Visible to Intel only — GUID: cai1690835912822
Ixiasoft
Visible to Intel only — GUID: cai1690835912822
Ixiasoft
3.3.3. Tensor Accumulation Mode
In Tensor accumulation mode, two columns of 64-bit input data are fed through port fp32_a{1..2} to perform addition or subtraction with the 32-bit floating-point values from the previous DSP block through cascade_data_in[63:0] by accumulator (FP32 ALU).
To perform addition or subtraction on accumulator, set the parameter dsp_fp32_sub_encorrespondingly based on the parameter table. To enable cascading connection from previous DSP block, enable the parameter dsp_chain_tensor in the settings.
The two 32-bit floating-point results are sent out through fp32_a{1..2} on first and second column or it can be cascaded to the next DSP block through cascade_data_out[63:0].
Accumulator Operation | Cascade Input Data Enabled |
---|---|
Addition | Column One = fp32_a1[31:0] + cascade_data_in_col_1[31:0] Column Two = fp32_a2[31:0] + cascade_data_in_col_2[31:0] |
Subtraction | Column One = fp32_a1[31:0] - cascade_data_in_col_1[31:0] Column Two = fp32_a2[31:0] - cascade_data_in_col_2[31:0] |