diff options
Diffstat (limited to 'src/mesa/drivers/dri/r200/r200_tcl.c')
| -rw-r--r-- | src/mesa/drivers/dri/r200/r200_tcl.c | 65 |
1 files changed, 56 insertions, 9 deletions
diff --git a/src/mesa/drivers/dri/r200/r200_tcl.c b/src/mesa/drivers/dri/r200/r200_tcl.c index ca9a8dbf8c..0f5e501c1a 100644 --- a/src/mesa/drivers/dri/r200/r200_tcl.c +++ b/src/mesa/drivers/dri/r200/r200_tcl.c @@ -109,7 +109,7 @@ static GLboolean discrete_prim[0x10] = { #define ELT_INIT(prim, hw_prim) \ r200TclPrimitive( ctx, prim, hw_prim | R200_VF_PRIM_WALK_IND ) -#define GET_MESA_ELTS() rmesa->tcl.Elts +#define GET_MESA_ELTS() TNL_CONTEXT(ctx)->vb.Elts /* Don't really know how many elts will fit in what's left of cmdbuf, @@ -156,8 +156,6 @@ static GLushort *r200AllocElts( r200ContextPtr rmesa, GLuint nr ) if (rmesa->radeon.dma.flush) rmesa->radeon.dma.flush( rmesa->radeon.glCtx ); - rcommonEnsureCmdBufSpace(&rmesa->radeon, AOS_BUFSZ(rmesa->radeon.tcl.aos_count), __FUNCTION__); - r200EmitAOS( rmesa, rmesa->radeon.tcl.aos_count, 0 ); @@ -187,9 +185,6 @@ static void r200EmitPrim( GLcontext *ctx, r200TclPrimitive( ctx, prim, hwprim ); // fprintf(stderr,"Emit prim %d\n", rmesa->radeon.tcl.aos_count); - rcommonEnsureCmdBufSpace( &rmesa->radeon, - AOS_BUFSZ(rmesa->radeon.tcl.aos_count) + - rmesa->radeon.hw.max_state_size + VBUF_BUFSZ, __FUNCTION__ ); r200EmitAOS( rmesa, rmesa->radeon.tcl.aos_count, @@ -206,6 +201,7 @@ static void r200EmitPrim( GLcontext *ctx, r200EmitPrim( ctx, prim, hwprim, start, count ); \ (void) rmesa; } while (0) +#define MAX_CONVERSION_SIZE 40 /* Try & join small primitives */ #if 0 @@ -368,6 +364,58 @@ r200ComputeFogBlendFactor( GLcontext *ctx, GLfloat fogcoord ) } } +/** + * Predict total emit size for next rendering operation so there is no flush in middle of rendering + * Prediction has to aim towards the best possible value that is worse than worst case scenario + */ +static void r200EnsureEmitSize( GLcontext * ctx , GLubyte* vimap_rev ) +{ + r200ContextPtr rmesa = R200_CONTEXT(ctx); + TNLcontext *tnl = TNL_CONTEXT(ctx); + struct vertex_buffer *VB = &tnl->vb; + GLuint space_required; + GLuint nr_aos = 0; + int i; + /* predict number of aos to emit */ + for (i = 0; i < 15; ++i) + { + if (vimap_rev[i] != 255) + { + ++nr_aos; + } + } + + { + /* count the prediction for state size */ + space_required = radeonCountStateEmitSize( &rmesa->radeon ); + /* vtx may be changed in r200EmitArrays so account for it if not dirty */ + if (!rmesa->hw.vtx.dirty) + space_required += rmesa->hw.vtx.check(rmesa->radeon.glCtx, &rmesa->hw.vtx); + /* predict size for elements */ + for (i = 0; i < VB->PrimitiveCount; ++i) + { + if (!VB->Primitive[i].count) + continue; + /* If primitive.count is less than MAX_CONVERSION_SIZE + rendering code may decide convert to elts. + In that case we have to make pessimistic prediction. + and use larger of 2 paths. */ + const GLuint elts = ELTS_BUFSZ(nr_aos); + const GLuint index = INDEX_BUFSZ; + const GLuint vbuf = VBUF_BUFSZ; + if ( (!VB->Elts && VB->Primitive[i].count >= MAX_CONVERSION_SIZE) + || vbuf > index + elts) + space_required += vbuf; + else + space_required += index + elts; + space_required += AOS_BUFSZ(nr_aos); + } + space_required += SCISSOR_BUFSZ; + } + /* flush the buffer in case we need more than is left. */ + rcommonEnsureCmdBufSpace(&rmesa->radeon, space_required, __FUNCTION__); +} + /**********************************************************************/ /* Render pipeline stage */ @@ -482,10 +530,9 @@ static GLboolean r200_run_tcl_render( GLcontext *ctx, /* Do the actual work: */ radeonReleaseArrays( ctx, ~0 /* stage->changed_inputs */ ); + r200EnsureEmitSize( ctx, vimap_rev ); r200EmitArrays( ctx, vimap_rev ); - rmesa->tcl.Elts = VB->Elts; - for (i = 0 ; i < VB->PrimitiveCount ; i++) { GLuint prim = _tnl_translate_prim(&VB->Primitive[i]); @@ -495,7 +542,7 @@ static GLboolean r200_run_tcl_render( GLcontext *ctx, if (!length) continue; - if (rmesa->tcl.Elts) + if (VB->Elts) r200EmitEltPrimitive( ctx, start, start+length, prim ); else r200EmitPrimitive( ctx, start, start+length, prim ); |