piwalker/dolphin
1
0
Fork 0
mirror of https://github.com/dolphin-emu/dolphin.git synced 2024-11-14 21:37:52 -07:00
Code Issues Actions Packages Projects Releases Wiki Activity

VideoBackends:Metal: FBFetch for Intel GPUs

Browse Source
This commit is contained in:
TellowKrinkle 2022-06-12 00:20:00 -05:00
parent a5ef9dfd53
commit 61705b05da
1 changed files with 88 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

88
Source/Core/VideoBackends/Metal/MTLUtil.mm
View File

@ -80,6 +80,89 @@ void Metal::Util::PopulateBackendInfoAdapters(VideoConfig* config,
}
}
static bool DetectIntelGPUFBFetch(id<MTLDevice> dev)
{
// Even though it's nowhere in the feature set tables, some Intel GPUs support fbfetch!
// Annoyingly, the Haswell compiler successfully makes a pipeline but actually miscompiles it and
// doesn't insert any fbfetch instructions.
// The Broadwell compiler inserts the Skylake fbfetch instruction,
// but Broadwell doesn't support that. It seems to make the shader not do anything.
// So we actually have to test the thing
static constexpr const char* shader = R"(
vertex float4 fs_triangle(uint vid [[vertex_id]]) {
return float4(vid & 1 ? 3 : -1, vid & 2 ? 3 : -1, 0, 1);
}
fragment float4 fbfetch_test(float4 in [[color(0), raster_order_group(0)]]) {
return in * 2;
}
)";
auto lib = MRCTransfer([dev newLibraryWithSource:[NSString stringWithUTF8String:shader]
options:nil
error:nil]);
if (!lib)
return false;
auto pdesc = MRCTransfer([MTLRenderPipelineDescriptor new]);
[pdesc setVertexFunction:MRCTransfer([lib newFunctionWithName:@"fs_triangle"])];
[pdesc setFragmentFunction:MRCTransfer([lib newFunctionWithName:@"fbfetch_test"])];
[[pdesc colorAttachments][0] setPixelFormat:MTLPixelFormatRGBA8Unorm];
auto pipe = MRCTransfer([dev newRenderPipelineStateWithDescriptor:pdesc error:nil]);
if (!pipe)
return false;
auto buf = MRCTransfer([dev newBufferWithLength:4 options:MTLResourceStorageModeShared]);
auto tdesc = [MTLTextureDescriptor texture2DDescriptorWithPixelFormat:MTLPixelFormatRGBA8Unorm
width:1
height:1
mipmapped:false];
[tdesc setUsage:MTLTextureUsageRenderTarget];
auto tex = MRCTransfer([dev newTextureWithDescriptor:tdesc]);
auto q = MRCTransfer([dev newCommandQueue]);
u32 px = 0x11223344;
memcpy([buf contents], &px, 4);
id<MTLCommandBuffer> cmdbuf = [q commandBuffer];
id<MTLBlitCommandEncoder> upload_encoder = [cmdbuf blitCommandEncoder];
[upload_encoder copyFromBuffer:buf
sourceOffset:0
sourceBytesPerRow:4
sourceBytesPerImage:4
sourceSize:MTLSizeMake(1, 1, 1)
toTexture:tex
destinationSlice:0
destinationLevel:0
destinationOrigin:MTLOriginMake(0, 0, 0)];
[upload_encoder endEncoding];
auto rpdesc = MRCTransfer([MTLRenderPassDescriptor new]);
[[rpdesc colorAttachments][0] setTexture:tex];
[[rpdesc colorAttachments][0] setLoadAction:MTLLoadActionLoad];
[[rpdesc colorAttachments][0] setStoreAction:MTLStoreActionStore];
id<MTLRenderCommandEncoder> renc = [cmdbuf renderCommandEncoderWithDescriptor:rpdesc];
[renc setRenderPipelineState:pipe];
[renc drawPrimitives:MTLPrimitiveTypeTriangle vertexStart:0 vertexCount:3];
[renc endEncoding];
id<MTLBlitCommandEncoder> download_encoder = [cmdbuf blitCommandEncoder];
[download_encoder copyFromTexture:tex
sourceSlice:0
sourceLevel:0
sourceOrigin:MTLOriginMake(0, 0, 0)
sourceSize:MTLSizeMake(1, 1, 1)
toBuffer:buf
destinationOffset:0
destinationBytesPerRow:4
destinationBytesPerImage:4];
[download_encoder endEncoding];
[cmdbuf commit];
[cmdbuf waitUntilCompleted];
u32 outpx;
memcpy(&outpx, [buf contents], 4);
// Proper fbfetch will double contents, Haswell will return black, and Broadwell will do nothing
if (outpx == 0x22446688)
return true; // Skylake+
else if (outpx == 0x11223344)
return false; // Broadwell
else
return false; // Haswell
}
void Metal::Util::PopulateBackendInfoFeatures(VideoConfig* config, id<MTLDevice> device)
{
// Initialize DriverDetails first so we can use it later
@ -135,6 +218,11 @@ void Metal::Util::PopulateBackendInfoFeatures(VideoConfig* config, id<MTLDevice>
}
if (DriverDetails::HasBug(DriverDetails::BUG_BROKEN_SUBGROUP_INVOCATION_ID))
g_features.subgroup_ops = false;
#if TARGET_OS_OSX
if (@available(macOS 11, *))
if (vendor == DriverDetails::VENDOR_INTEL)
config->backend_info.bSupportsFramebufferFetch |= DetectIntelGPUFBFetch(device);
#endif
if (DriverDetails::HasBug(DriverDetails::BUG_BROKEN_DYNAMIC_SAMPLER_INDEXING))
config->backend_info.bSupportsDynamicSamplerIndexing = false;
}