r300g: various other cleanups
[mesa.git] / src / gallium / drivers / r300 / r300_render.c
index 95b7cb5fa49d1a4ee6e2e92f38107aeeb0cb94b3..f6bc68c43ad968b0614a6e1322d14f73ab4194ab 100644 (file)
 #include "draw/draw_context.h"
 #include "draw/draw_vbuf.h"
 
-#include "pipe/p_inlines.h"
+#include "util/u_inlines.h"
 
 #include "util/u_format.h"
 #include "util/u_memory.h"
+#include "util/u_upload_mgr.h"
 #include "util/u_prim.h"
 
 #include "r300_cs.h"
 #include "r300_context.h"
+#include "r300_screen_buffer.h"
 #include "r300_emit.h"
 #include "r300_reg.h"
 #include "r300_render.h"
 #include "r300_state_derived.h"
 
-/* r300_render: Vertex and index buffer primitive emission. */
-#define R300_MAX_VBO_SIZE  (1024 * 1024)
+/* XXX The DRM rejects VAP_ALT_NUM_VERTICES.. */
+//#define ENABLE_ALT_NUM_VERTS
 
-uint32_t r300_translate_primitive(unsigned prim)
+static uint32_t r300_translate_primitive(unsigned prim)
 {
     switch (prim) {
         case PIPE_PRIM_POINTS:
@@ -115,21 +117,67 @@ static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300,
     return color_control;
 }
 
+/* Check if the requested number of dwords is available in the CS and
+ * if not, flush. Return TRUE if the flush occured. */
+static boolean r300_reserve_cs_space(struct r300_context *r300,
+                                     unsigned dwords)
+{
+    if (!r300->rws->check_cs(r300->rws, dwords)) {
+        r300->context.flush(&r300->context, 0, NULL);
+        return TRUE;
+    }
+    return FALSE;
+}
+
 static boolean immd_is_good_idea(struct r300_context *r300,
-                                      unsigned count)
+                                 unsigned count)
 {
-    return count <= 4;
+    struct pipe_vertex_element* velem;
+    struct pipe_vertex_buffer* vbuf;
+    boolean checked[PIPE_MAX_ATTRIBS] = {0};
+    unsigned vertex_element_count = r300->velems->count;
+    unsigned i, vbi;
+
+    if (count > 10 || DBG_ON(r300, DBG_NO_IMMD)) {
+        return FALSE;
+    }
+
+    /* We shouldn't map buffers referenced by CS, busy buffers,
+     * and ones placed in VRAM. */
+    /* XXX Check for VRAM buffers. */
+    for (i = 0; i < vertex_element_count; i++) {
+        velem = &r300->velems->velem[i];
+        vbi = velem->vertex_buffer_index;
+
+        if (!checked[vbi]) {
+            vbuf = &r300->vertex_buffer[vbi];
+
+            if (r300->context.is_resource_referenced(&r300->context,
+                                                    vbuf->buffer,
+                                                    0, 0)) {
+                /* It's a very bad idea to map it... */
+                return FALSE;
+            }
+            checked[vbi] = TRUE;
+        }
+    }
+    return TRUE;
 }
 
-static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
-                                            unsigned mode,
-                                            unsigned start,
-                                            unsigned count)
+/*****************************************************************************
+ * The emission of draw packets for r500. Older GPUs may use these functions *
+ * after resolving fallback issues (e.g. stencil ref two-sided).             *
+ ****************************************************************************/
+
+void r500_emit_draw_arrays_immediate(struct r300_context *r300,
+                                     unsigned mode,
+                                     unsigned start,
+                                     unsigned count)
 {
     struct pipe_vertex_element* velem;
     struct pipe_vertex_buffer* vbuf;
-    unsigned vertex_element_count = r300->vertex_element_count;
-    unsigned i, v, vbi, dw, elem_offset;
+    unsigned vertex_element_count = r300->velems->count;
+    unsigned i, v, vbi, dw, elem_offset, dwords;
 
     /* Size of the vertex, in dwords. */
     unsigned vertex_size = 0;
@@ -142,16 +190,17 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
 
     /* Stride to the same attrib in the next vertex in the vertex buffer,
      * in dwords. */
-    unsigned stride[PIPE_MAX_ATTRIBS];
+    unsigned stride[PIPE_MAX_ATTRIBS] = {0};
 
     /* Mapped vertex buffers. */
     uint32_t* map[PIPE_MAX_ATTRIBS] = {0};
+    struct pipe_transfer* transfer[PIPE_MAX_ATTRIBS] = {NULL};
 
     CS_LOCALS(r300);
 
     /* Calculate the vertex size, offsets, strides etc. and map the buffers. */
     for (i = 0; i < vertex_element_count; i++) {
-        velem = &r300->vertex_element[i];
+        velem = &r300->velems->velem[i];
         offset[i] = velem->src_offset / 4;
         size[i] = util_format_get_blocksize(velem->src_format) / 4;
         vertex_size += size[i];
@@ -160,22 +209,28 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
         /* Map the buffer. */
         if (!map[vbi]) {
             vbuf = &r300->vertex_buffer[vbi];
-            map[vbi] = (uint32_t*)pipe_buffer_map(r300->context.screen,
+            map[vbi] = (uint32_t*)pipe_buffer_map(&r300->context,
                                                   vbuf->buffer,
-                                                  PIPE_BUFFER_USAGE_CPU_READ);
+                                                  PIPE_TRANSFER_READ,
+                                                 &transfer[vbi]);
             map[vbi] += vbuf->buffer_offset / 4;
             stride[vbi] = vbuf->stride / 4;
         }
     }
 
+    dwords = 9 + count * vertex_size;
+
+    r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + dwords);
+    r300_emit_buffer_validate(r300, FALSE, NULL);
     r300_emit_dirty_state(r300);
 
-    BEGIN_CS(10 + count * vertex_size);
+    BEGIN_CS(dwords);
     OUT_CS_REG(R300_GA_COLOR_CONTROL,
             r300_provoking_vertex_fixes(r300, mode));
     OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
-    OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, 0);
-    OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
+    OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
+    OUT_CS(count - 1);
+    OUT_CS(0);
     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, count * vertex_size);
     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (count << 16) |
             r300_translate_primitive(mode));
@@ -183,7 +238,7 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
     /* Emit vertices. */
     for (v = 0; v < count; v++) {
         for (i = 0; i < vertex_element_count; i++) {
-            velem = &r300->vertex_element[i];
+            velem = &r300->velems->velem[i];
             vbi = velem->vertex_buffer_index;
             elem_offset = offset[i] + stride[vbi] * (v + start);
 
@@ -196,128 +251,238 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
 
     /* Unmap buffers. */
     for (i = 0; i < vertex_element_count; i++) {
-        vbi = r300->vertex_element[i].vertex_buffer_index;
+        vbi = r300->velems->velem[i].vertex_buffer_index;
 
         if (map[vbi]) {
             vbuf = &r300->vertex_buffer[vbi];
-            pipe_buffer_unmap(r300->context.screen, vbuf->buffer);
+            pipe_buffer_unmap(&r300->context, vbuf->buffer, transfer[vbi]);
             map[vbi] = NULL;
         }
     }
 }
 
-static void r300_emit_draw_arrays(struct r300_context *r300,
-                                  unsigned mode,
-                                  unsigned count)
+void r500_emit_draw_arrays(struct r300_context *r300,
+                           unsigned mode,
+                           unsigned count)
 {
+#if defined(ENABLE_ALT_NUM_VERTS)
+    boolean alt_num_verts = count > 65535;
+#else
+    boolean alt_num_verts = FALSE;
+#endif
     CS_LOCALS(r300);
 
-    BEGIN_CS(8);
+    if (alt_num_verts) {
+        if (count >= (1 << 24)) {
+            fprintf(stderr, "r300: Got a huge number of vertices: %i, "
+                    "refusing to render.\n", count);
+            return;
+        }
+        BEGIN_CS(9);
+        OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
+    } else {
+        BEGIN_CS(7);
+    }
     OUT_CS_REG(R300_GA_COLOR_CONTROL,
             r300_provoking_vertex_fixes(r300, mode));
-    OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, 0);
-    OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
+    OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
+    OUT_CS(count - 1);
+    OUT_CS(0);
     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
-           r300_translate_primitive(mode));
+           r300_translate_primitive(mode) |
+           (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
     END_CS;
 }
 
-static void r300_emit_draw_elements(struct r300_context *r300,
-                                    struct pipe_buffer* indexBuffer,
-                                    unsigned indexSize,
-                                    unsigned minIndex,
-                                    unsigned maxIndex,
-                                    unsigned mode,
-                                    unsigned start,
-                                    unsigned count)
+void r500_emit_draw_elements(struct r300_context *r300,
+                             struct pipe_resource* indexBuffer,
+                             unsigned indexSize,
+                             unsigned minIndex,
+                             unsigned maxIndex,
+                             unsigned mode,
+                             unsigned start,
+                             unsigned count)
 {
     uint32_t count_dwords;
     uint32_t offset_dwords = indexSize * start / sizeof(uint32_t);
+#if defined(ENABLE_ALT_NUM_VERTS)
+    boolean alt_num_verts = count > 65535;
+#else
+    boolean alt_num_verts = FALSE;
+#endif
     CS_LOCALS(r300);
 
-    /* XXX most of these are stupid */
-    assert(indexSize == 4 || indexSize == 2);
-    assert((start * indexSize)  % 4 == 0);
-    assert(offset_dwords == 0);
+    if (count >= (1 << 24)) {
+        fprintf(stderr, "r300: Got a huge number of vertices: %i, "
+                "refusing to render.\n", count);
+        return;
+    }
+
+    maxIndex = MIN2(maxIndex, r300->vertex_buffer_max_index);
+
+    DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, min %u max %u\n",
+        count, minIndex, maxIndex);
 
-    BEGIN_CS(14);
+    if (alt_num_verts) {
+        BEGIN_CS(15);
+        OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
+    } else {
+        BEGIN_CS(13);
+    }
     OUT_CS_REG(R300_GA_COLOR_CONTROL,
             r300_provoking_vertex_fixes(r300, mode));
-    OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, minIndex);
-    OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, maxIndex);
+    OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
+    OUT_CS(maxIndex);
+    OUT_CS(minIndex);
     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
     if (indexSize == 4) {
-        count_dwords = count + start;
+        count_dwords = count;
         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
                R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
-               r300_translate_primitive(mode));
+               r300_translate_primitive(mode) |
+               (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
     } else {
-        count_dwords = (count + start + 1) / 2;
+        count_dwords = (count + 1) / 2;
         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
-               r300_translate_primitive(mode));
+               r300_translate_primitive(mode) |
+               (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
     }
 
     /* INDX_BUFFER is a truly special packet3.
      * Unlike most other packet3, where the offset is after the count,
      * the order is reversed, so the relocation ends up carrying the
      * size of the indexbuf instead of the offset.
-     *
-     * XXX Fix offset
      */
     OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
     OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
            (0 << R300_INDX_BUFFER_SKIP_SHIFT));
-    OUT_CS(offset_dwords);
-    OUT_CS_RELOC(indexBuffer, count_dwords,
-        RADEON_GEM_DOMAIN_GTT, 0, 0);
+    OUT_CS(offset_dwords << 2);
+    OUT_CS_BUF_RELOC(indexBuffer, count_dwords,
+                    RADEON_GEM_DOMAIN_GTT, 0, 0);
 
     END_CS;
 }
 
-static boolean r300_setup_vertex_buffers(struct r300_context *r300)
+/*****************************************************************************
+ * The emission of draw packets for r300 which take care of the two-sided    *
+ * stencil ref fallback and call r500's functions.                           *
+ ****************************************************************************/
+
+/* Set drawing for front faces. */
+static void r300_begin_stencil_ref_fallback(struct r300_context *r300)
 {
-    struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
-    struct pipe_vertex_element *velem = r300->vertex_element;
-    struct pipe_buffer *pbuf;
+    struct r300_rs_state *rs = (struct r300_rs_state*)r300->rs_state.state;
+    CS_LOCALS(r300);
 
-validate:
-    for (int i = 0; i < r300->vertex_element_count; i++) {
-        pbuf = vbuf[velem[i].vertex_buffer_index].buffer;
+    BEGIN_CS(2);
+    OUT_CS_REG(R300_SU_CULL_MODE, rs->cull_mode | R300_CULL_BACK);
+    END_CS;
+}
 
-        if (!r300->winsys->add_buffer(r300->winsys, pbuf,
-                                      RADEON_GEM_DOMAIN_GTT, 0)) {
-            r300->context.flush(&r300->context, 0, NULL);
-            goto validate;
-        }
+/* Set drawing for back faces. */
+static void r300_switch_stencil_ref_side(struct r300_context *r300)
+{
+    struct r300_rs_state *rs = (struct r300_rs_state*)r300->rs_state.state;
+    struct r300_dsa_state *dsa = (struct r300_dsa_state*)r300->dsa_state.state;
+    CS_LOCALS(r300);
+
+    BEGIN_CS(4);
+    OUT_CS_REG(R300_SU_CULL_MODE, rs->cull_mode | R300_CULL_FRONT);
+    OUT_CS_REG(R300_ZB_STENCILREFMASK,
+               dsa->stencil_ref_bf | r300->stencil_ref.ref_value[1]);
+    END_CS;
+}
+
+/* Restore the original state. */
+static void r300_end_stencil_ref_fallback(struct r300_context *r300)
+{
+    struct r300_rs_state *rs = (struct r300_rs_state*)r300->rs_state.state;
+    struct r300_dsa_state *dsa = (struct r300_dsa_state*)r300->dsa_state.state;
+    CS_LOCALS(r300);
+
+    BEGIN_CS(4);
+    OUT_CS_REG(R300_SU_CULL_MODE, rs->cull_mode);
+    OUT_CS_REG(R300_ZB_STENCILREFMASK,
+               dsa->stencil_ref_mask | r300->stencil_ref.ref_value[0]);
+    END_CS;
+}
+
+void r300_emit_draw_arrays_immediate(struct r300_context *r300,
+                                     unsigned mode,
+                                     unsigned start,
+                                     unsigned count)
+{
+    if (!r300->stencil_ref_bf_fallback) {
+        r500_emit_draw_arrays_immediate(r300, mode, start, count);
+    } else {
+        r300_begin_stencil_ref_fallback(r300);
+        r500_emit_draw_arrays_immediate(r300, mode, start, count);
+        r300_switch_stencil_ref_side(r300);
+        r500_emit_draw_arrays_immediate(r300, mode, start, count);
+        r300_end_stencil_ref_fallback(r300);
     }
+}
 
-    if (!r300->winsys->validate(r300->winsys)) {
-        r300->context.flush(&r300->context, 0, NULL);
-        return r300->winsys->validate(r300->winsys);
+void r300_emit_draw_arrays(struct r300_context *r300,
+                           unsigned mode,
+                           unsigned count)
+{
+    if (!r300->stencil_ref_bf_fallback) {
+        r500_emit_draw_arrays(r300, mode, count);
+    } else {
+        r300_begin_stencil_ref_fallback(r300);
+        r500_emit_draw_arrays(r300, mode, count);
+        r300_switch_stencil_ref_side(r300);
+        r500_emit_draw_arrays(r300, mode, count);
+        r300_end_stencil_ref_fallback(r300);
     }
+}
 
-    return TRUE;
+void r300_emit_draw_elements(struct r300_context *r300,
+                             struct pipe_resource* indexBuffer,
+                             unsigned indexSize,
+                             unsigned minIndex,
+                             unsigned maxIndex,
+                             unsigned mode,
+                             unsigned start,
+                             unsigned count)
+{
+    if (!r300->stencil_ref_bf_fallback) {
+        r500_emit_draw_elements(r300, indexBuffer, indexSize, minIndex,
+                                maxIndex, mode, start, count);
+    } else {
+        r300_begin_stencil_ref_fallback(r300);
+        r500_emit_draw_elements(r300, indexBuffer, indexSize, minIndex,
+                                maxIndex, mode, start, count);
+        r300_switch_stencil_ref_side(r300);
+        r500_emit_draw_elements(r300, indexBuffer, indexSize, minIndex,
+                                maxIndex, mode, start, count);
+        r300_end_stencil_ref_fallback(r300);
+    }
 }
 
 static void r300_shorten_ubyte_elts(struct r300_context* r300,
-                                    struct pipe_buffer** elts,
+                                    struct pipe_resource** elts,
+                                    unsigned start,
                                     unsigned count)
 {
+    struct pipe_context* context = &r300->context;
     struct pipe_screen* screen = r300->context.screen;
-    struct pipe_buffer* new_elts;
+    struct pipe_resource* new_elts;
     unsigned char *in_map;
     unsigned short *out_map;
+    struct pipe_transfer *src_transfer, *dst_transfer;
     unsigned i;
 
-    new_elts = screen->buffer_create(screen, 32,
-                                     PIPE_BUFFER_USAGE_INDEX |
-                                     PIPE_BUFFER_USAGE_CPU_WRITE |
-                                     PIPE_BUFFER_USAGE_GPU_READ,
-                                     2 * count);
+    new_elts = pipe_buffer_create(screen,
+                                 PIPE_BIND_INDEX_BUFFER,
+                                 2 * count);
 
-    in_map = pipe_buffer_map(screen, *elts, PIPE_BUFFER_USAGE_CPU_READ);
-    out_map = pipe_buffer_map(screen, new_elts, PIPE_BUFFER_USAGE_CPU_WRITE);
+    in_map = pipe_buffer_map(context, *elts, PIPE_TRANSFER_READ, &src_transfer);
+    out_map = pipe_buffer_map(context, new_elts, PIPE_TRANSFER_WRITE, &dst_transfer);
+
+    in_map += start;
 
     for (i = 0; i < count; i++) {
         *out_map = (unsigned short)*in_map;
@@ -325,15 +490,43 @@ static void r300_shorten_ubyte_elts(struct r300_context* r300,
         out_map++;
     }
 
-    pipe_buffer_unmap(screen, *elts);
-    pipe_buffer_unmap(screen, new_elts);
+    pipe_buffer_unmap(context, *elts, src_transfer);
+    pipe_buffer_unmap(context, new_elts, dst_transfer);
+
+    *elts = new_elts;
+}
+
+static void r300_align_ushort_elts(struct r300_context *r300,
+                                   struct pipe_resource **elts,
+                                   unsigned start, unsigned count)
+{
+    struct pipe_context* context = &r300->context;
+    struct pipe_transfer *in_transfer = NULL;
+    struct pipe_transfer *out_transfer = NULL;
+    struct pipe_resource* new_elts;
+    unsigned short *in_map;
+    unsigned short *out_map;
+
+    new_elts = pipe_buffer_create(context->screen,
+                                 PIPE_BIND_INDEX_BUFFER,
+                                 2 * count);
+
+    in_map = pipe_buffer_map(context, *elts,
+                            PIPE_TRANSFER_READ, &in_transfer);
+    out_map = pipe_buffer_map(context, new_elts,
+                             PIPE_TRANSFER_WRITE, &out_transfer);
+
+    memcpy(out_map, in_map+start, 2 * count);
+
+    pipe_buffer_unmap(context, *elts, in_transfer);
+    pipe_buffer_unmap(context, new_elts, out_transfer);
 
     *elts = new_elts;
 }
 
 /* This is the fast-path drawing & emission for HW TCL. */
 void r300_draw_range_elements(struct pipe_context* pipe,
-                              struct pipe_buffer* indexBuffer,
+                              struct pipe_resource* indexBuffer,
                               unsigned indexSize,
                               unsigned minIndex,
                               unsigned maxIndex,
@@ -342,94 +535,135 @@ void r300_draw_range_elements(struct pipe_context* pipe,
                               unsigned count)
 {
     struct r300_context* r300 = r300_context(pipe);
-    struct pipe_buffer* orgIndexBuffer = indexBuffer;
-
-    if (!u_trim_pipe_prim(mode, &count)) {
-        return;
-    }
-
-    if (count > 65535) {
-       /* XXX: use aux/indices functions to split this into smaller
-        * primitives.
-        */
+    struct pipe_resource* orgIndexBuffer = indexBuffer;
+#if defined(ENABLE_ALT_NUM_VERTS)
+    boolean alt_num_verts = r300->screen->caps.is_r500 &&
+                            count > 65536;
+#else
+    boolean alt_num_verts = FALSE;
+#endif
+    unsigned short_count;
+
+    if (r300->skip_rendering) {
         return;
     }
 
-    r300_update_derived_state(r300);
-
-    r300_emit_buffer_validate(r300);
-
-    if (!r300_setup_vertex_buffers(r300)) {
+    if (!u_trim_pipe_prim(mode, &count)) {
         return;
     }
 
     if (indexSize == 1) {
-        r300_shorten_ubyte_elts(r300, &indexBuffer, count);
+        r300_shorten_ubyte_elts(r300, &indexBuffer, start, count);
         indexSize = 2;
+        start = 0;
+    } else if (indexSize == 2 && start % 2 != 0) {
+        r300_align_ushort_elts(r300, &indexBuffer, start, count);
+        start = 0;
     }
 
-    if (!r300->winsys->add_buffer(r300->winsys, indexBuffer,
-                                  RADEON_GEM_DOMAIN_GTT, 0)) {
-        goto cleanup;
-    }
+    r300_update_derived_state(r300);
 
-    if (!r300->winsys->validate(r300->winsys)) {
-        goto cleanup;
-    }
+    r300_upload_index_buffer(r300, &indexBuffer, indexSize, start, count);
 
+    /* 128 dwords for emit_aos and emit_draw_elements */
+    r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 128);
+    r300_emit_buffer_validate(r300, TRUE, indexBuffer);
     r300_emit_dirty_state(r300);
-
     r300_emit_aos(r300, 0);
 
-    r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex, maxIndex,
-                            mode, start, count);
+    u_upload_flush(r300->upload_vb);
+    u_upload_flush(r300->upload_ib);
+    if (alt_num_verts || count <= 65535) {
+        r300->emit_draw_elements(r300, indexBuffer, indexSize, minIndex,
+                                 maxIndex, mode, start, count);
+    } else {
+        do {
+            short_count = MIN2(count, 65534);
+            r300->emit_draw_elements(r300, indexBuffer, indexSize, minIndex,
+                                      maxIndex, mode, start, short_count);
+
+            start += short_count;
+            count -= short_count;
+
+            /* 16 spare dwords are enough for emit_draw_elements. */
+            if (count && r300_reserve_cs_space(r300, 16)) {
+                r300_emit_buffer_validate(r300, TRUE, indexBuffer);
+                r300_emit_dirty_state(r300);
+                r300_emit_aos(r300, 0);
+            }
+        } while (count);
+    }
 
-cleanup:
     if (indexBuffer != orgIndexBuffer) {
-        pipe->screen->buffer_destroy(indexBuffer);
+        pipe_resource_reference( &indexBuffer, NULL );
     }
 }
 
 /* Simple helpers for context setup. Should probably be moved to util. */
 void r300_draw_elements(struct pipe_context* pipe,
-                        struct pipe_buffer* indexBuffer,
+                        struct pipe_resource* indexBuffer,
                         unsigned indexSize, unsigned mode,
                         unsigned start, unsigned count)
 {
-   pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0, ~0,
-                             mode, start, count);
+    struct r300_context *r300 = r300_context(pipe);
+
+    pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0,
+                              r300->vertex_buffer_max_index,
+                              mode, start, count);
 }
 
 void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
                       unsigned start, unsigned count)
 {
     struct r300_context* r300 = r300_context(pipe);
-
-    if (!u_trim_pipe_prim(mode, &count)) {
+#if defined(ENABLE_ALT_NUM_VERTS)
+    boolean alt_num_verts = r300->screen->caps.is_r500 &&
+                            count > 65536;
+#else
+    boolean alt_num_verts = FALSE;
+#endif
+    unsigned short_count;
+
+    if (r300->skip_rendering) {
         return;
     }
 
-    if (count > 65535) {
-        /* XXX: driver needs to handle this -- use the functions in
-         * aux/indices to split this into several smaller primitives.
-         */
+    if (!u_trim_pipe_prim(mode, &count)) {
         return;
     }
 
     r300_update_derived_state(r300);
 
-    r300_emit_buffer_validate(r300);
-
     if (immd_is_good_idea(r300, count)) {
-        r300_emit_draw_arrays_immediate(r300, mode, start, count);
+        r300->emit_draw_arrays_immediate(r300, mode, start, count);
     } else {
-        if (!r300_setup_vertex_buffers(r300)) {
-            return;
-        }
-
+        /* Make sure there are at least 128 spare dwords in the command buffer.
+         * (most of it being consumed by emit_aos) */
+        r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 128);
+        r300_emit_buffer_validate(r300, TRUE, NULL);
         r300_emit_dirty_state(r300);
-        r300_emit_aos(r300, start);
-        r300_emit_draw_arrays(r300, mode, count);
+
+        if (alt_num_verts || count <= 65535) {
+            r300_emit_aos(r300, start);
+            r300->emit_draw_arrays(r300, mode, count);
+        } else {
+            do {
+                short_count = MIN2(count, 65535);
+                r300_emit_aos(r300, start);
+                r300->emit_draw_arrays(r300, mode, short_count);
+
+                start += short_count;
+                count -= short_count;
+
+                /* Again, we emit both AOS and draw_arrays so there should be
+                 * at least 128 spare dwords. */
+                if (count && r300_reserve_cs_space(r300, 128)) {
+                    r300_emit_buffer_validate(r300, TRUE, NULL);
+                    r300_emit_dirty_state(r300);
+                }
+            } while (count);
+        }
+       u_upload_flush(r300->upload_vb);
     }
 }
 
@@ -445,39 +679,39 @@ void r300_swtcl_draw_arrays(struct pipe_context* pipe,
                                unsigned count)
 {
     struct r300_context* r300 = r300_context(pipe);
+    struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS];
     int i;
 
+    if (r300->skip_rendering) {
+        return;
+    }
+
     if (!u_trim_pipe_prim(mode, &count)) {
         return;
     }
 
     for (i = 0; i < r300->vertex_buffer_count; i++) {
-        void* buf = pipe_buffer_map(pipe->screen,
+        void* buf = pipe_buffer_map(pipe,
                                     r300->vertex_buffer[i].buffer,
-                                    PIPE_BUFFER_USAGE_CPU_READ);
+                                    PIPE_TRANSFER_READ,
+                                   &vb_transfer[i]);
         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
     }
 
     draw_set_mapped_element_buffer(r300->draw, 0, NULL);
 
-    draw_set_mapped_constant_buffer(r300->draw,
-                                   PIPE_SHADER_VERTEX,
-                                    0,
-                                   r300->shader_constants[PIPE_SHADER_VERTEX].constants,
-                                   r300->shader_constants[PIPE_SHADER_VERTEX].count *
-                (sizeof(float) * 4));
-
     draw_arrays(r300->draw, mode, start, count);
 
     for (i = 0; i < r300->vertex_buffer_count; i++) {
-        pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
+        pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer,
+                         vb_transfer[i]);
         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
     }
 }
 
 /* SW TCL elements, using Draw. */
 void r300_swtcl_draw_range_elements(struct pipe_context* pipe,
-                                       struct pipe_buffer* indexBuffer,
+                                       struct pipe_resource* indexBuffer,
                                        unsigned indexSize,
                                        unsigned minIndex,
                                        unsigned maxIndex,
@@ -486,40 +720,42 @@ void r300_swtcl_draw_range_elements(struct pipe_context* pipe,
                                        unsigned count)
 {
     struct r300_context* r300 = r300_context(pipe);
+    struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS];
+    struct pipe_transfer *ib_transfer;
     int i;
     void* indices;
 
+    if (r300->skip_rendering) {
+        return;
+    }
+
     if (!u_trim_pipe_prim(mode, &count)) {
         return;
     }
 
     for (i = 0; i < r300->vertex_buffer_count; i++) {
-        void* buf = pipe_buffer_map(pipe->screen,
+        void* buf = pipe_buffer_map(pipe,
                                     r300->vertex_buffer[i].buffer,
-                                    PIPE_BUFFER_USAGE_CPU_READ);
+                                    PIPE_TRANSFER_READ,
+                                   &vb_transfer[i]);
         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
     }
 
-    indices = pipe_buffer_map(pipe->screen, indexBuffer,
-                              PIPE_BUFFER_USAGE_CPU_READ);
+    indices = pipe_buffer_map(pipe, indexBuffer,
+                              PIPE_TRANSFER_READ, &ib_transfer);
     draw_set_mapped_element_buffer_range(r300->draw, indexSize,
                                          minIndex, maxIndex, indices);
 
-    draw_set_mapped_constant_buffer(r300->draw,
-                                   PIPE_SHADER_VERTEX,
-                                    0,
-            r300->shader_constants[PIPE_SHADER_VERTEX].constants,
-            r300->shader_constants[PIPE_SHADER_VERTEX].count *
-                (sizeof(float) * 4));
-
     draw_arrays(r300->draw, mode, start, count);
 
     for (i = 0; i < r300->vertex_buffer_count; i++) {
-        pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
+        pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer,
+                         vb_transfer[i]);
         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
     }
 
-    pipe_buffer_unmap(pipe->screen, indexBuffer);
+    pipe_buffer_unmap(pipe, indexBuffer,
+                     ib_transfer);
     draw_set_mapped_element_buffer_range(r300->draw, 0, start,
                                          start + count - 1, NULL);
 }
@@ -538,11 +774,13 @@ struct r300_render {
     unsigned hwprim;
 
     /* VBO */
-    struct pipe_buffer* vbo;
+    struct pipe_resource* vbo;
     size_t vbo_size;
     size_t vbo_offset;
     size_t vbo_max_used;
     void * vbo_ptr;
+
+   struct pipe_transfer *vbo_transfer;
 };
 
 static INLINE struct r300_render*
@@ -559,7 +797,7 @@ r300_render_get_vertex_info(struct vbuf_render* render)
 
     r300_update_derived_state(r300);
 
-    return (struct vertex_info*)r300->vertex_format_state.state;
+    return &r300->vertex_info;
 }
 
 static boolean r300_render_allocate_vertices(struct vbuf_render* render,
@@ -573,13 +811,12 @@ static boolean r300_render_allocate_vertices(struct vbuf_render* render,
 
     if (size + r300render->vbo_offset > r300render->vbo_size)
     {
-        pipe_buffer_reference(&r300->vbo, NULL);
+        pipe_resource_reference(&r300->vbo, NULL);
         r300render->vbo = pipe_buffer_create(screen,
-                                             64,
-                                             PIPE_BUFFER_USAGE_VERTEX,
-                                             R300_MAX_VBO_SIZE);
+                                             PIPE_BIND_VERTEX_BUFFER,
+                                             R300_MAX_DRAW_VBO_SIZE);
         r300render->vbo_offset = 0;
-        r300render->vbo_size = R300_MAX_VBO_SIZE;
+        r300render->vbo_size = R300_MAX_DRAW_VBO_SIZE;
     }
 
     r300render->vertex_size = vertex_size;
@@ -592,10 +829,11 @@ static boolean r300_render_allocate_vertices(struct vbuf_render* render,
 static void* r300_render_map_vertices(struct vbuf_render* render)
 {
     struct r300_render* r300render = r300_render(render);
-    struct pipe_screen* screen = r300render->r300->context.screen;
 
-    r300render->vbo_ptr = pipe_buffer_map(screen, r300render->vbo,
-                                          PIPE_BUFFER_USAGE_CPU_WRITE);
+    r300render->vbo_ptr = pipe_buffer_map(&r300render->r300->context,
+                                         r300render->vbo,
+                                          PIPE_TRANSFER_WRITE,
+                                         &r300render->vbo_transfer);
 
     return ((uint8_t*)r300render->vbo_ptr + r300render->vbo_offset);
 }
@@ -605,7 +843,7 @@ static void r300_render_unmap_vertices(struct vbuf_render* render,
                                              ushort max)
 {
     struct r300_render* r300render = r300_render(render);
-    struct pipe_screen* screen = r300render->r300->context.screen;
+    struct pipe_context* context = &r300render->r300->context;
     CS_LOCALS(r300render->r300);
     BEGIN_CS(2);
     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max);
@@ -613,7 +851,7 @@ static void r300_render_unmap_vertices(struct vbuf_render* render,
 
     r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
                                     r300render->vertex_size * (max + 1));
-    pipe_buffer_unmap(screen, r300render->vbo);
+    pipe_buffer_unmap(context, r300render->vbo, r300render->vbo_transfer);
 }
 
 static void r300_render_release_vertices(struct vbuf_render* render)
@@ -635,15 +873,17 @@ static boolean r300_render_set_primitive(struct vbuf_render* render,
     return TRUE;
 }
 
-static void r300_render_draw_arrays(struct vbuf_render* render,
-                                          unsigned start,
-                                          unsigned count)
+static void r500_render_draw_arrays(struct vbuf_render* render,
+                                    unsigned start,
+                                    unsigned count)
 {
     struct r300_render* r300render = r300_render(render);
     struct r300_context* r300 = r300render->r300;
 
     CS_LOCALS(r300);
 
+    r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 2);
+    r300_emit_buffer_validate(r300, FALSE, NULL);
     r300_emit_dirty_state(r300);
 
     DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
@@ -655,19 +895,22 @@ static void r300_render_draw_arrays(struct vbuf_render* render,
     END_CS;
 }
 
-static void r300_render_draw(struct vbuf_render* render,
-                                   const ushort* indices,
-                                   uint count)
+static void r500_render_draw(struct vbuf_render* render,
+                             const ushort* indices,
+                             uint count)
 {
     struct r300_render* r300render = r300_render(render);
     struct r300_context* r300 = r300render->r300;
     int i;
+    unsigned dwords = 2 + (count+1)/2;
 
     CS_LOCALS(r300);
 
+    r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + dwords);
+    r300_emit_buffer_validate(r300, FALSE, NULL);
     r300_emit_dirty_state(r300);
 
-    BEGIN_CS(2 + (count+1)/2);
+    BEGIN_CS(dwords);
     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2);
     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
            r300render->hwprim);
@@ -680,6 +923,40 @@ static void r300_render_draw(struct vbuf_render* render,
     END_CS;
 }
 
+static void r300_render_draw_arrays(struct vbuf_render* render,
+                                    unsigned start,
+                                    unsigned count)
+{
+    struct r300_context* r300 = r300_render(render)->r300;
+
+    if (!r300->stencil_ref_bf_fallback) {
+        r500_render_draw_arrays(render, start, count);
+    } else {
+        r300_begin_stencil_ref_fallback(r300);
+        r500_render_draw_arrays(render, start, count);
+        r300_switch_stencil_ref_side(r300);
+        r500_render_draw_arrays(render, start, count);
+        r300_end_stencil_ref_fallback(r300);
+    }
+}
+
+static void r300_render_draw(struct vbuf_render* render,
+                             const ushort* indices,
+                             uint count)
+{
+    struct r300_context* r300 = r300_render(render)->r300;
+
+    if (!r300->stencil_ref_bf_fallback) {
+        r500_render_draw(render, indices, count);
+    } else {
+        r300_begin_stencil_ref_fallback(r300);
+        r500_render_draw(render, indices, count);
+        r300_switch_stencil_ref_side(r300);
+        r500_render_draw(render, indices, count);
+        r300_end_stencil_ref_fallback(r300);
+    }
+}
+
 static void r300_render_destroy(struct vbuf_render* render)
 {
     FREE(render);
@@ -700,8 +977,13 @@ static struct vbuf_render* r300_render_create(struct r300_context* r300)
     r300render->base.map_vertices = r300_render_map_vertices;
     r300render->base.unmap_vertices = r300_render_unmap_vertices;
     r300render->base.set_primitive = r300_render_set_primitive;
-    r300render->base.draw = r300_render_draw;
-    r300render->base.draw_arrays = r300_render_draw_arrays;
+    if (r300->screen->caps.is_r500) {
+        r300render->base.draw = r500_render_draw;
+        r300render->base.draw_arrays = r500_render_draw_arrays;
+    } else {
+        r300render->base.draw = r300_render_draw;
+        r300render->base.draw_arrays = r300_render_draw_arrays;
+    }
     r300render->base.release_vertices = r300_render_release_vertices;
     r300render->base.destroy = r300_render_destroy;