amd/blis
1
0
Fork 0
mirror of https://github.com/amd/blis.git synced 2026-05-11 09:39:59 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fixed minor bug in computing b_next in gemm.

Browse Source
This commit is contained in:
Field G. Van Zee
2013-04-23 19:12:14 -05:00
parent db072a5b4a
commit 3671528cf8
1 changed files with 7 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
frame/3/gemm/bli_gemm_ker_var2.c
View File

@@ -212,7 +212,7 @@ void PASTEMAC(ch,varname)( \
m_left = m % MR; \
\
/* Compute the number of elements in B to duplicate per iteration. */ \
k_nr = k * NR; \
k_nr = k * NR; \
\
/* Determine some increments used to step through A, B, and C. */ \
rstep_a = ps_a; \
@@ -222,8 +222,8 @@ void PASTEMAC(ch,varname)( \
rstep_c = rs_c * MR; \
cstep_c = cs_c * NR; \
\
b1 = b_cast; \
c1 = c_cast; \
b1 = b_cast; \
c1 = c_cast; \
\
/* If the micro-kernel needs elements of B duplicated, set bp to
point to the duplication buffer. If no duplication is called for,
@@ -250,12 +250,11 @@ void PASTEMAC(ch,varname)( \
{ \
/* Compute the addresses of the next panels of A and B. */ \
a2 = a1 + rstep_a; \
if ( bli_is_last_iter_f( i, m_iter, m_left ) ) \
if ( i == m_iter - 1 && m_left == 0 ) \
{ \
a2 = a_cast; \
b2 = b1 + cstep_b; \
/*if ( i == n_iter - 1 && n_left == 0 )*/ \
if ( bli_is_last_iter_f( i, n_iter, n_left ) ) \
if ( j == n_iter - 1 && n_left == 0 ) \
b2 = b_cast; \
} \
\
@@ -278,7 +277,7 @@ void PASTEMAC(ch,varname)( \
/* Compute the addresses of the next panels of A and B. */ \
a2 = a_cast; \
b2 = b1 + cstep_b; \
if ( bli_is_last_iter_f( i, n_iter, n_left ) ) \
if ( j == n_iter - 1 && n_left == 0 ) \
b2 = b_cast; \
\
\
@@ -320,7 +319,7 @@ void PASTEMAC(ch,varname)( \
{ \
/* Compute the addresses of the next panels of A and B. */ \
a2 = a1 + rstep_a; \
if ( bli_is_last_iter_f( i, m_iter, m_left ) ) \
if ( i == m_iter - 1 && m_left == 0 ) \
{ \
a2 = a_cast; \
b2 = b_cast; \