fork download
copy 
  1. /*
  2.  * CUDA上でのポインター配列テスト
  3.  */
  4.  
  5. #include <cuda_runtime_api.h>
  6. #include <stdlib.h>
  7. #include <stdio.h>
  8. #include <time.h>
  9.  
  10. __global__ void dev_set_pointer( float **array, float *pointer, int index )
  11. {
  12. 	array[index] = pointer;
  13. }
  14.  
  15. __global__ void dev_add_const( float **array, float add )
  16. {
  17. 	float *input;
  18. 	float *output;
  19.  
  20. 	input  = array[0];
  21. 	output = array[1];
  22.  
  23. 	output[threadIdx.x] = input[threadIdx.x] + add;
  24. }
  25.  
  26. int main( void )
  27. {
  28. 	dim3 grid;
  29. 	dim3 block;
  30.  
  31. 	cudaDeviceProp devprop;
  32.  
  33. 	int size;	
  34. 	int i;
  35. 	int memmode;
  36. 	int device;
  37. 	float add;
  38.  
  39. 	float *dev_array_0;
  40. 	float *dev_array_1;
  41. 	float **dev_pointer_array;
  42.  
  43. 	float *array_0;
  44. 	float *array_1;
  45. 	float **pointer_array;
  46.  
  47. 	size = 10;
  48. 	device = 1;
  49. 	memmode = 0;
  50. 	add = 7;
  51.  
  52. 	/* 動作設定 */
  53. 	printf( "device > " );
  54. 	scanf( "%d", &device );
  55. 	printf( "set pointer ( 0 : from host, 1 : from device ) > " );
  56. 	scanf( "%d", &memmode );
  57.  
  58. 	/* 乱数初期化 */
  59. 	srand( time( NULL ) );
  60.  
  61. 	/* デバイス設定 */
  62. 	cudaSetDevice( device );
  63.  
  64. 	/* デバイス名表示 */
  65. 	cudaGetDeviceProperties( &devprop, device );
  66. 	printf( "dev : %s\n\n", devprop.name );
  67.  
  68. 	/* 確保 */
  69. 	array_0 = (float *)malloc( sizeof(float) * size );
  70. 	array_1 = (float *)malloc( sizeof(float) * size );
  71. 	pointer_array = (float **)malloc( sizeof(float *) * 2 );
  72.  
  73. 	cudaMalloc( &dev_array_0, sizeof(float) * size );
  74. 	cudaMalloc( &dev_array_1, sizeof(float) * size );
  75. 	cudaMalloc( &dev_pointer_array, sizeof(float *) * 2 );
  76.  
  77. 	/* 初期値設定 */
  78. 	for ( i = 0; i < size; i++ ) {
  79. 		array_0[i] = (float)rand() / RAND_MAX;
  80. 	}
  81.  
  82. 	/* 転送 */
  83. 	cudaMemcpy( dev_array_0, array_0, sizeof(float) * size, cudaMemcpyHostToDevice );
  84.  
  85. 	/* カーネル起動準備 */
  86. 	grid.x = 1;
  87. 	grid.y = 1;
  88. 	grid.z = 1;
  89.  
  90. 	block.x = 1;
  91. 	block.y = 1;
  92. 	block.z = 1;
  93.  
  94. 	/* ポインターのセットをホストで行うかデバイスで行うか */
  95. 	if ( memmode ) {
  96. 		puts( "set pointer from device" );
  97.  
  98. 		dev_set_pointer<<< grid, block >>>( dev_pointer_array, dev_array_0, 0 );
  99. 		dev_set_pointer<<< grid, block >>>( dev_pointer_array, dev_array_1, 1 );
  100. 	} else {
  101. 		puts( "set pointer from host" );
  102.  
  103. 		pointer_array[0] = dev_array_0;
  104. 		pointer_array[1] = dev_array_1;
  105.  
  106. 		cudaMemcpy( dev_pointer_array, pointer_array, sizeof(float *) * 2, cudaMemcpyHostToDevice );
  107. 	}
  108.  
  109. 	/* 計算カーネル実行 */
  110. 	block.x = size;
  111.  
  112. 	dev_add_const<<< grid, block >>>( dev_pointer_array, add );
  113.  
  114. 	/* 結果取得 */
  115. 	cudaMemcpy( array_1, dev_array_1, sizeof(float) * size, cudaMemcpyDeviceToHost );
  116.  
  117. 	/* 結果確認 */
  118. 	for ( i = 0; i < size; i++ ) {
  119. 		printf( "i:%f\to:%f\n", array_0[i], array_1[i] );
  120. 	}
  121.  
  122. 	/* 解放 */
  123. 	free( array_0 );
  124. 	free( array_1 );
  125. 	free( pointer_array );
  126.  
  127. 	cudaFree( dev_array_0 );
  128. 	cudaFree( dev_array_1 );
  129. 	cudaFree( dev_pointer_array );
  130.  
  131. 	return 0;
  132. }
  133.  
Not running #stdin #stdout 0s 0KB
comments (?)
stdin
copy 
Standard input is empty
stdout
copy 
Standard output is empty
https://ideone.com/rShxo
language:
C++ (gcc 8.3)
created:
11 years ago
visibility:
public

Share or Embed source code

Discover > Sphere Engine API

The brand new service which powers Ideone!

Discover > IDE Widget

Widget for compiling and running the source code in a web browser!