Wasmtime GitHub notifications bot (Dec 01 2021 at 21:47): Wasmtime GitHub notifications bot (Dec 01 2021 at 21:47):alexcrichton labeled issue #3581:
I ran into an odd issue today where https://github.com/bytecodealliance/wasmtime/pull/3579 failed on CI, but it didn't change anything to the point that nothing in theory should have failed, much less a clif codegen test. I'm going to copy the failure here because I'm about to rerun CI at which point I suspect the logs will get overwritten:
<details>
... thread 'worker #0' panicked at 'shuffle mask values must be between 0 and 31', cranelift/codegen/src/isa/x64/lower.rs:5519:21 stack backtrace: 0: std::panicking::begin_panic at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/std/src/panicking.rs:543:12 1: cranelift_codegen::isa::x64::lower::lower_insn_to_regs at ./codegen/src/isa/x64/lower.rs:5519:21 2: cranelift_codegen::isa::x64::lower::<impl cranelift_codegen::machinst::lower::LowerBackend for cranelift_codegen::isa::x64::X64Backend>::lower at ./codegen/src/isa/x64/lower.rs:6115:9 3: cranelift_codegen::machinst::lower::Lower<I>::lower_clif_block at ./codegen/src/machinst/lower.rs:750:17 4: cranelift_codegen::machinst::lower::Lower<I>::lower at ./codegen/src/machinst/lower.rs:991:17 5: cranelift_codegen::machinst::compile::compile at ./codegen/src/machinst/compile.rs:29:9 6: cranelift_codegen::isa::x64::X64Backend::compile_vcode at ./codegen/src/isa/x64/mod.rs:53:9 7: <cranelift_codegen::isa::x64::X64Backend as cranelift_codegen::machinst::MachBackend>::compile_function at ./codegen/src/isa/x64/mod.rs:64:21 8: cranelift_codegen::context::Context::compile at ./codegen/src/context.rs:172:22 9: cranelift_filetests::function_runner::compile at ./filetests/src/function_runner.rs:247:21 10: cranelift_filetests::function_runner::SingleFunctionCompiler::compile at ./filetests/src/function_runner.rs:76:25 11: <cranelift_filetests::test_run::TestRun as cranelift_filetests::subtest::SubTest>::run at ./filetests/src/test_run.rs:64:35 12: cranelift_filetests::runone::run_one_test at ./filetests/src/runone.rs:187:5 13: cranelift_filetests::runone::run at ./filetests/src/runone.rs:132:9 14: cranelift_filetests::concurrent::worker_thread::{{closure}}::{{closure}} at ./filetests/src/concurrent.rs:150:46 15: std::panicking::try::do_call at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/std/src/panicking.rs:403:40 16: __rust_try 17: std::panicking::try at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/std/src/panicking.rs:367:19 18: std::panic::catch_unwind at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/std/src/panic.rs:129:14 19: cranelift_filetests::concurrent::worker_thread::{{closure}} at ./filetests/src/concurrent.rs:150:30 note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace. ... FAIL filetests/filetests/runtests/simd-shuffle.clif: panicked in worker #0: shuffle mask values must be between 0 and 31 ... 342 tests thread 'filetests' panicked at 'test harness: 1 failure', cranelift/tests/filetests.rs:5:10 stack backtrace: 0: rust_begin_unwind at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/std/src/panicking.rs:517:5 1: core::panicking::panic_fmt at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/core/src/panicking.rs:101:14 2: core::result::unwrap_failed at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/core/src/result.rs:1617:5 3: core::result::Result<T,E>::expect at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/core/src/result.rs:1259:23 4: filetests::filetests at ./tests/filetests.rs:4:5 5: filetests::filetests::{{closure}} at ./tests/filetests.rs:2:1 6: core::ops::function::FnOnce::call_once at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/core/src/ops/function.rs:227:5 7: core::ops::function::FnOnce::call_once at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/core/src/ops/function.rs:227:5 note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace. failures: filetests test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out; finished in 9.75s ...</details>
or, effectively:
... thread 'worker #0' panicked at 'shuffle mask values must be between 0 and 31', cranelift/codegen/src/isa/x64/lower.rs:5519:21 ... FAIL filetests/filetests/runtests/simd-shuffle.clif: panicked in worker #0: shuffle mask values must be between 0 and 31 ...This looks like a "spurious" error in that it doesn't reproduce on other PRs on CI. The code in question, though:
is specifically gated for avx512 features, and cranelift filetests do indeed use
cranelift-nativeso my guess as to why this only showed up on my PR is that it just-so-happened to run on a machine with avx512 features enabled on its CPU. Assuming that's enabled I think it explains why this panics.This leads me to the actual issue here, the clif test in question specifically uses out-of-bounds indices for a shuffle argument:
function %shuffle_i8x16(i8x16, i8x16) -> i8x16 { block0(v0: i8x16, v1: i8x16): v2 = shuffle v0, v1, [3 0 31 26 100 6 12 11 23 13 24 4 2 15 17 5] return v2 }which looks like what's tripping the assert in the avx512 branch I think. I can reproduce this myself with a clif test that looks like:
test compile set enable_simd target x86_64 has_avx512vl has_avx512vbmi function %shuffle_i8x16(i8x16, i8x16) -> i8x16 { block0(v0: i8x16, v1: i8x16): v2 = shuffle v0, v1, [3 0 31 26 100 6 12 11 23 13 24 4 2 15 17 5] return v2 }this will fail with the same panic in
clif-util test foo.clifcc @abrown @jlb6740
Wasmtime GitHub notifications bot (Dec 01 2021 at 21:47):alexcrichton opened issue #3581:
I ran into an odd issue today where https://github.com/bytecodealliance/wasmtime/pull/3579 failed on CI, but it didn't change anything to the point that nothing in theory should have failed, much less a clif codegen test. I'm going to copy the failure here because I'm about to rerun CI at which point I suspect the logs will get overwritten:
<details>
... thread 'worker #0' panicked at 'shuffle mask values must be between 0 and 31', cranelift/codegen/src/isa/x64/lower.rs:5519:21 stack backtrace: 0: std::panicking::begin_panic at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/std/src/panicking.rs:543:12 1: cranelift_codegen::isa::x64::lower::lower_insn_to_regs at ./codegen/src/isa/x64/lower.rs:5519:21 2: cranelift_codegen::isa::x64::lower::<impl cranelift_codegen::machinst::lower::LowerBackend for cranelift_codegen::isa::x64::X64Backend>::lower at ./codegen/src/isa/x64/lower.rs:6115:9 3: cranelift_codegen::machinst::lower::Lower<I>::lower_clif_block at ./codegen/src/machinst/lower.rs:750:17 4: cranelift_codegen::machinst::lower::Lower<I>::lower at ./codegen/src/machinst/lower.rs:991:17 5: cranelift_codegen::machinst::compile::compile at ./codegen/src/machinst/compile.rs:29:9 6: cranelift_codegen::isa::x64::X64Backend::compile_vcode at ./codegen/src/isa/x64/mod.rs:53:9 7: <cranelift_codegen::isa::x64::X64Backend as cranelift_codegen::machinst::MachBackend>::compile_function at ./codegen/src/isa/x64/mod.rs:64:21 8: cranelift_codegen::context::Context::compile at ./codegen/src/context.rs:172:22 9: cranelift_filetests::function_runner::compile at ./filetests/src/function_runner.rs:247:21 10: cranelift_filetests::function_runner::SingleFunctionCompiler::compile at ./filetests/src/function_runner.rs:76:25 11: <cranelift_filetests::test_run::TestRun as cranelift_filetests::subtest::SubTest>::run at ./filetests/src/test_run.rs:64:35 12: cranelift_filetests::runone::run_one_test at ./filetests/src/runone.rs:187:5 13: cranelift_filetests::runone::run at ./filetests/src/runone.rs:132:9 14: cranelift_filetests::concurrent::worker_thread::{{closure}}::{{closure}} at ./filetests/src/concurrent.rs:150:46 15: std::panicking::try::do_call at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/std/src/panicking.rs:403:40 16: __rust_try 17: std::panicking::try at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/std/src/panicking.rs:367:19 18: std::panic::catch_unwind at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/std/src/panic.rs:129:14 19: cranelift_filetests::concurrent::worker_thread::{{closure}} at ./filetests/src/concurrent.rs:150:30 note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace. ... FAIL filetests/filetests/runtests/simd-shuffle.clif: panicked in worker #0: shuffle mask values must be between 0 and 31 ... 342 tests thread 'filetests' panicked at 'test harness: 1 failure', cranelift/tests/filetests.rs:5:10 stack backtrace: 0: rust_begin_unwind at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/std/src/panicking.rs:517:5 1: core::panicking::panic_fmt at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/core/src/panicking.rs:101:14 2: core::result::unwrap_failed at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/core/src/result.rs:1617:5 3: core::result::Result<T,E>::expect at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/core/src/result.rs:1259:23 4: filetests::filetests at ./tests/filetests.rs:4:5 5: filetests::filetests::{{closure}} at ./tests/filetests.rs:2:1 6: core::ops::function::FnOnce::call_once at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/core/src/ops/function.rs:227:5 7: core::ops::function::FnOnce::call_once at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/core/src/ops/function.rs:227:5 note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace. failures: filetests test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out; finished in 9.75s ...</details>
or, effectively:
... thread 'worker #0' panicked at 'shuffle mask values must be between 0 and 31', cranelift/codegen/src/isa/x64/lower.rs:5519:21 ... FAIL filetests/filetests/runtests/simd-shuffle.clif: panicked in worker #0: shuffle mask values must be between 0 and 31 ...This looks like a "spurious" error in that it doesn't reproduce on other PRs on CI. The code in question, though:
is specifically gated for avx512 features, and cranelift filetests do indeed use
cranelift-nativeso my guess as to why this only showed up on my PR is that it just-so-happened to run on a machine with avx512 features enabled on its CPU. Assuming that's enabled I think it explains why this panics.This leads me to the actual issue here, the clif test in question specifically uses out-of-bounds indices for a shuffle argument:
function %shuffle_i8x16(i8x16, i8x16) -> i8x16 { block0(v0: i8x16, v1: i8x16): v2 = shuffle v0, v1, [3 0 31 26 100 6 12 11 23 13 24 4 2 15 17 5] return v2 }which looks like what's tripping the assert in the avx512 branch I think. I can reproduce this myself with a clif test that looks like:
test compile set enable_simd target x86_64 has_avx512vl has_avx512vbmi function %shuffle_i8x16(i8x16, i8x16) -> i8x16 { block0(v0: i8x16, v1: i8x16): v2 = shuffle v0, v1, [3 0 31 26 100 6 12 11 23 13 24 4 2 15 17 5] return v2 }this will fail with the same panic in
clif-util test foo.clifcc @abrown @jlb6740
Wasmtime GitHub notifications bot (Dec 01 2021 at 21:48):alexcrichton labeled issue #3581:
I ran into an odd issue today where https://github.com/bytecodealliance/wasmtime/pull/3579 failed on CI, but it didn't change anything to the point that nothing in theory should have failed, much less a clif codegen test. I'm going to copy the failure here because I'm about to rerun CI at which point I suspect the logs will get overwritten:
<details>
... thread 'worker #0' panicked at 'shuffle mask values must be between 0 and 31', cranelift/codegen/src/isa/x64/lower.rs:5519:21 stack backtrace: 0: std::panicking::begin_panic at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/std/src/panicking.rs:543:12 1: cranelift_codegen::isa::x64::lower::lower_insn_to_regs at ./codegen/src/isa/x64/lower.rs:5519:21 2: cranelift_codegen::isa::x64::lower::<impl cranelift_codegen::machinst::lower::LowerBackend for cranelift_codegen::isa::x64::X64Backend>::lower at ./codegen/src/isa/x64/lower.rs:6115:9 3: cranelift_codegen::machinst::lower::Lower<I>::lower_clif_block at ./codegen/src/machinst/lower.rs:750:17 4: cranelift_codegen::machinst::lower::Lower<I>::lower at ./codegen/src/machinst/lower.rs:991:17 5: cranelift_codegen::machinst::compile::compile at ./codegen/src/machinst/compile.rs:29:9 6: cranelift_codegen::isa::x64::X64Backend::compile_vcode at ./codegen/src/isa/x64/mod.rs:53:9 7: <cranelift_codegen::isa::x64::X64Backend as cranelift_codegen::machinst::MachBackend>::compile_function at ./codegen/src/isa/x64/mod.rs:64:21 8: cranelift_codegen::context::Context::compile at ./codegen/src/context.rs:172:22 9: cranelift_filetests::function_runner::compile at ./filetests/src/function_runner.rs:247:21 10: cranelift_filetests::function_runner::SingleFunctionCompiler::compile at ./filetests/src/function_runner.rs:76:25 11: <cranelift_filetests::test_run::TestRun as cranelift_filetests::subtest::SubTest>::run at ./filetests/src/test_run.rs:64:35 12: cranelift_filetests::runone::run_one_test at ./filetests/src/runone.rs:187:5 13: cranelift_filetests::runone::run at ./filetests/src/runone.rs:132:9 14: cranelift_filetests::concurrent::worker_thread::{{closure}}::{{closure}} at ./filetests/src/concurrent.rs:150:46 15: std::panicking::try::do_call at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/std/src/panicking.rs:403:40 16: __rust_try 17: std::panicking::try at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/std/src/panicking.rs:367:19 18: std::panic::catch_unwind at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/std/src/panic.rs:129:14 19: cranelift_filetests::concurrent::worker_thread::{{closure}} at ./filetests/src/concurrent.rs:150:30 note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace. ... FAIL filetests/filetests/runtests/simd-shuffle.clif: panicked in worker #0: shuffle mask values must be between 0 and 31 ... 342 tests thread 'filetests' panicked at 'test harness: 1 failure', cranelift/tests/filetests.rs:5:10 stack backtrace: 0: rust_begin_unwind at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/std/src/panicking.rs:517:5 1: core::panicking::panic_fmt at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/core/src/panicking.rs:101:14 2: core::result::unwrap_failed at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/core/src/result.rs:1617:5 3: core::result::Result<T,E>::expect at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/core/src/result.rs:1259:23 4: filetests::filetests at ./tests/filetests.rs:4:5 5: filetests::filetests::{{closure}} at ./tests/filetests.rs:2:1 6: core::ops::function::FnOnce::call_once at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/core/src/ops/function.rs:227:5 7: core::ops::function::FnOnce::call_once at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/core/src/ops/function.rs:227:5 note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace. failures: filetests test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out; finished in 9.75s ...</details>
or, effectively:
... thread 'worker #0' panicked at 'shuffle mask values must be between 0 and 31', cranelift/codegen/src/isa/x64/lower.rs:5519:21 ... FAIL filetests/filetests/runtests/simd-shuffle.clif: panicked in worker #0: shuffle mask values must be between 0 and 31 ...This looks like a "spurious" error in that it doesn't reproduce on other PRs on CI. The code in question, though:
is specifically gated for avx512 features, and cranelift filetests do indeed use
cranelift-nativeso my guess as to why this only showed up on my PR is that it just-so-happened to run on a machine with avx512 features enabled on its CPU. Assuming that's enabled I think it explains why this panics.This leads me to the actual issue here, the clif test in question specifically uses out-of-bounds indices for a shuffle argument:
function %shuffle_i8x16(i8x16, i8x16) -> i8x16 { block0(v0: i8x16, v1: i8x16): v2 = shuffle v0, v1, [3 0 31 26 100 6 12 11 23 13 24 4 2 15 17 5] return v2 }which looks like what's tripping the assert in the avx512 branch I think. I can reproduce this myself with a clif test that looks like:
test compile set enable_simd target x86_64 has_avx512vl has_avx512vbmi function %shuffle_i8x16(i8x16, i8x16) -> i8x16 { block0(v0: i8x16, v1: i8x16): v2 = shuffle v0, v1, [3 0 31 26 100 6 12 11 23 13 24 4 2 15 17 5] return v2 }this will fail with the same panic in
clif-util test foo.clifcc @abrown @jlb6740
Wasmtime GitHub notifications bot (Dec 01 2021 at 23:52):abrown commented on issue #3581:
Looks like that assertion was written under the assumption that the verifier would remove invalid lane indices: https://github.com/bytecodealliance/wasmtime/blob/main/cranelift/codegen/src/verifier/mod.rs#L56-L57.
Wasmtime GitHub notifications bot (Dec 01 2021 at 23:52):abrown edited a comment on issue #3581:
Looks like that assertion was written under the assumption that the verifier would remove invalid lane indices:
https://github.com/bytecodealliance/wasmtime/blob/main/cranelift/codegen/src/verifier/mod.rs#L56-L57.
Wasmtime GitHub notifications bot (Dec 01 2021 at 23:52):abrown edited a comment on issue #3581:
Looks like that assertion was written under the assumption that the verifier would remove invalid lane indices:
https://github.com/bytecodealliance/wasmtime/blob/main/cranelift/codegen/src/verifier/mod.rs#L56-L57
Wasmtime GitHub notifications bot (Dec 01 2021 at 23:53):abrown edited a comment on issue #3581:
Looks like that assertion was written under the assumption that the verifier would remove invalid lane indices:
Wasmtime GitHub notifications bot (Dec 02 2021 at 23:06):fitzgen commented on issue #3581:
That is something that the verifier checks/asserts, not something that it will make true for you. The verifier is only used in ~debug builds. (If you already know this, apologies, it isn't clear to me based on your comment.)
Wasmtime GitHub notifications bot (Dec 03 2021 at 00:22):abrown commented on issue #3581:
Yup. I may have not been aware of that when that code was originally written. What is worse is that I don't think the verifier actually does that check anyways--or at least not that I can find--so there also might be some documentation that needs to be changed here.
Wasmtime GitHub notifications bot (Dec 16 2021 at 18:36):abrown closed issue #3581:
I ran into an odd issue today where https://github.com/bytecodealliance/wasmtime/pull/3579 failed on CI, but it didn't change anything to the point that nothing in theory should have failed, much less a clif codegen test. I'm going to copy the failure here because I'm about to rerun CI at which point I suspect the logs will get overwritten:
<details>
... thread 'worker #0' panicked at 'shuffle mask values must be between 0 and 31', cranelift/codegen/src/isa/x64/lower.rs:5519:21 stack backtrace: 0: std::panicking::begin_panic at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/std/src/panicking.rs:543:12 1: cranelift_codegen::isa::x64::lower::lower_insn_to_regs at ./codegen/src/isa/x64/lower.rs:5519:21 2: cranelift_codegen::isa::x64::lower::<impl cranelift_codegen::machinst::lower::LowerBackend for cranelift_codegen::isa::x64::X64Backend>::lower at ./codegen/src/isa/x64/lower.rs:6115:9 3: cranelift_codegen::machinst::lower::Lower<I>::lower_clif_block at ./codegen/src/machinst/lower.rs:750:17 4: cranelift_codegen::machinst::lower::Lower<I>::lower at ./codegen/src/machinst/lower.rs:991:17 5: cranelift_codegen::machinst::compile::compile at ./codegen/src/machinst/compile.rs:29:9 6: cranelift_codegen::isa::x64::X64Backend::compile_vcode at ./codegen/src/isa/x64/mod.rs:53:9 7: <cranelift_codegen::isa::x64::X64Backend as cranelift_codegen::machinst::MachBackend>::compile_function at ./codegen/src/isa/x64/mod.rs:64:21 8: cranelift_codegen::context::Context::compile at ./codegen/src/context.rs:172:22 9: cranelift_filetests::function_runner::compile at ./filetests/src/function_runner.rs:247:21 10: cranelift_filetests::function_runner::SingleFunctionCompiler::compile at ./filetests/src/function_runner.rs:76:25 11: <cranelift_filetests::test_run::TestRun as cranelift_filetests::subtest::SubTest>::run at ./filetests/src/test_run.rs:64:35 12: cranelift_filetests::runone::run_one_test at ./filetests/src/runone.rs:187:5 13: cranelift_filetests::runone::run at ./filetests/src/runone.rs:132:9 14: cranelift_filetests::concurrent::worker_thread::{{closure}}::{{closure}} at ./filetests/src/concurrent.rs:150:46 15: std::panicking::try::do_call at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/std/src/panicking.rs:403:40 16: __rust_try 17: std::panicking::try at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/std/src/panicking.rs:367:19 18: std::panic::catch_unwind at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/std/src/panic.rs:129:14 19: cranelift_filetests::concurrent::worker_thread::{{closure}} at ./filetests/src/concurrent.rs:150:30 note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace. ... FAIL filetests/filetests/runtests/simd-shuffle.clif: panicked in worker #0: shuffle mask values must be between 0 and 31 ... 342 tests thread 'filetests' panicked at 'test harness: 1 failure', cranelift/tests/filetests.rs:5:10 stack backtrace: 0: rust_begin_unwind at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/std/src/panicking.rs:517:5 1: core::panicking::panic_fmt at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/core/src/panicking.rs:101:14 2: core::result::unwrap_failed at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/core/src/result.rs:1617:5 3: core::result::Result<T,E>::expect at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/core/src/result.rs:1259:23 4: filetests::filetests at ./tests/filetests.rs:4:5 5: filetests::filetests::{{closure}} at ./tests/filetests.rs:2:1 6: core::ops::function::FnOnce::call_once at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/core/src/ops/function.rs:227:5 7: core::ops::function::FnOnce::call_once at /rustc/59eed8a2aac0230a8b53e89d4e99d55912ba6b35/library/core/src/ops/function.rs:227:5 note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace. failures: filetests test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out; finished in 9.75s ...</details>
or, effectively:
... thread 'worker #0' panicked at 'shuffle mask values must be between 0 and 31', cranelift/codegen/src/isa/x64/lower.rs:5519:21 ... FAIL filetests/filetests/runtests/simd-shuffle.clif: panicked in worker #0: shuffle mask values must be between 0 and 31 ...This looks like a "spurious" error in that it doesn't reproduce on other PRs on CI. The code in question, though:
is specifically gated for avx512 features, and cranelift filetests do indeed use
cranelift-nativeso my guess as to why this only showed up on my PR is that it just-so-happened to run on a machine with avx512 features enabled on its CPU. Assuming that's enabled I think it explains why this panics.This leads me to the actual issue here, the clif test in question specifically uses out-of-bounds indices for a shuffle argument:
function %shuffle_i8x16(i8x16, i8x16) -> i8x16 { block0(v0: i8x16, v1: i8x16): v2 = shuffle v0, v1, [3 0 31 26 100 6 12 11 23 13 24 4 2 15 17 5] return v2 }which looks like what's tripping the assert in the avx512 branch I think. I can reproduce this myself with a clif test that looks like:
test compile set enable_simd target x86_64 has_avx512vl has_avx512vbmi function %shuffle_i8x16(i8x16, i8x16) -> i8x16 { block0(v0: i8x16, v1: i8x16): v2 = shuffle v0, v1, [3 0 31 26 100 6 12 11 23 13 24 4 2 15 17 5] return v2 }this will fail with the same panic in
clif-util test foo.clifcc @abrown @jlb6740
Last updated: Oct 23 2024 at 20:03 UTC