Re: [Emutos-devel] GEMBench on TT High

[Eero T. <oa...@he...>]

Hi,

On 03/28/2016 08:17 PM, Vincent Rivière wrote:
> On 27/03/2016 19:10, Roger Burrows wrote:
>> For most things, there is no significant difference between 68000 & 68030
>
> The only place where optimizations for a specific CPU would be useful would
> be VDI blits and text output. But those areas are already manually optimized
> in assembler... preventing any further optimization by the compiler!!

IMHO more interesting differences were to TOS v3.

I had earlier compared the performance between
EmuTOS and TOS v1.04 on high-res ST with Hatari,
for which Hatari should be accurate enough.

In that, the main differences were in EmuTOS being
faster with GEM dialogs & VDI enquire, and a bit
faster with VDI (small & normal) text.

EmuTOS was *much* slower in:
- GEM window test
- VDI graphics
- New dialogs, and
- justified text

In this order.

(And slightly slower in text effects.)


Looking at Roger's numbers against TOS v3, the gaps
are larger in TT-medium than high.  There are gaps
in same places, but also in some additional places.

TT specific performance problems are in:
- blitting
- VDI scroll
- ROM access

See attached screenshot + spreadsheet.

I guess VDI scroll depends on blitting, but
why ROM access would be slower with EmuTOS?


As to test-cases that are slow also on ST &
compared to older TOS, I think justified text
and VDI graphics in general aren't that important
for most apps.  GEM window and new dialog speed
sound something that should be looked into though.

Is there some way to run those individual test-cases
separately with GEM Bench so that they could be
profiled?


	- Eero