[missed optimization] Missed removing loop of setting zero when dereference pointer (memset)

[BreadTom]

Godbolt and GCC Bug

void f0(unsigned char *ptr, unsigned long len){
    for(unsigned long i = 0; i < len; ++i){
        ptr[i] = 0;
    }
}

void f1(unsigned char *ptr, unsigned long len){
    for(unsigned long i = 0; i < len; ++i){
        while(ptr[i])
            ptr[i]--;
    }
}

void f2(unsigned char *ptr){
    while(ptr[0])
        ptr[0]--;
}

When flags are -O3,
f1() compares if ptr[i] is zero before setting it to zero and f0() memset it to zero.

When flags are -O3 and -march=icelake-client, f1() doesn't call memset like f0().

[BreadTom]

Reduced godbolt code

void f3(unsigned char *ptr){
    if(ptr[0])
        ptr[0] = 0;
}

define dso_local void @f3(ptr noundef captures(none) %ptr) local_unnamed_addr {
entry:
  %0 = load i8, ptr %ptr, align 1
  %tobool.not = icmp eq i8 %0, 0
  br i1 %tobool.not, label %if.end, label %if.then

if.then:
  store i8 0, ptr %ptr, align 1
  br label %if.end

if.end:
  ret void
}

Just like GCC, LLVM does not have a pass for this.