[CUDA]两个任意大小的矩阵相乘 [英] [CUDA] Multiplication of two Arbitrarily Sized Matrices

问题描述

将两个矩阵A [4x3]和B [3x4]相乘.
结果矩阵C [4x4].

每个块的并行操作数为4.
因此，
THREADS_PER_BLOCK = 4

Multiplying two Matrices A[4x3] and B[3x4].
Resultant Matrix C[4x4].

no of parallel operations per block is 4.
Therefore,
THREADS_PER_BLOCK = 4

dim3 dimBlock(THREADS_PER_BLOCK, THREADS_PER_BLOCK);
dim3 dimGrid(B.width/dimBlock.x, A.height/dimBlock.y);
MatrixMultKernel<<<dimGrid, dimBlock>>>(d_A, d_B, d_C);

CUDA报告无效的配置错误".在这方面需要帮助.

CUDA reports "invalid configuration error". need help in this regard.

推荐答案

使用CUBLAS，我们可以将两个任意大小的矩阵相乘. 有关更多详细信息，请 CUBLAS [ ^ ]

Using CUBLAS we can multiply two arbitrarily sized Matrices.
For more details CUBLAS[^]

尝试此链接 [ ^ ]

希望对您有帮助

try this link[^]

Hope it helps

这篇关于[CUDA]两个任意大小的矩阵相乘的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！