3 files changed, 118 insertions, 1 deletions
diff --git a/software/led_fade/Makefile b/software/led_fade/Makefile
index c42049b..bbdba66 100644
--- a/software/led_fade/Makefile
+++ b/software/led_fade/Makefile
@@ -1,6 +1,6 @@
 TARGET = led_fade
 C_SRCS += led_fade.c
-CFLAGS += -O2 -fno-builtin-printf
+CFLAGS += -O2 -fno-builtin-printf -DNO_INIT
 
 BSP_BASE = ../../bsp
 include $(BSP_BASE)/env/common.mk
diff --git a/software/performance_counters/Makefile b/software/performance_counters/Makefile
new file mode 100644
index 0000000..eb3e8f1
--- /dev/null
+++ b/software/performance_counters/Makefile
@@ -0,0 +1,6 @@
+TARGET = performance_counters
+C_SRCS += performance_counters.c
+CFLAGS += -O2 -fno-builtin-printf
+
+BSP_BASE = ../../bsp
+include $(BSP_BASE)/env/common.mk
diff --git a/software/performance_counters/performance_counters.c b/software/performance_counters/performance_counters.c
new file mode 100644
index 0000000..e7c12a3
--- /dev/null
+++ b/software/performance_counters/performance_counters.c
@@ -0,0 +1,111 @@
+// See LICENSE for license details.
+
+// This demo shows how to use basic
+// RISC-V profiling counters, mcycle
+// (counts the number of processor cycles)
+// and minstret (counts the number of retired instructions). 
+// Note that both are writable as well.
+
+#include <stdio.h>
+
+// The CSR encodings are in this header.
+#include "encoding.h"
+
+// The mcycle counter is 64-bit counter, but since
+// Freedom E platforms use RV32, we must access it as
+// 2 32-bit registers. At 256MHz, the lower bits will
+// roll over approx. every 5 seconds, so we check for
+// rollover with this routine as suggested by the
+// RISC-V Priviledged Architecture Specification.
+
+#define rdmcycle(x)  {				       \
+    uint32_t lo, hi, hi2;			       \
+    __asm__ __volatile__ ("1:\n\t"		       \
+			  "csrr %0, mcycleh\n\t"       \
+			  "csrr %1, mcycle\n\t"	       \
+			  "csrr %2, mcycleh\n\t"       \
+			  "bne  %0, %2, 1b\n\t"			\
+			  : "=r" (hi), "=r" (lo), "=r" (hi2)) ;	\
+    *(x) = lo | ((uint64_t) hi << 32); 				\
+  }
+
+
+// The minstret counter is 64-bit counter, but
+// Freedom E platforms use RV32, we must access it as
+// 2 32-bit registers, same as for mcycle.
+
+#define rdminstret(x)  {			       \
+    uint32_t lo, hi, hi2;			       \
+    __asm__ __volatile__ ("1:\n\t"		       \
+			  "csrr %0, minstreth\n\t"       \
+			  "csrr %1, minstret\n\t"	       \
+			  "csrr %2, minstreth\n\t"       \
+			  "bne  %0, %2, 1b\n\t"			\
+			  : "=r" (hi), "=r" (lo), "=r" (hi2)) ;	\
+    *(x) = lo | ((uint64_t) hi << 32); 				\
+  }
+
+// Simple program to measure the performance of.
+
+int factorial(int i){
+
+  int result = 1;
+  for (int ii = 1; ii <= i; ii++) {
+    result = result * i;
+  }
+
+  return result;
+  
+}
+
+
+int main()
+{
+
+  uint64_t before_cycle;
+  uint64_t before_instret;
+  
+  uint64_t after_cycle;
+  uint64_t after_instret;
+    
+  printf("\n\nDemo 1: Using Counter Differences.\n");
+  
+  for (int ii = 0; ii < 3; ii++){
+    rdmcycle(&before_cycle);
+    rdminstret(&before_instret);
+
+    volatile int result = factorial (100);
+    
+    rdmcycle(&after_cycle);
+    rdminstret(&after_instret);
+    
+    printf("Loop %d: Retired %d instructions in %d cycles\n",
+	   ii,
+	   (uint32_t)(after_instret - before_instret), 
+	   (uint32_t)(after_cycle - before_cycle));
+  }
+
+  printf("\n\nDemo 2: Clearing Counters, Using Values Directly.\n");
+
+  for (int ii = 0; ii < 3; ii++){
+
+    write_csr(mcycle,  0);
+    write_csr(mcycleh, 0);
+    write_csr(minstret, 0);
+    write_csr(minstreth, 0);
+    
+    volatile int result = factorial (100);
+    
+    rdmcycle(&after_cycle);
+    rdminstret(&after_instret);
+    
+    printf("Loop %d: Retired %d instructions in %d cycles\n",
+	   ii,
+	   (uint32_t)(after_instret),
+	   (uint32_t)(after_cycle));
+
+  }
+
+  return 0;
+
+}