From e541026cfc0ee4be25e7093cb7ff3acfa3cbb6e7 Mon Sep 17 00:00:00 2001 From: negargoli93 Date: Mon, 16 Jul 2018 13:18:23 -0700 Subject: fix pipline for tensor_core and change config --- cuda-kernels/config_fermi_islip.icnt | 2 +- cuda-kernels/gpgpusim.config | 10 +++++----- 2 files changed, 6 insertions(+), 6 deletions(-) (limited to 'cuda-kernels') diff --git a/cuda-kernels/config_fermi_islip.icnt b/cuda-kernels/config_fermi_islip.icnt index a788090..3b8b496 100755 --- a/cuda-kernels/config_fermi_islip.icnt +++ b/cuda-kernels/config_fermi_islip.icnt @@ -7,7 +7,7 @@ network_count = 2; // Topology topology = fly; -k = 62; +k = 102; n = 1; // Routing diff --git a/cuda-kernels/gpgpusim.config b/cuda-kernels/gpgpusim.config index 272ad3d..2510d21 100755 --- a/cuda-kernels/gpgpusim.config +++ b/cuda-kernels/gpgpusim.config @@ -10,7 +10,7 @@ -gpgpu_ptx_save_converted_ptxplus 0 # high level architecture configuration --gpgpu_n_clusters 40 +-gpgpu_n_clusters 80 -gpgpu_n_cores_per_cluster 1 -gpgpu_n_mem 11 -gpgpu_n_sub_partition_per_mchannel 2 @@ -33,7 +33,7 @@ # ID_OC_SP,ID_OC_SFU,ID_OC_MEM,OC_EX_SP,OC_EX_SFU,OC_EX_MEM,EX_WB ## Pascal GP102 has 4 SP SIMD units and 1 SFU unit ## we need to scale the number of pipeline registers to be equal to the number of SP units --gpgpu_pipeline_widths 4,1,1,4,1,1,6 +-gpgpu_pipeline_widths 4,1,1,1,4,1,1,1,6 -gpgpu_num_sp_units 4 -gpgpu_num_sfu_units 1 # Instruction latencies and initiation intervals @@ -72,15 +72,15 @@ -gpgpu_operand_collector_num_units_sp 20 -gpgpu_operand_collector_num_units_sfu 4 #-gpgpu_operand_collector_num_units_tensor_core 24 --gpgpu_operand_collector_num_units_mem 8 +-gpgpu_operand_collector_num_units_mem 8 -gpgpu_operand_collector_num_in_ports_sp 4 -gpgpu_operand_collector_num_out_ports_sp 4 -gpgpu_operand_collector_num_in_ports_sfu 1 -gpgpu_operand_collector_num_out_ports_sfu 1 #-gpgpu_operand_collector_num_in_ports_tensor_core 1 #-gpgpu_operand_collector_num_out_ports_tensor_core 1 --gpgpu_operand_collector_num_in_ports_mem 1 --gpgpu_operand_collector_num_out_ports_mem 1 +-gpgpu_operand_collector_num_in_ports_mem 10 +-gpgpu_operand_collector_num_out_ports_mem 10 # gpgpu_num_reg_banks should be increased to 32, but it gives an error! -gpgpu_num_reg_banks 32 -- cgit v1.3