From 6bff37abacf95b60a1d8d4c83c9e59e42ae1a7b9 Mon Sep 17 00:00:00 2001
From: Josh Stone <jistone@redhat.com>
Date: Wed, 9 Jun 2010 21:44:42 -0400
Subject: [PATCH] Let probe_bench work on archs besides x86_64

The cycles code is too x86_64-centric, but for now at least it's been
neutered to let other archs report basic timing numbers.

* scripts/probe_bench/bench.stp: Use get_cycles() instead of rdtsc().
* scripts/probe_bench/bench.c (rdtsc): Return 0 unless on x86_64.  This
  needs more work and inline assembly to fit more architectures.
---
 scripts/probe_perf/bench.c   | 11 ++++++++--
 scripts/probe_perf/bench.stp | 42 ++++++++++++++++--------------------
 2 files changed, 27 insertions(+), 26 deletions(-)

diff --git a/scripts/probe_perf/bench.c b/scripts/probe_perf/bench.c
index ec391c875..1cb2ea8c7 100644
--- a/scripts/probe_perf/bench.c
+++ b/scripts/probe_perf/bench.c
@@ -139,8 +139,12 @@ float zr, zi;
 long rdtsc(void) __attribute__ ((__noinline__));
 long rdtsc(void) 
 {
+  // XXX this only works on x86_64
+  // -- not i386 because the long result is only 32-bit
+  // -- not other archs because they need other asm
+#if defined __x86_64__
   int res[2];
-   
+
   __asm__ __volatile__  ("xorl %%eax,%%eax \n push %%rbx \n cpuid \n"
 			 ::: "%rax", "%rcx", "%rdx");
   // read TSC, store edx:eax in res
@@ -148,8 +152,11 @@ long rdtsc(void)
 			 : "=a" (res[0]), "=d" (res[1]) );
   __asm__ __volatile__  ("xorl %%eax,%%eax \n cpuid \n pop %%rbx \n"
 			 ::: "%rax", "%rcx", "%rdx");
-   
+
   return *(long*)res;
+#else
+  return 0;
+#endif
 }
 
 Initrand ()
diff --git a/scripts/probe_perf/bench.stp b/scripts/probe_perf/bench.stp
index 7b9628aae..df4b6a668 100644
--- a/scripts/probe_perf/bench.stp
+++ b/scripts/probe_perf/bench.stp
@@ -1,86 +1,80 @@
-function rdtsc () %{
-   int res[2];
-   __asm__ __volatile__  ("rdtsc\n" : "=a" (res[0]), "=d" (res[1]) );
-   THIS->__retvalue = *(long*)res;
-%}
-
 global process_hits
 %( @2 == "NO_STAP_SDT" %?
 probe process(@1).function("*").label("BENCH_PERMUTE")
-{   process_hits <<< (rdtsc() - $cycles_start) }
+{   process_hits <<< (get_cycles() - $cycles_start) }
 %:
 probe process(@1).mark("permute")
-{   process_hits <<< (rdtsc() - $arg1) }
+{   process_hits <<< (get_cycles() - $arg1) }
 %)
 
 global tower_hits
 %( @2 == "NO_STAP_SDT" %?
 probe process(@1).function("*").label("BENCH_TOWER")
-{   tower_hits <<< (rdtsc() - $cycles_start) }
+{   tower_hits <<< (get_cycles() - $cycles_start) }
 %:
 probe process(@1).mark("tower")
-{   tower_hits <<< (rdtsc() - $arg1) }
+{   tower_hits <<< (get_cycles() - $arg1) }
 %)
 
 global try_hits
 %( @2 == "NO_STAP_SDT" %?
 probe process(@1).function("*").label("BENCH_TRY")
-{   try_hits <<< (rdtsc() - $cycles_start) }
+{   try_hits <<< (get_cycles() - $cycles_start) }
 %:
 probe process(@1).mark("try")
-{   try_hits <<< (rdtsc() - $arg1) }
+{   try_hits <<< (get_cycles() - $arg1) }
 %)
 
 global innerproduct_hits
 %( @2 == "NO_STAP_SDT" %?
 probe process(@1).function("*").label("BENCH_INNERPRODUCT")
-{   innerproduct_hits <<< (rdtsc() - $cycles_start) }
+{   innerproduct_hits <<< (get_cycles() - $cycles_start) }
 %:
 probe process(@1).mark("innerproduct")
-{   innerproduct_hits <<< (rdtsc() - $arg1) }
+{   innerproduct_hits <<< (get_cycles() - $arg1) }
 %)
 
 global trial_hits
 %( @2 == "NO_STAP_SDT" %?
 probe process(@1).function("*").label("BENCH_TRIAL")
-{   trial_hits <<< (rdtsc() - $cycles_start) }
+{   trial_hits <<< (get_cycles() - $cycles_start) }
 %:
 probe process(@1).mark("trial")
-{   trial_hits <<< (rdtsc() - $arg1) }
+{   trial_hits <<< (get_cycles() - $arg1) }
 %)
 
 global quicksort_hits
 %( @2 == "NO_STAP_SDT" %?
 probe process(@1).function("*").label("BENCH_QUICKSORT")
-{   quicksort_hits <<< (rdtsc() - $cycles_start) }
+{   quicksort_hits <<< (get_cycles() - $cycles_start) }
 %:
 probe process(@1).mark("quicksort")
-{   quicksort_hits <<< (rdtsc() - $arg1) }
+{   quicksort_hits <<< (get_cycles() - $arg1) }
 %)
 
 global bubble_hits
 %( @2 == "NO_STAP_SDT" %?
 probe process(@1).function("*").label("BENCH_BUBBLE")
-{   bubble_hits <<< (rdtsc() - $cycles_start) }
+{   bubble_hits <<< (get_cycles() - $cycles_start) }
 %:
 probe process(@1).mark("bubble")
-{   bubble_hits <<< (rdtsc() - $arg1) }
+{   bubble_hits <<< (get_cycles() - $arg1) }
 %)
 
 global insert_hits
 %( @2 == "NO_STAP_SDT" %?
 probe process(@1).function("*").label("BENCH_INSERT")
-{   insert_hits <<< (rdtsc() - $cycles_start) }
+{   insert_hits <<< (get_cycles() - $cycles_start) }
 %:
 probe process(@1).mark("insert")
-{   insert_hits <<< (rdtsc() - $arg1) }
+{   insert_hits <<< (get_cycles() - $arg1) }
 %)
 
 global fft_hits
 %( @2 == "NO_STAP_SDT" %?
 probe process(@1).function("*").label("BENCH_FFT")
-{   fft_hits <<< (rdtsc() - $cycles_start) }
+{   fft_hits <<< (get_cycles() - $cycles_start) }
 %:
 probe process(@1).mark("fft")
-{   fft_hits <<< (rdtsc() - $arg1) }
+{   fft_hits <<< (get_cycles() - $arg1) }
 %)
-- 
2.43.5