amd/blis
1
0
Fork 0
mirror of https://github.com/amd/blis.git synced 2026-05-11 17:50:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

Removed gemmtrsm-, trsm-specific blocksize macros.

Browse Source 
Details:
- Modified gemmtrsm micro-kernel wrappers to use new aliased blocksize macros
  instead of operation-specific ones.
- Removed local, gemmtrsm-specific blocksize macro definitions found in
  micro-kernel header files.
  (Meant to include above changes in 31b100e7bf4a.)
- Added comments to reference gemmtrsm micro-kernel wrapper implementation.
This commit is contained in:
Field G. Van Zee
2013-04-13 15:26:35 -05:00
parent 1a9f427b85
commit 0020ef7c82
8 changed files with 14 additions and 70 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
frame/3/trsm/ukernels/bli_gemmtrsm_l_ref_mxn.c
View File

@@ -49,13 +49,14 @@ void PASTEMAC(ch,varname)( \
ctype* restrict c, inc_t rs_c, inc_t cs_c \
) \
{ \
const dim_t NR = PASTEMAC2(ch,varname,_nr); \
const dim_t NR = PASTEMAC(ch,nr); \
\
const inc_t rs_b = NR; \
const inc_t cs_b = 1; \
\
ctype* restrict minus_one = PASTEMAC(ch,m1); \
\
/* b = alpha * b - aL * bdT; */ \
PASTEMAC(ch,gemmukr)( k, \
minus_one, \
aL, \
@@ -63,6 +64,9 @@ void PASTEMAC(ch,varname)( \
alpha, \
b, rs_b, cs_b ); \
\
/* b = inv(a) * b;
bd = b; (if gemm ukernel needs duplicated B)
c = b; */ \
PASTEMAC(ch,trsmukr)( a, \
b, \
bd, \

16
frame/3/trsm/ukernels/bli_gemmtrsm_l_ref_mxn.h
View File

@@ -33,22 +33,6 @@
*/
//
// Define micro-kernel blocksizes.
//
#define bli_sgemmtrsm_l_ref_mxn_mr BLIS_DEFAULT_MR_S
#define bli_sgemmtrsm_l_ref_mxn_nr BLIS_DEFAULT_NR_S
#define bli_dgemmtrsm_l_ref_mxn_mr BLIS_DEFAULT_MR_D
#define bli_dgemmtrsm_l_ref_mxn_nr BLIS_DEFAULT_NR_D
#define bli_cgemmtrsm_l_ref_mxn_mr BLIS_DEFAULT_MR_C
#define bli_cgemmtrsm_l_ref_mxn_nr BLIS_DEFAULT_NR_C
#define bli_zgemmtrsm_l_ref_mxn_mr BLIS_DEFAULT_MR_Z
#define bli_zgemmtrsm_l_ref_mxn_nr BLIS_DEFAULT_NR_Z
//
// Prototype micro-kernel interfaces.
//

6
frame/3/trsm/ukernels/bli_gemmtrsm_u_ref_mxn.c
View File

@@ -49,13 +49,14 @@ void PASTEMAC(ch,varname)( \
ctype* restrict c, inc_t rs_c, inc_t cs_c \
) \
{ \
const dim_t NR = PASTEMAC2(ch,varname,_nr); \
const dim_t NR = PASTEMAC(ch,nr); \
\
const inc_t rs_b = NR; \
const inc_t cs_b = 1; \
\
ctype* restrict minus_one = PASTEMAC(ch,m1); \
\
/* b = alpha * b - aR * bdB; */ \
PASTEMAC(ch,gemmukr)( k, \
minus_one, \
aR, \
@@ -63,6 +64,9 @@ void PASTEMAC(ch,varname)( \
alpha, \
b, rs_b, cs_b ); \
\
/* b = inv(a) * b;
bd = b; (if gemm ukernel needs duplicated B)
c = b; */ \
PASTEMAC(ch,trsmukr)( a, \
b, \
bd, \

16
frame/3/trsm/ukernels/bli_gemmtrsm_u_ref_mxn.h
View File

@@ -33,22 +33,6 @@
*/
//
// Define micro-kernel blocksizes.
//
#define bli_sgemmtrsm_u_ref_mxn_mr BLIS_DEFAULT_MR_S
#define bli_sgemmtrsm_u_ref_mxn_nr BLIS_DEFAULT_NR_S
#define bli_dgemmtrsm_u_ref_mxn_mr BLIS_DEFAULT_MR_D
#define bli_dgemmtrsm_u_ref_mxn_nr BLIS_DEFAULT_NR_D
#define bli_cgemmtrsm_u_ref_mxn_mr BLIS_DEFAULT_MR_C
#define bli_cgemmtrsm_u_ref_mxn_nr BLIS_DEFAULT_NR_C
#define bli_zgemmtrsm_u_ref_mxn_mr BLIS_DEFAULT_MR_Z
#define bli_zgemmtrsm_u_ref_mxn_nr BLIS_DEFAULT_NR_Z
//
// Prototype micro-kernel interfaces.
//

4
frame/3/trsm/ukernels/bli_trsm_l_ref_mxn.c
View File

@@ -45,8 +45,8 @@ void PASTEMAC(ch,varname)( \
ctype* restrict c, inc_t rs_c, inc_t cs_c \
) \
{ \
const dim_t MR = PASTEMAC2(ch,varname,_mr); \
const dim_t NR = PASTEMAC2(ch,varname,_nr); \
const dim_t MR = PASTEMAC(ch,mr); \
const dim_t NR = PASTEMAC(ch,nr); \
\
const dim_t m = MR; \
const dim_t n = NR; \

16
frame/3/trsm/ukernels/bli_trsm_l_ref_mxn.h
View File

@@ -33,22 +33,6 @@
*/
//
// Define micro-kernel blocksizes.
//
#define bli_strsm_l_ref_mxn_mr BLIS_DEFAULT_MR_S
#define bli_strsm_l_ref_mxn_nr BLIS_DEFAULT_NR_S
#define bli_dtrsm_l_ref_mxn_mr BLIS_DEFAULT_MR_D
#define bli_dtrsm_l_ref_mxn_nr BLIS_DEFAULT_NR_D
#define bli_ctrsm_l_ref_mxn_mr BLIS_DEFAULT_MR_C
#define bli_ctrsm_l_ref_mxn_nr BLIS_DEFAULT_NR_C
#define bli_ztrsm_l_ref_mxn_mr BLIS_DEFAULT_MR_Z
#define bli_ztrsm_l_ref_mxn_nr BLIS_DEFAULT_NR_Z
//
// Prototype micro-kernel interfaces.
//

4
frame/3/trsm/ukernels/bli_trsm_u_ref_mxn.c
View File

@@ -45,8 +45,8 @@ void PASTEMAC(ch,varname)( \
ctype* restrict c, inc_t rs_c, inc_t cs_c \
) \
{ \
const dim_t MR = PASTEMAC2(ch,varname,_mr); \
const dim_t NR = PASTEMAC2(ch,varname,_nr); \
const dim_t MR = PASTEMAC(ch,mr); \
const dim_t NR = PASTEMAC(ch,nr); \
\
const dim_t m = MR; \
const dim_t n = NR; \

16
frame/3/trsm/ukernels/bli_trsm_u_ref_mxn.h
View File

@@ -33,22 +33,6 @@
*/
//
// Define micro-kernel blocksizes.
//
#define bli_strsm_u_ref_mxn_mr BLIS_DEFAULT_MR_S
#define bli_strsm_u_ref_mxn_nr BLIS_DEFAULT_NR_S
#define bli_dtrsm_u_ref_mxn_mr BLIS_DEFAULT_MR_D
#define bli_dtrsm_u_ref_mxn_nr BLIS_DEFAULT_NR_D
#define bli_ctrsm_u_ref_mxn_mr BLIS_DEFAULT_MR_C
#define bli_ctrsm_u_ref_mxn_nr BLIS_DEFAULT_NR_C
#define bli_ztrsm_u_ref_mxn_mr BLIS_DEFAULT_MR_Z
#define bli_ztrsm_u_ref_mxn_nr BLIS_DEFAULT_NR_Z
//
// Prototype micro-kernel interfaces.
//