-
Notifications
You must be signed in to change notification settings - Fork 11.8k
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 our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[libc][math][c23] Add exp2f16 C23 math function #101217
Merged
Merged
Changes from all commits
Commits
File filter
Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,21 @@ | ||
//===-- Implementation header for exp2f16 -----------------------*- C++ -*-===// | ||
// | ||
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. | ||
// See https://llvm.org/LICENSE.txt for license information. | ||
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception | ||
// | ||
//===----------------------------------------------------------------------===// | ||
|
||
#ifndef LLVM_LIBC_SRC_MATH_EXP2F16_H | ||
#define LLVM_LIBC_SRC_MATH_EXP2F16_H | ||
|
||
#include "src/__support/macros/config.h" | ||
#include "src/__support/macros/properties/types.h" | ||
|
||
namespace LIBC_NAMESPACE_DECL { | ||
|
||
float16 exp2f16(float16 x); | ||
|
||
} // namespace LIBC_NAMESPACE_DECL | ||
|
||
#endif // LLVM_LIBC_SRC_MATH_EXP2F16_H |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,134 @@ | ||
//===-- Half-precision 2^x function ---------------------------------------===// | ||
// | ||
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. | ||
// See https://llvm.org/LICENSE.txt for license information. | ||
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception | ||
// | ||
//===----------------------------------------------------------------------===// | ||
|
||
#include "src/math/exp2f16.h" | ||
#include "hdr/errno_macros.h" | ||
#include "hdr/fenv_macros.h" | ||
#include "src/__support/CPP/array.h" | ||
#include "src/__support/FPUtil/FEnvImpl.h" | ||
#include "src/__support/FPUtil/FPBits.h" | ||
#include "src/__support/FPUtil/PolyEval.h" | ||
#include "src/__support/FPUtil/except_value_utils.h" | ||
#include "src/__support/FPUtil/multiply_add.h" | ||
#include "src/__support/FPUtil/nearest_integer.h" | ||
#include "src/__support/FPUtil/rounding_mode.h" | ||
#include "src/__support/common.h" | ||
#include "src/__support/macros/config.h" | ||
#include "src/__support/macros/optimization.h" | ||
|
||
namespace LIBC_NAMESPACE_DECL { | ||
|
||
static constexpr fputil::ExceptValues<float16, 3> EXP2F16_EXCEPTS = {{ | ||
// (input, RZ output, RU offset, RD offset, RN offset) | ||
// x = 0x1.714p-11, exp2f16(x) = 0x1p+0 (RZ) | ||
{0x11c5U, 0x3c00U, 1U, 0U, 1U}, | ||
// x = -0x1.558p-4, exp2f16(x) = 0x1.e34p-1 (RZ) | ||
{0xad56U, 0x3b8dU, 1U, 0U, 0U}, | ||
// x = -0x1.d5cp-4, exp2f16(x) = 0x1.d8cp-1 (RZ) | ||
{0xaf57U, 0x3b63U, 1U, 0U, 0U}, | ||
}}; | ||
|
||
// Generated by Sollya with the following commands: | ||
// > display = hexadecimal; | ||
// > for i from 0 to 7 do printsingle(round(2^(i * 2^-3), SG, RN)); | ||
static constexpr cpp::array<uint32_t, 8> EXP2_MID_BITS = { | ||
0x3f80'0000U, 0x3f8b'95c2U, 0x3f98'37f0U, 0x3fa5'fed7U, | ||
0x3fb5'04f3U, 0x3fc5'672aU, 0x3fd7'44fdU, 0x3fea'c0c7U, | ||
}; | ||
|
||
LLVM_LIBC_FUNCTION(float16, exp2f16, (float16 x)) { | ||
using FPBits = fputil::FPBits<float16>; | ||
FPBits x_bits(x); | ||
|
||
uint16_t x_u = x_bits.uintval(); | ||
uint16_t x_abs = x_u & 0x7fffU; | ||
|
||
// When |x| >= 16, or x is NaN. | ||
if (LIBC_UNLIKELY(x_abs >= 0x4c00U)) { | ||
// exp2(NaN) = NaN | ||
if (x_bits.is_nan()) { | ||
if (x_bits.is_signaling_nan()) { | ||
fputil::raise_except_if_required(FE_INVALID); | ||
return FPBits::quiet_nan().get_val(); | ||
} | ||
|
||
return x; | ||
} | ||
|
||
// When x >= 16. | ||
if (x_bits.is_pos()) { | ||
// exp2(+inf) = +inf | ||
if (x_bits.is_inf()) | ||
return FPBits::inf().get_val(); | ||
|
||
switch (fputil::quick_get_round()) { | ||
case FE_TONEAREST: | ||
case FE_UPWARD: | ||
fputil::set_errno_if_required(ERANGE); | ||
fputil::raise_except_if_required(FE_OVERFLOW); | ||
return FPBits::inf().get_val(); | ||
default: | ||
return FPBits::max_normal().get_val(); | ||
} | ||
} | ||
|
||
// When x <= -25. | ||
if (x_u >= 0xce40U) { | ||
// exp2(-inf) = +0 | ||
if (x_bits.is_inf()) | ||
return FPBits::zero().get_val(); | ||
|
||
fputil::set_errno_if_required(ERANGE); | ||
fputil::raise_except_if_required(FE_UNDERFLOW | FE_INEXACT); | ||
|
||
if (fputil::fenv_is_round_up()) | ||
return FPBits::min_subnormal().get_val(); | ||
return FPBits::zero().get_val(); | ||
} | ||
} | ||
|
||
if (auto r = EXP2F16_EXCEPTS.lookup(x_u); LIBC_UNLIKELY(r.has_value())) | ||
return r.value(); | ||
|
||
// For -25 < x < 16, to compute 2^x, we perform the following range reduction: | ||
// find hi, mid, lo, such that: | ||
// x = hi + mid + lo, in which | ||
// hi is an integer, | ||
// mid * 2^3 is an integer, | ||
// -2^(-4) <= lo < 2^(-4). | ||
// In particular, | ||
// hi + mid = round(x * 2^3) * 2^(-3). | ||
// Then, | ||
// 2^x = 2^(hi + mid + lo) = 2^hi * 2^mid * 2^lo. | ||
// We store 2^mid in the lookup table EXP2_MID_BITS, and compute 2^hi * 2^mid | ||
// by adding hi to the exponent field of 2^mid. 2^lo is computed using a | ||
// degree-3 minimax polynomial generated by Sollya. | ||
|
||
float xf = x; | ||
float kf = fputil::nearest_integer(xf * 0x1.0p+3f); | ||
int x_hi_mid = static_cast<int>(kf); | ||
int x_hi = x_hi_mid >> 3; | ||
int x_mid = x_hi_mid & 0x7; | ||
// lo = x - (hi + mid) = round(x * 2^3) * (-2^(-3)) + x | ||
float lo = fputil::multiply_add(kf, -0x1.0p-3f, xf); | ||
|
||
uint32_t exp2_hi_mid_bits = | ||
EXP2_MID_BITS[x_mid] + | ||
static_cast<uint32_t>(x_hi << fputil::FPBits<float>::FRACTION_LEN); | ||
float exp2_hi_mid = fputil::FPBits<float>(exp2_hi_mid_bits).get_val(); | ||
// Degree-3 minimax polynomial generated by Sollya with the following | ||
// commands: | ||
// > display = hexadecimal; | ||
// > P = fpminimax((2^x - 1)/x, 2, [|SG...|], [-2^-4, 2^-4]); | ||
// > 1 + x * P; | ||
float exp2_lo = fputil::polyeval(lo, 0x1p+0f, 0x1.62e43p-1f, 0x1.ec0aa6p-3f, | ||
0x1.c6b4a6p-5f); | ||
return static_cast<float16>(exp2_hi_mid * exp2_lo); | ||
} | ||
|
||
} // namespace LIBC_NAMESPACE_DECL |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,40 @@ | ||
//===-- Exhaustive test for exp2f16 ---------------------------------------===// | ||
// | ||
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. | ||
// See https://llvm.org/LICENSE.txt for license information. | ||
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception | ||
// | ||
//===----------------------------------------------------------------------===// | ||
|
||
#include "src/math/exp2f16.h" | ||
#include "test/UnitTest/FPMatcher.h" | ||
#include "test/UnitTest/Test.h" | ||
#include "utils/MPFRWrapper/MPFRUtils.h" | ||
|
||
using LlvmLibcExp2f16Test = LIBC_NAMESPACE::testing::FPTest<float16>; | ||
|
||
namespace mpfr = LIBC_NAMESPACE::testing::mpfr; | ||
|
||
// Range: [0, Inf]; | ||
static constexpr uint16_t POS_START = 0x0000U; | ||
static constexpr uint16_t POS_STOP = 0x7c00U; | ||
|
||
// Range: [-Inf, 0]; | ||
static constexpr uint16_t NEG_START = 0x8000U; | ||
static constexpr uint16_t NEG_STOP = 0xfc00U; | ||
|
||
TEST_F(LlvmLibcExp2f16Test, PositiveRange) { | ||
for (uint16_t v = POS_START; v <= POS_STOP; ++v) { | ||
float16 x = FPBits(v).get_val(); | ||
EXPECT_MPFR_MATCH_ALL_ROUNDING(mpfr::Operation::Exp2, x, | ||
LIBC_NAMESPACE::exp2f16(x), 0.5); | ||
} | ||
} | ||
|
||
TEST_F(LlvmLibcExp2f16Test, NegativeRange) { | ||
for (uint16_t v = NEG_START; v <= NEG_STOP; ++v) { | ||
float16 x = FPBits(v).get_val(); | ||
EXPECT_MPFR_MATCH_ALL_ROUNDING(mpfr::Operation::Exp2, x, | ||
LIBC_NAMESPACE::exp2f16(x), 0.5); | ||
} | ||
} |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,22 @@ | ||
//===-- Performance test for exp2f16 --------------------------------------===// | ||
// | ||
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. | ||
// See https://llvm.org/LICENSE.txt for license information. | ||
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception | ||
// | ||
//===----------------------------------------------------------------------===// | ||
|
||
#include "SingleInputSingleOutputPerf.h" | ||
|
||
#include "src/math/exp2f16.h" | ||
|
||
// LLVM libc might be the only libc implementation with support for float16 math | ||
// functions currently. We can't compare our float16 functions against the | ||
// system libc, so we compare them against this placeholder function. | ||
static float16 placeholderf16(float16 x) { return x; } | ||
|
||
int main() { | ||
SINGLE_INPUT_SINGLE_OUTPUT_PERF_EX(float16, LIBC_NAMESPACE::exp2f16, | ||
::placeholderf16, 20'000, | ||
"exp2f16_perf.log") | ||
} |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This shouldn't change anything yet. I added it now so that we don't forget it when the day comes that we remove the
placeholderf16
function and benchmark against the system libc's function.There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
What's this testing exactly? And if we want to disable a single builtin it's better to just do
-fno-builtin-exp2f16
if that's what it's called.There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
All the math performance tests, except one, benchmark LLVM libc functions against respective system libc functions, so all of them except that one have
-fno-builtin
. I don't know why fine-grained-fno-builtin-<function>
flags weren't used.In this case, since the system libc shouldn't have
float16
math functions yet, I benchmark against a placeholder function for now. It would probably be better if at some point we just rewrote our performance testing code to use Google Benchmark, and also allow us to benchmark a single function instead of comparing two, while we're at it.