candymate opened issue #6786:
Test Case
Any cases with v128.loadXXX or v128.storeXXX would work
One example
(module (type (;0;) (func (param i32 v128))) (func (;0;) (type 0) (param i32 v128) local.get 0 local.get 1 v128.store16_lane align=1 0 ) (export "main" (func 0)) )
Steps to Reproduce
Compiling any cases containing v128.loadXXX or v128.storeXXX leads to reaching unreachable code.
Such compilations lead to reaching the following codes// cranelift/codegen/src/isa/riscv64/inst/args.rs:1289 pub(crate) fn from_type(t: Type) -> Self { if t.is_float() { return if t == F32 { Self::Flw } else { Self::Fld }; } match t { R32 => Self::Lwu, R64 | I64 => Self::Ld, I8 => Self::Lb, I16 => Self::Lh, I32 => Self::Lw, _ => unreachable!(), // reaches here } }
// cranelift/codegen/src/isa/riscv64/inst/args.rs:1338 pub(crate) fn from_type(t: Type) -> Self { if t.is_float() { return if t == F32 { Self::Fsw } else { Self::Fsd }; } match t.bits() { 1 | 8 => Self::Sb, 16 => Self::Sh, 32 => Self::Sw, 64 => Self::Sd, _ => unreachable!(), // reaches here } }
Steps are the following:
git submodule update --init
cargo run --release --target=riscv64gc-unknown-linux-gnu test.wasm
Expected Results
Successful compilation
Actual Results
Panics with 'internal error: entered unreachable code'
Versions and Environment
Wasmtime version or commit: v11.0.1 Release
Operating system: Ubuntu 20.04.6
Architecture: RISC-V64
candymate added the bug label to Issue #6786.
cfallin commented on issue #6786:
SIMD support on RISC-V is still a work-in-progress; @afonso360 was contributing code a few opcodes at a time recently. So it's expected that this doesn't yet work, though contributions are welcome if you're interested in tackling this!
afonso360 commented on issue #6786:
:wave: Hey,
This code should work, but SIMD support only works on machines with Vector extensions.
Are you running this natively on a RISC-V machine? Or just using QEMU?
If you are running on QEMU you can add the following parameter to make
-cpu rv64,v=true,vlen=128,vext_spec=v1.0
this compile and run.If you are running on an actual RISC-V machine, please make sure it supports version 1.0 of the Vector Extension. I haven't tested with 0.7.1 (popular in a lot of current CPU's) but I doubt it would work since they are somewhat incompatible.
That being said, we should probably have a better error message for this stuff.
afonso360 edited a comment on issue #6786:
:wave: Hey,
This code should work, but SIMD support only works on machines with Vector extensions.
Are you running this natively on a RISC-V machine? Or just using QEMU?
If you are running on QEMU you can add the following parameter
-cpu rv64,v=true,vlen=128,vext_spec=v1.0
to make this compile and run.If you are running on an actual RISC-V machine, please make sure it supports version 1.0 of the Vector Extension. I haven't tested with 0.7.1 (popular in a lot of current CPU's) but I doubt it would work since they are somewhat incompatible.
That being said, we should probably have a better error message for this stuff.
afonso360 added the cranelift:area:riscv64 label to Issue #6786.
afonso360 edited a comment on issue #6786:
:wave: Hey,
This code should work, but SIMD support only works on machines with Vector extensions.
Are you running this natively on a RISC-V machine, or just using QEMU?
If you are running on QEMU you can add the following parameter
-cpu rv64,v=true,vlen=128,vext_spec=v1.0
to make this compile and run.If you are running on an actual RISC-V machine, please make sure it supports version 1.0 of the Vector Extension. I haven't tested with 0.7.1 (popular in a lot of current CPU's) but I doubt it would work since they are somewhat incompatible.
That being said, we should probably have a better error message for this stuff.
afonso360 edited a comment on issue #6786:
:wave: Hey,
This code should work, but SIMD support only works on machines with Vector extensions.
Are you running this natively on a RISC-V machine, or just using QEMU?
If you are running on QEMU you can add the following parameter
-cpu rv64,v=true,vlen=128,vext_spec=v1.0
to the QEMU arguments to make this compile and run.If you are running on an actual RISC-V machine, please make sure it supports version 1.0 of the Vector Extension. I haven't tested with 0.7.1 (popular in a lot of current CPU's) but I doubt it would work since they are somewhat incompatible.
That being said, we should probably have a better error message for this stuff.
afonso360 edited a comment on issue #6786:
:wave: Hey,
This code should work, but SIMD support only works on machines with Vector extensions.
Are you running this natively on a RISC-V machine, or just using QEMU?
If you are running on QEMU you can add the following parameter
-cpu rv64,v=true,vlen=128,vext_spec=v1.0
to the QEMU arguments to make this compile and run.If you are running on an actual RISC-V machine, please make sure it supports version 1.0 of the Vector Extension and that this is reported in
/proc/cpuinfo
. I haven't tested with 0.7.1 (popular in a lot of current CPU's) but I doubt it would work since they are somewhat incompatible.That being said, we should probably have a better error message for this stuff.
candymate commented on issue #6786:
Hi,
I'm currently running QEMU, so I added the parameter you gave me. But, it doesn't change anything.
In ~/.cargo/config.toml, I have the following:
[target.riscv64gc-unknown-linux-gnu] linker = "riscv64-linux-gnu-gcc" runner = "qemu-riscv64 -cpu rv64,v=true,vlen=128,vext_spec=v1.0 -L /usr/riscv64-linux-gnu -E LD_LIBRARY_PATH=/usr/riscv64-linux-gnu/lib -E WASMTIME_TEST_NO_HOG_MEMORY=1"
Just in case, I upgraded Ubuntu to 22.04, and built
qemu-riscv64
from the latest version of riscv-gnu-toolchain. (https://github.com/riscv-collab/riscv-gnu-toolchain)Versions I'm using are the following:
$ qemu-riscv64 --version qemu-riscv64 version 7.1.0 (v7.1.0) $ riscv64-linux-gnu-gcc --version riscv64-linux-gnu-gcc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Still leads to the following error message. (Ran with a simple wrapper to provide Wasm memory)
Running `qemu-riscv64 -cpu rv64,v=true,vlen=128,vext_spec=v1.0 -L /usr/riscv64-linux-gnu -E LD_LIBRARY_PATH=/usr/riscv64-linux-gnu/lib -E WASMTIME_TEST_NO_HOG_MEMORY=1 target/riscv64gc-unknown-linux-gnu/debug/wasmtime-wrapper input` thread '<unnamed>' panicked at 'internal error: entered unreachable code', /home/candymate/.cargo/registry/src/index.crates.io-6f17d22bba15001f/cranelift-codegen-0.98.1/src/isa/riscv64/inst/args.rs:1347:18 note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
It would be helpful if you tell me what I'm missing.
candymate edited a comment on issue #6786:
Hi,
I'm currently running QEMU, so I added the parameter you gave me. But, it doesn't change anything.
In ~/.cargo/config.toml, I have the following:
[target.riscv64gc-unknown-linux-gnu] linker = "riscv64-linux-gnu-gcc" runner = "qemu-riscv64 -cpu rv64,v=true,vlen=128,vext_spec=v1.0 -L /usr/riscv64-linux-gnu -E LD_LIBRARY_PATH=/usr/riscv64-linux-gnu/lib -E WASMTIME_TEST_NO_HOG_MEMORY=1"
Just in case, I upgraded Ubuntu to 22.04, and built
qemu-riscv64
from the latest version of riscv-gnu-toolchain. (https://github.com/riscv-collab/riscv-gnu-toolchain)Versions I'm using are the following:
$ rustc --version rustc 1.71.0-nightly (ce5919fce 2023-05-15) $ qemu-riscv64 --version qemu-riscv64 version 7.1.0 (v7.1.0) $ riscv64-linux-gnu-gcc --version riscv64-linux-gnu-gcc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Still leads to the following error message. (Ran with a simple wrapper to provide Wasm memory)
Running `qemu-riscv64 -cpu rv64,v=true,vlen=128,vext_spec=v1.0 -L /usr/riscv64-linux-gnu -E LD_LIBRARY_PATH=/usr/riscv64-linux-gnu/lib -E WASMTIME_TEST_NO_HOG_MEMORY=1 target/riscv64gc-unknown-linux-gnu/debug/wasmtime-wrapper input` thread '<unnamed>' panicked at 'internal error: entered unreachable code', /home/candymate/.cargo/registry/src/index.crates.io-6f17d22bba15001f/cranelift-codegen-0.98.1/src/isa/riscv64/inst/args.rs:1347:18 note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
It would be helpful if you tell me what I'm missing.
candymate edited a comment on issue #6786:
Hi,
I'm currently running QEMU, so I added the parameter you gave me. But, it doesn't change anything.
In ~/.cargo/config.toml, I have the following:
[target.riscv64gc-unknown-linux-gnu] linker = "riscv64-linux-gnu-gcc" runner = "qemu-riscv64 -cpu rv64,v=true,vlen=128,vext_spec=v1.0 -L /usr/riscv64-linux-gnu -E LD_LIBRARY_PATH=/usr/riscv64-linux-gnu/lib -E WASMTIME_TEST_NO_HOG_MEMORY=1"
Just in case, I upgraded Ubuntu to 22.04, and built
qemu-riscv64
from the latest version of riscv-gnu-toolchain. (https://github.com/riscv-collab/riscv-gnu-toolchain)Versions I'm using are the following:
$ cargo --version cargo 1.71.0-nightly (13413c64f 2023-05-10) $ rustc --version rustc 1.71.0-nightly (ce5919fce 2023-05-15) $ qemu-riscv64 --version qemu-riscv64 version 7.1.0 (v7.1.0) $ riscv64-linux-gnu-gcc --version riscv64-linux-gnu-gcc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Still leads to the following error message. (Ran with a simple wrapper to provide Wasm memory)
Running `qemu-riscv64 -cpu rv64,v=true,vlen=128,vext_spec=v1.0 -L /usr/riscv64-linux-gnu -E LD_LIBRARY_PATH=/usr/riscv64-linux-gnu/lib -E WASMTIME_TEST_NO_HOG_MEMORY=1 target/riscv64gc-unknown-linux-gnu/debug/wasmtime-wrapper input` thread '<unnamed>' panicked at 'internal error: entered unreachable code', /home/candymate/.cargo/registry/src/index.crates.io-6f17d22bba15001f/cranelift-codegen-0.98.1/src/isa/riscv64/inst/args.rs:1347:18 note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
It would be helpful if you tell me what I'm missing.
afonso360 commented on issue #6786:
Oh! Right, QEMU only gained the ability to emulate
/proc/cpuinfo
in version 8.1, we use that to detect the features of the processor and automatically enable them.To manually force it on, you can use the wasmtime flag
--cranelift-enable has_v
, which should forcefully enable them on our side.
candymate commented on issue #6786:
Adding the
has_v
flag solved the problem. Thanks!
candymate closed issue #6786:
Test Case
Any cases with v128.loadXXX or v128.storeXXX would work
One example
(module (type (;0;) (func (param i32 v128))) (func (;0;) (type 0) (param i32 v128) local.get 0 local.get 1 v128.store16_lane align=1 0 ) (export "main" (func 0)) )
Steps to Reproduce
Compiling any cases containing v128.loadXXX or v128.storeXXX leads to reaching unreachable code.
Such compilations lead to reaching the following codes// cranelift/codegen/src/isa/riscv64/inst/args.rs:1289 pub(crate) fn from_type(t: Type) -> Self { if t.is_float() { return if t == F32 { Self::Flw } else { Self::Fld }; } match t { R32 => Self::Lwu, R64 | I64 => Self::Ld, I8 => Self::Lb, I16 => Self::Lh, I32 => Self::Lw, _ => unreachable!(), // reaches here } }
// cranelift/codegen/src/isa/riscv64/inst/args.rs:1338 pub(crate) fn from_type(t: Type) -> Self { if t.is_float() { return if t == F32 { Self::Fsw } else { Self::Fsd }; } match t.bits() { 1 | 8 => Self::Sb, 16 => Self::Sh, 32 => Self::Sw, 64 => Self::Sd, _ => unreachable!(), // reaches here } }
Steps are the following:
git submodule update --init
cargo run --release --target=riscv64gc-unknown-linux-gnu test.wasm
Expected Results
Successful compilation
Actual Results
Panics with 'internal error: entered unreachable code'
Versions and Environment
Wasmtime version or commit: v11.0.1 Release
Operating system: Ubuntu 20.04.6
Architecture: RISC-V64
Last updated: Nov 22 2024 at 16:03 UTC