From 35c028dbeb062db8feb3b0dbcd14b0a1cef30475 Mon Sep 17 00:00:00 2001
From: lucasdelimanogueira <lucasln96@hotmail.com>
Date: Mon, 29 Apr 2024 12:50:48 -0300
Subject: [PATCH] Makefile cuda

---
 build/Makefile  | 2 +-
 csrc/tensor.cpp | 6 ++++--
 2 files changed, 5 insertions(+), 3 deletions(-)

diff --git a/build/Makefile b/build/Makefile
index 9d38c99..cf3c755 100644
--- a/build/Makefile
+++ b/build/Makefile
@@ -19,7 +19,7 @@ CU_OBJS = $(patsubst $(SRCDIR)/%.cu, $(BUILDDIR)/%.cu.o, $(CU_SRCS))
 
 # CUDA flags and libraries
 CUDAFLAGS = -arch=sm_75
-CUDALIBS = -lcudart
+CUDALIBS = -L/usr/local/cuda-10.2/lib64/ -lcudart -lcuda
 
 # Rule to build the target
 $(TARGET): $(OBJS) $(CU_OBJS)
diff --git a/csrc/tensor.cpp b/csrc/tensor.cpp
index 7bf0005..6e2e336 100644
--- a/csrc/tensor.cpp
+++ b/csrc/tensor.cpp
@@ -2,6 +2,7 @@
 #include <stdlib.h>
 #include <string.h>
 #include <math.h>
+#include <cuda_runtime_api.h>
 #include "tensor.h"
 #include "cuda.h"
 #define THREADS_PER_BLOCK 128
@@ -76,14 +77,15 @@ extern "C" {
     void to_device(Tensor* tensor, char* device) {
         printf("Sending tensor to device: %s\n", device);
         #ifdef __CUDACC__
+
             if ((strcmp(device, "cuda") == 0) && (strcmp(tensor->device, "cpu") == 0)) {
-                
+            
                 float* data_tmp;
 
                 cudaMalloc((void **)&data_tmp, tensor->size * sizeof(float));
                 cudaMemcpy(data_tmp, tensor->data, tensor->size * sizeof(float), cudaMemcpyHostToDevice);
 
-                free(tensor_data);
+                free(tensor->data);
                 tensor->data = data_tmp;
                 tensor->device = device;
             }