// Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0 OR ISC

// ----------------------------------------------------------------------------
// Halve modulo p_384, z := (x / 2) mod p_384, assuming x reduced
// Input x[6]; output z[6]
//
//    extern void bignum_half_p384
//     (uint64_t z[static 6], uint64_t x[static 6]);
//
// Standard x86-64 ABI: RDI = z, RSI = x
// Microsoft x64 ABI:   RCX = z, RDX = x
// ----------------------------------------------------------------------------

#include "_internal_s2n_bignum.h"


        S2N_BN_SYM_VISIBILITY_DIRECTIVE(bignum_half_p384)
        S2N_BN_SYM_PRIVACY_DIRECTIVE(bignum_half_p384)
        .text

#define z %rdi
#define x %rsi

#define a %rax
#define d0 %rcx
#define d1 %rdx
#define d2 %r8
#define d3 %r9
#define d4 %r10
#define d5 %r11

#define d0short %ecx
#define d3short %r9d



S2N_BN_SYMBOL(bignum_half_p384):

#if WINDOWS_ABI
        pushq   %rdi
        pushq   %rsi
        movq    %rcx, %rdi
        movq    %rdx, %rsi
#endif

// Load lowest digit and get a mask for its lowest bit in d3

        movq    (x), a
        movl    $1, d3short
        andq    a, d3
        negq    d3

// Create a masked version of p_384 (top 3 words = the mask itself)

        movl    $0x00000000ffffffff, d0short
        andq    d3, d0
        movq    d0, d1
        xorq    d3, d1
        movq    d3, d2
        addq    d2, d2
        andq    d3, d2
        movq    d3, d4
        movq    d3, d5

// Perform addition with masked p_384. Catch the carry in a, as a bitmask
// for convenience though we only use its LSB below with SHRD

        addq    a, d0
        adcq    8(x), d1
        adcq    16(x), d2
        adcq    24(x), d3
        adcq    32(x), d4
        adcq    40(x), d5
        sbbq    a, a

// Shift right, pushing the carry back down, and store back

        shrdq   $1, d1, d0
        movq    d0, (z)
        shrdq   $1, d2, d1
        movq    d1, 8(z)
        shrdq   $1, d3, d2
        movq    d2, 16(z)
        shrdq   $1, d4, d3
        movq    d3, 24(z)
        shrdq   $1, d5, d4
        movq    d4, 32(z)
        shrdq   $1, a, d5
        movq    d5, 40(z)

// Return

#if WINDOWS_ABI
        popq   %rsi
        popq   %rdi
#endif
        ret

#if defined(__linux__) && defined(__ELF__)
.section .note.GNU-stack,"",%progbits
#endif