Commit ec310066 authored by Yunqing Wang's avatar Yunqing Wang
Browse files

Optimize 16x16 idct function

Wrote sse2 version of vp9_short_idct16x16 function. Compared to c
version, the sse2 version is over 2.5X faster.

Change-Id: I38536e2b846427a2cc5c5423aaf305fd0e605d61
parent 66eff0aa
......@@ -295,7 +295,7 @@ prototype void vp9_short_idct1_8x8 "int16_t *input, int16_t *output"
specialize vp9_short_idct1_8x8
prototype void vp9_short_idct16x16 "int16_t *input, int16_t *output, int pitch"
specialize vp9_short_idct16x16
specialize vp9_short_idct16x16 sse2
prototype void vp9_short_idct10_16x16 "int16_t *input, int16_t *output, int pitch"
specialize vp9_short_idct10_16x16
......
......@@ -234,6 +234,36 @@ void vp9_idct4_1d_sse2(int16_t *input, int16_t *output) {
_mm_storel_epi64((__m128i *)output, in);
}
#define TRANSPOSE_8X8(in0, in1, in2, in3, in4, in5, in6, in7, \
out0, out1, out2, out3, out4, out5, out6, out7) \
{ \
const __m128i tr0_0 = _mm_unpacklo_epi16(in0, in1); \
const __m128i tr0_1 = _mm_unpacklo_epi16(in2, in3); \
const __m128i tr0_2 = _mm_unpackhi_epi16(in0, in1); \
const __m128i tr0_3 = _mm_unpackhi_epi16(in2, in3); \
const __m128i tr0_4 = _mm_unpacklo_epi16(in4, in5); \
const __m128i tr0_5 = _mm_unpacklo_epi16(in6, in7); \
const __m128i tr0_6 = _mm_unpackhi_epi16(in4, in5); \
const __m128i tr0_7 = _mm_unpackhi_epi16(in6, in7); \
\
const __m128i tr1_0 = _mm_unpacklo_epi32(tr0_0, tr0_1); \
const __m128i tr1_1 = _mm_unpacklo_epi32(tr0_2, tr0_3); \
const __m128i tr1_2 = _mm_unpackhi_epi32(tr0_0, tr0_1); \
const __m128i tr1_3 = _mm_unpackhi_epi32(tr0_2, tr0_3); \
const __m128i tr1_4 = _mm_unpacklo_epi32(tr0_4, tr0_5); \
const __m128i tr1_5 = _mm_unpacklo_epi32(tr0_6, tr0_7); \
const __m128i tr1_6 = _mm_unpackhi_epi32(tr0_4, tr0_5); \
const __m128i tr1_7 = _mm_unpackhi_epi32(tr0_6, tr0_7); \
\
out0 = _mm_unpacklo_epi64(tr1_0, tr1_4); \
out1 = _mm_unpackhi_epi64(tr1_0, tr1_4); \
out2 = _mm_unpacklo_epi64(tr1_2, tr1_6); \
out3 = _mm_unpackhi_epi64(tr1_2, tr1_6); \
out4 = _mm_unpacklo_epi64(tr1_1, tr1_5); \
out5 = _mm_unpackhi_epi64(tr1_1, tr1_5); \
out6 = _mm_unpacklo_epi64(tr1_3, tr1_7); \
out7 = _mm_unpackhi_epi64(tr1_3, tr1_7); \
}
#define IDCT8x8_1D \
/* Stage1 */ \
......@@ -392,34 +422,8 @@ void vp9_short_idct8x8_sse2(int16_t *input, int16_t *output, int pitch) {
// 2-D
for (i = 0; i < 2; i++) {
// 8x8 Transpose is copied from vp9_short_fdct8x8_sse2()
{
const __m128i tr0_0 = _mm_unpacklo_epi16(in0, in1);
const __m128i tr0_1 = _mm_unpacklo_epi16(in2, in3);
const __m128i tr0_2 = _mm_unpackhi_epi16(in0, in1);
const __m128i tr0_3 = _mm_unpackhi_epi16(in2, in3);
const __m128i tr0_4 = _mm_unpacklo_epi16(in4, in5);
const __m128i tr0_5 = _mm_unpacklo_epi16(in6, in7);
const __m128i tr0_6 = _mm_unpackhi_epi16(in4, in5);
const __m128i tr0_7 = _mm_unpackhi_epi16(in6, in7);
const __m128i tr1_0 = _mm_unpacklo_epi32(tr0_0, tr0_1);
const __m128i tr1_1 = _mm_unpacklo_epi32(tr0_2, tr0_3);
const __m128i tr1_2 = _mm_unpackhi_epi32(tr0_0, tr0_1);
const __m128i tr1_3 = _mm_unpackhi_epi32(tr0_2, tr0_3);
const __m128i tr1_4 = _mm_unpacklo_epi32(tr0_4, tr0_5);
const __m128i tr1_5 = _mm_unpacklo_epi32(tr0_6, tr0_7);
const __m128i tr1_6 = _mm_unpackhi_epi32(tr0_4, tr0_5);
const __m128i tr1_7 = _mm_unpackhi_epi32(tr0_6, tr0_7);
in0 = _mm_unpacklo_epi64(tr1_0, tr1_4);
in1 = _mm_unpackhi_epi64(tr1_0, tr1_4);
in2 = _mm_unpacklo_epi64(tr1_2, tr1_6);
in3 = _mm_unpackhi_epi64(tr1_2, tr1_6);
in4 = _mm_unpacklo_epi64(tr1_1, tr1_5);
in5 = _mm_unpackhi_epi64(tr1_1, tr1_5);
in6 = _mm_unpacklo_epi64(tr1_3, tr1_7);
in7 = _mm_unpackhi_epi64(tr1_3, tr1_7);
}
TRANSPOSE_8X8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
in4, in5, in6, in7);
// 4-stage 1D idct8x8
IDCT8x8_1D
......@@ -633,4 +637,507 @@ void vp9_short_idct10_8x8_sse2(int16_t *input, int16_t *output, int pitch) {
_mm_store_si128((__m128i *)(output + half_pitch * 6), in6);
_mm_store_si128((__m128i *)(output + half_pitch * 7), in7);
}
void vp9_short_idct16x16_sse2(int16_t *input, int16_t *output, int pitch) {
const int half_pitch = pitch >> 1;
const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING);
const __m128i final_rounding = _mm_set1_epi16(1<<5);
const __m128i zero = _mm_setzero_si128();
const __m128i stg2_0 = pair_set_epi16(cospi_30_64, -cospi_2_64);
const __m128i stg2_1 = pair_set_epi16(cospi_2_64, cospi_30_64);
const __m128i stg2_2 = pair_set_epi16(cospi_14_64, -cospi_18_64);
const __m128i stg2_3 = pair_set_epi16(cospi_18_64, cospi_14_64);
const __m128i stg2_4 = pair_set_epi16(cospi_22_64, -cospi_10_64);
const __m128i stg2_5 = pair_set_epi16(cospi_10_64, cospi_22_64);
const __m128i stg2_6 = pair_set_epi16(cospi_6_64, -cospi_26_64);
const __m128i stg2_7 = pair_set_epi16(cospi_26_64, cospi_6_64);
const __m128i stg3_0 = pair_set_epi16(cospi_28_64, -cospi_4_64);
const __m128i stg3_1 = pair_set_epi16(cospi_4_64, cospi_28_64);
const __m128i stg3_2 = pair_set_epi16(cospi_12_64, -cospi_20_64);
const __m128i stg3_3 = pair_set_epi16(cospi_20_64, cospi_12_64);
const __m128i stg4_0 = pair_set_epi16(cospi_16_64, cospi_16_64);
const __m128i stg4_1 = pair_set_epi16(cospi_16_64, -cospi_16_64);
const __m128i stg4_2 = pair_set_epi16(cospi_24_64, -cospi_8_64);
const __m128i stg4_3 = pair_set_epi16(cospi_8_64, cospi_24_64);
const __m128i stg4_4 = pair_set_epi16(-cospi_8_64, cospi_24_64);
const __m128i stg4_5 = pair_set_epi16(cospi_24_64, cospi_8_64);
const __m128i stg4_6 = pair_set_epi16(-cospi_24_64, -cospi_8_64);
const __m128i stg4_7 = pair_set_epi16(-cospi_8_64, cospi_24_64);
const __m128i stg6_0 = pair_set_epi16(-cospi_16_64, cospi_16_64);
__m128i in0 = zero, in1 = zero, in2 = zero, in3 = zero, in4 = zero,
in5 = zero, in6 = zero, in7 = zero, in8 = zero, in9 = zero,
in10 = zero, in11 = zero, in12 = zero, in13 = zero,
in14 = zero, in15 = zero;
__m128i l0 = zero, l1 = zero, l2 = zero, l3 = zero, l4 = zero, l5 = zero,
l6 = zero, l7 = zero, l8 = zero, l9 = zero, l10 = zero, l11 = zero,
l12 = zero, l13 = zero, l14 = zero, l15 = zero;
__m128i r0 = zero, r1 = zero, r2 = zero, r3 = zero, r4 = zero, r5 = zero,
r6 = zero, r7 = zero, r8 = zero, r9 = zero, r10 = zero, r11 = zero,
r12 = zero, r13 = zero, r14 = zero, r15 = zero;
__m128i stp1_0, stp1_1, stp1_2, stp1_3, stp1_4, stp1_5, stp1_6, stp1_7,
stp1_8, stp1_9, stp1_10, stp1_11, stp1_12, stp1_13, stp1_14, stp1_15,
stp1_8_0, stp1_12_0;
__m128i stp2_0, stp2_1, stp2_2, stp2_3, stp2_4, stp2_5, stp2_6, stp2_7,
stp2_8, stp2_9, stp2_10, stp2_11, stp2_12, stp2_13, stp2_14, stp2_15;
__m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
int i;
// We work on a 8x16 block each time, and loop 4 times for 2-D 16x16 idct.
for (i = 0; i < 4; i++) {
// 1-D idct
if (i < 2) {
if (i == 1) input += 128;
// Load input data.
in0 = _mm_load_si128((__m128i *)input);
in8 = _mm_load_si128((__m128i *)(input + 8 * 1));
in1 = _mm_load_si128((__m128i *)(input + 8 * 2));
in9 = _mm_load_si128((__m128i *)(input + 8 * 3));
in2 = _mm_load_si128((__m128i *)(input + 8 * 4));
in10 = _mm_load_si128((__m128i *)(input + 8 * 5));
in3 = _mm_load_si128((__m128i *)(input + 8 * 6));
in11 = _mm_load_si128((__m128i *)(input + 8 * 7));
in4 = _mm_load_si128((__m128i *)(input + 8 * 8));
in12 = _mm_load_si128((__m128i *)(input + 8 * 9));
in5 = _mm_load_si128((__m128i *)(input + 8 * 10));
in13 = _mm_load_si128((__m128i *)(input + 8 * 11));
in6 = _mm_load_si128((__m128i *)(input + 8 * 12));
in14 = _mm_load_si128((__m128i *)(input + 8 * 13));
in7 = _mm_load_si128((__m128i *)(input + 8 * 14));
in15 = _mm_load_si128((__m128i *)(input + 8 * 15));
TRANSPOSE_8X8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
in4, in5, in6, in7);
TRANSPOSE_8X8(in8, in9, in10, in11, in12, in13, in14, in15, in8, in9,
in10, in11, in12, in13, in14, in15);
}
if (i == 2) {
TRANSPOSE_8X8(l0, l1, l2, l3, l4, l5, l6, l7, in0, in1, in2, in3, in4,
in5, in6, in7);
TRANSPOSE_8X8(r0, r1, r2, r3, r4, r5, r6, r7, in8, in9, in10, in11, in12,
in13, in14, in15);
}
if (i == 3) {
TRANSPOSE_8X8(l8, l9, l10, l11, l12, l13, l14, l15, in0, in1, in2, in3,
in4, in5, in6, in7);
TRANSPOSE_8X8(r8, r9, r10, r11, r12, r13, r14, r15, in8, in9, in10, in11,
in12, in13, in14, in15);
}
// Stage2
{
const __m128i lo_1_15 = _mm_unpacklo_epi16(in1, in15);
const __m128i hi_1_15 = _mm_unpackhi_epi16(in1, in15);
const __m128i lo_9_7 = _mm_unpacklo_epi16(in9, in7);
const __m128i hi_9_7 = _mm_unpackhi_epi16(in9, in7);
const __m128i lo_5_11 = _mm_unpacklo_epi16(in5, in11);
const __m128i hi_5_11 = _mm_unpackhi_epi16(in5, in11);
const __m128i lo_13_3 = _mm_unpacklo_epi16(in13, in3);
const __m128i hi_13_3 = _mm_unpackhi_epi16(in13, in3);
tmp0 = _mm_madd_epi16(lo_1_15, stg2_0);
tmp1 = _mm_madd_epi16(hi_1_15, stg2_0);
tmp2 = _mm_madd_epi16(lo_1_15, stg2_1);
tmp3 = _mm_madd_epi16(hi_1_15, stg2_1);
tmp4 = _mm_madd_epi16(lo_9_7, stg2_2);
tmp5 = _mm_madd_epi16(hi_9_7, stg2_2);
tmp6 = _mm_madd_epi16(lo_9_7, stg2_3);
tmp7 = _mm_madd_epi16(hi_9_7, stg2_3);
tmp0 = _mm_add_epi32(tmp0, rounding);
tmp1 = _mm_add_epi32(tmp1, rounding);
tmp2 = _mm_add_epi32(tmp2, rounding);
tmp3 = _mm_add_epi32(tmp3, rounding);
tmp4 = _mm_add_epi32(tmp4, rounding);
tmp5 = _mm_add_epi32(tmp5, rounding);
tmp6 = _mm_add_epi32(tmp6, rounding);
tmp7 = _mm_add_epi32(tmp7, rounding);
tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS);
tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS);
tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS);
tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS);
tmp4 = _mm_srai_epi32(tmp4, DCT_CONST_BITS);
tmp5 = _mm_srai_epi32(tmp5, DCT_CONST_BITS);
tmp6 = _mm_srai_epi32(tmp6, DCT_CONST_BITS);
tmp7 = _mm_srai_epi32(tmp7, DCT_CONST_BITS);
stp2_8 = _mm_packs_epi32(tmp0, tmp1);
stp2_15 = _mm_packs_epi32(tmp2, tmp3);
stp2_9 = _mm_packs_epi32(tmp4, tmp5);
stp2_14 = _mm_packs_epi32(tmp6, tmp7);
tmp0 = _mm_madd_epi16(lo_5_11, stg2_4);
tmp1 = _mm_madd_epi16(hi_5_11, stg2_4);
tmp2 = _mm_madd_epi16(lo_5_11, stg2_5);
tmp3 = _mm_madd_epi16(hi_5_11, stg2_5);
tmp4 = _mm_madd_epi16(lo_13_3, stg2_6);
tmp5 = _mm_madd_epi16(hi_13_3, stg2_6);
tmp6 = _mm_madd_epi16(lo_13_3, stg2_7);
tmp7 = _mm_madd_epi16(hi_13_3, stg2_7);
tmp0 = _mm_add_epi32(tmp0, rounding);
tmp1 = _mm_add_epi32(tmp1, rounding);
tmp2 = _mm_add_epi32(tmp2, rounding);
tmp3 = _mm_add_epi32(tmp3, rounding);
tmp4 = _mm_add_epi32(tmp4, rounding);
tmp5 = _mm_add_epi32(tmp5, rounding);
tmp6 = _mm_add_epi32(tmp6, rounding);
tmp7 = _mm_add_epi32(tmp7, rounding);
tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS);
tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS);
tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS);
tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS);
tmp4 = _mm_srai_epi32(tmp4, DCT_CONST_BITS);
tmp5 = _mm_srai_epi32(tmp5, DCT_CONST_BITS);
tmp6 = _mm_srai_epi32(tmp6, DCT_CONST_BITS);
tmp7 = _mm_srai_epi32(tmp7, DCT_CONST_BITS);
stp2_10 = _mm_packs_epi32(tmp0, tmp1);
stp2_13 = _mm_packs_epi32(tmp2, tmp3);
stp2_11 = _mm_packs_epi32(tmp4, tmp5);
stp2_12 = _mm_packs_epi32(tmp6, tmp7);
}
// Stage3
{
const __m128i lo_2_14 = _mm_unpacklo_epi16(in2, in14);
const __m128i hi_2_14 = _mm_unpackhi_epi16(in2, in14);
const __m128i lo_10_6 = _mm_unpacklo_epi16(in10, in6);
const __m128i hi_10_6 = _mm_unpackhi_epi16(in10, in6);
tmp0 = _mm_madd_epi16(lo_2_14, stg3_0);
tmp1 = _mm_madd_epi16(hi_2_14, stg3_0);
tmp2 = _mm_madd_epi16(lo_2_14, stg3_1);
tmp3 = _mm_madd_epi16(hi_2_14, stg3_1);
tmp4 = _mm_madd_epi16(lo_10_6, stg3_2);
tmp5 = _mm_madd_epi16(hi_10_6, stg3_2);
tmp6 = _mm_madd_epi16(lo_10_6, stg3_3);
tmp7 = _mm_madd_epi16(hi_10_6, stg3_3);
tmp0 = _mm_add_epi32(tmp0, rounding);
tmp1 = _mm_add_epi32(tmp1, rounding);
tmp2 = _mm_add_epi32(tmp2, rounding);
tmp3 = _mm_add_epi32(tmp3, rounding);
tmp4 = _mm_add_epi32(tmp4, rounding);
tmp5 = _mm_add_epi32(tmp5, rounding);
tmp6 = _mm_add_epi32(tmp6, rounding);
tmp7 = _mm_add_epi32(tmp7, rounding);
tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS);
tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS);
tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS);
tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS);
tmp4 = _mm_srai_epi32(tmp4, DCT_CONST_BITS);
tmp5 = _mm_srai_epi32(tmp5, DCT_CONST_BITS);
tmp6 = _mm_srai_epi32(tmp6, DCT_CONST_BITS);
tmp7 = _mm_srai_epi32(tmp7, DCT_CONST_BITS);
stp1_4 = _mm_packs_epi32(tmp0, tmp1);
stp1_7 = _mm_packs_epi32(tmp2, tmp3);
stp1_5 = _mm_packs_epi32(tmp4, tmp5);
stp1_6 = _mm_packs_epi32(tmp6, tmp7);
stp1_8_0 = _mm_add_epi16(stp2_8, stp2_9);
stp1_9 = _mm_sub_epi16(stp2_8, stp2_9);
stp1_10 = _mm_sub_epi16(stp2_11, stp2_10);
stp1_11 = _mm_add_epi16(stp2_11, stp2_10);
stp1_12_0 = _mm_add_epi16(stp2_12, stp2_13);
stp1_13 = _mm_sub_epi16(stp2_12, stp2_13);
stp1_14 = _mm_sub_epi16(stp2_15, stp2_14);
stp1_15 = _mm_add_epi16(stp2_15, stp2_14);
}
// Stage4
{
const __m128i lo_0_8 = _mm_unpacklo_epi16(in0, in8);
const __m128i hi_0_8 = _mm_unpackhi_epi16(in0, in8);
const __m128i lo_4_12 = _mm_unpacklo_epi16(in4, in12);
const __m128i hi_4_12 = _mm_unpackhi_epi16(in4, in12);
const __m128i lo_9_14 = _mm_unpacklo_epi16(stp1_9, stp1_14);
const __m128i hi_9_14 = _mm_unpackhi_epi16(stp1_9, stp1_14);
const __m128i lo_10_13 = _mm_unpacklo_epi16(stp1_10, stp1_13);
const __m128i hi_10_13 = _mm_unpackhi_epi16(stp1_10, stp1_13);
tmp0 = _mm_madd_epi16(lo_0_8, stg4_0);
tmp1 = _mm_madd_epi16(hi_0_8, stg4_0);
tmp2 = _mm_madd_epi16(lo_0_8, stg4_1);
tmp3 = _mm_madd_epi16(hi_0_8, stg4_1);
tmp4 = _mm_madd_epi16(lo_4_12, stg4_2);
tmp5 = _mm_madd_epi16(hi_4_12, stg4_2);
tmp6 = _mm_madd_epi16(lo_4_12, stg4_3);
tmp7 = _mm_madd_epi16(hi_4_12, stg4_3);
tmp0 = _mm_add_epi32(tmp0, rounding);
tmp1 = _mm_add_epi32(tmp1, rounding);
tmp2 = _mm_add_epi32(tmp2, rounding);
tmp3 = _mm_add_epi32(tmp3, rounding);
tmp4 = _mm_add_epi32(tmp4, rounding);
tmp5 = _mm_add_epi32(tmp5, rounding);
tmp6 = _mm_add_epi32(tmp6, rounding);
tmp7 = _mm_add_epi32(tmp7, rounding);
tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS);
tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS);
tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS);
tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS);
tmp4 = _mm_srai_epi32(tmp4, DCT_CONST_BITS);
tmp5 = _mm_srai_epi32(tmp5, DCT_CONST_BITS);
tmp6 = _mm_srai_epi32(tmp6, DCT_CONST_BITS);
tmp7 = _mm_srai_epi32(tmp7, DCT_CONST_BITS);
stp2_0 = _mm_packs_epi32(tmp0, tmp1);
stp2_1 = _mm_packs_epi32(tmp2, tmp3);
stp2_2 = _mm_packs_epi32(tmp4, tmp5);
stp2_3 = _mm_packs_epi32(tmp6, tmp7);
stp2_4 = _mm_add_epi16(stp1_4, stp1_5);
stp2_5 = _mm_sub_epi16(stp1_4, stp1_5);
stp2_6 = _mm_sub_epi16(stp1_7, stp1_6);
stp2_7 = _mm_add_epi16(stp1_7, stp1_6);
tmp0 = _mm_madd_epi16(lo_9_14, stg4_4);
tmp1 = _mm_madd_epi16(hi_9_14, stg4_4);
tmp2 = _mm_madd_epi16(lo_9_14, stg4_5);
tmp3 = _mm_madd_epi16(hi_9_14, stg4_5);
tmp4 = _mm_madd_epi16(lo_10_13, stg4_6);
tmp5 = _mm_madd_epi16(hi_10_13, stg4_6);
tmp6 = _mm_madd_epi16(lo_10_13, stg4_7);
tmp7 = _mm_madd_epi16(hi_10_13, stg4_7);
tmp0 = _mm_add_epi32(tmp0, rounding);
tmp1 = _mm_add_epi32(tmp1, rounding);
tmp2 = _mm_add_epi32(tmp2, rounding);
tmp3 = _mm_add_epi32(tmp3, rounding);
tmp4 = _mm_add_epi32(tmp4, rounding);
tmp5 = _mm_add_epi32(tmp5, rounding);
tmp6 = _mm_add_epi32(tmp6, rounding);
tmp7 = _mm_add_epi32(tmp7, rounding);
tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS);
tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS);
tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS);
tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS);
tmp4 = _mm_srai_epi32(tmp4, DCT_CONST_BITS);
tmp5 = _mm_srai_epi32(tmp5, DCT_CONST_BITS);
tmp6 = _mm_srai_epi32(tmp6, DCT_CONST_BITS);
tmp7 = _mm_srai_epi32(tmp7, DCT_CONST_BITS);
stp2_9 = _mm_packs_epi32(tmp0, tmp1);
stp2_14 = _mm_packs_epi32(tmp2, tmp3);
stp2_10 = _mm_packs_epi32(tmp4, tmp5);
stp2_13 = _mm_packs_epi32(tmp6, tmp7);
}
// Stage5
{
const __m128i lo_6_5 = _mm_unpacklo_epi16(stp2_6, stp2_5);
const __m128i hi_6_5 = _mm_unpackhi_epi16(stp2_6, stp2_5);
stp1_0 = _mm_add_epi16(stp2_0, stp2_3);
stp1_1 = _mm_add_epi16(stp2_1, stp2_2);
stp1_2 = _mm_sub_epi16(stp2_1, stp2_2);
stp1_3 = _mm_sub_epi16(stp2_0, stp2_3);
tmp0 = _mm_madd_epi16(lo_6_5, stg4_1);
tmp1 = _mm_madd_epi16(hi_6_5, stg4_1);
tmp2 = _mm_madd_epi16(lo_6_5, stg4_0);
tmp3 = _mm_madd_epi16(hi_6_5, stg4_0);
tmp0 = _mm_add_epi32(tmp0, rounding);
tmp1 = _mm_add_epi32(tmp1, rounding);
tmp2 = _mm_add_epi32(tmp2, rounding);
tmp3 = _mm_add_epi32(tmp3, rounding);
tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS);
tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS);
tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS);
tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS);
stp1_5 = _mm_packs_epi32(tmp0, tmp1);
stp1_6 = _mm_packs_epi32(tmp2, tmp3);
stp1_8 = _mm_add_epi16(stp1_8_0, stp1_11);
stp1_9 = _mm_add_epi16(stp2_9, stp2_10);
stp1_10 = _mm_sub_epi16(stp2_9, stp2_10);
stp1_11 = _mm_sub_epi16(stp1_8_0, stp1_11);
stp1_12 = _mm_sub_epi16(stp1_15, stp1_12_0);
stp1_13 = _mm_sub_epi16(stp2_14, stp2_13);
stp1_14 = _mm_add_epi16(stp2_14, stp2_13);
stp1_15 = _mm_add_epi16(stp1_15, stp1_12_0);
}
// Stage6
{
const __m128i lo_10_13 = _mm_unpacklo_epi16(stp1_10, stp1_13);
const __m128i hi_10_13 = _mm_unpackhi_epi16(stp1_10, stp1_13);
const __m128i lo_11_12 = _mm_unpacklo_epi16(stp1_11, stp1_12);
const __m128i hi_11_12 = _mm_unpackhi_epi16(stp1_11, stp1_12);
stp2_0 = _mm_add_epi16(stp1_0, stp2_7);
stp2_1 = _mm_add_epi16(stp1_1, stp1_6);
stp2_2 = _mm_add_epi16(stp1_2, stp1_5);
stp2_3 = _mm_add_epi16(stp1_3, stp2_4);
stp2_4 = _mm_sub_epi16(stp1_3, stp2_4);
stp2_5 = _mm_sub_epi16(stp1_2, stp1_5);
stp2_6 = _mm_sub_epi16(stp1_1, stp1_6);
stp2_7 = _mm_sub_epi16(stp1_0, stp2_7);
tmp0 = _mm_madd_epi16(lo_10_13, stg6_0);
tmp1 = _mm_madd_epi16(hi_10_13, stg6_0);
tmp2 = _mm_madd_epi16(lo_10_13, stg4_0);
tmp3 = _mm_madd_epi16(hi_10_13, stg4_0);
tmp4 = _mm_madd_epi16(lo_11_12, stg6_0);
tmp5 = _mm_madd_epi16(hi_11_12, stg6_0);
tmp6 = _mm_madd_epi16(lo_11_12, stg4_0);
tmp7 = _mm_madd_epi16(hi_11_12, stg4_0);
tmp0 = _mm_add_epi32(tmp0, rounding);
tmp1 = _mm_add_epi32(tmp1, rounding);
tmp2 = _mm_add_epi32(tmp2, rounding);
tmp3 = _mm_add_epi32(tmp3, rounding);
tmp4 = _mm_add_epi32(tmp4, rounding);
tmp5 = _mm_add_epi32(tmp5, rounding);
tmp6 = _mm_add_epi32(tmp6, rounding);
tmp7 = _mm_add_epi32(tmp7, rounding);
tmp0 = _mm_srai_epi32(tmp0, DCT_CONST_BITS);
tmp1 = _mm_srai_epi32(tmp1, DCT_CONST_BITS);
tmp2 = _mm_srai_epi32(tmp2, DCT_CONST_BITS);
tmp3 = _mm_srai_epi32(tmp3, DCT_CONST_BITS);
tmp4 = _mm_srai_epi32(tmp4, DCT_CONST_BITS);
tmp5 = _mm_srai_epi32(tmp5, DCT_CONST_BITS);
tmp6 = _mm_srai_epi32(tmp6, DCT_CONST_BITS);
tmp7 = _mm_srai_epi32(tmp7, DCT_CONST_BITS);
stp2_10 = _mm_packs_epi32(tmp0, tmp1);
stp2_13 = _mm_packs_epi32(tmp2, tmp3);
stp2_11 = _mm_packs_epi32(tmp4, tmp5);
stp2_12 = _mm_packs_epi32(tmp6, tmp7);
}
// Stage7
if (i == 0) {
// Left 8x16
l0 = _mm_add_epi16(stp2_0, stp1_15);
l1 = _mm_add_epi16(stp2_1, stp1_14);
l2 = _mm_add_epi16(stp2_2, stp2_13);
l3 = _mm_add_epi16(stp2_3, stp2_12);
l4 = _mm_add_epi16(stp2_4, stp2_11);
l5 = _mm_add_epi16(stp2_5, stp2_10);
l6 = _mm_add_epi16(stp2_6, stp1_9);
l7 = _mm_add_epi16(stp2_7, stp1_8);
l8 = _mm_sub_epi16(stp2_7, stp1_8);
l9 = _mm_sub_epi16(stp2_6, stp1_9);
l10 = _mm_sub_epi16(stp2_5, stp2_10);
l11 = _mm_sub_epi16(stp2_4, stp2_11);
l12 = _mm_sub_epi16(stp2_3, stp2_12);
l13 = _mm_sub_epi16(stp2_2, stp2_13);
l14 = _mm_sub_epi16(stp2_1, stp1_14);
l15 = _mm_sub_epi16(stp2_0, stp1_15);
} else if (i == 1) {
// Right 8x16
r0 = _mm_add_epi16(stp2_0, stp1_15);
r1 = _mm_add_epi16(stp2_1, stp1_14);
r2 = _mm_add_epi16(stp2_2, stp2_13);
r3 = _mm_add_epi16(stp2_3, stp2_12);
r4 = _mm_add_epi16(stp2_4, stp2_11);
r5 = _mm_add_epi16(stp2_5, stp2_10);
r6 = _mm_add_epi16(stp2_6, stp1_9);
r7 = _mm_add_epi16(stp2_7, stp1_8);
r8 = _mm_sub_epi16(stp2_7, stp1_8);
r9 = _mm_sub_epi16(stp2_6, stp1_9);
r10 = _mm_sub_epi16(stp2_5, stp2_10);
r11 = _mm_sub_epi16(stp2_4, stp2_11);
r12 = _mm_sub_epi16(stp2_3, stp2_12);
r13 = _mm_sub_epi16(stp2_2, stp2_13);
r14 = _mm_sub_epi16(stp2_1, stp1_14);
r15 = _mm_sub_epi16(stp2_0, stp1_15);
} else {
// 2-D
in0 = _mm_add_epi16(stp2_0, stp1_15);
in1 = _mm_add_epi16(stp2_1, stp1_14);
in2 = _mm_add_epi16(stp2_2, stp2_13);
in3 = _mm_add_epi16(stp2_3, stp2_12);
in4 = _mm_add_epi16(stp2_4, stp2_11);
in5 = _mm_add_epi16(stp2_5, stp2_10);
in6 = _mm_add_epi16(stp2_6, stp1_9);
in7 = _mm_add_epi16(stp2_7, stp1_8);
in8 = _mm_sub_epi16(stp2_7, stp1_8);
in9 = _mm_sub_epi16(stp2_6, stp1_9);
in10 = _mm_sub_epi16(stp2_5, stp2_10);
in11 = _mm_sub_epi16(stp2_4, stp2_11);
in12 = _mm_sub_epi16(stp2_3, stp2_12);
in13 = _mm_sub_epi16(stp2_2, stp2_13);
in14 = _mm_sub_epi16(stp2_1, stp1_14);
in15 = _mm_sub_epi16(stp2_0, stp1_15);
// Final rounding and shift
in0 = _mm_add_epi16(in0, final_rounding);
in1 = _mm_add_epi16(in1, final_rounding);
in2 = _mm_add_epi16(in2, final_rounding);
in3 = _mm_add_epi16(in3, final_rounding);
in4 = _mm_add_epi16(in4, final_rounding);
in5 = _mm_add_epi16(in5, final_rounding);
in6 = _mm_add_epi16(in6, final_rounding);
in7 = _mm_add_epi16(in7, final_rounding);
in8 = _mm_add_epi16(in8, final_rounding);
in9 = _mm_add_epi16(in9, final_rounding);
in10 = _mm_add_epi16(in10, final_rounding);
in11 = _mm_add_epi16(in11, final_rounding);
in12 = _mm_add_epi16(in12, final_rounding);
in13 = _mm_add_epi16(in13, final_rounding);
in14 = _mm_add_epi16(in14, final_rounding);
in15 = _mm_add_epi16(in15, final_rounding);
in0 = _mm_srai_epi16(in0, 6);
in1 = _mm_srai_epi16(in1, 6);
in2 = _mm_srai_epi16(in2, 6);
in3 = _mm_srai_epi16(in3, 6);
in4 = _mm_srai_epi16(in4, 6);
in5 = _mm_srai_epi16(in5, 6);
in6 = _mm_srai_epi16(in6, 6);
in7 = _mm_srai_epi16(in7, 6);
in8 = _mm_srai_epi16(in8, 6);
in9 = _mm_srai_epi16(in9, 6);
in10 = _mm_srai_epi16(in10, 6);
in11 = _mm_srai_epi16(in11, 6);
in12 = _mm_srai_epi16(in12, 6);
in13 = _mm_srai_epi16(in13, 6);
in14 = _mm_srai_epi16(in14, 6);
in15 = _mm_srai_epi16(in15, 6);
// Store results
_mm_store_si128((__m128i *)output, in0);
_mm_store_si128((__m128i *)(output + half_pitch * 1), in1);
_mm_store_si128((__m128i *)(output + half_pitch * 2), in2);
_mm_store_si128((__m128i *)(output + half_pitch * 3), in3);
_mm_store_si128((__m128i *)(output + half_pitch * 4), in4);