* 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 "kernel_operator.h"
using namespace AscendC;
constexpr int32_t BUFFER_NUM = 1;
class TEST_COPY : public testing::Test {
protected:
void SetUp()
{
AscendC::SetGCoreType(2);
}
void TearDown()
{
AscendC::SetGCoreType(0);
}
};
template <typename Type, int32_t extent1>
struct CopyType {
using T = Type;
constexpr static int32_t extent = extent1;
};
template<class XType, int32_t isSetMask = 0, int32_t maskMode = 0, int32_t mode = 0, int32_t maskLen = 1, int32_t counterMask = 0>
class KernelCopy {
public:
using X_T = typename XType::T;
__aicore__ inline KernelCopy() {}
__aicore__ inline void Init(GM_ADDR src, GM_ADDR dst)
{
srcGm.SetGlobalBuffer((__gm__ X_T*)src, 256);
dstGm.SetGlobalBuffer((__gm__ X_T*)dst, 256);
pipe.InitBuffer(inQueueX, BUFFER_NUM, 256 * sizeof(X_T));
pipe.InitBuffer(outQueueZ, BUFFER_NUM, 256 * sizeof(X_T));
}
__aicore__ inline void Process()
{
CopyIn();
if constexpr(maskMode == 0) {
if constexpr(mode == 0) {
if constexpr(isSetMask == 1) {
ComputeNormalBitIsSetMaskTrue();
} else {
ComputeNormalBitIsSetMaskFalse();
}
} else if constexpr(mode == 1) {
if constexpr(isSetMask == 1) {
ComputeNormalContinuousIsSetMaskTrue();
} else {
ComputeNormalContinuousIsSetMaskFalse();
}
} else {
ASSERT(false && "Normal mode only support bit and continuous mode!!!");
}
} else {
if constexpr(maskLen == 1) {
if constexpr(isSetMask == 1) {
ComputeCounterMaskIsSetMaskTrue();
} else {
ComputeCounterMaskIsSetMaskFalse();
}
} else {
ComputeCounterMasksIsSetMaskFalse();
}
}
CopyOut();
}
private:
__aicore__ inline void CopyIn()
{
LocalTensor<X_T> srcLocal = inQueueX.AllocTensor<X_T>();
DataCopy(srcLocal, srcGm, 256);
inQueueX.EnQue(srcLocal);
}
__aicore__ inline void ComputeNormalBitIsSetMaskTrue() {
LocalTensor<X_T> srcLocal = inQueueX.DeQue<X_T>();
LocalTensor<X_T> dstLocal = outQueueZ.AllocTensor<X_T>();
uint8_t repeat = sizeof(X_T);
CopyRepeatParams repeatParams(1, 1, 8, 8);
Duplicate(dstLocal, static_cast<X_T>(0), 256);
pipe_barrier(PIPE_ALL);
if (sizeof(X_T) == 2) {
uint64_t maskList[2] = {0x5555555555555555, 0x5555555555555555};
Copy<X_T, true>(dstLocal, srcLocal, maskList, repeat, repeatParams);
} else {
uint64_t maskList[2] = {0x5555555555555555, 0};
Copy<X_T, true>(dstLocal, srcLocal, maskList, repeat, repeatParams);
}
outQueueZ.EnQue<X_T>(dstLocal);
inQueueX.FreeTensor(srcLocal);
}
__aicore__ inline void ComputeNormalBitIsSetMaskFalse() {
LocalTensor<X_T> srcLocal = inQueueX.DeQue<X_T>();
LocalTensor<X_T> dstLocal = outQueueZ.AllocTensor<X_T>();
uint8_t repeat = sizeof(X_T);
CopyRepeatParams repeatParams(1, 1, 8, 8);
Duplicate(dstLocal, static_cast<X_T>(0), 256);
pipe_barrier(PIPE_ALL);
uint64_t maskList[2] = {0, 0};
SetMaskNorm();
if (sizeof(X_T) == 2) {
SetVectorMask<X_T, MaskMode::NORMAL>(0x5555555555555555, 0x5555555555555555);
} else {
SetVectorMask<X_T, MaskMode::NORMAL>(0, 0x5555555555555555);
}
Copy<X_T, false>(dstLocal, srcLocal, maskList, repeat, repeatParams);
outQueueZ.EnQue<X_T>(dstLocal);
inQueueX.FreeTensor(srcLocal);
}
__aicore__ inline void ComputeNormalContinuousIsSetMaskTrue() {
LocalTensor<X_T> srcLocal = inQueueX.DeQue<X_T>();
LocalTensor<X_T> dstLocal = outQueueZ.AllocTensor<X_T>();
uint8_t repeat = sizeof(X_T);
CopyRepeatParams repeatParams(1, 1, 8, 8);
Duplicate(dstLocal, static_cast<X_T>(0), 256);
pipe_barrier(PIPE_ALL);
Copy<X_T, true>(dstLocal, srcLocal, count_mask, repeat, repeatParams);
outQueueZ.EnQue<X_T>(dstLocal);
inQueueX.FreeTensor(srcLocal);
}
__aicore__ inline void ComputeNormalContinuousIsSetMaskFalse() {
LocalTensor<X_T> srcLocal = inQueueX.DeQue<X_T>();
LocalTensor<X_T> dstLocal = outQueueZ.AllocTensor<X_T>();
uint8_t repeat = sizeof(X_T);
int32_t mask = 0;
CopyRepeatParams repeatParams(1, 1, 8, 8);
Duplicate(dstLocal, static_cast<X_T>(0), 256);
pipe_barrier(PIPE_ALL);
SetMaskNorm();
SetVectorMask<X_T, MaskMode::NORMAL>(count_mask);
Copy<X_T, false>(dstLocal, srcLocal, mask, repeat, repeatParams);
outQueueZ.EnQue<X_T>(dstLocal);
inQueueX.FreeTensor(srcLocal);
}
__aicore__ inline void ComputeCounterMaskIsSetMaskTrue() {
LocalTensor<X_T> srcLocal = inQueueX.DeQue<X_T>();
LocalTensor<X_T> dstLocal = outQueueZ.AllocTensor<X_T>();
uint8_t repeat = sizeof(X_T);
CopyRepeatParams repeatParams(1, 1, 8, 8);
Duplicate(dstLocal, static_cast<X_T>(0), 256);
pipe_barrier(PIPE_ALL);
Copy<X_T, true>(dstLocal, srcLocal, count_mask, repeat, repeatParams);
outQueueZ.EnQue<X_T>(dstLocal);
inQueueX.FreeTensor(srcLocal);
}
__aicore__ inline void ComputeCounterMaskIsSetMaskFalse() {
LocalTensor<X_T> srcLocal = inQueueX.DeQue<X_T>();
LocalTensor<X_T> dstLocal = outQueueZ.AllocTensor<X_T>();
uint8_t repeat = sizeof(X_T);
int32_t mask = 0;
CopyRepeatParams repeatParams(1, 1, 8, 8);
Duplicate(dstLocal, static_cast<X_T>(0), 256);
pipe_barrier(PIPE_ALL);
SetMaskCount();
SetVectorMask<X_T, MaskMode::COUNTER>(count_mask);
Copy<X_T, false>(dstLocal, srcLocal, mask, repeat, repeatParams);
SetMaskNorm();
outQueueZ.EnQue<X_T>(dstLocal);
inQueueX.FreeTensor(srcLocal);
}
__aicore__ inline void ComputeCounterMasksIsSetMaskFalse() {
LocalTensor<X_T> srcLocal = inQueueX.DeQue<X_T>();
LocalTensor<X_T> dstLocal = outQueueZ.AllocTensor<X_T>();
uint8_t repeat = sizeof(X_T);
int32_t mask = 0;
CopyRepeatParams repeatParams(1, 1, 8, 8);
Duplicate(dstLocal, static_cast<X_T>(0), 256);
pipe_barrier(PIPE_ALL);
SetMaskCount();
SetVectorMask<X_T, MaskMode::COUNTER>(0, count_mask);
Copy<X_T, false>(dstLocal, srcLocal, mask, repeat, repeatParams);
SetMaskNorm();
outQueueZ.EnQue<X_T>(dstLocal);
inQueueX.FreeTensor(srcLocal);
}
__aicore__ inline void CopyOut()
{
LocalTensor<X_T> dstLocal = outQueueZ.DeQue<X_T>();
DataCopy(dstGm, dstLocal, 256);
outQueueZ.FreeTensor(dstLocal);
}
private:
TPipe pipe;
TQue<TPosition::VECIN, BUFFER_NUM> inQueueX;
TQue<TPosition::VECOUT, BUFFER_NUM> outQueueZ;
GlobalTensor<X_T> srcGm;
GlobalTensor<X_T> dstGm;
int32_t count_mask{counterMask};
};
#define KERNEL_COPY_M310(dataType, isSetMask, maskMode, mode, maskLen, counterMask) \
TEST_F(TEST_COPY, copy_kernel_##dataType##_is_set_mask_##isSetMask##_mask_mode_##maskMode##_##mode##_mask_len_##maskLen##_##counterMask##_case) \
{ \
uint32_t dataSize = 256; \
uint8_t srcGm[dataSize * sizeof(dataType)]; \
uint8_t dstGm[dataSize * sizeof(dataType)]; \
typedef CopyType<dataType, 256> xType; \
KernelCopy<xType, isSetMask, maskMode, mode, maskLen, counterMask> op; \
op.Init(srcGm, dstGm); \
op.Process(); \
dataType golden[dataSize]; \
uint8_t repeat = sizeof(dataType); \
uint8_t vecLen = 256 / sizeof(dataType); \
if (maskMode == 0 && mode == 0) { \
for (int i = 0; i < repeat; i++) { \
for (int j = 0; j < vecLen; j++) { \
if (j % 2 == 0) { \
golden[i*vecLen + j] = srcGm[i*vecLen + j]; \
} else { \
golden[i*vecLen + j] = 0; \
} \
} \
} \
} else { \
if (isSetMask == 1) { \
for (int r = 0; r < repeat; r++) { \
for (int i = 0; i < vecLen; i++) { \
if (i < counterMask) { \
golden[r*vecLen + i] = srcGm[r*vecLen + i]; \
} else { \
golden[r*vecLen + i] = 0; \
} \
} \
} \
} else { \
for (int i = 0; i < dataSize; i++) { \
if (i < counterMask) { \
golden[i] = srcGm[i]; \
} else { \
golden[i] = 0; \
} \
} \
} \
} \
for (uint32_t i = 0; i < dataSize; i++) { \
EXPECT_EQ(dstGm[i], 0x00); \
} \
}
KERNEL_COPY_M310(int16_t, 0, 0, 0, 2, 0)
KERNEL_COPY_M310(uint16_t, 0, 0, 0, 2, 0)
KERNEL_COPY_M310(int32_t, 0, 0, 0, 2, 0)
KERNEL_COPY_M310(uint32_t, 0, 0, 0, 2, 0)
KERNEL_COPY_M310(float, 0, 0, 0, 2, 0)
KERNEL_COPY_M310(half, 0, 0, 0, 2, 0)
KERNEL_COPY_M310(int16_t, 1, 0, 0, 2, 0)
KERNEL_COPY_M310(uint16_t, 1, 0, 0, 2, 0)
KERNEL_COPY_M310(int32_t, 1, 0, 0, 2, 0)
KERNEL_COPY_M310(uint32_t, 1, 0, 0, 2, 0)
KERNEL_COPY_M310(float, 1, 0, 0, 2, 0)
KERNEL_COPY_M310(half, 1, 0, 0, 2, 0)
KERNEL_COPY_M310(int16_t, 0, 0, 1, 1, 40)
KERNEL_COPY_M310(uint16_t, 0, 0, 1, 1, 100)
KERNEL_COPY_M310(int32_t, 0, 0, 1, 1, 121)
KERNEL_COPY_M310(uint32_t, 0, 0, 1, 1, 200)
KERNEL_COPY_M310(float, 0, 0, 1, 1, 158)
KERNEL_COPY_M310(half, 0, 0, 1, 1, 39)
KERNEL_COPY_M310(int16_t, 1, 0, 1, 1, 80)
KERNEL_COPY_M310(uint16_t, 1, 0, 1, 1, 123)
KERNEL_COPY_M310(int32_t, 1, 0, 1, 1, 61)
KERNEL_COPY_M310(uint32_t, 1, 0, 1, 1, 63)
KERNEL_COPY_M310(float, 1, 0, 1, 1, 50)
KERNEL_COPY_M310(half, 1, 0, 1, 1, 68)
KERNEL_COPY_M310(int16_t, 0, 1, 1, 1, 92)
KERNEL_COPY_M310(uint16_t, 0, 1, 1, 1, 89)
KERNEL_COPY_M310(int32_t, 0, 1, 1, 1, 249)
KERNEL_COPY_M310(uint32_t, 0, 1, 1, 1, 201)
KERNEL_COPY_M310(float, 0, 1, 1, 1, 157)
KERNEL_COPY_M310(half, 0, 1, 1, 1, 132)
KERNEL_COPY_M310(int16_t, 1, 0, 1, 1, 123)
KERNEL_COPY_M310(uint16_t, 1, 0, 1, 1, 104)
KERNEL_COPY_M310(int32_t, 1, 0, 1, 1, 48)
KERNEL_COPY_M310(uint32_t, 1, 0, 1, 1, 32)
KERNEL_COPY_M310(float, 1, 0, 1, 1, 55)
KERNEL_COPY_M310(half, 1, 0, 1, 1, 101)
KERNEL_COPY_M310(int16_t, 0, 1, 1, 2, 199)
KERNEL_COPY_M310(uint16_t, 0, 1, 1, 2, 187)
KERNEL_COPY_M310(int32_t, 0, 1, 1, 2, 136)
KERNEL_COPY_M310(uint32_t, 0, 1, 1, 2, 144)
KERNEL_COPY_M310(float, 0, 1, 1, 2, 243)
KERNEL_COPY_M310(half, 0, 1, 1, 2, 205)
