#include <linux/linkage.h>
.text
.option arch, +zvknhb, +zvkb
#define STATEP a0
#define DATA a1
#define NUM_BLOCKS a2
#define STATEP_C a3
#define K a4
#define MASK v0
#define INDICES v1
#define W0 v10
#define W1 v12
#define W2 v14
#define W3 v16
#define VTMP v20
#define FEBA v22
#define HGDC v24
#define PREV_FEBA v26
#define PREV_HGDC v28
.macro sha512_4rounds last, w0, w1, w2, w3
vle64.v VTMP, (K)
addi K, K, 32
vadd.vv VTMP, VTMP, \w0
vsha2cl.vv HGDC, FEBA, VTMP
vsha2ch.vv FEBA, HGDC, VTMP
.if !\last
vmerge.vvm VTMP, \w2, \w1, MASK
vsha2ms.vv \w0, VTMP, \w3
.endif
.endm
.macro sha512_16rounds last
sha512_4rounds \last, W0, W1, W2, W3
sha512_4rounds \last, W1, W2, W3, W0
sha512_4rounds \last, W2, W3, W0, W1
sha512_4rounds \last, W3, W0, W1, W2
.endm
SYM_FUNC_START(sha512_transform_zvknhb_zvkb)
vsetivli zero, 1, e8, m1, ta, ma
vmv.v.i MASK, 0x01
li t0, 0x00082028
vsetivli zero, 1, e32, m1, ta, ma
vmv.v.x INDICES, t0
addi STATEP_C, STATEP, 16
vsetivli zero, 4, e64, m2, ta, ma
vluxei8.v FEBA, (STATEP), INDICES
vluxei8.v HGDC, (STATEP_C), INDICES
.Lnext_block:
la K, K512
addi NUM_BLOCKS, NUM_BLOCKS, -1
vmv.v.v PREV_FEBA, FEBA
vmv.v.v PREV_HGDC, HGDC
vle64.v W0, (DATA)
vrev8.v W0, W0
addi DATA, DATA, 32
vle64.v W1, (DATA)
vrev8.v W1, W1
addi DATA, DATA, 32
vle64.v W2, (DATA)
vrev8.v W2, W2
addi DATA, DATA, 32
vle64.v W3, (DATA)
vrev8.v W3, W3
addi DATA, DATA, 32
sha512_16rounds 0
sha512_16rounds 0
sha512_16rounds 0
sha512_16rounds 0
sha512_16rounds 1
vadd.vv FEBA, FEBA, PREV_FEBA
vadd.vv HGDC, HGDC, PREV_HGDC
bnez NUM_BLOCKS, .Lnext_block
vsuxei8.v FEBA, (STATEP), INDICES
vsuxei8.v HGDC, (STATEP_C), INDICES
ret
SYM_FUNC_END(sha512_transform_zvknhb_zvkb)
.section ".rodata"
.p2align 3
.type K512, @object
K512:
.dword 0x428a2f98d728ae22, 0x7137449123ef65cd
.dword 0xb5c0fbcfec4d3b2f, 0xe9b5dba58189dbbc
.dword 0x3956c25bf348b538, 0x59f111f1b605d019
.dword 0x923f82a4af194f9b, 0xab1c5ed5da6d8118
.dword 0xd807aa98a3030242, 0x12835b0145706fbe
.dword 0x243185be4ee4b28c, 0x550c7dc3d5ffb4e2
.dword 0x72be5d74f27b896f, 0x80deb1fe3b1696b1
.dword 0x9bdc06a725c71235, 0xc19bf174cf692694
.dword 0xe49b69c19ef14ad2, 0xefbe4786384f25e3
.dword 0x0fc19dc68b8cd5b5, 0x240ca1cc77ac9c65
.dword 0x2de92c6f592b0275, 0x4a7484aa6ea6e483
.dword 0x5cb0a9dcbd41fbd4, 0x76f988da831153b5
.dword 0x983e5152ee66dfab, 0xa831c66d2db43210
.dword 0xb00327c898fb213f, 0xbf597fc7beef0ee4
.dword 0xc6e00bf33da88fc2, 0xd5a79147930aa725
.dword 0x06ca6351e003826f, 0x142929670a0e6e70
.dword 0x27b70a8546d22ffc, 0x2e1b21385c26c926
.dword 0x4d2c6dfc5ac42aed, 0x53380d139d95b3df
.dword 0x650a73548baf63de, 0x766a0abb3c77b2a8
.dword 0x81c2c92e47edaee6, 0x92722c851482353b
.dword 0xa2bfe8a14cf10364, 0xa81a664bbc423001
.dword 0xc24b8b70d0f89791, 0xc76c51a30654be30
.dword 0xd192e819d6ef5218, 0xd69906245565a910
.dword 0xf40e35855771202a, 0x106aa07032bbd1b8
.dword 0x19a4c116b8d2d0c8, 0x1e376c085141ab53
.dword 0x2748774cdf8eeb99, 0x34b0bcb5e19b48a8
.dword 0x391c0cb3c5c95a63, 0x4ed8aa4ae3418acb
.dword 0x5b9cca4f7763e373, 0x682e6ff3d6b2b8a3
.dword 0x748f82ee5defb2fc, 0x78a5636f43172f60
.dword 0x84c87814a1f0ab72, 0x8cc702081a6439ec
.dword 0x90befffa23631e28, 0xa4506cebde82bde9
.dword 0xbef9a3f7b2c67915, 0xc67178f2e372532b
.dword 0xca273eceea26619c, 0xd186b8c721c0c207
.dword 0xeada7dd6cde0eb1e, 0xf57d4f7fee6ed178
.dword 0x06f067aa72176fba, 0x0a637dc5a2c898a6
.dword 0x113f9804bef90dae, 0x1b710b35131c471b
.dword 0x28db77f523047d84, 0x32caab7b40c72493
.dword 0x3c9ebe0a15c9bebc, 0x431d67c49c100d4c
.dword 0x4cc5d4becb3e42b6, 0x597f299cfc657e2a
.dword 0x5fcb6fab3ad6faec, 0x6c44198c4a475817
.size K512, . - K512