Commit a0ae2651 authored by Russell King's avatar Russell King

[ARM] Convert to constant-optimising udelay() implementation.

This allows us to eliminate a multiplication when we have a constant
delay value, as per x86.
parent da5508a5
...@@ -60,7 +60,8 @@ EXPORT_SYMBOL_ALIAS(fp_send_sig,send_sig); ...@@ -60,7 +60,8 @@ EXPORT_SYMBOL_ALIAS(fp_send_sig,send_sig);
EXPORT_SYMBOL(__backtrace); EXPORT_SYMBOL(__backtrace);
/* platform dependent support */ /* platform dependent support */
EXPORT_SYMBOL(udelay); EXPORT_SYMBOL(__udelay);
EXPORT_SYMBOL(__const_udelay);
/* networking */ /* networking */
EXPORT_SYMBOL(csum_partial); EXPORT_SYMBOL(csum_partial);
......
...@@ -16,15 +16,16 @@ LC0: .word loops_per_jiffy ...@@ -16,15 +16,16 @@ LC0: .word loops_per_jiffy
/* /*
* 0 <= r0 <= 2000 * 0 <= r0 <= 2000
*/ */
ENTRY(udelay) ENTRY(__udelay)
mov r2, #0x6800 mov r2, #0x6800
orr r2, r2, #0x00db orr r2, r2, #0x00db
mul r1, r0, r2 mul r0, r2, r0
ENTRY(__const_udelay) @ 0 <= r0 <= 0x01ffffff
ldr r2, LC0 ldr r2, LC0
ldr r2, [r2] ldr r2, [r2] @ max = 0x0fffffff
mov r1, r1, lsr #11 mov r0, r0, lsr #11 @ max = 0x00003fff
mov r2, r2, lsr #11 mov r2, r2, lsr #11 @ max = 0x0003ffff
mul r0, r1, r2 mul r0, r2, r0 @ max = 2^32-1
movs r0, r0, lsr #6 movs r0, r0, lsr #6
RETINSTR(moveq,pc,lr) RETINSTR(moveq,pc,lr)
......
#ifndef __ASM_ARM_DELAY_H
#define __ASM_ARM_DELAY_H
/* /*
* Copyright (C) 1995 Russell King * Copyright (C) 1995-2004 Russell King
* *
* Delay routines, using a pre-computed "loops_per_second" value. * Delay routines, using a pre-computed "loops_per_second" value.
*/ */
#ifndef __ASM_ARM_DELAY_H
#define __ASM_ARM_DELAY_H
extern void __delay(int loops); extern void __delay(int loops);
/*
* This function intentionally does not exist; if you see references to
* it, it means that you're calling udelay() with an out of range value.
*
* With currently imposed limits, this means that we support a max delay
* of 2000us and 671 bogomips
*/
extern void __bad_udelay(void);
/* /*
* division by multiplication: you don't have to worry about * division by multiplication: you don't have to worry about
* loss of precision. * loss of precision.
...@@ -19,14 +27,16 @@ extern void __delay(int loops); ...@@ -19,14 +27,16 @@ extern void __delay(int loops);
* first constant multiplications gets optimized away if the delay is * first constant multiplications gets optimized away if the delay is
* a constant) * a constant)
*/ */
extern void udelay(unsigned long usecs); extern void __udelay(unsigned long usecs);
extern void __const_udelay(unsigned long);
static inline unsigned long muldiv(unsigned long a, unsigned long b, unsigned long c) #define MAX_UDELAY_MS 2
{
return a * b / c;
}
#define udelay(n) \
(__builtin_constant_p(n) ? \
((n) > (MAX_UDELAY_MS * 1000) ? __bad_udelay() : \
__const_udelay((n) * 0x68dbul)) : \
__udelay(n))
#endif /* defined(_ARM_DELAY_H) */ #endif /* defined(_ARM_DELAY_H) */
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment