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

Merge "Removing vp9_idct4_1d_sse2 function."

Browse Source
This commit is contained in:
Dmitry Kovalev
2013-10-11 10:43:56 -07:00
committed by Gerrit Code Review
parent dc079ab138 ddf1b76205
commit 9c8f3063b1
3 changed files with 16 additions and 53 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
vp9/common/vp9_idct.c
View File

@@ -96,7 +96,7 @@ void vp9_iwht4x4_1_add_c(int16_t *in, uint8_t *dest, int dest_stride) {
} }
} }
void vp9_idct4_1d_c(int16_t *input, int16_t *output) { static void idct4_1d(int16_t *input, int16_t *output) {
int16_t step[4]; int16_t step[4];
int temp1, temp2; int temp1, temp2;
// stage 1 // stage 1
@@ -124,7 +124,7 @@ void vp9_idct4x4_16_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
// Rows // Rows
for (i = 0; i < 4; ++i) { for (i = 0; i < 4; ++i) {
vp9_idct4_1d(input, outptr); idct4_1d(input, outptr);
input += 4; input += 4;
outptr += 4; outptr += 4;
} }
@@ -133,7 +133,7 @@ void vp9_idct4x4_16_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
for (i = 0; i < 4; ++i) { for (i = 0; i < 4; ++i) {
for (j = 0; j < 4; ++j) for (j = 0; j < 4; ++j)
temp_in[j] = out[j * 4 + i]; temp_in[j] = out[j * 4 + i];
vp9_idct4_1d(temp_in, temp_out); idct4_1d(temp_in, temp_out);
for (j = 0; j < 4; ++j) for (j = 0; j < 4; ++j)
dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4) dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4)
+ dest[j * dest_stride + i]); + dest[j * dest_stride + i]);
@@ -174,7 +174,7 @@ static void idct8_1d(int16_t *input, int16_t *output) {
step1[6] = dct_const_round_shift(temp2); step1[6] = dct_const_round_shift(temp2);
// stage 2 & stage 3 - even half // stage 2 & stage 3 - even half
vp9_idct4_1d(step1, step1); idct4_1d(step1, step1);
// stage 2 - odd half // stage 2 - odd half
step2[4] = step1[4] + step1[5]; step2[4] = step1[4] + step1[5];
@@ -283,10 +283,10 @@ static void iadst4_1d(int16_t *input, int16_t *output) {
void vp9_short_iht4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride, void vp9_short_iht4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride,
int tx_type) { int tx_type) {
const transform_2d IHT_4[] = { const transform_2d IHT_4[] = {
{ vp9_idct4_1d, vp9_idct4_1d }, // DCT_DCT = 0 { idct4_1d, idct4_1d }, // DCT_DCT = 0
{ iadst4_1d, vp9_idct4_1d }, // ADST_DCT = 1 { iadst4_1d, idct4_1d }, // ADST_DCT = 1
{ vp9_idct4_1d, iadst4_1d }, // DCT_ADST = 2 { idct4_1d, iadst4_1d }, // DCT_ADST = 2
{ iadst4_1d, iadst4_1d } // ADST_ADST = 3 { iadst4_1d, iadst4_1d } // ADST_ADST = 3
}; };
int i, j; int i, j;

2
vp9/common/vp9_rtcd_defs.sh
View File

@@ -306,8 +306,6 @@ specialize vp9_short_iht8x8_add sse2 neon
prototype void vp9_short_iht16x16_add "int16_t *input, uint8_t *output, int pitch, int tx_type" prototype void vp9_short_iht16x16_add "int16_t *input, uint8_t *output, int pitch, int tx_type"
specialize vp9_short_iht16x16_add sse2 specialize vp9_short_iht16x16_add sse2
prototype void vp9_idct4_1d "int16_t *input, int16_t *output"
specialize vp9_idct4_1d sse2
# dct and add # dct and add
prototype void vp9_iwht4x4_1_add "int16_t *input, uint8_t *dest, int dest_stride" prototype void vp9_iwht4x4_1_add "int16_t *input, uint8_t *dest, int dest_stride"

51
vp9/common/x86/vp9_idct_intrin_sse2.c
View File

@@ -165,41 +165,6 @@ void vp9_idct4x4_1_add_sse2(int16_t *input, uint8_t *dest, int stride) {
RECON_AND_STORE4X4(dest, dc_value); RECON_AND_STORE4X4(dest, dc_value);
} }
void vp9_idct4_1d_sse2(int16_t *input, int16_t *output) {
const __m128i zero = _mm_setzero_si128();
const __m128i c1 = _mm_setr_epi16((int16_t)cospi_16_64, (int16_t)cospi_16_64,
(int16_t)cospi_16_64, (int16_t)-cospi_16_64,
(int16_t)cospi_24_64, (int16_t)-cospi_8_64,
(int16_t)cospi_8_64, (int16_t)cospi_24_64);
const __m128i c2 = _mm_setr_epi16(1, 1, 1, 1, 1, -1, 1, -1);
const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING);
__m128i in, temp;
// Load input data.
in = _mm_loadl_epi64((__m128i *)input);
// Construct i3, i1, i3, i1, i2, i0, i2, i0
in = _mm_shufflelo_epi16(in, 0xd8);
in = _mm_unpacklo_epi32(in, in);
// Stage 1
in = _mm_madd_epi16(in, c1);
in = _mm_add_epi32(in, rounding);
in = _mm_srai_epi32(in, DCT_CONST_BITS);
in = _mm_packs_epi32(in, zero);
// Stage 2
temp = _mm_shufflelo_epi16(in, 0x9c);
in = _mm_shufflelo_epi16(in, 0xc9);
in = _mm_unpacklo_epi64(temp, in);
in = _mm_madd_epi16(in, c2);
in = _mm_packs_epi32(in, zero);
// Store results
_mm_storel_epi64((__m128i *)output, in);
}
static INLINE void transpose_4x4(__m128i *res) { static INLINE void transpose_4x4(__m128i *res) {
const __m128i tr0_0 = _mm_unpacklo_epi16(res[0], res[1]); const __m128i tr0_0 = _mm_unpacklo_epi16(res[0], res[1]);
const __m128i tr0_1 = _mm_unpacklo_epi16(res[2], res[3]); const __m128i tr0_1 = _mm_unpacklo_epi16(res[2], res[3]);
@@ -210,7 +175,7 @@ static INLINE void transpose_4x4(__m128i *res) {
res[3] = _mm_unpackhi_epi64(res[2], res[2]); res[3] = _mm_unpackhi_epi64(res[2], res[2]);
} }
void idct4_1d_sse2(__m128i *in) { static void idct4_1d_sse2(__m128i *in) {
const __m128i k__cospi_p16_p16 = pair_set_epi16(cospi_16_64, cospi_16_64); const __m128i k__cospi_p16_p16 = pair_set_epi16(cospi_16_64, cospi_16_64);
const __m128i k__cospi_p16_m16 = pair_set_epi16(cospi_16_64, -cospi_16_64); const __m128i k__cospi_p16_m16 = pair_set_epi16(cospi_16_64, -cospi_16_64);
const __m128i k__cospi_p24_m08 = pair_set_epi16(cospi_24_64, -cospi_8_64); const __m128i k__cospi_p24_m08 = pair_set_epi16(cospi_24_64, -cospi_8_64);
@@ -249,7 +214,7 @@ void idct4_1d_sse2(__m128i *in) {
in[3] = _mm_sub_epi16(u[0], u[3]); in[3] = _mm_sub_epi16(u[0], u[3]);
} }
void iadst4_1d_sse2(__m128i *in) { static void iadst4_1d_sse2(__m128i *in) {
const __m128i k__sinpi_p01_p04 = pair_set_epi16(sinpi_1_9, sinpi_4_9); const __m128i k__sinpi_p01_p04 = pair_set_epi16(sinpi_1_9, sinpi_4_9);
const __m128i k__sinpi_p03_p02 = pair_set_epi16(sinpi_3_9, sinpi_2_9); const __m128i k__sinpi_p03_p02 = pair_set_epi16(sinpi_3_9, sinpi_2_9);
const __m128i k__sinpi_p02_m01 = pair_set_epi16(sinpi_2_9, -sinpi_1_9); const __m128i k__sinpi_p02_m01 = pair_set_epi16(sinpi_2_9, -sinpi_1_9);
@@ -648,7 +613,7 @@ static INLINE void array_transpose_8x8(__m128i *in, __m128i *res) {
res[7] = _mm_unpackhi_epi64(tr1_6, tr1_7); res[7] = _mm_unpackhi_epi64(tr1_6, tr1_7);
} }
void idct8_1d_sse2(__m128i *in) { static void idct8_1d_sse2(__m128i *in) {
const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING); const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING);
const __m128i stg1_0 = pair_set_epi16(cospi_28_64, -cospi_4_64); const __m128i stg1_0 = pair_set_epi16(cospi_28_64, -cospi_4_64);
const __m128i stg1_1 = pair_set_epi16(cospi_4_64, cospi_28_64); const __m128i stg1_1 = pair_set_epi16(cospi_4_64, cospi_28_64);
@@ -689,7 +654,7 @@ void idct8_1d_sse2(__m128i *in) {
in[7] = in7; in[7] = in7;
} }
void iadst8_1d_sse2(__m128i *in) { static void iadst8_1d_sse2(__m128i *in) {
const __m128i k__cospi_p02_p30 = pair_set_epi16(cospi_2_64, cospi_30_64); const __m128i k__cospi_p02_p30 = pair_set_epi16(cospi_2_64, cospi_30_64);
const __m128i k__cospi_p30_m02 = pair_set_epi16(cospi_30_64, -cospi_2_64); const __m128i k__cospi_p30_m02 = pair_set_epi16(cospi_30_64, -cospi_2_64);
const __m128i k__cospi_p10_p22 = pair_set_epi16(cospi_10_64, cospi_22_64); const __m128i k__cospi_p10_p22 = pair_set_epi16(cospi_10_64, cospi_22_64);
@@ -1519,7 +1484,7 @@ static INLINE void array_transpose_16x16(__m128i *res0, __m128i *res1) {
res0[15] = tbuf[7]; res0[15] = tbuf[7];
} }
void iadst16_1d_8col(__m128i *in) { static void iadst16_1d_8col(__m128i *in) {
// perform 16x16 1-D ADST for 8 columns // perform 16x16 1-D ADST for 8 columns
__m128i s[16], x[16], u[32], v[32]; __m128i s[16], x[16], u[32], v[32];
const __m128i k__cospi_p01_p31 = pair_set_epi16(cospi_1_64, cospi_31_64); const __m128i k__cospi_p01_p31 = pair_set_epi16(cospi_1_64, cospi_31_64);
@@ -1989,7 +1954,7 @@ void iadst16_1d_8col(__m128i *in) {
in[15] = _mm_sub_epi16(kZero, s[1]); in[15] = _mm_sub_epi16(kZero, s[1]);
} }
void idct16_1d_8col(__m128i *in) { static void idct16_1d_8col(__m128i *in) {
const __m128i k__cospi_p30_m02 = pair_set_epi16(cospi_30_64, -cospi_2_64); const __m128i k__cospi_p30_m02 = pair_set_epi16(cospi_30_64, -cospi_2_64);
const __m128i k__cospi_p02_p30 = pair_set_epi16(cospi_2_64, cospi_30_64); const __m128i k__cospi_p02_p30 = pair_set_epi16(cospi_2_64, cospi_30_64);
const __m128i k__cospi_p14_m18 = pair_set_epi16(cospi_14_64, -cospi_18_64); const __m128i k__cospi_p14_m18 = pair_set_epi16(cospi_14_64, -cospi_18_64);
@@ -2333,13 +2298,13 @@ void idct16_1d_8col(__m128i *in) {
in[15] = _mm_sub_epi16(s[0], s[15]); in[15] = _mm_sub_epi16(s[0], s[15]);
} }
void idct16_1d_sse2(__m128i *in0, __m128i *in1) { static void idct16_1d_sse2(__m128i *in0, __m128i *in1) {
array_transpose_16x16(in0, in1); array_transpose_16x16(in0, in1);
idct16_1d_8col(in0); idct16_1d_8col(in0);
idct16_1d_8col(in1); idct16_1d_8col(in1);
} }
void iadst16_1d_sse2(__m128i *in0, __m128i *in1) { static void iadst16_1d_sse2(__m128i *in0, __m128i *in1) {
array_transpose_16x16(in0, in1); array_transpose_16x16(in0, in1);
iadst16_1d_8col(in0); iadst16_1d_8col(in0);
iadst16_1d_8col(in1); iadst16_1d_8col(in1);