[llvm] [LoongArch] Enable LoopTermFold Pass (PR #130737)

[Lu Weining via llvm-commits]

SixWeining wrote:

> > What's the effect after enabling this pass?
> 
> This will reduce one `addi` instruction in loop.body but with extra setup cost. For most cases, this keep the same behavior with gcc.
> 
> clang temp.c -S -O1
> 
> ```c
> void foo(int *__restrict a, short int * __restrict b, int n) {
> 	for(int i = 0 ; i < n; i++ )
> 		a[i] = b[i];
> }
> ```
> 
> before
> 
> ```assembly
> # %bb.0:
> 	ori	$a3, $zero, 1
> 	blt	$a2, $a3, .LBB0_2
> 	.p2align	4, , 16
> .LBB0_1:                                # =>This Inner Loop Header: Depth=1
> 	ld.h	$a3, $a1, 0
> 	st.w	$a3, $a0, 0
> 	addi.d	$a0, $a0, 4
> 	addi.d	$a2, $a2, -1
> 	addi.d	$a1, $a1, 2
> 	bnez	$a2, .LBB0_1
> .LBB0_2:
> 	ret
> ```
> 
> after
> 
> ```assembly
> # %bb.0:                                # %entry
> 	ori	$a3, $zero, 1
> 	blt	$a2, $a3, .LBB0_3
> # %bb.1:                                # %for.body.preheader
> 	alsl.d	$a2, $a2, $a0, 2
> 	.p2align	4, , 16
> .LBB0_2:                                # %for.body
>                                         # =>This Inner Loop Header: Depth=1
> 	ld.h	$a3, $a1, 0
> 	st.w	$a3, $a0, 0
> 	addi.d	$a0, $a0, 4
> 	addi.d	$a1, $a1, 2
> 	bne	$a0, $a2, .LBB0_2
> .LBB0_3:                                # %for.cond.cleanup
> 	ret
> ```

I see. Could you pre-commit a dedicate IR testcase in llvm/test/CodeGen/LoongArch? 

https://github.com/llvm/llvm-project/pull/130737