Commit 37fdce8a authored by Steinar Midtskogen's avatar Steinar Midtskogen
Browse files

Add support for v256 intrinsics

Change-Id: I1da08afaa945ca1aaf4bf9f50cf649a7feef2e60
parent 1352bbb4
......@@ -373,9 +373,13 @@ DSP_SRCS-yes += simd/v64_intrinsics.h
DSP_SRCS-yes += simd/v64_intrinsics_c.h
DSP_SRCS-yes += simd/v128_intrinsics.h
DSP_SRCS-yes += simd/v128_intrinsics_c.h
DSP_SRCS-yes += simd/v256_intrinsics.h
DSP_SRCS-yes += simd/v256_intrinsics_c.h
DSP_SRCS-$(HAVE_SSE2) += simd/v64_intrinsics_x86.h
DSP_SRCS-$(HAVE_SSE2) += simd/v128_intrinsics_x86.h
DSP_SRCS-$(HAVE_SSE2) += simd/v256_intrinsics_x86.h
DSP_SRCS-$(HAVE_NEON) += simd/v64_intrinsics_arm.h
DSP_SRCS-$(HAVE_NEON) += simd/v128_intrinsics_arm.h
DSP_SRCS-$(HAVE_NEON) += simd/v256_intrinsics_arm.h
$(eval $(call rtcd_h_template,aom_dsp_rtcd,aom_dsp/aom_dsp_rtcd_defs.pl))
......@@ -22,11 +22,11 @@
#include "./aom_simd_inline.h"
#if HAVE_NEON
#include "simd/v128_intrinsics_arm.h"
#include "simd/v256_intrinsics_arm.h"
#elif HAVE_SSE2
#include "simd/v128_intrinsics_x86.h"
#include "simd/v256_intrinsics_x86.h"
#else
#include "simd/v128_intrinsics.h"
#include "simd/v256_intrinsics.h"
#endif
#endif // AOM_DSP_AOM_AOM_SIMD_H_
/*
* 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.
*/
#ifndef _V256_INTRINSICS_H
#define _V256_INTRINSICS_H
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "./v256_intrinsics_c.h"
#include "./v128_intrinsics.h"
#include "./v64_intrinsics.h"
/* Fallback to plain, unoptimised C. */
typedef c_v256 v256;
SIMD_INLINE uint32_t v256_low_u32(v256 a) { return c_v256_low_u32(a); }
SIMD_INLINE v64 v256_low_v64(v256 a) { return c_v256_low_v64(a); }
SIMD_INLINE v128 v256_low_v128(v256 a) { return c_v256_low_v128(a); }
SIMD_INLINE v128 v256_high_v128(v256 a) { return c_v256_high_v128(a); }
SIMD_INLINE v256 v256_from_v128(v128 hi, v128 lo) {
return c_v256_from_v128(hi, lo);
}
SIMD_INLINE v256 v256_from_64(uint64_t a, uint64_t b, uint64_t c, uint64_t d) {
return c_v256_from_64(a, b, c, d);
}
SIMD_INLINE v256 v256_from_v64(v64 a, v64 b, v64 c, v64 d) {
return c_v256_from_v64(a, b, c, d);
}
SIMD_INLINE v256 v256_load_unaligned(const void *p) {
return c_v256_load_unaligned(p);
}
SIMD_INLINE v256 v256_load_aligned(const void *p) {
return c_v256_load_aligned(p);
}
SIMD_INLINE void v256_store_unaligned(void *p, v256 a) {
c_v256_store_unaligned(p, a);
}
SIMD_INLINE void v256_store_aligned(void *p, v256 a) {
c_v256_store_aligned(p, a);
}
SIMD_INLINE v256 v256_align(v256 a, v256 b, const unsigned int c) {
return c_v256_align(a, b, c);
}
SIMD_INLINE v256 v256_zero() { return c_v256_zero(); }
SIMD_INLINE v256 v256_dup_8(uint8_t x) { return c_v256_dup_8(x); }
SIMD_INLINE v256 v256_dup_16(uint16_t x) { return c_v256_dup_16(x); }
SIMD_INLINE v256 v256_dup_32(uint32_t x) { return c_v256_dup_32(x); }
typedef uint32_t sad256_internal;
SIMD_INLINE sad256_internal v256_sad_u8_init() { return c_v256_sad_u8_init(); }
SIMD_INLINE sad256_internal v256_sad_u8(sad256_internal s, v256 a, v256 b) {
return c_v256_sad_u8(s, a, b);
}
SIMD_INLINE uint32_t v256_sad_u8_sum(sad256_internal s) {
return c_v256_sad_u8_sum(s);
}
typedef uint32_t ssd256_internal;
SIMD_INLINE ssd256_internal v256_ssd_u8_init() { return c_v256_ssd_u8_init(); }
SIMD_INLINE ssd256_internal v256_ssd_u8(ssd256_internal s, v256 a, v256 b) {
return c_v256_ssd_u8(s, a, b);
}
SIMD_INLINE uint32_t v256_ssd_u8_sum(ssd256_internal s) {
return c_v256_ssd_u8_sum(s);
}
SIMD_INLINE int64_t v256_dotp_s16(v256 a, v256 b) {
return c_v256_dotp_s16(a, b);
}
SIMD_INLINE uint64_t v256_hadd_u8(v256 a) { return c_v256_hadd_u8(a); }
SIMD_INLINE v256 v256_or(v256 a, v256 b) { return c_v256_or(a, b); }
SIMD_INLINE v256 v256_xor(v256 a, v256 b) { return c_v256_xor(a, b); }
SIMD_INLINE v256 v256_and(v256 a, v256 b) { return c_v256_and(a, b); }
SIMD_INLINE v256 v256_andn(v256 a, v256 b) { return c_v256_andn(a, b); }
SIMD_INLINE v256 v256_add_8(v256 a, v256 b) { return c_v256_add_8(a, b); }
SIMD_INLINE v256 v256_add_16(v256 a, v256 b) { return c_v256_add_16(a, b); }
SIMD_INLINE v256 v256_sadd_s16(v256 a, v256 b) { return c_v256_sadd_s16(a, b); }
SIMD_INLINE v256 v256_add_32(v256 a, v256 b) { return c_v256_add_32(a, b); }
SIMD_INLINE v256 v256_padd_s16(v256 a) { return c_v256_padd_s16(a); }
SIMD_INLINE v256 v256_sub_8(v256 a, v256 b) { return c_v256_sub_8(a, b); }
SIMD_INLINE v256 v256_ssub_u8(v256 a, v256 b) { return c_v256_ssub_u8(a, b); }
SIMD_INLINE v256 v256_ssub_s8(v256 a, v256 b) { return c_v256_ssub_s8(a, b); }
SIMD_INLINE v256 v256_sub_16(v256 a, v256 b) { return c_v256_sub_16(a, b); }
SIMD_INLINE v256 v256_ssub_s16(v256 a, v256 b) { return c_v256_ssub_s16(a, b); }
SIMD_INLINE v256 v256_sub_32(v256 a, v256 b) { return c_v256_sub_32(a, b); }
SIMD_INLINE v256 v256_abs_s16(v256 a) { return c_v256_abs_s16(a); }
SIMD_INLINE v256 v256_mul_s16(v128 a, v128 b) { return c_v256_mul_s16(a, b); }
SIMD_INLINE v256 v256_mullo_s16(v256 a, v256 b) {
return c_v256_mullo_s16(a, b);
}
SIMD_INLINE v256 v256_mulhi_s16(v256 a, v256 b) {
return c_v256_mulhi_s16(a, b);
}
SIMD_INLINE v256 v256_mullo_s32(v256 a, v256 b) {
return c_v256_mullo_s32(a, b);
}
SIMD_INLINE v256 v256_madd_s16(v256 a, v256 b) { return c_v256_madd_s16(a, b); }
SIMD_INLINE v256 v256_madd_us8(v256 a, v256 b) { return c_v256_madd_us8(a, b); }
SIMD_INLINE v256 v256_avg_u8(v256 a, v256 b) { return c_v256_avg_u8(a, b); }
SIMD_INLINE v256 v256_rdavg_u8(v256 a, v256 b) { return c_v256_rdavg_u8(a, b); }
SIMD_INLINE v256 v256_avg_u16(v256 a, v256 b) { return c_v256_avg_u16(a, b); }
SIMD_INLINE v256 v256_min_u8(v256 a, v256 b) { return c_v256_min_u8(a, b); }
SIMD_INLINE v256 v256_max_u8(v256 a, v256 b) { return c_v256_max_u8(a, b); }
SIMD_INLINE v256 v256_min_s8(v256 a, v256 b) { return c_v256_min_s8(a, b); }
SIMD_INLINE v256 v256_max_s8(v256 a, v256 b) { return c_v256_max_s8(a, b); }
SIMD_INLINE v256 v256_min_s16(v256 a, v256 b) { return c_v256_min_s16(a, b); }
SIMD_INLINE v256 v256_max_s16(v256 a, v256 b) { return c_v256_max_s16(a, b); }
SIMD_INLINE v256 v256_ziplo_8(v256 a, v256 b) { return c_v256_ziplo_8(a, b); }
SIMD_INLINE v256 v256_ziphi_8(v256 a, v256 b) { return c_v256_ziphi_8(a, b); }
SIMD_INLINE v256 v256_ziplo_16(v256 a, v256 b) { return c_v256_ziplo_16(a, b); }
SIMD_INLINE v256 v256_ziphi_16(v256 a, v256 b) { return c_v256_ziphi_16(a, b); }
SIMD_INLINE v256 v256_ziplo_32(v256 a, v256 b) { return c_v256_ziplo_32(a, b); }
SIMD_INLINE v256 v256_ziphi_32(v256 a, v256 b) { return c_v256_ziphi_32(a, b); }
SIMD_INLINE v256 v256_ziplo_64(v256 a, v256 b) { return c_v256_ziplo_64(a, b); }
SIMD_INLINE v256 v256_ziphi_64(v256 a, v256 b) { return c_v256_ziphi_64(a, b); }
SIMD_INLINE v256 v256_ziplo_128(v256 a, v256 b) {
return c_v256_ziplo_128(a, b);
}
SIMD_INLINE v256 v256_ziphi_128(v256 a, v256 b) {
return c_v256_ziphi_128(a, b);
}
SIMD_INLINE v256 v256_zip_8(v128 a, v128 b) { return c_v256_zip_8(a, b); }
SIMD_INLINE v256 v256_zip_16(v128 a, v128 b) { return c_v256_zip_16(a, b); }
SIMD_INLINE v256 v256_zip_32(v128 a, v128 b) { return c_v256_zip_32(a, b); }
SIMD_INLINE v256 v256_unziplo_8(v256 a, v256 b) {
return c_v256_unziplo_8(a, b);
}
SIMD_INLINE v256 v256_unziphi_8(v256 a, v256 b) {
return c_v256_unziphi_8(a, b);
}
SIMD_INLINE v256 v256_unziplo_16(v256 a, v256 b) {
return c_v256_unziplo_16(a, b);
}
SIMD_INLINE v256 v256_unziphi_16(v256 a, v256 b) {
return c_v256_unziphi_16(a, b);
}
SIMD_INLINE v256 v256_unziplo_32(v256 a, v256 b) {
return c_v256_unziplo_32(a, b);
}
SIMD_INLINE v256 v256_unziphi_32(v256 a, v256 b) {
return c_v256_unziphi_32(a, b);
}
SIMD_INLINE v256 v256_unpack_u8_s16(v128 a) { return c_v256_unpack_u8_s16(a); }
SIMD_INLINE v256 v256_unpacklo_u8_s16(v256 a) {
return c_v256_unpacklo_u8_s16(a);
}
SIMD_INLINE v256 v256_unpackhi_u8_s16(v256 a) {
return c_v256_unpackhi_u8_s16(a);
}
SIMD_INLINE v256 v256_pack_s32_s16(v256 a, v256 b) {
return c_v256_pack_s32_s16(a, b);
}
SIMD_INLINE v256 v256_pack_s16_u8(v256 a, v256 b) {
return c_v256_pack_s16_u8(a, b);
}
SIMD_INLINE v256 v256_pack_s16_s8(v256 a, v256 b) {
return c_v256_pack_s16_s8(a, b);
}
SIMD_INLINE v256 v256_unpack_u16_s32(v128 a) {
return c_v256_unpack_u16_s32(a);
}
SIMD_INLINE v256 v256_unpack_s16_s32(v128 a) {
return c_v256_unpack_s16_s32(a);
}
SIMD_INLINE v256 v256_unpacklo_u16_s32(v256 a) {
return c_v256_unpacklo_u16_s32(a);
}
SIMD_INLINE v256 v256_unpacklo_s16_s32(v256 a) {
return c_v256_unpacklo_s16_s32(a);
}
SIMD_INLINE v256 v256_unpackhi_u16_s32(v256 a) {
return c_v256_unpackhi_u16_s32(a);
}
SIMD_INLINE v256 v256_unpackhi_s16_s32(v256 a) {
return c_v256_unpackhi_s16_s32(a);
}
SIMD_INLINE v256 v256_shuffle_8(v256 a, v256 pattern) {
return c_v256_shuffle_8(a, pattern);
}
SIMD_INLINE v256 v256_pshuffle_8(v256 a, v256 pattern) {
return c_v256_pshuffle_8(a, pattern);
}
SIMD_INLINE v256 v256_cmpgt_s8(v256 a, v256 b) { return c_v256_cmpgt_s8(a, b); }
SIMD_INLINE v256 v256_cmplt_s8(v256 a, v256 b) { return c_v256_cmplt_s8(a, b); }
SIMD_INLINE v256 v256_cmpeq_8(v256 a, v256 b) { return c_v256_cmpeq_8(a, b); }
SIMD_INLINE v256 v256_cmpgt_s16(v256 a, v256 b) {
return c_v256_cmpgt_s16(a, b);
}
SIMD_INLINE v256 v256_cmplt_s16(v256 a, v256 b) {
return c_v256_cmplt_s16(a, b);
}
SIMD_INLINE v256 v256_cmpeq_16(v256 a, v256 b) { return c_v256_cmpeq_16(a, b); }
SIMD_INLINE v256 v256_shl_8(v256 a, unsigned int c) {
return c_v256_shl_8(a, c);
}
SIMD_INLINE v256 v256_shr_u8(v256 a, unsigned int c) {
return c_v256_shr_u8(a, c);
}
SIMD_INLINE v256 v256_shr_s8(v256 a, unsigned int c) {
return c_v256_shr_s8(a, c);
}
SIMD_INLINE v256 v256_shl_16(v256 a, unsigned int c) {
return c_v256_shl_16(a, c);
}
SIMD_INLINE v256 v256_shr_u16(v256 a, unsigned int c) {
return c_v256_shr_u16(a, c);
}
SIMD_INLINE v256 v256_shr_s16(v256 a, unsigned int c) {
return c_v256_shr_s16(a, c);
}
SIMD_INLINE v256 v256_shl_32(v256 a, unsigned int c) {
return c_v256_shl_32(a, c);
}
SIMD_INLINE v256 v256_shr_u32(v256 a, unsigned int c) {
return c_v256_shr_u32(a, c);
}
SIMD_INLINE v256 v256_shr_s32(v256 a, unsigned int c) {
return c_v256_shr_s32(a, c);
}
SIMD_INLINE v256 v256_shr_n_byte(v256 a, const unsigned int n) {
return c_v256_shr_n_byte(a, n);
}
SIMD_INLINE v256 v256_shl_n_byte(v256 a, const unsigned int n) {
return c_v256_shl_n_byte(a, n);
}
SIMD_INLINE v256 v256_shl_n_8(v256 a, const unsigned int n) {
return c_v256_shl_n_8(a, n);
}
SIMD_INLINE v256 v256_shl_n_16(v256 a, const unsigned int n) {
return c_v256_shl_n_16(a, n);
}
SIMD_INLINE v256 v256_shl_n_32(v256 a, const unsigned int n) {
return c_v256_shl_n_32(a, n);
}
SIMD_INLINE v256 v256_shr_n_u8(v256 a, const unsigned int n) {
return c_v256_shr_n_u8(a, n);
}
SIMD_INLINE v256 v256_shr_n_u16(v256 a, const unsigned int n) {
return c_v256_shr_n_u16(a, n);
}
SIMD_INLINE v256 v256_shr_n_u32(v256 a, const unsigned int n) {
return c_v256_shr_n_u32(a, n);
}
SIMD_INLINE v256 v256_shr_n_s8(v256 a, const unsigned int n) {
return c_v256_shr_n_s8(a, n);
}
SIMD_INLINE v256 v256_shr_n_s16(v256 a, const unsigned int n) {
return c_v256_shr_n_s16(a, n);
}
SIMD_INLINE v256 v256_shr_n_s32(v256 a, const unsigned int n) {
return c_v256_shr_n_s32(a, n);
}
#endif /* _V256_INTRINSICS_H */
/*
* 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.
*/
#ifndef _V256_INTRINSICS_H
#define _V256_INTRINSICS_H
#include "./v256_intrinsics_v128.h"
#endif /* _V256_INTRINSICS_H */
This diff is collapsed.
/*
* 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.
*/
#ifndef _V256_INTRINSICS_V128_H
#define _V256_INTRINSICS_V128_H
#if HAVE_NEON
#include "./v128_intrinsics_arm.h"
#elif HAVE_SSE2
#include "./v128_intrinsics_x86.h"
#else
#include "./v128_intrinsics.h"
#endif
typedef struct { v128 lo, hi; } v256;
SIMD_INLINE uint32_t v256_low_u32(v256 a) { return v128_low_u32(a.lo); }
SIMD_INLINE v64 v256_low_v64(v256 a) { return v128_low_v64(a.lo); }
SIMD_INLINE v128 v256_low_v128(v256 a) { return a.lo; }
SIMD_INLINE v128 v256_high_v128(v256 a) { return a.hi; }
SIMD_INLINE v256 v256_from_v128(v128 hi, v128 lo) {
v256 t;
t.hi = hi;
t.lo = lo;
return t;
}
SIMD_INLINE v256 v256_from_64(uint64_t a, uint64_t b, uint64_t c, uint64_t d) {
return v256_from_v128(v128_from_64(a, b), v128_from_64(c, d));
}
SIMD_INLINE v256 v256_from_v64(v64 a, v64 b, v64 c, v64 d) {
return v256_from_v128(v128_from_v64(a, b), v128_from_v64(c, d));
}
SIMD_INLINE v256 v256_load_unaligned(const void *p) {
return v256_from_v128(v128_load_unaligned((uint8_t *)p + 16),
v128_load_unaligned(p));
}
SIMD_INLINE v256 v256_load_aligned(const void *p) {
return v256_from_v128(v128_load_aligned((uint8_t *)p + 16),
v128_load_aligned(p));
}
SIMD_INLINE void v256_store_unaligned(void *p, v256 a) {
v128_store_unaligned(p, a.lo);
v128_store_unaligned((uint8_t *)p + 16, a.hi);
}
SIMD_INLINE void v256_store_aligned(void *p, v256 a) {
v128_store_aligned(p, a.lo);
v128_store_aligned((uint8_t *)p + 16, a.hi);
}
SIMD_INLINE v256 v256_zero() {
return v256_from_v128(v128_zero(), v128_zero());
}
SIMD_INLINE v256 v256_dup_8(uint8_t x) {
v128 t = v128_dup_8(x);
return v256_from_v128(t, t);
}
SIMD_INLINE v256 v256_dup_16(uint16_t x) {
v128 t = v128_dup_16(x);
return v256_from_v128(t, t);
}
SIMD_INLINE v256 v256_dup_32(uint32_t x) {
v128 t = v128_dup_32(x);
return v256_from_v128(t, t);
}
SIMD_INLINE int64_t v256_dotp_s16(v256 a, v256 b) {
return v128_dotp_s16(a.hi, b.hi) + v128_dotp_s16(a.lo, b.lo);
}
SIMD_INLINE uint64_t v256_hadd_u8(v256 a) {
return v128_hadd_u8(a.hi) + v128_hadd_u8(a.lo);
}
typedef struct {
sad128_internal hi;
sad128_internal lo;
} sad256_internal;
SIMD_INLINE sad256_internal v256_sad_u8_init() {
sad256_internal t;
t.hi = v128_sad_u8_init();
t.lo = v128_sad_u8_init();
return t;
}
/* Implementation dependent return value. Result must be finalised with
v256_sad_u8_sum().
The result for more than 16 v256_sad_u8() calls is undefined. */
SIMD_INLINE sad256_internal v256_sad_u8(sad256_internal s, v256 a, v256 b) {
sad256_internal t;
t.hi = v128_sad_u8(s.hi, a.hi, b.hi);
t.lo = v128_sad_u8(s.lo, a.lo, b.lo);
return t;
}
SIMD_INLINE uint32_t v256_sad_u8_sum(sad256_internal s) {
return v128_sad_u8_sum(s.hi) + v128_sad_u8_sum(s.lo);
}
typedef struct {
ssd128_internal hi;
ssd128_internal lo;
} ssd256_internal;
SIMD_INLINE ssd256_internal v256_ssd_u8_init() {
ssd256_internal t;
t.hi = v128_ssd_u8_init();
t.lo = v128_ssd_u8_init();
return t;
}
/* Implementation dependent return value. Result must be finalised with
* v256_ssd_u8_sum(). */
SIMD_INLINE ssd256_internal v256_ssd_u8(ssd256_internal s, v256 a, v256 b) {
ssd256_internal t;
t.hi = v128_ssd_u8(s.hi, a.hi, b.hi);
t.lo = v128_ssd_u8(s.lo, a.lo, b.lo);
return t;
}
SIMD_INLINE uint32_t v256_ssd_u8_sum(ssd256_internal s) {
return v128_ssd_u8_sum(s.hi) + v128_ssd_u8_sum(s.lo);
}
SIMD_INLINE v256 v256_or(v256 a, v256 b) {
return v256_from_v128(v128_or(a.hi, b.hi), v128_or(a.lo, b.lo));
}
SIMD_INLINE v256 v256_xor(v256 a, v256 b) {
return v256_from_v128(v128_xor(a.hi, b.hi), v128_xor(a.lo, b.lo));
}
SIMD_INLINE v256 v256_and(v256 a, v256 b) {
return v256_from_v128(v128_and(a.hi, b.hi), v128_and(a.lo, b.lo));
}
SIMD_INLINE v256 v256_andn(v256 a, v256 b) {
return v256_from_v128(v128_andn(a.hi, b.hi), v128_andn(a.lo, b.lo));
}
SIMD_INLINE v256 v256_add_8(v256 a, v256 b) {
return v256_from_v128(v128_add_8(a.hi, b.hi), v128_add_8(a.lo, b.lo));
}
SIMD_INLINE v256 v256_add_16(v256 a, v256 b) {
return v256_from_v128(v128_add_16(a.hi, b.hi), v128_add_16(a.lo, b.lo));
}
SIMD_INLINE v256 v256_sadd_s16(v256 a, v256 b) {
return v256_from_v128(v128_sadd_s16(a.hi, b.hi), v128_sadd_s16(a.lo, b.lo));
}
SIMD_INLINE v256 v256_add_32(v256 a, v256 b) {
return v256_from_v128(v128_add_32(a.hi, b.hi), v128_add_32(a.lo, b.lo));
}
SIMD_INLINE v256 v256_padd_s16(v256 a) {
return v256_from_v128(v128_padd_s16(a.hi), v128_padd_s16(a.lo));
}
SIMD_INLINE v256 v256_sub_8(v256 a, v256 b) {
return v256_from_v128(v128_sub_8(a.hi, b.hi), v128_sub_8(a.lo, b.lo));
}
SIMD_INLINE v256 v256_ssub_u8(v256 a, v256 b) {
return v256_from_v128(v128_ssub_u8(a.hi, b.hi), v128_ssub_u8(a.lo, b.lo));
}
SIMD_INLINE v256 v256_ssub_s8(v256 a, v256 b) {
return v256_from_v128(v128_ssub_s8(a.hi, b.hi), v128_ssub_s8(a.lo, b.lo));
}
SIMD_INLINE v256 v256_sub_16(v256 a, v256 b) {
return v256_from_v128(v128_sub_16(a.hi, b.hi), v128_sub_16(a.lo, b.lo));
}
SIMD_INLINE v256 v256_ssub_s16(v256 a, v256 b) {
return v256_from_v128(v128_ssub_s16(a.hi, b.hi), v128_ssub_s16(a.lo, b.lo));
}
SIMD_INLINE v256 v256_sub_32(v256 a, v256 b) {
return v256_from_v128(v128_sub_32(a.hi, b.hi), v128_sub_32(a.lo, b.lo));
}
SIMD_INLINE v256 v256_abs_s16(v256 a) {
return v256_from_v128(v128_abs_s16(a.hi), v128_abs_s16(a.lo));
}
SIMD_INLINE v256 v256_mul_s16(v128 a, v128 b) {
v128 lo_bits = v128_mullo_s16(a, b);
v128 hi_bits = v128_mulhi_s16(a, b);
return v256_from_v128(v128_ziphi_16(hi_bits, lo_bits),
v128_ziplo_16(hi_bits, lo_bits));
}
SIMD_INLINE v256 v256_mullo_s16(v256 a, v256 b) {
return v256_from_v128(v128_mullo_s16(a.hi, b.hi), v128_mullo_s16(a.lo, b.lo));
}
SIMD_INLINE v256 v256_mulhi_s16(v256 a, v256 b) {
return v256_from_v128(v128_mulhi_s16(a.hi, b.hi), v128_mulhi_s16(a.lo, b.lo));
}
SIMD_INLINE v256 v256_mullo_s32(v256 a, v256 b) {
return v256_from_v128(v128_mullo_s32(a.hi, b.hi), v128_mullo_s32(a.lo, b.lo));
}
SIMD_INLINE v256 v256_madd_s16(v256 a, v256 b) {
return v256_from_v128(v128_madd_s16(a.hi, b.hi), v128_madd_s16(a.lo, b.lo));
}
SIMD_INLINE v256 v256_madd_us8(v256 a, v256 b) {
return v256_from_v128(v128_madd_us8(a.hi, b.hi), v128_madd_us8(a.lo, b.lo));
}
SIMD_INLINE v256 v256_avg_u8(v256 a, v256 b) {
return v256_from_v128(v128_avg_u8(a.hi, b.hi), v128_avg_u8(a.lo, b.lo));
}
SIMD_INLINE v256 v256_rdavg_u8(v256 a, v256 b) {
return v256_from_v128(v128_rdavg_u8(a.hi, b.hi), v128_rdavg_u8(a.lo, b.lo));
}
SIMD_INLINE v256 v256_avg_u16(v256 a, v256 b) {
return v256_from_v128(v128_avg_u16(a.hi, b.hi), v128_avg_u16(a.lo, b.lo));
}
SIMD_INLINE v256 v256_min_u8(v256 a, v256 b) {
return v256_from_v128(v128_min_u8(a.hi, b.hi), v128_min_u8(a.lo, b.lo));
}
SIMD_INLINE v256 v256_max_u8(v256 a, v256 b) {
return v256_from_v128(v128_max_u8(a.hi, b.hi), v128_max_u8(a.lo, b.lo));
}
SIMD_INLINE v256 v256_min_s8(v256 a, v256 b) {
return v256_from_v128(v128_min_s8(a.hi, b.hi), v128_min_s8(a.lo, b.lo));
}
SIMD_INLINE v256 v256_max_s8(v256 a, v256 b) {
return v256_from_v128(v128_max_s8(a.hi, b.hi), v128_max_s8(a.lo, b.lo));