[BOLT] Instrumentation: Fix tests

[Elvina]

Summary

Tests for instrumentation

[rafauler]

bolt/test/X86/asm-dump.c

10–11

nit: revert this change as it breaks this test, or add RUN: on line 11

[rafauler]

We need a test that checks indirect calls are correctly handled, like the one we have in D153771

[treapster]

In D151920#4450871, @rafauler wrote:

We need a test that checks indirect calls are correctly handled, like the one we have in D153771

Or we can make that test work for all platforms.

[rafauler]

In D151920#4451480, @treapster wrote:

In D151920#4450871, @rafauler wrote:

We need a test that checks indirect calls are correctly handled, like the one we have in D153771

Or we can make that test work for all platforms.

I agree

[Elvina]

added more tests, moved some from X86 to the common level

[Elvina]

Remove comment in meta-merge-fdata.test

[rafauler]

LGTM. Thanks!

[Amir]

Hi Elvina, this stack broke aarch64 buildbot: https://lab.llvm.org/buildbot/#/builders/221/builds/17715. You should've received a notification. Please prioritize fixing it or revert if it can't be fixed quickly.

[Elvina]

In D151920#4622639, @Amir wrote:

Hi Elvina, this stack broke aarch64 buildbot: https://lab.llvm.org/buildbot/#/builders/221/builds/17715. You should've received a notification. Please prioritize fixing it or revert if it can't be fixed quickly.

Hi! I'm trying to disable these tests in case they're building on aarch64, but I'm not sure this is the right way. Could you please take a look? https://reviews.llvm.org/D159094

[Amir]

In D151920#4626533, @Elvina wrote:

In D151920#4622639, @Amir wrote:

Hi Elvina, this stack broke aarch64 buildbot: https://lab.llvm.org/buildbot/#/builders/221/builds/17715. You should've received a notification. Please prioritize fixing it or revert if it can't be fixed quickly.

Hi! I'm trying to disable these tests in case they're building on aarch64, but I'm not sure this is the right way. Could you please take a look? https://reviews.llvm.org/D159094

Sorry, I meant specifically these tests (the buildbot is aarch64):

• instrumentation-ind-call.c: instrumented binary crashes with segfault
• basic-instrumentation.test: illegal instruction
• meta-merge-fdata.test: segfault

If needed I can provide access to the buildbot host, but it's a pretty standard Ubuntu 22.04 installation, so I'd suggest reproducing the issues locally first.

instrumentation-eh_frame_hdr.cpp and patch-entries.c will be addressed separately.

bolt/test/X86/asm-dump.c

 /**
  * Test for asm-dump functionality.
  *
- * REQUIRES: system-linux,bolt-runtime
+ * REQUIRES: x86_64-linux,bolt-runtime
  *
  * Compile the source
  * RUN: %clang -fPIC %s -o %t.exe -Wl,-q
  *
  * Profile collection: instrument the binary
- * RUN: llvm-bolt %t.exe --instrument --instrumentation-file=%t.fdata -o %t.instr
+ * RUN: llvm-bolt %t.exe --instrument --instrumentation-file=%t.fdata -o
+ * %t.instr
  *
  * Profile collection: run instrumented binary (and capture output)
  * RUN: %t.instr > %t.result
  *
  * Run BOLT with asm-dump
  * RUN: llvm-bolt %t.exe -p %t.fdata --funcs=main --asm-dump=%t -o /dev/null \
  * RUN:   | FileCheck %s --check-prefix=CHECK-BOLT
  *

bolt/test/X86/bolt-address-translation-internal-call.test

 # This checks for an issue with internal calls and BAT (BOLT address
 # translation). BAT needs to map every output block back to an input
 # block, but passes that introduce new blocks (such as validate
 # internal calls) might create new blocks without a mapping to an
 # input block.
 
-# REQUIRES: system-linux,bolt-runtime
+# REQUIRES: x86_64-linux,bolt-runtime
 
 # RUN: llvm-mc -filetype=obj -triple x86_64-unknown-unknown %s -o %t.o
 # Delete our BB symbols so BOLT doesn't mark them as entry points
 # RUN: llvm-strip --strip-unneeded %t.o
 # RUN: %clang %t.o -o %t.exe -Wl,-q
 
 # RUN: llvm-bolt --enable-bat %t.exe --relocs -o %t.out | FileCheck %s
 # CHECK: BOLT-INFO: Wrote {{.*}} BAT maps

bolt/test/X86/internal-call-instrument.s

 # This reproduces a bug with instrumentation crashes on internal call
 
-# REQUIRES: system-linux,bolt-runtime
+# REQUIRES: x86_64-linux,bolt-runtime
 
 # RUN: llvm-mc -filetype=obj -triple x86_64-unknown-unknown %s -o %t.o
 # Delete our BB symbols so BOLT doesn't mark them as entry points
 # RUN: llvm-strip --strip-unneeded %t.o
 # RUN: %clang %t.o -o %t.exe -Wl,-q
 
 # RUN: llvm-bolt --instrument %t.exe --relocs -o %t.out
 

bolt/test/runtime/AArch64/Inputs/basic-instrumentation.s

+  .globl main
+  .type main, %function
+main:
+  sub   sp, sp, #16
+  mov   w0, wzr
+  str   wzr, [sp, #12]
+  add   sp, sp, #16
+  ret
+.size main, .-main

bolt/test/runtime/AArch64/basic-instrumentation.test

+# Try to instrument a very fast test. Input bin will not execute any code during
+# runtime besides returning zero in main, so it is a good trivial case.
+REQUIRES: system-linux,bolt-runtime
+
+RUN: %clang %p/Inputs/basic-instrumentation.s -Wl,-q -o %t.exe
+RUN: llvm-bolt %t.exe -o %t --instrument \
+RUN:   --instrumentation-file=%t \
+RUN:   --instrumentation-file-append-pid
+
+# Execute program to collect profile
+RUN: rm %t.*.fdata || echo Nothing to remove
+RUN: %t
+
+# Profile should be written to %t.PID.fdata, check it
+RUN: mv %t.*.fdata %t.fdata
+RUN: cat %t.fdata | FileCheck -check-prefix=CHECK %s
+
+# Check BOLT works with this profile
+RUN: llvm-bolt %t.exe --data %t.fdata -o %t.2 --reorder-blocks=cache
+
+# The instrumented profile should at least say main was called once
+CHECK: main 0 0 1{{$}}