target/linux/generic/pending-4.9/306-mips_mem_functions_performance.patch

   1 From: Felix Fietkau <nbd@nbd.name>
   2 Subject: [PATCH] mips: allow the compiler to optimize memset, memcmp, memcpy  for better performance and (in some instances) smaller code
   3
   4 lede-commit: 07e59c7bc7f375f792ec9734be42fe4fa391a8bb
   5 Signed-off-by: Felix Fietkau <nbd@nbd.name>
   6 ---
   7  arch/mips/boot/compressed/Makefile |  3 ++-
   8  arch/mips/include/asm/string.h     | 38 ++++++++++++++++++++++++++++++++++++++
   9  arch/mips/lib/Makefile             |  2 +-
  10  arch/mips/lib/memcmp.c             | 22 ++++++++++++++++++++++
  11  4 files changed, 63 insertions(+), 2 deletions(-)
  12  create mode 100644 arch/mips/lib/memcmp.c
  13
  14 diff --git a/arch/mips/boot/compressed/Makefile b/arch/mips/boot/compressed/Makefile
  15 index 90aca95fe314..3cd3b391ef49 100644
  16 --- a/arch/mips/boot/compressed/Makefile
  17 +++ b/arch/mips/boot/compressed/Makefile
  18 @@ -23,7 +23,8 @@ KBUILD_CFLAGS := $(shell echo $(KBUILD_CFLAGS) | sed -e "s/-pg//")
  19  KBUILD_CFLAGS := $(filter-out -fstack-protector, $(KBUILD_CFLAGS))
  20
  21  KBUILD_CFLAGS := $(LINUXINCLUDE) $(KBUILD_CFLAGS) -D__KERNEL__ \
  22 -       -DBOOT_HEAP_SIZE=$(BOOT_HEAP_SIZE) -D"VMLINUX_LOAD_ADDRESS_ULL=$(VMLINUX_LOAD_ADDRESS)ull"
  23 +       -DBOOT_HEAP_SIZE=$(BOOT_HEAP_SIZE) -D"VMLINUX_LOAD_ADDRESS_ULL=$(VMLINUX_LOAD_ADDRESS)ull" \
  24 +       -D__ZBOOT__
  25
  26  KBUILD_AFLAGS := $(LINUXINCLUDE) $(KBUILD_AFLAGS) -D__ASSEMBLY__ \
  27         -DBOOT_HEAP_SIZE=$(BOOT_HEAP_SIZE) \
  28 diff --git a/arch/mips/include/asm/string.h b/arch/mips/include/asm/string.h
  29 index 29030cb398ee..7b737f9b6d58 100644
  30 --- a/arch/mips/include/asm/string.h
  31 +++ b/arch/mips/include/asm/string.h
  32 @@ -140,4 +140,42 @@ extern void *memcpy(void *__to, __const__ void *__from, size_t __n);
  33  #define __HAVE_ARCH_MEMMOVE
  34  extern void *memmove(void *__dest, __const__ void *__src, size_t __n);
  35
  36 +#ifndef __ZBOOT__
  37 +#define memset(__s, __c, len)                                  \
  38 +({                                                             \
  39 +       size_t __len = (len);                                   \
  40 +       void *__ret;                                            \
  41 +       if (__builtin_constant_p(len) && __len >= 64)           \
  42 +               __ret = memset((__s), (__c), __len);            \
  43 +       else                                                    \
  44 +               __ret = __builtin_memset((__s), (__c), __len);  \
  45 +       __ret;                                                  \
  46 +})
  47 +
  48 +#define memcpy(dst, src, len)                                  \
  49 +({                                                             \
  50 +       size_t __len = (len);                                   \
  51 +       void *__ret;                                            \
  52 +       if (__builtin_constant_p(len) && __len >= 64)           \
  53 +               __ret = memcpy((dst), (src), __len);            \
  54 +       else                                                    \
  55 +               __ret = __builtin_memcpy((dst), (src), __len);  \
  56 +       __ret;                                                  \
  57 +})
  58 +
  59 +#define memmove(dst, src, len)                                 \
  60 +({                                                             \
  61 +       size_t __len = (len);                                   \
  62 +       void *__ret;                                            \
  63 +       if (__builtin_constant_p(len) && __len >= 64)           \
  64 +               __ret = memmove((dst), (src), __len);           \
  65 +       else                                                    \
  66 +               __ret = __builtin_memmove((dst), (src), __len); \
  67 +       __ret;                                                  \
  68 +})
  69 +
  70 +#define __HAVE_ARCH_MEMCMP
  71 +#define memcmp(src1, src2, len) __builtin_memcmp((src1), (src2), (len))
  72 +#endif
  73 +
  74  #endif /* _ASM_STRING_H */
  75 diff --git a/arch/mips/lib/Makefile b/arch/mips/lib/Makefile
  76 index 0344e575f522..33a0211e954a 100644
  77 --- a/arch/mips/lib/Makefile
  78 +++ b/arch/mips/lib/Makefile
  79 @@ -4,7 +4,7 @@
  80
  81  lib-y  += bitops.o csum_partial.o delay.o memcpy.o memset.o \
  82            mips-atomic.o strlen_user.o strncpy_user.o \
  83 -          strnlen_user.o uncached.o
  84 +          strnlen_user.o uncached.o memcmp.o
  85
  86  obj-y                  += iomap.o
  87  obj-$(CONFIG_PCI)      += iomap-pci.o
  88 diff --git a/arch/mips/lib/memcmp.c b/arch/mips/lib/memcmp.c
  89 new file mode 100644
  90 index 000000000000..35ef1646286e
  91 --- /dev/null
  92 +++ b/arch/mips/lib/memcmp.c
  93 @@ -0,0 +1,22 @@
  94 +/*
  95 + *  copied from linux/lib/string.c
  96 + *
  97 + *  Copyright (C) 1991, 1992  Linus Torvalds
  98 + */
  99 +
 100 +#include <linux/module.h>
 101 +#include <linux/string.h>
 102 +
 103 +#undef memcmp
 104 +int memcmp(const void *cs, const void *ct, size_t count)
 105 +{
 106 +       const unsigned char *su1, *su2;
 107 +       int res = 0;
 108 +
 109 +       for (su1 = cs, su2 = ct; 0 < count; ++su1, ++su2, count--)
 110 +               if ((res = *su1 - *su2) != 0)
 111 +                       break;
 112 +       return res;
 113 +}
 114 +EXPORT_SYMBOL(memcmp);
 115 +
 116 --
 117 2.11.0
 118