|
|
|
@ -207,15 +207,16 @@ static void predictor_decompress_fir_adapt(int32_t *error_buffer, |
|
|
|
|
int val = 0; |
|
|
|
|
int error_val = error_buffer[i + 1]; |
|
|
|
|
int error_sign; |
|
|
|
|
int d = buffer_out[i - predictor_coef_num]; |
|
|
|
|
|
|
|
|
|
for (j = 0; j < predictor_coef_num; j++) { |
|
|
|
|
val += (buffer_out[i - j] - buffer_out[i - predictor_coef_num]) * |
|
|
|
|
val += (buffer_out[i - j] - d) * |
|
|
|
|
predictor_coef_table[j]; |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
val = (val + (1 << (predictor_quantitization - 1))) >> |
|
|
|
|
predictor_quantitization; |
|
|
|
|
val += buffer_out[i - predictor_coef_num] + error_val; |
|
|
|
|
val += d + error_val; |
|
|
|
|
|
|
|
|
|
buffer_out[i + 1] = sign_extend(val, readsamplesize); |
|
|
|
|
|
|
|
|
@ -224,7 +225,7 @@ static void predictor_decompress_fir_adapt(int32_t *error_buffer, |
|
|
|
|
if (error_sign) { |
|
|
|
|
for (j = predictor_coef_num - 1; j >= 0 && error_val * error_sign > 0; j--) { |
|
|
|
|
int sign; |
|
|
|
|
val = buffer_out[i - predictor_coef_num] - buffer_out[i - j]; |
|
|
|
|
val = d - buffer_out[i - j]; |
|
|
|
|
sign = sign_only(val) * error_sign; |
|
|
|
|
predictor_coef_table[j] -= sign; |
|
|
|
|
val *= sign; |
|
|
|
|