From 7f9fec25811c98cd2892f639661b0822988c7de7 Mon Sep 17 00:00:00 2001
From: Max Horn <max@quendi.de>
Date: Sat, 19 Feb 2022 12:59:19 +0100
Subject: [PATCH] Fix get/set_fpcr_aarch64

On Aarch64, the `fpcr` register is 64bit wide, although the top 32bit
are currently unused and reserved for future usage. Nevertheless, we
should safe and restore the full 64 bit, not just 32 bit. This also
silences a compiler warning about this. Reference:
<https://developer.arm.com/documentation/ddi0595/2021-06/AArch64-Registers/FPCR--Floating-point-Control-Register>
---
 src/processor_arm.cpp | 18 +++++++++---------
 1 file changed, 9 insertions(+), 9 deletions(-)

diff --git a/src/processor_arm.cpp b/src/processor_arm.cpp
index f5cc2a42a4870..ea8dddf629d62 100644
--- a/src/processor_arm.cpp
+++ b/src/processor_arm.cpp
@@ -1843,20 +1843,20 @@ extern "C" int jl_test_cpu_feature(jl_cpu_feature_t feature)
 
 #ifdef _CPU_AARCH64_
 // FPCR FZ, bit [24]
-static constexpr uint32_t fpcr_fz_mask = 1 << 24;
+static constexpr uint64_t fpcr_fz_mask = 1 << 24;
 // FPCR FZ16, bit [19]
-static constexpr uint32_t fpcr_fz16_mask = 1 << 19;
+static constexpr uint64_t fpcr_fz16_mask = 1 << 19;
 // FPCR DN, bit [25]
-static constexpr uint32_t fpcr_dn_mask = 1 << 25;
+static constexpr uint64_t fpcr_dn_mask = 1 << 25;
 
-static inline uint32_t get_fpcr_aarch64(void)
+static inline uint64_t get_fpcr_aarch64(void)
 {
-    uint32_t fpcr;
+    uint64_t fpcr;
     asm volatile("mrs %0, fpcr" : "=r"(fpcr));
     return fpcr;
 }
 
-static inline void set_fpcr_aarch64(uint32_t fpcr)
+static inline void set_fpcr_aarch64(uint64_t fpcr)
 {
     asm volatile("msr fpcr, %0" :: "r"(fpcr));
 }
@@ -1868,8 +1868,8 @@ extern "C" JL_DLLEXPORT int32_t jl_get_zero_subnormals(void)
 
 extern "C" JL_DLLEXPORT int32_t jl_set_zero_subnormals(int8_t isZero)
 {
-    uint32_t fpcr = get_fpcr_aarch64();
-    static uint32_t mask = fpcr_fz_mask | (jl_test_cpu_feature(JL_AArch64_fullfp16) ? fpcr_fz16_mask : 0);
+    uint64_t fpcr = get_fpcr_aarch64();
+    static uint64_t mask = fpcr_fz_mask | (jl_test_cpu_feature(JL_AArch64_fullfp16) ? fpcr_fz16_mask : 0);
     fpcr = isZero ? (fpcr | mask) : (fpcr & ~mask);
     set_fpcr_aarch64(fpcr);
     return 0;
@@ -1882,7 +1882,7 @@ extern "C" JL_DLLEXPORT int32_t jl_get_default_nans(void)
 
 extern "C" JL_DLLEXPORT int32_t jl_set_default_nans(int8_t isDefault)
 {
-    uint32_t fpcr = get_fpcr_aarch64();
+    uint64_t fpcr = get_fpcr_aarch64();
     fpcr = isDefault ? (fpcr | fpcr_dn_mask) : (fpcr & ~fpcr_dn_mask);
     set_fpcr_aarch64(fpcr);
     return 0;