Permalink
Browse files

OGL/Streambuffer: Use coherent mapping by default.

Coherent mappings have a lower overhead and less GL codes.
So enables coherent mapping by default for all drivers.
Both Qualcomm and ARM performs very bad with explicit flushing, so this change helps them as well.

AFAIK there was one GPU generation which was slower on coherent mapping: nvidia tesla
So Geforce 200 and 300 series should be tested with this PR before merging.
As this was last tested many years ago, this issue might have been fixed as well.
Those GPUs are close to 10 years old and not supported any more by nvidia.
  • Loading branch information...
degasus committed May 23, 2018
1 parent dfb1dba commit fdfd8b19d62dbd254046aaabdc4c7a9ee167467a
@@ -216,7 +216,7 @@ class MapAndSync : public StreamBuffer
class BufferStorage : public StreamBuffer
{
public:
BufferStorage(u32 type, u32 size, bool _coherent = false)
BufferStorage(u32 type, u32 size, bool _coherent = true)
: StreamBuffer(type, size), coherent(_coherent)
{
CreateFences();
@@ -368,13 +368,12 @@ std::unique_ptr<StreamBuffer> StreamBuffer::Create(u32 type, u32 size)
return std::make_unique<PinnedMemory>(type, size);
// buffer storage works well in most situations
bool coherent = DriverDetails::HasBug(DriverDetails::BUG_BROKEN_EXPLICIT_FLUSH);
if (g_ogl_config.bSupportsGLBufferStorage &&
!(DriverDetails::HasBug(DriverDetails::BUG_BROKEN_BUFFER_STORAGE) &&
type == GL_ARRAY_BUFFER) &&
!(DriverDetails::HasBug(DriverDetails::BUG_INTEL_BROKEN_BUFFER_STORAGE) &&
type == GL_ELEMENT_ARRAY_BUFFER))
return std::make_unique<BufferStorage>(type, size, coherent);
return std::make_unique<BufferStorage>(type, size);
// don't fall back to MapAnd* for Nvidia drivers
if (DriverDetails::HasBug(DriverDetails::BUG_BROKEN_UNSYNC_MAPPING))
@@ -52,8 +52,6 @@ static BugInfo m_known_bugs[] = {
BUG_BROKEN_BUFFER_STREAM, -1.0, -1.0, true},
{API_OPENGL, OS_ALL, VENDOR_QUALCOMM, DRIVER_QUALCOMM, Family::UNKNOWN,
BUG_BROKEN_NEGATED_BOOLEAN, -1.0, -1.0, true},
{API_OPENGL, OS_ALL, VENDOR_QUALCOMM, DRIVER_QUALCOMM, Family::UNKNOWN,
BUG_BROKEN_EXPLICIT_FLUSH, -1.0, -1.0, true},
{API_OPENGL, OS_ALL, VENDOR_ARM, DRIVER_ARM, Family::UNKNOWN, BUG_BROKEN_BUFFER_STREAM, -1.0,
-1.0, true},
{API_OPENGL, OS_ALL, VENDOR_ARM, DRIVER_ARM, Family::UNKNOWN, BUG_BROKEN_VSYNC, -1.0, -1.0,
@@ -200,15 +200,6 @@ enum Bug
// Causes misrenderings on a large amount of things that draw lines.
BUG_BROKEN_GEOMETRY_SHADERS,
// Bug: Explicit flush is very slow on Qualcomm
// Started Version: -1
// Ended Version: -1
// Our ARB_buffer_storage code uses explicit flush to avoid coherent mapping.
// Qualcomm seems to have lots of overhead on explicit flushing, but the coherent mapping path is
// fine.
// So let's use coherent mapping there.
BUG_BROKEN_EXPLICIT_FLUSH,
// Bug: glGetBufferSubData for bounding box reads is slow on AMD drivers
// Started Version: -1
// Ended Version: -1

0 comments on commit fdfd8b1

Please sign in to comment.