From 31c5ff0f160eed9fa13093bc56f445fe5cf8c78b Mon Sep 17 00:00:00 2001
From: Ari Koivula <ari@koivu.la>
Date: Wed, 28 Sep 2016 23:05:21 +0300
Subject: [PATCH] Add cross-platform core number detection

Well, turns out pthread_num_processors_np isn't standard so we need to
do this crap. Threw in hyper threading detection as a bonus.
---
 src/cfg.c              |  4 +---
 src/encoder.c          | 20 ++++++++++++++++++++
 src/strategyselector.c | 25 +++++++++++++++++++++++++
 src/strategyselector.h |  5 +++++
 4 files changed, 51 insertions(+), 3 deletions(-)

diff --git a/src/cfg.c b/src/cfg.c
index 5a02fadb..82dc6135 100644
--- a/src/cfg.c
+++ b/src/cfg.c
@@ -23,7 +23,6 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
-#include <pthread.h>
 
 
 kvz_config *kvz_config_alloc(void)
@@ -99,8 +98,7 @@ int kvz_config_init(kvz_config *cfg)
   cfg->slice_addresses_in_ts = MALLOC(int32_t, 1);
   cfg->slice_addresses_in_ts[0] = 0;
   
-  // Set number of threads to equal number of processors available.
-  cfg->threads = pthread_num_processors_np();
+  cfg->threads = -1;
   cfg->cpuid = 1;
 
   // Defaults for what sizes of PUs are tried.
diff --git a/src/encoder.c b/src/encoder.c
index b686e158..0f5cb9fa 100644
--- a/src/encoder.c
+++ b/src/encoder.c
@@ -24,6 +24,7 @@
 #include <stdlib.h>
 
 #include "cfg.h"
+#include "strategyselector.h"
 
 
 static int encoder_control_init_gop_layer_weights(encoder_control_t * const);
@@ -98,6 +99,21 @@ static int select_owf_auto(const kvz_config *const cfg)
   }
 }
 
+
+static unsigned cfg_num_threads(void)
+{
+  unsigned cpus = kvz_g_hardware_flags.physical_cpu_count;
+  unsigned fake_cpus = kvz_g_hardware_flags.logical_cpu_count - cpus;
+
+  // Default to 4 if we don't know the number of CPUs.
+  if (cpus == 0) return 4;
+
+  // 1.5 times the number of physical cores seems to be a good compromise
+  // when hyperthreading is available on Haswell.
+  return cpus + fake_cpus / 2;
+}
+
+
 /**
  * \brief Allocate and initialize an encoder control structure.
  *
@@ -112,6 +128,10 @@ encoder_control_t* kvz_encoder_control_init(kvz_config *const cfg) {
     goto init_failed;
   }
 
+  if (cfg->threads == -1) {
+    cfg->threads = cfg_num_threads();
+  }
+
   if (cfg->gop_len > 0) {
     if (cfg->tmvp_enable) {
       cfg->tmvp_enable = false;
diff --git a/src/strategyselector.c b/src/strategyselector.c
index 60819ca9..e3a49e5f 100644
--- a/src/strategyselector.c
+++ b/src/strategyselector.c
@@ -24,6 +24,15 @@
 #include <stdlib.h>
 #include <string.h>
 
+#ifdef _WIN32
+#include <windows.h>
+#elif MACOS
+#include <sys/param.h>
+#include <sys/sysctl.h>
+#else
+#include <unistd.h>
+#endif
+
 hardware_flags_t kvz_g_hardware_flags;
 hardware_flags_t kvz_g_strategies_in_use;
 hardware_flags_t kvz_g_strategies_available;
@@ -410,6 +419,7 @@ static void set_hardware_flags(int32_t cpuid) {
       CPUID1_EDX_MMX = 1 << 23,
       CPUID1_EDX_SSE = 1 << 25,
       CPUID1_EDX_SSE2 = 1 << 26,
+      CPUID1_EDX_HYPER_THREADING = 1 << 28,
     };
     enum {
       CPUID1_ECX_SSE3 = 1 << 0,
@@ -430,6 +440,21 @@ static void set_hardware_flags(int32_t cpuid) {
 
     // Dig CPU features with cpuid
     get_cpuid(1, 0, &cpuid1);
+
+#ifdef _WIN32
+    SYSTEM_INFO systeminfo;
+    GetSystemInfo(&systeminfo);
+
+    kvz_g_hardware_flags.logical_cpu_count = systeminfo.dwNumberOfProcessors;
+#else
+    kvz_g_hardware_flags.logical_cpu_count = sysconf(_SC_NPROCESSORS_ONLN);
+#endif
+
+    kvz_g_hardware_flags.physical_cpu_count = kvz_g_hardware_flags.logical_cpu_count;
+    kvz_g_hardware_flags.intel_flags.hyper_threading = cpuid1.edx & CPUID1_EDX_HYPER_THREADING;
+    if (kvz_g_hardware_flags.intel_flags.hyper_threading) {
+      kvz_g_hardware_flags.physical_cpu_count /= 2;
+    }
     
     // EDX
     if (cpuid1.edx & CPUID1_EDX_MMX)   kvz_g_hardware_flags.intel_flags.mmx = 1;
diff --git a/src/strategyselector.h b/src/strategyselector.h
index 1eae4d5a..4c6819ea 100644
--- a/src/strategyselector.h
+++ b/src/strategyselector.h
@@ -63,6 +63,8 @@ typedef struct {
     int sse42;
     int avx;
     int avx2;
+
+    bool hyper_threading;
   } intel_flags;
   
   struct {
@@ -72,6 +74,9 @@ typedef struct {
   struct {
     int neon;
   } arm_flags;
+
+  int logical_cpu_count;
+  int physical_cpu_count;
 } hardware_flags_t;
 
 extern hardware_flags_t kvz_g_hardware_flags;