Implementation notes: x86, thoth, crypto_aead/norx6461v1

Computer: thoth
Architecture: x86
CPU ID: AuthenticAMD-00000622-0183f9ff
SUPERCOP version: 20160806
Operation: crypto_aead
Primitive: norx6461v1

Time	Implementation	Compiler	Benchmark date	SUPERCOP version
209871	`ref`	`clang -march=native -O3 -fomit-frame-pointer -fwrapv -Qunused-arguments`	20160727	20160724
391502	`ref`	`gcc -march=native -mtune=native -Os -fomit-frame-pointer -fwrapv`	20160727	20160724
429192	`ref`	`gcc -march=native -mtune=native -O -fomit-frame-pointer -fwrapv`	20160727	20160724
429903	`ref`	`gcc -march=native -mtune=native -O3 -fomit-frame-pointer -fwrapv`	20160727	20160724
432309	`ref`	`gcc -march=native -mtune=native -O2 -fomit-frame-pointer -fwrapv`	20160727	20160724

Compiler output

Implementation: crypto_aead/norx6461v1/ymm
Compiler: clang -march=native -O3 -fomit-frame-pointer -fwrapv -Qunused-arguments

norx.c: norx.c:349:24: error: always_inline function '_mm_loadu_si128' requires target feature 'sse2', but would be inlined into function 'crypto_aead_norx6461v1_ymm_encrypt' that is compiled without support for 'sse2'
norx.c: const __m128i N = LOADU128(npub);
norx.c: ^
norx.c: norx.c:52:22: note: expanded from macro 'LOADU128'
norx.c: #define LOADU128(in) _mm_loadu_si128((__m128i*)(in))
norx.c: ^
norx.c: norx.c:350:24: error: always_inline function '_mm256_loadu_si256' requires target feature 'sse3', but would be inlined into function 'crypto_aead_norx6461v1_ymm_encrypt' that is compiled without support for 'sse3'
norx.c: const __m256i K = LOADU(k + 0);
norx.c: ^
norx.c: norx.c:47:19: note: expanded from macro 'LOADU'
norx.c: #define LOADU(in) _mm256_loadu_si256((__m256i*)(in))
norx.c: ^
norx.c: norx.c:355:5: error: always_inline function '_mm256_castsi128_si256' requires target feature 'sse3', but would be inlined into function 'crypto_aead_norx6461v1_ymm_encrypt' that is compiled without support for 'sse3'
norx.c: INITIALIZE(A, B, C, D, N, K);
norx.c: ^
norx.c: norx.c:270:9: note: expanded from macro 'INITIALIZE'
norx.c: A = _mm256_castsi128_si256(N); \
norx.c: ^
norx.c: norx.c:355:5: error: always_inline function '_mm256_castsi128_si256' requires target feature 'sse3', but would be inlined into function 'crypto_aead_norx6461v1_ymm_encrypt' that is compiled without support for 'sse3'
norx.c: norx.c:271:9: note: expanded from macro 'INITIALIZE'
norx.c: A = _mm256_inserti128_si256(A, _mm_set_epi64x(U1, U0), 1); \
norx.c: ^
norx.c: /usr/lib/llvm-3.8/bin/../lib/clang/3.8.0/include/avx2intrin.h:892:44: note: expanded from macro '_mm256_inserti128_si256'
norx.c: (__v4di)_mm256_castsi128_si256((__m128i)(V2)), \
norx.c: ^
norx.c: ...

Number of similar (compiler,implementation) pairs: 1, namely:

Compiler	Implementations
clang -march=native -O3 -fomit-frame-pointer -fwrapv -Qunused-arguments	ymm

Compiler output

Implementation: crypto_aead/norx6461v1/xmm
Compiler: clang -march=native -O3 -fomit-frame-pointer -fwrapv -Qunused-arguments

norx.c: norx.c:411:24: error: always_inline function '_mm_loadu_si128' requires target feature 'sse2', but would be inlined into function 'crypto_aead_norx6461v1_xmm_encrypt' that is compiled without support for 'sse2'
norx.c: const __m128i N = LOADU(npub);
norx.c: ^
norx.c: norx.c:47:19: note: expanded from macro 'LOADU'
norx.c: #define LOADU(in) _mm_loadu_si128((__m128i*)(in))
norx.c: ^
norx.c: norx.c:412:24: error: always_inline function '_mm_loadu_si128' requires target feature 'sse2', but would be inlined into function 'crypto_aead_norx6461v1_xmm_encrypt' that is compiled without support for 'sse2'
norx.c: const __m128i K0 = LOADU(k + 0);
norx.c: ^
norx.c: norx.c:47:19: note: expanded from macro 'LOADU'
norx.c: #define LOADU(in) _mm_loadu_si128((__m128i*)(in))
norx.c: ^
norx.c: norx.c:413:24: error: always_inline function '_mm_loadu_si128' requires target feature 'sse2', but would be inlined into function 'crypto_aead_norx6461v1_xmm_encrypt' that is compiled without support for 'sse2'
norx.c: const __m128i K1 = LOADU(k + 16);
norx.c: ^
norx.c: norx.c:47:19: note: expanded from macro 'LOADU'
norx.c: #define LOADU(in) _mm_loadu_si128((__m128i*)(in))
norx.c: ^
norx.c: norx.c:418:5: error: always_inline function '_mm_xor_si128' requires target feature 'sse2', but would be inlined into function 'crypto_aead_norx6461v1_xmm_encrypt' that is compiled without support for 'sse2'
norx.c: INITIALIZE(A0, A1, B0, B1, C0, C1, D0, D1, N, K0, K1);
norx.c: ^
norx.c: norx.c:327:10: note: expanded from macro 'INITIALIZE'
norx.c: A0 = XOR(_mm_slli_si128(N, 8), _mm_set_epi64x(0, U0)); \
norx.c: ^
norx.c: norx.c:69:19: note: expanded from macro 'XOR'
norx.c: ...

Number of similar (compiler,implementation) pairs: 1, namely:

Compiler	Implementations
clang -march=native -O3 -fomit-frame-pointer -fwrapv -Qunused-arguments	xmm

Compiler output

Implementation: crypto_aead/norx6461v1/xmm
Compiler: gcc -march=native -mtune=native -O2 -fomit-frame-pointer -fwrapv

norx.c: norx.c: In function 'crypto_aead_norx6461v1_xmm_encrypt':
norx.c: norx.c:411:19: warning: SSE vector return without SSE enabled changes the ABI [-Wpsabi]
norx.c: const __m128i N = LOADU(npub);
norx.c: ^
norx.c: In file included from /usr/lib/gcc/i686-linux-gnu/5/include/xmmintrin.h:1249:0,
norx.c: from /usr/lib/gcc/i686-linux-gnu/5/include/x86intrin.h:31,
norx.c: from norx.c:26:
norx.c: norx.c: In function 'block_copy':
norx.c: /usr/lib/gcc/i686-linux-gnu/5/include/emmintrin.h:714:1: error: inlining failed in call to always_inline '_mm_storeu_si128': target specific option mismatch
norx.c: _mm_storeu_si128 (__m128i *__P, __m128i __B)
norx.c: ^
norx.c: norx.c:48:24: error: called from here
norx.c: #define STOREU(out, x) _mm_storeu_si128((__m128i*)(out), (x))
norx.c: ^
norx.c: norx.c:364:9: note: in expansion of macro 'STOREU'
norx.c: STOREU(out + 48, LOADU(in + 48));
norx.c: ^
norx.c: In file included from /usr/lib/gcc/i686-linux-gnu/5/include/xmmintrin.h:1249:0,
norx.c: from /usr/lib/gcc/i686-linux-gnu/5/include/x86intrin.h:31,
norx.c: from norx.c:26:
norx.c: /usr/lib/gcc/i686-linux-gnu/5/include/emmintrin.h:696:1: error: inlining failed in call to always_inline '_mm_loadu_si128': target specific option mismatch
norx.c: _mm_loadu_si128 (__m128i const *__P)
norx.c: ^
norx.c: norx.c:48:24: error: called from here
norx.c: #define STOREU(out, x) _mm_storeu_si128((__m128i*)(out), (x))
norx.c: ...

Number of similar (compiler,implementation) pairs: 4, namely:

Compiler	Implementations
gcc -march=native -mtune=native -O2 -fomit-frame-pointer -fwrapv	xmm
gcc -march=native -mtune=native -O3 -fomit-frame-pointer -fwrapv	xmm
gcc -march=native -mtune=native -O -fomit-frame-pointer -fwrapv	xmm
gcc -march=native -mtune=native -Os -fomit-frame-pointer -fwrapv	xmm

Compiler output

Implementation: crypto_aead/norx6461v1/ymm
Compiler: gcc -march=native -mtune=native -O2 -fomit-frame-pointer -fwrapv

norx.c: norx.c: In function 'crypto_aead_norx6461v1_ymm_encrypt':
norx.c: norx.c:349:19: warning: SSE vector return without SSE enabled changes the ABI [-Wpsabi]
norx.c: const __m128i N = LOADU128(npub);
norx.c: ^
norx.c: norx.c:350:19: warning: AVX vector return without AVX enabled changes the ABI [-Wpsabi]
norx.c: const __m256i K = LOADU(k + 0);
norx.c: ^
norx.c: In file included from /usr/lib/gcc/i686-linux-gnu/5/include/immintrin.h:41:0,
norx.c: from /usr/lib/gcc/i686-linux-gnu/5/include/x86intrin.h:46,
norx.c: from norx.c:26:
norx.c: norx.c: In function 'block_copy':
norx.c: /usr/lib/gcc/i686-linux-gnu/5/include/avxintrin.h:900:1: error: inlining failed in call to always_inline '_mm256_storeu_si256': target specific option mismatch
norx.c: _mm256_storeu_si256 (__m256i *__P, __m256i __A)
norx.c: ^
norx.c: norx.c:48:24: error: called from here
norx.c: #define STOREU(out, x) _mm256_storeu_si256((__m256i*)(out), (x))
norx.c: ^
norx.c: norx.c:303:9: note: in expansion of macro 'STOREU'
norx.c: STOREU(out + 32, LOADU(in + 32));
norx.c: ^
norx.c: In file included from /usr/lib/gcc/i686-linux-gnu/5/include/immintrin.h:41:0,
norx.c: from /usr/lib/gcc/i686-linux-gnu/5/include/x86intrin.h:46,
norx.c: from norx.c:26:
norx.c: /usr/lib/gcc/i686-linux-gnu/5/include/avxintrin.h:894:1: error: inlining failed in call to always_inline '_mm256_loadu_si256': target specific option mismatch
norx.c: _mm256_loadu_si256 (__m256i const *__P)
norx.c: ...

Number of similar (compiler,implementation) pairs: 4, namely:

Compiler	Implementations
gcc -march=native -mtune=native -O2 -fomit-frame-pointer -fwrapv	ymm
gcc -march=native -mtune=native -O3 -fomit-frame-pointer -fwrapv	ymm
gcc -march=native -mtune=native -O -fomit-frame-pointer -fwrapv	ymm
gcc -march=native -mtune=native -Os -fomit-frame-pointer -fwrapv	ymm