leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 27 Packages Projects Releases Wiki Activity

CUTLASS 2.1 (#83)

Browse Source 
CUTLASS 2.1 contributes:
- BLAS-style host-side API added to CUTLASS Library
- Planar Complex GEMM kernels targeting Volta and Turing Tensor Cores
- Minor enhancements and bug fixes
This commit is contained in:
Andrew Kerr
2020-04-07 13:51:25 -07:00
committed by GitHub
parent 7c0cd26d13
commit 96dab34ad9
196 changed files with 20653 additions and 1995 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/cutlass/layout/matrix.h
+6 -6
View File
@@ -98,7 +98,7 @@ public:
/// Assumes coordinate has convention (row, column)
CUTLASS_HOST_DEVICE
LongIndex operator()(MatrixCoord const &coord) const {
return coord.row() * stride_[0] + coord.column();
return LongIndex(coord.row()) * LongIndex(stride_[0]) + coord.column();
}
/// Inverse of layout function, mapping linear offset to logical coordinate
@@ -134,7 +134,7 @@ public:
/// Compute the number of contiguous elements needed to store a tensor with the given size
CUTLASS_HOST_DEVICE
LongIndex capacity(MatrixCoord const &extent) const {
return extent.row() * stride_[0];
return LongIndex(extent.row()) * LongIndex(stride_[0]);
}
};
@@ -191,7 +191,7 @@ public:
/// Assumes coordinate has convention (row, column)
CUTLASS_HOST_DEVICE
LongIndex operator()(MatrixCoord const &coord) const {
return coord.row() + coord.column() * stride_[0];
return LongIndex(coord.column()) * LongIndex(stride_[0]) + coord.row();
}
/// Inverse of layout function, mapping linear offset to logical coordinate
@@ -227,7 +227,7 @@ public:
/// Compute the number of contiguous elements needed to store a tensor with the given size
CUTLASS_HOST_DEVICE
LongIndex capacity(MatrixCoord const &extent) const {
return extent.column() * stride_[0];
return LongIndex(extent.column()) * LongIndex(stride_[0]);
}
};
@@ -290,7 +290,7 @@ public:
LongIndex operator()(MatrixCoord const &coord) const {
Index row_major = coord.row() / kInterleave;
Index row_minor = coord.row() % kInterleave;
return row_major * stride_[0] + coord.column() * kInterleave + row_minor;
return LongIndex(row_major) * LongIndex(stride_[0]) + LongIndex(coord.column()) * kInterleave + row_minor;
}
/// Inverse of layout function, mapping linear offset to logical coordinate
@@ -397,7 +397,7 @@ public:
LongIndex operator()(MatrixCoord const &coord) const {
Index column_major = coord.column() / kInterleave;
Index column_minor = coord.column() % kInterleave;
return column_major * stride_[0] + coord.row() * kInterleave + column_minor;
return LongIndex(column_major) * LongIndex(stride_[0]) + LongIndex(coord.row()) * kInterleave + column_minor;
}
/// Inverse of layout function, mapping linear offset to logical coordinate