* Copyright (c) 2025 Huawei Technologies Co., Ltd.
* This program is free software, you can redistribute it and/or modify it under the terms and conditions of
* CANN Open Software License Agreement Version 2.0 (the "License").
* Please refer to the License for details. You may not use this file except in compliance with the License.
* THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
* INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
* See LICENSE in the root of the software repository for the full text of the License.
*/
#include "gtest/gtest.h"
#include "tikicpulib.h"
#include "kernel_operator.h"
using namespace AscendC;
#include "utils.h"
#include "test_api_utils.h"
#include "reduce.h"
class TestApiReduceLast : public testing::Test, public testing::WithParamInterface<std::vector<int>> {
};
TEST_P(TestApiReduceLast, Test_reduce_min_ab_to_a1) {
int a = this->GetParam()[0];
int b = this->GetParam()[1];
int stride = this->GetParam()[2];
auto *x = (half*)AscendC::GmAlloc(sizeof(half) * a * stride);
auto *y = (half*)AscendC::GmAlloc(sizeof(half) * a * 1);
half expect[a];
for (int i = 0; i < a; i++) {
expect[i] = (double)(i * b);
for (int j = 0; j < stride; j++) {
x[i * stride + j] = (double)(i * b + j);
}
}
auto kernel = [](int32_t a, int32_t b, int32_t stride, half *x, half *y) {
TPipe tpipe;
TBuf<TPosition::VECCALC> xbuf, ybuf, tmp;
tpipe.InitBuffer(xbuf, sizeof(half) * a * stride);
tpipe.InitBuffer(ybuf, sizeof(half) * a * 1);
tpipe.InitBuffer(tmp, 8 * 1024);
auto l_x = xbuf.Get<half>();
auto l_y = ybuf.Get<half>();
auto l_tmp = tmp.Get<uint8_t>();
GmToUb(l_x, x, a * stride);
GmToUb(l_y, y, a);
ReduceLast<half, WholeReduceMinAdapt, Min>(l_y, l_x, a, b, stride, l_tmp);
UbToGm(y, l_y, a);
};
AscendC::SetKernelMode(KernelMode::AIV_MODE);
ICPU_RUN_KF(kernel, 1, a, b, stride, x, y);
int diff_count = 0;
for (int i = 0; i < a; i++) {
auto diff = (double)(y[i] - expect[i]);
if (diff < -1e-5 || diff > 1e-5) {
diff_count++;
}
}
EXPECT_EQ(diff_count, 0);
}
TEST_P(TestApiReduceLast, Test_reduce_mean_ab_to_a1) {
int a = this->GetParam()[0];
int b = this->GetParam()[1];
int stride = this->GetParam()[2];
auto *x = (float*)AscendC::GmAlloc(sizeof(float) * a * stride);
auto *y = (float*)AscendC::GmAlloc(sizeof(float) * a * 1);
float expect[a];
for (int i = 0; i < a; i++) {
expect[i] = 0;
for (int j = 0; j < stride; j++) {
x[i * stride + j] = (double)(i * b + j);
if (j < b) {
expect[i] += x[i * stride + j];
}
}
expect[i] /= b;
}
auto kernel = [](int32_t a, int32_t b, int32_t stride, float *x, float *y) {
TPipe tpipe;
TBuf<TPosition::VECCALC> xbuf, ybuf, tmp;
tpipe.InitBuffer(xbuf, sizeof(float) * a * stride);
tpipe.InitBuffer(ybuf, sizeof(float) * a * 1);
tpipe.InitBuffer(tmp, 8 * 1024);
auto l_x = xbuf.Get<float>();
auto l_y = ybuf.Get<float>();
auto l_tmp = tmp.Get<uint8_t>();
GmToUb(l_x, x, a * stride);
GmToUb(l_y, y, a);
ReduceLast<float, WholeReduceMeanAdapt, Add>(l_y, l_x, a, b, stride, l_tmp);
UbToGm(y, l_y, a);
};
AscendC::SetKernelMode(KernelMode::AIV_MODE);
ICPU_RUN_KF(kernel, 1, a, b, stride, x, y);
int diff_count = 0;
for (int i = 0; i < a; i++) {
auto diff = (double)(y[i] - expect[i]);
if (diff < -1e-5 || diff > 1e-5) {
diff_count++;
}
}
EXPECT_EQ(diff_count, 0);
}
INSTANTIATE_TEST_SUITE_P(CalcWithDifferentShape, TestApiReduceLast,
::testing::Values(std::vector<int>{32, 32, 32},
std::vector<int>{300, 32, 32},
std::vector<int>{32, 128, 128},
std::vector<int>{1, 50, 64}));
TEST(TestApiReduceSumInt32, test_ab_to_a) {
uint32_t a = 16, b = 32;
auto *x = (int32_t*)AscendC::GmAlloc(sizeof(int32_t) * a * b);
auto *y = (int32_t*)AscendC::GmAlloc(sizeof(int32_t) * a);
int32_t expect[a];
for (uint32_t i = 0; i < a * b; i++) {
x[i] = (int32_t)1;
}
x[5] = (int32_t)2;
x[6] = (int32_t)(-2);
x[20] = (int32_t)2;
x[52] = (int32_t)3;
for (uint32_t i = 0; i < a; i++) {
expect[i] = 0;
for (uint32_t j = 0; j < b; j++) {
uint32_t idx = i * b + j;
expect[i] += x[idx];
}
}
auto kernel = [](uint32_t a, uint32_t b, int32_t *x, int32_t *y) {
TPipe tpipe;
TBuf<TPosition::VECCALC> xbuf, ybuf, tmp;
tpipe.InitBuffer(xbuf, sizeof(int32_t) * a * b);
tpipe.InitBuffer(ybuf, sizeof(int32_t) * a);
tpipe.InitBuffer(tmp, 8 * 1024);
auto l_x = xbuf.Get<int32_t>();
auto l_y = ybuf.Get<int32_t>();
auto l_tmp = tmp.Get<uint8_t>();
GmToUb(l_x, x, a * b);
uint32_t shape[] = {a, b};
ReduceSumInt32<int32_t, AscendC::Pattern::Reduce::AR, false>(l_y, l_x, l_tmp, shape, true);
UbToGm(y, l_y, a);
};
AscendC::SetKernelMode(KernelMode::AIV_MODE);
ICPU_RUN_KF(kernel, 1, a, b, x, y);
uint32_t diff_count = 0;
for (uint32_t i = 0; i < a; i++) {
auto diff = (double)(y[i] - expect[i]);
if (diff < -1e-5 || diff > 1e-5) {
diff_count++;
}
}
EXPECT_EQ(diff_count, 0);
}
TEST(TestApiReduceSumInt32, test_ab_to_b) {
uint32_t a = 16, b = 64;
auto *x = (int32_t*)AscendC::GmAlloc(sizeof(int32_t) * a * b);
auto *y = (int32_t*)AscendC::GmAlloc(sizeof(int32_t) * b);
int32_t expect[b];
for (uint32_t i = 0; i < a * b; i++) {
x[i] = (int32_t)1;
}
x[5] = (int32_t)2;
x[6] = (int32_t)(-2);
x[20] = (int32_t)2;
x[52] = (int32_t)3;
for (uint32_t i = 0; i < b; i++) {
expect[i] = 0;
for (uint32_t j = 0; j < a; j++) {
uint32_t idx = j * b + i;
expect[i] += x[idx];
}
}
auto kernel = [](uint32_t a, uint32_t b, int32_t *x, int32_t *y) {
TPipe tpipe;
TBuf<TPosition::VECCALC> xbuf, ybuf, tmp;
tpipe.InitBuffer(xbuf, sizeof(int32_t) * a * b);
tpipe.InitBuffer(ybuf, sizeof(int32_t) * b);
tpipe.InitBuffer(tmp, 8 * 1024);
auto l_x = xbuf.Get<int32_t>();
auto l_y = ybuf.Get<int32_t>();
auto l_tmp = tmp.Get<uint8_t>();
GmToUb(l_x, x, a * b);
uint32_t shape[] = {a, b};
ReduceSumInt32<int32_t, AscendC::Pattern::Reduce::RA, false>(l_y, l_x, l_tmp, shape, true);
UbToGm(y, l_y, b);
};
AscendC::SetKernelMode(KernelMode::AIV_MODE);
ICPU_RUN_KF(kernel, 1, a, b, x, y);
uint32_t diff_count = 0;
for (uint32_t i = 0; i < b; i++) {
auto diff = (int32_t)(y[i] - expect[i]);
if (diff < -1e-5 || diff > 1e-5) {
diff_count++;
}
}
EXPECT_EQ(diff_count, 0);
}
TEST(TestApiReduceSumInt32, test_ab_to_a_unalign) {
uint32_t a = 8, b = 8;
auto *x = (int32_t*)AscendC::GmAlloc(sizeof(int32_t) * a * b);
auto *y = (int32_t*)AscendC::GmAlloc(sizeof(int32_t) * b);
int32_t expect[a];
for (uint32_t i = 0; i < a * b; i++) {
x[i] = i % b == 7 ? 0 : 2;
}
for (uint32_t i = 0; i < a; i++) {
expect[i] = 14;
}
auto kernel = [](uint32_t a, uint32_t b, int32_t *x, int32_t *y) {
TPipe tpipe;
TBuf<TPosition::VECCALC> xbuf, ybuf, tmp;
tpipe.InitBuffer(xbuf, sizeof(int32_t) * a * b);
tpipe.InitBuffer(ybuf, sizeof(int32_t) * b);
tpipe.InitBuffer(tmp, 8 * 1024);
auto l_x = xbuf.Get<int32_t>();
auto l_y = ybuf.Get<int32_t>();
auto l_tmp = tmp.Get<uint8_t>();
GmToUb(l_x, x, a * b);
uint32_t shape[] = {a, 7};
ReduceSumInt32<int32_t, AscendC::Pattern::Reduce::AR, false>(l_y, l_x, l_tmp, shape, true);
UbToGm(y, l_y, b);
};
AscendC::SetKernelMode(KernelMode::AIV_MODE);
ICPU_RUN_KF(kernel, 1, a, b, x, y);
uint32_t diff_count = 0;
for (uint32_t i = 0; i < a; i++) {
auto diff = (double)(y[i] - expect[i]);
if (diff < -1e-5 || diff > 1e-5) {
diff_count++;
}
}
EXPECT_EQ(diff_count, 0);
}