1 files changed, 324 insertions, 388 deletions

diff --git a/src/mesa/drivers/dri/r300/r300_swtcl.c b/src/mesa/drivers/dri/r300/r300_swtcl.c
index ba3621b16b..ee2c71e1a7 100644
--- a/src/mesa/drivers/dri/r300/r300_swtcl.c
+++ b/src/mesa/drivers/dri/r300/r300_swtcl.c
@@ -28,362 +28,264 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 /*
  * Authors:
  *   Dave Airlie <airlied@linux.ie>
+ *   Maciej Cencora <m.cencora@gmail.com>
  */
 
-/* derived from r200 swtcl path */
-
-
-
-#include "main/glheader.h"
-#include "main/mtypes.h"
-#include "main/colormac.h"
-#include "main/enums.h"
-#include "main/image.h"
-#include "main/imports.h"
-#include "main/light.h"
-#include "main/macros.h"
-
-#include "swrast/s_context.h"
-#include "swrast/s_fog.h"
-#include "swrast_setup/swrast_setup.h"
-#include "math/m_translate.h"
 #include "tnl/tnl.h"
-#include "tnl/t_context.h"
 #include "tnl/t_pipeline.h"
 
-#include "r300_context.h"
-#include "r300_swtcl.h"
 #include "r300_state.h"
-#include "r300_ioctl.h"
+#include "r300_swtcl.h"
 #include "r300_emit.h"
-#include "r300_mem.h"
-
-static void flush_last_swtcl_prim( r300ContextPtr rmesa  );
-
+#include "r300_tex.h"
+#include "r300_render.h"
+#include "main/simple_list.h"
 
-void r300EmitVertexAOS(r300ContextPtr rmesa, GLuint vertex_size, GLuint offset);
-void r300EmitVbufPrim(r300ContextPtr rmesa, GLuint primitive, GLuint vertex_nr);
 #define EMIT_ATTR( ATTR, STYLE )					\
 do {									\
-   rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].attrib = (ATTR);	\
-   rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].format = (STYLE);	\
-   rmesa->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 )							\
 do {									\
-   rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].attrib = 0;		\
-   rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].format = EMIT_PAD;	\
-   rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].offset = (N);		\
-   rmesa->swtcl.vertex_attr_count++;					\
+   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].attrib = 0;		\
+   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].format = EMIT_PAD;	\
+   rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].offset = (N);		\
+   rmesa->radeon.swtcl.vertex_attr_count++;					\
 } while (0)
 
-static void r300SetVertexFormat( GLcontext *ctx )
+#define ADD_ATTR(_attr, _format, _dst_loc, _swizzle, _write_mask, _normalize) \
+do { \
+	attrs[num_attrs].element = (_attr); \
+	attrs[num_attrs].data_type = (_format); \
+	attrs[num_attrs].dst_loc = (_dst_loc); \
+	attrs[num_attrs].swizzle = (_swizzle); \
+	attrs[num_attrs].write_mask = (_write_mask); \
+	attrs[num_attrs]._signed = 0; \
+	attrs[num_attrs].normalize = (_normalize); \
+	++num_attrs; \
+} while (0)
+
+void r300ChooseSwtclVertexFormat(GLcontext *ctx, GLuint *_InputsRead,  GLuint *_OutputsWritten)
 {
 	r300ContextPtr rmesa = R300_CONTEXT( ctx );
 	TNLcontext *tnl = TNL_CONTEXT(ctx);
 	struct vertex_buffer *VB = &tnl->vb;
-	DECLARE_RENDERINPUTS(index_bitset);
-	GLuint InputsRead = 0, OutputsWritten = 0;
-	int vap_fmt_1 = 0;
-	int offset = 0;
-	int vte = 0;
-	int fog_id;
-	GLint inputs[VERT_ATTRIB_MAX];
-	GLint tab[VERT_ATTRIB_MAX];
-	int swizzle[VERT_ATTRIB_MAX][4];
-	GLuint i, nr;
-	GLuint sz;
-
-	DECLARE_RENDERINPUTS(render_inputs_bitset);
-	RENDERINPUTS_COPY(render_inputs_bitset, tnl->render_inputs_bitset);
-	RENDERINPUTS_COPY( index_bitset, tnl->render_inputs_bitset );
-	RENDERINPUTS_COPY(rmesa->state.render_inputs_bitset, render_inputs_bitset);
-
-	vte = rmesa->hw.vte.cmd[1];
-	vte &= ~(R300_VTX_XY_FMT | R300_VTX_Z_FMT | R300_VTX_W0_FMT);
-	/* Important:
-	 */
-	if ( VB->NdcPtr != NULL ) {
-		VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
-		vte |= R300_VTX_XY_FMT | R300_VTX_Z_FMT;
-	}
-	else {
-		VB->AttribPtr[VERT_ATTRIB_POS] = VB->ClipPtr;
-		vte |= R300_VTX_W0_FMT;
-	}
-
-	assert( VB->AttribPtr[VERT_ATTRIB_POS] != NULL );
-	rmesa->swtcl.vertex_attr_count = 0;
-
-	/* EMIT_ATTR's must be in order as they tell t_vertex.c how to
-	 * build up a hardware vertex.
-	 */
-	if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_POS)) {
-		sz = VB->AttribPtr[VERT_ATTRIB_POS]->size;
-		InputsRead |= 1 << VERT_ATTRIB_POS;
-		OutputsWritten |= 1 << VERT_RESULT_HPOS;
-		EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_1F + sz - 1 );
-		offset = sz;
-	} else {
-		offset = 4;
-		EMIT_PAD(4 * sizeof(float));
-	}
-/*
-	if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_POINTSIZE )) {
-		EMIT_ATTR( _TNL_ATTRIB_POINTSIZE, EMIT_1F );
-		offset += 1;
-	}
-*/
-	if (RENDERINPUTS_TEST(index_bitset, _TNL_ATTRIB_COLOR0)) {
-		sz = VB->AttribPtr[VERT_ATTRIB_COLOR0]->size;
-	        rmesa->swtcl.coloroffset = offset;
+	int first_free_tex = 0;
+	GLuint InputsRead = 0;
+	GLuint OutputsWritten = 0;
+	int num_attrs = 0;
+	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;
+
+	/* Always write position vector */
+	InputsRead |= 1 << VERT_ATTRIB_POS;
+	OutputsWritten |= 1 << VERT_RESULT_HPOS;
+	EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_4F );
+	ADD_ATTR(VERT_ATTRIB_POS, R300_DATA_TYPE_FLOAT_4, SWTCL_OVM_POS, SWIZZLE_XYZW, MASK_XYZW, 0);
+	rmesa->swtcl.coloroffset = 4;
+
+	if (fp_reads & FRAG_BIT_COL0) {
 		InputsRead |= 1 << VERT_ATTRIB_COLOR0;
 		OutputsWritten |= 1 << VERT_RESULT_COL0;
-		EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_1F + sz - 1 );
-		offset += sz;
+#if MESA_LITTLE_ENDIAN
+		EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_RGBA );
+		ADD_ATTR(VERT_ATTRIB_COLOR0, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR0, SWIZZLE_XYZW, MASK_XYZW, 1);
+#else
+		EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_ABGR );
+		ADD_ATTR(VERT_ATTRIB_COLOR0, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR0, SWIZZLE_XYZW, MASK_XYZW, 1);
+#endif
 	}
 
-	rmesa->swtcl.specoffset = 0;
-	if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_COLOR1 )) {
-		sz = VB->AttribPtr[VERT_ATTRIB_COLOR1]->size;
-		rmesa->swtcl.specoffset = offset;
-		EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_1F + sz - 1 );
+	if (fp_reads & FRAG_BIT_COL1) {
+		GLuint swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ONE);
 		InputsRead |= 1 << VERT_ATTRIB_COLOR1;
 		OutputsWritten |= 1 << VERT_RESULT_COL1;
+#if MESA_LITTLE_ENDIAN
+		EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_4UB_4F_RGBA );
+		ADD_ATTR(VERT_ATTRIB_COLOR1, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR1, swiz, MASK_XYZW, 1);
+#else
+		EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_4UB_4F_ABGR );
+		ADD_ATTR(VERT_ATTRIB_COLOR1, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR1, swiz, MASK_XYZW, 1);
+#endif
+		rmesa->swtcl.specoffset = rmesa->swtcl.coloroffset + 1;
 	}
 
-	fog_id = -1;
-	if (RENDERINPUTS_TEST(index_bitset, _TNL_ATTRIB_FOG)) {
-		/* find first free tex coord slot */
-		if (RENDERINPUTS_TEST_RANGE( index_bitset, _TNL_FIRST_TEX, _TNL_LAST_TEX )) {
-			int i;
-			for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
-				if (!RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_TEX(i) )) {
-					fog_id = i;
-					break;
-				}
-			}
-		} else {
-			fog_id = 0;
+	if (ctx->Light.Enabled && ctx->Light.Model.TwoSide) {
+		VB->AttribPtr[VERT_ATTRIB_GENERIC0] = VB->ColorPtr[1];
+		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);
+#else
+		EMIT_ATTR( _TNL_ATTRIB_GENERIC0, EMIT_4UB_4F_ABGR );
+		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];
+			GLuint swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ONE);
+			OutputsWritten |= 1 << VERT_RESULT_BFC1;
+#if MESA_LITTLE_ENDIAN
+			EMIT_ATTR( _TNL_ATTRIB_GENERIC1, EMIT_4UB_4F_RGBA );
+			ADD_ATTR(VERT_ATTRIB_GENERIC1, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR3, swiz, MASK_XYZW, 1);
+#else
+			EMIT_ATTR( _TNL_ATTRIB_GENERIC1, EMIT_4UB_4F_ABGR );
+			ADD_ATTR(VERT_ATTRIB_GENERIC1, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR3, swiz, MASK_XYZW, 1);
+#endif
 		}
+	}
 
-		if (fog_id == -1) {
-			fprintf(stderr, "\tout of free texcoords to do fog\n");
-			_mesa_exit(-1);
-		}
+	if (RENDERINPUTS_TEST(tnl->render_inputs_bitset, _TNL_ATTRIB_POINTSIZE )) {
+		GLuint swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ZERO);
+		InputsRead |= 1 << VERT_ATTRIB_POINT_SIZE;
+		OutputsWritten |= 1 << VERT_RESULT_PSIZ;
+		EMIT_ATTR( _TNL_ATTRIB_POINTSIZE, EMIT_1F );
+		ADD_ATTR(VERT_ATTRIB_POINT_SIZE, R300_DATA_TYPE_FLOAT_1, SWTCL_OVM_POINT_SIZE, swiz, MASK_X, 0);
+	}
 
-		sz = VB->AttribPtr[VERT_ATTRIB_FOG]->size;
-		EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1F + sz - 1);
-		InputsRead |= 1 << VERT_ATTRIB_FOG;
-		OutputsWritten |= 1 << VERT_RESULT_FOGC;
-		vap_fmt_1 |= sz << (3 * fog_id);
+	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->TexCoordPtr[tex_id] = VB->AttribPtr[VERT_ATTRIB_POS];
+		RENDERINPUTS_SET(tnl->render_inputs_bitset, _TNL_ATTRIB_TEX0 + tex_id);
 	}
 
-	if (RENDERINPUTS_TEST_RANGE( index_bitset, _TNL_FIRST_TEX, _TNL_LAST_TEX )) {
-		int i;
+	if (rmesa->selected_fp->fog_attr != FRAG_ATTRIB_MAX) {
+		int tex_id = rmesa->selected_fp->fog_attr - FRAG_ATTRIB_TEX0;
 
-		for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
-			if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_TEX(i) )) {
-				sz = VB->TexCoordPtr[i]->size;
-				InputsRead |= 1 << (VERT_ATTRIB_TEX0 + i);
-				OutputsWritten |= 1 << (VERT_RESULT_TEX0 + i);
-				EMIT_ATTR( _TNL_ATTRIB_TEX0+i, EMIT_1F + sz - 1 );
-				vap_fmt_1 |= sz << (3 * i);
-			}
-		}
+		VB->AttribPtr[VERT_ATTRIB_TEX0 + tex_id] = VB->AttribPtr[VERT_ATTRIB_FOG];
+		VB->TexCoordPtr[tex_id] = VB->AttribPtr[VERT_ATTRIB_FOG];
+		RENDERINPUTS_SET(tnl->render_inputs_bitset, _TNL_ATTRIB_TEX0 + tex_id);
 	}
 
-	/* RS can't put fragment position on the pixel stack, so stuff it in texcoord if needed */
-	if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_POS) && (ctx->FragmentProgram._Current->Base.InputsRead & FRAG_BIT_WPOS)) {
-		int first_free_tex = -1;
-		if (fog_id >= 0) {
-			first_free_tex = fog_id+1;
-		} else {
-			if (RENDERINPUTS_TEST_RANGE( index_bitset, _TNL_FIRST_TEX, _TNL_LAST_TEX )) {
-				int i;
-				for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
-					if (!RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_TEX(i) )) {
-						first_free_tex = i;
+	/**
+	 *  Sending only one texcoord component may lead to lock up,
+	 *  so for all textures always output 4 texcoord components to RS.
+	 */
+	{
+		int i;
+		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) {
+					case 1:
+						format = EMIT_1F;
+						hw_format = R300_DATA_TYPE_FLOAT_1;
+						swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ONE);
+						break;
+					case 2:
+						format = EMIT_2F;
+						hw_format = R300_DATA_TYPE_FLOAT_2;
+						swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_ZERO, SWIZZLE_ONE);
+						break;
+					case 3:
+						format = EMIT_3F;
+						hw_format = R300_DATA_TYPE_FLOAT_3;
+						swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ONE);
+						break;
+					case 4:
+						format = EMIT_4F;
+						hw_format = R300_DATA_TYPE_FLOAT_4;
+						swiz = SWIZZLE_XYZW;
 						break;
-					}
+					default:
+						continue;
 				}
-			} else {
-				first_free_tex = 0;
+				InputsRead |= 1 << (VERT_ATTRIB_TEX0 + i);
+				OutputsWritten |= 1 << (VERT_RESULT_TEX0 + i);
+				EMIT_ATTR(_TNL_ATTRIB_TEX(i), format);
+				ADD_ATTR(VERT_ATTRIB_TEX0 + i, hw_format, SWTCL_OVM_TEX(first_free_tex), swiz, MASK_XYZW, 0);
+				++first_free_tex;
 			}
 		}
-
-		if (first_free_tex == -1) {
-			fprintf(stderr, "\tout of free texcoords to write w pos\n");
-			_mesa_exit(-1);
-		}
-
-		sz = VB->AttribPtr[VERT_ATTRIB_POS]->size;
-		InputsRead |= 1 << (VERT_ATTRIB_TEX0 + first_free_tex);
-		OutputsWritten |= 1 << (VERT_RESULT_TEX0 + first_free_tex);
-		EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_1F + sz - 1 );
-		vap_fmt_1 |= sz << (3 * first_free_tex);
-	}
-
-	for (i = 0, nr = 0; i < VERT_ATTRIB_MAX; i++) {
-		if (InputsRead & (1 << i)) {
-			inputs[i] = nr++;
-		} else {
-			inputs[i] = -1;
-		}
-	}
-
-	/* Fixed, apply to vir0 only */
-	if (InputsRead & (1 << VERT_ATTRIB_POS))
-		inputs[VERT_ATTRIB_POS] = 0;
-	if (InputsRead & (1 << VERT_ATTRIB_COLOR0))
-		inputs[VERT_ATTRIB_COLOR0] = 2;
-	if (InputsRead & (1 << VERT_ATTRIB_COLOR1))
-		inputs[VERT_ATTRIB_COLOR1] = 3;
-	if (InputsRead & (1 << VERT_ATTRIB_FOG))
-		inputs[VERT_ATTRIB_FOG] = 6 + fog_id;
-	for (i = VERT_ATTRIB_TEX0; i <= VERT_ATTRIB_TEX7; i++)
-		if (InputsRead & (1 << i))
-			inputs[i] = 6 + (i - VERT_ATTRIB_TEX0);
-
-	for (i = 0, nr = 0; i < VERT_ATTRIB_MAX; i++) {
-		if (InputsRead & (1 << i)) {
-			tab[nr++] = i;
-		}
 	}
 
-	for (i = 0; i < nr; i++) {
-		int ci;
-
-		swizzle[i][0] = SWIZZLE_ZERO;
-		swizzle[i][1] = SWIZZLE_ZERO;
-		swizzle[i][2] = SWIZZLE_ZERO;
-		swizzle[i][3] = SWIZZLE_ONE;
-
-		for (ci = 0; ci < VB->AttribPtr[tab[i]]->size; ci++) {
-			swizzle[i][ci] = ci;
-		}
+	if (first_free_tex >= ctx->Const.MaxTextureUnits) {
+		fprintf(stderr, "\tout of free texcoords to write fog coordinate\n");
+		_mesa_exit(-1);
 	}
 
 	R300_NEWPRIM(rmesa);
-	R300_STATECHANGE(rmesa, vir[0]);
-	((drm_r300_cmd_header_t *) rmesa->hw.vir[0].cmd)->packet0.count =
-		r300VAPInputRoute0(&rmesa->hw.vir[0].cmd[R300_VIR_CNTL_0],
-				   VB->AttribPtr, inputs, tab, nr);
-	R300_STATECHANGE(rmesa, vir[1]);
-	((drm_r300_cmd_header_t *) rmesa->hw.vir[1].cmd)->packet0.count =
-		r300VAPInputRoute1(&rmesa->hw.vir[1].cmd[R300_VIR_CNTL_0], swizzle,
-				   nr);
-
-	R300_STATECHANGE(rmesa, vic);
-	rmesa->hw.vic.cmd[R300_VIC_CNTL_0] = r300VAPInputCntl0(ctx, InputsRead);
-	rmesa->hw.vic.cmd[R300_VIC_CNTL_1] = r300VAPInputCntl1(ctx, InputsRead);
-
-	R300_STATECHANGE(rmesa, vof);
-	rmesa->hw.vof.cmd[R300_VOF_CNTL_0] = r300VAPOutputCntl0(ctx, OutputsWritten);
-	rmesa->hw.vof.cmd[R300_VOF_CNTL_1] = vap_fmt_1;
-
-	rmesa->swtcl.vertex_size =
-		_tnl_install_attrs( ctx,
-				    rmesa->swtcl.vertex_attrs,
-				    rmesa->swtcl.vertex_attr_count,
-				    NULL, 0 );
+	rmesa->vbuf.num_attribs = num_attrs;
+	*_InputsRead = InputsRead;
+	*_OutputsWritten = OutputsWritten;
 
-	rmesa->swtcl.vertex_size /= 4;
-
-	RENDERINPUTS_COPY( rmesa->tnl_index_bitset, index_bitset );
-
-
-	R300_STATECHANGE(rmesa, vte);
-	rmesa->hw.vte.cmd[1] = vte;
-	rmesa->hw.vte.cmd[2] = rmesa->swtcl.vertex_size;
+	RENDERINPUTS_COPY(rmesa->render_inputs_bitset, tnl->render_inputs_bitset);
 }
 
-
-/* Flush vertices in the current dma region.
- */
-static void flush_last_swtcl_prim( r300ContextPtr rmesa  )
+static void r300PrepareVertices(GLcontext *ctx)
 {
-	if (RADEON_DEBUG & DEBUG_IOCTL)
-		fprintf(stderr, "%s\n", __FUNCTION__);
-
-	rmesa->dma.flush = NULL;
-
-	if (rmesa->dma.current.buf) {
-		struct r300_dma_region *current = &rmesa->dma.current;
-		GLuint current_offset = GET_START(current);
-
-		assert (current->start +
-			rmesa->swtcl.numverts * rmesa->swtcl.vertex_size * 4 ==
-			current->ptr);
-
-		if (rmesa->dma.current.start != rmesa->dma.current.ptr) {
-
-			r300EnsureCmdBufSpace( rmesa, rmesa->hw.max_state_size + (12*sizeof(int)), __FUNCTION__);
-
-			r300EmitState(rmesa);
-
-			r300EmitVertexAOS( rmesa,
-					   rmesa->swtcl.vertex_size,
-					   current_offset);
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	GLuint InputsRead, OutputsWritten;
+	radeon_print(RADEON_SWRENDER, RADEON_TRACE, "%s\n", __func__);
 
-			r300EmitVbufPrim( rmesa,
-					  rmesa->swtcl.hw_primitive,
-					  rmesa->swtcl.numverts);
+	r300ChooseSwtclVertexFormat(ctx, &InputsRead, &OutputsWritten);
+	r300SetupVAP(ctx, InputsRead, OutputsWritten);
 
-			r300EmitCacheFlush(rmesa);
-		}
+	rmesa->radeon.swtcl.vertex_size =
+		_tnl_install_attrs( ctx,
+				    rmesa->radeon.swtcl.vertex_attrs,
+				    rmesa->radeon.swtcl.vertex_attr_count,
+				    NULL, 0 );
 
-		rmesa->swtcl.numverts = 0;
-		current->start = current->ptr;
-	}
+	rmesa->radeon.swtcl.vertex_size /= 4;
 }
 
-/* Alloc space in the current dma region.
- */
-static void *
-r300AllocDmaLowVerts( r300ContextPtr rmesa, int nverts, int vsize )
+static void r300_predict_emit_size( r300ContextPtr rmesa )
 {
-	GLuint bytes = vsize * nverts;
-
-	if ( rmesa->dma.current.ptr + bytes > rmesa->dma.current.end )
-		r300RefillCurrentDmaRegion( rmesa, bytes);
-
-	if (!rmesa->dma.flush) {
-		rmesa->radeon.glCtx->Driver.NeedFlush |= FLUSH_STORED_VERTICES;
-		rmesa->dma.flush = flush_last_swtcl_prim;
-	}
-
-	ASSERT( vsize == rmesa->swtcl.vertex_size * 4 );
-	ASSERT( rmesa->dma.flush == flush_last_swtcl_prim );
-	ASSERT( rmesa->dma.current.start +
-		rmesa->swtcl.numverts * rmesa->swtcl.vertex_size * 4 ==
-		rmesa->dma.current.ptr );
-
-	{
-		GLubyte *head = (GLubyte *) (rmesa->dma.current.address + rmesa->dma.current.ptr);
-		rmesa->dma.current.ptr += bytes;
-		rmesa->swtcl.numverts += nverts;
-		return head;
+	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,
-  GL_LINES,
-  GL_LINES,
-  GL_LINES,
-  GL_TRIANGLES,
-  GL_TRIANGLES,
-  GL_TRIANGLES,
-  GL_TRIANGLES,
-  GL_TRIANGLES,
-  GL_TRIANGLES,
+	GL_POINTS,
+	GL_LINES,
+	GL_LINES,
+	GL_LINES,
+	GL_TRIANGLES,
+	GL_TRIANGLES,
+	GL_TRIANGLES,
+	GL_TRIANGLES,
+	GL_TRIANGLES,
+	GL_TRIANGLES,
 };
 
 static void r300RasterPrimitive( GLcontext *ctx, GLuint prim );
-static void r300RenderPrimitive( GLcontext *ctx, GLenum prim );
-//static void r300ResetLineStipple( GLcontext *ctx );
 
 /***********************************************************************
  *                    Emit primitives as inline vertices               *
@@ -402,18 +304,26 @@ static void r300RenderPrimitive( GLcontext *ctx, GLenum prim );
 #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->swtcl.vertex_size
-#define ALLOC_VERTS( n, size ) r300AllocDmaLowVerts( rmesa, n, size * 4 )
+#define GET_VERTEX_DWORDS() rmesa->radeon.swtcl.vertex_size
+#define ALLOC_VERTS( n, size ) r300_alloc_verts(rmesa, n, size);
 #define LOCAL_VARS						\
    r300ContextPtr rmesa = R300_CONTEXT(ctx);		\
-   const char *r300verts = (char *)rmesa->swtcl.verts;
+   const char *r300verts = (char *)rmesa->radeon.swtcl.verts;
 #define VERT(x) (r300Vertex *)(r300verts + ((x) * vertsize * sizeof(int)))
 #define VERTEX r300Vertex
-#define DO_DEBUG_VERTS (1 && (RADEON_DEBUG & DEBUG_VERTS))
-#define PRINT_VERTEX(x)
 #undef TAG
 #define TAG(x) r300_##x
 #include "tnl_dd/t_dd_triemit.h"
@@ -433,9 +343,8 @@ static void r300RenderPrimitive( GLcontext *ctx, GLenum prim );
  *              Build render functions from dd templates               *
  ***********************************************************************/
 
-#define R300_TWOSIDE_BIT	0x01
-#define R300_UNFILLED_BIT	0x02
-#define R300_MAX_TRIFUNC	0x04
+#define R300_UNFILLED_BIT	0x01
+#define R300_MAX_TRIFUNC	0x02
 
 static struct {
    tnl_points_func	        points;
@@ -446,9 +355,9 @@ static struct {
 
 #define DO_FALLBACK  0
 #define DO_UNFILLED (IND & R300_UNFILLED_BIT)
-#define DO_TWOSIDE  (IND & R300_TWOSIDE_BIT)
+#define DO_TWOSIDE   0
 #define DO_FLAT      0
-#define DO_OFFSET     0
+#define DO_OFFSET    0
 #define DO_TRI       1
 #define DO_QUAD      1
 #define DO_LINE      1
@@ -468,33 +377,39 @@ static struct {
 #define VERT_Y(_v) _v->v.y
 #define VERT_Z(_v) _v->v.z
 #define AREA_IS_CCW( a ) (a < 0)
-#define GET_VERTEX(e) (rmesa->swtcl.verts + (e*rmesa->swtcl.vertex_size*sizeof(int)))
-
-/* Only used to pull back colors into vertices (ie, we know color is
- * floating point).
- */
-#define R300_COLOR( dst, src )				\
-do {							\
-   UNCLAMPED_FLOAT_TO_UBYTE((dst)[0], (src)[2]);	\
-   UNCLAMPED_FLOAT_TO_UBYTE((dst)[1], (src)[1]);	\
-   UNCLAMPED_FLOAT_TO_UBYTE((dst)[2], (src)[0]);	\
-   UNCLAMPED_FLOAT_TO_UBYTE((dst)[3], (src)[3]);	\
+#define GET_VERTEX(e) (rmesa->radeon.swtcl.verts + (e*rmesa->radeon.swtcl.vertex_size*sizeof(int)))
+
+#define VERT_SET_RGBA( v, c ) \
+do { \
+   r300_color_t *color = (r300_color_t *)&((v)->ui[coloroffset]); \
+   UNCLAMPED_FLOAT_TO_UBYTE(color->red, (c)[0]); \
+   UNCLAMPED_FLOAT_TO_UBYTE(color->green, (c)[1]); \
+   UNCLAMPED_FLOAT_TO_UBYTE(color->blue, (c)[2]); \
+   UNCLAMPED_FLOAT_TO_UBYTE(color->alpha, (c)[3]); \
 } while (0)
 
-#define VERT_SET_RGBA( v, c )    if (coloroffset) R300_COLOR( v->ub4[coloroffset], c )
-#define VERT_COPY_RGBA( v0, v1 ) if (coloroffset) v0->ui[coloroffset] = v1->ui[coloroffset]
-#define VERT_SAVE_RGBA( idx )    if (coloroffset) color[idx] = v[idx]->ui[coloroffset]
-#define VERT_RESTORE_RGBA( idx ) if (coloroffset) v[idx]->ui[coloroffset] = color[idx]
+#define VERT_COPY_RGBA( v0, v1 ) v0->ui[coloroffset] = v1->ui[coloroffset]
+
+#define VERT_SET_SPEC( v0, c ) \
+do { \
+   if (specoffset) { \
+   UNCLAMPED_FLOAT_TO_UBYTE(v0->v.specular.red, (c)[0]); \
+   UNCLAMPED_FLOAT_TO_UBYTE(v0->v.specular.green, (c)[1]); \
+   UNCLAMPED_FLOAT_TO_UBYTE(v0->v.specular.blue, (c)[2]); \
+   } \
+} while (0)
 
-#define R300_SPEC( dst, src )				\
-do {							\
-   UNCLAMPED_FLOAT_TO_UBYTE((dst)[0], (src)[2]);	\
-   UNCLAMPED_FLOAT_TO_UBYTE((dst)[1], (src)[1]);	\
-   UNCLAMPED_FLOAT_TO_UBYTE((dst)[2], (src)[0]);	\
+#define VERT_COPY_SPEC( v0, v1 ) \
+do { \
+   if (specoffset) { \
+       v0->v.specular.red = v1->v.specular.red; \
+       v0->v.specular.green = v1->v.specular.green; \
+       v0->v.specular.blue = v1->v.specular.blue; \
+   } \
 } while (0)
 
-#define VERT_SET_SPEC( v, c )    if (specoffset) R300_SPEC( v->ub4[specoffset], c )
-#define VERT_COPY_SPEC( v0, v1 ) if (specoffset) COPY_3V(v0->ub4[specoffset], v1->ub4[specoffset])
+#define VERT_SAVE_RGBA( idx )    color[idx] = v[idx]->ui[coloroffset]
+#define VERT_RESTORE_RGBA( idx ) v[idx]->ui[coloroffset] = color[idx]
 #define VERT_SAVE_SPEC( idx )    if (specoffset) spec[idx] = v[idx]->ui[specoffset]
 #define VERT_RESTORE_SPEC( idx ) if (specoffset) v[idx]->ui[specoffset] = spec[idx]
 
@@ -514,7 +429,7 @@ do {							\
  ***********************************************************************/
 
 #define RASTERIZE(x) r300RasterPrimitive( ctx, reduced_prim[x] )
-#define RENDER_PRIMITIVE rmesa->swtcl.render_primitive
+#define RENDER_PRIMITIVE rmesa->radeon.swtcl.render_primitive
 #undef TAG
 #define TAG(x) x
 #include "tnl_dd/t_dd_unfilled.h"
@@ -530,26 +445,15 @@ do {							\
 #define TAG(x) x
 #include "tnl_dd/t_dd_tritmp.h"
 
-#define IND (R300_TWOSIDE_BIT)
-#define TAG(x) x##_twoside
-#include "tnl_dd/t_dd_tritmp.h"
-
 #define IND (R300_UNFILLED_BIT)
 #define TAG(x) x##_unfilled
 #include "tnl_dd/t_dd_tritmp.h"
 
-#define IND (R300_TWOSIDE_BIT|R300_UNFILLED_BIT)
-#define TAG(x) x##_twoside_unfilled
-#include "tnl_dd/t_dd_tritmp.h"
-
-
 
 static void init_rast_tab( void )
 {
    init();
-   init_twoside();
    init_unfilled();
-   init_twoside_unfilled();
 }
 
 /**********************************************************************/
@@ -571,8 +475,8 @@ static void init_rast_tab( void )
 #undef LOCAL_VARS
 #define LOCAL_VARS						\
    r300ContextPtr rmesa = R300_CONTEXT(ctx);		\
-   const GLuint vertsize = rmesa->swtcl.vertex_size;		\
-   const char *r300verts = (char *)rmesa->swtcl.verts;		\
+   const GLuint vertsize = rmesa->radeon.swtcl.vertex_size;		\
+   const char *r300verts = (char *)rmesa->radeon.swtcl.verts;		\
    const GLuint * const elt = TNL_CONTEXT(ctx)->vb.Elts;	\
    const GLboolean stipple = ctx->Line.StippleFlag;		\
    (void) elt; (void) stipple;
@@ -600,11 +504,11 @@ static void r300ChooseRenderState( GLcontext *ctx )
 	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_LIGHT_TWOSIDE) index |= R300_TWOSIDE_BIT;
 	if (flags & DD_TRI_UNFILLED)      index |= R300_UNFILLED_BIT;
 
-	if (index != rmesa->swtcl.RenderIndex) {
+	if (index != rmesa->radeon.swtcl.RenderIndex) {
 		tnl->Driver.Render.Points = rast_tab[index].points;
 		tnl->Driver.Render.Line = rast_tab[index].line;
 		tnl->Driver.Render.ClippedLine = rast_tab[index].line;
@@ -621,59 +525,64 @@ static void r300ChooseRenderState( GLcontext *ctx )
 			tnl->Driver.Render.ClippedPolygon = _tnl_RenderClippedPolygon;
 		}
 
-		rmesa->swtcl.RenderIndex = index;
+		rmesa->radeon.swtcl.RenderIndex = index;
 	}
 }
 
-
-static void r300RenderStart(GLcontext *ctx)
+void r300RenderStart(GLcontext *ctx)
 {
-        r300ContextPtr rmesa = R300_CONTEXT( ctx );
+	radeon_print(RADEON_SWRENDER, RADEON_VERBOSE, "%s\n", __func__);
+	r300ContextPtr rmesa = R300_CONTEXT( ctx );
 
 	r300ChooseRenderState(ctx);
-	r300SetVertexFormat(ctx);
 
 	r300UpdateShaders(rmesa);
-	r300UpdateShaderStates(rmesa);
 
-	r300EmitCacheFlush(rmesa);
+	r300PrepareVertices(ctx);
+
+	r300ValidateBuffers(ctx);
+
+	r300UpdateShaderStates(rmesa);
 
-	if (rmesa->dma.flush != 0 &&
-	    rmesa->dma.flush != flush_last_swtcl_prim)
-		rmesa->dma.flush( rmesa );
 
+	/* investigate if we can put back flush optimisation if needed */
+	if (rmesa->radeon.dma.flush != NULL) {
+		rmesa->radeon.dma.flush(ctx);
+	}
 }
 
-static void r300RenderFinish(GLcontext *ctx)
+void r300RenderFinish(GLcontext *ctx)
 {
 }
 
 static void r300RasterPrimitive( GLcontext *ctx, GLuint hwprim )
 {
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	radeon_print(RADEON_SWRENDER, RADEON_TRACE, "%s\n", __func__);
 
-	if (rmesa->swtcl.hw_primitive != hwprim) {
-	        R300_NEWPRIM( rmesa );
-		rmesa->swtcl.hw_primitive = hwprim;
+	if (rmesa->radeon.swtcl.hw_primitive != hwprim) {
+		R300_NEWPRIM( rmesa );
+		rmesa->radeon.swtcl.hw_primitive = hwprim;
 	}
 }
 
-static void r300RenderPrimitive(GLcontext *ctx, GLenum prim)
+void r300RenderPrimitive(GLcontext *ctx, GLenum prim)
 {
 
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	rmesa->swtcl.render_primitive = prim;
+	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;
+		return;
 
 	r300RasterPrimitive( ctx, reduced_prim[prim] );
 }
 
-static void r300ResetLineStipple(GLcontext *ctx)
+void r300ResetLineStipple(GLcontext *ctx)
 {
-
-
+	if (RADEON_DEBUG & RADEON_VERTS)
+		fprintf(stderr, "%s\n", __func__);
 }
 
 void r300InitSwtcl(GLcontext *ctx)
@@ -681,11 +590,13 @@ 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;
@@ -699,50 +610,75 @@ void r300InitSwtcl(GLcontext *ctx)
 	_tnl_init_vertices( ctx, ctx->Const.MaxArrayLockSize + 12,
 			    48 * sizeof(GLfloat) );
 
-	rmesa->swtcl.verts = (GLubyte *)tnl->clipspace.vertex_buf;
-	rmesa->swtcl.RenderIndex = ~0;
-	rmesa->swtcl.render_primitive = GL_TRIANGLES;
-	rmesa->swtcl.hw_primitive = 0;
+	rmesa->radeon.swtcl.verts = (GLubyte *)tnl->clipspace.vertex_buf;
+	rmesa->radeon.swtcl.RenderIndex = ~0;
+	rmesa->radeon.swtcl.render_primitive = GL_TRIANGLES;
+	rmesa->radeon.swtcl.hw_primitive = 0;
 
 	_tnl_invalidate_vertex_state( ctx, ~0 );
 	_tnl_invalidate_vertices( ctx, ~0 );
-	RENDERINPUTS_ZERO( rmesa->tnl_index_bitset );
 
 	_tnl_need_projected_coords( ctx, GL_FALSE );
-	r300ChooseRenderState(ctx);
 }
 
 void r300DestroySwtcl(GLcontext *ctx)
 {
 }
 
-void r300EmitVertexAOS(r300ContextPtr rmesa, GLuint vertex_size, GLuint offset)
+static void r300EmitVertexAOS(r300ContextPtr rmesa, GLuint vertex_size, struct radeon_bo *bo, GLuint offset)
 {
-	int cmd_reserved = 0;
-	int cmd_written = 0;
-
-	drm_radeon_cmd_header_t *cmd = NULL;
-	if (RADEON_DEBUG & DEBUG_VERTS)
-	  fprintf(stderr, "%s:  vertex_size %d, offset 0x%x \n",
-		  __FUNCTION__, vertex_size, offset);
-
-	start_packet3(CP_PACKET3(R300_PACKET3_3D_LOAD_VBPNTR, 2), 2);
-	e32(1);
-	e32(vertex_size | (vertex_size << 8));
-	e32(offset);
+	BATCH_LOCALS(&rmesa->radeon);
+
+	radeon_print(RADEON_SWRENDER, RADEON_TRACE,
+		"%s:  vertex_size %d, offset 0x%x \n",
+			__FUNCTION__, vertex_size, offset);
+
+	BEGIN_BATCH(7);
+	OUT_BATCH_PACKET3(R300_PACKET3_3D_LOAD_VBPNTR, 2);
+	OUT_BATCH(1);
+	OUT_BATCH(vertex_size | (vertex_size << 8));
+	OUT_BATCH_RELOC(offset, bo, offset, RADEON_GEM_DOMAIN_GTT, 0, 0);
+	END_BATCH();
 }
 
-void r300EmitVbufPrim(r300ContextPtr rmesa, GLuint primitive, GLuint vertex_nr)
+static void r300EmitVbufPrim(r300ContextPtr rmesa, GLuint primitive, GLuint vertex_nr)
 {
-
-	int cmd_reserved = 0;
-	int cmd_written = 0;
+	BATCH_LOCALS(&rmesa->radeon);
 	int type, num_verts;
-	drm_radeon_cmd_header_t *cmd = NULL;
+	if (RADEON_DEBUG & RADEON_VERTS)
+		fprintf(stderr, "%s\n", __func__);
 
 	type = r300PrimitiveType(rmesa, primitive);
 	num_verts = r300NumVerts(rmesa, vertex_nr, primitive);
 
-	start_packet3(CP_PACKET3(R300_PACKET3_3D_DRAW_VBUF_2, 0), 0);
-	e32(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (num_verts << 16) | type);
+	BEGIN_BATCH(3);
+	OUT_BATCH_PACKET3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
+	OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (num_verts << 16) | type);
+	END_BATCH();
+}
+
+void r300_swtcl_flush(GLcontext *ctx, uint32_t current_offset)
+{
+	radeon_print(RADEON_SWRENDER, RADEON_TRACE, "%s\n", __func__);
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+
+	r300EmitCacheFlush(rmesa);
+
+	radeonEmitState(&rmesa->radeon);
+    r300_emit_scissor(ctx);
+	r300EmitVertexAOS(rmesa,
+			rmesa->radeon.swtcl.vertex_size,
+			first_elem(&rmesa->radeon.dma.reserved)->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();
 }