Hi, Attached it a patch that omits the copy to current in display lists that are built from draw calls that allow the current state to be undefined like glDrawArrays, ... That change moves the copy to current function down in profiles. Also ctx->NewState is set less often which might save some recomputations.
Please review/apply. Thanks! Mathias
From 823bdef074d83a8798621f80deea21c9e4c65179 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Mathias=20Fr=C3=B6hlich?= <[email protected]> Date: Sun, 17 Oct 2010 18:16:30 +0200 Subject: [PATCH] Avoid the copy to current in dlists if not required. The current state is allowed to be undefined past DrawElements et al. Consequently omit that copying at least in the display list code. This pays us some percents performance. --- src/mesa/vbo/vbo.h | 3 +- src/mesa/vbo/vbo_save.h | 4 ++- src/mesa/vbo/vbo_save_api.c | 54 ++++++++++++++++++++++++------------------ 3 files changed, 36 insertions(+), 25 deletions(-) diff --git a/src/mesa/vbo/vbo.h b/src/mesa/vbo/vbo.h index c1406b2..79f7665 100644 --- a/src/mesa/vbo/vbo.h +++ b/src/mesa/vbo/vbo.h @@ -40,7 +40,8 @@ struct _mesa_prim { GLuint begin:1; GLuint end:1; GLuint weak:1; - GLuint pad:20; + GLuint no_current_update:1; + GLuint pad:19; GLuint start; GLuint count; diff --git a/src/mesa/vbo/vbo_save.h b/src/mesa/vbo/vbo_save.h index f5a407c..23cbea2 100644 --- a/src/mesa/vbo/vbo_save.h +++ b/src/mesa/vbo/vbo_save.h @@ -96,7 +96,9 @@ struct vbo_save_vertex_list { */ #define VBO_SAVE_BUFFER_SIZE (8*1024) /* dwords */ #define VBO_SAVE_PRIM_SIZE 128 -#define VBO_SAVE_PRIM_WEAK 0x40 +#define VBO_SAVE_PRIM_MODE_MASK 0x3f +#define VBO_SAVE_PRIM_WEAK 0x40 +#define VBO_SAVE_PRIM_NO_CURRENT_UPDATE 0x80 #define VBO_SAVE_FALLBACK 0x10000000 diff --git a/src/mesa/vbo/vbo_save_api.c b/src/mesa/vbo/vbo_save_api.c index 817d478..bf5ceda 100644 --- a/src/mesa/vbo/vbo_save_api.c +++ b/src/mesa/vbo/vbo_save_api.c @@ -294,26 +294,30 @@ static void _save_compile_vertex_list( struct gl_context *ctx ) node->vertex_store->refcount++; node->prim_store->refcount++; - - node->current_size = node->vertex_size - node->attrsz[0]; - node->current_data = NULL; - - if (node->current_size) { - /* If the malloc fails, we just pull the data out of the VBO - * later instead. - */ - node->current_data = MALLOC( node->current_size * sizeof(GLfloat) ); - if (node->current_data) { - const char *buffer = (const char *)save->vertex_store->buffer; - unsigned attr_offset = node->attrsz[0] * sizeof(GLfloat); - unsigned vertex_offset = 0; - - if (node->count) - vertex_offset = (node->count-1) * node->vertex_size * sizeof(GLfloat); - - memcpy( node->current_data, - buffer + node->buffer_offset + vertex_offset + attr_offset, - node->current_size * sizeof(GLfloat) ); + if (node->prim[0].no_current_update) { + node->current_size = 0; + node->current_data = NULL; + } else { + node->current_size = node->vertex_size - node->attrsz[0]; + node->current_data = NULL; + + if (node->current_size) { + /* If the malloc fails, we just pull the data out of the VBO + * later instead. + */ + node->current_data = MALLOC( node->current_size * sizeof(GLfloat) ); + if (node->current_data) { + const char *buffer = (const char *)save->vertex_store->buffer; + unsigned attr_offset = node->attrsz[0] * sizeof(GLfloat); + unsigned vertex_offset = 0; + + if (node->count) + vertex_offset = (node->count-1) * node->vertex_size * sizeof(GLfloat); + + memcpy( node->current_data, + buffer + node->buffer_offset + vertex_offset + attr_offset, + node->current_size * sizeof(GLfloat) ); + } } } @@ -397,6 +401,7 @@ static void _save_wrap_buffers( struct gl_context *ctx ) GLint i = save->prim_count - 1; GLenum mode; GLboolean weak; + GLboolean no_current_update; assert(i < (GLint) save->prim_max); assert(i >= 0); @@ -407,6 +412,7 @@ static void _save_wrap_buffers( struct gl_context *ctx ) save->prim[i].start); mode = save->prim[i].mode; weak = save->prim[i].weak; + no_current_update = save->prim[i].no_current_update; /* store the copied vertices, and allocate a new list. */ @@ -416,6 +422,7 @@ static void _save_wrap_buffers( struct gl_context *ctx ) */ save->prim[0].mode = mode; save->prim[0].weak = weak; + save->prim[0].no_current_update = no_current_update; save->prim[0].begin = 0; save->prim[0].end = 0; save->prim[0].pad = 0; @@ -770,10 +777,11 @@ GLboolean vbo_save_NotifyBegin( struct gl_context *ctx, GLenum mode ) GLuint i = save->prim_count++; assert(i < save->prim_max); - save->prim[i].mode = mode & ~VBO_SAVE_PRIM_WEAK; + save->prim[i].mode = mode & VBO_SAVE_PRIM_MODE_MASK; save->prim[i].begin = 1; save->prim[i].end = 0; save->prim[i].weak = (mode & VBO_SAVE_PRIM_WEAK) ? 1 : 0; + save->prim[i].no_current_update = (mode & VBO_SAVE_PRIM_NO_CURRENT_UPDATE) ? 1 : 0; save->prim[i].pad = 0; save->prim[i].start = save->vert_count; save->prim[i].count = 0; @@ -934,7 +942,7 @@ static void GLAPIENTRY _save_OBE_DrawArrays(GLenum mode, GLint start, GLsizei co _ae_map_vbos( ctx ); - vbo_save_NotifyBegin( ctx, mode | VBO_SAVE_PRIM_WEAK ); + vbo_save_NotifyBegin( ctx, mode | VBO_SAVE_PRIM_WEAK | VBO_SAVE_PRIM_NO_CURRENT_UPDATE); for (i = 0; i < count; i++) CALL_ArrayElement(GET_DISPATCH(), (start + i)); @@ -960,7 +968,7 @@ static void GLAPIENTRY _save_OBE_DrawElements(GLenum mode, GLsizei count, GLenum if (_mesa_is_bufferobj(ctx->Array.ElementArrayBufferObj)) indices = ADD_POINTERS(ctx->Array.ElementArrayBufferObj->Pointer, indices); - vbo_save_NotifyBegin( ctx, mode | VBO_SAVE_PRIM_WEAK ); + vbo_save_NotifyBegin( ctx, mode | VBO_SAVE_PRIM_WEAK | VBO_SAVE_PRIM_NO_CURRENT_UPDATE ); switch (type) { case GL_UNSIGNED_BYTE: -- 1.7.2.3
_______________________________________________ mesa-dev mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/mesa-dev
