add python stats package

romnn · Sep 1, 2023 · 9fbed9d · 9fbed9d
1 parent 3d51cdd
commit 9fbed9d
Show file tree

Hide file tree

Showing 10 changed files with 486 additions and 263 deletions.
diff --git a/Cargo.lock b/Cargo.lock
diff --git a/README.md b/README.md
@@ -94,7 +94,9 @@ rustup component add llvm-tools-preview
 cargo install grcov
 
 # collect code coverage in tests (todo)
-cargo xtasks coverage
+cargo xtask coverage
+
+cargo xtask accelsim convert-config -c ./accelsim/gtx1080/gpgpusim.config -c ./accelsim/gtx1080/gpgpusim.trace.config
 ```
 
 Publishing traces (used by CI)

diff --git a/accelsim/gtx1080/gpgpusim.config.yml b/accelsim/gtx1080/gpgpusim.config.yml
@@ -0,0 +1,248 @@
+shader_core:
+  gpgpu_simd_model: 1
+  gpgpu_shader_core_pipeline: 2048:32
+  gpgpu_tex_cache_l1: N:16:128:24,L:R:m:N:L,F:128:4,128:2
+  gpgpu_const_cache_l1: N:128:64:2,L:R:f:N:L,A:2:64,4
+  gpgpu_cache_il1: N:8:128:4,L:R:f:N:L,A:2:48,4
+  gpgpu_cache_dl1: N:64:128:6,L:L:m:N:H,A:128:8,8
+  gpgpu_l1_cache_write_ratio: 0
+  gpgpu_l1_banks: 1
+  gpgpu_l1_banks_byte_interleaving: 32
+  gpgpu_l1_banks_hashing_function: 0
+  gpgpu_l1_latency: 1
+  gpgpu_smem_latency: 3
+  gpgpu_cache_dl1_pref_l1: none
+  gpgpu_cache_dl1_pref_shared: none
+  gpgpu_gmem_skip_l1d: true
+  gpgpu_perfect_mem: false
+  n_regfile_gating_group: 4
+  gpgpu_clock_gated_reg_file: false
+  gpgpu_clock_gated_lanes: false
+  gpgpu_shader_registers: 65536
+  gpgpu_registers_per_block: 8192
+  gpgpu_ignore_resources_limitation: false
+  gpgpu_shader_cta: 32
+  gpgpu_num_cta_barriers: 16
+  gpgpu_n_clusters: 20
+  gpgpu_n_cores_per_cluster: 1
+  gpgpu_n_cluster_ejection_buffer_size: 8
+  gpgpu_n_ldst_response_buffer_size: 2
+  gpgpu_shmem_per_block: 49152
+  gpgpu_shmem_size: 98304
+  gpgpu_shmem_option: 0
+  gpgpu_unified_l1d_size: 0
+  gpgpu_adaptive_cache_config: false
+  gpgpu_shmem_size_default: 16384
+  gpgpu_shmem_size_pref_l1: 16384
+  gpgpu_shmem_size_pref_shared: 16384
+  gpgpu_shmem_num_banks: 32
+  gpgpu_shmem_limited_broadcast: 0
+  gpgpu_shmem_warp_parts: 1
+  gpgpu_mem_unit_ports: 1
+  gpgpu_warpdistro_shader: -1
+  gpgpu_warp_issue_shader: 0
+  gpgpu_local_mem_map: true
+  gpgpu_num_reg_banks: 32
+  gpgpu_reg_bank_use_warp_id: false
+  gpgpu_sub_core_model: false
+  gpgpu_enable_specialized_operand_collector: true
+  gpgpu_operand_collector_num_units_sp: 20
+  gpgpu_operand_collector_num_units_dp: 0
+  gpgpu_operand_collector_num_units_sfu: 4
+  gpgpu_operand_collector_num_units_int: 0
+  gpgpu_operand_collector_num_units_tensor_core: 4
+  gpgpu_operand_collector_num_units_mem: 8
+  gpgpu_operand_collector_num_units_gen: 0
+  gpgpu_operand_collector_num_in_ports_sp: 4
+  gpgpu_operand_collector_num_in_ports_dp: 0
+  gpgpu_operand_collector_num_in_ports_sfu: 1
+  gpgpu_operand_collector_num_in_ports_int: 0
+  gpgpu_operand_collector_num_in_ports_tensor_core: 1
+  gpgpu_operand_collector_num_in_ports_mem: 1
+  gpgpu_operand_collector_num_in_ports_gen: 0
+  gpgpu_operand_collector_num_out_ports_sp: 4
+  gpgpu_operand_collector_num_out_ports_dp: 0
+  gpgpu_operand_collector_num_out_ports_sfu: 1
+  gpgpu_operand_collector_num_out_ports_int: 0
+  gpgpu_operand_collector_num_out_ports_tensor_core: 1
+  gpgpu_operand_collector_num_out_ports_mem: 1
+  gpgpu_operand_collector_num_out_ports_gen: 0
+  gpgpu_coalesce_arch: 13
+  gpgpu_num_sched_per_core: 2
+  gpgpu_max_insn_issue_per_warp: 2
+  gpgpu_dual_issue_diff_exec_units: true
+  gpgpu_simt_core_sim_order: 1
+  gpgpu_pipeline_widths: 4,0,0,1,1,4,0,0,1,1,6
+  gpgpu_tensor_core_avail: 0
+  gpgpu_num_sp_units: 4
+  gpgpu_num_dp_units: 0
+  gpgpu_num_int_units: 0
+  gpgpu_num_sfu_units: 1
+  gpgpu_num_tensor_core_units: 0
+  gpgpu_num_mem_units: 1
+  gpgpu_scheduler: gto
+  gpgpu_concurrent_kernel_sm: false
+  gpgpu_perfect_inst_const_cache: false
+  gpgpu_inst_fetch_throughput: 1
+  gpgpu_reg_file_port_throughput: 1
+  specialized_unit_1: 0,4,4,4,4,BRA
+  specialized_unit_2: 0,4,4,4,4,BRA
+  specialized_unit_3: 0,4,4,4,4,BRA
+  specialized_unit_4: 0,4,4,4,4,BRA
+  specialized_unit_5: 0,4,4,4,4,BRA
+  specialized_unit_6: 0,4,4,4,4,BRA
+  specialized_unit_7: 0,4,4,4,4,BRA
+  specialized_unit_8: 0,4,4,4,4,BRA
+ptx:
+  g_save_embedded_ptx: false
+  keep: false
+  g_ptx_save_converted_ptxplus: false
+  g_occupancy_sm_number: 60
+  opcode_latency_int: 4,13,4,5,145
+  opcode_latency_fp: 4,13,4,5,39
+  opcode_latency_dp: 8,19,8,8,330
+  opcode_latency_sfu: '8'
+  opcode_latency_tensor: '64'
+  opcode_initiation_int: 1,2,2,2,8
+  opcode_initiation_fp: 1,2,1,1,4
+  opcode_initiation_dp: 1,2,1,1,130
+  opcode_initiation_sfu: '8'
+  opcode_initiation_tensor: '64'
+  cdp_latency_str: 7200,8000,100,12000,1600
+trace:
+  traces_filename: ./traces/kernelslist.g
+  trace_opcode_latency_initiation_int: 4,1
+  trace_opcode_latency_initiation_sp: 4,1
+  trace_opcode_latency_initiation_dp: 4,1
+  trace_opcode_latency_initiation_sfu: 4,1
+  trace_opcode_latency_initiation_tensor: 4,1
+  trace_opcode_latency_initiation_spec_op_1: 4,4
+  trace_opcode_latency_initiation_spec_op_2: 4,4
+  trace_opcode_latency_initiation_spec_op_3: 4,4
+  trace_opcode_latency_initiation_spec_op_4: 4,4
+  trace_opcode_latency_initiation_spec_op_5: 4,4
+  trace_opcode_latency_initiation_spec_op_6: 4,4
+  trace_opcode_latency_initiation_spec_op_7: 4,4
+  trace_opcode_latency_initiation_spec_op_8: 4,4
+sim:
+  gpu_max_cycle_opt: 0
+  gpu_max_insn_opt: 0
+  gpu_max_cta_opt: 0
+  gpu_max_completed_cta_opt: 0
+  gpgpu_runtime_stat: '500'
+  liveness_message_freq: 1
+  gpgpu_compute_capability_major: 7
+  gpgpu_compute_capability_minor: 0
+  gpgpu_flush_l1_cache: false
+  gpgpu_flush_l2_cache: false
+  gpu_deadlock_detect: true
+  gpgpu_ptx_instruction_classification: 0
+  g_ptx_sim_mode: 0
+  gpgpu_clock_domains: 1607.0:1607.0:1607.0:2500.0
+  max_concurrent_kernel: 32
+  gpgpu_cflog_interval: 0
+  g_visualizer_enabled: true
+  g_visualizer_filename: null
+  g_visualizer_zlevel: 6
+  stack_size_limit: 1024
+  heap_size_limit: 8388608
+  runtime_sync_depth_limit: 2
+  runtime_pending_launch_count_limit: 2048
+  trace_enabled: false
+  trace_config_str: none
+  trace_sampling_core: 0
+  trace_sampling_memory_partition: -1
+  g_kernel_launch_latency: 0
+  g_cdp_enabled: false
+  g_tb_launch_latency: 0
+dram_timing:
+  nbk: null
+  t_ccd: null
+  t_rrd: null
+  t_rdc: null
+  t_ras: null
+  t_rp: null
+  t_rc: null
+  t_cdlr: null
+  t_wr: null
+  cl: null
+  wl: null
+  nbkgrp: 1
+  t_ccdl: 0
+  t_rtpl: 0
+functional:
+  m_ptx_use_cuobjdump: true
+  m_experimental_lib_support: false
+  checkpoint_option: 0
+  checkpoint_kernel: 1
+  checkpoint_cta: 0
+  resume_option: 0
+  resume_kernel: 0
+  resume_cta: 0
+  checkpoint_cta_t: 0
+  checkpoint_insn_y: 0
+  m_ptx_convert_to_ptxplus: false
+  m_ptx_force_max_capability: 60
+  g_ptx_inst_debug_to_file: false
+  g_ptx_inst_debug_file: inst_debug.txt
+  g_ptx_inst_debug_thread_uid: 1
+interconn:
+  g_network_mode: 1
+  g_network_config_filename: config_fermi_islip.icnt
+  in_buffer_limit: 64
+  out_buffer_limit: 64
+  subnets: 2
+  arbiter_algo: 1
+  verbose: 0
+  grant_cycles: 1
+memory:
+  perf_sim_memcpy: true
+  simple_dram_model: false
+  scheduler_type: 1
+  gpgpu_l2_queue_config: 8:8:8:8
+  l2_ideal: false
+  l2_config_string: N:64:128:16,L:B:m:W:L,A:1024:1024,4:0,32
+  l2_texure_only: false
+  n_mem: 8
+  n_sub_partition_per_memory_channel: 2
+  gpu_n_mem_per_ctrlr: 1
+  gpgpu_memlatency_stat: 14
+  gpgpu_frfcfs_dram_sched_queue_size: 64
+  gpgpu_dram_return_queue_size: 116
+  dram_bus_width: 4
+  dram_burst_length: 8
+  data_command_freq_ratio: 4
+  gpgpu_dram_timing_opt: |-
+    "nbk=16:CCD=2:RRD=6:RCD=12:RAS=28:RP=12:RC=40:
+                            CL=12:WL=4:CDLR=5:WR=12:nbkgrp=1:CCDL=0:RTPL=0"
+  rop_latency: 120
+  dram_latency: 100
+  dual_bus_interface: 0
+  dram_bnk_indexing_policy: 0
+  dram_bnkgrp_indexing_policy: 0
+  seperate_write_queue_enabled: false
+  write_queue_size_opt: 32:28:16
+  elimnate_rw_turnaround: false
+  icnt_flit_size: 32
+  address_mapping:
+    addrdec_option: dramid@8;00000000.00000000.00000000.00000000.0000RRRR.RRRRRRRR.RBBBCCCC.BCCSSSSS
+    run_test: false
+    gpgpu_mem_address_mask: 1
+    memory_partition_indexing: 0
+unknown:
+- --enable_ptx_file_line_stats=1
+- --visualizer_enabled=0
+- --power_simulation_enabled=0
+- --trace_opcode_latency_initiation_int=2,2
+- --trace_opcode_latency_initiation_sp=2,1
+- --trace_opcode_latency_initiation_dp=64,64
+- --trace_opcode_latency_initiation_sfu=21,8
+- --trace_opcode_latency_initiation_tensor=32,32
+- --specialized_unit_1=1,4,4,4,4,BRA
+- --trace_opcode_latency_initiation_spec_op_1=4,4
+- --specialized_unit_2=1,4,200,4,4,TEX
+- --trace_opcode_latency_initiation_spec_op_2=200,4
+- --specialized_unit_3=1,4,32,4,4,TENSOR
+- --trace_opcode_latency_initiation_spec_op_3=32,32
+- --specialized_unit_4=1,4,4,4,4,UDP
+- --trace_opcode_latency_initiation_spec_op_4=4,1
diff --git a/gpucachesim/benchmarks.py b/gpucachesim/benchmarks.py
@@ -2,14 +2,43 @@
 import yaml
 from pathlib import Path
 from os import PathLike
-from typing import Optional
+import typing
 
 from gpucachesim import ROOT_DIR
 
 REPO_ROOT_DIR = ROOT_DIR.parent
 DEFAULT_BENCH_FILE = REPO_ROOT_DIR / "test-apps/test-apps-materialized.yml"
 
 
+class SimConfig(typing.TypedDict):
+    gpgpu_clock_domains: str
+
+    # @property
+    # def core_clock_speed(self) -> int:
+    #     self.gpgpu_clock_domains()
+    #
+    # @property
+    # def num_cores(self) -> int:
+    #     kk
+
+
+class GPUConfig(typing.TypedDict):
+    sim: SimConfig
+
+
+class ProfileConfig(typing.TypedDict):
+    profile_dir: PathLike
+
+
+class SimulateConfig(typing.TypedDict):
+    profile_dir: PathLike
+
+
+class BenchConfig(typing.TypedDict):
+    profile: ProfileConfig
+    simulate: SimulateConfig
+
+
 class Benchmarks:
     def __init__(self, path: PathLike) -> None:
         """load the materialized benchmark config"""

diff --git a/gpucachesim/stats/__init__.py b/gpucachesim/stats/__init__.py
@@ -1,15 +1,20 @@
 import click
+import yaml
 
 import gpucachesim.stats.stats as stats
 import gpucachesim.stats.native as native
-from gpucachesim.benchmarks import Benchmarks
+from gpucachesim.benchmarks import Benchmarks, GPUConfig, REPO_ROOT_DIR
+
+
+DEFAULT_CONFIG_FILE = REPO_ROOT_DIR / "./accelsim/gtx1080/gpgpusim.config.yml"
 
 
 @click.command()
 @click.option("--path", help="Path to materialized benchmark config")
+@click.option("--config", default=DEFAULT_CONFIG_FILE, help="Path to GPU config")
 @click.option("--bench", help="Benchmark name")
 @click.option("--input", default=0, help="Input index")
-def main(path, bench, input):
+def main(path, config, bench, input):
     from pprint import pprint
 
     b = Benchmarks(path)
@@ -19,8 +24,15 @@ def main(path, bench, input):
     bench_config = b.get_bench_config(bench, input)
     # pprint(bench_config)
 
-    our_stats = stats.Stats(bench_config["simulate"])
-    native_stats = native.Stats(bench_config["simulate"])
+    with open(config, "rb") as f:
+        config: GPUConfig = yaml.safe_load(f)
+
+    pprint(config)
+    our_stats = stats.Stats(bench_config)
+    native_stats = native.Stats(config, bench_config)
+
+    print(native_stats.cycles())
+    print(our_stats.cycles())
 
 
 if __name__ == "__main__":