VideoBackends:Metal: Use unified memory path by default on all GPUs
This commit is contained in:
parent
6ee0248eab
commit
c48035908c
|
@ -211,12 +211,13 @@ void Metal::Util::PopulateBackendInfoFeatures(VideoConfig* config, id<MTLDevice>
|
|||
config->backend_info.AAModes.push_back(i);
|
||||
}
|
||||
|
||||
// The unified memory path (using shared buffers for everything) performs noticeably better with
|
||||
// bbox even on discrete GPUs (20fps vs 15fps in Super Paper Mario elevator), so default to that.
|
||||
// The separate buffer + manual upload path is left available for testing and comparison.
|
||||
if (char* env = getenv("MTL_UNIFIED_MEMORY"))
|
||||
g_features.unified_memory = env[0] == '1' || env[0] == 'y' || env[0] == 'Y';
|
||||
else if (@available(macOS 10.15, iOS 13.0, *))
|
||||
g_features.unified_memory = [device hasUnifiedMemory];
|
||||
else
|
||||
g_features.unified_memory = false;
|
||||
g_features.unified_memory = true;
|
||||
|
||||
g_features.subgroup_ops = false;
|
||||
if (@available(macOS 10.15, iOS 13, *))
|
||||
|
|
Loading…
Reference in New Issue