https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111725
Bug ID: 111725
Summary: Missed one vsetvl
Product: gcc
Version: 14.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: lehua.ding at rivai dot ai
Target Milestone: ---
online: https://godbolt.org/z/br456cPs8
C Code (compiler option: -march=rv32gcv -mabi=ilp32d -S -fno-schedule-insns
-fno-schedule-insns2 -fno-tree-vectorize -frename-registers -O2):
#include "riscv_vector.h"
float f (int8_t * restrict in, int8_t * restrict out, int n, int m, unsigned
cond, size_t vl, float scalar)
{
vbool64_t mask = *(vbool64_t*) (in + 1000000);
for (size_t i = 0; i < n; i++)
{
vfloat32mf2_t v = __riscv_vle32_v_f32mf2 ((float *)(in + i + 200),
__riscv_vsetvlmax_e32mf2 ());
__riscv_vse32_v_f32mf2 ((float *)(out + i + 200), v,
__riscv_vsetvlmax_e32mf2 ());
vfloat32mf2_t v2 = __riscv_vle32_v_f32mf2_tumu (mask, v, (float *)(in + i
+ 300), __riscv_vsetvlmax_e32mf2 ());
__riscv_vse32_v_f32mf2_m (mask, (float *)(out + i + 300), v2,
__riscv_vsetvlmax_e32mf2 ());
}
vfloat32m1_t v = *(vfloat32m1_t*)(in + 300000);
for (size_t i = 0; i < n; i++)
{
v = __riscv_vfmv_s_f_f32m1_tu (v, (scalar + i), 3);
}
v = __riscv_vfadd_vv_f32m1 (v,v, 3);
return __riscv_vfmv_f_s_f32m1_f32 (v);
}
Assembly Code:
f:
li a5,999424
addi t0,a5,576
add t1,a0,t0
vsetvli a4,zero,e32,mf2,tu,mu
vlm.v v0,0(t1)
beq a2,zero,.L2
addi t5,a0,200
addi t6,a1,300
addi a6,a2,200
add t3,a0,a6
.L3:
vle32.v v1,0(t5)
addi a3,t6,-100
vse32.v v1,0(a3)
addi t4,t5,100
vle32.v v1,0(t4),v0.t
vse32.v v1,0(t6),v0.t
addi t5,t5,1
addi t6,t6,1
bne t5,t3,.L3
li a5,299008
addi t0,a5,992
add t1,a0,t0
vl1re32.v v2,0(t1)
li a4,0
.L6:
fcvt.s.wu fa5,a4
fadd.s ft0,fa5,fa0
vfmv.s.f v2,ft0
addi a4,a4,1
bne a4,a2,.L6
vfadd.vv v3,v2,v2 # Miss vsetivli zero,3,e32,m1,ta,ma
vfmv.f.s fa0,v3
ret
.L2:
li t2,299008
addi a1,t2,992
add a0,a0,a1
vl1re32.v v2,0(a0)
vsetivli zero,3,e32,m1,ta,ma
vfadd.vv v3,v2,v2
vfmv.f.s fa0,v3
ret
