amd/blis
1
0
Fork 0
mirror of https://github.com/amd/blis.git synced 2026-05-11 17:50:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fixing some bugs with herk parallelization

Browse Source
This commit is contained in:
Tyler Smith
2014-03-17 17:15:35 -05:00
parent c51d011083
commit 5296f58975
5 changed files with 8 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
frame/3/gemm/bli_gemm_threading.c
View File

@@ -172,8 +172,8 @@ gemm_thrinfo_t* bli_create_gemm_thrinfo_paths( )
gemm_thrinfo_t* jc_info = &paths[global_comm_id];
bli_setup_gemm_thrinfo_node( jc_info, global_comm, global_comm_id,
jr_comm, jr_comm_id,
jr_way, a,
jc_comm, jc_comm_id,
jc_way, a,
NULL, NULL, kc_info);
}
}

4
frame/3/herk/bli_herk_blk_var2f.c
View File

@@ -81,9 +81,9 @@ void bli_herk_blk_var2f( obj_t* a,
c1S_pack = thread_ibroadcast( thread, &c1S_pack_s );
// Pack A (if instructed).
bli_packm_int( a, &a_pack,
bli_packm_int( a, a_pack,
cntl_sub_packm_a( cntl ),
hemm_thread_sub_opackm( thread ) );
herk_thread_sub_opackm( thread ) );
// Query dimension in partitioning direction.
n_trans = bli_obj_width_after_trans( *c );

1
frame/3/herk/bli_herk_blk_var3f.c
View File

@@ -101,6 +101,7 @@ void bli_herk_blk_var3f( obj_t* a,
bli_packm_init( &ah1, ah1_pack,
cntl_sub_packm_b( cntl ) );
}
thread_ibarrier( thread );
// Pack A1 (if instructed).
bli_packm_int( &a1, a1_pack,

2
frame/3/herk/bli_herk_front.c
View File

@@ -77,7 +77,7 @@ void bli_herk_front( obj_t* alpha,
bli_obj_induce_trans( c_local );
}
herk_thrinfo_t* infos = bli_herk_cntl_get_thrinfos();
herk_thrinfo_t* infos = bli_create_herk_thrinfo_paths();
dim_t n_threads = thread_num_threads( (&infos[0]) );
// Invoke the internal back-end.

15
frame/3/herk/bli_herk_threading.c
View File

@@ -84,17 +84,6 @@ herk_thrinfo_t* bli_create_herk_thrinfo_node( thread_comm_t* ocomm, dim_t ocomm_
return thread;
}
dim_t read_env( char* env )
{
dim_t number = 1;
char* str = getenv( env );
if( str != NULL )
{
number = strtol( str, NULL, 10 );
}
return number;
}
void bli_herk_thrinfo_free_paths( herk_thrinfo_t* threads )
{
}
@@ -172,8 +161,8 @@ herk_thrinfo_t* bli_create_herk_thrinfo_paths( )
herk_thrinfo_t* jc_info = &paths[global_comm_id];
bli_setup_herk_thrinfo_node( jc_info, global_comm, global_comm_id,
jr_comm, jr_comm_id,
jr_way, a,
jc_comm, jc_comm_id,
jc_way, a,
NULL, NULL, kc_info);
}
}