void x264_sub8x16_dct_dc_msa( int16_t pi_dct[8],
uint8_t *p_pix1, uint8_t *p_pix2 )
{
int32_t a0, a1, a2, a3, a4, a5, a6, a7;
int32_t b0, b1, b2, b3, b4, b5, b6, b7;
a0 = subtract_sum4x4_msa( &p_pix1[ 0 * FENC_STRIDE + 0], FENC_STRIDE,
&p_pix2[ 0 * FDEC_STRIDE + 0], FDEC_STRIDE );
a1 = subtract_sum4x4_msa( &p_pix1[ 0 * FENC_STRIDE + 4], FENC_STRIDE,
&p_pix2[ 0 * FDEC_STRIDE + 4], FDEC_STRIDE );
a2 = subtract_sum4x4_msa( &p_pix1[ 4 * FENC_STRIDE + 0], FENC_STRIDE,
&p_pix2[ 4 * FDEC_STRIDE + 0], FDEC_STRIDE );
a3 = subtract_sum4x4_msa( &p_pix1[ 4 * FENC_STRIDE + 4], FENC_STRIDE,
&p_pix2[ 4 * FDEC_STRIDE + 4], FDEC_STRIDE );
a4 = subtract_sum4x4_msa( &p_pix1[ 8 * FENC_STRIDE + 0], FENC_STRIDE,
&p_pix2[ 8 * FDEC_STRIDE + 0], FDEC_STRIDE );
a5 = subtract_sum4x4_msa( &p_pix1[ 8 * FENC_STRIDE + 4], FENC_STRIDE,
&p_pix2[ 8 * FDEC_STRIDE + 4], FDEC_STRIDE );
a6 = subtract_sum4x4_msa( &p_pix1[12 * FENC_STRIDE + 0], FENC_STRIDE,
&p_pix2[12 * FDEC_STRIDE + 0], FDEC_STRIDE );
a7 = subtract_sum4x4_msa( &p_pix1[12 * FENC_STRIDE + 4], FENC_STRIDE,
&p_pix2[12 * FDEC_STRIDE + 4], FDEC_STRIDE );
BUTTERFLY_8( a0, a2, a4, a6, a7, a5, a3, a1,
b0, b1, b2, b3, b7, b6, b5, b4 );
BUTTERFLY_8( b0, b2, b4, b6, b7, b5, b3, b1,
a0, a1, a2, a3, a7, a6, a5, a4 );
BUTTERFLY_8( a0, a2, a4, a6, a7, a5, a3, a1,
pi_dct[0], pi_dct[1], pi_dct[6], pi_dct[7],
pi_dct[5], pi_dct[4], pi_dct[3], pi_dct[2] );
}
static void fdct8x32_1d_column_load_butterfly(const int16_t *input,
int32_t src_stride,
int16_t *temp_buff) {
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
v8i16 step0, step1, step2, step3;
v8i16 in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1;
v8i16 step0_1, step1_1, step2_1, step3_1;
/* 1st and 2nd set */
LD_SH4(input, src_stride, in0, in1, in2, in3);
LD_SH4(input + (28 * src_stride), src_stride, in4, in5, in6, in7);
LD_SH4(input + (4 * src_stride), src_stride, in0_1, in1_1, in2_1, in3_1);
LD_SH4(input + (24 * src_stride), src_stride, in4_1, in5_1, in6_1, in7_1);
SLLI_4V(in0, in1, in2, in3, 2);
SLLI_4V(in4, in5, in6, in7, 2);
SLLI_4V(in0_1, in1_1, in2_1, in3_1, 2);
SLLI_4V(in4_1, in5_1, in6_1, in7_1, 2);
BUTTERFLY_8(in0, in1, in2, in3, in4, in5, in6, in7,
step0, step1, step2, step3, in4, in5, in6, in7);
BUTTERFLY_8(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1,
step0_1, step1_1, step2_1, step3_1, in4_1, in5_1, in6_1, in7_1);
ST_SH4(step0, step1, step2, step3, temp_buff, 8);
ST_SH4(in4, in5, in6, in7, temp_buff + (28 * 8), 8);
ST_SH4(step0_1, step1_1, step2_1, step3_1, temp_buff + (4 * 8), 8);
ST_SH4(in4_1, in5_1, in6_1, in7_1, temp_buff + (24 * 8), 8);
/* 3rd and 4th set */
LD_SH4(input + (8 * src_stride), src_stride, in0, in1, in2, in3);
LD_SH4(input + (20 * src_stride), src_stride, in4, in5, in6, in7);
LD_SH4(input + (12 * src_stride), src_stride, in0_1, in1_1, in2_1, in3_1);
LD_SH4(input + (16 * src_stride), src_stride, in4_1, in5_1, in6_1, in7_1);
SLLI_4V(in0, in1, in2, in3, 2);
SLLI_4V(in4, in5, in6, in7, 2);
SLLI_4V(in0_1, in1_1, in2_1, in3_1, 2);
SLLI_4V(in4_1, in5_1, in6_1, in7_1, 2);
BUTTERFLY_8(in0, in1, in2, in3, in4, in5, in6, in7,
step0, step1, step2, step3, in4, in5, in6, in7);
BUTTERFLY_8(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1,
step0_1, step1_1, step2_1, step3_1, in4_1, in5_1, in6_1, in7_1);
ST_SH4(step0, step1, step2, step3, temp_buff + (8 * 8), 8);
ST_SH4(in4, in5, in6, in7, temp_buff + (20 * 8), 8);
ST_SH4(step0_1, step1_1, step2_1, step3_1, temp_buff + (12 * 8), 8);
ST_SH4(in4_1, in5_1, in6_1, in7_1, temp_buff + (15 * 8) + 8, 8);
}
static void avc_idct8_addblk_msa( uint8_t *p_dst, int16_t *p_src,
int32_t i_dst_stride )
{
v8i16 src0, src1, src2, src3, src4, src5, src6, src7;
v8i16 vec0, vec1, vec2, vec3;
v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
v8i16 res0, res1, res2, res3, res4, res5, res6, res7;
v4i32 tmp0_r, tmp1_r, tmp2_r, tmp3_r, tmp4_r, tmp5_r, tmp6_r, tmp7_r;
v4i32 tmp0_l, tmp1_l, tmp2_l, tmp3_l, tmp4_l, tmp5_l, tmp6_l, tmp7_l;
v4i32 vec0_r, vec1_r, vec2_r, vec3_r, vec0_l, vec1_l, vec2_l, vec3_l;
v4i32 res0_r, res1_r, res2_r, res3_r, res4_r, res5_r, res6_r, res7_r;
v4i32 res0_l, res1_l, res2_l, res3_l, res4_l, res5_l, res6_l, res7_l;
v16i8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
v16i8 zeros = { 0 };
p_src[ 0 ] += 32;
LD_SH8( p_src, 8, src0, src1, src2, src3, src4, src5, src6, src7 );
vec0 = src0 + src4;
vec1 = src0 - src4;
vec2 = src2 >> 1;
vec2 = vec2 - src6;
vec3 = src6 >> 1;
vec3 = src2 + vec3;
BUTTERFLY_4( vec0, vec1, vec2, vec3, tmp0, tmp1, tmp2, tmp3 );
vec0 = src7 >> 1;
vec0 = src5 - vec0 - src3 - src7;
vec1 = src3 >> 1;
vec1 = src1 - vec1 + src7 - src3;
vec2 = src5 >> 1;
vec2 = vec2 - src1 + src7 + src5;
vec3 = src1 >> 1;
vec3 = vec3 + src3 + src5 + src1;
tmp4 = vec3 >> 2;
tmp4 += vec0;
tmp5 = vec2 >> 2;
tmp5 += vec1;
tmp6 = vec1 >> 2;
tmp6 -= vec2;
tmp7 = vec0 >> 2;
tmp7 = vec3 - tmp7;
BUTTERFLY_8( tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7,
res0, res1, res2, res3, res4, res5, res6, res7 );
TRANSPOSE8x8_SH_SH( res0, res1, res2, res3, res4, res5, res6, res7,
res0, res1, res2, res3, res4, res5, res6, res7 );
UNPCK_SH_SW( res0, tmp0_r, tmp0_l );
UNPCK_SH_SW( res1, tmp1_r, tmp1_l );
UNPCK_SH_SW( res2, tmp2_r, tmp2_l );
UNPCK_SH_SW( res3, tmp3_r, tmp3_l );
UNPCK_SH_SW( res4, tmp4_r, tmp4_l );
UNPCK_SH_SW( res5, tmp5_r, tmp5_l );
UNPCK_SH_SW( res6, tmp6_r, tmp6_l );
UNPCK_SH_SW( res7, tmp7_r, tmp7_l );
BUTTERFLY_4( tmp0_r, tmp0_l, tmp4_l, tmp4_r,
vec0_r, vec0_l, vec1_l, vec1_r );
vec2_r = tmp2_r >> 1;
vec2_l = tmp2_l >> 1;
vec2_r -= tmp6_r;
vec2_l -= tmp6_l;
vec3_r = tmp6_r >> 1;
vec3_l = tmp6_l >> 1;
vec3_r += tmp2_r;
vec3_l += tmp2_l;
BUTTERFLY_4( vec0_r, vec1_r, vec2_r, vec3_r,
tmp0_r, tmp2_r, tmp4_r, tmp6_r );
BUTTERFLY_4( vec0_l, vec1_l, vec2_l, vec3_l,
tmp0_l, tmp2_l, tmp4_l, tmp6_l );
vec0_r = tmp7_r >> 1;
vec0_l = tmp7_l >> 1;
vec0_r = tmp5_r - vec0_r - tmp3_r - tmp7_r;
vec0_l = tmp5_l - vec0_l - tmp3_l - tmp7_l;
vec1_r = tmp3_r >> 1;
vec1_l = tmp3_l >> 1;
vec1_r = tmp1_r - vec1_r + tmp7_r - tmp3_r;
vec1_l = tmp1_l - vec1_l + tmp7_l - tmp3_l;
vec2_r = tmp5_r >> 1;
vec2_l = tmp5_l >> 1;
vec2_r = vec2_r - tmp1_r + tmp7_r + tmp5_r;
vec2_l = vec2_l - tmp1_l + tmp7_l + tmp5_l;
vec3_r = tmp1_r >> 1;
vec3_l = tmp1_l >> 1;
vec3_r = vec3_r + tmp3_r + tmp5_r + tmp1_r;
vec3_l = vec3_l + tmp3_l + tmp5_l + tmp1_l;
tmp1_r = vec3_r >> 2;
tmp1_l = vec3_l >> 2;
tmp1_r += vec0_r;
tmp1_l += vec0_l;
tmp3_r = vec2_r >> 2;
tmp3_l = vec2_l >> 2;
tmp3_r += vec1_r;
tmp3_l += vec1_l;
tmp5_r = vec1_r >> 2;
tmp5_l = vec1_l >> 2;
tmp5_r -= vec2_r;
tmp5_l -= vec2_l;
tmp7_r = vec0_r >> 2;
tmp7_l = vec0_l >> 2;
tmp7_r = vec3_r - tmp7_r;
tmp7_l = vec3_l - tmp7_l;
BUTTERFLY_4( tmp0_r, tmp0_l, tmp7_l, tmp7_r,
res0_r, res0_l, res7_l, res7_r );
BUTTERFLY_4( tmp2_r, tmp2_l, tmp5_l, tmp5_r,
res1_r, res1_l, res6_l, res6_r );
BUTTERFLY_4( tmp4_r, tmp4_l, tmp3_l, tmp3_r,
res2_r, res2_l, res5_l, res5_r );
BUTTERFLY_4( tmp6_r, tmp6_l, tmp1_l, tmp1_r,
res3_r, res3_l, res4_l, res4_r );
SRA_4V( res0_r, res0_l, res1_r, res1_l, 6 );
SRA_4V( res2_r, res2_l, res3_r, res3_l, 6 );
SRA_4V( res4_r, res4_l, res5_r, res5_l, 6 );
SRA_4V( res6_r, res6_l, res7_r, res7_l, 6 );
PCKEV_H4_SH( res0_l, res0_r, res1_l, res1_r, res2_l, res2_r, res3_l, res3_r,
res0, res1, res2, res3 );
PCKEV_H4_SH( res4_l, res4_r, res5_l, res5_r, res6_l, res6_r, res7_l, res7_r,
res4, res5, res6, res7 );
LD_SB8( p_dst, i_dst_stride,
dst0, dst1, dst2, dst3,
dst4, dst5, dst6, dst7 );
ILVR_B4_SH( zeros, dst0, zeros, dst1, zeros, dst2, zeros, dst3,
tmp0, tmp1, tmp2, tmp3 );
ILVR_B4_SH( zeros, dst4, zeros, dst5, zeros, dst6, zeros, dst7,
tmp4, tmp5, tmp6, tmp7 );
ADD4( res0, tmp0, res1, tmp1, res2, tmp2, res3, tmp3,
res0, res1, res2, res3 );
ADD4( res4, tmp4, res5, tmp5, res6, tmp6, res7, tmp7,
res4, res5, res6, res7 );
CLIP_SH4_0_255( res0, res1, res2, res3 );
CLIP_SH4_0_255( res4, res5, res6, res7 );
PCKEV_B4_SB( res1, res0, res3, res2, res5, res4, res7, res6,
dst0, dst1, dst2, dst3 );
ST8x4_UB( dst0, dst1, p_dst, i_dst_stride );
p_dst += ( 4 * i_dst_stride );
ST8x4_UB( dst2, dst3, p_dst, i_dst_stride );
}
static void fdct8x32_1d_column_even_store(int16_t *input, int16_t *temp) {
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
v8i16 temp0, temp1;
/* fdct even */
LD_SH4(input, 8, in0, in1, in2, in3);
LD_SH4(input + 96, 8, in12, in13, in14, in15);
BUTTERFLY_8(in0, in1, in2, in3, in12, in13, in14, in15,
vec0, vec1, vec2, vec3, in12, in13, in14, in15);
LD_SH4(input + 32, 8, in4, in5, in6, in7);
LD_SH4(input + 64, 8, in8, in9, in10, in11);
BUTTERFLY_8(in4, in5, in6, in7, in8, in9, in10, in11,
vec4, vec5, vec6, vec7, in8, in9, in10, in11);
/* Stage 3 */
ADD4(vec0, vec7, vec1, vec6, vec2, vec5, vec3, vec4, in0, in1, in2, in3);
BUTTERFLY_4(in0, in1, in2, in3, temp0, in4, in1, in0);
DOTP_CONST_PAIR(temp0, in4, cospi_16_64, cospi_16_64, temp1, temp0);
FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
ST_SH(temp0, temp);
ST_SH(temp1, temp + 512);
DOTP_CONST_PAIR(in0, in1, cospi_24_64, cospi_8_64, temp1, temp0);
FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
ST_SH(temp0, temp + 256);
ST_SH(temp1, temp + 768);
SUB4(vec0, vec7, vec1, vec6, vec2, vec5, vec3, vec4, vec7, vec6, vec5, vec4);
DOTP_CONST_PAIR(vec6, vec5, cospi_16_64, cospi_16_64, vec5, vec6);
ADD2(vec4, vec5, vec7, vec6, vec0, vec1);
DOTP_CONST_PAIR(vec1, vec0, cospi_28_64, cospi_4_64, temp1, temp0);
FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
ST_SH(temp0, temp + 128);
ST_SH(temp1, temp + 896);
SUB2(vec4, vec5, vec7, vec6, vec4, vec7);
DOTP_CONST_PAIR(vec7, vec4, cospi_12_64, cospi_20_64, temp1, temp0);
FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
ST_SH(temp0, temp + 640);
ST_SH(temp1, temp + 384);
DOTP_CONST_PAIR(in13, in10, cospi_16_64, cospi_16_64, vec2, vec5);
DOTP_CONST_PAIR(in12, in11, cospi_16_64, cospi_16_64, vec3, vec4);
ADD4(in8, vec3, in9, vec2, in14, vec5, in15, vec4, in0, vec1, vec6, in2);
DOTP_CONST_PAIR(vec6, vec1, cospi_24_64, cospi_8_64, in1, in3);
ADD2(in0, in1, in2, in3, vec0, vec7);
DOTP_CONST_PAIR(vec7, vec0, cospi_30_64, cospi_2_64, temp1, temp0);
FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
ST_SH(temp0, temp + 64);
ST_SH(temp1, temp + 960);
SUB2(in0, in1, in2, in3, in0, in2);
DOTP_CONST_PAIR(in2, in0, cospi_14_64, cospi_18_64, temp1, temp0);
FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
ST_SH(temp0, temp + 576);
ST_SH(temp1, temp + 448);
SUB2(in9, vec2, in14, vec5, vec2, vec5);
DOTP_CONST_PAIR((-vec2), vec5, cospi_24_64, cospi_8_64, in2, in1);
SUB4(in8, vec3, in15, vec4, in3, in2, in0, in1, in3, in0, vec2, vec5);
DOTP_CONST_PAIR(vec5, vec2, cospi_22_64, cospi_10_64, temp1, temp0);
FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
ST_SH(temp0, temp + 320);
ST_SH(temp1, temp + 704);
ADD2(in3, in2, in0, in1, vec3, vec4);
DOTP_CONST_PAIR(vec4, vec3, cospi_6_64, cospi_26_64, temp0, temp1);
FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
ST_SH(temp0, temp + 192);
ST_SH(temp1, temp + 832);
}