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

Merge pull request #3695 from martin-frbg/ppc6nrm2 

PPC6: Fix DNRM2 returning INF instead of zero due to intermediate overflow
This commit is contained in:
Martin Kroeker 2022-07-25 06:14:30 +02:00 committed by GitHub
parent 31377d04f0 6c118b7977
commit a4303ae378
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 29 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

33
kernel/power/nrm2.S
View File

@ -50,8 +50,8 @@
#define FZERO 144(SP) #define FZERO 144(SP)
#define FONE 148(SP) #define FONE 148(SP)
#define FMAX 152(SP) #define FMAX 152(SP)
#define FINF 160(SP)
#define STACKSIZE 160 #define STACKSIZE 168
PROLOGUE PROLOGUE
PROFCODE PROFCODE
@ -87,6 +87,16 @@
stw r10, FZERO stw r10, FZERO
stw r11, FONE stw r11, FONE
stw r12, FMAX stw r12, FMAX
#ifdef DOUBLE
lis r12, 0x7ff0
ori r12, r12, 0x0
sldi r12, r12, 32
oris r12, r12,0x0
ori r12, r12, 0x0
std r12, FINF
#endif
stw r10, 4 + FMAX stw r10, 4 + FMAX
lfs f1, FZERO lfs f1, FZERO
@ -334,10 +344,17 @@ LL(100):
lfs f0, FONE lfs f0, FONE
fcmpu cr0, f1, f31 fcmpu cr0, f1, f31
nop
beq- cr0, LL(9999) beq- cr0, LL(9999)
fdiv f30, f0, f31 fdiv f30, f0, f31
#ifdef DOUBLE
lfd f1, FINF
fcmpu cr0, f1, f30
lfs f1, FZERO
beq- cr0, LL(9999)
#endif
fmr f0, f1 fmr f0, f1
fmr f2, f1 fmr f2, f1
fmr f3, f1 fmr f3, f1
@ -712,9 +729,16 @@ LL(1999):
fcmpu cr0, f1, f31 fcmpu cr0, f1, f31
beq- cr0, LL(9999) beq- cr0, LL(9999)
fdiv f30, f0, f31 fdiv f30, f0, f31
#ifdef DOUBLE
lfd f1, FINF
fcmpu cr0, f1, f30
lfs f1, FZERO
beq- cr0, LL(9999)
#endif
fmr f0, f1 fmr f0, f1
fmr f2, f1 fmr f2, f1
fmr f3, f1 fmr f3, f1
@ -875,6 +899,7 @@ LL(2170):
fadd f0, f0, f4 fadd f0, f0, f4
fsqrt f0, f0 fsqrt f0, f0
fmul f1, f31, f0 fmul f1, f31, f0
.align 4 .align 4