OPTIM: hpack: read 32 bits at once when possible.
As suggested in the comment, it's possible to read 32 bits at once in
big-endian order, and now we have the functions to do this, so let's
do it. This reduces the code on the fast path by 31 bytes on x86, and
more importantly performs single-operation 32-bit reads instead of
playing with shifts and additions.
diff --git a/src/hpack-huff.c b/src/hpack-huff.c
index 9c63e28..35fa52d 100644
--- a/src/hpack-huff.c
+++ b/src/hpack-huff.c
@@ -31,6 +31,7 @@
#include <haproxy/api.h>
#include <haproxy/hpack-huff.h>
+#include <haproxy/net_helper.h>
struct huff {
uint32_t c; /* code point */
@@ -1439,17 +1440,11 @@
while (shift >= 32) {
curr = next;
- /* read up to 4 bytes into next. FIXME: this should
- * later be optimized to perform a single 32-bit big
- * endian read when unaligned accesses are possible.
- */
+ /* read up to 4 bytes into next */
next = 0;
if (huff + 4 <= huff_end) {
- next = ((uint32_t)huff[0] << 24) +
- ((uint32_t)huff[1] << 16) +
- ((uint32_t)huff[2] << 8) +
- (uint32_t)huff[3];
+ next = read_n32(huff);
huff += 4;
}
else {