| Commit message (Collapse) | Author | Age |
|
|
|
|
|
|
| |
Accuracy is too poor - requires at least two Newton iterations, but then
it is no longer significantly faster than `vsqrt`.
Fixes #2094.
|
|
|
|
|
|
| |
The original implementation fails for 0, denormals, inf, and NaN.
See #2150
|
|
|
|
| |
kernel)
|
|
|
|
|
| |
It's slightly faster and slightly more accurate, allowing our current
packetmath tests to pass for sqrt with a single iteration.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Clang does a poor job of optimizing the GEBP microkernel on 32-bit ARM,
leading to excessive 16-byte register spills, slowing down basic f32
matrix multiplication by approx 50%.
By specializing `gebp_traits`, we can eliminate the register spills.
Volatile inline ASM both acts as a barrier to prevent reordering and
enforces strict register use. In a simple f32 matrix multiply example,
this modification reduces 16-byte spills from 109 instances to zero,
leading to a 1.5x speed increase (search for `16-byte Spill` in the
assembly in https://godbolt.org/z/chsPbE).
This is a replacement of !379. See there for further discussion.
Also moved `gebp_traits` specializations for NEON to
`Eigen/src/Core/arch/NEON/GeneralBlockPanelKernel.h` to be alongside
other NEON-specific code.
Fixes #2138.
|
|
|
|
|
|
|
|
|
|
| |
The recent addition of vectorized pow (!330) relies on `pfrexp` and
`pldexp`. This was missing for `Eigen::half` and `Eigen::bfloat16`.
Adding tests for these packet ops also exposed an issue with handling
negative values in `pfrexp`, returning an incorrect exponent.
Added the missing implementations, corrected the exponent in `pfrexp1`,
and added `packetmath` tests.
|
|
|
|
|
| |
2)make paddsub op support the Packet2cf/Packet4f/Packet2f in NEON
3)make paddsub op support the Packet2cf/Packet4f in SSE
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
provides a ~10% speedup.
* Write iterative sqrt explicitly in terms of pmadd. This gives up to 7% speedup for psqrt<float> with AVX & SSE with FMA.
* Remove iterative psqrt<double> for NEON, because the initial rsqrt apprimation is not accurate enough for convergence in 2 Newton-Raphson steps and with 3 steps, just calling the builtin sqrt insn is faster.
The following benchmarks were compiled with clang "-O2 -fast-math -mfma" and with and without -mavx.
AVX+FMA (float)
name old cpu/op new cpu/op delta
BM_eigen_sqrt_float/1 1.08ns ± 0% 1.09ns ± 1% ~
BM_eigen_sqrt_float/8 2.07ns ± 0% 2.08ns ± 1% ~
BM_eigen_sqrt_float/64 12.4ns ± 0% 12.4ns ± 1% ~
BM_eigen_sqrt_float/512 95.7ns ± 0% 95.5ns ± 0% ~
BM_eigen_sqrt_float/4k 776ns ± 0% 763ns ± 0% -1.67%
BM_eigen_sqrt_float/32k 6.57µs ± 1% 6.13µs ± 0% -6.69%
BM_eigen_sqrt_float/256k 83.7µs ± 3% 83.3µs ± 2% ~
BM_eigen_sqrt_float/1M 335µs ± 2% 332µs ± 2% ~
SSE+FMA (float)
name old cpu/op new cpu/op delta
BM_eigen_sqrt_float/1 1.08ns ± 0% 1.09ns ± 0% ~
BM_eigen_sqrt_float/8 2.07ns ± 0% 2.06ns ± 0% ~
BM_eigen_sqrt_float/64 12.4ns ± 0% 12.4ns ± 1% ~
BM_eigen_sqrt_float/512 95.7ns ± 0% 96.3ns ± 4% ~
BM_eigen_sqrt_float/4k 774ns ± 0% 763ns ± 0% -1.50%
BM_eigen_sqrt_float/32k 6.58µs ± 2% 6.11µs ± 0% -7.06%
BM_eigen_sqrt_float/256k 82.7µs ± 1% 82.6µs ± 1% ~
BM_eigen_sqrt_float/1M 330µs ± 1% 329µs ± 2% ~
SSE+FMA (double)
BM_eigen_sqrt_double/1 1.63ns ± 0% 1.63ns ± 0% ~
BM_eigen_sqrt_double/8 6.51ns ± 0% 6.08ns ± 0% -6.68%
BM_eigen_sqrt_double/64 52.1ns ± 0% 46.5ns ± 1% -10.65%
BM_eigen_sqrt_double/512 417ns ± 0% 374ns ± 1% -10.29%
BM_eigen_sqrt_double/4k 3.33µs ± 0% 2.97µs ± 1% -11.00%
BM_eigen_sqrt_double/32k 26.7µs ± 0% 23.7µs ± 0% -11.07%
BM_eigen_sqrt_double/256k 213µs ± 0% 206µs ± 1% -3.31%
BM_eigen_sqrt_double/1M 862µs ± 0% 870µs ± 2% +0.96%
AVX+FMA (double)
name old cpu/op new cpu/op delta
BM_eigen_sqrt_double/1 1.63ns ± 0% 1.63ns ± 0% ~
BM_eigen_sqrt_double/8 6.51ns ± 0% 6.06ns ± 0% -6.95%
BM_eigen_sqrt_double/64 52.1ns ± 0% 46.5ns ± 1% -10.80%
BM_eigen_sqrt_double/512 417ns ± 0% 373ns ± 1% -10.59%
BM_eigen_sqrt_double/4k 3.33µs ± 0% 2.97µs ± 1% -10.79%
BM_eigen_sqrt_double/32k 26.7µs ± 0% 23.8µs ± 0% -10.94%
BM_eigen_sqrt_double/256k 214µs ± 0% 208µs ± 2% -2.76%
BM_eigen_sqrt_double/1M 866µs ± 3% 923µs ± 7% ~
|
|
|
|
| |
otherwise has an error of ~1000 ulps.
|
|
|
|
| |
Simple typo, the max impl called pmin instead of pmax for floats.
|
| |
|
| |
|
| |
|
| |
|
|
|
|
| |
- Adding propagate tests to bfloat16.
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Current implementations fail to consider half-float packets, only
half-float scalars. Added specializations for packets on AVX, AVX512 and
NEON. Added tests to `special_packetmath`.
The current `special_functions` tests would fail for half and bfloat16 due to
lack of precision. The NEON tests also fail with precision issues and
due to different handling of `sqrt(inf)`, so special functions bessel, ndtri
have been disabled.
Tested with AVX, AVX512.
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
The current impl corrupts the comparison masks when converting
from float back to bfloat16. The resulting masks are then
no longer all zeros or all ones, which breaks when used with
`pselect` (e.g. in `pmin<PropagateNumbers>`). This was
causing `packetmath_15` to fail on arm.
Introducing a simple `F32MaskToBf16Mask` corrects this (takes
the lower 16-bits for each float mask).
|
|
|
|
|
|
|
| |
Missing inline breaks blas, since symbol generated in
`complex_single.cpp`, `complex_double.cpp`, `single.cpp`, `double.cpp`
Changed rest of inlines to `EIGEN_STRONG_INLINE`.
|
|
|
|
|
|
| |
- Add predux_half_dowto4
- Remove explicit casts in Half.h to match the behaviour of BFloat16.h
- Enable more packetmath tests for Eigen::half
|
|
|
|
|
|
| |
Using overloaded arithmetic operators for Arm __fp16 always
causes a promotion to float. We replace operator* by vmulh_f16
to avoid this.
|
|
|
|
| |
using PacketMath.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Armv8.2-a provides a native half-precision floating point (__fp16 aka.
float16_t). This patch introduces
* __fp16 as underlying type of Eigen::half if this type is available
* the packet types Packet4hf and Packet8hf representing float16x4_t and
float16x8_t respectively
* packet-math for the above packets with corresponding scalar type Eigen::half
The packet-math functionality has been implemented by Ashutosh Sharma
<ashutosh.sharma@amperecomputing.com>.
This closes #1940.
|
|
|
|
| |
(almost) all packetmath tests with SSE, AVX, and AVX512.
|
| |
|
|
|
| |
'vmvnq_u64' does not exist for some reason.
|
| |
|
| |
|
|
|
|
|
|
|
|
| |
CastXML simulates the preprocessors of other compilers, but actually
parses the translation unit with an internal Clang compiler.
Use the same `vld1q_u64` workaround that we do for Clang.
Fixes: #1979
|
|
|
|
| |
the comments here have long been fixed. The workarounds were now detrimental because (1) they prevented using fused mul-add on Clang/ARM32 and (2) the unnecessary 'volatile' in 'asm volatile' prevented legitimate reordering by the compiler.
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
|
| |
for large values.
The NEON implementation mimics the SSE implementation, but didn't mention the caveat that due to the unsigned of signed integer conversions, not all values in the original floating point represented are supported.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Added missing `pmadd<Packet2f>` for NEON. This leads to significant
improvement in precision than previous `pmul+padd`, which was causing
the `pcos` tests to fail. Also added an approx test with
`std::sin`/`std::cos` since otherwise returning any `a^2+b^2=1` would
pass.
Modified `log(denorm)` tests. Denorms are not always supported by all
systems (returns `::min`), are always flushed to zero on 32-bit arm,
and configurably flush to zero on sse/avx/aarch64. This leads to
inconsistent results across different systems (i.e. `-inf` vs `nan`).
Added a check for existence and exclude ARM.
Removed logistic exactness test, since scalar and vectorized versions
follow different code-paths due to differences in `pexp` and `pmadd`,
which result in slightly different values. For example, exactness always
fails on arm, aarch64, and altivec.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The current multiply (`pmul`) and comparison operators (`pcmp_lt`,
`pcmp_le`, `pcmp_eq`) are missing for packets `Packet2l` and
`Packet2ul`. This leads to compile errors for the `packetmath.cpp` tests
in clang. Here we add and test the missing ops.
Tested:
```
$ aarch64-linux-gnu-g++ -static -I./ '-DEIGEN_TEST_PART_9=1' '-DEIGEN_TEST_PART_10=1' test/packetmath.cpp -o packetmath
$ adb push packetmath /data/local/tmp/
$ adb shell "/data/local/tmp/packetmath"
$ arm-linux-gnueabihf-g++ -mfpu=neon -static -I./ '-DEIGEN_TEST_PART_9=1' '-DEIGEN_TEST_PART_10=1' test/packetmath.cpp -o packetmath
$ adb push packetmath /data/local/tmp/
$ adb shell "/data/local/tmp/packetmath"
$ clang++ -target aarch64-linux-android21 -static -I./ '-DEIGEN_TEST_PART_9=1' '-DEIGEN_TEST_PART_10=1' test/packetmath.cpp -o packetmath
$ adb push packetmath /data/local/tmp/
$ adb shell "/data/local/tmp/packetmath"
$ clang++ -target armv7-linux-android21 -static -mfpu=neon -I./ '-DEIGEN_TEST_PART_9=1' '-DEIGEN_TEST_PART_10=1' test/packetmath.cpp -o packetmath
$ adb push packetmath /data/local/tmp/
$ adb shell "/data/local/tmp/packetmath"
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The NEON `pcast` operators are all implemented and tested for existing
packets. This requires adding a `pcast(a,b,c,d,e,f,g,h)` for casting
between `int64_t` and `int8_t` in `GenericPacketMath.h`.
Removed incorrect `HasHalfPacket` definition for NEON's
`Packet2l`/`Packet2ul`.
Adjustments were also made to the `packetmath` tests. These include
- minor bug fixes for cast tests (i.e. 4:1 casts, only casting for
packets that are vectorizable)
- added 8:1 cast tests
- random number generation
- original had uninteresting 0 to 0 casts for many casts between
floating-point and integers, and exhibited signed overflow
undefined behavior
Tested:
```
$ aarch64-linux-gnu-g++ -static -I./ '-DEIGEN_TEST_PART_ALL=1' test/packetmath.cpp -o packetmath
$ adb push packetmath /data/local/tmp/
$ adb shell "/data/local/tmp/packetmath"
```
|
|
|
|
|
|
|
|
|
|
|
| |
The use of the `packet_traits<>::HasCast` field is currently inconsistent with
`type_casting_traits<>`, and is unused apart from within
`test/packetmath.cpp`. In addition, those packetmath cast tests do not
currently reflect how casts are performed in practice: they ignore the
`SrcCoeffRatio` and `TgtCoeffRatio` fields, assuming a 1:1 ratio.
Here we remove the unsed `HasCast`, and modify the packet cast tests to
better reflect their usage.
|
|
|
|
| |
ptranspose on NEON
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
tests as other arithmetic types.
This change also contains a few minor cleanups:
1. Remove packet op pnot, which is not needed for anything other than pcmp_le_or_nan,
which can be done in other ways.
2. Remove the "HasInsert" enum, which is no longer needed since we removed the
corresponding packet ops.
3. Add faster pselect op for Packet4i when SSE4.1 is supported.
Among other things, this makes the fast transposeInPlace() method available for Matrix<bool>.
Run on ************** (72 X 2994 MHz CPUs); 2020-05-09T10:51:02.372347913-07:00
CPU: Intel Skylake Xeon with HyperThreading (36 cores) dL1:32KB dL2:1024KB dL3:24MB
Benchmark Time(ns) CPU(ns) Iterations
-----------------------------------------------------------------------
BM_TransposeInPlace<float>/4 9.77 9.77 71670320
BM_TransposeInPlace<float>/8 21.9 21.9 31929525
BM_TransposeInPlace<float>/16 66.6 66.6 10000000
BM_TransposeInPlace<float>/32 243 243 2879561
BM_TransposeInPlace<float>/59 844 844 829767
BM_TransposeInPlace<float>/64 933 933 750567
BM_TransposeInPlace<float>/128 3944 3945 177405
BM_TransposeInPlace<float>/256 16853 16853 41457
BM_TransposeInPlace<float>/512 204952 204968 3448
BM_TransposeInPlace<float>/1k 1053889 1053861 664
BM_TransposeInPlace<bool>/4 14.4 14.4 48637301
BM_TransposeInPlace<bool>/8 36.0 36.0 19370222
BM_TransposeInPlace<bool>/16 31.5 31.5 22178902
BM_TransposeInPlace<bool>/32 111 111 6272048
BM_TransposeInPlace<bool>/59 626 626 1000000
BM_TransposeInPlace<bool>/64 428 428 1632689
BM_TransposeInPlace<bool>/128 1677 1677 417377
BM_TransposeInPlace<bool>/256 7126 7126 96264
BM_TransposeInPlace<bool>/512 29021 29024 24165
BM_TransposeInPlace<bool>/1k 116321 116330 6068
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
single place, and can be replaced by other ops when constructing the first/final packet in linspaced_op_impl::packetOp.
I cannot measure any performance changes for SSE, AVX, or AVX512.
name old time/op new time/op delta
BM_LinSpace<float>/1 1.63ns ± 0% 1.63ns ± 0% ~ (p=0.762 n=5+5)
BM_LinSpace<float>/8 4.92ns ± 3% 4.89ns ± 3% ~ (p=0.421 n=5+5)
BM_LinSpace<float>/64 34.6ns ± 0% 34.6ns ± 0% ~ (p=0.841 n=5+5)
BM_LinSpace<float>/512 217ns ± 0% 217ns ± 0% ~ (p=0.421 n=5+5)
BM_LinSpace<float>/4k 1.68µs ± 0% 1.68µs ± 0% ~ (p=1.000 n=5+5)
BM_LinSpace<float>/32k 13.3µs ± 0% 13.3µs ± 0% ~ (p=0.905 n=5+4)
BM_LinSpace<float>/256k 107µs ± 0% 107µs ± 0% ~ (p=0.841 n=5+5)
BM_LinSpace<float>/1M 427µs ± 0% 427µs ± 0% ~ (p=0.690 n=5+5)
|
|
|
|
| |
Clean up a compiler warning in c++03 mode in AVX512/Complex.h.
|
|
|
|
| |
packet op implementations.
|
| |
|
|
|
|
|
|
|
| |
SSE/AVX/AVX512 as it is already used for NEON.
This will allow us to define multiple packet types backed by the same vector type, e.g., __m128i.
Use this machanism to define packets for half and clean up the packet op implementations.
|
| |
|
|
|
|
|
|
| |
{uint8, int8} -> {int16, uint16, int32, uint32, float}
{uint16, int16} -> {int32, uint32, int64, uint64, float}
for NEON. These conversions were advertised as vectorized, but not actually implemented.
|