Merge pull request #347 from fjtrujy/fix_pg

Fix gprof profiling to work with both PBP and PRX executables

Author: fjtrujy

src/prof/prof.c

@@ -81,6 +81,17 @@ static int initialized = 0;
 extern int _ftext;
 extern int _etext;
 
+/// _start is the entry point defined in both crt0.c and crt0_prx.c
+extern int _start;
+
+/// module_start is only defined in PRX startup code (crt0_prx.c) as an alias for _start
+/// Using weak reference allows us to detect PRX vs PBP at runtime
+/// We also verify module_start == _start to handle the case where a PBP defines its own module_start
+extern int module_start __attribute__((weak));
+
+/// relocation offset: runtime_address - link_address (for PRX)
+static unsigned int reloc_offset;
+
 /* forward declarations */
 __attribute__((__no_instrument_function__, __no_profile_instrument_function__))
 void __gprof_cleanup(void);
@@ -104,20 +115,44 @@ static void initialize()
         memset(&gp, '\0', sizeof(gp));
         gp.state = GMON_PROF_ON;
 
-        /* For PRX modules, the code is relocated at load time.
-           &_ftext and &_etext give us runtime (relocated) addresses.
-           Link-time addresses start at 0 for PSP executables.
-
-           Runtime:   &_ftext = relocated_base, &_etext = relocated_base + text_size
-           Link-time: _ftext = 0, _etext = text_size
-
-           So: lowpc_link = 0
-               highpc_link = &_etext - &_ftext (which equals text_size)
+        /* Detect PRX vs PBP using the module_start symbol.
+           In crt0_prx.c, module_start is defined as an alias for _start,
+           so they share the same address. We check both conditions:
+           1. module_start exists (not NULL)
+           2. module_start and _start are at the same address
+           This handles edge cases where a PBP defines its own module_start function.
+
+           For PRX (relocatable modules):
+           - Linked at address 0 (or small offset like 0x20 for C++)
+           - Relocated at load time to a PAGE-ALIGNED address (e.g., 0x08804000)
+           - reloc_offset = page-aligned load base
+           - link_addr = runtime_addr - reloc_offset
+
+           For PBP (standard executables):
+           - Linked at address 0x08804000 + offset (typically _ftext = 0x0880403c)
+           - Loaded at same address (no relocation)
+           - reloc_offset = 0
+           - link_addr = runtime_addr
         */
         gp.lowpc = (unsigned int)&_ftext;
         gp.highpc = (unsigned int)&_etext;
-        gp.lowpc_link = 0;
-        gp.highpc_link = (unsigned int)&_etext - (unsigned int)&_ftext;
+
+        unsigned int runtime_base = (unsigned int)&_ftext & 0x0FFFFFFF;
+
+        if (&module_start != NULL && (void *)&module_start == (void *)&_start) {
+                /* PRX: code was relocated from near-zero to runtime address.
+                   The load base is page-aligned, so we can compute it by masking
+                   off the page offset. The link-time _ftext is the page offset. */
+                unsigned int page_offset = runtime_base & 0xFFF;
+                reloc_offset = runtime_base - page_offset;  /* Page-aligned load base */
+                gp.lowpc_link = page_offset;                /* Link-time _ftext */
+                gp.highpc_link = (gp.highpc - gp.lowpc) + page_offset;
+        } else {
+                /* PBP: no relocation, addresses match ELF */
+                reloc_offset = 0;
+                gp.lowpc_link = runtime_base;
+                gp.highpc_link = (unsigned int)&_etext & 0x0FFFFFFF;
+        }
         gp.textsize = gp.highpc - gp.lowpc;
         gp.hashfraction = HISTFRACTION;
 
@@ -278,16 +313,17 @@ void __mcount(unsigned int frompc, unsigned int selfpc)
                 return;
         }
 
-        /* Mask upper bits and convert to link-time addresses.
-           Link-time addresses = runtime addresses - gp.lowpc (since lowpc_link = 0) */
-        frompc = (frompc & 0x0FFFFFFF) - gp.lowpc;
-        selfpc = (selfpc & 0x0FFFFFFF) - gp.lowpc;
+        /* Mask upper bits to normalize cached/uncached addresses,
+           then subtract relocation offset to get link-time addresses. */
+        frompc = (frompc & 0x0FFFFFFF) - reloc_offset;
+        selfpc = (selfpc & 0x0FFFFFFF) - reloc_offset;
 
-        /* Check if within text section (using link-time range) */
-        if (frompc <= gp.highpc_link)
+        /* Check if within text section (using link-time addresses) */
+        if (frompc >= gp.lowpc_link && frompc < gp.highpc_link)
         {
                 gp.pc = selfpc;
-                e = frompc / gp.hashfraction;
+                /* Arc index is based on offset within text section */
+                e = (frompc - gp.lowpc_link) / gp.hashfraction;
                 arc = gp.arcs + e;
                 arc->frompc = frompc;
                 arc->selfpc = selfpc;
@@ -300,14 +336,15 @@ void __mcount(unsigned int frompc, unsigned int selfpc)
 __attribute__((__no_instrument_function__, __no_profile_instrument_function__))
 static SceUInt timer_handler(SceUID uid, SceKernelSysClock *requested, SceKernelSysClock *actual, void *common)
 {
-        unsigned int frompc = gp.pc;  /* Already in link-time address */
+        unsigned int pc = gp.pc;  /* Already converted to link-time address by __mcount */
 
         if (gp.state == GMON_PROF_ON)
         {
-                /* Check if within text section (using link-time range) */
-                if (frompc <= gp.highpc_link)
+                /* Check if within text section (using link-time addresses) */
+                if (pc >= gp.lowpc_link && pc < gp.highpc_link)
                 {
-                        int e = frompc / gp.hashfraction;
+                        /* Sample index is based on offset within text section */
+                        int e = (pc - gp.lowpc_link) / gp.hashfraction;
                         gp.samples[e]++;
                 }
         }

src/samples/gprof/basic/Makefile.sample

@@ -1,14 +1,17 @@
 TARGET = gprofbasic
 OBJS = main.o
 
-INCDIR = 
+INCDIR =
 CFLAGS = -O2 -Wall -pg -g
 CXXFLAGS = $(CFLAGS) -fno-exceptions -fno-rtti
 ASFLAGS = $(CFLAGS)
 
 LIBDIR =
 LDFLAGS = -pg -g
 
+BUILD_PRX = 1
+PSP_FW_VERSION = 600
+
 EXTRA_TARGETS = EBOOT.PBP
 PSP_EBOOT_TITLE = GProf Basic Example
 

src/samples/gprof/basic_cpp/Makefile.sample

@@ -0,0 +1,20 @@
+TARGET = gprofbasic_cpp
+OBJS = main.o
+
+INCDIR =
+CFLAGS = -O2 -Wall -pg -g
+CXXFLAGS = $(CFLAGS) -fno-exceptions -fno-rtti
+ASFLAGS = $(CFLAGS)
+
+LIBDIR =
+LDFLAGS = -pg -g
+LIBS = -lstdc++
+
+BUILD_PRX = 1
+PSP_FW_VERSION = 600
+
+EXTRA_TARGETS = EBOOT.PBP
+PSP_EBOOT_TITLE = GProf Basic C++ Example
+
+PSPSDK=$(shell psp-config --pspsdk-path)
+include $(PSPSDK)/lib/build.mak

src/samples/gprof/basic_cpp/README.md

@@ -0,0 +1,72 @@
+Sample program to show how to use the `gprof` feature with C++.
+
+The requirements are quite easy, just adding `-g -pg` flags to the `CFLAGS` and `LDFLAGS` is enough to make things work out of the box. For C++ programs, you also need to link with `-lstdc++`.
+
+This example demonstrates:
+- Object-oriented programming with a `PrimeCalculator` class
+- C++ standard library features (`std::vector`, `std::cout`)
+- Profiling C++ code with gprof
+
+Firstly execute your program, then once program ends it will automatically generate a `gmon.out` file at CWD level.
+
+In order to inspect the content of the generated file you need to use the `psp-gprof` binary.
+
+For instance, following the next syntax:
+```
+psp-gprof -b {binary.elf} gmon.out
+```
+
+like:
+```
+psp-gprof -b gprofbasic_cpp.elf gmon.out
+```
+
+
+It will show something like:
+```
+Flat profile:
+
+Each sample counts as 0.001 seconds.
+  %   cumulative   self              self     total           
+ time   seconds   seconds    calls  ms/call  ms/call  name    
+ 95.98      0.17     0.17   104728     0.00     0.00  PrimeCalculator::isPrime(int) const
+  4.02      0.17     0.01        1     7.00     7.00  dummyFunction()
+  0.00      0.17     0.00        1     0.00   174.00  main
+  0.00      0.17     0.00        1     0.00   167.00  PrimeCalculator::sumOfSquareRoots(int) const
+
+
+                        Call graph
+
+
+granularity: each sample hit covers 2 byte(s) for 0.57% of 0.17 seconds
+
+index % time    self  children    called     name
+                0.00    0.17       1/1           _main [2]
+[1]    100.0    0.00    0.17       1         main [1]
+                0.00    0.17       1/1           PrimeCalculator::sumOfSquareRoots(int) const [4]
+                0.01    0.00       1/1           dummyFunction() [5]
+-----------------------------------------------
+                                                 <spontaneous>
+[2]    100.0    0.00    0.17                 _main [2]
+                0.00    0.17       1/1           main [1]
+-----------------------------------------------
+                0.17    0.00  104728/104728      PrimeCalculator::sumOfSquareRoots(int) const [4]
+[3]     96.0    0.17    0.00  104728         PrimeCalculator::isPrime(int) const [3]
+-----------------------------------------------
+                0.00    0.17       1/1           main [1]
+[4]     96.0    0.00    0.17       1         PrimeCalculator::sumOfSquareRoots(int) const [4]
+                0.17    0.00  104728/104728      PrimeCalculator::isPrime(int) const [3]
+-----------------------------------------------
+                0.01    0.00       1/1           main [1]
+[5]      4.0    0.01    0.00       1         dummyFunction() [5]
+-----------------------------------------------
+
+
+Index by function name
+
+   [5] dummyFunction()          [1] main
+   [3] PrimeCalculator::isPrime(int) const
+   [4] PrimeCalculator::sumOfSquareRoots(int) const
+```
+
+Cheers

src/samples/gprof/basic_cpp/main.cpp

@@ -0,0 +1,107 @@
+/*
+ * PSP Software Development Kit - https://github.com/pspdev
+ * -----------------------------------------------------------------------
+ * Licensed under the BSD license, see LICENSE in PSPSDK root for details.
+ *
+ * main.cpp - A basic C++ example for checking the GProf profiler.
+ *
+ * Copyright (c) 2024 Francisco Javier Trujillo Mata - fjtrujy@gmail.com
+ *
+ */
+#include <cstdlib>
+#include <cstring>
+#include <cstdio>
+#include <cmath>
+#include <iostream>
+#include <vector>
+
+#include <pspmoduleinfo.h>
+#include <pspthreadman.h>
+
+PSP_MODULE_INFO("GProf Basic C++ Example", 0, 1, 1);
+PSP_MAIN_THREAD_ATTR(THREAD_ATTR_USER | THREAD_ATTR_VFPU);
+
+// Class to handle prime number operations
+class PrimeCalculator
+{
+public:
+	// Method to check if a number is prime
+	bool isPrime(int num) const
+	{
+		if (num <= 1)
+			return false;
+		if (num <= 3)
+			return true;
+		if (num % 2 == 0 || num % 3 == 0)
+			return false;
+		for (int i = 5; i * i <= num; i += 6)
+		{
+			if (num % i == 0 || num % (i + 2) == 0)
+				return false;
+		}
+		return true;
+	}
+
+	// Method to compute the sum of square roots of the first N prime numbers
+	double sumOfSquareRoots(int N) const
+	{
+		int count = 0;
+		int num = 2;
+		double sum = 0.0;
+
+		while (count < N)
+		{
+			if (isPrime(num))
+			{
+				sum += std::sqrt(num);
+				count++;
+			}
+			num++;
+		}
+		return sum;
+	}
+
+	// Method to get the first N prime numbers
+	std::vector<int> getFirstNPrimes(int N) const
+	{
+		std::vector<int> primes;
+		int num = 2;
+
+		while (primes.size() < static_cast<size_t>(N))
+		{
+			if (isPrime(num))
+			{
+				primes.push_back(num);
+			}
+			num++;
+		}
+		return primes;
+	}
+};
+
+void dummyFunction()
+{
+	for (int i = 0; i < 10000; i++)
+	{
+		std::cout << ".";
+	}
+	std::cout << std::endl;
+}
+
+int main(int argc, char *argv[])
+{
+	std::cout << "Hello from C++!" << std::endl;
+	
+	dummyFunction();
+	
+	PrimeCalculator calculator;
+	int N = 10000; // Large number of primes to compute
+	
+	double sum = calculator.sumOfSquareRoots(N);
+	std::cout << "Sum of square roots of the first " << N 
+	          << " prime numbers is " << sum << std::endl;
+	
+	std::cout << "Goodbye from C++!" << std::endl;
+
+	return 0;
+}