replica
/
hanchenye-llvm-project
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Mixed type atomic routines for unsigned integers. 

New routines should be used for atomics like "<int>OP=<float>" when <int> is
unsigned. Using functions __kmpc_atomic_fixed<bits>_<op>_fp) produces incorrect
results

Differential Revision: https://reviews.llvm.org/D24756

llvm-svn: 282509
This commit is contained in:
Jonathan Peyton 2016-09-27 17:38:48 +00:00
parent 789ea82fa3
commit be31337e9d
2 changed files with 36 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
openmp/runtime/src/kmp_atomic.c
View File

@ -184,6 +184,9 @@ There are versions here for integers of size 1,2,4 and 8 bytes both signed and u
__kmpc_atomic_fixed1_wr
__kmpc_atomic_fixed1_xor
__kmpc_atomic_fixed1_xor_cpt
__kmpc_atomic_fixed1u_add_fp
__kmpc_atomic_fixed1u_sub_fp
__kmpc_atomic_fixed1u_mul_fp
__kmpc_atomic_fixed1u_div
__kmpc_atomic_fixed1u_div_cpt
__kmpc_atomic_fixed1u_div_cpt_rev
@ -240,6 +243,9 @@ There are versions here for integers of size 1,2,4 and 8 bytes both signed and u
__kmpc_atomic_fixed2_wr
__kmpc_atomic_fixed2_xor
__kmpc_atomic_fixed2_xor_cpt
__kmpc_atomic_fixed2u_add_fp
__kmpc_atomic_fixed2u_sub_fp
__kmpc_atomic_fixed2u_mul_fp
__kmpc_atomic_fixed2u_div
__kmpc_atomic_fixed2u_div_cpt
__kmpc_atomic_fixed2u_div_cpt_rev
@ -296,6 +302,9 @@ There are versions here for integers of size 1,2,4 and 8 bytes both signed and u
__kmpc_atomic_fixed4_wr
__kmpc_atomic_fixed4_xor
__kmpc_atomic_fixed4_xor_cpt
__kmpc_atomic_fixed4u_add_fp
__kmpc_atomic_fixed4u_sub_fp
__kmpc_atomic_fixed4u_mul_fp
__kmpc_atomic_fixed4u_div
__kmpc_atomic_fixed4u_div_cpt
__kmpc_atomic_fixed4u_div_cpt_rev
@ -352,6 +361,9 @@ There are versions here for integers of size 1,2,4 and 8 bytes both signed and u
__kmpc_atomic_fixed8_wr
__kmpc_atomic_fixed8_xor
__kmpc_atomic_fixed8_xor_cpt
__kmpc_atomic_fixed8u_add_fp
__kmpc_atomic_fixed8u_sub_fp
__kmpc_atomic_fixed8u_mul_fp
__kmpc_atomic_fixed8u_div
__kmpc_atomic_fixed8u_div_cpt
__kmpc_atomic_fixed8u_div_cpt_rev
@ -1394,26 +1406,38 @@ ATOMIC_CMPXCHG_MIX( float4, kmp_real32, div, 32, /, float8, kmp_real64, 4r, 3, K
// RHS=float16 (deprecated, to be removed when we are sure the compiler does not use them)
#if KMP_HAVE_QUAD
ATOMIC_CMPXCHG_MIX( fixed1, char, add, 8, +, fp, _Quad, 1i, 0, KMP_ARCH_X86 ) // __kmpc_atomic_fixed1_add_fp
ATOMIC_CMPXCHG_MIX( fixed1u, uchar, add, 8, +, fp, _Quad, 1i, 0, KMP_ARCH_X86 ) // __kmpc_atomic_fixed1u_add_fp
ATOMIC_CMPXCHG_MIX( fixed1, char, sub, 8, -, fp, _Quad, 1i, 0, KMP_ARCH_X86 ) // __kmpc_atomic_fixed1_sub_fp
ATOMIC_CMPXCHG_MIX( fixed1u, uchar, sub, 8, -, fp, _Quad, 1i, 0, KMP_ARCH_X86 ) // __kmpc_atomic_fixed1u_sub_fp
ATOMIC_CMPXCHG_MIX( fixed1, char, mul, 8, *, fp, _Quad, 1i, 0, KMP_ARCH_X86 ) // __kmpc_atomic_fixed1_mul_fp
ATOMIC_CMPXCHG_MIX( fixed1u, uchar, mul, 8, *, fp, _Quad, 1i, 0, KMP_ARCH_X86 ) // __kmpc_atomic_fixed1u_mul_fp
ATOMIC_CMPXCHG_MIX( fixed1, char, div, 8, /, fp, _Quad, 1i, 0, KMP_ARCH_X86 ) // __kmpc_atomic_fixed1_div_fp
ATOMIC_CMPXCHG_MIX( fixed1u, uchar, div, 8, /, fp, _Quad, 1i, 0, KMP_ARCH_X86 ) // __kmpc_atomic_fixed1u_div_fp
ATOMIC_CMPXCHG_MIX( fixed2, short, add, 16, +, fp, _Quad, 2i, 1, KMP_ARCH_X86 ) // __kmpc_atomic_fixed2_add_fp
ATOMIC_CMPXCHG_MIX( fixed2u, ushort, add, 16, +, fp, _Quad, 2i, 1, KMP_ARCH_X86 ) // __kmpc_atomic_fixed2u_add_fp
ATOMIC_CMPXCHG_MIX( fixed2, short, sub, 16, -, fp, _Quad, 2i, 1, KMP_ARCH_X86 ) // __kmpc_atomic_fixed2_sub_fp
ATOMIC_CMPXCHG_MIX( fixed2u, ushort, sub, 16, -, fp, _Quad, 2i, 1, KMP_ARCH_X86 ) // __kmpc_atomic_fixed2u_sub_fp
ATOMIC_CMPXCHG_MIX( fixed2, short, mul, 16, *, fp, _Quad, 2i, 1, KMP_ARCH_X86 ) // __kmpc_atomic_fixed2_mul_fp
ATOMIC_CMPXCHG_MIX( fixed2u, ushort, mul, 16, *, fp, _Quad, 2i, 1, KMP_ARCH_X86 ) // __kmpc_atomic_fixed2u_mul_fp
ATOMIC_CMPXCHG_MIX( fixed2, short, div, 16, /, fp, _Quad, 2i, 1, KMP_ARCH_X86 ) // __kmpc_atomic_fixed2_div_fp
ATOMIC_CMPXCHG_MIX( fixed2u, ushort, div, 16, /, fp, _Quad, 2i, 1, KMP_ARCH_X86 ) // __kmpc_atomic_fixed2u_div_fp
ATOMIC_CMPXCHG_MIX( fixed4, kmp_int32, add, 32, +, fp, _Quad, 4i, 3, 0 ) // __kmpc_atomic_fixed4_add_fp
ATOMIC_CMPXCHG_MIX( fixed4u, kmp_uint32, add, 32, +, fp, _Quad, 4i, 3, 0 ) // __kmpc_atomic_fixed4u_add_fp
ATOMIC_CMPXCHG_MIX( fixed4, kmp_int32, sub, 32, -, fp, _Quad, 4i, 3, 0 ) // __kmpc_atomic_fixed4_sub_fp
ATOMIC_CMPXCHG_MIX( fixed4u, kmp_uint32, sub, 32, -, fp, _Quad, 4i, 3, 0 ) // __kmpc_atomic_fixed4u_sub_fp
ATOMIC_CMPXCHG_MIX( fixed4, kmp_int32, mul, 32, *, fp, _Quad, 4i, 3, 0 ) // __kmpc_atomic_fixed4_mul_fp
ATOMIC_CMPXCHG_MIX( fixed4u, kmp_uint32, mul, 32, *, fp, _Quad, 4i, 3, 0 ) // __kmpc_atomic_fixed4u_mul_fp
ATOMIC_CMPXCHG_MIX( fixed4, kmp_int32, div, 32, /, fp, _Quad, 4i, 3, 0 ) // __kmpc_atomic_fixed4_div_fp
ATOMIC_CMPXCHG_MIX( fixed4u, kmp_uint32, div, 32, /, fp, _Quad, 4i, 3, 0 ) // __kmpc_atomic_fixed4u_div_fp
ATOMIC_CMPXCHG_MIX( fixed8, kmp_int64, add, 64, +, fp, _Quad, 8i, 7, KMP_ARCH_X86 ) // __kmpc_atomic_fixed8_add_fp
ATOMIC_CMPXCHG_MIX( fixed8u, kmp_uint64, add, 64, +, fp, _Quad, 8i, 7, KMP_ARCH_X86 ) // __kmpc_atomic_fixed8u_add_fp
ATOMIC_CMPXCHG_MIX( fixed8, kmp_int64, sub, 64, -, fp, _Quad, 8i, 7, KMP_ARCH_X86 ) // __kmpc_atomic_fixed8_sub_fp
ATOMIC_CMPXCHG_MIX( fixed8u, kmp_uint64, sub, 64, -, fp, _Quad, 8i, 7, KMP_ARCH_X86 ) // __kmpc_atomic_fixed8u_sub_fp
ATOMIC_CMPXCHG_MIX( fixed8, kmp_int64, mul, 64, *, fp, _Quad, 8i, 7, KMP_ARCH_X86 ) // __kmpc_atomic_fixed8_mul_fp
ATOMIC_CMPXCHG_MIX( fixed8u, kmp_uint64, mul, 64, *, fp, _Quad, 8i, 7, KMP_ARCH_X86 ) // __kmpc_atomic_fixed8u_mul_fp
ATOMIC_CMPXCHG_MIX( fixed8, kmp_int64, div, 64, /, fp, _Quad, 8i, 7, KMP_ARCH_X86 ) // __kmpc_atomic_fixed8_div_fp
ATOMIC_CMPXCHG_MIX( fixed8u, kmp_uint64, div, 64, /, fp, _Quad, 8i, 7, KMP_ARCH_X86 ) // __kmpc_atomic_fixed8u_div_fp

12
openmp/runtime/src/kmp_atomic.h
View File

@ -677,26 +677,38 @@ void __kmpc_atomic_float4_div_float8( ident_t *id_ref, int gtid, kmp_real32 * lh
// RHS=float16 (deprecated, to be removed when we are sure the compiler does not use them)
#if KMP_HAVE_QUAD
void __kmpc_atomic_fixed1_add_fp( ident_t *id_ref, int gtid, char * lhs, _Quad rhs );
void __kmpc_atomic_fixed1u_add_fp( ident_t *id_ref, int gtid, unsigned char * lhs, _Quad rhs );
void __kmpc_atomic_fixed1_sub_fp( ident_t *id_ref, int gtid, char * lhs, _Quad rhs );
void __kmpc_atomic_fixed1u_sub_fp( ident_t *id_ref, int gtid, unsigned char * lhs, _Quad rhs );
void __kmpc_atomic_fixed1_mul_fp( ident_t *id_ref, int gtid, char * lhs, _Quad rhs );
void __kmpc_atomic_fixed1u_mul_fp( ident_t *id_ref, int gtid, unsigned char * lhs, _Quad rhs );
void __kmpc_atomic_fixed1_div_fp( ident_t *id_ref, int gtid, char * lhs, _Quad rhs );
void __kmpc_atomic_fixed1u_div_fp( ident_t *id_ref, int gtid, unsigned char * lhs, _Quad rhs );
void __kmpc_atomic_fixed2_add_fp( ident_t *id_ref, int gtid, short * lhs, _Quad rhs );
void __kmpc_atomic_fixed2u_add_fp( ident_t *id_ref, int gtid, unsigned short * lhs, _Quad rhs );
void __kmpc_atomic_fixed2_sub_fp( ident_t *id_ref, int gtid, short * lhs, _Quad rhs );
void __kmpc_atomic_fixed2u_sub_fp( ident_t *id_ref, int gtid, unsigned short * lhs, _Quad rhs );
void __kmpc_atomic_fixed2_mul_fp( ident_t *id_ref, int gtid, short * lhs, _Quad rhs );
void __kmpc_atomic_fixed2u_mul_fp( ident_t *id_ref, int gtid, unsigned short * lhs, _Quad rhs );
void __kmpc_atomic_fixed2_div_fp( ident_t *id_ref, int gtid, short * lhs, _Quad rhs );
void __kmpc_atomic_fixed2u_div_fp( ident_t *id_ref, int gtid, unsigned short * lhs, _Quad rhs );
void __kmpc_atomic_fixed4_add_fp( ident_t *id_ref, int gtid, kmp_int32 * lhs, _Quad rhs );
void __kmpc_atomic_fixed4u_add_fp( ident_t *id_ref, int gtid, kmp_uint32 * lhs, _Quad rhs );
void __kmpc_atomic_fixed4_sub_fp( ident_t *id_ref, int gtid, kmp_int32 * lhs, _Quad rhs );
void __kmpc_atomic_fixed4u_sub_fp( ident_t *id_ref, int gtid, kmp_uint32 * lhs, _Quad rhs );
void __kmpc_atomic_fixed4_mul_fp( ident_t *id_ref, int gtid, kmp_int32 * lhs, _Quad rhs );
void __kmpc_atomic_fixed4u_mul_fp( ident_t *id_ref, int gtid, kmp_uint32 * lhs, _Quad rhs );
void __kmpc_atomic_fixed4_div_fp( ident_t *id_ref, int gtid, kmp_int32 * lhs, _Quad rhs );
void __kmpc_atomic_fixed4u_div_fp( ident_t *id_ref, int gtid, kmp_uint32 * lhs, _Quad rhs );
void __kmpc_atomic_fixed8_add_fp( ident_t *id_ref, int gtid, kmp_int64 * lhs, _Quad rhs );
void __kmpc_atomic_fixed8u_add_fp( ident_t *id_ref, int gtid, kmp_uint64 * lhs, _Quad rhs );
void __kmpc_atomic_fixed8_sub_fp( ident_t *id_ref, int gtid, kmp_int64 * lhs, _Quad rhs );
void __kmpc_atomic_fixed8u_sub_fp( ident_t *id_ref, int gtid, kmp_uint64 * lhs, _Quad rhs );
void __kmpc_atomic_fixed8_mul_fp( ident_t *id_ref, int gtid, kmp_int64 * lhs, _Quad rhs );
void __kmpc_atomic_fixed8u_mul_fp( ident_t *id_ref, int gtid, kmp_uint64 * lhs, _Quad rhs );
void __kmpc_atomic_fixed8_div_fp( ident_t *id_ref, int gtid, kmp_int64 * lhs, _Quad rhs );
void __kmpc_atomic_fixed8u_div_fp( ident_t *id_ref, int gtid, kmp_uint64 * lhs, _Quad rhs );