0
我試圖讓這個代碼與3D類型結構一起工作。我正在使用Cuda的2D功能。所以主機端線性數據（'board'）的大小是width * height * depth，而2D mallocs是width x height * depth（這裏的寬度和高度都是DIMxDIM元素）。內核從A的數據以B.我得到一個非法的內存訪問錯誤（使用存儲器校驗），在該行CUDA非法內存訪問
dst[offset] = curr;
錯誤消失，如果我改變的malloc身高* 2，但大小似乎匹配。我錯過了什麼？其他批評也是受歡迎的，我對C++和CUDA都是新手。
#include "cuda_runtime.h"
#include "device_launch_parameters.h"
#include <iostream>
#include <stdio.h>
#include <stdlib.h>
typedef signed int sint;
typedef unsigned int uint;
#define DIM 512
#define TPB 32 // Threads per block
#define CLEARANCE 5
#define MAPLAYERS 2
#define WIDTH (sizeof(sint) * DIM)
#define HEIGHT (DIM * MAPLAYERS)
void route(sint *A, size_t &pitchA, sint *B, size_t &pitchB, sint *board, int src, int dest);
__global__ void map(sint *src, size_t pitchSrc, sint *dst, size_t pitchDst, unsigned long *index);
__device__ bool hasClearance(sint* src, sint x, sint y, sint z, size_t pitch);
__device__ bool inBounds(sint x, sint y, sint z, sint xoff, sint yoff, sint zoff);
__device__ inline long long calcOffset(sint x, sint y, sint z, sint xoff, sint yoff, sint zoff, size_t pitch);
dim3 blocks(DIM/TPB, DIM/TPB, MAPLAYERS);
dim3 threads(TPB, TPB);
/** CUDA Error Check */
#define CER(ans) { gpuAssert((ans), __FILE__, __LINE__); }
inline void gpuAssert(cudaError_t code, const char *file, int line)
{
if (code != cudaSuccess)
{
fprintf(stderr, "GPUassert: %s %s %d\n", cudaGetErrorString(code), file, line);
int tmp;
std::cin >> tmp;
exit(code);
}
}
int main(void) {
sint *A;
sint *B;
size_t pitchA, pitchB;
sint *board = new sint[WIDTH*HEIGHT];
CER(cudaMallocPitch(&A, &pitchA, WIDTH, HEIGHT));
CER(cudaMallocPitch(&B, &pitchB, WIDTH, HEIGHT));
CER(cudaMemset2D(A, pitchA, 0, WIDTH, HEIGHT));
CER(cudaMemset2D(B, pitchA, 0, WIDTH, HEIGHT));
route(A, pitchA, B, pitchB, board, 0, DIM*DIM - 1);
CER(cudaFree(A));
CER(cudaFree(B));
delete[] board;
}
void route(sint *A, size_t &pitchA, sint *B, size_t &pitchB, sint *board, int src, int dest) {
unsigned long *dev_index;
unsigned long index = NULL;
CER(cudaMalloc((void**)&dev_index, sizeof(unsigned long)));
CER(cudaMemcpy(dev_index, &index, sizeof(unsigned long), cudaMemcpyHostToDevice));
CER(cudaMemcpy2D(A, pitchA, board, WIDTH, WIDTH, HEIGHT, cudaMemcpyHostToDevice));
CER(cudaMemcpy2D(B, pitchB, board, WIDTH, WIDTH, HEIGHT, cudaMemcpyHostToDevice));
map << <blocks, threads >> >(B, pitchB, A, pitchA, dev_index);
CER(cudaPeekAtLastError());
CER(cudaMemcpy(&index, dev_index, sizeof(unsigned long), cudaMemcpyDeviceToHost));
if (index != NULL) {
// break condition
}
}
__global__ void map(sint *src, size_t pitchSrc, sint *dst, size_t pitchDst, unsigned long *index) {
unsigned int x = threadIdx.x + blockIdx.x * blockDim.x;
unsigned int y = threadIdx.y + blockIdx.y * blockDim.y;
unsigned int z = blockIdx.z + blockIdx.z * blockDim.z;
unsigned long long offset = calcOffset(x, y, z, 0, 0, 0, pitchDst);
sint curr;
if (!inBounds(x, y, z, 0, 0, 0))
return;
curr = src[calcOffset(x, y, z, 0, 0, 0, pitchSrc)];
if (z % 2 == 0 && curr == 0 && hasClearance(src, x, y, z, pitchSrc)) {
// Processing
}
else
dst[offset] = 1;
return;
}
/** Finds linear offset for a given pixel and offset. */
__device__ inline long long calcOffset(sint x, sint y, sint z, sint xoff, sint yoff, sint zoff, size_t pitch) {
return (x + xoff) + (y + yoff) * pitch + ((z + zoff) * pitch * (HEIGHT/MAPLAYERS));
}
/** Checks if position is valid on the map. */
__device__ bool inBounds(sint x, sint y, sint z, sint xoff, sint yoff, sint zoff) {
if (0 > (x + xoff) || (x + xoff) >= DIM || 0 > (y + yoff) || (y + yoff) >= DIM || 0 > (z + zoff) || (z + zoff) >= MAPLAYERS)
return false;
return true;
}
/** Returns true if a block has clearnace */
__device__ bool hasClearance(sint* src, sint x, sint y, sint z, size_t pitch) {
for (int c = -CLEARANCE; c <= CLEARANCE; c++) {
for (int r = -CLEARANCE; r <= CLEARANCE; r++){
if (inBounds(x, y, z, r, c, 0)){
if (src[calcOffset(x, y, z, r, c, 0, pitch)] == 2 || src[calcOffset(x, y, z, r, c, 0, pitch)] == 1)
return false;
}
else {
return false;
}
}
}
return true;
}
的CUDA調試器的輸出：
Memory Checker detected 384 access violations.
error = access violation on load (global memory)
gridid = 18
blockIdx = {0,8,0}
threadIdx = {0,4,0}
address = 0x05d08000
accessSize = 4
請提供其他人可以編譯的代碼。如果您不確定這是什麼，請將這個問題中的代碼複製到一個全新的項目中，並且繼續修復所有編譯錯誤，直到沒有任何編譯錯誤。然後確保代碼演示您詢問的訪問衝突。然後將該固定代碼粘貼回問題中。 –