commonlib: Add assembly optimization for ipchksum() on x86
This patch adds a bit of optimized assembly code to the ipchksum()
algorithm for x86 targets in order to take advantage of larger load
sizes and the add-with-carry instruction. The same assembly (with one
minor manual tweak) works for both 32 and 64 bit mode (with most of the
work being done by GCC which automatically inserts `rax` or `eax` in the
inline assembly depending on the build target).
Change-Id: I484620dc14679ff5ca02b2ced2f84650730a6efc
Signed-off-by: Julius Werner <jwerner@chromium.org>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/80255
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
diff --git a/src/commonlib/bsd/ipchksum.c b/src/commonlib/bsd/ipchksum.c
index 89d261f..b7434e5 100644
--- a/src/commonlib/bsd/ipchksum.c
+++ b/src/commonlib/bsd/ipchksum.c
@@ -34,7 +34,30 @@
:: "cc"
);
}
-#endif
+#elif defined(__i386__) || defined(__x86_64__)
+ size_t size8 = size / 8;
+ const uint64_t *p8 = data;
+ i = size8 * 8;
+ asm (
+ "clc\n\t"
+ "1:\n\t"
+ "jecxz 2f\n\t" /* technically RCX on 64, but not gonna be that big */
+ "adc (%[p8]), %[wsum]\n\t"
+#if defined(__i386__)
+ "adc 4(%[p8]), %[wsum]\n\t"
+#endif /* __i386__ */
+ "lea -1(%[size8]), %[size8]\n\t" /* Use LEA as a makeshift ADD that */
+ "lea 8(%[p8]), %[p8]\n\t" /* doesn't modify the carry flag. */
+ "jmp 1b\n\t"
+ "2:\n\t"
+ "setc %b[size8]\n\t" /* reuse size register to save last carry */
+ "add %[size8], %[wsum]\n\t"
+ : [wsum] "+r" (wide_sum),
+ [p8] "+r" (p8),
+ [size8] "+c" (size8) /* put size in ECX so we can JECXZ */
+ :: "cc"
+ );
+#endif /* __i386__ || __x86_64__ */
while (wide_sum) {
sum += wide_sum & 0xFFFF;