mirror of
https://github.com/amd/blis.git
synced 2026-05-11 17:50:00 +00:00
Changed default blocking factor to default double precision MR and NR
This commit is contained in:
@@ -83,7 +83,7 @@ void bli_gemm_blk_var1f( obj_t* a,
|
||||
// Query dimension in partitioning direction.
|
||||
m_trans = bli_obj_length_after_trans( *a );
|
||||
dim_t start, end;
|
||||
bli_get_range( thread, 0, m_trans, 8, &start, &end );
|
||||
bli_get_range( thread, 0, m_trans, BLIS_DEFAULT_MC_D, &start, &end );
|
||||
|
||||
// Partition along the m dimension.
|
||||
for ( i = start; i < end; i += b_alg )
|
||||
|
||||
@@ -82,7 +82,7 @@ void bli_gemm_blk_var2f( obj_t* a,
|
||||
// Query dimension in partitioning direction.
|
||||
n_trans = bli_obj_width_after_trans( *b );
|
||||
dim_t start, end;
|
||||
bli_get_range( thread, 0, n_trans, 8, &start, &end );
|
||||
bli_get_range( thread, 0, n_trans, BLIS_DEFAULT_NC_D, &start, &end );
|
||||
|
||||
// Partition along the n dimension.
|
||||
for ( i = start; i < end; i += b_alg )
|
||||
|
||||
@@ -82,7 +82,7 @@ void bli_herk_blk_var1f( obj_t* a,
|
||||
// Query dimension in partitioning direction.
|
||||
m_trans = bli_obj_length_after_trans( *c );
|
||||
dim_t start, end;
|
||||
bli_get_range_weighted( thread, 0, m_trans, 8, bli_obj_is_upper( *c ), &start, &end );
|
||||
bli_get_range_weighted( thread, 0, m_trans, BLIS_DEFAULT_MC_D, bli_obj_is_upper( *c ), &start, &end );
|
||||
|
||||
// Partition along the m dimension.
|
||||
for ( i = start; i < end; i += b_alg )
|
||||
|
||||
@@ -91,7 +91,7 @@ void bli_herk_blk_var2f( obj_t* a,
|
||||
|
||||
// Needs to be replaced with a weighted range because triangle
|
||||
//bli_get_range( thread, 0, n_trans, 8, &start, &end );
|
||||
bli_get_range_weighted( thread, 0, n_trans, 8, bli_obj_is_lower( *c ), &start, &end );
|
||||
bli_get_range_weighted( thread, 0, n_trans, BLIS_DEFAULT_NC_D, bli_obj_is_lower( *c ), &start, &end );
|
||||
|
||||
// Partition along the n dimension.
|
||||
for ( i = start; i < end; i += b_alg )
|
||||
|
||||
@@ -94,8 +94,7 @@ void bli_trmm_blk_var1f( obj_t* a,
|
||||
bli_obj_width_after_trans( *a );
|
||||
|
||||
dim_t start, end;
|
||||
bli_get_range( thread, offA, m_trans, 8, &start, &end );
|
||||
|
||||
bli_get_range( thread, offA, m_trans, BLIS_DEFAULT_MC_D, &start, &end );
|
||||
// Partition along the m dimension.
|
||||
for ( i = start; i < end; i += b_alg )
|
||||
{
|
||||
|
||||
@@ -83,7 +83,7 @@ void bli_trmm_blk_var2b( obj_t* a,
|
||||
n_trans = bli_obj_width_after_trans( *b );
|
||||
dim_t start, end;
|
||||
//bli_get_range( thread, 0, n_trans, 8, &start, &end );
|
||||
bli_get_range_weighted( thread, 0, n_trans, 8, 0, &start, &end );
|
||||
bli_get_range_weighted( thread, 0, n_trans, BLIS_DEFAULT_NC_D, 0, &start, &end );
|
||||
|
||||
// Partition along the n dimension.
|
||||
for ( i = start; i < end; i += b_alg )
|
||||
|
||||
@@ -83,7 +83,7 @@ void bli_trmm_blk_var2f( obj_t* a,
|
||||
n_trans = bli_obj_width_after_trans( *b );
|
||||
dim_t start, end;
|
||||
//bli_get_range( thread, 0, n_trans, 8, &start, &end );
|
||||
bli_get_range_weighted( thread, 0, n_trans, 8, 1, &start, &end );
|
||||
bli_get_range_weighted( thread, 0, n_trans, BLIS_DEFAULT_NC_D, 1, &start, &end );
|
||||
|
||||
// Partition along the n dimension.
|
||||
for ( i = start; i < end; i += b_alg )
|
||||
|
||||
@@ -82,7 +82,7 @@ void bli_trsm_blk_var1b( obj_t* a,
|
||||
bli_obj_width_after_trans( *a );
|
||||
|
||||
dim_t start, end;
|
||||
bli_get_range( thread, offA, m_trans, 8, &start, &end );
|
||||
bli_get_range( thread, offA, m_trans, BLIS_DEFAULT_MC_D, &start, &end );
|
||||
|
||||
// Partition along the remaining portion of the m dimension.
|
||||
for ( i = start; i < end; i += b_alg )
|
||||
|
||||
@@ -81,7 +81,7 @@ void bli_trsm_blk_var1f( obj_t* a,
|
||||
offA = bli_abs( bli_obj_diag_offset_after_trans( *a ) );
|
||||
|
||||
dim_t start, end;
|
||||
bli_get_range( thread, offA, m_trans, 8, &start, &end );
|
||||
bli_get_range( thread, offA, m_trans, BLIS_DEFAULT_MC_D, &start, &end );
|
||||
|
||||
// Partition along the remaining portion of the m dimension.
|
||||
for ( i = start; i < end; i += b_alg )
|
||||
|
||||
@@ -83,7 +83,7 @@ void bli_trsm_blk_var2b( obj_t* a,
|
||||
// Query dimension in partitioning direction.
|
||||
n_trans = bli_obj_width_after_trans( *b );
|
||||
dim_t start, end;
|
||||
bli_get_range_weighted( thread, 0, n_trans, 8, 0, &start, &end );
|
||||
bli_get_range_weighted( thread, 0, n_trans, BLIS_DEFAULT_NC_D, 0, &start, &end );
|
||||
|
||||
// Partition along the n dimension.
|
||||
for ( i = start; i < end; i += b_alg )
|
||||
|
||||
@@ -84,7 +84,7 @@ void bli_trsm_blk_var2f( obj_t* a,
|
||||
n_trans = bli_obj_width_after_trans( *b );
|
||||
dim_t start, end;
|
||||
//bli_get_range( thread, 0, n_trans, 8, &start, &end );
|
||||
bli_get_range_weighted( thread, 0, n_trans, 8, 1, &start, &end );
|
||||
bli_get_range_weighted( thread, 0, n_trans, BLIS_DEFAULT_NC_D, 1, &start, &end );
|
||||
|
||||
// Partition along the n dimension.
|
||||
for ( i = start; i < end; i += b_alg )
|
||||
|
||||
Reference in New Issue
Block a user