lk/linux-kernel-test
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
5d2bd7009f
linux-kernel-test/arch/x86/include/asm/xor_64.h
Suresh Siddha 841e3604d3 x86, fpu: always use kernel_fpu_begin/end() for in-kernel FPU usage 
use kernel_fpu_begin/end() instead of unconditionally accessing cr0 and
saving/restoring just the few used xmm/ymm registers.

This has some advantages like:
* If the task's FPU state is already active, then kernel_fpu_begin()
  will just save the user-state and avoiding the read/write of cr0.
  In general, cr0 accesses are much slower.

* Manual save/restore of xmm/ymm registers will affect the 'modified' and
  the 'init' optimizations brought in the by xsaveopt/xrstor
  infrastructure.

* Foward compatibility with future vector register extensions will be a
  problem if the xmm/ymm registers are manually saved and restored
  (corrupting the extended state of those vector registers).

With this patch, there was no significant difference in the xor throughput
using AVX, measured during boot.

Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Link: http://lkml.kernel.org/r/1345842782-24175-5-git-send-email-suresh.b.siddha@intel.com
Cc: Jim Kukunas <james.t.kukunas@linux.intel.com>
Cc: NeilBrown <neilb@suse.de>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2012-09-18 15:52:08 -07:00

325 lines
7.0 KiB
C
Raw Blame History

#ifndef _ASM_X86_XOR_64_H
#define _ASM_X86_XOR_64_H
/*
* Optimized RAID-5 checksumming functions for MMX and SSE.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2, or (at your option)
* any later version.
*
* You should have received a copy of the GNU General Public License
* (for example /usr/src/linux/COPYING); if not, write to the Free
* Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
*/
/*
* Cache avoiding checksumming functions utilizing KNI instructions
* Copyright (C) 1999 Zach Brown (with obvious credit due Ingo)
*/
/*
* Based on
* High-speed RAID5 checksumming functions utilizing SSE instructions.
* Copyright (C) 1998 Ingo Molnar.
*/
/*
* x86-64 changes / gcc fixes from Andi Kleen.
* Copyright 2002 Andi Kleen, SuSE Labs.
*
* This hasn't been optimized for the hammer yet, but there are likely
* no advantages to be gotten from x86-64 here anyways.
*/
#include <asm/i387.h>
#define OFFS(x) "16*("#x")"
#define PF_OFFS(x) "256+16*("#x")"
#define PF0(x) " prefetchnta "PF_OFFS(x)"(%[p1]) ;\n"
#define LD(x, y) " movaps "OFFS(x)"(%[p1]), %%xmm"#y" ;\n"
#define ST(x, y) " movaps %%xmm"#y", "OFFS(x)"(%[p1]) ;\n"
#define PF1(x) " prefetchnta "PF_OFFS(x)"(%[p2]) ;\n"
#define PF2(x) " prefetchnta "PF_OFFS(x)"(%[p3]) ;\n"
#define PF3(x) " prefetchnta "PF_OFFS(x)"(%[p4]) ;\n"
#define PF4(x) " prefetchnta "PF_OFFS(x)"(%[p5]) ;\n"
#define PF5(x) " prefetchnta "PF_OFFS(x)"(%[p6]) ;\n"
#define XO1(x, y) " xorps "OFFS(x)"(%[p2]), %%xmm"#y" ;\n"
#define XO2(x, y) " xorps "OFFS(x)"(%[p3]), %%xmm"#y" ;\n"
#define XO3(x, y) " xorps "OFFS(x)"(%[p4]), %%xmm"#y" ;\n"
#define XO4(x, y) " xorps "OFFS(x)"(%[p5]), %%xmm"#y" ;\n"
#define XO5(x, y) " xorps "OFFS(x)"(%[p6]), %%xmm"#y" ;\n"
static void
xor_sse_2(unsigned long bytes, unsigned long *p1, unsigned long *p2)
{
unsigned int lines = bytes >> 8;
kernel_fpu_begin();
asm volatile(
#undef BLOCK
#define BLOCK(i) \
LD(i, 0) \
LD(i + 1, 1) \
PF1(i) \
PF1(i + 2) \
LD(i + 2, 2) \
LD(i + 3, 3) \
PF0(i + 4) \
PF0(i + 6) \
XO1(i, 0) \
XO1(i + 1, 1) \
XO1(i + 2, 2) \
XO1(i + 3, 3) \
ST(i, 0) \
ST(i + 1, 1) \
ST(i + 2, 2) \
ST(i + 3, 3) \
PF0(0)
PF0(2)
" .align 32 ;\n"
" 1: ;\n"
BLOCK(0)
BLOCK(4)
BLOCK(8)
BLOCK(12)
" addq %[inc], %[p1] ;\n"
" addq %[inc], %[p2] ;\n"
" decl %[cnt] ; jnz 1b"
: [p1] "+r" (p1), [p2] "+r" (p2), [cnt] "+r" (lines)
: [inc] "r" (256UL)
: "memory");
kernel_fpu_end();
}
static void
xor_sse_3(unsigned long bytes, unsigned long *p1, unsigned long *p2,
unsigned long *p3)
{
unsigned int lines = bytes >> 8;
kernel_fpu_begin();
asm volatile(
#undef BLOCK
#define BLOCK(i) \
PF1(i) \
PF1(i + 2) \
LD(i, 0) \
LD(i + 1, 1) \
LD(i + 2, 2) \
LD(i + 3, 3) \
PF2(i) \
PF2(i + 2) \
PF0(i + 4) \
PF0(i + 6) \
XO1(i, 0) \
XO1(i + 1, 1) \
XO1(i + 2, 2) \
XO1(i + 3, 3) \
XO2(i, 0) \
XO2(i + 1, 1) \
XO2(i + 2, 2) \
XO2(i + 3, 3) \
ST(i, 0) \
ST(i + 1, 1) \
ST(i + 2, 2) \
ST(i + 3, 3) \
PF0(0)
PF0(2)
" .align 32 ;\n"
" 1: ;\n"
BLOCK(0)
BLOCK(4)
BLOCK(8)
BLOCK(12)
" addq %[inc], %[p1] ;\n"
" addq %[inc], %[p2] ;\n"
" addq %[inc], %[p3] ;\n"
" decl %[cnt] ; jnz 1b"
: [cnt] "+r" (lines),
[p1] "+r" (p1), [p2] "+r" (p2), [p3] "+r" (p3)
: [inc] "r" (256UL)
: "memory");
kernel_fpu_end();
}
static void
xor_sse_4(unsigned long bytes, unsigned long *p1, unsigned long *p2,
unsigned long *p3, unsigned long *p4)
{
unsigned int lines = bytes >> 8;
kernel_fpu_begin();
asm volatile(
#undef BLOCK
#define BLOCK(i) \
PF1(i) \
PF1(i + 2) \
LD(i, 0) \
LD(i + 1, 1) \
LD(i + 2, 2) \
LD(i + 3, 3) \
PF2(i) \
PF2(i + 2) \
XO1(i, 0) \
XO1(i + 1, 1) \
XO1(i + 2, 2) \
XO1(i + 3, 3) \
PF3(i) \
PF3(i + 2) \
PF0(i + 4) \
PF0(i + 6) \
XO2(i, 0) \
XO2(i + 1, 1) \
XO2(i + 2, 2) \
XO2(i + 3, 3) \
XO3(i, 0) \
XO3(i + 1, 1) \
XO3(i + 2, 2) \
XO3(i + 3, 3) \
ST(i, 0) \
ST(i + 1, 1) \
ST(i + 2, 2) \
ST(i + 3, 3) \
PF0(0)
PF0(2)
" .align 32 ;\n"
" 1: ;\n"
BLOCK(0)
BLOCK(4)
BLOCK(8)
BLOCK(12)
" addq %[inc], %[p1] ;\n"
" addq %[inc], %[p2] ;\n"
" addq %[inc], %[p3] ;\n"
" addq %[inc], %[p4] ;\n"
" decl %[cnt] ; jnz 1b"
: [cnt] "+c" (lines),
[p1] "+r" (p1), [p2] "+r" (p2), [p3] "+r" (p3), [p4] "+r" (p4)
: [inc] "r" (256UL)
: "memory" );
kernel_fpu_end();
}
static void
xor_sse_5(unsigned long bytes, unsigned long *p1, unsigned long *p2,
unsigned long *p3, unsigned long *p4, unsigned long *p5)
{
unsigned int lines = bytes >> 8;
kernel_fpu_begin();
asm volatile(
#undef BLOCK
#define BLOCK(i) \
PF1(i) \
PF1(i + 2) \
LD(i, 0) \
LD(i + 1, 1) \
LD(i + 2, 2) \
LD(i + 3, 3) \
PF2(i) \
PF2(i + 2) \
XO1(i, 0) \
XO1(i + 1, 1) \
XO1(i + 2, 2) \
XO1(i + 3, 3) \
PF3(i) \
PF3(i + 2) \
XO2(i, 0) \
XO2(i + 1, 1) \
XO2(i + 2, 2) \
XO2(i + 3, 3) \
PF4(i) \
PF4(i + 2) \
PF0(i + 4) \
PF0(i + 6) \
XO3(i, 0) \
XO3(i + 1, 1) \
XO3(i + 2, 2) \
XO3(i + 3, 3) \
XO4(i, 0) \
XO4(i + 1, 1) \
XO4(i + 2, 2) \
XO4(i + 3, 3) \
ST(i, 0) \
ST(i + 1, 1) \
ST(i + 2, 2) \
ST(i + 3, 3) \
PF0(0)
PF0(2)
" .align 32 ;\n"
" 1: ;\n"
BLOCK(0)
BLOCK(4)
BLOCK(8)
BLOCK(12)
" addq %[inc], %[p1] ;\n"
" addq %[inc], %[p2] ;\n"
" addq %[inc], %[p3] ;\n"
" addq %[inc], %[p4] ;\n"
" addq %[inc], %[p5] ;\n"
" decl %[cnt] ; jnz 1b"
: [cnt] "+c" (lines),
[p1] "+r" (p1), [p2] "+r" (p2), [p3] "+r" (p3), [p4] "+r" (p4),
[p5] "+r" (p5)
: [inc] "r" (256UL)
: "memory");
kernel_fpu_end();
}
static struct xor_block_template xor_block_sse = {
.name = "generic_sse",
.do_2 = xor_sse_2,
.do_3 = xor_sse_3,
.do_4 = xor_sse_4,
.do_5 = xor_sse_5,
};
/* Also try the AVX routines */
#include "xor_avx.h"
#undef XOR_TRY_TEMPLATES
#define XOR_TRY_TEMPLATES \
do { \
AVX_XOR_SPEED; \
xor_speed(&xor_block_sse); \
} while (0)
/* We force the use of the SSE xor block because it can write around L2.
We may also be able to load into the L1 only depending on how the cpu
deals with a load to a line that is being prefetched. */
#define XOR_SELECT_TEMPLATE(FASTEST) \
AVX_SELECT(&xor_block_sse)
#endif /* _ASM_X86_XOR_64_H */
Reference in New Issue View Git Blame Copy Permalink