diff options
author | Yi Kong <yikong@google.com> | 2022-02-25 17:02:53 +0000 |
---|---|---|
committer | Automerger Merge Worker <android-build-automerger-merge-worker@system.gserviceaccount.com> | 2022-02-25 17:02:53 +0000 |
commit | edb0ad5bb04b48aab7dd0978f0475edd3550de7c (patch) | |
tree | fb979fb4cf4f8052c8cc66b1ec9516d91fcd859b /unsupported/Eigen/CXX11/src/Tensor/TensorDeviceDefault.h | |
parent | 8fd413e275f78a4c240f1442ce5cf77c73a20a55 (diff) | |
parent | bc0f5df265caa21a2120c22453655a7fcc941991 (diff) | |
download | eigen-7ecd5586f9218fe07c97ede8e405561aca973d78.tar.gz |
Merge changes Iee153445,Iee274471 am: 79df15ea88 am: 10f298fc41 am: 7cb5001398 am: bc0f5df265aml_uwb_331910010aml_uwb_331820070aml_uwb_331613010aml_uwb_331611010aml_uwb_331410010aml_uwb_331310030aml_uwb_331115000aml_uwb_331015040aml_uwb_330810010aml_tz4_332714070aml_tz4_332714050aml_tz4_332714010aml_tz4_331910000aml_tz4_331314030aml_tz4_331314020aml_tz4_331314010aml_tz4_331012050aml_tz4_331012040aml_tz4_331012000aml_ase_331311020aml_ase_331112000aml_ase_331011020android13-mainline-uwb-releaseandroid13-mainline-tzdata4-releaseandroid13-mainline-appsearch-releaseaml_tz4_332714010
Original change: https://android-review.googlesource.com/c/platform/external/eigen/+/1999079
Change-Id: Ife39d10c8b23d3eeb174cd52f462f9d20527ad03
Diffstat (limited to 'unsupported/Eigen/CXX11/src/Tensor/TensorDeviceDefault.h')
-rw-r--r-- | unsupported/Eigen/CXX11/src/Tensor/TensorDeviceDefault.h | 33 |
1 files changed, 28 insertions, 5 deletions
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorDeviceDefault.h b/unsupported/Eigen/CXX11/src/Tensor/TensorDeviceDefault.h index 9d141395b..46b9d3ab2 100644 --- a/unsupported/Eigen/CXX11/src/Tensor/TensorDeviceDefault.h +++ b/unsupported/Eigen/CXX11/src/Tensor/TensorDeviceDefault.h @@ -21,6 +21,12 @@ struct DefaultDevice { EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void deallocate(void* buffer) const { internal::aligned_free(buffer); } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void* allocate_temp(size_t num_bytes) const { + return allocate(num_bytes); + } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void deallocate_temp(void* buffer) const { + deallocate(buffer); + } EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void memcpy(void* dst, const void* src, size_t n) const { ::memcpy(dst, src, n); } @@ -33,11 +39,18 @@ struct DefaultDevice { EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void memset(void* buffer, int c, size_t n) const { ::memset(buffer, c, n); } + template<typename Type> + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Type get(Type data) const { + return data; + } EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE size_t numThreads() const { -#ifndef __CUDA_ARCH__ +#if !defined(EIGEN_GPU_COMPILE_PHASE) // Running on the host CPU return 1; +#elif defined(EIGEN_HIP_DEVICE_COMPILE) + // Running on a HIP device + return 64; #else // Running on a CUDA device return 32; @@ -45,9 +58,12 @@ struct DefaultDevice { } EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE size_t firstLevelCacheSize() const { -#ifndef __CUDA_ARCH__ +#if !defined(EIGEN_GPU_COMPILE_PHASE) && !defined(SYCL_DEVICE_ONLY) // Running on the host CPU return l1CacheSize(); +#elif defined(EIGEN_HIP_DEVICE_COMPILE) + // Running on a HIP device + return 48*1024; // FIXME : update this number for HIP #else // Running on a CUDA device, return the amount of shared memory available. return 48*1024; @@ -55,9 +71,12 @@ struct DefaultDevice { } EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE size_t lastLevelCacheSize() const { -#ifndef __CUDA_ARCH__ +#if !defined(EIGEN_GPU_COMPILE_PHASE) && !defined(SYCL_DEVICE_ONLY) // Running single threaded on the host CPU return l3CacheSize(); +#elif defined(EIGEN_HIP_DEVICE_COMPILE) + // Running on a HIP device + return firstLevelCacheSize(); // FIXME : update this number for HIP #else // Running on a CUDA device return firstLevelCacheSize(); @@ -65,13 +84,17 @@ struct DefaultDevice { } EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE int majorDeviceVersion() const { -#ifndef __CUDA_ARCH__ +#if !defined(EIGEN_GPU_COMPILE_PHASE) // Running single threaded on the host CPU // Should return an enum that encodes the ISA supported by the CPU return 1; +#elif defined(EIGEN_HIP_DEVICE_COMPILE) + // Running on a HIP device + // return 1 as major for HIP + return 1; #else // Running on a CUDA device - return __CUDA_ARCH__ / 100; + return EIGEN_CUDA_ARCH / 100; #endif } }; |