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

Merge pull request #2280 from martin-frbg/iosfix 

Add overlooked part of IOS compilation fix
This commit is contained in:
Martin Kroeker 2019-10-08 10:25:25 +02:00 committed by GitHub
parent 3a2df19db6 a448884a63
commit ba7838d2e1
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 19 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

38
kernel/arm64/znrm2.S
View File

@ -123,69 +123,69 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#if !defined(DOUBLE) #if !defined(DOUBLE)
ldr s4, [X] ldr s4, [X]
fcmp s4, REGZERO fcmp s4, REGZERO
beq KERNEL_S1_NEXT_\@ beq KERNEL_S1_NEXT
fabs s4, s4 fabs s4, s4
fcmp SCALE, s4 fcmp SCALE, s4
bge KERNEL_S1_SCALE_GE_XR_\@ bge KERNEL_S1_SCALE_GE_XR
fdiv s2, SCALE, s4 fdiv s2, SCALE, s4
fmul s2, s2, s2 fmul s2, s2, s2
fmul s3, SSQ, s2 fmul s3, SSQ, s2
fadd SSQ, REGONE, s3 fadd SSQ, REGONE, s3
fmov SCALE, s4 fmov SCALE, s4
b KERNEL_S1_NEXT_\@ b KERNEL_S1_NEXT
KERNEL_S1_SCALE_GE_XR_\@: KERNEL_S1_SCALE_GE_XR:
fdiv s2, s4, SCALE fdiv s2, s4, SCALE
fmla SSQ, s2, v2.s[0] fmla SSQ, s2, v2.s[0]
KERNEL_S1_NEXT_\@: KERNEL_S1_NEXT:
ldr s5, [X, #4] ldr s5, [X, #4]
fcmp s5, REGZERO fcmp s5, REGZERO
beq KERNEL_S1_END_\@ beq KERNEL_S1_END
fabs s5, s5 fabs s5, s5
fcmp SCALE, s5 fcmp SCALE, s5
bge KERNEL_S1_SCALE_GE_XI_\@ bge KERNEL_S1_SCALE_GE_XI
fdiv s2, SCALE, s5 fdiv s2, SCALE, s5
fmul s2, s2, s2 fmul s2, s2, s2
fmul s3, SSQ, s2 fmul s3, SSQ, s2
fadd SSQ, REGONE, s3 fadd SSQ, REGONE, s3
fmov SCALE, s5 fmov SCALE, s5
b KERNEL_S1_END_\@ b KERNEL_S1_END
KERNEL_S1_SCALE_GE_XI_\@: KERNEL_S1_SCALE_GE_XI:
fdiv s2, s5, SCALE fdiv s2, s5, SCALE
fmla SSQ, s2, v2.s[0] fmla SSQ, s2, v2.s[0]
#else #else
ldr d4, [X] ldr d4, [X]
fcmp d4, REGZERO fcmp d4, REGZERO
beq KERNEL_S1_NEXT_\@ beq KERNEL_S1_NEXT
fabs d4, d4 fabs d4, d4
fcmp SCALE, d4 fcmp SCALE, d4
bge KERNEL_S1_SCALE_GE_XR_\@ bge KERNEL_S1_SCALE_GE_XR
fdiv d2, SCALE, d4 fdiv d2, SCALE, d4
fmul d2, d2, d2 fmul d2, d2, d2
fmul d3, SSQ, d2 fmul d3, SSQ, d2
fadd SSQ, REGONE, d3 fadd SSQ, REGONE, d3
fmov SCALE, d4 fmov SCALE, d4
b KERNEL_S1_NEXT_\@ b KERNEL_S1_NEXT
KERNEL_S1_SCALE_GE_XR_\@: KERNEL_S1_SCALE_GE_XR:
fdiv d2, d4, SCALE fdiv d2, d4, SCALE
fmla SSQ, d2, v2.d[0] fmla SSQ, d2, v2.d[0]
KERNEL_S1_NEXT_\@: KERNEL_S1_NEXT:
ldr d5, [X, #8] ldr d5, [X, #8]
fcmp d5, REGZERO fcmp d5, REGZERO
beq KERNEL_S1_END_\@ beq KERNEL_S1_END
fabs d5, d5 fabs d5, d5
fcmp SCALE, d5 fcmp SCALE, d5
bge KERNEL_S1_SCALE_GE_XI_\@ bge KERNEL_S1_SCALE_GE_XI
fdiv d2, SCALE, d5 fdiv d2, SCALE, d5
fmul d2, d2, d2 fmul d2, d2, d2
fmul d3, SSQ, d2 fmul d3, SSQ, d2
fadd SSQ, REGONE, d3 fadd SSQ, REGONE, d3
fmov SCALE, d5 fmov SCALE, d5
b KERNEL_S1_END_\@ b KERNEL_S1_END
KERNEL_S1_SCALE_GE_XI_\@: KERNEL_S1_SCALE_GE_XI:
fdiv d2, d5, SCALE fdiv d2, d5, SCALE
fmla SSQ, d2, v2.d[0] fmla SSQ, d2, v2.d[0]
#endif #endif
KERNEL_S1_END_\@: KERNEL_S1_END:
add X, X, INC_X add X, X, INC_X
.endm .endm