jkivilin/camellia-simd-aesni 开源项目中文解读：技术栈、适用场景与上手建议

项目介绍

jkivilin/camellia-simd-aesni 是一个托管在 GitHub 上的开源项目，主要使用 C 编写，采用 MIT License 许可证，目前获得 20 个 Star、4 次 Fork。

仓库全名: jkivilin/camellia-simd-aesni

README 摘要

About In this repository, you find x86 (AES-NI, VAES, GFNI), ARMv8 (Neon/AES Crypto Extension), PowerPC crypto instruction set and RISC-V (RVA23+Zvkb+Zvkned) accelerated vector implementations of Camellia cipher. For x86-64 and aarch64, both Intel C intrinsics and assembly implementations are provided, with the instrinsics implementation being easier to port to other instruction sets. For PowerPC and RISC-V, a…

完整内容见 GitHub 原始 README 或本页。

常见问题

camellia-simd-aesni 用什么语言开发？

主要使用 C 编写，采用 MIT License 许可证，目前有 20 个 Star。

camellia-simd-aesni 可以免费商用吗？

如何获取 camellia-simd-aesni 的源码？

可以直接克隆仓库：git clone https://github.com/jkivilin/camellia-simd-aesni.git，默认分支为 master；安装与运行步骤见仓库 README。

camellia-simd-aesni 最近还在维护吗？

仓库最近一次更新为 1月2日，当前没有未关闭的 Issue；Star 数为 20，可作为社区活跃度的参考。

关于

Camellia cipher SIMD vector implementations for x86 (with AES-NI, VAES and/or GFNI instructions), ARM (with ARMv8 Crypto Extension instructions), POWER (with VMX+VSX+crypto instructions) and RISC-V (RVA23+Zvkb+Zvkned).

aesni armv8-a assembly block-cipher camellia gfni intrinsics powerpc simd

MIT License
20 Stars
4 Forks
2 Watching

描述取自 GitHub 官方 API，中文内容由 GitRead 中文社区编辑团队整理。

简要分析

热度分

价值分

活跃状态

活跃

GitHub

语言占比

Assembly

Makefile

贡献者

jkivilin

yapolyak

About

In this repository, you find x86 (AES-NI, VAES, GFNI), ARMv8 (Neon/AES Crypto Extension), PowerPC crypto instruction set and RISC-V (RVA23+Zvkb+Zvkned) accelerated vector implementations of Camellia cipher. For x86-64 and aarch64, both Intel C intrinsics and assembly implementations are provided, with the instrinsics implementation being easier to port to other instruction sets. For PowerPC and RISC-V, a 128-bit vector instrinsics implementation is provided.

How it works

It happens to be that Camellia uses s-box construction is very similar to AES SubBytes. With help of affine transforms, one can perform Camellia s-boxes with AES SubBytes and implementations here use SubBytes from AES-NI and similar AES vector instructions. Newer x86-64 processors also support Galois Field New Instructions (GFNI) which allow implementing Camellia s-box more straightforward manner and yield even better performance.

Due to the structure of Camellia cipher, at least 16 blocks needs to be processed in parallel for the best performance. Details can be found in Block Ciphers: Fast Implementations on x86-64 Architecture (pages 42-50).

Because of the requirement for parallel input blocks, these implementations are best suited for parallelizable cipher modes of operation, such as CTR, CBC decryption, CFB decryption, XTS, OCB, etc.

Implementations

SIMD128 - 1 block non-parallel

These SIMD128 (128-bit vector) implementation variants process one block at time. These provide constant-time alternative for reference implementation which uses table look-ups.

camellia_simd128_x86-64_aesni_avx.S:
- GCC assembly implementation variants for x86-64:
  - x86-64+AVX512+GFNI variant:
    - On Intel Core i3-1115G4 (tigerlake), this implementation is ~1.2 times faster than reference.
    - On AMD Ryzen 9 9950X3D (zen5), this implementation is ~1.2 times slower than reference.
  - x86-64+AVX+AES-NI variant:
    - On Intel Core i3-1115G4 (tigerlake), this implementation is ~1.6 times slower than reference.
    - On AMD Ryzen 9 9950X3D (zen5), this implementation is ~2.8 times slower than reference.
- Includes vector intrinsics implementation of Camellia key-setup (for 128-bit, 192-bit and 256-bit keys).
camellia_simd128_with_aes_instruction_set.c:
- C intrinsics implementation for x86 with AES-NI, for ARMv8 with Crypto Extension (CE), for PowerPC with AES crypto instruction set and RISC-V with RVA23+Zvkb+Zvkned.
  - x86 implementation requires AES-NI and either SSE4.1 or AVX instruction set and gets best performance with x86-64 + AVX.
  - ARM implementation requires AArch64, NEON and ARMv8 AES CE instruction set.
  - PowerPC implementation requires VSX and AES crypto instruction set.
  - RISC-V implementation requires 64-bit RVA23 with vector cryptography Zvkb and Zvkned extensions.
- Includes vector intrinsics implementation of Camellia key-setup (for 128-bit, 192-bit and 256-bit keys).
- On ARM/Cortex-A53, this implementation is ~1.4 times slower than reference (compiled with gcc-15).
- On Intel Core i3-1115G4 (tigerlake), this implementation is ~1.6 times slower than reference (compiled with gcc-15).
- On AMD Ryzen 9 9950X3D (zen5), this implementation is ~2.9 times slower than reference (compiled with gcc-15).

SIMD128 - 16 block parallel

These SIMD128 (128-bit vector) implementation variants process 16 blocks in parallel.

camellia_simd128_with_aes_instruction_set.c:
- C intrinsics implementation for x86 with AES-NI, for ARMv8 with Crypto Extension (CE), for PowerPC with AES crypto instruction set and RISC-V with RVA23+Zvkb+Zvkned.
  - x86 implementation requires AES-NI and either SSE4.1 or AVX instruction set and gets best performance with x86-64 + AVX.
  - ARM implementation requires AArch64, NEON and ARMv8 AES CE instruction set.
  - PowerPC implementation requires VSX and AES crypto instruction set.
  - RISC-V implementation requires 64-bit RVA23 with vector cryptography Zvkb and Zvkned extensions.
- Includes vector intrinsics implementation of Camellia key-setup (for 128-bit, 192-bit and 256-bit keys).
- On Intel Core i5-6500 (skylake), this implementation is ~3.5 times faster than reference.
- On Intel Core i3-1115G4 (tigerlake), this implementation is ~4.0 times faster than reference (compiled with gcc-14).
- On AMD Ryzen 9 9950X3D (zen5), this implementation is ~3.1 times faster than reference (compiled with gcc-15).
- On ARM/ThunderX2, this implementation is ~3.0 times faster than reference (compiled with gcc-13).
- On ARM/Cortex-A53, this implementation is ~2.2 times faster than reference (compiled with gcc-15).
- On POWER9/ppc64le, this implementation is ~2.4 times faster than reference.
camellia_simd128_x86-64_aesni_avx.S:
- GCC assembly implementation for x86-64 with AES-NI and AVX.
- Includes vector assembly implementation of Camellia key-setup (for 128-bit, 192-bit and 256-bit keys).
- On Intel Core i5-6500 (skylake), this implementation is ~3.6 times faster than reference.
- On Intel Core i3-1115G4 (tigerlake), this implementation is ~4.3 times faster than reference.
- On AMD Ryzen 9 7900X (zen4), this implementation is ~4.5 times faster than reference.
- On AMD Ryzen 9 9950X3D (zen5), this implementation is ~3.6 times faster than reference.
camellia_simd128_armv8_neon_aese.S:
- GCC assembly implementation for armv8 with Neon and AES CE.
- Includes vector assembly implementation of Camellia key-setup (for 128-bit, 192-bit and 256-bit keys).
- On ARM/ThunderX2, this implementation is ~2.7 times faster than reference.
- On ARM/Cortex-A53, this implementation is ~2.3 times faster than reference.

SIMD256 - 32 block parallel

The SIMD256 (256-bit vector) implementation variants process 32 blocks in parallel.

camellia_simd256_x86_aesni.c:
- Intel C intrinsics implentation for x86 with AES-NI or VAES or GFNI. Requires either AVX2 instruction set and gets best performance with x86-64 + AVX512 + GFNI.
- On Intel Core i5-6500 (skylake), when compiled for x86-64+AVX2+AES-NI, this implementation is ~5.4 times faster than reference.
- On AMD Ryzen 9 7900X (zen4), when compiled for x86-64+AVX512+VAES, this implementation is ~8.9 times faster than reference.
- On AMD Ryzen 9 7900X (zen4), when compiled for x86-64+AVX512+GFNI, this implementation is ~18.7 times faster than reference.
- On Intel Core i3-1115G4 (tigerlake), when compiled for x86-64+AVX512+VAES, this implementation is ~8.6 times faster than reference (compiled with gcc-14).
- On Intel Core i3-1115G4 (tigerlake), when compiled for x86-64+AVX512+GFNI, this implementation is ~17.8 times faster than reference (compiled with gcc-14).
- On AMD Ryzen 9 9950X3D (zen5), when compiled for x86-64+AVX512+VAES, this implementation is ~7.4 times faster than reference (compiled with gcc-15).
- On AMD Ryzen 9 9950X3D (zen5), when compiled for x86-64+AVX512+GFNI, this implementation is ~14.1 times faster than reference (compiled with gcc-15).
camellia_simd256_x86-64_aesni_avx2.S:
- GCC assembly implementation for x86-64 with AES-NI/VAES/GFNI AVX2.
- On Intel Core i5-6500 (skylake), when compiled for x86-64+AVX2+AES-NI, this implementation is ~5.8 times faster than reference.
- On AMD Ryzen 9 7900X (zen4), when compiled for x86-64+AVX2+VAES, this implementation is ~9.2 times faster than reference.
- On AMD Ryzen 9 7900X (zen4), when compiled for x86-64+AVX2+GFNI, this implementation is ~18.2 times faster than reference (~0.92 cycles/byte).
- On AMD Ryzen 9 9950X3D (zen5), when compiled for x86-64+AVX2+VAES, this implementation is ~7.3 times faster than reference.
- On AMD Ryzen 9 9950X3D (zen5), when compiled for x86-64+AVX2+GFNI, this implementation is ~14.1 times faster than reference.

Compiling and testing

Prerequisites

GNU make
GCC x86-64
Optionally GCC i686
Optionally GCC aarch64
Optionally GCC powerpc64le
Optionally GCC riscv64
Ubuntu 22.04 packages: gcc gcc-i686-linux-gnu gcc-aarch64-linux-gnu gcc-powerpc64le-linux-gnu gcc-riscv64-linux-gnu make

Compiling

Clone repository and run ‘make’…

$ make
x86_64-linux-gnu-gcc -O2 -Wall -march=sandybridge -mtune=native -msse4.1 -maes -c camellia_simd128_with_aes_instruction_set.c -o camellia_simd128_with_x86_aesni.o
x86_64-linux-gnu-gcc -O2 -Wall -c main.c -o main_simd128.o
x86_64-linux-gnu-gcc -O2 -Wall -c camellia-BSD-1.2.0/camellia.c -o camellia_ref_x86-64.o
x86_64-linux-gnu-gcc camellia_simd128_with_x86_aesni.o main_simd128.o camellia_ref_x86-64.o -o test_simd128_intrinsics_x86_64
x86_64-linux-gnu-gcc -O2 -Wall -march=haswell -mtune=native -mavx2 -maes -c camellia_simd128_with_aes_instruction_set.c -o camellia_simd128_with_x86_aesni_avx2.o
x86_64-linux-gnu-gcc -O2 -Wall -march=haswell -mtune=native -mavx2 -maes -c camellia_simd256_x86_aesni.c -o camellia_simd256_x86_aesni.o
x86_64-linux-gnu-gcc -O2 -Wall -DUSE_SIMD256 -c main.c -o main_simd256.o
x86_64-linux-gnu-gcc camellia_simd128_with_x86_aesni_avx2.o camellia_simd256_x86_aesni.o main_simd256.o camellia_ref_x86-64.o -o test_simd256_intrinsics_x86_64
x86_64-linux-gnu-gcc -O2 -Wall -march=haswell -mtune=native -mavx2 -maes -mvaes -DUSE_VAES -c camellia_simd256_x86_aesni.c -o camellia_simd256_x86_vaes.o
x86_64-linux-gnu-gcc camellia_simd128_with_x86_aesni_avx2.o camellia_simd256_x86_vaes.o main_simd256.o camellia_ref_x86-64.o -o test_simd256_intrinsics_x86_64_vaes
x86_64-linux-gnu-gcc -O2 -Wall -march=znver4 -mavx512f -mavx512vl -mavx512bw -mavx512dq -mavx512vbmi -mavx512ifma -mavx512vpopcntdq -mavx512vbmi2 -mavx512bitalg -mavx512vnni -mprefer-vector-width=512 -mavx2 -maes -mvaes -mgfni -c camellia_simd128_with_aes_instruction_set.c -o camellia_simd128_with_x86_aesni_avx512.o
x86_64-linux-gnu-gcc -O2 -Wall -march=znver4 -mavx512f -mavx512vl -mavx512bw -mavx512dq -mavx512vbmi -mavx512ifma -mavx512vpopcntdq -mavx512vbmi2 -mavx512bitalg -mavx512vnni -mprefer-vector-width=512 -mavx2 -maes -mvaes -mgfni -DUSE_VAES -c camellia_simd256_x86_aesni.c -o camellia_simd256_x86_vaes_avx512.o
x86_64-linux-gnu-gcc camellia_simd128_with_x86_aesni_avx512.o camellia_simd256_x86_vaes_avx512.o main_simd256.o camellia_ref_x86-64.o -o test_simd256_intrinsics_x86_64_vaes_avx512
x86_64-linux-gnu-gcc -O2 -Wall -march=znver4 -mavx512f -mavx512vl -mavx512bw -mavx512dq -mavx512vbmi -mavx512ifma -mavx512vpopcntdq -mavx512vbmi2 -mavx512bitalg -mavx512vnni -mprefer-vector-width=512 -mavx2 -maes -mvaes -mgfni -DUSE_GFNI -c camellia_simd256_x86_aesni.c -o camellia_simd256_x86_gfni_avx512.o
x86_64-linux-gnu-gcc camellia_simd128_with_x86_aesni_avx512.o camellia_simd256_x86_gfni_avx512.o main_simd256.o camellia_ref_x86-64.o -o test_simd256_intrinsics_x86_64_gfni_avx512
x86_64-linux-gnu-gcc -O2 -Wall -c camellia_simd128_x86-64_aesni_avx.S -o camellia_simd128_x86-64_aesni_avx.o
x86_64-linux-gnu-gcc camellia_simd128_x86-64_aesni_avx.o main_simd128.o camellia_ref_x86-64.o -o test_simd128_asm_x86_64
x86_64-linux-gnu-gcc -O2 -Wall -c camellia_simd256_x86-64_aesni_avx2.S -o camellia_simd256_x86-64_aesni_avx2.o
x86_64-linux-gnu-gcc camellia_simd128_x86-64_aesni_avx.o camellia_simd256_x86-64_aesni_avx2.o main_simd256.o camellia_ref_x86-64.o -o test_simd256_asm_x86_64
x86_64-linux-gnu-gcc -O2 -Wall -DUSE_VAES -c camellia_simd256_x86-64_aesni_avx2.S -o camellia_simd256_x86-64_vaes_avx2.o
x86_64-linux-gnu-gcc camellia_simd128_x86-64_aesni_avx.o camellia_simd256_x86-64_vaes_avx2.o main_simd256.o camellia_ref_x86-64.o -o test_simd256_asm_x86_64_vaes
x86_64-linux-gnu-gcc -O2 -Wall -DUSE_GFNI -c camellia_simd256_x86-64_aesni_avx2.S -o camellia_simd256_x86-64_gfni_avx2.o
x86_64-linux-gnu-gcc camellia_simd128_x86-64_aesni_avx.o camellia_simd256_x86-64_gfni_avx2.o main_simd256.o camellia_ref_x86-64.o -o test_simd256_asm_x86_64_gfni
x86_64-linux-gnu-gcc -O2 -Wall -DUSE_GFNI -DUSE_AVX512 -c camellia_simd128_x86-64_aesni_avx.S -o camellia_simd128_x86-64_aesni_avx+avx512+gfni.o
x86_64-linux-gnu-gcc camellia_simd128_x86-64_aesni_avx+avx512+gfni.o camellia_simd256_x86-64_gfni_avx2.o main_simd256.o camellia_ref_x86-64.o -o test_simd256_asm_x86_64_gfni_avx512
i686-linux-gnu-gcc -O2 -Wall -march=sandybridge -mtune=native -msse4.1 -maes -c camellia_simd128_with_aes_instruction_set.c -o camellia_simd128_with_x86_aesni_i386.o
i686-linux-gnu-gcc -O2 -Wall -c main.c -o main_simd128_i386.o
i686-linux-gnu-gcc -O2 -Wall -c camellia-BSD-1.2.0/camellia.c -o camellia_ref_i386.o
i686-linux-gnu-gcc camellia_simd128_with_x86_aesni_i386.o main_simd128_i386.o camellia_ref_i386.o -o test_simd128_intrinsics_i386
i686-linux-gnu-gcc -O2 -Wall -march=haswell -mtune=native -mavx2 -maes -c camellia_simd128_with_aes_instruction_set.c -o camellia_simd128_with_x86_aesni_avx2_i386.o
i686-linux-gnu-gcc -O2 -Wall -march=haswell -mtune=native -mavx2 -maes -c camellia_simd256_x86_aesni.c -o camellia_simd256_x86_aesni_i386.o
i686-linux-gnu-gcc -O2 -Wall -DUSE_SIMD256 -c main.c -o main_simd256_i386.o
i686-linux-gnu-gcc camellia_simd128_with_x86_aesni_avx2_i386.o camellia_simd256_x86_aesni_i386.o main_simd256_i386.o camellia_ref_i386.o -o test_simd256_intrinsics_i386
aarch64-linux-gnu-gcc -O2 -Wall -march=armv8-a+crypto -mtune=cortex-a53 -c camellia_simd128_with_aes_instruction_set.c -o camellia_simd128_with_aarch64_ce.o
aarch64-linux-gnu-gcc -O2 -Wall -march=armv8-a+crypto -mtune=cortex-a53 -c main.c -o main_simd128_aarch64.o
aarch64-linux-gnu-gcc -O2 -Wall -march=armv8-a+crypto -mtune=cortex-a53 -c camellia-BSD-1.2.0/camellia.c -o camellia_ref_aarch64.o
aarch64-linux-gnu-gcc -static camellia_simd128_with_aarch64_ce.o main_simd128_aarch64.o camellia_ref_aarch64.o -o test_simd128_intrinsics_aarch64
aarch64-linux-gnu-gcc -O2 -Wall -march=armv8-a+crypto -mtune=cortex-a53 -c camellia_simd128_armv8_neon_aese.S -o camellia_simd128_armv8_neon_aese.o
aarch64-linux-gnu-gcc -static camellia_simd128_armv8_neon_aese.o main_simd128_aarch64.o camellia_ref_aarch64.o -o test_simd128_asm_armv8
powerpc64le-linux-gnu-gcc -O2 -Wall -mcpu=power8 -maltivec -mvsx -mcrypto -c camellia_simd128_with_aes_instruction_set.c -o camellia_simd128_with_ppc64le.o
powerpc64le-linux-gnu-gcc -O2 -Wall -mcpu=power8 -maltivec -mvsx -mcrypto -c main.c -o main_simd128_ppc64le.o
powerpc64le-linux-gnu-gcc -O2 -Wall -mcpu=power8 -maltivec -mvsx -mcrypto -c camellia-BSD-1.2.0/camellia.c -o camellia_ref_ppc64le.o
powerpc64le-linux-gnu-gcc camellia_simd128_with_ppc64le.o main_simd128_ppc64le.o camellia_ref_ppc64le.o -o test_simd128_intrinsics_ppc64le
riscv64-linux-gnu-gcc -O2 -Wall -mstrict-align -march=rv64imafdcv_zba_zbb_zbs_zvkb_zvkned  -c camellia_simd128_with_aes_instruction_set.c -o camellia_simd128_with_riscv64.o
riscv64-linux-gnu-gcc -O2 -Wall -mstrict-align -march=rv64imafdcv_zba_zbb_zbs_zvkb_zvkned  -c main.c -o main_simd128_riscv64.o
riscv64-linux-gnu-gcc -O2 -Wall -mstrict-align -march=rv64imafdcv_zba_zbb_zbs_zvkb_zvkned  -c camellia-BSD-1.2.0/camellia.c -o camellia_ref_riscv64.o
riscv64-linux-gnu-gcc camellia_simd128_with_riscv64.o main_simd128_riscv64.o camellia_ref_riscv64.o -o test_simd128_intrinsics_riscv64

Testing

Fifteen executables are build. Run executables to verify implementation against test-vectors (with 128-bit, 192-bit and 256-bit key lengths) and benchmark against reference implementation from OpenSSL (with 128-bit key length).

Executables are:

test_simd128_asm_x86_64: SIMD128 only, for testing assembly x86-64/AES-NI/AVX implementation without AVX2.
test_simd128_asm_armv8: SIMD128 only, for testing armv8 assembly (Neon/AES) implementation.
test_simd128_intrinsics_i386: SIMD128 only, for testing intrinsics implementation on i386/AES-NI/AVX without AVX2.
test_simd128_intrinsics_x86_64: SIMD128 only, for testing intrinsics implementation on x86_64/AES-NI/AVX without AVX2.
test_simd128_intrinsics_aarch64: SIMD128 only, for testing intrinsics implementation on ARMv8 AArch64 with Crypto Extensions.
test_simd128_intrinsics_ppc64le: SIMD128 only, for testing intrinsics implementation on little-endian 64-bit PowerPC with crypto instruction set.
test_simd128_intrinsics_riscv64: SIMD128 only, for testing intrinsics implementation on 64-bit RISC-V with RVA23+Zvkb+Zvkned.
test_simd256_asm_x86_64: SIMD256 and SIMD128, for testing assembly x86-64/AES-NI/AVX2 implementations.
test_simd256_asm_x86_64_gfni: SIMD256 and SIMD128, for testing assembly x86-64/AES-NI/AVX2 implementations.
test_simd256_asm_x86_64_vaes: SIMD256 and SIMD128, for testing assembly x86-64/AES-NI/AVX2 implementations.
test_simd256_intrinsics_i386: SIMD256 and SIMD128, for testing intrinsics implementations on i386/AES-NI/AVX2.
test_simd256_intrinsics_x86_64: SIMD256 and SIMD128, for testing intrinsics implementation on x86_64/AES-NI/AVX2.
test_simd256_intrinsics_x86_64_vaes: SIMD256 and SIMD128, for testing intrinsics implementation on x86_64/VAES/AVX2.
test_simd256_intrinsics_x86_64_vaes_avx512: SIMD256 and SIMD128, for testing intrinsics implementation on x86_64/VAES/AVX512.
test_simd256_intrinsics_x86_64_gfni_avx512: SIMD256 and SIMD128, for testing intrinsics implementation on x86_64/GFNI/AVX512.

For example, output of test_simd256_asm_x86_64 and test_simd256_intrinsics_x86_64_gfni_avx512 on AMD Ryzen 9 7900X:

$ ./test_simd256_asm_x86_64
./test_simd256_asm_x86_64:
selftest: comparing camellia-128 test vectors against reference implementation...
selftest: comparing camellia-192 test vectors against reference implementation...
selftest: comparing camellia-256 test vectors against reference implementation...
selftest: checking 1-block parallel camellia-128/SIMD128 against test vectors...
selftest: checking 1-block parallel camellia-192/SIMD128 against test vectors...
selftest: checking 1-block parallel camellia-256/SIMD128 against test vectors...
selftest: checking 16-block parallel camellia-128/SIMD128 against test vectors...
selftest: checking 16-block parallel camellia-192/SIMD128 against test vectors...
selftest: checking 16-block parallel camellia-256/SIMD128 against test vectors...
selftest: checking 32-block parallel camellia-128/SIMD256 against test vectors...
selftest: checking 32-block parallel camellia-192/SIMD256 against test vectors...
selftest: checking 32-block parallel camellia-256/SIMD256 against test vectors...
selftest: checking 16-block parallel camellia-128/SIMD128 against large test vectors...
selftest: checking 16-block parallel camellia-256/SIMD128 against large test vectors...
selftest: checking 32-block parallel camellia-128/SIMD256 against large test vectors...
selftest: checking 32-block parallel camellia-256/SIMD256 against large test vectors...
           camellia-128 reference encryption:    372.041 Mebibytes/s,    390.113 Megabytes/s
           camellia-128 reference decryption:    371.912 Mebibytes/s,    389.979 Megabytes/s
   camellia-128 SIMD128 (1 block) encryption:    238.747 Mebibytes/s,    250.345 Megabytes/s
   camellia-128 SIMD128 (1 block) decryption:    238.891 Mebibytes/s,    250.495 Megabytes/s
 camellia-128 SIMD128 (16 blocks) encryption:   1347.869 Mebibytes/s,   1413.343 Megabytes/s
 camellia-128 SIMD128 (16 blocks) decryption:   1345.991 Mebibytes/s,   1411.374 Megabytes/s
 camellia-128 SIMD256 (32 blocks) encryption:   5159.460 Mebibytes/s,   5410.086 Megabytes/s
 camellia-128 SIMD256 (32 blocks) decryption:   5155.399 Mebibytes/s,   5405.827 Megabytes/s
$ ./test_simd256_intrinsics_x86_64_gfni_avx512
./test_simd256_intrinsics_x86_64_gfni_avx512:
selftest: comparing camellia-128 test vectors against reference implementation...
selftest: comparing camellia-192 test vectors against reference implementation...
selftest: comparing camellia-256 test vectors against reference implementation...
selftest: checking 16-block parallel camellia-128/SIMD128 against test vectors...
selftest: checking 16-block parallel camellia-192/SIMD128 against test vectors...
selftest: checking 16-block parallel camellia-256/SIMD128 against test vectors...
selftest: checking 32-block parallel camellia-128/SIMD256 against test vectors...
selftest: checking 32-block parallel camellia-192/SIMD256 against test vectors...
selftest: checking 32-block parallel camellia-256/SIMD256 against test vectors...
selftest: checking 16-block parallel camellia-128/SIMD128 against large test vectors...
selftest: checking 16-block parallel camellia-256/SIMD128 against large test vectors...
selftest: checking 32-block parallel camellia-128/SIMD256 against large test vectors...
selftest: checking 32-block parallel camellia-256/SIMD256 against large test vectors...
           camellia-128 reference encryption:    313.426 Mebibytes/s,    328.651 Megabytes/s
           camellia-128 reference decryption:    313.192 Mebibytes/s,    328.406 Megabytes/s
 camellia-128 SIMD128 (16 blocks) encryption:   1454.039 Mebibytes/s,   1524.670 Megabytes/s
 camellia-128 SIMD128 (16 blocks) decryption:   1441.535 Mebibytes/s,   1511.559 Megabytes/s
 camellia-128 SIMD256 (32 blocks) encryption:   5977.907 Mebibytes/s,   6268.290 Megabytes/s
 camellia-128 SIMD256 (32 blocks) decryption:   5965.201 Mebibytes/s,   6254.967 Megabytes/s

For example, output of test_simd128_intrinsics_aarch64 on ARM Cortex-A53 (648 Mhz):

$ ./test_simd128_intrinsics_aarch64
./test_simd128_intrinsics_aarch64:
selftest: comparing camellia-128 test vectors against reference implementation...
selftest: comparing camellia-192 test vectors against reference implementation...
selftest: comparing camellia-256 test vectors against reference implementation...
selftest: checking 16-block parallel camellia-128/SIMD128 against test vectors...
selftest: checking 16-block parallel camellia-192/SIMD128 against test vectors...
selftest: checking 16-block parallel camellia-256/SIMD128 against test vectors...
selftest: checking 16-block parallel camellia-128/SIMD128 against large test vectors...
selftest: checking 16-block parallel camellia-256/SIMD128 against large test vectors...
           camellia-128 reference encryption:     29.027 Mebibytes/s,     30.437 Megabytes/s
           camellia-128 reference decryption:     29.372 Mebibytes/s,     30.799 Megabytes/s
 camellia-128 SIMD128 (16 blocks) encryption:     36.700 Mebibytes/s,     38.483 Megabytes/s
 camellia-128 SIMD128 (16 blocks) decryption:     36.164 Mebibytes/s,     37.921 Megabytes/s

On ThunderX2 (2.2GHz) output of test_simd128_intrinsics_aarch64 is as follows (compiled with -mtune=native):

$ taskset -c 0 ./test_simd128_intrinsics_aarch64
./test_simd128_intrinsics_aarch64:
selftest: comparing camellia-128 test vectors against reference implementation...
selftest: comparing camellia-192 test vectors against reference implementation...
selftest: comparing camellia-256 test vectors against reference implementation...
selftest: checking 16-block parallel camellia-128/SIMD128 against test vectors...
selftest: checking 16-block parallel camellia-192/SIMD128 against test vectors...
selftest: checking 16-block parallel camellia-256/SIMD128 against test vectors...
selftest: checking 16-block parallel camellia-128/SIMD128 against large test vectors...
selftest: checking 16-block parallel camellia-256/SIMD128 against large test vectors...
           camellia-128 reference encryption:     98.590 Mebibytes/s,    103.379 Megabytes/s
           camellia-128 reference decryption:     98.588 Mebibytes/s,    103.377 Megabytes/s
 camellia-128 SIMD128 (16 blocks) encryption:    296.305 Mebibytes/s,    310.699 Megabytes/s
 camellia-128 SIMD128 (16 blocks) decryption:    287.411 Mebibytes/s,    301.372 Megabytes/s

and output of test_simd128_asm_armv8 is as follows:

$ taskset -c 0 ./test_simd128_asm_armv8
./test_simd128_asm_armv8:
selftest: comparing camellia-128 test vectors against reference implementation...
selftest: comparing camellia-192 test vectors against reference implementation...
selftest: comparing camellia-256 test vectors against reference implementation...
selftest: checking 16-block parallel camellia-128/SIMD128 against test vectors...
selftest: checking 16-block parallel camellia-192/SIMD128 against test vectors...
selftest: checking 16-block parallel camellia-256/SIMD128 against test vectors...
selftest: checking 16-block parallel camellia-128/SIMD128 against large test vectors...
selftest: checking 16-block parallel camellia-256/SIMD128 against large test vectors...
           camellia-128 reference encryption:     98.563 Mebibytes/s,    103.350 Megabytes/s
           camellia-128 reference decryption:     98.579 Mebibytes/s,    103.367 Megabytes/s
 camellia-128 SIMD128 (16 blocks) encryption:    268.200 Mebibytes/s,    281.228 Megabytes/s
 camellia-128 SIMD128 (16 blocks) decryption:    265.601 Mebibytes/s,    278.503 Megabytes/s

暂无评论

camellia-simd-aesni