gpt4 book ai didi

python - PyCUDA 使用字符串但不使用字符串数组

转载 作者:行者123 更新时间:2023-11-30 18:14:10 25 4
gpt4 key购买 nike

如何让 PyC​​uda 提取字符串数组而不是一个字符字符串?如果取消注释 C 代码中的该行,您将看到它迭代每个字符而不是每个字符串。

现在我只是想计算每个字符串的长度,但最终会将其变成一个词频计数器。第一步:传入一个数组...现在所需的输出应该是 25,27,44

import pycuda.driver as drv
import pycuda.tools
import pycuda.autoinit
import numpy
from pycuda.compiler import SourceModule

# create an array of 1s
lines = numpy.array(['ok this is the first line','number two line is this one','alright last line is in the third place here'])
lines = numpy.array(lines)
blocks = len(lines)
block_size = 1
nbr_values = blocks * block_size

# create a destination array that will receive the result
a = numpy.zeros(nbr_values).astype(numpy.float32)
dest = numpy.zeros_like(a)

######################
# SourceModele SECTION
mod = SourceModule("""
__global__ void gpusin(float *dest, char *lines)
{
const int i = blockDim.x*blockIdx.x + threadIdx.x;
dest[i] = sizeof (lines[i]);
//dest[i] = lines[i]; //uncomment this line to see that its iterating through individual chars not strings
}
""")

#Run the sourc model
gpusin = mod.get_function("gpusin")
gpusin(drv.Out(dest), drv.In(lines), grid=(blocks,1), block=(block_size,1,1) )
print str(dest)
print lines

最佳答案

import pycuda.driver as cuda
import pycuda.autoinit
from pycuda.compiler import SourceModule

import numpy

# create an array of 1s
data1 = ['Apple 12345 this is a long string again','Ding dong, it"s Pyton!!','digity! gigity!']
lines = numpy.array( data1, dtype=str)
linesGPU = cuda.mem_alloc(lines.size * lines.dtype.itemsize)
cuda.memcpy_htod(linesGPU, lines)
blocks = len(data1)
threadsPerBlock = lines.dtype.itemsize
nbr_values = lines.size * lines.dtype.itemsize # blocks * block_size
print("lines size: " + str(lines.size) + " itemsize : " + str(lines.dtype.itemsize))

# create a destination array that will receive the result
dest = numpy.zeros((nbr_values,), dtype=numpy.str)
destGPU = cuda.mem_alloc(dest.size * dest.dtype.itemsize)

mod = SourceModule("""
__global__ void process(char **dest, char **line)
{
int tID = threadIdx.x ;//+ blockIdx.x * blockDim.x;
dest[tID] = line[tID];
}
""")

#Run the sourc model
gpusin = mod.get_function("process")
gpusin(destGPU, linesGPU, grid=(blocks,1), block=(threadsPerBlock,1,1))
cuda.memcpy_dtoh(dest, destGPU)
print str(len(dest))
print dest

关于python - PyCUDA 使用字符串但不使用字符串数组,我们在Stack Overflow上找到一个类似的问题: https://stackoverflow.com/questions/18808279/

25 4 0
Copyright 2021 - 2024 cfsdn All Rights Reserved 蜀ICP备2022000587号
广告合作:1813099741@qq.com 6ren.com