Hi,
I've built several patches trying to reduce the propagation of state changes.
They are proof-of-concept, and I'd like to know what you think about
them, mainly :
- do you think it's worth it ? it adds complexity to state trackers,
but it avoids some possibly expensive driver function calls.
- if yes, what would you be the best design approach for this ?
For reference, Airlied built a patch similar to my 4th one, but with a
different approach (adding an incrementing 'Generation' field to
struct instead of comparing pointers) : http://fpaste.org/QTng/
(all tests done using 2.6.38.3/64 bits/r600g)
== Regression ? ==
- piglit quick.tests with all 4 patches applied => no regression
== Some simple benchmarking ==
Successive patches applied :
* 01 : st_atom_constbuf
* 02 : st_atom_rasterizer
* 03 : st_atom_texture
* 04 : st_atom_sampler
mesa-demos: drawoverhead (draws/sec):
* Draw only : 2100t
* Draw w/ nop sc : 2100t
* Draw w/ sc :
|git from 05/18 | +01 | +01->02 | +01->03 | +01->04 |
| 227t | 328t | 328t | 350t | 365t |
openarena (anholt benchmark, 3 runs average) :
|git from 05/18 | +01 | +01->02 | +01->03 | +01->04 |
| 74.6 | 75.0 | 74.4 | 75.7 | 74.7 |
-- Pierre-Eric
From fc58de87ddaf943942a81b0ec4deb8a0ce205534 Mon Sep 17 00:00:00 2001
From: Pierre-Eric Pelloux-Prayer <pell...@gmail.com>
Date: Fri, 20 May 2011 22:52:43 +0200
Subject: [PATCH 1/4] st/mesa : reduced constant buffers upload
This patch add a checksum like computation before a upload constant buffer.
If the checksum equals the previous one, the upload is skipped.
---
src/mesa/state_tracker/st_atom_constbuf.c | 35 +++++++++++++++++++++++-----
src/mesa/state_tracker/st_atom_constbuf.h | 5 ++-
src/mesa/state_tracker/st_cb_bitmap.c | 2 +-
src/mesa/state_tracker/st_cb_drawpixels.c | 4 +-
4 files changed, 34 insertions(+), 12 deletions(-)
diff --git a/src/mesa/state_tracker/st_atom_constbuf.c b/src/mesa/state_tracker/st_atom_constbuf.c
index 05667a7..cccd7e7 100644
--- a/src/mesa/state_tracker/st_atom_constbuf.c
+++ b/src/mesa/state_tracker/st_atom_constbuf.c
@@ -51,9 +51,10 @@
* constant buffer.
* \param shader_type either PIPE_SHADER_VERTEX or PIPE_SHADER_FRAGMENT
*/
-void st_upload_constants( struct st_context *st,
+int st_upload_constants( struct st_context *st,
struct gl_program_parameter_list *params,
- unsigned shader_type)
+ unsigned shader_type,
+ int previousTotal)
{
struct pipe_context *pipe = st->pipe;
@@ -73,6 +74,24 @@ void st_upload_constants( struct st_context *st,
*/
_mesa_load_state_parameters(st->ctx, params);
+ unsigned int total = (((unsigned int) st) ^ params->NumParameters), tmp;
+ // fprintf(stdout, "[%p] Dump (%d bytes) : \n", st, paramBytes);
+ for(int i=0; i<params->NumParameters; i++) {
+ for(int j=0; j<4; j++) {
+ memcpy(&tmp, ¶ms->ParameterValues[i][j], sizeof(int));
+ total = (total>>(sizeof(int)-1))^(total<<1)^tmp;
+
+ /*for(int k=0; k<4; ++k)
+ fprintf(stdout, "%02x ", ((unsigned char*) &tmp)[k]);*/
+ }
+ //fprintf(stdout, "\n");
+ }
+ //fprintf(stdout, " Total : %08x\n", total);
+ if (previousTotal == total) {
+ //fprintf(stdout, "Early quit !\n");
+ return total;
+ }
+
/* We always need to get a new buffer, to keep the drivers simple and
* avoid gratuitous rendering synchronization.
* Let's use a user buffer to avoid an unnecessary copy.
@@ -81,7 +100,6 @@ void st_upload_constants( struct st_context *st,
params->ParameterValues,
paramBytes,
PIPE_BIND_CONSTANT_BUFFER);
-
if (ST_DEBUG & DEBUG_CONSTANTS) {
debug_printf("%s(shader=%d, numParams=%d, stateFlags=0x%x)\n",
__FUNCTION__, shader_type, params->NumParameters,
@@ -94,12 +112,14 @@ void st_upload_constants( struct st_context *st,
st->state.constants[shader_type].ptr = params->ParameterValues;
st->state.constants[shader_type].size = paramBytes;
+ return total;
}
else if (st->state.constants[shader_type].ptr) {
st->state.constants[shader_type].ptr = NULL;
st->state.constants[shader_type].size = 0;
st->pipe->set_constant_buffer(st->pipe, shader_type, 0, NULL);
}
+ return 0;
}
@@ -108,10 +128,11 @@ void st_upload_constants( struct st_context *st,
*/
static void update_vs_constants(struct st_context *st )
{
+ static int previous_total = -1;
struct st_vertex_program *vp = st->vp;
struct gl_program_parameter_list *params = vp->Base.Base.Parameters;
- st_upload_constants( st, params, PIPE_SHADER_VERTEX );
+ previous_total = st_upload_constants( st, params, PIPE_SHADER_VERTEX, previous_total );
}
@@ -131,10 +152,10 @@ const struct st_tracked_state st_update_vs_constants = {
*/
static void update_fs_constants(struct st_context *st )
{
+ static int previous_total = -1;
struct st_fragment_program *fp = st->fp;
struct gl_program_parameter_list *params = fp->Base.Base.Parameters;
-
- st_upload_constants( st, params, PIPE_SHADER_FRAGMENT );
+ previous_total = st_upload_constants( st, params, PIPE_SHADER_FRAGMENT, previous_total );
}
@@ -156,7 +177,7 @@ static void update_gs_constants(struct st_context *st )
if (gp) {
params = gp->Base.Base.Parameters;
- st_upload_constants( st, params, PIPE_SHADER_GEOMETRY );
+ st_upload_constants( st, params, PIPE_SHADER_GEOMETRY, -1 );
}
}
diff --git a/src/mesa/state_tracker/st_atom_constbuf.h b/src/mesa/state_tracker/st_atom_constbuf.h
index 97b0766..8749bf5 100644
--- a/src/mesa/state_tracker/st_atom_constbuf.h
+++ b/src/mesa/state_tracker/st_atom_constbuf.h
@@ -33,9 +33,10 @@ struct gl_program_parameter_list;
struct st_context;
-void st_upload_constants( struct st_context *st,
+int st_upload_constants( struct st_context *st,
struct gl_program_parameter_list *params,
- unsigned id);
+ unsigned id,
+ int previousTotal);
#endif /* ST_ATOM_CONSTBUF_H */
diff --git a/src/mesa/state_tracker/st_cb_bitmap.c b/src/mesa/state_tracker/st_cb_bitmap.c
index 49b1960..0ffc729 100644
--- a/src/mesa/state_tracker/st_cb_bitmap.c
+++ b/src/mesa/state_tracker/st_cb_bitmap.c
@@ -437,7 +437,7 @@ draw_bitmap_quad(struct gl_context *ctx, GLint x, GLint y, GLfloat z,
GLfloat colorSave[4];
COPY_4V(colorSave, ctx->Current.Attrib[VERT_ATTRIB_COLOR0]);
COPY_4V(ctx->Current.Attrib[VERT_ATTRIB_COLOR0], color);
- st_upload_constants(st, fpv->parameters, PIPE_SHADER_FRAGMENT);
+ st_upload_constants(st, fpv->parameters, PIPE_SHADER_FRAGMENT, -1);
COPY_4V(ctx->Current.Attrib[VERT_ATTRIB_COLOR0], colorSave);
}
diff --git a/src/mesa/state_tracker/st_cb_drawpixels.c b/src/mesa/state_tracker/st_cb_drawpixels.c
index 9948f8d..57ede72 100644
--- a/src/mesa/state_tracker/st_cb_drawpixels.c
+++ b/src/mesa/state_tracker/st_cb_drawpixels.c
@@ -1034,7 +1034,7 @@ st_DrawPixels(struct gl_context *ctx, GLint x, GLint y,
}
/* update fragment program constants */
- st_upload_constants(st, fpv->parameters, PIPE_SHADER_FRAGMENT);
+ st_upload_constants(st, fpv->parameters, PIPE_SHADER_FRAGMENT, -1);
/* draw with textured quad */
{
@@ -1370,7 +1370,7 @@ st_CopyPixels(struct gl_context *ctx, GLint srcx, GLint srcy,
}
/* update fragment program constants */
- st_upload_constants(st, fpv->parameters, PIPE_SHADER_FRAGMENT);
+ st_upload_constants(st, fpv->parameters, PIPE_SHADER_FRAGMENT, -1);
if (rbRead->Base.Wrapped)
--
1.7.5.1
From 50498d5fe884b1b5ee1a007fe21a5b3d6fe161ba Mon Sep 17 00:00:00 2001
From: Pierre-Eric Pelloux-Prayer <pell...@gmail.com>
Date: Fri, 20 May 2011 22:54:19 +0200
Subject: [PATCH 2/4] st/mesa : reduce 'cso_set_rasterizer' calls count
cso_set_rasterizer is now called only if "struct pipe_rasterizer_state" content actually changed.
---
src/mesa/state_tracker/st_atom_rasterizer.c | 178 ++++++++++++++++++---------
1 files changed, 117 insertions(+), 61 deletions(-)
diff --git a/src/mesa/state_tracker/st_atom_rasterizer.c b/src/mesa/state_tracker/st_atom_rasterizer.c
index 250cbb2..127d611 100644
--- a/src/mesa/state_tracker/st_atom_rasterizer.c
+++ b/src/mesa/state_tracker/st_atom_rasterizer.c
@@ -58,17 +58,25 @@ static GLuint translate_fill( GLenum mode )
static void update_raster_state( struct st_context *st )
{
struct gl_context *ctx = st->ctx;
- struct pipe_rasterizer_state *raster = &st->state.rasterizer;
+ struct pipe_rasterizer_state tmp;
+ struct pipe_rasterizer_state *raster = &tmp; // &st->state.rasterizer;
const struct gl_vertex_program *vertProg = ctx->VertexProgram._Current;
const struct gl_fragment_program *fragProg = ctx->FragmentProgram._Current;
uint i;
- memset(raster, 0, sizeof(*raster));
+ bool change_occured = false;
+#define ASSIGN(field, value) { if (field!=value) { change_occured = true; field = value; } }
+#define EVAL_AND_ASSIGN(field, value, T) { T __tmp = (value); if (field!=__tmp) { change_occured = true; field = __tmp; } }
+
+ /* Only the changing fields are going to be updated and the change_occured flag will be set it needed
+ */
+ memcpy(raster, &st->state.rasterizer, sizeof(*raster));
/* _NEW_POLYGON, _NEW_BUFFERS
*/
+ if ((st->dirty.mesa & _NEW_POLYGON) | (st->dirty.mesa & _NEW_BUFFERS))
{
- raster->front_ccw = (ctx->Polygon.FrontFace == GL_CCW);
+ unsigned v = (ctx->Polygon.FrontFace == GL_CCW);
/*
* Gallium's surfaces are Y=0=TOP orientation. OpenGL is the
@@ -78,17 +86,19 @@ static void update_raster_state( struct st_context *st )
*/
if (st_fb_orientation(ctx->DrawBuffer) == Y_0_BOTTOM) {
/* Drawing to an FBO. The viewport will be inverted. */
- raster->front_ccw ^= 1;
+ v ^= 1;
}
+
+ ASSIGN(raster->front_ccw, v)
}
/* _NEW_LIGHT
*/
- if (ctx->Light.ShadeModel == GL_FLAT)
- raster->flatshade = 1;
-
- if (ctx->Light.ProvokingVertex == GL_FIRST_VERTEX_CONVENTION_EXT)
- raster->flatshade_first = 1;
+ if (st->dirty.mesa & _NEW_LIGHT)
+ {
+ EVAL_AND_ASSIGN(raster->flatshade, (ctx->Light.ShadeModel == GL_FLAT), bool)
+ EVAL_AND_ASSIGN(raster->flatshade_first, (ctx->Light.ProvokingVertex == GL_FIRST_VERTEX_CONVENTION_EXT), bool)
+ }
/* _NEW_LIGHT | _NEW_PROGRAM
*
@@ -96,170 +106,216 @@ static void update_raster_state( struct st_context *st )
* GL_LIGHT_MODEL_TWO_SIDE is set) or from vertex programs/shaders (when
* GL_VERTEX_PROGRAM_TWO_SIDE is set). Note the logic here.
*/
+ if ((st->dirty.mesa & _NEW_LIGHT) | (st->dirty.mesa & _NEW_PROGRAM))
+ {
+ unsigned light_twoside = 0;
if (ctx->VertexProgram._Current) {
if (ctx->VertexProgram._Enabled ||
(ctx->Shader.CurrentVertexProgram &&
ctx->Shader.CurrentVertexProgram->LinkStatus)) {
/* user-defined vertex program or shader */
- raster->light_twoside = ctx->VertexProgram.TwoSideEnabled;
+ light_twoside = ctx->VertexProgram.TwoSideEnabled;
}
else {
/* TNL-generated program */
- raster->light_twoside = ctx->Light.Enabled && ctx->Light.Model.TwoSide;
+ light_twoside = ctx->Light.Enabled && ctx->Light.Model.TwoSide;
}
}
else if (ctx->Light.Enabled && ctx->Light.Model.TwoSide) {
- raster->light_twoside = 1;
+ light_twoside = 1;
+ }
+ ASSIGN(raster->light_twoside, light_twoside);
+ ASSIGN(raster->clamp_vertex_color, raster->clamp_vertex_color);
}
-
- raster->clamp_vertex_color = ctx->Light._ClampVertexColor;
/* _NEW_POLYGON
*/
+ if ((st->dirty.mesa & _NEW_POLYGON))
+ {
if (ctx->Polygon.CullFlag) {
switch (ctx->Polygon.CullFaceMode) {
case GL_FRONT:
- raster->cull_face = PIPE_FACE_FRONT;
+ ASSIGN(raster->cull_face, PIPE_FACE_FRONT)
break;
case GL_BACK:
- raster->cull_face = PIPE_FACE_BACK;
+ ASSIGN(raster->cull_face, PIPE_FACE_BACK)
break;
case GL_FRONT_AND_BACK:
- raster->cull_face = PIPE_FACE_FRONT_AND_BACK;
+ ASSIGN(raster->cull_face, PIPE_FACE_FRONT_AND_BACK)
break;
}
}
else {
- raster->cull_face = PIPE_FACE_NONE;
+ ASSIGN(raster->cull_face, PIPE_FACE_NONE)
}
+
/* _NEW_POLYGON
*/
{
- raster->fill_front = translate_fill( ctx->Polygon.FrontMode );
- raster->fill_back = translate_fill( ctx->Polygon.BackMode );
+ unsigned fill_front = translate_fill( ctx->Polygon.FrontMode );
+ unsigned fill_back = translate_fill( ctx->Polygon.BackMode );
/* Simplify when culling is active:
*/
if (raster->cull_face & PIPE_FACE_FRONT) {
- raster->fill_front = raster->fill_back;
+ fill_front = fill_back;
}
if (raster->cull_face & PIPE_FACE_BACK) {
- raster->fill_back = raster->fill_front;
+ fill_back = fill_front;
}
+ ASSIGN(raster->fill_front, fill_front)
+ ASSIGN(raster->fill_back, fill_back)
}
/* _NEW_POLYGON
*/
if (ctx->Polygon.OffsetUnits != 0.0 ||
ctx->Polygon.OffsetFactor != 0.0) {
- raster->offset_point = ctx->Polygon.OffsetPoint;
- raster->offset_line = ctx->Polygon.OffsetLine;
- raster->offset_tri = ctx->Polygon.OffsetFill;
+ ASSIGN(raster->offset_point, ctx->Polygon.OffsetPoint)
+ ASSIGN(raster->offset_line, ctx->Polygon.OffsetLine)
+ ASSIGN(raster->offset_tri, ctx->Polygon.OffsetFill)
+ } else {
+ ASSIGN(raster->offset_point, 0)
+ ASSIGN(raster->offset_line, 0)
+ ASSIGN(raster->offset_tri, 0)
}
if (ctx->Polygon.OffsetPoint ||
ctx->Polygon.OffsetLine ||
ctx->Polygon.OffsetFill) {
- raster->offset_units = ctx->Polygon.OffsetUnits;
- raster->offset_scale = ctx->Polygon.OffsetFactor;
+ ASSIGN(raster->offset_units, ctx->Polygon.OffsetUnits)
+ ASSIGN(raster->offset_scale, ctx->Polygon.OffsetFactor)
+ } else {
+ ASSIGN(raster->offset_units, 0)
+ ASSIGN(raster->offset_scale, 0)
}
- if (ctx->Polygon.SmoothFlag)
- raster->poly_smooth = 1;
-
- if (ctx->Polygon.StippleFlag)
- raster->poly_stipple_enable = 1;
+ ASSIGN(raster->poly_smooth, ctx->Polygon.SmoothFlag)
+ ASSIGN(raster->poly_stipple_enable, ctx->Polygon.StippleFlag)
+ }
/* _NEW_POINT
*/
- raster->point_size = ctx->Point.Size;
-
- if (!ctx->Point.PointSprite && ctx->Point.SmoothFlag)
- raster->point_smooth = 1;
+ if (st->dirty.mesa & _NEW_POINT)
+ {
+ EVAL_AND_ASSIGN(raster->point_smooth, (!ctx->Point.PointSprite && ctx->Point.SmoothFlag), bool)
+ }
/* _NEW_POINT | _NEW_PROGRAM
*/
+ if ((st->dirty.mesa & _NEW_POINT) | (st->dirty.mesa & _NEW_PROGRAM))
+ {
if (ctx->Point.PointSprite) {
/* origin */
if ((ctx->Point.SpriteOrigin == GL_UPPER_LEFT) ^
(st_fb_orientation(ctx->DrawBuffer) == Y_0_BOTTOM))
- raster->sprite_coord_mode = PIPE_SPRITE_COORD_UPPER_LEFT;
+ ASSIGN(raster->sprite_coord_mode, PIPE_SPRITE_COORD_UPPER_LEFT)
else
- raster->sprite_coord_mode = PIPE_SPRITE_COORD_LOWER_LEFT;
+ ASSIGN(raster->sprite_coord_mode, PIPE_SPRITE_COORD_LOWER_LEFT)
/* Coord replacement flags. If bit 'k' is set that means
* that we need to replace GENERIC[k] attrib with an automatically
* computed texture coord.
*/
+ unsigned sprite_coord_enable = 0;
for (i = 0; i < MAX_TEXTURE_COORD_UNITS; i++) {
if (ctx->Point.CoordReplace[i]) {
- raster->sprite_coord_enable |= 1 << i;
+ sprite_coord_enable |= 1 << i;
}
}
if (fragProg->Base.InputsRead & FRAG_BIT_PNTC) {
- raster->sprite_coord_enable |=
+ sprite_coord_enable |=
1 << (FRAG_ATTRIB_PNTC - FRAG_ATTRIB_TEX0);
}
-
- raster->point_quad_rasterization = 1;
+ ASSIGN(raster->sprite_coord_enable, sprite_coord_enable)
+ ASSIGN(raster->point_quad_rasterization, 1)
+ } else {
+ ASSIGN(raster->sprite_coord_mode, 0)
+ ASSIGN(raster->sprite_coord_enable, 0)
+ ASSIGN(raster->point_quad_rasterization, 0)
+ }
}
/* ST_NEW_VERTEX_PROGRAM
*/
+ if (st->dirty.st & ST_NEW_VERTEX_PROGRAM)
+ {
+ bool point_size_per_vertex = FALSE;
if (vertProg) {
if (vertProg->Base.Id == 0) {
if (vertProg->Base.OutputsWritten & BITFIELD64_BIT(VERT_RESULT_PSIZ)) {
/* generated program which emits point size */
- raster->point_size_per_vertex = TRUE;
+ point_size_per_vertex = TRUE;
}
}
else if (ctx->VertexProgram.PointSizeEnabled) {
/* user-defined program and GL_VERTEX_PROGRAM_POINT_SIZE set */
- raster->point_size_per_vertex = ctx->VertexProgram.PointSizeEnabled;
+ point_size_per_vertex = ctx->VertexProgram.PointSizeEnabled;
+ }
}
+
+ ASSIGN(raster->point_size_per_vertex, point_size_per_vertex)
}
+
if (!raster->point_size_per_vertex) {
/* clamp size now */
- raster->point_size = CLAMP(ctx->Point.Size,
+ EVAL_AND_ASSIGN(raster->point_size, CLAMP(
+ ctx->Point.Size,
ctx->Point.MinSize,
- ctx->Point.MaxSize);
+ ctx->Point.MaxSize), float)
+ } else {
+ ASSIGN(raster->point_size, ctx->Point.Size);
}
/* _NEW_LINE
*/
- raster->line_smooth = ctx->Line.SmoothFlag;
+ if (st->dirty.mesa & _NEW_LINE)
+ {
+ ASSIGN(raster->line_smooth, ctx->Line.SmoothFlag)
+
if (ctx->Line.SmoothFlag) {
- raster->line_width = CLAMP(ctx->Line.Width,
+ EVAL_AND_ASSIGN(raster->line_width, CLAMP(ctx->Line.Width,
ctx->Const.MinLineWidthAA,
- ctx->Const.MaxLineWidthAA);
+ ctx->Const.MaxLineWidthAA), float)
}
else {
- raster->line_width = CLAMP(ctx->Line.Width,
+ EVAL_AND_ASSIGN(raster->line_width, CLAMP(ctx->Line.Width,
ctx->Const.MinLineWidth,
- ctx->Const.MaxLineWidth);
+ ctx->Const.MaxLineWidth), float)
}
- raster->line_stipple_enable = ctx->Line.StippleFlag;
- raster->line_stipple_pattern = ctx->Line.StipplePattern;
+ ASSIGN(raster->line_stipple_enable, ctx->Line.StippleFlag)
+ ASSIGN(raster->line_stipple_pattern, ctx->Line.StipplePattern)
/* GL stipple factor is in [1,256], remap to [0, 255] here */
- raster->line_stipple_factor = ctx->Line.StippleFactor - 1;
+ ASSIGN(raster->line_stipple_factor, ctx->Line.StippleFactor - 1)
+ }
/* _NEW_MULTISAMPLE */
- if (ctx->Multisample._Enabled || st->force_msaa)
- raster->multisample = 1;
+ if (st->dirty.mesa & _NEW_MULTISAMPLE)
+ {
+ EVAL_AND_ASSIGN(raster->multisample, (ctx->Multisample._Enabled || st->force_msaa), bool)
+ }
/* _NEW_SCISSOR */
- if (ctx->Scissor.Enabled)
- raster->scissor = 1;
+ if (st->dirty.mesa & _NEW_SCISSOR)
+ {
+ ASSIGN(raster->scissor, ctx->Scissor.Enabled)
+ }
/* _NEW_FRAG_CLAMP */
- raster->clamp_fragment_color = ctx->Color._ClampFragmentColor;
+ if (st->dirty.mesa & _NEW_FRAG_CLAMP)
+ {
+ ASSIGN(raster->clamp_fragment_color, ctx->Color._ClampFragmentColor)
+ }
- raster->gl_rasterization_rules = 1;
+ ASSIGN(raster->gl_rasterization_rules, 1)
- cso_set_rasterizer(st->cso_context, raster);
+ if (change_occured) {
+ memcpy(&st->state.rasterizer, raster, sizeof(struct pipe_rasterizer_state));
+ cso_set_rasterizer(st->cso_context, &st->state.rasterizer); // raster);
+ }
}
const struct st_tracked_state st_update_rasterizer = {
--
1.7.5.1
From d94fff49a92401bb0c36828e1ab8e0d3925b9c28 Mon Sep 17 00:00:00 2001
From: Pierre-Eric Pelloux-Prayer <pell...@gmail.com>
Date: Fri, 20 May 2011 22:56:05 +0200
Subject: [PATCH 3/4] st/mesa : reduce
cso_set_vertex_sampler_views/cso_set_fragment_sampler_views
calls
Above functions called only if samplers_view/samplervertex_view content have actually changed.
---
src/gallium/auxiliary/cso_cache/cso_context.c | 1 +
src/mesa/state_tracker/st_atom_texture.c | 20 +++++++++++++++++++-
2 files changed, 20 insertions(+), 1 deletions(-)
diff --git a/src/gallium/auxiliary/cso_cache/cso_context.c b/src/gallium/auxiliary/cso_cache/cso_context.c
index fdd40fc..1b5f06b 100644
--- a/src/gallium/auxiliary/cso_cache/cso_context.c
+++ b/src/gallium/auxiliary/cso_cache/cso_context.c
@@ -1193,6 +1193,7 @@ set_sampler_views(struct cso_context *ctx,
/* reference new views */
for (i = 0; i < count; i++) {
+ if (views[i] != info->views[i])
pipe_sampler_view_reference(&info->views[i], views[i]);
}
/* unref extra old views, if any */
diff --git a/src/mesa/state_tracker/st_atom_texture.c b/src/mesa/state_tracker/st_atom_texture.c
index 072eb97..c711715 100644
--- a/src/mesa/state_tracker/st_atom_texture.c
+++ b/src/mesa/state_tracker/st_atom_texture.c
@@ -256,8 +256,10 @@ update_vertex_textures(struct st_context *st)
struct gl_vertex_program *vprog = st->ctx->VertexProgram._Current;
GLuint su;
+ int old_num_vertex_textures = st->state.num_vertex_textures;
st->state.num_vertex_textures = 0;
+ bool change_occured = false;
/* loop over sampler units (aka tex image units) */
for (su = 0; su < st->ctx->Const.MaxTextureImageUnits; su++) {
struct pipe_sampler_view *sampler_view = NULL;
@@ -274,10 +276,17 @@ update_vertex_textures(struct st_context *st)
st->state.num_vertex_textures = su + 1;
}
+
+ if (sampler_view != st->state.sampler_vertex_views[su]) {
pipe_sampler_view_reference(&st->state.sampler_vertex_views[su], sampler_view);
+ change_occured = true;
+ }
}
- if (st->ctx->Const.MaxVertexTextureImageUnits > 0) {
+ if (old_num_vertex_textures != st->state.num_vertex_textures)
+ change_occured = true;
+
+ if (change_occured && st->ctx->Const.MaxVertexTextureImageUnits > 0) {
GLuint numUnits = MIN2(st->state.num_vertex_textures,
st->ctx->Const.MaxVertexTextureImageUnits);
cso_set_vertex_sampler_views(st->cso_context,
@@ -292,8 +301,10 @@ update_fragment_textures(struct st_context *st)
struct gl_fragment_program *fprog = st->ctx->FragmentProgram._Current;
GLuint su;
+ int old_num_textures = st->state.num_textures;
st->state.num_textures = 0;
+ bool change_occured = false;
/* loop over sampler units (aka tex image units) */
for (su = 0; su < st->ctx->Const.MaxTextureImageUnits; su++) {
struct pipe_sampler_view *sampler_view = NULL;
@@ -309,9 +320,16 @@ update_fragment_textures(struct st_context *st)
st->state.num_textures = su + 1;
}
+ if (st->state.sampler_views[su] != sampler_view) {
pipe_sampler_view_reference(&st->state.sampler_views[su], sampler_view);
+ change_occured = true;
}
+ }
+
+ if (old_num_textures != st->state.num_textures)
+ change_occured = true;
+ if (change_occured)
cso_set_fragment_sampler_views(st->cso_context,
st->state.num_textures,
st->state.sampler_views);
--
1.7.5.1
From 9df2b0250725fa6cc3a6278a8b1fe210b81afb27 Mon Sep 17 00:00:00 2001
From: Pierre-Eric Pelloux-Prayer <pell...@gmail.com>
Date: Fri, 20 May 2011 22:58:30 +0200
Subject: [PATCH 4/4] st/mesa : reduce
cso_single_sampler/cso_single_vertex_sampler calls
Above functions called only if relevant samplers have changed.
---
src/gallium/auxiliary/cso_cache/cso_context.c | 59 -------------------------
src/gallium/auxiliary/cso_cache/cso_context.h | 57 +++++++++++++++++++++++-
src/mesa/state_tracker/st_atom_sampler.c | 18 +++++++-
3 files changed, 72 insertions(+), 62 deletions(-)
diff --git a/src/gallium/auxiliary/cso_cache/cso_context.c b/src/gallium/auxiliary/cso_cache/cso_context.c
index 1b5f06b..d155cea 100644
--- a/src/gallium/auxiliary/cso_cache/cso_context.c
+++ b/src/gallium/auxiliary/cso_cache/cso_context.c
@@ -48,65 +48,6 @@
#include "cso_context.h"
-/**
- * Info related to samplers and sampler views.
- * We have one of these for fragment samplers and another for vertex samplers.
- */
-struct sampler_info
-{
- struct {
- void *samplers[PIPE_MAX_SAMPLERS];
- unsigned nr_samplers;
- } hw;
-
- void *samplers[PIPE_MAX_SAMPLERS];
- unsigned nr_samplers;
-
- void *samplers_saved[PIPE_MAX_SAMPLERS];
- unsigned nr_samplers_saved;
-
- struct pipe_sampler_view *views[PIPE_MAX_SAMPLERS];
- unsigned nr_views;
-
- struct pipe_sampler_view *views_saved[PIPE_MAX_SAMPLERS];
- unsigned nr_views_saved;
-};
-
-
-
-struct cso_context {
- struct pipe_context *pipe;
- struct cso_cache *cache;
-
- struct sampler_info fragment_samplers;
- struct sampler_info vertex_samplers;
-
- uint nr_vertex_buffers;
- struct pipe_vertex_buffer vertex_buffers[PIPE_MAX_ATTRIBS];
-
- uint nr_vertex_buffers_saved;
- struct pipe_vertex_buffer vertex_buffers_saved[PIPE_MAX_ATTRIBS];
-
- /** Current and saved state.
- * The saved state is used as a 1-deep stack.
- */
- void *blend, *blend_saved;
- void *depth_stencil, *depth_stencil_saved;
- void *rasterizer, *rasterizer_saved;
- void *fragment_shader, *fragment_shader_saved, *geometry_shader;
- void *vertex_shader, *vertex_shader_saved, *geometry_shader_saved;
- void *velements, *velements_saved;
-
- struct pipe_clip_state clip;
- struct pipe_clip_state clip_saved;
-
- struct pipe_framebuffer_state fb, fb_saved;
- struct pipe_viewport_state vp, vp_saved;
- struct pipe_blend_color blend_color;
- unsigned sample_mask;
- struct pipe_stencil_ref stencil_ref, stencil_ref_saved;
-};
-
static boolean delete_blend_state(struct cso_context *ctx, void *state)
{
diff --git a/src/gallium/auxiliary/cso_cache/cso_context.h b/src/gallium/auxiliary/cso_cache/cso_context.h
index 00edc9f..6bc6564 100644
--- a/src/gallium/auxiliary/cso_cache/cso_context.h
+++ b/src/gallium/auxiliary/cso_cache/cso_context.h
@@ -38,7 +38,62 @@
extern "C" {
#endif
-struct cso_context;
+/**
+ * Info related to samplers and sampler views.
+ * We have one of these for fragment samplers and another for vertex samplers.
+ */
+struct sampler_info
+{
+ struct {
+ void *samplers[PIPE_MAX_SAMPLERS];
+ unsigned nr_samplers;
+ } hw;
+
+ void *samplers[PIPE_MAX_SAMPLERS];
+ unsigned nr_samplers;
+
+ void *samplers_saved[PIPE_MAX_SAMPLERS];
+ unsigned nr_samplers_saved;
+
+ struct pipe_sampler_view *views[PIPE_MAX_SAMPLERS];
+ unsigned nr_views;
+
+ struct pipe_sampler_view *views_saved[PIPE_MAX_SAMPLERS];
+ unsigned nr_views_saved;
+};
+
+struct cso_context {
+ struct pipe_context *pipe;
+ struct cso_cache *cache;
+
+ struct sampler_info fragment_samplers;
+ struct sampler_info vertex_samplers;
+
+ uint nr_vertex_buffers;
+ struct pipe_vertex_buffer vertex_buffers[PIPE_MAX_ATTRIBS];
+
+ uint nr_vertex_buffers_saved;
+ struct pipe_vertex_buffer vertex_buffers_saved[PIPE_MAX_ATTRIBS];
+
+ /** Current and saved state.
+ * The saved state is used as a 1-deep stack.
+ */
+ void *blend, *blend_saved;
+ void *depth_stencil, *depth_stencil_saved;
+ void *rasterizer, *rasterizer_saved;
+ void *fragment_shader, *fragment_shader_saved, *geometry_shader;
+ void *vertex_shader, *vertex_shader_saved, *geometry_shader_saved;
+ void *velements, *velements_saved;
+
+ struct pipe_clip_state clip;
+ struct pipe_clip_state clip_saved;
+
+ struct pipe_framebuffer_state fb, fb_saved;
+ struct pipe_viewport_state vp, vp_saved;
+ struct pipe_blend_color blend_color;
+ unsigned sample_mask;
+ struct pipe_stencil_ref stencil_ref, stencil_ref_saved;
+};
struct cso_context *cso_create_context( struct pipe_context *pipe );
diff --git a/src/mesa/state_tracker/st_atom_sampler.c b/src/mesa/state_tracker/st_atom_sampler.c
index 06024ad..43b67bc 100644
--- a/src/mesa/state_tracker/st_atom_sampler.c
+++ b/src/mesa/state_tracker/st_atom_sampler.c
@@ -204,6 +204,7 @@ update_vertex_samplers(struct st_context *st)
struct pipe_sampler_state *sampler = st->state.vertex_samplers + su;
if (vprog->Base.SamplersUsed & (1 << su)) {
+ if (st->cso_context->vertex_samplers.samplers[su] != sampler) {
GLuint texUnit;
texUnit = vprog->Base.SamplerUnits[su];
@@ -212,11 +213,18 @@ update_vertex_samplers(struct st_context *st)
st->state.num_vertex_samplers = su + 1;
+ //fprintf(stderr, "%s su=%u non-null %p [%p]\n", __FUNCTION__, su, sampler, st->cso_context->vertex_samplers.samplers[su]);
cso_single_vertex_sampler(st->cso_context, su, sampler);
} else {
+ st->state.num_vertex_samplers = su + 1;
+ }
+ } else {
+ if (st->cso_context->vertex_samplers.samplers[su] != NULL) {
+ //fprintf(stderr, "%s su=%u null\n", __FUNCTION__, su);
cso_single_vertex_sampler(st->cso_context, su, NULL);
}
}
+ }
cso_single_vertex_sampler_done(st->cso_context);
}
@@ -234,6 +242,7 @@ update_fragment_samplers(struct st_context *st)
if (fprog->Base.SamplersUsed & (1 << su)) {
+ if (st->cso_context->fragment_samplers.samplers[su] != sampler) {
GLuint texUnit;
texUnit = fprog->Base.SamplerUnits[su];
@@ -242,14 +251,19 @@ update_fragment_samplers(struct st_context *st)
st->state.num_samplers = su + 1;
- /*printf("%s su=%u non-null\n", __FUNCTION__, su);*/
+ // fprintf(stderr, "%s su=%u non-null %p [%p]\n", __FUNCTION__, su, sampler, st->cso_context->fragment_samplers.samplers[su]);
cso_single_sampler(st->cso_context, su, sampler);
+ } else {
+ st->state.num_samplers = su + 1;
+ }
}
else {
- /*printf("%s su=%u null\n", __FUNCTION__, su);*/
+ if (st->cso_context->fragment_samplers.samplers[su] != NULL) {
+ // fprintf(stderr, "%s su=%u null\n", __FUNCTION__, su);
cso_single_sampler(st->cso_context, su, NULL);
}
}
+ }
cso_single_sampler_done(st->cso_context);
}
--
1.7.5.1
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
