.macro mulcore r, i, slo, shi, d0, d1=nil, d2=nil, d3=nil
// Multiply R_I by the expanded operand SLO/SHI, and leave the pieces
// of the product in registers D0, D1, D2, D3.
.macro mulcore r, i, slo, shi, d0, d1=nil, d2=nil, d3=nil
// Multiply R_I by the expanded operand SLO/SHI, and leave the pieces
// of the product in registers D0, D1, D2, D3.
- pshufd \d0, \r, SHUF(3, \i, 3, \i) // (r_i, ?; r_i, ?)
+ pshufd \d0, \r, SHUF(\i, 3, \i, 3) // (r_i, ?; r_i, ?)
.ifnes "\d1", "nil"
movdqa \d1, \slo // (s'_0, s'_1; s''_0, s''_1)
.endif
.ifnes "\d1", "nil"
movdqa \d1, \slo // (s'_0, s'_1; s''_0, s''_1)
.endif
// lane 0 or 1 of D; the high two lanes of D are clobbered. On
// completion, XMM3 is clobbered. If CC is `nil', then the
// contribution which would have been added to it is left in C.
// lane 0 or 1 of D; the high two lanes of D are clobbered. On
// completion, XMM3 is clobbered. If CC is `nil', then the
// contribution which would have been added to it is left in C.
- pshufd xmm3, \c, SHUF(2, 3, 3, 3) // (?, ?; ?, t = c'' mod B)
+ pshufd xmm3, \c, SHUF(3, 3, 3, 2) // (?, ?; ?, t = c'' mod B)
psrldq xmm3, 12 // (t, 0; 0, 0) = (t; 0)
pslldq xmm3, 2 // (t b; 0)
paddq \c, xmm3 // (c' + t b; c'')
psrldq xmm3, 12 // (t, 0; 0, 0) = (t; 0)
pslldq xmm3, 2 // (t b; 0)
paddq \c, xmm3 // (c' + t b; c'')
punpcklwd \c, \z // (c'_0, c''_0; c'_1, c''_1)
punpckhwd \d, \z // (c'_2, c''_2; c'_3, c''_3)
.endif
punpcklwd \c, \z // (c'_0, c''_0; c'_1, c''_1)
punpckhwd \d, \z // (c'_2, c''_2; c'_3, c''_3)
.endif
- pshufd \a, \a, SHUF(3, 1, 2, 0) // (a'_0, a'_1; a''_0, a''_1)
- pshufd \b, \b, SHUF(3, 1, 2, 0) // (a'_2, a'_3; a''_2, a''_3)
+ pshufd \a, \a, SHUF(0, 2, 1, 3) // (a'_0, a'_1; a''_0, a''_1)
+ pshufd \b, \b, SHUF(0, 2, 1, 3) // (a'_2, a'_3; a''_2, a''_3)
- pshufd \c, \c, SHUF(3, 1, 2, 0) // (c'_0, c'_1; c''_0, c''_1)
- pshufd \d, \d, SHUF(3, 1, 2, 0) // (c'_2, c'_3; c''_2, c''_3)
+ pshufd \c, \c, SHUF(0, 2, 1, 3) // (c'_0, c'_1; c''_0, c''_1)
+ pshufd \d, \d, SHUF(0, 2, 1, 3) // (c'_2, c'_3; c''_2, c''_3)