Open monorimet opened 1 month ago
FWIW this also fails with vector distribution explicitly disabled:, same dispatch:
<unknown>:0: error: LLVM Translation failed for operation: builtin.unrealized_conversion_cast
/home/eagarvey/SHARK/models/stabilityai_sdxl_turbo_1_64_1024x1024_fp16_gfx942/EulerDiscrete_unet_4_dispatches/configured_compiled_scheduled_unet_run_forward$async_dispatch_50.mlir:2:2: error: failed to translate the MLIR LLVM dialect to the native llvm::Module
hal.executable.variant public @rocm_hsaco_fb target(<"rocm", "rocm-hsaco-fb", {mma_intrinsics = [#iree_gpu.mma_layout<MFMA_F16_16x16x16_F32>, #iree_gpu.mma_layout<MFMA_F16_32x32x8_F32>], target_arch = "gfx942", ukernels = "none", waves_per_eu = 2 : i64}>) {
^
/home/eagarvey/SHARK/models/stabilityai_sdxl_turbo_1_64_1024x1024_fp16_gfx942/EulerDiscrete_unet_4_dispatches/configured_compiled_scheduled_unet_run_forward$async_dispatch_50.mlir:2:2: error: failed to serialize executable for target backend rocm
hal.executable.variant public @rocm_hsaco_fb target(<"rocm", "rocm-hsaco-fb", {mma_intrinsics = [#iree_gpu.mma_layout<MFMA_F16_16x16x16_F32>, #iree_gpu.mma_layout<MFMA_F16_32x32x8_F32>], target_arch = "gfx942", ukernels = "none", waves_per_eu = 2 : i64}>) {
^
/home/eagarvey/SHARK/models/stabilityai_sdxl_turbo_1_64_1024x1024_fp16_gfx942/EulerDiscrete_unet_4_dispatches/configured_compiled_scheduled_unet_run_forward$async_dispatch_50.mlir:1:0: error: failed to serialize executables
What happened?
Module-level error:
Dispatch-level Error:
Reproducer: attention_and_matmul_spec_mfma.mlir compiled_scheduled_unet_run_forward$async_dispatch_50.mlir
Steps to reproduce your issue
No response
What component(s) does this issue relate to?
Compiler
Version information
commit 2a2a4d020f7dc2e12f7768c322975068f41bbba1 (HEAD -> main, origin/main, origin/HEAD)
Additional context
This is a new version of the IR, where batch dim is 1. Hence, we likely have a vector distribution issue for this shape of this op.