gl canvas renderer with buffer re-specification

build.bat

@@ -4,6 +4,6 @@ if not exist bin mkdir bin
 set gles_shaders=src\gles_canvas_shaders\gles_canvas_blit_vertex.glsl src\gles_canvas_shaders\gles_canvas_blit_fragment.glsl src\gles_canvas_shaders\gles_canvas_clear_counters.glsl src\gles_canvas_shaders\gles_canvas_tile.glsl src\gles_canvas_shaders\gles_canvas_sort.glsl src\gles_canvas_shaders\gles_canvas_draw.glsl
 call python scripts\embed_text.py %gles_shaders% --output src\gles_canvas_shaders.h
 
-set INCLUDES=/I src /I src/util /I src/platform /I ext /I ext/angle_headers
+set INCLUDES=/I src /I src/util /I src/platform /I ext
 cl /we4013 /Zi /Zc:preprocessor /DMG_IMPLEMENTS_BACKEND_GL /std:c11 %INCLUDES% /c /Fo:bin/milepost.obj src/milepost.c
 lib bin/milepost.obj /OUT:bin/milepost.lib

examples/perf_text/build.bat

@@ -1,4 +1,4 @@
 
-set INCLUDES=/I ..\..\src /I ..\..\src\util /I ..\..\src\platform /I ../../ext /I ../../ext/angle_headers
+set INCLUDES=/I ..\..\src /I ..\..\src\util /I ..\..\src\platform /I ../../ext
 
 cl /we4013 /Zi /Zc:preprocessor /DMG_IMPLEMENTS_BACKEND_GL /std:c11 %INCLUDES% main.c /link /LIBPATH:../../bin milepost.lib user32.lib opengl32.lib gdi32.lib /out:../../bin/perf_text.exe

src/gl_canvas.c

@@ -114,11 +114,12 @@ void mg_gl_canvas_update_vertex_layout(mg_gl_canvas_backend* backend)
 void mg_gl_send_buffers(mg_gl_canvas_backend* backend, int vertexCount, int indexCount)
 {
 	glBindBuffer(GL_SHADER_STORAGE_BUFFER, backend->vertexBuffer);
-	glBufferData(GL_SHADER_STORAGE_BUFFER, LAYOUT_VERTEX_SIZE*vertexCount, backend->vertexMapping, GL_DYNAMIC_DRAW);
+	glBufferData(GL_SHADER_STORAGE_BUFFER, LAYOUT_VERTEX_SIZE*vertexCount, backend->vertexMapping, GL_STREAM_DRAW);
 
 	glBindBuffer(GL_SHADER_STORAGE_BUFFER, backend->indexBuffer);
-	glBufferData(GL_SHADER_STORAGE_BUFFER, LAYOUT_INT_SIZE*indexCount, backend->indexMapping, GL_DYNAMIC_DRAW);
+	glBufferData(GL_SHADER_STORAGE_BUFFER, LAYOUT_INT_SIZE*indexCount, backend->indexMapping, GL_STREAM_DRAW);
 }
+
 void mg_gl_canvas_begin(mg_canvas_backend* interface)
 {
 	mg_gl_canvas_backend* backend = (mg_gl_canvas_backend*)interface;
@@ -306,11 +307,11 @@ mg_canvas_backend* mg_gl_canvas_create(mg_surface surface)
 
 		glGenBuffers(1, &backend->vertexBuffer);
 		glBindBuffer(GL_SHADER_STORAGE_BUFFER, backend->vertexBuffer);
-//		glBufferData(GL_SHADER_STORAGE_BUFFER, MG_GL_CANVAS_VERTEX_BUFFER_SIZE, 0, GL_DYNAMIC_DRAW);
+		glBufferData(GL_SHADER_STORAGE_BUFFER, MG_GL_CANVAS_VERTEX_BUFFER_SIZE, 0, GL_STREAM_DRAW);
 
 		glGenBuffers(1, &backend->indexBuffer);
 		glBindBuffer(GL_SHADER_STORAGE_BUFFER, backend->indexBuffer);
-//		glBufferData(GL_SHADER_STORAGE_BUFFER, MG_GL_CANVAS_INDEX_BUFFER_SIZE, 0, GL_DYNAMIC_DRAW);
+		glBufferData(GL_SHADER_STORAGE_BUFFER, MG_GL_CANVAS_INDEX_BUFFER_SIZE, 0, GL_STREAM_DRAW);
 
 		glGenBuffers(1, &backend->tileCounterBuffer);
 		glBindBuffer(GL_SHADER_STORAGE_BUFFER, backend->tileCounterBuffer);

src/win32_gl_loader.h

@@ -36,6 +36,7 @@
 	GL_PROC(GLGENBUFFERS, glGenBuffers) \
 	GL_PROC(GLBINDBUFFER, glBindBuffer) \
 	GL_PROC(GLBUFFERDATA, glBufferData) \
+	GL_PROC(GLBUFFERSUBDATA, glBufferSubData) \
 	GL_PROC(GLUNIFORMMATRIX4FV, glUniformMatrix4fv) \
 	GL_PROC(GLVERTEXATTRIBPOINTER, glVertexAttribPointer) \
 	GL_PROC(GLENABLEVERTEXATTRIBARRAY, glEnableVertexAttribArray) \
@@ -47,6 +48,9 @@
 	GL_PROC(GLACTIVETEXTURE, glActiveTexture) \
 	GL_PROC(GLUNIFORM1I, glUniform1i) \
 	GL_PROC(GLTEXSTORAGE2D, glTexStorage2D) \
+	GL_PROC(GLMAPBUFFERRANGE, glMapBufferRange) \
+	GL_PROC(GLUNMAPBUFFER, glUnmapBuffer)
+
 
 #ifdef WIN32_GL_LOADER_API
 	//NOTE: pointer declarations

todo.txt

@@ -1,4 +1,10 @@
 
+WIP
+---
+[ ] Split vertex data into per-vertex and per-shape data. Keep only pos, cubics, and shapeID in vertex data
+[?] use half-floats or short fixed-point for pos and uv, packing them in two ints
+
+-----------
 [.] Check changes in macos version
 	[x] Restructure macos version to use mp_app_internal.h/mp_app.c
 	[x] test new run loop structure on macos
@@ -41,9 +47,11 @@
 	[ ] Text rendering test app
 	[?] Little test drawing app?
 
-	[>] Assess perf
+	[.] Assess perf
-		[ ] Tiling
+		[x] Tiling
-		[ ] Precomputing triangle edges/biases?
+		[>] Reduce size of data sent to GPU
+		[ ] experiment with persistently mapped buffer?
+		[?] Precomputing triangle edges/biases?
 
 [ ] Implement surfaces with child windows on win32
 	[/] Maybe implement compositing directly in d3d and opengl compat extension...

vector_renderer_notes.txt

@@ -80,3 +80,15 @@ Ideally, we would do the vertex computation on the GPU (opportunity to paralleli
 
 -> Problem profiling canvas renderer with angle (not really well supported in renderdoc)
 	-> Maybe use an OpenGL backend instead?
+
+Quick measurement on perf_text.exe
+----------------------------------
+* Re-allocate and copy each time with glBufferData		--> ~23ms
+* Allocate big buffer and update with glBufferSubData	--> ~23ms
+* Map whole buffer										--> ~44ms
+* Map whole buffer with GL_MAP_INVALIDATE_BUFFER_BIT	--> ~19ms (but with some startup hiccups...)
+* Map whole buffer with GL_MAP_INVALIDATE_RANGE_BIT		--> ~44ms
+
+
+-> Stutter is with GL_MAP_INVALIDATE_BUFFER_BIT isn't reassuring. Stick to glBufferData for now.
+-> May be worth it to try persistently mapped buffers later.