp = dst; - LOAD_7VECS_SB(src_tmp, src_stride, - src0, src1, src2, src3, src4, src5, src6); + LD_SB7(src_tmp, src_stride, src0, src1, src2, src3, src4, src5, src6); + XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6); src_tmp += (7 * src_stride); - - XORI_B_7VECS_SB(src0, src1, src2, src3, src4, src5, src6, - src0, src1, src2, src3, src4, src5, src6, 128); - - ILVR_B_6VECS_SB(src0, src2, src4, src1, src3, src5, - src1, src3, src5, src2, src4, src6, - src10_r, src32_r, src54_r, src21_r, src43_r, src65_r); - - ILVL_B_6VECS_SB(src0, src2, src4, src1, src3, src5, - src1, src3, src5, src2, src4, src6, - src10_l, src32_l, src54_l, src21_l, src43_l, src65_l); + 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 >> 2); loop_cnt--;) { - LOAD_4VECS_SB(src_tmp, src_stride, src7, src8, src9, src10); + LD_SB4(src_tmp, src_stride, src7, src8, src9, src10); + XORI_B4_128_SB(src7, src8, src9, src10); src_tmp += (4 * src_stride); - - XORI_B_4VECS_SB(src7, src8, src9, src10, src7, src8, src9, src10, 128); - - ILVR_B_4VECS_SB(src6, src7, src8, src9, src7, src8, src9, src10, - src76_r, src87_r, src98_r, src109_r); - - ILVL_B_4VECS_SB(src6, src7, src8, src9, src7, src8, src9, src10, - src76_l, src87_l, src98_l, src109_l); - - out0_r = FILT_8TAP_DPADD_S_H(src10_r, src32_r, src54_r, src76_r, - filt0, filt1, filt2, filt3); - out1_r = FILT_8TAP_DPADD_S_H(src21_r, src43_r, src65_r, src87_r, - filt0, filt1, filt2, filt3); - out2_r = FILT_8TAP_DPADD_S_H(src32_r, src54_r, src76_r, src98_r, - filt0, filt1, filt2, filt3); - out3_r = FILT_8TAP_DPADD_S_H(src43_r, src65_r, src87_r, src109_r, - filt0, filt1, filt2, filt3); - - out0_l = FILT_8TAP_DPADD_S_H(src10_l, src32_l, src54_l, src76_l, - filt0, filt1, filt2, filt3); - out1_l = FILT_8TAP_DPADD_S_H(src21_l, src43_l, src65_l, src87_l, - filt0, filt1, filt2, filt3); - out2_l = FILT_8TAP_DPADD_S_H(src32_l, src54_l, src76_l, src98_l, - filt0, filt1, filt2, filt3); - out3_l = FILT_8TAP_DPADD_S_H(src43_l, src65_l, src87_l, src109_l, - filt0, filt1, filt2, filt3); - - out0_r = SRARI_SATURATE_SIGNED_H(out0_r, FILTER_BITS, 7); - out1_r = SRARI_SATURATE_SIGNED_H(out1_r, FILTER_BITS, 7); - out2_r = SRARI_SATURATE_SIGNED_H(out2_r, FILTER_BITS, 7); - out3_r = SRARI_SATURATE_SIGNED_H(out3_r, FILTER_BITS, 7); - out0_l = SRARI_SATURATE_SIGNED_H(out0_l, FILTER_BITS, 7); - out1_l = SRARI_SATURATE_SIGNED_H(out1_l, FILTER_BITS, 7); - out2_l = SRARI_SATURATE_SIGNED_H(out2_l, FILTER_BITS, 7); - out3_l = SRARI_SATURATE_SIGNED_H(out3_l, FILTER_BITS, 7); - - out0_r = (v8i16)__msa_pckev_b((v16i8)out0_l, (v16i8)out0_r); - out1_r = (v8i16)__msa_pckev_b((v16i8)out1_l, (v16i8)out1_r); - out2_r = (v8i16)__msa_pckev_b((v16i8)out2_l, (v16i8)out2_r); - out3_r = (v8i16)__msa_pckev_b((v16i8)out3_l, (v16i8)out3_r); - - XORI_B_4VECS_UB(out0_r, out1_r, out2_r, out3_r, - tmp0, tmp1, tmp2, tmp3, 128); - - STORE_4VECS_UB(dst_tmp, dst_stride, tmp0, tmp1, tmp2, tmp3); + 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); + out0_r = FILT_8TAP_DPADD_S_H(src10_r, src32_r, src54_r, src76_r, filt0, + filt1, filt2, filt3); + out1_r = FILT_8TAP_DPADD_S_H(src21_r, src43_r, src65_r, src87_r, filt0, + filt1, filt2, filt3); + out2_r = FILT_8TAP_DPADD_S_H(src32_r, src54_r, src76_r, src98_r, filt0, + filt1, filt2, filt3); + out3_r = FILT_8TAP_DPADD_S_H(src43_r, src65_r, src87_r, src109_r, filt0, + filt1, filt2, filt3); + out0_l = FILT_8TAP_DPADD_S_H(src10_l, src32_l, src54_l, src76_l, filt0, + filt1, filt2, filt3); + out1_l = FILT_8TAP_DPADD_S_H(src21_l, src43_l, src65_l, src87_l, filt0, + filt1, filt2, filt3); + out2_l = FILT_8TAP_DPADD_S_H(src32_l, src54_l, src76_l, src98_l, filt0, + filt1, filt2, filt3); + out3_l = FILT_8TAP_DPADD_S_H(src43_l, src65_l, src87_l, src109_l, filt0, + filt1, filt2, filt3); + SRARI_H4_SH(out0_r, out1_r, out2_r, out3_r, FILTER_BITS); + SRARI_H4_SH(out0_l, out1_l, out2_l, out3_l, FILTER_BITS); + SAT_SH4_SH(out0_r, out1_r, out2_r, out3_r, 7); + SAT_SH4_SH(out0_l, out1_l, out2_l, out3_l, 7); + PCKEV_B4_UB(out0_l, out0_r, out1_l, out1_r, out2_l, out2_r, out3_l, + out3_r, tmp0, tmp1, tmp2, tmp3); + XORI_B4_128_UB(tmp0, tmp1, tmp2, tmp3); + ST_UB4(tmp0, tmp1, tmp2, tmp3, dst_tmp, dst_stride); dst_tmp += (4 * dst_stride); src10_r = src54_r; @@ -238,14 +268,12 @@ static void common_vt_8t_16w_mult_msa(const uint8_t *src, int32_t src_stride, src21_r = src65_r; src43_r = src87_r; src65_r = src109_r; - src10_l = src54_l; src32_l = src76_l; src54_l = src98_l; src21_l = src65_l; src43_l = src87_l; src65_l = src109_l; - src6 = src10; } @@ -254,134 +282,77 @@ static void common_vt_8t_16w_mult_msa(const uint8_t *src, int32_t src_stride, } } -static void common_vt_8t_16w_msa(const uint8_t *src, int32_t src_stride, - uint8_t *dst, int32_t dst_stride, - int8_t *filter, int32_t height) { - common_vt_8t_16w_mult_msa(src, src_stride, dst, dst_stride, - filter, height, 16); -} - static void common_vt_8t_32w_msa(const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride, int8_t *filter, int32_t height) { - common_vt_8t_16w_mult_msa(src, src_stride, dst, dst_stride, - filter, height, 32); + common_vt_8t_16w_mult_msa(src, src_stride, dst, dst_stride, filter, height, + 32); } static void common_vt_8t_64w_msa(const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride, int8_t *filter, int32_t height) { - common_vt_8t_16w_mult_msa(src, src_stride, dst, dst_stride, - filter, height, 64); + common_vt_8t_16w_mult_msa(src, src_stride, dst, dst_stride, filter, height, + 64); } static void common_vt_2t_4x4_msa(const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride, int8_t *filter) { - uint32_t out0, out1, out2, out3; v16i8 src0, src1, src2, src3, src4; v16i8 src10_r, src32_r, src21_r, src43_r, src2110, src4332; - v16i8 filt0; - v8u16 filt; + v16u8 filt0; + v8i16 filt; + v8u16 tmp0, tmp1; - filt = LOAD_UH(filter); - filt0 = (v16i8)__msa_splati_h((v8i16)filt, 0); + filt = LD_SH(filter); + filt0 = (v16u8)__msa_splati_h(filt, 0); - LOAD_5VECS_SB(src, src_stride, src0, src1, src2, src3, src4); + LD_SB5(src, src_stride, src0, src1, src2, src3, src4); src += (5 * src_stride); - ILVR_B_4VECS_SB(src0, src1, src2, src3, src1, src2, src3, src4, - src10_r, src21_r, src32_r, src43_r); - - ILVR_D_2VECS_SB(src2110, src21_r, src10_r, src4332, src43_r, src32_r); - - src2110 = (v16i8)__msa_dotp_u_h((v16u8)src2110, (v16u8)filt0); - src4332 = (v16i8)__msa_dotp_u_h((v16u8)src4332, (v16u8)filt0); - - src2110 = (v16i8)SRARI_SATURATE_UNSIGNED_H(src2110, FILTER_BITS, 7); - src4332 = (v16i8)SRARI_SATURATE_UNSIGNED_H(src4332, FILTER_BITS, 7); - - src2110 = (v16i8)__msa_pckev_b((v16i8)src4332, (v16i8)src2110); - - out0 = __msa_copy_u_w((v4i32)src2110, 0); - out1 = __msa_copy_u_w((v4i32)src2110, 1); - out2 = __msa_copy_u_w((v4i32)src2110, 2); - out3 = __msa_copy_u_w((v4i32)src2110, 3); - - STORE_WORD(dst, out0); - dst += dst_stride; - STORE_WORD(dst, out1); - dst += dst_stride; - STORE_WORD(dst, out2); - dst += dst_stride; - STORE_WORD(dst, out3); + ILVR_B4_SB(src1, src0, src2, src1, src3, src2, src4, src3, src10_r, src21_r, + src32_r, src43_r); + ILVR_D2_SB(src21_r, src10_r, src43_r, src32_r, src2110, src4332); + DOTP_UB2_UH(src2110, src4332, filt0, filt0, tmp0, tmp1); + SRARI_H2_UH(tmp0, tmp1, FILTER_BITS); + SAT_UH2_UH(tmp0, tmp1, 7); + src2110 = __msa_pckev_b((v16i8)tmp1, (v16i8)tmp0); + ST4x4_UB(src2110, src2110, 0, 1, 2, 3, dst, dst_stride); } static void common_vt_2t_4x8_msa(const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride, int8_t *filter) { - uint32_t out0, out1, out2, out3, out4, out5, out6, out7; v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8; v16i8 src10_r, src32_r, src54_r, src76_r, src21_r, src43_r; v16i8 src65_r, src87_r, src2110, src4332, src6554, src8776; - v16i8 filt0; - v8u16 filt; + v8u16 tmp0, tmp1, tmp2, tmp3; + v16u8 filt0; + v8i16 filt; - filt = LOAD_UH(filter); - filt0 = (v16i8)__msa_splati_h((v8i16)filt, 0); + filt = LD_SH(filter); + filt0 = (v16u8)__msa_splati_h(filt, 0); - LOAD_8VECS_SB(src, src_stride, - src0, src1, src2, src3, src4, src5, src6, src7); + LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7); src += (8 * src_stride); - src8 = LOAD_SB(src); + src8 = LD_SB(src); src += src_stride; - ILVR_B_8VECS_SB(src0, src1, src2, src3, src4, src5, src6, src7, - src1, src2, src3, src4, src5, src6, src7, src8, - src10_r, src21_r, src32_r, src43_r, - src54_r, src65_r, src76_r, src87_r); - - ILVR_D_4VECS_SB(src2110, src21_r, src10_r, src4332, src43_r, src32_r, - src6554, src65_r, src54_r, src8776, src87_r, src76_r); - - src2110 = (v16i8)__msa_dotp_u_h((v16u8)src2110, (v16u8)filt0); - src4332 = (v16i8)__msa_dotp_u_h((v16u8)src4332, (v16u8)filt0); - src6554 = (v16i8)__msa_dotp_u_h((v16u8)src6554, (v16u8)filt0); - src8776 = (v16i8)__msa_dotp_u_h((v16u8)src8776, (v16u8)filt0); - - src2110 = (v16i8)SRARI_SATURATE_UNSIGNED_H(src2110, FILTER_BITS, 7); - src4332 = (v16i8)SRARI_SATURATE_UNSIGNED_H(src4332, FILTER_BITS, 7); - src6554 = (v16i8)SRARI_SATURATE_UNSIGNED_H(src6554, FILTER_BITS, 7); - src8776 = (v16i8)SRARI_SATURATE_UNSIGNED_H(src8776, FILTER_BITS, 7); - - src2110 = (v16i8)__msa_pckev_b((v16i8)src4332, (v16i8)src2110); - src4332 = (v16i8)__msa_pckev_b((v16i8)src8776, (v16i8)src6554); - - out0 = __msa_copy_u_w((v4i32)src2110, 0); - out1 = __msa_copy_u_w((v4i32)src2110, 1); - out2 = __msa_copy_u_w((v4i32)src2110, 2); - out3 = __msa_copy_u_w((v4i32)src2110, 3); - out4 = __msa_copy_u_w((v4i32)src4332, 0); - out5 = __msa_copy_u_w((v4i32)src4332, 1); - out6 = __msa_copy_u_w((v4i32)src4332, 2); - out7 = __msa_copy_u_w((v4i32)src4332, 3); - - STORE_WORD(dst, out0); - dst += dst_stride; - STORE_WORD(dst, out1); - dst += dst_stride; - STORE_WORD(dst, out2); - dst += dst_stride; - STORE_WORD(dst, out3); - dst += dst_stride; - STORE_WORD(dst, out4); - dst += dst_stride; - STORE_WORD(dst, out5); - dst += dst_stride; - STORE_WORD(dst, out6); - dst += dst_stride; - STORE_WORD(dst, out7); + ILVR_B4_SB(src1, src0, src2, src1, src3, src2, src4, src3, src10_r, src21_r, + src32_r, src43_r); + ILVR_B4_SB(src5, src4, src6, src5, src7, src6, src8, src7, src54_r, src65_r, + src76_r, src87_r); + ILVR_D4_SB(src21_r, src10_r, src43_r, src32_r, src65_r, src54_r, + src87_r, src76_r, src2110, src4332, src6554, src8776); + DOTP_UB4_UH(src2110, src4332, src6554, src8776, filt0, filt0, filt0, filt0, + tmp0, tmp1, tmp2, tmp3); + SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS); + SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7); + PCKEV_B2_SB(tmp1, tmp0, tmp3, tmp2, src2110, src4332); + ST4x4_UB(src2110, src2110, 0, 1, 2, 3, dst, dst_stride); + ST4x4_UB(src4332, src4332, 0, 1, 2, 3, dst + 4 * dst_stride, dst_stride); } static void common_vt_2t_4w_msa(const uint8_t *src, int32_t src_stride, @@ -397,32 +368,24 @@ static void common_vt_2t_4w_msa(const uint8_t *src, int32_t src_stride, static void common_vt_2t_8x4_msa(const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride, int8_t *filter) { - v16u8 src0, src1, src2, src3, src4; - v16u8 vec0, vec1, vec2, vec3, filt0; + v16u8 src0, src1, src2, src3, src4, vec0, vec1, vec2, vec3, filt0; + v16i8 out0, out1; v8u16 tmp0, tmp1, tmp2, tmp3; - v8u16 filt; + v8i16 filt; /* rearranging filter_y */ - filt = LOAD_UH(filter); - filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0); + filt = LD_SH(filter); + filt0 = (v16u8)__msa_splati_h(filt, 0); - LOAD_5VECS_UB(src, src_stride, src0, src1, src2, src3, src4); - - ILVR_B_2VECS_UB(src0, src1, src1, src2, vec0, vec1); - ILVR_B_2VECS_UB(src2, src3, src3, src4, vec2, vec3); - - /* filter calc */ - tmp0 = __msa_dotp_u_h(vec0, filt0); - tmp1 = __msa_dotp_u_h(vec1, filt0); - tmp2 = __msa_dotp_u_h(vec2, filt0); - tmp3 = __msa_dotp_u_h(vec3, filt0); - - tmp0 = SRARI_SATURATE_UNSIGNED_H(tmp0, FILTER_BITS, 7); - tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); - tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); - tmp3 = SRARI_SATURATE_UNSIGNED_H(tmp3, FILTER_BITS, 7); - - PCKEV_B_STORE_8_BYTES_4(tmp0, tmp1, tmp2, tmp3, dst, dst_stride); + LD_UB5(src, src_stride, src0, src1, src2, src3, src4); + ILVR_B2_UB(src1, src0, src2, src1, vec0, vec1); + ILVR_B2_UB(src3, src2, src4, src3, vec2, vec3); + DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, tmp0, tmp1, + tmp2, tmp3); + SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS); + SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7); + PCKEV_B2_SB(tmp1, tmp0, tmp3, tmp2, out0, out1); + ST8x4_UB(out0, out1, dst, dst_stride); } static void common_vt_2t_8x8mult_msa(const uint8_t *src, int32_t src_stride, @@ -431,51 +394,39 @@ static void common_vt_2t_8x8mult_msa(const uint8_t *src, int32_t src_stride, uint32_t loop_cnt; v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8; v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0; + v16i8 out0, out1; v8u16 tmp0, tmp1, tmp2, tmp3; - v8u16 filt; + v8i16 filt; /* rearranging filter_y */ - filt = LOAD_UH(filter); - filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0); + filt = LD_SH(filter); + filt0 = (v16u8)__msa_splati_h(filt, 0); - src0 = LOAD_UB(src); + src0 = LD_UB(src); src += src_stride; for (loop_cnt = (height >> 3); loop_cnt--;) { - LOAD_8VECS_UB(src, src_stride, - src1, src2, src3, src4, src5, src6, src7, src8); + LD_UB8(src, src_stride, src1, src2, src3, src4, src5, src6, src7, src8); src += (8 * src_stride); - ILVR_B_4VECS_UB(src0, src1, src2, src3, src1, src2, src3, src4, - vec0, vec1, vec2, vec3); - - ILVR_B_4VECS_UB(src4, src5, src6, src7, src5, src6, src7, src8, - vec4, vec5, vec6, vec7); - - tmp0 = __msa_dotp_u_h(vec0, filt0); - tmp1 = __msa_dotp_u_h(vec1, filt0); - tmp2 = __msa_dotp_u_h(vec2, filt0); - tmp3 = __msa_dotp_u_h(vec3, filt0); - - tmp0 = SRARI_SATURATE_UNSIGNED_H(tmp0, FILTER_BITS, 7); - tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); - tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); - tmp3 = SRARI_SATURATE_UNSIGNED_H(tmp3, FILTER_BITS, 7); - - PCKEV_B_STORE_8_BYTES_4(tmp0, tmp1, tmp2, tmp3, dst, dst_stride); + ILVR_B4_UB(src1, src0, src2, src1, src3, src2, src4, src3, vec0, vec1, + vec2, vec3); + ILVR_B4_UB(src5, src4, src6, src5, src7, src6, src8, src7, vec4, vec5, + vec6, vec7); + DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, tmp0, tmp1, + tmp2, tmp3); + SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS); + SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7); + PCKEV_B2_SB(tmp1, tmp0, tmp3, tmp2, out0, out1); + ST8x4_UB(out0, out1, dst, dst_stride); dst += (4 * dst_stride); - tmp0 = __msa_dotp_u_h(vec4, filt0); - tmp1 = __msa_dotp_u_h(vec5, filt0); - tmp2 = __msa_dotp_u_h(vec6, filt0); - tmp3 = __msa_dotp_u_h(vec7, filt0); - - tmp0 = SRARI_SATURATE_UNSIGNED_H(tmp0, FILTER_BITS, 7); - tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); - tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); - tmp3 = SRARI_SATURATE_UNSIGNED_H(tmp3, FILTER_BITS, 7); - - PCKEV_B_STORE_8_BYTES_4(tmp0, tmp1, tmp2, tmp3, dst, dst_stride); + DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, tmp0, tmp1, + tmp2, tmp3); + SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS); + SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7); + PCKEV_B2_SB(tmp1, tmp0, tmp3, tmp2, out0, out1); + ST8x4_UB(out0, out1, dst, dst_stride); dst += (4 * dst_stride); src0 = src8; @@ -499,57 +450,45 @@ static void common_vt_2t_16w_msa(const uint8_t *src, int32_t src_stride, v16u8 src0, src1, src2, src3, src4; v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0; v8u16 tmp0, tmp1, tmp2, tmp3; - v8u16 filt; + v8i16 filt; /* rearranging filter_y */ - filt = LOAD_UH(filter); - filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0); + filt = LD_SH(filter); + filt0 = (v16u8)__msa_splati_h(filt, 0); - src0 = LOAD_UB(src); + src0 = LD_UB(src); src += src_stride; for (loop_cnt = (height >> 2); loop_cnt--;) { - LOAD_4VECS_UB(src, src_stride, src1, src2, src3, src4); + LD_UB4(src, src_stride, src1, src2, src3, src4); src += (4 * src_stride); - ILV_B_LRLR_UB(src0, src1, src1, src2, vec1, vec0, vec3, vec2); - - tmp0 = __msa_dotp_u_h(vec0, filt0); - tmp1 = __msa_dotp_u_h(vec1, filt0); - - tmp0 = SRARI_SATURATE_UNSIGNED_H(tmp0, FILTER_BITS, 7); - tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); - - PCKEV_B_STORE_VEC(tmp1, tmp0, dst); + ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2); + ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3); + DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1); + SRARI_H2_UH(tmp0, tmp1, FILTER_BITS); + SAT_UH2_UH(tmp0, tmp1, 7); + PCKEV_ST_SB(tmp0, tmp1, dst); dst += dst_stride; - ILV_B_LRLR_UB(src2, src3, src3, src4, vec5, vec4, vec7, vec6); - - tmp2 = __msa_dotp_u_h(vec2, filt0); - tmp3 = __msa_dotp_u_h(vec3, filt0); - - tmp3 = SRARI_SATURATE_UNSIGNED_H(tmp3, FILTER_BITS, 7); - tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); - - PCKEV_B_STORE_VEC(tmp3, tmp2, dst); + ILVR_B2_UB(src3, src2, src4, src3, vec4, vec6); + ILVL_B2_UB(src3, src2, src4, src3, vec5, vec7); + DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3); + SRARI_H2_UH(tmp2, tmp3, FILTER_BITS); + SAT_UH2_UH(tmp2, tmp3, 7); + PCKEV_ST_SB(tmp2, tmp3, dst); dst += dst_stride; - tmp0 = __msa_dotp_u_h(vec4, filt0); - tmp1 = __msa_dotp_u_h(vec5, filt0); - - tmp0 = SRARI_SATURATE_UNSIGNED_H(tmp0, FILTER_BITS, 7); - tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); - - PCKEV_B_STORE_VEC(tmp1, tmp0, dst); + DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1); + SRARI_H2_UH(tmp0, tmp1, FILTER_BITS); + SAT_UH2_UH(tmp0, tmp1, 7); + PCKEV_ST_SB(tmp0, tmp1, dst); dst += dst_stride; - tmp2 = __msa_dotp_u_h(vec6, filt0); - tmp3 = __msa_dotp_u_h(vec7, filt0); - - tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); - tmp3 = SRARI_SATURATE_UNSIGNED_H(tmp3, FILTER_BITS, 7); - - PCKEV_B_STORE_VEC(tmp3, tmp2, dst); + DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3); + SRARI_H2_UH(tmp2, tmp3, FILTER_BITS); + SAT_UH2_UH(tmp2, tmp3, 7); + PCKEV_ST_SB(tmp2, tmp3, dst); dst += dst_stride; src0 = src4; @@ -563,93 +502,68 @@ static void common_vt_2t_32w_msa(const uint8_t *src, int32_t src_stride, v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9; v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0; v8u16 tmp0, tmp1, tmp2, tmp3; - v8u16 filt; + v8i16 filt; /* rearranging filter_y */ - filt = LOAD_UH(filter); - filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0); + filt = LD_SH(filter); + filt0 = (v16u8)__msa_splati_h(filt, 0); - src0 = LOAD_UB(src); - src5 = LOAD_UB(src + 16); + src0 = LD_UB(src); + src5 = LD_UB(src + 16); src += src_stride; for (loop_cnt = (height >> 2); loop_cnt--;) { - LOAD_4VECS_UB(src, src_stride, src1, src2, src3, src4); + LD_UB4(src, src_stride, src1, src2, src3, src4); + ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2); + ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3); - ILV_B_LRLR_UB(src0, src1, src1, src2, vec1, vec0, vec3, vec2); - - LOAD_4VECS_UB(src + 16, src_stride, src6, src7, src8, src9); + LD_UB4(src + 16, src_stride, src6, src7, src8, src9); src += (4 * src_stride); - tmp0 = __msa_dotp_u_h(vec0, filt0); - tmp1 = __msa_dotp_u_h(vec1, filt0); + DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1); + SRARI_H2_UH(tmp0, tmp1, FILTER_BITS); + SAT_UH2_UH(tmp0, tmp1, 7); + PCKEV_ST_SB(tmp0, tmp1, dst); + DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3); + SRARI_H2_UH(tmp2, tmp3, FILTER_BITS); + SAT_UH2_UH(tmp2, tmp3, 7); + PCKEV_ST_SB(tmp2, tmp3, dst + dst_stride); - tmp0 = SRARI_SATURATE_UNSIGNED_H(tmp0, FILTER_BITS, 7); - tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); + ILVR_B2_UB(src3, src2, src4, src3, vec4, vec6); + ILVL_B2_UB(src3, src2, src4, src3, vec5, vec7); + DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1); + SRARI_H2_UH(tmp0, tmp1, FILTER_BITS); + SAT_UH2_UH(tmp0, tmp1, 7); + PCKEV_ST_SB(tmp0, tmp1, dst + 2 * dst_stride); - PCKEV_B_STORE_VEC(tmp1, tmp0, dst); + DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3); + SRARI_H2_UH(tmp2, tmp3, FILTER_BITS); + SAT_UH2_UH(tmp2, tmp3, 7); + PCKEV_ST_SB(tmp2, tmp3, dst + 3 * dst_stride); - tmp2 = __msa_dotp_u_h(vec2, filt0); - tmp3 = __msa_dotp_u_h(vec3, filt0); + ILVR_B2_UB(src6, src5, src7, src6, vec0, vec2); + ILVL_B2_UB(src6, src5, src7, src6, vec1, vec3); + DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1); + SRARI_H2_UH(tmp0, tmp1, FILTER_BITS); + SAT_UH2_UH(tmp0, tmp1, 7); + PCKEV_ST_SB(tmp0, tmp1, dst + 16); - tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); - tmp3 = SRARI_SATURATE_UNSIGNED_H(tmp3, FILTER_BITS, 7); + DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3); + SRARI_H2_UH(tmp2, tmp3, FILTER_BITS); + SAT_UH2_UH(tmp2, tmp3, 7); + PCKEV_ST_SB(tmp2, tmp3, dst + 16 + dst_stride); - PCKEV_B_STORE_VEC(tmp3, tmp2, dst + dst_stride); + ILVR_B2_UB(src8, src7, src9, src8, vec4, vec6); + ILVL_B2_UB(src8, src7, src9, src8, vec5, vec7); + DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1); + SRARI_H2_UH(tmp0, tmp1, FILTER_BITS); + SAT_UH2_UH(tmp0, tmp1, 7); + PCKEV_ST_SB(tmp0, tmp1, dst + 16 + 2 * dst_stride); - ILV_B_LRLR_UB(src2, src3, src3, src4, vec5, vec4, vec7, vec6); - - tmp0 = __msa_dotp_u_h(vec4, filt0); - tmp1 = __msa_dotp_u_h(vec5, filt0); - - tmp0 = SRARI_SATURATE_UNSIGNED_H(tmp0, FILTER_BITS, 7); - tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); - - PCKEV_B_STORE_VEC(tmp1, tmp0, dst + 2 * dst_stride); - - tmp2 = __msa_dotp_u_h(vec6, filt0); - tmp3 = __msa_dotp_u_h(vec7, filt0); - - tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); - tmp3 = SRARI_SATURATE_UNSIGNED_H(tmp3, FILTER_BITS, 7); - - PCKEV_B_STORE_VEC(tmp3, tmp2, dst + 3 * dst_stride); - - ILV_B_LRLR_UB(src5, src6, src6, src7, vec1, vec0, vec3, vec2); - - tmp0 = __msa_dotp_u_h(vec0, filt0); - tmp1 = __msa_dotp_u_h(vec1, filt0); - - tmp0 = SRARI_SATURATE_UNSIGNED_H(tmp0, FILTER_BITS, 7); - tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); - - PCKEV_B_STORE_VEC(tmp1, tmp0, dst + 16); - - tmp2 = __msa_dotp_u_h(vec2, filt0); - tmp3 = __msa_dotp_u_h(vec3, filt0); - - tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); - tmp3 = SRARI_SATURATE_UNSIGNED_H(tmp3, FILTER_BITS, 7); - - PCKEV_B_STORE_VEC(tmp3, tmp2, dst + 16 + dst_stride); - - ILV_B_LRLR_UB(src7, src8, src8, src9, vec5, vec4, vec7, vec6); - - tmp0 = __msa_dotp_u_h(vec4, filt0); - tmp1 = __msa_dotp_u_h(vec5, filt0); - - tmp0 = SRARI_SATURATE_UNSIGNED_H(tmp0, FILTER_BITS, 7); - tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); - - PCKEV_B_STORE_VEC(tmp1, tmp0, dst + 16 + 2 * dst_stride); - - tmp2 = __msa_dotp_u_h(vec6, filt0); - tmp3 = __msa_dotp_u_h(vec7, filt0); - - tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); - tmp3 = SRARI_SATURATE_UNSIGNED_H(tmp3, FILTER_BITS, 7); - - PCKEV_B_STORE_VEC(tmp3, tmp2, dst + 16 + 3 * dst_stride); + DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3); + SRARI_H2_UH(tmp2, tmp3, FILTER_BITS); + SAT_UH2_UH(tmp2, tmp3, 7); + PCKEV_ST_SB(tmp2, tmp3, dst + 16 + 3 * dst_stride); dst += (4 * dst_stride); src0 = src4; @@ -661,97 +575,72 @@ static void common_vt_2t_64w_msa(const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride, int8_t *filter, int32_t height) { uint32_t loop_cnt; - v16u8 src0, src1, src2, src3, src4, src5, src6, src7; - v16u8 src8, src9, src10, src11; - v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0; + v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10; + v16u8 src11, vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0; v8u16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; - v8u16 filt; + v8i16 filt; /* rearranging filter_y */ - filt = LOAD_UH(filter); - filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0); + filt = LD_SH(filter); + filt0 = (v16u8)__msa_splati_h(filt, 0); - LOAD_4VECS_UB(src, 16, src0, src3, src6, src9); + LD_UB4(src, 16, src0, src3, src6, src9); src += src_stride; for (loop_cnt = (height >> 1); loop_cnt--;) { - LOAD_2VECS_UB(src, src_stride, src1, src2); - LOAD_2VECS_UB(src + 16, src_stride, src4, src5); - LOAD_2VECS_UB(src + 32, src_stride, src7, src8); - LOAD_2VECS_UB(src + 48, src_stride, src10, src11); + LD_UB2(src, src_stride, src1, src2); + LD_UB2(src + 16, src_stride, src4, src5); + LD_UB2(src + 32, src_stride, src7, src8); + LD_UB2(src + 48, src_stride, src10, src11); src += (2 * src_stride); - ILV_B_LRLR_UB(src0, src1, src1, src2, vec1, vec0, vec3, vec2); + ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2); + ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3); + DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1); + SRARI_H2_UH(tmp0, tmp1, FILTER_BITS); + SAT_UH2_UH(tmp0, tmp1, 7); + PCKEV_ST_SB(tmp0, tmp1, dst); - tmp0 = __msa_dotp_u_h(vec0, filt0); - tmp1 = __msa_dotp_u_h(vec1, filt0); + DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3); + SRARI_H2_UH(tmp2, tmp3, FILTER_BITS); + SAT_UH2_UH(tmp2, tmp3, 7); + PCKEV_ST_SB(tmp2, tmp3, dst + dst_stride); - tmp0 = SRARI_SATURATE_UNSIGNED_H(tmp0, FILTER_BITS, 7); - tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); + ILVR_B2_UB(src4, src3, src5, src4, vec4, vec6); + ILVL_B2_UB(src4, src3, src5, src4, vec5, vec7); + DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp4, tmp5); + SRARI_H2_UH(tmp4, tmp5, FILTER_BITS); + SAT_UH2_UH(tmp4, tmp5, 7); + PCKEV_ST_SB(tmp4, tmp5, dst + 16); - PCKEV_B_STORE_VEC(tmp1, tmp0, dst); + DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp6, tmp7); + SRARI_H2_UH(tmp6, tmp7, FILTER_BITS); + SAT_UH2_UH(tmp6, tmp7, 7); + PCKEV_ST_SB(tmp6, tmp7, dst + 16 + dst_stride); - tmp2 = __msa_dotp_u_h(vec2, filt0); - tmp3 = __msa_dotp_u_h(vec3, filt0); + ILVR_B2_UB(src7, src6, src8, src7, vec0, vec2); + ILVL_B2_UB(src7, src6, src8, src7, vec1, vec3); + DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1); + SRARI_H2_UH(tmp0, tmp1, FILTER_BITS); + SAT_UH2_UH(tmp0, tmp1, 7); + PCKEV_ST_SB(tmp0, tmp1, dst + 32); - tmp3 = SRARI_SATURATE_UNSIGNED_H(tmp3, FILTER_BITS, 7); - tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); + DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3); + SRARI_H2_UH(tmp2, tmp3, FILTER_BITS); + SAT_UH2_UH(tmp2, tmp3, 7); + PCKEV_ST_SB(tmp2, tmp3, dst + 32 + dst_stride); - PCKEV_B_STORE_VEC(tmp3, tmp2, dst + dst_stride); + ILVR_B2_UB(src10, src9, src11, src10, vec4, vec6); + ILVL_B2_UB(src10, src9, src11, src10, vec5, vec7); + DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp4, tmp5); + SRARI_H2_UH(tmp4, tmp5, FILTER_BITS); + SAT_UH2_UH(tmp4, tmp5, 7); + PCKEV_ST_SB(tmp4, tmp5, dst + 48); - ILV_B_LRLR_UB(src3, src4, src4, src5, vec5, vec4, vec7, vec6); - - tmp4 = __msa_dotp_u_h(vec4, filt0); - tmp5 = __msa_dotp_u_h(vec5, filt0); - - tmp4 = SRARI_SATURATE_UNSIGNED_H(tmp4, FILTER_BITS, 7); - tmp5 = SRARI_SATURATE_UNSIGNED_H(tmp5, FILTER_BITS, 7); - - PCKEV_B_STORE_VEC(tmp5, tmp4, dst + 16); - - tmp6 = __msa_dotp_u_h(vec6, filt0); - tmp7 = __msa_dotp_u_h(vec7, filt0); - - tmp6 = SRARI_SATURATE_UNSIGNED_H(tmp6, FILTER_BITS, 7); - tmp7 = SRARI_SATURATE_UNSIGNED_H(tmp7, FILTER_BITS, 7); - - PCKEV_B_STORE_VEC(tmp7, tmp6, dst + 16 + dst_stride); - - ILV_B_LRLR_UB(src6, src7, src7, src8, vec1, vec0, vec3, vec2); - - tmp0 = __msa_dotp_u_h(vec0, filt0); - tmp1 = __msa_dotp_u_h(vec1, filt0); - - tmp0 = SRARI_SATURATE_UNSIGNED_H(tmp0, FILTER_BITS, 7); - tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); - - PCKEV_B_STORE_VEC(tmp1, tmp0, dst + 32); - - tmp2 = __msa_dotp_u_h(vec2, filt0); - tmp3 = __msa_dotp_u_h(vec3, filt0); - - tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); - tmp3 = SRARI_SATURATE_UNSIGNED_H(tmp3, FILTER_BITS, 7); - - PCKEV_B_STORE_VEC(tmp3, tmp2, dst + 32 + dst_stride); - - ILV_B_LRLR_UB(src9, src10, src10, src11, vec5, vec4, vec7, vec6); - - tmp4 = __msa_dotp_u_h(vec4, filt0); - tmp5 = __msa_dotp_u_h(vec5, filt0); - - tmp4 = SRARI_SATURATE_UNSIGNED_H(tmp4, FILTER_BITS, 7); - tmp5 = SRARI_SATURATE_UNSIGNED_H(tmp5, FILTER_BITS, 7); - - PCKEV_B_STORE_VEC(tmp5, tmp4, dst + 48); - - tmp6 = __msa_dotp_u_h(vec6, filt0); - tmp7 = __msa_dotp_u_h(vec7, filt0); - - tmp6 = SRARI_SATURATE_UNSIGNED_H(tmp6, FILTER_BITS, 7); - tmp7 = SRARI_SATURATE_UNSIGNED_H(tmp7, FILTER_BITS, 7); - - PCKEV_B_STORE_VEC(tmp7, tmp6, dst + 48 + dst_stride); + DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp6, tmp7); + SRARI_H2_UH(tmp6, tmp7, FILTER_BITS); + SAT_UH2_UH(tmp6, tmp7, 7); + PCKEV_ST_SB(tmp6, tmp7, dst + 48 + dst_stride); dst += (2 * dst_stride); src0 = src2; diff --git a/vp9/common/mips/msa/vp9_convolve_avg_msa.c b/vp9/common/mips/msa/vp9_convolve_avg_msa.c index 72b8ab713..eb8776078 100644 --- a/vp9/common/mips/msa/vp9_convolve_avg_msa.c +++ b/vp9/common/mips/msa/vp9_convolve_avg_msa.c @@ -19,46 +19,35 @@ static void avg_width4_msa(const uint8_t *src, int32_t src_stride, if (0 == (height % 4)) { for (cnt = (height / 4); cnt--;) { - LOAD_4VECS_UB(src, src_stride, src0, src1, src2, src3); + LD_UB4(src, src_stride, src0, src1, src2, src3); src += (4 * src_stride); - LOAD_4VECS_UB(dst, dst_stride, dst0, dst1, dst2, dst3); + LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3); - dst0 = __msa_aver_u_b(src0, dst0); - dst1 = __msa_aver_u_b(src1, dst1); - dst2 = __msa_aver_u_b(src2, dst2); - dst3 = __msa_aver_u_b(src3, dst3); + AVER_UB4_UB(src0, dst0, src1, dst1, src2, dst2, src3, dst3, + dst0, dst1, dst2, dst3); out0 = __msa_copy_u_w((v4i32)dst0, 0); out1 = __msa_copy_u_w((v4i32)dst1, 0); out2 = __msa_copy_u_w((v4i32)dst2, 0); out3 = __msa_copy_u_w((v4i32)dst3, 0); - - STORE_WORD(dst, out0); - dst += dst_stride; - STORE_WORD(dst, out1); - dst += dst_stride; - STORE_WORD(dst, out2); - dst += dst_stride; - STORE_WORD(dst, out3); - dst += dst_stride; + SW4(out0, out1, out2, out3, dst, dst_stride); + dst += (4 * dst_stride); } } else if (0 == (height % 2)) { for (cnt = (height / 2); cnt--;) { - LOAD_2VECS_UB(src, src_stride, src0, src1); + LD_UB2(src, src_stride, src0, src1); src += (2 * src_stride); - LOAD_2VECS_UB(dst, dst_stride, dst0, dst1); + LD_UB2(dst, dst_stride, dst0, dst1); - dst0 = __msa_aver_u_b(src0, dst0); - dst1 = __msa_aver_u_b(src1, dst1); + AVER_UB2_UB(src0, dst0, src1, dst1, dst0, dst1); out0 = __msa_copy_u_w((v4i32)dst0, 0); out1 = __msa_copy_u_w((v4i32)dst1, 0); - - STORE_WORD(dst, out0); + SW(out0, dst); dst += dst_stride; - STORE_WORD(dst, out1); + SW(out1, dst); dst += dst_stride; } } @@ -72,29 +61,19 @@ static void avg_width8_msa(const uint8_t *src, int32_t src_stride, v16u8 dst0, dst1, dst2, dst3; for (cnt = (height / 4); cnt--;) { - LOAD_4VECS_UB(src, src_stride, src0, src1, src2, src3); + LD_UB4(src, src_stride, src0, src1, src2, src3); src += (4 * src_stride); + LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3); - LOAD_4VECS_UB(dst, dst_stride, dst0, dst1, dst2, dst3); - - dst0 = __msa_aver_u_b(src0, dst0); - dst1 = __msa_aver_u_b(src1, dst1); - dst2 = __msa_aver_u_b(src2, dst2); - dst3 = __msa_aver_u_b(src3, dst3); + AVER_UB4_UB(src0, dst0, src1, dst1, src2, dst2, src3, dst3, + dst0, dst1, dst2, dst3); out0 = __msa_copy_u_d((v2i64)dst0, 0); out1 = __msa_copy_u_d((v2i64)dst1, 0); out2 = __msa_copy_u_d((v2i64)dst2, 0); out3 = __msa_copy_u_d((v2i64)dst3, 0); - - STORE_DWORD(dst, out0); - dst += dst_stride; - STORE_DWORD(dst, out1); - dst += dst_stride; - STORE_DWORD(dst, out2); - dst += dst_stride; - STORE_DWORD(dst, out3); - dst += dst_stride; + SD4(out0, out1, out2, out3, dst, dst_stride); + dst += (4 * dst_stride); } } @@ -105,24 +84,15 @@ static void avg_width16_msa(const uint8_t *src, int32_t src_stride, v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7; for (cnt = (height / 8); cnt--;) { - LOAD_8VECS_UB(src, src_stride, - src0, src1, src2, src3, src4, src5, src6, src7); + LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7); src += (8 * src_stride); + LD_UB8(dst, dst_stride, dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7); - LOAD_8VECS_UB(dst, dst_stride, - dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7); - - dst0 = __msa_aver_u_b(src0, dst0); - dst1 = __msa_aver_u_b(src1, dst1); - dst2 = __msa_aver_u_b(src2, dst2); - dst3 = __msa_aver_u_b(src3, dst3); - dst4 = __msa_aver_u_b(src4, dst4); - dst5 = __msa_aver_u_b(src5, dst5); - dst6 = __msa_aver_u_b(src6, dst6); - dst7 = __msa_aver_u_b(src7, dst7); - - STORE_8VECS_UB(dst, dst_stride, - dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7); + AVER_UB4_UB(src0, dst0, src1, dst1, src2, dst2, src3, dst3, + dst0, dst1, dst2, dst3); + AVER_UB4_UB(src4, dst4, src5, dst5, src6, dst6, src7, dst7, + dst4, dst5, dst6, dst7); + ST_UB8(dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst, dst_stride); dst += (8 * dst_stride); } } @@ -137,99 +107,34 @@ static void avg_width32_msa(const uint8_t *src, int32_t src_stride, v16u8 dst8, dst9, dst10, dst11, dst12, dst13, dst14, dst15; for (cnt = (height / 8); cnt--;) { - src0 = LOAD_UB(src); - src1 = LOAD_UB(src + 16); - src += src_stride; - src2 = LOAD_UB(src); - src3 = LOAD_UB(src + 16); - src += src_stride; - src4 = LOAD_UB(src); - src5 = LOAD_UB(src + 16); - src += src_stride; - src6 = LOAD_UB(src); - src7 = LOAD_UB(src + 16); - src += src_stride; + LD_UB4(src, src_stride, src0, src2, src4, src6); + LD_UB4(src + 16, src_stride, src1, src3, src5, src7); + src += (4 * src_stride); + LD_UB4(dst_dup, dst_stride, dst0, dst2, dst4, dst6); + LD_UB4(dst_dup + 16, dst_stride, dst1, dst3, dst5, dst7); + dst_dup += (4 * dst_stride); + LD_UB4(src, src_stride, src8, src10, src12, src14); + LD_UB4(src + 16, src_stride, src9, src11, src13, src15); + src += (4 * src_stride); + LD_UB4(dst_dup, dst_stride, dst8, dst10, dst12, dst14); + LD_UB4(dst_dup + 16, dst_stride, dst9, dst11, dst13, dst15); + dst_dup += (4 * dst_stride); - dst0 = LOAD_UB(dst_dup); - dst1 = LOAD_UB(dst_dup + 16); - dst_dup += dst_stride; - dst2 = LOAD_UB(dst_dup); - dst3 = LOAD_UB(dst_dup + 16); - dst_dup += dst_stride; - dst4 = LOAD_UB(dst_dup); - dst5 = LOAD_UB(dst_dup + 16); - dst_dup += dst_stride; - dst6 = LOAD_UB(dst_dup); - dst7 = LOAD_UB(dst_dup + 16); - dst_dup += dst_stride; + AVER_UB4_UB(src0, dst0, src1, dst1, src2, dst2, src3, dst3, + dst0, dst1, dst2, dst3); + AVER_UB4_UB(src4, dst4, src5, dst5, src6, dst6, src7, dst7, + dst4, dst5, dst6, dst7); + AVER_UB4_UB(src8, dst8, src9, dst9, src10, dst10, src11, dst11, + dst8, dst9, dst10, dst11); + AVER_UB4_UB(src12, dst12, src13, dst13, src14, dst14, src15, dst15, + dst12, dst13, dst14, dst15); - src8 = LOAD_UB(src); - src9 = LOAD_UB(src + 16); - src += src_stride; - src10 = LOAD_UB(src); - src11 = LOAD_UB(src + 16); - src += src_stride; - src12 = LOAD_UB(src); - src13 = LOAD_UB(src + 16); - src += src_stride; - src14 = LOAD_UB(src); - src15 = LOAD_UB(src + 16); - src += src_stride; - - dst8 = LOAD_UB(dst_dup); - dst9 = LOAD_UB(dst_dup + 16); - dst_dup += dst_stride; - dst10 = LOAD_UB(dst_dup); - dst11 = LOAD_UB(dst_dup + 16); - dst_dup += dst_stride; - dst12 = LOAD_UB(dst_dup); - dst13 = LOAD_UB(dst_dup + 16); - dst_dup += dst_stride; - dst14 = LOAD_UB(dst_dup); - dst15 = LOAD_UB(dst_dup + 16); - dst_dup += dst_stride; - - dst0 = __msa_aver_u_b(src0, dst0); - dst1 = __msa_aver_u_b(src1, dst1); - dst2 = __msa_aver_u_b(src2, dst2); - dst3 = __msa_aver_u_b(src3, dst3); - dst4 = __msa_aver_u_b(src4, dst4); - dst5 = __msa_aver_u_b(src5, dst5); - dst6 = __msa_aver_u_b(src6, dst6); - dst7 = __msa_aver_u_b(src7, dst7); - dst8 = __msa_aver_u_b(src8, dst8); - dst9 = __msa_aver_u_b(src9, dst9); - dst10 = __msa_aver_u_b(src10, dst10); - dst11 = __msa_aver_u_b(src11, dst11); - dst12 = __msa_aver_u_b(src12, dst12); - dst13 = __msa_aver_u_b(src13, dst13); - dst14 = __msa_aver_u_b(src14, dst14); - dst15 = __msa_aver_u_b(src15, dst15); - - STORE_UB(dst0, dst); - STORE_UB(dst1, dst + 16); - dst += dst_stride; - STORE_UB(dst2, dst); - STORE_UB(dst3, dst + 16); - dst += dst_stride; - STORE_UB(dst4, dst); - STORE_UB(dst5, dst + 16); - dst += dst_stride; - STORE_UB(dst6, dst); - STORE_UB(dst7, dst + 16); - dst += dst_stride; - STORE_UB(dst8, dst); - STORE_UB(dst9, dst + 16); - dst += dst_stride; - STORE_UB(dst10, dst); - STORE_UB(dst11, dst + 16); - dst += dst_stride; - STORE_UB(dst12, dst); - STORE_UB(dst13, dst + 16); - dst += dst_stride; - STORE_UB(dst14, dst); - STORE_UB(dst15, dst + 16); - dst += dst_stride; + ST_UB4(dst0, dst2, dst4, dst6, dst, dst_stride); + ST_UB4(dst1, dst3, dst5, dst7, dst + 16, dst_stride); + dst += (4 * dst_stride); + ST_UB4(dst8, dst10, dst12, dst14, dst, dst_stride); + ST_UB4(dst9, dst11, dst13, dst15, dst + 16, dst_stride); + dst += (4 * dst_stride); } } @@ -243,48 +148,40 @@ static void avg_width64_msa(const uint8_t *src, int32_t src_stride, v16u8 dst8, dst9, dst10, dst11, dst12, dst13, dst14, dst15; for (cnt = (height / 4); cnt--;) { - LOAD_4VECS_UB(src, 16, src0, src1, src2, src3); + LD_UB4(src, 16, src0, src1, src2, src3); src += src_stride; - LOAD_4VECS_UB(src, 16, src4, src5, src6, src7); + LD_UB4(src, 16, src4, src5, src6, src7); src += src_stride; - LOAD_4VECS_UB(src, 16, src8, src9, src10, src11); + LD_UB4(src, 16, src8, src9, src10, src11); src += src_stride; - LOAD_4VECS_UB(src, 16, src12, src13, src14, src15); + LD_UB4(src, 16, src12, src13, src14, src15); src += src_stride; - LOAD_4VECS_UB(dst_dup, 16, dst0, dst1, dst2, dst3); + LD_UB4(dst_dup, 16, dst0, dst1, dst2, dst3); dst_dup += dst_stride; - LOAD_4VECS_UB(dst_dup, 16, dst4, dst5, dst6, dst7); + LD_UB4(dst_dup, 16, dst4, dst5, dst6, dst7); dst_dup += dst_stride; - LOAD_4VECS_UB(dst_dup, 16, dst8, dst9, dst10, dst11); + LD_UB4(dst_dup, 16, dst8, dst9, dst10, dst11); dst_dup += dst_stride; - LOAD_4VECS_UB(dst_dup, 16, dst12, dst13, dst14, dst15); + LD_UB4(dst_dup, 16, dst12, dst13, dst14, dst15); dst_dup += dst_stride; - dst0 = __msa_aver_u_b(src0, dst0); - dst1 = __msa_aver_u_b(src1, dst1); - dst2 = __msa_aver_u_b(src2, dst2); - dst3 = __msa_aver_u_b(src3, dst3); - dst4 = __msa_aver_u_b(src4, dst4); - dst5 = __msa_aver_u_b(src5, dst5); - dst6 = __msa_aver_u_b(src6, dst6); - dst7 = __msa_aver_u_b(src7, dst7); - dst8 = __msa_aver_u_b(src8, dst8); - dst9 = __msa_aver_u_b(src9, dst9); - dst10 = __msa_aver_u_b(src10, dst10); - dst11 = __msa_aver_u_b(src11, dst11); - dst12 = __msa_aver_u_b(src12, dst12); - dst13 = __msa_aver_u_b(src13, dst13); - dst14 = __msa_aver_u_b(src14, dst14); - dst15 = __msa_aver_u_b(src15, dst15); + AVER_UB4_UB(src0, dst0, src1, dst1, src2, dst2, src3, dst3, + dst0, dst1, dst2, dst3); + AVER_UB4_UB(src4, dst4, src5, dst5, src6, dst6, src7, dst7, + dst4, dst5, dst6, dst7); + AVER_UB4_UB(src8, dst8, src9, dst9, src10, dst10, src11, dst11, + dst8, dst9, dst10, dst11); + AVER_UB4_UB(src12, dst12, src13, dst13, src14, dst14, src15, dst15, + dst12, dst13, dst14, dst15); - STORE_4VECS_UB(dst, 16, dst0, dst1, dst2, dst3); + ST_UB4(dst0, dst1, dst2, dst3, dst, 16); dst += dst_stride; - STORE_4VECS_UB(dst, 16, dst4, dst5, dst6, dst7); + ST_UB4(dst4, dst5, dst6, dst7, dst, 16); dst += dst_stride; - STORE_4VECS_UB(dst, 16, dst8, dst9, dst10, dst11); + ST_UB4(dst8, dst9, dst10, dst11, dst, 16); dst += dst_stride; - STORE_4VECS_UB(dst, 16, dst12, dst13, dst14, dst15); + ST_UB4(dst12, dst13, dst14, dst15, dst, 16); dst += dst_stride; } } diff --git a/vp9/common/mips/msa/vp9_convolve_copy_msa.c b/vp9/common/mips/msa/vp9_convolve_copy_msa.c index 064ba762f..7a292c5ce 100644 --- a/vp9/common/mips/msa/vp9_convolve_copy_msa.c +++ b/vp9/common/mips/msa/vp9_convolve_copy_msa.c @@ -12,16 +12,14 @@ #include "vp9/common/mips/msa/vp9_macros_msa.h" static void copy_width8_msa(const uint8_t *src, int32_t src_stride, - uint8_t *dst, int32_t dst_stride, - int32_t height) { + uint8_t *dst, int32_t dst_stride, int32_t height) { int32_t cnt; uint64_t out0, out1, out2, out3, out4, out5, out6, out7; v16u8 src0, src1, src2, src3, src4, src5, src6, src7; if (0 == height % 12) { for (cnt = (height / 12); cnt--;) { - LOAD_8VECS_UB(src, src_stride, - src0, src1, src2, src3, src4, src5, src6, src7); + LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7); src += (8 * src_stride); out0 = __msa_copy_u_d((v2i64)src0, 0); @@ -33,44 +31,24 @@ static void copy_width8_msa(const uint8_t *src, int32_t src_stride, out6 = __msa_copy_u_d((v2i64)src6, 0); out7 = __msa_copy_u_d((v2i64)src7, 0); - STORE_DWORD(dst, out0); - dst += dst_stride; - STORE_DWORD(dst, out1); - dst += dst_stride; - STORE_DWORD(dst, out2); - dst += dst_stride; - STORE_DWORD(dst, out3); - dst += dst_stride; - STORE_DWORD(dst, out4); - dst += dst_stride; - STORE_DWORD(dst, out5); - dst += dst_stride; - STORE_DWORD(dst, out6); - dst += dst_stride; - STORE_DWORD(dst, out7); - dst += dst_stride; + SD4(out0, out1, out2, out3, dst, dst_stride); + dst += (4 * dst_stride); + SD4(out4, out5, out6, out7, dst, dst_stride); + dst += (4 * dst_stride); - LOAD_4VECS_UB(src, src_stride, src0, src1, src2, src3); + LD_UB4(src, src_stride, src0, src1, src2, src3); src += (4 * src_stride); out0 = __msa_copy_u_d((v2i64)src0, 0); out1 = __msa_copy_u_d((v2i64)src1, 0); out2 = __msa_copy_u_d((v2i64)src2, 0); out3 = __msa_copy_u_d((v2i64)src3, 0); - - STORE_DWORD(dst, out0); - dst += dst_stride; - STORE_DWORD(dst, out1); - dst += dst_stride; - STORE_DWORD(dst, out2); - dst += dst_stride; - STORE_DWORD(dst, out3); - dst += dst_stride; + SD4(out0, out1, out2, out3, dst, dst_stride); + dst += (4 * dst_stride); } } else if (0 == height % 8) { for (cnt = height >> 3; cnt--;) { - LOAD_8VECS_UB(src, src_stride, - src0, src1, src2, src3, src4, src5, src6, src7); + LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7); src += (8 * src_stride); out0 = __msa_copy_u_d((v2i64)src0, 0); @@ -82,53 +60,33 @@ static void copy_width8_msa(const uint8_t *src, int32_t src_stride, out6 = __msa_copy_u_d((v2i64)src6, 0); out7 = __msa_copy_u_d((v2i64)src7, 0); - STORE_DWORD(dst, out0); - dst += dst_stride; - STORE_DWORD(dst, out1); - dst += dst_stride; - STORE_DWORD(dst, out2); - dst += dst_stride; - STORE_DWORD(dst, out3); - dst += dst_stride; - STORE_DWORD(dst, out4); - dst += dst_stride; - STORE_DWORD(dst, out5); - dst += dst_stride; - STORE_DWORD(dst, out6); - dst += dst_stride; - STORE_DWORD(dst, out7); - dst += dst_stride; + SD4(out0, out1, out2, out3, dst, dst_stride); + dst += (4 * dst_stride); + SD4(out4, out5, out6, out7, dst, dst_stride); + dst += (4 * dst_stride); } } else if (0 == height % 4) { for (cnt = (height / 4); cnt--;) { - LOAD_4VECS_UB(src, src_stride, src0, src1, src2, src3); + LD_UB4(src, src_stride, src0, src1, src2, src3); src += (4 * src_stride); - out0 = __msa_copy_u_d((v2i64)src0, 0); out1 = __msa_copy_u_d((v2i64)src1, 0); out2 = __msa_copy_u_d((v2i64)src2, 0); out3 = __msa_copy_u_d((v2i64)src3, 0); - STORE_DWORD(dst, out0); - dst += dst_stride; - STORE_DWORD(dst, out1); - dst += dst_stride; - STORE_DWORD(dst, out2); - dst += dst_stride; - STORE_DWORD(dst, out3); - dst += dst_stride; + SD4(out0, out1, out2, out3, dst, dst_stride); + dst += (4 * dst_stride); } } else if (0 == height % 2) { for (cnt = (height / 2); cnt--;) { - LOAD_2VECS_UB(src, src_stride, src0, src1); + LD_UB2(src, src_stride, src0, src1); src += (2 * src_stride); - out0 = __msa_copy_u_d((v2i64)src0, 0); out1 = __msa_copy_u_d((v2i64)src1, 0); - STORE_DWORD(dst, out0); + SD(out0, dst); dst += dst_stride; - STORE_DWORD(dst, out1); + SD(out1, dst); dst += dst_stride; } } @@ -147,12 +105,12 @@ static void copy_16multx8mult_msa(const uint8_t *src, int32_t src_stride, dst_tmp = dst; for (loop_cnt = (height >> 3); loop_cnt--;) { - LOAD_8VECS_UB(src_tmp, src_stride, - src0, src1, src2, src3, src4, src5, src6, src7); + LD_UB8(src_tmp, src_stride, + src0, src1, src2, src3, src4, src5, src6, src7); src_tmp += (8 * src_stride); - STORE_8VECS_UB(dst_tmp, dst_stride, - src0, src1, src2, src3, src4, src5, src6, src7); + ST_UB8(src0, src1, src2, src3, src4, src5, src6, src7, + dst_tmp, dst_stride); dst_tmp += (8 * dst_stride); } @@ -162,90 +120,79 @@ static void copy_16multx8mult_msa(const uint8_t *src, int32_t src_stride, } static void copy_width16_msa(const uint8_t *src, int32_t src_stride, - uint8_t *dst, int32_t dst_stride, - int32_t height) { + uint8_t *dst, int32_t dst_stride, int32_t height) { int32_t cnt; v16u8 src0, src1, src2, src3, src4, src5, src6, src7; if (0 == height % 12) { for (cnt = (height / 12); cnt--;) { - LOAD_8VECS_UB(src, src_stride, - src0, src1, src2, src3, src4, src5, src6, src7); + LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7); src += (8 * src_stride); - - STORE_8VECS_UB(dst, dst_stride, - src0, src1, src2, src3, src4, src5, src6, src7); + ST_UB8(src0, src1, src2, src3, src4, src5, src6, src7, dst, dst_stride); dst += (8 * dst_stride); - LOAD_4VECS_UB(src, src_stride, src0, src1, src2, src3); + LD_UB4(src, src_stride, src0, src1, src2, src3); src += (4 * src_stride); - - STORE_4VECS_UB(dst, dst_stride, src0, src1, src2, src3); + ST_UB4(src0, src1, src2, src3, dst, dst_stride); dst += (4 * dst_stride); } } else if (0 == height % 8) { copy_16multx8mult_msa(src, src_stride, dst, dst_stride, height, 16); } else if (0 == height % 4) { for (cnt = (height >> 2); cnt--;) { - LOAD_4VECS_UB(src, src_stride, src0, src1, src2, src3); + LD_UB4(src, src_stride, src0, src1, src2, src3); src += (4 * src_stride); - STORE_4VECS_UB(dst, dst_stride, src0, src1, src2, src3); + ST_UB4(src0, src1, src2, src3, dst, dst_stride); dst += (4 * dst_stride); } } } static void copy_width32_msa(const uint8_t *src, int32_t src_stride, - uint8_t *dst, int32_t dst_stride, - int32_t height) { + uint8_t *dst, int32_t dst_stride, int32_t height) { int32_t cnt; v16u8 src0, src1, src2, src3, src4, src5, src6, src7; if (0 == height % 12) { for (cnt = (height / 12); cnt--;) { - LOAD_4VECS_UB(src, src_stride, src0, src1, src2, src3); - LOAD_4VECS_UB(src + 16, src_stride, src4, src5, src6, src7); + LD_UB4(src, src_stride, src0, src1, src2, src3); + LD_UB4(src + 16, src_stride, src4, src5, src6, src7); src += (4 * src_stride); - - STORE_4VECS_UB(dst, dst_stride, src0, src1, src2, src3); - STORE_4VECS_UB(dst + 16, dst_stride, src4, src5, src6, src7); + ST_UB4(src0, src1, src2, src3, dst, dst_stride); + ST_UB4(src4, src5, src6, src7, dst + 16, dst_stride); dst += (4 * dst_stride); - LOAD_4VECS_UB(src, src_stride, src0, src1, src2, src3); - LOAD_4VECS_UB(src + 16, src_stride, src4, src5, src6, src7); + LD_UB4(src, src_stride, src0, src1, src2, src3); + LD_UB4(src + 16, src_stride, src4, src5, src6, src7); src += (4 * src_stride); - - STORE_4VECS_UB(dst, dst_stride, src0, src1, src2, src3); - STORE_4VECS_UB(dst + 16, dst_stride, src4, src5, src6, src7); + ST_UB4(src0, src1, src2, src3, dst, dst_stride); + ST_UB4(src4, src5, src6, src7, dst + 16, dst_stride); dst += (4 * dst_stride); - LOAD_4VECS_UB(src, src_stride, src0, src1, src2, src3); - LOAD_4VECS_UB(src + 16, src_stride, src4, src5, src6, src7); + LD_UB4(src, src_stride, src0, src1, src2, src3); + LD_UB4(src + 16, src_stride, src4, src5, src6, src7); src += (4 * src_stride); - - STORE_4VECS_UB(dst, dst_stride, src0, src1, src2, src3); - STORE_4VECS_UB(dst + 16, dst_stride, src4, src5, src6, src7); + ST_UB4(src0, src1, src2, src3, dst, dst_stride); + ST_UB4(src4, src5, src6, src7, dst + 16, dst_stride); dst += (4 * dst_stride); } } else if (0 == height % 8) { copy_16multx8mult_msa(src, src_stride, dst, dst_stride, height, 32); } else if (0 == height % 4) { for (cnt = (height >> 2); cnt--;) { - LOAD_4VECS_UB(src, src_stride, src0, src1, src2, src3); - LOAD_4VECS_UB(src + 16, src_stride, src4, src5, src6, src7); + LD_UB4(src, src_stride, src0, src1, src2, src3); + LD_UB4(src + 16, src_stride, src4, src5, src6, src7); src += (4 * src_stride); - - STORE_4VECS_UB(dst, dst_stride, src0, src1, src2, src3); - STORE_4VECS_UB(dst + 16, dst_stride, src4, src5, src6, src7); + ST_UB4(src0, src1, src2, src3, dst, dst_stride); + ST_UB4(src4, src5, src6, src7, dst + 16, dst_stride); dst += (4 * dst_stride); } } } static void copy_width64_msa(const uint8_t *src, int32_t src_stride, - uint8_t *dst, int32_t dst_stride, - int32_t height) { + uint8_t *dst, int32_t dst_stride, int32_t height) { copy_16multx8mult_msa(src, src_stride, dst, dst_stride, height, 64); } @@ -264,8 +211,8 @@ void vp9_convolve_copy_msa(const uint8_t *src, ptrdiff_t src_stride, uint32_t cnt, tmp; /* 1 word storage */ for (cnt = h; cnt--;) { - tmp = LOAD_WORD(src); - STORE_WORD(dst, tmp); + tmp = LW(src); + SW(tmp, dst); src += src_stride; dst += dst_stride; } diff --git a/vp9/common/mips/msa/vp9_convolve_msa.h b/vp9/common/mips/msa/vp9_convolve_msa.h index b109a4014..40fe94d3b 100644 --- a/vp9/common/mips/msa/vp9_convolve_msa.h +++ b/vp9/common/mips/msa/vp9_convolve_msa.h @@ -16,142 +16,104 @@ extern const uint8_t mc_filt_mask_arr[16 * 3]; -#define HORIZ_8TAP_FILT(src, mask0, mask1, mask2, mask3, \ - filt_h0, filt_h1, filt_h2, filt_h3) ({ \ - v8i16 vec0, vec1, vec2, vec3, horiz_out; \ - \ - vec0 = (v8i16)__msa_vshf_b((v16i8)(mask0), (v16i8)(src), (v16i8)(src)); \ - vec0 = __msa_dotp_s_h((v16i8)vec0, (v16i8)(filt_h0)); \ - vec1 = (v8i16)__msa_vshf_b((v16i8)(mask1), (v16i8)(src), (v16i8)(src)); \ - vec0 = __msa_dpadd_s_h(vec0, (v16i8)(filt_h1), (v16i8)vec1); \ - vec2 = (v8i16)__msa_vshf_b((v16i8)(mask2), (v16i8)(src), (v16i8)(src)); \ - vec2 = __msa_dotp_s_h((v16i8)vec2, (v16i8)(filt_h2)); \ - vec3 = (v8i16)__msa_vshf_b((v16i8)(mask3), (v16i8)(src), (v16i8)(src)); \ - vec2 = __msa_dpadd_s_h(vec2, (v16i8)(filt_h3), (v16i8)vec3); \ - vec0 = __msa_adds_s_h(vec0, vec2); \ - horiz_out = SRARI_SATURATE_SIGNED_H(vec0, FILTER_BITS, 7); \ - \ - horiz_out; \ +#define FILT_8TAP_DPADD_S_H(vec0, vec1, vec2, vec3, \ + filt0, filt1, filt2, filt3) ({ \ + v8i16 tmp0, tmp1; \ + \ + tmp0 = __msa_dotp_s_h((v16i8)vec0, (v16i8)filt0); \ + tmp0 = __msa_dpadd_s_h(tmp0, (v16i8)vec1, (v16i8)filt1); \ + tmp1 = __msa_dotp_s_h((v16i8)vec2, (v16i8)filt2); \ + tmp1 = __msa_dpadd_s_h(tmp1, (v16i8)vec3, (v16i8)filt3); \ + tmp0 = __msa_adds_s_h(tmp0, tmp1); \ + \ + tmp0; \ }) -#define HORIZ_8TAP_FILT_2VECS(src0, src1, mask0, mask1, mask2, mask3, \ - filt_h0, filt_h1, filt_h2, filt_h3) ({ \ - v8i16 vec0, vec1, vec2, vec3, horiz_out; \ - \ - vec0 = (v8i16)__msa_vshf_b((v16i8)(mask0), (v16i8)(src1), (v16i8)(src0)); \ - vec0 = __msa_dotp_s_h((v16i8)vec0, (v16i8)(filt_h0)); \ - vec1 = (v8i16)__msa_vshf_b((v16i8)(mask1), (v16i8)(src1), (v16i8)(src0)); \ - vec0 = __msa_dpadd_s_h(vec0, (v16i8)(filt_h1), (v16i8)vec1); \ - vec2 = (v8i16)__msa_vshf_b((v16i8)(mask2), (v16i8)(src1), (v16i8)(src0)); \ - vec2 = __msa_dotp_s_h((v16i8)vec2, (v16i8)(filt_h2)); \ - vec3 = (v8i16)__msa_vshf_b((v16i8)(mask3), (v16i8)(src1), (v16i8)(src0)); \ - vec2 = __msa_dpadd_s_h(vec2, ((v16i8)filt_h3), (v16i8)vec3); \ - vec0 = __msa_adds_s_h(vec0, vec2); \ - horiz_out = (v8i16)SRARI_SATURATE_SIGNED_H(vec0, FILTER_BITS, 7); \ - \ - horiz_out; \ +#define HORIZ_8TAP_FILT(src0, src1, mask0, mask1, mask2, mask3, \ + filt_h0, filt_h1, filt_h2, filt_h3) ({ \ + v16i8 vec0_m, vec1_m, vec2_m, vec3_m; \ + v8i16 hz_out_m; \ + \ + VSHF_B4_SB(src0, src1, mask0, mask1, mask2, mask3, \ + vec0_m, vec1_m, vec2_m, vec3_m); \ + hz_out_m = FILT_8TAP_DPADD_S_H(vec0_m, vec1_m, vec2_m, vec3_m, \ + filt_h0, filt_h1, filt_h2, filt_h3); \ + \ + hz_out_m = __msa_srari_h(hz_out_m, FILTER_BITS); \ + hz_out_m = __msa_sat_s_h(hz_out_m, 7); \ + \ + hz_out_m; \ }) -#define FILT_8TAP_DPADD_S_H(vec0, vec1, vec2, vec3, \ - filt0, filt1, filt2, filt3) ({ \ - v8i16 tmp0, tmp1; \ - \ - tmp0 = __msa_dotp_s_h((v16i8)(vec0), (v16i8)(filt0)); \ - tmp0 = __msa_dpadd_s_h(tmp0, (v16i8)(vec1), (v16i8)(filt1)); \ - tmp1 = __msa_dotp_s_h((v16i8)(vec2), (v16i8)(filt2)); \ - tmp1 = __msa_dpadd_s_h(tmp1, (v16i8)(vec3), ((v16i8)filt3)); \ - tmp0 = __msa_adds_s_h(tmp0, tmp1); \ - \ - tmp0; \ -}) - -#define HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, \ - mask0, mask1, mask2, mask3, \ - filt0, filt1, filt2, filt3, \ - out0, out1) { \ - v8i16 vec0_m, vec1_m, vec2_m, vec3_m, vec4_m, vec5_m, vec6_m, vec7_m; \ - v8i16 res0_m, res1_m, res2_m, res3_m; \ - \ - vec0_m = (v8i16)__msa_vshf_b((v16i8)(mask0), (v16i8)(src1), (v16i8)(src0)); \ - vec1_m = (v8i16)__msa_vshf_b((v16i8)(mask0), (v16i8)(src3), (v16i8)(src2)); \ - \ - res0_m = __msa_dotp_s_h((v16i8)vec0_m, (v16i8)(filt0)); \ - res1_m = __msa_dotp_s_h((v16i8)vec1_m, (v16i8)(filt0)); \ - \ - vec2_m = (v8i16)__msa_vshf_b((v16i8)(mask1), (v16i8)(src1), (v16i8)(src0)); \ - vec3_m = (v8i16)__msa_vshf_b((v16i8)(mask1), (v16i8)(src3), (v16i8)(src2)); \ - \ - res0_m = __msa_dpadd_s_h(res0_m, (filt1), (v16i8)vec2_m); \ - res1_m = __msa_dpadd_s_h(res1_m, (filt1), (v16i8)vec3_m); \ - \ - vec4_m = (v8i16)__msa_vshf_b((v16i8)(mask2), (v16i8)(src1), (v16i8)(src0)); \ - vec5_m = (v8i16)__msa_vshf_b((v16i8)(mask2), (v16i8)(src3), (v16i8)(src2)); \ - \ - res2_m = __msa_dotp_s_h((v16i8)(filt2), (v16i8)vec4_m); \ - res3_m = __msa_dotp_s_h((v16i8)(filt2), (v16i8)vec5_m); \ - \ - vec6_m = (v8i16)__msa_vshf_b((v16i8)(mask3), (v16i8)(src1), (v16i8)(src0)); \ - vec7_m = (v8i16)__msa_vshf_b((v16i8)(mask3), (v16i8)(src3), (v16i8)(src2)); \ - \ - res2_m = __msa_dpadd_s_h(res2_m, (v16i8)(filt3), (v16i8)vec6_m); \ - res3_m = __msa_dpadd_s_h(res3_m, (v16i8)(filt3), (v16i8)vec7_m); \ - \ - out0 = __msa_adds_s_h(res0_m, res2_m); \ - out1 = __msa_adds_s_h(res1_m, res3_m); \ +#define HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, \ + mask0, mask1, mask2, mask3, \ + filt0, filt1, filt2, filt3, \ + out0, out1) { \ + v16i8 vec0_m, vec1_m, vec2_m, vec3_m, vec4_m, vec5_m, vec6_m, vec7_m; \ + v8i16 res0_m, res1_m, res2_m, res3_m; \ + \ + VSHF_B2_SB(src0, src1, src2, src3, mask0, mask0, vec0_m, vec1_m); \ + DOTP_SB2_SH(vec0_m, vec1_m, filt0, filt0, res0_m, res1_m); \ + VSHF_B2_SB(src0, src1, src2, src3, mask1, mask1, vec2_m, vec3_m); \ + DPADD_SB2_SH(vec2_m, vec3_m, filt1, filt1, res0_m, res1_m); \ + VSHF_B2_SB(src0, src1, src2, src3, mask2, mask2, vec4_m, vec5_m); \ + DOTP_SB2_SH(vec4_m, vec5_m, filt2, filt2, res2_m, res3_m); \ + VSHF_B2_SB(src0, src1, src2, src3, mask3, mask3, vec6_m, vec7_m); \ + DPADD_SB2_SH(vec6_m, vec7_m, filt3, filt3, res2_m, res3_m); \ + ADDS_SH2_SH(res0_m, res2_m, res1_m, res3_m, out0, out1); \ } -#define HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, \ - mask0, mask1, mask2, mask3, \ - filt0, filt1, filt2, filt3, \ - out0, out1, out2, out3) { \ - v8i16 vec0_m, vec1_m, vec2_m, vec3_m; \ - v8i16 vec4_m, vec5_m, vec6_m, vec7_m; \ - v8i16 res0_m, res1_m, res2_m, res3_m; \ - v8i16 res4_m, res5_m, res6_m, res7_m; \ - \ - vec0_m = (v8i16)__msa_vshf_b((v16i8)(mask0), (v16i8)(src0), (v16i8)(src0)); \ - vec1_m = (v8i16)__msa_vshf_b((v16i8)(mask0), (v16i8)(src1), (v16i8)(src1)); \ - vec2_m = (v8i16)__msa_vshf_b((v16i8)(mask0), (v16i8)(src2), (v16i8)(src2)); \ - vec3_m = (v8i16)__msa_vshf_b((v16i8)(mask0), (v16i8)(src3), (v16i8)(src3)); \ - \ - res0_m = __msa_dotp_s_h((v16i8)vec0_m, (v16i8)(filt0)); \ - res1_m = __msa_dotp_s_h((v16i8)vec1_m, (v16i8)(filt0)); \ - res2_m = __msa_dotp_s_h((v16i8)vec2_m, (v16i8)(filt0)); \ - res3_m = __msa_dotp_s_h((v16i8)vec3_m, (v16i8)(filt0)); \ - \ - vec0_m = (v8i16)__msa_vshf_b((v16i8)(mask2), (v16i8)(src0), (v16i8)(src0)); \ - vec1_m = (v8i16)__msa_vshf_b((v16i8)(mask2), (v16i8)(src1), (v16i8)(src1)); \ - vec2_m = (v8i16)__msa_vshf_b((v16i8)(mask2), (v16i8)(src2), (v16i8)(src2)); \ - vec3_m = (v8i16)__msa_vshf_b((v16i8)(mask2), (v16i8)(src3), (v16i8)(src3)); \ - \ - res4_m = __msa_dotp_s_h((v16i8)vec0_m, (v16i8)(filt2)); \ - res5_m = __msa_dotp_s_h((v16i8)vec1_m, (v16i8)(filt2)); \ - res6_m = __msa_dotp_s_h((v16i8)vec2_m, (v16i8)(filt2)); \ - res7_m = __msa_dotp_s_h((v16i8)vec3_m, (v16i8)(filt2)); \ - \ - vec4_m = (v8i16)__msa_vshf_b((v16i8)(mask1), (v16i8)(src0), (v16i8)(src0)); \ - vec5_m = (v8i16)__msa_vshf_b((v16i8)(mask1), (v16i8)(src1), (v16i8)(src1)); \ - vec6_m = (v8i16)__msa_vshf_b((v16i8)(mask1), (v16i8)(src2), (v16i8)(src2)); \ - vec7_m = (v8i16)__msa_vshf_b((v16i8)(mask1), (v16i8)(src3), (v16i8)(src3)); \ - \ - res0_m = __msa_dpadd_s_h(res0_m, (v16i8)(filt1), (v16i8)vec4_m); \ - res1_m = __msa_dpadd_s_h(res1_m, (v16i8)(filt1), (v16i8)vec5_m); \ - res2_m = __msa_dpadd_s_h(res2_m, (v16i8)(filt1), (v16i8)vec6_m); \ - res3_m = __msa_dpadd_s_h(res3_m, (v16i8)(filt1), (v16i8)vec7_m); \ - \ - vec4_m = (v8i16)__msa_vshf_b((v16i8)(mask3), (v16i8)(src0), (v16i8)(src0)); \ - vec5_m = (v8i16)__msa_vshf_b((v16i8)(mask3), (v16i8)(src1), (v16i8)(src1)); \ - vec6_m = (v8i16)__msa_vshf_b((v16i8)(mask3), (v16i8)(src2), (v16i8)(src2)); \ - vec7_m = (v8i16)__msa_vshf_b((v16i8)(mask3), (v16i8)(src3), (v16i8)(src3)); \ - \ - res4_m = __msa_dpadd_s_h(res4_m, (v16i8)(filt3), (v16i8)vec4_m); \ - res5_m = __msa_dpadd_s_h(res5_m, (v16i8)(filt3), (v16i8)vec5_m); \ - res6_m = __msa_dpadd_s_h(res6_m, (v16i8)(filt3), (v16i8)vec6_m); \ - res7_m = __msa_dpadd_s_h(res7_m, (v16i8)(filt3), (v16i8)vec7_m); \ - \ - out0 = __msa_adds_s_h(res0_m, res4_m); \ - out1 = __msa_adds_s_h(res1_m, res5_m); \ - out2 = __msa_adds_s_h(res2_m, res6_m); \ - out3 = __msa_adds_s_h(res3_m, res7_m); \ +#define HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, \ + mask0, mask1, mask2, mask3, \ + filt0, filt1, filt2, filt3, \ + out0, out1, out2, out3) { \ + v16i8 vec0_m, vec1_m, vec2_m, vec3_m, vec4_m, vec5_m, vec6_m, vec7_m; \ + v8i16 res0_m, res1_m, res2_m, res3_m, res4_m, res5_m, res6_m, res7_m; \ + \ + VSHF_B2_SB(src0, src0, src1, src1, mask0, mask0, vec0_m, vec1_m); \ + VSHF_B2_SB(src2, src2, src3, src3, mask0, mask0, vec2_m, vec3_m); \ + DOTP_SB4_SH(vec0_m, vec1_m, vec2_m, vec3_m, filt0, filt0, filt0, filt0, \ + res0_m, res1_m, res2_m, res3_m); \ + VSHF_B2_SB(src0, src0, src1, src1, mask2, mask2, vec0_m, vec1_m); \ + VSHF_B2_SB(src2, src2, src3, src3, mask2, mask2, vec2_m, vec3_m); \ + DOTP_SB4_SH(vec0_m, vec1_m, vec2_m, vec3_m, filt2, filt2, filt2, filt2, \ + res4_m, res5_m, res6_m, res7_m); \ + VSHF_B2_SB(src0, src0, src1, src1, mask1, mask1, vec4_m, vec5_m); \ + VSHF_B2_SB(src2, src2, src3, src3, mask1, mask1, vec6_m, vec7_m); \ + DPADD_SB4_SH(vec4_m, vec5_m, vec6_m, vec7_m, filt1, filt1, filt1, filt1, \ + res0_m, res1_m, res2_m, res3_m); \ + VSHF_B2_SB(src0, src0, src1, src1, mask3, mask3, vec4_m, vec5_m); \ + VSHF_B2_SB(src2, src2, src3, src3, mask3, mask3, vec6_m, vec7_m); \ + DPADD_SB4_SH(vec4_m, vec5_m, vec6_m, vec7_m, filt3, filt3, filt3, filt3, \ + res4_m, res5_m, res6_m, res7_m); \ + ADDS_SH4_SH(res0_m, res4_m, res1_m, res5_m, res2_m, res6_m, res3_m, \ + res7_m, out0, out1, out2, out3); \ +} + +#define PCKEV_XORI128_AVG_ST_UB(in0, in1, dst, pdst) { \ + v16u8 tmp_m; \ + \ + tmp_m = PCKEV_XORI128_UB(in1, in0); \ + tmp_m = __msa_aver_u_b(tmp_m, (v16u8)dst); \ + ST_UB(tmp_m, (pdst)); \ +} + +#define PCKEV_AVG_ST_UB(in0, in1, dst, pdst) { \ + v16u8 tmp_m; \ + \ + tmp_m = (v16u8)__msa_pckev_b((v16i8)in0, (v16i8)in1); \ + tmp_m = __msa_aver_u_b(tmp_m, (v16u8)dst); \ + ST_UB(tmp_m, (pdst)); \ +} + +#define PCKEV_AVG_ST8x4_UB(in1, dst0, in2, dst1, in3, dst2, in4, dst3, \ + pdst, stride) { \ + v16u8 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \ + uint8_t *pdst_m = (uint8_t *)(pdst); \ + \ + PCKEV_B2_UB(in2, in1, in4, in3, tmp0_m, tmp1_m); \ + PCKEV_D2_UB(dst1, dst0, dst3, dst2, tmp2_m, tmp3_m); \ + AVER_UB2_UB(tmp0_m, tmp2_m, tmp1_m, tmp3_m, tmp0_m, tmp1_m); \ + ST8x4_UB(tmp0_m, tmp1_m, pdst_m, stride); \ } #endif /* VP9_COMMON_MIPS_MSA_VP9_CONVOLVE_MSA_H_ */ diff --git a/vp9/common/vp9_rtcd_defs.pl b/vp9/common/vp9_rtcd_defs.pl index 638b83673..5cb1cd5af 100644 --- a/vp9/common/vp9_rtcd_defs.pl +++ b/vp9/common/vp9_rtcd_defs.pl @@ -286,19 +286,19 @@ specialize qw/vp9_filter_by_weight8x8 sse2/; # Sub Pixel Filters # add_proto qw/void vp9_convolve_copy/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"; -specialize qw/vp9_convolve_copy neon dspr2/, "$sse2_x86inc"; +specialize qw/vp9_convolve_copy neon dspr2 msa/, "$sse2_x86inc"; add_proto qw/void vp9_convolve_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"; -specialize qw/vp9_convolve_avg neon dspr2/, "$sse2_x86inc"; +specialize qw/vp9_convolve_avg neon dspr2 msa/, "$sse2_x86inc"; add_proto qw/void vp9_convolve8/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"; -specialize qw/vp9_convolve8 sse2 ssse3 neon dspr2/, "$avx2_ssse3"; +specialize qw/vp9_convolve8 sse2 ssse3 neon dspr2 msa/, "$avx2_ssse3"; add_proto qw/void vp9_convolve8_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"; -specialize qw/vp9_convolve8_horiz sse2 ssse3 neon dspr2/, "$avx2_ssse3"; +specialize qw/vp9_convolve8_horiz sse2 ssse3 neon dspr2 msa/, "$avx2_ssse3"; add_proto qw/void vp9_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"; -specialize qw/vp9_convolve8_vert sse2 ssse3 neon dspr2/, "$avx2_ssse3"; +specialize qw/vp9_convolve8_vert sse2 ssse3 neon dspr2 msa/, "$avx2_ssse3"; add_proto qw/void vp9_convolve8_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"; specialize qw/vp9_convolve8_avg sse2 ssse3 neon dspr2/; diff --git a/vp9/vp9_common.mk b/vp9/vp9_common.mk index 6b64f810c..fa7dcb2c7 100644 --- a/vp9/vp9_common.mk +++ b/vp9/vp9_common.mk @@ -132,6 +132,12 @@ VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_mblpf_vert_loopfilter_ds # common (msa) VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_macros_msa.h +VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve8_horiz_msa.c +VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve8_msa.c +VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve8_vert_msa.c +VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve_avg_msa.c +VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve_copy_msa.c +VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve_msa.h VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_idct8x8_msa.c VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_idct16x16_msa.c VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_idct32x32_msa.c