vector_ptr now requires alignment in bytes instead of elements

Author: stijnh

example.cu

@@ -0,0 +1,13 @@
+#include "kernel_float.h"
+#include <cuda_fp16.h>
+
+namespace kf = kernel_float;
+
+__global__ void kernel(
+    kf::vec_ptr<half, 4, const __nv_fp8_e5m2> input, 
+    float constant, 
+    kf::vec_ptr<half, 4> output
+) {
+    int i = blockIdx.x * blockDim.x + threadIdx.x;
+    output(i) = input[i] + kf::cast<half>(constant);
+}

examples/vector_add/main.cu

@@ -22,7 +22,7 @@ __global__ void my_kernel(
     int i = blockIdx.x * blockDim.x + threadIdx.x;
 
     if (i * N < length) {
-        output(i) = kf::fma(input[i], input[i], kf::cast<half>(constant));
+        output[i] = kf::fma(input[i], input[i], kf::cast<half>(constant));
     }
 }