feat: Allocate thread resource based on the physical processor

The original design has contradiction.
It uses almost all the logical processors for a single PoW,
but allows dcurl to handle 2 PoW at the same time based on the
hyper-threading.

The new function is added in cpu-utils.h to detect the thread number per
physical processor, which also shows whether the hyper-threading is
enabled or not.
Once we know the status of hyper-threading, we can determine the
threadpool size and the thread number for a single PoW.

Close #191.

Doxyfile

@@ -17,6 +17,7 @@ INPUT                  = . \
                          src
 FILE_PATTERNS          = dcurl.c \
                          dcurl.h \
+                         cpu-utils.h \
                          *.md
 USE_MDFILE_AS_MAINPAGE = README.md
 #---------------------------------------------------------------------------
@@ -27,6 +28,10 @@ INLINE_SOURCES         = YES
 REFERENCED_BY_RELATION = YES
 REFERENCES_RELATION    = YES
 #---------------------------------------------------------------------------
+# Build related configuration options
+#---------------------------------------------------------------------------
+EXTRACT_STATIC         = YES
+#---------------------------------------------------------------------------
 # Configuration options related to the LaTeX output
 #---------------------------------------------------------------------------
 GENERATE_LATEX         = NO

src/cpu-utils.h

@@ -15,6 +15,11 @@
 #include <sys/sysinfo.h>
 #endif
 
+/**
+ * @file cpu-utils.h
+ * @brief Utility functions for acquiring CPU information.
+ */
+
 /* On Mac OS X, define our own get_nprocs_conf() */
 #if defined(__APPLE__) || defined(__FreeBSD__)
 #include <sys/sysctl.h>
@@ -28,7 +33,16 @@ static unsigned int get_nprocs_conf()
 }
 #endif
 
-static inline int get_avail_nprocs()
+/**
+ * @brief Get the available number of logical processor.
+ *
+ * Detect the logical processor number with `get_nproc_conf()`
+ * or get the number with the environment variable **DCURL_NUM_CPU**.
+ * @return The available number of logical processor.
+ * If the environment variable **DCURL_NUM_CPU** is not set, return the
+ * **maximum logical processor number - 1**.
+ */
+static inline int get_avail_logic_nprocs()
 {
     size_t nproc = get_nprocs_conf() - 1;
 
@@ -55,3 +69,62 @@ static inline int get_avail_nprocs()
         nproc = 1;
     return nproc;
 }
+
+/**
+ * @brief Get the thread number per physical processor.
+ *
+ * - GNU/Linux: Acquire the thread number by parsing the CPU information.
+ * - macOS: Acquire the thread number by doing the calculation of
+ * (logical processor number / physical processor number).
+ * @return The thread number per physical processor.
+ * @retval 1 Hyperthreading disabled.
+ * @retval 2 Hyperthreading enabled.
+ * @retval -1 Unexpected error.
+ */
+static inline int get_nthds_per_physic_proc()
+{
+    FILE *fd;
+    int nthread;
+#if defined(__linux__)
+    char nthd[4];
+
+    fd = popen("LC_ALL=C lscpu | grep 'Thread(s) per core' | awk '{printf $4}'",
+               "r");
+    if (fd == NULL)
+        return -1;
+    if (fgets(nthd, sizeof(nthd), fd) == NULL)
+        return -1;
+    nthread = (int) strtol(nthd, NULL, 10);
+    if (errno == ERANGE || nthread == 0) {
+        return -1;
+    }
+#elif defined(__APPLE__)
+    char p_proc[4], l_proc[4];
+    int physic_proc, logic_proc;
+
+    fd = popen("sysctl hw.physicalcpu | awk '{printf $2}'", "r");
+    if (fd == NULL)
+        return -1;
+    if (fgets(p_proc, sizeof(p_proc), fd) == NULL)
+        return -1;
+    fd = popen("sysctl hw.logicalcpu | awk '{printf $2}'", "r");
+    if (fd == NULL)
+        return -1;
+    if (fgets(l_proc, sizeof(l_proc), fd) == NULL)
+        return -1;
+    physic_proc = (int) strtol(p_proc, NULL, 10);
+    if (errno == ERANGE || physic_proc == 0) {
+        return -1;
+    }
+    logic_proc = (int) strtol(l_proc, NULL, 10);
+    if (errno == ERANGE || logic_proc == 0) {
+        return -1;
+    }
+
+    nthread = logic_proc / physic_proc;
+#endif
+
+    if (pclose(fd) == -1)
+        return -1;
+    return nthread;
+}

src/pow_avx.c

@@ -573,7 +573,8 @@ static bool PowAVX(void *pow_ctx)
 
 static bool PoWAVX_Context_Initialize(ImplContext *impl_ctx)
 {
-    int nproc = get_avail_nprocs();
+    impl_ctx->num_max_thread = get_nthds_per_physic_proc();
+    int nproc = get_avail_logic_nprocs() / impl_ctx->num_max_thread;
     if (impl_ctx->num_max_thread <= 0 || nproc <= 0)
         return false;
 
@@ -609,7 +610,7 @@ static bool PoWAVX_Context_Initialize(ImplContext *impl_ctx)
         impl_ctx->bitmap = impl_ctx->bitmap << 1 | 0x1;
         uv_loop_init(&ctx[i].loop);
     }
-    uv_set_threadpool_size(nproc);
+    uv_set_threadpool_size(impl_ctx->num_max_thread * nproc);
     impl_ctx->context = ctx;
     uv_mutex_init(&impl_ctx->lock);
     return true;
@@ -693,7 +694,7 @@ ImplContext PoWAVX_Context = {
     .context = NULL,
     .description = "CPU (Intel AVX)",
     .bitmap = 0,
-    .num_max_thread = 2,
+    .num_max_thread = 0,
     .num_working_thread = 0,
     .initialize = PoWAVX_Context_Initialize,
     .destroy = PoWAVX_Context_Destroy,

src/pow_c.c

@@ -337,7 +337,8 @@ bool PowC(void *pow_ctx)
 
 static bool PoWC_Context_Initialize(ImplContext *impl_ctx)
 {
-    int nproc = get_avail_nprocs();
+    impl_ctx->num_max_thread = get_nthds_per_physic_proc();
+    int nproc = get_avail_logic_nprocs() / impl_ctx->num_max_thread;
     if (impl_ctx->num_max_thread <= 0 || nproc <= 0)
         return false;
 
@@ -373,7 +374,7 @@ static bool PoWC_Context_Initialize(ImplContext *impl_ctx)
         impl_ctx->bitmap = impl_ctx->bitmap << 1 | 0x1;
         uv_loop_init(&ctx[i].loop);
     }
-    uv_set_threadpool_size(nproc);
+    uv_set_threadpool_size(impl_ctx->num_max_thread * nproc);
     impl_ctx->context = ctx;
     uv_mutex_init(&impl_ctx->lock);
     return true;
@@ -456,7 +457,7 @@ ImplContext PoWC_Context = {
     .context = NULL,
     .description = "CPU (Pure C)",
     .bitmap = 0,
-    .num_max_thread = 2,
+    .num_max_thread = 0,
     .num_working_thread = 0,
     .initialize = PoWC_Context_Initialize,
     .destroy = PoWC_Context_Destroy,

src/pow_sse.c

@@ -355,7 +355,8 @@ static bool PowSSE(void *pow_ctx)
 
 static bool PoWSSE_Context_Initialize(ImplContext *impl_ctx)
 {
-    int nproc = get_avail_nprocs();
+    impl_ctx->num_max_thread = get_nthds_per_physic_proc();
+    int nproc = get_avail_logic_nprocs() / impl_ctx->num_max_thread;
     if (impl_ctx->num_max_thread <= 0 || nproc <= 0)
         return false;
 
@@ -391,7 +392,7 @@ static bool PoWSSE_Context_Initialize(ImplContext *impl_ctx)
         impl_ctx->bitmap = impl_ctx->bitmap << 1 | 0x1;
         uv_loop_init(&ctx[i].loop);
     }
-    uv_set_threadpool_size(nproc);
+    uv_set_threadpool_size(impl_ctx->num_max_thread * nproc);
     impl_ctx->context = ctx;
     uv_mutex_init(&impl_ctx->lock);
     return true;
@@ -475,7 +476,7 @@ ImplContext PoWSSE_Context = {
     .context = NULL,
     .description = "CPU (Intel SSE)",
     .bitmap = 0,
-    .num_max_thread = 2,
+    .num_max_thread = 0,
     .num_working_thread = 0,
     .initialize = PoWSSE_Context_Initialize,
     .destroy = PoWSSE_Context_Destroy,