Expanded vec128 and VectorSub #574
Conversation
Added double precision (f64) values to the vec128 structure Added cases for all formats of data for vector subtraction in the Value::VectorSub function NOTE: Unsure what the saturate function is for however maintained original functionality
|
I reckon ben will ask you to run |
|
Neat - this is cool. Like @Razzile said, just format your code according to this guide and we'll pull it! |
| assert_true(this->type == VEC128_TYPE && other->type == VEC128_TYPE); | ||
| switch (type) { | ||
| case FLOAT32_TYPE: | ||
| if (saturate) assert_always(); |
There was a problem hiding this comment.
Surround if statements in curly brackets please. Ditto for everything below.
|
This only fixes the wonky physics in sonic games, however the death issues and inverted model glitches are still apparent. Likely caused by #514. I think that commit was responsible for invisible level in sonic unleashed demo, and model depth issues in TLoS:DotD and various other games.
|
|
I've actually got some uncommitted code locally that implements just about everything that was unimplemented here. |
|
*Insert flattery statement, because your amazing! |
|
I've updated the code with Google formatting, for some reason PS. I'm going to write all of the SSE and AVX acceleration for this shortly, @DrChat should I wait until you've put up your latest file? |
| * Copyright 2013 Ben Vanik. All rights reserved. * | ||
| * Released under the BSD license - see LICENSE in the root for more details. * | ||
| ****************************************************************************** | ||
| */ |
There was a problem hiding this comment.
The space needs to come back - just undo this change.
|
To be honest I'd recommend not writing any AVX/SSE/etc acceleration for this code - it already runs fast enough. Also - If you already know how, could you squash the 2 formatting commits back to one? |
| @@ -90,7 +90,8 @@ typedef struct alignas(16) vec128_s { | |||
| uint32_t uz; | |||
| uint32_t uw; | |||
| }; | |||
| float f32[4]; | |||
| float f32[4]; | |||
| double f64[2]; | |||
There was a problem hiding this comment.
I thought the formatting for this file was okay?
There was a problem hiding this comment.
Should be spaces instead of tabs
There was a problem hiding this comment.
Fixed. Also squashed.
There was a problem hiding this comment.
On the SSE and AVX stuff. I noticed Halo Reach and Halo 3 use this code a lot and when I debugged it appeared to be positional co-ordinates, thats why I was eager to smash out some faster code for it
|
@ScorchEmber256 The code was originally written against missing Halo 3 and Halo Reach functionality with the processing of entity positions, I can only assume by the code that I have written that I've just added missing functionality I'm not entirely sure how it would affect other games. |
Added formatting to code
|
@HaydnTrigg This code's part of the recompiler - this is ran once during compilation to propagate constants through the generated code. The speed on this code has no bearing on the speed of the generated code. |
|
Merged in 7fe4ffc. |
Added double precision (f64) values to the vec128 structure
Added cases for all formats of data for vector subtraction in the
Value::VectorSub function
NOTE: Unsure what the saturate function is for however maintained
original functionality