From 696a24e6c7a8d9af5b818a4cd020836a14236bef Mon Sep 17 00:00:00 2001
From: Juan ESCOBAR <juan.escobar@aero.obs-mip.fr>
Date: Tue, 13 Sep 2022 17:30:16 +0200
Subject: [PATCH] Juan 13/09/2022: add new ARCH_SRC/cray_gu/gu.c , for
Automatic Gradual Underflow on CPU (like GPU)
---
src/ARCH_SRC/cray_gu/gu.c | 45 +++++++++++++++++++++++++++++++++++++++
1 file changed, 45 insertions(+)
create mode 100644 src/ARCH_SRC/cray_gu/gu.c
diff --git a/src/ARCH_SRC/cray_gu/gu.c b/src/ARCH_SRC/cray_gu/gu.c
new file mode 100644
index 000000000..24581c7e0
--- /dev/null
+++ b/src/ARCH_SRC/cray_gu/gu.c
@@ -0,0 +1,45 @@
+/*
+
+* Copyright -2022 Hewlett Packard Enterprise Development LP
+
+*/
+
+void __cce_floatingpt_gu( void ); // Gradual Underflow.
+
+
+
+static void (*p)(void) __attribute__((section(".init_array"))) = __cce_floatingpt_gu;
+
+
+
+#define DENORMS_ARE_ZEROS 6
+
+#define FLUSH_TO_ZERO_MASK 15
+
+
+
+void __cce_floatingpt_gu( void )
+
+{
+
+ unsigned long gu_src = (1 << DENORMS_ARE_ZEROS);
+
+ unsigned long gu_dst = (1 << FLUSH_TO_ZERO_MASK);
+
+
+
+ do {
+
+ unsigned long __mxcsr;
+
+ __asm__ volatile("stmxcsr %0" : : "m" (__mxcsr));
+
+ __mxcsr &= ~(gu_src | gu_dst);
+
+ __asm__ volatile("ldmxcsr %0" : : "m" (__mxcsr));
+
+ }
+
+ while (0);
+
+}
--
GitLab