From 4ebe29b2e3c78bed8fb29d644b920f0f1250f3f7 Mon Sep 17 00:00:00 2001 From: Scr3amer Date: Sat, 2 Sep 2023 09:22:20 -0400 Subject: [PATCH] Do not redefine _mm_roti_epi64 if xop target feature is enabled. (#1305) Undefine compiler macro if present and XOP not enable to avoid preprocessor warnings due to macro redefinition --- .../blake2b/ref/blake2b-compress-sse41.h | 29 +++++++++++-------- .../blake2b/ref/blake2b-compress-ssse3.h | 29 +++++++++++-------- .../crypto_pwhash/argon2/blamka-round-ssse3.h | 29 +++++++++++-------- 3 files changed, 51 insertions(+), 36 deletions(-) diff --git a/src/libsodium/crypto_generichash/blake2b/ref/blake2b-compress-sse41.h b/src/libsodium/crypto_generichash/blake2b/ref/blake2b-compress-sse41.h index ac78e5bb..07c60493 100644 --- a/src/libsodium/crypto_generichash/blake2b/ref/blake2b-compress-sse41.h +++ b/src/libsodium/crypto_generichash/blake2b/ref/blake2b-compress-sse41.h @@ -5,18 +5,23 @@ #define LOADU(p) _mm_loadu_si128((const __m128i *) (const void *) (p)) #define STOREU(p, r) _mm_storeu_si128((__m128i *) (void *) (p), r) -#define _mm_roti_epi64(x, c) \ - (-(c) == 32) \ - ? _mm_shuffle_epi32((x), _MM_SHUFFLE(2, 3, 0, 1)) \ - : (-(c) == 24) \ - ? _mm_shuffle_epi8((x), r24) \ - : (-(c) == 16) \ - ? _mm_shuffle_epi8((x), r16) \ - : (-(c) == 63) \ - ? _mm_xor_si128(_mm_srli_epi64((x), -(c)), \ - _mm_add_epi64((x), (x))) \ - : _mm_xor_si128(_mm_srli_epi64((x), -(c)), \ - _mm_slli_epi64((x), 64 - (-(c)))) +#ifndef __XOP__ // XOP feature set enabled implies _mm_roti_epi64 is existing AFAIK so no need to define the libsodium replacement. +# ifdef _mm_roti_epi64 // If we have the function macro but not the feature set macro it will fail compiling. +# undef _mm_roti_epi64 // So we hide the existing one and use the one provided by libsodium. Hiding avoids warnings for macro redefinitions. +# endif +# define _mm_roti_epi64(x, c) \ + (-(c) == 32) \ + ? _mm_shuffle_epi32((x), _MM_SHUFFLE(2, 3, 0, 1)) \ + : (-(c) == 24) \ + ? _mm_shuffle_epi8((x), r24) \ + : (-(c) == 16) \ + ? _mm_shuffle_epi8((x), r16) \ + : (-(c) == 63) \ + ? _mm_xor_si128(_mm_srli_epi64((x), -(c)), \ + _mm_add_epi64((x), (x))) \ + : _mm_xor_si128(_mm_srli_epi64((x), -(c)), \ + _mm_slli_epi64((x), 64 - (-(c)))) +#endif #define G1(row1l, row2l, row3l, row4l, row1h, row2h, row3h, row4h, b0, b1) \ row1l = _mm_add_epi64(_mm_add_epi64(row1l, b0), row2l); \ diff --git a/src/libsodium/crypto_generichash/blake2b/ref/blake2b-compress-ssse3.h b/src/libsodium/crypto_generichash/blake2b/ref/blake2b-compress-ssse3.h index 9a7164fe..bfb60862 100644 --- a/src/libsodium/crypto_generichash/blake2b/ref/blake2b-compress-ssse3.h +++ b/src/libsodium/crypto_generichash/blake2b/ref/blake2b-compress-ssse3.h @@ -5,18 +5,23 @@ #define LOADU(p) _mm_loadu_si128((const __m128i *) (const void *) (p)) #define STOREU(p, r) _mm_storeu_si128((__m128i *) (void *) (p), r) -#define _mm_roti_epi64(x, c) \ - (-(c) == 32) \ - ? _mm_shuffle_epi32((x), _MM_SHUFFLE(2, 3, 0, 1)) \ - : (-(c) == 24) \ - ? _mm_shuffle_epi8((x), r24) \ - : (-(c) == 16) \ - ? _mm_shuffle_epi8((x), r16) \ - : (-(c) == 63) \ - ? _mm_xor_si128(_mm_srli_epi64((x), -(c)), \ - _mm_add_epi64((x), (x))) \ - : _mm_xor_si128(_mm_srli_epi64((x), -(c)), \ - _mm_slli_epi64((x), 64 - (-(c)))) +#ifndef __XOP__ // XOP feature set enabled implies _mm_roti_epi64 is existing AFAIK so no need to define the libsodium replacement. +# ifdef _mm_roti_epi64 // If we have the function macro but not the feature set macro it will fail compiling. +# undef _mm_roti_epi64 // So we hide the existing one and use the one provided by libsodium. Hiding avoids warnings for macro redefinitions. +# endif +# define _mm_roti_epi64(x, c) \ + (-(c) == 32) \ + ? _mm_shuffle_epi32((x), _MM_SHUFFLE(2, 3, 0, 1)) \ + : (-(c) == 24) \ + ? _mm_shuffle_epi8((x), r24) \ + : (-(c) == 16) \ + ? _mm_shuffle_epi8((x), r16) \ + : (-(c) == 63) \ + ? _mm_xor_si128(_mm_srli_epi64((x), -(c)), \ + _mm_add_epi64((x), (x))) \ + : _mm_xor_si128(_mm_srli_epi64((x), -(c)), \ + _mm_slli_epi64((x), 64 - (-(c)))) +#endif #define G1(row1l, row2l, row3l, row4l, row1h, row2h, row3h, row4h, b0, b1) \ row1l = _mm_add_epi64(_mm_add_epi64(row1l, b0), row2l); \ diff --git a/src/libsodium/crypto_pwhash/argon2/blamka-round-ssse3.h b/src/libsodium/crypto_pwhash/argon2/blamka-round-ssse3.h index 98a47b93..18230637 100644 --- a/src/libsodium/crypto_pwhash/argon2/blamka-round-ssse3.h +++ b/src/libsodium/crypto_pwhash/argon2/blamka-round-ssse3.h @@ -8,18 +8,23 @@ (_mm_setr_epi8(2, 3, 4, 5, 6, 7, 0, 1, 10, 11, 12, 13, 14, 15, 8, 9)) #define r24 \ (_mm_setr_epi8(3, 4, 5, 6, 7, 0, 1, 2, 11, 12, 13, 14, 15, 8, 9, 10)) -#define _mm_roti_epi64(x, c) \ - (-(c) == 32) \ - ? _mm_shuffle_epi32((x), _MM_SHUFFLE(2, 3, 0, 1)) \ - : (-(c) == 24) \ - ? _mm_shuffle_epi8((x), r24) \ - : (-(c) == 16) \ - ? _mm_shuffle_epi8((x), r16) \ - : (-(c) == 63) \ - ? _mm_xor_si128(_mm_srli_epi64((x), -(c)), \ - _mm_add_epi64((x), (x))) \ - : _mm_xor_si128(_mm_srli_epi64((x), -(c)), \ - _mm_slli_epi64((x), 64 - (-(c)))) +#ifndef __XOP__ // XOP feature set enabled implies _mm_roti_epi64 is existing AFAIK so no need to define the libsodium replacement. +# ifdef _mm_roti_epi64 // If we have the function macro but not the feature set macro it will fail compiling. +# undef _mm_roti_epi64 // So we hide the existing one and use the one provided by libsodium. Hiding avoids warnings for macro redefinitions. +# endif +# define _mm_roti_epi64(x, c) \ + (-(c) == 32) \ + ? _mm_shuffle_epi32((x), _MM_SHUFFLE(2, 3, 0, 1)) \ + : (-(c) == 24) \ + ? _mm_shuffle_epi8((x), r24) \ + : (-(c) == 16) \ + ? _mm_shuffle_epi8((x), r16) \ + : (-(c) == 63) \ + ? _mm_xor_si128(_mm_srli_epi64((x), -(c)), \ + _mm_add_epi64((x), (x))) \ + : _mm_xor_si128(_mm_srli_epi64((x), -(c)), \ + _mm_slli_epi64((x), 64 - (-(c)))) +#endif static inline __m128i fBlaMka(__m128i x, __m128i y)