You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I am using the latest docker image. I want to create a input-stationary for large MM. Here is what I did
./autosa ./autosa_tests/large/mm/kernel.c --config=./autosa_config/autosa_config.json \
--target=autosa_hls_c --output-dir=./autosa.tmp/output \
--sa-sizes="{kernel[]->space_time[2];kernel[]->array_part[260,256,512];kernel[]->latency[20,16];kernel[]->simd[1]}"
--simd-info=./autosa_tests/large/mm/simd_info.json \
--host-serialize --local-reduce --reduce-op="+" --simd-touch-space --no-isl-sink;# Error in HLS synthesis
===>The following messages were generated while performing high-level synthesis for kernel: kernel0 Log file: /scratch/users/sx233/FPGA-test/gemm.autosa/temp.autosa.large.mm/_x/kernel0.hw/kernel0/vitis_hls.log :
ERROR: [v++ 214-124] use of undeclared identifier 'fifo_C_1_serialize': /scratch/users/sx233/FPGA-test/gemm.autosa/temp.autosa.large.mm/src/kernel_kernel.cpp:931
ERROR: [v++ 60-300] Failed to build kernel(ip) kernel0, see log for details: /scratch/users/sx233/FPGA-test/gemm.autosa/temp.autosa.large.mm/_x/kernel0.hw/kernel0/vitis_hls.log
Also another question for the PE dimension of the input-stationary SA. For example, if I have a 1024x1024 MatMul kernel, the array part factor is set to [256,256,512], latency hiding factor is [32,32] and SIMD factor is 8. Since the reduction loop is selected as the space loop, I was expecting the SA size to be 512/8 = 64? but in the generated code I can only see 32 PEs. Is that the correct way for calculating the PE dimension for that?
The text was updated successfully, but these errors were encountered:
I am using the latest docker image. I want to create a input-stationary for large MM. Here is what I did
Also another question for the PE dimension of the input-stationary SA. For example, if I have a 1024x1024 MatMul kernel, the array part factor is set to [256,256,512], latency hiding factor is [32,32] and SIMD factor is 8. Since the reduction loop is selected as the space loop, I was expecting the SA size to be 512/8 = 64? but in the generated code I can only see 32 PEs. Is that the correct way for calculating the PE dimension for that?
The text was updated successfully, but these errors were encountered: