Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Refactor FloatPyCast for improved performance using lookup table #151

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Refactor FloatPyCast for improved performance using lookup table #151

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
35 changes: 28 additions & 7 deletions ml_dtypes/_src/dtypes.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -178,16 +178,37 @@ struct TypeDescriptor<uint4> : Int4TypeDescriptor<uint4> {
};

namespace {
template <typename From, typename To, bool IsTable>
struct FloatPyCaster {
static void Cast(void* from_void, void* to_void, npy_intp n) {
const auto* from = static_cast<From*>(from_void);
auto* to = static_cast<To*>(to_void);
for (npy_intp i = 0; i < n; ++i) {
to[i] = static_cast<To>(static_cast<float>(from[i]));
}
}
};

// Performs a NumPy array cast from type 'From' to 'To' via float.
template <typename From, typename To>
void FloatPyCast(void* from_void, void* to_void, npy_intp n, void* fromarr,
void* toarr) {
const auto* from = static_cast<From*>(from_void);
auto* to = static_cast<To*>(to_void);
for (npy_intp i = 0; i < n; ++i) {
to[i] = static_cast<To>(static_cast<float>(from[i]));
struct FloatPyCaster<From, To, true> {
static void Cast(void* from_void, void* to_void, npy_intp n) {
const auto* from = static_cast<From*>(from_void);
auto* to = static_cast<To*>(to_void);
To table[256];
// Use int for loop index to avoid overflow.
for (int i = 0; i < 256; ++i) {
table[i] = static_cast<To>(static_cast<float>(__builtin_bit_cast(From, uint8_t(i))));
}
for (npy_intp i = 0; i < n; ++i) {
to[i] = table[__builtin_bit_cast(uint8_t, from[i])];
}
}
};

template <typename From, typename To>
void FloatPyCast(void* from_void, void* to_void, npy_intp n, void* fromarr, void* toarr) {
constexpr bool is_table = (sizeof(From) == 1);
FloatPyCaster<From, To, is_table>::Cast(from_void, to_void, n);
}

template <typename Type1, typename Type2>
Expand Down