aparler/PyNorch
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

transpose tensor and matmul autograd

Browse source
This commit is contained in:
lucasdelimanogueira 2024-05-01 02:24:06 -03:00
parent 6aa18f0e36
commit f7cc7c8203
2 changed files with 2 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

4
norch/csrc/cuda.cu
View file

@ -57,7 +57,7 @@ __host__ void add_tensor_cuda(Tensor* tensor1, Tensor* tensor2, float* result_da
cudaDeviceSynchronize();
}
__global__ void sum_tensor_cuda_kernel(float* data, float* result_data, int size) {
_global__ void sum_tensor_cuda_kernel(float* data, float* result_data, int size) {
extern __shared__ float sdata[];
unsigned int tid = threadIdx.x;
@ -73,7 +73,7 @@ __global__ void sum_tensor_cuda_kernel(float* data, float* result_data, int size
// Reduce within block using shared memory
for (unsigned int s = blockDim.x / 2; s > 0; s >>= 1) {
if (tid < s) {
if (tid < s && i + s < size) {
sdata[tid] += sdata[tid + s];
}
__syncthreads();

1
norch/tensor.py
View file

@ -20,7 +20,6 @@ class Tensor:
if data != None:
data, shape = self.flatten(data)
# Adjust the path to the shared library
self.data_ctype = (ctypes.c_float * len(data))(*data)
self.shape_ctype = (ctypes.c_int * len(shape))(*shape)
self.ndim_ctype = ctypes.c_int(len(shape))