mirror of
https://github.com/matrix-construct/construct
synced 2024-06-02 18:18:56 +02:00
ircd::simd: Categorical interface renaming for stream related.
This commit is contained in:
Jason Volk
parent
de0650d3f7
commit
c49e546be5
|
|
@ -40,7 +40,7 @@ noexcept
|
|||
{
|
||||
const u64x2 res
|
||||
{
|
||||
stream<block_t>(in, max, [&val, &closure]
|
||||
for_each<block_t>(in, max, [&val, &closure]
|
||||
(const auto block, const auto mask)
|
||||
{
|
||||
closure(val, block, mask);
|
||||
|
|
|
|||
|
|
@ -9,58 +9,58 @@
|
|||
// full license for this software is available in the LICENSE file.
|
||||
|
||||
#pragma once
|
||||
#define HAVE_IRCD_SIMD_ISTREAM_H
|
||||
#define HAVE_IRCD_SIMD_FOR_EACH_H
|
||||
|
||||
namespace ircd::simd
|
||||
{
|
||||
template<class block_t>
|
||||
using istream_fixed_proto = void (block_t, block_t mask);
|
||||
using for_each_fixed_proto = void (block_t, block_t mask);
|
||||
|
||||
template<class block_t>
|
||||
using istream_variable_proto = u64x2 (block_t, block_t mask);
|
||||
using for_each_variable_proto = u64x2 (block_t, block_t mask);
|
||||
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
using istream_is_fixed_stride = std::is_same
|
||||
using for_each_is_fixed_stride = std::is_same
|
||||
<
|
||||
std::invoke_result_t<lambda, block_t, block_t>, void
|
||||
>;
|
||||
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
using istream_is_variable_stride = std::is_same
|
||||
using for_each_is_variable_stride = std::is_same
|
||||
<
|
||||
std::invoke_result_t<lambda, block_t, block_t>, u64x2
|
||||
>;
|
||||
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
using istream_fixed_stride = std::enable_if
|
||||
using for_each_fixed_stride = std::enable_if
|
||||
<
|
||||
istream_is_fixed_stride<block_t, lambda>::value, u64x2
|
||||
for_each_is_fixed_stride<block_t, lambda>::value, u64x2
|
||||
>;
|
||||
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
using istream_variable_stride = std::enable_if
|
||||
using for_each_variable_stride = std::enable_if
|
||||
<
|
||||
istream_is_variable_stride<block_t, lambda>::value, u64x2
|
||||
for_each_is_variable_stride<block_t, lambda>::value, u64x2
|
||||
>;
|
||||
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
typename istream_fixed_stride<block_t, lambda>::type
|
||||
stream(const block_t *, const u64x2, lambda&&) noexcept;
|
||||
typename for_each_fixed_stride<block_t, lambda>::type
|
||||
for_each(const block_t *, const u64x2, lambda&&) noexcept;
|
||||
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
typename istream_fixed_stride<block_t, lambda>::type
|
||||
stream(const char *, const u64x2, lambda&&) noexcept;
|
||||
typename for_each_fixed_stride<block_t, lambda>::type
|
||||
for_each(const char *, const u64x2, lambda&&) noexcept;
|
||||
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
typename istream_variable_stride<block_t, lambda>::type
|
||||
stream(const char *, const u64x2, lambda&&) noexcept;
|
||||
typename for_each_variable_stride<block_t, lambda>::type
|
||||
for_each(const char *, const u64x2, lambda&&) noexcept;
|
||||
}
|
||||
|
||||
/// Streaming consumer
|
||||
|
|
@ -90,10 +90,10 @@ namespace ircd::simd
|
|||
///
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
inline typename ircd::simd::istream_variable_stride<block_t, lambda>::type
|
||||
ircd::simd::stream(const char *const __restrict__ in,
|
||||
const u64x2 max,
|
||||
lambda&& closure)
|
||||
inline typename ircd::simd::for_each_variable_stride<block_t, lambda>::type
|
||||
ircd::simd::for_each(const char *const __restrict__ in,
|
||||
const u64x2 max,
|
||||
lambda&& closure)
|
||||
noexcept
|
||||
{
|
||||
using block_t_u = unaligned<block_t>;
|
||||
|
|
@ -172,10 +172,10 @@ noexcept
|
|||
///
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
inline typename ircd::simd::istream_fixed_stride<block_t, lambda>::type
|
||||
ircd::simd::stream(const char *const __restrict__ in,
|
||||
const u64x2 max,
|
||||
lambda&& closure)
|
||||
inline typename ircd::simd::for_each_fixed_stride<block_t, lambda>::type
|
||||
ircd::simd::for_each(const char *const __restrict__ in,
|
||||
const u64x2 max,
|
||||
lambda&& closure)
|
||||
noexcept
|
||||
{
|
||||
using block_t_u = unaligned<block_t>;
|
||||
|
|
@ -246,10 +246,10 @@ noexcept
|
|||
////
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
inline typename ircd::simd::istream_fixed_stride<block_t, lambda>::type
|
||||
ircd::simd::stream(const block_t *const __restrict__ in,
|
||||
const u64x2 max,
|
||||
lambda&& closure)
|
||||
inline typename ircd::simd::for_each_fixed_stride<block_t, lambda>::type
|
||||
ircd::simd::for_each(const block_t *const __restrict__ in,
|
||||
const u64x2 max,
|
||||
lambda&& closure)
|
||||
noexcept
|
||||
{
|
||||
u64x2 count
|
||||
|
|
@ -9,58 +9,58 @@
|
|||
// full license for this software is available in the LICENSE file.
|
||||
|
||||
#pragma once
|
||||
#define HAVE_IRCD_SIMD_OSTREAM_H
|
||||
#define HAVE_IRCD_SIMD_GENERATE_H
|
||||
|
||||
namespace ircd::simd
|
||||
{
|
||||
template<class block_t>
|
||||
using ostream_fixed_proto = void (block_t &, block_t &mask);
|
||||
using generate_fixed_proto = void (block_t &);
|
||||
|
||||
template<class block_t>
|
||||
using ostream_variable_proto = u64x2 (block_t &, block_t &mask);
|
||||
using generate_variable_proto = u64x2 (block_t &);
|
||||
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
using ostream_is_fixed_stride = std::is_same
|
||||
using generate_is_fixed_stride = std::is_same
|
||||
<
|
||||
std::invoke_result_t<lambda, block_t &, block_t &>, void
|
||||
std::invoke_result_t<lambda, block_t &>, void
|
||||
>;
|
||||
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
using ostream_is_variable_stride = std::is_same
|
||||
using generate_is_variable_stride = std::is_same
|
||||
<
|
||||
std::invoke_result_t<lambda, block_t &, block_t &>, u64x2
|
||||
std::invoke_result_t<lambda, block_t &>, u64x2
|
||||
>;
|
||||
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
using ostream_fixed_stride = std::enable_if
|
||||
using generate_fixed_stride = std::enable_if
|
||||
<
|
||||
ostream_is_fixed_stride<block_t, lambda>::value, u64x2
|
||||
generate_is_fixed_stride<block_t, lambda>::value, u64x2
|
||||
>;
|
||||
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
using ostream_variable_stride = std::enable_if
|
||||
using generate_variable_stride = std::enable_if
|
||||
<
|
||||
ostream_is_variable_stride<block_t, lambda>::value, u64x2
|
||||
generate_is_variable_stride<block_t, lambda>::value, u64x2
|
||||
>;
|
||||
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
typename ostream_fixed_stride<block_t, lambda>::type
|
||||
stream(block_t *, const u64x2, lambda&&) noexcept;
|
||||
typename generate_fixed_stride<block_t, lambda>::type
|
||||
generate(block_t *, const u64x2, lambda&&) noexcept;
|
||||
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
typename ostream_fixed_stride<block_t, lambda>::type
|
||||
stream(char *, const u64x2, lambda&&) noexcept;
|
||||
typename generate_fixed_stride<block_t, lambda>::type
|
||||
generate(char *, const u64x2, lambda&&) noexcept;
|
||||
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
typename ostream_variable_stride<block_t, lambda>::type
|
||||
stream(char *, const u64x2, lambda&&) noexcept;
|
||||
typename generate_variable_stride<block_t, lambda>::type
|
||||
generate(char *, const u64x2, lambda&&) noexcept;
|
||||
}
|
||||
|
||||
/// Streaming generator
|
||||
|
|
@ -73,9 +73,7 @@ namespace ircd::simd
|
|||
///
|
||||
/// * variable-stride: progress for each iteration of the loop across the
|
||||
/// output and buffer is not fixed; the transform function may advance the
|
||||
/// pointer one to sizeof(block_t) bytes each iteration. Due to these
|
||||
/// characteristics, unaligned bytes may be redundantly loaded and non-temporal
|
||||
/// features are not used to optimize the operation.
|
||||
/// pointer one to sizeof(block_t) bytes each iteration.
|
||||
///
|
||||
/// u64x2 counter lanes = { output_length, available_to_user }; The argument
|
||||
/// `max` gives the buffer size in that format. The return value is the
|
||||
|
|
@ -90,10 +88,10 @@ namespace ircd::simd
|
|||
///
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
inline typename ircd::simd::ostream_variable_stride<block_t, lambda>::type
|
||||
ircd::simd::stream(char *const __restrict__ out,
|
||||
const u64x2 max,
|
||||
lambda&& closure)
|
||||
inline typename ircd::simd::generate_variable_stride<block_t, lambda>::type
|
||||
ircd::simd::generate(char *const __restrict__ out,
|
||||
const u64x2 max,
|
||||
lambda&& closure)
|
||||
noexcept
|
||||
{
|
||||
using block_t_u = unaligned<block_t>;
|
||||
|
|
@ -113,11 +111,6 @@ noexcept
|
|||
// primary broadband loop
|
||||
while(produce[0] && count[0] + sizeof(block_t) <= max[0])
|
||||
{
|
||||
auto mask
|
||||
{
|
||||
~block_t{0}
|
||||
};
|
||||
|
||||
const auto di
|
||||
{
|
||||
reinterpret_cast<block_t_u *>(out + count[0])
|
||||
|
|
@ -128,16 +121,16 @@ noexcept
|
|||
*di
|
||||
);
|
||||
|
||||
produce = closure(block, mask);
|
||||
produce = closure(block);
|
||||
count += produce;
|
||||
}
|
||||
|
||||
// trailing narrowband loop
|
||||
while(produce[0] && count[0] < max[0])
|
||||
{
|
||||
block_t block {0}, mask {0};
|
||||
produce = closure(block, mask);
|
||||
for(size_t i(0); count[0] + i < max[0] && i < sizeof(block_t) && mask[i]; ++i)
|
||||
block_t block {0};
|
||||
produce = closure(block);
|
||||
for(size_t i(0); count[0] + i < max[0] && i < sizeof(block_t); ++i)
|
||||
out[count[0] + i] = block[i];
|
||||
|
||||
count += produce;
|
||||
|
|
@ -169,10 +162,10 @@ noexcept
|
|||
///
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
inline typename ircd::simd::ostream_fixed_stride<block_t, lambda>::type
|
||||
ircd::simd::stream(char *const __restrict__ out,
|
||||
const u64x2 max,
|
||||
lambda&& closure)
|
||||
inline typename ircd::simd::generate_fixed_stride<block_t, lambda>::type
|
||||
ircd::simd::generate(char *const __restrict__ out,
|
||||
const u64x2 max,
|
||||
lambda&& closure)
|
||||
noexcept
|
||||
{
|
||||
using block_t_u = unaligned<block_t>;
|
||||
|
|
@ -201,12 +194,7 @@ noexcept
|
|||
*di
|
||||
);
|
||||
|
||||
auto mask
|
||||
{
|
||||
~block_t{0}
|
||||
};
|
||||
|
||||
closure(block, mask);
|
||||
closure(block);
|
||||
count += produce;
|
||||
}
|
||||
|
||||
|
|
@ -214,11 +202,11 @@ noexcept
|
|||
assert(count[0] + sizeof(block_t) > max[0]);
|
||||
if(likely(count[0] < max[0]))
|
||||
{
|
||||
block_t block {0}, mask {0};
|
||||
closure(block, mask);
|
||||
block_t block;
|
||||
closure(block);
|
||||
|
||||
size_t i(0);
|
||||
for(; count[0] + i < max[0] && mask[i]; ++i)
|
||||
for(; count[0] + i < max[0]; ++i)
|
||||
out[count[0] + i] = block[i];
|
||||
|
||||
count += u64x2
|
||||
|
|
@ -241,10 +229,10 @@ noexcept
|
|||
////
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
inline typename ircd::simd::ostream_fixed_stride<block_t, lambda>::type
|
||||
ircd::simd::stream(block_t *const __restrict__ out,
|
||||
const u64x2 max,
|
||||
lambda&& closure)
|
||||
inline typename ircd::simd::generate_fixed_stride<block_t, lambda>::type
|
||||
ircd::simd::generate(block_t *const __restrict__ out,
|
||||
const u64x2 max,
|
||||
lambda&& closure)
|
||||
noexcept
|
||||
{
|
||||
u64x2 count
|
||||
|
|
@ -271,12 +259,7 @@ noexcept
|
|||
*di
|
||||
);
|
||||
|
||||
auto mask
|
||||
{
|
||||
~block_t{0}
|
||||
};
|
||||
|
||||
closure(block, mask);
|
||||
closure(block);
|
||||
count += produce;
|
||||
}
|
||||
|
||||
|
|
@ -32,9 +32,9 @@
|
|||
#include "lateral.h"
|
||||
#include "any.h"
|
||||
#include "all.h"
|
||||
#include "istream.h"
|
||||
#include "iostream.h"
|
||||
#include "ostream.h"
|
||||
#include "for_each.h"
|
||||
#include "transform.h"
|
||||
#include "generate.h"
|
||||
#include "accumulate.h"
|
||||
#include "print.h"
|
||||
|
||||
|
|
|
|||
|
|
@ -9,53 +9,53 @@
|
|||
// full license for this software is available in the LICENSE file.
|
||||
|
||||
#pragma once
|
||||
#define HAVE_IRCD_SIMD_IOSTREAM_H
|
||||
#define HAVE_IRCD_SIMD_TRANSFORM_H
|
||||
|
||||
namespace ircd::simd
|
||||
{
|
||||
template<class block_t>
|
||||
using iostream_fixed_proto = void (block_t &, block_t mask);
|
||||
using transform_fixed_proto = void (block_t &, block_t mask);
|
||||
|
||||
template<class block_t>
|
||||
using iostream_variable_proto = u64x2 (block_t &, block_t mask);
|
||||
using transform_variable_proto = u64x2 (block_t &, block_t mask);
|
||||
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
using iostream_is_fixed_stride = std::is_same
|
||||
using transform_is_fixed_stride = std::is_same
|
||||
<
|
||||
std::invoke_result_t<lambda, block_t &, block_t>, void
|
||||
>;
|
||||
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
using iostream_is_variable_stride = std::is_same
|
||||
using transform_is_variable_stride = std::is_same
|
||||
<
|
||||
std::invoke_result_t<lambda, block_t &, block_t>, u64x2
|
||||
>;
|
||||
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
using iostream_fixed_stride = std::enable_if
|
||||
using transform_fixed_stride = std::enable_if
|
||||
<
|
||||
iostream_is_fixed_stride<block_t, lambda>::value, u64x2
|
||||
transform_is_fixed_stride<block_t, lambda>::value, u64x2
|
||||
>;
|
||||
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
using iostream_variable_stride = std::enable_if
|
||||
using transform_variable_stride = std::enable_if
|
||||
<
|
||||
iostream_is_variable_stride<block_t, lambda>::value, u64x2
|
||||
transform_is_variable_stride<block_t, lambda>::value, u64x2
|
||||
>;
|
||||
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
typename iostream_fixed_stride<block_t, lambda>::type
|
||||
stream(char *, const char *, const u64x2, lambda&&) noexcept;
|
||||
typename transform_fixed_stride<block_t, lambda>::type
|
||||
transform(char *, const char *, const u64x2, lambda&&) noexcept;
|
||||
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
typename iostream_variable_stride<block_t, lambda>::type
|
||||
stream(char *, const char *, const u64x2, lambda&&) noexcept;
|
||||
typename transform_variable_stride<block_t, lambda>::type
|
||||
transform(char *, const char *, const u64x2, lambda&&) noexcept;
|
||||
}
|
||||
|
||||
/// Streaming transform
|
||||
|
|
@ -82,11 +82,11 @@ namespace ircd::simd
|
|||
///
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
inline typename ircd::simd::iostream_variable_stride<block_t, lambda>::type
|
||||
ircd::simd::stream(char *const __restrict__ out,
|
||||
const char *const __restrict__ in,
|
||||
const u64x2 max,
|
||||
lambda&& closure)
|
||||
inline typename ircd::simd::transform_variable_stride<block_t, lambda>::type
|
||||
ircd::simd::transform(char *const __restrict__ out,
|
||||
const char *const __restrict__ in,
|
||||
const u64x2 max,
|
||||
lambda&& closure)
|
||||
noexcept
|
||||
{
|
||||
using block_t_u = unaligned<block_t>;
|
||||
|
|
@ -178,11 +178,11 @@ noexcept
|
|||
///
|
||||
template<class block_t,
|
||||
class lambda>
|
||||
inline typename ircd::simd::iostream_fixed_stride<block_t, lambda>::type
|
||||
ircd::simd::stream(char *const __restrict__ out,
|
||||
const char *const __restrict__ in,
|
||||
const u64x2 max,
|
||||
lambda&& closure)
|
||||
inline typename ircd::simd::transform_fixed_stride<block_t, lambda>::type
|
||||
ircd::simd::transform(char *const __restrict__ out,
|
||||
const char *const __restrict__ in,
|
||||
const u64x2 max,
|
||||
lambda&& closure)
|
||||
noexcept
|
||||
{
|
||||
using block_t_u = unaligned<block_t>;
|
||||
|
|
@ -529,7 +529,7 @@ const
|
|||
|
||||
const auto count
|
||||
{
|
||||
simd::stream<block_t>(start + 1, max, each_block)
|
||||
simd::for_each<block_t>(start + 1, max, each_block)
|
||||
};
|
||||
|
||||
const bool ok
|
||||
|
|
@ -3456,7 +3456,7 @@ noexcept
|
|||
|
||||
const auto consumed
|
||||
{
|
||||
simd::stream<block_t>(ircd::data(buf), ircd::data(input), max, string_stringify)
|
||||
simd::transform<block_t>(ircd::data(buf), ircd::data(input), max, string_stringify)
|
||||
};
|
||||
|
||||
return consumed[0]; // output pos (bytes written)
|
||||
|
|
@ -3701,7 +3701,7 @@ noexcept
|
|||
|
||||
const auto count
|
||||
{
|
||||
simd::stream<block_t>(ircd::data(input), max, string_serialized)
|
||||
simd::for_each<block_t>(ircd::data(input), max, string_serialized)
|
||||
};
|
||||
|
||||
return count[0];
|
||||
|
|
|
|||
Loading…
Reference in a new issue