Implementation notes: amd64, h3neo, crypto_kem/firesaber2
Computer: h3neo
Microarchitecture: amd64; K10 45nm (100f63)
Architecture: amd64
CPU ID: AuthenticAMD-00100f63-078bfbff
SUPERCOP version: 20240625
Operation: crypto_kem
Primitive: firesaber2
Time | Object size | Test size | Implementation | Compiler | Benchmark date | SUPERCOP version |
1340290 | 115286 0 0 | 134178 860 1728 | T:ref | clang_-march=native_-O3_-fwrapv_-Qunused-arguments_-fPIC_-fPIE_-gdwarf-4_-Wall | 20240716 | 20240625 |
1341094 | 45734 0 0 | 63738 860 1728 | T:ref | clang_-march=native_-O2_-fwrapv_-Qunused-arguments_-fPIC_-fPIE_-gdwarf-4_-Wall | 20240716 | 20240625 |
1353967 | 121711 0 0 | 141050 860 1728 | T:ref | clang_-mcpu=native_-O3_-fwrapv_-Qunused-arguments_-fPIC_-fPIE_-gdwarf-4_-Wall | 20240716 | 20240625 |
1391165 | 53733 0 0 | 73900 820 1760 | T:ref | gcc_-march=native_-mtune=native_-O3_-fwrapv_-fPIC_-fPIE_-gdwarf-4_-Wall | 20240716 | 20240625 |
2801449 | 17323 0 0 | 34706 860 1728 | T:ref | clang_-march=native_-O_-fwrapv_-Qunused-arguments_-fPIC_-fPIE_-gdwarf-4_-Wall | 20240716 | 20240625 |
2877952 | 13242 0 0 | 29868 852 1728 | T:ref | clang_-march=native_-Os_-fwrapv_-Qunused-arguments_-fPIC_-fPIE_-gdwarf-4_-Wall | 20240716 | 20240625 |
3343285 | 14104 0 0 | 31940 820 1760 | T:ref | gcc_-march=native_-mtune=native_-O_-fwrapv_-fPIC_-fPIE_-gdwarf-4_-Wall | 20240716 | 20240625 |
3442630 | 16623 0 0 | 35484 820 1760 | T:ref | gcc_-march=native_-mtune=native_-O2_-fwrapv_-fPIC_-fPIE_-gdwarf-4_-Wall | 20240716 | 20240625 |
3527129 | 13043 0 0 | 29548 812 1728 | T:ref | gcc_-march=native_-mtune=native_-Os_-fwrapv_-fPIC_-fPIE_-gdwarf-4_-Wall | 20240716 | 20240625 |
Compiler output
SABER_indcpa.c: In file included from SABER_indcpa.c:9:
SABER_indcpa.c: In file included from ././polymul/toom-cook_4way.c:6:
SABER_indcpa.c: ././polymul/scm_avx.c:40:13: warning: AVX vector argument of type '__m256i' (vector of 4 'long long' values) without 'avx' enabled changes the ABI [-Wpsabi]
SABER_indcpa.c: c_avx[0] = mul_add(a0, b0, c_avx[0]);
SABER_indcpa.c: ^
SABER_indcpa.c: ././polymul/scm_avx.c:43:9: error: always_inline function '_mm256_mullo_epi16' requires target feature 'avx2', but would be inlined into function 'schoolbook_avx_new3_acc' that is compiled without support for 'avx2'
SABER_indcpa.c: temp = _mm256_mullo_epi16 (a0, b1);
SABER_indcpa.c: ^
SABER_indcpa.c: ././polymul/scm_avx.c:43:9: error: AVX vector argument of type '__m256i' (vector of 4 'long long' values) without 'avx' enabled changes the ABI
SABER_indcpa.c: ././polymul/scm_avx.c:44:7: warning: AVX vector argument of type '__m256i' (vector of 4 'long long' values) without 'avx' enabled changes the ABI [-Wpsabi]
SABER_indcpa.c: temp=mul_add(a1, b0, temp);
SABER_indcpa.c: ^
SABER_indcpa.c: ././polymul/scm_avx.c:45:13: error: always_inline function '_mm256_add_epi16' requires target feature 'avx2', but would be inlined into function 'schoolbook_avx_new3_acc' that is compiled without support for 'avx2'
SABER_indcpa.c: c_avx[1] = _mm256_add_epi16(temp, c_avx[1]);
SABER_indcpa.c: ^
SABER_indcpa.c: ././polymul/scm_avx.c:45:13: error: AVX vector argument of type '__m256i' (vector of 4 'long long' values) without 'avx' enabled changes the ABI
SABER_indcpa.c: ././polymul/scm_avx.c:48:9: error: always_inline function '_mm256_mullo_epi16' requires target feature 'avx2', but would be inlined into function 'schoolbook_avx_new3_acc' that is compiled without support for 'avx2'
SABER_indcpa.c: temp = _mm256_mullo_epi16 (a0, b2);
SABER_indcpa.c: ^
SABER_indcpa.c: ././polymul/scm_avx.c:48:9: error: AVX vector argument of type '__m256i' (vector of 4 'long long' values) without 'avx' enabled changes the ABI
SABER_indcpa.c: ././polymul/scm_avx.c:49:9: warning: AVX vector argument of type '__m256i' (vector of 4 'long long' values) without 'avx' enabled changes the ABI [-Wpsabi]
SABER_indcpa.c: temp = mul_add(a1, b1, temp);
SABER_indcpa.c: ^
SABER_indcpa.c: ././polymul/scm_avx.c:50:7: warning: AVX vector argument of type '__m256i' (vector of 4 'long long' values) without 'avx' enabled changes the ABI [-Wpsabi]
SABER_indcpa.c: temp=mul_add(a2, b0, temp);
SABER_indcpa.c: ...
Number of similar (implementation,compiler) pairs: 5, namely:
Implementation | Compiler |
T:avx2 | clang -march=native -O2 -fwrapv -Qunused-arguments -fPIC -fPIE -gdwarf-4 -Wall (Ubuntu_Clang_14.0.0) |
T:avx2 | clang -march=native -O3 -fwrapv -Qunused-arguments -fPIC -fPIE -gdwarf-4 -Wall (Ubuntu_Clang_14.0.0) |
T:avx2 | clang -march=native -O -fwrapv -Qunused-arguments -fPIC -fPIE -gdwarf-4 -Wall (Ubuntu_Clang_14.0.0) |
T:avx2 | clang -march=native -Os -fwrapv -Qunused-arguments -fPIC -fPIE -gdwarf-4 -Wall (Ubuntu_Clang_14.0.0) |
T:avx2 | clang -mcpu=native -O3 -fwrapv -Qunused-arguments -fPIC -fPIE -gdwarf-4 -Wall (Ubuntu_Clang_14.0.0) |
Compiler output
SABER_indcpa.c: In file included from ./polymul/toom-cook_4way.c:6,
SABER_indcpa.c: from SABER_indcpa.c:9:
SABER_indcpa.c: ./polymul/scm_avx.c: In function 'mul_add':
SABER_indcpa.c: ./polymul/scm_avx.c:6:1: warning: AVX vector return without AVX enabled changes the ABI [-Wpsabi]
SABER_indcpa.c: 6 | inline __m256i mul_add(__m256i a, __m256i b, __m256i c) {
SABER_indcpa.c: | ^~~~~~
SABER_indcpa.c: SABER_indcpa.c: In function 'indcpa_kem_enc':
SABER_indcpa.c: SABER_indcpa.c:269:26: warning: unused variable 'CLOCK2' [-Wunused-variable]
SABER_indcpa.c: 269 | uint64_t CLOCK1, CLOCK2;
SABER_indcpa.c: | ^~~~~~
SABER_indcpa.c: SABER_indcpa.c:269:18: warning: unused variable 'CLOCK1' [-Wunused-variable]
SABER_indcpa.c: 269 | uint64_t CLOCK1, CLOCK2;
SABER_indcpa.c: | ^~~~~~
SABER_indcpa.c: SABER_indcpa.c: In function 'indcpa_kem_dec':
SABER_indcpa.c: SABER_indcpa.c:436:26: warning: unused variable 'CLOCK2' [-Wunused-variable]
SABER_indcpa.c: 436 | uint64_t CLOCK1, CLOCK2;
SABER_indcpa.c: | ^~~~~~
SABER_indcpa.c: SABER_indcpa.c:436:18: warning: unused variable 'CLOCK1' [-Wunused-variable]
SABER_indcpa.c: 436 | uint64_t CLOCK1, CLOCK2;
SABER_indcpa.c: | ^~~~~~
SABER_indcpa.c: In file included from ./polymul/toom-cook_4way.c:6,
SABER_indcpa.c: from SABER_indcpa.c:9:
SABER_indcpa.c: ./polymul/scm_avx.c: In function 'mul_add':
SABER_indcpa.c: ./polymul/scm_avx.c:6:16: note: the ABI for passing parameters with 32-byte alignment has changed in GCC 4.6
SABER_indcpa.c: 6 | inline __m256i mul_add(__m256i a, __m256i b, __m256i c) {
SABER_indcpa.c: ...
Number of similar (implementation,compiler) pairs: 4, namely:
Implementation | Compiler |
T:avx2 | gcc -march=native -mtune=native -O2 -fwrapv -fPIC -fPIE -gdwarf-4 -Wall (11.4.0) |
T:avx2 | gcc -march=native -mtune=native -O3 -fwrapv -fPIC -fPIE -gdwarf-4 -Wall (11.4.0) |
T:avx2 | gcc -march=native -mtune=native -O -fwrapv -fPIC -fPIE -gdwarf-4 -Wall (11.4.0) |
T:avx2 | gcc -march=native -mtune=native -Os -fwrapv -fPIC -fPIE -gdwarf-4 -Wall (11.4.0) |
Compiler output
poly.c: poly.c:31:26: error: always_inline function '_mm256_set1_epi16' requires target feature 'avx', but would be inlined into function 'nttmul_poly_crt' that is compiled without support for 'avx'
poly.c: const __m256i u_pinv = _mm256_set1_epi16(CRT_U_PINV);
poly.c: ^
poly.c: poly.c:31:26: error: AVX vector return of type '__m256i' (vector of 4 'long long' values) without 'avx' enabled changes the ABI
poly.c: poly.c:32:21: error: always_inline function '_mm256_set1_epi16' requires target feature 'avx', but would be inlined into function 'nttmul_poly_crt' that is compiled without support for 'avx'
poly.c: const __m256i u = _mm256_set1_epi16(CRT_U);
poly.c: ^
poly.c: poly.c:32:21: error: AVX vector return of type '__m256i' (vector of 4 'long long' values) without 'avx' enabled changes the ABI
poly.c: poly.c:33:22: error: always_inline function '_mm256_load_si256' requires target feature 'avx', but would be inlined into function 'nttmul_poly_crt' that is compiled without support for 'avx'
poly.c: const __m256i p0 = _mm256_load_si256((__m256i *)&PDATA0[_16XP]);
poly.c: ^
poly.c: poly.c:33:22: error: AVX vector return of type '__m256i' (vector of 4 'long long' values) without 'avx' enabled changes the ABI
poly.c: poly.c:34:22: error: always_inline function '_mm256_load_si256' requires target feature 'avx', but would be inlined into function 'nttmul_poly_crt' that is compiled without support for 'avx'
poly.c: const __m256i p1 = _mm256_load_si256((__m256i *)&PDATA1[_16XP]);
poly.c: ^
poly.c: poly.c:34:22: error: AVX vector return of type '__m256i' (vector of 4 'long long' values) without 'avx' enabled changes the ABI
poly.c: poly.c:35:23: error: always_inline function '_mm256_set1_epi16' requires target feature 'avx', but would be inlined into function 'nttmul_poly_crt' that is compiled without support for 'avx'
poly.c: const __m256i mod = _mm256_set1_epi16(KEM_Q-1);
poly.c: ^
poly.c: poly.c:35:23: error: AVX vector return of type '__m256i' (vector of 4 'long long' values) without 'avx' enabled changes the ABI
poly.c: poly.c:36:30: error: always_inline function '_mm256_load_si256' requires target feature 'avx', but would be inlined into function 'nttmul_poly_crt' that is compiled without support for 'avx'
poly.c: const __m256i mont0_pinv = _mm256_load_si256((__m256i *)&PDATA0[_16XMONT_PINV]);
poly.c: ^
poly.c: poly.c:36:30: error: AVX vector return of type '__m256i' (vector of 4 'long long' values) without 'avx' enabled changes the ABI
poly.c: poly.c:37:25: error: always_inline function '_mm256_load_si256' requires target feature 'avx', but would be inlined into function 'nttmul_poly_crt' that is compiled without support for 'avx'
poly.c: ...
Number of similar (implementation,compiler) pairs: 5, namely:
Implementation | Compiler |
T:avx2_nttmul | clang -march=native -O2 -fwrapv -Qunused-arguments -fPIC -fPIE -gdwarf-4 -Wall (Ubuntu_Clang_14.0.0) |
T:avx2_nttmul | clang -march=native -O3 -fwrapv -Qunused-arguments -fPIC -fPIE -gdwarf-4 -Wall (Ubuntu_Clang_14.0.0) |
T:avx2_nttmul | clang -march=native -O -fwrapv -Qunused-arguments -fPIC -fPIE -gdwarf-4 -Wall (Ubuntu_Clang_14.0.0) |
T:avx2_nttmul | clang -march=native -Os -fwrapv -Qunused-arguments -fPIC -fPIE -gdwarf-4 -Wall (Ubuntu_Clang_14.0.0) |
T:avx2_nttmul | clang -mcpu=native -O3 -fwrapv -Qunused-arguments -fPIC -fPIE -gdwarf-4 -Wall (Ubuntu_Clang_14.0.0) |
Compiler output
poly.c: poly.c: In function 'mulmod':
poly.c: poly.c:7:1: warning: AVX vector return without AVX enabled changes the ABI [-Wpsabi]
poly.c: 7 | static inline __m256i mulmod(const __m256i a, const __m256i b_pinv, const __m256i b, const __m256i p) {
poly.c: | ^~~~~~
poly.c: poly.c:7:23: note: the ABI for passing parameters with 32-byte alignment has changed in GCC 4.6
poly.c: 7 | static inline __m256i mulmod(const __m256i a, const __m256i b_pinv, const __m256i b, const __m256i p) {
poly.c: | ^~~~~~
poly.c: In file included from /usr/lib/gcc/x86_64-linux-gnu/11/include/immintrin.h:47,
poly.c: from poly.c:3:
poly.c: /usr/lib/gcc/x86_64-linux-gnu/11/include/avx2intrin.h:808:1: error: inlining failed in call to 'always_inline' '_mm256_sub_epi16': target specific option mismatch
poly.c: 808 | _mm256_sub_epi16 (__m256i __A, __m256i __B)
poly.c: | ^~~~~~~~~~~~~~~~
poly.c: poly.c:12:7: note: called from here
poly.c: 12 | t = _mm256_sub_epi16(u,t);
poly.c: | ^~~~~~~~~~~~~~~~~~~~~
poly.c: In file included from /usr/lib/gcc/x86_64-linux-gnu/11/include/immintrin.h:47,
poly.c: from poly.c:3:
poly.c: /usr/lib/gcc/x86_64-linux-gnu/11/include/avx2intrin.h:546:1: error: inlining failed in call to 'always_inline' '_mm256_mulhi_epi16': target specific option mismatch
poly.c: 546 | _mm256_mulhi_epi16 (__m256i __A, __m256i __B)
poly.c: | ^~~~~~~~~~~~~~~~~~
poly.c: poly.c:11:7: note: called from here
poly.c: 11 | t = _mm256_mulhi_epi16(t,p);
poly.c: | ^~~~~~~~~~~~~~~~~~~~~~~
poly.c: In file included from /usr/lib/gcc/x86_64-linux-gnu/11/include/immintrin.h:47,
poly.c: from poly.c:3:
poly.c: ...
Number of similar (implementation,compiler) pairs: 4, namely:
Implementation | Compiler |
T:avx2_nttmul | gcc -march=native -mtune=native -O2 -fwrapv -fPIC -fPIE -gdwarf-4 -Wall (11.4.0) |
T:avx2_nttmul | gcc -march=native -mtune=native -O3 -fwrapv -fPIC -fPIE -gdwarf-4 -Wall (11.4.0) |
T:avx2_nttmul | gcc -march=native -mtune=native -O -fwrapv -fPIC -fPIE -gdwarf-4 -Wall (11.4.0) |
T:avx2_nttmul | gcc -march=native -mtune=native -Os -fwrapv -fPIC -fPIE -gdwarf-4 -Wall (11.4.0) |