From 352ae7e65b6a74bcd768aa88b866a44c713284c8 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Wed, 26 Oct 2016 15:35:44 +0000
Subject: [PATCH] ADAM
---
src/blas_kernels.cu | 15 +++++++++++++++
1 files changed, 15 insertions(+), 0 deletions(-)
diff --git a/src/blas_kernels.cu b/src/blas_kernels.cu
index b4d520e..684e66d 100644
--- a/src/blas_kernels.cu
+++ b/src/blas_kernels.cu
@@ -140,6 +140,21 @@
}
+__global__ void adam_kernel(int N, float *x, float *m, float *v, float B1, float B2, float rate, float eps, int t)
+{
+ int index = (blockIdx.x + blockIdx.y*gridDim.x) * blockDim.x + threadIdx.x;
+ if (index >= N) return;
+
+ x[index] = x[index] - (rate * sqrt(1.-pow(B2, t)) / (1.-pow(B1, t)) * m[index] / (sqrt(v[index]) + eps));
+ //if(index == 0) printf("%f %f %f %f\n", m[index], v[index], (rate * sqrt(1.-pow(B2, t)) / (1.-pow(B1, t)) * m[index] / (sqrt(v[index]) + eps)));
+}
+
+extern "C" void adam_gpu(int n, float *x, float *m, float *v, float B1, float B2, float rate, float eps, int t)
+{
+ adam_kernel<<<cuda_gridsize(n), BLOCK>>>(n, x, m, v, B1, B2, rate, eps, t);
+ check_error(cudaPeekAtLastError());
+}
+
__global__ void normalize_kernel(int N, float *x, float *mean, float *variance, int batch, int filters, int spatial)
{
int index = (blockIdx.x + blockIdx.y*gridDim.x) * blockDim.x + threadIdx.x;
--
Gitblit v1.10.0