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];
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.
reinterpret_cast between unrelated types is undefined behavior and leads
to misoptimizations on some platforms.
Use the safer (and faster) version via bit_cast
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.
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.
We can't make guarantees on alignment for existing calls to `pset`,
so we should default to loading unaligned. But in that case, we should
just use `ploadu` directly. For loading constants, this load should hopefully
get optimized away.
This is causing segfaults in Google Maps.
