Makes e. g. matrix multiplication 2x faster:
name old cpu/op new cpu/op delta
BM_convers 181ms ± 1% 62ms ± 9% -65.82% (p=0.016 n=4+5)
Tested on all possible input values (not adding tests, since they
take a long time).
Activates vectorization of the Eigen::half versions of the tanh and
logistic functions when they run on Neon. Both functions convert their
inputs to float before computing the output, and as a result of this
commit, the conversions and the computation in float are vectorized.
We currently have plenty of type definitions with the alignment
qualifier coming after the type. The compiler warns about ignoring
them:
int EIGEN_ALIGN16 ai[4];
Turn this into:
EIGEN_ALIGN16 int ai[4];
Fixes compiler errors in expressions that look like
Eigen::Matrix<Eigen::half, 3, 1>::Random().maxCoeff()
The error comes from the code that creates the initial value for
vectorized reductions. The fix is to specify the scalar type of the
reduction's initial value.
The cahnge is necessary for Eigen::half because unlike other types,
Eigen::half scalars cannot be implicitly created from integers.
To elide the memcpy, we need to first load the `src` value into
registers by making a local copy. This avoids the need to resort
to potential UB by using `reinterpret_cast`.
This change doesn't seem to affect CPU (at least not with gcc/clang).
With optimizations on, the copy is also elided.
Looks like we need to update the
`EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR` for newer versions of MSVC as
well when compiling with NVCC. Fixes build issues for VS 2017.
VS2017 doesn't like deducing alias types, leading to a bunch of compile
errors for functions involving the `tuple` alias. Replacing with
`TupleImpl` seems to solve this, allowing the test to compile/pass.
The `Complex.h` file applies equally to HIP/CUDA, so placing under the
generic `GPU` folder.
The `TensorReductionCuda.h` has already been deprecated, now removing
for the next Eigen version.
MSVC does not support specializing compound assignments for
`std::complex`, since it already specializes them (contrary to the
standard).
Trying to use one of these on device will currently lead to a
duplicate definition error. This is still probably preferable
to no error though. If we remove the definitions for MSVC, then
it will compile, but the kernel will fail silently.
The only proper solution would be to define our own custom `Complex`
type.
The 2979 warning is yet another "calling a __host__ function from a
__host__ device__ function. Although we probably should eventually
address these, they are flooding the logs. Most of these are
harmless since we only call the original from the host.
In cases where these are actually called from device, an error is generated
instead anyways.
The 2977 warning is a bit strange - although the warning suggests the
`__device__` annotation is ignored, this doesn't actually seem to be
the case. Without the `__device__` declarations, the kernel actually
fails to run when attempting to construct such objects. Again,
these warnings are flooding the logs, so disabling for now.
reinterpret_cast between unrelated types is undefined behavior and leads
to misoptimizations on some platforms.
Use the safer (and faster) version via bit_cast
These names are so common, IMO they should not exist directly in the
`Eigen::` namespace. This prevents us from using the `last` or `all`
names for any parameters or local variables, otherwise spitting out
warnings about shadowing or hiding the global values. Many external
projects (and our own examples) also heavily use
```
using namespace Eigen;
```
which means these conflict with external libraries as well, e.g.
`std::fill(first,last,value)`.
It seems originally these were placed in a separate namespace
`Eigen::placeholders`, which has since been deprecated. I propose
to un-deprecate this, and restore the original locations.
These symbols are also imported into `Eigen::indexing`, which
additionally imports `fix` and `seq`. An alternative is to remove the
`placeholders` namespace and stick with `indexing`.
NOTE: this is an API-breaking change.
Fixes#2321.
An analogue of `std::tuple` that works on device.
Context: I've tried `std::tuple` in various versions of NVCC and clang,
and although code seems to compile, it often fails to run - generating
"illegal memory access" errors, or "illegal instruction" errors.
This replacement does work on device.
The `Serializer<T>` class implements a binary serialization that
can write to (`serialize`) and read from (`deserialize`) a byte
buffer. Also added convenience routines for serializing
a list of arguments.
This will mainly be for testing, specifically to transfer data to
and from the GPU.
The `Options` of the new `hCoeffs` vector do not necessarily match
those of the `MatrixType`, leading to build errors. Having the
`CoeffVectorType` be a template parameter relieves this restriction.
CUDA 9 seems to require labelling defaulted constructors as
`EIGEN_DEVICE_FUNC`, despite giving warnings that such labels are
ignored. Without these labels, the `gpu_basic` test fails to
compile, with errors about calling `__host__` functions from
`__host__ __device__` functions.
GCC 4.8 doesn't seem to like the `g` register constraint, failing to
compile with "error: 'asm' operand requires impossible reload".
Tested `r` instead, and that seems to work, even with latest compilers.
Also fixed some minor macro issues to eliminate warnings on armv7.
Fixes#2315.
All cuda `__half` functions are device-only in CUDA 9, including
conversions. Host-side conversions were added in CUDA 10.
The existing code doesn't build prior to 10.0.
All arithmetic functions are always device-only, so there's
therefore no reason to use vectorization on the host at all.
Modified the code to disable vectorization for `__half` on host,
which required also updating the `TensorReductionGpu` implementation
which previously made assumptions about available packets.
