| Issue |
55396
|
| Summary |
[riscv] Reload of vlenb during frame setup/teardown with scalable vector alloca
|
| Labels |
backend:RISC-V,
llvm:codegen,
performance
|
| Assignees |
|
| Reporter |
preames
|
Noticed this while looking at something unrelated. We're reading vlenb twice whereas this should effectively be a constant for a given platform.
```
$ cat rvv-aligned-alloca.ll
declare void @callee(<vscale x 1 x i8>* %p)
define void @caller() {
%a = alloca <vscale x 1 x i8>
call void @callee(<vscale x 1 x i8>* %a)
ret void
}
$ ./opt -S -O3 rvv-aligned-alloca.ll | ./llc -O3 -march=riscv64 -mattr=+v
.text
.attribute 4, 16
.attribute 5, "rv64i2p0_f2p0_d2p0_v1p0_zve32f1p0_zve32x1p0_zve64d1p0_zve64f1p0_zve64x1p0_zvl128b1p0_zvl32b1p0_zvl64b1p0"
.file "rvv-aligned-alloca.ll"
.globl caller # -- Begin function caller
.p2align 2
.type caller,@function
caller: # @caller
.cfi_startproc
# %bb.0:
addi sp, sp, -16
.cfi_def_cfa_offset 16
sd ra, 8(sp) # 8-byte Folded Spill
.cfi_offset ra, -8
csrr a0, vlenb
sub sp, sp, a0
addi a0, sp, 8
call callee@plt
csrr a0, vlenb
add sp, sp, a0
ld ra, 8(sp) # 8-byte Folded Reload
addi sp, sp, 16
ret
.Lfunc_end0:
.size caller, .Lfunc_end0-caller
.cfi_endproc
```
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
