mirrors/faiss
1
0
Fork 0
mirror of https://github.com/facebookresearch/faiss.git synced 2025-06-03 21:54:02 +08:00
Code Issues Projects Releases Wiki Activity

fix issue 320

Browse Source
This commit is contained in:
Matthijs Douze 2018-01-31 07:00:24 -07:00
parent 4fe9204683
commit df2edbe3ab
2 changed files with 9 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
gpu/impl/BroadcastSum.cu
View File

@ -47,7 +47,7 @@ __global__ void sumAlongColumns(Tensor<T, 1, true> input,
if (endRow) { if (endRow) {
for (int row = rowStart; row < output.getSize(0); ++row) { for (int row = rowStart; row < output.getSize(0); ++row) {
T out = output[row][col].ldg(); T out = output[row][col];
out = Math<T>::add(out, val); out = Math<T>::add(out, val);
output[row][col] = out; output[row][col] = out;
} }
@ -57,7 +57,7 @@ __global__ void sumAlongColumns(Tensor<T, 1, true> input,
for (int row = rowStart; row < rowEnd; row += kRowUnroll) { for (int row = rowStart; row < rowEnd; row += kRowUnroll) {
#pragma unroll #pragma unroll
for (int i = 0; i < kRowUnroll; ++i) { for (int i = 0; i < kRowUnroll; ++i) {
rows[i] = output[row + i][col].ldg(); rows[i] = output[row + i][col];
} }
#pragma unroll #pragma unroll
@ -86,7 +86,7 @@ __global__ void sumAlongColumns(Tensor<T, 1, true> input,
for (int row = rowStart; row < output.getSize(0); ++row) { for (int row = rowStart; row < output.getSize(0); ++row) {
#pragma unroll #pragma unroll
for (int i = 0; i < kColLoad; ++i) { for (int i = 0; i < kColLoad; ++i) {
T out = output[row][col + i * blockDim.x].ldg(); T out = output[row][col + i * blockDim.x];
out = Math<T>::add(out, val[i]); out = Math<T>::add(out, val[i]);
output[row][col + i * blockDim.x] = out; output[row][col + i * blockDim.x] = out;
} }
@ -100,7 +100,7 @@ __global__ void sumAlongColumns(Tensor<T, 1, true> input,
#pragma unroll #pragma unroll
for (int j = 0; j < kColLoad; ++j) { for (int j = 0; j < kColLoad; ++j) {
rows[i * kColLoad + j] = rows[i * kColLoad + j] =
output[row + i][col + j * blockDim.x].ldg(); output[row + i][col + j * blockDim.x];
} }
} }

5
gpu/utils/Tensor-inl.cuh
View File

@ -310,6 +310,11 @@ Tensor<T, Dim, InnerContig, IndexT, PtrTraits>::canCastResize() const {
static_assert(sizeof(U) >= sizeof(T), "only handles greater sizes"); static_assert(sizeof(U) >= sizeof(T), "only handles greater sizes");
constexpr int kMultiple = sizeof(U) / sizeof(T); constexpr int kMultiple = sizeof(U) / sizeof(T);
// Ensure that the base pointer is sizeof(U) aligned
if (((uintptr_t) data_) % sizeof(U) != 0) {
return false;
}
// Check all outer strides // Check all outer strides
for (int i = 0; i < Dim - 1; ++i) { for (int i = 0; i < Dim - 1; ++i) {
if (stride_[i] % kMultiple != 0) { if (stride_[i] % kMultiple != 0) {