[SOLVED] Csed211 – 조승혁

Table of Contents
§ GCC Optimization
§ Manual Optimization
GCC
§ GCC provides some optimization options.
• https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html
• The optimization option tends to improve the performance / reduce the size of the binary.
§ We can enable the optimization options by using the –O{level} argument.
• -O, -O1, -O2, -O3, -O0, -Os, -Ofast, …
• level controls the optimization level.
• Higher optimization level increases the compilation time.
GCC Optimization Levels (1)
-O0 (Default)
• Reduces compilation time.
• Makes debugging produce the expected results.
-O1, -O
• Consume more time and memory for compilation.
• Reduces the code size and execution time.
GCC Optimization Levels (2)
-O2
• Performs nearly all supported optimizations.
• Does not involve space-speed tradeoff.
• Improve the performance with increased compilation time.
• …
-O3
• Turns on more optimization flags.
• Does not guarantee better performance.
GCC Optimization Levels (3)
-Os
• Removed the optimizations that increases code size from –O2.
• Tunes for code size rather than execution speed.
-Ofast
• Included the optimizations which disregard strict standards compliance to –O3.
-Og
• Optimizes debugging experience.
• Only for edit-compile-debug cycle.
• Recommended when debugging the program instead of –O0.
GCC Optimization Levels: Summary

GCC Optimization Flags
Optimization level includes optimization flags.
• https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html For example, -O1 uses the following flags:

GCC Optimization Flags – Example
-funroll-loops
• Unrolls loops whose number of iterations can be determined at compile time or upon entry to the loop.
• Turns on complete loop peeling, i.e., complete removal of loops with a small constant number of iterations.
• The code size becomes larger.

GCC Optimization – Example
gcc –O1 –fno-defer-pop –o test test.c
• Use optimization level –O1 and optimization flag –fno-defer-pop.
• The further flags override the flags from the optimization level.
• i.e., -O1 set –fdefer-pop as default.
Manual Optimization
We can optimize the program by ourself!
• Remove redundant variables, reduce the number of function calls, exploit locality, …
• Or write the ASM code manually!
ASM Code in C (1)
§ We use __asm() or __asm__() function for inline assembler.
§ It can be used anywhere inside the C / C++ code.
§ We use inline ASM for:
• Spot-optimizing speed-critical sections of code.
• Making direct hardware access for device drivers.
ASM Code in C (2)
Assembler template is the literal string with assembler instructions.
• We divide each line with the character . %n: register mapped to the n th argument.
%, {, |, } should be used with %.

ASM Code in C (3)

1: GCC Optimization (1)
We will test the following GCC optimization levels with the code loops.c.
• -O1, -Os, -O2.

1: GCC Optimization (2)
§ Use time command to see the compilation time.
• ex) time gcc loops.c –O<level>
§ Run the binary file generated with gcc and check the spent time.
• ex) ./a.out
2: ASM Code in C
§ Fill in the blank parts of asm.c.
• The program should results 105.
§ Hints
• mov v1 v2 -> copy the value of v1 to v2.
• add v1 v2 -> v2 = v1 + v2

Quiz