Bump to depend on iree 3.0.0 packages.

[ScottTodd]

See the release tracker at iree-org/iree#19063.

[marbre]

Seems there are some failures which might be due to the update :/

[ScottTodd]

Seems there are some failures which might be due to the update :/

Ack. Fixed a few but looks like there are more.

[ScottTodd]

@qedawkins @kuhar I'm still getting errors here in some tests. Not sure what I'm doing wrong

https://github.com/iree-org/iree-turbine/actions/runs/11861587460/job/33059220943?pr=275#step:6:42863

            with InsertionPoint(self.def_module.body_block):
                def_func_op = func_d.FuncOp(name, def_ftype)
                def_func_block = def_func_op.add_entry_block()
                def_func_args = list(def_func_block.arguments)
                if workgroup_size is not None and subgroup_size is not None:
>                   def_func_op.attributes["translation_info"] = Attribute.parse(
                        f"#iree_codegen.translation_info<pipeline = None "
                        f"workgroup_size=[{','.join(str(x) for x in workgroup_size)}] "
                        f"subgroup_size={subgroup_size}>"
                    )
E                   iree.compiler._mlir_libs._site_initialize.<locals>.MLIRError: Unable to parse attribute:
E                   error: "#iree_codegen.translation_info<pipeline = None workgroup_size=[128,2,1] subgroup_size=64>":1:32: expected ::mlir::iree_compiler::IREE::Codegen::DispatchLoweringPassPipeline to be one of: CPUDefault, CPUDoubleTilingExpert, CPUConvTileAndDecomposeExpert, Mmt4dTilingExpert, CPUBufferOpsTileAndVectorize, CPUDataTiling, CPULinalgExtTileAndVectorize, LLVMGPUDefault, LLVMGPUBaseLowering, LLVMGPUDistribute, LLVMGPUVectorize, LLVMGPUMatmulSimt, LLVMGPUMatmulTensorCore, LLVMGPUTransposeSharedMem, LLVMGPUWarpReduction, LLVMGPUPackUnPack, LLVMGPUMatmulTensorCoreMmaSync, LLVMGPUVectorDistribute, LLVMGPUPadAndVectorDistribute, LLVMGPUWinogradVectorize, LLVMGPUTileAndFuse, SPIRVBaseLowering, SPIRVBaseDistribute, SPIRVBaseVectorize, SPIRVSubgroupReduce, SPIRVMatmulPromoteVectorize, SPIRVCooperativeMatrixVectorize, SPIRVWinogradVectorize, VMVXDefault, TransformDialectCodegen, Custom, None
E                   error: "#iree_codegen.translation_info<pipeline = None workgroup_size=[128,2,1] subgroup_size=64>":1:41: failed to parse DispatchLoweringPassPipelineAttr parameter 'value' which is to be a `::mlir::iree_compiler::IREE::Codegen::DispatchLoweringPassPipeline`
E                   error: "#iree_codegen.translation_info<pipeline = None workgroup_size=[128,2,1] subgroup_size=64>":1:41: failed to parse IREECodegen_TranslationInfoAttr parameter 'passPipeline' which is to be a `IREE::Codegen::DispatchLoweringPassPipelineAttr`

iree/turbine/kernel/compiler/dispatch_codegen.py:167: MLIRError

[ScottTodd]

Ah, I think those workflows still installed the older versions. Failures should go away after pushing a stable release, or I can get them installing the nightly releases too.

[ScottTodd]

Passing --upgrade in the workflows helped. Looks like some of the runners are persistent and don't clear our their venvs between runs. FYI @harsh-nod