CUDA性能优化系列——Kmeans算法调优（一）

àì夳堔傛蜴生んèń 2022-11-04 01:49 176阅读 0赞

本篇主要介绍基于CUDA实现的Kmeans算法，以及通过性能分析工具Nsight Compute对CUDA程序性能调优，加速算法。不对Kmeans算法原理性的内容展开。

### CUDA实现思路 ###

主要分成两个核函数实现。第一部分基于当前的聚类中心点坐标，计算训练数据数据所属类别；第二部分基于新的类别信息，计算新的聚类中心坐标。本片主要介绍第一部分的CUDA实现与调优过程。  
参数说明：为了更好的阅读性，原始数据维度为1，数据大小为1024 \*1024 ，聚类中心个数为 16.

const int kSrcCount = 1024 * 1024;
    const int kClusterCount = 16;
    const int kCoords = 1;

### V1 LocalMemory版本 ###

Base版本，主要实现与 性能优化方法如下：  
内存分布：共享内存保存聚类中心坐标，寄存器中保存数据点。  
内存使用：保证加载设备内存满足合并对齐访问，同时加载共享内存满足广播机制。  
计算逻辑：一次计算数据点到各聚类中心欧式距离，并将该数据点类别标记为距离最短的聚类中心类别。

__global__ void kKmeansClassifyMembership(int SrcCount,
    	int Coords/*[1]*/,
    	int ClusterCount/*[16]*/,
    	float* d_pSrc, /*[Coords][SrcCount]*/
    	int* d_pMembership,/*[SrcCount]*/
    	float* d_pCluster, /*[ClusterCount][Coords]*/
    	int* d_pChanged/*[1] output*/
    )
    { 
    	extern __shared__ float sm_cluster[];/*[ClusterCount][Coords]*/
    	__shared__ int sm_blockchanged;
    	if (threadIdx.x == 0)
    	{ 
    		sm_blockchanged = 0;
    	}
    
    	float regData[4] = {  0.0f,0.0f, 0.0f, 0.0f };//这里虽然定义成reg，实际上使用的是LocalMemory
    	int tid = threadIdx.x + blockIdx.x * blockDim.x;
    	//TODO:load cluster center point to sm
    	for (int i = threadIdx.x; i < Coords * ClusterCount; i += blockDim.x)
    	{ 
    		sm_cluster[i] = d_pCluster[i];
    	}
    	__syncthreads();
    	//TODO:find nearest cluster center point
    	float minDistance = 10000000.0f;
    	int clusterIndex = 0;
    	if (tid < SrcCount)
    	{ 
    		for (size_t i = 0; i < Coords; i++)
    		{ 
    			regData[i] = d_pSrc[tid + i * SrcCount];
    		}
    		for (int itcluster = 0; itcluster < ClusterCount; itcluster++)
    		{ 
    			float curDistance = 0.0f;
    			for (int idim = 0; idim < Coords; idim++)
    			{ 
    				curDistance += pow(regData[idim] - sm_cluster[itcluster * Coords + idim], 2);
    			}
    			if (curDistance < minDistance)
    			{ 
    				minDistance = curDistance;
    				clusterIndex = itcluster;
    			}
    		}
    	}
    	
    	//TODO:find if membership changed
    	bool bchanged = (clusterIndex == d_pMembership[tid]) ? false : true;
    	d_pMembership[tid] = clusterIndex;
    	if (bchanged)
    	{ 
    		atomicAdd(&sm_blockchanged, 1);
    	}
    	__syncthreads();
    	if (threadIdx.x == 0)
    	{ 
    		atomicAdd(d_pChanged, sm_blockchanged);
    	}
    }

执行情况如下

![在这里插入图片描述][20210323002013900.png]  
核心和内存使用率都比较低，执行时间为243us。  
![在这里插入图片描述][20210323001501724.png]  
调整block grid数量后耗时170us

![在这里插入图片描述][20210323001857451.png]  
将grid增大8倍后，核心和内存使用率得到提升。  
![在这里插入图片描述][20210323001907240.png]  
V1版本主要性能问题由加载内存延迟导致。瓶颈代码位置是比较数据点所属类别是否发生变化，如下：

bool bchanged = (clusterIndex == d_pMembership[tid]) ? false : true;

此外，V1版本在设计时regData是想保存在寄存器中，而编译器把它放在了局部内存，不符合预期。

### V2 提前加载数据点类别 ###

针对V1版本中的性能问题，将数据点的类别信息提前加载到寄存器中

/*V2，提前加载分类信息，修复stall_long_sb指标高问题*/
    __global__ void kKmeansClassifyMembershipPrefetch(int SrcCount,
    	int Coords/*[1]*/,
    	int ClusterCount/*[16]*/,
    	float* d_pSrc, /*[Coords][SrcCount]*/
    	int* d_pMembership,/*[SrcCount]*/
    	float* d_pCluster, /*[ClusterCount][Coords]*/
    	int* d_pChanged/*[1] output*/
    )
    { 
    	extern __shared__ float sm_cluster[];/*[ClusterCount][Coords]*/
    	__shared__ int sm_blockchanged;
    	if (threadIdx.x == 0)
    	{ 
    		sm_blockchanged = 0;
    	}
    
    	float regData[4] = {  0.0f,0.0f, 0.0f, 0.0f };//这里虽然定义成reg，实际上使用的是LocalMemory
    	int tid = threadIdx.x + blockIdx.x * blockDim.x;
    	//TODO:load cluster center point to sm
    	for (int i = threadIdx.x; i < Coords * ClusterCount; i += blockDim.x)
    	{ 
    		sm_cluster[i] = d_pCluster[i];
    	}
    	int preMembership = d_pMembership[tid];
    	__syncthreads();
    	//TODO:find nearest cluster center point
    	float minDistance = 10000000.0f;
    	int clusterIndex = 0;
    	if (tid < SrcCount)
    	{ 
    		for (size_t i = 0; i < Coords; i++)
    		{ 
    			regData[i] = d_pSrc[tid + i * SrcCount];
    		}
    		for (int itcluster = 0; itcluster < ClusterCount; itcluster++)
    		{ 
    			float curDistance = 0.0f;
    			for (int idim = 0; idim < Coords; idim++)
    			{ 
    				curDistance += pow(regData[idim] - sm_cluster[itcluster * Coords + idim], 2);
    			}
    			if (curDistance < minDistance)
    			{ 
    				minDistance = curDistance;
    				clusterIndex = itcluster;
    			}
    		}
    	}
    	
    	//TODO:find if membership changed
    	bool bchanged = (clusterIndex == preMembership) ? false : true;
    	d_pMembership[tid] = clusterIndex;
    	if (bchanged)
    	{ 
    		atomicAdd(&sm_blockchanged, 1);
    	}
    	__syncthreads();
    	if (threadIdx.x == 0)
    	{ 
    		atomicAdd(d_pChanged, sm_blockchanged);
    	}
    }

执行情况如下，耗时118us  
![在这里插入图片描述][20210323003001641.png]  
优化到这步，核心的使用率升高到峰值的70% +。  
![在这里插入图片描述][2021032300304549.png]

V2版本的主要性能问题是stall\_wait，代码位置如下，下一个版本尝试通过编译器优化去掉循环

for (int idim = 0; idim < Coords; idim++)
    { 
    	curDistance += pow(regData[idim] - sm_cluster[itcluster * Coords + idim], 2);
    }

### V3 模板与编译器展开 ###

前文已经做过说明，数据的维度固定为1，因此V2版本性能瓶颈的位置没必要做循环处理。尝试通过编译器展开技术优化，模板参数COORDS代表数据维度，代码如下

template<int COORDS>
    __global__ void kKmeansClassifyMembershipTmpl(int SrcCount,
    	int Coords/*[1]*/,
    	int ClusterCount/*[16]*/,
    	float* d_pSrc, /*[Coords][SrcCount]*/
    	int* d_pMembership,/*[SrcCount]*/
    	float* d_pCluster, /*[ClusterCount][Coords]*/
    	int* d_pChanged/*[1] output*/
    )
    { 
    	extern __shared__ float sm_cluster[];/*[ClusterCount][Coords]*/
    	__shared__ int sm_blockchanged;
    	if (threadIdx.x == 0)
    	{ 
    		sm_blockchanged = 0;
    	}
    
    	float regData[4] = {  0.0f,0.0f, 0.0f, 0.0f };//这里数据保存在寄存器
    	int tid = threadIdx.x + blockIdx.x * blockDim.x;
    	//TODO:load cluster center point to sm
    	for (int i = threadIdx.x; i < Coords * ClusterCount; i += blockDim.x)
    	{ 
    		sm_cluster[i] = d_pCluster[i];
    	}
    	int preMembership = d_pMembership[tid];
    	__syncthreads();
    	//TODO:find nearest cluster center point
    	float minDistance = 10000000.0f;
    	int clusterIndex = 0;
    	if (tid < SrcCount)
    	{ 
    #pragma unroll
    		for (size_t i = 0; i < COORDS; i++)
    		{ 
    			regData[i] = d_pSrc[tid + i * SrcCount];
    		}
    		for (int itcluster = 0; itcluster < ClusterCount; itcluster++)
    		{ 
    			float curDistance = 0.0f;
    #pragma unroll
    			for (int idim = 0; idim < COORDS; idim++)
    			{ 
    				curDistance += pow(regData[idim] - sm_cluster[itcluster * Coords + idim], 2);
    			}
    			if (curDistance < minDistance)
    			{ 
    				minDistance = curDistance;
    				clusterIndex = itcluster;
    			}
    		}
    	}
    	
    	//TODO:find if membership changed
    	bool bchanged = (clusterIndex == preMembership) ? false : true;
    	d_pMembership[tid] = clusterIndex;
    	if (bchanged)
    	{ 
    		atomicAdd(&sm_blockchanged, 1);
    	}
    	__syncthreads();
    	if (threadIdx.x == 0)
    	{ 
    		atomicAdd(d_pChanged, sm_blockchanged);
    	}
    }

执行情况如下，耗时45us  
![在这里插入图片描述][2021032300434683.png]  
V3版本的核心和内存的负载非常均衡，且都达到了峰值的70%。  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3lhbjMxNDE1_size_16_color_FFFFFF_t_70]  
优化到V3版本，Nsight Compute对于warp的执行状态已经不再提示问题。同时对比V1版本，通过模板与展开技术，编译器将regData放在了寄存器中，这也是执行速度对比V2提升近一倍的原因之一。

### V4 2 ^ n倍展开 ###

优化到V3版本已经无明确的性能问题提示，这里尝试通过2 ^ n 倍展开，观察是否能进一步减速算法。经验证，4倍展开性能最佳，即模板参数UNROLLSCALE=4。

/*V4，按照聚类数量展开,收益不大;在最外层逻辑独立展开4倍，快20%*/
    template<int COORDS,int UNROLLSCALE>
    __global__ void kKmeansClassifyMembershipTAdv(int SrcCount,
    	int Coords/*[1]*/,
    	int ClusterCount/*[16]*/,
    	float* d_pSrc, /*[Coords][SrcCount]*/
    	int* d_pMembership,/*[SrcCount]*/
    	float* d_pCluster, /*[ClusterCount][Coords]*/
    	int* d_pChanged/*[1] output*/
    )
    { 
    	extern __shared__ float sm_cluster[];/*[ClusterCount][Coords]*/
    	__shared__ int sm_blockchanged;
    	if (threadIdx.x == 0)
    	{ 
    		sm_blockchanged = 0;
    	}
    	float regData[UNROLLSCALE];
    	int unrollsize = blockDim.x * gridDim.x;
    	int tid = threadIdx.x + blockIdx.x * blockDim.x;
    	//TODO:load cluster center point to sm#pragma unroll
    	for (int i = threadIdx.x; i < COORDS * ClusterCount; i += blockDim.x)
    	{ 
    		sm_cluster[i] = d_pCluster[i];
    	}
    	int preMembership[UNROLLSCALE];
    #pragma unroll
    	for (size_t unroll = 0; unroll < UNROLLSCALE; unroll++)
    	{ 
    		preMembership[unroll] = d_pMembership[tid + unroll * unrollsize];
    	}
    	__syncthreads();
    	//TODO:find nearest cluster center point
    	float minDistance[UNROLLSCALE];
    #pragma unroll
    	for (size_t i = 0; i < UNROLLSCALE; i++)
    	{ 
    		minDistance[i] = 10000000.0f;
    	}
    	int clusterIndex[UNROLLSCALE];
    	if (tid < SrcCount)
    	{ 
    #pragma unroll
    		for (size_t unroll = 0; unroll < UNROLLSCALE; unroll++)
    		{ 
    #pragma unroll
    			for (size_t idim = 0; idim < COORDS; idim++)
    			{ 
    				regData[idim+unroll* COORDS] = d_pSrc[tid + unroll * unrollsize + idim * SrcCount];
    			}
    		}
    #pragma unroll
    		for (size_t unroll = 0; unroll < UNROLLSCALE; unroll++)
    		{ 
    			for (int itcluster = 0; itcluster < ClusterCount; itcluster++)
    			{ 
    				float curDistance = 0.0f;
    #pragma unroll
    				for (int idim = 0; idim < COORDS; idim++)
    				{ 
    					curDistance += pow(regData[idim + unroll * COORDS] - sm_cluster[itcluster * COORDS + idim], 2);
    				}
    				if (curDistance < minDistance[unroll])
    				{ 
    					minDistance[unroll] = curDistance;
    					clusterIndex[unroll] = itcluster;
    				}
    			}
    		}
    	}
    	//TODO:find if membership changed
    	bool bchanged = false;
    #pragma unroll
    	for (size_t unroll = 0; unroll < UNROLLSCALE; unroll++)
    	{ 
    		bchanged |= (clusterIndex[unroll] == preMembership[unroll]) ? false : true;
    		d_pMembership[tid+ unroll * unrollsize] = clusterIndex[unroll];
    	}
    	if (bchanged)
    	{ 
    		atomicAdd(&sm_blockchanged, 1);
    	}
    	__syncthreads();
    	if (threadIdx.x == 0)
    	{ 
    		atomicAdd(d_pChanged, sm_blockchanged);
    	}
    }

4倍展开时，执行情况如下  
![在这里插入图片描述][20210323223149517.png]  
此时运行时间为36us  
![在这里插入图片描述][20210323223231618.png]  
由于做了4倍展开block数量由变为2048

### V5进一步展开 ###

尝试在聚类个数这个维度对应的for循环上进行展开，经验证2倍展开性能最佳，即模板参数CLUSTERUNROLL=2.

/*V5，3个维度全部展开*/
    template<int COORDS, int UNROLLSCALE,int CLUSTERS,int CLUSTERUNROLL>
    __global__ void kKmeansClassifyMembershipTUnrollAll(int SrcCount,
    	int Coords/*[1]*/,
    	int ClusterCount/*[16]*/,
    	float* d_pSrc, /*[Coords][SrcCount]*/
    	int* d_pMembership,/*[SrcCount]*/
    	float* d_pCluster, /*[ClusterCount][Coords]*/
    	int* d_pChanged/*[1] output*/
    )
    { 
    	extern __shared__ float sm_cluster[];/*[ClusterCount][Coords]*/
    	__shared__ int sm_blockchanged;
    	if (threadIdx.x == 0)
    	{ 
    		sm_blockchanged = 0;
    	}
    	float regData[UNROLLSCALE];
    	int unrollsize = blockDim.x * gridDim.x;
    	int tid = threadIdx.x + blockIdx.x * blockDim.x;
    	//TODO:load cluster center point to sm#pragma unroll
    #pragma unroll
    	for (int i = threadIdx.x; i < COORDS * CLUSTERS; i += blockDim.x)
    	{ 
    		sm_cluster[i] = d_pCluster[i];
    	}
    	int preMembership[UNROLLSCALE];
    #pragma unroll
    	for (size_t unroll = 0; unroll < UNROLLSCALE; unroll++)
    	{ 
    		preMembership[unroll] = d_pMembership[tid + unroll * unrollsize];
    	}
    	__syncthreads();
    	//TODO:find nearest cluster center point
    	float minDistance[UNROLLSCALE];
    #pragma unroll
    	for (size_t i = 0; i < UNROLLSCALE; i++)
    	{ 
    		minDistance[i] = 10000000.0f;
    	}
    	int clusterIndex[UNROLLSCALE];
    	if (tid < SrcCount)
    	{ 
    #pragma unroll
    		for (size_t unroll = 0; unroll < UNROLLSCALE; unroll++)
    		{ 
    #pragma unroll
    			for (size_t idim = 0; idim < COORDS; idim++)
    			{ 
    				regData[idim + unroll * COORDS] = d_pSrc[tid + unroll * unrollsize + idim * SrcCount];
    			}
    		}
    #pragma unroll
    		for (size_t unroll = 0; unroll < UNROLLSCALE; unroll++)
    		{ 
    			for (int itclusterOut = 0; itclusterOut < CLUSTERS/CLUSTERUNROLL; itclusterOut++)
    			{ 
    #pragma unroll
    				for (size_t itclusterIn = 0; itclusterIn < CLUSTERUNROLL; itclusterIn++)
    				{ 
    					float curDistance = 0.0f;
    #pragma unroll
    					for (int idim = 0; idim < COORDS; idim++)
    					{ 
    						curDistance += pow(regData[idim + unroll * COORDS] - sm_cluster[(itclusterOut* CLUSTERUNROLL + itclusterIn) * COORDS + idim], 2);
    					}
    					if (curDistance < minDistance[unroll])
    					{ 
    						minDistance[unroll] = curDistance;
    						clusterIndex[unroll] = itclusterOut * CLUSTERUNROLL + itclusterIn;
    					}
    				}
    			}
    		}
    	}
    	//TODO:find if membership changed
    	bool bchanged = false;
    #pragma unroll
    	for (size_t unroll = 0; unroll < UNROLLSCALE; unroll++)
    	{ 
    		bchanged |= (clusterIndex[unroll] == preMembership[unroll]) ? false : true;
    		d_pMembership[tid + unroll * unrollsize] = clusterIndex[unroll];
    	}
    	if (bchanged)
    	{ 
    		atomicAdd(&sm_blockchanged, 1);//直接加1而不使用atomic原则上对最终的效果没影响
    	}
    	__syncthreads();
    	if (threadIdx.x == 0)
    	{ 
    		atomicAdd(d_pChanged, sm_blockchanged);
    	}
    }

执行情况  
![在这里插入图片描述][20210323224839615.png]  
执行速度得到进一步提升，耗时30us。但是过度展开引发了计算和内存加载二者之间不平衡的问题。如果只从均衡性的角度来看，V5版本进一步进行展开不如V4版本。

### V6 消除原子操作Atomic ###

之前的所有版本都是通过原子操作写入设备内存中，以标记本次迭代计算中数据点所属类别是否发生改变。实际上该块设备内存仅需包含所属分类是否变化的信息，无需统计出当前迭代有多少数据点发生了变化，因此没必要通过原子操作统计出确切的数值，只要核函数之前初始化为0，数据点类别变化所对应线程修改该块内存即可。

/*V6，去除atomic*/
    template<int COORDS, int UNROLLSCALE, int CLUSTERS, int CLUSTERUNROLL>
    __global__ void kKmeansClassifyMembershipRMAtomic(int SrcCount,
    	int Coords/*[1]*/,
    	int ClusterCount/*[16]*/,
    	float* d_pSrc, /*[Coords][SrcCount]*/
    	int* d_pMembership,/*[SrcCount]*/
    	float* d_pCluster, /*[ClusterCount][Coords]*/
    	int* d_pChanged/*[1] output*/
    )
    { 
    	extern __shared__ float sm_cluster[];/*[ClusterCount][Coords]*/
    	__shared__ int sm_blockchanged;
    	if (threadIdx.x == 0)
    	{ 
    		sm_blockchanged = 0;
    	}
    
    	float regData[UNROLLSCALE];
    	int unrollsize = blockDim.x * gridDim.x;
    	int tid = threadIdx.x + blockIdx.x * blockDim.x;
    	//TODO:load cluster center point to sm#pragma unroll
    #pragma unroll
    	for (int i = threadIdx.x; i < COORDS * CLUSTERS; i += blockDim.x)
    	{ 
    		sm_cluster[i] = d_pCluster[i];
    	}
    	int preMembership[UNROLLSCALE];
    #pragma unroll
    	for (size_t unroll = 0; unroll < UNROLLSCALE; unroll++)
    	{ 
    		preMembership[unroll] = d_pMembership[tid + unroll * unrollsize];
    	}
    	__syncthreads();
    	//TODO:find nearest cluster center point
    	float minDistance[UNROLLSCALE];
    #pragma unroll
    	for (size_t i = 0; i < UNROLLSCALE; i++)
    	{ 
    		minDistance[i] = 10000000.0f;
    	}
    	int clusterIndex[UNROLLSCALE];
    	if (tid < SrcCount)
    	{ 
    #pragma unroll
    		for (size_t unroll = 0; unroll < UNROLLSCALE; unroll++)
    		{ 
    #pragma unroll
    			for (size_t idim = 0; idim < COORDS; idim++)
    			{ 
    				regData[idim + unroll * COORDS] = d_pSrc[tid + unroll * unrollsize + idim * SrcCount];
    			}
    		}
    
    #pragma unroll
    		for (size_t unroll = 0; unroll < UNROLLSCALE; unroll++)
    		{ 
    #pragma unroll
    			for (int itclusterOut = 0; itclusterOut < CLUSTERS / CLUSTERUNROLL; itclusterOut++)
    			{ 
    				for (size_t itclusterIn = 0; itclusterIn < CLUSTERUNROLL; itclusterIn++)
    				{ 
    					float curDistance = 0.0f;
    #pragma unroll
    					for (int idim = 0; idim < COORDS; idim++)
    					{ 
    						curDistance += pow(regData[idim + unroll * COORDS] - sm_cluster[(itclusterOut * CLUSTERUNROLL + itclusterIn) * COORDS + idim], 2);
    					}
    					if (curDistance < minDistance[unroll])
    					{ 
    						minDistance[unroll] = curDistance;
    						clusterIndex[unroll] = itclusterOut * CLUSTERUNROLL + itclusterIn;
    					}
    				}
    
    
    
    			}
    		}
    
    	}
    	//TODO:find if membership changed
    	bool bchanged = false;
    #pragma unroll
    	for (size_t unroll = 0; unroll < UNROLLSCALE; unroll++)
    	{ 
    		bchanged |= (clusterIndex[unroll] == preMembership[unroll]) ? false : true;
    		d_pMembership[tid + unroll * unrollsize] = clusterIndex[unroll];
    	}
    
    	if (bchanged)
    	{ 
    		sm_blockchanged = 1;
    	}
    	__syncthreads();
    	if (threadIdx.x == 0 && sm_blockchanged == 1)
    	{ 
    		*d_pChanged = 1;
    	}
    }

执行情况如下  
![在这里插入图片描述][20210323225917651.png]  
V6版本的执行速度相对V5并没有进一步提升，本人猜测可能是数据点分布初始化设置的不够合理，导致这一步优化没有使算法进一步加速。

### 总结 ###

本篇基于CUDA实现了Kmeans算法，并进行性能调优。通过6个版本的迭代，使执行速度从243us下降至30us，得到8倍加速比。  
![在这里插入图片描述][20210323230603671.png]

[20210323002013900.png]: /images/20221023/8fe3d619e2c5457eafb70e6db6fb2ed9.png
[20210323001501724.png]: /images/20221023/0b57e5e3edd24a3e88a066b32e7cf4a8.png
[20210323001857451.png]: /images/20221023/1c65feda54884e30b0d6ac4591821283.png
[20210323001907240.png]: /images/20221023/dc4b6c52c9f54464b84ae577cd90aa49.png
[20210323003001641.png]: /images/20221023/336a930a52874ecbbde96e0ac717fd92.png
[2021032300304549.png]: /images/20221023/44b923474805477283bfc839821705dc.png
[2021032300434683.png]: /images/20221023/9fe7ce093a9d48fb9145963b552cb226.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3lhbjMxNDE1_size_16_color_FFFFFF_t_70]: /images/20221023/7d59d406a4924b068a955477e0aaf44d.png
[20210323223149517.png]: /images/20221023/1d12765e5f274e538a7d561f2eb19f80.png
[20210323223231618.png]: /images/20221023/f5a22c9e6664410e979f3e80037d2d4e.png
[20210323224839615.png]: /images/20221023/2574acfb1fc84191a88ae277f09c200a.png
[20210323225917651.png]: /images/20221023/edc4e1a41ab1451eb5bcfdb1d7bd04d2.png
[20210323230603671.png]: /images/20221023/41a580d0d3d74282870f34cdfcb2b40d.png