floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #4244 from XiWeiGu/loongarch64_fixed_dynamic_arch 

Loongarch64 fixed dynamic arch
This commit is contained in:
Martin Kroeker 2023-09-30 10:16:43 +02:00 committed by GitHub
parent 5ac0ec70a1 211bb35dee
commit 189627709e
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
6 changed files with 24 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
.github/workflows/loongarch64.yml vendored
View File

@ -18,6 +18,9 @@ jobs:
- target: LOONGSON2K1000 - target: LOONGSON2K1000
triple: loongarch64-unknown-linux-gnu triple: loongarch64-unknown-linux-gnu
opts: NO_SHARED=1 TARGET=LOONGSON2K1000 opts: NO_SHARED=1 TARGET=LOONGSON2K1000
- target: DYNAMIC_ARCH
triple: loongarch64-unknown-linux-gnu
opts: NO_SHARED=1 DYNAMIC_ARCH=1 TARGET=GENERIC
steps: steps:
- name: Checkout repository - name: Checkout repository

10
kernel/loongarch64/dgemv_n_8_lasx.S
View File

@ -341,7 +341,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
fmadd.d $f10, $f12, $f2, $f10 fmadd.d $f10, $f12, $f2, $f10
.endm .endm
.macro DGEMV_N XW:req, X_8:req, X_4:req, X_2:req, X_1:req, Y_8:req, Y_4:req, Y_1:req .macro DGEMV_N_LASX XW:req, X_8:req, X_4:req, X_2:req, X_1:req, Y_8:req, Y_4:req, Y_1:req
PTR_SRLI J, N, 3 PTR_SRLI J, N, 3
beqz J, .L_\XW\()_N_7 beqz J, .L_\XW\()_N_7
PTR_SLLI K_LDA, LDA, 3 PTR_SLLI K_LDA, LDA, 3
@ -541,13 +541,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
.hword .L_GAP_1_0 - .L_GAP_TABLE .hword .L_GAP_1_0 - .L_GAP_TABLE
.hword .L_GAP_1_1 - .L_GAP_TABLE .hword .L_GAP_1_1 - .L_GAP_TABLE
.L_GAP_0_0: /* if (inc_x == 1) && (incy == 1) */ .L_GAP_0_0: /* if (inc_x == 1) && (incy == 1) */
DGEMV_N GAP_0_0, X_8, X_4, X_2, X_1, Y_8, Y_4, Y_1 DGEMV_N_LASX GAP_0_0, X_8, X_4, X_2, X_1, Y_8, Y_4, Y_1
.L_GAP_0_1: /* if (inc_x == 1) && (incy != 1) */ .L_GAP_0_1: /* if (inc_x == 1) && (incy != 1) */
DGEMV_N GAP_0_1, X_8, X_4, X_2, X_1, Y_8_GAP, Y_4_GAP, Y_1 DGEMV_N_LASX GAP_0_1, X_8, X_4, X_2, X_1, Y_8_GAP, Y_4_GAP, Y_1
.L_GAP_1_0: /* if (inc_x != 1) && (incy == 1) */ .L_GAP_1_0: /* if (inc_x != 1) && (incy == 1) */
DGEMV_N GAP_1_0, X_8_GAP, X_4_GAP, X_2_GAP, X_1, Y_8, Y_4, Y_1 DGEMV_N_LASX GAP_1_0, X_8_GAP, X_4_GAP, X_2_GAP, X_1, Y_8, Y_4, Y_1
.L_GAP_1_1: /* if (inc_x != 1) && (incy != 1) */ .L_GAP_1_1: /* if (inc_x != 1) && (incy != 1) */
DGEMV_N GAP_1_1, X_8_GAP, X_4_GAP, X_2_GAP, X_1, Y_8_GAP, Y_4_GAP, Y_1 DGEMV_N_LASX GAP_1_1, X_8_GAP, X_4_GAP, X_2_GAP, X_1, Y_8_GAP, Y_4_GAP, Y_1
.L_END: .L_END:
pop_if_used 17 + 7, 24 + 4 pop_if_used 17 + 7, 24 + 4
jirl $r0, $r1, 0x0 jirl $r0, $r1, 0x0

6
kernel/loongarch64/dgemv_t_8_lasx.S
View File

@ -220,7 +220,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
GMADD xvf, d, TP0, A0, X0, TP0, TP1, A2, X0, TP1 GMADD xvf, d, TP0, A0, X0, TP0, TP1, A2, X0, TP1
.endm .endm
.macro DGEMV_T XW:req X8:req, X4:req .macro DGEMV_T_LASX XW:req X8:req, X4:req
PTR_SRLI J, N, 3 PTR_SRLI J, N, 3
beqz J, .L_\XW\()_N_7 beqz J, .L_\XW\()_N_7
PTR_SLLI K_LDA, LDA, 3 PTR_SLLI K_LDA, LDA, 3
@ -472,9 +472,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
.hword .L_GAP_0 - .L_GAP_TABLE .hword .L_GAP_0 - .L_GAP_TABLE
.hword .L_GAP_1 - .L_GAP_TABLE .hword .L_GAP_1 - .L_GAP_TABLE
.L_GAP_0: /* if (incx == 1) */ .L_GAP_0: /* if (incx == 1) */
DGEMV_T GAP_0, X8, X4 DGEMV_T_LASX GAP_0, X8, X4
.L_GAP_1: /* if (incx != 1) */ .L_GAP_1: /* if (incx != 1) */
DGEMV_T GAP_1, X8_GAP, X4_GAP DGEMV_T_LASX GAP_1, X8_GAP, X4_GAP
.L_END: .L_END:
pop_if_used 17 + 8, 24 + 3 pop_if_used 17 + 8, 24 + 3
jirl $r0, $r1, 0x0 jirl $r0, $r1, 0x0

10
kernel/loongarch64/sgemv_n_8_lasx.S
View File

@ -274,7 +274,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
GST f, s, Y0_F, Y, 0 GST f, s, Y0_F, Y, 0
.endm .endm
.macro SGEMV_N XW:req, X_8:req, X_4:req, X_2:req, X_1:req, Y_8:req, Y_4:req, Y_1:req .macro SGEMV_N_LASX XW:req, X_8:req, X_4:req, X_2:req, X_1:req, Y_8:req, Y_4:req, Y_1:req
PTR_SRLI J, N, 3 PTR_SRLI J, N, 3
beqz J, .L_\XW\()_N_7 beqz J, .L_\XW\()_N_7
PTR_SLLI K_LDA, LDA, 3 PTR_SLLI K_LDA, LDA, 3
@ -450,13 +450,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
.hword .L_GAP_1_0 - .L_GAP_TABLE .hword .L_GAP_1_0 - .L_GAP_TABLE
.hword .L_GAP_1_1 - .L_GAP_TABLE .hword .L_GAP_1_1 - .L_GAP_TABLE
.L_GAP_0_0: /* if (inc_x == 1) && (incy == 1) */ .L_GAP_0_0: /* if (inc_x == 1) && (incy == 1) */
SGEMV_N GAP_0_0, X_8, X_4, X_2, X_1, Y_8, Y_4, Y_1 SGEMV_N_LASX GAP_0_0, X_8, X_4, X_2, X_1, Y_8, Y_4, Y_1
.L_GAP_0_1: /* if (inc_x == 1) && (incy != 1) */ .L_GAP_0_1: /* if (inc_x == 1) && (incy != 1) */
SGEMV_N GAP_0_1, X_8, X_4, X_2, X_1, Y_8_GAP, Y_4_GAP, Y_1 SGEMV_N_LASX GAP_0_1, X_8, X_4, X_2, X_1, Y_8_GAP, Y_4_GAP, Y_1
.L_GAP_1_0: /* if (inc_x != 1) && (incy == 1) */ .L_GAP_1_0: /* if (inc_x != 1) && (incy == 1) */
SGEMV_N GAP_1_0, X_8_GAP, X_4_GAP, X_2_GAP, X_1, Y_8, Y_4, Y_1 SGEMV_N_LASX GAP_1_0, X_8_GAP, X_4_GAP, X_2_GAP, X_1, Y_8, Y_4, Y_1
.L_GAP_1_1: /* if (inc_x != 1) && (incy != 1) */ .L_GAP_1_1: /* if (inc_x != 1) && (incy != 1) */
SGEMV_N GAP_1_1, X_8_GAP, X_4_GAP, X_2_GAP, X_1, Y_8_GAP, Y_4_GAP, Y_1 SGEMV_N_LASX GAP_1_1, X_8_GAP, X_4_GAP, X_2_GAP, X_1, Y_8_GAP, Y_4_GAP, Y_1
.L_END: .L_END:
pop_if_used 17 + 7, 19 pop_if_used 17 + 7, 19
jirl $r0, $r1, 0x0 jirl $r0, $r1, 0x0

6
kernel/loongarch64/sgemv_t_8_lasx.S
View File

@ -160,7 +160,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
GMADD xvf, s, TP0, A0, X0, TP0, TP1, A1, X0, TP1 GMADD xvf, s, TP0, A0, X0, TP0, TP1, A1, X0, TP1
.endm .endm
.macro SGEMV_T XW:req X8:req, X4:req .macro SGEMV_T_LASX XW:req X8:req, X4:req
PTR_SRLI J, N, 3 PTR_SRLI J, N, 3
beqz J, .L_\XW\()_N_7 beqz J, .L_\XW\()_N_7
PTR_SLLI K_LDA, LDA, 3 PTR_SLLI K_LDA, LDA, 3
@ -396,9 +396,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
.hword .L_GAP_0 - .L_GAP_TABLE .hword .L_GAP_0 - .L_GAP_TABLE
.hword .L_GAP_1 - .L_GAP_TABLE .hword .L_GAP_1 - .L_GAP_TABLE
.L_GAP_0: /* if (incx == 1) */ .L_GAP_0: /* if (incx == 1) */
SGEMV_T GAP_0, X8, X4 SGEMV_T_LASX GAP_0, X8, X4
.L_GAP_1: /* if (incx != 1) */ .L_GAP_1: /* if (incx != 1) */
SGEMV_T GAP_1, X8_GAP, X4_GAP SGEMV_T_LASX GAP_1, X8_GAP, X4_GAP
.L_END: .L_END:
pop_if_used 17 + 8, 18 pop_if_used 17 + 8, 18
jirl $r0, $r1, 0x0 jirl $r0, $r1, 0x0

5
lapack/laswp/loongarch64/Makefile
View File

@ -1,6 +1,11 @@
TOPDIR = ../../.. TOPDIR = ../../..
include ../../../Makefile.system include ../../../Makefile.system
ifeq ($(DYNAMIC_ARCH), 1)
LASWP = ../generic/laswp_k_4.c
ZLASWP = ../generic/zlaswp_k_4.c
endif
ifndef LASWP ifndef LASWP
LASWP = ../generic/laswp_k.c LASWP = ../generic/laswp_k.c
endif endif