ref: e07e74fb0f548a2ceb72da4d3264541ffc807db4
parent: 429e6528097850b08b675e1fa8d75eef59a10e32
author: Linfeng Zhang <linfengz@google.com>
date: Tue Feb 14 10:39:37 EST 2017
Add vpx_highbd_idct16x16_38_add_c() When eob is less than or equal to 38 for high-bitdepth 16x16 idct, call this function. BUG=webm:1301 Change-Id: I09167f89d29c401f9c36710b0fd2d02644052060
--- a/test/partial_idct_test.cc
+++ b/test/partial_idct_test.cc
@@ -354,6 +354,15 @@
&highbd_wrapper<vpx_highbd_idct16x16_256_add_c>, TX_16X16, 256, 12, 2),
make_tuple(
&vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_256_add_c>,
+ &highbd_wrapper<vpx_highbd_idct16x16_38_add_c>, TX_16X16, 38, 8, 2),
+ make_tuple(
+ &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_256_add_c>,
+ &highbd_wrapper<vpx_highbd_idct16x16_38_add_c>, TX_16X16, 38, 10, 2),
+ make_tuple(
+ &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_256_add_c>,
+ &highbd_wrapper<vpx_highbd_idct16x16_38_add_c>, TX_16X16, 38, 12, 2),
+ make_tuple(
+ &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_256_add_c>,
&highbd_wrapper<vpx_highbd_idct16x16_10_add_c>, TX_16X16, 10, 8, 2),
make_tuple(
&vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_256_add_c>,
--- a/vp9/common/vp9_idct.c
+++ b/vp9/common/vp9_idct.c
@@ -349,6 +349,8 @@
vpx_highbd_idct16x16_1_add(input, dest, stride, bd);
} else if (eob <= 10) {vpx_highbd_idct16x16_10_add(input, dest, stride, bd);
+ } else if (eob <= 38) {+ vpx_highbd_idct16x16_38_add(input, dest, stride, bd);
} else {vpx_highbd_idct16x16_256_add(input, dest, stride, bd);
}
--- a/vpx_dsp/inv_txfm.c
+++ b/vpx_dsp/inv_txfm.c
@@ -2082,6 +2082,35 @@
output[15] = HIGHBD_WRAPLOW(-x1, bd);
}
+void vpx_highbd_idct16x16_38_add_c(const tran_low_t *input, uint8_t *dest8,
+ int stride, int bd) {+ int i, j;
+ tran_low_t out[16 * 16] = { 0 };+ tran_low_t *outptr = out;
+ tran_low_t temp_in[16], temp_out[16];
+ uint16_t *const dest = CONVERT_TO_SHORTPTR(dest8);
+
+ // First transform rows. Since all non-zero dct coefficients are in
+ // upper-left 8x8 area, we only need to calculate first 8 rows here.
+ for (i = 0; i < 8; ++i) {+ vpx_highbd_idct16_c(input, outptr, bd);
+ input += 16;
+ outptr += 16;
+ }
+
+ // Then transform columns
+ for (i = 0; i < 16; ++i) {+ uint16_t *destT = dest;
+ for (j = 0; j < 16; ++j) temp_in[j] = out[j * 16 + i];
+ vpx_highbd_idct16_c(temp_in, temp_out, bd);
+ for (j = 0; j < 16; ++j) {+ destT[i] = highbd_clip_pixel_add(destT[i],
+ ROUND_POWER_OF_TWO(temp_out[j], 6), bd);
+ destT += stride;
+ }
+ }
+}
+
void vpx_highbd_idct16x16_10_add_c(const tran_low_t *input, uint8_t *dest8,
int stride, int bd) {int i, j;
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -673,6 +673,8 @@
add_proto qw/void vpx_highbd_idct16x16_256_add/, "const tran_low_t *input, uint8_t *dest, int stride, int bd";
+ add_proto qw/void vpx_highbd_idct16x16_38_add/, "const tran_low_t *input, uint8_t *dest, int stride, int bd";
+
add_proto qw/void vpx_highbd_idct16x16_10_add/, "const tran_low_t *input, uint8_t *dest, int stride, int bd";
} else {add_proto qw/void vpx_idct4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int stride";
@@ -729,6 +731,11 @@
add_proto qw/void vpx_highbd_idct16x16_256_add/, "const tran_low_t *input, uint8_t *dest, int stride, int bd";
specialize qw/vpx_highbd_idct16x16_256_add neon sse2/;
+
+ add_proto qw/void vpx_highbd_idct16x16_38_add/, "const tran_low_t *input, uint8_t *dest, int stride, int bd";
+ specialize qw/vpx_highbd_idct16x16_38_add neon sse2/;
+ $vpx_highbd_idct16x16_38_add_neon=vpx_highbd_idct16x16_256_add_neon;
+ $vpx_highbd_idct16x16_38_add_sse2=vpx_highbd_idct16x16_256_add_sse2;
add_proto qw/void vpx_highbd_idct16x16_10_add/, "const tran_low_t *input, uint8_t *dest, int stride, int bd";
specialize qw/vpx_highbd_idct16x16_10_add sse2/;