radeonsi: Create CE IB.

Based on work by Marek Olšák.

v2: Add preamble IB.

Leaves the load packet in the space calculation as the
radeon winsys might not be able to support a premable.

The added space calculation may look expensive, but
is converted to a constant with (at least) -O2 and -O3.

v3: - Fix code style.
    - Remove needed space for vertex buffer descriptors.
    - Fail when the preamble cannot be created.

Signed-off-by: Bas Nieuwenhuizen <[email protected]>
Reviewed-by: Marek Olšák <[email protected]>

3 files changed, 52 insertions, 1 deletions

diff --git a/src/gallium/drivers/radeonsi/si_hw_context.c b/src/gallium/drivers/radeonsi/si_hw_context.c
index b621b55abd3..5294898aa7f 100644
--- a/src/gallium/drivers/radeonsi/si_hw_context.c
+++ b/src/gallium/drivers/radeonsi/si_hw_context.c
@@ -26,10 +26,41 @@
 
 #include "si_pipe.h"
 
+static unsigned si_descriptor_list_cs_space(unsigned count, unsigned element_size)
+{
+	/* Ensure we have enough space to start a new range in a hole */
+	assert(element_size >= 3);
+
+	/* 5 dwords for possible load to reinitialize when we have no preamble
+	 * IB + 5 dwords for write to L2 + 3 bytes for every range written to
+	 * CE RAM.
+	 */
+	return 5 + 5 + 3 + count * element_size;
+}
+
+static unsigned si_ce_needed_cs_space(void)
+{
+	unsigned space = 0;
+
+	space += si_descriptor_list_cs_space(SI_NUM_CONST_BUFFERS, 4);
+	space += si_descriptor_list_cs_space(SI_NUM_RW_BUFFERS, 4);
+	space += si_descriptor_list_cs_space(SI_NUM_SHADER_BUFFERS, 4);
+	space += si_descriptor_list_cs_space(SI_NUM_SAMPLERS, 16);
+	space += si_descriptor_list_cs_space(SI_NUM_IMAGES, 8);
+
+	space *= SI_NUM_SHADERS;
+
+	/* Increment CE counter packet */
+	space += 2;
+
+	return space;
+}
+
 /* initialize */
 void si_need_cs_space(struct si_context *ctx)
 {
 	struct radeon_winsys_cs *cs = ctx->b.gfx.cs;
+	struct radeon_winsys_cs *ce_ib = ctx->ce_ib;
 	struct radeon_winsys_cs *dma = ctx->b.dma.cs;
 
 	/* Flush the DMA IB if it's not empty. */
@@ -53,7 +84,9 @@ void si_need_cs_space(struct si_context *ctx)
 	/* If the CS is sufficiently large, don't count the space needed
 	 * and just flush if there is not enough space left.
 	 */
-	if (unlikely(cs->cdw > cs->max_dw - 2048))
+	if (unlikely(cs->cdw > cs->max_dw - 2048 ||
+                     (ce_ib && ce_ib->max_dw - ce_ib->cdw <
+                      si_ce_needed_cs_space())))
 		ctx->b.gfx.flush(ctx, RADEON_FLUSH_ASYNC, NULL);
 }
 
diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
index 6a990ed64c3..ddfa59fd128 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.c
+++ b/src/gallium/drivers/radeonsi/si_pipe.c
@@ -142,6 +142,21 @@ static struct pipe_context *si_create_context(struct pipe_screen *screen,
 
 	sctx->b.gfx.cs = ws->cs_create(sctx->b.ctx, RING_GFX,
 				       si_context_gfx_flush, sctx);
+
+	if (!(sscreen->b.debug_flags & DBG_NO_CE) && ws->cs_add_const_ib) {
+		sctx->ce_ib = ws->cs_add_const_ib(sctx->b.gfx.cs);
+		if (!sctx->ce_ib)
+			goto fail;
+
+		if (ws->cs_add_const_preamble_ib) {
+			sctx->ce_preamble_ib =
+			           ws->cs_add_const_preamble_ib(sctx->b.gfx.cs);
+
+			if (!sctx->ce_preamble_ib)
+				goto fail;
+		}
+	}
+
 	sctx->b.gfx.flush = si_context_gfx_flush;
 
 	/* Border colors. */
diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index 0398b1df61e..b8db3b2ce85 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -191,6 +191,9 @@ struct si_context {
 	void				*custom_blend_dcc_decompress;
 	void				*pstipple_sampler_state;
 	struct si_screen		*screen;
+	struct radeon_winsys_cs		*ce_ib;
+	struct radeon_winsys_cs		*ce_preamble_ib;
+
 	struct pipe_fence_handle	*last_gfx_fence;
 	struct si_shader_ctx_state	fixed_func_tcs_shader;
 	LLVMTargetMachineRef		tm;