thirdparty/ppsspp
2
0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Shrink the GLRRenderCommand struct from 152 to 88 bytes

Browse source 
Turns out the VR work bloated it a bit, which can't be good.

Think it's fine to allocate these view matrices on the heap to get them
out of the way, there won't be that crazy many per frame usually.
This commit is contained in:
Henrik Rydgård 2023-05-10 00:20:43 +02:00
parent 16fbb31a65
commit cc16eb8321
4 changed files with 28 additions and 17 deletions
Download patch file Download diff file Expand all files Collapse all files

19
Common/GPU/OpenGL/GLQueueRunner.cpp
View file

@ -1038,29 +1038,30 @@ void GLQueueRunner::PerformRenderPass(const GLRStep &step, bool first, bool last
{
_dbg_assert_(curProgram);
if (IsMultiviewSupported()) {
int layout = GetStereoBufferIndex(c.uniformMatrix4.name);
int layout = GetStereoBufferIndex(c.uniformStereoMatrix4.name);
if (layout >= 0) {
int size = 2 * 16 * sizeof(float);
glBindBufferBase(GL_UNIFORM_BUFFER, layout, *c.uniformMatrix4.loc);
glBindBuffer(GL_UNIFORM_BUFFER, *c.uniformMatrix4.loc);
glBindBufferBase(GL_UNIFORM_BUFFER, layout, *c.uniformStereoMatrix4.loc);
glBindBuffer(GL_UNIFORM_BUFFER, *c.uniformStereoMatrix4.loc);
void *matrices = glMapBufferRange(GL_UNIFORM_BUFFER, 0, size, GL_MAP_WRITE_BIT | GL_MAP_INVALIDATE_BUFFER_BIT);
memcpy(matrices, c.uniformMatrix4.m, size);
memcpy(matrices, c.uniformStereoMatrix4.mData, size);
glUnmapBuffer(GL_UNIFORM_BUFFER);
glBindBuffer(GL_UNIFORM_BUFFER, 0);
}
} else {
int loc = c.uniformMatrix4.loc ? *c.uniformMatrix4.loc : -1;
if (c.uniformMatrix4.name) {
loc = curProgram->GetUniformLoc(c.uniformMatrix4.name);
int loc = c.uniformStereoMatrix4.loc ? *c.uniformStereoMatrix4.loc : -1;
if (c.uniformStereoMatrix4.name) {
loc = curProgram->GetUniformLoc(c.uniformStereoMatrix4.name);
}
if (loc >= 0) {
if (GetVRFBOIndex() == 0) {
glUniformMatrix4fv(loc, 1, false, c.uniformMatrix4.m);
glUniformMatrix4fv(loc, 1, false, c.uniformStereoMatrix4.mData);
} else {
glUniformMatrix4fv(loc, 1, false, &c.uniformMatrix4.m[16]);
glUniformMatrix4fv(loc, 1, false, c.uniformStereoMatrix4.mData + 16);
}
}
}
delete [] c.uniformStereoMatrix4.mData;
CHECK_GL_ERROR_IF_DEBUG();
break;
}

9
Common/GPU/OpenGL/GLQueueRunner.h
View file

@ -25,7 +25,7 @@ struct GLOffset2D {
int x, y;
};
enum class GLRAllocType {
enum class GLRAllocType : uint8_t {
NONE,
NEW,
ALIGNED,
@ -130,8 +130,13 @@ struct GLRRenderData {
struct {
const char *name; // if null, use loc
const GLint *loc;
float m[32];
float m[16];
} uniformMatrix4;
struct {
const char *name; // if null, use loc
const GLint *loc;
float *mData; // new'd, 32 entries
} uniformStereoMatrix4;
struct {
uint32_t clearColor;
float clearZ;

5
Common/GPU/OpenGL/GLRenderManager.cpp
View file

@ -41,6 +41,11 @@ GLRTexture::~GLRTexture() {
}
}
GLRenderManager::GLRenderManager() {
// size_t sz = sizeof(GLRRenderData);
// _dbg_assert_(sz == 88);
}
GLRenderManager::~GLRenderManager() {
_dbg_assert_(!run_);

12
Common/GPU/OpenGL/GLRenderManager.h
View file

@ -390,10 +390,9 @@ struct GLRRenderThreadTask {
// directly in the destructor.
class GLRenderManager {
public:
GLRenderManager() {}
GLRenderManager();
~GLRenderManager();
void SetInvalidationCallback(InvalidationCallback callback) {
invalidationCallback_ = callback;
}
@ -783,10 +782,11 @@ public:
_dbg_assert_(curProgram_);
#endif
GLRRenderData data{ GLRRenderCommand::UNIFORMSTEREOMATRIX };
data.uniformMatrix4.name = name;
data.uniformMatrix4.loc = loc;
memcpy(&data.uniformMatrix4.m[0], left, sizeof(float) * 16);
memcpy(&data.uniformMatrix4.m[16], right, sizeof(float) * 16);
data.uniformStereoMatrix4.name = name;
data.uniformStereoMatrix4.loc = loc;
data.uniformStereoMatrix4.mData = new float[32];
memcpy(&data.uniformStereoMatrix4.mData[0], left, sizeof(float) * 16);
memcpy(&data.uniformStereoMatrix4.mData[16], right, sizeof(float) * 16);
curRenderStep_->commands.push_back(data);
}