codex-storage
/
constantine
mirror of https://github.com/codex-storage/constantine.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Add ARM64 fallback for Div128by64

This commit is contained in:
Mamy André-Ratsimbazafy 2020-03-01 15:10:23 +01:00
parent 3ce237234d
commit dfb007e6bc
No known key found for this signature in database
GPG Key ID: 7B88AD1FE79492E1
1 changed files with 17 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
constantine/primitives/extended_precision.nim
View File

@ -75,7 +75,9 @@ func unsafeFMA2_hi*(hi: var Ct[uint32], a1, b1, a2, b2, c1: Ct[uint32]) {.inline
# #
# ############################################################ # ############################################################
when defined(gcc) or defined(clang) or defined(llvm_gcc): const GccCompatible = defined(gcc) or defined(clang) or defined(llvm_gcc)
when sizeof(int) == 8 and GccCompatible:
type type
uint128*{.importc: "unsigned __int128".} = object uint128*{.importc: "unsigned __int128".} = object
@ -104,12 +106,20 @@ when defined(gcc) or defined(clang) or defined(llvm_gcc):
# 2. don't forget to dereference the var hidden pointer # 2. don't forget to dereference the var hidden pointer
# 3. - # 3. -
# 4. no clobbered registers beside explectly used RAX and RDX # 4. no clobbered registers beside explectly used RAX and RDX
when defined(amd64):
asm """ asm """
divq %[divisor] divq %[divisor]
: "=a" (`*q`), "=d" (`*r`) : "=a" (`*q`), "=d" (`*r`)
: "d" (`n_hi`), "a" (`n_lo`), [divisor] "rm" (`d`) : "d" (`n_hi`), "a" (`n_lo`), [divisor] "rm" (`d`)
: :
""" """
else:
var dblPrec {.noInit.}: uint128
{.emit:[dblPrec, " = (unsigned __int128)", n_hi," << 64 | (unsigned __int128)",n_lo,";"].}
# Don't forget to dereference the var param
{.emit:["*",q, " = (NU64)(", dblPrec," / ", d, ");"].}
{.emit:["*",r, " = (NU64)(", dblPrec," % ", d, ");"].}
func unsafeFMA*(hi, lo: var Ct[uint64], a, b, c: Ct[uint64]) {.inline.}= func unsafeFMA*(hi, lo: var Ct[uint64], a, b, c: Ct[uint64]) {.inline.}=
## Extended precision multiplication + addition ## Extended precision multiplication + addition