diff --git a/src/gallium/drivers/panfrost/pan_cmdstream.c b/src/gallium/drivers/panfrost/pan_cmdstream.c
index 20639cdee5a..7d7ec775dc0 100644
--- a/src/gallium/drivers/panfrost/pan_cmdstream.c
+++ b/src/gallium/drivers/panfrost/pan_cmdstream.c
@@ -262,6 +262,7 @@ panfrost_vt_set_draw_info(struct panfrost_context *ctx,
}
tiler_prefix->unknown_draw = draw_flags;
+ ctx->offset_start = vertex_postfix->offset_start;
/* Encode the padded vertex count */
@@ -1204,24 +1205,19 @@ panfrost_emit_sampler_descriptors(struct panfrost_batch *batch,
return T.gpu;
}
-void
+mali_ptr
panfrost_emit_vertex_data(struct panfrost_batch *batch,
- struct mali_vertex_tiler_postfix *vertex_postfix)
+ mali_ptr *buffers)
{
struct panfrost_context *ctx = batch->ctx;
struct panfrost_vertex_state *so = ctx->vertex;
struct panfrost_shader_state *vs = panfrost_get_shader_state(ctx, PIPE_SHADER_VERTEX);
- unsigned instance_shift = vertex_postfix->instance_shift;
- unsigned instance_odd = vertex_postfix->instance_odd;
-
/* Worst case: everything is NPOT, which is only possible if instancing
* is enabled. Otherwise single record is gauranteed */
- bool could_npot = instance_shift || instance_odd;
-
struct panfrost_transfer S = panfrost_pool_alloc_aligned(&batch->pool,
MALI_ATTRIBUTE_BUFFER_LENGTH * vs->attribute_count *
- (could_npot ? 2 : 1),
+ (ctx->instance_count > 1 ? 2 : 1),
MALI_ATTRIBUTE_BUFFER_LENGTH * 2);
struct panfrost_transfer T = panfrost_pool_alloc_aligned(&batch->pool,
@@ -1286,14 +1282,14 @@ panfrost_emit_vertex_data(struct panfrost_batch *batch,
if (!divisor || ctx->instance_count <= 1) {
pan_pack(bufs + k, ATTRIBUTE_BUFFER, cfg) {
- if (ctx->instance_count > 1)
+ if (ctx->instance_count > 1) {
cfg.type = MALI_ATTRIBUTE_TYPE_1D_MODULUS;
+ cfg.divisor = ctx->padded_count;
+ }
cfg.pointer = addr;
cfg.stride = stride;
cfg.size = size;
- cfg.divisor_r = instance_shift;
- cfg.divisor_p = instance_odd;
}
} else if (util_is_power_of_two_or_zero(hw_divisor)) {
pan_pack(bufs + k, ATTRIBUTE_BUFFER, cfg) {
@@ -1358,8 +1354,6 @@ panfrost_emit_vertex_data(struct panfrost_batch *batch,
* addressing modes and now base is 64 aligned.
*/
- unsigned start = vertex_postfix->offset_start;
-
for (unsigned i = 0; i < so->num_elements; ++i) {
unsigned vbi = so->pipe[i].vertex_buffer_index;
struct pipe_vertex_buffer *buf = &ctx->vertex_buffers[vbi];
@@ -1376,8 +1370,8 @@ panfrost_emit_vertex_data(struct panfrost_batch *batch,
/* Also, somewhat obscurely per-instance data needs to be
* offset in response to a delayed start in an indexed draw */
- if (so->pipe[i].instance_divisor && ctx->instance_count > 1 && start)
- src_offset -= buf->stride * start;
+ if (so->pipe[i].instance_divisor && ctx->instance_count > 1)
+ src_offset -= buf->stride * ctx->offset_start;
pan_pack(out + i, ATTRIBUTE, cfg) {
cfg.buffer_index = attrib_to_buffer[i];
@@ -1386,8 +1380,8 @@ panfrost_emit_vertex_data(struct panfrost_batch *batch,
}
}
- vertex_postfix->attributes = S.gpu;
- vertex_postfix->attribute_meta = T.gpu;
+ *buffers = S.gpu;
+ return T.gpu;
}
static mali_ptr
diff --git a/src/gallium/drivers/panfrost/pan_cmdstream.h b/src/gallium/drivers/panfrost/pan_cmdstream.h
index 430d47243b1..3603dcc2bc3 100644
--- a/src/gallium/drivers/panfrost/pan_cmdstream.h
+++ b/src/gallium/drivers/panfrost/pan_cmdstream.h
@@ -77,9 +77,9 @@ mali_ptr
panfrost_emit_sampler_descriptors(struct panfrost_batch *batch,
enum pipe_shader_type stage);
-void
+mali_ptr
panfrost_emit_vertex_data(struct panfrost_batch *batch,
- struct mali_vertex_tiler_postfix *vertex_postfix);
+ mali_ptr *buffers);
void
panfrost_emit_varying_descriptor(struct panfrost_batch *batch,
diff --git a/src/gallium/drivers/panfrost/pan_context.c b/src/gallium/drivers/panfrost/pan_context.c
index 215fe594406..9f22c2b698c 100644
--- a/src/gallium/drivers/panfrost/pan_context.c
+++ b/src/gallium/drivers/panfrost/pan_context.c
@@ -319,9 +319,10 @@ panfrost_draw_vbo(
1, 1, 1);
/* Emit all sort of descriptors. */
- mali_ptr push_vert = 0, push_frag = 0;
+ mali_ptr push_vert = 0, push_frag = 0, attribs = 0;
- panfrost_emit_vertex_data(batch, &vertex_postfix);
+ vertex_postfix.attribute_meta = panfrost_emit_vertex_data(batch, &attribs);
+ vertex_postfix.attributes = attribs;
panfrost_emit_varying_descriptor(batch,
ctx->padded_count *
ctx->instance_count,
diff --git a/src/gallium/drivers/panfrost/pan_context.h b/src/gallium/drivers/panfrost/pan_context.h
index b7314c59e5f..eb28e552db0 100644
--- a/src/gallium/drivers/panfrost/pan_context.h
+++ b/src/gallium/drivers/panfrost/pan_context.h
@@ -124,6 +124,7 @@ struct panfrost_context {
unsigned vertex_count;
unsigned instance_count;
+ unsigned offset_start;
enum pipe_prim_type active_prim;
/* If instancing is enabled, vertex count padded for instance; if
diff --git a/src/panfrost/lib/midgard.xml b/src/panfrost/lib/midgard.xml
index a6c96b35d73..af209576a66 100644
--- a/src/panfrost/lib/midgard.xml
+++ b/src/panfrost/lib/midgard.xml
@@ -257,6 +257,7 @@
+