 florian
|
2e86cdb620
* allow swapping of operands for three operand instructions if they are commutative
|
hai 1 ano |
|
florian
|
fd68d3bfbb
* ensure that left and right have a result type set in tx86addnode.pass_1, resolves #40727
|
hai 1 ano |
 J. Gareth "Curious Kit" Moreton
|
2b7df4237b
* nf_pass1_done, nf_error, nf_processing and nf_do_not_execute
|
hai 1 ano |
|
J. Gareth "Curious Kit" Moreton
|
b804fc5164
* Flags specific to TAddNode have been moved to their own field
|
hai 1 ano |
|
J. Gareth "Curious Kit" Moreton
|
2a44ffe51b
* x86: BZHI optimisation now sets the subtract and shift nodes to 'do not execute' as they are skipped over.
|
%!s(int64=2) %!d(string=hai) anos |
|
J. Gareth "Curious Kit" Moreton
|
16bd996e74
* x86: 'and not' optimisation now sets the NOT node to 'do not execute' as it is skipped over
|
%!s(int64=2) %!d(string=hai) anos |
|
J. Gareth "Curious Kit" Moreton
|
7da9b4a988
* x86: The code generator will now attempt to manipulate "x and ((1 shl y) - 1)" to use BZHI
|
%!s(int64=2) %!d(string=hai) anos |
|
J. Gareth "Curious Kit" Moreton
|
a1ea1dd98e
* x86: Changed BMI2 check to BMI1 for ordinal ANDN optimisation
|
%!s(int64=2) %!d(string=hai) anos |
|
J. Gareth "Curious Kit" Moreton
|
626e23d89f
* x86: The code generator will now attempt manipulate "x and (not y)" (where x and y are ordinals) to use ANDN.
|
%!s(int64=2) %!d(string=hai) anos |
|
florian
|
f097d1ea35
* use the bts instruction on x86 if possible, the reg,reg variant is better than a mov/shl/or sequence
|
%!s(int64=3) %!d(string=hai) anos |
|
J. Gareth "Curious Kit" Moreton
|
6f24c8b4ef
* x86: Code generation fixes where FLAGS
|
%!s(int64=3) %!d(string=hai) anos |
|
J. Gareth "Curious Kit" Moreton
|
9b4033fa80
Fast mod
|
%!s(int64=4) %!d(string=hai) anos |
|
J. Gareth "Curious Kit" Moreton
|
2dc0995067
- Bug fix to new ADD/SUB optimisation where conditions are concerned
|
%!s(int64=3) %!d(string=hai) anos |
|
florian
|
01d6358f93
* some fixes for expectloc
|
%!s(int64=4) %!d(string=hai) anos |
 Florian Klämpfl
|
e121a753e1
* mark temps as deallocated
|
%!s(int64=4) %!d(string=hai) anos |
 Yuriy Sydorov
|
b8befe3d1f
* Fixed a typo in the condition, though it seems this code path is not used now.
|
%!s(int64=4) %!d(string=hai) anos |
|
florian
|
c02b43fb93
* change fma cpu flag into an fpu flag
|
%!s(int64=4) %!d(string=hai) anos |
|
florian
|
e46b426ca1
+ tx86addnode.second_opvector makes use of AVX instructions
|
%!s(int64=4) %!d(string=hai) anos |
|
florian
|
637976e83f
* patch by Marģers to unify internal error numbers, resolves #37888
|
%!s(int64=5) %!d(string=hai) anos |
|
florian
|
043c3df292
* properly handle boolean64 xor boolean64 on i386
|
%!s(int64=5) %!d(string=hai) anos |
|
florian
|
b2ed667a86
* swap fpu stack correctly when moving data into xmm registers
|
%!s(int64=5) %!d(string=hai) anos |
|
florian
|
5f67d830e7
* better rules if a 3 op mul operation is tried
|
%!s(int64=5) %!d(string=hai) anos |
|
florian
|
b4de0b2e9c
* insert fwait after fcompp, resolves #37136
|
%!s(int64=5) %!d(string=hai) anos |
|
florian
|
97efb03ae6
* call x86 addordinal for non short boolean evaluated boolean nodes
|
%!s(int64=5) %!d(string=hai) anos |
|
florian
|
fca4d3315b
* let the operation do the type conversion for x87 floating point operations
|
%!s(int64=5) %!d(string=hai) anos |
|
florian
|
3c16324f80
* x86: avoid to put variables into registers if they are handled by the x87-FPU
|
%!s(int64=5) %!d(string=hai) anos |
|
florian
|
91f3f022fe
* MM registers which might not be changed can be still used in 3 op mathematical operations as source
|
%!s(int64=5) %!d(string=hai) anos |
|
florian
|
5f3050961f
* load left properly into a register in tx86addnode.second_addordinal, resolves #36287
|
%!s(int64=5) %!d(string=hai) anos |
|
florian
|
c8f746b881
* do not allocate an extra register for some integer operations if not needed
|
%!s(int64=5) %!d(string=hai) anos |
 Jonas Maebe
|
ce598c15ec
* factored out the conditions under which add nodes need to perform
|
%!s(int64=6) %!d(string=hai) anos |
