 J. Gareth "Curious Kit" Moreton
|
23217f4ac9
* x86: Added new post-peephole optimisations to convert BMI2 shift/rotate instructions to their smaller traditional versions if possible
|
3 weeks ago |
|
J. Gareth "Curious Kit" Moreton
|
54f7127048
* x86: New "Not/And -> Andn/Not" peephole optimisation
|
3 weeks ago |
 Pierre Muller
|
688e20f7a5
Try to fix bug report #41397
|
2 weeks ago |
|
Pierre Muller
|
e1d6a355b5
Set unititialized local boolean variables to false to avoid random behavior
|
2 weeks ago |
|
J. Gareth "Curious Kit" Moreton
|
84a4158979
* x86-64: Efficiency boost for "MovlMovq2MovlMovl 1" / "Mov2Nop 8" optimisation
|
1 month ago |
|
J. Gareth "Curious Kit" Moreton
|
a6af56cc0d
* x86-64: Fixed bug where the wrong operand was checked for the "Mov2Nop 8" peephole optimisation (fixes i41367)
|
1 month ago |
|
J. Gareth "Curious Kit" Moreton
|
37cd24f65a
* x86: Minor refactoring of "VmovdqxVmovdqxVmovdqxVmovdqx2VmovdqyVmovdqy" optimisations to avoid requirement of AVX2
|
3 months ago |
|
J. Gareth "Curious Kit" Moreton
|
5381fb2a4f
* x86_64: Moved "MovlMovq2MovlMovl 1" optimisation to pass 2 (fixes #41317)
|
2 months ago |
|
J. Gareth "Curious Kit" Moreton
|
90685b3b5d
* x86: MOV/AND optimisations now check for the flags being in use
|
3 months ago |
|
J. Gareth "Curious Kit" Moreton
|
833c668e29
* x86: New MOV/AND peephole optimisation and more efficient optimisation searching in "OptPass1MOV"
|
3 months ago |
|
J. Gareth "Curious Kit" Moreton
|
32b217bcd7
* x86_64: Deep MOV optimisation extended to attempt to replace registers after
|
3 months ago |
|
J. Gareth "Curious Kit" Moreton
|
63e2605928
* x86: AND %reg,%reg removal is now more thorough under -O3.
|
3 months ago |
|
J. Gareth "Curious Kit" Moreton
|
98d8830eb7
* x86: Refactored and optimised "PostPeepholeOptLea"
|
3 months ago |
|
J. Gareth "Curious Kit" Moreton
|
98563dd069
* x86: New post-peephole optimisation that converts 32-bit registers
|
3 months ago |
|
J. Gareth "Curious Kit" Moreton
|
6548dd0211
* x86: "MovMov2Mov 1" now removes the first instruction if the intermediate register is not in use afterwards.
|
4 months ago |
|
J. Gareth "Curious Kit" Moreton
|
d659ca7427
* x86: Fixed oversight in OptPass2Movx where a "movzwl/movzbl"
|
4 months ago |
|
J. Gareth "Curious Kit" Moreton
|
e7b6a08eae
* x86: New peephole optimisation for improving newly inserted (V)MOVD/(V)MOVQ instructions
|
5 months ago |
|
J. Gareth "Curious Kit" Moreton
|
ebe7bc8097
* x86: Arithmetic combination optimisations are now much more careful about the flags being in use
|
1 year ago |
|
J. Gareth "Curious Kit" Moreton
|
c39a3c6473
* x86: The ADD/SUB peephole optimisations that combine adjacent operations are now much more careful about the flags
|
1 year ago |
 florian
|
7afde9fe77
* don't do MovOpMov2MovOp optimization reducing bit width of the operation if symbols are involved, resolves #40269
|
5 months ago |
|
J. Gareth "Curious Kit" Moreton
|
1a65b058b3
* x86_64: Bug fix in "PrePeepholeOptSxx" where shifts greater than or equal to 32 weren't handled correctly
|
5 months ago |
|
J. Gareth "Curious Kit" Moreton
|
b611c0011d
* x86: Fixed bug and refactored optimisations where SHR instructions were
|
6 months ago |
|
J. Gareth "Curious Kit" Moreton
|
de51e6a32a
* x86_64: Extension to MovMov2Mov optimisations to allow more MOVL to MOVQ zero-extensions
|
7 months ago |
|
J. Gareth "Curious Kit" Moreton
|
a70e4ed9be
* x86: "OptPass2ADD" and "OptPass2SUB" now force another iteration if "OptPass2MOV" returned True
|
7 months ago |
|
J. Gareth "Curious Kit" Moreton
|
f85aa24ad0
* x86: Refactoring register update code in "OptPass2ADD" and "OptPass2SUB"
|
7 months ago |
|
J. Gareth "Curious Kit" Moreton
|
b2dd980329
* x86: Fixed bug where flags register wasn't tracked during
|
7 months ago |
|
J. Gareth "Curious Kit" Moreton
|
83c9b17ec5
* x86: Fixed oversight in LeaLea2LeaLea optimisations where bad code is
|
7 months ago |
|
J. Gareth "Curious Kit" Moreton
|
4648f967bd
* x86: Fixed bug in "MovMovMov2MovMovMov" optimisation where a register wasn't properly tracked
|
7 months ago |
|
florian
|
29b3b09251
* allow absolutes on open arrays (use of this is doubtful though), resolves #41147
|
7 months ago |
|
florian
|
125da5f10a
* SubLea2Lea has to check both times for equal super registers, resolves the second part of #41126
|
7 months ago |
