Merge pull request eembc#4 from sifive/align_linux64

Align linux64 performance tweak to freedom_metal

README.md

@@ -128,11 +128,11 @@ What is and is not allowed.
 ~~~
 4. All source files must be compiled with the same flags.
 5. All data type sizes must match size in bits such that:
-	* `ee_u8` is an 8 bits datatype.
-	* `ee_s16` is a 16 bits datatype.
-	* `ee_u16` is a 16 bits datatype.
-	* `ee_s32` is a 32 bits datatype.
-	* `ee_u32` is a 32 bits datatype.
+	* `ee_u8` is an unsigned 8-bit datatype.
+	* `ee_s16` is a signed 16-bit datatype.
+	* `ee_u16` is an unsigned 16-bit datatype.
+	* `ee_s32` is a signed 32-bit datatype.
+	* `ee_u32` is an unsigned 32-bit datatype.
 
 ## Allowed
 

barebones/ee_printf.c

@@ -29,9 +29,9 @@ limitations under the License.
 
 static char *digits = "0123456789abcdefghijklmnopqrstuvwxyz";
 static char *upper_digits = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ";
-static size_t strnlen(const char *s, size_t count);
+static ee_size_t strnlen(const char *s, ee_size_t count);
 
-static size_t strnlen(const char *s, size_t count)
+static ee_size_t strnlen(const char *s, ee_size_t count)
 {
   const char *sc;
   for (sc = s; *sc != '\0' && count--; ++sc);
@@ -183,7 +183,7 @@ static char *iaddr(char *str, unsigned char *addr, int size, int precision, int
   return str;
 }
 
-#ifdef HAS_FLOAT
+#if HAS_FLOAT
 
 char *ecvtbuf(double arg, int ndigits, int *decpt, int *sign, char *buf);
 char *fcvtbuf(double arg, int ndigits, int *decpt, int *sign, char *buf);
@@ -529,7 +529,7 @@ static int ee_vsprintf(char *buf, const char *fmt, va_list args)
       case 'u':
         break;
 
-#ifdef HAS_FLOAT
+#if HAS_FLOAT
 
       case 'f':
         str = flt(str, va_arg(args, double), field_width, precision, *fmt, flags | SIGN);

freedom-metal/core_portme.c

@@ -36,9 +36,11 @@ Original Author: Shay Gal-on
 #endif
 	volatile ee_s32 seed4_volatile=ITERATIONS;
 	volatile ee_s32 seed5_volatile=0;
+
+	unsigned long long timebase = 0;
 /* Porting : Timing functions
 	How to capture time and convert to seconds must be ported to whatever is supported by the platform.
-	e.g. Read value from on board RTC, read value from cpu clock cycles performance counter etc. 
+	e.g. Read value from on board RTC, read value from cpu clock cycles performance counter etc.
 	Sample implementation for standard time.h and windows.h definitions included.
 */
 CORETIMETYPE barebones_clock() {
@@ -54,36 +56,36 @@ CORETIMETYPE barebones_clock() {
 #define MYTIMEDIFF(fin,ini) ((fin)-(ini))
 #define TIMER_RES_DIVIDER 1
 #define SAMPLE_TIME_IMPLEMENTATION 1
-#define EE_TICKS_PER_SEC (CLOCKS_PER_SEC / TIMER_RES_DIVIDER)
+#define EE_TICKS_PER_SEC (timebase/ TIMER_RES_DIVIDER)
 
 /** Define Host specific (POSIX), or target specific global time variables. */
 static CORETIMETYPE start_time_val, stop_time_val;
 
 /* Function : start_time
 	This function will be called right before starting the timed portion of the benchmark.
 
-	Implementation may be capturing a system timer (as implemented in the example code) 
+	Implementation may be capturing a system timer (as implemented in the example code)
 	or zeroing some system parameters - e.g. setting the cpu clocks cycles to 0.
 */
 void start_time(void) {
-	GETMYTIME(&start_time_val );      
+	GETMYTIME(&start_time_val );
 }
 /* Function : stop_time
 	This function will be called right after ending the timed portion of the benchmark.
 
-	Implementation may be capturing a system timer (as implemented in the example code) 
+	Implementation may be capturing a system timer (as implemented in the example code)
 	or other system parameters - e.g. reading the current value of cpu cycles counter.
 */
 void stop_time(void) {
-	GETMYTIME(&stop_time_val );      
+	GETMYTIME(&stop_time_val );
 }
 /* Function : get_time
 	Return an abstract "ticks" number that signifies time on the system.
-	
+
 	Actual value returned may be cpu cycles, milliseconds or any other value,
 	as long as it can be converted to seconds by <time_in_secs>.
 	This methodology is taken to accomodate any hardware or simulated platform.
-	The sample implementation returns millisecs by default, 
+	The sample implementation returns millisecs by default,
 	and the resolution is controlled by <TIMER_RES_DIVIDER>
 */
 CORE_TICKS get_time(void) {
@@ -104,7 +106,7 @@ secs_ret time_in_secs(CORE_TICKS ticks) {
 ee_u32 default_num_contexts=1;
 
 /* Function : portable_init
-	Target specific initialization code 
+	Target specific initialization code
 	Test for some common mistakes.
 */
 void portable_init(core_portable *p, int *argc, char *argv[])
@@ -115,10 +117,13 @@ void portable_init(core_portable *p, int *argc, char *argv[])
 	if (sizeof(ee_u32) != 4) {
 		ee_printf("ERROR! Please define ee_u32 to a 32b unsigned type!\n");
 	}
+	//get timer frequency
+	metal_timer_get_timebase_frequency(0, &timebase);
+
 	p->portable_id=1;
 }
 /* Function : portable_fini
-	Target specific final code 
+	Target specific final code
 */
 void portable_fini(core_portable *p)
 {

freedom-metal/core_portme.h

@@ -23,10 +23,10 @@ Original Author: Shay Gal-on
 /************************/
 /* Data types and settings */
 /************************/
-/* Configuration : HAS_FLOAT 
+/* Configuration : HAS_FLOAT
 	Define to 1 if the platform supports floating point.
 */
-#ifndef HAS_FLOAT 
+#ifndef HAS_FLOAT
 #define HAS_FLOAT 1
 #endif
 /* Configuration : HAS_TIME_H
@@ -60,23 +60,23 @@ Original Author: Shay Gal-on
 /* Definitions : COMPILER_VERSION, COMPILER_FLAGS, MEM_LOCATION
 	Initialize these strings per platform
 */
-#ifndef COMPILER_VERSION 
+#ifndef COMPILER_VERSION
  #ifdef __GNUC__
  #define COMPILER_VERSION "GCC"__VERSION__
  #else
  #define COMPILER_VERSION "Please put compiler version here (e.g. gcc 4.1)"
  #endif
 #endif
-#ifndef COMPILER_FLAGS 
+#ifndef COMPILER_FLAGS
  #define COMPILER_FLAGS FLAGS_STR /* "Please put compiler flags here (e.g. -o3)" */
 #endif
-#ifndef MEM_LOCATION 
+#ifndef MEM_LOCATION
  #define MEM_LOCATION "STACK"
 #endif
 
 /* Data Types :
 	To avoid compiler issues, define the data types that need ot be used for 8b, 16b and 32b in <core_portme.h>.
-	
+
 	*Imprtant* :
 	ee_ptr_int needs to be the data type used to hold pointers, otherwise coremark may fail!!!
 */
@@ -85,7 +85,7 @@ typedef unsigned short ee_u16;
 typedef signed int ee_s32;
 typedef double ee_f32;
 typedef unsigned char ee_u8;
-typedef unsigned int ee_u32;
+typedef signed int ee_u32;
 typedef signed long ee_u64;
 #if __riscv_xlen == 32
 typedef ee_u32 ee_ptr_int;
@@ -102,12 +102,12 @@ typedef signed int ee_size_t;
 /* Configuration : CORE_TICKS
 	Define type of return from the timing functions.
  */
-#define CORETIMETYPE ee_u32 
+#define CORETIMETYPE ee_u32
 typedef ee_u32 CORE_TICKS;
 
 /* Configuration : SEED_METHOD
 	Defines method to get seed values that cannot be computed at compile time.
-	
+
 	Valid values :
 	SEED_ARG - from command line.
 	SEED_FUNC - from a system function.
@@ -119,7 +119,7 @@ typedef ee_u32 CORE_TICKS;
 
 /* Configuration : MEM_METHOD
 	Defines method to get a block of memry.
-	
+
 	Valid values :
 	MEM_MALLOC - for platforms that implement malloc and have malloc.h.
 	MEM_STATIC - to use a static memory array.
@@ -130,19 +130,19 @@ typedef ee_u32 CORE_TICKS;
 #endif
 
 /* Configuration : MULTITHREAD
-	Define for parallel execution 
-	
+	Define for parallel execution
+
 	Valid values :
 	1 - only one context (default).
 	N>1 - will execute N copies in parallel.
-	
-	Note : 
+
+	Note :
 	If this flag is defined to more then 1, an implementation for launching parallel contexts must be defined.
-	
+
 	Two sample implementations are provided. Use <USE_PTHREAD> or <USE_FORK> to enable them.
-	
+
 	It is valid to have a different implementation of <core_start_parallel> and <core_end_parallel> in <core_portme.c>,
-	to fit a particular architecture. 
+	to fit a particular architecture.
 */
 #ifndef MULTITHREAD
 #define MULTITHREAD 1
@@ -152,22 +152,22 @@ typedef ee_u32 CORE_TICKS;
 #endif
 
 /* Configuration : MAIN_HAS_NOARGC
-	Needed if platform does not support getting arguments to main. 
-	
+	Needed if platform does not support getting arguments to main.
+
 	Valid values :
 	0 - argc/argv to main is supported
 	1 - argc/argv to main is not supported
-	
-	Note : 
+
+	Note :
 	This flag only matters if MULTITHREAD has been defined to a value greater then 1.
 */
-#ifndef MAIN_HAS_NOARGC 
+#ifndef MAIN_HAS_NOARGC
 #define MAIN_HAS_NOARGC 0
 #endif
 
 /* Configuration : MAIN_HAS_NORETURN
-	Needed if platform does not support returning a value from main. 
-	
+	Needed if platform does not support returning a value from main.
+
 	Valid values :
 	0 - main returns an int, and return value will be 0.
 	1 - platform does not support returning a value from main
@@ -201,4 +201,5 @@ void portable_fini(core_portable *p);
 
 int ee_printf(const char *fmt, ...);
 
+extern int metal_timer_get_timebase_frequency(int hartid, unsigned long long *timebase);
 #endif /* CORE_PORTME_H */

freedom-metal/core_portme.mak

@@ -11,7 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-# 
+#
 # Original Author: Shay Gal-on
 
 #File : core_portme.mak
@@ -26,23 +26,23 @@ OUTFLAG= -o
 #CC 		= gcc
 # Flag : LD
 #	Use this flag to define compiler to use
-#LD		= 
+#LD		=
 # Flag : AS
 #	Use this flag to define compiler to use
 #AS		= gas
 # Flag : CFLAGS
 #	Use this flag to define compiler options. Note, you can add compiler options from the command line using XCFLAGS="other flags"
-PORT_CFLAGS = -specs=nano.specs -O3 -DITERATIONS=10000 -DPERFORMANCE_RUN=1 -DMAIN_HAS_NOARGC=1 -DMAIN_HAS_NORETURN=1 -DHAS_STDIO -DHAS_PRINTF -DHAS_TIME_H -DUSE_CLOCK
+PORT_CFLAGS = -O3 -DITERATIONS=3000 -DPERFORMANCE_RUN=1 -DMAIN_HAS_NOARGC=1 -DMAIN_HAS_NORETURN=1 -DHAS_STDIO -DHAS_PRINTF -DHAS_TIME_H -DUSE_CLOCK
 FLAGS_STR = "$(PORT_CFLAGS) $(XCFLAGS) $(XLFLAGS) $(LFLAGS_END)"
 override CFLAGS += $(PORT_CFLAGS) -I$(PORT_DIR) -I. -DFLAGS_STR=\"$(FLAGS_STR)\" -Xlinker --defsym=__stack_size=0x1000
 #Flag : LFLAGS_END
-#	Define any libraries needed for linking or other flags that should come at the end of the link line (e.g. linker scripts). 
+#	Define any libraries needed for linking or other flags that should come at the end of the link line (e.g. linker scripts).
 #	Note : On certain platforms, the default clock_gettime implementation is supported but requires linking of librt.
 SEPARATE_COMPILE =
 # Flag : SEPARATE_COMPILE
 # You must also define below how to create an object file, and how to link.
 OBJOUT 	= -o
-LFLAGS 	= 
+LFLAGS 	=
 ASFLAGS =
 OFLAG 	= -o
 COUT 	= -c
@@ -80,7 +80,7 @@ $(OPATH)$(PORT_DIR)/%$(OEXT) : %.s
 # For the purpose of this simple port, no pre or post steps needed.
 
 .PHONY : port_prebuild port_postbuild port_prerun port_postrun port_preload port_postload
-port_pre% port_post% : 
+port_pre% port_post% :
 
 # FLAG : OPATH
 # Path to the output folder. Default - current folder.