| Age | Commit message (Collapse) | Author |
|---|
|
accurate path is skipped. (#164522)
|
|
This is one more follow-up to PR
https://github.com/llvm/llvm-project/pull/157517 that cleans up the
usage of
libc_errno in math unit-tests (non-smoke). It follows the same rule:
* if you use libc_errno in code literally, include
src/__support/libc_errno.h
* if you only rely on errno constants, include hdr/errno_macros.h
Several tests for exp/log still retain the direct libc_errno usage,
since in some cases they skip doing any validation if the error was
raised. But the direct usage of libc_errno is removed from all other
cases.
|
|
<stdint.h> includes. (#150303)
https://github.com/llvm/llvm-project/issues/149993
|
|
LIBC_ERRNO_MODE_SYSTEM to be header-only. (#143187)
This is the first step in preparation for:
https://discourse.llvm.org/t/rfc-make-clang-builtin-math-functions-constexpr-with-llvm-libc-to-support-c-23-constexpr-math-functions/86450
|
|
|
|
Fix #98393
|
|
Context: https://github.com/llvm/llvm-project/pull/87017
- Add proxy header `libc/hdr/math_macros.h` that will:
- include `<math.h>` in overlay mode,
- include `"include/llvm-libc-macros/math-macros.h"` in full build mode.
- Its corresponding CMake target `libc.hdr.math_macros` will only depend
on `libc.include.math` and `libc.include.llvm-libc-macros.math_macros`
in full build mode.
- Replace all `#include "include/llvm-libc-macros/math-macros.h"` with
`#include "hdr/math_macros.h"`.
- Add dependency to `libc.hdr.math_macros` CMake target when using
`add_fp_unittest`.
- Update the remaining dependency.
- Update bazel overlay: add `libc:hdr_math_macros` target, and replacing
all dependency on `libc:llvm_libc_macros_math_macros` with
`libc:hdr_math_macros`.
|
|
Reland of #84991
A downstream overlay mode user ran into issues with the isnan macro not
working in our sources with a specific libc configuration. This patch
replaces the last direct includes of math.h with our internal
math_macros.h, along with the necessary build system changes.
|
|
Having libc_errno outside of the namespace causes versioning issues when
trying to link the tests against LLVM-libc. Most of this patch is just
moving libc_errno inside the namespace in tests. This isn't necessary in
the function implementations since those are already inside the
namespace.
|
|
The semantics for casting can range from "bitcast" (same representation)
to "different representation", to "type promotion". Here we remove the
cast operator and force usage of `get_val` as the only function to get
the floating point value, making the intent clearer and more consistent.
|
|
(#69558)
Clean up usage of `DECLARE_SPECIAL_CONSTANTS` in global scope.
|
|
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
|
|
all rounding modes.
Implement double precision exp function correctly rounded for all
rounding modes. Using 4 stages:
- Range reduction: reduce to `exp(x) = 2^hi * 2^mid1 * 2^mid2 * exp(lo)`.
- Use 64 + 64 LUT for 2^mid1 and 2^mid2, and use cubic Taylor polynomial to
approximate `(exp(lo) - 1) / lo` in double precision. Relative error in this
step is bounded by 1.5 * 2^-63.
- If the rounding test fails, use degree-6 Taylor polynomial to approximate
`exp(lo)` in double-double precision. Relative error in this step is bounded by
2^-99.
- If the rounding test still fails, use degree-7 Taylor polynomial to compute
`exp(lo)` in ~128-bit precision.
Reviewed By: zimmermann6
Differential Revision: https://reviews.llvm.org/D158551
|
|
Fixing an issue with LLVM libc's fenv.h defined rounding mode macros
differently from system libc, making get_round() return different values from
fegetround(). Also letting math tests to skip rounding modes that cannot be
set. This should allow math tests to be run on platforms in which fenv.h is not
implemented yet.
This allows us to re-enable hermatic floating point tests in
https://reviews.llvm.org/D151123 and reverting https://reviews.llvm.org/D152742.
Reviewed By: jhuber6
Differential Revision: https://reviews.llvm.org/D152873
|
|
This patch mimics the behavior of Google Test and allow users to log custom messages after all flavors of ASSERT_ / EXPECT_.
Reviewed By: sivachandra, lntue
Differential Revision: https://reviews.llvm.org/D152630
|
|
Failing buildbot https://lab.llvm.org/buildbot/#/builders/73/builds/49707
This reverts commit 9a7b4c934893d6bc571e1ce8efab2127ae5f4e45.
|
|
This patch mimics the behavior of Google Test and allow users to log custom messages after all flavors of ASSERT_ / EXPECT_.
Reviewed By: sivachandra, lntue
Differential Revision: https://reviews.llvm.org/D152630
|
|
Reviewed By: lntue
Differential Revision: https://reviews.llvm.org/D151875
|
|
Reviewed By: gchatelet
Differential Revision: https://reviews.llvm.org/D151256
|
|
itsperformance.
Make log10 correctly rounded for non-FMA targets and improve its
performance.
Implemented fast pass and accurate pass:
**Fast Pass**:
- Range reduction step 0: Extract exponent and mantissa
```
x = 2^(e_x) * m_x
```
- Range reduction step 1: Use lookup tables of size 2^7 = 128 to reduce the argument to:
```
-2^-8 <= v = r * m_x - 1 < 2^-7
where r = 2^-8 * ceil( 2^8 * (1 - 2^-8) / (1 + k * 2^-7) )
and k = trunc( (m_x - 1) * 2^7 )
```
- Polynomial approximation: approximate `log(1 + v)` by a degree-7 polynomial generated by Sollya with:
```
> P = fpminimax((log(1 + x) - x)/x^2, 5, [|D...|], [-2^-8, 2^-7]);
```
- Combine the results:
```
log10(x) ~ ( e_x * log(2) - log(r) + v + v^2 * P(v) ) * log10(e)
```
- Perform additive Ziv's test with errors bounded by `P_ERR * v^2`. Return the result if Ziv's test passed.
**Accurate Pass**:
- Take `e_x`, `v`, and the lookup table index from the range reduction step of fast pass.
- Perform 3 more range reduction steps:
- Range reduction step 2: Use look-up tables of size 193 to reduce the argument to `[-0x1.3ffcp-15, 0x1.3e3dp-15]`
```
v2 = r2 * (1 + v) - 1 = (1 + s2) * (1 + v) - 1 = s2 + v + s2 * v
where r2 = 2^-16 * round ( 2^16 / (1 + k * 2^-14) )
and k = trunc( v * 2^14 + 0.5 ).
```
- Range reduction step 3: Use look-up tables of size 161 to reduce the argument to `[-0x1.01928p-22 , 0x1p-22]`
```
v3 = r3 * (1 + v2) - 1 = (1 + s3) * (1 + v2) - 1 = s3 + v2 + s3 * v2
where r3 = 2^-21 * round ( 2^21 / (1 + k * 2^-21) )
and k = trunc( v * 2^21 + 0.5 ).
```
- Range reduction step 4: Use look-up tables of size 130 to reduce the argument to `[-0x1.0002143p-29 , 0x1p-29]`
```
v4 = r4 * (1 + v3) - 1 = (1 + s4) * (1 + v3) - 1 = s4 + v3 + s4 * v3
where r4 = 2^-28 * round ( 2^28 / (1 + k * 2^-28) )
and k = trunc( v * 2^28 + 0.5 ).
```
- Polynomial approximation: approximate `log10(1 + v4)` by a degree-4 minimax polynomial generated by Sollya with:
```
> P = fpminimax(log10(1 + x)/x, 3, [|128...|], [-0x1.0002143p-29 , 0x1p-29]);
```
- Combine the results:
```
log10(x) ~ e_x * log10(2) - log10(r) - log10(r2) - log10(r3) - log10(r4) + v * P(v)
```
- The combined results are computed using floating points of 128-bit precision.
**Performance**
- For `0.5 <= x <= 2`, the fast pass hitting rate is about 99.92%.
- Reciprocal throughput from CORE-MATH's perf tool on Ryzen 5900X:
```
$ ./perf.sh log10
GNU libc version: 2.35
GNU libc release: stable
-- CORE-MATH reciprocal throughput -- with FMA
[####################] 100 %
Ntrial = 20 ; Min = 20.402 + 0.589 clc/call; Median-Min = 0.277 clc/call; Max = 22.752 clc/call;
-- CORE-MATH reciprocal throughput -- without FMA (-march=x86-64-v2)
[####################] 100 %
Ntrial = 20 ; Min = 75.797 + 3.317 clc/call; Median-Min = 3.407 clc/call; Max = 79.371 clc/call;
-- System LIBC reciprocal throughput --
[####################] 100 %
Ntrial = 20 ; Min = 22.668 + 0.184 clc/call; Median-Min = 0.181 clc/call; Max = 23.205 clc/call;
-- LIBC reciprocal throughput -- with FMA
[####################] 100 %
Ntrial = 20 ; Min = 25.977 + 0.183 clc/call; Median-Min = 0.138 clc/call; Max = 26.283 clc/call;
-- LIBC reciprocal throughput -- without FMA
[####################] 100 %
Ntrial = 20 ; Min = 22.140 + 0.980 clc/call; Median-Min = 0.853 clc/call; Max = 23.790 clc/call;
```
- Latency from CORE-MATH's perf tool on Ryzen 5900X:
```
$ ./perf.sh log10 --latency
GNU libc version: 2.35
GNU libc release: stable
-- CORE-MATH latency -- with FMA
[####################] 100 %
Ntrial = 20 ; Min = 54.613 + 0.357 clc/call; Median-Min = 0.287 clc/call; Max = 55.701 clc/call;
-- CORE-MATH latency -- without FMA (-march=x86-64-v2)
[####################] 100 %
Ntrial = 20 ; Min = 79.681 + 0.482 clc/call; Median-Min = 0.294 clc/call; Max = 81.604 clc/call;
-- System LIBC latency --
[####################] 100 %
Ntrial = 20 ; Min = 61.532 + 0.208 clc/call; Median-Min = 0.199 clc/call; Max = 62.256 clc/call;
-- LIBC latency -- with FMA
[####################] 100 %
Ntrial = 20 ; Min = 41.510 + 0.205 clc/call; Median-Min = 0.244 clc/call; Max = 41.867 clc/call;
-- LIBC latency -- without FMA
[####################] 100 %
Ntrial = 20 ; Min = 55.669 + 0.240 clc/call; Median-Min = 0.280 clc/call; Max = 56.056 clc/call;
```
- Accurate pass latency:
```
$ ./perf.sh log10 --latency --simple_stat
GNU libc version: 2.35
GNU libc release: stable
-- CORE-MATH latency -- with FMA
640.688
-- CORE-MATH latency -- without FMA (-march=x86-64-v2)
667.354
-- LIBC latency -- with FMA
495.593
-- LIBC latency -- without FMA
504.143
```
Reviewed By: zimmermann6
Differential Revision: https://reviews.llvm.org/D150014
|
|
Reviewed By: lntue
Differential Revision: https://reviews.llvm.org/D148452
|
|
floating point exceptions.
Switch math functions to use libc_errno and fix some errno and
floating point exceptions
Reviewed By: sivachandra
Differential Revision: https://reviews.llvm.org/D145349
|
|
This part of the effort to make all test related pieces into the `test`
directory. This helps is excluding test related pieces in a straight
forward manner if LLVM_INCLUDE_TESTS is OFF. Future patches will also move
the MPFR wrapper and testutils into the 'test' directory.
|
|
|
|
|
|
Implement double precision log10 function correctly rounded for all
rounding modes. This implementation currently needs FMA instructions for
correctness.
Use 2 passes:
Fast pass:
- 1 step range reduction with a lookup table of `2^7 = 128` elements to reduce the ranges to `[-2^-7, 2^-7]`.
- Use a degree-7 minimax polynomial generated by Sollya, evaluated using a mixed of double-double and double precisions.
- Apply Ziv's test for accuracy.
Accurate pass:
- Apply 5 more range reduction steps to reduce the ranges further to [-2^-27, 2^-27].
- Use a degree-4 minimax polynomial generated by Sollya, evaluated using 192-bit precisions.
- By the result of Lefevre (add quote), this is more than enough for correct rounding to all rounding modes.
In progress: Adding detail documentations about the algorithm.
Depend on: https://reviews.llvm.org/D136799
Reviewed By: zimmermann6
Differential Revision: https://reviews.llvm.org/D139846
|
