Wed, 03 Oct 2018 00:52:37 -0400render: first (untested!) shot at converting D3D11 renderer to new interfaces. SDL-ryan-batching-renderer
Ryan C. Gordon <icculus@icculus.org> [Wed, 03 Oct 2018 00:52:37 -0400] rev 12283
render: first (untested!) shot at converting D3D11 renderer to new interfaces.

Probably doesn't even compile yet.

Mon, 01 Oct 2018 22:53:45 -0400render: Make the GL backends cache and defer more state changes. SDL-ryan-batching-renderer
Ryan C. Gordon <icculus@icculus.org> [Mon, 01 Oct 2018 22:53:45 -0400] rev 12282
render: Make the GL backends cache and defer more state changes.

Mon, 01 Oct 2018 13:41:15 -0400render: D3D9 doesn't need to check for stream offset support anymore. SDL-ryan-batching-renderer
Ryan C. Gordon <icculus@icculus.org> [Mon, 01 Oct 2018 13:41:15 -0400] rev 12281
render: D3D9 doesn't need to check for stream offset support anymore.

We don't use offsets at all now. Too slow.

Mon, 01 Oct 2018 11:32:08 -0400render: Set the D3D9 stream source once and choose offsets during draw calls. SDL-ryan-batching-renderer
Ryan C. Gordon <icculus@icculus.org> [Mon, 01 Oct 2018 11:32:08 -0400] rev 12280
render: Set the D3D9 stream source once and choose offsets during draw calls.

This is _much_ faster than setting the offsets with SetStreamSource!

Mon, 01 Oct 2018 03:02:54 -0400render: Patched to compile. SDL-ryan-batching-renderer
Ryan C. Gordon <icculus@icculus.org> [Mon, 01 Oct 2018 03:02:54 -0400] rev 12279
render: Patched to compile.

Mon, 01 Oct 2018 01:23:02 -0400render: first shot at moving Direct3D 9 backend to new interface. Untested! SDL-ryan-batching-renderer
Ryan C. Gordon <icculus@icculus.org> [Mon, 01 Oct 2018 01:23:02 -0400] rev 12278
render: first shot at moving Direct3D 9 backend to new interface. Untested!

Tue, 02 Oct 2018 20:51:33 -0700Make sure we don't read and write to HIDAPI at the same time, it's not thread-safe on Windows
Sam Lantinga <slouken@libsdl.org> [Tue, 02 Oct 2018 20:51:33 -0700] rev 12277
Make sure we don't read and write to HIDAPI at the same time, it's not thread-safe on Windows

Tue, 02 Oct 2018 13:17:31 -0700The Amlogic X96 is a set-top box
Sam Lantinga <slouken@libsdl.org> [Tue, 02 Oct 2018 13:17:31 -0700] rev 12276
The Amlogic X96 is a set-top box

Mon, 01 Oct 2018 14:52:28 -0700Fixed UnsatisfiedLinkError when initializing the HIDDeviceManager in some cases
Sam Lantinga <slouken@libsdl.org> [Mon, 01 Oct 2018 14:52:28 -0700] rev 12275
Fixed UnsatisfiedLinkError when initializing the HIDDeviceManager in some cases

Mon, 01 Oct 2018 14:43:03 -0700Working on bug 3921 - Add some Fastpath to BlitNtoNKey and BlitNtoNKeyCopyAlpha
Sam Lantinga <slouken@libsdl.org> [Mon, 01 Oct 2018 14:43:03 -0700] rev 12274
Working on bug 3921 - Add some Fastpath to BlitNtoNKey and BlitNtoNKeyCopyAlpha

Sylvain

I did various benches. with clang 6.0.0 on linux, and ndk-r16b on android (NDK_TOOLCHAIN_VERSION=clang).

- still see a x10 speed factor.
- with duff_loops, it does not use vectorisation (but doesn't seem to be a problem).

on linux my patch is already at full speed on -O2, whereas the duff_loops need -O3 (200 ms at -03, and 300ms at -02).

I realized that on Android, I had a slight variation which fits best.
both on linux with -O2 and -O3, and on android with 02/03 and armeabi-v7a/arm64.

Here's the patch.