With　the　rapid　development　of　artificial　intelligence　and　deep　learning,　particularly　with　the　emergence　of　pre-trained　large　models,　the　tremendous　pressure　on　computational　power　has　led　to　an　increasing　demand　for　hardware　accelerators.　This　trend　brings　significant　challenges　in　computational　efficiency　and　resource　consumption.　In　this　context,　research　on　parameter　sparsity　is　becoming　increasingly　important,　as　it　can　reduce　unnecessary　computations　in　models　and　lower　storage　requirements　through　sparse　matrix　multiplication.　However,　traditional　CPU　and　GPU　architectures　are　characterized　by　high　power　consumption,　while　FPGA　development　processes　are　often　overly　complex.　Therefore,　exploring　sparse　matrix　multiplication　on　new　platforms　and　architectures　has　become　essential.　The　Versal　ACAP　(Adaptive　Compute　Acceleration　Platform)　offers　a　promising　solution　for　high-performance　computing　due　to　its　low　power　consumption　and　shorter　development　cycles.　This　paper　aims　to　investigate　efficient　implementations　of　the　sparse　matrix-dense　matrix　multiplication　(SpMM)　algorithm　on　the　Versal　ACAP　platform,　leveraging　the　AIE　Graph　methodology　for　automatic　strategy　generation.　By　analyzing　and　optimizing　the　SpMM　algorithm　based　on　the　CSR　compression　format,　we　demonstrate　the　advantages　and　potential　of　the　Versal　ACAP　platform　in　the　research　of　parameter　sparsity　for　large　models.　Experimental　results　indicate　that　the　method　achieves　an　average　computation　time　of　1.64　µs　in　a　single-core　configuration,　while　reducing　PLIO　resource　consumption　by　approximately　73.4%　in　multi-core　scenarios,　highlighting　the　potential　of　the　AIE　architecture　in　SpMM　computations.　©　2024　Copyright　held　by　the　owner/author(s).