Skip to content

Commit

Permalink
Merge pull request #4288 from Rohde-Schwarz/rene/simd_load_store
Browse files Browse the repository at this point in the history 
Load/Store for SIMD type wrappers
  • Loading branch information
@reneme
reneme authored Aug 13, 2024
2 parents 1499274 + 5c84388 commit 5f4c2c3
Show file tree
Hide file tree
Showing 3 changed files with 119 additions and 6 deletions.
81 changes: 75 additions & 6 deletions src/lib/utils/loadstor.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -167,12 +167,34 @@ constexpr bool native_endianness_is_unknown() {
#endif
}

/**
* Models a custom type that provides factory methods to be loaded in big- or
* little-endian byte order.
*/
template <typename T>
concept custom_loadable = requires(std::span<const uint8_t, sizeof(T)> data) {
{ T::load_be(data) } -> std::same_as<T>;
{ T::load_le(data) } -> std::same_as<T>;
};

/**
* Models a custom type that provides store methods to be stored in big- or
* little-endian byte order.
*/
template <typename T>
concept custom_storable = requires(std::span<uint8_t, sizeof(T)> data, const T value) {
{ value.store_be(data) };
{ value.store_le(data) };
};

/**
* Models a type that can be loaded/stored from/to a byte range.
*/
template <typename T>
concept unsigned_integralish = std::unsigned_integral<strong_type_wrapped_type<T>> ||
(std::is_enum_v<T> && std::unsigned_integral<std::underlying_type_t<T>>);
concept unsigned_integralish =
std::unsigned_integral<strong_type_wrapped_type<T>> ||
(std::is_enum_v<T> && std::unsigned_integral<std::underlying_type_t<T>>) ||
(custom_loadable<strong_type_wrapped_type<T>> || custom_storable<strong_type_wrapped_type<T>>);

template <typename T>
struct wrapped_type_helper_with_enum {
Expand Down Expand Up @@ -276,6 +298,7 @@ inline constexpr void fallback_store_any(InT in, OutR&& out_range) {
* @return T loaded from @p in_range, as a big-endian value
*/
template <Endianness endianness, unsigned_integralish WrappedOutT, ranges::contiguous_range<uint8_t> InR>
requires(!custom_loadable<strong_type_wrapped_type<WrappedOutT>>)
inline constexpr WrappedOutT load_any(InR&& in_range) {
using OutT = detail::wrapped_type<WrappedOutT>;
ranges::assert_exact_byte_length<sizeof(OutT)>(in_range);
Expand All @@ -302,6 +325,28 @@ inline constexpr WrappedOutT load_any(InR&& in_range) {
}());
}

/**
* Load a custom object from a range in either big or little endian byte order
*
* This is the base implementation for custom objects (e.g. SIMD type wrappres),
* all other overloads are just convenience overloads.
*
* @param in_range a fixed-length byte range
* @return T loaded from @p in_range, as a big-endian value
*/
template <Endianness endianness, unsigned_integralish WrappedOutT, ranges::contiguous_range<uint8_t> InR>
requires(custom_loadable<strong_type_wrapped_type<WrappedOutT>>)
inline constexpr WrappedOutT load_any(InR&& in_range) {
using OutT = detail::wrapped_type<WrappedOutT>;
ranges::assert_exact_byte_length<sizeof(OutT)>(in_range);
std::span<const uint8_t, sizeof(OutT)> ins{in_range};
if constexpr(endianness == Endianness::Big) {
return wrap_strong_type<WrappedOutT>(OutT::load_be(ins));
} else {
return wrap_strong_type<WrappedOutT>(OutT::load_le(ins));
}
}

/**
* Load many unsigned integers
* @param in a fixed-length span to some bytes
Expand Down Expand Up @@ -335,9 +380,9 @@ template <Endianness endianness,
(std::same_as<AutoDetect, OutT> || std::same_as<OutT, std::ranges::range_value_t<OutR>>))
inline constexpr void load_any(OutR&& out, InR&& in) {
ranges::assert_equal_byte_lengths(out, in);
using element_type = std::ranges::range_value_t<OutR>;

auto load_elementwise = [&] {
using element_type = std::ranges::range_value_t<OutR>;
constexpr size_t bytes_per_element = sizeof(element_type);
std::span<const uint8_t> in_s(in);
for(auto& out_elem : out) {
Expand All @@ -352,7 +397,7 @@ inline constexpr void load_any(OutR&& out, InR&& in) {
if(std::is_constant_evaluated()) /* TODO: C++23: if consteval {} */ {
load_elementwise();
} else {
if constexpr(is_native(endianness)) {
if constexpr(is_native(endianness) && !custom_loadable<element_type>) {
typecast_copy(out, in);
} else {
load_elementwise();
Expand Down Expand Up @@ -502,6 +547,7 @@ namespace detail {
* @param out_range a byte range to store the word into
*/
template <Endianness endianness, unsigned_integralish WrappedInT, ranges::contiguous_output_range<uint8_t> OutR>
requires(!custom_storable<strong_type_wrapped_type<WrappedInT>>)
inline constexpr void store_any(WrappedInT wrapped_in, OutR&& out_range) {
const auto in = detail::unwrap_strong_type_or_enum(wrapped_in);
using InT = decltype(in);
Expand All @@ -527,6 +573,29 @@ inline constexpr void store_any(WrappedInT wrapped_in, OutR&& out_range) {
}
}

/**
* Store a custom word in either big or little endian byte order into a range
*
* This is the base implementation for storing custom objects, all other
* overloads are just convenience overloads.
*
* @param wrapped_in a custom object to be stored
* @param out_range a byte range to store the word into
*/
template <Endianness endianness, unsigned_integralish WrappedInT, ranges::contiguous_output_range<uint8_t> OutR>
requires(custom_storable<strong_type_wrapped_type<WrappedInT>>)
inline constexpr void store_any(WrappedInT wrapped_in, OutR&& out_range) {
const auto in = detail::unwrap_strong_type_or_enum(wrapped_in);
using InT = decltype(in);
ranges::assert_exact_byte_length<sizeof(in)>(out_range);
std::span<uint8_t, sizeof(InT)> outs{out_range};
if constexpr(endianness == Endianness::Big) {
in.store_be(outs);
} else {
in.store_le(outs);
}
}

/**
* Store many unsigned integers words into a byte range
* @param out a sized range of some bytes
Expand Down Expand Up @@ -561,9 +630,9 @@ template <Endianness endianness,
requires(std::same_as<AutoDetect, InT> || std::same_as<InT, std::ranges::range_value_t<InR>>)
inline constexpr void store_any(OutR&& out, InR&& in) {
ranges::assert_equal_byte_lengths(out, in);
using element_type = std::ranges::range_value_t<InR>;

auto store_elementwise = [&] {
using element_type = std::ranges::range_value_t<InR>;
constexpr size_t bytes_per_element = sizeof(element_type);
std::span<uint8_t> out_s(out);
for(auto in_elem : in) {
Expand All @@ -578,7 +647,7 @@ inline constexpr void store_any(OutR&& out, InR&& in) {
if(std::is_constant_evaluated()) /* TODO: C++23: if consteval {} */ {
store_elementwise();
} else {
if constexpr(is_native(endianness)) {
if constexpr(is_native(endianness) && !custom_storable<element_type>) {
typecast_copy(out, in);
} else {
store_elementwise();
Expand Down
10 changes: 10 additions & 0 deletions src/lib/utils/simd/simd_32.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,8 @@

#include <botan/types.h>

#include <span>

#if defined(BOTAN_TARGET_SUPPORTS_SSE2)
#include <emmintrin.h>
#define BOTAN_SIMD_USE_SSE2
Expand Down Expand Up @@ -186,6 +188,10 @@ class SIMD_4x32 final {
#endif
}

static SIMD_4x32 load_le(std::span<const uint8_t, 16> in) { return SIMD_4x32::load_le(in.data()); }

static SIMD_4x32 load_be(std::span<const uint8_t, 16> in) { return SIMD_4x32::load_be(in.data()); }

void store_le(uint32_t out[4]) const noexcept { this->store_le(reinterpret_cast<uint8_t*>(out)); }

void store_be(uint32_t out[4]) const noexcept { this->store_be(reinterpret_cast<uint8_t*>(out)); }
Expand Down Expand Up @@ -246,6 +252,10 @@ class SIMD_4x32 final {
#endif
}

void store_be(std::span<uint8_t, 16> out) const { this->store_be(out.data()); }

void store_le(std::span<uint8_t, 16> out) const { this->store_le(out.data()); }

/*
* This is used for SHA-2/SHACAL2
*/
Expand Down
34 changes: 34 additions & 0 deletions src/tests/test_simd.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -12,6 +12,7 @@
#include <botan/internal/loadstor.h>
#include <botan/internal/rotate.h>
#include <botan/internal/simd_32.h>
#include <botan/internal/stl_util.h>
#endif

namespace Botan_Tests {
Expand Down Expand Up @@ -134,6 +135,39 @@ class SIMD_32_Tests final : public Test {
test_eq(result, "shift right 2", input.shift_elems_right<2>(), pat3, pat4, 0, 0);
test_eq(result, "shift right 3", input.shift_elems_right<3>(), pat4, 0, 0, 0);

// Test load/stores SIMD wrapper types
const auto simd_le_in = Botan::hex_decode("ABCDEF01234567890123456789ABCDEF");
const auto simd_be_in = Botan::hex_decode("0123456789ABCDEFABCDEF0123456789");
const auto simd_le_array_in = Botan::concat(simd_le_in, simd_be_in);
const auto simd_be_array_in = Botan::concat(simd_be_in, simd_le_in);

auto simd_le = Botan::load_le<Botan::SIMD_4x32>(simd_le_in);
auto simd_be = Botan::load_be<Botan::SIMD_4x32>(simd_be_in);
auto simd_le_array = Botan::load_le<std::array<Botan::SIMD_4x32, 2>>(simd_le_array_in);
auto simd_be_array = Botan::load_be<std::array<Botan::SIMD_4x32, 2>>(simd_be_array_in);

auto simd_le_vec = Botan::store_le<std::vector<uint8_t>>(simd_le);
auto simd_be_vec = Botan::store_be(simd_be);
auto simd_le_array_vec = Botan::store_le<std::vector<uint8_t>>(simd_le_array);
auto simd_be_array_vec = Botan::store_be(simd_be_array);

result.test_is_eq("roundtrip SIMD little-endian", simd_le_vec, simd_le_in);
result.test_is_eq(
"roundtrip SIMD big-endian", std::vector(simd_be_vec.begin(), simd_be_vec.end()), simd_be_in);
result.test_is_eq("roundtrip SIMD array little-endian", simd_le_array_vec, simd_le_array_in);
result.test_is_eq("roundtrip SIMD array big-endian",
std::vector(simd_be_array_vec.begin(), simd_be_array_vec.end()),
simd_be_array_in);

using StrongSIMD = Botan::Strong<Botan::SIMD_4x32, struct StrongSIMD_>;
const auto simd_le_strong = Botan::load_le<StrongSIMD>(simd_le_in);
const auto simd_be_strong = Botan::load_be<StrongSIMD>(simd_be_in);

result.test_is_eq(
"roundtrip SIMD strong little-endian", Botan::store_le<std::vector<uint8_t>>(simd_le_strong), simd_le_in);
result.test_is_eq(
"roundtrip SIMD strong big-endian", Botan::store_be<std::vector<uint8_t>>(simd_be_strong), simd_be_in);

return {result};
}

Expand Down

0 comments on commit 5f4c2c3

Please sign in to comment.