[libc][math][c23] Add exp2m1f16 C23 math function #105690
Conversation
|
@llvm/pr-subscribers-libc Author: OverMighty (overmighty) ChangesPart of #95250. Patch is 21.96 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/105690.diff 13 Files Affected:
diff --git a/libc/config/linux/x86_64/entrypoints.txt b/libc/config/linux/x86_64/entrypoints.txt
index 65c5757efe6274..06ca526c671ea8 100644
--- a/libc/config/linux/x86_64/entrypoints.txt
+++ b/libc/config/linux/x86_64/entrypoints.txt
@@ -596,6 +596,7 @@ if(LIBC_TYPES_HAS_FLOAT16)
libc.src.math.copysignf16
libc.src.math.exp10f16
libc.src.math.exp2f16
+ libc.src.math.exp2m1f16
libc.src.math.expf16
libc.src.math.expm1f16
libc.src.math.f16add
diff --git a/libc/docs/math/index.rst b/libc/docs/math/index.rst
index 185d2d440849a0..0d2bbdd3381d19 100644
--- a/libc/docs/math/index.rst
+++ b/libc/docs/math/index.rst
@@ -292,7 +292,7 @@ Higher Math Functions
+-----------+------------------+-----------------+------------------------+----------------------+------------------------+------------------------+----------------------------+
| exp2 | |check| | |check| | | |check| | | 7.12.6.4 | F.10.3.4 |
+-----------+------------------+-----------------+------------------------+----------------------+------------------------+------------------------+----------------------------+
-| exp2m1 | |check| | | | | | 7.12.6.5 | F.10.3.5 |
+| exp2m1 | |check| | | | |check| | | 7.12.6.5 | F.10.3.5 |
+-----------+------------------+-----------------+------------------------+----------------------+------------------------+------------------------+----------------------------+
| expm1 | |check| | |check| | | |check| | | 7.12.6.6 | F.10.3.6 |
+-----------+------------------+-----------------+------------------------+----------------------+------------------------+------------------------+----------------------------+
diff --git a/libc/spec/stdc.td b/libc/spec/stdc.td
index 118dcce829be23..c75343d567d7f9 100644
--- a/libc/spec/stdc.td
+++ b/libc/spec/stdc.td
@@ -596,6 +596,7 @@ def StdC : StandardSpec<"stdc"> {
GuardedFunctionSpec<"exp2f16", RetValSpec<Float16Type>, [ArgSpec<Float16Type>], "LIBC_TYPES_HAS_FLOAT16">,
FunctionSpec<"exp2m1f", RetValSpec<FloatType>, [ArgSpec<FloatType>]>,
+ GuardedFunctionSpec<"exp2m1f16", RetValSpec<Float16Type>, [ArgSpec<Float16Type>], "LIBC_TYPES_HAS_FLOAT16">,
FunctionSpec<"expm1", RetValSpec<DoubleType>, [ArgSpec<DoubleType>]>,
FunctionSpec<"expm1f", RetValSpec<FloatType>, [ArgSpec<FloatType>]>,
diff --git a/libc/src/math/CMakeLists.txt b/libc/src/math/CMakeLists.txt
index 3cba34fc249322..bd18c68a5e2fd6 100644
--- a/libc/src/math/CMakeLists.txt
+++ b/libc/src/math/CMakeLists.txt
@@ -115,6 +115,7 @@ add_math_entrypoint_object(exp2f)
add_math_entrypoint_object(exp2f16)
add_math_entrypoint_object(exp2m1f)
+add_math_entrypoint_object(exp2m1f16)
add_math_entrypoint_object(exp10)
add_math_entrypoint_object(exp10f)
diff --git a/libc/src/math/exp2m1f16.h b/libc/src/math/exp2m1f16.h
new file mode 100644
index 00000000000000..39fcebd5a69300
--- /dev/null
+++ b/libc/src/math/exp2m1f16.h
@@ -0,0 +1,21 @@
+//===-- Implementation header for exp2m1f16 ---------------------*- 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_EXP2M1F16_H
+#define LLVM_LIBC_SRC_MATH_EXP2M1F16_H
+
+#include "src/__support/macros/config.h"
+#include "src/__support/macros/properties/types.h"
+
+namespace LIBC_NAMESPACE_DECL {
+
+float16 exp2m1f16(float16 x);
+
+} // namespace LIBC_NAMESPACE_DECL
+
+#endif // LLVM_LIBC_SRC_MATH_EXP2M1F16_H
diff --git a/libc/src/math/generic/CMakeLists.txt b/libc/src/math/generic/CMakeLists.txt
index 350072f4b9649d..c565dd5cb04f5a 100644
--- a/libc/src/math/generic/CMakeLists.txt
+++ b/libc/src/math/generic/CMakeLists.txt
@@ -1441,13 +1441,9 @@ add_entrypoint_object(
.expxf16
libc.hdr.errno_macros
libc.hdr.fenv_macros
- libc.src.__support.CPP.array
libc.src.__support.FPUtil.except_value_utils
libc.src.__support.FPUtil.fenv_impl
libc.src.__support.FPUtil.fp_bits
- libc.src.__support.FPUtil.multiply_add
- libc.src.__support.FPUtil.nearest_integer
- libc.src.__support.FPUtil.polyeval
libc.src.__support.FPUtil.rounding_mode
libc.src.__support.macros.optimization
COMPILE_OPTIONS
@@ -1475,6 +1471,28 @@ add_entrypoint_object(
-O3
)
+add_entrypoint_object(
+ exp2m1f16
+ SRCS
+ exp2m1f16.cpp
+ HDRS
+ ../exp2m1f16.h
+ DEPENDS
+ .expxf16
+ libc.hdr.errno_macros
+ libc.hdr.fenv_macros
+ libc.src.__support.common
+ libc.src.__support.FPUtil.except_value_utils
+ libc.src.__support.FPUtil.fenv_impl
+ libc.src.__support.FPUtil.fp_bits
+ libc.src.__support.FPUtil.multiply_add
+ libc.src.__support.FPUtil.polyeval
+ libc.src.__support.FPUtil.rounding_mode
+ libc.src.__support.macros.optimization
+ COMPILE_OPTIONS
+ -O3
+)
+
add_entrypoint_object(
exp10
SRCS
diff --git a/libc/src/math/generic/exp2f16.cpp b/libc/src/math/generic/exp2f16.cpp
index 66b79567040053..09d61d04d4cda9 100644
--- a/libc/src/math/generic/exp2f16.cpp
+++ b/libc/src/math/generic/exp2f16.cpp
@@ -10,13 +10,9 @@
#include "expxf16.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"
@@ -88,39 +84,8 @@ LLVM_LIBC_FUNCTION(float16, exp2f16, (float16 x)) {
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);
+ // exp2(x) = exp2(hi + mid) * exp2(lo)
+ auto [exp2_hi_mid, exp2_lo] = exp2_range_reduction(x);
return static_cast<float16>(exp2_hi_mid * exp2_lo);
}
diff --git a/libc/src/math/generic/exp2m1f16.cpp b/libc/src/math/generic/exp2m1f16.cpp
new file mode 100644
index 00000000000000..9d01e80ea59e2a
--- /dev/null
+++ b/libc/src/math/generic/exp2m1f16.cpp
@@ -0,0 +1,161 @@
+//===-- Half-precision 2^x - 1 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/exp2m1f16.h"
+#include "expxf16.h"
+#include "hdr/errno_macros.h"
+#include "hdr/fenv_macros.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/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, 6> EXP2M1F16_EXCEPTS_LO = {{
+ // (input, RZ output, RU offset, RD offset, RN offset)
+ // x = 0x1.cf4p-13, exp2m1f16(x) = 0x1.41p-13 (RZ)
+ {0x0b3dU, 0x0904U, 1U, 0U, 1U},
+ // x = 0x1.4fcp-12, exp2m1f16(x) = 0x1.d14p-13 (RZ)
+ {0x0d3fU, 0x0b45U, 1U, 0U, 1U},
+ // x = 0x1.63p-11, exp2m1f16(x) = 0x1.ec4p-12 (RZ)
+ {0x118cU, 0x0fb1U, 1U, 0U, 0U},
+ // x = 0x1.6fp-7, exp2m1f16(x) = 0x1.fe8p-8 (RZ)
+ {0x21bcU, 0x1ffaU, 1U, 0U, 1U},
+ // x = -0x1.c6p-10, exp2m1f16(x) = -0x1.3a8p-10 (RZ)
+ {0x9718U, 0x94eaU, 0U, 1U, 0U},
+ // x = -0x1.cfcp-10, exp2m1f16(x) = -0x1.414p-10 (RZ)
+ {0x973fU, 0x9505U, 0U, 1U, 0U},
+}};
+
+#ifdef LIBC_TARGET_CPU_HAS_FMA
+static constexpr size_t N_EXP2M1F16_EXCEPTS_HI = 6;
+#else
+static constexpr size_t N_EXP2M1F16_EXCEPTS_HI = 7;
+#endif
+
+static constexpr fputil::ExceptValues<float16, N_EXP2M1F16_EXCEPTS_HI>
+ EXP2M1F16_EXCEPTS_HI = {{
+ // (input, RZ output, RU offset, RD offset, RN offset)
+ // x = 0x1.e58p-3, exp2m1f16(x) = 0x1.6dcp-3 (RZ)
+ {0x3396U, 0x31b7U, 1U, 0U, 0U},
+#ifndef LIBC_TARGET_CPU_HAS_FMA
+ // x = 0x1.2e8p-2, exp2m1f16(x) = 0x1.d14p-3 (RZ)
+ {0x34baU, 0x3345U, 1U, 0U, 0U},
+#endif
+ // x = 0x1.ad8p-2, exp2m1f16(x) = 0x1.598p-2 (RZ)
+ {0x36b6U, 0x3566U, 1U, 0U, 0U},
+#ifdef LIBC_TARGET_CPU_HAS_FMA
+ // x = 0x1.edcp-2, exp2m1f16(x) = 0x1.964p-2 (RZ)
+ {0x37b7U, 0x3659U, 1U, 0U, 1U},
+#endif
+ // x = -0x1.804p-3, exp2m1f16(x) = -0x1.f34p-4 (RZ)
+ {0xb201U, 0xafcdU, 0U, 1U, 1U},
+ // x = -0x1.f3p-3, exp2m1f16(x) = -0x1.3e4p-3 (RZ)
+ {0xb3ccU, 0xb0f9U, 0U, 1U, 0U},
+ // x = -0x1.294p-1, exp2m1f16(x) = -0x1.53p-2 (RZ)
+ {0xb8a5U, 0xb54cU, 0U, 1U, 1U},
+#ifndef LIBC_TARGET_CPU_HAS_FMA
+ // x = -0x1.a34p-1, exp2m1f16(x) = -0x1.bb4p-2 (RZ)
+ {0xba8dU, 0xb6edU, 0U, 1U, 1U},
+#endif
+ }};
+
+LLVM_LIBC_FUNCTION(float16, exp2m1f16, (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| <= 2^(-3), or |x| >= 11, or x is NaN.
+ if (LIBC_UNLIKELY(x_abs <= 0x3000U || x_abs >= 0x4980U)) {
+ // exp2m1(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_u >= 0x4c00 && x_bits.is_pos()) {
+ // exp2m1(+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 | FE_INEXACT);
+ return FPBits::inf().get_val();
+ default:
+ return FPBits::max_normal().get_val();
+ }
+ }
+
+ // When x < -11.
+ if (x_u > 0xc980U) {
+ // exp2m1(-inf) = -1
+ if (x_bits.is_inf())
+ return FPBits::one(Sign::NEG).get_val();
+
+ // When -12 < x < -11, round(2^x - 1, HP, RN) = -0x1.ffcp-1.
+ if (x_u < 0xca00U) {
+ return fputil::round_result_slightly_down(
+ static_cast<float16>(-0x1.ffcp-1));
+ }
+
+ // When x <= -12, round(2^x - 1, HP, RN) = -1.
+ switch (fputil::quick_get_round()) {
+ case FE_TONEAREST:
+ case FE_DOWNWARD:
+ return FPBits::one(Sign::NEG).get_val();
+ default:
+ return static_cast<float16>(-0x1.ffcp-1);
+ }
+ }
+
+ // When |x| <= 2^(-3).
+ if (x_abs <= 0x3000U) {
+ if (auto r = EXP2M1F16_EXCEPTS_LO.lookup(x_u);
+ LIBC_UNLIKELY(r.has_value()))
+ return r.value();
+
+ float xf = x;
+ // Degree-5 minimax polynomial generated by Sollya with the following
+ // commands:
+ // > display = hexadecimal;
+ // > P = fpminimax((2^x - 1)/x, 4, [|SG...|], [-2^-3, 2^-3]);
+ // > x * P;
+ return static_cast<float16>(
+ xf * fputil::polyeval(xf, 0x1.62e43p-1f, 0x1.ebfbdep-3f,
+ 0x1.c6af88p-5f, 0x1.3b45d6p-7f,
+ 0x1.641e7cp-10f));
+ }
+ }
+
+ if (auto r = EXP2M1F16_EXCEPTS_HI.lookup(x_u); LIBC_UNLIKELY(r.has_value()))
+ return r.value();
+
+ // exp2(x) = exp2(hi + mid) * exp2(lo)
+ auto [exp2_hi_mid, exp2_lo] = exp2_range_reduction(x);
+ // exp2m1(x) = exp2(hi + mid) * exp2(lo) - 1
+ return static_cast<float16>(
+ fputil::multiply_add(exp2_hi_mid, exp2_lo, -1.0f));
+}
+
+} // namespace LIBC_NAMESPACE_DECL
diff --git a/libc/src/math/generic/expxf16.h b/libc/src/math/generic/expxf16.h
index a0db6cee438e92..35294130a15007 100644
--- a/libc/src/math/generic/expxf16.h
+++ b/libc/src/math/generic/expxf16.h
@@ -10,6 +10,7 @@
#define LLVM_LIBC_SRC_MATH_GENERIC_EXPXF16_H
#include "src/__support/CPP/array.h"
+#include "src/__support/FPUtil/FPBits.h"
#include "src/__support/FPUtil/PolyEval.h"
#include "src/__support/FPUtil/multiply_add.h"
#include "src/__support/FPUtil/nearest_integer.h"
@@ -89,6 +90,43 @@ constexpr cpp::array<uint32_t, 8> EXP2_MID_BITS = {
0x3fb5'04f3U, 0x3fc5'672aU, 0x3fd7'44fdU, 0x3fea'c0c7U,
};
+LIBC_INLINE ExpRangeReduction exp2_range_reduction(float16 x) {
+ // 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 {exp2_hi_mid, exp2_lo};
+}
+
} // namespace LIBC_NAMESPACE_DECL
#endif // LLVM_LIBC_SRC_MATH_GENERIC_EXPXF16_H
diff --git a/libc/test/src/math/CMakeLists.txt b/libc/test/src/math/CMakeLists.txt
index 0c4118c3694548..07a9405081f97d 100644
--- a/libc/test/src/math/CMakeLists.txt
+++ b/libc/test/src/math/CMakeLists.txt
@@ -1003,6 +1003,17 @@ add_fp_unittest(
libc.src.__support.FPUtil.fp_bits
)
+add_fp_unittest(
+ exp2m1f16_test
+ NEED_MPFR
+ SUITE
+ libc-math-unittests
+ SRCS
+ exp2m1f16_test.cpp
+ DEPENDS
+ libc.src.math.exp2m1f16
+)
+
add_fp_unittest(
exp10_test
NEED_MPFR
diff --git a/libc/test/src/math/exp2m1f16_test.cpp b/libc/test/src/math/exp2m1f16_test.cpp
new file mode 100644
index 00000000000000..bf299118429d7e
--- /dev/null
+++ b/libc/test/src/math/exp2m1f16_test.cpp
@@ -0,0 +1,40 @@
+//===-- Exhaustive test for exp2m1f16 -------------------------------------===//
+//
+// 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/exp2m1f16.h"
+#include "test/UnitTest/FPMatcher.h"
+#include "test/UnitTest/Test.h"
+#include "utils/MPFRWrapper/MPFRUtils.h"
+
+using LlvmLibcExp2m1f16Test = 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(LlvmLibcExp2m1f16Test, PositiveRange) {
+ for (uint16_t v = POS_START; v <= POS_STOP; ++v) {
+ float16 x = FPBits(v).get_val();
+ EXPECT_MPFR_MATCH_ALL_ROUNDING(mpfr::Operation::Exp2m1, x,
+ LIBC_NAMESPACE::exp2m1f16(x), 0.5);
+ }
+}
+
+TEST_F(LlvmLibcExp2m1f16Test, NegativeRange) {
+ for (uint16_t v = NEG_START; v <= NEG_STOP; ++v) {
+ float16 x = FPBits(v).get_val();
+ EXPECT_MPFR_MATCH_ALL_ROUNDING(mpfr::Operation::Exp2m1, x,
+ LIBC_NAMESPACE::exp2m1f16(x), 0.5);
+ }
+}
diff --git a/libc/test/src/math/smoke/CMakeLists.txt b/libc/test/src/math/smoke/CMakeLists.txt
index 7271e933b9311d..349716d44cfb20 100644
--- a/libc/test/src/math/smoke/CMakeLists.txt
+++ b/libc/test/src/math/smoke/CMakeLists.txt
@@ -1106,6 +1106,17 @@ add_fp_unittest(
libc.src.math.exp2m1f
)
+add_fp_unittest(
+ exp2m1f16_test
+ SUITE
+ libc-math-smoke-tests
+ SRCS
+ exp2m1f16_test.cpp
+ DEPENDS
+ libc.src.errno.errno
+ libc.src.math.exp2m1f16
+)
+
add_fp_unittest(
exp10_test
SUITE
diff --git a/libc/test/src/math/smoke/exp2m1f16_test.cpp b/libc/test/src/math/smoke/exp2m1f16_test.cpp
new file mode 100644
index 00000000000000..7e6c2050abbbad
--- /dev/null
+++ b/libc/test/src/math/smoke/exp2m1f16_test.cpp
@@ -0,0 +1,94 @@
+//===-- Unittests for exp2m1f16 -------------------------------------------===//
+//
+// 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 "hdr/errno_macros.h"
+#include "hdr/fenv_macros.h"
+#include "src/errno/libc_errno.h"
+#include "src/math/exp2m1f16.h"
+#include "test/UnitTest/FPMatcher.h"
+#include "test/UnitTest/Test.h"
+
+using LlvmLibcExp2m1f16Test = LIBC_NAMESPACE::testing::FPTest<float16>;
+
+TEST_F(LlvmLibcExp2m1f16Test, SpecialNumbers) {
+ LIBC_NAMESPACE::libc_errno = 0;
+
+ EXPECT_FP_EQ_ALL_ROUNDING(aNaN, LIBC_NAMESPACE::exp2m1f16(aNaN));
+ EXPECT_MATH_ERRNO(0);
+
+ EXPECT_FP_EQ_WITH_EXCEPTION(aNaN, LIBC_NAMESPACE::exp2m1f16(sNaN),
+ FE_INVALID);
+ EXPECT_MATH_ERRNO(0);
+
+ EXPECT_FP_EQ_ALL_ROUNDING(inf, LIBC_NAMESPACE::exp2m1f16(inf));
+ EXPECT_MATH_ERRNO(0);
+
+ EXPECT_FP_EQ_ALL_ROUNDING(static_cast<float16>(-1.0),
+ LIBC_NAMESPACE::exp2m1f16(neg_inf));
+ EXPECT_MATH_ERRNO(0);
+
+ EXPECT_FP_EQ_ALL_ROUNDING(zero, LIBC_NAMESPACE::exp2m1f16(zero));
+ EXPECT_MATH_ERRNO(0);
+
+ EXPECT_FP_EQ_ALL_ROUNDING(neg_zero, LIBC_NAMESPACE::exp2m1f16(neg_zero));
+ EX...
[truncated]
|
|
Done. I haven't run the GPU build in a while so I haven't tested it yet, as it'll probably take some time to run the GPU build again. |
I'll test it in a bit. |
libc/config/gpu/entrypoints.txt
Outdated
| @@ -486,6 +486,7 @@ if(LIBC_TYPES_HAS_FLOAT16) | |||
| libc.src.math.ceilf16 | |||
| libc.src.math.copysignf16 | |||
| libc.src.math.exp10f16 | |||
| libc.src.math.exp10m1f16 | |||
Hm, works fine on NVPTX but fails on AMDGPU. Wonder why the sign is off. |
|
I poked at it but couldn't figure out why it's giving |
exp2 should be correctly lowered and it's just a bug if it's not. Is the bug in whatever this emitted or the llvm.exp2.f16 lowering? Can you post IR/ISA output? |
|
This doesn't use |
Sorry for the (long) delay, here's the IR output: https://gist.github.com/overmighty/20ec6b34f36d6b02c8009d85fa6e0794. |
Why are there any volatile accesses? There shouldn't be any z But still, this should just emit llvm.exp2 |
Some of them are from https://github.com/llvm/llvm-project/blob/main/libc/src/__support/FPUtil/rounding_mode.h#L67. I'm not sure about the others.
I don't think any higher math function implementation in LLVM libc emits the respective LLVM intrinsic currently. |
These volatiles should be removed. The tests will work just fine without the forced stack access |
The usage seems to be with the rounding mode check, so it's important that nothing gets constant propagated. I actually don't know how the rounding mode i handled when
These try to get the rounding mode through math operations so it's really important that it's not constant propagated. If you make it a constant then it will go through the LLVM optimizer which just assumes it's to-nearest, see https://godbolt.org/z/ehejrv1zK. We could probably avoid the volatile with some inline assembly or something. |
They won't be if you are using strictfp (as you must be if you depend on the rounding mode). If you do know the rounding mode, they will correctly fold.
You need to enable fenv access |
That doesn't work on the GPU targets yet https://godbolt.org/z/6n3svs1ne. But I guess NVPTX doesn't have |
Correct, but you shouldn't design around working around unimplemented features. Full GPU strictfp support isn't coming anytime soon. It is still wrong to rely on this check without it |
The other really broken thing is the -fexperimental-strict-floating-point flag doesn't do anything other than for the hardcoded targets in clang, making it useless as a bringup aid |
We can omit the gpu targets for now. I'm going to add |
overmighty
force-pushed
the
users/overmighty/libc-math-exp2m1f16
branch
from
9c4fca7 to
f3ea05a
Compare
|
Rebased and added the last 3 commits. |
Part of #95250.