#include <limits.h>
+#define IMMD_DWORDS 32
+
static uint32_t r300_translate_primitive(unsigned prim)
{
switch (prim) {
return color_control;
}
-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)
+void r500_emit_index_bias(struct r300_context *r300, int index_bias)
{
CS_LOCALS(r300);
/* 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
+ * in r300_emit_vertex_arrays (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. */
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_EMIT_AOS = (1 << 2), /* call emit_vertex_arrays? */
+ PREP_EMIT_AOS_SWTCL = (1 << 3), /* call emit_vertex_arrays_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.
+ * if not, flush.
* \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.
+ * \return TRUE if the CS was flushed
*/
-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)
+static boolean r300_reserve_cs_dwords(struct r300_context *r300,
+ enum r300_prepare_flags flags,
+ unsigned cs_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);
+ boolean emit_vertex_arrays = flags & PREP_EMIT_AOS;
+ boolean emit_vertex_arrays_swtcl = flags & PREP_EMIT_AOS_SWTCL;
/* Add dirty state, index offset, and AOS. */
if (first_draw) {
cs_dwords += r300_get_num_dirty_dwords(r300);
- if (hw_index_bias)
+ if (r300->screen->caps.index_bias_supported)
cs_dwords += 2; /* emit_index_offset */
- if (emit_aos)
- cs_dwords += 55; /* emit_aos */
+ if (emit_vertex_arrays)
+ cs_dwords += 55; /* emit_vertex_arrays */
- if (emit_aos_swtcl)
- cs_dwords += 7; /* emit_aos_swtcl */
+ if (emit_vertex_arrays_swtcl)
+ cs_dwords += 7; /* emit_vertex_arrays_swtcl */
}
- /* Emitted in flush. */
- end_dwords += 26; /* emit_query_end */
-
- cs_dwords += end_dwords;
+ cs_dwords += r300_get_num_cs_end_dwords(r300);
/* Reserve requested CS space. */
- if (!r300_check_cs(r300, cs_dwords)) {
+ if (cs_dwords > (R300_MAX_CMDBUF_DWORDS - r300->cs->cdw)) {
r300->context.flush(&r300->context, 0, NULL);
flushed = TRUE;
}
+ return flushed;
+}
+
+/**
+ * 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 buffer_offset The offset passed to emit_vertex_arrays.
+ * \param index_bias The index bias to emit.
+ * \return TRUE if rendering should be skipped
+ */
+static boolean r300_emit_states(struct r300_context *r300,
+ enum r300_prepare_flags flags,
+ struct pipe_resource *index_buffer,
+ int buffer_offset,
+ int index_bias)
+{
+ boolean first_draw = flags & PREP_FIRST_DRAW;
+ boolean emit_vertex_arrays = flags & PREP_EMIT_AOS;
+ boolean emit_vertex_arrays_swtcl = flags & PREP_EMIT_AOS_SWTCL;
+ boolean indexed = flags & PREP_INDEXED;
+ boolean validate_vbos = flags & PREP_VALIDATE_VBOS;
+
/* Validate buffers and emit dirty state if needed. */
- if (first_draw || flushed) {
- r300_emit_buffer_validate(r300, flags & PREP_VALIDATE_VBOS, index_buffer);
+ if (first_draw) {
+ /* upload buffers first */
+ if (r300->screen->caps.has_tcl && r300->any_user_vbs) {
+ r300_upload_user_buffers(r300);
+ r300->any_user_vbs = false;
+ }
+
+ if (r300->validate_buffers) {
+ if (!r300_emit_buffer_validate(r300, validate_vbos,
+ index_buffer)) {
+ fprintf(stderr, "r300: CS space validation failed. "
+ "(not enough memory?) Skipping rendering.\n");
+ return FALSE;
+ }
+
+ /* Consider the validation done only if everything was validated. */
+ if (validate_vbos) {
+ r300->validate_buffers = FALSE;
+ if (r300->any_user_vbs)
+ r300->upload_vb_validated = TRUE;
+ if (r300->index_buffer.buffer &&
+ r300_buffer_is_user_buffer(r300->index_buffer.buffer)) {
+ r300->upload_ib_validated = TRUE;
+ }
+ }
+ }
+
r300_emit_dirty_state(r300);
- if (hw_index_bias) {
+ if (r300->screen->caps.index_bias_supported) {
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_vertex_arrays)
+ r300_emit_vertex_arrays(r300, buffer_offset, indexed);
- if (emit_aos_swtcl)
- r300_emit_aos_swtcl(r300, indexed);
+ if (emit_vertex_arrays_swtcl)
+ r300_emit_vertex_arrays_swtcl(r300, indexed);
}
- if (end_cs_dwords)
- *end_cs_dwords = end_dwords;
+ return TRUE;
+}
+
+/**
+ * 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 buffer_offset The offset passed to emit_vertex_arrays.
+ * \param index_bias The index bias to emit.
+ * \return TRUE if rendering should be skipped
+ */
+static boolean r300_prepare_for_rendering(struct r300_context *r300,
+ enum r300_prepare_flags flags,
+ struct pipe_resource *index_buffer,
+ unsigned cs_dwords,
+ int buffer_offset,
+ int index_bias)
+{
+ if (r300_reserve_cs_dwords(r300, flags, cs_dwords))
+ flags |= PREP_FIRST_DRAW;
+
+ return r300_emit_states(r300, flags, index_buffer, buffer_offset, index_bias);
}
static boolean immd_is_good_idea(struct r300_context *r300,
return FALSE;
}
- if (count > 10) {
+ if (count * r300->velems->vertex_size_dwords > IMMD_DWORDS) {
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(vbuf->buffer)->domain & R300_DOMAIN_GTT)) {
+ return FALSE;
+ }
+
if (r300_buffer_is_referenced(&r300->context,
vbuf->buffer,
R300_REF_CS | R300_REF_HW)) {
}
/*****************************************************************************
- * The emission of draw packets for r500. Older GPUs may use these functions *
- * after resolving fallback issues (e.g. stencil ref two-sided). *
+ * The HWTCL draw functions. *
****************************************************************************/
static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
struct pipe_vertex_element* velem;
struct pipe_vertex_buffer* vbuf;
unsigned vertex_element_count = r300->velems->count;
- unsigned i, v, vbi, dw, elem_offset, dwords;
+ unsigned i, v, vbi;
/* Size of the vertex, in dwords. */
- unsigned vertex_size = 0;
+ unsigned vertex_size = r300->velems->vertex_size_dwords;
- /* Offsets of the attribute, in dwords, from the start of the vertex. */
- unsigned offset[PIPE_MAX_ATTRIBS];
+ /* The number of dwords for this draw operation. */
+ unsigned dwords = 9 + count * vertex_size;
/* Size of the vertex element, in dwords. */
unsigned size[PIPE_MAX_ATTRIBS];
/* Stride to the same attrib in the next vertex in the vertex buffer,
* in dwords. */
- unsigned stride[PIPE_MAX_ATTRIBS] = {0};
+ unsigned stride[PIPE_MAX_ATTRIBS];
/* Mapped vertex buffers. */
- uint32_t* map[PIPE_MAX_ATTRIBS] = {0};
- struct pipe_transfer* transfer[PIPE_MAX_ATTRIBS] = {NULL};
+ uint32_t* map[PIPE_MAX_ATTRIBS];
+ uint32_t* mapelem[PIPE_MAX_ATTRIBS];
+ struct pipe_transfer* transfer[PIPE_MAX_ATTRIBS] = {0};
CS_LOCALS(r300);
+ if (!r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0))
+ return;
+
/* Calculate the vertex size, offsets, strides etc. and map the buffers. */
for (i = 0; i < vertex_element_count; i++) {
velem = &r300->velems->velem[i];
- offset[i] = velem->src_offset / 4;
size[i] = r300->velems->hw_format_size[i] / 4;
- vertex_size += size[i];
vbi = velem->vertex_buffer_index;
+ vbuf = &r300->vertex_buffer[vbi];
+ stride[i] = vbuf->stride / 4;
/* Map the buffer. */
- if (!map[vbi]) {
- vbuf = &r300->vertex_buffer[vbi];
+ if (!transfer[vbi]) {
map[vbi] = (uint32_t*)pipe_buffer_map(&r300->context,
vbuf->buffer,
PIPE_TRANSFER_READ,
&transfer[vbi]);
- map[vbi] += vbuf->buffer_offset / 4;
- stride[vbi] = vbuf->stride / 4;
+ map[vbi] += (vbuf->buffer_offset / 4) + stride[i] * start;
}
+ mapelem[i] = map[vbi] + (velem->src_offset / 4);
}
- dwords = 9 + count * vertex_size;
-
- 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));
/* Emit vertices. */
for (v = 0; v < count; v++) {
for (i = 0; i < vertex_element_count; i++) {
- velem = &r300->velems->velem[i];
- vbi = velem->vertex_buffer_index;
- elem_offset = offset[i] + stride[vbi] * (v + start);
-
- for (dw = 0; dw < size[i]; dw++) {
- OUT_CS(map[vbi][elem_offset + dw]);
- }
+ OUT_CS_TABLE(&mapelem[i][stride[i] * v], size[i]);
}
}
END_CS;
for (i = 0; i < vertex_element_count; i++) {
vbi = r300->velems->velem[i].vertex_buffer_index;
- if (map[vbi]) {
+ if (transfer[vbi]) {
vbuf = &r300->vertex_buffer[vbi];
- pipe_buffer_unmap(&r300->context, vbuf->buffer, transfer[vbi]);
- map[vbi] = NULL;
+ pipe_buffer_unmap(&r300->context, transfer[vbi]);
+ transfer[vbi] = NULL;
}
}
}
unsigned maxIndex,
unsigned mode,
unsigned start,
- unsigned count)
+ unsigned count,
+ uint16_t *imm_indices3)
{
- uint32_t count_dwords;
- uint32_t offset_dwords = indexSize * start / sizeof(uint32_t);
+ uint32_t count_dwords, offset_dwords;
boolean alt_num_verts = count > 65535;
CS_LOCALS(r300);
DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, min %u max %u\n",
count, minIndex, maxIndex);
- BEGIN_CS(13 + (alt_num_verts ? 2 : 0));
- if (alt_num_verts) {
- OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
- }
+ BEGIN_CS(5);
OUT_CS_REG(R300_GA_COLOR_CONTROL,
r300_provoking_vertex_fixes(r300, mode));
OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
OUT_CS(maxIndex);
OUT_CS(minIndex);
+ END_CS;
+
+ /* If start is odd, render the first triangle with indices embedded
+ * in the command stream. This will increase start by 3 and make it
+ * even. We can then proceed without a fallback. */
+ if (indexSize == 2 && (start & 1) &&
+ mode == PIPE_PRIM_TRIANGLES) {
+ BEGIN_CS(4);
+ OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 2);
+ OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (3 << 16) |
+ R300_VAP_VF_CNTL__PRIM_TRIANGLES);
+ OUT_CS(imm_indices3[1] << 16 | imm_indices3[0]);
+ OUT_CS(imm_indices3[2]);
+ END_CS;
+
+ start += 3;
+ count -= 3;
+ if (!count)
+ return;
+ }
+
+ offset_dwords = indexSize * start / sizeof(uint32_t);
+
+ BEGIN_CS(8 + (alt_num_verts ? 2 : 0));
+ if (alt_num_verts) {
+ OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
+ }
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
if (indexSize == 4) {
count_dwords = count;
(0 << R300_INDX_BUFFER_SKIP_SHIFT));
OUT_CS(offset_dwords << 2);
OUT_CS_BUF_RELOC(indexBuffer, count_dwords,
- r300_buffer(indexBuffer)->domain, 0, 0);
+ r300_buffer(indexBuffer)->domain, 0);
END_CS;
}
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 */
- boolean translate = FALSE;
-
- if (r300->skip_rendering) {
- return;
- }
-
- if (!u_trim_pipe_prim(mode, &count)) {
- return;
- }
-
- /* Set up fallback for incompatible vertex layout if needed. */
- if (r300->incompatible_vb_layout || r300->velems->incompatible_layout) {
- r300_begin_vertex_translate(r300);
- translate = TRUE;
- }
+ uint16_t indices3[3];
- if (indexBias && !index_bias_supported(r300)) {
+ if (indexBias && !r300->screen->caps.index_bias_supported) {
r300_split_index_bias(r300, indexBias, &buffer_offset, &index_offset);
}
&start, count);
r300_update_derived_state(r300);
- r300_upload_index_buffer(r300, &indexBuffer, indexSize, start, count);
- /* 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);
+ /* Fallback for misaligned ushort indices. */
+ if (indexSize == 2 && (start & 1) &&
+ !r300_buffer_is_user_buffer(indexBuffer)) {
+ struct pipe_transfer *transfer;
+ struct pipe_resource *userbuf;
+
+ uint16_t *ptr = pipe_buffer_map(pipe, indexBuffer,
+ PIPE_TRANSFER_READ, &transfer);
+
+ if (mode == PIPE_PRIM_TRIANGLES) {
+ memcpy(indices3, ptr + start, 6);
+ } else {
+ /* Copy the mapped index buffer directly to the upload buffer.
+ * The start index will be aligned simply from the fact that
+ * every sub-buffer in u_upload_mgr is aligned. */
+ userbuf = pipe->screen->user_buffer_create(pipe->screen,
+ ptr, count * 2,
+ PIPE_BIND_INDEX_BUFFER);
+ indexBuffer = userbuf;
+ r300_upload_index_buffer(r300, &indexBuffer, indexSize, &start, count);
+ pipe_resource_reference(&userbuf, NULL);
+ }
+ pipe_buffer_unmap(pipe, transfer);
+ } else {
+ if (r300_buffer_is_user_buffer(indexBuffer))
+ r300_upload_index_buffer(r300, &indexBuffer, indexSize, &start, count);
+ }
+
+ /* 19 dwords for emit_draw_elements. Give up if the function fails. */
+ if (!r300_prepare_for_rendering(r300,
+ PREP_FIRST_DRAW | PREP_VALIDATE_VBOS | PREP_EMIT_AOS |
+ PREP_INDEXED, indexBuffer, 19, buffer_offset, indexBias))
+ goto done;
- 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);
+ minIndex, maxIndex, mode, start, count, indices3);
} else {
do {
short_count = MIN2(count, 65534);
r300_emit_draw_elements(r300, indexBuffer, indexSize,
minIndex, maxIndex,
- mode, start, short_count);
+ mode, start, short_count, indices3);
start += short_count;
count -= short_count;
/* 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);
+ if (!r300_prepare_for_rendering(r300,
+ PREP_VALIDATE_VBOS | PREP_EMIT_AOS | PREP_INDEXED,
+ indexBuffer, 19, buffer_offset, indexBias))
+ goto done;
}
} while (count);
}
+done:
if (indexBuffer != orgIndexBuffer) {
pipe_resource_reference( &indexBuffer, NULL );
}
-
- if (translate) {
- r300_end_vertex_translate(r300);
- }
-}
-
-/* Simple helpers for context setup. Should probably be moved to util. */
-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, indexBias,
- 0, r300->vertex_buffer_max_index,
- mode, start, count);
}
static void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
count > 65536 &&
r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
unsigned short_count;
- boolean translate = FALSE;
-
- if (r300->skip_rendering) {
- return;
- }
-
- if (!u_trim_pipe_prim(mode, &count)) {
- return;
- }
-
- /* Set up fallback for incompatible vertex layout if needed. */
- if (r300->incompatible_vb_layout || r300->velems->incompatible_layout) {
- r300_begin_vertex_translate(r300);
- translate = TRUE;
- }
r300_update_derived_state(r300);
if (immd_is_good_idea(r300, count)) {
r300_emit_draw_arrays_immediate(r300, mode, start, count);
} else {
- /* 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);
+ /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */
+ if (!r300_prepare_for_rendering(r300,
+ PREP_FIRST_DRAW | PREP_VALIDATE_VBOS | PREP_EMIT_AOS,
+ NULL, 9, start, 0))
+ return;
if (alt_num_verts || count <= 65535) {
r300_emit_draw_arrays(r300, mode, count);
start += short_count;
count -= short_count;
- /* 9 spare dwords for emit_draw_arrays. */
+ /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */
if (count) {
- r300_prepare_for_rendering(r300,
- PREP_VALIDATE_VBOS | PREP_EMIT_AOS, NULL, 9,
- start, 0, NULL);
+ if (!r300_prepare_for_rendering(r300,
+ PREP_VALIDATE_VBOS | PREP_EMIT_AOS, NULL, 9,
+ start, 0))
+ return;
}
} while (count);
}
- u_upload_flush(r300->upload_vb);
- }
-
- if (translate) {
- r300_end_vertex_translate(r300);
}
}
-/****************************************************************************
- * The rest of this file is for SW TCL rendering only. Please be polite and *
- * keep these functions separated so that they are easier to locate. ~C. *
- ***************************************************************************/
-
-/* SW TCL arrays, using Draw. */
-static void r300_swtcl_draw_arrays(struct pipe_context* pipe,
- unsigned mode,
- unsigned start,
- unsigned count)
+static void r300_draw_vbo(struct pipe_context* pipe,
+ const struct pipe_draw_info *info)
{
struct r300_context* r300 = r300_context(pipe);
- struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS];
- int i;
+ unsigned count = info->count;
+ boolean translate = FALSE;
+ boolean indexed = info->indexed && r300->index_buffer.buffer;
+ unsigned start_indexed = 0;
if (r300->skip_rendering) {
return;
}
- if (!u_trim_pipe_prim(mode, &count)) {
+ if (!u_trim_pipe_prim(info->mode, &count)) {
return;
}
- r300_update_derived_state(r300);
+ /* Index buffer range checking. */
+ if (indexed) {
+ assert(r300->index_buffer.offset % r300->index_buffer.index_size == 0);
- for (i = 0; i < r300->vertex_buffer_count; i++) {
- void* buf = pipe_buffer_map(pipe,
- r300->vertex_buffer[i].buffer,
- PIPE_TRANSFER_READ,
- &vb_transfer[i]);
- draw_set_mapped_vertex_buffer(r300->draw, i, buf);
- }
+ /* Compute start for draw_elements, taking the offset into account. */
+ start_indexed =
+ info->start +
+ (r300->index_buffer.offset / r300->index_buffer.index_size);
- draw_set_mapped_element_buffer(r300->draw, 0, 0, NULL);
+ if ((start_indexed + count) * r300->index_buffer.index_size >
+ r300->index_buffer.buffer->width0) {
+ fprintf(stderr, "r300: Invalid index buffer range. Skipping rendering.\n");
+ return;
+ }
+ }
- draw_arrays(r300->draw, mode, start, count);
+ /* Set up fallback for incompatible vertex layout if needed. */
+ if (r300->incompatible_vb_layout || r300->velems->incompatible_layout) {
+ r300_begin_vertex_translate(r300);
+ translate = TRUE;
+ }
- /* XXX Not sure whether this is the best fix.
- * It prevents CS from being rejected and weird assertion failures. */
- draw_flush(r300->draw);
+ if (indexed) {
+ r300_draw_range_elements(pipe,
+ r300->index_buffer.buffer,
+ r300->index_buffer.index_size,
+ info->index_bias,
+ info->min_index,
+ info->max_index,
+ info->mode,
+ start_indexed,
+ count);
+ } else {
+ r300_draw_arrays(pipe,
+ info->mode,
+ info->start,
+ count);
+ }
- for (i = 0; i < r300->vertex_buffer_count; i++) {
- pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer,
- vb_transfer[i]);
- draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
+ if (translate) {
+ r300_end_vertex_translate(r300);
}
}
+/****************************************************************************
+ * The rest of this file is for SW TCL rendering only. Please be polite and *
+ * keep these functions separated so that they are easier to locate. ~C. *
+ ***************************************************************************/
+
/* SW TCL elements, using Draw. */
-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)
+static void r300_swtcl_draw_vbo(struct pipe_context* pipe,
+ const struct pipe_draw_info *info)
{
struct r300_context* r300 = r300_context(pipe);
struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS];
- struct pipe_transfer *ib_transfer;
+ struct pipe_transfer *ib_transfer = NULL;
+ unsigned count = info->count;
int i;
- void* indices;
+ void *indices = NULL;
+ boolean indexed = info->indexed && r300->index_buffer.buffer;
if (r300->skip_rendering) {
return;
}
- if (!u_trim_pipe_prim(mode, &count)) {
+ if (!u_trim_pipe_prim(info->mode, &count)) {
return;
}
r300_update_derived_state(r300);
+ r300_reserve_cs_dwords(r300,
+ PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL |
+ (indexed ? PREP_INDEXED : 0),
+ indexed ? 256 : 6);
+
for (i = 0; i < r300->vertex_buffer_count; i++) {
- void* buf = pipe_buffer_map(pipe,
- r300->vertex_buffer[i].buffer,
- PIPE_TRANSFER_READ,
- &vb_transfer[i]);
- draw_set_mapped_vertex_buffer(r300->draw, i, buf);
+ if (r300->vertex_buffer[i].buffer) {
+ void *buf = pipe_buffer_map(pipe,
+ r300->vertex_buffer[i].buffer,
+ PIPE_TRANSFER_READ,
+ &vb_transfer[i]);
+ draw_set_mapped_vertex_buffer(r300->draw, i, buf);
+ }
}
- indices = pipe_buffer_map(pipe, indexBuffer,
- PIPE_TRANSFER_READ, &ib_transfer);
- draw_set_mapped_element_buffer_range(r300->draw, indexSize, indexBias,
- minIndex, maxIndex, indices);
+ if (indexed) {
+ indices = pipe_buffer_map(pipe, r300->index_buffer.buffer,
+ PIPE_TRANSFER_READ, &ib_transfer);
+ }
- draw_arrays(r300->draw, mode, start, count);
+ draw_set_mapped_index_buffer(r300->draw, indices);
- /* XXX Not sure whether this is the best fix.
- * It prevents CS from being rejected and weird assertion failures. */
+ r300->draw_vbo_locked = TRUE;
+ r300->draw_first_emitted = FALSE;
+ draw_vbo(r300->draw, info);
draw_flush(r300->draw);
+ r300->draw_vbo_locked = FALSE;
for (i = 0; i < r300->vertex_buffer_count; i++) {
- pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer,
- vb_transfer[i]);
- draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
+ if (r300->vertex_buffer[i].buffer) {
+ pipe_buffer_unmap(pipe, vb_transfer[i]);
+ draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
+ }
}
- pipe_buffer_unmap(pipe, indexBuffer,
- ib_transfer);
- draw_set_mapped_element_buffer_range(r300->draw, 0, 0,
- start, start + count - 1,
- NULL);
+ if (indexed) {
+ pipe_buffer_unmap(pipe, ib_transfer);
+ draw_set_mapped_index_buffer(r300->draw, NULL);
+ }
}
/* Object for rendering using Draw. */
unsigned hwprim;
/* VBO */
- struct pipe_resource* vbo;
- size_t vbo_size;
- size_t vbo_offset;
size_t vbo_max_used;
void * vbo_ptr;
struct pipe_screen* screen = r300->context.screen;
size_t size = (size_t)vertex_size * (size_t)count;
- if (size + r300render->vbo_offset > r300render->vbo_size)
+ DBG(r300, DBG_DRAW, "r300: render_allocate_vertices (size: %d)\n", size);
+
+ if (size + r300->draw_vbo_offset > r300->draw_vbo_size)
{
- pipe_resource_reference(&r300->vbo, NULL);
- r300render->vbo = pipe_buffer_create(screen,
- PIPE_BIND_VERTEX_BUFFER,
- R300_MAX_DRAW_VBO_SIZE);
- r300render->vbo_offset = 0;
- r300render->vbo_size = R300_MAX_DRAW_VBO_SIZE;
+ pipe_resource_reference(&r300->vbo, NULL);
+ r300->vbo = pipe_buffer_create(screen,
+ PIPE_BIND_VERTEX_BUFFER,
+ R300_MAX_DRAW_VBO_SIZE);
+ r300->draw_vbo_offset = 0;
+ r300->draw_vbo_size = R300_MAX_DRAW_VBO_SIZE;
+ r300->validate_buffers = TRUE;
}
r300render->vertex_size = vertex_size;
- r300->vbo = r300render->vbo;
- r300->vbo_offset = r300render->vbo_offset;
- return (r300render->vbo) ? TRUE : FALSE;
+ return (r300->vbo) ? TRUE : FALSE;
}
static void* r300_render_map_vertices(struct vbuf_render* render)
{
struct r300_render* r300render = r300_render(render);
+ struct r300_context* r300 = r300render->r300;
assert(!r300render->vbo_transfer);
+ DBG(r300, DBG_DRAW, "r300: render_map_vertices\n");
+
r300render->vbo_ptr = pipe_buffer_map(&r300render->r300->context,
- r300render->vbo,
+ r300->vbo,
PIPE_TRANSFER_WRITE,
&r300render->vbo_transfer);
- return ((uint8_t*)r300render->vbo_ptr + r300render->vbo_offset);
+ assert(r300render->vbo_ptr);
+
+ return ((uint8_t*)r300render->vbo_ptr + r300->draw_vbo_offset);
}
static void r300_render_unmap_vertices(struct vbuf_render* render,
{
struct r300_render* r300render = r300_render(render);
struct pipe_context* context = &r300render->r300->context;
+ struct r300_context* r300 = r300render->r300;
assert(r300render->vbo_transfer);
+ DBG(r300, DBG_DRAW, "r300: render_unmap_vertices\n");
+
r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
r300render->vertex_size * (max + 1));
- pipe_buffer_unmap(context, r300render->vbo, r300render->vbo_transfer);
+ pipe_buffer_unmap(context, r300render->vbo_transfer);
r300render->vbo_transfer = NULL;
}
static void r300_render_release_vertices(struct vbuf_render* render)
{
struct r300_render* r300render = r300_render(render);
+ struct r300_context* r300 = r300render->r300;
+
+ DBG(r300, DBG_DRAW, "r300: render_release_vertices\n");
- r300render->vbo_offset += r300render->vbo_max_used;
+ r300->draw_vbo_offset += r300render->vbo_max_used;
r300render->vbo_max_used = 0;
}
unsigned dwords = 6;
CS_LOCALS(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: render_draw_arrays (count: %d)\n", count);
- DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
+ if (r300->draw_first_emitted) {
+ if (!r300_prepare_for_rendering(r300,
+ PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL,
+ NULL, 6, 0, 0))
+ return;
+ } else {
+ if (!r300_emit_states(r300,
+ PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL,
+ NULL, 0, 0))
+ return;
+ }
/* Uncomment to dump all VBOs rendered through this interface.
* Slow and noisy!
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
r300render->hwprim);
END_CS;
+
+ r300->draw_first_emitted = TRUE;
}
static void r300_render_draw_elements(struct vbuf_render* render,
struct r300_context* r300 = r300render->r300;
int i;
unsigned end_cs_dwords;
- unsigned max_index = (r300render->vbo_size - r300render->vbo_offset) /
+ unsigned max_index = (r300->draw_vbo_size - r300->draw_vbo_offset) /
(r300render->r300->vertex_info.size * 4) - 1;
unsigned short_count;
- struct r300_cs_info cs_info;
+ unsigned free_dwords;
CS_LOCALS(r300);
+ DBG(r300, DBG_DRAW, "r300: render_draw_elements (count: %d)\n", count);
- /* Reserve at least 256 dwords.
- *
- * Below we manage the CS space manually because there may be more
+ if (r300->draw_first_emitted) {
+ if (!r300_prepare_for_rendering(r300,
+ PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL | PREP_INDEXED,
+ NULL, 256, 0, 0))
+ return;
+ } else {
+ if (!r300_emit_states(r300,
+ PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL | PREP_INDEXED,
+ NULL, 0, 0))
+ return;
+ }
+
+ /* 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);
+
+ end_cs_dwords = r300_get_num_cs_end_dwords(r300);
while (count) {
- r300->rws->get_cs_info(r300->rws, &cs_info);
+ free_dwords = R300_MAX_CMDBUF_DWORDS - r300->cs->cdw;
- short_count = MIN2(count, (cs_info.free - end_cs_dwords - 6) * 2);
+ short_count = MIN2(count, (free_dwords - end_cs_dwords - 6) * 2);
BEGIN_CS(6 + (short_count+1)/2);
OUT_CS_REG(R300_GA_COLOR_CONTROL,
count -= short_count;
if (count) {
- r300_prepare_for_rendering(r300,
- PREP_EMIT_AOS_SWTCL | PREP_INDEXED,
- NULL, 256, 0, 0, &end_cs_dwords);
+ if (!r300_prepare_for_rendering(r300,
+ PREP_EMIT_AOS_SWTCL | PREP_INDEXED,
+ NULL, 256, 0, 0))
+ return;
+
+ end_cs_dwords = r300_get_num_cs_end_dwords(r300);
}
}
+
+ r300->draw_first_emitted = TRUE;
}
static void r300_render_destroy(struct vbuf_render* render)
r300render->r300 = r300;
- /* XXX find real numbers plz */
- r300render->base.max_vertex_buffer_bytes = 128 * 1024;
+ r300render->base.max_vertex_buffer_bytes = 1024 * 1024;
r300render->base.max_indices = 16 * 1024;
r300render->base.get_vertex_info = r300_render_get_vertex_info;
r300render->base.release_vertices = r300_render_release_vertices;
r300render->base.destroy = r300_render_destroy;
- r300render->vbo = NULL;
- r300render->vbo_size = 0;
- r300render->vbo_offset = 0;
-
return &r300render->base;
}
return stage;
}
-void r300_init_render_functions(struct r300_context *r300)
+void r300_draw_flush_vbuf(struct r300_context *r300)
+{
+ pipe_resource_reference(&r300->vbo, NULL);
+ r300->draw_vbo_size = 0;
+}
+
+/****************************************************************************
+ * End of SW TCL functions *
+ ***************************************************************************/
+
+/* This functions is used to draw a rectangle for the blitter module.
+ *
+ * If we rendered a quad, the pixels on the main diagonal
+ * would be computed and stored twice, which makes the clear/copy codepaths
+ * somewhat inefficient. Instead we use a rectangular point sprite. */
+static void r300_blitter_draw_rectangle(struct blitter_context *blitter,
+ unsigned x1, unsigned y1,
+ unsigned x2, unsigned y2,
+ float depth,
+ enum blitter_attrib_type type,
+ const float attrib[4])
+{
+ struct r300_context *r300 = r300_context(util_blitter_get_pipe(blitter));
+ unsigned last_sprite_coord_enable = r300->sprite_coord_enable;
+ unsigned width = x2 - x1;
+ unsigned height = y2 - y1;
+ unsigned vertex_size =
+ type == UTIL_BLITTER_ATTRIB_COLOR || !r300->draw ? 8 : 4;
+ unsigned dwords = 13 + vertex_size +
+ (type == UTIL_BLITTER_ATTRIB_TEXCOORD ? 7 : 0);
+ const float zeros[4] = {0, 0, 0, 0};
+ CS_LOCALS(r300);
+
+ r300->context.set_vertex_buffers(&r300->context, 0, NULL);
+
+ if (type == UTIL_BLITTER_ATTRIB_TEXCOORD)
+ r300->sprite_coord_enable = 1;
+
+ r300_update_derived_state(r300);
+
+ /* Mark some states we don't care about as non-dirty. */
+ r300->clip_state.dirty = FALSE;
+ r300->viewport_state.dirty = FALSE;
+
+ if (!r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0))
+ goto done;
+
+ DBG(r300, DBG_DRAW, "r300: draw_rectangle\n");
+
+ BEGIN_CS(dwords);
+ /* Set up GA. */
+ OUT_CS_REG(R300_GA_POINT_SIZE, (height * 6) | ((width * 6) << 16));
+
+ if (type == UTIL_BLITTER_ATTRIB_TEXCOORD) {
+ /* Set up the GA to generate texcoords. */
+ OUT_CS_REG(R300_GB_ENABLE, R300_GB_POINT_STUFF_ENABLE |
+ (R300_GB_TEX_STR << R300_GB_TEX0_SOURCE_SHIFT));
+ OUT_CS_REG_SEQ(R300_GA_POINT_S0, 4);
+ OUT_CS_32F(attrib[0]);
+ OUT_CS_32F(attrib[3]);
+ OUT_CS_32F(attrib[2]);
+ OUT_CS_32F(attrib[1]);
+ }
+
+ /* Set up VAP controls. */
+ OUT_CS_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
+ OUT_CS_REG(R300_VAP_VTE_CNTL, R300_VTX_XY_FMT | R300_VTX_Z_FMT);
+ OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
+ OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
+ OUT_CS(1);
+ OUT_CS(0);
+
+ /* Draw. */
+ OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, vertex_size);
+ OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (1 << 16) |
+ R300_VAP_VF_CNTL__PRIM_POINTS);
+
+ OUT_CS_32F(x1 + width * 0.5f);
+ OUT_CS_32F(y1 + height * 0.5f);
+ OUT_CS_32F(depth);
+ OUT_CS_32F(1);
+
+ if (vertex_size == 8) {
+ if (!attrib)
+ attrib = zeros;
+ OUT_CS_TABLE(attrib, 4);
+ }
+ END_CS;
+
+done:
+ /* Restore the state. */
+ r300_mark_atom_dirty(r300, &r300->clip_state);
+ r300_mark_atom_dirty(r300, &r300->rs_state);
+ r300_mark_atom_dirty(r300, &r300->viewport_state);
+
+ r300->sprite_coord_enable = last_sprite_coord_enable;
+}
+
+static void r300_resource_resolve(struct pipe_context* pipe,
+ struct pipe_resource* dest,
+ unsigned dst_layer,
+ struct pipe_resource* src,
+ unsigned src_layer)
{
- /* Set generic functions. */
- r300->context.draw_elements = r300_draw_elements;
+ struct r300_context* r300 = r300_context(pipe);
+ struct pipe_surface* srcsurf, surf_tmpl;
+ struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state;
+ float color[] = {0, 0, 0, 0};
+
+ memset(&surf_tmpl, 0, sizeof(surf_tmpl));
+ surf_tmpl.format = src->format;
+ surf_tmpl.usage = 0; /* not really a surface hence no bind flags */
+ surf_tmpl.u.tex.level = 0; /* msaa resources cannot have mipmaps */
+ surf_tmpl.u.tex.first_layer = src_layer;
+ surf_tmpl.u.tex.last_layer = src_layer;
+ srcsurf = pipe->create_surface(pipe, src, &surf_tmpl);
+ surf_tmpl.format = dest->format;
+ surf_tmpl.u.tex.first_layer = dst_layer;
+ surf_tmpl.u.tex.last_layer = dst_layer;
+
+ DBG(r300, DBG_DRAW, "r300: Resolving resource...\n");
+
+ /* Enable AA resolve. */
+ aa->dest = r300_surface(pipe->create_surface(pipe, dest, &surf_tmpl));
+
+ aa->aaresolve_ctl =
+ R300_RB3D_AARESOLVE_CTL_AARESOLVE_MODE_RESOLVE |
+ R300_RB3D_AARESOLVE_CTL_AARESOLVE_ALPHA_AVERAGE;
+ r300->aa_state.size = 12;
+ r300_mark_atom_dirty(r300, &r300->aa_state);
+
+ /* Resolve the surface. */
+ r300->context.clear_render_target(pipe,
+ srcsurf, color, 0, 0, src->width0, src->height0);
+
+ /* Disable AA resolve. */
+ aa->aaresolve_ctl = 0;
+ r300->aa_state.size = 4;
+ r300_mark_atom_dirty(r300, &r300->aa_state);
+
+ pipe_surface_reference((struct pipe_surface**)&srcsurf, NULL);
+ pipe_surface_reference((struct pipe_surface**)&aa->dest, NULL);
+}
+void r300_init_render_functions(struct r300_context *r300)
+{
/* 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;
+ r300->context.draw_vbo = r300_draw_vbo;
} else {
- r300->context.draw_arrays = r300_swtcl_draw_arrays;
- r300->context.draw_range_elements = r300_swtcl_draw_range_elements;
+ r300->context.draw_vbo = r300_swtcl_draw_vbo;
}
+ r300->context.resource_resolve = r300_resource_resolve;
+ r300->blitter->draw_rectangle = r300_blitter_draw_rectangle;
+
/* Plug in the two-sided stencil reference value fallback if needed. */
if (!r300->screen->caps.is_r500)
r300_plug_in_stencil_ref_fallback(r300);