Changes to support Intel AVX/AVX2 in cvResize().

This commit is contained in:
Richard Yoo
2014-06-06 13:37:13 -07:00
parent 9a5e9d3442
commit 11a09ef5cc
7 changed files with 1005 additions and 320 deletions

View File

@@ -1706,6 +1706,7 @@ CVAPI(double) cvGetTickFrequency( void );
#define CV_CPU_SSE4_2 7
#define CV_CPU_POPCNT 8
#define CV_CPU_AVX 10
#define CV_CPU_AVX2 11
#define CV_HARDWARE_MAX_FEATURE 255
CVAPI(int) cvCheckHardwareSupport(int feature);

View File

@@ -141,6 +141,10 @@ CV_INLINE IppiSize ippiSize(const cv::Size & _size)
# define __xgetbv() 0
# endif
# endif
# if defined __AVX2__
# include <immintrin.h>
# define CV_AVX2 1
# endif
#endif
@@ -176,6 +180,9 @@ CV_INLINE IppiSize ippiSize(const cv::Size & _size)
#ifndef CV_AVX
# define CV_AVX 0
#endif
#ifndef CV_AVX2
# define CV_AVX2 0
#endif
#ifndef CV_NEON
# define CV_NEON 0
#endif

View File

@@ -253,6 +253,41 @@ struct HWFeatures
f.have[CV_CPU_AVX] = (((cpuid_data[2] & (1<<28)) != 0)&&((cpuid_data[2] & (1<<27)) != 0));//OS uses XSAVE_XRSTORE and CPU support AVX
}
#if CV_AVX2
#if defined _MSC_VER && (defined _M_IX86 || defined _M_X64)
__cpuidex(cpuid_data, 7, 0);
#elif defined __GNUC__ && (defined __i386__ || defined __x86_64__)
#ifdef __x86_64__
asm __volatile__
(
"movl $7, %%eax\n\t"
"movl $0, %%ecx\n\t"
"cpuid\n\t"
:[eax]"=a"(cpuid_data[0]),[ebx]"=b"(cpuid_data[1]),[ecx]"=c"(cpuid_data[2]),[edx]"=d"(cpuid_data[3])
:
: "cc"
);
#else
asm volatile
(
"pushl %%ebx\n\t"
"movl $7,%%eax\n\t"
"movl $0,%%ecx\n\t"
"cpuid\n\t"
"popl %%ebx\n\t"
: "=a"(cpuid_data[0]), "=b"(cpuid_data[1]), "=c"(cpuid_data[2]), "=d"(cpuid_data[3])
:
: "cc"
);
#endif
#endif
if( f.x86_family >= 6 )
{
f.have[CV_CPU_AVX2] = (cpuid_data[1] & (1<<5)) != 0;
}
#endif
return f;
}

File diff suppressed because it is too large Load Diff

View File

@@ -3005,6 +3005,9 @@ void printVersionInfo(bool useStdOut)
#if CV_AVX
if (checkHardwareSupport(CV_CPU_AVX)) cpu_features += " avx";
#endif
#if CV_AVX2
if (checkHardwareSupport(CV_CPU_AVX2)) cpu_features += " avx2";
#endif
#if CV_NEON
cpu_features += " neon"; // NEON is currently not checked at runtime
#endif