Merge changes Iee153445,Iee274471 am: 79df15ea88 am: 10f298fc41 am: 7cb5001398t_frc_odp_330442040t_frc_odp_330442000t_frc_ase_330444010android-wear-13.0.0-gpl_r3android-wear-13.0.0-gpl_r2android-wear-13.0.0-gpl_r1android-vts-13.0_r8android-vts-13.0_r7android-vts-13.0_r6android-vts-13.0_r5android-vts-13.0_r4android-vts-13.0_r3android-vts-13.0_r2android-t-qpr3-beta-3-gplandroid-t-qpr3-beta-1-gplandroid-t-qpr2-beta-3-gplandroid-t-qpr2-beta-2-gplandroid-t-qpr1-beta-3-gplandroid-t-qpr1-beta-1-gplandroid-cts-13.0_r8android-cts-13.0_r7android-cts-13.0_r6android-cts-13.0_r5android-cts-13.0_r4android-cts-13.0_r3android-cts-13.0_r2android-13.0.0_r83android-13.0.0_r82android-13.0.0_r81android-13.0.0_r80android-13.0.0_r79android-13.0.0_r78android-13.0.0_r77android-13.0.0_r76android-13.0.0_r75android-13.0.0_r74android-13.0.0_r73android-13.0.0_r72android-13.0.0_r71android-13.0.0_r70android-13.0.0_r69android-13.0.0_r68android-13.0.0_r67android-13.0.0_r66android-13.0.0_r65android-13.0.0_r64android-13.0.0_r63android-13.0.0_r62android-13.0.0_r61android-13.0.0_r60android-13.0.0_r59android-13.0.0_r58android-13.0.0_r57android-13.0.0_r56android-13.0.0_r55android-13.0.0_r54android-13.0.0_r53android-13.0.0_r52android-13.0.0_r51android-13.0.0_r50android-13.0.0_r49android-13.0.0_r48android-13.0.0_r47android-13.0.0_r46android-13.0.0_r45android-13.0.0_r44android-13.0.0_r43android-13.0.0_r42android-13.0.0_r41android-13.0.0_r40android-13.0.0_r39android-13.0.0_r38android-13.0.0_r37android-13.0.0_r36android-13.0.0_r35android-13.0.0_r34android-13.0.0_r33android-13.0.0_r32android-13.0.0_r30android-13.0.0_r29android-13.0.0_r28android-13.0.0_r27android-13.0.0_r24android-13.0.0_r23android-13.0.0_r22android-13.0.0_r21android-13.0.0_r20android-13.0.0_r19android-13.0.0_r18android-13.0.0_r17android-13.0.0_r16aml_go_odp_330912000aml_go_ads_330915100aml_go_ads_330915000aml_go_ads_330913000android13-tests-releaseandroid13-tests-devandroid13-qpr3-s9-releaseandroid13-qpr3-s8-releaseandroid13-qpr3-s7-releaseandroid13-qpr3-s6-releaseandroid13-qpr3-s5-releaseandroid13-qpr3-s4-releaseandroid13-qpr3-s3-releaseandroid13-qpr3-s2-releaseandroid13-qpr3-s14-releaseandroid13-qpr3-s13-releaseandroid13-qpr3-s12-releaseandroid13-qpr3-s11-releaseandroid13-qpr3-s10-releaseandroid13-qpr3-s1-releaseandroid13-qpr3-releaseandroid13-qpr3-c-s8-releaseandroid13-qpr3-c-s7-releaseandroid13-qpr3-c-s6-releaseandroid13-qpr3-c-s5-releaseandroid13-qpr3-c-s4-releaseandroid13-qpr3-c-s3-releaseandroid13-qpr3-c-s2-releaseandroid13-qpr3-c-s12-releaseandroid13-qpr3-c-s11-releaseandroid13-qpr3-c-s10-releaseandroid13-qpr3-c-s1-releaseandroid13-qpr2-s9-releaseandroid13-qpr2-s8-releaseandroid13-qpr2-s7-releaseandroid13-qpr2-s6-releaseandroid13-qpr2-s5-releaseandroid13-qpr2-s3-releaseandroid13-qpr2-s2-releaseandroid13-qpr2-s12-releaseandroid13-qpr2-s11-releaseandroid13-qpr2-s10-releaseandroid13-qpr2-s1-releaseandroid13-qpr2-releaseandroid13-qpr2-b-s1-releaseandroid13-qpr1-s8-releaseandroid13-qpr1-s7-releaseandroid13-qpr1-s6-releaseandroid13-qpr1-s5-releaseandroid13-qpr1-s4-releaseandroid13-qpr1-s3-releaseandroid13-qpr1-s2-releaseandroid13-qpr1-s1-releaseandroid13-qpr1-releaseandroid13-mainline-go-adservices-releaseandroid13-frc-odp-releaseandroid13-devandroid13-d4-s2-releaseandroid13-d4-s1-releaseandroid13-d4-releaseandroid13-d3-s1-releaseandroid13-d2-releaseandroid-wear-13.0.0-gpl_r1

Original change: https://android-review.googlesource.com/c/platform/external/eigen/+/1999079

Change-Id: I4c76dc5ddc7fb0ae9fc42436f28bd8bf9de50a97

1 files changed, 176 insertions, 0 deletions

diff --git a/test/gpu_common.h b/test/gpu_common.h
new file mode 100644
index 000000000..c37eaa13f
--- /dev/null
+++ b/test/gpu_common.h
@@ -0,0 +1,176 @@
+#ifndef EIGEN_TEST_GPU_COMMON_H
+#define EIGEN_TEST_GPU_COMMON_H
+
+#ifdef EIGEN_USE_HIP
+  #include <hip/hip_runtime.h>
+  #include <hip/hip_runtime_api.h>
+#else
+  #include <cuda.h>
+  #include <cuda_runtime.h>
+  #include <cuda_runtime_api.h>
+#endif
+
+#include <iostream>
+
+#define EIGEN_USE_GPU
+#include <unsupported/Eigen/CXX11/src/Tensor/TensorGpuHipCudaDefines.h>
+
+#if !defined(__CUDACC__) && !defined(__HIPCC__)
+dim3 threadIdx, blockDim, blockIdx;
+#endif
+
+template<typename Kernel, typename Input, typename Output>
+void run_on_cpu(const Kernel& ker, int n, const Input& in, Output& out)
+{
+  for(int i=0; i<n; i++)
+    ker(i, in.data(), out.data());
+}
+
+
+template<typename Kernel, typename Input, typename Output>
+__global__
+EIGEN_HIP_LAUNCH_BOUNDS_1024
+void run_on_gpu_meta_kernel(const Kernel ker, int n, const Input* in, Output* out)
+{
+  int i = threadIdx.x + blockIdx.x*blockDim.x;
+  if(i<n) {
+    ker(i, in, out);
+  }
+}
+
+
+template<typename Kernel, typename Input, typename Output>
+void run_on_gpu(const Kernel& ker, int n, const Input& in, Output& out)
+{
+  typename Input::Scalar*  d_in;
+  typename Output::Scalar* d_out;
+  std::ptrdiff_t in_bytes  = in.size()  * sizeof(typename Input::Scalar);
+  std::ptrdiff_t out_bytes = out.size() * sizeof(typename Output::Scalar);
+  
+  gpuMalloc((void**)(&d_in),  in_bytes);
+  gpuMalloc((void**)(&d_out), out_bytes);
+  
+  gpuMemcpy(d_in,  in.data(),  in_bytes,  gpuMemcpyHostToDevice);
+  gpuMemcpy(d_out, out.data(), out_bytes, gpuMemcpyHostToDevice);
+  
+  // Simple and non-optimal 1D mapping assuming n is not too large
+  // That's only for unit testing!
+  dim3 Blocks(128);
+  dim3 Grids( (n+int(Blocks.x)-1)/int(Blocks.x) );
+
+  gpuDeviceSynchronize();
+  
+#ifdef EIGEN_USE_HIP
+  hipLaunchKernelGGL(HIP_KERNEL_NAME(run_on_gpu_meta_kernel<Kernel,
+				     typename std::decay<decltype(*d_in)>::type,
+				     typename std::decay<decltype(*d_out)>::type>), 
+		     dim3(Grids), dim3(Blocks), 0, 0, ker, n, d_in, d_out);
+#else
+  run_on_gpu_meta_kernel<<<Grids,Blocks>>>(ker, n, d_in, d_out);
+#endif
+  // Pre-launch errors.
+  gpuError_t err = gpuGetLastError();
+  if (err != gpuSuccess) {
+    printf("%s: %s\n", gpuGetErrorName(err), gpuGetErrorString(err));
+    gpu_assert(false);
+  }
+  
+  // Kernel execution errors.
+  err = gpuDeviceSynchronize();
+  if (err != gpuSuccess) {
+    printf("%s: %s\n", gpuGetErrorName(err), gpuGetErrorString(err));
+    gpu_assert(false);
+  }
+  
+  
+  // check inputs have not been modified
+  gpuMemcpy(const_cast<typename Input::Scalar*>(in.data()),  d_in,  in_bytes,  gpuMemcpyDeviceToHost);
+  gpuMemcpy(out.data(), d_out, out_bytes, gpuMemcpyDeviceToHost);
+  
+  gpuFree(d_in);
+  gpuFree(d_out);
+}
+
+
+template<typename Kernel, typename Input, typename Output>
+void run_and_compare_to_gpu(const Kernel& ker, int n, const Input& in, Output& out)
+{
+  Input  in_ref,  in_gpu;
+  Output out_ref, out_gpu;
+  #if !defined(EIGEN_GPU_COMPILE_PHASE)
+  in_ref = in_gpu = in;
+  out_ref = out_gpu = out;
+  #else
+  EIGEN_UNUSED_VARIABLE(in);
+  EIGEN_UNUSED_VARIABLE(out);
+  #endif
+  run_on_cpu (ker, n, in_ref,  out_ref);
+  run_on_gpu(ker, n, in_gpu, out_gpu);
+  #if !defined(EIGEN_GPU_COMPILE_PHASE)
+  VERIFY_IS_APPROX(in_ref, in_gpu);
+  VERIFY_IS_APPROX(out_ref, out_gpu);
+  #endif
+}
+
+struct compile_time_device_info {
+  EIGEN_DEVICE_FUNC
+  void operator()(int i, const int* /*in*/, int* info) const
+  {
+    if (i == 0) {
+      EIGEN_UNUSED_VARIABLE(info)
+      #if defined(__CUDA_ARCH__)
+      info[0] = int(__CUDA_ARCH__ +0);
+      #endif
+      #if defined(EIGEN_HIP_DEVICE_COMPILE)
+      info[1] = int(EIGEN_HIP_DEVICE_COMPILE +0);
+      #endif
+    }
+  }
+};
+
+void ei_test_init_gpu()
+{
+  int device = 0;
+  gpuDeviceProp_t deviceProp;
+  gpuGetDeviceProperties(&deviceProp, device);
+
+  ArrayXi dummy(1), info(10);
+  info = -1;
+  run_on_gpu(compile_time_device_info(),10,dummy,info);
+
+
+  std::cout << "GPU compile-time info:\n";
+  
+  #ifdef EIGEN_CUDACC
+  std::cout << "  EIGEN_CUDACC:                 " << int(EIGEN_CUDACC) << "\n";
+  #endif
+  
+  #ifdef EIGEN_CUDA_SDK_VER
+  std::cout << "  EIGEN_CUDA_SDK_VER:             " << int(EIGEN_CUDA_SDK_VER) << "\n";
+  #endif
+
+  #ifdef EIGEN_COMP_NVCC
+  std::cout << "  EIGEN_COMP_NVCC:             " << int(EIGEN_COMP_NVCC) << "\n";
+  #endif
+  
+  #ifdef EIGEN_HIPCC
+  std::cout << "  EIGEN_HIPCC:                 " << int(EIGEN_HIPCC) << "\n";
+  #endif
+
+  std::cout << "  EIGEN_CUDA_ARCH:             " << info[0] << "\n";  
+  std::cout << "  EIGEN_HIP_DEVICE_COMPILE:    " << info[1] << "\n";
+
+  std::cout << "GPU device info:\n";
+  std::cout << "  name:                        " << deviceProp.name << "\n";
+  std::cout << "  capability:                  " << deviceProp.major << "." << deviceProp.minor << "\n";
+  std::cout << "  multiProcessorCount:         " << deviceProp.multiProcessorCount << "\n";
+  std::cout << "  maxThreadsPerMultiProcessor: " << deviceProp.maxThreadsPerMultiProcessor << "\n";
+  std::cout << "  warpSize:                    " << deviceProp.warpSize << "\n";
+  std::cout << "  regsPerBlock:                " << deviceProp.regsPerBlock << "\n";
+  std::cout << "  concurrentKernels:           " << deviceProp.concurrentKernels << "\n";
+  std::cout << "  clockRate:                   " << deviceProp.clockRate << "\n";
+  std::cout << "  canMapHostMemory:            " << deviceProp.canMapHostMemory << "\n";
+  std::cout << "  computeMode:                 " << deviceProp.computeMode << "\n";
+}
+
+#endif // EIGEN_TEST_GPU_COMMON_H