| Issue |
86918
|
| Summary |
[Question] Weird AArch64 assemblies
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
wormtql
|
LLVM: 17.0.2
When I compile libgmp into AArch64 assemblies (clang++ -O3), I got some weird instructions:
```
.LBB28_35:
sub x8, x29, #312
ldur x8, [x8, #-256] // 8-byte Folded Reload
lsr x8, x8, #0 // that's weird
add x8, x8, #15
and x9, x8, #0xfffffffffffffff0
mov x8, sp
subs x0, x8, x9
mov sp, x0
sub x8, x29, #328
stur x0, [x8, #-256] // 8-byte Folded Spill
b .LBB28_37
```
the instruction `lsr x8, x8, #0` is confusing. To my best knowledge, I think it just did nothing
another problem:
```
.LBB200_16:
ldr x8, [sp, #72] // 8-byte Folded Reload
ldr x12, [sp, #88] // 8-byte Folded Reload
ldur x11, [x29, #-24] // 8-byte Folded Reload
ldur w13, [x29, #-68] // 4-byte Folded Reload
ldur w14, [x29, #-72] // 4-byte Folded Reload
ldr x9, [sp, #96] // 8-byte Folded Reload
mov w10, w9 // move to w10
and w10, w10, w14, lsl #1 // but immediately overwrite w10
eor w10, w10, w13
asr x13, x11, #63
add x11, x11, x13
eor x11, x11, x13
stur x12, [x29, #-112] // 8-byte Folded Spill
stur x11, [x29, #-104] // 8-byte Folded Spill
stur w10, [x29, #-92] // 4-byte Folded Spill
stur x9, [x29, #-88] // 8-byte Folded Spill
stur x8, [x29, #-80] // 8-byte Folded Spill
b .LBB200_17
```
why not use one instruction `and w10, w9, w14, lsl #1` instead of the two instruction above?
another problem:
```
add x11, x11, x13
subs x8, x8, x11
cset w14, ne
// implicit-def: $x8
mov w8, w14
sbfx x11, x8, #0, #1
mov w8, #2 // =0x2
// kill: def $x8 killed $w8
and w14, w14, #0x1 // I think this instruction is redundant?
ands w14, w14, #0x1
csinc x8, x8, xzr, ne
subs x10, x10, #31
add x12, x12, x13
subs x10, x10, x12
```
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
