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

arm: add softfp support in kernel/arm/swap_vfp.S

This commit is contained in:
Ashwin Sekhar T K 2017-07-01 20:37:40 +05:30
parent e25f4c01d6
commit 83bd547517
2 changed files with 42 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
kernel/arm/KERNEL.ARMV6
View File

@ -57,6 +57,11 @@ DNRM2KERNEL = nrm2_vfp.S
CNRM2KERNEL = nrm2_vfp.S
ZNRM2KERNEL = nrm2_vfp.S
SSWAPKERNEL = swap_vfp.S
DSWAPKERNEL = swap_vfp.S
CSWAPKERNEL = swap_vfp.S
ZSWAPKERNEL = swap_vfp.S
SGEMMKERNEL = ../generic/gemmkernel_4x2.c
ifneq ($(SGEMM_UNROLL_M), $(SGEMM_UNROLL_N))
SGEMMINCOPY = sgemm_ncopy_4_vfp.S
@ -96,11 +101,6 @@ ZGEMMOTCOPYOBJ = zgemm_otcopy.o
ifeq ($(ARM_ABI),hard)
SSWAPKERNEL = swap_vfp.S
DSWAPKERNEL = swap_vfp.S
CSWAPKERNEL = swap_vfp.S
ZSWAPKERNEL = swap_vfp.S
SGEMVNKERNEL = gemv_n_vfp.S
DGEMVNKERNEL = gemv_n_vfp.S
CGEMVNKERNEL = cgemv_n_vfp.S

37
kernel/arm/swap_vfp.S
View File

@ -38,9 +38,43 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define STACKSIZE 256
#if !defined(__ARM_PCS_VFP)
#if !defined(COMPLEX)
#if !defined(DOUBLE)
#define OLD_X [fp, #0 ]
#define OLD_INC_X [fp, #4 ]
#define OLD_Y [fp, #8 ]
#define OLD_INC_Y [fp, #12 ]
#else
#define OLD_X [fp, #8 ]
#define OLD_INC_X [fp, #12]
#define OLD_Y [fp, #16]
#define OLD_INC_Y [fp, #20]
#endif
#else //COMPLEX
#if !defined(DOUBLE)
#define OLD_X [fp, #4 ]
#define OLD_INC_X [fp, #8 ]
#define OLD_Y [fp, #12 ]
#define OLD_INC_Y [fp, #16 ]
#else
#define OLD_X [fp, #16]
#define OLD_INC_X [fp, #20]
#define OLD_Y [fp, #24]
#define OLD_INC_Y [fp, #28]
#endif
#endif // !defined(__ARM_PCS_VFP)
#else
#define OLD_INC_X [fp, #0 ]
#define OLD_Y [fp, #4 ]
#define OLD_INC_Y [fp, #8 ]
#endif
#define N r0
@ -229,6 +263,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
push {r4 , fp}
add fp, sp, #8
#if !defined(__ARM_PCS_VFP)
ldr X, OLD_X
#endif
ldr INC_X , OLD_INC_X
ldr Y, OLD_Y
ldr INC_Y , OLD_INC_Y