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 rare edge case bug in herk_l macro-kernel.

Browse Source 
Details:
- Fixed a potential bug in herk_l at the m_left edge case. If MR was
  chosen to be much larger than NR, then one could encounter edge cases
  in the the MC dimension that fall entirely below the diagonal, which
  the previous implementation of the herk_l macro-kernel was not allowing
  for.
This commit is contained in:
Field G. Van Zee
2013-04-23 17:49:10 -05:00
parent 1dab11e37d
commit db072a5b4a
1 changed files with 38 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

53
frame/3/herk/bli_herk_l_ker_var2.c
View File

@@ -302,27 +302,50 @@ void PASTEMAC(ch,varname)( \
c11 += rstep_c; \
} \
\
/* Bottom edge handling. (Note that the diagonal is guaranteed not
to factor in here.) */ \
/* Bottom edge handling. */ \
if ( m_left ) \
{ \
/* Compute the diagonal offset for the submatrix at (i,j). */ \
diagoffc_ij = diagoffc - (doff_t)j*NR + (doff_t)i*MR; \
\
/* Compute the address of the next panel of A. */ \
a2 = a_cast; \
\
/* Invoke the gemm micro-kernel. */ \
PASTEMAC(ch,ukrname)( k, \
alpha_cast, \
a1, \
bp, \
zero, \
ct, rs_ct, cs_ct, \
a2, b2 ); \
if ( bli_intersects_diag_n( diagoffc_ij, m_left, NR ) ) \
{ \
/* Invoke the gemm micro-kernel. */ \
PASTEMAC(ch,ukrname)( k, \
alpha_cast, \
a1, \
bp, \
zero, \
ct, rs_ct, cs_ct, \
a2, b2 ); \
\
/* Scale the bottom edge of C and add the result. */ \
PASTEMAC3(ch,ch,ch,xpbys_mxn)( m_left, NR, \
ct, rs_ct, cs_ct, \
beta_cast, \
c11, rs_c, cs_c ); \
/* Scale C and add the result to only the stored part. */ \
PASTEMAC3(ch,ch,ch,xpbys_mxn_l)( diagoffc_ij, \
m_left, NR, \
ct, rs_ct, cs_ct, \
beta_cast, \
c11, rs_c, cs_c ); \
} \
else if ( bli_is_strictly_below_diag_n( diagoffc_ij, m_left, NR ) ) \
{ \
/* Invoke the gemm micro-kernel. */ \
PASTEMAC(ch,ukrname)( k, \
alpha_cast, \
a1, \
bp, \
zero, \
ct, rs_ct, cs_ct, \
a2, b2 ); \
\
/* Scale the bottom edge of C and add the result. */ \
PASTEMAC3(ch,ch,ch,xpbys_mxn)( m_left, NR, \
ct, rs_ct, cs_ct, \
beta_cast, \
c11, rs_c, cs_c ); \
} \
} \
\
b1 += cstep_b; \