This repository contains code examples accompanying the paper "Lost in translation: Exposing hidden compiler optimization opportunities".
The files are organized according to the architecture and the MiBench benchmark discussed in the paper:
-
results/i5
: Intel i5-6300ubenchmark-8
:consumer-jpeg-c-src-jchuff-9-1
-
results/A53
: Cortex-A53 as implemented in Broadcom BCM2837 used in Raspberry Pi model 3B+benchmark-7
:consumer-jpeg-c-src-jcdctmgr-13-1
benchmark-9
:consumer-jpeg-c-src-jfdctint-2-1
benchmark-10
:consumer-lame-src-fft-2-1
benchmark-17
:consumer-lame-src-takehiro-5-1
benchmark-18
:consumer-mad-src-layer3-5-1
benchmark-31
:office-rsynth-src-nsynth-5-1
benchmark-36
:security-pgp-e-src-mpilib-4-1
benchmark-42
:telecomm-gsm-src-short-term-2-1
Each per-benchmark directory contains IR, raw object and object dump files corresponding
to the transformations discussed in the paper and to the standard -O3
code generation path.
For each benchmark the files are named after the transformation pass after which they were generated,
the prefix standard-O3
standing for the unmodified -O3
optimization flow.
File extensions:
.ll
: LLVM IR file.o
: raw object file.o.dump-Dhr
,.o.dump-dhr
: human-readable dump of the object file, respectively with/without data disassembly.