SDK1.1の結果
systemsim % mysim spu 6 display statistics SPU DD3.0 *** Total Cycle count 185063862 Total Instruction count 643 Total CPI 287813.15 *** Performance Cycle count 185063862 Performance Instruction count 105187041 (95980854) Performance CPI 1.76 (1.93) Branch instructions 5043446 Branch taken 3319266 Branch not taken 1724180 Hint instructions 1371998 Hint hit 2236813 Contention at LS between Load/Store and Prefetch 1864481 Single cycle 65274588 ( 35.3%) Dual cycle 15353133 ( 8.3%) Nop cycle 3521561 ( 1.9%) Stall due to branch miss 23404176 ( 12.6%) Stall due to prefetch miss 14748 ( 0.0%) Stall due to dependency 63853981 ( 34.5%) Stall due to fp resource conflict 26048 ( 0.0%) Stall due to waiting for hint target 2388059 ( 1.3%) Stall due to dp pipeline 2548824 ( 1.4%) Channel stall cycle 8678735 ( 4.7%) SPU Initialization cycle 9 ( 0.0%) ----------------------------------------------------------------------- Total cycle 185063862 (100.0%) Stall cycles due to dependency on each pipelines FX2 5283556 ( 8.3% of all dependency stalls) SHUF 18857474 ( 29.5% of all dependency stalls) FX3 1942732 ( 3.0% of all dependency stalls) LS 17491005 ( 27.4% of all dependency stalls) BR 243373 ( 0.4% of all dependency stalls) SPR 117 ( 0.0% of all dependency stalls) LNOP 0 ( 0.0% of all dependency stalls) NOP 0 ( 0.0% of all dependency stalls) FXB 0 ( 0.0% of all dependency stalls) FP6 15480338 ( 24.2% of all dependency stalls) FP7 3454611 ( 5.4% of all dependency stalls) FPD 1100775 ( 1.7% of all dependency stalls) The number of used registers are 128, the used ratio is 100.00 dumped pipeline stats systemsim %
なぜかSDK1.1(gcc-4.0.2)の方が遅いです。ストールが増えているのが特に悲しい。この結果だけでgcc-4.0.2がだめとは言えないですが、SDK1.0.1の方がよさげです。とはいえFC5に環境を移行したのでSDK1.1で作業を進めます。