This is likely a limitation on the IR we can currently process. We use some
of the PNaCl lowering passes to simplify LLVM IR, then run on that, which
works for basically everything clang emits, but some things clang doesn't
emit, might not work. This looks like that type of bug.
You can grep for that string ("Unrecognized struct value") in the fastcomp
llvm repo, to see where it happens. Likely in lib/Transform/NaCl somewhere.
- Alon
On Wed, Aug 27, 2014 at 1:45 PM, Tom Short <[email protected]> wrote:
>
> The Julia language [1] uses LLVM's JIT to dynamically create
> code. It's closer to being able to generate static code [2],
> including (I think) LLVM bitcode.
>
> As a trial, I used Julia to generate a large bitcode
> file (sys.bc) for much of Julia's standard set of functions. For
> simple functions that use standard LLVM types, a recent
> Emscripten (fastcomp) generates good JavaScript. For more complex
> functions that use other types, I'm getting a variety of errors.
> Here is an example of a common error:
>
> Value: %Complex.16 %0
> LLVM ERROR: Unrecognized struct value
> Traceback (most recent call last):
> File "/usr/bin/emcc", line 1200, in <module>
>
> The disassembled code for the function that triggered this error
> looks like this:
>
>
> define %Complex.15 @"julia_sin;108403"(%Complex.15) {
> top:
> %1 = load %jl_value_t** @"*Main.Core.Float3267512", align 8
> %2 = getelementptr inbounds %jl_value_t* %1, i64 1, i32 0
> %3 = load %jl_value_t** %2, align 8
> %4 = extractvalue %Complex.15 %0, 0, !julia_type !98
> %5 = call float @"julia_convert;85919"(%jl_value_t* %3, i16 %4)
> %6 = load %jl_value_t** @"*Main.Core.Float3267512", align 8
> %7 = getelementptr inbounds %jl_value_t* %6, i64 1, i32 0
> %8 = load %jl_value_t** %7, align 8
> %9 = extractvalue %Complex.15 %0, 1, !julia_type !98
> %10 = call float @"julia_convert;85919"(%jl_value_t* %8, i16 %9)
> %11 = insertvalue %Complex.16 undef, float %5, 0
> %12 = insertvalue %Complex.16 %11, float %10, 1, !julia_type !103
> %13 = call %Complex.16 @"julia_sin;89568"(%Complex.16 %12), !julia_type
> !103
> %14 = load %jl_value_t** @"*Main.Core.Float1667510", align 8
> %15 = getelementptr inbounds %jl_value_t* %14, i64 1, i32 0
> %16 = load %jl_value_t** %15, align 8
> %17 = extractvalue %Complex.16 %13, 0
> %18 = call i16 @"julia_convert;85922"(%jl_value_t* %16, float %17),
> !julia_type !98
> %19 = load %jl_value_t** @"*Main.Core.Float1667510", align 8
> %20 = getelementptr inbounds %jl_value_t* %19, i64 1, i32 0
> %21 = load %jl_value_t** %20, align 8
> %22 = extractvalue %Complex.16 %13, 1
> %23 = call i16 @"julia_convert;85922"(%jl_value_t* %21, float %22),
> !julia_type !98
> %24 = insertvalue %Complex.15 undef, i16 %18, 0
> %25 = insertvalue %Complex.15 %24, i16 %23, 1, !julia_type !104
> ret %Complex.15 %25
> }
>
> The type mentioned in the error was defined earlier as:
>
> %Complex.16 = type { float, float }
>
> My main question is: where do I start to try to debug LLVM errors
> like this? Are there LLVM references that'd help? Any way to use
> Clang or other LLVM tool to try to narrow it down or test this?
> The .bc code from Julia could well be wrong or be missing code,
> but I don't really know where to start here.
>
>
>
> [1] http://julialang.org
> [2] https://groups.google.com/d/msg/julia-dev/qdnggTuIp9s/BoQSNGNXxV0J
>
> --
> You received this message because you are subscribed to the Google Groups
> "emscripten-discuss" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to [email protected].
> For more options, visit https://groups.google.com/d/optout.
>
--
You received this message because you are subscribed to the Google Groups
"emscripten-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
