common : try to fix Android CI

[ggerganov]

ref #6414

https://github.com/ggerganov/llama.cpp/actions/runs/8757099433/job/24035025415#step:5:200

does not pass :/

[phymbert]

@ggerganov What about simply skipping __ANDROID__ definition ?

pthread_setaffinity_np is not supported on android ndk.

patch example

Index: common/common.cpp
IDEA additional info:
Subsystem: com.intellij.openapi.diff.impl.patch.CharsetEP
<+>UTF-8
===================================================================
diff --git a/common/common.cpp b/common/common.cpp
--- a/common/common.cpp	(revision 0e4802b2ecbaab04b4f829fde4a3096ca19c84b5)
+++ b/common/common.cpp	(date 1713342380677)
@@ -109,6 +109,7 @@
 }
 
 #if defined(__x86_64__) && defined(__linux__)
+#ifndef __ANDROID__
 #include <pthread.h>
 
 static void cpuid(unsigned leaf, unsigned subleaf,
@@ -156,6 +157,7 @@
     return result;
 }
 
+#endif // __ANDROID__
 #endif // __x86_64__ && __linux__
 
 /**
@@ -163,6 +165,7 @@
  */
 int get_math_cpu_count() {
 #if defined(__x86_64__) && defined(__linux__)
+#ifndef __ANDROID__
     int cpu_count = sysconf(_SC_NPROCESSORS_ONLN);
     if (cpu_count < 1) {
         return get_num_physical_cores();
@@ -177,7 +180,8 @@
             }
         }
     }
-#endif
+#endif // __ANDROID__
+#endif // __x86_64__ && __linux__
     return get_num_physical_cores();
 }
 
Index: examples/llama.android/README.md
IDEA additional info:
Subsystem: com.intellij.openapi.diff.impl.patch.CharsetEP
<+>UTF-8
===================================================================
diff --git a/examples/llama.android/README.md b/examples/llama.android/README.md
--- a/examples/llama.android/README.md	(revision 0e4802b2ecbaab04b4f829fde4a3096ca19c84b5)
+++ b/examples/llama.android/README.md	(date 1713605825665)
@@ -1,0 +1,3 @@
+# llama.cpp/examples/llama.android
+
+Demonstration of simple android app example.

Works for me locally.

BTW I will have a look to the android example. It does not work.

[ggerganov]

Thanks. I keep forgetting that the Android CI pulls from master, so maybe even the first attempt was correct:

diff --git a/common/common.cpp b/common/common.cpp
index b6143e41..3039e470 100644
--- a/common/common.cpp
+++ b/common/common.cpp
@@ -108,7 +108,7 @@ int32_t get_num_physical_cores() {
     return n_threads > 0 ? (n_threads <= 4 ? n_threads : n_threads / 2) : 4;
 }
 
-#if defined(__x86_64__) && defined(__linux__) && !defined(__ANDROID__)
+#if defined(__x86_64__) && defined(__gnu_linux__)
 #include <pthread.h>
 
 static void cpuid(unsigned leaf, unsigned subleaf,
@@ -162,7 +162,7 @@ static int count_math_cpus(int cpu_count) {
  * Returns number of CPUs on system that are useful for math.
  */
 int get_math_cpu_count() {
-#if defined(__x86_64__) && defined(__linux__) && !defined(__ANDROID__)
+#if defined(__x86_64__) && defined(__gnu_linux__)
     int cpu_count = sysconf(_SC_NPROCESSORS_ONLN);
     if (cpu_count < 1) {
         return get_num_physical_cores();

For now, will merge as it is to make the CI green and can improve in the future

[phymbert]

Noted, I will fix that also. At the moment it is impossible to load a model anyway. I will have fun with that :)

[phymbert]

Yes it is fixed, thanks:
https://github.com/ggerganov/llama.cpp/actions/runs/8764400637