Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

armv7l: ptrtoint not supported for non-integral pointers #36062

Closed
staticfloat opened this issue May 28, 2020 · 13 comments · Fixed by #36705
Closed

armv7l: ptrtoint not supported for non-integral pointers #36062

staticfloat opened this issue May 28, 2020 · 13 comments · Fixed by #36705
Labels
kind:bug Indicates an unexpected problem or unintended behavior kind:upstream The issue is with an upstream dependency, e.g. LLVM system:arm ARMv7 and AArch64 system:32-bit Affects only 32-bit systems

Comments

@staticfloat
Copy link
Sponsor Member

Armv7l bootstrap is failing right at the end during the link step with this:

ptrtoint not supported for non-integral pointers
inttoptr not supported for non-integral pointers
ptrtoint not supported for non-integral pointers
inttoptr not supported for non-integral pointers
in function julia__unsafe_copytoNOT._53532
LLVM ERROR: Broken function found, compilation aborted!

Exemplary build: https://build.julialang.org/#/builders/57/builds/231
@staticfloat staticfloat added kind:bug Indicates an unexpected problem or unintended behavior system:32-bit Affects only 32-bit systems system:arm ARMv7 and AArch64 labels May 28, 2020
@IanButterworth
Copy link
Sponsor Member

Looks very similar to this LLVM issue #31156 (comment)

ptrtoint not supported for non-integral pointers
inttoptr not supported for non-integral pointers
ptrtoint not supported for non-integral pointers
inttoptr not supported for non-integral pointers
in function julia_copyto!_18518
LLVM ERROR: Broken function found, compilation aborted!

which was patched by @Keno and ultimately merged in #35313

Seems like the patch didn't fix julia__unsafe_copytoNOT.?

@giordano
Copy link
Contributor

giordano commented Jun 3, 2020

@terasakisatoshi has been able to build julia v1.4.2 for armv7l: https://discourse.julialang.org/t/have-a-try-julia-v1-4-2-for-arm32bit/40284

@Keno
Copy link
Member

Keno commented Jun 3, 2020

Inttoptr issues like this indicate an un-soundness in some LLVM pass. They need to be fixed one by one.

@ViralBShah
Copy link
Member

Will this hurt us on those new arm macs? I imagine they are aarch64, and so this may not be an issue.

@ViralBShah
Copy link
Member

@Keno Is this something we can communicate to our various ARM contacts?

@staticfloat
Copy link
Sponsor Member Author

Correct; armv7l is separate from aarch64 (otherwise known as armv8)

@Keno
Copy link
Member

Keno commented Jul 1, 2020

It's not an ARM issue, it's an LLVM issue.

@mbauman mbauman added the kind:upstream The issue is with an upstream dependency, e.g. LLVM label Jul 1, 2020
@ViralBShah
Copy link
Member

Right - but is it only an LLVM issue on arm - and hence the ARM llvm team should look at it, or is it a general LLVM issue that needs to be fixed by the LLVM community?

@Keno
Copy link
Member

Keno commented Jul 1, 2020

It's specific to how we are using LLVM. It's unlikely the ARM LLVM team would care enough to understand how we're using LLVM.

@yuyichao
Copy link
Contributor

Bugpoint reduced repro:

; ModuleID = 'bugpoint-reduced-simplified.bc'
source_filename = "text"
target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64-ni:10:11:12:13"
target triple = "armv7l-unknown-linux-gnueabihf"

define void @julia_show_tuple_as_call_23179() local_unnamed_addr {
top:
  br i1 undef, label %L16.lr.ph, label %L51

L16.lr.ph:                                        ; preds = %top
  unreachable

L51:                                              ; preds = %top
  %0 = addrspacecast [2 x {} addrspace(10)*]* undef to [2 x {} addrspace(10)*] addrspace(11)*
  call fastcc void @julia_show_signature_function_32381([2 x {} addrspace(10)*] addrspace(11)* nocapture readonly %0)
  ret void
}

define dso_local fastcc void @julia_show_signature_function_32381([2 x {} addrspace(10)*] addrspace(11)* nocapture nonnull readonly align 4 dereferenceable(8) %0) unnamed_addr {
top:
  %1 = bitcast [2 x {} addrspace(10)*] addrspace(11)* %0 to i64 addrspace(11)*
  ret void
}

!llvm.module.flags = !{!0}

!0 = !{i32 1, !"Debug Info Version", i32 3}

reproducible with opt bugpoint-reduced-simplified.ll -inline.

Not 100% sure if this is the original one but this is what I got by running opt -O2 on the dumpped unoptimized IR. I had to strip other unrelated functions first before feeding it to bugpoint since it otherwise go OOM....

@yuyichao
Copy link
Contributor

Reported as https://bugs.llvm.org/show_bug.cgi?id=46737

@yuyichao
Copy link
Contributor

That's actually not related and is fixed on LLVM master already. The actual one seems to be caused by Loop Strength Reduction.

What I currently have is.

*** IR Dump Before Loop Strength Reduction ***
; Preheader:
L37.preheader:                                    ; preds = %idxend4
  %110 = bitcast {} addrspace(10)* %1 to { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(10)*, !dbg !8332
  %111 = addrspacecast { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(10)* %110 to { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(11)*, !dbg !8332
  %112 = getelementptr inbounds { i8 addrspace(13)*, i32, i16, i16, i32 }, { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(11)* %111, i32 0, i32 1, !dbg !8332
  %113 = load i32, i32 addrspace(11)* %112, align 4, !dbg !8332, !tbaa !8259
  %114 = bitcast {} addrspace(10)* %1 to i8 addrspace(13)* addrspace(10)*, !dbg !8332
  %115 = addrspacecast i8 addrspace(13)* addrspace(10)* %114 to i8 addrspace(13)* addrspace(11)*, !dbg !8332
  %116 = load i8 addrspace(13)*, i8 addrspace(13)* addrspace(11)* %115, align 4, !dbg !8332
  %117 = bitcast {} addrspace(10)* %1 to { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(10)*, !dbg !8333
  %118 = addrspacecast { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(10)* %117 to { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(11)*, !dbg !8333
  %119 = getelementptr inbounds { i8 addrspace(13)*, i32, i16, i16, i32 }, { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(11)* %118, i32 0, i32 1, !dbg !8333
  %120 = load i32, i32 addrspace(11)* %119, align 4, !dbg !8333
  %121 = bitcast {} addrspace(10)* %1 to i8 addrspace(13)* addrspace(10)*, !dbg !8333
  %122 = addrspacecast i8 addrspace(13)* addrspace(10)* %121 to i8 addrspace(13)* addrspace(11)*, !dbg !8333
  %123 = load i8 addrspace(13)*, i8 addrspace(13)* addrspace(11)* %122, align 4, !dbg !8333
  %124 = bitcast {} addrspace(10)* %1 to { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(10)*, !dbg !8334
  %125 = addrspacecast { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(10)* %124 to { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(11)*, !dbg !8334
  %126 = getelementptr inbounds { i8 addrspace(13)*, i32, i16, i16, i32 }, { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(11)* %125, i32 0, i32 1, !dbg !8334
  %127 = load i32, i32 addrspace(11)* %126, align 4, !dbg !8334
  %128 = bitcast {} addrspace(10)* %1 to i8 addrspace(13)* addrspace(10)*, !dbg !8334
  %129 = addrspacecast i8 addrspace(13)* addrspace(10)* %128 to i8 addrspace(13)* addrspace(11)*, !dbg !8334
  %130 = load i8 addrspace(13)*, i8 addrspace(13)* addrspace(11)* %129, align 4, !dbg !8334
  %131 = bitcast {} addrspace(10)* %1 to { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(10)*, !dbg !8335
  %132 = addrspacecast { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(10)* %131 to { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(11)*, !dbg !8335
  %133 = getelementptr inbounds { i8 addrspace(13)*, i32, i16, i16, i32 }, { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(11)* %132, i32 0, i32 1, !dbg !8335
  %134 = load i32, i32 addrspace(11)* %133, align 4, !dbg !8335
  %135 = bitcast {} addrspace(10)* %1 to i8 addrspace(13)* addrspace(10)*, !dbg !8335
  %136 = addrspacecast i8 addrspace(13)* addrspace(10)* %135 to i8 addrspace(13)* addrspace(11)*, !dbg !8335
  %137 = load i8 addrspace(13)*, i8 addrspace(13)* addrspace(11)* %136, align 4, !dbg !8335
  br label %L37, !dbg !8280

; Loop:
L37:                                              ; preds = %L37.preheader, %idxend16
  %value_phi7 = phi i32 [ %162, %idxend16 ], [ %107, %L37.preheader ]
  %44 = add i32 %value_phi7, -1, !dbg !8280
  %45 = icmp ult i32 %44, %113, !dbg !8280
  br i1 %45, label %idxend10, label %oob9, !dbg !8280

idxend10:                                         ; preds = %L37
  %140 = getelementptr inbounds i8, i8 addrspace(13)* %116, i32 %44, !dbg !8280
  %141 = load i8, i8 addrspace(13)* %140, align 1, !dbg !8280, !tbaa !8300
  %142 = icmp eq i8 %141, 58, !dbg !8336
  br i1 %142, label %L49, label %L47, !dbg !8281

L49:                                              ; preds = %idxend10
  %50 = add i32 %value_phi7, -1, !dbg !8282
  %51 = icmp ult i32 %50, %120, !dbg !8282
  br i1 %51, label %idxend12, label %oob11, !dbg !8282

idxend12:                                         ; preds = %L49
  %145 = getelementptr inbounds i8, i8 addrspace(13)* %123, i32 %50, !dbg !8282
  store i8 48, i8 addrspace(13)* %145, align 1, !dbg !8282, !tbaa !8300
  %146 = add i32 %value_phi7, -2, !dbg !8342
  %147 = icmp ult i32 %146, %127, !dbg !8342
  br i1 %147, label %idxend14, label %oob13, !dbg !8342

idxend14:                                         ; preds = %idxend12
  %151 = getelementptr inbounds i8, i8 addrspace(13)* %130, i32 %146, !dbg !8342
  %152 = load i8, i8 addrspace(13)* %151, align 1, !dbg !8342, !tbaa !8300
  %153 = zext i8 %152 to i32, !dbg !8344
  %154 = add nuw nsw i32 %153, 1, !dbg !8284
  %155 = trunc i32 %154 to i8, !dbg !8349
  %156 = icmp eq i8 %152, -1, !dbg !8288
  br i1 %156, label %L59, label %L64, !dbg !8288

L64:                                              ; preds = %idxend14
  %64 = add i32 %value_phi7, -2, !dbg !8292
  %65 = icmp ult i32 %64, %134, !dbg !8292
  br i1 %65, label %idxend16, label %oob15, !dbg !8292

idxend16:                                         ; preds = %L64
  %160 = getelementptr inbounds i8, i8 addrspace(13)* %137, i32 %64, !dbg !8292
  store i8 %155, i8 addrspace(13)* %160, align 1, !dbg !8292, !tbaa !8300
  %161 = icmp eq i32 %value_phi7, %108, !dbg !8350
  %162 = add i32 %value_phi7, -1, !dbg !8351
  br i1 %161, label %L77.loopexit, label %L37, !dbg !8287

; Exit blocks
oob9:                                             ; preds = %L37
  %value_phi7.lcssa = phi i32 [ %value_phi7, %L37 ]
  %138 = alloca i32, align 4, !dbg !8280
  store i32 %value_phi7.lcssa, i32* %138, align 4, !dbg !8280
  %139 = addrspacecast {} addrspace(10)* %1 to {} addrspace(12)*, !dbg !8280
  call void @jl_bounds_error_ints({} addrspace(12)* %139, i32* nonnull %138, i32 1), !dbg !8280
  unreachable, !dbg !8280

L47:                                              ; preds = %idxend10
  %.sroa.027.0..sroa_idx = getelementptr inbounds [2 x i32], [2 x i32]* %0, i32 0, i32 0, !dbg !8281
  store i32 %2, i32* %.sroa.027.0..sroa_idx, align 4, !dbg !8281
  %.sroa.228.0..sroa_idx29 = getelementptr inbounds [2 x i32], [2 x i32]* %0, i32 0, i32 1, !dbg !8281
  store i32 %3, i32* %.sroa.228.0..sroa_idx29, align 4, !dbg !8281
  %46 = getelementptr {} addrspace(10)*, {} addrspace(10)** %gcframe, i32 1
  %47 = load {} addrspace(10)*, {} addrspace(10)** %46, align 4, !tbaa !8249
  %48 = getelementptr {}**, {}*** %ptls, i32 0
  %49 = bitcast {}*** %48 to {} addrspace(10)**
  store {} addrspace(10)* %47, {} addrspace(10)** %49, align 4, !tbaa !8249
  ret void, !dbg !8281

oob11:                                            ; preds = %L49
  %value_phi7.lcssa39 = phi i32 [ %value_phi7, %L49 ]
  %143 = alloca i32, align 4, !dbg !8282
  store i32 %value_phi7.lcssa39, i32* %143, align 4, !dbg !8282
  %144 = addrspacecast {} addrspace(10)* %1 to {} addrspace(12)*, !dbg !8282
  call void @jl_bounds_error_ints({} addrspace(12)* %144, i32* nonnull %143, i32 1), !dbg !8282
  unreachable, !dbg !8282

oob13:                                            ; preds = %idxend12
  %value_phi7.lcssa51 = phi i32 [ %value_phi7, %idxend12 ]
  %148 = add i32 %value_phi7.lcssa51, -1, !dbg !8343
  %149 = alloca i32, align 4, !dbg !8342
  store i32 %148, i32* %149, align 4, !dbg !8342
  %150 = addrspacecast {} addrspace(10)* %1 to {} addrspace(12)*, !dbg !8342
  call void @jl_bounds_error_ints({} addrspace(12)* %150, i32* nonnull %149, i32 1), !dbg !8342
  unreachable, !dbg !8342

L59:                                              ; preds = %idxend14
  %.lcssa47 = phi i32 [ %154, %idxend14 ], !dbg !8284
  %52 = load {}*, {}** @jl_globalYY.16, align 4, !dbg !8288, !tbaa !8276, !nonnull !4
  %53 = addrspacecast {}* %52 to {} addrspace(10)*, !dbg !8288
  %54 = load {}*, {}** @jl_symYY.trunc61, align 4, !dbg !8288, !tbaa !8276, !nonnull !4
  %55 = addrspacecast {}* %54 to {} addrspace(10)*, !dbg !8288
  %56 = load {}*, {}** @SUM.CoreDOT.UInt8254, align 4, !dbg !8288, !tbaa !8276, !nonnull !4, !dereferenceable !8278, !align !8279
  %57 = addrspacecast {}* %56 to {} addrspace(10)*, !dbg !8288
  %58 = call {} addrspace(10)* @jl_box_int32(i32 signext %.lcssa47), !dbg !8288
  %59 = getelementptr {} addrspace(10)*, {} addrspace(10)** %gcframe, i32 2
  store {} addrspace(10)* %58, {} addrspace(10)** %59
  %60 = getelementptr {} addrspace(10)*, {} addrspace(10)** %4, i32 0, !dbg !8288
  store {} addrspace(10)* %55, {} addrspace(10)** %60, align 4, !dbg !8288
  %61 = getelementptr {} addrspace(10)*, {} addrspace(10)** %4, i32 1, !dbg !8288
  store {} addrspace(10)* %57, {} addrspace(10)** %61, align 4, !dbg !8288
  %62 = getelementptr {} addrspace(10)*, {} addrspace(10)** %4, i32 2, !dbg !8288
  store {} addrspace(10)* %58, {} addrspace(10)** %62, align 4, !dbg !8288
  %63 = call nonnull {} addrspace(10)* @jl_apply_generic({} addrspace(10)* %53, {} addrspace(10)** %4, i32 3), !dbg !8288
  call void @llvm.trap(), !dbg !8288
  unreachable, !dbg !8288

oob15:                                            ; preds = %L64
  %value_phi7.lcssa50 = phi i32 [ %value_phi7, %L64 ]
  %157 = add i32 %value_phi7.lcssa50, -1, !dbg !8343
  %158 = alloca i32, align 4, !dbg !8292
  store i32 %157, i32* %158, align 4, !dbg !8292
  %159 = addrspacecast {} addrspace(10)* %1 to {} addrspace(12)*, !dbg !8292
  call void @jl_bounds_error_ints({} addrspace(12)* %159, i32* nonnull %158, i32 1), !dbg !8292
  unreachable, !dbg !8292

L77.loopexit:                                     ; preds = %idxend16
  br label %L77, !dbg !8293
*** IR Dump After Loop Strength Reduction ***
; Preheader:
L37.preheader:                                    ; preds = %idxend4
  %112 = bitcast {} addrspace(10)* %1 to { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(10)*, !dbg !8332
  %113 = addrspacecast { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(10)* %112 to { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(11)*, !dbg !8332
  %114 = getelementptr inbounds { i8 addrspace(13)*, i32, i16, i16, i32 }, { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(11)* %113, i32 0, i32 1, !dbg !8332
  %115 = load i32, i32 addrspace(11)* %114, align 4, !dbg !8332, !tbaa !8259
  %116 = bitcast {} addrspace(10)* %1 to i8 addrspace(13)* addrspace(10)*, !dbg !8332
  %117 = addrspacecast i8 addrspace(13)* addrspace(10)* %116 to i8 addrspace(13)* addrspace(11)*, !dbg !8332
  %118 = load i8 addrspace(13)*, i8 addrspace(13)* addrspace(11)* %117, align 4, !dbg !8332
  %119 = bitcast {} addrspace(10)* %1 to { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(10)*, !dbg !8333
  %120 = addrspacecast { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(10)* %119 to { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(11)*, !dbg !8333
  %121 = getelementptr inbounds { i8 addrspace(13)*, i32, i16, i16, i32 }, { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(11)* %120, i32 0, i32 1, !dbg !8333
  %122 = load i32, i32 addrspace(11)* %121, align 4, !dbg !8333
  %123 = bitcast {} addrspace(10)* %1 to i8 addrspace(13)* addrspace(10)*, !dbg !8333
  %124 = addrspacecast i8 addrspace(13)* addrspace(10)* %123 to i8 addrspace(13)* addrspace(11)*, !dbg !8333
  %125 = load i8 addrspace(13)*, i8 addrspace(13)* addrspace(11)* %124, align 4, !dbg !8333
  %126 = bitcast {} addrspace(10)* %1 to { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(10)*, !dbg !8334
  %127 = addrspacecast { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(10)* %126 to { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(11)*, !dbg !8334
  %128 = getelementptr inbounds { i8 addrspace(13)*, i32, i16, i16, i32 }, { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(11)* %127, i32 0, i32 1, !dbg !8334
  %129 = load i32, i32 addrspace(11)* %128, align 4, !dbg !8334
  %130 = bitcast {} addrspace(10)* %1 to i8 addrspace(13)* addrspace(10)*, !dbg !8334
  %131 = addrspacecast i8 addrspace(13)* addrspace(10)* %130 to i8 addrspace(13)* addrspace(11)*, !dbg !8334
  %132 = load i8 addrspace(13)*, i8 addrspace(13)* addrspace(11)* %131, align 4, !dbg !8334
  %133 = ptrtoint i8 addrspace(13)* %132 to i32
  %134 = bitcast {} addrspace(10)* %1 to { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(10)*, !dbg !8335
  %135 = addrspacecast { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(10)* %134 to { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(11)*, !dbg !8335
  %136 = getelementptr inbounds { i8 addrspace(13)*, i32, i16, i16, i32 }, { i8 addrspace(13)*, i32, i16, i16, i32 } addrspace(11)* %135, i32 0, i32 1, !dbg !8335
  %137 = load i32, i32 addrspace(11)* %136, align 4, !dbg !8335
  %138 = bitcast {} addrspace(10)* %1 to i8 addrspace(13)* addrspace(10)*, !dbg !8335
  %139 = addrspacecast i8 addrspace(13)* addrspace(10)* %138 to i8 addrspace(13)* addrspace(11)*, !dbg !8335
  %140 = load i8 addrspace(13)*, i8 addrspace(13)* addrspace(11)* %139, align 4, !dbg !8335
  %141 = ptrtoint i8 addrspace(13)* %140 to i32
  %142 = add i32 %2, -2, !dbg !8292
  %143 = add i32 %2, -2, !dbg !8336
  %144 = sub i32 0, %141, !dbg !8280
  %145 = inttoptr i32 %144 to i8 addrspace(13)*
  %146 = sub i32 0, %133, !dbg !8280
  %147 = inttoptr i32 %146 to i8 addrspace(13)*
  %148 = add i32 %2, -2, !dbg !8282
  %149 = add i32 %2, -2, !dbg !8280
  %scevgep62 = getelementptr i8, i8 addrspace(13)* %125, i32 -2, !dbg !8280
  %scevgep66 = getelementptr i8, i8 addrspace(13)* %118, i32 -2, !dbg !8280
  %150 = sub i32 0, %110, !dbg !8280
  %151 = add i32 %150, %2, !dbg !8337
  br label %L37, !dbg !8280

; Loop:
L37:                                              ; preds = %L37.preheader, %idxend16
  %lsr.iv67 = phi i8 addrspace(13)* [ %scevgep66, %L37.preheader ], [ %scevgep68, %idxend16 ]
  %lsr.iv63 = phi i8 addrspace(13)* [ %scevgep62, %L37.preheader ], [ %scevgep64, %idxend16 ]
  %lsr.iv58 = phi i8 addrspace(13)* [ %147, %L37.preheader ], [ %scevgep59, %idxend16 ]
  %lsr.iv55 = phi i8 addrspace(13)* [ %145, %L37.preheader ], [ %scevgep, %idxend16 ]
  %lsr.iv53 = phi i32 [ 0, %L37.preheader ], [ %lsr.iv.next54, %idxend16 ]
  %lsr.iv = phi i32 [ -1, %L37.preheader ], [ %lsr.iv.next, %idxend16 ]
  %lsr.iv5860 = ptrtoint i8 addrspace(13)* %lsr.iv58 to i32
  %lsr.iv5556 = ptrtoint i8 addrspace(13)* %lsr.iv55 to i32
  %44 = add i32 %2, %lsr.iv, !dbg !8280
  %45 = add i32 %149, %lsr.iv53, !dbg !8280
  %46 = icmp ult i32 %45, %115, !dbg !8280
  br i1 %46, label %idxend10, label %oob9, !dbg !8280

idxend10:                                         ; preds = %L37
  %scevgep69 = getelementptr i8, i8 addrspace(13)* %lsr.iv67, i32 %2, !dbg !8280
  %154 = load i8, i8 addrspace(13)* %scevgep69, align 1, !dbg !8280, !tbaa !8300
  %155 = icmp eq i8 %154, 58, !dbg !8339
  br i1 %155, label %L49, label %L47, !dbg !8281

L49:                                              ; preds = %idxend10
  %51 = add i32 %148, %lsr.iv53, !dbg !8282
  %52 = icmp ult i32 %51, %122, !dbg !8282
  br i1 %52, label %idxend12, label %oob11, !dbg !8282

idxend12:                                         ; preds = %L49
  %scevgep65 = getelementptr i8, i8 addrspace(13)* %lsr.iv63, i32 %2, !dbg !8282
  store i8 48, i8 addrspace(13)* %scevgep65, align 1, !dbg !8282, !tbaa !8300
  %158 = add i32 %143, %lsr.iv53, !dbg !8336
  %159 = add i32 %158, -1, !dbg !8336
  %160 = icmp ult i32 %159, %129, !dbg !8336
  br i1 %160, label %idxend14, label %oob13, !dbg !8336

idxend14:                                         ; preds = %idxend12
  %164 = sub i32 0, %lsr.iv5860, !dbg !8336
  %165 = add i32 %164, %143, !dbg !8336
  %166 = inttoptr i32 %165 to i8 addrspace(13)*
  %scevgep61 = getelementptr i8, i8 addrspace(13)* %166, i32 -1, !dbg !8336
  %167 = load i8, i8 addrspace(13)* %scevgep61, align 1, !dbg !8336, !tbaa !8300
  %168 = zext i8 %167 to i32, !dbg !8346
  %169 = add nuw nsw i32 %168, 1, !dbg !8284
  %170 = trunc i32 %169 to i8, !dbg !8351
  %171 = icmp eq i8 %167, -1, !dbg !8288
  br i1 %171, label %L59, label %L64, !dbg !8288

L64:                                              ; preds = %idxend14
  %65 = add i32 %142, %lsr.iv53, !dbg !8292
  %66 = add i32 %65, -1, !dbg !8292
  %67 = icmp ult i32 %66, %137, !dbg !8292
  br i1 %67, label %idxend16, label %oob15, !dbg !8292

idxend16:                                         ; preds = %L64
  %175 = sub i32 0, %lsr.iv5556, !dbg !8292
  %176 = add i32 %175, %143, !dbg !8292
  %177 = inttoptr i32 %176 to i8 addrspace(13)*
  %scevgep57 = getelementptr i8, i8 addrspace(13)* %177, i32 -1, !dbg !8292
  store i8 %170, i8 addrspace(13)* %scevgep57, align 1, !dbg !8292, !tbaa !8300
  %lsr.iv.next = add i32 %lsr.iv, -1, !dbg !8337
  %lsr.iv.next54 = add i32 %lsr.iv53, -1, !dbg !8337
  %scevgep = getelementptr i8, i8 addrspace(13)* %lsr.iv55, i32 1, !dbg !8337
  %scevgep59 = getelementptr i8, i8 addrspace(13)* %lsr.iv58, i32 1, !dbg !8337
  %scevgep64 = getelementptr i8, i8 addrspace(13)* %lsr.iv63, i32 -1, !dbg !8337
  %scevgep68 = getelementptr i8, i8 addrspace(13)* %lsr.iv67, i32 -1, !dbg !8337
  %178 = add i32 %151, %lsr.iv.next54, !dbg !8337
  %179 = icmp eq i32 %178, 0, !dbg !8337
  br i1 %179, label %L77.loopexit, label %L37, !dbg !8287

; Exit blocks
oob9:                                             ; preds = %L37
  %value_phi7.lcssa = phi i32 [ %44, %L37 ]
  %152 = alloca i32, align 4, !dbg !8280
  store i32 %value_phi7.lcssa, i32* %152, align 4, !dbg !8280
  %153 = addrspacecast {} addrspace(10)* %1 to {} addrspace(12)*, !dbg !8280
  call void @jl_bounds_error_ints({} addrspace(12)* %153, i32* nonnull %152, i32 1), !dbg !8280
  unreachable, !dbg !8280

L47:                                              ; preds = %idxend10
  %.sroa.027.0..sroa_idx = getelementptr inbounds [2 x i32], [2 x i32]* %0, i32 0, i32 0, !dbg !8281
  store i32 %2, i32* %.sroa.027.0..sroa_idx, align 4, !dbg !8281
  %.sroa.228.0..sroa_idx29 = getelementptr inbounds [2 x i32], [2 x i32]* %0, i32 0, i32 1, !dbg !8281
  store i32 %3, i32* %.sroa.228.0..sroa_idx29, align 4, !dbg !8281
  %47 = getelementptr {} addrspace(10)*, {} addrspace(10)** %gcframe, i32 1
  %48 = load {} addrspace(10)*, {} addrspace(10)** %47, align 4, !tbaa !8249
  %49 = getelementptr {}**, {}*** %ptls, i32 0
  %50 = bitcast {}*** %49 to {} addrspace(10)**
  store {} addrspace(10)* %48, {} addrspace(10)** %50, align 4, !tbaa !8249
  ret void, !dbg !8281

oob11:                                            ; preds = %L49
  %value_phi7.lcssa39 = phi i32 [ %44, %L49 ]
  %156 = alloca i32, align 4, !dbg !8282
  store i32 %value_phi7.lcssa39, i32* %156, align 4, !dbg !8282
  %157 = addrspacecast {} addrspace(10)* %1 to {} addrspace(12)*, !dbg !8282
  call void @jl_bounds_error_ints({} addrspace(12)* %157, i32* nonnull %156, i32 1), !dbg !8282
  unreachable, !dbg !8282

oob13:                                            ; preds = %idxend12
  %value_phi7.lcssa51 = phi i32 [ %44, %idxend12 ]
  %161 = add i32 %value_phi7.lcssa51, -1, !dbg !8345
  %162 = alloca i32, align 4, !dbg !8336
  store i32 %161, i32* %162, align 4, !dbg !8336
  %163 = addrspacecast {} addrspace(10)* %1 to {} addrspace(12)*, !dbg !8336
  call void @jl_bounds_error_ints({} addrspace(12)* %163, i32* nonnull %162, i32 1), !dbg !8336
  unreachable, !dbg !8336

L59:                                              ; preds = %idxend14
  %.lcssa47 = phi i32 [ %169, %idxend14 ], !dbg !8284
  %53 = load {}*, {}** @jl_globalYY.16, align 4, !dbg !8288, !tbaa !8276, !nonnull !4
  %54 = addrspacecast {}* %53 to {} addrspace(10)*, !dbg !8288
  %55 = load {}*, {}** @jl_symYY.trunc61, align 4, !dbg !8288, !tbaa !8276, !nonnull !4
  %56 = addrspacecast {}* %55 to {} addrspace(10)*, !dbg !8288
  %57 = load {}*, {}** @SUM.CoreDOT.UInt8254, align 4, !dbg !8288, !tbaa !8276, !nonnull !4, !dereferenceable !8278, !align !8279
  %58 = addrspacecast {}* %57 to {} addrspace(10)*, !dbg !8288
  %59 = call {} addrspace(10)* @jl_box_int32(i32 signext %.lcssa47), !dbg !8288
  %60 = getelementptr {} addrspace(10)*, {} addrspace(10)** %gcframe, i32 2
  store {} addrspace(10)* %59, {} addrspace(10)** %60
  %61 = getelementptr {} addrspace(10)*, {} addrspace(10)** %4, i32 0, !dbg !8288
  store {} addrspace(10)* %56, {} addrspace(10)** %61, align 4, !dbg !8288
  %62 = getelementptr {} addrspace(10)*, {} addrspace(10)** %4, i32 1, !dbg !8288
  store {} addrspace(10)* %58, {} addrspace(10)** %62, align 4, !dbg !8288
  %63 = getelementptr {} addrspace(10)*, {} addrspace(10)** %4, i32 2, !dbg !8288
  store {} addrspace(10)* %59, {} addrspace(10)** %63, align 4, !dbg !8288
  %64 = call nonnull {} addrspace(10)* @jl_apply_generic({} addrspace(10)* %54, {} addrspace(10)** %4, i32 3), !dbg !8288
  call void @llvm.trap(), !dbg !8288
  unreachable, !dbg !8288

oob15:                                            ; preds = %L64
  %value_phi7.lcssa50 = phi i32 [ %44, %L64 ]
  %172 = add i32 %value_phi7.lcssa50, -1, !dbg !8345
  %173 = alloca i32, align 4, !dbg !8292
  store i32 %172, i32* %173, align 4, !dbg !8292
  %174 = addrspacecast {} addrspace(10)* %1 to {} addrspace(12)*, !dbg !8292
  call void @jl_bounds_error_ints({} addrspace(12)* %174, i32* nonnull %173, i32 1), !dbg !8292
  unreachable, !dbg !8292

L77.loopexit:                                     ; preds = %idxend16
  br label %L77, !dbg !8293

@yuyichao
Copy link
Contributor

Bugpoint reduced IR:

target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64-ni:10:11:12:13"
target triple = "armv7l-unknown-linux-gnueabihf"

define dso_local void @julia_roundup_19730() {
top:
  br label %L37

L37:                                              ; preds = %L49, %top
  %value_phi7 = phi i32 [ %3, %L49 ], [ 0, %top ]
  br i1 undef, label %L49, label %L47

L47:                                              ; preds = %L37
  unreachable

L49:                                              ; preds = %L37
  %0 = add i32 %value_phi7, -2
  %1 = getelementptr inbounds i8, i8 addrspace(13)* null, i32 %0
  %2 = load i8, i8 addrspace(13)* %1, align 1
  %3 = add i32 %value_phi7, -1
  br label %L37
}

Reported as https://bugs.llvm.org/show_bug.cgi?id=46757

yuyichao added a commit that referenced this issue Jul 17, 2020
@yuyichao
Disable Loop Strength Reduction on ARM
76356ec 
Works around #36062
@yuyichao yuyichao mentioned this issue Jul 17, 2020
Merged
vchuravy added a commit to JuliaPackaging/Yggdrasil that referenced this issue Jul 18, 2020
@vchuravy
[LLVM] add patches for JuliaLang/julia#36062
08b7620
vchuravy added a commit to JuliaPackaging/Yggdrasil that referenced this issue Jul 18, 2020
@vchuravy
[LLVM] add patches for JuliaLang/julia#36062
d15e15d
@yuyichao yuyichao mentioned this issue Jul 18, 2020
Merged
vchuravy added a commit to JuliaPackaging/Yggdrasil that referenced this issue Jul 18, 2020
@vchuravy
Merge pull request #1318 from JuliaPackaging/vc/llvm_armppc
47518e6 
[LLVM] add patches for JuliaLang/julia#36062
yuyichao added a commit that referenced this issue Jul 18, 2020
@yuyichao
Remove non-integral pointer from data layout before codegen
b17627d 
This allows LLVM codegen passes to insert `inttoptr` and `ptrtoint` as it wish,
and avoids hitting any illegal ones in those passes.

Fix #36062
yuyichao added a commit that referenced this issue Jul 24, 2020
@yuyichao
Remove non-integral pointer from data layout before codegen
91efeac 
This allows LLVM codegen passes to insert `inttoptr` and `ptrtoint` as it wish,
and avoids hitting any illegal ones in those passes.

Fix #36062
yuyichao added a commit that referenced this issue Jul 24, 2020
@yuyichao
Remove non-integral pointer from data layout before codegen (#36705)
235784a 
This allows LLVM codegen passes to insert `inttoptr` and `ptrtoint` as it wish,
and avoids hitting any illegal ones in those passes.

Fix #36062
@terasakisatoshi terasakisatoshi mentioned this issue Aug 5, 2020
Merged
KristofferC pushed a commit that referenced this issue Aug 10, 2020
@yuyichao @KristofferC
Remove non-integral pointer from data layout before codegen (#36705)
f3d8ffb 
This allows LLVM codegen passes to insert `inttoptr` and `ptrtoint` as it wish,
and avoids hitting any illegal ones in those passes.

Fix #36062

(cherry picked from commit 235784a)
simeonschaub pushed a commit to simeonschaub/julia that referenced this issue Aug 11, 2020
@yuyichao @simeonschaub
Remove non-integral pointer from data layout before codegen (JuliaLan…
05a2af2 
…g#36705)

This allows LLVM codegen passes to insert `inttoptr` and `ptrtoint` as it wish,
and avoids hitting any illegal ones in those passes.

Fix JuliaLang#36062
KristofferC pushed a commit that referenced this issue Aug 18, 2020
@yuyichao @KristofferC
Remove non-integral pointer from data layout before codegen (#36705)
511a53f 
This allows LLVM codegen passes to insert `inttoptr` and `ptrtoint` as it wish,
and avoids hitting any illegal ones in those passes.

Fix #36062

(cherry picked from commit 235784a)
KristofferC pushed a commit that referenced this issue Aug 19, 2020
@yuyichao @KristofferC
Remove non-integral pointer from data layout before codegen (#36705)
27408bc 
This allows LLVM codegen passes to insert `inttoptr` and `ptrtoint` as it wish,
and avoids hitting any illegal ones in those passes.

Fix #36062

(cherry picked from commit 235784a)
vchuravy added a commit to JuliaPackaging/Yggdrasil that referenced this issue Aug 22, 2020
@vchuravy
[LLVM] add patches for JuliaLang/julia#36062
a68722c
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
kind:bug Indicates an unexpected problem or unintended behavior kind:upstream The issue is with an upstream dependency, e.g. LLVM system:arm ARMv7 and AArch64 system:32-bit Affects only 32-bit systems
Projects
None yet
Milestone
No milestone
Development

Successfully merging a pull request may close this issue.

Strip non-integral address space from datalayout after GC lowering JuliaLang/julia
7 participants
@staticfloat @mbauman @yuyichao @ViralBShah @giordano @Keno @IanButterworth