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);
}
void vpx_fdct4x4_msa(const int16_t *input, int16_t *output,
int32_t src_stride) {
v8i16 in0, in1, in2, in3;
LD_SH4(input, src_stride, in0, in1, in2, in3);
/* fdct4 pre-process */
{
v8i16 vec, mask;
v16i8 zero = { 0 };
v16i8 one = __msa_ldi_b(1);
mask = (v8i16)__msa_sldi_b(zero, one, 15);
SLLI_4V(in0, in1, in2, in3, 4);
vec = __msa_ceqi_h(in0, 0);
vec = vec ^ 255;
vec = mask & vec;
in0 += vec;
}
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
ADD4(in0, 1, in1, 1, in2, 1, in3, 1, in0, in1, in2, in3);
SRA_4V(in0, in1, in2, in3, 2);
PCKEV_D2_SH(in1, in0, in3, in2, in0, in2);
ST_SH2(in0, in2, output, 8);
}
void vp8_short_walsh4x4_msa(int16_t *input, int16_t *output, int32_t pitch) {
v8i16 in0_h, in1_h, in2_h, in3_h;
v4i32 in0_w, in1_w, in2_w, in3_w, temp0, temp1, temp2, temp3;
LD_SH4(input, pitch / 2, in0_h, in1_h, in2_h, in3_h);
TRANSPOSE4x4_SH_SH(in0_h, in1_h, in2_h, in3_h, in0_h, in1_h, in2_h, in3_h);
UNPCK_R_SH_SW(in0_h, in0_w);
UNPCK_R_SH_SW(in1_h, in1_w);
UNPCK_R_SH_SW(in2_h, in2_w);
UNPCK_R_SH_SW(in3_h, in3_w);
BUTTERFLY_4(in0_w, in1_w, in3_w, in2_w, temp0, temp3, temp2, temp1);
SLLI_4V(temp0, temp1, temp2, temp3, 2);
BUTTERFLY_4(temp0, temp1, temp2, temp3, in0_w, in1_w, in2_w, in3_w);
temp0 = RET_1_IF_NZERO_W(temp0);
in0_w += temp0;
TRANSPOSE4x4_SW_SW(in0_w, in1_w, in2_w, in3_w, in0_w, in1_w, in2_w, in3_w);
BUTTERFLY_4(in0_w, in1_w, in3_w, in2_w, temp0, temp3, temp2, temp1);
BUTTERFLY_4(temp0, temp1, temp2, temp3, in0_w, in1_w, in2_w, in3_w);
in0_w += RET_1_IF_NEG_W(in0_w);
in1_w += RET_1_IF_NEG_W(in1_w);
in2_w += RET_1_IF_NEG_W(in2_w);
in3_w += RET_1_IF_NEG_W(in3_w);
ADD4(in0_w, 3, in1_w, 3, in2_w, 3, in3_w, 3, in0_w, in1_w, in2_w, in3_w);
SRA_4V(in0_w, in1_w, in2_w, in3_w, 3);
PCKEV_H2_SH(in1_w, in0_w, in3_w, in2_w, in0_h, in1_h);
ST_SH2(in0_h, in1_h, output, 8);
}
void vp9_fwht4x4_msa(const int16_t *input, int16_t *output,
int32_t src_stride) {
v8i16 in0, in1, in2, in3, in4;
LD_SH4(input, src_stride, in0, in1, in2, in3);
in0 += in1;
in3 -= in2;
in4 = (in0 - in3) >> 1;
SUB2(in4, in1, in4, in2, in1, in2);
in0 -= in2;
in3 += in1;
TRANSPOSE4x4_SH_SH(in0, in2, in3, in1, in0, in2, in3, in1);
in0 += in2;
in1 -= in3;
in4 = (in0 - in1) >> 1;
SUB2(in4, in2, in4, in3, in2, in3);
in0 -= in3;
in1 += in2;
SLLI_4V(in0, in1, in2, in3, 2);
TRANSPOSE4x4_SH_SH(in0, in3, in1, in2, in0, in3, in1, in2);
ST4x2_UB(in0, output, 4);
ST4x2_UB(in3, output + 4, 4);
ST4x2_UB(in1, output + 8, 4);
ST4x2_UB(in2, output + 12, 4);
}
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 vp8_short_fdct8x4_msa(int16_t *input, int16_t *output, int32_t pitch) {
v8i16 in0, in1, in2, in3;
v8i16 temp0, temp1, tmp0, tmp1;
v8i16 const0, const1, const2;
v8i16 coeff = { 2217, 5352, -5352, 14500, 7500, 12000, 25000, 26000 };
v8i16 zero = { 0 };
v4i32 vec0_w, vec1_w, vec2_w, vec3_w;
LD_SH4(input, pitch / 2, in0, in1, in2, in3);
TRANSPOSE4x4_H(in0, in1, in2, in3, in0, in1, in2, in3);
BUTTERFLY_4(in0, in1, in2, in3, temp0, temp1, in1, in3);
SLLI_4V(temp0, temp1, in1, in3, 3);
in0 = temp0 + temp1;
in2 = temp0 - temp1;
SET_DOTP_VALUES(coeff, 0, 1, 2, const1, const2);
temp0 = __msa_splati_h(coeff, 3);
vec1_w = (v4i32)__msa_ilvev_h(zero, temp0);
coeff = __msa_ilvl_h(zero, coeff);
vec3_w = __msa_splati_w((v4i32)coeff, 0);
ILVRL_H2_SH(in3, in1, tmp1, tmp0);
vec0_w = vec1_w;
vec2_w = vec3_w;
DPADD_SH4_SW(tmp1, tmp0, tmp1, tmp0, const1, const1, const2, const2, vec0_w,
vec1_w, vec2_w, vec3_w);
SRA_4V(vec1_w, vec0_w, vec3_w, vec2_w, 12);
PCKEV_H2_SH(vec1_w, vec0_w, vec3_w, vec2_w, in1, in3);
TRANSPOSE4x4_H(in0, in1, in2, in3, in0, in1, in2, in3);
BUTTERFLY_4(in0, in1, in2, in3, temp0, temp1, in1, in3);
in0 = temp0 + temp1 + 7;
in2 = temp0 - temp1 + 7;
in0 >>= 4;
in2 >>= 4;
SPLATI_W2_SW(coeff, 2, vec3_w, vec1_w);
vec3_w += vec1_w;
vec1_w = __msa_splati_w((v4i32)coeff, 1);
const0 = RET_1_IF_NZERO_H(in3);
ILVRL_H2_SH(in3, in1, tmp1, tmp0);
vec0_w = vec1_w;
vec2_w = vec3_w;
DPADD_SH4_SW(tmp1, tmp0, tmp1, tmp0, const1, const1, const2, const2, vec0_w,
vec1_w, vec2_w, vec3_w);
SRA_4V(vec1_w, vec0_w, vec3_w, vec2_w, 16);
PCKEV_H2_SH(vec1_w, vec0_w, vec3_w, vec2_w, in1, in3);
in1 += const0;
PCKEV_D2_SH(in1, in0, in3, in2, temp0, temp1);
ST_SH2(temp0, temp1, output, 8);
PCKOD_D2_SH(in1, in0, in3, in2, in0, in2);
ST_SH2(in0, in2, output + 16, 8);
}
void vp9_fht4x4_msa(const int16_t *input, int16_t *output, int32_t stride,
int32_t tx_type) {
v8i16 in0, in1, in2, in3;
LD_SH4(input, stride, in0, in1, in2, in3);
/* fdct4 pre-process */
{
v8i16 temp, mask;
v16i8 zero = { 0 };
v16i8 one = __msa_ldi_b(1);
mask = (v8i16)__msa_sldi_b(zero, one, 15);
SLLI_4V(in0, in1, in2, in3, 4);
temp = __msa_ceqi_h(in0, 0);
temp = (v8i16)__msa_xori_b((v16u8)temp, 255);
temp = mask & temp;
in0 += temp;
}
switch (tx_type) {
case DCT_DCT:
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
case ADST_DCT:
VP9_FADST4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
case DCT_ADST:
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_FADST4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
case ADST_ADST:
VP9_FADST4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_FADST4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
default:
assert(0);
break;
}
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
ADD4(in0, 1, in1, 1, in2, 1, in3, 1, in0, in1, in2, in3);
SRA_4V(in0, in1, in2, in3, 2);
PCKEV_D2_SH(in1, in0, in3, in2, in0, in2);
ST_SH2(in0, in2, output, 8);
}
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);
}
void vp8_short_fdct4x4_msa(int16_t *input, int16_t *output, int32_t pitch) {
v8i16 in0, in1, in2, in3;
v8i16 temp0, temp1;
v8i16 const0, const1;
v8i16 coeff = { 2217, 5352, -5352, 14500, 7500, 12000, 25000, 26000 };
v4i32 out0, out1, out2, out3;
v8i16 zero = { 0 };
LD_SH4(input, pitch / 2, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
BUTTERFLY_4(in0, in1, in2, in3, temp0, temp1, in1, in3);
SLLI_4V(temp0, temp1, in1, in3, 3);
in0 = temp0 + temp1;
in2 = temp0 - temp1;
SET_DOTP_VALUES(coeff, 0, 1, 2, const0, const1);
temp0 = __msa_ilvr_h(in3, in1);
in1 = __msa_splati_h(coeff, 3);
out0 = (v4i32)__msa_ilvev_h(zero, in1);
coeff = __msa_ilvl_h(zero, coeff);
out1 = __msa_splati_w((v4i32)coeff, 0);
DPADD_SH2_SW(temp0, temp0, const0, const1, out0, out1);
out0 >>= 12;
out1 >>= 12;
PCKEV_H2_SH(out0, out0, out1, out1, in1, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
BUTTERFLY_4(in0, in1, in2, in3, temp0, temp1, in1, in3);
in0 = temp0 + temp1 + 7;
in2 = temp0 - temp1 + 7;
in0 >>= 4;
in2 >>= 4;
ILVR_H2_SW(zero, in0, zero, in2, out0, out2);
temp1 = RET_1_IF_NZERO_H(in3);
ILVR_H2_SH(zero, temp1, in3, in1, temp1, temp0);
SPLATI_W2_SW(coeff, 2, out3, out1);
out3 += out1;
out1 = __msa_splati_w((v4i32)coeff, 1);
DPADD_SH2_SW(temp0, temp0, const0, const1, out1, out3);
out1 >>= 16;
out3 >>= 16;
out1 += (v4i32)temp1;
PCKEV_H2_SH(out1, out0, out3, out2, in0, in2);
ST_SH2(in0, in2, 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 fadst16_cols_step1_msa(const int16_t *input, int32_t stride,
const int32_t *const0, int16_t *int_buf) {
v8i16 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15;
v8i16 tp0, tp1, tp2, tp3, g0, g1, g2, g3, g8, g9, g10, g11, h0, h1, h2, h3;
v4i32 k0, k1, k2, k3;
/* load input data */
r0 = LD_SH(input);
r15 = LD_SH(input + 15 * stride);
r7 = LD_SH(input + 7 * stride);
r8 = LD_SH(input + 8 * stride);
SLLI_4V(r0, r15, r7, r8, 2);
/* stage 1 */
LD_SW2(const0, 4, k0, k1);
LD_SW2(const0 + 8, 4, k2, k3);
MADD_BF(r15, r0, r7, r8, k0, k1, k2, k3, g0, g1, g2, g3);
r3 = LD_SH(input + 3 * stride);
r4 = LD_SH(input + 4 * stride);
r11 = LD_SH(input + 11 * stride);
r12 = LD_SH(input + 12 * stride);
SLLI_4V(r3, r4, r11, r12, 2);
LD_SW2(const0 + 4 * 4, 4, k0, k1);
LD_SW2(const0 + 4 * 6, 4, k2, k3);
MADD_BF(r11, r4, r3, r12, k0, k1, k2, k3, g8, g9, g10, g11);
/* stage 2 */
BUTTERFLY_4(g0, g2, g10, g8, tp0, tp2, tp3, tp1);
ST_SH2(tp0, tp2, int_buf, 8);
ST_SH2(tp1, tp3, int_buf + 4 * 8, 8);
LD_SW2(const0 + 4 * 8, 4, k0, k1);
k2 = LD_SW(const0 + 4 * 10);
MADD_BF(g1, g3, g9, g11, k0, k1, k2, k0, h0, h1, h2, h3);
ST_SH2(h0, h1, int_buf + 8 * 8, 8);
ST_SH2(h3, h2, int_buf + 12 * 8, 8);
r9 = LD_SH(input + 9 * stride);
r6 = LD_SH(input + 6 * stride);
r1 = LD_SH(input + stride);
r14 = LD_SH(input + 14 * stride);
SLLI_4V(r9, r6, r1, r14, 2);
LD_SW2(const0 + 4 * 11, 4, k0, k1);
LD_SW2(const0 + 4 * 13, 4, k2, k3);
MADD_BF(r9, r6, r1, r14, k0, k1, k2, k3, g0, g1, g2, g3);
ST_SH2(g1, g3, int_buf + 3 * 8, 4 * 8);
r13 = LD_SH(input + 13 * stride);
r2 = LD_SH(input + 2 * stride);
r5 = LD_SH(input + 5 * stride);
r10 = LD_SH(input + 10 * stride);
SLLI_4V(r13, r2, r5, r10, 2);
LD_SW2(const0 + 4 * 15, 4, k0, k1);
LD_SW2(const0 + 4 * 17, 4, k2, k3);
MADD_BF(r13, r2, r5, r10, k0, k1, k2, k3, h0, h1, h2, h3);
ST_SH2(h1, h3, int_buf + 11 * 8, 4 * 8);
BUTTERFLY_4(h0, h2, g2, g0, tp0, tp1, tp2, tp3);
ST_SH4(tp0, tp1, tp2, tp3, int_buf + 2 * 8, 4 * 8);
}
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 );
}
}