CUDA编译出错，求助各位大大，感激！

wudiyu214

E:\study\delft\work\gasw\source\Cuda.targets(45,5): error MSB3721: 命令“echo "$(CUDA_BIN_PATH)\nvcc.exe"  --opencc-options -LIST:source=on  -ccbin "e:\IDEs\VS2010\VC\bin"  -D_DEBUG -D_WIN32   -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v5.5\include" -I"./" -I"../../common/inc" -I"../../../shared/inc" -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v5.5\include" -Xcompiler "/EHsc /W3 /nologo /Od /Zi   /MDd  " -maxrregcount=32 --ptxas-options=-v -gencode=arch=compute_13,code=\"sm_13,compute_13\"   --compile -o "$(IntDir)\$(InputName).cu.obj" "E:\study\delft\work\gasw\source\gpu\main.cu"

1>E:\study\delft\work\gasw\source\Cuda.targets(45,5): error MSB3721:  "$(CUDA_BIN_PATH)\nvcc.exe"  --opencc-options -LIST:source=on  -gencode=arch=compute_13,code=\"sm_13,compute_13\"    -ccbin "e:\IDEs\VS2010\VC\bin"  -D_DEBUG -D_WIN32   -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v5.5\include" -I"./" -I"../../common/inc" -I"../../../shared/inc" -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v5.5\include" -Xcompiler "/EHsc /W3 /nologo /Od /Zi   /MDd  " -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v5.5\include" -maxrregcount=32 --ptxas-options=-v -gencode=arch=compute_13,code=\"sm_13,compute_13\"   --compile -o "$(IntDir)\$(InputName).cu.obj" "E:\study\delft\work\gasw\source\gpu\main.cu"”已退出，返回代码为 1。
1>

这是错误代码，error MSB3721 返回值为1.。。
在网上搜了好多解决方法，都没什么效果，环境变量配置应该没什么问题，sdk里面的示例程序都可以运行

代码是smith waterman的gpu加速实现，cu文件代码如下

1. #include "stdafx.h"
   
2. #include "main_cu.h"
   
3. #include "gpudb.h"
   
4. #include "fastafile.h"
   
5. #include "main.h"
   
6. 
   
7. //__constant__ queryType c_query[63*1024/sizeof(queryType)]; /**< Query sequence in constant memory */
   
8. __constant__ int c_gapPenaltyTotal; //Gap + Extend penalties
   
9. __constant__ int c_gapExtendPenalty; //Extend penalty
   
10. __constant__ int c_queryLength; //Length of query sequence in chunks of 4
    
11. 
    
12. texture t_queryProfile;
    
13. 
    
14. /**
    
15. Align 4 query profile entries with a database residue.
    
16. */
    
17. __device__  void alignResidues(scoreType &maxScore, scoreType4& left, int4& ixLeft, scoreType& top, scoreType& topLeft, int &IyTop, const char4 &substScore)
    
18. {       
    
19.         //q0
    
20.         ixLeft.x = max(0,max(left.x+c_gapPenaltyTotal,ixLeft.x+c_gapExtendPenalty)); //Max(0,...) here so IxColumn[] can be unsigned
    
21.         IyTop = max(top+c_gapPenaltyTotal,IyTop+c_gapExtendPenalty);
    
22.         int align = topLeft+substScore.x;
    
23.         topLeft=left.x;
    
24.         left.x = max(align,max(ixLeft.x,IyTop));
    
25.        
    
26.         //q1
    
27.         ixLeft.y = max(0,max(left.y+c_gapPenaltyTotal,ixLeft.y+c_gapExtendPenalty)); //Max(0,...) here so IxColumn[] can be unsigned
    
28.         IyTop = max(left.x+c_gapPenaltyTotal,IyTop+c_gapExtendPenalty);
    
29.         align = topLeft+substScore.y;
    
30.         topLeft=left.y;
    
31.         left.y = max(align,max(ixLeft.y,IyTop));
    
32. 
    
33.         //q2
    
34.         ixLeft.z = max(0,max(left.z+c_gapPenaltyTotal,ixLeft.z+c_gapExtendPenalty)); //Max(0,...) here so IxColumn[] can be unsigned
    
35.         IyTop = max(left.y+c_gapPenaltyTotal,IyTop+c_gapExtendPenalty);
    
36.         align = topLeft+substScore.z;
    
37.         topLeft=left.z;
    
38.         left.z = max(align,max(ixLeft.z,IyTop));
    
39. 
    
40.         //q3
    
41.         ixLeft.w = max(0,max(left.w+c_gapPenaltyTotal,ixLeft.w+c_gapExtendPenalty)); //Max(0,...) here so IxColumn[] can be unsigned
    
42.         IyTop = max(left.z+c_gapPenaltyTotal,IyTop+c_gapExtendPenalty);
    
43.         align = topLeft+substScore.w;       
    
44.         left.w = max(align,max(ixLeft.w,IyTop));
    
45. 
    
46.         topLeft=top; //The next column is to the right of this one, so current top left becomes new top
    
47.         top = left.w; //Set top value for next query chunk
    
48.         maxScore = max(left.x,max(left.y,max(left.z,max(left.w,maxScore)))); //Update max score
    
49. }                       
    
50.                        
    
51. /**
    
52. Align a database sequence subblock with the entire query sequence.
    
53. The loading/aligning with the query sequence in the 'inner' function as query sequence (constant) memory is much faster than the global memory in which the db resides.
    
54. */
    
55. __device__ inline void alignWithQuery(const seqType8 &s, int column, TempData2* tempColumn, scoreType &maxScore)
    
56. {
    
57.                
    
58.                 //Set the top related values to 0 as we're at the top of the matrix
    
59.                 scoreType8 top = {0,0,0,0,0,0,0,0};
    
60.                 scoreType topLeft = 0;
    
61.                 int8 IyTop = {0,0,0,0,0,0,0,0};               
    
62. 
    
63.                 char4 substScores; //Query profile scores
    
64.                 scoreType4 left;
    
65.                 int4 ixLeft;
    
66.                 for(int j=0;j> GPUdb::LOG2_BLOCK_SIZE;
    
67.                 int groupNumInBlock = groupNum & (GPUdb::BLOCK_SIZE-1); //equals groupNum % GPUdb::BLOCK_SIZE
    
68.                 int groupOffset = blockOffsets[seqBlock]+__umul24(groupNumInBlock,GPUdb::SUBBLOCK_SIZE);
    
69.                 const GPUdb::seqType* group = &sequences[groupOffset];
    
70.        
    
71.                                
    
72.                 align(group,tempColumn,seqNum,scores); //Perform alignment
    
73.                 groupNum+= gridDim.x*blockDim.x;
    
74.        
    
75.         }
    
76. 
    
77. }
    
78. 
    
79. // main routine that executes on the host
    
80. bool launchSW(scoreType** scores, void* query, size_t queryLength, GPUdb& db, FastaMatrix& substitutionMatrix, int gapPenalty, int gapExtendPenalty, double& time)
    
81. {
    
82.         /*
    
83.         //Prepare substitution matrix
    
84.         if(!substitutionMatrix.copyToGPU())
    
85.         {
    
86.                 puts("Error uploading substitution matrix.");
    
87.                 return false;
    
88.         }
    
89.         t_substMatrix.addressMode[0] = cudaAddressModeWrap;
    
90.         t_substMatrix.addressMode[1] = cudaAddressModeWrap;
    
91.         t_substMatrix.filterMode = cudaFilterModePoint;
    
92.         t_substMatrix.normalized = false;
    
93.         cudaBindTextureToArray(t_substMatrix,substitutionMatrix.getCudaArray(),substitutionMatrix.getChannelDesc());       
    
94.         */
    
95. 
    
96.         //Prepare substitution matrix
    
97.        
    
98.         if(!substitutionMatrix.copyQueryProfileToGPU())
    
99.         {
    
100.                 puts("Error uploading query profile.");
     
101.                 return false;
     
102.         }
     
103.         t_queryProfile.addressMode[0] = cudaAddressModeWrap;
     
104.         t_queryProfile.addressMode[1] = cudaAddressModeWrap;
     
105.         t_queryProfile.filterMode = cudaFilterModePoint;
     
106.         t_queryProfile.normalized = false;
     
107.         cudaBindTextureToArray(t_queryProfile,substitutionMatrix.getQueryProfileCudaArray(),substitutionMatrix.getQueryProfileChannelDesc());       
     
108. 
     
109.         //Prepare database
     
110.         if(!db.copyToGPU())
     
111.         {
     
112.                 puts("Error uploading database.");
     
113.                 return false;
     
114.         }
     
115. 
     
116.         //Prepare query
     
117. /*        if(cudaMemcpyToSymbol(c_query,query,queryLength,0,cudaMemcpyHostToDevice)!=cudaSuccess)
     
118.         {
     
119.                 puts("Error copying query sequence.");
     
120.                 return false;
     
121.         }*/
     
122.         size_t queryLengthInChunks = WHOLE_AMOUNT_OF(queryLength,sizeof(queryType));
     
123.         size_t queryLengthDiv2InChunks = WHOLE_AMOUNT_OF(queryLength/2,sizeof(queryType));
     
124.         if(cudaMemcpyToSymbol(c_queryLength,&queryLengthInChunks,sizeof(queryLengthInChunks),0,cudaMemcpyHostToDevice)!=cudaSuccess)
     
125.         {
     
126.                 puts("Error copying query sequence length.");
     
127.                 return false;
     
128.         }
     
129. 
     
130.         //Prepare penalties
     
131.         int gapPenaltyTotal = gapPenalty + gapExtendPenalty;
     
132.         if(cudaMemcpyToSymbol(c_gapPenaltyTotal,&gapPenaltyTotal,sizeof(gapPenaltyTotal),0,cudaMemcpyHostToDevice)!=cudaSuccess
     
133.         | cudaMemcpyToSymbol(c_gapExtendPenalty,&gapExtendPenalty,sizeof(gapExtendPenalty),0,cudaMemcpyHostToDevice)!=cudaSuccess)
     
134.         {
     
135.                 puts("Error copying penalties.");
     
136.                 return false;
     
137.         }
     
138. 
     
139.         //Prepare score array
     
140.         size_t scoreArraySize = sizeof(scoreType)*db.getNumSequences();
     
141.         if(cudaMallocHost(scores,scoreArraySize)!=cudaSuccess)
     
142.         {
     
143.                 puts("Error allocating host score array.");
     
144.                 return false;
     
145.         }
     
146.        
     
147.         scoreType* d_scores;
     
148.         if(cudaMalloc(&d_scores,scoreArraySize)!=cudaSuccess)
     
149.         {
     
150.                 puts("Error allocating device score array: database too large?.");
     
151.                 return false;
     
152.         }
     
153.         cudaMemset(d_scores,-1,scoreArraySize); //Set scores to -1 so we can check if they were actually all written by the kernel.
     
154.        
     
155. 
     
156.         int matrixSize = queryLengthDiv2InChunks*sizeof(queryType)*sizeof(TempData2); //Size of temporary storage for one thread
     
157. 
     
158.         //Determine launch configuration
     
159.         cudaDeviceProp props;
     
160.         cudaGetDeviceProperties(&props,0);
     
161.         size_t free, total;
     
162.         cudaMemGetInfo(&free,&total);
     
163.                
     
164.         if(db.getNumSequences() < 16)
     
165.         {
     
166.                 puts("Database does not contain enough sequences to have a full thread block.");
     
167.                 return false;
     
168.         }
     
169.        
     
170.                
     
171.         int blocks = props.multiProcessorCount*4;
     
172.         int threadsPerBlock = props.maxThreadsPerBlock/8; //Blocksize should be a multiple of 32 threads
     
173. 
     
174.         int blocksPerhw = (int)ceil((double)db.getNumBlocks()/(double)(blocks*threadsPerBlock/16));
     
175. 
     
176. 
     
177. 
     
178.         //threadsPerBlock=1;blocks=32;
     
179.         printf("Using %d blocks of %d threads: %d threads for %d sequences in %d blocks.\n",blocks,threadsPerBlock,blocks*threadsPerBlock,db.getNumSequences(),db.getNumBlocks());
     
180.         printf("Processing %d blocks per half warp.\n",blocksPerhw);
     
181.         //printf("Processing %d sequences per thread.\n",symbolsPerThread);
     
182. 
     
183.         //Prepare temporary score matrices: one F and one Ix column per thread
     
184.         TempData2* d_tempColumns;
     
185.         matrixSize *= blocks*threadsPerBlock;
     
186.        
     
187.         if(cudaMalloc(&d_tempColumns,matrixSize)!=cudaSuccess)
     
188.         {
     
189.                 puts("Error allocating temporary matrix: too many threads for sequence size?");
     
190.                 return false;
     
191.         }
     
192.         /*if(cudaMemset(d_tempColumns,0,matrixSize)!=cudaSuccess)
     
193.         {
     
194.                 puts("Error allocating temporary matrix.");
     
195.                 return false;
     
196.         }*/
     
197. 
     
198.         //Run kernel
     
199.         fflush(stdout);
     
200.         puts("Running...");
     
201. 
     
202.        
     
203.         clock_t start = clock();
     
204.         smithWaterman <<>>(db.getNumBlocks()*GPUdb::BLOCK_SIZE,d_scores,db.get_d_BlockOffsets(),  db.get_d_SeqNums(),db.get_d_Sequences(), d_tempColumns);       
     
205. 
     
206.         if(cudaThreadSynchronize()!=cudaSuccess)
     
207.                 return false;
     
208. 
     
209.         clock_t stop = clock();
     
210.         time=(stop-start)/(double)CLOCKS_PER_SEC;
     
211. 
     
212.         //Get scores
     
213.         cudaMemcpy((void*) *scores,d_scores,scoreArraySize,cudaMemcpyDeviceToHost);
     
214.        
     
215. 
     
216.         cudaFree(d_tempColumns);       
     
217.         cudaFree(d_scores);
     
218.         return true;
     
219. }

复制代码

希望各位大大帮帮忙，这里先谢过

wudiyu214

仔细看了一下好想是自己的GPU arch选错了-gencode=arch=compute_13,code=\"sm_13,compute_13\这个地方应该是compute_20但是自己在VS2010上的项目属性配置上已经更改，但是却毫无效果，不知道为什么
C:\Users\wdy\Desktop\QQ截图20140124171051.jpg

giant_dragon

我都是直接用命令行编译的，好像不需要设置吧，是不是系统能自动搜索device？

路曼曼

680就是个低功耗版本的过渡产品，当然不是想当然的那样越来越好。