Skip to content

[libc][math] Refactor asinhf implementation to header-only in src/__support/math folder. #150843

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

Merged
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions libc/shared/math.h
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,7 @@
#include "math/asin.h"
#include "math/asinf.h"
#include "math/asinf16.h"
#include "math/asinhf.h"
#include "math/erff.h"
#include "math/exp.h"
#include "math/exp10.h"
Expand Down
23 changes: 23 additions & 0 deletions libc/shared/math/asinhf.h
Original file line number Diff line number Diff line change
@@ -0,0 +1,23 @@
//===-- Shared asinhf function ----------------------------------*- 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_SHARED_MATH_ASINHF_H
#define LLVM_LIBC_SHARED_MATH_ASINHF_H

#include "shared/libc_common.h"
#include "src/__support/math/asinhf.h"

namespace LIBC_NAMESPACE_DECL {
namespace shared {

using math::asinhf;

} // namespace shared
} // namespace LIBC_NAMESPACE_DECL

#endif // LLVM_LIBC_SHARED_MATH_ASINHF_H
14 changes: 14 additions & 0 deletions libc/src/__support/math/CMakeLists.txt
Original file line number Diff line number Diff line change
Expand Up @@ -140,6 +140,20 @@ add_header_library(
libc.src.__support.macros.properties.cpu_features
)

add_header_library(
asinhf
HDRS
asinhf.h
DEPENDS
.acoshf_utils
libc.src.__support.FPUtil.fp_bits
libc.src.__support.FPUtil.polyeval
libc.src.__support.FPUtil.multiply_add
libc.src.__support.FPUtil.sqrt
libc.src.__support.macros.config
libc.src.__support.macros.optimization
)

add_header_library(
asinf
HDRS
Expand Down
125 changes: 125 additions & 0 deletions libc/src/__support/math/asinhf.h
Original file line number Diff line number Diff line change
@@ -0,0 +1,125 @@
//===-- Implementation header for asinf -------------------------*- 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___SUPPORT_MATH_ASINHF_H
#define LLVM_LIBC_SRC___SUPPORT_MATH_ASINHF_H

#include "acoshf_utils.h"
#include "src/__support/FPUtil/FPBits.h"
#include "src/__support/FPUtil/PolyEval.h"
#include "src/__support/FPUtil/multiply_add.h"
#include "src/__support/FPUtil/sqrt.h"
#include "src/__support/macros/config.h"
#include "src/__support/macros/optimization.h" // LIBC_UNLIKELY

namespace LIBC_NAMESPACE_DECL {

namespace math {

LIBC_INLINE static constexpr float asinhf(float x) {
using namespace acoshf_internal;
using FPBits_t = typename fputil::FPBits<float>;
FPBits_t xbits(x);
uint32_t x_u = xbits.uintval();
uint32_t x_abs = xbits.abs().uintval();

// |x| <= 2^-3
if (LIBC_UNLIKELY(x_abs <= 0x3e80'0000U)) {
// |x| <= 2^-26
if (LIBC_UNLIKELY(x_abs <= 0x3280'0000U)) {
return static_cast<float>(LIBC_UNLIKELY(x_abs == 0)
? x
: (x - 0x1.5555555555555p-3 * x * x * x));
}

double x_d = x;
double x_sq = x_d * x_d;
// Generated by Sollya with:
// > P = fpminimax(asinh(x)/x, [|0, 2, 4, 6, 8, 10, 12, 14, 16|], [|D...|],
// [0, 2^-2]);
double p = fputil::polyeval(
x_sq, 0.0, -0x1.555555555551ep-3, 0x1.3333333325495p-4,
-0x1.6db6db5a7622bp-5, 0x1.f1c70f82928c6p-6, -0x1.6e893934266b7p-6,
0x1.1c0b41d3fbe78p-6, -0x1.c0f47810b3c4fp-7, 0x1.2c8602690143dp-7);
return static_cast<float>(fputil::multiply_add(x_d, p, x_d));
}

const double SIGN[2] = {1.0, -1.0};
double x_sign = SIGN[x_u >> 31];
double x_d = x;

#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS
// Helper functions to set results for exceptional cases.
auto round_result_slightly_down = [x_sign](float r) -> float {
return fputil::multiply_add(static_cast<float>(x_sign), r,
static_cast<float>(x_sign) * (-0x1.0p-24f));
};
auto round_result_slightly_up = [x_sign](float r) -> float {
return fputil::multiply_add(static_cast<float>(x_sign), r,
static_cast<float>(x_sign) * 0x1.0p-24f);
};

if (LIBC_UNLIKELY(x_abs >= 0x4bdd'65a5U)) {
if (LIBC_UNLIKELY(xbits.is_inf_or_nan())) {
if (xbits.is_signaling_nan()) {
fputil::raise_except_if_required(FE_INVALID);
return FPBits_t::quiet_nan().get_val();
}

return x;
}

// Exceptional cases when x > 2^24.
switch (x_abs) {
case 0x4bdd65a5: // |x| = 0x1.bacb4ap24f
return round_result_slightly_down(0x1.1e0696p4f);
case 0x4c803f2c: // |x| = 0x1.007e58p26f
return round_result_slightly_down(0x1.2b786cp4f);
case 0x4f8ffb03: // |x| = 0x1.1ff606p32f
return round_result_slightly_up(0x1.6fdd34p4f);
case 0x5c569e88: // |x| = 0x1.ad3d1p57f
return round_result_slightly_up(0x1.45c146p5f);
case 0x5e68984e: // |x| = 0x1.d1309cp61f
return round_result_slightly_up(0x1.5c9442p5f);
case 0x655890d3: // |x| = 0x1.b121a6p75f
return round_result_slightly_down(0x1.a9a3f2p5f);
case 0x65de7ca6: // |x| = 0x1.bcf94cp76f
return round_result_slightly_up(0x1.af66cp5f);
case 0x6eb1a8ec: // |x| = 0x1.6351d8p94f
return round_result_slightly_down(0x1.08b512p6f);
case 0x7997f30a: // |x| = 0x1.2fe614p116f
return round_result_slightly_up(0x1.451436p6f);
}
} else {
// Exceptional cases when x < 2^24.
if (LIBC_UNLIKELY(x_abs == 0x45abaf26)) {
// |x| = 0x1.575e4cp12f
return round_result_slightly_down(0x1.29becap3f);
}
if (LIBC_UNLIKELY(x_abs == 0x49d29048)) {
// |x| = 0x1.a5209p20f
return round_result_slightly_down(0x1.e1b92p3f);
}
}
#else
if (LIBC_UNLIKELY(xbits.is_inf_or_nan()))
return x;
#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS

// asinh(x) = log(x + sqrt(x^2 + 1))
return static_cast<float>(
x_sign * log_eval(fputil::multiply_add(
x_d, x_sign,
fputil::sqrt<double>(fputil::multiply_add(x_d, x_d, 1.0)))));
}

} // namespace math

} // namespace LIBC_NAMESPACE_DECL

#endif // LLVM_LIBC_SRC___SUPPORT_MATH_ASINHF_H
7 changes: 1 addition & 6 deletions libc/src/math/generic/CMakeLists.txt
Original file line number Diff line number Diff line change
Expand Up @@ -3889,12 +3889,7 @@ add_entrypoint_object(
HDRS
../asinhf.h
DEPENDS
.explogxf
libc.src.__support.FPUtil.fp_bits
libc.src.__support.FPUtil.multiply_add
libc.src.__support.FPUtil.polyeval
libc.src.__support.FPUtil.sqrt
libc.src.__support.macros.optimization
libc.src.__support.math.asinhf
)

add_entrypoint_object(
Expand Down
106 changes: 2 additions & 104 deletions libc/src/math/generic/asinhf.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -7,112 +7,10 @@
//===----------------------------------------------------------------------===//

#include "src/math/asinhf.h"
#include "src/__support/FPUtil/FPBits.h"
#include "src/__support/FPUtil/PolyEval.h"
#include "src/__support/FPUtil/multiply_add.h"
#include "src/__support/FPUtil/sqrt.h"
#include "src/__support/macros/config.h"
#include "src/__support/macros/optimization.h" // LIBC_UNLIKELY
#include "src/math/generic/common_constants.h"
#include "src/math/generic/explogxf.h"
#include "src/__support/math/asinhf.h"

namespace LIBC_NAMESPACE_DECL {

LLVM_LIBC_FUNCTION(float, asinhf, (float x)) {
using namespace acoshf_internal;
using FPBits_t = typename fputil::FPBits<float>;
FPBits_t xbits(x);
uint32_t x_u = xbits.uintval();
uint32_t x_abs = xbits.abs().uintval();

// |x| <= 2^-3
if (LIBC_UNLIKELY(x_abs <= 0x3e80'0000U)) {
// |x| <= 2^-26
if (LIBC_UNLIKELY(x_abs <= 0x3280'0000U)) {
return static_cast<float>(LIBC_UNLIKELY(x_abs == 0)
? x
: (x - 0x1.5555555555555p-3 * x * x * x));
}

double x_d = x;
double x_sq = x_d * x_d;
// Generated by Sollya with:
// > P = fpminimax(asinh(x)/x, [|0, 2, 4, 6, 8, 10, 12, 14, 16|], [|D...|],
// [0, 2^-2]);
double p = fputil::polyeval(
x_sq, 0.0, -0x1.555555555551ep-3, 0x1.3333333325495p-4,
-0x1.6db6db5a7622bp-5, 0x1.f1c70f82928c6p-6, -0x1.6e893934266b7p-6,
0x1.1c0b41d3fbe78p-6, -0x1.c0f47810b3c4fp-7, 0x1.2c8602690143dp-7);
return static_cast<float>(fputil::multiply_add(x_d, p, x_d));
}

const double SIGN[2] = {1.0, -1.0};
double x_sign = SIGN[x_u >> 31];
double x_d = x;

#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS
// Helper functions to set results for exceptional cases.
auto round_result_slightly_down = [x_sign](float r) -> float {
return fputil::multiply_add(static_cast<float>(x_sign), r,
static_cast<float>(x_sign) * (-0x1.0p-24f));
};
auto round_result_slightly_up = [x_sign](float r) -> float {
return fputil::multiply_add(static_cast<float>(x_sign), r,
static_cast<float>(x_sign) * 0x1.0p-24f);
};

if (LIBC_UNLIKELY(x_abs >= 0x4bdd'65a5U)) {
if (LIBC_UNLIKELY(xbits.is_inf_or_nan())) {
if (xbits.is_signaling_nan()) {
fputil::raise_except_if_required(FE_INVALID);
return FPBits_t::quiet_nan().get_val();
}

return x;
}

// Exceptional cases when x > 2^24.
switch (x_abs) {
case 0x4bdd65a5: // |x| = 0x1.bacb4ap24f
return round_result_slightly_down(0x1.1e0696p4f);
case 0x4c803f2c: // |x| = 0x1.007e58p26f
return round_result_slightly_down(0x1.2b786cp4f);
case 0x4f8ffb03: // |x| = 0x1.1ff606p32f
return round_result_slightly_up(0x1.6fdd34p4f);
case 0x5c569e88: // |x| = 0x1.ad3d1p57f
return round_result_slightly_up(0x1.45c146p5f);
case 0x5e68984e: // |x| = 0x1.d1309cp61f
return round_result_slightly_up(0x1.5c9442p5f);
case 0x655890d3: // |x| = 0x1.b121a6p75f
return round_result_slightly_down(0x1.a9a3f2p5f);
case 0x65de7ca6: // |x| = 0x1.bcf94cp76f
return round_result_slightly_up(0x1.af66cp5f);
case 0x6eb1a8ec: // |x| = 0x1.6351d8p94f
return round_result_slightly_down(0x1.08b512p6f);
case 0x7997f30a: // |x| = 0x1.2fe614p116f
return round_result_slightly_up(0x1.451436p6f);
}
} else {
// Exceptional cases when x < 2^24.
if (LIBC_UNLIKELY(x_abs == 0x45abaf26)) {
// |x| = 0x1.575e4cp12f
return round_result_slightly_down(0x1.29becap3f);
}
if (LIBC_UNLIKELY(x_abs == 0x49d29048)) {
// |x| = 0x1.a5209p20f
return round_result_slightly_down(0x1.e1b92p3f);
}
}
#else
if (LIBC_UNLIKELY(xbits.is_inf_or_nan()))
return x;
#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS

// asinh(x) = log(x + sqrt(x^2 + 1))
return static_cast<float>(
x_sign * log_eval(fputil::multiply_add(
x_d, x_sign,
fputil::sqrt<double>(fputil::multiply_add(x_d, x_d, 1.0)))));
}
LLVM_LIBC_FUNCTION(float, asinhf, (float x)) { return math::asinhf(x); }

} // namespace LIBC_NAMESPACE_DECL
1 change: 1 addition & 0 deletions libc/test/shared/CMakeLists.txt
Original file line number Diff line number Diff line change
Expand Up @@ -16,6 +16,7 @@ add_fp_unittest(
libc.src.__support.math.asin
libc.src.__support.math.asinf
libc.src.__support.math.asinf16
libc.src.__support.math.asinhf
libc.src.__support.math.erff
libc.src.__support.math.exp
libc.src.__support.math.exp10
Expand Down
1 change: 1 addition & 0 deletions libc/test/shared/shared_math_test.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -42,6 +42,7 @@ TEST(LlvmLibcSharedMathTest, AllFloat) {
EXPECT_FP_EQ(0x1.921fb6p+0, LIBC_NAMESPACE::shared::acosf(0.0f));
EXPECT_FP_EQ(0x0p+0f, LIBC_NAMESPACE::shared::acoshf(1.0f));
EXPECT_FP_EQ(0x0p+0f, LIBC_NAMESPACE::shared::asinf(0.0f));
EXPECT_FP_EQ(0x0p+0f, LIBC_NAMESPACE::shared::asinhf(0.0f));
EXPECT_FP_EQ(0x0p+0f, LIBC_NAMESPACE::shared::erff(0.0f));
EXPECT_FP_EQ(0x1p+0f, LIBC_NAMESPACE::shared::exp10f(0.0f));
EXPECT_FP_EQ(0x1p+0f, LIBC_NAMESPACE::shared::expf(0.0f));
Expand Down
23 changes: 15 additions & 8 deletions utils/bazel/llvm-project-overlay/libc/BUILD.bazel
Original file line number Diff line number Diff line change
Expand Up @@ -2231,6 +2231,20 @@ libc_support_library(
],
)

libc_support_library(
name = "__support_math_asinhf",
hdrs = ["src/__support/math/asinhf.h"],
deps = [
":__support_math_acoshf_utils",
":__support_fputil_fp_bits",
":__support_fputil_polyeval",
":__support_fputil_multiply_add",
":__support_fputil_sqrt",
":__support_macros_config",
":__support_macros_optimization",
],
)

libc_support_library(
name = "__support_math_asinf",
hdrs = ["src/__support/math/asinf.h"],
Expand Down Expand Up @@ -2829,14 +2843,7 @@ libc_math_function(
libc_math_function(
name = "asinhf",
additional_deps = [
":__support_fputil_fma",
":__support_fputil_multiply_add",
":__support_fputil_nearest_integer",
":__support_fputil_polyeval",
":__support_fputil_sqrt",
":__support_macros_optimization",
":common_constants",
":explogxf",
":__support_math_asinhf",
],
)

Expand Down
Loading