GCC Base Performance Optimization Guide
Overview
The optimization of compiler base performance is crucial to improving the development efficiency, running performance, and maintainability of applications. It is important in both computer science and software development. Based on the general compilation optimization capability, GCC for openEuler enhances middle- and back-end performance optimization technologies, including instruction optimization, vectorization enhancement, prefetch enhancement, and data flow analysis enhancement.
Installation and Deployment
Software Requirements
OS: openEuler 25.03
Hardware Requirements
AArch64 architecture
Software Installation
Install GCC and related components as needed. The following uses GCC as an example:
yum install gccHow to Use
Optimization for CRC
Description
GCC identifies cyclic redundancy check (CRC) code and generates efficient hardware instructions.
How to Use
Add -floop-crc during compilation.
Note: -floop-crc must be used together with -O3 -march=armv8.1-a.
If-conversion Enhancement
Description
If-conversion optimization is enhanced by using more registers to reduce conflicts.
How to Use
This enhancement is part of Register Transfer Language (RTL) if-conversion optimization. Enable the enhancement by using the following options:
-fifcvt-allow-complicated-cmps
--param=ifcvt-allow-register-renaming=[0,1,2], where the numbers are used to control the optimization scope
Note: This optimization requires the -O2 optimization level and must be used together with --param=max-rtl-if-conversion-unpredictable-cost=48 and --param=max-rtl-if-conversion-predictable-cost=48.
Optimization for Multiplication
Description
Arm instructions are combined to convert low-order 32-bit multiplications into high-order 64-bit multiplication instructions.
How to Use
Use the -fuaddsub-overflow-match-all and -fif-conversion-gimple options.
Note: This optimization requires the -O3 or higher optimization level.
Optimization for CMLT Instruction Generation
Description
cmlt instructions are generated for some elementary arithmetic operations to reduce the number of instructions.
How to Use
Use the -mcmlt-arith option.
Note: This optimization requires the -O3 or higher optimization level.
Optimization for Vectorization
Description
Redundant instructions generated during vectorization are identified and simplified, and shorter arrays can be vectorized.
How to Use
Use --param=vect-alias-flexible-segment-len=1. The default value is 0.
Note: This optimization requires the -O3 or higher optimization level.
Optimization for min max and uzp1/uzp2 Instructions
Description
The min max and uzp1/uzp2 instructions are optimized to reduce the total instructions and improve performance.
How to Use
Use the -fconvert-minmax option to enable min max optimization. uzp1/uzp2 instruction optimization is enabled by default at a level higher than -O3.
Note: This optimization requires the -O3 or higher optimization level.
Optimization for LDP and STP
Description
Each LDP and STP instruction with poor performance is split into two LDR and STR instructions.
How to Use
Use the -fsplit-ldp-stp option. Use --param=param-ldp-dependency-search-range=[1,32] to control the search range. The default value is 16.
Note: This optimization requires the -O1 or higher optimization level.
Optimization for AES Instruction
Description
The AES software instruction sequences are identified and accelerated using hardware instructions.
How to Use
Use the -fcrypto-accel-aes option.
Note: This optimization requires the -O3 or higher optimization level.
Optimization for Indirect Calls
Description
Indirect calls in programs are identified, analyzed, and then optimized into direct calls.
How to Use
Use the -ficp -ficp-speculatively option.
Note: This optimization must be used together with -O2 -flto -flto-partition=one.
IPA-prefetch
Description
Indirect memory accesses in a loop are identified, and a prefetch instruction is inserted to reduce the delay.
How to Use
Use the -fipa-prefetch -fipa-ic option.
Note: This optimization must be used together with -O3 -flto.
-fipa-struct-reorg
Description
This option optimizes memory layout. The structure members are rearranged in memory to improve the cache hit rate.
How to Use
Add -O3 -flto -flto-partition=one -fipa-struct-reorg to the option.
Note: The -fipa-struct-reorg option can be enabled only when -O3 -flto -flto-partition=one is enabled globally.
-fipa-reorder-fields
Description
The memory space layout is optimized by arranging structure members from largest to smallest based on their size. This reduces padding caused by alignment boundaries, decreases overall memory usage, and improves the cache hit rate.
How to Use
Add -O3 -flto -flto-partition=one -fipa-reorder-fields to the option.
Note: The -fipa-reorder-fields option can be enabled only when -O3 -flto -flto-partition=one is enabled globally.
-ftree-slp-transpose-vectorize
Description
In the loop splitting phase, temporary arrays are introduced to partition the loop, which enhances data-flow analysis for loops that read continuous memory. In the superword-level parallelism (SLP) vectorization phase, SLP analysis is performed for transposing grouped_stores.
How to Use
Add -O3 -ftree-slp-transpose-vectorize to the option.
Note: The -ftree-slp-transpose-vectorize option can be enabled only when -O3 is enabled.
LLC-prefetch
Description
In main execution paths of programs, memory-reuse patterns in loops are analyzed to identify and rank top hot data. The prefetch instruction is introduced to allocate the data to last-level cache (LLC), reducing LLC misses.
How to Use
Use the -fllc-allocate option. The -O2 or higher optimization level is required.
Other related interfaces:
| Option | Default Value | Description |
|---|---|---|
| --param=mem-access-ratio=[0,100] | 20 | Ratio of the number of memory accesses in a loop to the number of instructions. |
| --param=mem-access-num=unsigned | 3 | Number of memory accesses in a loop. |
| --param=outer-loop-nums=[1,10] | 1 | Maximum number of outer loop layers that can be unrolled. |
| --param=filter-kernels=[0,1] | 1 | Indicates whether to perform path series filtering on loops. |
| --param=branch-prob-threshold=[50,100] | 80 | Probability threshold for a branch to be considered highly probable. |
| --param=prefetch-offset=[1,999999] | 1024 | Prefetch offset distance, where the value is a power of 2. |
| --param=issue-topn=unsigned | 1 | Number of prefetch instructions. |
| --param=force-issue=[0,1] | 0 | Indicates whether to perform forcible prefetch, that is, the static mode. |
| --param=llc-capacity-per-core=[0,999999] | 107 | Average LLC capacity allocated to each core in multi-branch prefetch mode. |
-fipa-struct-sfc
Description
This option is used to statically compress structure members to reduce the structure size and improve the cache hit rate.
How to Use
Add -O3 -flto -flto-partition=one -fipa-reorder-fields -fipa-struct-sfc to the option. You can use -fipa-struct-sfc-bitfield and -fipa-struct-sfc-shadow for further optimization.
Note: The -fipa-struct-sfc option can be enabled only when -O3 -flto -flto-partition=one is enabled globally and -fipa-reorder-fields or -fipa-struct-reorg>=2 is enabled.
-fipa-struct-dfc
Description
This option is used to dynamically compress structure members by cloning the program path and heuristically minimizing the structure size. At runtime, it improves the cache hit rate by checking execution paths and selecting the optimal one.
How to Use
Add -O3 -flto -flto-partition=one -fipa-reorder-fields -fipa-struct-dfc to the option. You can use -fipa-struct-dfc-bitfield and -fipa-struct-dfc-shadow for further optimization.
Note: The -fipa-struct-dfc option can be enabled only when -O3 -flto -flto-partition=one is enabled globally and -fipa-reorder-fields or -fipa-struct-reorg>=2 is enabled.
-fipa-alignment-propagation
Description
This option is used to analyze and propagate the address-alignment values for local variables, optimizing the bitwise AND operations.
How to Use
Add -O3 -fipa-alignment-propagation to the option.
Note: The -fipa-alignment-propagation option can be enabled only when -O3 is enabled.
-fipa-localize-array
Description
This option is used to convert the global pointer variables allocated by calloc to local variables.
How to Use
Add -O3 -fipa-localize-array to the option.
Note: The -fipa-localize-array option can be enabled only when -O3 is enabled.
-fipa-array-dse
Description
This option is used to analyze the transfer of arrays between functions and the usage of the arrays in the called functions, removing redundant array writes.
How to Use
Add -O3 -fipa-array-dse to the option.
Note: The -fipa-array-dse option can be enabled only when -O3 is enabled.
-ffind-with-sve
Description
This option is used to identify std::find function calls and attempt to optimize them using SVE instructions.
How to Use
Add -ffind-with-sve to the option.
-floop-sve-mode-opt
Description
By analyzing static code characteristics, special scenarios can be identified. When the conditions are met, additional optimization opportunities leveraging the SVE instruction set are introduced, resulting in improved performance.
How to Use
Add -O3 -floop-sve-mode-opt to the option.
Note: The -floop-sve-mode-opt option can be enabled only when -O3 is enabled and SVE is included in the -march setting.
Licensed under the MulanPSL2
J.G.W.A.B. No. 11030102011597