Whitespace check-in
We also simplified the CPU_ProbeNEON logic a bit to a vmov.u32 and vshl.u32.pull/853/head
Jeffrey Walton
parent
40251d9b7f
commit
e8603143dc
199
neon_simd.cpp
199
neon_simd.cpp
|
|
@ -6,6 +6,7 @@
|
|||
// ARMv8a NEON instructions. A separate source file is needed
|
||||
// because additional CXXFLAGS are required to enable the
|
||||
// appropriate instructions sets in some build configurations.
|
||||
// For Linux and Unix additional flags are not required.
|
||||
|
||||
#include "pch.h"
|
||||
#include "config.h"
|
||||
|
|
@ -39,55 +40,55 @@ NAMESPACE_BEGIN(CryptoPP)
|
|||
extern "C" {
|
||||
typedef void (*SigHandler)(int);
|
||||
|
||||
static jmp_buf s_jmpSIGILL;
|
||||
static void SigIllHandler(int)
|
||||
{
|
||||
longjmp(s_jmpSIGILL, 1);
|
||||
}
|
||||
static jmp_buf s_jmpSIGILL;
|
||||
static void SigIllHandler(int)
|
||||
{
|
||||
longjmp(s_jmpSIGILL, 1);
|
||||
}
|
||||
}
|
||||
#endif // Not CRYPTOPP_MS_STYLE_INLINE_ASSEMBLY
|
||||
|
||||
bool CPU_ProbeARMv7()
|
||||
{
|
||||
#if defined(CRYPTOPP_NO_CPU_FEATURE_PROBES)
|
||||
return false;
|
||||
return false;
|
||||
#elif CRYPTOPP_BOOL_ARM32
|
||||
# if defined(CRYPTOPP_MS_STYLE_INLINE_ASSEMBLY)
|
||||
volatile bool result = true;
|
||||
__try
|
||||
{
|
||||
// Modern MS hardware is ARMv7
|
||||
result = true;
|
||||
}
|
||||
__except (EXCEPTION_EXECUTE_HANDLER)
|
||||
{
|
||||
return false;
|
||||
}
|
||||
return result;
|
||||
# elif defined(__GNUC__) || defined(__clang__)
|
||||
// longjmp and clobber warnings. Volatile is required.
|
||||
// http://github.com/weidai11/cryptopp/issues/24 and http://stackoverflow.com/q/7721854
|
||||
volatile bool result = true;
|
||||
volatile bool result = true;
|
||||
__try
|
||||
{
|
||||
// Modern MS hardware is ARMv7
|
||||
result = true;
|
||||
}
|
||||
__except (EXCEPTION_EXECUTE_HANDLER)
|
||||
{
|
||||
return false;
|
||||
}
|
||||
return result;
|
||||
# else
|
||||
// longjmp and clobber warnings. Volatile is required.
|
||||
// http://github.com/weidai11/cryptopp/issues/24 and http://stackoverflow.com/q/7721854
|
||||
volatile bool result = true;
|
||||
|
||||
volatile SigHandler oldHandler = signal(SIGILL, SigIllHandler);
|
||||
if (oldHandler == SIG_ERR)
|
||||
return false;
|
||||
volatile SigHandler oldHandler = signal(SIGILL, SigIllHandler);
|
||||
if (oldHandler == SIG_ERR)
|
||||
return false;
|
||||
|
||||
volatile sigset_t oldMask;
|
||||
if (sigprocmask(0, NULLPTR, (sigset_t*)&oldMask))
|
||||
return false;
|
||||
volatile sigset_t oldMask;
|
||||
if (sigprocmask(0, NULLPTR, (sigset_t*)&oldMask))
|
||||
return false;
|
||||
|
||||
if (setjmp(s_jmpSIGILL))
|
||||
result = false;
|
||||
else
|
||||
{
|
||||
if (setjmp(s_jmpSIGILL))
|
||||
result = false;
|
||||
else
|
||||
{
|
||||
|
||||
#if 0
|
||||
// ARMv7 added movt and movw
|
||||
int a;
|
||||
asm volatile("movw %0,%1 \n"
|
||||
"movt %0,%1 \n"
|
||||
: "=r"(a) : "i"(0x1234));
|
||||
// ARMv7 added movt and movw
|
||||
int a;
|
||||
asm volatile("movw %0,%1 \n"
|
||||
"movt %0,%1 \n"
|
||||
: "=r"(a) : "i"(0x1234));
|
||||
|
||||
00000010 <_Z5test2v>: // ARM
|
||||
10: e3010234 movw r0, #4660 ; 0x1234
|
||||
|
|
@ -100,96 +101,80 @@ bool CPU_ProbeARMv7()
|
|||
24: e12fff1e bx lr
|
||||
#endif
|
||||
|
||||
int a;
|
||||
asm volatile (
|
||||
".arm \n\t"
|
||||
".inst 0xe3010234 \n\t" // movw r0, 0x1234
|
||||
".inst 0xe3410234 \n\t" // movt r0, 0x1234
|
||||
"mov %0, r0 \n\t" // mov [a], r0
|
||||
: "=r" (a) : : "r0");
|
||||
volatile int a;
|
||||
asm volatile (
|
||||
".arm \n\t"
|
||||
".inst 0xe3010234 \n\t" // movw r0, 0x1234
|
||||
".inst 0xe3410234 \n\t" // movt r0, 0x1234
|
||||
"mov %0, r0 \n\t" // mov [a], r0
|
||||
: "=r" (a) : : "r0");
|
||||
|
||||
result = (a == 0x12341234);
|
||||
}
|
||||
result = (a == 0x12341234);
|
||||
}
|
||||
|
||||
sigprocmask(SIG_SETMASK, (sigset_t*)&oldMask, NULLPTR);
|
||||
signal(SIGILL, oldHandler);
|
||||
return result;
|
||||
sigprocmask(SIG_SETMASK, (sigset_t*)&oldMask, NULLPTR);
|
||||
signal(SIGILL, oldHandler);
|
||||
return result;
|
||||
# endif
|
||||
#else
|
||||
return false;
|
||||
return false;
|
||||
#endif // CRYPTOPP_BOOL_ARM32
|
||||
}
|
||||
|
||||
bool CPU_ProbeNEON()
|
||||
{
|
||||
#if defined(__aarch32__) || defined(__aarch64__)
|
||||
return true;
|
||||
return true;
|
||||
#elif defined(CRYPTOPP_NO_CPU_FEATURE_PROBES)
|
||||
return false;
|
||||
#elif (CRYPTOPP_ARM_NEON_AVAILABLE)
|
||||
return false;
|
||||
#elif CRYPTOPP_BOOL_ARM32
|
||||
# if defined(CRYPTOPP_MS_STYLE_INLINE_ASSEMBLY)
|
||||
volatile bool result = true;
|
||||
__try
|
||||
{
|
||||
uint32_t v1[4] = {1,1,1,1};
|
||||
uint32x4_t x1 = vld1q_u32(v1);
|
||||
uint64_t v2[2] = {1,1};
|
||||
uint64x2_t x2 = vld1q_u64(v2);
|
||||
|
||||
uint32x4_t x3 = vdupq_n_u32(2);
|
||||
x3 = vsetq_lane_u32(vgetq_lane_u32(x1,0),x3,0);
|
||||
x3 = vsetq_lane_u32(vgetq_lane_u32(x1,3),x3,3);
|
||||
uint64x2_t x4 = vdupq_n_u64(2);
|
||||
x4 = vsetq_lane_u64(vgetq_lane_u64(x2,0),x4,0);
|
||||
x4 = vsetq_lane_u64(vgetq_lane_u64(x2,1),x4,1);
|
||||
|
||||
result = !!(vgetq_lane_u32(x3,0) | vgetq_lane_u64(x4,1));
|
||||
}
|
||||
__except (EXCEPTION_EXECUTE_HANDLER)
|
||||
{
|
||||
return false;
|
||||
}
|
||||
return result;
|
||||
volatile bool result = true;
|
||||
__try
|
||||
{
|
||||
volatile uint32x4_t x = vdupq_n_u32(1);
|
||||
volatile uint32x4_t y = vshlq_n_u32(x, 4);
|
||||
return (y[0] & y[1] & y[2] & y[3]) == 16;
|
||||
}
|
||||
__except (EXCEPTION_EXECUTE_HANDLER)
|
||||
{
|
||||
return false;
|
||||
}
|
||||
return result;
|
||||
# else
|
||||
// longjmp and clobber warnings. Volatile is required.
|
||||
// http://github.com/weidai11/cryptopp/issues/24 and http://stackoverflow.com/q/7721854
|
||||
volatile bool result = true;
|
||||
|
||||
// longjmp and clobber warnings. Volatile is required.
|
||||
// http://github.com/weidai11/cryptopp/issues/24 and http://stackoverflow.com/q/7721854
|
||||
volatile bool result = true;
|
||||
volatile SigHandler oldHandler = signal(SIGILL, SigIllHandler);
|
||||
if (oldHandler == SIG_ERR)
|
||||
return false;
|
||||
|
||||
volatile SigHandler oldHandler = signal(SIGILL, SigIllHandler);
|
||||
if (oldHandler == SIG_ERR)
|
||||
return false;
|
||||
volatile sigset_t oldMask;
|
||||
if (sigprocmask(0, NULLPTR, (sigset_t*)&oldMask))
|
||||
return false;
|
||||
|
||||
volatile sigset_t oldMask;
|
||||
if (sigprocmask(0, NULLPTR, (sigset_t*)&oldMask))
|
||||
return false;
|
||||
if (setjmp(s_jmpSIGILL))
|
||||
result = false;
|
||||
else
|
||||
{
|
||||
// This is risky... When we hand encode the instructions
|
||||
// for vmov.u32 and vshl.u32 we get a SIGILL. Apparently
|
||||
// we need more than just the instructions. Using
|
||||
// intrinsics introduces the risk because the whole
|
||||
// file gets built with ISA options, and the higher ISA
|
||||
// may escape the try block with the SIGILL guard.
|
||||
uint32x4_t x = vdupq_n_u32(1);
|
||||
uint32x4_t y = vshlq_n_u32(x, 4);
|
||||
return (y[0] & y[1] & y[2] & y[3]) == 16;
|
||||
}
|
||||
|
||||
if (setjmp(s_jmpSIGILL))
|
||||
result = false;
|
||||
else
|
||||
{
|
||||
uint32_t v1[4] = {1,1,1,1};
|
||||
uint32x4_t x1 = vld1q_u32(v1);
|
||||
uint64_t v2[2] = {1,1};
|
||||
uint64x2_t x2 = vld1q_u64(v2);
|
||||
|
||||
uint32x4_t x3 = {0,0,0,0};
|
||||
x3 = vsetq_lane_u32(vgetq_lane_u32(x1,0),x3,0);
|
||||
x3 = vsetq_lane_u32(vgetq_lane_u32(x1,3),x3,3);
|
||||
uint64x2_t x4 = {0,0};
|
||||
x4 = vsetq_lane_u64(vgetq_lane_u64(x2,0),x4,0);
|
||||
x4 = vsetq_lane_u64(vgetq_lane_u64(x2,1),x4,1);
|
||||
|
||||
// Hack... GCC optimizes away the code and returns true
|
||||
result = !!(vgetq_lane_u32(x3,0) | vgetq_lane_u64(x4,1));
|
||||
}
|
||||
|
||||
sigprocmask(SIG_SETMASK, (sigset_t*)&oldMask, NULLPTR);
|
||||
signal(SIGILL, oldHandler);
|
||||
return result;
|
||||
sigprocmask(SIG_SETMASK, (sigset_t*)&oldMask, NULLPTR);
|
||||
signal(SIGILL, oldHandler);
|
||||
return result;
|
||||
# endif
|
||||
#else
|
||||
return false;
|
||||
return false;
|
||||
#endif // CRYPTOPP_ARM_NEON_AVAILABLE
|
||||
}
|
||||
|
||||
|
|
|
|||
Loading…
Reference in New Issue