fd_tickcount support on Arm via cntvct_el0

Author: riptl

config/extra/with-arm.mk

@@ -37,3 +37,6 @@ include config/extra/with-rocksdb.mk
 endif
 
 FD_ARCH_SUPPORTS_SANDBOX:=1
+
+CPPFLAGS+=-DFD_HAS_ARM=1
+FD_HAS_ARM:=1

src/app/firedancer/commands/shred_version.c

@@ -4,6 +4,7 @@
 #include "../../../disco/topo/fd_topo.h"
 #include "../../../discof/ipecho/fd_ipecho_client.h"
 
+#include <stdlib.h>
 #include <unistd.h>
 
 void

src/util/fd_util_base.h

@@ -181,6 +181,14 @@
 #define FD_HAS_AESNI 0
 #endif
 
+/* FD_HAS_ARM:  If the build target supports armv8-a specific features
+   and can benefit from aarch64 specific optimizations, define
+   FD_HAS_ARM. */
+
+#ifndef FD_HAS_ARM
+#define FD_HAS_ARM 0
+#endif
+
 /* FD_HAS_LZ4 indicates that the target supports LZ4 compression.
    Roughly, does "#include <lz4.h>" and the APIs therein work? */
 
@@ -1245,6 +1253,8 @@ fd_hash_memcpy( ulong                    seed,
 #ifndef FD_TICKCOUNT_STYLE
 #if FD_HAS_X86 /* Use RDTSC */
 #define FD_TICKCOUNT_STYLE 1
+#elif FD_HAS_ARM /* Use CNTVCT_EL0 */
+#define FD_TICKCOUNT_STYLE 2
 #else /* Use portable fallback */
 #define FD_TICKCOUNT_STYLE 0
 #endif
@@ -1290,6 +1300,23 @@ fd_hash_memcpy( ulong                    seed,
 
 #define fd_tickcount() ((long)__builtin_ia32_rdtsc())
 
+#elif FD_TICKCOUNT_STYLE==2 /* armv8 (fast) */
+
+/* fd_tickcount (ARM): https://developer.arm.com/documentation/ddi0601/2021-12/AArch64-Registers/CNTVCT-EL0--Counter-timer-Virtual-Count-register
+   Approx 24 MHz on Apple M1. */
+
+static inline long
+fd_tickcount( void ) {
+  /* consider using 'isb' */
+  ulong value;
+  __asm__ __volatile__ (
+    "isb\n"
+    "mrs %0, cntvct_el0\n"
+    "nop"
+    : "=r" (value) );
+  return (long)value;
+}
+
 #else
 #error "Unknown FD_TICKCOUNT_STYLE"
 #endif