summaryrefslogtreecommitdiff
path: root/src/mesa/drivers/dri/r200/r200_tcl.c
diff options
context:
space:
mode:
Diffstat (limited to 'src/mesa/drivers/dri/r200/r200_tcl.c')
-rw-r--r--src/mesa/drivers/dri/r200/r200_tcl.c143
1 files changed, 100 insertions, 43 deletions
diff --git a/src/mesa/drivers/dri/r200/r200_tcl.c b/src/mesa/drivers/dri/r200/r200_tcl.c
index 99aecfe1e9..c702910ef2 100644
--- a/src/mesa/drivers/dri/r200/r200_tcl.c
+++ b/src/mesa/drivers/dri/r200/r200_tcl.c
@@ -51,6 +51,8 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
#include "r200_swtcl.h"
#include "r200_maos.h"
+#include "radeon_common_context.h"
+
#define HAVE_POINTS 1
@@ -109,7 +111,7 @@ 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,
@@ -123,7 +125,7 @@ static GLboolean discrete_prim[0x10] = {
#define RESET_STIPPLE() do { \
R200_STATECHANGE( rmesa, lin ); \
- r200EmitState( rmesa ); \
+ radeonEmitState(&rmesa->radeon); \
} while (0)
#define AUTO_STIPPLE( mode ) do { \
@@ -134,7 +136,7 @@ static GLboolean discrete_prim[0x10] = {
else \
rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] &= \
~R200_LINE_PATTERN_AUTO_RESET; \
- r200EmitState( rmesa ); \
+ radeonEmitState(&rmesa->radeon); \
} while (0)
@@ -142,27 +144,24 @@ static GLboolean discrete_prim[0x10] = {
static GLushort *r200AllocElts( r200ContextPtr rmesa, GLuint nr )
{
- if (rmesa->dma.flush == r200FlushElts &&
- rmesa->store.cmd_used + nr*2 < R200_CMD_BUF_SZ) {
+ if (rmesa->radeon.dma.flush == r200FlushElts &&
+ rmesa->tcl.elt_used + nr*2 < R200_ELT_BUF_SZ) {
- GLushort *dest = (GLushort *)(rmesa->store.cmd_buf +
- rmesa->store.cmd_used);
+ GLushort *dest = (GLushort *)(rmesa->radeon.tcl.elt_dma_bo->ptr +
+ rmesa->radeon.tcl.elt_dma_offset + rmesa->tcl.elt_used);
- rmesa->store.cmd_used += nr*2;
+ rmesa->tcl.elt_used += nr*2;
return dest;
}
else {
- if (rmesa->dma.flush)
- rmesa->dma.flush( rmesa );
-
- r200EnsureCmdBufSpace( rmesa, AOS_BUFSZ(rmesa->tcl.nr_aos_components) +
- rmesa->hw.max_state_size + ELTS_BUFSZ(nr) );
+ if (rmesa->radeon.dma.flush)
+ rmesa->radeon.dma.flush( rmesa->radeon.glCtx );
r200EmitAOS( rmesa,
- rmesa->tcl.aos_components,
- rmesa->tcl.nr_aos_components, 0 );
+ rmesa->radeon.tcl.aos_count, 0 );
+ r200EmitMaxVtxIndex(rmesa, rmesa->radeon.tcl.aos[0].count);
return r200AllocEltsOpenEnded( rmesa, rmesa->tcl.hw_primitive, nr );
}
}
@@ -188,13 +187,11 @@ static void r200EmitPrim( GLcontext *ctx,
r200ContextPtr rmesa = R200_CONTEXT( ctx );
r200TclPrimitive( ctx, prim, hwprim );
- r200EnsureCmdBufSpace( rmesa, AOS_BUFSZ(rmesa->tcl.nr_aos_components) +
- rmesa->hw.max_state_size + VBUF_BUFSZ );
+ // fprintf(stderr,"Emit prim %d\n", rmesa->radeon.tcl.aos_count);
r200EmitAOS( rmesa,
- rmesa->tcl.aos_components,
- rmesa->tcl.nr_aos_components,
- start );
+ rmesa->radeon.tcl.aos_count,
+ start );
/* Why couldn't this packet have taken an offset param?
*/
@@ -207,6 +204,7 @@ 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
@@ -369,6 +367,66 @@ 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 GLuint 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 state_size;
+ 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 = 0;
+ state_size = radeonCountStateEmitSize( &rmesa->radeon );
+ /* vtx may be changed in r200EmitArrays so account for it if not dirty */
+ if (!rmesa->hw.vtx.dirty)
+ state_size += 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);
+ }
+ }
+
+ radeon_print(RADEON_RENDER,RADEON_VERBOSE,
+ "%s space %u, aos %d\n",
+ __func__, space_required, AOS_BUFSZ(nr_aos) );
+ /* flush the buffer in case we need more than is left. */
+ if (rcommonEnsureCmdBufSpace(&rmesa->radeon, space_required + state_size, __FUNCTION__))
+ return space_required + radeonCountStateEmitSize( &rmesa->radeon );
+ else
+ return space_required + state_size;
+}
+
/**********************************************************************/
/* Render pipeline stage */
@@ -394,19 +452,19 @@ static GLboolean r200_run_tcl_render( GLcontext *ctx,
/* TODO: separate this from the swtnl pipeline
*/
- if (rmesa->TclFallback)
+ if (rmesa->radeon.TclFallback)
return GL_TRUE; /* fallback to software t&l */
- if (R200_DEBUG & DEBUG_PRIMS)
- fprintf(stderr, "%s\n", __FUNCTION__);
+ radeon_print(RADEON_RENDER, RADEON_NORMAL, "%s\n", __FUNCTION__);
if (VB->Count == 0)
return GL_FALSE;
/* Validate state:
*/
- if (rmesa->NewGLState)
- r200ValidateState( ctx );
+ if (rmesa->radeon.NewGLState)
+ if (!r200ValidateState( ctx ))
+ return GL_TRUE; /* fallback to sw t&l */
if (!ctx->VertexProgram._Enabled) {
/* NOTE: inputs != tnl->render_inputs - these are the untransformed
@@ -481,11 +539,11 @@ static GLboolean r200_run_tcl_render( GLcontext *ctx,
/* Do the actual work:
*/
- r200ReleaseArrays( ctx, ~0 /* stage->changed_inputs */ );
+ radeonReleaseArrays( ctx, ~0 /* stage->changed_inputs */ );
+ GLuint emit_end = r200EnsureEmitSize( ctx, vimap_rev )
+ + rmesa->radeon.cmdbuf.cs->cdw;
r200EmitArrays( ctx, vimap_rev );
- rmesa->tcl.Elts = VB->Elts;
-
for (i = 0 ; i < VB->PrimitiveCount ; i++)
{
GLuint prim = _tnl_translate_prim(&VB->Primitive[i]);
@@ -495,11 +553,14 @@ 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 );
}
+ if ( emit_end < rmesa->radeon.cmdbuf.cs->cdw )
+ WARN_ONCE("Rendering was %d commands larger than predicted size."
+ " We might overflow command buffer.\n", rmesa->radeon.cmdbuf.cs->cdw - emit_end);
return GL_FALSE; /* finished the pipe */
}
@@ -545,7 +606,7 @@ static void transition_to_swtnl( GLcontext *ctx )
tnl->Driver.NotifyMaterialChange =
_mesa_validate_all_lighting_tables;
- r200ReleaseArrays( ctx, ~0 );
+ radeonReleaseArrays( ctx, ~0 );
/* Still using the D3D based hardware-rasterizer from the radeon;
* need to put the card into D3D mode to make it work:
@@ -565,15 +626,11 @@ static void transition_to_hwtnl( GLcontext *ctx )
tnl->Driver.NotifyMaterialChange = r200UpdateMaterial;
- if ( rmesa->dma.flush )
- rmesa->dma.flush( rmesa );
+ if ( rmesa->radeon.dma.flush )
+ rmesa->radeon.dma.flush( rmesa->radeon.glCtx );
- rmesa->dma.flush = NULL;
+ rmesa->radeon.dma.flush = NULL;
- if (rmesa->swtcl.indexed_verts.buf)
- r200ReleaseDmaRegion( rmesa, &rmesa->swtcl.indexed_verts,
- __FUNCTION__ );
-
R200_STATECHANGE( rmesa, vap );
rmesa->hw.vap.cmd[VAP_SE_VAP_CNTL] |= R200_VAP_TCL_ENABLE;
rmesa->hw.vap.cmd[VAP_SE_VAP_CNTL] &= ~R200_VAP_FORCE_W_TO_ONE;
@@ -594,7 +651,7 @@ static void transition_to_hwtnl( GLcontext *ctx )
rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL] &= ~(R200_VTX_XY_FMT|R200_VTX_Z_FMT);
rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL] |= R200_VTX_W0_FMT;
- if (R200_DEBUG & DEBUG_FALLBACKS)
+ if (R200_DEBUG & RADEON_FALLBACKS)
fprintf(stderr, "R200 end tcl fallback\n");
}
@@ -631,21 +688,21 @@ static char *getFallbackString(GLuint bit)
void r200TclFallback( GLcontext *ctx, GLuint bit, GLboolean mode )
{
r200ContextPtr rmesa = R200_CONTEXT(ctx);
- GLuint oldfallback = rmesa->TclFallback;
+ GLuint oldfallback = rmesa->radeon.TclFallback;
if (mode) {
- rmesa->TclFallback |= bit;
+ rmesa->radeon.TclFallback |= bit;
if (oldfallback == 0) {
- if (R200_DEBUG & DEBUG_FALLBACKS)
+ if (R200_DEBUG & RADEON_FALLBACKS)
fprintf(stderr, "R200 begin tcl fallback %s\n",
getFallbackString( bit ));
transition_to_swtnl( ctx );
}
}
else {
- rmesa->TclFallback &= ~bit;
+ rmesa->radeon.TclFallback &= ~bit;
if (oldfallback == bit) {
- if (R200_DEBUG & DEBUG_FALLBACKS)
+ if (R200_DEBUG & RADEON_FALLBACKS)
fprintf(stderr, "R200 end tcl fallback %s\n",
getFallbackString( bit ));
transition_to_hwtnl( ctx );
generated by cgit v1.2.3 (git 2.39.1) at 2024-05-29 10:53:50 +0000