chromium/third_party/xnnpack/src/src/xnnpack/simd/s32-avx2.h

// Copyright 2024 Google LLC
//
// This source code is licensed under the BSD-style license found in the
// LICENSE file in the root directory of this source tree.
//

#ifndef __XNNPACK_SRC_XNNPACK_SIMD_S32_AVX2_H_
#define __XNNPACK_SRC_XNNPACK_SIMD_S32_AVX2_H_

#include <assert.h>
#include <immintrin.h>
#include <stddef.h>
#include <stdint.h>

#include "xnnpack/common.h"
#include "xnnpack/unaligned.h"

// SIMD vector type for s32 using AVX2.
xnn_simd_s32_t;
#define xnn_simd_size_s32 …
#define xnn_simd_log2_size_s32 …
#define xnn_simd_bytes_s32 …

#define XNN_SIMD_CONST_S32(var, val) …

// Mask table used for masked load/store operations.
static const int32_t mask_table_avx_s32[14] = …;
// Arithmetic operations.
static XNN_INLINE xnn_simd_s32_t xnn_mul_s32(xnn_simd_s32_t a,
                                             xnn_simd_s32_t b) { … }

static XNN_INLINE xnn_simd_s32_t xnn_max_s32(xnn_simd_s32_t a,
                                             xnn_simd_s32_t b) { … }

static XNN_INLINE xnn_simd_s32_t xnn_min_s32(xnn_simd_s32_t a,
                                             xnn_simd_s32_t b) { … }

// Load/store operations.

static XNN_INLINE xnn_simd_s32_t xnn_loadu_s32(const int32_t* ptr) { … }

static XNN_INLINE xnn_simd_s32_t xnn_load_s32(const int32_t* ptr) { … }

static XNN_INLINE void xnn_storeu_s32(int32_t* ptr, xnn_simd_s32_t v) { … }

static XNN_INLINE void xnn_store_s32(int32_t* ptr, xnn_simd_s32_t v) { … }

static XNN_INLINE xnn_simd_s32_t xnn_set1_s32(int32_t v) { … }

static XNN_INLINE xnn_simd_s32_t xnn_set1_or_load_s32(const int32_t* v) { … }

// Tail load/store operations.

static XNN_INLINE xnn_simd_s32_t
xnn_load_tail_s32(const int32_t* input, size_t num_elements) { … }

static XNN_INLINE void xnn_store_tail_s32(int32_t* output, xnn_simd_s32_t v,
                                          size_t num_elements) { … }

#endif  // __XNNPACK_SRC_XNNPACK_SIMD_S32_AVX2_H_