bolt/test/X86/tail-duplication-cache.s

   1 # REQUIRES: system-linux
   2
   3 # RUN: llvm-mc -filetype=obj -triple x86_64-unknown-unknown \
   4 # RUN:   %s -o %t.o
   5 # RUN: link_fdata %s %t.o %t.fdata
   6 # RUN: link_fdata %s %t.o %t.fdata2 "FDATA2"
   7 # RUN: %clang %cflags %t.o -o %t.exe -Wl,-q
   8 # RUN: llvm-bolt %t.exe --data %t.fdata --reorder-blocks=none \
   9 # RUN:    --print-finalized --tail-duplication=cache -o %t.out | FileCheck %s
  10 # RUN: llvm-bolt %t.exe --data %t.fdata2 --reorder-blocks=none \
  11 # RUN:    --print-finalized --tail-duplication=cache -o %t.out2 \
  12 # RUN:    | FileCheck --check-prefix="CHECK2" %s
  13
  14 # A test where the tail is duplicated to eliminate an unconditional jump
  15 # FDATA: 1 main #.BB0_br# 1 main #.BB4# 0 100
  16 # FDATA: 1 main #.BB0_br# 1 main #.BB1# 0 100
  17 # FDATA: 1 main #.BB1_br# 1 main #.BB3# 0 50
  18 # FDATA: 1 main #.BB1_br# 1 main #.BB2# 0 50
  19 # FDATA: 1 main #.BB3_br# 1 main #.BB2# 0 50
  20 # CHECK: BOLT-INFO: tail duplication modified 1 ({{.*}}%) functions; duplicated 1 blocks (13 bytes) responsible for 50 dynamic executions ({{.*}}% of all block executions)
  21 # CHECK: BB Layout   : .LBB00, .Ltmp0, .Ltmp1, .Ltmp2, .Ltmp3, .Ltmp4, .Ltmp5, .Ltail-dup0, .Ltmp6
  22
  23 # A test where the tail is not duplicated due to the cache score
  24 # FDATA2: 1 main #.BB0_br# 1 main #.BB4# 0 100
  25 # FDATA2: 1 main #.BB0_br# 1 main #.BB1# 0 2
  26 # FDATA2: 1 main #.BB1_br# 1 main #.BB3# 0 1
  27 # FDATA2: 1 main #.BB1_br# 1 main #.BB2# 0 1
  28 # FDATA2: 1 main #.BB3_br# 1 main #.BB2# 0 1
  29 # CHECK2: BOLT-INFO: tail duplication modified 0 (0.00%) functions; duplicated 0 blocks (0 bytes) responsible for 0 dynamic executions (0.00% of all block executions)
  30 # CHECK2: BB Layout   : .LBB00, .Ltmp0, .Ltmp1, .Ltmp2, .Ltmp3, .Ltmp4, .Ltmp5, .Ltmp6
  31
  32     .text
  33     .globl main
  34     .type main, %function
  35     .size main, .Lend-main
  36 main:
  37 .BB0:
  38     xor %eax, %eax
  39         cmpl    %eax, %ebx
  40 .BB0_br:
  41     je      .BB4
  42 .BB1:
  43     inc %rax
  44 .BB1_br:
  45     je .BB3
  46 .BB2:
  47     inc %rax
  48     inc %rax
  49     inc %rax
  50     inc %rax
  51     retq
  52 .BB3:
  53     inc %rax
  54 .BB3_br:
  55     jmp .BB2
  56 .BB4:
  57     retq
  58 # For relocations against .text
  59     call exit
  60 .Lend: