74 lines · plain
1# NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py2# RUN: llvm-mca -march=aarch64 -mcpu=exynos-m3 -resource-pressure=false < %s | FileCheck %s -check-prefixes=ALL,M33# RUN: llvm-mca -march=aarch64 -mcpu=exynos-m4 -resource-pressure=false < %s | FileCheck %s -check-prefixes=ALL,M44# RUN: llvm-mca -march=aarch64 -mcpu=exynos-m5 -resource-pressure=false < %s | FileCheck %s -check-prefixes=ALL,M55 6fsqrt s31, s317 8# Newton series for sqrtf().9frsqrte s1, s010fmul s2, s1, s111frsqrts s2, s0, s212fmul s1, s1, s213fmul s2, s1, s114frsqrts s2, s0, s215fmul s2, s2, s016fmul s1, s1, s217fcmp s0, #0.018fcsel s0, s0, s1, eq19 20# ALL: Iterations: 10021# ALL-NEXT: Instructions: 110022 23# M3-NEXT: Total Cycles: 320324# M4-NEXT: Total Cycles: 310325# M5-NEXT: Total Cycles: 280326 27# ALL-NEXT: Total uOps: 120028 29# ALL: Dispatch Width: 630 31# M3-NEXT: uOps Per Cycle: 0.3732# M3-NEXT: IPC: 0.3433# M3-NEXT: Block RThroughput: 20.034 35# M4-NEXT: uOps Per Cycle: 0.3936# M4-NEXT: IPC: 0.3537# M4-NEXT: Block RThroughput: 2.338 39# M5-NEXT: uOps Per Cycle: 0.4340# M5-NEXT: IPC: 0.3941# M5-NEXT: Block RThroughput: 2.342 43# ALL: Instruction Info:44# ALL-NEXT: [1]: #uOps45# ALL-NEXT: [2]: Latency46# ALL-NEXT: [3]: RThroughput47# ALL-NEXT: [4]: MayLoad48# ALL-NEXT: [5]: MayStore49# ALL-NEXT: [6]: HasSideEffects (U)50 51# ALL: [1] [2] [3] [4] [5] [6] Instructions:52 53# M3-NEXT: 1 18 19.00 fsqrt s31, s3154# M3-NEXT: 1 4 0.50 frsqrte s1, s055 56# M4-NEXT: 1 8 1.75 fsqrt s31, s3157# M4-NEXT: 1 3 0.50 frsqrte s1, s058 59# M5-NEXT: 1 8 1.25 fsqrt s31, s3160# M5-NEXT: 1 3 0.50 frsqrte s1, s061 62# ALL-NEXT: 1 3 0.33 fmul s2, s1, s163# ALL-NEXT: 1 4 0.33 frsqrts s2, s0, s264# ALL-NEXT: 1 3 0.33 fmul s1, s1, s265# ALL-NEXT: 1 3 0.33 fmul s2, s1, s166# ALL-NEXT: 1 4 0.33 frsqrts s2, s0, s267# ALL-NEXT: 1 3 0.33 fmul s2, s2, s068# ALL-NEXT: 1 3 0.33 fmul s1, s1, s269# ALL-NEXT: 1 2 1.00 fcmp s0, #0.070 71# M3-NEXT: 2 5 1.00 fcsel s0, s0, s1, eq72# M4-NEXT: 2 5 1.00 fcsel s0, s0, s1, eq73# M5-NEXT: 2 2 1.00 fcsel s0, s0, s1, eq74