Issues

• [AMDGPU] Backport mix-instruction fpround fixes (LLVM PRs #236759, #214544)

  ROCm/llvm-project#5339 · Filter by author oanaoana opened on Nov 9, 2026

• [Issue]: [HIP][gfx1100] __fadd_rn(__fmul_rn(a, c), b) contracts into an FMA and flips RN-to-even tie cases

  ROCm/llvm-project#5330 · Filter by author xyzzing opened on Nov 9, 2026

• Backport request: gfx11 VOPD interlock hazard fix (llvm/llvm-project#220348, #225880) to ROCm 7.2.x and TheRock 7.14: silent wrong results on gfx1100

  ROCm/llvm-project#5318 · Filter by author vivienfanghuagood opened on Nov 8, 2026

• [Issue]: [Clang][AMDGPU] -fno-finite-math-only does not override -ffast-math for OCML device library selection**

  ROCm/llvm-project#5242 · Filter by author thelittlefireman opened on Nov 3, 2026
  1

  

• [AFAR 24.3.0] GPU libc free() ordering bug (llvm-project#238205) corrupts flang array-constructor temporaries on MI300A

  ROCm/llvm-project#5228 · Filter by author sbryngelson opened on Nov 3, 2026
  1

• [AMDGPU][gfx950] Missed MFMA→v_accvgpr_read hazard across a branch gives silently wrong results on ROCm 7.2.3; please backport llvm/llvm-project#225989

  ROCm/llvm-project#5153 · Filter by author vivienfanghuagood opened on Oct 28, 2026

• [Issue]: [AMDGPU][gfx1250] Too few wait states before a VALU write of an FP8 16x16x64 WMMA source

  bugExternalmi455
  ROCm/llvm-project#5089 · Filter by author JP-Fernando opened on Oct 23, 2026

• [Issue]: [AMDGPU][gfx950] Runtime-dimensional offset calculation regresses 8% and adds ~200 instructions

  ROCm/llvm-project#4675 · Filter by author lauri9 opened on Sep 28, 2026
  1

• [Feature]: extend transpiler support to Qwen3-0.6B

  ROCm/llvm-project#4601 · Filter by author martin-luecke opened on Sep 24, 2026
  

• [Issue]: Incorrect missing DI for optimized out variable

  bugdebug-info-fuzzing
  Bug

  ROCm/llvm-project#4571 · Filter by author mgcarrasco opened on Sep 22, 2026
  

• [Issue]: Debugger reports a live variable as optimized out

  bugdebug-info-fuzzing
  Bug

  ROCm/llvm-project#4570 · Filter by author mgcarrasco opened on Sep 22, 2026
  

• [Issue]: GPU codegen is not reproducible

  ROCm/llvm-project#4513 · Filter by author eble-amd opened on Sep 17, 2026
  1

  

• Clang backend failure when compiling hipblaslt on LoongArch64

  ROCm/llvm-project#4454 · Filter by author arrowd opened on Sep 15, 2026

• Full LTO default on AMDGCN causes register spills and 7.6x slowdown in MIOpen's hipconv solver (gfx950)

  ROCm/llvm-project#4434 · Filter by author amd-alavin opened on Sep 14, 2026
  6

• [Issue]: Debugger shows the wrong value for a variable

  bugdebug-info-fuzzing
  Bug

  ROCm/llvm-project#4380 · Filter by author mgcarrasco opened on Sep 11, 2026
  

• [Issue]: Latest documentation for AMD Clang out of date

  ROCm/llvm-project#4344 · Filter by author Danny213123 opened on Sep 9, 2026
  

• [Issue]: missing do concurrent feature in the rock 10.0 amdflang 23

  ROCm/llvm-project#4330 · Filter by author JorgeG94 opened on Sep 9, 2026

• [Feature]: extend transpiler support to attention

  transpiler
  ROCm/llvm-project#4314 · Filter by author ftynse opened on Sep 8, 2026
  

• [Feature]: transpiler pending refactors

  transpiler
  ROCm/llvm-project#4129 · Filter by author ftynse opened on Aug 26, 2026
  

• [Issue]: [flang][OpenMP][USM] amdflang 23 emits spurious omp_target_data_op_emi on Fortran USM target offload (regression from flang 22 and 20)

  bugflangopenmp
  ROCm/llvm-project#4109 · Filter by author anujshuk-amd opened on Aug 25, 2026
  

• [Issue]: [llvm-objdump] --offloading target filter broken on ROCm 7.14

  ROCm/llvm-project#4108 · Filter by author maartenarnst opened on Aug 25, 2026
  3

  

• [Attributor][OpenMP] attributor-max-pi-accesses cap degrades codegen image-wide once an offload image has ~512 kernels

  ROCm/llvm-project#4070 · Filter by author sbryngelson opened on Aug 24, 2026

• [Issue]: ICE during Greedy Register Allocator pass

  ROCm/llvm-project#3995 · Filter by author StewMH opened on Aug 19, 2026
  2

• [Issue]: Missed optimization for kernel generation when using IF in TARGET TEAMS DISTRIBUTE PARALLEL DO

  ROCm/llvm-project#3971 · Filter by author AdminXVII opened on Aug 18, 2026

• [Issue]: do concurrent compiler bug (frontier)

  ROCm/llvm-project#3700 · Filter by author JorgeG94 opened on Jul 30, 2026