J. Gareth "Curious Kit" Moreton
33dfb6cb4e
* x86: Converted bit-test optimisations to use
...
debug_hexstr so comments aren't truncated
(and put $ before BTx values)
2023-02-06 21:31:50 +00:00
J. Gareth "Curious Kit" Moreton
ae47cc2c88
* x86: New debug_hexstr instruction to help produce concise debug comments
2023-02-06 21:31:50 +00:00
florian
6e84ee5806
* patch by Jinyang He to fix building loongarch64 with -Oodfa/-O4
2023-02-06 22:26:02 +01:00
Jinyang He
783bbb66f9
Update Makefiles
2023-02-05 19:18:48 +00:00
Jinyang He
92e60d2107
add loongarch64 linux rtl support
2023-02-05 19:18:48 +00:00
Jinyang He
3d2c96ebbf
Add loongarch64 linux support to compiler
2023-02-05 19:18:48 +00:00
Jinyang He
12c4290ffe
Add loongarch64 architecture support to compiler
2023-02-05 19:18:48 +00:00
Jinyang He
8fac3d62ce
add has_double_field
2023-02-05 19:18:48 +00:00
Jinyang He
96f4300158
Fix the edge of first_imaginary in move op
2023-02-05 19:18:48 +00:00
Jinyang He
2d680afab7
Fix the ~ is NOR rather than NOT
2023-02-05 19:18:48 +00:00
Jinyang He
cda8971baa
Add dwarf cfa support
2023-02-05 19:18:48 +00:00
Jonas Maebe
7f41bb4718
JVM: add typesym for anonymous record typed constants
...
Fixes compilation of system unit with JVM compiler after b38d13577f
2023-02-05 10:58:59 +01:00
florian
9c10167b6f
+ CPUX86_HAS_BSWAP
2023-02-04 19:20:10 +01:00
Karoly Balogh
6bbb10e64d
m68k: do not generate LEA instructions with explicitly specified size. only long is possible anyway.
2023-01-30 09:56:41 +01:00
Karoly Balogh
c577ac5ce9
m68k: add tables to convert from some stub opcodes + condition into real opcodes. Use them in the external assembler writer for now. Later the internal assembler will need these tables too.
2023-01-30 09:56:41 +01:00
florian
e9ec4a8bb3
* more riscv32-freertos-esp32c3 stuff added
2023-01-29 19:30:11 +01:00
florian
09b435cdab
* transform a/a only into 1 if fastmath is on
2023-01-29 13:49:33 +01:00
florian
bedd4edc72
+ first work for esp32-c3 support
2023-01-28 21:28:19 +01:00
florian
048385fd7f
* compilation fixed
2023-01-28 11:42:52 +01:00
florian
3e75ba52f3
* initial preparations for riscv32-freertos
2023-01-27 22:39:15 +01:00
Sven/Sarah Barth
e8fcc5343b
* use new chartype_for_stringlike function to retrieve the char type for a stringlike def instead of casting to a stringdef
2023-01-26 07:07:34 +01:00
Sven/Sarah Barth
ea417fbaaf
+ add utility function to retrieve the char type def from a stringlike def
2023-01-26 07:07:34 +01:00
florian
dfb8794d4d
* compilation after merge fixed
2023-01-25 20:44:34 +01:00
Pierre Muller
aaa6f0d9c5
Only signed extension is needed
2023-01-25 19:36:45 +00:00
Pierre Muller
4793447be1
Add sign extension to 32-bit for unaligned OS_8 and OS_16 types (to try to solve #40102 )
2023-01-25 19:36:45 +00:00
florian
8ef0f59025
* mark constant symbols used in preprocessor expressions as used, resolves #40108
2023-01-24 22:36:26 +01:00
Jonas Maebe
821c22a38f
LLVM: workaround to fix make all
2023-01-23 22:44:48 +01:00
Jonas Maebe
e54197e9d2
Darwin: support -Xg for DWARFv4
2023-01-22 22:34:25 +01:00
florian
789031a6e6
* compile float*2 as float+float if possible as proposed by Rika, resolves #40116
2023-01-22 20:07:40 +01:00
florian
ec28b7586c
+ support procvar[] in tp/mac procvar mode if the procvar returns an array type
2023-01-22 16:24:29 +01:00
J. Gareth "Curious Kit" Moreton
f3f9c68ddb
* x86-64: Extended the movl/movq optimisation to cover more distance
2023-01-22 09:35:35 +00:00
J. Gareth "Curious Kit" Moreton
42b052eb83
* x86-64: movl/movq -> movl/movl zero-extension optimisation
2023-01-22 09:35:35 +00:00
Jonas Maebe
00a5d30300
LLVM: remove use of getelementptr in make_simple_ref_ptr
...
We cannot safely infer whether it needs to be indirect or not there,
it should be done at a higher level if appropriate.
2023-01-20 21:07:18 +01:00
Jonas Maebe
51fdbb5524
macOS: set max var/constalign to 64
...
For all non-ppc(64) targets. Fixes test/talignrec1.
2023-01-20 21:07:18 +01:00
Jonas Maebe
29bae2297f
tcginnode.pass_generate_code: use tlhcgobj.g_undefined_ok
...
Also override the code that uses this for LLVM with a variant that does not
need it for LLVM versions that do not support the freeze instruction.
2023-01-20 21:07:18 +01:00
Jonas Maebe
ab581c5c30
LLVM: override thlcgobj.g_undefined_ok
...
Uses the freeze instruction available in LLVM 10.0+. If we don't freeze undef/
poison values before using them in a calculation (even if that calculation is
something like "and 0", which masks the result completely), the result will
still be undef/poison and will keep propagating.
2023-01-20 21:07:18 +01:00
Jonas Maebe
ee0ad3ff86
thlcgobj.g_undefined_ok: new method
...
Specify that a register's contents may be undefined because of previous
operations (e.g. shifting by more bits than the register has, whose result
depends on the target architecture), but that the code generator will
later on ensure that if this is the case, that result will be overwritten/
masked.
2023-01-20 21:07:18 +01:00
Jonas Maebe
0b6942de8a
llvm:a fix opaque ptr IR for Objective-C protocol
...
Also for interfacecom functions and properties. Fixes test/tobjc9 and
test/tobjc9b with LLVM 15.0+
2023-01-20 21:07:18 +01:00
florian
55deefbab5
* do not crash on illegal with statement, resolves #40118
2023-01-18 23:03:28 +01:00
florian
5598ffc210
* x86_64: warn only on negative offsets if rsp is involed, second part of #40113
...
+ tests
2023-01-17 23:09:42 +01:00
florian
8b08486fa1
* do not warn on lea e/rsp with negative address offset, part of #40113
...
+ tests
2023-01-15 23:06:05 +01:00
Pierre Muller
5cedecbe25
Avoid float_raise call in FPC_SOFT_FPUX80 code
2023-01-14 11:41:20 +00:00
Pierre Muller
977ca19dc0
Set current_filepos in Internal assembler class also if DEBUG macro is set
2023-01-14 11:41:20 +00:00
florian
a9dca32689
* undid accidentely activated define
2023-01-13 22:34:57 +01:00
florian
e77ec69de4
* avoid internal error in case of an illegal generic declaration
2023-01-12 23:05:51 +01:00
J. Gareth "Curious Kit" Moreton
df80d4e612
* x86: CMOV optimisations now check optimizecputype rather than cputype
2023-01-10 22:23:58 +00:00
J. Gareth "Curious Kit" Moreton
4d57dee8d9
* x86: Extension to CanBeCMOV that permits a potentially unsafe reference if it appears in the previous comparison
2023-01-10 22:23:58 +00:00
J. Gareth "Curious Kit" Moreton
b108608b29
* x86: New CMP optimisation that mimics some of the CMOV optimisations and removes unnecessary conditions.
2023-01-10 22:23:58 +00:00
J. Gareth "Curious Kit" Moreton
bab60d819f
* x86: Refactored CMOV optimisations and permitted the use of constants in some cases.
2023-01-10 22:23:58 +00:00
J. Gareth "Curious Kit" Moreton
5f3749dc49
* x86: Introduced TrySwapMovOp method, and redesigned TrySwapMovCmp
...
to use it while also trying to move one more instruction back
2023-01-10 22:23:58 +00:00
