generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

added checkPtxVersion into gpu module

Browse Source
This commit is contained in:
Alexey Spizhevoy
2011-01-18 14:52:35 +00:00
parent cbb132ccb1
commit 1a0d41fb53
5 changed files with 107 additions and 46 deletions
Download Patch File Download Diff File Expand all files Collapse all files

100
modules/gpu/src/initialization.cpp
View File

@@ -68,6 +68,7 @@ CV_EXPORTS int cv::gpu::getCudaEnabledDeviceCount()
return count;
}
CV_EXPORTS string cv::gpu::getDeviceName(int device)
{
cudaDeviceProp prop;
@@ -75,10 +76,13 @@ CV_EXPORTS string cv::gpu::getDeviceName(int device)
return prop.name;
}
CV_EXPORTS void cv::gpu::setDevice(int device)
{
cudaSafeCall( cudaSetDevice( device ) );
}
CV_EXPORTS int cv::gpu::getDevice()
{
int device;
@@ -86,6 +90,7 @@ CV_EXPORTS int cv::gpu::getDevice()
return device;
}
CV_EXPORTS void cv::gpu::getComputeCapability(int device, int& major, int& minor)
{
cudaDeviceProp prop;
@@ -95,6 +100,7 @@ CV_EXPORTS void cv::gpu::getComputeCapability(int device, int& major, int& minor
minor = prop.minor;
}
CV_EXPORTS int cv::gpu::getNumberOfSMs(int device)
{
cudaDeviceProp prop;
@@ -108,6 +114,7 @@ CV_EXPORTS void cv::gpu::getGpuMemInfo(size_t& free, size_t& total)
cudaSafeCall( cudaMemGetInfo( &free, &total ) );
}
CV_EXPORTS bool cv::gpu::hasNativeDoubleSupport(int device)
{
int major, minor;
@@ -115,6 +122,7 @@ CV_EXPORTS bool cv::gpu::hasNativeDoubleSupport(int device)
return major > 1 || (major == 1 && minor >= 3);
}
CV_EXPORTS bool cv::gpu::hasAtomicsSupport(int device)
{
int major, minor;
@@ -122,36 +130,90 @@ CV_EXPORTS bool cv::gpu::hasAtomicsSupport(int device)
return major > 1 || (major == 1 && minor >= 1);
}
CV_EXPORTS bool cv::gpu::hasPtxFor(int major, int minor)
namespace
{
#ifdef HAVE_PTX_FOR_NVIDIA_CC_10
if (major == 1 && minor == 0) return true;
template <unsigned int cmp_op>
bool comparePairs(int lhs1, int lhs2, int rhs1, int rhs2);
template <>
bool comparePairs<CMP_EQ>(int lhs1, int lhs2, int rhs1, int rhs2)
{
return lhs1 == rhs1 && lhs2 == rhs2;
}
template <>
bool comparePairs<CMP_GT>(int lhs1, int lhs2, int rhs1, int rhs2)
{
return lhs1 > rhs1 || (lhs1 == rhs1 && lhs2 > rhs2);
}
template <>
bool comparePairs<CMP_GE>(int lhs1, int lhs2, int rhs1, int rhs2)
{
return lhs1 > rhs1 || (lhs1 == rhs1 && lhs2 >= rhs2);
}
template <>
bool comparePairs<CMP_LT>(int lhs1, int lhs2, int rhs1, int rhs2)
{
return lhs1 < rhs1 || (lhs1 == rhs1 && lhs2 < rhs2);
}
template <>
bool comparePairs<CMP_LE>(int lhs1, int lhs2, int rhs1, int rhs2)
{
return lhs1 < rhs1 || (lhs1 == rhs1 && lhs2 <= rhs2);
}
template <>
bool comparePairs<CMP_NE>(int lhs1, int lhs2, int rhs1, int rhs2)
{
return lhs1 < rhs1 || (lhs1 == rhs1 && lhs2 <= rhs2);
}
}
template <unsigned int cmp_op>
CV_EXPORTS bool cv::gpu::checkPtxVersion(int major, int minor)
{
#ifdef OPENCV_GPU_CUDA_ARCH_10
if (comparePairs<cmp_op>(1, 0, major, minor)) return true;
#endif
#ifdef HAVE_PTX_FOR_NVIDIA_CC_11
if (major == 1 && minor == 1) return true;
#ifdef OPENCV_GPU_CUDA_ARCH_11
if (comparePairs<cmp_op>(1, 1, major, minor)) return true;
#endif
#ifdef HAVE_PTX_FOR_NVIDIA_CC_12
if (major == 1 && minor == 2) return true;
#ifdef OPENCV_GPU_CUDA_ARCH_12
if (comparePairs<cmp_op>(1, 2, major, minor)) return true;
#endif
#ifdef HAVE_PTX_FOR_NVIDIA_CC_13
if (major == 1 && minor == 3) return true;
#ifdef OPENCV_GPU_CUDA_ARCH_13
if (comparePairs<cmp_op>(1, 3, major, minor)) return true;
#endif
#ifdef HAVE_PTX_FOR_NVIDIA_CC_20
if (major == 2 && minor == 0) return true;
#ifdef OPENCV_GPU_CUDA_ARCH_20
if (comparePairs<cmp_op>(2, 0, major, minor)) return true;
#endif
#ifdef HAVE_PTX_FOR_NVIDIA_CC_21
if (major == 2 && minor == 1) return true;
#ifdef OPENCV_GPU_CUDA_ARCH_21
if (comparePairs<cmp_op>(2, 1, major, minor)) return true;
#endif
return false;
}
template CV_EXPORTS bool cv::gpu::checkPtxVersion<CMP_EQ>(int major, int minor);
template CV_EXPORTS bool cv::gpu::checkPtxVersion<CMP_GT>(int major, int minor);
template CV_EXPORTS bool cv::gpu::checkPtxVersion<CMP_GE>(int major, int minor);
template CV_EXPORTS bool cv::gpu::checkPtxVersion<CMP_LT>(int major, int minor);
template CV_EXPORTS bool cv::gpu::checkPtxVersion<CMP_LE>(int major, int minor);
template CV_EXPORTS bool cv::gpu::checkPtxVersion<CMP_NE>(int major, int minor);
CV_EXPORTS bool isCompatibleWith(int device)
{
// According to the CUDA C Programming Guide Version 3.2: "PTX code
@@ -161,17 +223,7 @@ CV_EXPORTS bool isCompatibleWith(int device)
int major, minor;
getComputeCapability(device, major, minor);
for (; major >= 1; --major)
{
for (; minor >= 0; --minor)
{
if (hasPtxFor(major, minor))
return true;
}
minor = 9;
}
return false;
return checkPtxVersion<CMP_LE>(major, minor);
}
#endif