CL: Remaining functions for OpenCL 1.2

include/angle_cl.h

@@ -29,11 +29,10 @@ using ContextErrorCB = void(CL_CALLBACK *)(const char *errinfo,
                                            size_t cb,
                                            void *user_data);
 
+using MemoryCB  = void(CL_CALLBACK *)(cl_mem memobj, void *user_data);
 using ProgramCB = void(CL_CALLBACK *)(cl_program program, void *user_data);
-
-using EventCB = void(CL_CALLBACK *)(cl_event event, cl_int event_command_status, void *user_data);
-
-using UserFunc = void(CL_CALLBACK *)(void *args);
+using EventCB   = void(CL_CALLBACK *)(cl_event event, cl_int event_command_status, void *user_data);
+using UserFunc  = void(CL_CALLBACK *)(void *args);
 
 template <typename T = void>
 struct Dispatch
@@ -80,6 +79,11 @@ struct NativeObject : public Dispatch<>
 
     NativeObjectType *getNative() { return static_cast<NativeObjectType *>(this); }
 
+    const NativeObjectType *getNative() const
+    {
+        return static_cast<const NativeObjectType *>(this);
+    }
+
     static NativeObjectType *CastNative(NativeObjectType *p) { return p; }
 };
 

src/libANGLE/CLBuffer.h

@@ -31,7 +31,7 @@ class Buffer final : public Memory
   public:
     ~Buffer() override;
 
-    cl_mem_object_type getType() const final;
+    MemObjectType getType() const final;
 
     bool isSubBuffer() const;
 
@@ -60,12 +60,12 @@ inline bool Buffer::isRegionValid(const cl_buffer_region &region) const
 
 inline bool Buffer::IsValid(const _cl_mem *buffer)
 {
-    return Memory::IsValid(buffer) && buffer->cast<Memory>().getType() == CL_MEM_OBJECT_BUFFER;
+    return Memory::IsValid(buffer) && buffer->cast<Memory>().getType() == MemObjectType::Buffer;
 }
 
-inline cl_mem_object_type Buffer::getType() const
+inline MemObjectType Buffer::getType() const
 {
-    return CL_MEM_OBJECT_BUFFER;
+    return MemObjectType::Buffer;
 }
 
 inline bool Buffer::isSubBuffer() const

src/libANGLE/CLContext.cpp

@@ -131,12 +131,17 @@ cl_mem Context::createImage(const cl_mem_properties *properties,
                             void *hostPtr,
                             cl_int &errorCode)
 {
-    const ImageDescriptor imageDesc = {
-        desc->image_type,        desc->image_width,      desc->image_height,
-        desc->image_depth,       desc->image_array_size, desc->image_row_pitch,
-        desc->image_slice_pitch, desc->num_mip_levels,   desc->num_samples};
+    const ImageDescriptor imageDesc = {FromCLenum<MemObjectType>(desc->image_type),
+                                       desc->image_width,
+                                       desc->image_height,
+                                       desc->image_depth,
+                                       desc->image_array_size,
+                                       desc->image_row_pitch,
+                                       desc->image_slice_pitch,
+                                       desc->num_mip_levels,
+                                       desc->num_samples};
     return Object::Create<Image>(errorCode, *this, Memory::PropArray{}, flags, *format, imageDesc,
-                                 &desc->buffer->cast<Memory>(), hostPtr);
+                                 Memory::Cast(desc->buffer), hostPtr);
 }
 
 cl_mem Context::createImage2D(MemFlags flags,
@@ -148,7 +153,7 @@ cl_mem Context::createImage2D(MemFlags flags,
                               cl_int &errorCode)
 {
     const ImageDescriptor imageDesc = {
-        CL_MEM_OBJECT_IMAGE2D, width, height, 0u, 0u, rowPitch, 0u, 0u, 0u};
+        MemObjectType::Image2D, width, height, 0u, 0u, rowPitch, 0u, 0u, 0u};
     return Object::Create<Image>(errorCode, *this, Memory::PropArray{}, flags, *format, imageDesc,
                                  nullptr, hostPtr);
 }
@@ -164,11 +169,21 @@ cl_mem Context::createImage3D(MemFlags flags,
                               cl_int &errorCode)
 {
     const ImageDescriptor imageDesc = {
-        CL_MEM_OBJECT_IMAGE3D, width, height, depth, 0u, rowPitch, slicePitch, 0u, 0u};
+        MemObjectType::Image3D, width, height, depth, 0u, rowPitch, slicePitch, 0u, 0u};
     return Object::Create<Image>(errorCode, *this, Memory::PropArray{}, flags, *format, imageDesc,
                                  nullptr, hostPtr);
 }
 
+cl_int Context::getSupportedImageFormats(MemFlags flags,
+                                         MemObjectType imageType,
+                                         cl_uint numEntries,
+                                         cl_image_format *imageFormats,
+                                         cl_uint *numImageFormats)
+{
+    return mImpl->getSupportedImageFormats(flags, imageType, numEntries, imageFormats,
+                                           numImageFormats);
+}
+
 cl_sampler Context::createSamplerWithProperties(const cl_sampler_properties *properties,
                                                 cl_int &errorCode)
 {

src/libANGLE/CLContext.h

@@ -62,6 +62,12 @@ class Context final : public _cl_context, public Object
                          void *hostPtr,
                          cl_int &errorCode);
 
+    cl_int getSupportedImageFormats(MemFlags flags,
+                                    MemObjectType imageType,
+                                    cl_uint numEntries,
+                                    cl_image_format *imageFormats,
+                                    cl_uint *numImageFormats);
+
     cl_sampler createSamplerWithProperties(const cl_sampler_properties *properties,
                                            cl_int &errorCode);
 

src/libANGLE/CLDevice.cpp

@@ -285,7 +285,7 @@ cl_int Device::getInfo(DeviceInfo name, size_t valueSize, void *value, size_t *v
             copySize   = sizeof(valPointer);
             break;
         case DeviceInfo::ParentDevice:
-            valPointer = mParent->getNative();
+            valPointer = Device::CastNative(mParent.get());
             copyValue  = &valPointer;
             copySize   = sizeof(valPointer);
             break;
@@ -388,25 +388,25 @@ bool Device::supportsBuiltInKernel(const std::string &name) const
 
 bool Device::supportsNativeImageDimensions(const cl_image_desc &desc) const
 {
-    switch (desc.image_type)
+    switch (FromCLenum<MemObjectType>(desc.image_type))
     {
-        case CL_MEM_OBJECT_IMAGE1D:
+        case MemObjectType::Image1D:
             return desc.image_width <= mInfo.mImage2D_MaxWidth;
-        case CL_MEM_OBJECT_IMAGE2D:
+        case MemObjectType::Image2D:
             return desc.image_width <= mInfo.mImage2D_MaxWidth &&
                    desc.image_height <= mInfo.mImage2D_MaxHeight;
-        case CL_MEM_OBJECT_IMAGE3D:
+        case MemObjectType::Image3D:
             return desc.image_width <= mInfo.mImage3D_MaxWidth &&
                    desc.image_height <= mInfo.mImage3D_MaxHeight &&
                    desc.image_depth <= mInfo.mImage3D_MaxDepth;
-        case CL_MEM_OBJECT_IMAGE1D_ARRAY:
+        case MemObjectType::Image1D_Array:
             return desc.image_width <= mInfo.mImage2D_MaxWidth &&
                    desc.image_array_size <= mInfo.mImageMaxArraySize;
-        case CL_MEM_OBJECT_IMAGE2D_ARRAY:
+        case MemObjectType::Image2D_Array:
             return desc.image_width <= mInfo.mImage2D_MaxWidth &&
                    desc.image_height <= mInfo.mImage2D_MaxHeight &&
                    desc.image_array_size <= mInfo.mImageMaxArraySize;
-        case CL_MEM_OBJECT_IMAGE1D_BUFFER:
+        case MemObjectType::Image1D_Buffer:
             return desc.image_width <= mInfo.mImageMaxBufferSize;
         default:
             ASSERT(false);
@@ -419,21 +419,21 @@ bool Device::supportsImageDimensions(const ImageDescriptor &desc) const
 {
     switch (desc.type)
     {
-        case CL_MEM_OBJECT_IMAGE1D:
+        case MemObjectType::Image1D:
             return desc.width <= mInfo.mImage2D_MaxWidth;
-        case CL_MEM_OBJECT_IMAGE2D:
+        case MemObjectType::Image2D:
             return desc.width <= mInfo.mImage2D_MaxWidth && desc.height <= mInfo.mImage2D_MaxHeight;
-        case CL_MEM_OBJECT_IMAGE3D:
+        case MemObjectType::Image3D:
             return desc.width <= mInfo.mImage3D_MaxWidth &&
                    desc.height <= mInfo.mImage3D_MaxHeight && desc.depth <= mInfo.mImage3D_MaxDepth;
-        case CL_MEM_OBJECT_IMAGE1D_ARRAY:
+        case MemObjectType::Image1D_Array:
             return desc.width <= mInfo.mImage2D_MaxWidth &&
                    desc.arraySize <= mInfo.mImageMaxArraySize;
-        case CL_MEM_OBJECT_IMAGE2D_ARRAY:
+        case MemObjectType::Image2D_Array:
             return desc.width <= mInfo.mImage2D_MaxWidth &&
                    desc.height <= mInfo.mImage2D_MaxHeight &&
                    desc.arraySize <= mInfo.mImageMaxArraySize;
-        case CL_MEM_OBJECT_IMAGE1D_BUFFER:
+        case MemObjectType::Image1D_Buffer:
             return desc.width <= mInfo.mImageMaxBufferSize;
         default:
             ASSERT(false);

src/libANGLE/CLEvent.cpp

@@ -106,6 +106,14 @@ cl_int Event::setCallback(cl_int commandExecCallbackType, EventCB pfnNotify, voi
     return CL_SUCCESS;
 }
 
+cl_int Event::getProfilingInfo(ProfilingInfo name,
+                               size_t valueSize,
+                               void *value,
+                               size_t *valueSizeRet)
+{
+    return mImpl->getProfilingInfo(name, valueSize, value, valueSizeRet);
+}
+
 Event::~Event() = default;
 
 void Event::callback(cl_int commandStatus)

src/libANGLE/CLEvent.h

@@ -28,6 +28,11 @@ class Event final : public _cl_event, public Object
 
     cl_int setCallback(cl_int commandExecCallbackType, EventCB pfnNotify, void *userData);
 
+    cl_int getProfilingInfo(ProfilingInfo name,
+                            size_t valueSize,
+                            void *value,
+                            size_t *valueSizeRet);
+
   public:
     ~Event() override;
 

src/libANGLE/CLImage.cpp

@@ -93,20 +93,16 @@ cl_int Image::getInfo(ImageInfo name, size_t valueSize, void *value, size_t *val
     return CL_SUCCESS;
 }
 
-bool Image::IsValid(const _cl_mem *image)
+bool Image::IsTypeValid(MemObjectType imageType)
 {
-    if (!Memory::IsValid(image))
+    switch (imageType)
     {
-        return false;
-    }
-    switch (image->cast<Memory>().getType())
-    {
-        case CL_MEM_OBJECT_IMAGE1D:
-        case CL_MEM_OBJECT_IMAGE2D:
-        case CL_MEM_OBJECT_IMAGE3D:
-        case CL_MEM_OBJECT_IMAGE1D_ARRAY:
-        case CL_MEM_OBJECT_IMAGE2D_ARRAY:
-        case CL_MEM_OBJECT_IMAGE1D_BUFFER:
+        case MemObjectType::Image1D:
+        case MemObjectType::Image2D:
+        case MemObjectType::Image3D:
+        case MemObjectType::Image1D_Array:
+        case MemObjectType::Image2D_Array:
+        case MemObjectType::Image1D_Buffer:
             break;
         default:
             return false;
@@ -120,17 +116,17 @@ bool Image::isRegionValid(const size_t origin[3], const size_t region[3]) const
 {
     switch (getType())
     {
-        case CL_MEM_OBJECT_IMAGE1D:
-        case CL_MEM_OBJECT_IMAGE1D_BUFFER:
+        case MemObjectType::Image1D:
+        case MemObjectType::Image1D_Buffer:
             return origin[0] + region[0] <= mDesc.width;
-        case CL_MEM_OBJECT_IMAGE2D:
+        case MemObjectType::Image2D:
             return origin[0] + region[0] <= mDesc.width && origin[1] + region[1] <= mDesc.height;
-        case CL_MEM_OBJECT_IMAGE3D:
+        case MemObjectType::Image3D:
             return origin[0] + region[0] <= mDesc.width && origin[1] + region[1] <= mDesc.height &&
                    origin[2] + region[2] <= mDesc.depth;
-        case CL_MEM_OBJECT_IMAGE1D_ARRAY:
+        case MemObjectType::Image1D_Array:
             return origin[0] + region[0] <= mDesc.width && origin[1] + region[1] <= mDesc.arraySize;
-        case CL_MEM_OBJECT_IMAGE2D_ARRAY:
+        case MemObjectType::Image2D_Array:
             return origin[0] + region[0] <= mDesc.width && origin[1] + region[1] <= mDesc.height &&
                    origin[2] + region[2] <= mDesc.arraySize;
         default:

src/libANGLE/CLImage.h

@@ -22,12 +22,13 @@ class Image final : public Memory
 
     cl_int getInfo(ImageInfo name, size_t valueSize, void *value, size_t *valueSizeRet) const;
 
+    static bool IsTypeValid(MemObjectType imageType);
     static bool IsValid(const _cl_mem *image);
 
   public:
     ~Image() override;
 
-    cl_mem_object_type getType() const final;
+    MemObjectType getType() const final;
 
     const cl_image_format &getFormat() const;
     const ImageDescriptor &getDescriptor() const;
@@ -54,7 +55,12 @@ class Image final : public Memory
     friend class Object;
 };
 
-inline cl_mem_object_type Image::getType() const
+inline bool Image::IsValid(const _cl_mem *image)
+{
+    return Memory::IsValid(image) && IsTypeValid(image->cast<Memory>().getType());
+}
+
+inline MemObjectType Image::getType() const
 {
     return mDesc.type;
 }

src/libANGLE/CLKernel.cpp

@@ -15,6 +15,11 @@
 namespace cl
 {
 
+cl_int Kernel::setArg(cl_uint argIndex, size_t argSize, const void *argValue)
+{
+    return mImpl->setArg(argIndex, argSize, argValue);
+}
+
 cl_int Kernel::getInfo(KernelInfo name, size_t valueSize, void *value, size_t *valueSizeRet) const
 {
     cl_uint valUInt       = 0u;

src/libANGLE/CLKernel.h

@@ -19,6 +19,8 @@ class Kernel final : public _cl_kernel, public Object
   public:
     // Front end entry functions, only called from OpenCL entry points
 
+    cl_int setArg(cl_uint argIndex, size_t argSize, const void *argValue);
+
     cl_int getInfo(KernelInfo name, size_t valueSize, void *value, size_t *valueSizeRet) const;
 
     cl_int getWorkGroupInfo(cl_device_id device,

src/libANGLE/CLMemory.cpp

@@ -15,6 +15,12 @@
 namespace cl
 {
 
+cl_int Memory::setDestructorCallback(MemoryCB pfnNotify, void *userData)
+{
+    mDestructorCallbacks.emplace(pfnNotify, userData);
+    return CL_SUCCESS;
+}
+
 cl_int Memory::getInfo(MemInfo name, size_t valueSize, void *value, size_t *valueSizeRet) const
 {
     static_assert(
@@ -29,7 +35,7 @@ cl_int Memory::getInfo(MemInfo name, size_t valueSize, void *value, size_t *valu
     switch (name)
     {
         case MemInfo::Type:
-            valUInt   = getType();
+            valUInt   = ToCLenum(getType());
             copyValue = &valUInt;
             copySize  = sizeof(valUInt);
             break;
@@ -101,7 +107,16 @@ cl_int Memory::getInfo(MemInfo name, size_t valueSize, void *value, size_t *valu
     return CL_SUCCESS;
 }
 
-Memory::~Memory() = default;
+Memory::~Memory()
+{
+    while (!mDestructorCallbacks.empty())
+    {
+        const MemoryCB callback = mDestructorCallbacks.top().first;
+        void *const userData    = mDestructorCallbacks.top().second;
+        mDestructorCallbacks.pop();
+        callback(this, userData);
+    }
+}
 
 MemFlags Memory::getEffectiveFlags() const
 {

src/libANGLE/CLMemory.h

@@ -12,6 +12,8 @@
 #include "libANGLE/CLObject.h"
 #include "libANGLE/renderer/CLMemoryImpl.h"
 
+#include <stack>
+
 namespace cl
 {
 
@@ -20,6 +22,8 @@ class Memory : public _cl_mem, public Object
   public:
     // Front end entry functions, only called from OpenCL entry points
 
+    cl_int setDestructorCallback(MemoryCB pfnNotify, void *userData);
+
     cl_int getInfo(MemInfo name, size_t valueSize, void *value, size_t *valueSizeRet) const;
 
   public:
@@ -27,7 +31,7 @@ class Memory : public _cl_mem, public Object
 
     ~Memory() override;
 
-    virtual cl_mem_object_type getType() const = 0;
+    virtual MemObjectType getType() const = 0;
 
     const Context &getContext() const;
     const PropArray &getProperties() const;
@@ -42,7 +46,11 @@ class Memory : public _cl_mem, public Object
 
     MemFlags getEffectiveFlags() const;
 
+    static Memory *Cast(cl_mem memobj);
+
   protected:
+    using CallbackData = std::pair<MemoryCB, void *>;
+
     Memory(const Buffer &buffer,
            Context &context,
            PropArray &&properties,
@@ -77,6 +85,7 @@ class Memory : public _cl_mem, public Object
     const rx::CLMemoryImpl::Ptr mImpl;
     const size_t mSize;
 
+    std::stack<CallbackData> mDestructorCallbacks;
     cl_uint mMapCount = 0u;
 
     friend class Buffer;
@@ -124,6 +133,11 @@ inline T &Memory::getImpl() const
     return static_cast<T &>(*mImpl);
 }
 
+inline Memory *Memory::Cast(cl_mem memobj)
+{
+    return static_cast<Memory *>(memobj);
+}
+
 }  // namespace cl
 
 #endif  // LIBANGLE_CLMEMORY_H_

src/libANGLE/CLPlatform.cpp

@@ -70,16 +70,7 @@ void Platform::Initialize(const cl_icd_dispatch &dispatch,
         ERR() << "Already initialized";
         return;
     }
-
-    _cl_platform_id::sDispatch   = &dispatch;
-    _cl_device_id::sDispatch     = &dispatch;
-    _cl_context::sDispatch       = &dispatch;
-    _cl_command_queue::sDispatch = &dispatch;
-    _cl_mem::sDispatch           = &dispatch;
-    _cl_program::sDispatch       = &dispatch;
-    _cl_kernel::sDispatch        = &dispatch;
-    _cl_event::sDispatch         = &dispatch;
-    _cl_sampler::sDispatch       = &dispatch;
+    Dispatch::sDispatch = &dispatch;
 
     platforms.reserve(createFuncs.size());
     while (!createFuncs.empty())

src/libANGLE/CLtypes.h

@@ -17,6 +17,7 @@
 
 // Include frequently used standard headers
 #include <algorithm>
+#include <array>
 #include <functional>
 #include <list>
 #include <memory>
@@ -61,7 +62,7 @@ using ProgramPtrs  = std::vector<ProgramPtr>;
 
 struct ImageDescriptor
 {
-    cl_mem_object_type type;
+    MemObjectType type;
     size_t width;
     size_t height;
     size_t depth;

src/libANGLE/renderer/CLContextImpl.h

@@ -41,6 +41,12 @@ class CLContextImpl : angle::NonCopyable
                                           void *hostPtr,
                                           cl_int &errorCode) = 0;
 
+    virtual cl_int getSupportedImageFormats(cl::MemFlags flags,
+                                            cl::MemObjectType imageType,
+                                            cl_uint numEntries,
+                                            cl_image_format *imageFormats,
+                                            cl_uint *numImageFormats) = 0;
+
     virtual CLSamplerImpl::Ptr createSampler(const cl::Sampler &sampler, cl_int &errorCode) = 0;
 
     virtual CLProgramImpl::Ptr createProgramWithSource(const cl::Program &program,

src/libANGLE/renderer/CLEventImpl.h

@@ -28,6 +28,11 @@ class CLEventImpl : angle::NonCopyable
 
     virtual cl_int setCallback(cl::Event &event, cl_int commandExecCallbackType) = 0;
 
+    virtual cl_int getProfilingInfo(cl::ProfilingInfo name,
+                                    size_t valueSize,
+                                    void *value,
+                                    size_t *valueSizeRet) = 0;
+
   protected:
     const cl::Event &mEvent;
 };

src/libANGLE/renderer/CLKernelImpl.h

@@ -31,12 +31,12 @@ class CLKernelImpl : angle::NonCopyable
         WorkGroupInfo(WorkGroupInfo &&);
         WorkGroupInfo &operator=(WorkGroupInfo &&);
 
-        size_t mGlobalWorkSize[3]         = {};
-        size_t mWorkGroupSize             = 0u;
-        size_t mCompileWorkGroupSize[3]   = {};
-        cl_ulong mLocalMemSize            = 0u;
-        size_t mPrefWorkGroupSizeMultiple = 0u;
-        cl_ulong mPrivateMemSize          = 0u;
+        std::array<size_t, 3u> mGlobalWorkSize       = {};
+        size_t mWorkGroupSize                        = 0u;
+        std::array<size_t, 3u> mCompileWorkGroupSize = {};
+        cl_ulong mLocalMemSize                       = 0u;
+        size_t mPrefWorkGroupSizeMultiple            = 0u;
+        cl_ulong mPrivateMemSize                     = 0u;
     };
 
     struct ArgInfo
@@ -83,6 +83,8 @@ class CLKernelImpl : angle::NonCopyable
     CLKernelImpl(const cl::Kernel &kernel);
     virtual ~CLKernelImpl();
 
+    virtual cl_int setArg(cl_uint argIndex, size_t argSize, const void *argValue) = 0;
+
     virtual Info createInfo(cl_int &errorCode) const = 0;
 
   protected:

src/libANGLE/renderer/cl/CLCommandQueueCL.cpp

@@ -7,11 +7,14 @@
 
 #include "libANGLE/renderer/cl/CLCommandQueueCL.h"
 
+#include "libANGLE/renderer/cl/CLContextCL.h"
 #include "libANGLE/renderer/cl/CLEventCL.h"
 #include "libANGLE/renderer/cl/CLKernelCL.h"
 #include "libANGLE/renderer/cl/CLMemoryCL.h"
 
 #include "libANGLE/CLBuffer.h"
+#include "libANGLE/CLCommandQueue.h"
+#include "libANGLE/CLContext.h"
 #include "libANGLE/CLImage.h"
 #include "libANGLE/CLKernel.h"
 #include "libANGLE/CLMemory.h"
@@ -36,10 +39,24 @@ void CheckCreateEvent(cl_int errorCode, cl_event nativeEvent, CLEventImpl::Creat
 
 CLCommandQueueCL::CLCommandQueueCL(const cl::CommandQueue &commandQueue, cl_command_queue native)
     : CLCommandQueueImpl(commandQueue), mNative(native)
-{}
+{
+    if (commandQueue.getProperties().isSet(CL_QUEUE_ON_DEVICE))
+    {
+        commandQueue.getContext().getImpl<CLContextCL>().mDeviceQueues.emplace(
+            commandQueue.getNative());
+    }
+}
 
 CLCommandQueueCL::~CLCommandQueueCL()
 {
+    if (mCommandQueue.getProperties().isSet(CL_QUEUE_ON_DEVICE))
+    {
+        const size_t numRemoved =
+            mCommandQueue.getContext().getImpl<CLContextCL>().mDeviceQueues.erase(
+                mCommandQueue.getNative());
+        ASSERT(numRemoved == 1u);
+    }
+
     if (mNative->getDispatch().clReleaseCommandQueue(mNative) != CL_SUCCESS)
     {
         ERR() << "Error while releasing CL command-queue";

src/libANGLE/renderer/cl/CLCommandQueueCL.h

@@ -20,6 +20,8 @@ class CLCommandQueueCL : public CLCommandQueueImpl
     CLCommandQueueCL(const cl::CommandQueue &commandQueue, cl_command_queue native);
     ~CLCommandQueueCL() override;
 
+    cl_command_queue getNative() const;
+
     cl_int setProperty(cl::CommandQueueProperties properties, cl_bool enable) override;
 
     cl_int enqueueReadBuffer(const cl::Buffer &buffer,
@@ -212,6 +214,11 @@ class CLCommandQueueCL : public CLCommandQueueImpl
     const cl_command_queue mNative;
 };
 
+inline cl_command_queue CLCommandQueueCL::getNative() const
+{
+    return mNative;
+}
+
 }  // namespace rx
 
 #endif  // LIBANGLE_RENDERER_CL_CLCOMMANDQUEUECL_H_

src/libANGLE/renderer/cl/CLContextCL.cpp

@@ -135,12 +135,12 @@ CLMemoryImpl::Ptr CLContextCL::createImage(const cl::Image &image,
 
     if (mContext.getPlatform().isVersionOrNewer(1u, 2u))
     {
+        const cl_mem_object_type nativeType = cl::ToCLenum(desc.type);
+        const cl_mem nativeParent =
+            image.getParent() ? image.getParent()->getImpl<CLMemoryCL>().getNative() : nullptr;
         const cl_image_desc nativeDesc = {
-            desc.type,       desc.width,
-            desc.height,     desc.depth,
-            desc.arraySize,  desc.rowPitch,
-            desc.slicePitch, desc.numMipLevels,
-            desc.numSamples, {cl::Memory::CastNative(image.getParent().get())}};
+            nativeType,    desc.width,      desc.height,       desc.depth,      desc.arraySize,
+            desc.rowPitch, desc.slicePitch, desc.numMipLevels, desc.numSamples, {nativeParent}};
 
         if (image.getProperties().empty())
         {
@@ -158,12 +158,12 @@ CLMemoryImpl::Ptr CLContextCL::createImage(const cl::Image &image,
     {
         switch (desc.type)
         {
-            case CL_MEM_OBJECT_IMAGE2D:
+            case cl::MemObjectType::Image2D:
                 nativeImage = mNative->getDispatch().clCreateImage2D(
                     mNative, image.getFlags().get(), &format, desc.width, desc.height,
                     desc.rowPitch, hostPtr, &errorCode);
                 break;
-            case CL_MEM_OBJECT_IMAGE3D:
+            case cl::MemObjectType::Image3D:
                 nativeImage = mNative->getDispatch().clCreateImage3D(
                     mNative, image.getFlags().get(), &format, desc.width, desc.height, desc.depth,
                     desc.rowPitch, desc.slicePitch, hostPtr, &errorCode);
@@ -178,6 +178,16 @@ CLMemoryImpl::Ptr CLContextCL::createImage(const cl::Image &image,
     return CLMemoryImpl::Ptr(nativeImage != nullptr ? new CLMemoryCL(image, nativeImage) : nullptr);
 }
 
+cl_int CLContextCL::getSupportedImageFormats(cl::MemFlags flags,
+                                             cl::MemObjectType imageType,
+                                             cl_uint numEntries,
+                                             cl_image_format *imageFormats,
+                                             cl_uint *numImageFormats)
+{
+    return mNative->getDispatch().clGetSupportedImageFormats(
+        mNative, flags.get(), cl::ToCLenum(imageType), numEntries, imageFormats, numImageFormats);
+}
+
 CLSamplerImpl::Ptr CLContextCL::createSampler(const cl::Sampler &sampler, cl_int &errorCode)
 {
     cl_sampler nativeSampler = nullptr;

src/libANGLE/renderer/cl/CLContextCL.h

@@ -8,8 +8,12 @@
 #ifndef LIBANGLE_RENDERER_CL_CLCONTEXTCL_H_
 #define LIBANGLE_RENDERER_CL_CLCONTEXTCL_H_
 
+#include "libANGLE/renderer/cl/cl_types.h"
+
 #include "libANGLE/renderer/CLContextImpl.h"
 
+#include <unordered_set>
+
 namespace rx
 {
 
@@ -19,6 +23,10 @@ class CLContextCL : public CLContextImpl
     CLContextCL(const cl::Context &context, cl_context native);
     ~CLContextCL() override;
 
+    bool hasMemory(cl_mem memory) const;
+    bool hasSampler(cl_sampler sampler) const;
+    bool hasDeviceQueue(cl_command_queue queue) const;
+
     cl::DevicePtrs getDevices(cl_int &errorCode) const override;
 
     CLCommandQueueImpl::Ptr createCommandQueue(const cl::CommandQueue &commandQueue,
@@ -35,6 +43,12 @@ class CLContextCL : public CLContextImpl
                                   void *hostPtr,
                                   cl_int &errorCode) override;
 
+    cl_int getSupportedImageFormats(cl::MemFlags flags,
+                                    cl::MemObjectType imageType,
+                                    cl_uint numEntries,
+                                    cl_image_format *imageFormats,
+                                    cl_uint *numImageFormats) override;
+
     CLSamplerImpl::Ptr createSampler(const cl::Sampler &sampler, cl_int &errorCode) override;
 
     CLProgramImpl::Ptr createProgramWithSource(const cl::Program &program,
@@ -69,8 +83,31 @@ class CLContextCL : public CLContextImpl
 
   private:
     const cl_context mNative;
+
+    std::unordered_set<const _cl_mem *> mMemories;
+    std::unordered_set<const _cl_sampler *> mSamplers;
+    std::unordered_set<const _cl_command_queue *> mDeviceQueues;
+
+    friend class CLCommandQueueCL;
+    friend class CLMemoryCL;
+    friend class CLSamplerCL;
 };
 
+inline bool CLContextCL::hasMemory(cl_mem memory) const
+{
+    return mMemories.find(memory) != mMemories.cend();
+}
+
+inline bool CLContextCL::hasSampler(cl_sampler sampler) const
+{
+    return mSamplers.find(sampler) != mSamplers.cend();
+}
+
+inline bool CLContextCL::hasDeviceQueue(cl_command_queue queue) const
+{
+    return mDeviceQueues.find(queue) != mDeviceQueues.cend();
+}
+
 }  // namespace rx
 
 #endif  // LIBANGLE_RENDERER_CL_CLCONTEXTCL_H_

src/libANGLE/renderer/cl/CLDeviceCL.cpp

@@ -101,6 +101,14 @@ CLDeviceImpl::Info CLDeviceCL::createInfo(cl::DeviceType type) const
         return Info{};
     }
     info.mVersionStr.assign(valString.data());
+
+    // Limit version number to supported version
+    if (info.mVersionStr[7] != '1')
+    {
+        info.mVersionStr[7] = '1';
+        info.mVersionStr[9] = '2';
+    }
+
     info.mVersion = ExtractCLVersion(info.mVersionStr);
     if (info.mVersion == 0u)
     {

src/libANGLE/renderer/cl/CLEventCL.cpp

@@ -41,6 +41,15 @@ cl_int CLEventCL::setCallback(cl::Event &event, cl_int commandExecCallbackType)
                                                      &event);
 }
 
+cl_int CLEventCL::getProfilingInfo(cl::ProfilingInfo name,
+                                   size_t valueSize,
+                                   void *value,
+                                   size_t *valueSizeRet)
+{
+    return mNative->getDispatch().clGetEventProfilingInfo(mNative, cl::ToCLenum(name), valueSize,
+                                                          value, valueSizeRet);
+}
+
 std::vector<cl_event> CLEventCL::Cast(const cl::EventPtrs &events)
 {
     std::vector<cl_event> nativeEvents;

src/libANGLE/renderer/cl/CLEventCL.h

@@ -29,6 +29,11 @@ class CLEventCL : public CLEventImpl
 
     cl_int setCallback(cl::Event &event, cl_int commandExecCallbackType) override;
 
+    cl_int getProfilingInfo(cl::ProfilingInfo name,
+                            size_t valueSize,
+                            void *value,
+                            size_t *valueSizeRet) override;
+
     static std::vector<cl_event> Cast(const cl::EventPtrs &events);
 
   private:

src/libANGLE/renderer/cl/CLKernelCL.cpp

@@ -7,13 +7,19 @@
 
 #include "libANGLE/renderer/cl/CLKernelCL.h"
 
+#include "libANGLE/renderer/cl/CLCommandQueueCL.h"
+#include "libANGLE/renderer/cl/CLContextCL.h"
 #include "libANGLE/renderer/cl/CLDeviceCL.h"
+#include "libANGLE/renderer/cl/CLMemoryCL.h"
+#include "libANGLE/renderer/cl/CLSamplerCL.h"
 
+#include "libANGLE/CLCommandQueue.h"
 #include "libANGLE/CLContext.h"
 #include "libANGLE/CLKernel.h"
+#include "libANGLE/CLMemory.h"
 #include "libANGLE/CLPlatform.h"
 #include "libANGLE/CLProgram.h"
-#include "libANGLE/Debug.h"
+#include "libANGLE/CLSampler.h"
 
 namespace rx
 {
@@ -42,7 +48,11 @@ bool GetArgInfo(cl_kernel kernel,
 {
     errorCode = kernel->getDispatch().clGetKernelArgInfo(kernel, index, cl::ToCLenum(name),
                                                          sizeof(T), &value, nullptr);
-    return errorCode == CL_SUCCESS || errorCode == CL_KERNEL_ARG_INFO_NOT_AVAILABLE;
+    if (errorCode == CL_KERNEL_ARG_INFO_NOT_AVAILABLE)
+    {
+        errorCode = CL_SUCCESS;
+    }
+    return errorCode == CL_SUCCESS;
 }
 
 template <typename T>
@@ -64,6 +74,7 @@ bool GetArgString(cl_kernel kernel,
                                                          nullptr, &size);
     if (errorCode == CL_KERNEL_ARG_INFO_NOT_AVAILABLE)
     {
+        errorCode = CL_SUCCESS;
         return true;
     }
     else if (errorCode != CL_SUCCESS)
@@ -115,6 +126,47 @@ CLKernelCL::~CLKernelCL()
     }
 }
 
+cl_int CLKernelCL::setArg(cl_uint argIndex, size_t argSize, const void *argValue)
+{
+    void *value = nullptr;
+    if (argValue != nullptr)
+    {
+        // If argument is a CL object, fetch the mapped value
+        const CLContextCL &ctx = mKernel.getProgram().getContext().getImpl<CLContextCL>();
+        if (argSize == sizeof(cl_mem))
+        {
+            cl_mem memory = *static_cast<const cl_mem *>(argValue);
+            if (ctx.hasMemory(memory))
+            {
+                value = memory->cast<cl::Memory>().getImpl<CLMemoryCL>().getNative();
+            }
+        }
+        if (value == nullptr && argSize == sizeof(cl_sampler))
+        {
+            cl_sampler sampler = *static_cast<const cl_sampler *>(argValue);
+            if (ctx.hasSampler(sampler))
+            {
+                value = sampler->cast<cl::Sampler>().getImpl<CLSamplerCL>().getNative();
+            }
+        }
+        if (value == nullptr && argSize == sizeof(cl_command_queue))
+        {
+            cl_command_queue queue = *static_cast<const cl_command_queue *>(argValue);
+            if (ctx.hasDeviceQueue(queue))
+            {
+                value = queue->cast<cl::CommandQueue>().getImpl<CLCommandQueueCL>().getNative();
+            }
+        }
+    }
+
+    // If mapped value was found, use it instead of original value
+    if (value != nullptr)
+    {
+        argValue = &value;
+    }
+    return mNative->getDispatch().clSetKernelArg(mNative, argIndex, argSize, argValue);
+}
+
 CLKernelImpl::Info CLKernelCL::createInfo(cl_int &errorCode) const
 {
     const cl::Context &ctx = mKernel.getProgram().getContext();
@@ -133,7 +185,9 @@ CLKernelImpl::Info CLKernelCL::createInfo(cl_int &errorCode) const
     {
         const cl_device_id device = ctx.getDevices()[index]->getImpl<CLDeviceCL>().getNative();
         WorkGroupInfo &workGroup  = info.mWorkGroups[index];
+
         if ((ctx.getPlatform().isVersionOrNewer(1u, 2u) &&
+             ctx.getDevices()[index]->supportsBuiltInKernel(info.mFunctionName) &&
              !GetWorkGroupInfo(mNative, device, cl::KernelWorkGroupInfo::GlobalWorkSize,
                                workGroup.mGlobalWorkSize, errorCode)) ||
             !GetWorkGroupInfo(mNative, device, cl::KernelWorkGroupInfo::WorkGroupSize,

src/libANGLE/renderer/cl/CLKernelCL.h

@@ -23,6 +23,8 @@ class CLKernelCL : public CLKernelImpl
 
     cl_kernel getNative() const;
 
+    cl_int setArg(cl_uint argIndex, size_t argSize, const void *argValue) override;
+
     Info createInfo(cl_int &errorCode) const override;
 
   private:

src/libANGLE/renderer/cl/CLMemoryCL.cpp

@@ -7,17 +7,26 @@
 
 #include "libANGLE/renderer/cl/CLMemoryCL.h"
 
+#include "libANGLE/renderer/cl/CLContextCL.h"
+
 #include "libANGLE/CLBuffer.h"
+#include "libANGLE/CLContext.h"
 
 namespace rx
 {
 
 CLMemoryCL::CLMemoryCL(const cl::Memory &memory, cl_mem native)
     : CLMemoryImpl(memory), mNative(native)
-{}
+{
+    memory.getContext().getImpl<CLContextCL>().mMemories.emplace(memory.getNative());
+}
 
 CLMemoryCL::~CLMemoryCL()
 {
+    const size_t numRemoved =
+        mMemory.getContext().getImpl<CLContextCL>().mMemories.erase(mMemory.getNative());
+    ASSERT(numRemoved == 1u);
+
     if (mNative->getDispatch().clReleaseMemObject(mNative) != CL_SUCCESS)
     {
         ERR() << "Error while releasing CL memory object";

src/libANGLE/renderer/cl/CLPlatformCL.cpp

@@ -129,6 +129,13 @@ CLPlatformImpl::Info CLPlatformCL::createInfo() const
     info.mName               = GetPlatformString(mNative, cl::PlatformInfo::Name);
     info.mExtensions         = GetPlatformString(mNative, cl::PlatformInfo::Extensions);
 
+    // Limit version number to supported version
+    if (info.mVersionStr[7] != '1')
+    {
+        info.mVersionStr[7] = '1';
+        info.mVersionStr[9] = '2';
+    }
+
     if (vendor.empty() || info.mProfile.empty() || info.mVersionStr.empty() || info.mName.empty() ||
         info.mExtensions.empty())
     {

src/libANGLE/renderer/cl/CLProgramCL.cpp

@@ -134,7 +134,7 @@ cl_int CLProgramCL::createKernels(cl_uint numKernels,
 
     std::vector<cl_kernel> nativeKernels(numKernels, nullptr);
     const cl_int errorCode = mNative->getDispatch().clCreateKernelsInProgram(
-        mNative, numKernels, nativeKernels.data(), nullptr);
+        mNative, numKernels, nativeKernels.data(), numKernelsRet);
     if (errorCode == CL_SUCCESS)
     {
         for (cl_kernel nativeKernel : nativeKernels)

src/libANGLE/renderer/cl/CLSamplerCL.cpp

@@ -7,17 +7,26 @@
 
 #include "libANGLE/renderer/cl/CLSamplerCL.h"
 
-#include "libANGLE/Debug.h"
+#include "libANGLE/renderer/cl/CLContextCL.h"
+
+#include "libANGLE/CLContext.h"
+#include "libANGLE/CLSampler.h"
 
 namespace rx
 {
 
 CLSamplerCL::CLSamplerCL(const cl::Sampler &sampler, cl_sampler native)
     : CLSamplerImpl(sampler), mNative(native)
-{}
+{
+    sampler.getContext().getImpl<CLContextCL>().mSamplers.emplace(sampler.getNative());
+}
 
 CLSamplerCL::~CLSamplerCL()
 {
+    const size_t numRemoved =
+        mSampler.getContext().getImpl<CLContextCL>().mSamplers.erase(mSampler.getNative());
+    ASSERT(numRemoved == 1u);
+
     if (mNative->getDispatch().clReleaseSampler(mNative) != CL_SUCCESS)
     {
         ERR() << "Error while releasing CL sampler";

src/libANGLE/renderer/cl/CLSamplerCL.h

@@ -21,10 +21,17 @@ class CLSamplerCL : public CLSamplerImpl
     CLSamplerCL(const cl::Sampler &sampler, cl_sampler native);
     ~CLSamplerCL() override;
 
+    cl_sampler getNative() const;
+
   private:
     const cl_sampler mNative;
 };
 
+inline cl_sampler CLSamplerCL::getNative() const
+{
+    return mNative;
+}
+
 }  // namespace rx
 
 #endif  // LIBANGLE_RENDERER_CL_CLSAMPLERCL_H_

src/libANGLE/renderer/cl/cl_types.h

@@ -13,9 +13,15 @@
 namespace rx
 {
 
+class CLCommandQueueCL;
 class CLContextCL;
 class CLDeviceCL;
+class CLEventCL;
+class CLKernelCL;
+class CLMemoryCL;
 class CLPlatformCL;
+class CLProgramCL;
+class CLSamplerCL;
 
 }  // namespace rx
 

src/libGLESv2/cl_stubs.cpp

@@ -309,8 +309,8 @@ cl_int GetSupportedImageFormats(cl_context context,
                                 cl_image_format *image_formats,
                                 cl_uint *num_image_formats)
 {
-    WARN_NOT_SUPPORTED(GetSupportedImageFormats);
-    return 0;
+    return context->cast<Context>().getSupportedImageFormats(flags, image_type, num_entries,
+                                                             image_formats, num_image_formats);
 }
 
 cl_int GetMemObjectInfo(cl_mem memobj,
@@ -347,8 +347,7 @@ cl_int SetMemObjectDestructorCallback(cl_mem memobj,
                                       void(CL_CALLBACK *pfn_notify)(cl_mem memobj, void *user_data),
                                       void *user_data)
 {
-    WARN_NOT_SUPPORTED(SetMemObjectDestructorCallback);
-    return 0;
+    return memobj->cast<Memory>().setDestructorCallback(pfn_notify, user_data);
 }
 
 void *SVMAlloc(cl_context context, SVM_MemFlags flags, size_t size, cl_uint alignment)
@@ -567,8 +566,7 @@ cl_int ReleaseKernel(cl_kernel kernel)
 
 cl_int SetKernelArg(cl_kernel kernel, cl_uint arg_index, size_t arg_size, const void *arg_value)
 {
-    WARN_NOT_SUPPORTED(SetKernelArg);
-    return 0;
+    return kernel->cast<Kernel>().setArg(arg_index, arg_size, arg_value);
 }
 
 cl_int SetKernelArgSVMPointer(cl_kernel kernel, cl_uint arg_index, const void *arg_value)
@@ -688,8 +686,8 @@ cl_int GetEventProfilingInfo(cl_event event,
                              void *param_value,
                              size_t *param_value_size_ret)
 {
-    WARN_NOT_SUPPORTED(GetEventProfilingInfo);
-    return 0;
+    return event->cast<Event>().getProfilingInfo(param_name, param_value_size, param_value,
+                                                 param_value_size_ret);
 }
 
 cl_int Flush(cl_command_queue command_queue)