[ViewVC] Annotation of: cvs/libecb/Changes

TODO: naming of cutnions, index_to_coord, xy_to_s, xy_to_linear/seq, 32/64
TODO: returning packed coordinates, really
TODO: support order=0?
0x0001000c
        - added ecb_hilbert2d function family.
        - use __builtin_popcountl on 64 bit clang and gcc.
        - use 64 bit builtins, not 128 bit builtins, for ctz/clz on 64 bit cpus.
        - add ecb_clz* functions,
        - renamed ecb_gray* functions to have trailing bit width.
        - use ecb_function_ consistently for newly added functions.

0x0001000b
        - added ecb_gray*_{de,en}code functions.

0x0001000a
        - added ecb_ptrmix.

0x00010009
        - added ecb_i2a family of functions.
        - added ECB_64BIT_NATIVE.

TODO: 
08:30:06 <b_jonas> I think it could be worth to add a macro that works like alignof or _Alignof on sane 
                   compilers, and like __alignof on MS compilers that support it, see 
                   http://msdn.microsoft.com/en-us/library/45t0s5f4.aspx
08:30:24 <b_jonas> even if you can't support it on all the old compilers
08:31:17 <b_jonas> I'd also like a macro for alignas, but sadly, that seems impossible in general, because 
                   the MS compiler only has some half-attempt to do something similar but with different and 
                   more broken semantics, see http://msdn.microsoft.com/en-us/library/83ythb65.aspx
08:31:35 <b_jonas> but I wonder if some special case could still be worth to support
08:32:23 <b_jonas> probably not, because it would just account to making a union with a highly aligned type, 
                   which is something I can do on any compiler portably

TODO: #define ECB_IS_INTEGRAL(x) !((1 ? 1 : (x)) / 2)
      #define ECB_IS_INTEGRAL(x) (sizeof ((x) + 1.0f) != sizeof((x) + 1ULL))

TODO: ecb_minpot, either using bit tricks or ecb_ldXX

TODO: __builtin_popcountll exists...

TODO: __builtin_powi

TODO: https://gustedt.wordpress.com/2010/06/08/detect-empty-macro-arguments/

implement is_constant for c11: https://gustedt.wordpress.com/2013/08/22/testing-compile-time-constness-and-null-pointers-with-c11s-_generic/

#ifdef _MSC_VER

#include <stdlib.h>
#define bswap_32(x) _byteswap_ulong(x)
#define bswap_64(x) _byteswap_uint64(x)

#elif defined(__APPLE__)

// Mac OS X / Darwin features
#include <libkern/OSByteOrder.h>
#define bswap_32(x) OSSwapInt32(x)
#define bswap_64(x) OSSwapInt64(x)

#elif defined(__sun) || defined(sun)

#include <sys/byteorder.h>
#define bswap_32(x) BSWAP_32(x)
#define bswap_64(x) BSWAP_64(x)

#elif defined(__FreeBSD__)

#include <sys/endian.h>
#define bswap_32(x) bswap32(x)
#define bswap_64(x) bswap64(x)

#elif defined(__OpenBSD__)

#include <sys/types.h>
#define bswap_32(x) swap32(x)
#define bswap_64(x) swap64(x)

#elif defined(__NetBSD__)

#include <sys/types.h>
#include <machine/bswap.h>
#if defined(__BSWAP_RENAME) && !defined(__bswap_32)
#define bswap_32(x) bswap32(x)
#define bswap_64(x) bswap64(x)
#endif

#else

#include <byteswap.h>

#endif

TODO: generic poprcount etc., also  fast_t for them?

        - allow any rotate count in rcb_rot*, at the expense
          of bad optimisation results on some platforms or with
          some word sizes.
        - try to use inttypes.h on mingw.

0x00010008
        - aligned/unaligned load/store, bswap, host order
          conversion.
        - generic C++ ecb_rot[lr], ecb_popcount, ecb_ctz,
          ecb_is_pot32 and ecb_bitrev functions.

0x00010007
        - new ECB_OPTIMIZE_SIZE symbol.

0x00010006
        - new ECB_MEMORY_FENCE_RELAXED memory fence.
        - use acquire/eelease memory barriers on sun workshop pro, not read/write.
        - rely on c++ compiler barriers to do the right thing in gcc/clang.
        - change release memory fence to memory barrier on ia32/ia64.
        - apply ctz/ld patch for msc by Zsbán Ambrus.
        - ECB_PTRSIZE erroneously was 8 on most 32bit systems (
          found by Zsbán Ambrus).
        - improved compiletime detection of endianness, also, allow
          runtime detection to indicate other-than-big/little endianness.
        - no memory barrier neded on arm < 6.

0x00010005
        - improve ecb_binary16_to_float.
        - add ecb_float_to_binary16.
        - add ecb_binary16_to_binary32 and ecb_binary32_to_binary16 pair.

0x00010001
        - add ecb_is_pot32/64.
        - add intptr_t/uintptr_t.
        - add ECB_PTRSIZE.
        - more macros for C/C++ version checks.
        - support C11 atomics for memory fences.
        - support gcc-4.7 atomics for memory fences.
        - support m68k, m88k and sh (patch by Miod Vallat).
        - add ecb_binary16_to_float.

TODO: ecb_restrict_array etc. http://ue.tst.eu/5093eafd713ec5fda776d8065070aa4c.txt
TODO: ffs
64 bit variants of everything
TODO: examples from X for clz/ctz
TODO: arithmetic right shift
TODO: template/generic functions for x32/x64 and so on
TODO: #define ecb_integer_multiples_of(n,d) ((char (*)[d])(n) - (char (*)[d])0)
TODO: generalised shift
TODO: #define ECB_FAST_UNALIGNED_ACCESS
unsigned long gensh(unsigned long v, int x) {
int a, b;
    a = (v << x) & -(((unsigned int)x) < 32);
    x = -x;
    b = (v >> x) & -(((unsigned int)x) < 32);
    return a|b;
}

TODO: export(=dllexport) & hidden
TODO: flatten
TODO: warning(msg)
TODO: error(msg)
TODO: leaf (uh), noclone (hmmm)
TODO: nonnull, returns_nonnull
TODO: nothrow
TODO: used
TODO: trap
TODO: http://llvm.org/docs/doxygen/html/Compiler_8h_source.html

TODO: read/write unaligned macros
TODO: htonl and friends
TODO: macro to convert from unsigned to signed "the natural way"
TODO: ecb_static_assert, with message (just like boost), or somesuch, using array-declaration
TODO: alignof


Revision:	1.44
Committed:	Fri Mar 25 15:30:02 2022 UTC (2 years, 4 months ago) by root
Branch:	MAIN
Changes since 1.43:	+1 -0 lines
Log Message:	* empty log message *
#	User	Rev	Content
1	root	1.39	TODO: naming of cutnions, index_to_coord, xy_to_s, xy_to_linear/seq, 32/64
2			TODO: returning packed coordinates, really
3			TODO: support order=0?
4			0x0001000c
5			- added ecb_hilbert2d function family.
6	root	1.40	- use __builtin_popcountl on 64 bit clang and gcc.
7	root	1.41	- use 64 bit builtins, not 128 bit builtins, for ctz/clz on 64 bit cpus.
8	root	1.42	- add ecb_clz* functions,
9	root	1.43	- renamed ecb_gray* functions to have trailing bit width.
10	root	1.44	- use ecb_function_ consistently for newly added functions.
11	root	1.39
12	root	1.38	0x0001000b
13			- added ecb_gray*_{de,en}code functions.
14
15	root	1.37	0x0001000a
16			- added ecb_ptrmix.
17
18	root	1.34	0x00010009
19	root	1.35	- added ecb_i2a family of functions.
20	root	1.34	- added ECB_64BIT_NATIVE.
21
22	root	1.15	TODO:
23			08:30:06 <b_jonas> I think it could be worth to add a macro that works like alignof or _Alignof on sane
24			compilers, and like __alignof on MS compilers that support it, see
25			http://msdn.microsoft.com/en-us/library/45t0s5f4.aspx
26			08:30:24 <b_jonas> even if you can't support it on all the old compilers
27			08:31:17 <b_jonas> I'd also like a macro for alignas, but sadly, that seems impossible in general, because
28			the MS compiler only has some half-attempt to do something similar but with different and
29			more broken semantics, see http://msdn.microsoft.com/en-us/library/83ythb65.aspx
30			08:31:35 <b_jonas> but I wonder if some special case could still be worth to support
31			08:32:23 <b_jonas> probably not, because it would just account to making a union with a highly aligned type,
32			which is something I can do on any compiler portably
33
34			TODO: #define ECB_IS_INTEGRAL(x) !((1 ? 1 : (x)) / 2)
35			#define ECB_IS_INTEGRAL(x) (sizeof ((x) + 1.0f) != sizeof((x) + 1ULL))
36
37	root	1.28	TODO: ecb_minpot, either using bit tricks or ecb_ldXX
38
39			TODO: __builtin_popcountll exists...
40
41	root	1.26	TODO: __builtin_powi
42
43	root	1.15	TODO: https://gustedt.wordpress.com/2010/06/08/detect-empty-macro-arguments/
44
45	root	1.13	implement is_constant for c11: https://gustedt.wordpress.com/2013/08/22/testing-compile-time-constness-and-null-pointers-with-c11s-_generic/
46	root	1.12
47			#ifdef _MSC_VER
48
49			#include <stdlib.h>
50			#define bswap_32(x) _byteswap_ulong(x)
51			#define bswap_64(x) _byteswap_uint64(x)
52
53			#elif defined(__APPLE__)
54
55			// Mac OS X / Darwin features
56			#include <libkern/OSByteOrder.h>
57			#define bswap_32(x) OSSwapInt32(x)
58			#define bswap_64(x) OSSwapInt64(x)
59
60			#elif defined(__sun) \|\| defined(sun)
61
62			#include <sys/byteorder.h>
63			#define bswap_32(x) BSWAP_32(x)
64			#define bswap_64(x) BSWAP_64(x)
65
66			#elif defined(__FreeBSD__)
67
68			#include <sys/endian.h>
69			#define bswap_32(x) bswap32(x)
70			#define bswap_64(x) bswap64(x)
71
72			#elif defined(__OpenBSD__)
73
74			#include <sys/types.h>
75			#define bswap_32(x) swap32(x)
76			#define bswap_64(x) swap64(x)
77
78			#elif defined(__NetBSD__)
79
80			#include <sys/types.h>
81			#include <machine/bswap.h>
82			#if defined(__BSWAP_RENAME) && !defined(__bswap_32)
83			#define bswap_32(x) bswap32(x)
84			#define bswap_64(x) bswap64(x)
85			#endif
86
87			#else
88
89			#include <byteswap.h>
90
91			#endif
92
93	root	1.32	TODO: generic poprcount etc., also fast_t for them?
94	root	1.33
95	root	1.36	- allow any rotate count in rcb_rot*, at the expense
96			of bad optimisation results on some platforms or with
97			some word sizes.
98	root	1.33	- try to use inttypes.h on mingw.
99
100	root	1.30	0x00010008
101			- aligned/unaligned load/store, bswap, host order
102			conversion.
103	root	1.32	- generic C++ ecb_rot[lr], ecb_popcount, ecb_ctz,
104			ecb_is_pot32 and ecb_bitrev functions.
105	root	1.30
106	root	1.29	0x00010007
107			- new ECB_OPTIMIZE_SIZE symbol.
108
109	root	1.28	0x00010006
110	root	1.29	- new ECB_MEMORY_FENCE_RELAXED memory fence.
111	root	1.28	- use acquire/eelease memory barriers on sun workshop pro, not read/write.
112			- rely on c++ compiler barriers to do the right thing in gcc/clang.
113	root	1.27	- change release memory fence to memory barrier on ia32/ia64.
114	root	1.23	- apply ctz/ld patch for msc by Zsbán Ambrus.
115	root	1.24	- ECB_PTRSIZE erroneously was 8 on most 32bit systems (
116			found by Zsbán Ambrus).
117	root	1.25	- improved compiletime detection of endianness, also, allow
118			runtime detection to indicate other-than-big/little endianness.
119	root	1.26	- no memory barrier neded on arm < 6.
120	root	1.23
121	root	1.22	0x00010005
122			- improve ecb_binary16_to_float.
123			- add ecb_float_to_binary16.
124			- add ecb_binary16_to_binary32 and ecb_binary32_to_binary16 pair.
125
126	root	1.6	0x00010001
127			- add ecb_is_pot32/64.
128			- add intptr_t/uintptr_t.
129			- add ECB_PTRSIZE.
130	root	1.7	- more macros for C/C++ version checks.
131			- support C11 atomics for memory fences.
132			- support gcc-4.7 atomics for memory fences.
133	root	1.10	- support m68k, m88k and sh (patch by Miod Vallat).
134	root	1.11	- add ecb_binary16_to_float.
135	root	1.6
136	root	1.9	TODO: ecb_restrict_array etc. http://ue.tst.eu/5093eafd713ec5fda776d8065070aa4c.txt
137	sf-exg	1.20	TODO: ffs
138	root	1.1	64 bit variants of everything
139	root	1.2	TODO: examples from X for clz/ctz
140	root	1.3	TODO: arithmetic right shift
141			TODO: template/generic functions for x32/x64 and so on
142	root	1.4	TODO: #define ecb_integer_multiples_of(n,d) ((char ()[d])(n) - (char ()[d])0)
143	root	1.3	TODO: generalised shift
144	root	1.8	TODO: #define ECB_FAST_UNALIGNED_ACCESS
145	root	1.3	unsigned long gensh(unsigned long v, int x) {
146			int a, b;
147			a = (v << x) & -(((unsigned int)x) < 32);
148			x = -x;
149			b = (v >> x) & -(((unsigned int)x) < 32);
150			return a\|b;
151			}
152	root	1.14
153	root	1.16	TODO: export(=dllexport) & hidden
154			TODO: flatten
155			TODO: warning(msg)
156			TODO: error(msg)
157	root	1.17	TODO: leaf (uh), noclone (hmmm)
158			TODO: nonnull, returns_nonnull
159			TODO: nothrow
160			TODO: used
161			TODO: trap
162			TODO: http://llvm.org/docs/doxygen/html/Compiler_8h_source.html
163	root	1.16
164	root	1.14	TODO: read/write unaligned macros
165			TODO: htonl and friends
166	sf-exg	1.20	TODO: macro to convert from unsigned to signed "the natural way"
167	root	1.14	TODO: ecb_static_assert, with message (just like boost), or somesuch, using array-declaration
168			TODO: alignof
169
170