diff options
Diffstat (limited to 'src/gallium/drivers/r300/r300_render.c')
-rw-r--r-- | src/gallium/drivers/r300/r300_render.c | 641 |
1 files changed, 373 insertions, 268 deletions
diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c index b4197e0352..26594dabe4 100644 --- a/src/gallium/drivers/r300/r300_render.c +++ b/src/gallium/drivers/r300/r300_render.c @@ -39,7 +39,6 @@ #include "r300_screen_buffer.h" #include "r300_emit.h" #include "r300_reg.h" -#include "r300_state_derived.h" #include <limits.h> @@ -128,16 +127,30 @@ void r500_emit_index_bias(struct r300_context *r300, int index_bias) END_CS; } +static void r300_emit_draw_init(struct r300_context *r300, unsigned mode, + unsigned min_index, unsigned max_index) +{ + CS_LOCALS(r300); + + 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(max_index); + OUT_CS(min_index); + 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 + * 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. */ 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_buffer *vb, *vbufs = r300->vbuf_mgr->vertex_buffer; struct pipe_vertex_element *velem = r300->velems->velem; unsigned i, size; int max_neg_bias; @@ -164,10 +177,10 @@ static void r300_split_index_bias(struct r300_context *r300, int index_bias, } enum r300_prepare_flags { - PREP_FIRST_DRAW = (1 << 0), /* call emit_dirty_state and friends? */ + PREP_EMIT_STATES = (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? */ }; @@ -180,33 +193,33 @@ enum r300_prepare_flags { * \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) + 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 emit_aos_swtcl = flags & PREP_EMIT_AOS_SWTCL; + boolean first_draw = flags & PREP_EMIT_STATES; + 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 (r300->screen->caps.index_bias_supported) + if (r300->screen->caps.is_r500) 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 */ } cs_dwords += r300_get_num_cs_end_dwords(r300); /* Reserve requested CS space. */ if (cs_dwords > (R300_MAX_CMDBUF_DWORDS - r300->cs->cdw)) { - r300->context.flush(&r300->context, 0, NULL); + r300_flush(&r300->context, R300_FLUSH_ASYNC, NULL); flushed = TRUE; } @@ -218,57 +231,55 @@ static boolean r300_reserve_cs_dwords(struct r300_context *r300, * \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 buffer_offset The offset passed to emit_vertex_arrays. * \param index_bias The index bias to emit. + * \param instance_id Index of instance to render * \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) + int buffer_offset, + int index_bias, int instance_id) { - boolean first_draw = flags & PREP_FIRST_DRAW; - boolean emit_aos = flags & PREP_EMIT_AOS; - boolean emit_aos_swtcl = flags & PREP_EMIT_AOS_SWTCL; + boolean first_draw = flags & PREP_EMIT_STATES; + 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) { - /* 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_emit_buffer_validate(r300, validate_vbos, + index_buffer)) { + fprintf(stderr, "r300: CS space validation failed. " + "(not enough memory?) Skipping rendering.\n"); + return FALSE; } r300_emit_dirty_state(r300); - if (r300->screen->caps.index_bias_supported) { + if (r300->screen->caps.is_r500) { 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->vertex_arrays_dirty || + r300->vertex_arrays_indexed != indexed || + r300->vertex_arrays_offset != buffer_offset || + r300->vertex_arrays_instance_id != instance_id)) { + r300_emit_vertex_arrays(r300, buffer_offset, indexed, instance_id); + + r300->vertex_arrays_dirty = FALSE; + r300->vertex_arrays_indexed = indexed; + r300->vertex_arrays_offset = buffer_offset; + r300->vertex_arrays_instance_id = instance_id; + } - if (emit_aos_swtcl) - r300_emit_aos_swtcl(r300, indexed); + if (emit_vertex_arrays_swtcl) + r300_emit_vertex_arrays_swtcl(r300, indexed); } return TRUE; @@ -281,28 +292,32 @@ static boolean r300_emit_states(struct r300_context *r300, * \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 buffer_offset The offset passed to emit_vertex_arrays. * \param index_bias The index bias to emit. + * \param instance_id The instance to render. * \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) + int buffer_offset, + int index_bias, + int instance_id) { + /* Make sure there is enough space in the command stream and emit states. */ if (r300_reserve_cs_dwords(r300, flags, cs_dwords)) - flags |= PREP_FIRST_DRAW; + flags |= PREP_EMIT_STATES; - return r300_emit_states(r300, flags, index_buffer, aos_offset, index_bias); + return r300_emit_states(r300, flags, index_buffer, buffer_offset, + index_bias, instance_id); } static boolean immd_is_good_idea(struct r300_context *r300, unsigned count) { struct pipe_vertex_element* velem; - struct pipe_vertex_buffer* vbuf; + struct pipe_resource *buf; boolean checked[PIPE_MAX_ATTRIBS] = {0}; unsigned vertex_element_count = r300->velems->count; unsigned i, vbi; @@ -326,18 +341,12 @@ static boolean immd_is_good_idea(struct r300_context *r300, vbi = velem->vertex_buffer_index; if (!checked[vbi]) { - vbuf = &r300->vertex_buffer[vbi]; + buf = r300->vbuf_mgr->real_vertex_buffer[vbi]; - if (!(r300_buffer(vbuf->buffer)->domain & R300_DOMAIN_GTT)) { + if ((r300_resource(buf)->domain != R300_DOMAIN_GTT)) { return FALSE; } - if (r300_buffer_is_referenced(&r300->context, - vbuf->buffer, - R300_REF_CS | R300_REF_HW)) { - /* It's a very bad idea to map it... */ - return FALSE; - } checked[vbi] = TRUE; } } @@ -348,10 +357,8 @@ static boolean immd_is_good_idea(struct r300_context *r300, * The HWTCL draw functions. * ****************************************************************************/ -static void r300_emit_draw_arrays_immediate(struct r300_context *r300, - unsigned mode, - unsigned start, - unsigned count) +static void r300_draw_arrays_immediate(struct r300_context *r300, + const struct pipe_draw_info *info) { struct pipe_vertex_element* velem; struct pipe_vertex_buffer* vbuf; @@ -362,7 +369,7 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300, unsigned vertex_size = r300->velems->vertex_size_dwords; /* The number of dwords for this draw operation. */ - unsigned dwords = 9 + count * vertex_size; + unsigned dwords = 4 + info->count * vertex_size; /* Size of the vertex element, in dwords. */ unsigned size[PIPE_MAX_ATTRIBS]; @@ -372,47 +379,42 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300, unsigned stride[PIPE_MAX_ATTRIBS]; /* Mapped vertex buffers. */ - uint32_t* map[PIPE_MAX_ATTRIBS]; + uint32_t* map[PIPE_MAX_ATTRIBS] = {0}; 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)) + if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0, -1)) 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]; - size[i] = r300->velems->hw_format_size[i] / 4; + size[i] = r300->velems->format_size[i] / 4; vbi = velem->vertex_buffer_index; - vbuf = &r300->vertex_buffer[vbi]; + vbuf = &r300->vbuf_mgr->vertex_buffer[vbi]; stride[i] = vbuf->stride / 4; /* Map the buffer. */ - 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[i] * start; + if (!map[vbi]) { + map[vbi] = (uint32_t*)r300->rws->buffer_map( + r300_resource(r300->vbuf_mgr->real_vertex_buffer[vbi])->buf, + r300->cs, PIPE_TRANSFER_READ | PIPE_TRANSFER_UNSYNCHRONIZED); + map[vbi] += (vbuf->buffer_offset / 4) + stride[i] * info->start; } mapelem[i] = map[vbi] + (velem->src_offset / 4); } + r300_emit_draw_init(r300, info->mode, 0, info->count-1); + BEGIN_CS(dwords); - OUT_CS_REG(R300_GA_COLOR_CONTROL, - r300_provoking_vertex_fixes(r300, mode)); OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size); - OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2); - OUT_CS(count - 1); - OUT_CS(0); - OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, count * vertex_size); - OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (count << 16) | - r300_translate_primitive(mode)); + OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, info->count * vertex_size); + OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (info->count << 16) | + r300_translate_primitive(info->mode)); /* Emit vertices. */ - for (v = 0; v < count; v++) { + for (v = 0; v < info->count; v++) { for (i = 0; i < vertex_element_count; i++) { OUT_CS_TABLE(&mapelem[i][stride[i] * v], size[i]); } @@ -423,10 +425,9 @@ 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 (transfer[vbi]) { - vbuf = &r300->vertex_buffer[vbi]; - pipe_buffer_unmap(&r300->context, vbuf->buffer, transfer[vbi]); - transfer[vbi] = NULL; + if (map[vbi]) { + r300->rws->buffer_unmap(r300_resource(r300->vbuf_mgr->real_vertex_buffer[vbi])->buf); + map[vbi] = NULL; } } } @@ -444,15 +445,12 @@ static void r300_emit_draw_arrays(struct r300_context *r300, return; } - BEGIN_CS(7 + (alt_num_verts ? 2 : 0)); + r300_emit_draw_init(r300, mode, 0, count-1); + + BEGIN_CS(2 + (alt_num_verts ? 2 : 0)); if (alt_num_verts) { OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count); } - 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(count - 1); - OUT_CS(0); OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0); OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) | r300_translate_primitive(mode) | @@ -463,37 +461,53 @@ static void r300_emit_draw_arrays(struct r300_context *r300, static void r300_emit_draw_elements(struct r300_context *r300, struct pipe_resource* indexBuffer, unsigned indexSize, - unsigned minIndex, - unsigned maxIndex, + unsigned min_index, + unsigned max_index, 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); - if (count >= (1 << 24)) { + if (count >= (1 << 24) || max_index >= (1 << 24)) { fprintf(stderr, "r300: Got a huge number of vertices: %i, " - "refusing to render.\n", count); + "refusing to render (max_index: %i).\n", count, max_index); 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); + count, min_index, max_index); + + r300_emit_draw_init(r300, mode, min_index, max_index); + + /* 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; - BEGIN_CS(13 + (alt_num_verts ? 2 : 0)); + 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_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); OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0); if (indexSize == 4) { count_dwords = count; @@ -508,68 +522,169 @@ static void r300_emit_draw_elements(struct r300_context *r300, (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0)); } - /* INDX_BUFFER is a truly special packet3. - * Unlike most other packet3, where the offset is after the count, - * the order is reversed, so the relocation ends up carrying the - * size of the indexbuf instead of the offset. - */ OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2); OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) | (0 << R300_INDX_BUFFER_SKIP_SHIFT)); OUT_CS(offset_dwords << 2); - OUT_CS_BUF_RELOC(indexBuffer, count_dwords, - r300_buffer(indexBuffer)->domain, 0); + OUT_CS(count_dwords); + OUT_CS_RELOC(r300_resource(indexBuffer)); + END_CS; +} + +static void r300_draw_elements_immediate(struct r300_context *r300, + const struct pipe_draw_info *info) +{ + uint8_t *ptr1; + uint16_t *ptr2; + uint32_t *ptr4; + unsigned index_size = r300->index_buffer.index_size; + unsigned i, count_dwords = index_size == 4 ? info->count : + (info->count + 1) / 2; + CS_LOCALS(r300); + + /* 19 dwords for r300_draw_elements_immediate. Give up if the function fails. */ + if (!r300_prepare_for_rendering(r300, + PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_AOS | + PREP_INDEXED, NULL, 2+count_dwords, 0, info->index_bias, -1)) + return; + + r300_emit_draw_init(r300, info->mode, info->min_index, info->max_index); + + BEGIN_CS(2 + count_dwords); + OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, count_dwords); + switch (index_size) { + case 1: + ptr1 = r300_resource(r300->index_buffer.buffer)->b.user_ptr; + ptr1 += info->start; + + OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) | + r300_translate_primitive(info->mode)); + + if (info->index_bias && !r300->screen->caps.is_r500) { + for (i = 0; i < info->count-1; i += 2) + OUT_CS(((ptr1[i+1] + info->index_bias) << 16) | + (ptr1[i] + info->index_bias)); + + if (info->count & 1) + OUT_CS(ptr1[i] + info->index_bias); + } else { + for (i = 0; i < info->count-1; i += 2) + OUT_CS(((ptr1[i+1]) << 16) | + (ptr1[i] )); + + if (info->count & 1) + OUT_CS(ptr1[i]); + } + break; + + case 2: + ptr2 = (uint16_t*)r300_resource(r300->index_buffer.buffer)->b.user_ptr; + ptr2 += info->start; + + OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) | + r300_translate_primitive(info->mode)); + + if (info->index_bias && !r300->screen->caps.is_r500) { + for (i = 0; i < info->count-1; i += 2) + OUT_CS(((ptr2[i+1] + info->index_bias) << 16) | + (ptr2[i] + info->index_bias)); + + if (info->count & 1) + OUT_CS(ptr2[i] + info->index_bias); + } else { + OUT_CS_TABLE(ptr2, count_dwords); + } + break; + + case 4: + ptr4 = (uint32_t*)r300_resource(r300->index_buffer.buffer)->b.user_ptr; + ptr4 += info->start; + + OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) | + R300_VAP_VF_CNTL__INDEX_SIZE_32bit | + r300_translate_primitive(info->mode)); + + if (info->index_bias && !r300->screen->caps.is_r500) { + for (i = 0; i < info->count; i++) + OUT_CS(ptr4[i] + info->index_bias); + } else { + OUT_CS_TABLE(ptr4, count_dwords); + } + break; + } END_CS; } -/* This is the fast-path drawing & emission for HW TCL. */ -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) +static void r300_draw_elements(struct r300_context *r300, + const struct pipe_draw_info *info, + int instance_id) { - struct r300_context* r300 = r300_context(pipe); + struct pipe_resource *indexBuffer = r300->index_buffer.buffer; + unsigned indexSize = r300->index_buffer.index_size; struct pipe_resource* orgIndexBuffer = indexBuffer; + unsigned start = info->start; + unsigned count = info->count; boolean alt_num_verts = r300->screen->caps.is_r500 && - count > 65536 && - r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0); + count > 65536; unsigned short_count; int buffer_offset = 0, index_offset = 0; /* for index bias emulation */ - unsigned new_offset; + uint16_t indices3[3]; - if (indexBias && !r300->screen->caps.index_bias_supported) { - r300_split_index_bias(r300, indexBias, &buffer_offset, &index_offset); + if (info->index_bias && !r300->screen->caps.is_r500) { + r300_split_index_bias(r300, info->index_bias, &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, &new_offset); - - start = new_offset; + /* Fallback for misaligned ushort indices. */ + if (indexSize == 2 && (start & 1) && + !r300_resource(indexBuffer)->b.user_ptr) { + /* If we got here, then orgIndexBuffer == indexBuffer. */ + uint16_t *ptr = r300->rws->buffer_map(r300_resource(orgIndexBuffer)->buf, + r300->cs, + PIPE_TRANSFER_READ | + PIPE_TRANSFER_UNSYNCHRONIZED); + + if (info->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 the upload buffer is aligned. */ + r300_upload_index_buffer(r300, &indexBuffer, indexSize, &start, + count, (uint8_t*)ptr); + } + r300->rws->buffer_unmap(r300_resource(orgIndexBuffer)->buf); + } else { + if (r300_resource(indexBuffer)->b.user_ptr) + r300_upload_index_buffer(r300, &indexBuffer, indexSize, + &start, count, + r300_resource(indexBuffer)->b.user_ptr); + } - /* 15 dwords for emit_draw_elements. Give up if the function fails. */ + /* 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, 15, buffer_offset, indexBias)) + PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_AOS | + PREP_INDEXED, indexBuffer, 19, buffer_offset, info->index_bias, + instance_id)) goto done; 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, info->min_index, + info->max_index, info->mode, start, count, + indices3); } else { do { - short_count = MIN2(count, 65534); + if (indexSize == 2 && (start & 1)) + short_count = MIN2(count, 65535); + else + short_count = MIN2(count, 65534); + r300_emit_draw_elements(r300, indexBuffer, indexSize, - minIndex, maxIndex, - mode, start, short_count); + info->min_index, info->max_index, + info->mode, start, short_count, indices3); start += short_count; count -= short_count; @@ -578,7 +693,8 @@ static void r300_draw_range_elements(struct pipe_context* pipe, if (count) { if (!r300_prepare_for_rendering(r300, PREP_VALIDATE_VBOS | PREP_EMIT_AOS | PREP_INDEXED, - indexBuffer, 15, buffer_offset, indexBias)) + indexBuffer, 19, buffer_offset, info->index_bias, + instance_id)) goto done; } } while (count); @@ -590,107 +706,112 @@ done: } } -static void r300_draw_arrays(struct pipe_context* pipe, unsigned mode, - unsigned start, unsigned count) +static void r300_draw_arrays(struct r300_context *r300, + const struct pipe_draw_info *info, + int instance_id) { - struct r300_context* r300 = r300_context(pipe); boolean alt_num_verts = r300->screen->caps.is_r500 && - count > 65536 && - r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0); + info->count > 65536; + unsigned start = info->start; + unsigned count = info->count; unsigned short_count; - r300_update_derived_state(r300); + /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */ + if (!r300_prepare_for_rendering(r300, + PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_AOS, + NULL, 9, start, 0, instance_id)) + return; - if (immd_is_good_idea(r300, count)) { - r300_emit_draw_arrays_immediate(r300, mode, start, count); + if (alt_num_verts || count <= 65535) { + r300_emit_draw_arrays(r300, info->mode, count); } else { - /* 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; + do { + short_count = MIN2(count, 65535); + r300_emit_draw_arrays(r300, info->mode, short_count); - if (alt_num_verts || count <= 65535) { - r300_emit_draw_arrays(r300, mode, count); - } else { - do { - short_count = MIN2(count, 65535); - r300_emit_draw_arrays(r300, mode, short_count); - - start += short_count; - count -= short_count; - - /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */ - if (count) { - if (!r300_prepare_for_rendering(r300, - PREP_VALIDATE_VBOS | PREP_EMIT_AOS, NULL, 9, - start, 0)) - return; - } - } while (count); - } + start += short_count; + count -= short_count; + + /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */ + if (count) { + if (!r300_prepare_for_rendering(r300, + PREP_VALIDATE_VBOS | PREP_EMIT_AOS, NULL, 9, + start, 0, instance_id)) + return; + } + } while (count); } } +static void r300_draw_arrays_instanced(struct r300_context *r300, + const struct pipe_draw_info *info) +{ + int i; + + for (i = 0; i < info->instance_count; i++) + r300_draw_arrays(r300, info, i); +} + +static void r300_draw_elements_instanced(struct r300_context *r300, + const struct pipe_draw_info *info) +{ + int i; + + for (i = 0; i < info->instance_count; i++) + r300_draw_elements(r300, info, i); +} + static void r300_draw_vbo(struct pipe_context* pipe, - const struct pipe_draw_info *info) + const struct pipe_draw_info *dinfo) { struct r300_context* r300 = r300_context(pipe); - unsigned count = info->count; - boolean translate = FALSE; - boolean indexed = info->indexed && r300->index_buffer.buffer; - unsigned start_indexed = 0; + struct pipe_draw_info info = *dinfo; + boolean buffers_updated, uploader_flushed; - if (r300->skip_rendering) { - return; - } + info.indexed = info.indexed && r300->index_buffer.buffer; - if (!u_trim_pipe_prim(info->mode, &count)) { + if (r300->skip_rendering || + !u_trim_pipe_prim(info.mode, &info.count)) { return; } - /* Index buffer range checking. */ - if (indexed) { - assert(r300->index_buffer.offset % r300->index_buffer.index_size == 0); - - /* Compute start for draw_elements, taking the offset into account. */ - start_indexed = - info->start + - (r300->index_buffer.offset / r300->index_buffer.index_size); - - 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; - } - } + r300_update_derived_state(r300); - /* 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; + /* Start the vbuf manager and update buffers if needed. */ + u_vbuf_mgr_draw_begin(r300->vbuf_mgr, &info, + &buffers_updated, &uploader_flushed); + if (buffers_updated) { + r300->vertex_arrays_dirty = TRUE; } - 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); + /* Draw. */ + if (info.indexed) { + info.start += r300->index_buffer.offset; + info.max_index = MIN2(r300->vbuf_mgr->max_index, info.max_index); + + if (info.instance_count <= 1) { + if (info.count <= 8 && + r300_resource(r300->index_buffer.buffer)->b.user_ptr) { + r300_draw_elements_immediate(r300, &info); + } else { + r300_draw_elements(r300, &info, -1); + } + } else { + r300_draw_elements_instanced(r300, &info); + } } else { - r300_draw_arrays(pipe, - info->mode, - info->start, - count); + if (info.instance_count <= 1) { + if (immd_is_good_idea(r300, info.count)) { + r300_draw_arrays_immediate(r300, &info); + } else { + r300_draw_arrays(r300, &info, -1); + } + } else { + r300_draw_arrays_instanced(r300, &info); + } } - if (translate) { - r300_end_vertex_translate(r300); - } + u_vbuf_mgr_draw_end(r300->vbuf_mgr); } /**************************************************************************** @@ -721,15 +842,16 @@ static void r300_swtcl_draw_vbo(struct pipe_context* pipe, r300_update_derived_state(r300); r300_reserve_cs_dwords(r300, - PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL | + PREP_EMIT_STATES | PREP_EMIT_AOS_SWTCL | (indexed ? PREP_INDEXED : 0), indexed ? 256 : 6); - for (i = 0; i < r300->vertex_buffer_count; i++) { - if (r300->vertex_buffer[i].buffer) { + for (i = 0; i < r300->vbuf_mgr->nr_vertex_buffers; i++) { + if (r300->vbuf_mgr->vertex_buffer[i].buffer) { void *buf = pipe_buffer_map(pipe, - r300->vertex_buffer[i].buffer, - PIPE_TRANSFER_READ, + r300->vbuf_mgr->vertex_buffer[i].buffer, + PIPE_TRANSFER_READ | + PIPE_TRANSFER_UNSYNCHRONIZED, &vb_transfer[i]); draw_set_mapped_vertex_buffer(r300->draw, i, buf); } @@ -737,7 +859,8 @@ static void r300_swtcl_draw_vbo(struct pipe_context* pipe, if (indexed) { indices = pipe_buffer_map(pipe, r300->index_buffer.buffer, - PIPE_TRANSFER_READ, &ib_transfer); + PIPE_TRANSFER_READ | + PIPE_TRANSFER_UNSYNCHRONIZED, &ib_transfer); } draw_set_mapped_index_buffer(r300->draw, indices); @@ -748,16 +871,15 @@ static void r300_swtcl_draw_vbo(struct pipe_context* pipe, draw_flush(r300->draw); r300->draw_vbo_locked = FALSE; - for (i = 0; i < r300->vertex_buffer_count; i++) { - if (r300->vertex_buffer[i].buffer) { - pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer, - vb_transfer[i]); + for (i = 0; i < r300->vbuf_mgr->nr_vertex_buffers; i++) { + if (r300->vbuf_mgr->vertex_buffer[i].buffer) { + pipe_buffer_unmap(pipe, vb_transfer[i]); draw_set_mapped_vertex_buffer(r300->draw, i, NULL); } } if (indexed) { - pipe_buffer_unmap(pipe, r300->index_buffer.buffer, ib_transfer); + pipe_buffer_unmap(pipe, ib_transfer); draw_set_mapped_index_buffer(r300->draw, NULL); } } @@ -813,10 +935,10 @@ static boolean r300_render_allocate_vertices(struct vbuf_render* render, pipe_resource_reference(&r300->vbo, NULL); r300->vbo = pipe_buffer_create(screen, PIPE_BIND_VERTEX_BUFFER, + PIPE_USAGE_STREAM, 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; @@ -835,7 +957,8 @@ static void* r300_render_map_vertices(struct vbuf_render* render) r300render->vbo_ptr = pipe_buffer_map(&r300render->r300->context, r300->vbo, - PIPE_TRANSFER_WRITE, + PIPE_TRANSFER_WRITE | + PIPE_TRANSFER_UNSYNCHRONIZED, &r300render->vbo_transfer); assert(r300render->vbo_ptr); @@ -857,7 +980,7 @@ static void r300_render_unmap_vertices(struct vbuf_render* render, r300render->vbo_max_used = MAX2(r300render->vbo_max_used, r300render->vertex_size * (max + 1)); - pipe_buffer_unmap(context, r300->vbo, r300render->vbo_transfer); + pipe_buffer_unmap(context, r300render->vbo_transfer); r300render->vbo_transfer = NULL; } @@ -901,33 +1024,16 @@ static void r300_render_draw_arrays(struct vbuf_render* render, if (r300->draw_first_emitted) { if (!r300_prepare_for_rendering(r300, - PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL, - NULL, 6, 0, 0)) + PREP_EMIT_STATES | PREP_EMIT_AOS_SWTCL, + NULL, dwords, 0, 0, -1)) return; } else { if (!r300_emit_states(r300, - PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL, - NULL, 0, 0)) + PREP_EMIT_STATES | PREP_EMIT_AOS_SWTCL, + NULL, 0, 0, -1)) return; } - /* 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)); @@ -958,13 +1064,13 @@ static void r300_render_draw_elements(struct vbuf_render* render, if (r300->draw_first_emitted) { if (!r300_prepare_for_rendering(r300, - PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL | PREP_INDEXED, - NULL, 256, 0, 0)) + PREP_EMIT_STATES | PREP_EMIT_AOS_SWTCL | PREP_INDEXED, + NULL, 256, 0, 0, -1)) return; } else { if (!r300_emit_states(r300, - PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL | PREP_INDEXED, - NULL, 0, 0)) + PREP_EMIT_STATES | PREP_EMIT_AOS_SWTCL | PREP_INDEXED, + NULL, 0, 0, -1)) return; } @@ -1001,7 +1107,7 @@ static void r300_render_draw_elements(struct vbuf_render* render, if (count) { if (!r300_prepare_for_rendering(r300, PREP_EMIT_AOS_SWTCL | PREP_INDEXED, - NULL, 256, 0, 0)) + NULL, 256, 0, 0, -1)) return; end_cs_dwords = r300_get_num_cs_end_dwords(r300); @@ -1022,8 +1128,7 @@ static struct vbuf_render* r300_render_create(struct r300_context* r300) 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; @@ -1106,7 +1211,7 @@ static void r300_blitter_draw_rectangle(struct blitter_context *blitter, r300->clip_state.dirty = FALSE; r300->viewport_state.dirty = FALSE; - if (!r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0)) + if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0, -1)) goto done; DBG(r300, DBG_DRAW, "r300: draw_rectangle\n"); @@ -1190,7 +1295,7 @@ static void r300_resource_resolve(struct pipe_context* pipe, 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.size = 10; r300_mark_atom_dirty(r300, &r300->aa_state); /* Resolve the surface. */ |