Star1730
1645
代码
代码
Issues491
Pull Requests549
Actions
项目讨论
Wiki
分析
Star1730
1645
已合并
fix(pow/sub): A2/A3补充out dtype白名单校验（#3964） #6800
baijinxiang_hw创建于 10 天前
fix(pow/sub): A2/A3补充out dtype白名单校验（#3964） #6800
已合并
baijinxiang_hw创建于 10 天前
从doufloat/ops-mathfix/pow-out-dtype-whitelist合入到cann/ops-mathmaster
共 7 个文件变更+172-2
Mmath/pow/op_api/aclnn_pow.cpp+12-0
@@ -49,6 +49,13 @@ static const std::initializer_list<op::DataType> DTYPE_SUPPORT_LIST = {
49 op::DataType::DT_INT8, op::DataType::DT_UINT8, op::DataType::DT_DOUBLE, op::DataType::DT_BOOL,49 op::DataType::DT_INT8, op::DataType::DT_UINT8, op::DataType::DT_DOUBLE, op::DataType::DT_BOOL,
50 op::DataType::DT_INT16, op::DataType::DT_COMPLEX64, op::DataType::DT_COMPLEX128, op::DataType::DT_BF16};50 op::DataType::DT_INT16, op::DataType::DT_COMPLEX64, op::DataType::DT_COMPLEX128, op::DataType::DT_BF16};
51 51 
52+// out的数据类型支持列表（比输入列表多UINT16/UINT32/UINT64，与API资料一致）
53+static const std::initializer_list<op::DataType> OUT_DTYPE_SUPPORT_LIST = {
54+ op::DataType::DT_FLOAT, op::DataType::DT_INT32, op::DataType::DT_INT64, op::DataType::DT_FLOAT16,
55+ op::DataType::DT_INT8, op::DataType::DT_UINT8, op::DataType::DT_DOUBLE, op::DataType::DT_INT16,
56+ op::DataType::DT_COMPLEX64, op::DataType::DT_COMPLEX128, op::DataType::DT_BF16, op::DataType::DT_UINT16,
57+ op::DataType::DT_UINT32, op::DataType::DT_UINT64};
58+ 
52static const std::initializer_list<op::DataType> SQUARE_NEED_CAST_DTYPE_LIST = {59static const std::initializer_list<op::DataType> SQUARE_NEED_CAST_DTYPE_LIST = {
53 op::DataType::DT_INT8, op::DataType::DT_UINT8, op::DataType::DT_BOOL, op::DataType::DT_INT16};60 op::DataType::DT_INT8, op::DataType::DT_UINT8, op::DataType::DT_BOOL, op::DataType::DT_INT16};
54 61 
@@ -184,6 +191,11 @@ static bool CheckDtypeValid(const op::DataType selfDtype, const op::DataType exp
184 op::ToString(selfDtype).GetString(), op::ToString(expDtype).GetString());191 op::ToString(selfDtype).GetString(), op::ToString(expDtype).GetString());
185 return false;192 return false;
186 }193 }
194+ if (GetCurrentPlatformInfo().GetCurNpuArch() == NpuArch::DAV_2201 && !CheckType(outDtype, OUT_DTYPE_SUPPORT_LIST)) {
195+ OP_LOGE(ACLNN_ERR_PARAM_INVALID, "out dtype %s should be in out dtype support list %s.",
196+ op::ToString(outDtype).GetString(), op::ToString(OUT_DTYPE_SUPPORT_LIST).GetString());
197+ return false;
198+ }
187 OP_CHECK_RESULT_DTYPE_CAST_FAILED(promoteType, outDtype, return false);199 OP_CHECK_RESULT_DTYPE_CAST_FAILED(promoteType, outDtype, return false);
188 return true;200 return true;
189}201}
Mmath/pow/op_api/aclnn_pow_tensor_tensor.cpp+13-0
@@ -54,6 +54,13 @@ static const std::initializer_list<op::DataType> DTYPE_SUPPORT_LIST = {
54 op::DataType::DT_INT8, op::DataType::DT_UINT8, op::DataType::DT_DOUBLE, op::DataType::DT_BOOL,54 op::DataType::DT_INT8, op::DataType::DT_UINT8, op::DataType::DT_DOUBLE, op::DataType::DT_BOOL,
55 op::DataType::DT_COMPLEX64, op::DataType::DT_COMPLEX128, op::DataType::DT_INT16, op::DataType::DT_BF16};55 op::DataType::DT_COMPLEX64, op::DataType::DT_COMPLEX128, op::DataType::DT_INT16, op::DataType::DT_BF16};
56 56 
57+// out的数据类型支持列表（比输入列表多UINT16/UINT32/UINT64，与API资料一致）
58+static const std::initializer_list<op::DataType> OUT_DTYPE_SUPPORT_LIST = {
59+ op::DataType::DT_FLOAT, op::DataType::DT_INT32, op::DataType::DT_INT64, op::DataType::DT_FLOAT16,
60+ op::DataType::DT_INT8, op::DataType::DT_UINT8, op::DataType::DT_DOUBLE, op::DataType::DT_INT16,
61+ op::DataType::DT_COMPLEX64, op::DataType::DT_COMPLEX128, op::DataType::DT_BF16, op::DataType::DT_UINT16,
62+ op::DataType::DT_UINT32, op::DataType::DT_UINT64};
63+ 
57static bool CheckNotNull(const aclTensor* self, const aclTensor* exponent, const aclTensor* out)64static bool CheckNotNull(const aclTensor* self, const aclTensor* exponent, const aclTensor* out)
58{65{
59 OP_CHECK_NULL(self, return false);66 OP_CHECK_NULL(self, return false);
@@ -132,6 +139,12 @@ static aclnnStatus CheckParams(const aclTensor* self, const aclTensor* exponent,
132 139 
133 // 2. 检查输入的数据类型是否在API支持的数据类型范围之内，需要根据api定义校验140 // 2. 检查输入的数据类型是否在API支持的数据类型范围之内，需要根据api定义校验
134 CHECK_RET(CheckDtypeValid(self, exponent), ACLNN_ERR_PARAM_INVALID);141 CHECK_RET(CheckDtypeValid(self, exponent), ACLNN_ERR_PARAM_INVALID);
142+ if (GetCurrentPlatformInfo().GetCurNpuArch() == NpuArch::DAV_2201 &&
143+ !CheckType(out->GetDataType(), OUT_DTYPE_SUPPORT_LIST)) {
144+ OP_LOGE(ACLNN_ERR_PARAM_INVALID, "out dtype %s should be in out dtype support list %s.",
145+ op::ToString(out->GetDataType()).GetString(), op::ToString(OUT_DTYPE_SUPPORT_LIST).GetString());
146+ return ACLNN_ERR_PARAM_INVALID;
147+ }
135 148 
136 // 3. 检查self和exponent能否做数据类型推导以及推导的数据类型能否转换为输出数据类型149 // 3. 检查self和exponent能否做数据类型推导以及推导的数据类型能否转换为输出数据类型
137 CHECK_RET(CheckPromoteType(self, exponent, out), ACLNN_ERR_PARAM_INVALID);150 CHECK_RET(CheckPromoteType(self, exponent, out), ACLNN_ERR_PARAM_INVALID);
Mmath/pow/tests/ut/op_api/test_aclnn_pow_scalar_tensor.cpp+47-0
@@ -16,6 +16,7 @@
16#include "op_api_ut_common/op_api_ut.h"16#include "op_api_ut_common/op_api_ut.h"
17#include "op_api_ut_common/scalar_desc.h"17#include "op_api_ut_common/scalar_desc.h"
18#include "op_api_ut_common/tensor_desc.h"18#include "op_api_ut_common/tensor_desc.h"
19+#include "opdev/platform.h"
19 20 
20using namespace std;21using namespace std;
21 22 
@@ -445,3 +446,49 @@ TEST_F(l2_pow_scalar_tensor_test, Ascend950PR_89_aclnnPowScalarTensor_scalar_flo
445 EXPECT_EQ(aclRet, ACL_SUCCESS);446 EXPECT_EQ(aclRet, ACL_SUCCESS);
446}447}
447 */448 */
449+ 
450+TEST_F(l2_pow_scalar_tensor_test, aclnnPowScalarTensor_out_dtype_not_in_support_list)
451+{
452+ auto expTensorDesc = TensorDesc({4, 8}, ACL_FLOAT, ACL_FORMAT_ND).ValueRange(0, 2);
453+ auto selfScalarDesc = ScalarDesc(2);
454+ // INT4不在out支持列表中，校验报错
455+ {
456+ auto outTensorDesc = TensorDesc({4, 8}, ACL_INT4, ACL_FORMAT_ND);
457+ auto ut = OP_API_UT(aclnnPowScalarTensor, INPUT(selfScalarDesc, expTensorDesc), OUTPUT(outTensorDesc));
458+ uint64_t workspaceSize = 0;
459+ EXPECT_EQ(ut.TestGetWorkspaceSize(&workspaceSize), ACLNN_ERR_PARAM_INVALID);
460+ }
461+ // BOOL不在out支持列表中，校验报错
462+ {
463+ auto outTensorDesc = TensorDesc({4, 8}, ACL_BOOL, ACL_FORMAT_ND);
464+ auto ut = OP_API_UT(aclnnPowScalarTensor, INPUT(selfScalarDesc, expTensorDesc), OUTPUT(outTensorDesc));
465+ uint64_t workspaceSize = 0;
466+ EXPECT_EQ(ut.TestGetWorkspaceSize(&workspaceSize), ACLNN_ERR_PARAM_INVALID);
467+ }
468+ // 列表内dtype（UINT16）但推导类型不可转换，仍返回161002
469+ {
470+ auto outTensorDesc = TensorDesc({4, 8}, ACL_UINT16, ACL_FORMAT_ND);
471+ auto ut = OP_API_UT(aclnnPowScalarTensor, INPUT(selfScalarDesc, expTensorDesc), OUTPUT(outTensorDesc));
472+ uint64_t workspaceSize = 0;
473+ EXPECT_EQ(ut.TestGetWorkspaceSize(&workspaceSize), ACLNN_ERR_PARAM_INVALID);
474+ }
475+ // 合法out dtype
476+ {
477+ auto outTensorDesc = TensorDesc({4, 8}, ACL_FLOAT, ACL_FORMAT_ND).Precision(0.0001, 0.0001);
478+ auto ut = OP_API_UT(aclnnPowScalarTensor, INPUT(selfScalarDesc, expTensorDesc), OUTPUT(outTensorDesc));
479+ uint64_t workspaceSize = 0;
480+ EXPECT_EQ(ut.TestGetWorkspaceSize(&workspaceSize), ACL_SUCCESS);
481+ }
482+}
483+ 
484+TEST_F(l2_pow_scalar_tensor_test, Ascend950_out_dtype_check_not_applied)
485+{
486+ op::SetPlatformSocVersion(op::SocVersion::ASCEND950);
487+ auto expTensorDesc = TensorDesc({4, 8}, ACL_FLOAT, ACL_FORMAT_ND).ValueRange(0, 2);
488+ auto selfScalarDesc = ScalarDesc(2);
489+ auto outTensorDesc = TensorDesc({4, 8}, ACL_INT4, ACL_FORMAT_ND);
490+ auto ut = OP_API_UT(aclnnPowScalarTensor, INPUT(selfScalarDesc, expTensorDesc), OUTPUT(outTensorDesc));
491+ uint64_t workspaceSize = 0;
492+ EXPECT_NE(ut.TestGetWorkspaceSize(&workspaceSize), ACLNN_ERR_PARAM_INVALID);
493+ op::SetPlatformSocVersion(op::SocVersion::ASCEND910B);
494+}
Mmath/pow/tests/ut/op_api/test_aclnn_pow_tensor_scalar.cpp+14-2
@@ -252,7 +252,7 @@ TEST_F(l2_pow_tensor_scalar_test, aclnnPowTensorScalar_1_1_16_129_in64_hwcn_1_1_
252 252 
253TEST_F(l2_pow_tensor_scalar_test, Ascend950_aclnnPowTensorScalar_square_complex64)253TEST_F(l2_pow_tensor_scalar_test, Ascend950_aclnnPowTensorScalar_square_complex64)
254{254{
255- op::SetPlatformNpuArch(NpuArch::DAV_3510);255+ op::SetPlatformSocVersion(op::SocVersion::ASCEND950);
256 256 
257 const vector<int64_t>& selfShape = {1, 1, 16, 129};257 const vector<int64_t>& selfShape = {1, 1, 16, 129};
258 aclDataType selfDtype = ACL_COMPLEX64;258 aclDataType selfDtype = ACL_COMPLEX64;
@@ -269,7 +269,7 @@ TEST_F(l2_pow_tensor_scalar_test, Ascend950_aclnnPowTensorScalar_square_complex6
269 auto ut = OP_API_UT(aclnnPowTensorScalar, INPUT(selfTensorDesc, expScalarDesc), OUTPUT(outTensorDesc));269 auto ut = OP_API_UT(aclnnPowTensorScalar, INPUT(selfTensorDesc, expScalarDesc), OUTPUT(outTensorDesc));
270 uint64_t workspaceSize = 0;270 uint64_t workspaceSize = 0;
271 aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspaceSize);271 aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspaceSize);
272- op::SetPlatformNpuArch(NpuArch::DAV_2201);272+ op::SetPlatformSocVersion(op::SocVersion::ASCEND910B);
273 EXPECT_EQ(aclRet, ACL_SUCCESS);273 EXPECT_EQ(aclRet, ACL_SUCCESS);
274}274}
275 275 
@@ -295,6 +295,18 @@ TEST_F(l2_pow_tensor_scalar_test, Ascend950_aclnnPowTensorScalar_square_complex1
295 op::SetPlatformNpuArch(NpuArch::DAV_2201);295 op::SetPlatformNpuArch(NpuArch::DAV_2201);
296 EXPECT_EQ(aclRet, ACL_SUCCESS);296 EXPECT_EQ(aclRet, ACL_SUCCESS);
297}297}
298+ 
299+TEST_F(l2_pow_tensor_scalar_test, Ascend950_out_dtype_check_not_applied)
300+{
301+ op::SetPlatformNpuArch(NpuArch::DAV_3510);
302+ auto selfTensorDesc = TensorDesc({4, 8}, ACL_FLOAT, ACL_FORMAT_ND);
303+ auto expScalarDesc = ScalarDesc(2);
304+ auto outTensorDesc = TensorDesc({4, 8}, ACL_INT4, ACL_FORMAT_ND);
305+ auto ut = OP_API_UT(aclnnPowTensorScalar, INPUT(selfTensorDesc, expScalarDesc), OUTPUT(outTensorDesc));
306+ uint64_t workspaceSize = 0;
307+ EXPECT_NE(ut.TestGetWorkspaceSize(&workspaceSize), ACLNN_ERR_PARAM_INVALID);
308+ op::SetPlatformNpuArch(NpuArch::DAV_2201);
309+}
298/*310/*
299TEST_F(l2_pow_tensor_scalar_test, aclnnPowTensorScalar_1_1_16_129_complex64_hwcn_1_1_16_129_complex64_hwcn) {311TEST_F(l2_pow_tensor_scalar_test, aclnnPowTensorScalar_1_1_16_129_complex64_hwcn_1_1_16_129_complex64_hwcn) {
300 // input312 // input
Mmath/pow/tests/ut/op_api/test_aclnn_pow_tensor_tensor.cpp+13-0
@@ -16,6 +16,7 @@
16#include "op_api_ut_common/op_api_ut.h"16#include "op_api_ut_common/op_api_ut.h"
17#include "op_api_ut_common/scalar_desc.h"17#include "op_api_ut_common/scalar_desc.h"
18#include "op_api_ut_common/tensor_desc.h"18#include "op_api_ut_common/tensor_desc.h"
19+#include "opdev/platform.h"
19 20 
20using namespace std;21using namespace std;
21 22 
@@ -630,3 +631,15 @@ TEST_F(l2_pow_tensor_tensor_test, aclnnPowTensorTensor_2_7_bool_nd_2_7_bool_nd)
630 aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspaceSize);631 aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspaceSize);
631 EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_INVALID);632 EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_INVALID);
632}633}
634+ 
635+TEST_F(l2_pow_tensor_tensor_test, Ascend950_out_dtype_check_not_applied)
636+{
637+ op::SetPlatformSocVersion(op::SocVersion::ASCEND950);
638+ auto selfTensorDesc = TensorDesc({4, 8}, ACL_FLOAT, ACL_FORMAT_ND);
639+ auto otherTensorDesc = TensorDesc({4, 8}, ACL_FLOAT, ACL_FORMAT_ND);
640+ auto outTensorDesc = TensorDesc({4, 8}, ACL_INT4, ACL_FORMAT_ND);
641+ auto ut = OP_API_UT(aclnnPowTensorTensor, INPUT(selfTensorDesc, otherTensorDesc), OUTPUT(outTensorDesc));
642+ uint64_t workspaceSize = 0;
643+ EXPECT_NE(ut.TestGetWorkspaceSize(&workspaceSize), ACLNN_ERR_PARAM_INVALID);
644+ op::SetPlatformSocVersion(op::SocVersion::ASCEND910B);
645+}
Mmath/sub/op_api/aclnn_sub.cpp+6-0
@@ -247,6 +247,9 @@ static aclnnStatus CheckParams(const aclTensor* self, const aclTensor* other, co
247 const std::initializer_list<op::DataType> dtypeSupportList = GetDtypeSupportListBySocVersion();247 const std::initializer_list<op::DataType> dtypeSupportList = GetDtypeSupportListBySocVersion();
248 OP_CHECK_DTYPE_NOT_SUPPORT(self, dtypeSupportList, return ACLNN_ERR_PARAM_INVALID);248 OP_CHECK_DTYPE_NOT_SUPPORT(self, dtypeSupportList, return ACLNN_ERR_PARAM_INVALID);
249 OP_CHECK_DTYPE_NOT_SUPPORT(other, dtypeSupportList, return ACLNN_ERR_PARAM_INVALID);249 OP_CHECK_DTYPE_NOT_SUPPORT(other, dtypeSupportList, return ACLNN_ERR_PARAM_INVALID);
250+ if (GetCurrentPlatformInfo().GetCurNpuArch() == NpuArch::DAV_2201) {
251+ OP_CHECK_DTYPE_NOT_SUPPORT(out, dtypeSupportList, return ACLNN_ERR_PARAM_INVALID);
252+ }
250 253 
251 // 3. 检查self和other能否做数据类型推导以及推导的数据类型能否转换为输出数据类型254 // 3. 检查self和other能否做数据类型推导以及推导的数据类型能否转换为输出数据类型
252 op::DataType promoteType = op::PromoteType(self->GetDataType(), other->GetDataType());255 op::DataType promoteType = op::PromoteType(self->GetDataType(), other->GetDataType());
@@ -432,6 +435,9 @@ static aclnnStatus CheckParamsScalar(const aclTensor* self, const aclScalar* oth
432 // 2. 检查输入的数据类型是否在API支持的数据类型范围之内，需要根据api定义校验435 // 2. 检查输入的数据类型是否在API支持的数据类型范围之内，需要根据api定义校验
433 const std::initializer_list<op::DataType> dtypeSupportList = GetDtypeSupportListBySocVersion();436 const std::initializer_list<op::DataType> dtypeSupportList = GetDtypeSupportListBySocVersion();
434 OP_CHECK_DTYPE_NOT_SUPPORT(self, dtypeSupportList, return ACLNN_ERR_PARAM_INVALID);437 OP_CHECK_DTYPE_NOT_SUPPORT(self, dtypeSupportList, return ACLNN_ERR_PARAM_INVALID);
438+ if (GetCurrentPlatformInfo().GetCurNpuArch() == NpuArch::DAV_2201) {
439+ OP_CHECK_DTYPE_NOT_SUPPORT(out, dtypeSupportList, return ACLNN_ERR_PARAM_INVALID);
440+ }
435 441 
436 // 3. 检查self和other能否做数据类型推导以及推导的数据类型能否转换为输出数据类型442 // 3. 检查self和other能否做数据类型推导以及推导的数据类型能否转换为输出数据类型
437 auto promoteType = PromoteTypeScalar(self, other, alpha, out);443 auto promoteType = PromoteTypeScalar(self, other, alpha, out);
Mmath/sub/tests/ut/op_api/test_aclnn_sub.cpp+67-0
@@ -15,6 +15,7 @@
15#include "op_api_ut_common/tensor_desc.h"15#include "op_api_ut_common/tensor_desc.h"
16#include "op_api_ut_common/scalar_desc.h"16#include "op_api_ut_common/scalar_desc.h"
17#include "op_api_ut_common/op_api_ut.h"17#include "op_api_ut_common/op_api_ut.h"
18+#include "opdev/platform.h"
18 19 
19class l2_sub_test : public testing::Test {20class l2_sub_test : public testing::Test {
20protected:21protected:
@@ -529,3 +530,69 @@ TEST_F(l2_sub_test, Ascend950PR_89_case_005)
529 // SAMPLE: precision simulate530 // SAMPLE: precision simulate
530 // ut.TestPrecision();531 // ut.TestPrecision();
531}532}
533+ 
534+TEST_F(l2_sub_test, abnormal_out_dtype_not_in_support_list)
535+{
536+ // INT4不在out支持列表中，校验报错
537+ {
538+ auto selfDesc = TensorDesc({10, 5}, ACL_FLOAT, ACL_FORMAT_ND);
539+ auto otherDesc = ScalarDesc(1.0f);
540+ auto scalarDesc = ScalarDesc(1.0f);
541+ auto outDesc = TensorDesc({10, 5}, ACL_INT4, ACL_FORMAT_ND);
542+ auto ut = OP_API_UT(aclnnSubs, INPUT(selfDesc, otherDesc, scalarDesc), OUTPUT(outDesc));
543+ uint64_t workspaceSize = 0;
544+ EXPECT_EQ(ut.TestGetWorkspaceSize(&workspaceSize), ACLNN_ERR_PARAM_INVALID);
545+ }
546+ // 合法out dtype不受影响
547+ {
548+ auto selfDesc = TensorDesc({10, 5}, ACL_FLOAT, ACL_FORMAT_ND);
549+ auto otherDesc = ScalarDesc(1.0f);
550+ auto scalarDesc = ScalarDesc(1.0f);
551+ auto outDesc = TensorDesc({10, 5}, ACL_FLOAT, ACL_FORMAT_ND);
552+ auto ut = OP_API_UT(aclnnSubs, INPUT(selfDesc, otherDesc, scalarDesc), OUTPUT(outDesc));
553+ uint64_t workspaceSize = 0;
554+ EXPECT_EQ(ut.TestGetWorkspaceSize(&workspaceSize), ACL_SUCCESS);
555+ }
556+ // inplace场景out与self同dtype，不受新增校验影响
557+ {
558+ auto selfDesc = TensorDesc({10, 5}, ACL_FLOAT, ACL_FORMAT_ND);
559+ auto otherDesc = ScalarDesc(1.0f);
560+ auto scalarDesc = ScalarDesc(1.0f);
561+ auto ut = OP_API_UT(aclnnInplaceSubs, INPUT(selfDesc, otherDesc, scalarDesc), OUTPUT());
562+ uint64_t workspaceSize = 0;
563+ EXPECT_EQ(ut.TestGetWorkspaceSize(&workspaceSize), ACL_SUCCESS);
564+ }
565+ // tensor-tensor入口aclnnSub：INT4不在out支持列表中，校验报错
566+ {
567+ auto selfDesc = TensorDesc({10, 5}, ACL_FLOAT, ACL_FORMAT_ND);
568+ auto otherDesc = TensorDesc({10, 5}, ACL_FLOAT, ACL_FORMAT_ND);
569+ auto scalarDesc = ScalarDesc(1.0f);
570+ auto outDesc = TensorDesc({10, 5}, ACL_INT4, ACL_FORMAT_ND);
571+ auto ut = OP_API_UT(aclnnSub, INPUT(selfDesc, otherDesc, scalarDesc), OUTPUT(outDesc));
572+ uint64_t workspaceSize = 0;
573+ EXPECT_EQ(ut.TestGetWorkspaceSize(&workspaceSize), ACLNN_ERR_PARAM_INVALID);
574+ }
575+ // tensor-tensor入口aclnnSub：合法out dtype不受影响
576+ {
577+ auto selfDesc = TensorDesc({10, 5}, ACL_FLOAT, ACL_FORMAT_ND);
578+ auto otherDesc = TensorDesc({10, 5}, ACL_FLOAT, ACL_FORMAT_ND);
579+ auto scalarDesc = ScalarDesc(1.0f);
580+ auto outDesc = TensorDesc({10, 5}, ACL_FLOAT, ACL_FORMAT_ND);
581+ auto ut = OP_API_UT(aclnnSub, INPUT(selfDesc, otherDesc, scalarDesc), OUTPUT(outDesc));
582+ uint64_t workspaceSize = 0;
583+ EXPECT_EQ(ut.TestGetWorkspaceSize(&workspaceSize), ACL_SUCCESS);
584+ }
585+}
586+ 
587+TEST_F(l2_sub_test, Ascend950_out_dtype_check_not_applied)
588+{
589+ op::SetPlatformSocVersion(op::SocVersion::ASCEND950);
590+ auto selfDesc = TensorDesc({10, 5}, ACL_FLOAT, ACL_FORMAT_ND);
591+ auto otherDesc = ScalarDesc(1.0f);
592+ auto scalarDesc = ScalarDesc(1.0f);
593+ auto outDesc = TensorDesc({10, 5}, ACL_INT4, ACL_FORMAT_ND);
594+ auto ut = OP_API_UT(aclnnSubs, INPUT(selfDesc, otherDesc, scalarDesc), OUTPUT(outDesc));
595+ uint64_t workspaceSize = 0;
596+ EXPECT_NE(ut.TestGetWorkspaceSize(&workspaceSize), ACLNN_ERR_PARAM_INVALID);
597+ op::SetPlatformSocVersion(op::SocVersion::ASCEND910B);
598+}