How to detect X32 ABI or environment in the preprocessor?

Question

X32 is an ABI for amd64/x86_64 CPUs using 32-bit pointers. The idea is to combine the larger register set of x86_64 with the smaller memory and cache footprint resulting from 32-bit pointers. It provides up to about a 40% speedup. See Difference between x86, x32, and x64 architectures on Stack Overflow, and the Debian X32 Ports wiki page for details and setting it up as a chroot environment.

We have a bug report from a Debian maintainer under the environment. The report is adcq is an illegal instruction. The inline assembly is activated based on preprocessor macros, so we are not detecting X32 properly (or more correctly, not at all until now).

The most obvious choice (to me) for a preprocessor macro is something like __X32__, but that's not offered. Based on Clang's patch and Debian's suggestion, it looks like __ILP32__ can be used. But I'd like a more canonical answer since _ILP32 and __code_model_small__ look interesting, too. (And I'm aware of issues with SSE2, where the compiler supported it but the OS did not).

What are the preprocessor macros that can be used to reliably detect an X32 ABI and environment when using Clang and GCC?

To be clear, I'm not trying to fix the code at this point. I just want to know the macros that can be used in a complete remediation.

---

# cpp -dM < /dev/null | sort
#define __amd64 1
#define __amd64__ 1
#define __ATOMIC_ACQ_REL 4
#define __ATOMIC_ACQUIRE 2
#define __ATOMIC_CONSUME 1
#define __ATOMIC_HLE_ACQUIRE 65536
#define __ATOMIC_HLE_RELEASE 131072
#define __ATOMIC_RELAXED 0
#define __ATOMIC_RELEASE 3
#define __ATOMIC_SEQ_CST 5
#define __BIGGEST_ALIGNMENT__ 16
#define __BYTE_ORDER__ __ORDER_LITTLE_ENDIAN__
#define __CHAR16_TYPE__ short unsigned int
#define __CHAR32_TYPE__ unsigned int
#define __CHAR_BIT__ 8
#define __code_model_small__ 1
#define __DBL_DECIMAL_DIG__ 17
#define __DBL_DENORM_MIN__ ((double)4.94065645841246544177e-324L)
#define __DBL_DIG__ 15
#define __DBL_EPSILON__ ((double)2.22044604925031308085e-16L)
#define __DBL_HAS_DENORM__ 1
#define __DBL_HAS_INFINITY__ 1
#define __DBL_HAS_QUIET_NAN__ 1
#define __DBL_MANT_DIG__ 53
#define __DBL_MAX_10_EXP__ 308
#define __DBL_MAX__ ((double)1.79769313486231570815e+308L)
#define __DBL_MAX_EXP__ 1024
#define __DBL_MIN_10_EXP__ (-307)
#define __DBL_MIN__ ((double)2.22507385850720138309e-308L)
#define __DBL_MIN_EXP__ (-1021)
#define __DEC128_EPSILON__ 1E-33DL
#define __DEC128_MANT_DIG__ 34
#define __DEC128_MAX__ 9.999999999999999999999999999999999E6144DL
#define __DEC128_MAX_EXP__ 6145
#define __DEC128_MIN__ 1E-6143DL
#define __DEC128_MIN_EXP__ (-6142)
#define __DEC128_SUBNORMAL_MIN__ 0.000000000000000000000000000000001E-6143DL
#define __DEC32_EPSILON__ 1E-6DF
#define __DEC32_MANT_DIG__ 7
#define __DEC32_MAX__ 9.999999E96DF
#define __DEC32_MAX_EXP__ 97
#define __DEC32_MIN__ 1E-95DF
#define __DEC32_MIN_EXP__ (-94)
#define __DEC32_SUBNORMAL_MIN__ 0.000001E-95DF
#define __DEC64_EPSILON__ 1E-15DD
#define __DEC64_MANT_DIG__ 16
#define __DEC64_MAX__ 9.999999999999999E384DD
#define __DEC64_MAX_EXP__ 385
#define __DEC64_MIN__ 1E-383DD
#define __DEC64_MIN_EXP__ (-382)
#define __DEC64_SUBNORMAL_MIN__ 0.000000000000001E-383DD
#define __DEC_EVAL_METHOD__ 2
#define __DECIMAL_BID_FORMAT__ 1
#define __DECIMAL_DIG__ 21
#define __ELF__ 1
#define __FINITE_MATH_ONLY__ 0
#define __FLOAT_WORD_ORDER__ __ORDER_LITTLE_ENDIAN__
#define __FLT_DECIMAL_DIG__ 9
#define __FLT_DENORM_MIN__ 1.40129846432481707092e-45F
#define __FLT_DIG__ 6
#define __FLT_EPSILON__ 1.19209289550781250000e-7F
#define __FLT_EVAL_METHOD__ 0
#define __FLT_HAS_DENORM__ 1
#define __FLT_HAS_INFINITY__ 1
#define __FLT_HAS_QUIET_NAN__ 1
#define __FLT_MANT_DIG__ 24
#define __FLT_MAX_10_EXP__ 38
#define __FLT_MAX__ 3.40282346638528859812e+38F
#define __FLT_MAX_EXP__ 128
#define __FLT_MIN_10_EXP__ (-37)
#define __FLT_MIN__ 1.17549435082228750797e-38F
#define __FLT_MIN_EXP__ (-125)
#define __FLT_RADIX__ 2
#define __FXSR__ 1
#define __GCC_ATOMIC_BOOL_LOCK_FREE 2
#define __GCC_ATOMIC_CHAR16_T_LOCK_FREE 2
#define __GCC_ATOMIC_CHAR32_T_LOCK_FREE 2
#define __GCC_ATOMIC_CHAR_LOCK_FREE 2
#define __GCC_ATOMIC_INT_LOCK_FREE 2
#define __GCC_ATOMIC_LLONG_LOCK_FREE 2
#define __GCC_ATOMIC_LONG_LOCK_FREE 2
#define __GCC_ATOMIC_POINTER_LOCK_FREE 2
#define __GCC_ATOMIC_SHORT_LOCK_FREE 2
#define __GCC_ATOMIC_TEST_AND_SET_TRUEVAL 1
#define __GCC_ATOMIC_WCHAR_T_LOCK_FREE 2
#define __GCC_HAVE_DWARF2_CFI_ASM 1
#define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_1 1
#define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_2 1
#define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_4 1
#define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_8 1
#define __GCC_IEC_559 2
#define __GCC_IEC_559_COMPLEX 2
#define __GNUC__ 5
#define __GNUC_MINOR__ 2
#define __GNUC_PATCHLEVEL__ 1
#define __GNUC_STDC_INLINE__ 1
#define __gnu_linux__ 1
#define __GXX_ABI_VERSION 1009
#define __has_include_next(STR) __has_include_next__(STR)
#define __has_include(STR) __has_include__(STR)
#define __ILP32__ 1
#define _ILP32 1
#define __INT16_C(c) c
#define __INT16_MAX__ 0x7fff
#define __INT16_TYPE__ short int
#define __INT32_C(c) c
#define __INT32_MAX__ 0x7fffffff
#define __INT32_TYPE__ int
#define __INT64_C(c) c ## LL
#define __INT64_MAX__ 0x7fffffffffffffffLL
#define __INT64_TYPE__ long long int
#define __INT8_C(c) c
#define __INT8_MAX__ 0x7f
#define __INT8_TYPE__ signed char
#define __INT_FAST16_MAX__ 0x7fffffff
#define __INT_FAST16_TYPE__ int
#define __INT_FAST32_MAX__ 0x7fffffff
#define __INT_FAST32_TYPE__ int
#define __INT_FAST64_MAX__ 0x7fffffffffffffffLL
#define __INT_FAST64_TYPE__ long long int
#define __INT_FAST8_MAX__ 0x7f
#define __INT_FAST8_TYPE__ signed char
#define __INT_LEAST16_MAX__ 0x7fff
#define __INT_LEAST16_TYPE__ short int
#define __INT_LEAST32_MAX__ 0x7fffffff
#define __INT_LEAST32_TYPE__ int
#define __INT_LEAST64_MAX__ 0x7fffffffffffffffLL
#define __INT_LEAST64_TYPE__ long long int
#define __INT_LEAST8_MAX__ 0x7f
#define __INT_LEAST8_TYPE__ signed char
#define __INT_MAX__ 0x7fffffff
#define __INTMAX_C(c) c ## LL
#define __INTMAX_MAX__ 0x7fffffffffffffffLL
#define __INTMAX_TYPE__ long long int
#define __INTPTR_MAX__ 0x7fffffff
#define __INTPTR_TYPE__ int
#define __k8 1
#define __k8__ 1
#define __LDBL_DENORM_MIN__ 3.64519953188247460253e-4951L
#define __LDBL_DIG__ 18
#define __LDBL_EPSILON__ 1.08420217248550443401e-19L
#define __LDBL_HAS_DENORM__ 1
#define __LDBL_HAS_INFINITY__ 1
#define __LDBL_HAS_QUIET_NAN__ 1
#define __LDBL_MANT_DIG__ 64
#define __LDBL_MAX_10_EXP__ 4932
#define __LDBL_MAX__ 1.18973149535723176502e+4932L
#define __LDBL_MAX_EXP__ 16384
#define __LDBL_MIN_10_EXP__ (-4931)
#define __LDBL_MIN__ 3.36210314311209350626e-4932L
#define __LDBL_MIN_EXP__ (-16381)
#define __linux 1
#define __linux__ 1
#define linux 1
#define __LONG_LONG_MAX__ 0x7fffffffffffffffLL
#define __LONG_MAX__ 0x7fffffffL
#define __MMX__ 1
#define __NO_INLINE__ 1
#define __ORDER_BIG_ENDIAN__ 4321
#define __ORDER_LITTLE_ENDIAN__ 1234
#define __ORDER_PDP_ENDIAN__ 3412
#define __PRAGMA_REDEFINE_EXTNAME 1
#define __PTRDIFF_MAX__ 0x7fffffff
#define __PTRDIFF_TYPE__ int
#define __REGISTER_PREFIX__ 
#define __SCHAR_MAX__ 0x7f
#define __SHRT_MAX__ 0x7fff
#define __SIG_ATOMIC_MAX__ 0x7fffffff
#define __SIG_ATOMIC_MIN__ (-__SIG_ATOMIC_MAX__ - 1)
#define __SIG_ATOMIC_TYPE__ int
#define __SIZE_MAX__ 0xffffffffU
#define __SIZEOF_DOUBLE__ 8
#define __SIZEOF_FLOAT128__ 16
#define __SIZEOF_FLOAT__ 4
#define __SIZEOF_FLOAT80__ 16
#define __SIZEOF_INT128__ 16
#define __SIZEOF_INT__ 4
#define __SIZEOF_LONG__ 4
#define __SIZEOF_LONG_DOUBLE__ 16
#define __SIZEOF_LONG_LONG__ 8
#define __SIZEOF_POINTER__ 4
#define __SIZEOF_PTRDIFF_T__ 4
#define __SIZEOF_SHORT__ 2
#define __SIZEOF_SIZE_T__ 4
#define __SIZEOF_WCHAR_T__ 4
#define __SIZEOF_WINT_T__ 4
#define __SIZE_TYPE__ unsigned int
#define __SSE__ 1
#define __SSE2__ 1
#define __SSE2_MATH__ 1
#define __SSE_MATH__ 1
#define __STDC__ 1
#define __STDC_HOSTED__ 1
#define __STDC_IEC_559__ 1
#define __STDC_IEC_559_COMPLEX__ 1
#define __STDC_ISO_10646__ 201103L
#define __STDC_NO_THREADS__ 1
#define _STDC_PREDEF_H 1
#define __STDC_UTF_16__ 1
#define __STDC_UTF_32__ 1
#define __STDC_VERSION__ 201112L
#define __UINT16_C(c) c
#define __UINT16_MAX__ 0xffff
#define __UINT16_TYPE__ short unsigned int
#define __UINT32_C(c) c ## U
#define __UINT32_MAX__ 0xffffffffU
#define __UINT32_TYPE__ unsigned int
#define __UINT64_C(c) c ## ULL
#define __UINT64_MAX__ 0xffffffffffffffffULL
#define __UINT64_TYPE__ long long unsigned int
#define __UINT8_C(c) c
#define __UINT8_MAX__ 0xff
#define __UINT8_TYPE__ unsigned char
#define __UINT_FAST16_MAX__ 0xffffffffU
#define __UINT_FAST16_TYPE__ unsigned int
#define __UINT_FAST32_MAX__ 0xffffffffU
#define __UINT_FAST32_TYPE__ unsigned int
#define __UINT_FAST64_MAX__ 0xffffffffffffffffULL
#define __UINT_FAST64_TYPE__ long long unsigned int
#define __UINT_FAST8_MAX__ 0xff
#define __UINT_FAST8_TYPE__ unsigned char
#define __UINT_LEAST16_MAX__ 0xffff
#define __UINT_LEAST16_TYPE__ short unsigned int
#define __UINT_LEAST32_MAX__ 0xffffffffU
#define __UINT_LEAST32_TYPE__ unsigned int
#define __UINT_LEAST64_MAX__ 0xffffffffffffffffULL
#define __UINT_LEAST64_TYPE__ long long unsigned int
#define __UINT_LEAST8_MAX__ 0xff
#define __UINT_LEAST8_TYPE__ unsigned char
#define __UINTMAX_C(c) c ## ULL
#define __UINTMAX_MAX__ 0xffffffffffffffffULL
#define __UINTMAX_TYPE__ long long unsigned int
#define __UINTPTR_MAX__ 0xffffffffU
#define __UINTPTR_TYPE__ unsigned int
#define __unix 1
#define __unix__ 1
#define unix 1
#define __USER_LABEL_PREFIX__ 
#define __VERSION__ "5.2.1 20150911"
#define __WCHAR_MAX__ 0x7fffffffL
#define __WCHAR_MIN__ (-__WCHAR_MAX__ - 1)
#define __WCHAR_TYPE__ long int
#define __WINT_MAX__ 0xffffffffU
#define __WINT_MIN__ 0U
#define __WINT_TYPE__ unsigned int
#define __x86_64 1
#define __x86_64__ 1

Byelostok

Are you quite sure it's adcq that is at fault? x32 allows native 64-bit arithmetic. – Sallie

EOF - Yes, I'm quite sure that's the instruction. – Byelostok

Cannot reproduce. I've written a small program in x32, using inline assembly adcq. No errors. – Sallie

@EOF - You can reproduce it with Crypto++ 5.6.3 Release Candidate available (RC3). The download link is included with the announcement. (But I think the issue has always been there; its finally been tested). But to be clear, I want to stay focused on the question at hand: properly detecting X32 from the preprocessor. (I still remember SSE2, where the compiler supported it but the OS did not). – Byelostok

@EOF - no problem. I added a download link. All you should need to do is unzip -a cryptopp563rc3.zip -d <dir>, then cd <dir>, followed by make. Crypto++ does not have external dependencies, so there's no need for Autotools, Boost, etc. Its cd && make. – Byelostok

I think you've covered it with _IPL32 (or __ILP32__) in conjunction with __x86_64__ . Sounds like the library is not correctly determining the native 'word' size to be [unsigned] long long in this case; otherwise, gcc and clang will define _LP64. You will have to fix the code or submit a bug report. – Kimon

Answer 1

There doesn't appear to be a predefined macro that explicitly specifies an x32 environment. Comparing the output of cpp -dM and cpp -dM -mx32, the symbols _ILP32 and __ILP32__ are defined only for x32, and _LP64 and __LP64__ are defined only for x86_64 without x32. A number of other predefined macros have different values for the two environments.

I think the most straightforward way to detect x32 at compile time is to examine the __x86_64__ and SIZE_MAX macros. The former is predefined (or not) by gcc, and the latter is defined in <stdint.h>.

The following program demonstrates this. It works correctly with gcc -m64, gcc -m32, and gcc -mx32 on an x86_64 system,, and with gcc on a non-x86_64 system (SPARC).

#include <stdio.h>
#include <stdint.h>
int main(void) {
#ifdef __x86_64__
    #if SIZE_MAX == 0xFFFFFFFF
        puts("This is x32");
    #else
        puts("This is x86_64 but not x32");
    #endif
#else
    puts("This is not x64_64");
#endif
}

Maltose

Thanks Keith. I forgot about this question. Clang defines __ILP32__ whenever a 32-bit data model is used (unlike GCC, ICC and Comeau), so it shows up everywhere, like i686. Also, __ILP32__ can be defined on other platforms as needed, like AARCH64/ARM64. So to detect X32, we need to look for both __ILP32__ and one of (__x86_64__ or __amd64__). If you update your answer, I'd be happy to accept. – Byelostok

@jww: __ILP32__ would also work, but I think using SIZE_MAX is clearer. – Maltose

Answer 2

Seems The Debian pages have been updated.
From https://wiki.debian.org/X32Port

When X32 is being targeted, both X32 and x86_64 preprocessor macros are defined. To detect X32, you should use both __ILP32__ and __x86_64__ rather than __amd64__, __x86_64__ or __LP64__.

Regale