From 9588625c43c86ef1bde8140f620a30f52420e6a6 Mon Sep 17 00:00:00 2001 From: "Field G. Van Zee" Date: Wed, 30 May 2018 15:19:53 -0500 Subject: [PATCH] Renamed "next micropanel" macros in _l3_thrinfo.h. Details: - Renamed several macros defined in bli_l3_thrinfo.h designed to compute the values of a_next and b_next to insert into an auxinfo_t struct in level-3 macrokernels. (Previously, the macros did not use a bli_ prefix.) - Updated instances of above macro usage within various macrokernels. --- frame/3/bli_l3_thrinfo.h | 18 +++++++++--------- frame/3/gemm/bli_gemm_ker_var2.c | 4 ++-- frame/3/gemm/ind/bli_gemm4mb_ker_var2.c | 4 ++-- frame/3/gemm/ind/old/bli_gemm3m2_ker_var2.c | 4 ++-- frame/3/herk/bli_herk_l_ker_var2.c | 4 ++-- frame/3/herk/bli_herk_u_ker_var2.c | 4 ++-- frame/3/trmm/bli_trmm_ll_ker_var2.c | 6 +++--- frame/3/trmm/bli_trmm_lu_ker_var2.c | 6 +++--- frame/3/trmm/bli_trmm_rl_ker_var2.c | 8 ++++---- frame/3/trmm/bli_trmm_ru_ker_var2.c | 8 ++++---- frame/3/trsm/bli_trsm_ll_ker_var2.c | 2 +- frame/3/trsm/bli_trsm_lu_ker_var2.c | 2 +- frame/3/trsm/bli_trsm_rl_ker_var2.c | 4 ++-- frame/3/trsm/bli_trsm_ru_ker_var2.c | 4 ++-- 14 files changed, 39 insertions(+), 39 deletions(-) diff --git a/frame/3/bli_l3_thrinfo.h b/frame/3/bli_l3_thrinfo.h index fcf1f507d..58733bcf5 100644 --- a/frame/3/bli_l3_thrinfo.h +++ b/frame/3/bli_l3_thrinfo.h @@ -38,24 +38,24 @@ // gemm -#define gemm_get_next_a_micropanel( thread, a1, step ) ( a1 + step * thread->n_way ) -#define gemm_get_next_b_micropanel( thread, b1, step ) ( b1 + step * thread->n_way ) +#define bli_gemm_get_next_a_upanel( thread, a1, step ) ( a1 + step * thread->n_way ) +#define bli_gemm_get_next_b_upanel( thread, b1, step ) ( b1 + step * thread->n_way ) // herk -#define herk_get_next_a_micropanel( thread, a1, step ) ( a1 + step * thread->n_way ) -#define herk_get_next_b_micropanel( thread, b1, step ) ( b1 + step * thread->n_way ) +#define bli_herk_get_next_a_upanel( thread, a1, step ) ( a1 + step * thread->n_way ) +#define bli_herk_get_next_b_upanel( thread, b1, step ) ( b1 + step * thread->n_way ) // trmm -#define trmm_r_ir_my_iter( index, thread ) ( index % thread->n_way == thread->work_id % thread->n_way ) -#define trmm_r_jr_my_iter( index, thread ) ( index % thread->n_way == thread->work_id % thread->n_way ) -#define trmm_l_ir_my_iter( index, thread ) ( index % thread->n_way == thread->work_id % thread->n_way ) -#define trmm_l_jr_my_iter( index, thread ) ( index % thread->n_way == thread->work_id % thread->n_way ) +#define bli_trmm_r_ir_my_iter( index, thread ) ( index % thread->n_way == thread->work_id % thread->n_way ) +#define bli_trmm_r_jr_my_iter( index, thread ) ( index % thread->n_way == thread->work_id % thread->n_way ) +#define bli_trmm_l_ir_my_iter( index, thread ) ( index % thread->n_way == thread->work_id % thread->n_way ) +#define bli_trmm_l_jr_my_iter( index, thread ) ( index % thread->n_way == thread->work_id % thread->n_way ) // trsm -#define trsm_my_iter( index, thread ) ( index % thread->n_way == thread->work_id % thread->n_way ) +#define bli_trsm_my_iter( index, thread ) ( index % thread->n_way == thread->work_id % thread->n_way ) // // thrinfo_t APIs specific to level-3 operations. diff --git a/frame/3/gemm/bli_gemm_ker_var2.c b/frame/3/gemm/bli_gemm_ker_var2.c index 3f54ef031..2e05deaad 100644 --- a/frame/3/gemm/bli_gemm_ker_var2.c +++ b/frame/3/gemm/bli_gemm_ker_var2.c @@ -295,11 +295,11 @@ void PASTEMAC(ch,varname) \ m_cur = ( bli_is_not_edge_f( i, m_iter, m_left ) ? MR : m_left ); \ \ /* Compute the addresses of the next panels of A and B. */ \ - a2 = gemm_get_next_a_micropanel( caucus, a1, rstep_a ); \ + a2 = bli_gemm_get_next_a_upanel( caucus, a1, rstep_a ); \ if ( bli_is_last_iter( i, m_iter, ir_thread_id, ir_num_threads ) ) \ { \ a2 = a_cast; \ - b2 = gemm_get_next_b_micropanel( thread, b1, cstep_b ); \ + b2 = bli_gemm_get_next_b_upanel( thread, b1, cstep_b ); \ if ( bli_is_last_iter( j, n_iter, jr_thread_id, jr_num_threads ) ) \ b2 = b_cast; \ } \ diff --git a/frame/3/gemm/ind/bli_gemm4mb_ker_var2.c b/frame/3/gemm/ind/bli_gemm4mb_ker_var2.c index c57144392..7cb809904 100644 --- a/frame/3/gemm/ind/bli_gemm4mb_ker_var2.c +++ b/frame/3/gemm/ind/bli_gemm4mb_ker_var2.c @@ -291,11 +291,11 @@ void PASTEMAC(ch,varname) \ m_cur = ( bli_is_not_edge_f( i, m_iter, m_left ) ? MR : m_left ); \ \ /* Compute the addresses of the next panels of A and B. */ \ - a2 = gemm_get_next_a_micropanel( caucus, a1, rstep_a ); \ + a2 = bli_gemm_get_next_a_upanel( caucus, a1, rstep_a ); \ if ( bli_is_last_iter( i, m_iter, ir_thread_id, ir_num_threads ) ) \ { \ a2 = a_cast; \ - b2 = gemm_get_next_b_micropanel( thread, b1, cstep_b ); \ + b2 = bli_gemm_get_next_b_upanel( thread, b1, cstep_b ); \ if ( bli_is_last_iter( j, n_iter, jr_thread_id, jr_num_threads ) ) \ b2 = b_cast; \ } \ diff --git a/frame/3/gemm/ind/old/bli_gemm3m2_ker_var2.c b/frame/3/gemm/ind/old/bli_gemm3m2_ker_var2.c index 9ecb4cf5a..d711fd946 100644 --- a/frame/3/gemm/ind/old/bli_gemm3m2_ker_var2.c +++ b/frame/3/gemm/ind/old/bli_gemm3m2_ker_var2.c @@ -300,11 +300,11 @@ void PASTEMAC(ch,varname) \ m_cur = ( bli_is_not_edge_f( i, m_iter, m_left ) ? MR : m_left ); \ \ /* Compute the addresses of the next panels of A and B. */ \ - a2 = gemm_get_next_a_micropanel( caucus, a1, rstep_a ); \ + a2 = bli_gemm_get_next_a_upanel( caucus, a1, rstep_a ); \ if ( bli_is_last_iter( i, m_iter, ir_thread_id, ir_num_threads ) ) \ { \ a2 = a_cast; \ - b2 = gemm_get_next_b_micropanel( thread, b1, cstep_b ); \ + b2 = bli_gemm_get_next_b_upanel( thread, b1, cstep_b ); \ if ( bli_is_last_iter( j, n_iter, jr_thread_id, jr_num_threads ) ) \ b2 = b_cast; \ } \ diff --git a/frame/3/herk/bli_herk_l_ker_var2.c b/frame/3/herk/bli_herk_l_ker_var2.c index 05a0e300e..3ee105140 100644 --- a/frame/3/herk/bli_herk_l_ker_var2.c +++ b/frame/3/herk/bli_herk_l_ker_var2.c @@ -312,11 +312,11 @@ void PASTEMAC(ch,varname) \ m_cur = ( bli_is_not_edge_f( i, m_iter, m_left ) ? MR : m_left ); \ \ /* Compute the addresses of the next panels of A and B. */ \ - a2 = herk_get_next_a_micropanel( caucus, a1, rstep_a ); \ + a2 = bli_herk_get_next_a_upanel( caucus, a1, rstep_a ); \ if ( bli_is_last_iter( i, m_iter, ir_thread_id, ir_num_threads ) ) \ { \ a2 = a_cast; \ - b2 = herk_get_next_b_micropanel( thread, b1, cstep_b ); \ + b2 = bli_herk_get_next_b_upanel( thread, b1, cstep_b ); \ if ( bli_is_last_iter( j, n_iter, jr_thread_id, jr_num_threads ) ) \ b2 = b_cast; \ } \ diff --git a/frame/3/herk/bli_herk_u_ker_var2.c b/frame/3/herk/bli_herk_u_ker_var2.c index 51600d839..b58f600e5 100644 --- a/frame/3/herk/bli_herk_u_ker_var2.c +++ b/frame/3/herk/bli_herk_u_ker_var2.c @@ -312,11 +312,11 @@ void PASTEMAC(ch,varname) \ m_cur = ( bli_is_not_edge_f( i, m_iter, m_left ) ? MR : m_left ); \ \ /* Compute the addresses of the next panels of A and B. */ \ - a2 = herk_get_next_a_micropanel( caucus, a1, rstep_a ); \ + a2 = bli_herk_get_next_a_upanel( caucus, a1, rstep_a ); \ if ( bli_is_last_iter( i, m_iter, ir_thread_id, ir_num_threads ) ) \ { \ a2 = a_cast; \ - b2 = herk_get_next_b_micropanel( thread, b1, cstep_b ); \ + b2 = bli_herk_get_next_b_upanel( thread, b1, cstep_b ); \ if ( bli_is_last_iter( j, n_iter, jr_thread_id, jr_num_threads ) ) \ b2 = b_cast; \ } \ diff --git a/frame/3/trmm/bli_trmm_ll_ker_var2.c b/frame/3/trmm/bli_trmm_ll_ker_var2.c index 94dd233b0..8d5c4d0f0 100644 --- a/frame/3/trmm/bli_trmm_ll_ker_var2.c +++ b/frame/3/trmm/bli_trmm_ll_ker_var2.c @@ -322,7 +322,7 @@ void PASTEMAC(ch,varname) \ /* Loop over the n dimension (NR columns at a time). */ \ for ( j = 0; j < n_iter; ++j ) \ { \ - if ( trmm_l_jr_my_iter( j, jr_thread ) ) { \ + if ( bli_trmm_l_jr_my_iter( j, jr_thread ) ) { \ \ ctype* restrict a1; \ ctype* restrict c11; \ @@ -364,7 +364,7 @@ void PASTEMAC(ch,varname) \ is_a_cur += ( bli_is_odd( is_a_cur ) ? 1 : 0 ); \ ps_a_cur = ( is_a_cur * ss_a_num ) / ss_a_den; \ \ - if ( trmm_l_ir_my_iter( i, ir_thread ) ) { \ + if ( bli_trmm_l_ir_my_iter( i, ir_thread ) ) { \ \ b1_i = b1 + ( off_a1011 * PACKNR ) / off_scl; \ \ @@ -434,7 +434,7 @@ void PASTEMAC(ch,varname) \ } \ else if ( bli_is_strictly_below_diag_n( diagoffa_i, MR, k ) ) \ { \ - if ( trmm_l_ir_my_iter( i, ir_thread ) ) { \ + if ( bli_trmm_l_ir_my_iter( i, ir_thread ) ) { \ \ ctype* restrict a2; \ \ diff --git a/frame/3/trmm/bli_trmm_lu_ker_var2.c b/frame/3/trmm/bli_trmm_lu_ker_var2.c index 924e75d4f..e54d7d582 100644 --- a/frame/3/trmm/bli_trmm_lu_ker_var2.c +++ b/frame/3/trmm/bli_trmm_lu_ker_var2.c @@ -329,7 +329,7 @@ void PASTEMAC(ch,varname) \ /* Loop over the n dimension (NR columns at a time). */ \ for ( j = 0; j < n_iter; ++j ) \ { \ - if ( trmm_l_jr_my_iter( j, jr_thread ) ) { \ + if ( bli_trmm_l_jr_my_iter( j, jr_thread ) ) { \ \ ctype* restrict a1; \ ctype* restrict c11; \ @@ -371,7 +371,7 @@ void PASTEMAC(ch,varname) \ is_a_cur += ( bli_is_odd( is_a_cur ) ? 1 : 0 ); \ ps_a_cur = ( is_a_cur * ss_a_num ) / ss_a_den; \ \ - if ( trmm_l_ir_my_iter( i, ir_thread ) ) { \ + if ( bli_trmm_l_ir_my_iter( i, ir_thread ) ) { \ \ b1_i = b1 + ( off_a1112 * PACKNR ) / off_scl; \ \ @@ -441,7 +441,7 @@ void PASTEMAC(ch,varname) \ } \ else if ( bli_is_strictly_above_diag_n( diagoffa_i, MR, k ) ) \ { \ - if ( trmm_l_ir_my_iter( i, ir_thread ) ) { \ + if ( bli_trmm_l_ir_my_iter( i, ir_thread ) ) { \ \ ctype* restrict a2; \ \ diff --git a/frame/3/trmm/bli_trmm_rl_ker_var2.c b/frame/3/trmm/bli_trmm_rl_ker_var2.c index 26ef9a13e..919eab1a3 100644 --- a/frame/3/trmm/bli_trmm_rl_ker_var2.c +++ b/frame/3/trmm/bli_trmm_rl_ker_var2.c @@ -361,7 +361,7 @@ void PASTEMAC(ch,varname) \ is_b_cur += ( bli_is_odd( is_b_cur ) ? 1 : 0 ); \ ps_b_cur = ( is_b_cur * ss_b_num ) / ss_b_den; \ \ - if ( trmm_r_jr_my_iter( j, jr_thread ) ) { \ + if ( bli_trmm_r_jr_my_iter( j, jr_thread ) ) { \ \ /* Save the 4m1/3m1 imaginary stride of B to the auxinfo_t object. */ \ @@ -370,7 +370,7 @@ void PASTEMAC(ch,varname) \ /* Loop over the m dimension (MR rows at a time). */ \ for ( i = 0; i < m_iter; ++i ) \ { \ - if ( trmm_r_ir_my_iter( i, ir_thread ) ) { \ + if ( bli_trmm_r_ir_my_iter( i, ir_thread ) ) { \ \ ctype* restrict a1_i; \ ctype* restrict a2; \ @@ -446,7 +446,7 @@ void PASTEMAC(ch,varname) \ } \ else if ( bli_is_strictly_below_diag_n( diagoffb_j, k, NR ) ) \ { \ - if ( trmm_r_jr_my_iter( j, jr_thread ) ) { \ + if ( bli_trmm_r_jr_my_iter( j, jr_thread ) ) { \ \ /* Save the 4m1/3m1 imaginary stride of B to the auxinfo_t object. */ \ @@ -455,7 +455,7 @@ void PASTEMAC(ch,varname) \ /* Loop over the m dimension (MR rows at a time). */ \ for ( i = 0; i < m_iter; ++i ) \ { \ - if ( trmm_r_ir_my_iter( i, ir_thread ) ) { \ + if ( bli_trmm_r_ir_my_iter( i, ir_thread ) ) { \ \ ctype* restrict a2; \ \ diff --git a/frame/3/trmm/bli_trmm_ru_ker_var2.c b/frame/3/trmm/bli_trmm_ru_ker_var2.c index 4c9af0757..bc4907772 100644 --- a/frame/3/trmm/bli_trmm_ru_ker_var2.c +++ b/frame/3/trmm/bli_trmm_ru_ker_var2.c @@ -361,7 +361,7 @@ void PASTEMAC(ch,varname) \ is_b_cur += ( bli_is_odd( is_b_cur ) ? 1 : 0 ); \ ps_b_cur = ( is_b_cur * ss_b_num ) / ss_b_den; \ \ - if ( trmm_r_jr_my_iter( j, jr_thread ) ) { \ + if ( bli_trmm_r_jr_my_iter( j, jr_thread ) ) { \ \ /* Save the 4m1/3m1 imaginary stride of B to the auxinfo_t object. */ \ @@ -370,7 +370,7 @@ void PASTEMAC(ch,varname) \ /* Loop over the m dimension (MR rows at a time). */ \ for ( i = 0; i < m_iter; ++i ) \ { \ - if ( trmm_r_ir_my_iter( i, ir_thread ) ) { \ + if ( bli_trmm_r_ir_my_iter( i, ir_thread ) ) { \ \ ctype* restrict a1_i; \ ctype* restrict a2; \ @@ -446,7 +446,7 @@ void PASTEMAC(ch,varname) \ } \ else if ( bli_is_strictly_above_diag_n( diagoffb_j, k, NR ) ) \ { \ - if ( trmm_r_jr_my_iter( j, jr_thread ) ) { \ + if ( bli_trmm_r_jr_my_iter( j, jr_thread ) ) { \ \ /* Save the 4m1/3m1 imaginary stride of B to the auxinfo_t object. */ \ @@ -455,7 +455,7 @@ void PASTEMAC(ch,varname) \ /* Loop over the m dimension (MR rows at a time). */ \ for ( i = 0; i < m_iter; ++i ) \ { \ - if ( trmm_r_ir_my_iter( i, ir_thread ) ) { \ + if ( bli_trmm_r_ir_my_iter( i, ir_thread ) ) { \ \ ctype* restrict a2; \ \ diff --git a/frame/3/trsm/bli_trsm_ll_ker_var2.c b/frame/3/trsm/bli_trsm_ll_ker_var2.c index 17c4b6d7a..bf7f647de 100644 --- a/frame/3/trsm/bli_trsm_ll_ker_var2.c +++ b/frame/3/trsm/bli_trsm_ll_ker_var2.c @@ -340,7 +340,7 @@ void PASTEMAC(ch,varname) \ /* Loop over the n dimension (NR columns at a time). */ \ for ( j = 0; j < n_iter; ++j ) \ { \ - if( trsm_my_iter( j, thread ) ) { \ + if( bli_trsm_my_iter( j, thread ) ) { \ \ ctype* restrict a1; \ ctype* restrict c11; \ diff --git a/frame/3/trsm/bli_trsm_lu_ker_var2.c b/frame/3/trsm/bli_trsm_lu_ker_var2.c index f12bbb194..4b3c4c4b3 100644 --- a/frame/3/trsm/bli_trsm_lu_ker_var2.c +++ b/frame/3/trsm/bli_trsm_lu_ker_var2.c @@ -348,7 +348,7 @@ void PASTEMAC(ch,varname) \ /* Loop over the n dimension (NR columns at a time). */ \ for ( j = 0; j < n_iter; ++j ) \ { \ - if( trsm_my_iter( j, thread ) ) { \ + if( bli_trsm_my_iter( j, thread ) ) { \ \ ctype* restrict a1; \ ctype* restrict c11; \ diff --git a/frame/3/trsm/bli_trsm_rl_ker_var2.c b/frame/3/trsm/bli_trsm_rl_ker_var2.c index be7904936..7f4b93bd3 100644 --- a/frame/3/trsm/bli_trsm_rl_ker_var2.c +++ b/frame/3/trsm/bli_trsm_rl_ker_var2.c @@ -422,7 +422,7 @@ void PASTEMAC(ch,varname) \ /* Loop over the m dimension (MR rows at a time). */ \ for ( i = 0; i < m_iter; ++i ) \ { \ - if( trsm_my_iter( i, thread ) ){ \ + if( bli_trsm_my_iter( i, thread ) ){ \ \ ctype* restrict a11; \ ctype* restrict a12; \ @@ -508,7 +508,7 @@ void PASTEMAC(ch,varname) \ /* Loop over the m dimension (MR rows at a time). */ \ for ( i = 0; i < m_iter; ++i ) \ { \ - if( trsm_my_iter( i, thread ) ){ \ + if( bli_trsm_my_iter( i, thread ) ){ \ \ ctype* restrict a2; \ \ diff --git a/frame/3/trsm/bli_trsm_ru_ker_var2.c b/frame/3/trsm/bli_trsm_ru_ker_var2.c index e1f2694b0..d91b4b0d0 100644 --- a/frame/3/trsm/bli_trsm_ru_ker_var2.c +++ b/frame/3/trsm/bli_trsm_ru_ker_var2.c @@ -415,7 +415,7 @@ void PASTEMAC(ch,varname) \ /* Loop over the m dimension (MR rows at a time). */ \ for ( i = 0; i < m_iter; ++i ) \ { \ - if( trsm_my_iter( i, thread ) ){ \ + if( bli_trsm_my_iter( i, thread ) ){ \ \ ctype* restrict a10; \ ctype* restrict a11; \ @@ -501,7 +501,7 @@ void PASTEMAC(ch,varname) \ /* Loop over the m dimension (MR rows at a time). */ \ for ( i = 0; i < m_iter; ++i ) \ { \ - if( trsm_my_iter( i, thread ) ){ \ + if( bli_trsm_my_iter( i, thread ) ){ \ \ ctype* restrict a2; \ \