From 0f85e9123e781b6ec7fa78fa1c54b1de40179a23 Mon Sep 17 00:00:00 2001
From: Malcolm <malcolmjestadt@gmail.com>
Date: Thu, 15 Jan 2026 04:20:25 +0000
Subject: [PATCH] utils: Scale busy_wait according to arm timer frequency > -
 This is a fatal issue that was impacting arm builds, since busy_waits were
 written assuming an approx 3Ghz x86 machine > and most arm machines have a
 hardware timer that runs south of 100mhz, meaning the top items in the
 profiler were calls to busy_wait(); > all over the code. Fixing this is a
 very significant speedup, on my snapdragon 8 gen 2 device. 27->37FPS in Metal
 Gear Rising, but almost > all games benefit when run on ARM.

---
 rpcs3/rpcs3.cpp    |  5 +++++
 rpcs3/util/asm.hpp | 29 +++++++++++++++++++++++++++--
 2 files changed, 32 insertions(+), 2 deletions(-)

diff --git a/rpcs3/rpcs3.cpp b/rpcs3/rpcs3.cpp
index 007de26bdb..6c2130eca6 100644
--- a/rpcs3/rpcs3.cpp
+++ b/rpcs3/rpcs3.cpp
@@ -26,6 +26,7 @@
 #include "Utilities/sema.h"
 #include "Utilities/date_time.h"
 #include "util/console.h"
+#include "util/asm.hpp"
 #include "Crypto/decrypt_binaries.h"
 #ifdef _WIN32
 #include "module_verifier.hpp"
@@ -681,6 +682,10 @@ int run_rpcs3(int argc, char** argv)
 		logs::set_init({std::move(ver), std::move(sys), std::move(os), std::move(qt), std::move(time)});
 	}
 
+#ifdef ARCH_ARM64
+	utils::init_arm_timer_scale();
+#endif
+
 #ifdef _WIN32
 	sys_log.notice("Initialization times before main(): %fGc", intro_cycles / 1000000000.);
 #elif defined(RUSAGE_THREAD)
diff --git a/rpcs3/util/asm.hpp b/rpcs3/util/asm.hpp
index 8942dc2a09..56aa955652 100644
--- a/rpcs3/util/asm.hpp
+++ b/rpcs3/util/asm.hpp
@@ -183,10 +183,35 @@ namespace utils
 #endif
 	}
 
-	// Synchronization helper (cache-friendly busy waiting)
-	inline void busy_wait(usz cycles = 3000)
+	// The hardware clock on many arm timers run south of 100mhz
+	// and the busy waits in RPCS3 were written assuming an x86 machine
+	// with hardware timers that run around 3GHz.
+	// For instance, on the snapdragon 8 gen 2, the hardware timer runs at 19.2mhz.
+	// This means that a busy wait that would have taken nanoseconds on x86 will run for
+	// many microseconds on many arm machines. 
+#ifdef ARCH_ARM64
+
+	inline u64 arm_timer_scale = 1;
+
+	inline void init_arm_timer_scale()
 	{
+		u64 freq = 0;
+		asm volatile("mrs %0, cntfrq_el0" : "=r"(freq));
+		
+		// Try to scale hardware timer to match 3GHz
+		u64 timer_scale = freq / 30000000;
+		if (timer_scale)
+			arm_timer_scale = timer_scale;
+	}
+#endif
+		
+	inline void busy_wait(u64 cycles = 3000)
+	{
+#ifdef ARCH_ARM64
+		const u64 stop = get_tsc() + ((cycles / 100) * arm_timer_scale);
+#else
 		const u64 stop = get_tsc() + cycles;
+#endif
 		do pause();
 		while (get_tsc() < stop);
 	}