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

Some fixes for the internal functions,

Browse Source 
was innappropriately only having thread chief do some things.
This commit is contained in:
Tyler Smith
2014-03-27 14:18:46 -05:00
parent a6fd483454
commit 9f78ec6e7e
3 changed files with 12 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
frame/3/gemm/bli_gemm_int.c
View File

@@ -110,28 +110,25 @@ void bli_gemm_int( obj_t* alpha,
// packed, this is our last chance to handle the transposition.
if ( cntl_is_leaf( cntl ) && bli_obj_has_trans( *c ) )
{
if( thread_am_ochief( thread ) ) {
//if( thread_am_ochief( thread ) ) {
bli_obj_induce_trans( c_local );
bli_obj_set_onlytrans( BLIS_NO_TRANSPOSE, c_local );
}
// }
}
// If alpha is non-unit, typecast and apply it to the scalar attached
// to B.
if ( !bli_obj_equals( alpha, &BLIS_ONE ) )
{
if( thread_am_ochief( thread ) )
bli_obj_scalar_apply_scalar( alpha, &b_local );
bli_obj_scalar_apply_scalar( alpha, &b_local );
}
// If beta is non-unit, typecast and apply it to the scalar attached
// to C.
if ( !bli_obj_equals( beta, &BLIS_ONE ) )
{
if( thread_am_ochief( thread ) )
bli_obj_scalar_apply_scalar( beta, &c_local );
bli_obj_scalar_apply_scalar( beta, &c_local );
}
thread_obarrier( thread );
// Extract the variant number and implementation type.
n = cntl_var_num( cntl );

14
frame/3/herk/bli_herk_int.c
View File

@@ -109,34 +109,28 @@ void bli_herk_int( obj_t* alpha,
// packed, this is our last chance to handle the transposition.
if ( cntl_is_leaf( cntl ) && bli_obj_has_trans( *c ) )
{
if( thread_am_ochief( thread ) ) {
bli_obj_induce_trans( c_local );
bli_obj_set_onlytrans( BLIS_NO_TRANSPOSE, c_local );
}
bli_obj_induce_trans( c_local );
bli_obj_set_onlytrans( BLIS_NO_TRANSPOSE, c_local );
}
// If alpha is non-unit, typecast and apply it to the scalar
// attached to A'.
if ( !bli_obj_equals( alpha, &BLIS_ONE ) )
{
if( thread_am_ochief( thread ) )
bli_obj_scalar_apply_scalar( alpha, &ah_local );
bli_obj_scalar_apply_scalar( alpha, &ah_local );
}
// If beta is non-unit, typecast and apply it to the scalar
// attached to C.
if ( !bli_obj_equals( beta, &BLIS_ONE ) )
{
if( thread_am_ochief( thread ) )
bli_obj_scalar_apply_scalar( beta, &c_local );
bli_obj_scalar_apply_scalar( beta, &c_local );
}
// Set a bool based on the uplo field of C's root object.
if ( bli_obj_root_is_lower( c_local ) ) uplo = 0;
else uplo = 1;
thread_obarrier( thread );
// Extract the variant number and implementation type.
n = cntl_var_num( cntl );
i = cntl_impl_type( cntl );

14
frame/3/trmm/bli_trmm_int.c
View File

@@ -131,26 +131,22 @@ void bli_trmm_int( obj_t* alpha,
// packed, this is our last chance to handle the transposition.
if ( cntl_is_leaf( cntl ) && bli_obj_has_trans( *c ) )
{
if( thread_am_ochief( thread ) ) {
bli_obj_induce_trans( c_local );
bli_obj_set_onlytrans( BLIS_NO_TRANSPOSE, c_local );
}
bli_obj_induce_trans( c_local );
bli_obj_set_onlytrans( BLIS_NO_TRANSPOSE, c_local );
}
// If alpha is non-unit, typecast and apply it to the scalar attached
// to B.
if ( !bli_obj_equals( alpha, &BLIS_ONE ) )
{
if( thread_am_ochief( thread ) )
bli_obj_scalar_apply_scalar( alpha, &b_local );
bli_obj_scalar_apply_scalar( alpha, &b_local );
}
// If beta is non-unit, typecast and apply it to the scalar attached
// to C.
if ( !bli_obj_equals( beta, &BLIS_ONE ) )
{
if( thread_am_ochief( thread ) )
bli_obj_scalar_apply_scalar( beta, &c_local );
bli_obj_scalar_apply_scalar( beta, &c_local );
}
// Set two bools: one based on the implied side parameter (the structure
@@ -170,8 +166,6 @@ void bli_trmm_int( obj_t* alpha,
else uplo = 1;
}
thread_obarrier( thread );
// Extract the variant number and implementation type.
n = cntl_var_num( cntl );
i = cntl_impl_type( cntl );