codex-storage/constantine
1
0
Fork 0
mirror of https://github.com/codex-storage/constantine.git synced 2025-02-06 16:05:05 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix inline assembly and inline __int128 in C++ mode

Browse Source
This commit is contained in:
Mamy André-Ratsimbazafy 2020-03-01 16:23:41 +01:00
parent c8e482f6d2
commit 4b7c6b552e
No known key found for this signature in database
GPG Key ID: 7B88AD1FE79492E1
1 changed files with 40 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

57
constantine/primitives/extended_precision.nim
View File

@ -108,19 +108,31 @@ when sizeof(int) == 8:
# 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): when defined(amd64):
asm """ when defined(cpp):
divq %[divisor] asm """
: "=a" (`*q`), "=d" (`*r`) divq %[divisor]
: "d" (`n_hi`), "a" (`n_lo`), [divisor] "rm" (`d`) : "=a" (`q`), "=d" (`r`)
: : "d" (`n_hi`), "a" (`n_lo`), [divisor] "rm" (`d`)
""" :
"""
else:
asm """
divq %[divisor]
: "=a" (`*q`), "=d" (`*r`)
: "d" (`n_hi`), "a" (`n_lo`), [divisor] "rm" (`d`)
:
"""
else: else:
var dblPrec {.noInit.}: uint128 var dblPrec {.noInit.}: uint128
{.emit:[dblPrec, " = (unsigned __int128)", n_hi," << 64 | (unsigned __int128)",n_lo,";"].} {.emit:[dblPrec, " = (unsigned __int128)", n_hi," << 64 | (unsigned __int128)",n_lo,";"].}
# Don't forget to dereference the var param # Don't forget to dereference the var param in C mode
{.emit:["*",q, " = (NU64)(", dblPrec," / ", d, ");"].} when defined(cpp):
{.emit:["*",r, " = (NU64)(", dblPrec," % ", d, ");"].} {.emit:[q, " = (NU64)(", dblPrec," / ", d, ");"].}
{.emit:[r, " = (NU64)(", dblPrec," % ", d, ");"].}
else:
{.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
@ -132,9 +144,13 @@ when sizeof(int) == 8:
var dblPrec {.noInit.}: uint128 var dblPrec {.noInit.}: uint128
{.emit:[dblPrec, " = (unsigned __int128)", a," * (unsigned __int128)", b, " + (unsigned __int128)",c,";"].} {.emit:[dblPrec, " = (unsigned __int128)", a," * (unsigned __int128)", b, " + (unsigned __int128)",c,";"].}
# Don't forget to dereference the var param # Don't forget to dereference the var param in C mode
{.emit:["*",hi, " = (NU64)(", dblPrec," >> ", 63'u64, ");"].} when defined(cpp):
{.emit:["*",lo, " = (NU64)", dblPrec," & ", 1'u64 shl 63 - 1, ";"].} {.emit:[hi, " = (NU64)(", dblPrec," >> ", 63'u64, ");"].}
{.emit:[lo, " = (NU64)", dblPrec," & ", 1'u64 shl 63 - 1, ";"].}
else:
{.emit:["*",hi, " = (NU64)(", dblPrec," >> ", 63'u64, ");"].}
{.emit:["*",lo, " = (NU64)", dblPrec," & ", 1'u64 shl 63 - 1, ";"].}
func unsafeFMA2*(hi, lo: var Ct[uint64], a1, b1, a2, b2, c1, c2: Ct[uint64]) {.inline.}= func unsafeFMA2*(hi, lo: var Ct[uint64], a1, b1, a2, b2, c1, c2: Ct[uint64]) {.inline.}=
## (hi, lo) <- a1 * b1 + a2 * b2 + c1 + c2 ## (hi, lo) <- a1 * b1 + a2 * b2 + c1 + c2
@ -145,9 +161,13 @@ when sizeof(int) == 8:
" + (unsigned __int128)", a2," * (unsigned __int128)", b2, " + (unsigned __int128)", a2," * (unsigned __int128)", b2,
" + (unsigned __int128)", c1, " + (unsigned __int128)", c1,
" + (unsigned __int128)", c2, ";"].} " + (unsigned __int128)", c2, ";"].}
# Don't forget to dereference the var param # Don't forget to dereference the var param in C mode
{.emit:["*",hi, " = (NU64)(", dblPrec," >> ", 63'u64, ");"].} when defined(cpp):
{.emit:["*",lo, " = (NU64)", dblPrec," & ", (1'u64 shl 63 - 1), ";"].} {.emit:[hi, " = (NU64)(", dblPrec," >> ", 63'u64, ");"].}
{.emit:[lo, " = (NU64)", dblPrec," & ", (1'u64 shl 63 - 1), ";"].}
else:
{.emit:["*",hi, " = (NU64)(", dblPrec," >> ", 63'u64, ");"].}
{.emit:["*",lo, " = (NU64)", dblPrec," & ", (1'u64 shl 63 - 1), ";"].}
func unsafeFMA2_hi*(hi: var Ct[uint64], a1, b1, a2, b2, c: Ct[uint64]) {.inline.}= func unsafeFMA2_hi*(hi: var Ct[uint64], a1, b1, a2, b2, c: Ct[uint64]) {.inline.}=
## Returns the high word of the sum of extended precision multiply-adds ## Returns the high word of the sum of extended precision multiply-adds
@ -157,8 +177,11 @@ when sizeof(int) == 8:
{.emit:[dblPrec, " = (unsigned __int128)", a1," * (unsigned __int128)", b1, {.emit:[dblPrec, " = (unsigned __int128)", a1," * (unsigned __int128)", b1,
" + (unsigned __int128)", a2," * (unsigned __int128)", b2, " + (unsigned __int128)", a2," * (unsigned __int128)", b2,
" + (unsigned __int128)", c, ";"].} " + (unsigned __int128)", c, ";"].}
# Don't forget to dereference the var param # Don't forget to dereference the var param in C mode
{.emit:["*",hi, " = (NU64)(", dblPrec," >> ", 63'u64, ");"].} when defined(cpp):
{.emit:[hi, " = (NU64)(", dblPrec," >> ", 63'u64, ");"].}
else:
{.emit:["*",hi, " = (NU64)(", dblPrec," >> ", 63'u64, ");"].}
elif defined(vcc): elif defined(vcc):
func udiv128(highDividend, lowDividend, divisor: uint64, remainder: var uint64): uint64 {.importc:"_udiv128", header: "<immintrin.h>", nodecl.} func udiv128(highDividend, lowDividend, divisor: uint64, remainder: var uint64): uint64 {.importc:"_udiv128", header: "<immintrin.h>", nodecl.}