From dc771107d4c016c900584c63093fc113487f30b1 Mon Sep 17 00:00:00 2001
From: Nicholas Rianto Putra <nicholasmagbanua@gmail.com>
Date: Fri, 12 Apr 2019 19:55:45 +0700
Subject: [PATCH] edit makefile
---
Makefile | 1 +
src/radix_sort_parallel.cu | 18 +++++++++---------
2 files changed, 10 insertions(+), 9 deletions(-)
diff --git a/Makefile b/Makefile
index 9243cc0..ff6c8b2 100644
--- a/Makefile
+++ b/Makefile
@@ -1,2 +1,3 @@
radix_sort:
+ gcc -g -Wall -o serial_radix_sort src/serial_radix_sort.c -lm
nvcc src/radix_sort_parallel.cu -o radix_sort
diff --git a/src/radix_sort_parallel.cu b/src/radix_sort_parallel.cu
index e857a87..fd4227e 100644
--- a/src/radix_sort_parallel.cu
+++ b/src/radix_sort_parallel.cu
@@ -55,15 +55,15 @@ __global__ void assign_index(int* arr, int* arr_idx, int* down_sweep_i, int* up_
}
__global__ void assign_flags(int* arr, int n, int idx, int* flags) {
- for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < n; i += blockDim.x * gridDim.x) {
- flags[i] = ((arr[i] & 1 << idx) == 1 << idx) ? 0 : 1;
- }
+ for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < n; i += blockDim.x * gridDim.x) {
+ flags[i] = ((arr[i] & 1 << idx) == 1 << idx) ? 0 : 1;
+ }
}
__global__ void scatter(int* arr, int* arr_temp, int* arr_idx, int n) {
- for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < n; i += blockDim.x * gridDim.x) {
- arr[arr_idx[i]] = arr_temp[i];
- }
+ for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < n; i += blockDim.x * gridDim.x) {
+ arr[arr_idx[i]] = arr_temp[i];
+ }
}
void count_sort(int* d_arr, int n, int idx) {
@@ -129,11 +129,11 @@ int main(int argc, char *argv[]) {
// Move sorting CPU
int* d_arr;
- cudaMalloc(&d_arr, sizeof(int) * n);
- cudaMemcpy(d_arr, arr, sizeof(int) * n, cudaMemcpyHostToDevice);
+ cudaMalloc(&d_arr, sizeof(int) * n);
+ cudaMemcpy(d_arr, arr, sizeof(int) * n, cudaMemcpyHostToDevice);
radix_sort(d_arr, n);
cudaMemcpy(arr, d_arr, n * sizeof(int), cudaMemcpyDeviceToHost);
- cudaFree(d_arr);
+ cudaFree(d_arr);
end = clock();
cpu_time_used = ((double) (end - start)) * 1000000 / CLOCKS_PER_SEC;
--
GitLab