[Mesa-dev] [PATCH 5/6] r600g: simplify dirty atom tracking

[Grazvydas Ignotas]

Now that R600_NUM_ATOMS is below 64, dirty atom tracking can be
simplified.
---
 src/gallium/drivers/r600/r600_hw_context.c   |  9 +++---
 src/gallium/drivers/r600/r600_pipe.h         | 45 ++++------------------------
 src/gallium/drivers/r600/r600_state_common.c |  9 +++---
 3 files changed, 14 insertions(+), 49 deletions(-)

diff --git a/src/gallium/drivers/r600/r600_hw_context.c b/src/gallium/drivers/r600/r600_hw_context.c
index 092f261..cf71597 100644
--- a/src/gallium/drivers/r600/r600_hw_context.c
+++ b/src/gallium/drivers/r600/r600_hw_context.c
@@ -48,16 +48,15 @@ void r600_need_cs_space(struct r600_context *ctx, unsigned num_dw,
 	num_dw += ctx->b.rings.gfx.cs->cdw;
 
 	if (count_draw_in) {
-		unsigned i;
+		uint64_t mask;
 
 		/* The number of dwords all the dirty states would take. */
-		i = r600_next_dirty_atom(ctx, 0);
-		while (i < R600_NUM_ATOMS) {
-			num_dw += ctx->atoms[i]->num_dw;
+		mask = ctx->dirty_atoms;
+		while (mask != 0) {
+			num_dw += ctx->atoms[u_bit_scan64(&mask)]->num_dw;
 			if (ctx->screen->b.trace_bo) {
 				num_dw += R600_TRACE_CS_DWORDS;
 			}
-			i = r600_next_dirty_atom(ctx, i + 1);
 		}
 
 		/* The upper-bound of how much space a draw command would take. */
diff --git a/src/gallium/drivers/r600/r600_pipe.h b/src/gallium/drivers/r600/r600_pipe.h
index 9860eff..76539d6 100644
--- a/src/gallium/drivers/r600/r600_pipe.h
+++ b/src/gallium/drivers/r600/r600_pipe.h
@@ -87,9 +87,6 @@
 #define R600_BIG_ENDIAN 0
 #endif
 
-#define R600_DIRTY_ATOM_WORD_BITS (sizeof(unsigned long) * 8)
-#define R600_DIRTY_ATOM_ARRAY_LEN DIV_ROUND_UP(R600_NUM_ATOMS, R600_DIRTY_ATOM_WORD_BITS)
-
 struct r600_context;
 struct r600_bytecode;
 union  r600_shader_key;
@@ -438,7 +435,7 @@ struct r600_context {
 	/* State binding slots are here. */
 	struct r600_atom		*atoms[R600_NUM_ATOMS];
 	/* Dirty atom bitmask for fast tests */
-	unsigned long			dirty_atoms[R600_DIRTY_ATOM_ARRAY_LEN];
+	uint64_t			dirty_atoms;
 	/* States for CS initialization. */
 	struct r600_command_buffer	start_cs_cmd; /* invariant state mostly */
 	/** Compute specific registers initializations.  The start_cs_cmd atom
@@ -515,18 +512,17 @@ static inline void r600_set_atom_dirty(struct r600_context *rctx,
 				       struct r600_atom *atom,
 				       bool dirty)
 {
-	unsigned long mask;
-	unsigned int w;
+	uint64_t mask;
 
 	atom->dirty = dirty;
 
 	assert(atom->id != 0);
-	w = atom->id / R600_DIRTY_ATOM_WORD_BITS;
-	mask = 1ul << (atom->id % R600_DIRTY_ATOM_WORD_BITS);
+	assert(atom->id < sizeof(mask) * 8);
+	mask = 1ull << atom->id;
 	if (dirty)
-		rctx->dirty_atoms[w] |= mask;
+		rctx->dirty_atoms |= mask;
 	else
-		rctx->dirty_atoms[w] &= ~mask;
+		rctx->dirty_atoms &= ~mask;
 }
 
 static inline void r600_mark_atom_dirty(struct r600_context *rctx,
@@ -535,35 +531,6 @@ static inline void r600_mark_atom_dirty(struct r600_context *rctx,
 	r600_set_atom_dirty(rctx, atom, true);
 }
 
-static inline unsigned int r600_next_dirty_atom(struct r600_context *rctx,
-						unsigned int id)
-{
-#if !defined(DEBUG) && defined(HAVE___BUILTIN_CTZ)
-	unsigned int w = id / R600_DIRTY_ATOM_WORD_BITS;
-	unsigned int bit = id % R600_DIRTY_ATOM_WORD_BITS;
-	unsigned long bits, mask = (1ul << bit) - 1;
-
-	for (; w < R600_DIRTY_ATOM_ARRAY_LEN; w++, mask = 0ul) {
-		bits = rctx->dirty_atoms[w] & ~mask;
-		if (bits == 0)
-			continue;
-		return w * R600_DIRTY_ATOM_WORD_BITS + __builtin_ctzl(bits);
-	}
-
-	return R600_NUM_ATOMS;
-#else
-	for (; id < R600_NUM_ATOMS; id++) {
-		bool dirty = !!(rctx->dirty_atoms[id / R600_DIRTY_ATOM_WORD_BITS] &
-			(1ul << (id % R600_DIRTY_ATOM_WORD_BITS)));
-		assert(dirty == (rctx->atoms[id] && rctx->atoms[id]->dirty));
-		if (dirty)
-			break;
-	}
-
-	return id;
-#endif
-}
-
 void r600_trace_emit(struct r600_context *rctx);
 
 static inline void r600_emit_atom(struct r600_context *rctx, struct r600_atom *atom)
diff --git a/src/gallium/drivers/r600/r600_state_common.c b/src/gallium/drivers/r600/r600_state_common.c
index 1f96ed6..39c65ae 100644
--- a/src/gallium/drivers/r600/r600_state_common.c
+++ b/src/gallium/drivers/r600/r600_state_common.c
@@ -1427,8 +1427,8 @@ static void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info
 	struct r600_context *rctx = (struct r600_context *)ctx;
 	struct pipe_draw_info info = *dinfo;
 	struct pipe_index_buffer ib = {};
-	unsigned i;
 	struct radeon_winsys_cs *cs = rctx->b.rings.gfx.cs;
+	uint64_t mask;
 
 	if (!info.indirect && !info.count && (info.indexed || !info.count_from_stream_output)) {
 		return;
@@ -1538,10 +1538,9 @@ static void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info
 	r600_need_cs_space(rctx, ib.user_buffer ? 5 : 0, TRUE);
 	r600_flush_emit(rctx);
 
-	i = r600_next_dirty_atom(rctx, 0);
-	while (i < R600_NUM_ATOMS) {
-		r600_emit_atom(rctx, rctx->atoms[i]);
-		i = r600_next_dirty_atom(rctx, i + 1);
+	mask = rctx->dirty_atoms;
+	while (mask != 0) {
+		r600_emit_atom(rctx, rctx->atoms[u_bit_scan64(&mask)]);
 	}
 
 	if (rctx->b.chip_class == CAYMAN) {
-- 
1.9.1