fix: 修复 FFT 算子 host 侧缓冲区尺寸计算的整数溢出-ops-fft-AtomGit

Star71

Star71

已合并

fix: 修复 FFT 算子 host 侧缓冲区尺寸计算的整数溢出 #33

Tian_1122创建于 14 天前从Tian_1122/ops-fft： fix/fft-integer-overflow 合入到 cann/ops-fft： master

fix: 修复 FFT 算子 host 侧缓冲区尺寸计算的整数溢出 #33

已合并

Tian_1122创建于 14 天前

从Tian_1122/ops-fft fix/fft-integer-overflow合入到cann/ops-fft master

共 8 个文件变更+19-19

Msrc/fft1_d/arch32/fft_b/fft1_d_b.cpp+2-2

	@@ -142,8 +142,8 @@ extern "C" aclError aclfftFft1DB(float x, float y, uint32_t n,
142	std::vector<uint32_t> radixVec;	142	std::vector<uint32_t> radixVec;
143	InitRadixB(n, radixVec);	143	InitRadixB(n, radixVec);
144		144
145	- uint32_t inputSize = n * batches * sizeof(float) * 2;	145	+ size_t inputSize = static_cast<size_t>(n) * batches * sizeof(float) * 2;
146	- uint32_t outputSize = inputSize;	146	+ size_t outputSize = inputSize;
147		147
148	std::vector<float> wMatrixHost;	148	std::vector<float> wMatrixHost;
149	for (size_t i = 0; i < radixVec.size(); i++) {	149	for (size_t i = 0; i < radixVec.size(); i++) {

Msrc/fft1_d/arch32/fft_mix/fft1_d_mix.cpp+1-1

	@@ -69,7 +69,7 @@ extern "C" aclError aclfftFft1DMix(float x, float y, uint32_t n,
69	int64_t totalWs = wsIn + wsOut + wsSync + wsC2c + wsAux;	69	int64_t totalWs = wsIn + wsOut + wsSync + wsC2c + wsAux;
70		70
71	// 6. Allocate & copy	71	// 6. Allocate & copy
72	- uint32_t inputSize = n * batches * sizeof(float) * 2;	72	+ size_t inputSize = static_cast<size_t>(n) * batches * sizeof(float) * 2;
73	void dIn=nullptr,dOut=nullptr,dDft=nullptr,dTw=nullptr,dRadix=nullptr,dWs=nullptr,*dTil=nullptr;	73	void dIn=nullptr,dOut=nullptr,dDft=nullptr,dTw=nullptr,dRadix=nullptr,dWs=nullptr,*dTil=nullptr;
74	CHECK_ACL(aclrtMalloc(&dIn, inputSize, ACL_MEM_MALLOC_HUGE_FIRST));	74	CHECK_ACL(aclrtMalloc(&dIn, inputSize, ACL_MEM_MALLOC_HUGE_FIRST));
75	CHECK_ACL(aclrtMalloc(&dOut, inputSize, ACL_MEM_MALLOC_HUGE_FIRST));	75	CHECK_ACL(aclrtMalloc(&dOut, inputSize, ACL_MEM_MALLOC_HUGE_FIRST));

Msrc/fft1_d/arch32/fft_n/fft1_d_n.cpp+2-2

	@@ -265,8 +265,8 @@ extern "C" aclError aclfftFft1DN(float x, float y, uint32_t n, int32_t norm,
265	}	265	}
266	int32_t repeatBatchSize = 0;	266	int32_t repeatBatchSize = 0;
267	ComputeRepeatBatchSize(n, batches, repeatBatchSize);	267	ComputeRepeatBatchSize(n, batches, repeatBatchSize);
268	- const uint32_t inputSize = n * batches * sizeof(float) * 2;	268	+ const size_t inputSize = static_cast<size_t>(n) * batches * sizeof(float) * 2;
269	- const uint32_t outputSize = inputSize;	269	+ const size_t outputSize = inputSize;
270	std::vector<float> wMatrixHost;	270	std::vector<float> wMatrixHost;
271	for (size_t i = 0; i < radixVec.size(); i++) {	271	for (size_t i = 0; i < radixVec.size(); i++) {
272	uint32_t radix = radixVec[i];	272	uint32_t radix = radixVec[i];

Msrc/fft1_d/arch32/fft_stride/fft1_d_stride.cpp+4-4

	@@ -131,13 +131,13 @@ aclError aclfftFft1DStride(float x, float y, uint32_t n, uint32_t stride,
131		131
132	uint32_t s0 = ComputeS0(n, stride);	132	uint32_t s0 = ComputeS0(n, stride);
133		133
134	- uint32_t inputSize = n * stride * sizeof(float) * 2;	134	+ size_t inputSize = static_cast<size_t>(n) * stride * sizeof(float) * 2;
135	uint32_t sMatrixSize = sMatrixHost.size() * sizeof(float);	135	uint32_t sMatrixSize = sMatrixHost.size() * sizeof(float);
136	- uint32_t outputSize = inputSize;	136	+ size_t outputSize = inputSize;
137	- uint32_t kernelWorkspaceSize = n * s0 * sizeof(float) * 4;	137	+ size_t kernelWorkspaceSize = static_cast<size_t>(n) * s0 * sizeof(float) * 4;
138	uint32_t tilingSize = sizeof(Fft1DStrideTilingData);	138	uint32_t tilingSize = sizeof(Fft1DStrideTilingData);
139	uint32_t sysWorkspaceSize = ascendcPlatform->GetLibApiWorkSpaceSize();	139	uint32_t sysWorkspaceSize = ascendcPlatform->GetLibApiWorkSpaceSize();
140	- uint32_t totalWorkspaceSize = kernelWorkspaceSize + sysWorkspaceSize;	140	+ size_t totalWorkspaceSize = kernelWorkspaceSize + sysWorkspaceSize;
141		141
142	void *dev_input = nullptr;	142	void *dev_input = nullptr;
143	void *dev_s_matrix = nullptr;	143	void *dev_s_matrix = nullptr;

Msrc/fft2_d/arch32/dd/fft2_d_dd.cpp+3-3

	@@ -76,15 +76,15 @@ extern "C" aclError aclfftFft2DDd(float x, float y, uint32_t fftX, uint32_t ff
76	return ACL_ERROR_INVALID_PARAM;	76	return ACL_ERROR_INVALID_PARAM;
77	}	77	}
78		78
79	- uint32_t inputSize = batches * fftX * fftY * sizeof(float) * 2;	79	+ size_t inputSize = static_cast<size_t>(batches) * fftX * fftY * sizeof(float) * 2;
80	- uint32_t outputSize = inputSize;	80	+ size_t outputSize = inputSize;
81		81
82	std::vector<float> pMatrixHost = InitPQMatrix(fftX, isForward != 0, true);	82	std::vector<float> pMatrixHost = InitPQMatrix(fftX, isForward != 0, true);
83	std::vector<float> qMatrixHost = InitPQMatrix(fftY, isForward != 0, false);	83	std::vector<float> qMatrixHost = InitPQMatrix(fftY, isForward != 0, false);
84	uint32_t pMatrixSize = pMatrixHost.size() * sizeof(float);	84	uint32_t pMatrixSize = pMatrixHost.size() * sizeof(float);
85	uint32_t qMatrixSize = qMatrixHost.size() * sizeof(float);	85	uint32_t qMatrixSize = qMatrixHost.size() * sizeof(float);
86		86
87	- uint64_t workspaceSize = batches * fftX * fftY * sizeof(float) * 2;	87	+ uint64_t workspaceSize = static_cast<uint64_t>(batches) * fftX * fftY * sizeof(float) * 2;
88		88
89	void *dev_input = nullptr;	89	void *dev_input = nullptr;
90	void *dev_output = nullptr;	90	void *dev_output = nullptr;

Msrc/irfft1_d/arch32/c2r_fft/irfft1_d_c2r_fft.cpp+2-2

	@@ -172,8 +172,8 @@ extern "C" aclError aclfftIrfft1DC2RFft(float x, float y, uint32_t n,
172	int64_t totalWs = wsIn + wsOut + wsSync + wsC2c + wsAux;	172	int64_t totalWs = wsIn + wsOut + wsSync + wsC2c + wsAux;
173		173
174	// 7. Allocate & copy	174	// 7. Allocate & copy
175	- uint32_t inputSize = (n / 2 + 1) * batches * sizeof(float) * 2;	175	+ size_t inputSize = static_cast<size_t>(n / 2 + 1) * batches * sizeof(float) * 2;
176	- uint32_t outputSize = n * batches * sizeof(float);	176	+ size_t outputSize = static_cast<size_t>(n) * batches * sizeof(float);
177	void dIn=nullptr,dOut=nullptr,dDft=nullptr,dTw=nullptr,dRadix=nullptr,dWs=nullptr,*dTil=nullptr;	177	void dIn=nullptr,dOut=nullptr,dDft=nullptr,dTw=nullptr,dRadix=nullptr,dWs=nullptr,*dTil=nullptr;
178	void dInIdx=nullptr,dA=nullptr,dB=nullptr,dOutIdx=nullptr;	178	void dInIdx=nullptr,dA=nullptr,dB=nullptr,dOutIdx=nullptr;
179	CHECK_ACL(aclrtMalloc(&dIn, inputSize, ACL_MEM_MALLOC_HUGE_FIRST));	179	CHECK_ACL(aclrtMalloc(&dIn, inputSize, ACL_MEM_MALLOC_HUGE_FIRST));

Msrc/irfft1_d/arch35/fft/irfft1_d_fft.cpp+3-3

	@@ -141,12 +141,12 @@ extern "C" aclError aclfftIrfft1DFft(float x, float y, uint32_t n, int32_t nor
141	tempN = M;	141	tempN = M;
142	}	142	}
143		143
144	- const uint32_t inputSize = batches * (n / 2 + 1) * sizeof(float) * 2;	144	+ const size_t inputSize = static_cast<size_t>(batches) * (n / 2 + 1) * sizeof(float) * 2;
145	- const uint32_t outputSize = batches * n * sizeof(float);	145	+ const size_t outputSize = static_cast<size_t>(batches) * n * sizeof(float);
146	const uint32_t fftMatrixSize = allfftMatrices.size() * sizeof(float);	146	const uint32_t fftMatrixSize = allfftMatrices.size() * sizeof(float);
147	const uint32_t twSize = allTwiddleFactors.size() * sizeof(float);	147	const uint32_t twSize = allTwiddleFactors.size() * sizeof(float);
148	const uint32_t radixListSize = tilingData.radixListLen * sizeof(float);	148	const uint32_t radixListSize = tilingData.radixListLen * sizeof(float);
149	- const uint32_t workspaceSize = 2 * batches * n * sizeof(float) * 2;	149	+ const size_t workspaceSize = 2 * static_cast<size_t>(batches) * n * sizeof(float) * 2;
150	const uint32_t tilingSize = sizeof(Irfft1DfftTilingData);	150	const uint32_t tilingSize = sizeof(Irfft1DfftTilingData);
151		151
152	void *dev_input = nullptr;	152	void *dev_input = nullptr;

Msrc/rfft1_d/arch32/r2c_fft/rfft1_d_r2c_fft.cpp+2-2

	@@ -146,8 +146,8 @@ extern "C" aclError aclfftRfft1DR2CFft(float x, float y, uint32_t n,
146		146
147	// 7. Allocate & copy	147	// 7. Allocate & copy
148	// R2C: input is n real floats, output is (n/2+1) complex = (n/2+1)*2 floats	148	// R2C: input is n real floats, output is (n/2+1) complex = (n/2+1)*2 floats
149	- uint32_t inputSize = n * batches * sizeof(float);	149	+ size_t inputSize = static_cast<size_t>(n) * batches * sizeof(float);
150	- uint32_t outputSize = (n / 2 + 1) * batches * sizeof(float) * 2;	150	+ size_t outputSize = static_cast<size_t>(n / 2 + 1) * batches * sizeof(float) * 2;
151	void dIn=nullptr,dOut=nullptr,dDft=nullptr,dTw=nullptr,dRadix=nullptr,dWs=nullptr,*dTil=nullptr;	151	void dIn=nullptr,dOut=nullptr,dDft=nullptr,dTw=nullptr,dRadix=nullptr,dWs=nullptr,*dTil=nullptr;
152	void dInIdx=nullptr,dA=nullptr,dB=nullptr,dOutIdx=nullptr;	152	void dInIdx=nullptr,dA=nullptr,dB=nullptr,dOutIdx=nullptr;
153	CHECK_ACL(aclrtMalloc(&dIn, inputSize, ACL_MEM_MALLOC_HUGE_FIRST));	153	CHECK_ACL(aclrtMalloc(&dIn, inputSize, ACL_MEM_MALLOC_HUGE_FIRST));