[llvm-dev] Mapping between instruction ins/outs to instruction bits-field (original) (raw)
Yaniv via llvm-dev llvm-dev at lists.llvm.org
Wed Nov 8 21:09:11 PST 2017
- Previous message: [llvm-dev] Mapping between instruction ins/outs to instruction bits-field
- Next message: [llvm-dev] Questions about code-size optimizations in ARM backend
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
CGIOperandList::CGIOperandList() ignores names that are specified in MIOperandInfo (Not my preferred logic).
I did find "noNamedPositionallyEncodedOperands" that will probably do fine in most of my cases.
Anyway, thanks for your help.
On Tue, 2017-11-07 at 17:27 +0200, Yaniv wrote:
Hi,
How does the mapping between instruction inputs and outputs to the instruction bits-field is done? From what I saw it either by name matching or by positional matching of unassigned bits-fields, and the two methods can't be mixed. Is it correct? Supposing the above is correct then, setting inputs names in MIOperandInfo (of custom Operand) should behave as if they were specified directly in the out/ins list? Thanks, Yaniv
- Previous message: [llvm-dev] Mapping between instruction ins/outs to instruction bits-field
- Next message: [llvm-dev] Questions about code-size optimizations in ARM backend
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]