Returning [i128::MIN; 1] from a function actually returns [0i128; 1]

[cppforliving]

Array created using repeat expression [i128::MIN; 1] and returned from a function actually returns [0i128; 1].
Interestingly the problem does not happen when using the list syntax [i128::MIN] to create the array.

I tried this code:

fn main() {
    fn min_array_ok() -> [i128; 1] {
        [i128::MIN]
    }
    assert_eq!(min_array_ok(), [-170141183460469231731687303715884105728i128]);
    
    fn min_array_nok() -> [i128; 1] {
        [i128::MIN; 1]
    }
    assert_eq!(min_array_nok(), [-170141183460469231731687303715884105728i128]); // panic!
}

I expected to see this happen: the second assert_eq should pass

Instead, this happened: the second assert_eq fails screaming that

thread 'main' panicked at 'assertion failed: `(left == right)`
  left: `[0]`,
 right: `[-170141183460469231731687303715884105728]`', src/main.rs:10:5

Meta

Problem exists on stable, beta and nightly

rustc --version --verbose:

rustc 1.63.0 (4b91a6ea7 2022-08-08)
binary: rustc
commit-hash: 4b91a6ea7258a947e59c6522cd5898e7c0a6a88f
commit-date: 2022-08-08
host: x86_64-unknown-linux-gnu
release: 1.63.0
LLVM version: 14.0.5

Backtrace

   0: rust_begin_unwind
             at /rustc/289279de116707f28cf9c18e4bbb8c6ec84ad75b/library/std/src/panicking.rs:584:5
   1: core::panicking::panic_fmt
             at /rustc/289279de116707f28cf9c18e4bbb8c6ec84ad75b/library/core/src/panicking.rs:142:14
   2: core::panicking::assert_failed_inner
   3: core::panicking::assert_failed
             at /rustc/289279de116707f28cf9c18e4bbb8c6ec84ad75b/library/core/src/panicking.rs:181:5
   4: playground::main
             at ./[src/main.rs:10](https://play.rust-lang.org/?version=nightly&mode=debug&edition=2021#):5
   5: core::ops::function::FnOnce::call_once
             at /rustc/289279de116707f28cf9c18e4bbb8c6ec84ad75b/library/core/src/ops/function.rs:248:5

I wonder if

rust/compiler/rustc_codegen_ssa/src/mir/rvalue.rs

Lines 90 to 94 in 8778809

	if bx.cx().const_to_opt_uint(v) == Some(0) {
	let fill = bx.cx().const_u8(0);
	bx.memset(start, fill, size, dest.align, MemFlags::empty());
	return bx;
	}

is at fault here.

It does look to be inappropriately turning the repeat into a memset (when looking at the llvm IR)

[saethlin]

https://godbolt.org/z/4xz1d4Tz9
Compiling this:

pub fn min_array_ok() -> [i128; 1] {
    [i128::MIN]
}

pub fn min_array_nok() -> [i128; 1] {
    [i128::MIN; 1]
}

with -Cno-prepopulate-passes -Copt-level=0 --emit=llvm-ir
Yields:

define void @_ZN7example12min_array_ok17h64ef6889d5bfe023E(ptr sret([1 x i128]) %0) unnamed_addr #0 !dbg !6 {
start:
  %1 = getelementptr inbounds [1 x i128], ptr %0, i64 0, i64 0, !dbg !11
  store i128 -170141183460469231731687303715884105728, ptr %1, align 8, !dbg !11
  ret void, !dbg !12
}

define void @_ZN7example13min_array_nok17h5f347fb6a42ac96fE(ptr sret([1 x i128]) %0) unnamed_addr #0 !dbg !13 {
start:
  %1 = getelementptr inbounds [1 x i128], ptr %0, i64 0, i64 0, !dbg !14
  call void @llvm.memset.p0.i64(ptr align 8 %1, i8 0, i64 16, i1 false), !dbg !14
  ret void, !dbg !15
}

Which does not seem right. We shouldn't be emitting a memset to 0 there.

[saethlin]

This program hits an LLVM assertion if compiled with an LLVM that has assertions enabled:

╰ ➤ cargo +master-stage1 run
   Compiling scratch v0.1.0 (/tmp/scratch)
rustc: /checkout/src/llvm-project/llvm/include/llvm/ADT/APInt.h:1469: uint64_t llvm::APInt::getZExtValue() const: Assertion `getActiveBits() <= 64 && "Too many bits for uint64_t"' failed.
error: could not compile `scratch`

Caused by:
  process didn't exit successfully: `rustc --crate-name scratch --edition=2021 src/main.rs --error-format=json --json=diagnostic-rendered-ansi,artifacts,future-incompat --crate-type bin --emit=dep-info,link -C embed-bitcode=no -C debuginfo=2 -C metadata=604a9a6ba3499cae -C extra-filename=-604a9a6ba3499cae --out-dir /tmp/scratch/target/debug/deps -C linker=clang -C incremental=/tmp/scratch/target/debug/incremental -L dependency=/tmp/scratch/target/debug/deps -C link-arg=-fuse-ld=mold` (signal: 6, SIGABRT: process abort signal)

[apiraino]

This appeared in 1.43.0. Until 1.42 the sample would code emit a compile error (godbolt link), maybe that would shadow the error?

A bisection between those two stable releases points to a group of old PR but I can't really pinpoint one :/

  commit[0] 2020-01-29UTC: Auto merge of #68634 - JohnTitor:clippy-up, r=JohnTitor
  commit[1] 2020-01-29UTC: Auto merge of #67688 - cjgillot:passes, r=Zoxc
  commit[2] 2020-01-30UTC: Auto merge of #68659 - Dylan-DPC:rollup-zo7zi9f, r=Dylan-DPC
  commit[3] 2020-01-30UTC: Auto merge of #68325 - faern:move-numeric-consts-to-associated-consts-step1, r=LukasKalbertodt
  commit[4] 2020-01-30UTC: Auto merge of #68661 - nnethercote:rm-unused-read_uleb128-param, r=Mark-Simulacrum

pub fn min_array_nok() -> [i128; 1] {
    [std::i128::MIN; 1]
}

fails even under 1.26. And before that, i128 didn't even exist. So this has been a problem since 128 bit ints were introduced, as far as I can see.