Fix bad dot group initialiser in HIP and CUDA
This commit is contained in:
parent
e347d2ff6c
commit
f2f7f3a3de
@ -253,7 +253,7 @@ __global__ void dot_kernel(const T * a, const T * b, T * sum, int array_size)
|
|||||||
int i = blockDim.x * blockIdx.x + threadIdx.x;
|
int i = blockDim.x * blockIdx.x + threadIdx.x;
|
||||||
const size_t local_i = threadIdx.x;
|
const size_t local_i = threadIdx.x;
|
||||||
|
|
||||||
tb_sum[local_i] = 0.0;
|
tb_sum[local_i] = {};
|
||||||
for (; i < array_size; i += blockDim.x*gridDim.x)
|
for (; i < array_size; i += blockDim.x*gridDim.x)
|
||||||
tb_sum[local_i] += a[i] * b[i];
|
tb_sum[local_i] += a[i] * b[i];
|
||||||
|
|
||||||
|
|||||||
@ -244,7 +244,7 @@ __global__ void dot_kernel(const T * a, const T * b, T * sum, int array_size)
|
|||||||
const size_t local_i = threadIdx.x;
|
const size_t local_i = threadIdx.x;
|
||||||
size_t i = blockDim.x * blockIdx.x + local_i;
|
size_t i = blockDim.x * blockIdx.x + local_i;
|
||||||
|
|
||||||
tb_sum[local_i]{};
|
tb_sum[local_i] = {};
|
||||||
for (; i < array_size; i += blockDim.x*gridDim.x)
|
for (; i < array_size; i += blockDim.x*gridDim.x)
|
||||||
tb_sum[local_i] += a[i] * b[i];
|
tb_sum[local_i] += a[i] * b[i];
|
||||||
|
|
||||||
|
|||||||
Loading…
Reference in New Issue
Block a user