[AMDGPU][SDAG ISel] clang/llc hangs on rocPRIM adjacent_difference kernel at gfx950 -O3

## Summary

clang 23.0.0git (and standalone `llc`) hangs indefinitely in **AMDGPU DAG->DAG Pattern Instruction Selection** when compiling a rocPRIM `adjacent_difference_impl` trampoline kernel for `gfx950` at `-O3`. The compiler thread sits at 100% CPU with no output. The same IR compiles in ROCm 7.2 (older LLVM 23-era amd-staging) and hangs starting in ROCm 7.13.

Original downstream tracking: ROCM-24721. Reproduced cleanly on a workstation with the publicly available `rocm-sdk-devel==7.13.0a20260515` wheel.

## Affected versions

- **AMD clang 23.0.0git** as shipped in `rocm-sdk-devel==7.13.0a20260515`
  - upstream SHA: `43215c73116c407735c85a180d174f718798c328`
  - AMD patch SHA: `2506c552d8428e2cc1778bef048b20f818e06bb3`
- Also reproduced with a local `amd-staging`-derived `llc` (AOMP-23.0-60, SHA `8b7c0c42edfe`) — so the bug is in shared/upstream AMDGPU codegen, not unique to any one downstream patch set.

## Known good vs known bad

| State | Branch / wheel | Upstream merge-base | Date |
|---|---|---|---|
| :white_check_mark: works | `release/rocm-rel-7.2` | `c793782b03ab` | 2025-10-05 |
| :x: hangs | `rocm-sdk-devel==7.13.0a20260515` (per-family `gfx950-dcgpu` index) | `5b1eed6278de` | 2026-03-12 |

Bisect window is roughly 5 months of upstream commits plus the AMD patches that landed on `amd-staging` between those merge-bases. (Not bisected to a single commit yet — see "What we tried".)

## Reproducer

### Setup (no GPU required — compile-only)

```bash
python3 -m venv ~/.rocm-7.13-repro
source ~/.rocm-7.13-repro/bin/activate
pip install --index-url https://rocm.nightlies.amd.com/v2/gfx950-dcgpu/ \
  'rocm[libraries,devel]==7.13.0a20260515'
```

### Reproduce with `llc`

Save the IR below as `reduced.ll`, then:

```bash
llc -O3 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx950 reduced.ll -o /tmp/out.s
# Hangs forever at 100% CPU. Kill with Ctrl-C.
```

### Reproduce end-to-end with clang (matches original repro)

```bash
git clone https://github.com/ROCm/rocm-libraries.git --branch therock-7.13 --depth=1
cd rocm-libraries/projects/rocprim
SDK=$(rocm-sdk path --root)
mkdir build && cd build
PATH="$SDK/bin:$PATH" cmake \
  -DCMAKE_PREFIX_PATH="$(rocm-sdk path --cmake)" \
  -DCMAKE_CXX_COMPILER="$SDK/bin/hipcc" \
  -DBUILD_BENCHMARK=ON -DBUILD_TEST=OFF -DBUILD_EXAMPLE=OFF \
  -DAMDGPU_TARGETS=gfx950 ..
make benchmark_device_adjacent_difference
# Hangs at 100% CPU on the single TU, never returns.
```

## Diagnostic output

With `-mllvm -debug-pass=Executions`, the last pass entered before the hang is:

```
Executing Pass 'AMDGPU DAG->DAG Pattern Instruction Selection' on Function
  '_ZN7rocprim..._trampoline_kernel...
   adjacent_difference_impl<..., true, true, short*, short*, plus<void>>...
   gfx950 wavefront target...'
```

i.e. the rocPRIM `adjacent_difference_impl` trampoline kernel instantiated for `short*` + `plus<void>` at `gfx950`.

`ps` confirms `clang-23` is pinned at 100% CPU; the compile completes through the rest of the TU when codegen is bypassed (`-emit-llvm` succeeds in ~9 s on the same TU). So the hang is purely in SDAG ISel for this one function.

## What we tried

- **Hypothesis (rejected):** the new combine #177370 (`5d018e93fe98 AMDGPU: Perform zero/any extend combine into permute`) looked suspicious — it adds `ISD::ANY_EXTEND` to the DAG-combine list with an `AfterLegalizeTypes` threshold, creates `AMDGPUISD::PERM` nodes, and lands inside the affected window (Jan 28 2026). Reverting it from the source tree and rebuilding `llc` did **not** fix the hang. So either the trigger is a different commit, or it's a multi-commit interaction.

## Reduction process used

1. Captured exact `clang -cc1` invocation from the rocPRIM build (`hipcc -###`).
2. Replaced `-emit-obj` with `-emit-llvm` → IR dumped in 9 s (so the hang is purely codegen).
3. `llvm-extract --func=… --recursive` reduced 602 functions → 1.
4. `llvm-reduce --test=<hangs.sh>` shrunk the single function from 1316 → 505 lines.
5. Verified `llc` still hangs on the final 505-line IR.

## Environment

- Host: Linux 6.8 (Ubuntu 24.04), `x86_64`
- No AMD GPU required on the build host — pure codegen hang.

## Reduced IR

Save the block below as `reduced.ll` and run:

```bash
llc -O2 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx950 reduced.ll -o /tmp/out.s
# Hangs forever at 100% CPU. Kill with Ctrl-C. (llc must be built with the suspect commit in its source.)
```

<details>
<summary>reduced.ll (505 lines, single function, 412 instructions)</summary>

```llvm
; ModuleID = '/tmp/single_func.ll'
source_filename = "/home/lambj/git/repro-rocprim-7.13/projects/rocprim/benchmark/benchmark_device_adjacent_difference.cpp"
target datalayout = "e-m:e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-p7:160:256:256:32-p8:128:128:128:48-p9:192:256:256:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5-G1-ni:7:8:9"
target triple = "amdgcn-amd-amdhsa"

; Function Attrs: convergent nocallback nofree nounwind willreturn
declare void @llvm.amdgcn.s.barrier() #0

; Function Attrs: convergent mustprogress nofree norecurse nounwind willreturn
define amdgpu_kernel void @_ZN7rocprim17ROCPRIM_400400_NS6detail17trampoline_kernelINS0_14default_configENS1_35adjacent_difference_config_selectorILb1EsEEZNS1_24adjacent_difference_implIS3_Lb1ELb1EPsS7_NS0_4plusIvEEEE10hipError_tPvRmT2_T3_mT4_P12ihipStream_tbEUlT_E_NS1_11comp_targetILNS1_3genE0ELNS1_11target_archE4294967295ELNS1_3gpuE0ELNS1_3repE0EEENS1_30default_config_static_selectorELNS0_4arch9wavefront6targetE1EEEvT1_() #1 {
  %1 = addrspacecast ptr null to ptr addrspace(1)
  %2 = load i64, ptr addrspace(4) null, align 8
  %3 = load ptr, ptr addrspace(4) null, align 8
  %4 = addrspacecast ptr %3 to ptr addrspace(1)
  %5 = load i64, ptr addrspace(4) null, align 8
  %6 = load ptr, ptr addrspace(4) null, align 8
  %7 = addrspacecast ptr %6 to ptr addrspace(1)
  %8 = load i64, ptr addrspace(4) null, align 8
  %9 = getelementptr inbounds [2 x i8], ptr addrspace(1) %1, i64 %2
  %10 = getelementptr inbounds [2 x i8], ptr addrspace(1) %4, i64 %2
  %11 = getelementptr inbounds [2 x i8], ptr addrspace(1) %7, i64 %8
  %12 = tail call noundef i32 @llvm.amdgcn.workgroup.id.x()
  %13 = mul i32 %12, 736
  %14 = freeze i64 %5
  %15 = udiv i64 %14, 736
  %16 = mul i64 %15, 736
  %17 = sub i64 %14, %16
  %18 = icmp ne i64 %17, 0
  %19 = zext i1 %18 to i64
  %20 = zext i32 %12 to i64
  %21 = add i64 %8, %20
  %22 = add nsw i64 %15, -1
  %23 = add nsw i64 %22, %19
  %24 = icmp ult i64 %21, %23
  %25 = zext i32 %13 to i64
  %26 = getelementptr inbounds nuw [2 x i8], ptr addrspace(1) %9, i64 %25
  %27 = tail call noundef range(i32 0, 1024) i32 @llvm.amdgcn.workitem.id.x()
  %28 = zext nneg i32 %27 to i64
  %29 = getelementptr inbounds nuw [2 x i8], ptr addrspace(1) %26, i64 %28
  %30 = load i16, ptr addrspace(1) %29, align 2
  %31 = getelementptr inbounds nuw i8, ptr addrspace(1) %29, i64 64
  %32 = load i16, ptr addrspace(1) %31, align 2
  %33 = getelementptr inbounds nuw i8, ptr addrspace(1) %29, i64 128
  %34 = load i16, ptr addrspace(1) %33, align 2
  %35 = getelementptr inbounds nuw i8, ptr addrspace(1) %29, i64 192
  %36 = load i16, ptr addrspace(1) %35, align 2
  %37 = getelementptr inbounds nuw i8, ptr addrspace(1) %29, i64 256
  %38 = load i16, ptr addrspace(1) %37, align 2
  %39 = getelementptr inbounds nuw i8, ptr addrspace(1) %29, i64 320
  %40 = load i16, ptr addrspace(1) %39, align 2
  %41 = getelementptr inbounds nuw i8, ptr addrspace(1) %29, i64 384
  %42 = load i16, ptr addrspace(1) %41, align 2
  %43 = getelementptr inbounds nuw i8, ptr addrspace(1) %29, i64 448
  %44 = load i16, ptr addrspace(1) %43, align 2
  %45 = getelementptr inbounds nuw i8, ptr addrspace(1) %29, i64 512
  %46 = load i16, ptr addrspace(1) %45, align 2
  %47 = getelementptr inbounds nuw i8, ptr addrspace(1) %29, i64 576
  %48 = load i16, ptr addrspace(1) %47, align 2
  %49 = getelementptr inbounds nuw i8, ptr addrspace(1) %29, i64 640
  %50 = load i16, ptr addrspace(1) %49, align 2
  %51 = getelementptr inbounds nuw i8, ptr addrspace(1) %29, i64 704
  %52 = load i16, ptr addrspace(1) %51, align 2
  %53 = getelementptr inbounds nuw i8, ptr addrspace(1) %29, i64 768
  %54 = load i16, ptr addrspace(1) %53, align 2
  %55 = getelementptr inbounds nuw i8, ptr addrspace(1) %29, i64 832
  %56 = load i16, ptr addrspace(1) %55, align 2
  %57 = getelementptr inbounds nuw i8, ptr addrspace(1) %29, i64 896
  %58 = load i16, ptr addrspace(1) %57, align 2
  %59 = getelementptr inbounds nuw i8, ptr addrspace(1) %29, i64 960
  %60 = load i16, ptr addrspace(1) %59, align 2
  %61 = getelementptr inbounds nuw i8, ptr addrspace(1) %29, i64 1024
  %62 = load i16, ptr addrspace(1) %61, align 2
  %63 = getelementptr inbounds nuw i8, ptr addrspace(1) %29, i64 1088
  %64 = load i16, ptr addrspace(1) %63, align 2
  %65 = getelementptr inbounds nuw i8, ptr addrspace(1) %29, i64 1152
  %66 = load i16, ptr addrspace(1) %65, align 2
  %67 = getelementptr inbounds nuw i8, ptr addrspace(1) %29, i64 1216
  %68 = load i16, ptr addrspace(1) %67, align 2
  %69 = getelementptr inbounds nuw i8, ptr addrspace(1) %29, i64 1280
  %70 = load i16, ptr addrspace(1) %69, align 2
  %71 = getelementptr inbounds nuw i8, ptr addrspace(1) %29, i64 1344
  %72 = load i16, ptr addrspace(1) %71, align 2
  %73 = getelementptr inbounds nuw i8, ptr addrspace(1) %29, i64 1408
  %74 = load i16, ptr addrspace(1) %73, align 2
  %75 = getelementptr inbounds nuw [2 x i8], ptr addrspace(3) null, i32 %27
  store i16 %30, ptr addrspace(3) %75, align 2
  %76 = getelementptr inbounds nuw i8, ptr addrspace(3) %75, i32 64
  store i16 %32, ptr addrspace(3) %76, align 2
  %77 = getelementptr inbounds nuw i8, ptr addrspace(3) %75, i32 128
  store i16 %34, ptr addrspace(3) %77, align 2
  %78 = getelementptr inbounds nuw i8, ptr addrspace(3) %75, i32 192
  store i16 %36, ptr addrspace(3) %78, align 2
  %79 = getelementptr inbounds nuw i8, ptr addrspace(3) %75, i32 256
  store i16 %38, ptr addrspace(3) %79, align 2
  %80 = getelementptr inbounds nuw i8, ptr addrspace(3) %75, i32 320
  store i16 %40, ptr addrspace(3) %80, align 2
  %81 = getelementptr inbounds nuw i8, ptr addrspace(3) %75, i32 384
  store i16 %42, ptr addrspace(3) %81, align 2
  %82 = getelementptr inbounds nuw i8, ptr addrspace(3) %75, i32 448
  store i16 %44, ptr addrspace(3) %82, align 2
  %83 = getelementptr inbounds nuw i8, ptr addrspace(3) %75, i32 512
  store i16 %46, ptr addrspace(3) %83, align 2
  %84 = getelementptr inbounds nuw i8, ptr addrspace(3) %75, i32 576
  store i16 %48, ptr addrspace(3) %84, align 2
  %85 = getelementptr inbounds nuw i8, ptr addrspace(3) %75, i32 640
  store i16 %50, ptr addrspace(3) %85, align 2
  %86 = getelementptr inbounds nuw i8, ptr addrspace(3) %75, i32 704
  store i16 %52, ptr addrspace(3) %86, align 2
  %87 = getelementptr inbounds nuw i8, ptr addrspace(3) %75, i32 768
  store i16 %54, ptr addrspace(3) %87, align 2
  %88 = getelementptr inbounds nuw i8, ptr addrspace(3) %75, i32 832
  store i16 %56, ptr addrspace(3) %88, align 2
  %89 = getelementptr inbounds nuw i8, ptr addrspace(3) %75, i32 896
  store i16 %58, ptr addrspace(3) %89, align 2
  %90 = getelementptr inbounds nuw i8, ptr addrspace(3) %75, i32 960
  store i16 %60, ptr addrspace(3) %90, align 2
  %91 = getelementptr inbounds nuw i8, ptr addrspace(3) %75, i32 1024
  store i16 %62, ptr addrspace(3) %91, align 2
  %92 = getelementptr inbounds nuw i8, ptr addrspace(3) %75, i32 1088
  store i16 %64, ptr addrspace(3) %92, align 2
  %93 = getelementptr inbounds nuw i8, ptr addrspace(3) %75, i32 1152
  store i16 %66, ptr addrspace(3) %93, align 2
  %94 = getelementptr inbounds nuw i8, ptr addrspace(3) %75, i32 1216
  store i16 %68, ptr addrspace(3) %94, align 2
  %95 = getelementptr inbounds nuw i8, ptr addrspace(3) %75, i32 1280
  store i16 %70, ptr addrspace(3) %95, align 2
  %96 = getelementptr inbounds nuw i8, ptr addrspace(3) %75, i32 1344
  store i16 %72, ptr addrspace(3) %96, align 2
  %97 = getelementptr inbounds nuw i8, ptr addrspace(3) %75, i32 1408
  store i16 %74, ptr addrspace(3) %97, align 2
  fence syncscope("workgroup") release
  tail call void @llvm.amdgcn.s.barrier()
  fence syncscope("workgroup") acquire
  %98 = mul nuw nsw i32 %27, 46
  %99 = getelementptr inbounds nuw i8, ptr addrspace(3) null, i32 %98
  %100 = getelementptr inbounds nuw i8, ptr addrspace(3) %99, i32 2
  %101 = load <2 x i16>, ptr addrspace(3) %99, align 2
  %102 = load <2 x i16>, ptr addrspace(3) %100, align 2
  %103 = getelementptr inbounds nuw i8, ptr addrspace(3) %99, i32 6
  %104 = load <2 x i16>, ptr addrspace(3) %103, align 2
  %105 = getelementptr inbounds nuw i8, ptr addrspace(3) %99, i32 10
  %106 = load <2 x i16>, ptr addrspace(3) %105, align 2
  %107 = getelementptr inbounds nuw i8, ptr addrspace(3) %99, i32 14
  %108 = load <2 x i16>, ptr addrspace(3) %107, align 2
  %109 = getelementptr inbounds nuw i8, ptr addrspace(3) %99, i32 18
  %110 = load <2 x i16>, ptr addrspace(3) %109, align 2
  %111 = getelementptr inbounds nuw i8, ptr addrspace(3) %99, i32 22
  %112 = load <2 x i16>, ptr addrspace(3) %111, align 2
  %113 = getelementptr inbounds nuw i8, ptr addrspace(3) %99, i32 26
  %114 = load <2 x i16>, ptr addrspace(3) %113, align 2
  %115 = getelementptr inbounds nuw i8, ptr addrspace(3) %99, i32 30
  %116 = load <2 x i16>, ptr addrspace(3) %115, align 2
  %117 = getelementptr inbounds nuw i8, ptr addrspace(3) %99, i32 34
  %118 = load <2 x i16>, ptr addrspace(3) %117, align 2
  %119 = getelementptr inbounds nuw i8, ptr addrspace(3) %99, i32 38
  %120 = load <2 x i16>, ptr addrspace(3) %119, align 2
  %121 = getelementptr inbounds nuw i8, ptr addrspace(3) %99, i32 42
  %122 = load <2 x i16>, ptr addrspace(3) %121, align 2
  fence syncscope("workgroup") release
  tail call void @llvm.amdgcn.s.barrier()
  fence syncscope("workgroup") acquire
  %123 = icmp eq i64 %21, %23
  %124 = shufflevector <2 x i16> %101, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %125 = mul i64 %21, 4294966560
  %126 = add i64 %125, %5
  %127 = trunc i64 %126 to i32
  %128 = tail call noundef range(i32 0, 1024) i32 @llvm.amdgcn.workitem.id.x()
  %129 = getelementptr inbounds nuw [2 x i8], ptr addrspace(3) getelementptr inbounds nuw (i8, ptr addrspace(3) null, i32 64), i32 %128
  %130 = extractelement <2 x i16> %101, i64 0
  store i16 %130, ptr addrspace(3) %129, align 2
  %131 = mul nuw nsw i32 %128, 23
  %132 = add nuw nsw i32 %131, 1
  %133 = shufflevector <23 x i16> <i16 0, i16 poison, i16 poison, i16 poison, i16 poison, i16 poison, i16 poison, i16 poison, i16 poison, i16 poison, i16 poison, i16 poison, i16 poison, i16 poison, i16 poison, i16 poison, i16 poison, i16 poison, i16 poison, i16 poison, i16 poison, i16 poison, i16 poison>, <23 x i16> %124, <23 x i32> <i32 23, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %134 = icmp ult i32 %132, %127
  %135 = add <2 x i16> %102, %101
  %136 = shufflevector <2 x i16> %135, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %137 = select i1 %134, <23 x i16> %136, <23 x i16> %133
  %138 = shufflevector <23 x i16> %137, <23 x i16> %133, <23 x i32> <i32 0, i32 poison, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45>
  %139 = add nuw nsw i32 %131, 2
  %140 = shufflevector <2 x i16> %102, <2 x i16> poison, <23 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %141 = shufflevector <23 x i16> %138, <23 x i16> %140, <23 x i32> <i32 0, i32 23, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %142 = icmp ult i32 %139, %127
  %143 = shufflevector <2 x i16> %102, <2 x i16> poison, <2 x i32> <i32 1, i32 poison>
  %144 = add <2 x i16> %143, %102
  %145 = shufflevector <2 x i16> %144, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %146 = shufflevector <23 x i16> %141, <23 x i16> %145, <23 x i32> <i32 0, i32 23, i32 poison, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %147 = select i1 %142, <23 x i16> %146, <23 x i16> %141
  %148 = add nuw nsw i32 %131, 3
  %149 = shufflevector <23 x i16> %147, <23 x i16> %140, <23 x i32> <i32 0, i32 1, i32 24, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %150 = icmp ult i32 %148, %127
  %151 = shufflevector <2 x i16> %102, <2 x i16> poison, <2 x i32> <i32 1, i32 poison>
  %152 = add <2 x i16> %104, %151
  %153 = shufflevector <2 x i16> %152, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %154 = shufflevector <23 x i16> %149, <23 x i16> %153, <23 x i32> <i32 0, i32 1, i32 23, i32 poison, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %155 = select i1 %150, <23 x i16> %154, <23 x i16> %149
  %156 = add nuw nsw i32 %131, 4
  %157 = shufflevector <2 x i16> %104, <2 x i16> poison, <23 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %158 = shufflevector <23 x i16> %155, <23 x i16> %157, <23 x i32> <i32 0, i32 1, i32 2, i32 23, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %159 = icmp ult i32 %156, %127
  %160 = shufflevector <2 x i16> %104, <2 x i16> poison, <2 x i32> <i32 1, i32 poison>
  %161 = add <2 x i16> %160, %104
  %162 = shufflevector <2 x i16> %161, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %163 = shufflevector <23 x i16> %158, <23 x i16> %162, <23 x i32> <i32 0, i32 1, i32 2, i32 23, i32 poison, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %164 = select i1 %159, <23 x i16> %163, <23 x i16> %158
  %165 = add nuw nsw i32 %131, 5
  %166 = shufflevector <23 x i16> %164, <23 x i16> %157, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 24, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %167 = icmp ult i32 %165, %127
  %168 = shufflevector <2 x i16> %104, <2 x i16> poison, <2 x i32> <i32 1, i32 poison>
  %169 = add <2 x i16> %106, %168
  %170 = shufflevector <2 x i16> %169, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %171 = shufflevector <23 x i16> %166, <23 x i16> %170, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 23, i32 poison, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %172 = select i1 %167, <23 x i16> %171, <23 x i16> %166
  %173 = add nuw nsw i32 %131, 6
  %174 = shufflevector <2 x i16> %106, <2 x i16> poison, <23 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %175 = shufflevector <23 x i16> %172, <23 x i16> %174, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 23, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %176 = icmp ult i32 %173, %127
  %177 = shufflevector <2 x i16> %106, <2 x i16> poison, <2 x i32> <i32 1, i32 poison>
  %178 = add <2 x i16> %177, %106
  %179 = shufflevector <2 x i16> %178, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %180 = shufflevector <23 x i16> %175, <23 x i16> %179, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 23, i32 poison, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %181 = select i1 %176, <23 x i16> %180, <23 x i16> %175
  %182 = add nuw nsw i32 %131, 7
  %183 = shufflevector <23 x i16> %181, <23 x i16> %174, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 24, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %184 = icmp ult i32 %182, %127
  %185 = shufflevector <2 x i16> %106, <2 x i16> poison, <2 x i32> <i32 1, i32 poison>
  %186 = add <2 x i16> %108, %185
  %187 = shufflevector <2 x i16> %186, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %188 = shufflevector <23 x i16> %183, <23 x i16> %187, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 23, i32 poison, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %189 = select i1 %184, <23 x i16> %188, <23 x i16> %183
  %190 = add nuw nsw i32 %131, 8
  %191 = shufflevector <2 x i16> %108, <2 x i16> poison, <23 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %192 = shufflevector <23 x i16> %189, <23 x i16> %191, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 23, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %193 = icmp ult i32 %190, %127
  %194 = shufflevector <2 x i16> %108, <2 x i16> poison, <2 x i32> <i32 1, i32 poison>
  %195 = add <2 x i16> %194, %108
  %196 = shufflevector <2 x i16> %195, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %197 = shufflevector <23 x i16> %192, <23 x i16> %196, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 23, i32 poison, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %198 = select i1 %193, <23 x i16> %197, <23 x i16> %192
  %199 = add nuw nsw i32 %131, 9
  %200 = shufflevector <23 x i16> %198, <23 x i16> %191, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 24, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %201 = icmp ult i32 %199, %127
  %202 = shufflevector <2 x i16> %108, <2 x i16> poison, <2 x i32> <i32 1, i32 poison>
  %203 = add <2 x i16> %110, %202
  %204 = shufflevector <2 x i16> %203, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %205 = shufflevector <23 x i16> %200, <23 x i16> %204, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 23, i32 poison, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %206 = select i1 %201, <23 x i16> %205, <23 x i16> %200
  %207 = add nuw nsw i32 %131, 10
  %208 = shufflevector <2 x i16> %110, <2 x i16> poison, <23 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %209 = shufflevector <23 x i16> %206, <23 x i16> %208, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 23, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %210 = icmp ult i32 %207, %127
  %211 = shufflevector <2 x i16> %110, <2 x i16> poison, <2 x i32> <i32 1, i32 poison>
  %212 = add <2 x i16> %211, %110
  %213 = shufflevector <2 x i16> %212, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %214 = shufflevector <23 x i16> %209, <23 x i16> %213, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 23, i32 poison, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %215 = select i1 %210, <23 x i16> %214, <23 x i16> %209
  %216 = add nuw nsw i32 %131, 11
  %217 = shufflevector <23 x i16> %215, <23 x i16> %208, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 24, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %218 = icmp ult i32 %216, %127
  %219 = shufflevector <2 x i16> %110, <2 x i16> poison, <2 x i32> <i32 1, i32 poison>
  %220 = add <2 x i16> %112, %219
  %221 = shufflevector <2 x i16> %220, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %222 = shufflevector <23 x i16> %217, <23 x i16> %221, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 23, i32 poison, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %223 = select i1 %218, <23 x i16> %222, <23 x i16> %217
  %224 = add nuw nsw i32 %131, 12
  %225 = shufflevector <2 x i16> %112, <2 x i16> poison, <23 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %226 = shufflevector <23 x i16> %223, <23 x i16> %225, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 23, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %227 = icmp ult i32 %224, %127
  %228 = shufflevector <2 x i16> %112, <2 x i16> poison, <2 x i32> <i32 1, i32 poison>
  %229 = add <2 x i16> %228, %112
  %230 = shufflevector <2 x i16> %229, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %231 = shufflevector <23 x i16> %226, <23 x i16> %230, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 23, i32 poison, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %232 = select i1 %227, <23 x i16> %231, <23 x i16> %226
  %233 = add nuw nsw i32 %131, 13
  %234 = shufflevector <23 x i16> %232, <23 x i16> %225, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 24, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %235 = icmp ult i32 %233, %127
  %236 = shufflevector <2 x i16> %112, <2 x i16> poison, <2 x i32> <i32 1, i32 poison>
  %237 = add <2 x i16> %114, %236
  %238 = shufflevector <2 x i16> %237, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %239 = shufflevector <23 x i16> %234, <23 x i16> %238, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 23, i32 poison, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %240 = select i1 %235, <23 x i16> %239, <23 x i16> %234
  %241 = add nuw nsw i32 %131, 14
  %242 = shufflevector <2 x i16> %114, <2 x i16> poison, <23 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %243 = shufflevector <23 x i16> %240, <23 x i16> %242, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 23, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %244 = icmp ult i32 %241, %127
  %245 = shufflevector <2 x i16> %114, <2 x i16> poison, <2 x i32> <i32 1, i32 poison>
  %246 = add <2 x i16> %245, %114
  %247 = shufflevector <2 x i16> %246, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %248 = shufflevector <23 x i16> %243, <23 x i16> %247, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 23, i32 poison, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %249 = select i1 %244, <23 x i16> %248, <23 x i16> %243
  %250 = add nuw nsw i32 %131, 15
  %251 = shufflevector <23 x i16> %249, <23 x i16> %242, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 24, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %252 = icmp ult i32 %250, %127
  %253 = shufflevector <2 x i16> %114, <2 x i16> poison, <2 x i32> <i32 1, i32 poison>
  %254 = add <2 x i16> %116, %253
  %255 = shufflevector <2 x i16> %254, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %256 = shufflevector <23 x i16> %251, <23 x i16> %255, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 23, i32 poison, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %257 = select i1 %252, <23 x i16> %256, <23 x i16> %251
  %258 = add nuw nsw i32 %131, 16
  %259 = shufflevector <2 x i16> %116, <2 x i16> poison, <23 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %260 = shufflevector <23 x i16> %257, <23 x i16> %259, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 23, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %261 = icmp ult i32 %258, %127
  %262 = shufflevector <2 x i16> %116, <2 x i16> poison, <2 x i32> <i32 1, i32 poison>
  %263 = add <2 x i16> %262, %116
  %264 = shufflevector <2 x i16> %263, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %265 = shufflevector <23 x i16> %260, <23 x i16> %264, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 23, i32 poison, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %266 = select i1 %261, <23 x i16> %265, <23 x i16> %260
  %267 = add nuw nsw i32 %131, 17
  %268 = shufflevector <23 x i16> %266, <23 x i16> %259, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 24, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
  %269 = icmp ult i32 %267, %127
  %270 = shufflevector <2 x i16> %116, <2 x i16> poison, <2 x i32> <i32 1, i32 poison>
  %271 = add <2 x i16> %118, %270
  %272 = shufflevector <2 x i16> %271, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %273 = shufflevector <23 x i16> %268, <23 x i16> %272, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 23, i32 poison, i32 18, i32 19, i32 20, i32 21, i32 22>
  %274 = select i1 %269, <23 x i16> %273, <23 x i16> %268
  %275 = add nuw nsw i32 %131, 18
  %276 = shufflevector <2 x i16> %118, <2 x i16> poison, <23 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %277 = shufflevector <23 x i16> %274, <23 x i16> %276, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 23, i32 18, i32 19, i32 20, i32 21, i32 22>
  %278 = icmp ult i32 %275, %127
  %279 = shufflevector <2 x i16> %118, <2 x i16> poison, <2 x i32> <i32 1, i32 poison>
  %280 = add <2 x i16> %279, %118
  %281 = shufflevector <2 x i16> %280, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %282 = shufflevector <23 x i16> %277, <23 x i16> %281, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 23, i32 poison, i32 19, i32 20, i32 21, i32 22>
  %283 = select i1 %278, <23 x i16> %282, <23 x i16> %277
  %284 = add nuw nsw i32 %131, 19
  %285 = shufflevector <23 x i16> %283, <23 x i16> %276, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 24, i32 19, i32 20, i32 21, i32 22>
  %286 = icmp ult i32 %284, %127
  %287 = shufflevector <2 x i16> %118, <2 x i16> poison, <2 x i32> <i32 1, i32 poison>
  %288 = add <2 x i16> %120, %287
  %289 = shufflevector <2 x i16> %288, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %290 = shufflevector <23 x i16> %285, <23 x i16> %289, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 23, i32 poison, i32 20, i32 21, i32 22>
  %291 = select i1 %286, <23 x i16> %290, <23 x i16> %285
  %292 = add nuw nsw i32 %131, 20
  %293 = shufflevector <2 x i16> %120, <2 x i16> poison, <23 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %294 = shufflevector <23 x i16> %291, <23 x i16> %293, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 23, i32 20, i32 21, i32 22>
  %295 = icmp ult i32 %292, %127
  %296 = shufflevector <2 x i16> %120, <2 x i16> poison, <2 x i32> <i32 1, i32 poison>
  %297 = add <2 x i16> %296, %120
  %298 = shufflevector <2 x i16> %297, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %299 = shufflevector <23 x i16> %294, <23 x i16> %298, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 23, i32 poison, i32 21, i32 22>
  %300 = select i1 %295, <23 x i16> %299, <23 x i16> %294
  %301 = add nuw nsw i32 %131, 21
  %302 = shufflevector <23 x i16> %300, <23 x i16> %293, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 24, i32 21, i32 22>
  %303 = icmp ult i32 %301, %127
  %304 = shufflevector <2 x i16> %120, <2 x i16> poison, <2 x i32> <i32 1, i32 poison>
  %305 = add <2 x i16> %122, %304
  %306 = shufflevector <2 x i16> %305, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %307 = shufflevector <23 x i16> %302, <23 x i16> %306, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 23, i32 poison, i32 22>
  %308 = select i1 %303, <23 x i16> %307, <23 x i16> %302
  %309 = add nuw nsw i32 %131, 22
  %310 = shufflevector <2 x i16> %122, <2 x i16> poison, <23 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %311 = shufflevector <23 x i16> %308, <23 x i16> %310, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 23, i32 22>
  %312 = icmp ult i32 %309, %127
  %313 = shufflevector <2 x i16> %122, <2 x i16> poison, <2 x i32> <i32 1, i32 poison>
  %314 = add <2 x i16> %313, %122
  %315 = shufflevector <2 x i16> %314, <2 x i16> poison, <23 x i32> <i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %316 = shufflevector <23 x i16> %311, <23 x i16> %315, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 23, i32 poison>
  %317 = select i1 %312, <23 x i16> %316, <23 x i16> %311
  fence syncscope("workgroup") release
  tail call void @llvm.amdgcn.s.barrier()
  fence syncscope("workgroup") acquire
  %318 = shufflevector <23 x i16> %317, <23 x i16> %310, <23 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 24>
  %319 = add nuw nsw i32 %131, 23
  %320 = icmp ult i32 %319, %127
  %321 = extractelement <2 x i16> %122, i64 1
  %322 = getelementptr inbounds nuw i8, ptr addrspace(3) %129, i32 2
  %323 = load i16, ptr addrspace(3) %322, align 2
  %324 = add i16 %323, %321
  %325 = insertelement <23 x i16> %318, i16 %324, i64 22
  fence syncscope("workgroup") release
  tail call void @llvm.amdgcn.s.barrier()
  fence syncscope("workgroup") acquire
  %326 = zext i32 %13 to i64
  %327 = getelementptr inbounds nuw [2 x i8], ptr addrspace(1) %10, i64 %326
  %328 = tail call noundef range(i32 0, 1024) i32 @llvm.amdgcn.workitem.id.x()
  %329 = mul nuw nsw i32 %328, 46
  %330 = getelementptr inbounds nuw i8, ptr addrspace(3) null, i32 %329
  %331 = shufflevector <23 x i16> %325, <23 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  store <4 x i16> %331, ptr addrspace(3) %330, align 2
  %332 = getelementptr inbounds nuw i8, ptr addrspace(3) %330, i32 8
  %333 = shufflevector <23 x i16> %325, <23 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
  store <4 x i16> %333, ptr addrspace(3) %332, align 2
  %334 = getelementptr inbounds nuw i8, ptr addrspace(3) %330, i32 16
  %335 = shufflevector <23 x i16> %325, <23 x i16> poison, <4 x i32> <i32 8, i32 9, i32 10, i32 11>
  store <4 x i16> %335, ptr addrspace(3) %334, align 2
  %336 = getelementptr inbounds nuw i8, ptr addrspace(3) %330, i32 24
  %337 = shufflevector <23 x i16> %325, <23 x i16> poison, <4 x i32> <i32 12, i32 13, i32 14, i32 15>
  store <4 x i16> %337, ptr addrspace(3) %336, align 2
  %338 = getelementptr inbounds nuw i8, ptr addrspace(3) %330, i32 32
  %339 = shufflevector <23 x i16> %325, <23 x i16> poison, <4 x i32> <i32 16, i32 17, i32 18, i32 19>
  store <4 x i16> %339, ptr addrspace(3) %338, align 2
  %340 = getelementptr inbounds nuw i8, ptr addrspace(3) %330, i32 40
  %341 = shufflevector <23 x i16> %325, <23 x i16> poison, <2 x i32> <i32 20, i32 21>
  store <2 x i16> %341, ptr addrspace(3) %340, align 2
  %342 = extractelement <23 x i16> %325, i64 22
  %343 = getelementptr inbounds nuw i8, ptr addrspace(3) %330, i32 44
  store i16 %342, ptr addrspace(3) %343, align 2
  fence syncscope("workgroup") release
  tail call void @llvm.amdgcn.s.barrier()
  fence syncscope("workgroup") acquire
  %344 = zext nneg i32 %328 to i64
  %345 = getelementptr inbounds nuw [2 x i8], ptr addrspace(3) null, i32 %328
  %346 = load i16, ptr addrspace(3) %345, align 2
  %347 = getelementptr inbounds nuw i8, ptr addrspace(3) %345, i32 64
  %348 = load i16, ptr addrspace(3) %347, align 2
  %349 = getelementptr inbounds nuw i8, ptr addrspace(3) %345, i32 128
  %350 = load i16, ptr addrspace(3) %349, align 2
  %351 = getelementptr inbounds nuw i8, ptr addrspace(3) %345, i32 192
  %352 = load i16, ptr addrspace(3) %351, align 2
  %353 = getelementptr inbounds nuw i8, ptr addrspace(3) %345, i32 256
  %354 = load i16, ptr addrspace(3) %353, align 2
  %355 = getelementptr inbounds nuw i8, ptr addrspace(3) %345, i32 320
  %356 = load i16, ptr addrspace(3) %355, align 2
  %357 = getelementptr inbounds nuw i8, ptr addrspace(3) %345, i32 384
  %358 = load i16, ptr addrspace(3) %357, align 2
  %359 = getelementptr inbounds nuw i8, ptr addrspace(3) %345, i32 448
  %360 = load i16, ptr addrspace(3) %359, align 2
  %361 = getelementptr inbounds nuw i8, ptr addrspace(3) %345, i32 512
  %362 = load i16, ptr addrspace(3) %361, align 2
  %363 = getelementptr inbounds nuw i8, ptr addrspace(3) %345, i32 576
  %364 = load i16, ptr addrspace(3) %363, align 2
  %365 = getelementptr inbounds nuw i8, ptr addrspace(3) %345, i32 640
  %366 = load i16, ptr addrspace(3) %365, align 2
  %367 = getelementptr inbounds nuw i8, ptr addrspace(3) %345, i32 704
  %368 = load i16, ptr addrspace(3) %367, align 2
  %369 = getelementptr inbounds nuw i8, ptr addrspace(3) %345, i32 768
  %370 = load i16, ptr addrspace(3) %369, align 2
  %371 = getelementptr inbounds nuw i8, ptr addrspace(3) %345, i32 832
  %372 = load i16, ptr addrspace(3) %371, align 2
  %373 = getelementptr inbounds nuw i8, ptr addrspace(3) %345, i32 896
  %374 = load i16, ptr addrspace(3) %373, align 2
  %375 = getelementptr inbounds nuw i8, ptr addrspace(3) %345, i32 960
  %376 = load i16, ptr addrspace(3) %375, align 2
  %377 = getelementptr inbounds nuw i8, ptr addrspace(3) %345, i32 1024
  %378 = load i16, ptr addrspace(3) %377, align 2
  %379 = getelementptr inbounds nuw i8, ptr addrspace(3) %345, i32 1088
  %380 = load i16, ptr addrspace(3) %379, align 2
  %381 = getelementptr inbounds nuw i8, ptr addrspace(3) %345, i32 1152
  %382 = load i16, ptr addrspace(3) %381, align 2
  %383 = getelementptr inbounds nuw i8, ptr addrspace(3) %345, i32 1216
  %384 = load i16, ptr addrspace(3) %383, align 2
  %385 = getelementptr inbounds nuw i8, ptr addrspace(3) %345, i32 1280
  %386 = load i16, ptr addrspace(3) %385, align 2
  %387 = getelementptr inbounds nuw i8, ptr addrspace(3) %345, i32 1344
  %388 = load i16, ptr addrspace(3) %387, align 2
  %389 = getelementptr inbounds nuw i8, ptr addrspace(3) %345, i32 1408
  %390 = load i16, ptr addrspace(3) %389, align 2
  %391 = getelementptr inbounds nuw [2 x i8], ptr addrspace(1) %327, i64 %344
  store i16 %346, ptr addrspace(1) %391, align 2
  %392 = getelementptr inbounds nuw i8, ptr addrspace(1) %391, i64 64
  store i16 %348, ptr addrspace(1) %392, align 2
  %393 = getelementptr inbounds nuw i8, ptr addrspace(1) %391, i64 128
  store i16 %350, ptr addrspace(1) %393, align 2
  %394 = getelementptr inbounds nuw i8, ptr addrspace(1) %391, i64 192
  store i16 %352, ptr addrspace(1) %394, align 2
  %395 = getelementptr inbounds nuw i8, ptr addrspace(1) %391, i64 256
  store i16 %354, ptr addrspace(1) %395, align 2
  %396 = getelementptr inbounds nuw i8, ptr addrspace(1) %391, i64 320
  store i16 %356, ptr addrspace(1) %396, align 2
  %397 = getelementptr inbounds nuw i8, ptr addrspace(1) %391, i64 384
  store i16 %358, ptr addrspace(1) %397, align 2
  %398 = getelementptr inbounds nuw i8, ptr addrspace(1) %391, i64 448
  store i16 %360, ptr addrspace(1) %398, align 2
  %399 = getelementptr inbounds nuw i8, ptr addrspace(1) %391, i64 512
  store i16 %362, ptr addrspace(1) %399, align 2
  %400 = getelementptr inbounds nuw i8, ptr addrspace(1) %391, i64 576
  store i16 %364, ptr addrspace(1) %400, align 2
  %401 = getelementptr inbounds nuw i8, ptr addrspace(1) %391, i64 640
  store i16 %366, ptr addrspace(1) %401, align 2
  %402 = getelementptr inbounds nuw i8, ptr addrspace(1) %391, i64 704
  store i16 %368, ptr addrspace(1) %402, align 2
  %403 = getelementptr inbounds nuw i8, ptr addrspace(1) %391, i64 768
  store i16 %370, ptr addrspace(1) %403, align 2
  %404 = getelementptr inbounds nuw i8, ptr addrspace(1) %391, i64 832
  store i16 %372, ptr addrspace(1) %404, align 2
  %405 = getelementptr inbounds nuw i8, ptr addrspace(1) %391, i64 896
  store i16 %374, ptr addrspace(1) %405, align 2
  %406 = getelementptr inbounds nuw i8, ptr addrspace(1) %391, i64 960
  store i16 %376, ptr addrspace(1) %406, align 2
  %407 = getelementptr inbounds nuw i8, ptr addrspace(1) %391, i64 1024
  store i16 %378, ptr addrspace(1) %407, align 2
  %408 = getelementptr inbounds nuw i8, ptr addrspace(1) %391, i64 1088
  store i16 %380, ptr addrspace(1) %408, align 2
  %409 = getelementptr inbounds nuw i8, ptr addrspace(1) %391, i64 1152
  store i16 %382, ptr addrspace(1) %409, align 2
  %410 = getelementptr inbounds nuw i8, ptr addrspace(1) %391, i64 1216
  store i16 %384, ptr addrspace(1) %410, align 2
  %411 = getelementptr inbounds nuw i8, ptr addrspace(1) %391, i64 1280
  store i16 %386, ptr addrspace(1) %411, align 2
  %412 = getelementptr inbounds nuw i8, ptr addrspace(1) %391, i64 1344
  store i16 %388, ptr addrspace(1) %412, align 2
  ret void
}

; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
declare noundef i32 @llvm.amdgcn.workgroup.id.x() #2

; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
declare noundef range(i32 0, 1024) i32 @llvm.amdgcn.workitem.id.x() #2

; uselistorder directives
uselistorder ptr @llvm.amdgcn.workitem.id.x, { 2, 1, 0 }

attributes #0 = { convergent nocallback nofree nounwind willreturn }
attributes #1 = { convergent mustprogress nofree norecurse nounwind willreturn "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,32" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="gfx950" "uniform-work-group-size" }
attributes #2 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
```

</details>

cc: AMDGPU codegen


Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[AMDGPU][SDAG ISel] clang/llc hangs on rocPRIM adjacent_difference kernel at gfx950 -O3 #2616

Summary

Affected versions

Known good vs known bad

Reproducer

Setup (no GPU required — compile-only)

Reproduce with `llc`

Reproduce end-to-end with clang (matches original repro)

Diagnostic output

What we tried

Reduction process used

Environment

Reduced IR

Metadata

Assignees

Labels

Type

Fields

Projects

Milestone

Relationships

Development

State	Branch / wheel	Upstream merge-base	Date
✅ works	`release/rocm-rel-7.2`	`c793782b03ab`	2025-10-05
❌ hangs	`rocm-sdk-devel==7.13.0a20260515` (per-family `gfx950-dcgpu` index)	`5b1eed6278de`	2026-03-12

[AMDGPU][SDAG ISel] clang/llc hangs on rocPRIM adjacent_difference kernel at gfx950 -O3 #2616

Description

Summary

Affected versions

Known good vs known bad

Reproducer

Setup (no GPU required — compile-only)

Reproduce with llc

Reproduce end-to-end with clang (matches original repro)

Diagnostic output

What we tried

Reduction process used

Environment

Reduced IR

Metadata

Metadata

Assignees

Labels

Type

Fields

Projects

Milestone

Relationships

Development

Issue actions

Reproduce with `llc`