summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--src/gallium/drivers/freedreno/a3xx/fd3_draw.c6
-rw-r--r--src/gallium/drivers/freedreno/a3xx/fd3_emit.c5
-rw-r--r--src/gallium/drivers/freedreno/a3xx/fd3_emit.h29
-rw-r--r--src/gallium/drivers/freedreno/a3xx/fd3_gmem.c22
-rw-r--r--src/gallium/drivers/freedreno/freedreno_context.c1
-rw-r--r--src/gallium/drivers/freedreno/freedreno_context.h32
-rw-r--r--src/gallium/drivers/freedreno/freedreno_draw.h2
-rw-r--r--src/gallium/drivers/freedreno/freedreno_gmem.c2
8 files changed, 66 insertions, 33 deletions
diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_draw.c b/src/gallium/drivers/freedreno/a3xx/fd3_draw.c
index d808777a1fb..c5d8b774552 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_draw.c
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_draw.c
@@ -80,8 +80,6 @@ fd3_draw(struct fd_context *ctx, const struct pipe_draw_info *info)
OUT_PKT0(ring, REG_A3XX_PC_VERTEX_REUSE_BLOCK_CNTL, 1);
OUT_RING(ring, 0x0000000b); /* PC_VERTEX_REUSE_BLOCK_CNTL */
- OUT_WFI (ring);
-
OUT_PKT0(ring, REG_A3XX_VFD_INDEX_MIN, 4);
OUT_RING(ring, info->min_index); /* VFD_INDEX_MIN */
OUT_RING(ring, info->max_index); /* VFD_INDEX_MAX */
@@ -111,7 +109,7 @@ fd3_clear(struct fd_context *ctx, unsigned buffers,
OUT_RING(ring, A3XX_RB_BLEND_ALPHA_UINT(0xff) |
A3XX_RB_BLEND_ALPHA_FLOAT(1.0));
- fd3_emit_rbrc_draw_state(ring,
+ fd3_emit_rbrc_draw_state(ctx, ring,
A3XX_RB_RENDER_CONTROL_ALPHA_TEST_FUNC(FUNC_NEVER));
if (buffers & PIPE_CLEAR_DEPTH) {
@@ -220,8 +218,6 @@ fd3_clear(struct fd_context *ctx, unsigned buffers,
fd_draw(ctx, DI_PT_RECTLIST, DI_SRC_SEL_AUTO_INDEX, 2,
INDEX_SIZE_IGN, 0, 0, NULL);
-
- OUT_WFI (ring);
}
void
diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_emit.c b/src/gallium/drivers/freedreno/a3xx/fd3_emit.c
index 825656ae62b..91993725ea6 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_emit.c
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_emit.c
@@ -243,8 +243,6 @@ emit_cache_flush(struct fd_ringbuffer *ring)
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
-
- OUT_WFI (ring);
}
/* emit texture state for mem->gmem restore operation.. eventually it would
@@ -356,7 +354,7 @@ fd3_emit_state(struct fd_context *ctx, uint32_t dirty)
struct fd3_zsa_stateobj *zsa = fd3_zsa_stateobj(ctx->zsa);
struct pipe_stencil_ref *sr = &ctx->stencil_ref;
- fd3_emit_rbrc_draw_state(ring, zsa->rb_render_control);
+ fd3_emit_rbrc_draw_state(ctx, ring, zsa->rb_render_control);
OUT_PKT0(ring, REG_A3XX_RB_ALPHA_REF, 1);
OUT_RING(ring, zsa->rb_alpha_ref);
@@ -607,4 +605,5 @@ fd3_emit_restore(struct fd_context *ctx)
}
emit_cache_flush(ring);
+ fd_rmw_wfi(ctx, ring);
}
diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_emit.h b/src/gallium/drivers/freedreno/a3xx/fd3_emit.h
index 668e5ddd095..bf7787ab6f7 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_emit.h
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_emit.h
@@ -66,24 +66,23 @@ void fd3_emit_restore(struct fd_context *ctx);
* GMEM/binning code is deciding on the bin-width (and whether to
* use binning) after the draw/clear state is emitted.
*/
-static inline void
-fd3_emit_rbrc_draw_state(struct fd_ringbuffer *ring, uint32_t val)
-{
- OUT_PKT3(ring, CP_REG_RMW, 3);
- OUT_RING(ring, REG_A3XX_RB_RENDER_CONTROL);
- OUT_RING(ring, A3XX_RB_RENDER_CONTROL_BIN_WIDTH__MASK |
- A3XX_RB_RENDER_CONTROL_ENABLE_GMEM);
- OUT_RING(ring, val);
-}
+
+#define RBRC_DRAW_STATE (A3XX_RB_RENDER_CONTROL_ALPHA_TEST | \
+ A3XX_RB_RENDER_CONTROL_ALPHA_TEST_FUNC__MASK)
static inline void
-fd3_emit_rbrc_tile_state(struct fd_ringbuffer *ring, uint32_t val)
+fd3_emit_rbrc_draw_state(struct fd_context *ctx,
+ struct fd_ringbuffer *ring, uint32_t val)
{
- OUT_PKT3(ring, CP_REG_RMW, 3);
- OUT_RING(ring, REG_A3XX_RB_RENDER_CONTROL);
- OUT_RING(ring, ~(A3XX_RB_RENDER_CONTROL_BIN_WIDTH__MASK |
- A3XX_RB_RENDER_CONTROL_ENABLE_GMEM));
- OUT_RING(ring, val);
+ assert(!(val & ~RBRC_DRAW_STATE));
+ if (val != ctx->rmw.rbrc_draw) {
+ fd_rmw_wfi(ctx, ring);
+ OUT_PKT3(ring, CP_REG_RMW, 3);
+ OUT_RING(ring, REG_A3XX_RB_RENDER_CONTROL);
+ OUT_RING(ring, ~RBRC_DRAW_STATE);
+ OUT_RING(ring, val);
+ ctx->rmw.rbrc_draw = val;
+ }
}
#endif /* FD3_EMIT_H */
diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_gmem.c b/src/gallium/drivers/freedreno/a3xx/fd3_gmem.c
index 4a5a241bac9..3d0a607ed28 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_gmem.c
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_gmem.c
@@ -181,9 +181,11 @@ fd3_emit_tile_gmem2mem(struct fd_context *ctx, struct fd_tile *tile)
OUT_RING(ring, A3XX_RB_MODE_CONTROL_RENDER_MODE(RB_RESOLVE_PASS) |
A3XX_RB_MODE_CONTROL_MARB_CACHE_SPLIT_MODE);
- fd3_emit_rbrc_draw_state(ring,
- A3XX_RB_RENDER_CONTROL_DISABLE_COLOR_PIPE |
- A3XX_RB_RENDER_CONTROL_ALPHA_TEST_FUNC(FUNC_NEVER));
+ OUT_PKT0(ring, REG_A3XX_RB_RENDER_CONTROL, 1);
+ OUT_RING(ring, A3XX_RB_RENDER_CONTROL_DISABLE_COLOR_PIPE |
+ A3XX_RB_RENDER_CONTROL_ENABLE_GMEM |
+ A3XX_RB_RENDER_CONTROL_ALPHA_TEST_FUNC(FUNC_NEVER) |
+ A3XX_RB_RENDER_CONTROL_BIN_WIDTH(ctx->gmem.bin_w));
OUT_PKT0(ring, REG_A3XX_GRAS_SC_CONTROL, 1);
OUT_RING(ring, A3XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RESOLVE_PASS) |
@@ -295,7 +297,8 @@ fd3_emit_tile_mem2gmem(struct fd_context *ctx, struct fd_tile *tile)
A3XX_RB_MRT_BLEND_CONTROL_CLAMP_ENABLE);
}
- fd3_emit_rbrc_tile_state(ring,
+ OUT_PKT0(ring, REG_A3XX_RB_RENDER_CONTROL, 1);
+ OUT_RING(ring, A3XX_RB_RENDER_CONTROL_ALPHA_TEST_FUNC(FUNC_ALWAYS) |
A3XX_RB_RENDER_CONTROL_BIN_WIDTH(gmem->bin_w));
OUT_PKT0(ring, REG_A3XX_RB_DEPTH_CONTROL, 1);
@@ -335,9 +338,6 @@ fd3_emit_tile_mem2gmem(struct fd_context *ctx, struct fd_tile *tile)
A3XX_RB_STENCIL_CONTROL_ZPASS_BF(STENCIL_KEEP) |
A3XX_RB_STENCIL_CONTROL_ZFAIL_BF(STENCIL_KEEP));
- fd3_emit_rbrc_draw_state(ring,
- A3XX_RB_RENDER_CONTROL_ALPHA_TEST_FUNC(FUNC_ALWAYS));
-
OUT_PKT0(ring, REG_A3XX_GRAS_SC_CONTROL, 1);
OUT_RING(ring, A3XX_GRAS_SC_CONTROL_RENDER_MODE(RB_RENDERING_PASS) |
A3XX_GRAS_SC_CONTROL_MSAA_SAMPLES(MSAA_ONE) |
@@ -423,7 +423,8 @@ fd3_emit_sysmem_prep(struct fd_context *ctx)
emit_mrt(ring, pfb->nr_cbufs, pfb->cbufs, NULL, 0);
- fd3_emit_rbrc_tile_state(ring,
+ OUT_PKT0(ring, REG_A3XX_RB_RENDER_CONTROL, 1);
+ OUT_RING(ring, A3XX_RB_RENDER_CONTROL_ALPHA_TEST_FUNC(FUNC_NEVER) |
A3XX_RB_RENDER_CONTROL_BIN_WIDTH(pitch));
/* setup scissor/offset for current tile: */
@@ -514,8 +515,9 @@ fd3_emit_tile_renderprep(struct fd_context *ctx, struct fd_tile *tile)
emit_mrt(ring, pfb->nr_cbufs, pfb->cbufs, NULL, gmem->bin_w);
- fd3_emit_rbrc_tile_state(ring,
- A3XX_RB_RENDER_CONTROL_ENABLE_GMEM |
+ OUT_PKT0(ring, REG_A3XX_RB_RENDER_CONTROL, 1);
+ OUT_RING(ring, A3XX_RB_RENDER_CONTROL_ENABLE_GMEM |
+ A3XX_RB_RENDER_CONTROL_ALPHA_TEST_FUNC(FUNC_NEVER) |
A3XX_RB_RENDER_CONTROL_BIN_WIDTH(gmem->bin_w));
/* setup scissor/offset for current tile: */
diff --git a/src/gallium/drivers/freedreno/freedreno_context.c b/src/gallium/drivers/freedreno/freedreno_context.c
index c959ccfcc89..28be508e329 100644
--- a/src/gallium/drivers/freedreno/freedreno_context.c
+++ b/src/gallium/drivers/freedreno/freedreno_context.c
@@ -174,6 +174,7 @@ fd_context_init(struct fd_context *ctx, struct pipe_screen *pscreen,
}
fd_context_next_rb(pctx);
+ fd_reset_rmw_state(ctx);
util_slab_create(&ctx->transfer_pool, sizeof(struct pipe_transfer),
16, UTIL_SLAB_SINGLETHREADED);
diff --git a/src/gallium/drivers/freedreno/freedreno_context.h b/src/gallium/drivers/freedreno/freedreno_context.h
index db37f9cb8a3..a8abbca7a62 100644
--- a/src/gallium/drivers/freedreno/freedreno_context.h
+++ b/src/gallium/drivers/freedreno/freedreno_context.h
@@ -38,6 +38,7 @@
#include "freedreno_screen.h"
#include "freedreno_gmem.h"
+#include "freedreno_util.h"
struct fd_vertex_stateobj;
@@ -150,6 +151,20 @@ struct fd_context {
struct fd_ringbuffer *ring;
struct fd_ringmarker *draw_start, *draw_end;
+ /* Keep track if WAIT_FOR_IDLE is needed for registers we need
+ * to update via RMW:
+ */
+ struct {
+ bool need_wfi;
+ /* note: would be nicer to have in fd3_context, fd2_context,
+ * etc, because the registered modified via RMR differ across
+ * generation. But as long as it is a small set of registers
+ * that might be more hassle than it's worth.
+ */
+ /* state for RB_RENDER_CONTROL: */
+ uint32_t rbrc_draw;
+ } rmw;
+
struct pipe_scissor_state scissor;
/* we don't have a disable/enable bit for scissor, so instead we keep
@@ -249,6 +264,23 @@ fd_supported_prim(struct fd_context *ctx, unsigned prim)
return (1 << prim) & ctx->primtype_mask;
}
+static INLINE void
+fd_reset_rmw_state(struct fd_context *ctx)
+{
+ ctx->rmw.need_wfi = true;
+ ctx->rmw.rbrc_draw = ~0;
+}
+
+/* emit before a RMW a WAIT_FOR_IDLE only if needed: */
+static inline void
+fd_rmw_wfi(struct fd_context *ctx, struct fd_ringbuffer *ring)
+{
+ if (ctx->rmw.need_wfi) {
+ OUT_WFI(ring);
+ ctx->rmw.need_wfi = false;
+ }
+}
+
struct pipe_context * fd_context_init(struct fd_context *ctx,
struct pipe_screen *pscreen, const uint8_t *primtypes,
void *priv);
diff --git a/src/gallium/drivers/freedreno/freedreno_draw.h b/src/gallium/drivers/freedreno/freedreno_draw.h
index 9ccc246395f..190c0e52d24 100644
--- a/src/gallium/drivers/freedreno/freedreno_draw.h
+++ b/src/gallium/drivers/freedreno/freedreno_draw.h
@@ -85,6 +85,8 @@ fd_draw(struct fd_context *ctx, enum pc_di_primtype primtype,
}
emit_marker(ring, 7);
+
+ ctx->rmw.need_wfi = true;
}
#endif /* FREEDRENO_DRAW_H_ */
diff --git a/src/gallium/drivers/freedreno/freedreno_gmem.c b/src/gallium/drivers/freedreno/freedreno_gmem.c
index 68b4eb607a6..47d99d3429c 100644
--- a/src/gallium/drivers/freedreno/freedreno_gmem.c
+++ b/src/gallium/drivers/freedreno/freedreno_gmem.c
@@ -272,6 +272,8 @@ fd_gmem_render_tiles(struct pipe_context *pctx)
/* mark start for next draw cmds: */
fd_ringmarker_mark(ctx->draw_start);
+ fd_reset_rmw_state(ctx);
+
/* update timestamps on render targets: */
timestamp = fd_ringbuffer_timestamp(ctx->ring);
if (pfb->cbufs[0])
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-18 07:45:05 +0000