sglang_v0.5.2/pytorch_2.8.0/third_party/XNNPACK/test/s32-simd-hvx.cc

74 lines
2.0 KiB
C++

// Auto-generated file. Do not edit!
// Template: test/s32-simd.cc.in
// Generator: tools/xngen
//
// 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.
// This header needs to go first for the arch test macros.
#include "xnnpack/common.h"
#if XNN_ENABLE_HVX && XNN_ARCH_HEXAGON
#include <algorithm>
#include <cmath>
#include <cstddef>
#include <cstdint>
#include <limits>
#include <random>
#include <vector>
#include <gmock/gmock.h>
#include <gtest/gtest.h>
#include "xnnpack/isa-checks.h"
#include "xnnpack/simd/s32-hvx.h"
#include "replicable_random_device.h"
namespace xnnpack {
class S32SimdHVXTest : public ::testing::Test {
protected:
void SetUp() override {
TEST_REQUIRES_HVX;
inputs_.resize(3 * xnn_simd_size_s32);
output_.resize(xnn_simd_size_s32);
std::uniform_int_distribution<int32_t> s32dist(-10000, 10000);
std::generate(inputs_.begin(), inputs_.end(),
[&]() { return s32dist(rng_); });
}
xnnpack::ReplicableRandomDevice rng_;
std::vector<int32_t> inputs_;
std::vector<int32_t> output_;
};
TEST_F(S32SimdHVXTest, Mul) {
const xnn_simd_s32_t a = xnn_loadu_s32(inputs_.data());
const xnn_simd_s32_t b = xnn_loadu_s32(inputs_.data() + xnn_simd_size_s32);
const xnn_simd_s32_t res = xnn_mul_s32(a, b);
xnn_storeu_s32(output_.data(), res);
for (size_t k = 0; k < xnn_simd_size_s32; k++) {
ASSERT_EQ(output_[k], inputs_[k] * inputs_[k + xnn_simd_size_s32]);
}
}
TEST_F(S32SimdHVXTest, StoreTail) {
const xnn_simd_s32_t a = xnn_loadu_s32(inputs_.data());
for (size_t num_elements = 1; num_elements < xnn_simd_size_s32;
num_elements++) {
xnn_store_tail_s32(output_.data(), a, num_elements);
for (size_t k = 0; k < num_elements; k++) {
ASSERT_EQ(output_[k], inputs_[k]);
}
for (size_t k = num_elements; k < xnn_simd_size_s32; k++) {
ASSERT_EQ(output_[k], 0.0f);
}
}
}
} // namespace xnnpack
#endif // XNN_ENABLE_HVX && XNN_ARCH_HEXAGON