From 679e53a9f12e0b40956568f17d2149dd400c5969 Mon Sep 17 00:00:00 2001 From: Samuel Neves Date: Wed, 6 Dec 2017 23:45:58 +0000 Subject: chacha20: avx512vl implementation Signed-off-by: Samuel Neves --- src/crypto/chacha20-x86_64.S | 570 ++++++++++++++++++++++++++++++++++++++++++ src/crypto/chacha20poly1305.c | 1 + 2 files changed, 571 insertions(+) diff --git a/src/crypto/chacha20-x86_64.S b/src/crypto/chacha20-x86_64.S index d69625f..39d7080 100644 --- a/src/crypto/chacha20-x86_64.S +++ b/src/crypto/chacha20-x86_64.S @@ -59,6 +59,10 @@ .align 64 .Lsixteen: .long 16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16 +.section .rodata.cst32.Ltwoy, "aM", @progbits, 32 +.align 64 +.Ltwoy: +.long 2,0,0,0, 2,0,0,0 .text @@ -2057,4 +2061,570 @@ ENTRY(chacha20_avx512) .L16x_epilogue: ret ENDPROC(chacha20_avx512) + +.align 32 +ENTRY(chacha20_avx512vl) + cmpq $0,%rdx + je .Lavx512vl_epilogue + + leaq 8(%rsp),%r10 + + cmpq $128,%rdx + ja .Lchacha20_8xvl + + subq $64+8,%rsp + andq $-64,%rsp + vbroadcasti128 .Lsigma(%rip),%ymm0 + vbroadcasti128 (%rcx),%ymm1 + vbroadcasti128 16(%rcx),%ymm2 + vbroadcasti128 (%r8),%ymm3 + + vmovdqa32 %ymm0,%ymm16 + vmovdqa32 %ymm1,%ymm17 + vmovdqa32 %ymm2,%ymm18 + vpaddd .Lzeroz(%rip),%ymm3,%ymm3 + vmovdqa32 .Ltwoy(%rip),%ymm20 + movq $10,%r8 + vmovdqa32 %ymm3,%ymm19 + jmp .Loop_avx512vl + +.align 16 +.Loop_outer_avx512vl: + vmovdqa32 %ymm18,%ymm2 + vpaddd %ymm20,%ymm19,%ymm3 + movq $10,%r8 + vmovdqa32 %ymm3,%ymm19 + jmp .Loop_avx512vl + +.align 32 +.Loop_avx512vl: + vpaddd %ymm1,%ymm0,%ymm0 + vpxor %ymm0,%ymm3,%ymm3 + vprold $16,%ymm3,%ymm3 + vpaddd %ymm3,%ymm2,%ymm2 + vpxor %ymm2,%ymm1,%ymm1 + vprold $12,%ymm1,%ymm1 + vpaddd %ymm1,%ymm0,%ymm0 + vpxor %ymm0,%ymm3,%ymm3 + vprold $8,%ymm3,%ymm3 + vpaddd %ymm3,%ymm2,%ymm2 + vpxor %ymm2,%ymm1,%ymm1 + vprold $7,%ymm1,%ymm1 + vpshufd $78,%ymm2,%ymm2 + vpshufd $57,%ymm1,%ymm1 + vpshufd $147,%ymm3,%ymm3 + vpaddd %ymm1,%ymm0,%ymm0 + vpxor %ymm0,%ymm3,%ymm3 + vprold $16,%ymm3,%ymm3 + vpaddd %ymm3,%ymm2,%ymm2 + vpxor %ymm2,%ymm1,%ymm1 + vprold $12,%ymm1,%ymm1 + vpaddd %ymm1,%ymm0,%ymm0 + vpxor %ymm0,%ymm3,%ymm3 + vprold $8,%ymm3,%ymm3 + vpaddd %ymm3,%ymm2,%ymm2 + vpxor %ymm2,%ymm1,%ymm1 + vprold $7,%ymm1,%ymm1 + vpshufd $78,%ymm2,%ymm2 + vpshufd $147,%ymm1,%ymm1 + vpshufd $57,%ymm3,%ymm3 + decq %r8 + jnz .Loop_avx512vl + vpaddd %ymm16,%ymm0,%ymm0 + vpaddd %ymm17,%ymm1,%ymm1 + vpaddd %ymm18,%ymm2,%ymm2 + vpaddd %ymm19,%ymm3,%ymm3 + + subq $64,%rdx + jb .Ltail64_avx512vl + + vpxor 0(%rsi),%xmm0,%xmm4 + vpxor 16(%rsi),%xmm1,%xmm5 + vpxor 32(%rsi),%xmm2,%xmm6 + vpxor 48(%rsi),%xmm3,%xmm7 + leaq 64(%rsi),%rsi + + vmovdqu %xmm4,0(%rdi) + vmovdqu %xmm5,16(%rdi) + vmovdqu %xmm6,32(%rdi) + vmovdqu %xmm7,48(%rdi) + leaq 64(%rdi),%rdi + + jz .Ldone_avx512vl + + vextracti128 $1,%ymm0,%xmm4 + vextracti128 $1,%ymm1,%xmm5 + vextracti128 $1,%ymm2,%xmm6 + vextracti128 $1,%ymm3,%xmm7 + + subq $64,%rdx + jb .Ltail_avx512vl + + vpxor 0(%rsi),%xmm4,%xmm4 + vpxor 16(%rsi),%xmm5,%xmm5 + vpxor 32(%rsi),%xmm6,%xmm6 + vpxor 48(%rsi),%xmm7,%xmm7 + leaq 64(%rsi),%rsi + + vmovdqu %xmm4,0(%rdi) + vmovdqu %xmm5,16(%rdi) + vmovdqu %xmm6,32(%rdi) + vmovdqu %xmm7,48(%rdi) + leaq 64(%rdi),%rdi + + vmovdqa32 %ymm16,%ymm0 + vmovdqa32 %ymm17,%ymm1 + jnz .Loop_outer_avx512vl + + jmp .Ldone_avx512vl + +.align 16 +.Ltail64_avx512vl: + vmovdqa %xmm0,0(%rsp) + vmovdqa %xmm1,16(%rsp) + vmovdqa %xmm2,32(%rsp) + vmovdqa %xmm3,48(%rsp) + addq $64,%rdx + jmp .Loop_tail_avx512vl + +.align 16 +.Ltail_avx512vl: + vmovdqa %xmm4,0(%rsp) + vmovdqa %xmm5,16(%rsp) + vmovdqa %xmm6,32(%rsp) + vmovdqa %xmm7,48(%rsp) + addq $64,%rdx + +.Loop_tail_avx512vl: + movzbl (%rsi,%r8,1),%eax + movzbl (%rsp,%r8,1),%ecx + leaq 1(%r8),%r8 + xorl %ecx,%eax + movb %al,-1(%rdi,%r8,1) + decq %rdx + jnz .Loop_tail_avx512vl + + vmovdqa32 %ymm16,0(%rsp) + vmovdqa32 %ymm16,32(%rsp) + +.Ldone_avx512vl: + vzeroall + leaq -8(%r10),%rsp +.Lavx512vl_epilogue: + ret + +.align 32 +.Lchacha20_8xvl: + leaq 8(%rsp),%r10 + subq $64+8,%rsp + andq $-64,%rsp + vzeroupper + + leaq .Lsigma(%rip),%r9 + vbroadcasti128 (%r9),%ymm3 + vbroadcasti128 (%rcx),%ymm7 + vbroadcasti128 16(%rcx),%ymm11 + vbroadcasti128 (%r8),%ymm15 + + vpshufd $0x00,%ymm3,%ymm0 + vpshufd $0x55,%ymm3,%ymm1 + vpshufd $0xaa,%ymm3,%ymm2 + vpshufd $0xff,%ymm3,%ymm3 + vmovdqa64 %ymm0,%ymm16 + vmovdqa64 %ymm1,%ymm17 + vmovdqa64 %ymm2,%ymm18 + vmovdqa64 %ymm3,%ymm19 + + vpshufd $0x00,%ymm7,%ymm4 + vpshufd $0x55,%ymm7,%ymm5 + vpshufd $0xaa,%ymm7,%ymm6 + vpshufd $0xff,%ymm7,%ymm7 + vmovdqa64 %ymm4,%ymm20 + vmovdqa64 %ymm5,%ymm21 + vmovdqa64 %ymm6,%ymm22 + vmovdqa64 %ymm7,%ymm23 + + vpshufd $0x00,%ymm11,%ymm8 + vpshufd $0x55,%ymm11,%ymm9 + vpshufd $0xaa,%ymm11,%ymm10 + vpshufd $0xff,%ymm11,%ymm11 + vmovdqa64 %ymm8,%ymm24 + vmovdqa64 %ymm9,%ymm25 + vmovdqa64 %ymm10,%ymm26 + vmovdqa64 %ymm11,%ymm27 + + vpshufd $0x00,%ymm15,%ymm12 + vpshufd $0x55,%ymm15,%ymm13 + vpshufd $0xaa,%ymm15,%ymm14 + vpshufd $0xff,%ymm15,%ymm15 + vpaddd .Lincy(%rip),%ymm12,%ymm12 + vmovdqa64 %ymm12,%ymm28 + vmovdqa64 %ymm13,%ymm29 + vmovdqa64 %ymm14,%ymm30 + vmovdqa64 %ymm15,%ymm31 + + movl $10,%eax + jmp .Loop8xvl + +.align 32 +.Loop_outer8xvl: + + + vpbroadcastd 8(%r9),%ymm2 + vpbroadcastd 12(%r9),%ymm3 + vpaddd .Leight(%rip),%ymm28,%ymm28 + vmovdqa64 %ymm20,%ymm4 + vmovdqa64 %ymm21,%ymm5 + vmovdqa64 %ymm22,%ymm6 + vmovdqa64 %ymm23,%ymm7 + vmovdqa64 %ymm24,%ymm8 + vmovdqa64 %ymm25,%ymm9 + vmovdqa64 %ymm26,%ymm10 + vmovdqa64 %ymm27,%ymm11 + vmovdqa64 %ymm28,%ymm12 + vmovdqa64 %ymm29,%ymm13 + vmovdqa64 %ymm30,%ymm14 + vmovdqa64 %ymm31,%ymm15 + + vmovdqa64 %ymm0,%ymm16 + vmovdqa64 %ymm1,%ymm17 + vmovdqa64 %ymm2,%ymm18 + vmovdqa64 %ymm3,%ymm19 + + movl $10,%eax + jmp .Loop8xvl + +.align 32 +.Loop8xvl: + vpaddd %ymm4,%ymm0,%ymm0 + vpaddd %ymm5,%ymm1,%ymm1 + vpaddd %ymm6,%ymm2,%ymm2 + vpaddd %ymm7,%ymm3,%ymm3 + vpxor %ymm0,%ymm12,%ymm12 + vpxor %ymm1,%ymm13,%ymm13 + vpxor %ymm2,%ymm14,%ymm14 + vpxor %ymm3,%ymm15,%ymm15 + vprold $16,%ymm12,%ymm12 + vprold $16,%ymm13,%ymm13 + vprold $16,%ymm14,%ymm14 + vprold $16,%ymm15,%ymm15 + vpaddd %ymm12,%ymm8,%ymm8 + vpaddd %ymm13,%ymm9,%ymm9 + vpaddd %ymm14,%ymm10,%ymm10 + vpaddd %ymm15,%ymm11,%ymm11 + vpxor %ymm8,%ymm4,%ymm4 + vpxor %ymm9,%ymm5,%ymm5 + vpxor %ymm10,%ymm6,%ymm6 + vpxor %ymm11,%ymm7,%ymm7 + vprold $12,%ymm4,%ymm4 + vprold $12,%ymm5,%ymm5 + vprold $12,%ymm6,%ymm6 + vprold $12,%ymm7,%ymm7 + vpaddd %ymm4,%ymm0,%ymm0 + vpaddd %ymm5,%ymm1,%ymm1 + vpaddd %ymm6,%ymm2,%ymm2 + vpaddd %ymm7,%ymm3,%ymm3 + vpxor %ymm0,%ymm12,%ymm12 + vpxor %ymm1,%ymm13,%ymm13 + vpxor %ymm2,%ymm14,%ymm14 + vpxor %ymm3,%ymm15,%ymm15 + vprold $8,%ymm12,%ymm12 + vprold $8,%ymm13,%ymm13 + vprold $8,%ymm14,%ymm14 + vprold $8,%ymm15,%ymm15 + vpaddd %ymm12,%ymm8,%ymm8 + vpaddd %ymm13,%ymm9,%ymm9 + vpaddd %ymm14,%ymm10,%ymm10 + vpaddd %ymm15,%ymm11,%ymm11 + vpxor %ymm8,%ymm4,%ymm4 + vpxor %ymm9,%ymm5,%ymm5 + vpxor %ymm10,%ymm6,%ymm6 + vpxor %ymm11,%ymm7,%ymm7 + vprold $7,%ymm4,%ymm4 + vprold $7,%ymm5,%ymm5 + vprold $7,%ymm6,%ymm6 + vprold $7,%ymm7,%ymm7 + vpaddd %ymm5,%ymm0,%ymm0 + vpaddd %ymm6,%ymm1,%ymm1 + vpaddd %ymm7,%ymm2,%ymm2 + vpaddd %ymm4,%ymm3,%ymm3 + vpxor %ymm0,%ymm15,%ymm15 + vpxor %ymm1,%ymm12,%ymm12 + vpxor %ymm2,%ymm13,%ymm13 + vpxor %ymm3,%ymm14,%ymm14 + vprold $16,%ymm15,%ymm15 + vprold $16,%ymm12,%ymm12 + vprold $16,%ymm13,%ymm13 + vprold $16,%ymm14,%ymm14 + vpaddd %ymm15,%ymm10,%ymm10 + vpaddd %ymm12,%ymm11,%ymm11 + vpaddd %ymm13,%ymm8,%ymm8 + vpaddd %ymm14,%ymm9,%ymm9 + vpxor %ymm10,%ymm5,%ymm5 + vpxor %ymm11,%ymm6,%ymm6 + vpxor %ymm8,%ymm7,%ymm7 + vpxor %ymm9,%ymm4,%ymm4 + vprold $12,%ymm5,%ymm5 + vprold $12,%ymm6,%ymm6 + vprold $12,%ymm7,%ymm7 + vprold $12,%ymm4,%ymm4 + vpaddd %ymm5,%ymm0,%ymm0 + vpaddd %ymm6,%ymm1,%ymm1 + vpaddd %ymm7,%ymm2,%ymm2 + vpaddd %ymm4,%ymm3,%ymm3 + vpxor %ymm0,%ymm15,%ymm15 + vpxor %ymm1,%ymm12,%ymm12 + vpxor %ymm2,%ymm13,%ymm13 + vpxor %ymm3,%ymm14,%ymm14 + vprold $8,%ymm15,%ymm15 + vprold $8,%ymm12,%ymm12 + vprold $8,%ymm13,%ymm13 + vprold $8,%ymm14,%ymm14 + vpaddd %ymm15,%ymm10,%ymm10 + vpaddd %ymm12,%ymm11,%ymm11 + vpaddd %ymm13,%ymm8,%ymm8 + vpaddd %ymm14,%ymm9,%ymm9 + vpxor %ymm10,%ymm5,%ymm5 + vpxor %ymm11,%ymm6,%ymm6 + vpxor %ymm8,%ymm7,%ymm7 + vpxor %ymm9,%ymm4,%ymm4 + vprold $7,%ymm5,%ymm5 + vprold $7,%ymm6,%ymm6 + vprold $7,%ymm7,%ymm7 + vprold $7,%ymm4,%ymm4 + decl %eax + jnz .Loop8xvl + + vpaddd %ymm16,%ymm0,%ymm0 + vpaddd %ymm17,%ymm1,%ymm1 + vpaddd %ymm18,%ymm2,%ymm2 + vpaddd %ymm19,%ymm3,%ymm3 + + vpunpckldq %ymm1,%ymm0,%ymm18 + vpunpckldq %ymm3,%ymm2,%ymm19 + vpunpckhdq %ymm1,%ymm0,%ymm0 + vpunpckhdq %ymm3,%ymm2,%ymm2 + vpunpcklqdq %ymm19,%ymm18,%ymm1 + vpunpckhqdq %ymm19,%ymm18,%ymm18 + vpunpcklqdq %ymm2,%ymm0,%ymm3 + vpunpckhqdq %ymm2,%ymm0,%ymm0 + vpaddd %ymm20,%ymm4,%ymm4 + vpaddd %ymm21,%ymm5,%ymm5 + vpaddd %ymm22,%ymm6,%ymm6 + vpaddd %ymm23,%ymm7,%ymm7 + + vpunpckldq %ymm5,%ymm4,%ymm2 + vpunpckldq %ymm7,%ymm6,%ymm19 + vpunpckhdq %ymm5,%ymm4,%ymm4 + vpunpckhdq %ymm7,%ymm6,%ymm6 + vpunpcklqdq %ymm19,%ymm2,%ymm5 + vpunpckhqdq %ymm19,%ymm2,%ymm2 + vpunpcklqdq %ymm6,%ymm4,%ymm7 + vpunpckhqdq %ymm6,%ymm4,%ymm4 + vshufi32x4 $0,%ymm5,%ymm1,%ymm19 + vshufi32x4 $3,%ymm5,%ymm1,%ymm5 + vshufi32x4 $0,%ymm2,%ymm18,%ymm1 + vshufi32x4 $3,%ymm2,%ymm18,%ymm2 + vshufi32x4 $0,%ymm7,%ymm3,%ymm18 + vshufi32x4 $3,%ymm7,%ymm3,%ymm7 + vshufi32x4 $0,%ymm4,%ymm0,%ymm3 + vshufi32x4 $3,%ymm4,%ymm0,%ymm4 + vpaddd %ymm24,%ymm8,%ymm8 + vpaddd %ymm25,%ymm9,%ymm9 + vpaddd %ymm26,%ymm10,%ymm10 + vpaddd %ymm27,%ymm11,%ymm11 + + vpunpckldq %ymm9,%ymm8,%ymm6 + vpunpckldq %ymm11,%ymm10,%ymm0 + vpunpckhdq %ymm9,%ymm8,%ymm8 + vpunpckhdq %ymm11,%ymm10,%ymm10 + vpunpcklqdq %ymm0,%ymm6,%ymm9 + vpunpckhqdq %ymm0,%ymm6,%ymm6 + vpunpcklqdq %ymm10,%ymm8,%ymm11 + vpunpckhqdq %ymm10,%ymm8,%ymm8 + vpaddd %ymm28,%ymm12,%ymm12 + vpaddd %ymm29,%ymm13,%ymm13 + vpaddd %ymm30,%ymm14,%ymm14 + vpaddd %ymm31,%ymm15,%ymm15 + + vpunpckldq %ymm13,%ymm12,%ymm10 + vpunpckldq %ymm15,%ymm14,%ymm0 + vpunpckhdq %ymm13,%ymm12,%ymm12 + vpunpckhdq %ymm15,%ymm14,%ymm14 + vpunpcklqdq %ymm0,%ymm10,%ymm13 + vpunpckhqdq %ymm0,%ymm10,%ymm10 + vpunpcklqdq %ymm14,%ymm12,%ymm15 + vpunpckhqdq %ymm14,%ymm12,%ymm12 + vperm2i128 $0x20,%ymm13,%ymm9,%ymm0 + vperm2i128 $0x31,%ymm13,%ymm9,%ymm13 + vperm2i128 $0x20,%ymm10,%ymm6,%ymm9 + vperm2i128 $0x31,%ymm10,%ymm6,%ymm10 + vperm2i128 $0x20,%ymm15,%ymm11,%ymm6 + vperm2i128 $0x31,%ymm15,%ymm11,%ymm15 + vperm2i128 $0x20,%ymm12,%ymm8,%ymm11 + vperm2i128 $0x31,%ymm12,%ymm8,%ymm12 + cmpq $512,%rdx + jb .Ltail8xvl + + movl $0x80,%eax + vpxord 0(%rsi),%ymm19,%ymm19 + vpxor 32(%rsi),%ymm0,%ymm0 + vpxor 64(%rsi),%ymm5,%ymm5 + vpxor 96(%rsi),%ymm13,%ymm13 + leaq (%rsi,%rax,1),%rsi + vmovdqu32 %ymm19,0(%rdi) + vmovdqu %ymm0,32(%rdi) + vmovdqu %ymm5,64(%rdi) + vmovdqu %ymm13,96(%rdi) + leaq (%rdi,%rax,1),%rdi + + vpxor 0(%rsi),%ymm1,%ymm1 + vpxor 32(%rsi),%ymm9,%ymm9 + vpxor 64(%rsi),%ymm2,%ymm2 + vpxor 96(%rsi),%ymm10,%ymm10 + leaq (%rsi,%rax,1),%rsi + vmovdqu %ymm1,0(%rdi) + vmovdqu %ymm9,32(%rdi) + vmovdqu %ymm2,64(%rdi) + vmovdqu %ymm10,96(%rdi) + leaq (%rdi,%rax,1),%rdi + + vpxord 0(%rsi),%ymm18,%ymm18 + vpxor 32(%rsi),%ymm6,%ymm6 + vpxor 64(%rsi),%ymm7,%ymm7 + vpxor 96(%rsi),%ymm15,%ymm15 + leaq (%rsi,%rax,1),%rsi + vmovdqu32 %ymm18,0(%rdi) + vmovdqu %ymm6,32(%rdi) + vmovdqu %ymm7,64(%rdi) + vmovdqu %ymm15,96(%rdi) + leaq (%rdi,%rax,1),%rdi + + vpxor 0(%rsi),%ymm3,%ymm3 + vpxor 32(%rsi),%ymm11,%ymm11 + vpxor 64(%rsi),%ymm4,%ymm4 + vpxor 96(%rsi),%ymm12,%ymm12 + leaq (%rsi,%rax,1),%rsi + vmovdqu %ymm3,0(%rdi) + vmovdqu %ymm11,32(%rdi) + vmovdqu %ymm4,64(%rdi) + vmovdqu %ymm12,96(%rdi) + leaq (%rdi,%rax,1),%rdi + + vpbroadcastd 0(%r9),%ymm0 + vpbroadcastd 4(%r9),%ymm1 + + subq $512,%rdx + jnz .Loop_outer8xvl + + jmp .Ldone8xvl + +.align 32 +.Ltail8xvl: + vmovdqa64 %ymm19,%ymm8 + xorq %r9,%r9 + subq %rsi,%rdi + cmpq $64,%rdx + jb .Less_than_64_8xvl + vpxor 0(%rsi),%ymm8,%ymm8 + vpxor 32(%rsi),%ymm0,%ymm0 + vmovdqu %ymm8,0(%rdi,%rsi,1) + vmovdqu %ymm0,32(%rdi,%rsi,1) + je .Ldone8xvl + vmovdqa %ymm5,%ymm8 + vmovdqa %ymm13,%ymm0 + leaq 64(%rsi),%rsi + + cmpq $128,%rdx + jb .Less_than_64_8xvl + vpxor 0(%rsi),%ymm5,%ymm5 + vpxor 32(%rsi),%ymm13,%ymm13 + vmovdqu %ymm5,0(%rdi,%rsi,1) + vmovdqu %ymm13,32(%rdi,%rsi,1) + je .Ldone8xvl + vmovdqa %ymm1,%ymm8 + vmovdqa %ymm9,%ymm0 + leaq 64(%rsi),%rsi + + cmpq $192,%rdx + jb .Less_than_64_8xvl + vpxor 0(%rsi),%ymm1,%ymm1 + vpxor 32(%rsi),%ymm9,%ymm9 + vmovdqu %ymm1,0(%rdi,%rsi,1) + vmovdqu %ymm9,32(%rdi,%rsi,1) + je .Ldone8xvl + vmovdqa %ymm2,%ymm8 + vmovdqa %ymm10,%ymm0 + leaq 64(%rsi),%rsi + + cmpq $256,%rdx + jb .Less_than_64_8xvl + vpxor 0(%rsi),%ymm2,%ymm2 + vpxor 32(%rsi),%ymm10,%ymm10 + vmovdqu %ymm2,0(%rdi,%rsi,1) + vmovdqu %ymm10,32(%rdi,%rsi,1) + je .Ldone8xvl + vmovdqa32 %ymm18,%ymm8 + vmovdqa %ymm6,%ymm0 + leaq 64(%rsi),%rsi + + cmpq $320,%rdx + jb .Less_than_64_8xvl + vpxord 0(%rsi),%ymm18,%ymm18 + vpxor 32(%rsi),%ymm6,%ymm6 + vmovdqu32 %ymm18,0(%rdi,%rsi,1) + vmovdqu %ymm6,32(%rdi,%rsi,1) + je .Ldone8xvl + vmovdqa %ymm7,%ymm8 + vmovdqa %ymm15,%ymm0 + leaq 64(%rsi),%rsi + + cmpq $384,%rdx + jb .Less_than_64_8xvl + vpxor 0(%rsi),%ymm7,%ymm7 + vpxor 32(%rsi),%ymm15,%ymm15 + vmovdqu %ymm7,0(%rdi,%rsi,1) + vmovdqu %ymm15,32(%rdi,%rsi,1) + je .Ldone8xvl + vmovdqa %ymm3,%ymm8 + vmovdqa %ymm11,%ymm0 + leaq 64(%rsi),%rsi + + cmpq $448,%rdx + jb .Less_than_64_8xvl + vpxor 0(%rsi),%ymm3,%ymm3 + vpxor 32(%rsi),%ymm11,%ymm11 + vmovdqu %ymm3,0(%rdi,%rsi,1) + vmovdqu %ymm11,32(%rdi,%rsi,1) + je .Ldone8xvl + vmovdqa %ymm4,%ymm8 + vmovdqa %ymm12,%ymm0 + leaq 64(%rsi),%rsi + +.Less_than_64_8xvl: + vmovdqa %ymm8,0(%rsp) + vmovdqa %ymm0,32(%rsp) + leaq (%rdi,%rsi,1),%rdi + andq $63,%rdx + +.Loop_tail8xvl: + movzbl (%rsi,%r9,1),%eax + movzbl (%rsp,%r9,1),%ecx + leaq 1(%r9),%r9 + xorl %ecx,%eax + movb %al,-1(%rdi,%r9,1) + decq %rdx + jnz .Loop_tail8xvl + + vpxor %ymm8,%ymm8,%ymm8 + vmovdqa %ymm8,0(%rsp) + vmovdqa %ymm8,32(%rsp) + +.Ldone8xvl: + vzeroall + leaq -8(%r10),%rsp +.L8xvl_epilogue: + ret +ENDPROC(chacha20_avx512vl) + #endif /* CONFIG_AS_AVX512 */ diff --git a/src/crypto/chacha20poly1305.c b/src/crypto/chacha20poly1305.c index af3ce35..bde923b 100644 --- a/src/crypto/chacha20poly1305.c +++ b/src/crypto/chacha20poly1305.c @@ -33,6 +33,7 @@ asmlinkage void poly1305_blocks_avx2(void *ctx, const u8 *inp, size_t len, u32 p #endif #ifdef CONFIG_AS_AVX512 asmlinkage void chacha20_avx512(u8 *out, const u8 *in, size_t len, const u32 key[8], const u32 counter[4]); +asmlinkage void chacha20_avx512vl(u8 *out, const u8 *in, size_t len, const u32 key[8], const u32 counter[4]); asmlinkage void poly1305_blocks_avx512(void *ctx, const u8 *inp, size_t len, u32 padbit); #endif -- cgit v1.2.3-59-g8ed1b