[libc] Implement expf function that is correctly rounded for all rounding modes.

[lntue]

Summary

Implement expf function that is correctly rounded for all rounding modes.

[lntue]

Make sure hi and mid are non-negative.

[lntue]

Return MAX_NORMAL when x >= 89 in FE_DOWNWARD and FE_TOWARDZERO rounding modes.

[zimmermann6]

my exhaustive search confirms it is correctly rounded for all four rounding modes, great!

libc/src/math/generic/CMakeLists.txt

 
 add_entrypoint_object(
   expf
   SRCS
     expf.cpp
   HDRS
     ../expf.h
   DEPENDS
-    .exp_utils
-    .math_utils
+    libc.src.__support.FPUtil.fputil
     libc.include.math
+  COMPILE_OPTIONS
+    -O3
 )
 
 add_entrypoint_object(
   exp2f
   SRCS
     exp2f.cpp
   HDRS
     ../exp2f.h

libc/src/math/generic/expf.cpp

 //===-- Single-precision e^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/expf.h"
-#include "exp_utils.h"
-#include "math_utils.h"
-
+#include "src/__support/FPUtil/BasicOperations.h"
+#include "src/__support/FPUtil/FEnvImpl.h"
+#include "src/__support/FPUtil/FMA.h"
+#include "src/__support/FPUtil/FPBits.h"
+#include "src/__support/FPUtil/PolyEval.h"
 #include "src/__support/common.h"
-#include <math.h>
-
-#include <stdint.h>
 
-#define InvLn2N exp2f_data.invln2_scaled
-#define T exp2f_data.tab
-#define C exp2f_data.poly_scaled
-#define SHIFT exp2f_data.shift
+#include <errno.h>
 
 namespace __llvm_libc {
 
+// Lookup table for exp(m) with m = -104, ..., 89.
+//   -104 = floor(log(single precision's min denormal))
+//     89 = ceil(log(single precision's max normal))
+// Table is generated with Sollya as follow:
+// > display = hexadecimal;
+// > for i from -104 to 89 do { D(exp(i)); };
+static constexpr double EXP_M1[195] = {
+    0x1.f1e6b68529e33p-151, 0x1.525be4e4e601dp-149, 0x1.cbe0a45f75eb1p-148,
+    0x1.3884e838aea68p-146, 0x1.a8c1f14e2af5dp-145, 0x1.20a717e64a9bdp-143,
+    0x1.8851d84118908p-142, 0x1.0a9bdfb02d240p-140, 0x1.6a5bea046b42ep-139,
+    0x1.ec7f3b269efa8p-138, 0x1.4eafb87eab0f2p-136, 0x1.c6e2d05bbc000p-135,
+    0x1.35208867c2683p-133, 0x1.a425b317eeacdp-132, 0x1.1d8508fa8246ap-130,
+    0x1.840fbc08fdc8ap-129, 0x1.07b7112bc1ffep-127, 0x1.666d0dad2961dp-126,
+    0x1.e726c3f64d0fep-125, 0x1.4b0dc07cabf98p-123, 0x1.c1f2daf3b6a46p-122,
+    0x1.31c5957a47de2p-120, 0x1.9f96445648b9fp-119, 0x1.1a6baeadb4fd1p-117,
+    0x1.7fd974d372e45p-116, 0x1.04da4d1452919p-114, 0x1.62891f06b3450p-113,
+    0x1.e1dd273aa8a4ap-112, 0x1.4775e0840bfddp-110, 0x1.bd109d9d94bdap-109,
+    0x1.2e73f53fba844p-107, 0x1.9b138170d6bfep-106, 0x1.175af0cf60ec5p-104,
+    0x1.7baee1bffa80bp-103, 0x1.02057d1245cebp-101, 0x1.5eafffb34ba31p-100,
+    0x1.dca23bae16424p-99,  0x1.43e7fc88b8056p-97,  0x1.b83bf23a9a9ebp-96,
+    0x1.2b2b8dd05b318p-94,  0x1.969d47321e4ccp-93,  0x1.1452b7723aed2p-91,
+    0x1.778fe2497184cp-90,  0x1.fe7116182e9ccp-89,  0x1.5ae191a99585ap-87,
+    0x1.d775d87da854dp-86,  0x1.4063f8cc8bb98p-84,  0x1.b374b315f87c1p-83,
+    0x1.27ec458c65e3cp-81,  0x1.923372c67a074p-80,  0x1.1152eaeb73c08p-78,
+    0x1.737c5645114b5p-77,  0x1.f8e6c24b5592ep-76,  0x1.571db733a9d61p-74,
+    0x1.d257d547e083fp-73,  0x1.3ce9b9de78f85p-71,  0x1.aebabae3a41b5p-70,
+    0x1.24b6031b49bdap-68,  0x1.8dd5e1bb09d7ep-67,  0x1.0e5b73d1ff53dp-65,
+    0x1.6f741de1748ecp-64,  0x1.f36bd37f42f3ep-63,  0x1.536452ee2f75cp-61,
+    0x1.cd480a1b74820p-60,  0x1.39792499b1a24p-58,  0x1.aa0de4bf35b38p-57,
+    0x1.2188ad6ae3303p-55,  0x1.898471fca6055p-54,  0x1.0b6c3afdde064p-52,
+    0x1.6b7719a59f0e0p-51,  0x1.ee001eed62aa0p-50,  0x1.4fb547c775da8p-48,
+    0x1.c8464f7616468p-47,  0x1.36121e24d3bbap-45,  0x1.a56e0c2ac7f75p-44,
+    0x1.1e642baeb84a0p-42,  0x1.853f01d6d53bap-41,  0x1.0885298767e9ap-39,
+    0x1.67852a7007e42p-38,  0x1.e8a37a45fc32ep-37,  0x1.4c1078fe9228ap-35,
+    0x1.c3527e433fab1p-34,  0x1.32b48bf117da2p-32,  0x1.a0db0d0ddb3ecp-31,
+    0x1.1b48655f37267p-29,  0x1.81056ff2c5772p-28,  0x1.05a628c699fa1p-26,
+    0x1.639e3175a689dp-25,  0x1.e355bbaee85cbp-24,  0x1.4875ca227ec38p-22,
+    0x1.be6c6fdb01612p-21,  0x1.2f6053b981d98p-19,  0x1.9c54c3b43bc8bp-18,
+    0x1.18354238f6764p-16,  0x1.7cd79b5647c9bp-15,  0x1.02cf22526545ap-13,
+    0x1.5fc21041027adp-12,  0x1.de16b9c24a98fp-11,  0x1.44e51f113d4d6p-9,
+    0x1.b993fe00d5376p-8,   0x1.2c155b8213cf4p-6,   0x1.97db0ccceb0afp-5,
+    0x1.152aaa3bf81ccp-3,   0x1.78b56362cef38p-2,   0x1.0000000000000p+0,
+    0x1.5bf0a8b145769p+1,   0x1.d8e64b8d4ddaep+2,   0x1.415e5bf6fb106p+4,
+    0x1.b4c902e273a58p+5,   0x1.28d389970338fp+7,   0x1.936dc5690c08fp+8,
+    0x1.122885aaeddaap+10,  0x1.749ea7d470c6ep+11,  0x1.fa7157c470f82p+12,
+    0x1.5829dcf950560p+14,  0x1.d3c4488ee4f7fp+15,  0x1.3de1654d37c9ap+17,
+    0x1.b00b5916ac955p+18,  0x1.259ac48bf05d7p+20,  0x1.8f0ccafad2a87p+21,
+    0x1.0f2ebd0a80020p+23,  0x1.709348c0ea4f9p+24,  0x1.f4f22091940bdp+25,
+    0x1.546d8f9ed26e1p+27,  0x1.ceb088b68e804p+28,  0x1.3a6e1fd9eecfdp+30,
+    0x1.ab5adb9c43600p+31,  0x1.226af33b1fdc1p+33,  0x1.8ab7fb5475fb7p+34,
+    0x1.0c3d3920962c9p+36,  0x1.6c932696a6b5dp+37,  0x1.ef822f7f6731dp+38,
+    0x1.50bba3796379ap+40,  0x1.c9aae4631c056p+41,  0x1.370470aec28edp+43,
+    0x1.a6b765d8cdf6dp+44,  0x1.1f43fcc4b662cp+46,  0x1.866f34a725782p+47,
+    0x1.0953e2f3a1ef7p+49,  0x1.689e221bc8d5bp+50,  0x1.ea215a1d20d76p+51,
+    0x1.4d13fbb1a001ap+53,  0x1.c4b334617cc67p+54,  0x1.33a43d282a519p+56,
+    0x1.a220d397972ebp+57,  0x1.1c25c88df6862p+59,  0x1.8232558201159p+60,
+    0x1.0672a3c9eb871p+62,  0x1.64b41c6d37832p+63,  0x1.e4cf766fe49bep+64,
+    0x1.49767bc0483e3p+66,  0x1.bfc951eb8bb76p+67,  0x1.304d6aeca254bp+69,
+    0x1.9d97010884251p+70,  0x1.19103e4080b45p+72,  0x1.7e013cd114461p+73,
+    0x1.03996528e074cp+75,  0x1.60d4f6fdac731p+76,  0x1.df8c5af17ba3bp+77,
+    0x1.45e3076d61699p+79,  0x1.baed16a6e0da7p+80,  0x1.2cffdfebde1a1p+82,
+    0x1.9919cabefcb69p+83,  0x1.160345c9953e3p+85,  0x1.79dbc9dc53c66p+86,
+    0x1.00c810d464097p+88,  0x1.5d009394c5c27p+89,  0x1.da57de8f107a8p+90,
+    0x1.425982cf597cdp+92,  0x1.b61e5ca3a5e31p+93,  0x1.29bb825dfcf87p+95,
+    0x1.94a90db0d6fe2p+96,  0x1.12fec759586fdp+98,  0x1.75c1dc469e3afp+99,
+    0x1.fbfd219c43b04p+100, 0x1.5936d44e1a146p+102, 0x1.d531d8a7ee79cp+103,
+    0x1.3ed9d24a2d51bp+105, 0x1.b15cfe5b6e17bp+106, 0x1.268038c2c0e00p+108,
+    0x1.9044a73545d48p+109, 0x1.1002ab6218b38p+111, 0x1.71b3540cbf921p+112,
+    0x1.f6799ea9c414ap+113, 0x1.55779b984f3ebp+115, 0x1.d01a210c44aa4p+116,
+    0x1.3b63da8e91210p+118, 0x1.aca8d6b0116b8p+119, 0x1.234de9e0c74e9p+121,
+    0x1.8bec7503ca477p+122, 0x1.0d0eda9796b90p+124, 0x1.6db0118477245p+125,
+    0x1.f1056dc7bf22dp+126, 0x1.51c2cc3433801p+128, 0x1.cb108ffbec164p+129,
+};
+
+// Lookup table for exp(m * 2^(-7)) with m = 0, ..., 127.
+// Table is generated with Sollya as follow:
+// > display = hexadecimal;
+// > for i from 0 to 127 do { D(exp(i / 128)); };
+static constexpr double EXP_M2[128] = {
+    0x1.0000000000000p0, 0x1.0202015600446p0, 0x1.04080ab55de39p0,
+    0x1.06122436410ddp0, 0x1.08205601127edp0, 0x1.0a32a84e9c1f6p0,
+    0x1.0c49236829e8cp0, 0x1.0e63cfa7ab09dp0, 0x1.1082b577d34edp0,
+    0x1.12a5dd543ccc5p0, 0x1.14cd4fc989cd6p0, 0x1.16f9157587069p0,
+    0x1.192937074e0cdp0, 0x1.1b5dbd3f68122p0, 0x1.1d96b0eff0e79p0,
+    0x1.1fd41afcba45ep0, 0x1.2216045b6f5cdp0, 0x1.245c7613b8a9bp0,
+    0x1.26a7793f60164p0, 0x1.28f7170a755fdp0, 0x1.2b4b58b372c79p0,
+    0x1.2da4478b620c7p0, 0x1.3001ecf601af7p0, 0x1.32645269ea829p0,
+    0x1.34cb8170b5835p0, 0x1.373783a722012p0, 0x1.39a862bd3c106p0,
+    0x1.3c1e2876834aap0, 0x1.3e98deaa11dccp0, 0x1.41188f42c3e32p0,
+    0x1.439d443f5f159p0, 0x1.462707b2bac21p0, 0x1.48b5e3c3e8186p0,
+    0x1.4b49e2ae5ac67p0, 0x1.4de30ec211e60p0, 0x1.50817263c13cdp0,
+    0x1.5325180cfacf7p0, 0x1.55ce0a4c58c7cp0, 0x1.587c53c5a7af0p0,
+    0x1.5b2fff3210fd9p0, 0x1.5de9176045ff5p0, 0x1.60a7a734ab0e8p0,
+    0x1.636bb9a983258p0, 0x1.663559cf1bc7cp0, 0x1.690492cbf9433p0,
+    0x1.6bd96fdd034a2p0, 0x1.6eb3fc55b1e76p0, 0x1.719443a03acb9p0,
+    0x1.747a513dbef6ap0, 0x1.776630c678bc1p0, 0x1.7a57ede9ea23ep0,
+    0x1.7d4f946f0ba8dp0, 0x1.804d30347b546p0, 0x1.8350cd30ac390p0,
+    0x1.865a7772164c5p0, 0x1.896a3b1f66a0ep0, 0x1.8c802477b0010p0,
+    0x1.8f9c3fd29beafp0, 0x1.92be99a09bf00p0, 0x1.95e73e6b1b75ep0,
+    0x1.99163ad4b1dccp0, 0x1.9c4b9b995509bp0, 0x1.9f876d8e8c566p0,
+    0x1.a2c9bda3a3e78p0, 0x1.a61298e1e069cp0, 0x1.a9620c6cb3374p0,
+    0x1.acb82581eee54p0, 0x1.b014f179fc3b8p0, 0x1.b3787dc80f95fp0,
+    0x1.b6e2d7fa5eb18p0, 0x1.ba540dba56e56p0, 0x1.bdcc2cccd3c85p0,
+    0x1.c14b431256446p0, 0x1.c4d15e873c193p0, 0x1.c85e8d43f7cd0p0,
+    0x1.cbf2dd7d490f2p0, 0x1.cf8e5d84758a9p0, 0x1.d3311bc7822b4p0,
+    0x1.d6db26d16cd67p0, 0x1.da8c8d4a66969p0, 0x1.de455df80e3c0p0,
+    0x1.e205a7bdab73ep0, 0x1.e5cd799c6a54ep0, 0x1.e99ce2b397649p0,
+    0x1.ed73f240dc142p0, 0x1.f152b7a07bb76p0, 0x1.f539424d90f5ep0,
+    0x1.f927a1e24bb76p0, 0x1.fd1de6182f8c9p0, 0x1.008e0f64294abp1,
+    0x1.02912df5ce72ap1, 0x1.049856cd84339p1, 0x1.06a39207f0a09p1,
+    0x1.08b2e7d2035cfp1, 0x1.0ac6606916501p1, 0x1.0cde041b0e9aep1,
+    0x1.0ef9db467dcf8p1, 0x1.1119ee5ac36b6p1, 0x1.133e45d82e952p1,
+    0x1.1566ea50201d7p1, 0x1.1793e4652cc50p1, 0x1.19c53ccb3fc6bp1,
+    0x1.1bfafc47bda73p1, 0x1.1e352bb1a74adp1, 0x1.2073d3f1bd518p1,
+    0x1.22b6fe02a3b9cp1, 0x1.24feb2f105cb8p1, 0x1.274afbdbba4a6p1,
+    0x1.299be1f3e7f1cp1, 0x1.2bf16e7d2a38cp1, 0x1.2e4baacdb6614p1,
+    0x1.30aaa04e80d05p1, 0x1.330e587b62b28p1, 0x1.3576dce33feadp1,
+    0x1.37e437282d4eep1, 0x1.3a5670ff972edp1, 0x1.3ccd9432682b4p1,
+    0x1.3f49aa9d30590p1, 0x1.41cabe304cb34p1, 0x1.4450d8f00edd4p1,
+    0x1.46dc04f4e5338p1, 0x1.496c4c6b832dap1, 0x1.4c01b9950a111p1,
+    0x1.4e9c56c731f5dp1, 0x1.513c2e6c731d7p1, 0x1.53e14b042f9cap1,
+    0x1.568bb722dd593p1, 0x1.593b7d72305bbp1,
+};
+
+INLINE_FMA
 LLVM_LIBC_FUNCTION(float, expf, (float x)) {
-  uint32_t abstop;
-  uint64_t ki, t;
-  // double_t for better performance on targets with FLT_EVAL_METHOD == 2.
-  double_t kd, xd, z, r, r2, y, s;
-
-  xd = static_cast<double_t>(x);
-  abstop = top12_bits(x) & 0x7ff;
-  if (unlikely(abstop >= top12_bits(88.0f))) {
-    // |x| >= 88 or x is nan.
-    if (as_uint32_bits(x) == as_uint32_bits(-INFINITY))
+  using FPBits = typename fputil::FPBits<float>;
+  FPBits xbits(x);
+
+  // When x < log(2^-150) or nan
+  if (unlikely(xbits.uintval() >= 0xc2cf'f1b5U)) {
+    // exp(-Inf) = 0
+    if (xbits.is_inf())
       return 0.0f;
-    if (abstop >= top12_bits(INFINITY))
-      return x + x;
-    if (x > as_float(0x42b17217)) // x > log(0x1p128) ~= 88.72
-      return overflow<float>(0);
-    if (x < as_float(0xc2cff1b4)) // x < log(0x1p-150) ~= -103.97
-      return underflow<float>(0);
-    if (x < as_float(0xc2ce8ecf)) // x < log(0x1p-149) ~= -103.28
-      return may_underflow<float>(0);
-  }
-
-  // x*N/Ln2 = k + r with r in [-1/2, 1/2] and int k.
-  z = InvLn2N * xd;
-
-  // Round and convert z to int, the result is in [-150*N, 128*N] and
-  // ideally nearest int is used, otherwise the magnitude of r can be
-  // bigger which gives larger approximation error.
-  kd = static_cast<double>(z + SHIFT);
-  ki = as_uint64_bits(kd);
-  kd -= SHIFT;
-  r = z - kd;
-
-  // exp(x) = 2^(k/N) * 2^(r/N) ~= s *(C0*r^3 + C1*r^2 + C2*r + 1)
-  t = T[ki % N];
-  t += ki << (52 - EXP2F_TABLE_BITS);
-  s = as_double(t);
-  z = C[0] * r + C[1];
-  r2 = r * r;
-  y = C[2] * r + 1;
-  y = z * r2 + y;
-  y = y * s;
-  return static_cast<float>(y);
+    // exp(nan) = nan
+    if (xbits.is_nan())
+      return x;
+    if (fputil::get_round() == FE_UPWARD)
+      return static_cast<float>(FPBits(FPBits::MIN_SUBNORMAL));
+    if (x != 0.0f)
+      errno = ERANGE;
+    return 0.0f;
+  }
+  // x >= 89 or nan
+  if (unlikely(!xbits.get_sign() && (xbits.uintval() >= 0x42b2'0000))) {
+    if (xbits.uintval() < 0x7f80'0000U) {
+      int rounding = fputil::get_round();
+      if (rounding == FE_DOWNWARD || rounding == FE_TOWARDZERO)
+        return static_cast<float>(FPBits(FPBits::MAX_NORMAL));
+
+      errno = ERANGE;
+    }
+    return x + static_cast<float>(FPBits::inf());
+  }
+  // |x| < 2^-25
+  if (unlikely(xbits.get_unbiased_exponent() <= 101)) {
+    return 1.0f + x;
+  }
+
+  // For -104 < x < 89, to compute exp(x), we perform the following range
+  // reduction: find hi, mid, lo such that:
+  //   x = hi + mid + lo, in which
+  //     hi is an integer,
+  //     mid * 2^7 is an integer
+  //     -2^(-8) <= lo < 2^-8.
+  // In particular,
+  //   hi + mid = round(x * 2^7) * 2^(-7).
+  // Then,
+  //   exp(x) = exp(hi + mid + lo) = exp(hi) * exp(mid) * exp(lo).
+  // We store exp(hi) and exp(mid) in the lookup tables EXP_M1 and EXP_M2
+  // respectively.  exp(lo) is computed using a degree-7 minimax polynomial
+  // generated by Sollya.
+
+  // x_hi = hi + mid.
+  int x_hi = static_cast<int>(x * 0x1.0p7f);
+  // Subtract (hi + mid) from x to get lo.
+  x -= static_cast<float>(x_hi) * 0x1.0p-7f;
+  double xd = static_cast<double>(x);
+  // Make sure that -2^(-8) <= lo < 2^-8.
+  if (x >= 0x1.0p-8f) {
+    ++x_hi;
+    xd -= 0x1.0p-7;
+  }
+  if (x < -0x1.0p-8f) {
+    --x_hi;
+    xd += 0x1.0p-7;
+  }
+  x_hi += 104 << 7;
+  // hi = x_hi >> 7
+  double exp_hi = EXP_M1[x_hi >> 7];
+  // lo = x_hi & 0x0000'007fU;
+  double exp_mid = EXP_M2[x_hi & 0x7f];
+  // Degree-7 minimax polynomial generated by Sollya with the following
+  // commands:
+  //   > display = hexadecimal;
+  //   > Q = fpminimax(expm1(x)/x, 6, [|D...|], [-2^-8, 2^-8]);
+  //   > Q;
+  double exp_lo = fputil::polyeval(
+      xd, 0x1p0, 0x1p0, 0x1p-1, 0x1.5555555555555p-3, 0x1.55555555553ap-5,
+      0x1.1111111204dfcp-7, 0x1.6c16cb2da593ap-10, 0x1.9ff1648996d2ep-13);
+  return static_cast<float>(exp_hi * exp_mid * exp_lo);
 }
 
 } // namespace __llvm_libc

libc/test/src/math/exhaustive/CMakeLists.txt

SRCS
     cosf_test.cpp
   DEPENDS
     libc.include.math
     libc.src.math.cosf
     libc.src.__support.FPUtil.fputil
 )
 
 add_fp_unittest(
+  expf_test
+  NO_RUN_POSTBUILD
+  NEED_MPFR
+  SUITE
+    libc_math_exhaustive_tests
+  SRCS
+    expf_test.cpp
+  DEPENDS
+    .exhaustive_test
+    libc.include.math
+    libc.src.math.expf
+    libc.src.__support.FPUtil.fputil
+  LINK_OPTIONS
+    -lpthread
+)
+
+add_fp_unittest(
   expm1f_test
   NEED_MPFR
   SUITE
     libc_math_exhaustive_tests
   SRCS
   expm1f_test.cpp
   DEPENDS
     libc.include.math

libc/test/src/math/exhaustive/expf_test.cpp

+//===-- Exhaustive test for expf ------------------------------------------===//
+//
+// 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 "exhaustive_test.h"
+#include "src/__support/FPUtil/FPBits.h"
+#include "src/math/expf.h"
+#include "utils/MPFRWrapper/MPFRUtils.h"
+#include "utils/UnitTest/FPMatcher.h"
+
+using FPBits = __llvm_libc::fputil::FPBits<float>;
+
+namespace mpfr = __llvm_libc::testing::mpfr;
+
+struct LlvmLibcExpfExhaustiveTest : public LlvmLibcExhaustiveTest<uint32_t> {
+  void check(uint32_t start, uint32_t stop,
+             mpfr::RoundingMode rounding) override {
+    mpfr::ForceRoundingMode r(rounding);
+    uint32_t bits = start;
+    do {
+      FPBits xbits(bits);
+      float x = float(xbits);
+      EXPECT_MPFR_MATCH(mpfr::Operation::Exp, x, __llvm_libc::expf(x), 0.5,
+                        rounding);
+    } while (bits++ < stop);
+  }
+};
+
+static constexpr int NUM_THREADS = 16;
+
+// Range: [0, 89];
+static constexpr uint32_t POS_START = 0x0000'0000U;
+static constexpr uint32_t POS_STOP = 0x42b2'0000U;
+
+TEST_F(LlvmLibcExpfExhaustiveTest, PostiveRangeRoundNearestTieToEven) {
+  test_full_range(POS_START, POS_STOP, NUM_THREADS,
+                  mpfr::RoundingMode::Nearest);
+}
+
+TEST_F(LlvmLibcExpfExhaustiveTest, PostiveRangeRoundUp) {
+  test_full_range(POS_START, POS_STOP, NUM_THREADS, mpfr::RoundingMode::Upward);
+}
+
+TEST_F(LlvmLibcExpfExhaustiveTest, PostiveRangeRoundDown) {
+  test_full_range(POS_START, POS_STOP, NUM_THREADS,
+                  mpfr::RoundingMode::Downward);
+}
+
+TEST_F(LlvmLibcExpfExhaustiveTest, PostiveRangeRoundTowardZero) {
+  test_full_range(POS_START, POS_STOP, NUM_THREADS,
+                  mpfr::RoundingMode::TowardZero);
+}
+
+// Range: [-104, 0];
+static constexpr uint32_t NEG_START = 0x8000'0000U;
+static constexpr uint32_t NEG_STOP = 0xc2d0'0000U;
+
+TEST_F(LlvmLibcExpfExhaustiveTest, NegativeRangeRoundNearestTieToEven) {
+  test_full_range(NEG_START, NEG_STOP, NUM_THREADS,
+                  mpfr::RoundingMode::Nearest);
+}
+
+TEST_F(LlvmLibcExpfExhaustiveTest, NegativeRangeRoundUp) {
+  test_full_range(NEG_START, NEG_STOP, NUM_THREADS, mpfr::RoundingMode::Upward);
+}
+
+TEST_F(LlvmLibcExpfExhaustiveTest, NegativeRangeRoundDown) {
+  test_full_range(NEG_START, NEG_STOP, NUM_THREADS,
+                  mpfr::RoundingMode::Downward);
+}
+
+TEST_F(LlvmLibcExpfExhaustiveTest, NegativeRangeRoundTowardZero) {
+  test_full_range(NEG_START, NEG_STOP, NUM_THREADS,
+                  mpfr::RoundingMode::TowardZero);
+}

libc/test/src/math/expf_test.cpp

 }
 
 TEST(LlvmLibcExpfTest, Underflow) {
   errno = 0;
   EXPECT_FP_EQ(0.0f, __llvm_libc::expf(float(FPBits(0xff7fffffU))));
   EXPECT_MATH_ERRNO(ERANGE);
 
   float x = float(FPBits(0xc2cffff8U));
-  EXPECT_MPFR_MATCH(mpfr::Operation::Exp, x, __llvm_libc::expf(x), 1.0);
+  EXPECT_MPFR_MATCH_ALL_ROUNDING(mpfr::Operation::Exp, x, __llvm_libc::expf(x),
+                                 0.5);
   EXPECT_MATH_ERRNO(ERANGE);
 
   x = float(FPBits(0xc2d00008U));
-  EXPECT_MPFR_MATCH(mpfr::Operation::Exp, x, __llvm_libc::expf(x), 1.0);
+  EXPECT_MPFR_MATCH_ALL_ROUNDING(mpfr::Operation::Exp, x, __llvm_libc::expf(x),
+                                 0.5);
   EXPECT_MATH_ERRNO(ERANGE);
 }
 
 // Test with inputs which are the borders of underflow/overflow but still
 // produce valid results without setting errno.
 TEST(LlvmLibcExpfTest, Borderline) {
   float x;
 
   errno = 0;
   x = float(FPBits(0x42affff8U));
-  ASSERT_MPFR_MATCH(mpfr::Operation::Exp, x, __llvm_libc::expf(x), 1.0);
+  ASSERT_MPFR_MATCH_ALL_ROUNDING(mpfr::Operation::Exp, x, __llvm_libc::expf(x),
+                                 0.5);
   EXPECT_MATH_ERRNO(0);
 
   x = float(FPBits(0x42b00008U));
-  ASSERT_MPFR_MATCH(mpfr::Operation::Exp, x, __llvm_libc::expf(x), 1.0);
+  ASSERT_MPFR_MATCH_ALL_ROUNDING(mpfr::Operation::Exp, x, __llvm_libc::expf(x),
+                                 0.5);
   EXPECT_MATH_ERRNO(0);
 
   x = float(FPBits(0xc2affff8U));
-  ASSERT_MPFR_MATCH(mpfr::Operation::Exp, x, __llvm_libc::expf(x), 1.0);
+  ASSERT_MPFR_MATCH_ALL_ROUNDING(mpfr::Operation::Exp, x, __llvm_libc::expf(x),
+                                 0.5);
   EXPECT_MATH_ERRNO(0);
 
   x = float(FPBits(0xc2b00008U));
-  ASSERT_MPFR_MATCH(mpfr::Operation::Exp, x, __llvm_libc::expf(x), 1.0);
+  ASSERT_MPFR_MATCH_ALL_ROUNDING(mpfr::Operation::Exp, x, __llvm_libc::expf(x),
+                                 0.5);
   EXPECT_MATH_ERRNO(0);
 }
 
 TEST(LlvmLibcExpfTest, InFloatRange) {
   constexpr uint32_t COUNT = 1000000;
   constexpr uint32_t STEP = UINT32_MAX / COUNT;
   for (uint32_t i = 0, v = 0; i <= COUNT; ++i, v += STEP) {
     float x = float(FPBits(v));
     if (isnan(x) || isinf(x))
       continue;
     errno = 0;
     float result = __llvm_libc::expf(x);
 
     // If the computation resulted in an error or did not produce valid result
     // in the single-precision floating point range, then ignore comparing with
     // MPFR result as MPFR can still produce valid results because of its
     // wider precision.
     if (isnan(result) || isinf(result) || errno != 0)
       continue;
-    ASSERT_MPFR_MATCH(mpfr::Operation::Exp, x, __llvm_libc::expf(x), 1.0);
+    EXPECT_MPFR_MATCH_ALL_ROUNDING(mpfr::Operation::Exp, x,
+                                   __llvm_libc::expf(x), 0.5);
   }
 }