@@ -25,15 +25,11 @@
#endif
#ifdef CONFIG_ARC_HAS_LL64
-# define PREFETCH_READ(RX) prefetch [RX, 56]
-# define PREFETCH_WRITE(RX) prefetchw [RX, 64]
# define LOADX(DST,RX) ldd.ab DST, [RX, 8]
# define STOREX(SRC,RX) std.ab SRC, [RX, 8]
# define ZOLSHFT 5
# define ZOLAND 0x1F
#else
-# define PREFETCH_READ(RX) prefetch [RX, 28]
-# define PREFETCH_WRITE(RX) prefetchw [RX, 32]
# define LOADX(DST,RX) ld.ab DST, [RX, 4]
# define STOREX(SRC,RX) st.ab SRC, [RX, 4]
# define ZOLSHFT 4
@@ -41,8 +37,6 @@
#endif
ENTRY_CFI(memcpy)
- prefetch [r1] ; Prefetch the read location
- prefetchw [r0] ; Prefetch the write location
mov.f 0, r2
;;; if size is zero
jz.d [blink]
@@ -72,8 +66,6 @@ ENTRY_CFI(memcpy)
lpnz @.Lcopy32_64bytes
;; LOOP START
LOADX (r6, r1)
- PREFETCH_READ (r1)
- PREFETCH_WRITE (r3)
LOADX (r8, r1)
LOADX (r10, r1)
LOADX (r4, r1)
@@ -117,9 +109,7 @@ ENTRY_CFI(memcpy)
lpnz @.Lcopy8bytes_1
;; LOOP START
ld.ab r6, [r1, 4]
- prefetch [r1, 28] ;Prefetch the next read location
ld.ab r8, [r1,4]
- prefetchw [r3, 32] ;Prefetch the next write location
SHIFT_1 (r7, r6, 24)
or r7, r7, r5
@@ -162,9 +152,7 @@ ENTRY_CFI(memcpy)
lpnz @.Lcopy8bytes_2
;; LOOP START
ld.ab r6, [r1, 4]
- prefetch [r1, 28] ;Prefetch the next read location
ld.ab r8, [r1,4]
- prefetchw [r3, 32] ;Prefetch the next write location
SHIFT_1 (r7, r6, 16)
or r7, r7, r5
@@ -204,9 +192,7 @@ ENTRY_CFI(memcpy)
lpnz @.Lcopy8bytes_3
;; LOOP START
ld.ab r6, [r1, 4]
- prefetch [r1, 28] ;Prefetch the next read location
ld.ab r8, [r1,4]
- prefetchw [r3, 32] ;Prefetch the next write location
SHIFT_1 (r7, r6, 8)
or r7, r7, r5
ARCv2 optimized memcpy uses PREFETCHW instruction for prefetching the next cache line but doesn't ensure that the line is not past the end of the buffer. PRETECHW changes the line ownership and marks it dirty, which can cause data corruption if this area is used for DMA IO. Fix the issue by avoiding the PREFETCHW. This leads to performance degradation but it is OK as we'll introduce new memcpy implementation optimized for unaligned memory access using. We also cut off all PREFETCH instructions at they are quite useless according to benchmark. Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com> --- arch/arc/lib/memcpy-archs.S | 14 -------------- 1 file changed, 14 deletions(-)