Mesa (master): dri/nouveau: Optimize VBO binding re-emission.
Francisco Jerez
currojerez at kemper.freedesktop.org
Sun Oct 31 01:08:47 UTC 2010
Module: Mesa
Branch: master
Commit: e89af209261e51988b99d954d09f2cbc59e55358
URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=e89af209261e51988b99d954d09f2cbc59e55358
Author: Francisco Jerez <currojerez at riseup.net>
Date: Fri Oct 29 21:36:42 2010 +0200
dri/nouveau: Optimize VBO binding re-emission.
---
src/mesa/drivers/dri/nouveau/nouveau_vbo_t.c | 94 +++++++++++++++++++++-----
src/mesa/drivers/dri/nouveau/nv10_render.c | 8 +-
src/mesa/drivers/dri/nouveau/nv20_render.c | 8 +-
3 files changed, 84 insertions(+), 26 deletions(-)
diff --git a/src/mesa/drivers/dri/nouveau/nouveau_vbo_t.c b/src/mesa/drivers/dri/nouveau/nouveau_vbo_t.c
index c00bd31..bf5885f 100644
--- a/src/mesa/drivers/dri/nouveau/nouveau_vbo_t.c
+++ b/src/mesa/drivers/dri/nouveau/nouveau_vbo_t.c
@@ -67,7 +67,7 @@ vbo_init_arrays(struct gl_context *ctx, const struct _mesa_index_buffer *ib,
nouveau_init_array(&render->attrs[attr], attr,
get_array_stride(ctx, array),
array->Size, array->Type,
- array->BufferObj,
+ imm ? array->BufferObj : NULL,
array->Ptr, imm);
}
}
@@ -85,12 +85,14 @@ vbo_deinit_arrays(struct gl_context *ctx, const struct _mesa_index_buffer *ib,
FOR_EACH_BOUND_ATTR(render, i, attr) {
struct nouveau_array *a = &render->attrs[attr];
+ if (render->mode == IMM)
+ nouveau_bo_ref(NULL, &a->bo);
+
nouveau_deinit_array(a);
render->map[i] = -1;
}
render->attr_count = 0;
- context_bctx(ctx, VERTEX);
}
/* Make some rendering decisions from the GL context. */
@@ -180,7 +182,7 @@ vbo_choose_attrs(struct gl_context *ctx, const struct gl_client_array **arrays)
(ctx->Texture._GenFlags & TEXGEN_NEED_NORMALS))
vbo_emit_attr(ctx, arrays, VERT_ATTRIB_NORMAL);
- if (ctx->Light.Enabled) {
+ if (ctx->Light.Enabled && render->mode == IMM) {
vbo_emit_attr(ctx, arrays, MAT(FRONT_AMBIENT));
vbo_emit_attr(ctx, arrays, MAT(FRONT_DIFFUSE));
vbo_emit_attr(ctx, arrays, MAT(FRONT_SPECULAR));
@@ -258,38 +260,94 @@ vbo_maybe_split(struct gl_context *ctx, const struct gl_client_array **arrays,
/* VBO rendering path. */
+static GLboolean
+check_update_array(struct nouveau_array *a, unsigned offset,
+ struct nouveau_bo *bo, int *pdelta)
+{
+ int delta = *pdelta;
+ GLboolean dirty;
+
+ if (a->bo == bo) {
+ if (delta < 0)
+ delta = (offset - (long)a->offset) / a->stride;
+
+ dirty = (delta < 0 ||
+ offset != (a->offset + delta * a->stride));
+ } else {
+ dirty = GL_TRUE;
+ }
+
+ *pdelta = (dirty ? 0 : delta);
+ return dirty;
+}
+
static void
vbo_bind_vertices(struct gl_context *ctx, const struct gl_client_array **arrays,
- GLint basevertex, GLuint min_index, GLuint max_index)
+ int base, unsigned min_index, unsigned max_index, int *pdelta)
{
struct nouveau_render_state *render = to_render_state(ctx);
- int i, attr;
+ struct nouveau_channel *chan = context_chan(ctx);
+ struct nouveau_bo *bo[NUM_VERTEX_ATTRS];
+ unsigned offset[NUM_VERTEX_ATTRS];
+ GLboolean dirty = GL_FALSE;
+ int i, j, attr;
+ RENDER_LOCALS(ctx);
+
+ *pdelta = -1;
FOR_EACH_BOUND_ATTR(render, i, attr) {
const struct gl_client_array *array = arrays[attr];
struct nouveau_array *a = &render->attrs[attr];
- unsigned delta = (basevertex + min_index)
- * array->StrideB;
+ unsigned delta = (base + min_index) * array->StrideB;
+
+ bo[i] = NULL;
+
+ if (_mesa_is_bufferobj(array->BufferObj)) {
+ struct nouveau_bufferobj *nbo =
+ to_nouveau_bufferobj(array->BufferObj);
- if (a->bo) {
/* Array in a buffer obj. */
- a->offset = (intptr_t)array->Ptr + delta;
+ nouveau_bo_ref(nbo->bo, &bo[i]);
+ offset[i] = delta + (intptr_t)array->Ptr;
+
} else {
- int j, n = max_index - min_index + 1;
+ int n = max_index - min_index + 1;
char *sp = (char *)array->Ptr + delta;
- char *dp = nouveau_get_scratch(
- ctx, n * a->stride, &a->bo, &a->offset);
+ char *dp = nouveau_get_scratch(ctx, n * a->stride,
+ &bo[i], &offset[i]);
- /* Array in client memory, move it to
- * a scratch buffer obj. */
+ /* Array in client memory, move it to a
+ * scratch buffer obj. */
for (j = 0; j < n; j++)
memcpy(dp + j * a->stride,
sp + j * array->StrideB,
a->stride);
}
+
+ dirty |= check_update_array(a, offset[i], bo[i], pdelta);
+ }
+
+ *pdelta -= min_index;
+
+ if (dirty) {
+ /* Buffers changed, update the attribute binding. */
+ FOR_EACH_BOUND_ATTR(render, i, attr) {
+ struct nouveau_array *a = &render->attrs[attr];
+
+ nouveau_bo_ref(NULL, &a->bo);
+ a->offset = offset[i];
+ a->bo = bo[i];
+ }
+
+ TAG(render_bind_vertices)(ctx);
+
+ } else {
+ /* Just cleanup. */
+ FOR_EACH_BOUND_ATTR(render, i, attr)
+ nouveau_bo_ref(NULL, &bo[i]);
}
- TAG(render_bind_vertices)(ctx);
+ BATCH_VALIDATE();
}
static void
@@ -300,7 +358,7 @@ vbo_draw_vbo(struct gl_context *ctx, const struct gl_client_array **arrays,
{
struct nouveau_channel *chan = context_chan(ctx);
dispatch_t dispatch = get_array_dispatch(&to_render_state(ctx)->ib);
- int i, delta = -min_index, basevertex = 0;
+ int i, delta = 0, basevertex = 0;
RENDER_LOCALS(ctx);
TAG(render_set_format)(ctx);
@@ -311,8 +369,8 @@ vbo_draw_vbo(struct gl_context *ctx, const struct gl_client_array **arrays,
if (i == 0 || basevertex != prims[i].basevertex) {
basevertex = prims[i].basevertex;
- vbo_bind_vertices(ctx, arrays, basevertex,
- min_index, max_index);
+ vbo_bind_vertices(ctx, arrays, basevertex, min_index,
+ max_index, &delta);
}
if (count > get_max_vertices(ctx, ib, AVAIL_RING(chan)))
diff --git a/src/mesa/drivers/dri/nouveau/nv10_render.c b/src/mesa/drivers/dri/nouveau/nv10_render.c
index 4a396f8..94ff3ed 100644
--- a/src/mesa/drivers/dri/nouveau/nv10_render.c
+++ b/src/mesa/drivers/dri/nouveau/nv10_render.c
@@ -138,7 +138,6 @@ nv10_render_bind_vertices(struct gl_context *ctx)
{
struct nouveau_render_state *render = to_render_state(ctx);
struct nouveau_bo_context *bctx = context_bctx(ctx, VERTEX);
- struct nouveau_channel *chan = context_chan(ctx);
struct nouveau_grobj *celsius = context_eng3d(ctx);
int i, attr;
@@ -150,15 +149,16 @@ nv10_render_bind_vertices(struct gl_context *ctx)
a->bo, a->offset,
NOUVEAU_BO_GART | NOUVEAU_BO_RD);
}
-
- BEGIN_RING(chan, celsius, NV10TCL_VERTEX_ARRAY_VALIDATE, 1);
- OUT_RING(chan, 0);
}
/* Vertex array rendering defs. */
#define RENDER_LOCALS(ctx) \
struct nouveau_grobj *celsius = context_eng3d(ctx)
+#define BATCH_VALIDATE() \
+ BEGIN_RING(chan, celsius, NV10TCL_VERTEX_ARRAY_VALIDATE, 1); \
+ OUT_RING(chan, 0)
+
#define BATCH_BEGIN(prim) \
BEGIN_RING(chan, celsius, NV10TCL_VERTEX_BUFFER_BEGIN_END, 1); \
OUT_RING(chan, prim)
diff --git a/src/mesa/drivers/dri/nouveau/nv20_render.c b/src/mesa/drivers/dri/nouveau/nv20_render.c
index 44625ab..19fc1e8 100644
--- a/src/mesa/drivers/dri/nouveau/nv20_render.c
+++ b/src/mesa/drivers/dri/nouveau/nv20_render.c
@@ -160,7 +160,6 @@ nv20_render_bind_vertices(struct gl_context *ctx)
{
struct nouveau_render_state *render = to_render_state(ctx);
struct nouveau_bo_context *bctx = context_bctx(ctx, VERTEX);
- struct nouveau_channel *chan = context_chan(ctx);
struct nouveau_grobj *kelvin = context_eng3d(ctx);
int i, attr;
@@ -174,15 +173,16 @@ nv20_render_bind_vertices(struct gl_context *ctx)
NOUVEAU_BO_LOW | NOUVEAU_BO_OR |
NOUVEAU_BO_GART | NOUVEAU_BO_RD);
}
-
- BEGIN_RING(chan, kelvin, NV20TCL_VTX_CACHE_INVALIDATE, 1);
- OUT_RING(chan, 0);
}
/* Vertex array rendering defs. */
#define RENDER_LOCALS(ctx) \
struct nouveau_grobj *kelvin = context_eng3d(ctx)
+#define BATCH_VALIDATE() \
+ BEGIN_RING(chan, kelvin, NV20TCL_VTX_CACHE_INVALIDATE, 1); \
+ OUT_RING(chan, 0)
+
#define BATCH_BEGIN(prim) \
BEGIN_RING(chan, kelvin, NV20TCL_VERTEX_BEGIN_END, 1); \
OUT_RING(chan, prim)
More information about the mesa-commit
mailing list