florian
|
f00a55ae08
+ tinlinenode.createintern
|
пре 10 година |
Jonas Maebe
|
67b8aceaee
* synchronized with privatetrunk till r30095
|
пре 10 година |
Jonas Maebe
|
68303b8df1
* add explicit check that argument of sizeof(x) is a simple load node (this
|
пре 10 година |
florian
|
0dfbba86f3
* properly firstpass nodes when inc(...) is converted into ...:=...+x, resolves #27471
|
пре 10 година |
florian
|
c9ba9665dd
* set nf_internal for if nodes generated by assert, prevent warning for assert(true); resolves #26467
|
пре 10 година |
nickysn
|
a9e1419ee3
* extracted the pass 1 handling of inc/dec to a virtual method first_IncDec
|
пре 10 година |
pierre
|
a77735d7af
Fix bug in in_const_swap_qword handling
|
пре 10 година |
Jonas Maebe
|
e8026ad8d3
* fixed the resultdef of length() intrinsic: it's ossinttype rather than
|
пре 11 година |
sergei
|
04d8e8a5dc
* On 64-bit targets, handle abs(int64) internally, using the same code as for abs(longint), i.e. without branching. Both generic and x86-specific pass 2 code is already suitable for different operand sizes, only type checking needs removal of excessive conversions to 32 bits.
|
пре 11 година |
svenbarth
|
34394d6925
Fix for Mantis #26180. Accept undefineddef as first parameter type of an Assert if it is used inside a generic.
|
пре 11 година |
sergei
|
274c2fc82f
* handle_ln_const: if/else branches were swapped, correct values are: ln(0)=-Inf and ln(-#)=NaN.
|
пре 11 година |
sergei
|
743326dc70
* abs(): when translated to generic fpc_abs_real() helper, typecast result back to original precision, similar to how it is done for sqr and sqrt. Fixes compilation on m68k after r27808 (m86k looks like the only target still needing this generic handling).
|
пре 11 година |
sergei
|
37625211f5
* Fixed sqrt procedure selection on arm-wince after r27809.
|
пре 11 година |
sergei
|
22e099d000
* sqrt(real): for targets with emulated floating point, invoke float64_sqrt or float32_sqrt from softfpu code. Testing on ARM CPU shows that float64_sqrt executes about twice faster and offers better accuracy than fpc_sqrt_real from genmath.inc.
|
пре 11 година |
sergei
|
fc5f45f65c
* sqr(real) and sqrt(real): remove typeconv node inserted by initial call processing (see explanation in comments), allowing these functions to be evaluated using precision of argument. In particular, sqrt(single) and sqrt(double) now emit 'sqrtss' and 'sqrtsd' instructions on x86 targets with -Cfsse3. Non-x86 targets already have the necessary support in code generators.
|
пре 11 година |
florian
|
d88d644925
+ support for FMA intrinsic: if there is no hardware support, the compiler throws an error.
|
пре 11 година |
svenbarth
|
256c852631
Fix for Mantis #23776.
|
пре 11 година |
Jonas Maebe
|
d452686c39
* moved pbestrealtype from symdef to symcpu
|
пре 11 година |
Jonas Maebe
|
b57c95043f
+ support overriding tdef/tsym methods with target-specific functionality:
|
пре 11 година |
Jonas Maebe
|
0edb7b2288
* fixed r27320 for darwin, as its mangled names get prefixed with '_'
|
пре 11 година |
svenbarth
|
a08e8ff982
Fix for Mantis #25929
|
пре 11 година |
florian
|
ab3f5744e0
* skip dec/inc type checking in unspecialized generic methods, resolves #25603
|
пре 11 година |
florian
|
e841027a48
* fixes handling of typed files in iso mode
|
пре 11 година |
nickysn
|
e1332304ef
* when converting Hi() to a shr node, convert the right shr const parameter to
|
пре 11 година |
Jonas Maebe
|
d2b8275b99
+ support for arbitrary encodings in readstr/writestr
|
пре 11 година |
sergei
|
8ecad469a6
* Reverted r26256 since it produced wrong results if argument of inc/dec is actually unaligned. Added description of the issue and slightly increased node complexity at which address is cached, so operations on record/class fields do not take address.
|
пре 11 година |
sergei
|
b376f56bf3
* Use typed pointer to store address of x when processing inc(x)/dec(x) with range checking enabled. This preserves alignment of x and prevents generating unaligned code on alignment-sensitive targets.
|
пре 11 година |
florian
|
4d5119bf1c
* fixes several issues which cause warnings by the dfa code when using it to detect uninitialized variables
|
пре 11 година |
florian
|
412dc6586f
* the writestr helper initializes the target string, so mark it as being written
|
пре 11 година |
sergei
|
9ad98a2c4d
* Handle "SarInt64" always as internal procedure, introduced virtual method tinlinenode.first_sar that by default converts it into call to "fpc_sarint64" unless target CPU is 64-bit. This provides a point to insert target-specific optimizations.
|
пре 11 година |