Merge pull request #789 from pao/topic/zebras

RFC: Optionally set FZ/DAZ on SSE(2) processors
JuliaLang · May 3, 2012 · 887d701 · 887d701
2 parents 75529ab + 4158d61
commit 887d701
Show file tree

Hide file tree

Showing 2 changed files with 28 additions and 0 deletions.
diff --git a/src/julia.expmap b/src/julia.expmap
@@ -151,6 +151,7 @@
     jl_buf_mutex_lock;
     jl_buf_mutex_unlock;
     jl_start_io_thread;
+    jl_zero_denormals;
     jl_save_system_image;
     jl_restore_system_image;
     jl_compress_ast;

diff --git a/src/sys.c b/src/sys.c
@@ -16,6 +16,10 @@
 #include <pthread.h>
 #include "julia.h"
 
+#ifdef __SSE__
+#include <xmmintrin.h>
+#endif
+
 // --- io and select ---
 
 void jl__not__used__(void)
@@ -378,3 +382,26 @@ DLLEXPORT void jl_start_io_thread(void)
     pthread_cond_init(&wake_cond, NULL);
     pthread_create(&io_thread, NULL, run_io_thr, NULL);
 }
+
+DLLEXPORT uint8_t jl_zero_denormals(uint8_t isZero)
+{
+#ifdef __SSE2__
+    // SSE2 supports both FZ and DAZ
+    uint32_t flags = 0x8040;
+#elif __SSE__
+    // SSE supports only the FZ flag
+    uint32_t flags = 0x8000;
+#endif
+
+#ifdef __SSE__
+    if (isZero) {
+	_mm_setcsr(_mm_getcsr() | flags);
+    }
+    else {
+	_mm_setcsr(_mm_getcsr() & ~flags);
+    }
+    return 1;
+#else
+    return 0;
+#endif
+}