/*
- * Copyright 2008 Ben Skeggs
+ * Copyright 2010 Christoph Bumiller
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
#include "pipe/p_context.h"
#include "pipe/p_state.h"
-#include "pipe/p_inlines.h"
-
+#include "util/u_inlines.h"
#include "util/u_format.h"
+#include "translate/translate.h"
#include "nv50_context.h"
+#include "nv50_resource.h"
-static boolean
-nv50_push_elements_u08(struct nv50_context *, uint8_t *, unsigned);
-
-static boolean
-nv50_push_elements_u16(struct nv50_context *, uint16_t *, unsigned);
-
-static boolean
-nv50_push_elements_u32(struct nv50_context *, uint32_t *, unsigned);
-
-static boolean
-nv50_push_arrays(struct nv50_context *, unsigned, unsigned);
-
-#define NV50_USING_LOATHED_EDGEFLAG(ctx) ((ctx)->vertprog->cfg.edgeflag_in < 16)
-
-static INLINE unsigned
-nv50_prim(unsigned mode)
-{
- switch (mode) {
- case PIPE_PRIM_POINTS: return NV50TCL_VERTEX_BEGIN_POINTS;
- case PIPE_PRIM_LINES: return NV50TCL_VERTEX_BEGIN_LINES;
- case PIPE_PRIM_LINE_LOOP: return NV50TCL_VERTEX_BEGIN_LINE_LOOP;
- case PIPE_PRIM_LINE_STRIP: return NV50TCL_VERTEX_BEGIN_LINE_STRIP;
- case PIPE_PRIM_TRIANGLES: return NV50TCL_VERTEX_BEGIN_TRIANGLES;
- case PIPE_PRIM_TRIANGLE_STRIP:
- return NV50TCL_VERTEX_BEGIN_TRIANGLE_STRIP;
- case PIPE_PRIM_TRIANGLE_FAN: return NV50TCL_VERTEX_BEGIN_TRIANGLE_FAN;
- case PIPE_PRIM_QUADS: return NV50TCL_VERTEX_BEGIN_QUADS;
- case PIPE_PRIM_QUAD_STRIP: return NV50TCL_VERTEX_BEGIN_QUAD_STRIP;
- case PIPE_PRIM_POLYGON: return NV50TCL_VERTEX_BEGIN_POLYGON;
- case PIPE_PRIM_LINES_ADJACENCY:
- return NV50TCL_VERTEX_BEGIN_LINES_ADJACENCY;
- case PIPE_PRIM_LINE_STRIP_ADJACENCY:
- return NV50TCL_VERTEX_BEGIN_LINE_STRIP_ADJACENCY;
- case PIPE_PRIM_TRIANGLES_ADJACENCY:
- return NV50TCL_VERTEX_BEGIN_TRIANGLES_ADJACENCY;
- case PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY:
- return NV50TCL_VERTEX_BEGIN_TRIANGLE_STRIP_ADJACENCY;
- default:
- break;
- }
-
- NOUVEAU_ERR("invalid primitive type %d\n", mode);
- return NV50TCL_VERTEX_BEGIN_POINTS;
-}
-
-static INLINE uint32_t
-nv50_vbo_type_to_hw(enum pipe_format format)
-{
- const struct util_format_description *desc;
-
- desc = util_format_description(format);
- assert(desc);
-
- switch (desc->channel[0].type) {
- case UTIL_FORMAT_TYPE_FLOAT:
- return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_FLOAT;
- case UTIL_FORMAT_TYPE_UNSIGNED:
- if (desc->channel[0].normalized) {
- return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_UNORM;
- }
- return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_USCALED;
- case UTIL_FORMAT_TYPE_SIGNED:
- if (desc->channel[0].normalized) {
- return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_SNORM;
- }
- return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_SSCALED;
- /*
- case PIPE_FORMAT_TYPE_UINT:
- return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_UINT;
- case PIPE_FORMAT_TYPE_SINT:
- return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_SINT; */
- default:
- return 0;
- }
-}
-
-static INLINE uint32_t
-nv50_vbo_size_to_hw(unsigned size, unsigned nr_c)
-{
- static const uint32_t hw_values[] = {
- 0, 0, 0, 0,
- NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_8,
- NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_8_8,
- NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_8_8_8,
- NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_8_8_8_8,
- NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_16,
- NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_16_16,
- NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_16_16_16,
- NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_16_16_16_16,
- 0, 0, 0, 0,
- NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_32,
- NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_32_32,
- NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_32_32_32,
- NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_32_32_32_32 };
-
- /* we'd also have R11G11B10 and R10G10B10A2 */
-
- assert(nr_c > 0 && nr_c <= 4);
-
- if (size > 32)
- return 0;
- size >>= (3 - 2);
-
- return hw_values[size + (nr_c - 1)];
-}
-
-static INLINE uint32_t
-nv50_vbo_vtxelt_to_hw(struct pipe_vertex_element *ve)
-{
- uint32_t hw_type, hw_size;
- enum pipe_format pf = ve->src_format;
- const struct util_format_description *desc;
- unsigned size;
-
- desc = util_format_description(pf);
- assert(desc);
-
- size = util_format_get_component_bits(pf, UTIL_FORMAT_COLORSPACE_RGB, 0);
-
- hw_type = nv50_vbo_type_to_hw(pf);
- hw_size = nv50_vbo_size_to_hw(size, ve->nr_components);
-
- if (!hw_type || !hw_size) {
- NOUVEAU_ERR("unsupported vbo format: %s\n", pf_name(pf));
- abort();
- return 0x24e80000;
- }
-
- if (desc->swizzle[0] == UTIL_FORMAT_SWIZZLE_Z) /* BGRA */
- hw_size |= (1 << 31); /* no real swizzle bits :-( */
-
- return (hw_type | hw_size);
-}
+#include "nv50_3d.xml.h"
-/* For instanced drawing from user buffers, hitting the FIFO repeatedly
- * with the same vertex data is probably worse than uploading all data.
- */
-static boolean
-nv50_upload_vtxbuf(struct nv50_context *nv50, unsigned i)
-{
- struct nv50_screen *nscreen = nv50->screen;
- struct pipe_screen *pscreen = &nscreen->base.base;
- struct pipe_buffer *buf = nscreen->strm_vbuf[i];
- struct pipe_vertex_buffer *vb = &nv50->vtxbuf[i];
- uint8_t *src;
- unsigned size = align(vb->buffer->size, 4096);
-
- if (buf && buf->size < size)
- pipe_buffer_reference(&nscreen->strm_vbuf[i], NULL);
-
- if (!nscreen->strm_vbuf[i]) {
- nscreen->strm_vbuf[i] = pipe_buffer_create(
- pscreen, 0, PIPE_BUFFER_USAGE_VERTEX, size);
- buf = nscreen->strm_vbuf[i];
- }
-
- src = pipe_buffer_map(pscreen, vb->buffer, PIPE_BUFFER_USAGE_CPU_READ);
- if (!src)
- return FALSE;
- src += vb->buffer_offset;
-
- size = (vb->max_index + 1) * vb->stride + 16; /* + 16 is for stride 0 */
- if (vb->buffer_offset + size > vb->buffer->size)
- size = vb->buffer->size - vb->buffer_offset;
-
- pipe_buffer_write(pscreen, buf, vb->buffer_offset, size, src);
- pipe_buffer_unmap(pscreen, vb->buffer);
-
- vb->buffer = buf; /* don't pipe_reference, this is a private copy */
- return TRUE;
-}
-
-static void
-nv50_upload_user_vbufs(struct nv50_context *nv50)
-{
- unsigned i;
-
- if (nv50->vbo_fifo)
- nv50->dirty |= NV50_NEW_ARRAYS;
- if (!(nv50->dirty & NV50_NEW_ARRAYS))
- return;
-
- for (i = 0; i < nv50->vtxbuf_nr; ++i) {
- if (nv50->vtxbuf[i].buffer->usage & PIPE_BUFFER_USAGE_VERTEX)
- continue;
- nv50_upload_vtxbuf(nv50, i);
- }
-}
+void
+nv50_vertex_state_delete(struct pipe_context *pipe,
+ void *hwcso)
+{
+ struct nv50_vertex_stateobj *so = hwcso;
+
+ if (so->translate)
+ so->translate->release(so->translate);
+ FREE(hwcso);
+}
+
+void *
+nv50_vertex_state_create(struct pipe_context *pipe,
+ unsigned num_elements,
+ const struct pipe_vertex_element *elements)
+{
+ struct nv50_vertex_stateobj *so;
+ struct translate_key transkey;
+ unsigned i;
+
+ so = MALLOC(sizeof(*so) +
+ num_elements * sizeof(struct nv50_vertex_element));
+ if (!so)
+ return NULL;
+ so->num_elements = num_elements;
+ so->instance_elts = 0;
+ so->instance_bufs = 0;
+ so->need_conversion = FALSE;
+
+ transkey.nr_elements = 0;
+ transkey.output_stride = 0;
+
+ for (i = 0; i < num_elements; ++i) {
+ const struct pipe_vertex_element *ve = &elements[i];
+ const unsigned vbi = ve->vertex_buffer_index;
+ enum pipe_format fmt = ve->src_format;
+
+ so->element[i].pipe = elements[i];
+ so->element[i].state = nv50_format_table[fmt].vtx;
+
+ if (!so->element[i].state) {
+ switch (util_format_get_nr_components(fmt)) {
+ case 1: fmt = PIPE_FORMAT_R32_FLOAT; break;
+ case 2: fmt = PIPE_FORMAT_R32G32_FLOAT; break;
+ case 3: fmt = PIPE_FORMAT_R32G32B32_FLOAT; break;
+ case 4: fmt = PIPE_FORMAT_R32G32B32A32_FLOAT; break;
+ default:
+ assert(0);
+ return NULL;
+ }
+ so->element[i].state = nv50_format_table[fmt].vtx;
+ so->need_conversion = TRUE;
+ }
+ so->element[i].state |= i;
+
+ if (1) {
+ unsigned j = transkey.nr_elements++;
+
+ transkey.element[j].type = TRANSLATE_ELEMENT_NORMAL;
+ transkey.element[j].input_format = ve->src_format;
+ transkey.element[j].input_buffer = vbi;
+ transkey.element[j].input_offset = ve->src_offset;
+ transkey.element[j].instance_divisor = ve->instance_divisor;
+
+ transkey.element[j].output_format = fmt;
+ transkey.element[j].output_offset = transkey.output_stride;
+ transkey.output_stride += (util_format_get_stride(fmt, 1) + 3) & ~3;
+
+ if (unlikely(ve->instance_divisor)) {
+ so->instance_elts |= 1 << i;
+ so->instance_bufs |= 1 << vbi;
+ }
+ }
+ }
+
+ so->translate = translate_create(&transkey);
+ so->vertex_size = transkey.output_stride / 4;
+ so->packet_vertex_limit = NV04_PFIFO_MAX_PACKET_LEN /
+ MAX2(so->vertex_size, 1);
+
+ return so;
+}
+
+#define NV50_3D_VERTEX_ATTRIB_INACTIVE \
+ NV50_3D_VERTEX_ARRAY_ATTRIB_TYPE_FLOAT | \
+ NV50_3D_VERTEX_ARRAY_ATTRIB_FORMAT_32_32_32_32 | \
+ NV50_3D_VERTEX_ARRAY_ATTRIB_CONST
static void
-nv50_set_static_vtxattr(struct nv50_context *nv50, unsigned i, void *data)
-{
- struct nouveau_grobj *tesla = nv50->screen->tesla;
- struct nouveau_channel *chan = tesla->channel;
- float v[4];
-
- util_format_read_4f(nv50->vtxelt[i].src_format,
- v, 0, data, 0, 0, 0, 1, 1);
-
- switch (nv50->vtxelt[i].nr_components) {
- case 4:
- BEGIN_RING(chan, tesla, NV50TCL_VTX_ATTR_4F_X(i), 4);
- OUT_RINGf (chan, v[0]);
- OUT_RINGf (chan, v[1]);
- OUT_RINGf (chan, v[2]);
- OUT_RINGf (chan, v[3]);
- break;
- case 3:
- BEGIN_RING(chan, tesla, NV50TCL_VTX_ATTR_3F_X(i), 3);
- OUT_RINGf (chan, v[0]);
- OUT_RINGf (chan, v[1]);
- OUT_RINGf (chan, v[2]);
- break;
- case 2:
- BEGIN_RING(chan, tesla, NV50TCL_VTX_ATTR_2F_X(i), 2);
- OUT_RINGf (chan, v[0]);
- OUT_RINGf (chan, v[1]);
- break;
- case 1:
- BEGIN_RING(chan, tesla, NV50TCL_VTX_ATTR_1F(i), 1);
- OUT_RINGf (chan, v[0]);
- break;
- default:
- assert(0);
- break;
- }
+nv50_emit_vtxattr(struct nv50_context *nv50, struct pipe_vertex_buffer *vb,
+ struct pipe_vertex_element *ve, unsigned attr)
+{
+ const void *data;
+ struct nouveau_channel *chan = nv50->screen->base.channel;
+ struct nv04_resource *res = nv04_resource(vb->buffer);
+ float v[4];
+ const unsigned nc = util_format_get_nr_components(ve->src_format);
+
+ data = nouveau_resource_map_offset(&nv50->base, res, vb->buffer_offset +
+ ve->src_offset, NOUVEAU_BO_RD);
+
+ util_format_read_4f(ve->src_format, v, 0, data, 0, 0, 0, 1, 1);
+
+ switch (nc) {
+ case 4:
+ BEGIN_RING(chan, RING_3D(VTX_ATTR_4F_X(attr)), 4);
+ OUT_RINGf (chan, v[0]);
+ OUT_RINGf (chan, v[1]);
+ OUT_RINGf (chan, v[2]);
+ OUT_RINGf (chan, v[3]);
+ break;
+ case 3:
+ BEGIN_RING(chan, RING_3D(VTX_ATTR_3F_X(attr)), 3);
+ OUT_RINGf (chan, v[0]);
+ OUT_RINGf (chan, v[1]);
+ OUT_RINGf (chan, v[2]);
+ break;
+ case 2:
+ BEGIN_RING(chan, RING_3D(VTX_ATTR_2F_X(attr)), 2);
+ OUT_RINGf (chan, v[0]);
+ OUT_RINGf (chan, v[1]);
+ break;
+ case 1:
+ if (attr == nv50->vertprog->vp.edgeflag) {
+ BEGIN_RING(chan, RING_3D(EDGEFLAG_ENABLE), 1);
+ OUT_RING (chan, v[0] ? 1 : 0);
+ }
+ BEGIN_RING(chan, RING_3D(VTX_ATTR_1F(attr)), 1);
+ OUT_RINGf (chan, v[0]);
+ break;
+ default:
+ assert(0);
+ break;
+ }
}
-static unsigned
-init_per_instance_arrays_immd(struct nv50_context *nv50,
- unsigned startInstance,
- unsigned pos[16], unsigned step[16])
-{
- struct nouveau_bo *bo;
- unsigned i, b, count = 0;
-
- for (i = 0; i < nv50->vtxelt_nr; ++i) {
- if (!nv50->vtxelt[i].instance_divisor)
- continue;
- ++count;
- b = nv50->vtxelt[i].vertex_buffer_index;
-
- pos[i] = nv50->vtxelt[i].src_offset +
- nv50->vtxbuf[b].buffer_offset +
- startInstance * nv50->vtxbuf[b].stride;
- step[i] = startInstance % nv50->vtxelt[i].instance_divisor;
-
- bo = nouveau_bo(nv50->vtxbuf[b].buffer);
- if (!bo->map)
- nouveau_bo_map(bo, NOUVEAU_BO_RD);
-
- nv50_set_static_vtxattr(nv50, i, (uint8_t *)bo->map + pos[i]);
- }
-
- return count;
-}
-
-static unsigned
-init_per_instance_arrays(struct nv50_context *nv50,
- unsigned startInstance,
- unsigned pos[16], unsigned step[16])
+static INLINE void
+nv50_vbuf_range(struct nv50_context *nv50, int vbi,
+ uint32_t *base, uint32_t *size)
{
- struct nouveau_grobj *tesla = nv50->screen->tesla;
- struct nouveau_channel *chan = tesla->channel;
- struct nouveau_bo *bo;
- struct nouveau_stateobj *so;
- unsigned i, b, count = 0;
- const uint32_t rl = NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD;
-
- if (nv50->vbo_fifo)
- return init_per_instance_arrays_immd(nv50, startInstance,
- pos, step);
-
- so = so_new(nv50->vtxelt_nr, nv50->vtxelt_nr * 2, nv50->vtxelt_nr * 2);
-
- for (i = 0; i < nv50->vtxelt_nr; ++i) {
- if (!nv50->vtxelt[i].instance_divisor)
- continue;
- ++count;
- b = nv50->vtxelt[i].vertex_buffer_index;
-
- pos[i] = nv50->vtxelt[i].src_offset +
- nv50->vtxbuf[b].buffer_offset +
- startInstance * nv50->vtxbuf[b].stride;
-
- if (!startInstance) {
- step[i] = 0;
- continue;
- }
- step[i] = startInstance % nv50->vtxelt[i].instance_divisor;
-
- bo = nouveau_bo(nv50->vtxbuf[b].buffer);
-
- so_method(so, tesla, NV50TCL_VERTEX_ARRAY_START_HIGH(i), 2);
- so_reloc (so, bo, pos[i], rl | NOUVEAU_BO_HIGH, 0, 0);
- so_reloc (so, bo, pos[i], rl | NOUVEAU_BO_LOW, 0, 0);
- }
-
- if (count && startInstance) {
- so_ref (so, &nv50->state.instbuf); /* for flush notify */
- so_emit(chan, nv50->state.instbuf);
- }
- so_ref (NULL, &so);
-
- return count;
+ if (unlikely(nv50->vertex->instance_bufs & (1 << vbi))) {
+ /* TODO: use min and max instance divisor to get a proper range */
+ *base = 0;
+ *size = nv50->vtxbuf[vbi].buffer->width0;
+ } else {
+ assert(nv50->vbo_max_index != ~0);
+ *base = nv50->vbo_min_index * nv50->vtxbuf[vbi].stride;
+ *size = (nv50->vbo_max_index -
+ nv50->vbo_min_index + 1) * nv50->vtxbuf[vbi].stride;
+ }
}
static void
-step_per_instance_arrays_immd(struct nv50_context *nv50,
- unsigned pos[16], unsigned step[16])
-{
- struct nouveau_bo *bo;
- unsigned i, b;
-
- for (i = 0; i < nv50->vtxelt_nr; ++i) {
- if (!nv50->vtxelt[i].instance_divisor)
- continue;
- if (++step[i] != nv50->vtxelt[i].instance_divisor)
- continue;
- b = nv50->vtxelt[i].vertex_buffer_index;
- bo = nouveau_bo(nv50->vtxbuf[b].buffer);
-
- step[i] = 0;
- pos[i] += nv50->vtxbuf[b].stride;
-
- nv50_set_static_vtxattr(nv50, i, (uint8_t *)bo->map + pos[i]);
- }
+nv50_prevalidate_vbufs(struct nv50_context *nv50)
+{
+ struct pipe_vertex_buffer *vb;
+ struct nv04_resource *buf;
+ int i;
+ uint32_t base, size;
+
+ nv50->vbo_fifo = nv50->vbo_user = 0;
+
+ nv50_bufctx_reset(nv50, NV50_BUFCTX_VERTEX);
+
+ for (i = 0; i < nv50->num_vtxbufs; ++i) {
+ vb = &nv50->vtxbuf[i];
+ if (!vb->stride)
+ continue;
+ buf = nv04_resource(vb->buffer);
+
+ /* NOTE: user buffers with temporary storage count as mapped by GPU */
+ if (!nouveau_resource_mapped_by_gpu(vb->buffer)) {
+ if (nv50->vbo_push_hint) {
+ nv50->vbo_fifo = ~0;
+ continue;
+ } else {
+ if (buf->status & NOUVEAU_BUFFER_STATUS_USER_MEMORY) {
+ nv50->vbo_user |= 1 << i;
+ assert(vb->stride > vb->buffer_offset);
+ nv50_vbuf_range(nv50, i, &base, &size);
+ nouveau_user_buffer_upload(buf, base, size);
+ } else {
+ nouveau_buffer_migrate(&nv50->base, buf, NOUVEAU_BO_GART);
+ }
+ nv50->base.vbo_dirty = TRUE;
+ }
+ }
+ nv50_bufctx_add_resident(nv50, NV50_BUFCTX_VERTEX, buf, NOUVEAU_BO_RD);
+ nouveau_buffer_adjust_score(&nv50->base, buf, 1);
+ }
}
static void
-step_per_instance_arrays(struct nv50_context *nv50,
- unsigned pos[16], unsigned step[16])
-{
- struct nouveau_grobj *tesla = nv50->screen->tesla;
- struct nouveau_channel *chan = tesla->channel;
- struct nouveau_bo *bo;
- struct nouveau_stateobj *so;
- unsigned i, b;
- const uint32_t rl = NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD;
-
- if (nv50->vbo_fifo) {
- step_per_instance_arrays_immd(nv50, pos, step);
- return;
- }
-
- so = so_new(nv50->vtxelt_nr, nv50->vtxelt_nr * 2, nv50->vtxelt_nr * 2);
-
- for (i = 0; i < nv50->vtxelt_nr; ++i) {
- if (!nv50->vtxelt[i].instance_divisor)
- continue;
- b = nv50->vtxelt[i].vertex_buffer_index;
-
- if (++step[i] == nv50->vtxelt[i].instance_divisor) {
- step[i] = 0;
- pos[i] += nv50->vtxbuf[b].stride;
- }
-
- bo = nouveau_bo(nv50->vtxbuf[b].buffer);
-
- so_method(so, tesla, NV50TCL_VERTEX_ARRAY_START_HIGH(i), 2);
- so_reloc (so, bo, pos[i], rl | NOUVEAU_BO_HIGH, 0, 0);
- so_reloc (so, bo, pos[i], rl | NOUVEAU_BO_LOW, 0, 0);
- }
-
- so_ref (so, &nv50->state.instbuf); /* for flush notify */
- so_ref (NULL, &so);
-
- so_emit(chan, nv50->state.instbuf);
+nv50_update_user_vbufs(struct nv50_context *nv50)
+{
+ struct nouveau_channel *chan = nv50->screen->base.channel;
+ uint32_t base, offset, size;
+ int i;
+ uint32_t written = 0;
+
+ for (i = 0; i < nv50->vertex->num_elements; ++i) {
+ struct pipe_vertex_element *ve = &nv50->vertex->element[i].pipe;
+ const int b = ve->vertex_buffer_index;
+ struct pipe_vertex_buffer *vb = &nv50->vtxbuf[b];
+ struct nv04_resource *buf = nv04_resource(vb->buffer);
+
+ if (!(nv50->vbo_user & (1 << b)))
+ continue;
+
+ if (!vb->stride) {
+ nv50_emit_vtxattr(nv50, vb, ve, i);
+ continue;
+ }
+ nv50_vbuf_range(nv50, b, &base, &size);
+
+ if (!(written & (1 << b))) {
+ written |= 1 << b;
+ nouveau_user_buffer_upload(buf, base, size);
+ }
+ offset = vb->buffer_offset + ve->src_offset;
+
+ MARK_RING (chan, 6, 4);
+ BEGIN_RING(chan, RING_3D(VERTEX_ARRAY_LIMIT_HIGH(i)), 2);
+ OUT_RESRCh(chan, buf, base + size - 1, NOUVEAU_BO_RD);
+ OUT_RESRCl(chan, buf, base + size - 1, NOUVEAU_BO_RD);
+ BEGIN_RING(chan, RING_3D(VERTEX_ARRAY_START_HIGH(i)), 2);
+ OUT_RESRCh(chan, buf, offset, NOUVEAU_BO_RD);
+ OUT_RESRCl(chan, buf, offset, NOUVEAU_BO_RD);
+ }
+ nv50->base.vbo_dirty = TRUE;
}
static INLINE void
-nv50_unmap_vbufs(struct nv50_context *nv50)
+nv50_release_user_vbufs(struct nv50_context *nv50)
{
- unsigned i;
+ uint32_t vbo_user = nv50->vbo_user;
- for (i = 0; i < nv50->vtxbuf_nr; ++i)
- if (nouveau_bo(nv50->vtxbuf[i].buffer)->map)
- nouveau_bo_unmap(nouveau_bo(nv50->vtxbuf[i].buffer));
-}
+ while (vbo_user) {
+ int i = ffs(vbo_user) - 1;
+ vbo_user &= ~(1 << i);
-void
-nv50_draw_arrays_instanced(struct pipe_context *pipe,
- unsigned mode, unsigned start, unsigned count,
- unsigned startInstance, unsigned instanceCount)
-{
- struct nv50_context *nv50 = nv50_context(pipe);
- struct nouveau_channel *chan = nv50->screen->tesla->channel;
- struct nouveau_grobj *tesla = nv50->screen->tesla;
- unsigned i, nz_divisors;
- unsigned step[16], pos[16];
-
- if (!NV50_USING_LOATHED_EDGEFLAG(nv50))
- nv50_upload_user_vbufs(nv50);
-
- nv50_state_validate(nv50);
-
- nz_divisors = init_per_instance_arrays(nv50, startInstance, pos, step);
-
- BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 2);
- OUT_RING (chan, NV50_CB_AUX | (24 << 8));
- OUT_RING (chan, startInstance);
-
- BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
- OUT_RING (chan, nv50_prim(mode));
-
- if (nv50->vbo_fifo)
- nv50_push_arrays(nv50, start, count);
- else {
- BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BUFFER_FIRST, 2);
- OUT_RING (chan, start);
- OUT_RING (chan, count);
- }
- BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
- OUT_RING (chan, 0);
-
- for (i = 1; i < instanceCount; i++) {
- if (nz_divisors) /* any non-zero array divisors ? */
- step_per_instance_arrays(nv50, pos, step);
-
- BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
- OUT_RING (chan, nv50_prim(mode) | (1 << 28));
-
- if (nv50->vbo_fifo)
- nv50_push_arrays(nv50, start, count);
- else {
- BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BUFFER_FIRST, 2);
- OUT_RING (chan, start);
- OUT_RING (chan, count);
- }
- BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
- OUT_RING (chan, 0);
- }
- nv50_unmap_vbufs(nv50);
-
- so_ref(NULL, &nv50->state.instbuf);
+ nouveau_buffer_release_gpu_storage(nv04_resource(nv50->vtxbuf[i].buffer));
+ }
}
void
-nv50_draw_arrays(struct pipe_context *pipe, unsigned mode, unsigned start,
- unsigned count)
-{
- struct nv50_context *nv50 = nv50_context(pipe);
- struct nouveau_channel *chan = nv50->screen->tesla->channel;
- struct nouveau_grobj *tesla = nv50->screen->tesla;
- boolean ret;
-
- nv50_state_validate(nv50);
-
- BEGIN_RING(chan, tesla, 0x142c, 1);
- OUT_RING (chan, 0);
- BEGIN_RING(chan, tesla, 0x142c, 1);
- OUT_RING (chan, 0);
-
- BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
- OUT_RING (chan, nv50_prim(mode));
-
- if (nv50->vbo_fifo)
- ret = nv50_push_arrays(nv50, start, count);
- else {
- BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BUFFER_FIRST, 2);
- OUT_RING (chan, start);
- OUT_RING (chan, count);
- ret = TRUE;
- }
- BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
- OUT_RING (chan, 0);
-
- nv50_unmap_vbufs(nv50);
-
- /* XXX: not sure what to do if ret != TRUE: flush and retry?
- */
- assert(ret);
-}
-
-static INLINE boolean
-nv50_draw_elements_inline_u08(struct nv50_context *nv50, uint8_t *map,
- unsigned start, unsigned count)
-{
- struct nouveau_channel *chan = nv50->screen->tesla->channel;
- struct nouveau_grobj *tesla = nv50->screen->tesla;
-
- map += start;
-
- if (nv50->vbo_fifo)
- return nv50_push_elements_u08(nv50, map, count);
-
- if (count & 1) {
- BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32, 1);
- OUT_RING (chan, map[0]);
- map++;
- count--;
- }
-
- while (count) {
- unsigned nr = count > 2046 ? 2046 : count;
- int i;
-
- BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U16, nr >> 1);
- for (i = 0; i < nr; i += 2)
- OUT_RING (chan, (map[i + 1] << 16) | map[i]);
-
- count -= nr;
- map += nr;
- }
- return TRUE;
-}
-
-static INLINE boolean
-nv50_draw_elements_inline_u16(struct nv50_context *nv50, uint16_t *map,
- unsigned start, unsigned count)
-{
- struct nouveau_channel *chan = nv50->screen->tesla->channel;
- struct nouveau_grobj *tesla = nv50->screen->tesla;
-
- map += start;
-
- if (nv50->vbo_fifo)
- return nv50_push_elements_u16(nv50, map, count);
-
- if (count & 1) {
- BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32, 1);
- OUT_RING (chan, map[0]);
- map++;
- count--;
- }
-
- while (count) {
- unsigned nr = count > 2046 ? 2046 : count;
- int i;
-
- BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U16, nr >> 1);
- for (i = 0; i < nr; i += 2)
- OUT_RING (chan, (map[i + 1] << 16) | map[i]);
-
- count -= nr;
- map += nr;
- }
- return TRUE;
-}
-
-static INLINE boolean
-nv50_draw_elements_inline_u32(struct nv50_context *nv50, uint32_t *map,
- unsigned start, unsigned count)
-{
- struct nouveau_channel *chan = nv50->screen->tesla->channel;
- struct nouveau_grobj *tesla = nv50->screen->tesla;
-
- map += start;
-
- if (nv50->vbo_fifo)
- return nv50_push_elements_u32(nv50, map, count);
-
- while (count) {
- unsigned nr = count > 2047 ? 2047 : count;
-
- BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U32, nr);
- OUT_RINGp (chan, map, nr);
-
- count -= nr;
- map += nr;
- }
- return TRUE;
-}
-
-static INLINE void
-nv50_draw_elements_inline(struct nv50_context *nv50,
- void *map, unsigned indexSize,
- unsigned start, unsigned count)
-{
- switch (indexSize) {
- case 1:
- nv50_draw_elements_inline_u08(nv50, map, start, count);
- break;
- case 2:
- nv50_draw_elements_inline_u16(nv50, map, start, count);
- break;
- case 4:
- nv50_draw_elements_inline_u32(nv50, map, start, count);
- break;
- }
-}
-
-void
-nv50_draw_elements_instanced(struct pipe_context *pipe,
- struct pipe_buffer *indexBuffer,
- unsigned indexSize,
- unsigned mode, unsigned start, unsigned count,
- unsigned startInstance, unsigned instanceCount)
-{
- struct nv50_context *nv50 = nv50_context(pipe);
- struct nouveau_grobj *tesla = nv50->screen->tesla;
- struct nouveau_channel *chan = tesla->channel;
- struct pipe_screen *pscreen = pipe->screen;
- void *map;
- unsigned i, nz_divisors;
- unsigned step[16], pos[16];
-
- map = pipe_buffer_map(pscreen, indexBuffer, PIPE_BUFFER_USAGE_CPU_READ);
-
- if (!NV50_USING_LOATHED_EDGEFLAG(nv50))
- nv50_upload_user_vbufs(nv50);
-
- nv50_state_validate(nv50);
-
- nz_divisors = init_per_instance_arrays(nv50, startInstance, pos, step);
+nv50_vertex_arrays_validate(struct nv50_context *nv50)
+{
+ struct nouveau_channel *chan = nv50->screen->base.channel;
+ struct nv50_vertex_stateobj *vertex = nv50->vertex;
+ struct pipe_vertex_buffer *vb;
+ struct nv50_vertex_element *ve;
+ unsigned i;
+
+ if (unlikely(vertex->need_conversion)) {
+ nv50->vbo_fifo = ~0;
+ nv50->vbo_user = 0;
+ } else {
+ nv50_prevalidate_vbufs(nv50);
+ }
+
+ BEGIN_RING(chan, RING_3D(VERTEX_ARRAY_ATTRIB(0)), vertex->num_elements);
+ for (i = 0; i < vertex->num_elements; ++i) {
+ ve = &vertex->element[i];
+ vb = &nv50->vtxbuf[ve->pipe.vertex_buffer_index];
+
+ if (likely(vb->stride) || nv50->vbo_fifo) {
+ OUT_RING(chan, ve->state);
+ } else {
+ OUT_RING(chan, ve->state | NV50_3D_VERTEX_ARRAY_ATTRIB_CONST);
+ nv50->vbo_fifo &= ~(1 << i);
+ }
+ }
+
+ for (i = 0; i < vertex->num_elements; ++i) {
+ struct nv04_resource *res;
+ unsigned size, offset;
+
+ ve = &vertex->element[i];
+ vb = &nv50->vtxbuf[ve->pipe.vertex_buffer_index];
+
+ if (unlikely(ve->pipe.instance_divisor)) {
+ if (!(nv50->state.instance_elts & (1 << i))) {
+ BEGIN_RING(chan, RING_3D(VERTEX_ARRAY_PER_INSTANCE(i)), 1);
+ OUT_RING (chan, 1);
+ }
+ BEGIN_RING(chan, RING_3D(VERTEX_ARRAY_DIVISOR(i)), 1);
+ OUT_RING (chan, ve->pipe.instance_divisor);
+ } else
+ if (unlikely(nv50->state.instance_elts & (1 << i))) {
+ BEGIN_RING(chan, RING_3D(VERTEX_ARRAY_PER_INSTANCE(i)), 1);
+ OUT_RING (chan, 0);
+ }
+
+ res = nv04_resource(vb->buffer);
+
+ if (nv50->vbo_fifo || unlikely(vb->stride == 0)) {
+ if (!nv50->vbo_fifo)
+ nv50_emit_vtxattr(nv50, vb, &ve->pipe, i);
+ BEGIN_RING(chan, RING_3D(VERTEX_ARRAY_FETCH(i)), 1);
+ OUT_RING (chan, 0);
+ continue;
+ }
+
+ size = vb->buffer->width0;
+ offset = ve->pipe.src_offset + vb->buffer_offset;
+
+ MARK_RING (chan, 8, 4);
+ BEGIN_RING(chan, RING_3D(VERTEX_ARRAY_FETCH(i)), 1);
+ OUT_RING (chan, NV50_3D_VERTEX_ARRAY_FETCH_ENABLE | vb->stride);
+ BEGIN_RING(chan, RING_3D(VERTEX_ARRAY_LIMIT_HIGH(i)), 2);
+ OUT_RESRCh(chan, res, size - 1, NOUVEAU_BO_RD);
+ OUT_RESRCl(chan, res, size - 1, NOUVEAU_BO_RD);
+ BEGIN_RING(chan, RING_3D(VERTEX_ARRAY_START_HIGH(i)), 2);
+ OUT_RESRCh(chan, res, offset, NOUVEAU_BO_RD);
+ OUT_RESRCl(chan, res, offset, NOUVEAU_BO_RD);
+ }
+ for (; i < nv50->state.num_vtxelts; ++i) {
+ BEGIN_RING(chan, RING_3D(VERTEX_ARRAY_ATTRIB(i)), 1);
+ OUT_RING (chan, NV50_3D_VERTEX_ATTRIB_INACTIVE);
+ BEGIN_RING(chan, RING_3D(VERTEX_ARRAY_FETCH(i)), 1);
+ OUT_RING (chan, 0);
+ }
+
+ nv50->state.num_vtxelts = vertex->num_elements;
+ nv50->state.instance_elts = vertex->instance_elts;
+}
+
+#define NV50_PRIM_GL_CASE(n) \
+ case PIPE_PRIM_##n: return NV50_3D_VERTEX_BEGIN_GL_PRIMITIVE_##n
- BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 2);
- OUT_RING (chan, NV50_CB_AUX | (24 << 8));
- OUT_RING (chan, startInstance);
-
- BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
- OUT_RING (chan, nv50_prim(mode));
-
- nv50_draw_elements_inline(nv50, map, indexSize, start, count);
-
- BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
- OUT_RING (chan, 0);
-
- for (i = 1; i < instanceCount; ++i) {
- if (nz_divisors) /* any non-zero array divisors ? */
- step_per_instance_arrays(nv50, pos, step);
-
- BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
- OUT_RING (chan, nv50_prim(mode) | (1 << 28));
-
- nv50_draw_elements_inline(nv50, map, indexSize, start, count);
-
- BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
- OUT_RING (chan, 0);
- }
- nv50_unmap_vbufs(nv50);
-
- so_ref(NULL, &nv50->state.instbuf);
+static INLINE unsigned
+nv50_prim_gl(unsigned prim)
+{
+ switch (prim) {
+ NV50_PRIM_GL_CASE(POINTS);
+ NV50_PRIM_GL_CASE(LINES);
+ NV50_PRIM_GL_CASE(LINE_LOOP);
+ NV50_PRIM_GL_CASE(LINE_STRIP);
+ NV50_PRIM_GL_CASE(TRIANGLES);
+ NV50_PRIM_GL_CASE(TRIANGLE_STRIP);
+ NV50_PRIM_GL_CASE(TRIANGLE_FAN);
+ NV50_PRIM_GL_CASE(QUADS);
+ NV50_PRIM_GL_CASE(QUAD_STRIP);
+ NV50_PRIM_GL_CASE(POLYGON);
+ NV50_PRIM_GL_CASE(LINES_ADJACENCY);
+ NV50_PRIM_GL_CASE(LINE_STRIP_ADJACENCY);
+ NV50_PRIM_GL_CASE(TRIANGLES_ADJACENCY);
+ NV50_PRIM_GL_CASE(TRIANGLE_STRIP_ADJACENCY);
+ default:
+ return NV50_3D_VERTEX_BEGIN_GL_PRIMITIVE_POINTS;
+ break;
+ }
}
-void
-nv50_draw_elements(struct pipe_context *pipe,
- struct pipe_buffer *indexBuffer, unsigned indexSize,
- unsigned mode, unsigned start, unsigned count)
+static void
+nv50_draw_vbo_flush_notify(struct nouveau_channel *chan)
{
- struct nv50_context *nv50 = nv50_context(pipe);
- struct nouveau_channel *chan = nv50->screen->tesla->channel;
- struct nouveau_grobj *tesla = nv50->screen->tesla;
- struct pipe_screen *pscreen = pipe->screen;
- void *map;
-
- map = pipe_buffer_map(pscreen, indexBuffer, PIPE_BUFFER_USAGE_CPU_READ);
-
- nv50_state_validate(nv50);
+ struct nv50_context *nv50 = chan->user_private;
- BEGIN_RING(chan, tesla, 0x142c, 1);
- OUT_RING (chan, 0);
- BEGIN_RING(chan, tesla, 0x142c, 1);
- OUT_RING (chan, 0);
+ nouveau_fence_update(&nv50->screen->base, TRUE);
- BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
- OUT_RING (chan, nv50_prim(mode));
-
- nv50_draw_elements_inline(nv50, map, indexSize, start, count);
-
- BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
- OUT_RING (chan, 0);
-
- nv50_unmap_vbufs(nv50);
-
- pipe_buffer_unmap(pscreen, indexBuffer);
-}
-
-static INLINE boolean
-nv50_vbo_static_attrib(struct nv50_context *nv50, unsigned attrib,
- struct nouveau_stateobj **pso,
- struct pipe_vertex_element *ve,
- struct pipe_vertex_buffer *vb)
-
-{
- struct nouveau_stateobj *so;
- struct nouveau_grobj *tesla = nv50->screen->tesla;
- struct nouveau_bo *bo = nouveau_bo(vb->buffer);
- float v[4];
- int ret;
-
- ret = nouveau_bo_map(bo, NOUVEAU_BO_RD);
- if (ret)
- return FALSE;
-
- util_format_read_4f(ve->src_format, v, 0, (uint8_t *)bo->map +
- (vb->buffer_offset + ve->src_offset), 0,
- 0, 0, 1, 1);
- so = *pso;
- if (!so)
- *pso = so = so_new(nv50->vtxelt_nr, nv50->vtxelt_nr * 4, 0);
-
- switch (ve->nr_components) {
- case 4:
- so_method(so, tesla, NV50TCL_VTX_ATTR_4F_X(attrib), 4);
- so_data (so, fui(v[0]));
- so_data (so, fui(v[1]));
- so_data (so, fui(v[2]));
- so_data (so, fui(v[3]));
- break;
- case 3:
- so_method(so, tesla, NV50TCL_VTX_ATTR_3F_X(attrib), 3);
- so_data (so, fui(v[0]));
- so_data (so, fui(v[1]));
- so_data (so, fui(v[2]));
- break;
- case 2:
- so_method(so, tesla, NV50TCL_VTX_ATTR_2F_X(attrib), 2);
- so_data (so, fui(v[0]));
- so_data (so, fui(v[1]));
- break;
- case 1:
- if (attrib == nv50->vertprog->cfg.edgeflag_in) {
- so_method(so, tesla, NV50TCL_EDGEFLAG_ENABLE, 1);
- so_data (so, v[0] ? 1 : 0);
- }
- so_method(so, tesla, NV50TCL_VTX_ATTR_1F(attrib), 1);
- so_data (so, fui(v[0]));
- break;
- default:
- nouveau_bo_unmap(bo);
- return FALSE;
- }
-
- nouveau_bo_unmap(bo);
- return TRUE;
-}
-
-void
-nv50_vbo_validate(struct nv50_context *nv50)
-{
- struct nouveau_grobj *tesla = nv50->screen->tesla;
- struct nouveau_stateobj *vtxbuf, *vtxfmt, *vtxattr;
- unsigned i, n_ve;
-
- /* don't validate if Gallium took away our buffers */
- if (nv50->vtxbuf_nr == 0)
- return;
- nv50->vbo_fifo = 0;
-
- for (i = 0; i < nv50->vtxbuf_nr; ++i)
- if (nv50->vtxbuf[i].stride &&
- !(nv50->vtxbuf[i].buffer->usage & PIPE_BUFFER_USAGE_VERTEX))
- nv50->vbo_fifo = 0xffff;
-
- if (NV50_USING_LOATHED_EDGEFLAG(nv50))
- nv50->vbo_fifo = 0xffff; /* vertprog can't set edgeflag */
-
- n_ve = MAX2(nv50->vtxelt_nr, nv50->state.vtxelt_nr);
-
- vtxattr = NULL;
- vtxbuf = so_new(n_ve * 2, n_ve * 5, nv50->vtxelt_nr * 4);
- vtxfmt = so_new(1, n_ve, 0);
- so_method(vtxfmt, tesla, NV50TCL_VERTEX_ARRAY_ATTRIB(0), n_ve);
-
- for (i = 0; i < nv50->vtxelt_nr; i++) {
- struct pipe_vertex_element *ve = &nv50->vtxelt[i];
- struct pipe_vertex_buffer *vb =
- &nv50->vtxbuf[ve->vertex_buffer_index];
- struct nouveau_bo *bo = nouveau_bo(vb->buffer);
- uint32_t hw = nv50_vbo_vtxelt_to_hw(ve);
-
- if (!vb->stride &&
- nv50_vbo_static_attrib(nv50, i, &vtxattr, ve, vb)) {
- so_data(vtxfmt, hw | (1 << 4));
-
- so_method(vtxbuf, tesla,
- NV50TCL_VERTEX_ARRAY_FORMAT(i), 1);
- so_data (vtxbuf, 0);
-
- nv50->vbo_fifo &= ~(1 << i);
- continue;
- }
-
- if (nv50->vbo_fifo) {
- so_data (vtxfmt, hw |
- (ve->instance_divisor ? (1 << 4) : i));
- so_method(vtxbuf, tesla,
- NV50TCL_VERTEX_ARRAY_FORMAT(i), 1);
- so_data (vtxbuf, 0);
- continue;
- }
- so_data(vtxfmt, hw | i);
-
- so_method(vtxbuf, tesla, NV50TCL_VERTEX_ARRAY_FORMAT(i), 3);
- so_data (vtxbuf, 0x20000000 |
- (ve->instance_divisor ? 0 : vb->stride));
- so_reloc (vtxbuf, bo, vb->buffer_offset +
- ve->src_offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART |
- NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0);
- so_reloc (vtxbuf, bo, vb->buffer_offset +
- ve->src_offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART |
- NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0);
-
- /* vertex array limits */
- so_method(vtxbuf, tesla, NV50TCL_VERTEX_ARRAY_LIMIT_HIGH(i), 2);
- so_reloc (vtxbuf, bo, vb->buffer->size - 1,
- NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD |
- NOUVEAU_BO_HIGH, 0, 0);
- so_reloc (vtxbuf, bo, vb->buffer->size - 1,
- NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD |
- NOUVEAU_BO_LOW, 0, 0);
- }
- for (; i < n_ve; ++i) {
- so_data (vtxfmt, 0x7e080010);
-
- so_method(vtxbuf, tesla, NV50TCL_VERTEX_ARRAY_FORMAT(i), 1);
- so_data (vtxbuf, 0);
- }
- nv50->state.vtxelt_nr = nv50->vtxelt_nr;
-
- so_ref (vtxfmt, &nv50->state.vtxfmt);
- so_ref (vtxbuf, &nv50->state.vtxbuf);
- so_ref (vtxattr, &nv50->state.vtxattr);
- so_ref (NULL, &vtxbuf);
- so_ref (NULL, &vtxfmt);
- so_ref (NULL, &vtxattr);
+ nv50_bufctx_emit_relocs(nv50);
}
-typedef void (*pfn_push)(struct nouveau_channel *, void *);
-
-struct nv50_vbo_emitctx
+static void
+nv50_draw_arrays(struct nv50_context *nv50,
+ unsigned mode, unsigned start, unsigned count,
+ unsigned instance_count)
{
- pfn_push push[16];
- uint8_t *map[16];
- unsigned stride[16];
- unsigned nr_ve;
- unsigned vtx_dwords;
- unsigned vtx_max;
-
- float edgeflag;
- unsigned ve_edgeflag;
-};
+ struct nouveau_channel *chan = nv50->screen->base.channel;
+ unsigned prim;
-static INLINE void
-emit_vtx_next(struct nouveau_channel *chan, struct nv50_vbo_emitctx *emit)
-{
- unsigned i;
+ chan->flush_notify = nv50_draw_vbo_flush_notify;
+ chan->user_private = nv50;
- for (i = 0; i < emit->nr_ve; ++i) {
- emit->push[i](chan, emit->map[i]);
- emit->map[i] += emit->stride[i];
- }
-}
+ prim = nv50_prim_gl(mode);
-static INLINE void
-emit_vtx(struct nouveau_channel *chan, struct nv50_vbo_emitctx *emit,
- uint32_t vi)
-{
- unsigned i;
+ while (instance_count--) {
+ BEGIN_RING(chan, RING_3D(VERTEX_BEGIN_GL), 1);
+ OUT_RING (chan, prim);
+ BEGIN_RING(chan, RING_3D(VERTEX_BUFFER_FIRST), 2);
+ OUT_RING (chan, start);
+ OUT_RING (chan, count);
+ BEGIN_RING(chan, RING_3D(VERTEX_END_GL), 1);
+ OUT_RING (chan, 0);
- for (i = 0; i < emit->nr_ve; ++i)
- emit->push[i](chan, emit->map[i] + emit->stride[i] * vi);
-}
+ prim |= NV50_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT;
+ }
-static INLINE boolean
-nv50_map_vbufs(struct nv50_context *nv50)
-{
- int i;
-
- for (i = 0; i < nv50->vtxbuf_nr; ++i) {
- struct pipe_vertex_buffer *vb = &nv50->vtxbuf[i];
- unsigned size = vb->stride * (vb->max_index + 1) + 16;
-
- if (nouveau_bo(vb->buffer)->map)
- continue;
-
- size = vb->stride * (vb->max_index + 1) + 16;
- size = MIN2(size, vb->buffer->size);
- if (!size)
- size = vb->buffer->size;
-
- if (nouveau_bo_map_range(nouveau_bo(vb->buffer),
- 0, size, NOUVEAU_BO_RD))
- break;
- }
-
- if (i == nv50->vtxbuf_nr)
- return TRUE;
- for (; i >= 0; --i)
- nouveau_bo_unmap(nouveau_bo(nv50->vtxbuf[i].buffer));
- return FALSE;
+ chan->flush_notify = nv50_default_flush_notify;
}
static void
-emit_b32_1(struct nouveau_channel *chan, void *data)
-{
- uint32_t *v = data;
-
- OUT_RING(chan, v[0]);
+nv50_draw_elements_inline_u08(struct nouveau_channel *chan, uint8_t *map,
+ unsigned start, unsigned count)
+{
+ map += start;
+
+ if (count & 3) {
+ unsigned i;
+ BEGIN_RING_NI(chan, RING_3D(VB_ELEMENT_U32), count & 3);
+ for (i = 0; i < (count & 3); ++i)
+ OUT_RING(chan, *map++);
+ count &= ~3;
+ }
+ while (count) {
+ unsigned i, nr = MIN2(count, NV04_PFIFO_MAX_PACKET_LEN * 4) / 4;
+
+ BEGIN_RING_NI(chan, RING_3D(VB_ELEMENT_U8), nr);
+ for (i = 0; i < nr; ++i) {
+ OUT_RING(chan,
+ (map[3] << 24) | (map[2] << 16) | (map[1] << 8) | map[0]);
+ map += 4;
+ }
+ count -= nr * 4;
+ }
}
static void
-emit_b32_2(struct nouveau_channel *chan, void *data)
+nv50_draw_elements_inline_u16(struct nouveau_channel *chan, uint16_t *map,
+ unsigned start, unsigned count)
{
- uint32_t *v = data;
+ map += start;
- OUT_RING(chan, v[0]);
- OUT_RING(chan, v[1]);
-}
-
-static void
-emit_b32_3(struct nouveau_channel *chan, void *data)
-{
- uint32_t *v = data;
+ if (count & 1) {
+ count &= ~1;
+ BEGIN_RING(chan, RING_3D(VB_ELEMENT_U32), 1);
+ OUT_RING (chan, *map++);
+ }
+ while (count) {
+ unsigned i, nr = MIN2(count, NV04_PFIFO_MAX_PACKET_LEN * 2) / 2;
- OUT_RING(chan, v[0]);
- OUT_RING(chan, v[1]);
- OUT_RING(chan, v[2]);
+ BEGIN_RING_NI(chan, RING_3D(VB_ELEMENT_U16), nr);
+ for (i = 0; i < nr; ++i) {
+ OUT_RING(chan, (map[1] << 16) | map[0]);
+ map += 2;
+ }
+ count -= nr * 2;
+ }
}
static void
-emit_b32_4(struct nouveau_channel *chan, void *data)
+nv50_draw_elements_inline_u32(struct nouveau_channel *chan, uint32_t *map,
+ unsigned start, unsigned count)
{
- uint32_t *v = data;
+ map += start;
- OUT_RING(chan, v[0]);
- OUT_RING(chan, v[1]);
- OUT_RING(chan, v[2]);
- OUT_RING(chan, v[3]);
-}
+ while (count) {
+ const unsigned nr = MIN2(count, NV04_PFIFO_MAX_PACKET_LEN);
-static void
-emit_b16_1(struct nouveau_channel *chan, void *data)
-{
- uint16_t *v = data;
+ BEGIN_RING_NI(chan, RING_3D(VB_ELEMENT_U32), nr);
+ OUT_RINGp (chan, map, nr);
- OUT_RING(chan, v[0]);
+ map += nr;
+ count -= nr;
+ }
}
static void
-emit_b16_3(struct nouveau_channel *chan, void *data)
+nv50_draw_elements_inline_u32_short(struct nouveau_channel *chan, uint32_t *map,
+ unsigned start, unsigned count)
{
- uint16_t *v = data;
-
- OUT_RING(chan, (v[1] << 16) | v[0]);
- OUT_RING(chan, v[2]);
-}
+ map += start;
-static void
-emit_b08_1(struct nouveau_channel *chan, void *data)
-{
- uint8_t *v = data;
+ if (count & 1) {
+ count--;
+ BEGIN_RING(chan, RING_3D(VB_ELEMENT_U32), 1);
+ OUT_RING (chan, *map++);
+ }
+ while (count) {
+ unsigned i, nr = MIN2(count, NV04_PFIFO_MAX_PACKET_LEN * 2) / 2;
- OUT_RING(chan, v[0]);
+ BEGIN_RING_NI(chan, RING_3D(VB_ELEMENT_U16), nr);
+ for (i = 0; i < nr; ++i) {
+ OUT_RING(chan, (map[1] << 16) | map[0]);
+ map += 2;
+ }
+ count -= nr * 2;
+ }
}
static void
-emit_b08_3(struct nouveau_channel *chan, void *data)
-{
- uint8_t *v = data;
-
- OUT_RING(chan, (v[2] << 16) | (v[1] << 8) | v[0]);
-}
-
-static boolean
-emit_prepare(struct nv50_context *nv50, struct nv50_vbo_emitctx *emit,
- unsigned start)
-{
- unsigned i;
-
- if (nv50_map_vbufs(nv50) == FALSE)
- return FALSE;
-
- emit->ve_edgeflag = nv50->vertprog->cfg.edgeflag_in;
-
- emit->edgeflag = 0.5f;
- emit->nr_ve = 0;
- emit->vtx_dwords = 0;
-
- for (i = 0; i < nv50->vtxelt_nr; ++i) {
- struct pipe_vertex_element *ve;
- struct pipe_vertex_buffer *vb;
- unsigned n, size;
- const struct util_format_description *desc;
-
- ve = &nv50->vtxelt[i];
- vb = &nv50->vtxbuf[ve->vertex_buffer_index];
- if (!(nv50->vbo_fifo & (1 << i)) || ve->instance_divisor)
- continue;
- n = emit->nr_ve++;
-
- emit->stride[n] = vb->stride;
- emit->map[n] = (uint8_t *)nouveau_bo(vb->buffer)->map +
- vb->buffer_offset +
- (start * vb->stride + ve->src_offset);
-
- desc = util_format_description(ve->src_format);
- assert(desc);
-
- size = util_format_get_component_bits(
- ve->src_format, UTIL_FORMAT_COLORSPACE_RGB, 0);
-
- assert(ve->nr_components > 0 && ve->nr_components <= 4);
-
- /* It shouldn't be necessary to push the implicit 1s
- * for case 3 and size 8 cases 1, 2, 3.
- */
- switch (size) {
- default:
- NOUVEAU_ERR("unsupported vtxelt size: %u\n", size);
- return FALSE;
- case 32:
- switch (ve->nr_components) {
- case 1: emit->push[n] = emit_b32_1; break;
- case 2: emit->push[n] = emit_b32_2; break;
- case 3: emit->push[n] = emit_b32_3; break;
- case 4: emit->push[n] = emit_b32_4; break;
- }
- emit->vtx_dwords += ve->nr_components;
- break;
- case 16:
- switch (ve->nr_components) {
- case 1: emit->push[n] = emit_b16_1; break;
- case 2: emit->push[n] = emit_b32_1; break;
- case 3: emit->push[n] = emit_b16_3; break;
- case 4: emit->push[n] = emit_b32_2; break;
- }
- emit->vtx_dwords += (ve->nr_components + 1) >> 1;
- break;
- case 8:
- switch (ve->nr_components) {
- case 1: emit->push[n] = emit_b08_1; break;
- case 2: emit->push[n] = emit_b16_1; break;
- case 3: emit->push[n] = emit_b08_3; break;
- case 4: emit->push[n] = emit_b32_1; break;
- }
- emit->vtx_dwords += 1;
- break;
- }
- }
-
- emit->vtx_max = 512 / emit->vtx_dwords;
- if (emit->ve_edgeflag < 16)
- emit->vtx_max = 1;
-
- return TRUE;
-}
-
-static INLINE void
-set_edgeflag(struct nouveau_channel *chan,
- struct nouveau_grobj *tesla,
- struct nv50_vbo_emitctx *emit, uint32_t index)
-{
- unsigned i = emit->ve_edgeflag;
-
- if (i < 16) {
- float f = *((float *)(emit->map[i] + index * emit->stride[i]));
-
- if (emit->edgeflag != f) {
- emit->edgeflag = f;
-
- BEGIN_RING(chan, tesla, 0x15e4, 1);
- OUT_RING (chan, f ? 1 : 0);
- }
- }
+nv50_draw_elements(struct nv50_context *nv50, boolean shorten,
+ unsigned mode, unsigned start, unsigned count,
+ unsigned instance_count, int32_t index_bias)
+{
+ struct nouveau_channel *chan = nv50->screen->base.channel;
+ void *data;
+ unsigned prim;
+ const unsigned index_size = nv50->idxbuf.index_size;
+
+ chan->flush_notify = nv50_draw_vbo_flush_notify;
+ chan->user_private = nv50;
+
+ prim = nv50_prim_gl(mode);
+
+ if (index_bias != nv50->state.index_bias) {
+ BEGIN_RING(chan, RING_3D(VB_ELEMENT_BASE), 1);
+ OUT_RING (chan, index_bias);
+ nv50->state.index_bias = index_bias;
+ }
+
+ if (nouveau_resource_mapped_by_gpu(nv50->idxbuf.buffer)) {
+ struct nv04_resource *res = nv04_resource(nv50->idxbuf.buffer);
+
+ start += nv50->idxbuf.offset >> (index_size >> 1);
+
+ nouveau_buffer_adjust_score(&nv50->base, res, 1);
+
+ while (instance_count--) {
+ BEGIN_RING(chan, RING_3D(VERTEX_BEGIN_GL), 1);
+ OUT_RING (chan, mode);
+
+ switch (index_size) {
+ case 4:
+ {
+ WAIT_RING (chan, 2);
+ BEGIN_RING(chan, RING_3D(VB_ELEMENT_U32) | 0x30000, 0);
+ OUT_RING (chan, count);
+ nouveau_pushbuf_submit(chan, res->bo, res->offset + start * 4,
+ count * 4);
+ }
+ break;
+ case 2:
+ {
+ unsigned pb_start = (start & ~1);
+ unsigned pb_words = (((start + count + 1) & ~1) - pb_start) >> 1;
+
+ BEGIN_RING(chan, RING_3D(VB_ELEMENT_U16_SETUP), 1);
+ OUT_RING (chan, (start << 31) | count);
+ WAIT_RING (chan, 2);
+ BEGIN_RING(chan, RING_3D(VB_ELEMENT_U16) | 0x30000, 0);
+ OUT_RING (chan, pb_words);
+ nouveau_pushbuf_submit(chan, res->bo, res->offset + pb_start * 2,
+ pb_words * 4);
+ BEGIN_RING(chan, RING_3D(VB_ELEMENT_U16_SETUP), 1);
+ OUT_RING (chan, 0);
+ break;
+ }
+ case 1:
+ {
+ unsigned pb_start = (start & ~3);
+ unsigned pb_words = (((start + count + 3) & ~3) - pb_start) >> 1;
+
+ BEGIN_RING(chan, RING_3D(VB_ELEMENT_U8_SETUP), 1);
+ OUT_RING (chan, (start << 30) | count);
+ WAIT_RING (chan, 2);
+ BEGIN_RING(chan, RING_3D(VB_ELEMENT_U8) | 0x30000, 0);
+ OUT_RING (chan, pb_words);
+ nouveau_pushbuf_submit(chan, res->bo, res->offset + pb_start,
+ pb_words * 4);
+ BEGIN_RING(chan, RING_3D(VB_ELEMENT_U8_SETUP), 1);
+ OUT_RING (chan, 0);
+ break;
+ }
+ default:
+ assert(0);
+ return;
+ }
+ BEGIN_RING(chan, RING_3D(VERTEX_END_GL), 1);
+ OUT_RING (chan, 0);
+
+ nv50_resource_fence(res, NOUVEAU_BO_RD);
+
+ mode |= NV50_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT;
+ }
+ } else {
+ data = nouveau_resource_map_offset(&nv50->base,
+ nv04_resource(nv50->idxbuf.buffer),
+ nv50->idxbuf.offset, NOUVEAU_BO_RD);
+ if (!data)
+ return;
+
+ while (instance_count--) {
+ BEGIN_RING(chan, RING_3D(VERTEX_BEGIN_GL), 1);
+ OUT_RING (chan, prim);
+ switch (index_size) {
+ case 1:
+ nv50_draw_elements_inline_u08(chan, data, start, count);
+ break;
+ case 2:
+ nv50_draw_elements_inline_u16(chan, data, start, count);
+ break;
+ case 4:
+ if (shorten)
+ nv50_draw_elements_inline_u32_short(chan, data, start, count);
+ else
+ nv50_draw_elements_inline_u32(chan, data, start, count);
+ break;
+ default:
+ assert(0);
+ return;
+ }
+ BEGIN_RING(chan, RING_3D(VERTEX_END_GL), 1);
+ OUT_RING (chan, 0);
+
+ prim |= NV50_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT;
+ }
+ }
+
+ chan->flush_notify = nv50_default_flush_notify;
}
-static boolean
-nv50_push_arrays(struct nv50_context *nv50, unsigned start, unsigned count)
-{
- struct nouveau_channel *chan = nv50->screen->base.channel;
- struct nouveau_grobj *tesla = nv50->screen->tesla;
- struct nv50_vbo_emitctx emit;
-
- if (emit_prepare(nv50, &emit, start) == FALSE)
- return FALSE;
-
- while (count) {
- unsigned i, dw, nr = MIN2(count, emit.vtx_max);
- dw = nr * emit.vtx_dwords;
-
- set_edgeflag(chan, tesla, &emit, 0); /* nr will be 1 */
-
- BEGIN_RING_NI(chan, tesla, NV50TCL_VERTEX_DATA, dw);
- for (i = 0; i < nr; ++i)
- emit_vtx_next(chan, &emit);
-
- count -= nr;
- }
-
- return TRUE;
-}
-
-static boolean
-nv50_push_elements_u32(struct nv50_context *nv50, uint32_t *map, unsigned count)
-{
- struct nouveau_channel *chan = nv50->screen->base.channel;
- struct nouveau_grobj *tesla = nv50->screen->tesla;
- struct nv50_vbo_emitctx emit;
-
- if (emit_prepare(nv50, &emit, 0) == FALSE)
- return FALSE;
-
- while (count) {
- unsigned i, dw, nr = MIN2(count, emit.vtx_max);
- dw = nr * emit.vtx_dwords;
-
- set_edgeflag(chan, tesla, &emit, *map);
-
- BEGIN_RING_NI(chan, tesla, NV50TCL_VERTEX_DATA, dw);
- for (i = 0; i < nr; ++i)
- emit_vtx(chan, &emit, *map++);
-
- count -= nr;
- }
-
- return TRUE;
-}
-
-static boolean
-nv50_push_elements_u16(struct nv50_context *nv50, uint16_t *map, unsigned count)
-{
- struct nouveau_channel *chan = nv50->screen->base.channel;
- struct nouveau_grobj *tesla = nv50->screen->tesla;
- struct nv50_vbo_emitctx emit;
-
- if (emit_prepare(nv50, &emit, 0) == FALSE)
- return FALSE;
-
- while (count) {
- unsigned i, dw, nr = MIN2(count, emit.vtx_max);
- dw = nr * emit.vtx_dwords;
-
- set_edgeflag(chan, tesla, &emit, *map);
-
- BEGIN_RING_NI(chan, tesla, NV50TCL_VERTEX_DATA, dw);
- for (i = 0; i < nr; ++i)
- emit_vtx(chan, &emit, *map++);
-
- count -= nr;
- }
-
- return TRUE;
-}
-
-static boolean
-nv50_push_elements_u08(struct nv50_context *nv50, uint8_t *map, unsigned count)
-{
- struct nouveau_channel *chan = nv50->screen->base.channel;
- struct nouveau_grobj *tesla = nv50->screen->tesla;
- struct nv50_vbo_emitctx emit;
-
- if (emit_prepare(nv50, &emit, 0) == FALSE)
- return FALSE;
-
- while (count) {
- unsigned i, dw, nr = MIN2(count, emit.vtx_max);
- dw = nr * emit.vtx_dwords;
-
- set_edgeflag(chan, tesla, &emit, *map);
-
- BEGIN_RING_NI(chan, tesla, NV50TCL_VERTEX_DATA, dw);
- for (i = 0; i < nr; ++i)
- emit_vtx(chan, &emit, *map++);
-
- count -= nr;
- }
-
- return TRUE;
+void
+nv50_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
+{
+ struct nv50_context *nv50 = nv50_context(pipe);
+ struct nouveau_channel *chan = nv50->screen->base.channel;
+
+ /* For picking only a few vertices from a large user buffer, push is better,
+ * if index count is larger and we expect repeated vertices, suggest upload.
+ */
+ nv50->vbo_push_hint = /* the 64 is heuristic */
+ !(info->indexed &&
+ ((info->max_index - info->min_index + 64) < info->count));
+
+ nv50->vbo_min_index = info->min_index;
+ nv50->vbo_max_index = info->max_index;
+
+ if (nv50->vbo_push_hint != !!nv50->vbo_fifo)
+ nv50->dirty |= NV50_NEW_ARRAYS;
+
+ if (nv50->vbo_user && !(nv50->dirty & (NV50_NEW_VERTEX | NV50_NEW_ARRAYS)))
+ nv50_update_user_vbufs(nv50);
+
+ nv50_state_validate(nv50);
+
+ if (nv50->vbo_fifo) {
+ nv50_push_vbo(nv50, info);
+ return;
+ }
+
+ if (nv50->state.instance_base != info->start_instance) {
+ nv50->state.instance_base = info->start_instance;
+ /* NOTE: this does not affect the shader input, should it ? */
+ BEGIN_RING(chan, RING_3D(VB_INSTANCE_BASE), 1);
+ OUT_RING (chan, info->start_instance);
+ }
+
+ if (nv50->base.vbo_dirty) {
+ BEGIN_RING(chan, RING_3D(VERTEX_ARRAY_FLUSH), 1);
+ OUT_RING (chan, 0);
+ nv50->base.vbo_dirty = FALSE;
+ }
+
+ if (!info->indexed) {
+ nv50_draw_arrays(nv50,
+ info->mode, info->start, info->count,
+ info->instance_count);
+ } else {
+ boolean shorten = info->max_index <= 65535;
+
+ assert(nv50->idxbuf.buffer);
+
+ if (info->primitive_restart != nv50->state.prim_restart) {
+ if (info->primitive_restart) {
+ BEGIN_RING(chan, RING_3D(PRIM_RESTART_ENABLE), 2);
+ OUT_RING (chan, 1);
+ OUT_RING (chan, info->restart_index);
+
+ if (info->restart_index > 65535)
+ shorten = FALSE;
+ } else {
+ BEGIN_RING(chan, RING_3D(PRIM_RESTART_ENABLE), 1);
+ OUT_RING (chan, 0);
+ }
+ nv50->state.prim_restart = info->primitive_restart;
+ } else
+ if (info->primitive_restart) {
+ BEGIN_RING(chan, RING_3D(PRIM_RESTART_INDEX), 1);
+ OUT_RING (chan, info->restart_index);
+
+ if (info->restart_index > 65535)
+ shorten = FALSE;
+ }
+
+ nv50_draw_elements(nv50, shorten,
+ info->mode, info->start, info->count,
+ info->instance_count, info->index_bias);
+ }
+
+ nv50_release_user_vbufs(nv50);
}
projects / mesa.git / blobdiff