///--------------------------------------------------------------------------
/// Main code.
- .arch pentium4
.text
+FUNC(salsa20_core_x86ish_avx)
+ .arch .avx
+ vzeroupper
+ endprologue
+ // drop through...
+ENDFUNC
+
+ .arch pentium4
+
FUNC(salsa20_core_x86ish_sse2)
// Initial setup.
// d ^= (c + b) <<< 13
movdqa xmm4, xmm2
paddd xmm4, xmm1
- pshufd xmm1, xmm1, SHUF(2, 1, 0, 3)
+ pshufd xmm1, xmm1, SHUF(3, 0, 1, 2)
movdqa xmm5, xmm4
pslld xmm4, 13
psrld xmm5, 19
// a ^= (d + c) <<< 18
movdqa xmm4, xmm3
- pshufd xmm3, xmm3, SHUF(0, 3, 2, 1)
+ pshufd xmm3, xmm3, SHUF(1, 2, 3, 0)
paddd xmm4, xmm2
- pshufd xmm2, xmm2, SHUF(1, 0, 3, 2)
+ pshufd xmm2, xmm2, SHUF(2, 3, 0, 1)
movdqa xmm5, xmm4
pslld xmm4, 18
psrld xmm5, 14
// d ^= (c + b) <<< 13
movdqa xmm4, xmm2
paddd xmm4, xmm3
- pshufd xmm3, xmm3, SHUF(2, 1, 0, 3)
+ pshufd xmm3, xmm3, SHUF(3, 0, 1, 2)
movdqa xmm5, xmm4
pslld xmm4, 13
psrld xmm5, 19
// a ^= (d + c) <<< 18
movdqa xmm4, xmm1
- pshufd xmm1, xmm1, SHUF(0, 3, 2, 1)
+ pshufd xmm1, xmm1, SHUF(1, 2, 3, 0)
paddd xmm4, xmm2
- pshufd xmm2, xmm2, SHUF(1, 0, 3, 2)
+ pshufd xmm2, xmm2, SHUF(2, 3, 0, 1)
movdqa xmm5, xmm4
pslld xmm4, 18
psrld xmm5, 14
// input. This can be done by juggling values in registers, with the
// following fancy footwork: some row rotations, a transpose, and
// some more rotations.
- pshufd xmm1, xmm1, SHUF(2, 1, 0, 3) // 3, 4, 9, 14
- pshufd xmm2, xmm2, SHUF(1, 0, 3, 2) // 2, 7, 8, 13
- pshufd xmm3, xmm3, SHUF(0, 3, 2, 1) // 1, 6, 11, 12
+ pshufd xmm1, xmm1, SHUF(3, 0, 1, 2) // 3, 4, 9, 14
+ pshufd xmm2, xmm2, SHUF(2, 3, 0, 1) // 2, 7, 8, 13
+ pshufd xmm3, xmm3, SHUF(1, 2, 3, 0) // 1, 6, 11, 12
movdqa xmm4, xmm0
movdqa xmm5, xmm3
punpckhdq xmm1, xmm3 // 5, 6, 7, 4
punpckhdq xmm2, xmm5 // 15, 12, 13, 14
- pshufd xmm1, xmm1, SHUF(2, 1, 0, 3) // 4, 5, 6, 7
- pshufd xmm4, xmm4, SHUF(1, 0, 3, 2) // 8, 9, 10, 11
- pshufd xmm2, xmm2, SHUF(0, 3, 2, 1) // 12, 13, 14, 15
+ pshufd xmm1, xmm1, SHUF(3, 0, 1, 2) // 4, 5, 6, 7
+ pshufd xmm4, xmm4, SHUF(2, 3, 0, 1) // 8, 9, 10, 11
+ pshufd xmm2, xmm2, SHUF(1, 2, 3, 0) // 12, 13, 14, 15
// Finally we have to write out the result.
movdqu [OUT + 0], xmm0