wu-arch/vortex
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Divide by per-breakdown cycle for avg stall cycles

Browse Source
This commit is contained in:
Hansung Kim
2024-04-03 15:29:51 -07:00
parent 62c7d1f4cf
commit 6c632200d5
2 changed files with 11 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
hw/rtl/core/VX_core.sv
View File

@@ -436,31 +436,31 @@ module VX_core import VX_gpu_pkg::*; #(
pipeline_perf_if.scb_stalls, `ISSUE_WIDTH); pipeline_perf_if.scb_stalls, `ISSUE_WIDTH);
$display("issue scoreboard: stalls by operand hazard: alu %d (%2.2f cycles per issue)", $display("issue scoreboard: stalls by operand hazard: alu %d (%2.2f cycles per issue)",
scrb_alu_per_core, scrb_alu_per_core,
$itor(scrb_alu_per_core) / $itor(dispatch_fires_total)); $itor(scrb_alu_per_core) / $itor(pipeline_perf_if.dispatch_fires[`EX_ALU]));
$display("issue scoreboard: stalls by operand hazard: fpu %d (%2.2f cycles per issue)", $display("issue scoreboard: stalls by operand hazard: fpu %d (%2.2f cycles per issue)",
scrb_fpu_per_core, scrb_fpu_per_core,
$itor(scrb_fpu_per_core) / $itor(dispatch_fires_total)); $itor(scrb_fpu_per_core) / $itor(pipeline_perf_if.dispatch_fires[`EX_FPU]));
$display("issue scoreboard: stalls by operand hazard: lsu %d (%2.2f cycles per issue)", $display("issue scoreboard: stalls by operand hazard: lsu %d (%2.2f cycles per issue)",
scrb_lsu_per_core, scrb_lsu_per_core,
$itor(scrb_lsu_per_core) / $itor(dispatch_fires_total)); $itor(scrb_lsu_per_core) / $itor(pipeline_perf_if.dispatch_fires[`EX_LSU]));
$display("issue scoreboard: stalls by operand hazard: sfu %d (%2.2f cycles per issue)", $display("issue scoreboard: stalls by operand hazard: sfu %d (%2.2f cycles per issue)",
scrb_sfu_per_core, scrb_sfu_per_core,
$itor(scrb_sfu_per_core) / $itor(dispatch_fires_total)); $itor(scrb_sfu_per_core) / $itor(pipeline_perf_if.dispatch_fires[`EX_SFU]));
$display("issue scoreboard: sfu stalls: %d (scrs=%f, wctl=%f)",pipeline_perf_if.units_uses[`EX_SFU], $display("issue scoreboard: sfu stalls: %d (scrs=%f, wctl=%f)",pipeline_perf_if.units_uses[`EX_SFU],
$itor(scrb_csrs_per_core) / $itor(sfu_tot) * 100.0, $itor(scrb_csrs_per_core) / $itor(sfu_tot) * 100.0,
$itor(scrb_wctl_per_core) / $itor(sfu_tot) * 100.0); $itor(scrb_wctl_per_core) / $itor(sfu_tot) * 100.0);
$display("issue dispatch: stalls by FU busy: alu %d (%2.2f cycles per issue)", $display("issue dispatch: stalls by FU busy: alu %d (%2.2f cycles per issue)",
pipeline_perf_if.dispatch_stalls[`EX_ALU], pipeline_perf_if.dispatch_stalls[`EX_ALU],
$itor(pipeline_perf_if.dispatch_stalls[`EX_ALU]) / $itor(dispatch_fires_total)); $itor(pipeline_perf_if.dispatch_stalls[`EX_ALU]) / $itor(pipeline_perf_if.dispatch_fires[`EX_ALU]));
$display("issue dispatch: stalls by FU busy: fpu %d (%2.2f cycles per issue)", $display("issue dispatch: stalls by FU busy: fpu %d (%2.2f cycles per issue)",
pipeline_perf_if.dispatch_stalls[`EX_FPU], pipeline_perf_if.dispatch_stalls[`EX_FPU],
$itor(pipeline_perf_if.dispatch_stalls[`EX_FPU]) / $itor(dispatch_fires_total)); $itor(pipeline_perf_if.dispatch_stalls[`EX_FPU]) / $itor(pipeline_perf_if.dispatch_fires[`EX_FPU]));
$display("issue dispatch: stalls by FU busy: lsu %d (%2.2f cycles per issue)", $display("issue dispatch: stalls by FU busy: lsu %d (%2.2f cycles per issue)",
pipeline_perf_if.dispatch_stalls[`EX_LSU], pipeline_perf_if.dispatch_stalls[`EX_LSU],
$itor(pipeline_perf_if.dispatch_stalls[`EX_LSU]) / $itor(dispatch_fires_total)); $itor(pipeline_perf_if.dispatch_stalls[`EX_LSU]) / $itor(pipeline_perf_if.dispatch_fires[`EX_LSU]));
$display("issue dispatch: stalls by FU busy: sfu %d (%2.2f cycles per issue)", $display("issue dispatch: stalls by FU busy: sfu %d (%2.2f cycles per issue)",
pipeline_perf_if.dispatch_stalls[`EX_SFU], pipeline_perf_if.dispatch_stalls[`EX_SFU],
$itor(pipeline_perf_if.dispatch_stalls[`EX_SFU]) / $itor(dispatch_fires_total)); $itor(pipeline_perf_if.dispatch_stalls[`EX_SFU]) / $itor(pipeline_perf_if.dispatch_fires[`EX_SFU]));
$display("issue dispatch: fires: total %d", $display("issue dispatch: fires: total %d",
dispatch_fires_total); dispatch_fires_total);
$display("issue dispatch: fires: alu %d", $display("issue dispatch: fires: alu %d",

3
hw/rtl/core/VX_operands.sv
View File

@@ -197,6 +197,9 @@ module VX_operands import VX_gpu_pkg::*; #(
assign stg_valid_in = scoreboard_if[i].valid && data_ready; assign stg_valid_in = scoreboard_if[i].valid && data_ready;
assign scoreboard_if[i].ready = stg_ready_in && data_ready; assign scoreboard_if[i].ready = stg_ready_in && data_ready;
// NOTE(hansung): toggle_buffer is 1-reg pipe without flow, halving
// throughput. Wouldn't this cap overall IPC? Or OK as long as
// ISSUE_WIDTH > 1?
VX_toggle_buffer #( VX_toggle_buffer #(
.DATAW (DATAW) .DATAW (DATAW)
) staging_buffer ( ) staging_buffer (