int32_t vp8_block_error_msa(int16_t *coeff_ptr, int16_t *dq_coeff_ptr)
{
int32_t err = 0;
uint32_t loop_cnt;
v8i16 coeff, dq_coeff, coeff0, coeff1;
v4i32 diff0, diff1;
v2i64 err0 = { 0 };
v2i64 err1 = { 0 };
for (loop_cnt = 2; loop_cnt--;)
{
coeff = LD_SH(coeff_ptr);
dq_coeff = LD_SH(dq_coeff_ptr);
ILVRL_H2_SH(coeff, dq_coeff, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
DPADD_SD2_SD(diff0, diff1, err0, err1);
coeff_ptr += 8;
dq_coeff_ptr += 8;
}
err0 += __msa_splati_d(err0, 1);
err1 += __msa_splati_d(err1, 1);
err = __msa_copy_s_d(err0, 0);
err += __msa_copy_s_d(err1, 0);
return err;
}
static void common_vt_8t_and_aver_dst_4w_msa(const uint8_t *src,
int32_t src_stride, uint8_t *dst,
int32_t dst_stride, int8_t *filter,
int32_t height) {
uint32_t loop_cnt;
v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
v16u8 dst0, dst1, dst2, dst3, out;
v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r;
v16i8 src65_r, src87_r, src109_r, src2110, src4332, src6554, src8776;
v16i8 src10998, filt0, filt1, filt2, filt3;
v8i16 filt, out10, out32;
src -= (3 * src_stride);
filt = LD_SH(filter);
SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
src += (7 * src_stride);
ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r, src32_r,
src54_r, src21_r);
ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
ILVR_D3_SB(src21_r, src10_r, src43_r, src32_r, src65_r, src54_r, src2110,
src4332, src6554);
XORI_B3_128_SB(src2110, src4332, src6554);
for (loop_cnt = (height >> 2); loop_cnt--;) {
static void common_hv_8ht_8vt_and_aver_dst_4w_msa(const uint8_t *src,
int32_t src_stride,
uint8_t *dst,
int32_t dst_stride,
int8_t *filter_horiz,
int8_t *filter_vert,
int32_t height) {
uint32_t loop_cnt;
v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
v16u8 dst0, dst1, dst2, dst3, mask0, mask1, mask2, mask3, tmp0, tmp1;
v16i8 filt_hz0, filt_hz1, filt_hz2, filt_hz3;
v8i16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, hz_out5, hz_out6;
v8i16 hz_out7, hz_out8, hz_out9, res0, res1, vec0, vec1, vec2, vec3, vec4;
v8i16 filt, filt_vt0, filt_vt1, filt_vt2, filt_vt3;
mask0 = LD_UB(&mc_filt_mask_arr[16]);
src -= (3 + 3 * src_stride);
/* rearranging filter */
filt = LD_SH(filter_horiz);
SPLATI_H4_SB(filt, 0, 1, 2, 3, filt_hz0, filt_hz1, filt_hz2, filt_hz3);
mask1 = mask0 + 2;
mask2 = mask0 + 4;
mask3 = mask0 + 6;
LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
src += (7 * src_stride);
hz_out0 = HORIZ_8TAP_FILT(src0, src1, mask0, mask1, mask2, mask3, filt_hz0,
filt_hz1, filt_hz2, filt_hz3);
hz_out2 = HORIZ_8TAP_FILT(src2, src3, mask0, mask1, mask2, mask3, filt_hz0,
filt_hz1, filt_hz2, filt_hz3);
hz_out4 = HORIZ_8TAP_FILT(src4, src5, mask0, mask1, mask2, mask3, filt_hz0,
filt_hz1, filt_hz2, filt_hz3);
hz_out5 = HORIZ_8TAP_FILT(src5, src6, mask0, mask1, mask2, mask3, filt_hz0,
filt_hz1, filt_hz2, filt_hz3);
SLDI_B2_SH(hz_out2, hz_out4, hz_out0, hz_out2, hz_out1, hz_out3, 8);
filt = LD_SH(filter_vert);
SPLATI_H4_SH(filt, 0, 1, 2, 3, filt_vt0, filt_vt1, filt_vt2, filt_vt3);
ILVEV_B2_SH(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
vec2 = (v8i16)__msa_ilvev_b((v16i8)hz_out5, (v16i8)hz_out4);
for (loop_cnt = (height >> 2); loop_cnt--;) {
static void fadst16_cols_step2_msa(int16_t *int_buf, const int32_t *const0,
int16_t *out) {
int16_t *out_ptr = out + 128;
v8i16 tp0, tp1, tp2, tp3, g5, g7, g13, g15;
v8i16 h0, h1, h2, h3, h4, h5, h6, h7, h10, h11;
v8i16 out0, out1, out2, out3, out4, out5, out6, out7;
v8i16 out8, out9, out10, out11, out12, out13, out14, out15;
v4i32 k0, k1, k2, k3;
LD_SH2(int_buf + 3 * 8, 4 * 8, g13, g15);
LD_SH2(int_buf + 11 * 8, 4 * 8, g5, g7);
LD_SW2(const0 + 4 * 19, 4, k0, k1);
k2 = LD_SW(const0 + 4 * 21);
MADD_BF(g7, g5, g15, g13, k0, k1, k2, k0, h4, h5, h6, h7);
tp0 = LD_SH(int_buf + 4 * 8);
tp1 = LD_SH(int_buf + 5 * 8);
tp3 = LD_SH(int_buf + 10 * 8);
tp2 = LD_SH(int_buf + 14 * 8);
LD_SW2(const0 + 4 * 22, 4, k0, k1);
k2 = LD_SW(const0 + 4 * 24);
MADD_BF(tp0, tp1, tp2, tp3, k0, k1, k2, k0, out4, out6, out5, out7);
out4 = -out4;
ST_SH(out4, (out + 3 * 16));
ST_SH(out5, (out_ptr + 4 * 16));
h1 = LD_SH(int_buf + 9 * 8);
h3 = LD_SH(int_buf + 12 * 8);
MADD_BF(h1, h3, h5, h7, k0, k1, k2, k0, out12, out14, out13, out15);
out13 = -out13;
ST_SH(out12, (out + 2 * 16));
ST_SH(out13, (out_ptr + 5 * 16));
tp0 = LD_SH(int_buf);
tp1 = LD_SH(int_buf + 8);
tp2 = LD_SH(int_buf + 2 * 8);
tp3 = LD_SH(int_buf + 6 * 8);
BUTTERFLY_4(tp0, tp1, tp3, tp2, out0, out1, h11, h10);
out1 = -out1;
ST_SH(out0, (out));
ST_SH(out1, (out_ptr + 7 * 16));
h0 = LD_SH(int_buf + 8 * 8);
h2 = LD_SH(int_buf + 13 * 8);
BUTTERFLY_4(h0, h2, h6, h4, out8, out9, out11, out10);
out8 = -out8;
ST_SH(out8, (out + 16));
ST_SH(out9, (out_ptr + 6 * 16));
/* stage 4 */
LD_SW2(const0 + 4 * 25, 4, k0, k1);
LD_SW2(const0 + 4 * 27, 4, k2, k3);
MADD_SHORT(h10, h11, k1, k2, out2, out3);
ST_SH(out2, (out + 7 * 16));
ST_SH(out3, (out_ptr));
MADD_SHORT(out6, out7, k0, k3, out6, out7);
ST_SH(out6, (out + 4 * 16));
ST_SH(out7, (out_ptr + 3 * 16));
MADD_SHORT(out10, out11, k0, k3, out10, out11);
ST_SH(out10, (out + 6 * 16));
ST_SH(out11, (out_ptr + 16));
MADD_SHORT(out14, out15, k1, k2, out14, out15);
ST_SH(out14, (out + 5 * 16));
ST_SH(out15, (out_ptr + 2 * 16));
}
static void fadst16_rows_step1_msa(int16_t *input, 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);
r7 = LD_SH(input + 7 * 8);
r8 = LD_SH(input + 8 * 8);
r15 = LD_SH(input + 15 * 8);
/* stage 1 */
LD_SW2(const0, 4, k0, k1);
LD_SW2(const0 + 4 * 2, 4, k2, k3);
MADD_BF(r15, r0, r7, r8, k0, k1, k2, k3, g0, g1, g2, g3);
r3 = LD_SH(input + 3 * 8);
r4 = LD_SH(input + 4 * 8);
r11 = LD_SH(input + 11 * 8);
r12 = LD_SH(input + 12 * 8);
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, tp1, int_buf, 4 * 8);
ST_SH2(tp2, tp3, int_buf + 8, 4 * 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, h3, int_buf + 8 * 8, 4 * 8);
ST_SH2(h1, h2, int_buf + 9 * 8, 4 * 8);
r1 = LD_SH(input + 8);
r6 = LD_SH(input + 6 * 8);
r9 = LD_SH(input + 9 * 8);
r14 = LD_SH(input + 14 * 8);
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);
r2 = LD_SH(input + 2 * 8);
r5 = LD_SH(input + 5 * 8);
r10 = LD_SH(input + 10 * 8);
r13 = LD_SH(input + 13 * 8);
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 fdct8x32_1d_row_odd_rd(int16_t *temp, int16_t *interm_ptr,
int16_t *out) {
v8i16 in16, in17, in18, in19, in20, in21, in22, in23;
v8i16 in24, in25, in26, in27, in28, in29, in30, in31;
v8i16 vec4, vec5;
in20 = LD_SH(temp + 32);
in21 = LD_SH(temp + 40);
in26 = LD_SH(temp + 80);
in27 = LD_SH(temp + 88);
DOTP_CONST_PAIR(in27, in20, cospi_16_64, cospi_16_64, in20, in27);
DOTP_CONST_PAIR(in26, in21, cospi_16_64, cospi_16_64, in21, in26);
FDCT_POSTPROC_2V_NEG_H(in20, in21);
FDCT_POSTPROC_2V_NEG_H(in26, in27);
in18 = LD_SH(temp + 16);
in19 = LD_SH(temp + 24);
in28 = LD_SH(temp + 96);
in29 = LD_SH(temp + 104);
FDCT_POSTPROC_2V_NEG_H(in18, in19);
FDCT_POSTPROC_2V_NEG_H(in28, in29);
vec4 = in19 - in20;
ST_SH(vec4, interm_ptr + 32);
vec4 = in18 - in21;
ST_SH(vec4, interm_ptr + 88);
vec4 = in29 - in26;
ST_SH(vec4, interm_ptr + 64);
vec4 = in28 - in27;
ST_SH(vec4, interm_ptr + 56);
ADD4(in18, in21, in19, in20, in28, in27, in29, in26, in21, in20, in27, in26);
in22 = LD_SH(temp + 48);
in23 = LD_SH(temp + 56);
in24 = LD_SH(temp + 64);
in25 = LD_SH(temp + 72);
DOTP_CONST_PAIR(in25, in22, cospi_16_64, cospi_16_64, in22, in25);
DOTP_CONST_PAIR(in24, in23, cospi_16_64, cospi_16_64, in23, in24);
FDCT_POSTPROC_2V_NEG_H(in22, in23);
FDCT_POSTPROC_2V_NEG_H(in24, in25);
in16 = LD_SH(temp);
in17 = LD_SH(temp + 8);
in30 = LD_SH(temp + 112);
in31 = LD_SH(temp + 120);
FDCT_POSTPROC_2V_NEG_H(in16, in17);
FDCT_POSTPROC_2V_NEG_H(in30, in31);
vec4 = in17 - in22;
ST_SH(vec4, interm_ptr + 40);
vec4 = in30 - in25;
ST_SH(vec4, interm_ptr + 48);
vec4 = in31 - in24;
ST_SH(vec4, interm_ptr + 72);
vec4 = in16 - in23;
ST_SH(vec4, interm_ptr + 80);
ADD4(in16, in23, in17, in22, in30, in25, in31, in24, in16, in17, in30, in31);
DOTP_CONST_PAIR(in26, in21, cospi_24_64, cospi_8_64, in18, in29);
DOTP_CONST_PAIR(in27, in20, cospi_24_64, cospi_8_64, in19, in28);
ADD4(in16, in19, in17, in18, in30, in29, in31, in28, in27, in22, in21, in25);
DOTP_CONST_PAIR(in21, in22, cospi_28_64, cospi_4_64, in26, in24);
ADD2(in27, in26, in25, in24, in23, in20);
DOTP_CONST_PAIR(in20, in23, cospi_31_64, cospi_1_64, vec4, vec5);
ST_SH(vec5, out);
ST_SH(vec4, out + 120);
SUB2(in27, in26, in25, in24, in22, in21);
DOTP_CONST_PAIR(in21, in22, cospi_15_64, cospi_17_64, vec5, vec4);
ST_SH(vec5, out + 112);
ST_SH(vec4, out + 8);
SUB4(in17, in18, in16, in19, in31, in28, in30, in29, in23, in26, in24, in20);
DOTP_CONST_PAIR((-in23), in20, cospi_28_64, cospi_4_64, in27, in25);
SUB2(in26, in27, in24, in25, in23, in20);
DOTP_CONST_PAIR(in20, in23, cospi_23_64, cospi_9_64, vec4, vec5);
ST_SH(vec4, out + 16);
ST_SH(vec5, out + 104);
ADD2(in26, in27, in24, in25, in22, in21);
DOTP_CONST_PAIR(in21, in22, cospi_7_64, cospi_25_64, vec4, vec5);
ST_SH(vec4, out + 24);
ST_SH(vec5, out + 96);
in20 = LD_SH(interm_ptr + 32);
in21 = LD_SH(interm_ptr + 88);
in27 = LD_SH(interm_ptr + 56);
in26 = LD_SH(interm_ptr + 64);
in16 = in20;
in17 = in21;
DOTP_CONST_PAIR(-in16, in27, cospi_24_64, cospi_8_64, in20, in27);
DOTP_CONST_PAIR(-in17, in26, cospi_24_64, cospi_8_64, in21, in26);
in22 = LD_SH(interm_ptr + 40);
in25 = LD_SH(interm_ptr + 48);
in24 = LD_SH(interm_ptr + 72);
in23 = LD_SH(interm_ptr + 80);
SUB4(in23, in20, in22, in21, in25, in26, in24, in27, in28, in17, in18, in31);
DOTP_CONST_PAIR(in18, in17, cospi_12_64, cospi_20_64, in29, in30);
in16 = in28 + in29;
in19 = in31 + in30;
DOTP_CONST_PAIR(in19, in16, cospi_27_64, cospi_5_64, vec5, vec4);
ST_SH(vec5, out + 32);
ST_SH(vec4, out + 88);
SUB2(in28, in29, in31, in30, in17, in18);
DOTP_CONST_PAIR(in18, in17, cospi_11_64, cospi_21_64, vec5, vec4);
ST_SH(vec5, out + 40);
ST_SH(vec4, out + 80);
ADD4(in22, in21, in23, in20, in24, in27, in25, in26, in16, in29, in30, in19);
DOTP_CONST_PAIR(-in16, in19, cospi_12_64, cospi_20_64, in28, in31);
SUB2(in29, in28, in30, in31, in16, in19);
DOTP_CONST_PAIR(in19, in16, cospi_19_64, cospi_13_64, vec5, vec4);
ST_SH(vec5, out + 72);
ST_SH(vec4, out + 48);
ADD2(in29, in28, in30, in31, in17, in18);
DOTP_CONST_PAIR(in18, in17, cospi_3_64, cospi_29_64, vec5, vec4);
ST_SH(vec4, out + 56);
ST_SH(vec5, out + 64);
}
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_column_odd_store(int16_t *input, int16_t *temp_ptr) {
v8i16 in16, in17, in18, in19, in20, in21, in22, in23;
v8i16 in24, in25, in26, in27, in28, in29, in30, in31, vec4, vec5;
in20 = LD_SH(input + 32);
in21 = LD_SH(input + 40);
in26 = LD_SH(input + 80);
in27 = LD_SH(input + 88);
DOTP_CONST_PAIR(in27, in20, cospi_16_64, cospi_16_64, in20, in27);
DOTP_CONST_PAIR(in26, in21, cospi_16_64, cospi_16_64, in21, in26);
in18 = LD_SH(input + 16);
in19 = LD_SH(input + 24);
in28 = LD_SH(input + 96);
in29 = LD_SH(input + 104);
vec4 = in19 - in20;
ST_SH(vec4, input + 32);
vec4 = in18 - in21;
ST_SH(vec4, input + 40);
vec4 = in29 - in26;
ST_SH(vec4, input + 80);
vec4 = in28 - in27;
ST_SH(vec4, input + 88);
in21 = in18 + in21;
in20 = in19 + in20;
in27 = in28 + in27;
in26 = in29 + in26;
LD_SH4(input + 48, 8, in22, in23, in24, in25);
DOTP_CONST_PAIR(in25, in22, cospi_16_64, cospi_16_64, in22, in25);
DOTP_CONST_PAIR(in24, in23, cospi_16_64, cospi_16_64, in23, in24);
in16 = LD_SH(input);
in17 = LD_SH(input + 8);
in30 = LD_SH(input + 112);
in31 = LD_SH(input + 120);
vec4 = in17 - in22;
ST_SH(vec4, input + 16);
vec4 = in16 - in23;
ST_SH(vec4, input + 24);
vec4 = in31 - in24;
ST_SH(vec4, input + 96);
vec4 = in30 - in25;
ST_SH(vec4, input + 104);
ADD4(in16, in23, in17, in22, in30, in25, in31, in24, in16, in17, in30, in31);
DOTP_CONST_PAIR(in26, in21, cospi_24_64, cospi_8_64, in18, in29);
DOTP_CONST_PAIR(in27, in20, cospi_24_64, cospi_8_64, in19, in28);
ADD4(in16, in19, in17, in18, in30, in29, in31, in28, in27, in22, in21, in25);
DOTP_CONST_PAIR(in21, in22, cospi_28_64, cospi_4_64, in26, in24);
ADD2(in27, in26, in25, in24, in23, in20);
DOTP_CONST_PAIR(in20, in23, cospi_31_64, cospi_1_64, vec4, vec5);
FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
ST_SH(vec5, temp_ptr);
ST_SH(vec4, temp_ptr + 960);
SUB2(in27, in26, in25, in24, in22, in21);
DOTP_CONST_PAIR(in21, in22, cospi_15_64, cospi_17_64, vec5, vec4);
FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
ST_SH(vec5, temp_ptr + 448);
ST_SH(vec4, temp_ptr + 512);
SUB4(in17, in18, in16, in19, in31, in28, in30, in29, in23, in26, in24, in20);
DOTP_CONST_PAIR((-in23), in20, cospi_28_64, cospi_4_64, in27, in25);
SUB2(in26, in27, in24, in25, in23, in20);
DOTP_CONST_PAIR(in20, in23, cospi_23_64, cospi_9_64, vec4, vec5);
FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
ST_SH(vec4, temp_ptr + 704);
ST_SH(vec5, temp_ptr + 256);
ADD2(in26, in27, in24, in25, in22, in21);
DOTP_CONST_PAIR(in21, in22, cospi_7_64, cospi_25_64, vec4, vec5);
FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
ST_SH(vec4, temp_ptr + 192);
ST_SH(vec5, temp_ptr + 768);
LD_SH4(input + 16, 8, in22, in23, in20, in21);
LD_SH4(input + 80, 8, in26, in27, in24, in25);
in16 = in20;
in17 = in21;
DOTP_CONST_PAIR(-in16, in27, cospi_24_64, cospi_8_64, in20, in27);
DOTP_CONST_PAIR(-in17, in26, cospi_24_64, cospi_8_64, in21, in26);
SUB4(in23, in20, in22, in21, in25, in26, in24, in27, in28, in17, in18, in31);
DOTP_CONST_PAIR(in18, in17, cospi_12_64, cospi_20_64, in29, in30);
ADD2(in28, in29, in31, in30, in16, in19);
DOTP_CONST_PAIR(in19, in16, cospi_27_64, cospi_5_64, vec5, vec4);
FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
ST_SH(vec5, temp_ptr + 832);
ST_SH(vec4, temp_ptr + 128);
SUB2(in28, in29, in31, in30, in17, in18);
DOTP_CONST_PAIR(in18, in17, cospi_11_64, cospi_21_64, vec5, vec4);
FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
ST_SH(vec5, temp_ptr + 320);
ST_SH(vec4, temp_ptr + 640);
ADD4(in22, in21, in23, in20, in24, in27, in25, in26, in16, in29, in30, in19);
DOTP_CONST_PAIR(-in16, in19, cospi_12_64, cospi_20_64, in28, in31);
SUB2(in29, in28, in30, in31, in16, in19);
DOTP_CONST_PAIR(in19, in16, cospi_19_64, cospi_13_64, vec5, vec4);
FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
ST_SH(vec5, temp_ptr + 576);
ST_SH(vec4, temp_ptr + 384);
ADD2(in29, in28, in30, in31, in17, in18);
DOTP_CONST_PAIR(in18, in17, cospi_3_64, cospi_29_64, vec5, vec4);
FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
ST_SH(vec5, temp_ptr + 64);
ST_SH(vec4, temp_ptr + 896);
}
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;
}
}
int32_t vp8_mbblock_error_msa(MACROBLOCK *mb, int32_t dc)
{
BLOCK *be;
BLOCKD *bd;
int16_t *coeff_ptr, *dq_coeff_ptr;
int32_t err = 0;
uint32_t loop_cnt;
v8i16 coeff, coeff0, coeff1, coeff2, coeff3, coeff4;
v8i16 dq_coeff, dq_coeff2, dq_coeff3, dq_coeff4;
v4i32 diff0, diff1;
v2i64 err0, err1;
v16u8 zero = { 0 };
v16u8 mask0 = (v16u8)__msa_ldi_b(255);
if (1 == dc)
{
mask0 = (v16u8)__msa_insve_w((v4i32)mask0, 0, (v4i32)zero);
}
for (loop_cnt = 0; loop_cnt < 8; loop_cnt++)
{
be = &mb->block[2 * loop_cnt];
bd = &mb->e_mbd.block[2 * loop_cnt];
coeff_ptr = be->coeff;
dq_coeff_ptr = bd->dqcoeff;
coeff = LD_SH(coeff_ptr);
dq_coeff = LD_SH(dq_coeff_ptr);
coeff_ptr += 8;
dq_coeff_ptr += 8;
coeff2 = LD_SH(coeff_ptr);
dq_coeff2 = LD_SH(dq_coeff_ptr);
be = &mb->block[2 * loop_cnt + 1];
bd = &mb->e_mbd.block[2 * loop_cnt + 1];
coeff_ptr = be->coeff;
dq_coeff_ptr = bd->dqcoeff;
coeff3 = LD_SH(coeff_ptr);
dq_coeff3 = LD_SH(dq_coeff_ptr);
coeff_ptr += 8;
dq_coeff_ptr += 8;
coeff4 = LD_SH(coeff_ptr);
dq_coeff4 = LD_SH(dq_coeff_ptr);
ILVRL_H2_SH(coeff, dq_coeff, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
diff0 = (v4i32)__msa_bmnz_v(zero, (v16u8)diff0, mask0);
DOTP_SW2_SD(diff0, diff1, diff0, diff1, err0, err1);
ILVRL_H2_SH(coeff2, dq_coeff2, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
DPADD_SD2_SD(diff0, diff1, err0, err1);
err0 += __msa_splati_d(err0, 1);
err1 += __msa_splati_d(err1, 1);
err += __msa_copy_s_d(err0, 0);
err += __msa_copy_s_d(err1, 0);
ILVRL_H2_SH(coeff3, dq_coeff3, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
diff0 = (v4i32)__msa_bmnz_v(zero, (v16u8)diff0, mask0);
DOTP_SW2_SD(diff0, diff1, diff0, diff1, err0, err1);
ILVRL_H2_SH(coeff4, dq_coeff4, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
DPADD_SD2_SD(diff0, diff1, err0, err1);
err0 += __msa_splati_d(err0, 1);
err1 += __msa_splati_d(err1, 1);
err += __msa_copy_s_d(err0, 0);
err += __msa_copy_s_d(err1, 0);
}
return err;
}
int32_t vp8_mbuverror_msa(MACROBLOCK *mb)
{
BLOCK *be;
BLOCKD *bd;
int16_t *coeff_ptr, *dq_coeff_ptr;
int32_t err = 0;
uint32_t loop_cnt;
v8i16 coeff, coeff0, coeff1, coeff2, coeff3, coeff4;
v8i16 dq_coeff, dq_coeff2, dq_coeff3, dq_coeff4;
v4i32 diff0, diff1;
v2i64 err0, err1, err_dup0, err_dup1;
for (loop_cnt = 16; loop_cnt < 24; loop_cnt += 2)
{
be = &mb->block[loop_cnt];
bd = &mb->e_mbd.block[loop_cnt];
coeff_ptr = be->coeff;
dq_coeff_ptr = bd->dqcoeff;
coeff = LD_SH(coeff_ptr);
dq_coeff = LD_SH(dq_coeff_ptr);
coeff_ptr += 8;
dq_coeff_ptr += 8;
coeff2 = LD_SH(coeff_ptr);
dq_coeff2 = LD_SH(dq_coeff_ptr);
be = &mb->block[loop_cnt + 1];
bd = &mb->e_mbd.block[loop_cnt + 1];
coeff_ptr = be->coeff;
dq_coeff_ptr = bd->dqcoeff;
coeff3 = LD_SH(coeff_ptr);
dq_coeff3 = LD_SH(dq_coeff_ptr);
coeff_ptr += 8;
dq_coeff_ptr += 8;
coeff4 = LD_SH(coeff_ptr);
dq_coeff4 = LD_SH(dq_coeff_ptr);
ILVRL_H2_SH(coeff, dq_coeff, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
DOTP_SW2_SD(diff0, diff1, diff0, diff1, err0, err1);
ILVRL_H2_SH(coeff2, dq_coeff2, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
DPADD_SD2_SD(diff0, diff1, err0, err1);
err_dup0 = __msa_splati_d(err0, 1);
err_dup1 = __msa_splati_d(err1, 1);
ADD2(err0, err_dup0, err1, err_dup1, err0, err1);
err += __msa_copy_s_d(err0, 0);
err += __msa_copy_s_d(err1, 0);
ILVRL_H2_SH(coeff3, dq_coeff3, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
DOTP_SW2_SD(diff0, diff1, diff0, diff1, err0, err1);
ILVRL_H2_SH(coeff4, dq_coeff4, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
DPADD_SD2_SD(diff0, diff1, err0, err1);
err_dup0 = __msa_splati_d(err0, 1);
err_dup1 = __msa_splati_d(err1, 1);
ADD2(err0, err_dup0, err1, err_dup1, err0, err1);
err += __msa_copy_s_d(err0, 0);
err += __msa_copy_s_d(err1, 0);
}
return err;
}
