#include <sel-sched-ir.h>
Data Fields | |
| int | type |
| rtx | lhs |
| rtx | rhs |
| regset | reg_sets |
| regset | reg_clobbers |
| regset | reg_uses |
Detailed Description
InstructionData. Contains information about insn pattern.
Field Documentation
| rtx idata_def::lhs |
If insn is a SET, this is its left hand side.
| regset idata_def::reg_clobbers |
| regset idata_def::reg_sets |
Registers that are set/used by this insn. This info is now gathered via sched-deps.c. The downside of this is that we also use live info from flow that is accumulated in the basic blocks. These two infos can be slightly inconsistent, hence in the beginning we make a pass through CFG and calculating the conservative solution for the info in basic blocks. When this scheduler will be switched to use dataflow, this can be unified as df gives us both per basic block and per instruction info. Actually, we don't do that pass and just hope for the best.
| regset idata_def::reg_uses |
| rtx idata_def::rhs |
If insn is a SET, this is its right hand side.
| int idata_def::type |
Type of the insn. o CALL_INSN - Call insn o JUMP_INSN - Jump insn o INSN - INSN that cannot be cloned o USE - INSN that can be cloned o SET - INSN that can be cloned and separable into lhs and rhs o PC - simplejump. Insns that simply redirect control flow should not have any dependencies. Sched-deps.c, though, might consider them as producers or consumers of certain registers. To avoid that we handle dependency for simple jumps ourselves.
The documentation for this struct was generated from the following file:
