@@ -32,7 +32,7 @@
Since a stf.spill f0 can store 16B in one go, we use this instruction
to get peak speed. */
-#include <sysdep.h>
+#include "sysdep.h"
#undef ret
#define dest in0
@@ -31,7 +31,7 @@
This implementation assumes that it is safe to do read ahead
in the src block, without getting beyond its limit. */
#define OP_T_THRES 16
@@ -40,7 +40,7 @@
All the loops in this function could have had the internal branch removed
if br.ctop and br.cloop could be predicated :-(. */
#define saved_pr r15
@@ -33,7 +33,7 @@
and all the mux1 instructions should be replaced by plain mov's. */
@@ -37,7 +37,7 @@
#define USE_LFETCH
#define USE_FLP
#define LFETCH_DIST 500
sh1 must be computed using an extra instruction: sub sh1 = 64, sh1
or the UM.be bit should be cleared at the beginning and set at the end. */
to get peak speed when value = 0. */
@@ -30,7 +30,7 @@
This implementation assumes little endian mode. For big endian mode,
the instruction czx1.r should be replaced by czx1.l. */
#define saved_lc r18
@@ -27,7 +27,7 @@
Unlike memcmp(), this function is optimized for mismatches within the
first few characters. */
#define s1 in0
shr.u tmp = r[0], sh2 // tmp = w1 >> sh2
*/
#define saved_lc r15
@@ -28,7 +28,7 @@
@@ -29,7 +29,7 @@
In this form, it assumes little endian mode.