From eedf013384faade2d092174029a871442ef9c36a Mon Sep 17 00:00:00 2001 From: foreman Date: Thu, 24 May 2018 10:45:00 -0400 Subject: [PATCH] P4 to Git Change 1559096 by gandryey@gera-w8 on 2018/05/24 10:34:47 SWDEV-79445 - OCL generic changes and code clean-up - Remove obsolete logic for aliased kernels, since it was a pre-SI optimization Affected files ... ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.cpp#418 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.cpp#98 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocvirtual.cpp#52 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/kernel.cpp#27 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/kernel.hpp#21 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/program.cpp#93 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/program.hpp#44 edit [ROCm/clr commit: bcea58f07f6bfd6003c06d6607138b39e685dab3] --- .../clr/rocclr/runtime/device/gpu/gpuvirtual.cpp | 8 +++----- .../clr/rocclr/runtime/device/pal/palvirtual.cpp | 3 +-- .../clr/rocclr/runtime/device/rocm/rocvirtual.cpp | 3 +-- projects/clr/rocclr/runtime/platform/kernel.cpp | 4 ++-- projects/clr/rocclr/runtime/platform/kernel.hpp | 3 +-- projects/clr/rocclr/runtime/platform/program.cpp | 13 ++++--------- projects/clr/rocclr/runtime/platform/program.hpp | 7 ++----- 7 files changed, 14 insertions(+), 27 deletions(-) diff --git a/projects/clr/rocclr/runtime/device/gpu/gpuvirtual.cpp b/projects/clr/rocclr/runtime/device/gpu/gpuvirtual.cpp index d435481061..9382b55e49 100644 --- a/projects/clr/rocclr/runtime/device/gpu/gpuvirtual.cpp +++ b/projects/clr/rocclr/runtime/device/gpu/gpuvirtual.cpp @@ -1879,9 +1879,8 @@ bool VirtualGPU::submitKernelInternal(const amd::NDRangeContainer& sizes, const GpuEvent gpuEvent; groupOffset = 0; - // Get the GPU kernel object with optimization enabled - bool noAlias = true; - device::Kernel* devKernel = const_cast(kernel.getDeviceKernel(dev(), noAlias)); + // Get the GPU kernel object + device::Kernel* devKernel = const_cast(kernel.getDeviceKernel(dev())); Kernel& gpuKernelOpt = static_cast(*devKernel); if (gpuKernelOpt.hsa()) { @@ -2922,9 +2921,8 @@ void VirtualGPU::profileEvent(EngineType engine, bool type) const { bool VirtualGPU::processMemObjectsHSA(const amd::Kernel& kernel, const_address params, bool nativeMem, std::vector* memList) { - static const bool NoAlias = true; const HSAILKernel& hsaKernel = - static_cast(*(kernel.getDeviceKernel(dev(), NoAlias))); + static_cast(*(kernel.getDeviceKernel(dev()))); const amd::KernelSignature& signature = kernel.signature(); const amd::KernelParameters& kernelParams = kernel.parameters(); diff --git a/projects/clr/rocclr/runtime/device/pal/palvirtual.cpp b/projects/clr/rocclr/runtime/device/pal/palvirtual.cpp index a83e969a97..7b7b8dced9 100644 --- a/projects/clr/rocclr/runtime/device/pal/palvirtual.cpp +++ b/projects/clr/rocclr/runtime/device/pal/palvirtual.cpp @@ -2905,9 +2905,8 @@ void VirtualGPU::profileEvent(EngineType engine, bool type) const { bool VirtualGPU::processMemObjectsHSA(const amd::Kernel& kernel, const_address params, bool nativeMem) { - static const bool NoAlias = true; const HSAILKernel& hsaKernel = - static_cast(*(kernel.getDeviceKernel(dev(), NoAlias))); + static_cast(*(kernel.getDeviceKernel(dev()))); const amd::KernelSignature& signature = kernel.signature(); const amd::KernelParameters& kernelParams = kernel.parameters(); std::vector memList; diff --git a/projects/clr/rocclr/runtime/device/rocm/rocvirtual.cpp b/projects/clr/rocclr/runtime/device/rocm/rocvirtual.cpp index 501357a5e7..33169dea09 100644 --- a/projects/clr/rocclr/runtime/device/rocm/rocvirtual.cpp +++ b/projects/clr/rocclr/runtime/device/rocm/rocvirtual.cpp @@ -171,8 +171,7 @@ void VirtualGPU::MemoryDependency::clear(bool all) { } bool VirtualGPU::processMemObjects(const amd::Kernel& kernel, const_address params) { - static const bool NoAlias = true; - const Kernel& hsaKernel = static_cast(*(kernel.getDeviceKernel(dev(), NoAlias))); + const Kernel& hsaKernel = static_cast(*(kernel.getDeviceKernel(dev()))); const amd::KernelSignature& signature = kernel.signature(); const amd::KernelParameters& kernelParams = kernel.parameters(); diff --git a/projects/clr/rocclr/runtime/platform/kernel.cpp b/projects/clr/rocclr/runtime/platform/kernel.cpp index 47b4a724a0..8d0005a727 100644 --- a/projects/clr/rocclr/runtime/platform/kernel.cpp +++ b/projects/clr/rocclr/runtime/platform/kernel.cpp @@ -29,8 +29,8 @@ Kernel::~Kernel() { delete parameters_; } -const device::Kernel* Kernel::getDeviceKernel(const Device& device, bool noAlias) const { - return symbol_.getDeviceKernel(device, noAlias); +const device::Kernel* Kernel::getDeviceKernel(const Device& device) const { + return symbol_.getDeviceKernel(device); } const KernelSignature& Kernel::signature() const { return symbol_.signature(); } diff --git a/projects/clr/rocclr/runtime/platform/kernel.hpp b/projects/clr/rocclr/runtime/platform/kernel.hpp index 45403bb335..d18187784d 100644 --- a/projects/clr/rocclr/runtime/platform/kernel.hpp +++ b/projects/clr/rocclr/runtime/platform/kernel.hpp @@ -273,8 +273,7 @@ class Kernel : public RuntimeObject { const KernelSignature& signature() const; //! Return the kernel entry point for the given device. - const device::Kernel* getDeviceKernel(const Device& device, //!< Device object - bool noAlias = true //!< Controls alias optimization + const device::Kernel* getDeviceKernel(const Device& device //!< Device object ) const; //! Return the parameters. diff --git a/projects/clr/rocclr/runtime/platform/program.cpp b/projects/clr/rocclr/runtime/platform/program.cpp index 6338f444ad..8515cf3ada 100644 --- a/projects/clr/rocclr/runtime/platform/program.cpp +++ b/projects/clr/rocclr/runtime/platform/program.cpp @@ -603,22 +603,17 @@ bool Program::ParseAllOptions(const std::string& options, option::Options& parse return amd::option::parseAllOptions(allOpts, parsedOptions, linkOptsOnly); } -bool Symbol::setDeviceKernel(const Device& device, const device::Kernel* func, bool noAlias) { +bool Symbol::setDeviceKernel(const Device& device, const device::Kernel* func) { // FIXME_lmoriche: check that the signatures are compatible if (deviceKernels_.size() == 0) { signature_ = func->signature(); } - - if (noAlias) { - deviceKernels_[&device] = func; - } else { - devKernelsNoOpt_[&device] = func; - } + deviceKernels_[&device] = func; return true; } -const device::Kernel* Symbol::getDeviceKernel(const Device& device, bool noAlias) const { - const devicekernels_t* devKernels = (noAlias) ? &deviceKernels_ : &devKernelsNoOpt_; +const device::Kernel* Symbol::getDeviceKernel(const Device& device) const { + const devicekernels_t* devKernels = &deviceKernels_; const auto itEnd = devKernels->cend(); auto it = devKernels->find(&device); if (it != itEnd) { diff --git a/projects/clr/rocclr/runtime/platform/program.hpp b/projects/clr/rocclr/runtime/platform/program.hpp index 597a03231c..b5b99aa181 100644 --- a/projects/clr/rocclr/runtime/platform/program.hpp +++ b/projects/clr/rocclr/runtime/platform/program.hpp @@ -39,7 +39,6 @@ class Symbol : public HeapObject { private: devicekernels_t deviceKernels_; //! All device kernels objects. - devicekernels_t devKernelsNoOpt_; //! Kernel objects without optimization KernelSignature signature_; //! Kernel signature. public: @@ -48,13 +47,11 @@ class Symbol : public HeapObject { //! Set the entry point and check or set the signature. bool setDeviceKernel(const Device& device, //!< Device object. - const device::Kernel* func, //!< Device kernel object. - bool noAlias = true //!< No-alias optimization + const device::Kernel* func //!< Device kernel object. ); //! Return the device kernel. - const device::Kernel* getDeviceKernel(const Device& device, //!< Device object. - bool noAlias = true //!< No-alias optimization + const device::Kernel* getDeviceKernel(const Device& device //!< Device object. ) const; //! Return this Symbol's signature.