diff --git a/src/bele.h b/src/bele.h
index 1b8ee586..ca2282cf 100644
--- a/src/bele.h
+++ b/src/bele.h
@@ -42,39 +42,103 @@
 #define bswap64 upx_bswap64
 #endif
 
+/*************************************************************************
+// XE - eXtended Endian compatibility
+// try to detect XX16 vs XX32 vs XX64 size mismatches
+**************************************************************************/
+
+#if 0 // permissive version using "void *"
+
+#define REQUIRE_XE16 /*empty*/
+#define REQUIRE_XE24 /*empty*/
+#define REQUIRE_XE32 /*empty*/
+#define REQUIRE_XE64 /*empty*/
+typedef void XE16;
+typedef void XE24;
+typedef void XE32;
+typedef void XE64;
+
+#else // permissive version
+
+// forward declarations
+struct BE16;
+struct BE32;
+struct BE64;
+struct LE16;
+struct LE32;
+struct LE64;
+
+// Note:
+//   void is explicitly allowed, but there is no automatic pointer conversion because of template
+//   char is explicitly allowed
+//   byte is explicitly allowed
+template <class T>
+static inline constexpr bool is_xe16_type =
+    upx::is_same_any_v<T, void, char, byte, upx_int16_t, upx_uint16_t, BE16, LE16>;
+template <class T>
+static inline constexpr bool is_xe24_type = upx::is_same_any_v<T, void, char, byte>;
+template <class T>
+static inline constexpr bool is_xe32_type =
+    upx::is_same_any_v<T, void, char, byte, upx_int32_t, upx_uint32_t, BE32, LE32>;
+template <class T>
+static inline constexpr bool is_xe64_type =
+    upx::is_same_any_v<T, void, char, byte, upx_int64_t, upx_uint64_t, BE64, LE64>;
+
+template <class T>
+using enable_if_xe16 = std::enable_if_t<is_xe16_type<T>, T>;
+template <class T>
+using enable_if_xe24 = std::enable_if_t<is_xe24_type<T>, T>;
+template <class T>
+using enable_if_xe32 = std::enable_if_t<is_xe32_type<T>, T>;
+template <class T>
+using enable_if_xe64 = std::enable_if_t<is_xe64_type<T>, T>;
+
+#define REQUIRE_XE16 template <class XE16, class = enable_if_xe16<XE16> >
+#define REQUIRE_XE24 template <class XE24, class = enable_if_xe24<XE24> >
+#define REQUIRE_XE32 template <class XE32, class = enable_if_xe32<XE32> >
+#define REQUIRE_XE64 template <class XE64, class = enable_if_xe64<XE64> >
+
+#endif // permissive version
+
 /*************************************************************************
 // core - NE
 **************************************************************************/
 
-static forceinline unsigned get_ne16(const void *p) noexcept {
+REQUIRE_XE16
+static forceinline unsigned get_ne16(const XE16 *p) noexcept {
     upx_uint16_t v = 0;
     upx_memcpy_inline(&v, p, sizeof(v));
     return v;
 }
 
-static forceinline unsigned get_ne32(const void *p) noexcept {
+REQUIRE_XE32
+static forceinline unsigned get_ne32(const XE32 *p) noexcept {
     upx_uint32_t v = 0;
     upx_memcpy_inline(&v, p, sizeof(v));
     return v;
 }
 
-static forceinline upx_uint64_t get_ne64(const void *p) noexcept {
+REQUIRE_XE64
+static forceinline upx_uint64_t get_ne64(const XE64 *p) noexcept {
     upx_uint64_t v = 0;
     upx_memcpy_inline(&v, p, sizeof(v));
     return v;
 }
 
-static forceinline void set_ne16(void *p, unsigned vv) noexcept {
+REQUIRE_XE16
+static forceinline void set_ne16(XE16 *p, unsigned vv) noexcept {
     upx_uint16_t v = (upx_uint16_t) (vv & 0xffff);
     upx_memcpy_inline(p, &v, sizeof(v));
 }
 
-static forceinline void set_ne32(void *p, unsigned vv) noexcept {
+REQUIRE_XE32
+static forceinline void set_ne32(XE32 *p, unsigned vv) noexcept {
     upx_uint32_t v = vv;
     upx_memcpy_inline(p, &v, sizeof(v));
 }
 
-static forceinline void set_ne64(void *p, upx_uint64_t vv) noexcept {
+REQUIRE_XE64
+static forceinline void set_ne64(XE64 *p, upx_uint64_t vv) noexcept {
     upx_uint64_t v = vv;
     upx_memcpy_inline(p, &v, sizeof(v));
 }
@@ -138,41 +202,58 @@ static forceinline constexpr upx_uint64_t no_bswap64(upx_uint64_t v) noexcept {
 // get/set 16/32/64
 **************************************************************************/
 
-inline unsigned get_be16(const void *p) noexcept { return ne16_to_be16(get_ne16(p)); }
-inline unsigned get_be32(const void *p) noexcept { return ne32_to_be32(get_ne32(p)); }
-inline upx_uint64_t get_be64(const void *p) noexcept { return ne64_to_be64(get_ne64(p)); }
-inline unsigned get_le16(const void *p) noexcept { return ne16_to_le16(get_ne16(p)); }
-inline unsigned get_le32(const void *p) noexcept { return ne32_to_le32(get_ne32(p)); }
-inline upx_uint64_t get_le64(const void *p) noexcept { return ne64_to_le64(get_ne64(p)); }
-inline void set_be16(void *p, unsigned v) noexcept { set_ne16(p, ne16_to_be16(v)); }
-inline void set_be32(void *p, unsigned v) noexcept { set_ne32(p, ne32_to_be32(v)); }
-inline void set_be64(void *p, upx_uint64_t v) noexcept { set_ne64(p, ne64_to_be64(v)); }
-inline void set_le16(void *p, unsigned v) noexcept { set_ne16(p, ne16_to_le16(v)); }
-inline void set_le32(void *p, unsigned v) noexcept { set_ne32(p, ne32_to_le32(v)); }
-inline void set_le64(void *p, upx_uint64_t v) noexcept { set_ne64(p, ne64_to_le64(v)); }
+REQUIRE_XE16
+inline unsigned get_be16(const XE16 *p) noexcept { return ne16_to_be16(get_ne16(p)); }
+REQUIRE_XE32
+inline unsigned get_be32(const XE32 *p) noexcept { return ne32_to_be32(get_ne32(p)); }
+REQUIRE_XE64
+inline upx_uint64_t get_be64(const XE64 *p) noexcept { return ne64_to_be64(get_ne64(p)); }
+REQUIRE_XE16
+inline unsigned get_le16(const XE16 *p) noexcept { return ne16_to_le16(get_ne16(p)); }
+REQUIRE_XE32
+inline unsigned get_le32(const XE32 *p) noexcept { return ne32_to_le32(get_ne32(p)); }
+REQUIRE_XE64
+inline upx_uint64_t get_le64(const XE64 *p) noexcept { return ne64_to_le64(get_ne64(p)); }
+
+REQUIRE_XE16
+inline void set_be16(XE16 *p, unsigned v) noexcept { set_ne16(p, ne16_to_be16(v)); }
+REQUIRE_XE32
+inline void set_be32(XE32 *p, unsigned v) noexcept { set_ne32(p, ne32_to_be32(v)); }
+REQUIRE_XE64
+inline void set_be64(XE64 *p, upx_uint64_t v) noexcept { set_ne64(p, ne64_to_be64(v)); }
+REQUIRE_XE16
+inline void set_le16(XE16 *p, unsigned v) noexcept { set_ne16(p, ne16_to_le16(v)); }
+REQUIRE_XE32
+inline void set_le32(XE32 *p, unsigned v) noexcept { set_ne32(p, ne32_to_le32(v)); }
+REQUIRE_XE64
+inline void set_le64(XE64 *p, upx_uint64_t v) noexcept { set_ne64(p, ne64_to_le64(v)); }
 
 /*************************************************************************
 // get/set 24/26
 **************************************************************************/
 
-inline unsigned get_be24(const void *p) noexcept {
+REQUIRE_XE24
+inline unsigned get_be24(const XE24 *p) noexcept {
     const byte *b = ACC_CCAST(const byte *, p);
     return (b[0] << 16) | (b[1] << 8) | (b[2] << 0);
 }
 
-inline unsigned get_le24(const void *p) noexcept {
+REQUIRE_XE24
+inline unsigned get_le24(const XE24 *p) noexcept {
     const byte *b = ACC_CCAST(const byte *, p);
     return (b[0] << 0) | (b[1] << 8) | (b[2] << 16);
 }
 
-inline void set_be24(void *p, unsigned v) noexcept {
+REQUIRE_XE24
+inline void set_be24(XE24 *p, unsigned v) noexcept {
     byte *b = ACC_PCAST(byte *, p);
     b[0] = ACC_ICONV(byte, (v >> 16) & 0xff);
     b[1] = ACC_ICONV(byte, (v >> 8) & 0xff);
     b[2] = ACC_ICONV(byte, (v >> 0) & 0xff);
 }
 
-inline void set_le24(void *p, unsigned v) noexcept {
+REQUIRE_XE24
+inline void set_le24(XE24 *p, unsigned v) noexcept {
     byte *b = ACC_PCAST(byte *, p);
     b[0] = ACC_ICONV(byte, (v >> 0) & 0xff);
     b[1] = ACC_ICONV(byte, (v >> 8) & 0xff);
@@ -201,56 +282,64 @@ inline void set_le14_5(void *p, unsigned v) noexcept {
 // get signed values
 **************************************************************************/
 
-static forceinline int sign_extend(unsigned v, unsigned bits) noexcept {
+static forceinline constexpr int sign_extend(unsigned v, unsigned bits) noexcept {
     const unsigned sign_bit = 1u << (bits - 1);
     v &= sign_bit | (sign_bit - 1);
     v |= 0u - (v & sign_bit);
     return ACC_ICAST(int, v);
 }
 
-static forceinline upx_int64_t sign_extend(upx_uint64_t v, unsigned bits) noexcept {
+static forceinline constexpr upx_int64_t sign_extend(upx_uint64_t v, unsigned bits) noexcept {
     const upx_uint64_t sign_bit = 1ull << (bits - 1);
     v &= sign_bit | (sign_bit - 1);
     v |= 0ull - (v & sign_bit);
     return ACC_ICAST(upx_int64_t, v);
 }
 
-inline int get_be16_signed(const void *p) noexcept {
+REQUIRE_XE16
+inline int get_be16_signed(const XE16 *p) noexcept {
     unsigned v = get_be16(p);
     return sign_extend(v, 16);
 }
 
-inline int get_be24_signed(const void *p) noexcept {
+REQUIRE_XE24
+inline int get_be24_signed(const XE24 *p) noexcept {
     unsigned v = get_be24(p);
     return sign_extend(v, 24);
 }
 
-inline int get_be32_signed(const void *p) noexcept {
+REQUIRE_XE32
+inline int get_be32_signed(const XE32 *p) noexcept {
     unsigned v = get_be32(p);
     return sign_extend(v, 32);
 }
 
-inline upx_int64_t get_be64_signed(const void *p) noexcept {
+REQUIRE_XE64
+inline upx_int64_t get_be64_signed(const XE64 *p) noexcept {
     upx_uint64_t v = get_be64(p);
     return sign_extend(v, 64);
 }
 
-inline int get_le16_signed(const void *p) noexcept {
+REQUIRE_XE16
+inline int get_le16_signed(const XE16 *p) noexcept {
     unsigned v = get_le16(p);
     return sign_extend(v, 16);
 }
 
-inline int get_le24_signed(const void *p) noexcept {
+REQUIRE_XE24
+inline int get_le24_signed(const XE24 *p) noexcept {
     unsigned v = get_le24(p);
     return sign_extend(v, 24);
 }
 
-inline int get_le32_signed(const void *p) noexcept {
+REQUIRE_XE32
+inline int get_le32_signed(const XE32 *p) noexcept {
     unsigned v = get_le32(p);
     return sign_extend(v, 32);
 }
 
-inline upx_int64_t get_le64_signed(const void *p) noexcept {
+REQUIRE_XE64
+inline upx_int64_t get_le64_signed(const XE64 *p) noexcept {
     upx_uint64_t v = get_le64(p);
     return sign_extend(v, 64);
 }
@@ -633,6 +722,7 @@ inline unsigned ALIGN_DOWN(const LE32 &a, unsigned b) { return ALIGN_DOWN(unsign
 inline unsigned ALIGN_UP(unsigned a, const LE32 &b) { return ALIGN_UP(a, unsigned(b)); }
 inline unsigned ALIGN_UP(const LE32 &a, unsigned b) { return ALIGN_UP(unsigned(a), b); }
 
+// TODO: introduce upx::umax() and upx::umin()
 inline unsigned UPX_MAX(unsigned a, const BE16 &b) { return UPX_MAX(a, unsigned(b)); }
 inline unsigned UPX_MAX(const BE16 &a, unsigned b) { return UPX_MAX(unsigned(a), b); }
 inline unsigned UPX_MIN(unsigned a, const BE16 &b) { return UPX_MIN(a, unsigned(b)); }
diff --git a/src/check/dt_check.cpp b/src/check/dt_check.cpp
index c66b8f21..8f2251e3 100644
--- a/src/check/dt_check.cpp
+++ b/src/check/dt_check.cpp
@@ -113,6 +113,17 @@ ACC_COMPILE_TIME_ASSERT_HEADER(bswap64(bswap64(0xf8f7f6f5f4f3f2f1ull)) ==
                                no_bswap64(0xf8f7f6f5f4f3f2f1ull))
 #endif
 
+ACC_COMPILE_TIME_ASSERT_HEADER(sign_extend(0u + 0, 8) == 0)
+ACC_COMPILE_TIME_ASSERT_HEADER(sign_extend(0u + 1, 8) == 1)
+ACC_COMPILE_TIME_ASSERT_HEADER(sign_extend(0u + 127, 8) == 127)
+ACC_COMPILE_TIME_ASSERT_HEADER(sign_extend(0u + 128, 8) == -128)
+ACC_COMPILE_TIME_ASSERT_HEADER(sign_extend(0u - 1, 8) == -1)
+ACC_COMPILE_TIME_ASSERT_HEADER(sign_extend(0u + 256, 8) == 0)
+ACC_COMPILE_TIME_ASSERT_HEADER(sign_extend(0u + 257, 8) == 1)
+ACC_COMPILE_TIME_ASSERT_HEADER(sign_extend(0u + 383, 8) == 127)
+ACC_COMPILE_TIME_ASSERT_HEADER(sign_extend(0u + 384, 8) == -128)
+ACC_COMPILE_TIME_ASSERT_HEADER(sign_extend(0u + 511, 8) == -1)
+
 ACC_COMPILE_TIME_ASSERT_HEADER(CHAR_BIT == 8)
 #if 0 // does not work with MSVC
 #if '\0' - 1 < 0
diff --git a/src/conf.h b/src/conf.h
index fc05d803..fd731f6e 100644
--- a/src/conf.h
+++ b/src/conf.h
@@ -109,8 +109,7 @@ inline constexpr bool upx_is_integral_v = upx_is_integral<T>::value;
 // protect against integer overflows and malicious header fields
 // see C 11 standard, Annex K
 //
-// this limits uncompressed_size to about 640 MiB which is perfectly fine:
-//   "640 MiB Ought to be Enough for Anyone" --Markus F.X.J. Oberhumer, 1981 ;-)
+// this limits uncompressed_size to about 682 MiB (715_128_832 bytes)
 typedef size_t upx_rsize_t;
 #define UPX_RSIZE_MAX       UPX_RSIZE_MAX_MEM
 #define UPX_RSIZE_MAX_MEM   (768 * 1024 * 1024)   // DO NOT CHANGE !!!
diff --git a/src/util/membuffer.cpp b/src/util/membuffer.cpp
index f5398082..061a9ac5 100644
--- a/src/util/membuffer.cpp
+++ b/src/util/membuffer.cpp
@@ -321,8 +321,11 @@ TEST_CASE("MemBuffer::getSizeForCompression") {
     CHECK(MemBuffer::getSizeForCompression(1024 * 1024) == 1180160);         // 0x00100000
     CHECK(MemBuffer::getSizeForCompression(64 * 1024 * 1024) == 75497984);   // 0x04000000
     CHECK(MemBuffer::getSizeForCompression(512 * 1024 * 1024) == 603980288); // 0x20000000
-    // "640 MiB Ought to be Enough for Anyone" --Markus F.X.J. Oberhumer, 1981 ;-)
     CHECK(MemBuffer::getSizeForCompression(640 * 1024 * 1024) == 754975232); // 0x28000000
+    // "682 MiB Ought to be Enough for Anyone" --Markus F.X.J. Oberhumer, 2023 ;-)
+    CHECK(MemBuffer::getSizeForCompression(682 * 1024 * 1024) == 804520448); // 0x2aa00000
+    CHECK(MemBuffer::getSizeForCompression(715827428) == UPX_RSIZE_MAX);     // 0x2aaaa8e4
+    CHECK_THROWS(MemBuffer::getSizeForCompression(715827428 + 1));           // 0x2aaaa8e4 + 1
 }
 
 /* vim:set ts=4 sw=4 et: */
diff --git a/src/util/util.cpp b/src/util/util.cpp
index e241a41e..3c737bb8 100644
--- a/src/util/util.cpp
+++ b/src/util/util.cpp
@@ -43,8 +43,7 @@
 // and malicious header fields
 // see C 11 standard, Annex K
 //
-// this limits uncompressed_size to about 640 MiB which is perfectly fine:
-//   "640 MiB Ought to be Enough for Anyone" --Markus F.X.J. Oberhumer, 1981 ;-)
+// this limits uncompressed_size to about 682 MiB (715_128_832 bytes)
 **************************************************************************/
 
 ACC_COMPILE_TIME_ASSERT_HEADER(UPX_RSIZE_MAX_MEM == UPX_RSIZE_MAX)