97 lines · plain
1# NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py2# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1500 -timeline -timeline-max-iterations=3 < %s | FileCheck %s3 4# perf stat reports a throughput of 0.60 IPC for this code snippet.5# Each lzcnt has a false dependency on %ecx; the first lzcnt has to wait on the6# imul. However, the folded load can start immediately.7# The last lzcnt has a false dependency on %cx. However, even in this case, the8# folded load can start immediately.9 10imul %edx, %ecx11lzcnt (%rsp), %cx12lzcnt 2(%rsp), %cx13 14# CHECK: Iterations: 150015# CHECK-NEXT: Instructions: 450016# CHECK-NEXT: Total Cycles: 1200317# CHECK-NEXT: Total uOps: 750018 19# CHECK: Dispatch Width: 420# CHECK-NEXT: uOps Per Cycle: 0.6221# CHECK-NEXT: IPC: 0.3722# CHECK-NEXT: Block RThroughput: 4.023 24# CHECK: Instruction Info:25# CHECK-NEXT: [1]: #uOps26# CHECK-NEXT: [2]: Latency27# CHECK-NEXT: [3]: RThroughput28# CHECK-NEXT: [4]: MayLoad29# CHECK-NEXT: [5]: MayStore30# CHECK-NEXT: [6]: HasSideEffects (U)31 32# CHECK: [1] [2] [3] [4] [5] [6] Instructions:33# CHECK-NEXT: 1 4 2.00 imull %edx, %ecx34# CHECK-NEXT: 2 6 2.00 * lzcntw (%rsp), %cx35# CHECK-NEXT: 2 6 2.00 * lzcntw 2(%rsp), %cx36 37# CHECK: Resources:38# CHECK-NEXT: [0.0] - PdAGLU0139# CHECK-NEXT: [0.1] - PdAGLU0140# CHECK-NEXT: [1] - PdBranch41# CHECK-NEXT: [2] - PdCount42# CHECK-NEXT: [3] - PdDiv43# CHECK-NEXT: [4] - PdEX044# CHECK-NEXT: [5] - PdEX145# CHECK-NEXT: [6] - PdFPCVT46# CHECK-NEXT: [7.0] - PdFPFMA47# CHECK-NEXT: [7.1] - PdFPFMA48# CHECK-NEXT: [8.0] - PdFPMAL49# CHECK-NEXT: [8.1] - PdFPMAL50# CHECK-NEXT: [9] - PdFPMMA51# CHECK-NEXT: [10] - PdFPSTO52# CHECK-NEXT: [11] - PdFPU053# CHECK-NEXT: [12] - PdFPU154# CHECK-NEXT: [13] - PdFPU255# CHECK-NEXT: [14] - PdFPU356# CHECK-NEXT: [15] - PdFPXBR57# CHECK-NEXT: [16.0] - PdLoad58# CHECK-NEXT: [16.1] - PdLoad59# CHECK-NEXT: [17] - PdMul60# CHECK-NEXT: [18] - PdStore61 62# CHECK: Resource pressure per iteration:63# CHECK-NEXT: [0.0] [0.1] [1] [2] [3] [4] [5] [6] [7.0] [7.1] [8.0] [8.1] [9] [10] [11] [12] [13] [14] [15] [16.0] [16.1] [17] [18]64# CHECK-NEXT: 3.00 3.00 - - - 4.00 1.00 - - - - - - - - - - - - 3.00 3.00 2.00 -65 66# CHECK: Resource pressure by instruction:67# CHECK-NEXT: [0.0] [0.1] [1] [2] [3] [4] [5] [6] [7.0] [7.1] [8.0] [8.1] [9] [10] [11] [12] [13] [14] [15] [16.0] [16.1] [17] [18] Instructions:68# CHECK-NEXT: - - - - - - 1.00 - - - - - - - - - - - - - - 2.00 - imull %edx, %ecx69# CHECK-NEXT: - 3.00 - - - 2.00 - - - - - - - - - - - - - - 3.00 - - lzcntw (%rsp), %cx70# CHECK-NEXT: 3.00 - - - - 2.00 - - - - - - - - - - - - - 3.00 - - - lzcntw 2(%rsp), %cx71 72# CHECK: Timeline view:73# CHECK-NEXT: 012345678974# CHECK-NEXT: Index 0123456789 012345675 76# CHECK: [0,0] DeeeeER . . . .. imull %edx, %ecx77# CHECK-NEXT: [0,1] DeeeeeeER . . . .. lzcntw (%rsp), %cx78# CHECK-NEXT: [0,2] .D=eeeeeeER . . .. lzcntw 2(%rsp), %cx79# CHECK-NEXT: [1,0] .D=======eeeeER. . .. imull %edx, %ecx80# CHECK-NEXT: [1,1] . D======eeeeeeER . .. lzcntw (%rsp), %cx81# CHECK-NEXT: [1,2] . D========eeeeeeER . .. lzcntw 2(%rsp), %cx82# CHECK-NEXT: [2,0] . D=============eeeeER .. imull %edx, %ecx83# CHECK-NEXT: [2,1] . D=============eeeeeeER.. lzcntw (%rsp), %cx84# CHECK-NEXT: [2,2] . D==============eeeeeeER lzcntw 2(%rsp), %cx85 86# CHECK: Average Wait times (based on the timeline view):87# CHECK-NEXT: [0]: Executions88# CHECK-NEXT: [1]: Average time spent waiting in a scheduler's queue89# CHECK-NEXT: [2]: Average time spent waiting in a scheduler's queue while ready90# CHECK-NEXT: [3]: Average time elapsed from WB until retire stage91 92# CHECK: [0] [1] [2] [3]93# CHECK-NEXT: 0. 3 7.7 0.3 0.0 imull %edx, %ecx94# CHECK-NEXT: 1. 3 7.3 0.0 0.0 lzcntw (%rsp), %cx95# CHECK-NEXT: 2. 3 8.7 1.0 0.0 lzcntw 2(%rsp), %cx96# CHECK-NEXT: 3 7.9 0.4 0.0 <total>97