diff --git a/libc/src/__support/FPUtil/Hypot.h b/libc/src/__support/FPUtil/Hypot.h
--- a/libc/src/__support/FPUtil/Hypot.h
+++ b/libc/src/__support/FPUtil/Hypot.h
@@ -10,6 +10,7 @@
 #define LLVM_LIBC_SRC_SUPPORT_FPUTIL_HYPOT_H
 
 #include "BasicOperations.h"
+#include "FEnvImpl.h"
 #include "FPBits.h"
 #include "src/__support/CPP/TypeTraits.h"
 
@@ -143,11 +144,22 @@
   if ((x_bits.get_unbiased_exponent() >=
        y_bits.get_unbiased_exponent() + MantissaWidth<T>::VALUE + 2) ||
       (y == 0)) {
+    // Check if the rounding mode is FE_UPWARD, will need -frounding-math so
+    // that the compiler does not optimize it away.
+    if ((y != 0) && (0x1p0f + 0x1p-24f != 0x1p0f)) {
+      UIntType out_bits = FPBits_t(abs(x)).uintval();
+      return T(FPBits_t(++out_bits));
+    }
     return abs(x);
   } else if ((y_bits.get_unbiased_exponent() >=
               x_bits.get_unbiased_exponent() + MantissaWidth<T>::VALUE + 2) ||
              (x == 0)) {
-    y_bits.set_sign(0);
+    // Check if the rounding mode is FE_UPWARD, will need -frounding-math so
+    // that the compiler does not optimize it away.
+    if ((x != 0) && (0x1p0f + 0x1p-24f != 0x1p0f)) {
+      UIntType out_bits = FPBits_t(abs(y)).uintval();
+      return T(FPBits_t(++out_bits));
+    }
     return abs(y);
   }
 
@@ -250,8 +262,16 @@
   y_new >>= 1;
 
   // Round to the nearest, tie to even.
-  if (round_bit && (lsb || sticky_bits || (r != 0))) {
-    ++y_new;
+  switch (get_round()) {
+  case FE_TONEAREST:
+    // Round to nearest, ties to even
+    if (round_bit && (lsb || sticky_bits || (r != 0)))
+      ++y_new;
+    break;
+  case FE_UPWARD:
+    if (round_bit || sticky_bits || (r != 0))
+      ++y_new;
+    break;
   }
 
   if (y_new >= (ONE >> 1)) {
diff --git a/libc/src/math/generic/CMakeLists.txt b/libc/src/math/generic/CMakeLists.txt
--- a/libc/src/math/generic/CMakeLists.txt
+++ b/libc/src/math/generic/CMakeLists.txt
@@ -954,7 +954,9 @@
   DEPENDS
     libc.src.__support.FPUtil.fputil
   COMPILE_OPTIONS
-    -O2
+    -O3
+    -frounding-math
+    -Wno-c++17-extensions
 )
 
 add_entrypoint_object(
@@ -1002,7 +1004,9 @@
   DEPENDS
     libc.src.__support.FPUtil.fputil
   COMPILE_OPTIONS
-    -O2
+    -O3
+    -frounding-math
+    -Wno-c++17-extensions
 )
 
 add_entrypoint_object(
diff --git a/libc/test/src/math/CMakeLists.txt b/libc/test/src/math/CMakeLists.txt
--- a/libc/test/src/math/CMakeLists.txt
+++ b/libc/test/src/math/CMakeLists.txt
@@ -1060,6 +1060,8 @@
     libc.include.math
     libc.src.math.hypotf
     libc.src.__support.FPUtil.fputil
+  COMPILE_OPTIONS
+    -Wno-c++17-extensions
 )
 
 add_fp_unittest(
@@ -1073,6 +1075,8 @@
     libc.include.math
     libc.src.math.hypot
     libc.src.__support.FPUtil.fputil
+  COMPILE_OPTIONS
+    -Wno-c++17-extensions
 )
 
 add_fp_unittest(
diff --git a/libc/test/src/math/HypotTest.h b/libc/test/src/math/HypotTest.h
--- a/libc/test/src/math/HypotTest.h
+++ b/libc/test/src/math/HypotTest.h
@@ -10,7 +10,6 @@
 #define LLVM_LIBC_TEST_SRC_MATH_HYPOTTEST_H
 
 #include "src/__support/FPUtil/FPBits.h"
-#include "src/__support/FPUtil/Hypot.h"
 #include "utils/MPFRWrapper/MPFRUtils.h"
 #include "utils/UnitTest/FPMatcher.h"
 #include "utils/UnitTest/Test.h"
@@ -62,9 +61,9 @@
             y = -y;
           }
 
-          T result = func(x, y);
           mpfr::BinaryInput<T> input{x, y};
-          ASSERT_MPFR_MATCH(mpfr::Operation::Hypot, input, result, 0.5);
+          ASSERT_MPFR_MATCH_ALL_ROUNDING(mpfr::Operation::Hypot, input,
+                                         func(x, y), 0.5);
         }
       }
     }
@@ -85,12 +84,38 @@
           y = -y;
         }
 
-        T result = func(x, y);
         mpfr::BinaryInput<T> input{x, y};
-        ASSERT_MPFR_MATCH(mpfr::Operation::Hypot, input, result, 0.5);
+        ASSERT_MPFR_MATCH_ALL_ROUNDING(mpfr::Operation::Hypot, input,
+                                       func(x, y), 0.5);
       }
     }
   }
+
+  void test_tricky_inputs(Func func) {
+    constexpr int N = 15;
+    constexpr mpfr::BinaryInput<T> INPUTS[N] = {
+        {0x1.ffffecp-1f, 0x1.000002p+27},
+        {0x1.900004p+34, 0x1.400002p+23}, /* 45 identical bits */
+        {0x1.05555p+34, 0x1.bffffep+23},  /* 44 identical bits */
+        {0x1.e5fffap+34, 0x1.affffep+23}, /* 45 identical bits */
+        {0x1.260002p+34, 0x1.500002p+23}, /* 45 identical bits */
+        {0x1.fffffap+34, 0x1.fffffep+23}, /* 45 identical bits */
+        {0x1.8ffffap+34, 0x1.3ffffep+23}, /* 45 identical bits */
+        {0x1.87fffcp+35, 0x1.bffffep+23}, /* 47 identical bits */
+        {0x1.b8e50ap-52, -0x1.db1e78p-64},
+        {0x1.03b54cp-33, 0x1.6ca6bep-45},
+        {0x1.e2eff6p+97, -0x1.044cb2p+108},
+        {-0x1.6b05c4p-127, 0x1.6b3146p-126},
+        {-0x1.6b05c4p-127, 0x1.6b3146p-126},
+        {0x1.26b188p-127, -0x1.a4f2fp-128},
+        {0x1.e2eff6p+97, -0x1.044cb2p+108},
+    };
+
+    for (int i = 0; i < N; ++i) {
+      ASSERT_MPFR_MATCH_ALL_ROUNDING(mpfr::Operation::Hypot, INPUTS[i],
+                                     func(INPUTS[i].x, INPUTS[i].y), 0.5);
+    }
+  }
 };
 
 #endif // LLVM_LIBC_TEST_SRC_MATH_HYPOTTEST_H
diff --git a/libc/test/src/math/differential_testing/CMakeLists.txt b/libc/test/src/math/differential_testing/CMakeLists.txt
--- a/libc/test/src/math/differential_testing/CMakeLists.txt
+++ b/libc/test/src/math/differential_testing/CMakeLists.txt
@@ -406,6 +406,7 @@
     libc.src.math.hypotf
   COMPILE_OPTIONS
     -fno-builtin
+    -Wno-c++17-extensions
 )
 
 add_diff_binary(
@@ -417,4 +418,5 @@
     libc.src.math.hypot
   COMPILE_OPTIONS
     -fno-builtin
+    -Wno-c++17-extensions
 )
diff --git a/libc/test/src/math/hypotf_test.cpp b/libc/test/src/math/hypotf_test.cpp
--- a/libc/test/src/math/hypotf_test.cpp
+++ b/libc/test/src/math/hypotf_test.cpp
@@ -23,3 +23,7 @@
 TEST_F(LlvmLibcHypotfTest, NormalRange) {
   test_normal_range(&__llvm_libc::hypotf);
 }
+
+TEST_F(LlvmLibcHypotfTest, TrickyInputs) {
+  test_tricky_inputs(&__llvm_libc::hypotf);
+}