Author: Remi Meier <remi.me...@gmail.com> Branch: extradoc Changeset: r5955:e71b2e056b2f Date: 2019-07-25 09:25 +0200 http://bitbucket.org/pypy/extradoc/changeset/e71b2e056b2f/
Log: some final improvements from my side diff --git a/blog/draft/2019-07-arm64.rst b/blog/draft/2019-07-arm64.rst --- a/blog/draft/2019-07-arm64.rst +++ b/blog/draft/2019-07-arm64.rst @@ -2,59 +2,64 @@ We are pleased to announce the availability of the new PyPy for AArch64. This port brings PyPy's high-performance just-in-time compiler to the AArch64 -platform, also known as 64-bit ARM. This work was funded by ARM Holdings Ltd. -and Crossbar.io. +platform, also known as 64-bit ARM. With the addition of AArch64, PyPy now +supports a total of 6 architectures: x86 (32 & 64bit), ARM (32 & 64bit), PPC64, +and s390x. The AArch64 work was funded by ARM Holdings Ltd. and Crossbar.io. -To show how well the new PyPy port performs, we compare the performance of PyPy -against CPython on a set of benchmarks. As a point of comparison, we include the -results of PyPy on x86_64. Note, however, that the results presented here were -measured on a Graviton A1 machine from AWS, which comes with a very serious -word of warning: Graviton A1's are -virtual machines and as such, are not suitable for benchmarking. If someone -has access to a beefy enough (16G) ARM64 server and is willing to give -us access to it, we are happy to redo the benchmarks on a real machine. -Our main concern here is that while a virtual CPU is 1-to-1 with a real CPU, it's -not clear to us how caches are shared, and how they cross CPU boundaries. +PyPy has a good record of boosting the performance of Python programs on the +existing platforms. To show how well the new PyPy port performs, we compare the +performance of PyPy against CPython on a set of benchmarks. As a point of +comparison, we include the results of PyPy on x86_64. -The following graph shows the speedup of PyPy (hg id 2417f925ce94) compared to -CPython (2.7.15) on AArch64, as well as the speedups on a x86_64 Linux laptop, +Note, however, that the results presented here were measured on a Graviton A1 +machine from AWS, which comes with a very serious word of warning: Graviton A1's +are virtual machines, and, as such, they are not suitable for benchmarking. If +someone has access to a beefy enough (16G) ARM64 server and is willing to give +us access to it, we are happy to redo the benchmarks on a real machine. One +major concern is that while a virtual CPU is 1-to-1 with a real CPU, it is not +clear to us how CPU caches are shared across virtual CPUs. Also, note that by no +means is this benchmark suite representative enough to average the results. Read +the numbers individually per benchmark. + +The following graph shows the speedups on AArch64 of PyPy (hg id 2417f925ce94) compared to +CPython (2.7.15), as well as the speedups on a x86_64 Linux laptop comparing the most recent release, PyPy 7.1.1, to CPython 2.7.16. .. image:: 2019-07-arm64-speedups.png In the majority of benchmarks, the speedups achieved on AArch64 match those achieved on the x86_64 laptop. Over CPython, PyPy on AArch64 achieves speedups -between 0.6x to 44.9x. These speedups are comparable to x86_64, where they are -between 0.6x and 58.9x. +between 0.6x to 44.9x. These speedups are comparable to x86_64, where the +numbers are between 0.6x and 58.9x. The next graph compares between the speedups achieved on AArch64 to the speedups -achieved on x86_64, i.e., how much the speedup is on AArch64 vs. the same -benchmark on x86_64. Note that by no means is this benchmark suite -representative enough to average the results. Read the numbers individually per -benchmark. +achieved on x86_64, i.e., how great the speedup is on AArch64 vs. the same +benchmark on x86_64. This comparison should give a rough idea about the +quality of the generated code for the new platform. .. image:: 2019-07-arm64-relative.png -Note also that we see a wide variance. There are generally three groups of +Note that we see a large variance: There are generally three groups of benchmarks - those that run at more or less the same speed, those that -run at 2x the speed and those that run at 0.5x the speed of x86_64. +run at 2x the speed, and those that run at 0.5x the speed of x86_64. The variance and disparity are likely related to a variety of issues, mostly due -to differences in architecture. What *is* however interesting is that compared +to differences in architecture. What *is* however interesting is that, compared to measurements performed on older ARM boards, the branch predictor on the Graviton A1 machine appears to have improved. As a result, the speedups achieved -by PyPy over CPython are smaller: "sophisticated" branchy code, like CPython -itself, simply runs a lot faster. +by PyPy over CPython are smaller than on older ARM boards: sufficiently branchy +code, like CPython itself, simply runs a lot faster. As a result, the advantage +of the non-branchy code generated by PyPy's just-in-time compiler is smaller. -One takeaway here is that there is a lot of improvement left to be done -in PyPy. This is true for both of the above platforms, but probably more -so for AArch64, which comes with a large number of registers. The PyPy -backend was written with x86 (the 32-bit variant) in mind, which is very -register poor. We think we can improve somewhat in the area of emitting -more modern machine code, which should have more impact for AArch64 -than for x86_64. There are also still a few missing features in the AArch64 -backend, which are implemented as calls instead of inlined instructions; -something we hope to improve. +One takeaway here is that many possible improvements for PyPy yet to be +implemented. This is true for both of the above platforms, but probably more so +for AArch64, which comes with a large number of CPU registers. The PyPy backend +was written with x86 (the 32-bit variant) in mind, which has a really low number +of registers. We think that we can improve in the area of emitting more modern +machine code, which may have a higher impact on AArch64 than on x86_64. There is +also a number of missing features in the AArch64 backend. These features are +currently implemented as expensive function calls instead of inlined native +instructions, something we intend to improve. Best, Maciej Fijalkowski, Armin Rigo and the PyPy team _______________________________________________ pypy-commit mailing list pypy-commit@python.org https://mail.python.org/mailman/listinfo/pypy-commit
