On 02/15/2012 07:01 PM, Eric Anholt wrote:
None of the consumers of this state will be called while the fs or vs
is in place, and the update_texture_state() call will get re-called on
fs/vs change, so it should be safe to skip the computation. Improves
the performance of a VS state change microbenchmark by 1.60186% +/-
0.443318% (n=20).
---
src/mesa/main/texstate.c | 117 +++++++++++++++++++++++++++------------------
1 files changed, 70 insertions(+), 47 deletions(-)
This looks good. Minor comment below.
diff --git a/src/mesa/main/texstate.c b/src/mesa/main/texstate.c
index 187ec9c..2dbf530 100644
--- a/src/mesa/main/texstate.c
+++ b/src/mesa/main/texstate.c
@@ -476,6 +476,68 @@ update_tex_combine(struct gl_context *ctx, struct
gl_texture_unit *texUnit)
}
}
It would be good if the following two functions had comments
indicating that the _ReallyEnabled and _EnabledCoordUnits field must
have already been computed first (as well as any other dependencies).
+static void
+update_fffs_texture_state(struct gl_context *ctx)
+{
+ GLuint unit;
+
+ ctx->Texture._EnabledUnits = 0;
+ ctx->Texture._EnabledCoordUnits = 0;
+
+ for (unit = 0; unit< ctx->Const.MaxCombinedTextureImageUnits; unit++) {
+ struct gl_texture_unit *texUnit =&ctx->Texture.Unit[unit];
+
+ if (!texUnit->_ReallyEnabled)
+ continue;
+
+ ctx->Texture._EnabledUnits |= (1<< unit);
+ ctx->Texture._EnabledCoordUnits |= (1<< unit);
+
+ update_tex_combine(ctx, texUnit);
+ }
+}
+
+static void
+update_ffvs_texture_state(struct gl_context *ctx)
+{
+ GLuint unit;
+
+ ctx->Texture._GenFlags = 0x0;
+ ctx->Texture._TexMatEnabled = 0x0;
+ ctx->Texture._TexGenEnabled = 0x0;
+
+ /* Setup texgen for those texture coordinate sets that are in use */
+ for (unit = 0; unit< ctx->Const.MaxTextureCoordUnits; unit++) {
+ struct gl_texture_unit *texUnit =&ctx->Texture.Unit[unit];
+
+ texUnit->_GenFlags = 0x0;
+
+ if (!(ctx->Texture._EnabledCoordUnits& (1<< unit)))
+ continue;
+
+ if (texUnit->TexGenEnabled) {
+ if (texUnit->TexGenEnabled& S_BIT) {
+ texUnit->_GenFlags |= texUnit->GenS._ModeBit;
+ }
+ if (texUnit->TexGenEnabled& T_BIT) {
+ texUnit->_GenFlags |= texUnit->GenT._ModeBit;
+ }
+ if (texUnit->TexGenEnabled& R_BIT) {
+ texUnit->_GenFlags |= texUnit->GenR._ModeBit;
+ }
+ if (texUnit->TexGenEnabled& Q_BIT) {
+ texUnit->_GenFlags |= texUnit->GenQ._ModeBit;
+ }
+
+ ctx->Texture._TexGenEnabled |= ENABLE_TEXGEN(unit);
+ ctx->Texture._GenFlags |= texUnit->_GenFlags;
+ }
+
+ ASSERT(unit< Elements(ctx->TextureMatrixStack));
+ if (ctx->TextureMatrixStack[unit].Top->type != MATRIX_IDENTITY)
+ ctx->Texture._TexMatEnabled |= ENABLE_TEXMAT(unit);
+ }
+}
/**
* \note This routine refers to derived texture matrix values to
@@ -491,7 +553,6 @@ update_texture_state( struct gl_context *ctx )
GLuint unit;
struct gl_program *fprog = NULL;
struct gl_program *vprog = NULL;
- GLbitfield enabledFragUnits = 0x0;
if (ctx->Shader.CurrentVertexProgram&&
ctx->Shader.CurrentVertexProgram->LinkStatus) {
@@ -518,11 +579,6 @@ update_texture_state( struct gl_context *ctx )
/* TODO: only set this if there are actual changes */
ctx->NewState |= _NEW_TEXTURE;
- ctx->Texture._EnabledUnits = 0x0;
- ctx->Texture._GenFlags = 0x0;
- ctx->Texture._TexMatEnabled = 0x0;
- ctx->Texture._TexGenEnabled = 0x0;
-
/*
* Update texture unit state.
*/
@@ -601,59 +657,26 @@ update_texture_state( struct gl_context *ctx )
continue;
}
}
-
- /* if we get here, we know this texture unit is enabled */
-
- ctx->Texture._EnabledUnits |= (1<< unit);
-
- if (enabledFragTargets)
- enabledFragUnits |= (1<< unit);
-
- update_tex_combine(ctx, texUnit);
}
/* Determine which texture coordinate sets are actually needed */
if (fprog) {
const GLuint coordMask = (1<< MAX_TEXTURE_COORD_UNITS) - 1;
+ /* Note that this gets consumed by update_ffvs_texture_state(). */
ctx->Texture._EnabledCoordUnits
= (fprog->InputsRead>> FRAG_ATTRIB_TEX0)& coordMask;
}
else {
- ctx->Texture._EnabledCoordUnits = enabledFragUnits;
+ update_fffs_texture_state(ctx);
}
- /* Setup texgen for those texture coordinate sets that are in use */
- for (unit = 0; unit< ctx->Const.MaxTextureCoordUnits; unit++) {
- struct gl_texture_unit *texUnit =&ctx->Texture.Unit[unit];
-
- texUnit->_GenFlags = 0x0;
-
- if (!(ctx->Texture._EnabledCoordUnits& (1<< unit)))
- continue;
-
- if (texUnit->TexGenEnabled) {
- if (texUnit->TexGenEnabled& S_BIT) {
- texUnit->_GenFlags |= texUnit->GenS._ModeBit;
- }
- if (texUnit->TexGenEnabled& T_BIT) {
- texUnit->_GenFlags |= texUnit->GenT._ModeBit;
- }
- if (texUnit->TexGenEnabled& R_BIT) {
- texUnit->_GenFlags |= texUnit->GenR._ModeBit;
- }
- if (texUnit->TexGenEnabled& Q_BIT) {
- texUnit->_GenFlags |= texUnit->GenQ._ModeBit;
- }
-
- ctx->Texture._TexGenEnabled |= ENABLE_TEXGEN(unit);
- ctx->Texture._GenFlags |= texUnit->_GenFlags;
- }
-
- ASSERT(unit< Elements(ctx->TextureMatrixStack));
- if (ctx->TextureMatrixStack[unit].Top->type != MATRIX_IDENTITY)
- ctx->Texture._TexMatEnabled |= ENABLE_TEXMAT(unit);
- }
+ /* Skip texgen and texmat computed state setup in the presence of a vertex
+ * program. Those state flags are only used in the case of fixed function
+ * vertex shading, in the tnl pipeline or ff_vertex_shader.
+ */
+ if (!vprog)
+ update_ffvs_texture_state(ctx);
}
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
