Avoid redundant uniform uploads (#2090)

* Avoid redundant uniform uploads

* Fix placement of GL_CHECK()s and import of glUniform4f.

* Fix typo

* Migrate GL uniform cache to use tinystl to conform to BGFX data structures

* Address review

* Address review
This commit is contained in:
juj 2020-04-08 09:08:09 +03:00 committed by GitHub
parent 66625bcf22
commit d9d9865ad9
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 178 additions and 19 deletions

View File

@ -205,6 +205,7 @@ typedef void (GL_APIENTRYP PFNGLUNIFORM1IVPROC) (GLint location, GLsiz
typedef void (GL_APIENTRYP PFNGLUNIFORM2FVPROC) (GLint location, GLsizei count, const GLfloat *value);
typedef void (GL_APIENTRYP PFNGLUNIFORM3FVPROC) (GLint location, GLsizei count, const GLfloat *value);
typedef void (GL_APIENTRYP PFNGLUNIFORM4FVPROC) (GLint location, GLsizei count, const GLfloat *value);
typedef void (GL_APIENTRYP PFNGLUNIFORM4FPROC) (GLint location, GLfloat x, GLfloat y, GLfloat z, GLfloat w);
typedef void (GL_APIENTRYP PFNGLUNIFORMMATRIX3FVPROC) (GLint location, GLsizei count, GLboolean transpose, const GLfloat *value);
typedef void (GL_APIENTRYP PFNGLUNIFORMMATRIX4FVPROC) (GLint location, GLsizei count, GLboolean transpose, const GLfloat *value);
typedef void (GL_APIENTRYP PFNGLUSEPROGRAMPROC) (GLuint program);
@ -402,6 +403,7 @@ GL_IMPORT______(false, PFNGLUNIFORM1FVPROC, glUniform1fv)
GL_IMPORT______(false, PFNGLUNIFORM2FVPROC, glUniform2fv);
GL_IMPORT______(false, PFNGLUNIFORM3FVPROC, glUniform3fv);
GL_IMPORT______(false, PFNGLUNIFORM4FVPROC, glUniform4fv);
GL_IMPORT______(false, PFNGLUNIFORM4FPROC, glUniform4f);
GL_IMPORT______(false, PFNGLUNIFORMMATRIX3FVPROC, glUniformMatrix3fv);
GL_IMPORT______(false, PFNGLUNIFORMMATRIX4FVPROC, glUniformMatrix4fv);
GL_IMPORT______(false, PFNGLUSEPROGRAMPROC, glUseProgram);

View File

@ -2735,7 +2735,7 @@ namespace bgfx { namespace gl
}
m_vaoSupport = !BX_ENABLED(BX_PLATFORM_EMSCRIPTEN)
&& (!!(BGFX_CONFIG_RENDERER_OPENGLES >= 30)
&& (!!(BGFX_CONFIG_RENDERER_OPENGLES >= 30)
|| s_extension[Extension::ARB_vertex_array_object].m_supported
|| s_extension[Extension::OES_vertex_array_object].m_supported
);
@ -3442,17 +3442,17 @@ namespace bgfx { namespace gl
GL_CHECK(glColorMask(GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE) );
ProgramGL& program = m_program[_blitter.m_program.idx];
GL_CHECK(glUseProgram(program.m_id) );
GL_CHECK(glUniform1i(program.m_sampler[0].loc, 0) );
GlUseProgram(program.m_id);
GlUniform1i(program.m_sampler[0].loc, 0);
float proj[16];
bx::mtxOrtho(proj, 0.0f, (float)width, (float)height, 0.0f, 0.0f, 1000.0f, 0.0f, g_caps.homogeneousDepth);
GL_CHECK(glUniformMatrix4fv(program.m_predefined[0].m_loc
GlUniformMatrix4fv(program.m_predefined[0].m_loc
, 1
, GL_FALSE
, proj
) );
);
GL_CHECK(glActiveTexture(GL_TEXTURE0) );
GL_CHECK(glBindTexture(GL_TEXTURE_2D, m_textures[_blitter.m_texture.idx].m_id) );
@ -3553,19 +3553,19 @@ namespace bgfx { namespace gl
void setShaderUniform4f(uint8_t /*_flags*/, uint32_t _regIndex, const void* _val, uint32_t _numRegs)
{
GL_CHECK(glUniform4fv(_regIndex
GlUniform4fv(_regIndex
, _numRegs
, (const GLfloat*)_val
) );
);
}
void setShaderUniform4x4f(uint8_t /*_flags*/, uint32_t _regIndex, const void* _val, uint32_t _numRegs)
{
GL_CHECK(glUniformMatrix4fv(_regIndex
GlUniformMatrix4fv(_regIndex
, _numRegs
, GL_FALSE
, (const GLfloat*)_val
) );
);
}
uint32_t setFrameBuffer(FrameBufferHandle _fbh, uint32_t _height, uint16_t _discard = BGFX_CLEAR_NONE, bool _msaa = true)
@ -4029,7 +4029,7 @@ namespace bgfx { namespace gl
case UniformType::_uniform: \
{ \
_type* value = (_type*)data; \
GL_CHECK(glUniform##_glsuffix(loc, num, value) ); \
GlUniform##_glsuffix(loc, num, value); \
} \
break;
@ -4037,7 +4037,7 @@ namespace bgfx { namespace gl
case UniformType::_uniform: \
{ \
_type* value = (_type*)data; \
GL_CHECK(glUniform##_glsuffix(loc, num, GL_FALSE, value) ); \
GlUniform##_glsuffix(loc, num, GL_FALSE, value); \
} \
break;
@ -4048,12 +4048,12 @@ namespace bgfx { namespace gl
// since they need to marshal an array over from Wasm to JS, so optimize the case when there is exactly one
// uniform to upload.
case UniformType::Sampler:
if (num > 1) glUniform1iv(loc, num, (int*)data);
else glUniform1i(loc, *(int*)data);
if (num > 1) GlUniform1iv(loc, num, (int*)data);
else GlUniform1i(loc, *(int*)data);
break;
case UniformType::Vec4:
if (num > 1) glUniform4fv(loc, num, (float*)data);
else glUniform4f(loc, ((float*)data)[0], ((float*)data)[1], ((float*)data)[2], ((float*)data)[3]);
if (num > 1) GlUniform4fv(loc, num, (float*)data);
else GlUniform4f(loc, ((float*)data)[0], ((float*)data)[1], ((float*)data)[2], ((float*)data)[3]);
break;
#else
CASE_IMPLEMENT_UNIFORM(Sampler, 1iv, I, int);
@ -4178,7 +4178,7 @@ namespace bgfx { namespace gl
GL_CHECK(glBindBuffer(GL_ARRAY_BUFFER, vb.m_id) );
ProgramGL& program = m_program[_clearQuad.m_program[numMrt-1].idx];
GL_CHECK(glUseProgram(program.m_id) );
GlUseProgram(program.m_id);
program.bindAttributesBegin();
program.bindAttributes(layout, 0);
program.bindAttributesEnd();
@ -4237,6 +4237,95 @@ namespace bgfx { namespace gl
}
}
void GlUseProgram(GLuint program)
{
m_uniformStateCache.saveCurrentProgram(program);
GL_CHECK(glUseProgram(program) );
}
// Cache uniform uploads to avoid redundant uploading of state that is
// already set to a shader program
void GlUniform1i(uint32_t loc, int value)
{
if (m_uniformStateCache.updateUniformCache(loc, value))
{
GL_CHECK(glUniform1i(loc, value) );
}
}
void GlUniform1iv(uint32_t loc, int num, const int *data)
{
bool changed = false;
for(int i = 0; i < num; ++i)
{
if (m_uniformStateCache.updateUniformCache(loc+i, data[i]))
{
changed = true;
}
}
if (changed)
{
GL_CHECK(glUniform1iv(loc, num, data) );
}
}
void GlUniform4f(uint32_t loc, float x, float y, float z, float w)
{
UniformStateCache::f4 f; f.val[0] = x; f.val[1] = y; f.val[2] = z; f.val[3] = w;
if (m_uniformStateCache.updateUniformCache(loc, f))
{
GL_CHECK(glUniform4f(loc, x, y, z, w));
}
}
void GlUniform4fv(uint32_t loc, int num, const float *data)
{
bool changed = false;
for(int i = 0; i < num; ++i)
{
if (m_uniformStateCache.updateUniformCache(loc+i, *(const UniformStateCache::f4*)&data[4*i]))
{
changed = true;
}
}
if (changed)
{
GL_CHECK(glUniform4fv(loc, num, data) );
}
}
void GlUniformMatrix3fv(uint32_t loc, int num, GLboolean transpose, const float *data)
{
bool changed = false;
for(int i = 0; i < num; ++i)
{
if (m_uniformStateCache.updateUniformCache(loc+i, *(const UniformStateCache::f3x3*)&data[9*i]))
{
changed = true;
}
}
if (changed)
{
GL_CHECK(glUniformMatrix3fv(loc, num, transpose, data) );
}
}
void GlUniformMatrix4fv(uint32_t loc, int num, GLboolean transpose, const float *data)
{
bool changed = false;
for(int i = 0; i < num; ++i)
{
if (m_uniformStateCache.updateUniformCache(loc+i, *(const UniformStateCache::f4x4*)&data[16*i]))
{
changed = true;
}
}
if (changed)
{
GL_CHECK(glUniformMatrix4fv(loc, num, transpose, data) );
}
}
void* m_renderdocdll;
uint16_t m_numWindows;
@ -4256,6 +4345,7 @@ namespace bgfx { namespace gl
OcclusionQueryGL m_occlusionQuery;
SamplerStateCache m_samplerStateCache;
UniformStateCache m_uniformStateCache;
TextVideoMem m_textVideoMem;
bool m_rtMsaa;
@ -4664,7 +4754,7 @@ namespace bgfx { namespace gl
if (0 != m_id)
{
GL_CHECK(glUseProgram(0) );
s_renderGL->GlUseProgram(0);
GL_CHECK(glDeleteProgram(m_id) );
m_id = 0;
}
@ -7155,7 +7245,7 @@ namespace bgfx { namespace gl
const RenderCompute& compute = renderItem.compute;
ProgramGL& program = m_program[key.m_program.idx];
GL_CHECK(glUseProgram(program.m_id) );
GlUseProgram(program.m_id);
GLbitfield barrier = 0;
for (uint32_t ii = 0; ii < maxComputeBindings; ++ii)
@ -7653,7 +7743,7 @@ namespace bgfx { namespace gl
// Skip rendering if program index is valid, but program is invalid.
currentProgram = 0 == id ? ProgramHandle{kInvalidHandle} : currentProgram;
GL_CHECK(glUseProgram(id) );
GlUseProgram(id);
programChanged =
constantsChanged =
bindAttribs = true;

View File

@ -32,6 +32,10 @@
|| BX_PLATFORM_WINDOWS \
)
// Keep a state cache of GL uniform values to avoid redundant uploads
// on the following platforms.
#define BGFX_GL_CONFIG_UNIFORM_CACHE BX_PLATFORM_EMSCRIPTEN
#define BGFX_GL_PROFILER_BEGIN(_view, _abgr) \
BX_MACRO_BLOCK_BEGIN \
GL_CHECK(glPushDebugGroup(GL_DEBUG_SOURCE_APPLICATION, 0, -1, s_viewName[view]) ); \
@ -1222,6 +1226,69 @@ namespace bgfx { namespace gl
#define GL_IMPORT(_optional, _proto, _func, _import) extern _proto _func
#include "glimports.h"
class UniformStateCache
{
public:
struct f4 { float val[4]; bool operator ==(const f4 &rhs) { const uint64_t *a = (const uint64_t *)this; const uint64_t *b = (const uint64_t *)&rhs; return a[0] == b[0] && a[1] == b[1]; }};
struct f3x3 { float val[9]; bool operator ==(const f3x3 &rhs) { const uint64_t *a = (const uint64_t *)this; const uint64_t *b = (const uint64_t *)&rhs; return a[0] == b[0] && a[1] == b[1] && a[2] == b[2] && a[3] == b[3] && ((const uint32_t*)a)[8] == ((const uint32_t*)b)[8]; }};
struct f4x4 { float val[16]; bool operator ==(const f4x4 &rhs) { const uint64_t *a = (const uint64_t *)this; const uint64_t *b = (const uint64_t *)&rhs; return a[0] == b[0] && a[1] == b[1] && a[2] == b[2] && a[3] == b[3] && a[4] == b[4] && a[5] == b[5] && a[6] == b[6] && a[7] == b[7]; }};
UniformStateCache()
:currentProgram(0)
{}
// Inserts the new value into the uniform cache, and returns true
// if the old value was different than the new one.
template<typename T>
bool updateUniformCache(uint32_t loc, const T &value)
{
#if BGFX_GL_CONFIG_UNIFORM_CACHE
// Uniform state cache for various types.
stl::unordered_map<uint64_t, T> &uniformCacheMap = getUniformCache<T>();
uint64_t key = ((uint64_t)currentProgram << 32) | loc;
auto iter = uniformCacheMap.find(key);
// Not found in the cache? Add it.
if (iter == uniformCacheMap.end())
{
uniformCacheMap[key] = value;
return true;
}
// Value in the cache was the same as new state? Skip reuploading this state.
if (iter->second == value)
{
return false;
}
iter->second = value;
#endif
return true;
}
void saveCurrentProgram(GLuint program)
{
#if BGFX_GL_CONFIG_UNIFORM_CACHE
currentProgram = program;
#endif
}
private:
GLuint currentProgram;
stl::unordered_map<uint64_t, int> uniformiCacheMap;
stl::unordered_map<uint64_t, f4> uniformf4CacheMap;
stl::unordered_map<uint64_t, f3x3> uniformf3x3CacheMap;
stl::unordered_map<uint64_t, f4x4> uniformf4x4CacheMap;
template<typename T>
stl::unordered_map<uint64_t, T> &getUniformCache();
};
template<> stl::unordered_map<uint64_t, int> &UniformStateCache::getUniformCache() { return uniformiCacheMap; }
template<> stl::unordered_map<uint64_t, UniformStateCache::f4> &UniformStateCache::getUniformCache() { return uniformf4CacheMap; }
template<> stl::unordered_map<uint64_t, UniformStateCache::f3x3> &UniformStateCache::getUniformCache() { return uniformf3x3CacheMap; }
template<> stl::unordered_map<uint64_t, UniformStateCache::f4x4> &UniformStateCache::getUniformCache() { return uniformf4x4CacheMap; }
class SamplerStateCache
{
public: