/*
* Copyright 2009 Corbin Simpson <MostAwesomeDude@gmail.com>
+ * Copyright 2010 Marek Olšák <maraeo@gmail.com>
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
#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)
+#include <limits.h>
-/* 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:
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)
+static boolean index_bias_supported(struct r300_context *r300)
+{
+ return r300->screen->caps.is_r500 &&
+ r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
+}
+
+static void r500_emit_index_bias(struct r300_context *r300, int index_bias)
+{
+ CS_LOCALS(r300);
+
+ BEGIN_CS(2);
+ OUT_CS_REG(R500_VAP_INDEX_OFFSET,
+ (index_bias & 0xFFFFFF) | (index_bias < 0 ? 1<<24 : 0));
+ END_CS;
+}
+
+/* This function splits the index bias value into two parts:
+ * - buffer_offset: the value that can be safely added to buffer offsets
+ * in r300_emit_aos (it must yield a positive offset when added to
+ * a vertex buffer offset)
+ * - index_offset: the value that must be manually subtracted from indices
+ * in an index buffer to achieve negative offsets. */
+static void r300_split_index_bias(struct r300_context *r300, int index_bias,
+ int *buffer_offset, int *index_offset)
+{
+ struct pipe_vertex_buffer *vb, *vbufs = r300->vertex_buffer;
+ struct pipe_vertex_element *velem = r300->velems->velem;
+ unsigned i, size;
+ int max_neg_bias;
+
+ if (index_bias < 0) {
+ /* See how large index bias we may subtract. We must be careful
+ * here because negative buffer offsets are not allowed
+ * by the DRM API. */
+ max_neg_bias = INT_MAX;
+ for (i = 0; i < r300->velems->count; i++) {
+ vb = &vbufs[velem[i].vertex_buffer_index];
+ size = (vb->buffer_offset + velem[i].src_offset) / vb->stride;
+ max_neg_bias = MIN2(max_neg_bias, size);
+ }
+
+ /* Now set the minimum allowed value. */
+ *buffer_offset = MAX2(-max_neg_bias, index_bias);
+ } else {
+ /* A positive index bias is OK. */
+ *buffer_offset = index_bias;
+ }
+
+ *index_offset = index_bias - *buffer_offset;
+}
+
+enum r300_prepare_flags {
+ PREP_FIRST_DRAW = (1 << 0), /* call emit_dirty_state and friends? */
+ PREP_VALIDATE_VBOS = (1 << 1), /* validate VBOs? */
+ PREP_EMIT_AOS = (1 << 2), /* call emit_aos? */
+ PREP_EMIT_AOS_SWTCL = (1 << 3), /* call emit_aos_swtcl? */
+ PREP_INDEXED = (1 << 4) /* is this draw_elements? */
+};
+
+/**
+ * Check if the requested number of dwords is available in the CS and
+ * if not, flush. Then validate buffers and emit dirty state.
+ * \param r300 The context.
+ * \param flags See r300_prepare_flags.
+ * \param index_buffer The index buffer to validate. The parameter may be NULL.
+ * \param cs_dwords The number of dwords to reserve in CS.
+ * \param aos_offset The offset passed to emit_aos.
+ * \param index_bias The index bias to emit.
+ * \param end_cs_dwords The number of free dwords which must be available
+ * at the end of CS after drawing in case the CS space
+ * management is performed by a draw_* function manually.
+ * The parameter may be NULL.
+ */
+static void r300_prepare_for_rendering(struct r300_context *r300,
+ enum r300_prepare_flags flags,
+ struct pipe_resource *index_buffer,
+ unsigned cs_dwords,
+ int aos_offset,
+ int index_bias,
+ unsigned *end_cs_dwords)
{
- if (!r300->winsys->check_cs(r300->winsys, dwords)) {
+ unsigned end_dwords = 0;
+ boolean flushed = FALSE;
+ boolean first_draw = flags & PREP_FIRST_DRAW;
+ boolean emit_aos = flags & PREP_EMIT_AOS;
+ boolean emit_aos_swtcl = flags & PREP_EMIT_AOS_SWTCL;
+ boolean indexed = flags & PREP_INDEXED;
+ boolean hw_index_bias = index_bias_supported(r300);
+
+ /* Add dirty state, index offset, and AOS. */
+ if (first_draw) {
+ cs_dwords += r300_get_num_dirty_dwords(r300);
+
+ if (hw_index_bias)
+ cs_dwords += 2; /* emit_index_offset */
+
+ if (emit_aos)
+ cs_dwords += 55; /* emit_aos */
+
+ if (emit_aos_swtcl)
+ cs_dwords += 7; /* emit_aos_swtcl */
+ }
+
+ /* Emitted in flush. */
+ end_dwords += 26; /* emit_query_end */
+
+ cs_dwords += end_dwords;
+
+ /* Reserve requested CS space. */
+ if (!r300_check_cs(r300, cs_dwords)) {
r300->context.flush(&r300->context, 0, NULL);
- return TRUE;
+ flushed = TRUE;
+ }
+
+ /* Validate buffers and emit dirty state if needed. */
+ if (first_draw || flushed) {
+ r300_emit_buffer_validate(r300, flags & PREP_VALIDATE_VBOS, index_buffer);
+ r300_emit_dirty_state(r300);
+ if (hw_index_bias) {
+ if (r300->screen->caps.has_tcl)
+ r500_emit_index_bias(r300, index_bias);
+ else
+ r500_emit_index_bias(r300, 0);
+ }
+
+ if (emit_aos)
+ r300_emit_aos(r300, aos_offset, indexed);
+
+ if (emit_aos_swtcl)
+ r300_emit_aos_swtcl(r300, indexed);
}
- return FALSE;
+
+ if (end_cs_dwords)
+ *end_cs_dwords = end_dwords;
}
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 (DBG_ON(r300, DBG_NO_IMMD)) {
+ return FALSE;
+ }
+
+ if (r300->draw) {
+ return FALSE;
+ }
+
+ if (count > 10) {
+ 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_buffer_is_referenced(&r300->context,
+ vbuf->buffer,
+ R300_REF_CS | R300_REF_HW)) {
+ /* It's a very bad idea to map it... */
+ return FALSE;
+ }
+ checked[vbi] = TRUE;
+ }
+ }
+ return TRUE;
}
+/*****************************************************************************
+ * The emission of draw packets for r500. Older GPUs may use these functions *
+ * after resolving fallback issues (e.g. stencil ref two-sided). *
+ ****************************************************************************/
+
static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
unsigned mode,
unsigned start,
{
struct pipe_vertex_element* velem;
struct pipe_vertex_buffer* vbuf;
- unsigned vertex_element_count = r300->vertex_element_count;
+ unsigned vertex_element_count = r300->velems->count;
unsigned i, v, vbi, dw, elem_offset, dwords;
/* Size of the vertex, in dwords. */
/* 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];
/* 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 = 10 + count * vertex_size;
+ 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);
+ r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0, NULL);
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));
/* 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);
/* 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;
}
}
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);
+ if (count >= (1 << 24)) {
+ fprintf(stderr, "r300: Got a huge number of vertices: %i, "
+ "refusing to render.\n", count);
+ return;
+ }
+
+ BEGIN_CS(7 + (alt_num_verts ? 2 : 0));
if (alt_num_verts) {
- assert(count < (1 << 24));
- BEGIN_CS(10);
OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
- } else {
- BEGIN_CS(8);
}
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) |
}
static void r300_emit_draw_elements(struct r300_context *r300,
- struct pipe_buffer* indexBuffer,
+ struct pipe_resource* indexBuffer,
unsigned indexSize,
unsigned minIndex,
unsigned maxIndex,
{
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);
- assert((start * indexSize) % 4 == 0);
- assert(count < (1 << 24));
+ 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);
- maxIndex = MIN2(maxIndex, r300->vertex_buffer_max_index);
-
+ BEGIN_CS(13 + (alt_num_verts ? 2 : 0));
if (alt_num_verts) {
- BEGIN_CS(16);
OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
- } else {
- BEGIN_CS(14);
}
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;
OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
(0 << R300_INDX_BUFFER_SKIP_SHIFT));
OUT_CS(offset_dwords << 2);
- OUT_CS_RELOC(indexBuffer, count_dwords,
- RADEON_GEM_DOMAIN_GTT, 0, 0);
+ OUT_CS_BUF_RELOC(indexBuffer, count_dwords,
+ r300_buffer(indexBuffer)->domain, 0, 0);
END_CS;
}
static void r300_shorten_ubyte_elts(struct r300_context* r300,
- struct pipe_buffer** elts,
+ struct pipe_resource** elts,
+ int index_bias,
+ 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;
+ *out_map = (unsigned short)(*in_map + index_bias);
in_map++;
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_rebuild_ushort_elts(struct r300_context *r300,
+ struct pipe_resource **elts,
+ int index_bias,
+ 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;
+ unsigned i;
+
+ 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);
+
+ in_map += start;
+ for (i = 0; i < count; i++) {
+ *out_map = (unsigned short)(*in_map + index_bias);
+ in_map++;
+ out_map++;
+ }
+
+ pipe_buffer_unmap(context, *elts, in_transfer);
+ pipe_buffer_unmap(context, new_elts, out_transfer);
+
+ *elts = new_elts;
+}
+
+static void r300_rebuild_uint_elts(struct r300_context *r300,
+ struct pipe_resource **elts,
+ int index_bias,
+ 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 int *in_map;
+ unsigned int *out_map;
+ unsigned i;
+
+ 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);
+
+ in_map += start;
+ for (i = 0; i < count; i++) {
+ *out_map = (unsigned int)(*in_map + index_bias);
+ in_map++;
+ out_map++;
+ }
+
+ 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,
- unsigned indexSize,
- unsigned minIndex,
- unsigned maxIndex,
- unsigned mode,
- unsigned start,
- unsigned count)
+static void r300_draw_range_elements(struct pipe_context* pipe,
+ struct pipe_resource* indexBuffer,
+ unsigned indexSize,
+ int indexBias,
+ unsigned minIndex,
+ unsigned maxIndex,
+ unsigned mode,
+ unsigned start,
+ unsigned count)
{
struct r300_context* r300 = r300_context(pipe);
- struct pipe_buffer* orgIndexBuffer = indexBuffer;
-#if defined(ENABLE_ALT_NUM_VERTS)
- boolean alt_num_verts = r300_screen(pipe->screen)->caps->is_r500 &&
- count > 65536;
-#else
- boolean alt_num_verts = FALSE;
-#endif
+ struct pipe_resource* orgIndexBuffer = indexBuffer;
+ boolean alt_num_verts = r300->screen->caps.is_r500 &&
+ count > 65536 &&
+ r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
unsigned short_count;
+ int buffer_offset = 0, index_offset = 0; /* for index bias emulation */
+
+ if (r300->skip_rendering) {
+ return;
+ }
if (!u_trim_pipe_prim(mode, &count)) {
return;
}
- if (indexSize == 1) {
- r300_shorten_ubyte_elts(r300, &indexBuffer, count);
- indexSize = 2;
+ if (indexBias && !index_bias_supported(r300)) {
+ r300_split_index_bias(r300, indexBias, &buffer_offset, &index_offset);
+ }
+
+ /* Rebuild the index buffer if needed. */
+ switch (indexSize) {
+ case 1:
+ r300_shorten_ubyte_elts(r300, &indexBuffer, index_offset, start, count);
+ indexSize = 2;
+ start = 0;
+ break;
+
+ case 2:
+ if (start % 2 != 0 || index_offset) {
+ r300_rebuild_ushort_elts(r300, &indexBuffer, index_offset, start, count);
+ start = 0;
+ }
+ break;
+
+ case 4:
+ if (index_offset) {
+ r300_rebuild_uint_elts(r300, &indexBuffer, index_offset, start, count);
+ start = 0;
+ }
+ break;
}
r300_update_derived_state(r300);
+ 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);
+ /* 15 dwords for emit_draw_elements */
+ r300_prepare_for_rendering(r300,
+ PREP_FIRST_DRAW | PREP_VALIDATE_VBOS | PREP_EMIT_AOS | PREP_INDEXED,
+ indexBuffer, 15, buffer_offset, indexBias, NULL);
+ 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);
+ 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);
+ 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);
+ /* 15 dwords for emit_draw_elements */
+ if (count) {
+ r300_prepare_for_rendering(r300,
+ PREP_VALIDATE_VBOS | PREP_EMIT_AOS | PREP_INDEXED,
+ indexBuffer, 15, buffer_offset, indexBias, NULL);
}
} while (count);
}
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,
- unsigned indexSize, unsigned mode,
- unsigned start, unsigned count)
+static void r300_draw_elements(struct pipe_context* pipe,
+ struct pipe_resource* indexBuffer,
+ unsigned indexSize, int indexBias, unsigned mode,
+ unsigned start, unsigned count)
{
struct r300_context *r300 = r300_context(pipe);
- pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0,
- r300->vertex_buffer_max_index,
+ pipe->draw_range_elements(pipe, indexBuffer, indexSize, indexBias,
+ 0, r300->vertex_buffer_max_index,
mode, start, count);
}
-void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
- unsigned start, unsigned count)
+static void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
+ unsigned start, unsigned count)
{
struct r300_context* r300 = r300_context(pipe);
-#if defined(ENABLE_ALT_NUM_VERTS)
- boolean alt_num_verts = r300_screen(pipe->screen)->caps->is_r500 &&
- count > 65536;
-#else
- boolean alt_num_verts = FALSE;
-#endif
+ boolean alt_num_verts = r300->screen->caps.is_r500 &&
+ count > 65536 &&
+ r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
unsigned short_count;
+ if (r300->skip_rendering) {
+ return;
+ }
+
if (!u_trim_pipe_prim(mode, &count)) {
return;
}
if (immd_is_good_idea(r300, count)) {
r300_emit_draw_arrays_immediate(r300, mode, start, count);
} else {
- /* 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);
+ /* 9 spare dwords for emit_draw_arrays. */
+ r300_prepare_for_rendering(r300, PREP_FIRST_DRAW | PREP_VALIDATE_VBOS | PREP_EMIT_AOS,
+ NULL, 9, start, 0, NULL);
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);
+ /* 9 spare dwords for emit_draw_arrays. */
+ if (count) {
+ r300_prepare_for_rendering(r300,
+ PREP_VALIDATE_VBOS | PREP_EMIT_AOS, NULL, 9,
+ start, 0, NULL);
}
} while (count);
}
+ u_upload_flush(r300->upload_vb);
}
}
***************************************************************************/
/* SW TCL arrays, using Draw. */
-void r300_swtcl_draw_arrays(struct pipe_context* pipe,
- unsigned mode,
- unsigned start,
- unsigned count)
+static void r300_swtcl_draw_arrays(struct pipe_context* pipe,
+ unsigned mode,
+ unsigned start,
+ 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;
}
+ r300_update_derived_state(r300);
+
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_set_mapped_element_buffer(r300->draw, 0, 0, NULL);
draw_arrays(r300->draw, mode, start, count);
+ /* XXX Not sure whether this is the best fix.
+ * It prevents CS from being rejected and weird assertion failures. */
+ draw_flush(r300->draw);
+
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,
- unsigned indexSize,
- unsigned minIndex,
- unsigned maxIndex,
- unsigned mode,
- unsigned start,
- unsigned count)
+static void r300_swtcl_draw_range_elements(struct pipe_context* pipe,
+ struct pipe_resource* indexBuffer,
+ unsigned indexSize,
+ int indexBias,
+ unsigned minIndex,
+ unsigned maxIndex,
+ unsigned mode,
+ unsigned start,
+ 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;
}
+ r300_update_derived_state(r300);
+
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);
- draw_set_mapped_element_buffer_range(r300->draw, indexSize,
+ indices = pipe_buffer_map(pipe, indexBuffer,
+ PIPE_TRANSFER_READ, &ib_transfer);
+ draw_set_mapped_element_buffer_range(r300->draw, indexSize, indexBias,
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);
+ /* XXX Not sure whether this is the best fix.
+ * It prevents CS from being rejected and weird assertion failures. */
+ draw_flush(r300->draw);
+
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);
- draw_set_mapped_element_buffer_range(r300->draw, 0, start,
- start + count - 1, NULL);
+ pipe_buffer_unmap(pipe, indexBuffer,
+ ib_transfer);
+ draw_set_mapped_element_buffer_range(r300->draw, 0, 0,
+ start, start + count - 1,
+ NULL);
}
/* Object for rendering using Draw. */
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*
struct r300_render* r300render = r300_render(render);
struct r300_context* r300 = r300render->r300;
- r300_update_derived_state(r300);
-
return &r300->vertex_info;
}
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;
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);
+ assert(!r300render->vbo_transfer);
+
+ 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);
}
ushort max)
{
struct r300_render* r300render = r300_render(render);
- struct pipe_screen* screen = r300render->r300->context.screen;
- CS_LOCALS(r300render->r300);
- BEGIN_CS(2);
- OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max);
- END_CS;
+ struct pipe_context* context = &r300render->r300->context;
+
+ assert(r300render->vbo_transfer);
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);
+
+ r300render->vbo_transfer = NULL;
}
static void r300_render_release_vertices(struct vbuf_render* render)
}
static void r300_render_draw_arrays(struct vbuf_render* render,
- unsigned start,
- unsigned count)
+ unsigned start,
+ unsigned count)
{
struct r300_render* r300render = r300_render(render);
struct r300_context* r300 = r300render->r300;
+ uint8_t* ptr;
+ unsigned i;
+ unsigned dwords = 6;
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);
+ (void) i; (void) ptr;
+
+ r300_prepare_for_rendering(r300, PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL,
+ NULL, dwords, 0, 0, NULL);
DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
- BEGIN_CS(2);
+ /* Uncomment to dump all VBOs rendered through this interface.
+ * Slow and noisy!
+ ptr = pipe_buffer_map(&r300render->r300->context,
+ r300render->vbo, PIPE_TRANSFER_READ,
+ &r300render->vbo_transfer);
+
+ for (i = 0; i < count; i++) {
+ printf("r300: Vertex %d\n", i);
+ draw_dump_emitted_vertex(&r300->vertex_info, ptr);
+ ptr += r300->vertex_info.size * 4;
+ printf("\n");
+ }
+
+ pipe_buffer_unmap(&r300render->r300->context, r300render->vbo,
+ r300render->vbo_transfer);
+ */
+
+ BEGIN_CS(dwords);
+ OUT_CS_REG(R300_GA_COLOR_CONTROL,
+ r300_provoking_vertex_fixes(r300, r300render->prim));
+ OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
r300render->hwprim);
END_CS;
}
-static void r300_render_draw(struct vbuf_render* render,
- const ushort* indices,
- uint count)
+static void r300_render_draw_elements(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;
+ unsigned end_cs_dwords;
+ unsigned max_index = (r300render->vbo_size - r300render->vbo_offset) /
+ (r300render->r300->vertex_info.size * 4) - 1;
+ unsigned short_count;
+ struct r300_cs_info cs_info;
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);
+ /* Reserve at least 256 dwords.
+ *
+ * Below we manage the CS space manually because there may be more
+ * indices than it can fit in CS. */
+ r300_prepare_for_rendering(r300,
+ PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL | PREP_INDEXED,
+ NULL, 256, 0, 0, &end_cs_dwords);
+
+ while (count) {
+ r300->rws->get_cs_info(r300->rws, &cs_info);
+
+ short_count = MIN2(count, (cs_info.free - end_cs_dwords - 6) * 2);
+
+ BEGIN_CS(6 + (short_count+1)/2);
+ OUT_CS_REG(R300_GA_COLOR_CONTROL,
+ r300_provoking_vertex_fixes(r300, r300render->prim));
+ OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max_index);
+ OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (short_count+1)/2);
+ OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (short_count << 16) |
+ r300render->hwprim);
+ for (i = 0; i < short_count-1; i += 2) {
+ OUT_CS(indices[i+1] << 16 | indices[i]);
+ }
+ if (short_count % 2) {
+ OUT_CS(indices[short_count-1]);
+ }
+ END_CS;
- 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);
- for (i = 0; i < count-1; i += 2) {
- OUT_CS(indices[i+1] << 16 | indices[i]);
- }
- if (count % 2) {
- OUT_CS(indices[count-1]);
+ /* OK now subtract the emitted indices and see if we need to emit
+ * another draw packet. */
+ indices += short_count;
+ count -= short_count;
+
+ if (count) {
+ r300_prepare_for_rendering(r300,
+ PREP_EMIT_AOS_SWTCL | PREP_INDEXED,
+ NULL, 256, 0, 0, &end_cs_dwords);
+ }
}
- END_CS;
}
static void r300_render_destroy(struct vbuf_render* render)
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_elements = r300_render_draw_elements;
r300render->base.draw_arrays = r300_render_draw_arrays;
r300render->base.release_vertices = r300_render_release_vertices;
r300render->base.destroy = r300_render_destroy;
return stage;
}
+
+/****************************************************************************
+ * Two-sided stencil reference value fallback. It's designed to be as much
+ * separate from rest of the driver as possible.
+ ***************************************************************************/
+
+struct r300_stencilref_context {
+ void (*draw_arrays)(struct pipe_context *pipe,
+ unsigned mode, unsigned start, unsigned count);
+
+ void (*draw_range_elements)(
+ struct pipe_context *pipe, struct pipe_resource *indexBuffer,
+ unsigned indexSize, int indexBias, unsigned minIndex, unsigned maxIndex,
+ unsigned mode, unsigned start, unsigned count);
+
+ uint32_t rs_cull_mode;
+ uint32_t zb_stencilrefmask;
+ ubyte ref_value_front;
+};
+
+static boolean r300_stencilref_needed(struct r300_context *r300)
+{
+ struct r300_dsa_state *dsa = (struct r300_dsa_state*)r300->dsa_state.state;
+
+ return dsa->two_sided_stencil_ref ||
+ (dsa->two_sided &&
+ r300->stencil_ref.ref_value[0] != r300->stencil_ref.ref_value[1]);
+}
+
+/* Set drawing for front faces. */
+static void r300_stencilref_begin(struct r300_context *r300)
+{
+ struct r300_stencilref_context *sr = r300->stencilref_fallback;
+ 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;
+
+ /* Save state. */
+ sr->rs_cull_mode = rs->cull_mode;
+ sr->zb_stencilrefmask = dsa->stencil_ref_mask;
+ sr->ref_value_front = r300->stencil_ref.ref_value[0];
+
+ /* We *cull* pixels, therefore no need to mask out the bits. */
+ rs->cull_mode |= R300_CULL_BACK;
+
+ r300->rs_state.dirty = TRUE;
+}
+
+/* Set drawing for back faces. */
+static void r300_stencilref_switch_side(struct r300_context *r300)
+{
+ struct r300_stencilref_context *sr = r300->stencilref_fallback;
+ 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;
+
+ rs->cull_mode = sr->rs_cull_mode | R300_CULL_FRONT;
+ dsa->stencil_ref_mask = dsa->stencil_ref_bf;
+ r300->stencil_ref.ref_value[0] = r300->stencil_ref.ref_value[1];
+
+ r300->rs_state.dirty = TRUE;
+ r300->dsa_state.dirty = TRUE;
+}
+
+/* Restore the original state. */
+static void r300_stencilref_end(struct r300_context *r300)
+{
+ struct r300_stencilref_context *sr = r300->stencilref_fallback;
+ 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;
+
+ /* Restore state. */
+ rs->cull_mode = sr->rs_cull_mode;
+ dsa->stencil_ref_mask = sr->zb_stencilrefmask;
+ r300->stencil_ref.ref_value[0] = sr->ref_value_front;
+
+ r300->rs_state.dirty = TRUE;
+ r300->dsa_state.dirty = TRUE;
+}
+
+static void r300_stencilref_draw_arrays(struct pipe_context *pipe, unsigned mode,
+ unsigned start, unsigned count)
+{
+ struct r300_context *r300 = r300_context(pipe);
+ struct r300_stencilref_context *sr = r300->stencilref_fallback;
+
+ if (!r300_stencilref_needed(r300)) {
+ sr->draw_arrays(pipe, mode, start, count);
+ } else {
+ r300_stencilref_begin(r300);
+ sr->draw_arrays(pipe, mode, start, count);
+ r300_stencilref_switch_side(r300);
+ sr->draw_arrays(pipe, mode, start, count);
+ r300_stencilref_end(r300);
+ }
+}
+
+static void r300_stencilref_draw_range_elements(
+ struct pipe_context *pipe, struct pipe_resource *indexBuffer,
+ unsigned indexSize, int indexBias, unsigned minIndex, unsigned maxIndex,
+ unsigned mode, unsigned start, unsigned count)
+{
+ struct r300_context *r300 = r300_context(pipe);
+ struct r300_stencilref_context *sr = r300->stencilref_fallback;
+
+ if (!r300_stencilref_needed(r300)) {
+ sr->draw_range_elements(pipe, indexBuffer, indexSize, indexBias,
+ minIndex, maxIndex, mode, start, count);
+ } else {
+ r300_stencilref_begin(r300);
+ sr->draw_range_elements(pipe, indexBuffer, indexSize, indexBias,
+ minIndex, maxIndex, mode, start, count);
+ r300_stencilref_switch_side(r300);
+ sr->draw_range_elements(pipe, indexBuffer, indexSize, indexBias,
+ minIndex, maxIndex, mode, start, count);
+ r300_stencilref_end(r300);
+ }
+}
+
+static void r300_plug_in_stencil_ref_fallback(struct r300_context *r300)
+{
+ r300->stencilref_fallback = CALLOC_STRUCT(r300_stencilref_context);
+
+ /* Save original draw functions. */
+ r300->stencilref_fallback->draw_arrays = r300->context.draw_arrays;
+ r300->stencilref_fallback->draw_range_elements = r300->context.draw_range_elements;
+
+ /* Override the draw functions. */
+ r300->context.draw_arrays = r300_stencilref_draw_arrays;
+ r300->context.draw_range_elements = r300_stencilref_draw_range_elements;
+}
+
+void r300_init_render_functions(struct r300_context *r300)
+{
+ /* Set generic functions. */
+ r300->context.draw_elements = r300_draw_elements;
+
+ /* Set draw functions based on presence of HW TCL. */
+ if (r300->screen->caps.has_tcl) {
+ r300->context.draw_arrays = r300_draw_arrays;
+ r300->context.draw_range_elements = r300_draw_range_elements;
+ } else {
+ r300->context.draw_arrays = r300_swtcl_draw_arrays;
+ r300->context.draw_range_elements = r300_swtcl_draw_range_elements;
+ }
+
+ /* Plug in two-sided stencil reference value fallback if needed. */
+ if (!r300->screen->caps.is_r500)
+ r300_plug_in_stencil_ref_fallback(r300);
+}