Apply targeted MapBufferRange optimization for NoOverwrite

- Only applies to NoOverwrite operations (not Discard)
- Uses ARB_map_buffer_range extension check for better portability
- More conservative approach to avoid driver-specific issues

Author: FireworkSky

src/FNA3D_Driver_OpenGL.c

@@ -4640,44 +4640,23 @@ static void OPENGL_SetVertexBufferData(
 
 	/* FIXME: Staging buffer for elementSizeInBytes < vertexStride! */
 
-	const GLsizeiptr updateSize = (GLsizeiptr) (elementCount * vertexStride);
-
-#ifdef USE_ES3
-	/* GLES3 optimization: Use glMapBufferRange to avoid CPU-GPU sync overhead.
-	 * GL_MAP_UNSYNCHRONIZED_BIT is critical for NoOverwrite (ring buffer) performance.
-	 * Desktop OpenGL doesn't benefit from this, so keep GLES3-only.
-	 */
-	if (renderer->supports_ARB_map_buffer_range && renderer->glMapBufferRange != NULL)
+	if (	options == FNA3D_SETDATAOPTIONS_NOOVERWRITE &&
+		renderer->supports_ARB_map_buffer_range	)
 	{
-		GLbitfield mapFlags = GL_MAP_WRITE_BIT;
-		
-		if (options == FNA3D_SETDATAOPTIONS_NOOVERWRITE)
-		{
-			mapFlags |= GL_MAP_UNSYNCHRONIZED_BIT;  /* No sync - huge speedup! */
-		}
-		else if (options == FNA3D_SETDATAOPTIONS_DISCARD)
-		{
-			mapFlags |= GL_MAP_INVALIDATE_RANGE_BIT;  /* Invalidate old data */
-		}
-		
-		void* ptr = renderer->glMapBufferRange(
+		void *ptr = renderer->glMapBufferRange(
 			GL_ARRAY_BUFFER,
 			(GLintptr) offsetInBytes,
-			updateSize,
-			mapFlags
+			(GLsizeiptr) (elementCount * vertexStride),
+			GL_MAP_WRITE_BIT | GL_MAP_UNSYNCHRONIZED_BIT
 		);
-		
 		if (ptr != NULL)
 		{
-			SDL_memcpy(ptr, data, updateSize);
+			SDL_memcpy(ptr, data, (elementCount * vertexStride));
 			renderer->glUnmapBuffer(GL_ARRAY_BUFFER);
-			return;
 		}
-		/* Fall through to glBufferSubData if map failed */
+		return;
 	}
-#endif
-	
-	/* Fallback: standard glBufferData/glBufferSubData path */
+
 	if (options == FNA3D_SETDATAOPTIONS_DISCARD)
 	{
 		renderer->glBufferData(
@@ -4691,7 +4670,7 @@ static void OPENGL_SetVertexBufferData(
 	renderer->glBufferSubData(
 		GL_ARRAY_BUFFER,
 		(GLintptr) offsetInBytes,
-		updateSize,
+		(GLsizeiptr) (elementCount * vertexStride),
 		data
 	);
 }
@@ -4862,39 +4841,23 @@ static void OPENGL_SetIndexBufferData(
 
 	BindIndexBuffer(renderer, glBuffer->handle);
 
-#ifdef USE_ES3
-	/* GLES3 optimization: Use glMapBufferRange for index buffers too */
-	if (renderer->supports_ARB_map_buffer_range && renderer->glMapBufferRange != NULL)
+	if (	options == FNA3D_SETDATAOPTIONS_NOOVERWRITE &&
+		renderer->supports_ARB_map_buffer_range	)
 	{
-		GLbitfield mapFlags = GL_MAP_WRITE_BIT;
-		
-		if (options == FNA3D_SETDATAOPTIONS_NOOVERWRITE)
-		{
-			mapFlags |= GL_MAP_UNSYNCHRONIZED_BIT;
-		}
-		else if (options == FNA3D_SETDATAOPTIONS_DISCARD)
-		{
-			mapFlags |= GL_MAP_INVALIDATE_RANGE_BIT;
-		}
-		
-		void* ptr = renderer->glMapBufferRange(
+		void *ptr = renderer->glMapBufferRange(
 			GL_ELEMENT_ARRAY_BUFFER,
 			(GLintptr) offsetInBytes,
 			(GLsizeiptr) dataLength,
-			mapFlags
+			GL_MAP_WRITE_BIT | GL_MAP_UNSYNCHRONIZED_BIT
 		);
-		
 		if (ptr != NULL)
 		{
 			SDL_memcpy(ptr, data, dataLength);
 			renderer->glUnmapBuffer(GL_ELEMENT_ARRAY_BUFFER);
-			return;
 		}
-		/* Fall through if map failed */
+		return;
 	}
-#endif
 
-	/* Fallback: standard path */
 	if (options == FNA3D_SETDATAOPTIONS_DISCARD)
 	{
 		renderer->glBufferData(
@@ -5377,16 +5340,8 @@ static uint8_t OPENGL_SupportsHardwareInstancing(FNA3D_Renderer *driverData)
 
 static uint8_t OPENGL_SupportsNoOverwrite(FNA3D_Renderer *driverData)
 {
-	/* NoOverwrite (ring buffer optimization) is critical for GLES3 SpriteBatch performance.
-	 * On desktop OpenGL, keep disabled as drivers handle buffer updates differently.
-	 * GLES3 benefits greatly from avoiding frequent glBufferData orphaning.
-	 */
-	#ifdef USE_ES3
-	return 1;  /* Enable for GLES3 only */
-	#else
-	(void) driverData;  /* Unused in desktop OpenGL */
-	return 0;  /* Keep disabled for desktop OpenGL */
-	#endif
+	OpenGLRenderer *renderer = (OpenGLRenderer*) driverData;
+	return renderer->supports_ARB_map_buffer_range;
 }
 
 static uint8_t OPENGL_SupportsSRGBRenderTargets(FNA3D_Renderer *driverData)

src/FNA3D_Driver_OpenGL.h

@@ -232,12 +232,9 @@ typedef unsigned char	GLboolean;
 #define GL_STATIC_DRAW  				0x88E4
 #define GL_MAX_VERTEX_ATTRIBS				0x8869
 
-/* Buffer map flags (GLES3 optimization) */
-#ifdef USE_ES3
+/* NoOverwrite Uploads */
 #define GL_MAP_WRITE_BIT				0x0002
-#define GL_MAP_INVALIDATE_RANGE_BIT			0x0004
 #define GL_MAP_UNSYNCHRONIZED_BIT			0x0020
-#endif
 
 /* Render targets */
 #define GL_FRAMEBUFFER  				0x8D40

src/FNA3D_Driver_OpenGL_glfuncs.h

@@ -38,14 +38,12 @@ GL_EXT(EXT_framebuffer_blit)
 GL_EXT(EXT_framebuffer_multisample)
 GL_EXT(ARB_internalformat_query)
 GL_EXT(ARB_invalidate_subdata)
-#ifdef USE_ES3
-GL_EXT(ARB_map_buffer_range)
-#endif
 GL_EXT(ARB_draw_instanced)
 GL_EXT(ARB_instanced_arrays)
 GL_EXT(ARB_draw_elements_base_vertex)
 GL_EXT(EXT_draw_buffers2)
 GL_EXT(ARB_texture_multisample)
+GL_EXT(ARB_map_buffer_range)
 GL_EXT(KHR_debug)
 GL_EXT(GREMEDY_string_marker)
 
@@ -155,11 +153,6 @@ GL_PROC(ARB_internalformat_query, void, glGetInternalformativ, (GLenum a, GLenum
 /* This is mostly needed by ES3, where loads/stores are a huge slowdown */
 GL_PROC(ARB_invalidate_subdata, void, glInvalidateFramebuffer, (GLenum a, GLsizei b, const GLenum *c))
 
-/* Buffer mapping for GLES3 optimization */
-#ifdef USE_ES3
-GL_PROC_EXT(ARB_map_buffer_range, EXT, GLvoid*, glMapBufferRange, (GLenum a, GLintptr b, GLsizeiptr c, GLbitfield d))
-#endif
-
 /* Hardware instancing is nice to have, but isn't used all the time */
 GL_PROC(ARB_draw_instanced, void, glDrawElementsInstanced, (GLenum a, GLsizei b, GLenum c, const GLvoid *d, GLsizei e))
 GL_PROC(ARB_instanced_arrays, void, glVertexAttribDivisor, (GLuint a, GLuint b))
@@ -178,6 +171,9 @@ GL_PROC(EXT_draw_buffers2, void, glColorMaski, (GLuint a, GLboolean b, GLboolean
 /* Probably used by nobody, honestly */
 GL_PROC(ARB_texture_multisample, void, glSampleMaski, (GLuint a, GLuint b))
 
+/* Technically UnmapBuffer is core, but useless without MapBufferRange */
+GL_PROC_EXT(ARB_map_buffer_range, EXT, GLvoid*, glMapBufferRange, (GLenum a, GLintptr b, GLsizeiptr c, GLbitfield d))
+
 /* "NOTE: when implemented in an OpenGL ES context, all entry points defined
  * by this extension must have a "KHR" suffix. When implemented in an
  * OpenGL context, all entry points must have NO suffix, as shown below."