Skip to content
GitLab
Commit ec4048ae authored by Urvang Joshi's avatar Urvang Joshi
Browse files

Move scale_sqrt2_* functions to a common location. 

All of these were duplicated in two places.

BUG=aomedia:442

Change-Id: I8eb027c814f58eb23f71a8b8b4691970a333ad96
parent 7f8feefd
Hide whitespace changes
Inline Side-by-side
aom_dsp/x86/txfm_common_sse2.h
View file @ ec4048ae
......@@ -14,6 +14,7 @@
#include <emmintrin.h>
#include "aom/aom_integer.h"
#include "aom_dsp/x86/synonyms.h"
#define pair_set_epi16(a, b) \
_mm_set_epi16((int16_t)(b), (int16_t)(a), (int16_t)(b), (int16_t)(a), \
......@@ -227,4 +228,99 @@ static INLINE void idtx16_8col(__m128i *in) {
}
#endif // CONFIG_EXT_TX
static INLINE void scale_sqrt2_8x4(__m128i *in) {
// Implements ROUND_POWER_OF_TWO(input * Sqrt2, DCT_CONST_BITS), for 32
// consecutive elements.
const __m128i v_scale_w = _mm_set1_epi16((int16_t)Sqrt2);
const __m128i v_p0l_w = _mm_mullo_epi16(in[0], v_scale_w);
const __m128i v_p0h_w = _mm_mulhi_epi16(in[0], v_scale_w);
const __m128i v_p1l_w = _mm_mullo_epi16(in[1], v_scale_w);
const __m128i v_p1h_w = _mm_mulhi_epi16(in[1], v_scale_w);
const __m128i v_p2l_w = _mm_mullo_epi16(in[2], v_scale_w);
const __m128i v_p2h_w = _mm_mulhi_epi16(in[2], v_scale_w);
const __m128i v_p3l_w = _mm_mullo_epi16(in[3], v_scale_w);
const __m128i v_p3h_w = _mm_mulhi_epi16(in[3], v_scale_w);
const __m128i v_p0a_d = _mm_unpacklo_epi16(v_p0l_w, v_p0h_w);
const __m128i v_p0b_d = _mm_unpackhi_epi16(v_p0l_w, v_p0h_w);
const __m128i v_p1a_d = _mm_unpacklo_epi16(v_p1l_w, v_p1h_w);
const __m128i v_p1b_d = _mm_unpackhi_epi16(v_p1l_w, v_p1h_w);
const __m128i v_p2a_d = _mm_unpacklo_epi16(v_p2l_w, v_p2h_w);
const __m128i v_p2b_d = _mm_unpackhi_epi16(v_p2l_w, v_p2h_w);
const __m128i v_p3a_d = _mm_unpacklo_epi16(v_p3l_w, v_p3h_w);
const __m128i v_p3b_d = _mm_unpackhi_epi16(v_p3l_w, v_p3h_w);
in[0] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p0a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p0b_d, DCT_CONST_BITS));
in[1] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p1a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p1b_d, DCT_CONST_BITS));
in[2] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p2a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p2b_d, DCT_CONST_BITS));
in[3] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p3a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p3b_d, DCT_CONST_BITS));
}
static INLINE void scale_sqrt2_8x8(__m128i *in) {
// Implements 'ROUND_POWER_OF_TWO_SIGNED(input * Sqrt2, DCT_CONST_BITS)'
// for each element.
const __m128i v_scale_w = _mm_set1_epi16((int16_t)Sqrt2);
const __m128i v_p0l_w = _mm_mullo_epi16(in[0], v_scale_w);
const __m128i v_p0h_w = _mm_mulhi_epi16(in[0], v_scale_w);
const __m128i v_p1l_w = _mm_mullo_epi16(in[1], v_scale_w);
const __m128i v_p1h_w = _mm_mulhi_epi16(in[1], v_scale_w);
const __m128i v_p2l_w = _mm_mullo_epi16(in[2], v_scale_w);
const __m128i v_p2h_w = _mm_mulhi_epi16(in[2], v_scale_w);
const __m128i v_p3l_w = _mm_mullo_epi16(in[3], v_scale_w);
const __m128i v_p3h_w = _mm_mulhi_epi16(in[3], v_scale_w);
const __m128i v_p4l_w = _mm_mullo_epi16(in[4], v_scale_w);
const __m128i v_p4h_w = _mm_mulhi_epi16(in[4], v_scale_w);
const __m128i v_p5l_w = _mm_mullo_epi16(in[5], v_scale_w);
const __m128i v_p5h_w = _mm_mulhi_epi16(in[5], v_scale_w);
const __m128i v_p6l_w = _mm_mullo_epi16(in[6], v_scale_w);
const __m128i v_p6h_w = _mm_mulhi_epi16(in[6], v_scale_w);
const __m128i v_p7l_w = _mm_mullo_epi16(in[7], v_scale_w);
const __m128i v_p7h_w = _mm_mulhi_epi16(in[7], v_scale_w);
const __m128i v_p0a_d = _mm_unpacklo_epi16(v_p0l_w, v_p0h_w);
const __m128i v_p0b_d = _mm_unpackhi_epi16(v_p0l_w, v_p0h_w);
const __m128i v_p1a_d = _mm_unpacklo_epi16(v_p1l_w, v_p1h_w);
const __m128i v_p1b_d = _mm_unpackhi_epi16(v_p1l_w, v_p1h_w);
const __m128i v_p2a_d = _mm_unpacklo_epi16(v_p2l_w, v_p2h_w);
const __m128i v_p2b_d = _mm_unpackhi_epi16(v_p2l_w, v_p2h_w);
const __m128i v_p3a_d = _mm_unpacklo_epi16(v_p3l_w, v_p3h_w);
const __m128i v_p3b_d = _mm_unpackhi_epi16(v_p3l_w, v_p3h_w);
const __m128i v_p4a_d = _mm_unpacklo_epi16(v_p4l_w, v_p4h_w);
const __m128i v_p4b_d = _mm_unpackhi_epi16(v_p4l_w, v_p4h_w);
const __m128i v_p5a_d = _mm_unpacklo_epi16(v_p5l_w, v_p5h_w);
const __m128i v_p5b_d = _mm_unpackhi_epi16(v_p5l_w, v_p5h_w);
const __m128i v_p6a_d = _mm_unpacklo_epi16(v_p6l_w, v_p6h_w);
const __m128i v_p6b_d = _mm_unpackhi_epi16(v_p6l_w, v_p6h_w);
const __m128i v_p7a_d = _mm_unpacklo_epi16(v_p7l_w, v_p7h_w);
const __m128i v_p7b_d = _mm_unpackhi_epi16(v_p7l_w, v_p7h_w);
in[0] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p0a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p0b_d, DCT_CONST_BITS));
in[1] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p1a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p1b_d, DCT_CONST_BITS));
in[2] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p2a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p2b_d, DCT_CONST_BITS));
in[3] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p3a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p3b_d, DCT_CONST_BITS));
in[4] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p4a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p4b_d, DCT_CONST_BITS));
in[5] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p5a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p5b_d, DCT_CONST_BITS));
in[6] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p6a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p6b_d, DCT_CONST_BITS));
in[7] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p7a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p7b_d, DCT_CONST_BITS));
}
static INLINE void scale_sqrt2_8x16(__m128i *in) {
scale_sqrt2_8x8(in);
scale_sqrt2_8x8(in + 8);
}
#endif // AOM_DSP_X86_TXFM_COMMON_SSE2_H_
av1/common/x86/idct_intrin_sse2.c
View file @ ec4048ae
......@@ -387,101 +387,6 @@ static INLINE void flip_buffer_lr_8x8(__m128i *in) {
}
#endif // CONFIG_EXT_TX
static INLINE void scale_sqrt2_8x4(__m128i *in) {
// Implements 'ROUND_POWER_OF_TWO(input * Sqrt2, DCT_CONST_BITS)'
// for each element
const __m128i v_scale_w = _mm_set1_epi16((int16_t)Sqrt2);
const __m128i v_p0l_w = _mm_mullo_epi16(in[0], v_scale_w);
const __m128i v_p0h_w = _mm_mulhi_epi16(in[0], v_scale_w);
const __m128i v_p1l_w = _mm_mullo_epi16(in[1], v_scale_w);
const __m128i v_p1h_w = _mm_mulhi_epi16(in[1], v_scale_w);
const __m128i v_p2l_w = _mm_mullo_epi16(in[2], v_scale_w);
const __m128i v_p2h_w = _mm_mulhi_epi16(in[2], v_scale_w);
const __m128i v_p3l_w = _mm_mullo_epi16(in[3], v_scale_w);
const __m128i v_p3h_w = _mm_mulhi_epi16(in[3], v_scale_w);
const __m128i v_p0a_d = _mm_unpacklo_epi16(v_p0l_w, v_p0h_w);
const __m128i v_p0b_d = _mm_unpackhi_epi16(v_p0l_w, v_p0h_w);
const __m128i v_p1a_d = _mm_unpacklo_epi16(v_p1l_w, v_p1h_w);
const __m128i v_p1b_d = _mm_unpackhi_epi16(v_p1l_w, v_p1h_w);
const __m128i v_p2a_d = _mm_unpacklo_epi16(v_p2l_w, v_p2h_w);
const __m128i v_p2b_d = _mm_unpackhi_epi16(v_p2l_w, v_p2h_w);
const __m128i v_p3a_d = _mm_unpacklo_epi16(v_p3l_w, v_p3h_w);
const __m128i v_p3b_d = _mm_unpackhi_epi16(v_p3l_w, v_p3h_w);
in[0] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p0a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p0b_d, DCT_CONST_BITS));
in[1] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p1a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p1b_d, DCT_CONST_BITS));
in[2] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p2a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p2b_d, DCT_CONST_BITS));
in[3] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p3a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p3b_d, DCT_CONST_BITS));
}
static INLINE void scale_sqrt2_8x8(__m128i *in) {
// Implements 'ROUND_POWER_OF_TWO_SIGNED(input * Sqrt2, DCT_CONST_BITS)'
// for each element
const __m128i v_scale_w = _mm_set1_epi16((int16_t)Sqrt2);
const __m128i v_p0l_w = _mm_mullo_epi16(in[0], v_scale_w);
const __m128i v_p0h_w = _mm_mulhi_epi16(in[0], v_scale_w);
const __m128i v_p1l_w = _mm_mullo_epi16(in[1], v_scale_w);
const __m128i v_p1h_w = _mm_mulhi_epi16(in[1], v_scale_w);
const __m128i v_p2l_w = _mm_mullo_epi16(in[2], v_scale_w);
const __m128i v_p2h_w = _mm_mulhi_epi16(in[2], v_scale_w);
const __m128i v_p3l_w = _mm_mullo_epi16(in[3], v_scale_w);
const __m128i v_p3h_w = _mm_mulhi_epi16(in[3], v_scale_w);
const __m128i v_p4l_w = _mm_mullo_epi16(in[4], v_scale_w);
const __m128i v_p4h_w = _mm_mulhi_epi16(in[4], v_scale_w);
const __m128i v_p5l_w = _mm_mullo_epi16(in[5], v_scale_w);
const __m128i v_p5h_w = _mm_mulhi_epi16(in[5], v_scale_w);
const __m128i v_p6l_w = _mm_mullo_epi16(in[6], v_scale_w);
const __m128i v_p6h_w = _mm_mulhi_epi16(in[6], v_scale_w);
const __m128i v_p7l_w = _mm_mullo_epi16(in[7], v_scale_w);
const __m128i v_p7h_w = _mm_mulhi_epi16(in[7], v_scale_w);
const __m128i v_p0a_d = _mm_unpacklo_epi16(v_p0l_w, v_p0h_w);
const __m128i v_p0b_d = _mm_unpackhi_epi16(v_p0l_w, v_p0h_w);
const __m128i v_p1a_d = _mm_unpacklo_epi16(v_p1l_w, v_p1h_w);
const __m128i v_p1b_d = _mm_unpackhi_epi16(v_p1l_w, v_p1h_w);
const __m128i v_p2a_d = _mm_unpacklo_epi16(v_p2l_w, v_p2h_w);
const __m128i v_p2b_d = _mm_unpackhi_epi16(v_p2l_w, v_p2h_w);
const __m128i v_p3a_d = _mm_unpacklo_epi16(v_p3l_w, v_p3h_w);
const __m128i v_p3b_d = _mm_unpackhi_epi16(v_p3l_w, v_p3h_w);
const __m128i v_p4a_d = _mm_unpacklo_epi16(v_p4l_w, v_p4h_w);
const __m128i v_p4b_d = _mm_unpackhi_epi16(v_p4l_w, v_p4h_w);
const __m128i v_p5a_d = _mm_unpacklo_epi16(v_p5l_w, v_p5h_w);
const __m128i v_p5b_d = _mm_unpackhi_epi16(v_p5l_w, v_p5h_w);
const __m128i v_p6a_d = _mm_unpacklo_epi16(v_p6l_w, v_p6h_w);
const __m128i v_p6b_d = _mm_unpackhi_epi16(v_p6l_w, v_p6h_w);
const __m128i v_p7a_d = _mm_unpacklo_epi16(v_p7l_w, v_p7h_w);
const __m128i v_p7b_d = _mm_unpackhi_epi16(v_p7l_w, v_p7h_w);
in[0] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p0a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p0b_d, DCT_CONST_BITS));
in[1] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p1a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p1b_d, DCT_CONST_BITS));
in[2] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p2a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p2b_d, DCT_CONST_BITS));
in[3] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p3a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p3b_d, DCT_CONST_BITS));
in[4] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p4a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p4b_d, DCT_CONST_BITS));
in[5] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p5a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p5b_d, DCT_CONST_BITS));
in[6] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p6a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p6b_d, DCT_CONST_BITS));
in[7] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p7a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p7b_d, DCT_CONST_BITS));
}
static INLINE void scale_sqrt2_8x16(__m128i *in) {
scale_sqrt2_8x8(in);
scale_sqrt2_8x8(in + 8);
}
void av1_iht8x16_128_add_sse2(const tran_low_t *input, uint8_t *dest,
int stride, int tx_type) {
__m128i in[16];
......
av1/encoder/x86/dct_intrin_sse2.c
View file @ ec4048ae
......@@ -2456,104 +2456,6 @@ void av1_fht16x16_sse2(const int16_t *input, tran_low_t *output, int stride,
}
}
static INLINE void scale_sqrt2_8x4(__m128i *in) {
// Implements fdct_round_shift(input * Sqrt2), which is equivalent to
// ROUND_POWER_OF_TWO(input * Sqrt2, DCT_CONST_BITS),
// for 32 consecutive elements.
const __m128i v_scale_w = _mm_set1_epi16((int16_t)Sqrt2);
const __m128i v_p0l_w = _mm_mullo_epi16(in[0], v_scale_w);
const __m128i v_p0h_w = _mm_mulhi_epi16(in[0], v_scale_w);
const __m128i v_p1l_w = _mm_mullo_epi16(in[1], v_scale_w);
const __m128i v_p1h_w = _mm_mulhi_epi16(in[1], v_scale_w);
const __m128i v_p2l_w = _mm_mullo_epi16(in[2], v_scale_w);
const __m128i v_p2h_w = _mm_mulhi_epi16(in[2], v_scale_w);
const __m128i v_p3l_w = _mm_mullo_epi16(in[3], v_scale_w);
const __m128i v_p3h_w = _mm_mulhi_epi16(in[3], v_scale_w);
const __m128i v_p0a_d = _mm_unpacklo_epi16(v_p0l_w, v_p0h_w);
const __m128i v_p0b_d = _mm_unpackhi_epi16(v_p0l_w, v_p0h_w);
const __m128i v_p1a_d = _mm_unpacklo_epi16(v_p1l_w, v_p1h_w);
const __m128i v_p1b_d = _mm_unpackhi_epi16(v_p1l_w, v_p1h_w);
const __m128i v_p2a_d = _mm_unpacklo_epi16(v_p2l_w, v_p2h_w);
const __m128i v_p2b_d = _mm_unpackhi_epi16(v_p2l_w, v_p2h_w);
const __m128i v_p3a_d = _mm_unpacklo_epi16(v_p3l_w, v_p3h_w);
const __m128i v_p3b_d = _mm_unpackhi_epi16(v_p3l_w, v_p3h_w);
in[0] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p0a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p0b_d, DCT_CONST_BITS));
in[1] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p1a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p1b_d, DCT_CONST_BITS));
in[2] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p2a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p2b_d, DCT_CONST_BITS));
in[3] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p3a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p3b_d, DCT_CONST_BITS));
}
static INLINE void scale_sqrt2_8x8_unsigned(__m128i *in) {
// Implements 'ROUND_POWER_OF_TWO_SIGNED(input * Sqrt2, DCT_CONST_BITS)'
// for each element
const __m128i v_scale_w = _mm_set1_epi16((int16_t)Sqrt2);
const __m128i v_p0l_w = _mm_mullo_epi16(in[0], v_scale_w);
const __m128i v_p0h_w = _mm_mulhi_epi16(in[0], v_scale_w);
const __m128i v_p1l_w = _mm_mullo_epi16(in[1], v_scale_w);
const __m128i v_p1h_w = _mm_mulhi_epi16(in[1], v_scale_w);
const __m128i v_p2l_w = _mm_mullo_epi16(in[2], v_scale_w);
const __m128i v_p2h_w = _mm_mulhi_epi16(in[2], v_scale_w);
const __m128i v_p3l_w = _mm_mullo_epi16(in[3], v_scale_w);
const __m128i v_p3h_w = _mm_mulhi_epi16(in[3], v_scale_w);
const __m128i v_p4l_w = _mm_mullo_epi16(in[4], v_scale_w);
const __m128i v_p4h_w = _mm_mulhi_epi16(in[4], v_scale_w);
const __m128i v_p5l_w = _mm_mullo_epi16(in[5], v_scale_w);
const __m128i v_p5h_w = _mm_mulhi_epi16(in[5], v_scale_w);
const __m128i v_p6l_w = _mm_mullo_epi16(in[6], v_scale_w);
const __m128i v_p6h_w = _mm_mulhi_epi16(in[6], v_scale_w);
const __m128i v_p7l_w = _mm_mullo_epi16(in[7], v_scale_w);
const __m128i v_p7h_w = _mm_mulhi_epi16(in[7], v_scale_w);
const __m128i v_p0a_d = _mm_unpacklo_epi16(v_p0l_w, v_p0h_w);
const __m128i v_p0b_d = _mm_unpackhi_epi16(v_p0l_w, v_p0h_w);
const __m128i v_p1a_d = _mm_unpacklo_epi16(v_p1l_w, v_p1h_w);
const __m128i v_p1b_d = _mm_unpackhi_epi16(v_p1l_w, v_p1h_w);
const __m128i v_p2a_d = _mm_unpacklo_epi16(v_p2l_w, v_p2h_w);
const __m128i v_p2b_d = _mm_unpackhi_epi16(v_p2l_w, v_p2h_w);
const __m128i v_p3a_d = _mm_unpacklo_epi16(v_p3l_w, v_p3h_w);
const __m128i v_p3b_d = _mm_unpackhi_epi16(v_p3l_w, v_p3h_w);
const __m128i v_p4a_d = _mm_unpacklo_epi16(v_p4l_w, v_p4h_w);
const __m128i v_p4b_d = _mm_unpackhi_epi16(v_p4l_w, v_p4h_w);
const __m128i v_p5a_d = _mm_unpacklo_epi16(v_p5l_w, v_p5h_w);
const __m128i v_p5b_d = _mm_unpackhi_epi16(v_p5l_w, v_p5h_w);
const __m128i v_p6a_d = _mm_unpacklo_epi16(v_p6l_w, v_p6h_w);
const __m128i v_p6b_d = _mm_unpackhi_epi16(v_p6l_w, v_p6h_w);
const __m128i v_p7a_d = _mm_unpacklo_epi16(v_p7l_w, v_p7h_w);
const __m128i v_p7b_d = _mm_unpackhi_epi16(v_p7l_w, v_p7h_w);
in[0] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p0a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p0b_d, DCT_CONST_BITS));
in[1] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p1a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p1b_d, DCT_CONST_BITS));
in[2] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p2a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p2b_d, DCT_CONST_BITS));
in[3] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p3a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p3b_d, DCT_CONST_BITS));
in[4] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p4a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p4b_d, DCT_CONST_BITS));
in[5] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p5a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p5b_d, DCT_CONST_BITS));
in[6] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p6a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p6b_d, DCT_CONST_BITS));
in[7] = _mm_packs_epi32(xx_roundn_epi32_unsigned(v_p7a_d, DCT_CONST_BITS),
xx_roundn_epi32_unsigned(v_p7b_d, DCT_CONST_BITS));
}
static INLINE void scale_sqrt2_8x16(__m128i *in) {
scale_sqrt2_8x4(in);
scale_sqrt2_8x4(in + 4);
scale_sqrt2_8x4(in + 8);
scale_sqrt2_8x4(in + 12);
}
static INLINE void prepare_4x8_row_first(__m128i *in) {
in[0] = _mm_unpacklo_epi64(in[0], in[2]);
in[1] = _mm_unpacklo_epi64(in[1], in[3]);
......@@ -2942,9 +2844,9 @@ static INLINE void load_buffer_8x16(const int16_t *input, __m128i *in,
}
load_buffer_8x8(t, in, stride, flipud, fliplr);
scale_sqrt2_8x8_unsigned(in);
scale_sqrt2_8x8(in);
load_buffer_8x8(b, in + 8, stride, flipud, fliplr);
scale_sqrt2_8x8_unsigned(in + 8);
scale_sqrt2_8x8(in + 8);
}
static INLINE void round_power_of_two_signed(__m128i *x, int n) {
......@@ -3135,9 +3037,9 @@ static INLINE void load_buffer_16x8(const int16_t *input, __m128i *in,
// load first 8 columns
load_buffer_8x8(l, in, stride, flipud, fliplr);
scale_sqrt2_8x8_unsigned(in);
scale_sqrt2_8x8(in);
load_buffer_8x8(r, in + 8, stride, flipud, fliplr);
scale_sqrt2_8x8_unsigned(in + 8);
scale_sqrt2_8x8(in + 8);
}
#define col_16x8_rounding row_8x16_rounding
......
Supports Markdown
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment