[Mesa-dev] [PATCH v2] nv50, nvc0: fix start_instance in manual push path

[Ilia Mirkin]

The start instance is applied as an offset into the buffer directly,
ignoring the divisor, not as an instance id offset that respects the
divisor.

Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu>
Cc: "11.2 12.0" <mesa-stable at lists.freedesktop.org>
---

v1 -> v2: use translate's start_instance parameter.

This relies on the earlier fix I just sent for start_instance processing in translate_sse.

 src/gallium/drivers/nouveau/nv50/nv50_push.c          | 16 +++++++++++-----
 src/gallium/drivers/nouveau/nvc0/nvc0_vbo_translate.c | 16 +++++++++++-----
 2 files changed, 22 insertions(+), 10 deletions(-)

diff --git a/src/gallium/drivers/nouveau/nv50/nv50_push.c b/src/gallium/drivers/nouveau/nv50/nv50_push.c
index cbef95d..6a53ad0 100644
--- a/src/gallium/drivers/nouveau/nv50/nv50_push.c
+++ b/src/gallium/drivers/nouveau/nv50/nv50_push.c
@@ -30,6 +30,7 @@ struct push_context {
 
    uint32_t prim;
    uint32_t restart_index;
+   uint32_t start_instance;
    uint32_t instance_id;
 };
 
@@ -85,7 +86,8 @@ emit_vertices_i08(struct push_context *ctx, unsigned start, unsigned count)
 
       BEGIN_NI04(ctx->push, NV50_3D(VERTEX_DATA), size);
 
-      ctx->translate->run_elts8(ctx->translate, elts, nr, 0, ctx->instance_id,
+      ctx->translate->run_elts8(ctx->translate, elts, nr,
+                                ctx->start_instance, ctx->instance_id,
                                 ctx->push->cur);
 
       ctx->push->cur += size;
@@ -123,7 +125,8 @@ emit_vertices_i16(struct push_context *ctx, unsigned start, unsigned count)
 
       BEGIN_NI04(ctx->push, NV50_3D(VERTEX_DATA), size);
 
-      ctx->translate->run_elts16(ctx->translate, elts, nr, 0, ctx->instance_id,
+      ctx->translate->run_elts16(ctx->translate, elts, nr,
+                                 ctx->start_instance, ctx->instance_id,
                                  ctx->push->cur);
 
       ctx->push->cur += size;
@@ -161,7 +164,8 @@ emit_vertices_i32(struct push_context *ctx, unsigned start, unsigned count)
 
       BEGIN_NI04(ctx->push, NV50_3D(VERTEX_DATA), size);
 
-      ctx->translate->run_elts(ctx->translate, elts, nr, 0, ctx->instance_id,
+      ctx->translate->run_elts(ctx->translate, elts, nr,
+                               ctx->start_instance, ctx->instance_id,
                                ctx->push->cur);
 
       ctx->push->cur += size;
@@ -194,7 +198,8 @@ emit_vertices_seq(struct push_context *ctx, unsigned start, unsigned count)
 
       BEGIN_NI04(ctx->push, NV50_3D(VERTEX_DATA), size);
 
-      ctx->translate->run(ctx->translate, start, push, 0, ctx->instance_id,
+      ctx->translate->run(ctx->translate, start, push,
+                          ctx->start_instance, ctx->instance_id,
                           ctx->push->cur);
       ctx->push->cur += size;
       count -= push;
@@ -247,6 +252,7 @@ nv50_push_vbo(struct nv50_context *nv50, const struct pipe_draw_info *info)
    ctx.need_vertex_id = nv50->screen->base.class_3d >= NV84_3D_CLASS &&
       nv50->vertprog->vp.need_vertex_id && (nv50->vertex->num_elements < 32);
    ctx.index_bias = info->index_bias;
+   ctx.instance_id = 0;
 
    /* For indexed draws, gl_VertexID must be emitted for every vertex. */
    ctx.packet_vertex_limit =
@@ -301,7 +307,7 @@ nv50_push_vbo(struct nv50_context *nv50, const struct pipe_draw_info *info)
       ctx.restart_index = 0;
    }
 
-   ctx.instance_id = info->start_instance;
+   ctx.start_instance = info->start_instance;
    ctx.prim = nv50_prim_gl(info->mode);
 
    if (info->primitive_restart) {
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_vbo_translate.c b/src/gallium/drivers/nouveau/nvc0/nvc0_vbo_translate.c
index 20b6742..fd2bcbb 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_vbo_translate.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_vbo_translate.c
@@ -19,6 +19,7 @@ struct push_context {
 
    uint32_t vertex_size;
    uint32_t restart_index;
+   uint32_t start_instance;
    uint32_t instance_id;
 
    bool prim_restart;
@@ -44,6 +45,7 @@ nvc0_push_context_init(struct nvc0_context *nvc0, struct push_context *ctx)
 
    ctx->translate = nvc0->vertex->translate;
    ctx->vertex_size = nvc0->vertex->size;
+   ctx->instance_id = 0;
 
    ctx->need_vertex_id =
       nvc0->vertprog->vp.need_vertex_id && (nvc0->vertex->num_elements < 32);
@@ -246,7 +248,8 @@ disp_vertices_i08(struct push_context *ctx, unsigned start, unsigned count)
       if (unlikely(ctx->prim_restart))
          nR = prim_restart_search_i08(elts, nR, ctx->restart_index);
 
-      translate->run_elts8(translate, elts, nR, 0, ctx->instance_id, ctx->dest);
+      translate->run_elts8(translate, elts, nR,
+                           ctx->start_instance, ctx->instance_id, ctx->dest);
       count -= nR;
       ctx->dest += nR * ctx->vertex_size;
 
@@ -302,7 +305,8 @@ disp_vertices_i16(struct push_context *ctx, unsigned start, unsigned count)
       if (unlikely(ctx->prim_restart))
          nR = prim_restart_search_i16(elts, nR, ctx->restart_index);
 
-      translate->run_elts16(translate, elts, nR, 0, ctx->instance_id, ctx->dest);
+      translate->run_elts16(translate, elts, nR,
+                            ctx->start_instance, ctx->instance_id, ctx->dest);
       count -= nR;
       ctx->dest += nR * ctx->vertex_size;
 
@@ -358,7 +362,8 @@ disp_vertices_i32(struct push_context *ctx, unsigned start, unsigned count)
       if (unlikely(ctx->prim_restart))
          nR = prim_restart_search_i32(elts, nR, ctx->restart_index);
 
-      translate->run_elts(translate, elts, nR, 0, ctx->instance_id, ctx->dest);
+      translate->run_elts(translate, elts, nR,
+                          ctx->start_instance, ctx->instance_id, ctx->dest);
       count -= nR;
       ctx->dest += nR * ctx->vertex_size;
 
@@ -410,7 +415,8 @@ disp_vertices_seq(struct push_context *ctx, unsigned start, unsigned count)
    /* XXX: This will read the data corresponding to the primitive restart index,
     *  maybe we should avoid that ?
     */
-   translate->run(translate, start, count, 0, ctx->instance_id, ctx->dest);
+   translate->run(translate, start, count,
+                  ctx->start_instance, ctx->instance_id, ctx->dest);
    do {
       unsigned nr = count;
 
@@ -515,7 +521,7 @@ nvc0_push_vbo(struct nvc0_context *nvc0, const struct pipe_draw_info *info)
       index_size = 0;
    }
 
-   ctx.instance_id = info->start_instance;
+   ctx.start_instance = info->start_instance;
 
    prim = nvc0_prim_gl(info->mode);
    do {
-- 
2.7.3