mirror of
https://github.com/amd/blis.git
synced 2026-05-11 17:50:00 +00:00
Minor updates to packm blocked, cxk_3m/4m code.
Details: - Added 'const' qualifier to inlined packing code that handles micro-panel packing that is too large for an existing packm ukernel. - Comment updates.
This commit is contained in:
@@ -214,8 +214,8 @@ void PASTEMAC(ch,varname)( \
|
||||
row_stored = bli_is_col_packed( schema ); \
|
||||
col_stored = bli_is_row_packed( schema ); \
|
||||
\
|
||||
/* If the strides of P indicate row storage, then we are packing to
|
||||
column panels; otherwise, if the strides indicate column storage,
|
||||
/* If the row storage flag indicates row storage, then we are packing
|
||||
to column panels; otherwise, if the strides indicate column storage,
|
||||
we are packing to row panels. */ \
|
||||
if ( row_stored ) \
|
||||
{ \
|
||||
|
||||
@@ -255,8 +255,8 @@ void PASTEMAC(ch,varname)( \
|
||||
row_stored = bli_is_col_packed( schema ); \
|
||||
col_stored = bli_is_row_packed( schema ); \
|
||||
\
|
||||
/* If the strides of P indicate row storage, then we are packing to
|
||||
column panels; otherwise, if the strides indicate column storage,
|
||||
/* If the row storage flag indicates row storage, then we are packing
|
||||
to column panels; otherwise, if the strides indicate column storage,
|
||||
we are packing to row panels. */ \
|
||||
if ( row_stored ) \
|
||||
{ \
|
||||
|
||||
@@ -255,8 +255,8 @@ void PASTEMAC(ch,varname)( \
|
||||
row_stored = bli_is_col_packed( schema ); \
|
||||
col_stored = bli_is_row_packed( schema ); \
|
||||
\
|
||||
/* If the strides of P indicate row storage, then we are packing to
|
||||
column panels; otherwise, if the strides indicate column storage,
|
||||
/* If the row storage flag indicates row storage, then we are packing
|
||||
to column panels; otherwise, if the strides indicate column storage,
|
||||
we are packing to row panels. */ \
|
||||
if ( row_stored ) \
|
||||
{ \
|
||||
|
||||
@@ -198,8 +198,8 @@ void PASTEMAC(ch,varname)( \
|
||||
ctype_r* restrict p_r = ( ctype_r* )p; \
|
||||
ctype_r* restrict p_i = ( ctype_r* )p + psp; \
|
||||
ctype_r* restrict p_rpi = ( ctype_r* )p + 2*psp; \
|
||||
dim_t inca2 = 2*inca; \
|
||||
dim_t lda2 = 2*lda; \
|
||||
const dim_t inca2 = 2*inca; \
|
||||
const dim_t lda2 = 2*lda; \
|
||||
\
|
||||
/* Treat the panel as m x n and column-stored (unit row stride). */ \
|
||||
\
|
||||
|
||||
@@ -198,8 +198,8 @@ void PASTEMAC(ch,varname)( \
|
||||
ctype_r* restrict a_i = ( ctype_r* )a + 1; \
|
||||
ctype_r* restrict p_r = ( ctype_r* )p; \
|
||||
ctype_r* restrict p_i = ( ctype_r* )p + psp; \
|
||||
dim_t inca2 = 2*inca; \
|
||||
dim_t lda2 = 2*lda; \
|
||||
const dim_t inca2 = 2*inca; \
|
||||
const dim_t lda2 = 2*lda; \
|
||||
\
|
||||
/* Treat the panel as m x n and column-stored (unit row stride). */ \
|
||||
\
|
||||
|
||||
Reference in New Issue
Block a user