diff --git a/libc/src/string/memory_utils/memcmp_implementations.h b/libc/src/string/memory_utils/memcmp_implementations.h
@@ -13,6 +13,7 @@
 #include "src/__support/macros/optimization.h" // LIBC_UNLIKELY LIBC_LOOP_NOUNROLL
 #include "src/__support/macros/properties/architectures.h"
 #include "src/string/memory_utils/op_generic.h"
+#include "src/string/memory_utils/op_riscv.h"
 #include "src/string/memory_utils/utils.h" // CPtr MemcmpReturnType
 
 #include <stddef.h> // size_t
@@ -26,21 +27,17 @@
 namespace __llvm_libc {
 
 [[maybe_unused]] LIBC_INLINE MemcmpReturnType
-inline_memcmp_byte_per_byte(CPtr p1, CPtr p2, size_t offset, size_t count) {
-  LIBC_LOOP_NOUNROLL
-  for (; offset < count; ++offset)
-    if (auto value = generic::Memcmp<1>::block(p1 + offset, p2 + offset))
-      return value;
-  return MemcmpReturnType::ZERO();
+inline_memcmp_byte_per_byte(CPtr p1, CPtr p2, size_t count, size_t offset = 0) {
+  return generic::Memcmp<uint8_t>::loop_and_tail_offset(p1, p2, count, offset);
 }
 
 [[maybe_unused]] LIBC_INLINE MemcmpReturnType
 inline_memcmp_aligned_access_64bit(CPtr p1, CPtr p2, size_t count) {
   constexpr size_t kAlign = sizeof(uint64_t);
   if (count <= 2 * kAlign)
-    return inline_memcmp_byte_per_byte(p1, p2, 0, count);
+    return inline_memcmp_byte_per_byte(p1, p2, count);
   size_t bytes_to_p1_align = distance_to_align_up<kAlign>(p1);
-  if (auto value = inline_memcmp_byte_per_byte(p1, p2, 0, bytes_to_p1_align))
+  if (auto value = inline_memcmp_byte_per_byte(p1, p2, bytes_to_p1_align))
     return value;
   size_t offset = bytes_to_p1_align;
   size_t p2_alignment = distance_to_align_down<kAlign>(p2 + offset);
@@ -56,21 +53,20 @@ inline_memcmp_aligned_access_64bit(CPtr p1, CPtr p2, size_t count) {
       b = load64_aligned<uint8_t, uint16_t, uint16_t, uint16_t, uint8_t>(
           p2, offset);
     uint64_t a = load64_aligned<uint64_t>(p1, offset);
-    if (a != b) {
-      // TODO use cmp_neq_uint64_t from D148717 once it's submitted.
-      return Endian::to_big_endian(a) < Endian::to_big_endian(b) ? -1 : 1;
-    }
+    if (a != b)
+      return cmp_neq_uint64_t(Endian::to_big_endian(a),
+                              Endian::to_big_endian(b));
   }
-  return inline_memcmp_byte_per_byte(p1, p2, offset, count);
+  return inline_memcmp_byte_per_byte(p1, p2, count, offset);
 }
 
 [[maybe_unused]] LIBC_INLINE MemcmpReturnType
 inline_memcmp_aligned_access_32bit(CPtr p1, CPtr p2, size_t count) {
   constexpr size_t kAlign = sizeof(uint32_t);
   if (count <= 2 * kAlign)
-    return inline_memcmp_byte_per_byte(p1, p2, 0, count);
+    return inline_memcmp_byte_per_byte(p1, p2, count);
   size_t bytes_to_p1_align = distance_to_align_up<kAlign>(p1);
-  if (auto value = inline_memcmp_byte_per_byte(p1, p2, 0, bytes_to_p1_align))
+  if (auto value = inline_memcmp_byte_per_byte(p1, p2, bytes_to_p1_align))
     return value;
   size_t offset = bytes_to_p1_align;
   size_t p2_alignment = distance_to_align_down<kAlign>(p2 + offset);
@@ -83,16 +79,10 @@ inline_memcmp_aligned_access_32bit(CPtr p1, CPtr p2, size_t count) {
     else
       b = load32_aligned<uint8_t, uint16_t, uint8_t>(p2, offset);
     uint32_t a = load32_aligned<uint32_t>(p1, offset);
-    if (a != b) {
-      // TODO use cmp_uint32_t from D148717 once it's submitted.
-      // We perform the difference as an uint64_t.
-      const int64_t diff = static_cast<int64_t>(Endian::to_big_endian(a)) -
-                           static_cast<int64_t>(Endian::to_big_endian(b));
-      // And reduce the uint64_t into an uint32_t.
-      return static_cast<int32_t>((diff >> 1) | (diff & 0xFFFF));
-    }
+    if (a != b)
+      return cmp_uint32_t(Endian::to_big_endian(a), Endian::to_big_endian(b));
   }
-  return inline_memcmp_byte_per_byte(p1, p2, offset, count);
+  return inline_memcmp_byte_per_byte(p1, p2, count, offset);
 }
 
 LIBC_INLINE MemcmpReturnType inline_memcmp(CPtr p1, CPtr p2, size_t count) {
@@ -105,7 +95,7 @@ LIBC_INLINE MemcmpReturnType inline_memcmp(CPtr p1, CPtr p2, size_t count) {
 #elif defined(LIBC_TARGET_ARCH_IS_RISCV32)
   return inline_memcmp_aligned_access_32bit(p1, p2, count);
 #else
-  return inline_memcmp_byte_per_byte(p1, p2, 0, count);
+  return inline_memcmp_byte_per_byte(p1, p2, count);
 #endif
 }
 

diff --git a/libc/src/string/memory_utils/memmove_implementations.h b/libc/src/string/memory_utils/memmove_implementations.h
@@ -38,27 +38,27 @@ LIBC_INLINE void inline_memmove(Ptr dst, CPtr src, size_t count) {
 #if defined(LIBC_TARGET_ARCH_IS_X86) || defined(LIBC_TARGET_ARCH_IS_AARCH64)
 #if defined(LIBC_TARGET_ARCH_IS_X86)
 #if defined(__AVX512F__)
-  using uint128_t = uint8x16_t;
-  using uint256_t = uint8x32_t;
-  using uint512_t = uint8x64_t;
+  using uint128_t = generic_v128;
+  using uint256_t = generic_v256;
+  using uint512_t = generic_v512;
 #elif defined(__AVX__)
-  using uint128_t = uint8x16_t;
-  using uint256_t = uint8x32_t;
-  using uint512_t = cpp::array<uint8x32_t, 2>;
+  using uint128_t = generic_v128;
+  using uint256_t = generic_v256;
+  using uint512_t = cpp::array<generic_v256, 2>;
 #elif defined(__SSE2__)
-  using uint128_t = uint8x16_t;
-  using uint256_t = cpp::array<uint8x16_t, 2>;
-  using uint512_t = cpp::array<uint8x16_t, 4>;
+  using uint128_t = generic_v128;
+  using uint256_t = cpp::array<generic_v128, 2>;
+  using uint512_t = cpp::array<generic_v128, 4>;
 #else
   using uint128_t = cpp::array<uint64_t, 2>;
   using uint256_t = cpp::array<uint64_t, 4>;
   using uint512_t = cpp::array<uint64_t, 8>;
 #endif
 #elif defined(LIBC_TARGET_ARCH_IS_AARCH64)
   static_assert(aarch64::kNeon, "aarch64 supports vector types");
-  using uint128_t = uint8x16_t;
-  using uint256_t = uint8x32_t;
-  using uint512_t = uint8x64_t;
+  using uint128_t = generic_v128;
+  using uint256_t = generic_v256;
+  using uint512_t = generic_v512;
 #endif
   if (count == 0)
     return;

diff --git a/libc/src/string/memory_utils/memset_implementations.h b/libc/src/string/memory_utils/memset_implementations.h
@@ -60,17 +60,17 @@ inline_memset_aligned_access_64bit(Ptr dst, uint8_t value, size_t count) {
 [[maybe_unused]] LIBC_INLINE static void
 inline_memset_x86(Ptr dst, uint8_t value, size_t count) {
 #if defined(__AVX512F__)
-  using uint128_t = uint8x16_t;
-  using uint256_t = uint8x32_t;
-  using uint512_t = uint8x64_t;
+  using uint128_t = generic_v128;
+  using uint256_t = generic_v256;
+  using uint512_t = generic_v512;
 #elif defined(__AVX__)
-  using uint128_t = uint8x16_t;
-  using uint256_t = uint8x32_t;
-  using uint512_t = cpp::array<uint8x32_t, 2>;
+  using uint128_t = generic_v128;
+  using uint256_t = generic_v256;
+  using uint512_t = cpp::array<generic_v256, 2>;
 #elif defined(__SSE2__)
-  using uint128_t = uint8x16_t;
-  using uint256_t = cpp::array<uint8x16_t, 2>;
-  using uint512_t = cpp::array<uint8x16_t, 4>;
+  using uint128_t = generic_v128;
+  using uint256_t = cpp::array<generic_v128, 2>;
+  using uint512_t = cpp::array<generic_v128, 4>;
 #else
   using uint128_t = cpp::array<uint64_t, 2>;
   using uint256_t = cpp::array<uint64_t, 4>;
@@ -106,9 +106,9 @@ inline_memset_x86(Ptr dst, uint8_t value, size_t count) {
 [[maybe_unused]] LIBC_INLINE static void
 inline_memset_aarch64(Ptr dst, uint8_t value, size_t count) {
   static_assert(aarch64::kNeon, "aarch64 supports vector types");
-  using uint128_t = uint8x16_t;
-  using uint256_t = uint8x32_t;
-  using uint512_t = uint8x64_t;
+  using uint128_t = generic_v128;
+  using uint256_t = generic_v256;
+  using uint512_t = generic_v512;
   if (count == 0)
     return;
   if (count <= 3) {

diff --git a/libc/src/string/memory_utils/op_aarch64.h b/libc/src/string/memory_utils/op_aarch64.h
@@ -48,7 +48,7 @@ struct BzeroCacheLine {
       offset += SIZE;
     } while (offset < count - SIZE);
     // Unaligned store, we can't use 'dc zva' here.
-    generic::Memset<uint8x64_t>::tail(dst, value, count);
+    generic::Memset<generic_v512>::tail(dst, value, count);
   }
 };
 
@@ -171,6 +171,100 @@ template <size_t Size> struct Bcmp {
 
 } // namespace __llvm_libc::aarch64
 
+namespace __llvm_libc::generic {
+
+///////////////////////////////////////////////////////////////////////////////
+// Specializations for uint16_t
+template <> struct cmp_is_expensive<uint16_t> : public cpp::false_type {};
+template <> LIBC_INLINE bool eq<uint16_t>(CPtr p1, CPtr p2, size_t offset) {
+  return load<uint16_t>(p1, offset) == load<uint16_t>(p2, offset);
+}
+template <>
+LIBC_INLINE uint32_t neq<uint16_t>(CPtr p1, CPtr p2, size_t offset) {
+  return load<uint16_t>(p1, offset) ^ load<uint16_t>(p2, offset);
+}
+template <>
+LIBC_INLINE MemcmpReturnType cmp<uint16_t>(CPtr p1, CPtr p2, size_t offset) {
+  return static_cast<int32_t>(load_be<uint16_t>(p1, offset)) -
+         static_cast<int32_t>(load_be<uint16_t>(p2, offset));
+}
+
+///////////////////////////////////////////////////////////////////////////////
+// Specializations for uint32_t
+template <> struct cmp_is_expensive<uint32_t> : cpp::false_type {};
+template <>
+LIBC_INLINE uint32_t neq<uint32_t>(CPtr p1, CPtr p2, size_t offset) {
+  return load<uint32_t>(p1, offset) ^ load<uint32_t>(p2, offset);
+}
+template <>
+LIBC_INLINE MemcmpReturnType cmp<uint32_t>(CPtr p1, CPtr p2, size_t offset) {
+  const auto a = load_be<uint32_t>(p1, offset);
+  const auto b = load_be<uint32_t>(p2, offset);
+  return a > b ? 1 : a < b ? -1 : 0;
+}
+
+///////////////////////////////////////////////////////////////////////////////
+// Specializations for uint64_t
+template <> struct cmp_is_expensive<uint64_t> : cpp::false_type {};
+template <>
+LIBC_INLINE uint32_t neq<uint64_t>(CPtr p1, CPtr p2, size_t offset) {
+  return load<uint64_t>(p1, offset) != load<uint64_t>(p2, offset);
+}
+template <>
+LIBC_INLINE MemcmpReturnType cmp<uint64_t>(CPtr p1, CPtr p2, size_t offset) {
+  const auto a = load_be<uint64_t>(p1, offset);
+  const auto b = load_be<uint64_t>(p2, offset);
+  if (a != b)
+    return a > b ? 1 : -1;
+  return MemcmpReturnType::ZERO();
+}
+
+///////////////////////////////////////////////////////////////////////////////
+// Specializations for uint8x16_t
+template <> struct is_vector<uint8x16_t> : cpp::true_type {};
+template <> struct cmp_is_expensive<uint8x16_t> : cpp::false_type {};
+template <>
+LIBC_INLINE uint32_t neq<uint8x16_t>(CPtr p1, CPtr p2, size_t offset) {
+  for (size_t i = 0; i < 2; ++i) {
+    auto a = load<uint64_t>(p1, offset);
+    auto b = load<uint64_t>(p2, offset);
+    uint32_t cond = a != b;
+    if (cond)
+      return cond;
+    offset += sizeof(uint64_t);
+  }
+  return 0;
+}
+template <>
+LIBC_INLINE MemcmpReturnType cmp<uint8x16_t>(CPtr p1, CPtr p2, size_t offset) {
+  for (size_t i = 0; i < 2; ++i) {
+    auto a = load_be<uint64_t>(p1, offset);
+    auto b = load_be<uint64_t>(p2, offset);
+    if (a != b)
+      return cmp_neq_uint64_t(a, b);
+    offset += sizeof(uint64_t);
+  }
+  return MemcmpReturnType::ZERO();
+}
+
+///////////////////////////////////////////////////////////////////////////////
+// Specializations for uint8x16x2_t
+template <> struct is_vector<uint8x16x2_t> : cpp::true_type {};
+template <> struct cmp_is_expensive<uint8x16x2_t> : cpp::false_type {};
+template <>
+LIBC_INLINE MemcmpReturnType cmp<uint8x16x2_t>(CPtr p1, CPtr p2,
+                                               size_t offset) {
+  for (size_t i = 0; i < 4; ++i) {
+    auto a = load_be<uint64_t>(p1, offset);
+    auto b = load_be<uint64_t>(p2, offset);
+    if (a != b)
+      return cmp_neq_uint64_t(a, b);
+    offset += sizeof(uint64_t);
+  }
+  return MemcmpReturnType::ZERO();
+}
+} // namespace __llvm_libc::generic
+
 #endif // LIBC_TARGET_ARCH_IS_AARCH64
 
 #endif // LLVM_LIBC_SRC_STRING_MEMORY_UTILS_OP_AARCH64_H
diff --git a/libc/src/string/memory_utils/op_generic.h b/libc/src/string/memory_utils/op_generic.h
diff --git a/libc/src/string/memory_utils/op_riscv.h b/libc/src/string/memory_utils/op_riscv.h
@@ -0,0 +1,84 @@
+//===-- RISC-V implementation of memory function building blocks ----------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+//
+// This file provides x86 specific building blocks to compose memory functions.
+//
+//===----------------------------------------------------------------------===//
+#ifndef LLVM_LIBC_SRC_STRING_MEMORY_UTILS_OP_RISCV_H
+#define LLVM_LIBC_SRC_STRING_MEMORY_UTILS_OP_RISCV_H
+
+#include "src/__support/macros/properties/architectures.h"
+
+#if defined(LIBC_TARGET_ARCH_IS_ANY_RISCV)
+
+#include "src/__support/common.h"
+#include "src/string/memory_utils/op_generic.h"
+
+namespace __llvm_libc::generic {
+
+///////////////////////////////////////////////////////////////////////////////
+// Specializations for uint16_t
+template <> struct cmp_is_expensive<uint16_t> : public cpp::false_type {};
+template <> LIBC_INLINE bool eq<uint16_t>(CPtr p1, CPtr p2, size_t offset) {
+  return load<uint16_t>(p1, offset) == load<uint16_t>(p2, offset);
+}
+template <>
+LIBC_INLINE uint32_t neq<uint16_t>(CPtr p1, CPtr p2, size_t offset) {
+  return load<uint16_t>(p1, offset) ^ load<uint16_t>(p2, offset);
+}
+template <>
+LIBC_INLINE MemcmpReturnType cmp<uint16_t>(CPtr p1, CPtr p2, size_t offset) {
+  return static_cast<int32_t>(load_be<uint16_t>(p1, offset)) -
+         static_cast<int32_t>(load_be<uint16_t>(p2, offset));
+}
+template <>
+LIBC_INLINE MemcmpReturnType cmp_neq<uint16_t>(CPtr p1, CPtr p2, size_t offset);
+
+///////////////////////////////////////////////////////////////////////////////
+// Specializations for uint32_t
+template <> struct cmp_is_expensive<uint32_t> : public cpp::false_type {};
+template <> LIBC_INLINE bool eq<uint32_t>(CPtr p1, CPtr p2, size_t offset) {
+  return load<uint32_t>(p1, offset) == load<uint32_t>(p2, offset);
+}
+template <>
+LIBC_INLINE uint32_t neq<uint32_t>(CPtr p1, CPtr p2, size_t offset) {
+  return load<uint32_t>(p1, offset) ^ load<uint32_t>(p2, offset);
+}
+template <>
+LIBC_INLINE MemcmpReturnType cmp<uint32_t>(CPtr p1, CPtr p2, size_t offset) {
+  const auto a = load_be<uint32_t>(p1, offset);
+  const auto b = load_be<uint32_t>(p2, offset);
+  return cmp_uint32_t(a, b);
+}
+template <>
+LIBC_INLINE MemcmpReturnType cmp_neq<uint32_t>(CPtr p1, CPtr p2, size_t offset);
+
+///////////////////////////////////////////////////////////////////////////////
+// Specializations for uint64_t
+template <> struct cmp_is_expensive<uint64_t> : public cpp::true_type {};
+template <> LIBC_INLINE bool eq<uint64_t>(CPtr p1, CPtr p2, size_t offset) {
+  return load<uint64_t>(p1, offset) == load<uint64_t>(p2, offset);
+}
+template <>
+LIBC_INLINE uint32_t neq<uint64_t>(CPtr p1, CPtr p2, size_t offset) {
+  return !eq<uint64_t>(p1, p2, offset);
+}
+template <>
+LIBC_INLINE MemcmpReturnType cmp<uint64_t>(CPtr p1, CPtr p2, size_t offset);
+template <>
+LIBC_INLINE MemcmpReturnType cmp_neq<uint64_t>(CPtr p1, CPtr p2,
+                                               size_t offset) {
+  const auto a = load_be<uint64_t>(p1, offset);
+  const auto b = load_be<uint64_t>(p2, offset);
+  return cmp_neq_uint64_t(a, b);
+}
+
+} // namespace __llvm_libc::generic
+
+#endif // LIBC_TARGET_ARCH_IS_ANY_RISCV
+#endif // LLVM_LIBC_SRC_STRING_MEMORY_UTILS_OP_RISCV_H
diff --git a/libc/src/string/memory_utils/op_x86.h b/libc/src/string/memory_utils/op_x86.h
diff --git a/libc/src/string/memory_utils/utils.h b/libc/src/string/memory_utils/utils.h
@@ -15,9 +15,10 @@
 #include "src/__support/endian.h"
 #include "src/__support/macros/attributes.h" // LIBC_INLINE
 #include "src/__support/macros/config.h"     // LIBC_HAS_BUILTIN
+#include "src/__support/macros/properties/architectures.h"
 
 #include <stddef.h> // size_t
-#include <stdint.h> // intptr_t / uintptr_t
+#include <stdint.h> // intptr_t / uintptr_t / INT32_MAX / INT32_MIN
 
 namespace __llvm_libc {
 
@@ -149,6 +150,56 @@ template <typename T> struct StrictIntegralType {
 using MemcmpReturnType = StrictIntegralType<int32_t>;
 using BcmpReturnType = StrictIntegralType<uint32_t>;
 
+// This implements the semantic of 'memcmp' returning a negative value when 'a'
+// is less than 'b', '0' when 'a' equals 'b' and a positive number otherwise.
+LIBC_INLINE MemcmpReturnType cmp_uint32_t(uint32_t a, uint32_t b) {
+  // We perform the difference as an int64_t.
+  const int64_t diff = static_cast<int64_t>(a) - static_cast<int64_t>(b);
+  // For the int64_t to int32_t conversion we want the following properties:
+  // - int32_t[31:31] == 1 iff diff < 0
+  // - int32_t[31:0] == 0 iff diff == 0
+
+  // We also observe that:
+  // - When diff < 0: diff[63:32] == 0xffffffff and diff[31:0] != 0
+  // - When diff > 0: diff[63:32] == 0 and diff[31:0] != 0
+  // - When diff == 0: diff[63:32] == 0 and diff[31:0] == 0
+  // - https://godbolt.org/z/8W7qWP6e5
+  // - This implies that we can only look at diff[32:32] for determining the
+  // sign bit for the returned int32_t.
+
+  // So, we do the following:
+  // - int32_t[31:31] = diff[32:32]
+  // - int32_t[30:0] = diff[31:0] == 0 ? 0 : non-0.
+
+  // And, we can achieve the above by the expression below. We could have also
+  // used (diff64 >> 1) | (diff64 & 0x1) but (diff64 & 0xFFFF) is faster than
+  // (diff64 & 0x1). https://godbolt.org/z/j3b569rW1
+  return static_cast<int32_t>((diff >> 1) | (diff & 0xFFFF));
+}
+
+// Returns a negative value if 'a' is less than 'b' and a positive value
+// otherwise. This implements the semantic of 'memcmp' when we know that 'a' and
+// 'b' differ.
+LIBC_INLINE MemcmpReturnType cmp_neq_uint64_t(uint64_t a, uint64_t b) {
+#if defined(LIBC_TARGET_ARCH_IS_X86_64)
+  // On x86, we choose the returned values so that they are just one unit appart
+  // as this allows for better code generation.
+  static constexpr int32_t POSITIVE = INT32_MAX;
+  static constexpr int32_t NEGATIVE = INT32_MIN;
+  static_assert(cpp::bit_cast<uint32_t>(NEGATIVE) -
+                    cpp::bit_cast<uint32_t>(POSITIVE) ==
+                1);
+#else
+  // On RISC-V we simply use '1' and '-1' as it leads to branchless code.
+  // On ARMv8, both strategies lead to the same performance.
+  static constexpr int32_t POSITIVE = 1;
+  static constexpr int32_t NEGATIVE = -1;
+#endif
+  static_assert(POSITIVE > 0);
+  static_assert(NEGATIVE < 0);
+  return a < b ? NEGATIVE : POSITIVE;
+}
+
 // Loads bytes from memory (possibly unaligned) and materializes them as
 // type.
 template <typename T> LIBC_INLINE T load(CPtr ptr) {
@@ -280,6 +331,16 @@ void align_to_next_boundary(T1 *__restrict &p1, T2 *__restrict &p2,
     deferred_static_assert("AlignOn must be either Arg::P1 or Arg::P2");
 }
 
+template <size_t SIZE> struct AlignHelper {
+  AlignHelper(CPtr ptr) : offset_(distance_to_next_aligned<SIZE>(ptr)) {}
+
+  LIBC_INLINE bool not_aligned() const { return offset_ != SIZE; }
+  LIBC_INLINE uintptr_t offset() const { return offset_; }
+
+private:
+  uintptr_t offset_;
+};
+
 } // namespace __llvm_libc
 
 #endif // LLVM_LIBC_SRC_MEMORY_UTILS_UTILS_H
diff --git a/libc/src/string/memory_utils/x86_64/memcmp_implementations.h b/libc/src/string/memory_utils/x86_64/memcmp_implementations.h
@@ -18,79 +18,76 @@ namespace __llvm_libc {
 
 [[maybe_unused]] LIBC_INLINE MemcmpReturnType
 inline_memcmp_generic_gt16(CPtr p1, CPtr p2, size_t count) {
-  if (LIBC_UNLIKELY(count >= 384)) {
-    if (auto value = generic::Memcmp<16>::block(p1, p2))
-      return value;
-    align_to_next_boundary<16, Arg::P1>(p1, p2, count);
-  }
-  return generic::Memcmp<16>::loop_and_tail(p1, p2, count);
+  return generic::Memcmp<uint64_t>::loop_and_tail_align_above(384, p1, p2,
+                                                              count);
 }
 
+#if defined(__SSE4_1__)
 [[maybe_unused]] LIBC_INLINE MemcmpReturnType
-inline_memcmp_x86_sse2_gt16(CPtr p1, CPtr p2, size_t count) {
-  if (LIBC_UNLIKELY(count >= 384)) {
-    if (auto value = x86::sse2::Memcmp<16>::block(p1, p2))
-      return value;
-    align_to_next_boundary<16, Arg::P1>(p1, p2, count);
-  }
-  return x86::sse2::Memcmp<16>::loop_and_tail(p1, p2, count);
+inline_memcmp_x86_sse41_gt16(CPtr p1, CPtr p2, size_t count) {
+  return generic::Memcmp<__m128i>::loop_and_tail_align_above(384, p1, p2,
+                                                             count);
 }
+#endif // __SSE4_1__
 
+#if defined(__AVX2__)
 [[maybe_unused]] LIBC_INLINE MemcmpReturnType
 inline_memcmp_x86_avx2_gt16(CPtr p1, CPtr p2, size_t count) {
   if (count <= 32)
-    return x86::sse2::Memcmp<16>::head_tail(p1, p2, count);
+    return generic::Memcmp<__m128i>::head_tail(p1, p2, count);
   if (count <= 64)
-    return x86::avx2::Memcmp<32>::head_tail(p1, p2, count);
-  if (count <= 128)
-    return x86::avx2::Memcmp<64>::head_tail(p1, p2, count);
-  if (LIBC_UNLIKELY(count >= 384)) {
-    if (auto value = x86::avx2::Memcmp<32>::block(p1, p2))
-      return value;
-    align_to_next_boundary<32, Arg::P1>(p1, p2, count);
-  }
-  return x86::avx2::Memcmp<32>::loop_and_tail(p1, p2, count);
+    return generic::Memcmp<__m256i>::head_tail(p1, p2, count);
+  return generic::Memcmp<__m256i>::loop_and_tail_align_above(384, p1, p2,
+                                                             count);
 }
+#endif // __AVX2__
 
+#if defined(__AVX512BW__)
 [[maybe_unused]] LIBC_INLINE MemcmpReturnType
 inline_memcmp_x86_avx512bw_gt16(CPtr p1, CPtr p2, size_t count) {
   if (count <= 32)
-    return x86::sse2::Memcmp<16>::head_tail(p1, p2, count);
+    return generic::Memcmp<__m128i>::head_tail(p1, p2, count);
   if (count <= 64)
-    return x86::avx2::Memcmp<32>::head_tail(p1, p2, count);
+    return generic::Memcmp<__m256i>::head_tail(p1, p2, count);
   if (count <= 128)
-    return x86::avx512bw::Memcmp<64>::head_tail(p1, p2, count);
-  if (LIBC_UNLIKELY(count >= 384)) {
-    if (auto value = x86::avx512bw::Memcmp<64>::block(p1, p2))
-      return value;
-    align_to_next_boundary<64, Arg::P1>(p1, p2, count);
-  }
-  return x86::avx512bw::Memcmp<64>::loop_and_tail(p1, p2, count);
+    return generic::Memcmp<__m512i>::head_tail(p1, p2, count);
+  return generic::Memcmp<__m512i>::loop_and_tail_align_above(384, p1, p2,
+                                                             count);
 }
+#endif // __AVX512BW__
 
 LIBC_INLINE MemcmpReturnType inline_memcmp_x86(CPtr p1, CPtr p2, size_t count) {
-
   if (count == 0)
     return MemcmpReturnType::ZERO();
   if (count == 1)
-    return generic::Memcmp<1>::block(p1, p2);
+    return generic::Memcmp<uint8_t>::block(p1, p2);
   if (count == 2)
-    return generic::Memcmp<2>::block(p1, p2);
+    return generic::Memcmp<uint16_t>::block(p1, p2);
   if (count == 3)
-    return generic::Memcmp<3>::block(p1, p2);
-  if (count <= 8)
-    return generic::Memcmp<4>::head_tail(p1, p2, count);
+    return generic::MemcmpSequence<uint16_t, uint8_t>::block(p1, p2);
+  if (count == 4)
+    return generic::Memcmp<uint32_t>::block(p1, p2);
+  if (count == 5)
+    return generic::MemcmpSequence<uint32_t, uint8_t>::block(p1, p2);
+  if (count == 6)
+    return generic::MemcmpSequence<uint32_t, uint16_t>::block(p1, p2);
+  if (count == 7)
+    return generic::Memcmp<uint32_t>::head_tail(p1, p2, 7);
+  if (count == 8)
+    return generic::Memcmp<uint64_t>::block(p1, p2);
   if (count <= 16)
-    return generic::Memcmp<8>::head_tail(p1, p2, count);
-  if constexpr (x86::kAvx512BW)
-    return inline_memcmp_x86_avx512bw_gt16(p1, p2, count);
-  else if constexpr (x86::kAvx2)
-    return inline_memcmp_x86_avx2_gt16(p1, p2, count);
-  else if constexpr (x86::kSse2)
-    return inline_memcmp_x86_sse2_gt16(p1, p2, count);
-  else
-    return inline_memcmp_generic_gt16(p1, p2, count);
+    return generic::Memcmp<uint64_t>::head_tail(p1, p2, count);
+#if defined(__AVX512BW__)
+  return inline_memcmp_x86_avx512bw_gt16(p1, p2, count);
+#elif defined(__AVX2__)
+  return inline_memcmp_x86_avx2_gt16(p1, p2, count);
+#elif defined(__SSE4_1__)
+  return inline_memcmp_x86_sse41_gt16(p1, p2, count);
+#else
+  return inline_memcmp_generic_gt16(p1, p2, count);
+#endif
 }
+
 } // namespace __llvm_libc
 
 #endif // LIBC_SRC_STRING_MEMORY_UTILS_X86_64_MEMCMP_IMPLEMENTATIONS_H
diff --git a/libc/test/src/string/memory_utils/op_tests.cpp b/libc/test/src/string/memory_utils/op_tests.cpp
@@ -9,14 +9,11 @@
 #include "memory_check_utils.h"
 #include "src/string/memory_utils/op_aarch64.h"
 #include "src/string/memory_utils/op_builtin.h"
-#include "src/string/memory_utils/op_generic.h"
+#include "src/string/memory_utils/op_generic.h" // LLVM_LIBC_HAS_UINT64
+#include "src/string/memory_utils/op_riscv.h"
 #include "src/string/memory_utils/op_x86.h"
 #include "test/UnitTest/Test.h"
 
-#if defined(LIBC_TARGET_ARCH_IS_X86_64) || defined(LIBC_TARGET_ARCH_IS_AARCH64)
-#define LLVM_LIBC_HAS_UINT64
-#endif
-
 namespace __llvm_libc {
 
 template <typename T> struct has_head_tail {
@@ -131,13 +128,13 @@ using MemsetImplementations = testing::TypeList<
     generic::Memset<uint64_t>, generic::Memset<cpp::array<uint64_t, 2>>,
 #endif
 #ifdef __AVX512F__
-    generic::Memset<uint8x64_t>, generic::Memset<cpp::array<uint8x64_t, 2>>,
+    generic::Memset<generic_v512>, generic::Memset<cpp::array<generic_v512, 2>>,
 #endif
 #ifdef __AVX__
-    generic::Memset<uint8x32_t>, generic::Memset<cpp::array<uint8x32_t, 2>>,
+    generic::Memset<generic_v256>, generic::Memset<cpp::array<generic_v256, 2>>,
 #endif
 #ifdef __SSE2__
-    generic::Memset<uint8x16_t>, generic::Memset<cpp::array<uint8x16_t, 2>>,
+    generic::Memset<generic_v128>, generic::Memset<cpp::array<generic_v128, 2>>,
 #endif
     generic::Memset<uint32_t>, generic::Memset<cpp::array<uint32_t, 2>>, //
     generic::Memset<uint16_t>, generic::Memset<cpp::array<uint16_t, 2>>, //
@@ -194,35 +191,36 @@ TYPED_TEST(LlvmLibcOpTest, Memset, MemsetImplementations) {
 }
 
 using BcmpImplementations = testing::TypeList<
-#ifdef __SSE2__
-    x86::sse2::Bcmp<16>,  //
-    x86::sse2::Bcmp<32>,  //
-    x86::sse2::Bcmp<64>,  //
-    x86::sse2::Bcmp<128>, //
-#endif
+#ifdef LIBC_TARGET_ARCH_IS_X86_64
+#ifdef __SSE4_1__
+    generic::Bcmp<__m128i>,
+#endif // __SSE4_1__
 #ifdef __AVX2__
-    x86::avx2::Bcmp<32>,  //
-    x86::avx2::Bcmp<64>,  //
-    x86::avx2::Bcmp<128>, //
-#endif
+    generic::Bcmp<__m256i>,
+#endif // __AVX2__
 #ifdef __AVX512BW__
-    x86::avx512bw::Bcmp<64>,  //
-    x86::avx512bw::Bcmp<128>, //
-#endif
+    generic::Bcmp<__m512i>,
+#endif // __AVX512BW__
+
+#endif // LIBC_TARGET_ARCH_IS_X86_64
 #ifdef LIBC_TARGET_ARCH_IS_AARCH64
     aarch64::Bcmp<16>, //
-    aarch64::Bcmp<32>, //
+    aarch64::Bcmp<32>,
 #endif
+#ifndef LIBC_TARGET_ARCH_IS_ARM // Removing non uint8_t types for ARM
+    generic::Bcmp<uint16_t>,
+    generic::Bcmp<uint32_t>, //
 #ifdef LLVM_LIBC_HAS_UINT64
-    generic::Bcmp<8>, //
-#endif
-    generic::Bcmp<1>,  //
-    generic::Bcmp<2>,  //
-    generic::Bcmp<4>,  //
-    generic::Bcmp<16>, //
-    generic::Bcmp<32>, //
-    generic::Bcmp<64>  //
-    >;
+    generic::Bcmp<uint64_t>,
+#endif // LLVM_LIBC_HAS_UINT64
+    generic::BcmpSequence<uint16_t, uint8_t>,
+    generic::BcmpSequence<uint32_t, uint8_t>,  //
+    generic::BcmpSequence<uint32_t, uint16_t>, //
+    generic::BcmpSequence<uint32_t, uint16_t, uint8_t>,
+#endif // LIBC_TARGET_ARCH_IS_ARM
+    generic::BcmpSequence<uint8_t, uint8_t>,
+    generic::BcmpSequence<uint8_t, uint8_t, uint8_t>, //
+    generic::Bcmp<uint8_t>>;
 
 // Adapt CheckBcmp signature to op implementation signatures.
 template <auto FnImpl>
@@ -247,7 +245,8 @@ TYPED_TEST(LlvmLibcOpTest, Bcmp, BcmpImplementations) {
         ASSERT_TRUE((CheckBcmp<BlockImpl>(span1, span2, kSize)));
     }
   }
-  { // Test head tail operations from kSize to 2 * kSize.
+  if constexpr (has_head_tail<Impl>::value) {
+    // Test head tail operations from kSize to 2 * kSize.
     static constexpr auto HeadTailImpl = CmpAdaptor<Impl::head_tail>;
     Buffer Buffer1(2 * kSize);
     Buffer Buffer2(2 * kSize);
@@ -258,7 +257,8 @@ TYPED_TEST(LlvmLibcOpTest, Bcmp, BcmpImplementations) {
       ASSERT_TRUE((CheckBcmp<HeadTailImpl>(span1, span2, size)));
     }
   }
-  { // Test loop operations from kSize to 3 * kSize.
+  if constexpr (has_loop_and_tail<Impl>::value) {
+    // Test loop operations from kSize to 3 * kSize.
     if constexpr (kSize > 1) {
       static constexpr auto LoopImpl = CmpAdaptor<Impl::loop_and_tail>;
       Buffer Buffer1(3 * kSize);
@@ -274,32 +274,33 @@ TYPED_TEST(LlvmLibcOpTest, Bcmp, BcmpImplementations) {
 }
 
 using MemcmpImplementations = testing::TypeList<
+#ifdef LIBC_TARGET_ARCH_IS_X86_64
 #ifdef __SSE2__
-    x86::sse2::Memcmp<16>,  //
-    x86::sse2::Memcmp<32>,  //
-    x86::sse2::Memcmp<64>,  //
-    x86::sse2::Memcmp<128>, //
+    generic::Memcmp<__m128i>, //
 #endif
 #ifdef __AVX2__
-    x86::avx2::Memcmp<32>,  //
-    x86::avx2::Memcmp<64>,  //
-    x86::avx2::Memcmp<128>, //
+    generic::Memcmp<__m256i>, //
 #endif
 #ifdef __AVX512BW__
-    x86::avx512bw::Memcmp<64>,  //
-    x86::avx512bw::Memcmp<128>, //
+    generic::Memcmp<__m512i>, //
 #endif
-#ifdef LLVM_LIBC_HAS_UINT64
-    generic::Memcmp<8>, //
+#endif // LIBC_TARGET_ARCH_IS_X86_64
+#ifdef LIBC_TARGET_ARCH_IS_AARCH64
+    generic::Memcmp<uint8x16_t>, //
+    generic::Memcmp<uint8x16x2_t>,
 #endif
-    generic::Memcmp<1>,  //
-    generic::Memcmp<2>,  //
-    generic::Memcmp<3>,  //
-    generic::Memcmp<4>,  //
-    generic::Memcmp<16>, //
-    generic::Memcmp<32>, //
-    generic::Memcmp<64>  //
-    >;
+#ifndef LIBC_TARGET_ARCH_IS_ARM // Removing non uint8_t types for ARM
+    generic::Memcmp<uint16_t>,
+    generic::Memcmp<uint32_t>, //
+#ifdef LLVM_LIBC_HAS_UINT64
+    generic::Memcmp<uint64_t>,
+#endif // LLVM_LIBC_HAS_UINT64
+    generic::MemcmpSequence<uint16_t, uint8_t>,
+    generic::MemcmpSequence<uint32_t, uint16_t, uint8_t>, //
+#endif // LIBC_TARGET_ARCH_IS_ARM
+    generic::MemcmpSequence<uint8_t, uint8_t>,
+    generic::MemcmpSequence<uint8_t, uint8_t, uint8_t>,
+    generic::Memcmp<uint8_t>>;
 
 TYPED_TEST(LlvmLibcOpTest, Memcmp, MemcmpImplementations) {
   using Impl = ParamType;
@@ -314,7 +315,8 @@ TYPED_TEST(LlvmLibcOpTest, Memcmp, MemcmpImplementations) {
         ASSERT_TRUE((CheckMemcmp<BlockImpl>(span1, span2, kSize)));
     }
   }
-  { // Test head tail operations from kSize to 2 * kSize.
+  if constexpr (has_head_tail<Impl>::value) {
+    // Test head tail operations from kSize to 2 * kSize.
     static constexpr auto HeadTailImpl = CmpAdaptor<Impl::head_tail>;
     Buffer Buffer1(2 * kSize);
     Buffer Buffer2(2 * kSize);
@@ -325,7 +327,8 @@ TYPED_TEST(LlvmLibcOpTest, Memcmp, MemcmpImplementations) {
       ASSERT_TRUE((CheckMemcmp<HeadTailImpl>(span1, span2, size)));
     }
   }
-  { // Test loop operations from kSize to 3 * kSize.
+  if constexpr (has_loop_and_tail<Impl>::value) {
+    // Test loop operations from kSize to 3 * kSize.
     if constexpr (kSize > 1) {
       static constexpr auto LoopImpl = CmpAdaptor<Impl::loop_and_tail>;
       Buffer Buffer1(3 * kSize);

diff --git a/utils/bazel/llvm-project-overlay/libc/BUILD.bazel b/utils/bazel/llvm-project-overlay/libc/BUILD.bazel
@@ -1960,6 +1960,7 @@ libc_support_library(
         "src/string/memory_utils/op_aarch64.h",
         "src/string/memory_utils/op_builtin.h",
         "src/string/memory_utils/op_generic.h",
+        "src/string/memory_utils/op_riscv.h",
         "src/string/memory_utils/op_x86.h",
         "src/string/memory_utils/utils.h",
     ],