NotificationsYou must be signed in to change notification settings
Fork33.3k
Star69.7k

Math hypot exactfloat fastpath#8949

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to ourterms of service andprivacy statement. We’ll occasionally send you account related emails.

Already on GitHub?Sign in to your account

Jump to bottom

Merged

rhettinger merged 2 commits intopython:masterfromrhettinger:math-hypot-exactfloat-fastpath

Aug 27, 2018

Merged

Math hypot exactfloat fastpath#8949

rhettinger merged 2 commits intopython:masterfromrhettinger:math-hypot-exactfloat-fastpath

Aug 27, 2018

Conversation

Copy link

Contributor

rhettinger commentedAug 27, 2018

Provide a fast path for the common case of exact float inputs. Saves the overhead of an external function call and of thex == 1.0 error check. Allows the inner loops to mostly use registers.

Speeds-up the overall function by approximately 25%:

$ ------ baseline -------$ py -m timeit -r7 -s 'from math import dist; p=tuple(map(float, range(20))); q=tuple(reversed(p))' 'dist(p, q)'1000000 loops, best of 7: 297 nsec per loop$ ------ patched -------$ py -m timeit -r7 -s 'from math import dist; p=tuple(map(float, range(20))); q=tuple(reversed(p))' 'dist(p, q)'1000000 loops, best of 7: 215 nsec per loop

Disassembly ofmath_hypot() using GCC 8.2 shows a very tight inner loop without unnecessary register spills and reloads and without external calls that have to save and restore registers:

L378:xorl%eax, %eaxandpdlC5(%rip), %xmm0         # x = fabs(x);ucomisd%xmm0, %xmm0movl$1, %ecxsetne%alcmovp%ecx, %eaxorl%eax, %ebx               # found_nan |= Py_IS_NAN(x);L377:movsd%xmm0, (%r12,%r15,8)     # coordinates[i] = x;maxsd%xmm1, %xmm0             # if (x > max) { max = x; }addq$1, %r15                 # i++cmpq%r15, %rbp               # i < nmovapd%xmm0, %xmm1jleL418L385:movq24(%r13,%r15,8), %rdi    # item = PyTuple_GET_ITEM(args, i);cmpq%r14, 8(%rdi)            # if (PyFloat_CheckExact(item))jneL375movsd16(%rdi), %xmm0          # x = PyFloat_AS_DOUBLE(item)jmpL378

rhettinger added2 commits

August 26, 2018 14:04

Add fast path for exact floats.

d2eb71e

Saves function call overhead and lets inner-loop be performedin registers with no spills/reloads.

Extend the exact float fastpath to math_dist().

f1cffbc

rhettinger added performance

Performance or resource usage

skip issue skip news labels

Aug 27, 2018

the-knights-who-say-ni added the CLA signed label

Aug 27, 2018

bedevere-bot added the awaiting merge label

Aug 27, 2018

rhettinger merged commit74734f7 intopython:master

Aug 27, 2018

bedevere-bot removed the awaiting merge label

Aug 27, 2018

rhettinger deleted the math-hypot-exactfloat-fastpath branch

August 27, 2018 00:38

CuriousLearner added a commit to CuriousLearner/cpython that referenced this pull request

Aug 27, 2018

Merge branch 'master' into fix-issue24255

c50bb33

* master: (104 commits)  Fast path for exact floats in math.hypot() and math.dist() (pythonGH-8949)  Remove AIX workaround test_subprocess (pythonGH-8939)  bpo-34503: Fix refleak in PyErr_SetObject() (pythonGH-8934)  closes bpo-34504: Remove the useless NULL check in PySequence_Check(). (pythonGH-8935)  closes bpo-34501: PyType_FromSpecWithBases: Check spec->name before dereferencing it. (pythonGH-8930)  closes bpo-34502: Remove a note about utf8_mode from sys.exit() docs. (pythonGH-8928)  Remove unneeded PyErr_Clear() in _winapi_SetNamedPipeHandleState_impl() (pythonGH-8281)  Fix markup in stdtypes documentation (pythonGH-8905)  bpo-34395: Don't free allocated memory on realloc fail in load_mark() in _pickle.c. (pythonGH-8788)  Fix upsizing of marks stack in pickle module. (pythonGH-8860)  bpo-34171: Prevent creating Lib/trace.cover when run the trace module. (pythonGH-8841)  closes bpo-34493: Objects/genobject.c: Add missing NULL check to compute_cr_origin() (pythonGH-8911)  Fixed typo with asynccontextmanager code example (pythonGH-8845)  bpo-34426: fix typo (__lltrace__ -> __ltrace__) (pythonGH-8822)  bpo-13312: Avoid int underflow in time year. (pythonGH-8912)  bpo-34492: Python/coreconfig.c: Fix _Py_wstrlist_copy() (pythonGH-8910)  bpo-34448: Improve output of usable wchar_t check (pythonGH-8846)  closes bpo-34471: _datetime: Add missing NULL check to tzinfo_from_isoformat_results. (pythonGH-8869)  bpo-6700: Fix inspect.getsourcelines for module level frames/tracebacks (pythonGH-8864)  Fix typo in the dataclasses's doc (pythonGH-8896)  ...