From 35be6880788d90bc2a93de5295391b001a8a6a46 Mon Sep 17 00:00:00 2001 From: David Benjamin Date: Fri, 19 Feb 2016 18:47:22 -0500 Subject: [PATCH] Enable upstream's ChaCha20 assembly for x86 and ARM (32- and 64-bit). This removes chacha_vec_arm.S and chacha_vec.c in favor of unifying on upstream's code. Upstream's is faster and this cuts down on the number of distinct codepaths. Our old scheme also didn't give vectorized code on Windows or aarch64. BoringSSL-specific modifications made to the assembly: - As usual, the shelling out to $CC is replaced with hardcoding $avx. I've tested up to the AVX2 codepath, so enable it all. - I've removed the AMD XOP code as I have not tested it. - As usual, the ARM file need the arm_arch.h include tweaked. Speed numbers follow. We can hope for further wins on these benchmarks after importing the Poly1305 assembly. x86 --- Old: Did 1422000 ChaCha20-Poly1305 (16 bytes) seal operations in 1000433us (1421384.5 ops/sec): 22.7 MB/s Did 123000 ChaCha20-Poly1305 (1350 bytes) seal operations in 1003803us (122534.0 ops/sec): 165.4 MB/s Did 22000 ChaCha20-Poly1305 (8192 bytes) seal operations in 1000282us (21993.8 ops/sec): 180.2 MB/s Did 1428000 ChaCha20-Poly1305-Old (16 bytes) seal operations in 1000214us (1427694.5 ops/sec): 22.8 MB/s Did 124000 ChaCha20-Poly1305-Old (1350 bytes) seal operations in 1006332us (123219.8 ops/sec): 166.3 MB/s Did 22000 ChaCha20-Poly1305-Old (8192 bytes) seal operations in 1020771us (21552.3 ops/sec): 176.6 MB/s New: Did 1520000 ChaCha20-Poly1305 (16 bytes) seal operations in 1000567us (1519138.6 ops/sec): 24.3 MB/s Did 152000 ChaCha20-Poly1305 (1350 bytes) seal operations in 1004216us (151361.9 ops/sec): 204.3 MB/s Did 31000 ChaCha20-Poly1305 (8192 bytes) seal operations in 1009085us (30720.9 ops/sec): 251.7 MB/s Did 1797000 ChaCha20-Poly1305-Old (16 bytes) seal operations in 1000141us (1796746.7 ops/sec): 28.7 MB/s Did 171000 ChaCha20-Poly1305-Old (1350 bytes) seal operations in 1003204us (170453.9 ops/sec): 230.1 MB/s Did 31000 ChaCha20-Poly1305-Old (8192 bytes) seal operations in 1005349us (30835.1 ops/sec): 252.6 MB/s x86_64, no AVX2 --- Old: Did 1782000 ChaCha20-Poly1305 (16 bytes) seal operations in 1000204us (1781636.5 ops/sec): 28.5 MB/s Did 317000 ChaCha20-Poly1305 (1350 bytes) seal operations in 1001579us (316500.2 ops/sec): 427.3 MB/s Did 62000 ChaCha20-Poly1305 (8192 bytes) seal operations in 1012146us (61256.0 ops/sec): 501.8 MB/s Did 1778000 ChaCha20-Poly1305-Old (16 bytes) seal operations in 1000220us (1777608.9 ops/sec): 28.4 MB/s Did 315000 ChaCha20-Poly1305-Old (1350 bytes) seal operations in 1002886us (314093.5 ops/sec): 424.0 MB/s Did 71000 ChaCha20-Poly1305-Old (8192 bytes) seal operations in 1014606us (69977.9 ops/sec): 573.3 MB/s New: Did 1866000 ChaCha20-Poly1305 (16 bytes) seal operations in 1000019us (1865964.5 ops/sec): 29.9 MB/s Did 399000 ChaCha20-Poly1305 (1350 bytes) seal operations in 1001017us (398594.6 ops/sec): 538.1 MB/s Did 84000 ChaCha20-Poly1305 (8192 bytes) seal operations in 1005645us (83528.5 ops/sec): 684.3 MB/s Did 1881000 ChaCha20-Poly1305-Old (16 bytes) seal operations in 1000325us (1880388.9 ops/sec): 30.1 MB/s Did 404000 ChaCha20-Poly1305-Old (1350 bytes) seal operations in 1000004us (403998.4 ops/sec): 545.4 MB/s Did 85000 ChaCha20-Poly1305-Old (8192 bytes) seal operations in 1010048us (84154.4 ops/sec): 689.4 MB/s x86_64, AVX2 --- Old: Did 2375000 ChaCha20-Poly1305 (16 bytes) seal operations in 1000282us (2374330.4 ops/sec): 38.0 MB/s Did 448000 ChaCha20-Poly1305 (1350 bytes) seal operations in 1001865us (447166.0 ops/sec): 603.7 MB/s Did 88000 ChaCha20-Poly1305 (8192 bytes) seal operations in 1005217us (87543.3 ops/sec): 717.2 MB/s Did 2409000 ChaCha20-Poly1305-Old (16 bytes) seal operations in 1000188us (2408547.2 ops/sec): 38.5 MB/s Did 446000 ChaCha20-Poly1305-Old (1350 bytes) seal operations in 1001003us (445553.1 ops/sec): 601.5 MB/s Did 90000 ChaCha20-Poly1305-Old (8192 bytes) seal operations in 1006722us (89399.1 ops/sec): 732.4 MB/s New: Did 2622000 ChaCha20-Poly1305 (16 bytes) seal operations in 1000266us (2621302.7 ops/sec): 41.9 MB/s Did 794000 ChaCha20-Poly1305 (1350 bytes) seal operations in 1000783us (793378.8 ops/sec): 1071.1 MB/s Did 173000 ChaCha20-Poly1305 (8192 bytes) seal operations in 1000176us (172969.6 ops/sec): 1417.0 MB/s Did 2623000 ChaCha20-Poly1305-Old (16 bytes) seal operations in 1000330us (2622134.7 ops/sec): 42.0 MB/s Did 783000 ChaCha20-Poly1305-Old (1350 bytes) seal operations in 1000531us (782584.4 ops/sec): 1056.5 MB/s Did 174000 ChaCha20-Poly1305-Old (8192 bytes) seal operations in 1000840us (173854.0 ops/sec): 1424.2 MB/s arm, Nexus 4 --- Old: Did 388550 ChaCha20-Poly1305 (16 bytes) seal operations in 1000580us (388324.8 ops/sec): 6.2 MB/s Did 90000 ChaCha20-Poly1305 (1350 bytes) seal operations in 1003816us (89657.9 ops/sec): 121.0 MB/s Did 19000 ChaCha20-Poly1305 (8192 bytes) seal operations in 1045750us (18168.8 ops/sec): 148.8 MB/s Did 398500 ChaCha20-Poly1305-Old (16 bytes) seal operations in 1000305us (398378.5 ops/sec): 6.4 MB/s Did 90500 ChaCha20-Poly1305-Old (1350 bytes) seal operations in 1000305us (90472.4 ops/sec): 122.1 MB/s Did 19000 ChaCha20-Poly1305-Old (8192 bytes) seal operations in 1043278us (18211.8 ops/sec): 149.2 MB/s New: Did 424788 ChaCha20-Poly1305 (16 bytes) seal operations in 1000641us (424515.9 ops/sec): 6.8 MB/s Did 115000 ChaCha20-Poly1305 (1350 bytes) seal operations in 1001526us (114824.8 ops/sec): 155.0 MB/s Did 27000 ChaCha20-Poly1305 (8192 bytes) seal operations in 1033023us (26136.9 ops/sec): 214.1 MB/s Did 447750 ChaCha20-Poly1305-Old (16 bytes) seal operations in 1000549us (447504.3 ops/sec): 7.2 MB/s Did 117500 ChaCha20-Poly1305-Old (1350 bytes) seal operations in 1001923us (117274.5 ops/sec): 158.3 MB/s Did 27000 ChaCha20-Poly1305-Old (8192 bytes) seal operations in 1025118us (26338.4 ops/sec): 215.8 MB/s aarch64, Nexus 6p (Note we didn't have aarch64 assembly before at all, and still don't have it for Poly1305. Hopefully once that's added this will be faster than the arm numbers...) --- Old: Did 145040 ChaCha20-Poly1305 (16 bytes) seal operations in 1003065us (144596.8 ops/sec): 2.3 MB/s Did 14000 ChaCha20-Poly1305 (1350 bytes) seal operations in 1042605us (13427.9 ops/sec): 18.1 MB/s Did 2618 ChaCha20-Poly1305 (8192 bytes) seal operations in 1093241us (2394.7 ops/sec): 19.6 MB/s Did 148000 ChaCha20-Poly1305-Old (16 bytes) seal operations in 1000709us (147895.1 ops/sec): 2.4 MB/s Did 14000 ChaCha20-Poly1305-Old (1350 bytes) seal operations in 1047294us (13367.8 ops/sec): 18.0 MB/s Did 2607 ChaCha20-Poly1305-Old (8192 bytes) seal operations in 1090745us (2390.1 ops/sec): 19.6 MB/s New: Did 358000 ChaCha20-Poly1305 (16 bytes) seal operations in 1000769us (357724.9 ops/sec): 5.7 MB/s Did 45000 ChaCha20-Poly1305 (1350 bytes) seal operations in 1021267us (44062.9 ops/sec): 59.5 MB/s Did 8591 ChaCha20-Poly1305 (8192 bytes) seal operations in 1047136us (8204.3 ops/sec): 67.2 MB/s Did 343000 ChaCha20-Poly1305-Old (16 bytes) seal operations in 1000489us (342832.4 ops/sec): 5.5 MB/s Did 44000 ChaCha20-Poly1305-Old (1350 bytes) seal operations in 1008326us (43636.7 ops/sec): 58.9 MB/s Did 8866 ChaCha20-Poly1305-Old (8192 bytes) seal operations in 1083341us (8183.9 ops/sec): 67.0 MB/s Change-Id: I629fe195d072f2c99e8f947578fad6d70823c4c8 Reviewed-on: https://boringssl-review.googlesource.com/7202 Reviewed-by: Adam Langley --- BUILDING.md | 10 - crypto/chacha/CMakeLists.txt | 34 +- crypto/chacha/asm/chacha-armv4.pl | 2 +- crypto/chacha/asm/chacha-armv8.pl | 2 +- crypto/chacha/asm/chacha-x86.pl | 385 +---- crypto/chacha/asm/chacha-x86_64.pl | 476 +----- crypto/chacha/{chacha_generic.c => chacha.c} | 64 +- crypto/chacha/chacha_vec.c | 328 ---- crypto/chacha/chacha_vec_arm.S | 1447 ------------------ crypto/chacha/chacha_vec_arm_generate.go | 153 -- util/generate_build_files.py | 1 - 11 files changed, 83 insertions(+), 2819 deletions(-) rename crypto/chacha/{chacha_generic.c => chacha.c} (73%) delete mode 100644 crypto/chacha/chacha_vec.c delete mode 100644 crypto/chacha/chacha_vec_arm.S delete mode 100644 crypto/chacha/chacha_vec_arm_generate.go diff --git a/BUILDING.md b/BUILDING.md index d40df9ef..e1113104 100644 --- a/BUILDING.md +++ b/BUILDING.md @@ -31,16 +31,6 @@ * [Go](https://golang.org/dl/) is required. If not found by CMake, the go executable may be configured explicitly by setting `GO_EXECUTABLE`. - * If you change crypto/chacha/chacha\_vec.c, you will need the - arm-linux-gnueabihf-gcc compiler: - - ``` - wget https://releases.linaro.org/14.11/components/toolchain/binaries/arm-linux-gnueabihf/gcc-linaro-4.9-2014.11-x86_64_arm-linux-gnueabihf.tar.xz && \ - echo bc4ca2ced084d2dc12424815a4442e19cb1422db87068830305d90075feb1a3b gcc-linaro-4.9-2014.11-x86_64_arm-linux-gnueabihf.tar.xz | sha256sum -c && \ - tar xf gcc-linaro-4.9-2014.11-x86_64_arm-linux-gnueabihf.tar.xz && \ - sudo mv gcc-linaro-4.9-2014.11-x86_64_arm-linux-gnueabihf /opt/ - ``` - ## Building Using Ninja (note the 'N' is capitalized in the cmake invocation): diff --git a/crypto/chacha/CMakeLists.txt b/crypto/chacha/CMakeLists.txt index 266e8699..f9ab0247 100644 --- a/crypto/chacha/CMakeLists.txt +++ b/crypto/chacha/CMakeLists.txt @@ -4,7 +4,31 @@ if (${ARCH} STREQUAL "arm") set( CHACHA_ARCH_SOURCES - chacha_vec_arm.S + chacha-armv4.${ASM_EXT} + ) +endif() + +if (${ARCH} STREQUAL "aarch64") + set( + CHACHA_ARCH_SOURCES + + chacha-armv8.${ASM_EXT} + ) +endif() + +if (${ARCH} STREQUAL "x86") + set( + CHACHA_ARCH_SOURCES + + chacha-x86.${ASM_EXT} + ) +endif() + +if (${ARCH} STREQUAL "x86_64") + set( + CHACHA_ARCH_SOURCES + + chacha-x86_64.${ASM_EXT} ) endif() @@ -13,8 +37,12 @@ add_library( OBJECT - chacha_generic.c - chacha_vec.c + chacha.c ${CHACHA_ARCH_SOURCES} ) + +perlasm(chacha-armv4.${ASM_EXT} asm/chacha-armv4.pl) +perlasm(chacha-armv8.${ASM_EXT} asm/chacha-armv8.pl) +perlasm(chacha-x86.${ASM_EXT} asm/chacha-x86.pl) +perlasm(chacha-x86_64.${ASM_EXT} asm/chacha-x86_64.pl) \ No newline at end of file diff --git a/crypto/chacha/asm/chacha-armv4.pl b/crypto/chacha/asm/chacha-armv4.pl index 4545101c..b190445d 100755 --- a/crypto/chacha/asm/chacha-armv4.pl +++ b/crypto/chacha/asm/chacha-armv4.pl @@ -162,7 +162,7 @@ my @ret; } $code.=<<___; -#include "arm_arch.h" +#include .text #if defined(__thumb2__) diff --git a/crypto/chacha/asm/chacha-armv8.pl b/crypto/chacha/asm/chacha-armv8.pl index 6ddb31fb..e6fa144c 100755 --- a/crypto/chacha/asm/chacha-armv8.pl +++ b/crypto/chacha/asm/chacha-armv8.pl @@ -111,7 +111,7 @@ my ($a3,$b3,$c3,$d3)=map(($_&~3)+(($_+1)&3),($a2,$b2,$c2,$d2)); } $code.=<<___; -#include "arm_arch.h" +#include .text diff --git a/crypto/chacha/asm/chacha-x86.pl b/crypto/chacha/asm/chacha-x86.pl index 850c9172..edce43dc 100755 --- a/crypto/chacha/asm/chacha-x86.pl +++ b/crypto/chacha/asm/chacha-x86.pl @@ -26,6 +26,8 @@ # Bulldozer 13.4/+50% 4.38(*) # # (*) Bulldozer actually executes 4xXOP code path that delivers 3.55; +# +# Modified from upstream OpenSSL to remove the XOP code. $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; push(@INC,"${dir}","${dir}../../perlasm"); @@ -36,22 +38,7 @@ require "x86asm.pl"; $xmm=$ymm=0; for (@ARGV) { $xmm=1 if (/-DOPENSSL_IA32_SSE2/); } -$ymm=1 if ($xmm && - `$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1` - =~ /GNU assembler version ([2-9]\.[0-9]+)/ && - $1>=2.19); # first version supporting AVX - -$ymm=1 if ($xmm && !$ymm && $ARGV[0] eq "win32n" && - `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/ && - $1>=2.03); # first version supporting AVX - -$ymm=1 if ($xmm && !$ymm && $ARGV[0] eq "win32" && - `ml 2>&1` =~ /Version ([0-9]+)\./ && - $1>=10); # first version supporting AVX - -$ymm=1 if ($xmm && !$ymm && - `$ENV{CC} -v 2>&1` =~ /(^clang version|based on LLVM) ([3-9]\.[0-9]+)/ && - $2>=3.0); # first version supporting AVX +$ymm=$xmm; $a="eax"; ($b,$b_)=("ebx","ebp"); @@ -118,7 +105,6 @@ my ($ap,$bp,$cp,$dp)=map(($_&~3)+(($_-1)&3),($ai,$bi,$ci,$di)); # previous } &static_label("ssse3_shortcut"); -&static_label("xop_shortcut"); &static_label("ssse3_data"); &static_label("pic_point"); @@ -434,9 +420,6 @@ my ($ap,$bp,$cp,$dp)=map(($_&~3)+(($_-1)&3),($ai,$bi,$ci,$di)); # previous &function_begin("ChaCha20_ssse3"); &set_label("ssse3_shortcut"); - &test (&DWP(4,"ebp"),1<<11); # test XOP bit - &jnz (&label("xop_shortcut")); - &mov ($out,&wparam(0)); &mov ($inp,&wparam(1)); &mov ($len,&wparam(2)); @@ -767,366 +750,4 @@ sub SSSE3ROUND { # critical path is 20 "SIMD ticks" per round } &asciz ("ChaCha20 for x86, CRYPTOGAMS by "); -if ($xmm) { -my ($xa,$xa_,$xb,$xb_,$xc,$xc_,$xd,$xd_)=map("xmm$_",(0..7)); -my ($out,$inp,$len)=("edi","esi","ecx"); - -sub QUARTERROUND_XOP { -my ($ai,$bi,$ci,$di,$i)=@_; -my ($an,$bn,$cn,$dn)=map(($_&~3)+(($_+1)&3),($ai,$bi,$ci,$di)); # next -my ($ap,$bp,$cp,$dp)=map(($_&~3)+(($_-1)&3),($ai,$bi,$ci,$di)); # previous - - # a b c d - # - # 0 4 8 12 < even round - # 1 5 9 13 - # 2 6 10 14 - # 3 7 11 15 - # 0 5 10 15 < odd round - # 1 6 11 12 - # 2 7 8 13 - # 3 4 9 14 - - if ($i==0) { - my $j=4; - ($ap,$bp,$cp,$dp)=map(($_&~3)+(($_-$j--)&3),($ap,$bp,$cp,$dp)); - } elsif ($i==3) { - my $j=0; - ($an,$bn,$cn,$dn)=map(($_&~3)+(($_+$j++)&3),($an,$bn,$cn,$dn)); - } elsif ($i==4) { - my $j=4; - ($ap,$bp,$cp,$dp)=map(($_&~3)+(($_+$j--)&3),($ap,$bp,$cp,$dp)); - } elsif ($i==7) { - my $j=0; - ($an,$bn,$cn,$dn)=map(($_&~3)+(($_-$j++)&3),($an,$bn,$cn,$dn)); - } - - #&vpaddd ($xa,$xa,$xb); # see elsewhere - #&vpxor ($xd,$xd,$xa); # see elsewhere - &vmovdqa (&QWP(16*$cp-128,"ebx"),$xc_) if ($ai>0 && $ai<3); - &vprotd ($xd,$xd,16); - &vmovdqa (&QWP(16*$bp-128,"ebx"),$xb_) if ($i!=0); - &vpaddd ($xc,$xc,$xd); - &vmovdqa ($xc_,&QWP(16*$cn-128,"ebx")) if ($ai>0 && $ai<3); - &vpxor ($xb,$i!=0?$xb:$xb_,$xc); - &vmovdqa ($xa_,&QWP(16*$an-128,"ebx")); - &vprotd ($xb,$xb,12); - &vmovdqa ($xb_,&QWP(16*$bn-128,"ebx")) if ($i<7); - &vpaddd ($xa,$xa,$xb); - &vmovdqa ($xd_,&QWP(16*$dn-128,"ebx")) if ($di!=$dn); - &vpxor ($xd,$xd,$xa); - &vpaddd ($xa_,$xa_,$xb_) if ($i<7); # elsewhere - &vprotd ($xd,$xd,8); - &vmovdqa (&QWP(16*$ai-128,"ebx"),$xa); - &vpaddd ($xc,$xc,$xd); - &vmovdqa (&QWP(16*$di-128,"ebx"),$xd) if ($di!=$dn); - &vpxor ($xb,$xb,$xc); - &vpxor ($xd_,$di==$dn?$xd:$xd_,$xa_) if ($i<7); # elsewhere - &vprotd ($xb,$xb,7); - - ($xa,$xa_)=($xa_,$xa); - ($xb,$xb_)=($xb_,$xb); - ($xc,$xc_)=($xc_,$xc); - ($xd,$xd_)=($xd_,$xd); -} - -&function_begin("ChaCha20_xop"); -&set_label("xop_shortcut"); - &mov ($out,&wparam(0)); - &mov ($inp,&wparam(1)); - &mov ($len,&wparam(2)); - &mov ("edx",&wparam(3)); # key - &mov ("ebx",&wparam(4)); # counter and nonce - &vzeroupper (); - - &mov ("ebp","esp"); - &stack_push (131); - &and ("esp",-64); - &mov (&DWP(512,"esp"),"ebp"); - - &lea ("eax",&DWP(&label("ssse3_data")."-". - &label("pic_point"),"eax")); - &vmovdqu ("xmm3",&QWP(0,"ebx")); # counter and nonce - - &cmp ($len,64*4); - &jb (&label("1x")); - - &mov (&DWP(512+4,"esp"),"edx"); # offload pointers - &mov (&DWP(512+8,"esp"),"ebx"); - &sub ($len,64*4); # bias len - &lea ("ebp",&DWP(256+128,"esp")); # size optimization - - &vmovdqu ("xmm7",&QWP(0,"edx")); # key - &vpshufd ("xmm0","xmm3",0x00); - &vpshufd ("xmm1","xmm3",0x55); - &vpshufd ("xmm2","xmm3",0xaa); - &vpshufd ("xmm3","xmm3",0xff); - &vpaddd ("xmm0","xmm0",&QWP(16*3,"eax")); # fix counters - &vpshufd ("xmm4","xmm7",0x00); - &vpshufd ("xmm5","xmm7",0x55); - &vpsubd ("xmm0","xmm0",&QWP(16*4,"eax")); - &vpshufd ("xmm6","xmm7",0xaa); - &vpshufd ("xmm7","xmm7",0xff); - &vmovdqa (&QWP(16*12-128,"ebp"),"xmm0"); - &vmovdqa (&QWP(16*13-128,"ebp"),"xmm1"); - &vmovdqa (&QWP(16*14-128,"ebp"),"xmm2"); - &vmovdqa (&QWP(16*15-128,"ebp"),"xmm3"); - &vmovdqu ("xmm3",&QWP(16,"edx")); # key - &vmovdqa (&QWP(16*4-128,"ebp"),"xmm4"); - &vmovdqa (&QWP(16*5-128,"ebp"),"xmm5"); - &vmovdqa (&QWP(16*6-128,"ebp"),"xmm6"); - &vmovdqa (&QWP(16*7-128,"ebp"),"xmm7"); - &vmovdqa ("xmm7",&QWP(16*2,"eax")); # sigma - &lea ("ebx",&DWP(128,"esp")); # size optimization - - &vpshufd ("xmm0","xmm3",0x00); - &vpshufd ("xmm1","xmm3",0x55); - &vpshufd ("xmm2","xmm3",0xaa); - &vpshufd ("xmm3","xmm3",0xff); - &vpshufd ("xmm4","xmm7",0x00); - &vpshufd ("xmm5","xmm7",0x55); - &vpshufd ("xmm6","xmm7",0xaa); - &vpshufd ("xmm7","xmm7",0xff); - &vmovdqa (&QWP(16*8-128,"ebp"),"xmm0"); - &vmovdqa (&QWP(16*9-128,"ebp"),"xmm1"); - &vmovdqa (&QWP(16*10-128,"ebp"),"xmm2"); - &vmovdqa (&QWP(16*11-128,"ebp"),"xmm3"); - &vmovdqa (&QWP(16*0-128,"ebp"),"xmm4"); - &vmovdqa (&QWP(16*1-128,"ebp"),"xmm5"); - &vmovdqa (&QWP(16*2-128,"ebp"),"xmm6"); - &vmovdqa (&QWP(16*3-128,"ebp"),"xmm7"); - - &lea ($inp,&DWP(128,$inp)); # size optimization - &lea ($out,&DWP(128,$out)); # size optimization - &jmp (&label("outer_loop")); - -&set_label("outer_loop",32); - #&vmovdqa ("xmm0",&QWP(16*0-128,"ebp")); # copy key material - &vmovdqa ("xmm1",&QWP(16*1-128,"ebp")); - &vmovdqa ("xmm2",&QWP(16*2-128,"ebp")); - &vmovdqa ("xmm3",&QWP(16*3-128,"ebp")); - #&vmovdqa ("xmm4",&QWP(16*4-128,"ebp")); - &vmovdqa ("xmm5",&QWP(16*5-128,"ebp")); - &vmovdqa ("xmm6",&QWP(16*6-128,"ebp")); - &vmovdqa ("xmm7",&QWP(16*7-128,"ebp")); - #&vmovdqa (&QWP(16*0-128,"ebx"),"xmm0"); - &vmovdqa (&QWP(16*1-128,"ebx"),"xmm1"); - &vmovdqa (&QWP(16*2-128,"ebx"),"xmm2"); - &vmovdqa (&QWP(16*3-128,"ebx"),"xmm3"); - #&vmovdqa (&QWP(16*4-128,"ebx"),"xmm4"); - &vmovdqa (&QWP(16*5-128,"ebx"),"xmm5"); - &vmovdqa (&QWP(16*6-128,"ebx"),"xmm6"); - &vmovdqa (&QWP(16*7-128,"ebx"),"xmm7"); - #&vmovdqa ("xmm0",&QWP(16*8-128,"ebp")); - #&vmovdqa ("xmm1",&QWP(16*9-128,"ebp")); - &vmovdqa ("xmm2",&QWP(16*10-128,"ebp")); - &vmovdqa ("xmm3",&QWP(16*11-128,"ebp")); - &vmovdqa ("xmm4",&QWP(16*12-128,"ebp")); - &vmovdqa ("xmm5",&QWP(16*13-128,"ebp")); - &vmovdqa ("xmm6",&QWP(16*14-128,"ebp")); - &vmovdqa ("xmm7",&QWP(16*15-128,"ebp")); - &vpaddd ("xmm4","xmm4",&QWP(16*4,"eax")); # counter value - #&vmovdqa (&QWP(16*8-128,"ebx"),"xmm0"); - #&vmovdqa (&QWP(16*9-128,"ebx"),"xmm1"); - &vmovdqa (&QWP(16*10-128,"ebx"),"xmm2"); - &vmovdqa (&QWP(16*11-128,"ebx"),"xmm3"); - &vmovdqa (&QWP(16*12-128,"ebx"),"xmm4"); - &vmovdqa (&QWP(16*13-128,"ebx"),"xmm5"); - &vmovdqa (&QWP(16*14-128,"ebx"),"xmm6"); - &vmovdqa (&QWP(16*15-128,"ebx"),"xmm7"); - &vmovdqa (&QWP(16*12-128,"ebp"),"xmm4"); # save counter value - - &vmovdqa ($xa, &QWP(16*0-128,"ebp")); - &vmovdqa ($xd, "xmm4"); - &vmovdqa ($xb_,&QWP(16*4-128,"ebp")); - &vmovdqa ($xc, &QWP(16*8-128,"ebp")); - &vmovdqa ($xc_,&QWP(16*9-128,"ebp")); - - &mov ("edx",10); # loop counter - &nop (); - -&set_label("loop",32); - &vpaddd ($xa,$xa,$xb_); # elsewhere - &vpxor ($xd,$xd,$xa); # elsewhere - &QUARTERROUND_XOP(0, 4, 8, 12, 0); - &QUARTERROUND_XOP(1, 5, 9, 13, 1); - &QUARTERROUND_XOP(2, 6,10, 14, 2); - &QUARTERROUND_XOP(3, 7,11, 15, 3); - &QUARTERROUND_XOP(0, 5,10, 15, 4); - &QUARTERROUND_XOP(1, 6,11, 12, 5); - &QUARTERROUND_XOP(2, 7, 8, 13, 6); - &QUARTERROUND_XOP(3, 4, 9, 14, 7); - &dec ("edx"); - &jnz (&label("loop")); - - &vmovdqa (&QWP(16*4-128,"ebx"),$xb_); - &vmovdqa (&QWP(16*8-128,"ebx"),$xc); - &vmovdqa (&QWP(16*9-128,"ebx"),$xc_); - &vmovdqa (&QWP(16*12-128,"ebx"),$xd); - &vmovdqa (&QWP(16*14-128,"ebx"),$xd_); - - my ($xa0,$xa1,$xa2,$xa3,$xt0,$xt1,$xt2,$xt3)=map("xmm$_",(0..7)); - - #&vmovdqa ($xa0,&QWP(16*0-128,"ebx")); # it's there - &vmovdqa ($xa1,&QWP(16*1-128,"ebx")); - &vmovdqa ($xa2,&QWP(16*2-128,"ebx")); - &vmovdqa ($xa3,&QWP(16*3-128,"ebx")); - - for($i=0;$i<256;$i+=64) { - &vpaddd ($xa0,$xa0,&QWP($i+16*0-128,"ebp")); # accumulate key material - &vpaddd ($xa1,$xa1,&QWP($i+16*1-128,"ebp")); - &vpaddd ($xa2,$xa2,&QWP($i+16*2-128,"ebp")); - &vpaddd ($xa3,$xa3,&QWP($i+16*3-128,"ebp")); - - &vpunpckldq ($xt2,$xa0,$xa1); # "de-interlace" data - &vpunpckldq ($xt3,$xa2,$xa3); - &vpunpckhdq ($xa0,$xa0,$xa1); - &vpunpckhdq ($xa2,$xa2,$xa3); - &vpunpcklqdq ($xa1,$xt2,$xt3); # "a0" - &vpunpckhqdq ($xt2,$xt2,$xt3); # "a1" - &vpunpcklqdq ($xt3,$xa0,$xa2); # "a2" - &vpunpckhqdq ($xa3,$xa0,$xa2); # "a3" - - &vpxor ($xt0,$xa1,&QWP(64*0-128,$inp)); - &vpxor ($xt1,$xt2,&QWP(64*1-128,$inp)); - &vpxor ($xt2,$xt3,&QWP(64*2-128,$inp)); - &vpxor ($xt3,$xa3,&QWP(64*3-128,$inp)); - &lea ($inp,&QWP($i<192?16:(64*4-16*3),$inp)); - &vmovdqa ($xa0,&QWP($i+16*4-128,"ebx")) if ($i<192); - &vmovdqa ($xa1,&QWP($i+16*5-128,"ebx")) if ($i<192); - &vmovdqa ($xa2,&QWP($i+16*6-128,"ebx")) if ($i<192); - &vmovdqa ($xa3,&QWP($i+16*7-128,"ebx")) if ($i<192); - &vmovdqu (&QWP(64*0-128,$out),$xt0); # store output - &vmovdqu (&QWP(64*1-128,$out),$xt1); - &vmovdqu (&QWP(64*2-128,$out),$xt2); - &vmovdqu (&QWP(64*3-128,$out),$xt3); - &lea ($out,&QWP($i<192?16:(64*4-16*3),$out)); - } - &sub ($len,64*4); - &jnc (&label("outer_loop")); - - &add ($len,64*4); - &jz (&label("done")); - - &mov ("ebx",&DWP(512+8,"esp")); # restore pointers - &lea ($inp,&DWP(-128,$inp)); - &mov ("edx",&DWP(512+4,"esp")); - &lea ($out,&DWP(-128,$out)); - - &vmovd ("xmm2",&DWP(16*12-128,"ebp")); # counter value - &vmovdqu ("xmm3",&QWP(0,"ebx")); - &vpaddd ("xmm2","xmm2",&QWP(16*6,"eax"));# +four - &vpand ("xmm3","xmm3",&QWP(16*7,"eax")); - &vpor ("xmm3","xmm3","xmm2"); # counter value -{ -my ($a,$b,$c,$d,$t,$t1,$rot16,$rot24)=map("xmm$_",(0..7)); - -sub XOPROUND { - &vpaddd ($a,$a,$b); - &vpxor ($d,$d,$a); - &vprotd ($d,$d,16); - - &vpaddd ($c,$c,$d); - &vpxor ($b,$b,$c); - &vprotd ($b,$b,12); - - &vpaddd ($a,$a,$b); - &vpxor ($d,$d,$a); - &vprotd ($d,$d,8); - - &vpaddd ($c,$c,$d); - &vpxor ($b,$b,$c); - &vprotd ($b,$b,7); -} - -&set_label("1x"); - &vmovdqa ($a,&QWP(16*2,"eax")); # sigma - &vmovdqu ($b,&QWP(0,"edx")); - &vmovdqu ($c,&QWP(16,"edx")); - #&vmovdqu ($d,&QWP(0,"ebx")); # already loaded - &vmovdqa ($rot16,&QWP(0,"eax")); - &vmovdqa ($rot24,&QWP(16,"eax")); - &mov (&DWP(16*3,"esp"),"ebp"); - - &vmovdqa (&QWP(16*0,"esp"),$a); - &vmovdqa (&QWP(16*1,"esp"),$b); - &vmovdqa (&QWP(16*2,"esp"),$c); - &vmovdqa (&QWP(16*3,"esp"),$d); - &mov ("edx",10); - &jmp (&label("loop1x")); - -&set_label("outer1x",16); - &vmovdqa ($d,&QWP(16*5,"eax")); # one - &vmovdqa ($a,&QWP(16*0,"esp")); - &vmovdqa ($b,&QWP(16*1,"esp")); - &vmovdqa ($c,&QWP(16*2,"esp")); - &vpaddd ($d,$d,&QWP(16*3,"esp")); - &mov ("edx",10); - &vmovdqa (&QWP(16*3,"esp"),$d); - &jmp (&label("loop1x")); - -&set_label("loop1x",16); - &XOPROUND(); - &vpshufd ($c,$c,0b01001110); - &vpshufd ($b,$b,0b00111001); - &vpshufd ($d,$d,0b10010011); - - &XOPROUND(); - &vpshufd ($c,$c,0b01001110); - &vpshufd ($b,$b,0b10010011); - &vpshufd ($d,$d,0b00111001); - - &dec ("edx"); - &jnz (&label("loop1x")); - - &vpaddd ($a,$a,&QWP(16*0,"esp")); - &vpaddd ($b,$b,&QWP(16*1,"esp")); - &vpaddd ($c,$c,&QWP(16*2,"esp")); - &vpaddd ($d,$d,&QWP(16*3,"esp")); - - &cmp ($len,64); - &jb (&label("tail")); - - &vpxor ($a,$a,&QWP(16*0,$inp)); # xor with input - &vpxor ($b,$b,&QWP(16*1,$inp)); - &vpxor ($c,$c,&QWP(16*2,$inp)); - &vpxor ($d,$d,&QWP(16*3,$inp)); - &lea ($inp,&DWP(16*4,$inp)); # inp+=64 - - &vmovdqu (&QWP(16*0,$out),$a); # write output - &vmovdqu (&QWP(16*1,$out),$b); - &vmovdqu (&QWP(16*2,$out),$c); - &vmovdqu (&QWP(16*3,$out),$d); - &lea ($out,&DWP(16*4,$out)); # inp+=64 - - &sub ($len,64); - &jnz (&label("outer1x")); - - &jmp (&label("done")); - -&set_label("tail"); - &vmovdqa (&QWP(16*0,"esp"),$a); - &vmovdqa (&QWP(16*1,"esp"),$b); - &vmovdqa (&QWP(16*2,"esp"),$c); - &vmovdqa (&QWP(16*3,"esp"),$d); - - &xor ("eax","eax"); - &xor ("edx","edx"); - &xor ("ebp","ebp"); - -&set_label("tail_loop"); - &movb ("al",&BP(0,"esp","ebp")); - &movb ("dl",&BP(0,$inp,"ebp")); - &lea ("ebp",&DWP(1,"ebp")); - &xor ("al","dl"); - &movb (&BP(-1,$out,"ebp"),"al"); - &dec ($len); - &jnz (&label("tail_loop")); -} -&set_label("done"); - &vzeroupper (); - &mov ("esp",&DWP(512,"esp")); -&function_end("ChaCha20_xop"); -} - &asm_finish(); diff --git a/crypto/chacha/asm/chacha-x86_64.pl b/crypto/chacha/asm/chacha-x86_64.pl index 107fc708..55b726d2 100755 --- a/crypto/chacha/asm/chacha-x86_64.pl +++ b/crypto/chacha/asm/chacha-x86_64.pl @@ -36,6 +36,8 @@ # limitations, SSE2 can do better, but gain is considered too # low to justify the [maintenance] effort; # (iv) Bulldozer actually executes 4xXOP code path that delivers 2.20; +# +# Modified from upstream OpenSSL to remove the XOP code. $flavour = shift; $output = shift; @@ -48,24 +50,7 @@ $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or die "can't locate x86_64-xlate.pl"; -if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1` - =~ /GNU assembler version ([2-9]\.[0-9]+)/) { - $avx = ($1>=2.19) + ($1>=2.22); -} - -if (!$avx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) && - `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) { - $avx = ($1>=2.09) + ($1>=2.10); -} - -if (!$avx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) && - `ml64 2>&1` =~ /Version ([0-9]+)\./) { - $avx = ($1>=10) + ($1>=11); -} - -if (!$avx && `$ENV{CC} -v 2>&1` =~ /((?:^clang|LLVM) version|.*based on LLVM) ([3-9]\.[0-9]+)/) { - $avx = ($2>=3.0) + ($2>3.0); -} +$avx = 2; open OUT,"| \"$^X\" $xlate $flavour $output"; *STDOUT=*OUT; @@ -419,10 +404,6 @@ $code.=<<___; ChaCha20_ssse3: .LChaCha20_ssse3: ___ -$code.=<<___ if ($avx); - test \$`1<<(43-32)`,%r10d - jnz .LChaCha20_4xop # XOP is fastest even if we use 1/4 -___ $code.=<<___; cmp \$128,$len # we might throw away some data, ja .LChaCha20_4x # but overall it won't be slower @@ -1133,457 +1114,6 @@ $code.=<<___; ___ } -######################################################################## -# XOP code path that handles all lengths. -if ($avx) { -# There is some "anomaly" observed depending on instructions' size or -# alignment. If you look closely at below code you'll notice that -# sometimes argument order varies. The order affects instruction -# encoding by making it larger, and such fiddling gives 5% performance -# improvement. This is on FX-4100... - -my ($xb0,$xb1,$xb2,$xb3, $xd0,$xd1,$xd2,$xd3, - $xa0,$xa1,$xa2,$xa3, $xt0,$xt1,$xt2,$xt3)=map("%xmm$_",(0..15)); -my @xx=($xa0,$xa1,$xa2,$xa3, $xb0,$xb1,$xb2,$xb3, - $xt0,$xt1,$xt2,$xt3, $xd0,$xd1,$xd2,$xd3); - -sub XOP_lane_ROUND { -my ($a0,$b0,$c0,$d0)=@_; -my ($a1,$b1,$c1,$d1)=map(($_&~3)+(($_+1)&3),($a0,$b0,$c0,$d0)); -my ($a2,$b2,$c2,$d2)=map(($_&~3)+(($_+1)&3),($a1,$b1,$c1,$d1)); -my ($a3,$b3,$c3,$d3)=map(($_&~3)+(($_+1)&3),($a2,$b2,$c2,$d2)); -my @x=map("\"$_\"",@xx); - - ( - "&vpaddd (@x[$a0],@x[$a0],@x[$b0])", # Q1 - "&vpaddd (@x[$a1],@x[$a1],@x[$b1])", # Q2 - "&vpaddd (@x[$a2],@x[$a2],@x[$b2])", # Q3 - "&vpaddd (@x[$a3],@x[$a3],@x[$b3])", # Q4 - "&vpxor (@x[$d0],@x[$a0],@x[$d0])", - "&vpxor (@x[$d1],@x[$a1],@x[$d1])", - "&vpxor (@x[$d2],@x[$a2],@x[$d2])", - "&vpxor (@x[$d3],@x[$a3],@x[$d3])", - "&vprotd (@x[$d0],@x[$d0],16)", - "&vprotd (@x[$d1],@x[$d1],16)", - "&vprotd (@x[$d2],@x[$d2],16)", - "&vprotd (@x[$d3],@x[$d3],16)", - - "&vpaddd (@x[$c0],@x[$c0],@x[$d0])", - "&vpaddd (@x[$c1],@x[$c1],@x[$d1])", - "&vpaddd (@x[$c2],@x[$c2],@x[$d2])", - "&vpaddd (@x[$c3],@x[$c3],@x[$d3])", - "&vpxor (@x[$b0],@x[$c0],@x[$b0])", - "&vpxor (@x[$b1],@x[$c1],@x[$b1])", - "&vpxor (@x[$b2],@x[$b2],@x[$c2])", # flip - "&vpxor (@x[$b3],@x[$b3],@x[$c3])", # flip - "&vprotd (@x[$b0],@x[$b0],12)", - "&vprotd (@x[$b1],@x[$b1],12)", - "&vprotd (@x[$b2],@x[$b2],12)", - "&vprotd (@x[$b3],@x[$b3],12)", - - "&vpaddd (@x[$a0],@x[$b0],@x[$a0])", # flip - "&vpaddd (@x[$a1],@x[$b1],@x[$a1])", # flip - "&vpaddd (@x[$a2],@x[$a2],@x[$b2])", - "&vpaddd (@x[$a3],@x[$a3],@x[$b3])", - "&vpxor (@x[$d0],@x[$a0],@x[$d0])", - "&vpxor (@x[$d1],@x[$a1],@x[$d1])", - "&vpxor (@x[$d2],@x[$a2],@x[$d2])", - "&vpxor (@x[$d3],@x[$a3],@x[$d3])", - "&vprotd (@x[$d0],@x[$d0],8)", - "&vprotd (@x[$d1],@x[$d1],8)", - "&vprotd (@x[$d2],@x[$d2],8)", - "&vprotd (@x[$d3],@x[$d3],8)", - - "&vpaddd (@x[$c0],@x[$c0],@x[$d0])", - "&vpaddd (@x[$c1],@x[$c1],@x[$d1])", - "&vpaddd (@x[$c2],@x[$c2],@x[$d2])", - "&vpaddd (@x[$c3],@x[$c3],@x[$d3])", - "&vpxor (@x[$b0],@x[$c0],@x[$b0])", - "&vpxor (@x[$b1],@x[$c1],@x[$b1])", - "&vpxor (@x[$b2],@x[$b2],@x[$c2])", # flip - "&vpxor (@x[$b3],@x[$b3],@x[$c3])", # flip - "&vprotd (@x[$b0],@x[$b0],7)", - "&vprotd (@x[$b1],@x[$b1],7)", - "&vprotd (@x[$b2],@x[$b2],7)", - "&vprotd (@x[$b3],@x[$b3],7)" - ); -} - -my $xframe = $win64 ? 0xa0 : 0; - -$code.=<<___; -.type ChaCha20_4xop,\@function,5 -.align 32 -ChaCha20_4xop: -.LChaCha20_4xop: - lea -0x78(%rsp),%r11 - sub \$0x148+$xframe,%rsp -___ - ################ stack layout - # +0x00 SIMD equivalent of @x[8-12] - # ... - # +0x40 constant copy of key[0-2] smashed by lanes - # ... - # +0x100 SIMD counters (with nonce smashed by lanes) - # ... - # +0x140 -$code.=<<___ if ($win64); - movaps %xmm6,-0x30(%r11) - movaps %xmm7,-0x20(%r11) - movaps %xmm8,-0x10(%r11) - movaps %xmm9,0x00(%r11) - movaps %xmm10,0x10(%r11) - movaps %xmm11,0x20(%r11) - movaps %xmm12,0x30(%r11) - movaps %xmm13,0x40(%r11) - movaps %xmm14,0x50(%r11) - movaps %xmm15,0x60(%r11) -___ -$code.=<<___; - vzeroupper - - vmovdqa .Lsigma(%rip),$xa3 # key[0] - vmovdqu ($key),$xb3 # key[1] - vmovdqu 16($key),$xt3 # key[2] - vmovdqu ($counter),$xd3 # key[3] - lea 0x100(%rsp),%rcx # size optimization - - vpshufd \$0x00,$xa3,$xa0 # smash key by lanes... - vpshufd \$0x55,$xa3,$xa1 - vmovdqa $xa0,0x40(%rsp) # ... and offload - vpshufd \$0xaa,$xa3,$xa2 - vmovdqa $xa1,0x50(%rsp) - vpshufd \$0xff,$xa3,$xa3 - vmovdqa $xa2,0x60(%rsp) - vmovdqa $xa3,0x70(%rsp) - - vpshufd \$0x00,$xb3,$xb0 - vpshufd \$0x55,$xb3,$xb1 - vmovdqa $xb0,0x80-0x100(%rcx) - vpshufd \$0xaa,$xb3,$xb2 - vmovdqa $xb1,0x90-0x100(%rcx) - vpshufd \$0xff,$xb3,$xb3 - vmovdqa $xb2,0xa0-0x100(%rcx) - vmovdqa $xb3,0xb0-0x100(%rcx) - - vpshufd \$0x00,$xt3,$xt0 # "$xc0" - vpshufd \$0x55,$xt3,$xt1 # "$xc1" - vmovdqa $xt0,0xc0-0x100(%rcx) - vpshufd \$0xaa,$xt3,$xt2 # "$xc2" - vmovdqa $xt1,0xd0-0x100(%rcx) - vpshufd \$0xff,$xt3,$xt3 # "$xc3" - vmovdqa $xt2,0xe0-0x100(%rcx) - vmovdqa $xt3,0xf0-0x100(%rcx) - - vpshufd \$0x00,$xd3,$xd0 - vpshufd \$0x55,$xd3,$xd1 - vpaddd .Linc(%rip),$xd0,$xd0 # don't save counters yet - vpshufd \$0xaa,$xd3,$xd2 - vmovdqa $xd1,0x110-0x100(%rcx) - vpshufd \$0xff,$xd3,$xd3 - vmovdqa $xd2,0x120-0x100(%rcx) - vmovdqa $xd3,0x130-0x100(%rcx) - - jmp .Loop_enter4xop - -.align 32 -.Loop_outer4xop: - vmovdqa 0x40(%rsp),$xa0 # re-load smashed key - vmovdqa 0x50(%rsp),$xa1 - vmovdqa 0x60(%rsp),$xa2 - vmovdqa 0x70(%rsp),$xa3 - vmovdqa 0x80-0x100(%rcx),$xb0 - vmovdqa 0x90-0x100(%rcx),$xb1 - vmovdqa 0xa0-0x100(%rcx),$xb2 - vmovdqa 0xb0-0x100(%rcx),$xb3 - vmovdqa 0xc0-0x100(%rcx),$xt0 # "$xc0" - vmovdqa 0xd0-0x100(%rcx),$xt1 # "$xc1" - vmovdqa 0xe0-0x100(%rcx),$xt2 # "$xc2" - vmovdqa 0xf0-0x100(%rcx),$xt3 # "$xc3" - vmovdqa 0x100-0x100(%rcx),$xd0 - vmovdqa 0x110-0x100(%rcx),$xd1 - vmovdqa 0x120-0x100(%rcx),$xd2 - vmovdqa 0x130-0x100(%rcx),$xd3 - vpaddd .Lfour(%rip),$xd0,$xd0 # next SIMD counters - -.Loop_enter4xop: - mov \$10,%eax - vmovdqa $xd0,0x100-0x100(%rcx) # save SIMD counters - jmp .Loop4xop - -.align 32 -.Loop4xop: -___ - foreach (&XOP_lane_ROUND(0, 4, 8,12)) { eval; } - foreach (&XOP_lane_ROUND(0, 5,10,15)) { eval; } -$code.=<<___; - dec %eax - jnz .Loop4xop - - vpaddd 0x40(%rsp),$xa0,$xa0 # accumulate key material - vpaddd 0x50(%rsp),$xa1,$xa1 - vpaddd 0x60(%rsp),$xa2,$xa2 - vpaddd 0x70(%rsp),$xa3,$xa3 - - vmovdqa $xt2,0x20(%rsp) # offload $xc2,3 - vmovdqa $xt3,0x30(%rsp) - - vpunpckldq $xa1,$xa0,$xt2 # "de-interlace" data - vpunpckldq $xa3,$xa2,$xt3 - vpunpckhdq $xa1,$xa0,$xa0 - vpunpckhdq $xa3,$xa2,$xa2 - vpunpcklqdq $xt3,$xt2,$xa1 # "a0" - vpunpckhqdq $xt3,$xt2,$xt2 # "a1" - vpunpcklqdq $xa2,$xa0,$xa3 # "a2" - vpunpckhqdq $xa2,$xa0,$xa0 # "a3" -___ - ($xa0,$xa1,$xa2,$xa3,$xt2)=($xa1,$xt2,$xa3,$xa0,$xa2); -$code.=<<___; - vpaddd 0x80-0x100(%rcx),$xb0,$xb0 - vpaddd 0x90-0x100(%rcx),$xb1,$xb1 - vpaddd 0xa0-0x100(%rcx),$xb2,$xb2 - vpaddd 0xb0-0x100(%rcx),$xb3,$xb3 - - vmovdqa $xa0,0x00(%rsp) # offload $xa0,1 - vmovdqa $xa1,0x10(%rsp) - vmovdqa 0x20(%rsp),$xa0 # "xc2" - vmovdqa 0x30(%rsp),$xa1 # "xc3" - - vpunpckldq $xb1,$xb0,$xt2 - vpunpckldq $xb3,$xb2,$xt3 - vpunpckhdq $xb1,$xb0,$xb0 - vpunpckhdq $xb3,$xb2,$xb2 - vpunpcklqdq $xt3,$xt2,$xb1 # "b0" - vpunpckhqdq $xt3,$xt2,$xt2 # "b1" - vpunpcklqdq $xb2,$xb0,$xb3 # "b2" - vpunpckhqdq $xb2,$xb0,$xb0 # "b3" -___ - ($xb0,$xb1,$xb2,$xb3,$xt2)=($xb1,$xt2,$xb3,$xb0,$xb2); - my ($xc0,$xc1,$xc2,$xc3)=($xt0,$xt1,$xa0,$xa1); -$code.=<<___; - vpaddd 0xc0-0x100(%rcx),$xc0,$xc0 - vpaddd 0xd0-0x100(%rcx),$xc1,$xc1 - vpaddd 0xe0-0x100(%rcx),$xc2,$xc2 - vpaddd 0xf0-0x100(%rcx),$xc3,$xc3 - - vpunpckldq $xc1,$xc0,$xt2 - vpunpckldq $xc3,$xc2,$xt3 - vpunpckhdq $xc1,$xc0,$xc0 - vpunpckhdq $xc3,$xc2,$xc2 - vpunpcklqdq $xt3,$xt2,$xc1 # "c0" - vpunpckhqdq $xt3,$xt2,$xt2 # "c1" - vpunpcklqdq $xc2,$xc0,$xc3 # "c2" - vpunpckhqdq $xc2,$xc0,$xc0 # "c3" -___ - ($xc0,$xc1,$xc2,$xc3,$xt2)=($xc1,$xt2,$xc3,$xc0,$xc2); -$code.=<<___; - vpaddd 0x100-0x100(%rcx),$xd0,$xd0 - vpaddd 0x110-0x100(%rcx),$xd1,$xd1 - vpaddd 0x120-0x100(%rcx),$xd2,$xd2 - vpaddd 0x130-0x100(%rcx),$xd3,$xd3 - - vpunpckldq $xd1,$xd0,$xt2 - vpunpckldq $xd3,$xd2,$xt3 - vpunpckhdq $xd1,$xd0,$xd0 - vpunpckhdq $xd3,$xd2,$xd2 - vpunpcklqdq $xt3,$xt2,$xd1 # "d0" - vpunpckhqdq $xt3,$xt2,$xt2 # "d1" - vpunpcklqdq $xd2,$xd0,$xd3 # "d2" - vpunpckhqdq $xd2,$xd0,$xd0 # "d3" -___ - ($xd0,$xd1,$xd2,$xd3,$xt2)=($xd1,$xt2,$xd3,$xd0,$xd2); - ($xa0,$xa1)=($xt2,$xt3); -$code.=<<___; - vmovdqa 0x00(%rsp),$xa0 # restore $xa0,1 - vmovdqa 0x10(%rsp),$xa1 - - cmp \$64*4,$len - jb .Ltail4xop - - vpxor 0x00($inp),$xa0,$xa0 # xor with input - vpxor 0x10($inp),$xb0,$xb0 - vpxor 0x20($inp),$xc0,$xc0 - vpxor 0x30($inp),$xd0,$xd0 - vpxor 0x40($inp),$xa1,$xa1 - vpxor 0x50($inp),$xb1,$xb1 - vpxor 0x60($inp),$xc1,$xc1 - vpxor 0x70($inp),$xd1,$xd1 - lea 0x80($inp),$inp # size optimization - vpxor 0x00($inp),$xa2,$xa2 - vpxor 0x10($inp),$xb2,$xb2 - vpxor 0x20($inp),$xc2,$xc2 - vpxor 0x30($inp),$xd2,$xd2 - vpxor 0x40($inp),$xa3,$xa3 - vpxor 0x50($inp),$xb3,$xb3 - vpxor 0x60($inp),$xc3,$xc3 - vpxor 0x70($inp),$xd3,$xd3 - lea 0x80($inp),$inp # inp+=64*4 - - vmovdqu $xa0,0x00($out) - vmovdqu $xb0,0x10($out) - vmovdqu $xc0,0x20($out) - vmovdqu $xd0,0x30($out) - vmovdqu $xa1,0x40($out) - vmovdqu $xb1,0x50($out) - vmovdqu $xc1,0x60($out) - vmovdqu $xd1,0x70($out) - lea 0x80($out),$out # size optimization - vmovdqu $xa2,0x00($out) - vmovdqu $xb2,0x10($out) - vmovdqu $xc2,0x20($out) - vmovdqu $xd2,0x30($out) - vmovdqu $xa3,0x40($out) - vmovdqu $xb3,0x50($out) - vmovdqu $xc3,0x60($out) - vmovdqu $xd3,0x70($out) - lea 0x80($out),$out # out+=64*4 - - sub \$64*4,$len - jnz .Loop_outer4xop - - jmp .Ldone4xop - -.align 32 -.Ltail4xop: - cmp \$192,$len - jae .L192_or_more4xop - cmp \$128,$len - jae .L128_or_more4xop - cmp \$64,$len - jae .L64_or_more4xop - - xor %r10,%r10 - vmovdqa $xa0,0x00(%rsp) - vmovdqa $xb0,0x10(%rsp) - vmovdqa $xc0,0x20(%rsp) - vmovdqa $xd0,0x30(%rsp) - jmp .Loop_tail4xop - -.align 32 -.L64_or_more4xop: - vpxor 0x00($inp),$xa0,$xa0 # xor with input - vpxor 0x10($inp),$xb0,$xb0 - vpxor 0x20($inp),$xc0,$xc0 - vpxor 0x30($inp),$xd0,$xd0 - vmovdqu $xa0,0x00($out) - vmovdqu $xb0,0x10($out) - vmovdqu $xc0,0x20($out) - vmovdqu $xd0,0x30($out) - je .Ldone4xop - - lea 0x40($inp),$inp # inp+=64*1 - vmovdqa $xa1,0x00(%rsp) - xor %r10,%r10 - vmovdqa $xb1,0x10(%rsp) - lea 0x40($out),$out # out+=64*1 - vmovdqa $xc1,0x20(%rsp) - sub \$64,$len # len-=64*1 - vmovdqa $xd1,0x30(%rsp) - jmp .Loop_tail4xop - -.align 32 -.L128_or_more4xop: - vpxor 0x00($inp),$xa0,$xa0 # xor with input - vpxor 0x10($inp),$xb0,$xb0 - vpxor 0x20($inp),$xc0,$xc0 - vpxor 0x30($inp),$xd0,$xd0 - vpxor 0x40($inp),$xa1,$xa1 - vpxor 0x50($inp),$xb1,$xb1 - vpxor 0x60($inp),$xc1,$xc1 - vpxor 0x70($inp),$xd1,$xd1 - - vmovdqu $xa0,0x00($out) - vmovdqu $xb0,0x10($out) - vmovdqu $xc0,0x20($out) - vmovdqu $xd0,0x30($out) - vmovdqu $xa1,0x40($out) - vmovdqu $xb1,0x50($out) - vmovdqu $xc1,0x60($out) - vmovdqu $xd1,0x70($out) - je .Ldone4xop - - lea 0x80($inp),$inp # inp+=64*2 - vmovdqa $xa2,0x00(%rsp) - xor %r10,%r10 - vmovdqa $xb2,0x10(%rsp) - lea 0x80($out),$out # out+=64*2 - vmovdqa $xc2,0x20(%rsp) - sub \$128,$len # len-=64*2 - vmovdqa $xd2,0x30(%rsp) - jmp .Loop_tail4xop - -.align 32 -.L192_or_more4xop: - vpxor 0x00($inp),$xa0,$xa0 # xor with input - vpxor 0x10($inp),$xb0,$xb0 - vpxor 0x20($inp),$xc0,$xc0 - vpxor 0x30($inp),$xd0,$xd0 - vpxor 0x40($inp),$xa1,$xa1 - vpxor 0x50($inp),$xb1,$xb1 - vpxor 0x60($inp),$xc1,$xc1 - vpxor 0x70($inp),$xd1,$xd1 - lea 0x80($inp),$inp # size optimization - vpxor 0x00($inp),$xa2,$xa2 - vpxor 0x10($inp),$xb2,$xb2 - vpxor 0x20($inp),$xc2,$xc2 - vpxor 0x30($inp),$xd2,$xd2 - - vmovdqu $xa0,0x00($out) - vmovdqu $xb0,0x10($out) - vmovdqu $xc0,0x20($out) - vmovdqu $xd0,0x30($out) - vmovdqu $xa1,0x40($out) - vmovdqu $xb1,0x50($out) - vmovdqu $xc1,0x60($out) - vmovdqu $xd1,0x70($out) - lea 0x80($out),$out # size optimization - vmovdqu $xa2,0x00($out) - vmovdqu $xb2,0x10($out) - vmovdqu $xc2,0x20($out) - vmovdqu $xd2,0x30($out) - je .Ldone4xop - - lea 0x40($inp),$inp # inp+=64*3 - vmovdqa $xa2,0x00(%rsp) - xor %r10,%r10 - vmovdqa $xb2,0x10(%rsp) - lea 0x40($out),$out # out+=64*3 - vmovdqa $xc2,0x20(%rsp) - sub \$192,$len # len-=64*3 - vmovdqa $xd2,0x30(%rsp) - -.Loop_tail4xop: - movzb ($inp,%r10),%eax - movzb (%rsp,%r10),%ecx - lea 1(%r10),%r10 - xor %ecx,%eax - mov %al,-1($out,%r10) - dec $len - jnz .Loop_tail4xop - -.Ldone4xop: - vzeroupper -___ -$code.=<<___ if ($win64); - lea 0x140+0x30(%rsp),%r11 - movaps -0x30(%r11),%xmm6 - movaps -0x20(%r11),%xmm7 - movaps -0x10(%r11),%xmm8 - movaps 0x00(%r11),%xmm9 - movaps 0x10(%r11),%xmm10 - movaps 0x20(%r11),%xmm11 - movaps 0x30(%r11),%xmm12 - movaps 0x40(%r11),%xmm13 - movaps 0x50(%r11),%xmm14 - movaps 0x60(%r11),%xmm15 -___ -$code.=<<___; - add \$0x148+$xframe,%rsp - ret -.size ChaCha20_4xop,.-ChaCha20_4xop -___ -} - ######################################################################## # AVX2 code path if ($avx>1) { diff --git a/crypto/chacha/chacha_generic.c b/crypto/chacha/chacha.c similarity index 73% rename from crypto/chacha/chacha_generic.c rename to crypto/chacha/chacha.c index f262033c..afe1b2ad 100644 --- a/crypto/chacha/chacha_generic.c +++ b/crypto/chacha/chacha.c @@ -21,7 +21,49 @@ #include -#if defined(OPENSSL_WINDOWS) || (!defined(OPENSSL_X86_64) && !defined(OPENSSL_X86)) || !defined(__SSE2__) +#define U8TO32_LITTLE(p) \ + (((uint32_t)((p)[0])) | ((uint32_t)((p)[1]) << 8) | \ + ((uint32_t)((p)[2]) << 16) | ((uint32_t)((p)[3]) << 24)) + +#if !defined(OPENSSL_NO_ASM) && \ + (defined(OPENSSL_X86) || defined(OPENSSL_X86_64) || \ + defined(OPENSSL_ARM) || defined(OPENSSL_AARCH64)) + +/* ChaCha20_ctr32 is defined in asm/chacha-*.pl. */ +void ChaCha20_ctr32(uint8_t *out, const uint8_t *in, size_t in_len, + const uint32_t key[8], const uint32_t counter[4]); + +void CRYPTO_chacha_20(uint8_t *out, const uint8_t *in, size_t in_len, + const uint8_t key[32], const uint8_t nonce[12], + uint32_t counter) { + uint32_t counter_nonce[4]; + counter_nonce[0] = counter; + counter_nonce[1] = U8TO32_LITTLE(nonce + 0); + counter_nonce[2] = U8TO32_LITTLE(nonce + 4); + counter_nonce[3] = U8TO32_LITTLE(nonce + 8); + + const uint32_t *key_ptr = (const uint32_t *)key; +#if !defined(OPENSSL_X86) && !defined(OPENSSL_X86_64) + /* The assembly expects the key to be four-byte aligned. */ + uint32_t key_u32[8]; + if ((((uintptr_t)key) & 3) != 0) { + key_u32[0] = U8TO32_LITTLE(key + 0); + key_u32[1] = U8TO32_LITTLE(key + 4); + key_u32[2] = U8TO32_LITTLE(key + 8); + key_u32[3] = U8TO32_LITTLE(key + 12); + key_u32[4] = U8TO32_LITTLE(key + 16); + key_u32[5] = U8TO32_LITTLE(key + 20); + key_u32[6] = U8TO32_LITTLE(key + 24); + key_u32[7] = U8TO32_LITTLE(key + 28); + + key_ptr = key_u32; + } +#endif + + ChaCha20_ctr32(out, in, in_len, key_ptr, counter_nonce); +} + +#else /* sigma contains the ChaCha constants, which happen to be an ASCII string. */ static const uint8_t sigma[16] = { 'e', 'x', 'p', 'a', 'n', 'd', ' ', '3', @@ -40,10 +82,6 @@ static const uint8_t sigma[16] = { 'e', 'x', 'p', 'a', 'n', 'd', ' ', '3', (p)[3] = (v >> 24) & 0xff; \ } -#define U8TO32_LITTLE(p) \ - (((uint32_t)((p)[0])) | ((uint32_t)((p)[1]) << 8) | \ - ((uint32_t)((p)[2]) << 16) | ((uint32_t)((p)[3]) << 24)) - /* QUARTERROUND updates a, b, c, d with a ChaCha "quarter" round. */ #define QUARTERROUND(a,b,c,d) \ x[a] = PLUS(x[a],x[b]); x[d] = ROTATE(XOR(x[d],x[a]),16); \ @@ -51,13 +89,6 @@ static const uint8_t sigma[16] = { 'e', 'x', 'p', 'a', 'n', 'd', ' ', '3', x[a] = PLUS(x[a],x[b]); x[d] = ROTATE(XOR(x[d],x[a]), 8); \ x[c] = PLUS(x[c],x[d]); x[b] = ROTATE(XOR(x[b],x[c]), 7); -#if defined(OPENSSL_ARM) && !defined(OPENSSL_NO_ASM) -/* Defined in chacha_vec.c */ -void CRYPTO_chacha_20_neon(uint8_t *out, const uint8_t *in, size_t in_len, - const uint8_t key[32], const uint8_t nonce[12], - uint32_t counter); -#endif - /* chacha_core performs 20 rounds of ChaCha on the input words in * |input| and writes the 64 output bytes to |output|. */ static void chacha_core(uint8_t output[64], const uint32_t input[16]) { @@ -91,13 +122,6 @@ void CRYPTO_chacha_20(uint8_t *out, const uint8_t *in, size_t in_len, uint8_t buf[64]; size_t todo, i; -#if defined(OPENSSL_ARM) && !defined(OPENSSL_NO_ASM) - if (CRYPTO_is_NEON_capable()) { - CRYPTO_chacha_20_neon(out, in, in_len, key, nonce, counter); - return; - } -#endif - input[0] = U8TO32_LITTLE(sigma + 0); input[1] = U8TO32_LITTLE(sigma + 4); input[2] = U8TO32_LITTLE(sigma + 8); @@ -137,4 +161,4 @@ void CRYPTO_chacha_20(uint8_t *out, const uint8_t *in, size_t in_len, } } -#endif /* OPENSSL_WINDOWS || !OPENSSL_X86_64 && !OPENSSL_X86 || !__SSE2__ */ +#endif diff --git a/crypto/chacha/chacha_vec.c b/crypto/chacha/chacha_vec.c deleted file mode 100644 index 90d62002..00000000 --- a/crypto/chacha/chacha_vec.c +++ /dev/null @@ -1,328 +0,0 @@ -/* Copyright (c) 2014, Google Inc. - * - * Permission to use, copy, modify, and/or distribute this software for any - * purpose with or without fee is hereby granted, provided that the above - * copyright notice and this permission notice appear in all copies. - * - * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES - * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF - * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY - * SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES - * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION - * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN - * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. */ - -/* ==================================================================== - * - * When updating this file, also update chacha_vec_arm.S - * - * ==================================================================== */ - - -/* This implementation is by Ted Krovetz and was submitted to SUPERCOP and - * marked as public domain. It was been altered to allow for non-aligned inputs - * and to allow the block counter to be passed in specifically. */ - -#include - -#include "../internal.h" - - -#if defined(ASM_GEN) || \ - !defined(OPENSSL_WINDOWS) && \ - (defined(OPENSSL_X86_64) || defined(OPENSSL_X86)) && defined(__SSE2__) - -#define CHACHA_RNDS 20 /* 8 (high speed), 20 (conservative), 12 (middle) */ - -/* Architecture-neutral way to specify 16-byte vector of ints */ -typedef unsigned vec __attribute__((vector_size(16))); - -/* This implementation is designed for Neon, SSE and AltiVec machines. The - * following specify how to do certain vector operations efficiently on - * each architecture, using intrinsics. - * This implementation supports parallel processing of multiple blocks, - * including potentially using general-purpose registers. */ -#if __ARM_NEON__ -#include -#include -#define GPR_TOO 1 -#define VBPI 2 -#define ONE (vec) vsetq_lane_u32(1, vdupq_n_u32(0), 0) -#define LOAD_ALIGNED(m) (vec)(*((vec *)(m))) -#define LOAD(m) ({ \ - memcpy(alignment_buffer, m, 16); \ - LOAD_ALIGNED(alignment_buffer); \ - }) -#define STORE(m, r) ({ \ - (*((vec *)(alignment_buffer))) = (r); \ - memcpy(m, alignment_buffer, 16); \ - }) -#define ROTV1(x) (vec) vextq_u32((uint32x4_t)x, (uint32x4_t)x, 1) -#define ROTV2(x) (vec) vextq_u32((uint32x4_t)x, (uint32x4_t)x, 2) -#define ROTV3(x) (vec) vextq_u32((uint32x4_t)x, (uint32x4_t)x, 3) -#define ROTW16(x) (vec) vrev32q_u16((uint16x8_t)x) -#if __clang__ -#define ROTW7(x) (x << ((vec) {7, 7, 7, 7})) ^ (x >> ((vec) {25, 25, 25, 25})) -#define ROTW8(x) (x << ((vec) {8, 8, 8, 8})) ^ (x >> ((vec) {24, 24, 24, 24})) -#define ROTW12(x) \ - (x << ((vec) {12, 12, 12, 12})) ^ (x >> ((vec) {20, 20, 20, 20})) -#else -#define ROTW7(x) \ - (vec) vsriq_n_u32(vshlq_n_u32((uint32x4_t)x, 7), (uint32x4_t)x, 25) -#define ROTW8(x) \ - (vec) vsriq_n_u32(vshlq_n_u32((uint32x4_t)x, 8), (uint32x4_t)x, 24) -#define ROTW12(x) \ - (vec) vsriq_n_u32(vshlq_n_u32((uint32x4_t)x, 12), (uint32x4_t)x, 20) -#endif -#elif __SSE2__ -#include -#define GPR_TOO 0 -#if __clang__ -#define VBPI 4 -#else -#define VBPI 3 -#endif -#define ONE (vec) _mm_set_epi32(0, 0, 0, 1) -#define LOAD(m) (vec) _mm_loadu_si128((const __m128i *)(m)) -#define LOAD_ALIGNED(m) (vec) _mm_load_si128((const __m128i *)(m)) -#define STORE(m, r) _mm_storeu_si128((__m128i *)(m), (__m128i)(r)) -#define ROTV1(x) (vec) _mm_shuffle_epi32((__m128i)x, _MM_SHUFFLE(0, 3, 2, 1)) -#define ROTV2(x) (vec) _mm_shuffle_epi32((__m128i)x, _MM_SHUFFLE(1, 0, 3, 2)) -#define ROTV3(x) (vec) _mm_shuffle_epi32((__m128i)x, _MM_SHUFFLE(2, 1, 0, 3)) -#define ROTW7(x) \ - (vec)(_mm_slli_epi32((__m128i)x, 7) ^ _mm_srli_epi32((__m128i)x, 25)) -#define ROTW12(x) \ - (vec)(_mm_slli_epi32((__m128i)x, 12) ^ _mm_srli_epi32((__m128i)x, 20)) -#if __SSSE3__ -#include -#define ROTW8(x) \ - (vec) _mm_shuffle_epi8((__m128i)x, _mm_set_epi8(14, 13, 12, 15, 10, 9, 8, \ - 11, 6, 5, 4, 7, 2, 1, 0, 3)) -#define ROTW16(x) \ - (vec) _mm_shuffle_epi8((__m128i)x, _mm_set_epi8(13, 12, 15, 14, 9, 8, 11, \ - 10, 5, 4, 7, 6, 1, 0, 3, 2)) -#else -#define ROTW8(x) \ - (vec)(_mm_slli_epi32((__m128i)x, 8) ^ _mm_srli_epi32((__m128i)x, 24)) -#define ROTW16(x) \ - (vec)(_mm_slli_epi32((__m128i)x, 16) ^ _mm_srli_epi32((__m128i)x, 16)) -#endif -#else -#error-- Implementation supports only machines with neon or SSE2 -#endif - -#ifndef REVV_BE -#define REVV_BE(x) (x) -#endif - -#ifndef REVW_BE -#define REVW_BE(x) (x) -#endif - -#define BPI (VBPI + GPR_TOO) /* Blocks computed per loop iteration */ - -#define DQROUND_VECTORS(a,b,c,d) \ - a += b; d ^= a; d = ROTW16(d); \ - c += d; b ^= c; b = ROTW12(b); \ - a += b; d ^= a; d = ROTW8(d); \ - c += d; b ^= c; b = ROTW7(b); \ - b = ROTV1(b); c = ROTV2(c); d = ROTV3(d); \ - a += b; d ^= a; d = ROTW16(d); \ - c += d; b ^= c; b = ROTW12(b); \ - a += b; d ^= a; d = ROTW8(d); \ - c += d; b ^= c; b = ROTW7(b); \ - b = ROTV3(b); c = ROTV2(c); d = ROTV1(d); - -#define QROUND_WORDS(a,b,c,d) \ - a = a+b; d ^= a; d = d<<16 | d>>16; \ - c = c+d; b ^= c; b = b<<12 | b>>20; \ - a = a+b; d ^= a; d = d<< 8 | d>>24; \ - c = c+d; b ^= c; b = b<< 7 | b>>25; - -#define WRITE_XOR(in, op, d, v0, v1, v2, v3) \ - STORE(op + d + 0, LOAD(in + d + 0) ^ REVV_BE(v0)); \ - STORE(op + d + 4, LOAD(in + d + 4) ^ REVV_BE(v1)); \ - STORE(op + d + 8, LOAD(in + d + 8) ^ REVV_BE(v2)); \ - STORE(op + d +12, LOAD(in + d +12) ^ REVV_BE(v3)); - -#if __ARM_NEON__ -/* For ARM, we can't depend on NEON support, so this function is compiled with - * a different name, along with the generic code, and can be enabled at - * run-time. */ -void CRYPTO_chacha_20_neon( -#else -void CRYPTO_chacha_20( -#endif - uint8_t *out, - const uint8_t *in, - size_t inlen, - const uint8_t key[32], - const uint8_t nonce[12], - uint32_t counter) - { - unsigned iters, i; - unsigned *op = (unsigned *)out; - const unsigned *ip = (const unsigned *)in; - const unsigned *kp = (const unsigned *)key; -#if defined(__ARM_NEON__) - uint32_t np[3]; - alignas(16) uint8_t alignment_buffer[16]; -#endif - vec s0, s1, s2, s3; - alignas(16) unsigned chacha_const[] = - {0x61707865,0x3320646E,0x79622D32,0x6B206574}; -#if defined(__ARM_NEON__) - memcpy(np, nonce, 12); -#endif - s0 = LOAD_ALIGNED(chacha_const); - s1 = LOAD(&((const vec*)kp)[0]); - s2 = LOAD(&((const vec*)kp)[1]); - s3 = (vec){ - counter, - ((const uint32_t*)nonce)[0], - ((const uint32_t*)nonce)[1], - ((const uint32_t*)nonce)[2] - }; - - for (iters = 0; iters < inlen/(BPI*64); iters++) - { -#if GPR_TOO - register unsigned x0, x1, x2, x3, x4, x5, x6, x7, x8, - x9, x10, x11, x12, x13, x14, x15; -#endif -#if VBPI > 2 - vec v8,v9,v10,v11; -#endif -#if VBPI > 3 - vec v12,v13,v14,v15; -#endif - - vec v0,v1,v2,v3,v4,v5,v6,v7; - v4 = v0 = s0; v5 = v1 = s1; v6 = v2 = s2; v3 = s3; - v7 = v3 + ONE; -#if VBPI > 2 - v8 = v4; v9 = v5; v10 = v6; - v11 = v7 + ONE; -#endif -#if VBPI > 3 - v12 = v8; v13 = v9; v14 = v10; - v15 = v11 + ONE; -#endif -#if GPR_TOO - x0 = chacha_const[0]; x1 = chacha_const[1]; - x2 = chacha_const[2]; x3 = chacha_const[3]; - x4 = kp[0]; x5 = kp[1]; x6 = kp[2]; x7 = kp[3]; - x8 = kp[4]; x9 = kp[5]; x10 = kp[6]; x11 = kp[7]; - x12 = counter+BPI*iters+(BPI-1); x13 = np[0]; - x14 = np[1]; x15 = np[2]; -#endif - for (i = CHACHA_RNDS/2; i; i--) - { - DQROUND_VECTORS(v0,v1,v2,v3) - DQROUND_VECTORS(v4,v5,v6,v7) -#if VBPI > 2 - DQROUND_VECTORS(v8,v9,v10,v11) -#endif -#if VBPI > 3 - DQROUND_VECTORS(v12,v13,v14,v15) -#endif -#if GPR_TOO - QROUND_WORDS( x0, x4, x8,x12) - QROUND_WORDS( x1, x5, x9,x13) - QROUND_WORDS( x2, x6,x10,x14) - QROUND_WORDS( x3, x7,x11,x15) - QROUND_WORDS( x0, x5,x10,x15) - QROUND_WORDS( x1, x6,x11,x12) - QROUND_WORDS( x2, x7, x8,x13) - QROUND_WORDS( x3, x4, x9,x14) -#endif - } - - WRITE_XOR(ip, op, 0, v0+s0, v1+s1, v2+s2, v3+s3) - s3 += ONE; - WRITE_XOR(ip, op, 16, v4+s0, v5+s1, v6+s2, v7+s3) - s3 += ONE; -#if VBPI > 2 - WRITE_XOR(ip, op, 32, v8+s0, v9+s1, v10+s2, v11+s3) - s3 += ONE; -#endif -#if VBPI > 3 - WRITE_XOR(ip, op, 48, v12+s0, v13+s1, v14+s2, v15+s3) - s3 += ONE; -#endif - ip += VBPI*16; - op += VBPI*16; -#if GPR_TOO - op[0] = REVW_BE(REVW_BE(ip[0]) ^ (x0 + chacha_const[0])); - op[1] = REVW_BE(REVW_BE(ip[1]) ^ (x1 + chacha_const[1])); - op[2] = REVW_BE(REVW_BE(ip[2]) ^ (x2 + chacha_const[2])); - op[3] = REVW_BE(REVW_BE(ip[3]) ^ (x3 + chacha_const[3])); - op[4] = REVW_BE(REVW_BE(ip[4]) ^ (x4 + kp[0])); - op[5] = REVW_BE(REVW_BE(ip[5]) ^ (x5 + kp[1])); - op[6] = REVW_BE(REVW_BE(ip[6]) ^ (x6 + kp[2])); - op[7] = REVW_BE(REVW_BE(ip[7]) ^ (x7 + kp[3])); - op[8] = REVW_BE(REVW_BE(ip[8]) ^ (x8 + kp[4])); - op[9] = REVW_BE(REVW_BE(ip[9]) ^ (x9 + kp[5])); - op[10] = REVW_BE(REVW_BE(ip[10]) ^ (x10 + kp[6])); - op[11] = REVW_BE(REVW_BE(ip[11]) ^ (x11 + kp[7])); - op[12] = REVW_BE(REVW_BE(ip[12]) ^ (x12 + counter+BPI*iters+(BPI-1))); - op[13] = REVW_BE(REVW_BE(ip[13]) ^ (x13 + np[0])); - op[14] = REVW_BE(REVW_BE(ip[14]) ^ (x14 + np[1])); - op[15] = REVW_BE(REVW_BE(ip[15]) ^ (x15 + np[2])); - s3 += ONE; - ip += 16; - op += 16; -#endif - } - - for (iters = inlen%(BPI*64)/64; iters != 0; iters--) - { - vec v0 = s0, v1 = s1, v2 = s2, v3 = s3; - for (i = CHACHA_RNDS/2; i; i--) - { - DQROUND_VECTORS(v0,v1,v2,v3); - } - WRITE_XOR(ip, op, 0, v0+s0, v1+s1, v2+s2, v3+s3) - s3 += ONE; - ip += 16; - op += 16; - } - - inlen = inlen % 64; - if (inlen) - { - alignas(16) vec buf[4]; - vec v0,v1,v2,v3; - v0 = s0; v1 = s1; v2 = s2; v3 = s3; - for (i = CHACHA_RNDS/2; i; i--) - { - DQROUND_VECTORS(v0,v1,v2,v3); - } - - if (inlen >= 16) - { - STORE(op + 0, LOAD(ip + 0) ^ REVV_BE(v0 + s0)); - if (inlen >= 32) - { - STORE(op + 4, LOAD(ip + 4) ^ REVV_BE(v1 + s1)); - if (inlen >= 48) - { - STORE(op + 8, LOAD(ip + 8) ^ - REVV_BE(v2 + s2)); - buf[3] = REVV_BE(v3 + s3); - } - else - buf[2] = REVV_BE(v2 + s2); - } - else - buf[1] = REVV_BE(v1 + s1); - } - else - buf[0] = REVV_BE(v0 + s0); - - for (i=inlen & ~15; i 1 { - compiler = os.Args[1] - } - - args := []string{ - "-O3", - "-mcpu=cortex-a8", - "-mfpu=neon", - "-fpic", - "-DASM_GEN", - "-I", "../../include", - "-S", "chacha_vec.c", - "-o", "-", - } - - output, err := os.OpenFile("chacha_vec_arm.S", os.O_CREATE|os.O_TRUNC|os.O_WRONLY, 0644) - if err != nil { - panic(err) - } - defer output.Close() - - output.WriteString(preamble) - output.WriteString(compiler) - output.WriteString(" ") - output.WriteString(strings.Join(args, " ")) - output.WriteString("\n\n#if !defined(OPENSSL_NO_ASM)\n") - output.WriteString("#if defined(__arm__)\n\n") - - cmd := exec.Command(compiler, args...) - cmd.Stderr = os.Stderr - asm, err := cmd.StdoutPipe() - if err != nil { - panic(err) - } - if err := cmd.Start(); err != nil { - panic(err) - } - - attr28 := []byte(".eabi_attribute 28,") - globalDirective := []byte(".global\t") - newLine := []byte("\n") - attr28Handled := false - - scanner := bufio.NewScanner(asm) - for scanner.Scan() { - line := scanner.Bytes() - - if bytes.Contains(line, attr28) { - output.WriteString(attr28Block) - attr28Handled = true - continue - } - - output.Write(line) - output.Write(newLine) - - if i := bytes.Index(line, globalDirective); i >= 0 { - output.Write(line[:i]) - output.WriteString(".hidden\t") - output.Write(line[i+len(globalDirective):]) - output.Write(newLine) - } - } - - if err := scanner.Err(); err != nil { - panic(err) - } - - if !attr28Handled { - panic("EABI attribute 28 not seen in processing") - } - - if err := cmd.Wait(); err != nil { - panic(err) - } - - output.WriteString(trailer) -} - -const preamble = `# Copyright (c) 2014, Google Inc. -# -# Permission to use, copy, modify, and/or distribute this software for any -# purpose with or without fee is hereby granted, provided that the above -# copyright notice and this permission notice appear in all copies. -# -# THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES -# WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF -# MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY -# SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES -# WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION -# OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN -# CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. - -# This file contains a pre-compiled version of chacha_vec.c for ARM. This is -# needed to support switching on NEON code at runtime. If the whole of OpenSSL -# were to be compiled with the needed flags to build chacha_vec.c, then it -# wouldn't be possible to run on non-NEON systems. -# -# This file was generated by chacha_vec_arm_generate.go using the following -# compiler command: -# -# ` - -const attr28Block = ` -# EABI attribute 28 sets whether VFP register arguments were used to build this -# file. If object files are inconsistent on this point, the linker will refuse -# to link them. Thus we report whatever the compiler expects since we don't use -# VFP arguments. - -#if defined(__ARM_PCS_VFP) - .eabi_attribute 28, 1 -#else - .eabi_attribute 28, 0 -#endif - -` - -const trailer = ` -#endif /* __arm__ */ -#endif /* !OPENSSL_NO_ASM */ -` diff --git a/util/generate_build_files.py b/util/generate_build_files.py index acc693af..3220122b 100644 --- a/util/generate_build_files.py +++ b/util/generate_build_files.py @@ -39,7 +39,6 @@ OS_ARCH_COMBOS = [ # perlasm system. NON_PERL_FILES = { ('linux', 'arm'): [ - 'src/crypto/chacha/chacha_vec_arm.S', 'src/crypto/cpu-arm-asm.S', 'src/crypto/curve25519/asm/x25519-asm-arm.S', 'src/crypto/poly1305/poly1305_arm_asm.S',