Differentiate dsyevr_64_

[Roger-luo]

I'm not sure why it complains the return type is Union{}, the return type of this foo function is quite simple? I got

julia> foo(x) = sum(exp(fill(x, 10, 10)))
foo (generic function with 1 method)

julia> Enzyme.autodiff(foo, Active(2.0))
ERROR: return type is Union{}, giving up.
Stacktrace:
 [1] error(s::String)
   @ Base ./error.jl:33
 [2] enzyme!(job::GPUCompiler.CompilerJob{Enzyme.Compiler.EnzymeTarget, Enzyme.Compiler.EnzymeCompilerParams, GPUCompiler.FunctionSpec{typeof(Cassette.overdub), Tuple{Cassette.Context{nametype(EnzymeCtx), Nothing, Nothing, Enzyme.Compiler.var"##PassType#257", Nothing, Cassette.DisableHooks}, typeof(foo), Float64}}}, mod::LLVM.Module, primalf::LLVM.Function, adjoint::GPUCompiler.FunctionSpec{typeof(foo), Tuple{Active{Float64}}}, split::Bool, parallel::Bool)
   @ Enzyme.Compiler ~/.julia/packages/Enzyme/A97If/src/compiler.jl:143
 [3] codegen(output::Symbol, job::GPUCompiler.CompilerJob{Enzyme.Compiler.EnzymeTarget, Enzyme.Compiler.EnzymeCompilerParams, GPUCompiler.FunctionSpec{typeof(Cassette.overdub), Tuple{Cassette.Context{nametype(EnzymeCtx), Nothing, Nothing, Enzyme.Compiler.var"##PassType#257", Nothing, Cassette.DisableHooks}, typeof(foo), Float64}}}; libraries::Bool, deferred_codegen::Bool, optimize::Bool, strip::Bool, validate::Bool, only_entry::Bool, parent_job::Nothing)
   @ Enzyme.Compiler ~/.julia/packages/Enzyme/A97If/src/compiler.jl:435
 [4] _thunk(job::GPUCompiler.CompilerJob{Enzyme.Compiler.EnzymeTarget, Enzyme.Compiler.EnzymeCompilerParams, GPUCompiler.FunctionSpec{typeof(Cassette.overdub), Tuple{Cassette.Context{nametype(EnzymeCtx), Nothing, Nothing, Enzyme.Compiler.var"##PassType#257", Nothing, Cassette.DisableHooks}, typeof(foo), Float64}}})
   @ Enzyme.Compiler ~/.julia/packages/Enzyme/A97If/src/compiler.jl:761
 [5] callback(orc_ref::Ptr{LLVM.API.LLVMOrcOpaqueJITStack}, callback_ctx::Ptr{Nothing})
   @ Enzyme.Compiler ~/.julia/packages/Enzyme/A97If/src/compiler.jl:822
 [6] top-level scope
   @ REPL[11]:1

julia> @code_warntype foo(2.0)
Variables
  #self#::Core.Const(foo)
  x::Float64

Body::Float64
1 ─ %1 = Main.fill(x, 10, 10)::Matrix{Float64}
│   %2 = Main.exp(%1)::Matrix{Float64}
│   %3 = Main.sum(%2)::Float64
└──      return %3

[wsmoses]

What happens if you do autodiff_no_casette?

[Roger-luo]

It crash with like over 1k line error for autodiff_no_cassette (this is on another machine on MacOS, I can try it on the Linux machine if needed)

784768, i64 -6940216133781273352, i64 -6832105780228080109, i64 -5841289834741351288, i64 -8696547905883959080, i64 -5989860371902311149, i64 7129149566058721777, i64 8408196149339104217]: {[-1]:Pointer}, intvals: {}
  %13 = getelementptr inbounds [256 x i64], [256 x i64]* @_j_const5, i64 0, i64 %12, !dbg !165: {[-1]:Pointer}, intvals: {}
  %merge = phi double [ 0x7FF0000000000000, %L44 ], [ 0.000000e+00, %L47 ]: {[-1]:Float@double}, intvals: {}
  %35 = icmp sgt i32 %5, -13313, !dbg !211: {[-1]:Integer}, intvals: {}
  %10 = fsub contract double %8, %9, !dbg !155: {[-1]:Float@double}, intvals: {}
  %8 = fsub contract double %0, %7, !dbg !153: {[-1]:Float@double}, intvals: {}
</analysis>
Illegal updateAnalysis prev:{[-1]:Float@double} new: {[-1]:Integer}
val:   %5 = trunc i64 %4 to i32, !dbg !147 origin=  %35 = icmp sgt i32 %5, -13313, !dbg !211
Assertion failed: (0 && "Performed illegal updateAnalysis"), function updateAnalysis, file /workspace/srcdir/Enzyme/enzyme/Enzyme/TypeAnalysis/TypeAnalysis.cpp, line 511.

signal (6): Abort trap: 6
in expression starting at REPL[3]:1
__pthread_kill at /usr/lib/system/libsystem_kernel.dylib (unknown line)
Allocations: 35817871 (Pool: 35799479; Big: 18392); GC: 38
[1]    19138 abort      julia

[wsmoses]

Yeah this is primarily a cassette bug and unfortunately doesn't appear to have a good fix without cassette fixing the bug on their side. The with cassette version fails per that bug. The version without cassette means differentiating the bithack internals of exp (which we can, but shouldn't do)

[wsmoses]

Having since worked around cassette this now reached the point where it cannot be differentiated due to the following blas function not having a gradient defined:

declare void @dsyevr_64_(i8*, i8*, i8*, i8*, i64, i8*, i8*, i8*, i8*, i8*, i8*, i64, i64, i64, i8*, i64, i64, i8*, i64, i8*, i64, i64, i64, i64) local_unnamed_addr

@reikdas perhaps one to consider after ddot goes through.

[vchuravy]

bump. Does Enzyme now support dsyevr?