GLushort *radeonAllocEltsOpenEnded( r100ContextPtr rmesa,
GLuint vertex_format,
GLuint primitive,
GLuint min_nr )
{
GLushort *retval;
int align_min_nr;
BATCH_LOCALS(&rmesa->radeon);
if (RADEON_DEBUG & RADEON_IOCTL)
fprintf(stderr, "%s %d prim %x\n", __FUNCTION__, min_nr, primitive);
assert((primitive & RADEON_CP_VC_CNTL_PRIM_WALK_IND));
radeonEmitState(&rmesa->radeon);
radeonEmitScissor(rmesa);
rmesa->tcl.elt_cmd_start = rmesa->radeon.cmdbuf.cs->cdw;
/* round up min_nr to align the state */
align_min_nr = (min_nr + 1) & ~1;
#if RADEON_OLD_PACKETS
BEGIN_BATCH_NO_AUTOSTATE(2+ELTS_BUFSZ(align_min_nr)/4);
OUT_BATCH_PACKET3_CLIP(RADEON_CP_PACKET3_3D_RNDR_GEN_INDX_PRIM, 0);
OUT_BATCH(rmesa->ioctl.vertex_offset);
OUT_BATCH(rmesa->ioctl.vertex_max);
OUT_BATCH(vertex_format);
OUT_BATCH(primitive |
RADEON_CP_VC_CNTL_PRIM_WALK_IND |
RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA |
RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE);
#else
BEGIN_BATCH_NO_AUTOSTATE(ELTS_BUFSZ(align_min_nr)/4);
OUT_BATCH_PACKET3_CLIP(RADEON_CP_PACKET3_DRAW_INDX, 0);
OUT_BATCH(vertex_format);
OUT_BATCH(primitive |
RADEON_CP_VC_CNTL_PRIM_WALK_IND |
RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA |
RADEON_CP_VC_CNTL_MAOS_ENABLE |
RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE);
#endif
rmesa->tcl.elt_cmd_offset = rmesa->radeon.cmdbuf.cs->cdw;
rmesa->tcl.elt_used = min_nr;
retval = (GLushort *)(rmesa->radeon.cmdbuf.cs->packets + rmesa->tcl.elt_cmd_offset);
if (RADEON_DEBUG & RADEON_RENDER)
fprintf(stderr, "%s: header prim %x \n",
__FUNCTION__, primitive);
assert(!rmesa->radeon.dma.flush);
rmesa->radeon.glCtx->Driver.NeedFlush |= FLUSH_STORED_VERTICES;
rmesa->radeon.dma.flush = radeonFlushElts;
return retval;
}
static GLushort *r200AllocElts( r200ContextPtr rmesa, GLuint nr )
{
if (rmesa->dma.flush == r200FlushElts &&
rmesa->store.cmd_used + nr*2 < R200_CMD_BUF_SZ) {
GLushort *dest = (GLushort *)(rmesa->store.cmd_buf +
rmesa->store.cmd_used);
rmesa->store.cmd_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) );
r200EmitAOS( rmesa,
rmesa->tcl.aos_components,
rmesa->tcl.nr_aos_components, 0 );
return r200AllocEltsOpenEnded( rmesa, rmesa->tcl.hw_primitive, nr );
}
}
/**
* 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( struct gl_context * 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 elt_count =(VB->Primitive[i].count/GET_MAX_HW_ELTS() + 1);
const GLuint elts = ELTS_BUFSZ(nr_aos) * elt_count;
const GLuint index = INDEX_BUFSZ * elt_count;
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, __func__))
return space_required + radeonCountStateEmitSize( &rmesa->radeon );
else
return space_required + state_size;
}
static GLushort *radeonAllocElts( radeonContextPtr rmesa, GLuint nr )
{
if (rmesa->dma.flush)
rmesa->dma.flush( rmesa );
radeonEnsureCmdBufSpace(rmesa, AOS_BUFSZ(rmesa->tcl.nr_aos_components) +
rmesa->hw.max_state_size + ELTS_BUFSZ(nr));
radeonEmitAOS( rmesa,
rmesa->tcl.aos_components,
rmesa->tcl.nr_aos_components, 0 );
return radeonAllocEltsOpenEnded( rmesa,
rmesa->tcl.vertex_format,
rmesa->tcl.hw_primitive, nr );
}
/**
* 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 radeonEnsureEmitSize( struct gl_context * ctx , GLuint inputs )
{
r100ContextPtr rmesa = R100_CONTEXT(ctx);
TNLcontext *tnl = TNL_CONTEXT(ctx);
struct vertex_buffer *VB = &tnl->vb;
GLuint space_required;
GLuint state_size;
GLuint nr_aos = 1; /* radeonEmitArrays does always emit one */
int i;
/* list of flags that are allocating aos object */
const GLuint flags_to_check[] = {
VERT_BIT_NORMAL,
VERT_BIT_COLOR0,
VERT_BIT_COLOR1,
VERT_BIT_FOG
};
/* predict number of aos to emit */
for (i=0; i < sizeof(flags_to_check)/sizeof(flags_to_check[0]); ++i)
{
if (inputs & flags_to_check[i])
++nr_aos;
}
for (i = 0; i < ctx->Const.MaxTextureUnits; ++i)
{
if (inputs & VERT_BIT_TEX(i))
++nr_aos;
}
{
/* count the prediction for state size */
space_required = 0;
state_size = radeonCountStateEmitSize( &rmesa->radeon );
/* tcl may be changed in radeonEmitArrays so account for it if not dirty */
if (!rmesa->hw.tcl.dirty)
state_size += rmesa->hw.tcl.check( rmesa->radeon.glCtx, &rmesa->hw.tcl );
/* 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 += VB->Primitive[i].count * 3;
space_required += AOS_BUFSZ(nr_aos);
}
space_required += SCISSOR_BUFSZ;
}
/* flush the buffer in case we need more than is left. */
if (rcommonEnsureCmdBufSpace(&rmesa->radeon, space_required, __FUNCTION__))
return space_required + radeonCountStateEmitSize( &rmesa->radeon );
else
return space_required + state_size;
}
