namespace {
void fdct4x4(int16_t *in, int16_t *out, uint8_t* /*dst*/,
int stride, int /*tx_type*/) {
- vp9_short_fdct4x4_c(in, out, stride);
+ vp9_fdct4x4_c(in, out, stride);
}
void idct4x4_add(int16_t* /*in*/, int16_t *out, uint8_t *dst,
int stride, int /*tx_type*/) {
prototype void vp9_short_fht16x16 "int16_t *InputData, int16_t *OutputData, int pitch, int tx_type"
specialize vp9_short_fht16x16 sse2
+prototype void vp9_fwht4x4 "int16_t *input, int16_t *output, int stride"
+specialize vp9_fwht4x4
+
+prototype void vp9_fdct4x4 "int16_t *input, int16_t *output, int stride"
+specialize vp9_fdct4x4 sse2
+
prototype void vp9_fdct8x8 "int16_t *input, int16_t *output, int stride"
specialize vp9_fdct8x8 sse2
-prototype void vp9_short_fdct4x4 "int16_t *InputData, int16_t *OutputData, int stride"
-specialize vp9_short_fdct4x4 sse2
+prototype void vp9_fdct16x16 "int16_t *input, int16_t *output, int stride"
+specialize vp9_fdct16x16 sse2
prototype void vp9_fdct32x32 "int16_t *input, int16_t *output, int stride"
specialize vp9_fdct32x32 sse2
prototype void vp9_fdct32x32_rd "int16_t *input, int16_t *output, int stride"
specialize vp9_fdct32x32_rd sse2
-prototype void vp9_fdct16x16 "int16_t *input, int16_t *output, int stride"
-specialize vp9_fdct16x16 sse2
-
-prototype void vp9_short_walsh4x4 "int16_t *InputData, int16_t *OutputData, int pitch"
-specialize vp9_short_walsh4x4
-
#
# Motion search
#
output[3] = dct_const_round_shift(temp2);
}
-void vp9_short_fdct4x4_c(int16_t *input, int16_t *output, int stride) {
+void vp9_fdct4x4_c(int16_t *input, int16_t *output, int stride) {
// The 2D transform is done with two passes which are actually pretty
// similar. In the first one, we transform the columns and transpose
// the results. In the second one, we transform the rows. To achieve that,
/* 4-point reversible, orthonormal Walsh-Hadamard in 3.5 adds, 0.5 shifts per
pixel. */
-void vp9_short_walsh4x4_c(int16_t *input, int16_t *output, int stride) {
+void vp9_fwht4x4_c(int16_t *input, int16_t *output, int stride) {
int i;
int a1, b1, c1, d1, e1;
int16_t *ip = input;
static void switch_lossless_mode(VP9_COMP *cpi, int lossless) {
if (lossless) {
// printf("Switching to lossless\n");
- cpi->mb.fwd_txm4x4 = vp9_short_walsh4x4;
+ cpi->mb.fwd_txm4x4 = vp9_fwht4x4;
cpi->mb.e_mbd.itxm_add = vp9_iwht4x4_add;
cpi->mb.optimize = 0;
cpi->common.lf.filter_level = 0;
cpi->common.tx_mode = ONLY_4X4;
} else {
// printf("Not lossless\n");
- cpi->mb.fwd_txm4x4 = vp9_short_fdct4x4;
+ cpi->mb.fwd_txm4x4 = vp9_fdct4x4;
cpi->mb.e_mbd.itxm_add = vp9_idct4x4_add;
}
}
sf->optimize_coefficients = 0;
}
- cpi->mb.fwd_txm4x4 = vp9_short_fdct4x4;
+ cpi->mb.fwd_txm4x4 = vp9_fdct4x4;
if (cpi->oxcf.lossless || cpi->mb.e_mbd.lossless) {
- cpi->mb.fwd_txm4x4 = vp9_short_walsh4x4;
+ cpi->mb.fwd_txm4x4 = vp9_fwht4x4;
}
if (cpi->sf.subpel_search_method == SUBPEL_ITERATIVE) {
#include "vp9/common/vp9_idct.h" // for cospi constants
#include "vpx_ports/mem.h"
-void vp9_short_fdct4x4_sse2(int16_t *input, int16_t *output, int stride) {
+void vp9_fdct4x4_sse2(int16_t *input, int16_t *output, int stride) {
// The 2D transform is done with two passes which are actually pretty
// similar. In the first one, we transform the columns and transpose
// the results. In the second one, we transform the rows. To achieve that,