Remove unused convolve8_avg functions from aom_dsp
BUG=aomedia:1575
Change-Id: I21e8308b561b5054974e6e1c6aac4d9ef5615f11
diff --git a/aom_dsp/aom_convolve.c b/aom_dsp/aom_convolve.c
index a0ec277..38942fe 100644
--- a/aom_dsp/aom_convolve.c
+++ b/aom_dsp/aom_convolve.c
@@ -52,26 +52,6 @@
}
}
-static void convolve_avg_horiz(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const InterpKernel *x_filters, int x0_q4,
- int x_step_q4, int w, int h) {
- src -= SUBPEL_TAPS / 2 - 1;
- for (int y = 0; y < h; ++y) {
- int x_q4 = x0_q4;
- for (int x = 0; x < w; ++x) {
- const uint8_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
- const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
- const int sum = horz_scalar_product(src_x, x_filter);
- dst[x] = ROUND_POWER_OF_TWO(
- dst[x] + clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)), 1);
- x_q4 += x_step_q4;
- }
- src += src_stride;
- dst += dst_stride;
- }
-}
-
static void convolve_vert(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const InterpKernel *y_filters, int y0_q4,
@@ -92,29 +72,6 @@
}
}
-static void convolve_avg_vert(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const InterpKernel *y_filters, int y0_q4,
- int y_step_q4, int w, int h) {
- src -= src_stride * (SUBPEL_TAPS / 2 - 1);
-
- for (int x = 0; x < w; ++x) {
- int y_q4 = y0_q4;
- for (int y = 0; y < h; ++y) {
- const unsigned char *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
- const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
- const int sum = vert_scalar_product(src_y, src_stride, y_filter);
- dst[y * dst_stride] = ROUND_POWER_OF_TWO(
- dst[y * dst_stride] +
- clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)),
- 1);
- y_q4 += y_step_q4;
- }
- ++src;
- ++dst;
- }
-}
-
static void convolve(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
ptrdiff_t dst_stride, const InterpKernel *const x_filters,
int x0_q4, int x_step_q4,
@@ -174,21 +131,6 @@
w, h);
}
-void aom_convolve8_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4, int w,
- int h) {
- const InterpKernel *const filters_x = get_filter_base(filter_x);
- const int x0_q4 = get_filter_offset(filter_x, filters_x);
-
- (void)filter_y;
- (void)y_step_q4;
-
- convolve_avg_horiz(src, src_stride, dst, dst_stride, filters_x, x0_q4,
- x_step_q4, w, h);
-}
-
void aom_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
@@ -204,21 +146,6 @@
w, h);
}
-void aom_convolve8_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4, int w,
- int h) {
- const InterpKernel *const filters_y = get_filter_base(filter_y);
- const int y0_q4 = get_filter_offset(filter_y, filters_y);
-
- (void)filter_x;
- (void)x_step_q4;
-
- convolve_avg_vert(src, src_stride, dst, dst_stride, filters_y, y0_q4,
- y_step_q4, w, h);
-}
-
void aom_convolve8_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
ptrdiff_t dst_stride, const int16_t *filter_x,
int x_step_q4, const int16_t *filter_y, int y_step_q4,
@@ -233,21 +160,6 @@
filters_y, y0_q4, y_step_q4, w, h);
}
-void aom_convolve8_avg_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
- ptrdiff_t dst_stride, const int16_t *filter_x,
- int x_step_q4, const int16_t *filter_y, int y_step_q4,
- int w, int h) {
- /* Fixed size intermediate buffer places limits on parameters. */
- DECLARE_ALIGNED(16, uint8_t, temp[MAX_SB_SIZE * MAX_SB_SIZE]);
- assert(w <= MAX_SB_SIZE);
- assert(h <= MAX_SB_SIZE);
-
- aom_convolve8_c(src, src_stride, temp, MAX_SB_SIZE, filter_x, x_step_q4,
- filter_y, y_step_q4, w, h);
- aom_convolve_avg_c(temp, MAX_SB_SIZE, dst, dst_stride, NULL, 0, NULL, 0, w,
- h);
-}
-
void aom_convolve_copy_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
ptrdiff_t dst_stride, const int16_t *filter_x,
int filter_x_stride, const int16_t *filter_y,
@@ -548,29 +460,6 @@
}
}
-static void highbd_convolve_avg_horiz(const uint8_t *src8, ptrdiff_t src_stride,
- uint8_t *dst8, ptrdiff_t dst_stride,
- const InterpKernel *x_filters, int x0_q4,
- int x_step_q4, int w, int h, int bd) {
- uint16_t *src = CONVERT_TO_SHORTPTR(src8);
- uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
- src -= SUBPEL_TAPS / 2 - 1;
- for (int y = 0; y < h; ++y) {
- int x_q4 = x0_q4;
- for (int x = 0; x < w; ++x) {
- const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
- const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
- const int sum = highbd_horz_scalar_product(src_x, x_filter);
- dst[x] = ROUND_POWER_OF_TWO(
- dst[x] + clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd),
- 1);
- x_q4 += x_step_q4;
- }
- src += src_stride;
- dst += dst_stride;
- }
-}
-
static void highbd_convolve_vert(const uint8_t *src8, ptrdiff_t src_stride,
uint8_t *dst8, ptrdiff_t dst_stride,
const InterpKernel *y_filters, int y0_q4,
@@ -593,30 +482,6 @@
}
}
-static void highbd_convolve_avg_vert(const uint8_t *src8, ptrdiff_t src_stride,
- uint8_t *dst8, ptrdiff_t dst_stride,
- const InterpKernel *y_filters, int y0_q4,
- int y_step_q4, int w, int h, int bd) {
- uint16_t *src = CONVERT_TO_SHORTPTR(src8);
- uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
- src -= src_stride * (SUBPEL_TAPS / 2 - 1);
- for (int x = 0; x < w; ++x) {
- int y_q4 = y0_q4;
- for (int y = 0; y < h; ++y) {
- const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
- const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
- const int sum = highbd_vert_scalar_product(src_y, src_stride, y_filter);
- dst[y * dst_stride] = ROUND_POWER_OF_TWO(
- dst[y * dst_stride] +
- clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd),
- 1);
- y_q4 += y_step_q4;
- }
- ++src;
- ++dst;
- }
-}
-
static void highbd_convolve(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const InterpKernel *const x_filters, int x0_q4,
@@ -665,20 +530,6 @@
x_step_q4, w, h, bd);
}
-void aom_highbd_convolve8_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h, int bd) {
- const InterpKernel *const filters_x = get_filter_base(filter_x);
- const int x0_q4 = get_filter_offset(filter_x, filters_x);
- (void)filter_y;
- (void)y_step_q4;
-
- highbd_convolve_avg_horiz(src, src_stride, dst, dst_stride, filters_x, x0_q4,
- x_step_q4, w, h, bd);
-}
-
void aom_highbd_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
@@ -693,20 +544,6 @@
y_step_q4, w, h, bd);
}
-void aom_highbd_convolve8_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h, int bd) {
- const InterpKernel *const filters_y = get_filter_base(filter_y);
- const int y0_q4 = get_filter_offset(filter_y, filters_y);
- (void)filter_x;
- (void)x_step_q4;
-
- highbd_convolve_avg_vert(src, src_stride, dst, dst_stride, filters_y, y0_q4,
- y_step_q4, w, h, bd);
-}
-
void aom_highbd_convolve8_c(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
@@ -722,22 +559,6 @@
filters_y, y0_q4, y_step_q4, w, h, bd);
}
-void aom_highbd_convolve8_avg_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4, int w,
- int h, int bd) {
- // Fixed size intermediate buffer places limits on parameters.
- DECLARE_ALIGNED(16, uint16_t, temp[MAX_SB_SIZE * MAX_SB_SIZE]);
- assert(w <= MAX_SB_SIZE);
- assert(h <= MAX_SB_SIZE);
-
- aom_highbd_convolve8_c(src, src_stride, CONVERT_TO_BYTEPTR(temp), MAX_SB_SIZE,
- filter_x, x_step_q4, filter_y, y_step_q4, w, h, bd);
- aom_highbd_convolve_avg_c(CONVERT_TO_BYTEPTR(temp), MAX_SB_SIZE, dst,
- dst_stride, NULL, 0, NULL, 0, w, h, bd);
-}
-
void aom_highbd_convolve_copy_c(const uint8_t *src8, ptrdiff_t src_stride,
uint8_t *dst8, ptrdiff_t dst_stride,
const int16_t *filter_x, int filter_x_stride,
diff --git a/aom_dsp/aom_dsp.cmake b/aom_dsp/aom_dsp.cmake
index 10348e3..222f4b9 100644
--- a/aom_dsp/aom_dsp.cmake
+++ b/aom_dsp/aom_dsp.cmake
@@ -119,13 +119,9 @@
set(AOM_DSP_COMMON_INTRIN_DSPR2
"${AOM_ROOT}/aom_dsp/mips/common_dspr2.c"
"${AOM_ROOT}/aom_dsp/mips/common_dspr2.h"
- "${AOM_ROOT}/aom_dsp/mips/convolve2_avg_dspr2.c"
- "${AOM_ROOT}/aom_dsp/mips/convolve2_avg_horiz_dspr2.c"
"${AOM_ROOT}/aom_dsp/mips/convolve2_dspr2.c"
"${AOM_ROOT}/aom_dsp/mips/convolve2_horiz_dspr2.c"
"${AOM_ROOT}/aom_dsp/mips/convolve2_vert_dspr2.c"
- "${AOM_ROOT}/aom_dsp/mips/convolve8_avg_dspr2.c"
- "${AOM_ROOT}/aom_dsp/mips/convolve8_avg_horiz_dspr2.c"
"${AOM_ROOT}/aom_dsp/mips/convolve8_dspr2.c"
"${AOM_ROOT}/aom_dsp/mips/convolve8_horiz_dspr2.c"
"${AOM_ROOT}/aom_dsp/mips/convolve8_vert_dspr2.c"
@@ -136,9 +132,6 @@
"${AOM_ROOT}/aom_dsp/mips/inv_txfm_dspr2.h")
set(AOM_DSP_COMMON_INTRIN_MSA
- "${AOM_ROOT}/aom_dsp/mips/aom_convolve8_avg_horiz_msa.c"
- "${AOM_ROOT}/aom_dsp/mips/aom_convolve8_avg_msa.c"
- "${AOM_ROOT}/aom_dsp/mips/aom_convolve8_avg_vert_msa.c"
"${AOM_ROOT}/aom_dsp/mips/aom_convolve8_horiz_msa.c"
"${AOM_ROOT}/aom_dsp/mips/aom_convolve8_msa.c"
"${AOM_ROOT}/aom_dsp/mips/aom_convolve8_vert_msa.c"
diff --git a/aom_dsp/aom_dsp_rtcd_defs.pl b/aom_dsp/aom_dsp_rtcd_defs.pl
index 5e8bb4e..6f274c6 100755
--- a/aom_dsp/aom_dsp_rtcd_defs.pl
+++ b/aom_dsp/aom_dsp_rtcd_defs.pl
@@ -272,18 +272,12 @@
add_proto qw/void aom_convolve8/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
add_proto qw/void aom_convolve8_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
add_proto qw/void aom_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-add_proto qw/void aom_convolve8_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-add_proto qw/void aom_convolve8_avg_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-add_proto qw/void aom_convolve8_avg_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
specialize qw/aom_convolve_copy sse2 /;
specialize qw/aom_convolve_avg sse2 /;
specialize qw/aom_convolve8 sse2 ssse3/, "$avx2_ssse3";
specialize qw/aom_convolve8_horiz sse2 ssse3/, "$avx2_ssse3";
specialize qw/aom_convolve8_vert sse2 ssse3/, "$avx2_ssse3";
-specialize qw/aom_convolve8_avg sse2 ssse3/;
-specialize qw/aom_convolve8_avg_horiz sse2 ssse3/;
-specialize qw/aom_convolve8_avg_vert sse2 ssse3/;
add_proto qw/void aom_convolve8_add_src/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
add_proto qw/void aom_convolve8_add_src_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
@@ -305,9 +299,6 @@
specialize qw/aom_convolve8 neon dspr2 msa/;
specialize qw/aom_convolve8_horiz neon dspr2 msa/;
specialize qw/aom_convolve8_vert neon dspr2 msa/;
- specialize qw/aom_convolve8_avg neon dspr2 msa/;
- specialize qw/aom_convolve8_avg_horiz neon dspr2 msa/;
- specialize qw/aom_convolve8_avg_vert neon dspr2 msa/;
}
add_proto qw/void aom_highbd_convolve_copy/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
@@ -325,15 +316,6 @@
add_proto qw/void aom_highbd_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
specialize qw/aom_highbd_convolve8_vert avx2/, "$sse2_x86_64";
-add_proto qw/void aom_highbd_convolve8_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
-specialize qw/aom_highbd_convolve8_avg avx2/, "$sse2_x86_64";
-
-add_proto qw/void aom_highbd_convolve8_avg_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
-specialize qw/aom_highbd_convolve8_avg_horiz avx2/, "$sse2_x86_64";
-
-add_proto qw/void aom_highbd_convolve8_avg_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
-specialize qw/aom_highbd_convolve8_avg_vert avx2/, "$sse2_x86_64";
-
add_proto qw/void aom_highbd_convolve8_add_src/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
add_proto qw/void aom_highbd_convolve8_add_src_hip/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
diff --git a/aom_dsp/arm/aom_convolve8_avg_neon.c b/aom_dsp/arm/aom_convolve8_avg_neon.c
deleted file mode 100644
index 09429d6..0000000
--- a/aom_dsp/arm/aom_convolve8_avg_neon.c
+++ /dev/null
@@ -1,364 +0,0 @@
-/*
- * Copyright (c) 2016, Alliance for Open Media. All rights reserved
- *
- * This source code is subject to the terms of the BSD 2 Clause License and
- * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
- * was not distributed with this source code in the LICENSE file, you can
- * obtain it at www.aomedia.org/license/software. If the Alliance for Open
- * Media Patent License 1.0 was not distributed with this source code in the
- * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
- */
-
-#include <arm_neon.h>
-#include <assert.h>
-
-#include "./aom_config.h"
-#include "./aom_dsp_rtcd.h"
-#include "aom/aom_integer.h"
-#include "aom_ports/mem.h"
-
-static INLINE int32x4_t MULTIPLY_BY_Q0(int16x4_t dsrc0, int16x4_t dsrc1,
- int16x4_t dsrc2, int16x4_t dsrc3,
- int16x4_t dsrc4, int16x4_t dsrc5,
- int16x4_t dsrc6, int16x4_t dsrc7,
- int16x8_t q0s16) {
- int32x4_t qdst;
- int16x4_t d0s16, d1s16;
-
- d0s16 = vget_low_s16(q0s16);
- d1s16 = vget_high_s16(q0s16);
-
- qdst = vmull_lane_s16(dsrc0, d0s16, 0);
- qdst = vmlal_lane_s16(qdst, dsrc1, d0s16, 1);
- qdst = vmlal_lane_s16(qdst, dsrc2, d0s16, 2);
- qdst = vmlal_lane_s16(qdst, dsrc3, d0s16, 3);
- qdst = vmlal_lane_s16(qdst, dsrc4, d1s16, 0);
- qdst = vmlal_lane_s16(qdst, dsrc5, d1s16, 1);
- qdst = vmlal_lane_s16(qdst, dsrc6, d1s16, 2);
- qdst = vmlal_lane_s16(qdst, dsrc7, d1s16, 3);
- return qdst;
-}
-
-void aom_convolve8_avg_horiz_neon(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, // unused
- int y_step_q4, // unused
- int w, int h) {
- int width;
- const uint8_t *s;
- uint8_t *d;
- uint8x8_t d2u8, d3u8, d24u8, d25u8, d26u8, d27u8, d28u8, d29u8;
- uint32x2_t d2u32, d3u32, d6u32, d7u32, d28u32, d29u32, d30u32, d31u32;
- uint8x16_t q1u8, q3u8, q12u8, q13u8, q14u8, q15u8;
- int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d22s16, d23s16;
- int16x4_t d24s16, d25s16, d26s16, d27s16;
- uint16x4_t d2u16, d3u16, d4u16, d5u16, d16u16, d17u16, d18u16, d19u16;
- int16x8_t q0s16;
- uint16x8_t q1u16, q2u16, q8u16, q9u16, q10u16, q11u16, q12u16, q13u16;
- int32x4_t q1s32, q2s32, q14s32, q15s32;
- uint16x8x2_t q0x2u16;
- uint8x8x2_t d0x2u8, d1x2u8;
- uint32x2x2_t d0x2u32;
- uint16x4x2_t d0x2u16, d1x2u16;
- uint32x4x2_t q0x2u32;
-
- assert(x_step_q4 == 16);
-
- (void)x_step_q4;
- (void)y_step_q4;
- (void)filter_y;
-
- q0s16 = vld1q_s16(filter_x);
-
- src -= 3; // adjust for taps
- for (; h > 0; h -= 4) { // loop_horiz_v
- s = src;
- d24u8 = vld1_u8(s);
- s += src_stride;
- d25u8 = vld1_u8(s);
- s += src_stride;
- d26u8 = vld1_u8(s);
- s += src_stride;
- d27u8 = vld1_u8(s);
-
- q12u8 = vcombine_u8(d24u8, d25u8);
- q13u8 = vcombine_u8(d26u8, d27u8);
-
- q0x2u16 =
- vtrnq_u16(vreinterpretq_u16_u8(q12u8), vreinterpretq_u16_u8(q13u8));
- d24u8 = vreinterpret_u8_u16(vget_low_u16(q0x2u16.val[0]));
- d25u8 = vreinterpret_u8_u16(vget_high_u16(q0x2u16.val[0]));
- d26u8 = vreinterpret_u8_u16(vget_low_u16(q0x2u16.val[1]));
- d27u8 = vreinterpret_u8_u16(vget_high_u16(q0x2u16.val[1]));
- d0x2u8 = vtrn_u8(d24u8, d25u8);
- d1x2u8 = vtrn_u8(d26u8, d27u8);
-
- __builtin_prefetch(src + src_stride * 4);
- __builtin_prefetch(src + src_stride * 5);
-
- q8u16 = vmovl_u8(d0x2u8.val[0]);
- q9u16 = vmovl_u8(d0x2u8.val[1]);
- q10u16 = vmovl_u8(d1x2u8.val[0]);
- q11u16 = vmovl_u8(d1x2u8.val[1]);
-
- src += 7;
- d16u16 = vget_low_u16(q8u16);
- d17u16 = vget_high_u16(q8u16);
- d18u16 = vget_low_u16(q9u16);
- d19u16 = vget_high_u16(q9u16);
- q8u16 = vcombine_u16(d16u16, d18u16); // vswp 17 18
- q9u16 = vcombine_u16(d17u16, d19u16);
-
- d20s16 = vreinterpret_s16_u16(vget_low_u16(q10u16));
- d23s16 = vreinterpret_s16_u16(vget_high_u16(q10u16)); // vmov 23 21
- for (width = w; width > 0; width -= 4, src += 4, dst += 4) { // loop_horiz
- s = src;
- d28u32 = vld1_dup_u32((const uint32_t *)s);
- s += src_stride;
- d29u32 = vld1_dup_u32((const uint32_t *)s);
- s += src_stride;
- d31u32 = vld1_dup_u32((const uint32_t *)s);
- s += src_stride;
- d30u32 = vld1_dup_u32((const uint32_t *)s);
-
- __builtin_prefetch(src + 64);
-
- d0x2u16 =
- vtrn_u16(vreinterpret_u16_u32(d28u32), vreinterpret_u16_u32(d31u32));
- d1x2u16 =
- vtrn_u16(vreinterpret_u16_u32(d29u32), vreinterpret_u16_u32(d30u32));
- d0x2u8 = vtrn_u8(vreinterpret_u8_u16(d0x2u16.val[0]), // d28
- vreinterpret_u8_u16(d1x2u16.val[0])); // d29
- d1x2u8 = vtrn_u8(vreinterpret_u8_u16(d0x2u16.val[1]), // d31
- vreinterpret_u8_u16(d1x2u16.val[1])); // d30
-
- __builtin_prefetch(src + 64 + src_stride);
-
- q14u8 = vcombine_u8(d0x2u8.val[0], d0x2u8.val[1]);
- q15u8 = vcombine_u8(d1x2u8.val[1], d1x2u8.val[0]);
- q0x2u32 =
- vtrnq_u32(vreinterpretq_u32_u8(q14u8), vreinterpretq_u32_u8(q15u8));
-
- d28u8 = vreinterpret_u8_u32(vget_low_u32(q0x2u32.val[0]));
- d29u8 = vreinterpret_u8_u32(vget_high_u32(q0x2u32.val[0]));
- q12u16 = vmovl_u8(d28u8);
- q13u16 = vmovl_u8(d29u8);
-
- __builtin_prefetch(src + 64 + src_stride * 2);
-
- d = dst;
- d6u32 = vld1_lane_u32((const uint32_t *)d, d6u32, 0);
- d += dst_stride;
- d7u32 = vld1_lane_u32((const uint32_t *)d, d7u32, 0);
- d += dst_stride;
- d6u32 = vld1_lane_u32((const uint32_t *)d, d6u32, 1);
- d += dst_stride;
- d7u32 = vld1_lane_u32((const uint32_t *)d, d7u32, 1);
-
- d16s16 = vreinterpret_s16_u16(vget_low_u16(q8u16));
- d17s16 = vreinterpret_s16_u16(vget_high_u16(q8u16));
- d18s16 = vreinterpret_s16_u16(vget_low_u16(q9u16));
- d19s16 = vreinterpret_s16_u16(vget_high_u16(q9u16));
- d22s16 = vreinterpret_s16_u16(vget_low_u16(q11u16));
- d24s16 = vreinterpret_s16_u16(vget_low_u16(q12u16));
- d25s16 = vreinterpret_s16_u16(vget_high_u16(q12u16));
- d26s16 = vreinterpret_s16_u16(vget_low_u16(q13u16));
- d27s16 = vreinterpret_s16_u16(vget_high_u16(q13u16));
-
- q1s32 = MULTIPLY_BY_Q0(d16s16, d17s16, d20s16, d22s16, d18s16, d19s16,
- d23s16, d24s16, q0s16);
- q2s32 = MULTIPLY_BY_Q0(d17s16, d20s16, d22s16, d18s16, d19s16, d23s16,
- d24s16, d26s16, q0s16);
- q14s32 = MULTIPLY_BY_Q0(d20s16, d22s16, d18s16, d19s16, d23s16, d24s16,
- d26s16, d27s16, q0s16);
- q15s32 = MULTIPLY_BY_Q0(d22s16, d18s16, d19s16, d23s16, d24s16, d26s16,
- d27s16, d25s16, q0s16);
-
- __builtin_prefetch(src + 64 + src_stride * 3);
-
- d2u16 = vqrshrun_n_s32(q1s32, 7);
- d3u16 = vqrshrun_n_s32(q2s32, 7);
- d4u16 = vqrshrun_n_s32(q14s32, 7);
- d5u16 = vqrshrun_n_s32(q15s32, 7);
-
- q1u16 = vcombine_u16(d2u16, d3u16);
- q2u16 = vcombine_u16(d4u16, d5u16);
-
- d2u8 = vqmovn_u16(q1u16);
- d3u8 = vqmovn_u16(q2u16);
-
- d0x2u16 = vtrn_u16(vreinterpret_u16_u8(d2u8), vreinterpret_u16_u8(d3u8));
- d0x2u32 = vtrn_u32(vreinterpret_u32_u16(d0x2u16.val[0]),
- vreinterpret_u32_u16(d0x2u16.val[1]));
- d0x2u8 = vtrn_u8(vreinterpret_u8_u32(d0x2u32.val[0]),
- vreinterpret_u8_u32(d0x2u32.val[1]));
-
- q1u8 = vcombine_u8(d0x2u8.val[0], d0x2u8.val[1]);
- q3u8 = vreinterpretq_u8_u32(vcombine_u32(d6u32, d7u32));
-
- q1u8 = vrhaddq_u8(q1u8, q3u8);
-
- d2u32 = vreinterpret_u32_u8(vget_low_u8(q1u8));
- d3u32 = vreinterpret_u32_u8(vget_high_u8(q1u8));
-
- d = dst;
- vst1_lane_u32((uint32_t *)d, d2u32, 0);
- d += dst_stride;
- vst1_lane_u32((uint32_t *)d, d3u32, 0);
- d += dst_stride;
- vst1_lane_u32((uint32_t *)d, d2u32, 1);
- d += dst_stride;
- vst1_lane_u32((uint32_t *)d, d3u32, 1);
-
- q8u16 = q9u16;
- d20s16 = d23s16;
- q11u16 = q12u16;
- q9u16 = q13u16;
- d23s16 = vreinterpret_s16_u16(vget_high_u16(q11u16));
- }
- src += src_stride * 4 - w - 7;
- dst += dst_stride * 4 - w;
- }
- return;
-}
-
-void aom_convolve8_avg_vert_neon(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, // unused
- int x_step_q4, // unused
- const int16_t *filter_y, int y_step_q4, int w,
- int h) {
- int height;
- const uint8_t *s;
- uint8_t *d;
- uint8x8_t d2u8, d3u8;
- uint32x2_t d2u32, d3u32, d6u32, d7u32;
- uint32x2_t d16u32, d18u32, d20u32, d22u32, d24u32, d26u32;
- uint8x16_t q1u8, q3u8;
- int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16;
- int16x4_t d24s16, d25s16, d26s16, d27s16;
- uint16x4_t d2u16, d3u16, d4u16, d5u16;
- int16x8_t q0s16;
- uint16x8_t q1u16, q2u16, q8u16, q9u16, q10u16, q11u16, q12u16, q13u16;
- int32x4_t q1s32, q2s32, q14s32, q15s32;
-
- assert(y_step_q4 == 16);
-
- (void)x_step_q4;
- (void)y_step_q4;
- (void)filter_x;
-
- src -= src_stride * 3;
- q0s16 = vld1q_s16(filter_y);
- for (; w > 0; w -= 4, src += 4, dst += 4) { // loop_vert_h
- s = src;
- d16u32 = vld1_lane_u32((const uint32_t *)s, d16u32, 0);
- s += src_stride;
- d16u32 = vld1_lane_u32((const uint32_t *)s, d16u32, 1);
- s += src_stride;
- d18u32 = vld1_lane_u32((const uint32_t *)s, d18u32, 0);
- s += src_stride;
- d18u32 = vld1_lane_u32((const uint32_t *)s, d18u32, 1);
- s += src_stride;
- d20u32 = vld1_lane_u32((const uint32_t *)s, d20u32, 0);
- s += src_stride;
- d20u32 = vld1_lane_u32((const uint32_t *)s, d20u32, 1);
- s += src_stride;
- d22u32 = vld1_lane_u32((const uint32_t *)s, d22u32, 0);
- s += src_stride;
-
- q8u16 = vmovl_u8(vreinterpret_u8_u32(d16u32));
- q9u16 = vmovl_u8(vreinterpret_u8_u32(d18u32));
- q10u16 = vmovl_u8(vreinterpret_u8_u32(d20u32));
- q11u16 = vmovl_u8(vreinterpret_u8_u32(d22u32));
-
- d18s16 = vreinterpret_s16_u16(vget_low_u16(q9u16));
- d19s16 = vreinterpret_s16_u16(vget_high_u16(q9u16));
- d22s16 = vreinterpret_s16_u16(vget_low_u16(q11u16));
- d = dst;
- for (height = h; height > 0; height -= 4) { // loop_vert
- d24u32 = vld1_lane_u32((const uint32_t *)s, d24u32, 0);
- s += src_stride;
- d26u32 = vld1_lane_u32((const uint32_t *)s, d26u32, 0);
- s += src_stride;
- d26u32 = vld1_lane_u32((const uint32_t *)s, d26u32, 1);
- s += src_stride;
- d24u32 = vld1_lane_u32((const uint32_t *)s, d24u32, 1);
- s += src_stride;
-
- q12u16 = vmovl_u8(vreinterpret_u8_u32(d24u32));
- q13u16 = vmovl_u8(vreinterpret_u8_u32(d26u32));
-
- d6u32 = vld1_lane_u32((const uint32_t *)d, d6u32, 0);
- d += dst_stride;
- d6u32 = vld1_lane_u32((const uint32_t *)d, d6u32, 1);
- d += dst_stride;
- d7u32 = vld1_lane_u32((const uint32_t *)d, d7u32, 0);
- d += dst_stride;
- d7u32 = vld1_lane_u32((const uint32_t *)d, d7u32, 1);
- d -= dst_stride * 3;
-
- d16s16 = vreinterpret_s16_u16(vget_low_u16(q8u16));
- d17s16 = vreinterpret_s16_u16(vget_high_u16(q8u16));
- d20s16 = vreinterpret_s16_u16(vget_low_u16(q10u16));
- d21s16 = vreinterpret_s16_u16(vget_high_u16(q10u16));
- d24s16 = vreinterpret_s16_u16(vget_low_u16(q12u16));
- d25s16 = vreinterpret_s16_u16(vget_high_u16(q12u16));
- d26s16 = vreinterpret_s16_u16(vget_low_u16(q13u16));
- d27s16 = vreinterpret_s16_u16(vget_high_u16(q13u16));
-
- __builtin_prefetch(s);
- __builtin_prefetch(s + src_stride);
- q1s32 = MULTIPLY_BY_Q0(d16s16, d17s16, d18s16, d19s16, d20s16, d21s16,
- d22s16, d24s16, q0s16);
- __builtin_prefetch(s + src_stride * 2);
- __builtin_prefetch(s + src_stride * 3);
- q2s32 = MULTIPLY_BY_Q0(d17s16, d18s16, d19s16, d20s16, d21s16, d22s16,
- d24s16, d26s16, q0s16);
- __builtin_prefetch(d);
- __builtin_prefetch(d + dst_stride);
- q14s32 = MULTIPLY_BY_Q0(d18s16, d19s16, d20s16, d21s16, d22s16, d24s16,
- d26s16, d27s16, q0s16);
- __builtin_prefetch(d + dst_stride * 2);
- __builtin_prefetch(d + dst_stride * 3);
- q15s32 = MULTIPLY_BY_Q0(d19s16, d20s16, d21s16, d22s16, d24s16, d26s16,
- d27s16, d25s16, q0s16);
-
- d2u16 = vqrshrun_n_s32(q1s32, 7);
- d3u16 = vqrshrun_n_s32(q2s32, 7);
- d4u16 = vqrshrun_n_s32(q14s32, 7);
- d5u16 = vqrshrun_n_s32(q15s32, 7);
-
- q1u16 = vcombine_u16(d2u16, d3u16);
- q2u16 = vcombine_u16(d4u16, d5u16);
-
- d2u8 = vqmovn_u16(q1u16);
- d3u8 = vqmovn_u16(q2u16);
-
- q1u8 = vcombine_u8(d2u8, d3u8);
- q3u8 = vreinterpretq_u8_u32(vcombine_u32(d6u32, d7u32));
-
- q1u8 = vrhaddq_u8(q1u8, q3u8);
-
- d2u32 = vreinterpret_u32_u8(vget_low_u8(q1u8));
- d3u32 = vreinterpret_u32_u8(vget_high_u8(q1u8));
-
- vst1_lane_u32((uint32_t *)d, d2u32, 0);
- d += dst_stride;
- vst1_lane_u32((uint32_t *)d, d2u32, 1);
- d += dst_stride;
- vst1_lane_u32((uint32_t *)d, d3u32, 0);
- d += dst_stride;
- vst1_lane_u32((uint32_t *)d, d3u32, 1);
- d += dst_stride;
-
- q8u16 = q10u16;
- d18s16 = d22s16;
- d19s16 = d24s16;
- q10u16 = q13u16;
- d22s16 = d25s16;
- }
- }
- return;
-}
diff --git a/aom_dsp/arm/aom_convolve8_avg_neon_asm.asm b/aom_dsp/arm/aom_convolve8_avg_neon_asm.asm
deleted file mode 100644
index 80aef99..0000000
--- a/aom_dsp/arm/aom_convolve8_avg_neon_asm.asm
+++ /dev/null
@@ -1,295 +0,0 @@
-;
-; Copyright (c) 2016, Alliance for Open Media. All rights reserved
-;
-; This source code is subject to the terms of the BSD 2 Clause License and
-; the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
-; was not distributed with this source code in the LICENSE file, you can
-; obtain it at www.aomedia.org/license/software. If the Alliance for Open
-; Media Patent License 1.0 was not distributed with this source code in the
-; PATENTS file, you can obtain it at www.aomedia.org/license/patent.
-;
-
-;
-
-
- ; These functions are only valid when:
- ; x_step_q4 == 16
- ; w%4 == 0
- ; h%4 == 0
- ; taps == 8
- ; AV1_FILTER_WEIGHT == 128
- ; AV1_FILTER_SHIFT == 7
-
- EXPORT |aom_convolve8_avg_horiz_neon|
- EXPORT |aom_convolve8_avg_vert_neon|
- ARM
- REQUIRE8
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-
- ; Multiply and accumulate by q0
- MACRO
- MULTIPLY_BY_Q0 $dst, $src0, $src1, $src2, $src3, $src4, $src5, $src6, $src7
- vmull.s16 $dst, $src0, d0[0]
- vmlal.s16 $dst, $src1, d0[1]
- vmlal.s16 $dst, $src2, d0[2]
- vmlal.s16 $dst, $src3, d0[3]
- vmlal.s16 $dst, $src4, d1[0]
- vmlal.s16 $dst, $src5, d1[1]
- vmlal.s16 $dst, $src6, d1[2]
- vmlal.s16 $dst, $src7, d1[3]
- MEND
-
-; r0 const uint8_t *src
-; r1 int src_stride
-; r2 uint8_t *dst
-; r3 int dst_stride
-; sp[]const int16_t *filter_x
-; sp[]int x_step_q4
-; sp[]const int16_t *filter_y ; unused
-; sp[]int y_step_q4 ; unused
-; sp[]int w
-; sp[]int h
-
-|aom_convolve8_avg_horiz_neon| PROC
- push {r4-r10, lr}
-
- sub r0, r0, #3 ; adjust for taps
-
- ldr r5, [sp, #32] ; filter_x
- ldr r6, [sp, #48] ; w
- ldr r7, [sp, #52] ; h
-
- vld1.s16 {q0}, [r5] ; filter_x
-
- sub r8, r1, r1, lsl #2 ; -src_stride * 3
- add r8, r8, #4 ; -src_stride * 3 + 4
-
- sub r4, r3, r3, lsl #2 ; -dst_stride * 3
- add r4, r4, #4 ; -dst_stride * 3 + 4
-
- rsb r9, r6, r1, lsl #2 ; reset src for outer loop
- sub r9, r9, #7
- rsb r12, r6, r3, lsl #2 ; reset dst for outer loop
-
- mov r10, r6 ; w loop counter
-
-aom_convolve8_avg_loop_horiz_v
- vld1.8 {d24}, [r0], r1
- vld1.8 {d25}, [r0], r1
- vld1.8 {d26}, [r0], r1
- vld1.8 {d27}, [r0], r8
-
- vtrn.16 q12, q13
- vtrn.8 d24, d25
- vtrn.8 d26, d27
-
- pld [r0, r1, lsl #2]
-
- vmovl.u8 q8, d24
- vmovl.u8 q9, d25
- vmovl.u8 q10, d26
- vmovl.u8 q11, d27
-
- ; save a few instructions in the inner loop
- vswp d17, d18
- vmov d23, d21
-
- add r0, r0, #3
-
-aom_convolve8_avg_loop_horiz
- add r5, r0, #64
-
- vld1.32 {d28[]}, [r0], r1
- vld1.32 {d29[]}, [r0], r1
- vld1.32 {d31[]}, [r0], r1
- vld1.32 {d30[]}, [r0], r8
-
- pld [r5]
-
- vtrn.16 d28, d31
- vtrn.16 d29, d30
- vtrn.8 d28, d29
- vtrn.8 d31, d30
-
- pld [r5, r1]
-
- ; extract to s16
- vtrn.32 q14, q15
- vmovl.u8 q12, d28
- vmovl.u8 q13, d29
-
- pld [r5, r1, lsl #1]
-
- ; slightly out of order load to match the existing data
- vld1.u32 {d6[0]}, [r2], r3
- vld1.u32 {d7[0]}, [r2], r3
- vld1.u32 {d6[1]}, [r2], r3
- vld1.u32 {d7[1]}, [r2], r3
-
- sub r2, r2, r3, lsl #2 ; reset for store
-
- ; src[] * filter_x
- MULTIPLY_BY_Q0 q1, d16, d17, d20, d22, d18, d19, d23, d24
- MULTIPLY_BY_Q0 q2, d17, d20, d22, d18, d19, d23, d24, d26
- MULTIPLY_BY_Q0 q14, d20, d22, d18, d19, d23, d24, d26, d27
- MULTIPLY_BY_Q0 q15, d22, d18, d19, d23, d24, d26, d27, d25
-
- pld [r5, -r8]
-
- ; += 64 >> 7
- vqrshrun.s32 d2, q1, #7
- vqrshrun.s32 d3, q2, #7
- vqrshrun.s32 d4, q14, #7
- vqrshrun.s32 d5, q15, #7
-
- ; saturate
- vqmovn.u16 d2, q1
- vqmovn.u16 d3, q2
-
- ; transpose
- vtrn.16 d2, d3
- vtrn.32 d2, d3
- vtrn.8 d2, d3
-
- ; average the new value and the dst value
- vrhadd.u8 q1, q1, q3
-
- vst1.u32 {d2[0]}, [r2@32], r3
- vst1.u32 {d3[0]}, [r2@32], r3
- vst1.u32 {d2[1]}, [r2@32], r3
- vst1.u32 {d3[1]}, [r2@32], r4
-
- vmov q8, q9
- vmov d20, d23
- vmov q11, q12
- vmov q9, q13
-
- subs r6, r6, #4 ; w -= 4
- bgt aom_convolve8_avg_loop_horiz
-
- ; outer loop
- mov r6, r10 ; restore w counter
- add r0, r0, r9 ; src += src_stride * 4 - w
- add r2, r2, r12 ; dst += dst_stride * 4 - w
- subs r7, r7, #4 ; h -= 4
- bgt aom_convolve8_avg_loop_horiz_v
-
- pop {r4-r10, pc}
-
- ENDP
-
-|aom_convolve8_avg_vert_neon| PROC
- push {r4-r8, lr}
-
- ; adjust for taps
- sub r0, r0, r1
- sub r0, r0, r1, lsl #1
-
- ldr r4, [sp, #32] ; filter_y
- ldr r6, [sp, #40] ; w
- ldr lr, [sp, #44] ; h
-
- vld1.s16 {q0}, [r4] ; filter_y
-
- lsl r1, r1, #1
- lsl r3, r3, #1
-
-aom_convolve8_avg_loop_vert_h
- mov r4, r0
- add r7, r0, r1, asr #1
- mov r5, r2
- add r8, r2, r3, asr #1
- mov r12, lr ; h loop counter
-
- vld1.u32 {d16[0]}, [r4], r1
- vld1.u32 {d16[1]}, [r7], r1
- vld1.u32 {d18[0]}, [r4], r1
- vld1.u32 {d18[1]}, [r7], r1
- vld1.u32 {d20[0]}, [r4], r1
- vld1.u32 {d20[1]}, [r7], r1
- vld1.u32 {d22[0]}, [r4], r1
-
- vmovl.u8 q8, d16
- vmovl.u8 q9, d18
- vmovl.u8 q10, d20
- vmovl.u8 q11, d22
-
-aom_convolve8_avg_loop_vert
- ; always process a 4x4 block at a time
- vld1.u32 {d24[0]}, [r7], r1
- vld1.u32 {d26[0]}, [r4], r1
- vld1.u32 {d26[1]}, [r7], r1
- vld1.u32 {d24[1]}, [r4], r1
-
- ; extract to s16
- vmovl.u8 q12, d24
- vmovl.u8 q13, d26
-
- vld1.u32 {d6[0]}, [r5@32], r3
- vld1.u32 {d6[1]}, [r8@32], r3
- vld1.u32 {d7[0]}, [r5@32], r3
- vld1.u32 {d7[1]}, [r8@32], r3
-
- pld [r7]
- pld [r4]
-
- ; src[] * filter_y
- MULTIPLY_BY_Q0 q1, d16, d17, d18, d19, d20, d21, d22, d24
-
- pld [r7, r1]
- pld [r4, r1]
-
- MULTIPLY_BY_Q0 q2, d17, d18, d19, d20, d21, d22, d24, d26
-
- pld [r5]
- pld [r8]
-
- MULTIPLY_BY_Q0 q14, d18, d19, d20, d21, d22, d24, d26, d27
-
- pld [r5, r3]
- pld [r8, r3]
-
- MULTIPLY_BY_Q0 q15, d19, d20, d21, d22, d24, d26, d27, d25
-
- ; += 64 >> 7
- vqrshrun.s32 d2, q1, #7
- vqrshrun.s32 d3, q2, #7
- vqrshrun.s32 d4, q14, #7
- vqrshrun.s32 d5, q15, #7
-
- ; saturate
- vqmovn.u16 d2, q1
- vqmovn.u16 d3, q2
-
- ; average the new value and the dst value
- vrhadd.u8 q1, q1, q3
-
- sub r5, r5, r3, lsl #1 ; reset for store
- sub r8, r8, r3, lsl #1
-
- vst1.u32 {d2[0]}, [r5@32], r3
- vst1.u32 {d2[1]}, [r8@32], r3
- vst1.u32 {d3[0]}, [r5@32], r3
- vst1.u32 {d3[1]}, [r8@32], r3
-
- vmov q8, q10
- vmov d18, d22
- vmov d19, d24
- vmov q10, q13
- vmov d22, d25
-
- subs r12, r12, #4 ; h -= 4
- bgt aom_convolve8_avg_loop_vert
-
- ; outer loop
- add r0, r0, #4
- add r2, r2, #4
- subs r6, r6, #4 ; w -= 4
- bgt aom_convolve8_avg_loop_vert_h
-
- pop {r4-r8, pc}
-
- ENDP
- END
diff --git a/aom_dsp/arm/aom_convolve_neon.c b/aom_dsp/arm/aom_convolve_neon.c
index 6c2997e..faaac7c 100644
--- a/aom_dsp/arm/aom_convolve_neon.c
+++ b/aom_dsp/arm/aom_convolve_neon.c
@@ -43,24 +43,3 @@
aom_convolve8_vert_neon(temp + 64 * 3, 64, dst, dst_stride, filter_x,
x_step_q4, filter_y, y_step_q4, w, h);
}
-
-void aom_convolve8_avg_neon(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4, int w,
- int h) {
- DECLARE_ALIGNED(8, uint8_t, temp[64 * 72]);
- int intermediate_height = h + 7;
-
- assert(y_step_q4 == 16);
- assert(x_step_q4 == 16);
-
- /* This implementation has the same issues as above. In addition, we only want
- * to average the values after both passes.
- */
- aom_convolve8_horiz_neon(src - src_stride * 3, src_stride, temp, 64, filter_x,
- x_step_q4, filter_y, y_step_q4, w,
- intermediate_height);
- aom_convolve8_avg_vert_neon(temp + 64 * 3, 64, dst, dst_stride, filter_x,
- x_step_q4, filter_y, y_step_q4, w, h);
-}
diff --git a/aom_dsp/mips/aom_convolve8_avg_horiz_msa.c b/aom_dsp/mips/aom_convolve8_avg_horiz_msa.c
deleted file mode 100644
index 847394a..0000000
--- a/aom_dsp/mips/aom_convolve8_avg_horiz_msa.c
+++ /dev/null
@@ -1,704 +0,0 @@
-/*
- * Copyright (c) 2016, Alliance for Open Media. All rights reserved
- *
- * This source code is subject to the terms of the BSD 2 Clause License and
- * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
- * was not distributed with this source code in the LICENSE file, you can
- * obtain it at www.aomedia.org/license/software. If the Alliance for Open
- * Media Patent License 1.0 was not distributed with this source code in the
- * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
- */
-
-#include <assert.h>
-#include "./aom_dsp_rtcd.h"
-#include "aom_dsp/mips/aom_convolve_msa.h"
-
-static void common_hz_8t_and_aver_dst_4x4_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter) {
- v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
- v16u8 dst0, dst1, dst2, dst3, res2, res3;
- v16u8 mask0, mask1, mask2, mask3;
- v8i16 filt, res0, res1;
-
- mask0 = LD_UB(&mc_filt_mask_arr[16]);
- src -= 3;
-
- /* rearranging filter */
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- XORI_B4_128_SB(src0, src1, src2, src3);
- HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3,
- filt0, filt1, filt2, filt3, res0, res1);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- SRARI_H2_SH(res0, res1, FILTER_BITS);
- SAT_SH2_SH(res0, res1, 7);
- PCKEV_B2_UB(res0, res0, res1, res1, res2, res3);
- ILVR_W2_UB(dst1, dst0, dst3, dst2, dst0, dst2);
- XORI_B2_128_UB(res2, res3);
- AVER_UB2_UB(res2, dst0, res3, dst2, res2, res3);
- ST4x4_UB(res2, res3, 0, 1, 0, 1, dst, dst_stride);
-}
-
-static void common_hz_8t_and_aver_dst_4x8_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter) {
- v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
- v16u8 mask0, mask1, mask2, mask3, res0, res1, res2, res3;
- v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
- v8i16 filt, vec0, vec1, vec2, vec3;
-
- mask0 = LD_UB(&mc_filt_mask_arr[16]);
- src -= 3;
-
- /* rearranging filter */
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- XORI_B4_128_SB(src0, src1, src2, src3);
- src += (4 * src_stride);
- LD_UB8(dst, dst_stride, dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
- HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3,
- filt0, filt1, filt2, filt3, vec0, vec1);
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- XORI_B4_128_SB(src0, src1, src2, src3);
- HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3,
- filt0, filt1, filt2, filt3, vec2, vec3);
- SRARI_H4_SH(vec0, vec1, vec2, vec3, FILTER_BITS);
- SAT_SH4_SH(vec0, vec1, vec2, vec3, 7);
- PCKEV_B4_UB(vec0, vec0, vec1, vec1, vec2, vec2, vec3, vec3, res0, res1, res2,
- res3);
- ILVR_D2_UB(res1, res0, res3, res2, res0, res2);
- XORI_B2_128_UB(res0, res2);
- ILVR_W4_UB(dst1, dst0, dst3, dst2, dst5, dst4, dst7, dst6, dst0, dst2, dst4,
- dst6);
- ILVR_D2_UB(dst2, dst0, dst6, dst4, dst0, dst4);
- AVER_UB2_UB(res0, dst0, res2, dst4, res0, res2);
- ST4x8_UB(res0, res2, dst, dst_stride);
-}
-
-static void common_hz_8t_and_aver_dst_4w_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride, int8_t *filter,
- int32_t height) {
- if (4 == height) {
- common_hz_8t_and_aver_dst_4x4_msa(src, src_stride, dst, dst_stride, filter);
- } else if (8 == height) {
- common_hz_8t_and_aver_dst_4x8_msa(src, src_stride, dst, dst_stride, filter);
- }
-}
-
-static void common_hz_8t_and_aver_dst_8w_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride, int8_t *filter,
- int32_t height) {
- int32_t loop_cnt;
- v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
- v16u8 mask0, mask1, mask2, mask3, dst0, dst1, dst2, dst3;
- v8i16 filt, out0, out1, out2, out3;
-
- mask0 = LD_UB(&mc_filt_mask_arr[0]);
- src -= 3;
-
- /* rearranging filter */
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- XORI_B4_128_SB(src0, src1, src2, src3);
- src += (4 * src_stride);
- HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2,
- mask3, filt0, filt1, filt2, filt3, out0, out1,
- out2, out3);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
- SAT_SH4_SH(out0, out1, out2, out3, 7);
- CONVERT_UB_AVG_ST8x4_UB(out0, out1, out2, out3, dst0, dst1, dst2, dst3, dst,
- dst_stride);
- dst += (4 * dst_stride);
- }
-}
-
-static void common_hz_8t_and_aver_dst_16w_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter, int32_t height) {
- int32_t loop_cnt;
- v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
- v16u8 mask0, mask1, mask2, mask3, dst0, dst1;
- v8i16 filt, out0, out1, out2, out3;
- v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
- v8i16 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15;
-
- mask0 = LD_UB(&mc_filt_mask_arr[0]);
- src -= 3;
-
- /* rearranging filter */
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- for (loop_cnt = height >> 1; loop_cnt--;) {
- LD_SB2(src, src_stride, src0, src2);
- LD_SB2(src + 8, src_stride, src1, src3);
- src += (2 * src_stride);
-
- XORI_B4_128_SB(src0, src1, src2, src3);
- VSHF_B4_SH(src0, src0, mask0, mask1, mask2, mask3, vec0, vec4, vec8, vec12);
- VSHF_B4_SH(src1, src1, mask0, mask1, mask2, mask3, vec1, vec5, vec9, vec13);
- VSHF_B4_SH(src2, src2, mask0, mask1, mask2, mask3, vec2, vec6, vec10,
- vec14);
- VSHF_B4_SH(src3, src3, mask0, mask1, mask2, mask3, vec3, vec7, vec11,
- vec15);
- DOTP_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
- vec2, vec3);
- DOTP_SB4_SH(vec8, vec9, vec10, vec11, filt2, filt2, filt2, filt2, vec8,
- vec9, vec10, vec11);
- DPADD_SB4_SH(vec4, vec5, vec6, vec7, filt1, filt1, filt1, filt1, vec0, vec1,
- vec2, vec3);
- DPADD_SB4_SH(vec12, vec13, vec14, vec15, filt3, filt3, filt3, filt3, vec8,
- vec9, vec10, vec11);
- ADDS_SH4_SH(vec0, vec8, vec1, vec9, vec2, vec10, vec3, vec11, out0, out1,
- out2, out3);
- LD_UB2(dst, dst_stride, dst0, dst1);
- SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
- SAT_SH4_SH(out0, out1, out2, out3, 7);
- PCKEV_XORI128_AVG_ST_UB(out1, out0, dst0, dst);
- dst += dst_stride;
- PCKEV_XORI128_AVG_ST_UB(out3, out2, dst1, dst);
- dst += dst_stride;
- }
-}
-
-static void common_hz_8t_and_aver_dst_32w_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter, int32_t height) {
- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
- v16u8 dst1, dst2, mask0, mask1, mask2, mask3;
- v8i16 filt, out0, out1, out2, out3;
- v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
- v8i16 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15;
-
- mask0 = LD_UB(&mc_filt_mask_arr[0]);
- src -= 3;
-
- /* rearranging filter */
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- for (loop_cnt = height; loop_cnt--;) {
- src0 = LD_SB(src);
- src2 = LD_SB(src + 16);
- src3 = LD_SB(src + 24);
- src1 = __msa_sldi_b(src2, src0, 8);
- src += src_stride;
-
- XORI_B4_128_SB(src0, src1, src2, src3);
- VSHF_B4_SH(src0, src0, mask0, mask1, mask2, mask3, vec0, vec4, vec8, vec12);
- VSHF_B4_SH(src1, src1, mask0, mask1, mask2, mask3, vec1, vec5, vec9, vec13);
- VSHF_B4_SH(src2, src2, mask0, mask1, mask2, mask3, vec2, vec6, vec10,
- vec14);
- VSHF_B4_SH(src3, src3, mask0, mask1, mask2, mask3, vec3, vec7, vec11,
- vec15);
- DOTP_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
- vec2, vec3);
- DOTP_SB4_SH(vec8, vec9, vec10, vec11, filt2, filt2, filt2, filt2, vec8,
- vec9, vec10, vec11);
- DPADD_SB4_SH(vec4, vec5, vec6, vec7, filt1, filt1, filt1, filt1, vec0, vec1,
- vec2, vec3);
- DPADD_SB4_SH(vec12, vec13, vec14, vec15, filt3, filt3, filt3, filt3, vec8,
- vec9, vec10, vec11);
- ADDS_SH4_SH(vec0, vec8, vec1, vec9, vec2, vec10, vec3, vec11, out0, out1,
- out2, out3);
- SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
- SAT_SH4_SH(out0, out1, out2, out3, 7);
- LD_UB2(dst, 16, dst1, dst2);
- PCKEV_XORI128_AVG_ST_UB(out1, out0, dst1, dst);
- PCKEV_XORI128_AVG_ST_UB(out3, out2, dst2, dst + 16);
- dst += dst_stride;
- }
-}
-
-static void common_hz_8t_and_aver_dst_64w_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter, int32_t height) {
- uint32_t loop_cnt, cnt;
- v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
- v16u8 dst1, dst2, mask0, mask1, mask2, mask3;
- v8i16 filt, out0, out1, out2, out3;
- v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
- v8i16 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15;
-
- mask0 = LD_UB(&mc_filt_mask_arr[0]);
- src -= 3;
-
- /* rearranging filter */
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- for (loop_cnt = height; loop_cnt--;) {
- for (cnt = 0; cnt < 2; ++cnt) {
- src0 = LD_SB(&src[cnt << 5]);
- src2 = LD_SB(&src[16 + (cnt << 5)]);
- src3 = LD_SB(&src[24 + (cnt << 5)]);
- src1 = __msa_sldi_b(src2, src0, 8);
-
- XORI_B4_128_SB(src0, src1, src2, src3);
- VSHF_B4_SH(src0, src0, mask0, mask1, mask2, mask3, vec0, vec4, vec8,
- vec12);
- VSHF_B4_SH(src1, src1, mask0, mask1, mask2, mask3, vec1, vec5, vec9,
- vec13);
- VSHF_B4_SH(src2, src2, mask0, mask1, mask2, mask3, vec2, vec6, vec10,
- vec14);
- VSHF_B4_SH(src3, src3, mask0, mask1, mask2, mask3, vec3, vec7, vec11,
- vec15);
- DOTP_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0,
- vec1, vec2, vec3);
- DOTP_SB4_SH(vec8, vec9, vec10, vec11, filt2, filt2, filt2, filt2, vec8,
- vec9, vec10, vec11);
- DPADD_SB4_SH(vec4, vec5, vec6, vec7, filt1, filt1, filt1, filt1, vec0,
- vec1, vec2, vec3);
- DPADD_SB4_SH(vec12, vec13, vec14, vec15, filt3, filt3, filt3, filt3, vec8,
- vec9, vec10, vec11);
- ADDS_SH4_SH(vec0, vec8, vec1, vec9, vec2, vec10, vec3, vec11, out0, out1,
- out2, out3);
- SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
- SAT_SH4_SH(out0, out1, out2, out3, 7);
- LD_UB2(&dst[cnt << 5], 16, dst1, dst2);
- PCKEV_XORI128_AVG_ST_UB(out1, out0, dst1, &dst[cnt << 5]);
- PCKEV_XORI128_AVG_ST_UB(out3, out2, dst2, &dst[16 + (cnt << 5)]);
- }
-
- src += src_stride;
- dst += dst_stride;
- }
-}
-
-static void common_hz_2t_and_aver_dst_4x4_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter) {
- v16i8 src0, src1, src2, src3, mask;
- v16u8 filt0, dst0, dst1, dst2, dst3, vec0, vec1, res0, res1;
- v8u16 vec2, vec3, filt;
-
- mask = LD_SB(&mc_filt_mask_arr[16]);
-
- /* rearranging filter */
- filt = LD_UH(filter);
- filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- VSHF_B2_UB(src0, src1, src2, src3, mask, mask, vec0, vec1);
- DOTP_UB2_UH(vec0, vec1, filt0, filt0, vec2, vec3);
- SRARI_H2_UH(vec2, vec3, FILTER_BITS);
- PCKEV_B2_UB(vec2, vec2, vec3, vec3, res0, res1);
- ILVR_W2_UB(dst1, dst0, dst3, dst2, dst0, dst2);
- AVER_UB2_UB(res0, dst0, res1, dst2, res0, res1);
- ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
-}
-
-static void common_hz_2t_and_aver_dst_4x8_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter) {
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
- v16u8 filt0, vec0, vec1, vec2, vec3, res0, res1, res2, res3;
- v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
- v8u16 vec4, vec5, vec6, vec7, filt;
-
- mask = LD_SB(&mc_filt_mask_arr[16]);
-
- /* rearranging filter */
- filt = LD_UH(filter);
- filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
- LD_UB8(dst, dst_stride, dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
- VSHF_B2_UB(src0, src1, src2, src3, mask, mask, vec0, vec1);
- VSHF_B2_UB(src4, src5, src6, src7, mask, mask, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec4, vec5,
- vec6, vec7);
- SRARI_H4_UH(vec4, vec5, vec6, vec7, FILTER_BITS);
- PCKEV_B4_UB(vec4, vec4, vec5, vec5, vec6, vec6, vec7, vec7, res0, res1, res2,
- res3);
- ILVR_W4_UB(dst1, dst0, dst3, dst2, dst5, dst4, dst7, dst6, dst0, dst2, dst4,
- dst6);
- AVER_UB4_UB(res0, dst0, res1, dst2, res2, dst4, res3, dst6, res0, res1, res2,
- res3);
- ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
- dst += (4 * dst_stride);
- ST4x4_UB(res2, res3, 0, 1, 0, 1, dst, dst_stride);
-}
-
-static void common_hz_2t_and_aver_dst_4w_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride, int8_t *filter,
- int32_t height) {
- if (4 == height) {
- common_hz_2t_and_aver_dst_4x4_msa(src, src_stride, dst, dst_stride, filter);
- } else if (8 == height) {
- common_hz_2t_and_aver_dst_4x8_msa(src, src_stride, dst, dst_stride, filter);
- }
-}
-
-static void common_hz_2t_and_aver_dst_8x4_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter) {
- v16i8 src0, src1, src2, src3, mask;
- v16u8 filt0, dst0, dst1, dst2, dst3;
- v8u16 vec0, vec1, vec2, vec3, filt;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_UH(filter);
- filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
- vec2, vec3);
- SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- PCKEV_AVG_ST8x4_UB(vec0, dst0, vec1, dst1, vec2, dst2, vec3, dst3, dst,
- dst_stride);
-}
-
-static void common_hz_2t_and_aver_dst_8x8mult_msa(
- const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {
- v16i8 src0, src1, src2, src3, mask;
- v16u8 filt0, dst0, dst1, dst2, dst3;
- v8u16 vec0, vec1, vec2, vec3, filt;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_UH(filter);
- filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- src += (4 * src_stride);
- VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
- vec2, vec3);
- SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- src += (4 * src_stride);
- PCKEV_AVG_ST8x4_UB(vec0, dst0, vec1, dst1, vec2, dst2, vec3, dst3, dst,
- dst_stride);
- dst += (4 * dst_stride);
-
- VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
- vec2, vec3);
- SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- PCKEV_AVG_ST8x4_UB(vec0, dst0, vec1, dst1, vec2, dst2, vec3, dst3, dst,
- dst_stride);
- dst += (4 * dst_stride);
-
- if (16 == height) {
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- src += (4 * src_stride);
-
- VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
- vec2, vec3);
- SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- PCKEV_AVG_ST8x4_UB(vec0, dst0, vec1, dst1, vec2, dst2, vec3, dst3, dst,
- dst_stride);
- dst += (4 * dst_stride);
-
- VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
- vec2, vec3);
- SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- PCKEV_AVG_ST8x4_UB(vec0, dst0, vec1, dst1, vec2, dst2, vec3, dst3, dst,
- dst_stride);
- }
-}
-
-static void common_hz_2t_and_aver_dst_8w_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride, int8_t *filter,
- int32_t height) {
- if (4 == height) {
- common_hz_2t_and_aver_dst_8x4_msa(src, src_stride, dst, dst_stride, filter);
- } else {
- common_hz_2t_and_aver_dst_8x8mult_msa(src, src_stride, dst, dst_stride,
- filter, height);
- }
-}
-
-static void common_hz_2t_and_aver_dst_16w_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter, int32_t height) {
- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
- v16u8 filt0, dst0, dst1, dst2, dst3;
- v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
- v8u16 res0, res1, res2, res3, res4, res5, res6, res7, filt;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_UH(filter);
- filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- LD_SB4(src, src_stride, src0, src2, src4, src6);
- LD_SB4(src + 8, src_stride, src1, src3, src5, src7);
- src += (4 * src_stride);
-
- VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3);
- VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5);
- VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, res0, res1,
- res2, res3);
- DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, res4, res5,
- res6, res7);
- SRARI_H4_UH(res0, res1, res2, res3, FILTER_BITS);
- SRARI_H4_UH(res4, res5, res6, res7, FILTER_BITS);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- PCKEV_AVG_ST_UB(res1, res0, dst0, dst);
- dst += dst_stride;
- PCKEV_AVG_ST_UB(res3, res2, dst1, dst);
- dst += dst_stride;
- PCKEV_AVG_ST_UB(res5, res4, dst2, dst);
- dst += dst_stride;
- PCKEV_AVG_ST_UB(res7, res6, dst3, dst);
- dst += dst_stride;
-
- for (loop_cnt = (height >> 2) - 1; loop_cnt--;) {
- LD_SB4(src, src_stride, src0, src2, src4, src6);
- LD_SB4(src + 8, src_stride, src1, src3, src5, src7);
- src += (4 * src_stride);
-
- VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3);
- VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5);
- VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, res0, res1,
- res2, res3);
- DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, res4, res5,
- res6, res7);
- SRARI_H4_UH(res0, res1, res2, res3, FILTER_BITS);
- SRARI_H4_UH(res4, res5, res6, res7, FILTER_BITS);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- PCKEV_AVG_ST_UB(res1, res0, dst0, dst);
- dst += dst_stride;
- PCKEV_AVG_ST_UB(res3, res2, dst1, dst);
- dst += dst_stride;
- PCKEV_AVG_ST_UB(res5, res4, dst2, dst);
- dst += dst_stride;
- PCKEV_AVG_ST_UB(res7, res6, dst3, dst);
- dst += dst_stride;
- }
-}
-
-static void common_hz_2t_and_aver_dst_32w_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter, int32_t height) {
- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
- v16u8 filt0, dst0, dst1, dst2, dst3;
- v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
- v8u16 res0, res1, res2, res3, res4, res5, res6, res7, filt;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_UH(filter);
- filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- for (loop_cnt = (height >> 1); loop_cnt--;) {
- src0 = LD_SB(src);
- src2 = LD_SB(src + 16);
- src3 = LD_SB(src + 24);
- src1 = __msa_sldi_b(src2, src0, 8);
- src += src_stride;
- src4 = LD_SB(src);
- src6 = LD_SB(src + 16);
- src7 = LD_SB(src + 24);
- src5 = __msa_sldi_b(src6, src4, 8);
- src += src_stride;
-
- VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3);
- VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5);
- VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, res0, res1,
- res2, res3);
- DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, res4, res5,
- res6, res7);
- SRARI_H4_UH(res0, res1, res2, res3, FILTER_BITS);
- SRARI_H4_UH(res4, res5, res6, res7, FILTER_BITS);
- LD_UB2(dst, 16, dst0, dst1);
- PCKEV_AVG_ST_UB(res1, res0, dst0, dst);
- PCKEV_AVG_ST_UB(res3, res2, dst1, (dst + 16));
- dst += dst_stride;
- LD_UB2(dst, 16, dst2, dst3);
- PCKEV_AVG_ST_UB(res5, res4, dst2, dst);
- PCKEV_AVG_ST_UB(res7, res6, dst3, (dst + 16));
- dst += dst_stride;
- }
-}
-
-static void common_hz_2t_and_aver_dst_64w_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter, int32_t height) {
- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
- v16u8 filt0, dst0, dst1, dst2, dst3;
- v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
- v8u16 out0, out1, out2, out3, out4, out5, out6, out7, filt;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_UH(filter);
- filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- for (loop_cnt = height; loop_cnt--;) {
- LD_SB4(src, 16, src0, src2, src4, src6);
- src7 = LD_SB(src + 56);
- SLDI_B3_SB(src2, src4, src6, src0, src2, src4, src1, src3, src5, 8);
- src += src_stride;
-
- VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3);
- VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5);
- VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, out0, out1,
- out2, out3);
- DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, out4, out5,
- out6, out7);
- SRARI_H4_UH(out0, out1, out2, out3, FILTER_BITS);
- SRARI_H4_UH(out4, out5, out6, out7, FILTER_BITS);
- LD_UB4(dst, 16, dst0, dst1, dst2, dst3);
- PCKEV_AVG_ST_UB(out1, out0, dst0, dst);
- PCKEV_AVG_ST_UB(out3, out2, dst1, dst + 16);
- PCKEV_AVG_ST_UB(out5, out4, dst2, dst + 32);
- PCKEV_AVG_ST_UB(out7, out6, dst3, dst + 48);
- dst += dst_stride;
- }
-}
-
-void aom_convolve8_avg_horiz_msa(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4, int w,
- int h) {
- int8_t cnt, filt_hor[8];
-
- assert(x_step_q4 == 16);
- assert(((const int32_t *)filter_x)[1] != 0x800000);
-
- for (cnt = 0; cnt < 8; ++cnt) {
- filt_hor[cnt] = filter_x[cnt];
- }
-
- if (((const int32_t *)filter_x)[0] == 0) {
- switch (w) {
- case 4:
- common_hz_2t_and_aver_dst_4w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, &filt_hor[3], h);
- break;
- case 8:
- common_hz_2t_and_aver_dst_8w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, &filt_hor[3], h);
- break;
- case 16:
- common_hz_2t_and_aver_dst_16w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, &filt_hor[3], h);
- break;
- case 32:
- common_hz_2t_and_aver_dst_32w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, &filt_hor[3], h);
- break;
- case 64:
- common_hz_2t_and_aver_dst_64w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, &filt_hor[3], h);
- break;
- default:
- aom_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride, filter_x,
- x_step_q4, filter_y, y_step_q4, w, h);
- break;
- }
- } else {
- switch (w) {
- case 4:
- common_hz_8t_and_aver_dst_4w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, filt_hor, h);
- break;
- case 8:
- common_hz_8t_and_aver_dst_8w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, filt_hor, h);
- break;
- case 16:
- common_hz_8t_and_aver_dst_16w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, filt_hor, h);
- break;
- case 32:
- common_hz_8t_and_aver_dst_32w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, filt_hor, h);
- break;
- case 64:
- common_hz_8t_and_aver_dst_64w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, filt_hor, h);
- break;
- default:
- aom_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride, filter_x,
- x_step_q4, filter_y, y_step_q4, w, h);
- break;
- }
- }
-}
diff --git a/aom_dsp/mips/aom_convolve8_avg_msa.c b/aom_dsp/mips/aom_convolve8_avg_msa.c
deleted file mode 100644
index bed600d..0000000
--- a/aom_dsp/mips/aom_convolve8_avg_msa.c
+++ /dev/null
@@ -1,605 +0,0 @@
-/*
- * Copyright (c) 2016, Alliance for Open Media. All rights reserved
- *
- * This source code is subject to the terms of the BSD 2 Clause License and
- * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
- * was not distributed with this source code in the LICENSE file, you can
- * obtain it at www.aomedia.org/license/software. If the Alliance for Open
- * Media Patent License 1.0 was not distributed with this source code in the
- * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
- */
-
-#include <assert.h>
-#include "./aom_dsp_rtcd.h"
-#include "aom_dsp/mips/aom_convolve_msa.h"
-
-static void common_hv_8ht_8vt_and_aver_dst_4w_msa(
- const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert, int32_t height) {
- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
- v16u8 dst0, dst1, dst2, dst3, mask0, mask1, mask2, mask3, tmp0, tmp1;
- v16i8 filt_hz0, filt_hz1, filt_hz2, filt_hz3;
- v8i16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, hz_out5, hz_out6;
- v8i16 hz_out7, hz_out8, hz_out9, res0, res1, vec0, vec1, vec2, vec3, vec4;
- v8i16 filt, filt_vt0, filt_vt1, filt_vt2, filt_vt3;
-
- mask0 = LD_UB(&mc_filt_mask_arr[16]);
- src -= (3 + 3 * src_stride);
-
- /* rearranging filter */
- filt = LD_SH(filter_horiz);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt_hz0, filt_hz1, filt_hz2, filt_hz3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
- XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
- src += (7 * src_stride);
-
- hz_out0 = HORIZ_8TAP_FILT(src0, src1, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out2 = HORIZ_8TAP_FILT(src2, src3, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out4 = HORIZ_8TAP_FILT(src4, src5, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out5 = HORIZ_8TAP_FILT(src5, src6, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- SLDI_B2_SH(hz_out2, hz_out4, hz_out0, hz_out2, hz_out1, hz_out3, 8);
-
- filt = LD_SH(filter_vert);
- SPLATI_H4_SH(filt, 0, 1, 2, 3, filt_vt0, filt_vt1, filt_vt2, filt_vt3);
-
- ILVEV_B2_SH(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
- vec2 = (v8i16)__msa_ilvev_b((v16i8)hz_out5, (v16i8)hz_out4);
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {
- LD_SB4(src, src_stride, src7, src8, src9, src10);
- XORI_B4_128_SB(src7, src8, src9, src10);
- src += (4 * src_stride);
-
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- hz_out7 = HORIZ_8TAP_FILT(src7, src8, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out6 = (v8i16)__msa_sldi_b((v16i8)hz_out7, (v16i8)hz_out5, 8);
- vec3 = (v8i16)__msa_ilvev_b((v16i8)hz_out7, (v16i8)hz_out6);
- res0 = FILT_8TAP_DPADD_S_H(vec0, vec1, vec2, vec3, filt_vt0, filt_vt1,
- filt_vt2, filt_vt3);
-
- hz_out9 = HORIZ_8TAP_FILT(src9, src10, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out8 = (v8i16)__msa_sldi_b((v16i8)hz_out9, (v16i8)hz_out7, 8);
- vec4 = (v8i16)__msa_ilvev_b((v16i8)hz_out9, (v16i8)hz_out8);
- res1 = FILT_8TAP_DPADD_S_H(vec1, vec2, vec3, vec4, filt_vt0, filt_vt1,
- filt_vt2, filt_vt3);
- ILVR_W2_UB(dst1, dst0, dst3, dst2, dst0, dst2);
-
- SRARI_H2_SH(res0, res1, FILTER_BITS);
- SAT_SH2_SH(res0, res1, 7);
- PCKEV_B2_UB(res0, res0, res1, res1, tmp0, tmp1);
- XORI_B2_128_UB(tmp0, tmp1);
- AVER_UB2_UB(tmp0, dst0, tmp1, dst2, tmp0, tmp1);
- ST4x4_UB(tmp0, tmp1, 0, 1, 0, 1, dst, dst_stride);
- dst += (4 * dst_stride);
-
- hz_out5 = hz_out9;
- vec0 = vec2;
- vec1 = vec3;
- vec2 = vec4;
- }
-}
-
-static void common_hv_8ht_8vt_and_aver_dst_8w_msa(
- const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert, int32_t height) {
- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
- v16i8 filt_hz0, filt_hz1, filt_hz2, filt_hz3;
- v8i16 filt, filt_vt0, filt_vt1, filt_vt2, filt_vt3;
- v16u8 dst0, dst1, dst2, dst3, mask0, mask1, mask2, mask3;
- v8i16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, hz_out5, hz_out6;
- v8i16 hz_out7, hz_out8, hz_out9, hz_out10, tmp0, tmp1, tmp2, tmp3;
- v8i16 out0, out1, out2, out3, out4, out5, out6, out7, out8, out9;
-
- mask0 = LD_UB(&mc_filt_mask_arr[0]);
- src -= (3 + 3 * src_stride);
-
- /* rearranging filter */
- filt = LD_SH(filter_horiz);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt_hz0, filt_hz1, filt_hz2, filt_hz3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
- src += (7 * src_stride);
-
- XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
- hz_out0 = HORIZ_8TAP_FILT(src0, src0, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out1 = HORIZ_8TAP_FILT(src1, src1, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out2 = HORIZ_8TAP_FILT(src2, src2, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out3 = HORIZ_8TAP_FILT(src3, src3, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out4 = HORIZ_8TAP_FILT(src4, src4, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out5 = HORIZ_8TAP_FILT(src5, src5, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out6 = HORIZ_8TAP_FILT(src6, src6, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
-
- filt = LD_SH(filter_vert);
- SPLATI_H4_SH(filt, 0, 1, 2, 3, filt_vt0, filt_vt1, filt_vt2, filt_vt3);
-
- ILVEV_B2_SH(hz_out0, hz_out1, hz_out2, hz_out3, out0, out1);
- ILVEV_B2_SH(hz_out4, hz_out5, hz_out1, hz_out2, out2, out4);
- ILVEV_B2_SH(hz_out3, hz_out4, hz_out5, hz_out6, out5, out6);
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {
- LD_SB4(src, src_stride, src7, src8, src9, src10);
- XORI_B4_128_SB(src7, src8, src9, src10);
- src += (4 * src_stride);
-
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
-
- hz_out7 = HORIZ_8TAP_FILT(src7, src7, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- out3 = (v8i16)__msa_ilvev_b((v16i8)hz_out7, (v16i8)hz_out6);
- tmp0 = FILT_8TAP_DPADD_S_H(out0, out1, out2, out3, filt_vt0, filt_vt1,
- filt_vt2, filt_vt3);
-
- hz_out8 = HORIZ_8TAP_FILT(src8, src8, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- out7 = (v8i16)__msa_ilvev_b((v16i8)hz_out8, (v16i8)hz_out7);
- tmp1 = FILT_8TAP_DPADD_S_H(out4, out5, out6, out7, filt_vt0, filt_vt1,
- filt_vt2, filt_vt3);
-
- hz_out9 = HORIZ_8TAP_FILT(src9, src9, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- out8 = (v8i16)__msa_ilvev_b((v16i8)hz_out9, (v16i8)hz_out8);
- tmp2 = FILT_8TAP_DPADD_S_H(out1, out2, out3, out8, filt_vt0, filt_vt1,
- filt_vt2, filt_vt3);
-
- hz_out10 = HORIZ_8TAP_FILT(src10, src10, mask0, mask1, mask2, mask3,
- filt_hz0, filt_hz1, filt_hz2, filt_hz3);
- out9 = (v8i16)__msa_ilvev_b((v16i8)hz_out10, (v16i8)hz_out9);
- tmp3 = FILT_8TAP_DPADD_S_H(out5, out6, out7, out9, filt_vt0, filt_vt1,
- filt_vt2, filt_vt3);
-
- SRARI_H4_SH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
- SAT_SH4_SH(tmp0, tmp1, tmp2, tmp3, 7);
- CONVERT_UB_AVG_ST8x4_UB(tmp0, tmp1, tmp2, tmp3, dst0, dst1, dst2, dst3, dst,
- dst_stride);
- dst += (4 * dst_stride);
-
- hz_out6 = hz_out10;
- out0 = out2;
- out1 = out3;
- out2 = out8;
- out4 = out6;
- out5 = out7;
- out6 = out9;
- }
-}
-
-static void common_hv_8ht_8vt_and_aver_dst_16w_msa(
- const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert, int32_t height) {
- int32_t multiple8_cnt;
- for (multiple8_cnt = 2; multiple8_cnt--;) {
- common_hv_8ht_8vt_and_aver_dst_8w_msa(src, src_stride, dst, dst_stride,
- filter_horiz, filter_vert, height);
- src += 8;
- dst += 8;
- }
-}
-
-static void common_hv_8ht_8vt_and_aver_dst_32w_msa(
- const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert, int32_t height) {
- int32_t multiple8_cnt;
- for (multiple8_cnt = 4; multiple8_cnt--;) {
- common_hv_8ht_8vt_and_aver_dst_8w_msa(src, src_stride, dst, dst_stride,
- filter_horiz, filter_vert, height);
- src += 8;
- dst += 8;
- }
-}
-
-static void common_hv_8ht_8vt_and_aver_dst_64w_msa(
- const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert, int32_t height) {
- int32_t multiple8_cnt;
- for (multiple8_cnt = 8; multiple8_cnt--;) {
- common_hv_8ht_8vt_and_aver_dst_8w_msa(src, src_stride, dst, dst_stride,
- filter_horiz, filter_vert, height);
- src += 8;
- dst += 8;
- }
-}
-
-static void common_hv_2ht_2vt_and_aver_dst_4x4_msa(
- const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert) {
- v16i8 src0, src1, src2, src3, src4, mask;
- v16u8 filt_hz, filt_vt, vec0, vec1;
- v16u8 dst0, dst1, dst2, dst3, res0, res1;
- v8u16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, tmp0, tmp1, filt;
-
- mask = LD_SB(&mc_filt_mask_arr[16]);
-
- /* rearranging filter */
- filt = LD_UH(filter_horiz);
- filt_hz = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- filt = LD_UH(filter_vert);
- filt_vt = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src0, src1, mask, filt_hz, FILTER_BITS);
- hz_out2 = HORIZ_2TAP_FILT_UH(src2, src3, mask, filt_hz, FILTER_BITS);
- hz_out4 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS);
- hz_out1 = (v8u16)__msa_sldi_b((v16i8)hz_out2, (v16i8)hz_out0, 8);
- hz_out3 = (v8u16)__msa_pckod_d((v2i64)hz_out4, (v2i64)hz_out2);
- ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
-
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- ILVR_W2_UB(dst1, dst0, dst3, dst2, dst0, dst2);
- DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- PCKEV_B2_UB(tmp0, tmp0, tmp1, tmp1, res0, res1);
- AVER_UB2_UB(res0, dst0, res1, dst2, res0, res1);
- ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
-}
-
-static void common_hv_2ht_2vt_and_aver_dst_4x8_msa(
- const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert) {
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, mask;
- v16u8 filt_hz, filt_vt, vec0, vec1, vec2, vec3, res0, res1, res2, res3;
- v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
- v8u16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, hz_out5, hz_out6;
- v8u16 hz_out7, hz_out8, tmp0, tmp1, tmp2, tmp3;
- v8i16 filt;
-
- mask = LD_SB(&mc_filt_mask_arr[16]);
-
- /* rearranging filter */
- filt = LD_SH(filter_horiz);
- filt_hz = (v16u8)__msa_splati_h(filt, 0);
-
- filt = LD_SH(filter_vert);
- filt_vt = (v16u8)__msa_splati_h(filt, 0);
-
- LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
- src += (8 * src_stride);
- src8 = LD_SB(src);
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src0, src1, mask, filt_hz, FILTER_BITS);
- hz_out2 = HORIZ_2TAP_FILT_UH(src2, src3, mask, filt_hz, FILTER_BITS);
- hz_out4 = HORIZ_2TAP_FILT_UH(src4, src5, mask, filt_hz, FILTER_BITS);
- hz_out6 = HORIZ_2TAP_FILT_UH(src6, src7, mask, filt_hz, FILTER_BITS);
- hz_out8 = HORIZ_2TAP_FILT_UH(src8, src8, mask, filt_hz, FILTER_BITS);
- SLDI_B3_UH(hz_out2, hz_out4, hz_out6, hz_out0, hz_out2, hz_out4, hz_out1,
- hz_out3, hz_out5, 8);
- hz_out7 = (v8u16)__msa_pckod_d((v2i64)hz_out8, (v2i64)hz_out6);
-
- LD_UB8(dst, dst_stride, dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
- ILVR_W4_UB(dst1, dst0, dst3, dst2, dst5, dst4, dst7, dst6, dst0, dst2, dst4,
- dst6);
- ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
- ILVEV_B2_UB(hz_out4, hz_out5, hz_out6, hz_out7, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt_vt, filt_vt, filt_vt, filt_vt, tmp0,
- tmp1, tmp2, tmp3);
- SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
- PCKEV_B4_UB(tmp0, tmp0, tmp1, tmp1, tmp2, tmp2, tmp3, tmp3, res0, res1, res2,
- res3);
- AVER_UB4_UB(res0, dst0, res1, dst2, res2, dst4, res3, dst6, res0, res1, res2,
- res3);
- ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
- dst += (4 * dst_stride);
- ST4x4_UB(res2, res3, 0, 1, 0, 1, dst, dst_stride);
-}
-
-static void common_hv_2ht_2vt_and_aver_dst_4w_msa(
- const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert, int32_t height) {
- if (4 == height) {
- common_hv_2ht_2vt_and_aver_dst_4x4_msa(src, src_stride, dst, dst_stride,
- filter_horiz, filter_vert);
- } else if (8 == height) {
- common_hv_2ht_2vt_and_aver_dst_4x8_msa(src, src_stride, dst, dst_stride,
- filter_horiz, filter_vert);
- }
-}
-
-static void common_hv_2ht_2vt_and_aver_dst_8x4_msa(
- const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert) {
- v16i8 src0, src1, src2, src3, src4, mask;
- v16u8 filt_hz, filt_vt, dst0, dst1, dst2, dst3, vec0, vec1, vec2, vec3;
- v8u16 hz_out0, hz_out1, tmp0, tmp1, tmp2, tmp3;
- v8i16 filt;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_SH(filter_horiz);
- filt_hz = (v16u8)__msa_splati_h(filt, 0);
-
- filt = LD_SH(filter_vert);
- filt_vt = (v16u8)__msa_splati_h(filt, 0);
-
- LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
- src += (5 * src_stride);
-
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- hz_out0 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS);
- hz_out1 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS);
- vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
- tmp0 = __msa_dotp_u_h(vec0, filt_vt);
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz, FILTER_BITS);
- vec1 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
- tmp1 = __msa_dotp_u_h(vec1, filt_vt);
-
- hz_out1 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz, FILTER_BITS);
- vec2 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
- tmp2 = __msa_dotp_u_h(vec2, filt_vt);
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS);
- vec3 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
- tmp3 = __msa_dotp_u_h(vec3, filt_vt);
-
- SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
- PCKEV_AVG_ST8x4_UB(tmp0, dst0, tmp1, dst1, tmp2, dst2, tmp3, dst3, dst,
- dst_stride);
-}
-
-static void common_hv_2ht_2vt_and_aver_dst_8x8mult_msa(
- const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert, int32_t height) {
- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, mask;
- v16u8 filt_hz, filt_vt, vec0, dst0, dst1, dst2, dst3;
- v8u16 hz_out0, hz_out1, tmp0, tmp1, tmp2, tmp3;
- v8i16 filt;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_SH(filter_horiz);
- filt_hz = (v16u8)__msa_splati_h(filt, 0);
-
- filt = LD_SH(filter_vert);
- filt_vt = (v16u8)__msa_splati_h(filt, 0);
-
- src0 = LD_SB(src);
- src += src_stride;
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS);
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {
- LD_SB4(src, src_stride, src1, src2, src3, src4);
- src += (4 * src_stride);
-
- hz_out1 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS);
- vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
- tmp0 = __msa_dotp_u_h(vec0, filt_vt);
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz, FILTER_BITS);
- vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
- tmp1 = __msa_dotp_u_h(vec0, filt_vt);
-
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
-
- hz_out1 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz, FILTER_BITS);
- vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
- tmp2 = __msa_dotp_u_h(vec0, filt_vt);
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS);
- vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
- tmp3 = __msa_dotp_u_h(vec0, filt_vt);
-
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- PCKEV_AVG_ST8x4_UB(tmp0, dst0, tmp1, dst1, tmp2, dst2, tmp3, dst3, dst,
- dst_stride);
- dst += (4 * dst_stride);
- }
-}
-
-static void common_hv_2ht_2vt_and_aver_dst_8w_msa(
- const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert, int32_t height) {
- if (4 == height) {
- common_hv_2ht_2vt_and_aver_dst_8x4_msa(src, src_stride, dst, dst_stride,
- filter_horiz, filter_vert);
- } else {
- common_hv_2ht_2vt_and_aver_dst_8x8mult_msa(
- src, src_stride, dst, dst_stride, filter_horiz, filter_vert, height);
- }
-}
-
-static void common_hv_2ht_2vt_and_aver_dst_16w_msa(
- const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert, int32_t height) {
- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
- v16u8 filt_hz, filt_vt, vec0, vec1, dst0, dst1, dst2, dst3;
- v8u16 hz_out0, hz_out1, hz_out2, hz_out3, tmp0, tmp1;
- v8i16 filt;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_SH(filter_horiz);
- filt_hz = (v16u8)__msa_splati_h(filt, 0);
-
- filt = LD_SH(filter_vert);
- filt_vt = (v16u8)__msa_splati_h(filt, 0);
-
- LD_SB2(src, 8, src0, src1);
- src += src_stride;
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS);
- hz_out2 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS);
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {
- LD_SB4(src, src_stride, src0, src2, src4, src6);
- LD_SB4(src + 8, src_stride, src1, src3, src5, src7);
- src += (4 * src_stride);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
-
- hz_out1 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS);
- hz_out3 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS);
- ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
- DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst0, dst);
- dst += dst_stride;
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz, FILTER_BITS);
- hz_out2 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz, FILTER_BITS);
- ILVEV_B2_UB(hz_out1, hz_out0, hz_out3, hz_out2, vec0, vec1);
- DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst1, dst);
- dst += dst_stride;
-
- hz_out1 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS);
- hz_out3 = HORIZ_2TAP_FILT_UH(src5, src5, mask, filt_hz, FILTER_BITS);
- ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
- DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst2, dst);
- dst += dst_stride;
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src6, src6, mask, filt_hz, FILTER_BITS);
- hz_out2 = HORIZ_2TAP_FILT_UH(src7, src7, mask, filt_hz, FILTER_BITS);
- ILVEV_B2_UB(hz_out1, hz_out0, hz_out3, hz_out2, vec0, vec1);
- DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst3, dst);
- dst += dst_stride;
- }
-}
-
-static void common_hv_2ht_2vt_and_aver_dst_32w_msa(
- const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert, int32_t height) {
- int32_t multiple8_cnt;
- for (multiple8_cnt = 2; multiple8_cnt--;) {
- common_hv_2ht_2vt_and_aver_dst_16w_msa(src, src_stride, dst, dst_stride,
- filter_horiz, filter_vert, height);
- src += 16;
- dst += 16;
- }
-}
-
-static void common_hv_2ht_2vt_and_aver_dst_64w_msa(
- const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert, int32_t height) {
- int32_t multiple8_cnt;
- for (multiple8_cnt = 4; multiple8_cnt--;) {
- common_hv_2ht_2vt_and_aver_dst_16w_msa(src, src_stride, dst, dst_stride,
- filter_horiz, filter_vert, height);
- src += 16;
- dst += 16;
- }
-}
-
-void aom_convolve8_avg_msa(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4, int w,
- int h) {
- int8_t cnt, filt_hor[8], filt_ver[8];
-
- assert(x_step_q4 == 16);
- assert(y_step_q4 == 16);
- assert(((const int32_t *)filter_x)[1] != 0x800000);
- assert(((const int32_t *)filter_y)[1] != 0x800000);
-
- for (cnt = 0; cnt < 8; ++cnt) {
- filt_hor[cnt] = filter_x[cnt];
- filt_ver[cnt] = filter_y[cnt];
- }
-
- if (((const int32_t *)filter_x)[0] == 0 &&
- ((const int32_t *)filter_y)[0] == 0) {
- switch (w) {
- case 4:
- common_hv_2ht_2vt_and_aver_dst_4w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, &filt_hor[3],
- &filt_ver[3], h);
- break;
- case 8:
- common_hv_2ht_2vt_and_aver_dst_8w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, &filt_hor[3],
- &filt_ver[3], h);
- break;
- case 16:
- common_hv_2ht_2vt_and_aver_dst_16w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride,
- &filt_hor[3], &filt_ver[3], h);
- break;
- case 32:
- common_hv_2ht_2vt_and_aver_dst_32w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride,
- &filt_hor[3], &filt_ver[3], h);
- break;
- case 64:
- common_hv_2ht_2vt_and_aver_dst_64w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride,
- &filt_hor[3], &filt_ver[3], h);
- break;
- default:
- aom_convolve8_avg_c(src, src_stride, dst, dst_stride, filter_x,
- x_step_q4, filter_y, y_step_q4, w, h);
- break;
- }
- } else if (((const int32_t *)filter_x)[0] == 0 ||
- ((const int32_t *)filter_y)[0] == 0) {
- aom_convolve8_avg_c(src, src_stride, dst, dst_stride, filter_x, x_step_q4,
- filter_y, y_step_q4, w, h);
- } else {
- switch (w) {
- case 4:
- common_hv_8ht_8vt_and_aver_dst_4w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, filt_hor,
- filt_ver, h);
- break;
- case 8:
- common_hv_8ht_8vt_and_aver_dst_8w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, filt_hor,
- filt_ver, h);
- break;
- case 16:
- common_hv_8ht_8vt_and_aver_dst_16w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, filt_hor,
- filt_ver, h);
- break;
- case 32:
- common_hv_8ht_8vt_and_aver_dst_32w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, filt_hor,
- filt_ver, h);
- break;
- case 64:
- common_hv_8ht_8vt_and_aver_dst_64w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, filt_hor,
- filt_ver, h);
- break;
- default:
- aom_convolve8_avg_c(src, src_stride, dst, dst_stride, filter_x,
- x_step_q4, filter_y, y_step_q4, w, h);
- break;
- }
- }
-}
diff --git a/aom_dsp/mips/aom_convolve8_avg_vert_msa.c b/aom_dsp/mips/aom_convolve8_avg_vert_msa.c
deleted file mode 100644
index dae7711..0000000
--- a/aom_dsp/mips/aom_convolve8_avg_vert_msa.c
+++ /dev/null
@@ -1,677 +0,0 @@
-/*
- * Copyright (c) 2016, Alliance for Open Media. All rights reserved
- *
- * This source code is subject to the terms of the BSD 2 Clause License and
- * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
- * was not distributed with this source code in the LICENSE file, you can
- * obtain it at www.aomedia.org/license/software. If the Alliance for Open
- * Media Patent License 1.0 was not distributed with this source code in the
- * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
- */
-
-#include <assert.h>
-#include "./aom_dsp_rtcd.h"
-#include "aom_dsp/mips/aom_convolve_msa.h"
-
-static void common_vt_8t_and_aver_dst_4w_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride, int8_t *filter,
- int32_t height) {
- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
- v16u8 dst0, dst1, dst2, dst3, out;
- v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r;
- v16i8 src65_r, src87_r, src109_r, src2110, src4332, src6554, src8776;
- v16i8 src10998, filt0, filt1, filt2, filt3;
- v8i16 filt, out10, out32;
-
- src -= (3 * src_stride);
-
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
- src += (7 * src_stride);
-
- ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r, src32_r,
- src54_r, src21_r);
- ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
- ILVR_D3_SB(src21_r, src10_r, src43_r, src32_r, src65_r, src54_r, src2110,
- src4332, src6554);
- XORI_B3_128_SB(src2110, src4332, src6554);
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {
- LD_SB4(src, src_stride, src7, src8, src9, src10);
- src += (4 * src_stride);
-
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r,
- src87_r, src98_r, src109_r);
- ILVR_D2_SB(src87_r, src76_r, src109_r, src98_r, src8776, src10998);
- XORI_B2_128_SB(src8776, src10998);
- out10 = FILT_8TAP_DPADD_S_H(src2110, src4332, src6554, src8776, filt0,
- filt1, filt2, filt3);
- out32 = FILT_8TAP_DPADD_S_H(src4332, src6554, src8776, src10998, filt0,
- filt1, filt2, filt3);
- SRARI_H2_SH(out10, out32, FILTER_BITS);
- SAT_SH2_SH(out10, out32, 7);
- out = PCKEV_XORI128_UB(out10, out32);
- ILVR_W2_UB(dst1, dst0, dst3, dst2, dst0, dst2);
-
- dst0 = (v16u8)__msa_ilvr_d((v2i64)dst2, (v2i64)dst0);
- out = __msa_aver_u_b(out, dst0);
-
- ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride);
- dst += (4 * dst_stride);
-
- src2110 = src6554;
- src4332 = src8776;
- src6554 = src10998;
- src6 = src10;
- }
-}
-
-static void common_vt_8t_and_aver_dst_8w_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride, int8_t *filter,
- int32_t height) {
- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
- v16u8 dst0, dst1, dst2, dst3;
- v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r;
- v16i8 src65_r, src87_r, src109_r, filt0, filt1, filt2, filt3;
- v8i16 filt, out0, out1, out2, out3;
-
- src -= (3 * src_stride);
-
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
- src += (7 * src_stride);
-
- XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
- ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r, src32_r,
- src54_r, src21_r);
- ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {
- LD_SB4(src, src_stride, src7, src8, src9, src10);
- src += (4 * src_stride);
-
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- XORI_B4_128_SB(src7, src8, src9, src10);
- ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r,
- src87_r, src98_r, src109_r);
- out0 = FILT_8TAP_DPADD_S_H(src10_r, src32_r, src54_r, src76_r, filt0, filt1,
- filt2, filt3);
- out1 = FILT_8TAP_DPADD_S_H(src21_r, src43_r, src65_r, src87_r, filt0, filt1,
- filt2, filt3);
- out2 = FILT_8TAP_DPADD_S_H(src32_r, src54_r, src76_r, src98_r, filt0, filt1,
- filt2, filt3);
- out3 = FILT_8TAP_DPADD_S_H(src43_r, src65_r, src87_r, src109_r, filt0,
- filt1, filt2, filt3);
- SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
- SAT_SH4_SH(out0, out1, out2, out3, 7);
- CONVERT_UB_AVG_ST8x4_UB(out0, out1, out2, out3, dst0, dst1, dst2, dst3, dst,
- dst_stride);
- dst += (4 * dst_stride);
-
- src10_r = src54_r;
- src32_r = src76_r;
- src54_r = src98_r;
- src21_r = src65_r;
- src43_r = src87_r;
- src65_r = src109_r;
- src6 = src10;
- }
-}
-
-static void common_vt_8t_and_aver_dst_16w_mult_msa(
- const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height, int32_t width) {
- const uint8_t *src_tmp;
- uint8_t *dst_tmp;
- uint32_t loop_cnt, cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
- v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r;
- v16i8 src65_r, src87_r, src109_r, src10_l, src32_l, src54_l, src76_l;
- v16i8 src98_l, src21_l, src43_l, src65_l, src87_l, src109_l;
- v16i8 filt0, filt1, filt2, filt3;
- v16u8 dst0, dst1, dst2, dst3, tmp0, tmp1, tmp2, tmp3;
- v8i16 out0_r, out1_r, out2_r, out3_r, out0_l, out1_l, out2_l, out3_l, filt;
-
- src -= (3 * src_stride);
-
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- for (cnt = (width >> 4); cnt--;) {
- src_tmp = src;
- dst_tmp = dst;
-
- LD_SB7(src_tmp, src_stride, src0, src1, src2, src3, src4, src5, src6);
- XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
- src_tmp += (7 * src_stride);
-
- ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r, src32_r,
- src54_r, src21_r);
- ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
- ILVL_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_l, src32_l,
- src54_l, src21_l);
- ILVL_B2_SB(src4, src3, src6, src5, src43_l, src65_l);
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {
- LD_SB4(src_tmp, src_stride, src7, src8, src9, src10);
- src_tmp += (4 * src_stride);
-
- LD_UB4(dst_tmp, dst_stride, dst0, dst1, dst2, dst3);
- XORI_B4_128_SB(src7, src8, src9, src10);
- ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r,
- src87_r, src98_r, src109_r);
- ILVL_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_l,
- src87_l, src98_l, src109_l);
- out0_r = FILT_8TAP_DPADD_S_H(src10_r, src32_r, src54_r, src76_r, filt0,
- filt1, filt2, filt3);
- out1_r = FILT_8TAP_DPADD_S_H(src21_r, src43_r, src65_r, src87_r, filt0,
- filt1, filt2, filt3);
- out2_r = FILT_8TAP_DPADD_S_H(src32_r, src54_r, src76_r, src98_r, filt0,
- filt1, filt2, filt3);
- out3_r = FILT_8TAP_DPADD_S_H(src43_r, src65_r, src87_r, src109_r, filt0,
- filt1, filt2, filt3);
- out0_l = FILT_8TAP_DPADD_S_H(src10_l, src32_l, src54_l, src76_l, filt0,
- filt1, filt2, filt3);
- out1_l = FILT_8TAP_DPADD_S_H(src21_l, src43_l, src65_l, src87_l, filt0,
- filt1, filt2, filt3);
- out2_l = FILT_8TAP_DPADD_S_H(src32_l, src54_l, src76_l, src98_l, filt0,
- filt1, filt2, filt3);
- out3_l = FILT_8TAP_DPADD_S_H(src43_l, src65_l, src87_l, src109_l, filt0,
- filt1, filt2, filt3);
- SRARI_H4_SH(out0_r, out1_r, out2_r, out3_r, FILTER_BITS);
- SRARI_H4_SH(out0_l, out1_l, out2_l, out3_l, FILTER_BITS);
- SAT_SH4_SH(out0_r, out1_r, out2_r, out3_r, 7);
- SAT_SH4_SH(out0_l, out1_l, out2_l, out3_l, 7);
- PCKEV_B4_UB(out0_l, out0_r, out1_l, out1_r, out2_l, out2_r, out3_l,
- out3_r, tmp0, tmp1, tmp2, tmp3);
- XORI_B4_128_UB(tmp0, tmp1, tmp2, tmp3);
- AVER_UB4_UB(tmp0, dst0, tmp1, dst1, tmp2, dst2, tmp3, dst3, dst0, dst1,
- dst2, dst3);
- ST_UB4(dst0, dst1, dst2, dst3, dst_tmp, dst_stride);
- dst_tmp += (4 * dst_stride);
-
- src10_r = src54_r;
- src32_r = src76_r;
- src54_r = src98_r;
- src21_r = src65_r;
- src43_r = src87_r;
- src65_r = src109_r;
- src10_l = src54_l;
- src32_l = src76_l;
- src54_l = src98_l;
- src21_l = src65_l;
- src43_l = src87_l;
- src65_l = src109_l;
- src6 = src10;
- }
-
- src += 16;
- dst += 16;
- }
-}
-
-static void common_vt_8t_and_aver_dst_16w_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter, int32_t height) {
- common_vt_8t_and_aver_dst_16w_mult_msa(src, src_stride, dst, dst_stride,
- filter, height, 16);
-}
-
-static void common_vt_8t_and_aver_dst_32w_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter, int32_t height) {
- common_vt_8t_and_aver_dst_16w_mult_msa(src, src_stride, dst, dst_stride,
- filter, height, 32);
-}
-
-static void common_vt_8t_and_aver_dst_64w_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter, int32_t height) {
- common_vt_8t_and_aver_dst_16w_mult_msa(src, src_stride, dst, dst_stride,
- filter, height, 64);
-}
-
-static void common_vt_2t_and_aver_dst_4x4_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter) {
- v16i8 src0, src1, src2, src3, src4;
- v16u8 dst0, dst1, dst2, dst3, out, filt0, src2110, src4332;
- v16i8 src10_r, src32_r, src21_r, src43_r;
- v8i16 filt;
- v8u16 tmp0, tmp1;
-
- filt = LD_SH(filter);
- filt0 = (v16u8)__msa_splati_h(filt, 0);
-
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- src += (4 * src_stride);
-
- src4 = LD_SB(src);
- src += src_stride;
-
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- ILVR_W2_UB(dst1, dst0, dst3, dst2, dst0, dst1);
- dst0 = (v16u8)__msa_ilvr_d((v2i64)dst1, (v2i64)dst0);
- ILVR_B4_SB(src1, src0, src2, src1, src3, src2, src4, src3, src10_r, src21_r,
- src32_r, src43_r);
- ILVR_D2_UB(src21_r, src10_r, src43_r, src32_r, src2110, src4332);
- DOTP_UB2_UH(src2110, src4332, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
-
- out = (v16u8)__msa_pckev_b((v16i8)tmp1, (v16i8)tmp0);
- out = __msa_aver_u_b(out, dst0);
-
- ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride);
-}
-
-static void common_vt_2t_and_aver_dst_4x8_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter) {
- v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src87_r;
- v16i8 src10_r, src32_r, src54_r, src76_r, src21_r, src43_r, src65_r;
- v16u8 src2110, src4332, src6554, src8776, filt0;
- v8u16 tmp0, tmp1, tmp2, tmp3;
- v8i16 filt;
-
- filt = LD_SH(filter);
- filt0 = (v16u8)__msa_splati_h(filt, 0);
-
- LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
- src += (8 * src_stride);
- src8 = LD_SB(src);
-
- LD_UB8(dst, dst_stride, dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
- ILVR_W4_UB(dst1, dst0, dst3, dst2, dst5, dst4, dst7, dst6, dst0, dst1, dst2,
- dst3);
- ILVR_D2_UB(dst1, dst0, dst3, dst2, dst0, dst1);
- ILVR_B4_SB(src1, src0, src2, src1, src3, src2, src4, src3, src10_r, src21_r,
- src32_r, src43_r);
- ILVR_B4_SB(src5, src4, src6, src5, src7, src6, src8, src7, src54_r, src65_r,
- src76_r, src87_r);
- ILVR_D4_UB(src21_r, src10_r, src43_r, src32_r, src65_r, src54_r, src87_r,
- src76_r, src2110, src4332, src6554, src8776);
- DOTP_UB4_UH(src2110, src4332, src6554, src8776, filt0, filt0, filt0, filt0,
- tmp0, tmp1, tmp2, tmp3);
- SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
- PCKEV_B2_UB(tmp1, tmp0, tmp3, tmp2, src2110, src4332);
- AVER_UB2_UB(src2110, dst0, src4332, dst1, src2110, src4332);
- ST4x4_UB(src2110, src2110, 0, 1, 2, 3, dst, dst_stride);
- dst += (4 * dst_stride);
- ST4x4_UB(src4332, src4332, 0, 1, 2, 3, dst, dst_stride);
-}
-
-static void common_vt_2t_and_aver_dst_4w_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride, int8_t *filter,
- int32_t height) {
- if (4 == height) {
- common_vt_2t_and_aver_dst_4x4_msa(src, src_stride, dst, dst_stride, filter);
- } else if (8 == height) {
- common_vt_2t_and_aver_dst_4x8_msa(src, src_stride, dst, dst_stride, filter);
- }
-}
-
-static void common_vt_2t_and_aver_dst_8x4_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter) {
- v16u8 src0, src1, src2, src3, src4;
- v16u8 dst0, dst1, dst2, dst3, vec0, vec1, vec2, vec3, filt0;
- v8u16 tmp0, tmp1, tmp2, tmp3;
- v8i16 filt;
-
- /* rearranging filter_y */
- filt = LD_SH(filter);
- filt0 = (v16u8)__msa_splati_h(filt, 0);
-
- LD_UB5(src, src_stride, src0, src1, src2, src3, src4);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- ILVR_B2_UB(src1, src0, src2, src1, vec0, vec1);
- ILVR_B2_UB(src3, src2, src4, src3, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, tmp0, tmp1,
- tmp2, tmp3);
- SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
- PCKEV_AVG_ST8x4_UB(tmp0, dst0, tmp1, dst1, tmp2, dst2, tmp3, dst3, dst,
- dst_stride);
-}
-
-static void common_vt_2t_and_aver_dst_8x8mult_msa(
- const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {
- uint32_t loop_cnt;
- v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
- v16u8 dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8;
- v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0;
- v8u16 tmp0, tmp1, tmp2, tmp3;
- v8i16 filt;
-
- /* rearranging filter_y */
- filt = LD_SH(filter);
- filt0 = (v16u8)__msa_splati_h(filt, 0);
-
- src0 = LD_UB(src);
- src += src_stride;
-
- for (loop_cnt = (height >> 3); loop_cnt--;) {
- LD_UB8(src, src_stride, src1, src2, src3, src4, src5, src6, src7, src8);
- src += (8 * src_stride);
- LD_UB8(dst, dst_stride, dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8);
-
- ILVR_B4_UB(src1, src0, src2, src1, src3, src2, src4, src3, vec0, vec1, vec2,
- vec3);
- ILVR_B4_UB(src5, src4, src6, src5, src7, src6, src8, src7, vec4, vec5, vec6,
- vec7);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, tmp0, tmp1,
- tmp2, tmp3);
- SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
- PCKEV_AVG_ST8x4_UB(tmp0, dst1, tmp1, dst2, tmp2, dst3, tmp3, dst4, dst,
- dst_stride);
- dst += (4 * dst_stride);
-
- DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, tmp0, tmp1,
- tmp2, tmp3);
- SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
- PCKEV_AVG_ST8x4_UB(tmp0, dst5, tmp1, dst6, tmp2, dst7, tmp3, dst8, dst,
- dst_stride);
- dst += (4 * dst_stride);
-
- src0 = src8;
- }
-}
-
-static void common_vt_2t_and_aver_dst_8w_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride, int8_t *filter,
- int32_t height) {
- if (4 == height) {
- common_vt_2t_and_aver_dst_8x4_msa(src, src_stride, dst, dst_stride, filter);
- } else {
- common_vt_2t_and_aver_dst_8x8mult_msa(src, src_stride, dst, dst_stride,
- filter, height);
- }
-}
-
-static void common_vt_2t_and_aver_dst_16w_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter, int32_t height) {
- uint32_t loop_cnt;
- v16u8 src0, src1, src2, src3, src4, dst0, dst1, dst2, dst3, filt0;
- v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
- v8u16 tmp0, tmp1, tmp2, tmp3, filt;
-
- /* rearranging filter_y */
- filt = LD_UH(filter);
- filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- src0 = LD_UB(src);
- src += src_stride;
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {
- LD_UB4(src, src_stride, src1, src2, src3, src4);
- src += (4 * src_stride);
-
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2);
- ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3);
- DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst0, dst);
- dst += dst_stride;
-
- ILVR_B2_UB(src3, src2, src4, src3, vec4, vec6);
- ILVL_B2_UB(src3, src2, src4, src3, vec5, vec7);
- DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp3, tmp2, dst1, dst);
- dst += dst_stride;
-
- DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst2, dst);
- dst += dst_stride;
-
- DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp3, tmp2, dst3, dst);
- dst += dst_stride;
-
- src0 = src4;
- }
-}
-
-static void common_vt_2t_and_aver_dst_32w_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter, int32_t height) {
- uint32_t loop_cnt;
- v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9;
- v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
- v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0;
- v8u16 tmp0, tmp1, tmp2, tmp3, filt;
-
- /* rearranging filter_y */
- filt = LD_UH(filter);
- filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- LD_UB2(src, 16, src0, src5);
- src += src_stride;
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {
- LD_UB4(src, src_stride, src1, src2, src3, src4);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2);
- ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3);
-
- LD_UB4(src + 16, src_stride, src6, src7, src8, src9);
- LD_UB4(dst + 16, dst_stride, dst4, dst5, dst6, dst7);
- src += (4 * src_stride);
-
- DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst0, dst);
-
- DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp3, tmp2, dst1, dst + dst_stride);
-
- ILVR_B2_UB(src3, src2, src4, src3, vec4, vec6);
- ILVL_B2_UB(src3, src2, src4, src3, vec5, vec7);
- DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst2, dst + 2 * dst_stride);
-
- DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp3, tmp2, dst3, dst + 3 * dst_stride);
-
- ILVR_B2_UB(src6, src5, src7, src6, vec0, vec2);
- ILVL_B2_UB(src6, src5, src7, src6, vec1, vec3);
- DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst4, dst + 16);
-
- DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp3, tmp2, dst5, dst + 16 + dst_stride);
-
- ILVR_B2_UB(src8, src7, src9, src8, vec4, vec6);
- ILVL_B2_UB(src8, src7, src9, src8, vec5, vec7);
- DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst6, dst + 16 + 2 * dst_stride);
-
- DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp3, tmp2, dst7, dst + 16 + 3 * dst_stride);
- dst += (4 * dst_stride);
-
- src0 = src4;
- src5 = src9;
- }
-}
-
-static void common_vt_2t_and_aver_dst_64w_msa(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter, int32_t height) {
- uint32_t loop_cnt;
- v16u8 src0, src1, src2, src3, src4, src5;
- v16u8 src6, src7, src8, src9, src10, src11, filt0;
- v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
- v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
- v8u16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
- v8u16 filt;
-
- /* rearranging filter_y */
- filt = LD_UH(filter);
- filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- LD_UB4(src, 16, src0, src3, src6, src9);
- src += src_stride;
-
- for (loop_cnt = (height >> 1); loop_cnt--;) {
- LD_UB2(src, src_stride, src1, src2);
- LD_UB2(dst, dst_stride, dst0, dst1);
- LD_UB2(src + 16, src_stride, src4, src5);
- LD_UB2(dst + 16, dst_stride, dst2, dst3);
- LD_UB2(src + 32, src_stride, src7, src8);
- LD_UB2(dst + 32, dst_stride, dst4, dst5);
- LD_UB2(src + 48, src_stride, src10, src11);
- LD_UB2(dst + 48, dst_stride, dst6, dst7);
- src += (2 * src_stride);
-
- ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2);
- ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3);
- DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst0, dst);
-
- DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp3, tmp2, dst1, dst + dst_stride);
-
- ILVR_B2_UB(src4, src3, src5, src4, vec4, vec6);
- ILVL_B2_UB(src4, src3, src5, src4, vec5, vec7);
- DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp4, tmp5);
- SRARI_H2_UH(tmp4, tmp5, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp5, tmp4, dst2, dst + 16);
-
- DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp6, tmp7);
- SRARI_H2_UH(tmp6, tmp7, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp7, tmp6, dst3, dst + 16 + dst_stride);
-
- ILVR_B2_UB(src7, src6, src8, src7, vec0, vec2);
- ILVL_B2_UB(src7, src6, src8, src7, vec1, vec3);
- DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst4, dst + 32);
-
- DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp3, tmp2, dst5, dst + 32 + dst_stride);
-
- ILVR_B2_UB(src10, src9, src11, src10, vec4, vec6);
- ILVL_B2_UB(src10, src9, src11, src10, vec5, vec7);
- DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp4, tmp5);
- SRARI_H2_UH(tmp4, tmp5, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp5, tmp4, dst6, (dst + 48));
-
- DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp6, tmp7);
- SRARI_H2_UH(tmp6, tmp7, FILTER_BITS);
- PCKEV_AVG_ST_UB(tmp7, tmp6, dst7, dst + 48 + dst_stride);
- dst += (2 * dst_stride);
-
- src0 = src2;
- src3 = src5;
- src6 = src8;
- src9 = src11;
- }
-}
-
-void aom_convolve8_avg_vert_msa(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4, int w,
- int h) {
- int8_t cnt, filt_ver[8];
-
- assert(y_step_q4 == 16);
- assert(((const int32_t *)filter_y)[1] != 0x800000);
-
- for (cnt = 0; cnt < 8; ++cnt) {
- filt_ver[cnt] = filter_y[cnt];
- }
-
- if (((const int32_t *)filter_y)[0] == 0) {
- switch (w) {
- case 4:
- common_vt_2t_and_aver_dst_4w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, &filt_ver[3], h);
- break;
- case 8:
- common_vt_2t_and_aver_dst_8w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, &filt_ver[3], h);
- break;
- case 16:
- common_vt_2t_and_aver_dst_16w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, &filt_ver[3], h);
- break;
- case 32:
- common_vt_2t_and_aver_dst_32w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, &filt_ver[3], h);
- break;
- case 64:
- common_vt_2t_and_aver_dst_64w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, &filt_ver[3], h);
- break;
- default:
- aom_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, filter_x,
- x_step_q4, filter_y, y_step_q4, w, h);
- break;
- }
- } else {
- switch (w) {
- case 4:
- common_vt_8t_and_aver_dst_4w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, filt_ver, h);
- break;
- case 8:
- common_vt_8t_and_aver_dst_8w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, filt_ver, h);
- break;
- case 16:
- common_vt_8t_and_aver_dst_16w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, filt_ver, h);
-
- break;
- case 32:
- common_vt_8t_and_aver_dst_32w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, filt_ver, h);
- break;
- case 64:
- common_vt_8t_and_aver_dst_64w_msa(src, (int32_t)src_stride, dst,
- (int32_t)dst_stride, filt_ver, h);
- break;
- default:
- aom_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, filter_x,
- x_step_q4, filter_y, y_step_q4, w, h);
- break;
- }
- }
-}
diff --git a/aom_dsp/mips/convolve2_avg_dspr2.c b/aom_dsp/mips/convolve2_avg_dspr2.c
deleted file mode 100644
index d557115..0000000
--- a/aom_dsp/mips/convolve2_avg_dspr2.c
+++ /dev/null
@@ -1,256 +0,0 @@
-/*
- * Copyright (c) 2016, Alliance for Open Media. All rights reserved
- *
- * This source code is subject to the terms of the BSD 2 Clause License and
- * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
- * was not distributed with this source code in the LICENSE file, you can
- * obtain it at www.aomedia.org/license/software. If the Alliance for Open
- * Media Patent License 1.0 was not distributed with this source code in the
- * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
- */
-
-#include <assert.h>
-#include <stdio.h>
-
-#include "./aom_dsp_rtcd.h"
-#include "aom_dsp/mips/convolve_common_dspr2.h"
-#include "aom_dsp/aom_convolve.h"
-#include "aom_dsp/aom_dsp_common.h"
-#include "aom_ports/mem.h"
-
-#if HAVE_DSPR2
-static void convolve_bi_avg_vert_4_dspr2(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- const int16_t *filter_y, int32_t w,
- int32_t h) {
- int32_t x, y;
- const uint8_t *src_ptr;
- uint8_t *dst_ptr;
- uint8_t *cm = aom_ff_cropTbl;
- uint32_t vector4a = 64;
- uint32_t load1, load2;
- uint32_t p1, p2;
- uint32_t scratch1, scratch2;
- uint32_t store1, store2;
- int32_t Temp1, Temp2;
- const int16_t *filter = &filter_y[3];
- uint32_t filter45;
-
- filter45 = ((const int32_t *)filter)[0];
-
- for (y = h; y--;) {
- /* prefetch data to cache memory */
- prefetch_store(dst + dst_stride);
-
- for (x = 0; x < w; x += 4) {
- src_ptr = src + x;
- dst_ptr = dst + x;
-
- __asm__ __volatile__(
- "ulw %[load1], 0(%[src_ptr]) \n\t"
- "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
- "ulw %[load2], 0(%[src_ptr]) \n\t"
-
- "mtlo %[vector4a], $ac0 \n\t"
- "mtlo %[vector4a], $ac1 \n\t"
- "mtlo %[vector4a], $ac2 \n\t"
- "mtlo %[vector4a], $ac3 \n\t"
- "mthi $zero, $ac0 \n\t"
- "mthi $zero, $ac1 \n\t"
- "mthi $zero, $ac2 \n\t"
- "mthi $zero, $ac3 \n\t"
-
- "preceu.ph.qbr %[scratch1], %[load1] \n\t"
- "preceu.ph.qbr %[p1], %[load2] \n\t"
- "precrq.ph.w %[p2], %[p1], %[scratch1] \n\t" /* pixel 2 */
- "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
-
- "dpa.w.ph $ac0, %[p1], %[filter45] \n\t"
- "dpa.w.ph $ac1, %[p2], %[filter45] \n\t"
-
- "preceu.ph.qbl %[scratch1], %[load1] \n\t"
- "preceu.ph.qbl %[p1], %[load2] \n\t"
- "precrq.ph.w %[p2], %[p1], %[scratch1] \n\t" /* pixel 2 */
- "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
-
- "dpa.w.ph $ac2, %[p1], %[filter45] \n\t"
- "dpa.w.ph $ac3, %[p2], %[filter45] \n\t"
-
- "extp %[Temp1], $ac0, 31 \n\t"
- "extp %[Temp2], $ac1, 31 \n\t"
-
- "lbu %[scratch1], 0(%[dst_ptr]) \n\t"
- "lbu %[scratch2], 1(%[dst_ptr]) \n\t"
-
- "lbux %[store1], %[Temp1](%[cm]) \n\t"
- "addqh_r.w %[store1], %[store1], %[scratch1] \n\t" /* pixel 1 */
- "extp %[Temp1], $ac2, 31 \n\t"
-
- "lbux %[store2], %[Temp2](%[cm]) \n\t"
- "addqh_r.w %[store2], %[store2], %[scratch2] \n\t" /* pixel 2 */
- "extp %[Temp2], $ac3, 31 \n\t"
- "lbu %[scratch1], 2(%[dst_ptr]) \n\t"
-
- "sb %[store1], 0(%[dst_ptr]) \n\t"
- "sb %[store2], 1(%[dst_ptr]) \n\t"
- "lbu %[scratch2], 3(%[dst_ptr]) \n\t"
-
- "lbux %[store1], %[Temp1](%[cm]) \n\t"
- "lbux %[store2], %[Temp2](%[cm]) \n\t"
- "addqh_r.w %[store1], %[store1], %[scratch1] \n\t" /* pixel 3 */
- "addqh_r.w %[store2], %[store2], %[scratch2] \n\t" /* pixel 4 */
-
- "sb %[store1], 2(%[dst_ptr]) \n\t"
- "sb %[store2], 3(%[dst_ptr]) \n\t"
-
- : [load1] "=&r"(load1), [load2] "=&r"(load2), [p1] "=&r"(p1),
- [p2] "=&r"(p2), [scratch1] "=&r"(scratch1),
- [scratch2] "=&r"(scratch2), [Temp1] "=&r"(Temp1),
- [Temp2] "=&r"(Temp2), [store1] "=&r"(store1),
- [store2] "=&r"(store2), [src_ptr] "+r"(src_ptr)
- : [filter45] "r"(filter45), [vector4a] "r"(vector4a),
- [src_stride] "r"(src_stride), [cm] "r"(cm), [dst_ptr] "r"(dst_ptr));
- }
-
- /* Next row... */
- src += src_stride;
- dst += dst_stride;
- }
-}
-
-static void convolve_bi_avg_vert_64_dspr2(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- const int16_t *filter_y, int32_t h) {
- int32_t x, y;
- const uint8_t *src_ptr;
- uint8_t *dst_ptr;
- uint8_t *cm = aom_ff_cropTbl;
- uint32_t vector4a = 64;
- uint32_t load1, load2;
- uint32_t p1, p2;
- uint32_t scratch1, scratch2;
- uint32_t store1, store2;
- int32_t Temp1, Temp2;
- const int16_t *filter = &filter_y[3];
- uint32_t filter45;
-
- filter45 = ((const int32_t *)filter)[0];
-
- for (y = h; y--;) {
- /* prefetch data to cache memory */
- prefetch_store(dst + dst_stride);
- prefetch_store(dst + dst_stride + 32);
-
- for (x = 0; x < 64; x += 4) {
- src_ptr = src + x;
- dst_ptr = dst + x;
-
- __asm__ __volatile__(
- "ulw %[load1], 0(%[src_ptr]) \n\t"
- "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
- "ulw %[load2], 0(%[src_ptr]) \n\t"
-
- "mtlo %[vector4a], $ac0 \n\t"
- "mtlo %[vector4a], $ac1 \n\t"
- "mtlo %[vector4a], $ac2 \n\t"
- "mtlo %[vector4a], $ac3 \n\t"
- "mthi $zero, $ac0 \n\t"
- "mthi $zero, $ac1 \n\t"
- "mthi $zero, $ac2 \n\t"
- "mthi $zero, $ac3 \n\t"
-
- "preceu.ph.qbr %[scratch1], %[load1] \n\t"
- "preceu.ph.qbr %[p1], %[load2] \n\t"
- "precrq.ph.w %[p2], %[p1], %[scratch1] \n\t" /* pixel 2 */
- "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
-
- "dpa.w.ph $ac0, %[p1], %[filter45] \n\t"
- "dpa.w.ph $ac1, %[p2], %[filter45] \n\t"
-
- "preceu.ph.qbl %[scratch1], %[load1] \n\t"
- "preceu.ph.qbl %[p1], %[load2] \n\t"
- "precrq.ph.w %[p2], %[p1], %[scratch1] \n\t" /* pixel 2 */
- "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
-
- "dpa.w.ph $ac2, %[p1], %[filter45] \n\t"
- "dpa.w.ph $ac3, %[p2], %[filter45] \n\t"
-
- "extp %[Temp1], $ac0, 31 \n\t"
- "extp %[Temp2], $ac1, 31 \n\t"
-
- "lbu %[scratch1], 0(%[dst_ptr]) \n\t"
- "lbu %[scratch2], 1(%[dst_ptr]) \n\t"
-
- "lbux %[store1], %[Temp1](%[cm]) \n\t"
- "addqh_r.w %[store1], %[store1], %[scratch1] \n\t" /* pixel 1 */
- "extp %[Temp1], $ac2, 31 \n\t"
-
- "lbux %[store2], %[Temp2](%[cm]) \n\t"
- "addqh_r.w %[store2], %[store2], %[scratch2] \n\t" /* pixel 2 */
- "extp %[Temp2], $ac3, 31 \n\t"
- "lbu %[scratch1], 2(%[dst_ptr]) \n\t"
-
- "sb %[store1], 0(%[dst_ptr]) \n\t"
- "sb %[store2], 1(%[dst_ptr]) \n\t"
- "lbu %[scratch2], 3(%[dst_ptr]) \n\t"
-
- "lbux %[store1], %[Temp1](%[cm]) \n\t"
- "lbux %[store2], %[Temp2](%[cm]) \n\t"
- "addqh_r.w %[store1], %[store1], %[scratch1] \n\t" /* pixel 3 */
- "addqh_r.w %[store2], %[store2], %[scratch2] \n\t" /* pixel 4 */
-
- "sb %[store1], 2(%[dst_ptr]) \n\t"
- "sb %[store2], 3(%[dst_ptr]) \n\t"
-
- : [load1] "=&r"(load1), [load2] "=&r"(load2), [p1] "=&r"(p1),
- [p2] "=&r"(p2), [scratch1] "=&r"(scratch1),
- [scratch2] "=&r"(scratch2), [Temp1] "=&r"(Temp1),
- [Temp2] "=&r"(Temp2), [store1] "=&r"(store1),
- [store2] "=&r"(store2), [src_ptr] "+r"(src_ptr)
- : [filter45] "r"(filter45), [vector4a] "r"(vector4a),
- [src_stride] "r"(src_stride), [cm] "r"(cm), [dst_ptr] "r"(dst_ptr));
- }
-
- /* Next row... */
- src += src_stride;
- dst += dst_stride;
- }
-}
-
-void aom_convolve2_avg_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4, int w,
- int h) {
- uint32_t pos = 38;
-
- assert(y_step_q4 == 16);
-
- /* bit positon for extract from acc */
- __asm__ __volatile__("wrdsp %[pos], 1 \n\t"
- :
- : [pos] "r"(pos));
-
- prefetch_store(dst);
-
- switch (w) {
- case 4:
- case 8:
- case 16:
- case 32:
- convolve_bi_avg_vert_4_dspr2(src, src_stride, dst, dst_stride, filter_y,
- w, h);
- break;
- case 64:
- prefetch_store(dst + 32);
- convolve_bi_avg_vert_64_dspr2(src, src_stride, dst, dst_stride, filter_y,
- h);
- break;
- default:
- aom_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, filter_x,
- x_step_q4, filter_y, y_step_q4, w, h);
- break;
- }
-}
-#endif
diff --git a/aom_dsp/mips/convolve2_avg_horiz_dspr2.c b/aom_dsp/mips/convolve2_avg_horiz_dspr2.c
deleted file mode 100644
index efbdcf6..0000000
--- a/aom_dsp/mips/convolve2_avg_horiz_dspr2.c
+++ /dev/null
@@ -1,802 +0,0 @@
-/*
- * Copyright (c) 2016, Alliance for Open Media. All rights reserved
- *
- * This source code is subject to the terms of the BSD 2 Clause License and
- * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
- * was not distributed with this source code in the LICENSE file, you can
- * obtain it at www.aomedia.org/license/software. If the Alliance for Open
- * Media Patent License 1.0 was not distributed with this source code in the
- * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
- */
-
-#include <assert.h>
-#include <stdio.h>
-
-#include "./aom_dsp_rtcd.h"
-#include "aom_dsp/mips/convolve_common_dspr2.h"
-#include "aom_dsp/aom_convolve.h"
-#include "aom_dsp/aom_dsp_common.h"
-#include "aom_ports/mem.h"
-
-#if HAVE_DSPR2
-static void convolve_bi_avg_horiz_4_dspr2(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- const int16_t *filter_x0, int32_t h) {
- int32_t y;
- uint8_t *cm = aom_ff_cropTbl;
- int32_t Temp1, Temp2, Temp3, Temp4;
- uint32_t vector4a = 64;
- uint32_t tp1, tp2;
- uint32_t p1, p2, p3;
- uint32_t tn1, tn2;
- const int16_t *filter = &filter_x0[3];
- uint32_t filter45;
-
- filter45 = ((const int32_t *)filter)[0];
-
- for (y = h; y--;) {
- /* prefetch data to cache memory */
- prefetch_load(src + src_stride);
- prefetch_load(src + src_stride + 32);
- prefetch_store(dst + dst_stride);
-
- __asm__ __volatile__(
- "ulw %[tp1], 0(%[src]) \n\t"
- "ulw %[tp2], 4(%[src]) \n\t"
-
- /* even 1. pixel */
- "mtlo %[vector4a], $ac3 \n\t"
- "mthi $zero, $ac3 \n\t"
- "preceu.ph.qbr %[p1], %[tp1] \n\t"
- "preceu.ph.qbl %[p2], %[tp1] \n\t"
- "dpa.w.ph $ac3, %[p1], %[filter45] \n\t"
- "extp %[Temp1], $ac3, 31 \n\t"
-
- /* even 2. pixel */
- "mtlo %[vector4a], $ac2 \n\t"
- "mthi $zero, $ac2 \n\t"
- "balign %[tp2], %[tp1], 3 \n\t"
- "dpa.w.ph $ac2, %[p2], %[filter45] \n\t"
- "extp %[Temp3], $ac2, 31 \n\t"
-
- "lbu %[p2], 3(%[dst]) \n\t" /* load odd 2 */
-
- /* odd 1. pixel */
- "lbux %[tp1], %[Temp1](%[cm]) \n\t" /* even 1 */
- "mtlo %[vector4a], $ac3 \n\t"
- "mthi $zero, $ac3 \n\t"
- "lbu %[Temp1], 1(%[dst]) \n\t" /* load odd 1 */
- "preceu.ph.qbr %[p1], %[tp2] \n\t"
- "preceu.ph.qbl %[p3], %[tp2] \n\t"
- "dpa.w.ph $ac3, %[p1], %[filter45] \n\t"
- "extp %[Temp2], $ac3, 31 \n\t"
-
- "lbu %[tn2], 0(%[dst]) \n\t" /* load even 1 */
-
- /* odd 2. pixel */
- "lbux %[tp2], %[Temp3](%[cm]) \n\t" /* even 2 */
- "mtlo %[vector4a], $ac2 \n\t"
- "mthi $zero, $ac2 \n\t"
- "lbux %[tn1], %[Temp2](%[cm]) \n\t" /* odd 1 */
- "addqh_r.w %[tn2], %[tn2], %[tp1] \n\t" /* average even 1 */
- "dpa.w.ph $ac2, %[p3], %[filter45] \n\t"
- "extp %[Temp4], $ac2, 31 \n\t"
-
- "lbu %[tp1], 2(%[dst]) \n\t" /* load even 2 */
- "sb %[tn2], 0(%[dst]) \n\t" /* store even 1 */
-
- /* clamp */
- "addqh_r.w %[Temp1], %[Temp1], %[tn1] \n\t" /* average odd 1 */
- "lbux %[p3], %[Temp4](%[cm]) \n\t" /* odd 2 */
- "sb %[Temp1], 1(%[dst]) \n\t" /* store odd 1 */
-
- "addqh_r.w %[tp1], %[tp1], %[tp2] \n\t" /* average even 2 */
- "sb %[tp1], 2(%[dst]) \n\t" /* store even 2 */
-
- "addqh_r.w %[p2], %[p2], %[p3] \n\t" /* average odd 2 */
- "sb %[p2], 3(%[dst]) \n\t" /* store odd 2 */
-
- : [tp1] "=&r"(tp1), [tp2] "=&r"(tp2), [tn1] "=&r"(tn1),
- [tn2] "=&r"(tn2), [p1] "=&r"(p1), [p2] "=&r"(p2), [p3] "=&r"(p3),
- [Temp1] "=&r"(Temp1), [Temp2] "=&r"(Temp2), [Temp3] "=&r"(Temp3),
- [Temp4] "=&r"(Temp4)
- : [filter45] "r"(filter45), [vector4a] "r"(vector4a), [cm] "r"(cm),
- [dst] "r"(dst), [src] "r"(src));
-
- /* Next row... */
- src += src_stride;
- dst += dst_stride;
- }
-}
-
-static void convolve_bi_avg_horiz_8_dspr2(const uint8_t *src,
- int32_t src_stride, uint8_t *dst,
- int32_t dst_stride,
- const int16_t *filter_x0, int32_t h) {
- int32_t y;
- uint8_t *cm = aom_ff_cropTbl;
- uint32_t vector4a = 64;
- int32_t Temp1, Temp2, Temp3;
- uint32_t tp1, tp2, tp3, tp4;
- uint32_t p1, p2, p3, p4, n1;
- uint32_t st0, st1;
- const int16_t *filter = &filter_x0[3];
- uint32_t filter45;
-
- filter45 = ((const int32_t *)filter)[0];
-
- for (y = h; y--;) {
- /* prefetch data to cache memory */
- prefetch_load(src + src_stride);
- prefetch_load(src + src_stride + 32);
- prefetch_store(dst + dst_stride);
-
- __asm__ __volatile__(
- "ulw %[tp1], 0(%[src]) \n\t"
- "ulw %[tp2], 4(%[src]) \n\t"
-
- /* even 1. pixel */
- "mtlo %[vector4a], $ac3 \n\t"
- "mthi $zero, $ac3 \n\t"
- "mtlo %[vector4a], $ac2 \n\t"
- "mthi $zero, $ac2 \n\t"
- "preceu.ph.qbr %[p1], %[tp1] \n\t"
- "preceu.ph.qbl %[p2], %[tp1] \n\t"
- "preceu.ph.qbr %[p3], %[tp2] \n\t"
- "preceu.ph.qbl %[p4], %[tp2] \n\t"
- "ulw %[tp3], 8(%[src]) \n\t"
- "dpa.w.ph $ac3, %[p1], %[filter45] \n\t"
- "extp %[Temp1], $ac3, 31 \n\t"
- "lbu %[Temp2], 0(%[dst]) \n\t"
- "lbu %[tp4], 2(%[dst]) \n\t"
-
- /* even 2. pixel */
- "dpa.w.ph $ac2, %[p2], %[filter45] \n\t"
- "extp %[Temp3], $ac2, 31 \n\t"
-
- /* even 3. pixel */
- "lbux %[st0], %[Temp1](%[cm]) \n\t"
- "mtlo %[vector4a], $ac1 \n\t"
- "mthi $zero, $ac1 \n\t"
- "lbux %[st1], %[Temp3](%[cm]) \n\t"
- "dpa.w.ph $ac1, %[p3], %[filter45] \n\t"
- "extp %[Temp1], $ac1, 31 \n\t"
-
- "addqh_r.w %[Temp2], %[Temp2], %[st0] \n\t"
- "addqh_r.w %[tp4], %[tp4], %[st1] \n\t"
- "sb %[Temp2], 0(%[dst]) \n\t"
- "sb %[tp4], 2(%[dst]) \n\t"
-
- /* even 4. pixel */
- "mtlo %[vector4a], $ac2 \n\t"
- "mthi $zero, $ac2 \n\t"
- "mtlo %[vector4a], $ac3 \n\t"
- "mthi $zero, $ac3 \n\t"
-
- "balign %[tp3], %[tp2], 3 \n\t"
- "balign %[tp2], %[tp1], 3 \n\t"
-
- "lbux %[st0], %[Temp1](%[cm]) \n\t"
- "lbu %[Temp2], 4(%[dst]) \n\t"
- "addqh_r.w %[Temp2], %[Temp2], %[st0] \n\t"
-
- "dpa.w.ph $ac2, %[p4], %[filter45] \n\t"
- "extp %[Temp3], $ac2, 31 \n\t"
-
- /* odd 1. pixel */
- "mtlo %[vector4a], $ac1 \n\t"
- "mthi $zero, $ac1 \n\t"
- "sb %[Temp2], 4(%[dst]) \n\t"
- "preceu.ph.qbr %[p1], %[tp2] \n\t"
- "preceu.ph.qbl %[p2], %[tp2] \n\t"
- "preceu.ph.qbr %[p3], %[tp3] \n\t"
- "preceu.ph.qbl %[p4], %[tp3] \n\t"
- "dpa.w.ph $ac3, %[p1], %[filter45] \n\t"
- "extp %[Temp2], $ac3, 31 \n\t"
-
- "lbu %[tp1], 6(%[dst]) \n\t"
-
- /* odd 2. pixel */
- "mtlo %[vector4a], $ac3 \n\t"
- "mthi $zero, $ac3 \n\t"
- "mtlo %[vector4a], $ac2 \n\t"
- "mthi $zero, $ac2 \n\t"
- "lbux %[st0], %[Temp3](%[cm]) \n\t"
- "dpa.w.ph $ac1, %[p2], %[filter45] \n\t"
- "extp %[Temp3], $ac1, 31 \n\t"
-
- "lbu %[tp2], 1(%[dst]) \n\t"
- "lbu %[tp3], 3(%[dst]) \n\t"
- "addqh_r.w %[tp1], %[tp1], %[st0] \n\t"
-
- /* odd 3. pixel */
- "lbux %[st1], %[Temp2](%[cm]) \n\t"
- "dpa.w.ph $ac3, %[p3], %[filter45] \n\t"
- "addqh_r.w %[tp2], %[tp2], %[st1] \n\t"
- "extp %[Temp2], $ac3, 31 \n\t"
-
- "lbu %[tp4], 5(%[dst]) \n\t"
-
- /* odd 4. pixel */
- "sb %[tp2], 1(%[dst]) \n\t"
- "sb %[tp1], 6(%[dst]) \n\t"
- "dpa.w.ph $ac2, %[p4], %[filter45] \n\t"
- "extp %[Temp1], $ac2, 31 \n\t"
-
- "lbu %[tp1], 7(%[dst]) \n\t"
-
- /* clamp */
- "lbux %[p4], %[Temp3](%[cm]) \n\t"
- "addqh_r.w %[tp3], %[tp3], %[p4] \n\t"
-
- "lbux %[p2], %[Temp2](%[cm]) \n\t"
- "addqh_r.w %[tp4], %[tp4], %[p2] \n\t"
-
- "lbux %[p1], %[Temp1](%[cm]) \n\t"
- "addqh_r.w %[tp1], %[tp1], %[p1] \n\t"
-
- /* store bytes */
- "sb %[tp3], 3(%[dst]) \n\t"
- "sb %[tp4], 5(%[dst]) \n\t"
- "sb %[tp1], 7(%[dst]) \n\t"
-
- : [tp1] "=&r"(tp1), [tp2] "=&r"(tp2), [tp3] "=&r"(tp3),
- [tp4] "=&r"(tp4), [st0] "=&r"(st0), [st1] "=&r"(st1), [p1] "=&r"(p1),
- [p2] "=&r"(p2), [p3] "=&r"(p3), [p4] "=&r"(p4), [n1] "=&r"(n1),
- [Temp1] "=&r"(Temp1), [Temp2] "=&r"(Temp2), [Temp3] "=&r"(Temp3)
- : [filter45] "r"(filter45), [vector4a] "r"(vector4a), [cm] "r"(cm),
- [dst] "r"(dst), [src] "r"(src));
-
- /* Next row... */
- src += src_stride;
- dst += dst_stride;
- }
-}
-
-static void convolve_bi_avg_horiz_16_dspr2(const uint8_t *src_ptr,
- int32_t src_stride, uint8_t *dst_ptr,
- int32_t dst_stride,
- const int16_t *filter_x0, int32_t h,
- int32_t count) {
- int32_t y, c;
- const uint8_t *src;
- uint8_t *dst;
- uint8_t *cm = aom_ff_cropTbl;
- uint32_t vector_64 = 64;
- int32_t Temp1, Temp2, Temp3;
- uint32_t qload1, qload2, qload3;
- uint32_t p1, p2, p3, p4, p5;
- uint32_t st1, st2, st3;
- const int16_t *filter = &filter_x0[3];
- uint32_t filter45;
-
- filter45 = ((const int32_t *)filter)[0];
-
- for (y = h; y--;) {
- src = src_ptr;
- dst = dst_ptr;
-
- /* prefetch data to cache memory */
- prefetch_load(src_ptr + src_stride);
- prefetch_load(src_ptr + src_stride + 32);
- prefetch_store(dst_ptr + dst_stride);
-
- for (c = 0; c < count; c++) {
- __asm__ __volatile__(
- "ulw %[qload1], 0(%[src]) \n\t"
- "ulw %[qload2], 4(%[src]) \n\t"
-
- /* even 1. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* even 1 */
- "mthi $zero, $ac1 \n\t"
- "mtlo %[vector_64], $ac2 \n\t" /* even 2 */
- "mthi $zero, $ac2 \n\t"
- "preceu.ph.qbr %[p1], %[qload1] \n\t"
- "preceu.ph.qbl %[p2], %[qload1] \n\t"
- "preceu.ph.qbr %[p3], %[qload2] \n\t"
- "preceu.ph.qbl %[p4], %[qload2] \n\t"
- "ulw %[qload3], 8(%[src]) \n\t"
- "dpa.w.ph $ac1, %[p1], %[filter45] \n\t" /* even 1 */
- "extp %[Temp1], $ac1, 31 \n\t" /* even 1 */
- "lbu %[st2], 0(%[dst]) \n\t" /* load even 1 from dst */
-
- /* even 2. pixel */
- "mtlo %[vector_64], $ac3 \n\t" /* even 3 */
- "mthi $zero, $ac3 \n\t"
- "preceu.ph.qbr %[p1], %[qload3] \n\t"
- "preceu.ph.qbl %[p5], %[qload3] \n\t"
- "ulw %[qload1], 12(%[src]) \n\t"
- "dpa.w.ph $ac2, %[p2], %[filter45] \n\t" /* even 1 */
- "extp %[Temp2], $ac2, 31 \n\t" /* even 1 */
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 1 */
-
- "lbu %[qload3], 2(%[dst]) \n\t" /* load even 2 from dst */
-
- /* even 3. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* even 4 */
- "mthi $zero, $ac1 \n\t"
- "addqh_r.w %[st2], %[st2], %[st1] \n\t" /* average even 1 */
- "preceu.ph.qbr %[p2], %[qload1] \n\t"
- "sb %[st2], 0(%[dst]) \n\t" /* store even 1 to dst */
- "dpa.w.ph $ac3, %[p3], %[filter45] \n\t" /* even 3 */
- "extp %[Temp3], $ac3, 31 \n\t" /* even 3 */
- "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 1 */
-
- /* even 4. pixel */
- "mtlo %[vector_64], $ac2 \n\t" /* even 5 */
- "mthi $zero, $ac2 \n\t"
- "addqh_r.w %[qload3], %[qload3], %[st2] \n\t" /* average even 2 */
- "preceu.ph.qbl %[p3], %[qload1] \n\t"
- "sb %[qload3], 2(%[dst]) \n\t" /* store even 2 to dst */
- "lbu %[qload3], 4(%[dst]) \n\t" /* load even 3 from dst */
- "lbu %[qload1], 6(%[dst]) \n\t" /* load even 4 from dst */
- "dpa.w.ph $ac1, %[p4], %[filter45] \n\t" /* even 4 */
- "extp %[Temp1], $ac1, 31 \n\t" /* even 4 */
- "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 3 */
-
- /* even 5. pixel */
- "mtlo %[vector_64], $ac3 \n\t" /* even 6 */
- "mthi $zero, $ac3 \n\t"
- "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average even 3 */
- "sb %[qload3], 4(%[dst]) \n\t" /* store even 3 to dst */
- "dpa.w.ph $ac2, %[p1], %[filter45] \n\t" /* even 5 */
- "extp %[Temp2], $ac2, 31 \n\t" /* even 5 */
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 4 */
-
- /* even 6. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* even 7 */
- "mthi $zero, $ac1 \n\t"
- "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average even 4 */
- "sb %[qload1], 6(%[dst]) \n\t" /* store even 4 to dst */
- "dpa.w.ph $ac3, %[p5], %[filter45] \n\t" /* even 6 */
- "lbu %[qload2], 8(%[dst]) \n\t" /* load even 5 from dst */
- "extp %[Temp3], $ac3, 31 \n\t" /* even 6 */
- "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 5 */
-
- /* even 7. pixel */
- "mtlo %[vector_64], $ac2 \n\t" /* even 8 */
- "mthi $zero, $ac2 \n\t"
- "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average even 5 */
- "sb %[qload2], 8(%[dst]) \n\t" /* store even 5 to dst */
- "dpa.w.ph $ac1, %[p2], %[filter45] \n\t" /* even 7 */
- "lbu %[qload3], 10(%[dst]) \n\t" /* load even 6 from dst */
- "extp %[Temp1], $ac1, 31 \n\t" /* even 7 */
- "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 6 */
-
- "lbu %[st2], 12(%[dst]) \n\t" /* load even 7 from dst */
-
- /* even 8. pixel */
- "mtlo %[vector_64], $ac3 \n\t" /* odd 1 */
- "mthi $zero, $ac3 \n\t"
- "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average even 6 */
- "dpa.w.ph $ac2, %[p3], %[filter45] \n\t" /* even 8 */
- "sb %[qload3], 10(%[dst]) \n\t" /* store even 6 to dst */
- "extp %[Temp2], $ac2, 31 \n\t" /* even 8 */
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 7 */
-
- /* ODD pixels */
- "ulw %[qload1], 1(%[src]) \n\t"
- "ulw %[qload2], 5(%[src]) \n\t"
-
- "addqh_r.w %[st2], %[st2], %[st1] \n\t" /* average even 7 */
-
- /* odd 1. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* odd 2 */
- "mthi $zero, $ac1 \n\t"
- "preceu.ph.qbr %[p1], %[qload1] \n\t"
- "preceu.ph.qbl %[p2], %[qload1] \n\t"
- "preceu.ph.qbr %[p3], %[qload2] \n\t"
- "preceu.ph.qbl %[p4], %[qload2] \n\t"
- "sb %[st2], 12(%[dst]) \n\t" /* store even 7 to dst */
- "ulw %[qload3], 9(%[src]) \n\t"
- "dpa.w.ph $ac3, %[p1], %[filter45] \n\t" /* odd 1 */
- "lbu %[qload2], 14(%[dst]) \n\t" /* load even 8 from dst */
- "extp %[Temp3], $ac3, 31 \n\t" /* odd 1 */
- "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 8 */
-
- "lbu %[st1], 1(%[dst]) \n\t" /* load odd 1 from dst */
-
- /* odd 2. pixel */
- "mtlo %[vector_64], $ac2 \n\t" /* odd 3 */
- "mthi $zero, $ac2 \n\t"
- "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average even 8 */
- "preceu.ph.qbr %[p1], %[qload3] \n\t"
- "preceu.ph.qbl %[p5], %[qload3] \n\t"
- "sb %[qload2], 14(%[dst]) \n\t" /* store even 8 to dst */
- "ulw %[qload1], 13(%[src]) \n\t"
- "dpa.w.ph $ac1, %[p2], %[filter45] \n\t" /* odd 2 */
- "lbu %[qload3], 3(%[dst]) \n\t" /* load odd 2 from dst */
- "extp %[Temp1], $ac1, 31 \n\t" /* odd 2 */
- "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 1 */
-
- /* odd 3. pixel */
- "mtlo %[vector_64], $ac3 \n\t" /* odd 4 */
- "mthi $zero, $ac3 \n\t"
- "addqh_r.w %[st3], %[st3], %[st1] \n\t" /* average odd 1 */
- "preceu.ph.qbr %[p2], %[qload1] \n\t"
- "dpa.w.ph $ac2, %[p3], %[filter45] \n\t" /* odd 3 */
- "sb %[st3], 1(%[dst]) \n\t" /* store odd 1 to dst */
- "extp %[Temp2], $ac2, 31 \n\t" /* odd 3 */
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 2 */
-
- /* odd 4. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* odd 5 */
- "mthi $zero, $ac1 \n\t"
- "addqh_r.w %[qload3], %[qload3], %[st1] \n\t" /* average odd 2 */
- "preceu.ph.qbl %[p3], %[qload1] \n\t"
- "sb %[qload3], 3(%[dst]) \n\t" /* store odd 2 to dst */
- "lbu %[qload1], 5(%[dst]) \n\t" /* load odd 3 from dst */
- "dpa.w.ph $ac3, %[p4], %[filter45] \n\t" /* odd 4 */
- "extp %[Temp3], $ac3, 31 \n\t" /* odd 4 */
- "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 3 */
-
- "lbu %[st1], 7(%[dst]) \n\t" /* load odd 4 from dst */
-
- /* odd 5. pixel */
- "mtlo %[vector_64], $ac2 \n\t" /* odd 6 */
- "mthi $zero, $ac2 \n\t"
- "addqh_r.w %[qload1], %[qload1], %[st2] \n\t" /* average odd 3 */
- "sb %[qload1], 5(%[dst]) \n\t" /* store odd 3 to dst */
- "dpa.w.ph $ac1, %[p1], %[filter45] \n\t" /* odd 5 */
- "extp %[Temp1], $ac1, 31 \n\t" /* odd 5 */
- "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 4 */
-
- "lbu %[qload1], 9(%[dst]) \n\t" /* load odd 5 from dst */
-
- /* odd 6. pixel */
- "mtlo %[vector_64], $ac3 \n\t" /* odd 7 */
- "mthi $zero, $ac3 \n\t"
- "addqh_r.w %[st1], %[st1], %[st3] \n\t" /* average odd 4 */
- "sb %[st1], 7(%[dst]) \n\t" /* store odd 4 to dst */
- "dpa.w.ph $ac2, %[p5], %[filter45] \n\t" /* odd 6 */
- "extp %[Temp2], $ac2, 31 \n\t" /* odd 6 */
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 5 */
-
- /* odd 7. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* odd 8 */
- "mthi $zero, $ac1 \n\t"
- "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average odd 5 */
- "sb %[qload1], 9(%[dst]) \n\t" /* store odd 5 to dst */
- "lbu %[qload2], 11(%[dst]) \n\t" /* load odd 6 from dst */
- "dpa.w.ph $ac3, %[p2], %[filter45] \n\t" /* odd 7 */
- "extp %[Temp3], $ac3, 31 \n\t" /* odd 7 */
-
- "lbu %[qload3], 13(%[dst]) \n\t" /* load odd 7 from dst */
-
- /* odd 8. pixel */
- "dpa.w.ph $ac1, %[p3], %[filter45] \n\t" /* odd 8 */
- "extp %[Temp1], $ac1, 31 \n\t" /* odd 8 */
-
- "lbu %[qload1], 15(%[dst]) \n\t" /* load odd 8 from dst */
-
- "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 6 */
- "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average odd 6 */
-
- "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 7 */
- "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average odd 7 */
-
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 8 */
- "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average odd 8 */
-
- "sb %[qload2], 11(%[dst]) \n\t" /* store odd 6 to dst */
- "sb %[qload3], 13(%[dst]) \n\t" /* store odd 7 to dst */
- "sb %[qload1], 15(%[dst]) \n\t" /* store odd 8 to dst */
-
- : [qload1] "=&r"(qload1), [qload2] "=&r"(qload2), [st1] "=&r"(st1),
- [st2] "=&r"(st2), [st3] "=&r"(st3), [p1] "=&r"(p1), [p2] "=&r"(p2),
- [p3] "=&r"(p3), [p4] "=&r"(p4), [qload3] "=&r"(qload3),
- [p5] "=&r"(p5), [Temp1] "=&r"(Temp1), [Temp2] "=&r"(Temp2),
- [Temp3] "=&r"(Temp3)
- : [filter45] "r"(filter45), [vector_64] "r"(vector_64), [cm] "r"(cm),
- [dst] "r"(dst), [src] "r"(src));
-
- src += 16;
- dst += 16;
- }
-
- /* Next row... */
- src_ptr += src_stride;
- dst_ptr += dst_stride;
- }
-}
-
-static void convolve_bi_avg_horiz_64_dspr2(const uint8_t *src_ptr,
- int32_t src_stride, uint8_t *dst_ptr,
- int32_t dst_stride,
- const int16_t *filter_x0,
- int32_t h) {
- int32_t y, c;
- const uint8_t *src;
- uint8_t *dst;
- uint8_t *cm = aom_ff_cropTbl;
- uint32_t vector_64 = 64;
- int32_t Temp1, Temp2, Temp3;
- uint32_t qload1, qload2, qload3;
- uint32_t p1, p2, p3, p4, p5;
- uint32_t st1, st2, st3;
- const int16_t *filter = &filter_x0[3];
- uint32_t filter45;
-
- filter45 = ((const int32_t *)filter)[0];
-
- for (y = h; y--;) {
- src = src_ptr;
- dst = dst_ptr;
-
- /* prefetch data to cache memory */
- prefetch_load(src_ptr + src_stride);
- prefetch_load(src_ptr + src_stride + 32);
- prefetch_load(src_ptr + src_stride + 64);
- prefetch_store(dst_ptr + dst_stride);
- prefetch_store(dst_ptr + dst_stride + 32);
-
- for (c = 0; c < 4; c++) {
- __asm__ __volatile__(
- "ulw %[qload1], 0(%[src]) \n\t"
- "ulw %[qload2], 4(%[src]) \n\t"
-
- /* even 1. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* even 1 */
- "mthi $zero, $ac1 \n\t"
- "mtlo %[vector_64], $ac2 \n\t" /* even 2 */
- "mthi $zero, $ac2 \n\t"
- "preceu.ph.qbr %[p1], %[qload1] \n\t"
- "preceu.ph.qbl %[p2], %[qload1] \n\t"
- "preceu.ph.qbr %[p3], %[qload2] \n\t"
- "preceu.ph.qbl %[p4], %[qload2] \n\t"
- "ulw %[qload3], 8(%[src]) \n\t"
- "dpa.w.ph $ac1, %[p1], %[filter45] \n\t" /* even 1 */
- "extp %[Temp1], $ac1, 31 \n\t" /* even 1 */
- "lbu %[st2], 0(%[dst]) \n\t" /* load even 1 from dst */
-
- /* even 2. pixel */
- "mtlo %[vector_64], $ac3 \n\t" /* even 3 */
- "mthi $zero, $ac3 \n\t"
- "preceu.ph.qbr %[p1], %[qload3] \n\t"
- "preceu.ph.qbl %[p5], %[qload3] \n\t"
- "ulw %[qload1], 12(%[src]) \n\t"
- "dpa.w.ph $ac2, %[p2], %[filter45] \n\t" /* even 1 */
- "extp %[Temp2], $ac2, 31 \n\t" /* even 1 */
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 1 */
-
- "lbu %[qload3], 2(%[dst]) \n\t" /* load even 2 from dst */
-
- /* even 3. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* even 4 */
- "mthi $zero, $ac1 \n\t"
- "addqh_r.w %[st2], %[st2], %[st1] \n\t" /* average even 1 */
- "preceu.ph.qbr %[p2], %[qload1] \n\t"
- "sb %[st2], 0(%[dst]) \n\t" /* store even 1 to dst */
- "dpa.w.ph $ac3, %[p3], %[filter45] \n\t" /* even 3 */
- "extp %[Temp3], $ac3, 31 \n\t" /* even 3 */
- "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 1 */
-
- /* even 4. pixel */
- "mtlo %[vector_64], $ac2 \n\t" /* even 5 */
- "mthi $zero, $ac2 \n\t"
- "addqh_r.w %[qload3], %[qload3], %[st2] \n\t" /* average even 2 */
- "preceu.ph.qbl %[p3], %[qload1] \n\t"
- "sb %[qload3], 2(%[dst]) \n\t" /* store even 2 to dst */
- "lbu %[qload3], 4(%[dst]) \n\t" /* load even 3 from dst */
- "lbu %[qload1], 6(%[dst]) \n\t" /* load even 4 from dst */
- "dpa.w.ph $ac1, %[p4], %[filter45] \n\t" /* even 4 */
- "extp %[Temp1], $ac1, 31 \n\t" /* even 4 */
- "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 3 */
-
- /* even 5. pixel */
- "mtlo %[vector_64], $ac3 \n\t" /* even 6 */
- "mthi $zero, $ac3 \n\t"
- "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average even 3 */
- "sb %[qload3], 4(%[dst]) \n\t" /* store even 3 to dst */
- "dpa.w.ph $ac2, %[p1], %[filter45] \n\t" /* even 5 */
- "extp %[Temp2], $ac2, 31 \n\t" /* even 5 */
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 4 */
-
- /* even 6. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* even 7 */
- "mthi $zero, $ac1 \n\t"
- "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average even 4 */
- "sb %[qload1], 6(%[dst]) \n\t" /* store even 4 to dst */
- "dpa.w.ph $ac3, %[p5], %[filter45] \n\t" /* even 6 */
- "lbu %[qload2], 8(%[dst]) \n\t" /* load even 5 from dst */
- "extp %[Temp3], $ac3, 31 \n\t" /* even 6 */
- "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 5 */
-
- /* even 7. pixel */
- "mtlo %[vector_64], $ac2 \n\t" /* even 8 */
- "mthi $zero, $ac2 \n\t"
- "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average even 5 */
- "sb %[qload2], 8(%[dst]) \n\t" /* store even 5 to dst */
- "dpa.w.ph $ac1, %[p2], %[filter45] \n\t" /* even 7 */
- "lbu %[qload3], 10(%[dst]) \n\t" /* load even 6 from dst */
- "extp %[Temp1], $ac1, 31 \n\t" /* even 7 */
- "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 6 */
-
- "lbu %[st2], 12(%[dst]) \n\t" /* load even 7 from dst */
-
- /* even 8. pixel */
- "mtlo %[vector_64], $ac3 \n\t" /* odd 1 */
- "mthi $zero, $ac3 \n\t"
- "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average even 6 */
- "dpa.w.ph $ac2, %[p3], %[filter45] \n\t" /* even 8 */
- "sb %[qload3], 10(%[dst]) \n\t" /* store even 6 to dst */
- "extp %[Temp2], $ac2, 31 \n\t" /* even 8 */
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 7 */
-
- /* ODD pixels */
- "ulw %[qload1], 1(%[src]) \n\t"
- "ulw %[qload2], 5(%[src]) \n\t"
-
- "addqh_r.w %[st2], %[st2], %[st1] \n\t" /* average even 7 */
-
- /* odd 1. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* odd 2 */
- "mthi $zero, $ac1 \n\t"
- "preceu.ph.qbr %[p1], %[qload1] \n\t"
- "preceu.ph.qbl %[p2], %[qload1] \n\t"
- "preceu.ph.qbr %[p3], %[qload2] \n\t"
- "preceu.ph.qbl %[p4], %[qload2] \n\t"
- "sb %[st2], 12(%[dst]) \n\t" /* store even 7 to dst */
- "ulw %[qload3], 9(%[src]) \n\t"
- "dpa.w.ph $ac3, %[p1], %[filter45] \n\t" /* odd 1 */
- "lbu %[qload2], 14(%[dst]) \n\t" /* load even 8 from dst */
- "extp %[Temp3], $ac3, 31 \n\t" /* odd 1 */
- "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 8 */
-
- "lbu %[st1], 1(%[dst]) \n\t" /* load odd 1 from dst */
-
- /* odd 2. pixel */
- "mtlo %[vector_64], $ac2 \n\t" /* odd 3 */
- "mthi $zero, $ac2 \n\t"
- "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average even 8 */
- "preceu.ph.qbr %[p1], %[qload3] \n\t"
- "preceu.ph.qbl %[p5], %[qload3] \n\t"
- "sb %[qload2], 14(%[dst]) \n\t" /* store even 8 to dst */
- "ulw %[qload1], 13(%[src]) \n\t"
- "dpa.w.ph $ac1, %[p2], %[filter45] \n\t" /* odd 2 */
- "lbu %[qload3], 3(%[dst]) \n\t" /* load odd 2 from dst */
- "extp %[Temp1], $ac1, 31 \n\t" /* odd 2 */
- "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 1 */
-
- /* odd 3. pixel */
- "mtlo %[vector_64], $ac3 \n\t" /* odd 4 */
- "mthi $zero, $ac3 \n\t"
- "addqh_r.w %[st3], %[st3], %[st1] \n\t" /* average odd 1 */
- "preceu.ph.qbr %[p2], %[qload1] \n\t"
- "dpa.w.ph $ac2, %[p3], %[filter45] \n\t" /* odd 3 */
- "sb %[st3], 1(%[dst]) \n\t" /* store odd 1 to dst */
- "extp %[Temp2], $ac2, 31 \n\t" /* odd 3 */
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 2 */
-
- /* odd 4. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* odd 5 */
- "mthi $zero, $ac1 \n\t"
- "addqh_r.w %[qload3], %[qload3], %[st1] \n\t" /* average odd 2 */
- "preceu.ph.qbl %[p3], %[qload1] \n\t"
- "sb %[qload3], 3(%[dst]) \n\t" /* store odd 2 to dst */
- "lbu %[qload1], 5(%[dst]) \n\t" /* load odd 3 from dst */
- "dpa.w.ph $ac3, %[p4], %[filter45] \n\t" /* odd 4 */
- "extp %[Temp3], $ac3, 31 \n\t" /* odd 4 */
- "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 3 */
-
- "lbu %[st1], 7(%[dst]) \n\t" /* load odd 4 from dst */
-
- /* odd 5. pixel */
- "mtlo %[vector_64], $ac2 \n\t" /* odd 6 */
- "mthi $zero, $ac2 \n\t"
- "addqh_r.w %[qload1], %[qload1], %[st2] \n\t" /* average odd 3 */
- "sb %[qload1], 5(%[dst]) \n\t" /* store odd 3 to dst */
- "dpa.w.ph $ac1, %[p1], %[filter45] \n\t" /* odd 5 */
- "extp %[Temp1], $ac1, 31 \n\t" /* odd 5 */
- "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 4 */
-
- "lbu %[qload1], 9(%[dst]) \n\t" /* load odd 5 from dst */
-
- /* odd 6. pixel */
- "mtlo %[vector_64], $ac3 \n\t" /* odd 7 */
- "mthi $zero, $ac3 \n\t"
- "addqh_r.w %[st1], %[st1], %[st3] \n\t" /* average odd 4 */
- "sb %[st1], 7(%[dst]) \n\t" /* store odd 4 to dst */
- "dpa.w.ph $ac2, %[p5], %[filter45] \n\t" /* odd 6 */
- "extp %[Temp2], $ac2, 31 \n\t" /* odd 6 */
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 5 */
-
- /* odd 7. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* odd 8 */
- "mthi $zero, $ac1 \n\t"
- "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average odd 5 */
- "sb %[qload1], 9(%[dst]) \n\t" /* store odd 5 to dst */
- "lbu %[qload2], 11(%[dst]) \n\t" /* load odd 6 from dst */
- "dpa.w.ph $ac3, %[p2], %[filter45] \n\t" /* odd 7 */
- "extp %[Temp3], $ac3, 31 \n\t" /* odd 7 */
-
- "lbu %[qload3], 13(%[dst]) \n\t" /* load odd 7 from dst */
-
- /* odd 8. pixel */
- "dpa.w.ph $ac1, %[p3], %[filter45] \n\t" /* odd 8 */
- "extp %[Temp1], $ac1, 31 \n\t" /* odd 8 */
-
- "lbu %[qload1], 15(%[dst]) \n\t" /* load odd 8 from dst */
-
- "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 6 */
- "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average odd 6 */
-
- "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 7 */
- "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average odd 7 */
-
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 8 */
- "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average odd 8 */
-
- "sb %[qload2], 11(%[dst]) \n\t" /* store odd 6 to dst */
- "sb %[qload3], 13(%[dst]) \n\t" /* store odd 7 to dst */
- "sb %[qload1], 15(%[dst]) \n\t" /* store odd 8 to dst */
-
- : [qload1] "=&r"(qload1), [qload2] "=&r"(qload2), [st1] "=&r"(st1),
- [st2] "=&r"(st2), [st3] "=&r"(st3), [p1] "=&r"(p1), [p2] "=&r"(p2),
- [p3] "=&r"(p3), [p4] "=&r"(p4), [qload3] "=&r"(qload3),
- [p5] "=&r"(p5), [Temp1] "=&r"(Temp1), [Temp2] "=&r"(Temp2),
- [Temp3] "=&r"(Temp3)
- : [filter45] "r"(filter45), [vector_64] "r"(vector_64), [cm] "r"(cm),
- [dst] "r"(dst), [src] "r"(src));
-
- src += 16;
- dst += 16;
- }
-
- /* Next row... */
- src_ptr += src_stride;
- dst_ptr += dst_stride;
- }
-}
-
-void aom_convolve2_avg_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h) {
- uint32_t pos = 38;
-
- assert(x_step_q4 == 16);
-
- /* bit positon for extract from acc */
- __asm__ __volatile__("wrdsp %[pos], 1 \n\t"
- :
- : [pos] "r"(pos));
-
- /* prefetch data to cache memory */
- prefetch_load(src);
- prefetch_load(src + 32);
- prefetch_store(dst);
-
- switch (w) {
- case 4:
- convolve_bi_avg_horiz_4_dspr2(src, src_stride, dst, dst_stride, filter_x,
- h);
- break;
- case 8:
- convolve_bi_avg_horiz_8_dspr2(src, src_stride, dst, dst_stride, filter_x,
- h);
- break;
- case 16:
- convolve_bi_avg_horiz_16_dspr2(src, src_stride, dst, dst_stride, filter_x,
- h, 1);
- break;
- case 32:
- convolve_bi_avg_horiz_16_dspr2(src, src_stride, dst, dst_stride, filter_x,
- h, 2);
- break;
- case 64:
- prefetch_load(src + 64);
- prefetch_store(dst + 32);
-
- convolve_bi_avg_horiz_64_dspr2(src, src_stride, dst, dst_stride, filter_x,
- h);
- break;
- default:
- aom_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride, filter_x,
- x_step_q4, filter_y, y_step_q4, w, h);
- break;
- }
-}
-#endif
diff --git a/aom_dsp/mips/convolve8_avg_dspr2.c b/aom_dsp/mips/convolve8_avg_dspr2.c
deleted file mode 100644
index 3574da1..0000000
--- a/aom_dsp/mips/convolve8_avg_dspr2.c
+++ /dev/null
@@ -1,646 +0,0 @@
-/*
- * Copyright (c) 2016, Alliance for Open Media. All rights reserved
- *
- * This source code is subject to the terms of the BSD 2 Clause License and
- * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
- * was not distributed with this source code in the LICENSE file, you can
- * obtain it at www.aomedia.org/license/software. If the Alliance for Open
- * Media Patent License 1.0 was not distributed with this source code in the
- * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
- */
-
-#include <assert.h>
-#include <stdio.h>
-
-#include "./aom_dsp_rtcd.h"
-#include "aom_dsp/mips/convolve_common_dspr2.h"
-#include "aom_dsp/aom_convolve.h"
-#include "aom_dsp/aom_dsp_common.h"
-#include "aom_ports/mem.h"
-
-#if HAVE_DSPR2
-static void convolve_avg_vert_4_dspr2(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- const int16_t *filter_y, int32_t w,
- int32_t h) {
- int32_t x, y;
- const uint8_t *src_ptr;
- uint8_t *dst_ptr;
- uint8_t *cm = aom_ff_cropTbl;
- uint32_t vector4a = 64;
- uint32_t load1, load2, load3, load4;
- uint32_t p1, p2;
- uint32_t n1, n2;
- uint32_t scratch1, scratch2;
- uint32_t store1, store2;
- int32_t vector1b, vector2b, vector3b, vector4b;
- int32_t Temp1, Temp2;
-
- vector1b = ((const int32_t *)filter_y)[0];
- vector2b = ((const int32_t *)filter_y)[1];
- vector3b = ((const int32_t *)filter_y)[2];
- vector4b = ((const int32_t *)filter_y)[3];
-
- src -= 3 * src_stride;
-
- for (y = h; y--;) {
- /* prefetch data to cache memory */
- prefetch_store(dst + dst_stride);
-
- for (x = 0; x < w; x += 4) {
- src_ptr = src + x;
- dst_ptr = dst + x;
-
- __asm__ __volatile__(
- "ulw %[load1], 0(%[src_ptr]) \n\t"
- "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
- "ulw %[load2], 0(%[src_ptr]) \n\t"
- "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
- "ulw %[load3], 0(%[src_ptr]) \n\t"
- "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
- "ulw %[load4], 0(%[src_ptr]) \n\t"
-
- "mtlo %[vector4a], $ac0 \n\t"
- "mtlo %[vector4a], $ac1 \n\t"
- "mtlo %[vector4a], $ac2 \n\t"
- "mtlo %[vector4a], $ac3 \n\t"
- "mthi $zero, $ac0 \n\t"
- "mthi $zero, $ac1 \n\t"
- "mthi $zero, $ac2 \n\t"
- "mthi $zero, $ac3 \n\t"
-
- "preceu.ph.qbr %[scratch1], %[load1] \n\t"
- "preceu.ph.qbr %[p1], %[load2] \n\t"
- "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
- "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
- "preceu.ph.qbr %[scratch2], %[load3] \n\t"
- "preceu.ph.qbr %[p2], %[load4] \n\t"
- "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
- "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
-
- "dpa.w.ph $ac0, %[p1], %[vector1b] \n\t"
- "dpa.w.ph $ac0, %[p2], %[vector2b] \n\t"
- "dpa.w.ph $ac1, %[n1], %[vector1b] \n\t"
- "dpa.w.ph $ac1, %[n2], %[vector2b] \n\t"
-
- "preceu.ph.qbl %[scratch1], %[load1] \n\t"
- "preceu.ph.qbl %[p1], %[load2] \n\t"
- "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
- "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
- "preceu.ph.qbl %[scratch2], %[load3] \n\t"
- "preceu.ph.qbl %[p2], %[load4] \n\t"
- "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
- "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
-
- "dpa.w.ph $ac2, %[p1], %[vector1b] \n\t"
- "dpa.w.ph $ac2, %[p2], %[vector2b] \n\t"
- "dpa.w.ph $ac3, %[n1], %[vector1b] \n\t"
- "dpa.w.ph $ac3, %[n2], %[vector2b] \n\t"
-
- "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
- "ulw %[load1], 0(%[src_ptr]) \n\t"
- "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
- "ulw %[load2], 0(%[src_ptr]) \n\t"
- "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
- "ulw %[load3], 0(%[src_ptr]) \n\t"
- "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
- "ulw %[load4], 0(%[src_ptr]) \n\t"
-
- "preceu.ph.qbr %[scratch1], %[load1] \n\t"
- "preceu.ph.qbr %[p1], %[load2] \n\t"
- "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
- "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
- "preceu.ph.qbr %[scratch2], %[load3] \n\t"
- "preceu.ph.qbr %[p2], %[load4] \n\t"
- "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
- "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
-
- "dpa.w.ph $ac0, %[p1], %[vector3b] \n\t"
- "dpa.w.ph $ac0, %[p2], %[vector4b] \n\t"
- "extp %[Temp1], $ac0, 31 \n\t"
- "dpa.w.ph $ac1, %[n1], %[vector3b] \n\t"
- "dpa.w.ph $ac1, %[n2], %[vector4b] \n\t"
- "extp %[Temp2], $ac1, 31 \n\t"
-
- "preceu.ph.qbl %[scratch1], %[load1] \n\t"
- "preceu.ph.qbl %[p1], %[load2] \n\t"
- "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
- "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
- "lbu %[scratch1], 0(%[dst_ptr]) \n\t"
- "preceu.ph.qbl %[scratch2], %[load3] \n\t"
- "preceu.ph.qbl %[p2], %[load4] \n\t"
- "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
- "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
- "lbu %[scratch2], 1(%[dst_ptr]) \n\t"
-
- "lbux %[store1], %[Temp1](%[cm]) \n\t"
- "dpa.w.ph $ac2, %[p1], %[vector3b] \n\t"
- "dpa.w.ph $ac2, %[p2], %[vector4b] \n\t"
- "addqh_r.w %[store1], %[store1], %[scratch1] \n\t" /* pixel 1 */
- "extp %[Temp1], $ac2, 31 \n\t"
-
- "lbux %[store2], %[Temp2](%[cm]) \n\t"
- "dpa.w.ph $ac3, %[n1], %[vector3b] \n\t"
- "dpa.w.ph $ac3, %[n2], %[vector4b] \n\t"
- "addqh_r.w %[store2], %[store2], %[scratch2] \n\t" /* pixel 2 */
- "extp %[Temp2], $ac3, 31 \n\t"
- "lbu %[scratch1], 2(%[dst_ptr]) \n\t"
-
- "sb %[store1], 0(%[dst_ptr]) \n\t"
- "sb %[store2], 1(%[dst_ptr]) \n\t"
- "lbu %[scratch2], 3(%[dst_ptr]) \n\t"
-
- "lbux %[store1], %[Temp1](%[cm]) \n\t"
- "lbux %[store2], %[Temp2](%[cm]) \n\t"
- "addqh_r.w %[store1], %[store1], %[scratch1] \n\t" /* pixel 3 */
- "addqh_r.w %[store2], %[store2], %[scratch2] \n\t" /* pixel 4 */
-
- "sb %[store1], 2(%[dst_ptr]) \n\t"
- "sb %[store2], 3(%[dst_ptr]) \n\t"
-
- : [load1] "=&r"(load1), [load2] "=&r"(load2), [load3] "=&r"(load3),
- [load4] "=&r"(load4), [p1] "=&r"(p1), [p2] "=&r"(p2),
- [n1] "=&r"(n1), [n2] "=&r"(n2), [scratch1] "=&r"(scratch1),
- [scratch2] "=&r"(scratch2), [Temp1] "=&r"(Temp1),
- [Temp2] "=&r"(Temp2), [store1] "=&r"(store1),
- [store2] "=&r"(store2), [src_ptr] "+r"(src_ptr)
- : [vector1b] "r"(vector1b), [vector2b] "r"(vector2b),
- [vector3b] "r"(vector3b), [vector4b] "r"(vector4b),
- [vector4a] "r"(vector4a), [src_stride] "r"(src_stride),
- [cm] "r"(cm), [dst_ptr] "r"(dst_ptr));
- }
-
- /* Next row... */
- src += src_stride;
- dst += dst_stride;
- }
-}
-
-static void convolve_avg_vert_64_dspr2(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- const int16_t *filter_y, int32_t h) {
- int32_t x, y;
- const uint8_t *src_ptr;
- uint8_t *dst_ptr;
- uint8_t *cm = aom_ff_cropTbl;
- uint32_t vector4a = 64;
- uint32_t load1, load2, load3, load4;
- uint32_t p1, p2;
- uint32_t n1, n2;
- uint32_t scratch1, scratch2;
- uint32_t store1, store2;
- int32_t vector1b, vector2b, vector3b, vector4b;
- int32_t Temp1, Temp2;
-
- vector1b = ((const int32_t *)filter_y)[0];
- vector2b = ((const int32_t *)filter_y)[1];
- vector3b = ((const int32_t *)filter_y)[2];
- vector4b = ((const int32_t *)filter_y)[3];
-
- src -= 3 * src_stride;
-
- for (y = h; y--;) {
- /* prefetch data to cache memory */
- prefetch_store(dst + dst_stride);
- prefetch_store(dst + dst_stride + 32);
-
- for (x = 0; x < 64; x += 4) {
- src_ptr = src + x;
- dst_ptr = dst + x;
-
- __asm__ __volatile__(
- "ulw %[load1], 0(%[src_ptr]) \n\t"
- "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
- "ulw %[load2], 0(%[src_ptr]) \n\t"
- "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
- "ulw %[load3], 0(%[src_ptr]) \n\t"
- "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
- "ulw %[load4], 0(%[src_ptr]) \n\t"
-
- "mtlo %[vector4a], $ac0 \n\t"
- "mtlo %[vector4a], $ac1 \n\t"
- "mtlo %[vector4a], $ac2 \n\t"
- "mtlo %[vector4a], $ac3 \n\t"
- "mthi $zero, $ac0 \n\t"
- "mthi $zero, $ac1 \n\t"
- "mthi $zero, $ac2 \n\t"
- "mthi $zero, $ac3 \n\t"
-
- "preceu.ph.qbr %[scratch1], %[load1] \n\t"
- "preceu.ph.qbr %[p1], %[load2] \n\t"
- "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
- "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
- "preceu.ph.qbr %[scratch2], %[load3] \n\t"
- "preceu.ph.qbr %[p2], %[load4] \n\t"
- "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
- "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
-
- "dpa.w.ph $ac0, %[p1], %[vector1b] \n\t"
- "dpa.w.ph $ac0, %[p2], %[vector2b] \n\t"
- "dpa.w.ph $ac1, %[n1], %[vector1b] \n\t"
- "dpa.w.ph $ac1, %[n2], %[vector2b] \n\t"
-
- "preceu.ph.qbl %[scratch1], %[load1] \n\t"
- "preceu.ph.qbl %[p1], %[load2] \n\t"
- "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
- "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
- "preceu.ph.qbl %[scratch2], %[load3] \n\t"
- "preceu.ph.qbl %[p2], %[load4] \n\t"
- "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
- "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
-
- "dpa.w.ph $ac2, %[p1], %[vector1b] \n\t"
- "dpa.w.ph $ac2, %[p2], %[vector2b] \n\t"
- "dpa.w.ph $ac3, %[n1], %[vector1b] \n\t"
- "dpa.w.ph $ac3, %[n2], %[vector2b] \n\t"
-
- "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
- "ulw %[load1], 0(%[src_ptr]) \n\t"
- "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
- "ulw %[load2], 0(%[src_ptr]) \n\t"
- "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
- "ulw %[load3], 0(%[src_ptr]) \n\t"
- "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
- "ulw %[load4], 0(%[src_ptr]) \n\t"
-
- "preceu.ph.qbr %[scratch1], %[load1] \n\t"
- "preceu.ph.qbr %[p1], %[load2] \n\t"
- "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
- "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
- "preceu.ph.qbr %[scratch2], %[load3] \n\t"
- "preceu.ph.qbr %[p2], %[load4] \n\t"
- "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
- "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
-
- "dpa.w.ph $ac0, %[p1], %[vector3b] \n\t"
- "dpa.w.ph $ac0, %[p2], %[vector4b] \n\t"
- "extp %[Temp1], $ac0, 31 \n\t"
- "dpa.w.ph $ac1, %[n1], %[vector3b] \n\t"
- "dpa.w.ph $ac1, %[n2], %[vector4b] \n\t"
- "extp %[Temp2], $ac1, 31 \n\t"
-
- "preceu.ph.qbl %[scratch1], %[load1] \n\t"
- "preceu.ph.qbl %[p1], %[load2] \n\t"
- "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
- "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
- "lbu %[scratch1], 0(%[dst_ptr]) \n\t"
- "preceu.ph.qbl %[scratch2], %[load3] \n\t"
- "preceu.ph.qbl %[p2], %[load4] \n\t"
- "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
- "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
- "lbu %[scratch2], 1(%[dst_ptr]) \n\t"
-
- "lbux %[store1], %[Temp1](%[cm]) \n\t"
- "dpa.w.ph $ac2, %[p1], %[vector3b] \n\t"
- "dpa.w.ph $ac2, %[p2], %[vector4b] \n\t"
- "addqh_r.w %[store1], %[store1], %[scratch1] \n\t" /* pixel 1 */
- "extp %[Temp1], $ac2, 31 \n\t"
-
- "lbux %[store2], %[Temp2](%[cm]) \n\t"
- "dpa.w.ph $ac3, %[n1], %[vector3b] \n\t"
- "dpa.w.ph $ac3, %[n2], %[vector4b] \n\t"
- "addqh_r.w %[store2], %[store2], %[scratch2] \n\t" /* pixel 2 */
- "extp %[Temp2], $ac3, 31 \n\t"
- "lbu %[scratch1], 2(%[dst_ptr]) \n\t"
-
- "sb %[store1], 0(%[dst_ptr]) \n\t"
- "sb %[store2], 1(%[dst_ptr]) \n\t"
- "lbu %[scratch2], 3(%[dst_ptr]) \n\t"
-
- "lbux %[store1], %[Temp1](%[cm]) \n\t"
- "lbux %[store2], %[Temp2](%[cm]) \n\t"
- "addqh_r.w %[store1], %[store1], %[scratch1] \n\t" /* pixel 3 */
- "addqh_r.w %[store2], %[store2], %[scratch2] \n\t" /* pixel 4 */
-
- "sb %[store1], 2(%[dst_ptr]) \n\t"
- "sb %[store2], 3(%[dst_ptr]) \n\t"
-
- : [load1] "=&r"(load1), [load2] "=&r"(load2), [load3] "=&r"(load3),
- [load4] "=&r"(load4), [p1] "=&r"(p1), [p2] "=&r"(p2),
- [n1] "=&r"(n1), [n2] "=&r"(n2), [scratch1] "=&r"(scratch1),
- [scratch2] "=&r"(scratch2), [Temp1] "=&r"(Temp1),
- [Temp2] "=&r"(Temp2), [store1] "=&r"(store1),
- [store2] "=&r"(store2), [src_ptr] "+r"(src_ptr)
- : [vector1b] "r"(vector1b), [vector2b] "r"(vector2b),
- [vector3b] "r"(vector3b), [vector4b] "r"(vector4b),
- [vector4a] "r"(vector4a), [src_stride] "r"(src_stride),
- [cm] "r"(cm), [dst_ptr] "r"(dst_ptr));
- }
-
- /* Next row... */
- src += src_stride;
- dst += dst_stride;
- }
-}
-
-void aom_convolve8_avg_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4, int w,
- int h) {
- assert(y_step_q4 == 16);
- assert(((const int32_t *)filter_y)[1] != 0x800000);
-
- if (((const int32_t *)filter_y)[0] == 0) {
- aom_convolve2_avg_vert_dspr2(src, src_stride, dst, dst_stride, filter_x,
- x_step_q4, filter_y, y_step_q4, w, h);
- } else {
- uint32_t pos = 38;
-
- /* bit positon for extract from acc */
- __asm__ __volatile__("wrdsp %[pos], 1 \n\t"
- :
- : [pos] "r"(pos));
-
- prefetch_store(dst);
-
- switch (w) {
- case 4:
- case 8:
- case 16:
- case 32:
- convolve_avg_vert_4_dspr2(src, src_stride, dst, dst_stride, filter_y, w,
- h);
- break;
- case 64:
- prefetch_store(dst + 32);
- convolve_avg_vert_64_dspr2(src, src_stride, dst, dst_stride, filter_y,
- h);
- break;
- default:
- aom_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, filter_x,
- x_step_q4, filter_y, y_step_q4, w, h);
- break;
- }
- }
-}
-
-void aom_convolve8_avg_dspr2(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4, int w,
- int h) {
- /* Fixed size intermediate buffer places limits on parameters. */
- DECLARE_ALIGNED(32, uint8_t, temp[64 * 135]);
- int32_t intermediate_height = ((h * y_step_q4) >> 4) + 7;
-
- assert(w <= 64);
- assert(h <= 64);
- assert(x_step_q4 == 16);
- assert(y_step_q4 == 16);
-
- if (intermediate_height < h) intermediate_height = h;
-
- aom_convolve8_horiz(src - (src_stride * 3), src_stride, temp, 64, filter_x,
- x_step_q4, filter_y, y_step_q4, w, intermediate_height);
-
- aom_convolve8_avg_vert(temp + 64 * 3, 64, dst, dst_stride, filter_x,
- x_step_q4, filter_y, y_step_q4, w, h);
-}
-
-void aom_convolve_avg_dspr2(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int filter_x_stride,
- const int16_t *filter_y, int filter_y_stride, int w,
- int h) {
- int x, y;
- uint32_t tp1, tp2, tn1;
- uint32_t tp3, tp4, tn2;
-
- (void)filter_x;
- (void)filter_x_stride;
- (void)filter_y;
- (void)filter_y_stride;
-
- /* prefetch data to cache memory */
- prefetch_load(src);
- prefetch_load(src + 32);
- prefetch_store(dst);
-
- switch (w) {
- case 4:
- /* 1 word storage */
- for (y = h; y--;) {
- prefetch_load(src + src_stride);
- prefetch_load(src + src_stride + 32);
- prefetch_store(dst + dst_stride);
-
- __asm__ __volatile__(
- "ulw %[tp1], 0(%[src]) \n\t"
- "ulw %[tp2], 0(%[dst]) \n\t"
- "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
- "sw %[tn1], 0(%[dst]) \n\t" /* store */
-
- : [tn1] "=&r"(tn1), [tp1] "=&r"(tp1), [tp2] "=&r"(tp2)
- : [src] "r"(src), [dst] "r"(dst));
-
- src += src_stride;
- dst += dst_stride;
- }
- break;
- case 8:
- /* 2 word storage */
- for (y = h; y--;) {
- prefetch_load(src + src_stride);
- prefetch_load(src + src_stride + 32);
- prefetch_store(dst + dst_stride);
-
- __asm__ __volatile__(
- "ulw %[tp1], 0(%[src]) \n\t"
- "ulw %[tp2], 0(%[dst]) \n\t"
- "ulw %[tp3], 4(%[src]) \n\t"
- "ulw %[tp4], 4(%[dst]) \n\t"
- "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
- "sw %[tn1], 0(%[dst]) \n\t" /* store */
- "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
- "sw %[tn2], 4(%[dst]) \n\t" /* store */
-
- : [tp1] "=&r"(tp1), [tp2] "=&r"(tp2), [tp3] "=&r"(tp3),
- [tp4] "=&r"(tp4), [tn1] "=&r"(tn1), [tn2] "=&r"(tn2)
- : [src] "r"(src), [dst] "r"(dst));
-
- src += src_stride;
- dst += dst_stride;
- }
- break;
- case 16:
- /* 4 word storage */
- for (y = h; y--;) {
- prefetch_load(src + src_stride);
- prefetch_load(src + src_stride + 32);
- prefetch_store(dst + dst_stride);
-
- __asm__ __volatile__(
- "ulw %[tp1], 0(%[src]) \n\t"
- "ulw %[tp2], 0(%[dst]) \n\t"
- "ulw %[tp3], 4(%[src]) \n\t"
- "ulw %[tp4], 4(%[dst]) \n\t"
- "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
- "ulw %[tp1], 8(%[src]) \n\t"
- "ulw %[tp2], 8(%[dst]) \n\t"
- "sw %[tn1], 0(%[dst]) \n\t" /* store */
- "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
- "sw %[tn2], 4(%[dst]) \n\t" /* store */
- "ulw %[tp3], 12(%[src]) \n\t"
- "ulw %[tp4], 12(%[dst]) \n\t"
- "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
- "sw %[tn1], 8(%[dst]) \n\t" /* store */
- "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
- "sw %[tn2], 12(%[dst]) \n\t" /* store */
-
- : [tp1] "=&r"(tp1), [tp2] "=&r"(tp2), [tp3] "=&r"(tp3),
- [tp4] "=&r"(tp4), [tn1] "=&r"(tn1), [tn2] "=&r"(tn2)
- : [src] "r"(src), [dst] "r"(dst));
-
- src += src_stride;
- dst += dst_stride;
- }
- break;
- case 32:
- /* 8 word storage */
- for (y = h; y--;) {
- prefetch_load(src + src_stride);
- prefetch_load(src + src_stride + 32);
- prefetch_store(dst + dst_stride);
-
- __asm__ __volatile__(
- "ulw %[tp1], 0(%[src]) \n\t"
- "ulw %[tp2], 0(%[dst]) \n\t"
- "ulw %[tp3], 4(%[src]) \n\t"
- "ulw %[tp4], 4(%[dst]) \n\t"
- "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
- "ulw %[tp1], 8(%[src]) \n\t"
- "ulw %[tp2], 8(%[dst]) \n\t"
- "sw %[tn1], 0(%[dst]) \n\t" /* store */
- "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
- "sw %[tn2], 4(%[dst]) \n\t" /* store */
- "ulw %[tp3], 12(%[src]) \n\t"
- "ulw %[tp4], 12(%[dst]) \n\t"
- "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
- "ulw %[tp1], 16(%[src]) \n\t"
- "ulw %[tp2], 16(%[dst]) \n\t"
- "sw %[tn1], 8(%[dst]) \n\t" /* store */
- "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
- "sw %[tn2], 12(%[dst]) \n\t" /* store */
- "ulw %[tp3], 20(%[src]) \n\t"
- "ulw %[tp4], 20(%[dst]) \n\t"
- "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
- "ulw %[tp1], 24(%[src]) \n\t"
- "ulw %[tp2], 24(%[dst]) \n\t"
- "sw %[tn1], 16(%[dst]) \n\t" /* store */
- "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
- "sw %[tn2], 20(%[dst]) \n\t" /* store */
- "ulw %[tp3], 28(%[src]) \n\t"
- "ulw %[tp4], 28(%[dst]) \n\t"
- "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
- "sw %[tn1], 24(%[dst]) \n\t" /* store */
- "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
- "sw %[tn2], 28(%[dst]) \n\t" /* store */
-
- : [tp1] "=&r"(tp1), [tp2] "=&r"(tp2), [tp3] "=&r"(tp3),
- [tp4] "=&r"(tp4), [tn1] "=&r"(tn1), [tn2] "=&r"(tn2)
- : [src] "r"(src), [dst] "r"(dst));
-
- src += src_stride;
- dst += dst_stride;
- }
- break;
- case 64:
- prefetch_load(src + 64);
- prefetch_store(dst + 32);
-
- /* 16 word storage */
- for (y = h; y--;) {
- prefetch_load(src + src_stride);
- prefetch_load(src + src_stride + 32);
- prefetch_load(src + src_stride + 64);
- prefetch_store(dst + dst_stride);
- prefetch_store(dst + dst_stride + 32);
-
- __asm__ __volatile__(
- "ulw %[tp1], 0(%[src]) \n\t"
- "ulw %[tp2], 0(%[dst]) \n\t"
- "ulw %[tp3], 4(%[src]) \n\t"
- "ulw %[tp4], 4(%[dst]) \n\t"
- "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
- "ulw %[tp1], 8(%[src]) \n\t"
- "ulw %[tp2], 8(%[dst]) \n\t"
- "sw %[tn1], 0(%[dst]) \n\t" /* store */
- "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
- "sw %[tn2], 4(%[dst]) \n\t" /* store */
- "ulw %[tp3], 12(%[src]) \n\t"
- "ulw %[tp4], 12(%[dst]) \n\t"
- "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
- "ulw %[tp1], 16(%[src]) \n\t"
- "ulw %[tp2], 16(%[dst]) \n\t"
- "sw %[tn1], 8(%[dst]) \n\t" /* store */
- "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
- "sw %[tn2], 12(%[dst]) \n\t" /* store */
- "ulw %[tp3], 20(%[src]) \n\t"
- "ulw %[tp4], 20(%[dst]) \n\t"
- "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
- "ulw %[tp1], 24(%[src]) \n\t"
- "ulw %[tp2], 24(%[dst]) \n\t"
- "sw %[tn1], 16(%[dst]) \n\t" /* store */
- "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
- "sw %[tn2], 20(%[dst]) \n\t" /* store */
- "ulw %[tp3], 28(%[src]) \n\t"
- "ulw %[tp4], 28(%[dst]) \n\t"
- "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
- "ulw %[tp1], 32(%[src]) \n\t"
- "ulw %[tp2], 32(%[dst]) \n\t"
- "sw %[tn1], 24(%[dst]) \n\t" /* store */
- "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
- "sw %[tn2], 28(%[dst]) \n\t" /* store */
- "ulw %[tp3], 36(%[src]) \n\t"
- "ulw %[tp4], 36(%[dst]) \n\t"
- "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
- "ulw %[tp1], 40(%[src]) \n\t"
- "ulw %[tp2], 40(%[dst]) \n\t"
- "sw %[tn1], 32(%[dst]) \n\t" /* store */
- "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
- "sw %[tn2], 36(%[dst]) \n\t" /* store */
- "ulw %[tp3], 44(%[src]) \n\t"
- "ulw %[tp4], 44(%[dst]) \n\t"
- "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
- "ulw %[tp1], 48(%[src]) \n\t"
- "ulw %[tp2], 48(%[dst]) \n\t"
- "sw %[tn1], 40(%[dst]) \n\t" /* store */
- "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
- "sw %[tn2], 44(%[dst]) \n\t" /* store */
- "ulw %[tp3], 52(%[src]) \n\t"
- "ulw %[tp4], 52(%[dst]) \n\t"
- "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
- "ulw %[tp1], 56(%[src]) \n\t"
- "ulw %[tp2], 56(%[dst]) \n\t"
- "sw %[tn1], 48(%[dst]) \n\t" /* store */
- "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
- "sw %[tn2], 52(%[dst]) \n\t" /* store */
- "ulw %[tp3], 60(%[src]) \n\t"
- "ulw %[tp4], 60(%[dst]) \n\t"
- "adduh_r.qb %[tn1], %[tp2], %[tp1] \n\t" /* average */
- "sw %[tn1], 56(%[dst]) \n\t" /* store */
- "adduh_r.qb %[tn2], %[tp3], %[tp4] \n\t" /* average */
- "sw %[tn2], 60(%[dst]) \n\t" /* store */
-
- : [tp1] "=&r"(tp1), [tp2] "=&r"(tp2), [tp3] "=&r"(tp3),
- [tp4] "=&r"(tp4), [tn1] "=&r"(tn1), [tn2] "=&r"(tn2)
- : [src] "r"(src), [dst] "r"(dst));
-
- src += src_stride;
- dst += dst_stride;
- }
- break;
- default:
- for (y = h; y > 0; --y) {
- for (x = 0; x < w; ++x) {
- dst[x] = (dst[x] + src[x] + 1) >> 1;
- }
-
- src += src_stride;
- dst += dst_stride;
- }
- break;
- }
-}
-#endif
diff --git a/aom_dsp/mips/convolve8_avg_horiz_dspr2.c b/aom_dsp/mips/convolve8_avg_horiz_dspr2.c
deleted file mode 100644
index f6534b4..0000000
--- a/aom_dsp/mips/convolve8_avg_horiz_dspr2.c
+++ /dev/null
@@ -1,998 +0,0 @@
-/*
- * Copyright (c) 2016, Alliance for Open Media. All rights reserved
- *
- * This source code is subject to the terms of the BSD 2 Clause License and
- * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
- * was not distributed with this source code in the LICENSE file, you can
- * obtain it at www.aomedia.org/license/software. If the Alliance for Open
- * Media Patent License 1.0 was not distributed with this source code in the
- * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
- */
-
-#include <assert.h>
-#include <stdio.h>
-
-#include "./aom_dsp_rtcd.h"
-#include "aom_dsp/mips/convolve_common_dspr2.h"
-#include "aom_dsp/aom_convolve.h"
-#include "aom_dsp/aom_dsp_common.h"
-#include "aom_ports/mem.h"
-
-#if HAVE_DSPR2
-static void convolve_avg_horiz_4_dspr2(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- const int16_t *filter_x0, int32_t h) {
- int32_t y;
- uint8_t *cm = aom_ff_cropTbl;
- int32_t vector1b, vector2b, vector3b, vector4b;
- int32_t Temp1, Temp2, Temp3, Temp4;
- uint32_t vector4a = 64;
- uint32_t tp1, tp2;
- uint32_t p1, p2, p3, p4;
- uint32_t n1, n2, n3, n4;
- uint32_t tn1, tn2;
-
- vector1b = ((const int32_t *)filter_x0)[0];
- vector2b = ((const int32_t *)filter_x0)[1];
- vector3b = ((const int32_t *)filter_x0)[2];
- vector4b = ((const int32_t *)filter_x0)[3];
-
- for (y = h; y--;) {
- /* prefetch data to cache memory */
- prefetch_load(src + src_stride);
- prefetch_load(src + src_stride + 32);
- prefetch_store(dst + dst_stride);
-
- __asm__ __volatile__(
- "ulw %[tp1], 0(%[src]) \n\t"
- "ulw %[tp2], 4(%[src]) \n\t"
-
- /* even 1. pixel */
- "mtlo %[vector4a], $ac3 \n\t"
- "mthi $zero, $ac3 \n\t"
- "preceu.ph.qbr %[p1], %[tp1] \n\t"
- "preceu.ph.qbl %[p2], %[tp1] \n\t"
- "preceu.ph.qbr %[p3], %[tp2] \n\t"
- "preceu.ph.qbl %[p4], %[tp2] \n\t"
- "dpa.w.ph $ac3, %[p1], %[vector1b] \n\t"
- "dpa.w.ph $ac3, %[p2], %[vector2b] \n\t"
- "dpa.w.ph $ac3, %[p3], %[vector3b] \n\t"
- "ulw %[tn2], 8(%[src]) \n\t"
- "dpa.w.ph $ac3, %[p4], %[vector4b] \n\t"
- "extp %[Temp1], $ac3, 31 \n\t"
-
- /* even 2. pixel */
- "mtlo %[vector4a], $ac2 \n\t"
- "mthi $zero, $ac2 \n\t"
- "preceu.ph.qbr %[p1], %[tn2] \n\t"
- "balign %[tn1], %[tn2], 3 \n\t"
- "balign %[tn2], %[tp2], 3 \n\t"
- "balign %[tp2], %[tp1], 3 \n\t"
- "dpa.w.ph $ac2, %[p2], %[vector1b] \n\t"
- "dpa.w.ph $ac2, %[p3], %[vector2b] \n\t"
- "dpa.w.ph $ac2, %[p4], %[vector3b] \n\t"
- "dpa.w.ph $ac2, %[p1], %[vector4b] \n\t"
- "extp %[Temp3], $ac2, 31 \n\t"
-
- "lbu %[p2], 3(%[dst]) \n\t" /* load odd 2 */
-
- /* odd 1. pixel */
- "lbux %[tp1], %[Temp1](%[cm]) \n\t" /* even 1 */
- "mtlo %[vector4a], $ac3 \n\t"
- "mthi $zero, $ac3 \n\t"
- "lbu %[Temp1], 1(%[dst]) \n\t" /* load odd 1 */
- "preceu.ph.qbr %[n1], %[tp2] \n\t"
- "preceu.ph.qbl %[n2], %[tp2] \n\t"
- "preceu.ph.qbr %[n3], %[tn2] \n\t"
- "preceu.ph.qbl %[n4], %[tn2] \n\t"
- "dpa.w.ph $ac3, %[n1], %[vector1b] \n\t"
- "dpa.w.ph $ac3, %[n2], %[vector2b] \n\t"
- "dpa.w.ph $ac3, %[n3], %[vector3b] \n\t"
- "dpa.w.ph $ac3, %[n4], %[vector4b] \n\t"
- "extp %[Temp2], $ac3, 31 \n\t"
-
- "lbu %[tn2], 0(%[dst]) \n\t" /* load even 1 */
-
- /* odd 2. pixel */
- "lbux %[tp2], %[Temp3](%[cm]) \n\t" /* even 2 */
- "mtlo %[vector4a], $ac2 \n\t"
- "mthi $zero, $ac2 \n\t"
- "preceu.ph.qbr %[n1], %[tn1] \n\t"
- "lbux %[tn1], %[Temp2](%[cm]) \n\t" /* odd 1 */
- "addqh_r.w %[tn2], %[tn2], %[tp1] \n\t" /* average even 1 */
- "dpa.w.ph $ac2, %[n2], %[vector1b] \n\t"
- "dpa.w.ph $ac2, %[n3], %[vector2b] \n\t"
- "dpa.w.ph $ac2, %[n4], %[vector3b] \n\t"
- "dpa.w.ph $ac2, %[n1], %[vector4b] \n\t"
- "extp %[Temp4], $ac2, 31 \n\t"
-
- "lbu %[tp1], 2(%[dst]) \n\t" /* load even 2 */
- "sb %[tn2], 0(%[dst]) \n\t" /* store even 1 */
-
- /* clamp */
- "addqh_r.w %[Temp1], %[Temp1], %[tn1] \n\t" /* average odd 1 */
- "lbux %[n2], %[Temp4](%[cm]) \n\t" /* odd 2 */
- "sb %[Temp1], 1(%[dst]) \n\t" /* store odd 1 */
-
- "addqh_r.w %[tp1], %[tp1], %[tp2] \n\t" /* average even 2 */
- "sb %[tp1], 2(%[dst]) \n\t" /* store even 2 */
-
- "addqh_r.w %[p2], %[p2], %[n2] \n\t" /* average odd 2 */
- "sb %[p2], 3(%[dst]) \n\t" /* store odd 2 */
-
- : [tp1] "=&r"(tp1), [tp2] "=&r"(tp2), [tn1] "=&r"(tn1),
- [tn2] "=&r"(tn2), [p1] "=&r"(p1), [p2] "=&r"(p2), [p3] "=&r"(p3),
- [p4] "=&r"(p4), [n1] "=&r"(n1), [n2] "=&r"(n2), [n3] "=&r"(n3),
- [n4] "=&r"(n4), [Temp1] "=&r"(Temp1), [Temp2] "=&r"(Temp2),
- [Temp3] "=&r"(Temp3), [Temp4] "=&r"(Temp4)
- : [vector1b] "r"(vector1b), [vector2b] "r"(vector2b),
- [vector3b] "r"(vector3b), [vector4b] "r"(vector4b),
- [vector4a] "r"(vector4a), [cm] "r"(cm), [dst] "r"(dst),
- [src] "r"(src));
-
- /* Next row... */
- src += src_stride;
- dst += dst_stride;
- }
-}
-
-static void convolve_avg_horiz_8_dspr2(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- const int16_t *filter_x0, int32_t h) {
- int32_t y;
- uint8_t *cm = aom_ff_cropTbl;
- uint32_t vector4a = 64;
- int32_t vector1b, vector2b, vector3b, vector4b;
- int32_t Temp1, Temp2, Temp3;
- uint32_t tp1, tp2;
- uint32_t p1, p2, p3, p4, n1;
- uint32_t tn1, tn2, tn3;
- uint32_t st0, st1;
-
- vector1b = ((const int32_t *)filter_x0)[0];
- vector2b = ((const int32_t *)filter_x0)[1];
- vector3b = ((const int32_t *)filter_x0)[2];
- vector4b = ((const int32_t *)filter_x0)[3];
-
- for (y = h; y--;) {
- /* prefetch data to cache memory */
- prefetch_load(src + src_stride);
- prefetch_load(src + src_stride + 32);
- prefetch_store(dst + dst_stride);
-
- __asm__ __volatile__(
- "ulw %[tp1], 0(%[src]) \n\t"
- "ulw %[tp2], 4(%[src]) \n\t"
-
- /* even 1. pixel */
- "mtlo %[vector4a], $ac3 \n\t"
- "mthi $zero, $ac3 \n\t"
- "mtlo %[vector4a], $ac2 \n\t"
- "mthi $zero, $ac2 \n\t"
- "preceu.ph.qbr %[p1], %[tp1] \n\t"
- "preceu.ph.qbl %[p2], %[tp1] \n\t"
- "preceu.ph.qbr %[p3], %[tp2] \n\t"
- "preceu.ph.qbl %[p4], %[tp2] \n\t"
- "ulw %[tn2], 8(%[src]) \n\t"
- "dpa.w.ph $ac3, %[p1], %[vector1b] \n\t"
- "dpa.w.ph $ac3, %[p2], %[vector2b] \n\t"
- "dpa.w.ph $ac3, %[p3], %[vector3b] \n\t"
- "dpa.w.ph $ac3, %[p4], %[vector4b] \n\t"
- "extp %[Temp1], $ac3, 31 \n\t"
- "lbu %[Temp2], 0(%[dst]) \n\t"
- "lbu %[tn3], 2(%[dst]) \n\t"
-
- /* even 2. pixel */
- "preceu.ph.qbr %[p1], %[tn2] \n\t"
- "preceu.ph.qbl %[n1], %[tn2] \n\t"
- "ulw %[tn1], 12(%[src]) \n\t"
- "dpa.w.ph $ac2, %[p2], %[vector1b] \n\t"
- "dpa.w.ph $ac2, %[p3], %[vector2b] \n\t"
- "dpa.w.ph $ac2, %[p4], %[vector3b] \n\t"
- "dpa.w.ph $ac2, %[p1], %[vector4b] \n\t"
- "extp %[Temp3], $ac2, 31 \n\t"
-
- /* even 3. pixel */
- "lbux %[st0], %[Temp1](%[cm]) \n\t"
- "mtlo %[vector4a], $ac1 \n\t"
- "mthi $zero, $ac1 \n\t"
- "preceu.ph.qbr %[p2], %[tn1] \n\t"
- "lbux %[st1], %[Temp3](%[cm]) \n\t"
- "dpa.w.ph $ac1, %[p3], %[vector1b] \n\t"
- "dpa.w.ph $ac1, %[p4], %[vector2b] \n\t"
- "dpa.w.ph $ac1, %[p1], %[vector3b] \n\t"
- "dpa.w.ph $ac1, %[n1], %[vector4b] \n\t"
- "extp %[Temp1], $ac1, 31 \n\t"
-
- "addqh_r.w %[Temp2], %[Temp2], %[st0] \n\t"
- "addqh_r.w %[tn3], %[tn3], %[st1] \n\t"
- "sb %[Temp2], 0(%[dst]) \n\t"
- "sb %[tn3], 2(%[dst]) \n\t"
-
- /* even 4. pixel */
- "mtlo %[vector4a], $ac2 \n\t"
- "mthi $zero, $ac2 \n\t"
- "mtlo %[vector4a], $ac3 \n\t"
- "mthi $zero, $ac3 \n\t"
-
- "balign %[tn3], %[tn1], 3 \n\t"
- "balign %[tn1], %[tn2], 3 \n\t"
- "balign %[tn2], %[tp2], 3 \n\t"
- "balign %[tp2], %[tp1], 3 \n\t"
-
- "lbux %[st0], %[Temp1](%[cm]) \n\t"
- "lbu %[Temp2], 4(%[dst]) \n\t"
- "addqh_r.w %[Temp2], %[Temp2], %[st0] \n\t"
-
- "dpa.w.ph $ac2, %[p4], %[vector1b] \n\t"
- "dpa.w.ph $ac2, %[p1], %[vector2b] \n\t"
- "dpa.w.ph $ac2, %[n1], %[vector3b] \n\t"
- "dpa.w.ph $ac2, %[p2], %[vector4b] \n\t"
- "extp %[Temp3], $ac2, 31 \n\t"
-
- /* odd 1. pixel */
- "mtlo %[vector4a], $ac1 \n\t"
- "mthi $zero, $ac1 \n\t"
- "sb %[Temp2], 4(%[dst]) \n\t"
- "preceu.ph.qbr %[p1], %[tp2] \n\t"
- "preceu.ph.qbl %[p2], %[tp2] \n\t"
- "preceu.ph.qbr %[p3], %[tn2] \n\t"
- "preceu.ph.qbl %[p4], %[tn2] \n\t"
- "dpa.w.ph $ac3, %[p1], %[vector1b] \n\t"
- "dpa.w.ph $ac3, %[p2], %[vector2b] \n\t"
- "dpa.w.ph $ac3, %[p3], %[vector3b] \n\t"
- "dpa.w.ph $ac3, %[p4], %[vector4b] \n\t"
- "extp %[Temp2], $ac3, 31 \n\t"
-
- "lbu %[tp1], 6(%[dst]) \n\t"
-
- /* odd 2. pixel */
- "mtlo %[vector4a], $ac3 \n\t"
- "mthi $zero, $ac3 \n\t"
- "mtlo %[vector4a], $ac2 \n\t"
- "mthi $zero, $ac2 \n\t"
- "preceu.ph.qbr %[p1], %[tn1] \n\t"
- "preceu.ph.qbl %[n1], %[tn1] \n\t"
- "lbux %[st0], %[Temp3](%[cm]) \n\t"
- "dpa.w.ph $ac1, %[p2], %[vector1b] \n\t"
- "dpa.w.ph $ac1, %[p3], %[vector2b] \n\t"
- "dpa.w.ph $ac1, %[p4], %[vector3b] \n\t"
- "dpa.w.ph $ac1, %[p1], %[vector4b] \n\t"
- "extp %[Temp3], $ac1, 31 \n\t"
-
- "lbu %[tp2], 1(%[dst]) \n\t"
- "lbu %[tn2], 3(%[dst]) \n\t"
- "addqh_r.w %[tp1], %[tp1], %[st0] \n\t"
-
- /* odd 3. pixel */
- "lbux %[st1], %[Temp2](%[cm]) \n\t"
- "preceu.ph.qbr %[p2], %[tn3] \n\t"
- "dpa.w.ph $ac3, %[p3], %[vector1b] \n\t"
- "dpa.w.ph $ac3, %[p4], %[vector2b] \n\t"
- "dpa.w.ph $ac3, %[p1], %[vector3b] \n\t"
- "dpa.w.ph $ac3, %[n1], %[vector4b] \n\t"
- "addqh_r.w %[tp2], %[tp2], %[st1] \n\t"
- "extp %[Temp2], $ac3, 31 \n\t"
-
- "lbu %[tn3], 5(%[dst]) \n\t"
-
- /* odd 4. pixel */
- "sb %[tp2], 1(%[dst]) \n\t"
- "sb %[tp1], 6(%[dst]) \n\t"
- "dpa.w.ph $ac2, %[p4], %[vector1b] \n\t"
- "dpa.w.ph $ac2, %[p1], %[vector2b] \n\t"
- "dpa.w.ph $ac2, %[n1], %[vector3b] \n\t"
- "dpa.w.ph $ac2, %[p2], %[vector4b] \n\t"
- "extp %[Temp1], $ac2, 31 \n\t"
-
- "lbu %[tn1], 7(%[dst]) \n\t"
-
- /* clamp */
- "lbux %[p4], %[Temp3](%[cm]) \n\t"
- "addqh_r.w %[tn2], %[tn2], %[p4] \n\t"
-
- "lbux %[p2], %[Temp2](%[cm]) \n\t"
- "addqh_r.w %[tn3], %[tn3], %[p2] \n\t"
-
- "lbux %[n1], %[Temp1](%[cm]) \n\t"
- "addqh_r.w %[tn1], %[tn1], %[n1] \n\t"
-
- /* store bytes */
- "sb %[tn2], 3(%[dst]) \n\t"
- "sb %[tn3], 5(%[dst]) \n\t"
- "sb %[tn1], 7(%[dst]) \n\t"
-
- : [tp1] "=&r"(tp1), [tp2] "=&r"(tp2), [tn1] "=&r"(tn1),
- [tn2] "=&r"(tn2), [tn3] "=&r"(tn3), [st0] "=&r"(st0),
- [st1] "=&r"(st1), [p1] "=&r"(p1), [p2] "=&r"(p2), [p3] "=&r"(p3),
- [p4] "=&r"(p4), [n1] "=&r"(n1), [Temp1] "=&r"(Temp1),
- [Temp2] "=&r"(Temp2), [Temp3] "=&r"(Temp3)
- : [vector1b] "r"(vector1b), [vector2b] "r"(vector2b),
- [vector3b] "r"(vector3b), [vector4b] "r"(vector4b),
- [vector4a] "r"(vector4a), [cm] "r"(cm), [dst] "r"(dst),
- [src] "r"(src));
-
- /* Next row... */
- src += src_stride;
- dst += dst_stride;
- }
-}
-
-static void convolve_avg_horiz_16_dspr2(const uint8_t *src_ptr,
- int32_t src_stride, uint8_t *dst_ptr,
- int32_t dst_stride,
- const int16_t *filter_x0, int32_t h,
- int32_t count) {
- int32_t y, c;
- const uint8_t *src;
- uint8_t *dst;
- uint8_t *cm = aom_ff_cropTbl;
- uint32_t vector_64 = 64;
- int32_t filter12, filter34, filter56, filter78;
- int32_t Temp1, Temp2, Temp3;
- uint32_t qload1, qload2, qload3;
- uint32_t p1, p2, p3, p4, p5;
- uint32_t st1, st2, st3;
-
- filter12 = ((const int32_t *)filter_x0)[0];
- filter34 = ((const int32_t *)filter_x0)[1];
- filter56 = ((const int32_t *)filter_x0)[2];
- filter78 = ((const int32_t *)filter_x0)[3];
-
- for (y = h; y--;) {
- src = src_ptr;
- dst = dst_ptr;
-
- /* prefetch data to cache memory */
- prefetch_load(src_ptr + src_stride);
- prefetch_load(src_ptr + src_stride + 32);
- prefetch_store(dst_ptr + dst_stride);
-
- for (c = 0; c < count; c++) {
- __asm__ __volatile__(
- "ulw %[qload1], 0(%[src]) \n\t"
- "ulw %[qload2], 4(%[src]) \n\t"
-
- /* even 1. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* even 1 */
- "mthi $zero, $ac1 \n\t"
- "mtlo %[vector_64], $ac2 \n\t" /* even 2 */
- "mthi $zero, $ac2 \n\t"
- "preceu.ph.qbr %[p1], %[qload1] \n\t"
- "preceu.ph.qbl %[p2], %[qload1] \n\t"
- "preceu.ph.qbr %[p3], %[qload2] \n\t"
- "preceu.ph.qbl %[p4], %[qload2] \n\t"
- "ulw %[qload3], 8(%[src]) \n\t"
- "dpa.w.ph $ac1, %[p1], %[filter12] \n\t" /* even 1 */
- "dpa.w.ph $ac1, %[p2], %[filter34] \n\t" /* even 1 */
- "dpa.w.ph $ac1, %[p3], %[filter56] \n\t" /* even 1 */
- "dpa.w.ph $ac1, %[p4], %[filter78] \n\t" /* even 1 */
- "extp %[Temp1], $ac1, 31 \n\t" /* even 1 */
- "lbu %[st2], 0(%[dst]) \n\t" /* load even 1 from dst */
-
- /* even 2. pixel */
- "mtlo %[vector_64], $ac3 \n\t" /* even 3 */
- "mthi $zero, $ac3 \n\t"
- "preceu.ph.qbr %[p1], %[qload3] \n\t"
- "preceu.ph.qbl %[p5], %[qload3] \n\t"
- "ulw %[qload1], 12(%[src]) \n\t"
- "dpa.w.ph $ac2, %[p2], %[filter12] \n\t" /* even 1 */
- "dpa.w.ph $ac2, %[p3], %[filter34] \n\t" /* even 1 */
- "dpa.w.ph $ac2, %[p4], %[filter56] \n\t" /* even 1 */
- "dpa.w.ph $ac2, %[p1], %[filter78] \n\t" /* even 1 */
- "extp %[Temp2], $ac2, 31 \n\t" /* even 1 */
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 1 */
-
- "lbu %[qload3], 2(%[dst]) \n\t" /* load even 2 from dst */
-
- /* even 3. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* even 4 */
- "mthi $zero, $ac1 \n\t"
- "addqh_r.w %[st2], %[st2], %[st1] \n\t" /* average even 1 */
- "preceu.ph.qbr %[p2], %[qload1] \n\t"
- "sb %[st2], 0(%[dst]) \n\t" /* store even 1 to dst */
- "dpa.w.ph $ac3, %[p3], %[filter12] \n\t" /* even 3 */
- "dpa.w.ph $ac3, %[p4], %[filter34] \n\t" /* even 3 */
- "dpa.w.ph $ac3, %[p1], %[filter56] \n\t" /* even 3 */
- "dpa.w.ph $ac3, %[p5], %[filter78] \n\t" /* even 3 */
- "extp %[Temp3], $ac3, 31 \n\t" /* even 3 */
- "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 1 */
-
- /* even 4. pixel */
- "mtlo %[vector_64], $ac2 \n\t" /* even 5 */
- "mthi $zero, $ac2 \n\t"
- "addqh_r.w %[qload3], %[qload3], %[st2] \n\t" /* average even 2 */
- "preceu.ph.qbl %[p3], %[qload1] \n\t"
- "sb %[qload3], 2(%[dst]) \n\t" /* store even 2 to dst */
- "ulw %[qload2], 16(%[src]) \n\t"
- "lbu %[qload3], 4(%[dst]) \n\t" /* load even 3 from dst */
- "lbu %[qload1], 6(%[dst]) \n\t" /* load even 4 from dst */
- "dpa.w.ph $ac1, %[p4], %[filter12] \n\t" /* even 4 */
- "dpa.w.ph $ac1, %[p1], %[filter34] \n\t" /* even 4 */
- "dpa.w.ph $ac1, %[p5], %[filter56] \n\t" /* even 4 */
- "dpa.w.ph $ac1, %[p2], %[filter78] \n\t" /* even 4 */
- "extp %[Temp1], $ac1, 31 \n\t" /* even 4 */
- "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 3 */
-
- /* even 5. pixel */
- "mtlo %[vector_64], $ac3 \n\t" /* even 6 */
- "mthi $zero, $ac3 \n\t"
- "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average even 3 */
- "preceu.ph.qbr %[p4], %[qload2] \n\t"
- "sb %[qload3], 4(%[dst]) \n\t" /* store even 3 to dst */
- "dpa.w.ph $ac2, %[p1], %[filter12] \n\t" /* even 5 */
- "dpa.w.ph $ac2, %[p5], %[filter34] \n\t" /* even 5 */
- "dpa.w.ph $ac2, %[p2], %[filter56] \n\t" /* even 5 */
- "dpa.w.ph $ac2, %[p3], %[filter78] \n\t" /* even 5 */
- "extp %[Temp2], $ac2, 31 \n\t" /* even 5 */
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 4 */
-
- /* even 6. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* even 7 */
- "mthi $zero, $ac1 \n\t"
- "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average even 4 */
- "preceu.ph.qbl %[p1], %[qload2] \n\t"
- "sb %[qload1], 6(%[dst]) \n\t" /* store even 4 to dst */
- "ulw %[qload3], 20(%[src]) \n\t"
- "dpa.w.ph $ac3, %[p5], %[filter12] \n\t" /* even 6 */
- "dpa.w.ph $ac3, %[p2], %[filter34] \n\t" /* even 6 */
- "dpa.w.ph $ac3, %[p3], %[filter56] \n\t" /* even 6 */
- "dpa.w.ph $ac3, %[p4], %[filter78] \n\t" /* even 6 */
- "lbu %[qload2], 8(%[dst]) \n\t" /* load even 5 from dst */
- "extp %[Temp3], $ac3, 31 \n\t" /* even 6 */
- "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 5 */
-
- /* even 7. pixel */
- "mtlo %[vector_64], $ac2 \n\t" /* even 8 */
- "mthi $zero, $ac2 \n\t"
- "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average even 5 */
- "preceu.ph.qbr %[p5], %[qload3] \n\t"
- "sb %[qload2], 8(%[dst]) \n\t" /* store even 5 to dst */
- "dpa.w.ph $ac1, %[p2], %[filter12] \n\t" /* even 7 */
- "dpa.w.ph $ac1, %[p3], %[filter34] \n\t" /* even 7 */
- "dpa.w.ph $ac1, %[p4], %[filter56] \n\t" /* even 7 */
- "dpa.w.ph $ac1, %[p1], %[filter78] \n\t" /* even 7 */
- "lbu %[qload3], 10(%[dst]) \n\t" /* load even 6 from dst */
- "extp %[Temp1], $ac1, 31 \n\t" /* even 7 */
- "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 6 */
-
- "lbu %[st2], 12(%[dst]) \n\t" /* load even 7 from dst */
-
- /* even 8. pixel */
- "mtlo %[vector_64], $ac3 \n\t" /* odd 1 */
- "mthi $zero, $ac3 \n\t"
- "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average even 6 */
- "dpa.w.ph $ac2, %[p3], %[filter12] \n\t" /* even 8 */
- "dpa.w.ph $ac2, %[p4], %[filter34] \n\t" /* even 8 */
- "sb %[qload3], 10(%[dst]) \n\t" /* store even 6 to dst */
- "dpa.w.ph $ac2, %[p1], %[filter56] \n\t" /* even 8 */
- "dpa.w.ph $ac2, %[p5], %[filter78] \n\t" /* even 8 */
- "extp %[Temp2], $ac2, 31 \n\t" /* even 8 */
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 7 */
-
- /* ODD pixels */
- "ulw %[qload1], 1(%[src]) \n\t"
- "ulw %[qload2], 5(%[src]) \n\t"
-
- "addqh_r.w %[st2], %[st2], %[st1] \n\t" /* average even 7 */
-
- /* odd 1. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* odd 2 */
- "mthi $zero, $ac1 \n\t"
- "preceu.ph.qbr %[p1], %[qload1] \n\t"
- "preceu.ph.qbl %[p2], %[qload1] \n\t"
- "preceu.ph.qbr %[p3], %[qload2] \n\t"
- "preceu.ph.qbl %[p4], %[qload2] \n\t"
- "sb %[st2], 12(%[dst]) \n\t" /* store even 7 to dst */
- "ulw %[qload3], 9(%[src]) \n\t"
- "dpa.w.ph $ac3, %[p1], %[filter12] \n\t" /* odd 1 */
- "dpa.w.ph $ac3, %[p2], %[filter34] \n\t" /* odd 1 */
- "lbu %[qload2], 14(%[dst]) \n\t" /* load even 8 from dst */
- "dpa.w.ph $ac3, %[p3], %[filter56] \n\t" /* odd 1 */
- "dpa.w.ph $ac3, %[p4], %[filter78] \n\t" /* odd 1 */
- "extp %[Temp3], $ac3, 31 \n\t" /* odd 1 */
- "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 8 */
-
- "lbu %[st1], 1(%[dst]) \n\t" /* load odd 1 from dst */
-
- /* odd 2. pixel */
- "mtlo %[vector_64], $ac2 \n\t" /* odd 3 */
- "mthi $zero, $ac2 \n\t"
- "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average even 8 */
- "preceu.ph.qbr %[p1], %[qload3] \n\t"
- "preceu.ph.qbl %[p5], %[qload3] \n\t"
- "sb %[qload2], 14(%[dst]) \n\t" /* store even 8 to dst */
- "ulw %[qload1], 13(%[src]) \n\t"
- "dpa.w.ph $ac1, %[p2], %[filter12] \n\t" /* odd 2 */
- "dpa.w.ph $ac1, %[p3], %[filter34] \n\t" /* odd 2 */
- "dpa.w.ph $ac1, %[p4], %[filter56] \n\t" /* odd 2 */
- "dpa.w.ph $ac1, %[p1], %[filter78] \n\t" /* odd 2 */
- "lbu %[qload3], 3(%[dst]) \n\t" /* load odd 2 from dst */
- "extp %[Temp1], $ac1, 31 \n\t" /* odd 2 */
- "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 1 */
-
- /* odd 3. pixel */
- "mtlo %[vector_64], $ac3 \n\t" /* odd 4 */
- "mthi $zero, $ac3 \n\t"
- "addqh_r.w %[st3], %[st3], %[st1] \n\t" /* average odd 1 */
- "preceu.ph.qbr %[p2], %[qload1] \n\t"
- "dpa.w.ph $ac2, %[p3], %[filter12] \n\t" /* odd 3 */
- "dpa.w.ph $ac2, %[p4], %[filter34] \n\t" /* odd 3 */
- "dpa.w.ph $ac2, %[p1], %[filter56] \n\t" /* odd 3 */
- "dpa.w.ph $ac2, %[p5], %[filter78] \n\t" /* odd 3 */
- "sb %[st3], 1(%[dst]) \n\t" /* store odd 1 to dst */
- "extp %[Temp2], $ac2, 31 \n\t" /* odd 3 */
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 2 */
-
- /* odd 4. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* odd 5 */
- "mthi $zero, $ac1 \n\t"
- "addqh_r.w %[qload3], %[qload3], %[st1] \n\t" /* average odd 2 */
- "preceu.ph.qbl %[p3], %[qload1] \n\t"
- "sb %[qload3], 3(%[dst]) \n\t" /* store odd 2 to dst */
- "lbu %[qload1], 5(%[dst]) \n\t" /* load odd 3 from dst */
- "ulw %[qload2], 17(%[src]) \n\t"
- "dpa.w.ph $ac3, %[p4], %[filter12] \n\t" /* odd 4 */
- "dpa.w.ph $ac3, %[p1], %[filter34] \n\t" /* odd 4 */
- "dpa.w.ph $ac3, %[p5], %[filter56] \n\t" /* odd 4 */
- "dpa.w.ph $ac3, %[p2], %[filter78] \n\t" /* odd 4 */
- "extp %[Temp3], $ac3, 31 \n\t" /* odd 4 */
- "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 3 */
-
- "lbu %[st1], 7(%[dst]) \n\t" /* load odd 4 from dst */
-
- /* odd 5. pixel */
- "mtlo %[vector_64], $ac2 \n\t" /* odd 6 */
- "mthi $zero, $ac2 \n\t"
- "addqh_r.w %[qload1], %[qload1], %[st2] \n\t" /* average odd 3 */
- "preceu.ph.qbr %[p4], %[qload2] \n\t"
- "sb %[qload1], 5(%[dst]) \n\t" /* store odd 3 to dst */
- "dpa.w.ph $ac1, %[p1], %[filter12] \n\t" /* odd 5 */
- "dpa.w.ph $ac1, %[p5], %[filter34] \n\t" /* odd 5 */
- "dpa.w.ph $ac1, %[p2], %[filter56] \n\t" /* odd 5 */
- "dpa.w.ph $ac1, %[p3], %[filter78] \n\t" /* odd 5 */
- "extp %[Temp1], $ac1, 31 \n\t" /* odd 5 */
- "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 4 */
-
- "lbu %[qload1], 9(%[dst]) \n\t" /* load odd 5 from dst */
-
- /* odd 6. pixel */
- "mtlo %[vector_64], $ac3 \n\t" /* odd 7 */
- "mthi $zero, $ac3 \n\t"
- "addqh_r.w %[st1], %[st1], %[st3] \n\t" /* average odd 4 */
- "preceu.ph.qbl %[p1], %[qload2] \n\t"
- "sb %[st1], 7(%[dst]) \n\t" /* store odd 4 to dst */
- "ulw %[qload3], 21(%[src]) \n\t"
- "dpa.w.ph $ac2, %[p5], %[filter12] \n\t" /* odd 6 */
- "dpa.w.ph $ac2, %[p2], %[filter34] \n\t" /* odd 6 */
- "dpa.w.ph $ac2, %[p3], %[filter56] \n\t" /* odd 6 */
- "dpa.w.ph $ac2, %[p4], %[filter78] \n\t" /* odd 6 */
- "extp %[Temp2], $ac2, 31 \n\t" /* odd 6 */
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 5 */
-
- /* odd 7. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* odd 8 */
- "mthi $zero, $ac1 \n\t"
- "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average odd 5 */
- "preceu.ph.qbr %[p5], %[qload3] \n\t"
- "sb %[qload1], 9(%[dst]) \n\t" /* store odd 5 to dst */
- "lbu %[qload2], 11(%[dst]) \n\t" /* load odd 6 from dst */
- "dpa.w.ph $ac3, %[p2], %[filter12] \n\t" /* odd 7 */
- "dpa.w.ph $ac3, %[p3], %[filter34] \n\t" /* odd 7 */
- "dpa.w.ph $ac3, %[p4], %[filter56] \n\t" /* odd 7 */
- "dpa.w.ph $ac3, %[p1], %[filter78] \n\t" /* odd 7 */
- "extp %[Temp3], $ac3, 31 \n\t" /* odd 7 */
-
- "lbu %[qload3], 13(%[dst]) \n\t" /* load odd 7 from dst */
-
- /* odd 8. pixel */
- "dpa.w.ph $ac1, %[p3], %[filter12] \n\t" /* odd 8 */
- "dpa.w.ph $ac1, %[p4], %[filter34] \n\t" /* odd 8 */
- "dpa.w.ph $ac1, %[p1], %[filter56] \n\t" /* odd 8 */
- "dpa.w.ph $ac1, %[p5], %[filter78] \n\t" /* odd 8 */
- "extp %[Temp1], $ac1, 31 \n\t" /* odd 8 */
-
- "lbu %[qload1], 15(%[dst]) \n\t" /* load odd 8 from dst */
-
- "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 6 */
- "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average odd 6 */
-
- "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 7 */
- "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average odd 7 */
-
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 8 */
- "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average odd 8 */
-
- "sb %[qload2], 11(%[dst]) \n\t" /* store odd 6 to dst */
- "sb %[qload3], 13(%[dst]) \n\t" /* store odd 7 to dst */
- "sb %[qload1], 15(%[dst]) \n\t" /* store odd 8 to dst */
-
- : [qload1] "=&r"(qload1), [qload2] "=&r"(qload2), [st1] "=&r"(st1),
- [st2] "=&r"(st2), [st3] "=&r"(st3), [p1] "=&r"(p1), [p2] "=&r"(p2),
- [p3] "=&r"(p3), [p4] "=&r"(p4), [qload3] "=&r"(qload3),
- [p5] "=&r"(p5), [Temp1] "=&r"(Temp1), [Temp2] "=&r"(Temp2),
- [Temp3] "=&r"(Temp3)
- : [filter12] "r"(filter12), [filter34] "r"(filter34),
- [filter56] "r"(filter56), [filter78] "r"(filter78),
- [vector_64] "r"(vector_64), [cm] "r"(cm), [dst] "r"(dst),
- [src] "r"(src));
-
- src += 16;
- dst += 16;
- }
-
- /* Next row... */
- src_ptr += src_stride;
- dst_ptr += dst_stride;
- }
-}
-
-static void convolve_avg_horiz_64_dspr2(const uint8_t *src_ptr,
- int32_t src_stride, uint8_t *dst_ptr,
- int32_t dst_stride,
- const int16_t *filter_x0, int32_t h) {
- int32_t y, c;
- const uint8_t *src;
- uint8_t *dst;
- uint8_t *cm = aom_ff_cropTbl;
- uint32_t vector_64 = 64;
- int32_t filter12, filter34, filter56, filter78;
- int32_t Temp1, Temp2, Temp3;
- uint32_t qload1, qload2, qload3;
- uint32_t p1, p2, p3, p4, p5;
- uint32_t st1, st2, st3;
-
- filter12 = ((const int32_t *)filter_x0)[0];
- filter34 = ((const int32_t *)filter_x0)[1];
- filter56 = ((const int32_t *)filter_x0)[2];
- filter78 = ((const int32_t *)filter_x0)[3];
-
- for (y = h; y--;) {
- src = src_ptr;
- dst = dst_ptr;
-
- /* prefetch data to cache memory */
- prefetch_load(src_ptr + src_stride);
- prefetch_load(src_ptr + src_stride + 32);
- prefetch_load(src_ptr + src_stride + 64);
- prefetch_store(dst_ptr + dst_stride);
- prefetch_store(dst_ptr + dst_stride + 32);
-
- for (c = 0; c < 4; c++) {
- __asm__ __volatile__(
- "ulw %[qload1], 0(%[src]) \n\t"
- "ulw %[qload2], 4(%[src]) \n\t"
-
- /* even 1. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* even 1 */
- "mthi $zero, $ac1 \n\t"
- "mtlo %[vector_64], $ac2 \n\t" /* even 2 */
- "mthi $zero, $ac2 \n\t"
- "preceu.ph.qbr %[p1], %[qload1] \n\t"
- "preceu.ph.qbl %[p2], %[qload1] \n\t"
- "preceu.ph.qbr %[p3], %[qload2] \n\t"
- "preceu.ph.qbl %[p4], %[qload2] \n\t"
- "ulw %[qload3], 8(%[src]) \n\t"
- "dpa.w.ph $ac1, %[p1], %[filter12] \n\t" /* even 1 */
- "dpa.w.ph $ac1, %[p2], %[filter34] \n\t" /* even 1 */
- "dpa.w.ph $ac1, %[p3], %[filter56] \n\t" /* even 1 */
- "dpa.w.ph $ac1, %[p4], %[filter78] \n\t" /* even 1 */
- "extp %[Temp1], $ac1, 31 \n\t" /* even 1 */
- "lbu %[st2], 0(%[dst]) \n\t" /* load even 1 from dst */
-
- /* even 2. pixel */
- "mtlo %[vector_64], $ac3 \n\t" /* even 3 */
- "mthi $zero, $ac3 \n\t"
- "preceu.ph.qbr %[p1], %[qload3] \n\t"
- "preceu.ph.qbl %[p5], %[qload3] \n\t"
- "ulw %[qload1], 12(%[src]) \n\t"
- "dpa.w.ph $ac2, %[p2], %[filter12] \n\t" /* even 1 */
- "dpa.w.ph $ac2, %[p3], %[filter34] \n\t" /* even 1 */
- "dpa.w.ph $ac2, %[p4], %[filter56] \n\t" /* even 1 */
- "dpa.w.ph $ac2, %[p1], %[filter78] \n\t" /* even 1 */
- "extp %[Temp2], $ac2, 31 \n\t" /* even 1 */
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 1 */
-
- "lbu %[qload3], 2(%[dst]) \n\t" /* load even 2 from dst */
-
- /* even 3. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* even 4 */
- "mthi $zero, $ac1 \n\t"
- "addqh_r.w %[st2], %[st2], %[st1] \n\t" /* average even 1 */
- "preceu.ph.qbr %[p2], %[qload1] \n\t"
- "sb %[st2], 0(%[dst]) \n\t" /* store even 1 to dst */
- "dpa.w.ph $ac3, %[p3], %[filter12] \n\t" /* even 3 */
- "dpa.w.ph $ac3, %[p4], %[filter34] \n\t" /* even 3 */
- "dpa.w.ph $ac3, %[p1], %[filter56] \n\t" /* even 3 */
- "dpa.w.ph $ac3, %[p5], %[filter78] \n\t" /* even 3 */
- "extp %[Temp3], $ac3, 31 \n\t" /* even 3 */
- "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 1 */
-
- /* even 4. pixel */
- "mtlo %[vector_64], $ac2 \n\t" /* even 5 */
- "mthi $zero, $ac2 \n\t"
- "addqh_r.w %[qload3], %[qload3], %[st2] \n\t" /* average even 2 */
- "preceu.ph.qbl %[p3], %[qload1] \n\t"
- "sb %[qload3], 2(%[dst]) \n\t" /* store even 2 to dst */
- "ulw %[qload2], 16(%[src]) \n\t"
- "lbu %[qload3], 4(%[dst]) \n\t" /* load even 3 from dst */
- "lbu %[qload1], 6(%[dst]) \n\t" /* load even 4 from dst */
- "dpa.w.ph $ac1, %[p4], %[filter12] \n\t" /* even 4 */
- "dpa.w.ph $ac1, %[p1], %[filter34] \n\t" /* even 4 */
- "dpa.w.ph $ac1, %[p5], %[filter56] \n\t" /* even 4 */
- "dpa.w.ph $ac1, %[p2], %[filter78] \n\t" /* even 4 */
- "extp %[Temp1], $ac1, 31 \n\t" /* even 4 */
- "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 3 */
-
- /* even 5. pixel */
- "mtlo %[vector_64], $ac3 \n\t" /* even 6 */
- "mthi $zero, $ac3 \n\t"
- "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average even 3 */
- "preceu.ph.qbr %[p4], %[qload2] \n\t"
- "sb %[qload3], 4(%[dst]) \n\t" /* store even 3 to dst */
- "dpa.w.ph $ac2, %[p1], %[filter12] \n\t" /* even 5 */
- "dpa.w.ph $ac2, %[p5], %[filter34] \n\t" /* even 5 */
- "dpa.w.ph $ac2, %[p2], %[filter56] \n\t" /* even 5 */
- "dpa.w.ph $ac2, %[p3], %[filter78] \n\t" /* even 5 */
- "extp %[Temp2], $ac2, 31 \n\t" /* even 5 */
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 4 */
-
- /* even 6. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* even 7 */
- "mthi $zero, $ac1 \n\t"
- "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average even 4 */
- "preceu.ph.qbl %[p1], %[qload2] \n\t"
- "sb %[qload1], 6(%[dst]) \n\t" /* store even 4 to dst */
- "ulw %[qload3], 20(%[src]) \n\t"
- "dpa.w.ph $ac3, %[p5], %[filter12] \n\t" /* even 6 */
- "dpa.w.ph $ac3, %[p2], %[filter34] \n\t" /* even 6 */
- "dpa.w.ph $ac3, %[p3], %[filter56] \n\t" /* even 6 */
- "dpa.w.ph $ac3, %[p4], %[filter78] \n\t" /* even 6 */
- "lbu %[qload2], 8(%[dst]) \n\t" /* load even 5 from dst */
- "extp %[Temp3], $ac3, 31 \n\t" /* even 6 */
- "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 5 */
-
- /* even 7. pixel */
- "mtlo %[vector_64], $ac2 \n\t" /* even 8 */
- "mthi $zero, $ac2 \n\t"
- "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average even 5 */
- "preceu.ph.qbr %[p5], %[qload3] \n\t"
- "sb %[qload2], 8(%[dst]) \n\t" /* store even 5 to dst */
- "dpa.w.ph $ac1, %[p2], %[filter12] \n\t" /* even 7 */
- "dpa.w.ph $ac1, %[p3], %[filter34] \n\t" /* even 7 */
- "dpa.w.ph $ac1, %[p4], %[filter56] \n\t" /* even 7 */
- "dpa.w.ph $ac1, %[p1], %[filter78] \n\t" /* even 7 */
- "lbu %[qload3], 10(%[dst]) \n\t" /* load even 6 from dst */
- "extp %[Temp1], $ac1, 31 \n\t" /* even 7 */
- "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 6 */
-
- "lbu %[st2], 12(%[dst]) \n\t" /* load even 7 from dst */
-
- /* even 8. pixel */
- "mtlo %[vector_64], $ac3 \n\t" /* odd 1 */
- "mthi $zero, $ac3 \n\t"
- "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average even 6 */
- "dpa.w.ph $ac2, %[p3], %[filter12] \n\t" /* even 8 */
- "dpa.w.ph $ac2, %[p4], %[filter34] \n\t" /* even 8 */
- "sb %[qload3], 10(%[dst]) \n\t" /* store even 6 to dst */
- "dpa.w.ph $ac2, %[p1], %[filter56] \n\t" /* even 8 */
- "dpa.w.ph $ac2, %[p5], %[filter78] \n\t" /* even 8 */
- "extp %[Temp2], $ac2, 31 \n\t" /* even 8 */
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 7 */
-
- /* ODD pixels */
- "ulw %[qload1], 1(%[src]) \n\t"
- "ulw %[qload2], 5(%[src]) \n\t"
-
- "addqh_r.w %[st2], %[st2], %[st1] \n\t" /* average even 7 */
-
- /* odd 1. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* odd 2 */
- "mthi $zero, $ac1 \n\t"
- "preceu.ph.qbr %[p1], %[qload1] \n\t"
- "preceu.ph.qbl %[p2], %[qload1] \n\t"
- "preceu.ph.qbr %[p3], %[qload2] \n\t"
- "preceu.ph.qbl %[p4], %[qload2] \n\t"
- "sb %[st2], 12(%[dst]) \n\t" /* store even 7 to dst */
- "ulw %[qload3], 9(%[src]) \n\t"
- "dpa.w.ph $ac3, %[p1], %[filter12] \n\t" /* odd 1 */
- "dpa.w.ph $ac3, %[p2], %[filter34] \n\t" /* odd 1 */
- "lbu %[qload2], 14(%[dst]) \n\t" /* load even 8 from dst */
- "dpa.w.ph $ac3, %[p3], %[filter56] \n\t" /* odd 1 */
- "dpa.w.ph $ac3, %[p4], %[filter78] \n\t" /* odd 1 */
- "extp %[Temp3], $ac3, 31 \n\t" /* odd 1 */
- "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 8 */
-
- "lbu %[st1], 1(%[dst]) \n\t" /* load odd 1 from dst */
-
- /* odd 2. pixel */
- "mtlo %[vector_64], $ac2 \n\t" /* odd 3 */
- "mthi $zero, $ac2 \n\t"
- "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average even 8 */
- "preceu.ph.qbr %[p1], %[qload3] \n\t"
- "preceu.ph.qbl %[p5], %[qload3] \n\t"
- "sb %[qload2], 14(%[dst]) \n\t" /* store even 8 to dst */
- "ulw %[qload1], 13(%[src]) \n\t"
- "dpa.w.ph $ac1, %[p2], %[filter12] \n\t" /* odd 2 */
- "dpa.w.ph $ac1, %[p3], %[filter34] \n\t" /* odd 2 */
- "dpa.w.ph $ac1, %[p4], %[filter56] \n\t" /* odd 2 */
- "dpa.w.ph $ac1, %[p1], %[filter78] \n\t" /* odd 2 */
- "lbu %[qload3], 3(%[dst]) \n\t" /* load odd 2 from dst */
- "extp %[Temp1], $ac1, 31 \n\t" /* odd 2 */
- "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 1 */
-
- /* odd 3. pixel */
- "mtlo %[vector_64], $ac3 \n\t" /* odd 4 */
- "mthi $zero, $ac3 \n\t"
- "addqh_r.w %[st3], %[st3], %[st1] \n\t" /* average odd 1 */
- "preceu.ph.qbr %[p2], %[qload1] \n\t"
- "dpa.w.ph $ac2, %[p3], %[filter12] \n\t" /* odd 3 */
- "dpa.w.ph $ac2, %[p4], %[filter34] \n\t" /* odd 3 */
- "dpa.w.ph $ac2, %[p1], %[filter56] \n\t" /* odd 3 */
- "dpa.w.ph $ac2, %[p5], %[filter78] \n\t" /* odd 3 */
- "sb %[st3], 1(%[dst]) \n\t" /* store odd 1 to dst */
- "extp %[Temp2], $ac2, 31 \n\t" /* odd 3 */
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 2 */
-
- /* odd 4. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* odd 5 */
- "mthi $zero, $ac1 \n\t"
- "addqh_r.w %[qload3], %[qload3], %[st1] \n\t" /* average odd 2 */
- "preceu.ph.qbl %[p3], %[qload1] \n\t"
- "sb %[qload3], 3(%[dst]) \n\t" /* store odd 2 to dst */
- "lbu %[qload1], 5(%[dst]) \n\t" /* load odd 3 from dst */
- "ulw %[qload2], 17(%[src]) \n\t"
- "dpa.w.ph $ac3, %[p4], %[filter12] \n\t" /* odd 4 */
- "dpa.w.ph $ac3, %[p1], %[filter34] \n\t" /* odd 4 */
- "dpa.w.ph $ac3, %[p5], %[filter56] \n\t" /* odd 4 */
- "dpa.w.ph $ac3, %[p2], %[filter78] \n\t" /* odd 4 */
- "extp %[Temp3], $ac3, 31 \n\t" /* odd 4 */
- "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 3 */
-
- "lbu %[st1], 7(%[dst]) \n\t" /* load odd 4 from dst */
-
- /* odd 5. pixel */
- "mtlo %[vector_64], $ac2 \n\t" /* odd 6 */
- "mthi $zero, $ac2 \n\t"
- "addqh_r.w %[qload1], %[qload1], %[st2] \n\t" /* average odd 3 */
- "preceu.ph.qbr %[p4], %[qload2] \n\t"
- "sb %[qload1], 5(%[dst]) \n\t" /* store odd 3 to dst */
- "dpa.w.ph $ac1, %[p1], %[filter12] \n\t" /* odd 5 */
- "dpa.w.ph $ac1, %[p5], %[filter34] \n\t" /* odd 5 */
- "dpa.w.ph $ac1, %[p2], %[filter56] \n\t" /* odd 5 */
- "dpa.w.ph $ac1, %[p3], %[filter78] \n\t" /* odd 5 */
- "extp %[Temp1], $ac1, 31 \n\t" /* odd 5 */
- "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 4 */
-
- "lbu %[qload1], 9(%[dst]) \n\t" /* load odd 5 from dst */
-
- /* odd 6. pixel */
- "mtlo %[vector_64], $ac3 \n\t" /* odd 7 */
- "mthi $zero, $ac3 \n\t"
- "addqh_r.w %[st1], %[st1], %[st3] \n\t" /* average odd 4 */
- "preceu.ph.qbl %[p1], %[qload2] \n\t"
- "sb %[st1], 7(%[dst]) \n\t" /* store odd 4 to dst */
- "ulw %[qload3], 21(%[src]) \n\t"
- "dpa.w.ph $ac2, %[p5], %[filter12] \n\t" /* odd 6 */
- "dpa.w.ph $ac2, %[p2], %[filter34] \n\t" /* odd 6 */
- "dpa.w.ph $ac2, %[p3], %[filter56] \n\t" /* odd 6 */
- "dpa.w.ph $ac2, %[p4], %[filter78] \n\t" /* odd 6 */
- "extp %[Temp2], $ac2, 31 \n\t" /* odd 6 */
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 5 */
-
- /* odd 7. pixel */
- "mtlo %[vector_64], $ac1 \n\t" /* odd 8 */
- "mthi $zero, $ac1 \n\t"
- "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average odd 5 */
- "preceu.ph.qbr %[p5], %[qload3] \n\t"
- "sb %[qload1], 9(%[dst]) \n\t" /* store odd 5 to dst */
- "lbu %[qload2], 11(%[dst]) \n\t" /* load odd 6 from dst */
- "dpa.w.ph $ac3, %[p2], %[filter12] \n\t" /* odd 7 */
- "dpa.w.ph $ac3, %[p3], %[filter34] \n\t" /* odd 7 */
- "dpa.w.ph $ac3, %[p4], %[filter56] \n\t" /* odd 7 */
- "dpa.w.ph $ac3, %[p1], %[filter78] \n\t" /* odd 7 */
- "extp %[Temp3], $ac3, 31 \n\t" /* odd 7 */
-
- "lbu %[qload3], 13(%[dst]) \n\t" /* load odd 7 from dst */
-
- /* odd 8. pixel */
- "dpa.w.ph $ac1, %[p3], %[filter12] \n\t" /* odd 8 */
- "dpa.w.ph $ac1, %[p4], %[filter34] \n\t" /* odd 8 */
- "dpa.w.ph $ac1, %[p1], %[filter56] \n\t" /* odd 8 */
- "dpa.w.ph $ac1, %[p5], %[filter78] \n\t" /* odd 8 */
- "extp %[Temp1], $ac1, 31 \n\t" /* odd 8 */
-
- "lbu %[qload1], 15(%[dst]) \n\t" /* load odd 8 from dst */
-
- "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 6 */
- "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average odd 6 */
-
- "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 7 */
- "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average odd 7 */
-
- "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 8 */
- "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average odd 8 */
-
- "sb %[qload2], 11(%[dst]) \n\t" /* store odd 6 to dst */
- "sb %[qload3], 13(%[dst]) \n\t" /* store odd 7 to dst */
- "sb %[qload1], 15(%[dst]) \n\t" /* store odd 8 to dst */
-
- : [qload1] "=&r"(qload1), [qload2] "=&r"(qload2), [st1] "=&r"(st1),
- [st2] "=&r"(st2), [st3] "=&r"(st3), [p1] "=&r"(p1), [p2] "=&r"(p2),
- [p3] "=&r"(p3), [p4] "=&r"(p4), [qload3] "=&r"(qload3),
- [p5] "=&r"(p5), [Temp1] "=&r"(Temp1), [Temp2] "=&r"(Temp2),
- [Temp3] "=&r"(Temp3)
- : [filter12] "r"(filter12), [filter34] "r"(filter34),
- [filter56] "r"(filter56), [filter78] "r"(filter78),
- [vector_64] "r"(vector_64), [cm] "r"(cm), [dst] "r"(dst),
- [src] "r"(src));
-
- src += 16;
- dst += 16;
- }
-
- /* Next row... */
- src_ptr += src_stride;
- dst_ptr += dst_stride;
- }
-}
-
-void aom_convolve8_avg_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h) {
- assert(x_step_q4 == 16);
- assert(((const int32_t *)filter_x)[1] != 0x800000);
-
- if (((const int32_t *)filter_x)[0] == 0) {
- aom_convolve2_avg_horiz_dspr2(src, src_stride, dst, dst_stride, filter_x,
- x_step_q4, filter_y, y_step_q4, w, h);
- } else {
- uint32_t pos = 38;
-
- src -= 3;
-
- /* bit positon for extract from acc */
- __asm__ __volatile__("wrdsp %[pos], 1 \n\t"
- :
- : [pos] "r"(pos));
-
- /* prefetch data to cache memory */
- prefetch_load(src);
- prefetch_load(src + 32);
- prefetch_store(dst);
-
- switch (w) {
- case 4:
- convolve_avg_horiz_4_dspr2(src, src_stride, dst, dst_stride, filter_x,
- h);
- break;
- case 8:
- convolve_avg_horiz_8_dspr2(src, src_stride, dst, dst_stride, filter_x,
- h);
- break;
- case 16:
- convolve_avg_horiz_16_dspr2(src, src_stride, dst, dst_stride, filter_x,
- h, 1);
- break;
- case 32:
- convolve_avg_horiz_16_dspr2(src, src_stride, dst, dst_stride, filter_x,
- h, 2);
- break;
- case 64:
- prefetch_load(src + 64);
- prefetch_store(dst + 32);
-
- convolve_avg_horiz_64_dspr2(src, src_stride, dst, dst_stride, filter_x,
- h);
- break;
- default:
- aom_convolve8_avg_horiz_c(src + 3, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4, w,
- h);
- break;
- }
- }
-}
-#endif
diff --git a/aom_dsp/mips/convolve_common_dspr2.h b/aom_dsp/mips/convolve_common_dspr2.h
index f8fd9e2..c6895fb 100644
--- a/aom_dsp/mips/convolve_common_dspr2.h
+++ b/aom_dsp/mips/convolve_common_dspr2.h
@@ -29,18 +29,6 @@
const int16_t *filter_y, int y_step_q4, int w,
int h);
-void aom_convolve2_avg_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h);
-
-void aom_convolve2_avg_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4, int w,
- int h);
-
void aom_convolve2_dspr2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
ptrdiff_t dst_stride, const int16_t *filter, int w,
int h);
diff --git a/aom_dsp/x86/aom_asm_stubs.c b/aom_dsp/x86/aom_asm_stubs.c
index 522e6ef..2c51cb9 100644
--- a/aom_dsp/x86/aom_asm_stubs.c
+++ b/aom_dsp/x86/aom_asm_stubs.c
@@ -20,12 +20,6 @@
filter8_1dfunction aom_filter_block1d8_h8_sse2;
filter8_1dfunction aom_filter_block1d4_v8_sse2;
filter8_1dfunction aom_filter_block1d4_h8_sse2;
-filter8_1dfunction aom_filter_block1d16_v8_avg_sse2;
-filter8_1dfunction aom_filter_block1d16_h8_avg_sse2;
-filter8_1dfunction aom_filter_block1d8_v8_avg_sse2;
-filter8_1dfunction aom_filter_block1d8_h8_avg_sse2;
-filter8_1dfunction aom_filter_block1d4_v8_avg_sse2;
-filter8_1dfunction aom_filter_block1d4_h8_avg_sse2;
filter8_1dfunction aom_filter_block1d16_v2_sse2;
filter8_1dfunction aom_filter_block1d16_h2_sse2;
@@ -33,12 +27,6 @@
filter8_1dfunction aom_filter_block1d8_h2_sse2;
filter8_1dfunction aom_filter_block1d4_v2_sse2;
filter8_1dfunction aom_filter_block1d4_h2_sse2;
-filter8_1dfunction aom_filter_block1d16_v2_avg_sse2;
-filter8_1dfunction aom_filter_block1d16_h2_avg_sse2;
-filter8_1dfunction aom_filter_block1d8_v2_avg_sse2;
-filter8_1dfunction aom_filter_block1d8_h2_avg_sse2;
-filter8_1dfunction aom_filter_block1d4_v2_avg_sse2;
-filter8_1dfunction aom_filter_block1d4_h2_avg_sse2;
// void aom_convolve8_horiz_sse2(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
@@ -50,33 +38,15 @@
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// int w, int h);
-// void aom_convolve8_avg_horiz_sse2(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h);
-// void aom_convolve8_avg_vert_sse2(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h);
FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , sse2);
FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , sse2);
-FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, sse2);
-FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_, sse2);
// void aom_convolve8_sse2(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// int w, int h);
-// void aom_convolve8_avg_sse2(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h);
FUN_CONV_2D(, sse2);
-FUN_CONV_2D(avg_, sse2);
#if ARCH_X86_64
highbd_filter8_1dfunction aom_highbd_filter_block1d16_v8_sse2;
@@ -85,12 +55,6 @@
highbd_filter8_1dfunction aom_highbd_filter_block1d8_h8_sse2;
highbd_filter8_1dfunction aom_highbd_filter_block1d4_v8_sse2;
highbd_filter8_1dfunction aom_highbd_filter_block1d4_h8_sse2;
-highbd_filter8_1dfunction aom_highbd_filter_block1d16_v8_avg_sse2;
-highbd_filter8_1dfunction aom_highbd_filter_block1d16_h8_avg_sse2;
-highbd_filter8_1dfunction aom_highbd_filter_block1d8_v8_avg_sse2;
-highbd_filter8_1dfunction aom_highbd_filter_block1d8_h8_avg_sse2;
-highbd_filter8_1dfunction aom_highbd_filter_block1d4_v8_avg_sse2;
-highbd_filter8_1dfunction aom_highbd_filter_block1d4_h8_avg_sse2;
highbd_filter8_1dfunction aom_highbd_filter_block1d16_v2_sse2;
highbd_filter8_1dfunction aom_highbd_filter_block1d16_h2_sse2;
@@ -98,12 +62,6 @@
highbd_filter8_1dfunction aom_highbd_filter_block1d8_h2_sse2;
highbd_filter8_1dfunction aom_highbd_filter_block1d4_v2_sse2;
highbd_filter8_1dfunction aom_highbd_filter_block1d4_h2_sse2;
-highbd_filter8_1dfunction aom_highbd_filter_block1d16_v2_avg_sse2;
-highbd_filter8_1dfunction aom_highbd_filter_block1d16_h2_avg_sse2;
-highbd_filter8_1dfunction aom_highbd_filter_block1d8_v2_avg_sse2;
-highbd_filter8_1dfunction aom_highbd_filter_block1d8_h2_avg_sse2;
-highbd_filter8_1dfunction aom_highbd_filter_block1d4_v2_avg_sse2;
-highbd_filter8_1dfunction aom_highbd_filter_block1d4_h2_avg_sse2;
// void aom_highbd_convolve8_horiz_sse2(const uint8_t *src,
// ptrdiff_t src_stride,
@@ -123,42 +81,15 @@
// const int16_t *filter_y,
// int y_step_q4,
// int w, int h, int bd);
-// void aom_highbd_convolve8_avg_horiz_sse2(const uint8_t *src,
-// ptrdiff_t src_stride,
-// uint8_t *dst,
-// ptrdiff_t dst_stride,
-// const int16_t *filter_x,
-// int x_step_q4,
-// const int16_t *filter_y,
-// int y_step_q4,
-// int w, int h, int bd);
-// void aom_highbd_convolve8_avg_vert_sse2(const uint8_t *src,
-// ptrdiff_t src_stride,
-// uint8_t *dst,
-// ptrdiff_t dst_stride,
-// const int16_t *filter_x,
-// int x_step_q4,
-// const int16_t *filter_y,
-// int y_step_q4,
-// int w, int h, int bd);
HIGH_FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , sse2);
HIGH_FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , sse2);
-HIGH_FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, sse2);
-HIGH_FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_,
- sse2);
// void aom_highbd_convolve8_sse2(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// int w, int h, int bd);
-// void aom_highbd_convolve8_avg_sse2(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h, int bd);
HIGH_FUN_CONV_2D(, sse2);
-HIGH_FUN_CONV_2D(avg_, sse2);
// The SSE2 highbd convolve functions can deal with coefficients up to 32767.
// So redirect highbd_convolve8_add_src to regular highbd_convolve8.
diff --git a/aom_dsp/x86/aom_high_subpixel_8t_sse2.asm b/aom_dsp/x86/aom_high_subpixel_8t_sse2.asm
index 11889bb..b6f0407 100644
--- a/aom_dsp/x86/aom_high_subpixel_8t_sse2.asm
+++ b/aom_dsp/x86/aom_high_subpixel_8t_sse2.asm
@@ -394,169 +394,6 @@
pop rbp
ret
-global sym(aom_highbd_filter_block1d4_v8_avg_sse2) PRIVATE
-sym(aom_highbd_filter_block1d4_v8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 7
- %define k0k6 [rsp + 16 * 0]
- %define k2k5 [rsp + 16 * 1]
- %define k3k4 [rsp + 16 * 2]
- %define k1k7 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define max [rsp + 16 * 5]
- %define min [rsp + 16 * 6]
-
- HIGH_GET_FILTERS_4
-
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rbx, DWORD PTR arg(3) ;out_pitch
- lea rax, [rax + rax] ;bytes per line
- lea rbx, [rbx + rbx]
- lea rdx, [rax + rax * 2]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movq xmm0, [rsi] ;load src: row 0
- movq xmm1, [rsi + rax] ;1
- movq xmm6, [rsi + rdx * 2] ;6
- lea rsi, [rsi + rax]
- movq xmm7, [rsi + rdx * 2] ;7
- movq xmm2, [rsi + rax] ;2
- movq xmm3, [rsi + rax * 2] ;3
- movq xmm4, [rsi + rdx] ;4
- movq xmm5, [rsi + rax * 4] ;5
-
- HIGH_APPLY_FILTER_4 1
-
- lea rdi, [rdi + rbx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 7
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(aom_highbd_filter_block1d8_v8_avg_sse2) PRIVATE
-sym(aom_highbd_filter_block1d8_v8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 8
- %define k0k1 [rsp + 16 * 0]
- %define k6k7 [rsp + 16 * 1]
- %define k2k5 [rsp + 16 * 2]
- %define k3k4 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define temp [rsp + 16 * 5]
- %define max [rsp + 16 * 6]
- %define min [rsp + 16 * 7]
-
- HIGH_GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rbx, DWORD PTR arg(3) ;out_pitch
- lea rax, [rax + rax] ;bytes per line
- lea rbx, [rbx + rbx]
- lea rdx, [rax + rax * 2]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-.loop:
- LOAD_VERT_8 0
- HIGH_APPLY_FILTER_8 1, 0
-
- lea rdi, [rdi + rbx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 8
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(aom_highbd_filter_block1d16_v8_avg_sse2) PRIVATE
-sym(aom_highbd_filter_block1d16_v8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 8
- %define k0k1 [rsp + 16 * 0]
- %define k6k7 [rsp + 16 * 1]
- %define k2k5 [rsp + 16 * 2]
- %define k3k4 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define temp [rsp + 16 * 5]
- %define max [rsp + 16 * 6]
- %define min [rsp + 16 * 7]
-
- HIGH_GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rbx, DWORD PTR arg(3) ;out_pitch
- lea rax, [rax + rax] ;bytes per line
- lea rbx, [rbx + rbx]
- lea rdx, [rax + rax * 2]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-.loop:
- LOAD_VERT_8 0
- HIGH_APPLY_FILTER_8 1, 0
- sub rsi, rax
-
- LOAD_VERT_8 16
- HIGH_APPLY_FILTER_8 1, 16
- add rdi, rbx
-
- dec rcx
- jnz .loop
-
- add rsp, 16 * 8
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
;void aom_filter_block1d4_h8_sse2
;(
; unsigned char *src_ptr,
@@ -774,194 +611,3 @@
UNSHADOW_ARGS
pop rbp
ret
-
-global sym(aom_highbd_filter_block1d4_h8_avg_sse2) PRIVATE
-sym(aom_highbd_filter_block1d4_h8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 7
- %define k0k6 [rsp + 16 * 0]
- %define k2k5 [rsp + 16 * 1]
- %define k3k4 [rsp + 16 * 2]
- %define k1k7 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define max [rsp + 16 * 5]
- %define min [rsp + 16 * 6]
-
- HIGH_GET_FILTERS_4
-
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- lea rax, [rax + rax] ;bytes per line
- lea rdx, [rdx + rdx]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movdqu xmm0, [rsi - 6] ;load src
- movdqu xmm4, [rsi + 2]
- movdqa xmm1, xmm0
- movdqa xmm6, xmm4
- movdqa xmm7, xmm4
- movdqa xmm2, xmm0
- movdqa xmm3, xmm0
- movdqa xmm5, xmm4
-
- psrldq xmm1, 2
- psrldq xmm6, 4
- psrldq xmm7, 6
- psrldq xmm2, 4
- psrldq xmm3, 6
- psrldq xmm5, 2
-
- HIGH_APPLY_FILTER_4 1
-
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 7
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(aom_highbd_filter_block1d8_h8_avg_sse2) PRIVATE
-sym(aom_highbd_filter_block1d8_h8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 8
- %define k0k1 [rsp + 16 * 0]
- %define k6k7 [rsp + 16 * 1]
- %define k2k5 [rsp + 16 * 2]
- %define k3k4 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define temp [rsp + 16 * 5]
- %define max [rsp + 16 * 6]
- %define min [rsp + 16 * 7]
-
- HIGH_GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- lea rax, [rax + rax] ;bytes per line
- lea rdx, [rdx + rdx]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movdqu xmm0, [rsi - 6] ;load src
- movdqu xmm1, [rsi - 4]
- movdqu xmm2, [rsi - 2]
- movdqu xmm3, [rsi]
- movdqu xmm4, [rsi + 2]
- movdqu xmm5, [rsi + 4]
- movdqu xmm6, [rsi + 6]
- movdqu xmm7, [rsi + 8]
-
- HIGH_APPLY_FILTER_8 1, 0
-
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 8
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(aom_highbd_filter_block1d16_h8_avg_sse2) PRIVATE
-sym(aom_highbd_filter_block1d16_h8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 8
- %define k0k1 [rsp + 16 * 0]
- %define k6k7 [rsp + 16 * 1]
- %define k2k5 [rsp + 16 * 2]
- %define k3k4 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define temp [rsp + 16 * 5]
- %define max [rsp + 16 * 6]
- %define min [rsp + 16 * 7]
-
- HIGH_GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- lea rax, [rax + rax] ;bytes per line
- lea rdx, [rdx + rdx]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movdqu xmm0, [rsi - 6] ;load src
- movdqu xmm1, [rsi - 4]
- movdqu xmm2, [rsi - 2]
- movdqu xmm3, [rsi]
- movdqu xmm4, [rsi + 2]
- movdqu xmm5, [rsi + 4]
- movdqu xmm6, [rsi + 6]
- movdqu xmm7, [rsi + 8]
-
- HIGH_APPLY_FILTER_8 1, 0
-
- movdqu xmm0, [rsi + 10] ;load src
- movdqu xmm1, [rsi + 12]
- movdqu xmm2, [rsi + 14]
- movdqu xmm3, [rsi + 16]
- movdqu xmm4, [rsi + 18]
- movdqu xmm5, [rsi + 20]
- movdqu xmm6, [rsi + 22]
- movdqu xmm7, [rsi + 24]
-
- HIGH_APPLY_FILTER_8 1, 16
-
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 8
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
diff --git a/aom_dsp/x86/aom_high_subpixel_bilinear_sse2.asm b/aom_dsp/x86/aom_high_subpixel_bilinear_sse2.asm
index c6479d7..7b3fe64 100644
--- a/aom_dsp/x86/aom_high_subpixel_bilinear_sse2.asm
+++ b/aom_dsp/x86/aom_high_subpixel_bilinear_sse2.asm
@@ -256,86 +256,6 @@
ret
%endif
-global sym(aom_highbd_filter_block1d4_v2_avg_sse2) PRIVATE
-sym(aom_highbd_filter_block1d4_v2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- push rsi
- push rdi
- ; end prolog
-
- HIGH_GET_PARAM_4
-.loop:
- movq xmm0, [rsi] ;load src
- movq xmm1, [rsi + 2*rax]
-
- HIGH_APPLY_FILTER_4 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-%if ARCH_X86_64
-global sym(aom_highbd_filter_block1d8_v2_avg_sse2) PRIVATE
-sym(aom_highbd_filter_block1d8_v2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 8
- push rsi
- push rdi
- ; end prolog
-
- HIGH_GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;0
- movdqu xmm1, [rsi + 2*rax] ;1
-
- HIGH_APPLY_FILTER_8 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(aom_highbd_filter_block1d16_v2_avg_sse2) PRIVATE
-sym(aom_highbd_filter_block1d16_v2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 9
- push rsi
- push rdi
- ; end prolog
-
- HIGH_GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;0
- movdqu xmm1, [rsi + 2*rax] ;1
- movdqu xmm2, [rsi + 16]
- movdqu xmm3, [rsi + 2*rax + 16]
-
- HIGH_APPLY_FILTER_16 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-%endif
-
global sym(aom_highbd_filter_block1d4_h2_sse2) PRIVATE
sym(aom_highbd_filter_block1d4_h2_sse2):
push rbp
@@ -416,84 +336,3 @@
pop rbp
ret
%endif
-
-global sym(aom_highbd_filter_block1d4_h2_avg_sse2) PRIVATE
-sym(aom_highbd_filter_block1d4_h2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- push rsi
- push rdi
- ; end prolog
-
- HIGH_GET_PARAM_4
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqa xmm1, xmm0
- psrldq xmm1, 2
-
- HIGH_APPLY_FILTER_4 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-%if ARCH_X86_64
-global sym(aom_highbd_filter_block1d8_h2_avg_sse2) PRIVATE
-sym(aom_highbd_filter_block1d8_h2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 8
- push rsi
- push rdi
- ; end prolog
-
- HIGH_GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqu xmm1, [rsi + 2]
-
- HIGH_APPLY_FILTER_8 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(aom_highbd_filter_block1d16_h2_avg_sse2) PRIVATE
-sym(aom_highbd_filter_block1d16_h2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 9
- push rsi
- push rdi
- ; end prolog
-
- HIGH_GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqu xmm1, [rsi + 2]
- movdqu xmm2, [rsi + 16]
- movdqu xmm3, [rsi + 18]
-
- HIGH_APPLY_FILTER_16 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-%endif
diff --git a/aom_dsp/x86/aom_subpixel_8t_intrin_ssse3.c b/aom_dsp/x86/aom_subpixel_8t_intrin_ssse3.c
index a9a15fd..7dd1f1d 100644
--- a/aom_dsp/x86/aom_subpixel_8t_intrin_ssse3.c
+++ b/aom_dsp/x86/aom_subpixel_8t_intrin_ssse3.c
@@ -285,12 +285,6 @@
filter8_1dfunction aom_filter_block1d8_h8_ssse3;
filter8_1dfunction aom_filter_block1d4_v8_ssse3;
filter8_1dfunction aom_filter_block1d4_h8_ssse3;
-filter8_1dfunction aom_filter_block1d16_v8_avg_ssse3;
-filter8_1dfunction aom_filter_block1d16_h8_avg_ssse3;
-filter8_1dfunction aom_filter_block1d8_v8_avg_ssse3;
-filter8_1dfunction aom_filter_block1d8_h8_avg_ssse3;
-filter8_1dfunction aom_filter_block1d4_v8_avg_ssse3;
-filter8_1dfunction aom_filter_block1d4_h8_avg_ssse3;
filter8_1dfunction aom_filter_block1d16_v8_add_src_ssse3;
filter8_1dfunction aom_filter_block1d16_h8_add_src_ssse3;
filter8_1dfunction aom_filter_block1d8_v8_add_src_ssse3;
@@ -304,12 +298,6 @@
filter8_1dfunction aom_filter_block1d8_h2_ssse3;
filter8_1dfunction aom_filter_block1d4_v2_ssse3;
filter8_1dfunction aom_filter_block1d4_h2_ssse3;
-filter8_1dfunction aom_filter_block1d16_v2_avg_ssse3;
-filter8_1dfunction aom_filter_block1d16_h2_avg_ssse3;
-filter8_1dfunction aom_filter_block1d8_v2_avg_ssse3;
-filter8_1dfunction aom_filter_block1d8_h2_avg_ssse3;
-filter8_1dfunction aom_filter_block1d4_v2_avg_ssse3;
-filter8_1dfunction aom_filter_block1d4_h2_avg_ssse3;
// void aom_convolve8_horiz_ssse3(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
@@ -321,21 +309,8 @@
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// int w, int h);
-// void aom_convolve8_avg_horiz_ssse3(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h);
-// void aom_convolve8_avg_vert_ssse3(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h);
FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , ssse3);
FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , ssse3);
-FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, ssse3);
-FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_,
- ssse3);
FUN_CONV_1D_NO_BILINEAR(add_src_horiz, x_step_q4, filter_x, h, src, add_src_,
ssse3);
@@ -347,11 +322,5 @@
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// int w, int h);
-// void aom_convolve8_avg_ssse3(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h);
FUN_CONV_2D(, ssse3);
-FUN_CONV_2D(avg_, ssse3);
FUN_CONV_2D_NO_BILINEAR(add_src_, add_src_, ssse3);
diff --git a/aom_dsp/x86/aom_subpixel_8t_sse2.asm b/aom_dsp/x86/aom_subpixel_8t_sse2.asm
index b4bf4ee..c88fc9f 100644
--- a/aom_dsp/x86/aom_subpixel_8t_sse2.asm
+++ b/aom_dsp/x86/aom_subpixel_8t_sse2.asm
@@ -370,166 +370,6 @@
pop rbp
ret
-global sym(aom_filter_block1d4_v8_avg_sse2) PRIVATE
-sym(aom_filter_block1d4_v8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 6
- %define k0k1 [rsp + 16 * 0]
- %define k2k3 [rsp + 16 * 1]
- %define k5k4 [rsp + 16 * 2]
- %define k6k7 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define zero [rsp + 16 * 5]
-
- GET_FILTERS_4
-
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rbx, DWORD PTR arg(3) ;out_pitch
- lea rdx, [rax + rax * 2]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movd xmm0, [rsi] ;load src: row 0
- movd xmm1, [rsi + rax] ;1
- movd xmm6, [rsi + rdx * 2] ;6
- lea rsi, [rsi + rax]
- movd xmm7, [rsi + rdx * 2] ;7
- movd xmm2, [rsi + rax] ;2
- movd xmm3, [rsi + rax * 2] ;3
- movd xmm4, [rsi + rdx] ;4
- movd xmm5, [rsi + rax * 4] ;5
-
- APPLY_FILTER_4 1
-
- lea rdi, [rdi + rbx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 6
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(aom_filter_block1d8_v8_avg_sse2) PRIVATE
-sym(aom_filter_block1d8_v8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 10
- %define k0 [rsp + 16 * 0]
- %define k1 [rsp + 16 * 1]
- %define k2 [rsp + 16 * 2]
- %define k3 [rsp + 16 * 3]
- %define k4 [rsp + 16 * 4]
- %define k5 [rsp + 16 * 5]
- %define k6 [rsp + 16 * 6]
- %define k7 [rsp + 16 * 7]
- %define krd [rsp + 16 * 8]
- %define zero [rsp + 16 * 9]
-
- GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rbx, DWORD PTR arg(3) ;out_pitch
- lea rdx, [rax + rax * 2]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-.loop:
- LOAD_VERT_8 0
- APPLY_FILTER_8 1, 0
-
- lea rdi, [rdi + rbx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 10
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(aom_filter_block1d16_v8_avg_sse2) PRIVATE
-sym(aom_filter_block1d16_v8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 10
- %define k0 [rsp + 16 * 0]
- %define k1 [rsp + 16 * 1]
- %define k2 [rsp + 16 * 2]
- %define k3 [rsp + 16 * 3]
- %define k4 [rsp + 16 * 4]
- %define k5 [rsp + 16 * 5]
- %define k6 [rsp + 16 * 6]
- %define k7 [rsp + 16 * 7]
- %define krd [rsp + 16 * 8]
- %define zero [rsp + 16 * 9]
-
- GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rbx, DWORD PTR arg(3) ;out_pitch
- lea rdx, [rax + rax * 2]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-.loop:
- LOAD_VERT_8 0
- APPLY_FILTER_8 1, 0
- sub rsi, rax
-
- LOAD_VERT_8 8
- APPLY_FILTER_8 1, 8
- add rdi, rbx
-
- dec rcx
- jnz .loop
-
- add rsp, 16 * 10
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
;void aom_filter_block1d4_h8_sse2
;(
; unsigned char *src_ptr,
@@ -773,220 +613,3 @@
UNSHADOW_ARGS
pop rbp
ret
-
-global sym(aom_filter_block1d4_h8_avg_sse2) PRIVATE
-sym(aom_filter_block1d4_h8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 6
- %define k0k1 [rsp + 16 * 0]
- %define k2k3 [rsp + 16 * 1]
- %define k5k4 [rsp + 16 * 2]
- %define k6k7 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define zero [rsp + 16 * 5]
-
- GET_FILTERS_4
-
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movdqu xmm0, [rsi - 3] ;load src
-
- movdqa xmm1, xmm0
- movdqa xmm6, xmm0
- movdqa xmm7, xmm0
- movdqa xmm2, xmm0
- movdqa xmm3, xmm0
- movdqa xmm5, xmm0
- movdqa xmm4, xmm0
-
- psrldq xmm1, 1
- psrldq xmm6, 6
- psrldq xmm7, 7
- psrldq xmm2, 2
- psrldq xmm3, 3
- psrldq xmm5, 5
- psrldq xmm4, 4
-
- APPLY_FILTER_4 1
-
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 6
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(aom_filter_block1d8_h8_avg_sse2) PRIVATE
-sym(aom_filter_block1d8_h8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 10
- %define k0 [rsp + 16 * 0]
- %define k1 [rsp + 16 * 1]
- %define k2 [rsp + 16 * 2]
- %define k3 [rsp + 16 * 3]
- %define k4 [rsp + 16 * 4]
- %define k5 [rsp + 16 * 5]
- %define k6 [rsp + 16 * 6]
- %define k7 [rsp + 16 * 7]
- %define krd [rsp + 16 * 8]
- %define zero [rsp + 16 * 9]
-
- GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movdqu xmm0, [rsi - 3] ;load src
-
- movdqa xmm1, xmm0
- movdqa xmm6, xmm0
- movdqa xmm7, xmm0
- movdqa xmm2, xmm0
- movdqa xmm5, xmm0
- movdqa xmm3, xmm0
- movdqa xmm4, xmm0
-
- psrldq xmm1, 1
- psrldq xmm6, 6
- psrldq xmm7, 7
- psrldq xmm2, 2
- psrldq xmm5, 5
- psrldq xmm3, 3
- psrldq xmm4, 4
-
- APPLY_FILTER_8 1, 0
-
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 10
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(aom_filter_block1d16_h8_avg_sse2) PRIVATE
-sym(aom_filter_block1d16_h8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 10
- %define k0 [rsp + 16 * 0]
- %define k1 [rsp + 16 * 1]
- %define k2 [rsp + 16 * 2]
- %define k3 [rsp + 16 * 3]
- %define k4 [rsp + 16 * 4]
- %define k5 [rsp + 16 * 5]
- %define k6 [rsp + 16 * 6]
- %define k7 [rsp + 16 * 7]
- %define krd [rsp + 16 * 8]
- %define zero [rsp + 16 * 9]
-
- GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movdqu xmm0, [rsi - 3] ;load src
-
- movdqa xmm1, xmm0
- movdqa xmm6, xmm0
- movdqa xmm7, xmm0
- movdqa xmm2, xmm0
- movdqa xmm5, xmm0
- movdqa xmm3, xmm0
- movdqa xmm4, xmm0
-
- psrldq xmm1, 1
- psrldq xmm6, 6
- psrldq xmm7, 7
- psrldq xmm2, 2
- psrldq xmm5, 5
- psrldq xmm3, 3
- psrldq xmm4, 4
-
- APPLY_FILTER_8 1, 0
-
- movdqu xmm0, [rsi + 5] ;load src
-
- movdqa xmm1, xmm0
- movdqa xmm6, xmm0
- movdqa xmm7, xmm0
- movdqa xmm2, xmm0
- movdqa xmm5, xmm0
- movdqa xmm3, xmm0
- movdqa xmm4, xmm0
-
- psrldq xmm1, 1
- psrldq xmm6, 6
- psrldq xmm7, 7
- psrldq xmm2, 2
- psrldq xmm5, 5
- psrldq xmm3, 3
- psrldq xmm4, 4
-
- APPLY_FILTER_8 1, 8
-
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 10
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
diff --git a/aom_dsp/x86/aom_subpixel_8t_ssse3.asm b/aom_dsp/x86/aom_subpixel_8t_ssse3.asm
index 05bed72..6dc2310 100644
--- a/aom_dsp/x86/aom_subpixel_8t_ssse3.asm
+++ b/aom_dsp/x86/aom_subpixel_8t_ssse3.asm
@@ -375,11 +375,8 @@
INIT_XMM ssse3
SUBPIX_HFILTER16 h8
-SUBPIX_HFILTER16 h8_avg
SUBPIX_HFILTER8 h8
-SUBPIX_HFILTER8 h8_avg
SUBPIX_HFILTER4 h8
-SUBPIX_HFILTER4 h8_avg
SUBPIX_HFILTER16 h8_add_src
SUBPIX_HFILTER8 h8_add_src
@@ -873,11 +870,8 @@
INIT_XMM ssse3
SUBPIX_VFILTER16 v8
-SUBPIX_VFILTER16 v8_avg
SUBPIX_VFILTER v8, 8
-SUBPIX_VFILTER v8_avg, 8
SUBPIX_VFILTER v8, 4
-SUBPIX_VFILTER v8_avg, 4
%if (ARCH_X86 || X86_SUBPIX_VFILTER_PREFER_SLOW_CELERON)
SUBPIX_VFILTER16 v8_add_src
diff --git a/aom_dsp/x86/aom_subpixel_bilinear_sse2.asm b/aom_dsp/x86/aom_subpixel_bilinear_sse2.asm
index 57aedcd..d0b4b28 100644
--- a/aom_dsp/x86/aom_subpixel_bilinear_sse2.asm
+++ b/aom_dsp/x86/aom_subpixel_bilinear_sse2.asm
@@ -214,84 +214,6 @@
pop rbp
ret
-global sym(aom_filter_block1d4_v2_avg_sse2) PRIVATE
-sym(aom_filter_block1d4_v2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM_4
-.loop:
- movd xmm0, [rsi] ;load src
- movd xmm1, [rsi + rax]
-
- APPLY_FILTER_4 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(aom_filter_block1d8_v2_avg_sse2) PRIVATE
-sym(aom_filter_block1d8_v2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movq xmm0, [rsi] ;0
- movq xmm1, [rsi + rax] ;1
-
- APPLY_FILTER_8 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(aom_filter_block1d16_v2_avg_sse2) PRIVATE
-sym(aom_filter_block1d16_v2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;0
- movdqu xmm1, [rsi + rax] ;1
- movdqa xmm2, xmm0
- movdqa xmm3, xmm1
-
- APPLY_FILTER_16 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
global sym(aom_filter_block1d4_h2_sse2) PRIVATE
sym(aom_filter_block1d4_h2_sse2):
push rbp
@@ -371,83 +293,3 @@
UNSHADOW_ARGS
pop rbp
ret
-
-global sym(aom_filter_block1d4_h2_avg_sse2) PRIVATE
-sym(aom_filter_block1d4_h2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM_4
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqa xmm1, xmm0
- psrldq xmm1, 1
-
- APPLY_FILTER_4 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(aom_filter_block1d8_h2_avg_sse2) PRIVATE
-sym(aom_filter_block1d8_h2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqa xmm1, xmm0
- psrldq xmm1, 1
-
- APPLY_FILTER_8 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(aom_filter_block1d16_h2_avg_sse2) PRIVATE
-sym(aom_filter_block1d16_h2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqu xmm1, [rsi + 1]
- movdqa xmm2, xmm0
- movdqa xmm3, xmm1
-
- APPLY_FILTER_16 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
diff --git a/aom_dsp/x86/aom_subpixel_bilinear_ssse3.asm b/aom_dsp/x86/aom_subpixel_bilinear_ssse3.asm
index 727fb30..59edc49 100644
--- a/aom_dsp/x86/aom_subpixel_bilinear_ssse3.asm
+++ b/aom_dsp/x86/aom_subpixel_bilinear_ssse3.asm
@@ -187,83 +187,6 @@
pop rbp
ret
-global sym(aom_filter_block1d4_v2_avg_ssse3) PRIVATE
-sym(aom_filter_block1d4_v2_avg_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM_4
-.loop:
- movd xmm0, [rsi] ;load src
- movd xmm1, [rsi + rax]
-
- APPLY_FILTER_4 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(aom_filter_block1d8_v2_avg_ssse3) PRIVATE
-sym(aom_filter_block1d8_v2_avg_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movq xmm0, [rsi] ;0
- movq xmm1, [rsi + rax] ;1
-
- APPLY_FILTER_8 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(aom_filter_block1d16_v2_avg_ssse3) PRIVATE
-sym(aom_filter_block1d16_v2_avg_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;0
- movdqu xmm1, [rsi + rax] ;1
- movdqa xmm2, xmm0
-
- APPLY_FILTER_16 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
global sym(aom_filter_block1d4_h2_ssse3) PRIVATE
sym(aom_filter_block1d4_h2_ssse3):
push rbp
@@ -342,82 +265,3 @@
UNSHADOW_ARGS
pop rbp
ret
-
-global sym(aom_filter_block1d4_h2_avg_ssse3) PRIVATE
-sym(aom_filter_block1d4_h2_avg_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM_4
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqa xmm1, xmm0
- psrldq xmm1, 1
-
- APPLY_FILTER_4 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(aom_filter_block1d8_h2_avg_ssse3) PRIVATE
-sym(aom_filter_block1d8_h2_avg_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqa xmm1, xmm0
- psrldq xmm1, 1
-
- APPLY_FILTER_8 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(aom_filter_block1d16_h2_avg_ssse3) PRIVATE
-sym(aom_filter_block1d16_h2_avg_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqu xmm1, [rsi + 1]
- movdqa xmm2, xmm0
-
- APPLY_FILTER_16 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
diff --git a/aom_dsp/x86/highbd_convolve_avx2.c b/aom_dsp/x86/highbd_convolve_avx2.c
index a19af49..81586df 100644
--- a/aom_dsp/x86/highbd_convolve_avx2.c
+++ b/aom_dsp/x86/highbd_convolve_avx2.c
@@ -1066,250 +1066,6 @@
} while (height > 0);
}
-// Calculation with averaging the input pixels
-
-static INLINE void store_8x1_avg_pixels(const __m256i *y0, const __m256i *mask,
- uint16_t *dst) {
- const __m128i a0 = _mm256_castsi256_si128(*y0);
- const __m128i a1 = _mm256_extractf128_si256(*y0, 1);
- __m128i res = _mm_packus_epi32(a0, a1);
- const __m128i pix = _mm_loadu_si128((const __m128i *)dst);
- res = _mm_min_epi16(res, _mm256_castsi256_si128(*mask));
- res = _mm_avg_epu16(res, pix);
- _mm_storeu_si128((__m128i *)dst, res);
-}
-
-static INLINE void store_8x2_avg_pixels(const __m256i *y0, const __m256i *y1,
- const __m256i *mask, uint16_t *dst,
- ptrdiff_t pitch) {
- __m256i a = _mm256_packus_epi32(*y0, *y1);
- const __m128i pix0 = _mm_loadu_si128((const __m128i *)dst);
- const __m128i pix1 = _mm_loadu_si128((const __m128i *)(dst + pitch));
- const __m256i pix =
- _mm256_insertf128_si256(_mm256_castsi128_si256(pix0), pix1, 1);
- a = _mm256_min_epi16(a, *mask);
- a = _mm256_avg_epu16(a, pix);
- _mm_storeu_si128((__m128i *)dst, _mm256_castsi256_si128(a));
- _mm_storeu_si128((__m128i *)(dst + pitch), _mm256_extractf128_si256(a, 1));
-}
-
-static INLINE void store_16x1_avg_pixels(const __m256i *y0, const __m256i *y1,
- const __m256i *mask, uint16_t *dst) {
- __m256i a = _mm256_packus_epi32(*y0, *y1);
- const __m256i pix = _mm256_loadu_si256((const __m256i *)dst);
- a = _mm256_min_epi16(a, *mask);
- a = _mm256_avg_epu16(a, pix);
- _mm256_storeu_si256((__m256i *)dst, a);
-}
-
-static INLINE void store_16x2_avg_pixels(const __m256i *y0, const __m256i *y1,
- const __m256i *mask, uint16_t *dst,
- ptrdiff_t pitch) {
- const __m256i pix0 = _mm256_loadu_si256((const __m256i *)dst);
- const __m256i pix1 = _mm256_loadu_si256((const __m256i *)(dst + pitch));
- __m256i p = _mm256_min_epi16(*y0, *mask);
- p = _mm256_avg_epu16(p, pix0);
- _mm256_storeu_si256((__m256i *)dst, p);
-
- p = _mm256_min_epi16(*y1, *mask);
- p = _mm256_avg_epu16(p, pix1);
- _mm256_storeu_si256((__m256i *)(dst + pitch), p);
-}
-
-static INLINE void store_8x1_2t_avg_pixels_ver(const __m128i *y0,
- const __m128i *y1,
- const __m128i *mask,
- uint16_t *dst) {
- __m128i res = _mm_packus_epi32(*y0, *y1);
- const __m128i pix = _mm_loadu_si128((const __m128i *)dst);
- res = _mm_min_epi16(res, *mask);
- res = _mm_avg_epu16(res, pix);
- _mm_storeu_si128((__m128i *)dst, res);
-}
-
-static void aom_highbd_filter_block1d8_h8_avg_avx2(
- const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
- ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
- __m256i signal[8], res0, res1;
- const __m256i max = _mm256_set1_epi16((1 << bd) - 1);
-
- __m256i ff[4];
- pack_filters(filter, ff);
-
- src_ptr -= 3;
- do {
- pack_8x2_pixels(src_ptr, src_pitch, signal);
- filter_8x1_pixels(signal, ff, &res0);
- filter_8x1_pixels(&signal[4], ff, &res1);
- store_8x2_avg_pixels(&res0, &res1, &max, dst_ptr, dst_pitch);
- height -= 2;
- src_ptr += src_pitch << 1;
- dst_ptr += dst_pitch << 1;
- } while (height > 1);
-
- if (height > 0) {
- pack_8x1_pixels(src_ptr, signal);
- filter_8x1_pixels(signal, ff, &res0);
- store_8x1_avg_pixels(&res0, &max, dst_ptr);
- }
-}
-
-static void aom_highbd_filter_block1d16_h8_avg_avx2(
- const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
- ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
- __m256i signal[8], res0, res1;
- const __m256i max = _mm256_set1_epi16((1 << bd) - 1);
-
- __m256i ff[4];
- pack_filters(filter, ff);
-
- src_ptr -= 3;
- do {
- pack_16x1_pixels(src_ptr, signal);
- filter_8x1_pixels(signal, ff, &res0);
- filter_8x1_pixels(&signal[4], ff, &res1);
- store_16x1_avg_pixels(&res0, &res1, &max, dst_ptr);
- height -= 1;
- src_ptr += src_pitch;
- dst_ptr += dst_pitch;
- } while (height > 0);
-}
-
-static void aom_highbd_filter_block1d8_v8_avg_avx2(
- const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
- ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
- __m256i signal[9], res0, res1;
- const __m256i max = _mm256_set1_epi16((1 << bd) - 1);
-
- __m256i ff[4];
- pack_filters(filter, ff);
-
- pack_8x9_init(src_ptr, src_pitch, signal);
-
- do {
- pack_8x9_pixels(src_ptr, src_pitch, signal);
-
- filter_8x9_pixels(signal, ff, &res0, &res1);
- store_8x2_avg_pixels(&res0, &res1, &max, dst_ptr, dst_pitch);
- update_pixels(signal);
-
- src_ptr += src_pitch << 1;
- dst_ptr += dst_pitch << 1;
- height -= 2;
- } while (height > 0);
-}
-
-static void aom_highbd_filter_block1d16_v8_avg_avx2(
- const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
- ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
- __m256i signal[17], res0, res1;
- const __m256i max = _mm256_set1_epi16((1 << bd) - 1);
-
- __m256i ff[4];
- pack_filters(filter, ff);
-
- pack_16x9_init(src_ptr, src_pitch, signal);
-
- do {
- pack_16x9_pixels(src_ptr, src_pitch, signal);
- filter_16x9_pixels(signal, ff, &res0, &res1);
- store_16x2_avg_pixels(&res0, &res1, &max, dst_ptr, dst_pitch);
- update_16x9_pixels(signal);
-
- src_ptr += src_pitch << 1;
- dst_ptr += dst_pitch << 1;
- height -= 2;
- } while (height > 0);
-}
-
-static void aom_highbd_filter_block1d8_h2_avg_avx2(
- const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
- ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
- __m256i signal[2], res0, res1;
- const __m256i max = _mm256_set1_epi16((1 << bd) - 1);
-
- __m256i ff;
- pack_2t_filter(filter, &ff);
-
- src_ptr -= 3;
- do {
- pack_8x2_2t_pixels(src_ptr, src_pitch, signal);
- filter_16_2t_pixels(signal, &ff, &res0, &res1);
- store_8x2_avg_pixels(&res0, &res1, &max, dst_ptr, dst_pitch);
- height -= 2;
- src_ptr += src_pitch << 1;
- dst_ptr += dst_pitch << 1;
- } while (height > 1);
-
- if (height > 0) {
- pack_8x1_2t_pixels(src_ptr, signal);
- filter_8x1_2t_pixels(signal, &ff, &res0);
- store_8x1_avg_pixels(&res0, &max, dst_ptr);
- }
-}
-
-static void aom_highbd_filter_block1d16_h2_avg_avx2(
- const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
- ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
- __m256i signal[2], res0, res1;
- const __m256i max = _mm256_set1_epi16((1 << bd) - 1);
-
- __m256i ff;
- pack_2t_filter(filter, &ff);
-
- src_ptr -= 3;
- do {
- pack_16x1_2t_pixels(src_ptr, signal);
- filter_16_2t_pixels(signal, &ff, &res0, &res1);
- store_16x1_avg_pixels(&res0, &res1, &max, dst_ptr);
- height -= 1;
- src_ptr += src_pitch;
- dst_ptr += dst_pitch;
- } while (height > 0);
-}
-
-static void aom_highbd_filter_block1d16_v2_avg_avx2(
- const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
- ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
- __m256i signal[3], res0, res1;
- const __m256i max = _mm256_set1_epi16((1 << bd) - 1);
- __m256i ff;
-
- pack_2t_filter(filter, &ff);
- pack_16x2_init(src_ptr, signal);
-
- do {
- pack_16x2_2t_pixels(src_ptr, src_pitch, signal);
- filter_16x2_2t_pixels(signal, &ff, &res0, &res1);
- store_16x1_avg_pixels(&res0, &res1, &max, dst_ptr);
-
- src_ptr += src_pitch;
- dst_ptr += dst_pitch;
- height -= 1;
- } while (height > 0);
-}
-
-static void aom_highbd_filter_block1d8_v2_avg_avx2(
- const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
- ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
- __m128i signal[3], res0, res1;
- const __m128i max = _mm_set1_epi16((1 << bd) - 1);
- __m128i ff;
-
- pack_8x1_2t_filter(filter, &ff);
- pack_8x2_init(src_ptr, signal);
-
- do {
- pack_8x2_2t_pixels_ver(src_ptr, src_pitch, signal);
- filter_8_2t_pixels(signal, &ff, &res0, &res1);
- store_8x1_2t_avg_pixels_ver(&res0, &res1, &max, dst_ptr);
-
- src_ptr += src_pitch;
- dst_ptr += dst_pitch;
- height -= 1;
- } while (height > 0);
-}
-
void aom_highbd_filter_block1d4_h8_sse2(const uint16_t *, ptrdiff_t, uint16_t *,
ptrdiff_t, uint32_t, const int16_t *,
int);
@@ -1331,30 +1087,4 @@
HIGH_FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , avx2);
HIGH_FUN_CONV_2D(, avx2);
-void aom_highbd_filter_block1d4_h8_avg_sse2(const uint16_t *, ptrdiff_t,
- uint16_t *, ptrdiff_t, uint32_t,
- const int16_t *, int);
-void aom_highbd_filter_block1d4_h2_avg_sse2(const uint16_t *, ptrdiff_t,
- uint16_t *, ptrdiff_t, uint32_t,
- const int16_t *, int);
-void aom_highbd_filter_block1d4_v8_avg_sse2(const uint16_t *, ptrdiff_t,
- uint16_t *, ptrdiff_t, uint32_t,
- const int16_t *, int);
-void aom_highbd_filter_block1d4_v2_avg_sse2(const uint16_t *, ptrdiff_t,
- uint16_t *, ptrdiff_t, uint32_t,
- const int16_t *, int);
-#define aom_highbd_filter_block1d4_h8_avg_avx2 \
- aom_highbd_filter_block1d4_h8_avg_sse2
-#define aom_highbd_filter_block1d4_h2_avg_avx2 \
- aom_highbd_filter_block1d4_h2_avg_sse2
-#define aom_highbd_filter_block1d4_v8_avg_avx2 \
- aom_highbd_filter_block1d4_v8_avg_sse2
-#define aom_highbd_filter_block1d4_v2_avg_avx2 \
- aom_highbd_filter_block1d4_v2_avg_sse2
-
-HIGH_FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, avx2);
-HIGH_FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_,
- avx2);
-HIGH_FUN_CONV_2D(avg_, avx2);
-
#undef HIGHBD_FUNC
diff --git a/av1/common/av1_rtcd_defs.pl b/av1/common/av1_rtcd_defs.pl
index d0dffdd..c3e2fdc 100755
--- a/av1/common/av1_rtcd_defs.pl
+++ b/av1/common/av1_rtcd_defs.pl
@@ -164,15 +164,6 @@
add_proto qw/void av1_highbd_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
specialize qw/av1_highbd_convolve8_vert/, "$sse2_x86_64";
-add_proto qw/void av1_highbd_convolve8_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
-specialize qw/av1_highbd_convolve8_avg/, "$sse2_x86_64";
-
-add_proto qw/void av1_highbd_convolve8_avg_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
-specialize qw/av1_highbd_convolve8_avg_horiz/, "$sse2_x86_64";
-
-add_proto qw/void av1_highbd_convolve8_avg_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
-specialize qw/av1_highbd_convolve8_avg_vert/, "$sse2_x86_64";
-
#
# dct
#
diff --git a/test/convolve_test.cc b/test/convolve_test.cc
index 103bfb9..52df913 100644
--- a/test/convolve_test.cc
+++ b/test/convolve_test.cc
@@ -38,19 +38,14 @@
struct ConvolveFunctions {
ConvolveFunctions(ConvolveFunc copy, ConvolveFunc avg, ConvolveFunc h8,
- ConvolveFunc h8_avg, ConvolveFunc v8, ConvolveFunc v8_avg,
- ConvolveFunc hv8, ConvolveFunc hv8_avg, int bd)
- : copy_(copy), avg_(avg), h8_(h8), v8_(v8), hv8_(hv8), h8_avg_(h8_avg),
- v8_avg_(v8_avg), hv8_avg_(hv8_avg), use_highbd_(bd) {}
+ ConvolveFunc v8, ConvolveFunc hv8, int bd)
+ : copy_(copy), avg_(avg), h8_(h8), v8_(v8), hv8_(hv8), use_highbd_(bd) {}
ConvolveFunc copy_;
ConvolveFunc avg_;
ConvolveFunc h8_;
ConvolveFunc v8_;
ConvolveFunc hv8_;
- ConvolveFunc h8_avg_;
- ConvolveFunc v8_avg_;
- ConvolveFunc hv8_avg_;
int use_highbd_; // 0 if high bitdepth not used, else the actual bit depth.
};
@@ -600,78 +595,6 @@
}
}
-TEST_P(ConvolveTest, MatchesReferenceAveragingSubpixelFilter) {
- uint8_t *const in = input();
- uint8_t *const out = output();
- uint8_t ref8[kOutputStride * kMaxDimension];
- uint16_t ref16[kOutputStride * kMaxDimension];
- uint8_t *ref;
- if (UUT_->use_highbd_ == 0) {
- ref = ref8;
- } else {
- ref = CONVERT_TO_BYTEPTR(ref16);
- }
-
- // Populate ref and out with some random data
- ::libaom_test::ACMRandom prng;
- for (int y = 0; y < Height(); ++y) {
- for (int x = 0; x < Width(); ++x) {
- uint16_t r;
- if (UUT_->use_highbd_ == 0 || UUT_->use_highbd_ == 8) {
- r = prng.Rand8Extremes();
- } else {
- r = prng.Rand16() & mask_;
- }
- assign_val(out, y * kOutputStride + x, r);
- assign_val(ref, y * kOutputStride + x, r);
- }
- }
-
- for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
- const InterpFilter filter = (InterpFilter)filter_bank;
- const InterpKernel *filters =
- (const InterpKernel *)av1_get_interp_filter_kernel(filter);
- const InterpFilterParams filter_params =
- av1_get_interp_filter_params(filter);
- if (filter_params.taps != SUBPEL_TAPS) continue;
-
- for (int filter_x = 0; filter_x < kNumFilters; ++filter_x) {
- for (int filter_y = 0; filter_y < kNumFilters; ++filter_y) {
- wrapper_filter_average_block2d_8_c(in, kInputStride, filters[filter_x],
- filters[filter_y], ref,
- kOutputStride, Width(), Height());
-
- if (filter_x && filter_y)
- ASM_REGISTER_STATE_CHECK(UUT_->hv8_avg_(
- in, kInputStride, out, kOutputStride, filters[filter_x], 16,
- filters[filter_y], 16, Width(), Height()));
- else if (filter_y)
- ASM_REGISTER_STATE_CHECK(UUT_->v8_avg_(
- in, kInputStride, out, kOutputStride, kInvalidFilter, 16,
- filters[filter_y], 16, Width(), Height()));
- else if (filter_x)
- ASM_REGISTER_STATE_CHECK(UUT_->h8_avg_(
- in, kInputStride, out, kOutputStride, filters[filter_x], 16,
- kInvalidFilter, 16, Width(), Height()));
- else
- ASM_REGISTER_STATE_CHECK(
- UUT_->avg_(in, kInputStride, out, kOutputStride, kInvalidFilter,
- 0, kInvalidFilter, 0, Width(), Height()));
-
- CheckGuardBlocks();
-
- for (int y = 0; y < Height(); ++y)
- for (int x = 0; x < Width(); ++x)
- ASSERT_EQ(lookup(ref, y * kOutputStride + x),
- lookup(out, y * kOutputStride + x))
- << "mismatch at (" << x << "," << y << "), "
- << "filters (" << filter_bank << "," << filter_x << ","
- << filter_y << ")";
- }
- }
- }
-}
-
TEST_P(ConvolveTest, FilterExtremes) {
uint8_t *const in = input();
uint8_t *const out = output();
@@ -874,7 +797,7 @@
using std::tr1::make_tuple;
#define WRAP(func, bd) \
- void wrap_##func##_##bd( \
+ static void wrap_##func##_##bd( \
const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, \
ptrdiff_t dst_stride, const int16_t *filter_x, int filter_x_stride, \
const int16_t *filter_y, int filter_y_stride, int w, int h) { \
@@ -889,95 +812,69 @@
WRAP(convolve_copy_sse2, 12)
WRAP(convolve_avg_sse2, 12)
WRAP(convolve8_horiz_sse2, 8)
-WRAP(convolve8_avg_horiz_sse2, 8)
WRAP(convolve8_vert_sse2, 8)
-WRAP(convolve8_avg_vert_sse2, 8)
WRAP(convolve8_sse2, 8)
-WRAP(convolve8_avg_sse2, 8)
WRAP(convolve8_horiz_sse2, 10)
-WRAP(convolve8_avg_horiz_sse2, 10)
WRAP(convolve8_vert_sse2, 10)
-WRAP(convolve8_avg_vert_sse2, 10)
WRAP(convolve8_sse2, 10)
-WRAP(convolve8_avg_sse2, 10)
WRAP(convolve8_horiz_sse2, 12)
-WRAP(convolve8_avg_horiz_sse2, 12)
WRAP(convolve8_vert_sse2, 12)
-WRAP(convolve8_avg_vert_sse2, 12)
WRAP(convolve8_sse2, 12)
-WRAP(convolve8_avg_sse2, 12)
#endif // HAVE_SSE2 && ARCH_X86_64
WRAP(convolve_copy_c, 8)
WRAP(convolve_avg_c, 8)
WRAP(convolve8_horiz_c, 8)
-WRAP(convolve8_avg_horiz_c, 8)
WRAP(convolve8_vert_c, 8)
-WRAP(convolve8_avg_vert_c, 8)
WRAP(convolve8_c, 8)
-WRAP(convolve8_avg_c, 8)
WRAP(convolve_copy_c, 10)
WRAP(convolve_avg_c, 10)
WRAP(convolve8_horiz_c, 10)
-WRAP(convolve8_avg_horiz_c, 10)
WRAP(convolve8_vert_c, 10)
-WRAP(convolve8_avg_vert_c, 10)
WRAP(convolve8_c, 10)
-WRAP(convolve8_avg_c, 10)
WRAP(convolve_copy_c, 12)
WRAP(convolve_avg_c, 12)
WRAP(convolve8_horiz_c, 12)
-WRAP(convolve8_avg_horiz_c, 12)
WRAP(convolve8_vert_c, 12)
-WRAP(convolve8_avg_vert_c, 12)
WRAP(convolve8_c, 12)
-WRAP(convolve8_avg_c, 12)
#if HAVE_AVX2
WRAP(convolve_copy_avx2, 8)
WRAP(convolve_avg_avx2, 8)
WRAP(convolve8_horiz_avx2, 8)
-WRAP(convolve8_avg_horiz_avx2, 8)
WRAP(convolve8_vert_avx2, 8)
-WRAP(convolve8_avg_vert_avx2, 8)
WRAP(convolve8_avx2, 8)
-WRAP(convolve8_avg_avx2, 8)
WRAP(convolve_copy_avx2, 10)
WRAP(convolve_avg_avx2, 10)
WRAP(convolve8_avx2, 10)
WRAP(convolve8_horiz_avx2, 10)
WRAP(convolve8_vert_avx2, 10)
-WRAP(convolve8_avg_avx2, 10)
-WRAP(convolve8_avg_horiz_avx2, 10)
-WRAP(convolve8_avg_vert_avx2, 10)
WRAP(convolve_copy_avx2, 12)
WRAP(convolve_avg_avx2, 12)
WRAP(convolve8_avx2, 12)
WRAP(convolve8_horiz_avx2, 12)
WRAP(convolve8_vert_avx2, 12)
-WRAP(convolve8_avg_avx2, 12)
-WRAP(convolve8_avg_horiz_avx2, 12)
-WRAP(convolve8_avg_vert_avx2, 12)
#endif // HAVE_AVX2
#undef WRAP
-const ConvolveFunctions convolve8_c(
- wrap_convolve_copy_c_8, wrap_convolve_avg_c_8, wrap_convolve8_horiz_c_8,
- wrap_convolve8_avg_horiz_c_8, wrap_convolve8_vert_c_8,
- wrap_convolve8_avg_vert_c_8, wrap_convolve8_c_8, wrap_convolve8_avg_c_8, 8);
-const ConvolveFunctions convolve10_c(
- wrap_convolve_copy_c_10, wrap_convolve_avg_c_10, wrap_convolve8_horiz_c_10,
- wrap_convolve8_avg_horiz_c_10, wrap_convolve8_vert_c_10,
- wrap_convolve8_avg_vert_c_10, wrap_convolve8_c_10, wrap_convolve8_avg_c_10,
- 10);
-const ConvolveFunctions convolve12_c(
- wrap_convolve_copy_c_12, wrap_convolve_avg_c_12, wrap_convolve8_horiz_c_12,
- wrap_convolve8_avg_horiz_c_12, wrap_convolve8_vert_c_12,
- wrap_convolve8_avg_vert_c_12, wrap_convolve8_c_12, wrap_convolve8_avg_c_12,
- 12);
+const ConvolveFunctions convolve8_c(wrap_convolve_copy_c_8,
+ wrap_convolve_avg_c_8,
+ wrap_convolve8_horiz_c_8,
+ wrap_convolve8_vert_c_8, wrap_convolve8_c_8,
+ 8);
+const ConvolveFunctions convolve10_c(wrap_convolve_copy_c_10,
+ wrap_convolve_avg_c_10,
+ wrap_convolve8_horiz_c_10,
+ wrap_convolve8_vert_c_10,
+ wrap_convolve8_c_10, 10);
+const ConvolveFunctions convolve12_c(wrap_convolve_copy_c_12,
+ wrap_convolve_avg_c_12,
+ wrap_convolve8_horiz_c_12,
+ wrap_convolve8_vert_c_12,
+ wrap_convolve8_c_12, 12);
const ConvolveParam kArrayConvolve_c[] = {
ALL_SIZES(convolve8_c), ALL_SIZES(convolve10_c), ALL_SIZES(convolve12_c)
};
@@ -985,21 +882,21 @@
INSTANTIATE_TEST_CASE_P(C, ConvolveTest, ::testing::ValuesIn(kArrayConvolve_c));
#if HAVE_SSE2 && ARCH_X86_64
-const ConvolveFunctions convolve8_sse2(
- wrap_convolve_copy_sse2_8, wrap_convolve_avg_sse2_8,
- wrap_convolve8_horiz_sse2_8, wrap_convolve8_avg_horiz_sse2_8,
- wrap_convolve8_vert_sse2_8, wrap_convolve8_avg_vert_sse2_8,
- wrap_convolve8_sse2_8, wrap_convolve8_avg_sse2_8, 8);
-const ConvolveFunctions convolve10_sse2(
- wrap_convolve_copy_sse2_10, wrap_convolve_avg_sse2_10,
- wrap_convolve8_horiz_sse2_10, wrap_convolve8_avg_horiz_sse2_10,
- wrap_convolve8_vert_sse2_10, wrap_convolve8_avg_vert_sse2_10,
- wrap_convolve8_sse2_10, wrap_convolve8_avg_sse2_10, 10);
-const ConvolveFunctions convolve12_sse2(
- wrap_convolve_copy_sse2_12, wrap_convolve_avg_sse2_12,
- wrap_convolve8_horiz_sse2_12, wrap_convolve8_avg_horiz_sse2_12,
- wrap_convolve8_vert_sse2_12, wrap_convolve8_avg_vert_sse2_12,
- wrap_convolve8_sse2_12, wrap_convolve8_avg_sse2_12, 12);
+const ConvolveFunctions convolve8_sse2(wrap_convolve_copy_sse2_8,
+ wrap_convolve_avg_sse2_8,
+ wrap_convolve8_horiz_sse2_8,
+ wrap_convolve8_vert_sse2_8,
+ wrap_convolve8_sse2_8, 8);
+const ConvolveFunctions convolve10_sse2(wrap_convolve_copy_sse2_10,
+ wrap_convolve_avg_sse2_10,
+ wrap_convolve8_horiz_sse2_10,
+ wrap_convolve8_vert_sse2_10,
+ wrap_convolve8_sse2_10, 10);
+const ConvolveFunctions convolve12_sse2(wrap_convolve_copy_sse2_12,
+ wrap_convolve_avg_sse2_12,
+ wrap_convolve8_horiz_sse2_12,
+ wrap_convolve8_vert_sse2_12,
+ wrap_convolve8_sse2_12, 12);
const ConvolveParam kArrayConvolve_sse2[] = { ALL_SIZES(convolve8_sse2),
ALL_SIZES(convolve10_sse2),
ALL_SIZES(convolve12_sse2) };
@@ -1010,11 +907,8 @@
#if HAVE_SSSE3
const ConvolveFunctions convolve8_ssse3(aom_convolve_copy_c, aom_convolve_avg_c,
aom_convolve8_horiz_ssse3,
- aom_convolve8_avg_horiz_ssse3,
aom_convolve8_vert_ssse3,
- aom_convolve8_avg_vert_ssse3,
- aom_convolve8_ssse3,
- aom_convolve8_avg_ssse3, 0);
+ aom_convolve8_ssse3, 0);
const ConvolveParam kArrayConvolve8_ssse3[] = { ALL_SIZES(convolve8_ssse3) };
INSTANTIATE_TEST_CASE_P(SSSE3, ConvolveTest,
@@ -1022,21 +916,21 @@
#endif
#if HAVE_AVX2
-const ConvolveFunctions convolve8_avx2(
- wrap_convolve_copy_avx2_8, wrap_convolve_avg_avx2_8,
- wrap_convolve8_horiz_avx2_8, wrap_convolve8_avg_horiz_avx2_8,
- wrap_convolve8_vert_avx2_8, wrap_convolve8_avg_vert_avx2_8,
- wrap_convolve8_avx2_8, wrap_convolve8_avg_avx2_8, 8);
-const ConvolveFunctions convolve10_avx2(
- wrap_convolve_copy_avx2_10, wrap_convolve_avg_avx2_10,
- wrap_convolve8_horiz_avx2_10, wrap_convolve8_avg_horiz_avx2_10,
- wrap_convolve8_vert_avx2_10, wrap_convolve8_avg_vert_avx2_10,
- wrap_convolve8_avx2_10, wrap_convolve8_avg_avx2_10, 10);
-const ConvolveFunctions convolve12_avx2(
- wrap_convolve_copy_avx2_12, wrap_convolve_avg_avx2_12,
- wrap_convolve8_horiz_avx2_12, wrap_convolve8_avg_horiz_avx2_12,
- wrap_convolve8_vert_avx2_12, wrap_convolve8_avg_vert_avx2_12,
- wrap_convolve8_avx2_12, wrap_convolve8_avg_avx2_12, 12);
+const ConvolveFunctions convolve8_avx2(wrap_convolve_copy_avx2_8,
+ wrap_convolve_avg_avx2_8,
+ wrap_convolve8_horiz_avx2_8,
+ wrap_convolve8_vert_avx2_8,
+ wrap_convolve8_avx2_8, 8);
+const ConvolveFunctions convolve10_avx2(wrap_convolve_copy_avx2_10,
+ wrap_convolve_avg_avx2_10,
+ wrap_convolve8_horiz_avx2_10,
+ wrap_convolve8_vert_avx2_10,
+ wrap_convolve8_avx2_10, 10);
+const ConvolveFunctions convolve12_avx2(wrap_convolve_copy_avx2_12,
+ wrap_convolve_avg_avx2_12,
+ wrap_convolve8_horiz_avx2_12,
+ wrap_convolve8_vert_avx2_12,
+ wrap_convolve8_avx2_12, 12);
const ConvolveParam kArrayConvolve8_avx2[] = { ALL_SIZES_64(convolve8_avx2),
ALL_SIZES_64(convolve10_avx2),
ALL_SIZES_64(convolve12_avx2) };
@@ -1047,15 +941,17 @@
// TODO(any): Make NEON versions support 128x128 128x64 64x128 block sizes
#if HAVE_NEON && !(CONFIG_AV1)
#if HAVE_NEON_ASM
-const ConvolveFunctions convolve8_neon(
- aom_convolve_copy_neon, aom_convolve_avg_neon, aom_convolve8_horiz_neon,
- aom_convolve8_avg_horiz_neon, aom_convolve8_vert_neon,
- aom_convolve8_avg_vert_neon, aom_convolve8_neon, aom_convolve8_avg_neon, 0);
+const ConvolveFunctions convolve8_neon(aom_convolve_copy_neon,
+ aom_convolve_avg_neon,
+ aom_convolve8_horiz_neon,
+ aom_convolve8_vert_neon,
+ aom_convolve8_neon, 0);
#else // HAVE_NEON
-const ConvolveFunctions convolve8_neon(
- aom_convolve_copy_neon, aom_convolve_avg_neon, aom_convolve8_horiz_neon,
- aom_convolve8_avg_horiz_neon, aom_convolve8_vert_neon,
- aom_convolve8_avg_vert_neon, aom_convolve8_neon, aom_convolve8_avg_neon, 0);
+const ConvolveFunctions convolve8_neon(aom_convolve_copy_neon,
+ aom_convolve_avg_neon,
+ aom_convolve8_horiz_neon,
+ aom_convolve8_vert_neon,
+ aom_convolve8_neon, 0);
#endif // HAVE_NEON_ASM
const ConvolveParam kArrayConvolve8_neon[] = { ALL_SIZES_64(convolve8_neon) };
@@ -1065,11 +961,11 @@
// TODO(any): Make DSPR2 versions support 128x128 128x64 64x128 block sizes
#if HAVE_DSPR2 && !(CONFIG_AV1)
-const ConvolveFunctions convolve8_dspr2(
- aom_convolve_copy_dspr2, aom_convolve_avg_dspr2, aom_convolve8_horiz_dspr2,
- aom_convolve8_avg_horiz_dspr2, aom_convolve8_vert_dspr2,
- aom_convolve8_avg_vert_dspr2, aom_convolve8_dspr2, aom_convolve8_avg_dspr2,
- 0);
+const ConvolveFunctions convolve8_dspr2(aom_convolve_copy_dspr2,
+ aom_convolve_avg_dspr2,
+ aom_convolve8_horiz_dspr2,
+ aom_convolve8_vert_dspr2,
+ aom_convolve8_dspr2, 0);
const ConvolveParam kArrayConvolve8_dspr2[] = { ALL_SIZES_64(convolve8_dspr2) };
INSTANTIATE_TEST_CASE_P(DSPR2, ConvolveTest,
@@ -1078,10 +974,11 @@
// TODO(any): Make MSA versions support 128x128 128x64 64x128 block sizes
#if HAVE_MSA && !(CONFIG_AV1)
-const ConvolveFunctions convolve8_msa(
- aom_convolve_copy_msa, aom_convolve_avg_msa, aom_convolve8_horiz_msa,
- aom_convolve8_avg_horiz_msa, aom_convolve8_vert_msa,
- aom_convolve8_avg_vert_msa, aom_convolve8_msa, aom_convolve8_avg_msa, 0);
+const ConvolveFunctions convolve8_msa(aom_convolve_copy_msa,
+ aom_convolve_avg_msa,
+ aom_convolve8_horiz_msa,
+ aom_convolve8_vert_msa, aom_convolve8_msa,
+ 0);
const ConvolveParam kArrayConvolve8_msa[] = { ALL_SIZES_64(convolve8_msa) };
INSTANTIATE_TEST_CASE_P(MSA, ConvolveTest,
