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

Some fixes for the bgq kernels

Browse Source
This commit is contained in:
Tyler Michael Smith
2014-04-08 17:50:44 +00:00
parent bde697f75e
commit 20e24430a7
3 changed files with 5 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
kernels/bgq/1/bli_axpyv_opt_var1.c
View File

@@ -75,4 +75,3 @@ void bli_daxpyv_opt_var1(
y[4*n_run + i] += *alpha * x[4*n_run + i];
}
}
}

4
kernels/bgq/1/bli_dotv_opt_var1.c
View File

@@ -85,11 +85,11 @@ void bli_ddotv_opt_var1(
rhos += vec_extract( rhov, 2 );
rhos += vec_extract( rhov, 3 );
}
for ( dim_t i = n_left; i < n_left; i++ )
for ( dim_t i = 0; i < n_left; i++ )
{
rhos += x[4*n_run + i] * y[4*n_run + i];
}
*rho = rhos;
}}
}

7
kernels/bgq/3/bli_gemm_8x8.c
View File

@@ -226,7 +226,7 @@ void bli_zgemm_8x8(
double * a = (double*) a_z;
double * b = (double*) b_z;
double * c = (double*) c_z;
//Registers for storing C.
//2 2x4 subblocks of C, c0, and c1
//Each sub-block has 4 columns, 0, 1, 2, 3
@@ -253,7 +253,6 @@ void bli_zgemm_8x8(
vector4double b0, b1, b2, b3;
vector4double a0, a1;
double _Complex tmp = 0.0;
for( dim_t i = 0; i < k; i++ )
{
@@ -334,13 +333,13 @@ void bli_zgemm_8x8(
\
/* Scale by alpha */ \
REG1 = vec_xmadd( alphav, AB, zed ); \
REG2 = vec_xxcpnmadd( AB, alphav, zed ); \
REG2 = vec_xxcpnmadd( AB, alphav, zed ); \
AB = vec_sub(REG1, REG2 ); \
\
\
/* Scale by beta */ \
REG1 = vec_xmadd( betav, C, zed ); \
REG2 = vec_xxcpnmadd( C, betav, zed ); \
REG2 = vec_xxcpnmadd( C, betav, zed ); \
C = vec_sub(REG1, REG2 ); \
\
/* Add AB to C */ \