this repo has no description
More progress
+436
-702
+12
-2
src/duct/include/commpage.h
+12
-2
src/duct/include/commpage.h
···
7
7
extern uint8_t __commpage[];
8
8
#endif
9
9
10
-
#define _COMM_PAGE_BASE_ADDRESS (__commpage)
10
+
//#ifndef __ASSEMBLER__
11
+
//#define _COMM_PAGE_BASE_ADDRESS (__commpage)
12
+
//#else
13
+
//#define _COMM_PAGE_BASE_ADDRESS ___commpage@GOTPCREL(%rip)
14
+
//#endif
11
15
#define _COMM_PAGE_AREA_LENGTH ( 1 * 4096 ) /* reserved length of entire comm area */
12
-
//#define _COMM_PAGE_BASE_ADDRESS ( 0xffff0000 ) /* base address of allocated memory */
16
+
17
+
#ifdef __x86_64__
18
+
#define _COMM_PAGE_BASE_ADDRESS ( 0x7fffffe00000 ) /* base address of allocated memory */
19
+
#else
20
+
#define _COMM_PAGE_BASE_ADDRESS ( 0xaaaa0000 ) /* base address of allocated memory */
21
+
#endif
22
+
13
23
#define _COMM_PAGE_START_ADDRESS ( _COMM_PAGE_BASE_ADDRESS ) /* address traditional commpage code starts on */
14
24
#define _COMM_PAGE_AREA_USED ( 1 * 4096 ) /* this is the amt actually allocated */
15
25
#define _COMM_PAGE_SIGS_OFFSET 0x8000 /* offset to routine signatures */
+1
-9
src/duct/src/CMakeLists.txt
+1
-9
src/duct/src/CMakeLists.txt
···
2
2
3
3
cmake_minimum_required(VERSION 2.4.0)
4
4
5
-
if(CMAKE_SIZEOF_VOID_P EQUAL 4 OR CMAKE_INSTALL_LIBDIR STREQUAL "lib32")
6
-
set(BITS 32)
7
-
add_definitions(-DTARGET_CPU_X86=1)
8
-
else(CMAKE_SIZEOF_VOID_P EQUAL 4 OR CMAKE_INSTALL_LIBDIR STREQUAL "lib32")
9
-
set(BITS 64)
10
-
add_definitions(-DTARGET_CPU_X86_64=1)
11
-
endif(CMAKE_SIZEOF_VOID_P EQUAL 4 OR CMAKE_INSTALL_LIBDIR STREQUAL "lib32")
12
-
13
5
add_definitions(-DTARGET_OS_MAC=1)
14
6
add_definitions(-DHAVE_STDINT_H=1)
15
7
add_definitions(-D__APPLE__ -D__DYNAMIC__)
16
8
add_definitions(-D__ENVIRONMENT_MAC_OS_X_VERSION_MIN_REQUIRED__=1080)
17
9
18
10
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -nostdinc -D__DARWIN_UNIX03 -fPIC -w")
19
-
set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -nostdlib")
11
+
set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -nostdlib -Wl,-flat_namespace -Wl,-undefined,suppress")
20
12
21
13
SET(CMAKE_INSTALL_RPATH "${CMAKE_INSTALL_PREFIX}/${CMAKE_INSTALL_LIBDIR}/darling")
22
14
SET(CMAKE_BUILD_WITH_INSTALL_RPATH TRUE)
+7
-3
src/duct/src/commpage.c
+7
-3
src/duct/src/commpage.c
···
1
1
#include <commpage.h>
2
+
#include <sys/mman.h>
2
3
3
-
uint8_t __commpage[_COMM_PAGE_AREA_LENGTH];
4
+
//uint8_t __commpage[_COMM_PAGE_AREA_LENGTH];
4
5
5
6
__attribute__((constructor))
6
7
void _darling_initialize_commpage(void)
7
8
{
8
-
*(_COMM_PAGE_NCPUS) = 1;
9
-
*(_COMM_PAGE_VERSION) = 12;
9
+
mmap(_COMM_PAGE_BASE_ADDRESS, _COMM_PAGE_AREA_LENGTH, PROT_READ|PROT_WRITE,
10
+
MAP_ANON | MAP_PRIVATE, -1, 0);
11
+
12
+
*((uint8_t*)_COMM_PAGE_NCPUS) = 1;
13
+
*((uint8_t*)_COMM_PAGE_VERSION) = 12;
10
14
*((uint64_t*) _COMM_PAGE_MEMORY_SIZE) = 1024LL*1024LL*1024LL;
11
15
}
12
16
+2
-3
src/libc/CMakeLists.txt
+2
-3
src/libc/CMakeLists.txt
···
19
19
#add_definitions("'-D__weak_reference(sym,alias)=__asm__(\".weak \" \\\#alias );__asm__(\".equ \" \\\#alias \", \" \\\#sym)'")
20
20
#add_definitions("'-D__weak_reference(sym,alias)='")
21
21
add_definitions("'-D__warn_references(name,warning)='")
22
-
add_definitions(-Ddladdr=__darwin_dladdr)
22
+
#add_definitions(-Ddladdr=__darwin_dladdr)
23
23
add_definitions(-DBUILDING_LIBC)
24
24
25
25
# Hide warnings that get spammed
···
35
35
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -nostdinc -fPIC -w ")
36
36
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -include ${CMAKE_CURRENT_SOURCE_DIR}/weak_reference.h")
37
37
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -ggdb -O0") # development flags
38
-
set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -nostdlib -Wl,--version-script=${DARLING_TOP_DIRECTORY}/darwin.map -Bsymbolic-functions")
39
-
set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -Wl,--defsym=__darwin_pthread_self=pthread_self")
38
+
set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -nostdlib")
40
39
41
40
include_directories("${CMAKE_CURRENT_SOURCE_DIR}/fbsdcompat")
42
41
include_directories("${CMAKE_CURRENT_SOURCE_DIR}/pthreads")
+17
-61
src/libc/gen/platfunc.h
+17
-61
src/libc/gen/platfunc.h
···
1
-
// Modified by Lubos Dolezel for Darling
2
-
3
1
/*
4
2
* Copyright (c) 2003-2010 Apple Inc. All rights reserved.
5
3
*
···
40
38
// This is a shared macro which calls PLATFUNC_VARIANT_NAME which has different
41
39
// implementations in __ASSEMBLER__ and !__ASSEMBLER__
42
40
#define PLATFUNC_DESCRIPTOR_NAME(name, variant) \
43
-
PLATFUNC_VARIANT_NAME(name, variant)
41
+
PLATFUNC_VARIANT_NAME(platfunc_##name,variant)
44
42
45
43
#ifdef __ASSEMBLER__
46
44
···
54
52
*/
55
53
#define MP_SPIN_TRIES 1000
56
54
57
-
#ifndef DARLING
58
-
59
-
#define PLATFUNC_VARIANT_NAME(name, variant) _ ## name ## $VARIANT$ ## variant
60
-
61
-
#else
62
-
63
-
#define PLATFUNC_VARIANT_NAME(name, variant) name ##$VARIANT$ ## variant
64
-
65
-
#endif
55
+
#define PLATFUNC_VARIANT_NAME(name, variant) _##name##$VARIANT$##variant
66
56
67
57
#if defined (__i386__)
68
58
#define PLATFUNC_DESCRIPTOR_FIELD_POINTER .long
69
-
#define PLATFUNC_DESCRIPTOR_REFERENCE(name, variant) \
70
-
.long PLATFUNC_DESCRIPTOR_NAME(name, variant)
59
+
#define PLATFUNC_DESCRIPTOR_REFERENCE(name,variant) \
60
+
.long PLATFUNC_DESCRIPTOR_NAME(name,variant)
71
61
#elif defined (__x86_64__)
72
62
#define PLATFUNC_DESCRIPTOR_FIELD_POINTER .quad
73
63
#define PLATFUNC_DESCRIPTOR_REFERENCE(name, variant) \
74
-
.quad PLATFUNC_DESCRIPTOR_NAME(name, variant)
64
+
.quad PLATFUNC_DESCRIPTOR_NAME(name,variant)
75
65
#else
76
66
#error unsupported architecture
77
67
#endif
···
79
69
#ifdef VARIANT_DYLD
80
70
81
71
#define PLATFUNC_FUNCTION_START_GENERIC(name, variant, codetype, alignment) \
82
-
PLATFUNC_FUNCTION_START(name, variant, codetype, alignment) \
72
+
PLATFUNC_FUNCTION_START(name,variant, codetype, alignment) \
83
73
.globl _ ## name ;\
84
74
_ ## name ## :
85
75
86
76
#define PLATFUNC_DESCRIPTOR(name, variant, must, cant)
87
77
88
-
#elif defined(DARLING)
89
-
90
-
#undef PLATFUNC_DESCRIPTOR_NAME
91
-
#define PLATFUNC_DESCRIPTOR_NAME(name, variant) name##$VARIANT_DESC$##variant
92
-
93
-
#define PLATFUNC_FUNCTION_START_GENERIC(name, variant, codetype, alignment) \
94
-
PLATFUNC_FUNCTION_START(name, variant, codetype, alignment)
95
-
96
-
#define PLATFUNC_DESCRIPTOR(name, variant, must, cant) \
97
-
.globl PLATFUNC_DESCRIPTOR_NAME(name,variant) ;\
98
-
.hidden PLATFUNC_DESCRIPTOR_NAME(name,variant) ;\
99
-
PLATFUNC_DESCRIPTOR_NAME(name,variant) ## : ;\
100
-
PLATFUNC_DESCRIPTOR_FIELD_POINTER PLATFUNC_VARIANT_NAME(name,variant) ;\
101
-
.long must ;\
102
-
.long cant ;\
103
-
104
78
#else /* VARIANT_DYLD */
105
79
106
80
#define PLATFUNC_FUNCTION_START_GENERIC PLATFUNC_FUNCTION_START
107
81
108
82
#define PLATFUNC_DESCRIPTOR(name, variant, must, cant) \
109
83
.const_data ;\
110
-
.private_extern PLATFUNC_DESCRIPTOR_NAME(name, variant) ;\
111
-
PLATFUNC_DESCRIPTOR_NAME(name, variant) ## : ;\
112
-
PLATFUNC_DESCRIPTOR_FIELD_POINTER PLATFUNC_VARIANT_NAME(name, variant) ;\
84
+
.private_extern PLATFUNC_DESCRIPTOR_NAME(name,variant) ;\
85
+
PLATFUNC_DESCRIPTOR_NAME(name,variant) ## : ;\
86
+
PLATFUNC_DESCRIPTOR_FIELD_POINTER PLATFUNC_VARIANT_NAME(name,variant) ;\
113
87
.long must ;\
114
88
.long cant ;\
115
89
.text
116
90
117
91
#endif /* VARIANT_DYLD */
118
92
119
-
#ifdef DARLING
120
-
121
-
#define PLATFUNC_FUNCTION_START(name, variant, codetype, alignment) \
122
-
.text ;\
123
-
.align (2<<alignment), 0x90 ;\
124
-
.globl PLATFUNC_VARIANT_NAME(name,variant) ## ;\
125
-
PLATFUNC_VARIANT_NAME(name,variant) ##:
126
-
127
-
#else
128
-
129
93
#define PLATFUNC_FUNCTION_START(name, variant, codetype, alignment) \
130
94
.text ;\
131
95
.align alignment, 0x90 ;\
132
-
.private_extern PLATFUNC_VARIANT_NAME(name, variant) ;\
133
-
PLATFUNC_VARIANT_NAME(name, variant) ## :
134
-
135
-
#endif
96
+
.private_extern PLATFUNC_VARIANT_NAME(name,variant) ;\
97
+
PLATFUNC_VARIANT_NAME(name,variant) ## :
136
98
137
99
#else /* __ASSEMBLER__ */
138
100
139
-
#ifdef DARLING
140
-
#undef PLATFUNC_DESCRIPTOR_NAME
141
-
#define PLATFUNC_DESCRIPTOR_NAME(name, variant) name ## $VARIANT_DESC$ ## variant
142
-
#endif
143
-
144
-
#define PLATFUNC_VARIANT_NAME(name, variant) name ## $VARIANT$ ## variant
145
-
146
-
#define PLATFUNC_DESCRIPTOR_PROTOTYPE(name, variant) extern const platfunc_descriptor PLATFUNC_DESCRIPTOR_NAME(name, variant);
147
-
#define PLATFUNC_DESCRIPTOR_REFERENCE(name, variant) &PLATFUNC_DESCRIPTOR_NAME(name, variant)
101
+
#define PLATFUNC_VARIANT_NAME(name, variant) name##$VARIANT$##variant
102
+
#define PLATFUNC_DESCRIPTOR_PROTOTYPE(name, variant) extern const platfunc_descriptor PLATFUNC_DESCRIPTOR_NAME(name,variant);
103
+
#define PLATFUNC_DESCRIPTOR_REFERENCE(name, variant) &PLATFUNC_DESCRIPTOR_NAME(name,variant)
148
104
149
105
#define PLATFUNC_DESCRIPTOR(name, variant, must, cant) \
150
-
extern void PLATFUNC_VARIANT_NAME(name, variant) (void); \
151
-
const platfunc_descriptor PLATFUNC_DESCRIPTOR_NAME(name, variant) = { \
152
-
.address = PLATFUNC_VARIANT_NAME(name, variant), \
106
+
extern void PLATFUNC_VARIANT_NAME(name,variant) (void); \
107
+
const platfunc_descriptor PLATFUNC_DESCRIPTOR_NAME(name,variant) = { \
108
+
.address = PLATFUNC_VARIANT_NAME(name,variant), \
153
109
.musthave = must, \
154
110
.canthave = cant \
155
111
}
-1
src/libc/gen/stack_logging_disk.c
-1
src/libc/gen/stack_logging_disk.c
-1
src/libc/gen/wordexp.c
-1
src/libc/gen/wordexp.c
-8
src/libc/pthreads/pthread_tsd.c
-8
src/libc/pthreads/pthread_tsd.c
···
253
253
254
254
}
255
255
256
-
#ifdef DARLING
257
-
258
-
// Needed for Darling libdyld
259
-
260
-
int __darwin_pthread_key_create(pthread_key_t *key, void (*destructor)(void*)) __attribute__((weak, alias("pthread_key_create")));
261
-
int __darwin_pthread_key_delete(pthread_key_t key) __attribute__((weak, alias("pthread_key_delete")));
262
-
int __darwin_pthread_setspecific(pthread_key_t key, const void *value) __attribute__((weak, alias("pthread_setspecific")));
263
-
#endif
264
256
-1
src/libc/threads/cprocs.c
-1
src/libc/threads/cprocs.c
+20
-20
src/libc/x86_64/CMakeLists.txt
+20
-20
src/libc/x86_64/CMakeLists.txt
···
3
3
cmake_minimum_required(VERSION 2.8.0)
4
4
enable_language(C ASM)
5
5
6
-
add_definitions(-D_getcontext=getcontext)
7
-
add_definitions(-D_moncount=moncount)
8
-
add_definitions(-D__setcontext=_setcontext)
9
-
add_definitions(-D_getmcontext=getmcontext)
10
-
add_definitions(-D__ctx_done=_ctx_done)
11
-
add_definitions(-D__ctx_start=_ctx_start)
12
-
add_definitions(-D_cpu_number=cpu_number)
13
-
add_definitions(-D_sys_icache_invalidate=sys_icache_invalidate)
14
-
add_definitions(-D_sys_dcache_flush=sys_dcache_flush)
6
+
#add_definitions(-D_getcontext=getcontext)
7
+
#add_definitions(-D_moncount=moncount)
8
+
#add_definitions(-D__setcontext=_setcontext)
9
+
#add_definitions(-D_getmcontext=getmcontext)
10
+
#add_definitions(-D__ctx_done=_ctx_done)
11
+
#add_definitions(-D__ctx_start=_ctx_start)
12
+
#add_definitions(-D_cpu_number=cpu_number)
13
+
#add_definitions(-D_sys_icache_invalidate=sys_icache_invalidate)
14
+
#add_definitions(-D_sys_dcache_flush=sys_dcache_flush)
15
15
add_definitions(-DPRIVATE)
16
16
17
17
#set(CMAKE_C_FLAGS "-I${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_C_FLAGS} -include ${CMAKE_CURRENT_SOURCE_DIR}/../../duct/include/commpage.h -include ${DARLING_TOP_DIRECTORY}/platform-include/sys/time.h")
···
31
31
../i386/pthreads/init_cpu_capabilities.c
32
32
)
33
33
34
-
add_definitions(-D_preempt=preempt)
35
-
add_definitions(-D_backoff=backoff)
36
-
add_definitions(-D_pthread_getspecific=pthread_getspecific)
37
-
add_definitions(-D__commpage_pthread_mutex_lock=_commpage_pthread_mutex_lock)
38
-
add_definitions(-D_pthread_self=pthread_self)
39
-
add_definitions(-D___pthread_set_self=__pthread_set_self)
40
-
add_definitions(-D_start_wqthread=start_wqthread)
41
-
add_definitions(-D_thread_start=thread_start)
42
-
add_definitions(-D__pthread_start=_pthread_start)
43
-
add_definitions(-D__pthread_wqthread=_pthread_wqthread)
34
+
#add_definitions(-D_preempt=preempt)
35
+
#add_definitions(-D_backoff=backoff)
36
+
#add_definitions(-D_pthread_getspecific=pthread_getspecific)
37
+
#add_definitions(-D__commpage_pthread_mutex_lock=_commpage_pthread_mutex_lock)
38
+
#add_definitions(-D_pthread_self=pthread_self)
39
+
#add_definitions(-D___pthread_set_self=__pthread_set_self)
40
+
#add_definitions(-D_start_wqthread=start_wqthread)
41
+
#add_definitions(-D_thread_start=thread_start)
42
+
#add_definitions(-D__pthread_start=_pthread_start)
43
+
#add_definitions(-D__pthread_wqthread=_pthread_wqthread)
44
44
45
45
set(x86_64_sources ${x86_64_sources}
46
46
pthreads/preempt.S
···
85
85
sys/setjmp.S
86
86
sys/_sigtramp.S
87
87
sys/spinlocks_asm.S
88
-
#sys/spinlocks.c # aliases in other complation units don't work, work done in spinlocks_asm.S
88
+
sys/spinlocks.c # aliases in other complation units don't work, work done in spinlocks_asm.S
89
89
)
90
90
91
91
-5
src/libc/x86_64/gen/_ctx_start.S
-5
src/libc/x86_64/gen/_ctx_start.S
-4
src/libc/x86_64/gen/_setcontext.S
-4
src/libc/x86_64/gen/_setcontext.S
-4
src/libc/x86_64/gen/cpu_number.S
-4
src/libc/x86_64/gen/cpu_number.S
-4
src/libc/x86_64/gen/getcontext.S
-4
src/libc/x86_64/gen/getcontext.S
+1
-6
src/libc/x86_64/gen/icacheinval.S
+1
-6
src/libc/x86_64/gen/icacheinval.S
···
26
26
27
27
#include <machine/cpu_capabilities.h>
28
28
29
-
#ifdef DARLING
30
-
# define _sys_icache_invalidate sys_icache_invalidate
31
-
# define _sys_dcache_flush sys_dcache_flush
32
-
#endif
33
-
34
29
.text
35
-
.align (2<<4), 0x00
30
+
.align 4, 0x00
36
31
37
32
/* void sys_icache_invalidate(addr_t start, int length) */
38
33
+3
-5
src/libc/x86_64/gen/mcount.S
+3
-5
src/libc/x86_64/gen/mcount.S
···
20
20
*
21
21
* @APPLE_LICENSE_HEADER_END@
22
22
*/
23
-
#include <architecture/i386/asm_help.h>
23
+
#import <architecture/i386/asm_help.h>
24
24
25
25
.text
26
-
.globl mcount
27
-
mcount:
28
-
#ifndef DARLING /* DARLING-TODO, moncount is defunct/deprecated API */
26
+
.globl _mcount
27
+
_mcount:
29
28
pushq %rbp // setup mcount's frame
30
29
movq %rsp, %rbp
31
30
subq $64, %rsp // allocate space for storage and alignment
···
50
49
51
50
movq %rbp, %rsp
52
51
popq %rbp // tear down frame
53
-
#endif
54
52
ret
-4
src/libc/x86_64/pthreads/pthread_getspecific.S
-4
src/libc/x86_64/pthreads/pthread_getspecific.S
-4
src/libc/x86_64/pthreads/pthread_mutex_lock.S
-4
src/libc/x86_64/pthreads/pthread_mutex_lock.S
-4
src/libc/x86_64/pthreads/pthread_self.S
-4
src/libc/x86_64/pthreads/pthread_self.S
-4
src/libc/x86_64/pthreads/pthread_set_self.S
-4
src/libc/x86_64/pthreads/pthread_set_self.S
-5
src/libc/x86_64/pthreads/start_wqthread.S
-5
src/libc/x86_64/pthreads/start_wqthread.S
-5
src/libc/x86_64/pthreads/thread_start.S
-5
src/libc/x86_64/pthreads/thread_start.S
+3
-3
src/libc/x86_64/string/__bzero.S
+3
-3
src/libc/x86_64/string/__bzero.S
+3
-15
src/libc/x86_64/string/bcopy.c
+3
-15
src/libc/x86_64/string/bcopy.c
···
1
-
// Modified by Lubos Dolezel for Darling
2
1
/*
3
2
* Copyright (c) 2007, 2008, 2009, 2010 Apple Inc. All rights reserved.
4
3
*
···
23
22
*/
24
23
25
24
#include <machine/cpu_capabilities.h>
26
-
#include <sys/types.h>
27
25
#include "platfunc.h"
28
26
29
27
PLATFUNC_DESCRIPTOR_PROTOTYPE(bcopy, sse42)
···
35
33
0
36
34
};
37
35
38
-
void bcopy_chooser(const void *src, void *dest, size_t n)
39
-
#ifndef DARLING
40
-
__asm__("_bcopy");
41
-
#else
42
-
__asm__("bcopy");
43
-
#endif
44
-
void bcopy_chooser(const void *src, void *dest, size_t n) {
45
-
#ifndef DARLING
36
+
void *bcopy_chooser() __asm__("_bcopy");
37
+
void *bcopy_chooser() {
46
38
__asm__(".desc _bcopy, 0x100");
47
-
#endif
48
-
void (*impl)(const void *src, void *dest, size_t n);
49
-
impl = find_platform_function((const platfunc_descriptor **) bcopy_platfunc_descriptors);
50
-
51
-
impl(src, dest, n);
39
+
return find_platform_function((const platfunc_descriptor **) bcopy_platfunc_descriptors);
52
40
}
-5
src/libc/x86_64/string/bcopy_sse3x.S
-5
src/libc/x86_64/string/bcopy_sse3x.S
···
1
-
// Modified by Lubos Dolezel for Darling
2
1
/*
3
2
* Copyright (c) 2006 Apple Computer, Inc. All rights reserved.
4
3
*
···
36
35
*
37
36
* The following #defines are tightly coupled to the u-architecture:
38
37
*/
39
-
40
-
#ifdef DARLING
41
-
# define _longcopy longcopy@PLT
42
-
#endif
43
38
44
39
#define kShort 80 // too short to bother with SSE (must be >=80)
45
40
#define kVeryLong (500*1024) // large enough for non-temporal stores (>=8192 and <2GB)
+4
-9
src/libc/x86_64/string/bzero.c
+4
-9
src/libc/x86_64/string/bzero.c
···
1
-
// Modified by Lubos Dolezel for Darling
2
1
/*
3
2
* Copyright (c) 2007, 2008, 2009, 2010 Apple Inc. All rights reserved.
4
3
*
···
23
22
*/
24
23
25
24
#include <machine/cpu_capabilities.h>
26
-
#include <sys/types.h>
27
25
#include "platfunc.h"
28
26
29
27
PLATFUNC_DESCRIPTOR_PROTOTYPE(bzero, sse42)
···
35
33
0
36
34
};
37
35
38
-
void bzero_chooser(void *s, size_t n) __asm__("bzero");
39
-
void bzero_chooser(void *s, size_t n) {
40
-
//__asm__(".desc _bzero, 0x100");
41
-
void (*impl)(void*, size_t);
42
-
impl = find_platform_function((const platfunc_descriptor **) bzero_platfunc_descriptors);
43
-
44
-
impl(s, n);
36
+
void *bzero_chooser() __asm__("_bzero");
37
+
void *bzero_chooser() {
38
+
__asm__(".desc _bzero, 0x100");
39
+
return find_platform_function((const platfunc_descriptor **) bzero_platfunc_descriptors);
45
40
}
+4
-12
src/libc/x86_64/string/ffs.S
+4
-12
src/libc/x86_64/string/ffs.S
···
1
-
// Modified by Lubos Dolezel for Darling
2
1
/*
3
2
* Copyright (c) 2005-2006 Apple Computer, Inc. All rights reserved.
4
3
*
···
21
20
*
22
21
* @APPLE_LICENSE_HEADER_END@
23
22
*/
24
-
25
-
#ifdef DARLING
26
-
# define _ffs ffs
27
-
# define _ffsl ffsl
28
-
# define _fls fls
29
-
# define _flsl flsl
30
-
#endif
31
23
32
24
.text
33
25
34
-
.align (2<<2)
26
+
.align 2
35
27
.globl _ffs
36
28
_ffs:
37
29
movl $(-1), %edx
···
41
33
ret
42
34
43
35
44
-
.align (2<<2)
36
+
.align 2
45
37
.globl _ffsl
46
38
_ffsl:
47
39
movl $(-1), %edx
···
51
43
ret
52
44
53
45
54
-
.align (2<<2)
46
+
.align 2
55
47
.globl _fls
56
48
_fls:
57
49
movl $(-1), %edx
···
61
53
ret
62
54
63
55
64
-
.align (2<<2)
56
+
.align 2
65
57
.globl _flsl
66
58
_flsl:
67
59
movl $(-1), %edx
+2
-9
src/libc/x86_64/string/longcopy_sse3x.S
+2
-9
src/libc/x86_64/string/longcopy_sse3x.S
···
1
-
// Modified by Lubos Dolezel for Darling
2
1
/*
3
2
* Copyright (c) 2006 Apple Computer, Inc. All rights reserved.
4
3
*
···
56
55
57
56
// void longcopy(const void *dest, void *sou, size_t len)
58
57
59
-
#ifdef DARLING
60
-
# define _longcopy longcopy
61
-
# define _memcpy memcpy@PLT
62
-
.globl longcopy
63
-
#endif
64
-
65
-
// .private_extern _longcopy
58
+
.private_extern _longcopy
66
59
_longcopy:
67
60
pushq %rbp // set up a frame for backtraces
68
61
movq %rsp,%rbp
···
103
96
// ecx = scratch reg used to read a byte of each cache line
104
97
// eax = chunk offset
105
98
106
-
.align (2<<4),0x90 // 16-byte align inner loops
99
+
.align 4,0x90 // 16-byte align inner loops
107
100
LTouchLoop:
108
101
movzb (%rsi,%rax),%ecx // touch line 0, 2, 4, or 6 of page
109
102
movzb 1*64(%rsi,%rax),%ecx // touch line 1, 3, 5, or 7
+1
-9
src/libc/x86_64/string/memcmp.S
+1
-9
src/libc/x86_64/string/memcmp.S
···
1
-
// Modified by Lubos Dolezel for Darling
2
1
/*
3
2
* Copyright (c) 2005 Apple Computer, Inc. All rights reserved.
4
3
*
···
41
40
42
41
#define kShort 18 // too short for vectors (must be >16)
43
42
44
-
#ifdef DARLING
45
-
# define _memcmp memcmp
46
-
# define _bcmp bcmp
47
-
.type memcmp, @function
48
-
.type bcmp, @function
49
-
#endif
50
-
51
43
.text
52
-
.align (2<<4)
44
+
.align 4
53
45
54
46
.globl _memcmp
55
47
.globl _bcmp
+4
-9
src/libc/x86_64/string/memcpy.c
+4
-9
src/libc/x86_64/string/memcpy.c
···
1
-
// Modified by Lubos Dolezel for Darling
2
1
/*
3
2
* Copyright (c) 2007, 2008, 2009, 2010 Apple Inc. All rights reserved.
4
3
*
···
23
22
*/
24
23
25
24
#include <machine/cpu_capabilities.h>
26
-
#include <sys/types.h>
27
25
#include "platfunc.h"
28
26
29
27
PLATFUNC_DESCRIPTOR_PROTOTYPE(memcpy, sse42)
···
35
33
0
36
34
};
37
35
38
-
void *memcpy_chooser(void *dest, const void *src, size_t n) __asm__("memcpy");
39
-
void *memcpy_chooser(void *dest, const void *src, size_t n) {
40
-
// __asm__(".desc _memcpy, 0x100");
41
-
void *(*impl)(void *, const void *, size_t);
42
-
impl = find_platform_function((const platfunc_descriptor **) memcpy_platfunc_descriptors);
43
-
44
-
return impl(dest, src, n);
36
+
void *memcpy_chooser() __asm__("_memcpy");
37
+
void *memcpy_chooser() {
38
+
__asm__(".desc _memcpy, 0x100");
39
+
return find_platform_function((const platfunc_descriptor **) memcpy_platfunc_descriptors);
45
40
}
+4
-9
src/libc/x86_64/string/memmove.c
+4
-9
src/libc/x86_64/string/memmove.c
···
1
-
// Modified by Lubos Dolezel for Darling
2
1
/*
3
2
* Copyright (c) 2007, 2008, 2009, 2010 Apple Inc. All rights reserved.
4
3
*
···
23
22
*/
24
23
25
24
#include <machine/cpu_capabilities.h>
26
-
#include <sys/types.h>
27
25
#include "platfunc.h"
28
26
29
27
PLATFUNC_DESCRIPTOR_PROTOTYPE(memmove, sse42)
···
35
33
0
36
34
};
37
35
38
-
void *memmove_chooser(void *dest, const void *src, size_t n) __asm__("memmove");
39
-
void *memmove_chooser(void *dest, const void *src, size_t n) {
40
-
//__asm__(".desc _memmove, 0x100");
41
-
void *(*impl)(void *, const void *, size_t);
42
-
impl = find_platform_function((const platfunc_descriptor **) memmove_platfunc_descriptors);
43
-
44
-
return impl(dest, src, n);
36
+
void *memmove_chooser() __asm__("_memmove");
37
+
void *memmove_chooser() {
38
+
__asm__(".desc _memmove, 0x100");
39
+
return find_platform_function((const platfunc_descriptor **) memmove_platfunc_descriptors);
45
40
}
-10
src/libc/x86_64/string/memset.S
-10
src/libc/x86_64/string/memset.S
···
1
-
// Modified by Lubos Dolezel for Darling
2
1
/*
3
2
* Copyright (c) 2005 Apple Computer, Inc. All rights reserved.
4
3
*
···
42
41
* NB: we avoid "stos" family of instructions (stosl, stosb), as they are very slow
43
42
* on P4s and probably other processors.
44
43
*/
45
-
46
-
#ifdef DARLING
47
-
# define _memset memset
48
-
# define _memset_pattern4 memset_pattern4
49
-
# define _memset_pattern8 memset_pattern8
50
-
# define _memset_pattern16 memset_pattern16
51
-
# define _bzero bzero@PLT
52
-
.type memset, @function
53
-
#endif
54
44
55
45
#define kShort 255 // for nonzero memset(), too short for commpage
56
46
+1
-7
src/libc/x86_64/string/strcmp.S
+1
-7
src/libc/x86_64/string/strcmp.S
···
1
-
// Modified by Lubos Dolezel for Darling
2
1
/*
3
2
* Copyright (c) 2005 Apple Computer, Inc. All rights reserved.
4
3
*
···
35
34
// reading bytes past the difference. To avoid this, we never do a load
36
35
// that crosses a page boundary.
37
36
38
-
#ifdef DARLING
39
-
# define _strcmp strcmp
40
-
.type strcmp, @function
41
-
#endif
42
-
43
37
.text
44
38
.globl _strcmp
45
39
46
-
.align (2<<4)
40
+
.align 4
47
41
_strcmp: // int strcmp(const char *s1,const char *s2);
48
42
49
43
// In order to avoid spurious page faults, we loop over:
+1
-7
src/libc/x86_64/string/strcpy.S
+1
-7
src/libc/x86_64/string/strcpy.S
···
1
-
// Modified by Lubos Dolezel for Darling
2
1
/*
3
2
* Copyright (c) 2005 Apple Computer, Inc. All rights reserved.
4
3
*
···
37
36
//
38
37
// We align the destination, because unaligned vector stores are slow.
39
38
40
-
#ifdef DARLING
41
-
# define _strcpy strcpy
42
-
.type strcpy, @function
43
-
#endif
44
-
45
39
.text
46
40
.globl _strcpy
47
41
48
-
.align (2<<4)
42
+
.align 4
49
43
_strcpy: // char *strcpy(const char *dst, const char *src);
50
44
movq %rdi,%rcx // preserve dest ptr so we can return it
51
45
movl %edi,%edx // copy low 4 bytes of dest ptr
+1
-7
src/libc/x86_64/string/strlcat.S
+1
-7
src/libc/x86_64/string/strlcat.S
···
1
-
// Modified by Lubos Dolezel for Darling
2
1
/*
3
2
* Copyright (c) 2007 Apple Inc. All rights reserved.
4
3
*
···
47
46
// On Core2 class machines, this algorithm seems to be faster than the
48
47
// naive byte-by-byte version for operands longer than about 11 bytes.
49
48
50
-
#ifdef DARLING
51
-
# define _strlcat strlcat
52
-
.type strlcat, @function
53
-
#endif
54
-
55
49
.text
56
50
.globl _strlcat
57
51
···
64
58
// %rsi = source ptr
65
59
// %rdx = size
66
60
67
-
.align (2<<4)
61
+
.align 4
68
62
_strlcat: // size_t *strlcat(char *dst, const char *src, size_t size);
69
63
movl %edi,%ecx // copy buffer ptr
70
64
movq %rdi,%r10 // save copies of buffer ptr and length
+1
-7
src/libc/x86_64/string/strlcpy.S
+1
-7
src/libc/x86_64/string/strlcpy.S
···
1
-
// Modified by Lubos Dolezel for Darling
2
1
/*
3
2
* Copyright (c) 2007 Apple Inc. All rights reserved.
4
3
*
···
44
43
// with one exception: 0x01 bytes preceeding the first zero are also
45
44
// mapped to 0x80.
46
45
47
-
#ifdef DARLING
48
-
# define _strlcpy strlcpy
49
-
.type strlcpy,@function
50
-
#endif
51
-
52
46
.text
53
47
.globl _strlcpy
54
48
···
57
51
// %rsi = source ptr
58
52
// %rdx = length
59
53
60
-
.align (2<<4)
54
+
.align 4
61
55
_strlcpy: // size_t *strlcpy(char *dst, const char *src, size_t size);
62
56
movl %esi,%ecx // copy source ptr
63
57
movq %rdi,%r10 // copy dest ptr
+1
-7
src/libc/x86_64/string/strlen.S
+1
-7
src/libc/x86_64/string/strlen.S
···
1
-
// Modified by Lubos Dolezel for Darling
2
1
/*
3
2
* Copyright (c) 2005-2007 Apple Inc. All rights reserved.
4
3
*
···
32
31
* We favor the fall-through (ie, short operand) path.
33
32
*/
34
33
35
-
#ifdef DARLING
36
-
# define _strlen strlen
37
-
.type strlen, @function
38
-
#endif
39
-
40
34
.text
41
35
.globl _strlen
42
-
.align (2<<4), 0x90
36
+
.align 4, 0x90
43
37
_strlen: // size_t strlen(char *b);
44
38
pxor %xmm0,%xmm0 // zero %xmm0
45
39
movl %edi,%ecx // copy low half of ptr
+1
-7
src/libc/x86_64/string/strncmp.S
+1
-7
src/libc/x86_64/string/strncmp.S
···
1
-
// Modified by Lubos Dolezel for Darling
2
1
/*
3
2
* Copyright (c) 2005-2006 Apple Computer, Inc. All rights reserved.
4
3
*
···
35
34
// reading bytes past the difference. To avoid this, we never do a load
36
35
// that crosses a page boundary.
37
36
38
-
#ifdef DARLING
39
-
# define _strncmp strncmp
40
-
.type strncmp, @function
41
-
#endif
42
-
43
37
#define kShort 20 // too short for vectors (must be >16)
44
38
45
39
.text
46
40
.globl _strncmp
47
41
48
-
.align (2<<4)
42
+
.align 4
49
43
_strncmp: // int strncmp(const char *s1, const char *s2, size_t len);
50
44
cmpq $(kShort),%rdx // worth accelerating?
51
45
ja LNotShort // yes
-7
src/libc/x86_64/string/strncpy.S
-7
src/libc/x86_64/string/strncpy.S
···
1
-
// Modified by Lubos Dolezel for Darling
2
1
/*
3
2
* Copyright (c) 2005 Apple Computer, Inc. All rights reserved.
4
3
*
···
42
41
// Recall that strncpy() zero fills the remainder of the dest buffer,
43
42
// and does not terminate the string if its length is greater than or
44
43
// equal to n.
45
-
46
-
#ifdef DARLING
47
-
# define _strncpy strncpy
48
-
# define _bzero bzero@PLT
49
-
.type strncpy, @function
50
-
#endif
51
44
52
45
#define kShort 31 // too short to bother with vector loop
53
46
+2
-2
src/libc/x86_64/sys/OSAtomic.S
+2
-2
src/libc/x86_64/sys/OSAtomic.S
···
28
28
#include <architecture/i386/asm_help.h>
29
29
30
30
#define DECLARE(x) \
31
-
.align 1<<2 ; \
31
+
.align 2 ; \
32
32
.globl x ; \
33
33
.globl x ## Barrier ; \
34
34
x: ; \
···
41
41
#define ATOMIC_RET_ORIG 0
42
42
#define ATOMIC_RET_NEW 1
43
43
44
-
#ifndef DARLING
44
+
#ifndef DARLING__DISABLED
45
45
// compare and exchange 32-bit
46
46
// xchg32 <new> <dst> <mp>
47
47
.macro xchg32
-5
src/libc/x86_64/sys/_setjmp.S
-5
src/libc/x86_64/sys/_setjmp.S
+2
-18
src/libc/x86_64/sys/_sigtramp.S
+2
-18
src/libc/x86_64/sys/_sigtramp.S
···
42
42
#define MCONTEXT_SS_R8 80
43
43
#define MCONTEXT_SS_RIP 144
44
44
45
-
#ifdef DARLING
46
-
# define __sigtramp _sigtramp
47
-
# define ___sigreturn __sigreturn@PLT
48
-
# define ___in_sigtramp __in_sigtramp@GOTPCREL
49
-
#endif
50
-
51
45
/* register use:
52
46
%rbx uctx
53
47
···
60
54
ucontext_t *uctx %r8
61
55
)
62
56
*/
63
-
64
-
#ifdef DARLING
65
-
# define __sigtramp _sigtramp
66
-
#endif
67
57
68
58
.globl __sigtramp
69
59
.text
70
-
.align 1<<4
60
+
.align 4
71
61
__sigtramp:
72
62
Lstart:
73
63
/* Although this routine does not need any stack frame, various parts
···
162
152
DW_OP_breg(3), UCONTEXT_UC_MCONTEXT, DW_OP_deref, \
163
153
DW_OP_plus_uconst, MCONTEXT_SS_R8+(8*(regno-8)), 1
164
154
165
-
#ifndef DARLING
166
155
/* Unwind tables. */
167
156
.section __TEXT,__eh_frame,coalesced,no_toc+strip_static_syms+live_support
168
-
#else
169
-
.text
170
-
#endif
171
157
172
158
EH_frame1:
173
159
.set L$set$0,LECIE1-LSCIE1
···
264
250
.byte DW_OP_plus_uconst, MCONTEXT_SS_RSP, DW_OP_deref
265
251
Lcfa_end:
266
252
267
-
.align 1<<3
253
+
.align 3
268
254
LEFDE1:
269
255
270
-
#ifndef DARLING
271
256
.subsections_via_symbols
272
-
#endif
273
257
+1
-1
src/libc/x86_64/sys/atomic.c
+1
-1
src/libc/x86_64/sys/atomic.c
+1
-9
src/libc/x86_64/sys/i386_gettimeofday_asm.S
+1
-9
src/libc/x86_64/sys/i386_gettimeofday_asm.S
···
31
31
#include <machine/cpu_capabilities.h>
32
32
#include "platfunc.h"
33
33
34
-
#ifdef DARLING
35
-
# define _mach_absolute_time mach_absolute_time@PLT
36
-
#endif
37
-
38
34
#define NSEC_PER_SEC 1000*1000*1000
39
35
#define NSEC_PER_USEC 1000
40
36
41
-
#ifndef DARLING
42
37
.private_extern ___commpage_gettimeofday
43
-
#else
44
-
# define ___commpage_gettimeofday __commpage_gettimeofday
45
-
#endif
46
-
.align 1<<4
38
+
.align 4
47
39
___commpage_gettimeofday:
48
40
// %rdi = ptr to timeval
49
41
pushq %rbp // set up a frame for backtraces
-4
src/libc/x86_64/sys/nanotime.S
-4
src/libc/x86_64/sys/nanotime.S
+1
-1
src/libc/x86_64/sys/setjmp.S
+1
-1
src/libc/x86_64/sys/setjmp.S
+1
-1
src/libc/x86_64/sys/spinlocks.c
+1
-1
src/libc/x86_64/sys/spinlocks.c
-15
src/libc/x86_64/sys/spinlocks_asm.S
-15
src/libc/x86_64/sys/spinlocks_asm.S
···
43
43
PLATFUNC_DESCRIPTOR(OSSpinLockTry,up,kUP,0)
44
44
PLATFUNC_DESCRIPTOR(_spin_lock_try,up,kUP,0)
45
45
46
-
#ifdef DARLING
47
-
.globl OSSpinLockTry
48
-
.globl _spin_lock_try
49
-
OSSpinLockTry:
50
-
_spin_lock_try:
51
-
#endif
52
46
PLATFUNC_FUNCTION_START_GENERIC(OSSpinLockTry, mp, 64, 4)
53
47
PLATFUNC_FUNCTION_START_GENERIC(_spin_lock_try, mp, 64, 4)
54
48
xorl %eax, %eax
···
90
84
PLATFUNC_DESCRIPTOR(spin_lock,up,kUP,0)
91
85
92
86
93
-
#ifdef DARLING
94
-
.globl OSSpinLockLock
95
-
.globl _spin_lock
96
-
.globl spin_lock
97
-
98
-
OSSpinLockLock:
99
-
_spin_lock:
100
-
spin_lock:
101
-
#endif
102
87
PLATFUNC_FUNCTION_START_GENERIC(OSSpinLockLock, mp, 64, 4)
103
88
PLATFUNC_FUNCTION_START_GENERIC(_spin_lock, mp, 64, 4)
104
89
PLATFUNC_FUNCTION_START_GENERIC(spin_lock, mp, 64, 4)
+7
-6
src/libm/CMakeLists.txt
+7
-6
src/libm/CMakeLists.txt
···
5
5
6
6
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -msse -msse2 -msse3 -w -ggdb -O0")
7
7
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -D__DARWIN_UNIX03 -fPIC -w -DBUILDING_FOR_CARBONCORE_LEGACY -funroll-loops -msse3") # -DBUILDING_FOR_CARBONCORE_LEGACY")
8
-
set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -nostdlib -Wl,--version-script=${DARLING_TOP_DIRECTORY}/darwin.map")
8
+
set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -nostdlib -Wl,-flat_namespace -Wl,-undefined,suppress -Wl,-alias_list,${CMAKE_CURRENT_SOURCE_DIR}/Exports/libm_Intel.a.alias")
9
+
set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -Wl,-alias_list,${CMAKE_CURRENT_SOURCE_DIR}/Exports/libmathCommonIntel.alias")
9
10
10
11
include_directories(${CMAKE_CURRENT_SOURCE_DIR}/Source/Intel)
11
12
include_directories(${CMAKE_CURRENT_SOURCE_DIR}/../kernel-include)
12
-
add_definitions(-Dliteral8=text -Dliteral4=text -Dliteral16=text -Dmovsxw=movswl -DPRIVATE)
13
+
add_definitions(-Dmovsxw=movswl -DPRIVATE)
13
14
14
15
set(libm_sources
15
16
Source/abs.c
···
262
263
#add_library(system_m_extra_no_fenv OBJECT Source/Intel/xmm_log.c)
263
264
#set_target_properties(system_m_extra_no_fenv PROPERTIES COMPILE_FLAGS "-UBUILDING_FOR_CARBONCORE_LEGACY -ftrapping-math -fPIC -DNO_FENV")
264
265
265
-
add_library(system_m_no_fenv STATIC ${libm_nofenv_sources}) # $<TARGET_OBJECTS:system_m_extra_no_fenv>)
266
-
set_target_properties(system_m_no_fenv PROPERTIES COMPILE_FLAGS "-UBUILDING_FOR_CARBONCORE_LEGACY -ftrapping-math -fPIC -DNO_FENV")
266
+
#add_library(system_m_no_fenv OBJECT ${libm_nofenv_sources}) # $<TARGET_OBJECTS:system_m_extra_no_fenv>)
267
+
#set_target_properties(system_m_no_fenv PROPERTIES COMPILE_FLAGS "-UBUILDING_FOR_CARBONCORE_LEGACY -ftrapping-math -fPIC -DNO_FENV")
267
268
268
-
add_library(system_m SHARED ${libm_sources} $<TARGET_OBJECTS:system_m_extra>)
269
-
target_link_libraries(system_m PRIVATE "-Wl,--whole-archive" system_m_no_fenv)
269
+
add_darling_library(system_m SHARED ${libm_sources} $<TARGET_OBJECTS:system_m_extra>)
270
+
#target_link_libraries(system_m)
270
271
271
272
install(TARGETS system_m DESTINATION ${CMAKE_INSTALL_LIBDIR}/darling)
272
273
+5
-6
src/libm/Source/Intel/acosf.S
+5
-6
src/libm/Source/Intel/acosf.S
···
28
28
S1: .double -3.333333463718495862077843807
29
29
30
30
31
-
////.const
32
-
.align (1<<4)
31
+
.const
32
+
.align 4
33
33
34
34
/* Define some coefficients for center polynomial (used for x in [-.62,
35
35
+.62]). These are stored in pairs at aligned addresses for use in SIMD
···
158
158
showing that a valid result will be obtained regardless of which
159
159
value rsqrtss provides.
160
160
*/
161
-
#define _acosf acosf
162
-
.align (1<<5)
161
+
.align 5
163
162
#if !defined DevelopmentInstrumentation
164
163
// This is the regular name used in the deployed implementation.
165
-
.globl acosf
166
-
acosf:
164
+
.globl _acosf
165
+
_acosf:
167
166
#else
168
167
// This is the name used for a special test version of the routine.
169
168
.globl _acosfInstrumented
+4
-4
src/libm/Source/Intel/acoshf.S
+4
-4
src/libm/Source/Intel/acoshf.S
···
11
11
#include <machine/asm.h>
12
12
#include "abi.h"
13
13
14
-
////.const
14
+
.const
15
15
16
-
.align (1<<4)
16
+
.align 4
17
17
.quad 0x3e607933f5c9e7de, 0xbfb348d7f4d2c2c1 // c0, d
18
18
.quad 0xbec93b87a2f58777, 0x0000000000000000 // c1, 0
19
19
.quad 0x3ffb9a832f4eb8b5, 0x3ffec80a0deff17f // a0, b0
···
286
286
287
287
288
288
.literal8
289
-
.align (1<<3)
289
+
.align 3
290
290
one: .quad 0x3ff0000000000000
291
291
onehalf: .quad 0x3fe0000000000000
292
292
onethird: .quad 0x3fd5555555555555
···
297
297
298
298
299
299
.text
300
-
.align (1<<4)
300
+
.align 4
301
301
#if defined( __x86_64__ )
302
302
#define RELATIVE_ADDR( _a ) (_a)( %rip )
303
303
#else
+5
-5
src/libm/Source/Intel/asinf.S
+5
-5
src/libm/Source/Intel/asinf.S
···
27
27
S1: .double -3.333333463718495862077843807
28
28
29
29
30
-
////.const
31
-
.align (1<<4)
30
+
.const
31
+
.align 4
32
32
33
33
/* Define some coefficients for center polynomial (used for x in [-.57,
34
34
+.57]). These are stored in pairs at aligned addresses for use in SIMD
···
162
162
showing that a valid result will be obtained regardless of which
163
163
value rsqrtss provides.
164
164
*/
165
-
.align (1<<5)
165
+
.align 5
166
166
#if !defined DevelopmentInstrumentation
167
167
// This is the regular name used in the deployed implementation.
168
-
.globl asinf
169
-
asinf:
168
+
.globl _asinf
169
+
_asinf:
170
170
#else
171
171
// This is the name used for a special test version of the routine.
172
172
.globl _asinfInstrumented
+6
-6
src/libm/Source/Intel/asinhf.S
+6
-6
src/libm/Source/Intel/asinhf.S
···
52
52
#include <machine/asm.h>
53
53
#include "abi.h"
54
54
55
-
////.const
55
+
.const
56
56
57
57
// Coefficients for 7th order polynomial approximation on [0, 1/4]
58
58
// The polynomail is computed in packed factored form as follows:
···
61
61
// lo double: (cx * (x + alo)) * (x(x + b1lo) + b0lo)
62
62
//
63
63
// The high and low parts are then unpacked and multiplied.
64
-
.align (1<<4)
64
+
.align 4
65
65
asinhf_low: .quad 0x4008183efcaf7119, 0x4007eba0a6c21cf1 // b0hi, b0lo
66
66
.quad 0x3ffe9547f4507ace, 0xbffd2c173c2ad586 // ahi, alo
67
67
.quad 0x40021aa6afb79159, 0xc0015e2ed556dde5 // b1hi, b1lo
···
69
69
70
70
// Coefficients for rational approximation on [1/4, 4]
71
71
// p(x) and q(x) are computed side-by-side in packed arithmetic, then unpacked and divided.
72
-
.align (1<<4)
72
+
.align 4
73
73
asinhf_mid: .quad 0x3e4328ccef61bd30, 0x3f80f6f9cf323b3c // p[0], q[0]
74
74
.quad 0x3f80f6e561f06785, 0x3f85b29a3e277523 // p[1], q[1]
75
75
.quad 0x3f85b32a2f11b40a, 0x3f8e26416e925090 // p[2], q[2]
···
79
79
.quad 0x3f26c17c7b263d18, 0x3f001e11059bddca // p[6], q[6]
80
80
81
81
82
-
.align (1<<4)
82
+
.align 4
83
83
// Polynomial coefficients for the correction to log(2x) for the "large" case.
84
84
// The polynomial is computed in factored form as follows:
85
85
//
···
348
348
.quad 0x3fe61e3efda46467, 0x3fe0080402010080 // log(1.99609), 1/1.99609
349
349
350
350
.literal8
351
-
.align (1<<3)
351
+
.align 3
352
352
one: .quad 0x3ff0000000000000
353
353
onehalf: .quad 0x3fe0000000000000
354
354
onethird: .quad 0x3fd5555555555555
···
360
360
#else
361
361
#define RELATIVE_ADDR( _a ) (_a)-asinhf_body( %ecx )
362
362
363
-
.align (1<<4)
363
+
.align 4
364
364
asinhf_pic:
365
365
movl (%esp), %ecx // Copy address of this instruction to %ecx
366
366
ret
+6
-6
src/libm/Source/Intel/atan2f.S
+6
-6
src/libm/Source/Intel/atan2f.S
···
29
29
C2: .double 0.0029352921857004596570518
30
30
31
31
32
-
//.const
33
-
.align (1<<4)
32
+
.const
33
+
.align 4
34
34
35
35
/* Define some coefficients for center polynomial (used for x in [-1, +1]).
36
36
These are stored in pairs at aligned addresses for use in SIMD
···
42
42
C10: .double 5.4728447324456990092824269, 6.7197076223592378022736307
43
43
44
44
// This needs to be 16-byte aligned because it is used in an orpd instruction.
45
-
.align (1<<4)
45
+
.align 4
46
46
pPi: .double +3.141592653589793238462643 // pi.
47
47
48
48
···
209
209
(x1, y1) are in the same quadrant, then y0/x0 <= y1/x1 implies
210
210
atan2f(y0, x0) <= atan2f(y1, x1).)
211
211
*/
212
-
.align (1<<5)
213
-
.globl atan2f
214
-
atan2f:
212
+
.align 5
213
+
.globl _atan2f
214
+
_atan2f:
215
215
216
216
cvtss2sd Argy, y // Convert to double precision.
217
217
cvtss2sd Argx, x
+5
-5
src/libm/Source/Intel/atanf.S
+5
-5
src/libm/Source/Intel/atanf.S
···
21
21
C2: .double 0.0029352921857004596570518
22
22
23
23
24
-
//.const
25
-
.align (1<<4)
24
+
.const
25
+
.align 4
26
26
27
27
/* Define some coefficients for center polynomial (used for x in [-1, +1]).
28
28
These are stored in pairs at aligned addresses for use in SIMD
···
155
155
Exhaustive testing proved this routine returns faithfully rounded
156
156
results.
157
157
*/
158
-
.align (1<<5)
158
+
.align 5
159
159
#if !defined DevelopmentInstrumentation
160
160
// This is the regular name used in the deployed implementation.
161
-
.globl atanf
162
-
atanf:
161
+
.globl _atanf
162
+
_atanf:
163
163
#else
164
164
// This is the name used for a special test version of the routine.
165
165
.globl _atanfInstrumented
+5
-5
src/libm/Source/Intel/atanhf.S
+5
-5
src/libm/Source/Intel/atanhf.S
···
12
12
#include <machine/asm.h>
13
13
#include "abi.h"
14
14
15
-
//.const
16
-
.align (1<<4)
15
+
.const
16
+
.align 4
17
17
// Polynomial coefficients used for the 7/8 < |x| < 1 case. These are addressed by offset from big_table,
18
18
// so take care if you move things around.
19
19
···
554
554
// a = 7/8
555
555
556
556
.literal8
557
-
.align (1<<3)
557
+
.align 3
558
558
one_256th: .quad 0x3f70000000000000 // 1/256
559
559
one: .quad 0x3ff0000000000000
560
560
one_plus_eps: .quad 0x3ff0000000000001 // 1 + ulp
···
567
567
neglog2_2: .quad 0xbfd62e42fefa39ef // -ln(2)/2
568
568
569
569
.literal4
570
-
.align (1<<2)
570
+
.align 2
571
571
f256: .long 0x43800000 // 256.0f
572
572
573
573
.text
···
577
577
#elif defined( __i386__ )
578
578
#define RELATIVE_ADDR( _a) (_a)-atanhf_body( CX_P )
579
579
#define INDEX %edi
580
-
.align (1<<4)
580
+
.align 4
581
581
atanhf_pic:
582
582
movl (%esp), %ecx // copy address of local_addr to %ecx
583
583
ret
+3
-3
src/libm/Source/Intel/cbrtf.S
+3
-3
src/libm/Source/Intel/cbrtf.S
···
52
52
#include <machine/asm.h>
53
53
#include "abi.h"
54
54
55
-
//.const
56
-
.align (1<<4)
55
+
.const
56
+
.align 4
57
57
58
58
// Minimax polynomial coefficents (addressed by offset from cbrt_table)
59
59
.quad 0x4000204182c17486, 0xbff8b5b876f0d973 // b1lo b1hi
···
261
261
denormal_bias: .quad 0x3ff0000000000000, 0x3ab0000000000000
262
262
263
263
.text
264
-
.align (1<<4)
264
+
.align 4
265
265
266
266
#if defined( __x86_64__ )
267
267
#define RELATIVE_ADDR( _a ) (_a)( %rip )
+11
-13
src/libm/Source/Intel/ceil.S
+11
-13
src/libm/Source/Intel/ceil.S
···
7
7
* Copyright 2009, Apple Inc.
8
8
*/
9
9
10
-
.type ceil, @function
11
10
#ifdef __i386__
12
11
13
12
#ifdef __SSE3__
14
-
15
13
16
14
.text
17
-
.align (1<<4)
18
-
.globl ceil
19
-
ceil:
15
+
.align 4
16
+
.globl _ceil
17
+
_ceil:
20
18
mov 8(%esp), %ecx
21
19
fldl 4(%esp)
22
20
cmp $0x43300000, %ecx // if x is negative or x > 0x1.0p53
···
48
46
#else // i386, no SSE3
49
47
50
48
.text
51
-
.align (1<<4)
52
-
.globl ceil
53
-
ceil:
49
+
.align 4
50
+
.globl _ceil
51
+
_ceil:
54
52
mov 8(%esp), %ecx
55
53
fldl 4(%esp)
56
54
cmp $0x43300000, %ecx // if x is negative or x > 0x1.0p53
···
93
91
94
92
#else // x86_64
95
93
96
-
//.const
97
-
.align (1<<4)
94
+
.const
95
+
.align 4
98
96
one: .quad 0x3ff0000000000000
99
97
absmask:.quad 0x7fffffffffffffff
100
98
thresh: .quad 0x4330000000000000
101
99
102
100
.text
103
-
.align (1<<4)
104
-
.globl ceil
105
-
ceil:
101
+
.align 4
102
+
.globl _ceil
103
+
_ceil:
106
104
movd %xmm0, %rcx
107
105
andq absmask(%rip), %rcx // |x|
108
106
cmpq thresh(%rip), %rcx // if |x| >= 0x1.0p52 or isnan(x)
+12
-13
src/libm/Source/Intel/ceilf.S
+12
-13
src/libm/Source/Intel/ceilf.S
···
3
3
*
4
4
* Steve Canon, March 2009.
5
5
*/
6
-
7
-
.type ceilf, @function
6
+
8
7
#if defined __i386__
9
8
10
9
#include <System/i386/cpu_capabilities.h>
11
10
.set cpubits, _COMM_PAGE_CPU_CAPABILITIES
12
11
13
12
.text
14
-
.align (1<<4)
15
-
.globl ceilf
16
-
ceilf:
13
+
.align 4
14
+
.globl _ceilf
15
+
_ceilf:
17
16
movss 4(%esp), %xmm0 // load argument
18
17
testl $(kHasSSE4_1), cpubits
19
18
jz 0f
···
24
23
flds 4(%esp)
25
24
ret
26
25
27
-
.align (1<<4)
26
+
.align 4
28
27
0: // no SSE 4.1
29
28
mov 4(%esp), %eax
30
29
mov $23, %cl
···
48
47
cvttps2dq %xmm0, %xmm0 // set inexact
49
48
1: flds 4(%esp)
50
49
ret
51
-
.align (1<<4)
50
+
.align 4
52
51
2: cvttps2dq %xmm0, %xmm0 // set inexact
53
52
cmp $1, %edx // if x > 0, goto 3
54
53
jge 3f
55
54
andl $0x80000000, 4(%esp) // copysign(0.0, x)
56
55
flds 4(%esp)
57
56
ret
58
-
.align (1<<4)
57
+
.align 4
59
58
3: movl $0x3f800000, 4(%esp) // return 1.0
60
59
flds 4(%esp)
61
60
ret
62
61
63
62
#elif defined __x86_64__
64
63
65
-
//.const
66
-
.align (1<<4)
64
+
.const
65
+
.align 4
67
66
one: .long 0x3f800000
68
67
absmask:.long 0x7fffffff
69
68
70
69
.text
71
-
.align (1<<4)
72
-
.globl ceilf
73
-
ceilf:
70
+
.align 4
71
+
.globl _ceilf
72
+
_ceilf:
74
73
movd %xmm0, %eax
75
74
andl absmask(%rip), %eax
76
75
movd absmask(%rip), %xmm1
+4
-6
src/libm/Source/Intel/cosh.S
+4
-6
src/libm/Source/Intel/cosh.S
···
12
12
#include <machine/asm.h>
13
13
#include "abi.h"
14
14
15
-
.type cosh, @function
16
-
17
-
//.const
18
-
.align (1<<4)
15
+
.const
16
+
.align 4
19
17
// Polynomial coefficients, offset from exp2table
20
18
.quad 0x3f55e52272e0eaec, 0x3f55e52272e0eaec // c4
21
19
.quad 0x401cc9eea1e24220, 0x401cc9eea1e24220 // c3/c4
···
163
161
.double 0.20918130318061568990950557203211923657510680e-8
164
162
165
163
.literal8
166
-
.align (1<<3)
164
+
.align 3
167
165
one_n7: .quad 0x3f80000000000000
168
166
lge_p7: .quad 0x40671547652b82fe
169
167
lge_hi: .quad 0x3ff7154760000000
···
181
179
#else
182
180
#define RELATIVE_ADDR(_a) (_a)-cosh_body(%ecx)
183
181
184
-
.align (1<<4)
182
+
.align 4
185
183
cosh_pic:
186
184
movl (%esp), %ecx
187
185
ret
+3
-5
src/libm/Source/Intel/coshf.S
+3
-5
src/libm/Source/Intel/coshf.S
···
11
11
#include <machine/asm.h>
12
12
#include "abi.h"
13
13
14
-
.type coshf, @function
15
-
16
14
// This is identical to sinhf with some indices permuted to compute cosh instead.
17
15
// Read the detailed comments in that file for more information.
18
16
19
-
//.const
20
-
.align (1<<4)
17
+
.const
18
+
.align 4
21
19
.quad 0x3f811111111110fe, 0x3fa555f78359bc34 // c5, c4 = 0.0083333..., 0.0416715
22
20
.quad 0x4034000000000018, 0x4027ff4991a5ebc5 // c3/c5 = 0.1666666.../c5, c2/c4 = 0.5/c4
23
21
coshf_table: .quad 0x0000000000000000, 0x3ff0000000000000 // sinh(0/16), cosh(0/16)
···
220
218
#define RELATIVE_ADDR(_a) (_a)-coshf_body( %ecx )
221
219
#define RELATIVE_ADDR_B(_a) (_a)-coshf_body_eleven( %ecx )
222
220
223
-
.align (1<<4)
221
+
.align 4
224
222
coshf_pic:
225
223
movl (%esp), %ecx // Copy address of this instruction to %ecx
226
224
ret
+7
-7
src/libm/Source/Intel/e_cbrtl.S
+7
-7
src/libm/Source/Intel/e_cbrtl.S
···
8
8
*/
9
9
10
10
#define ENTRY(name) \
11
-
.globl name; \
12
-
.align (1<<4); \
13
-
name##:
11
+
.globl _##name; \
12
+
.align 4; \
13
+
_##name##:
14
14
15
15
#if defined( __LP64__ )
16
16
#define LOCAL_STACK_SIZE 20
···
20
20
21
21
#include "abi.h"
22
22
23
-
//.const
24
-
.align (1<<4)
23
+
.const
24
+
.align 4
25
25
onethird: .long 0xaaaaaaab, 0xaaaaaaaa, 0x00003ffd, 0x00000000 //(long double) 1.0L/3.0L
26
26
27
-
.align (1<<3)
27
+
.align 3
28
28
correction: .double 0.62996052494743658238361, 0.79370052598409973737585, 1.0, 1.2599210498948731647672, 1.5874010519681994747517
29
29
coeffs: .double 1.7830491344381518, -1.5730724799776633, 1.2536000054780357, -0.60460822457398278, 0.15834924310704463, -0.017322841453552703
30
30
31
-
.align (1<<2)
31
+
.align 2
32
32
infinity: .single +Infinity
33
33
34
34
// Stack:
+3
-3
src/libm/Source/Intel/exp.S
+3
-3
src/libm/Source/Intel/exp.S
···
20
20
Tables and Constants
21
21
22
22
****************************************************************************/
23
-
//.const
24
-
.align (1<<4)
23
+
.const
24
+
.align 4
25
25
26
26
27
27
// Coefficients for the 4th order minimax polynomial
···
175
175
.quad 0x3f8111116e99ac77
176
176
177
177
.literal8
178
-
.align (1<<3)
178
+
.align 3
179
179
lge_p7: .quad 0x40671547652b82fe
180
180
lge: .quad 0x3ff71547652b82fe
181
181
lge_hi: .quad 0x3ff7154760000000
+3
-3
src/libm/Source/Intel/exp2.S
+3
-3
src/libm/Source/Intel/exp2.S
···
17
17
Tables and Constants
18
18
19
19
****************************************************************************/
20
-
//.const
21
-
.align (1<<4)
20
+
.const
21
+
.align 4
22
22
23
23
24
24
// Coefficients for the 4th order minimax polynomial
···
176
176
177
177
178
178
.literal8
179
-
.align (1<<3)
179
+
.align 3
180
180
one_p7: .quad 0x4060000000000000
181
181
tiny_val: .quad 0x0010000000000001
182
182
huge_val: .quad 0x7fefffffffffffff
+3
-3
src/libm/Source/Intel/exp2f.S
+3
-3
src/libm/Source/Intel/exp2f.S
···
55
55
//
56
56
//
57
57
58
-
.align (1<<4)
58
+
.align 4
59
59
// 8th order minimax fit of exp2 on [-1.0,1.0]. |error| < 0.402865722354948566583852e-9:
60
60
exp2f_c: .quad 0x40bc03f30399c376, 0x3ff000000001ea2a // c4/c8 = 0.961813690023115610862381719985771e-2 / 0.134107709538786543922336536865157e-5, c0 = 1.0 + 0.278626872016317130037181614004e-10
61
61
.quad 0x408f10e7f73e6d8f, 0x3fe62e42fd0933ee // c5/c8 = 0.133318252930790403741964203236548e-2 / 0.134107709538786543922336536865157e-5, c1 = .693147176943623740308984004029708
···
64
64
.quad 0x3eb67fe1dc3105ba // c8 = 0.134107709538786543922336536865157e-5
65
65
66
66
67
-
.align (1<<3)
67
+
.align 3
68
68
exp2f_nofenv_c: .quad 0x3ff0000000058fca // 1.0 + -8.09329727503262700660348520172e-11
69
69
.double 0.693147206709644041218074094717934
70
70
.double 0.240226515050550309232521176082490
···
85
85
#define RELATIVE_ADDR_B( _a) (_a)-exp2f_no_fenv_body( CX_P )
86
86
87
87
//a short routine to get the local address
88
-
.align (1<<4)
88
+
.align 4
89
89
exp2f_pic: movl (%esp), %ecx //copy address of local_addr to %ecx
90
90
ret
91
91
#else
+3
-3
src/libm/Source/Intel/expf.S
+3
-3
src/libm/Source/Intel/expf.S
···
59
59
//
60
60
//
61
61
62
-
.align (1<<4)
62
+
.align 4
63
63
// 8th order minimax fit of exp2 on [-1.0,1.0]. |error| < 0.402865722354948566583852e-9:
64
64
expf_c: .quad 0x40bc03f30399c376, 0x3ff000000001ea2a // c4/c8 = 0.961813690023115610862381719985771e-2 / 0.134107709538786543922336536865157e-5, c0 = 1.0 + 0.278626872016317130037181614004e-10
65
65
.quad 0x408f10e7f73e6d8f, 0x3fe62e42fd0933ee // c5/c8 = 0.133318252930790403741964203236548e-2 / 0.134107709538786543922336536865157e-5, c1 = .693147176943623740308984004029708
···
68
68
.quad 0x3eb67fe1dc3105ba // c8 = 0.134107709538786543922336536865157e-5
69
69
70
70
71
-
.align (1<<3)
71
+
.align 3
72
72
expf_nofenv_c: .quad 0x3ff0000000058fca // 1.0 + -8.09329727503262700660348520172e-11
73
73
.double 0.693147206709644041218074094717934
74
74
.double 0.240226515050550309232521176082490
···
90
90
#define RELATIVE_ADDR_B( _a) (_a)-expf_no_fenv_body( CX_P )
91
91
92
92
//a short routine to get the local address
93
-
.align (1<<4)
93
+
.align 4
94
94
expf_pic: movl (%esp), %ecx //copy address of local_addr to %ecx
95
95
ret
96
96
#else
+2
-2
src/libm/Source/Intel/expl.S
+2
-2
src/libm/Source/Intel/expl.S
+3
-3
src/libm/Source/Intel/expm1.S
+3
-3
src/libm/Source/Intel/expm1.S
···
20
20
Tables and Constants
21
21
22
22
****************************************************************************/
23
-
//.const
24
-
.align (1<<4)
23
+
.const
24
+
.align 4
25
25
26
26
// Coefficients for the 4th order minimax polynomial
27
27
// approximation of exp2( ) on the interval [-0x1p-27, 0x1p-7]
···
176
176
// Other useful constants
177
177
178
178
.literal8
179
-
.align (1<<3)
179
+
.align 3
180
180
lge_p7: .quad 0x40671547652b82fe
181
181
lge: .quad 0x3ff71547652b82fe
182
182
lge_hi: .quad 0x3ff7154760000000
+3
-3
src/libm/Source/Intel/expm1f.S
+3
-3
src/libm/Source/Intel/expm1f.S
···
11
11
#include <machine/asm.h>
12
12
#include "abi.h"
13
13
14
-
//.const
14
+
.const
15
15
16
16
//minimax polynomial for exp2(x)-1
17
-
.align (1<<4)
17
+
.align 4
18
18
// 8th order minimax fit of exp2 on [-1.0,1.0]. |error| < 0.402865722354948566583852e-9:
19
19
expm1f_c: .quad 0x40bc03f30399c376, 0x3dbea2a63403aaa8 // c4/c8 = 0.961813690023115610862381719985771e-2 / 0.134107709538786543922336536865157e-5, c0 = 0.278626872016317130037181614004e-10
20
20
.quad 0x408f10e7f73e6d8f, 0x3fe62e42fd0933ee // c5/c8 = 0.133318252930790403741964203236548e-2 / 0.134107709538786543922336536865157e-5, c1 = .693147176943623740308984004029708
···
172
172
#define RELATIVE_ADDR_B( _a) (_a)-expm1f_no_fenv_body( CX_P )
173
173
174
174
//a short routine to get the local address
175
-
.align (1<<4)
175
+
.align 4
176
176
expm1f_pic: movl (%esp), %ecx //copy address of local_addr to %ecx
177
177
ret
178
178
#else
+11
-11
src/libm/Source/Intel/floor.S
+11
-11
src/libm/Source/Intel/floor.S
···
12
12
#ifdef __SSE3__
13
13
14
14
.text
15
-
.align (1<<4)
16
-
.globl floor
17
-
floor:
15
+
.align 4
16
+
.globl _floor
17
+
_floor:
18
18
mov 8(%esp), %ecx
19
19
fldl 4(%esp)
20
20
cmp $0x43300000, %ecx // if x is negative or x > 0x1.0p53
···
48
48
#else // i386, no SSE3
49
49
50
50
.text
51
-
.align (1<<4)
52
-
.globl floor
53
-
floor:
51
+
.align 4
52
+
.globl _floor
53
+
_floor:
54
54
mov 8(%esp), %ecx
55
55
and $0x7fffffff, %ecx // high word of |x|
56
56
fldl 4(%esp)
···
78
78
79
79
#else // x86_64
80
80
81
-
//.const
82
-
.align (1<<4)
81
+
.const
82
+
.align 4
83
83
mone: .quad 0xbff0000000000000
84
84
absmask:.quad 0x7fffffffffffffff
85
85
thresh: .quad 0x432fffffffffffff
86
86
87
87
.text
88
-
.align (1<<4)
89
-
.globl floor
90
-
floor:
88
+
.align 4
89
+
.globl _floor
90
+
_floor:
91
91
movd %xmm0, %rcx
92
92
andq absmask(%rip), %rcx // |x|
93
93
subq $1, %rcx
+11
-11
src/libm/Source/Intel/floorf.S
+11
-11
src/libm/Source/Intel/floorf.S
···
10
10
.set cpubits, _COMM_PAGE_CPU_CAPABILITIES
11
11
12
12
.text
13
-
.align (1<<4)
14
-
.globl floorf
15
-
floorf:
13
+
.align 4
14
+
.globl _floorf
15
+
_floorf:
16
16
movss 4(%esp), %xmm0 // load argument
17
17
testl $(kHasSSE4_1), cpubits
18
18
jz 0f
···
23
23
flds 4(%esp)
24
24
ret
25
25
26
-
.align (1<<4)
26
+
.align 4
27
27
0: // no SSE 4.1
28
28
mov 4(%esp), %eax
29
29
mov $23, %cl
···
48
48
cvttps2dq %xmm0, %xmm0 // set inexact
49
49
1: flds 4(%esp)
50
50
ret
51
-
.align (1<<4)
51
+
.align 4
52
52
2: cvttps2dq %xmm0, %xmm0 // set inexact
53
53
cmp $0xffffffff, %edx // if x < 0.0, goto 3
54
54
jl 3f
55
55
andl $0x80000000, 4(%esp) // copysign(0.0, x)
56
56
flds 4(%esp)
57
57
ret
58
-
.align (1<<4)
58
+
.align 4
59
59
3: movl $0xbf800000, 4(%esp) // return 1.0
60
60
flds 4(%esp)
61
61
ret
62
62
63
63
#elif defined __x86_64__
64
64
65
-
//.const
66
-
.align (1<<4)
65
+
.const
66
+
.align 4
67
67
mone: .long 0xbf800000
68
68
absmask:.long 0x7fffffff
69
69
70
70
.text
71
-
.align (1<<4)
72
-
.globl floorf
73
-
floorf:
71
+
.align 4
72
+
.globl _floorf
73
+
_floorf:
74
74
movd %xmm0, %ecx
75
75
andl $0x7fffffff, %ecx // |x|
76
76
subl $1, %ecx // subtract 1. This forces |+-0| to -0
+9
-9
src/libm/Source/Intel/fmod.S
+9
-9
src/libm/Source/Intel/fmod.S
···
6
6
*/
7
7
8
8
.text
9
-
.align (1<<4)
10
-
.globl fmodf
11
-
fmodf:
9
+
.align 4
10
+
.globl _fmodf
11
+
_fmodf:
12
12
#if defined __x86_64__
13
13
#define xf -4(%rsp)
14
14
#define yf -8(%rsp)
···
34
34
ret
35
35
36
36
.text
37
-
.align (1<<4)
38
-
.globl fmod
39
-
fmod:
37
+
.align 4
38
+
.globl _fmod
39
+
_fmod:
40
40
#if defined __x86_64__
41
41
#define x -8(%rsp)
42
42
#define y -16(%rsp)
···
62
62
ret
63
63
64
64
.text
65
-
.align (1<<4)
66
-
.globl fmodl
67
-
fmodl:
65
+
.align 4
66
+
.globl _fmodl
67
+
_fmodl:
68
68
#if defined __x86_64__
69
69
#define xl 8(%rsp)
70
70
#define yl 24(%rsp)
+11
-11
src/libm/Source/Intel/hypot.S
+11
-11
src/libm/Source/Intel/hypot.S
···
10
10
#endif
11
11
12
12
.text
13
-
.globl hypot
14
-
.globl cabs
13
+
.globl _hypot
14
+
.globl _cabs
15
15
16
16
#if defined __i386__
17
17
···
20
20
21
21
// Entry point --------------------------------------------------------
22
22
23
-
.align (1<<4)
24
-
cabs:
25
-
hypot:
23
+
.align 4
24
+
_cabs:
25
+
_hypot:
26
26
mov 8(%esp), %eax
27
27
mov 16(%esp), %edx
28
28
mov ABSHI, %ecx
···
55
55
// Special case handling ----------------------------------------------
56
56
57
57
L_xHiIsZero:
58
-
cmpl $0, 4(%esp) // is the low word of x zero?
58
+
cmp $0, 4(%esp) // is the low word of x zero?
59
59
jnz L_returnFromXHiIsZero // if not, jump back to mainline
60
60
L_returnAbsY:
61
61
and %ecx, 16(%esp)
···
63
63
ret
64
64
65
65
L_yHiIsZero:
66
-
cmpl $0, 12(%esp) // is the low word of y zero?
66
+
cmp $0, 12(%esp) // is the low word of y zero?
67
67
jnz L_returnFromYHiIsZero // if not, jump back to mainline
68
68
L_returnAbsX:
69
69
and %ecx, 8(%esp)
···
73
73
L_xIsSpecial:
74
74
cmp INFHI, %edx // check if y is infinity
75
75
jnz L_returnAbsX
76
-
cmpl $0, 12(%esp)
76
+
cmp $0, 12(%esp)
77
77
jz L_returnAbsY
78
78
jmp L_returnAbsX
79
79
···
84
84
85
85
// Entry point --------------------------------------------------------
86
86
87
-
.align (1<<4)
88
-
cabs:
89
-
hypot:
87
+
.align 4
88
+
_cabs:
89
+
_hypot:
90
90
movd %xmm0, %rax
91
91
movd %xmm1, %rdx
92
92
mov ABSMASK, %rcx
+9
-9
src/libm/Source/Intel/hypotf.S
+9
-9
src/libm/Source/Intel/hypotf.S
···
13
13
#define INFINITY $0x7f800000
14
14
15
15
.text
16
-
.globl cabsf
17
-
.globl hypotf
16
+
.globl _cabsf
17
+
.globl _hypotf
18
18
19
19
// Entry points -------------------------------------------------------
20
20
21
21
#if defined __i386__
22
-
.align (1<<4)
23
-
cabsf: // on i386, we can use the same code for
24
-
hypotf: // hypotf and cabsf, because the arguments
22
+
.align 4
23
+
_cabsf: // on i386, we can use the same code for
24
+
_hypotf: // hypotf and cabsf, because the arguments
25
25
mov 4(%esp), %eax // come in at the same stack offsets
26
26
mov 8(%esp), %edx //
27
27
movss 4(%esp), %xmm0 // real at esp + 4
28
28
movss 8(%esp), %xmm1 // imag at esp + 8
29
29
#else
30
-
.align (1<<4) // however, on x86_64, the registers used
31
-
cabsf: // are different. cabsf's arguments come
30
+
.align 4 // however, on x86_64, the registers used
31
+
_cabsf: // are different. cabsf's arguments come
32
32
pshufd $0xfd, %xmm0, %xmm1 // in packed in xmm0.
33
-
.align (1<<4)
34
-
hypotf: //
33
+
.align 4
34
+
_hypotf: //
35
35
movd %xmm0, %eax // hypotf, on the other hand, gets x in
36
36
movd %xmm1, %edx // xmm0 and y in xmm1.
37
37
#endif
+6
-6
src/libm/Source/Intel/hypotl.S
+6
-6
src/libm/Source/Intel/hypotl.S
···
26
26
// Entry point ---------------------------------------------------------------
27
27
28
28
.text
29
-
.globl hypotl
30
-
.globl cabsl
31
-
.align (1<<4)
32
-
hypotl:
33
-
cabsl:
29
+
.globl _hypotl
30
+
.globl _cabsl
31
+
.align 4
32
+
_hypotl:
33
+
_cabsl:
34
34
35
35
// Check argument scaling ----------------------------------------------------
36
36
//
···
262
262
// Useful constants ----------------------------------------------------------
263
263
264
264
.literal16
265
-
.align (1<<4)
265
+
.align 4
266
266
L_infinity:
267
267
.quad 0x8000000000000000, 0x0000000000007fff
268
268
L_tiny:
+6
-6
src/libm/Source/Intel/log.S
+6
-6
src/libm/Source/Intel/log.S
···
18
18
#include <machine/asm.h>
19
19
#include "abi.h"
20
20
21
-
//.const
22
-
.align (1<<5)
21
+
.const
22
+
.align 5
23
23
xone: .quad 0x3ff0000000000000, 0
24
24
frexp_exp_mask: .quad 0x7ff0000000000000, 0
25
25
frexp_mant_mask: .quad 0x800fffffffffffff, 0
···
28
28
//log10_key_mask: .quad 0x07e0000000000000, 0
29
29
log1p_not_ulp_mask: .quad 0xfffffffffffffffe, 0
30
30
31
-
.align (1<<5)
31
+
.align 5
32
32
lgel: .quad 0xB8AA3B295C17F0BC, 0x3fff
33
33
ln2l: .quad 0xB17217F7D1CF79AC, 0x3ffe //ln(2) rounded up to long double
34
34
···
37
37
38
38
c0: .quad 0xFFFFFFFFFFFFFFD7, 0xbffd //c0 = -.4999999999999999988974167423L
39
39
40
-
.align (1<<5)
40
+
.align 5
41
41
a01: .double .827742667285236703751556405085096, -2.00038644890076831031534988283768 //a0,a1
42
42
b01: .double 1.51843353412997067893915870795354, 1.54454569915832086827096843200102 //b0,b1
43
43
44
44
// The lookup table is in a funny format since it has 2 long double and a single.
45
45
// {10-byte va ; 2-byte pad ; 4-byte single a ; 10-byte lg1pa ; 6-byte pad}
46
-
.align (1<<5)
46
+
.align 5
47
47
LOOKUP:
48
48
// This is the table for a, ap1, va, lg1pa: a = (float)k*scale, ap1 = a + 1, va = (long double)1./(1.+a), lg1pa = (long double)log2(1.+a)
49
49
// In C this would be
···
375
375
.quad 0x3ff0000000000000, 0x0000000000000000 //{0x1p+0, 0x0p+0}, k=63
376
376
377
377
.literal8
378
-
.align (1<<3)
378
+
.align 3
379
379
one: .double 1.0
380
380
mone: .double -1.0
381
381
+4
-4
src/libm/Source/Intel/log10f.S
+4
-4
src/libm/Source/Intel/log10f.S
···
9
9
#include <machine/asm.h>
10
10
#include "abi.h"
11
11
12
-
//.const
13
-
.align (1<<4)
12
+
.const
13
+
.align 4
14
14
15
15
// 256 entry of Lookup table of values used for log10 calculation, generated as:
16
16
//
···
300
300
301
301
302
302
.literal8
303
-
.align (1<<3)
303
+
.align 3
304
304
one: .double 1.0
305
305
onehalf: .double 0.5
306
306
onethird: .quad 0x3fd5555555555555 // 1/3
···
321
321
#define RELATIVE_ADDR2( _a, _i, _step) (_a)-rel_addr( CX_P, _i, _step )
322
322
323
323
//a short routine to get the local address
324
-
.align (1<<4)
324
+
.align 4
325
325
log10f_pic: movl (%esp), %ecx //copy address of local_addr to %ecx
326
326
ret
327
327
#else
+4
-4
src/libm/Source/Intel/log2f.S
+4
-4
src/libm/Source/Intel/log2f.S
···
9
9
#include <machine/asm.h>
10
10
#include "abi.h"
11
11
12
-
//.const
13
-
.align (1<<4)
12
+
.const
13
+
.align 4
14
14
15
15
// 256 entry of Lookup table of values used for log2 calculation, generated as:
16
16
//
···
300
300
.quad 0x3fefe8e4f15bd1a0, 0x3fe0080402010080 //log2(1.99609), 1/1.99609
301
301
302
302
.literal8
303
-
.align (1<<3)
303
+
.align 3
304
304
one: .double 1.0
305
305
onehalf: .double 0.5
306
306
onethird: .quad 0x3fd5555555555555 // 1/3
···
321
321
#define RELATIVE_ADDR2( _a, _i, _step) (_a)-rel_addr( CX_P, _i, _step )
322
322
323
323
//a short routine to get the local address
324
-
.align (1<<4)
324
+
.align 4
325
325
log2f_pic: movl (%esp), %ecx //copy address of local_addr to %ecx
326
326
ret
327
327
#else
+5
-5
src/libm/Source/Intel/logf.S
+5
-5
src/libm/Source/Intel/logf.S
···
15
15
#include <machine/asm.h>
16
16
#include "abi.h"
17
17
18
-
//.const
19
-
.align (1<<4)
18
+
.const
19
+
.align 4
20
20
21
21
// 511 entry of Lookup table of values used for log and log1p calculation, generated as:
22
22
//
···
436
436
437
437
438
438
.literal8
439
-
.align (1<<3)
439
+
.align 3
440
440
one: .double 1.0
441
441
onehalf: .double 0.5
442
442
onethird: .quad 0x3fd5555555555555 // 1/3
···
448
448
log2: .quad 0x3fe62e42fefa39efULL // ln(2)
449
449
450
450
.literal4
451
-
.align (1<<2)
451
+
.align 2
452
452
f256: .long 0x43800000 //256.0f
453
453
r256: .long 0x3b800000 //1.0f/256.0f
454
454
···
465
465
#define RELATIVE_ADDR2( _a, _i, _step) (_a)-rel_addr( CX_P, _i, _step )
466
466
467
467
//a short routine to get the local address
468
-
.align (1<<4)
468
+
.align 4
469
469
logf_pic: movl (%esp), %ecx //copy address of local_addr to %ecx
470
470
ret
471
471
#else
+3
-3
src/libm/Source/Intel/lroundl.S
+3
-3
src/libm/Source/Intel/lroundl.S
···
11
11
#include "abi.h"
12
12
#include <machine/asm.h>
13
13
14
-
.align (1<<2)
14
+
.align 2
15
15
.literal4
16
16
two63: .long 0x5f000000
17
17
mtwo63: .long 0xdf000000
18
18
one: .long 1
19
19
inf: .long 0x7f800000
20
20
21
-
.align (1<<3)
21
+
.align 3
22
22
.literal8
23
23
cutoff32: .double 2147483647.5 // 2**31-0.5
24
24
mcutoff32: .double -2147483648.5 // 2**31-0.5
25
25
26
26
27
-
.align (1<<4)
27
+
.align 4
28
28
.literal16
29
29
cutoff: .quad 0xffffffffffffffff, 0x403d
30
30
sign: .quad 0x0, 0xffffffffffffffff
+6
-7
src/libm/Source/Intel/machine/asm.h
+6
-7
src/libm/Source/Intel/machine/asm.h
···
22
22
*/
23
23
24
24
#define ENTRY(name) \
25
-
.type name, @function; \
26
-
.globl name; \
27
-
.align (2<<4); \
28
-
name##:
25
+
.globl _##name; \
26
+
.align 4; \
27
+
_##name##:
29
28
30
29
#ifndef NO_FENV
31
30
#define PRIVATE_ENTRY(name) \
32
-
.globl name; .hidden name; \
33
-
.align (2<<4); \
34
-
name##:
31
+
.globl _##name; .private_extern _##name; \
32
+
.align 4; \
33
+
_##name##:
35
34
#else
36
35
#define PRIVATE_ENTRY(name)
37
36
#endif
+8
-8
src/libm/Source/Intel/modf.S
+8
-8
src/libm/Source/Intel/modf.S
···
25
25
#if defined __i386__
26
26
27
27
.text
28
-
.align (1<<4)
29
-
.globl modf
30
-
modf:
28
+
.align 4
29
+
.globl _modf
30
+
_modf:
31
31
movl 8(%esp), %eax // high word of x
32
32
andl $0x7fffffff, %eax // high word of |x|
33
33
subl $0x3ff00000, %eax // subtract off exponent bias
···
78
78
79
79
#else // __x86_64__
80
80
81
-
//.const
82
-
.align (1<<4)
81
+
.const
82
+
.align 4
83
83
absmask:
84
84
.quad 0x7fffffffffffffff
85
85
one:
86
86
.quad 0x3ff0000000000000
87
87
88
88
.text
89
-
.align (1<<4)
90
-
.globl modf
91
-
modf:
89
+
.align 4
90
+
.globl _modf
91
+
_modf:
92
92
movsd absmask(%rip), %xmm1
93
93
andpd %xmm0, %xmm1 // |x|
94
94
movd %xmm1, %rax
+1
-1
src/libm/Source/Intel/modff.S
+1
-1
src/libm/Source/Intel/modff.S
+4
-4
src/libm/Source/Intel/nearbyint.S
+4
-4
src/libm/Source/Intel/nearbyint.S
···
5
5
//
6
6
// -- Stephen Canon, January 2010
7
7
8
-
#include <i386/cpu_capabilities.h>
8
+
#include <System/i386/cpu_capabilities.h>
9
9
10
10
.text
11
-
.align (1<<4)
12
-
.globl nearbyint
13
-
nearbyint:
11
+
.align 4
12
+
.globl _nearbyint
13
+
_nearbyint:
14
14
15
15
// Fast path if SSE4.1 is available ------------------------------------------
16
16
+4
-4
src/libm/Source/Intel/nearbyintf.S
+4
-4
src/libm/Source/Intel/nearbyintf.S
···
5
5
//
6
6
// -- Stephen Canon, January 2010
7
7
8
-
#include <i386/cpu_capabilities.h>
8
+
#include <System/i386/cpu_capabilities.h>
9
9
10
10
.text
11
-
.align (1<<4)
12
-
.globl nearbyintf
13
-
nearbyintf:
11
+
.align 4
12
+
.globl _nearbyintf
13
+
_nearbyintf:
14
14
15
15
// Fast path if SSE4.1 is available ------------------------------------------
16
16
+3
-3
src/libm/Source/Intel/nextafterl.S
+3
-3
src/libm/Source/Intel/nextafterl.S
···
11
11
#include "abi.h"
12
12
13
13
#if defined( __x86_64__ )
14
-
.align (1<<3)
14
+
.align 3
15
15
.literal8
16
16
signBit: .quad 0x8000000000000000
17
17
nonSignBit: .quad 0x7fffffffffffffff
18
18
#endif
19
19
20
-
.align (1<<4)
21
-
//.const
20
+
.align 4
21
+
.const
22
22
23
23
tiny: .quad 0x8000000000000000, 0x1 //0x1.0p-16382
24
24
big: .quad 0x8000000000000000, 0x7ffe //0x1.0p16383
+5
-5
src/libm/Source/Intel/powf.S
+5
-5
src/libm/Source/Intel/powf.S
···
16
16
#include "machine/asm.h"
17
17
#include "abi.h"
18
18
19
-
//.const
19
+
.const
20
20
gMaskShift: .byte 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, \
21
21
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, \
22
22
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, \
···
34
34
31,31,31,31, 31,31,31,31,31,31,31,31,31,31,31,31, \
35
35
31,31,31,31, 31,31,31,31,31,31,31,31,31,31,31,31
36
36
37
-
.align (1<<4)
37
+
.align 4
38
38
// 8th order minimax fit of exp2 on [-1.0,1.0]. |error| < 0.402865722354948566583852e-9:
39
39
powf_exp_c: .quad 0x40bc03f30399c376, 0x3ff000000001ea2a // c4/c8 = 0.961813690023115610862381719985771e-2 / 0.134107709538786543922336536865157e-5, c0 = 1.0 + 0.278626872016317130037181614004e-10
40
40
.quad 0x408f10e7f73e6d8f, 0x3fe62e42fd0933ee // c5/c8 = 0.133318252930790403741964203236548e-2 / 0.134107709538786543922336536865157e-5, c1 = .693147176943623740308984004029708
···
98
98
// exact powers of two. (These are 1.0 + 1 ulp and 1.0 - 1ulp.) The other values will land somewhere in the
99
99
// range [ 1.0 - 2**-8, 1.0 + 2**-7 ]. (Experimentally verified for all floats 1.0 <= x < 2.0.)
100
100
//
101
-
.align (1<<3)
101
+
.align 3
102
102
powf_log2_c: .quad 0x3ff0000000000000, 0x8000000000000000 // 1, -log2l(1)
103
103
.quad 0x3fefc00000000000, 0x3f872c7ba20f7327 // 0.992188, -log2l(0.992188)
104
104
.quad 0x3fef80fe03f80fe0, 0x3f9715662c7f3dbc // 0.984496, -log2l(0.984496)
···
659
659
jmp 9f
660
660
661
661
662
-
.align (1<<4)
662
+
.align 4
663
663
// return value in %xmm0
664
664
9:
665
665
pop DI_P
···
697
697
movapd %xmm0, %xmm2 // r = x
698
698
jmp 3f
699
699
700
-
.align (1<<4)
700
+
.align 4
701
701
// do{
702
702
2: mulsd %xmm0, %xmm0 // x *= x
703
703
test $1, %edx
+5
-5
src/libm/Source/Intel/rint.S
+5
-5
src/libm/Source/Intel/rint.S
···
6
6
//
7
7
// -- Stephen Canon, January 2010
8
8
9
-
.globl rint
9
+
.globl _rint
10
10
11
11
#if defined __i386__
12
12
13
13
.text
14
-
.align (1<<4)
15
-
rint:
14
+
.align 4
15
+
_rint:
16
16
fldl 4(%esp)
17
17
frndint
18
18
ret
···
24
24
twop52: .quad 0x4330000000000000
25
25
26
26
.text
27
-
.align (1<<4)
28
-
rint:
27
+
.align 4
28
+
_rint:
29
29
movd %xmm0, %rcx
30
30
mov %rcx, %rdx
31
31
and absmask(%rip), %rcx // |x|
+3
-3
src/libm/Source/Intel/rintf.S
+3
-3
src/libm/Source/Intel/rintf.S
+3
-3
src/libm/Source/Intel/rintl.S
+3
-3
src/libm/Source/Intel/rintl.S
+8
-8
src/libm/Source/Intel/round.S
+8
-8
src/libm/Source/Intel/round.S
···
25
25
#endif
26
26
27
27
.text
28
-
.align (1<<4)
29
-
.globl round
30
-
round:
28
+
.align 4
29
+
.globl _round
30
+
_round:
31
31
movl 8(%esp), %ecx
32
32
cmpl $0x43300000, %ecx
33
33
fldl 4(%esp)
···
67
67
68
68
#else //x86_64
69
69
70
-
//.const
71
-
.align (1<<4)
70
+
.const
71
+
.align 4
72
72
one: .quad 0x3ff0000000000000
73
73
absmask:.quad 0x7fffffffffffffff
74
74
half: .quad 0x3fe0000000000000
75
75
thresh: .quad 0x4330000000000000
76
76
77
77
.text
78
-
.align (1<<4)
79
-
.globl round
80
-
round:
78
+
.align 4
79
+
.globl _round
80
+
_round:
81
81
movd %xmm0, %rcx
82
82
andq absmask(%rip), %rcx
83
83
movsd absmask(%rip), %xmm2
+14
-14
src/libm/Source/Intel/roundf.S
+14
-14
src/libm/Source/Intel/roundf.S
···
7
7
#if defined __i386__
8
8
9
9
.text
10
-
.align (1<<4)
11
-
.globl roundf
12
-
roundf:
10
+
.align 4
11
+
.globl _roundf
12
+
_roundf:
13
13
mov 4(%esp), %eax // load the input, x
14
14
movss 4(%esp), %xmm0
15
15
and $0x7f800000, %eax // |x|
···
31
31
cvttps2dq %xmm0, %xmm0 // raise inexact
32
32
1: flds 4(%esp)
33
33
ret
34
-
.align (1<<4)
34
+
.align 4
35
35
2: je 3f // if |x| >= 0.5f, goto 3
36
36
andl $0x80000000, 4(%esp) // copysign(0.0, x)
37
37
cvttps2dq %xmm0, %xmm0 // raise inexact
38
38
flds 4(%esp)
39
39
ret
40
-
.align (1<<4)
40
+
.align 4
41
41
3: addl $0x00800000, 4(%esp)
42
42
andl $0xff800000, 4(%esp) // copysign(1.0, x)
43
43
cvttps2dq %xmm0, %xmm0 // raise inexact
···
46
46
47
47
#elif defined __x86_64__
48
48
49
-
//.const
50
-
.align (1<<4)
49
+
.const
50
+
.align 4
51
51
mzero: .long 0x80000000
52
-
.align (1<<4)
52
+
.align 4
53
53
expbit: .long 0x00800000
54
-
.align (1<<4)
54
+
.align 4
55
55
expmask:.long 0xff800000
56
56
57
57
.text
58
-
.align (1<<4)
59
-
.globl roundf
60
-
roundf:
58
+
.align 4
59
+
.globl _roundf
60
+
_roundf:
61
61
movd %xmm0, %eax
62
62
and $0x7f800000, %eax // |x|
63
63
mov $23, %ecx
···
79
79
paddd %xmm2, %xmm0 // add 0.5 (ish -- non-integral bits are garbage)
80
80
pand %xmm3, %xmm0 // truncate
81
81
1: ret
82
-
.align (1<<4)
82
+
.align 4
83
83
2: je 3f // if |x| >= 0.5f, goto 3
84
84
cvttps2dq %xmm0, %xmm1 // raise inexact
85
85
andps mzero(%rip), %xmm0 // copysign(0.0, x)
86
86
ret
87
-
.align (1<<4)
87
+
.align 4
88
88
3: cvttps2dq %xmm0, %xmm1 // raise inexact
89
89
paddd expbit(%rip), %xmm0
90
90
pand expmask(%rip), %xmm0 // copysign(1.0, x)
+17
-17
src/libm/Source/Intel/scalbn.S
+17
-17
src/libm/Source/Intel/scalbn.S
···
8
8
//
9
9
// -- Stephen Canon, January 2010
10
10
11
-
.globl scalbn
12
-
.globl ldexp
13
-
.globl scalb
14
-
.globl scalbln
11
+
.globl _scalbn
12
+
.globl _ldexp
13
+
.globl _scalb
14
+
.globl _scalbln
15
15
16
16
#if defined __x86_64__
17
17
···
26
26
#define clamplo mthreek(%rip)
27
27
28
28
.text
29
-
.align (1<<4)
30
-
scalb:
29
+
.align 4
30
+
_scalb:
31
31
movsd %xmm0, floatval
32
32
pcmpeqd %xmm0, %xmm0
33
33
fldl floatval // load x on x87
···
45
45
movapd %xmm1, %xmm0 // since n is a NaN, we just return n
46
46
ret
47
47
48
-
.align (1<<4)
49
-
scalbln:
48
+
.align 4
49
+
_scalbln:
50
50
movsd %xmm0, floatval
51
51
pcmpeqd %xmm0, %xmm0
52
52
fldl floatval // load x on x87
···
59
59
cmovl %rdx, ln // n == -3000
60
60
jmp L_common
61
61
62
-
.align (1<<4)
63
-
scalbn:
64
-
ldexp:
62
+
.align 4
63
+
_scalbn:
64
+
_ldexp:
65
65
movsd %xmm0, floatval
66
66
fldl floatval // load x on x87
67
67
pcmpeqd %xmm0, %xmm0
···
97
97
#define clamplo (mthreek-0b)(%ecx)
98
98
99
99
.text
100
-
.align (1<<4)
101
-
scalb:
100
+
.align 4
101
+
_scalb:
102
102
call 0f
103
103
0: pop %ecx // conjure pic info
104
104
movsd 12(%esp), %xmm1 // load n
···
118
118
fldl 12(%esp) // n is nan, return n
119
119
ret
120
120
121
-
.align (1<<4)
122
-
scalbln:
123
-
scalbn:
124
-
ldexp:
121
+
.align 4
122
+
_scalbln:
123
+
_scalbn:
124
+
_ldexp:
125
125
mov 12(%esp), n // load n
126
126
fldl floatval // load x on x87
127
127
fld1
+12
-12
src/libm/Source/Intel/scalbnf.S
+12
-12
src/libm/Source/Intel/scalbnf.S
···
8
8
9
9
10
10
.text
11
-
.globl scalblnf
12
-
.globl scalbnf
13
-
.globl ldexpf
11
+
.globl _scalblnf
12
+
.globl _scalbnf
13
+
.globl _ldexpf
14
14
15
15
#if defined __x86_64__
16
16
17
17
#define n %edi
18
18
#define ln %rdi
19
19
20
-
.align (1<<4)
21
-
scalblnf:
20
+
.align 4
21
+
_scalblnf:
22
22
cvtss2sd %xmm0, %xmm0 // (double)x
23
23
mov $300, %rdx
24
24
cmp %rdx, ln // if n > 300
···
28
28
cmovl %rdx, ln // n == -300
29
29
jmp L_common
30
30
31
-
.align (1<<4)
32
-
scalbnf:
33
-
ldexpf:
31
+
.align 4
32
+
_scalbnf:
33
+
_ldexpf:
34
34
cvtss2sd %xmm0, %xmm0 // (double)x
35
35
mov $300, %edx
36
36
cmp %edx, n // if n > 300
···
50
50
51
51
#define n %eax
52
52
53
-
.align (1<<4)
54
-
scalblnf:
55
-
scalbnf:
56
-
ldexpf:
53
+
.align 4
54
+
_scalblnf:
55
+
_scalbnf:
56
+
_ldexpf:
57
57
cvtss2sd 4(%esp), %xmm0 // (double)x
58
58
mov 8(%esp), n
59
59
mov $300, %edx
+12
-12
src/libm/Source/Intel/scalbnl.S
+12
-12
src/libm/Source/Intel/scalbnl.S
···
8
8
9
9
10
10
.text
11
-
.globl scalblnl
12
-
.globl scalbnl
13
-
.globl ldexpl
11
+
.globl _scalblnl
12
+
.globl _scalbnl
13
+
.globl _ldexpl
14
14
15
15
#if defined __x86_64__
16
16
···
20
20
#define ln %rdi
21
21
#define nw %di
22
22
23
-
.align (1<<4)
24
-
scalblnl:
23
+
.align 4
24
+
_scalblnl:
25
25
fldt 8(%rsp) // load x
26
26
fld1
27
27
fstpt floatval // conjure 1.0L
···
32
32
jl L_bigNegativeN // carefully.
33
33
jmp L_scaleX
34
34
35
-
.align (1<<4)
36
-
scalbnl:
37
-
ldexpl:
35
+
.align 4
36
+
_scalbnl:
37
+
_ldexpl:
38
38
fldt 8(%rsp) // load x
39
39
40
40
#elif defined __i386__
···
44
44
#define n %eax
45
45
#define nw %ax
46
46
47
-
.align (1<<4)
48
-
scalblnl:
49
-
scalbnl:
50
-
ldexpl:
47
+
.align 4
48
+
_scalblnl:
49
+
_scalbnl:
50
+
_ldexpl:
51
51
mov 20(%esp), n // load n
52
52
fldt floatval // load x
53
53
+9
-9
src/libm/Source/Intel/sinfcosf.S
+9
-9
src/libm/Source/Intel/sinfcosf.S
···
9
9
10
10
.literal8
11
11
12
-
.align (1<<3)
12
+
.align 3
13
13
14
14
Half: .double .5
15
15
16
16
17
-
//.const
17
+
.const
18
18
19
-
.align (1<<4)
19
+
.align 4
20
20
21
21
/* Coefficients to calculate x*(x**4+c0*x**2+c1) * c4*(x**4+c2*x**2+c3).
22
22
···
154
154
Exhaustive testing proved this routine returns faithfully rounded
155
155
results.
156
156
*/
157
-
.align (1<<5)
158
-
.globl sinf
159
-
sinf:
157
+
.align 5
158
+
.globl _sinf
159
+
_sinf:
160
160
161
161
// Put x into an integer register.
162
162
#if defined __i386__
···
361
361
Exhaustive testing proved this routine returns faithfully rounded
362
362
results.
363
363
*/
364
-
.align (1<<5)
365
-
.globl cosf
366
-
cosf:
364
+
.align 5
365
+
.globl _cosf
366
+
_cosf:
367
367
// This code is identical to _sinf except for the addition of Half.
368
368
369
369
// Put x into an integer register.
+4
-4
src/libm/Source/Intel/sinh.S
+4
-4
src/libm/Source/Intel/sinh.S
···
12
12
#include <machine/asm.h>
13
13
#include "abi.h"
14
14
15
-
//.const
16
-
.align (1<<4)
15
+
.const
16
+
.align 4
17
17
// Polynomial coefficients, offset from exp2table
18
18
.quad 0x3f55e52272e0eaec, 0x3f55e52272e0eaec // c4
19
19
.quad 0x401cc9eea1e24220, 0x401cc9eea1e24220 // c3/c4
···
161
161
.quad 0x3de61bfa2e91919a
162
162
163
163
.literal8
164
-
.align (1<<3)
164
+
.align 3
165
165
one_n7: .quad 0x3f80000000000000
166
166
lge_p7: .quad 0x40671547652b82fe
167
167
lge_hi: .quad 0x3ff7154760000000
···
178
178
#else
179
179
#define RELATIVE_ADDR(_a) (_a)-sinh_body(%ecx)
180
180
181
-
.align (1<<4)
181
+
.align 4
182
182
sinh_pic:
183
183
movl (%esp), %ecx
184
184
ret
+3
-3
src/libm/Source/Intel/sinhf.S
+3
-3
src/libm/Source/Intel/sinhf.S
···
47
47
// For the "uninteresting range", the code is taken directly from expf. Read the
48
48
// comments in that file for more information.
49
49
50
-
//.const
50
+
.const
51
51
52
-
.align (1<<4)
52
+
.align 4
53
53
.quad 0x3f811111111110fe, 0x3fa555f78359bc34 // c5, c4 = 0.0083333..., 0.0416715
54
54
.quad 0x4034000000000018, 0x4027ff4991a5ebc5 // c3/c5 = 0.1666666.../c5, c2/c4 = 0.5/c4
55
55
.quad 0x0000000000000000, 0x3ff0000000000000 // 0, 1
···
252
252
#define RELATIVE_ADDR(_a) (_a)-sinhf_body( %ecx )
253
253
#define RELATIVE_ADDR_B(_a) (_a)-sinhf_body_eleven( %ecx )
254
254
255
-
.align (1<<4)
255
+
.align 4
256
256
sinhf_pic:
257
257
movl (%esp), %ecx // Copy address of this instruction to %ecx
258
258
ret
+6
-6
src/libm/Source/Intel/tanf.S
+6
-6
src/libm/Source/Intel/tanf.S
···
9
9
10
10
.literal8
11
11
12
-
.align (1<<3)
12
+
.align 3
13
13
14
14
NegativeOne:
15
15
.double -1
16
16
17
17
18
-
//.const
18
+
.const
19
19
20
-
.align (1<<4)
20
+
.align 4
21
21
22
22
/* Coefficients to calculate:
23
23
···
167
167
Exhaustive testing proved this routine returns faithfully rounded
168
168
results.
169
169
*/
170
-
.align (1<<5)
171
-
.globl tanf
172
-
tanf:
170
+
.align 5
171
+
.globl _tanf
172
+
_tanf:
173
173
174
174
// Put x into an integer register.
175
175
#if defined __i386__
+3
-3
src/libm/Source/Intel/tanh.S
+3
-3
src/libm/Source/Intel/tanh.S
···
17
17
Tables and Constants
18
18
19
19
****************************************************************************/
20
-
//.const
21
-
.align (1<<4)
20
+
.const
21
+
.align 4
22
22
abs_mask: .quad 0x7fffffffffffffff, 0x0000000000000000
23
23
four_exp: .quad 0x0020000000000000, 0x0000000000000000
24
24
one: .quad 0x3ff0000000000000, 0x0000000000000000
···
324
324
.quad 0xbf29111400a41fbf, 0x0000000000000000 // c
325
325
326
326
.literal8
327
-
.align (1<<3)
327
+
.align 3
328
328
m2_lge_p7: .quad 0xc0771547652b82fe
329
329
m2_lge: .quad 0xc0071547652b82fe
330
330
m2_lge_hi: .quad 0xc007154760000000
+3
-5
src/libm/Source/Intel/tanhf.S
+3
-5
src/libm/Source/Intel/tanhf.S
···
11
11
#include <machine/asm.h>
12
12
#include "abi.h"
13
13
14
-
.type tanhf, @function
15
-
16
14
// This is identical to sinhf with some indices permuted to compute tanh instead.
17
15
// Read the detailed comments in that file for more information.
18
16
19
-
//.const
20
-
.align (1<<4)
17
+
.const
18
+
.align 4
21
19
.quad 0x3f811111111110fe, 0x3fa555f78359bc34 // c5, c4 = 0.0083333..., 0.0416715
22
20
.quad 0x4034000000000018, 0x4027ff4991a5ebc5 // c3/c5 = 0.1666666.../c5, c2/c4 = 0.5/c4
23
21
tanhf_table: .quad 0x0000000000000000, 0x3ff0000000000000 // sinh(0/16), cosh(0/16)
···
210
208
#define RELATIVE_ADDR(_a) (_a)-tanhf_body( %ecx )
211
209
#define RELATIVE_ADDR_B(_a) (_a)-tanhf_body_eleven( %ecx )
212
210
213
-
.align (1<<4)
211
+
.align 4
214
212
tanhf_pic:
215
213
movl (%esp), %ecx // Copy address of this instruction to %ecx
216
214
ret
+8
-8
src/libm/Source/Intel/trunc.S
+8
-8
src/libm/Source/Intel/trunc.S
···
25
25
#endif
26
26
27
27
.text
28
-
.align (1<<4)
29
-
.globl trunc
30
-
trunc:
28
+
.align 4
29
+
.globl _trunc
30
+
_trunc:
31
31
movl 8(%esp), %ecx
32
32
cmpl $0x43300000, %ecx
33
33
fldl 4(%esp)
···
47
47
48
48
#else //x86_64
49
49
50
-
//.const
51
-
.align (1<<4)
50
+
.const
51
+
.align 4
52
52
one: .quad 0x3ff0000000000000
53
53
absmask:.quad 0x7fffffffffffffff
54
54
half: .quad 0x3fe0000000000000
55
55
thresh: .quad 0x4330000000000000
56
56
57
57
.text
58
-
.align (1<<4)
59
-
.globl trunc
60
-
trunc:
58
+
.align 4
59
+
.globl _trunc
60
+
_trunc:
61
61
movd %xmm0, %rcx
62
62
andq absmask(%rip), %rcx
63
63
movsd absmask(%rip), %xmm2
+11
-11
src/libm/Source/Intel/truncf.S
+11
-11
src/libm/Source/Intel/truncf.S
···
12
12
#if defined __i386__
13
13
14
14
.text
15
-
.align (1<<4)
16
-
.globl truncf
17
-
truncf:
15
+
.align 4
16
+
.globl _truncf
17
+
_truncf:
18
18
movss 4(%esp), %xmm0 // load argument
19
19
testl $(kHasSSE4_1), cpubits
20
20
jz 0f
···
24
24
flds 4(%esp) //
25
25
ret
26
26
27
-
.align (1<<4)
27
+
.align 4
28
28
0: mov 4(%esp), %eax // load the input, x
29
29
and $0x7f800000, %eax // |x|
30
30
mov $23, %ecx
···
40
40
1: flds 4(%esp)
41
41
ret
42
42
43
-
.align (1<<4)
43
+
.align 4
44
44
2: // Handle |x| < 1.0 here.
45
45
andl $0x80000000, 4(%esp) // copysign(0.0, x)
46
46
cvttps2dq %xmm0, %xmm0 // raise inexact
···
49
49
50
50
#elif defined __x86_64__
51
51
52
-
//.const
53
-
.align (1<<4)
52
+
.const
53
+
.align 4
54
54
mzero: .long 0x80000000
55
55
56
56
.text
57
-
.align (1<<4)
58
-
.globl truncf
59
-
truncf:
57
+
.align 4
58
+
.globl _truncf
59
+
_truncf:
60
60
movd %xmm0, %eax
61
61
and $0x7f800000, %eax // |x|
62
62
mov $23, %ecx
···
71
71
cvttps2dq %xmm0, %xmm1 // raise inexact
72
72
andps %xmm2, %xmm0 // mask off non-integral bits
73
73
1: ret
74
-
.align (1<<4)
74
+
.align 4
75
75
2: cvttps2dq %xmm0, %xmm1 // raise inexact
76
76
andps mzero(%rip), %xmm0 // copysign(0.0, x)
77
77
ret