[all-commits] [llvm/llvm-project] 407b61: [LoongArch] Add support for selecting constant mat...

[wangleiat via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 407b613d7344a37662a9933d5684dd207917e577
      https://github.com/llvm/llvm-project/commit/407b613d7344a37662a9933d5684dd207917e577
  Author: wanglei <wanglei at loongson.cn>
  Date:   2022-04-15 (Fri, 15 Apr 2022)

  Changed paths:
    M llvm/lib/Target/LoongArch/LoongArchISelDAGToDAG.cpp
    M llvm/lib/Target/LoongArch/MCTargetDesc/CMakeLists.txt
    A llvm/lib/Target/LoongArch/MCTargetDesc/LoongArchMatInt.cpp
    A llvm/lib/Target/LoongArch/MCTargetDesc/LoongArchMatInt.h
    A llvm/test/CodeGen/LoongArch/imm.ll

  Log Message:
  -----------
  [LoongArch] Add support for selecting constant materializations.

Integer materializing can generate LU12I_W, ORI, LU32I_D, LU52I_D and
ADDI_W instructions.

According to the sign-extended behavior of these instructions
(except ORI), the generated instruction sequence can be improved.

For example, load -1 into general register:
The ADDI_W instruction performs the operation that the [31:0] bit data
in the general register `rj` plus the 12-bit immediate `simm12` sign
extension 32-bit data; the resultant [31:0] bit is sign extension, then
written into the general register `rd`.

Normal sequence:

```
lu12i.w $a0, -1
ori $a0, $a0, 2048
```

Improved with sign-extended instruction:

```
addi.w $a0, $zero,  -1
```

Reviewed By: SixWeining, MaskRay

Differential Revision: https://reviews.llvm.org/D123290