1 files changed, 96 insertions, 137 deletions

diff --git a/src/mesa/drivers/dri/r300/r300_render.c b/src/mesa/drivers/dri/r300/r300_render.c
index a118dab740..1aaf43bfaf 100644
--- a/src/mesa/drivers/dri/r300/r300_render.c
+++ b/src/mesa/drivers/dri/r300/r300_render.c
@@ -45,6 +45,9 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
  * obviously this does work... Further investigation is needed.
  *
  * \author Nicolai Haehnle <prefect_@gmx.net>
+ *
+ * \todo Add immediate implementation back? Perhaps this is useful if there are
+ * no bugs...
  */
 
 #include "glheader.h"
@@ -70,7 +73,6 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "r300_state.h"
 #include "r300_reg.h"
 #include "r300_tex.h"
-#include "r300_maos.h"
 #include "r300_emit.h"
 extern int future_hw_tcl_on;
 
@@ -79,47 +81,42 @@ extern int future_hw_tcl_on;
  */
 static int r300PrimitiveType(r300ContextPtr rmesa, GLcontext * ctx, int prim)
 {
-	int type = -1;
-
 	switch (prim & PRIM_MODE_MASK) {
 	case GL_POINTS:
-		type = R300_VAP_VF_CNTL__PRIM_POINTS;
+		return R300_VAP_VF_CNTL__PRIM_POINTS;
 		break;
 	case GL_LINES:
-		type = R300_VAP_VF_CNTL__PRIM_LINES;
+		return R300_VAP_VF_CNTL__PRIM_LINES;
 		break;
 	case GL_LINE_STRIP:
-		type = R300_VAP_VF_CNTL__PRIM_LINE_STRIP;
+		return R300_VAP_VF_CNTL__PRIM_LINE_STRIP;
 		break;
 	case GL_LINE_LOOP:
-		type = R300_VAP_VF_CNTL__PRIM_LINE_LOOP;
+		return R300_VAP_VF_CNTL__PRIM_LINE_LOOP;
 		break;
 	case GL_TRIANGLES:
-		type = R300_VAP_VF_CNTL__PRIM_TRIANGLES;
+		return R300_VAP_VF_CNTL__PRIM_TRIANGLES;
 		break;
 	case GL_TRIANGLE_STRIP:
-		type = R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP;
+		return R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP;
 		break;
 	case GL_TRIANGLE_FAN:
-		type = R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN;
+		return R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN;
 		break;
 	case GL_QUADS:
-		type = R300_VAP_VF_CNTL__PRIM_QUADS;
+		return R300_VAP_VF_CNTL__PRIM_QUADS;
 		break;
 	case GL_QUAD_STRIP:
-		type = R300_VAP_VF_CNTL__PRIM_QUAD_STRIP;
+		return R300_VAP_VF_CNTL__PRIM_QUAD_STRIP;
 		break;
 	case GL_POLYGON:
-		type = R300_VAP_VF_CNTL__PRIM_POLYGON;
+		return R300_VAP_VF_CNTL__PRIM_POLYGON;
 		break;
 	default:
-		fprintf(stderr,
-			"%s:%s Do not know how to handle primitive 0x%04x - help me !\n",
-			__FILE__, __FUNCTION__, prim & PRIM_MODE_MASK);
+		assert(0);
 		return -1;
 		break;
 	}
-	return type;
 }
 
 static int r300NumVerts(r300ContextPtr rmesa, int num_verts, int prim)
@@ -166,89 +163,103 @@ static int r300NumVerts(r300ContextPtr rmesa, int num_verts, int prim)
 			verts_off = num_verts;
 		break;
 	default:
-		fprintf(stderr,
-			"%s:%s Do not know how to handle primitive 0x%04x - help me !\n",
-			__FILE__, __FUNCTION__, prim & PRIM_MODE_MASK);
+		assert(0);
 		return -1;
 		break;
 	}
 
-	if (num_verts - verts_off == 0) {
-		WARN_ONCE
-		    ("user error: Need more than %d vertices to draw primitive 0x%04x !\n",
-		     num_verts, prim & PRIM_MODE_MASK);
-		return -1;
-	}
+	return num_verts - verts_off;
+}
 
-	if (verts_off > 0) {
-		WARN_ONCE
-		    ("user error: %d is not a valid number of vertices for primitive 0x%04x !\n",
-		     num_verts, prim & PRIM_MODE_MASK);
-		return -1;
+static void r300EmitElts(GLcontext * ctx, void *elts, unsigned long n_elts)
+{
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	struct r300_dma_region *rvb = &rmesa->state.elt_dma;
+	void *out;
+
+	if (r300IsGartMemory(rmesa, elts, n_elts * 4)) {
+		rvb->address = rmesa->radeon.radeonScreen->gartTextures.map;
+		rvb->start = ((char *)elts) - rvb->address;
+		rvb->aos_offset =
+		    rmesa->radeon.radeonScreen->gart_texture_offset +
+		    rvb->start;
+		return;
+	} else if (r300IsGartMemory(rmesa, elts, 1)) {
+		WARN_ONCE("Pointer not within GART memory!\n");
+		_mesa_exit(-1);
 	}
 
-	return num_verts - verts_off;
+	r300AllocDmaRegion(rmesa, rvb, n_elts * 4, 4);
+	rvb->aos_offset = GET_START(rvb);
+
+	out = rvb->address + rvb->start;
+	memcpy(out, elts, n_elts * 4);
 }
 
-static void inline r300FireEB(r300ContextPtr rmesa, unsigned long addr,
-			      int vertex_count, int type, int elt_size)
+static void r300FireEB(r300ContextPtr rmesa, unsigned long addr,
+		       int vertex_count, int type)
 {
 	int cmd_reserved = 0;
 	int cmd_written = 0;
 	drm_radeon_cmd_header_t *cmd = NULL;
-	unsigned long t_addr;
-	unsigned long magic_1, magic_2;
 
-	assert(elt_size == 2 || elt_size == 4);
+	start_packet3(CP_PACKET3(R300_PACKET3_3D_DRAW_INDX_2, 0), 0);
+	e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (vertex_count << 16) | type | R300_VAP_VF_CNTL__INDEX_SIZE_32bit);
 
-	if (addr & (elt_size - 1)) {
-		WARN_ONCE("Badly aligned buffer\n");
-		return;
-	}
+	start_packet3(CP_PACKET3(R300_PACKET3_INDX_BUFFER, 2), 2);
+	e32(R300_EB_UNK1 | (0 << 16) | R300_EB_UNK2);
+	e32(addr);
+	e32(vertex_count);
+}
+
+static void r300EmitAOS(r300ContextPtr rmesa, GLuint nr, GLuint offset)
+{
+	int sz = 1 + (nr >> 1) * 3 + (nr & 1) * 2;
+	int i;
+	int cmd_reserved = 0;
+	int cmd_written = 0;
+	drm_radeon_cmd_header_t *cmd = NULL;
 
-	magic_1 = (addr % 32) / 4;
-	t_addr = addr & ~0x1d;
-	magic_2 = (vertex_count + 1 + (t_addr & 0x2)) / 2 + magic_1;
+	if (RADEON_DEBUG & DEBUG_VERTS)
+		fprintf(stderr, "%s: nr=%d, ofs=0x%08x\n", __FUNCTION__, nr,
+			offset);
 
-	start_packet3(RADEON_CP_PACKET3_3D_DRAW_INDX_2, 0);
-	if (elt_size == 4) {
-		e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES |
-		    (vertex_count << 16) | type |
-		    R300_VAP_VF_CNTL__INDEX_SIZE_32bit);
-	} else {
-		e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES |
-		    (vertex_count << 16) | type);
-	}
+	start_packet3(CP_PACKET3(R300_PACKET3_3D_LOAD_VBPNTR, sz - 1), sz - 1);
+	e32(nr);
 
-	start_packet3(RADEON_CP_PACKET3_INDX_BUFFER, 2);
-#ifdef OPTIMIZE_ELTS
-	if (elt_size == 4) {
-		e32(R300_EB_UNK1 | (0 << 16) | R300_EB_UNK2);
-		e32(addr);
-	} else {
-		e32(R300_EB_UNK1 | (magic_1 << 16) | R300_EB_UNK2);
-		e32(t_addr);
+	for (i = 0; i + 1 < nr; i += 2) {
+		e32((rmesa->state.aos[i].aos_size << 0) |
+		    (rmesa->state.aos[i].aos_stride << 8) |
+		    (rmesa->state.aos[i + 1].aos_size << 16) |
+		    (rmesa->state.aos[i + 1].aos_stride << 24));
+
+		e32(rmesa->state.aos[i].aos_offset + offset * 4 * rmesa->state.aos[i].aos_stride);
+		e32(rmesa->state.aos[i + 1].aos_offset + offset * 4 * rmesa->state.aos[i + 1].aos_stride);
 	}
-#else
-	e32(R300_EB_UNK1 | (0 << 16) | R300_EB_UNK2);
-	e32(addr);
-#endif
 
-	if (elt_size == 4) {
-		e32(vertex_count);
-	} else {
-#ifdef OPTIMIZE_ELTS
-		e32(magic_2);
-#else
-		e32((vertex_count + 1) / 2);
-#endif
+	if (nr & 1) {
+		e32((rmesa->state.aos[nr - 1].aos_size << 0) |
+		    (rmesa->state.aos[nr - 1].aos_stride << 8));
+		e32(rmesa->state.aos[nr - 1].aos_offset + offset * 4 * rmesa->state.aos[nr - 1].aos_stride);
 	}
 }
 
+static void r300FireAOS(r300ContextPtr rmesa, int vertex_count, int type)
+{
+	int cmd_reserved = 0;
+	int cmd_written = 0;
+	drm_radeon_cmd_header_t *cmd = NULL;
+
+	start_packet3(CP_PACKET3(R300_PACKET3_3D_DRAW_VBUF_2, 0), 0);
+	e32(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (vertex_count << 16) | type);
+}
+
 static void r300RunRenderPrimitive(r300ContextPtr rmesa, GLcontext * ctx,
 				   int start, int end, int prim)
 {
 	int type, num_verts;
+	TNLcontext *tnl = TNL_CONTEXT(ctx);
+	struct vertex_buffer *vb = &tnl->vb;
 
 	type = r300PrimitiveType(rmesa, ctx, prim);
 	num_verts = r300NumVerts(rmesa, end - start, prim);
@@ -256,88 +267,37 @@ static void r300RunRenderPrimitive(r300ContextPtr rmesa, GLcontext * ctx,
 	if (type < 0 || num_verts <= 0)
 		return;
 
-	if (rmesa->state.VB.Elts) {
+	if (vb->Elts) {
 		r300EmitAOS(rmesa, rmesa->state.aos_count, start);
 		if (num_verts > 65535) {
 			/* not implemented yet */
 			WARN_ONCE("Too many elts\n");
 			return;
 		}
-		r300EmitElts(ctx, rmesa->state.VB.Elts, num_verts,
-			     rmesa->state.VB.elt_size);
-		r300FireEB(rmesa, rmesa->state.elt_dma.aos_offset,
-			   num_verts, type, rmesa->state.VB.elt_size);
+		r300EmitElts(ctx, vb->Elts, num_verts);
+		r300FireEB(rmesa, rmesa->state.elt_dma.aos_offset, num_verts, type);
 	} else {
 		r300EmitAOS(rmesa, rmesa->state.aos_count, start);
-		fire_AOS(rmesa, num_verts, type);
-	}
-}
-
-#define CONV_VB(a, b) rvb->AttribPtr[(a)].size = vb->b->size, \
-			rvb->AttribPtr[(a)].type = GL_FLOAT, \
-			rvb->AttribPtr[(a)].stride = vb->b->stride, \
-			rvb->AttribPtr[(a)].data = vb->b->data
-
-static void radeon_vb_to_rvb(r300ContextPtr rmesa,
-			     struct radeon_vertex_buffer *rvb,
-			     struct vertex_buffer *vb)
-{
-	int i;
-	GLcontext *ctx;
-	ctx = rmesa->radeon.glCtx;
-
-	memset(rvb, 0, sizeof(*rvb));
-
-	rvb->Elts = vb->Elts;
-	rvb->elt_size = 4;
-	rvb->elt_min = 0;
-	rvb->elt_max = vb->Count;
-
-	rvb->Count = vb->Count;
-
-	if (hw_tcl_on) {
-		CONV_VB(VERT_ATTRIB_POS, ObjPtr);
-	} else {
-		assert(vb->ClipPtr);
-		CONV_VB(VERT_ATTRIB_POS, ClipPtr);
+		r300FireAOS(rmesa, num_verts, type);
 	}
-
-	CONV_VB(VERT_ATTRIB_NORMAL, NormalPtr);
-	CONV_VB(VERT_ATTRIB_COLOR0, ColorPtr[0]);
-	CONV_VB(VERT_ATTRIB_COLOR1, SecondaryColorPtr[0]);
-	CONV_VB(VERT_ATTRIB_FOG, FogCoordPtr);
-
-	for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++)
-		CONV_VB(VERT_ATTRIB_TEX0 + i, TexCoordPtr[i]);
-
-	for (i = 0; i < MAX_VERTEX_PROGRAM_ATTRIBS; i++)
-		CONV_VB(VERT_ATTRIB_GENERIC0 + i,
-			AttribPtr[VERT_ATTRIB_GENERIC0 + i]);
-
-	rvb->Primitive = vb->Primitive;
-	rvb->PrimitiveCount = vb->PrimitiveCount;
-	rvb->LockFirst = rvb->LockCount = 0;
-	rvb->lock_uptodate = GL_FALSE;
 }
 
 static GLboolean r300RunRender(GLcontext * ctx,
 			       struct tnl_pipeline_stage *stage)
 {
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	struct radeon_vertex_buffer *VB = &rmesa->state.VB;
 	int i;
 	int cmd_reserved = 0;
 	int cmd_written = 0;
 	drm_radeon_cmd_header_t *cmd = NULL;
+	TNLcontext *tnl = TNL_CONTEXT(ctx);
+	struct vertex_buffer *vb = &tnl->vb;
 
 	if (RADEON_DEBUG & DEBUG_PRIMS)
 		fprintf(stderr, "%s\n", __FUNCTION__);
 
-	if (stage) {
-		TNLcontext *tnl = TNL_CONTEXT(ctx);
-		radeon_vb_to_rvb(rmesa, VB, &tnl->vb);
-	}
-
+	if (hw_tcl_on == GL_FALSE)
+	  vb->AttribPtr[VERT_ATTRIB_POS] = vb->ClipPtr;
 	r300UpdateShaders(rmesa);
 	if (r300EmitArrays(ctx))
 		return GL_TRUE;
@@ -352,11 +312,10 @@ static GLboolean r300RunRender(GLcontext * ctx,
 
 	r300EmitState(rmesa);
 
-	for (i = 0; i < VB->PrimitiveCount; i++) {
-		GLuint prim = _tnl_translate_prim(&VB->Primitive[i]);
-		GLuint start = VB->Primitive[i].start;
-		GLuint length = VB->Primitive[i].count;
-		GLuint end = VB->Primitive[i].start + VB->Primitive[i].count;
+	for (i = 0; i < vb->PrimitiveCount; i++) {
+		GLuint prim = _tnl_translate_prim(&vb->Primitive[i]);
+		GLuint start = vb->Primitive[i].start;
+		GLuint end = vb->Primitive[i].start + vb->Primitive[i].count;
 		r300RunRenderPrimitive(rmesa, ctx, start, end, prim);
 	}