syscall_other: optimisations

Cache the inode calculation in execve, otherwise each loop pass must recompute
it as the compiler can't know the functions don't touch it.

Add a couple of staticfasts

diff --git a/Kernel/syscall_other.c b/Kernel/syscall_other.c
index e618621..b14276a 100644 (file)
--- a/Kernel/syscall_other.c
+++ b/Kernel/syscall_other.c
@@ -27,7 +27,7 @@ put inside itself!
 
 int16_t _rename(void)
 {
-       inoptr srci, srcp, dsti, dstp;
+       staticfast inoptr srci, srcp, dsti, dstp;
        char fname[FILENAME_LEN + 1];
        int ret = -1;
 
@@ -394,6 +394,7 @@ int16_t _execve(void)
        int j;
        uint16_t top = (uint16_t)ramtop;
        uint8_t c;
+       uint16_t blocks;
 
        kputs("execve\n");
 
@@ -547,7 +548,12 @@ int16_t _execve(void)
         *  will still be cached. - Hat tip to Steve Hosgood's OMU for that trick
         */
        progptr = PROGBASE + 512;       // we copied the first block already
-       for (blk = 1; blk <= ino->c_node.i_size >> 9; ++blk) {
+
+       /* Compute this once otherwise each loop we must recalculate this
+          as the compiler isn't entitled to assume the loop didn't change it */
+       blocks = ino->c_node.i_size >> 9;
+
+       for (blk = 1; blk <= blocks; ++blk) {
                buf = bread(ino->c_dev, bmap(ino, blk, 1), 0);
                uput(buf, progptr, 512);
                bufdiscard((bufptr) buf);