Skip to content

Commit b6e24b2

Browse files
fgvanzeefgvanzee
committed
Use PASTEMAC in macro-kernels (over MAC2 or MAC3).
Details: - Replaced multi-type invocations of copys_mxn, xpbys_mxn, etc. (PASTEMAC2 and PASTEMAC3) with those that only use a single type (PASTEMAC). - Added extra macros to bli_adds_mxn_uplo.h and bli_xpbys_mxn_uplo.h to accommodate above change. - Fixed comment typo in bli_config.h files. - Added .nfs* pattern to .gitignore.
1 parent df80acf commit b6e24b2

File tree

13 files changed

+132
-65
lines changed

13 files changed

+132
-65
lines changed

‎.gitignore

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -4,6 +4,8 @@
44
*~
55
# vim backup files
66
*.swp
7+
# NFS file
8+
.nfs*
79

810
# -- compiler-related --
911

‎config/clarksville/bli_kernel.h

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -77,7 +77,7 @@
7777
#define BLIS_DEFAULT_KC_Z 256
7878
#define BLIS_DEFAULT_NC_Z 2048
7979

80-
// -- Ccache blocksize extensions (for optimizing edge cases) --
80+
// -- Cache blocksize extensions (for optimizing edge cases) --
8181

8282
// NOTE: These cache blocksize "extensions" have the same constraints as
8383
// the corresponding default blocksizes above.

‎config/flame/bli_kernel.h

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -77,7 +77,7 @@
7777
#define BLIS_DEFAULT_KC_Z 256
7878
#define BLIS_DEFAULT_NC_Z 2048
7979

80-
// -- Ccache blocksize extensions (for optimizing edge cases) --
80+
// -- Cache blocksize extensions (for optimizing edge cases) --
8181

8282
// NOTE: These cache blocksize "extensions" have the same constraints as
8383
// the corresponding default blocksizes above.

‎config/reference/bli_kernel.h

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -77,7 +77,7 @@
7777
#define BLIS_DEFAULT_KC_Z 256
7878
#define BLIS_DEFAULT_NC_Z 2048
7979

80-
// -- Ccache blocksize extensions (for optimizing edge cases) --
80+
// -- Cache blocksize extensions (for optimizing edge cases) --
8181

8282
// NOTE: These cache blocksize "extensions" have the same constraints as
8383
// the corresponding default blocksizes above.

‎frame/3/gemm/bli_gemm_ker_var2.c

Lines changed: 12 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -291,10 +291,10 @@ void PASTEMAC(ch,varname)( \
291291
a2, b2 ); \
292292
\
293293
/* Scale the bottom edge of C and add the result from above. */ \
294-
PASTEMAC3(ch,ch,ch,xpbys_mxn)( m_left, NR, \
295-
ct, rs_ct, cs_ct, \
296-
beta_cast, \
297-
c11, rs_c, cs_c ); \
294+
PASTEMAC(ch,xpbys_mxn)( m_left, NR, \
295+
ct, rs_ct, cs_ct, \
296+
beta_cast, \
297+
c11, rs_c, cs_c ); \
298298
} \
299299
\
300300
b1 += cstep_b; \
@@ -335,10 +335,10 @@ void PASTEMAC(ch,varname)( \
335335
a2, b2 ); \
336336
\
337337
/* Scale the right edge of C and add the result from above. */ \
338-
PASTEMAC3(ch,ch,ch,xpbys_mxn)( MR, n_left, \
339-
ct, rs_ct, cs_ct, \
340-
beta_cast, \
341-
c11, rs_c, cs_c ); \
338+
PASTEMAC(ch,xpbys_mxn)( MR, n_left, \
339+
ct, rs_ct, cs_ct, \
340+
beta_cast, \
341+
c11, rs_c, cs_c ); \
342342
\
343343
a1 += rstep_a; \
344344
c11 += rstep_c; \
@@ -361,10 +361,10 @@ void PASTEMAC(ch,varname)( \
361361
a2, b2 ); \
362362
\
363363
/* Scale the bottom-right corner of C and add the result from above. */ \
364-
PASTEMAC3(ch,ch,ch,xpbys_mxn)( m_left, n_left, \
365-
ct, rs_ct, cs_ct, \
366-
beta_cast, \
367-
c11, rs_c, cs_c ); \
364+
PASTEMAC(ch,xpbys_mxn)( m_left, n_left, \
365+
ct, rs_ct, cs_ct, \
366+
beta_cast, \
367+
c11, rs_c, cs_c ); \
368368
} \
369369
} \
370370
\

‎frame/3/herk/bli_herk_l_ker_var2.c

Lines changed: 9 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -294,11 +294,11 @@ void PASTEMAC(ch,varname)( \
294294
a2, b2 ); \
295295
\
296296
/* Scale C and add the result to only the stored part. */ \
297-
PASTEMAC3(ch,ch,ch,xpbys_mxn_l)( diagoffc_ij, \
298-
m_cur, n_cur, \
299-
ct, rs_ct, cs_ct, \
300-
beta_cast, \
301-
c11, rs_c, cs_c ); \
297+
PASTEMAC(ch,xpbys_mxn_l)( diagoffc_ij, \
298+
m_cur, n_cur, \
299+
ct, rs_ct, cs_ct, \
300+
beta_cast, \
301+
c11, rs_c, cs_c ); \
302302
} \
303303
else if ( bli_is_strictly_below_diag_n( diagoffc_ij, m_cur, n_cur ) ) \
304304
{ \
@@ -326,10 +326,10 @@ void PASTEMAC(ch,varname)( \
326326
a2, b2 ); \
327327
\
328328
/* Scale the edge of C and add the result. */ \
329-
PASTEMAC3(ch,ch,ch,xpbys_mxn)( m_cur, n_cur, \
330-
ct, rs_ct, cs_ct, \
331-
beta_cast, \
332-
c11, rs_c, cs_c ); \
329+
PASTEMAC(ch,xpbys_mxn)( m_cur, n_cur, \
330+
ct, rs_ct, cs_ct, \
331+
beta_cast, \
332+
c11, rs_c, cs_c ); \
333333
} \
334334
} \
335335
\

‎frame/3/herk/bli_herk_u_ker_var2.c

Lines changed: 9 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -294,11 +294,11 @@ void PASTEMAC(ch,varname)( \
294294
a2, b2 ); \
295295
\
296296
/* Scale C and add the result to only the stored part. */ \
297-
PASTEMAC3(ch,ch,ch,xpbys_mxn_u)( diagoffc_ij, \
298-
m_cur, n_cur, \
299-
ct, rs_ct, cs_ct, \
300-
beta_cast, \
301-
c11, rs_c, cs_c ); \
297+
PASTEMAC(ch,xpbys_mxn_u)( diagoffc_ij, \
298+
m_cur, n_cur, \
299+
ct, rs_ct, cs_ct, \
300+
beta_cast, \
301+
c11, rs_c, cs_c ); \
302302
} \
303303
else if ( bli_is_strictly_above_diag_n( diagoffc_ij, m_cur, n_cur ) ) \
304304
{ \
@@ -326,10 +326,10 @@ void PASTEMAC(ch,varname)( \
326326
a2, b2 ); \
327327
\
328328
/* Scale the edge of C and add the result. */ \
329-
PASTEMAC3(ch,ch,ch,xpbys_mxn)( m_cur, n_cur, \
330-
ct, rs_ct, cs_ct, \
331-
beta_cast, \
332-
c11, rs_c, cs_c ); \
329+
PASTEMAC(ch,xpbys_mxn)( m_cur, n_cur, \
330+
ct, rs_ct, cs_ct, \
331+
beta_cast, \
332+
c11, rs_c, cs_c ); \
333333
} \
334334
} \
335335
\

‎frame/3/trmm/bli_trmm_l_ker_var2.c

Lines changed: 9 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -333,9 +333,9 @@ void PASTEMAC(ch,varname)( \
333333
else \
334334
{ \
335335
/* Copy edge elements of C to the temporary buffer. */ \
336-
PASTEMAC2(ch,ch,copys_mxn)( m_cur, n_cur, \
337-
c11, rs_c, cs_c, \
338-
ct, rs_ct, cs_ct ); \
336+
PASTEMAC(ch,copys_mxn)( m_cur, n_cur, \
337+
c11, rs_c, cs_c, \
338+
ct, rs_ct, cs_ct ); \
339339
\
340340
/* Invoke the gemm micro-kernel. */ \
341341
PASTEMAC(ch,ukrname)( k_a1011, \
@@ -347,9 +347,9 @@ void PASTEMAC(ch,varname)( \
347347
a2, b2 ); \
348348
\
349349
/* Copy the result to the edge of C. */ \
350-
PASTEMAC2(ch,ch,copys_mxn)( m_cur, n_cur, \
351-
ct, rs_ct, cs_ct, \
352-
c11, rs_c, cs_c ); \
350+
PASTEMAC(ch,copys_mxn)( m_cur, n_cur, \
351+
ct, rs_ct, cs_ct, \
352+
c11, rs_c, cs_c ); \
353353
} \
354354
\
355355
a1 += k_a1011 * PACKMR; \
@@ -390,9 +390,9 @@ void PASTEMAC(ch,varname)( \
390390
a2, b2 ); \
391391
\
392392
/* Add the result to the edge of C. */ \
393-
PASTEMAC2(ch,ch,adds_mxn)( m_cur, n_cur, \
394-
ct, rs_ct, cs_ct, \
395-
c11, rs_c, cs_c ); \
393+
PASTEMAC(ch,adds_mxn)( m_cur, n_cur, \
394+
ct, rs_ct, cs_ct, \
395+
c11, rs_c, cs_c ); \
396396
} \
397397
\
398398
a1 += rstep_a; \

‎frame/3/trmm/bli_trmm_u_ker_var2.c

Lines changed: 9 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -333,9 +333,9 @@ void PASTEMAC(ch,varname)( \
333333
else \
334334
{ \
335335
/* Copy edge elements of C to the temporary buffer. */ \
336-
PASTEMAC2(ch,ch,copys_mxn)( m_cur, n_cur, \
337-
c11, rs_c, cs_c, \
338-
ct, rs_ct, cs_ct ); \
336+
PASTEMAC(ch,copys_mxn)( m_cur, n_cur, \
337+
c11, rs_c, cs_c, \
338+
ct, rs_ct, cs_ct ); \
339339
\
340340
/* Invoke the gemm micro-kernel. */ \
341341
PASTEMAC(ch,ukrname)( k_a1112, \
@@ -347,9 +347,9 @@ void PASTEMAC(ch,varname)( \
347347
a2, b2 ); \
348348
\
349349
/* Copy the result to the edge of C. */ \
350-
PASTEMAC2(ch,ch,copys_mxn)( m_cur, n_cur, \
351-
ct, rs_ct, cs_ct, \
352-
c11, rs_c, cs_c ); \
350+
PASTEMAC(ch,copys_mxn)( m_cur, n_cur, \
351+
ct, rs_ct, cs_ct, \
352+
c11, rs_c, cs_c ); \
353353
} \
354354
\
355355
a1 += k_a1112 * PACKMR; \
@@ -390,9 +390,9 @@ void PASTEMAC(ch,varname)( \
390390
a2, b2 ); \
391391
\
392392
/* Add the result to the edge of C. */ \
393-
PASTEMAC2(ch,ch,adds_mxn)( m_cur, n_cur, \
394-
ct, rs_ct, cs_ct, \
395-
c11, rs_c, cs_c ); \
393+
PASTEMAC(ch,adds_mxn)( m_cur, n_cur, \
394+
ct, rs_ct, cs_ct, \
395+
c11, rs_c, cs_c ); \
396396
} \
397397
\
398398
a1 += rstep_a; \

‎frame/3/trsm/bli_trsm_l_ker_var2.c

Lines changed: 7 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -361,9 +361,9 @@ void PASTEMAC(ch,varname)( \
361361
a2, b2 ); \
362362
\
363363
/* Copy the result to the bottom edge of C. */ \
364-
PASTEMAC2(ch,ch,copys_mxn)( m_cur, n_cur, \
365-
ct, rs_ct, cs_ct, \
366-
c11, rs_c, cs_c ); \
364+
PASTEMAC(ch,copys_mxn)( m_cur, n_cur, \
365+
ct, rs_ct, cs_ct, \
366+
c11, rs_c, cs_c ); \
367367
} \
368368
\
369369
a1 += k_a1011 * PACKMR; \
@@ -404,10 +404,10 @@ void PASTEMAC(ch,varname)( \
404404
a2, b2 ); \
405405
\
406406
/* Add the result to the edge of C. */ \
407-
PASTEMAC3(ch,ch,ch,xpbys_mxn)( m_cur, n_cur, \
408-
ct, rs_ct, cs_ct, \
409-
alpha_cast, \
410-
c11, rs_c, cs_c ); \
407+
PASTEMAC(ch,xpbys_mxn)( m_cur, n_cur, \
408+
ct, rs_ct, cs_ct, \
409+
alpha_cast, \
410+
c11, rs_c, cs_c ); \
411411
} \
412412
\
413413
a1 += rstep_a; \

0 commit comments

Comments
 (0)