arne
/
cryptopp
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Disable BMI2 code paths in Integer class (GH #850)

pull/853/head
Jeffrey Walton 2019-06-01 00:14:19 -04:00
parent 8e27c6b3fa
commit 1a0d7c516d
No known key found for this signature in database
GPG Key ID: B36AB348921B1838
1 changed files with 7 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
integer.cpp
View File

@ -45,6 +45,12 @@
// thousands of times during the life of a program. Each load produces a // thousands of times during the life of a program. Each load produces a
// memory leak and they can add up quickly. If they library is being used in // memory leak and they can add up quickly. If they library is being used in
// Java or .Net then Singleton must be avoided at all costs. // Java or .Net then Singleton must be avoided at all costs.
//
// The code below has a path cut-in for BMI2 using mulx and adcx instructions.
// There was a modest speedup of approximately 0.03 ms in Integer operations.
// We had to disable BMI2 for the moment because some OS X machines were
// advertising BMI/BMI2 support but caused SIGILL's at runtime. Also see
// https://github.com/weidai11/cryptopp/issues/850.
#include "pch.h" #include "pch.h"
#include "config.h" #include "config.h"
@ -212,7 +218,7 @@ static word AtomicInverseModPower2(word A)
#if defined(__SUNPRO_CC) && __SUNPRO_CC < 0x5100 #if defined(__SUNPRO_CC) && __SUNPRO_CC < 0x5100
// Sun Studio's gcc-style inline assembly is heavily bugged as of version 5.9 Patch 124864-09 2008/12/16, but this one works // Sun Studio's gcc-style inline assembly is heavily bugged as of version 5.9 Patch 124864-09 2008/12/16, but this one works
#define MultiplyWordsLoHi(p0, p1, a, b) asm ("mulq %3" : "=a"(p0), "=d"(p1) : "a"(a), "r"(b) : "cc"); #define MultiplyWordsLoHi(p0, p1, a, b) asm ("mulq %3" : "=a"(p0), "=d"(p1) : "a"(a), "r"(b) : "cc");
#elif defined(__BMI2__) #elif defined(__BMI2__) && 0
#define MultiplyWordsLoHi(p0, p1, a, b) asm ("mulxq %3, %0, %1" : "=r"(p0), "=r"(p1) : "d"(a), "r"(b)); #define MultiplyWordsLoHi(p0, p1, a, b) asm ("mulxq %3, %0, %1" : "=r"(p0), "=r"(p1) : "d"(a), "r"(b));
#define MulAcc(c, d, a, b) asm ("mulxq %6, %3, %4; addq %3, %0; adcxq %4, %1; adcxq %7, %2;" : "+&r"(c), "+&r"(d##0), "+&r"(d##1), "=&r"(p0), "=&r"(p1) : "d"(a), "r"(b), "r"(W64LIT(0)) : "cc"); #define MulAcc(c, d, a, b) asm ("mulxq %6, %3, %4; addq %3, %0; adcxq %4, %1; adcxq %7, %2;" : "+&r"(c), "+&r"(d##0), "+&r"(d##1), "=&r"(p0), "=&r"(p1) : "d"(a), "r"(b), "r"(W64LIT(0)) : "cc");
#define Double3Words(c, d) asm ("addq %0, %0; adcxq %1, %1; adcxq %2, %2;" : "+r"(c), "+r"(d##0), "+r"(d##1) : : "cc"); #define Double3Words(c, d) asm ("addq %0, %0; adcxq %1, %1; adcxq %2, %2;" : "+r"(c), "+r"(d##0), "+r"(d##1) : : "cc");