Topics

Embench data for LLVM 11.0.0 release


Jeremy Bennett
 

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi all,

(cross posted to Code Size and Code Speed SIGs)

Clang/LLVM 11.0.0 was released a couple of weeks ago. We have evaluated
its performance using Embench, with RISC-V RV32IMC (specifically the
PULP RI5CY core) and Arm Cortex M4 (FPU disabled). We provide
comparative data against GCC 10.2 and LLVM 10.0.1.

The current Embench benchmark suite targets the smallest microcontroller
class processors. For full details see embench.org.

Summary
=======

LLVM 11.0.0 compiled code for RISC-V is 7% slower and 7% larger than
for Arm

LLVM 11.0.0 compiled code size for RISC-V has improved 7% compared to
LLVM 10.0.1.

LLVM 11.0.0 compiled code is 4% quicker than GCC 10.2, but also 6%
larger than GCC 10.2.

Details
=======

For speed higher is better
For size lower is better

LLVM 11.0.0 RV32IMC v Arm Cortex M4
- -----------------------------------

Speed Size
----- ----
Arm 1.04 1.02
RV32 0.97 1.09

LLVM 11.0.0 v LLVM 10.0.1 RV32IMC
- ---------------------------------

Speed Size
----- ----
LLVM 11.0.0 0.97 1.09
LLVM 10.0.1 0.98 1.16

LLVM 11.0.0 v GCC 10.2 RV32IMC
- ------------------------------

Speed Size
----- ----
LLVM 11.0.0 0.97 1.09
GCC 10.2 0.93 1.03

Best wishes,


Jeremy
- --
Cell: +44 (7970) 676050
SkypeID: jeremybennett
Twitter: @jeremypbennett
Email: jeremy.bennett@embecosm.com
Web: www.embecosm.com
PGP key: 1024D/BEF58172FB4754E1 2009-03-20
-----BEGIN PGP SIGNATURE-----

iF0EARECAB0WIQRASGDWqmhRZUfAaPW+9YFy+0dU4QUCX6Qp+gAKCRC+9YFy+0dU
4UVFAJ4x4fmDp2rkkEkQPnehwbtPfNY5VQCdH+IlY0l9diR+wHS1+xXXtz2MsNU=
=Qlz3
-----END PGP SIGNATURE-----