Explore SME matrix operations with interactive visualizations
Introduction
Explore the ZA tile layout
Transpose a 32-bit matrix in place with ZA
Transpose a matrix with an incomplete edge
Multiply matrices with vector outer products
Transpose and pack matrix A
Pad matrix B to the ZA tile width
Trace packed GEMM through ZA
Compare a GEMM with gather loads
Interleave matrix A
Trace GEMM with interleaved matrix A
Next Steps
Explore SME matrix operations with interactive visualizations
Introduction
Explore the ZA tile layout
Transpose a 32-bit matrix in place with ZA
Transpose a matrix with an incomplete edge
Multiply matrices with vector outer products
Transpose and pack matrix A
Pad matrix B to the ZA tile width
Trace packed GEMM through ZA
Compare a GEMM with gather loads
Interleave matrix A
Trace GEMM with interleaved matrix A
Next Steps
Review frequently asked questions
These FAQs were drafted with an approved AI-assisted workflow and reviewed by Arm contributors before publication. Human technical review remains part of the process so the final page reflects engineering rigor, accuracy, and Arm editorial standards.
Scalable Vector Length determines the height and width of the ZA storage. Data element size determines how many elements will fit into a row of ZA storage of a given Scalable Vector Length. That number of elements per row determines the height of the logical ZA tiles. The height of the ZA storage divided by the height of a logical ZA tile, determines the number of ZA tiles.
SME uses predicates to specify which rows or columns are active. SME operations only affect active rows. If a matrix is not an exact multiple of the ZA tile size, then the last few rows or columns will be set as inactive.
Review the section on
vectorizing matrix multiplication
Follow one column of matrix A and the corresponding row of matrix B. You’ll see their outer product add a partial result to multiple elements of matrix C at the same time. Repeat for each column and row pair to see the complete result accumulate in ZA.
In Output Packed Data, you should see the values of matrix A in the new order, with zeros where its final tile has no input rows. In Output Padded Data, check that the original values of matrix B remain in order and zeros fill the inactive columns.
Compare how each view loads matrix A. With gather loads, you read values from non-contiguous addresses. With the transposed-and-interleaved layout, you load from contiguous addresses. You can compare Output Matrix Data in both views to confirm that the results are identical.