Note that this is an ugly change because ALL index variables used inside
bitset_foreach() MUST be switched to size_t now, or the magical

elm != (size_t) -1;

will fail :-(

[r28199]

[r28198]

[r28197]

[r28196]

[r28195]

[r28194]

Use PTR_INT_TYPE instead of long which is 32bit on Win64.

[r28193]

[r28189]

[r28188]

Only makes it hard to search for arch_register_type_* flags.

[r28185]

[r28184]

[r28183]

- split out cmov and popcnt feature, can be detected by cpuid
- use feature flags even if not intel or amd cpu
- p6 instructions cannot be detected by feature bit, set them on
  architecture detection
- detect old Geode

[r28182]

[r28181]

[r28180]

Allowed native 32bit code generation when running on 64bit x86.
Removed ebx save, gcc should handle this.
Enable native detection on windows again.

[r28179]

[r28177]

[r28176]

[r28175]

[r28172]

use a 'low-tech' solution for emitting 8- and 16-bit register names. This also fixes a recently introduced bug in ia32_build_8bit_reg_map_high

[r28168]

create new_tarval_from_long_double function and change argument of new_tarval_from_double to double - so bindings can get away with C89 types only

[r28167]

[r28166]

- add support for MSVC compiler
- fixed cpuid toggle code
- fixed wrong ussage of feature mask register
- detect atom, core2, K10
- removed panics ...

Probably the whole detection code should be either moved into the frontend
OR at least into some extra file, so it can be reused for the amd64 and the
upcoming new x86 backends.

[r28160]

Results of review by Mallon

[r28156]

[r28155]

[r28154]

[r28153]

remove strange init_once/inited checks from arch_init functions. Calling it twice is invalid (and crashs anyway)

[r28152]

[r28149]

[r28148]

[r28147]

[r28146]

[r28145]

Put definition of get_base_sc() into .c file to remove the (non-standard) inline specifier from official header.

Most users of this function are inside teh same c file, so the performance hit is small.

[r28144]

[r28143]

[r28142]

[r28141]

[r28140]

[r28139]