#include "util/u_inlines.h"
#include "util/u_format.h"
+#include "nouveau/nouveau_util.h"
#include "nv50_context.h"
-#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)
{
if (!nv50_state_validate(nv50, 10 + 16*3))
return;
+ if (nv50->vbo_fifo) {
+ nv50_push_elements_instanced(pipe, NULL, 0, mode, start,
+ count, startInstance,
+ instanceCount);
+ return;
+ }
+
BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 2);
OUT_RING (chan, NV50_CB_AUX | (24 << 8));
OUT_RING (chan, startInstance);
nv50_draw_arrays_instanced(pipe, mode, start, count, 0, 1);
}
-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;
+struct inline_ctx {
+ struct nv50_context *nv50;
+ void *map;
+};
- map += start;
+static void
+inline_elt08(void *priv, unsigned start, unsigned count)
+{
+ struct inline_ctx *ctx = priv;
+ struct nouveau_grobj *tesla = ctx->nv50->screen->tesla;
+ struct nouveau_channel *chan = tesla->channel;
+ uint8_t *map = (uint8_t *)ctx->map + start;
if (count & 1) {
BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32, 1);
OUT_RING (chan, map[0]);
map++;
- count--;
+ count &= ~1;
}
- 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 >>= 1;
+ if (!count)
+ return;
- count -= nr;
- map += nr;
+ BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U16, count);
+ while (count--) {
+ OUT_RING(chan, (map[1] << 16) | map[0]);
+ map += 2;
}
- return TRUE;
}
-static INLINE boolean
-nv50_draw_elements_inline_u16(struct nv50_context *nv50, uint16_t *map,
- unsigned start, unsigned count)
+static void
+inline_elt16(void *priv, unsigned start, unsigned count)
{
- struct nouveau_channel *chan = nv50->screen->tesla->channel;
- struct nouveau_grobj *tesla = nv50->screen->tesla;
-
- map += start;
+ struct inline_ctx *ctx = priv;
+ struct nouveau_grobj *tesla = ctx->nv50->screen->tesla;
+ struct nouveau_channel *chan = tesla->channel;
+ uint16_t *map = (uint16_t *)ctx->map + start;
if (count & 1) {
BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32, 1);
OUT_RING (chan, map[0]);
+ count &= ~1;
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 >>= 1;
+ if (!count)
+ return;
- count -= nr;
- map += nr;
+ BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U16, count);
+ while (count--) {
+ OUT_RING(chan, (map[1] << 16) | map[0]);
+ map += 2;
}
- return TRUE;
}
-static INLINE boolean
-nv50_draw_elements_inline_u32(struct nv50_context *nv50, uint32_t *map,
- unsigned start, unsigned count)
+static void
+inline_elt32(void *priv, unsigned start, unsigned count)
{
- struct nouveau_channel *chan = nv50->screen->tesla->channel;
- struct nouveau_grobj *tesla = nv50->screen->tesla;
-
- map += start;
-
- 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;
-}
+ struct inline_ctx *ctx = priv;
+ struct nouveau_grobj *tesla = ctx->nv50->screen->tesla;
+ struct nouveau_channel *chan = tesla->channel;
-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;
- }
+ BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U32, count);
+ OUT_RINGp (chan, (uint32_t *)ctx->map + start, count);
}
-static unsigned
-init_per_instance_arrays(struct nv50_context *nv50,
- unsigned startInstance,
- unsigned pos[16], unsigned step[16])
+static void
+inline_edgeflag(void *priv, boolean enabled)
{
- struct nouveau_grobj *tesla = nv50->screen->tesla;
+ struct inline_ctx *ctx = priv;
+ struct nouveau_grobj *tesla = ctx->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;
-
- 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;
+ BEGIN_RING(chan, tesla, NV50TCL_EDGEFLAG_ENABLE, 1);
+ OUT_RING (chan, enabled ? 1 : 0);
}
static void
-step_per_instance_arrays(struct nv50_context *nv50,
- unsigned pos[16], unsigned step[16])
+nv50_draw_elements_inline(struct pipe_context *pipe,
+ struct pipe_buffer *indexBuffer, unsigned indexSize,
+ unsigned mode, unsigned start, unsigned count,
+ unsigned startInstance, unsigned instanceCount)
{
+ struct pipe_screen *pscreen = pipe->screen;
+ struct nv50_context *nv50 = nv50_context(pipe);
+ struct nouveau_channel *chan = nv50->screen->tesla->channel;
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;
-
- 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;
+ struct instance a[16];
+ struct inline_ctx ctx;
+ struct u_split_prim s;
+ boolean nzi = FALSE;
+ unsigned overhead;
+
+ overhead = 16*3; /* potential instance adjustments */
+ overhead += 4; /* Begin()/End() */
+ overhead += 4; /* potential edgeflag disable/reenable */
+ overhead += 3; /* potentially 3 VTX_ELT_U16/U32 packet headers */
+
+ s.priv = &ctx;
+ if (indexSize == 1)
+ s.emit = inline_elt08;
+ else
+ if (indexSize == 2)
+ s.emit = inline_elt16;
+ else
+ s.emit = inline_elt32;
+ s.edge = inline_edgeflag;
+
+ ctx.nv50 = nv50;
+ ctx.map = pipe_buffer_map(pscreen, indexBuffer, PIPE_BUFFER_USAGE_CPU_READ);
+ assert(ctx.map);
+ if (!ctx.map)
+ return;
- if (++step[i] == nv50->vtxelt[i].instance_divisor) {
- step[i] = 0;
- pos[i] += nv50->vtxbuf[b].stride;
- }
+ instance_init(nv50, a, startInstance);
+ if (!nv50_state_validate(nv50, overhead + 6 + 3))
+ return;
- bo = nouveau_bo(nv50->vtxbuf[b].buffer);
+ BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 2);
+ OUT_RING (chan, NV50_CB_AUX | (24 << 8));
+ OUT_RING (chan, startInstance);
+ while (instanceCount--) {
+ unsigned max_verts;
+ boolean done;
+
+ u_split_prim_init(&s, mode, start, count);
+ do {
+ if (AVAIL_RING(chan) < (overhead + 6)) {
+ FIRE_RING(chan);
+ if (!nv50_state_validate(nv50, (overhead + 6))) {
+ assert(0);
+ return;
+ }
+ }
- 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);
+ max_verts = AVAIL_RING(chan) - overhead;
+ if (max_verts > 2047)
+ max_verts = 2047;
+ if (indexSize != 4)
+ max_verts <<= 1;
+ instance_step(nv50, a);
+
+ BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
+ OUT_RING (chan, nv50_prim(s.mode) | (nzi ? (1<<28) : 0));
+ done = u_split_prim_next(&s, max_verts);
+ BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
+ OUT_RING (chan, 0);
+ } while (!done);
+
+ nzi = TRUE;
}
- so_ref (so, &nv50->state.instbuf); /* for flush notify */
- so_ref (NULL, &so);
-
- so_emit(chan, nv50->state.instbuf);
+ pipe_buffer_unmap(pscreen, indexBuffer);
}
void
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;
- 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);
+ struct instance a[16];
+ unsigned prim = nv50_prim(mode);
- if (!nv50_state_validate(nv50, 0))
+ instance_init(nv50, a, startInstance);
+ if (!nv50_state_validate(nv50, 13 + 16*3))
return;
- chan->flush_notify = nv50_state_flush_notify;
- nz_divisors = init_per_instance_arrays(nv50, startInstance, pos, step);
+ if (nv50->vbo_fifo) {
+ nv50_push_elements_instanced(pipe, indexBuffer, indexSize,
+ mode, start, count, startInstance,
+ instanceCount);
+ return;
+ } else
+ if (!(indexBuffer->usage & PIPE_BUFFER_USAGE_INDEX) || indexSize == 1) {
+ nv50_draw_elements_inline(pipe, indexBuffer, indexSize,
+ mode, start, count, startInstance,
+ instanceCount);
+ return;
+ }
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);
+ while (instanceCount--) {
+ if (AVAIL_RING(chan) < (7 + 16*3)) {
+ FIRE_RING(chan);
+ if (!nv50_state_validate(nv50, 10 + 16*3)) {
+ assert(0);
+ return;
+ }
+ }
+ instance_step(nv50, a);
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);
-
+ OUT_RING (chan, prim);
+ if (indexSize == 4) {
+ BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32 | 0x30000, 0);
+ OUT_RING (chan, count);
+ nouveau_pushbuf_submit(chan, nouveau_bo(indexBuffer),
+ start << 2, count << 2);
+ } else
+ if (indexSize == 2) {
+ unsigned vb_start = (start & ~1);
+ unsigned vb_end = (start + count + 1) & ~1;
+ unsigned dwords = (vb_end - vb_start) >> 1;
+
+ BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16_SETUP, 1);
+ OUT_RING (chan, ((start & 1) << 31) | count);
+ BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16 | 0x30000, 0);
+ OUT_RING (chan, dwords);
+ nouveau_pushbuf_submit(chan, nouveau_bo(indexBuffer),
+ vb_start << 1, dwords << 2);
+ BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16_SETUP, 1);
+ OUT_RING (chan, 0);
+ }
BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
OUT_RING (chan, 0);
- }
-
- chan->flush_notify = NULL;
- so_ref(NULL, &nv50->state.instbuf);
+ prim |= (1 << 28);
+ }
}
void
struct pipe_buffer *indexBuffer, unsigned indexSize,
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;
- struct pipe_screen *pscreen = pipe->screen;
- void *map;
-
- if (!nv50_state_validate(nv50, 14))
- return;
- chan->flush_notify = nv50_state_flush_notify;
-
- BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
- OUT_RING (chan, nv50_prim(mode));
-
- if (indexSize == 4) {
- BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32 | 0x30000, 0);
- OUT_RING (chan, count);
- nouveau_pushbuf_submit(chan, nouveau_bo(indexBuffer),
- start << 2, count << 2);
- } else
- if (indexSize == 2) {
- unsigned vb_start = (start & ~1);
- unsigned vb_end = (start + count + 1) & ~1;
- unsigned dwords = (vb_end - vb_start) >> 1;
-
- BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16_SETUP, 1);
- OUT_RING (chan, ((start & 1) << 31) | count);
- BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16 | 0x30000, 0);
- OUT_RING (chan, dwords);
- nouveau_pushbuf_submit(chan, nouveau_bo(indexBuffer),
- vb_start << 1, dwords << 2);
- BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16_SETUP, 1);
- OUT_RING (chan, 0);
- } else {
- map = pipe_buffer_map(pscreen, indexBuffer,
- PIPE_BUFFER_USAGE_CPU_READ);
- nv50_draw_elements_inline(nv50, map, indexSize, start, count);
- pipe_buffer_unmap(pscreen, indexBuffer);
- }
-
- BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
- OUT_RING (chan, 0);
- chan->flush_notify = NULL;
+ nv50_draw_elements_instanced(pipe, indexBuffer, indexSize,
+ mode, start, count, 0, 1);
}
static INLINE boolean
if (nv50->vtxbuf_nr == 0)
return NULL;
- assert(!NV50_USING_LOATHED_EDGEFLAG(nv50));
+ nv50->vbo_fifo = 0;
+ if (nv50->screen->force_push ||
+ nv50->vertprog->cfg.edgeflag_in < 16)
+ nv50->vbo_fifo = 0xffff;
+
+ 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;
+ }
n_ve = MAX2(nv50->vtxelt->num_elements, nv50->state.vtxelt_nr);
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);