void fdct16x8_1d_row(int16_t *input, int16_t *output) {
v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
LD_SH8(input, 16, in0, in1, in2, in3, in4, in5, in6, in7);
LD_SH8((input + 8), 16, in8, in9, in10, in11, in12, in13, in14, in15);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in8, in9, in10, in11, in12, in13, in14, in15, in8, in9,
in10, in11, in12, in13, in14, in15);
ADD4(in0, 1, in1, 1, in2, 1, in3, 1, in0, in1, in2, in3);
ADD4(in4, 1, in5, 1, in6, 1, in7, 1, in4, in5, in6, in7);
ADD4(in8, 1, in9, 1, in10, 1, in11, 1, in8, in9, in10, in11);
ADD4(in12, 1, in13, 1, in14, 1, in15, 1, in12, in13, in14, in15);
SRA_4V(in0, in1, in2, in3, 2);
SRA_4V(in4, in5, in6, in7, 2);
SRA_4V(in8, in9, in10, in11, 2);
SRA_4V(in12, in13, in14, in15, 2);
BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7, in8, in9, in10, in11,
in12, in13, in14, in15, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6,
tmp7, in8, in9, in10, in11, in12, in13, in14, in15);
ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, input, 16);
FDCT8x16_EVEN(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp0, tmp1,
tmp2, tmp3, tmp4, tmp5, tmp6, tmp7);
LD_SH8(input, 16, in8, in9, in10, in11, in12, in13, in14, in15);
FDCT8x16_ODD(in8, in9, in10, in11, in12, in13, in14, in15, in0, in1, in2, in3,
in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3, tmp0, in0,
tmp1, in1, tmp2, in2, tmp3, in3);
ST_SH8(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3, output, 16);
TRANSPOSE8x8_SH_SH(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7, tmp4, in4,
tmp5, in5, tmp6, in6, tmp7, in7);
ST_SH8(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7, output + 8, 16);
}
static void fdct8x32_1d_row_load_butterfly(int16_t *temp_buff,
int16_t *output) {
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
v8i16 step0, step1, step2, step3, step4, step5, step6, step7;
LD_SH8(temp_buff, 32, in0, in1, in2, in3, in4, in5, in6, in7);
LD_SH8(temp_buff + 24, 32, in8, in9, in10, in11, in12, in13, in14, in15);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in8, in9, in10, in11, in12, in13, in14, in15,
in8, in9, in10, in11, in12, in13, in14, in15);
BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7,
in8, in9, in10, in11, in12, in13, in14, in15,
step0, step1, step2, step3, step4, step5, step6, step7,
in8, in9, in10, in11, in12, in13, in14, in15);
ST_SH8(step0, step1, step2, step3, step4, step5, step6, step7, output, 8);
ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, (output + 24 * 8), 8);
/* 2nd set */
LD_SH8(temp_buff + 8, 32, in0, in1, in2, in3, in4, in5, in6, in7);
LD_SH8(temp_buff + 16, 32, in8, in9, in10, in11, in12, in13, in14, in15);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in8, in9, in10, in11, in12, in13, in14, in15,
in8, in9, in10, in11, in12, in13, in14, in15);
BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7,
in8, in9, in10, in11, in12, in13, in14, in15,
step0, step1, step2, step3, step4, step5, step6, step7,
in8, in9, in10, in11, in12, in13, in14, in15);
ST_SH8(step0, step1, step2, step3, step4, step5, step6, step7,
(output + 8 * 8), 8);
ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, (output + 16 * 8), 8);
}
static void fadst16_transpose_postproc_msa(int16_t *input, int16_t *out) {
v8i16 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15;
v8i16 l0, l1, l2, l3, l4, l5, l6, l7, l8, l9, l10, l11, l12, l13, l14, l15;
/* load input data */
LD_SH8(input, 16, l0, l1, l2, l3, l4, l5, l6, l7);
TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7, r0, r1, r2, r3, r4, r5, r6,
r7);
FDCT_POSTPROC_2V_NEG_H(r0, r1);
FDCT_POSTPROC_2V_NEG_H(r2, r3);
FDCT_POSTPROC_2V_NEG_H(r4, r5);
FDCT_POSTPROC_2V_NEG_H(r6, r7);
ST_SH8(r0, r1, r2, r3, r4, r5, r6, r7, out, 8);
out += 64;
LD_SH8(input + 8, 16, l8, l9, l10, l11, l12, l13, l14, l15);
TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15, r8, r9, r10, r11,
r12, r13, r14, r15);
FDCT_POSTPROC_2V_NEG_H(r8, r9);
FDCT_POSTPROC_2V_NEG_H(r10, r11);
FDCT_POSTPROC_2V_NEG_H(r12, r13);
FDCT_POSTPROC_2V_NEG_H(r14, r15);
ST_SH8(r8, r9, r10, r11, r12, r13, r14, r15, out, 8);
out += 64;
/* load input data */
input += 128;
LD_SH8(input, 16, l0, l1, l2, l3, l4, l5, l6, l7);
TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7, r0, r1, r2, r3, r4, r5, r6,
r7);
FDCT_POSTPROC_2V_NEG_H(r0, r1);
FDCT_POSTPROC_2V_NEG_H(r2, r3);
FDCT_POSTPROC_2V_NEG_H(r4, r5);
FDCT_POSTPROC_2V_NEG_H(r6, r7);
ST_SH8(r0, r1, r2, r3, r4, r5, r6, r7, out, 8);
out += 64;
LD_SH8(input + 8, 16, l8, l9, l10, l11, l12, l13, l14, l15);
TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15, r8, r9, r10, r11,
r12, r13, r14, r15);
FDCT_POSTPROC_2V_NEG_H(r8, r9);
FDCT_POSTPROC_2V_NEG_H(r10, r11);
FDCT_POSTPROC_2V_NEG_H(r12, r13);
FDCT_POSTPROC_2V_NEG_H(r14, r15);
ST_SH8(r8, r9, r10, r11, r12, r13, r14, r15, out, 8);
}
static void postproc_fdct16x8_1d_row(int16_t *intermediate, int16_t *output) {
int16_t *temp = intermediate;
int16_t *out = output;
v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
v8i16 in0, in1, in2, in3, in4, in5, in6, in7, in8, in9, in10, in11;
v8i16 in12, in13, in14, in15;
LD_SH8(temp, 16, in0, in1, in2, in3, in4, in5, in6, in7);
temp = intermediate + 8;
LD_SH8(temp, 16, in8, in9, in10, in11, in12, in13, in14, in15);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in8, in9, in10, in11, in12, in13, in14, in15, in8, in9,
in10, in11, in12, in13, in14, in15);
FDCT_POSTPROC_2V_NEG_H(in0, in1);
FDCT_POSTPROC_2V_NEG_H(in2, in3);
FDCT_POSTPROC_2V_NEG_H(in4, in5);
FDCT_POSTPROC_2V_NEG_H(in6, in7);
FDCT_POSTPROC_2V_NEG_H(in8, in9);
FDCT_POSTPROC_2V_NEG_H(in10, in11);
FDCT_POSTPROC_2V_NEG_H(in12, in13);
FDCT_POSTPROC_2V_NEG_H(in14, in15);
BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7, in8, in9, in10, in11,
in12, in13, in14, in15, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6,
tmp7, in8, in9, in10, in11, in12, in13, in14, in15);
temp = intermediate;
ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, temp, 16);
FDCT8x16_EVEN(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp0, tmp1,
tmp2, tmp3, tmp4, tmp5, tmp6, tmp7);
temp = intermediate;
LD_SH8(temp, 16, in8, in9, in10, in11, in12, in13, in14, in15);
FDCT8x16_ODD(in8, in9, in10, in11, in12, in13, in14, in15, in0, in1, in2, in3,
in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3, tmp0, in0,
tmp1, in1, tmp2, in2, tmp3, in3);
ST_SH8(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3, out, 16);
TRANSPOSE8x8_SH_SH(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7, tmp4, in4,
tmp5, in5, tmp6, in6, tmp7, in7);
out = output + 8;
ST_SH8(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7, out, 16);
}
static void hevc_idct_transpose_8x32_to_32x8(int16_t *tmp_buf, int16_t *coeffs)
{
uint8_t i;
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
for (i = 0; i < 4; i++) {
LD_SH8(tmp_buf + i * 8 * 8, 8, in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
ST_SH8(in0, in1, in2, in3, in4, in5, in6, in7, coeffs + i * 8, 32);
}
}
void vpx_fdct8x8_1_msa(const int16_t *input, int16_t *out, int32_t stride) {
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
v4i32 vec_w;
LD_SH8(input, stride, in0, in1, in2, in3, in4, in5, in6, in7);
ADD4(in0, in1, in2, in3, in4, in5, in6, in7, in0, in2, in4, in6);
ADD2(in0, in2, in4, in6, in0, in4);
vec_w = __msa_hadd_s_w(in0, in0);
vec_w += __msa_hadd_s_w(in4, in4);
out[0] = HADD_SW_S32(vec_w);
out[1] = 0;
}
void vp9_fht8x8_msa(const int16_t *input, int16_t *output, int32_t stride,
int32_t tx_type) {
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
LD_SH8(input, stride, in0, in1, in2, in3, in4, in5, in6, in7);
SLLI_4V(in0, in1, in2, in3, 2);
SLLI_4V(in4, in5, in6, in7, 2);
switch (tx_type) {
case DCT_DCT:
VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
break;
case ADST_DCT:
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
break;
case DCT_ADST:
VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
break;
case ADST_ADST:
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
break;
default:
assert(0);
break;
}
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_SRLI_AVE_S_4V_H(in0, in1, in2, in3, in4, in5, in6, in7);
ST_SH8(in0, in1, in2, in3, in4, in5, in6, in7, output, 8);
}
static void hevc_idct_8x8_msa(int16_t *coeffs)
{
const int16_t *filter = >8x8_cnst[0];
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
LD_SH8(coeffs, 8, in0, in1, in2, in3, in4, in5, in6, in7);
HEVC_IDCT8x8_COL(in0, in1, in2, in3, in4, in5, in6, in7, 7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
HEVC_IDCT8x8_COL(in0, in1, in2, in3, in4, in5, in6, in7, 12);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
ST_SH8(in0, in1, in2, in3, in4, in5, in6, in7, coeffs, 8);
}
void vpx_fdct8x8_msa(const int16_t *input, int16_t *output,
int32_t src_stride) {
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
LD_SH8(input, src_stride, in0, in1, in2, in3, in4, in5, in6, in7);
SLLI_4V(in0, in1, in2, in3, 2);
SLLI_4V(in4, in5, in6, in7, 2);
VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4,
in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
in4, in5, in6, in7);
VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4,
in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
in4, in5, in6, in7);
SRLI_AVE_S_4V_H(in0, in1, in2, in3, in4, in5, in6, in7);
ST_SH8(in0, in1, in2, in3, in4, in5, in6, in7, output, 8);
}
static int32_t avc_coeff_last64_msa( int16_t *p_src )
{
uint32_t u_res;
v8i16 src0, src1, src2, src3, src4, src5, src6, src7;
v8i16 tmp_h0, tmp_h1, tmp_h2, tmp_h3, tmp_h4, tmp_h5, tmp_h6, tmp_h7;
v16u8 tmp0, tmp1, tmp2, tmp3;
v8u16 vec0, vec1, vec2, vec3;
v4i32 out0;
v16u8 mask = { 1, 2, 4, 8, 16, 32, 64, 128, 1, 2, 4, 8, 16, 32, 64, 128 };
LD_SH8( p_src, 8, src0, src1, src2, src3, src4, src5, src6, src7 );
tmp_h0 = __msa_ceqi_h( src0, 0 );
tmp_h1 = __msa_ceqi_h( src1, 0 );
tmp_h2 = __msa_ceqi_h( src2, 0 );
tmp_h3 = __msa_ceqi_h( src3, 0 );
tmp_h4 = __msa_ceqi_h( src4, 0 );
tmp_h5 = __msa_ceqi_h( src5, 0 );
tmp_h6 = __msa_ceqi_h( src6, 0 );
tmp_h7 = __msa_ceqi_h( src7, 0 );
PCKEV_B4_UB( tmp_h1, tmp_h0, tmp_h3, tmp_h2, tmp_h5, tmp_h4, tmp_h7, tmp_h6,
tmp0, tmp1, tmp2, tmp3 );
tmp0 = tmp0 & mask;
tmp1 = tmp1 & mask;
tmp2 = tmp2 & mask;
tmp3 = tmp3 & mask;
HADD_UB4_UH( tmp0, tmp1, tmp2, tmp3, vec0, vec1, vec2, vec3 );
PCKEV_B2_UB( vec1, vec0, vec3, vec2, tmp0, tmp1 );
HADD_UB2_UH( tmp0, tmp1, vec0, vec1 );
tmp0 = ( v16u8 ) __msa_pckev_b( ( v16i8 ) vec1, ( v16i8 ) vec0 );
vec0 = __msa_hadd_u_h( tmp0, tmp0 );
tmp0 = ( v16u8 ) __msa_pckev_b( ( v16i8 ) vec0, ( v16i8 ) vec0 );
out0 = ( v4i32 ) __msa_nloc_d( ( v2i64 ) tmp0 );
u_res = __msa_copy_u_w( out0, 0 );
return ( 63 - u_res );
}
static void hevc_addblk_8x8_msa(int16_t *coeffs, uint8_t *dst, int32_t stride)
{
uint8_t *temp_dst = dst;
uint64_t dst0, dst1, dst2, dst3;
v2i64 dst_vec0 = { 0 };
v2i64 dst_vec1 = { 0 };
v8i16 dst_r0, dst_l0, dst_r1, dst_l1;
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
v16u8 zeros = { 0 };
LD_SH8(coeffs, 8, in0, in1, in2, in3, in4, in5, in6, in7);
LD4(temp_dst, stride, dst0, dst1, dst2, dst3);
temp_dst += (4 * stride);
INSERT_D2_SD(dst0, dst1, dst_vec0);
INSERT_D2_SD(dst2, dst3, dst_vec1);
ILVRL_B2_SH(zeros, dst_vec0, dst_r0, dst_l0);
ILVRL_B2_SH(zeros, dst_vec1, dst_r1, dst_l1);
ADD4(dst_r0, in0, dst_l0, in1, dst_r1, in2, dst_l1, in3,
dst_r0, dst_l0, dst_r1, dst_l1);
CLIP_SH4_0_255(dst_r0, dst_l0, dst_r1, dst_l1);
PCKEV_B2_SH(dst_l0, dst_r0, dst_l1, dst_r1, dst_r0, dst_r1);
ST8x4_UB(dst_r0, dst_r1, dst, stride);
dst += (4 * stride);
LD4(temp_dst, stride, dst0, dst1, dst2, dst3);
INSERT_D2_SD(dst0, dst1, dst_vec0);
INSERT_D2_SD(dst2, dst3, dst_vec1);
UNPCK_UB_SH(dst_vec0, dst_r0, dst_l0);
UNPCK_UB_SH(dst_vec1, dst_r1, dst_l1);
ADD4(dst_r0, in4, dst_l0, in5, dst_r1, in6, dst_l1, in7,
dst_r0, dst_l0, dst_r1, dst_l1);
CLIP_SH4_0_255(dst_r0, dst_l0, dst_r1, dst_l1);
PCKEV_B2_SH(dst_l0, dst_r0, dst_l1, dst_r1, dst_r0, dst_r1);
ST8x4_UB(dst_r0, dst_r1, dst, stride);
}
static void fdct8x32_1d_row_even_4x(int16_t *input, int16_t *interm_ptr,
int16_t *out) {
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;
v4i32 vec0_l, vec1_l, vec2_l, vec3_l, vec4_l, vec5_l, vec6_l, vec7_l;
v4i32 vec0_r, vec1_r, vec2_r, vec3_r, vec4_r, vec5_r, vec6_r, vec7_r;
v4i32 tmp0_w, tmp1_w, tmp2_w, tmp3_w;
/* fdct32 even */
/* stage 2 */
LD_SH8(input, 8, in0, in1, in2, in3, in4, in5, in6, in7);
LD_SH8(input + 64, 8, in8, in9, in10, in11, in12, in13, in14, in15);
BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7,
in8, in9, in10, in11, in12, in13, in14, in15,
vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7,
in8, in9, in10, in11, in12, in13, in14, in15);
ST_SH8(vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, interm_ptr, 8);
ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, interm_ptr + 64, 8);
/* Stage 3 */
UNPCK_SH_SW(vec0, vec0_l, vec0_r);
UNPCK_SH_SW(vec1, vec1_l, vec1_r);
UNPCK_SH_SW(vec2, vec2_l, vec2_r);
UNPCK_SH_SW(vec3, vec3_l, vec3_r);
UNPCK_SH_SW(vec4, vec4_l, vec4_r);
UNPCK_SH_SW(vec5, vec5_l, vec5_r);
UNPCK_SH_SW(vec6, vec6_l, vec6_r);
UNPCK_SH_SW(vec7, vec7_l, vec7_r);
ADD4(vec0_r, vec7_r, vec1_r, vec6_r, vec2_r, vec5_r, vec3_r, vec4_r,
tmp0_w, tmp1_w, tmp2_w, tmp3_w);
BUTTERFLY_4(tmp0_w, tmp1_w, tmp2_w, tmp3_w, vec4_r, vec6_r, vec7_r, vec5_r);
ADD4(vec0_l, vec7_l, vec1_l, vec6_l, vec2_l, vec5_l, vec3_l, vec4_l,
vec0_r, vec1_r, vec2_r, vec3_r);
tmp3_w = vec0_r + vec3_r;
vec0_r = vec0_r - vec3_r;
vec3_r = vec1_r + vec2_r;
vec1_r = vec1_r - vec2_r;
DOTP_CONST_PAIR_W(vec4_r, vec6_r, tmp3_w, vec3_r, cospi_16_64,
cospi_16_64, vec4_r, tmp3_w, vec6_r, vec3_r);
FDCT32_POSTPROC_NEG_W(vec4_r);
FDCT32_POSTPROC_NEG_W(tmp3_w);
FDCT32_POSTPROC_NEG_W(vec6_r);
FDCT32_POSTPROC_NEG_W(vec3_r);
PCKEV_H2_SH(vec4_r, tmp3_w, vec6_r, vec3_r, vec4, vec5);
ST_SH2(vec5, vec4, out, 8);
DOTP_CONST_PAIR_W(vec5_r, vec7_r, vec0_r, vec1_r, cospi_24_64,
cospi_8_64, vec4_r, tmp3_w, vec6_r, vec3_r);
FDCT32_POSTPROC_NEG_W(vec4_r);
FDCT32_POSTPROC_NEG_W(tmp3_w);
FDCT32_POSTPROC_NEG_W(vec6_r);
FDCT32_POSTPROC_NEG_W(vec3_r);
PCKEV_H2_SH(vec4_r, tmp3_w, vec6_r, vec3_r, vec4, vec5);
ST_SH2(vec5, vec4, out + 16, 8);
LD_SH8(interm_ptr, 8, vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7);
SUB4(vec3, vec4, vec2, vec5, vec1, vec6, vec0, vec7, vec4, vec5, vec6, vec7);
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, in5, in4);
FDCT_POSTPROC_2V_NEG_H(in4, in5);
ST_SH(in4, out + 32);
ST_SH(in5, out + 56);
SUB2(vec4, vec5, vec7, vec6, vec4, vec7);
DOTP_CONST_PAIR(vec7, vec4, cospi_12_64, cospi_20_64, in5, in4);
FDCT_POSTPROC_2V_NEG_H(in4, in5);
ST_SH(in4, out + 40);
ST_SH(in5, out + 48);
LD_SH8(interm_ptr + 64, 8, in8, in9, in10, in11, in12, in13, in14, in15);
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, in5, in4);
FDCT_POSTPROC_2V_NEG_H(in4, in5);
ST_SH(in4, out + 64);
ST_SH(in5, out + 120);
SUB2(in0, in1, in2, in3, in0, in2);
DOTP_CONST_PAIR(in2, in0, cospi_14_64, cospi_18_64, in5, in4);
FDCT_POSTPROC_2V_NEG_H(in4, in5);
ST_SH(in4, out + 72);
ST_SH(in5, out + 112);
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, in5, in4);
FDCT_POSTPROC_2V_NEG_H(in4, in5);
ST_SH(in4, out + 80);
ST_SH(in5, out + 104);
ADD2(in3, in2, in0, in1, vec3, vec4);
DOTP_CONST_PAIR(vec4, vec3, cospi_6_64, cospi_26_64, in4, in5);
FDCT_POSTPROC_2V_NEG_H(in4, in5);
ST_SH(in4, out + 96);
ST_SH(in5, out + 88);
}
static void hevc_idct_16x16_msa(int16_t *coeffs)
{
int16_t i, j, k;
int16_t buf[256];
int16_t *buf_ptr = &buf[0];
int16_t *src = coeffs;
const int16_t *filter = >16x16_cnst[0];
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 src0_r, src1_r, src2_r, src3_r, src4_r, src5_r, src6_r, src7_r;
v8i16 src0_l, src1_l, src2_l, src3_l, src4_l, src5_l, src6_l, src7_l;
for (i = 2; i--;) {
LD_SH16(src, 16, in0, in1, in2, in3, in4, in5, in6, in7,
in8, in9, in10, in11, in12, in13, in14, in15);
ILVR_H4_SH(in4, in0, in12, in8, in6, in2, in14, in10,
src0_r, src1_r, src2_r, src3_r);
ILVR_H4_SH(in5, in1, in13, in9, in3, in7, in11, in15,
src4_r, src5_r, src6_r, src7_r);
ILVL_H4_SH(in4, in0, in12, in8, in6, in2, in14, in10,
src0_l, src1_l, src2_l, src3_l);
ILVL_H4_SH(in5, in1, in13, in9, in3, in7, in11, in15,
src4_l, src5_l, src6_l, src7_l);
HEVC_IDCT16x16_COL(src0_r, src1_r, src2_r, src3_r, src4_r, src5_r,
src6_r, src7_r, src0_l, src1_l, src2_l, src3_l,
src4_l, src5_l, src6_l, src7_l, 7);
src += 8;
buf_ptr = (&buf[0] + 8);
filter = >16x16_cnst[0];
}
src = &buf[0];
buf_ptr = coeffs;
filter = >16x16_cnst[0];
for (i = 2; i--;) {
LD_SH16(src, 8, in0, in8, in1, in9, in2, in10, in3, in11,
in4, in12, in5, in13, in6, in14, in7, in15);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in8, in9, in10, in11, in12, in13, in14, in15,
in8, in9, in10, in11, in12, in13, in14, in15);
ILVR_H4_SH(in4, in0, in12, in8, in6, in2, in14, in10,
src0_r, src1_r, src2_r, src3_r);
ILVR_H4_SH(in5, in1, in13, in9, in3, in7, in11, in15,
src4_r, src5_r, src6_r, src7_r);
ILVL_H4_SH(in4, in0, in12, in8, in6, in2, in14, in10,
src0_l, src1_l, src2_l, src3_l);
ILVL_H4_SH(in5, in1, in13, in9, in3, in7, in11, in15,
src4_l, src5_l, src6_l, src7_l);
HEVC_IDCT16x16_COL(src0_r, src1_r, src2_r, src3_r, src4_r, src5_r,
src6_r, src7_r, src0_l, src1_l, src2_l, src3_l,
src4_l, src5_l, src6_l, src7_l, 12);
src += 128;
buf_ptr = coeffs + 8;
filter = >16x16_cnst[0];
}
LD_SH8(coeffs, 16, in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7);
ST_SH8(vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, coeffs, 16);
LD_SH8((coeffs + 8), 16, in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7);
LD_SH8((coeffs + 128), 16, in8, in9, in10, in11, in12, in13, in14, in15);
ST_SH8(vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, (coeffs + 128), 16);
TRANSPOSE8x8_SH_SH(in8, in9, in10, in11, in12, in13, in14, in15,
vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7);
ST_SH8(vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, (coeffs + 8), 16);
LD_SH8((coeffs + 136), 16, in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7);
ST_SH8(vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, (coeffs + 136), 16);
}
static void hevc_idct_8x32_column_msa(int16_t *coeffs, uint8_t buf_pitch,
uint8_t round)
{
uint8_t i;
const int16_t *filter_ptr0 = >32x32_cnst0[0];
const int16_t *filter_ptr1 = >32x32_cnst1[0];
const int16_t *filter_ptr2 = >32x32_cnst2[0];
const int16_t *filter_ptr3 = >8x8_cnst[0];
int16_t *src0 = (coeffs + buf_pitch);
int16_t *src1 = (coeffs + 2 * buf_pitch);
int16_t *src2 = (coeffs + 4 * buf_pitch);
int16_t *src3 = (coeffs);
int32_t cnst0, cnst1;
int32_t tmp_buf[8 * 32 + 15];
int32_t *tmp_buf_ptr = tmp_buf + 15;
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
v8i16 src0_r, src1_r, src2_r, src3_r, src4_r, src5_r, src6_r, src7_r;
v8i16 src0_l, src1_l, src2_l, src3_l, src4_l, src5_l, src6_l, src7_l;
v8i16 filt0, filter0, filter1, filter2, filter3;
v4i32 sum0_r, sum0_l, sum1_r, sum1_l, tmp0_r, tmp0_l, tmp1_r, tmp1_l;
/* Align pointer to 64 byte boundary */
tmp_buf_ptr = (int32_t *)(((uintptr_t) tmp_buf_ptr) & ~(uintptr_t) 63);
/* process coeff 4, 12, 20, 28 */
LD_SH4(src2, 8 * buf_pitch, in0, in1, in2, in3);
ILVR_H2_SH(in1, in0, in3, in2, src0_r, src1_r);
ILVL_H2_SH(in1, in0, in3, in2, src0_l, src1_l);
LD_SH2(src3, 16 * buf_pitch, in4, in6);
LD_SH2((src3 + 8 * buf_pitch), 16 * buf_pitch, in5, in7);
ILVR_H2_SH(in6, in4, in7, in5, src2_r, src3_r);
ILVL_H2_SH(in6, in4, in7, in5, src2_l, src3_l);
/* loop for all columns of constants */
for (i = 0; i < 2; i++) {
/* processing single column of constants */
cnst0 = LW(filter_ptr2);
cnst1 = LW(filter_ptr2 + 2);
filter0 = (v8i16) __msa_fill_w(cnst0);
filter1 = (v8i16) __msa_fill_w(cnst1);
DOTP_SH2_SW(src0_r, src0_l, filter0, filter0, sum0_r, sum0_l);
DPADD_SH2_SW(src1_r, src1_l, filter1, filter1, sum0_r, sum0_l);
ST_SW2(sum0_r, sum0_l, (tmp_buf_ptr + 2 * i * 8), 4);
/* processing single column of constants */
cnst0 = LW(filter_ptr2 + 4);
cnst1 = LW(filter_ptr2 + 6);
filter0 = (v8i16) __msa_fill_w(cnst0);
filter1 = (v8i16) __msa_fill_w(cnst1);
DOTP_SH2_SW(src0_r, src0_l, filter0, filter0, sum0_r, sum0_l);
DPADD_SH2_SW(src1_r, src1_l, filter1, filter1, sum0_r, sum0_l);
ST_SW2(sum0_r, sum0_l, (tmp_buf_ptr + (2 * i + 1) * 8), 4);
filter_ptr2 += 8;
}
/* process coeff 0, 8, 16, 24 */
/* loop for all columns of constants */
for (i = 0; i < 2; i++) {
/* processing first column of filter constants */
cnst0 = LW(filter_ptr3);
cnst1 = LW(filter_ptr3 + 2);
filter0 = (v8i16) __msa_fill_w(cnst0);
filter1 = (v8i16) __msa_fill_w(cnst1);
DOTP_SH4_SW(src2_r, src2_l, src3_r, src3_l, filter0, filter0, filter1,
filter1, sum0_r, sum0_l, tmp1_r, tmp1_l);
sum1_r = sum0_r - tmp1_r;
sum1_l = sum0_l - tmp1_l;
sum0_r = sum0_r + tmp1_r;
sum0_l = sum0_l + tmp1_l;
HEVC_EVEN16_CALC(tmp_buf_ptr, sum0_r, sum0_l, i, (7 - i));
HEVC_EVEN16_CALC(tmp_buf_ptr, sum1_r, sum1_l, (3 - i), (4 + i));
filter_ptr3 += 8;
}
/* process coeff 2 6 10 14 18 22 26 30 */
LD_SH8(src1, 4 * buf_pitch, in0, in1, in2, in3, in4, in5, in6, in7);
ILVR_H4_SH(in1, in0, in3, in2, in5, in4, in7, in6,
src0_r, src1_r, src2_r, src3_r);
ILVL_H4_SH(in1, in0, in3, in2, in5, in4, in7, in6,
src0_l, src1_l, src2_l, src3_l);
/* loop for all columns of constants */
for (i = 0; i < 8; i++) {
/* processing single column of constants */
filt0 = LD_SH(filter_ptr1);
SPLATI_W4_SH(filt0, filter0, filter1, filter2, filter3);
DOTP_SH2_SW(src0_r, src0_l, filter0, filter0, sum0_r, sum0_l);
DPADD_SH4_SW(src1_r, src1_l, src2_r, src2_l, filter1, filter1, filter2,
filter2, sum0_r, sum0_l, sum0_r, sum0_l);
DPADD_SH2_SW(src3_r, src3_l, filter3, filter3, sum0_r, sum0_l);
LD_SW2(tmp_buf_ptr + i * 8, 4, tmp0_r, tmp0_l);
tmp1_r = tmp0_r;
tmp1_l = tmp0_l;
tmp0_r += sum0_r;
tmp0_l += sum0_l;
ST_SW2(tmp0_r, tmp0_l, (tmp_buf_ptr + i * 8), 4);
tmp1_r -= sum0_r;
tmp1_l -= sum0_l;
ST_SW2(tmp1_r, tmp1_l, (tmp_buf_ptr + (15 - i) * 8), 4);
filter_ptr1 += 8;
}
/* process coeff 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 */
LD_SH8(src0, 2 * buf_pitch, in0, in1, in2, in3, in4, in5, in6, in7);
src0 += 16 * buf_pitch;
ILVR_H4_SH(in1, in0, in3, in2, in5, in4, in7, in6,
src0_r, src1_r, src2_r, src3_r);
ILVL_H4_SH(in1, in0, in3, in2, in5, in4, in7, in6,
src0_l, src1_l, src2_l, src3_l);
LD_SH8(src0, 2 * buf_pitch, in0, in1, in2, in3, in4, in5, in6, in7);
ILVR_H4_SH(in1, in0, in3, in2, in5, in4, in7, in6,
src4_r, src5_r, src6_r, src7_r);
ILVL_H4_SH(in1, in0, in3, in2, in5, in4, in7, in6,
src4_l, src5_l, src6_l, src7_l);
/* loop for all columns of filter constants */
for (i = 0; i < 16; i++) {
/* processing single column of constants */
filt0 = LD_SH(filter_ptr0);
SPLATI_W4_SH(filt0, filter0, filter1, filter2, filter3);
DOTP_SH2_SW(src0_r, src0_l, filter0, filter0, sum0_r, sum0_l);
DPADD_SH4_SW(src1_r, src1_l, src2_r, src2_l, filter1, filter1, filter2,
filter2, sum0_r, sum0_l, sum0_r, sum0_l);
DPADD_SH2_SW(src3_r, src3_l, filter3, filter3, sum0_r, sum0_l);
tmp1_r = sum0_r;
tmp1_l = sum0_l;
filt0 = LD_SH(filter_ptr0 + 8);
SPLATI_W4_SH(filt0, filter0, filter1, filter2, filter3);
DOTP_SH2_SW(src4_r, src4_l, filter0, filter0, sum0_r, sum0_l);
DPADD_SH4_SW(src5_r, src5_l, src6_r, src6_l, filter1, filter1, filter2,
filter2, sum0_r, sum0_l, sum0_r, sum0_l);
DPADD_SH2_SW(src7_r, src7_l, filter3, filter3, sum0_r, sum0_l);
sum0_r += tmp1_r;
sum0_l += tmp1_l;
LD_SW2(tmp_buf_ptr + i * 8, 4, tmp0_r, tmp0_l);
tmp1_r = tmp0_r;
tmp1_l = tmp0_l;
tmp0_r += sum0_r;
tmp0_l += sum0_l;
sum1_r = __msa_fill_w(round);
SRAR_W2_SW(tmp0_r, tmp0_l, sum1_r);
SAT_SW2_SW(tmp0_r, tmp0_l, 15);
in0 = __msa_pckev_h((v8i16) tmp0_l, (v8i16) tmp0_r);
ST_SH(in0, (coeffs + i * buf_pitch));
tmp1_r -= sum0_r;
tmp1_l -= sum0_l;
SRAR_W2_SW(tmp1_r, tmp1_l, sum1_r);
SAT_SW2_SW(tmp1_r, tmp1_l, 15);
in0 = __msa_pckev_h((v8i16) tmp1_l, (v8i16) tmp1_r);
ST_SH(in0, (coeffs + (31 - i) * buf_pitch));
filter_ptr0 += 16;
}
}
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 );
}
void vp10_iht8x8_64_add_msa(const int16_t *input, uint8_t *dst,
int32_t dst_stride, int32_t tx_type) {
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
/* load vector elements of 8x8 block */
LD_SH8(input, 8, in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
switch (tx_type) {
case DCT_DCT:
/* DCT in horizontal */
VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
/* DCT in vertical */
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
break;
case ADST_DCT:
/* DCT in horizontal */
VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
/* ADST in vertical */
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
break;
case DCT_ADST:
/* ADST in horizontal */
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
/* DCT in vertical */
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
break;
case ADST_ADST:
/* ADST in horizontal */
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
/* ADST in vertical */
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
break;
default:
assert(0);
break;
}
/* final rounding (add 2^4, divide by 2^5) and shift */
SRARI_H4_SH(in0, in1, in2, in3, 5);
SRARI_H4_SH(in4, in5, in6, in7, 5);
/* add block and store 8x8 */
VP9_ADDBLK_ST8x4_UB(dst, dst_stride, in0, in1, in2, in3);
dst += (4 * dst_stride);
VP9_ADDBLK_ST8x4_UB(dst, dst_stride, in4, in5, in6, in7);
}
static void avc_dequant_8x8_msa( int16_t *p_dct, int32_t pi_dequant_mf[6][64],
int32_t i_qp )
{
const int32_t i_mf = i_qp % 6;
const int32_t q_bits = i_qp / 6 - 6;
v8i16 dct0, dct1, dct2, dct3, dct4, dct5, dct6, dct7;
v4i32 dequant_m_f0, dequant_m_f1, dequant_m_f2, dequant_m_f3;
v4i32 dequant_m_f4, dequant_m_f5, dequant_m_f6, dequant_m_f7;
v4i32 dequant_m_f8, dequant_m_f9, dequant_m_f10, dequant_m_f11;
v4i32 dequant_m_f12, dequant_m_f13, dequant_m_f14, dequant_m_f15;
LD_SH8( p_dct, 8, dct0, dct1, dct2, dct3, dct4, dct5, dct6, dct7 );
LD_SW2( pi_dequant_mf[i_mf], 4, dequant_m_f0, dequant_m_f1 );
LD_SW2( pi_dequant_mf[i_mf] + 8, 4, dequant_m_f2, dequant_m_f3 );
LD_SW2( pi_dequant_mf[i_mf] + 16, 4, dequant_m_f4, dequant_m_f5 );
LD_SW2( pi_dequant_mf[i_mf] + 24, 4, dequant_m_f6, dequant_m_f7 );
LD_SW2( pi_dequant_mf[i_mf] + 32, 4, dequant_m_f8, dequant_m_f9 );
LD_SW2( pi_dequant_mf[i_mf] + 40, 4, dequant_m_f10, dequant_m_f11 );
LD_SW2( pi_dequant_mf[i_mf] + 48, 4, dequant_m_f12, dequant_m_f13 );
LD_SW2( pi_dequant_mf[i_mf] + 56, 4, dequant_m_f14, dequant_m_f15 );
if( q_bits >= 0 )
{
v8i16 q_bits_vec;
v8i16 dequant_mf_h0, dequant_mf_h1, dequant_mf_h2, dequant_mf_h3;
v8i16 dequant_mf_h4, dequant_mf_h5, dequant_mf_h6, dequant_mf_h7;
q_bits_vec = __msa_fill_h( q_bits );
PCKEV_H4_SH( dequant_m_f1, dequant_m_f0, dequant_m_f3, dequant_m_f2,
dequant_m_f5, dequant_m_f4, dequant_m_f7, dequant_m_f6,
dequant_mf_h0, dequant_mf_h1,
dequant_mf_h2, dequant_mf_h3 );
PCKEV_H4_SH( dequant_m_f9, dequant_m_f8, dequant_m_f11, dequant_m_f10,
dequant_m_f13, dequant_m_f12, dequant_m_f15, dequant_m_f14,
dequant_mf_h4, dequant_mf_h5,
dequant_mf_h6, dequant_mf_h7 );
dct0 *= dequant_mf_h0;
dct1 *= dequant_mf_h1;
dct2 *= dequant_mf_h2;
dct3 *= dequant_mf_h3;
dct4 *= dequant_mf_h4;
dct5 *= dequant_mf_h5;
dct6 *= dequant_mf_h6;
dct7 *= dequant_mf_h7;
SLLI_4V( dct0, dct1, dct2, dct3, q_bits_vec );
SLLI_4V( dct4, dct5, dct6, dct7, q_bits_vec );
ST_SH8( dct0, dct1, dct2, dct3, dct4, dct5, dct6, dct7, p_dct, 8 );
}
else
{
const int32_t q_bits_add = 1 << ( -q_bits - 1 );
v4i32 dct_signed_w0, dct_signed_w1, dct_signed_w2, dct_signed_w3;
v4i32 dct_signed_w4, dct_signed_w5, dct_signed_w6, dct_signed_w7;
v4i32 dct_signed_w8, dct_signed_w9, dct_signed_w10, dct_signed_w11;
v4i32 dct_signed_w12, dct_signed_w13, dct_signed_w14, dct_signed_w15;
v4i32 q_bits_vec, q_bits_vec_add;
q_bits_vec_add = __msa_fill_w( q_bits_add );
q_bits_vec = __msa_fill_w( -q_bits );
UNPCK_SH_SW( dct0, dct_signed_w0, dct_signed_w1 );
UNPCK_SH_SW( dct1, dct_signed_w2, dct_signed_w3 );
UNPCK_SH_SW( dct2, dct_signed_w4, dct_signed_w5 );
UNPCK_SH_SW( dct3, dct_signed_w6, dct_signed_w7 );
UNPCK_SH_SW( dct4, dct_signed_w8, dct_signed_w9 );
UNPCK_SH_SW( dct5, dct_signed_w10, dct_signed_w11 );
UNPCK_SH_SW( dct6, dct_signed_w12, dct_signed_w13 );
UNPCK_SH_SW( dct7, dct_signed_w14, dct_signed_w15 );
dct_signed_w0 *= dequant_m_f0;
dct_signed_w1 *= dequant_m_f1;
dct_signed_w2 *= dequant_m_f2;
dct_signed_w3 *= dequant_m_f3;
dct_signed_w4 *= dequant_m_f4;
dct_signed_w5 *= dequant_m_f5;
dct_signed_w6 *= dequant_m_f6;
dct_signed_w7 *= dequant_m_f7;
dct_signed_w8 *= dequant_m_f8;
dct_signed_w9 *= dequant_m_f9;
dct_signed_w10 *= dequant_m_f10;
dct_signed_w11 *= dequant_m_f11;
dct_signed_w12 *= dequant_m_f12;
dct_signed_w13 *= dequant_m_f13;
dct_signed_w14 *= dequant_m_f14;
dct_signed_w15 *= dequant_m_f15;
dct_signed_w0 += q_bits_vec_add;
dct_signed_w1 += q_bits_vec_add;
dct_signed_w2 += q_bits_vec_add;
dct_signed_w3 += q_bits_vec_add;
dct_signed_w4 += q_bits_vec_add;
dct_signed_w5 += q_bits_vec_add;
dct_signed_w6 += q_bits_vec_add;
dct_signed_w7 += q_bits_vec_add;
dct_signed_w8 += q_bits_vec_add;
dct_signed_w9 += q_bits_vec_add;
dct_signed_w10 += q_bits_vec_add;
dct_signed_w11 += q_bits_vec_add;
dct_signed_w12 += q_bits_vec_add;
dct_signed_w13 += q_bits_vec_add;
dct_signed_w14 += q_bits_vec_add;
dct_signed_w15 += q_bits_vec_add;
SRA_4V( dct_signed_w0, dct_signed_w1, dct_signed_w2, dct_signed_w3,
q_bits_vec );
SRA_4V( dct_signed_w4, dct_signed_w5, dct_signed_w6, dct_signed_w7,
q_bits_vec );
SRA_4V( dct_signed_w8, dct_signed_w9, dct_signed_w10, dct_signed_w11,
q_bits_vec );
SRA_4V( dct_signed_w12, dct_signed_w13, dct_signed_w14, dct_signed_w15,
q_bits_vec );
PCKEV_H4_SH( dct_signed_w1, dct_signed_w0, dct_signed_w3, dct_signed_w2,
dct_signed_w5, dct_signed_w4, dct_signed_w7, dct_signed_w6,
dct0, dct1, dct2, dct3 );
PCKEV_H4_SH( dct_signed_w9, dct_signed_w8, dct_signed_w11,
dct_signed_w10, dct_signed_w13, dct_signed_w12,
dct_signed_w15, dct_signed_w14, dct4, dct5, dct6, dct7 );
ST_SH8( dct0, dct1, dct2, dct3, dct4, dct5, dct6, dct7, p_dct, 8 );
}
}
static void fdct8x32_1d_row_even_rd(int16_t *temp, int16_t *out) {
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, temp0, temp1;
/* fdct32 even */
/* stage 2 */
LD_SH8(temp, 8, in0, in1, in2, in3, in4, in5, in6, in7);
LD_SH8(temp + 64, 8, in8, in9, in10, in11, in12, in13, in14, in15);
BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7,
in8, in9, in10, in11, in12, in13, in14, in15,
vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7,
in8, in9, in10, in11, in12, in13, in14, in15);
FDCT_POSTPROC_2V_NEG_H(vec0, vec1);
FDCT_POSTPROC_2V_NEG_H(vec2, vec3);
FDCT_POSTPROC_2V_NEG_H(vec4, vec5);
FDCT_POSTPROC_2V_NEG_H(vec6, vec7);
FDCT_POSTPROC_2V_NEG_H(in8, in9);
FDCT_POSTPROC_2V_NEG_H(in10, in11);
FDCT_POSTPROC_2V_NEG_H(in12, in13);
FDCT_POSTPROC_2V_NEG_H(in14, in15);
/* Stage 3 */
ADD4(vec0, vec7, vec1, vec6, vec2, vec5, vec3, vec4, in0, in1, in2, in3);
temp0 = in0 + in3;
in0 = in0 - in3;
in3 = in1 + in2;
in1 = in1 - in2;
DOTP_CONST_PAIR(temp0, in3, cospi_16_64, cospi_16_64, temp1, temp0);
ST_SH(temp0, out);
ST_SH(temp1, out + 8);
DOTP_CONST_PAIR(in0, in1, cospi_24_64, cospi_8_64, temp1, temp0);
ST_SH(temp0, out + 16);
ST_SH(temp1, out + 24);
SUB4(vec3, vec4, vec2, vec5, vec1, vec6, vec0, vec7, vec4, vec5, vec6, vec7);
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);
ST_SH(temp0, out + 32);
ST_SH(temp1, out + 56);
SUB2(vec4, vec5, vec7, vec6, vec4, vec7);
DOTP_CONST_PAIR(vec7, vec4, cospi_12_64, cospi_20_64, temp1, temp0);
ST_SH(temp0, out + 40);
ST_SH(temp1, out + 48);
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);
ST_SH(temp0, out + 64);
ST_SH(temp1, out + 120);
SUB2(in0, in1, in2, in3, in0, in2);
DOTP_CONST_PAIR(in2, in0, cospi_14_64, cospi_18_64, temp1, temp0);
ST_SH(temp0, out + 72);
ST_SH(temp1, out + 112);
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);
ST_SH(temp0, out + 80);
ST_SH(temp1, out + 104);
ADD2(in3, in2, in0, in1, vec3, vec4);
DOTP_CONST_PAIR(vec4, vec3, cospi_6_64, cospi_26_64, temp0, temp1);
ST_SH(temp0, out + 96);
ST_SH(temp1, out + 88);
}
