Merge remote branch 'origin/master' into gallium_draw_llvm
[mesa.git] / src / gallium / drivers / nv50 / nv50_vbo.c
index 54a2fa730a5a524da8c5c91cb94b32d4b34d33a4..50472868063f5bfa32c983300e8a5f61be2d98d8 100644 (file)
 #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)
 {
@@ -219,6 +187,13 @@ nv50_draw_arrays_instanced(struct pipe_context *pipe,
        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);
@@ -251,85 +226,83 @@ nv50_draw_arrays(struct pipe_context *pipe, unsigned mode, unsigned start,
        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;
+       struct inline_ctx *ctx = priv;
+       struct nouveau_grobj *tesla = ctx->nv50->screen->tesla;
+       struct nouveau_channel *chan = tesla->channel;
 
-       while (count) {
-               unsigned nr = count > 2047 ? 2047 : count;
+       BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U32, count);
+       OUT_RINGp    (chan, (uint32_t *)ctx->map + start, count);
+}
 
-               BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U32, nr);
-               OUT_RINGp (chan, map, nr);
+static void
+inline_edgeflag(void *priv, boolean enabled)
+{
+       struct inline_ctx *ctx = priv;
+       struct nouveau_grobj *tesla = ctx->nv50->screen->tesla;
+       struct nouveau_channel *chan = tesla->channel;
 
-               count -= nr;
-               map += nr;
-       }
-       return TRUE;
+       BEGIN_RING(chan, tesla, NV50TCL_EDGEFLAG_ENABLE, 1);
+       OUT_RING  (chan, enabled ? 1 : 0);
 }
 
 static void
@@ -343,51 +316,68 @@ nv50_draw_elements_inline(struct pipe_context *pipe,
        struct nouveau_channel *chan = nv50->screen->tesla->channel;
        struct nouveau_grobj *tesla = nv50->screen->tesla;
        struct instance a[16];
-       unsigned prim = nv50_prim(mode);
-       void *map;
-
-       map = pipe_buffer_map(pscreen, indexBuffer, PIPE_BUFFER_USAGE_CPU_READ);
-       assert(map);
-       if (!map)
+       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;
 
        instance_init(nv50, a, startInstance);
-       if (!nv50_state_validate(nv50, 0))
+       if (!nv50_state_validate(nv50, overhead + 6 + 3))
                return;
 
        BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 2);
        OUT_RING  (chan, NV50_CB_AUX | (24 << 8));
        OUT_RING  (chan, startInstance);
        while (instanceCount--) {
-               if (AVAIL_RING(chan) < (7 + 16*3)) {
-                       FIRE_RING(chan);
-                       if (!nv50_state_validate(nv50, 0)) {
-                               assert(0);
-                               return;
+               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;
+                               }
                        }
-               }
-               instance_step(nv50, a);
 
-               BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
-               OUT_RING  (chan, prim);
-               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;
-               default:
-                       assert(0);
-                       break;
-               }
-               BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
-               OUT_RING  (chan, 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);
 
-               prim |= (1 << 28);
+               nzi = TRUE;
        }
 
        pipe_buffer_unmap(pscreen, indexBuffer);
@@ -406,17 +396,23 @@ nv50_draw_elements_instanced(struct pipe_context *pipe,
        struct instance a[16];
        unsigned prim = nv50_prim(mode);
 
-       if (indexSize == 1) {
+       instance_init(nv50, a, startInstance);
+       if (!nv50_state_validate(nv50, 13 + 16*3))
+               return;
+
+       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;
        }
 
-       instance_init(nv50, a, startInstance);
-       if (!nv50_state_validate(nv50, 13 + 16*3))
-               return;
-
        BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 2);
        OUT_RING  (chan, NV50_CB_AUX | (24 << 8));
        OUT_RING  (chan, startInstance);
@@ -553,7 +549,16 @@ nv50_vbo_validate(struct nv50_context *nv50)
        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);
 
@@ -573,6 +578,16 @@ nv50_vbo_validate(struct nv50_context *nv50)
                    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);