mirror/gmsm
1
0
Fork 0
mirror of https://github.com/emmansun/gmsm.git synced 2025-04-27 04:36:19 +08:00
Code Issues Packages Projects Releases Wiki Activity

sm4: amd64, reduce VBROADCASTI128 usage

Browse Source
This commit is contained in:
Sun Yimin 2023-09-28 13:19:52 +08:00 committed by GitHub
parent 15d1b57a23
commit 5b3dfb6fa7
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 27 additions and 28 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
cipher/benchmark_test.go
View File

@ -25,7 +25,7 @@ func BenchmarkSM4ECBEncrypt1K(b *testing.B) {
benchmarkECBEncrypt1K(b, c) benchmarkECBEncrypt1K(b, c)
} }
func BenchmarkAES128EBCEncrypt1K(b *testing.B) { func BenchmarkAES128ECBEncrypt1K(b *testing.B) {
var key [16]byte var key [16]byte
c, _ := aes.NewCipher(key[:]) c, _ := aes.NewCipher(key[:])
benchmarkECBEncrypt1K(b, c) benchmarkECBEncrypt1K(b, c)

53
sm4/aesni_macros_amd64.s
View File

@ -16,48 +16,47 @@ GLOBL nibble_mask<>(SB), 8, $16
// inverse shift rows // inverse shift rows
DATA inverse_shift_rows<>+0x00(SB)/8, $0x0B0E0104070A0D00 DATA inverse_shift_rows<>+0x00(SB)/8, $0x0B0E0104070A0D00
DATA inverse_shift_rows<>+0x08(SB)/8, $0x0306090C0F020508 DATA inverse_shift_rows<>+0x08(SB)/8, $0x0306090C0F020508
GLOBL inverse_shift_rows<>(SB), 8, $16 DATA inverse_shift_rows<>+0x10(SB)/8, $0x0B0E0104070A0D00
DATA inverse_shift_rows<>+0x18(SB)/8, $0x0306090C0F020508
GLOBL inverse_shift_rows<>(SB), 8, $32
// Affine transform 1 (low and high hibbles) // Affine transform 1 (low and high hibbles)
DATA m1_low<>+0x00(SB)/8, $0x0A7FC3B6D5A01C69 DATA m1_low<>+0x00(SB)/8, $0x0A7FC3B6D5A01C69
DATA m1_low<>+0x08(SB)/8, $0x3045F98CEF9A2653 DATA m1_low<>+0x08(SB)/8, $0x3045F98CEF9A2653
GLOBL m1_low<>(SB), 8, $16 DATA m1_low<>+0x10(SB)/8, $0x0A7FC3B6D5A01C69
DATA m1_low<>+0x18(SB)/8, $0x3045F98CEF9A2653
GLOBL m1_low<>(SB), 8, $32
DATA m1_high<>+0x00(SB)/8, $0xC35BF46CAF379800 DATA m1_high<>+0x00(SB)/8, $0xC35BF46CAF379800
DATA m1_high<>+0x08(SB)/8, $0x68F05FC7049C33AB DATA m1_high<>+0x08(SB)/8, $0x68F05FC7049C33AB
GLOBL m1_high<>(SB), 8, $16 DATA m1_high<>+0x10(SB)/8, $0xC35BF46CAF379800
DATA m1_high<>+0x18(SB)/8, $0x68F05FC7049C33AB
GLOBL m1_high<>(SB), 8, $32
// Affine transform 2 (low and high hibbles) // Affine transform 2 (low and high hibbles)
DATA m2_low<>+0x00(SB)/8, $0x9A950A05FEF16E61 DATA m2_low<>+0x00(SB)/8, $0x9A950A05FEF16E61
DATA m2_low<>+0x08(SB)/8, $0x0E019E916A65FAF5 DATA m2_low<>+0x08(SB)/8, $0x0E019E916A65FAF5
GLOBL m2_low<>(SB), 8, $16 DATA m2_low<>+0x10(SB)/8, $0x9A950A05FEF16E61
DATA m2_low<>+0x18(SB)/8, $0x0E019E916A65FAF5
GLOBL m2_low<>(SB), 8, $32
DATA m2_high<>+0x00(SB)/8, $0x892D69CD44E0A400 DATA m2_high<>+0x00(SB)/8, $0x892D69CD44E0A400
DATA m2_high<>+0x08(SB)/8, $0x2C88CC68E14501A5 DATA m2_high<>+0x08(SB)/8, $0x2C88CC68E14501A5
GLOBL m2_high<>(SB), 8, $16 DATA m2_high<>+0x10(SB)/8, $0x892D69CD44E0A400
DATA m2_high<>+0x18(SB)/8, $0x2C88CC68E14501A5
GLOBL m2_high<>(SB), 8, $32
// left rotations of 32-bit words by 8-bit increments // left rotations of 32-bit words by 8-bit increments
DATA r08_mask<>+0x00(SB)/8, $0x0605040702010003 DATA r08_mask<>+0x00(SB)/8, $0x0605040702010003
DATA r08_mask<>+0x08(SB)/8, $0x0E0D0C0F0A09080B DATA r08_mask<>+0x08(SB)/8, $0x0E0D0C0F0A09080B
GLOBL r08_mask<>(SB), 8, $16 DATA r08_mask<>+0x10(SB)/8, $0x0605040702010003
DATA r08_mask<>+0x18(SB)/8, $0x0E0D0C0F0A09080B
GLOBL r08_mask<>(SB), 8, $32
DATA fk_mask<>+0x00(SB)/8, $0x56aa3350a3b1bac6 DATA fk_mask<>+0x00(SB)/8, $0x56aa3350a3b1bac6
DATA fk_mask<>+0x08(SB)/8, $0xb27022dc677d9197 DATA fk_mask<>+0x08(SB)/8, $0xb27022dc677d9197
GLOBL fk_mask<>(SB), 8, $16 GLOBL fk_mask<>(SB), 8, $16
// inverse shift rows
DATA inverse_shift_rows256<>+0x00(SB)/8, $0x0B0E0104070A0D00
DATA inverse_shift_rows256<>+0x08(SB)/8, $0x0306090C0F020508
DATA inverse_shift_rows256<>+0x10(SB)/8, $0x0B0E0104070A0D00
DATA inverse_shift_rows256<>+0x18(SB)/8, $0x0306090C0F020508
GLOBL inverse_shift_rows256<>(SB), 8, $32
DATA r08_mask256<>+0x00(SB)/8, $0x0605040702010003
DATA r08_mask256<>+0x08(SB)/8, $0x0E0D0C0F0A09080B
DATA r08_mask256<>+0x10(SB)/8, $0x0605040702010003
DATA r08_mask256<>+0x18(SB)/8, $0x0E0D0C0F0A09080B
GLOBL r08_mask256<>(SB), 8, $32
// Transpose matrix without PUNPCKHDQ/PUNPCKLDQ/PUNPCKHQDQ/PUNPCKLQDQ instructions, bad performance! // Transpose matrix without PUNPCKHDQ/PUNPCKLDQ/PUNPCKHQDQ/PUNPCKLQDQ instructions, bad performance!
// input: from high to low // input: from high to low
// r0 = [w3, w2, w1, w0] // r0 = [w3, w2, w1, w0]
@ -539,24 +538,24 @@ GLOBL r08_mask256<>(SB), 8, $32
// - yNibbleMask: 256 bits register stored nibble mask, should be loaded earlier. // - yNibbleMask: 256 bits register stored nibble mask, should be loaded earlier.
#define AVX2_SM4_SBOX(x, y, z, xw, yw, xNibbleMask, yNibbleMask) \ #define AVX2_SM4_SBOX(x, y, z, xw, yw, xNibbleMask, yNibbleMask) \
VPAND yNibbleMask, x, z; \ VPAND yNibbleMask, x, z; \
VBROADCASTI128 m1_low<>(SB), y; \ VMOVDQU m1_low<>(SB), y; \
VPSHUFB z, y, y; \ VPSHUFB z, y, y; \
VPSRLQ $4, x, x; \ VPSRLQ $4, x, x; \
VPAND yNibbleMask, x, x; \ VPAND yNibbleMask, x, x; \
VBROADCASTI128 m1_high<>(SB), z; \ VMOVDQU m1_high<>(SB), z; \
VPSHUFB x, z, x; \ VPSHUFB x, z, x; \
VPXOR y, x, x; \ VPXOR y, x, x; \
VPSHUFB inverse_shift_rows256<>(SB), x, x; \ VPSHUFB inverse_shift_rows<>(SB), x, x; \
VEXTRACTI128 $1, x, yw \ VEXTRACTI128 $1, x, yw \
VAESENCLAST xNibbleMask, xw, xw; \ VAESENCLAST xNibbleMask, xw, xw; \
VAESENCLAST xNibbleMask, yw, yw; \ VAESENCLAST xNibbleMask, yw, yw; \
VINSERTI128 $1, yw, x, x; \ VINSERTI128 $1, yw, x, x; \
VPANDN yNibbleMask, x, z; \ VPANDN yNibbleMask, x, z; \
VBROADCASTI128 m2_low<>(SB), y; \ VMOVDQU m2_low<>(SB), y; \
VPSHUFB z, y, y; \ VPSHUFB z, y, y; \
VPSRLQ $4, x, x; \ VPSRLQ $4, x, x; \
VPAND yNibbleMask, x, x; \ VPAND yNibbleMask, x, x; \
VBROADCASTI128 m2_high<>(SB), z; \ VMOVDQU m2_high<>(SB), z; \
VPSHUFB x, z, x; \ VPSHUFB x, z, x; \
VPXOR y, x, x VPXOR y, x, x
@ -571,11 +570,11 @@ GLOBL r08_mask256<>(SB), 8, $32
// - yNibbleMask: 256 bits register stored nibble mask, should be loaded earlier. // - yNibbleMask: 256 bits register stored nibble mask, should be loaded earlier.
#define AVX2_SM4_TAO_L1(x, y, z, xw, yw, xNibbleMask, yNibbleMask) \ #define AVX2_SM4_TAO_L1(x, y, z, xw, yw, xNibbleMask, yNibbleMask) \
AVX2_SM4_SBOX(x, y, z, xw, yw, xNibbleMask, yNibbleMask); \ AVX2_SM4_SBOX(x, y, z, xw, yw, xNibbleMask, yNibbleMask); \
VPSHUFB r08_mask256<>(SB), x, y; \ // y = x <<< 8 VPSHUFB r08_mask<>(SB), x, y; \ // y = x <<< 8
VPSHUFB r08_mask256<>(SB), y, z; \ // z = x <<< 16 VPSHUFB r08_mask<>(SB), y, z; \ // z = x <<< 16
VPXOR x, y, y; \ // y = x ^ (x <<< 8) VPXOR x, y, y; \ // y = x ^ (x <<< 8)
VPXOR z, y, y; \ // y = x ^ (x <<< 8) ^ (x <<< 16) VPXOR z, y, y; \ // y = x ^ (x <<< 8) ^ (x <<< 16)
VPSHUFB r08_mask256<>(SB), z, z; \ // z = x <<< 24 VPSHUFB r08_mask<>(SB), z, z; \ // z = x <<< 24
VPXOR x, z, x; \ // x = x ^ (x <<< 24) VPXOR x, z, x; \ // x = x ^ (x <<< 24)
VPSLLD $2, y, z; \ VPSLLD $2, y, z; \
VPSRLD $30, y, y; \ VPSRLD $30, y, y; \