X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fr300%2Fr300_render.c;h=53728431a635d59d98dc3c1d7ea59854b8378afd;hb=0e6d7ce0178d65787e3e10f56638c2f0a88296f1;hp=93bf388776c7b70279c86b7cdd08c845116aeb6e;hpb=2689dd304c6d644b04c941e6da63e466be5de0d6;p=mesa.git diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c index 93bf388776c..53728431a63 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"), @@ -34,17 +35,18 @@ #include "util/u_prim.h" #include "r300_cs.h" +#include "r300_cb.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" -/* XXX The DRM rejects VAP_ALT_NUM_VERTICES.. */ -//#define ENABLE_ALT_NUM_VERTS +#include -uint32_t r300_translate_primitive(unsigned prim) +#define IMMD_DWORDS 32 + +static uint32_t r300_translate_primitive(unsigned prim) { switch (prim) { case PIPE_PRIM_POINTS: @@ -117,16 +119,140 @@ static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300, return color_control; } -/* Check if the requested number of dwords is available in the CS and - * if not, flush. Return TRUE if the flush occured. */ -static boolean r300_reserve_cs_space(struct r300_context *r300, - unsigned dwords) +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->rws->check_cs(r300->rws, 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, @@ -138,7 +264,15 @@ static boolean immd_is_good_idea(struct r300_context *r300, unsigned vertex_element_count = r300->velems->count; unsigned i, vbi; - if (count > 10) { + if (DBG_ON(r300, DBG_NO_IMMD)) { + return FALSE; + } + + if (r300->draw) { + return FALSE; + } + + if (count * r300->velems->vertex_size_dwords > IMMD_DWORDS) { return FALSE; } @@ -152,7 +286,9 @@ static boolean immd_is_good_idea(struct r300_context *r300, if (!checked[vbi]) { vbuf = &r300->vertex_buffer[vbi]; - if (r300_buffer_is_referenced(r300, vbuf->buffer)) { + 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; } @@ -162,6 +298,11 @@ static boolean immd_is_good_idea(struct r300_context *r300, 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, @@ -170,84 +311,75 @@ 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, dwords; /* Size of the vertex, in dwords. */ - unsigned vertex_size = 0; - - /* Offsets of the attribute, in dwords, from the start of the vertex. */ - unsigned offset[PIPE_MAX_ATTRIBS]; + unsigned vertex_size = r300->velems->vertex_size_dwords; /* 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}; + uint32_t* map[PIPE_MAX_ATTRIBS]; + uint32_t* mapelem[PIPE_MAX_ATTRIBS]; + struct pipe_transfer* transfer[PIPE_MAX_ATTRIBS] = {0}; - CS_LOCALS(r300); + CB_LOCALS; /* 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]; - map[vbi] = (uint32_t*)pipe_buffer_map(r300->context.screen, + if (!transfer[vbi]) { + map[vbi] = (uint32_t*)pipe_buffer_map(&r300->context, vbuf->buffer, - PIPE_BUFFER_USAGE_CPU_READ); - map[vbi] += vbuf->buffer_offset / 4; - stride[vbi] = vbuf->stride / 4; + PIPE_TRANSFER_READ, + &transfer[vbi]); + map[vbi] += (vbuf->buffer_offset / 4) + stride[i] * start; } + mapelem[i] = map[vbi] + (velem->src_offset / 4); } dwords = 9 + count * vertex_size; - r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + dwords); - r300_emit_buffer_validate(r300, FALSE, NULL); - r300_emit_dirty_state(r300); + r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0, NULL); - BEGIN_CS(dwords); - OUT_CS_REG(R300_GA_COLOR_CONTROL, + BEGIN_CS_AS_CB(r300, dwords); + OUT_CB_REG(R300_GA_COLOR_CONTROL, r300_provoking_vertex_fixes(r300, mode)); - OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size); - 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) | + OUT_CB_REG(R300_VAP_VTX_SIZE, vertex_size); + OUT_CB_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2); + OUT_CB(count - 1); + OUT_CB(0); + OUT_CB_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, count * vertex_size); + OUT_CB(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->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_CB_TABLE(&mapelem[i][stride[i] * v], size[i]); } } - END_CS; + END_CB; /* Unmap buffers. */ 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.screen, vbuf->buffer); - map[vbi] = NULL; + pipe_buffer_unmap(&r300->context, vbuf->buffer, transfer[vbi]); + transfer[vbi] = NULL; } } } @@ -256,19 +388,18 @@ static void r300_emit_draw_arrays(struct r300_context *r300, unsigned mode, unsigned count) { -#if defined(ENABLE_ALT_NUM_VERTS) boolean alt_num_verts = count > 65535; -#else - boolean alt_num_verts = FALSE; -#endif CS_LOCALS(r300); + 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(9); OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count); - } else { - BEGIN_CS(7); } OUT_CS_REG(R300_GA_COLOR_CONTROL, r300_provoking_vertex_fixes(r300, mode)); @@ -283,7 +414,7 @@ static void r300_emit_draw_arrays(struct r300_context *r300, } static void r300_emit_draw_elements(struct r300_context *r300, - struct pipe_buffer* indexBuffer, + struct pipe_resource* indexBuffer, unsigned indexSize, unsigned minIndex, unsigned maxIndex, @@ -293,25 +424,23 @@ static void r300_emit_draw_elements(struct r300_context *r300, { 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(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); + BEGIN_CS(13 + (alt_num_verts ? 2 : 0)); if (alt_num_verts) { - BEGIN_CS(15); OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count); - } else { - BEGIN_CS(13); } OUT_CS_REG(R300_GA_COLOR_CONTROL, r300_provoking_vertex_fixes(r300, mode)); @@ -342,204 +471,169 @@ 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, - RADEON_GEM_DOMAIN_GTT, 0, 0); + r300_buffer(indexBuffer)->domain, 0, 0); END_CS; } -static void r300_shorten_ubyte_elts(struct r300_context* r300, - struct pipe_buffer** elts, - unsigned start, - unsigned count) -{ - struct pipe_screen* screen = r300->context.screen; - struct pipe_buffer* new_elts; - unsigned char *in_map; - unsigned short *out_map; - 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); - - 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 += start; - - for (i = 0; i < count; i++) { - *out_map = (unsigned short)*in_map; - in_map++; - out_map++; - } - - pipe_buffer_unmap(screen, *elts); - pipe_buffer_unmap(screen, new_elts); - - *elts = new_elts; -} - -static void r300_align_ushort_elts(struct r300_context *r300, - struct pipe_buffer **elts, - unsigned start, unsigned count) -{ - struct pipe_screen* screen = r300->context.screen; - struct pipe_buffer* new_elts; - unsigned short *in_map; - unsigned short *out_map; - - new_elts = screen->buffer_create(screen, 32, - PIPE_BUFFER_USAGE_INDEX | - PIPE_BUFFER_USAGE_CPU_WRITE | - PIPE_BUFFER_USAGE_GPU_READ, - 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); - - memcpy(out_map, in_map+start, 2 * count); - - pipe_buffer_unmap(screen, *elts); - pipe_buffer_unmap(screen, new_elts); - - *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 */ + boolean translate = FALSE; + + if (r300->skip_rendering) { + return; + } 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; + /* Index buffer range checking. */ + if ((start + count) * indexSize > indexBuffer->width0) { + fprintf(stderr, "r300: Invalid index buffer range. Skipping rendering.\n"); + 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 (indexBias && !index_bias_supported(r300)) { + r300_split_index_bias(r300, indexBias, &buffer_offset, &index_offset); + } + + 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); - /* 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_buffer_reference( &indexBuffer, NULL ); + pipe_resource_reference( &indexBuffer, NULL ); + } + + if (translate) { + r300_end_vertex_translate(r300); } } /* 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; + 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 { - /* 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); } + + if (translate) { + r300_end_vertex_translate(r300); + } } /**************************************************************************** @@ -548,89 +642,105 @@ void r300_draw_arrays(struct pipe_context* pipe, unsigned mode, ***************************************************************************/ /* 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. */ @@ -647,11 +757,13 @@ struct r300_render { unsigned hwprim; /* VBO */ - struct pipe_buffer* vbo; + struct pipe_resource* vbo; size_t vbo_size; size_t vbo_offset; size_t vbo_max_used; void * vbo_ptr; + + struct pipe_transfer *vbo_transfer; }; static INLINE struct r300_render* @@ -666,8 +778,6 @@ r300_render_get_vertex_info(struct vbuf_render* render) struct r300_render* r300render = r300_render(render); struct r300_context* r300 = r300render->r300; - r300_update_derived_state(r300); - return &r300->vertex_info; } @@ -682,10 +792,9 @@ static boolean r300_render_allocate_vertices(struct vbuf_render* render, if (size + r300render->vbo_offset > r300render->vbo_size) { - pipe_buffer_reference(&r300->vbo, NULL); + pipe_resource_reference(&r300->vbo, NULL); r300render->vbo = pipe_buffer_create(screen, - 64, - PIPE_BUFFER_USAGE_VERTEX, + PIPE_BIND_VERTEX_BUFFER, R300_MAX_DRAW_VBO_SIZE); r300render->vbo_offset = 0; r300render->vbo_size = R300_MAX_DRAW_VBO_SIZE; @@ -701,10 +810,13 @@ static boolean r300_render_allocate_vertices(struct vbuf_render* render, static void* r300_render_map_vertices(struct vbuf_render* render) { struct r300_render* r300render = r300_render(render); - struct pipe_screen* screen = r300render->r300->context.screen; - r300render->vbo_ptr = pipe_buffer_map(screen, r300render->vbo, - PIPE_BUFFER_USAGE_CPU_WRITE); + 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); } @@ -714,15 +826,15 @@ static void r300_render_unmap_vertices(struct vbuf_render* render, ushort max) { struct r300_render* r300render = r300_render(render); - struct pipe_screen* screen = r300render->r300->context.screen; - 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) @@ -745,53 +857,105 @@ static boolean r300_render_set_primitive(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; + unsigned free_dwords; 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) { + free_dwords = r300->rws->get_cs_free_dwords(r300->rws); + + 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, + 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) @@ -814,7 +978,7 @@ static struct vbuf_render* r300_render_create(struct r300_context* r300) r300render->base.map_vertices = r300_render_map_vertices; r300render->base.unmap_vertices = r300_render_unmap_vertices; r300render->base.set_primitive = r300_render_set_primitive; - r300render->base.draw = r300_render_draw; + r300render->base.draw_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; @@ -848,3 +1012,66 @@ struct draw_stage* r300_draw_stage(struct r300_context* r300) return stage; } + +/**************************************************************************** + * End of SW TCL functions * + ***************************************************************************/ + +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) +{ + 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; + } else { + r300->context.draw_arrays = r300_swtcl_draw_arrays; + r300->context.draw_range_elements = r300_swtcl_draw_range_elements; + } + + r300->context.resource_resolve = r300_resource_resolve; + + /* Plug in the two-sided stencil reference value fallback if needed. */ + if (!r300->screen->caps.is_r500) + r300_plug_in_stencil_ref_fallback(r300); +}