ref: ddf1b7620511d21389ca3ef1d6dca2248dc78666
parent: 2be3b84aed9fe25a2962dd3d61107ec9c798d155
author: Dmitry Kovalev <[email protected]>
date: Thu Oct 10 12:50:43 EDT 2013
Removing vp9_idct4_1d_sse2 function.
We have two SSE2-optimized functions for idct4_1d:
vp9_idct4_1d_sse2 <-- removing this one
idct4_1d_sse2
vp9_idct4_1d_sse2 was used only by the following functions which already
have SSE2 optimized variants:
vp9_idct4x4_16_add_c -> vp9_idct4x4_16_add_see2
idct8_1d -> vp9_idct8x8_{16, 10, 1}_see2
vp9_short_iht4x4_add_c -> vp9_short_iht4x4_add_see2
Change-Id: Ib0a7f6d1373dbaf7a4a41208cd9d0671fdf15edb
--- a/vp9/common/vp9_idct.c
+++ b/vp9/common/vp9_idct.c
@@ -96,7 +96,7 @@
}
}
-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];
int temp1, temp2;
// stage 1
@@ -124,7 +124,7 @@
// Rows
for (i = 0; i < 4; ++i) {- vp9_idct4_1d(input, outptr);
+ idct4_1d(input, outptr);
input += 4;
outptr += 4;
}
@@ -133,7 +133,7 @@
for (i = 0; i < 4; ++i) {for (j = 0; j < 4; ++j)
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)
dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4)
+ dest[j * dest_stride + i]);
@@ -174,7 +174,7 @@
step1[6] = dct_const_round_shift(temp2);
// stage 2 & stage 3 - even half
- vp9_idct4_1d(step1, step1);
+ idct4_1d(step1, step1);
// stage 2 - odd half
step2[4] = step1[4] + step1[5];
@@ -283,10 +283,10 @@
void vp9_short_iht4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride,
int tx_type) { const transform_2d IHT_4[] = {- { vp9_idct4_1d, vp9_idct4_1d }, // DCT_DCT = 0- { iadst4_1d, vp9_idct4_1d }, // ADST_DCT = 1- { vp9_idct4_1d, iadst4_1d }, // DCT_ADST = 2- { iadst4_1d, iadst4_1d } // ADST_ADST = 3+ { idct4_1d, idct4_1d }, // DCT_DCT = 0+ { iadst4_1d, idct4_1d }, // ADST_DCT = 1+ { idct4_1d, iadst4_1d }, // DCT_ADST = 2+ { iadst4_1d, iadst4_1d } // ADST_ADST = 3};
int i, j;
--- a/vp9/common/vp9_rtcd_defs.sh
+++ b/vp9/common/vp9_rtcd_defs.sh
@@ -306,8 +306,6 @@
prototype void vp9_short_iht16x16_add "int16_t *input, uint8_t *output, int pitch, int tx_type"
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
prototype void vp9_iwht4x4_1_add "int16_t *input, uint8_t *dest, int dest_stride"
--- a/vp9/common/x86/vp9_idct_intrin_sse2.c
+++ b/vp9/common/x86/vp9_idct_intrin_sse2.c
@@ -165,41 +165,6 @@
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) {const __m128i tr0_0 = _mm_unpacklo_epi16(res[0], res[1]);
const __m128i tr0_1 = _mm_unpacklo_epi16(res[2], res[3]);
@@ -210,7 +175,7 @@
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_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);
@@ -249,7 +214,7 @@
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_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);
@@ -648,7 +613,7 @@
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 stg1_0 = pair_set_epi16(cospi_28_64, -cospi_4_64);
const __m128i stg1_1 = pair_set_epi16(cospi_4_64, cospi_28_64);
@@ -689,7 +654,7 @@
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_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);
@@ -1519,7 +1484,7 @@
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
__m128i s[16], x[16], u[32], v[32];
const __m128i k__cospi_p01_p31 = pair_set_epi16(cospi_1_64, cospi_31_64);
@@ -1989,7 +1954,7 @@
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_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);
@@ -2333,13 +2298,13 @@
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);
idct16_1d_8col(in0);
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);
iadst16_1d_8col(in0);
iadst16_1d_8col(in1);