116 lines · plain
1# NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py2# RUN: llvm-mca -mtriple=amdgcn -mcpu=gfx1010 --timeline --iterations=1 --timeline-max-cycles=0 < %s | FileCheck %s -check-prefixes=CHECK,GFX103# RUN: llvm-mca -mtriple=amdgcn -mcpu=gfx1100 --timeline --iterations=1 --timeline-max-cycles=0 < %s | FileCheck %s -check-prefixes=CHECK,GFX114 5v_log_f32 v0, v06v_rcp_f32 v0, v07v_rsq_f32 v1, v18v_sqrt_f32 v2, v09v_rcp_f64 v[0:1], v[0:1]10v_rsq_f64 v[1:2], v[1:2]11v_sqrt_f64 v[2:3], v[0:1]12 13# CHECK: Iterations: 114# CHECK-NEXT: Instructions: 715 16# GFX10-NEXT: Total Cycles: 9417# GFX11-NEXT: Total Cycles: 14218 19# CHECK-NEXT: Total uOps: 720 21# CHECK: Dispatch Width: 122 23# GFX10-NEXT: uOps Per Cycle: 0.0724# GFX10-NEXT: IPC: 0.0725 26# GFX11-NEXT: uOps Per Cycle: 0.0527# GFX11-NEXT: IPC: 0.0528 29# CHECK-NEXT: Block RThroughput: 7.030 31# CHECK: Instruction Info:32# CHECK-NEXT: [1]: #uOps33# CHECK-NEXT: [2]: Latency34# CHECK-NEXT: [3]: RThroughput35# CHECK-NEXT: [4]: MayLoad36# CHECK-NEXT: [5]: MayStore37# CHECK-NEXT: [6]: HasSideEffects (U)38 39# CHECK: [1] [2] [3] [4] [5] [6] Instructions:40# CHECK-NEXT: 1 10 1.00 U v_log_f32_e32 v0, v041# CHECK-NEXT: 1 10 1.00 U v_rcp_f32_e32 v0, v042# CHECK-NEXT: 1 10 1.00 U v_rsq_f32_e32 v1, v143# CHECK-NEXT: 1 10 1.00 U v_sqrt_f32_e32 v2, v044 45# GFX10-NEXT: 1 24 1.00 U v_rcp_f64_e32 v[0:1], v[0:1]46# GFX10-NEXT: 1 24 1.00 U v_rsq_f64_e32 v[1:2], v[1:2]47# GFX10-NEXT: 1 24 1.00 U v_sqrt_f64_e32 v[2:3], v[0:1]48 49# GFX11-NEXT: 1 40 1.00 U v_rcp_f64_e32 v[0:1], v[0:1]50# GFX11-NEXT: 1 40 1.00 U v_rsq_f64_e32 v[1:2], v[1:2]51# GFX11-NEXT: 1 40 1.00 U v_sqrt_f64_e32 v[2:3], v[0:1]52 53# CHECK: Resources:54# CHECK-NEXT: [0] - HWBranch55# CHECK-NEXT: [1] - HWExport56# CHECK-NEXT: [2] - HWLGKM57# CHECK-NEXT: [3] - HWRC58# CHECK-NEXT: [4] - HWSALU59# CHECK-NEXT: [5] - HWTransVALU60# CHECK-NEXT: [6] - HWVALU61# CHECK-NEXT: [7] - HWVMEM62 63# CHECK: Resource pressure per iteration:64# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7]65# CHECK-NEXT: - - - 7.00 - 7.00 3.00 -66 67# CHECK: Resource pressure by instruction:68# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] Instructions:69# CHECK-NEXT: - - - 1.00 - 1.00 - - v_log_f32_e32 v0, v070# CHECK-NEXT: - - - 1.00 - 1.00 - - v_rcp_f32_e32 v0, v071# CHECK-NEXT: - - - 1.00 - 1.00 - - v_rsq_f32_e32 v1, v172# CHECK-NEXT: - - - 1.00 - 1.00 - - v_sqrt_f32_e32 v2, v073# CHECK-NEXT: - - - 1.00 - 1.00 1.00 - v_rcp_f64_e32 v[0:1], v[0:1]74# CHECK-NEXT: - - - 1.00 - 1.00 1.00 - v_rsq_f64_e32 v[1:2], v[1:2]75# CHECK-NEXT: - - - 1.00 - 1.00 1.00 - v_sqrt_f64_e32 v[2:3], v[0:1]76 77# CHECK: Timeline view:78 79# GFX10-NEXT: 0123456789 0123456789 0123456789 0123456789 012380# GFX10-NEXT: Index 0123456789 0123456789 0123456789 0123456789 012345678981 82# GFX11-NEXT: 0123456789 0123456789 0123456789 0123456789 0123456789 0123456789 012345678983# GFX11-NEXT: Index 0123456789 0123456789 0123456789 0123456789 0123456789 0123456789 0123456789 0184 85# GFX10: [0,0] DeeeeeeeeeE . . . . . . . . . . . . . . . . . v_log_f32_e32 v0, v086# GFX10-NEXT: [0,1] . . DeeeeeeeeeE . . . . . . . . . . . . . . . v_rcp_f32_e32 v0, v087# GFX10-NEXT: [0,2] . . .DeeeeeeeeeE . . . . . . . . . . . . . . . v_rsq_f32_e32 v1, v188# GFX10-NEXT: [0,3] . . . . DeeeeeeeeeE . . . . . . . . . . . . . v_sqrt_f32_e32 v2, v089# GFX10-NEXT: [0,4] . . . . .DeeeeeeeeeeeeeeeeeeeeeeeE . . . . . . . . . . v_rcp_f64_e32 v[0:1], v[0:1]90# GFX10-NEXT: [0,5] . . . . . . . . . DeeeeeeeeeeeeeeeeeeeeeeeE. . . . . . v_rsq_f64_e32 v[1:2], v[1:2]91# GFX10-NEXT: [0,6] . . . . . . . . . . . . . . DeeeeeeeeeeeeeeeeeeeeeeeE v_sqrt_f64_e32 v[2:3], v[0:1]92 93# GFX11: [0,0] DeeeeeeeeeE . . . . . . . . . . . . . . . . . . . . . . . . . .. v_log_f32_e32 v0, v094# GFX11-NEXT: [0,1] . . DeeeeeeeeeE . . . . . . . . . . . . . . . . . . . . . . . .. v_rcp_f32_e32 v0, v095# GFX11-NEXT: [0,2] . . .DeeeeeeeeeE . . . . . . . . . . . . . . . . . . . . . . . .. v_rsq_f32_e32 v1, v196# GFX11-NEXT: [0,3] . . . . DeeeeeeeeeE . . . . . . . . . . . . . . . . . . . . . .. v_sqrt_f32_e32 v2, v097# GFX11-NEXT: [0,4] . . . . .DeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeE . . . . . . . . . . . . . . . .. v_rcp_f64_e32 v[0:1], v[0:1]98# GFX11-NEXT: [0,5] . . . . . . . . . . . . .DeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeE . . . . . . . .. v_rsq_f64_e32 v[1:2], v[1:2]99# GFX11-NEXT: [0,6] . . . . . . . . . . . . . . . . . . . . .DeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeE v_sqrt_f64_e32 v[2:3], v[0:1]100 101# CHECK: Average Wait times (based on the timeline view):102# CHECK-NEXT: [0]: Executions103# CHECK-NEXT: [1]: Average time spent waiting in a scheduler's queue104# CHECK-NEXT: [2]: Average time spent waiting in a scheduler's queue while ready105# CHECK-NEXT: [3]: Average time elapsed from WB until retire stage106 107# CHECK: [0] [1] [2] [3]108# CHECK-NEXT: 0. 1 0.0 0.0 0.0 v_log_f32_e32 v0, v0109# CHECK-NEXT: 1. 1 0.0 0.0 0.0 v_rcp_f32_e32 v0, v0110# CHECK-NEXT: 2. 1 0.0 0.0 0.0 v_rsq_f32_e32 v1, v1111# CHECK-NEXT: 3. 1 0.0 0.0 0.0 v_sqrt_f32_e32 v2, v0112# CHECK-NEXT: 4. 1 0.0 0.0 0.0 v_rcp_f64_e32 v[0:1], v[0:1]113# CHECK-NEXT: 5. 1 0.0 0.0 0.0 v_rsq_f64_e32 v[1:2], v[1:2]114# CHECK-NEXT: 6. 1 0.0 0.0 0.0 v_sqrt_f64_e32 v[2:3], v[0:1]115# CHECK-NEXT: 1 0.0 0.0 0.0 <total>116