Implementation notes: amd64, margaux, crypto_kem/ntskem1264

Computer: margaux
Microarchitecture: amd64; Core 2 65nm (6fb)
Architecture: amd64
CPU ID: GenuineIntel-000006fb-bfebfbff
SUPERCOP version: 20240107
Operation: crypto_kem
Primitive: ntskem1264

Time	Object size	Test size	Implementation	Compiler	Benchmark date	SUPERCOP version
787561	150404 6228 16	169117 7124 1728	`T:sse2`	`clang_-march=native_-O3_-fomit-frame-pointer_-fwrapv_-Qunused-arguments_-fPIC_-fPIE`	20231229	20231212
803316	78586 6228 16	96229 7124 1728	`T:sse2`	`clang_-march=native_-O2_-fomit-frame-pointer_-fwrapv_-Qunused-arguments_-fPIC_-fPIE`	20231229	20231212
818860	150971 6228 16	169565 7124 1728	`T:sse2`	`clang_-mcpu=native_-O3_-fomit-frame-pointer_-fwrapv_-Qunused-arguments_-fPIC_-fPIE`	20231229	20231212
885840	58383 6228 16	74039 7116 1728	`T:sse2`	`clang_-march=native_-Os_-fomit-frame-pointer_-fwrapv_-Qunused-arguments_-fPIC_-fPIE`	20231229	20231212
996848	82599 6228 16	99085 7124 1728	`T:sse2`	`clang_-march=native_-O_-fomit-frame-pointer_-fwrapv_-Qunused-arguments_-fPIC_-fPIE`	20231229	20231212
1085547	144411 6228 16	164485 7124 1728	`T:opt`	`clang_-march=native_-O3_-fomit-frame-pointer_-fwrapv_-Qunused-arguments_-fPIC_-fPIE`	20231229	20231212
1117481	145497 6228 16	165445 7124 1728	`T:opt`	`clang_-mcpu=native_-O3_-fomit-frame-pointer_-fwrapv_-Qunused-arguments_-fPIC_-fPIE`	20231229	20231212
1184084	77368 6228 16	96525 7124 1728	`T:opt`	`clang_-march=native_-O2_-fomit-frame-pointer_-fwrapv_-Qunused-arguments_-fPIC_-fPIE`	20231229	20231212
1200436	53271 6228 16	70423 7116 1728	`T:opt`	`clang_-march=native_-Os_-fomit-frame-pointer_-fwrapv_-Qunused-arguments_-fPIC_-fPIE`	20231229	20231212
1256326	91001 6228 16	111015 7100 1792	`T:opt`	`gcc_-march=native_-mtune=native_-O3_-fomit-frame-pointer_-fwrapv_-fPIC_-fPIE`	20231229	20231212
1258999	57361 6228 16	76351 7100 1792	`T:opt`	`gcc_-march=native_-mtune=native_-O2_-fomit-frame-pointer_-fwrapv_-fPIC_-fPIE`	20231229	20231212
1331383	55195 6228 16	73503 7100 1792	`T:opt`	`gcc_-march=native_-mtune=native_-O_-fomit-frame-pointer_-fwrapv_-fPIC_-fPIE`	20231229	20231212
1346095	77529 6228 16	95589 7124 1728	`T:opt`	`clang_-march=native_-O_-fomit-frame-pointer_-fwrapv_-Qunused-arguments_-fPIC_-fPIE`	20231229	20231212
1633752	50901 6228 16	67798 7092 1760	`T:opt`	`gcc_-march=native_-mtune=native_-Os_-fomit-frame-pointer_-fwrapv_-fPIC_-fPIE`	20231229	20231212
9679363	49820 76 16	69863 924 1792	`T:ref`	`gcc_-march=native_-mtune=native_-O3_-fomit-frame-pointer_-fwrapv_-fPIC_-fPIE`	20231229	20231212
9829724	28794 76 16	46837 964 1728	`T:ref`	`clang_-march=native_-O_-fomit-frame-pointer_-fwrapv_-Qunused-arguments_-fPIC_-fPIE`	20231229	20231212
10296814	37691 76 16	58165 964 1728	`T:ref`	`clang_-march=native_-O3_-fomit-frame-pointer_-fwrapv_-Qunused-arguments_-fPIC_-fPIE`	20231229	20231212
10360188	36112 76 16	55477 964 1728	`T:ref`	`clang_-march=native_-O2_-fomit-frame-pointer_-fwrapv_-Qunused-arguments_-fPIC_-fPIE`	20231229	20231212
10659165	25824 76 16	44719 924 1792	`T:ref`	`gcc_-march=native_-mtune=native_-O2_-fomit-frame-pointer_-fwrapv_-fPIC_-fPIE`	20231229	20231212
10841271	38066 76 16	58421 964 1728	`T:ref`	`clang_-mcpu=native_-O3_-fomit-frame-pointer_-fwrapv_-Qunused-arguments_-fPIC_-fPIE`	20231229	20231212
11443242	23411 76 16	41671 924 1792	`T:ref`	`gcc_-march=native_-mtune=native_-O_-fomit-frame-pointer_-fwrapv_-fPIC_-fPIE`	20231229	20231212
11491930	20328 76 16	37278 916 1760	`T:ref`	`gcc_-march=native_-mtune=native_-Os_-fomit-frame-pointer_-fwrapv_-fPIC_-fPIE`	20231229	20231212
11675157	22193 76 16	39359 956 1728	`T:ref`	`clang_-march=native_-Os_-fomit-frame-pointer_-fwrapv_-Qunused-arguments_-fPIC_-fPIE`	20231229	20231212

Compiler output

Implementation: T:avx2
Security model: timingleaks
Compiler: clang -march=native -O2 -fomit-frame-pointer -fwrapv -Qunused-arguments -fPIC -fPIE

bitslice_fft_256.c: bitslice_fft_256.c:87:25: error: always_inline function '_mm256_set_epi64x' requires target feature 'avx', but would be inlined into function 'bitslice_butterflies12_256' that is compiled without support for 'avx'
bitslice_fft_256.c: out[i][b] = _mm256_set_epi64x(-((in[0][b] >> reversal[4*i+3]) & 1),
bitslice_fft_256.c: ^
bitslice_fft_256.c: bitslice_fft_256.c:87:25: error: AVX vector return of type '__m256i' (vector of 4 'long long' values) without 'avx' enabled changes the ABI
bitslice_fft_256.c: bitslice_fft_256.c:99:22: error: '__builtin_ia32_pshufd256' needs target feature avx2
bitslice_fft_256.c: vb = _mm256_shuffle_epi32(tmp[b], _MM_SHUFFLE(3, 2, 3, 2));
bitslice_fft_256.c: ^
bitslice_fft_256.c: /usr/lib/llvm-14/lib/clang/14.0.0/include/avx2intrin.h:470:13: note: expanded from macro '_mm256_shuffle_epi32'
bitslice_fft_256.c: ((__m256i)__builtin_ia32_pshufd256((__v8si)(__m256i)(a), (int)(imm)))
bitslice_fft_256.c: ^
bitslice_fft_256.c: bitslice_fft_256.c:100:22: error: '__builtin_ia32_pslldqi256_byteshift' needs target feature avx2
bitslice_fft_256.c: va = _mm256_slli_si256(out[k][b], 8);
bitslice_fft_256.c: ^
bitslice_fft_256.c: /usr/lib/llvm-14/lib/clang/14.0.0/include/avx2intrin.h:497:13: note: expanded from macro '_mm256_slli_si256'
bitslice_fft_256.c: ((__m256i)__builtin_ia32_pslldqi256_byteshift((__v4di)(__m256i)(a), (int)(imm)))
bitslice_fft_256.c: ^
bitslice_fft_256.c: bitslice_fft_256.c:101:22: error: always_inline function '_mm256_xor_si256' requires target feature 'avx2', but would be inlined into function 'bitslice_butterflies12_256' that is compiled without support for 'avx2'
bitslice_fft_256.c: vb = _mm256_xor_si256(va, vb);
bitslice_fft_256.c: ^
bitslice_fft_256.c: bitslice_fft_256.c:101:22: error: AVX vector argument of type '__m256i' (vector of 4 'long long' values) without 'avx' enabled changes the ABI
bitslice_fft_256.c: bitslice_fft_256.c:102:29: error: always_inline function '_mm256_xor_si256' requires target feature 'avx2', but would be inlined into function 'bitslice_butterflies12_256' that is compiled without support for 'avx2'
bitslice_fft_256.c: out[k][b] = _mm256_xor_si256(out[k][b], vb);
bitslice_fft_256.c: ^
bitslice_fft_256.c: bitslice_fft_256.c:102:29: error: AVX vector argument of type '__m256i' (vector of 4 'long long' values) without 'avx' enabled changes the ABI
bitslice_fft_256.c: bitslice_fft_256.c:112:22: error: always_inline function '_mm256_set_epi64x' requires target feature 'avx', but would be inlined into function 'bitslice_butterflies12_256' that is compiled without support for 'avx'
bitslice_fft_256.c: ...

Number of similar (compiler,implementation) pairs: 5, namely:

Compiler	Implementations
clang -march=native -O2 -fomit-frame-pointer -fwrapv -Qunused-arguments -fPIC -fPIE	T:avx2
clang -march=native -O3 -fomit-frame-pointer -fwrapv -Qunused-arguments -fPIC -fPIE	T:avx2
clang -march=native -O -fomit-frame-pointer -fwrapv -Qunused-arguments -fPIC -fPIE	T:avx2
clang -march=native -Os -fomit-frame-pointer -fwrapv -Qunused-arguments -fPIC -fPIE	T:avx2
clang -mcpu=native -O3 -fomit-frame-pointer -fwrapv -Qunused-arguments -fPIC -fPIE	T:avx2

Compiler output

Implementation: T:avx2
Security model: timingleaks
Compiler: gcc -march=native -mtune=native -O2 -fomit-frame-pointer -fwrapv -fPIC -fPIE

bitslice_fft_256.c: bitslice_fft_256.c: In function 'bitslice_butterflies12_256':
bitslice_fft_256.c: bitslice_fft_256.c:87:23: warning: AVX vector return without AVX enabled changes the ABI [-Wpsabi]
bitslice_fft_256.c: 87 | out[i][b] = _mm256_set_epi64x(-((in[0][b] >> reversal[4*i+3]) & 1),
bitslice_fft_256.c: | ~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
bitslice_fft_256.c: 88 | -((in[0][b] >> reversal[4*i+2]) & 1),
bitslice_fft_256.c: | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
bitslice_fft_256.c: 89 | -((in[0][b] >> reversal[4*i+1]) & 1),
bitslice_fft_256.c: | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
bitslice_fft_256.c: 90 | -((in[0][b] >> reversal[4*i+0]) & 1));
bitslice_fft_256.c: | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
bitslice_fft_256.c: In file included from /usr/lib/gcc/x86_64-linux-gnu/11/include/immintrin.h:43,
bitslice_fft_256.c: from bitslice_fft_256.h:21,
bitslice_fft_256.c: from bitslice_fft_256.c:18:
bitslice_fft_256.c: /usr/lib/gcc/x86_64-linux-gnu/11/include/avxintrin.h:1303:1: error: inlining failed in call to 'always_inline' '_mm256_set_epi64x': target specific option mismatch
bitslice_fft_256.c: 1303 | _mm256_set_epi64x (long long __A, long long __B, long long __C,
bitslice_fft_256.c: | ^~~~~~~~~~~~~~~~~
bitslice_fft_256.c: bitslice_fft_256.c:87:25: note: called from here
bitslice_fft_256.c: 87 | out[i][b] = _mm256_set_epi64x(-((in[0][b] >> reversal[4*i+3]) & 1),
bitslice_fft_256.c: | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
bitslice_fft_256.c: 88 | -((in[0][b] >> reversal[4*i+2]) & 1),
bitslice_fft_256.c: | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
bitslice_fft_256.c: 89 | -((in[0][b] >> reversal[4*i+1]) & 1),
bitslice_fft_256.c: | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
bitslice_fft_256.c: 90 | -((in[0][b] >> reversal[4*i+0]) & 1));
bitslice_fft_256.c: | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
bitslice_fft_256.c: ...

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

Compiler	Implementations
gcc -march=native -mtune=native -O2 -fomit-frame-pointer -fwrapv -fPIC -fPIE	T:avx2
gcc -march=native -mtune=native -O3 -fomit-frame-pointer -fwrapv -fPIC -fPIE	T:avx2
gcc -march=native -mtune=native -O -fomit-frame-pointer -fwrapv -fPIC -fPIE	T:avx2
gcc -march=native -mtune=native -Os -fomit-frame-pointer -fwrapv -fPIC -fPIE	T:avx2

Compiler output

Implementation: T:sse2
Security model: timingleaks
Compiler: gcc -march=native -mtune=native -O2 -fomit-frame-pointer -fwrapv -fPIC -fPIE

nts_kem.c: In file included from /usr/lib/gcc/x86_64-linux-gnu/11/include/immintrin.h:39,
nts_kem.c: from ff.h:18,
nts_kem.c: from nts_kem.c:18:
nts_kem.c: nts_kem.c: In function 'create_matrix_G':
nts_kem.c: /usr/lib/gcc/x86_64-linux-gnu/11/include/smmintrin.h:455:1: error: inlining failed in call to 'always_inline' '_mm_extract_epi64': target specific option mismatch
nts_kem.c: 455 | _mm_extract_epi64 (__m128i __X, const int __N)
nts_kem.c: | ^~~~~~~~~~~~~~~~~
nts_kem.c: nts_kem.c:859:57: note: called from here
nts_kem.c: 859 | v[i >> 6] |= (((_mm_extract_epi64(vh[l >> 7][j], 1) & (1ULL << (l & 63))) >> (l & 63)) << (i & 63));
nts_kem.c: | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
nts_kem.c: In file included from /usr/lib/gcc/x86_64-linux-gnu/11/include/immintrin.h:39,
nts_kem.c: from ff.h:18,
nts_kem.c: from nts_kem.c:18:
nts_kem.c: /usr/lib/gcc/x86_64-linux-gnu/11/include/smmintrin.h:455:1: error: inlining failed in call to 'always_inline' '_mm_extract_epi64': target specific option mismatch
nts_kem.c: 455 | _mm_extract_epi64 (__m128i __X, const int __N)
nts_kem.c: | ^~~~~~~~~~~~~~~~~
nts_kem.c: nts_kem.c:862:57: note: called from here
nts_kem.c: 862 | v[i >> 6] |= (((_mm_extract_epi64(vh[l >> 7][j], 0) & (1ULL << (l & 63))) >> (l & 63)) << (i & 63));
nts_kem.c: | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
nts_kem.c: In file included from /usr/lib/gcc/x86_64-linux-gnu/11/include/immintrin.h:39,
nts_kem.c: from ff.h:18,
nts_kem.c: from nts_kem.c:18:
nts_kem.c: /usr/lib/gcc/x86_64-linux-gnu/11/include/smmintrin.h:455:1: error: inlining failed in call to 'always_inline' '_mm_extract_epi64': target specific option mismatch
nts_kem.c: 455 | _mm_extract_epi64 (__m128i __X, const int __N)
nts_kem.c: | ^~~~~~~~~~~~~~~~~
nts_kem.c: ...

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

Compiler	Implementations
gcc -march=native -mtune=native -O2 -fomit-frame-pointer -fwrapv -fPIC -fPIE	T:sse2
gcc -march=native -mtune=native -O3 -fomit-frame-pointer -fwrapv -fPIC -fPIE	T:sse2
gcc -march=native -mtune=native -O -fomit-frame-pointer -fwrapv -fPIC -fPIE	T:sse2
gcc -march=native -mtune=native -Os -fomit-frame-pointer -fwrapv -fPIC -fPIE	T:sse2