From f9e65c53a29b89c167e713afa6d5eb92709047af Mon Sep 17 00:00:00 2001 From: "Mahmoud Khairy A. Abdallah" Date: Thu, 2 Apr 2020 21:23:07 -0400 Subject: updating config files --- .../SM75_RTX2060/config_turing_islip.icnt | 73 ++++++++ configs/tested-cfgs/SM75_RTX2060/gpgpusim.config | 187 +++++++++++++++++++++ 2 files changed, 260 insertions(+) create mode 100644 configs/tested-cfgs/SM75_RTX2060/config_turing_islip.icnt create mode 100644 configs/tested-cfgs/SM75_RTX2060/gpgpusim.config (limited to 'configs/tested-cfgs/SM75_RTX2060') diff --git a/configs/tested-cfgs/SM75_RTX2060/config_turing_islip.icnt b/configs/tested-cfgs/SM75_RTX2060/config_turing_islip.icnt new file mode 100644 index 0000000..eed1c34 --- /dev/null +++ b/configs/tested-cfgs/SM75_RTX2060/config_turing_islip.icnt @@ -0,0 +1,73 @@ +//52*1 fly with 32 flits per packet under gpgpusim injection mode +use_map = 0; +flit_size = 40; + +// currently we do not use this, see subnets below +network_count = 2; + +// Topology +topology = fly; +k = 52; +n = 1; + +// Routing + +routing_function = dest_tag; + +// Flow control + +num_vcs = 1; +vc_buf_size = 64; +input_buffer_size = 256; +ejection_buffer_size = 64; +boundary_buffer_size = 64; + +wait_for_tail_credit = 0; + +// Router architecture + +vc_allocator = islip; //separable_input_first; +sw_allocator = islip; //separable_input_first; +alloc_iters = 1; + +credit_delay = 0; +routing_delay = 0; +vc_alloc_delay = 1; +sw_alloc_delay = 1; + +input_speedup = 1; +output_speedup = 1; +internal_speedup = 2.0; + +// Traffic, GPGPU-Sim does not use this + +traffic = uniform; +packet_size ={{1,2,3,4},{10,20}}; +packet_size_rate={{1,1,1,1},{2,1}}; + +// Simulation - Don't change + +sim_type = gpgpusim; +//sim_type = latency; +injection_rate = 0.1; + +subnets = 2; + +// Always use read and write no matter following line +//use_read_write = 1; + + +read_request_subnet = 0; +read_reply_subnet = 1; +write_request_subnet = 0; +write_reply_subnet = 1; + +read_request_begin_vc = 0; +read_request_end_vc = 0; +write_request_begin_vc = 0; +write_request_end_vc = 0; +read_reply_begin_vc = 0; +read_reply_end_vc = 0; +write_reply_begin_vc = 0; +write_reply_end_vc = 0; + diff --git a/configs/tested-cfgs/SM75_RTX2060/gpgpusim.config b/configs/tested-cfgs/SM75_RTX2060/gpgpusim.config new file mode 100644 index 0000000..9d3992a --- /dev/null +++ b/configs/tested-cfgs/SM75_RTX2060/gpgpusim.config @@ -0,0 +1,187 @@ +# This config models the Turing RTX 2060 +# For more info about turing architecture: +# https://www.nvidia.com/content/dam/en-zz/Solutions/design-visualization/technologies/turing-architecture/NVIDIA-Turing-Architecture-Whitepaper.pdf +# "RTX on—The NVIDIA Turing GPU", IEEE MICRO 2020 + +# functional simulator specification +-gpgpu_ptx_instruction_classification 0 +-gpgpu_ptx_sim_mode 0 +-gpgpu_ptx_force_max_capability 70 + +# Device Limits +-gpgpu_stack_size_limit 1024 +-gpgpu_heap_size_limit 8388608 +-gpgpu_runtime_sync_depth_limit 2 +-gpgpu_runtime_pending_launch_count_limit 2048 +-gpgpu_kernel_launch_latency 5000 + +# Compute Capability +-gpgpu_compute_capability_major 7 +-gpgpu_compute_capability_minor 5 + +# PTX execution-driven +-gpgpu_ptx_convert_to_ptxplus 0 +-gpgpu_ptx_save_converted_ptxplus 0 +# SASS trace-driven mode execution +#-trace_driven_mode 1 + +# high level architecture configuration +-gpgpu_n_clusters 30 +-gpgpu_n_cores_per_cluster 1 +-gpgpu_n_mem 12 +-gpgpu_n_sub_partition_per_mchannel 2 + +# volta clock domains +#-gpgpu_clock_domains ::: +-gpgpu_clock_domains 1365.0:1365.0:1365.0:3500.0 +# boost mode +# -gpgpu_clock_domains 1680.0:1680.0:1680.0:3500.0 + +# shader core pipeline config +-gpgpu_shader_registers 65536 +-gpgpu_registers_per_block 65536 +-gpgpu_occupancy_sm_number 75 + +# This implies a maximum of 64 warps/SM +-gpgpu_shader_core_pipeline 2048:32 +-gpgpu_shader_cta 32 +-gpgpu_simd_model 1 + +# Pipeline widths and number of FUs +# ID_OC_SP,ID_OC_DP,ID_OC_INT,ID_OC_SFU,ID_OC_MEM,OC_EX_SP,OC_EX_DP,OC_EX_INT,OC_EX_SFU,OC_EX_MEM,EX_WB,ID_OC_TENSOR_CORE,OC_EX_TENSOR_CORE +## Turing has 4 SP SIMD units, 4 INT units, 4 SFU units, 8 Tensor core units +## We need to scale the number of pipeline registers to be equal to the number of SP units +-gpgpu_pipeline_widths 4,0,4,4,4,4,0,4,4,4,8,4,4 +-gpgpu_num_sp_units 4 +-gpgpu_num_sfu_units 4 +-gpgpu_num_int_units 4 +-gpgpu_tensor_core_avail 1 +-gpgpu_num_tensor_core_units 4 + +# Instruction latencies and initiation intervals +# "ADD,MAX,MUL,MAD,DIV" +# All Div operations are executed on SFU unit +-ptx_opcode_latency_int 4,13,4,5,145 +-ptx_opcode_initiation_int 2,2,2,2,8 +-ptx_opcode_latency_fp 4,13,4,5,39 +-ptx_opcode_initiation_fp 2,2,2,2,4 +-ptx_opcode_latency_dp 8,19,8,8,330 +-ptx_opcode_initiation_dp 4,4,4,4,130 +-ptx_opcode_latency_sfu 100 +-ptx_opcode_initiation_sfu 8 +-ptx_opcode_latency_tesnor 64 +-ptx_opcode_initiation_tensor 64 + +# Trung has sub core model, in which each scheduler has its own register file and EUs +# i.e. schedulers are isolated +-sub_core_model 1 +# disable specialized operand collectors and use generic operand collectors instead +-enable_specialized_operand_collector 0 +-gpgpu_operand_collector_num_units_gen 8 +-gpgpu_operand_collector_num_in_ports_gen 8 +-gpgpu_operand_collector_num_out_ports_gen 8 +# turing has 8 banks dual-port, 4 schedulers, two banks per scheduler +# we increase #banks to 16 to mitigate the effect of Regisrer File Cache (RFC) which we do not implement in the current version +-gpgpu_num_reg_banks 16 +-gpgpu_reg_file_port_throughput 2 + +# shared memory bankconflict detection +-gpgpu_shmem_num_banks 32 +-gpgpu_shmem_limited_broadcast 0 +-gpgpu_shmem_warp_parts 1 +-gpgpu_coalesce_arch 60 + +## In Turing, a warp scheduler can issue 1 inst per cycle +-gpgpu_max_insn_issue_per_warp 1 +-gpgpu_dual_issue_diff_exec_units 1 + +# Turing has four schedulers per core +-gpgpu_num_sched_per_core 4 +# Greedy then oldest scheduler +-gpgpu_scheduler gto + +# ::,::::,::,:** +# ** Optional parameter - Required when mshr_type==Texture Fifo +-adaptive_cache_config 0 +-l1_banks 4 +-gpgpu_cache:dl1 S:1:128:512,L:L:s:N:L,A:256:8,16:0,32 +-gpgpu_shmem_size 65536 +-gpgpu_shmem_sizeDefault 65536 +-gpgpu_shmem_per_block 65536 +-gmem_skip_L1D 0 +-icnt_flit_size 40 +-gpgpu_n_cluster_ejection_buffer_size 32 +-l1_latency 20 +-smem_latency 20 +-gpgpu_flush_l1_cache 1 + +# 32 sets, each 128 bytes 32-way for each memory sub partition (96 KB per memory sub partition). This gives us 6MB L2 cache +-gpgpu_cache:dl2 S:32:128:32,L:B:m:L:L,A:192:4,32:0,32 +-gpgpu_cache:dl2_texture_only 0 +-gpgpu_dram_partition_queues 64:64:64:64 +-perf_sim_memcpy 1 +-memory_partition_indexing 0 + +# 128 KB Inst. +-gpgpu_cache:il1 N:64:128:16,L:R:f:N:L,S:2:48,4 +-inst_fetch_throughput 4 +# 128 KB Tex +# Note, TEX is deprected in Volta, It is used for legacy apps only. Use L1D cache instead with .nc modifier or __ldg mehtod +-gpgpu_tex_cache:l1 N:4:128:256,L:R:m:N:L,T:512:8,128:2 +# 64 KB Const +-gpgpu_const_cache:l1 N:128:64:8,L:R:f:N:L,S:2:64,4 +-perfect_inst_const_cache 1 + +# interconnection +#-network_mode 1 +#-inter_config_file config_turing_islip.icnt +# use built-in local xbar +-network_mode 2 +-inct_in_buffer_limit 512 +-inct_out_buffer_limit 512 +-inct_subnets 2 +-arbiter_algo 1 + +# memory partition latency config +-rop_latency 160 +-dram_latency 100 + +# dram model config +-gpgpu_dram_scheduler 1 +-gpgpu_frfcfs_dram_sched_queue_size 64 +-gpgpu_dram_return_queue_size 192 + +# Turing has GDDR6 +# http://monitorinsider.com/GDDR6.html +-gpgpu_n_mem_per_ctrlr 1 +-gpgpu_dram_buswidth 2 +-gpgpu_dram_burst_length 16 +-dram_data_command_freq_ratio 4 # GDDR6 is QDR +-gpgpu_mem_address_mask 1 +-gpgpu_mem_addr_mapping dramid@8;00000000.00000000.00000000.00000000.0000RRRR.RRRRRRRR.RBBBCCCC.BCCSSSSS + +# Use the same GDDR5 timing, scaled to 3500MHZ +-gpgpu_dram_timing_opt "nbk=16:CCD=4:RRD=10:RCD=20:RAS=50:RP=20:RC=62: + CL=20:WL=8:CDLR=9:WR=20:nbkgrp=4:CCDL=4:RTPL=4" + +# select lower bits for bnkgrp to increase bnkgrp parallelism +-dram_bnk_indexing_policy 0 +-dram_bnkgrp_indexing_policy 1 + +#-Seperate_Write_Queue_Enable 1 +#-Write_Queue_Size 64:56:32 + +# stat collection +-gpgpu_memlatency_stat 14 +-gpgpu_runtime_stat 500 +-enable_ptx_file_line_stats 1 +-visualizer_enabled 0 + +# power model configs, disable it untill we create a real energy model for Volta +-power_simulation_enabled 0 + +# tracing functionality +#-trace_enabled 1 +#-trace_components WARP_SCHEDULER,SCOREBOARD +#-trace_sampling_core 0 + -- cgit v1.3 From fe8dd7eadbb67b9e917c92c9c540812db35fce37 Mon Sep 17 00:00:00 2001 From: "Mahmoud Khairy A. Abdallah" Date: Thu, 2 Apr 2020 21:41:01 -0400 Subject: updating the turing config --- configs/tested-cfgs/SM3_KEPLER_TITAN/gpgpusim.config | 2 +- configs/tested-cfgs/SM75_RTX2060/gpgpusim.config | 18 +++++++++--------- 2 files changed, 10 insertions(+), 10 deletions(-) (limited to 'configs/tested-cfgs/SM75_RTX2060') diff --git a/configs/tested-cfgs/SM3_KEPLER_TITAN/gpgpusim.config b/configs/tested-cfgs/SM3_KEPLER_TITAN/gpgpusim.config index 1951a3f..77617d6 100644 --- a/configs/tested-cfgs/SM3_KEPLER_TITAN/gpgpusim.config +++ b/configs/tested-cfgs/SM3_KEPLER_TITAN/gpgpusim.config @@ -40,7 +40,7 @@ -gpgpu_shader_registers 65536 -gpgpu_occupancy_sm_number 62 -# This implies a maximum of 32 warps/SM +# This implies a maximum of 64 warps/SM -gpgpu_shader_core_pipeline 2048:32 -gpgpu_shader_cta 16 -gpgpu_simd_model 1 diff --git a/configs/tested-cfgs/SM75_RTX2060/gpgpusim.config b/configs/tested-cfgs/SM75_RTX2060/gpgpusim.config index 9d3992a..e8329dd 100644 --- a/configs/tested-cfgs/SM75_RTX2060/gpgpusim.config +++ b/configs/tested-cfgs/SM75_RTX2060/gpgpusim.config @@ -1,12 +1,12 @@ # This config models the Turing RTX 2060 # For more info about turing architecture: -# https://www.nvidia.com/content/dam/en-zz/Solutions/design-visualization/technologies/turing-architecture/NVIDIA-Turing-Architecture-Whitepaper.pdf -# "RTX on—The NVIDIA Turing GPU", IEEE MICRO 2020 +# 1- https://www.nvidia.com/content/dam/en-zz/Solutions/design-visualization/technologies/turing-architecture/NVIDIA-Turing-Architecture-Whitepaper.pdf +# 2- "RTX on—The NVIDIA Turing GPU", IEEE MICRO 2020 # functional simulator specification -gpgpu_ptx_instruction_classification 0 -gpgpu_ptx_sim_mode 0 --gpgpu_ptx_force_max_capability 70 +-gpgpu_ptx_force_max_capability 75 # Device Limits -gpgpu_stack_size_limit 1024 @@ -33,17 +33,17 @@ # volta clock domains #-gpgpu_clock_domains ::: --gpgpu_clock_domains 1365.0:1365.0:1365.0:3500.0 +-gpgpu_clock_domains 1365.0:1365.0:1365.0:7000.0 # boost mode -# -gpgpu_clock_domains 1680.0:1680.0:1680.0:3500.0 +# -gpgpu_clock_domains 1680.0:1680.0:1680.0:7000.0 # shader core pipeline config -gpgpu_shader_registers 65536 -gpgpu_registers_per_block 65536 -gpgpu_occupancy_sm_number 75 -# This implies a maximum of 64 warps/SM --gpgpu_shader_core_pipeline 2048:32 +# This implies a maximum of 32 warps/SM +-gpgpu_shader_core_pipeline 1024:32 -gpgpu_shader_cta 32 -gpgpu_simd_model 1 @@ -89,7 +89,7 @@ -gpgpu_shmem_num_banks 32 -gpgpu_shmem_limited_broadcast 0 -gpgpu_shmem_warp_parts 1 --gpgpu_coalesce_arch 60 +-gpgpu_coalesce_arch 75 ## In Turing, a warp scheduler can issue 1 inst per cycle -gpgpu_max_insn_issue_per_warp 1 @@ -156,7 +156,7 @@ -gpgpu_n_mem_per_ctrlr 1 -gpgpu_dram_buswidth 2 -gpgpu_dram_burst_length 16 --dram_data_command_freq_ratio 4 # GDDR6 is QDR +-dram_data_command_freq_ratio 2 # GDDR6 is configured as DDR in Turing -gpgpu_mem_address_mask 1 -gpgpu_mem_addr_mapping dramid@8;00000000.00000000.00000000.00000000.0000RRRR.RRRRRRRR.RBBBCCCC.BCCSSSSS -- cgit v1.3 From 52204ff08a9c9a21a99fee3f976d2a419c014fec Mon Sep 17 00:00:00 2001 From: "Mahmoud Khairy A. Abdallah" Date: Mon, 18 May 2020 21:49:34 -0400 Subject: fixing some failing apps --- configs/tested-cfgs/SM75_RTX2060/gpgpusim.config | 6 +++--- configs/tested-cfgs/SM7_QV100_SASS/gpgpusim.config | 4 ++-- src/gpgpu-sim/gpu-sim.cc | 3 ++- src/gpgpu-sim/shader.cc | 2 +- 4 files changed, 8 insertions(+), 7 deletions(-) (limited to 'configs/tested-cfgs/SM75_RTX2060') diff --git a/configs/tested-cfgs/SM75_RTX2060/gpgpusim.config b/configs/tested-cfgs/SM75_RTX2060/gpgpusim.config index e8329dd..b89971e 100644 --- a/configs/tested-cfgs/SM75_RTX2060/gpgpusim.config +++ b/configs/tested-cfgs/SM75_RTX2060/gpgpusim.config @@ -33,9 +33,9 @@ # volta clock domains #-gpgpu_clock_domains ::: --gpgpu_clock_domains 1365.0:1365.0:1365.0:7000.0 +-gpgpu_clock_domains 1365.0:1365.0:1365.0:3500.0 # boost mode -# -gpgpu_clock_domains 1680.0:1680.0:1680.0:7000.0 +# -gpgpu_clock_domains 1680.0:1680.0:1680.0:3500.0 # shader core pipeline config -gpgpu_shader_registers 65536 @@ -156,7 +156,7 @@ -gpgpu_n_mem_per_ctrlr 1 -gpgpu_dram_buswidth 2 -gpgpu_dram_burst_length 16 --dram_data_command_freq_ratio 2 # GDDR6 is configured as DDR in Turing +-dram_data_command_freq_ratio 4 -gpgpu_mem_address_mask 1 -gpgpu_mem_addr_mapping dramid@8;00000000.00000000.00000000.00000000.0000RRRR.RRRRRRRR.RBBBCCCC.BCCSSSSS diff --git a/configs/tested-cfgs/SM7_QV100_SASS/gpgpusim.config b/configs/tested-cfgs/SM7_QV100_SASS/gpgpusim.config index 0255f76..0df3eec 100644 --- a/configs/tested-cfgs/SM7_QV100_SASS/gpgpusim.config +++ b/configs/tested-cfgs/SM7_QV100_SASS/gpgpusim.config @@ -74,8 +74,8 @@ -ptx_opcode_initiation_dp 4,4,4,4,130 -ptx_opcode_latency_sfu 100 -ptx_opcode_initiation_sfu 8 --ptx_opcode_latency_tesnor 6 --ptx_opcode_initiation_tensor 2 +-ptx_opcode_latency_tesnor 8 +-ptx_opcode_initiation_tensor 4 # Volta has sub core model, in which each scheduler has its own register file and EUs # i.e. schedulers are isolated diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc index e5b9c9d..cd5fa56 100644 --- a/src/gpgpu-sim/gpu-sim.cc +++ b/src/gpgpu-sim/gpu-sim.cc @@ -1828,7 +1828,8 @@ void shader_core_ctx::dump_warp_state( FILE *fout ) const void gpgpu_sim::perf_memcpy_to_gpu( size_t dst_start_addr, size_t count ) { if (m_memory_config->m_perf_sim_memcpy) { - assert (dst_start_addr % 32 == 0); + //if(!m_config.trace_driven_mode) //in trace-driven mode, CUDA runtime can start nre data structure at any position + // assert (dst_start_addr % 32 == 0); for ( unsigned counter = 0; counter < count; counter += 32 ) { const unsigned wr_addr = dst_start_addr + counter; diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc index ed0c25e..65ec113 100644 --- a/src/gpgpu-sim/shader.cc +++ b/src/gpgpu-sim/shader.cc @@ -3091,7 +3091,7 @@ unsigned int shader_core_config::max_cta( const kernel_info_t &k ) const case VOLTA: { //For Volta, we assign the remaining shared memory to L1 cache //For more info about adaptive cache, see https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#shared-memory-7-x - assert(gpgpu_shmem_size == 98304); //Volta has 96 KB shared + //assert(gpgpu_shmem_size == 98304); //Volta has 96 KB shared //To Do: make it flexible and not tuned to 9KB share memory unsigned max_assoc = m_L1D_config.get_max_assoc(); -- cgit v1.3