diff --git a/libc/src/__support/FPUtil/generic/sqrt.h b/libc/src/__support/FPUtil/generic/sqrt.h
diff --git a/libc/src/__support/FPUtil/sqrt.h b/libc/src/__support/FPUtil/sqrt.h
diff --git a/.../FPUtil/generic/sqrt_80_bit_long_double.h → .../__support/FPUtil/x86_64/SqrtLongDouble.h b/.../FPUtil/generic/sqrt_80_bit_long_double.h → .../__support/FPUtil/x86_64/SqrtLongDouble.h
@@ -6,17 +6,26 @@
 //
 //===----------------------------------------------------------------------===//
 
-#ifndef LLVM_LIBC_SRC_SUPPORT_FPUTIL_GENERIC_SQRT_80_BIT_LONG_DOUBLE_H
-#define LLVM_LIBC_SRC_SUPPORT_FPUTIL_GENERIC_SQRT_80_BIT_LONG_DOUBLE_H
+#ifndef LLVM_LIBC_SRC_SUPPORT_FPUTIL_X86_64_SQRT_LONG_DOUBLE_H
+#define LLVM_LIBC_SRC_SUPPORT_FPUTIL_X86_64_SQRT_LONG_DOUBLE_H
 
-#include "src/__support/FPUtil/FEnvImpl.h"
+#include "src/__support/architectures.h"
+
+#if !defined(LLVM_LIBC_ARCH_X86)
+#error "Invalid include"
+#endif
+
+#include "src/__support/CPP/TypeTraits.h"
 #include "src/__support/FPUtil/FPBits.h"
+#include "src/__support/FPUtil/Sqrt.h"
 
 namespace __llvm_libc {
 namespace fputil {
-namespace x86 {
 
-inline void normalize(int &exponent, __uint128_t &mantissa) {
+namespace internal {
+
+template <>
+inline void normalize<long double>(int &exponent, __uint128_t &mantissa) {
   // Use binary search to shift the leading 1 bit similar to float.
   // With MantissaWidth<long double> = 63, it will take
   // ceil(log2(63)) = 6 steps checking the mantissa bits.
@@ -34,9 +43,11 @@ inline void normalize(int &exponent, __uint128_t &mantissa) {
   }
 }
 
-// Correctly rounded SQRT for all rounding modes.
+} // namespace internal
+
+// Correctly rounded SQRT with round to nearest, ties to even.
 // Shift-and-add algorithm.
-static inline long double sqrt(long double x) {
+template <> inline long double sqrt<long double, 0>(long double x) {
   using UIntType = typename FPBits<long double>::UIntType;
   constexpr UIntType ONE = UIntType(1)
                            << int(MantissaWidth<long double>::VALUE);
@@ -67,7 +78,7 @@ static inline long double sqrt(long double x) {
     if (bits.get_implicit_bit()) {
       x_mant |= ONE;
     } else if (bits.get_unbiased_exponent() == 0) {
-      normalize(x_exp, x_mant);
+      internal::normalize<long double>(x_exp, x_mant);
     }
 
     // Step 1b: Make sure the exponent is even.
@@ -115,16 +126,9 @@ static inline long double sqrt(long double x) {
     y |= (static_cast<UIntType>(x_exp)
           << (MantissaWidth<long double>::VALUE + 1));
 
-    switch (get_round()) {
-    case FE_TONEAREST:
-      // Round to nearest, ties to even
-      if (rb && (lsb || (r != 0)))
-        ++y;
-      break;
-    case FE_UPWARD:
-      if (rb || (r != 0))
-        ++y;
-      break;
+    // Round to nearest, ties to even
+    if (rb && (lsb || (r != 0))) {
+      ++y;
     }
 
     // Extract output
@@ -137,8 +141,7 @@ static inline long double sqrt(long double x) {
   }
 }
 
-} // namespace x86
 } // namespace fputil
 } // namespace __llvm_libc
 
-#endif // LLVM_LIBC_SRC_SUPPORT_FPUTIL_GENERIC_SQRT_80_BIT_LONG_DOUBLE_H
+#endif // LLVM_LIBC_SRC_SUPPORT_FPUTIL_X86_64_SQRT_LONG_DOUBLE_H
diff --git a/libc/src/__support/FPUtil/x86_64/sqrt.h b/libc/src/__support/FPUtil/x86_64/sqrt.h
diff --git a/libc/src/math/aarch64/CMakeLists.txt b/libc/src/math/aarch64/CMakeLists.txt
@@ -77,3 +77,23 @@ add_entrypoint_object(
   COMPILE_OPTIONS
     -O2
 )
+
+add_entrypoint_object(
+  sqrt
+  SRCS
+    sqrt.cpp
+  HDRS
+    ../sqrt.h
+  COMPILE_OPTIONS
+    -O2
+)
+
+add_entrypoint_object(
+  sqrtf
+  SRCS
+    sqrtf.cpp
+  HDRS
+    ../sqrtf.h
+  COMPILE_OPTIONS
+    -O2
+)
diff --git a/libc/src/math/generic/CMakeLists.txt b/libc/src/math/generic/CMakeLists.txt
@@ -859,10 +859,8 @@ add_entrypoint_object(
     ../sqrt.h
   DEPENDS
     libc.src.__support.FPUtil.fputil
-    libc.src.__support.FPUtil.sqrt
   COMPILE_OPTIONS
-    -O3
-    -Wno-c++17-extensions
+    -O2
 )
 
 add_entrypoint_object(
@@ -873,10 +871,8 @@ add_entrypoint_object(
     ../sqrtf.h
   DEPENDS
     libc.src.__support.FPUtil.fputil
-    libc.src.__support.FPUtil.sqrt
   COMPILE_OPTIONS
-    -O3
-    -Wno-c++17-extensions
+    -O2
 )
 
 add_entrypoint_object(
@@ -887,10 +883,8 @@ add_entrypoint_object(
     ../sqrtl.h
   DEPENDS
     libc.src.__support.FPUtil.fputil
-    libc.src.__support.FPUtil.sqrt
   COMPILE_OPTIONS
-    -O3
-    -Wno-c++17-extensions
+    -O2
 )
 
 add_entrypoint_object(

diff --git a/libc/src/math/generic/sqrt.cpp b/libc/src/math/generic/sqrt.cpp
@@ -7,7 +7,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "src/math/sqrt.h"
-#include "src/__support/FPUtil/sqrt.h"
+#include "src/__support/FPUtil/Sqrt.h"
 #include "src/__support/common.h"
 
 namespace __llvm_libc {

diff --git a/libc/src/math/generic/sqrtf.cpp b/libc/src/math/generic/sqrtf.cpp
@@ -7,7 +7,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "src/math/sqrtf.h"
-#include "src/__support/FPUtil/sqrt.h"
+#include "src/__support/FPUtil/Sqrt.h"
 #include "src/__support/common.h"
 
 namespace __llvm_libc {

diff --git a/libc/src/math/generic/sqrtl.cpp b/libc/src/math/generic/sqrtl.cpp
@@ -7,7 +7,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "src/math/sqrtl.h"
-#include "src/__support/FPUtil/sqrt.h"
+#include "src/__support/FPUtil/Sqrt.h"
 #include "src/__support/common.h"
 
 namespace __llvm_libc {

diff --git a/libc/src/math/x86_64/CMakeLists.txt b/libc/src/math/x86_64/CMakeLists.txt
@@ -27,3 +27,33 @@ add_entrypoint_object(
   COMPILE_OPTIONS
     -O2
 )
+
+add_entrypoint_object(
+  sqrt
+  SRCS
+    sqrt.cpp
+  HDRS
+    ../sqrt.h
+  COMPILE_OPTIONS
+    -O2
+)
+
+add_entrypoint_object(
+  sqrtf
+  SRCS
+    sqrtf.cpp
+  HDRS
+    ../sqrtf.h
+  COMPILE_OPTIONS
+    -O2
+)
+
+add_entrypoint_object(
+  sqrtl
+  SRCS
+    sqrtl.cpp
+  HDRS
+    ../sqrtl.h
+  COMPILE_OPTIONS
+    -O2
+)
diff --git a/libc/src/math/x86_64/sqrt.cpp b/libc/src/math/x86_64/sqrt.cpp
@@ -0,0 +1,20 @@
+//===-- Implementation of the sqrt function for x86_64 --------------------===//
+//
+// 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/sqrt.h"
+#include "src/__support/common.h"
+
+namespace __llvm_libc {
+
+LLVM_LIBC_FUNCTION(double, sqrt, (double x)) {
+  double result;
+  __asm__ __volatile__("sqrtsd %x1, %x0" : "=x"(result) : "x"(x));
+  return result;
+}
+
+} // namespace __llvm_libc
diff --git a/libc/src/math/x86_64/sqrtf.cpp b/libc/src/math/x86_64/sqrtf.cpp
@@ -0,0 +1,20 @@
+//===-- Implementation of the sqrtf function for x86_64 -------------------===//
+//
+// 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/sqrtf.h"
+#include "src/__support/common.h"
+
+namespace __llvm_libc {
+
+LLVM_LIBC_FUNCTION(float, sqrtf, (float x)) {
+  float result;
+  __asm__ __volatile__("sqrtss %x1, %x0" : "=x"(result) : "x"(x));
+  return result;
+}
+
+} // namespace __llvm_libc
diff --git a/libc/src/math/x86_64/sqrtl.cpp b/libc/src/math/x86_64/sqrtl.cpp
@@ -0,0 +1,20 @@
+//===-- Implementation of the sqrtl function for x86_64 -------------------===//
+//
+// 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/sqrtl.h"
+#include "src/__support/common.h"
+
+namespace __llvm_libc {
+
+LLVM_LIBC_FUNCTION(long double, sqrtl, (long double x)) {
+  long double result;
+  __asm__ __volatile__("fsqrt" : "=t"(result) : "t"(x));
+  return result;
+}
+
+} // namespace __llvm_libc
diff --git a/libc/test/src/math/CMakeLists.txt b/libc/test/src/math/CMakeLists.txt
@@ -983,63 +983,26 @@ add_fp_unittest(
     libc.src.__support.FPUtil.fputil
 )
 
-add_fp_unittest(
-  sqrtl_test
-  NEED_MPFR
-  SUITE
-    libc_math_unittests
-  SRCS
-    sqrtl_test.cpp
-  DEPENDS
-    libc.include.math
-    libc.src.math.sqrtl
-    libc.src.__support.FPUtil.fputil
-)
-
-add_fp_unittest(
-  generic_sqrtf_test
-  NEED_MPFR
-  SUITE
-    libc_math_unittests
-  SRCS
-    generic_sqrtf_test.cpp
-  DEPENDS
-    libc.src.__support.FPUtil.fputil
-    libc.src.__support.FPUtil.generic.sqrt
-  COMPILE_OPTIONS
-    -O3
-    -Wno-c++17-extensions
-)
-
-add_fp_unittest(
-  generic_sqrt_test
-  NEED_MPFR
-  SUITE
-    libc_math_unittests
-  SRCS
-    generic_sqrt_test.cpp
-  DEPENDS
-    libc.src.__support.FPUtil.fputil
-    libc.src.__support.FPUtil.generic.sqrt
-  COMPILE_OPTIONS
-    -O3
-    -Wno-c++17-extensions
-)
-
-add_fp_unittest(
-  generic_sqrtl_test
-  NEED_MPFR
-  SUITE
-    libc_math_unittests
-  SRCS
-    generic_sqrtl_test.cpp
-  DEPENDS
-    libc.src.__support.FPUtil.fputil
-    libc.src.__support.FPUtil.generic.sqrt
-  COMPILE_OPTIONS
-    -O3
-    -Wno-c++17-extensions
-)
+# The quad precision test for sqrt against MPFR currently suffers
+# from insufficient precision in MPFR calculations leading to
+# https://hal.archives-ouvertes.fr/hal-01091186/document. We will
+# renable after fixing the precision issue.
+if(${LIBC_TARGET_ARCHITECTURE_IS_X86})
+  add_fp_unittest(
+    sqrtl_test
+    NEED_MPFR
+    SUITE
+      libc_math_unittests
+    SRCS
+      sqrtl_test.cpp
+    DEPENDS
+      libc.include.math
+      libc.src.math.sqrtl
+      libc.src.__support.FPUtil.fputil
+  )
+else()
+  message(STATUS "Skipping sqrtl_test")
+endif()
 
 add_fp_unittest(
   remquof_test

diff --git a/libc/test/src/math/generic_sqrt_test.cpp b/libc/test/src/math/generic_sqrt_test.cpp
diff --git a/libc/test/src/math/generic_sqrtf_test.cpp b/libc/test/src/math/generic_sqrtf_test.cpp
diff --git a/libc/test/src/math/generic_sqrtl_test.cpp b/libc/test/src/math/generic_sqrtl_test.cpp
diff --git a/utils/bazel/llvm-project-overlay/libc/BUILD.bazel b/utils/bazel/llvm-project-overlay/libc/BUILD.bazel
@@ -80,13 +80,15 @@ fputil_common_hdrs = [
     "src/__support/FPUtil/NearestIntegerOperations.h",
     "src/__support/FPUtil/NormalFloat.h",
     "src/__support/FPUtil/PlatformDefs.h",
+    "src/__support/FPUtil/Sqrt.h",
 ]
 
 fputil_hdrs = selects.with_or({
     "//conditions:default": fputil_common_hdrs,
     PLATFORM_CPU_X86_64: fputil_common_hdrs + [
         "src/__support/FPUtil/x86_64/LongDoubleBits.h",
         "src/__support/FPUtil/x86_64/NextAfterLongDouble.h",
+        "src/__support/FPUtil/x86_64/SqrtLongDouble.h",
         "src/__support/FPUtil/x86_64/FEnvImpl.h",
     ],
     PLATFORM_CPU_ARM64: fputil_common_hdrs + [
@@ -104,31 +106,6 @@ cc_library(
     ],
 )
 
-sqrt_common_hdrs = [
-    "src/__support/FPUtil/sqrt.h",
-    "src/__support/FPUtil/generic/sqrt.h",
-    "src/__support/FPUtil/generic/sqrt_80_bit_long_double.h",
-]
-
-sqrt_hdrs = selects.with_or({
-    "//conditions:default": sqrt_common_hdrs,
-    PLATFORM_CPU_X86_64: sqrt_common_hdrs + [
-        "src/__support/FPUtil/x86_64/sqrt.h",
-    ],
-    PLATFORM_CPU_ARM64: sqrt_common_hdrs + [
-        "src/__support/FPUtil/aarch64/sqrt.h",
-    ],
-})
-
-cc_library(
-    name = "__support_fputil_sqrt",
-    hdrs = sqrt_hdrs,
-    deps = [
-        ":__support_fputil",
-        ":libc_root",
-    ],
-)
-
 ################################ fenv targets ################################
 
 libc_function(
@@ -461,23 +438,28 @@ libc_math_function(
 
 libc_math_function(
     name = "sqrt",
-    additional_deps = [
-        ":__support_fputil_sqrt",
-    ]
+    specializations = [
+        "aarch64",
+        "generic",
+        "x86_64",
+    ],
 )
 
 libc_math_function(
     name = "sqrtf",
-    additional_deps = [
-        ":__support_fputil_sqrt",
-    ]
+    specializations = [
+        "aarch64",
+        "generic",
+        "x86_64",
+    ],
 )
 
 libc_math_function(
     name = "sqrtl",
-    additional_deps = [
-        ":__support_fputil_sqrt",
-    ]
+    specializations = [
+        "generic",
+        "x86_64",
+    ],
 )
 
 libc_math_function(name = "copysign")