X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fr300%2Fr300_render.c;h=60700cf30372888d7f726f85854860395194df98;hb=ab130400cf91ab471e265e58193c95f04c7aeeda;hp=952ff9938b05fc70e9f81f75df2e7b58dcbfd18e;hpb=bb45a15168e6c6678776fea5d8d0085454089260;p=mesa.git diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c index 952ff9938b0..60700cf3037 100644 --- a/src/gallium/drivers/r300/r300_render.c +++ b/src/gallium/drivers/r300/r300_render.c @@ -1,5 +1,6 @@ /* * Copyright 2009 Corbin Simpson + * Copyright 2010 Marek Olšák * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -40,6 +41,10 @@ #include "r300_reg.h" #include "r300_state_derived.h" +#include + +#define IMMD_DWORDS 32 + static uint32_t r300_translate_primitive(unsigned prim) { switch (prim) { @@ -113,55 +118,88 @@ static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300, return color_control; } -static void r500_emit_index_offset(struct r300_context *r300, int index_bias) +boolean r500_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); +} + +void r500_emit_index_bias(struct r300_context *r300, int index_bias) { CS_LOCALS(r300); - if (r300->screen->caps.is_r500 && - r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0)) { - BEGIN_CS(2); - OUT_CS_REG(R500_VAP_INDEX_OFFSET, - (index_bias & 0xFFFFFF) | (index_bias < 0 ? 1<<24 : 0)); - END_CS; - } else { - if (index_bias) { - fprintf(stderr, "r300: Non-zero index bias is unsupported " - "on this hardware.\n"); - assert(0); + 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), - PREP_VALIDATE_VBOS = (1 << 1), - PREP_EMIT_AOS = (1 << 2), - PREP_EMIT_AOS_SWTCL = (1 << 3), - PREP_INDEXED = (1 << 4) + 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. - * Return TRUE if flush occured. */ -static void r300_prepare_for_rendering(struct r300_context *r300, - enum r300_prepare_flags flags, - struct pipe_resource *index_buffer, - unsigned cs_dwords, - unsigned aos_offset, - int index_bias, - unsigned *end_cs_dwords) +/** + * Check if the requested number of dwords is available in the CS and + * if not, flush. + * \param r300 The context. + * \param flags See r300_prepare_flags. + * \param cs_dwords The number of dwords to reserve in CS. + * \return TRUE if the CS was flushed + */ +static boolean r300_reserve_cs_dwords(struct r300_context *r300, + enum r300_prepare_flags flags, + unsigned cs_dwords) { - boolean flushed = FALSE; - boolean first_draw = flags & PREP_FIRST_DRAW; - boolean emit_aos = flags & PREP_EMIT_AOS; + 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; - unsigned end_dwords = 0; + boolean hw_index_bias = r500_index_bias_supported(r300); /* Add dirty state, index offset, and AOS. */ if (first_draw) { cs_dwords += r300_get_num_dirty_dwords(r300); - if (r300->screen->caps.is_r500) + if (hw_index_bias) cs_dwords += 2; /* emit_index_offset */ if (emit_aos) @@ -171,30 +209,87 @@ static void r300_prepare_for_rendering(struct r300_context *r300, cs_dwords += 7; /* emit_aos_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->cs->ndw - 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 aos_offset The offset passed to emit_aos. + * \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 aos_offset, + int index_bias) +{ + 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 = r500_index_bias_supported(r300); + /* 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) { + if (!r300_emit_buffer_validate(r300, flags & PREP_VALIDATE_VBOS, + index_buffer)) { + fprintf(stderr, "r300: CS space validation failed. " + "(not enough memory?) Skipping rendering.\n"); + return FALSE; + } + r300_emit_dirty_state(r300); - r500_emit_index_offset(r300, index_bias); + 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, flags & PREP_INDEXED); + r300_emit_aos(r300, aos_offset, indexed); + if (emit_aos_swtcl) - r300_emit_aos_swtcl(r300, flags & PREP_INDEXED); + r300_emit_aos_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 aos_offset The offset passed to emit_aos. + * \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 aos_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, aos_offset, index_bias); } static boolean immd_is_good_idea(struct r300_context *r300, @@ -214,13 +309,12 @@ 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; @@ -228,6 +322,10 @@ static boolean immd_is_good_idea(struct r300_context *r300, 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)) { @@ -241,8 +339,7 @@ static boolean immd_is_good_idea(struct r300_context *r300, } /***************************************************************************** - * 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, @@ -253,51 +350,50 @@ 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] = util_format_get_blocksize(velem->src_format) / 4; - vertex_size += size[i]; + size[i] = r300->velems->hw_format_size[i] / 4; 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)); @@ -312,13 +408,7 @@ 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->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; @@ -327,10 +417,10 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300, 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; + transfer[vbi] = NULL; } } } @@ -422,73 +512,11 @@ static void r300_emit_draw_elements(struct r300_context *r300, (0 << R300_INDX_BUFFER_SKIP_SHIFT)); OUT_CS(offset_dwords << 2); OUT_CS_BUF_RELOC(indexBuffer, count_dwords, - R300_DOMAIN_GTT, 0, 0); + r300_buffer(indexBuffer)->domain, 0); END_CS; } -static void r300_shorten_ubyte_elts(struct r300_context* r300, - struct pipe_resource** elts, - unsigned start, - unsigned count) -{ - struct pipe_context* context = &r300->context; - struct pipe_screen* screen = r300->context.screen; - struct pipe_resource* new_elts; - unsigned char *in_map; - unsigned short *out_map; - struct pipe_transfer *src_transfer, *dst_transfer; - unsigned i; - - new_elts = pipe_buffer_create(screen, - PIPE_BIND_INDEX_BUFFER, - 2 * count); - - 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; - in_map++; - out_map++; - } - - 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. */ static void r300_draw_range_elements(struct pipe_context* pipe, struct pipe_resource* indexBuffer, @@ -506,37 +534,30 @@ static void r300_draw_range_elements(struct pipe_context* pipe, 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 */ + unsigned new_offset; - if (r300->skip_rendering) { - return; + if (indexBias && !r500_index_bias_supported(r300)) { + r300_split_index_bias(r300, indexBias, &buffer_offset, &index_offset); } - if (!u_trim_pipe_prim(mode, &count)) { - return; - } - - if (indexSize == 1) { - 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; - } + r300_translate_index_buffer(r300, &indexBuffer, &indexSize, index_offset, + &start, count); r300_update_derived_state(r300); - r300_upload_index_buffer(r300, &indexBuffer, indexSize, start, count); + r300_upload_index_buffer(r300, &indexBuffer, indexSize, start, count, &new_offset); + + start = new_offset; - /* 15 dwords for emit_draw_elements */ - r300_prepare_for_rendering(r300, - PREP_FIRST_DRAW | PREP_VALIDATE_VBOS | PREP_EMIT_AOS | PREP_INDEXED, - indexBuffer, 15, 0, indexBias, NULL); + /* 15 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, 15, 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); } else { do { short_count = MIN2(count, 65534); @@ -549,31 +570,20 @@ static void r300_draw_range_elements(struct pipe_context* pipe, /* 15 dwords for emit_draw_elements */ if (count) { - r300_prepare_for_rendering(r300, - PREP_VALIDATE_VBOS | PREP_EMIT_AOS | PREP_INDEXED, - indexBuffer, 15, 0, indexBias, NULL); + if (!r300_prepare_for_rendering(r300, + PREP_VALIDATE_VBOS | PREP_EMIT_AOS | PREP_INDEXED, + indexBuffer, 15, buffer_offset, indexBias)) + goto done; } } while (count); } +done: if (indexBuffer != orgIndexBuffer) { pipe_resource_reference( &indexBuffer, NULL ); } } -/* 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, unsigned start, unsigned count) { @@ -583,22 +593,16 @@ static void r300_draw_arrays(struct pipe_context* pipe, unsigned mode, 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; - } - 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); @@ -610,123 +614,146 @@ static void r300_draw_arrays(struct pipe_context* pipe, unsigned mode, 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); } } -/**************************************************************************** - * 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, r300->vertex_buffer[i].buffer, + 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, r300->index_buffer.buffer, ib_transfer); + draw_set_mapped_index_buffer(r300->draw, NULL); + } } /* Object for rendering using Draw. */ @@ -743,9 +770,6 @@ struct r300_render { unsigned hwprim; /* VBO */ - struct pipe_resource* vbo; - size_t vbo_size; - size_t vbo_offset; size_t vbo_max_used; void * vbo_ptr; @@ -776,35 +800,40 @@ static boolean r300_render_allocate_vertices(struct vbuf_render* render, 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; } 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, @@ -813,12 +842,15 @@ 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, r300->vbo, r300render->vbo_transfer); r300render->vbo_transfer = NULL; } @@ -826,8 +858,11 @@ static void r300_render_unmap_vertices(struct vbuf_render* render, 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; } @@ -853,13 +888,21 @@ static void r300_render_draw_arrays(struct vbuf_render* render, 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! @@ -886,6 +929,8 @@ static void r300_render_draw_arrays(struct vbuf_render* render, 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, @@ -896,25 +941,35 @@ 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->cs->ndw - 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, @@ -937,11 +992,16 @@ static void r300_render_draw_elements(struct vbuf_render* render, 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) @@ -969,10 +1029,6 @@ static struct vbuf_render* r300_render_create(struct r300_context* r300) 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; } @@ -999,150 +1055,153 @@ struct draw_stage* r300_draw_stage(struct r300_context* r300) 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) +void r300_draw_flush_vbuf(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]); + pipe_resource_reference(&r300->vbo, NULL); + r300->draw_vbo_size = 0; } -/* 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; +/**************************************************************************** + * End of SW TCL functions * + ***************************************************************************/ - /* 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]; +/* 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); - /* We *cull* pixels, therefore no need to mask out the bits. */ - rs->cull_mode |= R300_CULL_BACK; + if (type == UTIL_BLITTER_ATTRIB_TEXCOORD) + r300->sprite_coord_enable = 1; - r300->rs_state.dirty = TRUE; -} + r300_update_derived_state(r300); -/* 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; + /* Mark some states we don't care about as non-dirty. */ + r300->clip_state.dirty = FALSE; + r300->viewport_state.dirty = FALSE; - 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]; + if (!r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0)) + goto done; - r300->rs_state.dirty = TRUE; - r300->dsa_state.dirty = TRUE; -} + DBG(r300, DBG_DRAW, "r300: draw_rectangle\n"); -/* 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; + 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]); + } - /* 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; + /* 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); - r300->rs_state.dirty = TRUE; - r300->dsa_state.dirty = TRUE; -} + /* 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); -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; + OUT_CS_32F(x1 + width * 0.5f); + OUT_CS_32F(y1 + height * 0.5f); + OUT_CS_32F(depth); + OUT_CS_32F(1); - 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); + if (vertex_size == 8) { + if (!attrib) + attrib = zeros; + OUT_CS_TABLE(attrib, 4); } -} + END_CS; -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; +done: + /* Restore the state. */ + r300->clip_state.dirty = TRUE; + r300->rs_state.dirty = TRUE; + r300->viewport_state.dirty = TRUE; - 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); - } + r300->sprite_coord_enable = last_sprite_coord_enable; } -static void r300_plug_in_stencil_ref_fallback(struct r300_context *r300) +static void r300_resource_resolve(struct pipe_context* pipe, + struct pipe_resource* dest, + struct pipe_subresource subdest, + struct pipe_resource* src, + struct pipe_subresource subsrc) { - 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; + struct r300_context* r300 = r300_context(pipe); + struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state; + struct pipe_surface* srcsurf = src->screen->get_tex_surface(src->screen, + src, subsrc.face, subsrc.level, 0, 0); + float color[] = {0, 0, 0, 0}; + + DBG(r300, DBG_DRAW, "r300: Resolving resource...\n"); + + /* Enable AA resolve. */ + aa->dest = r300_surface( + dest->screen->get_tex_surface(dest->screen, dest, subdest.face, + subdest.level, 0, 0)); + + aa->aaresolve_ctl = + R300_RB3D_AARESOLVE_CTL_AARESOLVE_MODE_RESOLVE | + R300_RB3D_AARESOLVE_CTL_AARESOLVE_ALPHA_AVERAGE; + r300->aa_state.size = 12; + r300->aa_state.dirty = TRUE; + + /* 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->aa_state.dirty = TRUE; + + 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 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; + 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; } - /* Plug in two-sided stencil reference value fallback if needed. */ + 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); }