Improved SCons script for building the userspace stdlib.

targets/_any/stdlib/SConscript

@@ -1,16 +1,20 @@
 
 Import('env')
 
-shared_sources = Split('''
+env['CRTI_PATH'] = env.File(f'src/crti.{env["TARGET_ARCH"]}.s').abspath
+env['CRTN_PATH'] = env.File(f'src/crtn.{env["TARGET_ARCH"]}.s').abspath
+env['CRT0_PATH'] = env.File(f'src/crt0.{env["TARGET_ARCH"]}.s').abspath
+
+stdlib_sources = Split(f'''
     src/assert.cpp
     src/stdio.cpp
     src/stdlib.cpp
     src/string.cpp
+
+    src/memory.{env['TARGET_ARCH']}.s
 ''')
 
-kernel_sources = shared_sources
-
-env.Append(KERNEL_SOURCES = [env.File(f) for f in kernel_sources])
+env.Append(KERNEL_SOURCES = [env.File(f) for f in stdlib_sources])
 
 ### Stdlib
 stdlib_env = env.Clone()
@@ -18,20 +22,17 @@ stdlib_env['AS'] = 'x86_64-baos-elf-as'
 stdlib_env['CC'] = 'x86_64-baos-elf-gcc'
 stdlib_env['CXX'] = 'x86_64-baos-elf-g++'
 stdlib_env['LD'] = 'x86_64-baos-elf-g++'
+stdlib_env['OBJSUFFIX'] = f'.baos{stdlib_env["OBJSUFFIX"]}'
 stdlib_env.Append(CXXFLAGS = ['-fno-exceptions', '-fno-rtti', '-std=c++20'])
 stdlib_env.Append(CPPPATH = ['#targets/_any/bastl/include', '#targets/_any/stdlib/include', '#targets/_any/kernel/include'])
 
-stdlib_sources = Split('''
-    src/crt0.s
-''')
-for src in shared_sources:
-    new_src = stdlib_env.File(f'#.stdlib_variant/{stdlib_env.File(src).name}').abspath
-    stdlib_env.Command(new_src, src, Copy("$TARGET", "$SOURCE"))
-    stdlib_sources.append(new_src)
 lib_stdlib = stdlib_env.StaticLibrary(
-    target = stdlib_env.File('#stdlib.a'),
+    target = stdlib_env.File('#stdlib/stdlib.a'),
     source = stdlib_sources
 )
-stdlib_env.Default(lib_stdlib)
+crti_o = stdlib_env.Object(target = '#stdlib/crti.o', source = stdlib_env['CRTI_PATH'])
+crtn_o = stdlib_env.Object(target = '#stdlib/crtn.o', source = stdlib_env['CRTN_PATH'])
+crt0_o = stdlib_env.Object(target = '#stdlib/crt0.o', source = stdlib_env['CRT0_PATH'])
+stdlib_env.Default([lib_stdlib, crti_o, crtn_o, crt0_o])
 
 Return('env')

targets/_any/stdlib/src/crti.x86_64.s

@@ -0,0 +1,16 @@
+/* x86_64 crti.s */
+.section .init
+.global _init
+.type _init, @function
+_init:
+	push %rbp
+	movq %rsp, %rbp
+	/* gcc will nicely put the contents of crtbegin.o's .init section here. */
+
+.section .fini
+.global _fini
+.type _fini, @function
+_fini:
+	push %rbp
+	movq %rsp, %rbp
+	/* gcc will nicely put the contents of crtbegin.o's .fini section here. */

targets/_any/stdlib/src/crtn.x86_64.s

@@ -0,0 +1,10 @@
+/* x86_64 crtn.s */
+.section .init
+	/* gcc will nicely put the contents of crtend.o's .init section here. */
+	popq %rbp
+	ret
+
+.section .fini
+	/* gcc will nicely put the contents of crtend.o's .fini section here. */
+	popq %rbp
+	ret

targets/_any/stdlib/src/memory.x86_64.s

@@ -0,0 +1,66 @@
+
+.section .text
+.global memcpy
+.type memcpy @function
+
+// void* memcpy(void* dest [%rdi], void* src [%rsi], size_t count [%rdx])
+//
+memcpy:
+    movq %rdi, %rax // return value
+memcpy_qwords:
+    cmpq $8, %rdx      // check if the remaining bytes are at least 8
+    jl memcpy_bytes    // if not copy the remaining bytes per byte
+    movq (%rsi), %rcx  // if yes, copy using movq
+    movq %rcx, (%rdi)
+    addq $8, %rsi
+    addq $8, %rdi
+    subq $8, %rdx
+    jmp memcpy_qwords
+memcpy_bytes:
+    cmpq $0, %rdx
+    je memcpy_end
+    movb (%rsi), %ch
+    movb %ch, (%rdi)
+    addq $1, %rsi
+    addq $1, %rdi
+    subq $1, %rdx
+    jmp memcpy_bytes
+memcpy_end:
+    ret
+
+.global memmove
+.type memmove @function
+// void* memmove(void* dest [%rdi], void* src [%rsi], size_t count [%rdx])
+//
+memmove:
+    movq %rdi, %rax     // preserve dest as return value
+    cmpq %rsi, %rdi     // check if dest > src
+    jg memmove_backward // if yes, do everything backwards
+    movq %rsi, %rcx     // check if (src - dest) < 8
+    subq %rdi, %rcx
+    cmpq $8, %rdi
+    jl memcpy_bytes     // if yes, we have to do a bytewise copy
+    jmp memcpy_qwords   // otherwise copy whole qwords
+memmove_backward:       // dest > src, copy backwards
+    addq %rdx, %rdi     // dest = dest + count
+    addq %rdx, %rsi     // src = src + count
+memmove_qwords:
+    cmpq $8, %rdx        // check if the remaining bytes are at least 8
+    jl memmove_bytes    // if not copy the remaining bytes per byte
+    movq -8(%rsi), %rcx // if yes, copy using movq
+    movq %rcx, -8(%rdi)
+    subq $8, %rsi
+    subq $8, %rdi
+    subq $8, %rdx
+    jmp memmove_qwords
+memmove_bytes:
+    cmpq $0, %rdx
+    je memmove_end
+    movb (%rsi), %ch
+    movb %ch, (%rdi)
+    subq $1, %rsi
+    subq $1, %rdi
+    subq $1, %rdx
+    jmp memmove_bytes
+memmove_end:
+    ret