leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 25 Packages Projects Releases Wiki Activity

CUTLASS 2.3 initial commit (#134)

Browse Source 
CUTLASS 2.3 adds GEMMs targeting Sparse Tensor Cores on the NVIDIA Ampere Architecture, fast SGEMM, and small matrix classes, bug fixes, and performance enhancements.
This commit is contained in:
Andrew Kerr
2020-09-23 14:00:58 -07:00
committed by GitHub
parent 4dac7490e6
commit c53f3339bb
209 changed files with 46922 additions and 1677 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/cutlass/integer_subbyte.h
+20 -3
View File
@@ -83,11 +83,10 @@ struct integer_subbyte {
integer_subbyte(unsigned value)
: storage(reinterpret_cast<Storage const &>(value) & kMask) {}
/// Conversion from double
CUTLASS_HOST_DEVICE
integer_subbyte(double value) {
T tmp = (T)value;
storage = reinterpret_cast<Storage const &>(tmp) & kMask;
T tmp = static_cast<T>(value);
storage = Storage(reinterpret_cast<unsigned const &>(tmp) & kMask);
}
///
@@ -155,6 +154,12 @@ struct integer_subbyte {
/// 1-bit Unsigned integer type
using uint1b_t = integer_subbyte<1, false>;
/// 2-bit Integer type
using int2b_t = integer_subbyte<2, true>;
/// 2-bit Unsigned integer type
using uint2b_t = integer_subbyte<2, false>;
/// 4-bit Integer type
using int4b_t = integer_subbyte<4, true>;
@@ -169,6 +174,18 @@ struct sizeof_bits<uint1b_t> {
static int const value = 1;
};
/// Defines the size of an element in bits - specialized for int2b_t
template <>
struct sizeof_bits<int2b_t> {
static int const value = 2;
};
/// Defines the size of an element in bits - specialized for uint2b_t
template <>
struct sizeof_bits<uint2b_t> {
static int const value = 2;
};
/// Defines the size of an element in bits - specialized for int4b_t
template <>
struct sizeof_bits<int4b_t> {