From 0b610097b4342a5fdf6070615b736bae030a4e05 Mon Sep 17 00:00:00 2001
From: mykola2312 <49044616+mykola2312@users.noreply.github.com>
Date: Wed, 11 May 2022 18:48:46 +0300
Subject: [PATCH] add achitecture-dependent assembly to build, implement
 hardware byte-swapping

---
 CMakeLists.txt |  6 ++++--
 arch/x86_64.S  | 21 +++++++++++++++++++++
 endian.c       | 25 +++++++------------------
 test.c         |  5 +++--
 4 files changed, 35 insertions(+), 22 deletions(-)
 create mode 100644 arch/x86_64.S

diff --git a/CMakeLists.txt b/CMakeLists.txt
index e1e629a..84f112a 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1,5 +1,5 @@
 cmake_minimum_required(VERSION 3.10)
-project(cutil)
+project(cutil C ASM)
 
 set(HEADERS
     cutypes.h
@@ -20,7 +20,9 @@ set(SOURCES
     struct.c
 )
 
-add_library(cutil STATIC ${SOURCES} ${HEADERS})
+add_library(cutil STATIC ${SOURCES} ${HEADERS}
+    "arch/${CMAKE_SYSTEM_PROCESSOR}.S"
+)
 
 add_executable(test test.c)
 target_link_libraries(test cutil)
\ No newline at end of file
diff --git a/arch/x86_64.S b/arch/x86_64.S
new file mode 100644
index 0000000..2bc995f
--- /dev/null
+++ b/arch/x86_64.S
@@ -0,0 +1,21 @@
+.globl hw_bswap16
+.globl hw_bswap32
+.globl hw_bswap64
+
+.text
+
+hw_bswap16:
+    mov %edi, %eax
+    xchg %al, %ah
+    ret
+
+hw_bswap32:
+    mov %edi, %eax
+    bswap %eax
+    ret
+
+hw_bswap64:
+    mov %rdi, %rax
+    bswap %rax
+    ret
+
diff --git a/endian.c b/endian.c
index d86904c..1cf1d6f 100644
--- a/endian.c
+++ b/endian.c
@@ -46,29 +46,18 @@ static uint64_t sw_bswap64(uint64_t val)
 }
 
 #if (CU_ARCH == x86_32)
-static uint32_t hw_bswap32(uint32_t val)
-{
-    __asm__ ("bswap %0" : "=r" (val) : "r" (val));
-    return val;
-}
+extern uint16_t hw_bswap16(uint16_t val);
+extern uint32_t hw_bswap32(uint32_t val);
 
-bswap16_t cu_bswap16 = sw_bswap16;
+bswap16_t cu_bswap16 = hw_bswap16;
 bswap32_t cu_bswap32 = hw_bswap32;
 bswap64_t cu_bswap64 = sw_bswap64;
 #elif (CU_ARCH == x86_64)
-static uint32_t hw_bswap32(uint32_t val)
-{
-    __asm__ ("bswap %0" : "=r" (val) : "r" (val));
-    return val;
-}
+extern uint16_t hw_bswap16(uint16_t val);
+extern uint32_t hw_bswap32(uint32_t val);
+extern uint64_t hw_bswap64(uint64_t val);
 
-static uint64_t hw_bswap64(uint64_t val)
-{
-    __asm__ ("bswap %0" : "=r" (val) : "r" (val));
-    return val;
-}
-
-bswap16_t cu_bswap16 = sw_bswap16;
+bswap16_t cu_bswap16 = hw_bswap16;
 bswap32_t cu_bswap32 = hw_bswap32;
 bswap64_t cu_bswap64 = hw_bswap64;
 #else
diff --git a/test.c b/test.c
index 1c0c2a9..78bd65d 100644
--- a/test.c
+++ b/test.c
@@ -111,8 +111,9 @@ int main()
 
     printf("[endian]\n");
     printf("cutil_endian\t%u\n", cu_endian);
-    printf("%x\t%x\n", 0x12345678, l32toh32(0x12345678));
-    printf("%lx\t%lx\n", 0x12345678abcd8765, h64ton64(0x12345678abcd8765));
+    printf("%x\t%x\n", 0x1234, cu_bswap16(0x1234));
+    printf("%x\t%x\n", 0x12345678, cu_bswap32(0x12345678));
+    printf("%lx\t%lx\n", 0x12345678abcd8765, cu_bswap64(0x12345678abcd8765));
 
     printf("[struct]\n");
     printf("value_offset\t%u\n", value_offset(CU_STRUCT(test_s), 4, &test));