diff --git a/main.c b/main.c
index 0b7fdf7..d7eace9 100644
--- a/main.c
+++ b/main.c
@@ -9,6 +9,7 @@
 #include "texture.h"
 
 typedef uint8_t u8;
+typedef int32_t i32;
 
 #define CDEPTH 255
 
@@ -53,18 +54,18 @@ void drawTexInto(struct texture tex, struct texture buffer, struct rect texbound
    rect_dims(bufbounds, &outwidth, &outheight);
    // We precalculate the step through the texture to avoid division and
    // multiplication per pixel
-   float stepx = (float)texwidth / outwidth;
-   float stepy = (float)texheight / outheight;
-   float texx = texbounds.x1 * 3;
-   float texy = texbounds.y1 * 3;
+   i32 stepx = 256 * (float)texwidth / outwidth;
+   i32 stepy = 256 * (float)texheight / outheight;
+   i32 texx = 256 * texbounds.x1;
+   i32 texy = 256 * texbounds.y1;
    //log("%f,%f step %f,%f", texx, texy, stepx, stepy);
    // Buffer is R G B for each pixel, then across then down. If you want X, Y,
    // it's: 3 * (x + y * width)
    for(int y = bufbounds.y1; y < bufbounds.y2; y++) {
       texx = texbounds.x1;
       for(int x = bufbounds.x1; x < bufbounds.x2; x++) {
-         int texi = 3 * (floor(texx) + tex.width * floor(texy));
-         int bufi = 3 * (x + buffer.width * y);
+         i32 texi = 3 * ((texx >> 8) + tex.width * (texy >> 8));
+         i32 bufi = 3 * (x + buffer.width * y);
          // You can get rid of this for loop for more performance, but probably
          // not a big deal with -O3, it'll be unrolled
          for(int i = 0; i < 3; i++) {