diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_conv/pooling/pooling_fp32.cpp')
-rw-r--r-- | src/core/NEON/kernels/arm_conv/pooling/pooling_fp32.cpp | 58 |
1 files changed, 58 insertions, 0 deletions
diff --git a/src/core/NEON/kernels/arm_conv/pooling/pooling_fp32.cpp b/src/core/NEON/kernels/arm_conv/pooling/pooling_fp32.cpp index 5ee0884294..99d106583e 100644 --- a/src/core/NEON/kernels/arm_conv/pooling/pooling_fp32.cpp +++ b/src/core/NEON/kernels/arm_conv/pooling/pooling_fp32.cpp @@ -30,6 +30,12 @@ #include "kernels/cpp_nhwc_1x1_stride_any_depthfirst.hpp" #if defined(__aarch64__) +#if defined(ARM_COMPUTE_ENABLE_SME) +#include "kernels/sme_fp32_nhwc_max_2x2_s1_output2x2_depthfirst.hpp" +#include "kernels/sme_fp32_nhwc_avg_3x3_s1_output2x2_depthfirst.hpp" +#include "kernels/sme_fp32_nhwc_avg_generic_depthfirst.hpp" +#include "kernels/sme_fp32_nhwc_max_generic_depthfirst.hpp" +#endif // defined(ARM_COMPUTE_ENABLE_SME) #if defined(ARM_COMPUTE_ENABLE_SVE) #include "kernels/sve_fp32_nhwc_max_2x2_s1_output2x2_depthfirst.hpp" #include "kernels/sve_fp32_nhwc_avg_3x3_s1_output2x2_depthfirst.hpp" @@ -59,6 +65,58 @@ static const PoolingImplementation<float, float> pooling_fp32_methods[] = { }, }, #if defined(__aarch64__) +#if defined(ARM_COMPUTE_ENABLE_SME) + { + PoolingMethod::DEPTHFIRST, + "sme_fp32_nhwc_max_2x2_s1_output2x2_depthfirst", + [] (const PoolingArgs &args, const Nothing &os) -> bool { + return args.cpu_info->has_sme() && + is_supported<sme_fp32_nhwc_max_2x2_s1_output2x2_depthfirst>(args, os); + }, + nullptr, + [] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<float, float> * { + auto strat = new sme_fp32_nhwc_max_2x2_s1_output2x2_depthfirst(args.cpu_info); + return new PoolingDepthfirst<float>(strat, args); + }, + }, + { + PoolingMethod::DEPTHFIRST, + "sme_fp32_nhwc_avg_3x3_s1_output2x2_depthfirst", + [] (const PoolingArgs &args, const Nothing &os) -> bool { + return args.cpu_info->has_sme() && + is_supported<sme_fp32_nhwc_avg_3x3_s1_output2x2_depthfirst>(args, os); + }, + nullptr, + [] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<float, float> * { + auto strat = new sme_fp32_nhwc_avg_3x3_s1_output2x2_depthfirst(args.cpu_info); + return new PoolingDepthfirst<float>(strat, args); + }, + }, + { + PoolingMethod::DEPTHFIRST, + "sme_fp32_nhwc_avg_generic_depthfirst", + [] (const PoolingArgs &args, const Nothing &) -> bool { + return args.cpu_info->has_sme() && args.pool_type == PoolingType::AVERAGE; + }, + nullptr, + [] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<float, float> * { + auto strat = new sme_fp32_nhwc_avg_generic_depthfirst(args.cpu_info); + return new PoolingDepthfirstGeneric<float>(strat, args); + }, + }, + { + PoolingMethod::DEPTHFIRST, + "sme_fp32_nhwc_max_generic_depthfirst", + [] (const PoolingArgs &args, const Nothing &) -> bool { + return args.cpu_info->has_sme() && args.pool_type == PoolingType::MAX; + }, + nullptr, + [] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<float, float> * { + auto strat = new sme_fp32_nhwc_max_generic_depthfirst(args.cpu_info); + return new PoolingDepthfirstGeneric<float>(strat, args); + }, + }, +#endif // defined(ARM_COMPUTE_ENABLE_SME) #if defined(ARM_COMPUTE_ENABLE_SVE) { PoolingMethod::DEPTHFIRST, |