From d07f54b967650b71e54f599e7f97a2a03d414990 Mon Sep 17 00:00:00 2001 From: "Wladimir J. van der Laan" Date: Sun, 24 Jun 2018 15:30:07 +0200 Subject: [PATCH] Merge #13471: For AVX2 code, also check for AVX, XSAVE, and OS support 32d153fa360f73b4999701b97d55b12318fd2659 For AVX2 code, also check for AVX, XSAVE, and OS support (Pieter Wuille) Pull request description: Fixes #12903. Tree-SHA512: 01e71efb5d3a43c49a145a5b1dc4fe7d0a491e1e78479e7df830a2aaac57c3dcfc316e28984c695206c76f93b68e4350fc037ca36756ca579b7070e39c835da2 --- src/crypto/sha256.cpp | 21 +++++++++++++++++---- 1 file changed, 17 insertions(+), 4 deletions(-) diff --git a/src/crypto/sha256.cpp b/src/crypto/sha256.cpp index 2516064e20..2bf624042c 100644 --- a/src/crypto/sha256.cpp +++ b/src/crypto/sha256.cpp @@ -536,6 +536,14 @@ void inline cpuid(uint32_t leaf, uint32_t subleaf, uint32_t& a, uint32_t& b, uin { __asm__ ("cpuid" : "=a"(a), "=b"(b), "=c"(c), "=d"(d) : "0"(leaf), "2"(subleaf)); } + +/** Check whether the OS has enabled AVX registers. */ +bool AVXEnabled() +{ + uint32_t a, d; + __asm__("xgetbv" : "=a"(a), "=d"(d) : "c"(0)); + return (a & 6) == 6; +} #endif } // namespace @@ -544,6 +552,7 @@ std::string SHA256AutoDetect() { std::string ret = "standard"; #if defined(USE_ASM) && (defined(__x86_64__) || defined(__amd64__)) + (void)AVXEnabled; // Silence unused warning (in case ENABLE_AVX2 is not defined) uint32_t eax, ebx, ecx, edx; cpuid(1, 0, eax, ebx, ecx, edx); if ((ecx >> 19) & 1) { @@ -553,10 +562,14 @@ std::string SHA256AutoDetect() TransformD64_4way = sha256d64_sse41::Transform_4way; ret = "sse4(1way+4way)"; #if defined(ENABLE_AVX2) && !defined(BUILD_BITCOIN_INTERNAL) - cpuid(7, 0, eax, ebx, ecx, edx); - if ((ebx >> 5) & 1) { - TransformD64_8way = sha256d64_avx2::Transform_8way; - ret += ",avx2(8way)"; + if (((ecx >> 27) & 1) && ((ecx >> 28) & 1)) { // XSAVE and AVX + cpuid(7, 0, eax, ebx, ecx, edx); + if ((ebx >> 5) & 1) { // AVX2 flag + if (AVXEnabled()) { // OS has enabled AVX registers + TransformD64_8way = sha256d64_avx2::Transform_8way; + ret += ",avx2(8way)"; + } + } } #endif #else