Messages in this thread Patch in this message |  | | Date | Sun, 08 Sep 2002 13:20:18 +0900 (JST) | Subject | Re: TCP Segmentation Offloading (TSO) | From | Hirokazu Takahashi <> |
| |
Hello,
I updated the csum_partial() for x86. csum_partial() for standard x86 can also handle odd buffer better.
> > > Now that is grossly inefficient ;-) since you can save one instruction by > > > moving roll after adcl (hand edited partial patch hunk, won't apply):
I applied it. But it will be trivial for its performance on most packets.
> > I've been doing some PPro assembly lately, and I'm reminded that > > sometimes inserting instructions can reduce the timing by up to 8 cycles > > or so. > > The one instruction that you can still be moved around easily is the > pointer increment. But I would never try to improve code paths that I > consider non critical.
I wish recent x86 processor can reorder instructions in it.
--- linux/arch/i386/lib/checksum.S.BUG Sun Sep 1 17:00:59 2030 +++ linux/arch/i386/lib/checksum.S Fri Sep 6 16:19:27 2030 @@ -55,8 +55,21 @@ csum_partial: movl 20(%esp),%eax # Function arg: unsigned int sum movl 16(%esp),%ecx # Function arg: int len movl 12(%esp),%esi # Function arg: unsigned char *buff - testl $2, %esi # Check alignment. + testl $3, %esi # Check alignment. jz 2f # Jump if alignment is ok. + testl $1, %esi # Check alignment. + jz 10f # Jump if alignment is boundary of 2bytes. + + # buf is odd + dec %ecx + jl 8f + movzbl (%esi), %ebx + adcl %ebx, %eax + roll $8, %eax + inc %esi + testl $2, %esi + jz 2f +10: subl $2, %ecx # Alignment uses up two bytes. jae 1f # Jump if we had at least two bytes. addl $2, %ecx # ecx was < 2. Deal with it. @@ -111,6 +124,10 @@ csum_partial: 6: addl %ecx,%eax adcl $0, %eax 7: + testl $1, 12(%esp) + jz 8f + roll $8, %eax +8: popl %ebx popl %esi ret @@ -126,8 +143,8 @@ csum_partial: movl 16(%esp),%ecx # Function arg: int len movl 12(%esp),%esi # Function arg: const unsigned char *buf - testl $2, %esi - jnz 30f + testl $3, %esi + jnz 25f 10: movl %ecx, %edx movl %ecx, %ebx @@ -145,6 +162,19 @@ csum_partial: lea 2(%esi), %esi adcl $0, %eax jmp 10b +25: + testl $1, %esi + jz 30f + # buf is odd + dec %ecx + jl 90f + movzbl (%esi), %ebx + addl %ebx, %eax + adcl $0, %eax + roll $8, %eax + inc %esi + testl $2, %esi + jz 10b 30: subl $2, %ecx ja 20b @@ -211,6 +241,10 @@ csum_partial: addl %ebx,%eax adcl $0,%eax 80: + testl $1, 12(%esp) + jz 90f + roll $8, %eax +90: popl %ebx popl %esi ret - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
|  |