Le 24/05/2018 à 19:24, Segher Boessenkool a écrit :

On Wed, May 23, 2018 at 09:47:32AM +0200, Christophe Leroy wrote:

quoted

At the time being, memcmp() compares two chunks of memory
byte per byte.

This patch optimises the comparison by comparing word by word.

A small benchmark performed on an 8xx comparing two chuncks
of 512 bytes performed 100000 times gives:

Before : 5852274 TB ticks
After:   1488638 TB ticks

quoted

diff --git a/arch/powerpc/lib/string_32.S b/arch/powerpc/lib/string_32.S
index 40a576d56ac7..542e6cecbcaf 100644
--- a/arch/powerpc/lib/string_32.S
+++ b/arch/powerpc/lib/string_32.S

@@ -16,17 +16,45 @@
  	.text

  _GLOBAL(memcmp)
-	cmpwi	cr0, r5, 0
-	beq-	2f
-	mtctr	r5
-	addi	r6,r3,-1
-	addi	r4,r4,-1
-1:	lbzu	r3,1(r6)
-	lbzu	r0,1(r4)
-	subf.	r3,r0,r3
-	bdnzt	2,1b
+	srawi.	r7, r5, 2		/* Divide len by 4 */
+	mr	r6, r3
+	beq-	3f
+	mtctr	r7
+	li	r7, 0
+1:
+#ifdef __LITTLE_ENDIAN__
+	lwbrx	r3, r6, r7
+	lwbrx	r0, r4, r7
+#else
+	lwzx	r3, r6, r7
+	lwzx	r0, r4, r7
+#endif

You don't test whether the pointers are word-aligned.  Does that work?

copy_tofrom_user() word-aligns the store address and doesn't take care 
of the load address, so I believe it works.

Now, I just read in the MPC885 Ref Manual that unaligned access 
generates alignment exception when the processor is running in LE mode.

Ref. made to the discussion on patch "powerpc/32be: use stmw/lmw for 
registers save/restore in asm" 
(https://patchwork.ozlabs.org/patch/899465/), I will drop the handling 
for LE mode.

Christophe

Say, when a load is crossing a page boundary, or segment boundary.


Segher