#include "r300_emit.h"
#include "r300_tex.h"
#include "r300_render.h"
+#include "main/simple_list.h"
#define EMIT_ATTR( ATTR, STYLE ) \
do { \
- rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].attrib = (ATTR); \
- rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].format = (STYLE); \
- rmesa->radeon.swtcl.vertex_attr_count++; \
+ rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].attrib = (ATTR); \
+ rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].format = (STYLE); \
+ rmesa->radeon.swtcl.vertex_attr_count++; \
} while (0)
#define EMIT_PAD( N ) \
GLuint InputsRead = 0;
GLuint OutputsWritten = 0;
int num_attrs = 0;
- GLuint fp_reads = ctx->FragmentProgram._Current->Base.InputsRead;
+ GLuint fp_reads = rmesa->selected_fp->InputsRead;
struct vertex_attribute *attrs = rmesa->vbuf.attribs;
+ radeon_print(RADEON_SWRENDER, RADEON_VERBOSE, "%s\n", __func__);
rmesa->swtcl.coloroffset = rmesa->swtcl.specoffset = 0;
rmesa->radeon.swtcl.vertex_attr_count = 0;
+ if (RADEON_DEBUG & RADEON_VERTS)
+ fprintf(stderr, "%s\n", __func__);
+
/* We always want non Ndc coords format */
VB->AttribPtr[VERT_ATTRIB_POS] = VB->ClipPtr;
}
if (ctx->Light.Enabled && ctx->Light.Model.TwoSide) {
- VB->AttribPtr[VERT_ATTRIB_GENERIC0] = VB->ColorPtr[1];
+ VB->AttribPtr[VERT_ATTRIB_GENERIC0] = VB->BackfaceColorPtr;
OutputsWritten |= 1 << VERT_RESULT_BFC0;
#if MESA_LITTLE_ENDIAN
EMIT_ATTR( _TNL_ATTRIB_GENERIC0, EMIT_4UB_4F_RGBA );
ADD_ATTR(VERT_ATTRIB_GENERIC0, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR2, SWIZZLE_XYZW, MASK_XYZW, 1);
#endif
if (fp_reads & FRAG_BIT_COL1) {
- VB->AttribPtr[VERT_ATTRIB_GENERIC1] = VB->SecondaryColorPtr[1];
+ VB->AttribPtr[VERT_ATTRIB_GENERIC1] = VB->BackfaceSecondaryColorPtr;
GLuint swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ONE);
OutputsWritten |= 1 << VERT_RESULT_BFC1;
#if MESA_LITTLE_ENDIAN
ADD_ATTR(VERT_ATTRIB_POINT_SIZE, R300_DATA_TYPE_FLOAT_1, SWTCL_OVM_POINT_SIZE, swiz, MASK_X, 0);
}
+ if (rmesa->selected_fp->wpos_attr != FRAG_ATTRIB_MAX) {
+ int tex_id = rmesa->selected_fp->wpos_attr - FRAG_ATTRIB_TEX0;
+
+ VB->AttribPtr[VERT_ATTRIB_TEX0 + tex_id] = VB->AttribPtr[VERT_ATTRIB_POS];
+ VB->AttribPtr[_TNL_ATTRIB_TEX0 + tex_id] = VB->AttribPtr[VERT_ATTRIB_POS];
+ RENDERINPUTS_SET(tnl->render_inputs_bitset, _TNL_ATTRIB_TEX0 + tex_id);
+ }
+
+ if (rmesa->selected_fp->fog_attr != FRAG_ATTRIB_MAX) {
+ int tex_id = rmesa->selected_fp->fog_attr - FRAG_ATTRIB_TEX0;
+
+ VB->AttribPtr[VERT_ATTRIB_TEX0 + tex_id] = VB->AttribPtr[VERT_ATTRIB_FOG];
+ VB->AttribPtr[_TNL_ATTRIB_TEX0 + tex_id] = VB->AttribPtr[VERT_ATTRIB_FOG];
+ RENDERINPUTS_SET(tnl->render_inputs_bitset, _TNL_ATTRIB_TEX0 + tex_id);
+ }
+
/**
* Sending only one texcoord component may lead to lock up,
* so for all textures always output 4 texcoord components to RS.
GLuint swiz, format, hw_format;
for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
if (fp_reads & FRAG_BIT_TEX(i)) {
- switch (VB->TexCoordPtr[i]->size) {
+ switch (VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]->size) {
case 1:
format = EMIT_1F;
hw_format = R300_DATA_TYPE_FLOAT_1;
}
}
- /* RS can't put fragment position on the pixel stack, so stuff it in texcoord if needed */
- if (fp_reads & FRAG_BIT_WPOS) {
- if (first_free_tex >= ctx->Const.MaxTextureUnits) {
- fprintf(stderr, "\tout of free texcoords to write w pos\n");
- _mesa_exit(-1);
- }
-
- InputsRead |= 1 << (VERT_ATTRIB_TEX0 + first_free_tex);
- OutputsWritten |= 1 << (VERT_RESULT_TEX0 + first_free_tex);
- EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_4F );
- ADD_ATTR(VERT_ATTRIB_POS, R300_DATA_TYPE_FLOAT_4, SWTCL_OVM_TEX(first_free_tex), SWIZZLE_XYZW, MASK_XYZW, 0);
- ++first_free_tex;
- }
-
- if (fp_reads & FRAG_BIT_FOGC) {
- if (first_free_tex >= ctx->Const.MaxTextureUnits) {
- fprintf(stderr, "\tout of free texcoords to write fog coordinate\n");
- _mesa_exit(-1);
- }
-
- InputsRead |= 1 << VERT_ATTRIB_FOG;
- OutputsWritten |= 1 << VERT_RESULT_FOGC;
- GLuint swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ZERO);
- EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1F );
- ADD_ATTR(VERT_ATTRIB_FOG, R300_DATA_TYPE_FLOAT_1, SWTCL_OVM_TEX(first_free_tex), swiz, MASK_XYZW, 0);
+ if (first_free_tex >= ctx->Const.MaxTextureUnits) {
+ fprintf(stderr, "\tout of free texcoords to write fog coordinate\n");
+ _mesa_exit(-1);
}
R300_NEWPRIM(rmesa);
{
r300ContextPtr rmesa = R300_CONTEXT(ctx);
GLuint InputsRead, OutputsWritten;
+ radeon_print(RADEON_SWRENDER, RADEON_TRACE, "%s\n", __func__);
r300ChooseSwtclVertexFormat(ctx, &InputsRead, &OutputsWritten);
r300SetupVAP(ctx, InputsRead, OutputsWritten);
rmesa->radeon.swtcl.vertex_size /= 4;
}
+static void r300_predict_emit_size( r300ContextPtr rmesa )
+{
+ if (!rmesa->radeon.swtcl.emit_prediction) {
+ const int vertex_size = 7;
+ const int prim_size = 3;
+ const int cache_flush_size = 4;
+ const int pre_emit_state = 4;
+ const int scissor_size = 3;
+ const int state_size = radeonCountStateEmitSize(&rmesa->radeon);
+
+ if (rcommonEnsureCmdBufSpace(&rmesa->radeon,
+ state_size + pre_emit_state + scissor_size
+ + vertex_size + prim_size + cache_flush_size * 2,
+ __FUNCTION__))
+ rmesa->radeon.swtcl.emit_prediction = radeonCountStateEmitSize(&rmesa->radeon);
+ else
+ rmesa->radeon.swtcl.emit_prediction = state_size;
+
+ rmesa->radeon.swtcl.emit_prediction += rmesa->radeon.cmdbuf.cs->cdw
+ + vertex_size + scissor_size + prim_size + cache_flush_size * 2 + pre_emit_state;
+ radeon_print(RADEON_SWRENDER, RADEON_VERBOSE,
+ "%s, size %d\n",
+ __func__, rmesa->radeon.cmdbuf.cs->cdw
+ + vertex_size + scissor_size + prim_size + cache_flush_size * 2 + pre_emit_state);
+ }
+}
+
static GLuint reduced_prim[] = {
GL_POINTS,
#define HAVE_POLYGONS 1
#define HAVE_ELTS 1
+static void* r300_alloc_verts(r300ContextPtr rmesa, GLuint n, GLuint size)
+{
+ void *rv;
+ do {
+ r300_predict_emit_size( rmesa );
+ rv = rcommonAllocDmaLowVerts( &rmesa->radeon, n, size * 4 );
+ } while (!rv);
+ return rv;
+}
+
#undef LOCAL_VARS
#undef ALLOC_VERTS
#define CTX_ARG r300ContextPtr rmesa
#define GET_VERTEX_DWORDS() rmesa->radeon.swtcl.vertex_size
-#define ALLOC_VERTS( n, size ) rcommonAllocDmaLowVerts( &rmesa->radeon, n, size * 4 )
+#define ALLOC_VERTS( n, size ) r300_alloc_verts(rmesa, n, size);
#define LOCAL_VARS \
r300ContextPtr rmesa = R300_CONTEXT(ctx); \
const char *r300verts = (char *)rmesa->radeon.swtcl.verts;
r300ContextPtr rmesa = R300_CONTEXT(ctx);
GLuint index = 0;
GLuint flags = ctx->_TriangleCaps;
+ radeon_print(RADEON_SWRENDER, RADEON_VERBOSE, "%s\n", __func__);
if (flags & DD_TRI_UNFILLED) index |= R300_UNFILLED_BIT;
}
}
-
void r300RenderStart(GLcontext *ctx)
{
+ radeon_print(RADEON_SWRENDER, RADEON_VERBOSE, "%s\n", __func__);
r300ContextPtr rmesa = R300_CONTEXT( ctx );
r300ChooseRenderState(ctx);
+
+ r300UpdateShaders(rmesa);
+
r300PrepareVertices(ctx);
r300ValidateBuffers(ctx);
- r300UpdateShaders(rmesa);
r300UpdateShaderStates(rmesa);
- r300EmitCacheFlush(rmesa);
/* investigate if we can put back flush optimisation if needed */
if (rmesa->radeon.dma.flush != NULL) {
static void r300RasterPrimitive( GLcontext *ctx, GLuint hwprim )
{
r300ContextPtr rmesa = R300_CONTEXT(ctx);
+ radeon_print(RADEON_SWRENDER, RADEON_TRACE, "%s\n", __func__);
if (rmesa->radeon.swtcl.hw_primitive != hwprim) {
R300_NEWPRIM( rmesa );
r300ContextPtr rmesa = R300_CONTEXT(ctx);
rmesa->radeon.swtcl.render_primitive = prim;
+ radeon_print(RADEON_SWRENDER, RADEON_TRACE, "%s\n", __func__);
if ((prim == GL_TRIANGLES) && (ctx->_TriangleCaps & DD_TRI_UNFILLED))
return;
void r300ResetLineStipple(GLcontext *ctx)
{
+ if (RADEON_DEBUG & RADEON_VERTS)
+ fprintf(stderr, "%s\n", __func__);
}
void r300InitSwtcl(GLcontext *ctx)
TNLcontext *tnl = TNL_CONTEXT(ctx);
r300ContextPtr rmesa = R300_CONTEXT(ctx);
static int firsttime = 1;
+ radeon_print(RADEON_SWRENDER, RADEON_NORMAL, "%s\n", __func__);
if (firsttime) {
init_rast_tab();
firsttime = 0;
}
+ rmesa->radeon.swtcl.emit_prediction = 0;
tnl->Driver.Render.Start = r300RenderStart;
tnl->Driver.Render.Finish = r300RenderFinish;
{
BATCH_LOCALS(&rmesa->radeon);
- if (RADEON_DEBUG & DEBUG_VERTS)
- fprintf(stderr, "%s: vertex_size %d, offset 0x%x \n",
+ radeon_print(RADEON_SWRENDER, RADEON_TRACE,
+ "%s: vertex_size %d, offset 0x%x \n",
__FUNCTION__, vertex_size, offset);
BEGIN_BATCH(7);
{
BATCH_LOCALS(&rmesa->radeon);
int type, num_verts;
+ if (RADEON_DEBUG & RADEON_VERTS)
+ fprintf(stderr, "%s\n", __func__);
type = r300PrimitiveType(rmesa, primitive);
num_verts = r300NumVerts(rmesa, vertex_nr, primitive);
void r300_swtcl_flush(GLcontext *ctx, uint32_t current_offset)
{
+ radeon_print(RADEON_SWRENDER, RADEON_TRACE, "%s\n", __func__);
r300ContextPtr rmesa = R300_CONTEXT(ctx);
- rcommonEnsureCmdBufSpace(&rmesa->radeon,
- rmesa->radeon.hw.max_state_size + (12*sizeof(int)),
- __FUNCTION__);
+ r300EmitCacheFlush(rmesa);
+
radeonEmitState(&rmesa->radeon);
- r300_emit_scissor(ctx);
+ r300_emit_scissor(ctx);
r300EmitVertexAOS(rmesa,
- rmesa->radeon.swtcl.vertex_size,
- rmesa->radeon.dma.current,
- current_offset);
+ rmesa->radeon.swtcl.vertex_size,
+ rmesa->radeon.swtcl.bo,
+ current_offset);
r300EmitVbufPrim(rmesa,
rmesa->radeon.swtcl.hw_primitive,
rmesa->radeon.swtcl.numverts);
r300EmitCacheFlush(rmesa);
+ if ( rmesa->radeon.swtcl.emit_prediction < 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 - rmesa->radeon.swtcl.emit_prediction );
+ rmesa->radeon.swtcl.emit_prediction = 0;
COMMIT_BATCH();
}