Faster PC Histogram (#644)

* macro for declaring dpi functions in profiler

* vanilla_core_trace.def

* adds pc histogram collector

* adds PC_HIST argument to make

* comment note

* refactor histogram structure

* fixes instance name setting

* moves histogram code out of profiler

* fast-functional pc histogram

* cleanup

* scoreboard tracker module

* parameter for classifying long ops

* cleanup in profiler

* simv-pc-histogram

* detailed stall breakdown

* remove pc hist from trace

* bugfix for clearing scoreboard

* counts icache miss in id bubble tracking

* track stall_icache_store

* prioritize stall_all stalls

* revert

* revert

* remove unused file

* revert

* revert

* cleanup

* remove unused ifdef

* revert

* revert

* revert

* revert

* cleanup

* fixup

* stall_ifetch_wait reports pc in mem stage

* scoreboard tracker pkg

* remove debug

* [spmd] stall_remote_ld_wb test

* [testbenches] verilator workaround

* [vanilla_scoreboard_tracker] enum to struct

* [vanilla_core_pc_histogram] omit wb stalls

* [scoreboard_tracker] fixup

* restores wb stalls

* [vanilla_pc_histogram] removes macros, writes out instead instead

* [scoreboard_tracker] fix

* [pc_histogram] style fix

* [pc_histogram] BSG_ABSTRACT_MODULE

* [pc_histogram] bug fix

machines/.gitignore

@@ -1,3 +1,5 @@
+simv-pc-histogram
+simv-pc-histogram.daidir
 simv-debug
 simv-debug.daidir
 simv-profile
@@ -6,11 +8,13 @@ simv-coverage
 simv-coverage.daidir
 vc_hdrs.h
 stack.info.*
+*/build-pc-histogram
 */build-profile
 */build-debug
 */build-coverage
 */build
 */csrc
+*/csrc-pc-histogram
 */csrc-debug
 */csrc-profile
 */csrc-coverage

machines/Makefile.vcs

@@ -119,3 +119,14 @@ ASCII_TO_ROM_PY = $(BASEJUMP_STL_DIR)/bsg_mem/bsg_ascii_to_rom.py
 		$(VCS_INCLUDES) $(VCS_DEFINES) $(CSOURCES) $(VCS_SOURCES) $*/bsg_tag_boot_rom.v \
 		-l $*/build-coverage.log -Mdir=$*/csrc-coverage
 
+
+# build simv-pc-histogram
+# without debug option for faster simulation.
+%/simv-pc-histogram : %/Makefile.machine.include $(VSOURCES) $(CSOURCES) $(VINCLUDES) $(VHEADERS) 
+	$(call set_vcs_machine_variables,$<)
+	python $(POD_TRACE_GEN_PY) $(BSG_MACHINE_PODS_X) $(BSG_MACHINE_PODS_Y) > $*/pod_trace.tr
+	python $(ASCII_TO_ROM_PY) $*/pod_trace.tr bsg_tag_boot_rom > $*/bsg_tag_boot_rom.v
+	$(eval VCS_FLAGS += +define+BSG_ENABLE_PC_HISTOGRAM)
+	$(VCS) $(VCS_FLAGS) $(VCS_CFLAGS) -o $@ \
+		$(VCS_INCLUDES) $(VCS_DEFINES) $(CSOURCES) $(VCS_SOURCES) $*/bsg_tag_boot_rom.v \
+		-l $*/build.log -Mdir=$*/csrc

machines/Makefile.verilator

@@ -117,3 +117,13 @@ ASCII_TO_ROM_PY = $(BASEJUMP_STL_DIR)/bsg_mem/bsg_ascii_to_rom.py
 %/simsc-coverage : %/Makefile.machine.include $(VSOURCES) $(CSOURCES) $(VINCLUDES) $(VHEADERS) 
 	$(info Coverage is not currently supported for Verilator)
 
+# build simsc
+# without debug option for faster simulation.
+%/simsc-pc-histogram : %/Makefile.machine.include $(VSOURCES) $(CSOURCES) $(VINCLUDES) $(VHEADERS)
+	$(call set_verilator_machine_variables,$<)
+	python $(POD_TRACE_GEN_PY) $(BSG_MACHINE_PODS_X) $(BSG_MACHINE_PODS_Y) > $*/pod_trace.tr
+	python $(ASCII_TO_ROM_PY) $*/pod_trace.tr bsg_tag_boot_rom > $*/bsg_tag_boot_rom.v
+	$(eval VERILATOR_FLAGS += +define+BSG_ENABLE_PC_HISTOGRAM)
+	$(VERILATOR) $(VERILATOR_FLAGS) $(VERILATOR_CFLAGS) -Mdir $(@D)/obj_dir -o $(@F) \
+		$(VERILATOR_INCLUDES) $(VERILATOR_DEFINES) $(CSOURCES) $(VERILATOR_SOURCES) $*/bsg_tag_boot_rom.v
+	ln -s $(abspath $(@D))/obj_dir/simsc $@

machines/sim_filelist.mk

@@ -5,6 +5,8 @@ VINCLUDES += $(BSG_MANYCORE_DIR)/testbenches/common/v
 VHEADERS += $(BSG_MANYCORE_DIR)/testbenches/common/v/bsg_manycore_mem_cfg_pkg.v
 VHEADERS += $(BSG_MANYCORE_DIR)/testbenches/common/v/bsg_manycore_network_cfg_pkg.v
 VHEADERS += $(BSG_MANYCORE_DIR)/testbenches/common/v/bsg_manycore_profile_pkg.v
+VHEADERS += $(BSG_MANYCORE_DIR)/testbenches/common/v/vanilla_exe_bubble_classifier_pkg.v
+VHEADERS += $(BSG_MANYCORE_DIR)/testbenches/common/v/vanilla_scoreboard_tracker_pkg.v
 VHEADERS += $(BASEJUMP_STL_DIR)/bsg_test/bsg_dramsim3_pkg.v
 
 VSOURCES += $(BSG_MANYCORE_DIR)/testbenches/common/v/bsg_nonsynth_manycore_axi_mem.v
@@ -59,10 +61,14 @@ VSOURCES += $(BSG_MANYCORE_DIR)/testbenches/common/v/remote_load_trace.v
 CSOURCES += $(BSG_MANYCORE_DIR)/testbenches/common/v/remote_load_profiler.cpp
 VSOURCES += $(BSG_MANYCORE_DIR)/testbenches/common/v/instr_trace.v
 VSOURCES += $(BSG_MANYCORE_DIR)/testbenches/common/v/vanilla_core_profiler.v
+VSOURCES += $(BSG_MANYCORE_DIR)/testbenches/common/v/vanilla_exe_bubble_classifier.v
+VSOURCES += $(BSG_MANYCORE_DIR)/testbenches/common/v/vanilla_scoreboard_tracker.v
 CSOURCES += $(BSG_MANYCORE_DIR)/testbenches/common/v/vanilla_core_profiler.cpp
+CSOURCES += $(BSG_MANYCORE_DIR)/testbenches/common/v/vanilla_core_pc_histogram.cpp
 VSOURCES += $(BSG_MANYCORE_DIR)/testbenches/common/v/vcache_profiler.v
 CSOURCES += $(BSG_MANYCORE_DIR)/testbenches/common/v/vcache_profiler.cpp
 VSOURCES += $(BSG_MANYCORE_DIR)/testbenches/common/v/vcache_non_blocking_profiler.v
+VSOURCES += $(BSG_MANYCORE_DIR)/testbenches/common/v/vanilla_core_pc_histogram.v
 VSOURCES += $(BSG_MANYCORE_DIR)/testbenches/common/v/infinite_mem_profiler.v
 VSOURCES += $(BSG_MANYCORE_DIR)/testbenches/common/v/bsg_nonsynth_manycore_tag_master.v
 VSOURCES += $(BSG_MANYCORE_DIR)/testbenches/common/v/bsg_nonsynth_manycore_io_complex.v

software/mk/Makefile.verilog

@@ -4,6 +4,7 @@
 MAX_CYCLES ?= 100000000
 WAVE       ?= 0
 TRACE      ?= 0
+PC_HIST    ?= 0
 COVERAGE   ?= 0
 PROG_NAME  ?= main
 
@@ -47,6 +48,7 @@ else ifeq ($(BSG_PLATFORM),verilator)
 endif
 
 ifeq ($(WAVE),0)
+ifeq ($(PC_HIST),0)
 ifeq ($(TRACE),0)
 ifeq ($(COVERAGE),0)
   BSG_SIM_EXE = $(BSG_MACHINE_PATH)/$(BSG_SIM_BASE)
@@ -56,6 +58,9 @@ endif
 else
   BSG_SIM_EXE = $(BSG_MACHINE_PATH)/$(BSG_SIM_BASE)-profile
 endif
+else
+  BSG_SIM_EXE = $(BSG_MACHINE_PATH)/$(BSG_SIM_BASE)-pc-histogram
+endif
 else
   BSG_SIM_EXE = $(BSG_MACHINE_PATH)/$(BSG_SIM_BASE)-debug
 endif

software/spmd/stall_remote_ld_wb/Makefile

@@ -0,0 +1,15 @@
+bsg_tiles_X= 1
+bsg_tiles_Y= 1
+
+all: main.run
+
+include ../Makefile.include
+
+RISCV_LINK_OPTS = -march=rv32imaf -nostdlib -nostartfiles
+
+main.riscv: $(LINK_SCRIPT)  main.o
+	$(RISCV_LINK)  main.o -o $@ $(RISCV_LINK_OPTS)
+
+
+include ../../mk/Makefile.tail_rules
+

software/spmd/stall_remote_ld_wb/main.S

@@ -0,0 +1,67 @@
+#include "bsg_manycore_arch.h"
+#include "bsg_manycore_asm.h"
+
+//.data
+//  float_dmem_arr: .space 64 // array of 16 floats
+//.section .dram, "aw"
+//  float_dram_arr: .space 64
+
+.section .dram, "aw"
+dram_word: .word -1
+
+.text
+        bsg_asm_init_regfile
+
+// test1
+test1:
+        la       x1, dram_word
+        addi     x2, x0, -1
+        // prime
+        lw       x3,   0(x1)
+        lw       x4,  64(x1)
+        bne      x2, x3, fail
+        // do many remote loads
+        // first block
+        lw       x4,   0(x1)
+        lw       x5,   4(x1)
+        lw       x6,   8(x1)
+        lw       x7,  12(x1)
+        lw       x8,  16(x1)
+        lw       x9,  20(x1)
+        lw      x10,  24(x1)
+        lw      x11,  28(x1)
+        lw      x12,  32(x1)
+        lw      x13,  36(x1)
+        lw      x14,  40(x1)
+        lw      x15,  44(x1)
+        lw      x16,  48(x1)
+        lw      x17,  52(x1)
+        lw      x18,  56(x1)
+        lw      x19,  60(x1)
+        // second block
+        lw      x20,  64(x1)
+        lw      x21,  68(x1)
+        lw      x22,  72(x1)
+        lw      x23,  76(x1)
+        lw      x24,  80(x1)
+        lw      x25,  84(x1)
+        lw      x26,  88(x1)
+        lw      x27,  92(x1)
+        lw      x28,  96(x1)
+        lw      x29, 100(x1)
+        lw      x30, 104(x1)
+        lw      x31, 108(x1)
+        div     x1, x3, x2
+        // lw      x4, 112(x1)
+        // lw      x4, 116(x1)
+        // lw      x4, 120(x1)
+        // lw      x4, 124(x1)
+        bne     x3, x2, fail
+pass:
+        bsg_asm_finish(IO_X_INDEX, 0)
+pass_loop:
+        beq     x0, x0, pass_loop
+fail:
+        bsg_asm_fail(IO_X_INDEX, 0)
+fail_loop:
+        beq     x0, x0, fail_loop

testbenches/common/v/bsg_nonsynth_manycore_testbench.v

@@ -58,6 +58,8 @@ module bsg_nonsynth_manycore_testbench
 
     , parameter enable_vcore_pc_coverage_p=0
 
+    , parameter enable_vanilla_core_pc_histogram_p=0
+
     , parameter cache_bank_addr_width_lp = `BSG_SAFE_CLOG2(bsg_dram_size_p/(2*num_tiles_x_p*num_vcache_rows_p)*4) // byte addr
     , parameter link_sif_width_lp =
       `bsg_manycore_link_sif_width(addr_width_p,data_width_p,x_cord_width_p,y_cord_width_p)
@@ -913,6 +915,24 @@ if (enable_vanilla_core_trace_p) begin
 end
 `endif
 
+  //////////////////
+  // PC Histogram //
+  //////////////////
+`ifndef VERILATOR_WORKAROUND_DISABLE_PC_HISTOGRAM
+if (enable_vanilla_core_pc_histogram_p) begin
+  bind vanilla_core vanilla_core_pc_histogram
+    #(.x_cord_width_p(x_cord_width_p)
+      ,.y_cord_width_p(y_cord_width_p)
+      ,.data_width_p(data_width_p)
+      ,.icache_tag_width_p(icache_tag_width_p)
+      ,.icache_entries_p(icache_entries_p)
+      ,.origin_x_cord_p(`BSG_MACHINE_ORIGIN_X_CORD)
+      ,.origin_y_cord_p(`BSG_MACHINE_ORIGIN_Y_CORD)
+      )
+  vcore_pc_hist
+    (.*);
+`endif
+end // if (enable_vanilla_core_pc_histogram_p)
 
 endmodule
 

testbenches/common/v/profiler.vh

@@ -29,6 +29,14 @@ import "DPI-C" context function \
 import "DPI-C" context function \
   void profiler_name``_unlock();
 
+`define DECLARE_PROFILER_DPI_FUNCTIONS(profiler_name) \
+  `DECLARE_PROFILER_INIT_FUNC(profiler_name) \
+  `DECLARE_PROFILER_EXIT_FUNC(profiler_name) \
+  `DECLARE_PROFILER_IS_INIT_FUNC(profiler_name) \
+  `DECLARE_PROFILER_IS_EXIT_FUNC(profiler_name) \
+  `DECLARE_PROFILER_TRACE_FD_FUNC(profiler_name) \
+  `DECLARE_PROFILER_LOCK_FUNC(profiler_name) \
+  `DECLARE_PROFILER_UNLOCK_FUNC(profiler_name)
 
 `define DEFINE_PROFILER_INITIAL_BLOCK(profiler_name, trace_file_name, trace_file_header) \
 int init_trace_fd; \
@@ -55,13 +63,7 @@ end
   end
 
 `define DEFINE_PROFILER(profiler_name, trace_file_name, trace_file_header) \
-  `DECLARE_PROFILER_INIT_FUNC(profiler_name) \
-  `DECLARE_PROFILER_EXIT_FUNC(profiler_name) \
-  `DECLARE_PROFILER_IS_INIT_FUNC(profiler_name) \
-  `DECLARE_PROFILER_IS_EXIT_FUNC(profiler_name) \
-  `DECLARE_PROFILER_TRACE_FD_FUNC(profiler_name) \
-  `DECLARE_PROFILER_LOCK_FUNC(profiler_name) \
-  `DECLARE_PROFILER_UNLOCK_FUNC(profiler_name) \
+  `DECLARE_PROFILER_DPI_FUNCTIONS(profiler_name) \
   `DEFINE_PROFILER_INITIAL_BLOCK(profiler_name, trace_file_name, trace_file_header) \
   `DEFINE_PROFILER_FINAL_BLOCK(profiler_name)
 

testbenches/common/v/spmd_testbench.v

@@ -136,6 +136,9 @@ module spmd_testbench
 `endif
 `ifdef BSG_ENABLE_VANILLA_CORE_TRACE
     ,.enable_vanilla_core_trace_p(1)
+`endif
+`ifdef BSG_ENABLE_PC_HISTOGRAM
+    ,.enable_vanilla_core_pc_histogram_p(1)
 `endif
   // DR: If the instance name is changed, the bind statements in the
   // file where this module is defined, and header strings in the