static void fdct4(const tran_low_t *input, tran_low_t *output) {
tran_high_t temp;
tran_low_t step[4];
// stage 0
range_check(input, 4, 14);
// stage 1
output[0] = input[0] + input[3];
output[1] = input[1] + input[2];
output[2] = input[1] - input[2];
output[3] = input[0] - input[3];
range_check(output, 4, 15);
// stage 2
temp = output[0] * cospi_16_64 + output[1] * cospi_16_64;
step[0] = (tran_low_t)fdct_round_shift(temp);
temp = output[1] * -cospi_16_64 + output[0] * cospi_16_64;
step[1] = (tran_low_t)fdct_round_shift(temp);
temp = output[2] * cospi_24_64 + output[3] * cospi_8_64;
step[2] = (tran_low_t)fdct_round_shift(temp);
temp = output[3] * cospi_24_64 + output[2] * -cospi_8_64;
step[3] = (tran_low_t)fdct_round_shift(temp);
range_check(step, 4, 16);
// stage 3
output[0] = step[0];
output[1] = step[2];
output[2] = step[1];
output[3] = step[3];
range_check(output, 4, 16);
}
static void fdct8(const tran_low_t *input, tran_low_t *output) {
tran_high_t s0, s1, s2, s3, s4, s5, s6, s7; // canbe16
tran_high_t t0, t1, t2, t3; // needs32
tran_high_t x0, x1, x2, x3; // canbe16
// stage 1
s0 = input[0] + input[7];
s1 = input[1] + input[6];
s2 = input[2] + input[5];
s3 = input[3] + input[4];
s4 = input[3] - input[4];
s5 = input[2] - input[5];
s6 = input[1] - input[6];
s7 = input[0] - input[7];
// fdct4(step, step);
x0 = s0 + s3;
x1 = s1 + s2;
x2 = s1 - s2;
x3 = s0 - s3;
t0 = (x0 + x1) * cospi_16_64;
t1 = (x0 - x1) * cospi_16_64;
t2 = x2 * cospi_24_64 + x3 * cospi_8_64;
t3 = -x2 * cospi_8_64 + x3 * cospi_24_64;
output[0] = (tran_low_t)fdct_round_shift(t0);
output[2] = (tran_low_t)fdct_round_shift(t2);
output[4] = (tran_low_t)fdct_round_shift(t1);
output[6] = (tran_low_t)fdct_round_shift(t3);
// Stage 2
t0 = (s6 - s5) * cospi_16_64;
t1 = (s6 + s5) * cospi_16_64;
t2 = (tran_low_t)fdct_round_shift(t0);
t3 = (tran_low_t)fdct_round_shift(t1);
// Stage 3
x0 = s4 + t2;
x1 = s4 - t2;
x2 = s7 - t3;
x3 = s7 + t3;
// Stage 4
t0 = x0 * cospi_28_64 + x3 * cospi_4_64;
t1 = x1 * cospi_12_64 + x2 * cospi_20_64;
t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
t3 = x3 * cospi_28_64 + x0 * -cospi_4_64;
output[1] = (tran_low_t)fdct_round_shift(t0);
output[3] = (tran_low_t)fdct_round_shift(t2);
output[5] = (tran_low_t)fdct_round_shift(t1);
output[7] = (tran_low_t)fdct_round_shift(t3);
}
static void fdct4(const tran_low_t *input, tran_low_t *output) {
tran_high_t step[4];
tran_high_t temp1, temp2;
step[0] = input[0] + input[3];
step[1] = input[1] + input[2];
step[2] = input[1] - input[2];
step[3] = input[0] - input[3];
temp1 = (step[0] + step[1]) * cospi_16_64;
temp2 = (step[0] - step[1]) * cospi_16_64;
output[0] = (tran_low_t)fdct_round_shift(temp1);
output[2] = (tran_low_t)fdct_round_shift(temp2);
temp1 = step[2] * cospi_24_64 + step[3] * cospi_8_64;
temp2 = -step[2] * cospi_8_64 + step[3] * cospi_24_64;
output[1] = (tran_low_t)fdct_round_shift(temp1);
output[3] = (tran_low_t)fdct_round_shift(temp2);
}
static void fadst4(const tran_low_t *input, tran_low_t *output) {
tran_high_t x0, x1, x2, x3;
tran_high_t s0, s1, s2, s3, s4, s5, s6, s7;
x0 = input[0];
x1 = input[1];
x2 = input[2];
x3 = input[3];
if (!(x0 | x1 | x2 | x3)) {
output[0] = output[1] = output[2] = output[3] = 0;
return;
}
s0 = sinpi_1_9 * x0;
s1 = sinpi_4_9 * x0;
s2 = sinpi_2_9 * x1;
s3 = sinpi_1_9 * x1;
s4 = sinpi_3_9 * x2;
s5 = sinpi_4_9 * x3;
s6 = sinpi_2_9 * x3;
s7 = x0 + x1 - x3;
x0 = s0 + s2 + s5;
x1 = sinpi_3_9 * s7;
x2 = s1 - s3 + s6;
x3 = s4;
s0 = x0 + x3;
s1 = x1;
s2 = x2 - x3;
s3 = x2 - x0 + x3;
// 1-D transform scaling factor is sqrt(2).
output[0] = (tran_low_t)fdct_round_shift(s0);
output[1] = (tran_low_t)fdct_round_shift(s1);
output[2] = (tran_low_t)fdct_round_shift(s2);
output[3] = (tran_low_t)fdct_round_shift(s3);
}
static void fdct16(const tran_low_t in[16], tran_low_t out[16]) {
tran_high_t step1[8]; // canbe16
tran_high_t step2[8]; // canbe16
tran_high_t step3[8]; // canbe16
tran_high_t input[8]; // canbe16
tran_high_t temp1, temp2; // needs32
// step 1
input[0] = in[0] + in[15];
input[1] = in[1] + in[14];
input[2] = in[2] + in[13];
input[3] = in[3] + in[12];
input[4] = in[4] + in[11];
input[5] = in[5] + in[10];
input[6] = in[6] + in[9];
input[7] = in[7] + in[8];
step1[0] = in[7] - in[8];
step1[1] = in[6] - in[9];
step1[2] = in[5] - in[10];
step1[3] = in[4] - in[11];
step1[4] = in[3] - in[12];
step1[5] = in[2] - in[13];
step1[6] = in[1] - in[14];
step1[7] = in[0] - in[15];
// fdct8(step, step);
{
tran_high_t s0, s1, s2, s3, s4, s5, s6, s7; // canbe16
tran_high_t t0, t1, t2, t3; // needs32
tran_high_t x0, x1, x2, x3; // canbe16
// stage 1
s0 = input[0] + input[7];
s1 = input[1] + input[6];
s2 = input[2] + input[5];
s3 = input[3] + input[4];
s4 = input[3] - input[4];
s5 = input[2] - input[5];
s6 = input[1] - input[6];
s7 = input[0] - input[7];
// fdct4(step, step);
x0 = s0 + s3;
x1 = s1 + s2;
x2 = s1 - s2;
x3 = s0 - s3;
t0 = (x0 + x1) * cospi_16_64;
t1 = (x0 - x1) * cospi_16_64;
t2 = x3 * cospi_8_64 + x2 * cospi_24_64;
t3 = x3 * cospi_24_64 - x2 * cospi_8_64;
out[0] = (tran_low_t)fdct_round_shift(t0);
out[4] = (tran_low_t)fdct_round_shift(t2);
out[8] = (tran_low_t)fdct_round_shift(t1);
out[12] = (tran_low_t)fdct_round_shift(t3);
// Stage 2
t0 = (s6 - s5) * cospi_16_64;
t1 = (s6 + s5) * cospi_16_64;
t2 = fdct_round_shift(t0);
t3 = fdct_round_shift(t1);
// Stage 3
x0 = s4 + t2;
x1 = s4 - t2;
x2 = s7 - t3;
x3 = s7 + t3;
// Stage 4
t0 = x0 * cospi_28_64 + x3 * cospi_4_64;
t1 = x1 * cospi_12_64 + x2 * cospi_20_64;
t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
t3 = x3 * cospi_28_64 + x0 * -cospi_4_64;
out[2] = (tran_low_t)fdct_round_shift(t0);
out[6] = (tran_low_t)fdct_round_shift(t2);
out[10] = (tran_low_t)fdct_round_shift(t1);
out[14] = (tran_low_t)fdct_round_shift(t3);
}
// step 2
temp1 = (step1[5] - step1[2]) * cospi_16_64;
temp2 = (step1[4] - step1[3]) * cospi_16_64;
step2[2] = fdct_round_shift(temp1);
step2[3] = fdct_round_shift(temp2);
temp1 = (step1[4] + step1[3]) * cospi_16_64;
temp2 = (step1[5] + step1[2]) * cospi_16_64;
step2[4] = fdct_round_shift(temp1);
step2[5] = fdct_round_shift(temp2);
// step 3
step3[0] = step1[0] + step2[3];
step3[1] = step1[1] + step2[2];
step3[2] = step1[1] - step2[2];
step3[3] = step1[0] - step2[3];
step3[4] = step1[7] - step2[4];
step3[5] = step1[6] - step2[5];
step3[6] = step1[6] + step2[5];
step3[7] = step1[7] + step2[4];
// step 4
temp1 = step3[1] * -cospi_8_64 + step3[6] * cospi_24_64;
temp2 = step3[2] * cospi_24_64 + step3[5] * cospi_8_64;
step2[1] = fdct_round_shift(temp1);
step2[2] = fdct_round_shift(temp2);
temp1 = step3[2] * cospi_8_64 - step3[5] * cospi_24_64;
temp2 = step3[1] * cospi_24_64 + step3[6] * cospi_8_64;
step2[5] = fdct_round_shift(temp1);
step2[6] = fdct_round_shift(temp2);
// step 5
step1[0] = step3[0] + step2[1];
step1[1] = step3[0] - step2[1];
step1[2] = step3[3] + step2[2];
step1[3] = step3[3] - step2[2];
step1[4] = step3[4] - step2[5];
step1[5] = step3[4] + step2[5];
step1[6] = step3[7] - step2[6];
step1[7] = step3[7] + step2[6];
// step 6
temp1 = step1[0] * cospi_30_64 + step1[7] * cospi_2_64;
temp2 = step1[1] * cospi_14_64 + step1[6] * cospi_18_64;
out[1] = (tran_low_t)fdct_round_shift(temp1);
out[9] = (tran_low_t)fdct_round_shift(temp2);
temp1 = step1[2] * cospi_22_64 + step1[5] * cospi_10_64;
temp2 = step1[3] * cospi_6_64 + step1[4] * cospi_26_64;
out[5] = (tran_low_t)fdct_round_shift(temp1);
out[13] = (tran_low_t)fdct_round_shift(temp2);
temp1 = step1[3] * -cospi_26_64 + step1[4] * cospi_6_64;
temp2 = step1[2] * -cospi_10_64 + step1[5] * cospi_22_64;
out[3] = (tran_low_t)fdct_round_shift(temp1);
out[11] = (tran_low_t)fdct_round_shift(temp2);
temp1 = step1[1] * -cospi_18_64 + step1[6] * cospi_14_64;
temp2 = step1[0] * -cospi_2_64 + step1[7] * cospi_30_64;
out[7] = (tran_low_t)fdct_round_shift(temp1);
out[15] = (tran_low_t)fdct_round_shift(temp2);
}
void vp9_fdct8x8_quant_c(const int16_t *input, int stride,
tran_low_t *coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t *round_ptr,
const int16_t *quant_ptr, tran_low_t *qcoeff_ptr,
tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr,
uint16_t *eob_ptr, const int16_t *scan,
const int16_t *iscan) {
int eob = -1;
int i, j;
tran_low_t intermediate[64];
(void)iscan;
// Transform columns
{
tran_low_t *output = intermediate;
tran_high_t s0, s1, s2, s3, s4, s5, s6, s7; // canbe16
tran_high_t t0, t1, t2, t3; // needs32
tran_high_t x0, x1, x2, x3; // canbe16
int i;
for (i = 0; i < 8; i++) {
// stage 1
s0 = (input[0 * stride] + input[7 * stride]) * 4;
s1 = (input[1 * stride] + input[6 * stride]) * 4;
s2 = (input[2 * stride] + input[5 * stride]) * 4;
s3 = (input[3 * stride] + input[4 * stride]) * 4;
s4 = (input[3 * stride] - input[4 * stride]) * 4;
s5 = (input[2 * stride] - input[5 * stride]) * 4;
s6 = (input[1 * stride] - input[6 * stride]) * 4;
s7 = (input[0 * stride] - input[7 * stride]) * 4;
// fdct4(step, step);
x0 = s0 + s3;
x1 = s1 + s2;
x2 = s1 - s2;
x3 = s0 - s3;
t0 = (x0 + x1) * cospi_16_64;
t1 = (x0 - x1) * cospi_16_64;
t2 = x2 * cospi_24_64 + x3 * cospi_8_64;
t3 = -x2 * cospi_8_64 + x3 * cospi_24_64;
output[0 * 8] = (tran_low_t)fdct_round_shift(t0);
output[2 * 8] = (tran_low_t)fdct_round_shift(t2);
output[4 * 8] = (tran_low_t)fdct_round_shift(t1);
output[6 * 8] = (tran_low_t)fdct_round_shift(t3);
// Stage 2
t0 = (s6 - s5) * cospi_16_64;
t1 = (s6 + s5) * cospi_16_64;
t2 = fdct_round_shift(t0);
t3 = fdct_round_shift(t1);
// Stage 3
x0 = s4 + t2;
x1 = s4 - t2;
x2 = s7 - t3;
x3 = s7 + t3;
// Stage 4
t0 = x0 * cospi_28_64 + x3 * cospi_4_64;
t1 = x1 * cospi_12_64 + x2 * cospi_20_64;
t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
t3 = x3 * cospi_28_64 + x0 * -cospi_4_64;
output[1 * 8] = (tran_low_t)fdct_round_shift(t0);
output[3 * 8] = (tran_low_t)fdct_round_shift(t2);
output[5 * 8] = (tran_low_t)fdct_round_shift(t1);
output[7 * 8] = (tran_low_t)fdct_round_shift(t3);
input++;
output++;
}
}
// Rows
for (i = 0; i < 8; ++i) {
fdct8(&intermediate[i * 8], &coeff_ptr[i * 8]);
for (j = 0; j < 8; ++j) coeff_ptr[j + i * 8] /= 2;
}
memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
if (!skip_block) {
// Quantization pass: All coefficients with index >= zero_flag are
// skippable. Note: zero_flag can be zero.
for (i = 0; i < n_coeffs; i++) {
const int rc = scan[i];
const int coeff = coeff_ptr[rc];
const int coeff_sign = (coeff >> 31);
const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
int tmp = clamp(abs_coeff + round_ptr[rc != 0], INT16_MIN, INT16_MAX);
tmp = (tmp * quant_ptr[rc != 0]) >> 16;
qcoeff_ptr[rc] = (tmp ^ coeff_sign) - coeff_sign;
dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0];
if (tmp) eob = i;
}
}
*eob_ptr = eob + 1;
}
static void fadst16(const tran_low_t *input, tran_low_t *output) {
tran_high_t s0, s1, s2, s3, s4, s5, s6, s7, s8;
tran_high_t s9, s10, s11, s12, s13, s14, s15;
tran_high_t x0 = input[15];
tran_high_t x1 = input[0];
tran_high_t x2 = input[13];
tran_high_t x3 = input[2];
tran_high_t x4 = input[11];
tran_high_t x5 = input[4];
tran_high_t x6 = input[9];
tran_high_t x7 = input[6];
tran_high_t x8 = input[7];
tran_high_t x9 = input[8];
tran_high_t x10 = input[5];
tran_high_t x11 = input[10];
tran_high_t x12 = input[3];
tran_high_t x13 = input[12];
tran_high_t x14 = input[1];
tran_high_t x15 = input[14];
// stage 1
s0 = x0 * cospi_1_64 + x1 * cospi_31_64;
s1 = x0 * cospi_31_64 - x1 * cospi_1_64;
s2 = x2 * cospi_5_64 + x3 * cospi_27_64;
s3 = x2 * cospi_27_64 - x3 * cospi_5_64;
s4 = x4 * cospi_9_64 + x5 * cospi_23_64;
s5 = x4 * cospi_23_64 - x5 * cospi_9_64;
s6 = x6 * cospi_13_64 + x7 * cospi_19_64;
s7 = x6 * cospi_19_64 - x7 * cospi_13_64;
s8 = x8 * cospi_17_64 + x9 * cospi_15_64;
s9 = x8 * cospi_15_64 - x9 * cospi_17_64;
s10 = x10 * cospi_21_64 + x11 * cospi_11_64;
s11 = x10 * cospi_11_64 - x11 * cospi_21_64;
s12 = x12 * cospi_25_64 + x13 * cospi_7_64;
s13 = x12 * cospi_7_64 - x13 * cospi_25_64;
s14 = x14 * cospi_29_64 + x15 * cospi_3_64;
s15 = x14 * cospi_3_64 - x15 * cospi_29_64;
x0 = fdct_round_shift(s0 + s8);
x1 = fdct_round_shift(s1 + s9);
x2 = fdct_round_shift(s2 + s10);
x3 = fdct_round_shift(s3 + s11);
x4 = fdct_round_shift(s4 + s12);
x5 = fdct_round_shift(s5 + s13);
x6 = fdct_round_shift(s6 + s14);
x7 = fdct_round_shift(s7 + s15);
x8 = fdct_round_shift(s0 - s8);
x9 = fdct_round_shift(s1 - s9);
x10 = fdct_round_shift(s2 - s10);
x11 = fdct_round_shift(s3 - s11);
x12 = fdct_round_shift(s4 - s12);
x13 = fdct_round_shift(s5 - s13);
x14 = fdct_round_shift(s6 - s14);
x15 = fdct_round_shift(s7 - s15);
// stage 2
s0 = x0;
s1 = x1;
s2 = x2;
s3 = x3;
s4 = x4;
s5 = x5;
s6 = x6;
s7 = x7;
s8 = x8 * cospi_4_64 + x9 * cospi_28_64;
s9 = x8 * cospi_28_64 - x9 * cospi_4_64;
s10 = x10 * cospi_20_64 + x11 * cospi_12_64;
s11 = x10 * cospi_12_64 - x11 * cospi_20_64;
s12 = -x12 * cospi_28_64 + x13 * cospi_4_64;
s13 = x12 * cospi_4_64 + x13 * cospi_28_64;
s14 = -x14 * cospi_12_64 + x15 * cospi_20_64;
s15 = x14 * cospi_20_64 + x15 * cospi_12_64;
x0 = s0 + s4;
x1 = s1 + s5;
x2 = s2 + s6;
x3 = s3 + s7;
x4 = s0 - s4;
x5 = s1 - s5;
x6 = s2 - s6;
x7 = s3 - s7;
x8 = fdct_round_shift(s8 + s12);
x9 = fdct_round_shift(s9 + s13);
x10 = fdct_round_shift(s10 + s14);
x11 = fdct_round_shift(s11 + s15);
x12 = fdct_round_shift(s8 - s12);
x13 = fdct_round_shift(s9 - s13);
x14 = fdct_round_shift(s10 - s14);
x15 = fdct_round_shift(s11 - s15);
// stage 3
s0 = x0;
s1 = x1;
s2 = x2;
s3 = x3;
s4 = x4 * cospi_8_64 + x5 * cospi_24_64;
s5 = x4 * cospi_24_64 - x5 * cospi_8_64;
s6 = -x6 * cospi_24_64 + x7 * cospi_8_64;
s7 = x6 * cospi_8_64 + x7 * cospi_24_64;
s8 = x8;
s9 = x9;
s10 = x10;
s11 = x11;
s12 = x12 * cospi_8_64 + x13 * cospi_24_64;
s13 = x12 * cospi_24_64 - x13 * cospi_8_64;
s14 = -x14 * cospi_24_64 + x15 * cospi_8_64;
s15 = x14 * cospi_8_64 + x15 * cospi_24_64;
x0 = s0 + s2;
x1 = s1 + s3;
x2 = s0 - s2;
x3 = s1 - s3;
x4 = fdct_round_shift(s4 + s6);
x5 = fdct_round_shift(s5 + s7);
x6 = fdct_round_shift(s4 - s6);
x7 = fdct_round_shift(s5 - s7);
x8 = s8 + s10;
x9 = s9 + s11;
x10 = s8 - s10;
x11 = s9 - s11;
x12 = fdct_round_shift(s12 + s14);
x13 = fdct_round_shift(s13 + s15);
x14 = fdct_round_shift(s12 - s14);
x15 = fdct_round_shift(s13 - s15);
// stage 4
s2 = (-cospi_16_64) * (x2 + x3);
s3 = cospi_16_64 * (x2 - x3);
s6 = cospi_16_64 * (x6 + x7);
s7 = cospi_16_64 * (-x6 + x7);
s10 = cospi_16_64 * (x10 + x11);
s11 = cospi_16_64 * (-x10 + x11);
s14 = (-cospi_16_64) * (x14 + x15);
s15 = cospi_16_64 * (x14 - x15);
x2 = fdct_round_shift(s2);
x3 = fdct_round_shift(s3);
x6 = fdct_round_shift(s6);
x7 = fdct_round_shift(s7);
x10 = fdct_round_shift(s10);
x11 = fdct_round_shift(s11);
x14 = fdct_round_shift(s14);
x15 = fdct_round_shift(s15);
output[0] = (tran_low_t)x0;
output[1] = (tran_low_t)-x8;
output[2] = (tran_low_t)x12;
output[3] = (tran_low_t)-x4;
output[4] = (tran_low_t)x6;
output[5] = (tran_low_t)x14;
output[6] = (tran_low_t)x10;
output[7] = (tran_low_t)x2;
output[8] = (tran_low_t)x3;
output[9] = (tran_low_t)x11;
output[10] = (tran_low_t)x15;
output[11] = (tran_low_t)x7;
output[12] = (tran_low_t)x5;
output[13] = (tran_low_t)-x13;
output[14] = (tran_low_t)x9;
output[15] = (tran_low_t)-x1;
}
static void fadst8(const tran_low_t *input, tran_low_t *output) {
tran_high_t s0, s1, s2, s3, s4, s5, s6, s7;
tran_high_t x0 = input[7];
tran_high_t x1 = input[0];
tran_high_t x2 = input[5];
tran_high_t x3 = input[2];
tran_high_t x4 = input[3];
tran_high_t x5 = input[4];
tran_high_t x6 = input[1];
tran_high_t x7 = input[6];
// stage 1
s0 = cospi_2_64 * x0 + cospi_30_64 * x1;
s1 = cospi_30_64 * x0 - cospi_2_64 * x1;
s2 = cospi_10_64 * x2 + cospi_22_64 * x3;
s3 = cospi_22_64 * x2 - cospi_10_64 * x3;
s4 = cospi_18_64 * x4 + cospi_14_64 * x5;
s5 = cospi_14_64 * x4 - cospi_18_64 * x5;
s6 = cospi_26_64 * x6 + cospi_6_64 * x7;
s7 = cospi_6_64 * x6 - cospi_26_64 * x7;
x0 = fdct_round_shift(s0 + s4);
x1 = fdct_round_shift(s1 + s5);
x2 = fdct_round_shift(s2 + s6);
x3 = fdct_round_shift(s3 + s7);
x4 = fdct_round_shift(s0 - s4);
x5 = fdct_round_shift(s1 - s5);
x6 = fdct_round_shift(s2 - s6);
x7 = fdct_round_shift(s3 - s7);
// stage 2
s0 = x0;
s1 = x1;
s2 = x2;
s3 = x3;
s4 = cospi_8_64 * x4 + cospi_24_64 * x5;
s5 = cospi_24_64 * x4 - cospi_8_64 * x5;
s6 = -cospi_24_64 * x6 + cospi_8_64 * x7;
s7 = cospi_8_64 * x6 + cospi_24_64 * x7;
x0 = s0 + s2;
x1 = s1 + s3;
x2 = s0 - s2;
x3 = s1 - s3;
x4 = fdct_round_shift(s4 + s6);
x5 = fdct_round_shift(s5 + s7);
x6 = fdct_round_shift(s4 - s6);
x7 = fdct_round_shift(s5 - s7);
// stage 3
s2 = cospi_16_64 * (x2 + x3);
s3 = cospi_16_64 * (x2 - x3);
s6 = cospi_16_64 * (x6 + x7);
s7 = cospi_16_64 * (x6 - x7);
x2 = fdct_round_shift(s2);
x3 = fdct_round_shift(s3);
x6 = fdct_round_shift(s6);
x7 = fdct_round_shift(s7);
output[0] = (tran_low_t)x0;
output[1] = (tran_low_t)-x4;
output[2] = (tran_low_t)x6;
output[3] = (tran_low_t)-x2;
output[4] = (tran_low_t)x3;
output[5] = (tran_low_t)-x7;
output[6] = (tran_low_t)x5;
output[7] = (tran_low_t)-x1;
}
static INLINE void fwd_txfm2d_c(const int16_t *input, int32_t *output,
const int stride, const TXFM_2D_FLIP_CFG *cfg,
int32_t *buf, int bd) {
int c, r;
// Note when assigning txfm_size_col, we use the txfm_size from the
// row configuration and vice versa. This is intentionally done to
// accurately perform rectangular transforms. When the transform is
// rectangular, the number of columns will be the same as the
// txfm_size stored in the row cfg struct. It will make no difference
// for square transforms.
const int txfm_size_col = cfg->row_cfg->txfm_size;
const int txfm_size_row = cfg->col_cfg->txfm_size;
// Take the shift from the larger dimension in the rectangular case.
const int8_t *shift = (txfm_size_col > txfm_size_row) ? cfg->row_cfg->shift
: cfg->col_cfg->shift;
int8_t stage_range_col[MAX_TXFM_STAGE_NUM];
int8_t stage_range_row[MAX_TXFM_STAGE_NUM];
assert(cfg->col_cfg->stage_num <= MAX_TXFM_STAGE_NUM);
assert(cfg->row_cfg->stage_num <= MAX_TXFM_STAGE_NUM);
av1_gen_fwd_stage_range(stage_range_col, stage_range_row, cfg, bd);
const int8_t *cos_bit_col = cfg->col_cfg->cos_bit;
const int8_t *cos_bit_row = cfg->row_cfg->cos_bit;
const TxfmFunc txfm_func_col = fwd_txfm_type_to_func(cfg->col_cfg->txfm_type);
const TxfmFunc txfm_func_row = fwd_txfm_type_to_func(cfg->row_cfg->txfm_type);
// use output buffer as temp buffer
int32_t *temp_in = output;
int32_t *temp_out = output + txfm_size_row;
// Columns
for (c = 0; c < txfm_size_col; ++c) {
if (cfg->ud_flip == 0) {
for (r = 0; r < txfm_size_row; ++r) temp_in[r] = input[r * stride + c];
} else {
for (r = 0; r < txfm_size_row; ++r)
// flip upside down
temp_in[r] = input[(txfm_size_row - r - 1) * stride + c];
}
round_shift_array(temp_in, txfm_size_row, -shift[0]);
// Multiply everything by Sqrt2 on the larger dimension if the
// transform is rectangular
if (txfm_size_col > txfm_size_row) {
for (r = 0; r < txfm_size_row; ++r)
temp_in[r] = (int32_t)fdct_round_shift(temp_in[r] * Sqrt2);
}
txfm_func_col(temp_in, temp_out, cos_bit_col, stage_range_col);
round_shift_array(temp_out, txfm_size_row, -shift[1]);
if (cfg->lr_flip == 0) {
for (r = 0; r < txfm_size_row; ++r)
buf[r * txfm_size_col + c] = temp_out[r];
} else {
for (r = 0; r < txfm_size_row; ++r)
// flip from left to right
buf[r * txfm_size_col + (txfm_size_col - c - 1)] = temp_out[r];
}
}
// Rows
for (r = 0; r < txfm_size_row; ++r) {
// Multiply everything by Sqrt2 on the larger dimension if the
// transform is rectangular
if (txfm_size_row > txfm_size_col) {
for (c = 0; c < txfm_size_col; ++c)
buf[r * txfm_size_col + c] =
(int32_t)fdct_round_shift(buf[r * txfm_size_col + c] * Sqrt2);
}
txfm_func_row(buf + r * txfm_size_col, output + r * txfm_size_col,
cos_bit_row, stage_range_row);
round_shift_array(output + r * txfm_size_col, txfm_size_col, -shift[2]);
}
}
static void fdct8(const tran_low_t *input, tran_low_t *output) {
tran_high_t temp;
tran_low_t step[8];
// stage 0
range_check(input, 8, 13);
// stage 1
output[0] = input[0] + input[7];
output[1] = input[1] + input[6];
output[2] = input[2] + input[5];
output[3] = input[3] + input[4];
output[4] = input[3] - input[4];
output[5] = input[2] - input[5];
output[6] = input[1] - input[6];
output[7] = input[0] - input[7];
range_check(output, 8, 14);
// stage 2
step[0] = output[0] + output[3];
step[1] = output[1] + output[2];
step[2] = output[1] - output[2];
step[3] = output[0] - output[3];
step[4] = output[4];
temp = output[5] * -cospi_16_64 + output[6] * cospi_16_64;
step[5] = (tran_low_t)fdct_round_shift(temp);
temp = output[6] * cospi_16_64 + output[5] * cospi_16_64;
step[6] = (tran_low_t)fdct_round_shift(temp);
step[7] = output[7];
range_check(step, 8, 15);
// stage 3
temp = step[0] * cospi_16_64 + step[1] * cospi_16_64;
output[0] = (tran_low_t)fdct_round_shift(temp);
temp = step[1] * -cospi_16_64 + step[0] * cospi_16_64;
output[1] = (tran_low_t)fdct_round_shift(temp);
temp = step[2] * cospi_24_64 + step[3] * cospi_8_64;
output[2] = (tran_low_t)fdct_round_shift(temp);
temp = step[3] * cospi_24_64 + step[2] * -cospi_8_64;
output[3] = (tran_low_t)fdct_round_shift(temp);
output[4] = step[4] + step[5];
output[5] = step[4] - step[5];
output[6] = step[7] - step[6];
output[7] = step[7] + step[6];
range_check(output, 8, 16);
// stage 4
step[0] = output[0];
step[1] = output[1];
step[2] = output[2];
step[3] = output[3];
temp = output[4] * cospi_28_64 + output[7] * cospi_4_64;
step[4] = (tran_low_t)fdct_round_shift(temp);
temp = output[5] * cospi_12_64 + output[6] * cospi_20_64;
step[5] = (tran_low_t)fdct_round_shift(temp);
temp = output[6] * cospi_12_64 + output[5] * -cospi_20_64;
step[6] = (tran_low_t)fdct_round_shift(temp);
temp = output[7] * cospi_28_64 + output[4] * -cospi_4_64;
step[7] = (tran_low_t)fdct_round_shift(temp);
range_check(step, 8, 16);
// stage 5
output[0] = step[0];
output[1] = step[4];
output[2] = step[2];
output[3] = step[6];
output[4] = step[1];
output[5] = step[5];
output[6] = step[3];
output[7] = step[7];
range_check(output, 8, 16);
}
static void fdct16(const tran_low_t *input, tran_low_t *output) {
tran_high_t temp;
tran_low_t step[16];
// stage 0
range_check(input, 16, 13);
// stage 1
output[0] = input[0] + input[15];
output[1] = input[1] + input[14];
output[2] = input[2] + input[13];
output[3] = input[3] + input[12];
output[4] = input[4] + input[11];
output[5] = input[5] + input[10];
output[6] = input[6] + input[9];
output[7] = input[7] + input[8];
output[8] = input[7] - input[8];
output[9] = input[6] - input[9];
output[10] = input[5] - input[10];
output[11] = input[4] - input[11];
output[12] = input[3] - input[12];
output[13] = input[2] - input[13];
output[14] = input[1] - input[14];
output[15] = input[0] - input[15];
range_check(output, 16, 14);
// stage 2
step[0] = output[0] + output[7];
step[1] = output[1] + output[6];
step[2] = output[2] + output[5];
step[3] = output[3] + output[4];
step[4] = output[3] - output[4];
step[5] = output[2] - output[5];
step[6] = output[1] - output[6];
step[7] = output[0] - output[7];
step[8] = output[8];
step[9] = output[9];
temp = output[10] * -cospi_16_64 + output[13] * cospi_16_64;
step[10] = (tran_low_t)fdct_round_shift(temp);
temp = output[11] * -cospi_16_64 + output[12] * cospi_16_64;
step[11] = (tran_low_t)fdct_round_shift(temp);
temp = output[12] * cospi_16_64 + output[11] * cospi_16_64;
step[12] = (tran_low_t)fdct_round_shift(temp);
temp = output[13] * cospi_16_64 + output[10] * cospi_16_64;
step[13] = (tran_low_t)fdct_round_shift(temp);
step[14] = output[14];
step[15] = output[15];
range_check(step, 16, 15);
// stage 3
output[0] = step[0] + step[3];
output[1] = step[1] + step[2];
output[2] = step[1] - step[2];
output[3] = step[0] - step[3];
output[4] = step[4];
temp = step[5] * -cospi_16_64 + step[6] * cospi_16_64;
output[5] = (tran_low_t)fdct_round_shift(temp);
temp = step[6] * cospi_16_64 + step[5] * cospi_16_64;
output[6] = (tran_low_t)fdct_round_shift(temp);
output[7] = step[7];
output[8] = step[8] + step[11];
output[9] = step[9] + step[10];
output[10] = step[9] - step[10];
output[11] = step[8] - step[11];
output[12] = step[15] - step[12];
output[13] = step[14] - step[13];
output[14] = step[14] + step[13];
output[15] = step[15] + step[12];
range_check(output, 16, 16);
// stage 4
temp = output[0] * cospi_16_64 + output[1] * cospi_16_64;
step[0] = (tran_low_t)fdct_round_shift(temp);
temp = output[1] * -cospi_16_64 + output[0] * cospi_16_64;
step[1] = (tran_low_t)fdct_round_shift(temp);
temp = output[2] * cospi_24_64 + output[3] * cospi_8_64;
step[2] = (tran_low_t)fdct_round_shift(temp);
temp = output[3] * cospi_24_64 + output[2] * -cospi_8_64;
step[3] = (tran_low_t)fdct_round_shift(temp);
step[4] = output[4] + output[5];
step[5] = output[4] - output[5];
step[6] = output[7] - output[6];
step[7] = output[7] + output[6];
step[8] = output[8];
temp = output[9] * -cospi_8_64 + output[14] * cospi_24_64;
step[9] = (tran_low_t)fdct_round_shift(temp);
temp = output[10] * -cospi_24_64 + output[13] * -cospi_8_64;
step[10] = (tran_low_t)fdct_round_shift(temp);
step[11] = output[11];
step[12] = output[12];
temp = output[13] * cospi_24_64 + output[10] * -cospi_8_64;
step[13] = (tran_low_t)fdct_round_shift(temp);
temp = output[14] * cospi_8_64 + output[9] * cospi_24_64;
step[14] = (tran_low_t)fdct_round_shift(temp);
step[15] = output[15];
range_check(step, 16, 16);
// stage 5
output[0] = step[0];
output[1] = step[1];
output[2] = step[2];
output[3] = step[3];
temp = step[4] * cospi_28_64 + step[7] * cospi_4_64;
output[4] = (tran_low_t)fdct_round_shift(temp);
temp = step[5] * cospi_12_64 + step[6] * cospi_20_64;
output[5] = (tran_low_t)fdct_round_shift(temp);
temp = step[6] * cospi_12_64 + step[5] * -cospi_20_64;
output[6] = (tran_low_t)fdct_round_shift(temp);
temp = step[7] * cospi_28_64 + step[4] * -cospi_4_64;
output[7] = (tran_low_t)fdct_round_shift(temp);
output[8] = step[8] + step[9];
output[9] = step[8] - step[9];
output[10] = step[11] - step[10];
output[11] = step[11] + step[10];
output[12] = step[12] + step[13];
output[13] = step[12] - step[13];
output[14] = step[15] - step[14];
output[15] = step[15] + step[14];
range_check(output, 16, 16);
// stage 6
step[0] = output[0];
step[1] = output[1];
step[2] = output[2];
step[3] = output[3];
step[4] = output[4];
step[5] = output[5];
step[6] = output[6];
step[7] = output[7];
temp = output[8] * cospi_30_64 + output[15] * cospi_2_64;
step[8] = (tran_low_t)fdct_round_shift(temp);
temp = output[9] * cospi_14_64 + output[14] * cospi_18_64;
step[9] = (tran_low_t)fdct_round_shift(temp);
temp = output[10] * cospi_22_64 + output[13] * cospi_10_64;
step[10] = (tran_low_t)fdct_round_shift(temp);
temp = output[11] * cospi_6_64 + output[12] * cospi_26_64;
step[11] = (tran_low_t)fdct_round_shift(temp);
temp = output[12] * cospi_6_64 + output[11] * -cospi_26_64;
step[12] = (tran_low_t)fdct_round_shift(temp);
temp = output[13] * cospi_22_64 + output[10] * -cospi_10_64;
step[13] = (tran_low_t)fdct_round_shift(temp);
temp = output[14] * cospi_14_64 + output[9] * -cospi_18_64;
step[14] = (tran_low_t)fdct_round_shift(temp);
temp = output[15] * cospi_30_64 + output[8] * -cospi_2_64;
step[15] = (tran_low_t)fdct_round_shift(temp);
range_check(step, 16, 16);
// stage 7
output[0] = step[0];
output[1] = step[8];
output[2] = step[4];
output[3] = step[12];
output[4] = step[2];
output[5] = step[10];
output[6] = step[6];
output[7] = step[14];
output[8] = step[1];
output[9] = step[9];
output[10] = step[5];
output[11] = step[13];
output[12] = step[3];
output[13] = step[11];
output[14] = step[7];
output[15] = step[15];
range_check(output, 16, 16);
}
void vp10_fdct8x8_quant_c(const int16_t *input, int stride,
tran_low_t *coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t *zbin_ptr,
const int16_t *round_ptr, const int16_t *quant_ptr,
const int16_t *quant_shift_ptr,
tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
const int16_t *dequant_ptr, uint16_t *eob_ptr,
const int16_t *scan, const int16_t *iscan
#if CONFIG_AOM_QM
,
const qm_val_t *qm_ptr, const qm_val_t *iqm_ptr
#endif
) {
int eob = -1;
int i, j;
tran_low_t intermediate[64];
// Transform columns
{
tran_low_t *output = intermediate;
tran_high_t s0, s1, s2, s3, s4, s5, s6, s7; // canbe16
tran_high_t t0, t1, t2, t3; // needs32
tran_high_t x0, x1, x2, x3; // canbe16
int i;
for (i = 0; i < 8; i++) {
// stage 1
s0 = (input[0 * stride] + input[7 * stride]) * 4;
s1 = (input[1 * stride] + input[6 * stride]) * 4;
s2 = (input[2 * stride] + input[5 * stride]) * 4;
s3 = (input[3 * stride] + input[4 * stride]) * 4;
s4 = (input[3 * stride] - input[4 * stride]) * 4;
s5 = (input[2 * stride] - input[5 * stride]) * 4;
s6 = (input[1 * stride] - input[6 * stride]) * 4;
s7 = (input[0 * stride] - input[7 * stride]) * 4;
// fdct4(step, step);
x0 = s0 + s3;
x1 = s1 + s2;
x2 = s1 - s2;
x3 = s0 - s3;
t0 = (x0 + x1) * cospi_16_64;
t1 = (x0 - x1) * cospi_16_64;
t2 = x2 * cospi_24_64 + x3 * cospi_8_64;
t3 = -x2 * cospi_8_64 + x3 * cospi_24_64;
output[0 * 8] = (tran_low_t)fdct_round_shift(t0);
output[2 * 8] = (tran_low_t)fdct_round_shift(t2);
output[4 * 8] = (tran_low_t)fdct_round_shift(t1);
output[6 * 8] = (tran_low_t)fdct_round_shift(t3);
// stage 2
t0 = (s6 - s5) * cospi_16_64;
t1 = (s6 + s5) * cospi_16_64;
t2 = fdct_round_shift(t0);
t3 = fdct_round_shift(t1);
// stage 3
x0 = s4 + t2;
x1 = s4 - t2;
x2 = s7 - t3;
x3 = s7 + t3;
// stage 4
t0 = x0 * cospi_28_64 + x3 * cospi_4_64;
t1 = x1 * cospi_12_64 + x2 * cospi_20_64;
t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
t3 = x3 * cospi_28_64 + x0 * -cospi_4_64;
output[1 * 8] = (tran_low_t)fdct_round_shift(t0);
output[3 * 8] = (tran_low_t)fdct_round_shift(t2);
output[5 * 8] = (tran_low_t)fdct_round_shift(t1);
output[7 * 8] = (tran_low_t)fdct_round_shift(t3);
input++;
output++;
}
}
// Rows
for (i = 0; i < 8; ++i) {
fdct8(&intermediate[i * 8], &coeff_ptr[i * 8]);
for (j = 0; j < 8; ++j) coeff_ptr[j + i * 8] /= 2;
}
// TODO(jingning) Decide the need of these arguments after the
// quantization process is completed.
(void)zbin_ptr;
(void)quant_shift_ptr;
(void)iscan;
memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
if (!skip_block) {
// Quantization pass: All coefficients with index >= zero_flag are
// skippable. Note: zero_flag can be zero.
for (i = 0; i < n_coeffs; i++) {
const int rc = scan[i];
const int coeff = coeff_ptr[rc];
#if CONFIG_AOM_QM
const qm_val_t wt = qm_ptr[rc];
const qm_val_t iwt = iqm_ptr[rc];
const int dequant =
(dequant_ptr[rc != 0] * iwt + (1 << (AOM_QM_BITS - 1))) >>
AOM_QM_BITS;
#endif
const int coeff_sign = (coeff >> 31);
const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
int64_t tmp = clamp(abs_coeff + round_ptr[rc != 0], INT16_MIN, INT16_MAX);
int tmp32;
#if CONFIG_AOM_QM
tmp32 = (tmp * quant_ptr[rc != 0] * wt) >> (16 + AOM_QM_BITS);
qcoeff_ptr[rc] = (tmp32 ^ coeff_sign) - coeff_sign;
dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant;
#else
tmp32 = (tmp * quant_ptr[rc != 0]) >> 16;
qcoeff_ptr[rc] = (tmp32 ^ coeff_sign) - coeff_sign;
dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0];
#endif
if (tmp32) eob = i;
}
}
*eob_ptr = eob + 1;
}
