From 563b6ee873b898c0f3e3671cf6adaf91def5d92a Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Mon, 27 Mar 2017 14:59:06 -0700 Subject: [PATCH] Rewrite the byte swapping macros. The clever pointer tricks were actually not working, and we were doing the byte-by-byte moves in general. By just doing the memcpy and obvious byte swap code, we end up generating actual byte swap instructions, thanks to optimizing compilers. text data bss dec hex filename before: 2240807 51552 132016 2424375 24fe37 hw/xfree86/Xorg after: 2215167 51552 132016 2398735 249a0f hw/xfree86/Xorg Signed-off-by: Eric Anholt Reviewed-by: Keith Packard --- doc/Xserver-spec.xml | 2 +- glx/glxbyteorder.h | 21 ---------- include/misc.h | 97 ++++++++++++++++---------------------------- os/io.c | 4 +- 4 files changed, 37 insertions(+), 87 deletions(-) diff --git a/doc/Xserver-spec.xml b/doc/Xserver-spec.xml index 7867544e4..3dde65178 100644 --- a/doc/Xserver-spec.xml +++ b/doc/Xserver-spec.xml @@ -600,7 +600,7 @@ are: REQUEST, REQUEST_SIZE_MATCH, REQUEST_AT_LEAST_SIZE, REQUEST_FIXED_SIZE, LEGAL_NEW_RESOURCE, and VALIDATE_DRAWABLE_AND_GC. Useful byte swapping macros can be found in Xserver/include/dix.h: WriteReplyToClient and WriteSwappedDataToClient; and -in Xserver/include/misc.h: lswapl, lswaps, LengthRestB, LengthRestS, +in Xserver/include/misc.h: bswap_64, bswap_32, bswap_16, LengthRestB, LengthRestS, LengthRestL, SwapRestS, SwapRestL, swapl, swaps, cpswapl, and cpswaps. diff --git a/glx/glxbyteorder.h b/glx/glxbyteorder.h index 5e94e8626..8f0cd8a4b 100644 --- a/glx/glxbyteorder.h +++ b/glx/glxbyteorder.h @@ -37,25 +37,4 @@ #include "misc.h" -static inline uint16_t -bswap_16(uint16_t val) -{ - swap_uint16(&val); - return val; -} - -static inline uint32_t -bswap_32(uint32_t val) -{ - swap_uint32(&val); - return val; -} - -static inline uint64_t -bswap_64(uint64_t val) -{ - swap_uint64(&val); - return val; -} - #endif /* !defined(__GLXBYTEORDER_H__) */ diff --git a/include/misc.h b/include/misc.h index 01747fd38..a75eb617c 100644 --- a/include/misc.h +++ b/include/misc.h @@ -128,21 +128,6 @@ typedef struct _xReq *xReqPtr; #define USE_BACKGROUND_PIXEL 3 #define USE_BORDER_PIXEL 3 -/* byte swap a 32-bit literal */ -static inline uint32_t -lswapl(uint32_t x) -{ - return ((x & 0xff) << 24) | - ((x & 0xff00) << 8) | ((x & 0xff0000) >> 8) | ((x >> 24) & 0xff); -} - -/* byte swap a 16-bit literal */ -static inline uint16_t -lswaps(uint16_t x) -{ - return (uint16_t)((x & 0xff) << 8) | ((x >> 8) & 0xff); -} - #undef min #undef max @@ -311,88 +296,74 @@ __builtin_constant_p(int x) } #endif -/* byte swap a 64-bit value */ -static inline void -swap_uint64(uint64_t *x) +static inline uint64_t +bswap_64(uint64_t x) { - char n; - - n = ((char *) x)[0]; - ((char *) x)[0] = ((char *) x)[7]; - ((char *) x)[7] = n; - - n = ((char *) x)[1]; - ((char *) x)[1] = ((char *) x)[6]; - ((char *) x)[6] = n; - - n = ((char *) x)[2]; - ((char *) x)[2] = ((char *) x)[5]; - ((char *) x)[5] = n; - - n = ((char *) x)[3]; - ((char *) x)[3] = ((char *) x)[4]; - ((char *) x)[4] = n; + return (((x & 0xFF00000000000000ull) >> 56) | + ((x & 0x00FF000000000000ull) >> 40) | + ((x & 0x0000FF0000000000ull) >> 24) | + ((x & 0x000000FF00000000ull) >> 8) | + ((x & 0x00000000FF000000ull) << 8) | + ((x & 0x0000000000FF0000ull) << 24) | + ((x & 0x000000000000FF00ull) << 40) | + ((x & 0x00000000000000FFull) << 56)); } #define swapll(x) do { \ + uint64_t temp; \ if (sizeof(*(x)) != 8) \ wrong_size(); \ - swap_uint64((uint64_t *)(x)); \ + memcpy(&temp, x, 8); \ + temp = bswap_64(temp); \ + memcpy(x, &temp, 8); \ } while (0) -/* byte swap a 32-bit value */ -static inline void -swap_uint32(uint32_t * x) +static inline uint32_t +bswap_32(uint32_t x) { - char n = ((char *) x)[0]; - - ((char *) x)[0] = ((char *) x)[3]; - ((char *) x)[3] = n; - n = ((char *) x)[1]; - ((char *) x)[1] = ((char *) x)[2]; - ((char *) x)[2] = n; + return (((x & 0xFF000000) >> 24) | + ((x & 0x00FF0000) >> 8) | + ((x & 0x0000FF00) << 8) | + ((x & 0x000000FF) << 24)); } #define swapl(x) do { \ + uint32_t temp; \ if (sizeof(*(x)) != 4) \ wrong_size(); \ - if (__builtin_constant_p((uintptr_t)(x) & 3) && ((uintptr_t)(x) & 3) == 0) \ - *(x) = lswapl(*(x)); \ - else \ - swap_uint32((uint32_t *)(x)); \ + memcpy(&temp, x, 4); \ + temp = bswap_32(temp); \ + memcpy(x, &temp, 4); \ } while (0) -/* byte swap a 16-bit value */ -static inline void -swap_uint16(uint16_t * x) +static inline uint16_t +bswap_16(uint16_t x) { - char n = ((char *) x)[0]; - - ((char *) x)[0] = ((char *) x)[1]; - ((char *) x)[1] = n; + return (((x & 0xFF00) >> 8) | + ((x & 0x00FF) << 8)); } #define swaps(x) do { \ + uint16_t temp; \ if (sizeof(*(x)) != 2) \ wrong_size(); \ - if (__builtin_constant_p((uintptr_t)(x) & 1) && ((uintptr_t)(x) & 1) == 0) \ - *(x) = lswaps(*(x)); \ - else \ - swap_uint16((uint16_t *)(x)); \ + memcpy(&temp, x, 2); \ + temp = bswap_16(temp); \ + memcpy(x, &temp, 2); \ } while (0) /* copy 32-bit value from src to dst byteswapping on the way */ #define cpswapl(src, dst) do { \ if (sizeof((src)) != 4 || sizeof((dst)) != 4) \ wrong_size(); \ - (dst) = lswapl((src)); \ + (dst) = bswap_32((src)); \ } while (0) /* copy short from src to dst byteswapping on the way */ #define cpswaps(src, dst) do { \ if (sizeof((src)) != 2 || sizeof((dst)) != 2) \ wrong_size(); \ - (dst) = lswaps((src)); \ + (dst) = bswap_16((src)); \ } while (0) extern _X_EXPORT void SwapLongs(CARD32 *list, unsigned long count); diff --git a/os/io.c b/os/io.c index 8aa51a107..46c7e2371 100644 --- a/os/io.c +++ b/os/io.c @@ -108,12 +108,12 @@ static ConnectionOutputPtr FreeOutputs = (ConnectionOutputPtr) NULL; static OsCommPtr AvailableInput = (OsCommPtr) NULL; #define get_req_len(req,cli) ((cli)->swapped ? \ - lswaps((req)->length) : (req)->length) + bswap_16((req)->length) : (req)->length) #include #define get_big_req_len(req,cli) ((cli)->swapped ? \ - lswapl(((xBigReq *)(req))->length) : \ + bswap_32(((xBigReq *)(req))->length) : \ ((xBigReq *)(req))->length) #define BUFSIZE 16384