33d6bda86d
Differential Revision: https://phabricator.services.mozilla.com/D115697 |
||
---|---|---|
.. | ||
debian | ||
hwy | ||
BUILD | ||
CMakeLists.txt | ||
CMakeLists.txt.in | ||
CONTRIBUTING | ||
libhwy-contrib.pc.in | ||
libhwy-test.pc.in | ||
libhwy.pc.in | ||
LICENSE | ||
README.md | ||
run_tests.bat | ||
run_tests.sh | ||
test.py | ||
WORKSPACE |
Efficient and performance-portable SIMD
Highway is a C++ library for SIMD (Single Instruction, Multiple Data), i.e. applying the same operation to 'lanes'.
Why Highway?
- more portable (same source code) than platform-specific intrinsics,
- works on a wider range of compilers than compiler-specific vector extensions,
- more dependable than autovectorization,
- easier to write/maintain than assembly language,
- supports runtime dispatch,
- supports variable-length vector architectures.
Current status
Supported targets: scalar, SSE4, AVX2, AVX-512, NEON (ARMv7 and v8), WASM SIMD. Ports to RVV and SVE/SVE2 are in progress.
Version 0.11 is considered stable enough to use in other projects, and is expected to remain backwards compatible unless serious issues are discovered while implementing SVE/RVV targets. After these targets are added, Highway will reach version 1.0.
Continuous integration tests build with a recent version of Clang (running on x86 and QEMU for ARM) and MSVC from VS2015 (running on x86).
Before releases, we also test on x86 with Clang and GCC, and ARMv7/8 via GCC cross-compile and QEMU. See the testing process for details.
The contrib
directory contains SIMD-related utilities: an image class with
aligned rows, and a math library (16 functions already implemented, mostly
trigonometry).
Installation
This project uses cmake to generate and build. In a Debian-based system you can install it via:
sudo apt install cmake
Highway's unit tests use googletest.
By default, Highway's CMake downloads this dependency at configuration time.
You can disable this by setting the HWY_SYSTEM_GTEST
CMake variable to ON and
installing gtest separately:
sudo apt install libgtest-dev
To build and test the library the standard cmake workflow can be used:
mkdir -p build && cd build
cmake ..
make -j && make test
Or you can run run_tests.sh
(run_tests.bat
on Windows).
To test on all the attainable targets for your platform, use
cmake .. -DCMAKE_CXX_FLAGS="-DHWY_COMPILE_ALL_ATTAINABLE"
. Otherwise, the
default configuration skips baseline targets (e.g. scalar) that are superseded
by another baseline target.
Bazel is also supported for building, but it is not as widely used/tested.
Quick start
You can use the benchmark
inside examples/ as a starting point.
A quick-reference page briefly lists all operations and their parameters, and the instruction_matrix indicates the number of instructions per operation.
We recommend using full SIMD vectors whenever possible for maximum performance
portability. To obtain them, pass a HWY_FULL(float)
tag to functions such as
Zero/Set/Load
. There is also the option of a vector of up to N
(a power of
two) lanes: HWY_CAPPED(T, N)
. 128-bit vectors are guaranteed to be available
for lanes of type T
if HWY_TARGET != HWY_SCALAR
and N == 16 / sizeof(T)
.
Functions using Highway must be inside a namespace namespace HWY_NAMESPACE {
(possibly nested in one or more other namespaces defined by the project), and
additionally either prefixed with HWY_ATTR
, or residing between
HWY_BEFORE_NAMESPACE()
and HWY_AFTER_NAMESPACE()
.
-
For static dispatch,
HWY_TARGET
will be the best available target amongHWY_BASELINE_TARGETS
, i.e. those allowed for use by the compiler (see quick-reference). Functions insideHWY_NAMESPACE
can be called usingHWY_STATIC_DISPATCH(func)(args)
within the same module they are defined in. You can call the function from other modules by wrapping it in a regular function and declaring the regular function in a header. -
For dynamic dispatch, a table of function pointers is generated via the
HWY_EXPORT
macro that is used byHWY_DYNAMIC_DISPATCH(func)(args)
to call the best function pointer for the current CPU supported targets. A module is automatically compiled for each target inHWY_TARGETS
(see quick-reference) ifHWY_TARGET_INCLUDE
is defined and foreach_target.h is included.
Strip-mining loops
To vectorize a loop, "strip-mining" transforms it into an outer loop and inner loop with number of iterations matching the preferred vector width.
In this section, let T
denote the element type, d = HWY_FULL(T)
, count
the
number of elements to process, and N = Lanes(d)
the number of lanes in a full
vector. Assume the loop body is given as a function template<bool partial, class D> void LoopBody(D d, size_t max_n)
.
Highway offers several ways to express loops where N
need not divide count
:
-
Ensure all inputs/outputs are padded. Then the loop is simply
for (size_t i = 0; i < count; i += N) LoopBody<false>(d, 0);
Here, the template parameter and second function argument are not needed.
This is the preferred option, unless
N
is in the thousands and vector operations are pipelined with long latencies. This was the case for supercomputers in the 90s, but nowadays ALUs are cheap and we see most implementations split vectors into 1, 2 or 4 parts, so there is little cost to processing entire vectors even if we do not need all their lanes. Indeed this avoids the (potentially large) cost of predication or partial loads/stores on older targets, and does not duplicate code. -
Process whole vectors as above, followed by a scalar loop:
size_t i = 0; for (; i + N <= count; i += N) LoopBody<false>(d, 0); for (; i < count; ++i) LoopBody<false>(HWY_CAPPED(T, 1)(), 0);
The template parameter and second function arguments are again not needed.
This avoids duplicating code, and is reasonable if
count
is large. Otherwise, multiple iterations may be slower than oneLoopBody
variant with masking, especially because theHWY_SCALAR
target selected byHWY_CAPPED(T, 1)
is slower for some operations due to workarounds for undefined behavior in C++. -
Process whole vectors as above, followed by a single call to a modified
LoopBody
with masking:size_t i = 0; for (; i + N <= count; i += N) { LoopBody<false>(d, 0); } if (i < count) { LoopBody<true>(d, count - i); }
Now the template parameter and second function argument can be used inside
LoopBody
to replaceLoad/Store
of full aligned vectors withLoadN/StoreN(n)
that affect no more than1 <= n <= N
aligned elements (pending implementation).This is a good default when it is infeasible to ensure vectors are padded. In contrast to the scalar loop, only a single final iteration is needed.
Design philosophy
-
Performance is important but not the sole consideration. Anyone who goes to the trouble of using SIMD clearly cares about speed. However, portability, maintainability and readability also matter, otherwise we would write in assembly. We aim for performance within 10-20% of a hand-written assembly implementation on the development platform.
-
The guiding principles of C++ are "pay only for what you use" and "leave no room for a lower-level language below C++". We apply these by defining a SIMD API that ensures operation costs are visible, predictable and minimal.
-
Performance portability is important, i.e. the API should be efficient on all target platforms. Unfortunately, common idioms for one platform can be inefficient on others. For example: summing lanes horizontally versus shuffling. Documenting which operations are expensive does not prevent their use, as evidenced by widespread use of
HADDPS
. Performance acceptance tests may detect large regressions, but do not help choose the approach during initial development. Analysis tools can warn about some potential inefficiencies, but likely not all. We instead provide a carefully chosen set of vector types and operations that are efficient on all target platforms (PPC8, SSE4/AVX2+, ARMv8). -
Future SIMD hardware features are difficult to predict. For example, AVX2 came with surprising semantics (almost no interaction between 128-bit blocks) and AVX-512 added two kinds of predicates (writemask and zeromask). To ensure the API reflects hardware realities, we suggest a flexible approach that adds new operations as they become commonly available, with scalar fallbacks where not supported.
-
Masking is not yet widely supported on current CPUs. It is difficult to define an interface that provides access to all platform features while retaining performance portability. The P0214R5 proposal lacks support for AVX-512/ARM SVE zeromasks. We suggest limiting usage of masks to the
IfThen[Zero]Else[Zero]
functions until the community has gained more experience with them. -
"Width-agnostic" SIMD is more future-proof than user-specified fixed sizes. For example, valarray-like code can iterate over a 1D array with a library-specified vector width. This will result in better code when vector sizes increase, and matches the direction taken by ARM SVE and RiscV V as well as Agner Fog's ForwardCom instruction set proposal. However, some applications may require fixed sizes, so we also guarantee support for 128-bit vectors in each instruction set.
-
The API and its implementation should be usable and efficient with commonly used compilers, including MSVC. For example, we write
ShiftLeft<3>(v)
instead ofv << 3
because MSVC 2017 (ARM64) does not propagate the literal (https://godbolt.org/g/rKx5Ga). Highway requires function-specific target attributes, supported by GCC 4.9 / Clang 3.9 / MSVC 2015. -
Efficient and safe runtime dispatch is important. Modules such as image or video codecs are typically embedded into larger applications such as browsers, so they cannot require separate binaries for each CPU. Libraries also cannot predict whether the application already uses AVX2 (and pays the frequency throttling cost), so this decision must be left to the application. Using only the lowest-common denominator instructions sacrifices too much performance. Therefore, we provide code paths for multiple instruction sets and choose the most suitable at runtime. To reduce overhead, dispatch should be hoisted to higher layers instead of checking inside every low-level function. Highway supports inlining functions in the same file or in *-inl.h headers. We generate all code paths from the same source to reduce implementation- and debugging cost.
-
Not every CPU need be supported. For example, pre-SSE4.1 CPUs are increasingly rare and the AVX instruction set is limited to floating-point operations. To reduce code size and compile time, we provide specializations for SSE4, AVX2 and AVX-512 instruction sets on x86, plus a scalar fallback.
-
Access to platform-specific intrinsics is necessary for acceptance in performance-critical projects. We provide conversions to and from intrinsics to allow utilizing specialized platform-specific functionality, and simplify incremental porting of existing code.
-
The core API should be compact and easy to learn. We provide only the few dozen operations which are necessary and sufficient for most of the 150+ SIMD applications we examined.
Prior API designs
The author has been writing SIMD code since 2002: first via assembly language,
then intrinsics, later Intel's F32vec4
wrapper, followed by three generations
of custom vector classes. The first used macros to generate the classes, which
reduces duplication but also readability. The second used templates instead.
The third (used in highwayhash and PIK) added support for AVX2 and runtime
dispatch. The current design (used in JPEG XL) enables code generation for
multiple platforms and/or instruction sets from the same source, and improves
runtime dispatch.
Differences versus P0214R5 proposal
-
Adding widely used and portable operations such as
AndNot
,AverageRound
, bit-shift by immediates andIfThenElse
. -
Designing the API to avoid or minimize overhead on AVX2/AVX-512 caused by crossing 128-bit 'block' boundaries.
-
Avoiding the need for non-native vectors. By contrast, P0214R5's
simd_cast
returnsfixed_size<>
vectors which are more expensive to access because they reside on the stack. We can avoid this plus additional overhead on ARM/AVX2 by defining width-expanding operations as functions of a vector part, e.g. promoting half a vector ofuint8_t
lanes to one full vector ofuint16_t
, or demoting full vectors to half vectors with half-width lanes. -
Guaranteeing access to the underlying intrinsic vector type. This ensures all platform-specific capabilities can be used. P0214R5 instead only 'encourages' implementations to provide access.
-
Enabling safe runtime dispatch and inlining in the same binary. P0214R5 is based on the Vc library, which does not provide assistance for linking multiple instruction sets into the same binary. The Vc documentation suggests compiling separate executables for each instruction set or using GCC's ifunc (indirect functions). The latter is compiler-specific and risks crashes due to ODR violations when compiling the same function with different compiler flags. We solve this problem via target-specific namespaces and attributes (see HOWTO section below). We also permit a mix of static target selection and runtime dispatch for hotspots that may benefit from newer instruction sets if available.
-
Using built-in PPC vector types without a wrapper class. This leads to much better code generation with GCC 6.3: https://godbolt.org/z/pd2PNP. By contrast, P0214R5 requires a wrapper. We avoid this by using only the member operators provided by the PPC vectors; all other functions and typedefs are non-members. 2019-04 update: Clang power64le does not have this issue, so we simplified get_part(d, v) to GetLane(v).
-
Omitting inefficient or non-performance-portable operations such as
hmax
,operator[]
, and unsupported integer comparisons. Applications can often replace these operations at lower cost than emulating that exact behavior. -
Omitting
long double
types: these are not commonly available in hardware. -
Ensuring signed integer overflow has well-defined semantics (wraparound).
-
Simple header-only implementation and less than a tenth of the size of the Vc library from which P0214 was derived (98,000 lines in https://github.com/VcDevel/Vc according to the gloc Chrome extension).
-
Avoiding hidden performance costs. P0214R5 allows implicit conversions from integer to float, which costs 3-4 cycles on x86. We make these conversions explicit to ensure their cost is visible.
Other related work
-
Neat SIMD adopts a similar approach with interchangeable vector/scalar types and a compact interface. It allows access to the underlying intrinsics, but does not appear to be designed for other platforms than x86.
-
UME::SIMD (code, paper) also adopts an explicit vectorization model with vector classes. However, it exposes the union of all platform capabilities, which makes the API harder to learn (209-page spec) and implement (the estimated LOC count is 500K). The API is less performance-portable because it allows applications to use operations that are inefficient on other platforms.
-
Inastemp (code, paper) is a vector library for scientific computing with some innovative features: automatic FLOPS counting, and "if/else branches" using lambda functions. It supports IBM Power8, but only provides float and double types.
Overloaded function API
Most C++ vector APIs rely on class templates. However, the ARM SVE vector
type is sizeless and cannot be wrapped in a class. We instead rely on overloaded
functions. Overloading based on vector types is also undesirable because SVE
vectors cannot be default-constructed. We instead use a dedicated 'descriptor'
type Simd
for overloading, abbreviated to D
for template arguments and
d
in lvalues.
Note that generic function templates are possible (see highway.h).
Masks
AVX-512 introduced a major change to the SIMD interface: special mask registers (one bit per lane) that serve as predicates. It would be expensive to force AVX-512 implementations to conform to the prior model of full vectors with lanes set to all one or all zero bits. We instead provide a Mask type that emulates a subset of this functionality on other platforms at zero cost.
Masks are returned by comparisons and TestBit
; they serve as the input to
IfThen*
. We provide conversions between masks and vector lanes. For clarity
and safety, we use FF..FF as the definition of true. To also benefit from
x86 instructions that only require the sign bit of floating-point inputs to be
set, we provide a special ZeroIfNegative
function.
Additional resources
This is not an officially supported Google product. Contact: janwas@google.com