- 19 Aug, 2014 1 commit
-
-
Matthias Braun authored
-
- 18 Aug, 2014 3 commits
-
-
Matthias Braun authored
-
Matthias Braun authored
-
Matthias Braun authored
-
- 17 Aug, 2014 1 commit
-
-
Tobias Rapp authored
-
- 15 Aug, 2014 5 commits
-
-
yb9976 authored
-
Matthias Braun authored
This also allows to cleanup the finish neg/add carry flag fix code a big.
-
Matthias Braun authored
-
Tobias Rapp authored
-
yb9976 authored
-
- 14 Aug, 2014 3 commits
- 13 Aug, 2014 1 commit
-
-
Matthias Braun authored
-
- 12 Aug, 2014 11 commits
-
-
Matthias Braun authored
- For equal destinations merge the switch table entries and Proj nodes - Handle Switches with just a default Proj like Jmp nodes
-
Matthias Braun authored
- Normalize switch table now (sort + remove default entries) - Correctly handle the case of multiple entries mapping to the same proj number.
-
Matthias Braun authored
-
Matthias Braun authored
-
Matthias Braun authored
- No strange split between no-phi and phi variants anymore. The phi variant can perform everything now. - Repeat till we reach the fixpoint.
-
Matthias Braun authored
-
Matthias Braun authored
The were some comments in the code indicating that in case of SSA destruction problems critical edges were not created. But we deal with these cases anyway during SSA destruction so no need to avoid them earlier.
-
Matthias Braun authored
-
Matthias Braun authored
-
Matthias Braun authored
-
yb9976 authored
This fixes backend/ia32_scheduled2.c.
-
- 11 Aug, 2014 3 commits
- 08 Aug, 2014 12 commits
-
-
Andreas Fried authored
-
Andreas Fried authored
-
Andreas Fried authored
Other code still expects DestAM operations to have mode M.
-
Andreas Fried authored
NotMem now has two unused outputs but shares its mode and its Proj numbers with the other DestAM operations.
-
Andreas Fried authored
-
Andreas Fried authored
-
Andreas Fried authored
Implemented for now: When the flags of one Sub or Cmp node are needed, but the flags of another one are currently alive, invert the condition code and avoid rematerialization.
-
Andreas Fried authored
-
Andreas Fried authored
When a Sub node is rematerialized, only its flags will be needed. In this case, the Sub can be replaced by a Cmp and there is no need to have a scratch register to store the Sub's result.
-
Andreas Fried authored
-
Andreas Fried authored
When a node is rematerialized because its flag result is needed, rewire later uses of this node (but not the flags) to use the rematerialized node's results. This reduces the lifetime of the node's results.
-
Andreas Fried authored
Now that the flag outputs of arithmetic operations are used, a node may have multiple connections to the flag producer, but only one of these actually reads the flags.
-