Prepare matmul global func for test

Author: vxst

main.cu

@@ -0,0 +1,40 @@
+#include <cuda_runtime.h>
+#include <cstdio>
+
+#include "matmulf8.cuh"
+#include "load_core.cuh"
+
+
+int main() {
+    int n = 4096, m = 4096, p = 4096;
+    int *A, *B, *C;
+    cudaSetDevice(0);
+    cudaFree(0);
+    
+    cudaMallocHost(&A, n * m * sizeof(int) / 4);
+    cudaMallocHost(&B, m * p * sizeof(int) / 4);
+    cudaMallocHost(&C, n * p * sizeof(int) / 4);
+#ifdef DB
+    int* acore = load_core("addcore.bin");
+#else
+    int* acore = load_core("apdcore.bin");
+#endif
+    int* mcore = load_core("mltcore.bin");
+    for(int i = 0; i < n * m / 4; i++) {
+        A[i] = rand();
+        // A[i] &= 0x7f7f7f7f;
+        A[i] = 0;
+    }
+    for(int i = 0; i < m * p / 4; i++) {
+        B[i] = rand();
+        // B[i] &= 0x7f7f7f7f;
+        B[i] = 0;
+    }
+    float t = matmul(A, B, C, n, m, p, acore, mcore);
+    printf("Time: %f ms\n", t);
+    float flops = 2.0 * n * m * p / t / 1e6;
+    printf("FLOPS: %f GFLOPS\n", flops);
+    cudaFreeHost(A); cudaFreeHost(B); cudaFreeHost(C);
+    cudaFreeHost(acore); cudaFreeHost(mcore);
+    return 0;
+}

matmulf8.cu

@@ -51,37 +51,3 @@ float matmul(int* A, int* B, int* C, int n, int m, int p, int* acore, int* mcore
 
     return t;
 }
-
-int main() {
-    int n = 4096, m = 4096, p = 4096;
-    int *A, *B, *C;
-    cudaSetDevice(0);
-    cudaFree(0);
-    
-    cudaMallocHost(&A, n * m * sizeof(int) / 4);
-    cudaMallocHost(&B, m * p * sizeof(int) / 4);
-    cudaMallocHost(&C, n * p * sizeof(int) / 4);
-#ifdef DB
-    int* acore = load_core("addcore.bin");
-#else
-    int* acore = load_core("apdcore.bin");
-#endif
-    int* mcore = load_core("mltcore.bin");
-    for(int i = 0; i < n * m / 4; i++) {
-        A[i] = rand();
-        // A[i] &= 0x7f7f7f7f;
-        A[i] = 0;
-    }
-    for(int i = 0; i < m * p / 4; i++) {
-        B[i] = rand();
-        // B[i] &= 0x7f7f7f7f;
-        B[i] = 0;
-    }
-    float t = matmul(A, B, C, n, m, p, acore, mcore);
-    printf("Time: %f ms\n", t);
-    float flops = 2.0 * n * m * p / t / 1e6;
-    printf("FLOPS: %f GFLOPS\n", flops);
-    cudaFreeHost(A); cudaFreeHost(B); cudaFreeHost(C);
-    cudaFreeHost(acore); cudaFreeHost(mcore);
-    return 0;
-}

matmulf8.cuh

@@ -0,0 +1,3 @@
+#pragma once
+
+float matmul(int* A, int* B, int* C, int n, int m, int p, int* acore, int* mcore);