Star340
76
代码
代码
Issues22
Pull Requests9
Actions
项目讨论
Wiki
分析
Star340
76
已合并
fix: 修复代码审查发现的健壮性与文档问题(#96-#105) #113
syy_3597创建于 8月20日
fix: 修复代码审查发现的健壮性与文档问题(#96-#105) #113
已合并
syy_3597创建于 8月20日
从yang_3597/sipfix/code-review-issues-96-105合入到cann/sipmaster
共 66 个文件变更+1449-1452
M.gitignore+6-0
@@ -31,3 +31,9 @@ output/
31oat_reports/31oat_reports/
32 32 
33log/33log/
34+ 
35+# coverage tool and lint caches
36+lcov/
37+lcov-1.16/
38+.ruff_cache/
39+oat_reports*/
Mcore/base/swaplast2axes.cpp+21-20
@@ -8,6 +8,7 @@
8 * See LICENSE in the root of the software repository for the full text of the License.8 * See LICENSE in the root of the software repository for the full text of the License.
9 */9 */
10#include <iostream>10#include <iostream>
11+#include <optional>
11#include "utils/assert.h"12#include "utils/assert.h"
12#include "log/log.h"13#include "log/log.h"
13#include "utils/common_check.h"14#include "utils/common_check.h"
@@ -22,7 +23,7 @@ using namespace AsdSip;
22 23 
23namespace AsdSip {24namespace AsdSip {
24 25 
25-AspbStatus SwapLast2AxesDtypeCheck(const aclTensor *inTensor, aclTensor *outTensor)26+AspbStatus SwapLast2AxesDtypeCheck(const aclTensor* inTensor, aclTensor* outTensor)
26{27{
27 auto ret = ErrorType::ACL_ERROR_UNSUPPORTED_DATA_TYPE;28 auto ret = ErrorType::ACL_ERROR_UNSUPPORTED_DATA_TYPE;
28 SIP_OP_CHECK_DTYPE_NOT_MATCH(inTensor, aclDataType::ACL_COMPLEX64, ret);29 SIP_OP_CHECK_DTYPE_NOT_MATCH(inTensor, aclDataType::ACL_COMPLEX64, ret);
@@ -31,28 +32,28 @@ AspbStatus SwapLast2AxesDtypeCheck(const aclTensor *inTensor, aclTensor *outTens
31 return ErrorType::ACL_SUCCESS;32 return ErrorType::ACL_SUCCESS;
32}33}
33 34 
34-AspbStatus SwapLast2AxesShapeCheck(35+AspbStatus SwapLast2AxesShapeCheck(const aclTensor* inTensor, aclTensor* outTensor, int64_t*& storageDims,
35- const aclTensor *inTensor, aclTensor *outTensor, int64_t *&storageDims, uint64_t &inStorageDimsNum)36+ uint64_t& inStorageDimsNum)
36{37{
37 uint64_t outStorageDimsNum = 0;38 uint64_t outStorageDimsNum = 0;
38- CHECK_STATUS_WITH_ACL_RETURN(39+ CHECK_STATUS_WITH_ACL_RETURN(aclGetStorageShape(outTensor, &storageDims, &outStorageDimsNum),
39- aclGetStorageShape(outTensor, &storageDims, &outStorageDimsNum), "swapLast2Axes: aclGetStorageShape");40+ "swapLast2Axes: aclGetStorageShape");
40- int64_t outNum = GetTensorNum(outStorageDimsNum, storageDims);41+ std::optional<int64_t> outNumOpt = GetTensorNum(outStorageDimsNum, storageDims);
41 42 
42 delete[] storageDims;43 delete[] storageDims;
43 storageDims = nullptr;44 storageDims = nullptr;
44 45 
45- CHECK_STATUS_WITH_ACL_RETURN(46+ CHECK_STATUS_WITH_ACL_RETURN(aclGetStorageShape(inTensor, &storageDims, &inStorageDimsNum),
46- aclGetStorageShape(inTensor, &storageDims, &inStorageDimsNum), "swapLast2Axes: aclGetStorageShape");47+ "swapLast2Axes: aclGetStorageShape");
47 48 
48 // element num check49 // element num check
49- int64_t inNum = GetTensorNum(inStorageDimsNum, storageDims);50+ std::optional<int64_t> inNumOpt = GetTensorNum(inStorageDimsNum, storageDims);
50- if (outNum < inNum) {51+ if (!outNumOpt.has_value() || !inNumOpt.has_value() || outNumOpt.value() < inNumOpt.value()) {
51 delete[] storageDims;52 delete[] storageDims;
52 storageDims = nullptr;53 storageDims = nullptr;
53- ASDSIP_LOG(ERROR) << "swapLast2Axes get wrong inTensor or ouTensor num."54+ ASDSIP_LOG(ERROR) << "swapLast2Axes get wrong inTensor or outTensor num."
54- << "In element num is " << inNum << " , and out element num is " << outNum55+ << "In element num is " << inNumOpt.value_or(-1) << " , and out element num is "
55- << " . Out num must be no less than in num";56+ << outNumOpt.value_or(-1) << " . Out num must be no less than in num";
56 return ErrorType::ACL_ERROR_INVALID_PARAM;57 return ErrorType::ACL_ERROR_INVALID_PARAM;
57 }58 }
58 59 
@@ -67,14 +68,14 @@ AspbStatus SwapLast2AxesShapeCheck(
67 return ErrorType::ACL_SUCCESS;68 return ErrorType::ACL_SUCCESS;
68}69}
69 70 
70-AspbStatus swapLast2Axes(const aclTensor *inTensor, aclTensor *outTensor, void *stream, void *workspace)71+AspbStatus swapLast2Axes(const aclTensor* inTensor, aclTensor* outTensor, void* stream, void* workspace)
71{72{
72 auto ret = SwapLast2AxesDtypeCheck(inTensor, outTensor);73 auto ret = SwapLast2AxesDtypeCheck(inTensor, outTensor);
73 if (ret != ErrorType::ACL_SUCCESS) {74 if (ret != ErrorType::ACL_SUCCESS) {
74 return ret;75 return ret;
75 }76 }
76 77 
77- int64_t *storageDims = nullptr;78+ int64_t* storageDims = nullptr;
78 uint64_t inStorageDimsNum = 0;79 uint64_t inStorageDimsNum = 0;
79 ret = SwapLast2AxesShapeCheck(inTensor, outTensor, storageDims, inStorageDimsNum);80 ret = SwapLast2AxesShapeCheck(inTensor, outTensor, storageDims, inStorageDimsNum);
80 if (ret != ErrorType::ACL_SUCCESS) {81 if (ret != ErrorType::ACL_SUCCESS) {
@@ -96,24 +97,24 @@ AspbStatus swapLast2Axes(const aclTensor *inTensor, aclTensor *outTensor, void *
96 opDesc.specificParam = param;97 opDesc.specificParam = param;
97 ASDSIP_LOG(DEBUG) << "OpDesc: " << opDesc.opName << "; OpDesc info: " << param.ToString();98 ASDSIP_LOG(DEBUG) << "OpDesc: " << opDesc.opName << "; OpDesc info: " << param.ToString();
98 99 
99- SVector<aclTensor *> inTensors = {const_cast<aclTensor *>(inTensor)};100+ SVector<aclTensor*> inTensors = {const_cast<aclTensor*>(inTensor)};
100- SVector<aclTensor *> outTensors = {outTensor};101+ SVector<aclTensor*> outTensors = {outTensor};
101 102 
102 if (storageDims != nullptr) {103 if (storageDims != nullptr) {
103 delete[] storageDims;104 delete[] storageDims;
104 storageDims = nullptr;105 storageDims = nullptr;
105 }106 }
106 107 
107- Status status = RunAsdOpsV2(stream, opDesc, inTensors, outTensors, (uint8_t *)workspace);108+ Status status = RunAsdOpsV2(stream, opDesc, inTensors, outTensors, (uint8_t*)workspace);
108 ASDSIP_ECHECK(status.Ok(), status.Message(), ErrorType::ACL_ERROR_INTERNAL_ERROR);109 ASDSIP_ECHECK(status.Ok(), status.Message(), ErrorType::ACL_ERROR_INTERNAL_ERROR);
109 110 
110 ASDSIP_LOG(INFO) << "Execute swapLast2Axes success.";111 ASDSIP_LOG(INFO) << "Execute swapLast2Axes success.";
111 return ErrorType::ACL_SUCCESS;112 return ErrorType::ACL_SUCCESS;
112}113}
113 114 
114-AsdSip::AspbStatus swapLast2AxesGetWorkspaceSize(size_t &size)115+AsdSip::AspbStatus swapLast2AxesGetWorkspaceSize(size_t& size)
115{116{
116 size = ASYNC_WORKSPACE_SIZE;117 size = ASYNC_WORKSPACE_SIZE;
117 return ErrorType::ACL_SUCCESS;118 return ErrorType::ACL_SUCCESS;
118}119}
119-} // namespace AsdSip120+} // namespace AsdSip
Mcore/blas/blas_api.cpp+20-20
@@ -12,7 +12,7 @@
12#include "blas_common.h"12#include "blas_common.h"
13 13 
14namespace AsdSip {14namespace AsdSip {
15-AspbStatus asdBlasCreate(asdBlasHandle &handle)15+AspbStatus asdBlasCreate(asdBlasHandle& handle)
16{16{
17 std::lock_guard<std::mutex> lock(blas_mtx);17 std::lock_guard<std::mutex> lock(blas_mtx);
18 handle = BlasPlanCache::InitHandle();18 handle = BlasPlanCache::InitHandle();
@@ -20,13 +20,13 @@ AspbStatus asdBlasCreate(asdBlasHandle &handle)
20 return ErrorType::ACL_SUCCESS;20 return ErrorType::ACL_SUCCESS;
21}21}
22 22 
23-AspbStatus asdBlasSetStream(asdBlasHandle handle, void *stream)23+AspbStatus asdBlasSetStream(asdBlasHandle handle, void* stream)
24{24{
25 std::lock_guard<std::mutex> lock(blas_mtx);25 std::lock_guard<std::mutex> lock(blas_mtx);
26- ASDSIP_ECHECK(26+ ASDSIP_ECHECK(BlasPlanCache::doesPlanExist(handle), "blas plan does not exist.",
27- BlasPlanCache::doesPlanExist(handle), "blas plan does not exist.", ErrorType::ACL_ERROR_INTERNAL_ERROR);27+ ErrorType::ACL_ERROR_INTERNAL_ERROR);
28 ASDSIP_ECHECK(stream != nullptr, "stream is empty.", ErrorType::ACL_ERROR_INTERNAL_ERROR);28 ASDSIP_ECHECK(stream != nullptr, "stream is empty.", ErrorType::ACL_ERROR_INTERNAL_ERROR);
29- BlasPlan &plan = BlasPlanCache::getPlan(handle);29+ BlasPlan& plan = BlasPlanCache::getPlan(handle);
30 30 
31 plan.SetStream(stream);31 plan.SetStream(stream);
32 ASDSIP_LOG(INFO) << "Blas set stream success.";32 ASDSIP_LOG(INFO) << "Blas set stream success.";
@@ -41,7 +41,7 @@ AspbStatus asdBlasDestroy(asdBlasHandle handle)
41 return ErrorType::ACL_ERROR_INVALID_PARAM;41 return ErrorType::ACL_ERROR_INVALID_PARAM;
42 }42 }
43 BlasPlanCache::destroy_plan(handle);43 BlasPlanCache::destroy_plan(handle);
44- delete static_cast<int *>(handle);44+ delete static_cast<int*>(handle);
45 ASDSIP_LOG(INFO) << "BlasHandle destroy.";45 ASDSIP_LOG(INFO) << "BlasHandle destroy.";
46 return ErrorType::ACL_SUCCESS;46 return ErrorType::ACL_SUCCESS;
47}47}
@@ -49,33 +49,33 @@ AspbStatus asdBlasDestroy(asdBlasHandle handle)
49AspbStatus asdBlasSynchronize(asdBlasHandle handle)49AspbStatus asdBlasSynchronize(asdBlasHandle handle)
50{50{
51 std::lock_guard<std::mutex> lock(blas_mtx);51 std::lock_guard<std::mutex> lock(blas_mtx);
52- ASDSIP_ECHECK(52+ ASDSIP_ECHECK(BlasPlanCache::doesPlanExist(handle), "blas plan does not exist.",
53- BlasPlanCache::doesPlanExist(handle), "blas plan does not exist.", ErrorType::ACL_ERROR_INTERNAL_ERROR);53+ ErrorType::ACL_ERROR_INTERNAL_ERROR);
54- BlasPlan &plan = BlasPlanCache::getPlan(handle);54+ BlasPlan& plan = BlasPlanCache::getPlan(handle);
55 Mki::MkiRtStreamSynchronize(plan.GetStream());55 Mki::MkiRtStreamSynchronize(plan.GetStream());
56- ASDSIP_LOG(INFO) << "Blas get synchronized.";56+ ASDSIP_LOG(INFO) << "Blas stream synchronized.";
57 return ErrorType::ACL_SUCCESS;57 return ErrorType::ACL_SUCCESS;
58}58}
59 59 
60-AspbStatus asdBlasGetWorkspaceSize(asdBlasHandle handle, size_t &workspaceSize)60+AspbStatus asdBlasGetWorkspaceSize(asdBlasHandle handle, size_t& workspaceSize)
61{61{
62 std::lock_guard<std::mutex> lock(blas_mtx);62 std::lock_guard<std::mutex> lock(blas_mtx);
63- ASDSIP_ECHECK(63+ ASDSIP_ECHECK(BlasPlanCache::doesPlanExist(handle), "blas plan does not exist.",
64- BlasPlanCache::doesPlanExist(handle), "blas plan does not exist.", ErrorType::ACL_ERROR_INTERNAL_ERROR);64+ ErrorType::ACL_ERROR_INTERNAL_ERROR);
65- BlasPlan &plan = BlasPlanCache::getPlan(handle);65+ BlasPlan& plan = BlasPlanCache::getPlan(handle);
66 workspaceSize = static_cast<size_t>(plan.GetWorkspaceSize());66 workspaceSize = static_cast<size_t>(plan.GetWorkspaceSize());
67 ASDSIP_LOG(INFO) << "Blas get workspaceSize success.";67 ASDSIP_LOG(INFO) << "Blas get workspaceSize success.";
68 return ErrorType::ACL_SUCCESS;68 return ErrorType::ACL_SUCCESS;
69}69}
70 70 
71-AspbStatus asdBlasSetWorkspace(asdBlasHandle handle, void *workSpace)71+AspbStatus asdBlasSetWorkspace(asdBlasHandle handle, void* workSpace)
72{72{
73 std::lock_guard<std::mutex> lock(blas_mtx);73 std::lock_guard<std::mutex> lock(blas_mtx);
74- ASDSIP_ECHECK(74+ ASDSIP_ECHECK(BlasPlanCache::doesPlanExist(handle), "blas plan does not exist.",
75- BlasPlanCache::doesPlanExist(handle), "blas plan does not exist.", ErrorType::ACL_ERROR_INTERNAL_ERROR);75+ ErrorType::ACL_ERROR_INTERNAL_ERROR);
76- BlasPlan &plan = BlasPlanCache::getPlan(handle);76+ BlasPlan& plan = BlasPlanCache::getPlan(handle);
77- plan.SetWorkspace((uint8_t *)workSpace);77+ plan.SetWorkspace((uint8_t*)workSpace);
78 ASDSIP_LOG(INFO) << "Blas set workspace success.";78 ASDSIP_LOG(INFO) << "Blas set workspace success.";
79 return ErrorType::ACL_SUCCESS;79 return ErrorType::ACL_SUCCESS;
80}80}
81-} // namespace AsdSip81+} // namespace AsdSip
Mcore/blas/blasplan/BlasCalPlan.cpp+8-18
@@ -13,7 +13,7 @@
13#include "log/log.h"13#include "log/log.h"
14#include "utils/aspb_status.h"14#include "utils/aspb_status.h"
15 15 
16-constexpr uint32_t MAX_LENG_PER_UB_PROC = 6144; // 单次处理的最大数据量（复数） 6.75k ( 6912 -> 对齐32个complex元素)16+constexpr uint32_t MAX_LENG_PER_UB_PROC = 6144; // 单次处理的最大数据量（复数） 6.75k ( 6912 -> 对齐32个complex元素)
17constexpr uint32_t ELEMENTS_EACH_COMPLEX64 = 2;17constexpr uint32_t ELEMENTS_EACH_COMPLEX64 = 2;
18constexpr uint32_t PING_PONG_NUM = 2;18constexpr uint32_t PING_PONG_NUM = 2;
19constexpr uint32_t BLAS_SCAL_WORKSPACE_SIZE = 16 * 1024;19constexpr uint32_t BLAS_SCAL_WORKSPACE_SIZE = 16 * 1024;
@@ -27,10 +27,7 @@ BlasCalPlan::BlasCalPlan() : m(0), n(0), incy(1), transA('N')
27 maskTensor.hostData = nullptr;27 maskTensor.hostData = nullptr;
28};28};
29 29 
30-AsdSip::AspbStatus BlasCalPlan::CreateTensor()30+AsdSip::AspbStatus BlasCalPlan::CreateTensor() { return SetMaskTensor(); }
31-{
32- return SetMaskTensor();
33-}
34 31 
35AsdSip::AspbStatus BlasCalPlan::SetMaskTensor()32AsdSip::AspbStatus BlasCalPlan::SetMaskTensor()
36{33{
@@ -38,7 +35,7 @@ AsdSip::AspbStatus BlasCalPlan::SetMaskTensor()
38 MAX_LENG_PER_UB_PROC * sizeof(std::complex<float>) * PING_PONG_NUM;35 MAX_LENG_PER_UB_PROC * sizeof(std::complex<float>) * PING_PONG_NUM;
39 uint32_t realBaseAddr = imagBaseAddr + MAX_LENG_PER_UB_PROC * sizeof(uint32_t);36 uint32_t realBaseAddr = imagBaseAddr + MAX_LENG_PER_UB_PROC * sizeof(uint32_t);
40 37 
41- uint32_t *maskData = nullptr;38+ uint32_t* maskData = nullptr;
42 try {39 try {
43 maskData = new uint32_t[MAX_LENG_PER_UB_PROC * ELEMENTS_EACH_COMPLEX64];40 maskData = new uint32_t[MAX_LENG_PER_UB_PROC * ELEMENTS_EACH_COMPLEX64];
44 } catch (std::bad_alloc& e) {41 } catch (std::bad_alloc& e) {
@@ -56,14 +53,13 @@ AsdSip::AspbStatus BlasCalPlan::SetMaskTensor()
56 maskTensor.dataSize = MAX_LENG_PER_UB_PROC * ELEMENTS_EACH_COMPLEX64 * sizeof(uint32_t);53 maskTensor.dataSize = MAX_LENG_PER_UB_PROC * ELEMENTS_EACH_COMPLEX64 * sizeof(uint32_t);
57 54 
58 if (!MallocTensorInDevice(maskTensor).Ok()) {55 if (!MallocTensorInDevice(maskTensor).Ok()) {
59- ASDSIP_LOG(ERROR) << "BlasCalPlan maskTensor malloc failed: ";56+ ASDSIP_LOG(ERROR) << "BlasCalPlan maskTensor malloc failed";
60 delete[] maskData;57 delete[] maskData;
61 maskData = nullptr;58 maskData = nullptr;
62 maskTensor.hostData = nullptr;59 maskTensor.hostData = nullptr;
63 return ErrorType::ACL_ERROR_INTERNAL_ERROR;60 return ErrorType::ACL_ERROR_INTERNAL_ERROR;
64 }61 }
65 toAclTensor(maskTensor, maskAclTensor);62 toAclTensor(maskTensor, maskAclTensor);
66- maskData = nullptr;
67 return ErrorType::ACL_SUCCESS;63 return ErrorType::ACL_SUCCESS;
68};64};
69 65 
@@ -74,19 +70,13 @@ AsdSip::AspbStatus BlasCalPlan::FreeTensor()
74 maskTensor.data = nullptr;70 maskTensor.data = nullptr;
75 }71 }
76 if (maskTensor.hostData != nullptr) {72 if (maskTensor.hostData != nullptr) {
77- delete[] static_cast<uint32_t *>(maskTensor.hostData);73+ delete[] static_cast<uint32_t*>(maskTensor.hostData);
78 maskTensor.hostData = nullptr;74 maskTensor.hostData = nullptr;
79 }75 }
80 return ErrorType::ACL_SUCCESS;76 return ErrorType::ACL_SUCCESS;
81}77}
82 78 
83-BlasCalPlan::~BlasCalPlan()79+BlasCalPlan::~BlasCalPlan() { BlasPlan::DestroyPlanData(); }
84-{
85- BlasPlan::DestroyPlanData();
86-}
87 80 
88-int64_t BlasCalPlan::GetWorkspaceSize()81+int64_t BlasCalPlan::GetWorkspaceSize() { return BLAS_SCAL_WORKSPACE_SIZE; }
89-{82+} // namespace AsdSip
90- return BLAS_SCAL_WORKSPACE_SIZE;
91-}
92-}
Mcore/fft/fft_api.cpp+128-137
Mcore/fft/fftcore/dft_r2c_core.cpp+19-19
@@ -33,11 +33,11 @@ size_t DftR2CCore::EstimateWorkspaceSize()
33 ASDSIP_LOG(INFO) << "ASCEND_950 DftR2CCore workspace size.";33 ASDSIP_LOG(INFO) << "ASCEND_950 DftR2CCore workspace size.";
34 return MATMUL_WORKSIZE;34 return MATMUL_WORKSIZE;
35 }35 }
36- const KernelInfo &kernelInfo = kernel->GetKernelInfo();36+ const KernelInfo& kernelInfo = kernel->GetKernelInfo();
37 return getAlignedSize(kernelInfo.GetTotalScratchSize());37 return getAlignedSize(kernelInfo.GetTotalScratchSize());
38}38}
39 39 
40-void DftR2CCore::Run(void *input, void *output, void *stream, workspace::Workspace &workspace)40+void DftR2CCore::Run(void* input, void* output, void* stream, workspace::Workspace& workspace)
41{41{
42 if (Mki::PlatformInfo::Instance().GetPlatformType() == Mki::PlatformType::ASCEND_910B) {42 if (Mki::PlatformInfo::Instance().GetPlatformType() == Mki::PlatformType::ASCEND_910B) {
43 FftOperation::Run(input, output, stream, workspace);43 FftOperation::Run(input, output, stream, workspace);
@@ -51,7 +51,7 @@ void DftR2CCore::Run(void *input, void *output, void *stream, workspace::Workspa
51void DftR2CCore::DestroyInDevice() const51void DftR2CCore::DestroyInDevice() const
52{52{
53 // destroy tiling data in device53 // destroy tiling data in device
54- uint8_t *deviceLaunchBuffer = nullptr;54+ uint8_t* deviceLaunchBuffer = nullptr;
55 deviceLaunchBuffer = runInfo.GetTilingDeviceAddr();55 deviceLaunchBuffer = runInfo.GetTilingDeviceAddr();
56 if (deviceLaunchBuffer != nullptr) {56 if (deviceLaunchBuffer != nullptr) {
57 MkiRtMemFreeDevice(deviceLaunchBuffer);57 MkiRtMemFreeDevice(deviceLaunchBuffer);
@@ -64,17 +64,17 @@ AspbStatus DftR2CCore::InitRotationMatrix()
64 int64_t inSize = fftN;64 int64_t inSize = fftN;
65 int64_t outSize = 2 * (fftN / 2 + 1);65 int64_t outSize = 2 * (fftN / 2 + 1);
66 66 
67- std::function<AsdSip::FFTensor *()> func = [=]() -> AsdSip::FFTensor* {67+ std::function<AsdSip::FFTensor*()> func = [=]() -> AsdSip::FFTensor* {
68- AsdSip::FFTensor *rotationMatrixPtr = new AsdSip::FFTensor;68+ AsdSip::FFTensor* rotationMatrixPtr = new AsdSip::FFTensor;
69- AsdSip::FFTensor &rotationMatrix_ = *rotationMatrixPtr;69+ AsdSip::FFTensor& rotationMatrix_ = *rotationMatrixPtr;
70 70 
71- float *rotationMatrixHost = nullptr;71+ float* rotationMatrixHost = nullptr;
72 try {72 try {
73 rotationMatrixHost = new float[outSize * inSize];73 rotationMatrixHost = new float[outSize * inSize];
74- } catch(std::bad_alloc& e) {74+ } catch (std::bad_alloc& e) {
75 delete rotationMatrixPtr;75 delete rotationMatrixPtr;
76- ASDSIP_LOG(ERROR) << "rotationMatrixHost nalloc failed: " << e.what();76+ ASDSIP_LOG(ERROR) << "rotationMatrixHost alloc failed: " << e.what();
77- throw std::runtime_error("rotationMatrixHost nalloc failed:.");77+ throw std::runtime_error("rotationMatrixHost alloc failed:.");
78 }78 }
79 79 
80 float cosTable[fftN];80 float cosTable[fftN];
@@ -86,8 +86,8 @@ AspbStatus DftR2CCore::InitRotationMatrix()
86 for (int64_t i = 0; i < inSize; i++) {86 for (int64_t i = 0; i < inSize; i++) {
87 for (int64_t j = 0; j < (fftN / 2 + 1); j++) {87 for (int64_t j = 0; j < (fftN / 2 + 1); j++) {
88 *(rotationMatrixHost + i * outSize + 2 * j) = *(cosTable + (i * j) % fftN);88 *(rotationMatrixHost + i * outSize + 2 * j) = *(cosTable + (i * j) % fftN);
89- *(rotationMatrixHost + i * outSize + 2 * j + 1) =89+ *(rotationMatrixHost + i * outSize + 2 * j + 1) = (problemDesc.forward ? (-1.0) : (1.0)) *
90- (problemDesc.forward ? (-1.0) : (1.0)) * (*(sinTable + (i * j) % fftN));90+ (*(sinTable + (i * j) % fftN));
91 }91 }
92 }92 }
93 93 
@@ -133,19 +133,19 @@ AspbStatus DftR2CCore::InitTactic()
133 Tensor tensorIn;133 Tensor tensorIn;
134 Tensor tensorOut;134 Tensor tensorOut;
135 tensorIn.desc = {TENSOR_DTYPE_FLOAT, TENSOR_FORMAT_ND, {problemDesc.batch, problemDesc.nDoing}, {}, 0};135 tensorIn.desc = {TENSOR_DTYPE_FLOAT, TENSOR_FORMAT_ND, {problemDesc.batch, problemDesc.nDoing}, {}, 0};
136- tensorIn.dataSize =136+ tensorIn.dataSize = problemDesc.batch * problemDesc.nDoing *
137- problemDesc.batch * problemDesc.nDoing * GetTensorElementSize(AsdSip::TensorDType::TENSOR_DTYPE_FLOAT);137+ GetTensorElementSize(AsdSip::TensorDType::TENSOR_DTYPE_FLOAT);
138 138 
139 unsigned outputN = problemDesc.nDoing / 2 + 1;139 unsigned outputN = problemDesc.nDoing / 2 + 1;
140- tensorOut.dataSize =140+ tensorOut.dataSize = problemDesc.batch * outputN *
141- problemDesc.batch * outputN * GetTensorElementSize(AsdSip::TensorDType::TENSOR_DTYPE_COMPLEX64);141+ GetTensorElementSize(AsdSip::TensorDType::TENSOR_DTYPE_COMPLEX64);
142 142 
143 launchParam.SetParam(param);143 launchParam.SetParam(param);
144 launchParam.AddInTensor(tensorIn);144 launchParam.AddInTensor(tensorIn);
145 launchParam.AddInTensor(*rotationMatrix);145 launchParam.AddInTensor(*rotationMatrix);
146 launchParam.AddOutTensor(tensorOut);146 launchParam.AddOutTensor(tensorOut);
147 147 
148- Operation *op = Ops::Instance().GetOperationByName(std::string("DftR2COperation"));148+ Operation* op = Ops::Instance().GetOperationByName(std::string("DftR2COperation"));
149 if (op == nullptr) {149 if (op == nullptr) {
150 return AsdSip::ErrorType::ACL_ERROR_INTERNAL_ERROR;150 return AsdSip::ErrorType::ACL_ERROR_INTERNAL_ERROR;
151 }151 }
@@ -154,7 +154,7 @@ AspbStatus DftR2CCore::InitTactic()
154 ASDSIP_ECHECK(kernel != nullptr, "Get best kernel failed", AsdSip::ErrorType::ACL_ERROR_INTERNAL_ERROR);154 ASDSIP_ECHECK(kernel != nullptr, "Get best kernel failed", AsdSip::ErrorType::ACL_ERROR_INTERNAL_ERROR);
155 155 
156 // allocate and initialize tiling workspace156 // allocate and initialize tiling workspace
157- uint8_t *deviceLaunchBuffer = nullptr;157+ uint8_t* deviceLaunchBuffer = nullptr;
158 kernel->SetLaunchWithTiling(false);158 kernel->SetLaunchWithTiling(false);
159 uint32_t launchBufferSize = kernel->GetTilingSize(launchParam);159 uint32_t launchBufferSize = kernel->GetTilingSize(launchParam);
160 if (launchBufferSize == 0) {160 if (launchBufferSize == 0) {
@@ -171,7 +171,7 @@ AspbStatus DftR2CCore::InitTactic()
171 ASDSIP_LOG(ERROR) << "malloc device memory fail";171 ASDSIP_LOG(ERROR) << "malloc device memory fail";
172 return AsdSip::ErrorType::ACL_ERROR_INTERNAL_ERROR;172 return AsdSip::ErrorType::ACL_ERROR_INTERNAL_ERROR;
173 }173 }
174- deviceLaunchBuffer = static_cast<uint8_t *>(tempDevicePtr);174+ deviceLaunchBuffer = static_cast<uint8_t*>(tempDevicePtr);
175 st = MkiRtMemCopy(deviceLaunchBuffer, launchBufferSize, hostLaunchBuffer, launchBufferSize,175 st = MkiRtMemCopy(deviceLaunchBuffer, launchBufferSize, hostLaunchBuffer, launchBufferSize,
176 MKIRT_MEMCOPY_HOST_TO_DEVICE);176 MKIRT_MEMCOPY_HOST_TO_DEVICE);
177 if (st != MKIRT_SUCCESS) {177 if (st != MKIRT_SUCCESS) {
Mcore/fft/fftcore/fft_core_n.cpp+22-22
@@ -28,11 +28,11 @@
28using namespace AsdSip;28using namespace AsdSip;
29 29 
30constexpr int RADIXVEC_SIZE_THREE = 3;30constexpr int RADIXVEC_SIZE_THREE = 3;
31-constexpr int N_DOING_24 = 16777216; // pow(2, 24)31+constexpr int N_DOING_24 = 16777216; // pow(2, 24)
32constexpr int N_DOING_27 = 134217728; // pow(2, 27)32constexpr int N_DOING_27 = 134217728; // pow(2, 27)
33-constexpr int N_DOING_15 = 32768; // pow(2, 15)33+constexpr int N_DOING_15 = 32768; // pow(2, 15)
34-constexpr int N_DOING_19 = 524288; // pow(2, 19)34+constexpr int N_DOING_19 = 524288; // pow(2, 19)
35-constexpr int N_DOING_25 = 33554432; // pow(2, 25)35+constexpr int N_DOING_25 = 33554432; // pow(2, 25)
36constexpr int RADIX_INDEX2 = 2;36constexpr int RADIX_INDEX2 = 2;
37constexpr int RADIX_INDEX3 = 3;37constexpr int RADIX_INDEX3 = 3;
38constexpr int RADIX_INDEX4 = 4;38constexpr int RADIX_INDEX4 = 4;
@@ -60,16 +60,16 @@ constexpr int LOGN_27 = 27;
60 60 
61size_t FFTCoreN::EstimateWorkspaceSize()61size_t FFTCoreN::EstimateWorkspaceSize()
62{62{
63- const KernelInfo &kernelInfo = kernel->GetKernelInfo();63+ const KernelInfo& kernelInfo = kernel->GetKernelInfo();
64 return getAlignedSize(kernelInfo.GetTotalScratchSize());64 return getAlignedSize(kernelInfo.GetTotalScratchSize());
65}65}
66 66 
67-void FFTCoreN::Run(Tensor &input, Tensor &output, void *stream, workspace::Workspace &workspace)67+void FFTCoreN::Run(Tensor& input, Tensor& output, void* stream, workspace::Workspace& workspace)
68{68{
69- const KernelInfo &kernelInfo = kernel->GetKernelInfo();69+ const KernelInfo& kernelInfo = kernel->GetKernelInfo();
70 // set workspace70 // set workspace
71 size_t bufferSize = kernelInfo.GetTotalScratchSize();71 size_t bufferSize = kernelInfo.GetTotalScratchSize();
72- runInfo.SetScratchDeviceAddr((uint8_t *)workspace.allocate(bufferSize));72+ runInfo.SetScratchDeviceAddr((uint8_t*)workspace.allocate(bufferSize));
73 73 
74 runInfo.SetStream(stream);74 runInfo.SetStream(stream);
75 launchParam.GetInTensor(0).data = input.data;75 launchParam.GetInTensor(0).data = input.data;
@@ -119,7 +119,7 @@ void FFTCoreN::InitTilingArgs()
119 }119 }
120 120 
121 float batchDataSize = float(n) * 2 * 4 / 1024 / 1024;121 float batchDataSize = float(n) * 2 * 4 / 1024 / 1024;
122- float l2CacheSize = 92; // 910B4 cache size122+ float l2CacheSize = 92; // 910B4 cache size
123 repeatBatchSize = static_cast<uint32_t>(floor((l2CacheSize - 10 - CALCUL_TWO) / CALCUL_TWO / batchDataSize));123 repeatBatchSize = static_cast<uint32_t>(floor((l2CacheSize - 10 - CALCUL_TWO) / CALCUL_TWO / batchDataSize));
124 repeatBatchSize = repeatBatchSize < 1 ? problemDesc.batch : repeatBatchSize;124 repeatBatchSize = repeatBatchSize < 1 ? problemDesc.batch : repeatBatchSize;
125}125}
@@ -149,7 +149,7 @@ AspbStatus FFTCoreN::InitTactic()
149 ssyncTilingNum.push_back(x);149 ssyncTilingNum.push_back(x);
150 }150 }
151 OpParam::FftN param = {problemDesc.nDoing, problemDesc.batch, repeatBatchSize, problemDesc.forward, sradixVec,151 OpParam::FftN param = {problemDesc.nDoing, problemDesc.batch, repeatBatchSize, problemDesc.forward, sradixVec,
152- saicInputAddr, saivOutputAddr, slessTCopy, ssyncTilingNum};152+ saicInputAddr, saivOutputAddr, slessTCopy, ssyncTilingNum};
153 153 
154 Tensor tensorIn;154 Tensor tensorIn;
155 Tensor tensorOut;155 Tensor tensorOut;
@@ -163,7 +163,7 @@ AspbStatus FFTCoreN::InitTactic()
163 launchParam.AddInTensor(*index);163 launchParam.AddInTensor(*index);
164 launchParam.AddOutTensor(tensorOut);164 launchParam.AddOutTensor(tensorOut);
165 165 
166- Operation *op = Ops::Instance().GetOperationByName(std::string("FftNOperation"));166+ Operation* op = Ops::Instance().GetOperationByName(std::string("FftNOperation"));
167 if (op == nullptr) {167 if (op == nullptr) {
168 return AsdSip::ErrorType::ACL_ERROR_INTERNAL_ERROR;168 return AsdSip::ErrorType::ACL_ERROR_INTERNAL_ERROR;
169 }169 }
@@ -172,7 +172,7 @@ AspbStatus FFTCoreN::InitTactic()
172 ASDSIP_ECHECK(kernel != nullptr, "Get best kernel failed", AsdSip::ErrorType::ACL_ERROR_INTERNAL_ERROR);172 ASDSIP_ECHECK(kernel != nullptr, "Get best kernel failed", AsdSip::ErrorType::ACL_ERROR_INTERNAL_ERROR);
173 173 
174 // allocate and initialize tiling workspace174 // allocate and initialize tiling workspace
175- uint8_t *deviceLaunchBuffer = nullptr;175+ uint8_t* deviceLaunchBuffer = nullptr;
176 kernel->SetLaunchWithTiling(false);176 kernel->SetLaunchWithTiling(false);
177 uint32_t launchBufferSize = kernel->GetTilingSize(launchParam);177 uint32_t launchBufferSize = kernel->GetTilingSize(launchParam);
178 if (launchBufferSize == 0) {178 if (launchBufferSize == 0) {
@@ -189,7 +189,7 @@ AspbStatus FFTCoreN::InitTactic()
189 ASDSIP_LOG(ERROR) << "malloc device memory fail";189 ASDSIP_LOG(ERROR) << "malloc device memory fail";
190 return AsdSip::ErrorType::ACL_ERROR_INTERNAL_ERROR;190 return AsdSip::ErrorType::ACL_ERROR_INTERNAL_ERROR;
191 }191 }
192- deviceLaunchBuffer = static_cast<uint8_t *>(tempDevicePtr);192+ deviceLaunchBuffer = static_cast<uint8_t*>(tempDevicePtr);
193 st = MkiRtMemCopy(deviceLaunchBuffer, launchBufferSize, hostLaunchBuffer, launchBufferSize,193 st = MkiRtMemCopy(deviceLaunchBuffer, launchBufferSize, hostLaunchBuffer, launchBufferSize,
194 MKIRT_MEMCOPY_HOST_TO_DEVICE);194 MKIRT_MEMCOPY_HOST_TO_DEVICE);
195 if (st != MKIRT_SUCCESS) {195 if (st != MKIRT_SUCCESS) {
@@ -282,26 +282,26 @@ AspbStatus FFTCoreN::InitIndex()
282 tN = tN < N0 ? tN : N0;282 tN = tN < N0 ? tN : N0;
283 int64_t tilingNum = (tM / 2) * tN;283 int64_t tilingNum = (tM / 2) * tN;
284 284 
285- std::function<AsdSip::FFTensor *()> func = [=]() -> AsdSip::FFTensor* {285+ std::function<AsdSip::FFTensor*()> func = [=]() -> AsdSip::FFTensor* {
286- AsdSip::FFTensor *indexMatrixPtr = new(std::nothrow) AsdSip::FFTensor;286+ AsdSip::FFTensor* indexMatrixPtr = new (std::nothrow) AsdSip::FFTensor;
287 if (indexMatrixPtr == nullptr) {287 if (indexMatrixPtr == nullptr) {
288 ASDSIP_LOG(ERROR) << "indexMatrixPtr new failed";288 ASDSIP_LOG(ERROR) << "indexMatrixPtr new failed";
289 throw std::runtime_error("indexMatrixPtr new failed");289 throw std::runtime_error("indexMatrixPtr new failed");
290 }290 }
291- AsdSip::FFTensor &indexMatrix = *indexMatrixPtr;291+ AsdSip::FFTensor& indexMatrix = *indexMatrixPtr;
292 292 
293 indexMatrix.desc.dtype = TENSOR_DTYPE_INT32;293 indexMatrix.desc.dtype = TENSOR_DTYPE_INT32;
294 indexMatrix.desc.format = TENSOR_FORMAT_ND;294 indexMatrix.desc.format = TENSOR_FORMAT_ND;
295 indexMatrix.desc.dims = {tilingNum};295 indexMatrix.desc.dims = {tilingNum};
296 indexMatrix.dataSize = sizeof(int32_t) * tilingNum;296 indexMatrix.dataSize = sizeof(int32_t) * tilingNum;
297 297 
298- int32_t *indexMatrixHost = nullptr;298+ int32_t* indexMatrixHost = nullptr;
299 try {299 try {
300- indexMatrixHost = new int32_t[tilingNum];300+ indexMatrixHost = new int32_t[tilingNum];
301- } catch(std::bad_alloc& e) {301+ } catch (std::bad_alloc& e) {
302 delete indexMatrixPtr;302 delete indexMatrixPtr;
303- ASDSIP_LOG(ERROR) << "indexMatrixHost nalloc failed: " << e.what();303+ ASDSIP_LOG(ERROR) << "indexMatrixHost alloc failed: " << e.what();
304- throw std::runtime_error("indexMatrixHost nalloc failed:.");304+ throw std::runtime_error("indexMatrixHost alloc failed:.");
305 }305 }
306 306 
307 for (int64_t i = 0; i < tilingNum / 2; i++) {307 for (int64_t i = 0; i < tilingNum / 2; i++) {
@@ -355,7 +355,7 @@ bool FFTCoreN::PreAllocateInDevice()
355 355 
356void FFTCoreN::DestroyInDevice() const356void FFTCoreN::DestroyInDevice() const
357{357{
358- uint8_t *deviceBuffer = runInfo.GetTilingDeviceAddr();358+ uint8_t* deviceBuffer = runInfo.GetTilingDeviceAddr();
359 if (deviceBuffer != nullptr) {359 if (deviceBuffer != nullptr) {
360 MkiRtMemFreeDevice(deviceBuffer);360 MkiRtMemFreeDevice(deviceBuffer);
361 }361 }
Mcore/utils/common_check.cpp+34-29
@@ -8,42 +8,41 @@
8 * See LICENSE in the root of the software repository for the full text of the License.8 * See LICENSE in the root of the software repository for the full text of the License.
9 */9 */
10 10 
11+#include <optional>
11#include "utils/common_check.h"12#include "utils/common_check.h"
12#include "utils/assert.h"13#include "utils/assert.h"
13 14 
14using namespace AsdSip;15using namespace AsdSip;
15 16 
16-int64_t GetTensorNum(uint64_t dimsNum, int64_t *storageDims)17+std::optional<int64_t> GetTensorNum(uint64_t dimsNum, int64_t* storageDims)
17{18{
18 if (storageDims == nullptr) {19 if (storageDims == nullptr) {
19- return -1;20+ return std::nullopt;
20 }21 }
21- int64_t elenmentNum = 1;22+ int64_t elementNum = 1;
22 for (size_t i = 0; i < dimsNum; i++) {23 for (size_t i = 0; i < dimsNum; i++) {
23- elenmentNum *= storageDims[i];24+ elementNum *= storageDims[i];
24 }25 }
25- return elenmentNum;26+ return elementNum;
26}27}
27 28 
28-AsdSip::AspbStatus AsdTensorDtypeCheck(const aclTensor *tensor, const aclDataType expectedDtype)29+AsdSip::AspbStatus AsdTensorDtypeCheck(const aclTensor* tensor, const aclDataType expectedDtype)
29{30{
30 aclDataType dataType = aclDataType::ACL_DT_UNDEFINED;31 aclDataType dataType = aclDataType::ACL_DT_UNDEFINED;
31 CHECK_STATUS_WITH_ACL_RETURN(aclGetDataType(tensor, &dataType), "aclGetDataType");32 CHECK_STATUS_WITH_ACL_RETURN(aclGetDataType(tensor, &dataType), "aclGetDataType");
32 if (dataType != expectedDtype) {33 if (dataType != expectedDtype) {
33- ASDSIP_LOG(ERROR) << "Tensor expected dtype is " << expectedDtype34+ ASDSIP_LOG(ERROR) << "Tensor expected dtype is " << expectedDtype << " , but found is: " << dataType;
34- << " , but found is: " << dataType;
35 return AsdSip::ErrorType::ACL_ERROR_UNSUPPORTED_DATA_TYPE;35 return AsdSip::ErrorType::ACL_ERROR_UNSUPPORTED_DATA_TYPE;
36 }36 }
37 return AsdSip::ErrorType::ACL_SUCCESS;37 return AsdSip::ErrorType::ACL_SUCCESS;
38}38}
39 39 
40-AsdSip::AspbStatus AsdTensorShapeCheck(const aclTensor *tensor, const std::vector<int64_t> expectedDims)40+AsdSip::AspbStatus AsdTensorShapeCheck(const aclTensor* tensor, const std::vector<int64_t> expectedDims)
41{41{
42- int64_t *storageDims = nullptr;42+ int64_t* storageDims = nullptr;
43 uint64_t storageDimsNum = 0;43 uint64_t storageDimsNum = 0;
44 uint64_t expectedDimNum = expectedDims.size();44 uint64_t expectedDimNum = expectedDims.size();
45- CHECK_STATUS_WITH_ACL_RETURN(aclGetStorageShape(tensor, &storageDims, &storageDimsNum),45+ CHECK_STATUS_WITH_ACL_RETURN(aclGetStorageShape(tensor, &storageDims, &storageDimsNum), "aclGetStorageShape");
46- "aclGetStorageShape");
47 if (storageDimsNum != expectedDimNum) {46 if (storageDimsNum != expectedDimNum) {
48 delete[] storageDims;47 delete[] storageDims;
49 storageDims = nullptr;48 storageDims = nullptr;
@@ -54,7 +53,7 @@ AsdSip::AspbStatus AsdTensorShapeCheck(const aclTensor *tensor, const std::vecto
54 for (size_t i = 0; i < expectedDimNum; i++) {53 for (size_t i = 0; i < expectedDimNum; i++) {
55 if (*(storageDims + i) != expectedDims[i]) {54 if (*(storageDims + i) != expectedDims[i]) {
56 ASDSIP_LOG(ERROR) << "Tensor dim [" << i << "] value is " << *(storageDims + i)55 ASDSIP_LOG(ERROR) << "Tensor dim [" << i << "] value is " << *(storageDims + i)
57- << " , wthich is not be expected";56+ << " , which is not as expected";
58 delete[] storageDims;57 delete[] storageDims;
59 storageDims = nullptr;58 storageDims = nullptr;
60 return AsdSip::ErrorType::ACL_ERROR_INVALID_PARAM;59 return AsdSip::ErrorType::ACL_ERROR_INVALID_PARAM;
@@ -66,17 +65,17 @@ AsdSip::AspbStatus AsdTensorShapeCheck(const aclTensor *tensor, const std::vecto
66 return AsdSip::ErrorType::ACL_SUCCESS;65 return AsdSip::ErrorType::ACL_SUCCESS;
67}66}
68 67 
69-AsdSip::AspbStatus AsdTensorNumCheck(const aclTensor *tensor, int64_t expectedNum)68+AsdSip::AspbStatus AsdTensorNumCheck(const aclTensor* tensor, int64_t expectedNum)
70{69{
71- int64_t *storageDims = nullptr;70+ int64_t* storageDims = nullptr;
72 uint64_t storageDimsNum = 0;71 uint64_t storageDimsNum = 0;
73- CHECK_STATUS_WITH_ACL_RETURN(aclGetStorageShape(tensor, &storageDims, &storageDimsNum),72+ CHECK_STATUS_WITH_ACL_RETURN(aclGetStorageShape(tensor, &storageDims, &storageDimsNum), "aclGetStorageShape");
74- "aclGetStorageShape");73+ std::optional<int64_t> numOpt = GetTensorNum(storageDimsNum, storageDims);
75- int64_t num = GetTensorNum(storageDimsNum, storageDims);74+ if (!numOpt.has_value() || numOpt.value() != expectedNum) {
76- if (num != expectedNum) {
77 delete[] storageDims;75 delete[] storageDims;
78 storageDims = nullptr;76 storageDims = nullptr;
79- ASDSIP_LOG(ERROR) << "Tensor num is " << num << " , which is not be expected.";77+ ASDSIP_LOG(ERROR) << "Tensor num is " << (numOpt.has_value() ? std::to_string(numOpt.value()) : "invalid")
78+ << " , which is not as expected.";
80 return AsdSip::ErrorType::ACL_ERROR_INVALID_PARAM;79 return AsdSip::ErrorType::ACL_ERROR_INVALID_PARAM;
81 }80 }
82 delete[] storageDims;81 delete[] storageDims;
@@ -84,20 +83,26 @@ AsdSip::AspbStatus AsdTensorNumCheck(const aclTensor *tensor, int64_t expectedNu
84 return AsdSip::ErrorType::ACL_SUCCESS;83 return AsdSip::ErrorType::ACL_SUCCESS;
85}84}
86 85 
87-AsdSip::AspbStatus AsdTensorInvalidShapeCheck(const aclTensor *tensor)86+AsdSip::AspbStatus AsdTensorInvalidShapeCheck(const aclTensor* tensor)
88{87{
89- int64_t *storageDims = nullptr;88+ int64_t* storageDims = nullptr;
90 uint64_t storageDimsNum = 0;89 uint64_t storageDimsNum = 0;
91- CHECK_STATUS_WITH_ACL_RETURN(aclGetStorageShape(tensor, &storageDims, &storageDimsNum),90+ CHECK_STATUS_WITH_ACL_RETURN(aclGetStorageShape(tensor, &storageDims, &storageDimsNum), "aclGetStorageShape");
92- "aclGetStorageShape");91+ // 防护: 0 维张量(rank=0)时 storageDims 可能为 nullptr, 直接解引用会崩溃
93- if (*storageDims <= 0) {92+ if (storageDims == nullptr || storageDimsNum == 0) {
94- delete[] storageDims;93+ ASDSIP_LOG(ERROR) << "Tensor shape is invalid, storage dims is empty.";
95- storageDims = nullptr;
96- ASDSIP_LOG(ERROR) << "Tensor shape is invalid.";
97 return AsdSip::ErrorType::ACL_ERROR_INVALID_PARAM;94 return AsdSip::ErrorType::ACL_ERROR_INVALID_PARAM;
98 }95 }
96+ for (uint64_t i = 0; i < storageDimsNum; i++) {
97+ if (storageDims[i] <= 0) {
98+ delete[] storageDims;
99+ storageDims = nullptr;
100+ ASDSIP_LOG(ERROR) << "Tensor shape is invalid.";
101+ return AsdSip::ErrorType::ACL_ERROR_INVALID_PARAM;
102+ }
103+ }
99 104 
100 delete[] storageDims;105 delete[] storageDims;
101 storageDims = nullptr;106 storageDims = nullptr;
102 return AsdSip::ErrorType::ACL_SUCCESS;107 return AsdSip::ErrorType::ACL_SUCCESS;
103-}108+}
Mcore/utils/include/utils/common_check.h+19-18
@@ -11,43 +11,44 @@
11#define ASDSIP_COMMON_CHECK_H11#define ASDSIP_COMMON_CHECK_H
12 12 
13#include <cstdint>13#include <cstdint>
14+#include <optional>
14#include <vector>15#include <vector>
15#include "utils/aspb_status.h"16#include "utils/aspb_status.h"
16#include "utils/mem_base.h"17#include "utils/mem_base.h"
17#include "acl/acl.h"18#include "acl/acl.h"
18#include "aclnn/acl_meta.h"19#include "aclnn/acl_meta.h"
19 20 
20-int64_t GetTensorNum(uint64_t dimsNum, int64_t *storageDims);21+std::optional<int64_t> GetTensorNum(uint64_t dimsNum, int64_t* storageDims);
21 22 
22-AsdSip::AspbStatus AsdTensorDtypeCheck(const aclTensor *tensor, const aclDataType expectedDtype);23+AsdSip::AspbStatus AsdTensorDtypeCheck(const aclTensor* tensor, const aclDataType expectedDtype);
23 24 
24-AsdSip::AspbStatus AsdTensorShapeCheck(const aclTensor *tensor, const std::vector<int64_t> expectedDims);25+AsdSip::AspbStatus AsdTensorShapeCheck(const aclTensor* tensor, const std::vector<int64_t> expectedDims);
25 26 
26-AsdSip::AspbStatus AsdTensorNumCheck(const aclTensor *tensor, int64_t expectedNum);27+AsdSip::AspbStatus AsdTensorNumCheck(const aclTensor* tensor, int64_t expectedNum);
27 28 
28-AsdSip::AspbStatus AsdTensorInvalidShapeCheck(const aclTensor *tensor);29+AsdSip::AspbStatus AsdTensorInvalidShapeCheck(const aclTensor* tensor);
29 30 
30-#define SIP_OP_CHECK_DTYPE_NOT_MATCH(tensor, expectedDtype, ret) \31+#define SIP_OP_CHECK_DTYPE_NOT_MATCH(tensor, expectedDtype, ret) \
31 if (AsdTensorDtypeCheck(tensor, expectedDtype) != ErrorType::ACL_SUCCESS) { \32 if (AsdTensorDtypeCheck(tensor, expectedDtype) != ErrorType::ACL_SUCCESS) { \
32- ASDSIP_ELOG(ret) << #tensor << " get wrong dtype."; \33+ ASDSIP_ELOG(ret) << #tensor << " get wrong dtype."; \
33- return ret; \34+ return ret; \
34 }35 }
35 36 
36-#define SIP_OP_CHECK_SHAPE_NOT_MATCH(tensor, expectedDims, ret) \37+#define SIP_OP_CHECK_SHAPE_NOT_MATCH(tensor, expectedDims, ret) \
37 if (AsdTensorShapeCheck(tensor, expectedDims) != ErrorType::ACL_SUCCESS) { \38 if (AsdTensorShapeCheck(tensor, expectedDims) != ErrorType::ACL_SUCCESS) { \
38- ASDSIP_ELOG(ret) << #tensor << " get wrong shape."; \39+ ASDSIP_ELOG(ret) << #tensor << " get wrong shape."; \
39- return ret; \40+ return ret; \
40 }41 }
41 42 
42-#define SIP_OP_CHECK_NUM_NOT_MATCH(tensor, expectedNum, ret) \43+#define SIP_OP_CHECK_NUM_NOT_MATCH(tensor, expectedNum, ret) \
43 if (AsdTensorNumCheck(tensor, expectedNum) != ErrorType::ACL_SUCCESS) { \44 if (AsdTensorNumCheck(tensor, expectedNum) != ErrorType::ACL_SUCCESS) { \
44- ASDSIP_ELOG(ret) << #tensor << " get wrong element num."; \45+ ASDSIP_ELOG(ret) << #tensor << " get wrong element num."; \
45- return ret; \46+ return ret; \
46 }47 }
47 48 
48-#define SIP_OP_CHECK_INVALID_SHAPE(tensor, ret) \49+#define SIP_OP_CHECK_INVALID_SHAPE(tensor, ret) \
49 if (AsdTensorInvalidShapeCheck(tensor) != ErrorType::ACL_SUCCESS) { \50 if (AsdTensorInvalidShapeCheck(tensor) != ErrorType::ACL_SUCCESS) { \
50- ASDSIP_ELOG(ret) << #tensor << " get invalid shape."; \51+ ASDSIP_ELOG(ret) << #tensor << " get invalid shape."; \
51- return ret; \52+ return ret; \
52 }53 }
53-#endif54+#endif
Mcore/utils/include/utils/the_tensor.h+24-51
@@ -18,24 +18,21 @@ namespace wten {
18template <typename T>18template <typename T>
19class TheTensor {19class TheTensor {
20public:20public:
21- TheTensor(const std::vector<int64_t> &shape)21+ // 禁用拷贝: data_ 为裸指针, 浅拷贝会导致析构时 double-free
22- : shape_(shape),22+ TheTensor(const TheTensor&) = delete;
23- strides_(std::vector<int64_t>(shape.size())),23+ TheTensor& operator=(const TheTensor&) = delete;
24- size_(compute_size(shape)),24+ 
25- data_(new T[size_])25+ TheTensor(const std::vector<int64_t>& shape)
26+ : shape_(shape), strides_(std::vector<int64_t>(shape.size())), size_(compute_size(shape)), data_(new T[size_])
26 {27 {
27 update_strides();28 update_strides();
28 }29 }
29 30 
30- TheTensor(TheTensor &&tensor)31+ TheTensor(TheTensor&& tensor)
31- : shape_(tensor.shape_),32+ : shape_(tensor.shape_), strides_(tensor.strides_), size_(tensor.size_), data_(tensor.move_data())
32- strides_(tensor.strides_),33+ {}
33- size_(tensor.size_),
34- data_(tensor.move_data())
35- {
36- }
37 34 
38- TheTensor<T> &operator=(TheTensor<T> &&tensor)35+ TheTensor<T>& operator=(TheTensor<T>&& tensor)
39 {36 {
40 delete[] data_;37 delete[] data_;
41 data_ = tensor.data_;38 data_ = tensor.data_;
@@ -46,49 +43,28 @@ public:
46 return *this;43 return *this;
47 }44 }
48 45 
49- T &operator[](size_t index) const46+ T& operator[](size_t index) const { return data_[index]; }
50- {
51- return data_[index];
52- }
53 47 
54- T *data() const48+ T* data() const { return data_; }
55- {
56- return data_;
57- }
58 49 
59- T *move_data()50+ T* move_data()
60 {51 {
61- T *data = data_;52+ T* data = data_;
62 data_ = nullptr;53 data_ = nullptr;
63 return data;54 return data;
64 }55 }
65 56 
66- void free_data()57+ void free_data() { delete[] move_data(); }
67- {
68- delete[] move_data();
69- }
70 58 
71- size_t size() const59+ size_t size() const { return size_; }
72- {
73- return size_;
74- }
75 60 
76- size_t ndim() const61+ size_t ndim() const { return shape_.size(); }
77- {
78- return shape_.size();
79- }
80 62 
81- const std::vector<int64_t> &shape() const63+ const std::vector<int64_t>& shape() const { return shape_; }
82- {
83- return shape_;
84- }
85 64 
86- const std::vector<int64_t> &strides() const65+ const std::vector<int64_t>& strides() const { return strides_; }
87- {
88- return strides_;
89- }
90 66 
91- void reshape(const std::vector<int64_t> &shape)67+ void reshape(const std::vector<int64_t>& shape)
92 {68 {
93 if (size_ != compute_size(shape)) {69 if (size_ != compute_size(shape)) {
94 ASDSIP_LOG(ERROR) << "Invalid shape.";70 ASDSIP_LOG(ERROR) << "Invalid shape.";
@@ -99,18 +75,15 @@ public:
99 update_strides();75 update_strides();
100 }76 }
101 77 
102- ~TheTensor()78+ ~TheTensor() { delete[] data_; };
103- {
104- delete [] data_;
105- };
106 79 
107private:80private:
108 std::vector<int64_t> shape_;81 std::vector<int64_t> shape_;
109 std::vector<int64_t> strides_;82 std::vector<int64_t> strides_;
110 size_t size_;83 size_t size_;
111- T *data_;84+ T* data_;
112 85 
113- size_t compute_size(const std::vector<int64_t> &shape) const86+ size_t compute_size(const std::vector<int64_t>& shape) const
114 {87 {
115 size_t size = 1;88 size_t size = 1;
116 for (size_t dim : shape) {89 for (size_t dim : shape) {
@@ -131,4 +104,4 @@ private:
131 }104 }
132};105};
133 106 
134-}107+} // namespace wten
Mdocs/zh/Installation_Operation_Guide/environment_variable.md+9-5
@@ -50,8 +50,12 @@
50 <td>设置应用类日志的日志级别及各模块日志级别，仅支持调试日志。</td>50 <td>设置应用类日志的日志级别及各模块日志级别，仅支持调试日志。</td>
51 </tr>51 </tr>
52 <tr>52 <tr>
53- <td>ASCEND_MODULE_LOG_LEVEL</td>53+ <td>ASCEND_MODULE_LOG_LEVEL</td>
54- <td>设置应用类日志的各模块日志级别，仅支持调试日志。</td>54+ <td>设置应用类日志的各模块日志级别，仅支持调试日志。</td>
55- </tr>55+ </tr>
56- </tbody>56+ </tbody>
57- </table>57+ </table>
58+ 
59+- **环境变量长度限制：**
60+ 
61+ 环境变量值的长度上限为 **12800 字节**。超长的环境变量将被忽略（不影响其它变量），并在日志中输出 WARN 级别提示。若发现设置的环境变量未生效，请检查是否超出长度限制。
Mdocs/zh/context/SiP返回码.md+14-0
@@ -77,6 +77,20 @@
77 77 
78信号处理加速库的日志系统支持日志分级、日志输出到标准输出、日志输出到文件。78信号处理加速库的日志系统支持日志分级、日志输出到标准输出、日志输出到文件。
79 79 
80+> **说明**：不同接口对同类错误场景返回的错误码可能不同。例如，对无效 handle 场景，
81+> `asdBlasSetStream` 返回 `ACL_ERROR_INTERNAL_ERROR`，而 `asdFftSetStream` 返回 `ACL_ERROR_INVALID_PARAM`。
82+> 排查问题时请以各接口的日志输出为准。
83+ 
84+## 日志文件运维参数
85+ 
86+| 参数 | 值 | 说明 |
87+|------|----|------|
88+| 单个日志文件大小上限 | 20MB | 超过后自动滚动创建新文件 |
89+| 日志文件保留个数 | 50 个 | 超过后删除最旧的日志文件 |
90+| 磁盘剩余空间门限 | 1GB | 磁盘剩余空间低于该值时停止写入日志 |
91+ 
92+注：日志文件默认保存路径由环境变量 `ASCEND_PROCESS_LOG_PATH` 控制；环境变量值的长度上限为 12800 字节，超长的环境变量将被忽略并输出 WARN 日志。
93+ 
80- **日志分级**：94- **日志分级**：
81 95 
82日志的严重级别从高到低分为ERROR、WARN、INFO、DEBUG四个级别，如下表所示。日志级别由环境变量“ASCEND_GLOBAL_LOG_LEVEL”控制，默认为“INFO”。\96日志的严重级别从高到低分为ERROR、WARN、INFO、DEBUG四个级别，如下表所示。日志级别由环境变量“ASCEND_GLOBAL_LOG_LEVEL”控制，默认为“INFO”。\
Mexample/A2/BASE/asd_mul/example_mul_complex32.cpp+19-30
@@ -22,7 +22,7 @@ using namespace AsdSip;
22 AsdSip::AspbStatus err_ = (err); \22 AsdSip::AspbStatus err_ = (err); \
23 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \23 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
24 std::cout << "Execute failed." << std::endl; \24 std::cout << "Execute failed." << std::endl; \
25- exit(-1); \25+ return -1; \
26 } \26 } \
27 } while (0)27 } while (0)
28 28 
@@ -38,7 +38,7 @@ using namespace AsdSip;
38 printf(message, ##__VA_ARGS__); \38 printf(message, ##__VA_ARGS__); \
39 } while (0)39 } while (0)
40 40 
41-int64_t GetShapeSize(const std::vector<int64_t> &shape)41+int64_t GetShapeSize(const std::vector<int64_t>& shape)
42{42{
43 int64_t shapeSize = 1;43 int64_t shapeSize = 1;
44 for (auto i : shape) {44 for (auto i : shape) {
@@ -47,7 +47,7 @@ int64_t GetShapeSize(const std::vector<int64_t> &shape)
47 return shapeSize;47 return shapeSize;
48}48}
49 49 
50-int Init(int32_t deviceId, aclrtStream *stream)50+int Init(int32_t deviceId, aclrtStream* stream)
51{51{
52 // 固定写法，acl初始化52 // 固定写法，acl初始化
53 auto ret = aclInit(nullptr);53 auto ret = aclInit(nullptr);
@@ -60,8 +60,8 @@ int Init(int32_t deviceId, aclrtStream *stream)
60}60}
61 61 
62template <typename T>62template <typename T>
63-int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr,63+int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
64- aclDataType dataType, aclTensor **tensor)64+ aclDataType dataType, aclTensor** tensor)
65{65{
66 auto size = GetShapeSize(shape) * sizeof(T);66 auto size = GetShapeSize(shape) * sizeof(T);
67 // 调用aclrtMalloc申请device侧内存67 // 调用aclrtMalloc申请device侧内存
@@ -78,19 +78,12 @@ int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &
78 }78 }
79 79 
80 // 调用aclCreateTensor接口创建aclTensor80 // 调用aclCreateTensor接口创建aclTensor
81- *tensor = aclCreateTensor(shape.data(),81+ *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
82- shape.size(),82+ shape.data(), shape.size(), *deviceAddr);
83- dataType,
84- strides.data(),
85- 0,
86- aclFormat::ACL_FORMAT_ND,
87- shape.data(),
88- shape.size(),
89- *deviceAddr);
90 return 0;83 return 0;
91}84}
92 85 
93-void printTensor(const std::complex<op::fp16_t> *tensorData, int64_t nums)86+void printTensor(const std::complex<op::fp16_t>* tensorData, int64_t nums)
94{87{
95 for (int64_t i = 0; i < nums; i++) {88 for (int64_t i = 0; i < nums; i++) {
96 std::cout << "(" << (float)tensorData[i].real() << "," << (float)tensorData[i].imag() << ")" << " ";89 std::cout << "(" << (float)tensorData[i].real() << "," << (float)tensorData[i].imag() << ")" << " ";
@@ -98,7 +91,7 @@ void printTensor(const std::complex<op::fp16_t> *tensorData, int64_t nums)
98 std::cout << std::endl;91 std::cout << std::endl;
99}92}
100 93 
101-int main(int argc, char **argv)94+int main(int argc, char** argv)
102{95{
103 int deviceId = 0;96 int deviceId = 0;
104 97 
@@ -119,8 +112,7 @@ int main(int argc, char **argv)
119 for (int64_t i = 0; i < vecSize; i++) {112 for (int64_t i = 0; i < vecSize; i++) {
120 tensorInYData.push_back({(op::fp16_t)(22.0f + i), (op::fp16_t)(33.0f * (i + 1))});113 tensorInYData.push_back({(op::fp16_t)(22.0f + i), (op::fp16_t)(33.0f * (i + 1))});
121 }114 }
122- std::vector<std::complex<op::fp16_t>> tensorOutZData(115+ std::vector<std::complex<op::fp16_t>> tensorOutZData(vecSize, {(op::fp16_t)0.0f, (op::fp16_t)0.0f});
123- vecSize, {(op::fp16_t)0.0f, (op::fp16_t)0.0f});
124 116 
125 std::cout << "------- input X -------" << std::endl;117 std::cout << "------- input X -------" << std::endl;
126 printTensor(tensorInXData.data(), vecSize);118 printTensor(tensorInXData.data(), vecSize);
@@ -131,12 +123,12 @@ int main(int argc, char **argv)
131 std::vector<int64_t> yShape = {vecSize};123 std::vector<int64_t> yShape = {vecSize};
132 std::vector<int64_t> zShape = {vecSize};124 std::vector<int64_t> zShape = {vecSize};
133 125 
134- aclTensor *inputX = nullptr;126+ aclTensor* inputX = nullptr;
135- aclTensor *inputY = nullptr;127+ aclTensor* inputY = nullptr;
136- aclTensor *outputZ = nullptr;128+ aclTensor* outputZ = nullptr;
137- void *inputXDeviceAddr = nullptr;129+ void* inputXDeviceAddr = nullptr;
138- void *inputYDeviceAddr = nullptr;130+ void* inputYDeviceAddr = nullptr;
139- void *outputZDeviceAddr = nullptr;131+ void* outputZDeviceAddr = nullptr;
140 ret = CreateAclTensor(tensorInXData, xShape, &inputXDeviceAddr, aclDataType::ACL_COMPLEX32, &inputX);132 ret = CreateAclTensor(tensorInXData, xShape, &inputXDeviceAddr, aclDataType::ACL_COMPLEX32, &inputX);
141 CHECK_RET(ret == ::ACL_SUCCESS, return ret);133 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
142 ret = CreateAclTensor(tensorInYData, yShape, &inputYDeviceAddr, aclDataType::ACL_COMPLEX32, &inputY);134 ret = CreateAclTensor(tensorInYData, yShape, &inputYDeviceAddr, aclDataType::ACL_COMPLEX32, &inputY);
@@ -149,11 +141,8 @@ int main(int argc, char **argv)
149 ret = aclrtSynchronizeStream(stream);141 ret = aclrtSynchronizeStream(stream);
150 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("aclrtSynchronizeStream failed. ERROR: %d\n", ret); return ret);142 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("aclrtSynchronizeStream failed. ERROR: %d\n", ret); return ret);
151 143 
152- ret = aclrtMemcpy(tensorOutZData.data(),144+ ret = aclrtMemcpy(tensorOutZData.data(), vecSize * sizeof(std::complex<op::fp16_t>), outputZDeviceAddr,
153- vecSize * sizeof(std::complex<op::fp16_t>),145+ vecSize * sizeof(std::complex<op::fp16_t>), ACL_MEMCPY_DEVICE_TO_HOST);
154- outputZDeviceAddr,
155- vecSize * sizeof(std::complex<op::fp16_t>),
156- ACL_MEMCPY_DEVICE_TO_HOST);
157 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("copy z from device to host failed. ERROR: %d\n", ret); return ret);146 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("copy z from device to host failed. ERROR: %d\n", ret); return ret);
158 std::cout << "------- output Z -------" << std::endl;147 std::cout << "------- output Z -------" << std::endl;
159 148 
@@ -170,4 +159,4 @@ int main(int argc, char **argv)
170 aclrtResetDevice(deviceId);159 aclrtResetDevice(deviceId);
171 aclFinalize();160 aclFinalize();
172 return 0;161 return 0;
173-}162+}
Mexample/A2/BASE/asd_mul/example_mul_complex64.cpp+1-1
@@ -22,7 +22,7 @@ using namespace AsdSip;
22 AsdSip::AspbStatus err_ = (err); \22 AsdSip::AspbStatus err_ = (err); \
23 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \23 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
24 std::cout << "Execute failed." << std::endl; \24 std::cout << "Execute failed." << std::endl; \
25- exit(-1); \25+ return -1; \
26 } \26 } \
27 } while (0)27 } while (0)
28 28 
Mexample/A2/BASE/swap_last2_axes/example_swap_last2_axes.cpp+1-1
@@ -21,7 +21,7 @@ using namespace AsdSip;
21 AsdSip::AspbStatus err_ = (err); \21 AsdSip::AspbStatus err_ = (err); \
22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
23 std::cout << "Execute failed." << std::endl; \23 std::cout << "Execute failed." << std::endl; \
24- exit(-1); \24+ return -1; \
25 } \25 } \
26 } while (0)26 } while (0)
27 27 
Mexample/A2/BLAS/asum/example_sasum.cpp+17-27
@@ -23,7 +23,7 @@ using namespace AsdSip;
23 AsdSip::AspbStatus err_ = (err); \23 AsdSip::AspbStatus err_ = (err); \
24 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \24 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
25 std::cout << "Execute failed." << std::endl; \25 std::cout << "Execute failed." << std::endl; \
26- exit(-1); \26+ return -1; \
27 } \27 } \
28 } while (0)28 } while (0)
29 29 
@@ -39,7 +39,7 @@ using namespace AsdSip;
39 printf(message, ##__VA_ARGS__); \39 printf(message, ##__VA_ARGS__); \
40 } while (0)40 } while (0)
41 41 
42-int64_t GetShapeSize(const std::vector<int64_t> &shape)42+int64_t GetShapeSize(const std::vector<int64_t>& shape)
43{43{
44 int64_t shapeSize = 1;44 int64_t shapeSize = 1;
45 for (auto i : shape) {45 for (auto i : shape) {
@@ -48,7 +48,7 @@ int64_t GetShapeSize(const std::vector<int64_t> &shape)
48 return shapeSize;48 return shapeSize;
49}49}
50 50 
51-int Init(int32_t deviceId, aclrtStream *stream)51+int Init(int32_t deviceId, aclrtStream* stream)
52{52{
53 // 固定写法，acl初始化53 // 固定写法，acl初始化
54 auto ret = aclInit(nullptr);54 auto ret = aclInit(nullptr);
@@ -61,8 +61,8 @@ int Init(int32_t deviceId, aclrtStream *stream)
61}61}
62 62 
63template <typename T>63template <typename T>
64-int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr,64+int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
65- aclDataType dataType, aclTensor **tensor)65+ aclDataType dataType, aclTensor** tensor)
66{66{
67 auto size = GetShapeSize(shape) * sizeof(T);67 auto size = GetShapeSize(shape) * sizeof(T);
68 // 调用aclrtMalloc申请device侧内存68 // 调用aclrtMalloc申请device侧内存
@@ -79,19 +79,12 @@ int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &
79 }79 }
80 80 
81 // 调用aclCreateTensor接口创建aclTensor81 // 调用aclCreateTensor接口创建aclTensor
82- *tensor = aclCreateTensor(shape.data(),82+ *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
83- shape.size(),83+ shape.data(), shape.size(), *deviceAddr);
84- dataType,
85- strides.data(),
86- 0,
87- aclFormat::ACL_FORMAT_ND,
88- shape.data(),
89- shape.size(),
90- *deviceAddr);
91 return 0;84 return 0;
92}85}
93 86 
94-int main(int argc, char **argv)87+int main(int argc, char** argv)
95{88{
96 int deviceId = 0;89 int deviceId = 0;
97 90 
@@ -127,10 +120,10 @@ int main(int argc, char **argv)
127 std::vector<int64_t> xShape = {xSize};120 std::vector<int64_t> xShape = {xSize};
128 std::vector<int64_t> yShape = {ySize};121 std::vector<int64_t> yShape = {ySize};
129 122 
130- aclTensor *inputX = nullptr;123+ aclTensor* inputX = nullptr;
131- aclTensor *inputY = nullptr;124+ aclTensor* inputY = nullptr;
132- void *inputXDeviceAddr = nullptr;125+ void* inputXDeviceAddr = nullptr;
133- void *inputYDeviceAddr = nullptr;126+ void* inputYDeviceAddr = nullptr;
134 127 
135 ret = CreateAclTensor<float>(tensorInXData, xShape, &inputXDeviceAddr, aclDataType::ACL_FLOAT, &inputX);128 ret = CreateAclTensor<float>(tensorInXData, xShape, &inputXDeviceAddr, aclDataType::ACL_FLOAT, &inputX);
136 CHECK_RET(ret == ::ACL_SUCCESS, return ret);129 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
@@ -142,7 +135,7 @@ int main(int argc, char **argv)
142 asdBlasCreate(handle);135 asdBlasCreate(handle);
143 136 
144 size_t lwork = 0;137 size_t lwork = 0;
145- void *buffer = nullptr;138+ void* buffer = nullptr;
146 asdBlasMakeAsumPlan(handle);139 asdBlasMakeAsumPlan(handle);
147 asdBlasGetWorkspaceSize(handle, lwork);140 asdBlasGetWorkspaceSize(handle, lwork);
148 std::cout << "lwork = " << lwork << std::endl;141 std::cout << "lwork = " << lwork << std::endl;
@@ -157,12 +150,9 @@ int main(int argc, char **argv)
157 150 
158 asdBlasSynchronize(handle);151 asdBlasSynchronize(handle);
159 asdBlasDestroy(handle);152 asdBlasDestroy(handle);
160- 153+ 
161- ret = aclrtMemcpy(tensorInYData.data(),154+ ret = aclrtMemcpy(tensorInYData.data(), ySize * sizeof(float), inputYDeviceAddr, ySize * sizeof(float),
162- ySize * sizeof(float),155+ ACL_MEMCPY_DEVICE_TO_HOST);
163- inputYDeviceAddr,
164- ySize * sizeof(float),
165- ACL_MEMCPY_DEVICE_TO_HOST);
166 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return ret);156 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return ret);
167 157 
168 std::cout << "------- result -------" << std::endl;158 std::cout << "------- result -------" << std::endl;
@@ -178,4 +168,4 @@ int main(int argc, char **argv)
178 aclrtResetDevice(deviceId);168 aclrtResetDevice(deviceId);
179 aclFinalize();169 aclFinalize();
180 return 0;170 return 0;
181-}171+}
Mexample/A2/BLAS/asum/example_scasum.cpp+16-26
@@ -23,7 +23,7 @@ using namespace AsdSip;
23 AsdSip::AspbStatus err_ = (err); \23 AsdSip::AspbStatus err_ = (err); \
24 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \24 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
25 std::cout << "Execute failed." << std::endl; \25 std::cout << "Execute failed." << std::endl; \
26- exit(-1); \26+ return -1; \
27 } \27 } \
28 } while (0)28 } while (0)
29 29 
@@ -39,7 +39,7 @@ using namespace AsdSip;
39 printf(message, ##__VA_ARGS__); \39 printf(message, ##__VA_ARGS__); \
40 } while (0)40 } while (0)
41 41 
42-int64_t GetShapeSize(const std::vector<int64_t> &shape)42+int64_t GetShapeSize(const std::vector<int64_t>& shape)
43{43{
44 int64_t shapeSize = 1;44 int64_t shapeSize = 1;
45 for (auto i : shape) {45 for (auto i : shape) {
@@ -48,7 +48,7 @@ int64_t GetShapeSize(const std::vector<int64_t> &shape)
48 return shapeSize;48 return shapeSize;
49}49}
50 50 
51-int Init(int32_t deviceId, aclrtStream *stream)51+int Init(int32_t deviceId, aclrtStream* stream)
52{52{
53 // 固定写法，acl初始化53 // 固定写法，acl初始化
54 auto ret = aclInit(nullptr);54 auto ret = aclInit(nullptr);
@@ -61,8 +61,8 @@ int Init(int32_t deviceId, aclrtStream *stream)
61}61}
62 62 
63template <typename T>63template <typename T>
64-int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr,64+int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
65- aclDataType dataType, aclTensor **tensor)65+ aclDataType dataType, aclTensor** tensor)
66{66{
67 auto size = GetShapeSize(shape) * sizeof(T);67 auto size = GetShapeSize(shape) * sizeof(T);
68 // 调用aclrtMalloc申请device侧内存68 // 调用aclrtMalloc申请device侧内存
@@ -79,19 +79,12 @@ int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &
79 }79 }
80 80 
81 // 调用aclCreateTensor接口创建aclTensor81 // 调用aclCreateTensor接口创建aclTensor
82- *tensor = aclCreateTensor(shape.data(),82+ *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
83- shape.size(),83+ shape.data(), shape.size(), *deviceAddr);
84- dataType,
85- strides.data(),
86- 0,
87- aclFormat::ACL_FORMAT_ND,
88- shape.data(),
89- shape.size(),
90- *deviceAddr);
91 return 0;84 return 0;
92}85}
93 86 
94-int main(int argc, char **argv)87+int main(int argc, char** argv)
95{88{
96 int deviceId = 0;89 int deviceId = 0;
97 90 
@@ -127,10 +120,10 @@ int main(int argc, char **argv)
127 std::vector<int64_t> xShape = {xSize};120 std::vector<int64_t> xShape = {xSize};
128 std::vector<int64_t> yShape = {ySize};121 std::vector<int64_t> yShape = {ySize};
129 122 
130- aclTensor *inputX = nullptr;123+ aclTensor* inputX = nullptr;
131- aclTensor *inputY = nullptr;124+ aclTensor* inputY = nullptr;
132- void *inputXDeviceAddr = nullptr;125+ void* inputXDeviceAddr = nullptr;
133- void *inputYDeviceAddr = nullptr;126+ void* inputYDeviceAddr = nullptr;
134 127 
135 ret = CreateAclTensor(tensorInXData, xShape, &inputXDeviceAddr, aclDataType::ACL_COMPLEX64, &inputX);128 ret = CreateAclTensor(tensorInXData, xShape, &inputXDeviceAddr, aclDataType::ACL_COMPLEX64, &inputX);
136 CHECK_RET(ret == ::ACL_SUCCESS, return ret);129 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
@@ -142,7 +135,7 @@ int main(int argc, char **argv)
142 asdBlasCreate(handle);135 asdBlasCreate(handle);
143 136 
144 size_t lwork = 0;137 size_t lwork = 0;
145- void *buffer = nullptr;138+ void* buffer = nullptr;
146 asdBlasMakeAsumPlan(handle);139 asdBlasMakeAsumPlan(handle);
147 asdBlasGetWorkspaceSize(handle, lwork);140 asdBlasGetWorkspaceSize(handle, lwork);
148 std::cout << "lwork = " << lwork << std::endl;141 std::cout << "lwork = " << lwork << std::endl;
@@ -158,11 +151,8 @@ int main(int argc, char **argv)
158 asdBlasSynchronize(handle);151 asdBlasSynchronize(handle);
159 asdBlasDestroy(handle);152 asdBlasDestroy(handle);
160 153 
161- ret = aclrtMemcpy(tensorInYData.data(),154+ ret = aclrtMemcpy(tensorInYData.data(), ySize * sizeof(float), inputYDeviceAddr, ySize * sizeof(float),
162- ySize * sizeof(float),155+ ACL_MEMCPY_DEVICE_TO_HOST);
163- inputYDeviceAddr,
164- ySize * sizeof(float),
165- ACL_MEMCPY_DEVICE_TO_HOST);
166 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return ret);156 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return ret);
167 157 
168 std::cout << "------- result -------" << std::endl;158 std::cout << "------- result -------" << std::endl;
@@ -178,4 +168,4 @@ int main(int argc, char **argv)
178 aclrtResetDevice(deviceId);168 aclrtResetDevice(deviceId);
179 aclFinalize();169 aclFinalize();
180 return 0;170 return 0;
181-}171+}
Mexample/A2/BLAS/cal/example_cscal.cpp+1-1
@@ -22,7 +22,7 @@ using namespace AsdSip;
22 AsdSip::AspbStatus err_ = (err); \22 AsdSip::AspbStatus err_ = (err); \
23 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \23 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
24 std::cout << "Execute failed." << std::endl; \24 std::cout << "Execute failed." << std::endl; \
25- exit(-1); \25+ return -1; \
26 } else { \26 } else { \
27 std::cout << "Execute successfully." << std::endl; \27 std::cout << "Execute successfully." << std::endl; \
28 } \28 } \
Mexample/A2/BLAS/cal/example_csscal.cpp+1-1
@@ -24,7 +24,7 @@ using namespace AsdSip;
24 AsdSip::AspbStatus err_ = (err); \24 AsdSip::AspbStatus err_ = (err); \
25 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \25 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
26 std::cout << "Execute failed." << std::endl; \26 std::cout << "Execute failed." << std::endl; \
27- exit(-1); \27+ return -1; \
28 } else { \28 } else { \
29 std::cout << "Execute successfully." << std::endl; \29 std::cout << "Execute successfully." << std::endl; \
30 } \30 } \
Mexample/A2/BLAS/cal/example_sscal.cpp+1-1
@@ -21,7 +21,7 @@ using namespace AsdSip;
21 AsdSip::AspbStatus err_ = (err); \21 AsdSip::AspbStatus err_ = (err); \
22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
23 std::cout << "Execute failed." << std::endl; \23 std::cout << "Execute failed." << std::endl; \
24- exit(-1); \24+ return -1; \
25 } \25 } \
26 } while (0)26 } while (0)
27 27 
Mexample/A2/BLAS/caxpy/example_caxpy.cpp+20-31
@@ -20,7 +20,7 @@ using namespace AsdSip;
20 AsdSip::AspbStatus err_ = (err); \20 AsdSip::AspbStatus err_ = (err); \
21 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \21 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
22 std::cout << "Execute failed." << std::endl; \22 std::cout << "Execute failed." << std::endl; \
23- exit(-1); \23+ return -1; \
24 } \24 } \
25 } while (0)25 } while (0)
26 26 
@@ -36,7 +36,7 @@ using namespace AsdSip;
36 printf(message, ##__VA_ARGS__); \36 printf(message, ##__VA_ARGS__); \
37 } while (0)37 } while (0)
38 38 
39-int64_t GetShapeSize(const std::vector<int64_t> &shape)39+int64_t GetShapeSize(const std::vector<int64_t>& shape)
40{40{
41 int64_t shapeSize = 1;41 int64_t shapeSize = 1;
42 for (auto i : shape) {42 for (auto i : shape) {
@@ -45,7 +45,7 @@ int64_t GetShapeSize(const std::vector<int64_t> &shape)
45 return shapeSize;45 return shapeSize;
46}46}
47 47 
48-int Init(int32_t deviceId, aclrtStream *stream)48+int Init(int32_t deviceId, aclrtStream* stream)
49{49{
50 // 固定写法，acl初始化50 // 固定写法，acl初始化
51 auto ret = aclInit(nullptr);51 auto ret = aclInit(nullptr);
@@ -58,8 +58,8 @@ int Init(int32_t deviceId, aclrtStream *stream)
58}58}
59 59 
60template <typename T>60template <typename T>
61-int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr,61+int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
62- aclDataType dataType, aclTensor **tensor)62+ aclDataType dataType, aclTensor** tensor)
63{63{
64 auto size = GetShapeSize(shape) * sizeof(T);64 auto size = GetShapeSize(shape) * sizeof(T);
65 // 调用aclrtMalloc申请device侧内存65 // 调用aclrtMalloc申请device侧内存
@@ -76,15 +76,8 @@ int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &
76 }76 }
77 77 
78 // 调用aclCreateTensor接口创建aclTensor78 // 调用aclCreateTensor接口创建aclTensor
79- *tensor = aclCreateTensor(shape.data(),79+ *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
80- shape.size(),80+ shape.data(), shape.size(), *deviceAddr);
81- dataType,
82- strides.data(),
83- 0,
84- aclFormat::ACL_FORMAT_ND,
85- shape.data(),
86- shape.size(),
87- *deviceAddr);
88 return 0;81 return 0;
89}82}
90 83 
@@ -96,9 +89,8 @@ void printTensor(std::vector<std::complex<float>> tensorData, int64_t tensorSize
96 std::cout << std::endl;89 std::cout << std::endl;
97}90}
98 91 
99-int main(int argc, char **argv)92+int main(int argc, char** argv)
100{93{
101- 
102 int deviceId = 0;94 int deviceId = 0;
103 95 
104 aclrtStream stream;96 aclrtStream stream;
@@ -129,24 +121,24 @@ int main(int argc, char **argv)
129 std::vector<int64_t> vecShape = {vecSize};121 std::vector<int64_t> vecShape = {vecSize};
130 std::vector<int64_t> resultShape = {resultSize};122 std::vector<int64_t> resultShape = {resultSize};
131 123 
132- aclTensor *inVec = nullptr;124+ aclTensor* inVec = nullptr;
133- aclTensor *outResult = nullptr;125+ aclTensor* outResult = nullptr;
134- void *inpVecDeviceAddr = nullptr;126+ void* inpVecDeviceAddr = nullptr;
135- void *outResultDeviceAddr = nullptr;127+ void* outResultDeviceAddr = nullptr;
136 128 
137- ret = CreateAclTensor<std::complex<float>>(129+ ret = CreateAclTensor<std::complex<float>>(tensorInVecData, vecShape, &inpVecDeviceAddr, aclDataType::ACL_COMPLEX64,
138- tensorInVecData, vecShape, &inpVecDeviceAddr, aclDataType::ACL_COMPLEX64, &inVec);130+ &inVec);
139 CHECK_RET(ret == ::ACL_SUCCESS, return ret);131 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
140 132 
141- ret = CreateAclTensor<std::complex<float>>(133+ ret = CreateAclTensor<std::complex<float>>(tensorResultData, resultShape, &outResultDeviceAddr,
142- tensorResultData, resultShape, &outResultDeviceAddr, aclDataType::ACL_COMPLEX64, &outResult);134+ aclDataType::ACL_COMPLEX64, &outResult);
143 CHECK_RET(ret == ::ACL_SUCCESS, return ret);135 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
144 136 
145 asdBlasHandle handle;137 asdBlasHandle handle;
146 asdBlasCreate(handle);138 asdBlasCreate(handle);
147 139 
148 size_t lwork = 0;140 size_t lwork = 0;
149- void *buffer = nullptr;141+ void* buffer = nullptr;
150 asdBlasMakeCaxpyPlan(handle);142 asdBlasMakeCaxpyPlan(handle);
151 asdBlasGetWorkspaceSize(handle, lwork);143 asdBlasGetWorkspaceSize(handle, lwork);
152 std::cout << "lwork = " << lwork << std::endl;144 std::cout << "lwork = " << lwork << std::endl;
@@ -162,11 +154,8 @@ int main(int argc, char **argv)
162 asdBlasSynchronize(handle);154 asdBlasSynchronize(handle);
163 asdBlasDestroy(handle);155 asdBlasDestroy(handle);
164 156 
165- ret = aclrtMemcpy(tensorResultData.data(),157+ ret = aclrtMemcpy(tensorResultData.data(), resultSize * sizeof(std::complex<float>), outResultDeviceAddr,
166- resultSize * sizeof(std::complex<float>),158+ resultSize * sizeof(std::complex<float>), ACL_MEMCPY_DEVICE_TO_HOST);
167- outResultDeviceAddr,
168- resultSize * sizeof(std::complex<float>),
169- ACL_MEMCPY_DEVICE_TO_HOST);
170 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return ret);159 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return ret);
171 160 
172 std::cout << "------- result -------" << std::endl;161 std::cout << "------- result -------" << std::endl;
@@ -181,4 +170,4 @@ int main(int argc, char **argv)
181 aclrtResetDevice(deviceId);170 aclrtResetDevice(deviceId);
182 aclFinalize();171 aclFinalize();
183 return 0;172 return 0;
184-}173+}
Mexample/A2/BLAS/cgemm/example_cgemm.cpp+24-34
@@ -21,7 +21,7 @@ using namespace AsdSip;
21 AsdSip::AspbStatus err_ = (err); \21 AsdSip::AspbStatus err_ = (err); \
22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
23 std::cout << "Execute failed." << std::endl; \23 std::cout << "Execute failed." << std::endl; \
24- exit(-1); \24+ return -1; \
25 } else { \25 } else { \
26 std::cout << "Execute successfully." << std::endl; \26 std::cout << "Execute successfully." << std::endl; \
27 } \27 } \
@@ -39,7 +39,7 @@ using namespace AsdSip;
39 printf(message, ##__VA_ARGS__); \39 printf(message, ##__VA_ARGS__); \
40 } while (0)40 } while (0)
41 41 
42-int64_t GetShapeSize(const std::vector<int64_t> &shape)42+int64_t GetShapeSize(const std::vector<int64_t>& shape)
43{43{
44 int64_t shapeSize = 1;44 int64_t shapeSize = 1;
45 for (auto i : shape) {45 for (auto i : shape) {
@@ -48,7 +48,7 @@ int64_t GetShapeSize(const std::vector<int64_t> &shape)
48 return shapeSize;48 return shapeSize;
49}49}
50 50 
51-int Init(int32_t deviceId, aclrtStream *stream)51+int Init(int32_t deviceId, aclrtStream* stream)
52{52{
53 // 固定写法，acl初始化53 // 固定写法，acl初始化
54 auto ret = aclInit(nullptr);54 auto ret = aclInit(nullptr);
@@ -61,8 +61,8 @@ int Init(int32_t deviceId, aclrtStream *stream)
61}61}
62 62 
63template <typename T>63template <typename T>
64-int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr,64+int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
65- aclDataType dataType, aclTensor **tensor)65+ aclDataType dataType, aclTensor** tensor)
66{66{
67 auto size = GetShapeSize(shape) * sizeof(T);67 auto size = GetShapeSize(shape) * sizeof(T);
68 // 调用aclrtMalloc申请device侧内存68 // 调用aclrtMalloc申请device侧内存
@@ -79,15 +79,8 @@ int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &
79 }79 }
80 80 
81 // 调用aclCreateTensor接口创建aclTensor81 // 调用aclCreateTensor接口创建aclTensor
82- *tensor = aclCreateTensor(shape.data(),82+ *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
83- shape.size(),83+ shape.data(), shape.size(), *deviceAddr);
84- dataType,
85- strides.data(),
86- 0,
87- aclFormat::ACL_FORMAT_ND,
88- shape.data(),
89- shape.size(),
90- *deviceAddr);
91 return 0;84 return 0;
92}85}
93 86 
@@ -101,7 +94,7 @@ void printTensor(std::vector<std::complex<float>> tensorData, int64_t rows, int6
101 }94 }
102}95}
103 96 
104-int main(int argc, char **argv)97+int main(int argc, char** argv)
105{98{
106 int deviceId = 0;99 int deviceId = 0;
107 100 
@@ -147,23 +140,23 @@ int main(int argc, char **argv)
147 std::vector<int64_t> matBShape = {k, n};140 std::vector<int64_t> matBShape = {k, n};
148 std::vector<int64_t> matCShape = {m, n};141 std::vector<int64_t> matCShape = {m, n};
149 142 
150- aclTensor *matA = nullptr;143+ aclTensor* matA = nullptr;
151- aclTensor *matB = nullptr;144+ aclTensor* matB = nullptr;
152- aclTensor *matC = nullptr;145+ aclTensor* matC = nullptr;
153- void *matADeviceAddr = nullptr;146+ void* matADeviceAddr = nullptr;
154- void *matBDeviceAddr = nullptr;147+ void* matBDeviceAddr = nullptr;
155- void *matCDeviceAddr = nullptr;148+ void* matCDeviceAddr = nullptr;
156 149 
157- ret = CreateAclTensor<std::complex<float>>(150+ ret = CreateAclTensor<std::complex<float>>(tensorInAData, matAShape, &matADeviceAddr, aclDataType::ACL_COMPLEX64,
158- tensorInAData, matAShape, &matADeviceAddr, aclDataType::ACL_COMPLEX64, &matA);151+ &matA);
159 CHECK_RET(ret == ::ACL_SUCCESS, return ret);152 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
160 153 
161- ret = CreateAclTensor<std::complex<float>>(154+ ret = CreateAclTensor<std::complex<float>>(tensorInBData, matBShape, &matBDeviceAddr, aclDataType::ACL_COMPLEX64,
162- tensorInBData, matBShape, &matBDeviceAddr, aclDataType::ACL_COMPLEX64, &matB);155+ &matB);
163 CHECK_RET(ret == ::ACL_SUCCESS, return ret);156 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
164 157 
165- ret = CreateAclTensor<std::complex<float>>(158+ ret = CreateAclTensor<std::complex<float>>(tensorInCData, matCShape, &matCDeviceAddr, aclDataType::ACL_COMPLEX64,
166- tensorInCData, matCShape, &matCDeviceAddr, aclDataType::ACL_COMPLEX64, &matC);159+ &matC);
167 CHECK_RET(ret == ::ACL_SUCCESS, return ret);160 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
168 161 
169 std::cout << "alpha = " << alpha << std::endl;162 std::cout << "alpha = " << alpha << std::endl;
@@ -177,7 +170,7 @@ int main(int argc, char **argv)
177 asdBlasCreate(handle);170 asdBlasCreate(handle);
178 171 
179 size_t lwork = 0;172 size_t lwork = 0;
180- void *buffer = nullptr;173+ void* buffer = nullptr;
181 asdBlasMakeCgemmPlan(handle, transA, transB, m, n, k, lda, ldb, ldc);174 asdBlasMakeCgemmPlan(handle, transA, transB, m, n, k, lda, ldb, ldc);
182 asdBlasGetWorkspaceSize(handle, lwork);175 asdBlasGetWorkspaceSize(handle, lwork);
183 std::cout << "lwork = " << lwork << std::endl;176 std::cout << "lwork = " << lwork << std::endl;
@@ -193,11 +186,8 @@ int main(int argc, char **argv)
193 asdBlasSynchronize(handle);186 asdBlasSynchronize(handle);
194 asdBlasDestroy(handle);187 asdBlasDestroy(handle);
195 188 
196- ret = aclrtMemcpy(tensorInCData.data(),189+ ret = aclrtMemcpy(tensorInCData.data(), tensorCSize * sizeof(std::complex<float>), matCDeviceAddr,
197- tensorCSize * sizeof(std::complex<float>),190+ tensorCSize * sizeof(std::complex<float>), ACL_MEMCPY_DEVICE_TO_HOST);
198- matCDeviceAddr,
199- tensorCSize * sizeof(std::complex<float>),
200- ACL_MEMCPY_DEVICE_TO_HOST);
201 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return ret);191 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return ret);
202 192 
203 std::cout << "------- output TensorInC -------" << std::endl;193 std::cout << "------- output TensorInC -------" << std::endl;
@@ -214,4 +204,4 @@ int main(int argc, char **argv)
214 aclrtResetDevice(deviceId);204 aclrtResetDevice(deviceId);
215 aclFinalize();205 aclFinalize();
216 return 0;206 return 0;
217-}207+}
Mexample/A2/BLAS/cgemm_batched/example_cgemm_batched.cpp+27-37
@@ -14,7 +14,6 @@
14#include "acl/acl.h"14#include "acl/acl.h"
15#include "acl_meta.h"15#include "acl_meta.h"
16 16 
17- 
18using namespace AsdSip;17using namespace AsdSip;
19 18 
20#define ASD_STATUS_CHECK(err) \19#define ASD_STATUS_CHECK(err) \
@@ -22,7 +21,7 @@ using namespace AsdSip;
22 AsdSip::AspbStatus err_ = (err); \21 AsdSip::AspbStatus err_ = (err); \
23 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
24 std::cout << "Execute failed." << std::endl; \23 std::cout << "Execute failed." << std::endl; \
25- exit(-1); \24+ return -1; \
26 } else { \25 } else { \
27 std::cout << "Execute successfully." << std::endl; \26 std::cout << "Execute successfully." << std::endl; \
28 } \27 } \
@@ -40,7 +39,7 @@ using namespace AsdSip;
40 printf(message, ##__VA_ARGS__); \39 printf(message, ##__VA_ARGS__); \
41 } while (0)40 } while (0)
42 41 
43-int64_t GetShapeSize(const std::vector<int64_t> &shape)42+int64_t GetShapeSize(const std::vector<int64_t>& shape)
44{43{
45 int64_t shapeSize = 1;44 int64_t shapeSize = 1;
46 for (auto i : shape) {45 for (auto i : shape) {
@@ -49,7 +48,7 @@ int64_t GetShapeSize(const std::vector<int64_t> &shape)
49 return shapeSize;48 return shapeSize;
50}49}
51 50 
52-int Init(int32_t deviceId, aclrtStream *stream)51+int Init(int32_t deviceId, aclrtStream* stream)
53{52{
54 // 固定写法，acl初始化53 // 固定写法，acl初始化
55 auto ret = aclInit(nullptr);54 auto ret = aclInit(nullptr);
@@ -62,8 +61,8 @@ int Init(int32_t deviceId, aclrtStream *stream)
62}61}
63 62 
64template <typename T>63template <typename T>
65-int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr,64+int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
66- aclDataType dataType, aclTensor **tensor)65+ aclDataType dataType, aclTensor** tensor)
67{66{
68 auto size = GetShapeSize(shape) * sizeof(T);67 auto size = GetShapeSize(shape) * sizeof(T);
69 // 调用aclrtMalloc申请device侧内存68 // 调用aclrtMalloc申请device侧内存
@@ -80,15 +79,8 @@ int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &
80 }79 }
81 80 
82 // 调用aclCreateTensor接口创建aclTensor81 // 调用aclCreateTensor接口创建aclTensor
83- *tensor = aclCreateTensor(shape.data(),82+ *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
84- shape.size(),83+ shape.data(), shape.size(), *deviceAddr);
85- dataType,
86- strides.data(),
87- 0,
88- aclFormat::ACL_FORMAT_ND,
89- shape.data(),
90- shape.size(),
91- *deviceAddr);
92 return 0;84 return 0;
93}85}
94 86 
@@ -106,7 +98,7 @@ void printTensor(std::vector<std::complex<float>> tensorData, int64_t batch, int
106 }98 }
107}99}
108 100 
109-int main(int argc, char **argv)101+int main(int argc, char** argv)
110{102{
111 int deviceId = 0;103 int deviceId = 0;
112 104 
@@ -153,26 +145,26 @@ int main(int argc, char **argv)
153 std::vector<int64_t> matBShape = {batch, k, n};145 std::vector<int64_t> matBShape = {batch, k, n};
154 std::vector<int64_t> matCShape = {batch, m, n};146 std::vector<int64_t> matCShape = {batch, m, n};
155 147 
156- aclTensor *matA = nullptr;148+ aclTensor* matA = nullptr;
157- aclTensor *matB = nullptr;149+ aclTensor* matB = nullptr;
158- aclTensor *matC = nullptr;150+ aclTensor* matC = nullptr;
159- void *matADeviceAddr = nullptr;151+ void* matADeviceAddr = nullptr;
160- void *matBDeviceAddr = nullptr;152+ void* matBDeviceAddr = nullptr;
161- void *matCDeviceAddr = nullptr;153+ void* matCDeviceAddr = nullptr;
162 154 
163- ret = CreateAclTensor<std::complex<float>>(155+ ret = CreateAclTensor<std::complex<float>>(tensorInAData, matAShape, &matADeviceAddr, aclDataType::ACL_COMPLEX64,
164- tensorInAData, matAShape, &matADeviceAddr, aclDataType::ACL_COMPLEX64, &matA);156+ &matA);
165 CHECK_RET(ret == ::ACL_SUCCESS, return ret);157 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
166 158 
167- ret = CreateAclTensor<std::complex<float>>(159+ ret = CreateAclTensor<std::complex<float>>(tensorInBData, matBShape, &matBDeviceAddr, aclDataType::ACL_COMPLEX64,
168- tensorInBData, matBShape, &matBDeviceAddr, aclDataType::ACL_COMPLEX64, &matB);160+ &matB);
169 CHECK_RET(ret == ::ACL_SUCCESS, return ret);161 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
170 162 
171- ret = CreateAclTensor<std::complex<float>>(163+ ret = CreateAclTensor<std::complex<float>>(tensorInCData, matCShape, &matCDeviceAddr, aclDataType::ACL_COMPLEX64,
172- tensorInCData, matCShape, &matCDeviceAddr, aclDataType::ACL_COMPLEX64, &matC);164+ &matC);
173 CHECK_RET(ret == ::ACL_SUCCESS, return ret);165 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
174- std::cout << "alpha = " << "(" << (float) alpha.real() << "," << (float) alpha.imag() << ")" << std::endl;166+ std::cout << "alpha = " << "(" << (float)alpha.real() << "," << (float)alpha.imag() << ")" << std::endl;
175- std::cout << "beta = " << "(" << (float) beta.real() << "," << (float) beta.imag() << ")" << std::endl;167+ std::cout << "beta = " << "(" << (float)beta.real() << "," << (float)beta.imag() << ")" << std::endl;
176 std::cout << "------- input TensorInA -------" << std::endl;168 std::cout << "------- input TensorInA -------" << std::endl;
177 printTensor(tensorInAData, batch, m, k);169 printTensor(tensorInAData, batch, m, k);
178 std::cout << "------- input TensorInB -------" << std::endl;170 std::cout << "------- input TensorInB -------" << std::endl;
@@ -182,7 +174,7 @@ int main(int argc, char **argv)
182 asdBlasCreate(handle);174 asdBlasCreate(handle);
183 175 
184 size_t lwork = 0;176 size_t lwork = 0;
185- void *buffer = nullptr;177+ void* buffer = nullptr;
186 asdBlasMakeCgemmBatchedPlan(handle);178 asdBlasMakeCgemmBatchedPlan(handle);
187 asdBlasGetWorkspaceSize(handle, lwork);179 asdBlasGetWorkspaceSize(handle, lwork);
188 std::cout << "lwork = " << lwork << std::endl;180 std::cout << "lwork = " << lwork << std::endl;
@@ -193,16 +185,14 @@ int main(int argc, char **argv)
193 asdBlasSetWorkspace(handle, buffer);185 asdBlasSetWorkspace(handle, buffer);
194 asdBlasSetStream(handle, stream);186 asdBlasSetStream(handle, stream);
195 187 
196- ASD_STATUS_CHECK(asdBlasCgemmBatched(handle, transA, transB, m, n, k, alpha, matA, lda, matB, ldb, beta, matC, ldc, batch));188+ ASD_STATUS_CHECK(
189+ asdBlasCgemmBatched(handle, transA, transB, m, n, k, alpha, matA, lda, matB, ldb, beta, matC, ldc, batch));
197 190 
198 asdBlasSynchronize(handle);191 asdBlasSynchronize(handle);
199 asdBlasDestroy(handle);192 asdBlasDestroy(handle);
200 193 
201- ret = aclrtMemcpy(tensorInCData.data(),194+ ret = aclrtMemcpy(tensorInCData.data(), tensorCSize * sizeof(std::complex<float>), matCDeviceAddr,
202- tensorCSize * sizeof(std::complex<float>),195+ tensorCSize * sizeof(std::complex<float>), ACL_MEMCPY_DEVICE_TO_HOST);
203- matCDeviceAddr,
204- tensorCSize * sizeof(std::complex<float>),
205- ACL_MEMCPY_DEVICE_TO_HOST);
206 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return ret);196 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return ret);
207 197 
208 std::cout << "------- output TensorInC -------" << std::endl;198 std::cout << "------- output TensorInC -------" << std::endl;
Mexample/A2/BLAS/cgemv/example_cgemv.cpp+1-1
@@ -22,7 +22,7 @@ using namespace AsdSip;
22 AsdSip::AspbStatus err_ = (err); \22 AsdSip::AspbStatus err_ = (err); \
23 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \23 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
24 std::cout << "Execute failed." << std::endl; \24 std::cout << "Execute failed." << std::endl; \
25- exit(-1); \25+ return -1; \
26 } else { \26 } else { \
27 std::cout << "Execute successfully." << std::endl; \27 std::cout << "Execute successfully." << std::endl; \
28 } \28 } \
Mexample/A2/BLAS/cgemv_batched/example_cgemv_batched.cpp+1-1
@@ -22,7 +22,7 @@ using namespace AsdSip;
22 AsdSip::AspbStatus err_ = (err); \22 AsdSip::AspbStatus err_ = (err); \
23 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \23 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
24 std::cout << "Execute failed." << std::endl; \24 std::cout << "Execute failed." << std::endl; \
25- exit(-1); \25+ return -1; \
26 } else { \26 } else { \
27 std::cout << "Execute successfully." << std::endl; \27 std::cout << "Execute successfully." << std::endl; \
28 } \28 } \
Mexample/A2/BLAS/cgerc/example_cgerc.cpp+1-1
@@ -23,7 +23,7 @@ using namespace AsdSip;
23 AsdSip::AspbStatus err_ = (err); \23 AsdSip::AspbStatus err_ = (err); \
24 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \24 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
25 std::cout << "Execute failed." << std::endl; \25 std::cout << "Execute failed." << std::endl; \
26- exit(-1); \26+ return -1; \
27 } else { \27 } else { \
28 std::cout << "Execute successfully." << std::endl; \28 std::cout << "Execute successfully." << std::endl; \
29 } \29 } \
Mexample/A2/BLAS/colwise_mul/example_colwise_mul.cpp+1-1
@@ -22,7 +22,7 @@ using namespace AsdSip;
22 AsdSip::AspbStatus err_ = (err); \22 AsdSip::AspbStatus err_ = (err); \
23 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \23 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
24 std::cout << "Execute failed." << std::endl; \24 std::cout << "Execute failed." << std::endl; \
25- exit(-1); \25+ return -1; \
26 } else { \26 } else { \
27 std::cout << "Execute successfully." << std::endl; \27 std::cout << "Execute successfully." << std::endl; \
28 } \28 } \
Mexample/A2/BLAS/complex_mat_dot/example_complex_mat_dot.cpp+1-1
@@ -22,7 +22,7 @@ using namespace AsdSip;
22 AsdSip::AspbStatus err_ = (err); \22 AsdSip::AspbStatus err_ = (err); \
23 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \23 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
24 std::cout << "Execute failed." << std::endl; \24 std::cout << "Execute failed." << std::endl; \
25- exit(-1); \25+ return -1; \
26 } else { \26 } else { \
27 std::cout << "Execute successfully." << std::endl; \27 std::cout << "Execute successfully." << std::endl; \
28 } \28 } \
Mexample/A2/BLAS/copy/example_ccopy.cpp+1-1
@@ -23,7 +23,7 @@ using namespace AsdSip;
23 AsdSip::AspbStatus err_ = (err); \23 AsdSip::AspbStatus err_ = (err); \
24 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \24 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
25 std::cout << "Execute failed." << std::endl; \25 std::cout << "Execute failed." << std::endl; \
26- exit(-1); \26+ return -1; \
27 } else { \27 } else { \
28 std::cout << "Execute successfully." << std::endl; \28 std::cout << "Execute successfully." << std::endl; \
29 } \29 } \
Mexample/A2/BLAS/copy/example_scopy.cpp+1-1
@@ -21,7 +21,7 @@ using namespace AsdSip;
21 AsdSip::AspbStatus err_ = (err); \21 AsdSip::AspbStatus err_ = (err); \
22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
23 std::cout << "Execute failed." << std::endl; \23 std::cout << "Execute failed." << std::endl; \
24- exit(-1); \24+ return -1; \
25 } \25 } \
26 } while (0)26 } while (0)
27 27 
Mexample/A2/BLAS/csrot/example_csrot.cpp+1-1
@@ -22,7 +22,7 @@ using namespace AsdSip;
22 AsdSip::AspbStatus err_ = (err); \22 AsdSip::AspbStatus err_ = (err); \
23 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \23 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
24 std::cout << "Execute failed." << std::endl; \24 std::cout << "Execute failed." << std::endl; \
25- exit(-1); \25+ return -1; \
26 } else { \26 } else { \
27 std::cout << "Execute successfully." << std::endl; \27 std::cout << "Execute successfully." << std::endl; \
28 } \28 } \
Mexample/A2/BLAS/ctrmv/example_ctrmv.cpp+1-1
@@ -21,7 +21,7 @@ using namespace AsdSip;
21 AsdSip::AspbStatus err_ = (err); \21 AsdSip::AspbStatus err_ = (err); \
22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
23 std::cout << "Execute failed." << std::endl; \23 std::cout << "Execute failed." << std::endl; \
24- exit(-1); \24+ return -1; \
25 } \25 } \
26 } while (0)26 } while (0)
27 27 
Mexample/A2/BLAS/dot/example_cdotc.cpp+1-1
@@ -25,7 +25,7 @@ using namespace AsdSip;
25 AsdSip::AspbStatus err_ = (err); \25 AsdSip::AspbStatus err_ = (err); \
26 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \26 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
27 std::cout << "Execute failed." << std::endl; \27 std::cout << "Execute failed." << std::endl; \
28- exit(-1); \28+ return -1; \
29 } else { \29 } else { \
30 std::cout << "Execute successfully." << std::endl; \30 std::cout << "Execute successfully." << std::endl; \
31 } \31 } \
Mexample/A2/BLAS/dot/example_cdotu.cpp+1-1
@@ -25,7 +25,7 @@ using namespace AsdSip;
25 AsdSip::AspbStatus err_ = (err); \25 AsdSip::AspbStatus err_ = (err); \
26 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \26 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
27 std::cout << "Execute failed." << std::endl; \27 std::cout << "Execute failed." << std::endl; \
28- exit(-1); \28+ return -1; \
29 } else { \29 } else { \
30 std::cout << "Execute successfully." << std::endl; \30 std::cout << "Execute successfully." << std::endl; \
31 } \31 } \
Mexample/A2/BLAS/dot/example_sdot.cpp+1-1
@@ -21,7 +21,7 @@ using namespace AsdSip;
21 AsdSip::AspbStatus err_ = (err); \21 AsdSip::AspbStatus err_ = (err); \
22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
23 std::cout << "Execute failed." << std::endl; \23 std::cout << "Execute failed." << std::endl; \
24- exit(-1); \24+ return -1; \
25 } \25 } \
26 } while (0)26 } while (0)
27 27 
Mexample/A2/BLAS/hcgemm_batched/example_hcgemm_batched.cpp+27-37
@@ -14,7 +14,6 @@
14#include "acl/acl.h"14#include "acl/acl.h"
15#include "acl_meta.h"15#include "acl_meta.h"
16 16 
17- 
18using namespace AsdSip;17using namespace AsdSip;
19 18 
20#define ASD_STATUS_CHECK(err) \19#define ASD_STATUS_CHECK(err) \
@@ -22,7 +21,7 @@ using namespace AsdSip;
22 AsdSip::AspbStatus err_ = (err); \21 AsdSip::AspbStatus err_ = (err); \
23 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
24 std::cout << "Execute failed." << std::endl; \23 std::cout << "Execute failed." << std::endl; \
25- exit(-1); \24+ return -1; \
26 } else { \25 } else { \
27 std::cout << "Execute successfully." << std::endl; \26 std::cout << "Execute successfully." << std::endl; \
28 } \27 } \
@@ -40,7 +39,7 @@ using namespace AsdSip;
40 printf(message, ##__VA_ARGS__); \39 printf(message, ##__VA_ARGS__); \
41 } while (0)40 } while (0)
42 41 
43-int64_t GetShapeSize(const std::vector<int64_t> &shape)42+int64_t GetShapeSize(const std::vector<int64_t>& shape)
44{43{
45 int64_t shapeSize = 1;44 int64_t shapeSize = 1;
46 for (auto i : shape) {45 for (auto i : shape) {
@@ -49,7 +48,7 @@ int64_t GetShapeSize(const std::vector<int64_t> &shape)
49 return shapeSize;48 return shapeSize;
50}49}
51 50 
52-int Init(int32_t deviceId, aclrtStream *stream)51+int Init(int32_t deviceId, aclrtStream* stream)
53{52{
54 // 固定写法，acl初始化53 // 固定写法，acl初始化
55 auto ret = aclInit(nullptr);54 auto ret = aclInit(nullptr);
@@ -62,8 +61,8 @@ int Init(int32_t deviceId, aclrtStream *stream)
62}61}
63 62 
64template <typename T>63template <typename T>
65-int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr,64+int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
66- aclDataType dataType, aclTensor **tensor)65+ aclDataType dataType, aclTensor** tensor)
67{66{
68 auto size = GetShapeSize(shape) * sizeof(T);67 auto size = GetShapeSize(shape) * sizeof(T);
69 // 调用aclrtMalloc申请device侧内存68 // 调用aclrtMalloc申请device侧内存
@@ -80,15 +79,8 @@ int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &
80 }79 }
81 80 
82 // 调用aclCreateTensor接口创建aclTensor81 // 调用aclCreateTensor接口创建aclTensor
83- *tensor = aclCreateTensor(shape.data(),82+ *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
84- shape.size(),83+ shape.data(), shape.size(), *deviceAddr);
85- dataType,
86- strides.data(),
87- 0,
88- aclFormat::ACL_FORMAT_ND,
89- shape.data(),
90- shape.size(),
91- *deviceAddr);
92 return 0;84 return 0;
93}85}
94 86 
@@ -106,7 +98,7 @@ void printTensor(std::vector<std::complex<op::fp16_t>> tensorData, int64_t batch
106 }98 }
107}99}
108 100 
109-int main(int argc, char **argv)101+int main(int argc, char** argv)
110{102{
111 int deviceId = 0;103 int deviceId = 0;
112 104 
@@ -153,26 +145,26 @@ int main(int argc, char **argv)
153 std::vector<int64_t> matBShape = {batch, k, n};145 std::vector<int64_t> matBShape = {batch, k, n};
154 std::vector<int64_t> matCShape = {batch, m, n};146 std::vector<int64_t> matCShape = {batch, m, n};
155 147 
156- aclTensor *matA = nullptr;148+ aclTensor* matA = nullptr;
157- aclTensor *matB = nullptr;149+ aclTensor* matB = nullptr;
158- aclTensor *matC = nullptr;150+ aclTensor* matC = nullptr;
159- void *matADeviceAddr = nullptr;151+ void* matADeviceAddr = nullptr;
160- void *matBDeviceAddr = nullptr;152+ void* matBDeviceAddr = nullptr;
161- void *matCDeviceAddr = nullptr;153+ void* matCDeviceAddr = nullptr;
162 154 
163- ret = CreateAclTensor<std::complex<op::fp16_t>>(155+ ret = CreateAclTensor<std::complex<op::fp16_t>>(tensorInAData, matAShape, &matADeviceAddr,
164- tensorInAData, matAShape, &matADeviceAddr, aclDataType::ACL_COMPLEX32, &matA);156+ aclDataType::ACL_COMPLEX32, &matA);
165 CHECK_RET(ret == ::ACL_SUCCESS, return ret);157 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
166 158 
167- ret = CreateAclTensor<std::complex<op::fp16_t>>(159+ ret = CreateAclTensor<std::complex<op::fp16_t>>(tensorInBData, matBShape, &matBDeviceAddr,
168- tensorInBData, matBShape, &matBDeviceAddr, aclDataType::ACL_COMPLEX32, &matB);160+ aclDataType::ACL_COMPLEX32, &matB);
169 CHECK_RET(ret == ::ACL_SUCCESS, return ret);161 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
170 162 
171- ret = CreateAclTensor<std::complex<op::fp16_t>>(163+ ret = CreateAclTensor<std::complex<op::fp16_t>>(tensorInCData, matCShape, &matCDeviceAddr,
172- tensorInCData, matCShape, &matCDeviceAddr, aclDataType::ACL_COMPLEX32, &matC);164+ aclDataType::ACL_COMPLEX32, &matC);
173 CHECK_RET(ret == ::ACL_SUCCESS, return ret);165 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
174- std::cout << "alpha = " << "(" << (float) alpha.real() << "," << (float) alpha.imag() << ")" << std::endl;166+ std::cout << "alpha = " << "(" << (float)alpha.real() << "," << (float)alpha.imag() << ")" << std::endl;
175- std::cout << "beta = " << "(" << (float) beta.real() << "," << (float) beta.imag() << ")" << std::endl;167+ std::cout << "beta = " << "(" << (float)beta.real() << "," << (float)beta.imag() << ")" << std::endl;
176 std::cout << "------- input TensorInA -------" << std::endl;168 std::cout << "------- input TensorInA -------" << std::endl;
177 printTensor(tensorInAData, batch, m, k);169 printTensor(tensorInAData, batch, m, k);
178 std::cout << "------- input TensorInB -------" << std::endl;170 std::cout << "------- input TensorInB -------" << std::endl;
@@ -182,7 +174,7 @@ int main(int argc, char **argv)
182 asdBlasCreate(handle);174 asdBlasCreate(handle);
183 175 
184 size_t lwork = 0;176 size_t lwork = 0;
185- void *buffer = nullptr;177+ void* buffer = nullptr;
186 asdBlasMakeHCgemmBatchedPlan(handle);178 asdBlasMakeHCgemmBatchedPlan(handle);
187 asdBlasGetWorkspaceSize(handle, lwork);179 asdBlasGetWorkspaceSize(handle, lwork);
188 std::cout << "lwork = " << lwork << std::endl;180 std::cout << "lwork = " << lwork << std::endl;
@@ -193,16 +185,14 @@ int main(int argc, char **argv)
193 asdBlasSetWorkspace(handle, buffer);185 asdBlasSetWorkspace(handle, buffer);
194 asdBlasSetStream(handle, stream);186 asdBlasSetStream(handle, stream);
195 187 
196- ASD_STATUS_CHECK(asdBlasHCgemmBatched(handle, transA, transB, m, n, k, alpha, matA, lda, matB, ldb, beta, matC, ldc, batch));188+ ASD_STATUS_CHECK(
189+ asdBlasHCgemmBatched(handle, transA, transB, m, n, k, alpha, matA, lda, matB, ldb, beta, matC, ldc, batch));
197 190 
198 asdBlasSynchronize(handle);191 asdBlasSynchronize(handle);
199 asdBlasDestroy(handle);192 asdBlasDestroy(handle);
200 193 
201- ret = aclrtMemcpy(tensorInCData.data(),194+ ret = aclrtMemcpy(tensorInCData.data(), tensorCSize * sizeof(std::complex<op::fp16_t>), matCDeviceAddr,
202- tensorCSize * sizeof(std::complex<op::fp16_t>),195+ tensorCSize * sizeof(std::complex<op::fp16_t>), ACL_MEMCPY_DEVICE_TO_HOST);
203- matCDeviceAddr,
204- tensorCSize * sizeof(std::complex<op::fp16_t>),
205- ACL_MEMCPY_DEVICE_TO_HOST);
206 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return ret);196 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return ret);
207 197 
208 std::cout << "------- output TensorInC -------" << std::endl;198 std::cout << "------- output TensorInC -------" << std::endl;
Mexample/A2/BLAS/hcgemv_batched/example_hcgemv_batched.cpp+1-1
@@ -22,7 +22,7 @@ using namespace AsdSip;
22 AsdSip::AspbStatus err_ = (err); \22 AsdSip::AspbStatus err_ = (err); \
23 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \23 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
24 std::cout << "Execute failed." << std::endl; \24 std::cout << "Execute failed." << std::endl; \
25- exit(-1); \25+ return -1; \
26 } else { \26 } else { \
27 std::cout << "Execute successfully." << std::endl; \27 std::cout << "Execute successfully." << std::endl; \
28 } \28 } \
Mexample/A2/BLAS/iamax/example_icamax.cpp+1-1
@@ -21,7 +21,7 @@ using namespace AsdSip;
21 AsdSip::AspbStatus err_ = (err); \21 AsdSip::AspbStatus err_ = (err); \
22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
23 std::cout << "Execute failed." << std::endl; \23 std::cout << "Execute failed." << std::endl; \
24- exit(-1); \24+ return -1; \
25 } \25 } \
26 } while (0)26 } while (0)
27 27 
Mexample/A2/BLAS/iamax/example_isamax.cpp+1-1
@@ -21,7 +21,7 @@ using namespace AsdSip;
21 AsdSip::AspbStatus err_ = (err); \21 AsdSip::AspbStatus err_ = (err); \
22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
23 std::cout << "Execute failed." << std::endl; \23 std::cout << "Execute failed." << std::endl; \
24- exit(-1); \24+ return -1; \
25 } \25 } \
26 } while (0)26 } while (0)
27 27 
Mexample/A2/BLAS/nrm2/example_scnrm2.cpp+1-1
@@ -21,7 +21,7 @@ using namespace AsdSip;
21 AsdSip::AspbStatus err_ = (err); \21 AsdSip::AspbStatus err_ = (err); \
22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
23 std::cout << "Execute failed." << std::endl; \23 std::cout << "Execute failed." << std::endl; \
24- exit(-1); \24+ return -1; \
25 } \25 } \
26 } while (0)26 } while (0)
27 27 
Mexample/A2/BLAS/nrm2/example_snrm2.cpp+1-1
@@ -21,7 +21,7 @@ using namespace AsdSip;
21 AsdSip::AspbStatus err_ = (err); \21 AsdSip::AspbStatus err_ = (err); \
22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
23 std::cout << "Execute failed." << std::endl; \23 std::cout << "Execute failed." << std::endl; \
24- exit(-1); \24+ return -1; \
25 } \25 } \
26 } while (0)26 } while (0)
27 27 
Mexample/A2/BLAS/ssyr/example_ssyr.cpp+16-26
@@ -21,7 +21,7 @@ using namespace AsdSip;
21 AsdSip::AspbStatus err_ = (err); \21 AsdSip::AspbStatus err_ = (err); \
22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
23 std::cout << "Execute failed." << std::endl; \23 std::cout << "Execute failed." << std::endl; \
24- exit(-1); \24+ return -1; \
25 } \25 } \
26 } while (0)26 } while (0)
27 27 
@@ -37,7 +37,7 @@ using namespace AsdSip;
37 printf(message, ##__VA_ARGS__); \37 printf(message, ##__VA_ARGS__); \
38 } while (0)38 } while (0)
39 39 
40-int64_t GetShapeSize(const std::vector<int64_t> &shape)40+int64_t GetShapeSize(const std::vector<int64_t>& shape)
41{41{
42 int64_t shapeSize = 1;42 int64_t shapeSize = 1;
43 for (auto i : shape) {43 for (auto i : shape) {
@@ -46,7 +46,7 @@ int64_t GetShapeSize(const std::vector<int64_t> &shape)
46 return shapeSize;46 return shapeSize;
47}47}
48 48 
49-int Init(int32_t deviceId, aclrtStream *stream)49+int Init(int32_t deviceId, aclrtStream* stream)
50{50{
51 // 固定写法，acl初始化51 // 固定写法，acl初始化
52 auto ret = aclInit(nullptr);52 auto ret = aclInit(nullptr);
@@ -59,8 +59,8 @@ int Init(int32_t deviceId, aclrtStream *stream)
59}59}
60 60 
61template <typename T>61template <typename T>
62-int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr,62+int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
63- aclDataType dataType, aclTensor **tensor)63+ aclDataType dataType, aclTensor** tensor)
64{64{
65 auto size = GetShapeSize(shape) * sizeof(T);65 auto size = GetShapeSize(shape) * sizeof(T);
66 // 调用aclrtMalloc申请device侧内存66 // 调用aclrtMalloc申请device侧内存
@@ -77,19 +77,12 @@ int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &
77 }77 }
78 78 
79 // 调用aclCreateTensor接口创建aclTensor79 // 调用aclCreateTensor接口创建aclTensor
80- *tensor = aclCreateTensor(shape.data(),80+ *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
81- shape.size(),81+ shape.data(), shape.size(), *deviceAddr);
82- dataType,
83- strides.data(),
84- 0,
85- aclFormat::ACL_FORMAT_ND,
86- shape.data(),
87- shape.size(),
88- *deviceAddr);
89 return 0;82 return 0;
90}83}
91 84 
92-int main(int argc, char **argv)85+int main(int argc, char** argv)
93{86{
94 int deviceId = 0;87 int deviceId = 0;
95 88 
@@ -136,10 +129,10 @@ int main(int argc, char **argv)
136 std::vector<int64_t> xShape = {n};129 std::vector<int64_t> xShape = {n};
137 std::vector<int64_t> matAShape = {n, n};130 std::vector<int64_t> matAShape = {n, n};
138 131 
139- aclTensor *inputX = nullptr;132+ aclTensor* inputX = nullptr;
140- aclTensor *inputA = nullptr;133+ aclTensor* inputA = nullptr;
141- void *inputXDeviceAddr = nullptr;134+ void* inputXDeviceAddr = nullptr;
142- void *inputADeviceAddr = nullptr;135+ void* inputADeviceAddr = nullptr;
143 136 
144 ret = CreateAclTensor<float>(tensorInXData, xShape, &inputXDeviceAddr, aclDataType::ACL_FLOAT, &inputX);137 ret = CreateAclTensor<float>(tensorInXData, xShape, &inputXDeviceAddr, aclDataType::ACL_FLOAT, &inputX);
145 CHECK_RET(ret == ::ACL_SUCCESS, return ret);138 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
@@ -151,7 +144,7 @@ int main(int argc, char **argv)
151 asdBlasCreate(handle);144 asdBlasCreate(handle);
152 145 
153 size_t lwork = 0;146 size_t lwork = 0;
154- void *buffer = nullptr;147+ void* buffer = nullptr;
155 asdBlasMakeSsyrPlan(handle);148 asdBlasMakeSsyrPlan(handle);
156 asdBlasGetWorkspaceSize(handle, lwork);149 asdBlasGetWorkspaceSize(handle, lwork);
157 std::cout << "lwork = " << lwork << std::endl;150 std::cout << "lwork = " << lwork << std::endl;
@@ -168,11 +161,8 @@ int main(int argc, char **argv)
168 asdBlasSynchronize(handle);161 asdBlasSynchronize(handle);
169 asdBlasDestroy(handle);162 asdBlasDestroy(handle);
170 163 
171- ret = aclrtMemcpy(tensorInAData.data(),164+ ret = aclrtMemcpy(tensorInAData.data(), n * n * sizeof(float), inputADeviceAddr, n * n * sizeof(float),
172- n * n * sizeof(float),165+ ACL_MEMCPY_DEVICE_TO_HOST);
173- inputADeviceAddr,
174- n * n * sizeof(float),
175- ACL_MEMCPY_DEVICE_TO_HOST);
176 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return ret);166 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return ret);
177 167 
178 std::cout << "------- output A -------" << std::endl;168 std::cout << "------- output A -------" << std::endl;
@@ -193,4 +183,4 @@ int main(int argc, char **argv)
193 aclrtResetDevice(deviceId);183 aclrtResetDevice(deviceId);
194 aclFinalize();184 aclFinalize();
195 return 0;185 return 0;
196-}186+}
Mexample/A2/BLAS/ssyr2/example_ssyr2.cpp+18-28
@@ -21,7 +21,7 @@ using namespace AsdSip;
21 AsdSip::AspbStatus err_ = (err); \21 AsdSip::AspbStatus err_ = (err); \
22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
23 std::cout << "Execute failed." << std::endl; \23 std::cout << "Execute failed." << std::endl; \
24- exit(-1); \24+ return -1; \
25 } \25 } \
26 } while (0)26 } while (0)
27 27 
@@ -37,7 +37,7 @@ using namespace AsdSip;
37 printf(message, ##__VA_ARGS__); \37 printf(message, ##__VA_ARGS__); \
38 } while (0)38 } while (0)
39 39 
40-int64_t GetShapeSize(const std::vector<int64_t> &shape)40+int64_t GetShapeSize(const std::vector<int64_t>& shape)
41{41{
42 int64_t shapeSize = 1;42 int64_t shapeSize = 1;
43 for (auto i : shape) {43 for (auto i : shape) {
@@ -46,7 +46,7 @@ int64_t GetShapeSize(const std::vector<int64_t> &shape)
46 return shapeSize;46 return shapeSize;
47}47}
48 48 
49-int Init(int32_t deviceId, aclrtStream *stream)49+int Init(int32_t deviceId, aclrtStream* stream)
50{50{
51 // 固定写法，acl初始化51 // 固定写法，acl初始化
52 auto ret = aclInit(nullptr);52 auto ret = aclInit(nullptr);
@@ -59,8 +59,8 @@ int Init(int32_t deviceId, aclrtStream *stream)
59}59}
60 60 
61template <typename T>61template <typename T>
62-int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr,62+int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
63- aclDataType dataType, aclTensor **tensor)63+ aclDataType dataType, aclTensor** tensor)
64{64{
65 auto size = GetShapeSize(shape) * sizeof(T);65 auto size = GetShapeSize(shape) * sizeof(T);
66 // 调用aclrtMalloc申请device侧内存66 // 调用aclrtMalloc申请device侧内存
@@ -77,19 +77,12 @@ int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &
77 }77 }
78 78 
79 // 调用aclCreateTensor接口创建aclTensor79 // 调用aclCreateTensor接口创建aclTensor
80- *tensor = aclCreateTensor(shape.data(),80+ *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
81- shape.size(),81+ shape.data(), shape.size(), *deviceAddr);
82- dataType,
83- strides.data(),
84- 0,
85- aclFormat::ACL_FORMAT_ND,
86- shape.data(),
87- shape.size(),
88- *deviceAddr);
89 return 0;82 return 0;
90}83}
91 84 
92-int main(int argc, char **argv)85+int main(int argc, char** argv)
93{86{
94 int deviceId = 0;87 int deviceId = 0;
95 88 
@@ -151,12 +144,12 @@ int main(int argc, char **argv)
151 std::vector<int64_t> yShape = {n};144 std::vector<int64_t> yShape = {n};
152 std::vector<int64_t> matAShape = {n, n};145 std::vector<int64_t> matAShape = {n, n};
153 146 
154- aclTensor *inputX = nullptr;147+ aclTensor* inputX = nullptr;
155- aclTensor *inputY = nullptr;148+ aclTensor* inputY = nullptr;
156- aclTensor *inputA = nullptr;149+ aclTensor* inputA = nullptr;
157- void *inputXDeviceAddr = nullptr;150+ void* inputXDeviceAddr = nullptr;
158- void *inputYDeviceAddr = nullptr;151+ void* inputYDeviceAddr = nullptr;
159- void *inputADeviceAddr = nullptr;152+ void* inputADeviceAddr = nullptr;
160 153 
161 ret = CreateAclTensor<float>(tensorInXData, xShape, &inputXDeviceAddr, aclDataType::ACL_FLOAT, &inputX);154 ret = CreateAclTensor<float>(tensorInXData, xShape, &inputXDeviceAddr, aclDataType::ACL_FLOAT, &inputX);
162 CHECK_RET(ret == ::ACL_SUCCESS, return ret);155 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
@@ -171,7 +164,7 @@ int main(int argc, char **argv)
171 asdBlasCreate(handle);164 asdBlasCreate(handle);
172 165 
173 size_t lwork = 0;166 size_t lwork = 0;
174- void *buffer = nullptr;167+ void* buffer = nullptr;
175 asdBlasMakeSsyr2Plan(handle);168 asdBlasMakeSsyr2Plan(handle);
176 asdBlasGetWorkspaceSize(handle, lwork);169 asdBlasGetWorkspaceSize(handle, lwork);
177 std::cout << "lwork = " << lwork << std::endl;170 std::cout << "lwork = " << lwork << std::endl;
@@ -187,11 +180,8 @@ int main(int argc, char **argv)
187 asdBlasSynchronize(handle);180 asdBlasSynchronize(handle);
188 asdBlasDestroy(handle);181 asdBlasDestroy(handle);
189 182 
190- ret = aclrtMemcpy(tensorInAData.data(),183+ ret = aclrtMemcpy(tensorInAData.data(), n * n * sizeof(float), inputADeviceAddr, n * n * sizeof(float),
191- n * n * sizeof(float),184+ ACL_MEMCPY_DEVICE_TO_HOST);
192- inputADeviceAddr,
193- n * n * sizeof(float),
194- ACL_MEMCPY_DEVICE_TO_HOST);
195 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return ret);185 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return ret);
196 186 
197 std::cout << "------- output A -------" << std::endl;187 std::cout << "------- output A -------" << std::endl;
@@ -214,4 +204,4 @@ int main(int argc, char **argv)
214 aclrtResetDevice(deviceId);204 aclrtResetDevice(deviceId);
215 aclFinalize();205 aclFinalize();
216 return 0;206 return 0;
217-}207+}
Mexample/A2/BLAS/strmm/example_strmm.cpp+18-28
@@ -21,7 +21,7 @@ using namespace AsdSip;
21 AsdSip::AspbStatus err_ = (err); \21 AsdSip::AspbStatus err_ = (err); \
22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
23 std::cout << "Execute failed." << std::endl; \23 std::cout << "Execute failed." << std::endl; \
24- exit(-1); \24+ return -1; \
25 } \25 } \
26 } while (0)26 } while (0)
27 27 
@@ -37,7 +37,7 @@ using namespace AsdSip;
37 printf(message, ##__VA_ARGS__); \37 printf(message, ##__VA_ARGS__); \
38 } while (0)38 } while (0)
39 39 
40-int64_t GetShapeSize(const std::vector<int64_t> &shape)40+int64_t GetShapeSize(const std::vector<int64_t>& shape)
41{41{
42 int64_t shapeSize = 1;42 int64_t shapeSize = 1;
43 for (auto i : shape) {43 for (auto i : shape) {
@@ -46,7 +46,7 @@ int64_t GetShapeSize(const std::vector<int64_t> &shape)
46 return shapeSize;46 return shapeSize;
47}47}
48 48 
49-int Init(int32_t deviceId, aclrtStream *stream)49+int Init(int32_t deviceId, aclrtStream* stream)
50{50{
51 // 固定写法，acl初始化51 // 固定写法，acl初始化
52 auto ret = aclInit(nullptr);52 auto ret = aclInit(nullptr);
@@ -59,8 +59,8 @@ int Init(int32_t deviceId, aclrtStream *stream)
59}59}
60 60 
61template <typename T>61template <typename T>
62-int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr,62+int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
63- aclDataType dataType, aclTensor **tensor)63+ aclDataType dataType, aclTensor** tensor)
64{64{
65 auto size = GetShapeSize(shape) * sizeof(T);65 auto size = GetShapeSize(shape) * sizeof(T);
66 // 调用aclrtMalloc申请device侧内存66 // 调用aclrtMalloc申请device侧内存
@@ -77,19 +77,12 @@ int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &
77 }77 }
78 78 
79 // 调用aclCreateTensor接口创建aclTensor79 // 调用aclCreateTensor接口创建aclTensor
80- *tensor = aclCreateTensor(shape.data(),80+ *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
81- shape.size(),81+ shape.data(), shape.size(), *deviceAddr);
82- dataType,
83- strides.data(),
84- 0,
85- aclFormat::ACL_FORMAT_ND,
86- shape.data(),
87- shape.size(),
88- *deviceAddr);
89 return 0;82 return 0;
90}83}
91 84 
92-int main(int argc, char **argv)85+int main(int argc, char** argv)
93{86{
94 int deviceId = 0;87 int deviceId = 0;
95 88 
@@ -150,12 +143,12 @@ int main(int argc, char **argv)
150 std::vector<int64_t> bShape = {tensorBSize};143 std::vector<int64_t> bShape = {tensorBSize};
151 std::vector<int64_t> cShape = {tensorCSize};144 std::vector<int64_t> cShape = {tensorCSize};
152 145 
153- aclTensor *inputA = nullptr;146+ aclTensor* inputA = nullptr;
154- aclTensor *inputB = nullptr;147+ aclTensor* inputB = nullptr;
155- aclTensor *outputC = nullptr;148+ aclTensor* outputC = nullptr;
156- void *inputADeviceAddr = nullptr;149+ void* inputADeviceAddr = nullptr;
157- void *inputBDeviceAddr = nullptr;150+ void* inputBDeviceAddr = nullptr;
158- void *outputCDeviceAddr = nullptr;151+ void* outputCDeviceAddr = nullptr;
159 152 
160 ret = CreateAclTensor(tensorInAData, aShape, &inputADeviceAddr, aclDataType::ACL_FLOAT, &inputA);153 ret = CreateAclTensor(tensorInAData, aShape, &inputADeviceAddr, aclDataType::ACL_FLOAT, &inputA);
161 CHECK_RET(ret == ::ACL_SUCCESS, return ret);154 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
@@ -168,7 +161,7 @@ int main(int argc, char **argv)
168 asdBlasCreate(handle);161 asdBlasCreate(handle);
169 162 
170 size_t lwork = 0;163 size_t lwork = 0;
171- void *buffer = nullptr;164+ void* buffer = nullptr;
172 asdBlasMakeStrmmPlan(handle);165 asdBlasMakeStrmmPlan(handle);
173 asdBlasGetWorkspaceSize(handle, lwork);166 asdBlasGetWorkspaceSize(handle, lwork);
174 std::cout << "lwork = " << lwork << std::endl;167 std::cout << "lwork = " << lwork << std::endl;
@@ -185,11 +178,8 @@ int main(int argc, char **argv)
185 asdBlasSynchronize(handle);178 asdBlasSynchronize(handle);
186 asdBlasDestroy(handle);179 asdBlasDestroy(handle);
187 180 
188- ret = aclrtMemcpy(tensorCData.data(),181+ ret = aclrtMemcpy(tensorCData.data(), tensorCSize * sizeof(float), outputCDeviceAddr, tensorCSize * sizeof(float),
189- tensorCSize * sizeof(float),182+ ACL_MEMCPY_DEVICE_TO_HOST);
190- outputCDeviceAddr,
191- tensorCSize * sizeof(float),
192- ACL_MEMCPY_DEVICE_TO_HOST);
193 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return ret);183 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return ret);
194 184 
195 std::cout << "------- output C -------" << std::endl;185 std::cout << "------- output C -------" << std::endl;
@@ -212,4 +202,4 @@ int main(int argc, char **argv)
212 aclrtResetDevice(deviceId);202 aclrtResetDevice(deviceId);
213 aclFinalize();203 aclFinalize();
214 return 0;204 return 0;
215-}205+}
Mexample/A2/BLAS/strmv/example_strmv.cpp+1-1
@@ -21,7 +21,7 @@ using namespace AsdSip;
21 AsdSip::AspbStatus err_ = (err); \21 AsdSip::AspbStatus err_ = (err); \
22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
23 std::cout << "Execute failed." << std::endl; \23 std::cout << "Execute failed." << std::endl; \
24- exit(-1); \24+ return -1; \
25 } \25 } \
26 } while (0)26 } while (0)
27 27 
Mexample/A2/BLAS/swap/example_cswap.cpp+1-1
@@ -22,7 +22,7 @@ using namespace AsdSip;
22 AsdSip::AspbStatus err_ = (err); \22 AsdSip::AspbStatus err_ = (err); \
23 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \23 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
24 std::cout << "Execute failed." << std::endl; \24 std::cout << "Execute failed." << std::endl; \
25- exit(-1); \25+ return -1; \
26 } else { \26 } else { \
27 std::cout << "Execute successfully." << std::endl; \27 std::cout << "Execute successfully." << std::endl; \
28 } \28 } \
Mexample/A2/BLAS/swap/example_sswap.cpp+1-1
@@ -21,7 +21,7 @@ using namespace AsdSip;
21 AsdSip::AspbStatus err_ = (err); \21 AsdSip::AspbStatus err_ = (err); \
22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \22 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
23 std::cout << "Execute failed." << std::endl; \23 std::cout << "Execute failed." << std::endl; \
24- exit(-1); \24+ return -1; \
25 } \25 } \
26 } while (0)26 } while (0)
27 27 
Mexample/A2/Domain/rsInterpolationBySinc/example_rsInterpolationBySinc.cpp+1-1
@@ -23,7 +23,7 @@ using namespace AsdSip;
23 AsdSip::AspbStatus err_ = (err); \23 AsdSip::AspbStatus err_ = (err); \
24 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \24 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
25 std::cout << "Execute failed." << std::endl; \25 std::cout << "Execute failed." << std::endl; \
26- exit(-1); \26+ return -1; \
27 } \27 } \
28 } while (0)28 } while (0)
29 29 
Mexample/A2/FFT/1d/example_acl_fft_c2c_1d.cpp+16-26
@@ -32,11 +32,11 @@ using namespace AsdSip;
32 AsdSip::AspbStatus err_ = (err); \32 AsdSip::AspbStatus err_ = (err); \
33 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \33 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
34 std::cout << "Execute failed." << std::endl; \34 std::cout << "Execute failed." << std::endl; \
35- exit(-1); \35+ return -1; \
36 } \36 } \
37 } while (0)37 } while (0)
38 38 
39-int64_t GetShapeSize(const std::vector<int64_t> &shape)39+int64_t GetShapeSize(const std::vector<int64_t>& shape)
40{40{
41 int64_t shapeSize = 1;41 int64_t shapeSize = 1;
42 for (auto i : shape) {42 for (auto i : shape) {
@@ -45,7 +45,7 @@ int64_t GetShapeSize(const std::vector<int64_t> &shape)
45 return shapeSize;45 return shapeSize;
46}46}
47 47 
48-int Init(int32_t deviceId, aclrtStream *stream)48+int Init(int32_t deviceId, aclrtStream* stream)
49{49{
50 // 固定写法，AscendCL初始化50 // 固定写法，AscendCL初始化
51 auto ret = aclInit(nullptr);51 auto ret = aclInit(nullptr);
@@ -58,8 +58,8 @@ int Init(int32_t deviceId, aclrtStream *stream)
58}58}
59 59 
60template <typename T>60template <typename T>
61-int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr,61+int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
62- aclDataType dataType, aclTensor **tensor)62+ aclDataType dataType, aclTensor** tensor)
63{63{
64 auto size = GetShapeSize(shape) * sizeof(T);64 auto size = GetShapeSize(shape) * sizeof(T);
65 // 调用aclrtMalloc申请device侧内存65 // 调用aclrtMalloc申请device侧内存
@@ -76,15 +76,8 @@ int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &
76 }76 }
77 77 
78 // 调用aclCreateTensor接口创建aclTensor78 // 调用aclCreateTensor接口创建aclTensor
79- *tensor = aclCreateTensor(shape.data(),79+ *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
80- shape.size(),80+ shape.data(), shape.size(), *deviceAddr);
81- dataType,
82- strides.data(),
83- 0,
84- aclFormat::ACL_FORMAT_ND,
85- shape.data(),
86- shape.size(),
87- *deviceAddr);
88 return 0;81 return 0;
89}82}
90 83 
@@ -96,7 +89,7 @@ int main()
96 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("Init acl failed. ERROR: %d\n", ret); return ret);89 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("Init acl failed. ERROR: %d\n", ret); return ret);
97 90 
98 // 创造tensor的Host侧数据91 // 创造tensor的Host侧数据
99- int batch = 32, Nfft = 128; // c2c dft92+ int batch = 32, Nfft = 128; // c2c dft
100 // int batch = 32, Nfft = 8192; // c2c fftb93 // int batch = 32, Nfft = 8192; // c2c fftb
101 // int batch = 32, Nfft = 15000; // c2c mixed94 // int batch = 32, Nfft = 15000; // c2c mixed
102 // int batch = 32, Nfft = 32768; // c2c fftn95 // int batch = 32, Nfft = 32768; // c2c fftn
@@ -109,10 +102,10 @@ int main()
109 inputHostData[i] = std::complex<float>(i, i + 1);102 inputHostData[i] = std::complex<float>(i, i + 1);
110 }103 }
111 std::vector<std::complex<float>> outHostData(tensorInSize, std::complex<float>(0, 0));104 std::vector<std::complex<float>> outHostData(tensorInSize, std::complex<float>(0, 0));
112- void *inputDeviceAddr = nullptr;105+ void* inputDeviceAddr = nullptr;
113- void *outDeviceAddr = nullptr;106+ void* outDeviceAddr = nullptr;
114- aclTensor *input = nullptr;107+ aclTensor* input = nullptr;
115- aclTensor *out = nullptr;108+ aclTensor* out = nullptr;
116 ret = CreateAclTensor(inputHostData, selfShape, &inputDeviceAddr, aclDataType::ACL_COMPLEX64, &input);109 ret = CreateAclTensor(inputHostData, selfShape, &inputDeviceAddr, aclDataType::ACL_COMPLEX64, &input);
117 CHECK_RET(ret == ::ACL_SUCCESS, return ret);110 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
118 ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_COMPLEX64, &out);111 ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_COMPLEX64, &out);
@@ -125,12 +118,12 @@ int main()
125 118 
126 size_t work_size;119 size_t work_size;
127 asdFftGetWorkspaceSize(handle, work_size);120 asdFftGetWorkspaceSize(handle, work_size);
128- void *workspaceAddr = nullptr;121+ void* workspaceAddr = nullptr;
129 if (work_size > 0) {122 if (work_size > 0) {
130 ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST);123 ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST);
131 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);124 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);
132 }125 }
133- asdFftSetWorkspace(handle, (uint8_t *)workspaceAddr);126+ asdFftSetWorkspace(handle, (uint8_t*)workspaceAddr);
134 127 
135 asdFftSetStream(handle, stream);128 asdFftSetStream(handle, stream);
136 ASD_STATUS_CHECK(asdFftExecC2C(handle, input, out));129 ASD_STATUS_CHECK(asdFftExecC2C(handle, input, out));
@@ -142,11 +135,8 @@ int main()
142 135 
143 auto size = GetShapeSize(outShape);136 auto size = GetShapeSize(outShape);
144 std::vector<std::complex<float>> outData(size, 0);137 std::vector<std::complex<float>> outData(size, 0);
145- ret = aclrtMemcpy(outData.data(),138+ ret = aclrtMemcpy(outData.data(), outData.size() * sizeof(outData[0]), outDeviceAddr, size * sizeof(outData[0]),
146- outData.size() * sizeof(outData[0]),139+ ACL_MEMCPY_DEVICE_TO_HOST);
147- outDeviceAddr,
148- size * sizeof(outData[0]),
149- ACL_MEMCPY_DEVICE_TO_HOST);
150 140 
151 // 打印输出tensor值中前16个141 // 打印输出tensor值中前16个
152 for (int64_t i = 0; i < 16; i++) {142 for (int64_t i = 0; i < 16; i++) {
Mexample/A2/FFT/1d/example_acl_fft_c2r_1d.cpp+15-25
@@ -32,11 +32,11 @@ using namespace AsdSip;
32 AsdSip::AspbStatus err_ = (err); \32 AsdSip::AspbStatus err_ = (err); \
33 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \33 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
34 std::cout << "Execute failed." << std::endl; \34 std::cout << "Execute failed." << std::endl; \
35- exit(-1); \35+ return -1; \
36 } \36 } \
37 } while (0)37 } while (0)
38 38 
39-int64_t GetShapeSize(const std::vector<int64_t> &shape)39+int64_t GetShapeSize(const std::vector<int64_t>& shape)
40{40{
41 int64_t shapeSize = 1;41 int64_t shapeSize = 1;
42 for (auto i : shape) {42 for (auto i : shape) {
@@ -45,7 +45,7 @@ int64_t GetShapeSize(const std::vector<int64_t> &shape)
45 return shapeSize;45 return shapeSize;
46}46}
47 47 
48-int Init(int32_t deviceId, aclrtStream *stream)48+int Init(int32_t deviceId, aclrtStream* stream)
49{49{
50 // 固定写法，AscendCL初始化50 // 固定写法，AscendCL初始化
51 auto ret = aclInit(nullptr);51 auto ret = aclInit(nullptr);
@@ -58,8 +58,8 @@ int Init(int32_t deviceId, aclrtStream *stream)
58}58}
59 59 
60template <typename T>60template <typename T>
61-int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr,61+int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
62- aclDataType dataType, aclTensor **tensor)62+ aclDataType dataType, aclTensor** tensor)
63{63{
64 auto size = GetShapeSize(shape) * sizeof(T);64 auto size = GetShapeSize(shape) * sizeof(T);
65 // 调用aclrtMalloc申请device侧内存65 // 调用aclrtMalloc申请device侧内存
@@ -76,15 +76,8 @@ int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &
76 }76 }
77 77 
78 // 调用aclCreateTensor接口创建aclTensor78 // 调用aclCreateTensor接口创建aclTensor
79- *tensor = aclCreateTensor(shape.data(),79+ *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
80- shape.size(),80+ shape.data(), shape.size(), *deviceAddr);
81- dataType,
82- strides.data(),
83- 0,
84- aclFormat::ACL_FORMAT_ND,
85- shape.data(),
86- shape.size(),
87- *deviceAddr);
88 return 0;81 return 0;
89}82}
90 83 
@@ -115,10 +108,10 @@ int main()
115 inputHostData[i] = std::complex<float>(i, i + 1);108 inputHostData[i] = std::complex<float>(i, i + 1);
116 }109 }
117 std::vector<float> outHostData(tensorOutSize, 0);110 std::vector<float> outHostData(tensorOutSize, 0);
118- void *inputDeviceAddr = nullptr;111+ void* inputDeviceAddr = nullptr;
119- void *outDeviceAddr = nullptr;112+ void* outDeviceAddr = nullptr;
120- aclTensor *input = nullptr;113+ aclTensor* input = nullptr;
121- aclTensor *out = nullptr;114+ aclTensor* out = nullptr;
122 ret = CreateAclTensor(inputHostData, selfShape, &inputDeviceAddr, aclDataType::ACL_COMPLEX64, &input);115 ret = CreateAclTensor(inputHostData, selfShape, &inputDeviceAddr, aclDataType::ACL_COMPLEX64, &input);
123 CHECK_RET(ret == ::ACL_SUCCESS, return ret);116 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
124 ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_FLOAT, &out);117 ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_FLOAT, &out);
@@ -131,12 +124,12 @@ int main()
131 124 
132 size_t work_size;125 size_t work_size;
133 asdFftGetWorkspaceSize(handle, work_size);126 asdFftGetWorkspaceSize(handle, work_size);
134- void *workspaceAddr = nullptr;127+ void* workspaceAddr = nullptr;
135 if (work_size > 0) {128 if (work_size > 0) {
136 ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST);129 ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST);
137 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);130 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);
138 }131 }
139- asdFftSetWorkspace(handle, (uint8_t *)workspaceAddr);132+ asdFftSetWorkspace(handle, (uint8_t*)workspaceAddr);
140 133 
141 asdFftSetStream(handle, stream);134 asdFftSetStream(handle, stream);
142 ASD_STATUS_CHECK(asdFftExecC2R(handle, input, out));135 ASD_STATUS_CHECK(asdFftExecC2R(handle, input, out));
@@ -148,11 +141,8 @@ int main()
148 141 
149 auto size = GetShapeSize(outShape);142 auto size = GetShapeSize(outShape);
150 std::vector<float> outData(size, 0);143 std::vector<float> outData(size, 0);
151- ret = aclrtMemcpy(outData.data(),144+ ret = aclrtMemcpy(outData.data(), outData.size() * sizeof(outData[0]), outDeviceAddr, size * sizeof(outData[0]),
152- outData.size() * sizeof(outData[0]),145+ ACL_MEMCPY_DEVICE_TO_HOST);
153- outDeviceAddr,
154- size * sizeof(outData[0]),
155- ACL_MEMCPY_DEVICE_TO_HOST);
156 146 
157 // 打印输出tensor值中前16个147 // 打印输出tensor值中前16个
158 for (int64_t i = 0; i < 16; i++) {148 for (int64_t i = 0; i < 16; i++) {
Mexample/A2/FFT/1d/example_acl_fft_r2c_1d.cpp+15-25
@@ -32,11 +32,11 @@ using namespace AsdSip;
32 AsdSip::AspbStatus err_ = (err); \32 AsdSip::AspbStatus err_ = (err); \
33 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \33 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
34 std::cout << "Execute failed." << std::endl; \34 std::cout << "Execute failed." << std::endl; \
35- exit(-1); \35+ return -1; \
36 } \36 } \
37 } while (0)37 } while (0)
38 38 
39-int64_t GetShapeSize(const std::vector<int64_t> &shape)39+int64_t GetShapeSize(const std::vector<int64_t>& shape)
40{40{
41 int64_t shapeSize = 1;41 int64_t shapeSize = 1;
42 for (auto i : shape) {42 for (auto i : shape) {
@@ -45,7 +45,7 @@ int64_t GetShapeSize(const std::vector<int64_t> &shape)
45 return shapeSize;45 return shapeSize;
46}46}
47 47 
48-int Init(int32_t deviceId, aclrtStream *stream)48+int Init(int32_t deviceId, aclrtStream* stream)
49{49{
50 // 固定写法，AscendCL初始化50 // 固定写法，AscendCL初始化
51 auto ret = aclInit(nullptr);51 auto ret = aclInit(nullptr);
@@ -58,8 +58,8 @@ int Init(int32_t deviceId, aclrtStream *stream)
58}58}
59 59 
60template <typename T>60template <typename T>
61-int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr,61+int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
62- aclDataType dataType, aclTensor **tensor)62+ aclDataType dataType, aclTensor** tensor)
63{63{
64 auto size = GetShapeSize(shape) * sizeof(T);64 auto size = GetShapeSize(shape) * sizeof(T);
65 // 调用aclrtMalloc申请device侧内存65 // 调用aclrtMalloc申请device侧内存
@@ -76,15 +76,8 @@ int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &
76 }76 }
77 77 
78 // 调用aclCreateTensor接口创建aclTensor78 // 调用aclCreateTensor接口创建aclTensor
79- *tensor = aclCreateTensor(shape.data(),79+ *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
80- shape.size(),80+ shape.data(), shape.size(), *deviceAddr);
81- dataType,
82- strides.data(),
83- 0,
84- aclFormat::ACL_FORMAT_ND,
85- shape.data(),
86- shape.size(),
87- *deviceAddr);
88 return 0;81 return 0;
89}82}
90 83 
@@ -109,10 +102,10 @@ int main()
109 inputHostData[i] = i;102 inputHostData[i] = i;
110 }103 }
111 std::vector<std::complex<float>> outHostData(tensorOutSize, std::complex<float>(0, 0));104 std::vector<std::complex<float>> outHostData(tensorOutSize, std::complex<float>(0, 0));
112- void *inputDeviceAddr = nullptr;105+ void* inputDeviceAddr = nullptr;
113- void *outDeviceAddr = nullptr;106+ void* outDeviceAddr = nullptr;
114- aclTensor *input = nullptr;107+ aclTensor* input = nullptr;
115- aclTensor *out = nullptr;108+ aclTensor* out = nullptr;
116 ret = CreateAclTensor(inputHostData, selfShape, &inputDeviceAddr, aclDataType::ACL_FLOAT, &input);109 ret = CreateAclTensor(inputHostData, selfShape, &inputDeviceAddr, aclDataType::ACL_FLOAT, &input);
117 CHECK_RET(ret == ::ACL_SUCCESS, return ret);110 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
118 ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_COMPLEX64, &out);111 ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_COMPLEX64, &out);
@@ -125,12 +118,12 @@ int main()
125 118 
126 size_t work_size;119 size_t work_size;
127 asdFftGetWorkspaceSize(handle, work_size);120 asdFftGetWorkspaceSize(handle, work_size);
128- void *workspaceAddr = nullptr;121+ void* workspaceAddr = nullptr;
129 if (work_size > 0) {122 if (work_size > 0) {
130 ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST);123 ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST);
131 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);124 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);
132 }125 }
133- asdFftSetWorkspace(handle, (uint8_t *)workspaceAddr);126+ asdFftSetWorkspace(handle, (uint8_t*)workspaceAddr);
134 127 
135 asdFftSetStream(handle, stream);128 asdFftSetStream(handle, stream);
136 ASD_STATUS_CHECK(asdFftExecR2C(handle, input, out));129 ASD_STATUS_CHECK(asdFftExecR2C(handle, input, out));
@@ -142,11 +135,8 @@ int main()
142 135 
143 auto size = GetShapeSize(outShape);136 auto size = GetShapeSize(outShape);
144 std::vector<std::complex<float>> outData(size, 0);137 std::vector<std::complex<float>> outData(size, 0);
145- ret = aclrtMemcpy(outData.data(),138+ ret = aclrtMemcpy(outData.data(), outData.size() * sizeof(outData[0]), outDeviceAddr, size * sizeof(outData[0]),
146- outData.size() * sizeof(outData[0]),139+ ACL_MEMCPY_DEVICE_TO_HOST);
147- outDeviceAddr,
148- size * sizeof(outData[0]),
149- ACL_MEMCPY_DEVICE_TO_HOST);
150 140 
151 // 打印输出tensor值中前16个141 // 打印输出tensor值中前16个
152 for (int64_t i = 0; i < 16; i++) {142 for (int64_t i = 0; i < 16; i++) {
Mexample/A2/FFT/2d/example_acl_fft_c2c_2d.cpp+15-25
@@ -32,11 +32,11 @@ using namespace AsdSip;
32 AsdSip::AspbStatus err_ = (err); \32 AsdSip::AspbStatus err_ = (err); \
33 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \33 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
34 std::cout << "Execute failed." << std::endl; \34 std::cout << "Execute failed." << std::endl; \
35- exit(-1); \35+ return -1; \
36 } \36 } \
37 } while (0)37 } while (0)
38 38 
39-int64_t GetShapeSize(const std::vector<int64_t> &shape)39+int64_t GetShapeSize(const std::vector<int64_t>& shape)
40{40{
41 int64_t shapeSize = 1;41 int64_t shapeSize = 1;
42 for (auto i : shape) {42 for (auto i : shape) {
@@ -45,7 +45,7 @@ int64_t GetShapeSize(const std::vector<int64_t> &shape)
45 return shapeSize;45 return shapeSize;
46}46}
47 47 
48-int Init(int32_t deviceId, aclrtStream *stream)48+int Init(int32_t deviceId, aclrtStream* stream)
49{49{
50 // 固定写法，AscendCL初始化50 // 固定写法，AscendCL初始化
51 auto ret = aclInit(nullptr);51 auto ret = aclInit(nullptr);
@@ -58,8 +58,8 @@ int Init(int32_t deviceId, aclrtStream *stream)
58}58}
59 59 
60template <typename T>60template <typename T>
61-int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr,61+int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
62- aclDataType dataType, aclTensor **tensor)62+ aclDataType dataType, aclTensor** tensor)
63{63{
64 auto size = GetShapeSize(shape) * sizeof(T);64 auto size = GetShapeSize(shape) * sizeof(T);
65 // 调用aclrtMalloc申请device侧内存65 // 调用aclrtMalloc申请device侧内存
@@ -76,15 +76,8 @@ int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &
76 }76 }
77 77 
78 // 调用aclCreateTensor接口创建aclTensor78 // 调用aclCreateTensor接口创建aclTensor
79- *tensor = aclCreateTensor(shape.data(),79+ *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
80- shape.size(),80+ shape.data(), shape.size(), *deviceAddr);
81- dataType,
82- strides.data(),
83- 0,
84- aclFormat::ACL_FORMAT_ND,
85- shape.data(),
86- shape.size(),
87- *deviceAddr);
88 return 0;81 return 0;
89}82}
90 83 
@@ -109,10 +102,10 @@ int main()
109 inputHostData[i] = std::complex<float>(i, i + 1);102 inputHostData[i] = std::complex<float>(i, i + 1);
110 }103 }
111 std::vector<std::complex<float>> outHostData(tensorOutSize, std::complex<float>(0, 0));104 std::vector<std::complex<float>> outHostData(tensorOutSize, std::complex<float>(0, 0));
112- void *inputDeviceAddr = nullptr;105+ void* inputDeviceAddr = nullptr;
113- void *outDeviceAddr = nullptr;106+ void* outDeviceAddr = nullptr;
114- aclTensor *input = nullptr;107+ aclTensor* input = nullptr;
115- aclTensor *out = nullptr;108+ aclTensor* out = nullptr;
116 ret = CreateAclTensor(inputHostData, selfShape, &inputDeviceAddr, aclDataType::ACL_COMPLEX64, &input);109 ret = CreateAclTensor(inputHostData, selfShape, &inputDeviceAddr, aclDataType::ACL_COMPLEX64, &input);
117 CHECK_RET(ret == ::ACL_SUCCESS, return ret);110 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
118 ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_COMPLEX64, &out);111 ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_COMPLEX64, &out);
@@ -123,12 +116,12 @@ int main()
123 asdFftMakePlan2D(handle, Nfft1, Nfft2, asdFftType::ASCEND_FFT_C2C, asdFftDirection::ASCEND_FFT_FORWARD, batch);116 asdFftMakePlan2D(handle, Nfft1, Nfft2, asdFftType::ASCEND_FFT_C2C, asdFftDirection::ASCEND_FFT_FORWARD, batch);
124 size_t work_size;117 size_t work_size;
125 asdFftGetWorkspaceSize(handle, work_size);118 asdFftGetWorkspaceSize(handle, work_size);
126- void *workspaceAddr = nullptr;119+ void* workspaceAddr = nullptr;
127 if (work_size > 0) {120 if (work_size > 0) {
128 ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST);121 ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST);
129 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);122 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);
130 }123 }
131- asdFftSetWorkspace(handle, (uint8_t *)workspaceAddr);124+ asdFftSetWorkspace(handle, (uint8_t*)workspaceAddr);
132 asdFftSetStream(handle, stream);125 asdFftSetStream(handle, stream);
133 ASD_STATUS_CHECK(asdFftExecC2C(handle, input, out));126 ASD_STATUS_CHECK(asdFftExecC2C(handle, input, out));
134 ret = aclrtSynchronizeStream(stream);127 ret = aclrtSynchronizeStream(stream);
@@ -137,11 +130,8 @@ int main()
137 130 
138 auto size = GetShapeSize(outShape);131 auto size = GetShapeSize(outShape);
139 std::vector<std::complex<float>> outData(size, 0);132 std::vector<std::complex<float>> outData(size, 0);
140- ret = aclrtMemcpy(outData.data(),133+ ret = aclrtMemcpy(outData.data(), outData.size() * sizeof(outData[0]), outDeviceAddr, size * sizeof(outData[0]),
141- outData.size() * sizeof(outData[0]),134+ ACL_MEMCPY_DEVICE_TO_HOST);
142- outDeviceAddr,
143- size * sizeof(outData[0]),
144- ACL_MEMCPY_DEVICE_TO_HOST);
145 135 
146 // 打印输出tensor值中前16个136 // 打印输出tensor值中前16个
147 for (int64_t i = 0; i < 16; i++) {137 for (int64_t i = 0; i < 16; i++) {
Mexample/A2/FFT/2d/example_acl_fft_c2r_2d.cpp+15-25
@@ -32,11 +32,11 @@ using namespace AsdSip;
32 AsdSip::AspbStatus err_ = (err); \32 AsdSip::AspbStatus err_ = (err); \
33 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \33 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
34 std::cout << "Execute failed." << std::endl; \34 std::cout << "Execute failed." << std::endl; \
35- exit(-1); \35+ return -1; \
36 } \36 } \
37 } while (0)37 } while (0)
38 38 
39-int64_t GetShapeSize(const std::vector<int64_t> &shape)39+int64_t GetShapeSize(const std::vector<int64_t>& shape)
40{40{
41 int64_t shapeSize = 1;41 int64_t shapeSize = 1;
42 for (auto i : shape) {42 for (auto i : shape) {
@@ -45,7 +45,7 @@ int64_t GetShapeSize(const std::vector<int64_t> &shape)
45 return shapeSize;45 return shapeSize;
46}46}
47 47 
48-int Init(int32_t deviceId, aclrtStream *stream)48+int Init(int32_t deviceId, aclrtStream* stream)
49{49{
50 // 固定写法，AscendCL初始化50 // 固定写法，AscendCL初始化
51 auto ret = aclInit(nullptr);51 auto ret = aclInit(nullptr);
@@ -58,8 +58,8 @@ int Init(int32_t deviceId, aclrtStream *stream)
58}58}
59 59 
60template <typename T>60template <typename T>
61-int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr,61+int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
62- aclDataType dataType, aclTensor **tensor)62+ aclDataType dataType, aclTensor** tensor)
63{63{
64 auto size = GetShapeSize(shape) * sizeof(T);64 auto size = GetShapeSize(shape) * sizeof(T);
65 // 调用aclrtMalloc申请device侧内存65 // 调用aclrtMalloc申请device侧内存
@@ -76,15 +76,8 @@ int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &
76 }76 }
77 77 
78 // 调用aclCreateTensor接口创建aclTensor78 // 调用aclCreateTensor接口创建aclTensor
79- *tensor = aclCreateTensor(shape.data(),79+ *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
80- shape.size(),80+ shape.data(), shape.size(), *deviceAddr);
81- dataType,
82- strides.data(),
83- 0,
84- aclFormat::ACL_FORMAT_ND,
85- shape.data(),
86- shape.size(),
87- *deviceAddr);
88 return 0;81 return 0;
89}82}
90 83 
@@ -113,10 +106,10 @@ int main()
113 inputHostData[i] = std::complex<float>(i, i + 1);106 inputHostData[i] = std::complex<float>(i, i + 1);
114 }107 }
115 std::vector<float> outHostData(tensorOutSize, 0);108 std::vector<float> outHostData(tensorOutSize, 0);
116- void *inputDeviceAddr = nullptr;109+ void* inputDeviceAddr = nullptr;
117- void *outDeviceAddr = nullptr;110+ void* outDeviceAddr = nullptr;
118- aclTensor *input = nullptr;111+ aclTensor* input = nullptr;
119- aclTensor *out = nullptr;112+ aclTensor* out = nullptr;
120 ret = CreateAclTensor(inputHostData, selfShape, &inputDeviceAddr, aclDataType::ACL_COMPLEX64, &input);113 ret = CreateAclTensor(inputHostData, selfShape, &inputDeviceAddr, aclDataType::ACL_COMPLEX64, &input);
121 CHECK_RET(ret == ::ACL_SUCCESS, return ret);114 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
122 ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_FLOAT, &out);115 ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_FLOAT, &out);
@@ -127,12 +120,12 @@ int main()
127 asdFftMakePlan2D(handle, Nfft1, Nfft2, asdFftType::ASCEND_FFT_C2R, asdFftDirection::ASCEND_FFT_FORWARD, batch);120 asdFftMakePlan2D(handle, Nfft1, Nfft2, asdFftType::ASCEND_FFT_C2R, asdFftDirection::ASCEND_FFT_FORWARD, batch);
128 size_t work_size;121 size_t work_size;
129 asdFftGetWorkspaceSize(handle, work_size);122 asdFftGetWorkspaceSize(handle, work_size);
130- void *workspaceAddr = nullptr;123+ void* workspaceAddr = nullptr;
131 if (work_size > 0) {124 if (work_size > 0) {
132 ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST);125 ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST);
133 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);126 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);
134 }127 }
135- asdFftSetWorkspace(handle, (uint8_t *)workspaceAddr);128+ asdFftSetWorkspace(handle, (uint8_t*)workspaceAddr);
136 asdFftSetStream(handle, stream);129 asdFftSetStream(handle, stream);
137 ASD_STATUS_CHECK(asdFftExecC2R(handle, input, out));130 ASD_STATUS_CHECK(asdFftExecC2R(handle, input, out));
138 ret = aclrtSynchronizeStream(stream);131 ret = aclrtSynchronizeStream(stream);
@@ -141,11 +134,8 @@ int main()
141 134 
142 auto size = GetShapeSize(outShape);135 auto size = GetShapeSize(outShape);
143 std::vector<float> outData(size, 0);136 std::vector<float> outData(size, 0);
144- ret = aclrtMemcpy(outData.data(),137+ ret = aclrtMemcpy(outData.data(), outData.size() * sizeof(outData[0]), outDeviceAddr, size * sizeof(outData[0]),
145- outData.size() * sizeof(outData[0]),138+ ACL_MEMCPY_DEVICE_TO_HOST);
146- outDeviceAddr,
147- size * sizeof(outData[0]),
148- ACL_MEMCPY_DEVICE_TO_HOST);
149 139 
150 // 打印输出tensor值中前16个140 // 打印输出tensor值中前16个
151 for (int64_t i = 0; i < 16; i++) {141 for (int64_t i = 0; i < 16; i++) {
Mexample/A2/FFT/2d/example_acl_fft_r2c_2d.cpp+15-25
@@ -32,11 +32,11 @@ using namespace AsdSip;
32 AsdSip::AspbStatus err_ = (err); \32 AsdSip::AspbStatus err_ = (err); \
33 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \33 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
34 std::cout << "Execute failed." << std::endl; \34 std::cout << "Execute failed." << std::endl; \
35- exit(-1); \35+ return -1; \
36 } \36 } \
37 } while (0)37 } while (0)
38 38 
39-int64_t GetShapeSize(const std::vector<int64_t> &shape)39+int64_t GetShapeSize(const std::vector<int64_t>& shape)
40{40{
41 int64_t shapeSize = 1;41 int64_t shapeSize = 1;
42 for (auto i : shape) {42 for (auto i : shape) {
@@ -45,7 +45,7 @@ int64_t GetShapeSize(const std::vector<int64_t> &shape)
45 return shapeSize;45 return shapeSize;
46}46}
47 47 
48-int Init(int32_t deviceId, aclrtStream *stream)48+int Init(int32_t deviceId, aclrtStream* stream)
49{49{
50 // 固定写法，AscendCL初始化50 // 固定写法，AscendCL初始化
51 auto ret = aclInit(nullptr);51 auto ret = aclInit(nullptr);
@@ -58,8 +58,8 @@ int Init(int32_t deviceId, aclrtStream *stream)
58}58}
59 59 
60template <typename T>60template <typename T>
61-int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr,61+int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
62- aclDataType dataType, aclTensor **tensor)62+ aclDataType dataType, aclTensor** tensor)
63{63{
64 auto size = GetShapeSize(shape) * sizeof(T);64 auto size = GetShapeSize(shape) * sizeof(T);
65 // 调用aclrtMalloc申请device侧内存65 // 调用aclrtMalloc申请device侧内存
@@ -76,15 +76,8 @@ int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &
76 }76 }
77 77 
78 // 调用aclCreateTensor接口创建aclTensor78 // 调用aclCreateTensor接口创建aclTensor
79- *tensor = aclCreateTensor(shape.data(),79+ *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
80- shape.size(),80+ shape.data(), shape.size(), *deviceAddr);
81- dataType,
82- strides.data(),
83- 0,
84- aclFormat::ACL_FORMAT_ND,
85- shape.data(),
86- shape.size(),
87- *deviceAddr);
88 return 0;81 return 0;
89}82}
90 83 
@@ -108,10 +101,10 @@ int main()
108 inputHostData[i] = i;101 inputHostData[i] = i;
109 }102 }
110 std::vector<std::complex<float>> outHostData(tensorOutSize, std::complex<float>(0, 0));103 std::vector<std::complex<float>> outHostData(tensorOutSize, std::complex<float>(0, 0));
111- void *inputDeviceAddr = nullptr;104+ void* inputDeviceAddr = nullptr;
112- void *outDeviceAddr = nullptr;105+ void* outDeviceAddr = nullptr;
113- aclTensor *input = nullptr;106+ aclTensor* input = nullptr;
114- aclTensor *out = nullptr;107+ aclTensor* out = nullptr;
115 ret = CreateAclTensor(inputHostData, selfShape, &inputDeviceAddr, aclDataType::ACL_FLOAT, &input);108 ret = CreateAclTensor(inputHostData, selfShape, &inputDeviceAddr, aclDataType::ACL_FLOAT, &input);
116 CHECK_RET(ret == ::ACL_SUCCESS, return ret);109 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
117 ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_COMPLEX64, &out);110 ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_COMPLEX64, &out);
@@ -122,12 +115,12 @@ int main()
122 asdFftMakePlan2D(handle, Nfft1, Nfft2, asdFftType::ASCEND_FFT_R2C, asdFftDirection::ASCEND_FFT_FORWARD, batch);115 asdFftMakePlan2D(handle, Nfft1, Nfft2, asdFftType::ASCEND_FFT_R2C, asdFftDirection::ASCEND_FFT_FORWARD, batch);
123 size_t work_size;116 size_t work_size;
124 asdFftGetWorkspaceSize(handle, work_size);117 asdFftGetWorkspaceSize(handle, work_size);
125- void *workspaceAddr = nullptr;118+ void* workspaceAddr = nullptr;
126 if (work_size > 0) {119 if (work_size > 0) {
127 ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST);120 ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST);
128 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);121 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);
129 }122 }
130- asdFftSetWorkspace(handle, (uint8_t *)workspaceAddr);123+ asdFftSetWorkspace(handle, (uint8_t*)workspaceAddr);
131 asdFftSetStream(handle, stream);124 asdFftSetStream(handle, stream);
132 ASD_STATUS_CHECK(asdFftExecR2C(handle, input, out));125 ASD_STATUS_CHECK(asdFftExecR2C(handle, input, out));
133 ret = aclrtSynchronizeStream(stream);126 ret = aclrtSynchronizeStream(stream);
@@ -136,11 +129,8 @@ int main()
136 129 
137 auto size = GetShapeSize(outShape);130 auto size = GetShapeSize(outShape);
138 std::vector<std::complex<float>> outData(size, 0);131 std::vector<std::complex<float>> outData(size, 0);
139- ret = aclrtMemcpy(outData.data(),132+ ret = aclrtMemcpy(outData.data(), outData.size() * sizeof(outData[0]), outDeviceAddr, size * sizeof(outData[0]),
140- outData.size() * sizeof(outData[0]),133+ ACL_MEMCPY_DEVICE_TO_HOST);
141- outDeviceAddr,
142- size * sizeof(outData[0]),
143- ACL_MEMCPY_DEVICE_TO_HOST);
144 134 
145 // 打印输出tensor值中前16个135 // 打印输出tensor值中前16个
146 for (int64_t i = 0; i < 16; i++) {136 for (int64_t i = 0; i < 16; i++) {
Mexample/A2/FFT/3d/example_acl_fft_c2c_3d.cpp+25-34
@@ -27,16 +27,16 @@ using namespace AsdSip;
27 printf(message, ##__VA_ARGS__); \27 printf(message, ##__VA_ARGS__); \
28 } while (0)28 } while (0)
29 29 
30-#define ASD_STATUS_CHECK(err) \30+#define ASD_STATUS_CHECK(err) \
31- do { \31+ do { \
32- AsdSip::AspbStatus err_ = (err); \32+ AsdSip::AspbStatus err_ = (err); \
33- if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \33+ if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
34 std::cout << "Execute failed." << std::endl; \34 std::cout << "Execute failed." << std::endl; \
35- exit(-1); \35+ return -1; \
36- } \36+ } \
37 } while (0)37 } while (0)
38 38 
39-int64_t GetShapeSize(const std::vector<int64_t> &shape)39+int64_t GetShapeSize(const std::vector<int64_t>& shape)
40{40{
41 int64_t shapeSize = 1;41 int64_t shapeSize = 1;
42 for (auto i : shape) {42 for (auto i : shape) {
@@ -45,7 +45,7 @@ int64_t GetShapeSize(const std::vector<int64_t> &shape)
45 return shapeSize;45 return shapeSize;
46}46}
47 47 
48-int Init(int32_t deviceId, aclrtStream *stream)48+int Init(int32_t deviceId, aclrtStream* stream)
49{49{
50 // 固定写法，AscendCL初始化50 // 固定写法，AscendCL初始化
51 auto ret = aclInit(nullptr);51 auto ret = aclInit(nullptr);
@@ -58,8 +58,8 @@ int Init(int32_t deviceId, aclrtStream *stream)
58}58}
59 59 
60template <typename T>60template <typename T>
61-int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr,61+int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
62- aclDataType dataType, aclTensor **tensor)62+ aclDataType dataType, aclTensor** tensor)
63{63{
64 auto size = GetShapeSize(shape) * sizeof(T);64 auto size = GetShapeSize(shape) * sizeof(T);
65 // 调用aclrtMalloc申请device侧内存65 // 调用aclrtMalloc申请device侧内存
@@ -76,20 +76,13 @@ int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &
76 }76 }
77 77 
78 // 调用aclCreateTensor接口创建aclTensor78 // 调用aclCreateTensor接口创建aclTensor
79- *tensor = aclCreateTensor(shape.data(),79+ *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
80- shape.size(),80+ shape.data(), shape.size(), *deviceAddr);
81- dataType,
82- strides.data(),
83- 0,
84- aclFormat::ACL_FORMAT_ND,
85- shape.data(),
86- shape.size(),
87- *deviceAddr);
88 return 0;81 return 0;
89}82}
90 83 
91- 84+int main()
92-int main() {85+{
93 int32_t deviceId = 0;86 int32_t deviceId = 0;
94 aclrtStream stream;87 aclrtStream stream;
95 auto ret = Init(deviceId, &stream);88 auto ret = Init(deviceId, &stream);
@@ -107,10 +100,10 @@ int main() {
107 }100 }
108 std::vector<std::complex<float>> outHostData(tensorInSize, std::complex<float>(0, 0));101 std::vector<std::complex<float>> outHostData(tensorInSize, std::complex<float>(0, 0));
109 102 
110- void *inputDeviceAddr = nullptr;103+ void* inputDeviceAddr = nullptr;
111- void *outDeviceAddr = nullptr;104+ void* outDeviceAddr = nullptr;
112- aclTensor *input = nullptr;105+ aclTensor* input = nullptr;
113- aclTensor *out = nullptr;106+ aclTensor* out = nullptr;
114 ret = CreateAclTensor(inputHostData, selfShape, &inputDeviceAddr, aclDataType::ACL_COMPLEX64, &input);107 ret = CreateAclTensor(inputHostData, selfShape, &inputDeviceAddr, aclDataType::ACL_COMPLEX64, &input);
115 CHECK_RET(ret == ::ACL_SUCCESS, return ret);108 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
116 ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_COMPLEX64, &out);109 ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_COMPLEX64, &out);
@@ -119,16 +112,17 @@ int main() {
119 asdFftHandle handle;112 asdFftHandle handle;
120 asdFftCreate(handle);113 asdFftCreate(handle);
121 114 
122- asdFftMakePlan3D(handle, Nfft1, Nfft2, Nfft3, asdFftType::ASCEND_FFT_C2C, asdFftDirection::ASCEND_FFT_FORWARD, batch);115+ asdFftMakePlan3D(handle, Nfft1, Nfft2, Nfft3, asdFftType::ASCEND_FFT_C2C, asdFftDirection::ASCEND_FFT_FORWARD,
116+ batch);
123 117 
124 size_t work_size;118 size_t work_size;
125 asdFftGetWorkspaceSize(handle, work_size);119 asdFftGetWorkspaceSize(handle, work_size);
126- void *workspaceAddr = nullptr;120+ void* workspaceAddr = nullptr;
127 if (work_size > 0) {121 if (work_size > 0) {
128 ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST);122 ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST);
129 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);123 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);
130 }124 }
131- asdFftSetWorkspace(handle, (uint8_t *)workspaceAddr);125+ asdFftSetWorkspace(handle, (uint8_t*)workspaceAddr);
132 126 
133 asdFftSetStream(handle, stream);127 asdFftSetStream(handle, stream);
134 ASD_STATUS_CHECK(asdFftExecC2C(handle, input, out));128 ASD_STATUS_CHECK(asdFftExecC2C(handle, input, out));
@@ -140,11 +134,8 @@ int main() {
140 134 
141 auto size = GetShapeSize(outShape);135 auto size = GetShapeSize(outShape);
142 std::vector<std::complex<float>> outData(size, 0);136 std::vector<std::complex<float>> outData(size, 0);
143- ret = aclrtMemcpy(outData.data(),137+ ret = aclrtMemcpy(outData.data(), outData.size() * sizeof(outData[0]), outDeviceAddr, size * sizeof(outData[0]),
144- outData.size() * sizeof(outData[0]),138+ ACL_MEMCPY_DEVICE_TO_HOST);
145- outDeviceAddr,
146- size * sizeof(outData[0]),
147- ACL_MEMCPY_DEVICE_TO_HOST);
148 139 
149 // 打印输出tensor值中前16个140 // 打印输出tensor值中前16个
150 for (int64_t i = 0; i < std::min(static_cast<int64_t>(16), tensorInSize); i++) {141 for (int64_t i = 0; i < std::min(static_cast<int64_t>(16), tensorInSize); i++) {
@@ -165,4 +156,4 @@ int main() {
165 aclrtResetDevice(deviceId);156 aclrtResetDevice(deviceId);
166 aclFinalize();157 aclFinalize();
167 return 0;158 return 0;
168-}159+}
Mexample/A2/FFT/3d/example_acl_fft_c2c_3d_sep.cpp+31-46
@@ -29,16 +29,16 @@ using namespace AsdSip;
29 printf(message, ##__VA_ARGS__); \29 printf(message, ##__VA_ARGS__); \
30 } while (0)30 } while (0)
31 31 
32-#define ASD_STATUS_CHECK(err) \32+#define ASD_STATUS_CHECK(err) \
33- do { \33+ do { \
34- AsdSip::AspbStatus err_ = (err); \34+ AsdSip::AspbStatus err_ = (err); \
35- if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \35+ if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
36 std::cout << "Execute failed." << std::endl; \36 std::cout << "Execute failed." << std::endl; \
37- exit(-1); \37+ return -1; \
38- } \38+ } \
39 } while (0)39 } while (0)
40 40 
41-int64_t GetShapeSize(const std::vector<int64_t> &shape)41+int64_t GetShapeSize(const std::vector<int64_t>& shape)
42{42{
43 int64_t shapeSize = 1;43 int64_t shapeSize = 1;
44 for (auto i : shape) {44 for (auto i : shape) {
@@ -47,7 +47,7 @@ int64_t GetShapeSize(const std::vector<int64_t> &shape)
47 return shapeSize;47 return shapeSize;
48}48}
49 49 
50-int Init(int32_t deviceId, aclrtStream *stream)50+int Init(int32_t deviceId, aclrtStream* stream)
51{51{
52 // 固定写法，AscendCL初始化52 // 固定写法，AscendCL初始化
53 auto ret = aclInit(nullptr);53 auto ret = aclInit(nullptr);
@@ -60,8 +60,8 @@ int Init(int32_t deviceId, aclrtStream *stream)
60}60}
61 61 
62template <typename T>62template <typename T>
63-int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr,63+int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
64- aclDataType dataType, aclTensor **tensor)64+ aclDataType dataType, aclTensor** tensor)
65{65{
66 auto size = GetShapeSize(shape) * sizeof(T);66 auto size = GetShapeSize(shape) * sizeof(T);
67 // 调用aclrtMalloc申请device侧内存67 // 调用aclrtMalloc申请device侧内存
@@ -78,15 +78,8 @@ int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &
78 }78 }
79 79 
80 // 调用aclCreateTensor接口创建aclTensor80 // 调用aclCreateTensor接口创建aclTensor
81- *tensor = aclCreateTensor(shape.data(),81+ *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
82- shape.size(),82+ shape.data(), shape.size(), *deviceAddr);
83- dataType,
84- strides.data(),
85- 0,
86- aclFormat::ACL_FORMAT_ND,
87- shape.data(),
88- shape.size(),
89- *deviceAddr);
90 return 0;83 return 0;
91}84}
92 85 
@@ -123,15 +116,15 @@ int main()
123 inputImagHostData[i] = dis(gen);116 inputImagHostData[i] = dis(gen);
124 }117 }
125 118 
126- void *inputRealDeviceAddr = nullptr;119+ void* inputRealDeviceAddr = nullptr;
127- void *inputImagDeviceAddr = nullptr;120+ void* inputImagDeviceAddr = nullptr;
128- void *outputRealDeviceAddr = nullptr;121+ void* outputRealDeviceAddr = nullptr;
129- void *outputImagDeviceAddr = nullptr;122+ void* outputImagDeviceAddr = nullptr;
130 123 
131- aclTensor *inputReal = nullptr;124+ aclTensor* inputReal = nullptr;
132- aclTensor *inputImag = nullptr;125+ aclTensor* inputImag = nullptr;
133- aclTensor *outputReal = nullptr;126+ aclTensor* outputReal = nullptr;
134- aclTensor *outputImag = nullptr;127+ aclTensor* outputImag = nullptr;
135 128 
136 ret = CreateAclTensor(inputRealHostData, selfShape, &inputRealDeviceAddr, aclDataType::ACL_FLOAT, &inputReal);129 ret = CreateAclTensor(inputRealHostData, selfShape, &inputRealDeviceAddr, aclDataType::ACL_FLOAT, &inputReal);
137 CHECK_RET(ret == ::ACL_SUCCESS, return ret);130 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
@@ -146,16 +139,17 @@ int main()
146 asdFftHandle handle;139 asdFftHandle handle;
147 asdFftCreate(handle);140 asdFftCreate(handle);
148 141 
149- asdFftMakePlan3D(handle, Nfft1, Nfft2, Nfft3, asdFftType::ASCEND_FFT_C2C_SEP, asdFftDirection::ASCEND_FFT_FORWARD, batch);142+ asdFftMakePlan3D(handle, Nfft1, Nfft2, Nfft3, asdFftType::ASCEND_FFT_C2C_SEP, asdFftDirection::ASCEND_FFT_FORWARD,
143+ batch);
150 144 
151 size_t work_size;145 size_t work_size;
152 asdFftGetWorkspaceSize(handle, work_size);146 asdFftGetWorkspaceSize(handle, work_size);
153- void *workspaceAddr = nullptr;147+ void* workspaceAddr = nullptr;
154 if (work_size > 0) {148 if (work_size > 0) {
155 ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST);149 ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST);
156 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);150 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);
157 }151 }
158- asdFftSetWorkspace(handle, (uint8_t *)workspaceAddr);152+ asdFftSetWorkspace(handle, (uint8_t*)workspaceAddr);
159 153 
160 asdFftSetStream(handle, stream);154 asdFftSetStream(handle, stream);
161 ASD_STATUS_CHECK(asdFftExecC2CSeparated(handle, inputReal, inputImag, outputReal, outputImag));155 ASD_STATUS_CHECK(asdFftExecC2CSeparated(handle, inputReal, inputImag, outputReal, outputImag));
@@ -170,23 +164,14 @@ int main()
170 std::vector<float> outImagData(size, 0);164 std::vector<float> outImagData(size, 0);
171 std::vector<float> workspaceData(size * 2, -1);165 std::vector<float> workspaceData(size * 2, -1);
172 166 
173- ret = aclrtMemcpy(outRealData.data(),167+ ret = aclrtMemcpy(outRealData.data(), outRealData.size() * sizeof(outRealData[0]), outputRealDeviceAddr,
174- outRealData.size() * sizeof(outRealData[0]),168+ size * sizeof(outRealData[0]), ACL_MEMCPY_DEVICE_TO_HOST);
175- outputRealDeviceAddr,
176- size * sizeof(outRealData[0]),
177- ACL_MEMCPY_DEVICE_TO_HOST);
178 169 
179- ret = aclrtMemcpy(outImagData.data(),170+ ret = aclrtMemcpy(outImagData.data(), outImagData.size() * sizeof(outImagData[0]), outputImagDeviceAddr,
180- outImagData.size() * sizeof(outImagData[0]),171+ size * sizeof(outImagData[0]), ACL_MEMCPY_DEVICE_TO_HOST);
181- outputImagDeviceAddr,
182- size * sizeof(outImagData[0]),
183- ACL_MEMCPY_DEVICE_TO_HOST);
184 172 
185- ret = aclrtMemcpy(workspaceData.data(),173+ ret = aclrtMemcpy(workspaceData.data(), workspaceData.size() * sizeof(workspaceData[0]), workspaceAddr,
186- workspaceData.size() * sizeof(workspaceData[0]),174+ workspaceData.size() * sizeof(workspaceData[0]), ACL_MEMCPY_DEVICE_TO_HOST);
187- workspaceAddr,
188- workspaceData.size() * sizeof(workspaceData[0]),
189- ACL_MEMCPY_DEVICE_TO_HOST);
190 175 
191 // 打印输出tensor值中前16个176 // 打印输出tensor值中前16个
192 std::cout << "real part:" << std::endl;177 std::cout << "real part:" << std::endl;
@@ -227,4 +212,4 @@ int main()
227 aclrtResetDevice(deviceId);212 aclrtResetDevice(deviceId);
228 aclFinalize();213 aclFinalize();
229 return 0;214 return 0;
230-}215+}
Mexample/A2/FFT/3d/example_acl_fft_c2r_3d.cpp+23-33
@@ -27,16 +27,16 @@ using namespace AsdSip;
27 printf(message, ##__VA_ARGS__); \27 printf(message, ##__VA_ARGS__); \
28 } while (0)28 } while (0)
29 29 
30-#define ASD_STATUS_CHECK(err) \30+#define ASD_STATUS_CHECK(err) \
31- do { \31+ do { \
32- AsdSip::AspbStatus err_ = (err); \32+ AsdSip::AspbStatus err_ = (err); \
33- if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \33+ if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
34 std::cout << "Execute failed." << std::endl; \34 std::cout << "Execute failed." << std::endl; \
35- exit(-1); \35+ return -1; \
36- } \36+ } \
37 } while (0)37 } while (0)
38 38 
39-int64_t GetShapeSize(const std::vector<int64_t> &shape)39+int64_t GetShapeSize(const std::vector<int64_t>& shape)
40{40{
41 int64_t shapeSize = 1;41 int64_t shapeSize = 1;
42 for (auto i : shape) {42 for (auto i : shape) {
@@ -45,7 +45,7 @@ int64_t GetShapeSize(const std::vector<int64_t> &shape)
45 return shapeSize;45 return shapeSize;
46}46}
47 47 
48-int Init(int32_t deviceId, aclrtStream *stream)48+int Init(int32_t deviceId, aclrtStream* stream)
49{49{
50 // 固定写法，AscendCL初始化50 // 固定写法，AscendCL初始化
51 auto ret = aclInit(nullptr);51 auto ret = aclInit(nullptr);
@@ -58,8 +58,8 @@ int Init(int32_t deviceId, aclrtStream *stream)
58}58}
59 59 
60template <typename T>60template <typename T>
61-int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr,61+int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
62- aclDataType dataType, aclTensor **tensor)62+ aclDataType dataType, aclTensor** tensor)
63{63{
64 auto size = GetShapeSize(shape) * sizeof(T);64 auto size = GetShapeSize(shape) * sizeof(T);
65 // 调用aclrtMalloc申请device侧内存65 // 调用aclrtMalloc申请device侧内存
@@ -76,19 +76,11 @@ int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &
76 }76 }
77 77 
78 // 调用aclCreateTensor接口创建aclTensor78 // 调用aclCreateTensor接口创建aclTensor
79- *tensor = aclCreateTensor(shape.data(),79+ *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
80- shape.size(),80+ shape.data(), shape.size(), *deviceAddr);
81- dataType,
82- strides.data(),
83- 0,
84- aclFormat::ACL_FORMAT_ND,
85- shape.data(),
86- shape.size(),
87- *deviceAddr);
88 return 0;81 return 0;
89}82}
90 83 
91- 
92int main()84int main()
93{85{
94 int32_t deviceId = 0;86 int32_t deviceId = 0;
@@ -109,10 +101,10 @@ int main()
109 inputHostData[i] = std::complex<float>(i, i + 1);101 inputHostData[i] = std::complex<float>(i, i + 1);
110 }102 }
111 std::vector<float> outHostData(tensorOutSize, 0);103 std::vector<float> outHostData(tensorOutSize, 0);
112- void *inputDeviceAddr = nullptr;104+ void* inputDeviceAddr = nullptr;
113- void *outDeviceAddr = nullptr;105+ void* outDeviceAddr = nullptr;
114- aclTensor *input = nullptr;106+ aclTensor* input = nullptr;
115- aclTensor *out = nullptr;107+ aclTensor* out = nullptr;
116 ret = CreateAclTensor(inputHostData, selfShape, &inputDeviceAddr, aclDataType::ACL_COMPLEX64, &input);108 ret = CreateAclTensor(inputHostData, selfShape, &inputDeviceAddr, aclDataType::ACL_COMPLEX64, &input);
117 CHECK_RET(ret == ::ACL_SUCCESS, return ret);109 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
118 ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_FLOAT, &out);110 ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_FLOAT, &out);
@@ -120,16 +112,17 @@ int main()
120 asdFftHandle handle;112 asdFftHandle handle;
121 asdFftCreate(handle);113 asdFftCreate(handle);
122 114 
123- asdFftMakePlan3D(handle, Nfft1, Nfft2, Nfft3, asdFftType::ASCEND_FFT_C2R, asdFftDirection::ASCEND_FFT_FORWARD, batch);115+ asdFftMakePlan3D(handle, Nfft1, Nfft2, Nfft3, asdFftType::ASCEND_FFT_C2R, asdFftDirection::ASCEND_FFT_FORWARD,
116+ batch);
124 117 
125 size_t work_size;118 size_t work_size;
126 asdFftGetWorkspaceSize(handle, work_size);119 asdFftGetWorkspaceSize(handle, work_size);
127- void *workspaceAddr = nullptr;120+ void* workspaceAddr = nullptr;
128 if (work_size > 0) {121 if (work_size > 0) {
129 ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST);122 ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST);
130 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);123 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);
131 }124 }
132- asdFftSetWorkspace(handle, (uint8_t *)workspaceAddr);125+ asdFftSetWorkspace(handle, (uint8_t*)workspaceAddr);
133 126 
134 asdFftSetStream(handle, stream);127 asdFftSetStream(handle, stream);
135 ASD_STATUS_CHECK(asdFftExecC2R(handle, input, out));128 ASD_STATUS_CHECK(asdFftExecC2R(handle, input, out));
@@ -141,11 +134,8 @@ int main()
141 134 
142 auto size = GetShapeSize(outShape);135 auto size = GetShapeSize(outShape);
143 std::vector<float> outData(size, 0);136 std::vector<float> outData(size, 0);
144- ret = aclrtMemcpy(outData.data(),137+ ret = aclrtMemcpy(outData.data(), outData.size() * sizeof(outData[0]), outDeviceAddr, size * sizeof(outData[0]),
145- outData.size() * sizeof(outData[0]),138+ ACL_MEMCPY_DEVICE_TO_HOST);
146- outDeviceAddr,
147- size * sizeof(outData[0]),
148- ACL_MEMCPY_DEVICE_TO_HOST);
149 139 
150 // 打印输出tensor值中前16个140 // 打印输出tensor值中前16个
151 for (int64_t i = 0; i < std::min(static_cast<int64_t>(16), tensorOutSize); i++) {141 for (int64_t i = 0; i < std::min(static_cast<int64_t>(16), tensorOutSize); i++) {
@@ -166,4 +156,4 @@ int main()
166 aclrtResetDevice(deviceId);156 aclrtResetDevice(deviceId);
167 aclFinalize();157 aclFinalize();
168 return 0;158 return 0;
169-}159+}
Mexample/A2/FFT/3d/example_acl_fft_r2c_3d.cpp+23-32
@@ -27,16 +27,16 @@ using namespace AsdSip;
27 printf(message, ##__VA_ARGS__); \27 printf(message, ##__VA_ARGS__); \
28 } while (0)28 } while (0)
29 29 
30-#define ASD_STATUS_CHECK(err) \30+#define ASD_STATUS_CHECK(err) \
31- do { \31+ do { \
32- AsdSip::AspbStatus err_ = (err); \32+ AsdSip::AspbStatus err_ = (err); \
33- if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \33+ if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
34 std::cout << "Execute failed." << std::endl; \34 std::cout << "Execute failed." << std::endl; \
35- exit(-1); \35+ return -1; \
36- } \36+ } \
37 } while (0)37 } while (0)
38 38 
39-int64_t GetShapeSize(const std::vector<int64_t> &shape)39+int64_t GetShapeSize(const std::vector<int64_t>& shape)
40{40{
41 int64_t shapeSize = 1;41 int64_t shapeSize = 1;
42 for (auto i : shape) {42 for (auto i : shape) {
@@ -45,7 +45,7 @@ int64_t GetShapeSize(const std::vector<int64_t> &shape)
45 return shapeSize;45 return shapeSize;
46}46}
47 47 
48-int Init(int32_t deviceId, aclrtStream *stream)48+int Init(int32_t deviceId, aclrtStream* stream)
49{49{
50 // 固定写法，AscendCL初始化50 // 固定写法，AscendCL初始化
51 auto ret = aclInit(nullptr);51 auto ret = aclInit(nullptr);
@@ -58,8 +58,8 @@ int Init(int32_t deviceId, aclrtStream *stream)
58}58}
59 59 
60template <typename T>60template <typename T>
61-int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr,61+int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
62- aclDataType dataType, aclTensor **tensor)62+ aclDataType dataType, aclTensor** tensor)
63{63{
64 auto size = GetShapeSize(shape) * sizeof(T);64 auto size = GetShapeSize(shape) * sizeof(T);
65 // 调用aclrtMalloc申请device侧内存65 // 调用aclrtMalloc申请device侧内存
@@ -76,15 +76,8 @@ int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &
76 }76 }
77 77 
78 // 调用aclCreateTensor接口创建aclTensor78 // 调用aclCreateTensor接口创建aclTensor
79- *tensor = aclCreateTensor(shape.data(),79+ *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
80- shape.size(),80+ shape.data(), shape.size(), *deviceAddr);
81- dataType,
82- strides.data(),
83- 0,
84- aclFormat::ACL_FORMAT_ND,
85- shape.data(),
86- shape.size(),
87- *deviceAddr);
88 return 0;81 return 0;
89}82}
90 83 
@@ -108,10 +101,10 @@ int main()
108 }101 }
109 std::vector<std::complex<float>> outHostData(tensorInSize, std::complex<float>(0, 0));102 std::vector<std::complex<float>> outHostData(tensorInSize, std::complex<float>(0, 0));
110 103 
111- void *inputDeviceAddr = nullptr;104+ void* inputDeviceAddr = nullptr;
112- void *outDeviceAddr = nullptr;105+ void* outDeviceAddr = nullptr;
113- aclTensor *input = nullptr;106+ aclTensor* input = nullptr;
114- aclTensor *out = nullptr;107+ aclTensor* out = nullptr;
115 ret = CreateAclTensor(inputHostData, selfShape, &inputDeviceAddr, aclDataType::ACL_FLOAT, &input);108 ret = CreateAclTensor(inputHostData, selfShape, &inputDeviceAddr, aclDataType::ACL_FLOAT, &input);
116 CHECK_RET(ret == ::ACL_SUCCESS, return ret);109 CHECK_RET(ret == ::ACL_SUCCESS, return ret);
117 ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_COMPLEX64, &out);110 ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_COMPLEX64, &out);
@@ -120,16 +113,17 @@ int main()
120 asdFftHandle handle;113 asdFftHandle handle;
121 asdFftCreate(handle);114 asdFftCreate(handle);
122 115 
123- asdFftMakePlan3D(handle, Nfft1, Nfft2, Nfft3, asdFftType::ASCEND_FFT_R2C, asdFftDirection::ASCEND_FFT_FORWARD, batch);116+ asdFftMakePlan3D(handle, Nfft1, Nfft2, Nfft3, asdFftType::ASCEND_FFT_R2C, asdFftDirection::ASCEND_FFT_FORWARD,
117+ batch);
124 118 
125 size_t work_size;119 size_t work_size;
126 asdFftGetWorkspaceSize(handle, work_size);120 asdFftGetWorkspaceSize(handle, work_size);
127- void *workspaceAddr = nullptr;121+ void* workspaceAddr = nullptr;
128 if (work_size > 0) {122 if (work_size > 0) {
129 ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST);123 ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST);
130 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);124 CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);
131 }125 }
132- asdFftSetWorkspace(handle, (uint8_t *)workspaceAddr);126+ asdFftSetWorkspace(handle, (uint8_t*)workspaceAddr);
133 127 
134 asdFftSetStream(handle, stream);128 asdFftSetStream(handle, stream);
135 ASD_STATUS_CHECK(asdFftExecR2C(handle, input, out));129 ASD_STATUS_CHECK(asdFftExecR2C(handle, input, out));
@@ -141,11 +135,8 @@ int main()
141 135 
142 auto size = GetShapeSize(outShape);136 auto size = GetShapeSize(outShape);
143 std::vector<std::complex<float>> outData(size, 0);137 std::vector<std::complex<float>> outData(size, 0);
144- ret = aclrtMemcpy(outData.data(),138+ ret = aclrtMemcpy(outData.data(), outData.size() * sizeof(outData[0]), outDeviceAddr, size * sizeof(outData[0]),
145- outData.size() * sizeof(outData[0]),139+ ACL_MEMCPY_DEVICE_TO_HOST);
146- outDeviceAddr,
147- size * sizeof(outData[0]),
148- ACL_MEMCPY_DEVICE_TO_HOST);
149 140 
150 // 打印输出tensor值中前16个141 // 打印输出tensor值中前16个
151 for (int64_t i = 0; i < std::min(static_cast<int64_t>(16), tensorOutSize); i++) {142 for (int64_t i = 0; i < std::min(static_cast<int64_t>(16), tensorOutSize); i++) {
@@ -166,4 +157,4 @@ int main()
166 aclrtResetDevice(deviceId);157 aclrtResetDevice(deviceId);
167 aclFinalize();158 aclFinalize();
168 return 0;159 return 0;
169-}160+}
Mexample/A2/Filter/asd_convolve/example_asd_convolve_complex32.cpp+1-1
@@ -25,7 +25,7 @@ using half = op::fp16_t;
25 AsdSip::AspbStatus err_ = (err); \25 AsdSip::AspbStatus err_ = (err); \
26 if (err_ != AsdSip::ACL_SUCCESS) { \26 if (err_ != AsdSip::ACL_SUCCESS) { \
27 std::cout << "Execute failed." << std::endl; \27 std::cout << "Execute failed." << std::endl; \
28- exit(-1); \28+ return -1; \
29 } \29 } \
30 } while (0)30 } while (0)
31 31 
Mexample/A2/Filter/asd_convolve/example_asd_convolve_complex64.cpp+1-1
@@ -24,7 +24,7 @@ using namespace AsdSip;
24 AsdSip::AspbStatus err_ = (err); \24 AsdSip::AspbStatus err_ = (err); \
25 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \25 if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \
26 std::cout << "Execute failed." << std::endl; \26 std::cout << "Execute failed." << std::endl; \
27- exit(-1); \27+ return -1; \
28 } \28 } \
29 } while (0)29 } while (0)
30 30 
Minstall_deps.sh+1-2
@@ -362,8 +362,7 @@ install_dos2unix() {
362 if command -v dos2unix &> /dev/null; then362 if command -v dos2unix &> /dev/null; then
363 echo "dos2unix installed successfully"363 echo "dos2unix installed successfully"
364 else364 else
365- echo "dos2unix installation failed"365+ echo "dos2unix installation failed, can be ignored"
366- exit 1
367 fi366 fi
368}367}
369 368 
Mscripts/build_util.py+75-54
@@ -1,5 +1,5 @@
1# -*- coding: UTF-8 -*-1# -*- coding: UTF-8 -*-
2-# Copyright (c) 2025 Huawei Technologies Co., Ltd.2+# Copyright (c) 2026 Huawei Technologies Co., Ltd.
3# This program is free software, you can redistribute it and/or modify it under the terms and conditions of3# This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4# CANN Open Software License Agreement Version 2.0 (the "License").4# CANN Open Software License Agreement Version 2.0 (the "License").
5# Please refer to the License for details. You may not use this file except in compliance with the License.5# Please refer to the License for details. You may not use this file except in compliance with the License.
@@ -7,45 +7,46 @@
7# INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7# INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9 9 
10-import os
11import configparser10import configparser
12import json11import json
13import logging12import logging
13+import os
14import shutil14import shutil
15import stat15import stat
16-import re16+import sys
17 17 
18 18 
19# sycl-target --show-targets19# sycl-target --show-targets
20def get_build_target_list():20def get_build_target_list():
21- usr_config_file_path = os.getenv("BUILD_CONFIG_FILE", '')21+ usr_config_file_path = os.getenv("BUILD_CONFIG_FILE", "")
22- if usr_config_file_path == '':22+ if usr_config_file_path == "":
23 script_file_path = os.path.realpath(__file__)23 script_file_path = os.path.realpath(__file__)
24- build_config_json_file_path = os.path.join(os.path.dirname(24+ build_config_json_file_path = os.path.join(
25- script_file_path), "../configs/build_config.json")25+ os.path.dirname(script_file_path), "../configs/build_config.json"
26+ )
26 else:27 else:
27 build_config_json_file_path = usr_config_file_path28 build_config_json_file_path = usr_config_file_path
28 device_list = []29 device_list = []
29 try:30 try:
30 with open(build_config_json_file_path) as conf_file:31 with open(build_config_json_file_path) as conf_file:
31 conf = json.load(conf_file)32 conf = json.load(conf_file)
32- target_option = conf['targets']33+ target_option = conf["targets"]
33 for target, switch in target_option.items():34 for target, switch in target_option.items():
34 if switch is True:35 if switch is True:
35 device_list.append(target)36 device_list.append(target)
36 except FileNotFoundError:37 except FileNotFoundError:
37 logging.error("file %s is not found!", build_config_json_file_path)38 logging.error("file %s is not found!", build_config_json_file_path)
38- exit(1)39+ sys.exit(1)
39 except json.decoder.JSONDecodeError:40 except json.decoder.JSONDecodeError:
40 logging.error("file %s is not json file!", build_config_json_file_path)41 logging.error("file %s is not json file!", build_config_json_file_path)
41- exit(1)42+ sys.exit(1)
42 except KeyError:43 except KeyError:
43 logging.error("key 'targets' is not found in %s!", build_config_json_file_path)44 logging.error("key 'targets' is not found in %s!", build_config_json_file_path)
44- exit(1)45+ sys.exit(1)
45 46 
46 if len(device_list) == 0:47 if len(device_list) == 0:
47 logging.error("no target device is set")48 logging.error("no target device is set")
48- exit(1)49+ sys.exit(1)
49 50 
50 device_list = list(set(device_list))51 device_list = list(set(device_list))
51 return device_list52 return device_list
@@ -53,10 +54,12 @@ def get_build_target_list():
53 54 
54def get_info_from_file(file_path):55def get_info_from_file(file_path):
55 result = True56 result = True
56- tactic_info = dict()57+ tactic_info = {}
57- magic_dict = {"RT_DEV_BINARY_MAGIC_ELF": str(0x43554245),58+ magic_dict = {
58- "RT_DEV_BINARY_MAGIC_ELF_AIVEC": str(0x41415246),59+ "RT_DEV_BINARY_MAGIC_ELF": str(0x43554245),
59- "RT_DEV_BINARY_MAGIC_ELF_AICUBE": str(0x41494343)}60+ "RT_DEV_BINARY_MAGIC_ELF_AIVEC": str(0x41415246),
61+ "RT_DEV_BINARY_MAGIC_ELF_AICUBE": str(0x41494343),
62+ }
60 try:63 try:
61 with open(file_path) as f:64 with open(file_path) as f:
62 text = json.load(f)65 text = json.load(f)
@@ -71,7 +74,7 @@ def get_info_from_file(file_path):
71 else:74 else:
72 tactic_info["magic"] = magic_dict[magic]75 tactic_info["magic"] = magic_dict[magic]
73 if "kernelList" in text:76 if "kernelList" in text:
74- tactic_info["kernelList"] = ','.join(77+ tactic_info["kernelList"] = ",".join(
75 item["kernelName"] for item in text["kernelList"]78 item["kernelName"] for item in text["kernelList"]
76 )79 )
77 else:80 else:
@@ -89,8 +92,10 @@ def get_info_from_file(file_path):
89 92 
90 93 
91def write_meta(meta_info, output_path, target_version):94def write_meta(meta_info, output_path, target_version):
92- meta_path = os.path.join(output_path, 'meta.ini')95+ meta_path = os.path.join(output_path, "meta.ini")
93- with os.fdopen(os.open(meta_path, os.O_WRONLY | os.O_CREAT, stat.S_IWUSR | stat.S_IRUSR), 'w+') as fmeta:96+ with os.fdopen(
97+ os.open(meta_path, os.O_WRONLY | os.O_CREAT, stat.S_IWUSR | stat.S_IRUSR), "w+"
98+ ) as fmeta:
94 fmeta.write("$Version=1.0\n")99 fmeta.write("$Version=1.0\n")
95 fmeta.write(f"DeviceKernelVersion={target_version}\n")100 fmeta.write(f"DeviceKernelVersion={target_version}\n")
96 fmeta.write(f"$Object.Count={len(meta_info)}\n")101 fmeta.write(f"$Object.Count={len(meta_info)}\n")
@@ -107,8 +112,7 @@ def write_meta(meta_info, output_path, target_version):
107 112 
108# 目前只支持一个tactic文件夹下一个.o和.json文件113# 目前只支持一个tactic文件夹下一个.o和.json文件
109def copy_ascendc_code(meta_info, env_cache_dir, target_version, output_path):114def copy_ascendc_code(meta_info, env_cache_dir, target_version, output_path):
110- op_kernels_version_dir = os.path.join(115+ op_kernels_version_dir = os.path.join(env_cache_dir, "op_kernels", target_version)
111- env_cache_dir, "op_kernels", target_version)
112 if not os.path.exists(op_kernels_version_dir):116 if not os.path.exists(op_kernels_version_dir):
113 return 0117 return 0
114 code_file_count = 0118 code_file_count = 0
@@ -120,28 +124,34 @@ def copy_ascendc_code(meta_info, env_cache_dir, target_version, output_path):
120 for tactic in os.listdir(operation_dir):124 for tactic in os.listdir(operation_dir):
121 tactic_dir = os.path.join(operation_dir, tactic)125 tactic_dir = os.path.join(operation_dir, tactic)
122 for file in os.listdir(tactic_dir):126 for file in os.listdir(tactic_dir):
123- if not file.endswith('.json'):127+ if not file.endswith(".json"):
124 continue128 continue
125- code_file = os.path.join(tactic_dir, file[:-4] + 'o')129+ code_file = os.path.join(tactic_dir, file[:-4] + "o")
126 if not os.path.exists(code_file):130 if not os.path.exists(code_file):
127 logging.error("file %s has no object file.", file)131 logging.error("file %s has no object file.", file)
128- exit(1)132+ sys.exit(1)
129 json_file = os.path.join(tactic_dir, file)133 json_file = os.path.join(tactic_dir, file)
130 tactic_info, result = get_info_from_file(json_file)134 tactic_info, result = get_info_from_file(json_file)
131 if not result:135 if not result:
132 logging.error("failed to parse file %s.", json_file)136 logging.error("failed to parse file %s.", json_file)
133- exit(1)137+ sys.exit(1)
134 relative_to_path = os.path.join(operation, file)138 relative_to_path = os.path.join(operation, file)
135 to_path = os.path.join(output_operation_dir, file)139 to_path = os.path.join(output_operation_dir, file)
136 shutil.copyfile(code_file, to_path)140 shutil.copyfile(code_file, to_path)
137 try:141 try:
138 compile_info_str = json.dumps(tactic_info["compileInfo"])142 compile_info_str = json.dumps(tactic_info["compileInfo"])
139 meta_info[tactic] = (143 meta_info[tactic] = (
140- relative_to_path, tactic, tactic_info["kernelList"], compile_info_str,144+ relative_to_path,
141- tactic_info["opParaSize"], tactic_info["coreType"], tactic_info["magic"])145+ tactic,
146+ tactic_info["kernelList"],
147+ compile_info_str,
148+ tactic_info["opParaSize"],
149+ tactic_info["coreType"],
150+ tactic_info["magic"],
151+ )
142 except KeyError:152 except KeyError:
143 logging.error("%s get compile or meta info error", tactic)153 logging.error("%s get compile or meta info error", tactic)
144- exit(1)154+ sys.exit(1)
145 code_file_count += 1155 code_file_count += 1
146 return code_file_count156 return code_file_count
147 157 
@@ -149,13 +159,13 @@ def copy_ascendc_code(meta_info, env_cache_dir, target_version, output_path):
149def copy_tbe_code_all_version(input_paras):159def copy_tbe_code_all_version(input_paras):
150 tbe_sections = input_paras["tbe_ini"].sections()160 tbe_sections = input_paras["tbe_ini"].sections()
151 for target_version in input_paras["target_version_list"]:161 for target_version in input_paras["target_version_list"]:
152- output_path = os.path.join(162+ output_path = os.path.join(input_paras["env_cache_dir"], "obj", target_version)
153- input_paras["env_cache_dir"], "obj", target_version)
154 if not os.path.exists(output_path):163 if not os.path.exists(output_path):
155 os.makedirs(output_path)164 os.makedirs(output_path)
156 meta_info = {}165 meta_info = {}
157 target_version_path = os.path.join(166 target_version_path = os.path.join(
158- input_paras["tbe_kernel_path"], target_version)167+ input_paras["tbe_kernel_path"], target_version
168+ )
159 169 
160 for op_name in tbe_sections:170 for op_name in tbe_sections:
161 op_dir_path = os.path.join(output_path, op_name)171 op_dir_path = os.path.join(output_path, op_name)
@@ -163,19 +173,21 @@ def copy_tbe_code_all_version(input_paras):
163 os.mkdir(op_dir_path)173 os.mkdir(op_dir_path)
164 items = dict(input_paras["tbe_ini"].items(op_name))174 items = dict(input_paras["tbe_ini"].items(op_name))
165 for op_key, relative_op_path in items.items():175 for op_key, relative_op_path in items.items():
166- if '.' in op_key:176+ if "." in op_key:
167- op_key, version_op_key = op_key.split('.')177+ op_key, version_op_key = op_key.split(".")
168 if version_op_key != target_version:178 if version_op_key != target_version:
169 continue179 continue
170 180 
171- tactic_info, ret = get_info_from_file(os.path.join(181+ tactic_info, ret = get_info_from_file(
172- target_version_path, relative_op_path))182+ os.path.join(target_version_path, relative_op_path)
183+ )
173 if not ret:184 if not ret:
174 logging.error("failed to parse json file %s", relative_op_path)185 logging.error("failed to parse json file %s", relative_op_path)
175- exit(1)186+ sys.exit(1)
176 187 
177 from_path = os.path.join(188 from_path = os.path.join(
178- target_version_path, relative_op_path[:-4] + 'o')189+ target_version_path, relative_op_path[:-4] + "o"
190+ )
179 object_name = os.path.basename(from_path)191 object_name = os.path.basename(from_path)
180 to_path = os.path.join(op_dir_path, object_name)192 to_path = os.path.join(op_dir_path, object_name)
181 relative_to_path = os.path.join(op_name, object_name)193 relative_to_path = os.path.join(op_name, object_name)
@@ -184,16 +196,22 @@ def copy_tbe_code_all_version(input_paras):
184 try:196 try:
185 compile_info_str = json.dumps(tactic_info["compileInfo"])197 compile_info_str = json.dumps(tactic_info["compileInfo"])
186 meta_info[op_key] = (198 meta_info[op_key] = (
187- relative_to_path, op_name, tactic_info["kernelList"], compile_info_str,199+ relative_to_path,
188- tactic_info["opParaSize"], tactic_info["coreType"], tactic_info["magic"])200+ op_name,
201+ tactic_info["kernelList"],
202+ compile_info_str,
203+ tactic_info["opParaSize"],
204+ tactic_info["coreType"],
205+ tactic_info["magic"],
206+ )
189 except KeyError:207 except KeyError:
190 logging.error("%s get compile or meta info error", op_name)208 logging.error("%s get compile or meta info error", op_name)
191- exit(1)209+ sys.exit(1)
192 210 
193 ascendc_file_count = copy_ascendc_code(211 ascendc_file_count = copy_ascendc_code(
194- meta_info, input_paras["env_cache_dir"], target_version, output_path)212+ meta_info, input_paras["env_cache_dir"], target_version, output_path
195- logging.info(213+ )
196- f"{target_version} has {ascendc_file_count} AscendC tactics.")214+ logging.info(f"{target_version} has {ascendc_file_count} AscendC tactics.")
197 215 
198 write_meta(meta_info, output_path, target_version)216 write_meta(meta_info, output_path, target_version)
199 217 
@@ -203,31 +221,34 @@ def copy_tbe_device_code():
203 env_cache_dir = os.getenv("CACHE_DIR")221 env_cache_dir = os.getenv("CACHE_DIR")
204 tbe_kernel_path = os.getenv("ASDOPS_KERNEL_PATH")222 tbe_kernel_path = os.getenv("ASDOPS_KERNEL_PATH")
205 if not (env_code_root and env_cache_dir and tbe_kernel_path):223 if not (env_code_root and env_cache_dir and tbe_kernel_path):
206- logging.error(224+ logging.error("env CODE_ROOT | OUTPUT_DIR | ASDOPS_SOURCE_DIR not exist!")
207- "env CODE_ROOT | OUTPUT_DIR | ASDOPS_SOURCE_DIR not exist!")225+ sys.exit(1)
208- exit(1)
209 logging.info(f"tbe_kernel_path: {tbe_kernel_path}")226 logging.info(f"tbe_kernel_path: {tbe_kernel_path}")
210 input_path = os.path.join(env_code_root, "configs/tbe_tactic_json.ini")227 input_path = os.path.join(env_code_root, "configs/tbe_tactic_json.ini")
211 if not os.path.exists(input_path):228 if not os.path.exists(input_path):
212 logging.error("ini file: %s not exist!", input_path)229 logging.error("ini file: %s not exist!", input_path)
213- exit(1)230+ sys.exit(1)
214 tbe_ini = configparser.RawConfigParser()231 tbe_ini = configparser.RawConfigParser()
215 tbe_ini.optionxform = lambda option: option232 tbe_ini.optionxform = lambda option: option
216 try:233 try:
217 tbe_ini.read(input_path)234 tbe_ini.read(input_path)
218 except configparser.MissingSectionHeaderError:235 except configparser.MissingSectionHeaderError:
219 logging.error("ini file: %s format error!", input_path)236 logging.error("ini file: %s format error!", input_path)
220- exit(1)237+ sys.exit(1)
221 except configparser.ParsingError:238 except configparser.ParsingError:
222 logging.error("ini file: %s format error!", input_path)239 logging.error("ini file: %s format error!", input_path)
223- exit(1)240+ sys.exit(1)
224 241 
225 target_version_list = get_build_target_list()242 target_version_list = get_build_target_list()
226- copy_tbe_code_all_version({"env_code_root": env_code_root,243+ copy_tbe_code_all_version(
227- "target_version_list": target_version_list,244+ {
228- "env_cache_dir": env_cache_dir,245+ "env_code_root": env_code_root,
229- "tbe_kernel_path": tbe_kernel_path,246+ "target_version_list": target_version_list,
230- "tbe_ini": tbe_ini})247+ "env_cache_dir": env_cache_dir,
248+ "tbe_kernel_path": tbe_kernel_path,
249+ "tbe_ini": tbe_ini,
250+ }
251+ )
231 os.remove(input_path)252 os.remove(input_path)
232 253 
233 254 
Mscripts/compile_ascendc.py+235-112
@@ -1,6 +1,6 @@
1#!/usr/bin/env python1#!/usr/bin/env python
2# -*- coding: UTF-8 -*-2# -*- coding: UTF-8 -*-
3-# Copyright (c) 2024 Huawei Technologies Co., Ltd.3+# Copyright (c) 2026 Huawei Technologies Co., Ltd.
4# This program is free software, you can redistribute it and/or modify it under the terms and conditions of4# This program is free software, you can redistribute it and/or modify it under the terms and conditions of
5# CANN Open Software License Agreement Version 2.0 (the "License").5# CANN Open Software License Agreement Version 2.0 (the "License").
6# Please refer to the License for details. You may not use this file except in compliance with the License.6# Please refer to the License for details. You may not use this file except in compliance with the License.
@@ -10,47 +10,62 @@
10 10 
11import argparse11import argparse
12import json12import json
13+import logging
13import os14import os
14import re15import re
15import stat16import stat
16-import logging17+import sys
17 18 
18 19 
19def parse_args():20def parse_args():
20 parser = argparse.ArgumentParser()21 parser = argparse.ArgumentParser()
21- parser.add_argument('--soc', type=str, required=True)22+ parser.add_argument("--soc", type=str, required=True)
22- parser.add_argument('--channel', type=str, required=True)23+ parser.add_argument("--channel", type=str, required=True)
23- parser.add_argument('--srcs', type=str, required=True)24+ parser.add_argument("--srcs", type=str, required=True)
24- parser.add_argument('--dst', type=str, required=True)25+ parser.add_argument("--dst", type=str, required=True)
25- parser.add_argument('--code_root', type=str, required=True)26+ parser.add_argument("--code_root", type=str, required=True)
26- parser.add_argument('--kernel', type=str, required=True)27+ parser.add_argument("--kernel", type=str, required=True)
27- parser.add_argument('--use_msdebug', type=str)28+ parser.add_argument("--use_msdebug", type=str)
28- parser.add_argument('--use_mssanitizer', type=str, required=True)29+ parser.add_argument("--use_mssanitizer", type=str, required=True)
29- parser.add_argument('--no_warning', action='store_true')30+ parser.add_argument("--no_warning", action="store_true")
30- parser.add_argument('--include_directories', type=str, required=False, nargs="+")31+ parser.add_argument("--include_directories", type=str, required=False, nargs="+")
31- parser.add_argument('--use_ascendc_dump', action='store_true')32+ parser.add_argument("--use_ascendc_dump", action="store_true")
32 return parser.parse_args()33 return parser.parse_args()
33 34 
34 35 
35def gen_compile_cmd(args, dst: str, sub_arch: str, compile_options):36def gen_compile_cmd(args, dst: str, sub_arch: str, compile_options):
36- compile_cmd = [os.path.join(args.code_root, '3rdparty', 'compiler',37+ compile_cmd = [
37- 'ccec_compiler', 'bin', 'ccec'),38+ os.path.join(
38- '-c']39+ args.code_root, "3rdparty", "compiler", "ccec_compiler", "bin", "ccec"
40+ ),
41+ "-c",
42+ ]
39 if args.use_msdebug == "ON":43 if args.use_msdebug == "ON":
40- compile_cmd += ['-O0', '-g', '--cce-ignore-always-inline=true']44+ compile_cmd += ["-O0", "-g", "--cce-ignore-always-inline=true"]
41 else:45 else:
42- compile_cmd += ['-O2']46+ compile_cmd += ["-O2"]
43 compile_cmd += compile_options47 compile_cmd += compile_options
44- compile_cmd += [args.srcs, "--cce-aicore-arch=%s" % sub_arch,48+ compile_cmd += [
45- "--cce-aicore-only", "-o", dst,49+ args.srcs,
46- "-mllvm", "-cce-aicore-fp-ceiling=2"]50+ "--cce-aicore-arch=%s" % sub_arch,
51+ "--cce-aicore-only",
52+ "-o",
53+ dst,
54+ "-mllvm",
55+ "-cce-aicore-fp-ceiling=2",
56+ ]
47 if args.use_mssanitizer == "ON" and args.soc in ["ascend310p", "ascend910b"]:57 if args.use_mssanitizer == "ON" and args.soc in ["ascend310p", "ascend910b"]:
48- compile_cmd += ["-g", "--cce-enable-sanitizer",58+ compile_cmd += [
49- "-mllvm", "-cce-aicore-long-call",59+ "-g",
50- "-mllvm", "-cce-aicore-jump-expand=true"]60+ "--cce-enable-sanitizer",
61+ "-mllvm",
62+ "-cce-aicore-long-call",
63+ "-mllvm",
64+ "-cce-aicore-jump-expand=true",
65+ ]
51 compile_cmd += ["-std=c++17"]66 compile_cmd += ["-std=c++17"]
52 compile_cmd += ["--cce-mask-opt"]67 compile_cmd += ["--cce-mask-opt"]
53- if (args.use_ascendc_dump):68+ if args.use_ascendc_dump:
54 compile_cmd += ["-mllvm", "-cce-aicore-function-stack-size=0x4000"]69 compile_cmd += ["-mllvm", "-cce-aicore-function-stack-size=0x4000"]
55 compile_cmd += ["--cce-long-call=true"]70 compile_cmd += ["--cce-long-call=true"]
56 compile_cmd += ["-DASCENDC_DUMP=1"]71 compile_cmd += ["-DASCENDC_DUMP=1"]
@@ -59,27 +74,45 @@ def gen_compile_cmd(args, dst: str, sub_arch: str, compile_options):
59 74 
60 75 
61def gen_compile_cmd_v220(args, dst: str, sub_arch: str, compile_options):76def gen_compile_cmd_v220(args, dst: str, sub_arch: str, compile_options):
62- compile_cmd = [os.path.join(args.code_root, '3rdparty', 'compiler',77+ compile_cmd = [
63- 'ccec_compiler', 'bin', 'ccec'),78+ os.path.join(
64- '-c']79+ args.code_root, "3rdparty", "compiler", "ccec_compiler", "bin", "ccec"
80+ ),
81+ "-c",
82+ ]
65 if args.use_msdebug == "ON":83 if args.use_msdebug == "ON":
66- compile_cmd += ['-O0', '-g', '--cce-ignore-always-inline=true']84+ compile_cmd += ["-O0", "-g", "--cce-ignore-always-inline=true"]
67 else:85 else:
68- compile_cmd += ['-O3']86+ compile_cmd += ["-O3"]
69 compile_cmd += compile_options87 compile_cmd += compile_options
70- compile_cmd += [args.srcs, "--cce-aicore-arch=%s" % sub_arch,88+ compile_cmd += [
71- "--cce-aicore-only", "-o", dst,89+ args.srcs,
72- "-mllvm", "-cce-aicore-stack-size=0x8000",90+ "--cce-aicore-arch=%s" % sub_arch,
73- "-mllvm", "-cce-aicore-function-stack-size=0x8000",91+ "--cce-aicore-only",
74- "-mllvm", "-cce-aicore-record-overflow=true",92+ "-o",
75- "-mllvm", "-cce-aicore-addr-transform",93+ dst,
76- "-mllvm", "-cce-aicore-dcci-insert-for-scalar=false"]94+ "-mllvm",
95+ "-cce-aicore-stack-size=0x8000",
96+ "-mllvm",
97+ "-cce-aicore-function-stack-size=0x8000",
98+ "-mllvm",
99+ "-cce-aicore-record-overflow=true",
100+ "-mllvm",
101+ "-cce-aicore-addr-transform",
102+ "-mllvm",
103+ "-cce-aicore-dcci-insert-for-scalar=false",
104+ ]
77 if args.use_mssanitizer == "ON":105 if args.use_mssanitizer == "ON":
78- compile_cmd += ["-g", "--cce-enable-sanitizer",106+ compile_cmd += [
79- "-mllvm", "-cce-aicore-long-call",107+ "-g",
80- "-mllvm", "-cce-aicore-jump-expand=true"]108+ "--cce-enable-sanitizer",
109+ "-mllvm",
110+ "-cce-aicore-long-call",
111+ "-mllvm",
112+ "-cce-aicore-jump-expand=true",
113+ ]
81 compile_cmd += ["-std=c++17"]114 compile_cmd += ["-std=c++17"]
82- if (args.use_ascendc_dump):115+ if args.use_ascendc_dump:
83 compile_cmd += ["--cce-long-call=true"]116 compile_cmd += ["--cce-long-call=true"]
84 compile_cmd += ["-DASCENDC_DUMP=1"]117 compile_cmd += ["-DASCENDC_DUMP=1"]
85 compile_cmd += ["-DASCENDC_DEBUG"]118 compile_cmd += ["-DASCENDC_DEBUG"]
@@ -87,24 +120,38 @@ def gen_compile_cmd_v220(args, dst: str, sub_arch: str, compile_options):
87 120 
88 121 
89def gen_compile_cmd_v300(args, dst: str, sub_arch: str, compile_options):122def gen_compile_cmd_v300(args, dst: str, sub_arch: str, compile_options):
90- compile_cmd = [os.path.join(args.code_root, '3rdparty', 'compiler',123+ compile_cmd = [
91- 'ccec_compiler', 'bin', 'ccec'),124+ os.path.join(
92- '-c']125+ args.code_root, "3rdparty", "compiler", "ccec_compiler", "bin", "ccec"
126+ ),
127+ "-c",
128+ ]
93 if args.use_msdebug == "ON":129 if args.use_msdebug == "ON":
94- compile_cmd += ['-O0', '-g', '--cce-ignore-always-inline=true']130+ compile_cmd += ["-O0", "-g", "--cce-ignore-always-inline=true"]
95 else:131 else:
96- compile_cmd += ['-O3']132+ compile_cmd += ["-O3"]
97 compile_cmd += compile_options133 compile_cmd += compile_options
98- compile_cmd += [args.srcs, "--cce-aicore-arch=%s" % sub_arch,134+ compile_cmd += [
99- "--cce-aicore-only", "-o", dst,135+ args.srcs,
100- "-mllvm", "-cce-aicore-function-stack-size=0x4000",136+ "--cce-aicore-arch=%s" % sub_arch,
101- "-mllvm", "-cce-aicore-addr-transform",137+ "--cce-aicore-only",
102- "-mllvm", "--cce-aicore-or-combine=false",138+ "-o",
103- "-mllvm", "-instcombine-code-sinking=false",139+ dst,
104- "-mllvm", "-cce-aicore-jump-expand=false",140+ "-mllvm",
105- "-mllvm", "-cce-aicore-mask-opt=false"]141+ "-cce-aicore-function-stack-size=0x4000",
142+ "-mllvm",
143+ "-cce-aicore-addr-transform",
144+ "-mllvm",
145+ "--cce-aicore-or-combine=false",
146+ "-mllvm",
147+ "-instcombine-code-sinking=false",
148+ "-mllvm",
149+ "-cce-aicore-jump-expand=false",
150+ "-mllvm",
151+ "-cce-aicore-mask-opt=false",
152+ ]
106 compile_cmd += ["-std=c++17"]153 compile_cmd += ["-std=c++17"]
107- if (args.use_ascendc_dump):154+ if args.use_ascendc_dump:
108 compile_cmd += ["--cce-long-call=true"]155 compile_cmd += ["--cce-long-call=true"]
109 compile_cmd += ["-DASCENDC_DUMP=1"]156 compile_cmd += ["-DASCENDC_DUMP=1"]
110 compile_cmd += ["-DASCENDC_DEBUG"]157 compile_cmd += ["-DASCENDC_DEBUG"]
@@ -112,29 +159,49 @@ def gen_compile_cmd_v300(args, dst: str, sub_arch: str, compile_options):
112 159 
113 160 
114def gen_compile_cmd_c310(args, dst: str, sub_arch: str, compile_options):161def gen_compile_cmd_c310(args, dst: str, sub_arch: str, compile_options):
115- compile_cmd = [os.path.join(args.code_root, '3rdparty', 'compiler',162+ compile_cmd = [
116- 'ccec_compiler', 'bin', 'bisheng'),163+ os.path.join(
117- '-c']164+ args.code_root, "3rdparty", "compiler", "ccec_compiler", "bin", "bisheng"
165+ ),
166+ "-c",
167+ ]
118 compile_cmd += compile_options168 compile_cmd += compile_options
119- compile_cmd += [args.srcs, "--cce-aicore-arch=%s" % sub_arch,169+ compile_cmd += [
120- "--cce-aicore-only", "-o", dst,170+ args.srcs,
121- "-mllvm", "-cce-aicore-stack-size=0x8000",171+ "--cce-aicore-arch=%s" % sub_arch,
122- "-mllvm", "-cce-aicore-function-stack-size=0x8000",172+ "--cce-aicore-only",
123- "-mllvm", "-cce-aicore-record-overflow=true",173+ "-o",
124- "-mllvm", "-cce-aicore-addr-transform",174+ dst,
125- "-mllvm", "-cce-aicore-jump-expand=true",175+ "-mllvm",
126- "-mllvm", "-cce-aicore-dcci-insert-for-scalar=false",176+ "-cce-aicore-stack-size=0x8000",
127- "-mllvm", "-cce-aicore-dcci-before-kernel-end=false"]177+ "-mllvm",
178+ "-cce-aicore-function-stack-size=0x8000",
179+ "-mllvm",
180+ "-cce-aicore-record-overflow=true",
181+ "-mllvm",
182+ "-cce-aicore-addr-transform",
183+ "-mllvm",
184+ "-cce-aicore-jump-expand=true",
185+ "-mllvm",
186+ "-cce-aicore-dcci-insert-for-scalar=false",
187+ "-mllvm",
188+ "-cce-aicore-dcci-before-kernel-end=false",
189+ ]
128 compile_cmd += ["-std=c++17"]190 compile_cmd += ["-std=c++17"]
129 return compile_cmd191 return compile_cmd
130 192 
131 193 
132def gen_fatbin_cmd(args, obj_file: list, dst_file: str):194def gen_fatbin_cmd(args, obj_file: list, dst_file: str):
133- compile_cmd = [os.path.join(args.code_root, '3rdparty', 'compiler',195+ compile_cmd = [
134- 'ccec_compiler', 'bin', 'ld.lld'),196+ os.path.join(
135- '-m', 'aicorelinux', '-Ttext=0']197+ args.code_root, "3rdparty", "compiler", "ccec_compiler", "bin", "ld.lld"
198+ ),
199+ "-m",
200+ "aicorelinux",
201+ "-Ttext=0",
202+ ]
136 compile_cmd += obj_file203 compile_cmd += obj_file
137- compile_cmd += ['-static', '-o', "%s" % dst_file]204+ compile_cmd += ["-static", "-o", "%s" % dst_file]
138 return compile_cmd205 return compile_cmd
139 206 
140 207 
@@ -152,7 +219,7 @@ def gen_json(args, kernels):
152 "parameters": [],219 "parameters": [],
153 "sha256": "",220 "sha256": "",
154 "kernelList": [],221 "kernelList": [],
155- "compileInfo": {}222+ "compileInfo": {},
156 }223 }
157 json_template["binFileName"] = args.kernel224 json_template["binFileName"] = args.kernel
158 for kernel in kernels:225 for kernel in kernels:
@@ -174,14 +241,20 @@ def gen_json(args, kernels):
174 json_template["core_type"] = "MIX"241 json_template["core_type"] = "MIX"
175 json_template["magic"] = "RT_DEV_BINARY_MAGIC_ELF"242 json_template["magic"] = "RT_DEV_BINARY_MAGIC_ELF"
176 243 
177- with os.fdopen(os.open(os.path.splitext(args.dst)[0] + ".json",244+ with os.fdopen(
178- os.O_TRUNC | os.O_WRONLY | os.O_CREAT, stat.S_IWUSR | stat.S_IRUSR), 'w') as f:245+ os.open(
246+ os.path.splitext(args.dst)[0] + ".json",
247+ os.O_TRUNC | os.O_WRONLY | os.O_CREAT,
248+ stat.S_IWUSR | stat.S_IRUSR,
249+ ),
250+ "w",
251+ ) as f:
179 json.dump(json_template, f, indent=4)252 json.dump(json_template, f, indent=4)
180 253 
181 254 
182def get_common_options(args):255def get_common_options(args):
183 tikcpp_path = os.path.join(args.code_root, "3rdparty", "compiler", "tikcpp")256 tikcpp_path = os.path.join(args.code_root, "3rdparty", "compiler", "tikcpp")
184- options = ['-x', 'cce']257+ options = ["-x", "cce"]
185 options.append("-I.")258 options.append("-I.")
186 options.append("-I" + tikcpp_path)259 options.append("-I" + tikcpp_path)
187 options.append("-I" + os.path.join(tikcpp_path, "tikcfw"))260 options.append("-I" + os.path.join(tikcpp_path, "tikcfw"))
@@ -212,9 +285,9 @@ def get_arch(soc, channel):
212 arch_dict = {285 arch_dict = {
213 "ascend310b": {"vector": "dav-m300", "cube": "dav-m300"},286 "ascend310b": {"vector": "dav-m300", "cube": "dav-m300"},
214 "ascend310p": {"vector": "dav-m200", "cube": "dav-m200"},287 "ascend310p": {"vector": "dav-m200", "cube": "dav-m200"},
215- "ascend910": {"vector": "dav-c100", "cube": "dav-c100"},288+ "ascend910": {"vector": "dav-c100", "cube": "dav-c100"},
216 "ascend910b": {"vector": "dav-c220-vec", "cube": "dav-c220-cube", "mix": "mix"},289 "ascend910b": {"vector": "dav-c220-vec", "cube": "dav-c220-cube", "mix": "mix"},
217- "ascend950": {"vector": "dav-c310", "cube": "dav-c310", "mix": "mix"}290+ "ascend950": {"vector": "dav-c310", "cube": "dav-c310", "mix": "mix"},
218 }291 }
219 try:292 try:
220 return arch_dict[soc][channel]293 return arch_dict[soc][channel]
@@ -248,86 +321,136 @@ def compile_ascendc_operation(args):
248 for key in tiling_key_ids:321 for key in tiling_key_ids:
249 if args.soc == "ascend310p" or args.soc == "ascend910":322 if args.soc == "ascend310p" or args.soc == "ascend910":
250 dst = os.path.splitext(args.dst)[0] + f"_{key}.o"323 dst = os.path.splitext(args.dst)[0] + f"_{key}.o"
251- opt = options + [f'-D{args.kernel}={args.kernel}_{key}', f'-DTILING_KEY_VAR={key}']324+ opt = options + [
252- compile_cmd = ' '.join(gen_compile_cmd(args, dst, arch, opt))325+ f"-D{args.kernel}={args.kernel}_{key}",
253- if(exe_cmd(compile_cmd)) != 0:326+ f"-DTILING_KEY_VAR={key}",
327+ ]
328+ compile_cmd = " ".join(gen_compile_cmd(args, dst, arch, opt))
329+ if (exe_cmd(compile_cmd)) != 0:
254 return -1330 return -1
255 dsts.append(dst)331 dsts.append(dst)
256 if args.use_mssanitizer == "ON" and args.soc == "ascend310p":332 if args.use_mssanitizer == "ON" and args.soc == "ascend310p":
257 dsts.append("--dependent-libraries")333 dsts.append("--dependent-libraries")
258- dsts.append(os.path.join(mssanitizer_path, "libsanitizer_stub_dav-m200.a"))334+ dsts.append(
335+ os.path.join(mssanitizer_path, "libsanitizer_stub_dav-m200.a")
336+ )
259 elif args.soc == "ascend910b":337 elif args.soc == "ascend910b":
260 if args.channel != "mix":338 if args.channel != "mix":
261 dst = os.path.splitext(args.dst)[0] + f"_{key}.o"339 dst = os.path.splitext(args.dst)[0] + f"_{key}.o"
262- opt = options + [f'-D{args.kernel}={args.kernel}_{key}', f'-DTILING_KEY_VAR={key}']340+ opt = options + [
263- compile_cmd = ' '.join(gen_compile_cmd_v220(args, dst, arch, opt))341+ f"-D{args.kernel}={args.kernel}_{key}",
264- if(exe_cmd(compile_cmd)) != 0:342+ f"-DTILING_KEY_VAR={key}",
343+ ]
344+ compile_cmd = " ".join(gen_compile_cmd_v220(args, dst, arch, opt))
345+ if (exe_cmd(compile_cmd)) != 0:
265 return -1346 return -1
266 dsts.append(dst)347 dsts.append(dst)
267 if args.use_mssanitizer == "ON":348 if args.use_mssanitizer == "ON":
268 dsts.append("--dependent-libraries")349 dsts.append("--dependent-libraries")
269- dsts.append(os.path.join(mssanitizer_path, "libsanitizer_stub_dav-c220-cube.a"))350+ dsts.append(
270- dsts.append(os.path.join(mssanitizer_path, "libsanitizer_stub_dav-c220-vec.a"))351+ os.path.join(
352+ mssanitizer_path, "libsanitizer_stub_dav-c220-cube.a"
353+ )
354+ )
355+ dsts.append(
356+ os.path.join(
357+ mssanitizer_path, "libsanitizer_stub_dav-c220-vec.a"
358+ )
359+ )
271 else:360 else:
272 dst = os.path.splitext(args.dst)[0] + f"_mix_aic_{key}.o"361 dst = os.path.splitext(args.dst)[0] + f"_mix_aic_{key}.o"
273- aic_opt = options + [f'-D{args.kernel}={args.kernel}_{key}_mix_aic', f'-DTILING_KEY_VAR={key}']362+ aic_opt = options + [
274- compile_cmd = ' '.join(gen_compile_cmd_v220(args, dst, "dav-c220-cube", aic_opt))363+ f"-D{args.kernel}={args.kernel}_{key}_mix_aic",
275- if(exe_cmd(compile_cmd)) != 0:364+ f"-DTILING_KEY_VAR={key}",
365+ ]
366+ compile_cmd = " ".join(
367+ gen_compile_cmd_v220(args, dst, "dav-c220-cube", aic_opt)
368+ )
369+ if (exe_cmd(compile_cmd)) != 0:
276 return -1370 return -1
277 dsts.append(dst)371 dsts.append(dst)
278 if args.use_mssanitizer == "ON":372 if args.use_mssanitizer == "ON":
279 dsts.append("--dependent-libraries")373 dsts.append("--dependent-libraries")
280- dsts.append(os.path.join(mssanitizer_path, "libsanitizer_stub_dav-c220-cube.a"))374+ dsts.append(
375+ os.path.join(
376+ mssanitizer_path, "libsanitizer_stub_dav-c220-cube.a"
377+ )
378+ )
281 dst = os.path.splitext(args.dst)[0] + f"_mix_aiv_{key}.o"379 dst = os.path.splitext(args.dst)[0] + f"_mix_aiv_{key}.o"
282- aiv_opt = options + [f'-D{args.kernel}={args.kernel}_{key}_mix_aiv', f'-DTILING_KEY_VAR={key}']380+ aiv_opt = options + [
283- compile_cmd = ' '.join(gen_compile_cmd_v220(args, dst, "dav-c220-vec", aiv_opt))381+ f"-D{args.kernel}={args.kernel}_{key}_mix_aiv",
284- if(exe_cmd(compile_cmd)) != 0:382+ f"-DTILING_KEY_VAR={key}",
383+ ]
384+ compile_cmd = " ".join(
385+ gen_compile_cmd_v220(args, dst, "dav-c220-vec", aiv_opt)
386+ )
387+ if (exe_cmd(compile_cmd)) != 0:
285 return -1388 return -1
286 dsts.append(dst)389 dsts.append(dst)
287 if args.use_mssanitizer == "ON":390 if args.use_mssanitizer == "ON":
288 dsts.append("--dependent-libraries")391 dsts.append("--dependent-libraries")
289- dsts.append(os.path.join(mssanitizer_path, "libsanitizer_stub_dav-c220-vec.a"))392+ dsts.append(
393+ os.path.join(
394+ mssanitizer_path, "libsanitizer_stub_dav-c220-vec.a"
395+ )
396+ )
290 elif args.soc == "ascend310b":397 elif args.soc == "ascend310b":
291 dst = os.path.splitext(args.dst)[0] + f"_{key}.o"398 dst = os.path.splitext(args.dst)[0] + f"_{key}.o"
292- opt = options + [f'-D{args.kernel}={args.kernel}_{key}', f'-DTILING_KEY_VAR={key}']399+ opt = options + [
293- compile_cmd = ' '.join(gen_compile_cmd_v300(args, dst, arch, opt))400+ f"-D{args.kernel}={args.kernel}_{key}",
294- if(exe_cmd(compile_cmd)) != 0:401+ f"-DTILING_KEY_VAR={key}",
402+ ]
403+ compile_cmd = " ".join(gen_compile_cmd_v300(args, dst, arch, opt))
404+ if (exe_cmd(compile_cmd)) != 0:
295 return -1405 return -1
296 dsts.append(dst)406 dsts.append(dst)
297 elif args.soc == "ascend950":407 elif args.soc == "ascend950":
298 if args.channel != "mix":408 if args.channel != "mix":
299 dst = os.path.splitext(args.dst)[0] + f"_{key}.o"409 dst = os.path.splitext(args.dst)[0] + f"_{key}.o"
300- opt = options + [f'-D{args.kernel}={args.kernel}_{key}', f'-DTILING_KEY_VAR={key}']410+ opt = options + [
301- compile_cmd = ' '.join(gen_compile_cmd_c310(args, dst, arch, opt))411+ f"-D{args.kernel}={args.kernel}_{key}",
302- if(exe_cmd(compile_cmd)) != 0:412+ f"-DTILING_KEY_VAR={key}",
413+ ]
414+ compile_cmd = " ".join(gen_compile_cmd_c310(args, dst, arch, opt))
415+ if (exe_cmd(compile_cmd)) != 0:
303 return -1416 return -1
304 dsts.append(dst)417 dsts.append(dst)
305 else:418 else:
306 dst = os.path.splitext(args.dst)[0] + f"_mix_aic_{key}.o"419 dst = os.path.splitext(args.dst)[0] + f"_mix_aic_{key}.o"
307- aic_opt = options + [f'-D{args.kernel}={args.kernel}_{key}_mix_aic', f'-DTILING_KEY_VAR={key}']420+ aic_opt = options + [
308- compile_cmd = ' '.join(gen_compile_cmd_c310(args, dst, "dav-c310", aic_opt))421+ f"-D{args.kernel}={args.kernel}_{key}_mix_aic",
309- if(exe_cmd(compile_cmd)) != 0:422+ f"-DTILING_KEY_VAR={key}",
423+ ]
424+ compile_cmd = " ".join(
425+ gen_compile_cmd_c310(args, dst, "dav-c310", aic_opt)
426+ )
427+ if (exe_cmd(compile_cmd)) != 0:
310 return -1428 return -1
311 dsts.append(dst)429 dsts.append(dst)
312 dst = os.path.splitext(args.dst)[0] + f"_mix_aiv_{key}.o"430 dst = os.path.splitext(args.dst)[0] + f"_mix_aiv_{key}.o"
313- aiv_opt = options + [f'-D{args.kernel}={args.kernel}_{key}_mix_aiv', f'-DTILING_KEY_VAR={key}']431+ aiv_opt = options + [
314- compile_cmd = ' '.join(gen_compile_cmd_c310(args, dst, "dav-c310", aiv_opt))432+ f"-D{args.kernel}={args.kernel}_{key}_mix_aiv",
315- if(exe_cmd(compile_cmd)) != 0:433+ f"-DTILING_KEY_VAR={key}",
434+ ]
435+ compile_cmd = " ".join(
436+ gen_compile_cmd_c310(args, dst, "dav-c310", aiv_opt)
437+ )
438+ if (exe_cmd(compile_cmd)) != 0:
316 return -1439 return -1
317 dsts.append(dst)440 dsts.append(dst)
318 else:441 else:
319 logging.error("soc version %s is not supported", args.soc)442 logging.error("soc version %s is not supported", args.soc)
320- exit(1)443+ sys.exit(1)
321- kernels.append(f'{args.kernel}_{key}')444+ kernels.append(f"{args.kernel}_{key}")
322 445 
323- link_cmd = ' '.join(gen_fatbin_cmd(args, dsts, args.dst))446+ link_cmd = " ".join(gen_fatbin_cmd(args, dsts, args.dst))
324- if(exe_cmd(link_cmd)) != 0:447+ if (exe_cmd(link_cmd)) != 0:
325 return -1448 return -1
326 gen_json(args, kernels)449 gen_json(args, kernels)
327 return 0450 return 0
328 451 
329 452 
330-if __name__ == '__main__':453+if __name__ == "__main__":
331 input_args = parse_args()454 input_args = parse_args()
332 output_dir = os.path.dirname(input_args.dst)455 output_dir = os.path.dirname(input_args.dst)
333 if not os.path.exists(output_dir):456 if not os.path.exists(output_dir):
@@ -335,4 +458,4 @@ if __name__ == '__main__':
335 res = compile_ascendc_operation(input_args)458 res = compile_ascendc_operation(input_args)
336 if res != 0:459 if res != 0:
337 logging.error("compile ascend C failed!")460 logging.error("compile ascend C failed!")
338- exit(1)461+ sys.exit(1)
Mscripts/install.sh+19-8
@@ -61,7 +61,7 @@ function log() {
61 if [ x"$log_file" = x ]; then61 if [ x"$log_file" = x ]; then
62 echo -e "[cann-asdsip] [$(date +%Y%m%d-%H:%M:%S)] [$1] $2"62 echo -e "[cann-asdsip] [$(date +%Y%m%d-%H:%M:%S)] [$1] $2"
63 else63 else
64- if [ $(stat -c %s $log_file) -gt $MAX_LOG_SIZE ];then 64+ if [ $(stat -c %s $log_file) -gt $MAX_LOG_SIZE ];then
65 echo -e "[cann-asdsip] [$(date +%Y%m%d-%H:%M:%S)] [$1] log file is bigger than $MAX_LOG_SIZE, stop write log to file"65 echo -e "[cann-asdsip] [$(date +%Y%m%d-%H:%M:%S)] [$1] log file is bigger than $MAX_LOG_SIZE, stop write log to file"
66 else66 else
67 echo -e "[cann-asdsip] [$(date +%Y%m%d-%H:%M:%S)] [$1] $2" >>$log_file67 echo -e "[cann-asdsip] [$(date +%Y%m%d-%H:%M:%S)] [$1] $2" >>$log_file
@@ -80,7 +80,7 @@ function print() {
80 if [ x"$log_file" = x ]; then80 if [ x"$log_file" = x ]; then
81 echo -e "[cann-asdsip] [$(date +%Y%m%d-%H:%M:%S)] [$1] $2"81 echo -e "[cann-asdsip] [$(date +%Y%m%d-%H:%M:%S)] [$1] $2"
82 else82 else
83- if [ $(stat -c %s $log_file) -gt $MAX_LOG_SIZE ];then 83+ if [ $(stat -c %s $log_file) -gt $MAX_LOG_SIZE ];then
84 echo -e "[cann-asdsip] [$(date +%Y%m%d-%H:%M:%S)] [$1] log file is bigger than $MAX_LOG_SIZE, stop write log to file"84 echo -e "[cann-asdsip] [$(date +%Y%m%d-%H:%M:%S)] [$1] log file is bigger than $MAX_LOG_SIZE, stop write log to file"
85 echo -e "[cann-asdsip] [$(date +%Y%m%d-%H:%M:%S)] [$1] $2"85 echo -e "[cann-asdsip] [$(date +%Y%m%d-%H:%M:%S)] [$1] $2"
86 else86 else
@@ -141,7 +141,7 @@ function chmod_authority() {
141 # 修改文件和目录权限141 # 修改文件和目录权限
142 chmod_file ${default_install_path}142 chmod_file ${default_install_path}
143 chmod_file ${install_dir}143 chmod_file ${install_dir}
144- if [ "${install_for_all_flag}" == "y" ]; then 144+ if [ "${install_for_all_flag}" == "y" ]; then
145 chmod 444 ${install_dir}/scripts/filelist.csv145 chmod 444 ${install_dir}/scripts/filelist.csv
146 else146 else
147 chmod 440 ${install_dir}/scripts/filelist.csv147 chmod 440 ${install_dir}/scripts/filelist.csv
@@ -159,7 +159,7 @@ function chmod_authority() {
159 chmod ${path_rights} "${install_dir}"159 chmod ${path_rights} "${install_dir}"
160}160}
161 161 
162-function chmod_file() { 162+function chmod_file() {
163 chmod_recursion ${1} "550" "file" "*.sh"163 chmod_recursion ${1} "550" "file" "*.sh"
164 chmod_recursion ${1} "440" "file" "*.bin"164 chmod_recursion ${1} "440" "file" "*.bin"
165 chmod_recursion ${1} "440" "file" "*.h"165 chmod_recursion ${1} "440" "file" "*.h"
@@ -171,7 +171,7 @@ function chmod_file() {
171 171 
172function chmod_dir() {172function chmod_dir() {
173 chmod_recursion ${1} ${2} "dir"173 chmod_recursion ${1} ${2} "dir"
174-} 174+}
175 175 
176function chmod_recursion() {176function chmod_recursion() {
177 # install-for-all 实际上是给other组用户赋予了和同组用户相同的权限177 # install-for-all 实际上是给other组用户赋予了和同组用户相同的权限
@@ -424,7 +424,7 @@ function install_process() {
424 fi424 fi
425 if [ -n "${ARCH}" ]; then425 if [ -n "${ARCH}" ]; then
426 if [ "${arch_pkg}" != "${ARCH}" ]; then426 if [ "${arch_pkg}" != "${ARCH}" ]; then
427- print "ERROR" "Install failed, pkg arch ${arch_pkg} is not consistent with the current enviroment architecture ${ARCH}."427+ print "ERROR" "Install failed, pkg arch ${arch_pkg} is not consistent with the current environment architecture ${ARCH}."
428 exit 1428 exit 1
429 fi429 fi
430 fi430 fi
@@ -474,9 +474,20 @@ function check_owner() {
474function uninstall() {474function uninstall() {
475 # 使用awk读取version.info文件中的version键对应的值475 # 使用awk读取version.info文件中的version键对应的值
476 VERSION_INFO_PATH=${default_install_path}/latest/version.info476 VERSION_INFO_PATH=${default_install_path}/latest/version.info
477+ # 容错: version.info 缺失(latest软链损坏/版本目录被删)时给出明确报错, 避免 grep 报错中断
478+ if [ ! -f "$VERSION_INFO_PATH" ]; then
479+ print "ERROR" "Uninstall failed, can not find ${VERSION_INFO_PATH}, the installation may be broken. Please reinstall or remove ${default_install_path} manually."
480+ exit 1
481+ fi
477 old_version=$(grep -E 'Ascend-cann-asdsip :' $VERSION_INFO_PATH | cut -d ':' -f2)482 old_version=$(grep -E 'Ascend-cann-asdsip :' $VERSION_INFO_PATH | cut -d ':' -f2)
478 old_version=$(echo "$old_version" | sed 's/^[ ]*//')483 old_version=$(echo "$old_version" | sed 's/^[ ]*//')
479 484 
485+ # 容错: 提取到的版本号为空时直接报错, 避免拼接出带尾斜杠的无效路径
486+ if [ -z "$old_version" ]; then
487+ print "ERROR" "Uninstall failed, can not parse version from ${VERSION_INFO_PATH}. Please remove ${default_install_path} manually."
488+ exit 1
489+ fi
490+ 
480 # 输出version值491 # 输出version值
481 echo "Version is: $old_version"492 echo "Version is: $old_version"
482 493 
@@ -494,7 +505,7 @@ function check_uninstall_path() {
494 default_install_path="${HOME}/Ascend/asdsip"505 default_install_path="${HOME}/Ascend/asdsip"
495 fi506 fi
496 fi507 fi
497- 508+ 
498 if [ ! -d "${default_install_path}" ]; then509 if [ ! -d "${default_install_path}" ]; then
499 print "ERROR" "Uninstall failed, can not find the path of Ascend-cann-asdsip."510 print "ERROR" "Uninstall failed, can not find the path of Ascend-cann-asdsip."
500 exit 1511 exit 1
@@ -583,7 +594,7 @@ function upgrade() {
583 remove_back_up_version594 remove_back_up_version
584 print "INFO" "Ascend-cann-asdsip upgrade success!"595 print "INFO" "Ascend-cann-asdsip upgrade success!"
585}596}
586- 597+ 
587function main() {598function main() {
588 parse_script_args $*599 parse_script_args $*
589 if [ "${uninstall_flag}" == "y" ]; then600 if [ "${uninstall_flag}" == "y" ]; then
Mscripts/update_tbe_tactic_json.py+102-56
@@ -1,5 +1,5 @@
1# -*- coding: UTF-8 -*-1# -*- coding: UTF-8 -*-
2-# Copyright (c) 2025 Huawei Technologies Co., Ltd.2+# Copyright (c) 2026 Huawei Technologies Co., Ltd.
3# This program is free software, you can redistribute it and/or modify it under the terms and conditions of3# This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4# CANN Open Software License Agreement Version 2.0 (the "License").4# CANN Open Software License Agreement Version 2.0 (the "License").
5# Please refer to the License for details. You may not use this file except in compliance with the License.5# Please refer to the License for details. You may not use this file except in compliance with the License.
@@ -7,25 +7,40 @@
7# INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7# INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9 9 
10-import os
11import configparser10import configparser
12import json11import json
13import logging12import logging
13+import os
14import stat14import stat
15import sys15import sys
16from collections import namedtuple16from collections import namedtuple
17+ 
17from build_util import get_build_target_list18from build_util import get_build_target_list
18 19 
19- 20+logging.basicConfig(
20-logging.basicConfig(level=logging.INFO,21+ level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s"
21- format='%(asctime)s - %(levelname)s - %(message)s')22+)
22 23 
23JsonSpecification = namedtuple(24JsonSpecification = namedtuple(
24- "JsonSpecification", ["mode", "inputs", "outputs", "attrs", "dir", "deterministic"])25+ "JsonSpecification", ["mode", "inputs", "outputs", "attrs", "dir", "deterministic"]
26+)
25 27 
26-TacticDef = namedtuple("TacticDef", [28+TacticDef = namedtuple(
27- "ops_name", "operation", "input_num", "output_num", "dtypes_in", "dtypes_out", "formats_in",29+ "TacticDef",
28- "formats_out", "mode", "attrs", "soc_support"])30+ [
31+ "ops_name",
32+ "operation",
33+ "input_num",
34+ "output_num",
35+ "dtypes_in",
36+ "dtypes_out",
37+ "formats_in",
38+ "formats_out",
39+ "mode",
40+ "attrs",
41+ "soc_support",
42+ ],
43+)
29 44 
30 45 
31TARGET_INI = "configs/tbe_tactic_json.ini"46TARGET_INI = "configs/tbe_tactic_json.ini"
@@ -81,9 +96,17 @@ def read_tbe_json_file(json_file_path):
81 outputs = item["outputs"]96 outputs = item["outputs"]
82 mode = item["implMode"] if "implMode" in item else None97 mode = item["implMode"] if "implMode" in item else None
83 attrs = item["attrs"] if "attrs" in item else None98 attrs = item["attrs"] if "attrs" in item else None
84- deterministic = item["deterministic"] if "deterministic" in item else None99+ deterministic = (
100+ item["deterministic"] if "deterministic" in item else None
101+ )
85 json_info = JsonSpecification(102 json_info = JsonSpecification(
86- mode=mode, inputs=inputs, outputs=outputs, attrs=attrs, dir=file_name, deterministic=deterministic)103+ mode=mode,
104+ inputs=inputs,
105+ outputs=outputs,
106+ attrs=attrs,
107+ dir=file_name,
108+ deterministic=deterministic,
109+ )
87 ops_specification_list.append(json_info)110 ops_specification_list.append(json_info)
88 except FileNotFoundError:111 except FileNotFoundError:
89 logging.error("file %s is not found!", json_file)112 logging.error("file %s is not found!", json_file)
@@ -170,13 +193,23 @@ def get_match_json(json_info_dir, tactic_info):
170 return match_json_dir, result193 return match_json_dir, result
171 count_check = 0194 count_check = 0
172 for json_info in ops_specification_list:195 for json_info in ops_specification_list:
173- matched = impl_mode_matched_or_not(json_info.mode, tactic_info.mode) \196+ matched = (
197+ impl_mode_matched_or_not(json_info.mode, tactic_info.mode)
174 and inputs_outputs_matched_or_not(198 and inputs_outputs_matched_or_not(
175- json_info.inputs, tactic_info.input_num, tactic_info.dtypes_in, tactic_info.formats_in) \199+ json_info.inputs,
200+ tactic_info.input_num,
201+ tactic_info.dtypes_in,
202+ tactic_info.formats_in,
203+ )
176 and inputs_outputs_matched_or_not(204 and inputs_outputs_matched_or_not(
177- json_info.outputs, tactic_info.output_num, tactic_info.dtypes_out, tactic_info.formats_out) \205+ json_info.outputs,
178- and attrs_matched_or_not(json_info.attrs, tactic_info.attrs) \206+ tactic_info.output_num,
207+ tactic_info.dtypes_out,
208+ tactic_info.formats_out,
209+ )
210+ and attrs_matched_or_not(json_info.attrs, tactic_info.attrs)
179 and deterministic_matched_or_not(json_info.deterministic)211 and deterministic_matched_or_not(json_info.deterministic)
212+ )
180 213 
181 if matched:214 if matched:
182 match_json_dir, result = json_info.dir, True215 match_json_dir, result = json_info.dir, True
@@ -184,7 +217,8 @@ def get_match_json(json_info_dir, tactic_info):
184 217 
185 if count_check != 1:218 if count_check != 1:
186 logging.error(219 logging.error(
187- f"{json_info_dir}: matched json file number is {count_check}, which should be 1")220+ f"{json_info_dir}: matched json file number is {count_check}, which should be 1"
221+ )
188 result = False222 result = False
189 return match_json_dir, result223 return match_json_dir, result
190 224 
@@ -211,30 +245,28 @@ def get_tbe_tactic_json(tbe_config_ini):
211 target_version_list = get_build_target_list()245 target_version_list = get_build_target_list()
212 logging.info("target version list: %s", target_version_list)246 logging.info("target version list: %s", target_version_list)
213 for target_version in target_version_list:247 for target_version in target_version_list:
214- if target_version == 'ascend310b':248+ if target_version == "ascend310b":
215 logging.warning("ascend310b opp kernel is not ready")249 logging.warning("ascend310b opp kernel is not ready")
216 try:250 try:
217 for tactic_name in tbe_config_ini.sections():251 for tactic_name in tbe_config_ini.sections():
218 try:252 try:
219 ops = tbe_config_ini.get(tactic_name, "ops")253 ops = tbe_config_ini.get(tactic_name, "ops")
220- operation_name = tbe_config_ini.get(254+ operation_name = tbe_config_ini.get(tactic_name, "operationName")
221- tactic_name, "operationName")255+ input_num = int(tbe_config_ini.get(tactic_name, "inputCount"))
222- input_num = int(tbe_config_ini.get(256+ output_num = int(tbe_config_ini.get(tactic_name, "outputCount"))
223- tactic_name, "inputCount"))
224- output_num = int(tbe_config_ini.get(
225- tactic_name, "outputCount"))
226 input_dtypes = tbe_config_ini.get(tactic_name, "dtypeIn")257 input_dtypes = tbe_config_ini.get(tactic_name, "dtypeIn")
227 output_dtypes = tbe_config_ini.get(tactic_name, "dtypeOut")258 output_dtypes = tbe_config_ini.get(tactic_name, "dtypeOut")
228 input_formats = tbe_config_ini.get(259 input_formats = tbe_config_ini.get(
229- tactic_name, "formatIn", fallback=None)260+ tactic_name, "formatIn", fallback=None
261+ )
230 output_formats = tbe_config_ini.get(262 output_formats = tbe_config_ini.get(
231- tactic_name, "formatOut", fallback=None)263+ tactic_name, "formatOut", fallback=None
232- mode = tbe_config_ini.get(264+ )
233- tactic_name, "mode", fallback=None)265+ mode = tbe_config_ini.get(tactic_name, "mode", fallback=None)
234- attrs = tbe_config_ini.get(266+ attrs = tbe_config_ini.get(tactic_name, "attrs", fallback=None)
235- tactic_name, "attrs", fallback=None)
236 soc_support = tbe_config_ini.get(267 soc_support = tbe_config_ini.get(
237- tactic_name, "socSupport", fallback=None)268+ tactic_name, "socSupport", fallback=None
269+ )
238 except configparser.NoOptionError:270 except configparser.NoOptionError:
239 logging.error("configparser option is not found: %s", tactic_name)271 logging.error("configparser option is not found: %s", tactic_name)
240 continue272 continue
@@ -252,32 +284,44 @@ def get_tbe_tactic_json(tbe_config_ini):
252 284 
253 input_dtype_arr = input_dtypes.split(",")285 input_dtype_arr = input_dtypes.split(",")
254 output_dtype_arr = output_dtypes.split(",")286 output_dtype_arr = output_dtypes.split(",")
255- input_format_arr = input_formats.split(287+ input_format_arr = input_formats.split(",") if input_formats else None
256- ",") if input_formats else None288+ output_format_arr = (
257- output_format_arr = output_formats.split(289+ output_formats.split(",") if output_formats else None
258- ",") if output_formats else None290+ )
259- attr_arr = attrs.split(',') if attrs else None291+ attr_arr = attrs.split(",") if attrs else None
260 292 
261- tactic_info = TacticDef(ops_name=ops, operation=operation_name,293+ tactic_info = TacticDef(
262- input_num=input_num, output_num=output_num,294+ ops_name=ops,
263- dtypes_in=input_dtype_arr, dtypes_out=output_dtype_arr,295+ operation=operation_name,
264- formats_in=input_format_arr, formats_out=output_format_arr,296+ input_num=input_num,
265- mode=mode, attrs=attr_arr, soc_support=soc_support)297+ output_num=output_num,
266- if tactic_info.soc_support and target_version not in tactic_info.soc_support:298+ dtypes_in=input_dtype_arr,
299+ dtypes_out=output_dtype_arr,
300+ formats_in=input_format_arr,
301+ formats_out=output_format_arr,
302+ mode=mode,
303+ attrs=attr_arr,
304+ soc_support=soc_support,
305+ )
306+ if (
307+ tactic_info.soc_support
308+ and target_version not in tactic_info.soc_support
309+ ):
267 continue310 continue
268- json_info_dir = os.path.join(311+ json_info_dir = os.path.join(tbe_kernel_path, target_version, ops)
269- tbe_kernel_path, target_version, ops)312+ match_json_dir, ret = get_match_json(json_info_dir, tactic_info)
270- match_json_dir, ret = get_match_json(
271- json_info_dir, tactic_info)
272 if not ret:313 if not ret:
273 logging.error(314 logging.error(
274- f"[{target_version}] get tactic failed: {tactic_name}")315+ f"[{target_version}] get tactic failed: {tactic_name}"
275- exit(1)316+ )
317+ sys.exit(1)
276 if not json_paths_info.has_section(operation_name):318 if not json_paths_info.has_section(operation_name):
277 json_paths_info.add_section(operation_name)319 json_paths_info.add_section(operation_name)
278 json_paths_info.set(320 json_paths_info.set(
279- operation_name, tactic_name + "." + target_version,321+ operation_name,
280- os.path.join(ops, match_json_dir))322+ tactic_name + "." + target_version,
323+ os.path.join(ops, match_json_dir),
324+ )
281 325 
282 except configparser.NoSectionError:326 except configparser.NoSectionError:
283 result = False327 result = False
@@ -290,11 +334,13 @@ def get_tbe_tactic_json(tbe_config_ini):
290 334 
291def write_tbe_tactic_json(json_paths_info):335def write_tbe_tactic_json(json_paths_info):
292 code_root_dir = get_code_root()336 code_root_dir = get_code_root()
293- tbe_tactic_json_path = os.path.join(337+ tbe_tactic_json_path = os.path.join(code_root_dir, "configs/tbe_tactic_json.ini")
294- code_root_dir, "configs/tbe_tactic_json.ini")338+ fd = os.open(
295- fd = os.open(tbe_tactic_json_path, os.O_WRONLY | os.O_CREAT |339+ tbe_tactic_json_path,
296- os.O_TRUNC, stat.S_IWUSR | stat.S_IRUSR)340+ os.O_WRONLY | os.O_CREAT | os.O_TRUNC,
297- with os.fdopen(fd, 'w+') as f:341+ stat.S_IWUSR | stat.S_IRUSR,
342+ )
343+ with os.fdopen(fd, "w+") as f:
298 try:344 try:
299 json_paths_info.write(f, space_around_delimiters=False)345 json_paths_info.write(f, space_around_delimiters=False)
300 except configparser.Error as e:346 except configparser.Error as e:
@@ -307,11 +353,11 @@ def main():
307 tbe_config_ini, ret = read_tbe_config_file()353 tbe_config_ini, ret = read_tbe_config_file()
308 if not ret:354 if not ret:
309 logging.error("get tbe tactic info failed!")355 logging.error("get tbe tactic info failed!")
310- exit(1)356+ sys.exit(1)
311 json_paths_info, ret = get_tbe_tactic_json(tbe_config_ini)357 json_paths_info, ret = get_tbe_tactic_json(tbe_config_ini)
312 if not ret:358 if not ret:
313 logging.error("get tbe tactic json failed!")359 logging.error("get tbe tactic json failed!")
314- exit(1)360+ sys.exit(1)
315 write_tbe_tactic_json(json_paths_info)361 write_tbe_tactic_json(json_paths_info)
316 362 
317 363