Issues with Wishart from Distributions.jl

[torfjelde]

using Distributions, Enzyme

let dist = Wishart(7, [1 0.5; 0.5 1]), x = rand(dist), ∂f∂x = zero(x)
    _, y = Enzyme.autodiff(Enzyme.ReverseWithPrimal, logpdf, Enzyme.Active,
                           Enzyme.Const(dist), Enzyme.Duplicated(x, ∂f∂x))
    y, ∂f∂x
end

fails with a stacktrace that is too long to paste, but here's the beginning of it:

LLVM error: broken gc calling conv fix
Range types must match instruction type!
  call fastcc void @augmented_julia_chkuplofinite_16058({} addrspace(10)* nocapture noundef nonnull readonly align 16 dereferenceable(40) %0), !dbg !700, !range !166

; ModuleID = 'start'
source_filename = "start"
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128-ni:10:11:12:13"
target triple = "x86_64-linux-gnu"

@_j_str5 = private unnamed_addr constant [11 x i8] c"typeassert\00"
@_j_const7 = private unnamed_addr constant [129 x [2 x double]] [[2 x double] zeroinitializer, [2 x double] [double 0x3F7FE02A6B200000, double 0xBD6F30EE07912DF9], [2 x double] [double 0x3F8FC0A8B1000000, double 0xBD5FE0E183092C59], [2 x double] [double 0x3F97B91B07D80000, double 0xBD62772AB6C0559C], [2 x double] [double 0x3F9F829B0E780000, double 0x3D2980267C7E09E4], [2 x double] [double 0x3FA39E87BA000000, double 0xBD642A056FEA4DFD], [2 x double] [double 0x3FA77458F6340000, double 0xBD62303B9CB0D5E1], [2 x double] [double 0x3FAB42DD71180000, double 0x3D671BEC28D14C7E], [2 x double] [double 0x3FAF0A30C0100000, double 0x3D662A6617CC9717], [2 x double] [double 0x3FB16536EEA40000, double 0xBD60A3E2F3B47D18], [2 x double] [double 0x3FB341D7961C0000, double 0xBD4717B6B33E44F8], [2 x double] [double 0x3FB51B073F060000, double 0x3D383F69278E686A], [2 x double] [double 0x3FB6F0D28AE60000, double 0xBD62968C836CC8C2], [2 x double] [double 0x3FB8C345D6320000, double 0xBD5937C294D2F567], [2 x double] [double 0x3FBA926D3A4A0000, double 0x3D6AAC6CA17A4554], [2 x double] [double 0x3FBC5E548F5C0000, double 0xBD4C5E7514F4083F], [2 x double] [double 0x3FBE27076E2A0000, double 0x3D6E5CBD3D50FFFC], [2 x double] [double 0x3FBFEC9131DC0000, double 0xBD354555D1AE6607], [2 x double] [double 0x3FC0D77E7CD10000, double 0xBD6C69A65A23A170], [2 x double] [double 0x3FC1B72AD52F0000, double 0x3D69E80A41811A39], [2 x double] [double 0x3FC29552F8200000, double 0xBD35B967F4471DFC], [2 x double] [double 0x3FC371FC201F0000, double 0xBD6C22F10C9A4EA8], [2 x double] [double 0x3FC44D2B6CCB0000, double 0x3D6F4799F4F6543E], [2 x double] [double 0x3FC526E5E3A20000, double 0xBD62F21746FF8A47], [2 x double] [double 0x3FC5FF3070A80000, double 0xBD6B0B0DE3077D7E], [2 x double] [double 0x3FC6D60FE71A0000, double 0xBD56F1B955C4D1DA], [2 x double] [double 0x3FC7AB8902110000, double 0xBD537B720E4A694B], [2 x double] [double 0x3FC87FA065210000, double 0xBD5B77B7EFFB7F41], [2 x double] [double 0x3FC9525A9CF40000, double 0x3D65AD1D904C1D4E], [2 x double] [double 0x3FCA23BC1FE30000, double 0xBD62A739B23B93E1], [2 x double] [double 0x3FCAF3C94E810000, double 0xBD600349CC67F9B2], [2 x double] [double 0x3FCBC286742E0000, double 0xBD6CCA75818C5DBC], [2 x double] [double 0x3FCC8FF7C79B0000, double 0xBD697794F689F843], [2 x double] [double 0x3FCD5C216B500000, double 0xBD611BA91BBCA682], [2 x double] [double 0x3FCE27076E2B0000, double 0xBD3A342C2AF0003C], [2 x double] [double 0x3FCEF0ADCBDC0000, double 0x3D664D948637950E], [2 x double] [double 0x3FCFB9186D5E0000, double 0x3D5F1546AAA3361C], [2 x double] [double 0x3FD0402594B50000, double 0xBD67DF928EC217A5], [2 x double] [double 0x3FD0A324E2738000, double 0x3D50E35F73F7A018], [2 x double] [double 0x3FD1058BF9AE8000, double 0xBD6A9573B02FAA5A], [2 x double] [double 0x3FD1675CABAB8000, double 0x3D630701CE63EAB9], [2 x double] [double 0x3FD1C898C1698000, double 0x3D59FAFBC68E7540], [2 x double] [double 0x3FD22941FBCF8000, double 0xBD3A6976F5EB0963], [2 x double] [double 0x3FD2895A13DE8000, double 0x3D3A8D7AD24C13F0], [2 x double] [double 0x3FD2E8E2BAE10000, double 0x3D5D309C2CC91A85], [2 x double] [double 0x3FD347DD9A988000, double 0xBD25594DD4C58092], [2 x double] [double 0x3FD3A64C55698000, double 0xBD6D0B1C68651946], [2 x double] [double 0x3FD4043086868000, double 0x3D63F1DE86093EFA], [2 x double] [double 0x3FD4618BC21C8000, double 0xBD609EC17A426426], [2 x double] [double 0x3FD4BE5F95778000, double 0xBD3D7C92CD9AD824], [2 x double] [double 0x3FD51AAD872E0000, double 0xBD3F4BD8DB0A7CC1], [2 x double] [double 0x3FD5767717458000, double 0xBD62C9D5B2A49AF9], [2 x double] [double 0x3FD5D1BDBF580000, double 0x3D4394A11B1C1EE4], [2 x double] [double 0x3FD62C82F2BA0000, double 0xBD6C356848506EAD], [2 x double] [double 0x3FD686C81E9B0000, double 0x3D54AEC442BE1015], [2 x double] [double 0x3FD6E08EAA2B8000, double 0x3D60F1C609C98C6C], [2 x double] [double 0x3FD739D7F6BC0000, double 0xBD67FCB18ED9D603], [2 x double] [double 0x3FD792A55FDD8000, double 0xBD6C2EC1F512DC03], [2 x double] [double 0x3FD7EAF83B828000, double 0x3D67E1B259D2F3DA], [2 x double] [double 0x3FD842D1DA1E8000, double 0x3D462E927628CBC2], [2 x double] [double 0x3FD89A3386C18000, double 0xBD6ED2A52C73BF78], [2 x double] [double 0x3FD8F11E87368000, double 0xBD5D3881E8962A96], [2 x double] [double 0x3FD947941C210000, double 0x3D56FABA4CDD147D], [2 x double] [double 0x3FD99D9581180000, double 0xBD5F753456D113B8], [2 x double] [double 0x3FD9F323ECBF8000, double 0x3D584BF2B68D766F], [2 x double] [double 0x3FDA484090E58000, double 0x3D6D8515FE535B87], [2 x double] [double 0x3FDA9CEC9A9A0000, double 0x3D40931A909FEA5E], [2 x double] [double 0x3FDAF12932478000, double 0xBD3E53BB31EED7A9], [2 x double] [double 0x3FDB44F77BCC8000, double 0x3D4EC5197DDB55D3], [2 x double] [double 0x3FDB985896930000, double 0x3D50FB598FB14F89], [2 x double] [double 0x3FDBEB4D9DA70000, double 0x3D5B7BF7861D37AC], [2 x double] [double 0x3FDC3DD7A7CD8000, double 0x3D66A6B9D9E0A5BD], [2 x double] [double 0x3FDC8FF7C79A8000, double 0x3D5A21AC25D81EF3], [2 x double] [double 0x3FDCE1AF0B860000, double 0xBD48290905A86AA6], [2 x double] [double 0x3FDD32FE7E010000, double 0xBD542A9E21373414], [2 x double] [double 0x3FDD83E7258A0000, double 0x3D679F2828ADD176], [2 x double] [double 0x3FDDD46A04C20000, double 0xBD6DAFA08CECADB1], [2 x double] [double 0x3FDE24881A7C8000, double 0xBD53D9E34270BA6B], [2 x double] [double 0x3FDE744261D68000, double 0x3D3E1F8DF68DBCF3], [2 x double] [double 0x3FDEC399D2468000, double 0x3D49802EB9DCA7E7], [2 x double] [double 0x3FDF128F5FAF0000, double 0x3D3BB2CD720EC44C], [2 x double] [double 0x3FDF6123FA700000, double 0x3D645630A2B61E5B], [2 x double] [double 0x3FDFAF588F790000, double 0xBD49C24CA098362B], [2 x double] [double 0x3FDFFD2E08580000, double 0xBD46CF54D05F9367], [2 x double] [double 0x3FE02552A5A5C000, double 0x3D60FEC69C695D7F], [2 x double] [double 0x3FE04BDF9DA94000, double 0xBD692D9A033EFF75], [2 x double] [double 0x3FE0723E5C1CC000, double 0x3D6F404E57963891], [2 x double] [double 0x3FE0986F4F574000, double 0xBD55BE8DC04AD601], [2 x double] [double 0x3FE0BE72E4254000, double 0xBD657D49676844CC], [2 x double] [double 0x3FE0E44985D1C000, double 0x3D5917EDD5CBBD2D], [2 x double] [double 0x3FE109F39E2D4000, double 0x3D592DFBC7D93617], [2 x double] [double 0x3FE12F7195940000, double 0xBD6043ACFEDCE638], [2 x double] [double 0x3FE154C3D2F4C000, double 0x3D65E9A98F33A396], [2 x double] [double 0x3FE179EABBD88000, double 0x3D69A0BFC60E6FA0], [2 x double] [double 0x3FE19EE6B467C000, double 0x3D52DD98B97BAEF0], [2 x double] [double 0x3FE1C3B81F714000, double 0xBD3EDA1B58389902], [2 x double] [double 0x3FE1E85F5E704000, double 0x3D1A07BD8B34BE7C], [2 x double] [double 0x3FE20CDCD192C000, double 0xBD64926CAFC2F08A], [2 x double] [double 0x3FE23130D7BEC000, double 0xBD17AFA4392F1BA7], [2 x double] [double 0x3FE2555BCE990000, double 0xBD506987F78A4A5E], [2 x double] [double 0x3FE2795E1289C000, double 0xBD5DCA290F81848D], [2 x double] [double 0x3FE29D37FEC2C000, double 0xBD5EEA6F465268B4], [2 x double] [double 0x3FE2C0E9ED448000, double 0x3D5D1772F5386374], [2 x double] [double 0x3FE2E47436E40000, double 0x3D334202A10C3491], [2 x double] [double 0x3FE307D7334F0000, double 0x3D60BE1FB590A1F5], [2 x double] [double 0x3FE32B1339120000, double 0x3D6D71320556B67B], [2 x double] [double 0x3FE34E289D9D0000, double 0xBD6E2CE9146D277A], [2 x double] [double 0x3FE37117B5474000, double 0x3D4ED71774092113], [2 x double] [double 0x3FE393E0D3564000, double 0xBD65E6563BBD9FC9], [2 x double] [double 0x3FE3B6844A000000, double 0xBD3EEA838909F3D3], [2 x double] [double 0x3FE3D9026A714000, double 0x3D66FAA404263D0B], [2 x double] [double 0x3FE3FB5B84D18000, double 0xBD60BDA4B162AFA3], [2 x double] [double 0x3FE41D8FE8468000, double 0xBD5AA33736867A17], [2 x double] [double 0x3FE43F9FE2F9C000, double 0x3D5CCEF4E4F736C2], [2 x double] [double 0x3FE4618BC21C4000, double 0x3D6EC27D0B7B37B3], [2 x double] [double 0x3FE48353D1EA8000, double 0x3D51BEE7ABD17660], [2 x double] [double 0x3FE4A4F85DB04000, double 0xBD244FDD840B8591], [2 x double] [double 0x3FE4C679AFCD0000, double 0xBD61C64E971322CE], [2 x double] [double 0x3FE4E7D811B74000, double 0x3D6BB09CB0985646], [2 x double] [double 0x3FE50913CC018000, double 0xBD6794B434C5A4F5], [2 x double] [double 0x3FE52A2D265BC000, double 0x3D46ABB9DF22BC57], [2 x double] [double 0x3FE54B2467998000, double 0x3D6497A915428B44], [2 x double] [double 0x3FE56BF9D5B40000, double 0xBD58CD7DC73BD194], [2 x double] [double 0x3FE58CADB5CD8000, double 0xBD49DB3DB43689B4], [2 x double] [double 0x3FE5AD404C358000, double 0x3D6F2CFB29AAA5F0], [2 x double] [double 0x3FE5CDB1DC6C0000, double 0x3D67648CF6E3C5D7], [2 x double] [double 0x3FE5EE02A9240000, double 0x3D667570D6095FD2], [2 x double] [double 0x3FE60E32F4478000, double 0x3D51B194F912B417], [2 x double] [double 0x3FE62E42FEFA4000, double 0xBD48432A1B0E2634]]
@0 = private unnamed_addr constant [17879 x i8] c"Enzyme compilation failed.\0ACurrent scope: \0A; Function Attrs: mustprogress willreturn\0Adefine internal fastcc void @preprocess_julia_potrs__15971(i32 zeroext %0, {} addrspace(10)* noundef nonnull align 16 dereferenceable(40) %1, {} addrspace(10)* noundef nonnull align 16 dereferenceable(40) %2) unnamed_addr #71 !dbg !3896 {\0Atop:\0A  %3 = call noalias nonnull dereferenceable(8) dereferenceable_or_null(8) i8* @malloc(i64 8), !enzyme_fromstack !360\0A  %4 = bitcast i8* %3 to i64*, !enzyme_caststack !70\0A  %5 = bitcast i64* %4 to i8*\0A  %6 = call noalias nonnull dereferenceable(1) dereferenceable_or_null(1) i8* @malloc(i64 1), !enzyme_fromstack !3897\0A  %7 = call noalias nonnull dereferenceable(8) dereferenceable_or_null(8) i8* @malloc(i64 8), !enzyme_fromstack !360\0A  %8 = bitcast i8* %7 to i64*, !enzyme_caststack !70\0A  %9 = bitcast i64* %8 to i8*\0A  %10 = call noalias nonnull dereferenceable(8) dereferenceable_or_null(8) i8* @malloc(i64 8), !enzyme_fromstack !360\0A  %11 = bitcast i8* %10 to i64*, !enzyme_caststack !70\0A  %12 = bitcast i64* %11 to i8*\0A  %13 = call noalias nonnull dereferenceable(8) dereferenceable_or_null(8) i8* @malloc(i64 8), !enzyme_fromstack !360\0A  %14 = bitcast i8* %13 to i64*, !enzyme_caststack !70\0A  %15 = bitcast i64* %14 to i8*\0A  %16 = call noalias nonnull dereferenceable(8) dereferenceable_or_null(8) i8* @malloc(i64 8), !enzyme_fromstack !360\0A  %17 = bitcast i8* %16 to i64*, !enzyme_caststack !70\0A  %18 = bitcast i64* %17 to i8*\0A  %19 = call noalias nonnull dereferenceable(16) dereferenceable_or_null(16) i8* @malloc(i64 16), !enzyme_fromstack !2661\0A  %20 = bitcast i8* %19 to [2 x i64]*, !enzyme_caststack !70\0A  %21 = call {}*** @julia.get_pgcstack() #72\0A  %22 = addrspacecast {} addrspace(10)* %1 to {} addrspace(11)*, !dbg !3898\0A  %23 = bitcast {} addrspace(10)* %1 to {} addrspace(10)* addrspace(10)*, !dbg !3898\0A  %24 = addrspacecast {} addrspace(10)* addrspace(10)* %23 to {} addrspace(10)* addrspace(11)*, !dbg !3898\0A  %25 = getelementptr inbounds {} addrspace(10)*, {} addrspace(10)* addrspace(11)* %24, i64 3, !dbg !3898\0A  %26 = bitcast {} addrspace(10)* addrspace(11)* %25 to i64 addrspace(11)*, !dbg !3898\0A  %27 = load i64, i64 addrspace(11)* %26, align 8, !dbg !3898, !tbaa !74, !range !78, !invariant.load !70, !alias.scope !79, !noalias !82\0A  %28 = getelementptr inbounds {} addrspace(10)*, {} addrspace(10)* addrspace(11)* %24, i64 4, !dbg !3898\0A  %29 = bitcast {} addrspace(10)* addrspace(11)* %28 to i64 addrspace(11)*, !dbg !3898\0A  %30 = load i64, i64 addrspace(11)* %29, align 16, !dbg !3898, !tbaa !74, !range !78, !invariant.load !70, !alias.scope !79, !noalias !82\0A  %.not = icmp eq i64 %27, %30, !dbg !3901\0A  br i1 %.not, label %L13, label %L6, !dbg !3902\0A\0AL6:                                               ; preds = %top\0A  %31 = getelementptr inbounds [2 x i64], [2 x i64]* %20, i64 0, i64 0, !dbg !3903\0A  store i64 %27, i64* %31, align 8, !dbg !3903, !tbaa !129, !alias.scope !131, !noalias !3904\0A  %32 = getelementptr inbounds [2 x i64], [2 x i64]* %20, i64 0, i64 1, !dbg !3903\0A  store i64 %30, i64* %32, align 8, !dbg !3903, !tbaa !129, !alias.scope !131, !noalias !3904\0A  %33 = addrspacecast [2 x i64]* %20 to [2 x i64] addrspace(11)*, !dbg !3902\0A  %34 = call nonnull {} addrspace(10)* @julia_string_15923({} addrspace(10)* nofree noundef nonnull align 128 addrspacecast ({}* inttoptr (i64 140273867056768 to {}*) to {} addrspace(10)*), [2 x i64] addrspace(11)* nocapture nofree noundef nonnull readonly align 8 dereferenceable(16) %33) #73, !dbg !3902\0A  %current_task2143 = getelementptr inbounds {}**, {}*** %21, i64 -13, !dbg !3902\0A  %current_task21 = bitcast {}*** %current_task2143 to {}**, !dbg !3902\0A  %35 = call noalias nonnull {} addrspace(10)* @julia.gc_alloc_obj({}** nonnull %current_task21, i64 noundef 8, {} addrspace(10)* noundef addrspacecast ({}* inttoptr (i64 140273859838656 to {}*) 

[wsmoses]

What version of enzyme are you on? (the output of ] st`). This bug doesn't reproduce for me (and from the looks was fixed a long time ago).

Reopen if it persists on current release/main

[torfjelde]

Sorry, forgot to include the deps as I had already added it to my original comment in the PR.
But aight, will check with main when I find some time 👍

[yebai]

I can reproduce this bug on MacOS.

(jl_V30byf) pkg> st
Status `/private/var/folders/m8/xtrwgfnd6n17xhdx98_9y5nc0000gn/T/jl_V30byf/Project.toml`
  [31c24e10] Distributions v0.25.100
  [7da242da] Enzyme v0.11.8

julia> versioninfo()
Julia Version 1.9.3
Commit bed2cd540a1 (2023-08-24 14:43 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: macOS (arm64-apple-darwin22.4.0)
  CPU: 20 × Apple M1 Ultra
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-14.0.6 (ORCJIT, apple-m1)
  Threads: 1 on 16 virtual cores

Error:

julia> using Distributions, Enzyme

julia> let dist = Wishart(7, [1 0.5; 0.5 1]), x = rand(dist), ∂f∂x = zero(x)
           _, y = Enzyme.autodiff(Enzyme.ReverseWithPrimal, logpdf, Enzyme.Active,
                                  Enzyme.Const(dist), Enzyme.Duplicated(x, ∂f∂x))
           y, ∂f∂x
       end
warning: didn't implement memmove, using memcpy as fallback which can result in errors
warning: didn't implement memmove, using memcpy as fallback which can result in errors
warning: didn't implement memmove, using memcpy as fallback which can result in errors
warning: didn't implement memmove, using memcpy as fallback which can result in errors
ERROR: LLVM error: broken gc calling conv fix
Range types must match instruction type!
  call fastcc void @augmented_julia_chkuplofinite_4765({} addrspace(10)* nocapture noundef nonnull readonly align 16 dereferenceable(40) %0), !dbg !700, !range !166

; ModuleID = 'start'
source_filename = "start"
target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128-ni:10:11:12:13"
target triple = "arm64-apple-darwin22.4.0"

@_j_str5 = private unnamed_addr constant [11 x i8] c"typeassert\00"
@_j_const7 = private unnamed_addr constant [129 x [2 x double]] [[2 x double] zeroinitializer, [2 x double] [double 0x3F7FE02A6B200000, double 0xBD6F30EE07912DF9], [2 x double] [double 0x3F8FC0A8B1000000, double 0xBD5FE0E183092C59], [2 x double] [double 0x3F97B91B07D80000, double 0xBD62772AB6C0559C], [2 x double] [double 0x3F9F829B0E780000, double 0x3D2980267C7E09E4], [2 x double] [double 0x3FA39E87BA000000, double 0xBD642A056FEA4DFD], [2 x double] [double 0x3FA77458F6340000, double 0xBD62303B9CB0D5E1], [2 x double] [double 0x3FAB42DD71180000, double 0x3D671BEC28D14C7E], [2 x double] [double 0x3FAF0A30C0100000, double 0x3D662A6617CC9717], [2 x double] [double 0x3FB16536EEA40000, double 0xBD60A3E2F3B47D18], [2 x double] [double 0x3FB341D7961C0000, double 0xBD4717B6B33E44F8], [2 x double] [double 0x3FB51B073F060000, double 0x3D383F69278E686A], [2 x double] [double 0x3FB6F0D28AE60000, double 0xBD62968C836CC8C2], [2 x double] [double 0x3FB8C345D6320000, double 0xBD5937C294D2F567], [2 x double] [double 0x3FBA926D3A4A0000, double 0x3D6AAC6CA17A4554], [2 x double] ......

[wsmoses]

Let me try my Mac laptop to try to reproduce as well.

[wsmoses]

If not I may ask to hop on a zoom or something to try to figure out the root cause.

[simsurace]

I can reproduce with the same error:

julia> versioninfo()
Julia Version 1.9.3
Commit bed2cd540a1 (2023-08-24 14:43 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 8 × Intel(R) Core(TM) i7-9700K CPU @ 3.60GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-14.0.6 (ORCJIT, skylake)
  Threads: 1 on 8 virtual cores
Environment:
  JULIA_IMAGE_THREADS = 1

(@gp) pkg> st
Status `~/.julia/environments/gp/Project.toml`
  [31c24e10] Distributions v0.25.102
  [7da242da] Enzyme v0.11.8 `~/.julia/dev/Enzyme`

[wsmoses]

Looks like it was 1.9 specific. Fixed by the above PR.

Note that this then goes hits an unimplemented lapack call:

No augmented forward pass found for dpotrf_64_
 at context:   call void @dpotrf_64_(i8* nonnull %7, i8* nonnull %10, i64 %46, i8* nonnull %13, i64 %bitcast_coercion32, i64 1) #86 [ "jl_roots"({} addrspace(10)* null, {} addrspace(10)* null, {} addrspace(10)* %3, {} addrspace(10)* null, {} addrspace(10)* null) ], !dbg !132

Stacktrace:
 [1] potrf!
   @ ~/git/Enzyme.jl/julia-1.10.0-beta2/share/julia/stdlib/v1.10/LinearAlgebra/src/lapack.jl:3226


Stacktrace:
  [1] throwerr(cstr::Cstring)
    @ Enzyme.Compiler ~/git/Enzyme.jl/src/compiler.jl:3012
  [2] potrf!
    @ ~/git/Enzyme.jl/julia-1.10.0-beta2/share/julia/stdlib/v1.10/LinearAlgebra/src/lapack.jl:3226
  [3] _chol!
    @ ~/git/Enzyme.jl/julia-1.10.0-beta2/share/julia/stdlib/v1.10/LinearAlgebra/src/cholesky.jl:187 [inlined]
  [4] #cholesky!#140
    @ ~/git/Enzyme.jl/julia-1.10.0-beta2/share/julia/stdlib/v1.10/LinearAlgebra/src/cholesky.jl:268
  [5] cholesky!
    @ ~/git/Enzyme.jl/julia-1.10.0-beta2/share/julia/stdlib/v1.10/LinearAlgebra/src/cholesky.jl:267 [inlined]
  [6] #cholesky!#141
    @ ~/git/Enzyme.jl/julia-1.10.0-beta2/share/julia/stdlib/v1.10/LinearAlgebra/src/cholesky.jl:301
  [7] cholesky! (repeats 2 times)
    @ ~/git/Enzyme.jl/julia-1.10.0-beta2/share/julia/stdlib/v1.10/LinearAlgebra/src/cholesky.jl:295 [inlined]
  [8] cholesky (repeats 2 times)
    @ ~/git/Enzyme.jl/julia-1.10.0-beta2/share/julia/stdlib/v1.10/LinearAlgebra/src/cholesky.jl:401 [inlined]
  [9] nonsingular_wishart_logkernel
    @ ~/.julia/packages/Distributions/fYgbJ/src/matrix/wishart.jl:189
 [10] logkernel
    @ ~/.julia/packages/Distributions/fYgbJ/src/matrix/wishart.jl:163 [inlined]
 [11] _logpdf
    @ ~/.julia/packages/Distributions/fYgbJ/src/matrixvariates.jl:82 [inlined]
 [12] logpdf
    @ ~/.julia/packages/Distributions/fYgbJ/src/common.jl:250 [inlined]
 [13] logpdf
    @ ~/.julia/packages/Distributions/fYgbJ/src/common.jl:0 [inlined]
 [14] diffejulia_logpdf_1054_inner_1wrap
    @ ~/.julia/packages/Distributions/fYgbJ/src/common.jl:0
 [15] macro expansion

[yebai]

@wsmoses I suggest that we always add a test to similar bugfix PRs, so avoid accidentally breaking these again. This is particularly useful, given that subtle Julia internal changes often affect Enzyme.

[devmotion]

I was happy when I saw that this issue was closed - but when checking the MWE I noticed that it still doesn't work and one runs into the BLAS issue mentioned in #1081 (comment). So maybe this issue should be kept open until the potrf! issue is fixed?

[wsmoses]

So these are two separate issues, one of which is that we had the calling convention mismatch (here), the second of which is an implementation of potrf and/or cholesky, etc.

I'd like to keep this issue just to the calling conv fix. You can open another for cholesky, though I think we may have ones for those already. In any case, it should be fixed by #1220 which I was just finishing up now and hopefully should be merged in a few mins.

[wsmoses]

and @devmotion FYI cholesky support is now merged. If you retry turing back on main, that error should go away.

The final remaining error is an issue that comes from Julia1.10/GPUCompiler causing an infinite recursion for hvcat, that doesn't have any good fixes immediately. I'd recommend commenting that final stackoverflow error out, but then I think the remaining Turing.jl should pass