diff options
Diffstat (limited to 'src/xine-utils/ppcasm_string.S')
| -rw-r--r-- | src/xine-utils/ppcasm_string.S | 86 |
1 files changed, 33 insertions, 53 deletions
diff --git a/src/xine-utils/ppcasm_string.S b/src/xine-utils/ppcasm_string.S index ff4013e07..3091a9d68 100644 --- a/src/xine-utils/ppcasm_string.S +++ b/src/xine-utils/ppcasm_string.S @@ -16,16 +16,11 @@ #ifdef ARCH_PPC -#warning Be forewarned - using PowerPC assembly - -#define __KERNEL__ -#define __ASSEMBLY__ +//#warning Be forewarned - using PowerPC assembly #include "ppc_asm.tmpl" -#include <linux/config.h> -#include <asm/processor.h> -#include <asm/cache.h> -#include <asm/errno.h> +#define N_FUN 36 +#define N_SO 100 #define COPY_16_BYTES \ lwz r7,4(r4); \ @@ -37,60 +32,42 @@ stw r9,12(r6); \ stwu r10,16(r6) -#define COPY_16_BYTES_WITHEX(n) \ -8 ## n ## 0: \ - lwz r7,4(r4); \ -8 ## n ## 1: \ - lwz r8,8(r4); \ -8 ## n ## 2: \ - lwz r9,12(r4); \ -8 ## n ## 3: \ - lwzu r10,16(r4); \ -8 ## n ## 4: \ - stw r7,4(r6); \ -8 ## n ## 5: \ - stw r8,8(r6); \ -8 ## n ## 6: \ - stw r9,12(r6); \ -8 ## n ## 7: \ - stwu r10,16(r6) +#define __stringify_1(x) #x +#define __stringify(x) __stringify_1(x) -#define COPY_16_BYTES_EXCODE(n) \ -9 ## n ## 0: \ - addi r5,r5,-(16 * n); \ - b 104f; \ -9 ## n ## 1: \ - addi r5,r5,-(16 * n); \ - b 105f; \ -.section __ex_table,"a"; \ - .align 2; \ - .long 8 ## n ## 0b,9 ## n ## 0b; \ - .long 8 ## n ## 1b,9 ## n ## 0b; \ - .long 8 ## n ## 2b,9 ## n ## 0b; \ - .long 8 ## n ## 3b,9 ## n ## 0b; \ - .long 8 ## n ## 4b,9 ## n ## 1b; \ - .long 8 ## n ## 5b,9 ## n ## 1b; \ - .long 8 ## n ## 6b,9 ## n ## 1b; \ - .long 8 ## n ## 7b,9 ## n ## 1b; \ - .text +#define _GLOBFN(n)\ + .stabs __stringify(n:F-1),N_FUN,0,0,n;\ + .type n,@function; \ + .globl n;\ +n: + +#define _SIZE(n) \ + .size n, .-n .text + .stabs "src/xine-utils",N_SO,0,0,. + .stabs "ppcasm_string.S",N_SO,0,0,. + +#warning FIXME: Get cache line sizes from /proc +#define L1_CACHE_LINE_SIZE 32 CACHELINE_BYTES = 32 LG_CACHELINE_BYTES = 5 CACHELINE_MASK = (32 -1) - - .global ppcasm_cacheable_memcpy -ppcasm_cacheable_memcpy: -#if 0 /* this part causes "error loading shared library: unexpected reloc type - 0x0b (???) */ +/* + * This version uses dcbz on the complete cache lines in the + * destination area to reduce memory traffic. This requires that + * the destination area is cacheable. + * We only use this version if the source and dest don't overlap. + * -- paulus. + */ +_GLOBFN(ppcasm_cacheable_memcpy) add r7,r3,r5 /* test if the src & dst overlap */ add r8,r4,r5 cmplw 0,r4,r7 cmplw 1,r3,r8 crand 0,0,4 /* cr0.lt &= cr1.lt */ - blt ppcasm_memcpy /* if regions overlap */ -#endif + blt 66f //ppcasm_memcpy /* if regions overlap */ addi r4,r4,-4 addi r6,r3,-4 neg r0,r3 @@ -157,9 +134,10 @@ ppcasm_cacheable_memcpy: bdnz 40b 65: blr - .globl ppcasm_memcpy -ppcasm_memcpy: - srwi. r7,r5,3 +_SIZE(ppcasm_cacheable_memcpy) + +_GLOBFN(ppcasm_memcpy) +66: srwi. r7,r5,3 addi r6,r3,-4 addi r4,r4,-4 beq 2f /* if less than 8 bytes to do */ @@ -199,4 +177,6 @@ ppcasm_memcpy: mtctr r7 b 1b +_SIZE(ppcasm_memcpy) + #endif /* ARCH_PPC */ |