Thanks for doing this, this is good information.
On 7/31/2013 2:24 PM, Rainer Schuetze wrote:
I have just tried yesterdays dmd to build Visual D (it builds some libraries and
contains a few short non-compiling tasks in between):
Debug build dmd_dmc: 23 sec, std new 43 sec
Debug build dmd_msc: 19 sec, std new 20 sec
That makes it clear that the dmc malloc() was the dominator, not code gen.
"std new" is the version without the "block allocator".
Release build dmd_dmc: 3 min 30, std new 5 min 25
Release build dmd_msc: 1 min 32, std new 1 min 40
The release builds use "-release -O -inline" and need a bit more than 1 GB
memory for two of the libraries (I still had to patch dmd_dmc to be
large-address-aware).
This shows that removing most of the allocations was a good optimization for the
dmc-Runtime, but does not have a large, but still notable impact on a faster
heap implementation (the VS runtime usually maps directly to the Windows API for
non-Debug builds). I suspect the backend and the optimizer do not use "new" a
lot, but plain "malloc" calls, so they still suffer from the slow runtime.
Actually, dmc still should give a better showing. All the optimizations I've put
into dmd also went into dmc, and do result in significantly better code speed.
For example, the hash modulus optimization has a significant impact, but I
haven't released that dmc yet.
Optimized builds have an entirely different profile than debug builds, and I
haven't investigated that.