/// MA 02111-1307, USA.
///--------------------------------------------------------------------------
-/// External definitions.
+/// Preliminaries.
#include "config.h"
#include "asm-common.h"
-///--------------------------------------------------------------------------
-/// Main.code.
-
.arch armv7-a
.fpu neon
+
.text
+///--------------------------------------------------------------------------
+/// Main code.
+
FUNC(salsa20_core_arm_neon)
// Arguments are in registers.
// [ 8 9 10 11] [ 8 13 2 7] (c, q10)
// [12 13 14 15] [12 1 6 11] (d, q11)
//
- // [ 0 1 2 3] (a, q8)
- // [ 4 5 6 7] (b, q9)
- // [ 8 9 10 11] (c, q10)
- // [12 13 14 15] (d, q11)
- //
// We need a copy for later. Rather than waste time copying them by
// hand, we'll use the three-address nature of the instruction set.
// But this means that the main loop is offset by a bit.
// Almost there. Firstly the feedfoward addition. Also, establish a
// constant which will be useful later.
9: vadd.u32 q0, q8, q12 // 0, 5, 10, 15
- vmov.i64 q12, #0xffffffff // = (-1, 0, -1, 0)
+ vmov.i64 q12, #0xffffffff // = (0, -1; 0, -1)
vadd.u32 q1, q9, q13 // 4, 9, 14, 3
vadd.u32 q2, q10, q14 // 8, 13, 2, 7
vadd.u32 q3, q11, q15 // 12, 1, 6, 11