freedreno/a6xx: single cmdstream for draw+binning

Now that state which is different for draw vs binning pass is split out
into different state-groups with appropriate enable_mask (so the
appropriate one is chosen for draw vs binning), switch over to using a
single cmdstream for both passes.

This should significantly lower draw overhead for CPU bound benchmarks.

Signed-off-by: Rob Clark <robdclark@gmail.com>
This commit is contained in:
Rob Clark 2018-10-13 13:56:05 -04:00
parent 72f6164fef
commit 24d57a6d8f
3 changed files with 3 additions and 15 deletions

View File

@ -229,14 +229,13 @@ fd6_draw_vbo(struct fd_context *ctx, const struct pipe_draw_info *info,
fixup_shader_state(ctx, &emit.key.key);
unsigned dirty = ctx->dirty;
if (!(dirty & FD_DIRTY_PROG)) {
if (!(ctx->dirty & FD_DIRTY_PROG)) {
emit.prog = fd6_ctx->prog;
} else {
fd6_ctx->prog = fd6_emit_get_prog(&emit);
}
emit.dirty = ctx->dirty; /* *after* fixup_shader_state() */
emit.bs = fd6_emit_get_prog(&emit)->bs;
emit.vs = fd6_emit_get_prog(&emit)->vs;
emit.fs = fd6_emit_get_prog(&emit)->fs;
@ -257,18 +256,8 @@ fd6_draw_vbo(struct fd_context *ctx, const struct pipe_draw_info *info,
*/
emit.no_lrz_write = fp->writes_pos || fp->has_kill;
emit.binning_pass = false;
emit.dirty = dirty;
draw_impl(ctx, ctx->batch->draw, &emit, index_offset);
/* and now binning pass: */
emit.binning_pass = true;
emit.dirty = dirty & ~(FD_DIRTY_BLEND);
emit.vs = fd6_emit_get_prog(&emit)->bs;
draw_impl(ctx, ctx->batch->binning, &emit, index_offset);
if (emit.streamout_mask) {
struct fd_ringbuffer *ring = ctx->batch->draw;

View File

@ -66,7 +66,6 @@ struct fd6_emit {
struct fd_context *ctx;
const struct fd_vertex_state *vtx;
const struct pipe_draw_info *info;
bool binning_pass;
struct ir3_cache_key key;
enum fd_dirty_3d_state dirty;

View File

@ -408,7 +408,7 @@ emit_binning_pass(struct fd_batch *batch)
A6XX_SP_TP_WINDOW_OFFSET_Y(0));
/* emit IB to binning drawcmds: */
ctx->emit_ib(ring, batch->binning);
ctx->emit_ib(ring, batch->draw);
fd_reset_wfi(batch);