1
0
Fork 0
mirror of https://github.com/anrieff/libcpuid synced 2025-10-03 11:01:30 +00:00

Implemented busy_sse_loop for win32. Need to do for x64 though.

git-svn-id: https://svn.code.sf.net/p/libcpuid/code/HEAD/libcpuid@90 3b4be424-7ac5-41d7-8526-f4ddcb85d872
This commit is contained in:
Veselin Georgiev 2011-01-10 13:37:51 +00:00
commit f398770f79

View file

@ -513,6 +513,311 @@ void busy_sse_loop(int cycles)
);
#else
# ifdef COMPILER_MICROSOFT
__asm {
mov eax, cycles
xorps xmm0, xmm0
xorps xmm1, xmm1
xorps xmm2, xmm2
xorps xmm3, xmm3
xorps xmm4, xmm4
xorps xmm5, xmm5
xorps xmm6, xmm6
xorps xmm7, xmm7
//--
align 16
bsLoop:
// 0:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 1:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 2:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 3:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 4:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 5:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 6:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 7:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 8:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 9:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 10:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 11:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 12:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 13:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 14:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 15:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 16:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 17:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 18:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 19:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 20:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 21:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 22:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 23:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 24:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 25:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 26:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 27:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 28:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 29:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 30:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
// 31:
addps xmm0, xmm1
addps xmm1, xmm2
addps xmm2, xmm3
addps xmm3, xmm4
addps xmm4, xmm5
addps xmm5, xmm6
addps xmm6, xmm7
addps xmm7, xmm0
//----------------------
dec eax
jnz bsLoop
}
# else
# error "Unsupported compiler"
# endif /* COMPILER_MICROSOFT */