static void fadst16_transpose_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_SH16(input, 8, l0, l8, l1, l9, l2, l10, l3, l11, l4, l12, l5, l13, l6, l14,
l7, l15);
TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7, r0, r1, r2, r3, r4, r5, r6,
r7);
TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15, r8, r9, r10, r11,
r12, r13, r14, r15);
ST_SH8(r0, r8, r1, r9, r2, r10, r3, r11, out, 8);
ST_SH8(r4, r12, r5, r13, r6, r14, r7, r15, (out + 64), 8);
out += 16 * 8;
/* load input data */
input += 128;
LD_SH16(input, 8, l0, l8, l1, l9, l2, l10, l3, l11, l4, l12, l5, l13, l6, l14,
l7, l15);
TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7, r0, r1, r2, r3, r4, r5, r6,
r7);
TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15, r8, r9, r10, r11,
r12, r13, r14, r15);
ST_SH8(r0, r8, r1, r9, r2, r10, r3, r11, out, 8);
ST_SH8(r4, r12, r5, r13, r6, r14, r7, r15, (out + 64), 8);
}
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 hevc_idct_dc_8x8_msa(int16_t *coeffs)
{
int32_t val;
v8i16 dst;
val = (coeffs[0] + 1) >> 1;
val = (val + 32) >> 6;
dst = __msa_fill_h(val);
ST_SH8(dst, dst, dst, dst, dst, dst, dst, dst, coeffs, 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 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);
}
static void hevc_idct_dc_32x32_msa(int16_t *coeffs)
{
uint8_t loop;
int32_t val;
v8i16 dst;
val = (coeffs[0] + 1) >> 1;
val = (val + 32) >> 6;
dst = __msa_fill_h(val);
for (loop = 16; loop--;) {
ST_SH8(dst, dst, dst, dst, dst, dst, dst, dst, coeffs, 8);
coeffs += 8 * 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);
}
void fdct8x16_1d_column(const int16_t *input, int16_t *tmp_ptr,
int32_t src_stride) {
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;
v8i16 stp21, stp22, stp23, stp24, stp25, stp26, stp30;
v8i16 stp31, stp32, stp33, stp34, stp35, stp36, stp37;
v8i16 vec0, vec1, vec2, vec3, vec4, vec5, cnst0, cnst1, cnst4, cnst5;
v8i16 coeff = { cospi_16_64, -cospi_16_64, cospi_8_64, cospi_24_64,
-cospi_8_64, -cospi_24_64, cospi_12_64, cospi_20_64 };
v8i16 coeff1 = { cospi_2_64, cospi_30_64, cospi_14_64, cospi_18_64,
cospi_10_64, cospi_22_64, cospi_6_64, cospi_26_64 };
v8i16 coeff2 = {
-cospi_2_64, -cospi_10_64, -cospi_18_64, -cospi_26_64, 0, 0, 0, 0
};
LD_SH16(input, src_stride, in0, in1, in2, in3, in4, in5, in6, in7, in8, in9,
in10, in11, in12, in13, in14, in15);
SLLI_4V(in0, in1, in2, in3, 2);
SLLI_4V(in4, in5, in6, in7, 2);
SLLI_4V(in8, in9, in10, in11, 2);
SLLI_4V(in12, in13, in14, in15, 2);
ADD4(in0, in15, in1, in14, in2, in13, in3, in12, tmp0, tmp1, tmp2, tmp3);
ADD4(in4, in11, in5, in10, in6, in9, in7, in8, tmp4, tmp5, tmp6, tmp7);
FDCT8x16_EVEN(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp0, tmp1,
tmp2, tmp3, tmp4, tmp5, tmp6, tmp7);
ST_SH8(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp_ptr, 32);
SUB4(in0, in15, in1, in14, in2, in13, in3, in12, in15, in14, in13, in12);
SUB4(in4, in11, in5, in10, in6, in9, in7, in8, in11, in10, in9, in8);
tmp_ptr += 16;
/* stp 1 */
ILVL_H2_SH(in10, in13, in11, in12, vec2, vec4);
ILVR_H2_SH(in10, in13, in11, in12, vec3, vec5);
cnst4 = __msa_splati_h(coeff, 0);
stp25 = DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst4);
cnst5 = __msa_splati_h(coeff, 1);
cnst5 = __msa_ilvev_h(cnst5, cnst4);
stp22 = DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst5);
stp24 = DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst4);
stp23 = DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst5);
/* stp2 */
BUTTERFLY_4(in8, in9, stp22, stp23, stp30, stp31, stp32, stp33);
BUTTERFLY_4(in15, in14, stp25, stp24, stp37, stp36, stp35, stp34);
ILVL_H2_SH(stp36, stp31, stp35, stp32, vec2, vec4);
ILVR_H2_SH(stp36, stp31, stp35, stp32, vec3, vec5);
SPLATI_H2_SH(coeff, 2, 3, cnst0, cnst1);
cnst0 = __msa_ilvev_h(cnst0, cnst1);
stp26 = DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst0);
cnst0 = __msa_splati_h(coeff, 4);
cnst1 = __msa_ilvev_h(cnst1, cnst0);
stp21 = DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst1);
BUTTERFLY_4(stp30, stp37, stp26, stp21, in8, in15, in14, in9);
ILVRL_H2_SH(in15, in8, vec1, vec0);
SPLATI_H2_SH(coeff1, 0, 1, cnst0, cnst1);
cnst0 = __msa_ilvev_h(cnst0, cnst1);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr);
cnst0 = __msa_splati_h(coeff2, 0);
cnst0 = __msa_ilvev_h(cnst1, cnst0);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr + 224);
ILVRL_H2_SH(in14, in9, vec1, vec0);
SPLATI_H2_SH(coeff1, 2, 3, cnst0, cnst1);
cnst1 = __msa_ilvev_h(cnst1, cnst0);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst1);
ST_SH(in8, tmp_ptr + 128);
cnst1 = __msa_splati_h(coeff2, 2);
cnst0 = __msa_ilvev_h(cnst0, cnst1);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr + 96);
SPLATI_H2_SH(coeff, 2, 5, cnst0, cnst1);
cnst1 = __msa_ilvev_h(cnst1, cnst0);
stp25 = DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst1);
cnst1 = __msa_splati_h(coeff, 3);
cnst1 = __msa_ilvev_h(cnst0, cnst1);
stp22 = DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst1);
/* stp4 */
ADD2(stp34, stp25, stp33, stp22, in13, in10);
ILVRL_H2_SH(in13, in10, vec1, vec0);
SPLATI_H2_SH(coeff1, 4, 5, cnst0, cnst1);
cnst0 = __msa_ilvev_h(cnst0, cnst1);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr + 64);
cnst0 = __msa_splati_h(coeff2, 1);
cnst0 = __msa_ilvev_h(cnst1, cnst0);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr + 160);
SUB2(stp34, stp25, stp33, stp22, in12, in11);
ILVRL_H2_SH(in12, in11, vec1, vec0);
SPLATI_H2_SH(coeff1, 6, 7, cnst0, cnst1);
cnst1 = __msa_ilvev_h(cnst1, cnst0);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst1);
ST_SH(in8, tmp_ptr + 192);
cnst1 = __msa_splati_h(coeff2, 3);
cnst0 = __msa_ilvev_h(cnst0, cnst1);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr + 32);
}
static void fdct8x32_1d_row_transpose_store(int16_t *temp, int16_t *output) {
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
v8i16 in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1;
/* 1st set */
in0 = LD_SH(temp);
in4 = LD_SH(temp + 32);
in2 = LD_SH(temp + 64);
in6 = LD_SH(temp + 96);
in1 = LD_SH(temp + 128);
in7 = LD_SH(temp + 152);
in3 = LD_SH(temp + 192);
in5 = LD_SH(temp + 216);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
/* 2nd set */
in0_1 = LD_SH(temp + 16);
in1_1 = LD_SH(temp + 232);
in2_1 = LD_SH(temp + 80);
in3_1 = LD_SH(temp + 168);
in4_1 = LD_SH(temp + 48);
in5_1 = LD_SH(temp + 176);
in6_1 = LD_SH(temp + 112);
in7_1 = LD_SH(temp + 240);
ST_SH8(in0, in1, in2, in3, in4, in5, in6, in7, output, 32);
TRANSPOSE8x8_SH_SH(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1,
in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1);
/* 3rd set */
in0 = LD_SH(temp + 8);
in1 = LD_SH(temp + 136);
in2 = LD_SH(temp + 72);
in3 = LD_SH(temp + 200);
in4 = LD_SH(temp + 40);
in5 = LD_SH(temp + 208);
in6 = LD_SH(temp + 104);
in7 = LD_SH(temp + 144);
ST_SH8(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1,
output + 8, 32);
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, output + 16, 32);
/* 4th set */
in0_1 = LD_SH(temp + 24);
in1_1 = LD_SH(temp + 224);
in2_1 = LD_SH(temp + 88);
in3_1 = LD_SH(temp + 160);
in4_1 = LD_SH(temp + 56);
in5_1 = LD_SH(temp + 184);
in6_1 = LD_SH(temp + 120);
in7_1 = LD_SH(temp + 248);
TRANSPOSE8x8_SH_SH(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1,
in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1);
ST_SH8(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1,
output + 24, 32);
}
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 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 );
}
}
