mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2025-01-18 06:03:15 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Merge pull request #12565 from dkurt:dnn_non_intel_gpu

Browse Source 
* Remove isIntel check from deep learning layers

* Remove fp16->fp32 fallbacks where it's not necessary

* Fix Kernel::run to prevent localsize > globalsize
This commit is contained in:
Dmitry Kurtaev 2018-09-26 16:27:00 +03:00 committed by Alexander Alekhin
parent c8f3579f93
commit 24ab751547
29 changed files with 137 additions and 173 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

150
modules/core/include/opencv2/core/ocl.hpp
View File

@ -59,7 +59,7 @@ CV_EXPORTS_W void finish();
CV_EXPORTS bool haveSVM();
class CV_EXPORTS Context;
class CV_EXPORTS Device;
class CV_EXPORTS_W_SIMPLE Device;
class CV_EXPORTS Kernel;
class CV_EXPORTS Program;
class CV_EXPORTS ProgramSource;
@ -67,14 +67,14 @@ class CV_EXPORTS Queue;
class CV_EXPORTS PlatformInfo;
class CV_EXPORTS Image2D;
class CV_EXPORTS Device
class CV_EXPORTS_W_SIMPLE Device
{
public:
Device();
CV_WRAP Device();
explicit Device(void* d);
Device(const Device& d);
Device& operator = (const Device& d);
~Device();
CV_WRAP ~Device();
void set(void* d);
@ -89,24 +89,24 @@ public:
TYPE_ALL = 0xFFFFFFFF
};
String name() const;
String extensions() const;
bool isExtensionSupported(const String& extensionName) const;
String version() const;
String vendorName() const;
String OpenCL_C_Version() const;
String OpenCLVersion() const;
int deviceVersionMajor() const;
int deviceVersionMinor() const;
String driverVersion() const;
CV_WRAP String name() const;
CV_WRAP String extensions() const;
CV_WRAP bool isExtensionSupported(const String& extensionName) const;
CV_WRAP String version() const;
CV_WRAP String vendorName() const;
CV_WRAP String OpenCL_C_Version() const;
CV_WRAP String OpenCLVersion() const;
CV_WRAP int deviceVersionMajor() const;
CV_WRAP int deviceVersionMinor() const;
CV_WRAP String driverVersion() const;
void* ptr() const;
int type() const;
CV_WRAP int type() const;
int addressBits() const;
bool available() const;
bool compilerAvailable() const;
bool linkerAvailable() const;
CV_WRAP int addressBits() const;
CV_WRAP bool available() const;
CV_WRAP bool compilerAvailable() const;
CV_WRAP bool linkerAvailable() const;
enum
{
@ -119,21 +119,21 @@ public:
FP_SOFT_FLOAT=(1 << 6),
FP_CORRECTLY_ROUNDED_DIVIDE_SQRT=(1 << 7)
};
int doubleFPConfig() const;
int singleFPConfig() const;
int halfFPConfig() const;
CV_WRAP int doubleFPConfig() const;
CV_WRAP int singleFPConfig() const;
CV_WRAP int halfFPConfig() const;
bool endianLittle() const;
bool errorCorrectionSupport() const;
CV_WRAP bool endianLittle() const;
CV_WRAP bool errorCorrectionSupport() const;
enum
{
EXEC_KERNEL=(1 << 0),
EXEC_NATIVE_KERNEL=(1 << 1)
};
int executionCapabilities() const;
CV_WRAP int executionCapabilities() const;
size_t globalMemCacheSize() const;
CV_WRAP size_t globalMemCacheSize() const;
enum
{
@ -141,38 +141,38 @@ public:
READ_ONLY_CACHE=1,
READ_WRITE_CACHE=2
};
int globalMemCacheType() const;
int globalMemCacheLineSize() const;
size_t globalMemSize() const;
CV_WRAP int globalMemCacheType() const;
CV_WRAP int globalMemCacheLineSize() const;
CV_WRAP size_t globalMemSize() const;
size_t localMemSize() const;
CV_WRAP size_t localMemSize() const;
enum
{
NO_LOCAL_MEM=0,
LOCAL_IS_LOCAL=1,
LOCAL_IS_GLOBAL=2
};
int localMemType() const;
bool hostUnifiedMemory() const;
CV_WRAP int localMemType() const;
CV_WRAP bool hostUnifiedMemory() const;
bool imageSupport() const;
CV_WRAP bool imageSupport() const;
bool imageFromBufferSupport() const;
CV_WRAP bool imageFromBufferSupport() const;
uint imagePitchAlignment() const;
uint imageBaseAddressAlignment() const;
/// deprecated, use isExtensionSupported() method (probably with "cl_khr_subgroups" value)
bool intelSubgroupsSupport() const;
CV_WRAP bool intelSubgroupsSupport() const;
size_t image2DMaxWidth() const;
size_t image2DMaxHeight() const;
CV_WRAP size_t image2DMaxWidth() const;
CV_WRAP size_t image2DMaxHeight() const;
size_t image3DMaxWidth() const;
size_t image3DMaxHeight() const;
size_t image3DMaxDepth() const;
CV_WRAP size_t image3DMaxWidth() const;
CV_WRAP size_t image3DMaxHeight() const;
CV_WRAP size_t image3DMaxDepth() const;
size_t imageMaxBufferSize() const;
size_t imageMaxArraySize() const;
CV_WRAP size_t imageMaxBufferSize() const;
CV_WRAP size_t imageMaxArraySize() const;
enum
{
@ -181,53 +181,53 @@ public:
VENDOR_INTEL=2,
VENDOR_NVIDIA=3
};
int vendorID() const;
CV_WRAP int vendorID() const;
// FIXIT
// dev.isAMD() doesn't work for OpenCL CPU devices from AMD OpenCL platform.
// This method should use platform name instead of vendor name.
// After fix restore code in arithm.cpp: ocl_compare()
inline bool isAMD() const { return vendorID() == VENDOR_AMD; }
inline bool isIntel() const { return vendorID() == VENDOR_INTEL; }
inline bool isNVidia() const { return vendorID() == VENDOR_NVIDIA; }
CV_WRAP inline bool isAMD() const { return vendorID() == VENDOR_AMD; }
CV_WRAP inline bool isIntel() const { return vendorID() == VENDOR_INTEL; }
CV_WRAP inline bool isNVidia() const { return vendorID() == VENDOR_NVIDIA; }
int maxClockFrequency() const;
int maxComputeUnits() const;
int maxConstantArgs() const;
size_t maxConstantBufferSize() const;
CV_WRAP int maxClockFrequency() const;
CV_WRAP int maxComputeUnits() const;
CV_WRAP int maxConstantArgs() const;
CV_WRAP size_t maxConstantBufferSize() const;
size_t maxMemAllocSize() const;
size_t maxParameterSize() const;
CV_WRAP size_t maxMemAllocSize() const;
CV_WRAP size_t maxParameterSize() const;
int maxReadImageArgs() const;
int maxWriteImageArgs() const;
int maxSamplers() const;
CV_WRAP int maxReadImageArgs() const;
CV_WRAP int maxWriteImageArgs() const;
CV_WRAP int maxSamplers() const;
size_t maxWorkGroupSize() const;
int maxWorkItemDims() const;
CV_WRAP size_t maxWorkGroupSize() const;
CV_WRAP int maxWorkItemDims() const;
void maxWorkItemSizes(size_t*) const;
int memBaseAddrAlign() const;
CV_WRAP int memBaseAddrAlign() const;
int nativeVectorWidthChar() const;
int nativeVectorWidthShort() const;
int nativeVectorWidthInt() const;
int nativeVectorWidthLong() const;
int nativeVectorWidthFloat() const;
int nativeVectorWidthDouble() const;
int nativeVectorWidthHalf() const;
CV_WRAP int nativeVectorWidthChar() const;
CV_WRAP int nativeVectorWidthShort() const;
CV_WRAP int nativeVectorWidthInt() const;
CV_WRAP int nativeVectorWidthLong() const;
CV_WRAP int nativeVectorWidthFloat() const;
CV_WRAP int nativeVectorWidthDouble() const;
CV_WRAP int nativeVectorWidthHalf() const;
int preferredVectorWidthChar() const;
int preferredVectorWidthShort() const;
int preferredVectorWidthInt() const;
int preferredVectorWidthLong() const;
int preferredVectorWidthFloat() const;
int preferredVectorWidthDouble() const;
int preferredVectorWidthHalf() const;
CV_WRAP int preferredVectorWidthChar() const;
CV_WRAP int preferredVectorWidthShort() const;
CV_WRAP int preferredVectorWidthInt() const;
CV_WRAP int preferredVectorWidthLong() const;
CV_WRAP int preferredVectorWidthFloat() const;
CV_WRAP int preferredVectorWidthDouble() const;
CV_WRAP int preferredVectorWidthHalf() const;
size_t printfBufferSize() const;
size_t profilingTimerResolution() const;
CV_WRAP size_t printfBufferSize() const;
CV_WRAP size_t profilingTimerResolution() const;
static const Device& getDefault();
CV_WRAP static const Device& getDefault();
protected:
struct Impl;

2
modules/core/src/ocl.cpp
View File

@ -3078,7 +3078,7 @@ bool Kernel::run(int dims, size_t _globalsize[], size_t _localsize[],
dims == 1 ? 64 : dims == 2 ? (i == 0 ? 256 : 8) : dims == 3 ? (8>>(int)(i>0)) : 1;
CV_Assert( val > 0 );
total *= _globalsize[i];
if (_globalsize[i] == 1)
if (_globalsize[i] == 1 && !_localsize)
val = 1;
globalsize[i] = divUp(_globalsize[i], (unsigned int)val) * val;
}

15
modules/core/test/ocl/test_gemm.cpp
View File

@ -145,6 +145,21 @@ OCL_INSTANTIATE_TEST_CASE_P(Core, Gemm, ::testing::Combine(
testing::Values(CV_32FC1, CV_32FC2, CV_64FC1, CV_64FC2),
Bool(), Bool(), Bool(), Bool()));
// Test for non-Intel GPUs to check CL_INVALID_WORK_GROUP_SIZE when localsize > globalsize
OCL_TEST(Gemm, small)
{
UMat A(2, 3, CV_32F), B(4, 3, CV_32F), uC(2, 4, CV_32F);
Mat C(2, 4, CV_32F);
randu(A, -1, 1);
randu(B, -1, 1);
OCL_OFF(cv::gemm(A, B, 1, noArray(), 0, C, GEMM_2_T));
OCL_ON(cv::gemm(A, B, 1, noArray(), 0, uC, GEMM_2_T));
EXPECT_LE(cvtest::norm(C, uC, cv::NORM_INF), 1e-5);
}
} } // namespace opencv_test::ocl
#endif // HAVE_OPENCL

20
modules/dnn/src/dnn.cpp
View File

@ -1078,12 +1078,22 @@ struct Net::Impl
}
#else
{
if (!DNN_OPENCL_ALLOW_ALL_DEVICES
&& !(ocl::Device::getDefault().isIntel() && ocl::Device::getDefault().type() == ocl::Device::TYPE_GPU) // Current implementation is only valid for Intel GPU (#11494)
)
if (!DNN_OPENCL_ALLOW_ALL_DEVICES)
{
CV_LOG_WARNING(NULL, "DNN: OpenCL target is not supported with current OpenCL device (tested with Intel GPUs only), switching to CPU.");
preferableTarget = DNN_TARGET_CPU;
// Current implementation is only valid for GPU (#11494)
if (ocl::Device::getDefault().type() != ocl::Device::TYPE_GPU)
{
CV_LOG_WARNING(NULL, "DNN: OpenCL target is not supported with current OpenCL device (tested with GPUs only), switching to CPU.");
preferableTarget = DNN_TARGET_CPU;
}
else if (preferableTarget == DNN_TARGET_OPENCL_FP16 && !ocl::Device::getDefault().isIntel())
{
CV_LOG_WARNING(NULL,
"DNN: OpenCL target with fp16 precision is not supported "
"with current OpenCL device (tested with Intel GPUs only), "
"switching to OpenCL with fp32 precision.");
preferableTarget = DNN_TARGET_OPENCL;
}
}
}
#endif

3
modules/dnn/src/layers/batch_norm_layer.cpp
View File

@ -230,8 +230,7 @@ public:
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget) &&
OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()),
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
forward_ocl(inputs_arr, outputs_arr, internals_arr))
if (inputs_arr.depth() == CV_16S)

9
modules/dnn/src/layers/blank_layer.cpp
View File

@ -95,16 +95,9 @@ public:
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget) &&
OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()),
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
forward_ocl(inputs_arr, outputs_arr, internals_arr))
if (inputs_arr.depth() == CV_16S)
{
forward_fallback(inputs_arr, outputs_arr, internals_arr);
return;
}
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);

9
modules/dnn/src/layers/concat_layer.cpp
View File

@ -237,16 +237,9 @@ public:
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget) &&
OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()),
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
forward_ocl(inputs_arr, outputs_arr, internals_arr))
if (inputs_arr.depth() == CV_16S)
{
forward_fallback(inputs_arr, outputs_arr, internals_arr);
return;
}
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);

3
modules/dnn/src/layers/convolution_layer.cpp
View File

@ -1529,8 +1529,7 @@ public:
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget) &&
OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()),
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
forward_ocl(inputs_arr, outputs_arr, internals_arr));
if (inputs_arr.depth() == CV_16S)

6
modules/dnn/src/layers/crop_layer.cpp
View File

@ -137,12 +137,6 @@ public:
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
if (inputs_arr.depth() == CV_16S)
{
forward_fallback(inputs_arr, outputs_arr, internals_arr);
return;
}
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);

3
modules/dnn/src/layers/detection_output_layer.cpp
View File

@ -415,8 +415,7 @@ public:
if (_bboxesNormalized)
{
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget) &&
OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()),
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
forward_ocl(inputs_arr, outputs_arr, internals_arr))
}
if (inputs_arr.depth() == CV_16S)

3
modules/dnn/src/layers/eltwise_layer.cpp
View File

@ -354,8 +354,7 @@ public:
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget) &&
OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()),
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
forward_ocl(inputs_arr, outputs_arr, internals_arr))
if (inputs_arr.depth() == CV_16S)

9
modules/dnn/src/layers/flatten_layer.cpp
View File

@ -135,16 +135,9 @@ public:
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget) &&
outputs_arr.isUMatVector() &&
OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()),
outputs_arr.isUMatVector(),
forward_ocl(inputs_arr, outputs_arr, internals_arr))
if (inputs_arr.depth() == CV_16S)
{
forward_fallback(inputs_arr, outputs_arr, internals_arr);
return;
}
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);

3
modules/dnn/src/layers/fully_connected_layer.cpp
View File

@ -389,8 +389,7 @@ public:
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget) &&
OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()),
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
forward_ocl(inputs_arr, outputs_arr, internals_arr))
if (inputs_arr.depth() == CV_16S)

3
modules/dnn/src/layers/lrn_layer.cpp
View File

@ -148,8 +148,7 @@ public:
CV_Assert(inputs_arr.total() == outputs_arr.total());
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget) &&
OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()),
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
forward_ocl(inputs_arr, outputs_arr, internals_arr))
if (inputs_arr.depth() == CV_16S)

3
modules/dnn/src/layers/normalize_bbox_layer.cpp
View File

@ -184,8 +184,7 @@ public:
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget) &&
OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()),
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
forward_ocl(inputs_arr, outputs_arr, internals_arr))
if (inputs_arr.depth() == CV_16S)

16
modules/dnn/src/layers/padding_layer.cpp
View File

@ -99,19 +99,21 @@ public:
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
if (inputs_arr.depth() == CV_16S)
{
forward_fallback(inputs_arr, outputs_arr, internals_arr);
return;
}
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);
if (paddingType == "constant")
{
outputs[0].setTo(paddingValue);
if (inputs_arr.depth() == CV_16S)
{
std::vector<float> paddingValue_fp32(1, paddingValue);
std::vector<int16_t> paddingValue_fp16(1);
convertFp16(paddingValue_fp32, paddingValue_fp16);
outputs[0].setTo(paddingValue_fp16[0]);
}
else
outputs[0].setTo(paddingValue);
inputs[0].copyTo(outputs[0](dstRanges));
}
else if (paddingType == "reflect")

3
modules/dnn/src/layers/permute_layer.cpp
View File

@ -304,8 +304,7 @@ public:
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget) &&
OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()),
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
forward_ocl(inputs_arr, outputs_arr, internals_arr))
if (inputs_arr.depth() == CV_16S)

3
modules/dnn/src/layers/prior_box_layer.cpp
View File

@ -402,8 +402,7 @@ public:
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget) &&
OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()),
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
forward_ocl(inputs_arr, outputs_arr, internals_arr))
if (inputs_arr.depth() == CV_16S)

3
modules/dnn/src/layers/region_layer.cpp
View File

@ -196,8 +196,7 @@ public:
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget) &&
OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()),
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
forward_ocl(inputs_arr, outputs_arr, internals_arr))
if (inputs_arr.depth() == CV_16S)

3
modules/dnn/src/layers/reorg_layer.cpp
View File

@ -160,8 +160,7 @@ public:
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget) &&
OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()),
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
forward_ocl(inputs_arr, outputs_arr, internals_arr))
if (inputs_arr.depth() == CV_16S)

9
modules/dnn/src/layers/reshape_layer.cpp
View File

@ -233,16 +233,9 @@ public:
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget) &&
OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()),
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
forward_ocl(inputs_arr, outputs_arr, internals_arr))
if (inputs_arr.depth() == CV_16S)
{
forward_fallback(inputs_arr, outputs_arr, internals_arr);
return;
}
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);

3
modules/dnn/src/layers/shuffle_channel_layer.cpp
View File

@ -92,8 +92,7 @@ public:
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget) &&
OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()),
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
forward_ocl(inputs_arr, outputs_arr, internals_arr))
if (inputs_arr.depth() == CV_16S)

9
modules/dnn/src/layers/slice_layer.cpp
View File

@ -239,16 +239,9 @@ public:
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget) &&
OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()),
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
forward_ocl(inputs_arr, outputs_arr, internals_arr))
if (inputs_arr.depth() == CV_16S)
{
forward_fallback(inputs_arr, outputs_arr, internals_arr);
return;
}
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);

3
modules/dnn/src/layers/softmax_layer.cpp
View File

@ -187,8 +187,7 @@ public:
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget) &&
OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()),
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
forward_ocl(inputs_arr, outputs_arr, internals_arr))
if (inputs_arr.depth() == CV_16S)

6
modules/dnn/src/layers/split_layer.cpp
View File

@ -83,12 +83,6 @@ public:
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
if (inputs_arr.depth() == CV_16S)
{
forward_fallback(inputs_arr, outputs_arr, internals_arr);
return;
}
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);

3
modules/dnn/src/ocl4dnn/src/ocl4dnn_lrn.cpp
View File

@ -69,9 +69,6 @@ bool OCL4DNNLRN<Dtype>::Forward(const UMat& bottom, UMat& top)
{
bool ret = true;
if (!ocl::Device::getDefault().intelSubgroupsSupport())
return false;
switch (lrn_type_)
{
case LRNParameter_NormRegion_ACROSS_CHANNELS:

2
modules/dnn/src/opencl/prior_box.cl
View File

@ -114,6 +114,6 @@ __kernel void clip(const int nthreads,
for (int index = get_global_id(0); index < nthreads; index += get_global_size(0))
{
Dtype4 vec = vload4(index, dst);
vstore4(clamp(vec, 0, 1), index, dst);
vstore4(clamp(vec, 0.0f, 1.0f), index, dst);
}
}

4
modules/dnn/test/test_onnx_importer.cpp
View File

@ -295,7 +295,7 @@ TEST_P(Test_ONNX_nets, TinyYolov2)
TEST_P(Test_ONNX_nets, CNN_MNIST)
{
// output range: [-1952; 6574]
const double l1 = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 3.82 : 4.3e-4;
const double l1 = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 3.82 : 4.4e-4;
const double lInf = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 13.5 : 2e-3;
testONNXModels("cnn_mnist", pb, l1, lInf);
@ -341,7 +341,7 @@ TEST_P(Test_ONNX_nets, Inception_v2)
TEST_P(Test_ONNX_nets, DenseNet121)
{
// output range: [-87; 138]
const double l1 = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.12 : 1.88e-5;
const double l1 = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.12 : 2.2e-5;
const double lInf = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.74 : 1.23e-4;
testONNXModels("densenet121", pb, l1, lInf);
}

2
modules/python/test/test_dnn.py
View File

@ -95,7 +95,7 @@ if haveInfEngine:
if cv.ocl.haveOpenCL() and cv.ocl.useOpenCL():
dnnBackendsAndTargets.append([cv.dnn.DNN_BACKEND_OPENCV, cv.dnn.DNN_TARGET_OPENCL])
dnnBackendsAndTargets.append([cv.dnn.DNN_BACKEND_OPENCV, cv.dnn.DNN_TARGET_OPENCL_FP16])
if haveInfEngine: # FIXIT Check Intel iGPU only
if haveInfEngine and cv.ocl_Device.getDefault().isIntel():
dnnBackendsAndTargets.append([cv.dnn.DNN_BACKEND_INFERENCE_ENGINE, cv.dnn.DNN_TARGET_OPENCL])
dnnBackendsAndTargets.append([cv.dnn.DNN_BACKEND_INFERENCE_ENGINE, cv.dnn.DNN_TARGET_OPENCL_FP16])