From c370c754ebd133cfed19ad1773ef540e48bd4240 Mon Sep 17 00:00:00 2001 From: Georgios Pinitas Date: Thu, 12 Sep 2019 12:26:11 +0100 Subject: COMPMID-2570: Add hybrid precision flag in GEMMKernelInfo. Adds fp_mixed_precision flag in GEMMKernelInfo in order to indicate use of wider accumulators for FP16 workloads where supported. Change-Id: I8d69bc62d945fd40af77395ff673d07880b7aeb6 Signed-off-by: Georgios Pinitas Reviewed-on: https://review.mlplatform.org/c/1914 Reviewed-by: Michele Di Giorgio Comments-Addressed: Arm Jenkins Tested-by: Arm Jenkins --- arm_compute/core/KernelDescriptors.h | 1 + 1 file changed, 1 insertion(+) diff --git a/arm_compute/core/KernelDescriptors.h b/arm_compute/core/KernelDescriptors.h index 3affc30f71..215f0f1651 100644 --- a/arm_compute/core/KernelDescriptors.h +++ b/arm_compute/core/KernelDescriptors.h @@ -60,6 +60,7 @@ struct GEMMKernelInfo unsigned int depth_output_gemm3d{ 0 }; /**< Depth of the output tensor in case is reinterpreted as 3D */ bool reinterpret_input_as_3d{ false }; /**< Flag used to reinterpret the input as 3D */ bool broadcast_bias{ false }; /**< Flag used to broadcase the bias addition */ + bool fp_mixed_precision{ false }; /**< Flag used to indicate wider accumulators (32 bit instead of 16 for FP16). */ ActivationLayerInfo activation_info{}; /**< Activation function to perform after the matrix multiplication */ }; -- cgit v1.2.1