RDMA/ocrdma: Help gcc generate better code for ocrdma_srq_toggle_bit
authorRasmus Villemoes <linux@rasmusvillemoes.dk>
Fri, 16 Jan 2015 14:39:55 +0000 (15:39 +0100)
committerRoland Dreier <roland@purestorage.com>
Wed, 18 Feb 2015 16:31:05 +0000 (08:31 -0800)
gcc emits a surprising amount of code in order to flip a bit. One
would think that a single instruction is enough.

$ scripts/bloat-o-meter /tmp/ocrdma_verbs.o drivers/infiniband/hw/ocrdma/ocrdma_verbs.o
add/remove: 0/0 grow/shrink: 0/3 up/down: 0/-142 (-142)
function                                     old     new   delta
ocrdma_post_srq_recv                         498     460     -38
ocrdma_poll_cq                              2010    1962     -48
ocrdma_discard_cqes                          495     439     -56

All three calls of ocrdma_srq_toggle_bit happen within spinlocks, so
saving a few useless instructions might be worthwhile.

Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Acked-by: Selvin Xavier <selvin.xavier@emulex.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
drivers/infiniband/hw/ocrdma/ocrdma_verbs.c

index 5f656b9..c1e9f96 100644 (file)
@@ -1591,10 +1591,7 @@ static void ocrdma_srq_toggle_bit(struct ocrdma_srq *srq, int idx)
        int i = idx / 32;
        unsigned int mask = (1 << (idx % 32));
 
-       if (srq->idx_bit_fields[i] & mask)
-               srq->idx_bit_fields[i] &= ~mask;
-       else
-               srq->idx_bit_fields[i] |= mask;
+       srq->idx_bit_fields[i] ^= mask;
 }
 
 static int ocrdma_hwq_free_cnt(struct ocrdma_qp_hwq_info *q)