amd/blis
1
0
Fork 0
mirror of https://github.com/amd/blis.git synced 2026-07-02 21:27:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fixed warnings and compilation issues with GCC in TRSM

Browse Source 
- Current implementation uses macros to expand the code at
  compile time, but this is causing some false warning in GCC12 and 14.
- Added switch case in trsm right variants for n_remainder.
- This ensures that n_rem is compile time constant, therefore
   warnings related to array subscript out of bounds are fixed.
- mtune=znver3 flag is causing compilation issue in GCC 9.1,
  therefore this flag is removed.
- Remaned the file bli_trsm_small to bli_trsm_small_zen5 in order
  to avoid possibily of missing symbols.

AMD-Internal: [CPUPL-6199]
Change-Id: Ib8e90196ce0a41d38c2b29226df5ab6c2d8ba996
This commit is contained in:
Shubham Sharma
2024-12-18 06:21:10 -05:00
committed by Shubham Sharma
parent 4ce708c316
commit 8f99d8a5bb
1 changed files with 174 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

193
kernels/zen5/3/bli_trsm_small.c → kernels/zen5/3/bli_trsm_small_zen5.c
View File

@@ -589,15 +589,97 @@ BLIS_INLINE void runn_n_rem
L_ = p;
}
#endif
for( i = 0; (i+d_mr-1) < m; i += d_mr )
switch (n_rem)
{
RUNN_FRINGE(D_MR_, n_rem);
}
m_rem = m - i;
if( m_rem > 0 )
{
RUNN_FRINGE( m_rem, n_rem );
case 7:
for( i = 0; (i+d_mr-1) < m; i += d_mr )
{
RUNN_FRINGE(D_MR_, 7 );
}
m_rem = m - i;
if( m_rem > 0 )
{
RUNN_FRINGE( m_rem, 7 );
}
break;
case 6:
for( i = 0; (i+d_mr-1) < m; i += d_mr )
{
RUNN_FRINGE(D_MR_, 6 );
}
m_rem = m - i;
if( m_rem > 0 )
{
RUNN_FRINGE( m_rem, 6 );
}
break;
case 5:
for( i = 0; (i+d_mr-1) < m; i += d_mr )
{
RUNN_FRINGE(D_MR_, 5 );
}
m_rem = m - i;
if( m_rem > 0 )
{
RUNN_FRINGE( m_rem, 5 );
}
break;
case 4:
for( i = 0; (i+d_mr-1) < m; i += d_mr )
{
RUNN_FRINGE(D_MR_, 4 );
}
m_rem = m - i;
if( m_rem > 0 )
{
RUNN_FRINGE( m_rem, 4 );
}
break;
case 3:
for( i = 0; (i+d_mr-1) < m; i += d_mr )
{
RUNN_FRINGE(D_MR_, 3 );
}
m_rem = m - i;
if( m_rem > 0 )
{
RUNN_FRINGE( m_rem, 3 );
}
break;
case 2:
for( i = 0; (i+d_mr-1) < m; i += d_mr )
{
RUNN_FRINGE(D_MR_, 2 );
}
m_rem = m - i;
if( m_rem > 0 )
{
RUNN_FRINGE( m_rem, 2 );
}
break;
case 1:
for( i = 0; (i+d_mr-1) < m; i += d_mr )
{
RUNN_FRINGE(D_MR_, 1 );
}
m_rem = m - i;
if( m_rem > 0 )
{
RUNN_FRINGE( m_rem, 1 );
}
break;
default:
break;
}
}
/*
@@ -665,21 +747,96 @@ BLIS_INLINE void rlnn_n_rem
L_ = p;
}
#endif
for( i = (m - d_mr); (i + 1) > 0; i -= d_mr )
switch (n_rem)
{
RLNN_FRINGE(D_MR_, n_rem);
}
m_rem = i + d_mr;
if( m_rem > 0 )
{
RLNN_FRINGE( m_rem, n_rem );
case 7:
for( i = (m - d_mr); (i + 1) > 0; i -= d_mr )
{
RLNN_FRINGE(D_MR_, 7);
}
m_rem = i + d_mr;
if( m_rem > 0 )
{
RLNN_FRINGE( m_rem, 7);
}
break;
case 6:
for( i = (m - d_mr); (i + 1) > 0; i -= d_mr )
{
RLNN_FRINGE(D_MR_, 6);
}
m_rem = i + d_mr;
if( m_rem > 0 )
{
RLNN_FRINGE( m_rem, 6);
}
break;
case 5:
for( i = (m - d_mr); (i + 1) > 0; i -= d_mr )
{
RLNN_FRINGE(D_MR_, 5);
}
m_rem = i + d_mr;
if( m_rem > 0 )
{
RLNN_FRINGE( m_rem, 5);
}
break;
case 4:
for( i = (m - d_mr); (i + 1) > 0; i -= d_mr )
{
RLNN_FRINGE(D_MR_, 4);
}
m_rem = i + d_mr;
if( m_rem > 0 )
{
RLNN_FRINGE( m_rem, 4);
}
break;
case 3:
for( i = (m - d_mr); (i + 1) > 0; i -= d_mr )
{
RLNN_FRINGE(D_MR_, 3);
}
m_rem = i + d_mr;
if( m_rem > 0 )
{
RLNN_FRINGE( m_rem, 3);
}
break;
case 2:
for( i = (m - d_mr); (i + 1) > 0; i -= d_mr )
{
RLNN_FRINGE(D_MR_, 2);
}
m_rem = i + d_mr;
if( m_rem > 0 )
{
RLNN_FRINGE( m_rem, 2);
}
break;
case 1:
for( i = (m - d_mr); (i + 1) > 0; i -= d_mr )
{
RLNN_FRINGE(D_MR_, 1);
}
m_rem = i + d_mr;
if( m_rem > 0 )
{
RLNN_FRINGE( m_rem, 1);
}
break;
default:
break;
}
}
// RUNN - RLTN
err_t __attribute__((target("tune=znver3")))
bli_dtrsm_small_XAltB_XAuB_ZEN5
err_t bli_dtrsm_small_XAltB_XAuB_ZEN5
(
obj_t* AlphaObj,
obj_t* a,
@@ -1422,9 +1579,7 @@ BLIS_INLINE void lunn_n_rem
// LLNN - LUTN
err_t
__attribute__((target("tune=skylake-avx512")))
bli_dtrsm_small_AutXB_AlXB_ZEN5
err_t bli_dtrsm_small_AutXB_AlXB_ZEN5
(
obj_t* AlphaObj,
obj_t* a,