| Age | Commit message (Collapse) | Author |
|
(#130700)
|
|
This is a part of #97655.
|
|
declaration" (#98593)
Reverts llvm/llvm-project#98075
bots are broken
|
|
This is a part of #97655.
|
|
We compute atan2f(y, x) in 2 stages:
- Fast step: perform computations in double precision , with relative
errors < 2^-50
- Accurate step: if the result from the Fast step fails Ziv's rounding
test, then we perform computations in double-double precision, with
relative errors < 2^-100.
On Ryzen 5900X, worst-case latency is ~ 200 clocks, compared to average
latency ~ 60 clocks, and average reciprocal throughput ~ 20 clocks.
|
|
|
|
gcc bot failures with `atanf(+-Inf)`:
https://lab.llvm.org/buildbot/#/builders/250/builds/20331/steps/8/logs/stdio
|
|
Simplify the range reduction logic and computations. Performance test
using `perf.sh` from CORE-MATH project on Ryzen 5900X:
Before:
```
$ ./perf.sh atanf --rdtsc --path1
LIBC-location: /home/lnt/experiment/llvm-project/build/projects/libc/lib/libllvmlibc.a
GNU libc version: 2.35
GNU libc release: stable
-- CORE-MATH reciprocal throughput --
[####################] 100 %
Ntrial = 20 ; Min = 14.369 + 0.556 clc/call; Median-Min = 0.613 clc/call; Max = 15.061 clc/call;
-- System LIBC reciprocal throughput --
[####################] 100 %
Ntrial = 20 ; Min = 26.180 + 0.015 clc/call; Median-Min = 0.014 clc/call; Max = 26.260 clc/call;
-- LIBC reciprocal throughput --
[####################] 100 %
Ntrial = 20 ; Min = 21.237 + 0.022 clc/call; Median-Min = 0.020 clc/call; Max = 21.272 clc/call;
$ ./perf.sh atanf --rdtsc --path1 --latency
LIBC-location: /home/lnt/experiment/llvm-project/build/projects/libc/lib/libllvmlibc.a
GNU libc version: 2.35
GNU libc release: stable
-- CORE-MATH latency --
[####################] 100 %
Ntrial = 20 ; Min = 50.505 + 0.045 clc/call; Median-Min = 0.037 clc/call; Max = 50.579 clc/call;
-- System LIBC latency --
[####################] 100 %
Ntrial = 20 ; Min = 62.438 + 0.836 clc/call; Median-Min = 0.049 clc/call; Max = 64.498 clc/call;
-- LIBC latency --
[####################] 100 %
Ntrial = 20 ; Min = 67.781 + 0.546 clc/call; Median-Min = 0.028 clc/call; Max = 68.844 clc/call;
```
After:
```
$ ./perf.sh atanf --rdtsc --path2
LIBC-location: /home/lnt/experiment/llvm/llvm-project/build/projects/libc/lib/libllvmlibc.a
GNU libc version: 2.35
GNU libc release: stable
-- CORE-MATH reciprocal throughput --
[####################] 100 %
Ntrial = 20 ; Min = 14.400 + 0.353 clc/call; Median-Min = 0.404 clc/call; Max = 14.863 clc/call;
-- System LIBC reciprocal throughput --
[####################] 100 %
Ntrial = 20 ; Min = 25.247 + 0.783 clc/call; Median-Min = 0.714 clc/call; Max = 26.238 clc/call;
-- LIBC reciprocal throughput --
[####################] 100 %
Ntrial = 20 ; Min = 13.751 + 0.158 clc/call; Median-Min = 0.140 clc/call; Max = 14.006 clc/call;
$ ./perf.sh atanf --rdtsc --path2 --latency
LIBC-location: /home/lnt/experiment/llvm/llvm-project/build/projects/libc/lib/libllvmlibc.a
GNU libc version: 2.35
GNU libc release: stable
-- CORE-MATH latency --
[####################] 100 %
Ntrial = 20 ; Min = 51.837 + 0.073 clc/call; Median-Min = 0.058 clc/call; Max = 52.000 clc/call;
-- System LIBC latency --
[####################] 100 %
Ntrial = 20 ; Min = 62.487 + 1.878 clc/call; Median-Min = 1.965 clc/call; Max = 64.569 clc/call;
OK
-- LIBC latency --
[####################] 100 %
Ntrial = 20 ; Min = 55.414 + 1.312 clc/call; Median-Min = 0.345 clc/call; Max = 58.362 clc/call;
```
|
|
Another patch is needed to cover `DyadicFloat` and `NormalFloat`
constructors.
|
|
This is step 4 of
https://discourse.llvm.org/t/rfc-customizable-namespace-to-allow-testing-the-libc-when-the-system-libc-is-also-llvms-libc/73079
|
|
Fix signed zeros for acosf, acoshf, and atanf in FE_DOWNWARD mode.
Reviewed By: sivachandra
Differential Revision: https://reviews.llvm.org/D159476
|
|
Many math functions need to check for floating point rounding modes to
return correct values. Currently most of them use the internal implementation
of `fegetround`, which is platform-dependent and blocking math functions to be
enabled on platforms with unimplemented `fegetround`. In this change, we add
platform independent rounding mode checks and switching math functions to use
them instead. https://github.com/llvm/llvm-project/issues/63016
Reviewed By: sivachandra
Differential Revision: https://reviews.llvm.org/D152280
|
|
Differential Revision: https://reviews.llvm.org/D146738
|
|
|
|
|
|
Differential Revision: https://reviews.llvm.org/D134917
|
|
Performance by core-math (core-math/glibc 2.31/current llvm-14):
28.879/20.843/20.15
Differential Revision: https://reviews.llvm.org/D132842
|