[patch] patch for sys-libs/libomp-6.0.9999 for building on x32 (version 3)

libomp-gentoo.patch (text/plain), 50.43 KB, created by Martin 'The Bishop' Scheffler on 2018-02-01 20:44:52 UTC

commit 6cf92498f9e9a06493c78e5a248b5a0f7d3d42c4
Author: Martin 'The Bishop' Scheffler <[e-mail]>
Date:   Thu Feb 1 20:39:22 2018 +0000

    add X32 as distinct architecture

diff --git a/runtime/cmake/LibompMicroTests.cmake b/runtime/cmake/LibompMicroTests.cmake
index 0918fdd..a29001c 100644
--- a/runtime/cmake/LibompMicroTests.cmake
+++ b/runtime/cmake/LibompMicroTests.cmake
@@ -198,6 +198,7 @@ else()
     elseif(${INTEL64})
       libomp_append(libomp_expected_library_deps libc.so.6)
       libomp_append(libomp_expected_library_deps ld-linux-x86-64.so.2)
+      libomp_append(libomp_expected_library_deps ld-linux-x32.so.2)
     elseif(${ARM})
       libomp_append(libomp_expected_library_deps libc.so.6)
       libomp_append(libomp_expected_library_deps libffi.so.6)
diff --git a/runtime/src/kmp.h b/runtime/src/kmp.h
index 0d57bf4..5904503 100644
--- a/runtime/src/kmp.h
+++ b/runtime/src/kmp.h
@@ -93,7 +93,7 @@ class kmp_stats_list;
 #endif
 #endif
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 #include <xmmintrin.h>
 #endif
 
@@ -468,7 +468,7 @@ enum mic_type { non_mic, mic1, mic2, mic3, dummy };
 #define KMP_FAST_REDUCTION_BARRIER 1
 
 #undef KMP_FAST_REDUCTION_CORE_DUO
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 #define KMP_FAST_REDUCTION_CORE_DUO 1
 #endif
 
@@ -721,7 +721,7 @@ enum affinity_top_method {
 #if KMP_ARCH_X86 || KMP_ARCH_X86_64
   affinity_top_method_apicid,
   affinity_top_method_x2apicid,
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
   affinity_top_method_cpuinfo, // KMP_CPUINFO_FILE is usable on Windows* OS, too
 #if KMP_GROUP_AFFINITY
   affinity_top_method_group,
@@ -875,7 +875,7 @@ extern int __kmp_hws_abs_flag; // absolute or per-item number requested
 
 #define KMP_MAX_STKSIZE (~((size_t)1 << ((sizeof(size_t) * (1 << 3)) - 1)))
 
-#if KMP_ARCH_X86
+#if KMP_ARCH_X86 || KMP_ARCH_X86_X32
 #define KMP_DEFAULT_STKSIZE ((size_t)(2 * 1024 * 1024))
 #elif KMP_ARCH_X86_64
 #define KMP_DEFAULT_STKSIZE ((size_t)(4 * 1024 * 1024))
@@ -931,7 +931,7 @@ extern int __kmp_hws_abs_flag; // absolute or per-item number requested
 #else
 #define KMP_BLOCKTIME(team, tid)                                               \
   (get__bt_set(team, tid) ? get__blocktime(team, tid) : __kmp_dflt_blocktime)
-#if KMP_OS_UNIX && (KMP_ARCH_X86 || KMP_ARCH_X86_64)
+#if KMP_OS_UNIX && (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32)
 // HW TSC is used to reduce overhead (clock tick instead of nanosecond).
 extern kmp_uint64 __kmp_ticks_per_msec;
 #if KMP_COMPILER_ICC
@@ -994,7 +994,7 @@ extern kmp_uint64 __kmp_now_nsec();
 /* Minimum number of threads before switch to TLS gtid (experimentally
    determined) */
 /* josh TODO: what about OS X* tuning? */
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 #define KMP_TLS_GTID_MIN 5
 #else
 #define KMP_TLS_GTID_MIN INT_MAX
@@ -1042,7 +1042,7 @@ extern kmp_uint64 __kmp_now_nsec();
 #define KMP_NEXT_WAIT 512U /* susequent number of spin-tests */
 #endif
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 typedef struct kmp_cpuid {
   kmp_uint32 eax;
   kmp_uint32 ebx;
@@ -1187,7 +1187,7 @@ typedef struct kmp_sys_info {
   long nivcsw; /* the number of times a context switch was forced           */
 } kmp_sys_info_t;
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 typedef struct kmp_cpuinfo {
   int initialized; // If 0, other fields are not initialized.
   int signature; // CPUID(1).EAX
@@ -2529,7 +2529,7 @@ typedef int (*launch_t)(int gtid);
 // t_inline_argv. Historically, we have supported at least 96 bytes. Using a
 // larger value for more space between the master write/worker read section and
 // read/write by all section seems to buy more performance on EPCC PARALLEL.
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 #define KMP_INLINE_ARGV_BYTES                                                  \
   (4 * CACHE_LINE -                                                            \
    ((3 * KMP_PTR_SKIP + 2 * sizeof(int) + 2 * sizeof(kmp_int8) +               \
@@ -2580,12 +2580,12 @@ typedef struct KMP_ALIGN_CACHE kmp_base_team {
   ompt_lw_taskteam_t *ompt_serialized_team_info;
 #endif
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
   kmp_int8 t_fp_control_saved;
   kmp_int8 t_pad2b;
   kmp_int16 t_x87_fpu_control_word; // FP control regs
   kmp_uint32 t_mxcsr;
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
   void *t_inline_argv[KMP_INLINE_ARGV_ENTRIES];
 
@@ -2609,7 +2609,7 @@ typedef struct KMP_ALIGN_CACHE kmp_base_team {
 // omp_set_num_threads() call
 
 // Read/write by workers as well
-#if (KMP_ARCH_X86 || KMP_ARCH_X86_64)
+#if (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32)
   // Using CACHE_LINE=64 reduces memory footprint, but causes a big perf
   // regression of epcc 'parallel' and 'barrier' on fxe256lin01. This extra
   // padding serves to fix the performance of epcc 'parallel' and 'barrier' when
@@ -2751,7 +2751,7 @@ extern int __kmp_storage_map_verbose; /* True means storage map includes
                                          placement info */
 extern int __kmp_storage_map_verbose_specified;
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 extern kmp_cpuinfo_t __kmp_cpuinfo;
 #endif
 
@@ -2905,11 +2905,11 @@ extern KMP_THREAD_LOCAL int __kmp_gtid;
 #endif
 extern int __kmp_tls_gtid_min; /* #threads below which use sp search for gtid */
 extern int __kmp_foreign_tp; // If true, separate TP var for each foreign thread
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 extern int __kmp_inherit_fp_control; // copy fp creg(s) parent->workers at fork
 extern kmp_int16 __kmp_init_x87_fpu_control_word; // init thread's FP ctrl reg
 extern kmp_uint32 __kmp_init_mxcsr; /* init thread's mxscr */
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
 extern int __kmp_dflt_max_active_levels; /* max_active_levels for nested
                                             parallelism enabled by default via
@@ -3222,7 +3222,7 @@ extern void __kmp_check_stack_overlap(kmp_info_t *thr);
 extern void __kmp_expand_host_name(char *buffer, size_t size);
 extern void __kmp_expand_file_name(char *result, size_t rlen, char *pattern);
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 extern void
 __kmp_initialize_system_tick(void); /* Initialize timer tick value */
 #endif
@@ -3371,7 +3371,7 @@ extern int __kmp_fork_call(ident_t *loc, int gtid,
                            enum fork_context_e fork_context, kmp_int32 argc,
                            microtask_t microtask, launch_t invoker,
 /* TODO: revert workaround for Intel(R) 64 tracker #96 */
-#if (KMP_ARCH_ARM || KMP_ARCH_X86_64 || KMP_ARCH_AARCH64) && KMP_OS_LINUX
+#if (KMP_ARCH_ARM || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 || KMP_ARCH_AARCH64) && KMP_OS_LINUX
                            va_list *ap
 #else
                            va_list ap
@@ -3483,7 +3483,7 @@ extern int __kmp_read_from_file(char const *path, char const *format, ...);
 // Assembly routines that have no compiler intrinsic replacement
 //
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 
 extern void __kmp_query_cpuid(kmp_cpuinfo_t *p);
 
@@ -3495,7 +3495,7 @@ extern void __kmp_store_x87_fpu_control_word(kmp_int16 *p);
 extern void __kmp_clear_x87_fpu_status_word();
 #define KMP_X86_MXCSR_MASK 0xffffffc0 /* ignore status flags (6 lsb) */
 
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
 extern int __kmp_invoke_microtask(microtask_t pkfn, int gtid, int npr, int argc,
                                   void *argv[]
diff --git a/runtime/src/kmp_affinity.cpp b/runtime/src/kmp_affinity.cpp
index 9cc504b..607e994 100644
--- a/runtime/src/kmp_affinity.cpp
+++ b/runtime/src/kmp_affinity.cpp
@@ -934,7 +934,7 @@ static int __kmp_affinity_create_proc_group_map(AddrUnsPair **address2os,
 
 #endif /* KMP_GROUP_AFFINITY */
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 
 static int __kmp_cpuid_mask_width(int count) {
   int r = 0;
@@ -1858,7 +1858,7 @@ static int __kmp_affinity_create_x2apicid_map(AddrUnsPair **address2os,
   return depth;
 }
 
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
 #define osIdIndex 0
 #define threadIdIndex 1
@@ -2038,7 +2038,7 @@ static int __kmp_affinity_create_cpuinfo_map(AddrUnsPair **address2os,
           goto dup_field;
 #endif
         threadInfo[num_avail][osIdIndex] = val;
-#if KMP_OS_LINUX && !(KMP_ARCH_X86 || KMP_ARCH_X86_64)
+#if KMP_OS_LINUX && !(KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32)
         char path[256];
         KMP_SNPRINTF(
             path, sizeof(path),
@@ -4068,7 +4068,7 @@ static void __kmp_aux_affinity_initialize(void) {
     }
 #endif
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 
     if (depth < 0) {
       if (__kmp_affinity_verbose) {
@@ -4101,7 +4101,7 @@ static void __kmp_aux_affinity_initialize(void) {
       }
     }
 
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
 #if KMP_OS_LINUX
 
@@ -4184,7 +4184,7 @@ static void __kmp_aux_affinity_initialize(void) {
 // used, then we abort if that method fails. The exception is group affinity,
 // which might have been implicitly set.
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 
   else if (__kmp_affinity_top_method == affinity_top_method_x2apicid) {
     if (__kmp_affinity_verbose) {
@@ -4214,7 +4214,7 @@ static void __kmp_aux_affinity_initialize(void) {
     }
   }
 
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
   else if (__kmp_affinity_top_method == affinity_top_method_cpuinfo) {
     const char *filename;
diff --git a/runtime/src/kmp_affinity.h b/runtime/src/kmp_affinity.h
index e890165..d70ed5e 100644
--- a/runtime/src/kmp_affinity.h
+++ b/runtime/src/kmp_affinity.h
@@ -201,6 +201,18 @@ public:
 #elif __NR_sched_getaffinity != 204
 #error Wrong code for getaffinity system call.
 #endif /* __NR_sched_getaffinity */
+#elif KMP_ARCH_X86_X32
+#define __X32_SYSCALL_BIT 0x40000000
+#ifndef __NR_sched_setaffinity
+#define __NR_sched_setaffinity  (__X32_SYSCALL_BIT + 203)
+#elif __NR_sched_setaffinity != (__X32_SYSCALL_BIT + 203)
+#error Wrong code for setaffinity system call.
+#endif /* __NR_sched_setaffinity */
+#ifndef __NR_sched_getaffinity
+#define __NR_sched_getaffinity  (__X32_SYSCALL_BIT + 204)
+#elif __NR_sched_getaffinity != (__X32_SYSCALL_BIT + 204)
+#error Wrong code for getaffinity system call.
+#endif /* __NR_sched_getaffinity */
 #elif KMP_ARCH_PPC64
 #ifndef __NR_sched_setaffinity
 #define __NR_sched_setaffinity 222
diff --git a/runtime/src/kmp_atomic.cpp b/runtime/src/kmp_atomic.cpp
index b099eb6..37bfce9 100644
--- a/runtime/src/kmp_atomic.cpp
+++ b/runtime/src/kmp_atomic.cpp
@@ -816,7 +816,7 @@ static inline void operator/=(kmp_cmplx128_a16_t &lhs,
 // end of the first part of the workaround for C78287
 #endif // USE_CMPXCHG_FIX
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 
 // ------------------------------------------------------------------------
 // X86 or X86_64: no alignment problems ====================================
@@ -889,7 +889,7 @@ static inline void operator/=(kmp_cmplx128_a16_t &lhs,
   }
 // end of the second part of the workaround for C78287
 #endif // USE_CMPXCHG_FIX
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
 // Routines for ATOMIC 4-byte operands addition and subtraction
 ATOMIC_FIXED_ADD(fixed4, add, kmp_int32, 32, +, 4i, 3,
@@ -1030,7 +1030,7 @@ ATOMIC_CMPXCHG(float8, mul, kmp_real64, 64, *, 8r, 7,
   OP_CRITICAL(= *lhs OP, LCK_ID)                                               \
   }
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 
 // ------------------------------------------------------------------------
 // X86 or X86_64: no alignment problems ===================================
@@ -1053,7 +1053,7 @@ ATOMIC_CMPXCHG(float8, mul, kmp_real64, 64, *, 8r, 7,
     OP_CRITICAL(= *lhs OP, LCK_ID) /* unaligned - use critical */              \
   }                                                                            \
   }
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
 ATOMIC_CMPX_L(fixed1, andl, char, 8, &&, 1i, 0,
               KMP_ARCH_X86) // __kmpc_atomic_fixed1_andl
@@ -1129,7 +1129,7 @@ ATOMIC_CMPX_L(fixed8, orl, kmp_int64, 64, ||, 8i, 7,
   }                                                                            \
   }
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 
 // -------------------------------------------------------------------------
 // X86 or X86_64: no alignment problems ====================================
@@ -1158,7 +1158,7 @@ ATOMIC_CMPX_L(fixed8, orl, kmp_int64, 64, ||, 8i, 7,
     }                                                                          \
   }                                                                            \
   }
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
 MIN_MAX_COMPXCHG(fixed1, max, char, 8, <, 1i, 0,
                  KMP_ARCH_X86) // __kmpc_atomic_fixed1_max
@@ -1206,7 +1206,7 @@ MIN_MAX_CRITICAL(float16, min_a16, Quad_a16_t, >, 16r,
   }
 
 // ------------------------------------------------------------------------
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 // ------------------------------------------------------------------------
 // X86 or X86_64: no alignment problems ===================================
 #define ATOMIC_CMPX_EQV(TYPE_ID, OP_ID, TYPE, BITS, OP, LCK_ID, MASK,          \
@@ -1230,7 +1230,7 @@ MIN_MAX_CRITICAL(float16, min_a16, Quad_a16_t, >, 16r,
     OP_CRITICAL(^= ~, LCK_ID) /* unaligned address - use critical */           \
   }                                                                            \
   }
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
 ATOMIC_CMPXCHG(fixed1, neqv, kmp_int8, 8, ^, 1i, 0,
                KMP_ARCH_X86) // __kmpc_atomic_fixed1_neqv
@@ -1349,7 +1349,7 @@ ATOMIC_CRITICAL(cmplx16, div_a16, kmp_cmplx128_a16_t, /, 32c,
 
 // OpenMP 4.0: x = expr binop x for non-commutative operations.
 // Supported only on IA-32 architecture and Intel(R) 64
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 
 // ------------------------------------------------------------------------
 // Operation on *lhs, rhs bound by critical section
@@ -1553,7 +1553,7 @@ ATOMIC_CRITICAL_REV(cmplx16, div_a16, kmp_cmplx128_a16_t, /, 32c,
 #endif
 #endif
 
-#endif // KMP_ARCH_X86 || KMP_ARCH_X86_64
+#endif // KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 // End of OpenMP 4.0: x = expr binop x for non-commutative operations.
 
 #endif // OMP_40_ENABLED
@@ -1586,7 +1586,7 @@ ATOMIC_CRITICAL_REV(cmplx16, div_a16, kmp_cmplx128_a16_t, /, 32c,
   }
 
 // -------------------------------------------------------------------------
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 // -------------------------------------------------------------------------
 // X86 or X86_64: no alignment problems ====================================
 #define ATOMIC_CMPXCHG_MIX(TYPE_ID, TYPE, OP_ID, BITS, OP, RTYPE_ID, RTYPE,    \
@@ -1610,10 +1610,10 @@ ATOMIC_CRITICAL_REV(cmplx16, div_a16, kmp_cmplx128_a16_t, /, 32c,
     OP_CRITICAL(OP## =, LCK_ID) /* unaligned address - use critical */         \
   }                                                                            \
   }
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
 // -------------------------------------------------------------------------
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 // -------------------------------------------------------------------------
 #define ATOMIC_CMPXCHG_REV_MIX(TYPE_ID, TYPE, OP_ID, BITS, OP, RTYPE_ID,       \
                                RTYPE, LCK_ID, MASK, GOMP_FLAG)                 \
@@ -1627,7 +1627,7 @@ ATOMIC_CRITICAL_REV(cmplx16, div_a16, kmp_cmplx128_a16_t, /, 32c,
   OP_GOMP_CRITICAL_REV(OP, GOMP_FLAG)                                          \
   OP_CRITICAL_REV(OP, LCK_ID)                                                  \
   }
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
 // RHS=float8
 ATOMIC_CMPXCHG_MIX(fixed1, char, mul, 8, *, float8, kmp_real64, 1i, 0,
@@ -1753,7 +1753,7 @@ ATOMIC_CRITICAL_FP(float10, long double, mul, *, fp, _Quad, 10r,
 ATOMIC_CRITICAL_FP(float10, long double, div, /, fp, _Quad, 10r,
                    1) // __kmpc_atomic_float10_div_fp
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 // Reverse operations
 ATOMIC_CMPXCHG_REV_MIX(fixed1, char, sub_rev, 8, -, fp, _Quad, 1i, 0,
                        KMP_ARCH_X86) // __kmpc_atomic_fixed1_sub_rev_fp
@@ -1805,11 +1805,11 @@ ATOMIC_CRITICAL_REV_FP(float10, long double, sub_rev, -, fp, _Quad, 10r,
                        1) // __kmpc_atomic_float10_sub_rev_fp
 ATOMIC_CRITICAL_REV_FP(float10, long double, div_rev, /, fp, _Quad, 10r,
                        1) // __kmpc_atomic_float10_div_rev_fp
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
 #endif
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 // ------------------------------------------------------------------------
 // X86 or X86_64: no alignment problems ====================================
 #if USE_CMPXCHG_FIX
@@ -1843,7 +1843,7 @@ ATOMIC_CRITICAL_REV_FP(float10, long double, div_rev, /, fp, _Quad, 10r,
     OP_CRITICAL(OP## =, LCK_ID) /* unaligned address - use critical */         \
   }                                                                            \
   }
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
 ATOMIC_CMPXCHG_CMPLX(cmplx4, kmp_cmplx32, add, 64, +, cmplx8, kmp_cmplx64, 8c,
                      7, KMP_ARCH_X86) // __kmpc_atomic_cmplx4_add_cmplx8
@@ -1855,7 +1855,7 @@ ATOMIC_CMPXCHG_CMPLX(cmplx4, kmp_cmplx32, div, 64, /, cmplx8, kmp_cmplx64, 8c,
                      7, KMP_ARCH_X86) // __kmpc_atomic_cmplx4_div_cmplx8
 
 // READ, WRITE, CAPTURE are supported only on IA-32 architecture and Intel(R) 64
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 
 // ------------------------------------------------------------------------
 // Atomic READ routines
@@ -3326,7 +3326,7 @@ ATOMIC_CRITICAL_SWP(cmplx16_a16, kmp_cmplx128_a16_t, 32c,
 
 #endif // OMP_40_ENABLED
 
-#endif // KMP_ARCH_X86 || KMP_ARCH_X86_64
+#endif // KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 
 #undef OP_CRITICAL
 
@@ -3385,7 +3385,7 @@ void __kmpc_atomic_2(ident_t *id_ref, int gtid, void *lhs, void *rhs,
   if (
 #if KMP_ARCH_X86 && defined(KMP_GOMP_COMPAT)
       FALSE /* must use lock */
-#elif KMP_ARCH_X86 || KMP_ARCH_X86_64
+#elif KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
       TRUE /* no alignment problems */
 #else
       !((kmp_uintptr_t)lhs & 0x1) /* make sure address is 2-byte aligned */
@@ -3434,7 +3434,7 @@ void __kmpc_atomic_4(ident_t *id_ref, int gtid, void *lhs, void *rhs,
   if (
 // FIXME: On IA-32 architecture, gcc uses cmpxchg only for 4-byte ints.
 // Gomp compatibility is broken if this routine is called for floats.
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
       TRUE /* no alignment problems */
 #else
       !((kmp_uintptr_t)lhs & 0x3) /* make sure address is 4-byte aligned */
@@ -3484,7 +3484,7 @@ void __kmpc_atomic_8(ident_t *id_ref, int gtid, void *lhs, void *rhs,
 
 #if KMP_ARCH_X86 && defined(KMP_GOMP_COMPAT)
       FALSE /* must use lock */
-#elif KMP_ARCH_X86 || KMP_ARCH_X86_64
+#elif KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
       TRUE /* no alignment problems */
 #else
       !((kmp_uintptr_t)lhs & 0x7) /* make sure address is 8-byte aligned */
diff --git a/runtime/src/kmp_atomic.h b/runtime/src/kmp_atomic.h
index 77df975..95ad070 100644
--- a/runtime/src/kmp_atomic.h
+++ b/runtime/src/kmp_atomic.h
@@ -692,7 +692,7 @@ void __kmpc_atomic_cmplx16_div_a16(ident_t *id_ref, int gtid,
 
 // OpenMP 4.0: x = expr binop x for non-commutative operations.
 // Supported only on IA-32 architecture and Intel(R) 64
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 
 void __kmpc_atomic_fixed1_sub_rev(ident_t *id_ref, int gtid, char *lhs,
                                   char rhs);
@@ -792,7 +792,7 @@ void __kmpc_atomic_cmplx16_div_a16_rev(ident_t *id_ref, int gtid,
 #endif
 #endif // KMP_HAVE_QUAD
 
-#endif // KMP_ARCH_X86 || KMP_ARCH_X86_64
+#endif // KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 
 #endif // OMP_40_ENABLED
 
@@ -999,7 +999,7 @@ void __kmpc_atomic_32(ident_t *id_ref, int gtid, void *lhs, void *rhs,
                       void (*f)(void *, void *, void *));
 
 // READ, WRITE, CAPTURE are supported only on IA-32 architecture and Intel(R) 64
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 
 //  Below routines for atomic READ are listed
 char __kmpc_atomic_fixed1_rd(ident_t *id_ref, int gtid, char *loc);
@@ -1763,7 +1763,7 @@ long double __kmpc_atomic_float10_div_cpt_rev_fp(ident_t *id_ref, int gtid,
 
 #endif // OMP_40_ENABLED
 
-#endif // KMP_ARCH_X86 || KMP_ARCH_X86_64
+#endif // KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 
 /* ------------------------------------------------------------------------ */
 
diff --git a/runtime/src/kmp_csupport.cpp b/runtime/src/kmp_csupport.cpp
index 815fe23..b0b1b24 100644
--- a/runtime/src/kmp_csupport.cpp
+++ b/runtime/src/kmp_csupport.cpp
@@ -300,7 +300,7 @@ void __kmpc_fork_call(ident_t *loc, kmp_int32 argc, kmpc_micro microtask, ...) {
                     VOLATILE_CAST(microtask_t) microtask, // "wrapped" task
                     VOLATILE_CAST(launch_t) __kmp_invoke_task_func,
 /* TODO: revert workaround for Intel(R) 64 tracker #96 */
-#if (KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64) && KMP_OS_LINUX
+#if (KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 || KMP_ARCH_ARM || KMP_ARCH_AARCH64) && KMP_OS_LINUX
                     &ap
 #else
                     ap
@@ -388,7 +388,7 @@ void __kmpc_fork_teams(ident_t *loc, kmp_int32 argc, kmpc_micro microtask,
                   VOLATILE_CAST(microtask_t)
                       __kmp_teams_master, // "wrapped" task
                   VOLATILE_CAST(launch_t) __kmp_invoke_teams_master,
-#if (KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64) && KMP_OS_LINUX
+#if (KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 || KMP_ARCH_ARM || KMP_ARCH_AARCH64) && KMP_OS_LINUX
                   &ap
 #else
                   ap
@@ -530,13 +530,13 @@ void __kmpc_end_serialized_parallel(ident_t *loc, kmp_int32 global_tid) {
 
 /* return to the parallel section */
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
     if (__kmp_inherit_fp_control && serial_team->t.t_fp_control_saved) {
       __kmp_clear_x87_fpu_status_word();
       __kmp_load_x87_fpu_control_word(&serial_team->t.t_x87_fpu_control_word);
       __kmp_load_mxcsr(&serial_team->t.t_mxcsr);
     }
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
     this_thr->th.th_team = serial_team->t.t_parent;
     this_thr->th.th_info.ds.ds_tid = serial_team->t.t_master_tid;
@@ -597,7 +597,7 @@ void __kmpc_flush(ident_t *loc) {
   /* need explicit __mf() here since use volatile instead in library */
   KMP_MB(); /* Flush all pending memory write invalidates.  */
 
-#if (KMP_ARCH_X86 || KMP_ARCH_X86_64)
+#if (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32)
 #if KMP_MIC
 // fence-style instructions do not exist, but lock; xaddl $0,(%rsp) can be used.
 // We shouldn't need it, though, since the ABI rules require that
@@ -1210,7 +1210,7 @@ static __forceinline kmp_dyna_lockseq_t __kmp_map_hint_to_lock(uintptr_t hint) {
 #define KMP_TSX_LOCK(seq) __kmp_user_lock_seq
 #endif
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 #define KMP_CPUINFO_RTM (__kmp_cpuinfo.rtm)
 #else
 #define KMP_CPUINFO_RTM 0
@@ -2678,7 +2678,7 @@ void __kmpc_unset_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
   if ((__kmp_user_lock_kind == lk_tas) &&
       (sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
 #if KMP_OS_LINUX &&                                                            \
-    (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64)
+    (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 || KMP_ARCH_ARM || KMP_ARCH_AARCH64)
 // "fast" path implemented to fix customer performance issue
 #if USE_ITT_BUILD
     __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
@@ -2772,7 +2772,7 @@ void __kmpc_unset_nest_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
       (sizeof(lck->tas.lk.poll) + sizeof(lck->tas.lk.depth_locked) <=
        OMP_NEST_LOCK_T_SIZE)) {
 #if KMP_OS_LINUX &&                                                            \
-    (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64)
+    (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 || KMP_ARCH_ARM || KMP_ARCH_AARCH64)
     // "fast" path implemented to fix customer performance issue
     kmp_tas_lock_t *tl = (kmp_tas_lock_t *)user_lock;
 #if USE_ITT_BUILD
diff --git a/runtime/src/kmp_global.cpp b/runtime/src/kmp_global.cpp
index ab14ddb..3496888 100644
--- a/runtime/src/kmp_global.cpp
+++ b/runtime/src/kmp_global.cpp
@@ -16,7 +16,7 @@
 
 kmp_key_t __kmp_gtid_threadprivate_key;
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 kmp_cpuinfo_t __kmp_cpuinfo = {0}; // Not initialized
 #endif
 
@@ -177,11 +177,11 @@ KMP_THREAD_LOCAL int __kmp_gtid = KMP_GTID_DNE;
 #endif /* KMP_TDATA_GTID */
 int __kmp_tls_gtid_min = INT_MAX;
 int __kmp_foreign_tp = TRUE;
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 int __kmp_inherit_fp_control = TRUE;
 kmp_int16 __kmp_init_x87_fpu_control_word = 0;
 kmp_uint32 __kmp_init_mxcsr = 0;
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
 #ifdef USE_LOAD_BALANCE
 double __kmp_load_balance_interval = 1.0;
diff --git a/runtime/src/kmp_gsupport.cpp b/runtime/src/kmp_gsupport.cpp
index 9fa05e6..0eaa454 100644
--- a/runtime/src/kmp_gsupport.cpp
+++ b/runtime/src/kmp_gsupport.cpp
@@ -376,7 +376,7 @@ static
 
   rc = __kmp_fork_call(loc, gtid, fork_context_gnu, argc, wrapper,
                        __kmp_invoke_task_func,
-#if (KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64) && KMP_OS_LINUX
+#if (KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 || KMP_ARCH_ARM || KMP_ARCH_AARCH64) && KMP_OS_LINUX
                        &ap
 #else
                        ap
diff --git a/runtime/src/kmp_itt.h b/runtime/src/kmp_itt.h
index e961194..02af57d 100644
--- a/runtime/src/kmp_itt.h
+++ b/runtime/src/kmp_itt.h
@@ -155,11 +155,11 @@ __kmp_inline void __kmp_itt_stack_callee_leave(__itt_caller);
    therefore uninteresting when collecting traces for architecture simulation.
  */
 #ifndef INCLUDE_SSC_MARKS
-#define INCLUDE_SSC_MARKS (KMP_OS_LINUX && KMP_ARCH_X86_64)
+#define INCLUDE_SSC_MARKS (KMP_OS_LINUX && (KMP_ARCH_X86_64 || KMP_ARCH_X86_X32))
 #endif
 
-/* Linux 64 only for now */
-#if (INCLUDE_SSC_MARKS && KMP_OS_LINUX && KMP_ARCH_X86_64)
+/* Linux 64 and Linux 64/x32 only for now */
+#if (INCLUDE_SSC_MARKS && KMP_OS_LINUX && (KMP_ARCH_X86_64 || KMP_ARCH_X86_X32))
 // Portable (at least for gcc and icc) code to insert the necessary instructions
 // to set %ebx and execute the unlikely no-op.
 #if defined(__INTEL_COMPILER)
diff --git a/runtime/src/kmp_lock.cpp b/runtime/src/kmp_lock.cpp
index fb9edb1..ac5b7da 100644
--- a/runtime/src/kmp_lock.cpp
+++ b/runtime/src/kmp_lock.cpp
@@ -2736,7 +2736,7 @@ static void __kmp_set_drdpa_lock_flags(kmp_drdpa_lock_t *lck,
 }
 
 // Time stamp counter
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 #define __kmp_tsc() __kmp_hardware_timestamp()
 // Runtime's default backoff parameters
 kmp_backoff_t __kmp_spin_backoff_params = {1, 4096, 100};
diff --git a/runtime/src/kmp_lock.h b/runtime/src/kmp_lock.h
index 76f1968..28edfc5 100644
--- a/runtime/src/kmp_lock.h
+++ b/runtime/src/kmp_lock.h
@@ -161,7 +161,7 @@ extern void __kmp_destroy_nested_tas_lock(kmp_tas_lock_t *lck);
 
 #define KMP_USE_FUTEX                                                          \
   (KMP_OS_LINUX && !KMP_OS_CNK &&                                              \
-   (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64))
+   (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 || KMP_ARCH_ARM || KMP_ARCH_AARCH64))
 
 #if KMP_USE_FUTEX
 
@@ -630,7 +630,7 @@ extern int (*__kmp_acquire_user_lock_with_checks_)(kmp_user_lock_p lck,
                                                    kmp_int32 gtid);
 
 #if KMP_OS_LINUX &&                                                            \
-    (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64)
+    (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 || KMP_ARCH_ARM || KMP_ARCH_AARCH64)
 
 #define __kmp_acquire_user_lock_with_checks(lck, gtid)                         \
   if (__kmp_user_lock_kind == lk_tas) {                                        \
@@ -684,7 +684,7 @@ extern int (*__kmp_test_user_lock_with_checks_)(kmp_user_lock_p lck,
                                                 kmp_int32 gtid);
 
 #if KMP_OS_LINUX &&                                                            \
-    (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64)
+    (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 || KMP_ARCH_ARM || KMP_ARCH_AARCH64)
 
 #include "kmp_i18n.h" /* AC: KMP_FATAL definition */
 extern int __kmp_env_consistency_check; /* AC: copy from kmp.h here */
@@ -748,7 +748,7 @@ static inline void __kmp_destroy_user_lock_with_checks(kmp_user_lock_p lck) {
 extern int (*__kmp_acquire_nested_user_lock_with_checks_)(kmp_user_lock_p lck,
                                                           kmp_int32 gtid);
 
-#if KMP_OS_LINUX && (KMP_ARCH_X86 || KMP_ARCH_X86_64)
+#if KMP_OS_LINUX && (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32)
 
 #define __kmp_acquire_nested_user_lock_with_checks(lck, gtid, depth)           \
   if (__kmp_user_lock_kind == lk_tas) {                                        \
@@ -806,7 +806,7 @@ __kmp_acquire_nested_user_lock_with_checks(kmp_user_lock_p lck, kmp_int32 gtid,
 extern int (*__kmp_test_nested_user_lock_with_checks_)(kmp_user_lock_p lck,
                                                        kmp_int32 gtid);
 
-#if KMP_OS_LINUX && (KMP_ARCH_X86 || KMP_ARCH_X86_64)
+#if KMP_OS_LINUX && (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32)
 static inline int __kmp_test_nested_user_lock_with_checks(kmp_user_lock_p lck,
                                                           kmp_int32 gtid) {
   if (__kmp_user_lock_kind == lk_tas) {
@@ -1052,7 +1052,7 @@ extern void __kmp_cleanup_user_locks();
 
 // Shortcuts
 #define KMP_USE_INLINED_TAS                                                    \
-  (KMP_OS_LINUX && (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_ARM)) && 1
+  (KMP_OS_LINUX && (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 || KMP_ARCH_ARM)) && 1
 #define KMP_USE_INLINED_FUTEX KMP_USE_FUTEX && 0
 
 // List of lock definitions; all nested locks are indirect locks.
diff --git a/runtime/src/kmp_os.h b/runtime/src/kmp_os.h
index fbd3513..7224619 100644
--- a/runtime/src/kmp_os.h
+++ b/runtime/src/kmp_os.h
@@ -153,7 +153,7 @@ typedef unsigned long long kmp_uint64;
 #error "Can't determine size_t printf format specifier."
 #endif
 
-#if KMP_ARCH_X86
+#if KMP_ARCH_X86 || ( KMP_ARCH_X86_64 && __ILP32__ )
 #define KMP_SIZE_T_MAX (0xFFFFFFFF)
 #else
 #define KMP_SIZE_T_MAX (0xFFFFFFFFFFFFFFFF)
diff --git a/runtime/src/kmp_platform.h b/runtime/src/kmp_platform.h
index 534409e..fe7b764 100644
--- a/runtime/src/kmp_platform.h
+++ b/runtime/src/kmp_platform.h
@@ -74,6 +74,7 @@
 
 #define KMP_ARCH_X86 0
 #define KMP_ARCH_X86_64 0
+#define KMP_ARCH_X86_X32 0
 #define KMP_ARCH_AARCH64 0
 #define KMP_ARCH_PPC64_BE 0
 #define KMP_ARCH_PPC64_LE 0
@@ -93,8 +94,13 @@
 
 #if KMP_OS_UNIX
 #if defined __x86_64
+#if defined __ILP32__
+#undef KMP_ARCH_X86_X32
+#define KMP_ARCH_X86_X32 1
+#else
 #undef KMP_ARCH_X86_64
 #define KMP_ARCH_X86_64 1
+#endif
 #elif defined __i386
 #undef KMP_ARCH_X86
 #define KMP_ARCH_X86 1
@@ -176,12 +182,12 @@
 
 // Platforms which support Intel(R) Many Integrated Core Architecture
 #define KMP_MIC_SUPPORTED                                                      \
-  ((KMP_ARCH_X86 || KMP_ARCH_X86_64) && (KMP_OS_LINUX || KMP_OS_WINDOWS))
+  ((KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32) && (KMP_OS_LINUX || KMP_OS_WINDOWS))
 
 // TODO: Fixme - This is clever, but really fugly
 #if (1 !=                                                                      \
-     KMP_ARCH_X86 + KMP_ARCH_X86_64 + KMP_ARCH_ARM + KMP_ARCH_PPC64 +          \
-         KMP_ARCH_AARCH64 + KMP_ARCH_MIPS + KMP_ARCH_MIPS64)
+     KMP_ARCH_X86 + KMP_ARCH_X86_64 + KMP_ARCH_X86_X32 + KMP_ARCH_ARM +	   \
+     	 KMP_ARCH_PPC64 + KMP_ARCH_AARCH64 + KMP_ARCH_MIPS + KMP_ARCH_MIPS64)
 #error Unknown or unsupported architecture
 #endif
 
diff --git a/runtime/src/kmp_runtime.cpp b/runtime/src/kmp_runtime.cpp
index cd2d421..add025f 100644
--- a/runtime/src/kmp_runtime.cpp
+++ b/runtime/src/kmp_runtime.cpp
@@ -1081,7 +1081,7 @@ static void __kmp_fork_team_threads(kmp_root_t *root, kmp_team_t *team,
   KMP_MB();
 }
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 // Propagate any changes to the floating point control registers out to the team
 // We try to avoid unnecessary writes to the relevant cache line in the team
 // structure, so we don't make changes unless they are needed.
@@ -1141,7 +1141,7 @@ inline static void updateHWFPControl(kmp_team_t *team) {
 #else
 #define propagateFPControl(x) ((void)0)
 #define updateHWFPControl(x) ((void)0)
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
 static void __kmp_alloc_argv_entries(int argc, kmp_team_t *team,
                                      int realloc); // forward declaration
@@ -1401,7 +1401,7 @@ int __kmp_fork_call(ident_t *loc, int gtid,
                     enum fork_context_e call_context, // Intel, GNU, ...
                     kmp_int32 argc, microtask_t microtask, launch_t invoker,
 /* TODO: revert workaround for Intel(R) 64 tracker #96 */
-#if (KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64) && KMP_OS_LINUX
+#if (KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 || KMP_ARCH_ARM || KMP_ARCH_AARCH64) && KMP_OS_LINUX
                     va_list *ap
 #else
                     va_list ap
@@ -1518,7 +1518,7 @@ int __kmp_fork_call(ident_t *loc, int gtid,
       argv = (void **)parent_team->t.t_argv;
       for (i = argc - 1; i >= 0; --i)
 /* TODO: revert workaround for Intel(R) 64 tracker #96 */
-#if (KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64) && KMP_OS_LINUX
+#if (KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 || KMP_ARCH_ARM || KMP_ARCH_AARCH64) && KMP_OS_LINUX
         *argv++ = va_arg(*ap, void *);
 #else
         *argv++ = va_arg(ap, void *);
@@ -1721,11 +1721,11 @@ int __kmp_fork_call(ident_t *loc, int gtid,
     if (nthreads == 1) {
 /* josh todo: hypothetical question: what do we do for OS X*? */
 #if KMP_OS_LINUX &&                                                            \
-    (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64)
+    (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 || KMP_ARCH_ARM || KMP_ARCH_AARCH64)
       void *args[argc];
 #else
       void **args = (void **)KMP_ALLOCA(argc * sizeof(void *));
-#endif /* KMP_OS_LINUX && ( KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_ARM || \
+#endif /* KMP_OS_LINUX && ( KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 || KMP_ARCH_ARM || \
           KMP_ARCH_AARCH64) */
 
       KA_TRACE(20,
@@ -1813,7 +1813,7 @@ int __kmp_fork_call(ident_t *loc, int gtid,
           if (ap) {
             for (i = argc - 1; i >= 0; --i)
 // TODO: revert workaround for Intel(R) 64 tracker #96
-#if (KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64) && KMP_OS_LINUX
+#if (KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 || KMP_ARCH_ARM || KMP_ARCH_AARCH64) && KMP_OS_LINUX
               *argv++ = va_arg(*ap, void *);
 #else
               *argv++ = va_arg(ap, void *);
@@ -1837,7 +1837,7 @@ int __kmp_fork_call(ident_t *loc, int gtid,
           argv = args;
           for (i = argc - 1; i >= 0; --i)
 // TODO: revert workaround for Intel(R) 64 tracker #96
-#if (KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64) && KMP_OS_LINUX
+#if (KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 || KMP_ARCH_ARM || KMP_ARCH_AARCH64) && KMP_OS_LINUX
             *argv++ = va_arg(*ap, void *);
 #else
           *argv++ = va_arg(ap, void *);
@@ -2145,7 +2145,7 @@ int __kmp_fork_call(ident_t *loc, int gtid,
 #endif /* OMP_40_ENABLED */
       for (i = argc - 1; i >= 0; --i) {
 // TODO: revert workaround for Intel(R) 64 tracker #96
-#if (KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64) && KMP_OS_LINUX
+#if (KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 || KMP_ARCH_ARM || KMP_ARCH_AARCH64) && KMP_OS_LINUX
         void *new_argv = va_arg(*ap, void *);
 #else
       void *new_argv = va_arg(ap, void *);
@@ -4470,11 +4470,11 @@ static void __kmp_initialize_team(kmp_team_t *team, int new_nproc,
   // TODO???: team->t.t_max_active_levels       = new_max_active_levels;
   team->t.t_sched.sched = new_icvs->sched.sched;
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
   team->t.t_fp_control_saved = FALSE; /* not needed */
   team->t.t_x87_fpu_control_word = 0; /* not needed */
   team->t.t_mxcsr = 0; /* not needed */
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
   team->t.t_construct = 0;
 
@@ -6264,7 +6264,7 @@ void __kmp_register_library_startup(void) {
     double dtime;
     long ltime;
   } time;
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
   __kmp_initialize_system_tick();
 #endif
   __kmp_read_system_time(&time.dtime);
@@ -6839,13 +6839,13 @@ void __kmp_parallel_initialize(void) {
   KA_TRACE(10, ("__kmp_parallel_initialize: enter\n"));
   KMP_ASSERT(KMP_UBER_GTID(gtid));
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
   // Save the FP control regs.
   // Worker threads will set theirs to these values at thread startup.
   __kmp_store_x87_fpu_control_word(&__kmp_init_x87_fpu_control_word);
   __kmp_store_mxcsr(&__kmp_init_mxcsr);
   __kmp_init_mxcsr &= KMP_X86_MXCSR_MASK;
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
 #if KMP_OS_UNIX
 #if KMP_HANDLE_SIGNALS
@@ -7673,9 +7673,10 @@ __kmp_determine_reduction_method(
 #endif // KMP_OS_LINUX || KMP_OS_FREEBSD || KMP_OS_NETBSD || KMP_OS_WINDOWS ||
 // KMP_OS_DARWIN
 
-#elif KMP_ARCH_X86 || KMP_ARCH_ARM || KMP_ARCH_AARCH || KMP_ARCH_MIPS
+#elif KMP_ARCH_X86 || KMP_ARCH_X86_X32 || KMP_ARCH_ARM || KMP_ARCH_AARCH || KMP_ARCH_MIPS
 
 #if KMP_OS_LINUX || KMP_OS_WINDOWS
+   http://software.intel.com/en-us/articles/intel-software-development-emulator
 
     // basic tuning
 
diff --git a/runtime/src/kmp_settings.cpp b/runtime/src/kmp_settings.cpp
index 10beee4..9e9ce55 100644
--- a/runtime/src/kmp_settings.cpp
+++ b/runtime/src/kmp_settings.cpp
@@ -686,7 +686,7 @@ static void __kmp_stg_print_duplicate_lib_ok(kmp_str_buf_t *buffer,
 // -----------------------------------------------------------------------------
 // KMP_INHERIT_FP_CONTROL
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 
 static void __kmp_stg_parse_inherit_fp_control(char const *name,
                                                char const *value, void *data) {
@@ -700,7 +700,7 @@ static void __kmp_stg_print_inherit_fp_control(kmp_str_buf_t *buffer,
 #endif /* KMP_DEBUG */
 } // __kmp_stg_print_inherit_fp_control
 
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
 // -----------------------------------------------------------------------------
 // KMP_LIBRARY, OMP_WAIT_POLICY
@@ -2900,7 +2900,7 @@ static void __kmp_stg_parse_topology_method(char const *name, char const *value,
     __kmp_affinity_top_method = affinity_top_method_hwloc;
   }
 #endif
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
   else if (__kmp_str_match("x2apic id", 9, value) ||
            __kmp_str_match("x2apic_id", 9, value) ||
            __kmp_str_match("x2apic-id", 9, value) ||
@@ -2948,7 +2948,7 @@ static void __kmp_stg_parse_topology_method(char const *name, char const *value,
              __kmp_str_match("leaf4", 5, value)) {
     __kmp_affinity_top_method = affinity_top_method_apicid;
   }
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
   else if (__kmp_str_match("/proc/cpuinfo", 2, value) ||
            __kmp_str_match("cpuinfo", 5, value)) {
     __kmp_affinity_top_method = affinity_top_method_cpuinfo;
@@ -2978,7 +2978,7 @@ static void __kmp_stg_print_topology_method(kmp_str_buf_t *buffer,
     value = "all";
     break;
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
   case affinity_top_method_x2apicid:
     value = "x2APIC id";
     break;
@@ -2986,7 +2986,7 @@ static void __kmp_stg_print_topology_method(kmp_str_buf_t *buffer,
   case affinity_top_method_apicid:
     value = "APIC id";
     break;
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
 #if KMP_USE_HWLOC
   case affinity_top_method_hwloc:
@@ -4469,10 +4469,10 @@ static kmp_setting_t __kmp_stg_table[] = {
      __kmp_stg_print_handle_signals, NULL, 0, 0},
 #endif
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
     {"KMP_INHERIT_FP_CONTROL", __kmp_stg_parse_inherit_fp_control,
      __kmp_stg_print_inherit_fp_control, NULL, 0, 0},
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
 #ifdef KMP_GOMP_COMPAT
     {"GOMP_STACKSIZE", __kmp_stg_parse_stacksize, NULL, NULL, 0, 0},
diff --git a/runtime/src/kmp_stats.cpp b/runtime/src/kmp_stats.cpp
index d75695f..7241e9a 100644
--- a/runtime/src/kmp_stats.cpp
+++ b/runtime/src/kmp_stats.cpp
@@ -561,7 +561,7 @@ void kmp_stats_output_module::printHeaderInfo(FILE *statsOut) {
   fprintf(statsOut, "# Time of run: %s\n", &buffer[0]);
   if (gethostname(&hostName[0], sizeof(hostName)) == 0)
     fprintf(statsOut, "# Hostname: %s\n", &hostName[0]);
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
   fprintf(statsOut, "# CPU:  %s\n", &__kmp_cpuinfo.name[0]);
   fprintf(statsOut, "# Family: %d, Model: %d, Stepping: %d\n",
           __kmp_cpuinfo.family, __kmp_cpuinfo.model, __kmp_cpuinfo.stepping);
diff --git a/runtime/src/kmp_stats_timing.cpp b/runtime/src/kmp_stats_timing.cpp
index fe85dc4..dfe5fef 100644
--- a/runtime/src/kmp_stats_timing.cpp
+++ b/runtime/src/kmp_stats_timing.cpp
@@ -29,7 +29,7 @@ double tsc_tick_count::tick_time() {
   // pretty bad assumption of 1GHz clock for MIC
   return 1 / ((double)1000 * 1.e6);
 }
-#elif KMP_ARCH_X86 || KMP_ARCH_X86_64
+#elif KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 #include <string.h>
 // Extract the value from the CPUID information
 double tsc_tick_count::tick_time() {
diff --git a/runtime/src/kmp_taskq.cpp b/runtime/src/kmp_taskq.cpp
index 6e8f2d5..a5b52a4 100644
--- a/runtime/src/kmp_taskq.cpp
+++ b/runtime/src/kmp_taskq.cpp
@@ -1918,7 +1918,7 @@ void __kmpc_end_taskq_task(ident_t *loc, kmp_int32 global_tid,
     __kmp_pop_workshare(global_tid, ct_taskq, loc);
 
   if (in_parallel) {
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
     KMP_TEST_THEN_OR32(RCAST(volatile kmp_uint32 *, &queue->tq_flags),
                        TQF_ALL_TASKS_QUEUED);
 #else
@@ -1949,7 +1949,7 @@ void __kmpc_end_taskq_task(ident_t *loc, kmp_int32 global_tid,
       /* No synchronization needed for serial context */
       queue->tq_flags |= TQF_IS_LAST_TASK;
     } else {
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
       KMP_TEST_THEN_OR32(RCAST(volatile kmp_uint32 *, &queue->tq_flags),
                          TQF_IS_LAST_TASK);
 #else
diff --git a/runtime/src/kmp_utility.cpp b/runtime/src/kmp_utility.cpp
index 06090e6..27c6bc7 100644
--- a/runtime/src/kmp_utility.cpp
+++ b/runtime/src/kmp_utility.cpp
@@ -19,7 +19,7 @@
 
 static const char *unknown = "unknown";
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 
 /* NOTE: If called before serial_initialize (i.e. from runtime_initialize), then
    the debugging package has not been initialized yet, and only "0" will print
@@ -288,7 +288,7 @@ void __kmp_query_cpuid(kmp_cpuinfo_t *p) {
   }
 }
 
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
 void __kmp_expand_host_name(char *buffer, size_t size) {
   KMP_DEBUG_ASSERT(size >= sizeof(unknown));
diff --git a/runtime/src/z_Linux_asm.S b/runtime/src/z_Linux_asm.S
index c9fbc23..94e3cc4 100644
--- a/runtime/src/z_Linux_asm.S
+++ b/runtime/src/z_Linux_asm.S
@@ -18,7 +18,7 @@
 
 #include "kmp_config.h"
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 
 # if KMP_MIC
 // the 'delay r16/r32/r64' should be used instead of the 'pause'.
@@ -213,7 +213,7 @@ __kmp_unnamed_critical_addr:
 #endif /* KMP_GOMP_COMPAT */
 
 
-#if KMP_ARCH_X86 && !KMP_ARCH_PPC64
+#if (KMP_ARCH_X86 || KMP_ARCH_X86_X32) && !KMP_ARCH_PPC64
 
 // -----------------------------------------------------------------------
 // microtasking routines specifically written for IA-32 architecture
@@ -671,7 +671,7 @@ KMP_LABEL(invoke_3):
 #endif /* KMP_ARCH_X86 */
 
 
-#if KMP_ARCH_X86_64
+#if KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 
 // -----------------------------------------------------------------------
 // microtasking routines specifically written for IA-32 architecture and
@@ -1361,7 +1361,7 @@ KMP_LABEL(kmp_1_exit):
 
 
 // -----------------------------------------------------------------------
-#endif /* KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
 // '
 #if (KMP_OS_LINUX || KMP_OS_DARWIN) && KMP_ARCH_AARCH64
diff --git a/runtime/src/z_Linux_util.cpp b/runtime/src/z_Linux_util.cpp
index fa9b1c5..4f694cb 100644
--- a/runtime/src/z_Linux_util.cpp
+++ b/runtime/src/z_Linux_util.cpp
@@ -302,7 +302,7 @@ int __kmp_futex_determine_capable() {
 
 #endif // KMP_USE_FUTEX
 
-#if (KMP_ARCH_X86 || KMP_ARCH_X86_64) && (!KMP_ASM_INTRINS)
+#if (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32) && (!KMP_ASM_INTRINS)
 /* Only 32-bit "add-exchange" instruction on IA-32 architecture causes us to
    use compare_and_store for these routines */
 
@@ -418,7 +418,7 @@ kmp_uint64 __kmp_test_then_and64(volatile kmp_uint64 *p, kmp_uint64 d) {
   return old_value;
 }
 
-#endif /* (KMP_ARCH_X86 || KMP_ARCH_X86_64) && (! KMP_ASM_INTRINS) */
+#endif /* (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32) && (! KMP_ASM_INTRINS) */
 
 void __kmp_terminate_thread(int gtid) {
   int status;
@@ -531,12 +531,12 @@ static void *__kmp_launch_worker(void *thr) {
   KMP_CHECK_SYSFAIL("pthread_setcancelstate", status);
 #endif
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
   // Set FP control regs to be a copy of the parallel initialization thread's.
   __kmp_clear_x87_fpu_status_word();
   __kmp_load_x87_fpu_control_word(&__kmp_init_x87_fpu_control_word);
   __kmp_load_mxcsr(&__kmp_init_mxcsr);
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
 #ifdef KMP_BLOCK_SIGNALS
   status = sigfillset(&new_set);
@@ -1806,11 +1806,11 @@ void __kmp_runtime_initialize(void) {
     return;
   }
 
-#if (KMP_ARCH_X86 || KMP_ARCH_X86_64)
+#if (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32)
   if (!__kmp_cpuinfo.initialized) {
     __kmp_query_cpuid(&__kmp_cpuinfo);
   }
-#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
+#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 */
 
   __kmp_xproc = __kmp_get_xproc();
 
@@ -1917,7 +1917,7 @@ kmp_uint64 __kmp_now_nsec() {
   return KMP_NSEC_PER_SEC * t.tv_sec + 1000 * t.tv_usec;
 }
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 /* Measure clock ticks per millisecond */
 void __kmp_initialize_system_tick() {
   kmp_uint64 delay = 100000; // 50~100 usec on most machines.
@@ -2267,7 +2267,7 @@ finish: // Clean up and exit.
 
 #endif // USE_LOAD_BALANCE
 
-#if !(KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_MIC ||                            \
+#if !(KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32 || KMP_MIC ||                            \
       ((KMP_OS_LINUX || KMP_OS_DARWIN) && KMP_ARCH_AARCH64) || KMP_ARCH_PPC64)
 
 // we really only need the case with 1 argument, because CLANG always build
diff --git a/runtime/test/ompt/callback.h b/runtime/test/ompt/callback.h
index 3befe7c..0c5a075 100755
--- a/runtime/test/ompt/callback.h
+++ b/runtime/test/ompt/callback.h
@@ -105,7 +105,7 @@ ompt_label_##id:
   define_ompt_label(id) \
   print_possible_return_addresses(get_ompt_label_address(id))
 
-#if KMP_ARCH_X86 || KMP_ARCH_X86_64
+#if KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_X86_X32
 // On X86 the NOP instruction is 1 byte long. In addition, the comiler inserts
 // a MOV instruction for non-void runtime functions which is 3 bytes long.
 #define print_possible_return_addresses(addr) \