From d8adaf8ea6a31a380f6bf1fe65e88b661d3bb51e Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Fri, 21 Oct 2016 20:16:43 +0000
Subject: [PATCH] tree stuff
---
src/blas_kernels.cu | 30 ++++++++++++++++++++++++++++++
1 files changed, 30 insertions(+), 0 deletions(-)
diff --git a/src/blas_kernels.cu b/src/blas_kernels.cu
index 0391e2e..59ec005 100644
--- a/src/blas_kernels.cu
+++ b/src/blas_kernels.cu
@@ -691,3 +691,33 @@
mult_add_into_kernel<<<cuda_gridsize(num), BLOCK>>>(num, a, b, c);
check_error(cudaPeekAtLastError());
}
+
+
+__global__ void softmax_kernel(int n, int batch, float *input, float temp, float *output)
+{
+ int b = (blockIdx.x + blockIdx.y*gridDim.x) * blockDim.x + threadIdx.x;
+ if(b >= batch) return;
+
+ int i;
+ float sum = 0;
+ float largest = -INFINITY;
+ for(i = 0; i < n; ++i){
+ int val = input[i+b*n];
+ largest = (val>largest) ? val : largest;
+ }
+ for(i = 0; i < n; ++i){
+ sum += exp(input[i+b*n]/temp-largest/temp);
+ }
+ sum = (sum != 0) ? largest/temp+log(sum) : largest-100;
+ for(i = 0; i < n; ++i){
+ output[i+b*n] = exp(input[i+b*n]/temp-sum);
+ }
+}
+
+extern "C" void softmax_gpu(float *input, int n, int groups, float temp, float *output, cudaStream_t stream)
+{
+ int inputs = n;
+ int batch = groups;
+ softmax_kernel<<<cuda_gridsize(batch), BLOCK, 0, stream>>>(inputs, batch, input, temp, output);
+ check_error(cudaPeekAtLastError());
+}
--
Gitblit v1.10.0