weixin_41908053
liuxin95
采纳率37.5%
2019-05-11 15:48

pycuda报错cuMemFree failed

10
  • c++
  • python
  • ubuntu

报错内存释失败:Traceback (most recent call last):
File "filter_cuda.py", line 103, in
bilateral = Bilateral_filter(img,template_size,sigma[0],sigma[1])
File "filter_cuda.py", line 93, in Bilateral_filter
block=(template_size,template_size,3),grid=(rows,cols))
File "/usr/local/lib/python3.5/dist-packages/pycuda/driver.py", line 405, in function_call
Context.synchronize()
pycuda._driver.LogicError: cuCtxSynchronize failed: an illegal memory access was encountered
PyCUDA WARNING: a clean-up operation failed (dead context maybe?)
cuMemFree failed: an illegal memory access was encountered
PyCUDA WARNING: a clean-up operation failed (dead context maybe?)
cuMemFree failed: an illegal memory access was encountered
PyCUDA WARNING: a clean-up operation failed (dead context maybe?)
cuMemFree failed: an illegal memory access was encountered
PyCUDA WARNING: a clean-up operation failed (dead context maybe?)
cuMemFree failed: an illegal memory access was encountered
PyCUDA WARNING: a clean-up operation failed (dead context maybe?)
cuModuleUnload failed: an illegal memory access was encountered

import os,math,cv2,numpy
from PIL import Image
import numpy as np
import skimage
import pycuda.autoinit
import pycuda.driver as drv
from pycuda.compiler import SourceModule
from timeit import default_timer as timer

mod = SourceModule("""
#include<math.h>
__global__ void Gauss_cuda(int ***img,int ***im,float **disTemplate,int template_size)
{
    int x = blockIdx.x * blockDim.x + threadIdx.x;
    int y = blockIdx.y * blockDim.y + threadIdx.y;
    int z = threadIdx.z;
    int count = 0;
    for(int i=x;i<x+template_size;i++)
    {
        for(int j=y;j<y+template_size;j++)
        {
            count += im[i][j][z]*disTemplate[i][j]; 
        }
    }
    img[x][y][z] = count;  
}
__global__ void bilateral_cuda(float ***img,float ***orig,float **disTemplate,float *ourArg)
{
    int x = blockIdx.x * blockDim.x + threadIdx.x;
    int y = blockIdx.y * blockDim.y + threadIdx.y;
    int z = threadIdx.z;
    int xmin = max(int(x-int(ourArg[0])),0);
    int xmax = min(int(x+int(ourArg[0])+1),int(ourArg[1]));
    int ymin = max(int(y-int(ourArg[0])),0);
    int ymax = min(int(y+int(ourArg[0])+1),int(ourArg[2]));
    int w = 0;
    int v = 0;
    for(int i=xmin;i<xmax;i++)
    {
        for(int j=ymin;j<ymax;j++)
        {
            w += w+disTemplate[xmin-x+int(ourArg[0])+i][ymin-y+int(ourArg[0])+i]*exp((orig[i][j][z]-orig[x][y][z])*(orig[i][j][z]-orig[x][y][z])/(2*ourArg[3]*ourArg[3]));
            v += orig[x][y][z]*w;               
        }
    }
    img[x][y][z] = v/w;  
}

""")

Gauss_cuda = mod.get_function("Gauss_cuda")
bilateral_cuda = mod.get_function("bilateral_cuda")


def generateGaussianTemplate(template_size,sigma):
    template = np.zeros((template_size,template_size))
    mid = template_size/2
    sum = 0
    pi = 3.1415926
    for i in range(template_size):
        x = pow(i-mid,2)
        for j in range(template_size):
            y = pow(j-mid,2)
            g = math.exp((x+y)/(-2*sigma*sigma))/(2*pi*sigma)
            template[i][j] = g
            sum+=g
    #归一化
    template = template/sum
    return template

def Gauss_filter(img,template_size,sigma):
    [rows,cols,channel] = img.shape
    border = int(template_size/2)
    template = generateGaussianTemplate(template_size,sigma)
    im = cv2.copyMakeBorder(img,border,border,border,border,cv2.BORDER_REPLICATE)
    Gauss_cuda(
                drv.InOut(img), drv.In(im), drv.In(template),template_size,
                block=(template_size,template_size,3),grid=(rows,cols))
    return img

def Bilateral_filter(img,template_size,sigma1,sigma2):
    img = img/255
    border = int(template_size/2)
    [rows,cols,channel] = img.shape
    tmp = np.arange(-border,border+1)
    [x,y] = np.meshgrid(tmp,tmp)
    X = np.power(x,2)
    Y = np.power(y,2)
    d = np.exp(-(X+Y)/(2*sigma1*sigma1))
    orig_img = img
    bilateral_cuda(
                drv.InOut(img), drv.In(orig_img), drv.In(d),drv.In(np.array([border,rows,cols,sigma2])),
                block=(template_size,template_size,3),grid=(rows,cols))
    return img*255

if __name__=='__main__':
    img = cv2.imread('demo2.jpg')
    #noise = skimage.util.random_noise(img,'gaussian')*255
    #cv2.imwrite('noise.jpg',noise)
    template_size = 3
    sigma_g = 0.8
    sigma = [3,0.1]
    bilateral = Bilateral_filter(img,template_size,sigma[0],sigma[1])
    Gauss = Gauss_filter(img,template_size,sigma_g)
    cv2.imwrite('bilateral_result_{}_{}_{}.jpg'.format(template_size,sigma[0],sigma[1]),bilateral)
    cv2.imwrite('Gauss_result_{}_{}.jpg'.format(template_size,sigma_g),Gauss)

  • 点赞
  • 写回答
  • 关注问题
  • 收藏
  • 复制链接分享
  • 邀请回答

2条回答

  • dabocaiqq dabocaiqq 2年前
  • qq_44220918 叛逆&无情 2年前

    import os,math,cv2,numpy
    from PIL import Image
    import numpy as np
    import skimage
    import pycuda.autoinit
    import pycuda.driver as drv
    from pycuda.compiler import SourceModule
    from timeit import default_timer as timer

    mod = SourceModule("""
    #include
    global void Gauss_cuda(int **img,int **im,float **disTemplate,int template_size)
    {
    int x = blockIdx.x * blockDim.x + threadIdx.x;
    int y = blockIdx.y * blockDim.y + threadIdx.y;
    int z = threadIdx.z;
    int count = 0;
    for(int i=x;i<x+template_size;i++)
    {
    for(int j=y;j<y+template_size;j++)
    {
    count += im[i][j][z]*disTemplate[i][j];
    }
    }
    img[x][y][z] = count;

    }
    global void bilateral_cuda(float **img,float **orig,float **disTemplate,float ourArg)
    {
    int x = blockIdx.x * blockDim.x + threadIdx.x;
    int y = blockIdx.y * blockDim.y + threadIdx.y;
    int z = threadIdx.z;
    int xmin = max(int(x-int(ourArg[0])),0);
    int xmax = min(int(x+int(ourArg[0])+1),int(ourArg[1]));
    int ymin = max(int(y-int(ourArg[0])),0);
    int ymax = min(int(y+int(ourArg[0])+1),int(ourArg[2]));
    int w = 0;
    int v = 0;
    for(int i=xmin;i<xmax;i++)
    {
    for(int j=ymin;j<ymax;j++)
    {
    w += w+disTemplate[xmin-x+int(ourArg[0])+i][ymin-y+int(ourArg[0])+i]*exp((orig[i][j][z]-orig[x][y][z])
    (orig[i][j][z]-orig[x][y][z])/(2*ourArg[3]*ourArg[3]));
    v += orig[x][y][z]*w;

    }
    }
    img[x][y][z] = v/w;

    }

    """)

    Gauss_cuda = mod.get_function("Gauss_cuda")
    bilateral_cuda = mod.get_function("bilateral_cuda")

    def generateGaussianTemplate(template_size,sigma):
    template = np.zeros((template_size,template_size))
    mid = template_size/2
    sum = 0
    pi = 3.1415926
    for i in range(template_size):
    x = pow(i-mid,2)
    for j in range(template_size):
    y = pow(j-mid,2)
    g = math.exp((x+y)/(-2*sigma*sigma))/(2*pi*sigma)
    template[i][j] = g
    sum+=g
    #归一化
    template = template/sum
    return template

    def Gauss_filter(img,template_size,sigma):
    [rows,cols,channel] = img.shape
    border = int(template_size/2)
    template = generateGaussianTemplate(template_size,sigma)
    im = cv2.copyMakeBorder(img,border,border,border,border,cv2.BORDER_REPLICATE)
    Gauss_cuda(
    drv.InOut(img), drv.In(im), drv.In(template),template_size,
    block=(template_size,template_size,3),grid=(rows,cols))
    return img

    def Bilateral_filter(img,template_size,sigma1,sigma2):
    img = img/255
    border = int(template_size/2)
    [rows,cols,channel] = img.shape
    tmp = np.arange(-border,border+1)
    [x,y] = np.meshgrid(tmp,tmp)
    X = np.power(x,2)
    Y = np.power(y,2)
    d = np.exp(-(X+Y)/(2*sigma1*sigma1))
    orig_img = img
    Gauss_cuda(
    drv.InOut(img), drv.In(orig_img), drv.In(d),drv.In(np.array([border,rows,cols,sigma2])),
    block=(template_size,template_size,3),grid=(rows,cols))
    return img*255

    if name=='__main__':
    img = cv2.imread('demo2.jpg')
    #noise = skimage.util.random_noise(img,'gaussian')*255
    #cv2.imwrite('noise.jpg',noise)
    template_size = 3
    sigma_g = 0.8
    sigma = [3,0.1]
    bilateral = Bilateral_filter(img,template_size,sigma[0],sigma[1])
    Gauss = Gauss_filter(img,template_size,sigma_g)
    cv2.imwrite('bilateral_result_{}_{}_{}.jpg'.format(template_size,sigma[0],sigma[1]),bilateral)
    cv2.imwrite('Gauss_result_{}_{}.jpg'.format(template_size,sigma_g),Gauss)

        试试这个!!!!!!!!!!!!!!!!!
    
    点赞 评论 复制链接分享

相关推荐