Add GPROF

environ_base.sh

@@ -44,6 +44,7 @@ export KOS_AR="${KOS_CC_BASE}/bin/${KOS_CC_PREFIX}-gcc-ar"
 export KOS_OBJCOPY="${KOS_CC_BASE}/bin/${KOS_CC_PREFIX}-objcopy"
 export KOS_OBJDUMP="${KOS_CC_BASE}/bin/${KOS_CC_PREFIX}-objdump"
 export KOS_ADDR2LINE="${KOS_CC_BASE}/bin/${KOS_CC_PREFIX}-addr2line"
+export KOS_GPROF="${KOS_CC_BASE}/bin/${KOS_CC_PREFIX}-gprof"
 export KOS_LD="${KOS_CC_BASE}/bin/${KOS_CC_PREFIX}-ld"
 export KOS_RANLIB="${KOS_CC_BASE}/bin/${KOS_CC_PREFIX}-gcc-ranlib"
 export KOS_STRIP="${KOS_CC_BASE}/bin/${KOS_CC_PREFIX}-strip"

include/sys/gmon.h

@@ -0,0 +1,143 @@
+/* KallistiOS ##version##
+
+   sys/gmon.h
+   Copyright (C) 2025 Andy Barajas
+
+*/
+
+#ifndef __SYS_GMON_H
+#define __SYS_GMON_H
+
+#include <sys/cdefs.h>
+__BEGIN_DECLS
+
+#include <stdint.h>
+#include <stdbool.h>
+
+/** \defgroup debugging_gprof GPROF
+    \brief    GPROF utilities
+    \ingroup  debugging
+
+
+    This file provides utilities for profiling applications using gprof. Gprof
+    is a profiling tool that allows you to analyze where your program spent 
+    its time and which functions called which other functions during execution. 
+    This can help you identify performance bottlenecks and optimize your code.
+
+    Profiling Steps:
+
+    1. Compile your program with profiling flags:
+
+        Use the following flags:
+        ```sh
+        CFLAGS = -pg -fno-omit-frame-pointer -fno-inline
+        ```
+        = or =
+        ```sh
+        CFLAGS = -g -finstrument-functions -fno-inline 
+        LDFLAGS = -lgprof
+        ```
+        These flags enable profiling and prevent function inlining to ensure 
+        accurate profiling data.  If you use the -pg flag, the GCC compiler 
+        inserts trapa #33 instructions into your build.
+
+    2. Running your program to create gmon.out:
+
+        Execute your program normally to completion. This will generate a file 
+        named `gmon.out` in the current directory, which contains the profiling 
+        data.
+
+    3. Running gprof:
+
+        Use the following command to analyze the profiling data:
+        ```sh
+        $(KOS_GPROF) $(TARGET) gmon.out > gprof_output.txt
+        ```
+        Replace `$(TARGET)` with the name of your compiled program. This 
+        command will generate a human-readable report in `gprof_output.txt`.
+
+    4. Visualizing profiling data with gprof2dot:
+
+        To create a graphical representation of the profiling data, use the 
+        `gprof2dot` tool:
+        ```sh
+        gprof2dot gprof_output.txt > graph.dot
+        ```
+        This converts the `gprof` output into a DOT format file, which can be 
+        used to generate various types of graphs.
+
+    5. Generating an image from the DOT file:
+
+        Use the `dot` command from Graphviz to create a JPEG image from the DOT 
+        file:
+        ```sh
+        dot -Tjpg graph.dot -o graph.jpg
+        ```
+        This command will generate a JPEG image (`graph.jpg`) that visually 
+        represents the profiling data.
+
+    By following these steps, you can effectively profile your program and 
+    identify performance bottlenecks for optimization.
+
+    \author Andy Barajas
+
+    @{
+*/
+
+/** \brief  GPROF Trapa Code
+
+    GCC generates this pattern before each profiled function when you compile 
+    your project with the -pg flag:
+
+    trapa #33 --- This is a 2-byte instruction
+    nop       --- This is a 2-byte no-op placeholder
+*/
+#define GPROF_TRAPA_CODE  33
+
+/** \brief  Environment variable for setting the gmon output file prefix.
+
+    This variable allows you to set a custom prefix for the gmon output file
+    generated after profiling. The default output filename is gmon.out.
+    If set, the specified prefix will be appended with the PID of the 
+    running program, which is always 1 for Dreamcast.
+
+    Example:
+    \code
+    setenv(GMON_OUT_PREFIX, "/pc/test.out", 1);
+    \endcode
+    This will generate a file called \c test.out.1 after profiling is complete.
+*/
+#define GMON_OUT_PREFIX   "GMON_OUT_PREFIX"
+
+/** \brief  Start gprof profiling for a specified address range.
+
+    This function is intended for programs not linked with the `-pg' linker 
+    switch. If `-pg' was used during linking, monstartup is automatically 
+    called by the startup code with arguments covering the entire range of 
+    executable addresses, from the program's entry point to the highest code 
+    segment address. Using this method to initialize gprof will only generate 
+    histogram profiling data, without producing a call graph. Profiling starts 
+    immediately after calling this function.
+
+    \param  lowpc   The lower bound of the address range to profile.
+    \param  highpc  The upper bound of the address range to profile.
+*/
+void monstartup(uintptr_t lowpc, uintptr_t highpc);
+
+/** \brief  Restart or stop gprof profiling.
+
+    This function restarts or stops gprof profiling. It does NOT start gprof 
+    profiling initially, as gprof profiling starts before the program enters 
+    the main function. You can use this function to stop profiling and then 
+    restart it later when you reach the section of code you want to profile.
+
+    \param  enable            A boolean value to restart (true) or stop (false) 
+                              gprof profiling.
+*/
+void moncontrol(bool enable);
+
+/** @} */
+
+__END_DECLS
+
+#endif  /* __SYS_GMON_H */

kernel/arch/dreamcast/kernel/init.c

@@ -140,7 +140,7 @@ KOS_INIT_FLAG_WEAK(fs_dclsocket_shutdown, true);
 /* Auto-init stuff: override with a non-weak symbol if you don't want all of
    this to be linked into your code (and do the same with the
    arch_auto_shutdown function too). */
-int  __weak arch_auto_init(void) {
+int __weak arch_auto_init(void) {
     /* Initialize memory management */
     mm_init();
 
@@ -172,7 +172,7 @@ int  __weak arch_auto_init(void) {
     timer_init();           /* Timers */
     hardware_sys_init();        /* DC low-level hardware init */
 
-    if (!KOS_PLATFORM_IS_NAOMI)
+    if(!KOS_PLATFORM_IS_NAOMI)
         syscall_sysinfo_init();
 
     /* Initialize our timer */
@@ -206,7 +206,7 @@ int  __weak arch_auto_init(void) {
 
     KOS_INIT_FLAG_CALL(dcload_init);
 
-    if (!KOS_PLATFORM_IS_NAOMI)
+    if(!KOS_PLATFORM_IS_NAOMI)
         KOS_INIT_FLAG_CALL(fs_iso9660_init);
 
     KOS_INIT_FLAG_CALL(vmu_fs_init);
@@ -220,15 +220,15 @@ int  __weak arch_auto_init(void) {
         KOS_INIT_FLAG_CALL(maple_wait_scan);  /* Wait for the maple scan to complete */
     }
 
-    if (!KOS_PLATFORM_IS_NAOMI)
+    if(!KOS_PLATFORM_IS_NAOMI)
         KOS_INIT_FLAG_CALL(arch_init_net);
 
     return 0;
 }
 
-void  __weak arch_auto_shutdown(void) {
+void __weak arch_auto_shutdown(void) {
     KOS_INIT_FLAG_CALL(fs_dclsocket_shutdown);
-    if (!KOS_PLATFORM_IS_NAOMI)
+    if(!KOS_PLATFORM_IS_NAOMI)
         KOS_INIT_FLAG_CALL(net_shutdown);
 
     snd_shutdown();
@@ -244,7 +244,7 @@ void  __weak arch_auto_shutdown(void) {
     library_shutdown();
     KOS_INIT_FLAG_CALL(fs_dcload_shutdown);
     KOS_INIT_FLAG_CALL(vmu_fs_shutdown);
-    if (!KOS_PLATFORM_IS_NAOMI)
+    if(!KOS_PLATFORM_IS_NAOMI)
         KOS_INIT_FLAG_CALL(fs_iso9660_shutdown);
 #if defined(__NEWLIB__) && !(__NEWLIB__ < 2 && __NEWLIB_MINOR__ < 4)
     fs_rnd_shutdown();
@@ -259,13 +259,16 @@ void  __weak arch_auto_shutdown(void) {
     rtc_shutdown();
 }
 
+/* Strongly defined in gcrt1.s and linked when compiling with -pg */
+void __weak gprof_init(void) { }
+
 /* This is the entry point inside the C program */
 void arch_main(void) {
     uint8 *bss_start = (uint8 *)(&_bss_start);
     uint8 *bss_end = (uint8 *)(&end);
     int rv;
 
-    if (KOS_PLATFORM_IS_NAOMI) {
+    if(KOS_PLATFORM_IS_NAOMI) {
         /* Ugh. I'm really not sure why we have to set up these DMA registers this
            way on boot, but failing to do so breaks maple... */
         DMAC_SAR2 = 0;
@@ -296,6 +299,9 @@ void arch_main(void) {
     /* Run ctors */
     _init();
 
+    /* Setup gprof */
+    gprof_init();
+
     /* Call the user's main function */
     rv = main(0, NULL);
 

kernel/arch/dreamcast/kernel/startup.s

@@ -1,7 +1,8 @@
 ! KallistiOS ##version##
 !
-! startup.s
-! (c)2000-2001 Megan Potter
+! startup.S
+! Copyright (C) 2000-2001 Megan Potter
+! Copyright (C) 2025 Andress Barajas
 !
 ! This file is added to GCC during the patching stage of toolchain
 ! compilation. Any changes to this file will not take effect until the
@@ -21,6 +22,9 @@
 .globl __arch_old_stack
 .globl __arch_old_fpscr
 .globl __arch_mem_top
+#ifdef PROFILE
+.globl _gprof_init
+#endif
 
 .weak   _arch_stack_16m
 .weak   _arch_stack_32m
@@ -179,8 +183,33 @@ normal_exit:
 	rts
 	nop
 
+#ifdef PROFILE
+
+! Call profiler startup code
+	.align 2
+_gprof_init:
+	mov.l lowpc, r4
+	mov.l highpc, r5
+	mov.l monstartup_addr, r2
+	jmp @r2
+	nop
+
+! GPROF variables
+	.align 2
+
+monstartup_addr:
+	.long __monstartup
+lowpc:
+	.long __executable_start
+highpc:
+	.long __etext
+
+#endif /* PROFILE */
+
+
 ! Misc variables
 	.align	2
+
 dcload_magic_addr:
 	.long	0x8c004004
 dcload_magic_value:

kernel/libc/koslib/Makefile

@@ -14,6 +14,6 @@ OBJS = abort.o byteorder.o memset2.o memset4.o memcpy2.o memcpy4.o \
 	creat.o sleep.o rmdir.o rename.o inet_pton.o inet_ntop.o \
 	inet_ntoa.o inet_aton.o poll.o select.o symlink.o readlink.o \
 	gethostbyname.o getaddrinfo.o dirfd.o nanosleep.o basename.o dirname.o \
-	sched_yield.o dup.o dup2.o pipe.o
+	sched_yield.o dup.o dup2.o pipe.o gmon.o
 
 include $(KOS_BASE)/Makefile.prefab