[powerpc64le] rustc 1.74.0 built by mrustc thinks u128::BITS is 64

[lnikkila]

First of all, thank you for your work on this monumental project!

I've been attempting to bootstrap rustc 1.74.0 on powerpc64le. I ran into a couple of small issues, will open a separate issue for the other one, which was easy to work around, but this latter one seems fairly mysterious to me.

The first sign is that the mrustc-built rustc fails to build the stdlib:

error[E0308]: mismatched types
    --> /var/tmp/mrustc/rustc-1.74.0-src/vendor/time-0.3.22/src/format_description/well_known/iso8601/adt_hack.rs:94:38
     |
94   |         EncodedConfig::from_be_bytes(bytes)
     |         ---------------------------- ^^^^^ expected an array with a fixed size of 16 elements, found one with 8 elements
     |         |
     |         arguments to this function are incorrect
     |

The related code comes from here, where EncodedConfig is an alias for u128:
https://github.com/time-rs/time/blob/v0.3.22/time/src/format_description/well_known/iso8601/adt_hack.rs#L58

I then wrote this test case, which succeeds with mrustc, but fails with the mrustc-built rustc:

$ cat foo.rs
fn main() {
  assert_eq!(128, u128::BITS);
}
$ ./bin/mrustc -L output-1.74.0/ foo.rs
$ ./foo
$ ./run_rustc/output-1.74.0/prefix-s/bin/rustc foo.rs
$ ./foo
thread 'main' panicked at foo.rs:2:3:
assertion `left == right` failed
  left: 128
 right: 64

u128::BITS comes from here:
https://github.com/rust-lang/rust/blob/1.74.0/library/core/src/num/uint_macros.rs#L59

For the record, swapping out u128::BITS with u128::MAX.count_ones() results in the correct value.

To my novice eyes this seems like some kind of problem with const evaluation, so I enabled MRUSTC_DEBUG='Constant Evaluate', but this output looks alright, assuming 80000000 is big-endian 0x80 = 128:

$ grep -R 'evaluate_constant: <u128>::BITS = ValueRef'
output-1.74.0/rustc-build/libcore.rlib_dbg.txt:Constant Evaluate-     evaluate_constant: <u128>::BITS = ValueRef({0+4}80000000)
output-1.74.0/libcore.rlib_dbg.txt:Constant Evaluate-     evaluate_constant: <u128>::BITS = ValueRef({0+4}80000000)

I'll keep updating this issue once I look into this more.

Build info

I'm building this on GNU Guix with this environment:

guix shell -e '(@ (gnu system) %base-packages)' \
  -e '(map cdr ((@ (guix build-system gnu) standard-packages)))' \
  --container --emulate-fhs --network --pure \
  cmake curl git perl pkg-config python zlib

Lightly edited build-1.74.0.sh:

#!/bin/bash
set -e
export CC=gcc CXX=g++ LLVM_TARGETS=PowerPC MRUSTC_DEBUG='Constant Evaluate' PARLEVEL=64 RUSTC_TARGET=powerpc64le-unknown-linux-gnu
export RUSTC_VERSION=1.74.0 MRUSTC_TARGET_VER=1.74 OUTDIR_SUF=-1.74.0
# Enables use of ccache in mrustc if it's available (i.e. ccache is on PATH)
command -v ccache >/dev/null && export MRUSTC_CCACHE=1
make
make RUSTCSRC
make -f minicargo.mk LIBS $@
make test $@
make local_tests $@

RUSTC_INSTALL_BINDIR=bin make -f minicargo.mk output-1.74.0/rustc $@
./output-1.74.0/rustc --version

LIBGIT2_SYS_USE_PKG_CONFIG=1 make -f minicargo.mk -j ${PARLEVEL:-1} output-1.74.0/cargo $@
./output-1.74.0/cargo --version

#./output-1.74.0/rustc samples/no_core.rs
#./output-1.74.0/rustc samples/1.rs
make -C run_rustc -j ${PARLEVEL:-1} $@

Whatever other diagnostic info I could think of:

$ uname -a
Linux guix 6.12.21-gnu #1 SMP 1 ppc64le GNU/Linux
$ gcc -v
Using built-in specs.
COLLECT_GCC=gcc
COLLECT_LTO_WRAPPER=/gnu/store/hfmcgblfdpwghgzr6v9mp7253r51dalj-gcc-11.4.0/libexec/gcc/powerpc64le-unknown-linux-gnu/11.4.0/lto-wrapper
Target: powerpc64le-unknown-linux-gnu
Configured with: 
Thread model: posix
Supported LTO compression algorithms: zlib
gcc version 11.4.0 (GCC) 
$ ./bin/mrustc -Z print-cfgs
Setup: V V V
(0.00 s) Setup: DONE
Target Load: V V V
(0.00 s) Target Load: DONE
>rust_compiler=mrustc
>target_abi=llvm
>target_arch=powerpc64
>target_endian=little
>target_env=gnu
>target_family=unix
>target_has_atomic=8
>target_has_atomic=16
>target_has_atomic=32
>target_has_atomic=64
>target_has_atomic=ptr
>target_has_atomic=cas
>target_has_atomic_equal_alignment=8
>target_has_atomic_equal_alignment=16
>target_has_atomic_equal_alignment=32
>target_has_atomic_equal_alignment=64
>target_has_atomic_equal_alignment=ptr
>target_has_atomic_load_store=8
>target_has_atomic_load_store=16
>target_has_atomic_load_store=32
>target_has_atomic_load_store=64
>target_has_atomic_load_store=ptr
>target_os=linux
>target_pointer_width=64
>target_vendor=gnu
>target_vendor=
>linux
>unix

[bjorn3]

80000000

Isn't this u64::MAX as hex number?

[lnikkila]

@bjorn3 I thought that too until I looked at the other values which look big-endian, here's a longer snippet:

$ grep -R 'evaluate_constant: <u[0-9]*>::BITS = ValueRef'
output-1.74.0/rustc-build/libcore.rlib_dbg.txt:Constant Evaluate-     evaluate_constant: <u8>::BITS = ValueRef({0+4}08000000)
output-1.74.0/rustc-build/libcore.rlib_dbg.txt:Constant Evaluate-     evaluate_constant: <u16>::BITS = ValueRef({0+4}10000000)
output-1.74.0/rustc-build/libcore.rlib_dbg.txt:Constant Evaluate-     evaluate_constant: <u32>::BITS = ValueRef({0+4}20000000)
output-1.74.0/rustc-build/libcore.rlib_dbg.txt:Constant Evaluate-     evaluate_constant: <u64>::BITS = ValueRef({0+4}40000000)
output-1.74.0/rustc-build/libcore.rlib_dbg.txt:Constant Evaluate-     evaluate_constant: <u128>::BITS = ValueRef({0+4}80000000)
output-1.74.0/libcore.rlib_dbg.txt:Constant Evaluate-     evaluate_constant: <u8>::BITS = ValueRef({0+4}08000000)
output-1.74.0/libcore.rlib_dbg.txt:Constant Evaluate-     evaluate_constant: <u16>::BITS = ValueRef({0+4}10000000)
output-1.74.0/libcore.rlib_dbg.txt:Constant Evaluate-     evaluate_constant: <u32>::BITS = ValueRef({0+4}20000000)
output-1.74.0/libcore.rlib_dbg.txt:Constant Evaluate-     evaluate_constant: <u64>::BITS = ValueRef({0+4}40000000)
output-1.74.0/libcore.rlib_dbg.txt:Constant Evaluate-     evaluate_constant: <u128>::BITS = ValueRef({0+4}80000000)

I believe that comes from here:

mrustc/src/hir_conv/constant_evaluation.cpp

Lines 234 to 242 in 06b87d1

	void fmt(::std::ostream& os, size_t ofs, size_t len) const override {
	assert(ofs <= length);
	assert(ofs+len <= length);
	for(size_t i = 0; i < len; i ++) {
	if(i != 0 && (ofs+i) % 8 == 0)
	os << " ";
	putb_hex(os, this->data[ofs + i]);
	}
	}