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

replaced has* methods in the GPU module with the TargetArchs monostate

Browse Source
This commit is contained in:
Alexey Spizhevoy
2011-01-27 10:06:38 +00:00
parent 91769d0ed4
commit 891e2ff310
11 changed files with 104 additions and 125 deletions
Download Patch File Download Diff File Expand all files Collapse all files

60
modules/gpu/src/initialization.cpp
View File

@@ -162,52 +162,62 @@ namespace
}
CV_EXPORTS bool cv::gpu::hasPtxVersion(int major, int minor)
CV_EXPORTS bool cv::gpu::TargetArchs::builtWith(cv::gpu::GpuFeature feature)
{
if (feature == NATIVE_DOUBLE)
return hasEqualOrGreater(1, 3);
if (feature == ATOMICS)
return hasEqualOrGreater(1, 1);
return true;
}
CV_EXPORTS bool cv::gpu::TargetArchs::has(int major, int minor)
{
return hasPtx(major, minor) || hasBin(major, minor);
}
CV_EXPORTS bool cv::gpu::TargetArchs::hasPtx(int major, int minor)
{
return ::compare(CUDA_ARCH_PTX, major * 10 + minor, std::equal_to<int>());
}
CV_EXPORTS bool cv::gpu::hasLessOrEqualPtxVersion(int major, int minor)
CV_EXPORTS bool cv::gpu::TargetArchs::hasBin(int major, int minor)
{
return ::compare(CUDA_ARCH_BIN, major * 10 + minor, std::equal_to<int>());
}
CV_EXPORTS bool cv::gpu::TargetArchs::hasEqualOrLessPtx(int major, int minor)
{
return ::compare(CUDA_ARCH_PTX, major * 10 + minor,
std::less_equal<int>());
}
CV_EXPORTS bool cv::gpu::hasGreaterOrEqualPtxVersion(int major, int minor)
CV_EXPORTS bool cv::gpu::TargetArchs::hasEqualOrGreater(int major, int minor)
{
return hasEqualOrGreaterPtx(major, minor) ||
hasEqualOrGreaterBin(major, minor);
}
CV_EXPORTS bool cv::gpu::TargetArchs::hasEqualOrGreaterPtx(int major, int minor)
{
return ::compare(CUDA_ARCH_PTX, major * 10 + minor,
std::greater_equal<int>());
}
CV_EXPORTS bool cv::gpu::hasCubinVersion(int major, int minor)
{
return ::compare(CUDA_ARCH_BIN, major * 10 + minor, std::equal_to<int>());
}
CV_EXPORTS bool cv::gpu::hasGreaterOrEqualCubinVersion(int major, int minor)
CV_EXPORTS bool cv::gpu::TargetArchs::hasEqualOrGreaterBin(int major, int minor)
{
return ::compare(CUDA_ARCH_BIN, major * 10 + minor,
std::greater_equal<int>());
}
CV_EXPORTS bool cv::gpu::hasVersion(int major, int minor)
{
return hasPtxVersion(major, minor) || hasCubinVersion(major, minor);
}
CV_EXPORTS bool cv::gpu::hasGreaterOrEqualVersion(int major, int minor)
{
return hasGreaterOrEqualPtxVersion(major, minor) ||
hasGreaterOrEqualCubinVersion(major, minor);
}
CV_EXPORTS bool cv::gpu::isCompatibleWith(int device)
{
// According to the CUDA C Programming Guide Version 3.2: "PTX code
@@ -218,12 +228,12 @@ CV_EXPORTS bool cv::gpu::isCompatibleWith(int device)
getComputeCapability(device, major, minor);
// Check PTX compatibility
if (hasLessOrEqualPtxVersion(major, minor))
if (TargetArchs::hasEqualOrLessPtx(major, minor))
return true;
// Check CUBIN compatibility
for (int i = minor; i >= 0; --i)
if (hasCubinVersion(major, i))
if (TargetArchs::hasBin(major, i))
return true;
return false;

20
modules/gpu/src/matrix_reductions.cpp
View File

@@ -170,7 +170,7 @@ Scalar cv::gpu::sum(const GpuMat& src, GpuMat& buf)
ensureSizeIsEnough(buf_size, CV_8U, buf);
Caller* callers = multipass_callers;
if (hasGreaterOrEqualVersion(1, 1) && hasAtomicsSupport(getDevice()))
if (TargetArchs::builtWith(ATOMICS) && hasAtomicsSupport(getDevice()))
callers = singlepass_callers;
Caller caller = callers[src.depth()];
@@ -206,7 +206,7 @@ Scalar cv::gpu::sqrSum(const GpuMat& src, GpuMat& buf)
sqrSumCaller<int>, sqrSumCaller<float>, 0 };
Caller* callers = multipass_callers;
if (hasGreaterOrEqualVersion(1, 1) && hasAtomicsSupport(getDevice()))
if (TargetArchs::builtWith(ATOMICS) && hasAtomicsSupport(getDevice()))
callers = singlepass_callers;
Size buf_size;
@@ -283,7 +283,7 @@ void cv::gpu::minMax(const GpuMat& src, double* minVal, double* maxVal, const Gp
CV_Assert(mask.empty() || (mask.type() == CV_8U && src.size() == mask.size()));
CV_Assert(src.type() != CV_64F || (hasGreaterOrEqualVersion(1, 3) &&
CV_Assert(src.type() != CV_64F || (TargetArchs::builtWith(NATIVE_DOUBLE) &&
hasNativeDoubleSupport(getDevice())));
double minVal_; if (!minVal) minVal = &minVal_;
@@ -296,7 +296,7 @@ void cv::gpu::minMax(const GpuMat& src, double* minVal, double* maxVal, const Gp
if (mask.empty())
{
Caller* callers = multipass_callers;
if (hasGreaterOrEqualVersion(1, 1) && hasAtomicsSupport(getDevice()))
if (TargetArchs::builtWith(ATOMICS) && hasAtomicsSupport(getDevice()))
callers = singlepass_callers;
Caller caller = callers[src.type()];
@@ -306,7 +306,7 @@ void cv::gpu::minMax(const GpuMat& src, double* minVal, double* maxVal, const Gp
else
{
MaskedCaller* callers = masked_multipass_callers;
if (hasGreaterOrEqualVersion(1, 1) && hasAtomicsSupport(getDevice()))
if (TargetArchs::builtWith(ATOMICS) && hasAtomicsSupport(getDevice()))
callers = masked_singlepass_callers;
MaskedCaller caller = callers[src.type()];
@@ -382,7 +382,7 @@ void cv::gpu::minMaxLoc(const GpuMat& src, double* minVal, double* maxVal, Point
CV_Assert(mask.empty() || (mask.type() == CV_8U && src.size() == mask.size()));
CV_Assert(src.type() != CV_64F || (hasGreaterOrEqualVersion(1, 3) &&
CV_Assert(src.type() != CV_64F || (TargetArchs::builtWith(NATIVE_DOUBLE) &&
hasNativeDoubleSupport(getDevice())));
double minVal_; if (!minVal) minVal = &minVal_;
@@ -399,7 +399,7 @@ void cv::gpu::minMaxLoc(const GpuMat& src, double* minVal, double* maxVal, Point
if (mask.empty())
{
Caller* callers = multipass_callers;
if (hasGreaterOrEqualVersion(1, 1) && hasAtomicsSupport(getDevice()))
if (TargetArchs::builtWith(ATOMICS) && hasAtomicsSupport(getDevice()))
callers = singlepass_callers;
Caller caller = callers[src.type()];
@@ -409,7 +409,7 @@ void cv::gpu::minMaxLoc(const GpuMat& src, double* minVal, double* maxVal, Point
else
{
MaskedCaller* callers = masked_multipass_callers;
if (hasGreaterOrEqualVersion(1, 1) && hasAtomicsSupport(getDevice()))
if (TargetArchs::builtWith(ATOMICS) && hasAtomicsSupport(getDevice()))
callers = masked_singlepass_callers;
MaskedCaller caller = callers[src.type()];
@@ -463,7 +463,7 @@ int cv::gpu::countNonZero(const GpuMat& src, GpuMat& buf)
CV_Assert(src.channels() == 1);
CV_Assert(src.type() != CV_64F || (hasGreaterOrEqualVersion(1, 3) &&
CV_Assert(src.type() != CV_64F || (TargetArchs::builtWith(NATIVE_DOUBLE) &&
hasNativeDoubleSupport(getDevice())));
Size buf_size;
@@ -471,7 +471,7 @@ int cv::gpu::countNonZero(const GpuMat& src, GpuMat& buf)
ensureSizeIsEnough(buf_size, CV_8U, buf);
Caller* callers = multipass_callers;
if (hasGreaterOrEqualVersion(1, 1) && hasAtomicsSupport(getDevice()))
if (TargetArchs::builtWith(ATOMICS) && hasAtomicsSupport(getDevice()))
callers = singlepass_callers;
Caller caller = callers[src.type()];

6
modules/gpu/src/split_merge.cpp
View File

@@ -72,8 +72,8 @@ namespace cv { namespace gpu { namespace split_merge
{
CV_Assert(src);
CV_Assert(n > 0);
bool double_ok = hasGreaterOrEqualVersion(1, 3) &&
bool double_ok = TargetArchs::builtWith(NATIVE_DOUBLE) &&
hasNativeDoubleSupport(getDevice());
CV_Assert(src[0].depth() != CV_64F || double_ok);
@@ -116,7 +116,7 @@ namespace cv { namespace gpu { namespace split_merge
{
CV_Assert(dst);
bool double_ok = hasGreaterOrEqualVersion(1, 3) &&
bool double_ok = TargetArchs::builtWith(NATIVE_DOUBLE) &&
hasNativeDoubleSupport(getDevice());
CV_Assert(src.depth() != CV_64F || double_ok);