generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Merge changes I6ef79702,Id332c641,I354b5d22,I84438013

Browse Source 
* changes:
  Use common transpose for vpx_idct32x32_1024_add_neon
  Use common transpose for vpx_idct8x8_[12|64]_add_neon
  Use common transpose for vp9_iht8x8_add_neon
  Use common transpose for vpx_idct16x16_[10|256]_add_neon
This commit is contained in:
Johann Koenig
2016-08-04 22:30:46 +00:00
committed by Gerrit Code Review
parent 17720b60bb 0325b95938
commit 57f49db81f
5 changed files with 39 additions and 272 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
test/dct16x16_test.cc
View File

@@ -850,7 +850,7 @@ INSTANTIATE_TEST_CASE_P(C, PartialTrans16x16Test,
VPX_BITS_8)));
#endif // CONFIG_VP9_HIGHBITDEPTH
#if HAVE_NEON_ASM && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
#if HAVE_NEON && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
INSTANTIATE_TEST_CASE_P(
NEON, Trans16x16DCT,
::testing::Values(make_tuple(&vpx_fdct16x16_c, &vpx_idct16x16_256_add_neon,

81
vp9/common/arm/neon/vp9_iht8x8_add_neon.c
View File

@@ -14,6 +14,7 @@
#include "./vp9_rtcd.h"
#include "./vpx_config.h"
#include "vp9/common/vp9_common.h"
#include "vpx_dsp/arm/transpose_neon.h"
static int16_t cospi_2_64 = 16305;
static int16_t cospi_4_64 = 16069;
@@ -31,70 +32,6 @@ static int16_t cospi_26_64 = 4756;
static int16_t cospi_28_64 = 3196;
static int16_t cospi_30_64 = 1606;
static INLINE void TRANSPOSE8X8(int16x8_t *q8s16, int16x8_t *q9s16,
int16x8_t *q10s16, int16x8_t *q11s16,
int16x8_t *q12s16, int16x8_t *q13s16,
int16x8_t *q14s16, int16x8_t *q15s16) {
int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16;
int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16;
int32x4x2_t q0x2s32, q1x2s32, q2x2s32, q3x2s32;
int16x8x2_t q0x2s16, q1x2s16, q2x2s16, q3x2s16;
d16s16 = vget_low_s16(*q8s16);
d17s16 = vget_high_s16(*q8s16);
d18s16 = vget_low_s16(*q9s16);
d19s16 = vget_high_s16(*q9s16);
d20s16 = vget_low_s16(*q10s16);
d21s16 = vget_high_s16(*q10s16);
d22s16 = vget_low_s16(*q11s16);
d23s16 = vget_high_s16(*q11s16);
d24s16 = vget_low_s16(*q12s16);
d25s16 = vget_high_s16(*q12s16);
d26s16 = vget_low_s16(*q13s16);
d27s16 = vget_high_s16(*q13s16);
d28s16 = vget_low_s16(*q14s16);
d29s16 = vget_high_s16(*q14s16);
d30s16 = vget_low_s16(*q15s16);
d31s16 = vget_high_s16(*q15s16);
*q8s16 = vcombine_s16(d16s16, d24s16); // vswp d17, d24
*q9s16 = vcombine_s16(d18s16, d26s16); // vswp d19, d26
*q10s16 = vcombine_s16(d20s16, d28s16); // vswp d21, d28
*q11s16 = vcombine_s16(d22s16, d30s16); // vswp d23, d30
*q12s16 = vcombine_s16(d17s16, d25s16);
*q13s16 = vcombine_s16(d19s16, d27s16);
*q14s16 = vcombine_s16(d21s16, d29s16);
*q15s16 = vcombine_s16(d23s16, d31s16);
q0x2s32 =
vtrnq_s32(vreinterpretq_s32_s16(*q8s16), vreinterpretq_s32_s16(*q10s16));
q1x2s32 =
vtrnq_s32(vreinterpretq_s32_s16(*q9s16), vreinterpretq_s32_s16(*q11s16));
q2x2s32 =
vtrnq_s32(vreinterpretq_s32_s16(*q12s16), vreinterpretq_s32_s16(*q14s16));
q3x2s32 =
vtrnq_s32(vreinterpretq_s32_s16(*q13s16), vreinterpretq_s32_s16(*q15s16));
q0x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[0]), // q8
vreinterpretq_s16_s32(q1x2s32.val[0])); // q9
q1x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[1]), // q10
vreinterpretq_s16_s32(q1x2s32.val[1])); // q11
q2x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[0]), // q12
vreinterpretq_s16_s32(q3x2s32.val[0])); // q13
q3x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[1]), // q14
vreinterpretq_s16_s32(q3x2s32.val[1])); // q15
*q8s16 = q0x2s16.val[0];
*q9s16 = q0x2s16.val[1];
*q10s16 = q1x2s16.val[0];
*q11s16 = q1x2s16.val[1];
*q12s16 = q2x2s16.val[0];
*q13s16 = q2x2s16.val[1];
*q14s16 = q3x2s16.val[0];
*q15s16 = q3x2s16.val[1];
return;
}
static INLINE void IDCT8x8_1D(int16x8_t *q8s16, int16x8_t *q9s16,
int16x8_t *q10s16, int16x8_t *q11s16,
int16x8_t *q12s16, int16x8_t *q13s16,
@@ -489,8 +426,8 @@ void vp9_iht8x8_64_add_neon(const tran_low_t *input, uint8_t *dest,
q14s16 = vld1q_s16(input + 8 * 6);
q15s16 = vld1q_s16(input + 8 * 7);
TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
transpose_s16_8x8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
switch (tx_type) {
case 0: // idct_idct is not supported. Fall back to C
@@ -506,8 +443,8 @@ void vp9_iht8x8_64_add_neon(const tran_low_t *input, uint8_t *dest,
&q15s16);
// transpose the matrix
TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
transpose_s16_8x8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16,
&q14s16, &q15s16);
// generate IADST constants
// GENERATE_IADST_CONSTANTS
@@ -525,8 +462,8 @@ void vp9_iht8x8_64_add_neon(const tran_low_t *input, uint8_t *dest,
&q15s16);
// transpose the matrix
TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
transpose_s16_8x8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16,
&q14s16, &q15s16);
// generate IDCT constants
// GENERATE_IDCT_CONSTANTS
@@ -544,8 +481,8 @@ void vp9_iht8x8_64_add_neon(const tran_low_t *input, uint8_t *dest,
&q15s16);
// transpose the matrix
TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
transpose_s16_8x8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16,
&q14s16, &q15s16);
// then transform columns
IADST8X8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,

81
vpx_dsp/arm/idct16x16_add_neon.c
View File

@@ -11,72 +11,9 @@
#include <arm_neon.h>
#include "./vpx_config.h"
#include "vpx_dsp/arm/transpose_neon.h"
#include "vpx_dsp/txfm_common.h"
static INLINE void TRANSPOSE8X8(int16x8_t *q8s16, int16x8_t *q9s16,
int16x8_t *q10s16, int16x8_t *q11s16,
int16x8_t *q12s16, int16x8_t *q13s16,
int16x8_t *q14s16, int16x8_t *q15s16) {
int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16;
int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16;
int32x4x2_t q0x2s32, q1x2s32, q2x2s32, q3x2s32;
int16x8x2_t q0x2s16, q1x2s16, q2x2s16, q3x2s16;
d16s16 = vget_low_s16(*q8s16);
d17s16 = vget_high_s16(*q8s16);
d18s16 = vget_low_s16(*q9s16);
d19s16 = vget_high_s16(*q9s16);
d20s16 = vget_low_s16(*q10s16);
d21s16 = vget_high_s16(*q10s16);
d22s16 = vget_low_s16(*q11s16);
d23s16 = vget_high_s16(*q11s16);
d24s16 = vget_low_s16(*q12s16);
d25s16 = vget_high_s16(*q12s16);
d26s16 = vget_low_s16(*q13s16);
d27s16 = vget_high_s16(*q13s16);
d28s16 = vget_low_s16(*q14s16);
d29s16 = vget_high_s16(*q14s16);
d30s16 = vget_low_s16(*q15s16);
d31s16 = vget_high_s16(*q15s16);
*q8s16 = vcombine_s16(d16s16, d24s16); // vswp d17, d24
*q9s16 = vcombine_s16(d18s16, d26s16); // vswp d19, d26
*q10s16 = vcombine_s16(d20s16, d28s16); // vswp d21, d28
*q11s16 = vcombine_s16(d22s16, d30s16); // vswp d23, d30
*q12s16 = vcombine_s16(d17s16, d25s16);
*q13s16 = vcombine_s16(d19s16, d27s16);
*q14s16 = vcombine_s16(d21s16, d29s16);
*q15s16 = vcombine_s16(d23s16, d31s16);
q0x2s32 =
vtrnq_s32(vreinterpretq_s32_s16(*q8s16), vreinterpretq_s32_s16(*q10s16));
q1x2s32 =
vtrnq_s32(vreinterpretq_s32_s16(*q9s16), vreinterpretq_s32_s16(*q11s16));
q2x2s32 =
vtrnq_s32(vreinterpretq_s32_s16(*q12s16), vreinterpretq_s32_s16(*q14s16));
q3x2s32 =
vtrnq_s32(vreinterpretq_s32_s16(*q13s16), vreinterpretq_s32_s16(*q15s16));
q0x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[0]), // q8
vreinterpretq_s16_s32(q1x2s32.val[0])); // q9
q1x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[1]), // q10
vreinterpretq_s16_s32(q1x2s32.val[1])); // q11
q2x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[0]), // q12
vreinterpretq_s16_s32(q3x2s32.val[0])); // q13
q3x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[1]), // q14
vreinterpretq_s16_s32(q3x2s32.val[1])); // q15
*q8s16 = q0x2s16.val[0];
*q9s16 = q0x2s16.val[1];
*q10s16 = q1x2s16.val[0];
*q11s16 = q1x2s16.val[1];
*q12s16 = q2x2s16.val[0];
*q13s16 = q2x2s16.val[1];
*q14s16 = q3x2s16.val[0];
*q15s16 = q3x2s16.val[1];
return;
}
void vpx_idct16x16_256_add_neon_pass1(int16_t *in, int16_t *out,
int output_stride) {
int16x4_t d0s16, d1s16, d2s16, d3s16;
@@ -115,8 +52,8 @@ void vpx_idct16x16_256_add_neon_pass1(int16_t *in, int16_t *out,
q0x2s16 = vld2q_s16(in);
q15s16 = q0x2s16.val[0];
TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
transpose_s16_8x8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
d16s16 = vget_low_s16(q8s16);
d17s16 = vget_high_s16(q8s16);
@@ -356,8 +293,8 @@ void vpx_idct16x16_256_add_neon_pass2(int16_t *src, int16_t *out,
q0x2s16 = vld2q_s16(src);
q15s16 = q0x2s16.val[0];
TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
transpose_s16_8x8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
d16s16 = vget_low_s16(q8s16);
d17s16 = vget_high_s16(q8s16);
@@ -898,8 +835,8 @@ void vpx_idct16x16_10_add_neon_pass1(int16_t *in, int16_t *out,
q0x2s16 = vld2q_s16(in);
q15s16 = q0x2s16.val[0];
TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
transpose_s16_8x8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
// stage 3
q0s16 = vdupq_n_s16(cospi_28_64 * 2);
@@ -1041,8 +978,8 @@ void vpx_idct16x16_10_add_neon_pass2(int16_t *src, int16_t *out,
q0x2s16 = vld2q_s16(src);
q15s16 = q0x2s16.val[0];
TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
transpose_s16_8x8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
// stage 3
q6s16 = vdupq_n_s16(cospi_30_64 * 2);

66
vpx_dsp/arm/idct32x32_add_neon.c
View File

@@ -11,6 +11,7 @@
#include <arm_neon.h>
#include "./vpx_config.h"
#include "vpx_dsp/arm/transpose_neon.h"
#include "vpx_dsp/txfm_common.h"
#define LOAD_FROM_TRANSPOSED(prev, first, second) \
@@ -155,11 +156,7 @@ static INLINE void idct32_transpose_pair(int16_t *input, int16_t *t_buf) {
int16_t *in;
int i;
const int stride = 32;
int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16;
int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16;
int16x8_t q8s16, q9s16, q10s16, q11s16, q12s16, q13s16, q14s16, q15s16;
int32x4x2_t q0x2s32, q1x2s32, q2x2s32, q3x2s32;
int16x8x2_t q0x2s16, q1x2s16, q2x2s16, q3x2s16;
for (i = 0; i < 4; i++, input += 8) {
in = input;
@@ -179,65 +176,24 @@ static INLINE void idct32_transpose_pair(int16_t *input, int16_t *t_buf) {
in += stride;
q15s16 = vld1q_s16(in);
d16s16 = vget_low_s16(q8s16);
d17s16 = vget_high_s16(q8s16);
d18s16 = vget_low_s16(q9s16);
d19s16 = vget_high_s16(q9s16);
d20s16 = vget_low_s16(q10s16);
d21s16 = vget_high_s16(q10s16);
d22s16 = vget_low_s16(q11s16);
d23s16 = vget_high_s16(q11s16);
d24s16 = vget_low_s16(q12s16);
d25s16 = vget_high_s16(q12s16);
d26s16 = vget_low_s16(q13s16);
d27s16 = vget_high_s16(q13s16);
d28s16 = vget_low_s16(q14s16);
d29s16 = vget_high_s16(q14s16);
d30s16 = vget_low_s16(q15s16);
d31s16 = vget_high_s16(q15s16);
transpose_s16_8x8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16,
&q14s16, &q15s16);
q8s16 = vcombine_s16(d16s16, d24s16); // vswp d17, d24
q9s16 = vcombine_s16(d18s16, d26s16); // vswp d19, d26
q10s16 = vcombine_s16(d20s16, d28s16); // vswp d21, d28
q11s16 = vcombine_s16(d22s16, d30s16); // vswp d23, d30
q12s16 = vcombine_s16(d17s16, d25s16);
q13s16 = vcombine_s16(d19s16, d27s16);
q14s16 = vcombine_s16(d21s16, d29s16);
q15s16 = vcombine_s16(d23s16, d31s16);
q0x2s32 =
vtrnq_s32(vreinterpretq_s32_s16(q8s16), vreinterpretq_s32_s16(q10s16));
q1x2s32 =
vtrnq_s32(vreinterpretq_s32_s16(q9s16), vreinterpretq_s32_s16(q11s16));
q2x2s32 =
vtrnq_s32(vreinterpretq_s32_s16(q12s16), vreinterpretq_s32_s16(q14s16));
q3x2s32 =
vtrnq_s32(vreinterpretq_s32_s16(q13s16), vreinterpretq_s32_s16(q15s16));
q0x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[0]), // q8
vreinterpretq_s16_s32(q1x2s32.val[0])); // q9
q1x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[1]), // q10
vreinterpretq_s16_s32(q1x2s32.val[1])); // q11
q2x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[0]), // q12
vreinterpretq_s16_s32(q3x2s32.val[0])); // q13
q3x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[1]), // q14
vreinterpretq_s16_s32(q3x2s32.val[1])); // q15
vst1q_s16(t_buf, q0x2s16.val[0]);
vst1q_s16(t_buf, q8s16);
t_buf += 8;
vst1q_s16(t_buf, q0x2s16.val[1]);
vst1q_s16(t_buf, q9s16);
t_buf += 8;
vst1q_s16(t_buf, q1x2s16.val[0]);
vst1q_s16(t_buf, q10s16);
t_buf += 8;
vst1q_s16(t_buf, q1x2s16.val[1]);
vst1q_s16(t_buf, q11s16);
t_buf += 8;
vst1q_s16(t_buf, q2x2s16.val[0]);
vst1q_s16(t_buf, q12s16);
t_buf += 8;
vst1q_s16(t_buf, q2x2s16.val[1]);
vst1q_s16(t_buf, q13s16);
t_buf += 8;
vst1q_s16(t_buf, q3x2s16.val[0]);
vst1q_s16(t_buf, q14s16);
t_buf += 8;
vst1q_s16(t_buf, q3x2s16.val[1]);
vst1q_s16(t_buf, q15s16);
t_buf += 8;
}
return;

81
vpx_dsp/arm/idct8x8_add_neon.c
View File

@@ -11,72 +11,9 @@
#include <arm_neon.h>
#include "./vpx_config.h"
#include "vpx_dsp/arm/transpose_neon.h"
#include "vpx_dsp/txfm_common.h"
static INLINE void TRANSPOSE8X8(int16x8_t *q8s16, int16x8_t *q9s16,
int16x8_t *q10s16, int16x8_t *q11s16,
int16x8_t *q12s16, int16x8_t *q13s16,
int16x8_t *q14s16, int16x8_t *q15s16) {
int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16;
int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16;
int32x4x2_t q0x2s32, q1x2s32, q2x2s32, q3x2s32;
int16x8x2_t q0x2s16, q1x2s16, q2x2s16, q3x2s16;
d16s16 = vget_low_s16(*q8s16);
d17s16 = vget_high_s16(*q8s16);
d18s16 = vget_low_s16(*q9s16);
d19s16 = vget_high_s16(*q9s16);
d20s16 = vget_low_s16(*q10s16);
d21s16 = vget_high_s16(*q10s16);
d22s16 = vget_low_s16(*q11s16);
d23s16 = vget_high_s16(*q11s16);
d24s16 = vget_low_s16(*q12s16);
d25s16 = vget_high_s16(*q12s16);
d26s16 = vget_low_s16(*q13s16);
d27s16 = vget_high_s16(*q13s16);
d28s16 = vget_low_s16(*q14s16);
d29s16 = vget_high_s16(*q14s16);
d30s16 = vget_low_s16(*q15s16);
d31s16 = vget_high_s16(*q15s16);
*q8s16 = vcombine_s16(d16s16, d24s16); // vswp d17, d24
*q9s16 = vcombine_s16(d18s16, d26s16); // vswp d19, d26
*q10s16 = vcombine_s16(d20s16, d28s16); // vswp d21, d28
*q11s16 = vcombine_s16(d22s16, d30s16); // vswp d23, d30
*q12s16 = vcombine_s16(d17s16, d25s16);
*q13s16 = vcombine_s16(d19s16, d27s16);
*q14s16 = vcombine_s16(d21s16, d29s16);
*q15s16 = vcombine_s16(d23s16, d31s16);
q0x2s32 =
vtrnq_s32(vreinterpretq_s32_s16(*q8s16), vreinterpretq_s32_s16(*q10s16));
q1x2s32 =
vtrnq_s32(vreinterpretq_s32_s16(*q9s16), vreinterpretq_s32_s16(*q11s16));
q2x2s32 =
vtrnq_s32(vreinterpretq_s32_s16(*q12s16), vreinterpretq_s32_s16(*q14s16));
q3x2s32 =
vtrnq_s32(vreinterpretq_s32_s16(*q13s16), vreinterpretq_s32_s16(*q15s16));
q0x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[0]), // q8
vreinterpretq_s16_s32(q1x2s32.val[0])); // q9
q1x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[1]), // q10
vreinterpretq_s16_s32(q1x2s32.val[1])); // q11
q2x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[0]), // q12
vreinterpretq_s16_s32(q3x2s32.val[0])); // q13
q3x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[1]), // q14
vreinterpretq_s16_s32(q3x2s32.val[1])); // q15
*q8s16 = q0x2s16.val[0];
*q9s16 = q0x2s16.val[1];
*q10s16 = q1x2s16.val[0];
*q11s16 = q1x2s16.val[1];
*q12s16 = q2x2s16.val[0];
*q13s16 = q2x2s16.val[1];
*q14s16 = q3x2s16.val[0];
*q15s16 = q3x2s16.val[1];
return;
}
static INLINE void IDCT8x8_1D(int16x8_t *q8s16, int16x8_t *q9s16,
int16x8_t *q10s16, int16x8_t *q11s16,
int16x8_t *q12s16, int16x8_t *q13s16,
@@ -244,14 +181,14 @@ void vpx_idct8x8_64_add_neon(int16_t *input, uint8_t *dest, int dest_stride) {
q14s16 = vld1q_s16(input + 48);
q15s16 = vld1q_s16(input + 56);
TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
transpose_s16_8x8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
IDCT8x8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
transpose_s16_8x8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
IDCT8x8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
@@ -350,8 +287,8 @@ void vpx_idct8x8_12_add_neon(int16_t *input, uint8_t *dest, int dest_stride) {
q14s16 = vld1q_s16(input + 48);
q15s16 = vld1q_s16(input + 56);
TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
transpose_s16_8x8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
// First transform rows
// stage 1
@@ -427,8 +364,8 @@ void vpx_idct8x8_12_add_neon(int16_t *input, uint8_t *dest, int dest_stride) {
q14s16 = vsubq_s16(q1s16, q6s16);
q15s16 = vsubq_s16(q0s16, q7s16);
TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
transpose_s16_8x8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
IDCT8x8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);